Object Oriented Programming

via Fortran 90/95

Ed Akin
Rice University
Mechanical Engineering and Materials Science Department
Houston, Texas

May 29, 2001

Draft # 4.2, Copyright c 2001, All rights reserved.
ii
Contents

Preface vii

1 Program Design 1
1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3. Modular Program Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4. Program Composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.1. Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.2. Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.3. Flow Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4.4. Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.5. Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.6. Dynamic Memory Management . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5. Program evaluation and testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6. Program documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.7. Object Oriented Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.8. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 Data Types 23
2.1. Intrinsic Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2. User Defined Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3. Abstract Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4. Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3 Object Oriented Programming Concepts 33

3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2. Encapsulation, Inheritance, and Polymorphism . . . . . . . . . . . . . . . . . . . . . 34
3.2.1. Example Date, Person, and Student Classes . . . . . . . . . . . . . . . . . . . 37
3.3. Object Oriented Numerical Calculations . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.1. A Rational Number Class and Operator Overloading . . . . . . . . . . . . . . 39
3.4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4 Features of Programming Languages 51

4.1. Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2. Statements and Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3. Flow Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.3.1. Explicit Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3.2. Implied Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.3. Conditionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4. Subprograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

c 2001 J.E. Akin iii

 4.4.1. Functions and Subroutines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4.2. Global Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.3. Bit Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.4.4. Exception Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.5. Interface Prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.6. Characters and Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.7. User Defined Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.7.1. Overloading Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.7.2. User Defined Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.8. Pointers and Targets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.8.1. Pointer Type Declaration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.8.2. Pointer Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.8.3. Using Pointers in Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.8.4. Pointers and Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.9. Accessing External Source Files and Functions . . . . . . . . . . . . . . . . . . . . . 89
4.10. Procedural Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.10.1. Fitting Curves to Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.10.2. Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.11. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5 Object Oriented Methods 103

5.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.2. The Drill Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3. Global Positioning Satellite Distances . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.4. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

6 Inheritance and Polymorphism 119

6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
6.2. Example Applications of Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2.1. The Professor Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
6.2.2. The Employee and Manager Classes . . . . . . . . . . . . . . . . . . . . . . . 121
6.3. Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.3.1. Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.3.2. Subtyping Objects (Dynamic Dispatching) . . . . . . . . . . . . . . . . . . . 130
6.4. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

7 OO Data Structures 135

7.1. Data Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.2. Stacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.3. Queues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.4. Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.4.1. Singly Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.4.2. Doubly Linked Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.5. Direct (Random) Access Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.6. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

8 Arrays and Matrices 155

8.1. Subscripted Variables: Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.1.1. Initializing Array Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.1.2. Intrinsic Array Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.1.3. Colon Operations on Arrays (Subscript Triplet) . . . . . . . . . . . . . . . . . 159
8.1.4. Array Logical Mask Operators . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.1.5. User Defined Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.1.6. Connectivity Lists and Vector Subscripts . . . . . . . . . . . . . . . . . . . . 166

c 2001 J.E. Akin iv

 8.1.7. Component Gather and Scatter . . . . . . . . . . . . . . . . . . . . . . . . . . 168
8.2. Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.2.1. Matrix Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
8.2.2. Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
8.2.3. Factorizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
8.2.4. Determinant of a Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
8.2.5. Matrix Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.2.6. Computation with Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
8.3. Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

9 Advanced Topics 181

9.1. Templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
9.2. Subtyping Objects (Dynamic Dispatching) . . . . . . . . . . . . . . . . . . . . . . . . 183
9.3. Non-standard Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

A Bibliography 187

B Fortran 90 Overview 191

B.1. List of Language Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
B.2. Alphabetical Table of Fortran 90 Intrinsic Routines . . . . . . . . . . . . . . . . . . . 17
B.3. Syntax of Fortran 90 Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

C Selected Exercise Solutions 47

C.1. Problem 1.8.1 : Checking trigonometric identities . . . . . . . . . . . . . . . . . . . 47
C.2. Problem 1.8.2 : Newton-Raphson algorithm . . . . . . . . . . . . . . . . . . . . . . 47
C.3. Problem 1.8.3 : Game of life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
C.4. Problem 2.5.1 : Conversion factors . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
C.5. Problem 3.5.3 : Creating a vector class . . . . . . . . . . . . . . . . . . . . . . . . . 50
C.6. Problem 3.5.4 : Creating a sparse vector class . . . . . . . . . . . . . . . . . . . . . 56
C.7. Problem 4.11.1 : Count the lines in an external file . . . . . . . . . . . . . . . . . . . 61
C.8. Problem 4.11.3 : Computing CPU time useage . . . . . . . . . . . . . . . . . . . . . 62
C.9. Problem 4.11.4 : Converting a string to upper case . . . . . . . . . . . . . . . . . . . 62
C.10. Problem 4.11.8 : Read two values from each line of an external file . . . . . . . . . . 63
C.11. Problem 4.11.14 : Two line least square fits . . . . . . . . . . . . . . . . . . . . . . . 63
C.12. Problem 4.11.15 : Find the next available file unit . . . . . . . . . . . . . . . . . . . 65
C.13. Problem 5.4.4 : Polymorphic interface for the class ‘Position Angle’ . . . . . . . . . 66
C.14. Problem 6.4.1 : Using a function with the same name in two classes . . . . . . . . . . 67
C.15. Problem 6.4.3 : Revising the employee-manager classes . . . . . . . . . . . . . . . . 67

D Companion C++ Examples 69

D.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

E Glossary of Object Oriented Terms 77

F Subject Index 0

G Program Index 1

c 2001 J.E. Akin v

vi
Preface

There has been an explosion of interest in, and books on object-oriented programming (OOP). Why have
yet another book on the subject? In the past a basic education was said to master the three r’s: reading,
’riting, and ’rithmetic. Today a sound education in engineering programming leads to producing code that
satisfy the four r’s: readability, reusability, reliability, and really-efficient. While some object-oriented
programming languages have some of these abilities Fortran 90/95 offers all of them for engineering
applications. Thus this book is intended to take a different tack by using the Fortran 90/95 language as its
main OOP tool. With more than one hundred pure and hybrid object-oriented languages available, one
must be selective in deciding which ones merit the effort of learning to utilize them. There are millions
of Fortran programmers, so it is logical to present the hybrid object-oriented features of Fortran 90/95 to
them to update and expand their programming skills. This work provides an introduction to Fortran 90
as well as to object-oriented programming concepts. Even with the current release (Fortran 95) we will
demonstrate that Fortran offers essentially all of the tools recommended for object-oriented programming
techniques. It is expected that Fortran 200X will offer additional object-oriented capabilities, such as
declaring ”extensible” (or virtual) functions. Thus, it is expected that the tools learned here will be of
value far into the future.
It is commonly agreed that the two decades old F77 standard for the language was missing several
useful and important concepts of computer science that evolved and were made popular after its release,
but it also had a large number of powerful and useful features. The following F90 standard included
a large number of improvements that have often been overlooked by many programmers. It is fully
compatible with all old F77 standard code, but it declared several features of that standard as obsolete.
That was done to encourage programmers to learn better methods, even though the standard still supports
those now obsolete language constructs. The F90 standards committee brought into the language most of
the best features of other more recent languages like Ada, C, C++, Eiffel, etc. Those additions included in
part: structures, dynamic memory management, recursion, pointers (references), and abstract data types
along with their supporting tools of encapsulation, inheritance, and the overloading of operators and
routines. Equally important for those involved in numerical analysis the F90 standard added several new
features for efficient array operations that are very similar to those of the popular M ATLAB environment.
Most of those features include additional options to employ logical filters on arrays. All of the new array
features were intended for use on vector or parallel computers and allow programmers to avoid the bad
habit of writing numerous serial loops. The current standard, F95, went on to add more specific parallel
array tools, provided “pure” routines for general parallel operations, simplified the use of pointers, and
made a few user friendly refinements of some F90 features. Indeed, at this time one can view F90/95 as
the only cross-platform international standard language for parallel computing. Thus Fortran continues
to be an important programming language that richly rewards the effort of learning to take advantage of
its power, clarity, and user friendlyness.
We begin that learning process in Chapter 1 with an overview of general programming techniques.
Primarily the older “procedural” approach is discussed there, but the chapter is closed with an outline of
the newer “object” approach to programming. An experienced programmer may want to skip directly to
the last section of Chapter 1 where we outline some object-oriented methods. In Chapter 2, we introduce
the concept of the abstract data types and their extension to classes. Chapter 3 provides a fairly detailed
introduction to the concepts and terminology of object-oriented programming. A much larger supporting
glossary is provided as an appendix.
For the sake of completeness Chapter 4 introduces language specific details of the topics discussed in

c 2002 J.E. Akin vii

the first chapter. The Fortran 90/95 syntax is used there, but in several cases cross-references are made to
similar constructs in the C++ language and the M ATLAB environment. While some readers may want to
skip Chapter 4, it will help others learn the Fortran 90/95 syntax and/or to read related publications that
use C++ or M ATLAB. All of the syntax of Fortran 90 is also given in an appendix.
Since many Fortran applications relate to manipulating arrays or doing numerical matrix analysis,
Chapter 5 presents a very detailed coverage of the powerful intrinsic features Fortran 90 has added to
provide for more efficient operations with arrays. It has been demonstrated in the literature that object-
oriented implementations of scientific projects requiring intensive operations with arrays execute much
faster in Fortran 90 than in C++. Since Fortran 90 was designed for operations on vector and parallel
machines that chapter encourages the programmer to avoid unneeded serial loops and to replace them
with more efficient intrinsic array functions. Readers not needing to use numerical matrix analysis may
skip Chapter 5.
Chapter 6 returns to object-oriented methods with a more detailed coverage of using object-oriented
analysis and object-oriented design to create classes and demonstrates how to implement them as an OOP
in Fortran 90. Additional Fortran 90 examples of inheritance and polymorphism are given in Chapter
7. Object-oriented programs often require the objects to be stored in some type of “container” or data
structure such as a stack or linked-list. Fortran 90 object-oriented examples of typical containers are
given in Chapter 8. Some specialized topics for more advanced users are given in Chapter 9, so beginning
programmers could skip it.
To summarize the two optional uses of this text; it is recommended that experienced Fortran program-
mers wishing to learn to use OOP cover Chapters 2, 3, 6, 7, 8, and 9, while persons studying Fortran for
the first time should cover Chapters 1, 2, 3, and. Anyone needing to use numerical matrix analysis should
also include Chapter 5.
A OO glossary is included in an appendix to aid in reading this text and the current literature on OOP.
Another appendix on Fortran 90 gives an alphabetical listing on its intrinsic routines, a subject based
list of them, a detailed syntax of all the F90 statements, and a set of example uses of every statement.
Selected solutions for most of the assignments are included in another appendix along with comments
on those solutions. The final appendix gives the C++ versions of several of the F90 examples in the
text. They are provided as an aid to understanding other OOP literature. Since F90 and M ATLAB are so
similar the corresponding M ATLAB versions often directly follow the F90 examples in the text.

Ed Akin, Rice University, 2002

c 2002 J.E. Akin viii

Index

abstract data type, 15, 23, 27 rectangular, 170

abstraction, 19, 27 reshape, 159
access, 36 shape, 159
access restriction, 19 shift, 172
accessibility, 19 size, 159
accessor, 18 unknown size, 77
actual argument, 56 variable rank, 160
Ada, 33 array operations, 163
addition, 56 ASCII, 23
ADT, see abstract data type ASCII character set, 77, 78, 99, 163
ADVANCE specifier, 42, 103 assembly language, 15
agent, 18 assignment operator, 10, 39
algorithm, 51 ASSOCIATED, 15
ALLOCATABLE, 15 ASSOCIATED function, 76, 89
allocatable array, 160, 161 ASSOCIATED intrinsic, 132, 134
ALLOCATE, 15 associative, 176, 177
allocate, 42 asterisk (*), 58
ALLOCATE statement, 75, 93 ATAN2, 13
ALLOCATED, 15 attribute, 105, 106, 109, 121, 125
allocation status, 75 private, 27, 125
AND operand, 42 public, 27
terminator, 25
area, 34
attribute terminator, 25
argument
attributes, 19, 27
inout, 71
automatic array, 90, 160, 161
input, 71
automatic deallocation, 29
interface, 76
none, 71 BACKSPACE statement, 76
number of, 76 bad style, 162
optional, 76, 77 base class, 121
order, 76 behavior, 106, 109
output, 71 binary file, 163
rank, 76 bit
returned value, 76 clear, 75
type, 76 extract, 75
array, 26, 60, 67, 83 set, 75
allocatable, 160 shift, 75
assumed shape, 77 test, 75
automatic, 90, 160 bit manipulation, 75
Boolean, 168 blanks
constant, 160 all, 78
dummy dimension, 160 leading, 78
flip, 170 trailing, 78
mask, 168, 183 Boolean, 53
rank, 77, 159, 161, 170 Boolean value, 23

4
bottom-up, 4 Drill, 106
bounds, 159 Global Position, 114
bubble sort, 93, 95 Great Arc, 114
ordered, 96 Position Angle, 114
bug, 9 clipping function, 14, 71
CLOSE statement, 75
C, 1, 33, 52 Coad/Yourdon method, 18
C++, 1, 10, 14, 24, 33, 52, 58, 60, 77, 82, 103, colon operator, 56, 61, 62, 78, 160, 163, 167,
123 170
CALL statement, 42 column major order, 181
CASE DEFAULT statement, 64 column matrix, 174
CASE statement, 64 column order, 162
cases, 62 comma, 99
central processor unit, 73 comment, 1, 2, 7, 9, 12, 52
character, 82 commutative, 101, 176, 177
case change, 81 compiler, 10, 15, 91
control, 77 complex, 10, 82, 165
from number, 81 COMPLEX type, 23, 53
functions, 78 COMPLEX type , 24
non-print, 103 composition, 34, 36
non-printable, 77 conditional, 7–9, 11, 51, 58
strings, 77 conformable, 176
to number, 81 connectivity, 170
character set, 23 constant array, 160
CHARACTER type, 23, 26, 53 constructor, 18, 29, 34, 125, 134, 135
chemical element, 25 default, 18
circuits, 170 intrinsic, 18, 26, 34, 39
circular shift, 172 manual, 36
class, 15, 19, 33 public, 37
base, 18 structure, 26
Date, 120, 123 CONTAINS statement, 29, 33, 34, 73, 76, 86
derived, 18 continuation marker, 10
Drill, 105 control key, 79
Employee, 125 conversion factors, 29
Geometric, 120 count-controlled DO, 12, 13
Global Position, 114 CPU, see central processor unit
Great Arc, 114 curve fit, 91
hierarchy, 33 CYCLE statement, 66
instance, 33
Manager, 125, 135 data abstraction, 19
Person, 120, 123 data hiding, 36
polymorphic, 133 data types, 10
Position Angle, 109, 114 intrinsic, 23
Professor, 123 user defined, 23
Student, 120, 123 date, 101
class code DEALLOCATE, 15
class Angle, 114 deallocate, 18, 42
class Circle, 34 DEALLOCATE statement, 75
class Date, 37 debugger, 17
class Fibonacci Number, 29 debugging, 16
class Person, 37 default case, 64
class Rational, 42 default value, 29
class Rectangle, 34 dereference, 58
class Student, 37 derived class, 121

c 2002 J.E. Akin 5

derived type, 15, 23 exception, 75
component, 83 exception handler, 75
nested, 83 exception handling, 18
print, 85 exercises, 48
read, 85 EXIT statement, 66, 67
destructor, 29, 34, 41, 48 explicit loop, 11
determinant, 179 exponent range, 24
diagonal matrix, 174 exponentiation, 56
dimension expression, 10, 51, 52, 89
constant, 161 external subprogram, 77
extent, 159
lower bound, 159 factorization, 178, 179
upper bound, 159 FALSE result, 63
distributive, 177 Fibonacci number, 29
division, 56 file, 75
DO statement, 29, 60, 62 column count, 100
DO WHILE statement, 67 internal, 81
DO-EXIT pair, 68, 69 line count, 100
documentation, 17 read status, 100
domain, 19 unit number, 101
dot product, 12 finite element, 43
double, 24 flip, 167, 170
DOUBLE PRECISION type, 23, 24, 53 float, 53
dummy argument, 56, 73 floating point, see real, 23, 24, 183
dummy dimension, 161 flow control, 11, 51, 58
dummy dimension array, 160 FORMAT statement, 34
dummy variable, 73 function, 7, 9, 51, 69, 70
dynamic binding, 18 argument, 13, 15
dynamic data structures, 38 extensible, 132
dynamic dispatching, 132 recursive, 42, 102
dynamic memory, 75 result, 70
allocation, 15 return, 13
de-allocation, 15 variable, 15
management, 15 function code
dynamic memory management, 89 Add, 29
add Rational, 42
e, 25 Angle , 114
EBCDIC, 23 circle area, 34
EBCDIC character set, 77 clip, 71
Eiffel, 18 convert, 42
electric drill, 105 copy Rational, 42
ELSE statement, 42, 63, 67 Date , 37
encapsulate, 15 Decimal min, 114
encapsulation, 27, 33 Decimal sec, 114
end off shift, 172 Default Angle, 114
end-of-file, 76 Drill , 106
end-of-record, 76 gcd, 42, 102
end-of-transmission, 78 get Arc, 114
EOF, see end-of-file get Denominator, 42
EOR, see end-of-record get Latitude, 114
EOT, see end of transmission get Longitude, 114
equation get mr rate, 106
number, 173 get next io unit, 103
error checking, 18 Get Next Unit, 99

c 2002 J.E. Akin 6

 get Numerator, 42 part of, 18
get person, 37 horizontal tab, 78
get torque, 106 Hubbard, J.R., 36
Great Arc , 114 hyperbolic tangent, 103
inputCount, 93
Int deg, 114 identity matrix, 182
Int deg min, 114 if, 12
Int deg min sec, 114 IF ELSE statement, 62
is equal to, 42 IF statement, 29, 37, 42, 62
make Person, 37 if-else, 12
make Rational, 42 IF-ELSE pair, 63
make Rectangle, 36 IF-ELSEIF, 132
make Student, 37 IMPLICIT COMPLEX, 53
mid value, 70 IMPLICIT DOUBLE PRECISION, 53
mult Fraction, 87 IMPLICIT INTEGER, 52
mult Rational, 42 implicit loop, 12
new Fibonacci Number, 29 IMPLICIT NONE, 26, 29
Person , 37 IMPLICIT REAL, 52
Rational , 42 implied loop, 61, 62, 160, 170
rectangle area, 34 INCLUDE line, 37, 42, 90
set Date, 37 INDEX intrinsic, 81
set Lat and Long at, 114 indexed loop, 11
Student , 37 infinite loop, 9, 68, 69
toc, 73 inheritance, 18, 33, 34, 73, 121
to Decimal Degrees, 114 inherited, 37
to lower, 81 inner loop, 62
to Radians, 114 INQUIRE intrinsic, 93, 98, 103
to upper, 81, 101 INQUIRE statement, 76
FUNCTION statement, 29 instance, 33, 124
integer, 10, 82, 165
Game of Life, 4 INTEGER type, 23, 24, 53
Gamma, 25 intent
gather-scatte, 172 in, 29
gcd, see greatest common divisor inout, 29
generic function, 33, 34 statement, 29
generic interface, 134 INTENT statement, 29, 58, 71, 94
generic name, 34 interface, 2, 6, 9, 13, 15, 27, 34, 76, 93, 106,
generic object, 42 109, 123
generic routine, 123 general form, 77
generic subprogram, 77 human, 18
geometric shape, 34 input/output, 18
global positioning satellite, 108 prototype, 18
global variable, 14, 73 INTERFACE ASSIGNMENT (=) block, 87
GO TO statement, 65, 66 interface block, 34, 77
GPS, see global positioning satellite interface body, 77
Graham method, 18 interface operator (*), 39
graphical representation, 27, 120 INTERFACE OPERATOR block, 86, 87
greatest common divisor, 42, 102 INTERFACE OPERATOR statement, 170
interface prototype, 105, 106, 125
Has-A, 109 INTERFACE statement, 34
header file, 131 internal file, 81
hello world, 52, 101 internal sub-programs, 73
hierarchie interpreter, 10, 15
kind of, 18 intrinsic, 170

c 2002 J.E. Akin 7

intrinsic constructor, 86, 99, 108 counter, 60
intrinsic function, 12, 70 cycle, 66
inverse, 182 exit, 60, 66
IOSTAT = variable, 75, 76 explicit, 59
Is-A, 108, 109 implied, 61
ISO VARIABLE LENGTH STRING, 23 index, 101
Is A, 126 infinite, 60, 68
nested, 62, 66
keyword, 123 pseudocode, 59
KIND intrinsic, 24 skip, 66
Kind-Of, 109, 125 until, 67, 68
variable, 60
latitude, 108 while, 67
least squares, 91 loop control, 61, 162
LEN intrinsic, 78, 81 loop index, 101
length loop variable, 11
line, 52 lower triangle, 175, 178
name, 52
LEN TRIM intrinsic, 78 manual constructor, 86, 106
lexical operator, 95 manual page, 17
lexically mask, 165, 168, 169, 183
greater than, 78 masks, 62
less than, 78 Mathematica, 51
less than or equal, 78 mathematical constants, 25
library function, 16 Matlab, 1, 10, 14, 52, 61, 70, 100, 103
line continuation, 101 MATMUL intrinsic, 177
linear equations, 177, 178 matrix, 159, 174
linked list, 38, 88, 89 addition, 176
linker, 16, 90 algebra, 159
list column, 174
doubly-linked, 89 compatible, 176
singly-linked, 89 determinant, 179
logarithm, 70, 92 diagonal, 174
logical, 82 factorization, 178
AND, 63 flip, 167
equal to, 63 identity, 178
EQV, 63 inverse, 90, 178
greater than, 63 multiplication, 163, 176
less than, 63 non-singular, 178
NEQV, 63 null, 174
NOT, 63 skew symmetric, 175
operator, 63 solve, 90
OR, 63 square, 174, 175
logical expression, 11 symmetric, 175
logical mask, 62 Toeplitz, 175
logical operator, 63 transpose, 163, 175
LOGICAL type, 23, 42 triangular, 175, 178
long, 24 matrix addition, 181, 182
long double, 24 matrix algebra, 159, 176
long int, 24 matrix multiplication, 169, 177, 182
longitude, 108 matrix operator, 38
loop, 5, 7–9, 11, 51, 58, 183 matrix transpose, 169
abort, 68 maximum values, 71
breakout, 66 MAXLOC intrinsic, 71

c 2002 J.E. Akin 8

MAXVAL intrinsic, 71 NULL function (f95), 89
mean, 70 NULLIFY, 15
member, 121 nullify, 134
message, 27 NULLIFY statement, 89
methods, 3 number
private, 27 bit width, 24
public, 27 common range, 24
military standards, 75 label, 60
minimum values, 71 significant digits, 24
MINLOC intrinsic, 71 truncating, 166
MINVAL intrinsic, 71 type, 24
modular design, 6 numeric types, 23
module, 15, 25, 33, 69 numerical computation, 38
module code
class Angle, 114 object, 15, 19, 33
class Circle, 34 object oriented
class Date, 37 analysis, 18, 43, 105, 109, 120
class Fibonacci Number, 29 approach, 18
class Global Position, 114 design, 18, 43, 105, 109, 120
class Great Arc, 114 language, 18
class Person, 37 programming, 18, 105
class Position Angle, 114 representation, 18
class Rational, 42 Object Pascal, 18
class Rectangle, 34 OOA, see object oriented analysis
class Student, 37 OOD, see object oriented design
exceptions, 76 OOP, see object oriented programming
Fractions, 87 OPEN statement, 75, 163
Math Constants, 25 operator, 27
record Module, 97 .op., 87, 169
tic toc, 73, 101 .solve., 90, 91
MODULE PROCEDURE statement, 34, 39, 86, .t., 170
87, 170 .x., 170
MODULE statement, 29 assignment, 39
module variable, 29 binary, 87
modulo function, 56 defined, 18, 87
multiple inheritanc, 121 extended, 87
multiplication, 56 overloaded, 18
Myer, B., 18 overloading, 39, 86
symbol, 87
NAG, see National Algorithms Group unary, 87
named user defined, 77, 169
CYCLE, 66, 67 operator overloading, 10
DO, 67 operator precedence, 52
DO loop, 60 operator symbol, 169
EXIT, 66, 67 optional argument, 29, 37, 76
IF, 64 OPTIONAL attribute, 29, 36, 106
SELECT CASE, 64 OR operand, 37
National Algorithms Group, 91 ordering array, 96
nested outer loop, 62
DO, 67 overloaded member, 123
IF, 62 overloading, 39, 48, 86
new line, 79, 103 testing, 87
Newton-Raphson method, 11
non-advancing I/O, 42 package, 15

c 2002 J.E. Akin 9

parallel computer, 43 array indexing, 60
PARAMETER attribute, 25 clip an array, 71
Part-Of, 109 create a type, 26
partial derivative, 180 declare interface, 77
partitioned matrix, 175 Fibonacci, 29
pass by reference, 57, 77, 88 geometry, 34
pass by value, 57, 58, 77 if else logic, 63
path name, 37 linear fit, 93
pi, 25 Logical operators, 63
pointer, 10, 23, 76, 87 main, 37, 42
allocatable, 15 operate on strings, 79
arithmetic, 88 relational operators, 63
assignment, 89 simple loop, 60
association, 88 string to numbers, 81
declaration, 88 structure components, 85
dereference, 58 test bubble, 98
detrimental effect, 88 test Drill, 108
in expression, 89 test Fractions, 87
inquiry, 89 test Great Arc, 114
nullify, 89 program keyword, 56
status, 15, 88 PROGRAM statement, 26, 29
target, 88 projectile, 102
pointer object, 133 prototype, 6, 76
pointer variable, 87 pseudo-pointer, 96
polymorphic class, 133 pseudocode, 5, 14, 51, 71, 102
polymorphic interface, 120 public, 33, 125
polymorphism, 18, 33, 34, 121, 126 PUBLIC attribute, 29
portability, 15 public constructor, 37
pre-processor, 131 public method, 27
precedence rules, 11 PUBLIC statement, 27
precision, 183
quadratic equation, 3
double, 82
queue, 89
kind, 24
portable, 82 rank, 161
single, 82 rational number, 38, 39
specified, 82 read error, 103
underscore, 24 READ statement, 29, 62, 76
user defined, 24 real, 10, 82, 165
precision kind, 24 REAL type, 23, 24, 53
PRESENT function, 76 recursive algorithm, 88
PRESENT intrinsic, 29, 36 RECURSIVE qualifier, 42, 102
PRINT * statement, 29 reference, 10
private, 33, 106 relational operator, 52, 63, 78
PRIVATE attribute, 29, 36 remainder, 56
private attributes, 37 rename modifier, 121
PRIVATE statement, 27 reshape, 162
procedural programming, 18 RESULT option, 29
procedure, 69 result value, 70
program return, 161
documentation, 17 RETURN statement, 66
executable, 17 REWIND statement, 76
scope, 14
program code, 114 sample data, 99

c 2002 J.E. Akin 10

scatter, 173 to number, 81
scope, 14 trim, 78
SELECT CASE statement1, 64 verify, 78
SELECTED INT KIND, 23, 24 strings, 77
SELECTED REAL KIND, 23, 24 strong typing, 53
selector symbol, 26, 29, 34 struct, 53
server, 18 structure, 23, 25, 33, 85
short, 24 structure constructor, 26
size, 12 structured programming, 13
SIZE intrinsic, 70, 90, 93, 159 submatrix, 175
Smalltalk, 18 subprogram, 69
sort, 87, 91, 93, 96, 127 recursive, 102
bubble, 93 subroutine, 69, 70
characters, 95 subroutine code, 114
object, 97 assign, 87
objects, 95 Change, 77
strings, 95 delete Rational, 42
sorting, 42 equal Fraction, 87
sparse vector, 49 equal Integer, 42
sparse vector class, 183 exception status, 76
specification, 4 in, 106
SQRT intrinsic, 27 Integer Sort, 96, 99
square root, 27, 56, 70 invert, 42
stack, 89 list, 42
STAT = variable, 75 List Angle, 114
statement, 2, 9 List Great Arc, 114
statement block, 12, 58 List Position, 114
statements, 1 List Position Angle, 114
status List Pt to Pt, 114
FILE, 76 lsq fit, 93
IOSTAT =, 76 mult Fraction, 87
MODE, 76 No Change, 77
OPENED =, 76 out, 106
status checking, 161 Print, 29
STOP statement, 37 print Date, 37
storage print DOB, 37
column wise, 159 print DOD, 37
row wise, 159 print DOM, 37
string, 23, 56 print GPA, 37
adjust, 78 print Name, 37
case change, 81 print Sex, 37
character number, 78 readData, 93, 101
collating sets, 78 read Date, 37
colon operator, 78 Read Position Angle, 114
concatenate, 78 reduce, 42
copy, 78 set DOB, 37
dynamic length, 77 set DOD, 37
from number, 81 set DOM, 37
functions, 78 set Latitude, 114
length, 78 set Longitude, 114
logic, 78 simple arithmetic, 56
repeat, 78 Sort Reals, 94
scan, 78 Sort String, 95

c 2002 J.E. Akin 11

 String Sort, 99 type, 10
test matrix, 90 variable rank array, 160
tic, 73 vector, 159
SUBROUTINE statement, 29 vector class, 48, 183
subroutines, 33 vector subscript, 62, 170
subscrip, 159 volume, 48
subscript, 26, 60
bounds, 159 WHERE construct, 169
range, 181 WHERE statement, 62, 67, 169
vector, 170 while-true, 68
subtraction, 56 wildcard, 128
subtype, 133 WRITE statement, 34, 62, 76
subtyping, 126, 132
sum, 12
SUM intrinsic, 70, 93, 169
super class, 121
syntactic error, 17
SYSTEM CLOCK intrinsic, 73

tab, 79, 99, 103

TARGET, 15
target, 23, 76, 88, 89
template, 43, 126, 128
tensor, 159
testing, 15
time of day, 101
Toeplitz matrix, 175
top-down, 4
transformational functions, 169
transpose, 163, 175, 177
TRANSPOSE intrinsic, 170
tree structure, 38, 88, 89
triplet, see colon operator
true, 12
TRUE result, 63
truss, 170
type
conversion, 81
default, 52
implicit, 52
TYPE declaration, 26, 29
TYPE statement, 27, 34

unexpected result, 169

upper triangle, 175, 178
USE association, 121, 125
USE statement, 29, 33, 34, 37, 86, 90
user defined operator, 169
user interface, 2

validation, 29
variable, 8, 10, 23, 51
global, 14
name, 10

c 2002 J.E. Akin 12

Chapter 1

Program Design

1.1 Introduction
The programming process is similar in approach and creativity to writing a paper. In composition, you
are writing to express ideas; in programming you are expressing a computation. Both the programmer
and the writer must adhere to the syntactic rules (grammar) of a particular language. In prose, the funda-
mental idea-expressing unit is the sentence; in programming, two units statements and comments are
available.
Standing back, composition from technical prose to fiction should be organized broadly, usually
through an outline. The outline should be expanded as the detail is elaborated, and the whole re-examined
and re-organized when structural or creative flaws arise. Once the outline settles, you begin the actual
composition process, using sentences to weave the fabric your outline expresses. Clarity in writing
occurs when your sentences, both internally and globally, communicate the outline succinctly and clearly.
We stress this approach here, with the aim of developing a programming style that produces efficient
programs that humans can easily understand.
To a great degree, no matter which language you choose for your composition, the idea can be ex-
pressed with the same degree of clarity. Some subtleties can be better expressed in one language than
another, but the fundamental reason for choosing your language is your audience: People do not know
many languages, and if you want to address the American population, you had better choose English
over Swahili. Similar situations happen in programming languages, but they are not nearly so complex
or diverse. The number of languages is far fewer, and their differences minor. Fortran is the oldest lan-
guage among those in use today. C and C++ differ from it somewhat, but there are more similarities
than not. M ATLAB’s language, written in C and Fortran, was created much later than these two, and its
structure is so similar to the others that it can be easily mastered. The C++ language is an extension of
the C language that places its emphasis on object oriented programming (OOP) methods. Fortran added
object oriented capabilities with its F90 standard, and additional enhancements for parallel machines
were issued with F95. The Fortran 2000 standard is planned to contain more user-friendly constructs for
polymorphism and will, thus, enhance its object-oriented capabilities. This creation of a new language
and its similarity to more established ones are this book’s main points: More computer programming lan-
guages will be created during your career, but these new languages will probably not be much different
than ones you already know. Why should new languages evolve? In MATLAB’s case, it was the desire to
express matrix-like expressions easily that motivated its creation. The difference between M ATLAB and
Fortran 90 is infinitesimally small compare to the gap between English and Swahili.
An important difference between programming and composition is that in programming you are writ-
ing for two audiences: people and computers. As for the computer audience, what you write is “read” by
interpreters and compilers specific to the language you used. They are very rigid about syntactic rules,
and perform exactly the calculations you say. It is like a document you write being read by the most de-
tailed, picky person you know; every pronoun is questioned, and if the antecedent is not perfectly clear,
then they throw up their hands, rigidly declaring that the entire document cannot be understood. Your
picky friend might interpret the sentence “Pick you up at eight” to mean that you will literally lift him or
her off the ground at precisely 8 o’clock, and then demand to know whether the time is in the morning or

c 2001 J.E. Akin 1
afternoon and what the date is.
Humans demand even more from programs. This audience consists of two main groups, whose goals
can conflict. The larger of the two groups consists of users. Users care about how the program presents
itself, its user interface, and how quickly the program runs, how efficient it is. To satisfy this audience,
programmers may use statements that are overly terse because they know how to make the program more
readable by the computer’s compiler, enabling the compiler to produce faster, but less human-intelligible
program. This approach causes the other portion of the audience programmers to boo and hiss. The
smaller audience, of which you are also a member, must be able to read the program so that they can
enhance and/or change it. A characteristic of programs, which further distinguishes it from prose, is
that you and others will seek to modify your program in the future. For example, in the 1960s when
the first version of Fortran was created, useful programs by today’s standards (such as matrix inversion)
were written. Back then, the user interface possibilities were quite limited, and the use of visual displays
was limited. Thirty years later, you would (conceivably) want to take an old program, and provide a
modern user interface. If the program is structurally sound (a good outline and organized well) and is
well-written, re-using the “good” portions is easy accomplished.
The three-audience situation has prompted most languages to support both computer-oriented and
human-oriented “prose”. The program’s meaning is conveyed by statements, and is what the computer
interprets. Humans read this part, which in virtually all languages bears a strong relationship to mathe-
matical equations, and also read comments. Comments are not read by the computer at all, but are there
to help explain what might be expressed in a complicated way by programming language syntax. The
document or program you write today should be understandable tomorrow, not only by you, but also by
others. Sentences and paragraphs should make sense after a day or so of gestation. Paragraphs and larger
conceptual units should not make assumptions or leaps that confuse the reader. Otherwise, the document
you write for yourself or others served no purpose. The same is true with programming; the program’s
organization should be easy to follow and the way you write the program, using both statements and com-
ments, should help you and others understand how the computation proceeds. The existence of comments
permits the writer to directly express the program’s outline in the program to help the reader comprehend
the computation.
These similarities highlight the parallels between composition and programming. Differences become
evident because programming is, in many ways, more demanding than prose writing. On one hand, the
components and structure of programming languages are far simpler than the grammar and syntax of any
verbal or written language. When reading a document, you can figure out the misspelled words, and not
be bothered about every little imprecision in interpreting what is written. On the other, simple errors, akin
to misspelled words or unclear antecedents, can completely obviate a program, rendering it senseless or
causing it to go wildly wrong during execution. For example, there is no real dictionary when it comes
to programming. You can define variable names containing virtually any combination of letters (upper
and lower case), underscores, and numbers. A typographical error in a variable’s name can therefore
lead to unpredictable program behavior. Furthermore, computer execution speeds are becoming faster
and faster, meaning that increasingly complex programs can run very quickly. For example, the program
(actually groups of programs) that run NASA’s space shuttle might be comparable in size to Hugo’s Les
Misérables, but its complexity and immediate importance to the “user” far exceeds that of the novel.
As a consequence, program design must be extremely structured, having the ultimate intentions of
performing a specific calculation efficiently with attractive, understandable, efficient programs. Achiev-
ing these general goals means breaking the program into components, writing and testing them separately,
then merging them according to the outline. Toward this end, we stress modular programming. Modules
can be on the scale of chapters or paragraphs, and share many of the same features. They consist of a se-
quence of statements that by themselves express a meaningful computation. They can be merged to form
larger programs by specifying what they do and how they interface to other packages of software. The
analogy in prose is agreeing on the character’s names and what events are to happen in each paragraph
so that events happen to the right people in the right sequence once the whole is formed. Modules can be
re-used in two ways. As with our program from the 1960s, we would “lift” the matrix inversion routine
and put a different user interface around it. We can also re-use a routine within a program several times.
For example, solving the equations of space flight involves the inversion of many matrices. We would

c 2001 J.E. Akin 2
want our program to use the matrix inversion routine over and over, presenting it with a different matrix
each time.
The fundamental components of good program design are

1. Problem definition, leading to a program specification

2. Modular program design, which refines the specification

3. Module composition, which translates specification into executable program

4. Module/program evaluation and testing, during which you refine the program and find errors

5. Program documentation, which pervades all other phases

The result of following these steps is an efficient, easy-to-use program that has a user’s guide (how does
someone else run your program) and internal documentation so that other programmers can decipher the
algorithm.
Today it is common in a university education to be required to learn at least one foreign language.
Global interactions in business, engineering, and government make such a skill valuable to one’s career.
So it is in programming. One often needs to be able to read two or three programming languages even
if you compose programs in only one language. It is common for different program modules, in different
languages, to be compiled separately and then brought together by a “linker” to form a single executable.
When something goes wrong in such a process it is usually helpful to have a reading knowledge of the
programming languages being used.
When composing to express ideas there are, at least, two different approaches to consider: poetry and
prose. Likewise, in employing programming languages to create software there are distinctly different
approaches available. The two most common ones are “procedural programming” and “object-oriented
programming.” The two approaches are conceptually sketched in Fig. 1.1. They differ in the way that the
software development and maintenance are planned and implemented. Procedures may use objects, and
objects usually use procedures, called methods. Usually the object-oriented code takes more planning
and is significantly larger, but it is generally accepted to be easier to maintain. Today when one can have
literally millions of users active for years or decades, maintenance considerations are very important.

1.2 Problem Definition

The problem the program is to solve must be well specified. The programmer must broadly frame the
program’s intent and context by answering several questions.

 What must the program accomplish?

From operating the space shuttle to inverting a small matrix, some thought must be given to how
the program will do what is needed. In technical terms, we need to define the algorithm employed
in small-scale programs. In particular, numeric programs need to consider well how calculations
are performed. For example, finding the roots of a general polynomial demands a numeric (non-
closed form) solution. The choice of algorithm is influenced by the variations in polynomial order
and the accuracy demanded.

 What inputs are required and in what forms?

Most programs interact with humans and/or other programs. This interaction needs to be clearly
specified as to what format the data will take and when the data need to be requested or arrive.

 What is the execution environment and what should be in the user interface?
Is the program a stand-alone program, calculating the quadratic formula for example, or do the
results need to be plotted? In the former case, simple user input is probably all that is needed, but
the programmer might want to write the program so that its key components could be used in other
programs. In the latter, the program probably needs to be written so that it meshes well with some
pre-written graphics environment.

c 2001 J.E. Akin 3
 AA
Generation n Generation n+1

AAA A
AAA
•••
A AAA
A AA
AAA
A
•••

Figure 1.1: Here, the game is played on an 8  8 square array, and the filled squares indicate the presence
of life. The arrows emanating from one cells radiate to its eight neighbors. The rules are applied to the
nth generation to yield the next. The row of three filled cells became a column of three, for example.
What is going to happen to this configuration the next generation?

 What are the required and optional outputs, and what are their formats (printed, magnetic, graph-
ical, audio)?
In many cases, output takes two forms: interactive and archival. Interactive output means that the
programs results must be provided to the user or to other programs. Data format must be defined
so that the user can quickly see or hear the programs results. Archival results need to be stored on
long-term media, such as disk, so that later interpretation of the file’s contents is easy (recall the
notion of being able to read tomorrow what is written today) and that the reading process is easy.
The answers to these questions help programmers organize their thoughts, and can lead to decisions
about programming language and operating environment. At this point in the programming process, the
programmer should know what the program is to do and for whom the program is written. We don’t yet
have a clear notion of how the program will accomplish these tasks; that comes down the road. This
approach to program organization and design is known as top-down design. Here, broad program goals
and context is defined first, with additional detail filled in as needed. This approach contrasts with bottom-
up design, where the detail is decided first, then merged into a functioning whole. For programming, top-
design makes more sense, but you as well as professional programmers are frequently lured into writing
code immediately, usually motivated by the desire to “get something running and figure out later how to
organize it all.” That approach is motivated by expediency, but usually winds up being more inefficient
than a more considered, top-down approach that takes longer to get off the ground, but with increased
likelihood of working more quickly. The result of defining the programming problem is a specification:
how is the program structured, what computations does it perform, and how should it interact with the
user.
An Extended Example: The Game of Life
To illustrate how to organize and write a simple program, let’s structure a program that plays The Game
of Life. Conway’s “Game of Life” was popularized in Martin Gardner’s Mathematical Games column in
the October 1970 and February 1971 issues of Scientific American. This game is an example of what is
known in computer science as cellular automata. An extensive description of the game can be found in
The Recursive Universe by William Poundstone (Oxford University Press, 1987).
The rules of the game are quite simple. Imagine a rectangular array of square cells that are either
empty (no living being present) or filled (a being lives there). As shown in Fig. 1.1, each cell has eight
neighboring cells. At each tick of the clock, a new generation of beings is produced according to how
many neighbors surround a given cell.
 If a cell is empty, fill it if three of its neighboring cells are filled; otherwise, leave it empty.
 If a cell is filled, it
dies of loneliness if it has zero or one neighbors,
continues to live if it has two or three neighbors,
dies of overcrowding if it has more than three neighbors.

c 2001 J.E. Akin 4
 The programming task is to allow the user to “play the game” by letting him or her define initial
configurations, start the program, which applies the rules and displays each generation, and stop the
game at any time the user wants, returning to the initialization stage so that a new configuration can be
tried. To understand the program task, we as programmers need to pose several questions, some of which
might be

 What computer(s) are preferred, and what kind of display facilities do they have?

 Is the size of the array arbitrary or fixed?

 Am I the only programmer?

No matter how these questions are answered, we start by forming the program’s basic outline. Here is
one way we might outline the program in a procedural fashion.

1. Allow the user to initialize the rectangular array or quit the program.

2. Start the calculation of the next generation.

(a) Apply game rules to the current array.

(b) Generate a new array.
(c) Display the array.
(d) Determine whether the user wants to stop or not.
i. If not, go back to 2a.
ii. If so, go to step 1

Note how the idea of reusing the portion of the program that applies game rules arises naturally. This
idea is peculiar to programming languages, having no counterpart in prose (It’s like being told at the end
of a chapter to reread it!). This kind of looping behavior also occurs when we go back and allow the user
to restart the program.

This kind of outline is a form of pseudocode: y A programming language-like expression of how

the program operates. Note that at this point, the programming process is language-independent. Thus
informal pseudocode allows us to determine the program’s broad structure. We have not yet resolved
the issue of how, or if, the array should be displayed: Should it be refreshed as soon as a generation
is calculated, or should we wait until a final state is reached or a step limit is exceeded? Furthermore,
if calculating each generation takes a fair amount of time, our candidate program organization will not
allow the user to stop the program until a generation’s calculations have been finished. Consequently, we
may, depending on the speed of the computer, want to limit the size of the array. A more detailed issue
is how to represent the array internally. These issues can be determined later; programmers frequently
make notes at this stage about how the program would behave with this structure. Informal pseudocode
should remain in the final program in the form of comments.
Writing a program’s outline is not a meaningless exercise. How the program will behave is deter-
mined at that point. An alternative would be to ask the user how many generations should be calculated,
then calculate all generations, and display the results as a movie, allowing the user to go backward, play
in slow motion, freeze-frame, etc. Our outline will not allow such visual fun. Thus, programmers usually
design several candidate program organizations, understand the consequences of each, and determine
which best meets the specifications.
y The use of the word “code” is interesting here. It means program as both a noun and a verb: From the earliest days of
programming, what the programmer produced was called code, and what he or she did was “code the algorithm.” The origin of
this word is somewhat mysterious. It may have arisen as an analogy to Morse code, which used a series of dots and dashes as an
alternative to the alphabet. This code is tedious to read, but ideal for telegraphic transmission. A program is an alternate form of an
algorithm better suited to computation.

c 2001 J.E. Akin 5
 Program
Main Control
Subprogram #1

Subprogram #2

Figure 1.2: Modular program organization relies on self-contained routines where the passage of data (or
messages) from one to the other is very well defined, and each routine’s (or objects) role in the program
becomes evident.

1.3 Modular Program Design

We now need to define what the routines are and how they are interwoven to archive the program’s goals.
(We will deepen this discussion to include objects and messages when we introduce object-oriented
formulations in Sec. 1.7.) What granularity how large should a routine be comes with programming
experience and depends somewhat on the language used to express it. A program typically begins with
a main segment that controls or directs the solution of the problem by dividing it into sub-tasks (see
Figure 1.2). Each of these may well be decomposed into other routines. This step-wise refinement
continues as long as necessary, as long as it benefits program clarity and/or efficiency. This modular
program design is the key feature of modern programming design practice. Furthermore, routines can be
tested individually, and replaced or rewritten as needed. Before actually writing each routine, a job known
in computer circles as the implementation, the program’s organization can be studied: Will the whole
satisfy design specifications? Will the program execute efficiently? As the implementation proceeds,
each routine’s interface is defined: How does it interact with its master the routine that called it and
how are data exchanged between the two? In some most languages, this interface can be prototyped:
The routine’s interface what it expects and what values it calculates can be defined and the whole
program merged together and compiled to check for consistency without performing any calculations. In
small programs, where you can have these routine definitions easily fitting onto one page, this prototyping
can almost be performed visually. In complex programs, where there may be hundreds or thousands of
routines, such prototyping really pays off. Once the interfaces begin to form, we ask whether they make
sense: Do they exchange information efficiently? Does each routine have the information it needs or
should the program be reorganized so that data exchange can be accomplished more efficiently?
From another viewpoint, you should develop a programming style that “hedges your bets:” Programs
should be written in such a way that allows their components to be used in a variety of contexts. Again,
using a modular programming style, the fundamental components of the calculation should be expressed
as a series of subroutines or functions, the interweaving of which is controlled by a main program that
reads the input information and produces the output. A modular program can have its components ex-
tracted, and used in other programs (program re-use) or interfaced to environments. So-called monolithic
programs, which tend not to use routines and express the calculation as a single, long-winded program,
should not be written.
We emphasize that this modular design process proceeds without actually writing program statements.
We use a programming-like language, known as formal pseudocode, to express in prose what routines call
others and how. This prose might re-express a graphic representation of program organization, such as
that shown in Figure 1.2. In addition, expressing the program’s design in pseudocode eases the transition
to program composition, the actual programming process. The components of formal pseudocode at this
point are few:

c 2001 J.E. Akin 6
[1] ! This is a comment line in Fortran 90
[2]
[3] program main ! a program called main
[4] ! begin the main program
[5] print *,"Hello, world" ! * means default format
[6] end program main ! end the main program
[1] // This is a comment line in C++
[2] #include <iostream.h> // standard input output library
[3]
[4] main () // a program called main
[5] // begin the main program
[6] cout << "Hello, world" << endl ; // endl means new line
[7] return 0; // needed by some compilers
[8] // end the main program
[1] % This is a comment line in MATLAB
[2]
[3] function main () % a program called main
[4] % begin the main program
[5] disp (’Hello, world’); % display the string
[6] % end the main program

Figure 1.3: ’Hello World’ Program and Comments in Three Languages

 comments that we allow to include the original outline and to describe computational details;

 functions that express each routine, whether it be computational or concerned with the user inter-
face;

 conditionals that express changing the flow of a program; and

 loops that express iteration.

Comments. A comment begins with a comment character, which in our pseudocode we take to be the
exclamation point !, and ends when the line ends. Comments can consume an entire line or the right
portion of some line.
! This is a comment: you can read it, but the computer won’t
statements
statement ! From the comment character to end of this line is a comment
statements

The statements cited in the above lines share the status of the sentence that characterizes most written
languages. It is made up of components specific to the syntax of the programming language in use. For
example, most programming books begin with a program that does nothing but print “Hello world” on
the screen (or other output device). The pseudocode for this might have the following form:
! if necessary, include the device library
initiate my program, say main
send the character string ‘‘Hello world’’ to the output device library
terminate my program

Figure 1.3 illustrates this in three common languages, beginning with F90. At this point one can now
say that they are multi-lingual in computer languages. Here, too, we may note that, unlike the other two
languages shown, in Fortran when we begin a specific type of software construct, we almost always ex-
plicitly declare where we are ending its scope. Here the construct pair was program and end program,
but the same style holds true for if and end if pairs, for example. All languages have rules and syntax
to terminate the scope of some construct, but when several types of different constructs occur in the same
program segment, it may be unclear in which order they are terminating.
Functions. To express a program’s organization through its component routines and routines, we use
the notation of mathematical functions. Each program routine accepts inputs, expressed as arguments of
a function, performs its calculations, and returns the computational results as functional values.
output 1 = routine (input 1,...,input m)

or

c 2001 J.E. Akin 7
 call routine (input 1,..., input m, output 1,..., output n)

In Fortran, a routine evaluating a single output object, as in the first style, is called a function and, oth-
erwise, it is called a subroutine. Other languages usually use the term function in both cases. Each
routines’s various inputs and results are represented by variables, which, in sharp contrast to mathemat-
ical variables, have text-like names that indicate what they contain. These names contain no spaces, but
may contain several words. There are two conventions for variable names containing two or more words:
either words are joined by the underbar character “ ” (like next generation) or each word begins
with an uppercase letter (like NextGeneration). The results of a routines’s computation are always
indicated by a sequence of variables on the left side of the equals sign =. The use of an equals sign does
not mean mathematical equality; it is a symbol in our pseudocode that means “assign a routines’s results
to the variables (in order) listed on the left.”
Conditionals. To create something other than a sequential execution of routines, conditionals form a
test on the values of one or more variables, and continue execution at one point or another depending
on whether the test was true or false. That is usually done with the if statement. It either performs the
instruction(s) that immediately follow (after the then keyword) if some condition is valid (like x > 0) or
those that follow the else statement if the condition is not true.
if test then
statement group A ! executed if true
else
statement group B ! executed if false
end if

The test here can be very complicated, but is always based on values of variables. Parentheses should be
used to clarify exactly what the test is. For example,
((x > 0) and (y = 2))

One special statement frequently found in if statements is stop: This command means to stop or abort
the program, usually with a fatal error message.
Conditionals allow the program to execute non-sequentially (the only mode allowed by statements).
Furthermore, program execution order can be data-dependent. In this way, how the program be-
haves what output it produces and how it computes the output depends on what data, or messages, it
is given. This means that exact statement execution order is determined by the data, and/or messages, and
the programmer not just the programmer. It is this aspect of programming languages that distinguishes
them from written or spoken languages. An analogy might be that chapters in a novel are read in the
order specified by the reader’s birthday; what that order might be is determined by the novelist through
logical constructs. The tricky part is that in programming languages, each execution order must make
sense and not lead to inconsistencies or, at worst, errors: The novel must make sense in all the ways the
novelist allows. This data- and message-dependent execution order can be applied at all programming
levels, from routine execution to statements. Returning to our analogy to the novel, chapter (routine)
order and sentence (statement) order depend on the reader’s birthday. Such complexity in prose has little
utility, but does in programming. How else can a program be written that informs the user on what day
of the week and under what phase of the moon she was born given the birth date?
Loops. Looping constructs in our formal pseudocode take the form of do loops, where the keyword
do is paired with the key phrase end do to mean that the expressions and routine invocations contained
therein are calculated in order (from top to bottom), then calculated again starting with the first, then
again, then again, . . . , forever. The loop ceases only when we explicitly exit it with the exit command.
The pseudocode loop shown below on the left has the execution history shown on the right.
do
y = routine 1(x) y = routine 1(x)
z = routine 2(y) z = routine 2(y)
x = routine 3(z) x = routine 3(z) [let’s say x=-1]
if x > 0 then y = routine 1(x)
exit z = routine 2(y)
end if x = routine 3(z) [let’s say x=1]
end do [program ends]

c 2001 J.E. Akin 8
 Loop Pseudocode
Indexed loop do index=b,i,e
statements
end do

Pre-test loop while (test)

statements
end while

Post-test loop do
statements
if test exit
end do

Table 1.1: Pseudocode loop constructs

Infinite loops occur when the Boolean expression always evaluates to true; these are usually not what
the programmer intended and represent one type of program error a “bug.”y The constructs enclosed
by the loop can be anything: statements, logical constructs, and other loops! Because of this variety,
programs can exhibit extremely complex behaviors. How a program behaves depends entirely on the
programmer and how their definition of the program flows based on user-supplied data and messages.
The pseudocode loops are defined in Table 1.1.

1.4 Program Composition

Composing a program is the process of expressing or translating the program design into computer lan-
guage(s) selected for the task. Whereas the program design can often be expressed as a broad outline,
each routine’s algorithm must be expressed in complete detail. This writing process elaborates the formal
pseudocode and contains more explicit statements that more greatly resemble generic program state-
ments.
Generic programming language elements fall into five basic categories: the four we had be-
fore comments, loops, conditionals, and functions and statements. We will expand the variety of
comments, conditionals, loops, and functions/subroutines, which define routines and their interfaces.
The new element is the statement, the workhorse of programming. It is the statement that actually per-
forms a concrete computation. In addition to expanding the repertoire of programming constructs for
formal pseudocode, we also introduce what these constructs are in M ATLAB, Fortran, and C++. As we
shall see, formal pseudocode parallels these languages; the translation from pseudocode to executable
program is generally easy.

1.4.1 Comments
Comments need no further elaboration for pseudocode. However, programmers are encouraged to make
heavy use of comments.

1.4.2 Statements
Calculation is expressed by statements, which share the structure (and the status) of the sentence that
characterizes virtually all written language. Statements that are always executed one after the other as
written. A statement in most languages has a simple, well-defined structure common to them all.
variable = expression
y This term was originated by Grace Hopper, one of the first programmers. In the early days of computers, they were partially
built with mechanical devices known as relays. A relay is a mechanical switch that controls which way electric current flows:
The realization of the logical construct in programming languages. One day, a previously working program stopped being so.
Investigation revealed that an insect had crawled into the computer and had become lodged in a relay’s contacts. She then coined
the term “bug” to refer not only to such hardware failures, but to software ones as well since the user becomes upset no matter
which occurs.

c 2001 J.E. Akin 9
Statements are intended to bear a great resemblance to mathematical equations. This analogy with math-
ematics can appear confusing to the first-time programmer. For example, the statement a = a+1, which
means “increment the variable a by one” makes perfect sense as a programming statement, but no sense
as an algebraic equality since it seems to say that 0 = 1. Once you become more fluent in programming
languages, what is mathematics and what is programming become easily apparent. Statements are said to
be terminated when a certain character is encountered by the interpreter or the compiler. In Fortran, the
termination character is a carriage return or a semicolon (;). In C++, all statements must be terminated
with a semicolon or a comma; carriage returns do not terminate statements. M ATLAB statements may
end with a semicolon ‘;’ to suppress display of the calculated expression’s value. Most statements in
M ATLAB programs end thusly.
Sometimes, statements become quite long, becoming unreadable. Two solutions to improve clarity
can be used: decompose the expression into simpler expressions or use continuation markers to allow
the statement to span more than one line of text. The first solution requires you to use intermediate vari-
ables, which only results in program clutter. Multiline statements can be broken at convenient arithmetic
operators, and this approach is generally preferred. C++ has no continuation character; statements can
span multiple text lines, and end only when the semicolon is encountered. In M ATLAB, the continuation
character sequence comprise three periods ‘...’ placed at the end of each text line (before the carriage
return or comment character). In Fortran, a statement is continued to the next line when an ampersand &
is the last character on the line.
Variables. A variable is a named sequence of memory locations to which values can be assigned. As
such, every variable has an address in memory, which most languages conceal from the programmer so as
to present the programmer with a storage model independent of the architecture of the computer running
the program. Program variables correspond roughly to mathematical variables that can be integer, real,
or, complex-valued. Program variables can be more general than this, being able in some languages
to have values equal to a user-defined data type or object which, in turn, contains sequences of other
variables. Variables in all languages have names: a sequence of alphanumeric characters that cannot
begin with a number. Thus, a, A, a2, and a9b are feasible variable names (i.e., the interpreter or compiler
will not complain about these) while 3d is not. Since programs are meant to be read by humans as well as
interpreters and compilers, such names may not lead to program clarity even if they are carefully defined
and documented. The compiler/interpreter does not care whether humans can read a program easily or
not, but you should: Use variable names that express what the variables represent. For example, use
force as a name rather than f; use i, j, and k for indices rather than ii or i1.
In most languages, variables have type: the kind of quantity stored in them. Frequently occurring
data types are integer and floating point, for example. Integer variables would be chosen if the variable
were only used as an array index; floating point if the variable might have a fractional part.
In addition to having a name, type, and address, each variable has a value of the proper type. The
value should be assigned before the variable is used elsewhere. Compilers should indicate an error if a
variable is used before it has been assigned a value. Some languages allow variables to have aliases which
are usually referred to as “pointers” or “references”. Most higher level languages also allow programmers
to create “user defined” data types.
Assignment Operator. The symbol = in a statement means assignment of the expression into the vari-
able provided on the left. This symbol does not mean algebraic equality; it means that once expression
is computed, its value is stored in the variable. Thus, statements that make programming sense, like
a=a+1, make no mathematical sense because ‘=’ means different things in the two contexts. Fortran
90, and other languages, allow the user to extend the meaning of the assignment symbol (=) to other
operations. Such advanced features are referred to as “operator overloading”.
Expressions. Just as in mathematics, expressions in programming languages can have a complicated
structure. Most encountered in engineering programs amount to a mathematical expression involving
variables, numbers, and functions of variables and/or numbers. For example the following are all valid
statements.
A = B
x = sin(2*z)
force = G*mass1*mass2/(r*r)

c 2001 J.E. Akin 10
Thus, mathematical expressions obey the usual mathematical conventions, but with one added complex-
ity: Vertical position cannot be used help express what the calculation is; program expressions have only
one dimension. For example, the notation ab c clearly expresses to you how to perform the calculation.
However, the one-dimensional equivalent, obtained by smashing this expression onto one line, becomes
ambiguous: does a=bc mean divide a by b then multiply by c, or divide a by the product of b and c?
This ambiguity is relieved in program expressions in two ways. The first, the human-oriented way, de-
mands the use of parentheses grouping constructs to clarify what is being meant, as in (a=b)c . The
language-oriented way makes use of precedence rules: What an expression means is inferred from a set
of rules that specify what operations take effect first. In our example, because division is stronger than
multiplication, a=bc means (a=b)c. Most people find that frequent reliance on precedence rules leads to
programs that take a long time to decipher; the compiler/interpreter is “happy” either way.
Expressions make use of the common arithmetic and relational operators. They may also involve
function evaluations; the sin function was called in the second expression given in the previous example.
Programming expressions can be as complicated as the arithmetic or Boolean-algebra ones they emulate.
1.4.3 Flow Control
If a program consisted of a series of statements, statements would be executed one after the other, in the
order they were written. Such is the structure of all prose, where the equivalent of a statement is the
sentence. Programming languages differ markedly from prose in that statements can be meaningfully
executed over and over, with details of each execution differing each time (the value of some variable
might be changed), or some statements skipped, with statement ordering dependent on which statements
were executed previously or upon external events (the user clicked the mouse). With this extra variability,
programming languages can be more difficult for the human to trace program execution than the effort
it takes to read a novel. In written languages, sentences can be incredibly complex, much more so
than program statements; in programming, the sequencing of statements program flow can be more
complex.
The basic flow control constructs present in virtually all programming languages are loops
repetitive execution of a series of statements and conditionals diversions around statements.
Loops. Historically, the loop has been a major tool in designing the flow control of a procedure and one
would often code a loop segment without giving it a second thought. Today massively parallel computers
are being widely used and one must learn to avoid coding explicit loops in order to take advantage of
the power of such machines. Later we will review which intrinsic tools are included in F90 for use on
parallel (and serial) computers to offer improved efficiency over explicit loops.
The loop allows the programmer to repeat a series of statements, with a parameter the loop vari-
able taking on a different value for each repetition. The loop variable can be an integer or a floating-
point number. Loops can be used to control iterative algorithms, such as the Newton-Raphson algorithm
for finding solutions to nonlinear equations, to accumulate results for a sequential calculation, or to
merely repeat a program phrase, such as awaiting for the next typed input. Loops are controlled by a
logical expression, which when evaluated to true allows the loop another iteration and when false
terminates the loop and commences program execution with the statement immediately following those
statements enclosed within the loop.
There are three basic kinds of looping constructs, the choice of which is determined by the kind
of iterative behavior most appropriate to the computation. The indexed loop occurs most frequently in
programs. Here, one loop variable varies across a range of values. In pseudocode, the index’s value
begins at b, increments each time through the loop by i, and the loop ends when the index exceeds e.
For example:
do j = b, e, i

or using the default increment of unity:

do j = b, e

As an example of an indexed loop, let’s explore summing the series of numbers stored in the array A.
If we knew the number of elements in the array when we write the program, the sum can be calculated

c 2001 J.E. Akin 11
explicitly without using a loop.
sum = A1 + A2 + A3 + A4
However, we have already said that our statements must be on a single line, so we need a way to repre-
sent the subscript attached to each number. We develop the convention that a subscript is placed inside
parentheses like
sum = A(1) + A(2) + A(3) + A(4)

Such programs are very inflexible, and this hard-wired programming style is discouraged. For example,
suppose in another problem the array contains 1,000 elements. With an indexed loop, a more flexible,
easier to read program can be obtained. Here, the index assumes a succession of values, its value tested
against the termination value before the enclosed statements are executed, with the loop terminating once
this test fails to be true. The following generic indexed loop also sums array elements, but in a much
more flexible, concise way.
sum = 0
for i = 1,n
sum = sum + A(i)
end for

Here, the variable n does not need to be known when the program is written; this value can wait until the
program executes, and can be established by the user or after data is read.
In F90 the extensive
PN support for matrix expressions allows implicit loops. For example, consider
the calculation of i=1 xi yi . The language provides at least three ways of performing this calculation.
Assuming the vectors x and y are column vectors,
1. sum xy = 0
N = size(x)
do i = 1,N
sum xy = sum xy + x(i)*y(i)
end do

2. sum xy = sum(x*y)
3. sum xy = dot product(x,y)

The first method is based on the basic loop construct, and yields the slowest running program of the
three versions. In fact, avoiding the do statement by using implicit loops will almost always lead to faster
running programs. The second, and third statements employ intrinsic functions and/or operators designed
for arrays. In many circumstances, calculation efficiency and clarity of expression must be balanced. In
practice, it is usually necessary to set aside memory to hold subscripted arrays, such as x and y above,
before they can be referenced or used.
Conditionals. Conditionals test the veracity of logical expressions, and execute blocks of statements
accordingly (see Table 1.2). The most basic operation occurs when we want to execute a series of state-
ments when a logical expression, say test, evaluates to true. We call that a simple if conditional; the
beginning and end of the statements to be executed when test evaluates to true are enclosed by special
delimiters, which differ according to language. When only one statement is needed, C++ and Fortran
allow that one statement to end the line that begins with the if conditional. When you want one group
of statements to be executed when test is true and another set to be executed when false, you use
the if-else construct. When you want to test a series of logical expressions that are not necessarily com-
plementary, the nested-if construct allows for essentially arbitrarily complex structure to be defined. In
such cases, the logical tests can interlock, thereby creating programs that are quite difficult to read. Here
is where program comments become essential. For example, suppose you want to sum only the positive
numbers less than or equal to 10 in a given sequence. Let’s assume the entire sequence is stored in array
A. In informal pseudocode, we might write
loop across A
if A(i) > 0 and A(i) < = 10 add to sum
end of loop

More formally, this program fragment as a complete pseudocode would be

sum = 0

c 2001 J.E. Akin 12
 Conditional Pseudocode
if if (test) statement

if if test then
statements
end if

if-else if test then

statements A
else
statements B
end if

nested if if test1 then

statements A
if test2 then
statements B
end if % end of test2
end if

Table 1.2: Syntax of pseudocode conditionals

do i=1,n
if (A(i) > 0) & (A(i) <= 10)
sum = sum + A(i)
end if
end do

Several points are illustrated by this pseudocode example. First of all, the statements that can be included
with a loop can be arbitrary, comprised of simple statements, loops, and conditionals in any order. This
same generality applies to statements within a conditional as well. Secondly, logical expressions can
themselves be quite complicated. Finally, note how each level of statements in the program is indented,
visually indicated the subordination of statements within higher level loops or conditionals. This stylistic
practice lies at the heart of structured programming: explicit indication of each statement within the sur-
rounding hierarchy. In modern programming, the structured approach has become the standard because
it leads to greater clarity of expression, allowing others to understand the program more quickly and the
programmer to find bugs more readily. Employing this style only requires the programmer to use the
space key liberally when typing the program. Since sums are computed so often you might expect that a
language would provide an intrinsic function to compute it. For F90 and M ATLAB you would be correct.
1.4.4 Functions
Functions, which define sub-programs having a well-defined interface to other parts of the program, are
an essential part of programming languages. For, if properly developed, these functions can be included
in future programs, and they allow several programmers to work on complex programs. The function
takes an ordered sequence of messages, objects, or variables as its arguments, and returns to the calling
program a value (or set of values) that can be assigned to an object or variable. Familiar examples of
a function are the mathematical ones: the sin function takes a real-valued argument, uses this value to
calculate the trigonometric sine, and returns that value, which can be assigned to a variable.
y = sin(x)

Note that the argument need not be a variable: a number can be explicitly provided or an expression
can be used. Thus, sin(2.3) and sin(2*cos(x)) are all valid. Functions may require more than one
argument. For example, the atan2 function, which computes the arctangent function in such a way that
the quadrant of the calculated angle is unambiguous, needs the x and y components of the triangle.
z = atan2(x, y)

Note that the order of the arguments the x component must be the first and the number of argu-
ments both x and y are needed matter for all functions: The calling program’s argument ordering

c 2001 J.E. Akin 13
 A
AAA
A
L
clip ( x, L)

AAAAAA
–L x
L

–L

Figure 1.4: Input-output relationship for the function clip(x). So long as jxj < L, this function equals
its argument; for larger values, the output equals the clipping constant L no matter how large the input
might be.

and number must agree with those imposed by the function’s definition. Said another way, the inter-
face between the two must agree. Analogous to plugs and electric sockets in the home, a three prong
plug won’t fit into a two-hole socket, and, if you have a two-prong plug, you must plug it in the right
way. A function is usually defined separately, outside the body of any program or other function. We
call a program’s extent its scope. In M ATLAB, a program’s scope is equivalent to what is in a file; in
C and C++, scope is defined by brace pairs; and in Fortran, scope equals what occurs between function
declaration and its corresponding end statement. Variables are also defined within a program’s and a
function’s scope. What this means is that a variable named x defined within a function is available to all
statements occurring within that function, and different functions can use the same variable name without
any conflict occurring. What this means is that two functions f1 and f2 can each make use of a variable
named x, and the value of x depends on which function is being referred to. In technical terms, the scope
of every variable is limited to its defining function. At first, this situation may seem terribly confusing
(“There are two variables both of which are named x?”); further thought brings the realization that this
convention is what you want. Because each function is to be a routine a program having a well-defined
interface, execution of the function’s internal statements must not depend on the program that uses it.
This convention becomes especially important when different people write the programs or functions.
Thus, such local variables those defined locally within a function do not conflict, and they are stored
in different memory locations by the compiler or interpreter.
This limited scope convention can be countermanded when you explicitly declare variables to be
global. Such variables are now potentially available to all functions, and each function cannot define a
variable having the same name. For example, you may well want a variable pointedly named pi to be
available to all functions; you can do so by declaring it to be a global variable. To demonstrate scope,
consider the following simple example. Here, we want to clip the values stored in the array x and store
the results in the array y.

Main Pseudocode Program Function Pseudocode Definition

! Clip the elements of an array ! function clip(x, edge)
limit = 3 ! x - input variable
do i=1,n ! edge - location of breakpoint
y(i) = clip(x(i), limit) function clip(x, edge)
end do if abs(x) > edge then
y = sign(x)*edge
else
y = x
end if
end

The clipping function has the generic form show in Figure 1.4. Thus, values of the argument that are
less than L in magnitude are not changed, while those exceeding this limit are set equal to the limiting
value. In the program example, note that the name of the array in the calling program x is the same
as the argument’s name used in the definition of the function. Within the scope of a program or function,
an array and a scalar variable cannot have the same name. In our case, because each variable’s scope is
limited to the function or program definition, no conflict occurs: Each is well defined and the meaning
should be unambiguous. Also note that the second argument has a different name in the program that in
the function. No matter how the arguments are defined, we say that they are passed to the function, with

c 2001 J.E. Akin 14
the function’s variables set equal to values specified in the calling program. These interface rules allows
the function to be used in other programs, which means that we can reuse functions whenever we like!
1.4.5 Modules
Another important programming concept is that of packaging a group of related routines and/or selective
variables into a larger programming entity. In the Ada language they are called packages, while C++
and M ATLAB call them classes. F90 has a generalization of this concept that it calls a module. As we
will see later the F90 module includes the functionality of the C++ classes, as well as other uses such as
defining global constants. Therefore, we will find the use of F90 modules critical to our object-oriented
programming goals. In that context modules provide us with the means to take several routines related
to a specific data type and to encapsulate them into a larger programming unit that has the potential to be
reused for more than one application.
1.4.6 Dynamic Memory Management
From the very beginning, several decades ago, there was a clear need to be able to dynamically allocate
and deallocate segments of memory for use by a program. The initial standards for Fortran did not allow
for this. It was necessary to invoke machine language programs to accomplish that task or to write tools
to directly manage arrays by defining “pseudo-pointers” to manually move things around in memory or to
overwrite space that was no longer needed. It was very disappointing that the F77 standard failed to offer
that ability, although several “non-standard” compilers offered such an option. Beginning with the F90
standard a full set of dynamic memory management abilities is now available within Fortran. Dynamic
memory management is mainly needed for arrays and pointers. Both of these will be considered late,
with a whole chapter devoted to arrays. Both of these entities can be declared as ALLOCATABLE and
later one will ALLOCATE and then DEALLOCATE them. There are also new “automatic arrays” that
have the necessary memory space supplied and then freed as needed.
Pointers are often used in “data structures”, abstract data types, and objects. To check on the status
of such features one can invoke the ALLOCATED intrinsic and use ASSOCIATED to check on the
status of pointers and apply NULLIFY to pointers that need to be freed or initialized. Within F90
allocatable arrays cannot be used in the definitions of derived types, abstract data types, or objects.
However, allocatable pointers to arrays can be used in such definitions. To assist in creating efficient
executable codes, entities that might be pointed at by a pointer must have the TARGET attribute.
Numerous examples of dynamic memory management will be seen later. Persons that write compilers
suggest that, in any language, it is wise to deallocate dynamic memory in the reverse order of its creation.
The F90 language standard does not require that procedure but you see that advice followed in most of
the examples.

1.5 Program evaluation and testing

Your fully commented program, written with the aid of an editor, must now come alive and be trans-
lated into another language that more closely matches computer instructions; it must be executed or run.
Statements expressed in M ATLAB, Fortran, or C++ may not directly correspond to computational instruc-
tions. However, the Fortran syntax was designed to more clearly match mathematical expressions. These
languages are designed to allow humans to define computations easily and also allow easy translation.
Writing programs in these languages provides some degree of portability: A program can be executed on
very different computers without modification. So-called assembly languages allow more direct expres-
sion of program execution, but are very computer specific. Programmers that write in assembly language
must worry about the exquisite details of computer organization, so much so that writing of what the
computation is doing takes much longer. What they produce might run more rapidly that the same com-
putation expressed in Fortran, for example, but no portability results and programs become incredibly
hard to debug.
Programs become executable machine instructions in two basic ways. They are either interpreted or
compiled. In the first case, an interpreter reads your program and translates it to executable instructions
“on the fly.” Because interpreters essentially examine programs on a line-by-line basis, they usually allow
instructions accept typed user instructions as well as fully written programs. M ATLAB is an example of

c 2001 J.E. Akin 15
an interpreter.y It can accept typed commands created as the user thinks of them (plot a graph, see that
a parameter must have been typed incorrectly, change it, and replot, for example) or entire programs.
Because interpreters examine programs locally (line-by-line), program execution tends to be slower than
when a compiler is used.
Compilers are programs that take your program in its entirety and produce an executable version of
it. Compiler output is known as an executable file that, in UNIX for example, can become a command
essentially indistinguishable from others that are available. C++ is an example of a language that is
frequently compiled rather than interpreted. Compilers will produce much more efficient (faster running)
programs than interpreters, but if you find an error, you must edit and re-compile before you can attempt
execution again. Because compilation takes time, this cycle can be time-consuming if the program is
large.
Interpreters are themselves executable files written in compiled languages: MATLAB is written in
C. Executable programs produced by compilers are stand-alone programs: Everything user input and
output, file reading, etc. must be handled by the user’s program. In an interpreter, you can supplement a
program’s execution by typed instructions. For example, in an interpreter you can type a simple command
to make the variable a equal to 1; in a compiled program, you must include a program that asks for the
value of a. Consequently, users frequently write programs in the context of an interpreter, understand
how to make the program better by interacting with it, and then re-express it in a compiled language.
Both interpreters and compilers make extensive use of what are known as library commands or func-
tions. A natural example of a library function is the sin function: Users typically do not want to program
explicitly the computation of the trigonometric sine function. Instead, they want to be able to pull it “off
the shelf” and use as need be. Library modules are just programs written in a computer language like
you would write. Consequently, both interpreters and compilers allow user programs to become part
of the library, which is usually written by many programmers over a long period of time. It is through
modules available in a library that programming teams cooperate. Library modules tend to be more ex-
tensive and do more things in an interpreter. For example, M ATLAB provides a program that produces
pseudo-three-dimensional plots of functions. Such routines usually do not come with a compiler, but
may be purchased separately from graphics programming specialists. For compiled languages, we refer
to linking the library routines to the user’s program (in interpreters, this happens as a matter of course). A
linker is a program that takes modules produced by the compiler, be they yours or others, associates the
modules, and produces the executable file we mentioned earlier. Most C++ compilers “hide” the linking
step from you; you may think you are typing just the command to compile the program, but it is actually
performing that step for you. When you are compiling a module not intended for stand-alone execution,
a compiler option that you type can prevent the compiler from performing the linking step.
Debugging is the process of discovering and removing program errors. Two main types of errors
occur in writing programs: what we would generally term “typos” and what are design errors. The first
kind may be readily found (where is the function sni?) or more subtle (you type aa instead of a for a
variable’s name and aa also exists!). The second kind of error can be hard or subtle to find. The main
components of this process are
1. Search the program module by eye as you do a “mental run through” of its task. This kind of error
searching begins when you first think about program organization, and continues as you refine the
program. Why write a program that is logically flawed?
2. If written in a compiled language, compile the program to find syntax errors or warnings about
unused or undefined variables. If in an interpreted language, attempt preliminary execution to
obtain similar error messages. Linking also can locate modules or libraries that are improperly
referenced.
3. Running the executable file with typical data sets often causes the program to abort a harsh
word that expresses the situation where the program goes crazy and ceases to behave and the
system to supply an error message, such as division by zero. Error messages may help locate the
programming error.
y This statement is only partially true. M ATLAB does have some features of a compiler, like looking ahead to determine if
interface errors exist with respect to functions called by the main program.

c 2001 J.E. Akin 16
Easy errors to find are syntactic errors: You have violated the language’s rules of what a well-formed
program must be. Usually, the interpreter or compiler complain bitterly on encountering a syntax error.
Compilers find these at compile time (when the program is compiled), interpreters at run time. Design
errors are only revealed when we supply the program with data. The programmer should design test data
that exercises each of the program’s valid operations. Invalid user input (asking for the logarithm of a
negative number, for example) should be caught by the program and warning messages sent to the user.
The previous description of generic programming languages indicates why finding bugs can be quite
complicated. Programs can exhibit quite complex behaviors, and tracing incorrect behaviors can be
correspondingly difficult. One often hears the (true) statement “Computers do what we say, not what
we want.” Users frequently want computers to be smart, fixing obvious design (mental) errors because
they obviously conflict with what we want. However, this situation is much like what the novelist faces.
Inexact meaning can confuse the reader; he or she does not have a direct pathway to the novelist’s mind.
As opposed to the novelist, extensive testing of your program can detect such errors and make your
program approach perfection. Many operating systems supply interactive debugger programs that can
trace the execution of a program in complete detail. They can display the values of any variable, stop
at selected positions for evaluation, execute parts of the code in a statement-by-statement fashion, etc.
These can be very helpful in finding difficult-to-locate bugs, but they still cannot read your mind.
Be that as it may, what can the programmer do when the program compiles (no syntactic errors),
doesn’t cause system error messages (no dividing by zero), but the results are not correct? The simplest
approach is to include extra statements in your program, referred to as debugging statements, that display
(somewhat verbosely) values of internal variables. For example, in a loop you would print the value of the
loop index and all variables that the loop might be affecting. Because this output can be voluminous, the
most fruitful approach is to debug smaller problems. With this debugging information, you can usually
figure out the error, correct it, and change the comments accordingly. Without the latter, your program
and your internal documentation are out-of-sync.
Once debugged, you could delete the debugging statements you added. A better approach is to just
hide them. You can do this two ways: Comment them out or encase them in a conditional that is true when
the program is in “debugging mode.” The commenting approach completely removes the debugging
statements from the program execution stream, and allows you to easily put them back if further program
elaborations result in errors. The use of conditionals does put an overhead on computational efficiency,
but usually a small one.

1.6 Program documentation

Comments inside a program are intended to help you and others understand program design and how
it is organized. Frequently, comments describe what each variable means, how program execution is to
proceed, and what each module’s interface might be (what are the expected inputs and their formats, and
what outputs are produced). Program comments occur in the midst of the program’s source, and tem-
porarily interrupts the highly restricted syntax of most programming languages. Comments are entirely
ignored by the interpreter or compiler, and are allowed to enhance program clarity for humans.
Documentation includes program comments, but also includes external prose that describes what the
program does, how the user interface controls program behavior, and how the display of results means.
Making an executable program available to users does not help them understand how to use it. In UNIX,
all provided commands are accompanied by what are referred to as manual pages: concise descriptions
of what the program does, all user options, and descriptions of what error messages means. Programs
are useless without such documentation. Many programs provide such documentation whenever the user
types something that clearly indicates a lack of knowledge about how to use the program. This kind
of documentation must also be supplemented by prose that a user can read. Professional programmers
frequently write the documentation as the program is being designed. This simultaneous development of
the program and documentation of how it is used often uncovers user interface design flaws.

c 2001 J.E. Akin 17
1.7 Object Oriented Formulations
The above discussion of subprograms follows the older programming style where the emphasis is placed
on the procedures that a subprogram is to apply to the supplied data. Thus, it is referred to as procedural
programming. The alternate approach focuses on the data and its supporting functions, and is known as
an object oriented approach and is the main emphasis of this work. It also generalizes the concept of
data types and is usually heavily dependent on user defined data types and their extension to abstract data
types. These concepts are sketched in Fig. 1.5.

data Procedure data

a) Procedural Based Programming

message Object data

b) Object-Oriented Programming

Figure 1.5: Two Approaches to Programming

The process or creating an “object-oriented” (OO) formulation involves at least three stages: Object-
Oriented Analysis (OOA), Object-Oriented Design (OOD), and Object-Oriented Programming (OOP).
Many books have been written on each of these three subjects. Formal graphical standards for represent-
ing the results of OOA and OOD have been established and are widely used in the literature. Here the
main emphasis will be placed on OOP on the assumption that the two earlier stages have been completed.
In an effort to give some level of completeness, summaries of OOA and OOD procedures are given in
Tables 1–1 and 1–2, respectively. Having completed OOA and OOD studies one must select a language
to actually implement the design. More than 100 objected-oriented languages are in existence and use
today. They include “pure” OO languages like Crisp, Eiffel, Rexx, Simula, Smalltalk, etc. and “hybrid”
OO languages like C++ , F90 , Object Pascal, etc. In which of them should you invest your time? To get
some insight into answers to this question we should study the advice of some of the recognized leaders
in the field. In his 1988 book on OO software construction B. Myers listed seven steps necessary to
achieve object-orientedness in an implementation language. They are summarized in Table 1-3 and are
all found to exist in F90 and F95 . Thus we proceed with F90 as our language of choice. The basic F90
procedures for OOP will be illustrated in some short examples in Chapter 3 after covering some prelim-
inary material on abstract data types in Chapter 2. Additional OOP applications will also be covered in
later chapters.

c 2001 J.E. Akin 18
 Table 1–1. OO Analysis Summary

Find objects and classes :

 Create an abstraction of the problem domain.
 Give attributes, behaviors, classes, and objects meaningful names.
 Identify structures pertinent to the system’s complexity and responsibilities.
 Observe information needed to interact with the system, as well as information to be stored.
 Look for information re-use; are there multiple structures; can sub-systems be inherited?

Define the attributes :

 Select meaningful names.
 Describe the attribute and any constraints.
 What knowledge does it possess or communicate?
 Put it in the type or class that best describes it.
 Select accessibility as public or private.
 Identify the default, lower and upper bounds.
 Identify the different states it may hold.
 Note items that can either be stored or re-computed.

Define the behavior :

 Give the behaviors meaningful names.
 What questions should each be able to answer?
 What services should it provide?
 Which attribute components should it access?
 Define its accessibility (public or private).
 Define its interface prototype.
 Define any input/output interfaces.
 Identify a constructor with error checking to supplement the intrinsic constructor.
 Identify a default constructor.

Diagram the system :

 Employ an OO graphical representation such as the Coad/Yourdon method or its extension by
Graham.

c 2001 J.E. Akin 19
 Table 1–2. OO Design Summary

 Improve and add to the OOA results during OOD.

 Divide the member functions into constructors, accessors, agents and servers.
 Design the human interaction components.
 Design the task management components.
 Design the data management components.
 Identify operators to be overloaded.
 Identify operators to be defined.
 Design the interface prototypes for member functions and for operators.
 Design code for re-use through “kind of” and “part of” hierarchies.
 Identify base classes from which other classes are derived.
 Establish the exception handling procedures for all possible errors.

Table 1–3. 7 Steps to Object-Orientedness (B. Myer, 1988)

1. Object-based modular structure :

 Systems are modularized on the basis of their data structure (in F90).
2. Data Abstraction :
 Objects should be described as implementations of abstract data types (in F90).
3. Automatic memory management :
 Unused objects should be deallocated by the language system (most in F90, in F95).
4. Classes :
 Every non-simple type is a module, and every high-level module is a type (in F90).

5. Inheritance :
 A class may be defined as an extension or restriction of another (in F90).
6. Polymorphism and dynamic binding :
 Entities are permitted to refer to objects of more than one class and operations can have different
realizations in different classes (partially in F90/F95, expected in Fortran 2000).
7. Multiple and repeated inheritance :

 Can declare a class as heir to more than one class, and more than once to the same class (in F90).

c 2001 J.E. Akin 20
1.8 Exercises
1 Checking trigonometric identities
We know that the sine and cosine functions obey the trigonometric identity sin2  + cos2  = 1
no matter what value of  is used. Write a pseudocode, or M ATLAB , or F90 program that checks
this identity. Let it consist of a loop that increments across N equally spaced angles between 0 and
 , and calculates the quantity in question, printing the angle and the result. Test your program for
several values of N . (Later we will write a second version of this program that does not contain
any analysis loops, using instead M ATLAB’s, or F90’s, ability to calculate functions of arrays.)
2 Newton-Raphson algorithm
A commonly used numerical method of solving the equation f (x) = 0 has its origins with the
beginnings of calculus. Newton noted that the slope of a function tended to cross the x-axis near a
function’s position of zero value (called a root).

f(x)
A
A
f(xi )

AA xi+1 xi x

Because the function’s slope at some point xi equals its
derivative f (xi ), the equation of the line
0

passing through f (xi ) is f 0 (xi )x + f (xi ) f 0 (xi )xi . Solving for the case when this expression
equals the next trial root xi+1 .
f (xi )
xi+1 = xi
f 0 (xi )
The algorithm proceeds by continually applying this iterative equation until the error is “small.”
The definition of “small” is usually taken to mean that the absolute relative difference between
successive iterates is less than some tolerance value . (Raphson extended these concepts to an
array of functions.)
(a) In pseudocode, write a program that performs the Newton-Raphson algorithm. Assume that
functions that evaluate the function and its derivative are available. What is the most conve-
nient form of loop to use in your program?
(b) Translate your pseudocode into F90, or M ATLAB, and apply your program to the simple
function f (x) = e2x 5x 1. Use the functional expressions directly in your program or
make use of functions.
3 Game of Life pseudocode
Develop a pseudocode outline for the main parts of the “Game of Life” which was discussed earlier
and shown in Fig. 1.3. Include pseudocode for a function to compute the next generation.

c 2001 J.E. Akin 21
Chapter 2

Data Types

Any computer program is going to have to operate on the available data. The valid data types that are
available will vary from one language to another. Here we will examine the intrinsic or built-in data types,
user-defined data types or structures and, finally, introduce the concept of the abstract data type which
is the basic foundation of object-oriented methods. We will also consider the precision associated with
numerical data types. The Fortran data types are listed in Table 2–1. Such data can be used as constants,
variables, pointers and targets.

Table 2–1. F90/95 Data Types and Pointer Attributes

Data Option
j j
j
Intrinsic Derived
[Components of intrinsic type and/or
j previously declared derived types.]
j
j j j
j
Character Logical Numeric

j j
j
Floating Point Integer
(Default Precision)
j Selected-Int-Kind
j
j j j
Complex Real Double Precision
(Default Precision) (Default Precision) [Obsolete]
Selected-Real-Kind’s Selected-Real-Kind

2.1 Intrinsic Types

The simplest data type is the LOGICAL type which has the Boolean values of either .true. or .false.
and is used for relational operations. The other non-numeric data type is the CHARACTER. The sets of valid
character values will be defined by the hardware system on which the compiler is installed. Character sets
may be available in multiple languages, such as English and Japanese. There are international standards
for computer character sets. The two most common ones are the English character sets defined in the
ASCII and EBCDIC standards that have been adapted by the International Standards Organization (ISO).
Both of these standards for defining single characters include the digits (0 to 9), the 26 upper case letters
(A to Z), the 26 lower case letters (a to z), common mathematical symbols, and many non-printable codes
known as control characters. We will see later that strings of characters are still referred to as being of
the CHARACTER type, but they have a length that is greater than one. In other languages such a data type
is often called a string. [While not part of the F95 standard, the ISO Committee created a user-defined
type known as the ISO VARIABLE LENGTH STRING which is available as a F95 source module.]

c 2001 J.E. Akin 23
 For numerical computations, numbers are represented as integers or decimal values known as floating
point numbers or floats. The former is called an INTEGER type. The decimal values supported in Fortran
are the REAL and COMPLEX types. The range and precision of these three types depends on the hardware
being employed. At the present, 1999, most computers have 32 bit processors, but some offer 64 bit
processors. This means that the precision of a calculated result from a single program could vary from
one brand of computer to another. One would like to have a portable precision control so as to get
the same answer from different hardware; whereas some languages, like C++, specify three ranges of
precision (with specific bit widths). Fortran provides default precision types as well as two functions to
allow the user to define the “kind” of precision desired.

Table 2–2. Numeric Types on 32 Bit Processors

Significant
Type Bit Width Digits Common Range
integer 16 10 –32,768 to 32,767
real 32 6 1037 to 1037
double precisiony 64 15 10307 to 10307
complex 2@32 2@6 two reals
yobsolete in F90, see selected real kind

Still, it is good programming practice to employ a precision that is of the default, double, or quad pre-
cision level. Table 2–2 lists the default precisions for 32 bit processors. The first three entries correspond
to types int, float, and double, respectively, of C++. Examples of F90 integer constants are
–32 0 4675123 24 short 24 long

while typical real constant examples are

–3. 0.123456 1.234567e+2 0.0 0.3 double
7.6543e+4 double 0.23567 quad 0.3d0

In both cases, we note that it is possible to impose a user-defined precision kind by appending an under-
score ( ) followed by the name of the integer variable that gives the precision kind number. For example,
one could define
long = selected int kind(9)

to denote an integer in the range of 109 to 109 , while

double = selected real kind(15,307)

defines a real with 15 significant digits with an exponent range of 307. Likewise, a higher precision
real might be defined by the integer kind
quad = selected real kind(18,4932)

to denote 18 significant digits over the exponent range of 4932. If these kinds of precision are available
on your processors, then the F90 types of “integer (long),” “real (double),” and “real (quad)” would
correspond to the C++ precision types of “long int,” “double,” and “long double,” respectively. If the
processor cannot produce the requested precision, then it returns a negative number as the integer kind
number. Thus, one should always check that the kind (i.e., the above integer values of long, double, or
quad) is not negative, and report an exception if it is negative.
The old F77 intrinsic type of DOUBLE PRECISION has been declared obsolete, since it is now easy
to set any level of precision available on a processor. Another way to always define a double precision
real on any processor is to use the “kind” function such as
double = kind(1.0d0)

where the symbol ‘d’ is used to denote the I/O of a double precision real. For completeness it should be
noted that it is possible on some processors to define different kinds of character types, such as “greek”
or “ascii”, but in that case, the kind value comes before the underscore and the character string such as:
ascii “a string”.

c 2001 J.E. Akin 24
[ 1] Module Math Constants ! Define double precision math constants
[ 2] implicit none
[ 3] ! INTEGER, PARAMETER :: DP = SELECTED REAL KIND (15,307)
[ 4] INTEGER, PARAMETER :: DP = KIND (1.d0) ! Alternate form
[ 5] real(DP), parameter:: Deg Per Rad = 57.295779513082320876798155 DP
[ 6] real(DP), parameter:: Rad Per Deg = 0.017453292519943295769237 DP
[ 7]
[ 8] real(DP), parameter:: e Value = 2.71828182845904523560287 DP
[ 9] real(DP), parameter:: e Recip = 0.3678794411714423215955238 DP
[10] real(DP), parameter:: e Squared = 7.389056098930650227230427 DP
[11] real(DP), parameter:: Log10 of e = 0.4342944819032518276511289 DP
[12]
[13] real(DP), parameter:: Euler = 0.5772156649015328606 DP
[14] real(DP), parameter:: Euler Log = -0.5495393129816448223 DP
[15] real(DP), parameter:: Gamma = 0.577215664901532860606512 DP
[16] real(DP), parameter:: Gamma Log = -0.549539312981644822337662 DP
[17] real(DP), parameter:: Golden Ratio = 1.618033988749894848 DP
[18]
[19] real(DP), parameter:: Ln 2 = 0.6931471805599453094172321 DP
[20] real(DP), parameter:: Ln 10 = 2.3025850929940456840179915 DP
[21] real(DP), parameter:: Log10 of 2 = 0.3010299956639811952137389 DP
[22]
[23] real(DP), parameter:: pi Value = 3.141592653589793238462643 DP
[24] real(DP), parameter:: pi Ln = 1.144729885849400174143427 DP
[25] real(DP), parameter:: pi Log10 = 0.4971498726941338543512683 DP
[26] real(DP), parameter:: pi Over 2 = 1.570796326794896619231322 DP
[27] real(DP), parameter:: pi Over 3 = 1.047197551196597746154214 DP
[28] real(DP), parameter:: pi Over 4 = 0.7853981633974483096156608 DP
[29] real(DP), parameter:: pi Recip = 0.3183098861837906715377675 DP
[30] real(DP), parameter:: pi Squared = 9.869604401089358618834491 DP
[31] real(DP), parameter:: pi Sq Root = 1.772453850905516027298167 DP
[32]
[33] real(DP), parameter:: Sq Root of 2 = 1.4142135623730950488 DP
[34] real(DP), parameter:: Sq Root of 3 = 1.7320508075688772935 DP
[35]
[36] End Module Math Constants
[37]
[38] Program Test
[39] use Math Constants ! Access all constants
[40] real :: pi ! Define local data type
[41] print *, ’pi Value: ’, pi Value ! Display a constant
[42] pi = pi Value; print *, ’pi = ’, pi ! Convert to lower precision
[43] End Program Test ! Running gives:
[44] ! pi Value: 3.1415926535897931 ! pi = 3.14159274

Figure 2.1: Defining Global Double Precision Constants

To illustrate the concept of a defined precision intrinsic data type, consider a program segment to
make available useful constants such as pi (3.1415: : :) or Avogadro’s number (6:02 : : :  1023 ). These
are real constants that should not be changed during the use of the program. In F90, an item of that nature
is known as a PARAMETER. In Fig. 2.1, a selected group of such constants have been declared to be of
double precision and stored in a MODULE named Math Constants. The parameters in that module can
be made available to any program one writes by including the statement “ use math constants” at the
beginning of the program segment. The figure actually ends with a short sample program that converts
the tabulated value of pi (line 23) to a default precision real (line 42) and prints both.

2.2 User Defined Data Types

While the above intrinsic data types have been successfully employed to solve a vast number of pro-
gramming requirements, it is logical to want to combine these types in some structured combination
that represents the way we think of a particular physical object or business process. For example, as-
sume we wish to think of a chemical element in terms of the combination of its standard symbol, atomic
number and atomic mass. We could create such a data structure type and assign it a name, say chemi-
cal element, so that we can refer to that type for other uses just like we might declare a real variable.
In F90 we would define the structure with a TYPE construct as shown below (in lines 3–7):
[ 1] program create a type
[ 2] implicit none
[ 3] type chemical element ! a user defined data type
[ 4] character (len=2) :: symbol
[ 5] integer :: atomic number
[ 6] real :: atomic mass

c 2001 J.E. Akin 25
 [ 7] end type

Having created the new data type, we would need ways to define its values and/or ways to refer to any of
its components. The latter is accomplished by using the component selection symbol “%”. Continuing
the above program segment we could write:
[ 8] type (chemical element) :: argon, carbon, neon ! elements
[ 9] type (chemical element) :: Periodic Table(109) ! an array
[10] real :: mass ! a scalar
[11]
[12] carbon%atomic mass = 12.010 ! set a component value
[13] carbon%atomic number = 6 ! set a component value
[14] carbon%symbol = "C" ! set a component value
[15]
[16] argon = chemical element ("Ar", 18, 26.98) ! construct element
[17]
[18] read *, neon ! get "Ne" 10 20.183
[19]
[20] Periodic Table( 5) = argon ! insert element into array
[21] Periodic Table(17) = carbon ! insert element into array
[22] Periodic Table(55) = neon ! insert element into array
[23]
[24] mass = Periodic Table(5) % atomic mass ! extract component
[25]
[26] print *, mass ! gives 26.9799995
[27] print *, neon ! gives Ne 10 20.1830006
[28] print *, Periodic Table(17) ! gives C 6 12.0100002
[29] end program create a type

In the above program segment, we have introduced some new concepts:

 define argon, carbon and neon to be of the chemical element type (line 7).
 define an array to contain 109 chemical element types (line 8).
 used the selector symbol, %, to assign a value to each of the components of the carbon structure
(line 15).
 used the intrinsic “structure constructor” to define the argon values (line 15). The intrinsic construct
or initializer function must have the same name as the user-defined type. It must be supplied with
all of the components, and they must occur in the order that they were defined in the TYPE block.

 read in all the neon components, in order (line 17). [The ‘*’ means that the system is expected
to automatically find the next character, integer and real, respectively, and to insert them into the
components of neon.]

 inserted argon, carbon and neon into their specific locations in the periodic table array (lines 19–
21).
 extracted the atomic mass of argon from the corresponding element in the periodic element
array (line 23).
 print the real variable, mass (line 25). [The ‘*’ means to use a default number of digits.]
 printed all components of neon (line 26). [Using a default number of digits.]
 printed all the components of carbon by citing its reference in the periodic table array (line 27).
[Note that the printed real value differs from the value assigned in line 12. This is due to the way
reals are represented in a computer, and will be considered elsewhere in the text.]
A defined type can also be used to define other data structures. This is but one small example of the
concept of code re-use. If we were developing a code that involved the history of chemistry, we might
use the above type to create a type called history as shown below.
type (chemical element) :: oxygen
type history ! a second type using the first
character (len=31) :: element name
integer :: year found
type (chemical element) :: chemistry

c 2001 J.E. Akin 26
 end type history
type (history) :: Joseph Priestley ! Discoverer
oxygen = chemical element ("O", 76, 190.2) ! construct element
Joseph Priestley = history ("Oxygen", 1774, oxygen) ! construct
print *, Joseph Priestley ! gives Oxygen 1774 O 76 1.9020000E+02

Shortly we will learn about other important aspects of user-defined types, such as how to define operators
that use them as operands.

2.3 Abstract Data Types

Clearly, data alone is of little value. We must also have the means to input and output the data, subpro-
grams to manipulate and query the data, and the ability to define operators for commonly used procedures.
The coupling or encapsulation of the data with a select group of functions that defines everything that can
be done with the data type introduces the concept of an abstract data type (ADT). An ADT goes a step
further in that it usually hides from the user the details of how functions accomplish their tasks. Only
knowledge of input and output interfaces to the functions are described in detail. Even the components
of the data types are kept private.
The word abstract in the term abstract data type is used to: 1) indicate that we are interested only
in the essential features of the data type, 2) to indicate that the features are defined in a manner that
is independent of any specific programming language, and 3) to indicate that the instances of the ADT
are being defined by their behavior, and that the actual implementation is secondary. An ADT is an
abstraction that describes a set of items in terms of a hidden or encapsulated data structure and a set of
operations on that data structure.
Previously we created user-defined entity types such as the chemical element. The primary dif-
ference between entity types and ADTs is that all ADTs include methods for operating on the type. While
entity types are defined by a name and a list of attributes, an ADT is described by its name, attributes,
encapsulated methods, and possibly encapsulated rules.
Object-oriented programming is primarily a data abstraction technique. The purpose of abstraction
and data hiding in programming is to separate behavior from implementation. For abstraction to work,
the implementation must be encapsulated so that no other programming module can depend on its imple-
mentation details. Such encapsulation guarantees that modules can be implemented and revised indepen-
dently. Hiding of the attributes and some or all of the methods of an ADT is also important in the process.
In F90 the PRIVATE statement is used to hide an attribute or a method; otherwise, both will default to
PUBLIC. Public methods can be used outside the program module that defines an ADT. We refer to the
set of public methods or operations belonging to an ADT as the public interface of the type.
The user-defined data type, as given above, in F90 is not an ADT even though each is created with
three intrinsic methods to construct a value, read a value, or print a value. Those methods cannot modify
a type; they can only instantiate the type by assigning it a value and display that value. (Unlike F90, in
C or C++ a user-defined type, or “struct”, does not have an intrinsic constructor method, or input/output
methods.) Generally ADTs will have methods that modify or query a type’s state or behavior.
From the above discussion we see that the intrinsic data types in any language (such as complex,
integer and real in F90 ) are actually ADTs. The system has hidden methods (operators) to assign them
values and to manipulate them. For example, we know that we can multiply any one of the numerical
types by any other numerical type.
We do not know how the system does the multiplication, and we don’t care. All computer languages
provide functions to manipulate the intrinsic data types. For example, in F90 a square root function,
named sqrt, is provided to compute the square root of a real or complex number. From basic mathematics
you probably know that two distinctly different algorithms must be used and the choice depends on the
type of the supplied argument. Thus, we call the sqrt function a generic function since its single name,
sqrt, is used to select related functions in a manner hidden from the user. In F90 you can not take the
square root of an integer; you must convert it to a real value and you receive back a real answer. The

c 2001 J.E. Akin 27
 ADT name

Public attributes

Public ADT with private attributes

Private attributes

Public members

Private members

Component Type Name

Send Send
Member Name
Message Type

Receive Receive
Member Name
Message Type

Receive, Modified
Member Name
Send Type

Figure 2.2: Graphical Representation of ADTs

above discussions of the methods (routines) that are coupled to a data type and describe what you can
and can not do with the data type should give the programmer good insight into what must be done to
plan and implement the functions needed to yield a relatively complete ADT.

chemical_element ADT

character symbol
integer atomic_number
real atomic_mass

chemical_element chemical_element

Figure 2.3: Representation of the Public Chemical Element ADT

It is common to have a graphical representation of the ADTs and there are several different graphical
formats suggested in the literature. We will use the form shown in Fig. 2.4 where a rectangular box begins
with the ADT name and is followed by two partitions of that box that represent the lists of attribute data
and associated member routines. Items that are available to the outside world are in sub-boxes that cross
over the right border of the ADT box. They are the parts of the public interface to the ADT. Likewise
those items that are strictly internal, or private, are contained fully within their respective partitions of
the ADT box. There is a common special case where the name of the data type itself is available for
external use, but its individual attribute components are not. In that case the right edge of the private
attributes lists lie on the right edge of the ADT box. In addition, we will often segment the smallest box
for an item to give its type (or the most important type for members) and the name of the item. Public

c 2001 J.E. Akin 28
member boxes are also supplemented with an arrow to indicate which take in information (<--), or send
out information (-->). Such a graphical representation of the previous chemical element ADT, with
all its items public, is shown in Fig. 2.4.
The sequence of numbers known as Fibonacci numbers is the set that begins with one and two and
where the next number in the set is the sum of the two previous numbers (1, 2, 3, 5, 8, 13, ...). A primarily
private ADT to print a list of Fibonacci numbers up to some limit is represented graphically in Fig. 2.5.

Figure 2.4: Representation of a Fibonacci Number ADT

2.4 Classes
A class is basically the extension of an ADT by providing additional member routines to serve as con-
structors. Usually those additional members should include a default constructor which has no argu-
ments. Its purpose is to assure that the class is created with acceptable default values assigned to all
its data attributes. If the data attributes involve the storage of large amounts of data (memory) then one
usually also provides a destructor member to free up the associated memory when it is no longer needed.
F95 has an automatic deallocation feature which is not present in F90 and thus we will often formally
deallocate memory associated with data attributes of classes.
As a short example we will consider an extension of the above Fibonacci Number ADT. The ADT
for Fibonacci numbers simply keeps up with three numbers (low, high, and limit). Its intrinsic ini-
tializer has the (default) name Fibonacci. We generalize that ADT to a class by adding a constructor
named new Fibonacci number. The constructor accepts a single number that indicates how many
values in the infinite list we wish to see. It is also a default constructor because if we omit the one
optional argument it will list a minimum number of terms set in the constructor. The graphical repre-
sentation of the Fibonacci Number class extends Fig. 2.4 for its ADT by at least adding one public
constructor, called new Fibonacci number, as shown in Fig. 2.5. Technically, it is generally accepted
that a constructor should only be able to construct a specific object once. This differs from the intrin-
sic initializer which could be invoked multiple times to assign different values to a single user-defined
type. Thus, an additional logical attribute has been added to the previous ADT to allow the constructor,
new Fibonacci number, to verify that it is being invoked only once for each instance of the class. The
coding for this simple class is illustrated in Fig. 2.6. There the access restrictions are given on lines 4, 5,
and 7 while the attributes are declared on line 8 and the member functions are given in lines 13-33. The
validation program is in lines 36–42, with the results shown as comments at the end (lines 44–48).

c 2001 J.E. Akin 29
 Figure 2.5: Representation of a Fibonacci Number Class

[ 1] ! Fortran 90 OOP to print list of Fibonacci Numbers

[ 2] Module class Fibonacci Number ! file: Fibonacci Number.f90
[ 3] implicit none
[ 4] public :: Print ! member access
[ 5] private :: Add ! member access
[ 6] type Fibonacci Number ! attributes
[ 7] private
[ 8] integer :: low, high, limit ! state variables & access
[ 9] end type Fibonacci Number
[10]
[11] Contains ! member functionality
[12]
[13] function new Fibonacci Number (max) result (num) ! constructor
[14] implicit none
[15] integer, optional :: max
[16] type (Fibonacci Number) :: num
[17] num = Fibonacci Number (0, 1, 0) ! intrinsic
[18] if ( present(max) ) num = Fibonacci Number (0, 1, max) ! intrinsic
[19] num%exists = .true.
[20] end function new Fibonacci Number
[21]
[22] function Add (this) result (sum)
[23] implicit none
[24] type (Fibonacci Number), intent(in) :: this ! cannot modify
[25] integer :: sum
[26] sum = this%low + this%high ; end function add ! add components
[27]
[28] subroutine Print (num)
[29] implicit none
[30] type (Fibonacci Number), intent(inout) :: num ! will modify
[31] integer :: j, sum ! loops
[32] if ( num%limit < 0 ) return ! no data to print
[33] print *, ’M Fibonacci(M)’ ! header
[34] do j = 1, num%limit ! loop over range
[35] sum = Add(num) ; print *, j, sum ! sum and print
[36] num%low = num%high ; num%high = sum ! update
[37] end do ; end subroutine Print
[38] End Module class Fibonacci Number
[39]
[40] program Fibonacci !** The main Fibonacci program
[41] implicit none
[42] use class Fibonacci Number ! inherit variables and members
[43] integer, parameter :: end = 8 ! unchangeable
[44] type (Fibonacci Number) :: num
[45] num = new Fibonacci Number(end) ! manual constructor
[46] call Print (num) ! create and print list
[47] end program Fibonacci ! Running gives:
[48]
[49] ! M Fibonacci(M) ; ! M Fibonacci(M)
[50] ! 1 1 ; ! 5 8
[51] ! 2 2 ; ! 6 13
[52] ! 3 3 ; ! 7 21
[53] ! 4 5 ; ! 8 34

Figure 2.6: A Simple Fibonacci Class

c 2001 J.E. Akin 30
2.5 Exercises
1. Create a module of global constants of common a) physical constants, b) common units conversion
factors.
2. Teams in a Sports League compete in matches that result in a tie or a winning and loosing team.
When the result is not a tie the status of the teams is updated. The winner is declared better that the looser
and better than any team that was previously bettered by the loser. Specify this process by ADTs for
the League, Team, and Match. Include a logical member function is better than which expresses
whether a team is better than another.

c 2001 J.E. Akin 31
Chapter 3

Object Oriented Programming Concepts

3.1 Introduction
The use of Object Oriented (OO) design and Object Oriented Programming (OOP) are becoming in-
creasingly popular. Thus, it is useful to have an introductory understanding of OOP and some of the
programming features of OO languages. You can develop OO software in any high level language, like
C or Pascal. However, newer languages such as Ada, C++, and F90 have enhanced features that make
OOP much more natural, practical, and maintainable. C++ appeared before F90 and currently, is prob-
ably the most popular OOP language, yet F90 was clearly designed to have almost all of the abilities of
C++ . However, rather than study the new standards many authors simply refer to the two decades old
F77 standard and declare that Fortran can not be used for OOP. Here we will overcome that misinformed
point of view.
Modern OO languages provide the programmer with three capabilities that improve and simplify
the design of such programs: encapsulation, inheritance, and polymorphism (or generic functionality).
Related topics involve objects, classes, and data hiding. An object combines various classical data types
into a set that defines a new variable type, or structure. A class unifies the new entity types and supporting
data that represents its state with routines (functions and subroutines) that access and/or modify those
data. Every object created from a class, by providing the necessary data, is called an instance of the
class. In older languages like C and F77, the data and functions are separate entities. An OO language
provides a way to couple or encapsulate the data and its functions into a unified entity. This is a more
natural way to model real-world entities which have both data and functionality. The encapsulation is
done with a “module” block in F90, and with a “class” block in C++. This encapsulation also includes
a mechanism whereby some or all of the data and supporting routines can be hidden from the user. The
accessibility of the specifications and routines of a class is usually controlled by optional “public” and
“private” qualifiers. Data hiding allows one the means to protect information in one part of a program
from access, and especially from being changed in other parts of the program. In C++ the default is
that data and functions are “private” unless declared “public,” while F90 makes the opposite choice for
its default protection mode. In a F90 “module” it is the “contains” statement that, among other things,
couples the data, specifications, and operators before it to the functions and subroutines that follow it.
Class hierarchies can be visualized when we realize that we can employ one or more previously
defined classes (of data and functionality) to organize additional classes. Functionality programmed into
the earlier classes may not need to be re-coded to be usable in the later classes. This mechanism is called
inheritance. For example, if we have defined an Employee class, then a Manager class would
inherit all of the data and functionality of an employee. We would then only be required to add only
the totally new data and functions needed for a manager. We may also need a mechanism to re-define
specific Employee class functions that differ for a Manager class. By using the concept of a class
hierarchy, less programming effort is required to create the final enhanced program. In F90 the earlier
class is brought into the later class hierarchy by the “use” statement followed by the name of the “module”
statement block that defined the class.
Polymorphism allows different classes of objects that share some common functionality to be used in
code that requires only that common functionality. In other words, routines having the same generic name

c 2001 J.E. Akin 33
are interpreted differently depending on the class of the objects presented as arguments to the routines.
This is useful in class hierarchies where a small number of meaningful function names can be used to
manipulate different, but related object classes. The above concepts are those essential to object oriented
design and OOP. In the later sections we will demonstrate by example additional F90 implementations
of these concepts.

3.2 Encapsulation, Inheritance, and Polymorphism

We often need to use existing classes to define new classes. The two ways to do this are called composition
and inheritance. We will use both methods in a series of examples. Consider a geometry program
that uses two different classes: class Circle and class Rectangle, as represented graphically in
Figs. 3.1 and 3.2. and as partially implemented in F90 as shown in Fig. 3.3. Each class shown has the
data types and specifications to define the object and the functionality to compute their respective areas
(lines 3–22). The operator % is employed to select specific components of a defined type. Within the
geometry (main) program a single routine, compute area, is invoked (lines 38 and 44) to return the
area for any of the defined geometry classes. That is, a generic function name is used for all classes
of its arguments and it, in turn, branches to the corresponding functionality supplied with the argument
class. To accomplish this branching the geometry program first brings in the functionality of the desired
classes via a “use” statement for each class module (lines 25 and 26). Those “modules” are coupled to
the generic function by an “interface” block which has the generic function name compute area (lines
28, 29). There is included a “module procedure” list which gives one class routine name for each of the
classes of argument(s) that the generic function is designed to accept. The ability of a function to respond
differently when supplied with arguments that are objects of different types is called polymorphism.
In this example we have employed different names, rectangular area and circle area, in their
respective class modules, but that is not necessary. The “use” statement allows one to rename the class
routines and/or to bring in only selected members of the functionality.

Circle Class

real radius

real pi

Circle make_Circle
real Circle_Area

Circle Circle

Figure 3.1: Representation of a Circle Class

Another terminology used in OOP is that of constructors and destructors for objects. An intrinsic
constructor is a system function that is automatically invoked when an object is declared with all of its
possible components in the defined order (see lines 37 and 43). In C++, and F90 the intrinsic constructor
has the same name as the “type” of the object. One is illustrated in the statement
four sides = Rectangle (2.1,4.3)

where previously we declared

type (Rectangle) :: four sides

which, in turn, was coupled to the class Rectangle which had two components, base and height,
defined in that order, respectively. The intrinsic constructor in the example statement sets component

c 2001 J.E. Akin 34
 Figure 3.2: Representation of a Rectangle Class

[ 1] ! Areas of shapes of different classes, using different

[ 2] ! function names in each class
[ 3] module class Rectangle ! define the first object class
[ 4] implicit none
[ 5] type Rectangle
[ 6] real :: base, height ; end type Rectangle
[ 7] contains ! Computation of area for rectangles.
[ 8] function rectangle area ( r ) result ( area )
[ 9] type ( Rectangle ), intent(in) :: r
[10] real :: area
[11] area = r%base * r%height ; end function rectangle area
[12] end module class Rectangle
[13]
[14] module class Circle ! define the second object class
[15] real :: pi = 3.1415926535897931d0 ! a circle constant
[16] type Circle
[17] real :: radius ; end type Circle
[18] contains ! Computation of area for circles.
[19] function circle area ( c ) result ( area )
[20] type ( Circle ), intent(in) :: c
[21] real :: area
[22] area = pi * c%radius**2 ; end function circle area
[23] end module class Circle
[24]
[25] program geometry ! for both types in a single function
[26] use class Circle
[27] implicit none
[28] use class Rectangle
[29] ! Interface to generic routine to compute area for any type
[30] interface compute area
[31] module procedure rectangle area, circle area ; end interface
[32]
[33] ! Declare a set geometric objects.
[34] type ( Rectangle ) :: four sides
[35] type ( Circle ) :: two sides ! inside, outside
[36] real :: area = 0.0 ! the result
[37]
[38] ! Initialize a rectangle and compute its area.
[39] four sides = Rectangle ( 2.1, 4.3 ) ! implicit constructor
[40] area = compute area ( four sides ) ! generic function
[41] write ( 6,100 ) four sides, area ! implicit components list
[42] 100 format ("Area of ",f3.1," by ",f3.1," rectangle is ",f5.2)
[43]
[44] ! Initialize a circle and compute its area.
[45] two sides = Circle ( 5.4 ) ! implicit constructor
[46] area = compute area ( two sides ) ! generic function
[47] write ( 6,200 ) two sides, area
[48] 200 format ("Area of circle with ",f3.1," radius is ",f9.5 )
[49] end program geometry ! Running gives:
[50] ! Area of 2.1 by 4.3 rectangle is 9.03
[51] ! Area of circle with 5.4 radius is 91.60885

Figure 3.3: Multiple Geometric Shape Classes

base = 2.1 and component height = 4.3 for that instance, four sides, of the type Rectangle.
This intrinsic construction is possible because all the expected components of the type were supplied. If
all the components are not supplied, then the object cannot be constructed unless the functionality of the

c 2001 J.E. Akin 35
[ 1] function make Rectangle (bottom, side) result (name)
[ 2] ! Constructor for a Rectangle type
[ 3] implicit none
[ 4] real, optional, intent(in) :: bottom, side
[ 5] type (Rectangle) :: name
[ 6] name = Rectangle (1.,1.) ! default to unit square
[ 7] if ( present(bottom) ) then ! default to square
[ 8] name = Rectangle (bottom, bottom) ; end if
[ 9] if ( present(side) ) name = Rectangle (bottom, side) ! intrinsic
[10] end function make Rectangle
[11] . . .
[12] type ( Rectangle ) :: four sides, square, unit sq
[13] ! Test manual constructors
[14] four sides = make Rectangle (2.1,4.3) ! manual constructor, 1
[15] area = compute area ( four sides) ! generic function
[16] write ( 6,100 ) four sides, area
[17] ! Make a square
[18] square = make Rectangle (2.1) ! manual constructor, 2
[19] area = compute area ( square) ! generic function
[20] write ( 6,100 ) square, area
[21] ! "Default constructor", here a unit square
[22] unit sq = make Rectangle () ! manual constructor, 3
[23] area = compute area (unit sq) ! generic function
[24] write ( 6,100 ) unit sq, area
[25] . . .
[26] ! Running gives:
[27] ! Area of 2.1 by 4.3 rectangle is 9.03
[28] ! Area of 2.1 by 2.1 rectangle is 4.41
[29] ! Area of 1.0 by 1.0 rectangle is 1.00

Figure 3.4: A Manual Constructor for Rectangles

class is expanded by the programmer to accept a different number of arguments.

Assume that we want a special member of the Rectangle class, a square, to be constructed if the
height is omitted. That is, we would use height = base in that case. Or, we may want to construct a
unit square if both are omitted so that the constructor defaults to base = height = 1. Such a manual
constructor, named make Rectangle, is illustrated in Fig. 3.4 (see lines 5, 6). It illustrates some
additional features of F90. Note that the last two arguments were declared to have the additional type
attributes of “optional” (line 3), and that an associated logical function “present” is utilized (lines 6 and 8)
to determine if the calling program supplied the argument in question. That figure also shows the results
of the area computations for the corresponding variables “square” and “unit sq” defined if the manual
constructor is called with one or no optional arguments (line 5), respectively.
In the next section we will illustrate the concept of data hiding by using the private attribute. The
reader is warned that the intrinsic constructor can not be employed if any of its arguments have been
hidden. In that case a manual constructor must be provided to deal with any hidden components. Since
data hiding is so common it is probably best to plan on prividing a manual constructor.

3.2.1 Example Date, Person, and Student Classes

Before moving to some mathematical examples we will introduce the concept of data hiding and combine
a series of classes to illustrate composition and inheritancey . First, consider a simple class to define dates
and to print them in a pretty fashion, as shown in Figs. 3.5 and 3.6. While other modules will have
access to the Date class they will not be given access to the number of components it contains (3),
nor their names (month, day, year), nor their types (integers) because they are declared “private” in the
defining module (lines 5 and 6). The compiler will not allow external access to data and/or routines
declared as private. The module, class Date, is presented as a source “include” file in Fig. 3.6, and
in the future will be reference by the file name class Date.f90. Since we have chosen to hide all
the user defined components we must decide what functionality we will provide to the users, who may
have only executable access. The supporting documentation would have to name the public routines and
describe their arguments and return results. The default intrinsic constructor would be available only to
those that know full details about the components of the data type, and if those components are “public.”
y These examples mimic those given in Chapter 11 and 8 of the J.R. Hubbard book “Programming with C++,” McGraw-Hill,
1994, and usually use the same data for verification.

c 2001 J.E. Akin 36
 Date Class

integer month
integer day
integer year

Date Date_
Date Print_Date
Date Read_Date
Date Set_Date

Date Date

Figure 3.5: Graphical Representation of a Date Class

The intrinsic constructor, Date (lines 14 and 34), requires all the components be supplied, but it does
no error or consistency checks. My practice is to also define a “public constructor” whose name is the
same as the intrinsic constructor except for an appended underscore, that is, Date . Its sole purpose is to
do data checking and invoke the intrinsic constructor, Date. If the function Date (line 10) is declared
“public” it can be used outside the module class Date to invoke the intrinsic constructor, even if the
components of the data type being constructed are all “private.” In this example we have provided another
manual constructor to set a date, set Date (line 31), with a variable number of optional arguments. Also
supplied are two subroutines to read and print dates, read Date (line 27) and print Date (line 16),
respectively.
A sample main program that employs this class is given in Fig. 3.7, which contains sample outputs
as comments. This program uses the default constructor as well as all three programs in the public class
functionality. Note that the definition of the class was copied in via an “include” (line 1) statement and
activated with the “use” statement (line 4).
Now we will employ the class Date within a class Person which will use it to set the date of
birth (DOB) and date of death (DOD) in addition to the other Person components of name, national-
ity, and sex. As shown in Fig. 3.8, we have made all the type components “private,” but make all the
supporting functionality public, as represented graphically in Fig. 3.8. The functionality shown provides
a manual constructor, make Person, routines to set the DOB or DOD, and those for the printing of
most components. The source code for the new Person class is given in Fig. 3.9. Note that the manual
constructor (line 12) utilizes “optional” arguments and initializes all components in case they are not
supplied to the constructor. The Date public function from the class Date is “inherited” to initial-
ize the DOB and DOD (lines 18, 57, and 62). That function member from the previous module was
activated with the combination of the “include” and “use” statements. Of course, the include could have
been omitted if the compile statement included the path name to that source. A sample main program
for testing the class Person is in Fig. 3.10 along with comments containing its output. It utilizes the
constructors Date (line 7), Person (line10), and make Person (line 24).
Next, we want to use the previous two classes to define a class Student which adds something
else special to the general class Person. The student person will have additional “private” compo-
nents for an identification number, the expected date of matriculation (DOM), the total course credit hours
earned (credits), and the overall grade point average (GPA), as represented in Fig. 3.11. The source lines
for the type definition and selected public functionality are given in Fig. 3.12. There the constructors
are make Student (line 19) and Student (line 47). A testing main program with sample output is
illustrated in Fig. 3.13. Since there are various ways to utilize the various constructors three alternate
methods have been included as comments to indicate some of the programmers options. The first two
include statements (lines 1, 2) are actually redundant because the third include automatically brings
in those first two classes.

c 2001 J.E. Akin 37
[ 1] module class Date ! filename: class Date.f90
[ 2] implicit none
[ 3] public :: Date ! and everything not "private"
[ 4]
[ 5] type Date
[ 6] private
[ 7] integer :: month, day, year ; end type Date
[ 8]
[ 9] contains ! encapsulated functionality
[10]
[11] function Date (m, d, y) result (x) ! public constructor
[12] integer, intent(in) :: m, d, y ! month, day, year
[13] type (Date) :: x ! from intrinsic constructor
[14] if ( m < 1 .or. d < 1 ) stop ’Invalid components, Date ’
[15] x = Date (m, d, y) ; end function Date
[16]
[17] subroutine print Date (x) ! check and pretty print a date
[18] type (Date), intent(in) :: x
[19] character (len=*),parameter :: month Name(12) = &
[20] (/ "January ", "February ", "March ", "April ",&
[21] "May ", "June ", "July ", "August ",&
[22] "September", "October ", "November ", "December "/)
[23] if ( x%month < 1 .or. x%month > 12 ) print *, "Invalid month"
[24] if ( x%day < 1 .or. x%day > 31 ) print *, "Invalid day "
[25] print *, trim(month Name(x%month)),’ ’, x%day, ", ", x%year;
[26] end subroutine print Date
[27]
[28] subroutine read Date (x) ! read month, day, and year
[29] type (Date), intent(out) :: x ! into intrinsic constructor
[30] read *, x ; end subroutine read Date
[31]
[32] function set Date (m, d, y) result (x) ! manual constructor
[33] integer, optional, intent(in) :: m, d, y ! month, day, year
[34] type (Date) :: x
[35] x = Date (1,1,1997) ! default, (or use current date)
[36] if ( present(m) ) x%month = m ; if ( present(d) ) x%day = d
[37] if ( present(y) ) x%year = y ; end function set Date
[38]
[39] end module class Date

Figure 3.6: Defining a Date Class

[ 1] include ’class Date.f90’ ! see previous figure

[ 2] program main
[ 3] use class Date
[ 4] implicit none
[ 5] type (Date) :: today, peace
[ 6]
[ 7] ! peace = Date (11,11,1918) ! NOT allowed for private components
[ 8] peace = Date (11,11,1918) ! public constructor
[ 9] print *, "World War I ended on " ; call print Date (peace)
[10] peace = set Date (8, 14, 1945) ! optional constructor
[11] print *, "World War II ended on " ; call print Date (peace)
[12] print *, "Enter today as integer month, day, and year: "
[13] call read Date(today) ! create today’s date
[14]
[15] print *, "The date is "; call print Date (today)
[16] end program main ! Running produces:
[17] ! World War I ended on November 11, 1918
[18] ! World War II ended on August 14, 1945
[19] ! Enter today as integer month, day, and year: 7 10 1997
[20] ! The date is July 10, 1997

Figure 3.7: Testing a Date Class

3.3 Object Oriented Numerical Calculations

OOP is often used for numerical computation, especially when the standard storage mode for arrays is
not practical or efficient. Often one will find specialized storage modes like linked lists, or tree structures
used for dynamic data structures. Here we should note that many matrix operators are intrinsic to F90,
so one is more likely to define a class sparse matrix than a class matrix. However, either
class would allow us to encapsulate several matrix functions and subroutines into a module that could be
reused easily in other software. Here, we will illustrate OOP applied to rational numbers and introduce

c 2001 J.E. Akin 38
 Person Class

character name
character nationality
integer sex
Date Date_Of_Birth
Date Date_Of_Death

Person Person_
Person make_Person
Person print_DOB
Person print_DOD
Person print_Name
Person print_Nationality
Person print_Sex
Person set_DOB
Person set_DOD

Person Person

Figure 3.8: Graphical Representation of a Person Class

the important topic of operator overloading. Additional numerical applications of OOP will be illustrated
in later chapters.
3.3.1 A Rational Number Class and Operator Overloading
To illustrate an OOP approach to simple numerical operations we will introduce a fairly complete rational
number class, called class Rational which is represented graphically in Fig. 3.14. The defining F90
module is given in Fig. 3.15. The type components have been made private (line 5), but not the type
itself, so we can illustrate the intrinsic constructor (lines 38 and 102), but extra functionality has been
provided to allow users to get either of the two components (lines 52 and 57). The provided routines
shown in that figure are:
add Rational convert copy Rational delete Rational
equal integer gcd get Denominator get Numerator
invert is equal to list make Rational
mult Rational Rational reduce

Procedures with only one return argument are usually implemented as functions instead of subroutines.
Note that we would form a new rational number, z , as the product of two other rational numbers, x
and y , by invoking the mult Rational function (line 90),
z = mult Rational (x, y)

which returns z as its result. A natural tendency at this point would be to simply write this as z =
x  y. However, before we could do that we would have to have to tell the operator, “*”, how to act
when provided with this new data type. This is known as overloading an intrinsic operator. We had the
foresight to do this when we set up the module by declaring which of the “module procedures” were
equivalent to this operator symbol. Thus, from the “interface operator (*)” statement block (line 14)
the system now knows that the left and right operands of the “*” symbol correspond to the first and
second arguments in the function mult Rational. Here it is not necessary to overload the assignment
operator, “=”, when both of its operands are of the same intrinsic or defined type. However, to convert

c 2001 J.E. Akin 39
[ 1] module class Person ! filename: class Person.f90
[ 2] use class Date
[ 3] implicit none
[ 4] public :: Person
[ 5] type Person
[ 6] private
[ 7] character (len=20) :: name
[ 8] character (len=20) :: nationality
[ 9] integer :: sex
[10] type (Date) :: dob, dod ! birth, death
[11] end type Person
[12] contains
[13] function make Person (nam, nation, s, b, d) result (who)
[14] ! Optional Constructor for a Person type
[15] character (len=*), optional, intent(in) :: nam, nation
[16] integer, optional, intent(in) :: s ! sex
[17] type (Date), optional, intent(in) :: b, d ! birth, death
[18] type (Person) :: who
[19] who = Person (" ","USA",1,Date (1,1,0),Date (1,1,0)) ! defaults
[20] if ( present(nam) ) who % name = nam
[21] if ( present(nation) ) who % nationality = nation
[22] if ( present(s) ) who % sex = s
[23] if ( present(b) ) who % dob = b
[24] if ( present(d) ) who % dod = d ; end function
[25]
[26] function Person (nam, nation, s, b, d) result (who)
[27] ! Public Constructor for a Person type
[28] character (len=*), intent(in) :: nam, nation
[29] integer, intent(in) :: s ! sex
[30] type (Date), intent(in) :: b, d ! birth, death
[31] type (Person) :: who
[32] who = Person (nam, nation, s, b, d) ; end function Person
[33]
[34] subroutine print DOB (who)
[35] type (Person), intent(in) :: who
[36] call print Date (who % dob) ; end subroutine print DOB
[37]
[38] subroutine print DOD (who)
[39] type (Person), intent(in) :: who
[40] call print Date (who % dod) ; end subroutine print DOD
[41]
[42] subroutine print Name (who)
[43] type (Person), intent(in) :: who
[44] print *, who % name ; end subroutine print Name
[45]
[46] subroutine print Nationality (who)
[47] type (Person), intent(in) :: who
[48] print *, who % nationality ; end subroutine print Nationality
[49]
[50] subroutine print Sex (who)
[51] type (Person), intent(in) :: who
[52] if ( who % sex == 1 ) then ; print *, "male"
[53] else ; print *, "female" ; end if ; end subroutine print Sex
[54]
[55] subroutine set DOB (who, m, d, y)
[56] type (Person), intent(inout) :: who
[57] integer, intent(in) :: m, d, y ! month, day, year
[58] who % dob = Date (m, d, y) ; end subroutine set DOB
[59]
[60] subroutine set DOD(who, m, d, y)
[61] type (Person), intent(inout) :: who
[62] integer, intent(in) :: m, d, y ! month, day, year
[63] who % dod = Date (m, d, y) ; end subroutine set DOD
[64] end module class Person

Figure 3.9: Definition of a Typical Person Class

an integer to a rational we could, and have, defined an overloaded assignment operator procedure (line
10). Here we have provided the procedure, equal Integer, which is automatically invoked when
we write : type(Rational)y; y = 4. That would be simpler than invoking the constructor called
make rational. Before moving on note that the system does not yet know how to multiply an integer
times a rational number, or visa versa. To do that one would have to add more functionality, such as a
function, say int mult rn, and add it to the “module procedure” list associated with the “*” operator.
A typical main program which exercises most of the rational number functionality is given in Fig. 3.16,
along with typical numerical output. It tests the constructors Rational (line 8), make Rational

c 2001 J.E. Akin 40
[ 1] include ’class Date.f90’
[ 2] include ’class Person.f90’ ! see previous figure
[ 3] program main
[ 4] use class Date ; use class Person ! inherit class members
[ 5] implicit none
[ 6] type (Person) :: author, creator
[ 7] type (Date) :: b, d ! birth, death
[ 8] b = Date (4,13,1743) ; d = Date (7, 4,1826) ! OPTIONAL
[ 9] ! Method 1
[10] ! author = Person ("Thomas Jefferson", "USA", 1, b, d) ! NOT if private
[11] author = Person ("Thomas Jefferson", "USA", 1, b, d) ! constructor
[12] print *, "The author of the Declaration of Independence was ";
[13] call print Name (author);
[14] print *, ". He was born on "; call print DOB (author);
[15] print *, " and died on "; call print DOD (author); print *, ".";
[16] ! Method 2
[17] author = make Person ("Thomas Jefferson", "USA") ! alternate
[18] call set DOB (author, 4, 13, 1743) ! add DOB
[19] call set DOD (author, 7, 4, 1826) ! add DOD
[20] print *, "The author of the Declaration of Independence was ";
[21] call print Name (author)
[22] print *, ". He was born on "; call print DOB (author);
[23] print *, " and died on "; call print DOD (author); print *, ".";
[24] ! Another Person
[25] creator = make Person ("John Backus", "USA") ! alternate
[26] print *, "The creator of Fortran was "; call print Name (creator);
[27] print *, " who was born in "; call print Nationality (creator);
[28] print *, ".";
[29] end program main ! Running gives:
[30] ! The author of the Declaration of Independence was Thomas Jefferson.
[31] ! He was born on April 13, 1743 and died on July 4, 1826.
[32] ! The author of the Declaration of Independence was Thomas Jefferson.
[33] ! He was born on April 13, 1743 and died on July 4, 1826.
[34] ! The creator of Fortran was John Backus who was born in the USA.

Figure 3.10: Testing the Date and Person Classes

Student Class

Person who
character id [SSN]
Date matriculation
integer credits
real gpa

Student Student_
Student make_Student
Student get_Person
Student print_DOM
Student print_GPA
Student set_DOM

Student Student

Figure 3.11: Graphical Representation of a Student Class

(lines 14, 18, 25), and a simple destructor delete Rational (line 38). The intrinsic constructor (line
6) could have been used only if all the attributes were public, and that is considered an undesirable
practice in OOP. The simple destructor actually just sets the “deleted” number to have a set of default
components. Later we will see that constructors and destructors often must dynamically allocate and
deallocate, respectively, memory associated with a specific instance of some object.

c 2001 J.E. Akin 41
[ 1] module class Student ! filename class Student.f90
[ 2] use class Person ! inherits class Date
[ 3] implicit none
[ 4] public :: Student, set DOM, print DOM
[ 5] type Student
[ 6] private
[ 7] type (Person) :: who ! name and sex
[ 8] character (len=9) :: id ! ssn digits
[ 9] type (Date) :: dom ! matriculation
[10] integer :: credits
[11] real :: gpa ! grade point average
[12] end type Student
[13] contains ! coupled functionality
[14]
[15] function get person (s) result (p)
[16] type (Student), intent(in) :: s
[17] type (Person) :: p ! name and sex
[18] p = s % who ; end function get person
[19]
[20] function make Student (w, n, d, c, g) result (x) ! constructor
[21] ! Optional Constructor for a Student type
[22] type (Person), intent(in) :: w ! who
[23] character (len=*), optional, intent(in) :: n ! ssn
[24] type (Date), optional, intent(in) :: d ! matriculation
[25] integer, optional, intent(in) :: c ! credits
[26] real, optional, intent(in) :: g ! grade point ave
[27] type (Student) :: x ! new student
[28] x = Student (w, " ", Date (1,1,1), 0, 0.) ! defaults
[29] if ( present(n) ) x % id = n ! optional values
[30] if ( present(d) ) x % dom = d
[31] if ( present(c) ) x % credits = c
[32] if ( present(g) ) x % gpa = g ; end function make Student
[33]
[34] subroutine print DOM (who)
[35] type (Student), intent(in) :: who
[36] call print Date(who%dom) ; end subroutine print DOM
[37]
[38] subroutine print GPA (x)
[39] type (Student), intent(in) :: x
[40] print *, "My name is "; call print Name (x % who)
[41] print *, ", and my G.P.A. is ", x % gpa, "." ; end subroutine
[42]
[43] subroutine set DOM (who, m, d, y)
[44] type (Student), intent(inout) :: who
[45] integer, intent(in) :: m, d, y
[46] who % dom = Date ( m, d, y) ; end subroutine set DOM
[47]
[48] function Student (w, n, d, c, g) result (x)
[49] ! Public Constructor for a Student type
[50] type (Person), intent(in) :: w ! who
[51] character (len=*), intent(in) :: n ! ssn
[52] type (Date), intent(in) :: d ! matriculation
[53] integer, intent(in) :: c ! credits
[54] real, intent(in) :: g ! grade point ave
[55] type (Student) :: x ! new student
[56] x = Student (w, n, d, c, g) ; end function Student
[57] end module class Student

Figure 3.12: Defining a Typical Student Class

When considering which operators to overload for a newly defined object one should consider those
that are used in sorting operations, such as the greater-than, >, and less-than, <, operators. They are
often useful because of the need to sort various types of objects. If those symbols have been correctly
overloaded then a generic object sorting routine might be used, or require few changes.

3.4 Discussion
The previous sections have only briefly touched on some important OOP concepts. More details will be
covered later after a general overview of the features of the Fortran language. There are more than one
hundred OOP languages. Persons involved in software development need to be aware that F90 can meet
almost all of their needs for a OOP language. At the same time it includes the F77 standard as a subset
and thus allows efficient use of the many millions of Fortran functions and subroutines developed in the
past. The newer F95 standard is designed to make efficient use of super computers and massively parallel

c 2001 J.E. Akin 42
[ 1] include ’class Date.f90’
[ 2] include ’class Person.f90’
[ 3] include ’class Student.f90’ ! see previous figure
[ 4] program main ! create or correct a student
[ 5] use class Student ! inherits class Person, class Date also
[ 6] implicit none
[ 7] type (Person) :: p ; type (Student) :: x
[ 8] ! Method 1
[ 9] p = make Person ("Ann Jones","",0) ! optional person constructor
[10] call set DOB (p, 5, 13, 1977) ! add birth to person data
[11] x = Student (p, "219360061", Date (8,29,1955), 9, 3.1) ! public
[12] call print Name (p) ! list name
[13] print *, "Born :"; call print DOB (p) ! list dob
[14] print *, "Sex :"; call print Sex (p) ! list sex
[15] print *, "Matriculated:"; call print DOM (x) ! list dom
[16] call print GPA (x) ! list gpa
[17] ! Method 2
[18] x = make Student (p, "219360061") ! optional student constructor
[19] call set DOM (x, 8, 29, 1995) ! correct matriculation
[20] call print Name (p) ! list name
[21] print *, "was born on :"; call print DOB (p) ! list dob
[22] print *, "Matriculated:"; call print DOM (x) ! list dom
[23] ! Method 3
[24] x = make Student (make Person("Ann Jones"), "219360061") ! optional
[25] p = get Person (x) ! get defaulted person data
[26] call set DOM (x, 8, 29, 1995) ! add matriculation
[27] call set DOB (p, 5, 13, 1977) ! add birth
[28] call print Name (p) ! list name
[29] print *, "Matriculated:"; call print DOM (x) ! list dom
[30] print *, "was born on :"; call print DOB (p) ! list dob
[31] end program main ! Running gives:
[32] ! Ann Jones
[33] ! Born : May 13, 1977
[34] ! Sex : female
[35] ! Matriculated: August 29, 1955
[36] ! My name is Ann Jones, and my G.P.A. is 3.0999999.
[37] ! Ann Jones was born on: May 13, 1977 , Matriculated: August 29, 1995
[38] ! Ann Jones Matriculated: August 29, 1995 , was born on: May 13, 1977

Figure 3.13: Testing the Student, Person, and Date Classes

machines. It includes most of the High Performance Fortran features that are in wide use. Thus, efficient
use of OOP on parallel machines is available through F90 and F95.
None of the OOP languages have all the features one might desire. For example, the useful concept
of a “template” which is standard in C++ is not in the F90 standard. Yet the author has found that a
few dozen lines of F90 code will define a preprocessor that allows templates to be defined in F90 and
expanded in line at compile time. The real challenge in OOP is the actual OOA and OOD that must be
completed before programming can begin, regardless of the language employed. For example, several
authors have described widely different approaches for defining classes to be used in constructing OO
finite element systems. Additional example applications of OOP in F90 will be given in the following
chapters.

c 2001 J.E. Akin 43
 Rational Class

integer numerator
integer denominator

Rational Rational_
Rational make_Rational
Rational add_Rational
Rational convert
Rational copy_Rational
Rational delete_Rational
Rational equal_Rational
Rational get_Denominator
Rational get_Numerator
Rational invert
Rational is_equal_to
Rational list
Rational mult_Rational

Rational Rational
integer gcd
Rational reduce

Figure 3.14: Representation of a Rational Number Class

c 2001 J.E. Akin 44
[ 1] module class Rational ! filename: class Rational.f90
[ 2] implicit none
[ 3] ! public, everything but following private routines
[ 4] private :: gcd, reduce
[ 5] type Rational
[ 6] private ! numerator and denominator
[ 7] integer :: num, den ; end type Rational
[ 8]
[ 9] ! overloaded operators interfaces
[ 10] interface assignment (=)
[ 11] module procedure equal Integer ; end interface
[ 12] interface operator (+) ! add unary versions & (-) later
[ 13] module procedure add Rational ; end interface
[ 14] interface operator (*) ! add integer mult Rational, etc
[ 15] module procedure mult Rational ; end interface
[ 16] interface operator (==)
[ 17] module procedure is equal to ; end interface
[ 18] contains ! inherited operational functionality
[ 19] function add Rational (a, b) result (c) ! to overload +
[ 20] type (Rational), intent(in) :: a, b ! left + right
[ 21] type (Rational) :: c
[ 22] c % num = a % num*b % den + a % den*b % num
[ 23] c % den = a % den*b % den
[ 24] call reduce (c) ; end function add Rational
[ 25]
[ 26] function convert (name) result (value) ! rational to real
[ 27] type (Rational), intent(in) :: name
[ 28] real :: value ! decimal form
[ 29] value = float(name % num)/name % den ; end function convert
[ 30]
[ 31] function copy Rational (name) result (new)
[ 32] type (Rational), intent(in) :: name
[ 33] type (Rational) :: new
[ 34] new % num = name % num
[ 35] new % den = name % den ; end function copy Rational
[ 36]
[ 37] subroutine delete Rational (name) ! deallocate allocated items
[ 38] type (Rational), intent(inout) :: name ! simply zero it here
[ 39] name = Rational (0, 1) ; end subroutine delete Rational
[ 40]
[ 41] subroutine equal Integer (new, I) ! overload =, with integer
[ 42] type (Rational), intent(out) :: new ! left side of operator
[ 43] integer, intent(in) :: I ! right side of operator
[ 44] new % num = I ; new % den = 1 ; end subroutine equal Integer
[ 45]
[ 46] recursive function gcd (j, k) result (g) ! Greatest Common Divisor
[ 47] integer, intent(in) :: j, k ! numerator, denominator
[ 48] integer :: g
[ 49] if ( k == 0 ) then ; g = j
[ 50] else ; g = gcd ( k, modulo(j,k) ) ! recursive call
[ 51] end if ; end function gcd
[ 52]
[ 53] function get Denominator (name) result (n) ! an access function
[ 54] type (Rational), intent(in) :: name
[ 55] integer :: n ! denominator
[ 56] n = name % den ; end function get Denominator

(Fig. 3.15, A Fairly Complete Rational Number Class (continued))

c 2001 J.E. Akin 45
[ 57] function get Numerator (name) result (n) ! an access function
[ 58] type (Rational), intent(in) :: name
[ 59] integer :: n ! numerator
[ 60] n = name % num ; end function get Numerator
[ 61]
[ 62] subroutine invert (name) ! rational to rational inversion
[ 63] type (Rational), intent(inout) :: name
[ 64] integer :: temp
[ 65] temp = name % num
[ 66] name % num = name % den
[ 67] name % den = temp ; end subroutine invert
[ 68]
[ 69] function is equal to (a given, b given) result (t f)
[ 70] type (Rational), intent(in) :: a given, b given ! left == right
[ 71] type (Rational) :: a, b ! reduced copies
[ 72] logical :: t f
[ 73] a = copy Rational (a given) ; b = copy Rational (b given)
[ 74] call reduce(a) ; call reduce(b) ! reduced to lowest terms
[ 75] t f = (a%num == b%num) .and. (a%den == b%den) ; end function
[ 76]
[ 77] subroutine list(name) ! as a pretty print fraction
[ 78] type (Rational), intent(in) :: name
[ 79] print *, name % num, "/", name % den ; end subroutine list
[ 80]
[ 81] function make Rational (numerator, denominator) result (name)
[ 82] ! Optional Constructor for a rational type
[ 83] integer, optional, intent(in) :: numerator, denominator
[ 84] type (Rational) :: name
[ 85] name = Rational(0, 1) ! set defaults
[ 86] if ( present(numerator) ) name % num = numerator
[ 87] if ( present(denominator)) name % den = denominator
[ 88] if ( name % den == 0 ) name % den = 1 ! now simplify
[ 89] call reduce (name) ; end function make Rational
[ 90]
[ 91] function mult Rational (a, b) result (c) ! to overload *
[ 92] type (Rational), intent(in) :: a, b
[ 93] type (Rational) :: c
[ 94] c % num = a % num * b % num
[ 95] c % den = a % den * b % den
[ 96] call reduce (c) ; end function mult Rational
[ 97]
[ 98] function Rational (numerator, denominator) result (name)
[ 99] ! Public Constructor for a rational type
[100] integer, optional, intent(in) :: numerator, denominator
[101] type (Rational) :: name
[102] if ( denominator == 0 ) then ; name = Rational (numerator, 1)
[103] else ; name = Rational (numerator, denominator) ; end if
[104] end function Rational
[105]
[106] subroutine reduce (name) ! to simplest rational form
[107] type (Rational), intent(inout) :: name
[108] integer :: g ! greatest common divisor
[109] g = gcd (name % num, name % den)
[110] name % num = name % num/g
[111] name % den = name % den/g ; end subroutine reduce
[112] end module class Rational

Figure 3.15: A Fairly Complete Rational Number Class

c 2001 J.E. Akin 46
[ 1] include ’class Rational.f90’
[ 2] program main
[ 3] use class Rational
[ 4] implicit none
[ 5] type (Rational) :: x, y, z
[ 6] ! ------- only if Rational is NOT private ----------
[ 7] ! x = Rational(22,7) ! intrinsic constructor if public components
[ 8]
[ 9] x = Rational (22,7) ! public constructor if private components
[10] write (*,’("public x = ")’,advance=’no’); call list(x)
[11] write (*,’("converted x = ", g9.4)’) convert(x)
[12] call invert(x)
[13] write (*,’("inverted 1/x = ")’,advance=’no’); call list(x)
[14]
[15] x = make Rational () ! default constructor
[16] write (*,’("made null x = ")’,advance=’no’); call list(x)
[17] y = 4 ! rational = integer overload
[18] write (*,’("integer y = ")’,advance=’no’); call list(y)
[19] z = make Rational (22,7) ! manual constructor
[20] write (*,’("made full z = ")’,advance=’no’); call list(z)
[21] ! Test Accessors
[22] write (*,’("top of z = ", g4.0)’) get numerator(z)
[23] write (*,’("bottom of z = ", g4.0)’) get denominator(z)
[24] ! Misc. Function Tests
[25] write (*,’("making x = 100/360, ")’,advance=’no’)
[26] x = make Rational (100,360)
[27] write (*,’("reduced x = ")’,advance=’no’); call list(x)
[28] write (*,’("copying x to y gives ")’,advance=’no’)
[29] y = copy Rational (x)
[30] write (*,’("a new y = ")’,advance=’no’); call list(y)
[31] ! Test Overloaded Operators
[32] write (*,’("z * x gives ")’,advance=’no’); call list(z*x) ! times
[33] write (*,’("z + x gives ")’,advance=’no’); call list(z+x) ! add
[34] y = z ! overloaded assignment
[35] write (*,’("y = z gives y as ")’,advance=’no’); call list(y)
[36] write (*,’("logic y == x gives ")’,advance=’no’); print *, y==x
[37] write (*,’("logic y == z gives ")’,advance=’no’); print *, y==z
[38] ! Destruct
[39] call delete Rational (y) ! actually only null it here
[40] write (*,’("deleting y gives y = ")’,advance=’no’); call list(y)
[41] end program main ! Running gives:
[42] ! public x = 22 / 7 ! converted x = 3.143
[43] ! inverted 1/x = 7 / 22 ! made null x = 0 / 1
[44] ! integer y = 4 / 1 ! made full z = 22 / 7
[45] ! top of z = 22 ! bottom of z = 7
[46] ! making x = 100/360, reduced x = 5 / 18
[47] ! copying x to y gives a new y = 5 / 18
[48] ! z * x gives 55 / 63 ! z + x gives 431 / 126
[49] ! y = z gives y as 22 / 7 ! logic y == x gives F
[50] ! logic y == z gives T ! deleting y gives y = 0 / 1

Figure 3.16: Testing the Rational Number Class

c 2001 J.E. Akin 47
3.5 Exercises
1. Use the class Circle to create a class Sphere that computes the volume of a sphere. Have
a method that accepts an argument of a Circle. Use the radius of the Circle via a new member
get Circle radius to be added to the class Circle.
2. Use the class Circle and class Rectangle to create a class Cylinder that computes
the volume of a right circular cylinder. Have a method that accepts arguments of a Circle and a height,
and a second method that accepts arguments of a Rectangle and a radius. In the latter member
use the height of the Rectangle via a new member get Rectangle height to be added to the
class Rectangle.
3. Create a vector class to treat vectors with an arbitrary number of real coefficients. Assume that the
class Vector is defined as follows:

Vector Class

integer size
real, pointer data (:)

Vector assign
Vector make_Vector
Vector add_Real_to_Vector
Vector add_Vector
Vector copy_Vector
Vector delete_Vector
real dot_Vector
Vector equal_Real
logical is_equal_to
real length
Vector list
Vector normalize_Vector
Vector read_Vector
Vector real_mult_Vector
integer size_Vector
Vector subtract_Real
Vector subtract_Vector
real values
Vector Vector_
real Vector_max_value
real Vector_min_value
Vector Vector_mult_real

Vector Vector

Overload the common operators of (+) with add Vector and add Real to Vector, (–) with
subtract Vector and subtract Real, (*) with dot Vector, real mult Vector and Vec-
tor mult real, (=) with equal Real to set all coefficients to a single real number, and (==) with
routine is equal to.
Include two constructors assign and make Vector. Let assign convert a real array into an instance
of a Vector. Provide a destructor, means to read and write a Vector, normalize a Vector, and determine its
extreme values.

c 2001 J.E. Akin 48
 4. Modify the above Vector class to extend it to a Sparse Vector Class where the vast majority
of the coefficients are zero. Store and operate only on the non-zero entries.

Sparse_Vector Class

integer non_zeros
integer, pointer rows (:)
real, pointer values (:)

Sparse_Vector make_Sparse_Vector
Sparse_Vector add_Real_to_Sparse_Vector
Sparse_Vector add_Sparse_Vector
Sparse_Vector delete_Sparse_Vector
real dot_Vector
Sparse_Vector el_by_el_Mult
Sparse_Vector equal_Vector
real get_element
logical is_equal_to
integer largest_index
real normalize_Vector
length
real norm
Sparse_Vector normalize_Vector
Sparse_Vector pretty
Sparse_Vector read_Vector
Sparse_Vector real_mult_Sparse
integer rows_of
Sparse_Vector set_element
Sparse_Vector show
Sparse_Vector show_r_v
integer size_of
Sparse_Vector Sparse_mult_real
Sparse_vector sub_Sparse_Vector
Sparse_Vector sum_Sparse_Vector
real Vector_max_value
real Vector_min_value
Sparse_Vector Vector_to_Sparse
Sparse_Vector zero_Sparse

Sparse_Vector Sparse_Vector

c 2001 J.E. Akin 49
Chapter 4

Features of Programming Languages

The preceding chapter described the programming process as starting with a clearly specified task, ex-
pressing it mathematically as a set of algorithms, translating the algorithms in pseudocode, and finally,
translating the pseudocode into a “real” programming language. The final stages of this prescription work
because most (if not all) computational languages have remarkable similarities: They have statements,
the sequencing of which are controlled by various loop and conditional constructs, and functions that
foster program modularization. We indicated how similar MATLAB, C++, and Fortran are at this level,
but these languages differ the more they are detailed. It is the purpose of this chapter to describe those
details, and bring you from a superficial acquaintance with a computational language to fluency. Today,
the practicing engineer needs more than one programming language or environment. Once achieving
familiarity with one, you will find that learning other languages is easy.
When selecting a programming tool for engineering calculations, one is often faced with two different
levels of need. One level is where you need to quickly solve a small problem once, such as a homework
assignment, and computational efficiency is not important. You may not care if your code takes ten
seconds or one hundred seconds to execute; you want convenience. At that level it may make sense
to use an engineering environment like M ATLAB, or Mathematica. At the other extreme you may be
involved in doing a wide area weather prediction where a one-day run time, instead of a ten-day run time,
defines a useful versus a non-useful product. You might be developing a hospital laboratory system for
reporting test results to an emergency room physician where an answer in ten seconds versus an answer in
ten minutes can literally mean the difference between life or death for a patient. For programming at this
level one wants an efficient language. Since such projects can involve programming teams in different
countries, you want your language to be based on an international standard. Then you would choose to
program a language such as C++ or F90. Since most students have experienced only the first need level,
they tend to overvalue the first approach and devalue the second. This chapter will illustrate that the skills
needed for either approach are similar.
The structure of this chapter follows our usual progression to learning a language: What are variables,
how can variables be combined into expressions, what constructs are available to control program flow,
and how are functions defined so that we can employ modularity. The basics are described in Chapter 1;
we assume you are familiar with the language basics described there. Initially, this chapter will parallel
the program composition section of Chapter 1 as applied in the C++, F90, and M ATLAB languages, and
then it will bring in more advanced topics.
The features of F90 that are to be discussed here have been combined in a series of tables and placed
in Appendix B. It is expected that we will want to refer to those tables as we read this section as well
as later when we program. At times, references to C++ and M ATLAB have been given to show the
similarities between most languages and to provide an aid for when having to interface in reading codes
in those languages.

4.1 Comments
In M ATLAB and Fortran, a single character ‘%’ in M ATLAB, ‘!’ in F90 located anywhere in a line of
text means that the remainder of the text on that line comprises the comment. In C, an entirely different

c 2001 J.E. Akin 51
 Language Syntax Location
M ATLAB % comment (to end of line) anywhere
C++ // comment (to end of line) anywhere
F90 ! comment (to end of line) anywhere
F77 * comment (to end of line) column 1

Table 4.1: Comment syntax

structure for comments occurs. Comments begin with the two-character sequence ‘/*’ and end with the
next occurrence of the two-character sequence ‘*/’. In C, comments can occur anywhere in a program;
they can consume a portion of a line, temporarily interrupting a statement, or they can span multiple
lines of text. C++ allows the use of the C comment syntax, but has added a more popular two-character
sequence ‘//’ to proceed a comment to the end of a line. Table 4.1 gives a summary of these comments
syntax. It is also in the “Fortran 90 Overview” for quick reference. Samples of comment statements are
shown in Fig. 1.3, which gives the corresponding versions of the classic “hello world” program included
in most introductory programming texts.

4.2 Statements and Expressions

Before introducing statements and expressions, a word about documenting what you program. We en-
courage the heavy usage of comments. The three languages of concern here all allow comment lines and
comments appended to the end of statements. Their form is given above in Fig. 1.3 and Table 4.1.
The above languages currently allow variable names to contain up to 31 characters and allow the use
of the underscore, ‘ ’, to aid in clarity by serving as a virtual space character, as in my name. Another
useful convention is to use uppercase first letters for words comprising part of a variable’s name: MyName.
Fortran and M ATLAB allow a program line to contain up to 132 characters, while C++ has no limit on
line length. Since the old F77 standard was physically limited to holes punched in a card, it allowed only
a line length of 72 characters, a maximum name length of six characters, and did not allow the use of the
underscore in a name. In this text, we will usually keep line lengths to less than 65 characters in order to
make the programs more readable.
A statement in these three languages has a structure common to them all:
variable = expression

The built-in, or intrinsic, data types allowed for variables are summarized in Table 4.2. Additional user
defined types will be considered later. The expressions usually involves the use of arithmetic operators
and/or relational operators which are given in Tables 4.3 and 4.4, respectively. The order in which the
language applies these operators is called their precedence, and they are shown in Table 4.5. They are
also in the “Fortran 90 Overview” for quick reference.
In moving from M ATLAB to high level languages one finds that it is necessary to define the type of
each variable. Fortran has a default naming convention for its variables and it allows an easy overriding
of that built in “implicit” convention. Since most engineering and mathematical publications used the
letters from “i” through “n” as subscripts, summation ranges, loop counters, etc. Fortran first was released
with implicit variable typing such that all variables whose name begin with the letters “i” through “n”,
inclusive, defaulted to integers, unless declared otherwise. All other variables default to be real, unless
declared otherwise. In other words, you can think of the default code as if it contained the statements:

IMPLICIT INTEGER (I-N) ! F77 and F90 Default

IMPLICIT REAL (A-H, O-Z) ! F77 and F90 Default

The effect is automatic even if the statements are omitted. Explicit type declarations override any given
IMPLICIT types. For example, if the code had the above implicit defaults one could also explicitly
identify the exceptions to those default rules, such as the statements:

INTEGER :: Temp row

c 2001 J.E. Akin 52
 Storage M ATLABa C++ F90 F77
byte char character:: character
integer int integer:: integer
single precision float real:: real
double precision double real*8:: double precision
b
complex complex:: complex
Boolean bool logical:: logical
argument parameter:: parameter
pointer * pointer::
structure struct type::
a M ATLAB 4 requires no variable type declaration; the only two distinct types in M ATLAB are strings and reals (which include

complex). Booleans are just 0s and 1s treated as reals. M ATLAB5 allows the user to select more types.
b There is no specific data type for a complex variable in C++; they must be created by the programmer.

Table 4.2: Intrinsic data types of variables

Description M ATLABa C++ Fortranb

addition + + +
subtractionc - - -
multiplication * and .* * *
division / and ./ / /
exponentiation ˆ and .ˆ powd **
remainder %
increment ++
decrement --
parentheses (expres- () () ()
sion grouping)
a When doing arithmetic operations on matrices in M ATLAB , a period (‘.’) must be put before the operator if scalar arithmetic

is desired. Otherwise, M ATLAB assumes matrix operations; figure out the difference between ‘*’ and ‘.*’. Note that since matrix
and scalar addition coincide, no ‘.+’ operator exists (same holds for subtraction).
b Fortran 90 allows the user to change operators and to define new operator symbols.
c In all languages the minus sign is used for negation (i.e., changing sign).
d In C++ the exponentiation is calculated by function pow (x; y ).

Table 4.3: Arithmetic operators

REAL :: Interest = 0.04 ! declare and initialize

CHARACTER (Len=8) :: Months of year(12)

We will also see that the programmer can define new data types and explicitly declare their type as well.
The F90 standard discourages the use of any IMPLICIT variables such as

IMPLICIT COMPLEX (X-Z) ! Complex variables

IMPLICIT DOUBLE PRECISION (A-H) ! Double Precision reals

and encourages the use of

IMPLICIT NONE

which forces the programmer to specifically declare the type of each and every variable used, and is
referred to as strong typing. However, you need to know that such default variable types exist because
they are used in many millions of lines of older Fortran code and at some point you will need to use or
change such an existing program.

c 2001 J.E. Akin 53
 Description M ATLAB C++ F90 F77
Equal to == == == .EQ.
Not equal to ˜= != /= .NE.
Less than < < < .LT.
Less or equal <= <= <= .LE.
Greater than > > > .GT.
Greater or equal >= >= >= .GE.
Logical NOT ˜ ! .NOT. .NOT.
Logical AND & && .AND. .AND.
Logical inclusive OR ! || .OR. .OR.
Logical exclusive OR xor .XOR. .XOR.
Logical equivalent == == .EQV. .EQV.
Logical not equivalent ˜= != .NEQV. .NEQV.

Table 4.4: Relational operators (arithmetic and logical)

M ATLAB
C++ Operators F90 Operatorsa F77 Operators
Operators
() () [] -> . () ()
+ - ! ++ -- + ** **
- * & (type)
sizeof
* / * / % * / * /
+ -b + -b + -b + -b
< <= > >= << >> // //
== ˜= < <= > => == /= < <= > .EQ. .NE.
>= .LT. .LE.
.GT. .GE.
˜ == != .NOT. .NOT.
& && .AND. .AND.
| || .OR. .OR.
= | .EQV. .NEQV. .EQV. .NEQV.
?:
= += -= *= /=
%= &= ˆ= |=
<<= >>=
,
a User-defined unary (binary) operators have the highest (lowest) precedence in F90.
b These are binary operators representing addition and subtraction. Unary operators + and - have higher precedence.

Table 4.5: Precedence pecking order

c 2001 J.E. Akin 54
[ 1] program main
[ 2] ! Examples of simple arithmetic in F90
[ 3] implicit none
[ 4] integer :: Integer Var 1, Integer Var 2 ! user inputs
[ 5] integer :: Mult Result, Div Result, Add Result
[ 6] integer :: Sub Result, Mod Result
[ 7] real :: Pow Result, Sqrt Result
[ 8]
[ 9] print *, ’Enter two integers:’
[10] read *, Integer Var 1, Integer Var 2
[11]
[12] Add Result = Integer Var 1 + Integer Var 2
[13] print *, Integer Var 1,’ + ’, Integer Var 2,’ = ’, Add
[14]
[15] Sub Result = Integer Var 1 - Integer Var 2
[16] print *, Integer Var 1,’ - ’, Integer Var 2,’ = ’, Sub
[17]
[18] Mult Result = Integer Var 1 * Integer Var 2
[19] print *, Integer Var 1,’ * ’, Integer Var 2,’ = ’, Mult
[20]
[21] Div Result = Integer Var 1 / Integer Var 2
[22] print *, Integer Var 1,’ / ’, Integer Var 2,’ = ’, Div
[23]
[24] Mod Result = mod (Integer Var 1, Integer Var 2) ! remai
[25] print *, Integer Var 1,’ mod ’, Integer Var 2,’ = ’, Mod
[26]
[27] Pow Result = Integer Var 1 ** Integer Var 2 ! raise t
[28] print *, Integer Var 1,’ ˆ ’, Integer Var 2,’ = ’, Pow
[29]
[30] Sqrt Result = sqrt( real(Integer Var 1))
[31] print *,’Square root of ’, Integer Var 1,’ = ’, Sqrt Result
[32]
[33] end program main ! Running produces:
[34] ! Enter two integers:
[35] ! 25 + 4 = 29
[36] ! 25 - 4 = 21
[37] ! 25 * 4 = 100
[38] ! 25 / 4 = 6, note integer
[39] ! 25 mod 4 = 1
[40] ! 25 ˆ 4 = 3.9062500E+05
[41] ! Square root of 25 = 5.0000000

Figure 4.1: Typical Math and Functions in F90

An example program that employs the typical math operators in F90 is shown in Fig. 4.1. It presents
examples of addition (line 11), subtraction (line 14), multiplication (line 17), division (line 20), as well as
the use of the remainder or modulo function (line 23), exponentiation (line 26), and square root operators
(line 29). In addition it shows a way of inputing data from the default input device (line 9). The results
are appended as comments (lines 33-40). Observe that a program must include one and only one segment
that begins with the word program (line 1) and ends with the line end program (line 32). If a name
is assigned to the program then it must be appended to both of these lines. Often the name of main is
used, as here, but it is not required as it is in C++ . A C++ formulation of this example is included for
comparison in the appendix as are several other examples from this chapter.

A special expression available in M ATLAB and F90 uses the colon operator (:) to indicate forming
a vector (row matrix) of numbers according to an arithmetic progression. In MATLAB, the expression
b:i:e means the vector [b (b + i) (b + 2i)


(b + N i)], where (b + N i) is the largest number less than
or equal to (greater than or equal to if i is negative) the value of the variable e. Thus, b means “beginning
value”, i means the increment, and e the end value. The expression b:e means that the increment equals
one. You can use this construct to excise a portion of a vector or matrix. For example, x(2:5) equals
the vector comprised by the second through fifth elements of x, and A(3:5,i:j) creates a matrix from
the third, fourth, and fifth rows, ith through j th columns of the matrix A. F90 uses the convention of
b:e:i and has the same defaults when :i is omitted. This operator, also known as the subscript triplet,
is described in Table 4.6.
Of course, expressions often involve the use of functions. A tabulation of the built-in functions in
our languages is given in Table 4.7 and the F90 overview, as are all the remaining tables of this chapter.
The arguments of functions and subprograms have some important properties that vary with the language
used. Primarily, we are interested in how actual arguments are passed to the dummy arguments in the
subprogram. This data passing happens by either of two fundamentally different ways: by reference, or

c 2001 J.E. Akin 55
 B = Beginning, E = Ending, I = Increment

Syntax F90 M ATLAB Use F90 M ATLAB

Default B:E:I B:I:E Array subscript ranges yes yes
B B: B: Character positions in a string yes yes
E :E :E Loop control no yes
Full range : : Array element generation no yes

Table 4.6: Colon Operator Syntax and its Applications.

Description M ATLAB C++ F90 F77

exponential exp(x) exp(x) exp(x) exp(x)
natural log log(x) log(x) log(x) log(x)
base 10 log log10(x) log10(x) log10(x) log10(x)
square root sqrt(x) sqrt(x) sqrt(x) sqrt(x)
raise to power (xr ) x.ˆr pow(x,r) x**r x**r
absolute value abs(x) fabs(x) abs(x) abs(x)
smallest integer>x ceil(x) ceil(x) ceiling(x)
largest integer<x floor(x) floor(x) floor(x)
division remainder rem(x,y) fmod(x,y) mod(x,y)a mod(x,y)
modulo modulo(x,y)a
complex conjugate conj(z) conjg(z) conjg(z)
imaginary part imag(z) imag(z) aimag(z)
drop fraction fix(x) aint(x) aint(x)
round number round(x) nint(x) nint(x)
cosine cos(x) cos(x) cos(x) cos(x)
sine sin(x) sin(x) sin(x) sin(x)
tangent tan(x) tan(x) tan(x) tan(x)
arc cosine acos(x) acos(x) acos(x) acos(x)
arc sine asin(x) asin(x) asin(x) asin(x)
arc tangent atan(x) atan(x) atan(x) atan(x)
arc tangentb atan2(x,y) atan2(x,y) atan2(x,y) atan2(x,y)
hyperbolic cosine cosh(x) cosh(x) cosh(x) cosh(x)
hyperbolic sine sinh(x) sinh(x) sinh(x) sinh(x)
hyperbolic tangent tanh(x) tanh(x) tanh(x) tanh(x)
hyperbolic arc cosine acosh(x)
hyperbolic arc sine asinh(x)
hyperbolic arctan atanh(x)
a Differ for x < 0.
b atan2(x,y) is used to calculate the arc tangent of x=y in the range [ ; +]. The one-argument function atan(x)
computes the arc tangent of x in the range [ =2; +=2].

Table 4.7: Mathematical functions

by value. One should understand the difference between these two mechanisms.
“Passing by reference” means that the address in memory of the actual argument is passed to the
subprogram instead of the value stored at that address. The corresponding dummy argument in the
subprogram has the same address. That is, both arguments refer to the same memory location so any
change to that argument within the subprogram is passed back to the calling code. A variable is passed
by reference to a subroutine whenever it is expected that it should be changed by the subprogram. A
related term is “dereferencing”. When you dereference a memory address, you are telling the computer
to get the information located at the address. Typically, one indirectly gives the address by citing the

c 2001 J.E. Akin 56
 Description C++ F90 F77 M ATLAB
Conditionally execute statements if if if if
f g end if end if end

Loop a specific number of times for k=1:n do k=1,n do # k=1,n for k=1:n
f g end do # continue end

Loop an indefinite number of times while do while — while

f g end do — end

Terminate and exit loop break exit go to break

Skip a cycle of loop continue cycle go to —

Display message and abort error() stop stop error

Return to invoking function return return return return

Conditional array action — where — if

Conditional alternate statements else else else else

else if elseif elseif elseif

Conditional array alternatives — elsewhere — else

— — — elseif

Conditional case selections switch f g select case if if

end select end if end

Table 4.8: Flow Control Statements.

name of a pointer variable or a reference variable.

“Passing by value” means that the value of the actual argument stored at its address in memory is
copied and the copy is passed to the dummy argument in the subprogram. Thus any change to the
argument within the subprogram is not passed back to the calling code. The two passing methods do
not clearly show the intended use of the argument within the subprogram. Is it to be passed in for use
only, passed in for changing and returned, or is it to be created in the subprogram and passed out for use
in the calling code? For additional safety and clarity modern languages provide some way to allow the
programmer to optionally specify such intent explicitly.
Both C++ and M ATLAB use the pass by value method as their default mode. This means the value
associated with the argument name, say arg name, is copied and passed to the function. That copying
could be very inefficient if the argument is a huge array. To denote that you want to have the C++ argu-
ment passed by reference you must precede the argument name with an ampersand (&), e.g. &arg name,
in the calling code. Then within the subprogram the corresponding dummy variable must be derefer-
enced by preceding the name with an asterisk (*), e.g. *arg name. Conversely, Fortran uses the passing
by reference method as its default mode. On the rare occasions when one wants to pass by value simply
surround the argument name with parentheses, e.g. (arg name), in the calling code. In either case it is
recommended that you cite each argument with the optional “intent” statement within the subprogram.
Examples of the two passing options are covered in Sec. 4.5.

4.3 Flow Control

The basic flow control constructs present in our selected engineering languages are loops repetitive
execution of a block of statements and conditionals diversions around blocks of statements. A typical
set of flow control statement types are summarized in Table 4.8. Most of these will be illustrated in detail
in the following sections.

c 2001 J.E. Akin 57
4.3.1 Explicit Loops
The following discussion will introduce the important concept of loops. These are required in most
programs. However, the reader is warned that today the writing of explicit loops are generally not the
most efficient way to execute a loop operation in Fortran90 and M ATLAB. Of course, older languages
like F77 and C do require them, so that the time spent here not only covers the explicit loop concepts but
aids one in reading older languages. Our pseudocode for the common loops is :

c 2001 J.E. Akin 58
 Loop M ATLAB C++ Fortran
Indexed loop
f
for index=matrix for (init;test;inc) do index=b,e,i
statements statements
statements
g
end end do

Pre-test loop while test while (test) f do while (test)

statements statements statements
end g end do

Post-test loop do f do
statements statements
g while (test) if (test) exit
end do

Table 4.9: Basic loop constructs

Loop Pseudocode
Indexed loop for index=b,i,e
statements
end for

Pre-test loop while (test)

statements
end while

Post-test loop do
statements
if test exit
end do

In engineering programming one often needs to repeatedly perform a group of operations. Most
computer languages have a statement to execute this powerful and widely-used feature. In Fortran this
is the DO statement, while in C++ and M ATLAB it is the FOR statement. This one statement provides
for the initialization, incrementing and testing of the loop variable, plus repeated execution of a group of
statements contained within the loop. In Fortran77, the loop always cites a label number that indicates
the extent of the statements enclosed in the loop. This is allowed in F90, but not recommended, and is
considered obsolete. Instead, the END DO indicates the extent of the loop, and the number label is omitted
in both places. F90 does allow one to give a name to a loop. Then the structure is denoted as NAME:DO
followed by END DO NAME. Examples of the syntax for these statements for the languages of interest are
given in Table 4.9.
A simple example of combining loops and array indexing is illustrated in Figs. 4.2 and 4.3. Note in
Fig. 4.2 that the final value of a loop counter (called Integer Var here) upon exiting the loop (line 10)
can be language or compiler dependent despite the fact that they are same here. In Fig. 4.3, we introduce
for the first time a variable with a single subscript (line 5) and containing five numbers (integers) to be
manually initialized (lines 8-10) and then to be listed in a loop (lines 12-15) over all their values. Note
that C++ stores the first entry in an array at position zero (see appendix listing), M ATLAB uses position
one, and F90 defaults to position one.
C++ and Fortran 90 allow a special option to create loops that run “forever.” These could be used, for
example, to read an unknown amount of data until terminated, in a non-fatal way, by the input statement.
In C++, one omits the three loop controls, such as
f
for (;;) // forever loop
loop block
g // end forever loop

while in F90, one simply omits the loop control and gives only the DO command:
do ! forever

c 2001 J.E. Akin 59
[ 1] program main
[ 2] ! Examples of a simple loop in F90
[ 3] implicit none
[ 4] integer Integer Var
[ 5]
[ 6] do Integer Var = 0,4,1
[ 7] print *, ’The loop variable is:’, Integer Var
[ 8] end do ! over Integer Var
[ 9]
[10] print *, ’The final loop variable is:’, Integer Var
[11]
[12] end program main ! Running produces:
[13] ! The loop variable is: 0
[14] ! The loop variable is: 1
[15] ! The loop variable is: 2
[16] ! The loop variable is: 3
[17] ! The loop variable is: 4
[18] ! The final loop variable is: 5 <- NOTE

Figure 4.2: Typical Looping Concepts in F90

[ 1] program main
[ 2] ! Examples of simple array indexing in F90
[ 3] implicit none
[ 4] integer, parameter :: max = 5
[ 5] integer Integer Array(max) ! =(/ 10 20 30 40 50 /), or set below
[ 6] integer loopcount
[ 7]
[ 8] Integer Array(1) = 10 ! F90 index starts at 1, usually
[ 9] Integer Array(2) = 20 ; Integer Array(3) = 30
[10] Integer Array(4) = 40 ; Integer Array(5) = 50
[11]
[12] do loopcount = 1, max ! & means continued
[13] print *, ’The loop counter is: ’, loopcount, &
[14] ’ with an array value of: ’,Integer Array(loopcount)
[15] end do ! over loopcount
[16]
[17] print *, ’The final loop counter is: ’, loopcount
[18]
[19] end program main
[20] ! Running produces:
[21] ! The loop counter is: 1 with an array value of: 10
[22] ! The loop counter is: 2 with an array value of: 20
[23] ! The loop counter is: 3 with an array value of: 30
[24] ! The loop counter is: 4 with an array value of: 40
[25] ! The loop counter is: 5 with an array value of: 50
[26] ! The final loop counter is: 6

Figure 4.3: Simple Array Indexing in F90

loop block
end do ! forever

Most of the time, an infinite loop is used as a loop while true or a loop until true construct. These
will be considered shortly.
4.3.2 Implied Loops
Fortran and M ATLAB have shorthand methods for constructing “implied loops.” Both languages offer
the colon operator to imply an incremental range of integer values. Its syntax and types of applications
are given in Table 4.6 (page 56). The allowed usages of the operator differ slightly between the two
languages. Note that this means that the loop controls are slightly different in that the do control employs
commas instead of colons. For example, two equivalent loops are

Fortran M ATLAB
do k=B,E,I for k=B:I:E
A(k) = k**2 A(k) = k 2 ^
end do
end

Fortran offers an additional formal implied do loop that replaces the do and end do with a closed
pair of parentheses in the syntax:
(object, k = B,E,I)

c 2001 J.E. Akin 60
where again the increment, I, defaults to unity if not supplied. The above implied do is equivalent to the
formal loop
do k=B,E,I
define object
end do

However, the object defined in the implied loop can only be utilized for four specific Fortran operations:
1) read actions, 2) print and write actions, 3) data variables (not value) definitions, and 4) defining
array elements. For example,
print *, (4*k-1, k=1,10,3) ! 3, 15, 27, 39
read *, (A(j,:), j=1,rows) ! read A by rows, sequentially

The implied do loops can be nested to any level like the standard do statement. One simply makes the
inner loop the object of the outer loop, so that
((object j k, j=min, max), k=k1,k2,inc)

implies the nested loop

do k=k1,k2,inc
do j=min, max
use object j k
end do ! over j
end do ! over k

For example,
print *, (((A(k)*B(j)+3), j=1,5), k=1,max)
! read array by rows in each plane
read *, (((A(i,j,k), j=1,cols), i=1,rows), k=1,max)

Actually, there is even a simpler default form of implied dos for reading and writing arrays. That default
is to access arrays by columns. That is, process the leftmost subscript first. Thus, for an array with three
subscripts,
read *, A () read *, (((A(i,j,k), i=1,rows), j=1,cols), k=1,planes)

Both languages allow the implied loops to be employed to create an array vector simply by placing
the implied loop inside the standard array delimit symbols. For example, we may want an array to equally
distribute N + 1 points over the distance from zero to D.
F90: X = (/(k,k=0,N)/)* D/(N+1)
M ATLAB: X = [0:N] * D / (N+1),

which illustrates that M ATLAB allows the use of the colon operator to define arrays, but F90 does not.
In addition to locating elements in an array by the regular incrementing of loop variables, both
Fortran90 and M ATLAB support even more specific selections of elements: by random location via vector
subscripts, or by value via logical masks such as where and if in F90 and M ATLAB, respectively.
4.3.3 Conditionals
Logic tests are frequently needed to control the execution of a block of statements. The most basic
operation occurs when we want to do something when a logic test gives a true answer. We call that a
simple IF statement. When the test is true, the program executes the block of statements following the
IF. Often only one statement is needed, so C++ and Fortran allow that one statement to end the line that
begins with the IF logic. Frequently we will nest another IF within the statements from a higher level
IF. The common language syntax forms for the simple IF are given below in Table 4.10, along with the
examples of where a second true group is nested inside the first as shown in Table 4.11.
The next simplest case is where we need to do one thing when the answer is true, and a different
thing when the logic test is false. Then the syntax changes simply to an IF ftrue groupg ELSE
ffalse groupg mode of execution. The typical IF-ELSE syntaxes of the various languages are given
in Table 4.12. Of course, the above statement groups can contain other IF or IF-ELSE statements nested
within them. They can also contain any valid statements, including DO or FOR loops.
The most complicated logic tests occur when the number of cases for the answer go beyond the two
(true-false) of the IF-ELSE control structure. These multiple case decisions can be handled with the IF-
ELSEIF-ELSE control structures whose syntax is given in Table 4.13. They involve a sequence of logic

c 2001 J.E. Akin 61
 M ATLAB Fortran C++

f
if l expression IF (l expression) THEN if (l expression)
true group true group
end END IF true group;
g
IF (l expression) true statement if (l expression)
true statement;

Table 4.10: IF Constructs. The quantity l expression means a logical expression having a value that
is either TRUE of FALSE. The term true statement or true group means that the statement or group
of statements, respectively, are executed if the conditional in the if statement evaluates to TRUE.

M ATLAB Fortran C++

if l expression1 IF (l expression1) THEN if (l expression1)
true group A true group A f
if l expression2 IF (l expression2) THEN true group A
true group B true group B
f
if (l expression2)
end END IF
true group C true group C true group B
end END IF g
statement group D statement group D true group C
g
statement group D

Table 4.11: Nested IF Constructs.

M ATLAB Fortran C++

f
if l expression IF (l expression) THEN if (l expression)
true group A true group A
else ELSE true group A
false group B false group B g
f
end END IF else
false group B
g

Table 4.12: Logical IF-ELSE Constructs.

tests, each of which is followed by a group of statements that are to be executed if, and only if, the test
answer is true. There can be any number of such tests. They are terminated with an ELSE group of default
statements to be executed if none of the logic tests are true. Actually, the ELSE action is optional. For
program clarity or debugging, it should be included even if it only prints a warning message or contains
a comment statement. Typical “if” and “if-else” coding is given in Figs. 4.4, 4.5, and 4.6. Figure 4.4
simply uses the three logical comparisons of “greater than” (line 9), “less than” (line 12), or “equal to”
(line 15), respectively. Figure 4.5 goes a step further by combining two tests with a logical “and” test
(line 9), and includes a second else branch (line 11) to handle the case where the if is false. While the
input to these programs were numbers (line 7), the third example program in Fig. 4.6 accepts logical
input (lines 6,8) that represents either true or false values and carries out Boolean operations to negate
an input (via NOT in line 9), or to compare two inputs (with an AND in line 11, or OR in line 17, etc.) to
produce a third logical value.

Since following the logic of many IF-ELSEIF-ELSE statements can be very confusing both the C++
and Fortran languages allow a CASE selection or “switching” operation based on the value (numerical or
character) of some expression. For any allowed specified CASE value, a group of statements is executed.
If the value does not match any of the specified allowed CASE values, then a default group of statements
are executed. These are illustrated in Table 4.14.

c 2001 J.E. Akin 62
 M ATLAB Fortran C++

f
if l expression1 IF (l expression1) THEN if (l expression1)
true group A true group A
elseif l expression2 ELSE IF (l expression2) THEN true group A
true group B true group B g
f
elseif l expression3 ELSE IF (l expression3) THEN else if (l expression2)
true group C true group C
else ELSE true group B
default group D default group D g
f
end END IF else if (l expression3)
true group C
g
f
else
default group D
g

Table 4.13: Logical IF-ELSE-IF Constructs.

[ 1] program main
[ 2] ! Examples of relational "if" operator in F90
[ 3] implicit none
[ 4] integer :: Integer Var 1, Integer Var 2 ! user inputs
[ 5]
[ 6] print *, ’Enter two integers:’
[ 7] read *, Integer Var 1, Integer Var 2
[ 8]
[ 9] if ( Integer Var 1 > Integer Var 2 ) &
[10] print *, Integer Var 1,’ is greater than ’, Integer Var 2
[11]
[12] if ( Integer Var 1 < Integer Var 2 ) &
[13] print *, Integer Var 1,’ is less than ’, Integer Var 2
[14]
[15] if ( Integer Var 1 == Integer Var 2 ) &
[16] print *, Integer Var 1,’ is equal to ’, Integer Var 2
[17]
[18] end program main
[19]
[20] ! Running with 25 and 4 produces:
[21] ! Enter two integers:
[22] ! 25 is greater than 4

Figure 4.4: Typical Relational Operators in F90

[ 1] program main
[ 2] ! Illustrate a simple if-else logic in F90
[ 3] implicit none
[ 4] integer Integer Var
[ 5]
[ 6] print *,’Enter an integer: ’
[ 7] read *, Integer Var
[ 8]
[ 9] if ( Integer Var > 5 .and. Integer Var < 10 ) then
[10] print *, Integer Var, ’ is greater than 5 and less than 10’
[11] else
[12] print *, Integer Var, ’ is not greater than 5 and less than 10’
[13] end if ! range of input
[14]
[15] end program main
[16] !
[17] ! Running with 3 gives: 3 is not greater than 5 and less than 10
[18] ! Running with 8 gives: 8 is greater than 5 and less than 10

Figure 4.5: Typical If-Else Uses in F90

Fortran90 offers an additional optional feature called construct names that can be employed with the
above IF and SELECT CASE constructs to improve the readability of the program. The optional name,
followed by a colon, precedes the key words IF and SELECT CASE. To be consistent, the name should
also follow the key words END IF or END SELECT which always close the constructs. The construct
name option also is available for loops where it offers an additional pair of control actions that will be
explained later. Examples of these optional F90 features are given in Table 4.15.
While C++ and M ATLAB do not formally offer this option, the same enhancement of readability can

c 2001 J.E. Akin 63
[ 1] program main
[ 2] ! Examples of Logical operators in F90
[ 3] implicit none
[ 4] logical :: Logic Var 1, Logic Var 2
[ 5] print *,’Print logical value of A (T or F):’
[ 6] read *, Logic Var 1
[ 7] print *,’Print logical value of B (T or F):’
[ 8] read *, Logic Var 2
[ 9] print *,’NOT A is ’, (.NOT. Logic Var 1)
[10]
[11] if ( Logic Var 1 .AND. Logic Var 2 ) then
[12] print *, ’A ANDed with B is true’
[13] else
[14] print *, ’A ANDed with B is false’
[15] end if ! for AND
[16]
[17] if ( Logic Var 1 .OR. Logic Var 2 ) then
[18] print *, ’A ORed with B is true’
[19] else
[20] print *, ’A ORed with B is false’
[21] end if ! for OR
[22]
[23] if ( Logic Var 1 .EQV. Logic Var 2 ) then
[24] print *, ’A EQiValent with B is true’
[25] else
[26] print *, ’A EQiValent with B is false’
[27] end if ! for EQV
[28]
[29] if ( Logic Var 1 .NEQV. Logic Var 2 ) then
[30] print *, ’A Not EQiValent with B is true’
[31] else
[32] print *, ’A Not EQiValent with B is false’
[33] end if ! for NEQV
[34]
[35] end program main
[36] ! Running with T and F produces:
[37] ! Print logical value of A (T or F): T
[38] ! Print logical value of B (T or F): F
[39] ! NOT A is F
[40] ! A ANDed with B is false
[41] ! A ORed with B is true
[42] ! A EQiValent with B is false
[43] ! A Not EQiValent with B is true

Figure 4.6: Typical Logical Operators in F90

F90 C++
SELECT CASE (expression) switch (expression)
CASE (value 1) f
group 1 case value 1 :
CASE (value 2) group 1
group 2 break;
. case value 2 :
.
. group 2
CASE (value n) break;
group n .
.
.
CASE DEFAULT
default group case value n :
END SELECT group n
break;
default:
default group
break;
g

Table 4.14: Case Selection Constructs.

be achieved by using the trailing comment feature to append a name or description at the beginning and
end of these logic construct blocks.
Both C++ and Fortran allow statement labels and provide controls to branch to specific labels. Today
you are generally advised not to use a GO TO and its associated label! However, they are common in
many F77 codes. There are a few cases where a GO TO is still considered acceptable. For example, the
pseudo-WHILE construct of F77 requires a GO TO.

c 2001 J.E. Akin 64
 F90 Named IF F90Named SELECT
name: IF (logical 1) THEN name: SELECT CASE (expression)
true group A CASE (value 1)
ELSE IF (logical 2) THEN group 1
true group B CASE (value 2)
ELSE group 2
default group C CASE DEFAULT
ENDIF name default group
END SELECT name

Table 4.15: F90 Optional Logic Block Names.

Fortran C++
DO 1 ... for (...) f
DO 2 ...
...
for (...) f
...
IF (disaster) THEN if (disaster)
GO TO 3 go to error
END IF ...
... g
2 END DO
1 END DO
g
error:
3 next statement

Table 4.16: GO TO Break-out of Nested Loops. This situation can be an exception to the general recom-
mendation to avoid GO TO statements.

F77 F90 C++

DO 1 I = 1,N
f
DO I = 1,N for (i=1; i<n; i++)
... ...
IF (skip condition) THEN IF (skip condition) THEN if (skip condition)
GO TO 1 CYCLE ! to next I continue; // to next
ELSE ELSE else if
false group false group false group
END IF END IF end
1 continue END DO
g

Table 4.17: Skip a Single Loop Cycle.

initialize test
IF (l expression) THEN
true statement group
modify logical value
GO TO #
END IF

The GO TO can also be effectively utilized in both Fortran and C++ to break out of several nested loops.
This is illustrated in Table 4.16. The “break-out” construct can be used in the situation when, as a part of
a subroutine, you wanted the program exit the loop and also exit the subroutine, returning control to the
calling program. To do that, one would simply replace the GO TO statement with the RETURN statement.
In F90, one should also append the comment “! to calling program” to assist in making the subroutine
more readable.
You may find it necessary to want to skip a cycle in loop execution and/or exit from a single loop.
Both Fortran and C++ provide these control options without requiring the use of a GO TO. To skip a loop
cycle, Fortran90 and C++ use the statements CYCLE and continue, respectively, and EXIT and break
to abort a loop. These constructs are shown in Tables 4.17 and 4.18. Other forms of the GO TO in F77
were declared obsolete in F90, and should not be used. The Fortran abort examples could also use the
RETURN option described above in the rare cases when it proves to be more desirable or efficient.
As mentioned earlier, F90 allows the programmer to use “named” DO constructs. In addition to im-

c 2001 J.E. Akin 65
 F77 F90 C++
DO 1 I = 1,N
f
DO I = 1,N for (i=1; i<n; i++)
IF (exit condition) THEN IF (exit condition) THEN
GO TO 2 EXIT ! this do if (exit condition)
ELSE ELSE break;// out of loop
false group false group else if
END IF END IF false group
g
1 CONTINUE END DO end
2 next statement next statement
next statement

Table 4.18: Abort a Single Loop.

main: DO ! forever
test: DO k=1,k max
third: DO m=m max,m min,-1
IF (test condition) THEN
CYCLE test ! loop on k
END IF
END DO third ! loop on m
fourth: DO n=n min,n max,2
IF (main condition) THEN
EXIT main ! forever loop
END DO fourth ! on n
END DO test ! over k
END DO main
next statement

Table 4.19: F90 DOs Named for Control.

proving readability, this feature also offers additional control over nested loops because we can associate
the CYCLE and EXIT commands with a specific loop (Table 4.19). Without the optional name, the CYCLE
and EXIT commands act only on the inner-most loop in which they lie. We will see later that Fortran90
allows another type of loop called WHERE that is designed to operate on arrays.

4.3.3.1 Looping While True or Until True

It is very common to need to perform a loop so long as a condition is true, or to run the loop until
a condition becomes true. The two are very similar and both represent loops that would run forever
unless specifically terminated. We will refer to these two approaches as WHILE loops and UNTIL loops.
The WHILE logic test is made first in order to determine if the loop will be entered. Clearly, this means that
if the logic test is false the first time it is tested, then the statement blocks controlled by the WHILE are
never executed. If the WHILE loop is entered, something in the loop must eventually change the value of
a variable in the logic test or the loop would run forever. Once a change causes the WHILE logic test to be
false control is transferred to the first statement following the WHILE structure. By way of comparison,
an UNTIL loop is always entered at least once. Upon entering the loop, a beginning statement group is
executed. Then the logic test is evaluated. If the test result is true, the loop is exited and control is
passed to the next statement after the group. If the test is false, then an optional second statement group
is executed before the loop returns to the beginning statement group. The pseudo-code for these two
similar structures are given as follows :

while true until true

logic variable = true logic variable = false
begin: begin:
if (logic variable) then % true statements
true group if (logic variable) then
re-evaluate logic variable exit the loop
go to begin else % false
else % false false group
exit loop re-evaluate logic variable
end if go to begin
end if

c 2001 J.E. Akin 66
Since these constructs are commonly needed, several programming languages offer some support for
them. For example, Pascal has a REPEAT UNTIL command and C++ has the DO-WHILE pair for the until-
true construct. For the more common while-true loops, C++ and M ATLAB offer a WHILE command, and
Fortran 90 includes the DO WHILE. F77, however, only has the obsolete IF-GO TO pairs as illustrated in
a previous example. Many current programmers consider the WHILE construct obsolete because it is less
clear than a DO-EXIT pair or a “for-break” pair. Indeed, the F90 standard has declared the DO WHILE
as obsolete and eligible for future deletion from the language. We can see how the loop-abort feature of
C++ and F90 includes both the WHILE and UNTIL concepts. For example, the F90 construct
initialize logical variable
DO WHILE (logical variable) ! is true
true group
re-evaluate logical variable
END DO ! while true
.
.
.

is entirely equivalent to the aborted endless loop

initialize logical variable
DO ! forever while true
IF (.NOT. logical variable) EXIT ! as false
true group
re-evaluate logical variable
END DO ! while true
.
.
.

Likewise, a minor change includes the UNTIL construct.

DO ! forever until true
beginning statements and initialization
IF (logical expression) EXIT ! as true
false group
re-evaluate logical variable
END DO ! until true

When approached in the C++ language, we have the WHILE loop.

initialize logical variable
while (logical variable)
f // is true
true group
re-evaluate logical variable
g // end while true

Recalling the standard for syntax,

for (expr 1; expr 2; expr 3)
f
true group
g // end for

could be viewed as equivalent to the above WHILE in for form.

expr 1;
while (expr 2)
f // is true
true group
expr 3;
g // end while true

If one omits all three for expressions, then it becomes an “infinite loop” or a “do forever” which can
represent a WHILE or UNTIL construct by proper placement of the break command. Furthermore, C has
the do-while construct that is equivalent to Pascal’s REPEAT-UNTIL.
do // forever until true
statements
evaluate logical variable
while (logical variable) // is true

The syntax for the classical WHILE statements in C++, Fortran and M ATLAB are given in Table 4.20.
Fortran90 has declared the DO WHILE as obsolete, and recommends the DO-EXIT pair instead! Using
infinite loops with clearly aborted stages is a less error-prone approach to programming.

c 2001 J.E. Akin 67
 M ATLAB C++
initialize test initialize test
f
while l expression while (l expression)
true group
change test true group
change test
g
end

F77 F90
initialize test initialize test
# continue do while (l expression)
IF (l expression) THEN true group
true group change test
change test end do
go to #
END IF

Table 4.20: Looping While a Condition is True.

Function
M ATLABa C++ Fortran
Type
program statements
f
main(argc,char **argv) program main
[y1...yn]=f(a1,...,am) type y
[end of file] statements type a1,...,type am
statements
g
y = f(a1,I,am);
y = f(a1,...,am)
call s(a1,...,am)
end program

subroutine void f subroutine s(a1,...,am)

(type a1,...,type am) type a1,...,type am
f statements
statements
g
end

function function [r1...rn] type f (type a1,...,type am) function f(a1,...,am)

=f(a1,...,am)
statements f statements g type f
type a1,...,type am
statements
end

a Every function or program in M ATLAB must be in separate files.

Table 4.21: Function definitions. In each case, the function being defined is named f and is called with
m arguments a1,...,am.

4.4 Subprograms
The concept of modular programming requires the use of numerous subprograms or procedures to execute
independent segments of the calculations or operations. Typically, these procedures fall into classes such
as functions, subroutines, and modules. We will consider examples of the procedures for each of our
target languages. These are shown in Table 4.21.
Recall that Table 8.6 compared several intrinsic functions that are common to both F90 and M ATLAB.
For completeness, all of the Fortran90 functions are listed both alphabetically and by subject in Ap-
pendix B. Similar listings for M ATLAB can be found in the M ATLAB Primer.
4.4.1 Functions and Subroutines
Historically, a function was a subprogram that employed one or more input arguments and returned a
single result value. For example, a square root or logarithm function would accept a single input value
and return a single result. All of the languages of interest allow the user to define such a function, and they

c 2001 J.E. Akin 68
 One-Input, One-Result Procedures
M ATLAB function out = name (in)
F90 function name (in) ! name = out
function name (in) result (out)
C++ name (in, out)

Multiple-Input, Multiple-Result Procedures

M ATLAB function [inout, out2] = name (in1, in2, inout)
F90 subroutine name (in1, in2, inout, out2)
C++ name(in1, in2, inout, out2)
 Other arrangements acceptable

Table 4.22: Arguments and return values of subprograms.

all provide numerous intrinsic or built-in functions of this type. As you might expect, such a procedure
is called a function in C++, Fortran and M ATLAB. As an example of such a procedure, consider the
calculation of the mean value of a sequence of numbers defined as

1 X
n
mean = xk :
n
k=1
In Fortran90, a subprogram to return the mean (average) could be
function mean(x)
! mean = sum of vector x, divided by its size
real :: mean, x(:)
mean = sum(x)/size(x)
end function mean

Note that our function has employed two other intrinsic functions: size to determine the number of
elements in the array x, and sum to carry out the summation of all elements in x. Originally in Fortran,
the result value was required to be assigned to the name of the function. That is still a valid option in F90,
but today it is considered better practice to specify a result value name to be returned by the function.
The mean function is a M ATLAB intrinsic and can be used directly.
To illustrate the use of a result value, consider the related “median” value in F90.
function mid value(x) result(median)
! return the middle value of vector x
real :: median, x(:)
median = x(size(x)/2) ! what if size = 1 ??
end function mid value

To apply these two functions to an array, say y, we would simply write y ave = mean(y), and y mid
= mid value(y), respectively. While Fortran allows a “function” to return only a single object, both
C++ and M ATLAB use that subprogram name to return any number of result objects. Fortran employs
the name “subroutine” for such a procedure. Such procedures are allowed to have multiple inputs and
multiple outputs (including none). The syntax of the first line of these two subprogram classes are shown
in Table 4.22. Note that a typical subprogram may have no arguments, multiple input arguments (in1,
in2, inout), multiple result arguments (inout, out2), and arguments that are used for both input
and result usage (inout). These example names have been selected to reflect the fact that a programmer
usually intends for arguments to be used for input only, or for result values only, or for input, modification,
and output. It is considered good programming practice to declare such intentions to aid the compiler in
detecting unintended uses. F90 provides the INTENT statement for this purpose, but does not require its
use.
Having outlined the concepts of subprograms, we will review some presented earlier and then give
some new examples. Figure 1.3 presented a clipping function which was earlier expressed in pseudocode.
A corresponding Fortran implementation of such a clipping function is given in Fig. 4.7. Note that it is
very similar to the pseudocode version.

c 2001 J.E. Akin 69
[ 1] program main
[ 2] ! clip the elements of an array
[ 3] implicit none
[ 4] real, parameter :: limit = 3
[ 5] integer, parameter :: n = 5
[ 6] real :: y(n), x(n)
[ 7] ! Define x values that will be clipped
[ 8] x = (/ (-8. + 3.*k, k = 1,n) /) ! an implied loop
[ 9] do i = 1, n
[10] y(i) = clip (x(i), limit)
[11] end do
[12] print *, x
[13] print *, y
[14]
[15] contains ! methods
[16]
[17] function clip (x, L) result (c)
[18] ! c = clip(x, L) - clip the variable x, output
[19] ! x = scalar variable, input
[20] ! L = limit of the clipper, input
[21] !
[22] real, intent(in) :: x, L ! variable types
[23] real :: c ! variable types
[24] intent (in) x, L ! argument uses
[25] if ( abs(x) <= L ) then ! abs of x less than or equal L
[26] c = x; ! then use x
[27] else ! absolute of x greater than L ?
[28] c = sign(L,x) ! sign of x times L
[29] end if ! of value of x
[30] end function ! clip
[31] end program main
[32] !
[33] ! produces:
[34] ! -5.0000000 -2.0000000 1.0000000 4.0000000 7.0000000
[35] ! -3.0000000 -2.0000000 1.0000000 3.0000000 3.0000000

Figure 4.7: Clipping a Set of Array Values in F90

For the purpose of illustration an alternate F90 version of the Game of Life, shown earlier in Chapter 1
as pseudocode, is given in the assignment solutions section. Clearly we have not introduced all the
features utilized in these example codes so the reader should continue to refer back to them as your
programming understanding grows.
A simple program that illustrates program composition is maximum.f90, which asks the user to
specify several integers from which the program finds the largest. It is given in Fig. 4.8. Note how
the main program accepts the user input (lines 15,20), with the maxint function (line 22) finding the
maximum (lines 25-34). Perhaps modularity would have been better served by expressing the input
portion by a separate function. Of course, this routine is not really needed since F90 provides intrinsic
functions to find maximum and minimum values (maxval, minval) and their locations in any array
(maxloc, minloc). A similar C++ program composition is shown for comparison in the appendix.

c 2001 J.E. Akin 70
[ 1] program maximum ! of a set of integers (see intrinsic maxval)
[ 2] implicit none
[ 3] interface ! declare function interface protype
[ 4] function maxint (input, input length) result(max)
[ 5] integer, intent(in) :: input length, input(:)
[ 6] integer :: max
[ 7] end function ! maxint
[ 8] end interface
[ 9]
[10] integer, parameter :: ARRAYLENGTH=100
[11] integer :: integers(ARRAYLENGTH);
[12] integer :: i, n;
[13]
[14] ! Read in the number of integers
[15] print *,’Find maximum; type n: ’; read *, n
[16] if ( n > ARRAYLENGTH .or. n < 0 ) &
[17] stop ’Value you typed is too large or negative.’
[18]
[19] do i = 1, n ! Read in the user’s integers
[20] print *, ’Integer ’, i, ’?’; read *, integers(i)
[21] end do ! over n values
[22] print *, ’Maximum: ’, maxint (integers, n)
[23] end program maximum
[24]
[25] function maxint (input, input length) result(max)
[26] ! Find the maximum of an array of integers
[27] integer, intent(in) :: input length, input(:)
[28] integer :: i, max
[29]
[30] max = input(1); ! initialize
[31] do i = 1, input length ! note could be only 1
[32] if ( input(i) > max ) max = input(i);
[33] end do ! over values
[34] end function maxint ! produces this result:
[35] ! Find maximum; type n: 4
[36] ! Integer 1? 9
[37] ! Integer 2? 6
[38] ! Integer 3? 4
[39] ! Integer 4? -99
[40] ! Maximum: 9

Figure 4.8: Search for Largest Value in F90

c 2001 J.E. Akin 71
 Global Variable Declaration
M ATLAB global list of variables
F77 common /set name/ list of variables
F90 module set name
save
type (type tag) :: list of variables
end module set name
C++ extern list of variables

Access to Global Variables

M ATLAB global list of variables
F77 common /set name/ list of variables
F90 use set name, only subset of variables
use set name2 list of variables
C++ extern list of variables

Table 4.23: Defining and referring to global variables.

4.4.2 Global Variables

We have seen that variables used inside a procedure can be thought of as dummy variable names that
exist only in the procedure, unless they are members of the argument list. Even if they are arguments to
the procedure, they can still have names different from the names employed in the calling program. This
approach can have disadvantages. For example, it might lead to a long list of arguments, say 20 lines,
in a complicated procedure. For this and other reasons, we sometimes desire to have variables that are
accessible by any and all procedures at any time. These are called global variables regardless of their
type.
Generally, we explicitly declare them to be global and provide some means by which they can be
accessed, and thus modified, by selected procedures. When a selected procedure needs, or benefits from,
access to a global variable, one may wish to control which subset of global variables are accessible by the
procedure. The typical initial identification of global variables and the ways to access them are shown in
Table 4.23, respectively.
An advanced aspect of the concept of global variables are the topics of inheritance and object-oriented
programming. Fortran90, and other languages like C++, offer these advanced concepts. In F90, inheri-
tance is available to a module and/or a main program and their “internal sub-programs” defined as those
procedures following a contains statement, but occurring before an end module or the end program
statement. Everything that appears before the contains statement is available to, and can be changed by,
the internal sub-programs. Those inherited variables are more than local in nature, but not quite global;
thus, they may be thought of as territorial variables. The structure of these internal sub-programs with
inheritance is shown in Fig. 4.9
Perhaps the most commonly used global variables are those necessary to calculate the amount of
central processor unit (cpu) time, in seconds, that a particular code segment used during its execution.
All systems provide utilities for that purpose but some are more friendly than others. M ATLAB provides
a pair of functions, called tic and toc, that act together to provide the desired information. To illustrate
the use of global variables we will develop a F90 module called tic toc to hold the necessary variables
along with the routines tic and toc. It is illustrated in Fig. 4.10 where the module constants (lines 2-6)
are set (lines 17, 26) and computed (line 28) in the two internal functions.

c 2001 J.E. Akin 72
 module or program name inherit
Optional territorial variable, type specification, and calls
contains

subroutine Internal 1
territorial specifications and calls
contains

subroutine Internal 2
local computations
end subroutine Internal 2

subroutine Internal 3
local computations
end subroutine Internal 3

end subroutine Internal 1

end name inherit

Figure 4.9: F90 Internal Subprogram Structure.

[ 1] module tic toc

[ 2] ! Define global constants for timing increments
[ 3] implicit none
[ 4] integer :: start ! current value of system clock
[ 5] integer :: rate ! system clock counts/sec
[ 6] integer :: finish ! ending value of system clock
[ 7] real :: sec ! increment in sec, (finish-start)/rate
[ 8] ! Useage: use tic toc ! global constant access
[ 9] ! call tic ! start clock
[10] ! . . . ! use some cpu time
[11] ! cputime = toc () ! for increment
[12] contains ! access to start, rate, finish, sec
[13] subroutine tic
[14] ! -------------------------------------------------
[15] ! Model the matlab tic function, for use with toc
[16] ! -------------------------------------------------
[17] implicit none
[18] call system clock ( start, rate ) ! Get start value and rate
[19] end subroutine tic
[20]
[21] function toc ( ) result(sec)
[22] ! -------------------------------------------------
[23] ! Model the matlab toc function, for use with tic
[24] ! -------------------------------------------------
[25] implicit none
[26] real :: sec
[27] call system clock ( finish ) ! Stop the execution timer
[28] sec = 0.0
[29] if ( finish >= start ) sec = float(finish - start) / float(rate)
[30] end function toc
[31] end module tic toc

Figure 4.10: A Module for Computing CPU Times

c 2001 J.E. Akin 73
 Action C++ F90
Bitwise AND & iand
Bitwise exclusive OR ^ ieor
Bitwise exclusive OR j ior
Circular bit shift ishftc
Clear bit ibclr
Combination of bits mvbits
Extract bit ibits
Logical complement  not
Number of bits in integer sizeof bit size
Set bit ibset
Shift bit left  ishft
Shift bit right  ishft
Test on or off btest
Transfer bits to integer transfer

Table 4.24: Bit Function Intrinsics.

4.4.3 Bit Functions

We have discussed the fact that the digital computer is based on the use of individual bits. The subject of
bit manipulation is one that we do not wish to pursue here. However, advanced applications do sometimes
require these abilities, and the most common uses have been declared in the so-called military standards
USDOD-MIL-STD-1753, and made part of the Fortran90 standard. Several of these features are also a
part of C++. Table 4.24 gives a list of those functions.

4.4.4 Exception Controls

An exception handler is a block of code that is invoked to process specific error conditions. Standard
exception control keywords in a language are usually associated with the allocation of resources, such
as files or memory space, or input/output operations. For many applications we simply want to catch an
unexpected result and output a message so that the programmer can correct the situation. In that case we
may not care if the exception aborts the execution. However, if one is using a commerical execute only
program then it is very distubing to have a code abort. We would at least expect the code to respond to a
fatal error by closing down the program in some gentle fashion that saves what was completed before the
error and maybe even offer us a restart option. Here we provide only the minimum form of an exceptions
module that can be used by other modules to pass warnings of fatal messages to the user. It includes an
integer flag that can be utilized to rank the severity of possible messages. It is shown in Fig. 4.11. Below
we will summarize the F90 optional error flags that should always be checked and are likely to lead to a
call to the exception handler.
Dynamic Memory: The ALLOCATE and DEALLOCATE statements both use the optional flag STAT = to
return an integer flag that can be tested to invoke an exception handler. The integer value is zero after
a successful (de)allocation, and a positive value otherwise. If STAT = is absent, an unsuccessful result
stops execution.
File Open/Close: The OPEN, CLOSE, and ENDFILE statements allow the use of the optional keyword
IOSTAT = to return an integer flag which is zero if the statement executes successfully, and a positive
value otherwise. They also allow the older standard exception keyword ERR = to be assigned a positive
integer constant label number of the statement to which control is passed if an error occurs. An exception
handler could be called by that statement.
File Input/Output: The READ, WRITE, BACKSPACE, and REWIND statements allow the IOSTAT =
keyword to return a negative integer if an end-of-record (EOR) or end-of-file (EOF) is encountered, a
zero if there is no error, and a positive integer if an error occurs (such as reading a character during an

c 2001 J.E. Akin 74
[ 1] module exceptions
[ 2] implicit none
[ 3] integer, parameter :: INFO = 1, WARN = 2, FATAL = 3
[ 4] integer :: error count = 0
[ 5] integer :: max level = 0
[ 6] contains
[ 7]
[ 8] subroutine exception (program, message, flag)
[ 9] character(len=*) :: program
[10] character(len=*) :: message
[11] integer, optional :: flag
[12]
[13] error count = error count + 1
[14]
[15] print *, ’Exception Status Thrown’
[16] print *, ’ Program :’, program
[17] print *, ’ Message :’, message
[18] if ( present(flag) ) then
[19] print *, ’ Level :’, flag
[20] if ( flag > max level ) max level = flag
[21] end if ! flag given
[22] end subroutine exception
[23]
[24] subroutine exception status ()
[25] print *
[26] print *, "Exception Summary:"
[27] print *, " Exception count = ", error count
[28] print *, " Highest level = ", max level
[29] end subroutine exception status
[30] end module exceptions

Figure 4.11: A Minimal Exception Handling Module

integer input). They also allow the ERR = error label branching described above for the file open/close
operations.
In addition, the READ statement also retains the old standard keyword END = to identify a label number
to which control transfers when an end-of-file (EOF) is detected.
Status Inquiry: Whether in UNIT mode or FILE mode, the INQUIRE statement for file operations allows
the IOSTAT = and ERR = keywords like the OPEN statement. In addition, either mode supports two logical
keywords : EXISTS = to determine if the UNIT (or FILE) exists, and OPENED = to determine if a (the)
file is connected to this (an) unit.
Optional Arguments: The PRESENT function returns a logical value to indicate whether or not an
optional argument was provided in the invocation of the procedure in which the function appears.
Pointers and Targets: The ASSOCIATED function returns a logical value to indicate whether a pointer
is associated with a specific target, or with any target.

4.5 Interface Prototype

Compiler languages are more efficient than interpreted languages. If the compiler is going to correctly
generate calls to functions, or subprograms, it needs to know certain things about the arguments and
returned values. The number of arguments, their type, their rank, their order, etc. must be the same. This
collection of information is called the “interface” to the function, or subprogram. In most of our example
codes the functions and subprograms have been included in a single file. In practice they are usually
stored in separate external files, and often written by others. Thus, the program that is going to use these
external files must be given a “prototype” description of them. In other words, a segment of prototype,
or interface, code is a definition that is used by the compiler to determine what parameters are required
by the subprogram as it is called by your program. The interface prototype code for any subprogram can
usually be created by simply copying the first few lines of the subprogram (and maybe the last one) and
placing them in an interface directory.
To successfully compile a subprogram modern computer science methods sometimes require the pro-
grammer to specifically declare the interface to be used in invoking a subprogram, even if that subprogram
is included in the same file. This information is called a “prototype” in C and C++, and an “interface”
in F90. If the subprogram already exists, one can easily create the needed interface details by making

c 2001 J.E. Akin 75
a copy of the program and deleting from the copy all information except that which describes the argu-
ments and subprogram type. If the program does not exist, you write the interface first to define what
will be expected of the subprogram regardless of who writes it. It is considered good programming style
to include explicit interfaces, or prototype code, even if they are not required.
If in doubt about the need for an explicit interface see if the compiler gives an error because it is not
present. In F90 the common reasons for needing an explicit interface are: 1) Passing an array that has
only its rank declared. For example, A(:,:), B(:). These are known as “assumed-shape” arrays; 2)
Using a function to return a result that is: a) an array of unknown size, or b) a pointer, or c) a character
string with a dynamically determined length. Advanced features like optional argument lists, user defined
operators, generic subprogram names (to allow differing argument types) also require explicit operators.
In C++ before calling an external function, it must be declared with a prototype of its parameters.
The general form for a function is
function type function name ( argument type list);

where the argument type list is the comma separated list of pairs of type and name for each
argument of the function. These names are effectively treated as comments, and may be different from
the names in the calling program, or even omitted. The use of a prototype was shown in Fig. 4.8 and is
used again in Fig. 4.12 which also illustrates passing arguments by reference or by value.
An interface block for external subprograms was not required by F77 (thereby leading to hard to find
errors), but is strongly recommended if F90 and is explicitly required in several situations. The general
form for a F90 interface is

interface interface name

function interface body
subroutine interface body
module procedure interface body
end interface interface name

where a typical function interface body would be

function type function name (argument name list) result ( name )

implicit none
argument type, intent class :: name list
end function function name

where the argument name list is the comma separated list of names. Of course, the function type
refers to the result argument name. These names may be different from the names in the calling program.
A typical subroutine interface body would be

subroutine subroutine name (argument name list)

implicit none
argument type, intent class :: name list
end subroutine subroutine name

where the argument name list is the comma separated list of names. The topic of a module procedure is
covered elsewhere. The use of a interface block was shown in Fig. 4.8 and used in two new codes, shown
in Fig. 4.12, and the corresponding C++ code in the appendix, which also illustrate passing arguments by
reference (line 23) and by value (line 19) in both F90 and C++. The important, and often confusing, topic
of passing by reference or value was discussed in Sec. 4.2 and is related to other topics to be considered
later, such as the use of “pointers” in C++ and F90, and the “intent” attribute of F90 arguments. Passing
by reference is default in F90 while passing by value is default in C++ .

4.6 Characters and Strings

All of our example languages offer convenient ways to manipulate and compare strings of characters.
The characters are defined by one of the international standards such as ASCII, which is usually used
on UNIX, or the EBCDIC set. These contain both printable and non-printable (control) characters. On
a UNIX system, the full set can be seen with the command man ascii. In the 256-character ASCII
set, the upper case letters begin at character number 65, ‘A’, and the corresponding lower case values are

c 2001 J.E. Akin 76
[ 1] program main
[ 2] implicit none
[ 3] ! declare the interface prototypes
[ 4] interface
[ 5] subroutine Change (Refer)
[ 6] integer :: Refer; end subroutine Change
[ 7] subroutine No Change (Value)
[ 8] integer :: Value; end subroutine No Change
[ 9] end interface
[10]
[11] ! illustrate passing by reference and by value in F90
[12]
[13] integer :: Input Val, Dummy Val
[14]
[15] print *, "Enter an integer: "
[16] read *, Input Val; print *, "Input value was ", Input Val
[17]
[18] ! pass by value
[19] call No Change ( (Input Val) ) ! Use but do not change
[20] print *, "After No Change it is ", Input Val
[21]
[22] ! pass by reference
[23] call Change ( Input Val ) ! Use and change
[24] print *, "After Change it is ", Input Val
[25] end program
[26]
[27] subroutine Change (Refer)
[28] ! changes Refer in calling code IF passed by reference
[29] integer :: Refer
[30] Refer = 100;
[31] print *, "Inside Change it is set to ", Refer
[32] end subroutine Change
[33]
[34] subroutine No Change (Value)
[35] ! does not change Value in calling code IF passed by value
[36] integer :: Value
[37] Value = 100;
[38] print *, "Inside No Change it is set to ", Value
[39] end subroutine No Change
[40]
[41] ! Running gives:
[42] ! Enter an integer: 12
[43] ! Input value was 12
[44] ! Inside No Change it is set to 100
[45] ! After No Change it is 12
[46] ! Inside Change it is set to 100
[47] ! After Change it is 100

Figure 4.12: Passing Arguments by Reference and by Value in F90

32 positions higher (character 97 is ‘a’). These printable characters begin at character 32, as shown in
Table 4.25 for the ASCII standard. The first 33 characters are “non-printing” special control characters.
For example, NUL = null, EOT = end of transmission, BEL = bell, BS = backspace, and HT = horizontal
tab. To enter a control character, one must simultaneously hold down the CONTROL key and hit the letter
that is 64 positions higher in the list. That is, an end of transmission EOT is typed as CONTROL-D. The
code SP denotes the space character, and we will use the underscore “ ” to represent a blank in strings.
We can employ the standard relational operators (e.g., less than) to compare strings and would find
that ’bad’ < ’dog’ < ’same’ == ’same ’, that ’word’ > ’WORD’, and that ’four’ < ’one’
< ’two’ while ’1’ < ’2’ < ’4’. Note that the above equality occurred because trailing blanks are not
considered in relational operations, but leading blanks are considered: ’same’ 6= ’ same’. The F90
function adjustL removes leading blanks and appends them to the right end. Thus, it adjusts the string
to the left, so that ’same’ == adjustL(’ same’). This and other F90 intrinsic character functions
are summarized in Table 4.26.
All blanks are considered when determining the length of a character string. In F90 the intrinsic
function LEN provides these data so that LEN(’same’) = 4, LEN(’ same’) = 6, and LEN(’same ’)
= 7. There is another intrinsic function, LEN TRIM, that provides the string length ignoring trail-
ing blanks. By way of comparison: LEN TRIM(’same’) = 4, LEN TRIM(’ same’) = 6, and
LEN TRIM(’same ’) = 4. Each character in a string or any internal substrings may be referenced
by the colon operator. Given a character variable we can define a substring, say sub as
sub = variable(K:L) for 0 < K,L <= LEN(variable)
= null for K > L

c 2001 J.E. Akin 77
 0 NUL 1 SOH 2 STX 3 ETX 4 EOT 5 ENQ 6 ACK 7 BEL
8 BS 9 HT 10 NL 11 VT 12 NP 13 CR 14 SO 15 SI
16 DLE 17 DC1 18 DC2 19 DC3 20 DC4 21 NAK 22 SYN 23 ETB
24 CAN 25 EM 26 SUB 27 ESC 28 FS 29 GS 30 RS 31 US
32 SP 33 ! 34 " 35 # 36 $ 37 % 38 & 39 ’
40 ( 41 ) 42 * 43 + 44 , 45 - 46 . 47 /
48 0 49 1 50 2 51 3 52 4 53 5 54 6 55 7
56 8 57 9 58 : 59 ; 60 < 61 = 62 > 63 ?
64 @ 65 A 66 B 67 C 68 D 69 E 70 F 71 G
72 H 73 I 74 J 75 K 76 L 77 M 78 N 79 O
80 P 81 Q 82 R 83 S 84 T 85 U 86 V 87 W
88 X 89 Y 90 Z 91 [ 92 \ 93 ] 94 ˆ 95 _
96 ‘ 97 a 98 b 99 c 100 d 101 e 102 f 103 g
104 h 105 i 106 j 107 k 108 l 109 m 110 n 111 o
112 p 113 q 114 r 115 s 116 t 117 u 118 v 119 w
120 x 121 y 122 z 123 { 124 | 125 } 126 ˜ 127 DEL

Table 4.25: The ACSII Character Set

ACHAR (I) Character number I in ASCII collating set

ADJUSTL (STRING) Adjust left
ADJUSTR (STRING) Adjust right
CHAR (I)  Character I in processor collating set
IACHAR (C) Position of C in ASCII collating set
ICHAR (C) Position of C in processor collating set
INDEX (STRING, SUBSTRING)a Starting position of a substring
LEN (STRING) Length of a character entity
LEN TRIM (STRING) Length without trailing blanks
LGE (STRING A, STRING B) Lexically greater than or equal
LGT (STRING A, STRING B) Lexically greater than
LLE (STRING A, STRING B) Lexically less than or equal
LLT (STRING A, STRING B) Lexically less than
REPEAT (STRING, NCOPIES) Repeated concatenation
SCAN (STRING, SET)a Scan a string for a character in a set
TRIM (STRING) Remove trailing blank characters
VERIFY (STRING, SET)a Verify the set of characters in a string
STRING A//STRING B Concatenate two strings
a Optional arguments not shown.

Table 4.26: F90 Character Functions

= error for K or L > LEN(variable).

For example, given the string ’howl’, then we can define bird = string(2:4) = ’owl’, and prep
= string(1:3) = ’how’.
The F90 and F77 operator used to concatenate strings into larger strings is “//”. Continuing the last
example, we see that the concatenation string(1:3)//’ ’//string(2:4)//’?’ is ’how owl?’,
while the concatenation ’same ’//’word’ becomes ’same word’ and ’bad’//’ ’//’dog’
becomes ’bad dog’. Programs illustrating the reading and concatenating two strings are given in
Fig. 4.13, and in the companion C++ code in the appendix.
Sometimes one needs to type in a non-printing character, such as a tab or a newline. To allow this,
special transmissions have been allowed for, as summarized in Table 4.27.
Remember the ASCII character features: the uppercase letters correspond to numbers 65 through 90
in the list, while the lowercase letters are numbers 97 through 122, so that if we wanted to convert “G” to

c 2001 J.E. Akin 78
[ 1] program main
[ 2] ! Compare two strings
[ 3] ! Concatenate two character strings together
[ 4] ! Get the combined length
[ 5] implicit none
[ 6] character(len=20) :: String1, String2
[ 7] character(len=40) :: String3
[ 8] integer :: length
[ 9]
[10] print *,’Enter first string (20 char max):’
[11] read ’(a)’, String1 ! formatted
[12]
[13] print *,’Enter second string (20 char max):’
[14] read ’(a)’, String2 ! formatted
[15]
[16] ! compare
[17] if ( String1 == String2 ) then
[18] print *, "They are the same."
[19] else
[20] print *, "They are different."
[21] end if
[22]
[23] ! concatenate
[24] String3 = trim (String1) // trim (String2)
[25]
[26] print *,’The combined string is:’, String3
[27] length = len trim (String3)
[28] print *,’The combined length is:’, length
[29]
[30] end program main
[31] ! Running with "red" and "bird" produces:
[32] ! Enter first string (20 char max): red
[33] ! Enter second string (20 char max): bird
[34] ! They are different.
[35] ! The combined string is: redbird
[36] ! The combined length is: 7
[37] ! Also "the red" and "bird" works

Figure 4.13: Using Two Strings in F90

Action ASCII Character F90 Inputa C++ Input

Alert (Bell) 7 Ctrl-G na
Backspace 8 Ctrl-H nb
Carriage Return 13 Ctrl-M nr
End of Transmission 4 Ctrl-D Ctrl-D
Form Feed 12 Ctrl-L nf
Horizontal Tab 9 Ctrl-I nt
New Line 10 Ctrl-J nn
Vertical Tab 11 Ctrl-K nv
a “Ctrl-” denotes control action. That is, simultaneous pressing of the CONTROL key and the letter following.

Table 4.27: How to type non-printing characters.

“g” we could use commands such as:

character (len = 1) :: lower_g, UPPER_G
lower_g = achar(iachar(’G’) + 32)

or visa versa:
UPPER_G = achar(iachar(’g’) - 32)

since they differ by 32 locations. Likewise, since the zero character “0” occurs in position 48 of the
ASCII set we could convert a single digit to the same numerical value with:
integer :: number_5
number_5 = iachar(’5’) - 48

and so forth for all ten digits. To convert a string of digits, such as ’5623’, to the corresponding number
5623, we could use a looping operation.

c 2001 J.E. Akin 79
[ 1] program main
[ 2] ! Convert a character string to an integer in F90
[ 3] implicit none
[ 4] character(len=5) :: Age Char
[ 5] integer :: age
[ 6]
[ 7] print *, "Enter your age: "
[ 8] read *, Age Char ! a character string
[ 9]
[10] ! convert using an internal file read
[11] read (Age Char, fmt = ’(i5)’) age ! convert to integer
[12]
[13] print *, "Your integer age is ", age
[14] print ’(" Your binary age is ", b8)’, age
[15] print ’(" Your hexadecimal age is ", z8)’, age
[16] print ’(" Your octal age is ", o8)’, age
[17]
[18] end program main
[19] !
[20] ! Running gives:
[21] ! Enter your age: 45
[22] ! Your integer age is 45
[23] ! Your binary age is 101101
[24] ! Your hexadecimal age is 2D
[25] ! Your octal age is 55

Figure 4.14: Converting a String to an Integer with F90

character (len = 132) :: digits

integer :: d_to_n, power, number
! Now build the number from its digits
if (digits == ’ ’) then
print *, ’warning, no number found’
number = 0
else
number = 0
k = len_trim(digits)
do m = k, 1, -1 ! right to left
d_to_n = iachar(digits(m:m)) - 48
power = 10**(k-m)
number = number + d_to_n*power
end do ! over digits
print *, ’number = ’, number

However, since loops can be inefficient, it is better to learn that, in F90, an “internal file” can be (and
should be) employed to convert one data type to another. Here we could simply code:
! internal file called convert
write(convert, ‘‘(A)’’) digit
read(convert, ‘‘(I4)’’) number

to convert a character to an integer (or real) number. Converting strings to integers is shown in the codes
given in Fig. 4.14 (line 11) and the corresponding C++ appendix routine. Similar procedures would be
used to convert strings to reals. The C++ version (see appendix) uses the intrinsic function “atoi” while
the F90 version uses an internal file for the conversion.
One often finds it useful to change the case of a string of characters. Some languages provide intrinsic
functions for that purpose. In C++ and M ATLAB the function to convert a string to all lower case letters
are called tolower and lower, respectively. Here we define a similar F90 function called to lower
which is shown in Fig. 4.15 along with a testing program in Fig. 4.16. Note that the testing program
uses an interface to tolower (lines 4-13) assuming that routine was compiled and stored external to the
testing program. The tolower function employs the intrinsic function index (line 16) to see if the k-th
character of the input string is an upper case letter. The intrinsic function len is also used (line 8) to
force the new string to be the same length as the original string.

4.7 User Defined Data Types

Variables, as in mathematics, represent some quantity; unlike mathematics, many languages force the
programmer to define what type the variable is. Generic kinds of type are integer, floating point (single,
double, and quadruple precision), and complex-valued floating point. Table 4.2 (page 53) presents the
data types inherent in the various languages. Most beginning programmers find the requirement most

c 2001 J.E. Akin 80
[ 1] function to lower (string) result (new string) ! like C
[ 2] ! -------------------------------------------------------------
[ 3] ! Convert a string or character to lower case
[ 4] ! (valid for ASCII or EBCDIC processors)
[ 5] ! -------------------------------------------------------------
[ 6] implicit none
[ 7] character (len = *), intent(in) :: string ! unknown length
[ 8] character (len = len(string)) :: new string ! same length
[ 9] character (len = 26), parameter :: &
[10] UPPER = ’ABCDEFGHIJKLMNOPQRSTUVWXYZ’, &
[11] lower = ’abcdefghijklmnopqrstuvwxyz’
[12] integer :: k ! loop counter
[13] integer :: loc ! position in alphabet
[14] new string = string ! copy everything
[15] do k = 1, len(string) ! to change letters
[16] loc = index ( UPPER, string(k:k)) ! first upper
[17] if (loc /= 0 ) new string(k:k) = lower(loc:loc) ! convert it
[18] end do ! over string characters
[19] end function to lower

Figure 4.15: Converting a String to Lower Case with F90

[ 1] program up down ! test character case inversion functions

[ 2] implicit none
[ 3] character (len = 24) :: test=’ABCDefgh1234abcdZYXWzyxw’
[ 4]
[ 5] interface
[ 6] function to lower (string) result (new string)
[ 7] character (len = *), intent(in) :: string
[ 8] character (len = len(string)) :: new string
[ 9] end function to lower
[10] function to upper (string) result (new string)
[11] character (len = *), intent(in) :: string
[12] character (len = len(string)) :: new string
[13] end function to upper
[14] end interface
[15]
[16] print *, test
[17] print *, to lower (test)
[18] print *, to upper (test)
[19] end program ! running gives
[20] ! ABCDefgh1234abcdZYXWzyxw
[21] ! abcdefgh1234abcdzyxwzyxw
[22] ! ABCDEFGH1234ABCDZYXWZYXW

Figure 4.16: Testing String Conversions with F90

languages impose of defining explicitly each variable’s type to be tedious, unnecessary, and a source of
bugs. It’s tedious because the programmer must think not only about what the variable represents, but
also how the computations calculate its value, unnecessary because mathematics doesn’t work that way
(the variable x represents a quantity regardless whether it turns out to be an integer or a complex value),
and bug-creating because computations involving different types and assigned to a typed variable can
yield nonmathematical results (for example, dividing the integers 1 with 3 and assigning the results to an
integer yields a zero value).
M ATLAB is one language in which variables are not explicitly typed. (Beginning programmers
cheer!) Internally, M ATLAB represents numbers in double precision floating point. If a variable’s value
corresponds to an integer, M ATLAB will gleefully print it that way, effectively hiding its floating point
representation. A surprise occurs when a calculation accidentality becomes complex: M ATLAB will
(silently) change what the variable represents from being real to being complex. For example, M ATLAB
will, without complaint, calculate x=log(-1) and assign the value 3:14159i to x. In many applications,
the expression that yielded the value of 1 because of an error, and M ATLAB will let the error propagate.
(Beginning programmers sigh!) Most, if not all typed languages will immediately announce the evalua-
tion of the logarithm of a negative number, and halt execution. By explicitly defining the kinds of values
a variable will assume helps programming clarity and run-time debugging to some degree.
C++ has four intrinsic (i.e., built-in) types of data integer, single and double precision reals, and
character and F90 has the similar set: integer, real, complex, logical, and character. F90 also allows
the user to create a specific precision level for integer and real data. C++ has specified byte sizes for
three character, six integer, one single precision real, and two double precision real data types for a total
of twelve intrinsic data types.

c 2001 J.E. Akin 81
 C, C++ Variable.component.sub component
F90 Variable%component%sub component

Table 4.28: Referencing Defined Data Type Structure Components.

C, C++ struct data tag f

intrinsic type 1 component names;
intrinsic type 2 component names;
g ;

F90 type data tag

intrinsic type 1 :: component names;
intrinsic type 2 :: component names;
end type data tag

Table 4.29: Defining New Types of Data Structure

C, C++ struct data tag f

intrinsic type 1 component names;
g
struct tag 2 component names;
;

F90 type data tag

intrinsic type :: component names;
type (tag 2) :: component names;
end type data tag

Table 4.30: Nested Data Structure Definitions.

In addition to intrinsic types, C, C++ and F90 allow the formation of new types of
data structures that are collections of values of not necessarily the same type. These procedures
are named struct or type in C and F90, respectively.
To go along with this freedom, F90 allows you to define new operations to act on the derived types.
While C++ retains the struct keyword, it is viewed as a class with only public data members and no
functions. In other words, in C++ class is a generalization of struct and, thus, class is the preferred
keyword to use. As an example of a task made easier by derived data, consider creating parts of a data
structure to be used in an address book. We will need a variable that can have components and sub-
components. They are referenced by a special syntax and defined as illustrated in Tables 4.28 and 4.29.
This procedure for defining a new type of data structure can be “nested” by referring to other derived
type entities defined earlier in the program. These concepts are shown in Table 4.30. One should declare
the data type of all variables used in a program module. This is also true for user defined data structures.
Table 4.31 outlines the forms of these statements, how structures are initialized, and how component
values are assigned.
There are times when either the derived type variable or its components, or both, could be subscripted
objects (i.e., arrays). Then care must be taken in the interpretation of which variable or component is
being addressed. Table 4.32 illustrates the typical combinations with the F90 syntax.
As a concrete example, consider a phone type and address type definition.

c 2001 J.E. Akin 82
 C, C++ struct data tag variable list; /* Definition */
f g
struct data tag variable = component values ; /* Initialization */
variable.component.sub component = value; /* Assignment */

F90 type (data tag) :: variable list ! Definition

variable = data tag (component values) ! Initialization
variable%component%sub component = value ! Assignment

Table 4.31: Declaring, initializing, and assigning components of user-defined datatypes.

INTEGER, PARAMETER :: j max = 6

TYPE meaning demo
INTEGER, PARAMETER :: k max = 9, word = 15
CHARACTER (LEN = word) :: name(k max)
END TYPE meaning demo
TYPE (meaning demo) derived(j max)
Construct Interpretation
derived All components of all derived’s elements
derived(j) All components of jth element of derived
derived(j)%name All k max components of name within jth element of derived
derived%name(k) Component k of the name array for all elements of derived
derived(j)%name(k) Component k of the name array of jth element of derived

Table 4.32: F90 Derived Type Component Interpretation.

F90 C++
type phone type struct phone type f
g
integer :: area code, number, extension int area code, number, extension;
end type phone type ;
type address type
integer :: number
struct address type f
int number;
character (len = 35) :: street, city char street[35], city[35];
character (len = 2) :: state char state[2];
integer :: zip code
g
int zip code;
end type address type ;

These could be used to define part of a person type

F90 C++
type person type struct person type f
character (len = 50) :: name char name[50];
type (phone type) :: phone struct phone type phone;
type (address type) :: address struct address type address;
g
integer :: born year int born year;
end type person type ;

We define two people with

F90 C++
type (person type) :: sammy, barney struct person type sammy, barney;

or build an address book array filled with the above data structures by defining

c 2001 J.E. Akin 83
 F90
integer, parameter :: number = 99
type (person type), dimension (number) :: address book

C++
#define NUMBER 99
struct person type address book[NUMBER];

and then initialize, or “construct” sammy’s phone and zip code as

F90 C++
sammy%phone = phone type (713, 5278100, 0) f
sammy.phone = 713, 5278100, 0 ; g
sammy%zip code = 770051892 sammy.zip code = 770051892;

and print them with

F90 C++
,
,
print sammy%phone printf("(%d)%d, extension %d",
print sammy%address%zip code sammy.area code,
sammy.number,
sammy.extension);
printf("%d", sammy.zip code);

and then define specific members for barney with the “constructor”

F90 C++
barney = person type("Barn Owl", & f
barney = "Barn Owl", 0,0,0 , f g
phone type(0,0,0), & sammy.address, 1892,
sammy%address, 1892, "Sammy’s cousin") "Sammy’s cousin" ; g

Note the difference in the defined type constructors. Two are actually used here because the second com-
ponent must be defined as a phone type. C++ just uses brackets to enclose the supplied components
of each user defined type. F90 has an intrinsic functon that is created automatically by the type definition
and it accepts all of the components required by the type. That is why the function name “phone type”
appears in the intrinsic constructor routine “person type”. Finally, put them in the book.

F90 C++
address book(1) = sammy address book[1] = sammy;
address book(2) = barney address book[2] = barney;

Fig. 4.17 presents a sample code for utilizing user defined structure types using F90 (there is a C++
version in the appendix). First a “person” structure is created (lines 4-7) by using only the intrinsic
types of integers and characters. It then is used in turn within an additional data structure (line 10). The
components of the structures are read (lines 18, 21, 24) and output (lines 26,27). For more general data,
suggested in the comments, formatted input/output controls would be necessary.

4.7.1 Overloading Operators

As a complete short example of utilizing many of the new programming features that come with user
defined data structures we will consider the use of a familiar old mathematics system, fractions. Recall
that a fraction is the ratio of two integers. We will therefore define a new data type called Fraction. It

c 2001 J.E. Akin 84
[ 1] program main ()
[ 2] ! Define structures and components, via F90
[ 3] implicit none
[ 4] type Person ! define a person structure type
[ 5] character (len=20) :: Name
[ 6] integer :: Age
[ 7] end type Person
[ 8]
[ 9] type Who Where ! use person type in a new structure
[10] type (Person) :: Guest
[11] character (len=40) :: Address
[12] end type Who Where
[13]
[14] ! Fill a record of the Who Where type components
[15] type (Who Where) Record;
[16]
[17] print *,"Enter your name: "
[18] read *, Record % Guest % Name
[19]
[20] print *,"Enter your city: "
[21] read *, Record % Address
[22]
[23] print *,"enter your age: "
[24] read *, Record % Guest % Age
[25]
[26] print *,"Hello ", Record % Guest % Age, " year old ", &
[27] Record % Guest % Name, " in ", Record % Address
[28]
[29] end program main
[30]
[31] ! Running with input: Sammy, Houston, 104 gives
[32] ! Hello 104 year old Sammy in Houston
[33] !
[34] ! But try: Sammy Owl, Houston, 104 for a bug

Figure 4.17: Using Multiple Structures in F90

will simply consist of two integer types, named num and denom, respectively. New data types can be
defined in any program unit. For maximum usefulness we will place the definition in a module named
Fractions. To use this new data type we will want to have subprograms to define a fraction, list its
components, and multiply two fractions together, and to equate one fraction to another. In addition to
the intrinsic constructor function fraction we will create a manual constructor function called assign
and it will have two arguments, the numerator value, and denominator value, and will use them to return
a fraction type. The listing subroutine, called list Fraction, simply needs the name of the fraction
to be printed. The function, mult Fraction, accepts two fraction names, and returns the third fraction
as their product. Finally, we provide a function that equates the components of one fraction to those in a
new fraction.
This data structure is presented in Fig. 4.18. There we note that the module starts with the definition
of the new data type (lines 2-4), and is followed with the “contains” statement (line 12). The subpro-
grams that provide the functionality of the fraction data type follow the “contains” statement and are thus
coupled to the definition of the new type. When we have completed defining the functionality to go with
the new data type we end the module.
In this example the program to invoke the fraction type follows in Fig. 4.19. To access the module,
which defines the new data type and its supporting functions, we simply employ a “use” statement at the
beginning of the program (line 2). The program declares three Fraction type variables (line 3): x, y ,
and z . The variable x is defined to be 22/7 with the intrinsic type constructor (line 5), while y is assigned
a value of 1/3 by using the function assign (line 7). Both values are listed for confirmation. Then we
form the new fraction, z = 22=21, by invoking the mult Fraction function (line 9),
z = mult Fraction (x, y)

which returns z as its result. A natural tendency at this point would be to simply write this as z = x  y .
However, before we could do that we would have to tell the operators, “*” and ”=”, how to act when
provided with this new data type. This is known as overloading an intrinsic operator. We had the foresight
to do this when we set up the module by declaring which of the “module procedure”s were equivalent
to each operator symbol. Thus from the “interface operator (*)” statement block the system now knows
that the left and right operands of the “*” symbol correspond to the first and second arguments in the

c 2001 J.E. Akin 85
[ 1] module Fractions ! F90 "Fraction" data structure and functionality
[ 2] implicit none
[ 3] type Fraction ! define a data structure
[ 4] integer :: num, den ! with two "components"
[ 5] end type Fraction
[ 6]
[ 7] interface operator (*) ! extend meaning to fraction
[ 8] module procedure mult Fraction ; end interface
[ 9]
[10] interface assignment (=) ! extend meaning to fraction
[11] module procedure equal Fraction ; end interface
[12]
[13] contains ! functionality
[14] subroutine assign (name, numerator, denominator)
[15] type (Fraction), intent(inout) :: name
[16] integer, intent(in) :: numerator, denominator
[17]
[18] name % num = numerator ! % denotes which "component"
[19] if ( denominator == 0 ) then
[20] print *, "0 denominator not allowed, set to 1"
[21] name % den = 1
[22] else; name % den = denominator
[23] end if ; end subroutine assign
[24]
[25] subroutine list(name)
[26] type (Fraction), intent(in) :: name
[27]
[28] print *, name % num, "/", name % den ; end subroutine list
[29]
[30] function mult Fraction (a, b) result (c)
[31] type (Fraction), intent(in) :: a, b
[32] type (Fraction) :: c
[33]
[34] c%num = a%num * b%num ! standard = and * here
[35] c%den = a%den * b%den ; end function mult Fraction
[36]
[37] subroutine equal Fraction (new, name)
[38] type (Fraction), intent(out) :: new
[39] type (Fraction), intent(in) :: name
[40]
[41] new % num = name % num ! standard = here
[42] new % den = name % den ; end subroutine equal Fraction
[43] end module Fractions

Figure 4.18: Overloading operations for new data types

function mult Fraction. Likewise, the left and right operands of “=” are coupled to the first and
second arguments, respectively, of subroutine equal Fraction. The testing main and verification
results are in Fig. 4.19 Before moving on note that the system does not yet know how to multiply a
integer times a fraction, or visa versa. To do that one would have to add more functionality, such as
a function, say int mult frac, and add it to the ”module procedure” list associated with the “*”
operator.
When considering which operators to overload for a newly defined data type one should consider
those that are used in sorting operations, such as the greater-than, >, and less-than, <, operators. They
are often useful because of the need to sort various types of data. If those symbols have been correctly
overloaded then a generic sorting routine might be used, or require few changes.
4.7.2 User Defined Operators
In addition to the many intrinsic operators and functions we have seen so far, the F90 user can also define
new operators or extend existing ones. User defined operators can employ intrinsic data types and/or user
defined data types. The user defined operators, or extensions, can be unary or binary (i.e., have one or
two arguments). The operator symbol must be included between two periods, such as ‘.op.’. Specific
examples will be given in the next chapter.

4.8 Pointers and Targets

The beginning of every data item must be stored in computer memory at a specific address. The address
of that data item is called a pointer to the data item, and a variable that can hold such an address is called
a pointer variable. Often it is convenient to have a pointer to a variable, an array, or a sub-array. F90,
C++ and M ATLAB provide this sophisticated feature. The major benefits of the use of pointers is that

c 2001 J.E. Akin 86
[ 1] program main
[ 2] use Fractions
[ 3] implicit none
[ 4] type (Fraction) :: x, y, z
[ 5]
[ 6] x = Fraction (22,7) ! default constructor
[ 7] write (*,’("default x = ")’, advance=’no’) ; call list(x)
[ 8] call assign(y,1,3) ! manual constructor
[ 9] write (*,’("assigned y = ")’, advance=’no’) ; call list(y)
[10] z = mult Fraction (x,y) ! function use
[11] write (*,’("x mult y = ")’, advance=’no’) ; call list(z);
[12] print *, "Trying overloaded * and = for fractions:"
[13] write (*,’("y * x gives ")’, advance=’no’) ; call list(y*x) ! multi
[14] z = x*y ! new operator uses
[15] write (*,’("z = x*y gives ")’, advance=’no’) ; call list(z) ! add
[16] end program main ! Running gives:
[17] ! default x = 22/7 ! assigned y = 1/3 ! x mult y = 22/21
[18] ! Trying overloaded * and = for fractions:
[19] ! y * x gives 22/21 ! z = x*y gives 22/21

Figure 4.19: Testing overloading for new data types

C++ F90
Declaration type tag *pointer name; type (type tag), pointer ::
pointer name
Target &target name type (type tag), target :: target name
Examples char *cp, c; character, pointer :: cp
int *ip, i; integer, pointer :: ip
float *fp, f; real, pointer :: fp
cp = & c; cp => c
ip = & i; ip => i
fp = & f; fp => f

Table 4.33: Definition of pointers and accessing their targets.

they allow dynamic data structures, such as “linked lists” and “tree structures,” and they allow recursive
algorithms. Note that rather than containing data themselves, pointer variables simply exist to point
to where some data are stored. Unlike C and M ATLAB the F90 pointers are more like the “reference
variables” of the C++ language in that they are mainly an alias or synonym for another variable, or part
of another variable. They do not allow one to easily get the literal address in memory as does C. This is
why programmers that write computer operating systems usually prefer C over F90. But F90 pointers
allow easy access to array partitions for computational efficiency, which C++ does not. Pointers are often
used to pass arguments by reference.
The item to which a pointer points is known as a target variable. Thus, every pointer has a logical
status associated with it which indicates whether or not it is currently pointing to a target. The initial
value of the association is .false., or undefined.
4.8.1 Pointer Type Declaration
For every type of data object that can be declared in the language, including derived types, a correspond-
ing type of pointer and target can be declared (Table 4.33).
While the use of pointers gives programmers more options for constructing algorithms, they also have
a potential severely detrimental effect on the program execution efficiency. To ensure that compilers can
produce code that execute efficiently, F90 restricts the variables, to which a pointer can point, to those
specifically declared to have the attribute target. This, in part, makes the use of pointers in F90 and
C++ somewhat different. Another major difference is that C++ allows arithmetic to be performed on the
pointer address, but F90 does not.
So far, we have seen that F90 requires specific declarations of a pointer and an potential target.
However, C++ employs two unary operators, & and *, to deal with pointers and targets, respectively.
Thus, in C++ the operator &variable name means “the address of” variable name, and the C++
operator *pointer name means “the value at the address of” pointer name.

c 2001 J.E. Akin 87
 C, C++ pointer name = NULL
F90 nullify (list of pointer names)
F95 pointer name = NULL()

Table 4.34: Nullifying a pointer to break target association.

[ 1] program pt expression
[ 2] !
[ 3] ! F90 example of using pointers in expressions
[ 4] implicit none
[ 5] integer, POINTER :: p, q, r
[ 6] integer, TARGET :: i = 1, j = 2, k = 3
[ 7]
[ 8] q => j ! q points to integer j
[ 9] p => i ! p points to integer i
[10] !
[11] ! An expression that "looks like" pointer arithmetic
[12] ! automatically substitutes the target value:
[13] !
[14] q = p + 2 ! means: j = i + 2 = 1 + 2 = 3
[15] print *, i, j, k ! print target values
[16] p => k ! p now points to k
[17] print *, (q-p) ! means print j - k = 3 - 3 = 0
[18] !
[19] ! Check associations of pointers
[20] print *, associated (r) ! false
[21] r => k ! now r points to k, also
[22] print *, associated (p,i) ! false
[23] print *, associated (p,k) ! true
[24] print *, associated (r,k) ! true
[25] end program pt expression

Figure 4.20: Using F90 Pointers in Expressions.

4.8.2 Pointer Assignment

F90 requires that a pointer be associated with a target by a single pointer assignment statement.
C allows, but does not require, a similar statement. (See Table 4.33). After such a statement,
the pointer has a new association status and one could employ the F90 intrinsic inquiry function
associated(pointer name, target name) to return .true. as the logical return value. If one
wishes to break or nullify a pointer’s association with a target, but not assign it another target, one can
nullify the pointer as shown in Table 4.34.
4.8.3 Using Pointers in Expressions
The most important rule about using pointers in F90 expressions is that, where ever a pointer occurs,
it is treated as its associated target. That is, the target is automatically substituted for the pointer when
the pointer occurs in an expression. For example, consider the actions in Fig. 4.20 (where the results are
stated as comments).
4.8.4 Pointers and Linked Lists
Pointers are the simplest available mechanism for dynamic memory management of arrays such as stacks,
queues, trees, and linked lists. These are extraordinarily flexible data structures because their size can
grow or shrink during the execution of a program. For linked lists the basic technique is to create a
derived type that consists of one or more data elements and at least one pointer. Memory is allocated to
contain the data and a pointer is set to reference the next occurrence of data. If one pointer is present, the
list is a singly-linked list and can only be traversed in one direction: head to tail, or vice versa. If two
pointers are present: the list is a doubly-linked list and can be traversed in either direction. Linked lists
allow the data of interest to be scattered all over memory and uses pointers to weave through memory,
gathering data as required. Detailed examples of the use of linked lists are covered in Chapter 8.
As a conceptual example of when one might need to use linked-lists think of applications where
one never knows in advance how many data entries will be needed. For example, when a surveyor
determines the exact perimeter of a building or plot of land, critical measurements are taken at each

c 2001 J.E. Akin 88
angle. If the perimeter has N sides, the surveyor measures the length of each side and the interior angle
each side forms with the next. Often the perimeter has visual obstructions and offsets around them must
be made, recorded, and corrected for later. Regardless of how careful the surveyor is, errors are invariably
introduced during the measurement process. However, the error in angle measurements can be bounded.
The program for implementing the recording and correcting of the angles in a survey could be written
using a singly linked list. A linked list is chosen because the programmer has no idea how many sides
the perimeter has, and linked lists can grow arbitrarily. Because of the linked list’s ability to absorb a
short or long data stream, the user does not have to be asked to count the number of legs in the traverse.
The program begins by declaring a derived type that contains one angle measurement and a pointer to
the next measurement. A count is kept of the number of legs in this loop and the forward pointer for the
last angle read is cleared (set to null) to signal the end of list. After all the data are read, the entire list of
angles is reviewed to get the total of the measurements. This starts by revisiting the head of the list and
adding together all the angle measurements until a null pointer is encountered, signaling the end of list.
Then the error can be computed and distributed equally among the legs of the traverse.

4.9 Accessing External Source Files and Functions

At times one finds it necessary, or efficient to utilize other software from libraries, other users, or different
paths in your directories. Of course, you could always use the brute force approach and use a text editor
to copy the desired source code into your program. However, this is unwise not only because it wastes
storage, but more importantly gives multiple copies of a module that must all be found and changed if
future revisions are needed or desired. Better methods of accessing such codes can be defined either
inside your program, or external to it in the “linking” phase after compiling has been completed.
High level languages like C, C++, and F90 allow one or more approaches for accessing such soft-
ware from within your code. One feature common to all these languages is the availability of an “include”
statement which gives the system path to the desired code file. At compile time, and only then, a tem-
porary copy of the indicated code from that file is literally copied and inserted into your program at the
location of the corresponding “include” statement.
It is common practice, but not required, to denote such code fragments with name extensions of “.h”
and ”.inc”, in C++ and F90, respectively. For example, to use a program called “class Person” one
could insert the following statement in your program:
C, C++: include <class Person.h>
F90 : include ’class Person.inc’

if the files, class Person.h or class Person.inc, were in the same directory as your program. Otherwise,
it is necessary to give the complete system path to the file, such as,
include ‘/home/caam211/Include/inv.f90’
include ‘/home/caam211/Include/SolveVector.f90’

which give source links to the caam211 course files for the function inv(A) for returning the inverse
of a matrix A, and the function SolveVector(A,B) which returns the solution vector X for the matrix
system A*X = B.
In F90 one can also provide a “module” that defines constants, user defined types, supporting sub-
programs, operators, etc. Any of those features can be accessed by first including such a F90 module
before the main program and later invoking it with a “use” statement which cites the “module” name. For
example, the F90 program segments:
include ‘/home/caam211/Include/caam211 operators.f90’
Program Lab2 A 2
. . .
call test matrix ( A, B, X ) ! form and invert test matrix
. . .
subroutine test matrix ( A, B, X )
use caam211 operators ! included above
implicit none
real :: A(:,:), B(:), X(:)
real :: A inv(size(A,1),size(A,1)) ! automatic array allocation
A inv = inv(A)
X = A .solve. B ! like X = A B in Matlab n
. . .

c 2001 J.E. Akin 89
gives a source link to the caam211 course “module” source file named caam211 operators.f90
which contains subprograms, such as the function inv(), and operator definitions like .solve. which
is equivalent to the “\” operator in M ATLAB.
In the last example the omission of the “include” statement would require a compiler dependent state-
ment to allow the system to locate the module cited in the “use” statement. For the National Algorithms
Group (NAG) F90 compiler that link would be given as
f90 -o go /home/caam211/Include/caam211 operators.f90 my.f90

if the above segment was stored in the file named my.f90, while for the Cray F90 compiler a path flag,
-p, to the compiled version is required, such as:
f90 -o go -p /home/caam211/Include/caam211 op CRAY.o my.f90

Either would produce an executable file, named “go” in this example.

4.10 Procedural Applications

In this section we will consider two common examples of procedural algorithms: fitting curves to exper-
imental data, and sorting numbers, strings, and derived types. Sorting concepts will be discussed again
in Chapter 7.

4.10.1 Fitting Curves to Data

We must ofter deal with measurements and what they result in: data. Measurements are never exact
because they are limited by instrument sensitivity and are contaminated by noise. To determine trends
(how measurements are related to each other), confirm theoretical predictions, and the like, engineers
must frequently fit functions to data. The “curve” fit is intended to be smoother than a raw plot of the
data, hopefully revealing more about the underlying relation between the variables than would otherwise
be apparent.
Often, these functions take parametric form: The functional form is specified, but has unknown
coefficients. Suppose you want to fit a straight line to a dataset. With y denoting the measurement and
x the independent variable, we wish to fit the function y = f (x) = mx + b to the data. The fitting
process amounts to determining a few quantities of the assumed linear functional form the parameters
m and b from the data. You know that two points define a straight line; consequently, only two of the
(x; y ) pairs need be used. But which two should be used? In virtually all real-world circumstances, the
measurements do not precisely conform to the assumed functional form. Thus, fitting a curve by selecting
a few values (two in the linear case) and solving for the function’s parameters produces a circumspect
“fit”, to say the least. Instead, the most common approach is to use all the data in the curve fitting
process. Because you frequently have much more data than parameters, you have what is known as an
over-determined problem. In most cases, no parameter values produce a function that will fit all the
data exactly. Over-determined problems can be solved by specifying an error criterion (what is an error
and how large is the deviation of data from the assumed curve) and finding the set of parameter values
that minimizes the error criterion. With this approach, we can justifiably claim to have found the best
parameter choices.

The “Least Squares” Approach

Far and away the most common error criterion is the mean-squared error: Given measurement pairs

2
(xi ; yi ), i = 1; : : : ; N , the mean squared error 2 equals the average across the dataset of yi f (xi ) ,
the squared error between the ith measurement and the assumed parametric function f (xi ).

1 X
N
2
2 = yi f (xi )
N
i=1
Least squares fitting of functions to data amounts to minimizing the dataset’s mean squared error with
respect to the parameters.

c 2001 J.E. Akin 90
 To illustrate the least-squares approach, let’s fit a linear function to a dataset. Substituting the assumed
functional form f (x) = mx + b into the expression for the mean-squared error, we have

1 X
N
2
2
 = yi (mx + b) i
N
i=1
We can find a set of equations for the parameters m and b that minimize this quantity by evaluating the
derivative of 2 with respect to each parameter and setting each to zero.

d2 1 X
N

dm
=
N
2x i yi (mx + b) = 0i
i=1
d2
N
1 X

db
=
N
2 y i i
(mx + b) = 0
i=1
After some simplification, we find that we have two linear equations to solve for the fitting parameters.
N ! N ! N
1 X 2 1 X 1 X
m
N
xi +b
N
xi =
N
xi yi
i=1 i=1! i=1
1 X
N XN
m
N
xi +b =
1
N
yi
i=1 i=1
Thus, finding the least-squares fit of a straight line to a set of data amounts to solving a set of two linear
equations, the coefficients of which are computed from the data. Note that the four summations in the
last equation have the same range count (N) and could be evaluated in a single explicit loop.
An Aside
Because fitting data with a linear equation yields a set of two easily solved equations for the parameters,
one approach to fitting nonlinear curves to data is to convert the nonlinear problem into a linear one. For
example, suppose we want to fit a power law to the data: f (x) = axb . Instead of minimizing the mean
squared error directly, we transform the data so that we are fitting it with a linear curve. In the power
law case, the logarithm of the fitting curve is linear in the parameters: log f (x) = log a + b log x. This
equation is not linear in the parameter a. For purposes of least-squares fits, we instead treat a0 = log a
as the linear fit parameter, solve the resulting set of linear equations for a0 , and calculate a = exp a0 to
determine the power law fitting parameter. By evaluating the logarithm of xi and yi and applying the least
squares equations governing the fitting of a linear curve to data, we can fit a power-law function to data.
Thus, calculating a linear least squares fit to data underlies general approximation of measurements by
smooth curves. For an insight to the types of relationships that can be determined, see the following
summary.
x-axis y -axis Relationship
Linear Linear y = mx + b linear
Linear Logarithmic log y = mx + b exponential: y = eb
emx
Logarithmic Linear y = m log x + b logarithmic
Logarithmic Logarithmic log y = m log x + b power-law: y = eb
xm

We can now specify the computations required by the least squares fitting algorithm mathematically.
Algorithm: Least-Squares Fitting of Straight Lines to Data
1. Given N pairs of data points (xi ; yi )
PN 1 P N x, 1 P N x,
2. Calculatey a11 = N1 2
i=1 xi , a12 = N i=1 i a21 = N i=1 i a22 = 1, c1 =
1 PN 1 PN y .
N i=1 xi yi , and c2 = N i=1 i
y Note that these calculations can be performed in one loop rather than four.

c 2001 J.E. Akin 91
 3. Solve the set of linear equations
    
a11 a12 m c1
=
a21 a22 b c2

which for two equations can be done by hand to yield

m = (a12 c2


N c1 )=(a12 a21
N
a11 )
b = (c2

m a12 )=N

4. Calculate the mean squared error 2 1 P N y
2

= N i=1 i i
(mx + b) .

Implementing the Least Squares Algorithm

In F90, such calculations can be performed two different ways: one expresses the looping construct
directly, the other uses more efficient intrinsic array routines inside F90. Assuming the fxi g are stored in
the vector x, the coefficient a12 can be calculated (at least) two ways.
1. sum x = 0
N = size(x)
do i = 1,N
sum x = sum x + x(i)
end do
a12 = sum x/N
2. a12 = sum(x)/size(x)

Clearly, the second method produces a somewhat simpler expression than the first, and is vastly superior
to the first. In the sample code that follows in Fig. 4.21 we use the intrinsic array functions but encourage
the reader to check the results with a single loop that computes all six terms need to find m and b.
There are a few new features demonstrated in this example code. In line 6 we have specified a fixed
unit number to associate with the data file to be specified by the user. But we did not do an INQUIRE to
see if that unit was already in use. We will accept a user input filename (lines 8, 25 and 28) that contains
the data to be fitted. An interface (lines 12-21) is provided to external routines that will determine the
number of lines of data in the file and the read those data into the two arrays. Those two routines are given
elsewhere. Of course, the memory for the data arrays must be dynamically allocated (line 35) before they
can be read (line 37). After the least squares fit is computed (line 40) and printed the memory space for
the data is freed (line 44).
In the lsq fit subroutine (line 47) the three items of interest are passed in the array fit. (Routine
lsq fit could have been written as a function, try it.) Observe that y must be the same length as array
x so the size intrinsic was used to ensure that (line 56). The data summations are evaluated with the sum
intrinsic (lines 62-64) and it is used again to evaluate the mean squared error mse (line 72) as described
in step 4 of the algorithm. The test data (lines 78-89) and results (lines 92-96) are given as comments as
usual. Since no explicit loops have been used this form would be more efficient on vector computers and
some parallel computers.
4.10.2 Sorting
One of the most useful computational routines is sorting: Ordering a sequence of data according to some
rule. For example, the alphabetized list of filenames producted by a system directory command is far
easier to read than an unsorted list would be. Furthermore, data can be fruitfully sorted in more than one
way. As an example, you can sort system files by their creation date.
Sorting algorithms have been well studied by computer scientists in a quest to find the most efficient.
We use here the bubble sort algorithm, perhaps the oldest, but not most efficient. This algorithm makes
multiple passes over a list, going down the list interchanging adjacent elements in the list if needed to
put them in order. For example, consider the list [b, e, a, d, f, c], shown in Fig. 4.22, that we

c 2001 J.E. Akin 92
[ 1] program linear fit
[ 2] ! ------------------------------------------------------
[ 3] ! F90 linear least-squares fit on data in file
[ 4] ! specified by the user.
[ 5] ! ------------------------------------------------------
[ 6] implicit none
[ 7] integer, parameter :: filenumber = 1 ! RISKY
[ 8] real, allocatable :: x(:), y(:) ! data arrays
[ 9] character (len = 64) :: filename ! name of file to read
[10] integer :: lines ! number of input lines
[11] real :: fit(3) ! final results
[12]
[13] interface
[14] function inputCount(unit) result(linesOfInput)
[15] integer, intent(in) :: unit ! file unit number
[16] integer :: linesOfInput ! result
[17] end function inputCount
[18] subroutine readData (inFile, lines, x, y)
[19] integer, intent(in) :: inFile, lines ! file unit, size
[20] real, intent(out) :: x(lines), y(lines) ! data read
[21] end subroutine readData
[22] end interface
[23]
[24] ! Get the name of the file containing the data.
[25] write (*,*) ’Enter the filename to read data from:’
[26] read (*,’(A64)’) filename
[27]
[28] ! Open that file for reading.
[29] open (unit = filenumber, file = filename)
[30]
[31] ! Find the number of lines in the file
[32] lines = inputCount (filenumber)
[33] write (*,*) ’There were ’,lines,’ records read.’
[34]
[35] ! Allocate that many entries in the x and y array
[36] allocate (x(lines), y(lines))
[37]
[38] call readData (filenumber, lines, x, y) ! Read data
[39] close (filenumber)
[40]
[41] call lsq fit (x, y, fit) ! least-squares fit
[42] print *, "the slope is ", fit(1) ! display the results
[43] print *, "the intercept is ", fit(2)
[44] print *, "the error is ", fit(3)
[45] deallocate (y, x)
[46] contains
[47]

Fig. 4.21, A Typical Least Squares Linear Fit Program (continued)

wish to sort to alphabetical order. In the first pass, the algorithm begins by examining the first two list
elements (b, e). Since they are in order, these two are left alone. The next two elements (e, a)
are not in order; these two elements of the list are interchanged. In this way, we “bubble” the element a
toward the top and e toward the bottom. The algorithm proceeds through the list, interchanging elements
if need be until the last element is reached. Note that the bottom of the list at the end of the first pass
contains the correct entry. This effect occurs because of the algorithm’s structure: The “greatest” element
will always propagate to the list’s end. Once through the pass, we see that the list is in better, but not
perfect, order. We must perform another pass just like the first to improve the ordering. Thus, the second
pass need consider only the first n 1 elements, the third n 2, etc. The second pass does make the
list better formed. After more passes, the list eventually becomes sorted. To produce a completely sorted
list, the bubble-sort algorithm requires no more passes than the number of elements in the list minus one.
The following F90 routines illustrate some of the initial features of a simple procedural approach
to a simple process like the bubble-sort algorithm. We begin by considering the sorting of a list of real
numbers as shown in subroutine Sort Reals in Fig. 4.22.
In line 1 we have passed in the size of the array, and the actual array (called database). Note that the
database has intent (inout) because we plan to overwrite the original database with the newly sorted order,
which is done in lines 18–20. For efficiency sake we have included an integer counter, swaps Made,
so that we can determine if the sort has terminated early. If we wished to apply the same bubble-sort
algorithm to an integer array all we would have to do is change the procedure name and lines 6 and 10
that describe the type of data being sorted (try it).

c 2001 J.E. Akin 93
[48] subroutine lsq fit (x, y, fit)
[49] ! ------------------------------------------------------
[50] ! Linear least-squares fit, A u = c
[51] ! ------------------------------------------------------
[52] ! fit = slope, intercept, and mean squared error of fit.
[53] ! lines = the length of the arrays x and y.
[54] ! x = array containing the independent variable.
[55] ! y = array containing the dependent variable data.
[56] implicit none
[57] real, intent(in) :: x(:), y(size(x))
[58] real, intent(out) :: fit(3)
[59] integer :: lines
[60] real :: m, b, mse
[61] real :: sumx, sumx2, sumy, sumxy
[62]
[63] ! Summations
[64] sumx = sum ( x ) ; sumx2 = sum ( x**2 )
[65] sumy = sum ( y ) ; sumxy = sum ( x*y )
[66]
[67] ! Calculate slope intercept
[68] lines = size(x)
[69] m = (sumx*sumy - lines*sumxy)/(sumx**2 - lines*sumx2)
[70] b = (sumy - m*sumx)/lines
[71]
[72] ! Predicted y points and the sum of squared errors.
[73] mse = sum ( (y - m*x - b)**2 )/lines
[74] fit(1) = m ; fit(2) = b ; fit(3) = mse ! returned
[75] end subroutine lsq fit
[76]
[77] end program linear fit
[78]
[79] ! Given test set 1 in file lsq 1.dat:
[80] ! -5.000000 -2.004481
[81] ! -4.000000 -1.817331
[82] ! -3.000000 -1.376481
[83] ! -2.000000 -0.508725
[84] ! -1.000000 -0.138670
[85] ! 0.000000 0.376678
[86] ! 1.000000 0.825759
[87] ! 2.000000 1.036343
[88] ! 3.000000 1.815817
[89] ! 4.000000 2.442354
[90] ! 5.000000 2.636355
[91] ! Running the program yields:
[92] !
[93] ! Enter the filename to read data from: lsq 1.dat
[94] ! There were 11 records read.
[95] ! the slope is 0.4897670746
[96] ! the intercept is 0.2988743484
[97] ! the error is 0.2139159478E-01

Figure 4.21: A Typical Least Squares Linear Fit Program

That is true because the compiler knows how to apply the > operator to all the standard numerical
types in the language. But what if we want to sort character strings, or other types of objects? Fortran has
lexical operators (like LGE) to deal with strings, but user defined objects would require that we overload
the > operator, if the expected users would not find the overloading to be confusing. In other words, you
could develop a fairly general sort routine if we changed lines 6 and 10 to be
[ 6] type (Object), intent(inout) :: database (lines)
[10] type (Object) :: temp

and provided an overloading of > so that line 17 makes sense for the defined Object (or for selected
component of it).
To illustrate the sort of change that is necessary to sort character strings consider subroutine
Sort String Fig. 4.23:
To keep the same style as the previous algorithm and overload the > operator we would have to have a
procedure that utilizes the lexical operators in lines 24 and 25, along with the interface definition on lines
12 through 17, do define the meaning of > in the context of a string. While the concept of a “template”
for a code to carry out a bubble-sort on any list of objects it may not always be obvious what > means
when it is overloaded by you or some other programmer.
Note that in the two above sorting examples we have assumed that we had the authority to change the
original database, and that it was efficient to do so. Often that is not the case. Imagine the case where
the database represents millions of credit card users, each with a large number components of numbers,

c 2001 J.E. Akin 94
 Pass 1
b b b b b b
? e a a a a
e
a a? e? d d d
d d d e e e
f f f f? f? c
c c c c c f

Pass 2
b a a a a
?
a b? b b b
d d d d d
?
e e e e c
c c c c? e
f f f f f

Figure 4.22: Example passes of the bubble-sort algorithm through data.

[ 1] subroutine Sort Reals (lines, database)

[ 2] ! Bubble Sort of (changed) Real Database
[ 3]
[ 4] implicit none
[ 5] integer, intent(in) :: lines ! number of records
[ 6] real, intent(inout) :: database (lines) ! records in database
[ 7]
[ 8] integer :: swaps Made ! number of swaps made in one pass
[ 9] integer :: count ! loop variable
[10] real :: temp ! temporary holder for making swap
[11]
[12] do ! Repeat this loop forever... (until we break out of it)
[13] swaps Made = 0 ! Initially, we’ve made no swaps
[14] ! Make one pass of the bubble sort algorithm
[15] do count = 1, (lines - 1)
[16] ! If item is greater than the one after it, swap them
[17] if ( database (count) > database (count + 1) ) then
[18] temp = database (count)
[19] database (count) = database (count + 1)
[20] database (count + 1) = temp
[21] swaps Made = swaps Made + 1
[22] end if
[23] end do
[24] ! If we made no swaps, break out of the loop.
[25] if ( swaps Made == 0 ) exit ! do count swaps
[26] end do
[27] end subroutine Sort Reals

Figure 4.23: Bubble Sort of a Real Array

character strings, or general objects. If many workers are accessing those data for various sorting needs
you probably would not allow the original dataset to be changed for reasons of safety or security. Then
we consider an alternative to moving around the actual database components. That is, we should consider
using moving pointers to large data components, or pseudo-pointers such as an ordering array. The use
of an ordering array is shown in Fig. 4.24 where subroutine Integer Sort now includes an additional
argument.
The third argument has intent (out), as shown in line 7, and is an integer array of the same length
as the original database which has now been changed to intent (in) so the compiler will not allow us to
change the original data. If the data are properly sorted as supplied then it should not be changed and the
new order should be the same as the original sequential input. That is why line 13 initializes the return
order to a sequential list. Then we slightly change the previous sort logic so that lines 19 through 23 now
check whats in an ordered location, and change the order number when necessary, but never change the
original data. After exiting this routine you could list the information, in sorted order, without changing
the original data simply by using vector subscripts in a print statement like:
print *, database (order).

c 2001 J.E. Akin 95
[ 1] subroutine Sort String (lines, database)
[ 2] ! Bubble Sort of (Changed) String Database
[ 3] implicit none
[ 4]
[ 5] integer, intent(in) :: lines ! input size
[ 6] character(len=*), intent(inout) :: database (lines) ! records
[ 7]
[ 8] character (len = len(database (1))) :: temp ! swap holder
[ 9] integer :: swaps Made ! number of swaps in a pass
[10] integer :: count ! loop variable
[11]
[12] interface ! to lower
[13] function to lower (string) result (new String)
[14] character (len = *), intent(in) :: string
[15] character (len = len(string)) :: new String
[16] end function to lower
[17] end interface ! to lower
[18]
[19] do ! Repeat this loop forever... (until we break out of it)
[20] swaps Made = 0 ! Initially, we’ve made no swaps
[21] ! Make one pass of the bubble sort algorithm
[22] do count = 1, (lines - 1)
[23] ! If the element is greater than the one after it, swap them
[24] if ( LGT (to lower (database (count )),
[25] to lower (database (count + 1))) ) then
[26] temp = database (count )
[27] database (count ) = database (count + 1)
[28] database (count + 1) = temp
[29] swaps Made = swaps Made + 1
[30] end if
[31] end do
[32] ! If we made no swaps, berak out of the loop.
[33] if ( swaps Made == 0) exit ! do count swaps
[34] end do
[35] end subroutine Sort String

Figure 4.24: Bubble Sort of an Array of Character Strings

[ 1] subroutine Integer Sort (lines, database, order)

[ 2] ! Ordered Bubble Sort of (Unchanged) Integer Database
[ 3]
[ 4] implicit none
[ 5] integer, intent(in) :: lines ! number of records
[ 6] integer, intent(in) :: database (lines) ! records in database
[ 7] integer, intent(out) :: order (lines) ! the order array
[ 8]
[ 9] integer :: swaps Made ! number of swaps made in one pass
[10] integer :: count ! loop variable
[11] integer :: temp ! temporary holder for making swap
[12]
[13] order = (/ (count, count = 1, lines) /) ! default order
[14] do ! Repeat this loop forever... (until we break out of it)
[15] swaps Made = 0 ! Initially, we’ve made no swaps
[16] ! Make one pass of the bubble sort algorithm
[17] do count = 1, (lines - 1)
[18] ! If item is greater than the one after it, swap them
[19] if ( database (order (count)) > &
[20] database (order (count + 1)) ) then
[21] temp = order (count)
[22] order (count) = order (count + 1)
[23] order (count + 1) = temp
[24] swaps Made = swaps Made + 1
[25] end if
[26] end do
[27] ! If we made no swaps, break out of the loop.
[28] if ( swaps Made == 0 ) exit ! do count swaps
[29] end do
[30] end subroutine Integer Sort

Figure 4.25: An Ordered Bubble Sort of an Integer Array

Clearly you could write a very similar program using a true “pointer” array since they are now standard
in Fortran.
Next we will start to generalize the idea of sorting to include the sorting of objects that may have
numerous components. Assume the each record object to be read is defined as in Fig. 4.25.
There may be thousands, or millions, of such records to be read from a file, sorted by name and/or
number, and then displayed in sorted order. Program test bubble, in Fig. 4.26 illustrates one approach to
such a problem. Here since the database of records are to read from a file we do not yet know how many

c 2001 J.E. Akin 96
[ 1] module record Module
[ 2] !-------------------------------------------------------------
[ 3] ! record Module holds the "record" type
[ 4] !-------------------------------------------------------------
[ 5] ! record is a data structure with two names and an id number.
[ 6] type record
[ 7] character (len=24) :: last Name ! last name
[ 8] character (len=24) :: first Name ! first name
[ 9] integer :: id ! id number
[10] end type record
[11] end module record Module

Figure 4.26: A Typical Record in a List to be Sorted

[ 1] program test bubble

[ 2] !-------------------------------------------------------------
[ 3] ! test bubble asks for a filename for a file of names and id
[ 4] ! numbers, loads in the data from a file into the database,
[ 5] ! finds sorting orders, and prints sorted data
[ 6] !-------------------------------------------------------------
[ 7] use record Module ! need this to use the ’record’ type
[ 8] implicit none
[ 9] ! We define the database as an allocatable array of records.
[10] type (record), allocatable :: database (:)
[11]
[12] ! These arrays hold the sorted order of the database entries.
[13] integer, allocatable :: sort by Name (:)
[14] integer, allocatable :: sort by Number (:)
[15]
[16] character (len = 64) :: file Name ! file to read data from
[17] integer :: lines ! number of lines of input
[18] integer :: file Number ! the input file number
[19] integer :: loop Count ! loop counter
[20]
[21] file Number = 1 ! arbitrarily set file Number to 1
[22]
[23] write (*,*) ’Enter the filename to read data from:’
[24] read (*,’(A64)’) file Name
[25]
[26] ! Open our file and assign the number to ’file Number’
[27] open (unit = file Number, file = file Name)
[28]
[29] ! Find the number of lines in the input file with input Count.
[30] lines = input Count (file Number)
[31] write (*,*) ’There are ’, lines,’ records.’
[32]
[33] ! Allocate that many entries in the database and order arrays
[34] allocate ( database (lines) )
[35] allocate ( sort by Name (lines), sort by Number (lines) )
[36]
[37] ! Read the data from file into the database and close the file.
[38] call read Data (file Number, lines, database)
[39] close (file Number)
[40]
[41] ! Sort the database by name; the order will be in sort by Name.
[42] call String Sort (lines, database (:)%last Name, sort by Name)
[43] write (*,*); write (*,*) ’Data sorted by name: ’; write (*,*)
[44]
[45] ! Print out the data in the database sorted by name
[46] call show Data (lines, database, sort by Name)
[47] write (*,*); write (*,*) ’Data sorted by number:’; write (*,*)
[48]
[49] ! Sort the database by id numbers; new order is sort by Number.
[50] call Integer Sort (lines, database (:)%id, sort by Number)
[51]
[52] ! Print out the data in the database sorted by number.
[53] call show Data (lines, database, sort by Number)
[54] end program test bubble

Figure 4.27: Testing of Ordered Bubble Sorts

there are to be stored. Therefore, it is declared allocatable in line 13, and allocated later in line 34 after
we have evaluated the file size of a file named by the user. Although not generally necessary we have
selected to have an order array for names and a different one for numbers. The are sort by Name, and
sort by Number, respectively and are treated in a similar fashion to the database memory allocation as
noted in lines 13–14, and line 35.
In line 21 we have arbitrarily set a unit number to be used for the file. That is okay for a very small
code, but an unnecessary and unwise practice in general. The Fortran intrinsic inquire allows one to

c 2001 J.E. Akin 97
determine which units are inactive and we could create a function, say Get Next Unit, to select a safe
unit number for our input operation. After accepting a file name we open the unit, and count the number
of lines present in the file (see line 30). Had the database been on the standard input device, and not
contained any non-printing control characters, we could have easily read it with the statement
read *, database

However, it does contain tabs (ASCII character number 9), and is in a user defined file instead of the
standard input device so line 38 invokes subroutine read Data to get the data base. Of course, once the
tabs and commas have been accounted for and the names and id number extracted it uses an intrinsic
constructor on each line to form its database entry like:
database (line Count) = Record (last, first, id)

After all the records have been red into the database note that line 42 extracts all the last names with the
syntax
database (:) last Name

so they are copied into subroutine String Sort, as its second argument, and the ordered list
sort by Name) is returned to allow operations that need a last name sort. Likewise, subroutine In-
teger Sort, shown above, is used in line 50 to sort the id numbers and save the data in order list
sort by Number. The ordered lists are used in show Data, in lines 46 and 53, to display the sorted
information, without changing the original data.
If the supplied file, say namelist, contained data in the format of (String comma String tab Number)
with the following entries:
[ 1] Indurain, Miguel 5623
[ 2] van der Aarden, Eric 1245
[ 3] Rominger, Tony 3411
[ 4] Sorensen, Rolf 341
[ 5] Yates, Sean 8998
[ 6] Vandiver, Frank 45
[ 7] Smith, Sally 3821
[ 8] Johnston, David 3421
[ 9] Gillis, Malcolm 3785
[10] Johns, William 7234
[11] Johnston, Jonathan 7234
[12] Johnson, Alexa 5190
[13] Kruger, Charlotte 2345
[14] Butera, Robert 7253
[15] Armstrong, Lance 2374
[16] Hegg, Steve 9231
[17] LeBlanc, Lucien 23
[18] Peiper, Alan 5674
[19] Smith-Jones, Nancy 9082

The output would be:

[ 1] ! Enter the filename to read data from: namelist
[ 2] ! There are 19 records.
[ 3] !
[ 4] ! Data sorted by name:
[ 5] !
[ 6] ! Armstrong Lance 2374
[ 7] ! Butera Robert 7253
[ 8] ! Gillis Malcolm 3785
[ 9] ! Hegg Steve 9231
[10] ! Indurain Miguel 5623
[11] ! Johns William 7234
[12] ! Johnson Alexa 5190
[13] ! Johnston David 3421
[14] ! Johnston Jonathan 7234
[15] ! Kruger Charlotte 2345
[16] ! LeBlanc Lucien 23
[17] ! Peiper Alan 5674
[18] ! Rominger Tony 3411
[19] ! Smith Sally 3821
[20] ! Smith-Jones Nancy 9082
[21] ! Sorensen Rolf 341
[22] ! van der Aarden Eric 1245
[23] ! Vandiver Frank 45
[24] ! Yates Sean 8998
[25] !

and

c 2001 J.E. Akin 98
 [26] ! Data sorted by number:
[27] !
[28] ! LeBlanc Lucien 23
[29] ! Vandiver Frank 45
[30] ! Sorensen Rolf 341
[31] ! van der Aarden Eric 1245
[32] ! Kruger Charlotte 2345
[33] ! Armstrong Lance 2374
[34] ! Rominger Tony 3411
[35] ! Johnston David 3421
[36] ! Gillis Malcolm 3785
[37] ! Smith Sally 3821
[38] ! Johnson Alexa 5190
[39] ! Indurain Miguel 5623
[40] ! Peiper Alan 5674
[41] ! Johns William 7234
[42] ! Johnston Jonathan 7234
[43] ! Butera Robert 7253
[44] ! Yates Sean 8998
[45] ! Smith-Jones Nancy 9082
[46] ! Hegg Steve 9231

Pass 1 Pass 2 Pass 3

Level Level Level
1 2 3 4 5 6 1 2 3 4 5 1 2
Sorted Sorted Sorted
b b b b b b b a a a a a a
e e a a a a a b b b b b b
a a e d d d d d d d d d c
d d d e e e e e e e c c d
f f f f f c c c c c e e e
c c c c c f f f f f f f f

Is Was  Is Was Is Was

1 1 1 1 1 3 3 3 3
2 3 3 3 3 1 1 1 1
3 2 4 4 4 4 4 4 6
4 4 2 2 2 2 6 6 4
5 5 5 6 6 6 2 2 2
6 6 6 5 5 5 5 5 5
 Is Was ( j ) = k . What is position j was position k .

Figure 4.28: Sorting via an Order Vector, Array (Is Was) ! a b c d e f

4.11 Exercises
1 Frequently we need to know how many lines exist in an external file that is to be used by our
program. Usually we need that information to dynamically allocate memory for the arrays
that will be constructed from the file data to be read. Write a F90 program or routine that will
accept a unit number as input, open that unit, loop over the lines of data in the file connected to the
unit, and return the number of lines found in the file. (A external file ends when the iostat from
a read is less than zero.)

2 A related problem is to read a table of data from an external file. In addition to knowing the number
of lines in the file it is necessary to know the number of entities (columns) per line and to verify
that all lines of the file have the same number of columns. Develop a F90 program for that purpose.
(This is the sort of checking that the M ATLAB load function must do before loading an array of
data.)

c 2001 J.E. Akin 99
3 Write a program that displays the current date and time and uses the module tic toc, in Fig. 4.10,
to display the CPU time required for a calculation.
4 Develop a companion function called to upper that converts a string to all upper case letters.
Test it with the above program.
5 Develop a function that will take an external file unit number and count the number of lines in the
file connected to that unit. This assumes that the file has been “opened” on that unit. The interface
to the function is to be:
interface
function inputCount(unit) result(linesOfInput)
integer, intent(in) :: unit ! file unit number
integer :: linesOfInput ! result
end function inputCount
end interface

6 Assume the file in the previous problem contains two real values per line. Develop a subroutine
that will read the file and return two vectors holding the first and second values, respectively. The
interface to the subroutine is to be:
interface
subroutine readData (inFile, lines, x, y)
integer, intent(in) :: inFile, lines ! file unit, size
real, intent(out) :: x(lines), y(lines) ! data read
end subroutine readData
end interface

7 Written replies to the questions given below will be required. All of the named files are provided
in source form as well as being listed in the text. The cited Figure number indicates where some or
all of the code is discussed in the text.

(a) Figure 1.3 — hello.f90

What is necessary to split the printing statement so that “Hello,” and “world” occur on differ-
ent program lines? That is, to continue it over two lines?
(b) Figure 4.1 — arithmetic.f90
What is the meaning of the symbol ( mod ) used to get the Mod Result?
What is the meaning of the symbol ( ** ) used to get the Pow Result?
(c) Figure 4.3 — array index.f90
Is it good practice to use a loop index outside the loop? Why?
(d) Figure 4.4 — more or less.f90
What does the symbol ( > ) mean here?
What does the symbol ( == ) mean here?
(e) Figure 4.5 — if else.f90
What does the symbol ( .and. ) mean here? Can its preceding and following arguments be
interchanged (is it commutative)?
(f) Figure 4.6 — and or not.f90
What does the symbol ( .not. ) mean here?
What does the symbol ( .or. ) mean here? Can its preceding and following arguments be
interchanged (is it commutative)?
(g) Figure 4.7 — clip.f90
What does the symbol ( <= ) mean here?
(h) Figure 4.8 — maximum.f90
What are the input and output arguments for the maxint function?

c 2001 J.E. Akin 100
8 The vertical motion of a projectile at any time, t, has a position given by y = y0 + V0  t 1=2  g  t2,
and a velocity of V = V0 g  t when upward is taken as positive, and where the initial conditions
on the starting position and velocity, at t = 0, are y0 and V0 , respectively. Here the gravitational
acceleration term, g , has been taken downward. Recall that the numerical value of g depends on
the units employed. Use metric units with g = 9:81m=s2 for distances measured in meters and
time in seconds.
Write a C++ or F90 program that will accept initial values of y0 and V0 , and then compute and
print y and V for each single input value of time, t. Print the results for y0 = 1:5 meters and
V0 = 5:0m=s for times t = 0:5, 2.0, and 4.0 seconds.

9 Modify the projectile program written in Problem 2 to have it print the time, position, and velocity
for times ranging from 0.0 to 2.0 seconds, in increments of 0.05 seconds. If you use a direct loop
do not use real loop variables. Conclude the program by having it list the approximate maximum
(positive) height reached and the time when that occurred. The initial data will be the same, but
should be printed for completeness. The three columns of numbers should be neat and right jus-
tified. In that case the default print format (print * in F90) will usually not be neat and one must
employ a “formatted” print or write statement.

10 The Greatest Common Divisor of two positive integers can be computed by at least two differ-
ent approaches. There is a looping approach known as the Euclidean Algorithm which has the
following pseudocode:

Rank two positive integers as max and min.

do while min > 0
Find remainder of max divided by min.
Replace max by min.
Replace min by the remainder
end do
Display max as the greatest common divisor.

Implement this approach and test with max = 532 = 28  19 and min = 112 = 28  8. The
names of the remainder functions are given in Table 4.7.
Another approach to some algorithms is to use a “recursive” method which employs a subprogram
which calls itself. This may have an advantage in clarifying the algorithm, and/or in reducing
the round off error associated with the computations. For example, in computer graphics Bernstein
Polynomials are often used to display curves and surfaces efficiently by using a recursive definition
in calculating their value at a point.
The Greatest Common Divisor evaluation can also be stated in terms of a recursive function, say
gcd, having max and min as its initial two arguments. The following pseudocode defines the
function:

gcd(max, min) is
a) max if min = 0, otherwise
b) gcd(min, remainder of max divided by min) if min > 0

Also implement this version and verify that it gives the same result as the Eulerian Algorithm. Note
that F90 requires the use of the word ”recursive” when defining the subprogram statement block.
For example,

recursive function gcd(...) result(g)

....
end function gcd

c 2001 J.E. Akin 101
11 It is not uncommon for data files to be prepared with embedded tabs. Since it is a non-printing
control character you can not see it in a listing. However, if you read the file expecting an integer,
real, or complex variable the tab will cause a fatal read error. So one needs a tool to clean up such
a file.
Write a program to read a file and output a duplicate copy, except that all tabs are replaced with a
single space. One could read a complete line and check its characters, or read the file character by
character. Remember that C++ and F90 have opposite defaults when advancing to a new line. That
is, F90 advances to the next line, after any read or write, unless you include the format control,
advance = ’no’, while C++ does not advance unless you include the new line control, “<<
endl”, and C does not advance unless you include the new line control, “\n”.
12 Engineering data files consisting of discrete groups of variable types often begin with a control line
that lists the number of rows and columns of data, of the first variable type, that follow beginning
with the next line. At the end of the data block, the format repeats: control line, variable type
data block, etc. until all the variable types are read (or an error occurs where the end of file is
encountered). Write a program that reads such a file which contains an integer set, a real set, and a
second real set.
13 Neither C++ or F90 provides an inverse hyperbolic tangent function. Write such a function, called
arctanh. Test it with three different arguments against the values given by M ATLAB.

14 Often if one is utilizing a large number of input/output file units it may be difficult to keep up with
which one you need. One approach to dealing with that problem may be to define a unit Class
or to create an units Module to provide functionality and global access to file information. In
the latter case assume that we want to provide a function to simply find a unit number that is not
currently in use and utilize it for our input/output action:

interface
function get next io unit () result (next)
integer :: next ! the next available unit number
end function get next io unit
end interface

Use the Fortran INQUIRE statement to build such a utility. If you are familiar with Matlab you
will see this is similar to its fopen feature.

c 2001 J.E. Akin 102
Chapter 5

Object Oriented Methods

5.1 Introduction
In Section 1.7 we outlined procedures that should be considered while conducting the object-oriented
analysis and object-oriented design phases that are necessary before the OOP can begin. Here we will
expand on those concepts, but the reader is encouraged to read some of the books on those subjects.
Many of the references on OOA and OOD rely heavily on detailed graphical diagrams to describe the
classes, their attributes ans states, and how they interact with other classes. Often those OO methods do
not go into any programming language specific approaches. Our interest is on OOP so we usually will
assume that the OOA and OOD have been completed and supplied to us as a set of tables that describe the
application, and possibly a software interface contract. Sometimes we will use a subset of the common
OO methods diagrams to graphically represent the attributes and members of our classes. Since they
being used for OOP the graphical representations will contain, in part, the intrinsic data type descriptions
of the language being employed, as well as the derived types created with them.

5.2 The Drill Class

Our first illustration of typical OO methods will be to apply them to a common electric drill. It feeds
a rotating cutting bit through a workpiece, thereby removing a volume of material. The effort (power
or torque) required to make the hole clearly depends on the material of the workpiece, as well as the
attributes of the drill.
Table 5.1 contains a summary of the result of an OO analysis for the drill object. They are further
processed in Table 5.2 which gives the results of the OO design phase. When the OOD phase is complete
we can create the graphical representation of our Drill class as shown in Fig. 5.1. At this point one can
begin the actual OOP in the target language. The coding required for this object is so small we could
directly put it all together in one programming session. However, that is usually not the case. Often
segments of the coding will be assigned to different programming groups that must interface with each
other to produce a working final code. Often this means that the OOP design starts with defining the
interfaces to each member function. That is, all of the given and return arguments must be defined with
respect to their type, whether they are changed by the member, etc. Such an interface can be viewed
as a contract between a software supplier and a client user of the software. Once the interface has been
finalized, it can be written and given to the programmer to flesh out the full routine, but the interface itself
can not be changed.
The interface prototype for our drill object members are given in Fig. 5.2. In this case the remaining
coding is defined by a set of equations that relate the object attributes, selected member results, material
data, and a few conversion constants to obtain the proper units. Those relationships are given as:

c 2001 J.E. Akin

103
Attributes
What knowledge does it possess or require?
 Rotational speed (revolutions per minute)

 Feed rate per revolution (mm/rev)

 Diameter of the bit (mm)

 Power consumed (W)

Behavior
What questions should it be able to answer?
3
 What is the volumetric material removal rate? (mm /s)

 What is the cutting torque? (N
m)

 What is the material being removed?

Interfaces
What entities need to be input or output?
 Material data

 Torque produced

 Power

Table 5.1: Electric Drill OO Analysis

Area : A =  d2 =4 (mm2 )
Angular velocity : ! ; 1 rev=min = 260 rad/s (rad/s)
Material removal rate : M = A
feed
! (mm3 /s)
Power : P =m
u=T
! (W)
Torque : T = P=! ; 1 m = 1000 mm (N
mm)
Diameter : d (mm)
Feed rate : feed (mm/rev)
Material dissipation : u (W
s/mm3 )

Figure 5.1: Graphical Representation of an Electric Drill Class

The full implementation of the drill class is given in Fig. 5.3, and a main program to test the drill
class is given in Fig. 5.4. When we wrote the manual constructor, Drill , in this example we chose to

c 2001 J.E. Akin

104
[ 1] interface
[ 2] ! type (Drill) :: x ; x = Drill (d, f, s) ! intrinsic constructor
[ 3]
[ 4] function Drill (d, f, s) result (x) ! default constructor
[ 5] real, optional :: d, f, s ! given diameter, feed, speed
[ 6] type (Drill) :: x ! the Drill instance
[ 7] end function Drill
[ 8]
[ 9] function get mr rate (x) result (r) ! material removal rate
[10] type (Drill), intent(in) :: x ! a given drill instance
[11] real :: r ! volume cut rate
[12] end function get mr rate
[13]
[14] function get torque (x, unit Power) result (t) ! torque from power
[15] type (Drill), intent(in) :: x ! given drill instance
[16] real, intent(in) :: unit Power ! dissipated in cutting
[17] real :: t ! resulting torque
[18] end function get torque
[19]
[20] subroutine in (x) ! read a Drill instance
[21] type (Drill), intent(out) :: x ; end subroutine in
[22]
[23] subroutine out (x) ! output a Drill instance
[24] type (Drill), intent(in) :: x ! given drill instance
[25] end subroutine out
[26] end interface

Figure 5.2: Drill Object Contract Interface Prototype

[ 1] module class Drill ! class name

[ 2] implicit none ! enforce strong typing
[ 3] real, parameter :: pi = 3.141592654 ! or use math constants
[ 4] public :: Drill, Drill , get mr rate, get torque
[ 5] real, private :: diameter, feed, speed
[ 6]
[ 7] type Drill ! defined type, private data
[ 8] real :: diameter, feed, speed ; end type
[ 9]
[10] contains ! member functions, overloaded & new operators
[11]
[12] ! type (Drill) :: x ; x = Drill (d, f, s) ! intrinsic constructor
[13]
[14] function Drill (d, f, s) result (x) ! default constructor
[15] real, optional :: d, f, s ! given diameter, feed, speed
[16] type (Drill) :: x ! the Drill instance
[17] if ( present(d) .and. present(f) .and. present(s) ) then
[18] x = Drill (d, f, s) ! intrinsic constructor
[19] else ! check various input options
[20] if ( .not. ( present(d) ) ) then ! no diameter given
[21] x = Drill (10., 0., 0.) ! default 10mm, at rest zero
[22] end if ! default form
[23] end if ! full form
[24] end function Drill
[25]
[26] function get mr rate (x) result (r) ! material removal rate, mm3̂/sec
[27] type (Drill), intent(in) :: x ! a given drill instance
[28] real :: r ! volume cut rate
[29] r = 0.25 * pi * x%diameter * x%diameter * x%feed * x%speed/60.
[30] end function get mr rate
[31]
[32] function get torque (x, unit Power) result (t) ! torque from power
[33] type (Drill), intent(in) :: x ! given drill instance
[34] real, intent(in) :: unit Power ! dissipated in cutting
[35] real :: t ! resulting torque
[36] real :: rad per sec ! radians per second
[37] rad per sec = 2 * pi * x%speed / 60.
[38] t = get mr rate(x) * unit Power / rad per sec ! torque
[39] end function get torque
[40]
[41] subroutine in (x) ! input a Drill instance
[42] type (Drill), intent(out) :: x ! given drill instance
[43] read *, x ! get intrinsic data
[44] end subroutine in
[45]
[46] subroutine out (x) ! output a Drill instance
[47] type (Drill), intent(in) :: x ! given drill instance
[48] print *,"Drill"; print *, " Diameter: ",x % diameter
[49] print *," Feed : ",x % feed; print *," Speed : ",x % speed
[50] end subroutine out
[51] end module class Drill ! close class definition

Figure 5.3: A Electrical Drill Class

c 2001 J.E. Akin

105
Attributes
Name Type Private Description
diameter real Y Bit diameter (mm)
feed real Y Bit feed rate (mm/rev)
speed real Y Bit rotational speed (rpm)
Behavior
Name Private Description
drill N Default constructor using all attributes, or none
get mr rate N Material removal rate (mm3 /sec)
get torque N Required torque (N
m)
power N Required power (W)
Data
Name Description
u Material power description per unit volume (W s/mm3 )
Interfaces
Name Description
read Input drill and material data
print Output object results

Table 5.2: Electric Drill OO Design

utilize the intrinsic constructor Drill (in lines 18 and 21) rather than including lines to assign values to
each of the components of our data type. If at some later time we add or delete a component in the type
declaration then the number of required arguments for the intrinsic constructor would also change. That
would require the revision of all members that used the intrinsic constructor. An advantage of the object-
oriented approach to programming is that we know that all such routines (that can access the intrinsic
constructor) are encapsulated within this class declaration module, and we can be sure that no other code
segments must be changed to remain consistent with the new version. That is, OOP helps with code
maintance.

5.3 Global Positioning Satellite Distances

Consider the problem of traveling by ship or airplane between two points on the earth. Here we assume
that there are no physical obstructions that prevent the vehicle from following the shortest path, which is
an arc of a ”great circle” on the earth’s surface. We will neglect the altitude of the airplane in comparison
to the earth’s radius. The original and final positions are to be defined in terms of their angles of latitude
(measured N or S from the equator) and longitude (measured E or W from Greenwich, England). These
two attributes define an angular position from a defined reference point on the spherical surface of the
earth. They are measured in terms of whole degrees, whole minutes (1 degree = 60 minutes), and seconds
(1 minute = 60 seconds). Historically, whole seconds are usually used, but they give positions that are
only accurate to about 300 meters. Thus, we will use a real variable for the seconds to allow for potential
reuse for the software for applications that require more accuracy, such as those using Global Positioning
Satellite (GPS) data. Recall that latitude and longitude have associated directional information of North
or South, and East or West, respectively. Also in defining a global position point it seems logical to
include a name for each position. Depending on the application the name may identify a city or port, or
a ”station number” in a land survey, or a ”path point number” for a directed robot motion.
Eventually, we want to compute the great arc distance between given pairs of latitude and longitude.
That solid geometry calculation requires that one use angles that are real numbers measured in radi-
ans (2pi = 360 degrees). Thus our problem description begins with an Angle class as its basic class.
Both latitude and longitude will be defined to be of the Position Angle class and we observe that
a Position Angle is a ”Kind-Of” Angle, or a Position Angle has an ”Is-A” relationship to an
Angle. The positions we seek are on a surface so only two measures (latitude and longitude) are needed

c 2001 J.E. Akin

106
[ 1] program main ! test the Drill class
[ 2] use class Drill ! i.e., all public members and public data
[ 3] implicit none
[ 4] type (Drill) :: drill A, drill B, drill C
[ 5] real :: unit Power
[ 6] print *, "Enter diameter (mm), feed (mm/rev), speed (rpm):"
[ 7] call in (drill A)
[ 8] print *, "Enter average power unit for material ( W.s/mm**3):"
[ 9] read *, unit Power ; call out (drill A) ! user input
[10] print *, "Material removal rate is: ", get mr rate(drill A), &
[11] " mm**3/sec"
[12] print *, "Torque in this material is: ", &
[13] & get torque (drill A, unit Power), " W.s"
[14] drill B = Drill (5., 4., 3.); call out (drill B) ! manual
[15] drill C = Drill (); call out (drill C) ! default
[16] end program ! Running gives
[17] ! Enter diameter (mm), feed (mm/rev), speed (rpm): 10 0.2 800
[18] ! Enter average power unit for material ( W.s/mm**3): 0.5
[19] ! Drill
[20] ! Diameter: 10.
[21] ! Feed : 0.200000003
[22] ! Speed : 800.
[23] ! Material removal rate is: 209.439514 mm**3/sec
[24] ! Torque in this material is: 1.25 W.s
[25] ! Drill
[26] ! Diameter: 5.
[27] ! Feed : 4.
[28] ! Speed : 3.
[29] ! Drill
[30] ! Diameter: 10.
[31] ! Feed : 0.E+0
[32] ! Speed : 0.E+0

Figure 5.4: Testing a Electrical Drill Class

to uniquely define the location, which we will refer to as the Global Position. Here we see that the
two Position Angle object values are a ”Part-Of” the Global Position class, or we can say that
a Global Position ”Has-A” Position Angle.
The sort of relationships between classes that we have noted above are quite common and relate to the
concept of inheritance as a means to reuse code. In an ”Is-A” relationship, the derived class is a variation
of the base class. Here the derived class Position Angle forms an ”Is-A” relation to the base class,
Angle. In a ”Has-A” relationship, the derived class has an attribute or property of the base class. Here
the derived class of Global Position forms a Has-A relation to its base class of Position Angle.
Also, the Great Arc class forms a ”Has-A” relation to the Global Position class.
Looking back at previous classes, in Chapter 3, we observe that the class Student ”Is-A” variation
of the class Person and the class Person forms at least one ”Has-A” relationship with the class Date.
In general we know that a graduate student is a ”Kind-Of” student, but not every student is a graduate
student. This subtyping, or ”Is-A” relationship is also called interface inheritance. Likewise, complicated
classes can be designed from simpler or composition inheritance.
The OO Analysis Tables for the classes of Great Arc, Global Position, Position Angle,
and Angle are given in Tables 5.3 through 5.6, respectively. Historically people have specified lati-
tude and longitude mainly in terms of whole (integer) degrees, minutes, and seconds. Sometimes you
find navigation charts that give positions in whole degrees and decimal minutes. Today GPS data are
being used for various high accuracy positioning such as land surveys, or the control of robots as they
move over distances of a few meters. The latter will clearly need decimal seconds values in their con-
structor. Thus, we will create a number of constructors for the position angles. In the next chapter we
will review how to access any of them, based on the signature of their arguments, through the use of
a single polymorphic routine name. These considerations and the OOA tables lead to the construction
of the corresponding set of OO Design Tables given in Tables 5.7 through 5.10. Those OOD tables
could lead to software interface contracts to be distributed to the programming groups. When combined
and tested they yield the corresponding class modules which are shown for the classes Angle, Posi-
tion Angle, Global Position, and Great Arc in Figs. 5.6 to 5.12, respectively. They in turn
are verified by the main program given in Fig. 5.13 along with its output.

c 2001 J.E. Akin

107
Attributes
What knowledge does it possess or require?
 Global position 1 (latitude, longitude)

 Global position 2 (latitude, longitude)

 Smallest arc (km)

 Radius of the earth (km)

Behavior
What questions should it be able to answer?
 What is the (smallest) great arc between the points

What services should it provide?

 Default value (Greenwich, Greenwich, 0.0)

 Initialize for two positions

 Convert kilometers to miles

Relationships
What are its related classes?
 Has-A pair of Global Positions

Interfaces
What entities need to be input or output?
 The distance between two positions.

Table 5.3: Great Arc OO Analysis

Attributes
What knowledge does it possess or require?
 Latitude (degrees, minutes, seconds, and direction)

 Longitude (degrees, minutes, seconds, and direction)

Behavior
What questions should it be able to answer?
 What is the latitude of the location

 What is the longitude of the location

What services should it provide?

 Default position (Greenwich)

 Initialize a position (latitude and longitude)

Relationships
What are its related classes?
 Part-Of GreatArc

 Has-A pair of Position Angles

Interfaces
What entities need to be input or output?
 The latitude and longitude, and a position name.

Table 5.4: Global Position OO Analysis

c 2001 J.E. Akin

108
Attributes
What knowledge does it possess or require?
 Magnitude (degrees, minutes, seconds)

 Direction (N or S or E or W)

Behavior
What questions should it be able to answer?
 What is its magnitude and direction

What services should it provide?

 Default value (0, 0, 0.0, N)

 Initialization to input value

Relationships
What are its related classes?
 Part-Of Global Positions

 Is-A Angle

Interfaces
What entities need to be input or output?
 None

Table 5.5: Position Angle OO Analysis

Attributes
What knowledge does it possess or require?
 Signed value (radians)

Behavior
What questions should it be able to answer?
 What is the current value

What services should it provide?

 default values (0.0)

 Conversion to signed decimal degrees

 Conversion to signed degree, minutes, and decimal seconds

 Conversion from signed decimal degrees

 Conversion from signed degree, minutes, and decimal seconds

Relationships
What are its related classes?
 Base Class for Position Angle

Interfaces
What entities need to be input or output?
 None

Table 5.6: Angle OO Analysis

c 2001 J.E. Akin

109
Attributes
Name Type Private Description
point 1 Global Position Y Lat-Long-Name of point 1
point 2 Global Position Y Lat-Long-Name of point 2
arc real Y Arc distance between points
Behavior
Name Private Description
Great Arc N Constructor for two position points
get Arc N Compute great arc between two points
Data
Name Description
Earth Radius Mean Conversion factor
m Per Mile Conversion factor
Interfaces
Name Description
List Great Arc Print arc values (two positions and distance)
List Pt to Pt Print distance and two points

Table 5.7: Class Great Arc OO Design

Attributes
Name Type Private Description
latitude Position Angle Y Latitude
longitude Position Angle Y Longtitude
name characters Y Point name
Behavior
Name Private Description
Global Position N Constructor for d-m-s pairs and point name
set Lat and Long at N Constructor for lat-long-name set
get Latitude N Return latitude of a point
get Longitude N Return longitude of a point
set Latitude N Insert latitude of a point
set Longitude N Insert longitude of a point
Data
Name Description
None
Interfaces
Name Description
List Position Print name and latitude, longitude of a position

Table 5.8: Class Global Position OO Design

c 2001 J.E. Akin

110
Attributes
Name Type Private Description
deg integer Y Degrees of angle
min integer Y Minutes of angle
sec real Y Seconds of angle
dir character Y Compass direction
Behavior
Name Private Description
Default Angle N Default constructor
Decimal min N Constructor for decimal minutes
Decimal sec N Constructor for decimal seconds
Int deg N Constructor for whole deg
Int deg min N Constructor for whole deg, min
Int deg min sec N Constructor for whole deg, min, sec
to Decimal Degrees N Convert position angle values to decimal degree
to Radians N Convert position angle values to decimal radian
Data
Name Description
None
Interfaces
Name Description
List Position Angle Print values for position angle
Read Position Angle Read values for position angle

Table 5.9: Class Position Angle OO Design

Attributes
Name Type Private Description
rad real Y Radian measure of the angle (rad)
Behavior
Name Private Description
Angle N A generic constructor
List Angle N List angle value in radians and degrees
Data
Name Description
Deg per Rad Unit conversion parameter

Table 5.10: Class Angle OO Design

c 2001 J.E. Akin

111
 Angle Class

real rad

real Deg_Per_Rad

Angle Angle_
Angle List_Angle

Angle Angle

Figure 5.5: Graphical Representation of an Angle Class

[ 1] module class Angle ! file: class Angle.f90

[ 2] implicit none
[ 3] type Angle ! angle in (signed) radians
[ 4] private
[ 5] real :: rad ! radians
[ 6] end type
[ 7] real, parameter:: Deg Per Rad = 57.2957795130823209d0
[ 8] contains
[ 9]
[10] function Angle (r) result (ang) ! public constructor
[11] real, optional :: r ! radians
[12] type (Angle) :: ang
[13] if ( present(r) ) then
[14] ang = Angle (r) ! intrinsic constructor
[15] else ; ang = Angle (0.0) ! intrinsic constructor
[16] end if ; end function Angle
[17]
[18] subroutine List Angle (ang)
[19] type (Angle), intent(in) :: ang
[20] print *, ’Angle = ’, ang % rad, ’ radians (’, &
[21] Deg Per Rad * ang % rad, ’ degrees)’
[22] end subroutine List Angle
[23] end module class Angle

Figure 5.6: A Definition of the Class Angle

c 2001 J.E. Akin

112
 Position_Angle Class

integer deg
integer min
real sec
character dir

Position_Angle Default_Angle
Position_Angle Decimal_min
Position_Angle Decimal_sec
Position_Angle Int_deg
Position_Angle Int_deg_min
Position_Angle Int_deg_min_sec
Position_Angle List_Position_Angle
Position_Angle Read_Position_Angle
real to_Decimal_Degrees
real to_Radians

Position_Angle Position_Angle

Figure 5.7: Graphical Representation of a Position Angle Class

[ 1] module class Position Angle ! file: class Position Angle.f90

[ 2] use class Angle
[ 3] implicit none
[ 4] type Position Angle ! angle in deg, min, sec
[ 5] private
[ 6] integer :: deg, min ! degrees, minutes
[ 7] real :: sec ! seconds
[ 8] character :: dir ! N | S, E | W
[ 9] end type
[10] contains
[11]
[12] function Default Angle () result (ang) ! default constructor
[13] type (Position Angle) :: ang
[14] ang = Position Angle (0, 0, 0., ’N’) ! intrinsic
[15] end function Default Angle
[16]
[17] function Decimal min (d, m, news) result (ang) ! public
[18] integer, intent(in) :: d ! degrees
[19] real, intent(in) :: m ! minutes
[20] character, intent(in) :: news ! N | S, E | W
[21] type (Position Angle) :: ang ! angle out
[22] integer :: min ! minutes
[23] real :: s ! seconds
[24] min = floor ( m ) ; s = (m - min)*60. ! convert
[25] ang = Position Angle (d, m, s, news) ! intrinsic
[26] end function Decimal min
[27]
[28] function Decimal sec (d, m, s, news) result (ang) ! public
[29] integer, intent(in) :: d, m ! degrees, minutes
[30] real, intent(in) :: s ! seconds
[31] character, intent(in) :: news ! N | S, E | W
[32] type (Position Angle) :: ang ! angle out
[33] ang = Position Angle (d, m, s, news) ! intrinsic
[34] end function Decimal sec
[35]

(Fig. 5.8, A Definition of the Class Position Angle (Continued))

c 2001 J.E. Akin

113
[36] function Int deg (d, news) result (ang) ! public
[37] integer, intent(in) :: d ! degrees, minutes
[38] character, intent(in) :: news ! N | S, E | W
[39] type (Position Angle) :: ang ! angle out
[40] ang = Position Angle (d, 0, 0.0, news) ! intrinsic
[41] end function Int deg
[42]
[43] function Int deg min (d, m, news) result (ang) ! public
[44] integer, intent(in) :: d, m ! degrees, minutes
[45] character, intent(in) :: news ! N | S, E | W
[46] type (Position Angle) :: ang ! angle out
[47] ang = Position Angle (d, m, 0.0, news) ! intrinsic
[48] end function Int deg min
[49]
[50] function Int deg min sec (d, m, s, news) result (ang) ! public
[51] integer, intent(in) :: d, m, s ! deg, min, seconds
[52] character, intent(in) :: news ! N | S, E | W
[53] type (Position Angle) :: ang ! angle out
[54] ang = Position Angle (d, m, real(s), news) ! intrinsic
[55] end function Int deg min sec
[56]
[57] subroutine List Position Angle (a)
[58] type (Position Angle) :: a ! angle
[59] print 5, a ; 5 format (i3, " ", i2,"’ ", f8.5, ’" ’, a1)
[60] end subroutine
[61]
[62] subroutine Read Position Angle (a)
[63] type (Position Angle) :: a ! angle
[64] read *, a%deg, a%min, a%sec, a%dir ; end subroutine
[65]
[66] function to Decimal Degrees (ang) result (degrees)
[67] type (Position Angle), intent(in) :: ang
[68] real :: degrees
[69] degrees = ang%deg + ang%min/60. + ang%sec/60.
[70] if (ang%dir == "S" .or. ang%dir == "s" .or. &
[71] ang%dir == "W" .or. ang%dir == "w") degrees = -degrees
[72] end function to Decimal Degrees
[73]
[74] function to Radians (ang) result (radians)
[75] type (Position Angle), intent(in) :: ang
[76] real :: radians
[77] radians = (ang%deg + ang%min/60. + ang%sec/60.)/Deg Per Rad
[78] if (ang%dir == "S" .or. ang%dir == "s" .or. &
[79] ang%dir == "W" .or. ang%dir == "w") radians = -radians
[80] end function to Radians
[81] end module class Position Angle

Figure 5.8: A Definition of the Class Position Angle

Global_Position Class

Position_Angle latitude
Position_Angle longitude
character name

Global_Position Global_Position_
Global_Position set_Lat_and_Long_at
Position_Angle get_Latitude
Position_Angle get_Longitude
Position_Angle set_Latitude
Position_Angle set_Longitude
Global_Position List_Position

Global_Position Global_Position

Figure 5.9: Graphical Representation of a Global Position Class

c 2001 J.E. Akin

114
[ 1] module class Global Position
[ 2] use class Position Angle
[ 3] implicit none
[ 4] type Global Position
[ 5] private
[ 6] type (Position Angle) :: latitude, longitude
[ 7] character (len=31) :: name
[ 8] end type Global Position
[ 9] contains
[10]
[11] function Global Position (d1, m1, s1, c1, & ! constructor
[12] d2, m2, s2, c2, n) result (GP)
[13] integer, intent(in) :: d1, m1, s1 ! deg, min, sec
[14] integer, intent(in) :: d2, m2, s2 ! deg, min, sec
[15] character, intent(in) :: c1, c2 ! compass
[16] character (len=*) :: n ! name
[17] type (Global Position) :: GP ! returned position
[18] GP % latitude = Int deg min sec (d1, m1, s1, c1)
[19] GP % longitude = Int deg min sec (d2, m2, s2, c2)
[20] GP % name = n ; end function Global Position
[21]
[22] function set Lat and Long at (lat, long, n) result (GP) ! cons
[23] type (Position Angle), intent(in) :: lat, long ! angles
[24] character (len=*), intent(in) :: n ! name
[25] type (Global Position) : GP ! position
[26] GP % latitude = lat ; GP % longitude = long
[27] GP % name = n ; end function set Lat and Long at
[28]
[29] function get Latitude (GP) result (lat)
[30] type (Global Position), intent(in) :: GP
[31] type (Position Angle) :: lat
[32] lat = GP % latitude ; end function get Latitude
[33]
[34] function get Longitude (GP) result (long)
[35] type (Global Position), intent(in) :: GP
[36] type (Position Angle) :: long
[37] long = GP % longitude ; end function get Longitude
[38]
[39] subroutine set Latitude (GP, lat)
[40] type (Global Position), intent(inout) :: GP
[41] type (Position Angle), intent(in) :: lat
[42] GP % latitude = lat ; end subroutine set Latitude
[43]
[44] subroutine set Longitude (GP, long)
[45] type (Global Position), intent(inout) :: GP
[46] type (Position Angle), intent(in) :: long
[47] GP % longitude = long ; end subroutine set Longitude
[48]
[49] subroutine List Position (GP)
[50] type (Global Position), intent(in) :: GP
[51] print *, ’Position at ’, GP % name
[52] write (*,’(" Latitude: ")’, advance = "no")
[53] call List Position Angle (GP % latitude)
[54] write (*,’(" Longitude: ")’, advance = "no")
[55] call List Position Angle (GP % longitude)
[56] end subroutine List Position
[57] end module class Global Position

Figure 5.10: A Definition of the Class Global Position

c 2001 J.E. Akin

115
 Great_Arc Class

Global_Position point_1
Global_Position point_2
real arc

real Earth_Radius_Mean
real m_Per_Mile

Great_Arc Great_Arc_
real get_Arc
Great_Arc List_Great_Arc
real List_PT_to_Pt

Great_Arc Great_Arc

Figure 5.11: Graphical Representation of a Great Arc Class

[ 1] module class Great Arc

[ 2] use class Global Position
[ 3] implicit none
[ 4] real, parameter :: Earth Radius Mean = 6.371d6 ! meters
[ 5] real, parameter :: m Per Mile = 1609.344
[ 6] type Great Arc
[ 7] type (Global Position) :: point 1, point 2
[ 8] real :: arc
[ 9] end type Great Arc
[10] contains
[11]
[12] function Great Arc (GP1, GP2) result (G A) ! constructor
[13] type (Global Position), intent(in) :: GP1, GP2 ! points
[14] type (Great Arc) :: G A ! earth arc
[15] G A = Great Arc (GP1, GP2, get Arc (GP1, GP2)) ! intrinsic
[16] end function Great Arc
[17]
[18] function get Arc (GP1, GP2) result (dist)
[19] type (Global Position), intent(in) :: GP1, GP2
[20] real :: dist
[21] real :: lat1, lat2, long1, long2
[22] ! convert latitude, longitude to radians
[23] lat1 = to Radians (get Latitude (GP1))
[24] lat2 = to Radians (get Latitude (GP2))
[25] long1 = to Radians (get Longitude (GP1))
[26] long2 = to Radians (get Longitude (GP2))
[27] ! compute great circle arc of earth
[28] dist = 2 * Earth Radius Mean &
[29] * asin( sqrt ( (sin((lat1 - lat2)/2.))**2 &
[30] + cos(lat1)*cos(lat2)*(sin((long1-long2)/2.))**2 ) )
[31] end function get Arc
[32]
[33] subroutine List Great Arc (A to B)
[34] type (Great Arc), intent(in) :: A to B
[35] real :: dist ! in meters
[36] print * ; print *, "The great circle arc between"
[37] call List Position (A to B % point 1)
[38] call List Position (A to B % point 2)
[39] dist = A to B % arc ! convert to km and miles
[40] print *, "is ", dist/1000, " km (", dist/m Per Mile, "miles)."
[41] end subroutine List Great Arc
[42]
[43] subroutine List Pt to Pt (GP1, GP2) ! alternate
[44] type (Global Position), intent(in) :: GP1, GP2 ! points
[45] real :: arc ! distance
[46] print * ; print *, "The great circle arc between"
[47] call List Position (GP1) ; call List Position (GP2)
[48] arc = get Arc (GP1, GP2) ! in meters
[49] print *, "is ", arc/1000, " km (", arc/m Per Mile, "miles)"
[50] end subroutine List Pt to Pt
[51] end module class Great Arc

Figure 5.12: Definition of the Class Great Arc

c 2001 J.E. Akin

116
[ 1] program main
[ 2] use class Great Arc
[ 3] implicit none
[ 4] type (Great Arc) :: arc
[ 5] type (Global Position) :: g1, g2
[ 6] type (Position Angle) :: a1, a2
[ 7] type (Angle) :: ang
[ 8] real :: deg, rad
[ 9] a1 = Decimal sec (10, 30, 0., "N"); call List Position Angle(a1
[10] a1 = Int deg min sec(10, 30, 0, "N"); call List Position
[11] a1 = Int deg min (10, 30, "N"); call List Position An
[12] a1 = Int deg (20, "N"); call List Position Angle(a1
[13] ! call Read Position Angle (a2)
[14] a2 = Decimal sec (30, 48, 0., "E"); call List Position Angle(a2
[15] ang = Angle (1.0) ; call List Angle (ang)
[16] deg = to Decimal Degrees (a1) ; print *, deg, deg/Deg Per Rad
[17] rad = to Radians (a1) ; print *, rad
[18] !
[19] g1 = set Lat and Long at (a1, a2, ’g1’)
[20] call List Position (g1)
[21] g2 = Global Position (20, 5, 40, "S", 75, 0, 20, "E", ’g2’)
[22] call List Position (g2)
[23] print *, "Arc = ", get Arc (g1, g2), " (meters)"
[24] g1 = Global Position ( 0, 0, 0, "N", 0, 0, 0, "E", ’equator’)
[25] g2 = Global Position (90, 0, 0, "N", 0, 0, 0, "E", ’N-pole’)
[26] call List Pt to Pt (g1, g2)
[27] arc = Great Arc (g1, g2) ; call List Great Arc (arc)
[28] end program main ! running gives:
[29] ! 10 30’ 0.00000" N ; ! 10 30’ 0.00000" N ; ! 10 30’ 0.00000" N
[30] ! 20 0’ 0.00000" N ; ! 30 48’ 0.00000" N
[31] ! Angle = 1.000000000 radians ( 57.29578018 degrees)
[32] ! 20.00000000 0.3490658402 ; ! 0.3490658402
[33] ! Position at g1 ; ! Position at g2
[34] ! Latitude: 20 0’ 0.00000" N ; ! Latitude: 20 5’ 40.00000" S
[35] ! Longitude: 30 48’ 0.00000" E ; ! Longitude: 75 0’ 20.00000" E
[36] ! Arc = 6633165.000 (meters)
[37] !
[38] ! The great circle arc between
[39] ! Position at equator ; ! Position at N-pole
[40] ! Latitude: 0 0’ 0.00000" N ; ! Latitude: 90 0’ 0.00000" N
[41] ! Longitude: 0 0’ 0.00000" E ; ! Longitude: 0 0’ 0.00000" E
[42] ! is 10007.54297 km ( 6218.398926 miles)
[43] !
[44] ! The great circle arc between
[45] ! Position at equator ; ! Position at N-pole
[46] ! Latitude: 0 0’ 0.00000" N ; ! Latitude: 90 0’ 0.00000" N
[47] ! Longitude: 0 0’ 0.00000" E ; ! Longitude: 0 0’ 0.00000" E
[48] ! is 10007.54297 km ( 6218.398926 miles)

Figure 5.13: Testing the Great Arc Class Interactions

c 2001 J.E. Akin

117
5.4 Exercises
1. Referring to Chapter 3, develop OOA and OOD tables for the a) Geometric class, b) Date class, c)
Person class, d) Student class.
2. Develop the graphical representations for the classes in the a) drill study, b) global position study.
3. Use the classes in the GPS study to develop a main program that will read a list (vector) of
Global Position types and use them to output a square table of great arc distances from one site
to each of the others. That is, the table entry in row j, column k gives the arc from site j to site k. Such
a table would be symmetric (with zeros along one diagonal) so you may want to give only half of it.
4. Modify the given Class Position Angle to provide a polymorphic interface for a constructor
Position Angle that will accept decimal, integer or no data for the seconds value. Also allow for
the omission of the minutes value.

c 2001 J.E. Akin

118
Chapter 6

Inheritance and Polymorphism

6.1 Introduction
As we have seen earlier in our introduction to OOP inheritance is a mechanism for deriving a new class
from an older base class. That is, the base class, sometimes called the super class, is supplemented or
selectively altered to create the new derived class. Inheritance provides a powerful code reuse mechanism
since a hierarchy of related classes can be created that share the same code. A class can be derived from
an existing base class using the module construct illustrated in Fig. 6.1.
We note that the inheritance is invoked by the USE statement. Sometimes an inherited entity (attribute
or member) needs to be slightly amended for the purposes of the new classes. Thus, at times one may
want to selectively bring into the new class only certain entities from the base class. The modifier ONLY
in a USE statement allows one to select the desired entities from the base class as illustrated below in
Fig. 6.2. It is also common to develop name conflicts when combining entities from one or more related
classes. Thus a rename modifier, =>, is also provided for a USE statement to allow the programmer to
pick a new local name for an entity onherited from the base class. The form for that modifier is given in
Fig. 6.3.
It is logical to extend any or all of the above inheritance mechanisms to produce multiple inheritance.
Multiple Inheritance allows a derived class to be created by using inheritance from more than a single
base class. While multiple inheritance may at first seem like a panacea for efficient code reuse, experi-
ence has shown that a heavy use of multiple inheritance can result in entity conflicts and be otherwise
counterproductive. Nevertheless it is a useful tool in OOP. In F90 the module form for selective multiple
inheritance would combine the above USE options in a single module as illustrated in Fig. 6.4.

module derived class name

use base class name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Figure 6.1: F90 Single Inheritance Form.

c 2001 J.E. Akin 119

 module derived class name
use base class name, only: list of entities
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Figure 6.2: F90 Selective Single Inheritance Form.

module derived class name

use base class name, local name => base entity name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Figure 6.3: F90 Single Inheritance Form, with Local Renaming.

module derived class name

use base1 class name
use base2 class name
use base3 class name, only: list of entities
use base4 class name, local name => base entity name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Figure 6.4: F90 Multiple Selective Inheritance with Renaming.

c 2001 J.E. Akin 120

[ 1] module class Professor ! file: class Professor.f90
[ 2] implicit none
[ 3] public :: print, name
[ 4] private :: publs
[ 5] type Professor
[ 6] character (len=20) :: name
[ 7] integer :: publs ! publications
[ 8] end type Professor
[ 9] contains
[10] function make Professor (n, p) result (who)
[11] character (len=*), optional, intent(in) :: n
[12] integer, optional, intent(in) :: p
[13] type (Professor) :: who ! out
[14] who%name = " " ! set defaults
[15] who%publs = 0.0
[16] if ( present(n) ) who%name = n ! construct
[17] if ( present(p) ) who%publs = p
[18] end function make Professor
[19]
[20] function print (who)
[21] type (Professor), intent(in) :: who
[22] print *, "My name is ", who%name, &
[23] ", and I have ", who%publs, " publications."
[24] end function print
[25] end module class Professor

Figure 6.5: A Professor Class

6.2 Example Applications of Inheritance

6.2.1 The Professor Class
In the introductory examples of OOP in Chapter 3 we introduced the concepts of inheritance and multiple
inheritance by the use of the Date class, Person class, and Student class. To reinforce those concepts
we will reuse those three classes and will have them be inherited by a Professor class. Acknowledging
the common “publish or perish” aspect of academic life the professor class must keep up with the number
of publications of the professor. The new class is given in Fig. 6.5 along with a small validation program
in Fig. 6.6.
Note that the validation program brings in three different versions of the “print” member (lines 7-9)
and renames two of them to allow a polymorphic print statement (lines 12-14) that selects the proper
member based solely on the class of its argument. Observe that the previous Date class is brought into
the main through the use of the Person class (in line 7). Of course, it is necessary to have an interface
defined for the overloaded member name so that the compiler knows which candidate routines to search
at run time. This example also serves to remind the reader that Fortran does not have keywords that are not
allowed to be used by the programmer. In this case the print function (lines 19, 22, 25) has automatically
replaced the intrinsic print function of Fortran. Most languages, including C++ do not allow one to do
that.

6.2.2 The Employee and Manager Classes

Next we will begin the popular employee-manager classes as examples of common related classes that
demonstrate the use of inheritance. Once again the idea behind encapsulating these data and their as-
sociated functionality is to model a pair of real world entities - an employee and a manager. As we go
through possible relations between these two simple classes it becomes clear that there is no unique way
to establish the classes and how they should interact. We begin with a minimal approach and then work
through two alternate versions to reach the point where an experienced OO programmer might have be-
gun. The first Employee class, shown in Fig. 6.7 has a name and pay rate as its attributes. Only the
intrinsic constructor is used within the member setDataE to concatenate a first name and last name to
form the complete name attribute and to accept the pay rate. To query members getNameE and getRate
are provided to extract either of the desired attributes. Finally, member payE is provided to compute the
pay earned by an employee. It assumes that an employee is paid by the hour. A simple testing main
program is shown in Fig. 6.8 It simply defines two employees (empl1 and empl2), assigns their names
and pay rates, and then computes and displays their pay based on the respective number of hours worked.

c 2001 J.E. Akin 121

[ 1] ! Multiple Inheritance and Polymorphism of the "print" function
[ 2] include ’class Person.inc’ ! also brings in class Date
[ 3] include ’class Student.inc’
[ 4] include ’class Professor.inc’
[ 5]
[ 6] program main
[ 7] use class Person ! no changes
[ 8] use class Student, print S => print ! renamed to print S
[ 9] use class Professor, print F => print ! renamed to print F
[10] implicit none
[11]
[12] ! Interface to generic routine, print, for any type argument
[13] interface print ! using renamed type dependent functions
[14] module procedure print Name, print S, print F
[15] end interface
[16]
[17] type (Person) :: x; type (Student) :: y; type (Professor) :: z
[18]
[19] x = Person ("Bob"); ! default constructor
[20] call print(x); ! print person type
[21]
[22] y = Student ("Tom", 3.47); ! default constructor
[23] call print(y); ! print student type
[24]
[25] z = Professor ("Ann", 7); ! default constructor
[26] call print(z); ! print professor type
[27] ! alternate constructors not used
[28] end program main ! Running gives:
[29] ! Bob
[30] ! My name is Tom, and my G.P.A. is 3.4700000.
[31] ! My name is Ann, and I have 7 publications.

Figure 6.6: Bringing Four Classes and Three Functions Together

[ 1] module class Employee

[ 2] ! The module class Employee contains both the
[ 3] ! data and functionality of an employee.
[ 4] !
[ 5] implicit none
[ 6] public :: setDataE, getNameE, payE ! the Functionality
[ 7]
[ 8] type Employee ! the Data
[ 9] private
[10] character(30) :: name
[11] real :: payRate ; end type Employee
[12]
[13] contains ! inherited internal variables and subprograms
[14]
[15] function setDataE (lastName, firstName, newPayRate) result (E)
[16] character(*), intent(in) :: lastName
[17] character(*), intent(in) :: firstName
[18] real, intent(in) :: newPayRate
[19] type (Employee) :: E ! employee
[20] ! use intrinsic constructor
[21] E = Employee((trim(firstName)//" "//trim(lastName)),newPayRate)
[22] end function setDataE
[23]
[24] function getNameE ( Person ) result (n)
[25] type (Employee), intent(in) :: Person
[26] character(30) :: n ! name
[27] n = Person % name ; end function getNameE
[28]
[29] function getRate ( Person ) result ( r )
[30] type (Employee), intent(in) :: Person
[31] real :: r ! rate
[32] r = Person % payRate ; end function getRate
[33]
[34] function payE ( Person, hoursWorked ) result ( amount )
[35] type (Employee), intent(in) :: Person
[36] real, intent(in) :: hoursWorked
[37] real :: amount
[38] amount = Person % payRate * hoursWorked ; end function payE
[39] end module class Employee

Figure 6.7: First Definition of an Employee Class

Note that both empl1 and empl2 are each an instance of a class, and therefore they are objects and thus
distinctly different from a class.

c 2001 J.E. Akin 122

[ 1] program main
[ 2] ! Example use of employees
[ 3] use class Employee
[ 4] type (Employee) empl1, empl2
[ 5]
[ 6] ! Set up 1st employee and print out his name and pay
[ 7] empl1 = setDataE ( "Jones", "Bill", 25.0 )
[ 8] print *, "Name: ", getNameE ( empl1 )
[ 9] print *, "Pay: ", payE ( empl1, 40.0 )
[10]
[11] ! Set up 2nd employee and print out her name and pay
[12] empl2 = setDataE ( "Doe", "Jane", 27.5 )
[13] print *, "Name: ", getNameE ( empl2 )
[14] print *, "Pay: ", payE ( empl2, 38.0 )
[15] end program main ! Running produces;
[16] ! Name: Bill Jones ! Pay: 1000.
[17] ! Name: Jane Doe ! Pay: 1045.

Figure 6.8: First Test of an Employee Class

Next we deal with a manager which Is-A “kind of” employee. One difference is that some managers
may be paid a salary rather than an hourly rate. Thus we have the Manager class inherit the attributes
of the Employee class and add a new logical attribute isSalaried which is true when the manager is
salary based. To support such a case we must add a new member setSalaried which can turn the new
attribute on or off, and a corresponding member payM that uses the isSalaried flag when computing
the pay. The class Manager module is shown in Fig. 6.9 Note that the constructor Manager defaults
to an hourly worker (line 33) and it uses the inherited employee constructor (line 31). Figure 6.10 shows
a test program to validate the manager class (and indirectly the employee class). It defines a salaried
manager, mgr1, an hourly manager mgr2, and prints the name and weekly pay for both. (Verify these
weekly pay amounts.)
With these two classes we have mainly used different program names for members that do similar
things in each class (the author’s preference). However, many programmers prefer to use a single member
name for a typical operation, regardless of the class of the operand. We also restricted all the attributes
to private and allowed all the members to be public. We could use several alternate approaches to
building our Employee and Manager classes. For example, assume we want a single member name
called pay to be invoked for an employee, or manager (or executive). Furthermore we will allow the
attributes to be public instead of private. Lowering the access restrictions to the attributes makes
it easier to write an alternate program, but it is not a recommended procedure since it breaks the data
hiding concept that has been shown to be important to OO software maintenance and reliability. The
alternate Employee and Manager classes are shown in Figs. 6.11 and 6.12, respectively. Note that
they both have a pay member but their arguments are of different classes and their internal calculations
are different. Now we want a validation program that will create both classes of individuals, and use
a single member name, PrintPay, to print the proper pay amount from the single member name pay.
This can be done in different ways. One problem that arises in our plan to reuse the code in the two
alternate class modules is that neither contained a pay printing member. We will need two new routines,
PrintPayEmployee and PrintPayManager, and a generic or polymorphic interface to them. We have
at least three ways to do this. One way is to place the two routines in an external file (or external to main
if in the same file), leave the two class modules unchanged, and have the main program begin with (or
INCLUDE) an external interface prototype. This first approach to main is shown in Fig. 6.13. Note that
the two new external routines must each use their respective class module.
A second approach would be to have the two new routines become internal to the main, after line
30, and occur before end program. Another change would be that each routine would have to omit its
use statement (such as lines 34 and 41). Why? Because they are now internal to main and it has already
made use of the two classes (in line 2). That approach is shown in Figs. 6.13
A third approach would be the most logical and consistent with OOP principles. It is to make all the
class attributes private, place the print members in each respective class, insert a single generic name
interface in each class, and modify the main program to use the polymorphic name regardless of the class
of the argument it acts upon. The improved version of the classes are given below in Figs. 6.14, 6.15, and
6.16. Observe that generic interfaces for PrintPay and getName have been added, but that we could

c 2001 J.E. Akin 123

[ 1] module class Manager
[ 2] ! Gets class Employee and add additional functionality
[ 3] use class Employee
[ 4] implicit none
[ 5] public :: setSalaried, payM
[ 6]
[ 7] type Manager ! the Data
[ 8] private
[ 9] type (Employee) :: Person
[10] integer :: isSalaried ! ( or logical )
[11] end type Manager
[12]
[13] contains ! inherited internal variables and subprograms
[14]
[15] function getEmployee ( M ) result (E)
[16] type (Manager ), intent(in) :: M
[17] type (Employee) :: E
[18] E = M % Person ; end function getEmployee
[19]
[20] function getNameM ( M ) result (n)
[21] type (Manager ), intent(in) :: M
[22] type (Employee) :: E
[23] character(30) :: n ! name
[24] n = getNameE(M % Person); end function getNameM
[25]
[26] function Manager (lastName, firstName, newPayRate) result (M)
[27] character(*), intent(in) :: lastName
[28] character(*), intent(in) :: firstName
[29] real, intent(in) :: newPayRate
[30] type (Employee) :: E ! employee
[31] type (Manager ) :: M ! manager constructor
[32] E = setDataE (lastName, firstName, newPayRate)
[33] ! use intrinsic constructor
[34] M = Manager(E, 0) ! default to no salary
[35] end function Manager
[36]
[37] function setDataM (lastName, firstName, newPayRate) result (M)
[38] character(*), intent(in) :: lastName
[39] character(*), intent(in) :: firstName
[40] real, intent(in) :: newPayRate
[41] type (Employee) :: E ! employee
[42] type (Manager ) :: M ! manager
[43] E = setDataE (lastName, firstName, newPayRate)
[44] M % Person = E
[45] end function setDataM
[46]
[47] subroutine setSalaried ( Who, salariedFlag )
[48] type (Manager), intent(inout) :: Who
[49] integer, intent(in) :: salariedFlag
[50] Who % isSalaried = salariedFlag ; end subroutine setSalaried
[51]
[52] function payM ( Human, hoursWorked ) result ( amount )
[53] type (Manager), intent(in) :: Human
[54] real, intent(in) :: hoursWorked
[55] real :: amount, value
[56] value = getRate( getEmployee(Human) )
[57] if ( Human % isSalaried == 1 ) then ! (or use logical)
[58] amount = value
[59] else
[60] amount = value * hoursWorked
[61] end if ; end function payM
[62] end module class Manager

Figure 6.9: A First Declaration of a Manager Class

not do that for a corresponding setData; do you know why? A final improvement will be given as an
assignment.

6.3 Polymorphism
Fortran 90 and 95 do not include the full range of polymorphism abilities that one would like to have in
an object-oriented language. It is expected that the Fortran 2000 standard will add those abilities.
Some of the code “re-use” features can be constructed through the concept of subprogram “tem-
plates,” which will be discussed below. The lack of a standard “Is A” polymorphism can be overcome in
F90/95 by the use of the SELECT CASE feature to define “sub-types” of objects. This approach of sub-
typing programming provides the desired additional functionality, but it is clearly not as easy to change
or extend as an inheritance feature built into the language standard. A short example will be provided.

c 2001 J.E. Akin 124

[ 1] program main ! Example use of managers
[ 2] use class Manager
[ 3] implicit none
[ 4] type (Manager) mgr1, mgr2
[ 5]
[ 6] ! Set up 1st manager and print out her name and pay
[ 7]
[ 8] mgr1 = setDataM ( "Smith", "Kimberly", 1900.0 )
[ 9] call setSalaried ( mgr1, 1 ) ! Has a salary
[10]
[11] print *, "Name: ", getNameM ( mgr1)
[12] print *, "Pay: ", payM ( mgr1, 40.0 )
[13]
[14] ! Set up 2nd manager and print out his name and pay
[15]
[16] ! mgr2 = setDataM ( "Danish", "Tom", 46.5 )
[17] ! call setSalaried ( mgr2, 0 ) ! Doesn’t have a salary
[18] ! or
[19] mgr2 = Manager ( "Danish", "Tom", 46.5 )
[20]
[21] print *, "Name: ", getNameM ( mgr2)
[22] print *, "Pay: ", payM ( mgr2, 40.0 )
[23] end program main ! Running produces;
[24] ! Name: Kimberly Smith ! Pay: 1900.
[25] ! Name: Tom Danish ! Pay: 1860.

Figure 6.10: First Test of a Manager Class

[ 1] module class Employee ! Alternate

[ 2] implicit none
[ 3] public :: setData, getName, pay ! the Functionality
[ 4]
[ 5] type Employee ! the Data
[ 6] character(30) :: name
[ 7] real :: payRate
[ 8] end type Employee
[ 9]
[10] contains ! inherited internal variables and subprograms
[11]
[12] subroutine setData ( Person, lastName, firstName, newPayRate )
[13] type (Employee) :: Person
[14] character(*) :: lastName
[15] character(*) :: firstName
[16] real :: newPayRate
[17] Person % name = trim (firstName) // " " // trim (lastName)
[18] Person % payRate = newPayRate
[19] end subroutine setData
[20]
[21] function getName ( Person )
[22] character(30) :: getName
[23] type (Employee) :: Person
[24] getName = Person % name
[25] end function getName
[26]
[27] function pay ( Person, hoursWorked )
[28] real :: pay
[29] type (Employee) :: Person
[30] real :: hoursWorked
[31] pay = Person % payRate * hoursWorked
[32] end function pay
[33] end module class Employee

Figure 6.11: Alternate Public Access Form of an Employee Class

6.3.1 Templates
One of our goals has been to develop software that can be reused for other applications. There are some
algorithms that are effectively independent of the object type on which they operate. For example, in a
sorting algorithm one often needs to interchange, or swap, two objects. A short routine for that purpose
follows:
subroutine swap integers (x, y)
implicit none
integer, intent(inout) :: x, y
integer :: temp
temp = x
x = y
y = temp
end subroutine swap integers

c 2001 J.E. Akin 125

[ 1] module class Manager ! Alternate
[ 2] use class Employee, payEmployee => pay ! renamed
[ 3] implicit none
[ 4] public :: setSalaried, payManager
[ 5]
[ 6] type Manager ! the Data
[ 7] type (Employee) :: Person
[ 8] integer :: isSalaried ! ( or logical )
[ 9] end type Manager
[10]
[11] contains ! inherited internal variables and subprograms
[12]
[13] subroutine setSalaried ( Who, salariedFlag )
[14] type (Manager) :: Who
[15] integer :: salariedFlag
[16] Who % isSalaried = salariedFlag
[17] end subroutine setSalaried
[18]
[19] function pay ( Human, hoursWorked )
[20] real :: pay
[21] type (Manager) :: Human
[22] real :: hoursWorked
[23]
[24] if ( Human % isSalaried == 1 ) then ! (or use logical)
[25] pay = Human % Person % payRate
[26] else
[27] pay = Human % Person % payRate * hoursWorked
[28] end if
[29] end function pay
[30] end module class Manager

Figure 6.12: Alternate Public Access Form of a Manager Class

Observe that in this form it appears necessary to have one version for integer arguments and another for
real arguments. Indeed we might need a different version of the routine for each type of argument that
you may need to swap. A slightly different approach would be to write our swap algorithm as:
subroutine swap objects (x, y)
implicit none
type (Object), intent(inout) :: x, y
type (Object) :: temp
temp = x
x = y
y = temp
end subroutine swap objects

which would be a single routine that would work for any Object, but it has the disadvantage that one
find a way to redefine the Object type for each application of the routine. That would not be an easy
task. (While we will continue with this example with the algorithm in the above forms it should be noted
that the above approaches would not be efficient if x and y were very large arrays or derived type objects.
In that case we would modify the algorithm slightly to employ pointers to the large data items and simply
swap the pointers for a significant increase in efficiency.)
Consider ways that we might be able to generalize the above routines so that they could accept and
swap any specific type of arguments. For example, the first two versions could be re-written in a so called
template form as:
subroutine swap Template$ (x, y)
implicit none
Template$, intent(inout) :: x, y
Template$ :: temp
temp = x
x = y
y = temp
end subroutine swap Template$

In the above template the dollar sign ($) was includes in the “wild card” because while it is a valid member
of the F90 character set it is not a valid character for inclusion in the name of a variable, derived type,
function, module, or subroutine. In other words, a template in the illustrated form would not compile, but
such a name could serve as a reminder that its purpose is to produce a code that can be compiled after the
“wild card” substitutions have been made.
With this type of template it would be very easy to use a modern text editor to do a global substitution
of any one of the intrinsic types character, complex, double precision, integer, logi-
cal, or real for the “wild card” keyword Template$ to produce a source code to swap any or all of

c 2001 J.E. Akin 126

[ 1] program main ! Alternate employee and manager classes
[ 2] use class Manager ! and thus Employee
[ 3] implicit none
[ 4] ! supply interface for external code not in classes
[ 5] interface PrintPay ! For TYPE dependent arguments
[ 6] subroutine PrintPayManager ( Human, hoursWorked )
[ 7] use class Manager
[ 8] type (Manager) :: Human
[ 9] real :: hoursWorked
[10] end subroutine
[11] subroutine PrintPayEmployee ( Person, hoursWorked )
[12] use class Employee
[13] type (Employee) :: Person
[14] real :: hoursWorked
[15] end subroutine
[16] end interface
[17]
[18] type (Employee) empl ; type (Manager) mgr
[19]
[20] ! Set up an employee and print out his name and pay
[21] call setData ( empl, "Burke", "John", 25.0 )
[22]
[23] print *, "Name: ", getName ( empl )
[24] call PrintPay ( empl, 40.0 )
[25]
[26] ! Set up a manager and print out her name and pay
[27] call setData ( mgr % Person, "Kovacs", "Jan", 1200.0 )
[28] call setSalaried ( mgr, 1 ) ! Has a salary
[29]
[30] print *, "Name: ", getName ( mgr % Person )
[31] call PrintPay ( mgr, 40.0 )
[32] end program
[33]
[34] subroutine PrintPayEmployee ( Person, hoursWorked )
[35] use class Employee
[36] type (Employee) :: Person
[37] real :: hoursWorked
[38] print *, "Pay: ", pay ( Person, hoursworked )
[39] end subroutine
[40]
[41] subroutine PrintPayManager ( Human, hoursWorked )
[42] use class Manager
[43] type (Manager) :: Human
[44] real :: hoursWorked
[45] print *, "Pay: ", pay ( Human , hoursworked )
[46] end subroutine
[47] ! Running produces;
[48] ! Name: John Burke
[49] ! Pay: 1000.
[50] ! Name: Jan Kovacs
[51] ! Pay: 1200.

Figure 6.13: Testing the Alternate Employee and Manager Classes

the intrinsic data types. There would be no need to keep up with all the different routine names if we
placed all of them in a single module and also created a generic interface to them such as:
module swap library
implicit none
interface swap ! the generic name
module procedure swap character, swap complex
module procedure swap double precision, swap integer
module procedure swap logical, swap real
end interface
contains
subroutine swap characters (x, y)
. . .
end subroutine swap characters
subroutine swap . . .
. . .
end module swap library

The use of a text editor to make such substitutions is not very elegant and we expect that there may
be a better way to pursue the concept of developing a reuseable software template. The concept of a text
editor substitution also fails when we go to the next logical step and try to use a derived type argument
instead of any of the intrinsic data types. For example, if we were to replace the “wild card” with our
previous type (chemical element) that would create:
subroutine swap type (chemical element) (x,y)
implicit none

c 2001 J.E. Akin 127

[ 1] module class Employee ! the base class
[ 2] implicit none ! strong typing
[ 3] private :: PrintPayEmployee, payE ! private members
[ 4] type Employee ! the Data
[ 5] private ! all attributes private
[ 6] character(30) :: name
[ 7] real :: payRate ; end type Employee
[ 8]
[ 9] interface PrintPay ! a polymorphic member
[10] module procedure PrintPayEmployee ; end interface
[11] interface getName ! a polymorphic member
[12] module procedure getNameE ; end interface
[13] ! NOTE: can not have polymorphic setData. Why ?
[14]
[15] contains ! inherited internal variables and subprograms
[16]
[17] function setDataE (lastName, firstName, newPayRate) result (E)
[18] character(*), intent(in) :: lastName
[19] character(*), intent(in) :: firstName
[20] real, intent(in) :: newPayRate ! amount per period
[21] type (Employee) :: E ! employee
[22] ! use intrinsic constructor
[23] E = Employee((trim(firstName)//" "//trim(lastName)),newPayRate)
[24] end function setDataE
[25]
[26] function getNameE ( Person ) result (n)
[27] type (Employee), intent(in) :: Person
[28] character(30) :: n ! name
[29] n = Person % name ; end function getNameE
[30]
[31] function getRate ( Person ) result ( r )
[32] type (Employee), intent(in) :: Person
[33] real :: r ! rate of pay
[34] r = Person % payRate ; end function getRate
[35]
[36] function payE ( Person, hoursWorked ) result ( amount )
[37] type (Employee), intent(in) :: Person
[38] real, intent(in) :: hoursWorked
[39] real :: amount
[40] amount = Person % payRate * hoursWorked ; end function payE
[41]
[42] subroutine PrintPayEmployee ( Person, hoursWorked )
[43] type (Employee) :: Person
[44] real :: hoursWorked
[45] print *, "Pay: ", payE ( Person, hoursworked )
[46] end subroutine
[47] end module class Employee

Figure 6.14: A Better Private Access Form of an Employee Class

[ 1] module class Manager ! the derived class

[ 2] ! Get class Employee, add additional attribute & members
[ 3] use class Employee ! inherited base class
[ 4] implicit none ! strong typing
[ 5] private :: PrintPayManager, payM, getNameM ! private members
[ 6]
[ 7] type Manager ! the Data
[ 8] private ! all attributes private
[ 9] type (Employee) :: Person
[10] integer :: isSalaried ! 1 if true (or use logical)
[11] end type Manager
[12]
[13] interface PrintPay ! a polymorphic member
[14] module procedure PrintPayManager ; end interface
[15] interface getName ! a polymorphic member
[16] module procedure getNameM ; end interface

Fig. 6.15: A Better Private Access Form of a Manager Class (continued)

type (chemical element), intent (inout)::x,y

type (chemical element) ::temp
temp = x
x = y
y = temp
end subroutine swap type (chemical element)

This would fail to compile because it violates the syntax for a valid function or subroutine name, as well
as the end function or end subroutine syntax. Except for the first and last line syntax errors this would be
a valid code. To correct the problem we simply need to add a little logic and omit the characters type

c 2001 J.E. Akin 128

[17]
[18] contains ! inherited internal variables and subprograms
[19]
[20] function getEmployee ( M ) result (E)
[21] type (Manager ), intent(in) :: M
[22] type (Employee) :: E
[23] E = M % Person ; end function getEmployee
[24]
[25] function getNameM ( M ) result (n)
[26] type (Manager ), intent(in) :: M
[27] type (Employee) :: E
[28] character(30) :: n ! name
[29] n = getNameE(M % Person); end function getNameM
[30]
[31] function Manager (lastName, firstName, newPayRate) result (M)
[32] character(*), intent(in) :: lastName
[33] character(*), intent(in) :: firstName
[34] real, intent(in) :: newPayRate
[35] type (Employee) :: E ! employee
[36] type (Manager ) :: M ! manager constructed
[37] E = setDataE (lastName, firstName, newPayRate)
[38] ! use intrinsic constructor
[39] M = Manager(E, 0) ! default to hourly
[40] end function Manager
[41]
[42] function setDataM (lastName, firstName, newPayRate) result (M)
[43] character(*), intent(in) :: lastName
[44] character(*), intent(in) :: firstName
[45] real, intent(in) :: newPayRate ! hourly OR weekly
[46] type (Employee) :: E ! employee
[47] type (Manager ) :: M ! manager constructed
[48] E = setDataE (lastName, firstName, newPayRate)
[49] M % Person = E ; M % isSalaried = 0 ! default to hourly
[50] end function setDataM
[51]
[52] subroutine setSalaried ( Who, salariedFlag ) ! 0=hourly, 1=weekly
[53] type (Manager), intent(inout) :: Who
[54] integer, intent(in) :: salariedFlag ! 0 OR 1
[55] Who % isSalaried = salariedFlag ; end subroutine setSalaried
[56]
[57] function payM ( Human, hoursWorked ) result ( amount )
[58] type (Manager), intent(in) :: Human
[59] real, intent(in) :: hoursWorked
[60] real :: amount, value
[61] value = getRate( getEmployee(Human) )
[62] if ( Human % isSalaried == 1 ) then
[63] amount = value ! for weekly person
[64] else
[65] amount = value * hoursWorked ! for hourly person
[66] end if ; end function payM
[67]
[68] subroutine PrintPayManager ( Human, hoursWorked )
[69] type (Manager) :: Human
[70] real :: hoursWorked
[71] print *, "Pay: ", payM ( Human , hoursworked )
[72] end subroutine
[73] end module class Manager

Figure 6.15: A Better Private Access Form of a Manager Class

( ) when we create a function, module, or subroutine name that is based on a derived type data entity.
Then we obtain
subroutine swap chemical element (x,y)
implicit none
type (chemical element), intent (inout)::x,y
type (chemical element) ::temp
temp = x
x = y
y = temp
end subroutine swap chemical element

which yields a completely valid routine.

Unfortunately, text editors do not offer us such logic capabilities. However, as we have seen, high
level programming languages like C++ and F90 do have those abilities. At this point you should be able
to envision writing a pre-processor program that would accept a file of template routines, replace the
template “wildcard” words with the desired generic forms to produce a module or header file con-
taining the expanded source files that can then be brought into the desired program with an include or
use statement. The C++ language includes a template pre-processor to expand template files as needed.

c 2001 J.E. Akin 129

[ 1] program main ! Final employee and manager classes
[ 2] use class Manager ! and thus class Employee
[ 3] implicit none
[ 4]
[ 5] type (Employee) empl ; type (Manager) mgr
[ 6]
[ 7] ! Set up a hourly employee and print out his name and pay
[ 8] empl = setDataE ( "Burke", "John", 25.0 )
[ 9]
[10] print *, "Name: ", getName ( empl )
[11] call PrintPay ( empl, 40.0 ) ! polymorphic
[12]
[13] ! Set up a weekly manager and print out her name and pay
[14] mgr = setDataM ( "Kovacs", "Jan", 1200.0 )
[15] call setSalaried ( mgr, 1 ) ! rate is weekly
[16]
[17] print *, "Name: ", getName ( mgr )
[18] call PrintPay ( mgr, 40.0 ) ! polymorphic
[19] end program ! Running produces;
[20] ! Name: John Burke
[21] ! Pay: 1000.
[22] ! Name: Jan Kovacs
[23] ! Pay: 1200.

Figure 6.16: Testing the Better Employee-Manager Forms

Some programmers criticize F90/95 for not offering this ability as part of the standard. A few C++ pro-
grammers criticize templates and advise against their use. Regardless of the merits of including template
pre-processors in a language standard it should be clear that it is desirable to plan software for its efficient
reuse.
With F90 if one wants to take advantage of the concepts of templates then the only choices are to carry
out a little text editing or develop a pre-processor with the outlined capabilities. The former is clearly
the simplest and for many projects may take less time than developing such a template pre-processor.
However, if one makes the time investment to produce a template pre-processor one would have a tool
that could be applied to basically any coding project.
6.3.2 Subtyping Objects (Dynamic Dispatching)
One polymorphic feature missing from the Fortran 90 standard (but expected in Fortran 2000) that is
common to most object oriented languages is called run-time polymorphism or dynamic dispatching. In
the C++ language this ability is introduced in the so-called “virtual function.” To emulate this ability
is quite straightforward in F90 but is not elegant since it usually requires a group of if-elseif statements
or other selection processes. It is only tedious if the inheritance hierarchy contains many unmodified
subroutines and functions. The importance of the lack of a standardized dynamic dispatching depends on
the problem domain to which it must be applied. For several applications demonstrated in the literature
the alternate use of subtyping has worked quite well and resulted in programs that have been shown to
run several times faster than equivalent C++ versions.
We implement dynamic dispatching in F90 by a process often called subtyping. Two features must be
constructed to do this. First, a pointer object, which can point to any subtype member in an inheritance
hierarchy, must be developed. Second, an if-elseif or other selection method is developed to serve as a
dispatch mechanism to select the unique appropriate procedure to be executed based on the actual class
referenced in the controlling pointer object. This subtyping process is also referred to as implementing
a polymorphic class. Of course, the details of the actual dispatching process can be hidden from the
procedures that utilize the polymorphic class.
This process will be illustrated buy creating a specific polymorphic class, called
Is A Member Class, which has polymorphic procedures named new, assign, and display. They will
construct a new instance of the object, assign it a value, and list its components. The minimum example
of such a process requires two members and is easily extended to any number of member classes. We
begin by defining each of the subtype classes of interest.
The first is a class, Member 1 Class, which has two real components and the encapsulated func-
tionality to construct a new instance and another to accept a pointer to such a subtype and display related
information. It is shown in Fig. 6.17. The next class, Member 2 Class, has three components: two
reals and one of type Member 1. It has the same sort of functionality, but clearly must act on more

c 2001 J.E. Akin 130

[ 1] Module Member 1 Class
[ 2] implicit none
[ 3] type member 1
[ 4] real :: real 1, real 2
[ 5] end type member 1
[ 6]
[ 7] contains
[ 8]
[ 9] subroutine new member 1 (member, a, b)
[10] real, intent(in) :: a, b
[11] type (member 1) :: member
[12] member%real 1 = a ; member%real 2 = b
[13] end subroutine new member 1
[14]
[15] subroutine display memb 1 (pt to memb 1, c)
[16] type (member 1), pointer :: pt to memb 1
[17] character(len=1), intent(in) :: c
[18] print *, ’display memb 1 ’, c
[19] end subroutine display memb 1
[20]
[21] End Module Member 1 Class

Figure 6.17: Defining Subtype 1

[ 1] Module Member 2 Class
[ 2] Use Member 1 class
[ 3] implicit none
[ 4] type member 2
[ 5] type (member 1) :: r 1 2
[ 6] real :: real 3, real 4
[ 7] end type member 2
[ 8]
[ 9] contains
[10]
[11] subroutine new member 2 (member, a, b, c, d)
[12] real, intent(in) :: a, b, c, d
[13] type (member 2) :: member
[14] call new member 1 (member%r 1 2, a, b)
[15] member%real 3 = c ; member%real 4 = d
[16] end subroutine new member 2
[17]
[18] subroutine display memb 2 (pt to memb 2, c)
[19] type (member 2), pointer :: pt to memb 2
[20] character(len=1), intent(in) :: c
[21] print *, ’display memb 2 ’, c
[22] end subroutine display memb 2
[23]
[24] End Module Member 2 Class

Figure 6.18: Defining Subtype 2

components. It has also inherited the functionally from the Member 1 Class; as displayed in Fig. 6.18.
The polymorphic class is called the Is A Member Class and is shown in Fig. 6.19. It includes all
of the encapsulated data and function’s of the above two subtypes by including their use statements. The
necessary pointer object is defined as an Is A Member type that has a unique pointer for each subtype
member (two in this case). It also defines a polymorphic interface to each of the common procedures to
be applied to the various subtype objects. In the polymorphic function assign the dispatching is done very
simply. First, all pointers to the family of subtypes are nullified, and then the unique pointer component
to the subtype of interest is set to point to the desired member. The dispatching process for the display
procedure is different. It requires an if-elseif construct that contains calls to all of the possible subtype
members (two here) and a failsafe default state to abort the process or undertake the necessary exception
handling. Since all but one of the subtype pointer objects have been nullified it employs the associated
intrinsic function to select the one, and only, procedure to call and passes the pointer object on to that
procedure. The validation of this dynamic dispatching through a polymorphic class is shown in Fig. 6.20.
There a target is declared for reach possible subtype and then each of them is constructed and sent on
to the other polymorphic functions. The results clearly show that different display procedures were used
depending on the class of object supplied as an argument. It is expected that the new Fortran 2000
standard will allow such dynamic dispatching in a much simpler fashion.

c 2001 J.E. Akin 131

[ 1] Module Is A Member Class
[ 2] Use Member 1 Class ; Use Member 2 Class
[ 3] implicit none
[ 4]
[ 5] type Is A Member
[ 6] private
[ 7] type (member 1), pointer :: pt to memb 1
[ 8] type (member 2), pointer :: pt to memb 2
[ 9] end type Is A Member
[10]
[11] interface new
[12] module procedure new member 1
[13] module procedure new member 2
[14] end interface
[15]
[16] interface assign
[17] module procedure assign memb 1
[18] module procedure assign memb 2
[19] end interface
[20]
[21] interface display
[22] module procedure display memb 1
[23] module procedure display memb 2
[24] end interface
[25]
[26] contains
[27]
[28] subroutine assign memb 1 (Family, member)
[29] type (member 1), target, intent(in) :: member
[30] type (Is A Member), intent(out) :: Family
[31] call nullify Is A Member (Family)
[32] Family%pt to memb 1 => member
[33] end subroutine assign memb 1
[34]
[35] subroutine assign memb 2 (Family, member)
[36] type (member 2), target, intent(in) :: member
[37] type (Is A Member), intent(out) :: Family
[38] call nullify Is A Member (Family)
[39] Family%pt to memb 2 => member
[40] end subroutine assign memb 2
[41]
[42] subroutine nullify Is A Member (Family)
[43] type (Is A Member), intent(inout) :: Family
[44] nullify (Family%pt to memb 1)
[45] nullify (Family%pt to memb 2)
[46] end subroutine nullify Is A Member
[47]
[48] subroutine display members (A Member, c)
[49] type (Is A Member), intent(in) :: A Member
[50] character(len=1), intent(in) :: c
[51]
[52] ! select the proper member
[53] if ( associated (A Member%pt to memb 1) ) then
[54] call display (A Member%pt to memb 1, c)
[55] else if ( associated (A Member%pt to memb 2) ) then
[56] call display (A Member%pt to memb 2, c)
[57] else ! default case
[58] stop ’Error, no member defined in Is A Member Class’
[59] end if
[60] end subroutine display members
[61] End Module Is A Member Class

Figure 6.19: Combining Subtypes in an Is A Class

c 2001 J.E. Akin 132

[ 1] program main
[ 2] use Is A Member Class
[ 3] implicit none
[ 4]
[ 5] type (Is A Member) :: generic member
[ 6] type (member 1), target :: pt to memb 1
[ 7] type (member 2), target :: pt to memb 2
[ 8] character(len=1) :: c
[ 9]
[10] c = ’A’
[11] call new (pt to memb 1, 1.0, 2.0)
[12] call assign (generic member, pt to memb 1)
[13] call display members (generic member, c)
[14]
[15] c = ’B’
[16] call new (pt to memb 2, 1.0, 2.0, 3.0, 4.0)
[17] call assign (generic member, pt to memb 2)
[18] call display members (generic member, c)
[19]
[20] end program main
[21] ! running gives
[22] ! display memb 1 A
[23] ! display memb 2 B

Figure 6.20: Testing the Is A Subtypes

6.4 Exercises
1. Write a main program that will use the Class X and Class Y, given below, to invoke each of the f(v)
routines and assign a value of 66 to the integer component in X, and 44 to the integer component in Y.
(Solution given.)
module class X
public :: f
type X ; integer a; end type X
contains ! functionality
subroutine f(v); type (X ), intent(in) :: v
print *,"X f() executing"; end subroutine
end module class X
module class Y
use class X, X f => f ! renamed
public :: f
type Y ; integer a; end type Y ! dominates X a
contains ! functionality, overrides X f()
subroutine f(v); type (Y ), intent(in) :: v
print *,"Y f() executing"; end subroutine
end module class Y

2. Create the generic interface that would allow a single constructor name, Position Angle , to be
used for all the constructors given in the previous chapter for the class Position Angle. Note that this
is possible because they all had unique argument signatures. Also provide a new main program to test
this polymorphic version.
3. Modify the last Manager class by deleting the member setDataM and replace its appearance in the
last main with an existing constructor (but not used) in that class. Also provide a generic setData
interface in the class Employee as a nicer name and to allow for other employees, like executives, that
may have different kinds of attributes that may need to be set in the future. Explain why we could not use
setDataM in the generic setData.

4. The final member setDataE in Employee is actually a constructor and the name is misleading since
it does not just set data values, it also builds the name. Rename setDataE to the constructor notation
Employee and provide a new member in Employee called setRateE that only sets the employee pay
rate.

c 2001 J.E. Akin 133

134
Chapter 7

OO Data Structures

7.1 Data Structures

We have seen that F90 has a very strong intrinsic base for supporting the use of subscripted arrays.
Fortran arrays can contain intrinsic data types as well as user defined types (i.e., ADT’s). One can not
directly have an array of pointers but you are allowed to have an array contain defined types that are
pointers or that have components that are pointers. Arrays offer an efficient way to contain information
and to insert and extract information. However, there are many times when creating an efficient algorithm
dictates that we use some specialized storage method, or container, and a set of operations to act with that
storage mode. The storage representation and the set of operations that are allowed for it are known as a
data structure. How you store and retrieve an item from a container is often independent of the nature of
the item itself. Thus, different instances of a data structure may produce containers for different types of
objects. Data structures have the potential for a large amount of code reuse, which is a basic goal of OOP
methods. In the following sections we will consider some of the more commonly used containers.

7.2 Stacks
A stack is a data structure where access is restricted to the last inserted object. It is referred to as a last-in
first-out (LIFO) container. In other words, a stack is a container to which elements may only be inserted
or removed at one end of the container, called the top of the stack. It behaves much like a pile of dinner
plates. You can place a new element on the pile (widely known as a push), remove the top element from
the pile (widely known as a pop), and identify the element on the top of the pile. You can also have the
general concept of an empty pile, and possibly a full pile if it is associated with some type of restrictive
container. Since at this point we only know about using arrays as containers we will construct a stack
container by using an array.
Assume that we have defined the attributes of the “Object” that is to use our container by building a
module called object type. Then we could declare the array implementation of a stack type to be:

module stack type

use object type ! to define objects in the stack
implicit none

integer, parameter :: limit = 999 ! stack size limit

type stack
private
integer :: size ! size of array
integer :: top ! top of stack
type (Object) :: a(limit) ! stack items array
end type stack
end module stack type

c 2001 J.E. Akin 135

 The interface contract to develop one such stack support system (or ADT) is given as:
module stack of objects
implicit none
public :: stack, push on Stack, pop from Stack, &
is Stack Empty, is Stack Full
interface ! for a class Stack contract
function make Stack (n) result (s) ! constructor
use stack type ! to define stack structure
integer, optional :: n ! size of stack
type (stack) :: s ! the new stack
end function make Stack
subroutine push on Stack (s, item) ! push item on top of stack
use stack type ! for stack structure
type (stack), intent(inout) :: s
type (Object), intent(in) :: item
end subroutine push on Stack
function pop from Stack (s) result (item) ! pop item from top
use stack type ! for stack structure
type (stack), intent(inout) :: s
type (Object) :: item
end function pop from Stack
function is Stack Empty (s) result (b) ! test stack
use stack type ! for stack structure
type (stack), intent(in) :: s
logical :: b
end function is Stack Empty
function is Stack Full (s) result (b) ! test stack
use stack type ! for stack structure
type (stack), intent(in) :: s
logical :: b
end function is Stack Full
end interface
end module stack of objects

In the interface we see that some of the member services ( is Stack Empty and is Stack Full)
are independent of the contained objects. Others (pop from Stack and push on Stack) explicitly
depend on the Object utilizing the container. Of course, the constructor (here make Stack) always
indirectly relates to the Object being contained in the array. The full details of a Stack class are given in
Fig. 7.1.
For a specific implementation test we will simply utilize objects that have a single integer attribute.
That is, we define the object of interest by a code segment like:
module object type
type Object
integer :: data ; end type ! one integer attribute
end module object type

Obviously, there are many other types of objects that one may want to create and place in a container like
a stack. At the present one would have to edit the above segment to define all the attributes of the object.
(Begin to think about how you might seek to automate such a process.) The new Stack class is tested
in Fig. 7.2, while a history of the example stack is sketched in Fig. 7.3. The only part of that code that
depends on a specific object is in line 7 where the (public) intrinsic constructor, Object, was utilized
rather that some more general constructor, say Object .
In Fig. 7.1 note that we have used an alternate syntax and specified the type of function result (logical,
Object, or stack) as a prefix to the function name (lines 16, 28, 36, 40). The author thinks that the form
used in the interface contract is easier to read and understand since it requires an extra line of code,
however some programmers prefer the condensed style of Fig. 7.1. Later we will examine an alternate
implementation of a stack by using a linked list.
The stack implementation shown here is not complete. For example, some programmers like to in-
clude a member, say show Stack top, to display the top element on the container without removing it
from the stack. Also we need to be concerned about pre-conditions that need to be satisfied for a member
and may require that we throw an exception message. You can not pop an item off of an empty stack, nor
can you push an item onto the top of a full stack. Only the member pop from Stack does such pre-
condition checking in the sample code. Note that members is Stack Empty and is Stack Full

c 2001 J.E. Akin 136

are called accessors, as would be show Stack top, since they query the container but do not change
it.
[ 1] module class Stack
[ 2] implicit none
[ 3] use exceptions ! to warn of errors
[ 4] use object type
[ 5] public :: stack, push on Stack, pop from Stack, &
[ 6] is Stack Empty, is Stack Full
[ 7] integer, parameter :: limit = 999 ! stack size limit
[ 8]
[ 9] type stack
[10] private
[11] integer :: size ! size of array
[12] integer :: top ! top of stack
[13] type (Object) :: a(limit) ! stack items array
[14] end type
[15] contains ! encapsulated functionality
[16]
[17] type (stack) function make Stack (n) result (s) ! constructor
[18] integer, optional :: n ! size of stack
[19] s%size = limit ; if ( present (n) ) s%size = n
[20] s%top = 0 ! object array not initialized
[21] end function make Stack
[22]
[23] subroutine push on Stack (s, item) ! push item on top of stack
[24] type (stack), intent(inout) :: s
[25] type (Object), intent(in) :: item
[26] s%top = s%top + 1 ; s%a(s%top) = item
[27] end subroutine push on Stack
[28]
[29] type (Object) function pop from Stack (s) result (item) ! off top
[30] type (stack), intent(inout) :: s
[31] if ( s%top < 1 ) then
[32] call exception ("pop from Stack","stack is empty")
[33] else
[34] item = s%a(s%top) ; s%top = s%top - 1
[35] end if ; end function pop from Stack
[36]
[37] logical function is Stack Empty (s) result (b)
[38] type (stack), intent(in) :: s
[39] b = ( s%top == 0 ) ; end function is Stack Empty
[40]
[41] logical function is Stack Full (s) result (b)
[42] type (stack), intent(in) :: s
[43] b = ( s%top == s%size ) ; end function is Stack Full
[44]
[45] end module class Stack

Figure 7.1: A Typical Stack Class

c 2001 J.E. Akin 137

[ 1] include ’class Stack.f’ ! previous figure
[ 2] program main
[ 3] use class Stack
[ 4] implicit none
[ 5] type (stack) :: b
[ 6] type (object) :: value, four, five, six
[ 7]
[ 8] four = Object(4) ; five = Object(5) ; six = Object(6) ! initialize
[ 9]
[10] b = make Stack(3) ! private constructor
[11] print *, is Stack Empty(b), is Stack Full(b) ! b = [], empty
[12]
[13] call push on Stack (b, four) ! b = [4]
[14] call push on Stack (b, five) ! b = [5,4]
[15] call push on Stack (b, six ) ! b = [6,5,4], full
[16] print *, is Stack Empty(b), is Stack Full(b) ! F T
[17]
[18] value = pop from Stack (b) ; print *, value ! b = [5,4]
[19] print *, is Stack Empty(b), is Stack Full(b) ! F F
[20]
[21] value = pop from Stack (b) ; print *, value ! b = [4]
[22] print *, is Stack Empty(b), is Stack Full(b) ! F F
[23]
[24] value = pop from Stack (b) ; print *, value ! b = [], empty
[25] print *, is Stack Empty(b), is Stack Full(b) ! T F
[26]
[27] value = pop from Stack (b) ! nothing to pop
[28] end program main ! running gives:
[29] ! T F ! F T
[30] ! 6 ! F F
[31] ! 5 ! F F
[32] ! 4 ! T F
[33] ! Exception occurred in subprogram pop from Stack
[34] ! With message: stack is empty

Figure 7.2: Testing a Stack of Objects

Full ? F F F T F F F F
Empty ? T F F F F F T T
Error ? N N N N N N N Y
Stack: | | |4| |5| |6| |5| |4| | | | |
| | | | |4| |5| |4| | | | | | |
| | | | | | |4| | | | | | | | |
--- --- --- --- --- --- --- ---
(Line) 9 12 13 14 17 20 23 26

Figure 7.3: Steps in the Stack Testing

<
E ----- > Top
|D|
|C| ---------
|B| Front <---- A B C D E <---- Rear
|A| ---------
---
a) Stack b) Queue

Figure 7.4: Simple Containers

c 2001 J.E. Akin 138

7.3 Queues
A comparison of a stack and another simple container, a queue, is given in Fig. 7.4. Its name queue
comes from the British word which means waiting in a line for service. A queue is a container into which
elements may be inserted at one end, called the rear, and leave only from the other end, called the front.
The first element in the queue expects to be the first serviced and, thus, be the first out of line. A queue
is a first-in first-out (FIFO) container system. In planning our first queue container we will again make
use of an array of objects. Doing so one quickly finds that you are much less likely to encounter a full
queue if it is stored as a so-called fixed circular array with a total of Q Size Limit storage slots. At
this point we define the structure of our queue to be:

module Queue type

! A queue stored as a so-called fixed circular array with a total
! of Q Size Limit storage slots; requires remainder function, mod.
! (version 1, i.e., without allocatable arrays and pointers)
use object type ! to define objects in the Container
implicit none
integer, parameter :: Q Size Limit = 999
type Queue
private
integer :: head ! index of first element
integer :: tail ! index of last element
integer :: length ! size of used storage
type (Object) :: store (Q Size Limit) ! a circular array
end type Queue
end module Queue type

An interface contract that will allow us to build a typical queue is:

module Queue of Objects

implicit none
public :: Queue, Add to Q, Create Q, Get Front of Q, Is Q Empty,&
Is Q Full, Get Length of Q, Remove from Q
interface ! for a class Queue contract
subroutine Add to Q (Q, item) ! add to tail of queue
use Queue type ! for Queue structure
type (Queue), intent(inout) :: Q
type (Object), intent(in) :: item ; end Subroutine Add to Q
function Create Q (N) result (Q) ! manual constructor
use Queue type ! for Queue structure
integer, intent(in) :: N ! size of the new array
type (Queue) :: Q ; end function Create Q
function Get Capacity of Q (Q) result (item)
use Queue type ! for Queue structure
type (Queue), intent(in) :: Q
type (Object) :: item ; end function Get Capacity of Q
function Get Front of Q (Q) result (item)
use Queue type ! for Queue structure
type (Queue), intent(in) :: Q
type (Object) :: item ; end function Get Front of Q
function Is Q Empty (Q) result(B)
use Queue type ! for Queue structure
type (Queue), intent(in) :: Q
logical :: B ; end function Is Q Empty
function Is Q Full (Q) result(B)
use Queue type ! for Queue structure
type (Queue), intent(in) :: Q
logical :: B ; end function Is Q Full
function Get Length of Q (Q) result (N)
use Queue type ! for Queue structure
type (Queue), intent(in) :: Q
integer :: N ; end function Get Length of Q
subroutine Remove from Q (Q) ! remove from head of queue
use Queue type ! for Queue structure
type (Queue), intent(inout) :: Q; end subroutine Remove from Q
end interface
end module Queue of Objects

c 2001 J.E. Akin 139

For a specific version we provide full details for objects containing an integer in Fig. 7.5, and test and
display the validity of the implementation in Fig. 7.6, where again the objects are taken to be integers
(lines 15, 19, 20).

[ 1] module class Queue ! file: class Queue.f90

[ 2]
[ 3] ! A queue stored as a so-called fixed circular array with a total of
[ 4] ! Q Size Limit storage slots; requires remainder function, mod.
[ 5] ! (i.e., without allocatable arrays and pointers)
[ 6]
[ 7] use exceptions ! inherit exception handler
[ 8] implicit none
[ 9]
[10] public :: Queue, Add to Q, Create Q, Get Front of Q
[11] Is Q Full, Get Length of Q, Remove from
[12]
[13] integer, parameter :: Q Size Limit = 3
[14]
[15] type Queue
[16] private
[17] integer :: head ! index of first element
[18] integer :: tail ! index of last element
[19] integer :: length ! size of used storage
[20] integer :: store (Q Size Limit) ! a circular array of elements
[21] end type Queue
[22]
[23] contains ! member functionality
[24]
[25] Subroutine Add to Q (Q, item) ! add to tail of queue
[26] type (Queue), intent(inout) :: Q
[27] integer, intent(in) :: item
[28]
[29] if ( Is Q Full(Q) ) call exception ("Add to Q","full Q")
[30] Q%store (Q%tail) = item
[31] Q%tail = 1 + mod (Q%tail, Q Size Limit)
[32] Q%length = Q%length + 1 ; end Subroutine Add to Q
[33]
[34] type (Queue) function Create Q (N) result (Q) ! manual constructor
[35] integer, intent(in) :: N ! size of the new array
[36] integer :: k ! implied loop
[37]
[38] if (N > Q Size Limit) call exception("Create Q","increase size")
[39] Q = Queue (1, 1, 0, (/ (0, k=1,N) /)) ! intrinsic constructor
[40] end function Create Q
[41]
[42] integer function Get Capacity of Q (Q) result (item)
[43] type (Queue), intent(in) :: Q
[44]
[45] item = Q size Limit - Q%length ; end function Get Capacity
[46]
[47] integer function Get Front of Q (Q) result (item)
[48] type (Queue), intent(in) :: Q
[49]
[50] if (Is Q Empty(Q)) call exception("Get Front of Q","em
[51] item = Q%store (Q%head) ; end function Get Front of Q
[52]
[53] logical function Is Q Empty (Q) result(B)
[54] type (Queue), intent(in) :: Q
[55]
[56] B = (Q%length == 0) ; end function Is Q Empty
[57]
[58] logical function Is Q Full (Q) result(B)
[59] type (Queue), intent(in) :: Q
[60]
[61] B = (Q%length == Q Size Limit) ; end function Is Q Full
[62]
[63] integer function Get Length of Q (Q) result (N)
[64] type (Queue), intent(in) :: Q
[65] N = Q%length ; end function Get Length of Q
[66]
[67] subroutine Remove from Q (Q) ! remove from head of queue
[68] type (Queue), intent(inout) :: Q
[69]
[70] if (Is Q Empty(Q)) call exception("Remove from Q","empty Q"
[71] Q%head = 1 + mod (Q%head, Q Size Limit)
[72] Q%length = Q%length - 1 ; end subroutine Remove from Q
[73]
[74] end module class Queue ! file: class Queue.f

Figure 7.5: A Typical Queue Class

c 2001 J.E. Akin 140

[ 1] program main
[ 2] use class Queue ! inherit its methods & class global constants
[ 3] implicit none
[ 4]
[ 5] type (Queue) :: C, B ! not used, used
[ 6] integer :: value, limit = 3 ! work items
[ 7]
[ 8] C = Create Q (limit) ! private constructor
[ 9] print *, "Length of C = ", Get Length of Q (C)
[10] print *, "Capacity of C = ", Get Capacity of Q (C)
[11] print *, "C empty? full? ", is Q Empty (C), is Q Full (C) !
[12]
[13] B = Create Q (3) ! private constructor
[14] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[15]
[16] call Add to Q (B, 4); print *, "B = [4]"
[17] print *, "Length of B = ", Get Length of Q (B)
[18] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[19]
[20] call Add to Q (B, 5); print *, " B = [4,5]"
[21] call Add to Q (B, 6); print *, " B = [4,5,6], full"
[22] print *, "Length of B = ", Get Length of Q (B)
[23] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[24] print *, "Capacity of B = ", Get Capacity of Q (B)
[25]
[26] value = Get Front of Q (B); print *, "Front Q value = ", value
[27]
[28] call Remove from Q (B); print *, "Removing from B"
[29] print *, "Length of B = ", Get Length of Q (B)
[30] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[31] value = Get Front of Q (B); print *, "Front Q value = ", value
[32]
[33] call Remove from Q (B); print *, "Removing from B"
[34] print *, "Length of B = ", Get Length of Q (B)
[35] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[36]
[37] call Remove from Q (B); print *, "Removing from B"
[38] print *, "Length of B = ", Get Length of Q (B)
[39] print *, "B empty? full? ", is Q Empty (B), is Q Full (B) !
[40]
[41] print *, "Removing from B"; call Remove from Q (B)
[42] call exception status
[43] end program main ! running gives:
[44] ! Length of C = 0 ! Capacity of C = 3 ! C empty? full? T, F
[45] ! B empty? full? T, F
[46] ! B = [4] ! Length of B = 1 ! B empty? full? F, F
[47] ! B = [4,5]
[48] ! B = [4,5,6], full ! Length of B = 3 ! B empty? full? F, T
[49] ! Capacity of B = 0 ! Front Q value = 4 ! Removing from B
[50] ! Length of B = 2 ! B empty? full? F, F ! Front Q value = 5
[51] ! Removing from B ! Length of B = 1 ! B empty? full? F, F
[52] ! Removing from B ! Length of B = 0 ! B empty? full? T, F
[53] ! Removing from B
[54] ! Exception Status Thrown
[55] ! Program :Remove from Q
[56] ! Message :empty Q
[57] ! Level : 5
[58] !
[59] ! Exception Summary:
[60] ! Exception count = 1
[61] ! Highest level = 5

Figure 7.6: Testing of the Queue Class

c 2001 J.E. Akin 141

7.4 Linked Lists
From our limited discussion of stacks and queues it should be easy to see that to try to insert or remove an
object at the middle of a stack or queue is not an efficient process. Linked lists are containers which make
it easy to perform the operations of insertion and deletion. A linked list of objects can be thought of as a
group of boxes, usually called nodes, each containing an object to stored and a pointer, or reference, to
the box containing the next object in the list. In most of our applications a list is referenced by a special
box, called the header or root node, which does not store an object but serves mainly to point to the first
linkable box, and thereby produces a condition where the list is never truly empty. This simplifies the
insertion scheme by removing an algorithmic special case. We will begin our introduction of these topics
with a singly linked list, also known as a simple list. It is capable of being traversed in only one direction,
from the beginning of the list to the end, or vice versa.
As we have seen, arrays of data objects work well so long as we know, or can compute, in advance
the amount of data to be stored. The data structures (linked lists and trees) to be considered here employ
pointers to store and change data objects when we do not know the required amount of storage in advance.
During program execution linked lists and trees allow separate memory allocations for each individual
data object. However, they do not permit direct access to an arbitrary object in the container. Instead
some searching must be performed and thus they incur an execution time penalty for such an access
operation. That penalty is smaller in tree structures than in linked lists (but is smallest of all in arrays).
Linked lists and trees must use pointer (reference) variables. Fortran pointers can simply be thought of
as an alias for other variables of the same type. We are beginning to see that pointers give a programmer
more power. However, that includes more power to “shoot yourself in the foot”; they make it hard to
find some errors; and can lead to new types of errors such as the so called memory leaks. Recall that
each pointer must be in one of three states: undefined, null, or associated. As dummy arguments within
routines pointer variables cannot be assigned the INTENT attribute. That means they have a greater
potential for undesired side effects. To avoid accidentally changing a pointer it is good programming
practice to clearly state in comments the INTENT of all dummy pointer arguments and to immediately
copy those with an INTENT IN attribute. Thereafter working with the copied pointer guarantees that an
error or later modification of the routine can not produce a side effect on the pointer. We also want to avoid
a dangling pointer which is caused by a deallocation that leaves its target object forever inaccessible. A
related problem is a memory leak or unreferenced storage such as the program segment:
real, pointer :: X ptr (:)
allocate ( X ptr(Big number) )
. . . ! use X ptr
nullify ( X ptr ) ! dangling pointer

because now there is no way to release memory for X ptr. To avoid this we need to free the memory
before the pointer is nullified, so the segment becomes:
real, pointer :: X ptr (:)
allocate ( X ptr(Big number) )
. . . ! use X ptr
deallocate ( X ptr ) ! memory released
nullify ( X ptr )

Remember that in F95 the memory is automatically deallocated at the end of the scope of the variable,
unless one retains the variable with a SAVE statement (and formally deallocates it elsewhere).
7.4.1 Singly Linked Lists
We begin the study of the singly linked list by showing the notations employed in Fig. 7.7. From experi-
enced we have chosen to have a dummy first node, called first, to simplify our algorithms so that a list
is never truly empty. Also as we scan through a list we will use one pointer, called current, to point to
the current object in the list and a companion, called previous, to point to the directly preceding object
(if any). If no objects have been placed in the list then both of these simply point to the first node.
The end of the list is denoted by the next pointer attribute taking on the null value. To insert or delete
objects one must be able to rank two objects. This means that in order to have a generic linked list one
must overload the relational operators, (< and ==) when the object to be placed in the container is de-
fined. Since most objects have different types of attributes the overloading process is clearly application

c 2001 J.E. Akin 142

 ------------
Type (Singly Linked Node) :: | Object | An Object instance attribute
| | and one pointer attribute.
| instance |
|----------|
| Next |
| ---+---- >
| pointer |
------------
a) Singly linked node

<
-----------
| First | --- Singly Linked List % First (An empty object instance)
----------- -----------
| Next --+---- > <
| Obj 1 | --- Previous pointer
> <
----------- ----------- -----------
| Next --+---- | Obj m | --- Current pointer
----------- ----------- -----------
| Next --+---- | Obj n > |
>
----------- -----------
| Next --+---- Null
----------- -----
---
-
b) List of singly linked nodes

<
-----------
| | --- Singly Linked List % First
| First <
| --- Previous pointer
| <
| --- Current pointer
>
-----------
| Next --+----- Null
----------- -----
---
-
c) An ‘‘empty’’ (one node) singly linked list

Figure 7.7: Singly linked list terminology

dependent. The process for inserting an object is sketched in Fig. 7.8 while that for deleting an object is
in Fig. 7.9.
The Singly Linked List class is given in Fig. 7.10. It starts with the definition of a singly linked
node (lines 4-8) that has an object attribute and a pointer attribute to locate the next node. Then a list
is begun (lines 10–13) by creating the dummy first node that is consider to represent an empty list. The
object deletion member must employ an overloaded operator (line 28), as must the insertion member
(line 52). Observe that a list never gets “full”, unless the system runs out of memory. The empty list test
member (line 62) depends on the pointer status, but is independent of the objects stored. The constructor
for a list (line 68) simply creates the first node and nullifies it. The printing member (line 74) is called
an iterator since it runs through all objects in the list. The testing program for this container type and
its output results are given in Fig. 7.11. In order to test such a container it is necessary to have an object
type defined. Here an object with a single integer value was selected, and thus it was easy to overload the
relational operators with a clear meaning as shown in Fig. 7.12.

c 2001 J.E. Akin 143

 INSERT

Figure 7.8: Inserting an Object in a Singly Linked List

DELETE

Figure 7.9: Deleting an Object from a Singly Linked List

c 2001 J.E. Akin 144

[ 1] module singly linked list
[ 2] use class Object
[ 3] implicit none
[ 4]
[ 5] type S L node ! Singly Linked Node
[ 6] private
[ 7] type (Object) :: value ! Object attribute
[ 8] type (S L node), pointer :: next ! Pointer to next node
[ 9] end type S L node
[10]
[11] type S L list ! Singly Linked List of Nodes
[12] private
[13] type (S L node), pointer :: first ! Dummy first object in list
[14] end type S L list
[15]
[16] contains
[17] subroutine S L delete (links, Obj, found)
[18] type (S L list), intent (inout) :: links
[19] type (Object), intent (in) :: Obj
[20] logical, intent (out) :: found
[21] type (S L node), pointer :: previous, current
[22]
[23] ! find location of Obj
[24] previous => links%first ! begin at top of list
[25] current => previous%next ! begin at top of list
[26] found = .false. ! initialize
[27] do
[28] if ( found .or. (.not. associated (current))) return ! list end
[29] if ( Obj == current%value ) then ! *** OVERLOADED ***
[30] found = .true. ; exit ! this location search
[31] else ! move the next node in list
[32] previous => previous%next
[33] current => current%next
[34] end if
[35] end do ! to find location of node with Obj
[36] ! delete if found
[37] if ( found ) then
[38] previous%next => current%next ! redirect pointer
[39] deallocate ( current ) ! free space for node
[40] end if
[41] end subroutine S L delete
[42]

Fig. 8.5, A Typical Singly Linked List Class of Objects (continued)

c 2001 J.E. Akin 145

[43] subroutine S L insert (links, Obj )
[44] type (S L list), intent (inout) :: links
[45] type (Object), intent(in) :: Obj
[46] type (S L node), pointer :: previous, current
[47]
[48] ! Find location to insert a new object
[49] previous => links%first ! initialize
[50] current => previous%next ! initialize
[51] do
[52] if ( .not. associated (current) ) exit ! insert at end
[53] if ( Obj < current%value ) exit ! *** OVERLOADED ***
[54] previous => current ! inserbefor current
[55] current => current%next ! move to next node
[56] end do ! to locate insert node
[57] ! Insert before current (duplicates allowed)
[58] allocate ( previous%next ) ! get new node space
[59] previous%next%value = Obj ! new object inserted
[60] previous%next%next => current ! new next pointer
[61] end subroutine S L insert
[62]
[63] function is S L empty (links) result (t or f)
[64] type (S L list), intent (in) :: links
[65] logical :: t or f
[66] t or f = .not. associated ( links%first%next )
[67] end function is S L empty
[68]
[69] function S L new () result (new list)
[70] type (S L list) :: new list
[71] allocate ( new list%first ) ! get memory for the object
[72] nullify ( new list%first%next ) ! begin with empty list
[73] end function S L new
[74]
[75] subroutine print S L list (links)
[76] type (S L list), intent (in) :: links
[77] type ( S L node), pointer :: current
[78] integer :: counter
[79] current => links%first%next
[80] counter = 0 ; print *,’Link Object Value’
[81] do
[82] if ( .not. associated (current) ) exit ! list end
[83] counter = counter + 1
[84] print *, counter, ’ ’, current%value
[85] current => current%next
[86] end do
[87] end subroutine print S L list
[88] end module singly linked list

Figure 7.10: A Typical Singly Linked List Class of Objects

c 2001 J.E. Akin 146

[ 1] program main ! test a singly linked object list
[ 2] use singly linked list
[ 3] implicit none
[ 4] type (S L list) :: container
[ 5] type (Object) :: Obj 1, Obj 2, Obj 3, Obj 4
[ 6] logical :: delete ok
[ 7]
[ 8] Obj 1 = Object(15) ; Obj 2 = Object(25) ! constructor
[ 9] Obj 3 = Object(35) ; Obj 4 = Object(45) ! constructor
[10] container = S L new()
[11] print *, ’Empty status is ’, is S L empty (container)
[12] call S L insert (container, Obj 4) ! insert object
[13] call S L insert (container, Obj 2) ! insert object
[14] call S L insert (container, Obj 1) ! insert object
[15] call print S L list (container)
[16]
[17] call S L delete (container, obj 2, delete ok)
[18] print *, ’Object: ’, Obj 2, ’ deleted status is ’, delete ok
[19] call print S L list (container)
[20] print *, ’Empty status is ’, is S L empty (container)
[21]
[22] call S L insert (container, Obj 3) ! insert object
[23] call print S L list (container)
[24] call S L delete (container, obj 1, delete ok)
[25] print *, ’Object: ’, Obj 1, ’ deleted status is ’, delete ok
[26] call S L delete (container, obj 4, delete ok)
[27] print *, ’Object: ’, Obj 4, ’ deleted status is ’, delete ok
[28] call print S L list (container)
[29] print *, ’Empty status is ’, is S L empty (container)
[30]
[31] call S L delete (container, obj 3, delete ok)
[32] print *, ’Object: ’, Obj 3, ’ deleted status is ’, delete ok
[33] print *, ’Empty status is ’, is S L empty (container)
[34] call print S L list (container)
[35] end program ! running yields
[36] ! Empty status is T
[37] ! Link Object Value
[38] ! 1 15
[39] ! 2 25
[40] ! 3 45
[41] ! Object: 25 deleted status is T
[42] ! Link Object Value
[43] ! 1 15
[44] ! 2 45
[45] ! Empty status is F
[46] ! Link Object Value
[47] ! 1 15
[48] ! 2 35
[49] ! 3 45
[50] ! Object: 15 deleted status is T
[51] ! Object: 45 deleted status is T
[52] ! Link Object Value
[53] ! 1 35
[54] ! Empty status is F
[55] ! Object: 35 deleted status is T
[56] ! Empty status is T
[57] ! Link Object Value

Figure 7.11: Testing the singly linked list with integers

[ 1] module class Object
[ 2] implicit none
[ 3] type Object ! An integer object for testing lists
[ 4] integer :: data ; end type Object
[ 5]
[ 6] interface operator (<) ! for sorting or insert
[ 7] module procedure less than Object ; end interface
[ 8] interface operator (==) ! for sorting or delete
[ 9] module procedure equal to Object ; end interface
[10]
[11] contains ! overload definitions only
[12] function less than Object (Obj 1, Obj 2) result (Boolean)
[13] type (Object), intent(in) :: Obj 1, Obj 2
[14] logical :: Boolean
[15] Boolean = Obj 1%data < Obj 2%data ! standard (<) here
[16] end function less than Object
[17] function equal to Object (Obj 1, Obj 2) result (Boolean)
[18] type (Object), intent(in) :: Obj 1, Obj 2
[19] logical :: Boolean
[20] Boolean = Obj 1%data == Obj 2%data ! standard (==) here
[21] end function equal to Object
[22] end module class Object

Figure 7.12: Typical object definition to test a singly linked list

c 2001 J.E. Akin 147

 ----------------------------------
|\ | | |
Doubly Linked List = > | \ | Header | | |
| \| | | |
-------------------------------|--
/|\ |
| \|/
--|-------------------------------
| | | | |
Trailing pointer --- > | | | Object 1 | | |
| | | | |
-------------------------------|--
/|\ |
| Previous pointer \|/
--|-------------------------------
| | | | |
Current pointer ---- > | | | Object | | |
| | | | |
-------------------------------|--
/|\ | Next pointer
| \|/
--|-------------------------------
| | | | |
| | | Object m | | |
| | | | |
-------------------------------|--
/|\ |
| \|/
--|-------------------------------
| | | |\ |
| | | Object n | \ | Null pointer
| | | \|
----------------------------------

Figure 7.13: Notations for a Doubly Linked List

7.4.1.1 Example: A List of Sparse Vectors

In this example we want to create a linked list to hold sparse vectors (singly subscripted arrays) where
the length of each vector is specified. We will do simple operations on all the vectors like input them,
normalize them, add them (if their sizes are the same), etc. In doing this we will make use of some of the
efficiencies that F90 provides for arrays, such as using the subscript array triplet to avoid serial loops,
and operating on arrays by name alone. This is an example where a similar C++ implementation would
be much longer in length because of the need to provide all the serial loops.

7.4.2 Doubly Linked Lists

The notations of the doubly linked list are shown in Fig. 7.13. Again we have chosen to have a dummy
first node, called header, to simplify our algorithms so that a list is never truly empty. Also as we scan
through a list we will use one pointer, called current, to point to the current object in the list and a
companion, called previous, to point to the directly preceding object (if any). If no objects have been
placed in the list then both of these simply point to the header node. The end of the list is denoted by
the next pointer attribute taking on the null value. To insert or delete objects one must be able to rank
two objects. This means that in order to have a generic linked list one must again overload the relational
operators, (< and ==) when the object to be placed in the container is defined.
An incomplete, but illustrative Doubly Linked List class is given in Fig. 7.14. It starts with the
definition of a doubly linked node (lines 4-8) that has an object attribute and a pair of pointer attributes to
locate the nodes on either side of the object. Then a list is begun (lines 10-13) by creating the dummy first
node that is consider to represent an empty list. The object insertion member must employ an overloaded
operator (line 53), as before. Observe that a list never gets “full”, unless the system runs out of memory.
The constructor for a list (line 17) simply creates the first node and nullifies its pointers. A corresponding
destructor (line 24) has been provided to delete every thing associated with the list when we are done

c 2001 J.E. Akin 148

[ 1] module doubly linked list
[ 2] use class Object
[ 3] implicit none
[ 4] type D L node
[ 5] private
[ 6] type (Object) :: Obj
[ 7] type (D L node), pointer :: previous
[ 8] type (D L node), pointer :: next
[ 9] end type D L node
[ 10]
[ 11] type D L list
[ 12] private
[ 13] type (D L node), pointer :: header
[ 14] end type D L list
[ 15]
[ 16] contains
[ 17]
[ 18] function D L new () result (new list) ! constructor
[ 19] type (D L list) :: new list
[ 20] allocate (new list % header)
[ 21] nullify (new list % header % previous)
[ 22] nullify (new list % header % next)
[ 23] end function D L new
[ 24]
[ 25] subroutine destroy D L List (links) ! destructor
[ 26] type (D L list), intent (in) :: links
[ 27] type (D L node), pointer :: current
[ 28] do
[ 29] current => links % header % next
[ 30] if ( .not. associated ( current ) ) exit
[ 31] current % previous % next => current % next
[ 32] if ( associated ( current % next ) ) then
[ 33] current % next % previous => current % previous
[ 34] end if
[ 35] nullify ( current % previous )
[ 36] nullify ( current % next )
[ 37] print *, ’Destroying object ’, current % Obj
[ 38] deallocate ( current )
[ 39] end do
[ 40] deallocate ( links % header )
[ 41] print *,’D L List destroyed’
[ 42] end subroutine destroy D L List
[ 43]
Fig. 7.14, A Typical Doubly Linked List Class of Objects (continued)

with it. The printing member (line 90) is called an iterator since it runs through all objects in the list.
The testing program for this container type and its output results are given in Fig. 7.15. Here an object
with a single integer value was selected, and thus it was easy to overload the relational operators with a
clear meaning as shown in Fig. 7.12.

7.5 Direct (Random) Access Files

Often it may not be necessary to create special object data structures such as those outlined above. From
its beginning Fortran has had the ability to create a sophisticated random access data structure where the
implementation details are hidden from its user. This was necessary originally since the language was
utilized on computers with memory sizes that are considered tiny by today’s standard (e.g., 16 Kb), but
it was still necessary to efficiently create and modify large amounts of data. The standard left the actual
implementation details to the compiler writers. That data structure is known as a “direct access file”. It
behaves like a single subscript array in that the object at any position can be read, modified, or written at
random so long as the user keeps up with the position of interest. The user simply supplies the position,
known as the record number, as additional information in the read and write statements. With today’s
hardware, if the file is stored on a virtual disk (stored in random access memory) there is practically no
difference in access times for arrays and direct files.
It should be noted here that since pointers are addresses in memory they can not be written to any
type of file. That, of course, means that no object having a pointer as an attribute can be written either.
Thus in some cases one must employ the other types of data structures illustrated earlier in the chapter.
To illustrate the basic concepts of a randon access file consider the program called ran-
dom access file which is given in Fig. 7.16. In this case the object is simply a character string, as

c 2001 J.E. Akin 149

[ 43] subroutine D L insert before (links, values)
[ 44] type (D L list), intent (in) :: links
[ 45] type (Object), intent (in) :: values
[ 46] type (D L node), pointer :: current ! Temp traversal pointer
[ 47] type (D L node), pointer :: trailing ! Preceding node pointer
[ 48] ! Find location to insert new node, in ascending order
[ 49] trailing => links % header ! initialize
[ 50] current => trailing % next ! initialize
[ 51] do
[ 52] if (.not. associated (current)) exit ! insert at end
[ 53] if (values < current % Obj ) exit ! insert before current
[ 54] trailing => current ! move to next node
[ 55] current => current % next ! move to next node
[ 56] end do
[ 57] ! Insert before current (duplicates allowed)
[ 58] allocate (trailing % next) ! get new node space
[ 59] trailing % next % Obj = values ! new object inserted
[ 60] ! Insert the new pointers
[ 61] if (.not. associated (current)) then ! End of list (special)
[ 62] nullify (trailing % next % next)
[ 63] trailing % next % previous => trailing
[ 64] else ! Not the end of the list
[ 65] trailing % next % next => current
[ 66] trailing % next % previous => trailing
[ 67] current % previous => trailing % next
[ 68] end if
[ 69] end subroutine D L insert before
[ 70]
[ 71] function Get Obj at Ptr (ptr to Obj) result ( values)
[ 72] type (D L node), intent (in) :: ptr to Obj
[ 73] type (Object) :: values ! intent out
[ 74] values = ptr to Obj % Obj
[ 75] end function Get Obj at Ptr
[ 76]
[ 77] function Get Ptr to Obj (links, values) result (ptr to Obj)
[ 78] type (D L list), intent (in) :: links ! D L list header
[ 79] type (Object), intent (in) :: values ! Node identifier Object
[ 80] type (D L node), pointer :: ptr to Obj ! Pointer to the Object
[ 81] type (D L node), pointer :: current ! list traversal pointer
[ 82] current => links % header % next
[ 83] do ! Search list, WARNING: runs forever if values not in list
[ 84] if (current % Obj == values) exit ! *** OVERLOADED ***
[ 85] current => current % next
[ 86] end do
[ 87] ptr to Obj => current ! Return pointer to node
[ 88] end function Get Ptr to Obj
[ 89]
[ 90] subroutine print D L list ( links )
[ 91] type (D L list), intent (in) :: links
[ 92] type (D L node), pointer :: current ! Node traversal pointer
[ 93] integer :: counter ! Link position
[ 94] ! Traverse the list and print its contents to standard output
[ 95] current => links % header % next
[ 96] counter = 0 ; print *,’Link Object Value’
[ 97] do
[ 98] if (.not. associated (current)) exit
[ 99] counter = counter + 1
[100] print *, counter, ’ ’, current % Obj
[101] current => current % next
[102] end do
[103] end subroutine print D L list
[104] end module doubly linked list

Figure 7.14: A Typical Doubly Linked List Class of Objects

defined in line 4. The hardware transportability of this code is assured by establishing the required con-
stant record with the intrinsic given in line 10. It is then used in opening the file, which is designated
as a direct file in line 12. Lines 16–24 create the object record numbers in a sequential fashion. They
also define the new object to be stored with each record. In lines 27–32 the records are accessed in a
backwards order, but could have been accessed in any random or partial order. In line 35 a random object
is given a new value. Finally, the changes are output in a sequential order in lines 37–42. Sample input
data and program outputs are included as comments at the end of the program.

c 2001 J.E. Akin 150

[ 1] program main
[ 2] use doubly linked list
[ 3] implicit none
[ 4] type (D L list) :: container
[ 5] type (Object) :: Obj 1, Obj 2, Obj 3, Obj 4
[ 6] type (Object) :: value at pointer
[ 7] type (D L node), pointer :: point to Obj 3
[ 8]
[ 9] Obj 1 = Object(15) ; Obj 2 = Object(25)
[10] Obj 3 = Object(35) ; Obj 4 = Object(45)
[11] container = D L new()
[12] ! print *, ’Empty status is ’, is D L empty (container)
[13] call D L insert before (container, Obj 4)
[14] call D L insert before (container, Obj 2)
[15] call D L insert before (container, Obj 1)
[16] call D L insert before (container, Obj 3)
[17] call print D L list (container)
[18]
[19] ! find and get Obj 3
[20] point to Obj 3 = Get Ptr to Obj (container, ObjL
[21] value at pointer = Get Obj at Ptr (point to
[22] print *, ’Object: ’, Obj 3, ’ has a value of ’, value at pointer
[23] call destroy D L List (container)
[24] end program main ! Running gives:
[25] ! Link Object Value
[26] ! 1 15
[27] ! 2 25
[28] ! 3 35
[29] ! 4 45
[30] ! Object: 35 has a value of 35
[31] ! Destroying object 15
[32] ! Destroying object 25
[33] ! Destroying object 35
[34] ! Destroying object 45
[35] ! D L List destroyed

Figure 7.15: Testing a Partial Doubly Linked List

c 2001 J.E. Akin 151

[ 1] program random access file
[ 2] ! create a file and access or modify it randomly
[ 3] implicit none
[ 4] character(len=10) :: name
[ 5] integer :: j, rec len, no name, no open
[ 6] integer :: names = 0, unit = 1
[ 7]
[ 8] ! find the hardware dependent record length of the object
[ 9] ! to be stored and modified. Then open a binary file.
[10] inquire (iolength = rec len) name
[11] open (unit, file = "random list", status = "replace",
[12] access = "direct", recl = rec len,
[13] form = "unformatted", iostat = no open)
[14] if ( no open > 0 ) stop ’open failed for random list’
[15]
[16] ! read and store the names sequentially
[17] print *, ’ ’; print *, ’Original order’
[18] do ! forever from standard input
[19] read (*, ’(a)’, iostat = no name) name
[20] if ( no name < 0 ) exit ! the read loop
[21] names = names + 1 ! record number
[22] write (unit, rec = names) name ! save record
[23] print *, name ! echo
[24] end do
[25] if ( names == 0 ) stop ’no records read’
[26]
[27] ! list names in reverse order
[28] print *, ’ ’; print *, ’Reverse order’
[29] do j = names, 1, -1
[30] read (unit, rec = j) name
[31] print *, name
[32] end do ! of random read
[33]
[34] ! change the middle name in random file
[35] write (unit, rec = (names + 1)/2) ’New Name’
[36]
[37] ! list names in original order
[38] print *, ’ ’; print *, ’Modified data’
[39] do j = 1, names
[40] read (unit, rec = j) name
[41] print *, name
[42] end do ! of random read
[43]
[44] close (unit) ! replace previous records and save
[45] end program random access file
[46] ! Running with input of: Name 1
[47] ! B name
[48] ! 3 name
[49] ! name 4
[50] ! Fifth
[51] ! Yields:
[52] ! Original order Reverse order Modified data
[53] ! Name 1 Fifth Name 1
[54] ! B name name 4 B name
[55] ! 3 name 3 name New Name
[56] ! name 4 B name name 4
[57] ! Fifth Name 1 Fifth

Figure 7.16: Utilizing a Random Access File as a Data Structure

c 2001 J.E. Akin 152

7.6 Exercises

c 2001 J.E. Akin 153

154
Chapter 8

Arrays and Matrices

8.1 Subscripted Variables: Arrays

It is common in engineering and mathematics to employ a notation where one or more subscripts are
appended to a variable which is a member of some larger set. Such a variable may be a member of a
list of scalars, or it may represent an element in a vector, matrix, or Cartesian tensor.y In engineering
computation, we usually refer to subscripted variables as arrays. Since programming languages do not
have a convenient way to append the subscripts, we actually denote them by placing them in parentheses
or square brackets. Thus, an element usually written as Ajk becomes A(j,k) in Fortran and M ATLAB,
and A[j][k] in C++.
Arrays have properties that need to be understood in order to utilize them correctly in any program-
ming language. The primary feature of an array is that it must have at least one subscript. The “rank” of
an array is the number of subscripts, or dimensions, it has. Fortran allows an array to have up to seven
subscripts, C++ allows four, and M ATLAB allows only two since it deals only with matrices. An array
with two subscripts is called a rank-two array, one with a single subscript is called a rank-one array, or
a vector. Matrices are rank-two arrays that obey special mathematical operations. A scalar variable has
no subscripts and is sometimes called a rank zero array. Rank-one arrays with an extent of one are also
viewed as a scalar.
The “extent” of a subscript or dimension is the number of elements allowed for that subscript. That
is, the extent is an integer that ranges from the lower bound of the subscript to its upper bound. The
lower bound of a subscript is zero in C++, and it defaults to unity in Fortran. However, Fortran allows
the programmer to assign any integer value to the lower and upper bounds of a subscript.
The “size” of an array is the number of elements in it. That is, the size is the product of the extents of
all of its subscripts. Most languages require the extend of each subscript be provided in order to allocate
memory storage for the array.
The “shape” of an array is defined by its rank and extents. The shape is a rank-one array where each
of its elements is the extent of the corresponding subscript of the array whose shape is being determined.
Both Fortran and M ATLAB have statements that return the shape and size of an array as well as statements
for defining a new array by re-shaping an existing array.
It is also important to know which of two “storage mode” options a language employs to store and
access array elements. This knowledge is especially useful when reading or writing full arrays. Arrays
are stored by either varying their leftmost subscript first or by varying the rightmost subscript first. These
are referred to as “column-wise” and “row-wise” access, respectively. Clearly, they are the same for
rank-one arrays and differ for arrays of higher rank. Column-wise storage is used by Fortran and C++,
while M ATLAB uses row-wise storage.
Matrices are arrays that usually have only two subscripts: the first represents the row number, and
the second the column number where the element is located. Matrix algebra places certain restrictions on
the subscripts of two matrices when they are added or multiplied, etc. The fundamentals of matrices are
covered in detail in this chapter.
y An n-th order tensor has n subscripts and transforms to different coordinate systems by a special law. The most common uses
are scalars (n = 0) and vectors (n = 1).

c 2001 J.E. Akin 155
 Action C++a F90 F77 M ATLAB
Pre-allocate
integer A[100] INTEGER A(100) INTEGER A(100) A(100)=0
Initialize for j=0,99 A=12 do 5 J=1,100 for j=1:100
A[j]=12 A(J)=12 A(j)=12
end 5 continue end

a Arrays in C++ have a starting index of zero.

Table 8.1: Typical Vector Initialization

Purpose F90 M ATLAB

Form subscripts () ()
Separates subscripts & elements , ,
Generates elements & subscripts : :
Separate commands ; ;
Forms arrays (/ /) []
Continue to new line & ...
Indicate comment ! %
Suppress printing default ;

Table 8.2: Special Array Characters

Both Fortran and C++ require you to specify the maximum range of each subscript of an array before
the array or its elements are used. MATLAB does not have this as a requirement, but pre-allocating the
array space can drastically improve the speed of M ATLAB, as well as making much more efficient use of
the available memory. If you do not pre-allocate M ATLAB arrays, then the interpreter must check at each
step to see if a position larger than the current maximum has been reached. If so, the maximum value is
increased and memory is found to store the new element. Thus, failure to pre-allocate M ATLAB arrays is
permissible but inefficient.
For example, assume we want to set a vector A having 100 elements, to an initial value of 12. The
procedures are compared in Table 8.1. This example could have also been done efficiently in F90 and
M ATLAB by using the colon operator: A(1:100) = 12. The programmer should be alert for the chance
to replace loops with the colon operator: it’s more concise while retaining readability and executes more
quickly. The joys of the colon operator are described more fully in x8.1.3 (page 159).
Array operations often use special characters and operators. Fortran has “implied” DO loops associ-
ated with its array operations (see x4.3.2, page 60). Similar features in M ATLAB and F90 are listed in
Table 8.2.
Fortran has always had efficient array handling features, but until the release of F90 it was not easy
to dynamically create and release the memory space needed to store arrays. That is a useful feature for
arrays that require large amounts of space but are not needed for the entire life of the program. F90 has
several types of arrays, with the most recent types being added to allow the use of array operations, and
intrinsic functions similar to those in M ATLAB. Without getting into the details of the F90 standards and
terminology we will introduce the most common array usages in a historical order:
F77: Constant Arrays, Dummy Dimension Arrays, Variable Rank Arrays
F90: Automatic Arrays, Allocatable Arrays.

These different approaches all have the common feature that memory space needed for an array must be
set aside (allocated) before any element in the array is utilized.
The new F90 array features include the so-called automatic arrays. An automatic array is one that
appears in a subroutine, or function and has its size, but not its name, provided in the argument list of the
subprogram. For example,
subroutine auto A B (M, N, Other arguments)
implicit none
integer :: M, N

c 2001 J.E. Akin 156
 real :: A(M, N), B(M) ! Automatic arrays
! Create arrays A & B and use them for some purpose
...
end subroutine auto A B

would automatically allocate space for the M rows and N columns of the array A and for the M rows of
array B. When the purpose of the subroutine is finished and it “returns” to the calling program the array
space is automatically released, and the arrays A and B cease to exist. This is a useful feature, especially
in Object Oriented programs. If the system does not have enough space available to allocate for the array
the program stops and gives an error message to that effect. With today’s large memory computers that
is unlikely to occur except for the common user error where the dimension argument is undefined.
An extension of this concept that allows more flexibility and control is the allocatable array. An
allocatable array is one that has a known rank (number of subscripts), but an initially unknown extent
(range over each subscript). It can appear in any program, function, or subroutine. For example,
program make A B ! Allocatable arrays
implicit none
real, allocatable :: A(:,:), B(:) ! Declares rank of each
integer :: M, N ! Row and column sizes
integer :: A B Status ! Optional status check
print *,"Enter the number of rows and columns: "
read *, M, N ! Now know the (default) extent of each subscript
allocate ( A(M, N), B(M), stat = A B Status ) ! dynamic storage
! Verify that the dynamic memory was available
if ( A B Status /= 0 ) stop "Memory not available in make A B"
! Create arrays A & B and use them for some purpose
...
deallocate (A, B) ! free the memory space
! Do other things
...
end program make A B

would specifically allocate space for the M rows and N columns of the array A and for the M rows of
array B, and optionally verify that the space was available. When the purpose of the arrays are finished
the space is specifically released, and the arrays A and B cease to exist. The optional status checking
feature is useful in the unlikely event that the array is so large that the system does not have that much
dynamic space available. Then the user has the option of closing down the program in some desirable
way, or simply stopping on the spot.
The old F77 standard often encouraged the use of dummy dimension arrays. The dummy dimension
array is one that appears in a subroutine, or function and has its size and its name provided in the argument
list of the subprogram. For example,
subroutine dummy A B (M, N, A, B, Other things)
implicit none
integer :: M, N
real :: A(M, N), B(M) ! dummy arrays
! Create arrays A & B and use them for some purpose
...
end subroutine dummy A B

would imply that existing space for the M rows and N columns of the array A and for the M rows of array
B (or more) was declared or allocated in the calling program. When the purpose of the subroutine is
finished and it “returns” to the calling program the space in the calling program for the arrays A and B
continues to exist until the declaring program unit terminates.
Of course the use of constant dimensioned arrays is always allowed. The constant dimension array is
one that appears in any program unit and has integer constants, or integer parameter variables (preferred)
as given extents for each subscript of an array. For example,
program main
implicit none
integer, parameter :: M max=20, N max=40 ! Maximum expected
integer :: Days per Month(12) ! Constant array
integer :: M, N ! User sizes
real :: A(M max, N max), B(M max) ! Constant arrays
print *,"Enter the number of rows and columns: "
read *, M, N ! The user extent of each subscript
! Verify that the constant memory is available
if ( M > M max ) stop "Row size exceeded in main"
if ( N > N max ) stop "Column size exceeded in main"
! Create arrays A & B and use them for some purpose
call dummy A B (M, N, A, B, Other things) ! dummy arrays

c 2001 J.E. Akin 157
 Action F90 M ATLAB
Define sizea
integer :: A (2, 3) A(2,3)=0;
Enter rows A(1,:)=(/1,7,-2/) A=[1,7,-2;
A(2,:)=(/3,4,6/) 3,4,6];

a Optional in M ATLAB, but improves efficiency.

Table 8.3: Example Array Definitions

F90 M ATLAB Result

1
2
43
data = (/(k, k=1,6)/) data = [1 : 6] M =4 2
6
5 75
M = reshape(data,(/3,2/)) M = reshape(data,3,2)
"
3 6 #
N = reshape(data,(/2,3/)) N = reshape(data,2,3) N = 12 3 5
4 6

Table 8.4: Array Reshape Intrinsics

...
end program main
subroutine dummy A B (M, N, A, B, Other things) ! dummy arrays
implicit none
integer :: M, N
real :: A(M, N), B(M)
! Create arrays A & B and use them for some purpose
...
end subroutine main

In general it is considered very bad style to use integer constants, like 12, in a dimension, or in a DO
loop control, except for the unusual case where its meaning is obvious, and where you never expect to
have to change the number. In the example declaration:
integer :: Days per Month(12) ! Constant array

It is obvious that we are thinking about 12 months per year and that we do not expect the number of
months per year to ever change in other potential applications of this program.
8.1.1 Initializing Array Elements
Explicit lists of the initial elements in an array are allowed by C++, Fortran, and M ATLAB. M ATLAB is
oriented to enter element values in the way that we read, that is, row by row. Fortran and C also allow
array input by rows, but the default procedure is to accept values by ranging over its subscripts from left
to right. That is, both F90 and C++ read by columns as their default mode. For example, consider the
2 3
 array " #

A = 13 74 2
6 :

This array could be typed as explicit input with the commands shown in Table 8.3. An alternative for F90
and M ATLAB is to define the full array by column order as a vector that is then reshaped into a matrix
with a specified number of rows and columns. The use of the RESHAPE operator is shown in Table 8.4.
Returning to the previous example, we see that the matrix A could have also been defined as

F90 A = reshape((/1,3,7,4, -2,6/), (/2,3/))

A = reshape((/1,3,7,4, -2,6/),shape(A))
M ATLAB A = reshape([1,3,7,4, -2,6], 2,3)

c 2001 J.E. Akin 158
 To initialize the elements of an array to zero or unity, F90 and M ATLAB have special constructs or
functions that fill the bill. For example, for A to be zero and B to have unity elements, we could use the
following commands.

Action F90 M ATLAB

Define size integer :: A (2, 3) A(2,3)=0;
integer :: B (3) B(3)=0;

Zero A A=0 A=zeros(2,3);

Initialize B B=1 B=ones(3);

If we want to create a new array B with the first three even numbers, we would use implied loops.

Action F90 M ATLAB

Even set B=(/(2*k,k=1,3)/) B=2*[1:1:3];
B=(/(k,k=2,6,2)/) B=[2:2:6];

Arrays can also be initialized by reading their element values from a stored data file. The two most
common types of files are ASCII (standard characters) and binary (machine language) files. ASCII files
are easy to read and edit, but binary files make more efficient use of storage, and are read or written much
faster than ASCII files. ASCII files are often denoted by the name extension of “dat”. Binary files are
denoted by the name extension “mat” in M ATLAB, while in Fortran the extension “bin” is commonly
employed.
For example, assume that the above A(2,3) array is to be initialized by reading its values from an
ASCII file created by a text editor and given the name of A.dat. Further, assume that we wish to
multiply all elements by 3 and store it as a new ASCII file. Then we could use read procedures like
those in Table 8.5 where the last M ATLAB command associated a file name and a file type with the desired
input/output (I/O) action. Fortran requires an OPEN statement to do this if the default I/O files (unit 5 to
read and unit 6 to write) are not used in the read or write.
8.1.2 Intrinsic Array Functions
Note that M ATLAB has intrinsic functions ones and zeros to carry out a task that F90 does with an op-
erator. Often the reverse is true. M ATLAB has several operators that in Fortran correspond to an intrinsic
function or a CALLed function. A comparison of the similar F90 and M ATLAB array mathematical oper-
ators are given in Table 8.5. They generally only differ slightly in syntax. For example, to transpose the
matrix A, the F90 construct is transpose(A) while in M ATLAB  1 3 5 it’s simply
 1A’ .y In F90, the * operator
means, for matrices, term by term multiplication: when A= 2 4 6 and B= 3 5 6 , A*B yields 16 20
2 4 6 20 .
36
In M ATLAB, the same operation is expressed as A.*B. To multiply the matrices A and B, Fortran requires
the use of the intrinsic function matmul (i.e., matmul(A,B)) while M ATLAB uses the * operator (A*B).
Another group of commonly used functions that operate on arrays in Fortran90 and M ATLAB are
briefly described in Table 8.6. Both languages have several other functions of a more specialized nature,
but those in Table 8.6 are probably the most commonly used in programs.
Often one needs to truncate a real number in some special fashion. Table 8.7 illustrates how to do
that using some of the functions common to the languages of interest. That table also implies how one
can convert reals to integers and vice versa.

8.1.3 Colon Operations on Arrays (Subscript Triplet)

The syntax of the colon operator, which is available in M ATLAB and F90, is detailed in Table 4.6.
What the colon operator concisely expresses is a sequence of numbers in an arithmetic progression. As
shown in the table, the M ATLAB expression B:I:E expresses the sequence B, B+I, B+2*I, ...,
B+b E I B cI. The complicated expression for the sequence’s last term simply means that the last value of
the sequence does not exceed (in magnitude) the end value E.
y In M ATLAB, A’ actually means conjugate transpose. If A is real, this operator performs the transpose as desired. If A is
complex and we want its transpose, the M ATLAB construct is A.’.

c 2001 J.E. Akin 159

c 2001
J.E. Akin

Description Equation Fortran90 Operator Matlab Operator Original Sizes Result Size
Scalar plus scalar c = a b c = a  b c = a  b; 1; 1 1; 1
Element plus scalar cjk = 
ajk b c = a  b c = a  b; m; n and 1; 1 m; n

Element plus element c jk = jk 

a jk b c = a  b c = a  b; m; n and m; n m; n

Scalar times scalar c =  a b c = a  b c = a  b; 1; 1 1; 1

Element times scalar c jk = jk 
a b c = a  b c = a  b; m; n and 1; 1 m; n

Element times element c jk = jk  jk

a b c = a  b c = a . b; m; n and m; n m; n

Scalar divide scalar c = a=b c = a=b c = a=b; 1; 1 1; 1

Scalar divide element cjk = ajk =b c = a=b c = a=b; m; n and 1; 1 m; n

Element divide element cjk = ajk =bjk c = a=b c = a =b;

. m; n and m; n m; n

Scalar power scalar c = a

b
c = a  b c = a ^ b; 1; 1 1; 1
Element power scalar b
cjk = ajk c = a  b c = a ^ b; m; n and 1; 1 m; n

Element power element cjk = a

bjk
jk c = a  b c = a.^ b; m; n and m; n m; n
160

Matrix transpose Ckj

P = Ajk C = transpose (A) C = A


0
; m; n n; m

Matrix times matrix

Vector dot vector
Cij

c
=
=
P k

k
Aik Bkj

Ak Bk
C

c
= matmul(A; B )
= sum(A  B) c
C

= sum(A.
= A B;

 B );
m; r and r; n
m; 1 and m; 1
m; n

1; 1
c = dot product(A; B ) c = A  B
0
; m; 1 and m; 1 1; 1

Table 8.5: Array Operations in Programming Constructs. Lower case letters denote scalars or scalar elements of arrays. Matlab arrays are allowed a maximum of
two subscripts while Fortran allows seven. Upper case letters denote matrices or scalar elements of matrices.
 You can also use the colon operator to extract smaller
 1 7 arrays from larger ones. If we wanted to extract
= 2 , to get, respectively,
the second row and third column of the array, A 3 4 6
( )

= [3 4 6] = 2
G ; C
6 ;

we could use the colon operator as follows.

Action F90 M ATLAB
Define size integer :: B (3) B(3)=0;
integer :: C (2) C(2)=0;

Extract row B=A(2,:) B=A(2,:);

Extract columns C=A(:,3) C=A(:,3);

Table 8.6: Equivalent Fortran90 and M ATLAB Intrinsic Functions.

The following KEY symbols are utilized to denote the TYPE of the in-
trinsic function, or subroutine, and its arguments: A-complex, integer,
or real; I-integer; L-logical; M-mask (logical); R-real; X-real; Y-real;
V-vector (rank 1 array); and Z-complex. Optional arguments are not
shown. Fortran90 and M ATLAB also have very similar array operations
and colon operators.

Type Fortran90 M ATLAB Brief Description

A ABS(A) abs(a) Absolute value of A.
R ACOS(X) acos(x) Arc cosine function of real X.
R AIMAG(Z) imag(z) Imaginary part of complex number.
R AINT(X) real(fix(x)) Truncate X to a real whole number.
L ALL(M) all(m) True if all mask elements, M, are true.
R ANINT(X) real(round(x)) Real whole number nearest to X.
L ANY(M) any(m) True if any mask element, M, is true.
R ASIN(X) asin(x) Arcsine function of real X.
R ATAN(X) atan(x) Arctangent function of real X.
R ATAN2(Y,X) atan2(y,x) Arctangent for complex number(X, Y).
I CEILING(X) ceil(x) Least integer >= real X.
Z CMPLX(X,Y) (x+yi) Convert real(s) to complex type.
Z CONJG(Z) conj(z) Conjugate of complex number Z.
R COS(R Z) cos(r z) Cosine of real or complex argument.
R COSH(X) cosh(x) Hyperbolic cosine function of real X.
I COUNT(M) sum(m==1) Number of true mask, M, elements.
R,L DOT PRODUCT(X,Y) x’?y Dot product of vectors X and Y.
R EPSILON(X) eps 1
Number, like X,  .
R,Z EXP(R Z) exp(r z) Exponential of real or complex number.
I FLOOR(X) floor Greatest integer  X.
R HUGE(X) realmax Largest number like X.
I INT(A) fix(a) Convert A to integer type.
R LOG(R Z) log(r z) Logarithm of real or complex number.
R LOG10(X) log10(x) Base 10 logarithm function of real X.
R MATMUL(X,Y) x?y Conformable matrix multiplication, X*Y.
I,V I=MAXLOC(X) [y,i]=max(x) Location(s) of maximum array element.
R Y=MAXVAL(X) y=max(x) Value of maximum array element.
I,V I=MINLOC(X) [y,i]=min(x) Location(s) of minimum array element.
R Y=MINVAL(X) y=min(x) Value of minimum array element.
(continued)

c 2001 J.E. Akin 161
 Type Fortran90 M ATLAB Brief Description
I NINT(X) round(x) Integer nearest to real X.
A PRODUCT(A) prod(a) Product of array elements.
call RANDOM NUMBER(X) x=rand Pseudo-random numbers in ; . (0 1)
call RANDOM SEED rand(’seed’) Initialize random number generator.
R REAL (A) real(a) Convert A to real type.
R RESHAPE(X, (/ I, I2 /)) reshape(x, i, i2) Reshape array X into II2 array.
I,V SHAPE(X) size(x) Array (or scalar) shape vector.
R SIGN(X,Y) Absolute value of X times sign of Y.
R SIGN(0.5,X)-SIGN(0.5,-X) sign(x) Signum, normalized sign, –1, 0, or 1.
R,Z SIN(R Z) sin(r z) Sine of real or complex number.
R SINH(X) sinh(x) Hyperbolic sine function of real X.
I SIZE(X) length(x) Total number of elements in array X.
R,Z SQRT(R Z) sqrt(r z) Square root, of real or complex number.
R SUM(X) sum(x) Sum of array elements.
R TAN(X) tan(x) Tangent function of real X.
R TANH(X) tanh(x) Hyperbolic tangent function of real X.
R TINY(X) realmin Smallest positive number like X.
R TRANSPOSE(X) x’ Matrix transpose of any type matrix.
R X=1 x=ones(length(x)) Set all elements to 1.
R X=0 x=zero(length(x)) Set all elements to 0.
For more detailed descriptions and example uses of these intrinsic functions see Adams, J.C., et al.,
Fortran 90 Handbook, McGraw-Hill, New York, 1992, ISBN 0–07–000406–4.

C++ – int – – floor ceil

F90 aint int anint nint floor ceiling
M ATLAB real (fix) fix real (round) round floor ceil
Argument Value of Result
–2.000 –2.0 –2 –2.0 –2 –2 –2
–1.999 –1.0 –1 –2.0 –2 –2 –1
–1.500 –1.0 –1 –2.0 –2 –2 –1
–1.499 –1.0 –1 –1.0 –1 –2 –1
–1.000 –1.0 –1 –1.0 –1 –1 –1
–0.999 0.0 0 –1.0 –1 –1 0
–0.500 0.0 0 –1.0 –1 –1 0
–0.499 0.0 0 0.0 0 –1 0
0.000 0.0 0 0.0 0 0 0
0.499 0.0 0 0.0 0 0 1
0.500 0.0 0 1.0 1 0 1
0.999 0.0 0 1.0 1 0 1
1.000 1.0 1 1.0 1 1 1
1.499 1.0 1 1.0 1 1 2
1.500 1.0 1 2.0 2 1 2
1.999 1.0 1 2.0 2 1 2
2.000 2.0 2 2.0 2 2 2

Table 8.7: Truncating Numbers

c 2001 J.E. Akin 162
 WHERE (logical array expression)
true array assignments
ELSEWHERE
false array assignments
END WHERE
WHERE (logical array expression)
true array assignment

Table 8.8: F90 WHERE Constructs

One can often use colon operators to avoid loops acting on arrays to define new arrays. For example,
consider a square matrix
2
1 2 33
A = 64 4 5 6 75 :

7 8 9
We can flip it left to right to create a new matrix (in F90 syntax)
2
3 2 13
n:1:-1) = 4 6 5 4 5
6 7
B=A(:,
9 8 7
or flip it up to down
2
7 8 93
C=A(n:1:-1, :) = 4
6
4 5 6 75
1 2 3
or flip it up to down, then left to right
2
9 8 73
D = A (n:1:-1, n:1:-1) = 4
6
6 5 4 75 ,
3 2 1
where n =3 is the number of rows in the matrix A. In the M ATLAB syntax, the second and third
numbers would be interchanged in the colon operator. Actually, MATLAB has intrinsic operators to flip
the matrices so that one could simply write
B = fliplr(A); C = flipud(A); D = rot90(A);

8.1.4 Array Logical Mask Operators

By default most M ATLAB commands are designed to operate on arrays. Fortran77 and C++ have no built
in array operations and it is necessary to program each loop. The Fortran90 standard has many of the
M ATLAB array commands and often with the identical syntax as shown in Table 8.5 and 8.6. Often the
F90 versions of these functions have optional features (arguments) that give the user more control than
M ATLAB does by including a logical control mask to be defined shortly.
To emphasize that an IF type of relational operator is to act on all elements of an array, Fortran90 also
includes an array WHERE block or statement control (that is, an IF statement acting on all array elements)
which is outlined in Table 8.8.
Note that the necessary loops are implied and need not be written. As an example, if
   
A = 07 34 58 ; B = 12 34 56

c 2001 J.E. Akin 163
 Function Description Opt Example
all Find if all values are true, for a fixed di- d all(B = A, DIM = 1)
mension. (true, false, false)
any Find if any value is true, for a fixed di- d any (B > 2, DIM = 1)
mension. (false, true, true)
count Count number of true elements for a d count(A = B, DIM = 2)
fixed dimension. (1, 2)
maxloc Locate first element with maximum m maxloc(A, A < 9)
value given by mask. (2, 3)
maxval Max element, for fixed dimension, given b maxval (B, DIM=1, B > 0)
by mask. (2, 4, 6)
merge Pick true array, A, or false array, B, ac- – merge(A, B,
  L)
cording to mask, L. 0 3 5
2 4 8
minloc Locate first element with minimum value m minloc(A, A > 3)
given by mask. (2, 2)
minval Min element, for fixed dimension, given b minval(B, DIM = 2)
by mask. (1, 2)
pack Pack array, A, into a vector under control v pack(A, B < 4)
of mask. (0, 7, 3)
product Product of all elements, for fixed dimen- b product(B) ; (720)
sion, controlled by mask. product(B, DIM = 1, T)
(2, 12, 30)
sum Sum all elements, for fixed dimension, b sum(B) ;(21)
controlled by mask. sum(B, DIM = 2, T)
(9, 12)
unpack Replace the true locations in array B con- – unpack(U,
 L, B)
trolled by mask L with elements from the 7 3 8
vector U. 2 4 9
     
A
0 3 5
= 7 4 8 ; B
1 3 5
= 2 4 6 ; L = T F T
; U = (7 8 9) ; ;
F F T

Table 8.9: F90 Array Operators with Logic Mask Control. T and F denote true and false, respectively.
Optional arguments: b -- DIM & MASK, d -- DIM, m -- MASK, v -- VECTOR and DIM = 1 implies
for any rows, DIM = 2 for any columns, and DIM = 3 for any plane.

 
then, WHERE (A > B) B = A gives a new B = 17 34 58 . By default, M ATLAB always acts
on matrices and considers scalars a special case. Thus, it would employ the standard syntax, if A >
B,B=A, to do the same task.
A more sophisticated way to selectively pick subscripts of an array is to use a mask array. A mask
array is the same size and shape as the array on which it will act. It is a Boolean array: All its elements
have either true or false values. When associated with an operator, the operator will only act on those
elements in the original array whose corresponding mask location is true (i.e., .true. in Fortran, true in
C++ and 1 in M ATLAB and C). Fortran90 has several operations that allow or require masks (Table 8.9).
M ATLAB functions with the same name exist in some cases, as seen in Table 8.6. Usually, they correspond
to the F90 operator where the mask is true everywhere.

c 2001 J.E. Akin 164
 ALL ANY COUNT
CSHIFT DOT PRODUCT EOSHIFT
MATMUL MAXLOC MAXVAL
MINLOC MINVAL PACK
PRODUCT REPEAT RESHAPE
SPREAD SUM TRANSFER
TRANSPOSE TRIM UNPACK

Table 8.10: Intrinsic Functions Allowing Logical Mask Control

A general Fortran principle underlies the fact that the array mentioned in the WHERE mask may be
changed within the WHERE construct. When an array appears in the WHERE statement mask, the logical
test is executed first and the host system retains the result independent of whatever happens later inside
the WHERE construct. Thus, in the program fragment
integer, parameter :: n = 5
real :: x (n) = (/ (k, k = 1, n) /)
where (x > 0.0)
x = -x
end where

the sign is reversed for all elements of x because they all pass the initial logical mask. It is as if a classic
DO sequence had been programmed
do i = 1, n, 1
if (x(i) > 0.0) x(i) = -x(i)
end do

instead of the WHERE construct.

A more ominous and subtle issue surrounds the use of other transformational intrinsic functions listed
in Table 8.10. The danger is that when these intrinsics appear inside the body of a WHERE construct, the
WHERE statement’s initial mask may no longer apply. Hence, in the following example the transforma-
tional intrinsic function SUM operates over all five elements of X rather than just the two elements of X
that exceed six.
integer, parameter :: n = 5
real :: x(n) = (/ 2, 4, 6, 8, 10 /)
where (x > 6.0)
x = x / sum(x)
end where

Thus, the new values for x are f 2, 4, 6, 8/30, 10/30 g rather than f2, 4, 6, 8/18, 10/18g. This standard-
conforming, but otherwise “unexpected”, result should raise a caution for the programmer. If one did not
want the above illustrated result, then it would be necessary to use the same mask of the WHERE as an
optional argument to SUM: sum(x, mask = x > 6.0). A lot of care needs to be taken to assure that
transformational intrinsics that appear in a WHERE construct use exactly the same mask.
8.1.5 User Defined Operators
In addition to the many intrinsic operators and functions we have seen so far, the F90 user can also define
new operators or extend existing ones. User defined operators can employ intrinsic data types and/or user
defined data types. The user defined operators, or extensions, can be unary or binary (i.e., have one or
two arguments). The operator symbol must be included between two periods, such as ‘.op.’. As an
example, consider a program to be used to create a shorthand notation to replace the standard F90 matrix
transpose and matrix multiplication functions so that we could write
B = .t. A
C = B .x. D
or C = (.t.A) .x. D

instead of B = TRANSPOSE(A)
C = MATMUL (B, D)
or C = MATMUL(TRANSPOSE (A), D)

c 2001 J.E. Akin 165
 Operator Action Use Algebra
T
.t. transpose .t.A A

.x. multiplication A.x.B AB

.i. inverse of matrix .i.A A

1
.ix. solution A.ix.B A
1B
T
.tx. transpose times matrix A.tx.B A B
T
.xt. matrix times transpose A.xt.B AB

.eye. identity matrix .eye.N I, N  N

Table 8.11: Definitions in Matrix Operators.

To do this, one must have a MODULE PROCEDURE to define the operator actions for all envisioned (and
incorrect) inputs and an INTERFACE OPERATOR that informs F90 what your operation symbol is.
Fig. 8.1 illustrates the code that would partially define the operator ‘.t.’. Note that while TRANSPOSE
accepts any type of matrix of any rank, our operator works only for real or integer rectangular arrays (of
rank 2). It would not transpose LOGICAL arrays or vectors. That oversight can be extended by adding
more functions to the interface.
If one works with matrices often, then one may want to define your own library of matrix operators.
Such operators are not standard in F90 as they are in M ATLAB, but can be easily added. To provide a
foundation for such a library, we provide a Matrix Operators module with the operators defined in
Table 8.11. The reader is encouraged to expand the initial support provided in that module.

8.1.6 Connectivity Lists and Vector Subscripts

When using an array with constant increments in its subscripts, we usually provide its subscript in the
form of a colon operator or a control variable in a DO or FOR loop. In either case, the array subscripts
are integers. There are several practical programming applications where the required subscripts are
not known in advance. Typically, this occurs when we are dealing with an assemblage of components
that can be connected together in an arbitrary fashion by the user (e.g., electric circuits, truss structures,
volume elements in a solid model). To get the subscripts necessary to build the assemblage we must read
an integer data file that lists the junction numbers to which each component is attached. We call those
data a connectivity file. If we assume each component has the same number of junction points, then the
list can be input as a two-dimensional array. One subscript will range over the number of components
and the other will range over the number of possible junctions per component. For ease of typing these
data, we usually assume that the th row of the array contains the integer junction, or connection, points
k
of that component. Such a row of connectivity data is often used in two related operations: gather
and scatter. A gather operation uses the lists of connections to gather or collect information from
the assembly necessary to describe the component or its action. The scatter operation has the reverse
effect. It takes information about the component and sends it back to the assembly. Usually, values from
the component are added into corresponding junction points of the assembly.
The main point of this discussion is that another way to define a non-sequential set of subscripts is to
use an integer vector array that contains the set. Then one can use the array name as a way to range over
the subscripts. This is a compact way to avoid an additional FOR or DO loop. The connectivity list for a
component is often employed to select the subscripts needed for that component.
To illustrate the concept of vector subscripts, we will repeat the array flip example shown in x8.1.3 via
the colon operators. Here we will define an integer vector called Reverse that has constant increments
to be used in operating on the original array A. By using the vector name as a subscript, it automatically
invokes an implied loop over the contents of that vector. As shown in Figure 8.2, this has the same effect
as employing the colon operator directly.
The real power of the vector subscripts comes in the case where it has integers in a random, or user
input, order rather than in an order that has a uniform increment. For example, if we repeat the above
example using a vector Random=[3 1 2], then both M ATLAB and F90 would give the result

c 2001 J.E. Akin 166
[ 1] MODULE Ops Example ! User defined matrix transpose example
[ 2]
[ 3] IMPLICIT NONE
[ 4] INTERFACE OPERATOR (.t.) ! transpose operator
[ 5] MODULE PROCEDURE Trans R, Trans I ! for real or integer matrix
[ 6] ! Remember to add logicals and vectors later
[ 7] END INTERFACE ! defining .t.
[ 8]
[ 9] CONTAINS ! the actual operator actions for argument types
[10]
[11] FUNCTION Trans R ( A ) ! defines .t. for real rank 2 matrix
[12] REAL, DIMENSION(:,:), INTENT(IN) :: A
[13] REAL, DIMENSION(SIZE(A,2), SIZE(A,1)) :: Trans R
[14] Trans R = TRANSPOSE (A)
[15] END FUNCTION Trans R ! for real rank 2 transpose via .t.
[16]
[17] FUNCTION Trans I ( A ) ! defines .t. for integer rank 2 matrix
[18] INTEGER, DIMENSION(:,:), INTENT(IN) :: A
[19] INTEGER, DIMENSION(SIZE(A,2), SIZE(A,1)) :: Trans I
[20] Trans I = TRANSPOSE (A)
[21] END FUNCTION Trans I ! for integer rank 2 transpose via .t.
[22]
[23] END MODULE Ops Example ! User defined matrix transpose example
[24]
[25] PROGRAM Demo Trans ! illustrate the .t. operator
[26] USE Ops Example ! module with user definitions
[27] IMPLICIT NONE
[28] INTEGER, PARAMETER :: M = 3, N = 2 ! rows, columns
[29] REAL, DIMENSION(M,N) :: A ; REAL, DIMENSION(N,M) :: B
[30]
[31] ! define A, test operator, print results
[32] A = RESHAPE ( (/ ((I*J , I=1,M), J=1,N) /), SHAPE(A) )
[33] B = .t. A
[34] PRINT *, ’MATRIX A’ ; CALL M print (A, M, N)
[35] PRINT *, ’MATRIX B’ ; CALL M print (B, N, M)
[36] ! Produces the result:
[37] ! MATRIX A
[38] ! RC 1 2
[39] ! 1 1.000 2.000
[40] ! 2 2.000 4.000
[41] ! 3 3.000 6.000
[42] !
[43] ! MATRIX B
[44] ! RC 1 2 3
[45] ! 1 1.000 2.000 3.000
[46] ! 2 2.000 4.000 6.000
[47] END PROGRAM Demo Trans

Figure 8.1: Creating and applying user defined operators

2
1 2 33
A = 64 4 5 6 75 Reverse = [321]
;

7 8 9
Flip left to right: 2
3 2 13
B=A(: , Reverse) = 4 6 5 4 5
6 7

9 8 7
Flip up to down: 2
7 8 93
C = A(Reverse, :) = 4 4 5 6 5
6 7

1 2 3
Flip up to down, left to right: 2
9 8 73
D = A (Reverse,Reverse) = 4 6 5 4 5
6 7

3 2 1
Figure 8.2: F90 and M ATLAB Vector Subscripts and Array Shifts.

c 2001 J.E. Akin 167
 five = (/ 1 2 3 4 5 /)
! without a pad
three = eoshift(five,2) ! = (/ 3 4 5 0 0 /)
three = eoshift(five,-2) ! = (/ 0 0 1 2 3 /)
! with a pad
pad = eoshift(five,2,9) ! = (/ 3 4 5 9 9 /)
pad = eoshift(five,-2,9) ! = (/ 9 9 1 2 3 /)

Figure 8.3: F90 end-off shift (eoshift) intrinsic.

five = (/ 1 2 3 4 5 /)
left 3 = cshift(five,3) ! = (/ 3 4 5 1 2 /)
right 3 = cshift(five,-3) ! = (/ 4 5 1 2 3 /)

Figure 8.4: F90 Circular shift (cshift) intrinsic.

2
3 1 23
E = A (:, Random) = 4
6
6 4 5 75 :

9 7 8
While the reshape option of F90 and M ATLAB allows the array elements to change from one rect-
angular storage mode to another, one can also move elements around in the fixed shape array by utilizing
the colon operators, or by the use of “shift operators.” The latter accept an integer to specify how many
locations to move or shift an element. A positive number moves an element up a column, a negative value
moves it down the column, and a zero leaves it unchanged. The elements that are moved out of the array
either move from the head of the queue to the tail of the queue (called a “circular shift”) or are replaced
by a user specified “pad” value (called an “end off shift”). If no pad is given, its value defaults to zero.
These concepts are illustrated for F90 in Figures 8.3 and 8.4.
8.1.7 Component Gather and Scatter
Often the equations governing a system balance principle are assembled from the relative contributions
of each component. When the answers for a complete system have been obtained, it is then possible to
recover the response of each component. The automation of these processes has six basic requirements:
1. a component balance principle written in matrix form,
2. a joint connectivity data list that defines where a given component type connects into the system,
3. a definition of a scatter operator that scatters the coefficients of the component matrices into
corresponding locations in the governing system equations,
4. an efficient system equation solver,
5. a gather operator to gather the answers from the system for those joints connected to a compo-
nent, and
6. a recovery of the internal results in the component.
The first of these is discipline-dependent. We are primarily interested in the gather-scatter operations.
These are opposites that both depend on the component connectivity list, which is often utilized as a
vector subscript. The number of rows in the component equations is less than the number of rows in the
assembled system, except for the special case where the system has only a single component. Thus, it
is the purpose of the gather-scatter operators to define the relation between a system row number and a
particular component row number. That is, they define the relation that defines the subset of component
:
unknowns, say Ve for component e, in terms of all the system unknowns, say V Ve e V. Here the
containment  is defined by the component’s connection list and the number of unknowns per joint. If
there is only one unknown per joint, then the subset involves only the connection list. The above process
gathers the subset of component unknowns from the full set of system unknowns.
Let the list of joints or nodes connected to the component be called Le . The k th member in this list
contains the corresponding system node number, K: i.e. K = L e(k). Thus, for a single unknown per

c 2001 J.E. Akin 168
 (2)

(1) (6) Component Nodes Property

1 2 (4) 3 5 (1) 1, 2 1
(2) 2, 3 2
(3) (3) 2, 4 1
4 (4) 4, 3 3
(5) (5) 4, 3 4
(6) 3, 5 1

Figure 8.5: Example Circuit or Axial Spring System

joint, one simply has Ve = V(L )  e

e V. Written in full loop form, the component gather operation
would be
DO k = 1, size(L e)
V e (k) = V(L e (k))
END DO ! OVER LOCAL JOINTS

while in F90 or M ATLAB vector subscript form, it is simply V e = V(L e), for a single unknown per
joint. When there is more than one unknown per joint, the relation can be written in two ways.
We pick the one that counts (assigns equation numbers to) all unknowns at a joint before going on to
the next joint. Let the number of unknowns per joint be N. Then by deduction, one finds that the equation
number for the j -th unknown at the K th system node is

(
E K; j ) = (
N K 1) + j; 1 
j N:

But to find which equation numbers go with a particular component, we must use the connection list
L e. For the k th local node, K = L e (k) and

( )= (
E k; j N L ( ) 1) +
e k j ; 1  j N :

If we loop over all nodes on a component, we can build an index list, say I e, that tells which equations
relate to the component.
INTEGER, ALLOCATABLE :: I e(:), V e(:)
ALLOCATE(I e(N * SIZE (L e)), V e (N*SIZE(L e)))
DO k = 1, SIZE(L e) ! component nodes
DO j = 1, N ! unknowns per node
LOCAL = N *(k-1) + j
SYSTEM = N *(L e (k) - 1) + j
I e (LOCAL) = SYSTEM
END DO ! on unknowns
END DO ! on local nodes.

Therefore, the generalization of the component gather process is

DO m = 1, SIZE(I e)
V e (m) = V(I e (m))
END DO ! over local unknowns

or in vector subscript form V e = V(I e) for an arbitrary number of unknowns per joint.
To illustrate the scatter concept, consider a system shown in Figure 8.5, which has six components
and five nodes. If there is only one unknown at each joint (like voltage or axial displacement), then
the system equations will have five rows. Since each component is connected to two nodes, each will
contribute to (scatter to) two of the system equation rows. Which two rows? That is determined by the
connection list shown in the figure. For example, component (4) is joined to nodes 4 and 3. Thus, the
coefficients in the first row of the local component balance low would scatter into (be added to) the fourth
row of the system, while the second row of the component would scatter to the third system equation row.
If the component balance law is symmetric, then the columns locations scatter in the same fashion.

c 2001 J.E. Akin 169
8.2 Matrices
Matrices are very commonly used in many areas of applied mathematics and engineering. While they
can be considered a special case of the subscripted arrays given above they have their on special algebra
and calculus notations that are useful to know. In the following sections we will describe matrices and
the intrinsic operations on them that are included in F90 and M ATLAB. Neither C nor C++ have such
useful intrinsics, but require the programmer to develop them or extract them from a special library.
A matrix is defined as a rectangular array of quantities arranged in rows and columns. The array is
enclosed in brackets, and thus if there are m rows and n columns, the matrix can be represented by
2 3
a11 a12 a13


1 a j


1 a n
6
6
a21 a22 a23


2 a j


2 a n 7
7
6 .. .. 7
6 7
A= = [A]
6 . . 7
6
6 ai 1 ai 2 ai 3


aij


ain 7
7
(8.1)
6 .. .. 7
4 . . 5
am 1 am 2 am 3


amj


amn

where the typical element aij has two subscripts, of which the first denotes the row (ith ) and where the
second denotes the column (j th ) which the element occupies in the matrix. A matrix with m rows and n
columns is defined as a matrix of order m  n, or simply an m  n matrix. The number of rows is always
specified first. In Equation 8.1, the symbol A stands for the matrix of m rows and n columns, and it is
usually printed in boldface type. If m n = =1
, then the matrix is equivalent to a scalar. If m , the =1
matrix A reduces to the single row

A = [ a11 a 12 a 13


a j1


1 ] = (A)a n

which is called a row matrix. Similarly, if n = 1, the matrix A reduces to the single column
2 3
a11
6 a21 7
A=6
6 .
7
7 = col[ a 11 a 21


am 1 ] = fAg
4 . . 5
am 1
which is called a column matrix, or vector. When all the elements of matrix are equal to zero, the matrix
is called null or zero and is indicated by 0. A null matrix serves the same function as zero does in
ordinary algebra. To set all the elements of A to zero, one writes A in F90, and A m; n =0 = zeros[ ]
in M ATLAB.
=
If m n, the matrix is said to be square.
2 3
a11 a 12


a n 1
A=6 .
4 ..
..
.
7
5
an 1 an 2


ann

Before considering some of the matrix algebra implied by the above equation, a few other matrix
types need definition. A diagonal matrix is a square matrix which has zero elements outside the principal
diagonal. It follows, therefore, that for a diagonal matrix aij =0
when i 6 j , and not all aii are zero. A =
typical diagonal matrix may be represented by
2
a 11 0


0 3

6 0 22


0 7
A=6
a
7
6 . .. 7 ;
4 . . . 5
0 0


ann

or more concisely as A = diag[ 11 a22


a

a nn ].

c 2001 J.E. Akin 170
 A unit or identity matrix is a diagonal matrix whose elements are equal to 0 except those located on
its main diagonal, which are equal to 1. That is, aij =1 =
if i j , and aij =0 =
if i 6 j . The unit matrix
3 3
will be given the symbol I throughout these notes. An example of a  unit matrix is
2
1 0 03
I = 4 0 1 0 5 = diag[1 1 1] :

0 0 1
A Toeplitz matrix has constant-valued diagonals. An identity matrix is Toeplitz as is the following
matrix. 2 3
1 2 3 5
A=
6
6 4 1 2 3 77
4 1 4 1 25
10 1 4 1
Note how the values of a Toeplitz matrix’s elements are determined by the first row and the first column.
M ATLABuses the Toeplitz function to create this unusual matrix.
=
A symmetric matrix is a square matrix whose elements aij aji for all i; j . For example,
2
12 2 1 3
A = 4 2 33 0 5
1 0 15
is symmetric: The first row equals the first column, the second row the second column, etc.
An antisymmetric or skew symmetric matrix is a square matrix whose elements aij aji for all=
i; j . Note that this condition means that the diagonal values of an antisymmetric matrix must equal zero.

An example of such a matrix is 2 3

0 2 1
A=4 2 0 10 5
1 10 0
The transpose of a matrix A, denoted by AT , is obtained by interchanging the rows and columns.
Thus, the transpose of an m  n matrix is an n  m matrix. For example,
2
2 13 
2 3 0
A=4 3 5 5 AT =
0 1 1 5 1 :

In M ATLAB an appended prime is used to denote the transpose of any matrix, such as B A0 , =
whereas in F90 we employ the intrinsic function B = transpose( )
A , or a user defined operator like
=
B .t.A which we defined earlier.
If all the elements on one side of the diagonal of a square matrix are zero, the matrix is called a
triangular matrix. There are two types of triangular matrices: (1) an upper triangular U, whose elements
below the diagonal are all zero, and (2) a lower triangular L, whose elements above the diagonal are all
zero. An example of a lower triangular matrix is
2
10 0 0 3
L=4 1 3 0 5 :

5 1 2
A matrix may be divided into smaller arrays by horizontal and vertical lines. Such a matrix is then
referred to as a partitioned matrix, and the smaller arrays are called submatrices. For example, we can
3 3
partition a  matrix into four submatrices as shown:
2 3 2 3
a11 a12 j a13   2 1 j 3
A=6
6 a21 a22 j a23 7
7 = A11 A12 6
= 64 10 5 j 0 7
7
4 j 5 A21 A22 j 5
a31 a 32 j a33 4 6 j 10

c 2001 J.E. Akin 171
where, in the F90 and M ATLABcolon notation;
   
A11 = 11
a21
a
= 102 15 = A(1 : 2 1 : 2)
a12
a22
;

   
A12 = = 30 = A(1 : 2 3)
a13

a23
;

   
A21 = a3132 = 4 6 = A(3 1 : 2)
a ;

A22 = [ 33] = [10] = A(3 3)

a ;

It should be noted that the elements of a partitioned matrix must be so ordered that they are compatible
with the whole matrix A and with each other. That is, A11 and A12 must have an equal number of rows.
Likewise, A21 and A22 must have an equal number of rows. Matrices A11 and A21 must have an equal
number of columns. Likewise, for A12 and A22 . Note that A22 is a matrix even though it consists of
only one element. Provided the general rules for matrix algebra are observed, the submatrices can be
treated as if they were ordinary matrix elements.
8.2.1 Matrix Algebra
To define what addition and multiplication means for matrices, we need to define an algebra for arrays
of numbers so that they become useful to us. Without an algebra, all we have is a sequence of definitions
without the ability to manipulate what they mean!
Addition of two matrices of the same order is accomplished by adding corresponding elements of
= +
each matrix. The matrix addition C A B (as we write it in F90 and M ATLAB), where A, B and C
are matrices of the same order m  n can be indicated by the equation

cij = aij + bij ; 1  i m; 1  j n

where cij , aij , and bij are typical elements of the C, A, and B matrices, respectively. An example of
matrix addition is
2
3 0 13 2 1 1 13 2 2 1 03
4 2 1 2 5+4 2 5 6 5=4 4 4 85 :

1 1 1 3 4 9 2 5 10
Matrix subtraction, C = A B, is performed in a similar manner.
Matrix addition and subtraction are associative and commutative. That is, with the previous defini-
tions for matrix addition and subtraction, grouping and ordering with respect to these operations does not
affect the result.
(
A BC )=( )
A  B  C and C  B  A
Multiplication of the matrix A by a scalar c is defined as the multiplication of every element of the matrix
by the scalar c. Thus, the elements of the product B =
cA are given by bij =
caij , and is written as

=
B C  A in both F90 and M ATLAB. Clearly, scalar multiplication distributes over matrix addition.
We could define special multiplication in the somewhat boring way as the term by term product of
two identical sized matrices: C =
AB ) cij = =
aij bij . This feature is allowed in both F90 and

= =
M ATLAB where it is written as C A*B, and C A.*B, respectively. Although this definition might
be useful in some applications, this choice for what multiplication means in our algebra does not give us
much power. Instead, we define the matrix product C AB to mean =
X
p

cij = aik bkj ; 1  1 

i m; j n :

k =1
A and B can be multiplied together as only when the number of columns in A; p, equals the number
of rows in B. When this condition is fulfilled, the matrices A and B are said to be conformable for
multiplication. Otherwise, matrix multiplication of two matrices cannot be defined. The product of two

c 2001 J.E. Akin 172
conformable matrices A and B having orders m  p and p  n, respectively, yields an m  n matrix C.
=
In M ATLAB this is simply written as C A*B, where as in F90 one would use the intrinsic function
C = matmul ( ) =
A; B , or a user defined operator such as C A.x.B which we defined earlier.
The reason why this definition for matrix multiplication was chosen so that we can concisely represent
a system of linear equations. The verbose form explicitly lists the equations.
a11 x1 +a12 x2 + a13 x3 +


+ a1 n xn = c1
a21 x1 +a22 x2 + a23 x3 +


+ a2 n xn = c2
a31 x1 + a32 x2 + a33 x3 +


+ a3 n xn = c3
.. ..
. .
an x 1 1 +a 2 x2 +a 3 x3 +


+a
n n nn xn = cn

where the aij ’s and ci ’s usually represent known coefficients and the xi ’s unknowns. To express these
equations more precisely, we define matrices for each of these arrays of numbers and lay them out as a
matrix-vector product equaling a vector.
2 32 3 2 3
a11 a12 a13


1
a n x1 c1
6
6 a21 a22 a23


a2 n
76
76 x2 7
7
6
6 c2 7
7
6
6 a31 a32 a33


a3 n
76
76 x3 7
7 = 6
6 c3 7
7
6 .. 76 .. 7 6 .. 7
4 . 54 . 5 4 . 5
an 1 an 2 an 3


ann xn cn

=
We thus obtain the more compact matrix form AX C. A represents the square matrix of coefficients,
X the vector (column matrix) of unknowns, and C the vector of known quantities.
Matrix multiplication is associative and distributive. For example,
(AB)C = A(BC)
A(B + C) = AB + AC
However, matrix multiplication is not commutative. In general, AB 6 BA. Consequently, the order =
in which matrix multiplication is specified is by no means arbitrary. Clearly, if the two matrices are
not conformable, attempting to commute the product makes no sense (the matrix multiplication BA is
not defined). In addition, when the matrices are conformable so that either product makes sense (the
matrices are both square and have the same dimensions, for example), the product cannot be guaranteed
to commute. You should try finding a simple example that illustrates this point. When two matrices A and
B are multiplied, the product AB is referred to either as B premultiplied by A, or as A postmultiplied
=
by B. When AB BA, the matrices A and B are then said to be commutable. For example, the unit
matrix I commutes with any square matrix of the same order: AI IA A.y = =
The process of matrix multiplication can also be extended to partitioned matrices, provided the indi-
vidual products of submatrices are conformable for multiplication. For example, the multiplication
    
A11 A12 B11 B12 A11 B11 + A12 B21 A11 B12 + A12 B22
AB = =
A21 A22 B21 B22 A21 B11 + A22 B21 A21 B12 + A22 B22
is possible provided the products A11 B11 , A12 B21 , etc. are conformable. For this condition to be ful-
filled, it is only necessary for the vertical partitions in A to include a number of columns equal to the
number of rows in the corresponding horizontal partitions in B.
T ZT YT


BT AT . As an example
( ) =
3 equals AB


YZ
The transpose of a product of2matrices
3
of matrix multiplication, let B =4 1 5 and A =  21 10 01 ; then
2
2
2 1 0 
 33 
7
AB =
1 0 1 4 12 5 = 6
y This result is why I is called the identity matrix: It is the identity element with respect to matrix multiplication.

c 2001 J.E. Akin 173
 2
2 13
BT AT =
  
3 1 2 4 1 05 = 7 6
0 1
8.2.2 Inversion
Every (non-singular) square matrix A has an inverse, indicated by A 1 , such that by definition the
product AA 1 is a unit matrix I. The reverse is also true: A 1 A I. Inverse matrices are very useful
=
=
in the solution of simultaneous equations AX C such as above where A and C are known and X is
unknown. If the inverse of A is known, the unknowns of the X matrix can be (symbolically) found by
premultiplying both sides of the equation by the inverse A 1 AX A 1 C so that =
X = A 1C :
In this way, in theory we have “solved” our system of linear equations. To employ this approach, we
must find the inverse of the matrix A, which is not any easy task. Despite this computational difficulty,
using matrix algebra to concisely express complicated linear combinations of quantities often provides
much insight into a problem and its solution techniques.
Various methods can be used to determine the inverse of a given matrix. For very large systems of
equations it is probably more practical to avoid the calculation of the inverse and solve the equations by
a procedure called factorization. Various procedures for computing an inverse matrix can be found in
2 2 3 3
texts on numerical analysis. The inverse of  or  matrices can easily be written in closed form
2 2
by using Cramer’s rule. For a  matrix, we have the classic formula, which no engineering student
should forget.
 
d b
  1
a

c
b

d
= ad
c

bc
a

However, finding the inverse of larger arrays using Cramer’s rule is very inefficient computationally. In
M ATLAB an inverse matrix of A is computed as inv(A), but this is only practical for matrices of a small
size, say <100. F90 does not have an intrinsic matrix inversion function but we provide such a function,
named inv, in our operator library.
8.2.3 Factorizations
=
We have indicated that we will frequently employ matrices to solve linear equation systems like A  x b,
where A is a known square matrix, B is a known vector, and X is an unknown vector. While in theory
the solution is simply the inverse of A times the vector B, x A( 1)  b, that is computationally the
=
least efficient way to find the vector X. In practice, one usually uses some form of factorization of the
matrix A. A very common method is to define A to be the product of two triangular matrices, defined
above, say L  U = A, where L is a square lower triangular matrix and U is a square upper triangular

matrix. Skipping the details of this “LU-factorization” we could rewrite the original matrix system as
LU x = b, which can be viewed as two matrix identities:

L  = h b

U  = x h;

where h is a new temporary vector, and where both L and U are much cheaper to compute than the
inverse of A. We do not need the inverse of L or U since, as triangular matrices, their first or last row
contains only one non-zero term. That allows us to find one term in the unknown vector from one scalar
equation. The processes of recovering the vectors from these two identities is called substitution.
We illustrate this process with a example set of four equations with A and b given as:
2
1800 600 360 900 3
6
A=6
0 4500 2700 2250 777
6
4 0 2700 2700 1890 5
6300 5250 1890 3795

c 2001 J.E. Akin 174
 bT = [ 6300 2250 1890 21405 ] :

The LU-factorization process mentioned above gives the first of two lower triangular systems; Lh =
: 2 38 9 8 9
60 0 0 0 6300
b
>
> h1 >
> >
> >
>
7> > > >
6
6
6
0 150 0 0 7<
7
>
h2
>
=
=>
>
< 2250 >
=
6
4 0 90 36 0 7>
5>>
>
h3 >
>
>
>
>
>
>
1890 >
>
>
>
:

210 105 42 10 :
h4
; :
21405 ;
Observe that the significant difference from A  x = is that the first row of this identity has one
b

equation and one unknown:

60  1 = 6300
h

which yields h1 = 105

. This process continues through all the rows solving for one unknown, hk in row
k , because all the above h values are known. For example, the next row gives   h2 0 105 + 150 =
2250, which yields h2 = 15
. This process is known as “forward substitution.” When completed the
substitution yields the intermediate answer:

h
T
= [ 105 15 15 30 ] :

Now that h is known we can write the upper triangular identity, U  = x h , as:
2 38 9 8 9
30 10 6 15 >
> x1 >
> >
> 105 >
>
7> > > >
6
6
6
0 30 18 15 7<
7
>
x2
>
=
=>
>
< 15 >
=
6
4 0 0 30 15 7>
5>>
>
x3 >
>
>
>
>
>
>
15 >
>
>
>
:

0 0 0 30 :
x4
; :
30 ;
This time the bottom row has only one unknown,  x4 30 = 30
, so the last unknown is x 4 = 1.
Working backward up to the next row again there is only one unknown:

30  3 + 15  ( 1) = 15
x

so that x3 =0
. Proceeding back up through the remaining rows to get all the unknowns is called “back
substitution.” It yields
x
T
= [4 0 0 1] :

By inspection you can verify that this satisfies the original system of linear equations, A  x b. With a =
little more work one can employ matrix multiplication to verify that L  U A. While we have not given =
the simple algorithm for computing L and U from A, it is widely known as the “LU Factorization,”
and is in many texts on numerical analysis. Other common factorizations are the “QR Factorization,” the
“Cholesky Factorization” for a symmetric positive definite A, and the “SVD Factorization” for the case
where A is rectangular, or ill-conditioned and one is seeking a best approximation to X.
The factorization process is relatively expensive to compute but is much less expensive that an inver-
sion. The forward and backward substitutions are very fast and cheap. In problems where you have many
different b vectors (and corresponding x vectors, such as time dependent problems), one carries out the
expensive factorization process only once and the executes the cheap forward and back substitution for
each b vector supplied.
8.2.4 Determinant of a Matrix
Every square matrix, say A, has a single scalar quantity associated with it. That scalar is called the
determinant, jAj, of the matrix. The determinant is important in solving equations and inverting matrices.
A very important result is that the inverse A 1 exists if and only if jAj 6= 0. If the determinant is zero,
the matrix A (and the equivalent set of equations) is said to be singular. Simple conditions on a matrix’s
structure can be used to infer the determinant or its properties.
 If two rows or columns are equal, the determinant is zero.

c 2001 J.E. Akin 175
  Interchanging two rows, or two columns, changes the sign of the determinant.
 The determinant is unchanged if any row, or column, is modified by adding to it a linear combina-
tion of any of the other rows, or columns.
 A singular square matrix may have nonsingular square partitions.
The last two items will become significant when we consider how to apply boundary conditions and how
to solve a system of equations.
8.2.5 Matrix Calculus
At times you might find it necessary to differentiate or integrate matrices. These operations are simply
carried out on each and every element of the matrix. Let the elements aij of A be a function of a
parameter t. Then, the derivative and integral of a matrix simply equals term-by-term differentiation and
integration, respectively.
A
B = d

dt
! bij = 1  1 
daij

dt
; i m; j n
Z Z
C = A dt ! cij = 1  1 
aij dt; i m; j n

When dealing with functional relations the concept of rate of change is often very important. If we
()
have a function f
of a single independent variable, say x, then we call the rate of change the derivative
with respect to x, which is written as df =dx. Generalizing this notion to functions of more than two
variables, say z = ( )
f x; y , we may define two distinct rates of change. One is the function’s rate of

change with respect to one variable with the other held constant. We thus define partial derivatives.
When x is allowed to vary, the derivative is called the partial derivative with respect to x, and is denoted
by @ f =@ x. By analogy with the usual definition of derivative, this partial derivative is mathemtically

=
lim!0 ( +

) ( )
defined as
fx = @f

@x x
f x x; y

x
f x; y
:

A similar definition describes the partial derivative with respect to y , denoted by @ f =@ y . The second
notion of rate-of-change is the total derivative, which is expressed as df .

df = @f

@x
dx + @f

@y
dy

These definitions can be extended to include a function of any number of independent variables.
Often one encounters a scalar u defined by a symmetric square n  n matrix, A, a column vector B,
and a column vector X of n parameters. The combination we have in mind has the form

u = 12 XT AX + XTB + C (8.2)

If we calculate the derivative of the scalar u with respect to each xi , the result is the column vector
@u

@ X
= AX + B ;

a result that can be verified by expanding Equation 8.2, differentiating with respect to every xi in X, and
rewriting the result as a matrix product.
8.2.6 Computation with Matrices
Clearly, matrices are useful in representing systems of linear equations and expressing the solution. As
said earlier, we need to be able to express linear equations in terms of matrix notation so that analytic
manipulations become easy. Furthermore, calculations with linear equations become easy if we can di-
rectly express our matrix formulas in terms of programs. This section describes programming constructs
for the simple matrix expressions and manipulations covered in this chapter.

c 2001 J.E. Akin 176
 M ATLAB C++ F90
a
Pre-allocate A(100)=0 int A[100]; integer A(100)
linear array
Initialize to a for j=1:100 % slow for (j=0; j<100; j++) A=12
A(j)=12
constant value of end A[j]=12;
12 % better way
A=12*ones(1,100)
Pre-allocate A=ones(10,10) int A[10][10]; integer A(10,10)
two-dimensional
array
a C++ has a starting subscript of 0, but the argument in the allocation statement is the array’s size.

Table 8.12: Array initialization constructs.

Action M ATLAB C++ F90

a
Define A=zeros(2,3) int A[2][3]; integer, dimension (2,3)::A
size
Enter A=[1,7,-2; int A[2][3]=f A(1,:)=(/1,7,-2/)
f1,7,2g,
rows 3, 4, 6]; A(2,:)=(/3,4,6/)
f3,4,6g

g ;

a Optional in M ATLAB, but improves efficiency.

Table 8.13: Array initialization constructs

In most languages, we must express the fact that a variable is an ordered array of numbers—a
matrix—rather than a scalar (or some other kind of variable). Such declaration statements usually occur
at the beginning of the program or function. Table 8.12 shows the declaration of an integer array for
our suite of programming languages. Both Fortran and C++ require you to specify the maximum range
of each subscript of an array before the array or its elements are used. Such range specification is not
required by M ATLAB, but pre-allocating the array space can drastically improve the speed of M ATLAB,
as well as making much more efficient use of the available memory. If you do not pre-allocate M ATLAB
arrays, the interpreter must check at each step if a position in a row or column is larger than the current
maximum. If so, the maximum value is increased and the memory found to store the new element. Thus,
failure to pre-allocate M ATLAB arrays is permissible but inefficient.
Array initialization is concisely expressed in both Fortran and M ATLAB; in C++, you must write
a small program to initialize an array to a nonzero value.y If an array contains a variety of different
numbers, we can concisely express the initialization; again, in C++, we must explicitly write statements
for each array element.
An Aside: Matrix Storage
Most computer languages do not make evident how matrices are stored. More frequently than you might
think, it becomes necessary to know how an array is actually stored in the computer’s memory and
retrieved. The procedure both Fortran and M ATLAB use to store the elements of an array is known as
column major order: all the elements of the first column are stored sequentially, then all of the second,
etc. Another way of saying this is that the first (left most) subscript ranges over all its values before the
second is incremented. After the second subscript has been incremented then the first again ranges over
all its values. In C++, row major order is used: The first row of an array is stored sequentially, then the
second, etc. Clearly, translating programs from Fortran to C++ or vice versa must be done with care.
However, the above knowledge can be used to execute some operations more efficiently. For example,
the matrix addition procedure could be written as ck ak bk , = + 1
 k  m  n. One circumstance
y Global arrays those declared outside of any function definition are initialized to zero in many versions of C++. Array
declared within the scope of a function have no predefined values.

c 2001 J.E. Akin 177
 M ATLAB C++ F90
Addition
C=A+B
C=A+B for (i=0; i<n; i++)f C=A+B
for (j=0; j<n; j++)f
C[i][j]=A[i][j]+B[i][j];
g

Multiplication
C = AB
C=A*B for (i=0; i<n; i++)f C=matmul(A,B)
for (j=0; j<n; j++)f
C[i][j] = 0;
for (k=0; k<n; k++)f
C[i][j] += A[i][k]*B[k][j];
g
g

Scalar
multiplication C=a*B for (i=0; i<n; i++)f C=a*B
C = aB for (j=0; j<n; j++)f
C[i][j] = a*B[i][j];
g

Matrix
inverse
B=A 1
B=inv(A) a B=inv(A)a

a Neither C++ nor F90 have matrix inverse functions as part of their language definitions nor as part of standard collections

of mathematical functions (like those listed in Table 4.7). Instead, a special function, usually drawn from a library of numerical
functions, or a user defined operation, must be used.

Table 8.14: Elementary matrix computational routines (for n  n matrices)

where knowing the storage format becomes crucial is extracting submatrices in partitioned arrays. Such
a Fortran subroutine would have to dimension the arrays with a single subscript.

Expressing the addition, subtraction, or multiplication of arrays in Fortran or M ATLAB is concise and
natural. Explicit programs must be written in C++ to accomplish these calculations. Table 8.14 displays
what these constructs are for the special case of square matrices with n rows.

8.3 Exercises
1. Often it is necessary to check computer programs that invert matrices. One approach is use
test matrices for which the inverse is known analytically. Few such matrices are known, but one is the
following n  n matrix.
2 +2 1 0 0


0 1 3 1 2
1 2


2 1 3
2 +2 2 2 +2
n
n n n
1 1
n n
6
6 2 1 2 0


0 0 7
7
6
6 n 1 n n 1


3 2 7
7
0 1 1 1
2


0 0
6 7 6 7
6
2 7 6 2 1


4 3 7
=
n n n
6 7 6 7
6 .. .. .. .. .. .. .. 7 6 .. .. 7
6 7 6 7
6 . . . . . . . 7 6 . . 7
6
4 0 0





21 1 1
2
7
5
6
4 2 3 4


n n 1 7
5
1 0





0 12 +2 1 2 3


1
2 +2 2 +2
n
n n
n n

Develop two routines that will create each of these two matrices for a given n value, and test them with
a main program that uses matmul to compute their matrix product. The result should be the identity
matrix.

c 2001 J.E. Akin 178
 2. The numerical accuracy in calculating an inverse is always an issue: To what extent can you
believe the accuracy of the numbers that computer programs calculate. Because of the finite precision
used to represent floating point numbers, floating point calculations can only rarely yield exact answers.
We want to empirically compute the difference between the inverse of the first matrix in the previous
exercise by using a library inversion routine and compare its result with the exact answer. Because the
error varies throughout the matrix, we need to summarize the error with a single quantity. Two measures
are
q
routinely used: the peak absolute error max
i;j jaij bij j and the root-mean-squared (rms) error

n
12 P
i;j
( aij bij )2.y The first captures the biggest difference between the elements of two matrices,
and the second summarizes the error throughout the entire difference. Clearly, the peak absolute error
is always larger than the rms error. Comparing these two error measures provides some insight into the
distribution of error: If the two are comparable, the errors have about the same size; if not, the errors
deviate greatly throughout the matrix.
3. Combine the intrinsic array features of F90 with the concepts of OO classes to create a Vector
Class that is built around a type that has attributes consisting of the integer length of a vector and an array
of its real components. Provide members to construct vectors, delete the arrays, real vectors, list vectors,
and carry out basic mathematics operations. Overload the operators +, -, *, =, and ==. Avoid writing any
serial loops.
4. Extend the above Vector Class concepts to a Sparse Vector Class where it is assumed that most
of the values in the vector are zero and for efficiency only the non-zero entries are to be stored. This
clearly exceeds the intrinsic array features of F90 and begins to show the usefulness of OOP. The defined
type must be extended to include an integer array that contains the location (row number) of the non-zero
values. In addition to changing the input and output routines to utilize the extra integer position list, all
the mathematical member functions such as addition will have to be changed so that the resulting vector
has non-zero terms in locations that are a union of the two given location sets (unless the operation creates
new zero values). Use the concept of logical array masks in computing the dot product. Avoid writing
any serial loops.

y The 1=n2 term occurs in this expression because that equals the number of terms in the sum. The rms error is used frequently
in the practice to measure error; you average the squared error across the dataset and evaluate the square-root of the result.

c 2001 J.E. Akin 179
180
Chapter 9

Advanced Topics

9.1 Templates
One of our goals has been to develop software that can be reused for other applications. There are some
algorithms that are effectively independent of the object type on which they operate. For example, in a
sorting algorithm one often needs to interchange, or swap, two objects. A short routine for that purpose
follows:
subroutine swap integers (x, y)
implicit none
integer, intent(inout) :: x, y
integer :: temp
temp = x
x = y
y = temp
end subroutine swap integers

Observe that in this form it appears necessary to have one version for integer arguments and another for
real arguments. Indeed we might need a different version of the routine for each type of argument that
you may need to swap. A slightly different approach would be to write our swap algorithm as:
subroutine swap objects (x, y)
implicit none
type (Object), intent(inout) :: x, y
type (Object) :: temp
temp = x
x = y
y = temp
end subroutine swap objects

which would be a single routine that would work for any Object, but it has the disadvantage that one
find a way to redefine the Object type for each application of the routine. That would not be an easy
task. (While we will continue with this example with the algorithm in the above forms it should be noted
that the above approaches would not be efficient if x and y were very large arrays or derived type objects.
In that case we would modify the algorithm slightly to employ pointers to the large data items and simply
swap the pointers for a significant increase in efficiency.)
Consider ways that we might be able to generalize the above routines so that they could accept and
swap any specific type of arguments. For example, the first two versions could be re-written in a so called
template form as:
subroutine swap Template$ (x, y)
implicit none
Template$, intent(inout) :: x, y
Template$ :: temp
temp = x
x = y
y = temp
end subroutine swap Template$

In the above template the dollar sign ($) was includes in the “wild card” because while it is a valid member
of the F90 character set it is not a valid character for inclusion in the name of a variable, derived type,
function, module, or subroutine. In other words, a template in the illustrated form would not compile, but

c 2001 J.E. Akin 181

such a name could serve as a reminder that its purpose is to produce a code that can be compiled after the
“wild card” substitutions have been made.
With this type of template it would be very easy to use a modern text editor to do a global substitution
of any one of the intrinsic types character, complex, double precision, integer, logi-
cal, or real for the “wild card” keyword Template$ to produce a source code to swap any or all of
the intrinsic data types. There would be no need to keep up with all the different routine names if we
placed all of them in a single module and also created a generic interface to them such as:
module swap library
implicit none
interface swap ! the generic name
module procedure swap character, swap complex
module procedure swap double precision, swap integer
module procedure swap logical, swap real
end interface
contains
subroutine swap characters (x, y)
. . .
end subroutine swap characters
subroutine swap . . .
. . .
end module swap library

The use of a text editor to make such substitutions is not very elegant and we expect that there may
be a better way to pursue the concept of developing a re-useable software template. The concept of a text
editor substitution also fails when we go to the next logical step and try to use a derived type argument
instead of any of the intrinsic data types. For example, if we were to replace the “wild card” with our
previous type (chemical element) that would create:
subroutine swap type (chemical element) (x,y)
implicit none
type (chemical element), intent (inout)::x,y
type (chemical element) ::temp
temp = x
x = y
y = temp
end subroutine swap type (chemical element)

This would fail to compile because it violates the syntax for a valid function or subroutine name, as well
as the end function or end subroutine syntax. Except for the first and last line syntax errors this would be
a valid code. To correct the problem we simply need to add a little logic and omit the characters type
( ) when we create a function, module, or subroutine name that is based on a derived type data entity.
Then we obtain
subroutine swap chemical element (x,y)
implicit none
type (chemical element), intent (inout)::x,y
type (chemical element) ::temp
temp = x
x = y
y = temp
end subroutine swap chemical element

which yields a completely valid routine.

Unfortunately, text editors do not offer us such logic capabilities. However, as we have seen, high
level programming languages like C++ and F90 do have those abilities. At this point you should be able
to envision writing a pre-processor program that would accept a file of template routines, replace the
template “wildcard” words with the desired generic forms to produce a module or header file con-
taining the expanded source files that can then be brought into the desired program with an include or
use statement. The C++ language includes a template pre-processor to expand template files as needed.
Some programmers criticize F90/95 for not offering this ability as part of the standard. A few C++ pro-
grammers criticize templates and advise against their use. Regardless of the merits of including template
pre-processors in a language standard it should be clear that it is desirable to plan software for its efficient
reuse.
With F90 if one wants to take advantage of the concepts of templates then the only choices are to carry
out a little text editing or develop a pre-processor with the outlined capabilities. The former is clearly
the simplest and for many projects may take less time than developing such a template pre-processor.
However, if one makes the time investment to produce a template pre-processor one would have a tool

c 2001 J.E. Akin 182

that could be applied to basically any coding project. In the following sections we will give one example
of an F90 template pre-processor and demonstrate its application. Reviewing this approach you will
probably notice alternate ways to solve the same problem.

9.2 Subtyping Objects (Dynamic Dispatching)

One polymorphic feature missing from the Fortran 90 standard that is common to most object oriented
languages is called run-time polymorphism or dynamic dispatching. (This feature is expected in Fortran
200X as an ”extensible” function.) In the C++ language this ability is introduced in the so-called ” virtual
function”. To emulate this ability is quite straightforward in F90 but is not elegant since it usually requires
a group of if-elseif statements or other selection processes. It is only tedious if the inheritance hierar-
chy contains many unmodified subroutines and functions. The importance of the lack of a standardized
dynamic dispatching depends on the problem domain to which it must be applied. For several applica-
tions demonstrated in the literature the alternate use of subtyping has worked quite well and resulted in
programs that have been shown to run several times faster than equivalent C++ versions.
We implement dynamic dispatching in F90 by a process often called subtyping. Two features must be
constructed to do this. First, a pointer object, which can point to any subtype member in an inheritance
hierarchy, must be developed. Remember that F90 uses the operator ’=>’ to assign pointers to objects,
and any object to be pointed at must have the TARGET attribute. Second, we must construct a (dynamic)
dispatching mechanism to select the single appropriate procedure to execute at any time during the dy-
namic execution of the program. This step is done by checking which of the pointers actually points to
an object and then passing that (unique) pointer to the corresponding appropriate procedure. In F90 the
necessary checking can be carried out by using the ASSOCIATED intrinsic. Here, an if-elseif or other
selection method is developed to serve as a dispatch mechanism to select the unique appropriate proce-
dure to be executed based on the actual class referenced in the controlling pointer object. This subtyping
process is also referred to as implementing a polymorphic class. Of course, the details of the actual dis-
patching process can be hidden from the procedures that utilize the polymorphic class. The polymorphic
class knows only about the interfaces and data types defined in the hierarchy and nothing about how those
procedures are implemented.
This process will be illustrated buy creating a specific polymorphic class, in this case called
Is A Member Class, which has polymorphic procedures named new, assign, and display. They
will construct a new instance of the object, assign it a value, and list its components. The minimum ex-
ample of such a process requires two members and is easily extended to any number of member classes.
We begin by illustrating a short dynamic dispatching program and then defining each of the subtype
classes of interest. The validation of this dynamic dispatching through a polymorphic class is shown in
Fig. 9.1. There a target is declared for reach possible subtype and then each of them is constructed and
sent on to the other polymorphic functions. The results clearly show that different display procedures
were used depending on the class of object supplied as an argument. It is expected that the new Fortran
200X standard will allow such dynamic dispatching in a much simpler fashion.
The first subtype is a class, Member 1 Class, which has two real components and the encapsulated
functionality to construct a new instance and another to accept a pointer to such a subtype and display
related information. It is shown in Fig. 9.2. The next subtype class, Member 2 Class, has three
components: two reals and one of type Member 1. It has the same sort of functionality, but clearly must
act on more components. It has also inherited the functionally from the Member 1 Class; as displayed
in Fig. 9.3.
The polymorphic class, Is A Member Class, is shown in Fig. 9.4. It includes all of the encap-
sulated data and function’s of the above two subtypes by including their use statements. The necessary
pointer object is defined as an Is A Member type that has a unique pointer for each subtype member
(two in this case). That is, at any given time during execution it will associate only one of the pointers in
this list with an actual pointer object, and the other pointers are nullified. That is why this dispatching is
referred to as ”dynamic”. It also defines a polymorphic interface to each of the common procedures to be
applied to the various subtype objects. In the polymorphic function assign the dispatching is done very
simply. First, all pointers to the family of subtypes are nullified, and then the unique pointer component

c 2001 J.E. Akin 183

[ 1] program main
[ 2] use Is A Member Class
[ 3] implicit none
[ 4]
[ 5] type (Is A Member) :: generic member
[ 6] type (member 1), target :: pt to memb 1
[ 7] type (member 2), target :: pt to memb 2
[ 8] character(len=1) :: c
[ 9]
[10] c = ’A’
[11] call new (pt to memb 1, 1.0, 2.0)
[12] call assign (generic member, pt to memb 1)
[13] call display members (generic member, c)
[14]
[15] c = ’B’
[16] call new (pt to memb 2, 1.0, 2.0, 3.0, 4.0)
[17] call assign (generic member, pt to memb 2)
[18] call display members (generic member, c)
[19]
[20] end program main
[21] ! running gives
[22] ! display memb 1 A
[23] ! display memb 2 B

Figure 9.1: Test of Dynamic Dispatching

[ 1] Module Member 1 Class

[ 2] implicit none
[ 3] type member 1
[ 4] real :: real 1, real 2
[ 5] end type member 1
[ 6]
[ 7] contains
[ 8]
[ 9] subroutine new member 1 (member, a, b)
[10] real, intent(in) :: a, b
[11] type (member 1) :: member
[12] member%real 1 = a ; member%real 2 = b
[13] end subroutine new member 1
[14]
[15] subroutine display memb 1 (pt to memb 1, c)
[16] type (member 1), pointer :: pt to memb 1
[17] character(len=1), intent(in) :: c
[18] print *, ’display memb 1 ’, c
[19] end subroutine display memb 1
[20]
[21] End Module Member 1 Class

Figure 9.2: The First Subtype Class Member

to the subtype of interest is set to point to the desired member. The dispatching process for the display
procedure is different. It requires an if-elseif construct that contains calls to all of the possible subtype
members (two here) and a failsafe default state to abort the process or undertake the necessary exception
handling. Since all but one of the subtype pointer objects have been nullified it employs the ASSOCI-
ATED intrinsic function to select the one, and only, procedure to call and passes the pointer object on to
that procedure. In F90 a pointer can be nullified by using the NULLIFY statement, while F95 allows the
alternative of pointing at the intrinsic NULL function with returns a disassociated pointer. The NULL
function can also be used to define the initial association status of a pointer at the point it is declared.
That is a better programming style.
The are other approaches for implementing the dynamic dispatching concepts. Several examples are
give in the publications by the group Decyk, Norton, and Szymanski (1995, 1997, 1999) and on Prof.
Szymanski’s Web site.

9.3 Non-standard Features

Elsewhere in this work only features of Fortran included in the 1995 standard have been utilized. It
is common for compiler developers to provide addition enhancements, that are hardware or environment

specific, and for the most useful of those features to appear in the next standard release. Compiler releases
by Cray c Digitial c and Silicon Graphics c computers are examples of versions with extensive en-

hancements. Some compilers, like the Digitial c Visual Fortran c are designed to develop applications

c 2001 J.E. Akin 184

[ 1] Module Member 2 Class
[ 2] Use Member 1 class
[ 3] implicit none
[ 4] type member 2
[ 5] type (member 1) :: r 1 2
[ 6] real :: real 3, real 4
[ 7] end type member 2
[ 8]
[ 9] contains
[10]
[11] subroutine new member 2 (member, a, b, c, d)
[12] real, intent(in) :: a, b, c, d
[13] type (member 2) :: member
[14] call new member 1 (member%r 1 2, a, b)
[15] member%real 3 = c ; member%real 4 = d
[16] end subroutine new member 2
[17]
[18] subroutine display memb 2 (pt to memb 2, c)
[19] type (member 2), pointer :: pt to memb 2
[20] character(len=1), intent(in) :: c
[21] print *, ’display memb 2 ’, c
[22] end subroutine display memb 2
[23]
[24] End Module Member 2 Class

Figure 9.3: The Second Subtype Class Member

for the Microsoft c Windows c system and contain library modules for ”standard” graphical displays

via QuickWin c for dialog routines to the Graphical User Interface (GUI), for interfacing with multiple
programming languages or the operation system, and for multiple ”thread” operations. Threads are not
currently in the F90 standard. They allow for response to the user interaction with any of a set of multiple
buttons or dials in an active GUI.
Fortran 90 is a subset of the High Performance Fortran (HPF) standard that has been developed for
use on massively parallel computers. We have not discussed those enhancements.
Even without these special enhancements the OOP abilities of F90 provide an important tool in en-
gineering and scientific programming. In support of that position we close with a quote from computer
scientist Professor Boleslaw K. Szymanski’s Web page on High Performance Object-Oriented Program-
ming in Fortran 90 where his group concludes: ”All of our Fortran 90 programs execute more quickly
than the equivalent C++ versions, yet the abstraction modeling capabilities that we needed were compa-
rably powerful.”

c 2001 J.E. Akin 185

[ 1] Module Is A Member Class
[ 2] Use Member 1 Class ; Use Member 2 Class
[ 3] implicit none
[ 4]
[ 5] type Is A Member
[ 6] private
[ 7] type (member 1), pointer :: pt to memb 1
[ 8] type (member 2), pointer :: pt to memb 2 ! etc for others
[ 9] end type Is A Member
[10]
[11] interface new
[12] module procedure new member 1
[13] module procedure new member 2 ! etc for others
[14] end interface
[15]
[16] interface assign
[17] module procedure assign memb 1
[18] module procedure assign memb 2 ! etc for others
[19] end interface
[20]
[21] interface display
[22] module procedure display memb 1
[23] module procedure display memb 2 ! etc for others
[24] end interface
[25]
[26] contains
[27]
[28] subroutine assign memb 1 (Family, member)
[29] type (member 1), target, intent(in) :: member
[30] type (Is A Member), intent(out) :: Family
[31] call nullify Is A Member (Family) ! nullify all
[32] Family%pt to memb 1 => member
[33] end subroutine assign memb 1
[34]
[35] subroutine assign memb 2 (Family, member)
[36] type (member 2), target, intent(in) :: member
[37] type (Is A Member), intent(out) :: Family
[38] call nullify Is A Member (Family) ! nullify all
[39] Family%pt to memb 2 => member
[40] end subroutine assign memb 2 ! etc for others
[41]
[42] subroutine nullify Is A Member (Family)
[43] type (Is A Member), intent(inout) :: Family
[44] nullify (Family%pt to memb 1)
[45] nullify (Family%pt to memb 2) ! etc for others
[46] end subroutine nullify Is A Member
[47]
[48] subroutine display members (A Member, c)
[49] type (Is A Member), intent(in) :: A Member
[50] character(len=1), intent(in) :: c
[51]
[52] ! select the one proper member
[53] if ( associated (A Member%pt to memb 1) ) then
[54] call display (A Member%pt to memb 1, c)
[55] else if ( associated (A Member%pt to memb 2) ) then
[56] call display (A Member%pt to memb 2, c) ! etc for others
[57] else ! default case
[58] stop ’Error, no member defined in Is A Member Class’
[59] end if
[60] end subroutine display members
[61] End Module Is A Member Class

Figure 9.4: The Polymorphic Class for Subtypes

c 2001 J.E. Akin 186

Appendix A

Bibliography

1. Adams, J.C., Brainerd, W.S., Martin, J.T., Smith, B.T. and Wagener, J.L., Fortran 90 Handbook:
Complete ANSI / ISO Reference, Intertext Publications, McGraw-Hill Book Company, New York,
1992.

2. Akin, J.E. “Object-oriented Programming via Fortran 90,” Engineering Computations, 16(1) 26-
48, 1999.

3. Angell, I.O. and Griffith, G., High Resolution Computer Graphics Using Fortran 77, Macmillan,
London, 1987.

4. Bar-David, T., Object-Oriented Design for C++, Prentice Hall, 1993.

5. Barton, J.J. and L.R. Nackman, Scientific and Engineering C++, Addison Wesley, 1994.

6. Cary, J.R., S.G. Shasharina, J.C. Cummings, J.V.W. Reynders, and P.J. Hinker, “A Comparison of
C++ and Fortran 90 for Object-Oriented Scientific Programming”, Computer Phys. Comm., 105,
20, 1997.

7. Coad, P. and E. Yourdon, Object Oriented Design, Prentice Hall, 1991.

8. Decyk, V.K., Norton,C.D. and B.K. Szymanski, “Expressing Object-Oriented Concepts in

Fortran90,” ACM Fortran Forum, 16,(1), April 1997.

9. Decyk, V.K., Norton,C.D. and B.K. Szymanski, “How to Express C++ Concepts in Fortran90,”
Scientific Programming, 6, 363–390, 1997.

10. Dubois-Pèlerin, Y. and T. Zimmermann, “Object-oriented finite element programming: III. An

efficient implementation in C++,” Comp. Meth. Appl. Mech. Engr., 108, 165–183, 1993.

11. Dubois-Pèlerin, Y. and P. Pegon, “Improving Modularity in Object-Oriented Finite Element Pro-
gramming,” Communications in Numerical Methods in Engineering, 13, 193–198, 1997.

12. Filho, J.S.R.A. and P.R.B. Devloo, “Object Oriented Programming in Scientific Computations,”
Engineering Computations, 8(1), 81–87, 1991.

13. Gray, M.G., and R.M. Roberts, “Object-Based Programming in Fortran 90”, Computers in Physics,
11, 355, 1997.

14. Gehrke, W., Fortran 90 Language Guide, Springer, London, 1995.

15. George, A. and J. Liu “An Object-Oriented Approach to the Design of a User Interface for a Sparse
Matrix Package”, em SIAM J. Matrix Anal. Appl., 20(4), 953–969, 1999.

16. Graham, I., Object Oriented Methods, Addison-Wesley, 1991.

c 2001 J.E. Akin 187

 17. Hahn, B.D., Fortran 90 for Scientists and Engineers, Edward Arnold, London, 1994.
18. Hanly, J.R., Essential C++ for Engineers and Scientists, Addison-Wesley, 1997.
19. Hanselman, D. and Littlefield, B., Mastering Matlab 5, Prentice Hall, 1998.
20. Hubbard, J.R., Programming with C++, McGraw Hill, 1994.
21. Kerrigan, J., Migrating to Fortran 90, O’Reilly & Associates, Sebastopol, CA, 1993.
22. Koelbel, C.H., Loveman, D.B., Schreiber, R.S., Steele, G.L., Jr. and Zosel, M.E., The High Perfor-
mance Fortran Handbook, MIT Press, Cambridge, MA, 1994.
23. Machiels, L. and M.O. Deville, “Fortran 90: On Entry to Object Oriented Programming for the
Solution of Partial Differential Equations,” ACM Trans. Math. Software, 23(1), 32–49, Mar. 1997.
24. Mossberg, E. K. Otto, and M. Thune, “Object-Oriented Software Tools for the Construction of
Preconditioners” Scientific Programming, 6, 285–295, 1997.
25. Nielsen, K., Object-Oriented Development with C++, International Thomson Computer Press,
1997.
26. Norton, C.D., B.K. Szymanski, and V.K. Decyk, “Object Oriented Parallel Computation for Plasma
Simulation”, em Comm. ACM, 38(10), 88, 1995.
27. Norton, C.D., V.K. Decyk, and B.K. Szymanski, “High Performance Object-Oriented Scientific
Programming in Fortran 90”, em Proc. Eighth SIAM Conf. on Parallell Processing for Scientific
Programming, (Ed. Heath et. al.), March 1997.
28. Pratap, R., Getting Started with Matlab, Saunders College Publishing, Ft. Worth, TX, 1996.
29. Press, W.H., Teukolsky, S.A., Vetterling, W.T. and Flannery, B.P., Numerical Recipes in Fortran
77, Cambridge University Press, 1989.
30. Press, W.H., Teukolsky, S.A., Vetterling, W.T. and Flannery, B.P., Numerical Recipes in Fortran
90, 2nd ed., Cambridge University Press, 1996.
31. Rumbaugh, J., M. Blaha, W. Premerlani, F. Eddy and W. Lorensen, Object Oriented Modeling and
Design, Prentice Hall, 1991.
32. Thomas, P. and Weedon, R., Object-Oriented Programming in Eiffel, Addison-Wesley, 1995.
Links to World Wide Web sites (as of 2001, subject to change):
33. http://blas.mcmaster.ca/ fred/oo.html
34. http://citeseer.nj.nec.com/242268.html
35. http://csep1.phy.ornl.gov/CSEP/PL/PL.html
36. http://epubs.siam.org/sam-bin/dbq/article/31773
37. http://kanaima.ciens.ucv.ve/hpf/HTMLNotesnode29.html
38. http://mis.ucd.ie/courses/mms402/prentice.htm
39. http://webserv.gsfc.nasa.gov/ESS/annual.reports/ess98/cdn.html
40. http://www.amath.washington.edu/ lf/software/CompCPP F90SciOOP.html
41. http://www.arithmetica.ch/Oberon/CFORTRANOberon.nhtml
42. http://www.cs.rpi.edu/ szymansk/OOF90

c 2001 J.E. Akin 188

 43. http://www.nasatech.com/Briefs/Mar98/NPO20180.html
44. http://www.owlnet.rice.edu/ mech517/F90 docs/EC oop f90.pdf
45. http://www.ssec.wisc.edu/ robert/Software/F90-ObjOrientProg.html
46. http://www.tdb.uu.se/ ngssc/OOP00/module2/
47. http://www.ticra.dk/ooa.htm

c 2001 J.E. Akin 189

 c 2001 J.E. Akin 190
Appendix B

Fortran 90 Overview

This overview of Fortran 90 (F90) features is presented as a series of tables that illustrate the syntax and
abilities of F90. Frequently comparisons are made to similar features in the C++ and F77 languages and
to the Matlab environment.
These tables show that F90 has significant improvements over F77 and matches or exceeds newer
software capabilities found in C++ and Matlab for dynamic memory management, user defined data
structures, matrix operations, operator definition and overloading, intrinsics for vector and parallel pro-
cessors and the basic requirements for object-oriented programming.
They are intended to serve as a condensed quick reference guide for programming in F90 and for
understanding programs developed by others.

B.1 List of Language Tables

1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.17 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.18 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

c 2001 J.E. Akin 191
 4.25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.27 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.28 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.29 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.30 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.31 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.33 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.34 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

5.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.10 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

8.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
8.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.10 Intrinsic Functions Allowing Logical Mask Control . . . . . . . . . . . . . . . . . . . 165
8.11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
8.13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
8.14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

B.1 Comment syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

B.2 Intrinsic data types of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
B.3 Arithmetic operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
B.4 Relational operators (arithmetic and logical) . . . . . . . . . . . . . . . . . . . . . . . 4
B.5 Precedence pecking order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
B.6 Colon Operator Syntax and its Applications . . . . . . . . . . . . . . . . . . . . . . . 4
B.7 Mathematical functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
B.8 Flow Control Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
B.9 Basic loop constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
B.10 IF Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
B.11 Nested IF Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
B.12 Logical IF-ELSE Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
B.13 Logical IF-ELSE-IF Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
B.14 Case Selection Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
B.15 F90 Optional Logic Block Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
B.16 GO TO Break-out of Nested Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
B.17 Skip a Single Loop Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
B.18 Abort a Single Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

c 2001 J.E. Akin 192
 B.19 F90 DOs Named for Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
B.20 Looping While a Condition is True . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
B.21 Function definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
B.22 Arguments and return values of subprograms . . . . . . . . . . . . . . . . . . . . . . 10
B.23 Defining and referring to global variables . . . . . . . . . . . . . . . . . . . . . . . . 11
B.24 Bit Function Intrinsics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
B.25 The ACSII Character Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
B.26 F90 Character Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
B.27 How to type non-printing characters . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
B.28 Referencing Structure Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
B.29 Defining New Types of Data Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 13
B.30 Nested Data Structure Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
B.31 Declaring, initializing, and assigning components of user-defined datatypes . . . . . . 13
B.32 F90 Derived Type Component Interpretation . . . . . . . . . . . . . . . . . . . . . . 14
B.33 Definition of pointers and accessing their targets . . . . . . . . . . . . . . . . . . . . . 14
B.34 Nullifing a Pointer to Break Association with Target . . . . . . . . . . . . . . . . . . . 14
B.35 Special Array Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
B.36 Array Operations in Programming Constructs . . . . . . . . . . . . . . . . . . . . . . 15
B.37 Equivalent Fortran 90 and M ATLAB Intrinsic Functions . . . . . . . . . . . . . . . . . 16
B.38 Truncating Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
B.39 F90 WHERE Constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
B.40 F90 Array Operators with Logic Mask Control . . . . . . . . . . . . . . . . . . . . . 18
B.41 Array initialization constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
B.42 Array initialization constructs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
B.43 Elementary matrix computational routines . . . . . . . . . . . . . . . . . . . . . . . . 34
B.44 Dynamic allocation of arrays and pointers . . . . . . . . . . . . . . . . . . . . . . . . 34
B.45 Automatic memory management of local scope arrays . . . . . . . . . . . . . . . . . . 35
B.46 F90 Single Inheritance Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
B.47 F90 Selective Single Inheritance Form . . . . . . . . . . . . . . . . . . . . . . . . . . 35
B.48 F90 Single Inheritance Form, with Local Renaming . . . . . . . . . . . . . . . . . . . 35
B.49 F90 Multiple Selective Inheritance with Renaming . . . . . . . . . . . . . . . . . . . 36

c 2001 J.E. Akin 2
 Language Syntax Location
M ATLAB % comment (to end of line) anywhere
C /*comment*/ anywhere
F90 ! comment (to end of line) anywhere
F77 * comment (to end of line) column 1

Table B.1: Comment syntax.

Storage M ATLABa C++ F90 F77

byte char character:: character
integer int integer:: integer
single precision float real:: real
double precision double real*8:: double precision
b
complex complex:: complex
Boolean bool logical:: logical
argument parameter:: parameter
pointer * pointer::
structure struct type::
a M ATLAB 4 requires no variable type declaration; the only two distinct types in M ATLAB are strings and reals (which include

complex). Booleans are just 0s and 1s treated as reals. M ATLAB5 allows the user to select more types.
b There is no specific data type for a complex variable in C++; they must be created by the programmer.

Table B.2: Intrinsic data types of variables.

Description M ATLABa C++ Fortranb

addition + + +
subtractionc - - -
multiplication * and .* * *
division / and ./ / /
exponentiation ˆ and .ˆ powd **
remainder %
increment ++
decrement --
parentheses (expres- () () ()
sion grouping)
a When doing arithmetic operations on matrices in M ATLAB , a period (‘.’) must be put before the operator if scalar arithmetic

is desired. Otherwise, M ATLAB assumes matrix operations; figure out the difference between ‘*’ and ‘.*’. Note that since matrix
and scalar addition coincide, no ‘.+’ operator exists (same holds for subtraction).
b Fortran 90 allows the user to change operators and to define new operator symbols.
c In all languages the minus sign is used for negation (i.e., changing sign).
d In C++ the exponentiation xy is calculated by function pow (x; y ).

Table B.3: Arithmetic operators.

c 2001 J.E. Akin 3
 Description M ATLAB C++ F90 F77
Equal to == == == .EQ.
Not equal to ˜= != /= .NE.
Less than < < < .LT.
Less or equal <= <= <= .LE.
Greater than > > > .GT.
Greater or equal >= >= >= .GE.
Logical NOT ˜ ! .NOT. .NOT.
Logical AND & && .AND. .AND.
Logical inclusive OR ! || .OR. .OR.
Logical exclusive OR xor .XOR. .XOR.
Logical equivalent == == .EQV. .EQV.
Logical not equivalent ˜= != .NEQV. .NEQV.

Table B.4: Relational operators (arithmetic and logical).

M ATLAB
C++ Operators F90 Operatorsa F77 Operators
Operators
() () [] -> . () ()
+ - ! ++ -- + ** **
- * & (type)
sizeof
* / * / % * / * /
+ -b + -b + -b + -b
< <= > >= << >> // //
== ˜= < <= > => == /= < <= > .EQ. .NE.
>= .LT. .LE.
.GT. .GE.
˜ == != .NOT. .NOT.
& && .AND. .AND.
| || .OR. .OR.
= | .EQV. .NEQV. .EQV. .NEQV.
?:
= += -= *= /=
%= &= ˆ= |=
<<= >>=
,
a User-defined unary (binary) operators have the highest (lowest) precedence in F90.
b These are binary operators representing addition and subtraction. Unary operators + and - have higher precedence.

Table B.5: Precedence pecking order.

B = Beginning, E = Ending, I = Increment

Syntax F90 M ATLAB Use F90 M ATLAB

Default B:E:I B:I:E Array subscript ranges yes yes
B B: B: Character positions in a string yes yes
E :E :E Loop control no yes
Full range : : Array element generation no yes

Table B.6: Colon Operator Syntax and its Applications.

c 2001 J.E. Akin 4
 Description M ATLAB C++ F90 F77
exponential exp(x) exp(x) exp(x) exp(x)
natural log log(x) log(x) log(x) log(x)
base 10 log log10(x) log10(x) log10(x) log10(x)
square root sqrt(x) sqrt(x) sqrt(x) sqrt(x)
raise to power (xr ) x.ˆr pow(x,r) x**r x**r
absolute value abs(x) fabs(x) abs(x) abs(x)
smallest integer>x ceil(x) ceil(x) ceiling(x)
largest integer<x floor(x) floor(x) floor(x)
division remainder rem(x,y) fmod(x,y) mod(x,y)a mod(x,y)
modulo modulo(x,y)a
complex conjugate conj(z) conjg(z) conjg(z)
imaginary part imag(z) imag(z) aimag(z)
drop fraction fix(x) aint(x) aint(x)
round number round(x) nint(x) nint(x)
cosine cos(x) cos(x) cos(x) cos(x)
sine sin(x) sin(x) sin(x) sin(x)
tangent tan(x) tan(x) tan(x) tan(x)
arc cosine acos(x) acos(x) acos(x) acos(x)
arc sine asin(x) asin(x) asin(x) asin(x)
arc tangent atan(x) atan(x) atan(x) atan(x)
arc tangentb atan2(x,y) atan2(x,y) atan2(x,y) atan2(x,y)
hyperbolic cosine cosh(x) cosh(x) cosh(x) cosh(x)
hyperbolic sine sinh(x) sinh(x) sinh(x) sinh(x)
hyperbolic tangent tanh(x) tanh(x) tanh(x) tanh(x)
hyperbolic arc cosine acosh(x)
hyperbolic arc sine asinh(x)
hyperbolic arctan atanh(x)
a Differ for x < 0.
b atan2(x,y) is used to calculate the arc tangent of x=y in the range [ ; +]. The one-argument function atan(x)
computes the arc tangent of x in the range [ =2; +=2].

Table B.7: Mathematical functions.

c 2001 J.E. Akin 5
 Description C++ F90 F77 M ATLAB
Conditionally execute statements if if if if
f g end if end if end

Loop a specific number of times for k=1:n do k=1,n do # k=1,n for k=1:n
f g end do # continue end

Loop an indefinite number of times while do while — while

f g end do — end

Terminate and exit loop break exit go to break

Skip a cycle of loop continue cycle go to —

Display message and abort error() stop stop error

Return to invoking function return return return return

Conditional array action — where — if

Conditional alternate statements else else else else

else if elseif elseif elseif

Conditional array alternatives — elsewhere — else

— — — elseif

Conditional case selections switch f g select case if if

end select end if end

Table B.8: Flow Control Statements.

Loop M ATLAB C++ Fortran

Indexed loop
f
for index=matrix for (init;test;inc) do index=b,e,i
statements statements
end statements end do
g
Pre-test loop while test while (test) f do while (test)
statements statements statements
end g end do

Post-test loop do f do
statements statements
g while (test) if (test) exit
end do

Table B.9: Basic loop constructs.

c 2001 J.E. Akin 6
 M ATLAB Fortran C++

f
if l expression IF (l expression) THEN if (l expression)
true group true group
end END IF true group;
g
IF (l expression) true statement if (l expression)
true statement;

Table B.10: IF Constructs. The quantity l expression means a logical expression having a value that
is either TRUE of FALSE. The term true statement or true group means that the statement or group
of statements, respectively, are executed if the conditional in the if statement evaluates to TRUE.

M ATLAB Fortran C++

f
if l expression1 IF (l expression1) THEN if (l expression1)
true group A true group A
if l expression2 IF (l expression2) THEN true group A
true group B true group B
f
if (l expression2)
end END IF
true group C true group C true group B
end END IF g
statement group D statement group D true group C
g
statement group D

Table B.11: Nested IF Constructs.

M ATLAB Fortran C++

f
if l expression IF (l expression) THEN if (l expression)
true group A true group A
else ELSE true group A
false group B false group B g
f
end END IF else
false group B
g

Table B.12: Logical IF-ELSE Constructs.

M ATLAB Fortran C++

f
if l expression1 IF (l expression1) THEN if (l expression1)
true group A true group A
true group A
g
elseif l expression2 ELSE IF (l expression2) THEN
true group B true group B
f
elseif l expression3 ELSE IF (l expression3) THEN else if (l expression2)
true group C true group C
true group B
g
else ELSE
default group D default group D
end END IF else if (l expression3)
f
true group C
g
else
f
default group D
g

Table B.13: Logical IF-ELSE-IF Constructs.

c 2001 J.E. Akin 7
 F90 C++

f
SELECT CASE (expression) switch (expression)
CASE (value 1)
group 1 case value 1 :
CASE (value 2) group 1
group 2 break;
. case value 2 :
. group 2
.
CASE (value n) break;
group n .
.
CASE DEFAULT .
default group case value n :
END SELECT group n
break;
default:
default group
g
break;

Table B.14: Case Selection Constructs.

F90 Named IF F90Named SELECT

name: IF (logical 1) THEN name: SELECT CASE (expression)
true group A CASE (value 1)
ELSE IF (logical 2) THEN group 1
true group B CASE (value 2)
ELSE group 2
default group C CASE DEFAULT
ENDIF name default group
END SELECT name

Table B.15: F90 Optional Logic Block Names.

Fortran C++
DO 1 ... for (...) f
DO 2 ...
...
for (...) f
...
IF (disaster) THEN if (disaster)
GO TO 3 go to error
END IF
g
...
...
2 END DO
1 END DO
g
error:
3 next statement

Table B.16: GO TO Break-out of Nested Loops. This situation can be an exception to the general recom-
mendation to avoid GO TO statements.

F77 F90 C++

DO 1 I = 1,N
f
DO I = 1,N for (i=1; i<n; i++)
... ...
IF (skip condition) THEN IF (skip condition) THEN if (skip condition)
GO TO 1 CYCLE ! to next I continue; // to next
ELSE ELSE else if
false group false group false group
END IF END IF end
1 continue END DO g

Table B.17: Skip a Single Loop Cycle.

c 2001 J.E. Akin 8
 F77 F90 C++
DO 1 I = 1,N
f
DO I = 1,N for (i=1; i<n; i++)
IF (exit condition) THEN IF (exit condition) THEN
GO TO 2 EXIT ! this do if (exit condition)
ELSE ELSE break;// out of loop
false group false group else if
END IF END IF false group
1 CONTINUE END DO end
2 next statement next statement g
next statement

Table B.18: Abort a Single Loop.

main: DO ! forever
test: DO k=1,k max
third: DO m=m max,m min,-1
IF (test condition) THEN
CYCLE test ! loop on k
END IF
END DO third ! loop on m
fourth: DO n=n min,n max,2
IF (main condition) THEN
EXIT main ! forever loop
END DO fourth ! on n
END DO test ! over k
END DO main
next statement

Table B.19: F90 DOs Named for Control.

M ATLAB C++
initialize test initialize test
f
while l expression while (l expression)
true group
change test true group
change test
g
end

F77 F90
initialize test initialize test
# continue do while (l expression)
IF (l expression) THEN true group
true group change test
change test end do
go to #
END IF

Table B.20: Looping While a Condition is True.

c 2001 J.E. Akin 9
 Function
M ATLABa C++ Fortran
Type
program statements
f
main(argc,char **argv) program main
[y1...yn]=f(a1,...,am) type y
[end of file] statements type a1,...,type am
statements
g
y = f(a1,I,am);
y = f(a1,...,am)
call s(a1,...,am)
end program

subroutine void f subroutine s(a1,...,am)

(type a1,...,type am) type a1,...,type am
f statements
statements
g
end

function function [r1...rn] type f (type a1,...,type am) function f(a1,...,am)

=f(a1,...,am)
statements f statements g type f
type a1,...,type am
statements
end

a Every function or program in M ATLAB must be in separate files.

Table B.21: Function definitions. In each case, the function being defined is named f and is called with
m arguments a1,...,am.

One-Input, One-Result Procedures

M ATLAB function out = name (in)
F90 function name (in) ! name = out
function name (in) result (out)
C++ name (in, out)

Multiple-Input, Multiple-Result Procedures

M ATLAB function [inout, out2] = name (in1, in2, inout)
F90 subroutine name (in1, in2, inout, out2)
C++ name(in1, in2, inout, out2)

Table B.22: Arguments and return values of subprograms.

c 2001 J.E. Akin 10
 Global Variable Declaration
M ATLAB global list of variables
F77 common /set name/ list of variables
F90 module set name
save
type (type tag) :: list of variables
end module set name
C++ extern list of variables

Access to Global Variables

M ATLAB global list of variables
F77 common /set name/ list of variables
F90 use set name, only subset of variables
use set name2 list of variables
C++ extern list of variables

Table B.23: Defining and referring to global variables.

Action C++ F90

Bitwise AND & iand
Bitwise exclusive OR ^ ieor
Bitwise exclusive OR j ior
Circular bit shift ishftc
Clear bit ibclr
Combination of bits mvbits
Extract bit ibits
Logical complement  not
Number of bits in integer sizeof bit size
Set bit ibset
Shift bit left  ishft
Shift bit right  ishft
Test on or off btest
Transfer bits to integer transfer

Table B.24: Bit Function Intrinsics.

c 2001 J.E. Akin 11
 0 NUL 1 SOH 2 STX 3 ETX 4 EOT 5 ENQ 6 ACK 7 BEL
8 BS 9 HT 10 NL 11 VT 12 NP 13 CR 14 SO 15 SI
16 DLE 17 DC1 18 DC2 19 DC3 20 DC4 21 NAK 22 SYN 23 ETB
24 CAN 25 EM 26 SUB 27 ESC 28 FS 29 GS 30 RS 31 US
32 SP 33 ! 34 " 35 # 36 $ 37 % 38 & 39 ’
40 ( 41 ) 42 * 43 + 44 , 45 - 46 . 47 /
48 0 49 1 50 2 51 3 52 4 53 5 54 6 55 7
56 8 57 9 58 : 59 ; 60 < 61 = 62 > 63 ?
64 @ 65 A 66 B 67 C 68 D 69 E 70 F 71 G
72 H 73 I 74 J 75 K 76 L 77 M 78 N 79 O
80 P 81 Q 82 R 83 S 84 T 85 U 86 V 87 W
88 X 89 Y 90 Z 91 [ 92 \ 93 ] 94 ˆ 95 _
96 ‘ 97 a 98 b 99 c 100 d 101 e 102 f 103 g
104 h 105 i 106 j 107 k 108 l 109 m 110 n 111 o
112 p 113 q 114 r 115 s 116 t 117 u 118 v 119 w
120 x 121 y 122 z 123 { 124 | 125 } 126 ˜ 127 DEL
Table B.25: The ACSII Character Set.

ACHAR (I) Character number I in ASCII collating set

ADJUSTL (STRING) Adjust left
ADJUSTR (STRING) Adjust right
CHAR (I)  Character I in processor collating set
IACHAR (C) Position of C in ASCII collating set
ICHAR (C) Position of C in processor collating set
INDEX (STRING, SUBSTRING)a Starting position of a substring
LEN (STRING) Length of a character entity
LEN TRIM (STRING) Length without trailing blanks
LGE (STRING A, STRING B) Lexically greater than or equal
LGT (STRING A, STRING B) Lexically greater than
LLE (STRING A, STRING B) Lexically less than or equal
LLT (STRING A, STRING B) Lexically less than
REPEAT (STRING, NCOPIES) Repeated concatenation
SCAN (STRING, SET)a Scan a string for a character in a set
TRIM (STRING) Remove trailing blank characters
VERIFY (STRING, SET)a Verify the set of characters in a string
STRING A//STRING B Concatenate two strings
a Optional arguments not shown.

Table B.26: F90 Character Functions.

Action ASCII Character F90 Inputa C++ Input

Alert (Bell) 7 Ctrl-G na
Backspace 8 Ctrl-H nb
Carriage Return 13 Ctrl-M nr
End of Transmission 4 Ctrl-D Ctrl-D
Form Feed 12 Ctrl-L nf
Horizontal Tab 9 Ctrl-I nt
New Line 10 Ctrl-J nn
Vertical Tab 11 Ctrl-K nv
a “Ctrl-” denotes control action. That is, simultaneous pressing of the CONTROL key and the letter following.

Table B.27: How to type non-printing characters.

c 2001 J.E. Akin 12
 C, C++ Variable.component.sub component
F90 Variable%component%sub component

Table B.28: Referencing Structure Components.

C, C++ struct data tag f

intrinsic type 1 component names;
intrinsic type 2 component names;
g ;

F90 type data tag

intrinsic type 1 :: component names;
intrinsic type 2 :: component names;
end type data tag

Table B.29: Defining New Types of Data Structure.

C, C++ struct data tag f

intrinsic type 1 component names;
g
struct tag 2 component names;
;

F90 type data tag

intrinsic type :: component names;
type (tag 2) :: component names;
end type data tag

Table B.30: Nested Data Structure Definitions.

C, C++
f g
struct data tag variable list; /* Definition */
struct data tag variable = component values ; /* Initialization */
variable.component.sub component = value; /* Assignment */

F90 type (data tag) :: variable list ! Definition

variable = data tag (component values) ! Initialization
variable%component%sub component = value ! Assignment

Table B.31: Declaring, initializing, and assigning components of user-defined datatypes.

c 2001 J.E. Akin 13
 INTEGER, PARAMETER :: j max = 6
TYPE meaning demo
INTEGER, PARAMETER :: k max = 9, word = 15
CHARACTER (LEN = word) :: name(k max)
END TYPE meaning demo
TYPE (meaning demo) derived(j max)
Construct Interpretation
derived All components of all derived’s elements
derived(j) All components of jth element of derived
derived(j)%name All k max components of name within jth element of derived
derived%name(k) Component k of the name array for all elements of derived
derived(j)%name(k) Component k of the name array of jth element of derived

Table B.32: F90 Derived Type Component Interpretation.

C++ F90
Declaration type tag *pointer name; type (type tag), pointer ::
pointer name
Target &target name type (type tag), target :: target name
Examples char *cp, c; character, pointer :: cp
int *ip, i; integer, pointer :: ip
float *fp, f; real, pointer :: fp
cp = & c; cp => c
ip = & i; ip => i
fp = & f; fp => f

Table B.33: Definition of pointers and accessing their targets.

C, C++ pointer name = NULL

F90 nullify (list of pointer names)
F95 pointer name = NULL()

Table B.34: Nullifing a Pointer to Break Association with Target.

Purpose F90 M ATLAB

Form subscripts () ()
Separates subscripts & elements , ,
Generates elements & subscripts : :
Separate commands ; ;
Forms arrays (/ /) []
Continue to new line & ...
Indicate comment ! %
Suppress printing default ;

Table B.35: Special Array Characters.

c 2001 J.E. Akin 14

c 2001
J.E. Akin

Description Equation Fortran90 Operator Matlab Operator Original Sizes Result Size
Scalar plus scalar c = a  b c = a  b c = a  b; 1; 1 1; 1
Element plus scalar cjk 
= ajk b c = a  b c = a  b; m; n and 1; 1 m; n

Element plus element cjk = jk  jk

a b c = a  b c = a  b; m; n and m; n m; n

Scalar times scalar = 

c a b c = a  b c = a  b; 1; 1 1; 1
Element times scalar cjk = jk  a b c = a  b c = a  b; m; n and 1; 1 m; n

Element times element cjk = jk  jka b c = a  b c = a . b; m; n and m; n m; n

Scalar divide scalar c = a=b c = a=b c = a=b; 1; 1 1; 1

Scalar divide element cjk = ajk =b c = a=b c = a=b; m; n and 1; 1 m; n

Element divide element cjk = ajk =bjk c = a=b c = a:=b; m; n and m; n m; n

Scalar power scalar c = a

b c = a  b c = a ^ b; 1; 1 1; 1
Element power scalar cjk
b=
jk a c = a  b c = a ^ b; m; n and 1; 1 m; n

Element power element cjk = a

bjk
jk c = a  b c = a .^ b; m; n and m; n m; n

Matrix transpose Ckj = Ajk = transpose (A) = 0;

15

C C A m; n n; m
P
Matrix times matrix Cij = k Aik Bkj C = matmul(A; B ) C = A  B; m; r and r; n m; n
P
Vector dot vector c = k Ak Bk c = sum(A  B) c = sum(A:  B ); m; 1 and m; 1 1; 1
c = dot product(A; B ) c = A  B
0; m; 1 and m; 1 1; 1

Table B.36: Array Operations in Programming Constructs. Lower case letters denote scalars or scalar elements of arrays. Matlab arrays are allowed a maximum
of two subscripts while Fortran allows seven. Upper case letters denote matrices or scalar elements of matrices.
 Table B.37: Equivalent Fortran90 and M ATLAB Intrinsic Functions.
The following KEY symbols are utilized to denote the TYPE of the in-
trinsic function, or subroutine, and its arguments: A-complex, integer,
or real; I-integer; L-logical; M-mask (logical); R-real; X-real; Y-real;
V-vector (rank 1 array); and Z-complex. Optional arguments are not
shown. Fortran 90 and MATLAB also have very similar array operations
and colon operators.

Type Fortran90 M ATLAB Brief Description

A ABS(A) abs(a) Absolute value of A.
R ACOS(X) acos(x) Arc cosine function of real X.
R AIMAG(Z) imag(z) Imaginary part of complex number.
R AINT(X) real(fix(x)) Truncate X to a real whole number.
L ALL(M) all(m) True if all mask elements, M, are true.
R ANINT(X) real(round(x)) Real whole number nearest to X.
L ANY(M) any(m) True if any mask element, M, is true.
R ASIN(X) asin(x) Arcsine function of real X.
R ATAN(X) atan(x) Arctangent function of real X.
R ATAN2(Y,X) atan2(y,x) Arctangent for complex number(X, Y).
I CEILING(X) ceil(x) Least integer >= real X.
Z CMPLX(X,Y) (x+yi) Convert real(s) to complex type.
Z CONJG(Z) conj(z) Conjugate of complex number Z.
R COS(R Z) cos(r z) Cosine of real or complex argument.
R COSH(X) cosh(x) Hyperbolic cosine function of real X.
I COUNT(M) sum(m==1) Number of true mask, M, elements.
R,L DOT PRODUCT(X,Y) x’?y Dot product of vectors X and Y.
R EPSILON(X) eps Number, like X,  1.
R,Z EXP(R Z) exp(r z) Exponential of real or complex number.
I FLOOR(X) floor Greatest integer  X.
R HUGE(X) realmax Largest number like X.
I INT(A) fix(a) Convert A to integer type.
R LOG(R Z) log(r z) Logarithm of real or complex number.
R LOG10(X) log10(x) Base 10 logarithm function of real X.
R MATMUL(X,Y) x?y Conformable matrix multiplication, X*Y.
I,V I=MAXLOC(X) [y,i]=max(x) Location(s) of maximum array element.
R Y=MAXVAL(X) y=max(x) Value of maximum array element.
I,V I=MINLOC(X) [y,i]=min(x) Location(s) of minimum array element.
R Y=MINVAL(X) y=min(x) Value of minimum array element.
I NINT(X) round(x) Integer nearest to real X.
A PRODUCT(A) prod(a) Product of array elements.
call RANDOM NUMBER(X) x=rand Pseudo-random numbers in (0; 1).
call RANDOM SEED rand(’seed’) Initialize random number generator.
R REAL (A) real(a) Convert A to real type.
R RESHAPE(X, (/ I, I2 /)) reshape(x, i, i2) Reshape array X into II2 array.
I,V SHAPE(X) size(x) Array (or scalar) shape vector.
R SIGN(X,Y) Absolute value of X times sign of Y.
R SIGN(0.5,X)-SIGN(0.5,-X) sign(x) Signum, normalized sign, –1, 0, or 1.
R,Z SIN(R Z) sin(r z) Sine of real or complex number.
R SINH(X) sinh(x) Hyperbolic sine function of real X.
I SIZE(X) length(x) Total number of elements in array X.
R,Z SQRT(R Z) sqrt(r z) Square root, of real or complex number.
R SUM(X) sum(x) Sum of array elements.
(continued)

c 2001 J.E. Akin 16
 Type Fortran90 M ATLAB Brief Description
R TAN(X) tan(x) Tangent function of real X.
R TANH(X) tanh(x) Hyperbolic tangent function of real X.
R TINY(X) realmin Smallest positive number like X.
R TRANSPOSE(X) x’ Matrix transpose of any type matrix.
R X=1 x=ones(length(x)) Set all elements to 1.
R X=0 x=zero(length(x)) Set all elements to 0.
For more detailed descriptions and example uses of these intrinsic functions see Adams, J.C., et al.,
Fortran 90 Handbook, McGraw-Hill, New York, 1992, ISBN 0–07–000406–4.

B.2 Alphabetical Table of Fortran 90 Intrinsic Routines

The following KEY symbols are utilized to denote the TYPE of the intrinsic function, or subroutine, and
its arguments: A-complex, integer, or real; B-integer bit; C-character; D-dimension; I-integer; K-kind;
L-logical; M-mask (logical); N-integer, or real; P-pointer; R-real; S-string; T-target; V-vector (rank one
array); X-real; Y-real; Z-complex; and *-any type. For more detailed descriptions and example uses of
these intrinsic functions see Adams, J.C., et al., Fortran 90 Handbook, McGraw-Hill, New York, 1992,
ISBN 0–07–000406–4.

C++ – int – – floor ceil

F90 aint int anint nint floor ceiling
M ATLAB real (fix) fix real (round) round floor ceil
Argument Value of Result
–2.000 –2.0 –2 –2.0 –2 –2 –2
–1.999 –1.0 –1 –2.0 –2 –2 –1
–1.500 –1.0 –1 –2.0 –2 –2 –1
–1.499 –1.0 –1 –1.0 –1 –2 –1
–1.000 –1.0 –1 –1.0 –1 –1 –1
–0.999 0.0 0 –1.0 –1 –1 0
–0.500 0.0 0 –1.0 –1 –1 0
–0.499 0.0 0 0.0 0 –1 0
0.000 0.0 0 0.0 0 0 0
0.499 0.0 0 0.0 0 0 1
0.500 0.0 0 1.0 1 0 1
0.999 0.0 0 1.0 1 0 1
1.000 1.0 1 1.0 1 1 1
1.499 1.0 1 1.0 1 1 2
1.500 1.0 1 2.0 2 1 2
1.999 1.0 1 2.0 2 1 2
2.000 2.0 2 2.0 2 2 2

Table B.38: Truncating Numbers.

WHERE (logical array expression)

true array assignments
ELSEWHERE
false array assignments
END WHERE

WHERE (logical array expression) true array assignment

Table B.39: F90 WHERE Constructs.

c 2001 J.E. Akin 17
 Function Description Opt Example
all Find if all values are true, for a fixed di- d all(B = A, DIM = 1)
mension. (true, false, false)
any Find if any value is true, for a fixed di- d any (B > 2, DIM = 1)
mension. (false, true, true)
count Count number of true elements for a d count(A = B, DIM = 2)
fixed dimension. (1, 2)
maxloc Locate first element with maximum m maxloc(A, A < 9)
value given by mask. (2, 3)
maxval Max element, for fixed dimension, given b maxval (B, DIM=1, B > 0)
by mask. (2, 4, 6)
merge Pick true array, A, or false array, B, ac- – merge(A, B,
 L)
cording to mask, L. 0 3 5
2 4 8
minloc Locate first element with minimum value m minloc(A, A > 3)
given by mask. (2, 2)
minval Min element, for fixed dimension, given b minval(B, DIM = 2)
by mask. (1, 2)
pack Pack array, A, into a vector under control v pack(A, B < 4)
of mask. (0, 7, 3)
product Product of all elements, for fixed dimen- b product(B) ; (720)
sion, controlled by mask. product(B, DIM = 1, T)
(2, 12, 30)
sum Sum all elements, for fixed dimension, b sum(B) ;(21)
controlled by mask. sum(B, DIM = 2, T)
(9, 12)
unpack Replace the true locations in array B con- – 
unpack(U, L, B)
trolled by mask L with elements from the 7 3 8
vector U. 2 4 9

     
0 3 5 1 3 5 T F T
A = ; B = ; L = ; U = (7; 8; 9)
7 4 8 2 4 6 F F T

Table B.40: F90 Array Operators with Logic Mask Control. T and F denote true and false, respectively.
Optional arguments: b -- DIM & MASK, d -- DIM, m -- MASK, v -- VECTOR and DIM = 1 implies
for any rows, DIM = 2 for any columns, and DIM = 3 for any plane.

Type Intrinsic Description

A ABS (A) Absolute value of A.
C ACHAR (I) Character in position I of ASCII collating sequence.
R ACOS (X) Arc cosine (inverse cosine) function of real X.
C ADJUSTL (S) Adjust S left, move leading blanks to trailing blanks.
C ADJUSTR (S) Adjust S right, move trailing blanks to leading blanks.
R AIMAG (Z) Imaginary part of complex number, Z.
R AINT (X [,K]) Truncate X to a real whole number, of the given kind.
L ALL (M [,D]) True if all mask, M, elements are true, in dimension D.
L ALLOCATED (* ARRAY P) True if the array or pointer is allocated.
(continued)

c 2001 J.E. Akin 18
 Alphabetic Table of Fortran90 Intrinsic Functions (continued)
Type Intrinsic Description
R ANINT (X [,K]) Real whole number nearest to X, of the given kind.
L ANY (M [,D]) True if any mask, M, element is true, in dimension D.
R ASIN (X) Arcsine (inverse sine) function of real X.
L ASSOCIATED (P [,T]) True if pointer, P, is associated with any target, or T.
R ATAN (X) Arctangent (inverse tangent) function of real X.
R ATAN2 (Y,X) Arctangent for argument of complex number (X, Y).
I BIT SIZE (I) Maximum number of bits integer I can hold, e.g. 32.
L BTEST (I,I POS) True if bit location I POS of integer I has value 1.
I CEILING (X) Least integer  real X, of the given kind.
C CHAR (I [,K]) Character in position I of processor collating sequence.
Z CMPLX (X [,Y][,K]) Convert real(s) to complex type, of given kind.
Z CONJG (Z) Conjugate of complex number Z.
R COS (R Z) Cosine function of real or complex argument.
R COSH (X) Hyperbolic cosine function of real X.
I COUNT (M [,D]) Number of true mask, M, elements, in dimension D.
? CSHIFT (? ARAY,I SHIF [,D]) Circular shift out and in for I SHIF elements.
call DATE AND TIME ([S DATE] Real-time clock date, time, zone, and vector
[,S TIME] [,S ZONE] with year, month, day, UTC, hour, minutes, seconds,
[,I V VALUES]) and milliseconds.
R DBLE (A) Convert A to double precision real.
N DIGITS (N) Number of significant digits for N, e.g. 31.
R DIM (X,Y) The difference, MAX (X – Y, 0.0).
N,L DOT PRODUCT (V,V 2) Dot product of vectors V and V 2.
R DPROD (X,Y) Double precision real product of two real scalars.
? EOSHIFT (? ARRAY, Perform vector end-off shift by  I shift terms,
I SHIFT [,? FILL][,D]) and fill, in dimension D.
R EPSILON (X) Number  1, for numbers like X, e.g. 2??–23.
R,Z EXP (R Z) Exponential function of real or complex argument.
I EXPONENT (X) Exponent part of the model for real X.
I FLOOR (X) Greatest integer less than or equal to X.
R FRACTION (X) Fractional part of the model for real X.
N HUGE (N) Largest number for numbers like N, e.g. 2??128.
I IACHAR (C) Position of character C in ASCII collation.
B IAND (I,I 2) Logical AND on the bits of I and I 2.
B IBCLR (I,I POS) Clear bit I POS to zero in integer I.
B IBITS (I,I POS,I LEN) Extract an I LEN sequence of bits at I POS in I.
B IBSET (I,I POS) Set bit I POS to one in integer I.
I ICHAR (C) Position of character C in processor collation.
B IEOR (I,I 2) Exclusive OR on the bits of I and I 2.
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (right).
I INT (A [,K]) Convert A to integer type, of given kind.
B IOR (I,I 2) Inclusive OR on the bits of I and I 2.
B ISHFT (I,I SHIFT) Logical shift of bits of I by I SHIFT, pad with 0.
B ISHFTC (I,I SHIFT [,I SIZE]) Logical circular shift of I SIZE rightmost bits of I.
I KIND (ANY) Kind type integer parameter value for any argument.
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
I LEN (S) Total character string length.
I LEN TRIM (S) Length of S without trailing blanks.
L LGE (S,S 2) True if S > or equal to S 2 in ASCII sequence.
L LGT (S,S 2) True if S follows S 2 in ASCII collating sequence.
(continued)

c 2001 J.E. Akin 19
 Alphabetic Table of Fortran90 Intrinsic Functions (continued)
Type Intrinsic Description
L LLE (S,S 2) True if S < or equal to S 2 in ASCII sequence.
L LLT (S,S 2 True if S precedes S 2 in ASCII collating sequence.
R LOG (R Z) Natural (base e) logarithm of real or complex number.
L LOGICAL (L [,K]) Convert L to logical of kind K.
R LOG10 (X) Common (base 10) logarithm function of real X.
N,L MATMUL (MATRIX,MATRIX 2) Conformable matrix multiplication.
N MAX (N,N 2 [,N 3,...]) Maximum value of two or more numbers same type.
I MAXEXPONENT (X) Maximum exponent for real numbers like X, e.g. 128.
I,V MAXLOC (N ARRAY [,M]) Location(s) of maximum ARRAY element, passing M.
N MAXVAL (N ARRAY [,D] [,M]) Maximum ARRAY term, in dimension D, passing M.
? MERGE (? TRUE,? FALSE,M) Use ? TRUE when M is true; ? FALSE otherwise.
N MIN (N,N 2 [,N 3,...]) Minimum value of two or more same type numbers.
I MINEXPONENT (X) Minimum exponent for real numbers like X, e.g. –125.
I,V MINLOC (N ARRAY [,M]) Location(s) of minimum ARRAY term, passing M.
N MINVAL (N ARRAY [,D] [,M]) Minimum ARRAY term, in dimension D, passing M.
N MOD (N,N 2) Remainder for N 2. That is, N–INT(N/N 2)?N 2.
N MODULO (N,N 2) Modulo, that is, N–FLOOR(N/N 2)?N 2.
call MVBITS (I FROM,I LOC, Copy I LEN bits at I LOC in I FROM to I TO
I LEN,I TO,I POS) at I POS.
R NEAREST (X,Y) Nearest number at X in the direction of sign Y.
I NINT (X [,K]) Integer nearest to real X, of the stated kind.
I NOT (I) Logical complement of the bits of integer I.
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY at true M into vector, using V PAD.
I PRECISION (R Z) Decimal precision for a real or complex R Z, e.g. 6.
L PRESENT (OPTIONAL) True if optional argument is present in call.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY elements, along D, for mask M.
I RADIX (N) Base of the model for numbers like N, e.g. 2.
call RANDOM NUMBER (X) Pseudo-random numbers in range 0 < X < 1.
call RANDOM SEED ([I SIZE] Initialize random number generator, defaults to
[,I V PUT][,I V GET]) processor initialization.
I RANGE (A) Decimal exponent range in the model for A, e.g. 37.
R REAL (A [,K]) Convert A to real type, of type K.
S REPEAT (S,I COPIES) Concatenates I COPIES of string S.
? RESHAPE (? ARAY,I V SHAP Reshape ARAY, using vector SHAP, pad from
[,? PAD] [,V ORDER]) an array, and re-order.
R RRSPACING (X) Relative spacing reciprocal of numbers near X.
R SCALE (X,I) Return X times b??I, for base of b = RADIX (X).
I SCAN (S,S SET [,L BACK]) Leftmost character index in S found in S SET; (right-
most).
I SELECTED INT KIND (I r) Integer kind with range, –(10??I r) to (10??I r).
I SELECTED REAL KIND Kind for real of decimal precision, I, and exponent
I ([I] [,I r]) range, I r.
R SET EXPONENT (X,I) Number with mantissa of X and exponent of I.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
N SIGN (N,N 2) Absolute value of N times sign of same type N 2.
R,Z SIN (R Z) Sine function of real or complex number.
R SINH (X) Hyperbolic sine function of real X.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
R SPACING (X) Absolute spacing of numbers near real X, e.g. 2??–17.
? SPREAD (? ARAY,D,I COPIES) I COPIES along dimension D of ARAY into an array
of rank 1 greater.
(continued)

c 2001 J.E. Akin 20
 Alphabetic Table of Fortran90 Intrinsic Functions (continued)
Type Intrinsic Description
R,Z SQRT (R Z) Square root function, of real or complex number.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY elements, along D, passing mask M.
call SYSTEM CLOCK ([I NOW] Integer data from real-time clock. CPU time is
[,I RATE] [,I MAX]) (finish now - start now) / rate.
R TAN (X) Tangent function of real X.
R TANH (X) Hyperbolic tangent function of real X.
R TINY (N) Smallest positive number, like N, e.g. 2??–126.
? TRANSFER (? ARAY, V MOLD Same representation as ARAY, but type of MOLD, in
[,I SIZE]) vector of length SIZE.
? TRANSPOSE (MATRIX) Matrix transpose of any type matrix.
S TRIM (S) Remove trailing blanks from a single string.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
? UNPACK (V,M,? USE) Unpack vector V at true elements of M, into USE.
I VERIFY (S,S SET [,L BACK]) First position in S not found in S SET (or last).

Subject Table of Fortran 90 Intrinsic Routines

The following KEY symbols are utilized to denote the TYPE of the intrinsic function, or subroutine, and
its arguments: A-complex, integer, or real; B-integer bit; C-character; D-dimension; I-integer; K-kind;
L-logical; M-mask (logical); N-integer, or real; P-pointer; R-real; S-string; T-target; V-vector (rank one
array); X-real; Y-real; Z-complex; and *-any type. For more detailed descriptions and example uses of
these intrinsic functions see Adams, J.C., et al., Fortran 90 Handbook, McGraw-Hill, New York, 1992,
ISBN 0–07–000406–4.

Type Intrinsic Description

ALLOCATION
L ALLOCATED (? ARRAY) True if the array is allocated.
ARGUMENT
L PRESENT (OPTIONAL) True if optional argument is present in the call.
ARRAY: CONSTRUCTION
? MERGE (? TRUE,? FALSE,M) Use ? TRUE if M is true; ? FALSE otherwise.
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from
V PAD.
? RESHAPE (? ARRAY,I V SHAPE Reshape ARRAY using vector SHAPE, pad from an
[,? PAD] [,V ORDER]) array, and re-order.
? SPREAD (? ARRAY,D,I COPIES) I COPIES along D of ARRAY to rank 1 greater
array.
? UNPACK (V,M,? USE) Unpack V at true elements of M, into USE.
ARRAY: DIMENSIONS
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s vector), along dimension D.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s vector), along dimension D.
ARRAY: INQUIRY
L ALL (M [,D]) True if all mask, M, elements are true, along D.
L ALLOCATED (? ARRAY) True if the array is allocated.
L ANY (M [,D]) True if any mask, M, element is true, along D.
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
(continued)

c 2001 J.E. Akin 21
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
ARRAY: LOCATION
I,V MAXLOC (N ARRAY [,M]) Location(s) of maximum ARRAY term, passing M.
I,V MINLOC (N ARRAY [,M]) Location(s) of minimum ARRAY term, passing M.
ARRAY: MANIPULATION
? CSHIFT (? ARRAY,I SHIFT [,D]) Circular shift out and in for I SHIFT elements.
? EOSHIFT (? ARRAY,I SHIFT End-off shift ARRAY, and fill, in dimension D.
[,? FIL][,D])
? TRANSPOSE (MATRIX) Matrix transpose of any type matrix.
ARRAY: MATHEMATICS
N,L DOT PRODUCT (V,V 2) Dot product of vectors V and V 2.
N,L MATMUL (MATRIX,MATRIX 2) Conformable matrix multiplication.
N MAXVAL (N ARRAY [,D] [,M]) Value of max ARRAY term, along D, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of min ARRAY term, along D, passing M.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY terms, along D, for mask M.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY terms, along D, passing mask M.
ARRAY: PACKING
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from
V PAD.
? UNPACK (V,M,? USE) Unpack V at true elements of M, into USE.
ARRAY: REDUCTION
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ANY (M [,D]) True if any mask, M, term is true, along D.
I COUNT (M [,D]) Number of true mask, M, terms, along dimension D.
N MAXVAL (N ARRAY [,D] [,M]) Value of max ARRAY term, along D, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of min ARRAY term, along D, passing M.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY terms, along D, for mask M.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY terms, along D, passing mask M.
BACK SCAN
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (or right).
I SCAN (S,S SET [,L BACK]) Left character index in S also in S SET (or right).
I VERIFY (S,S SET [,L BACK]) First position in S not belonging to S SET (or last).
BIT: INQUIRY
I BIT SIZE (I) Max number of bits possible in integer I, e.g. 32.
BIT: MANIPULATION
L BTEST (I,I POS) True if bit location I POS of integer I has value one.
B IAND (I,I 2) Logical AND on the bits of I and I 2.
B IBCLR (I,I POS) Clear bit I POS to zero in integer I.
B IBITS (I,I POS,I LEN) Extract I LEN bits at I POS in integer I.
B IBSET (I,I POS) Set bit I POS to one in integer I.
B IEOR (I,I 2) Exclusive OR on the bits of I and I 2.
B IOR (I,I 2) Inclusive OR on the bits of I and I 2.
B ISHFT (I,I SHIFT) Logical shift of bits of I by I SHIFT, pad with 0.
B ISHFTC (I,I SHIFT [,I SIZE]) Logical circular shift of I SIZE rightmost bits of I.
call MVBITS (I GET, I LOC, I,
I TO,I POS) Copy I bits at I LOC in I GET to I TO at I POS.
I NOT (I) Logical complement of the bits of integer I.
? TRANSFER (? ARRAY,
(continued)

c 2001 J.E. Akin 22
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
V MOLD [,I SIZE]) Same representation as ARRAY, but type of MOLD.
BOUNDS
I CEILING (X) Least integer greater than or equal to real X.
I FLOOR (X) Greatest integer less than or equal to X.
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
N MAX (N,N 2 [,N 3,...]) Maximum value of two or more numbers same type.
N MAXVAL (N ARRAY [,D] [,M]) Value of max ARRAY term, along D, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of min ARRAY term, along D, passing M.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
CALLS
call MVBITS (I GET,I LOC,I, Copy I bits at I LOC in I GET to I TO at I POS.
I TO,I POS)
call DATE AND TIME ([S DATE] Real-time clock data.
[,S TIME] [,S ZONE]
[,I V VALUES])
call RANDOM NUMBER (X) Pseudo-random numbers in range 0 < X < 1.
call RANDOM SEED ([I SIZE] Initialize random number generator.
[,I V P] [,I V G])
call SYSTEM CLOCK ([I NOW] Integer data from real-time clock.
[,I RAT] [,I MX])
CHARACTERS
C ACHAR (I) Character in position I of ASCII collating sequence.
C CHAR (I [,K]) Character in position I of processor collation.
I IACHAR (C) Position of character C in ASCII collating sequence.
I ICHAR (C) Position of character C in processor collation.
CLOCK
call SYSTEM CLOCK ([I NOW] Integer data from real-time clock.
[,I RAT] [,I MX])
COMBINING
? MERGE (? TRUE,? FALSE,M) Use ? TRUE term if M is true or ? FALSE
otherwise.
COMPLEX
R AIMAG (Z) Imaginary part of complex number.
Z CMPLX (X [,Y][,K]) Convert real(s) to complex type, of given kind.
Z CONJG (Z) Conjugate of complex number Z.
R COS (R Z) Cosine function of real or complex argument.
R,Z EXP (R Z) Exponential function of real or complex argument.
R LOG (R Z) Natural (base e) logarithm of real or complex num-
ber.
I PRECISION (R Z) Decimal precision of real or complex value, e.g. 6.
R,Z SIN (R Z) Sine function of real or complex number.
R,Z SQRT (R Z) Square root function, of real or complex number.
CONVERSIONS
R AIMAG (Z) Imaginary part of complex number.
R AINT (X [,K]) Truncate X to a real whole number.
Z CMPLX (X [,Y][,K]) Convert real (s) to complex type, of given kind.
R DBLE (A) Convert A to double precision real.
R DPROD (X,Y) Double precision product of two default real scalars.
(continued)

c 2001 J.E. Akin 23
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
I INT (A [,K]) Convert A to integer type, of given kind.
L LOGICAL (L [,K]) Convert L to logical of kind K.
I NINT (X [,K]) Integer nearest to real X, of the stated kind.
R REAL (A [,K]) Convert A to real type, of type K.
N SIGN (N,N 2) Absolute value of N times sign of same type N 2.
? TRANSFER (? ARRAY, Same representation as ARRAY, but type of MOLD.
V MOLD [,I SIZ])
COPIES
? MERGE (? TRUE,? FALSE,M) Use ? TRUE if M is true or ? FALSE otherwise.
call MVBITS (I FROM,I LOC, I, Copy I bits at I LOC in I FROM to I TO at
I TO,I POS) I POS.
S REPEAT (S,I COPIES) Concatenates I COPIES of string S.
? SPREAD (? ARRAY,D,I COPIES) I COPIES along D of ARRAY to rank 1 greater
array.
COUNTING
I COUNT (M [,D]) Number of true mask, M, terms, along dimension D.
DATE
call DATE AND TIME ([S DATE] Real-time clock data.
[,S TIME] [,S ZONE]
[,I V VALUES])
DIMENSION OPTIONAL ARGUMENT
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ANY (M [,D]) True if any mask, M, term is true, along D.
I COUNT (M [,D]) Number of true mask, M, terms, along dimension D.
? CSHIFT (? ARRAY,I SHIFT [,D]) Perform circular shift out and in for I SHIFT terms.
? EOSHIFT (? ARRAY, Perform end-off shift, and fill, in dimension D.
I SHIFT [,? FIL][,D])
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
N MAXVAL (N ARRAY [,D] [,M]) Value of max ARRAY term, along D, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of min ARRAY term, along D, passing M.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY terms, along D, for mask M.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY terms, along D, passing mask M.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
DIMENSIONS
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
DOUBLE PRECISION (see SELECTED REAL KIND)
R DBLE (A) Convert A to double precision real.
R DPROD (X,Y) Double precision product of two default real scalars.
EXISTENCE
L ALLOCATED (? ARRAY) True if the array is allocated.
L ASSOCIATED (P [,T]) True if pointer, P, is associated with any target, or T.
L PRESENT (OPTIONAL) True if optional argument is present in call.
FILE

(continued)

c 2001 J.E. Akin 24
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
FILL IN
? EOSHIFT (? ARRAY,I SHIFT
[,? FIL][,D]) End-off shift ARRAY, and fill, in dimension D.
INQUIRY: ARRAY
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ALLOCATED (? ARRAY) True if the array is allocated.
L ANY (M [,D]) True if any mask, M, term is true, along D.
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along dimension D.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.
INQUIRY: BIT
I BIT SIZE (I) Max number of bits possible in integer I, e.g. 32.
INQUIRY: CHARACTER
I LEN (S) Total character string length.
I LEN TRIM (S) Length of S without trailing blanks.
INQUIRY: NUMBER MODEL
N DIGITS (N) Number of significant digits in number N, e.g. 31.
R EPSILON (X) Number  1, for numbers like X, e.g. 2??–23.
N HUGE (N) Largest number for numbers like N, e.g. 2??128.
I MAXEXPONENT (X) Max exponent for real numbers like X, e.g. 128.
I MINEXPONENT (X) Min exponent for real numbers like X, e.g. –125.
I PRECISION (R Z) Decimal precision for real or complex value, e.g. 6.
I RADIX (N) Base of the model for numbers like N, e.g. 2.
I RANGE (A) Decimal exponent range for A, e.g. 37.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
I SIZE (? ARRAY [,D]) ARRAY size, along dimension D.
R TINY (N) Smallest positive number, like N, e.g. 2??–126.
INQUIRY: MISCELLANEOUS
I COUNT (M [,D]) Number of true mask, M, elements, along D.
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (or right).
I SCAN (S,S SET [,L BACK]) Left character index in S also in S SET; (or right).
I VERIFY (S,S SET [,L BACK]) First position in S not belonging to S SET, (or last).
INTEGERS
I CEILING (X) Least integer greater than or equal to real X.
I FLOOR (X) Greatest integer less than or equal to X.
I MAX1 (X,X2 [,X3]) Maximum integer from list of reals
I MIN1 (X,X2 [,X3]) Minimum integer from list of reals
N MODULO (N,N 2) Modulo, N-FLOOR(N/N 2)?N 2.
I SELECTED INT KIND (I r) Integer with exponent, –(10??I r) to (10??I r).
KIND: INQUIRY
I KIND (ANY) Kind type integer parameter value for any argument.
KIND: DEFINITION
I SELECTED INT KIND (I r) Integer with exponent, –(10??I r) to (10??I r).
I SELECTED REAL KIND ([I] Real with precision, I, and exponent range, I r.
[,I r])
KIND: USE OPTION
(continued)

c 2001 J.E. Akin 25
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
R AINT (X [,K]) Truncate X to a real whole number.
R ANINT (X [,K]) Real whole number nearest to X.
C CHAR (I [,K]) Character in position I of processor collation.
Z CMPLX (X [,Y][,K]) Convert real(s) to complex type, of given kind.
I INT (A [,K]) Convert A to integer type, of given kind.
L LOGICAL (L [,K]) Convert L to logical of kind K.
I NINT (X [,K]) Integer nearest to real X, of the stated kind.
R REAL (A [,K]) Convert A to real type, of type K.
LOCATION
I IACHAR (C) Position of character C in ASCII collating sequence.
I ICHAR (C) Position of character C in processor collation.
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (or right).
I,V MAXLOC (N ARRAY [,M]) Vector location(s) of ARRAY maximum, passing M.
I,V MINLOC (N ARRAY [,M]) Vector location(s) of ARRAY minimum, passing M.
I SCAN (S,S SET [,L BACK]) Left character index in S found in S SET; (or right).
LOGICAL
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ALLOCATED (? ARRAY) True if the array is allocated.
L ANY (M [,D]) True if any mask, M, term is true, along D.
L ASSOCIATED (P [,T]) True if pointer, P, is associated with any target, or T.
L BTEST (I,I POS) True if bit location I POS of integer I has value one.
N,L DOT PRODUCT (V,V 2) Dot product of vectors V and V 2.
B IAND (I,I 2) Logical AND on the bits of I and I 2.
B IEOR (I,I 2) Exclusive OR on the bits of I and I 2.
B IOR (I,I 2) Inclusive OR on the bits of I and I 2.
B ISHFT (I,I SHIFT) Logical shift of bits of I by I SHIFT, pad with 0.
L LGE (S,S 2) True if S is  S 2 in ASCII collating sequence.
L LGT (S,S 2) True if S follows S 2 in ASCII collating sequence.
L LLE (S,S 2) True if S is  to S 2 in ASCII collating sequence.
L LLT (S,S 2) True if S precedes S 2 in ASCII collating sequence.
N,L MATMUL (MATRIX,MATRIX 2) Conformable matrix multiplication.
L LOGICAL (L [,K]) Convert L to logical of kind K.
I NOT (I) Logical complement of the bits of integer I.
L PRESENT (OPTIONAL) True if optional argument is present in call.
MASK, or MASK OPTIONAL ARGUMENT
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ANY (M [,D]) True if any mask, M, term is true, along D.
I COUNT (M [,D]) Number of true mask, M, terms, along dimension D.
I,V MAXLOC (N ARRAY [,M]) Vector of location(s) of ARRAY max’s, passing M.
N MAXVAL (N ARRAY [,D] [,M]) Value of ARRAY maximum, along D, passing M.
? MERGE (? TRUE,? FALSE,M) Use ? TRUE if M is true or ? FALSE otherwise.
I,V MINLOC (N ARRAY [,M]) Vector location(s) of ARRAY minimum, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of ARRAY minimum, along D, passing M.
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from
V PAD.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY terms, along D, for mask M.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY terms, along D, passing mask M.
MATHEMATICAL FUNCTIONS
R ACOS (X) Arc cosine (inverse cosine) function of real X.
(continued)

c 2001 J.E. Akin 26
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
R ASIN (X) Arcsine (inverse sine) function of real X.
R ATAN (X) Arctangent (inverse tangent) function of real X.
R ATAN2 (Y,X) Arctangent for argument of complex number (X, Y).
R COS (R Z) Cosine function of real or complex argument.
R COSH (X) Hyperbolic cosine function of real X.
R,Z EXP (R Z) Exponential function of real or complex argument.
R LOG (R Z) Natural logarithm of real or complex number.
R LOG10 (X) Common (base 10) logarithm function of real X.
R,Z SIN (R Z) Sine function of real or complex number.
R SINH (X) Hyperbolic sine function of real X.
R TAN (X) Tangent function of real X.
R TANH (X) Hyperbolic tangent function of real X.
MATRICES (See ARRAYS)
N,L DOT PRODUCT (V,V 2) Dot product of vectors V and V 2.
N,L MATMUL (MATRIX,MATRIX 2) Conformable matrix multiplication.
? TRANSPOSE (MATRIX) Matrix transpose of any type matrix.
NUMBER MODEL
N DIGITS (N) Number of significant digits for N, e.g. 31.
R EPSILON (X) Number  1, for numbers like X, e.g. 2??–23.
I EXPONENT (X) Exponent part of the model for real X.
R FRACTION (X) Fractional part of the model for real X.
N HUGE (N) Largest number for numbers like N, e.g. 2??128.
R NEAREST (X,Y) Nearest number at X in the direction of sign Y.
I RADIX (N) Base of the model for numbers like N, e.g. 2.
I RANGE (A) Decimal exponent range for A, e.g. 37.
R RRSPACING (X) Reciprocal of relative spacing of numbers near X.
R SCALE (X,I) Return X times b??I, where base b = RADIX (X).
R SET EXPONENT (X,I) Real with mantissa part of X and exponent part of I.
R SPACING (X) Absolute spacing of numbers near X, e.g. 2??-17.
R TINY (N) Smallest positive number, like N, e.g. 2??–126.
NUMERIC FUNCTIONS
A ABS (A) Absolute value of A.
R AIMAG (Z) Imaginary part of complex number.
R ANINT (X [,K]) Real whole number nearest to X.
I CEILING (X) Least integer greater than or equal to real X.
Z CMPLX (X [,Y][,K]) Convert real(s) to complex type, of given kind.
Z CONJG (Z) Conjugate of complex number Z.
R DBLE (A) Convert A to double precision real.
R DPROD (X,Y) Double precision real product of two real scalars.
I FLOOR (X) Greatest integer less than or equal to X.
I INT (A [,K]) Convert A to integer type, of given kind.
N MAX (N,N 2 [,N 3,...]) Maximum value of two or more numbers same type.
N MIN (N,N 2 [,N 3,...]) Minimum value of two or more same type numbers.
N MOD (N,N 2) Remainder for N 2, i.e., N-INT(N/N 2)?N 2.
N MODULO (N,N 2) Modulo, N-FLOOR(N/N 2)?N 2.
R REAL (A [,K]) Convert A to real type, of type K.
N SIGN (N,N 2) Absolute value of N times sign of same type N 2.
PADDING
B ISHFT (I,I SHIFT) Logical shift of bits of I by I SHIFT, pad with 0.
(continued)

c 2001 J.E. Akin 27
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from V PAD.
? RESHAPE (? ARRAY,I V SHAPE
[,? PAD] [,V ORDER]) Reshape ARRAY to vector SHAPE, pad, re-order.
POINTER
L ASSOCIATED (P [,T]) True if pointer, P, is associated with any target, or T.
PRESENCE
L PRESENT (OPTIONAL) True if optional argument is present in call.
RANDOM NUMBER
call RANDOM NUMBER (X) Pseudo-random numbers in range 0 < X < 1.
call RANDOM SEED ([I SIZE]
[,I V P][,I V G]) Initialize random number generator.
REALS
R AINT (X [,K]) Truncate X to a real whole number.
R ANINT (X [,K]) Real whole number nearest to X.
R AMAX0 (I,I2 [,I3]) Maximum real from list of integers.
R AMIN0 (I,I2 [,I3]) Minimum real from list of integers.
R REAL (A [,K]) Convert A to real type, of type K.
I SELECTED REAL KIND ([I] Real with precision, I, and exponent range, I r.
[,I r])
REDUCTION
L ALL (M [,D]) True if all mask, M, terms are true, along D.
L ANY (M [,D]) True if any mask, M, term is true, along D.
I COUNT (M [,D]) Number of true mask, M, terms, along dimension D.
N MAXVAL (N ARRAY [,D] [,M]) Value of max ARRAY term, along D, passing M.
N MINVAL (N ARRAY [,D] [,M]) Value of min ARRAY term, along D, passing M.
A PRODUCT (A ARRAY [,D] [,M]) Product of ARRAY terms, along D, for mask M.
A SUM (A ARRAY [,D] [,M]) Sum of ARRAY terms, along D, passing mask M.
RESHAPING ARRAYS
? CSHIFT (? ARRAY,I SHIFT [,D]) Perform circular shift out and in for I SHIFT terms.
? EOSHIFT (? ARRAY,I SHFT
[,? FIL] [,D]) End-off shift ARRAY, and fill, in dimension D.
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from
V PAD.
? RESHAPE (? ARRAY,I V SHAPE
[,? PAD] [,V ORDER]) Reshape ARRAY to vector SHAPE, pad, re-order.
? UNPACK (V,M,? USE) Unpack V for true elements of M, into USE.
REVERSE ORDER
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (right-
most).
I SCAN (S,S SET [,L BACK]) Left character index in S found in S SET; (right-
most).
I VERIFY (S,S SET [,L BACK]) First position in S not found in S SET, (or last).
SHIFTS
? CSHIFT (? ARRAY,I SHIFT [,D]) Perform circular shift out and in for I SHIFT terms.
? EOSHIFT (? ARRAY,I SHIFT
[,? FILL][,D]) Perform end-off shift, and fill, in dimension D.
B ISHFT (I,I SHIFT) Logical shift of bits of I by I SHIFT, pad with 0.
B ISHFTC (I,I SHIFT [,I SIZE]) Logical circular shift of I SIZE rightmost bits of I.
(continued)

c 2001 J.E. Akin 28
 Subject Table of Fortran 90 Intrinsic Functions (continued)
Type Intrinsic Description
STRING
C ADJUSTL (S) Adjust S left, move leading blanks to trailing blanks.
C ADJUSTR (S) Adjust S right, move trailing to leading blanks.
I INDEX (S,S SUB [,L BACK]) Left starting position of S SUB within S (or right).
I LEN (S) Total character string length.
I LEN TRIM (S) Length of S without trailing blanks.
L LGE (S,S 2) True if S is  to S 2 in ASCII collating sequence.
L LGT (S,S 2) True if S follows S 2 in ASCII collating sequence.
L LLE (S,S 2) True if S is  to S 2 in ASCII collating sequence.
L LLT (S,S 2) True if S precedes S 2 in ASCII collating sequence.
S REPEAT (S,I COPIES) Concatenates I COPIES of string S.
I SCAN (S,S SET [,L BACK]) Left character index in S found in S SET; (or right).
S TRIM (S) Remove trailing blanks from a single string.
I VERIFY (S,S SET [,L BACK]) First position in S not found in S SET, (or last).
TARGET
L ASSOCIATED (P [,T]) True if pointer, P, is associated with any target, or T.
TIME
call DATE AND TIME ([S DATE] Real-time clock data.
[,S TIME] [,S ZONE]
[,I V VALUES])
call SYSTEM CLOCK ([I NOW] Integer data from real-time clock.
[,I RAT] [,I MX])
VECTOR (See ARRAYS)
N,L DOT PRODUCT (V,V 2) Dot product of vectors V and V 2.
I,V LBOUND (? ARRAY [,D]) ARRAY lower bound(s) vector, along D.
I,V MAXLOC (N ARRAY [,M]) Location(s) of maximum ARRAY term, passing M.
I,V MINLOC (N ARRAY [,M]) Location(s) of minimum ARRAY term, passing M.
?,V PACK (? ARRAY,M [,V PAD]) Pack ARRAY for true M into vector, pad from
V PAD.
? RESHAPE (? ARRAY,I V SHAPE
[,? PAD] [,V ORDER]) Reshape ARRAY to vector SHAPE, pad, re-order.
I,V SHAPE (? ARRAY) ARRAY (or scalar) shape vector.
? TRANSFER (? ARRAY, V MOLD
[,I SIZE]) Same representation as ARRAY, but type of MOLD.
I,V UBOUND (? ARRAY [,D]) ARRAY upper bound(s) vector, along dimension D.

B.3 Syntax of Fortran 90 Statements

The following is a list of the recommended Fortran90 statements. Additional statements are allowed, but
have been declared obsolete, and are expected to be deleted in future standards. Thus, they should not be
utilized in new programs. They are appended to the end of this list. Below we list the standard syntax for
the Fortran90 statements. In some cases the most common simple form of a statement is shown before
it’s more general options. Such optional features are shown included in brackets, [ ], and a vertical
bar j means “or.” Note that the new attribute terminator symbol :: is always optional, but its use is
recommended.
The following abbreviations are employed: arg=argument, attr=attribute, exp=expression,
i =integer, r =real, s =string, spec=specifier, and here [type] means CHARACTER j COMPLEX j
INTEGER j LOGICAL j REAL, or a user defined name given in a TYPE statement. Recall that F90
allows variable names to be 31 characters long and they may include an underscore (but F77 allows only
6 characters and no underscore). F90 lines may contain up to 132 characters (but just 72 in F77). All

c 2001 J.E. Akin 29
 M ATLAB C++ F90
a
Pre-allocate A(100)=0 int A[100]; integer A(100)
linear array
Initialize to a for j=1:100 % slow for (j=0; j<100; j++) A=12
A(j)=12
constant value of end A[j]=12;
12 % better way
A=12*ones(1,100)
Pre-allocate A=ones(10,10) int A[10][10]; integer A(10,10)
two-dimensional
array
a C++ has a starting subscript of 0, but the argument in the allocation statement is the array’s size.

Table B.41: Array initialization constructs.

Action M ATLAB C++ F90

a
Define A=zeros(2,3) int A[2][3]; integer,dimension(2,3)::A
size
Enter A=[1,7,-2; int A[2][3]= f A(1,:)=(/1,7,-2/)
rows f
1,7,2 , g
3, 4, 6];
f
3,4,6 g A(2,:)=(/3,4,6/)

g;

a Optional in M ATLAB, but improves efficiency.

Table B.42: Array initialization constructs.

standard F77 statements are a sub-set of F90. Attribute options, and their specifiers, for each statement
are given in the companion table ”Fortran 90 Attributes and Specifiers”. The numerous options for the
INQUIRE statement are given in the table entitled “Options for F90 INQUIRE.”
In addition to the statements given below F90 offers intrinsic array operations, implied do loops,
vector subscripts, and about 160 intrinsic functions. Those functions, with their arguments, are given
in tables “Alphabetical Table of Fortran 90 Intrinsic Functions and Subroutines,” and “Subject Table of
Fortran 90 Intrinsic Functions and Subroutines.”

F90 Syntax
! preceeds a comment in F90
in column one denotes a comment line in F77
& continues a line in F90 (must be in column 6 for F77)
; terminates a statement in F90 (allows multiple statements per line)
variable = expression or statement ! is an assignment (column 7 in F77)
ALLOCATABLE [::] array name[(extents)] [, array name[(extents)]]
ALLOCATE (array name)
ALLOCATE (array name [, STAT=status] [,array name [, STAT=status]])
BACKSPACE i exp ! file unit number
BACKSPACE ([UNIT=]i value [, IOSTAT=i variable] [, ERR=i label])
C in column one denotes a comment line in F77
CALL subroutine name [([args])]
CASE (range list) [select name] ! purpose
CASE DEFAULT [select name] ! purpose
CHARACTER LEN=i value [::] s list
CHARACTER [(LEN=i value j * [, KIND=]i kind)] [[, attr list] ::] s list
CHARACTER [(i value j *, [KIND=]i kind)] [[, attr list] ::] s list
(continued)

c 2001 J.E. Akin 30
 F90 Syntax (continued)
CHARACTER [([KIND=i kind] [, LEN=i value j *])] [[, attr list] ::] s list
CLOSE (i value) ! unit number
CLOSE ([UNIT=]i value [, ERR=i label] [, IOSTAT=i variable] [, STATUS=exp])
COMPLEX [::] variable list
COMPLEX [([KIND=]i kind)] [[, attr list] ::] variable list
CONTAINS ! internal definitions follow
CYCLE ! current do only for a purpose
CYCLE [nested do name] ! and terminate its sub do’s for a purpose
DEALLOCATE (array name)
DEALLOCATE (array name [, STAT=status] [, array name [, STAT=status]])
DIMENSION array name(extents) [, array name(extents)]
DO ! forever
DO i variable = i start, i stop ! loop name or purpose
DO [i variable = i start, i stop [, i inc]] ! loop name or purpose
DO [i label,] [i variable = i start, i stop [, i inc]] ! loop name
[loop name:] DO [i variable = i start, i stop [, i inc]] ! purpose
[loop name:] DO [i label,] [i variable = i start, i stop [, i inc]]
DO WHILE (logical expression) ! obsolete, use DO-EXIT pair
DO [i label,] WHILE (logical expression) ! obsolete-obsolete
[ name:] DO [i label,] WHILE (logical expression) ! obsolete
ELSE [if name]
ELSE IF (logical expression) THEN [if name]
ELSE WHERE (logical expression)
END [name] ! purpose
END DO [do name] ! purpose
END FUNCTION [function name] ! purpose
END IF [if name] ! purpose
END INTERFACE ! purpose
END MODULE [module name] ! purpose
END PROGRAM [program name] ! purpose
END SELECT [select name] ! purpose
END SUBROUTINE [name] ! purpose
END TYPE [type name] ! purpose
END WHERE ! purpose
ENDFILE i exp ! for file unit number
ENDFILE ([UNIT=]i value [, IOSTAT=i variable] [, ERR=i label])
ENTRY entry name [([args])] [RESULT(variable name)]
EXIT ! current do only for a purpose
EXIT [nested do name] ! and its sub do’s for a purpose
EXTERNAL program list
i label FORMAT (specification and edit list)
FUNCTION name ([args]) ! purpose
FUNCTION name ([args]) [RESULT(variable name)] ! purpose
[type] [RECURSIVE] FUNCTION name ([args]) [RESULT(variable name)]
[RECURSIVE] [type] FUNCTION name ([args]) [RESULT(variable name)]
GO TO i label ! for a reason
IF (logical expression) executable statement
[name:] IF (logical expression) THEN ! state purpose
IMPLICIT type (letter list) ! F77 (a-h,o-z) real, (i-n) integer
IMPLICIT NONE ! F90 recommended default
INCLUDE source file path name ! purpose
INQUIRE ([FILE=]’name string’ [, see INQUIRE table]) ! re file
(continued)

c 2001 J.E. Akin 31
 F90 Syntax (continued)
INQUIRE ([NAME=]s variable [, see INQUIRE table]) ! re file
INQUIRE (IOLENGTH=i variable [, see INQUIRE table]) ! re output
INQUIRE ([UNIT=]i value [, see INQUIRE table]) ! re unit
INTEGER [::] variable list
INTEGER [([KIND=]i kind)] [[, attr list] ::] variable list
INTENT ([IN j INOUT j OUT]) argument list
INTERFACE ASSIGNMENT (+ j - j * j / j = j **) ! user extension
INTERFACE OPERATOR (.operator.) ! user defined
INTERFACE [interface name]
INTRINSIC function list
LOGICAL [::] variable list
LOGICAL [([KIND=]i kind)] [[, attr list] ::] variable list
MODULE PROCEDURE program list
MODULE module name ! purpose
NULLIFY (pointer list)
OPEN (i value) ! unit number
OPEN ([UNIT=]i value [, ERR=i label] [, IOSTAT=i variable] [, other spec])
OPTIONAL [::] argument list
PARAMETER (variable=value [, variable=value])
POINTER [::] name[(extent)] [, name[(extent)]] ! purpose
PRINT * , output list ! default free format
PRINT * , (io implied do) ! default free format
PRINT ’(formats)’ , output list ! formatted
PRINT ’(formats)’ , (io implied do) ! formatted
PRIVATE [[::] module variable list] ! limit access
PROGRAM [program name] ! purpose
PUBLIC [[::] module variable list] ! default access
READ * , input list ! default free format
READ * , (io implied do) ! default free format
READ ’(formats)’, input list ! formatted
READ ’(formats)’, (io implied do) ! formatted
READ ([UNIT=]i value, [FMT=]i label [, io spec list]), input list ! formatted
READ ([UNIT=]i value, s variable [, io spec list]), input list ! formatted
READ ([UNIT=]i value, ’(formats)’ [, io spec list]), input list ! formatted
READ (i value), input list ! binary read
READ ([UNIT=]i value, [, io spec list]), input list ! binary read
READ (s variable, [FMT=]i label), input list ! internal file type change
READ ([UNIT=]s variable, [FMT=]i label [, io spec list]), input list ! internal file change
REAL [::] variable list
REAL [([KIND=]i kind)] [[, attr list] ::] variable list
RECURSIVE FUNCTION name ([args]) [RESULT(variable name)] ! purpose
[ type] RECURSIVE FUNCTION name ([args]) [RESULT(variable name)] ! purpose
RECURSIVE SUBROUTINE name [([args])] ! purpose
RETURN ! from subroutine name
REWIND i exp ! file unit number
REWIND ([UNIT=]i value [, IOSTAT=i variable] [, ERR=i label])
SAVE [[::] variable list]
[name:] SELECT CASE (value)
SEQUENCE
STOP [’stop message string’]
SUBROUTINE name [([args])] ! purpose
SUBROUTINE name [([args])] [args, optional args] ! purpose
(continued)

c 2001 J.E. Akin 32
 F90 Syntax (continued)
[RECURSIVE] SUBROUTINE name [([args])] ! purpose
TARGET [::] name[(extent)] [, name[(extent)]]
TYPE (type name) [[, attr list] ::] variable list
TYPE [, PRIVATE j PUBLIC] name
USE module name [, ONLY: list in module name] ! purpose
USE module name [, new var or sub=>old name] ! purpose
WHERE (logical array expression) ! then
WHERE (logical array expression) array variable = array expression
WRITE * , output list ! default free format
WRITE * , (io implied do) ! default free format
WRITE ’(formats)’, output list ! formatted write
WRITE ’(formats)’, (io implied do) ! formatted write
WRITE ([UNIT=]i value, [FMT=]i label [, io spec list]), output list ! formatted write
WRITE ([UNIT=]i value, s variable [, io spec list]), output list ! formatted write
WRITE ([UNIT=]i value, ’(formats)’ [, io spec list]), output list ! formatted write
WRITE (i value), output list ! binary write
WRITE (i value), (io implied do) ! binary write
WRITE ([UNIT=]i value, [, io spec list]), output list ! binary write
WRITE (s variable, [FMT=]i label), output list ! internal file type change
WRITE ([UNIT=]s variable, [FMT=]i label [, io spec list]), output list ! internal file change

Obsolescent statements are those from Fortran77 that are redundant and for which better methods are
available in both Fortran77 and Fortran90.

Obsolete Syntax
ASSIGN i label TO i variable
BLOCK DATA [block data name]
COMMON [/common block name/] r variable list, i variable list
[i label] CONTINUE ! from do [do name]
DATA variable list / value list /
DATA (array implied do) / value list /
DOUBLE PRECISION [[, attr list] :: ] variable list
DO [i label,] [r variable = r start, r stop [, r inc]] ! real control
DO CONTINUE pair
[name:] DO [i label,] WHILE (logical expression) ! obsolete
END BLOCK DATA [block data name]
EQUIVALENCE (variable 1, variable 2) [ , (variable 3, variable 4)]
GO TO (i label 1,i label 2,...,i label n)[ , ] i variable
IF (arithmetic exp) i label neg, i label zero, i label pos
NAMELIST /group name/ variable list
PAUSE ! for human action
RETURN alternates
statement function (args) = expression

The attributes lists for the type declarations, e.g. REAL, are ALLOCATABLE, DIMENSION, INTENT,
OPTIONAL, KIND, POINTER, PARAMETER, PRIVATE, PUBLIC, SAVE, and TARGET; those for OPEN and
CLOSE are ACCESS, ACTION, BLANK, and DELIM; while those for READ and WRITE are ADVANCE, END,
EOR, ERR, and FMT.

c 2001 J.E. Akin 33
 M ATLAB C++ F90
Addition
C=A+B C=A+B for (i=0; i<10; i++) f C=A+B
for (j=0; j<10; j++) f
g
C[i][j]=A[i][j]+B[i][j];

g
Multiplication
f
C = AB C=A*B for (i=0; i<10; i++)
for (j=0; j<10; j++) f
C=matmul(A,B)

f
C[i][j] = 0;
for (k=0; k<10; k++)
g
C[i][j] += A[i][k]*B[k][j];

g
g
Scalar
multiplication C=a*B for (i=0; i<10; i++) f C=a*B
C= B a
for (j=0; j < 10; j++) f
g
C[i][j] = a*B[i][j];

g
Matrix
inverse
B=A B=inv(A)a
B=inv(A) a
1

a Neither C++ nor F90 have matrix inverse functions as part of their language definitions nor as part of standard collections

of mathematical functions (like those listed in Table 4.7). Instead, a special function, usually drawn from a library of numerical
functions, or a user defined operation, must be used.

Table B.43: Elementary matrix computational routines.

C++ int* point, vector, matrix

...
point = new type tag

f g
vector = new type tag [space 1]
if (vector == 0) error process
matrix = new type tag [space 1 * space 2]
...
delete matrix
...
delete vector
delete point

F90 type tag, pointer, allocatable :: point

type tag, allocatable :: vector (:), matrix (:,:)
...
allocate (point)
allocate (vector (space 1), STAT = my int)
if (my int /= 0) error process
allocate (matrix (space 1, space 2))
...
deallocate (matrix)
if (associated (point, target name)) pointer action...
if (allocated (matrix)) matrix action...
...
deallocate (vector)
deallocate (point)

Table B.44: Dynamic allocation of arrays and pointers.

c 2001 J.E. Akin 34
 SUBROUTINE AUTO ARRAYS (M,N, OTHER)
USE GLOBAL CONSTANTS ! FOR INTEGER K
IMPLICIT NONE
INTEGER, INTENT (IN) :: M,N
type tag, INTENT (OUT) :: OTHER (M,N) ! dummy array
! Automatic array allocations
type tag :: FROM USE (K)
type tag :: FROM ARG (M)
type tag :: FROM MIX (K,N)
...
! Automatic deallocation at end of scope
END SUBROUTINE AUTO ARRAYS

Table B.45: Automatic memory management of local scope arrays.

module derived class name
use base class name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Table B.46: F90 Single Inheritance Form.

module derived class name

use base class name, only: list of entities
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Table B.47: F90 Selective Single Inheritance Form.

module derived class name

use base class name, local name => base entity name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Table B.48: F90 Single Inheritance Form, with Local Renaming.

c 2001 J.E. Akin 35
 module derived class name
use base1 class name
use base2 class name
use base3 class name, only: list of entities
use base4 class name, local name => base entity name
! new attribute declarations, if any
...
contains

! new member definitions

...
end module derived class name

Table B.49: F90 Multiple Selective Inheritance with Renaming.

c 2001 J.E. Akin 36
 Examples of F90 Statements
The following is a list of examples of the recommended Fortran90 statements. Some have been
declared obsolete, and are expected to be deleted in future standards. Thus, they should not be utilized
in new programs. They are noted in the comments. In some cases the most common simple form
of a statement is shown along with it’s more general options. Note that the new attribute terminator
symbol :: is always optional, but its use is recommended. While Fortran is not case-sensitive, this
table employs upper case letters to denote standard features, and lower case letters for user supplied
information. The following abbreviations are employed: arg=argument, attr=attribute, exp=expression,
i =integer, l =logical, r =real, s =string, spec=specifier, z =complex.
Recall that F90 allows variable names to be 31 characters long and they may include an underscore
(but F77 allows only six characters and no underscore). F90 lines may contain up to 132 characters (but
just 72 in F77). All standard F77 statements are a sub-set of F90.
The attributes lists for the type declariations, e.g. REAL, are ALLOCATABLE, DIMENSION, INTENT,
OPTIONAL, KIND, POINTER, PARAMETER, PRIVATE, PUBLIC, SAVE, and TARGET. Those op-
tional attributes for OPEN are ACCESS = [DIRECT, SEQUENTIAL], ACTION = [READ, READWRITE,
WRITE], BLANK = [NULL, ZERO], DELIM = [APOSTROPHE, NONE, QUOTE], ERR = i label, FILE =
s name, FORM = [FORMATTED, UNFORMATTED], IOSTAT = i var, PAD = [NO, YES], POSITION =
[APPEND, ASIS, REWIND], RECL = i len, STATUS = [NEW, OLD, REPLACE, SEARCH, UNKNOWN],
and UNIT = i unit; while CLOSE utilizes only ERR, IOSTAT, STATUS, and UNIT.
The io spec list options for READ and WRITE are ADVANCE = [NO, YES], END = i label, EOR =
i label, ERR = i label, FMT = [*, i label, s var], IOSTAT = i var, NML = var list, REC = i exp, SIZE
= i size, and UNIT = i unit.

Fortran Statement Examples

Name Examples Comments
Allocatable ALLOCATABLE :: force, stiffness By name
ALLOCATABLE :: force(:), stiffness(:,:) Ranks
Allocate ALLOCATE (hyper matrix(5, 10, 3))
ALLOCATE (force(m))
ALLOCATE (array name(3, 3, 3, 3), STAT=i err) Error status
Assign ASSIGN 9 TO k Obsolete
Assignment c = ’b’ Character
s = "abc" String
s = c // ’abc’ Concatenation
s = string(j:m) Sub-string
s fmt = ’(2F5.1)’ Stored format
l = l 1 .OR. l 2 Logical
l = m < = 80
poor = (final > = 60) .AND. (final < 70)
proceed = .TRUE.
n=n+1 Arithmetic
x = b’1010’ Binary
z = (0.0, 1.0) Complex
r = SQRT (5.) Function
converged = ( ABS (x0 – x) < 2*SPACING (x) )
x = z’B’ Hexadecimal
k = 123 Integer
x = o’12’ Octal
r = 321. Real
a = 23. ; j = 120 ; ans = .TRUE.; Semicolon
k = SELECTED INTEGER KIND (20) Kind
m = SELECTED REAL KIND (16, 30)
(continued)

c 2001 J.E. Akin 37
 Fortran Statement Examples (continued)
Name Examples Comments
long = SELECTED REAL KIND (9, 20)
pi = 3.1459265 long
a=b+c Matrix add
d = MATMUL (a, b) Matrix multiply
e = TRANSPOSE (d) Matrix transpose
f = 0 ; g = (/ 2. , 4. , 6. /) Matrix initialize
B = Al:, n:(:–1) Matrix flipped
x = (/ (k, k = 0, n) /) * d Implied do
kth row => a(k,:) Pointer
corners => a(1:n:(n-1), 1:m:(m-1))
p 2 => r
student record%rank = 51 Defined type
patient data%city = ’houston’
sqrt(x) = DSQRT(x) ! function statement Obsolete
Backspace BACKSPACE i exp Compute unit
BACKSPACE 8 Unit
BACKSPACE (UNIT=9, IOSTAT=i, ERR=5) Error go to
BACKSPACE (9, IOSTAT=io ok, ERR=99)
BACKSPACE (UNIT=9, IOSTAT=io ok, ERR=99)
BACKSPACE (8, IOSTAT=io ok) I/O status
Block Data BLOCK DATA ! Obsolete
BLOCK DATA winter ! Obsolete Named
C C in column one denotes a comment line in F77 Obsolete
* in column one denotes a comment line in F77 Obsolete
! anywhere starts a comment line in F90
Call CALL sub1 (a, b)
CALL sub2 (a, b, *5) ! Obsolete, use CASE Alt return to 5
CALL sub3 No arguments
CALL subroutine name (args, optional args) Optional arg
Case CASE (range list) See SELECT
CASE (range list) select name Named
Case CASE DEFAULT See SELECT
Default CASE DEFAULT select name Named
Character CHARACTER (80) s, s 2*3(4)
CHARACTER *16 a, b, c
CHARACTER * home team :: recommended
CHARACTER (*), INTENT(IN) :: home team Intent
CHARACTER (LEN=3) :: b = ’xyz’ Initialize b
CHARACTER LEN=40 :: monday, wednesday, friday
CHARACTER (LEN=40), attr list :: last, first, middle
CHARACTER (40), attr list :: name, state
CHARACTER (*), PARAMETER :: reply = “Invalid Data”
CHARACTER (*, KIND=greek), attr list :: s1 list Kind
CHARACTER (*, KIND=greek), attr list :: last, first, middle
CHARACTER (KIND=cyrillic, LEN=40) :: name, state
CHARACTER (KIND=cyrillic, *), attr list :: s list
Close CLOSE (7) Unit number
CLOSE (UNIT=k)
CLOSE (UNIT=8, ERR=90, IOSTAT=i) Error go to
CLOSE (8, ERR=99, IOSTAT=io ok, STATUS=’KEEP’) I/O status
(continued)

c 2001 J.E. Akin 38
 Fortran Statement Examples (continued)
Name Examples Comments
CLOSE (9, ERR=99, IOSTAT=io, STATUS=’DELETE’) File status
CLOSE (UNIT=8, ERR=95, IOSTAT=io ok)
Common COMMON / name / h, p, t ! Obsolete Named common
COMMON p, d, q(m,n) ! Obsolete Blank common
Complex COMPLEX u, v, w(3, 6) :: recommended
COMPLEX :: u = (1.0,1.0), v = (1.0,10.0) Initialize u and v
COMPLEX :: variable list
COMPLEX attr list :: variable list
COMPLEX (KIND=i2 kind), attr list :: variable list Kind
Contains CONTAINS Internal definitions
CONTAINS
FUNCTION mine (b) Or subroutines
:::

END FUNCTION mine

Continuation ! any non-block character in column 6 flags continuation F77 obsolete
& at the end flags continuation to next line F90 standard
& at the beginning flags continuation from above line
a long name = a constant value* &
another value ! on following line
a long name here is set to = value
& * another value ! continued from above
Continue 100 CONTINUE Obsolete
Cycle CYCLE Current do only
CYCLE nested do name Terminate sub dos
Data DATA a, s / 4.01, ’z’ / Obsolete
DATA s fmt / ’(2F5.1)’ / Stored format
DATA (r(k), k=1,3) / 0.7, 0.8, 1.9 / Implied do
DATA array (4,4) / 1.0 / Single value
DATA bit val / b’0011111’ / Binary
Deallocate DEALLOCATE (force) File name
DEALLOCATE (force, STAT=i err) Error status
Dimension DIMENSION array (4, 4)
DIMENSION v(1000), w(3) = (/ 1., 2., 4. /) Initialize w
DIMENSION force(20), stiffness(:,:)
DIMENSION (5,10,3) :: triplet :: recommended
INTEGER, DIMENSION (:,:) :: material, nodes list Typed
REAL, DIMENSION(m, n) :: a, b
REAL, DIMENSION (:,:) :: force, stiffness
REAL, DIMENSION (5,10,3), INTENT(IN) :: triplet Intent
Do DO 100 j = init, last, incr ! Obsolete Labeled do
:::

100 CONTINUE Obsolete

DO j = init, last Unlabeled do
:::

END DO
DO ! forever Unlabeled do
:::

END DO ! forever
DO WHILE (diff <= delta) Unlabeled while
:::

END DO
DO 100 WHILE (diff < = delta) ! Obsolete Labeled while
(continued)

c 2001 J.E. Akin 39
 Fortran Statement Examples (continued)
Name Examples Comments
:::

100 CONTINUE Obsolete

DO Forever
DO k = i start, i stop Integer range
DO k = i start, i stop, i inc Increment
DO 10, k = i start, i stop Obsolete
do name: DO k = i start, i stop, i inc Named
do name: DO 10, k = i start, i stop, i inc Named label
DO 10, r variable = r start, r stop, r inc ! Obsolete Real range
Do While DO WHILE (.NOT. converged) Use DO-EXIT pair
DO 10, WHILE (.NOT. converged) Obsolete
do name: DO 10, WHILE (.NOT. converged) Obsolete
Double DOUBLE PRECISION a, d, y(2) Obsolete
Precision DOUBLE PRECISION :: a, d = 1.2D3, y(2) Initialize D
DOUBLE PRECISION, attr list :: variable list Obsolete
Else ELSE Then
ELSE leap year Named
Else If ELSE IF (k > 50) THEN
ELSE IF (days in year == 364) THEN
ELSE IF (days in year == 364) THEN leap year Named
Elsewhere ELSEWHERE See WHERE
End END
END name Named
End Block END BLOCK DATA Obsolete
Data END BLOCK DATA block data name Obsolete
End Do END DO
END DO do name Named
End Function END FUNCTION function name
END FUNCTION
End If END IF leap year Named
END IF
End Interface END INTERFACE
End Module END MODULE my matrix operators
END MODULE
End Program END PROGRAM program name
END PROGRAM
End Select END SELECT select name Named
END SELECT
End END SUBROUTINE name
Subroutine END SUBROUTINE
End Type END TYPE type name See TYPE
END TYPE
End Where END WHERE See WHERE
Endfile ENDFILE i exp Compute unit
ENDFILE (UNIT=k) Unit number
ENDFILE k
ENDFILE (UNIT=8, ERR=95) Error go to
ENDFILE (7, IOSTAT=io ok, ERR=99) I/O status
ENDFILE (UNIT=8, IOSTAT=k, ERR=9)
ENDFILE (UNIT=9, IOSTAT=io ok, ERR=99)
(continued)

c 2001 J.E. Akin 40
 Fortran Statement Examples (continued)
Name Examples Comments
Entry ENTRY sec1 (x, y) Arguments
ENTRY sec2 (a1, a2, *4) ! Obsolete, use CASE Alternate return to 4
ENTRY section No arguments
ENTRY entry name RESULT(variable name) Result
Equivalence EQUIVALENCE (v (1), a (1,1)) Obsolete
EQUIVALENCE (v, a)
EQUIVALENCE (x, v(10)), (p, q, d)
Exit EXIT Current do only
EXIT nested do name Current & sub-dos
External EXTERNAL my program
Format 10 FORMAT (2X, 2I3, 3F6.1, 4E12.2, 2A6, 3L2 ) XIFEAL
10 FORMAT (// 2D6.1, 3G12.2) D, G
10 FORMAT (2I3.3, 3G6.1E3, 4E12.2E3) Exponent w
10 FORMAT (’a quoted string’, ”another”, I2) Strings
10 FORMAT (1X, T10, A1, T20, A1) Tabs
10 FORMAT (5X, TR10, A1, TR10, A1, TL5, A1) Tab right, left
10 FORMAT (”Init=”, I2, :, 3X, ”Last=”, I2) : stop if empty
10 FORMAT (’Octal ’, o6, ’, Hex ’ z6) Octal, hex
10 FORMAT (specification and edit list)
Function FUNCTION z (a, b) Arguments
FUNCTION w (e, d) RESULT (a) Result
FUNCTION name (args)
FUNCTION name No argument
FUNCTION name (args) RESULT(variable name)
INTEGER FUNCTION n (j, k) Type
INTEGER FUNCTION name (args)
COMPLEX RECURSIVE FUNCTION dat (args)
RECURSIVE REAL FUNCTION name (args)
Go To GO TO 99 Unconditional
GO TO (10,20,35,95), i variable ! Obsolete Computed
If IF (arithmetic exp) 95, 10, 20 ! Obsolete Arithmetic
IF (logic) RETURN Logical if
IF (logic) n = n + 2
IF (logic) THEN if block
n=n+1
k=k+1
END IF
leap year: IF (logical expression) THEN Named
IF (logic) THEN if else block
n=n+1
ELSE
k=k+1
END IF
IF (c == ’a’) THEN if else-if block
na = na + 1
CALL sub a
ELSE IF (c == ’b’) THEN (Use CASE)
nb = nb + 1
ELSE IF (c == ’c’) THEN
nc = nc + 1
(continued)

c 2001 J.E. Akin 41
 Fortran Statement Examples (continued)
Name Examples Comments
CALL sub c
END IF
Implicit IMPLICIT INTEGER (i-n) F77 default
Type IMPLICIT REAL (a-h,o-z) F77 default
IMPLICIT NONE Recommended F90
IMPLICIT CHARACTER *10 (f,l) Character
IMPLICIT COMPLEX (a-c,z) Complex
IMPLICIT TYPE (color) (b,g,r) Derived type
IMPLICIT LOGICAL (KIND=bit) (m) Logical
Include INCLUDE ’path/source.f’
Inquire INQUIRE (UNIT=3, OPENED=t or f) Opened
INQUIRE (FILE=’mydata’, EXIST=t or f) Exists
INQUIRE (UNIT=3, OPENED=ok, IOSTAT=k) I/O status
INQUIRE (FILE=’name string’, see INQUIRE table) Re file
INQUIRE (NAME=s variable, see INQUIRE table) Re file
INQUIRE (IOLENGTH=i var, see INQUIRE table) Re output
INQUIRE (7, see INQUIRE table) Re unit
INQUIRE (UNIT=8, see INQUIRE table) Re unit
Integer INTEGER c, d(4) :: Recommended
INTEGER (long), attr list :: variable list
INTEGER, DIMENSION (4) :: a, d, e
INTEGER, ALLOCATABLE, DIMENSION(:,:) :: a, b Allocatable
INTEGER :: a = 100, b, c = 9 Initialize a & c
INTEGER :: i, j, k, l, m, n, month, year = 1996
INTEGER, attr list :: variable list
INTEGER (KIND=i2 kind), attr list :: variable list Kind
Intent INTENT (IN) :: credit card owners
INTENT (INOUT) :: amount due
INTENT (OUT) income rank
Interface INTERFACE ASSIGNMENT (=) User extension
INTERFACE OPERATOR (+) User extension
INTERFACE OPERATOR (–) User extension
INTERFACE OPERATOR (/) User extension
INTERFACE OPERATOR (*) User extension
INTERFACE OPERATOR (**) User extension
INTERFACE OPERATOR (.operator.) User defined
INTERFACE
INTERFACE interface name
Intrinsic INTRINSIC SQRT, EXP Functions
Logical LOGICAL c :: recommended
LOGICAL, ALLOCATABLE :: mask(:), mask 2(:,:) Allocatable
LOGICAL (KIND = byte) :: flag, status Kind
LOGICAL :: b = .FALSE., c Initialize b
Module MODULE PROCEDURE mat x mat, mat x vec Generics
MODULE my matrix operators
Namelist NAMELIST /data/ s, n, d Obsolete
Nullify NULLIFY (pointer list)
Open OPEN (7) Unit number
OPEN (UNIT=3, FILE="data.test") Name
OPEN (UNIT=2, FILE="data", STATUS = "old") File status
(continued)

c 2001 J.E. Akin 42
 Fortran Statement Examples (continued)
Name Examples Comments
OPEN (UNIT=3, IOSTAT=k) I/O status
OPEN (9, ERR = 12, ACCESS ="direct") Access type
OPEN (8, ERR=99, IOSTAT=io ok) Error go to
OPEN (UNIT=8, ERR=99, IOSTAT=io ok)
Optional OPTIONAL slow, fast Argument list
OPTIONAL :: argument list
Parameter PARAMETER (a="xyz"), (pi=3.14159) Character
PARAMETER (a=”z”, pi=3.14159) Real
PARAMETER (x=11, y = x/3) Computed
PARAMETER, REAL :: weight = 245.6 Type
Pause PAUSE ! for human action Obsolete
Pointer POINTER current, last :: recommended
POINTER :: name(4,5) Rank
REAL, POINTER :: y(:), x(:,:,:) Type
Print PRINT *, a, j List-directed
PRINT *, output list Default unformatted
PRINT *, (io implied do) Implied do
PRINT *, “The squre root of”, n, ’is’, SQRT(n) Function
PRINT *, (4*k-1, k=1,10,3)
PRINT 10, a, j Formatted
PRINT 10, m array Array
PRINT 10, (m(i), i = j,k) Implied do
PRINT 10, s(j:k) Substring
PRINT ’(A6, I3)’, a, j Character, integer
PRINT FMT=’(A6, I3)’, a, j Included format
PRINT data namelist ! Obsolete Namelist
PRINT ’(formats)’, output list Formatted
PRINT ’(formats)’, (io implied do) Implied do
PRINT ’(I4)’, (2*k, k=1,5)
Private PRIVATE
PRIVATE :: module variable list Specific items
Program PROGRAM my job
PROGRAM
Public PUBLIC
PUBLIC :: module variable list Specific items
Read READ *, a, j List-directed
READ 1, a , j Formatted
READ 10, m array Formatted array
READ 10, (m(i), i=j, k) Implied do
READ 10, s(i:k) Substring
READ ’(A6, I3)’, a, i Character, integer
READ (1, 2) x, y Formatted file
READ (UNIT=1, FMT=2) x, y
READ (1, 2, ERR=8, END=9) x, y End of file go to
READ (UNIT=1, FMT=2, ERR=8, END=9) x, y Error go to
READ (*, 2) x, y Formatted, std out
READ (*, 10) m array Unformatted array
READ (*, 10) (m(i), i=j, k) Implied do
READ (*, 10) s(i:k) Substring
READ (1, *) x, y Unformatted file
(continued)

c 2001 J.E. Akin 43
 Fortran Statement Examples (continued)
Name Examples Comments
READ (*, *) x, y Unformatted, std out
READ (1, ’(A6, I3)’) x, y Character, integer
READ (1, FMT=’(A6, I3)’) x, y Included format
READ (1, s fmt) x, y Format in a string
READ (1, FMT=s fmt) x, y
READ (*, NML=data) ! Obsolete Namelist read
READ (1, NML=data) ! Obsolete Namelist from a file
READ (1, END=8, ERR=9) x, y Unformatted
READ (s2, 1, ERR=9) x Internal, formatted
READ (s2, *, ERR=9) x Unformatted
READ (s2, REC=4, END=8) x Internal, direct
READ (1, REC=3) v Unformatted direct
READ (1, 2, REC=3) v Formatted direct
READ *, input list Default unformatted
READ *, (io implied do) Implied do
READ *, (a(j,:), j=1, rows)
READ ’(formats)’, input list Formatted read
READ ’(formats)’, (io implied do) Formatted read
READ ’(5I5, (5I5))’, (num(k), k=1, n)
READ (8, FMT=20), input list Formatted
READ (8, FMT=20, ADVANCE=’NO’), input Advance
READ (9, FMT=20, io spec list), input list I/O Specification
READ (UNIT=7, 20, io spec list), input list
READ (UNIT=8, FMT=10, io spec list), input
READ (7, s fmt, io spec list), input list Stored format
READ (UNIT=7, s fmt, io spec list), input
READ (9, ’(formats)’, io spec list), input list Inline format
READ (UNIT=9, ’(formats)’, io spec list), input
READ (8), input list Binary read
READ (UNIT=7), input list
READ (8, io spec list), input list I/O Specification
READ (UNIT=9, io spec list), input list
READ (s variable, FMT=20), input list Internal file,
READ (UNIT=s variable, 10, io spec list), input type change
Real REAL*4 :: recommended
REAL :: r, m(9)
REAL*16 a, b, c Quad Precision
REAL*8, DIMENSION(n) :: a, b, c Double Precision
REAL :: a = 3.14, b, c = 100.0 Initialize a & c
REAL :: variable list
REAL, attr list :: variable list
REAL, POINTER :: a(:,:)
REAL (KIND=i2 kind), attr list :: variable list Kind
REAL (double), attr list :: variable list
Recursive RECURSIVE FUNCTION name Function
RECURSIVE FUNCTION a(n) RESULT(fac) Result
INTEGER RECURSIVE FUNCTION name (args)
RECURSIVE SUBROUTINE name (args) Subroutine
RECURSIVE SUBROUTINE name
Return RETURN Standard return
(continued)

c 2001 J.E. Akin 44
 Fortran Statement Examples (continued)
Name Examples Comments
Rewind REWIND i exp Compute unit
REWIND 2 Unit number
REWIND k
REWIND (UNIT=8, IOSTAT=k, ERR=9) Error go to
REWIND (UNIT=8, ERR=95)
REWIND (8, IOSTAT=io ok, ERR=99) I/O status
Save SAVE a, /name/, c Scalars, common
SAVE Everything
SAVE :: variable list
Select Case SELECT CASE (value)
name: SELECT CASE (value) Named
u or l SELECT CASE (letter) Block
CASE ("a":"z") ! lower case
lower = .TRUE.
CASE ("A":"Z") ! upper case
lower = .FALSE.
CASE DEFAULT ! not a letter
PRINT *, "Symbol is not a letter", letter
lower = .FALSE.
END SELECT u or l
Sequence SEQUENCE Forced storage
Stop STOP
STOP "invalid data" With message
Subroutine SUBROUTINE sub1 (a, b)
SUBROUTINE sub1 No arguments
SUBROUTINE name (args, optional args) Optional arguments
SUBROUTINE sub3 (a, b, *9) ! Obsolete, use CASE Return to 9
RECURSIVE SUBROUTINE sub2 (a, b) Recursive
Target TARGET :: name, name 2 See Pointer
TARGET :: name(4,5), name 2(3)
Type TYPE (person) car pool(5) User defined type
Declaration TYPE (color), DIMENSION(256) :: hues
TYPE (type name), attr list :: variable list
TYPE (person), DIMENSION (n) :: address book
TYPE (type name) :: variable list
TYPE (student record) Definition block
CHARACTER (name len) :: last, first
INTEGER :: rank
END TYPE student record
Type TYPE, PRIVATE name Access
Statement TYPE, PUBLIC :: name
Use USE module name
USE module name, ONLY: list in module name Only
USE module name, var subr fun name => old name Rename
Where WHERE (logical array mask) Then
WHERE ( a array > 0.0 ) Where block
sqrt a = SQRT(a array)
END WHERE
WHERE ( mask > 0.0 ) Elsewhere block
a array = mask
(continued)

c 2001 J.E. Akin 45
 Fortran Statement Examples (continued)
Name Examples Comments
ELSEWHERE
a array = 0.0
END WHERE
WHERE (a array>0) b array = SQRT(a array) Statement
Write WRITE (*, 10) s(j:k) Substring
WRITE (1, *) x, y Unformatted file
WRITE (*, *) x, y Unformatted
WRITE (1, ’(A6, I3)’) x, y Character, integer
WRITE (1, FMT=’(A6, I3)’) x, y Included format
WRITE (1, s fmt) x, y Stored format string
WRITE (1, FMT=s fmt) x, y
WRITE (*, NML=data) ! Obsolete Namelist to stdout
WRITE (1, NML=data) ! Obsolete Namelist to a file
WRITE (1, END=8, ERR=9) x, y Unformatted
WRITE (1, REC=3) v Unformatted direct
WRITE (1, 2, REC=3) v Formatted direct
WRITE (s2, 1, ERR=9) x Internal, format
WRITE (s2, *, ERR=9) x Unformatted
WRITE (s2, REC=4, END=8) x Internal, direct
WRITE *, output list Unformatted
WRITE *, (io implied do) Implied do
WRITE *, ((a(i, j), j=1, cols), i=1, rows)
WRITE ’(formats)’, output list Formatted write
WRITE ’(formats)’, (io implied do) Implied do
WRITE (7, 10, ADVANCE=’NO’), output list Advance
WRITE (8, 10, io spec list), output list I/O specification
WRITE (9, FMT=20, io spec list), output list
WRITE (UNIT=7, 10, io spec list), output list
WRITE (9, s fmt, io spec list), output list Stored format
WRITE (UNIT=8, s fmt, io spec list), output
WRITE (9, ’(formats)’, io spec list), output list Inline format
WRITE (UNIT=7, ’(formats)’, io spec list), output
WRITE (8), output list Binary write
WRITE (7), (io implied do) Implied do
WRITE (8, ADVANCE=’NO’), output list Advance
WRITE (9, io spec list), output list I/O specification
WRITE (UNIT=9, io spec list), output list
WRITE (s variable, FMT=20), output list Internal file
WRITE (UNIT=s variable, FMT=20), output list
WRITE (s variable, 20, io spec list), output list I/O specification
WRITE (UNIT=s var, FMT=20, io spec), output

c 2001 J.E. Akin 46
Appendix C

Selected Exercise Solutions

C.1 Problem 1.8.1 : Checking trigonometric identities

The Fortran 90 program and output follow. The error levels are due to the fact that F90 defaults to
single precision reals. F90 is easily extended to double precision, and in theory supports any level of user
specified precision. For simplicity the F77 default naming convention for integers and reals is used. That
is not a good practice since safety dictates declaring the type of each variable at the beginning of each
program. (Try changing the reals to double precision to verify that the error is indeed reduced.)
[ 1] implicit none
[ 2] integer :: k,n = 16
[ 3] real, parameter :: pi = 3.141592654 ! set constant
[ 4] print *,’ Theta sinˆ2+cosˆ2 error’
[ 4] do k = 0, n ! Loop over (n+1) points
[ 5] theta = k*pi/n
[ 6] sint = sin( theta )
[ 7] cost = cos( theta )
[ 8] test = sint*sint + cost*cost
[ 9] write (*, ’( 3(1pe14.5) )’) theta, test, 1.-test
[10] end do ! over k
Theta sinˆ2+cosˆ2 error
0.00000E+00 1.00000E+00 0.00000E+00
1.96350E-01 1.00000E+00 5.96046E-08
3.92699E-01 1.00000E+00 0.00000E+00
5.89049E-01 1.00000E+00 0.00000E+00
7.85398E-01 1.00000E+00 5.96046E-08
9.81748E-01 1.00000E+00 0.00000E+00
1.17810E+00 1.00000E+00 5.96046E-08
1.37445E+00 1.00000E+00 0.00000E+00
1.57080E+00 1.00000E+00 0.00000E+00
1.76715E+00 1.00000E+00 5.96046E-08
1.96350E+00 1.00000E+00 0.00000E+00
2.15985E+00 1.00000E+00 0.00000E+00
2.35619E+00 1.00000E+00 5.96046E-08
2.55254E+00 1.00000E+00 0.00000E+00
2.74889E+00 1.00000E+00 0.00000E+00
2.94524E+00 1.00000E+00 0.00000E+00
3.14159E+00 1.00000E+00 0.00000E+00

C.2 Problem 1.8.2 : Newton-Raphson algorithm

The most convenient form of loop is the post-test loop, which allows each iteration to be calculated and
the error checked at the end.
xnew = x
do f
x = xnew
xnew = x - f(x)/fprime(x)
g
while (abs(xnew-x) < tolerance)

The alternate logic constructs employ tests at the end of the loop and transfer out the end of the
loop when necessary. M ATLAB and C++ transfer using the “break” command while F90 uses the “exit”
command.

c 2001 J.E. Akin 47

 A F90 program with an infinite loop, named testnewton.f90, and its result is given below. Be
warned that this version uses the IMPLICIT name styles for integers and reals instead of the better strong
typing that results from the recommended use of IMPLICIT NONE.
[ 1] function f(x) result(y)
[ 2] real, intent (in) :: x
[ 3] real :: y
[ 4] y = exp(2*x) - 5*x - 1
[ 5] end function f
[ 6] !
[ 7] function fprime(x) result(y)
[ 8] real, intent (in) :: x
[ 9] real :: y
[10] y = 2*exp(2*x) - 5
[11] end function fprime
[12] !
[13] program main
[14] implicit none
[15] real, parameter :: tolerance = 1.e-6 ! set constant
[16] real :: x, xnew = 3. ! Initial value
[17] integer :: iteration
[18] iteration = 0
[19] ! Iteration count
[20] do ! forever until true
[21] iteration = iteration + 1
[22] x = xnew
[23] xnew = x - f(x)/fprime(x)
[24] if ( abs(xnew - x) < tolerance ) exit ! converged is true
[25] end do ! forever
[26] print *, ’Solution: ’, xnew, ’, Iterations:’, iteration
[27] end program main
>>f90 -o newton testnewton.f90
>>newton
Solution: 0.8093941 , Iterations: 10

C.3 Problem 1.8.3 : Game of life

[ 1] program game of life ! procedural version
[ 2] implicit none
[ 3] integer, parameter :: boardsize = 10
[ 4] integer :: board (boardsize, boardsize) = 0
[ 5] integer :: newboard (boardsize, boardsize)
[ 6] character(len=1) :: ok ! page prompt
[ 7] integer :: k, number ! loops
[ 8]
[ 9] ! Initial life data, the "Glider"
[ 10] board (3, 3) = 1; board (4, 4) = 1; board (5, 4) = 1
[ 11] board (5, 3) = 1; board (5, 2) = 1
[ 12]
[ 13] print *, "Initial Life Display:"
[ 14] call spy (board) ! show initial lifeforms
[ 15] print *, "Initially alive = ", sum (board); print *, " "
[ 16]
[ 17] print *, "Enter number of generations to display:"
[ 18] read *, number
[ 19] do k = 1, number
[ 20] newboard = next generation (board)
[ 21] board = newboard ! save current lifeforms
[ 22] call spy (board) ! show current lifeforms
[ 23] print * ; print *, "Generation number = ", k
[ 24] print *, "Currently alive = ", sum (newboard)
[ 25]
[ 26] print *, ’continue? (y, n)’
[ 27] read *, ok ! read any character to continue
[ 28] if ( ok == ’n’ ) exit ! this do loop only
[ 29] end do ! on k for number of generations
[ 30]
[ 31] contains ! internal (vs external) subprograms
[ 32]
[ 33] function next generation (board) result (newboard)
[ 34] ! Compute the next generation of life
[ 35] integer, intent(in) :: board (:, :)
[ 36] integer :: newboard (size(board, 1), size(board, 2))
[ 37] integer :: i, j, neighbors ! loops
[ 38]
[ 39] newboard = 0 ! initialize next generation
[ 40] do i = 2, boardsize - 1
[ 41] do j = 2, boardsize - 1
[ 42] neighbors = sum (board (i - 1:i + 1, j - 1:j + 1)) %
[ 43] - board (i, j)
[ 44] if ( board (i, j) == 1 ) then ! life in the cell

c 2001 J.E. Akin 48

 [ 45] if ( (neighbors > 3 .or. neighbors < 2) ) then
[ 46] newboard (i, j) = 0 ! it died
[ 47] else
[ 48] newboard (i, j) = 1 ! newborn
[ 49] end if ! on number of neighbors
[ 50] else ! no life in the cell
[ 51] if ( neighbors == 3 ) then
[ 52] newboard (i, j) = 1 ! newborn
[ 53] else
[ 54] newboard (i, j) = 0 ! died
[ 55] end if ! on number of neighbors
[ 56] end if ! life status
[ 57] end do ! on column j
[ 58] end do ! on row i
[ 59] end function next generation
[ 60]
[ 61] Subroutine spy (board) ! model matlab spy function
[ 62] ! Show an X at each non-zero entry of board, else show -
[ 63] integer, intent(in) :: board (:, :)
[ 64] character (len=1) :: line (size(board, 1)) ! a line on screen
[ 65] integer :: i ! loops
[ 66]
[ 67] line = ’ ’ ! blank out the line
[ 68] do i = 1, size (board, 1 ) ! loop over each row
[ 69] line (1:size (board, 2 )) = ’-’ ! current board width
[ 70] where ( board (i, :) /= 0 ) line = ’X’ ! mark non-zero columns
[ 71] write (*, ’(80a1)’) line ! print current row
[ 72] end do ! over all rows
[ 73] end subroutine spy
[ 74] end program ! game of life
[ 75]
[ 76] ! Running gives:
[ 77] ! Initial Life Display:
[ 78] ! ----------
[ 79] ! ----------
[ 80] ! --X-------
[ 81] ! ---X------
[ 82] ! -XXX------
[ 83] ! ----------
[ 84] ! ----------
[ 85] ! ----------
[ 86] ! ----------
[ 87] ! ----------
[ 88] ! Initially alive = 5
[ 89] !
[ 90] ! Enter number of generations to display: 4
[ 91] ! ----------
[ 92] ! ----------
[ 93] ! ----------
[ 94] ! -X-X------
[ 95] ! --XX------
[ 96] ! --X-------
[ 97] ! ----------
[ 98] ! ----------
[ 99] ! ----------
[100] ! ----------
[101] !
[102] ! Generation number = 1
[103] ! Currently alive = 5
[104] ! continue? (y, n) n

C.4 Problem 2.5.1 : Conversion factors

This code illustrates the type of global units conversion factors that you can define for your field of study.
They can be accessed by any program that includes a use Conversion Constants line and cites a
parameter name, as shown on line 16.
[ 1] Module Conversion Constants ! DefineUnits Conversion
[ 2] ! Define selected precision
[ 3] INTEGER, PARAMETER :: DP = KIND (1.d0) ! Alternate form
[ 4] ! ========== Metric Conversions ==========
[ 5] real(DP), parameter:: cm Per Inch = 2.54 DP
[ 6] real(DP), parameter:: kg Per Pound = 0.45359237 DP
[ 7] real(DP), parameter:: kg Per Short Ton = 907.18474 DP
[ 8] real(DP), parameter:: kg Per Long Ton = 1016.0469088 DP
[ 9] real(DP), parameter:: m Per Foot = 3.048 DP
[10] real(DP), parameter:: m Per Mile = 1609.344 DP
[11] real(DP), parameter:: m Per Naut Mile = 1852.0 DP
[12] real(DP), parameter:: m Per Yard = 0.9144 DP
[13] end Module Conversion Constants
[14] Program Test

c 2001 J.E. Akin 49

 [15] use Conversion Constants
[16] print *, ’cm Per Inch = ’, cm Per Inch ; End Program Test
[17] ! Running gives: cm Per Inch = 2.54000000000000004

This code illustrates the type of common physical constants that can be made available as global
variables that you can define for your field of study. They can be accessed by any program that includes
a use Physical Constants line and cites a parameter name, as shown on line 60 below.
[ 1] Module Physical Constants ! Define Physical Constants
[ 2] ! Define selected precision
[ 3] INTEGER, PARAMETER :: DP = KIND (1.d0) ! Alternate form
[ 4]
[ 5] ! ========== Physics Constants and units ==========
[ 6] real(DP), parameter:: AMU Value = 1.6605402E-27 DP ! kg
[ 7] real(DP), parameter:: Atmosphere Pres = 9.80665E+04 DP ! Pa
[ 8] real(DP), parameter:: Avogadro = 6.0221367E+23 DP ! 1/mol
[ 9] real(DP), parameter:: Bohr Magneton = 9.2740154E-24 DP ! J/T
[10] real(DP), parameter:: Bohr Radius = 5.29177249E-11 DP ! m
[11] real(DP), parameter:: Boltzmann = 1.380657E-23 DP ! J/K
[12] real(DP), parameter:: c Light = 2.997924580E+8 DP ! m/s
[13] real(DP), parameter:: Electron Compton = 2.42631058E-12 DP ! m
[14] real(DP), parameter:: Electron Angular = 5.2729E-35 DP ! J*s
[15] real(DP), parameter:: Electron Charge =-1.60217738E-19 DP ! coul
[16] real(DP), parameter:: Electron Mass Rest = 9.1093897E-31 DP ! kg
[17] real(DP), parameter:: Electron Moment = 9.2847700E-24 DP ! J/T
[18] real(DP), parameter:: Electron Radius = 2.81794092E-15 DP ! m
[19] real(DP), parameter:: Faraday = 9.6485309E+04 DP ! C/mo
[20] real(DP), parameter:: G Universal = 6.67260E-11 DP ! mˆ3/(sˆ2*kg)
[21] real(DP), parameter:: Light Year = 9.46073E+15 DP ! m
[22] real(DP), parameter:: Mech equiv Heat = 4.185E+3 DP ! J/kcal
[23] real(DP), parameter:: Molar Volume = 0.02241410 DP ! mˆ3/mol
[24] real(DP), parameter:: Neutron Mass = 1.6749286E-27 DP ! kg
[25] real(DP), parameter:: Permeability = 1.25663706143E-06 DP ! H/m
[26] real(DP), parameter:: Permittivity = 8.85418781762E-12 DP ! F/m
[27] real(DP), parameter:: Planck Const = 6.6260754E-34 DP ! J*s
[28] real(DP), parameter:: Proton Mass = 1.6726230E-27 DP ! kg
[29] real(DP), parameter:: Proton Moment = 1.41060761E-26 DP ! J/T
[30] real(DP), parameter:: Quantum charge r = 4.13556E+12 DP ! J*s/C
[31] real(DP), parameter:: Rydberg inf = 1.0973731534E+07 DP! 1/m
[32] real(DP), parameter:: Rydberg Hydrogen = 1.09678E+07 DP ! 1/m
[33] real(DP), parameter:: Std Atmosphere = 1.01325E+05 DP ! Pa
[34] real(DP), parameter:: Stefan Boltzmann = 5.67050E-08 DP ! W/(mˆ2*Kˆ4)
[35] real(DP), parameter:: Thomson cross sect = 6.6516E-29 DP ! mˆ2
[36] real(DP), parameter:: Universal Gas C = 8.314510 DP ! J/mol*K
[37]
[38] ! ========== Astronomy Constants and units ==========
[39] real(DP), parameter:: AU Earth Sun = 1.4959787E+11 DP ! m
[40] real(DP), parameter:: Anomal Month = 27.5546 DP ! days
[41] real(DP), parameter:: Anomal Year = 365.2596 DP ! days
[42] real(DP), parameter:: Dracon Month = 27.2122 DP ! days
[43] real(DP), parameter:: Earth G = 9.80665 DP ! m/sˆ2
[44] real(DP), parameter:: Earth Mass = 5.974E+24 DP ! kg
[45] real(DP), parameter:: Earth Radius Eq = 6.37814E+6 DP ! m
[46] real(DP), parameter:: Earth Radius Mean = 6.371E+6 DP ! m
[47] real(DP), parameter:: Earth Radius Polar = 6.356755E+6 DP ! m
[48] real(DP), parameter:: Julian Year = 365.25 DP ! days
[49] real(DP), parameter:: Rotation Day = 23.93447222 DP ! hours
[50] real(DP), parameter:: Sidereal Day = 23.93446944 DP ! hours
[51] real(DP), parameter:: Sidereal Month = 27.3217 DP ! days
[52] real(DP), parameter:: Sidereal Ratio = 1.0027379092558 DP
[53] real(DP), parameter:: Sidereal Year = 365.2564 DP ! days
[54] real(DP), parameter:: Solar Day = 24.06571111 DP ! hours
[55] real(DP), parameter:: Synodic Month = 29.5306 DP ! days
[56] real(DP), parameter:: Tropical Year = 365.2422 DP ! days
[57] end Module Physical Constants ! Define Physical Constants
[58] Program Test
[59] use Physical Constants
[60] print *, ’Avogadro = ’, Avogadro ; End Program Test
[61] ! Running gives: Avogadro = 0.602213669999999967E+24

C.5 Problem 3.5.3 : Creating a vector class

We begin by defining the components to be included in our vector object. They include the length of each
vector and a corresponding real array of pointers to the vector components:
[ 1] module class Vector ! filename: class Vector.f90
[ 2] ! public, everything by default, but can specify any
[ 3] implicit none
[ 4] type Vector
[ 5] private

c 2001 J.E. Akin 50

 [ 6] integer :: size ! vector length
[ 7] real, pointer, dimension(:) :: data ! component values
[ 8] end type Vector

For persons familiar with vectors the use of overloaded operators makes sense (but it often does not
make sense). Thus we overload the addition, subtraction, multiplication, assignment, and logical equal
to operators by defining the correct class members to be used for different argument types:
[ 9] ! Overload common operators
[ 10] interface operator (+) ! add others later
[ 11] module procedure add Vector, add Real to Vector; end interface
[ 12] interface operator (-) ! add unary versions later
[ 13] module procedure subtract Vector, subtract Real; end interface
[ 14] interface operator (*) ! overload *
[ 15] module procedure dot Vector, real mult Vector, Vector mult real
[ 16] end interface
[ 17] interface assignment (=) ! overload =
[ 18] module procedure equal Real; end interface
[ 19] interface operator (==) ! overload ==
[ 20] module procedure is equal to; end interface
[ 21]

Then we encapsulate the supporting member functions, beginning with two constructors, assign and
make Vector:
[ 22] contains ! functions & operators
[ 23]
[ 24] function assign (values) result (name) ! array to vector constructor
[ 25] real, intent(in) :: values(:) ! given rank 1 array
[ 26] integer :: length ! array size
[ 27] type (Vector) :: name ! Vector to create
[ 28] length = size(values); allocate ( name%data(length) )
[ 29] name % size = length; name % data = values; end function assign
[ 30]
[ 31] function make Vector (len, values) result(v) ! Optional Constructor
[ 32] integer, optional, intent(in) :: len ! number of values
[ 33] real, optional, intent(in) :: values(:) ! given values
[ 34] type (Vector) :: v
[ 35] if ( present (len) ) then ! create vector data
[ 36] v%size = len ; allocate ( v%data(len) )
[ 37] if ( present (values)) then ; v%data = values ! vector
[ 38] else ; v%data = 0.d0 ! null vector
[ 39] end if ! values present
[ 40] else ! scalar constant
[ 41] v%size = 1 ; allocate ( v%data(1) ) ! default
[ 42] if ( present (values)) then ; v%data(1) = values(1) ! scalar
[ 43] else ; v%data(1) = 0.d0 ! null
[ 44] end if ! value present
[ 45] end if ! len present
[ 46] end function make Vector
[ 47]

The remainder of the members are given in alphabetical order:

[ 48] function add Real to Vector (v, r) result (new) ! overload +
[ 49] type (Vector), intent(in) :: v
[ 50] real, intent(in) :: r
[ 51] type (Vector) :: new ! new = v + r
[ 52] if ( v%size < 1 ) stop "No sizes in add Real to Vector"
[ 53] allocate ( new%data(v%size) ) ; new%size = v%size
[ 54] ! new%data = v%data + r ! as array operation
[ 55] new%data(1:v%size) = v%data(1:v%size) + r ; end function
[ 56]
[ 57] function add Vector (a, b) result (new) ! vector + vector
[ 58] type (Vector), intent(in) :: a, b
[ 59] type (Vector) :: new ! new = a + b
[ 60] if ( a%size /= b%size ) stop "Sizes differ in add Vector"
[ 61] allocate ( new%data(a%size) ) ; new%size = a%size
[ 62] new%data = a%data + b%data ; end function add Vector

Note that lines 55 and 62 above are similar ways to avoid writing serial loops that would have to be used
in most languages. This keeps the code cleaner and shorter, and more importantly it lets the compiler
carry out those operations in parallel on some machines.
While copy members are very important to C++ programmers the following copy Vector should
probably be omitted since you would not usually pass big arrays as copies and F90 defaults to passing by
reference unless forced to pass by value.
[ 63]
[ 64] function copy Vector (name) result (new)
[ 65] type (Vector), intent(in) :: name

c 2001 J.E. Akin 51

 [ 66] type (Vector) :: new
[ 67] allocate ( new%data(name%size) ) ; new%size = name%size
[ 68] new%data = name%data ; end function copy Vector

The routine delete Vector is the destructor for this class. In some sense it is incomplete because it
does not delete the size attribute. It was decided that while the actual array of data may take a huge
amount of storage, the single integer was not important. To be more complete one would have to have to
make size an integer pointer and allocate and deallocate it at numerous locations within this module.
[ 69]
[ 70] subroutine delete Vector (name) ! deallocate allocated items
[ 71] type (Vector), intent(inout) :: name
[ 72] integer :: ok ! check deallocate status
[ 73] deallocate (name%data, stat = ok )
[ 74] if ( ok /= 0 ) stop "Vector not allocated in delete Vector"
[ 75] name%size = 0 ; end subroutine delete Vector
[ 76]
[ 77] function dot Vector (a, b) result (c) ! overload *
[ 78] type (Vector), intent(in) :: a, b
[ 79] real :: c
[ 80] if ( a%size /= b%size ) stop "Sizes differ in dot Vector"
[ 81] c = dot product(a%data, b%data); end function dot Vector
[ 82]
[ 83] subroutine equal Real (new, R) ! overload =, real to vector
[ 84] type (Vector), intent(inout) :: new
[ 85] real, intent(in) :: R
[ 86] if ( associated (new%data) ) deallocate (new%data)
[ 87] allocate ( new%data(1) ); new%size = 1
[ 88] new%data = R ; end subroutine equal Real
[ 89]
[ 90] logical function is equal to (a, b) result (t f) ! overload ==
[ 91] type (Vector), intent(in) :: a, b ! left & right of ==
[ 92] t f = .false. ! initialize
[ 93] if ( a%size /= b%size ) return ! same size ?
[ 94] t f = all ( a%data == b%data ) ! and all values match
[ 95] end function is equal to
[ 96]
[ 97] function length (name) result (n) ! accessor member
[ 98] type (Vector), intent(in) :: name
[ 99] integer :: n
[100] n = name % size ; end function length
[101]
[102] subroutine list (name) ! accessor member, for prettier printing
[103] type (Vector), intent(in) :: name
[104] print *,"[", name % data(1:name%size), "]"; end subroutine list
[105]
[106] function normalize Vector (name) result (new)
[107] type (Vector), intent(in) :: name
[108] type (Vector) :: new
[109] real :: total, nil = epsilon(1.0) ! tolerance
[110] allocate ( new%data(name%size) ) ; new%size = name%size
[111] total = sqrt ( sum ( name%data**2 ) ) ! intrinsic functions
[112] if ( total < nil ) then ; new%data = 0.d0 ! avoid division by 0
[113] else ; new%data = name%data/total
[114] end if ; end function normalize Vector
[115]
[116] subroutine read Vector (name) ! read array, assign
[117] type (Vector), intent(inout) :: name
[118] integer, parameter :: max = 999
[119] integer :: length
[120] read (*,’(i1)’, advance = ’no’) length
[121] if ( length <= 0 ) stop "Invalid length in read Vector"
[122] if ( length >= max ) stop "Maximum length in read Vector"
[123] allocate ( name % data(length) ) ; name % size = length
[124] read *, name % data(1:length) ; end subroutine read Vector
[125]
[126] function real mult Vector (r, v) result (new) ! overload *
[127] real, intent(in) :: r
[128] type (Vector), intent(in) :: v
[129] type (Vector) :: new ! new = r * v
[130] if ( v%size < 1 ) stop "Zero size in real mult Vector"
[131] allocate ( new%data(v%size) ) ; new%size = v%size
[132] new%data = r * v%data ; end function real mult Vector
[133]
[134] function size Vector (name) result (n) ! accessor member
[135] type (Vector), intent(in) :: name
[136] integer :: n
[137] n = name % size ; end function size Vector
[138]
[139] function subtract Real(v, r) result(new) ! vector-real, overload -
[140] type (Vector), intent(in) :: v
[141] real, intent(in) :: r
[142] type (Vector) :: new ! new = v + r

c 2001 J.E. Akin 52

 [143] if ( v%size < 1 ) stop "Zero length in subtract Real"
[144] allocate ( new%data(v%size) ) ; new%size = v%size
[145] new%data = v%data - r ; end function subtract Real
[146]
[147] function subtract Vector (a, b) result (new) ! overload -
[148] type (Vector), intent(in) :: a, b
[149] type (Vector) :: new
[150] if ( a%size /= b%size ) stop "Sizes differ in subtract Vector"
[151] allocate ( new%data(a%size) ) ; new%size = a%size
[152] new%data = a%data - b%data ; end function subtract Vector
[153]
[154] function values (name) result (array) ! accessor member
[155] type (Vector), intent(in) :: name
[156] real :: array(name%size)
[157] array = name % data ; end function values

The routine delete Vector is the manual constructor for this class. It has no optional arguments so
both arguments must be supplied, and it duplicates the constructor on line 31, but it uses the naming
convention preferred by the author.
[158]
[159] function Vector (length, values) result(name) ! constructor
[160] integer, intent(in) :: length ! array size
[161] real, target, intent(in) :: values(length) ! given array
[162] real, pointer :: pt to val(:) ! pointer to array
[163] type (Vector) :: name ! Vector to create
[164] integer :: get m ! allocate flag
[165] allocate ( pt to val (length), stat = get m ) ! allocate
[166] if ( get m /= 0 ) stop ’allocate error’ ! check
[167] pt to val = values ! dereference values
[168] name = Vector(length, pt to val) ! intrinsic constructor
[169] end function Vector
[170]
[171] function Vector max value (a) result (v) ! accessor member
[172] type (Vector), intent(in) :: a
[173] real :: v
[174] v = maxval ( a%data(1:a%size) ) ; end function Vector max value
[175]
[176] function Vector min value (a) result (v) ! accessor member
[177] type (Vector), intent(in) :: a
[178] real :: v
[179] v = minval ( a%data(1:a%size) ) ; end function Vector min value
[180]
[181] function Vector mult real(v, r) result(new) ! vec*real, overload *
[182] type (Vector), intent(in) :: v
[183] real, intent(in) :: r
[184] type (Vector) :: new ! new = v * r
[185] if ( v%size < 1 ) stop "Zero size in Vector mult real"
[186] new = Real mult Vector(r, v); end function Vector mult real
[187]
[188] end module class Vector

A first test of this class is given below along with comments that give the verifications of the members.
[ 1] ! Testing Vector Class Constructors & Operators
[ 2] include ’class Vector.f90’ ! see previous figure
[ 3] program check vector class
[ 4] use class Vector
[ 5] implicit none
[ 6]
[ 7] type (Vector) :: x, y, z
[ 8]
[ 9] ! test optional constructors: assign, and copy
[10] x = make Vector () ! single scalar zero
[11] write (*,’("made scalar x = ")’,advance=’no’); call list(x)
[12]
[13] call delete Vector (x) ; y = make Vector (4) ! 4 zeros
[14] write (*,’("made null y = ")’,advance=’no’); call list(y)
[15]
[16] z = make Vector (4, (/11., 12., 13., 14./) ) ! 4 non-zeros
[17] write (*,’("made full z = ")’,advance=’no’); call list(z)
[18] write (*,’("assign [ 31., 32., 33., 34. ] to x")’)
[19]
[20] x = assign( (/31., 32., 33., 34./) ) ! (4) non-zeros
[21] write (*,’("assigned x = ")’,advance=’no’); call list(x)
[22]
[23] x = Vector (4, (/31., 32., 33., 34./) ) ! 4 non-zeros
[24] write (*,’("public x = ")’,advance=’no’); call list(x)
[25] write (*,’("copy x to y =")’,advance=’no’)
[26] y = copy Vector (x) ; call list(y) ! copy
[27]
[28] ! test overloaded operators
[29] write (*,’("z * x gives ")’,advance=’no’); print *, z*x ! dot
[30] write (*,’("z + x gives ")’,advance=’no’); call list(z+x) ! add

c 2001 J.E. Akin 53

 [31] y = 25.6 ! real to vector
[32] write (*,’("y = 25.6 gives ")’,advance=’no’); call list(y)
[33] y = z ! equality
[34] write (*,’("y = z gives y as ")’, advance=’no’); call list(y)
[35] write (*,’("logic y == x gives ")’,advance=’no’); print *, y==x
[36] write (*,’("logic y == z gives ")’,advance=’no’); print *, y==z
[37]
[38] ! test destructor, accessors
[39] call delete Vector (y) ! destructor
[40] write (*,’("deleting y gives y = ")’,advance=’no’); call list(y)
[41] print *, "size of x is ", length (x) ! accessor
[42] print *, "data in x are [", values (x), "]" ! accessor
[43] write (*,’("2. times x is ")’,advance=’no’); call list(2.0*x)
[44] write (*,’("x times 2. is ")’,advance=’no’); call list(x*2.0)
[45] call delete Vector (x); call delete Vector (z) ! clean up
[46] end program check vector class
[47] ! Running gives the output: ! made scalar x = [0]
[48] ! made null y = [0, 0, 0, 0] ! made full z = [11, 12, 13, 14]
[49] ! assign [31, 32, 33, 34] to x ! assigned x = [31, 32, 33, 34]
[50] ! public x = [31, 32, 33, 34] ! copy x to y = [31, 32, 33, 34]
[51] ! z * x gives 1630 ! z + x gives [42, 44, 46, 48]
[52] ! y = 256 gives [256000004] ! y = z, y = [11, 12, 13, 14]
[53] ! logic y == x gives F ! logic y == z gives T
[54] ! deleting y gives y = [] ! size of x is 4
[55] ! data in x : [31, 32, 33, 34] ! 2 times x is [62, 64, 66, 68]
[56] ! x times 2 is [62, 64, 66, 68]

Having tested the vector class we will now use it in some typical vector operations. We want a program
that will work with arrays of vectors to read in the number of vectors. The array of vectors will use an
automatic storage mode. That could be risky because if the system runs out of memory we get a fatal
error message and the run aborts. If we made the alternate choice of allocatable arrays then we could
check the allocation status and have a chance (but not a good chance) of closing down the code is some
”friendly” manner. Once the code reads the number of vectors then for each one it reads the number of
components and the the component values. After testing some simple vector math we compute a more
complicated result know as the orthonormal basis for the given set of vectors:
[ 1] ! Test Vector Class Constructors, Operators and Basis
[ 2] include ’class Vector.f’
[ 3]
[ 4] program check basis ! demonstrate a typical Vector class
[ 5] use class Vector
[ 6] implicit none
[ 7]
[ 8] interface
[ 9] subroutine testing basis (N V)
[ 10] integer, intent(in) :: N V
[ 11] end subroutine testing basis
[ 12] end interface
[ 13]
[ 14] print *, "Test automatic allocate, deallocate"
[ 15] print *, " " ; read *, N V
[ 16] print *, "The number of vectors to be read is: ", N V
[ 17] call testing basis ( N V) ! to use automatic arrays
[ 18] end program check basis
[ 19]
[ 20] subroutine testing basis (N V)
[ 21] ! test vectors AND demo automatic allocation/deallocation
[ 22] use class Vector
[ 23]
[ 24] integer, intent(in) :: N V
[ 25] type (Vector) :: Input(N V) ! automatic array
[ 26] type (Vector) :: Ortho(N V) ! automatic array
[ 27] integer :: j
[ 28] real :: norm
[ 29]
[ 30] interface
[ 31] subroutine orthonormal basis (Input, Ortho, N given)
[ 32] use class Vector
[ 33] type (Vector), intent(in) :: Input(N given)
[ 34] type (Vector), intent(out) :: Ortho(N given)
[ 35] integer, intent(in) :: N given
[ 36] end subroutine orthonormal basis
[ 37] end interface
[ 38]
[ 39] print *, " " ; print *, "The given ", N V, " vectors:"
[ 40] do j = 1, N V
[ 41] call read Vector ( Input(j) )
[ 42] call list ( Input(j) )
[ 43] end do ! for j
[ 44]
[ 45] print *, " "

c 2001 J.E. Akin 54

 [ 46] print *, "The Orthogonal Basis of the original set is:"
[ 47]
[ 48] call orthonormal basis (Input, Ortho, N V)
[ 49] do j = 1, N V ! list new orthogonal basis
[ 50] call list ( Ortho(j) )
[ 51] end do ! for j
[ 52]
[ 53] ! use vector class features & operators
[ 54] print *, ’ ’ ; print *,"vector 1 + vector 2 = "
[ 55] call list (Input(1)+Input(2))
[ 56] print *,"vector 1 - vector 2 = "
[ 57] call list (Input(1)-Input(2))
[ 58] print *,"vector 1 dot vector 2 = ", Input(1)*Input(2)
[ 59] print *,"vector 1 * 3.5 = "
[ 60] call list (3.5*Input(1))
[ 61] norm = sqrt ( dot Vector( Input(1), Input(1) ))
[ 62] print *,"norm(vector 1) = ", norm
[ 63] print *,"normalized vector 1 = "
[ 64] call list (normalize Vector(Input(1)))
[ 65] print *,"max(vector 1) = ", vector max value (Input(1))
[ 66] print *,"min(vector 1) = ", vector min value (Input(1))
[ 67] print *,"length of vector 1 = ", length ( Input(1) )
[ 68] end subroutine testing basis
[ 69]
[ 70] subroutine orthonormal basis (Input, Ortho, N given)
[ 71] ! Find Orthonormal Basis of a Set of Vector Classes
[ 72] use class Vector
[ 73] !*******************************************************
[ 74] ! =, -, +, * are overloaded operators from class Vector
[ 75] !*******************************************************
[ 76]
[ 77] type (Vector), intent(in) :: Input(N given)
[ 78] type (Vector), intent(out) :: Ortho(N given)
[ 79] integer, intent(in) :: N given
[ 80] integer :: i, j ! loops
[ 81] real :: dot
[ 82] do i = 1, N given ! original set of vectors
[ 83] Ortho(i) = Input(i) ! copy input vector class
[ 84] do j = 1, i ! for previous copies
[ 85] dot = dot Vector(Ortho(i), Ortho(j))
[ 86] Ortho(i) = Ortho(i) - (dot*Ortho(j))
[ 87] end do ! for j
[ 88] Ortho(i) = normalize Vector ( Ortho(i) )
[ 89] end do ! over i
[ 90] end subroutine orthonormal basis
[ 91]
[ 92] ! Compiling and inputting :
[ 93] ! 4
[ 94] ! 3 0.625 0 0
[ 95] ! 3 7.5 3.125 0
[ 96] ! 3 13.25 -7.8125 6.5
[ 97] ! 3 14.0 3.5 -7.5
[ 98] ! Gives:
[ 99] ! Test automatic allocate, deallocate
[100] !
[101] ! The number of vectors to be read is: 4
[102] ! The given 4 vectors:
[103] ! [ 0.6250 0.0000 0.0000 ]
[104] ! [ 7.5000 3.1250 0.0000 ]
[105] ! [ 13.2500 -7.8125 6.5000 ]
[106] ! [ 14.0000 3.5000 -7.5000 ]
[107] !
[108] ! The Orthogonal Basis of the original set is:
[109] ! [ 1.0000 0.0000 0.0000 ]
[110] ! [ 0.0000 -1.0000 0.0000 ]
[111] ! [ 0.0000 0.0000 -1.0000 ]
[112] ! [ 0.0000 0.0000 0.0000 ]
[113] !
[114] ! vector 1 + vector 2 = [ 8.1250 3.1250 0.0000 ]
[115] ! vector 1 - vector 2 = [-6.8750 -3.1250 0.0000 ]
[116] ! vector 1 dot vector 2 = 4.6875
[117] ! vector 1 * 3.5 = [ 2.1875 0.0000 0.0000 ]
[118] ! norm(vector 1) = 0.6250
[119] ! normalized vector 1 = [ 1.0000 0.0000 0.0000 ]
[120] ! max(vector 1) = 0.6250
[121] ! min(vector 1) = 0.0000
[122] ! length of vector 1 = 3

c 2001 J.E. Akin 55

C.6 Problem 3.5.4 : Creating a sparse vector class
This class begins like the previous Vector class except that we must add a row entry (line 4) for each data
value entry (line 5). This is done for efficiency since we expect most values in sparse vectors to be zero
(and hence their name). The attribute non zero is the size of both rows and values.
[ 1] module class sparse Vector
[ 2] implicit none
[ 3] type sv ! a sparse vector
[ 4] integer :: non zeros
[ 5] integer, pointer :: rows(:)
[ 6] real, pointer :: values(:)
[ 7] end type
[ 8]

The overloading process is similar, but now we will see that much more logic is required to deal with the
zero entries and new zeros created by addition or multiplication.
[ 8] interface assignment (=)
[ 9] module procedure equal Vector ; end interface
[ 10] interface operator (.dot.) ! define dot product operator
[ 11] module procedure dot Vector ; end interface
[ 12] interface operator (==) ! Boolean equal to
[ 13] module procedure is equal to ; end interface
[ 14] interface operator (*) ! term by term product
[ 15] module procedure el by el Mult, real mult Sparse
[ 16] module procedure Sparse mult real
[ 17] end interface
[ 18] interface operator (-) ! for sparse vectors
[ 19] module procedure Sub Sparse Vectors ; end interface
[ 20] interface operator (+) ! for sparse vectors
[ 21] module procedure Sum Sparse Vectors ; end interface
[ 22]
[ 23] contains ! operators and functionality

In the following constructor for the class note that both the pointer array attributes are allocated (line 32)
the same amount of storage in memory. One should also include the allocation status flag here and checks
its value to raise a possible exception (as seen in lines 41-46).
[ 24] subroutine make Sparse Vector (s,n,r,v)
[ 25] ! allows zero length vectors
[ 26] type (sv) :: s ! name
[ 27] integer, intent(in) :: n ! size
[ 28] integer, intent(in) :: r(n) ! rows
[ 29] real, intent(in) :: v(n) ! values
[ 30] if ( n < 0 ) stop &
[ 31] "Error, negative rows in make Sparse Vector"
[ 32] allocate (s%rows(n), s%values(n))
[ 33] s%non zeros = n ! copy size
[ 34] s%rows = r ! row array assignment
[ 35] s%values = v ! value array assignment
[ 36] end subroutine make Sparse Vector
[ 37]

This is really a destructor. Again, it is incomplete because the integer array size was not made allocatable
for simplicity.
[ 38] subroutine delete Sparse Vector (s)
[ 39] type (sv) :: s ! name of sparse vector
[ 40] integer :: error ! deallocate status flag, 0 no error
[ 41] deallocate (s%rows, s%values, stat = error) ! memory released
[ 42] if ( error == 0 ) then
[ 43] s%non zeros = 0 ! reset size
[ 44] else ! never created
[ 45] stop "Sparse vector to delete does not exist"
[ 46] end if ; end subroutine delete Sparse Vector
[ 47]

This creates a user defined operator call .dot. to be applied to sparse vectors.
[ 48] function dot Vector (u, v) result (d) ! defines .dot.
[ 49] ! dot product of sparse vectors
[ 50] type (sv), intent(in) :: u, v ! sparse vectors
[ 51] type (sv) :: w ! sparse vector, temporary
[ 52] real :: d ! dot product value
[ 53] d = 0.0 ! default
[ 54] if ( u%non zeros < 1 .or. v%non zeros < 1 ) return ! null
[ 55] w = el by el Mult (u, v) ! element by element sparse product
[ 56] if ( w%non zeros > 0 ) &
[ 57] d = sum( w%values(1:w%non zeros) ) ! summed
[ 58] call delete Sparse Vector (w) ! delete temp

c 2001 J.E. Akin 56

 [ 59] end function dot Vector
[ 60]

The above dot Vector is more complicated in this format because it is likely that stored non-zero
values will be multiplied by (unstored) zeros. Thus, the real work is done in the following member
function that employs Boolean logic. The terms for the summation that creates the scalar dot product are
first computed in a full vector equal in length to the minimum row number given. Observe that its size
is established through the use of the min intrinsic, acting on the two given sizes, within the dimension
attribute for the full array (lines 67,68). Three logical arrays (line 68) are used as “masks” which are
true when a non-zero exists in the corresponding row of their associated sparse vector (down to the
minimum row cited above). The three logical vectors are initialized in lines 77 to 92. That process ends
with the third vector being created as a Boolean product (line 91) and the maximum possible number of
non-zero products is found from the count intrinsic (line 92).
It is also important to note that the working space vector full is an automatic array and memory
for it is automatically allocated for it each time the function is called. It could be an extremely long
vector and thus it is possible (but not likely) that there would not be enough memory available. Then the
system would abort with an error message. To avoid that possibility one could have declared full to be
an allocatable vector and then allocate its memory by using a similar min construct. That allocation
request should (always) include the STAT flag so that if the memory allocation fails it would be possible
to issue an exception to try to avoid a fatal crash of the system (not likely).
[ 61] function el by el Mult (u, v) result (w) ! defines * operator
[ 62] ! element by element product of sparse vectors: 0 * real ?
[ 63] type (sv), intent(in) :: u, v ! given vectors
[ 64] type (sv) :: w ! new vector
[ 65] real :: full( min( u%rows(u%non zeros), & ! automatic
[ 66] & v%rows(v%non zeros) ) ) ! workspace
[ 67] logical, dimension( min( u%rows(u%non zeros), &
[ 68] v%rows(v%non zeros))) :: u m, v m, w m ! logical product masks
[ 69] integer :: j, k, last, n, row
[ 70] ! is either u or v null ?
[ 71] if ( u%non zeros < 1 .or. v%non zeros < 1 ) then ! w is null
[ 72] allocate ( w%rows(0), w%values(0) )
[ 73] w%non zeros = 0
[ 74] return ! a null sparse vector
[ 75] end if ! no calculation necessary
[ 76]
[ 77] ! Initialize logic masks
[ 78] last = min( u%rows(u%non zeros), v%rows(v%non zeros) ) ! max size
[ 79] u m = .false. ! assume no contributions
[ 80] do j = 1, size(u%rows)
[ 81] row = u%rows(j) ! get row number to flag
[ 82] if ( row > last ) exit ! j loop
[ 83] u m(row) = .true. ! possible contribution
[ 84] end do ! to initalize u mask
[ 85] v m = .false. ! assume no contributions
[ 86] do j = 1, size(v%rows)
[ 87] row = v%rows(j) ! get row number to flag
[ 88] if ( row > last ) exit ! j loop
[ 89] v m(row) = .true. ! possible contribution
[ 90] end do ! to initalize v mask
[ 91] w m = (u m .and. v m ) ! Boolean product logic
[ 92] n = count ( w m ) ! count possible products
[ 93] ! if ( n == 0 ) print *,"Warning: zero length sparse" ! debug
[ 94]

The vector full is set to zero (line 96) and comparison DO loops (lines 97,101) over the two given
vectors are minimized (lines 100,103) by testing where the mask vector w m is true (thereby indicating
a non-zero product). When all the products are stored in the full vector it is converted to the sparse
vector storage mode (line 109) for release as the return result. Because full is an automatic array
its memory is automatically released when the function is exited.
[ 95] ! Fill the product workspace, full
[ 96] full = 0.0 ! initialize
[ 97] do j = 1, size(u%rows) ! loop over u
[ 98] row = u%rows(j) ! row in u
[ 99] if ( row > last ) exit ! this loop in u ! past end of w
[100] if ( .not. w m(row) ) cycle ! to next j ! not in product
[101] do k = 1, size(v%rows) ! loop over v
[102] if ( v%rows(k) > last ) exit ! this loop ! past end of w
[103] if ( .not. w m(v%rows(k)) ) cycle ! to k+1 ! not in product
[104] if ( row == v%rows(k) ) then ! same row, u & v
[105] full(row) = u%values(j)*v%values(k) ! get product

c 2001 J.E. Akin 57

 [106] end if
[107] end do ! on k in v
[108] end do ! on j in u
[109] w = Vector To Sparse (full) !delete any zeros
[110] end function el by el Mult ! deletes full & 3 masks
[111]

The operator overloading members are given with the next function (line 112) as well as in lines 140,
231, and 320.
[112] subroutine equal Vector (new, s) ! overload =
[113] type (sv), intent(inout) :: new
[114] type (sv), intent(in) :: s
[115] allocate ( new%rows(s%non zeros) )
[116] allocate ( new%values(s%non zeros) )
[117] new%non zeros = s%non zeros
[118] if ( s%non zeros > 0 ) then
[119] new%rows (1:s%non zeros) = s%rows (1:s%non zeros) ! array copy
[120] new%values(1:s%non zeros) = s%values(1:s%non zeros) ! copy
[121] end if ; end subroutine equal Vector
[122]
[123] function get element (name, row) result (v)
[124] type (sv), intent(in) :: name ! sparse vector
[125] integer, intent(in) :: row ! row in sparse vector
[126] integer :: j ! loops
[127] real :: v ! value at row
[128] v = 0.0 ! default
[129] if ( row < 1 ) stop "Invalid row number, get element"
[130] if ( name%non zeros < 1 ) return ! not here
[131] if ( row > name%rows(name%non zeros) ) return ! not here
[132] do j = 1, name%non zeros
[133] if ( row == name%rows(j) ) then
[134] v = name%values(j) ! found the value
[135] return ! search done
[136] end if ! in the vector
[137] end do ! over possible values
[138] end function get element
[139]
[140] function is equal to (a, b) result (t or f) ! define ==
[141] type (sv), intent(in) :: a, b ! two sparse vectors
[142] logical :: t or f
[143] integer :: i ! loops
[144] t or f = .true. ! default
[145] if ( a%non zeros == b%non zeros ) then ! also check values
[146] do i = 1, a%non zeros ! or use count function for simplicity
[147] if (a%rows(i) /= b%rows(i) .or. &
[148] a%values(i) /= b%values(i)) then
[149] t or f = .false. ! because rows and/or values differ
[150] return ! no additional checks needed
[151] end if ! same values
[152] end do ! over sparse rows
[153] else ! sizes differ so vectors must be different
[154] t or f = .false.
[155] end if ! sizes match
[156] end function is equal to
[157]
[158] function largest index (s) result(row)
[159] type (sv), intent(in) :: s ! sparse vector
[160] integer :: row ! last non-zero in full vector
[161] integer :: j ! loops
[162] row = 0 ! initalize
[163] if ( s%non zeros < 1 ) return ! null vector
[164] do j = s%non zeros, 1, -1 ! loop backward
[165] if ( s%values(j) /= 0.0 ) then ! last non-zero term
[166] row = s%rows(j) ! actual row number
[167] return ! search done
[168] end if
[169] end do
[170] end function largest index
[171]
[172] function length (name) result (n)
[173] type (sv), intent(in) :: name
[174] integer :: n
[175] n = name % non zeros ! read access to size, if private
[176] end function length
[177]

Once again we observe that the next two functions employ the colon operator (lines 185,196,199,201)
to avoid explicit serial loops which would make them faster on certain vector and parallel computers.
[178] function norm (name) result (total)
[179] type (sv), intent(in) :: name
[180] real :: total
[181] if ( name%non zeros < 1 ) then

c 2001 J.E. Akin 58

 [182] ! print *, "Warning: empty vector in norm"
[183] total = 0.0
[184] else
[185] total = sqrt( sum( name%values(1:name%non zeros)**2 ))
[186] end if ! a null vector
[187] end function norm
[188]
[189] function normalize Vector (s) result (new)
[190] type (sv), intent(in) :: s
[191] type (sv) :: new
[192] real :: total, epsilon = 1.e-6
[193] allocate ( new%rows (s%non zeros) )
[194] allocate ( new%values(s%non zeros) )
[195] new%non zeros = s%non zeros ! copy size
[196] new%rows(1:s%non zeros) = s%rows(1:s%non zeros) ! copy rows
[197] total = sqrt( sum( s%values(1:s%non zeros)**2 )) ! norm
[198] if ( total <= epsilon ) then ! divide by 0 ?
[199] new%values(1:s%non zeros) = 0.d0 ! set to zero
[200] else ! or real values
[201] new%values(1:s%non zeros) = s%values(1:s%non zeros)/total
[202] end if ! division by zero
[203] end function normalize Vector
[204]
[205] subroutine pretty (s) ! print all values if space allows
[206] type (sv), intent(in) :: s ! sparse vector
[207] integer, parameter :: limit = 20 ! for print size
[208] integer :: n
[209] real :: full( s%rows(s%non zeros) ) ! temp
[210] n = s%non zeros
[211] if ( s%non zeros < 1 .or. s%rows(s%non zeros) > limit ) then
[212] print *, "Wrong size to pretty print"
[213] else
[214] full = 0. ! initialize to zero
[215] if ( n > 0 ) full(s%rows) = s%values ! array copy non zeros
[216] print *,"[", full,"]" ! pretty print
[217] end if ; end subroutine pretty ! automatic deallocate of full
[218]
[219] subroutine read Vector (name) ! sparse vector data on unit 1
[220] type (sv), intent(inout) :: name
[221] integer :: length, j
[222] read (1,’(i1)’, advance = ’no’) length
[223] if ( length <= 0 ) stop "Invalid length in read Vector"
[224] name % non zeros = length
[225] allocate ( name % rows (length) )
[226] allocate ( name % values (length) )
[227] read (1,*) ( name%rows(j), name%values(j), j = 1, length)
[228] name%rows = name%rows + 1 ! default to 1 not 0 in F90
[229] end subroutine read Vector
[230]
[231] function real mult Sparse (a, b) result (new)
[232] ! scalar * vector
[233] real, intent(in) :: a
[234] type (sv), intent(in) :: b
[235] type (sv) :: new
[236] allocate ( new%rows (b%non zeros) )
[237] allocate ( new%values(b%non zeros) )
[238] new%non zeros = b%non zeros
[239] if ( b%non zeros < 1 ) then
[240] print *, "Warning: zero size in real mult Sparse "
[241] else ! copy array components
[242] new%rows (1:b%non zeros) = b%rows (1:b%non zeros)
[243] new%values(1:b%non zeros) = a * b%values(1:b%non zeros)
[244] end if ! null vector
[245] end function real mult Sparse
[246]
[247] function rows of (s) result(n) ! copy rows array of s
[248] type (sv) :: s ! sparse vector
[249] integer :: n(s%non zeros) ! standard array
[250] if ( s%non zeros < 1 ) stop "No rows to extract, rows of"
[251] n = s%rows ! array copy
[252] end function rows of
[253]
[254] subroutine set element (s, row, value)
[255] ! Set, or insert, value into row of a sparse vector, s
[256] type (sv), intent(inout) :: s ! sparse vector
[257] integer, intent(in) :: row ! full vector row
[258] real, intent(in) :: value ! full vector value
[259] type (sv) :: new ! workspace
[260] logical :: found ! true if row exists
[261] integer :: j, where ! loops, locator
[262] found = .false. ! initialize
[263] where = 0 ! initialize
[264] do j = 1, s%non zeros
[265] if ( s%rows(j) == row ) then ! found it
[266] s%values(j) = value ! value changed

c 2001 J.E. Akin 59

 [267] return ! no insert needed
[268] end if
[269] if ( s%rows(j) > row ) then
[270] where = j ! insert before j
[271] exit ! the loop search
[272] else ! s%rows(j) < row, may be next or last
[273] where = j + 1
[274] end if
[275] end do ! over current rows in s
[276] if ( .not. found ) then ! expand and insert at where
[277] if ( where == 0 ) stop "Logic error, set element"
[278] new%non zeros = s%non zeros + 1
[279] allocate ( new%rows (new%non zeros) )
[280] allocate ( new%values(new%non zeros) )
[281] ! copy preceeding rows
[282] if ( where > 1 ) then ! copy to front of new
[283] new%rows (1:where-1) = s%rows (1:where-1) ! array copy
[284] new%values(1:where-1) = s%values(1:where-1) ! array copy
[285] end if ! copy to front of new
[286] ! insert, copy following rows of s
[287] new%rows (where ) = row ! insert
[288] new%values(where ) = value ! insert
[289] new%rows (where+1:) = s%rows (where:) ! array copy
[290] new%values(where+1:) = s%values(where:) ! array copy
[291] ! deallocate s, move new to s, deallocate new
[292] call delete Sparse Vector (s) ! delete s
[293] call equal Vector (s, new) ! s <- new
[294] call delete Sparse Vector (new) ! delete new
[295] end if ! an insert is required
[296] end subroutine set element
[297]
[298] subroutine show (s) ! alternating row number and value
[299] type (sv) :: s ! sparse vector
[300] integer :: j, k ! implied loops
[301] k = length (s)
[302] if ( k == 0 ) then
[303] print *, k ; else ; ! print in C++ style rows
[304] print *, k, ( (s%rows(j)-1), s%values(j), j = 1, k )
[305] end if ; end subroutine show
[306]
[307] subroutine show r v (s) ! all rows then all values
[308] type (sv) :: s ! sparse vector
[309] print *, "Vector has ", s%non zeros, " non zero terms."
[310] if ( s%non zeros > 0 ) then
[311] print *, "Rows: ", s%rows - 1 ! to look like C++
[312] print *, "Values: ", s%values
[313] end if ; end subroutine show r v
[314]
[315] function size of (s) result(n)
[316] type (sv) :: s
[317] integer :: n
[318] n = s%non zeros ; end function size of
[319]
[320] function Sparse mult real (a, b) result (new)
[321] ! vector * scalar
[322] real, intent(in) :: b
[323] type (sv), intent(in) :: a
[324] type (sv) :: new
[325] new = real mult Sparse ( b, a) ! reverse the order
[326] end function Sparse mult real
[327]

In the following subtraction and addition functions we again note that sparse terms with the same
values but opposite signs can result in new zero terms in the resulting vector. A temporary automatic
workspace vector, full, is used to hold the preliminary results. In this case it must have a size that is the
maximum of the two given vectors. Thus, the max intrinsic is employed in its dimension attribute (lines
331,344) which is opposite the earlier multiplication example (line 65).
[328] function Sub Sparse Vectors (u, v) result (w) ! defines -
[329] type (sv), intent(in) :: u, v
[330] type (sv) :: w
[331] real :: full( max( u%rows(u%non zeros), & ! automatic
[332] & v%rows(v%non zeros) ) ) ! workspace
[333] if ( u%non zeros <= 0 ) stop "First vector doesn’t exist"
[334] if ( v%non zeros <= 0 ) stop "Second vector doesn’t exist"
[335] full = 0.0 ! set to zero
[336] full(u%rows) = u%values ! copy first values
[337] full(v%rows) = full(v%rows) - v%values ! less second values
[338] w = Vector To Sparse (full) ! delete any zeros
[339] end function Sub Sparse Vectors ! automatically deletes full
[340]
[341] function Sum Sparse Vectors (u, v) result (w) ! defines +
[342] type (sv), intent(in) :: u, v

c 2001 J.E. Akin 60

 [343] type (sv) :: w
[344] real :: full( max( u%rows(u%non zeros), & ! automatic
[345] & v%rows(v%non zeros) ) ) ! workspace
[346] if ( u%non zeros <= 0 ) stop "First vector doesn’t exist"
[347] if ( v%non zeros <= 0 ) stop "Second vector doesn’t exist"
[348] full = 0. ! set to zero
[349] full(u%rows) = u%values ! copy first values
[350] full(v%rows) = full(v%rows) + v%values ! add second values
[351] w = Vector To Sparse (full) ! delete any zeros
[352] end function Sum Sparse Vectors ! automatically deletes full
[353]
[354] function values of (s) result(v) ! copy values of s
[355] type (sv) :: s ! sparse vector
[356] real :: v(s%non zeros) ! standard array
[357] if ( s%non zeros < 1 ) &
[358] stop "No values to extract, in values of"
[359] v = s%values ! array copy
[360] end function values of
[361]
[362] function Vector max value (a) result (v)
[363] type (sv), intent(in) :: a
[364] real :: v
[365] v = maxval (a%values(1:a%non zeros)) ! intrinsic function
[366] ! is it a sparse vector with a false negative maximum ?
[367] if ( a%non zeros < a%rows(a%non zeros) .and. v < 0. ) v = 0.0
[368] end function Vector max value
[369]
[370] function Vector min value (a) result (v)
[371] type (sv), intent(in) :: a
[372] real :: v
[373] v = minval ( a%values(1:a%non zeros) ) ! intrinsic function
[374] ! is it a sparse vector with a false positive minimum ?
[375] if ( a%non zeros < a%rows(a%non zeros) &
[376] .and. v > 0. ) v = 0.0
[377] end function Vector min value
[378]

This function is invoked several times in other member functions. It simply accepts a standard (dense)
vector and converts it to the sparse storage mode in the return result.
[379] function Vector To Sparse (full) result (sparse)
[380] real, intent(in) :: full(:) ! standard array
[381] type (sv) :: sparse ! sparse vector copy
[382] integer :: j, n, number ! loops and counters
[383] n = count ( full /= 0.0 ) ! count non zeros
[384] ! if ( n == 0 ) print *, "Warning: null full vector "
[385] allocate ( sparse%rows(n), sparse%values(n) )
[386] sparse%non zeros = n ! sparse size
[387] number = 0 ! non zeros inserted
[388] do j = 1, size(full)
[389] if ( full(j) == 0.0 ) cycle ! to next j value
[390] number = number + 1 ! non zeros inserted
[391] sparse%rows(number) = j ! row number in full
[392] sparse%values(number) = full(j) ! value
[393] if ( number == n ) exit ! all non zeros found
[394] end do ; end function Vector To Sparse
[395]
[396] function zero sparse () result (s)
[397] type (sv) :: s ! create sparse null vector
[398] s%non zeros = 0
[399] allocate (s%rows(0), s%values(0)); end function zero sparse
[400] end module class sparse Vector

C.7 Problem 4.11.1 : Count the lines in an external file

[ 1] function inputCount(unit) result(linesOfInput)

[ 2] !-------------------------------------------------------------
[ 3] ! takes a file number, counts the number of lines in that
[ 4] ! file, and returns the number of lines.
[ 5] !-------------------------------------------------------------
[ 6] implicit none
[ 7] integer, intent(in) :: unit ! file unit number
[ 8] integer :: linesOfInput ! result
[ 9] integer ioResult ! system I/O action error code
[10] character temp ! place to hold the character read
[11]
[12] rewind (unit) ! go to the front of the file
[13] linesOfInput = 0 ! initially, there are 0 lines
[14]
[15] do ! Until iostat says we’ve hit the end of file
[16] read (unit,’(A)’, iostat = ioResult) temp ! one char

c 2001 J.E. Akin 61

 [17]
[18] if ( ioResult == 0 ) then ! there were no errors
[19] linesOfInput = linesOfInput + 1 ! increment lines
[20] else if ( ioResult < 0 ) then ! we’ve hit end-of-file
[21] exit ! so exit this loop.
[22] else ! ioResult is positive, which is a user error
[23] write (*,*) ’inputCount: no data at unit =’, unit
[24] stop ’user read error’
[25] end if
[26] end do
[27] rewind(unit) ! go to the front of the file
[28] end Function inputCount

C.8 Problem 4.11.3 : Computing CPU time useage

While this is mainly designed to show the use of the module tic toc you should note that the intrinsic
way of printing a date or time is not “pretty” and could be easily improved.
[ 1] program watch
[ 2] ! -------------------------------------------------
[ 3] ! Exercise DATE AND TIME and SYSTEM CLOCK functions.
[ 4] ! -------------------------------------------------
[ 5] use tic toc
[ 6] implicit none
[ 7] character* 8 :: the date
[ 8] character*10 :: the time
[ 9] integer :: j, k
[10] !
[11] call date and time ( DATE = the date )
[12] call date and time ( TIME = the time )
[13] print *, ’The date is ’, the date, &
[14] & ’ and the time is now ’, the time
[15] ! Display facts about the system clock.
[16] print *, ’ ’
[17] call system clock ( COUNT RATE = rate )
[18] print *, ’System clock runs at ’, rate,&
[19] & ’ ticks per second’
[20] !
[21] ! Call the system clock to start an execution timer.
[22] call tic
[23] !
[24] ! call run the job, or test with next 3 lines
[25] do k = 1, 9999
[26] j = sqrt ( real(k*k) )
[27] end do
[28] ! Stop the execution timer and report execution time.
[29] print *, ’ ’
[30] print *, ’Job took ’, toc (), ’ seconds to execute.’
[31] end program watch ! Running gives
[32] ! The date is 19980313 and the time is now 171837.792
[33] ! System clock runs at 100 ticks per second
[34] ! Job took 0.9999999776E-02 seconds to execute.

C.9 Problem 4.11.4 : Converting a string to upper case

The change from the to lower should be obvious here. It seems desirable to place these two routines,
and others that deal with strings into a single strings utility module.
[ 1] function to upper (string) result (new string) ! like C
[ 2] ! -------------------------------------------------------------
[ 3] ! Convert a string or character to upper case
[ 4] ! (valid for ASCII or EBCDIC processors)
[ 5] ! -------------------------------------------------------------
[ 6] implicit none
[ 7] character (len = *), intent(in) :: string ! unknown length
[ 8] character (len = len(string)) :: new string ! same length
[ 9] character (len = 26), parameter :: &
[10] UPPER = ’ABCDEFGHIJKLMNOPQRSTUVWXYZ’, &
[11] lower = ’abcdefghijklmnopqrstuvwxyz’
[12] integer :: k ! loop counter
[13] integer :: loc ! position in alphabet
[14] new string = string ! copy everything
[15] do k = 1, len(string) ! to change letters
[16] loc = index ( lower, string(k:k)) ! locate
[17] if (loc /= 0 ) new string(k:k) = UPPER(loc:loc) ! convert
[18] end do ! over string characters
[19] end function to upper

c 2001 J.E. Akin 62

C.10 Problem 4.11.8 : Read two values from each line of an external file

[ 1] subroutine readData (inFile, lines, x, y)

[ 2] ! ------------------------------------------------------
[ 3] ! Take a file number, the number of lines to be read,
[ 4] ! and put the data into the arrays x and y
[ 5] ! ------------------------------------------------------
[ 6] ! inFile is unit number to be read
[ 7] ! lines is number of lines in the file
[ 8] ! x is independent data
[ 9] ! y is dependent data
[10] implicit none
[11] integer, intent(in) :: inFile, lines
[12] real, intent(out) :: x(lines), y(lines)
[13] integer :: j
[14]
[15] rewind (inFile) ! go to front of the file
[16] do j = 1, lines ! for the entire file
[17] read (inFile, *) x(j), y(j) ! get the x and y values
[18] end do ! over all lines
[19] end subroutine readData

C.11 Problem 4.11.14 : Two line least square fits

The extension of the single-line least squares fit shown in Fig. 4.21 is rather straightforward in that we
will call subroutine lsq fit multiple times. In line 37 we first call it in case a single-line fit may be more
accurate than the expected two-line fit.
[ 1] program two line lsq fit
[ 2] !------------------------------------------------------
[ 3] ! Best two-line linear least-squares fit of data in
[ 4] ! file specified by the user, and split in two sets
[ 5] !------------------------------------------------------
[ 6] implicit none
[ 7] real, allocatable :: x (:) ! independent data
[ 8] real, allocatable :: y (:) ! dependent data
[ 9]
[ 10] real :: fit(3), fit1(3), fit2(3) ! error results
[ 11] real :: left(3), right(3) ! best results
[ 12] real :: error ! current error
[ 13] real :: error min ! best error
[ 14] integer :: split ! best division
[ 15]
[ 16] integer, parameter :: filenumber = 1 ! input unit
[ 17] character (len = 64) :: filename ! input file
[ 18] integer :: lines ! of input
[ 19] integer :: inputCount, j ! loops
[ 20]
[ 21] ! Get the name of the file containing the data.
[ 22] write (*, *) ’Enter the data input filename:’
[ 23] read (*, *) filename
[ 24]
[ 25] ! Open that file for reading.
[ 26] open (unit = filenumber, file = filename)
[ 27]
[ 28] ! Find the number of lines in the file
[ 29] lines = inputCount (filenumber)
[ 30] write (*, *) ’There were ’,lines,’ records read.’
[ 31]
[ 32] ! Allocate that many entries in the x and y array
[ 33] allocate (x(lines), y(lines))
[ 34] call read xy file (filenumber, lines, x, y) ! Read data
[ 35] close (filenumber)
[ 36]
[ 37] call lsq fit (lines, x, y, fit) ! single line fit
[ 38] print *, "Single line fit"
[ 39] print *, "the slope is ", fit(1)
[ 40] print *, "the intercept is ", fit(2)
[ 41] print *, "the error is ", fit(3)
[ 42]

After that we want to try all the reasonable choices for breading the data set into two adjacent regions
that are each to be fit with a different straight line. Trial variables were defined in lines 10 and 12, while
the best results found are in variables declared in lines 11, 13, and 14. Note that on line 48 we have
required that at least three points be used to define an approximate straight line. If we allowed two points
to be employed we would get a false (or misleading) indication of zero error for such a choice. Thus, in

c 2001 J.E. Akin 63

line 48 we begin a loop over all possible sets of three or more data points and call lsq fit for each of the
two segments, as seen in lines 50 and 51.
[ 43] ! Loop to determine the mean squared error for each
[ 44] ! of the possible two divisions of the data
[ 45] !
[ 46] error min = HUGE(error min) ! initialize the error min
[ 47] split = 3 ! initialize split point
[ 48] do j = 3, lines-3 ! 3 pts to approximate a line
[ 49] ! least-squares fit of two data subsets
[ 50] call lsq fit (j, x(1:j), y(1:j), fit1)
[ 51] call lsq fit (lines-j, x(j+1:lines), y(j+1:lines), fit2)
[ 52] error = fit1(3) + fit2(3)
[ 53]

In splitting up the two data regions not that it was not necessary to copy segments of the independent
and dependent data. Instead the colon operator, or implied do loops, were used in lines 50 and 51 to pass
vectors with j and (lines – j) entries, respectively to the two calls to lsq fit. After combining the two
errors, in line 52, we update the current best choice for the data set division point in lines 55 through 58.
[ 54] ! does this division gives you a smaller error ?
[ 55] if ( error < error min ) then
[ 56] error min = error ; split = j
[ 57] left = fit1 ; right = fit2
[ 58] end if ! current best choice
[ 59] end do ! of split choices

After we exit the loop, at line 59, we simply list the best results obtained. In line 73 we have also
deallocated the data arrays even though it is just a formality at this point since all memory is released at the
program terminates immediately afterwards. Had this been a subroutine or function then we would need
to be sure that allocated variables are released when their access scope has terminated. Later versions of
Fortran will do that for you, but good programmers should keep up with memory allocations.
[ 60] ! Display the results
[ 61] print *, "Two line best fit; combined error is ", error min
[ 62] print *, "Best division of the data is:"
[ 63] print *, "data(:j), data(j+1:), where j = ", split
[ 64] print *, "Left line fit:"
[ 65] print *, "the slope is ", left(1)
[ 66] print *, "the intercept is ", left(2)
[ 67] print *, "the error is ", left(3)
[ 68] print *, "Right line fit:"
[ 69] print *, "the slope is ", right(1)
[ 70] print *, "the intercept is ", right(2)
[ 71] print *, "the error is ", right(3)
[ 72]
[ 73] deallocate (y, x)
[ 74] end program two line lsq fit
[ 75]

For completeness an input routine, read xy file, is illustrated. It is elementary since it does not
check for any read errors, and thus does not allow for any exception control if the read somehow fails.
[ 76] subroutine read xy file (infile, lines, x, y)
[ 77] !------------------------------------------------------
[ 78] ! Take a file number, the number of lines to be read,
[ 79] ! and put the data into the arrays x and y
[ 80] !------------------------------------------------------
[ 81] implicit none
[ 82] integer, intent(in) :: inFile ! unit to read
[ 83] integer, intent(in) :: lines ! length of the file
[ 84] real, intent(out) :: x(lines) ! independent data
[ 85] real, intent(out) :: y(lines) ! dependent data
[ 86] integer j
[ 87] rewind (inFile) ! go to front of the file
[ 88] do j = 1, lines ! for the entire file
[ 89] read (infile, *) x(j), y(j) ! get the x and y values
[ 90] end do ! over all lines
[ 91] end subroutine read xy file
[ 92]

If the supplied data file was huge, say argument lines has a value of ten million, the such data would
probably have been stored in a binary rather that a formatted file. In that case we would simply invoke a
binary read by re-writing line 89 as
[ 89] read (infile) x(j), y(j) ! binary read of x and y

c 2001 J.E. Akin 64

Such a change would yield a much faster input, but would still be relatively slow due to being in the loop
starting at line 88. To get the fastest posible input we would have had to have saved the binary data on the
file such that all the x values were stored first, followed by all the corresponding y values. In that case,
we avoid the loop and get the fastest possible input by replacing lines 88–90 with:
[ 88] ! sequential binary read of x and y values
[ 89] read (infile) x, y
[ 90] ! input complete, add iostat for exceptions

Here we will not go into the details about how we would have to replace subroutine inputCount an
equivalent one for binary files. To do that you will have to study the Fortran INQUIRE statement for
files, and its IOLENGTH option to get a hardware independent record lenght of a real variable.
[ 93] ! Given test data in file two line.dat:
[ 94] ! 0.0000000e+00 1.7348276e+01
[ 95] ! 1.0000000e+00 6.5017349e+01
[ 96] ! 2.0000000e+00 8.7237749e+01
[ 97] ! 3.0000000e+00 1.2433478e+02
[ 98] ! 4.0000000e+00 1.5456681e+02
[ 99] ! 5.0000000e+00 1.8956219e+02
[100] ! 6.0000000e+00 2.1740486e+02
[101] ! 7.0000000e+00 2.3138619e+02
[102] ! 8.0000000e+00 2.7995041e+02
[103] ! 9.0000000e+00 3.1885162e+02
[104] ! 1.0000000e+01 3.4628642e+02
[105] ! 1.1000000e+01 3.3522546e+02
[106] ! 1.2000000e+01 3.7626218e+02
[107] ! 1.3000000e+01 3.9577060e+02
[108] ! 1.4000000e+01 4.2217988e+02
[109] ! 1.5000000e+01 4.3388828e+02
[110] ! 1.6000000e+01 4.5897959e+02
[111] ! 1.7000000e+01 4.9506511e+02
[112] ! 1.8000000e+01 5.0747649e+02
[113] ! 1.9000000e+01 5.2168101e+02
[114] ! 2.0000000e+01 5.2976511e+02

Assuming the formatted data are stored in file two line.dat, as shown above we obtain the best two
straight ine fit.
[115] ! Running the program gives:
[116] !
[117] ! Enter the data input filename: two line.dat
[118] ! There were 21 records read.
[119] ! Single line fit
[120] ! the slope is 25.6630135
[121] ! the intercept is 53.2859993
[122] ! the error is 343.854675
[123] ! Two line best fit; combined error is 126.096634
[124] ! Best division of the data is:
[125] ! data(:j), data(j+1:), where j = 11
[126] ! Left line fit:
[127] ! the slope is 31.9555302
[128] ! the intercept is 24.9447269
[129] ! the error is 46.060421
[130] ! Right line fit:
[131] ! the slope is 21.6427555
[132] ! the intercept is 112.166664
[133] ! the error is 80.0362091
[134]

Check this out by plotting the data points and the three straight line segments. Just remember that the
first line covers the whole domain, while the second goes only up to halfway between points 11 and 12
while the third line runs from there to the end of the independent data.

C.12 Problem 4.11.15 : Find the next available file unit

The INQUIRE statement has a lot of very useful features that return information based on the unit number,
or the file name. It can also tell you how much storage a particular type of record requires (like the sizeof
function in C and C++). Here we use only the ability to determine if a unit number is currently open. To
do that we begin by checking the unit number that follows the last one we utilized. Line 9 declares that
variable, last unit and initializes it to 0. The save attribute in that line assures that the latest value of
last unit will always be saved and available on each subsequent use of the function. Since the standard
input/output units have numbers less that ten we allow the unit numbers to be used to range from 10 to
999, as seen in line 8. However, the upper limit could be changed.

c 2001 J.E. Akin 65

 Lines 14–18 determine if the unit after last unit is closed. If so that unit will be used and we are
basically finished. We set the return value, next, update last unit, and return.
[ 1] function get next io unit () result (next)
[ 2] ! * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
[ 3] ! find a unit number available for i/o action
[ 4] ! * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
[ 5] implicit none
[ 6] integer :: next ! the next available unit number
[ 7]
[ 8] integer, parameter :: min unit = 10, max unit = 999
[ 9] integer, save :: last unit = 0 ! initialize
[10] integer :: count ! number of failures
[11] logical :: open ! file status
[12]
[13] count = 0 ; next = min unit - 1
[14] if ( last unit > 0 ) then ! check next in line
[15] next = last unit + 1
[16] inquire (unit=next, opened=open)
[17] if ( .not. open ) last unit = next ! found it
[18] return

Otherwise, if the unit after last unit is open we must loop over all the higher unit numbers in search of
one that is closed. If we succeed then we update last unit and return by exiting the forever loop, as seen
in lines 24 and 25.
[19] else ! loop through allowed units
[20] do ! forever
[21] next = next + 1
[22] inquire (unit=next, opened=open)
[23] if ( .not. open ) then
[24] last unit = next ! found it
[25] exit ! the unit loop
[26] end if

At this point it may be impossible to find a unit. However, with 999 units available it is likely that
one that was previously in use has now been closed and is available again. Before aborting we reset the
search and allow three cycles to find a unit that is now free. That is done in lines 27–31.
[27] if ( next == max unit ) then ! attempt reset 3 times
[28] last unit = 0
[29] count = count + 1
[30] if ( count <= 3 ) next = min unit - 1
[31] end if ! reset try

In the unlikely event that all allowed units are still in use we abort the function after giving some
insight to why.
[32] if ( next > max unit ) then ! abort
[33] print *,’ERROR: max unit exceeded in get next io unit’
[34] stop ’ERROR: max unit exceeded in get next io unit’
[35] end if ! abort
[36] end do ! over unit numbers
[37] end if ! last unit
[38] end function get next io unit

C.13 Problem 5.4.4 : Polymorphic interface for the class ‘Position Angle’
[ 1] module class Position Angle ! file: class Position Angle.f90
[ 2] use class Angle
[ 3] implicit none
[ 4] type Position Angle ! angle in deg, min, sec
[ 5] private
[ 6] integer :: deg, min ! degrees, minutes
[ 7] real :: sec ! seconds
[ 8] character :: dir ! N | S, E | W
[ 9] end type

The above type definitions are unchanged. The only new part of the module for this class is the INTER-
FACE given in the following four lines.
[10] interface Position Angle ! generic constructor
[11] module procedure Decimal sec, Decimal min
[12] module procedure Int deg, Int deg min, Int deg min sec
[13] end interface
[14] contains . . .

Returning to the original main program:

c 2001 J.E. Akin 66

 [ 1] program main
[ 2] use class Great Arc
[ 3] implicit none
[ 4] type (Great Arc) :: arc
[ 5] type (Global Position) :: g1, g2
[ 6] type (Position Angle) :: a1, a2
[ 7] type (Angle) :: ang
[ 8] real :: deg, rad

We simply replace all the previous constructor calls with the generic function Position Angle as
shown on lines 8 through 17 below.
[ 9] a1 = Position Angle (10, 30, 0., "N") ! note decimal point
[10] call List Position Angle (a1)
[11] a1 = Position Angle (10, 30, 0, "N")
[12] call List Position Angle (a1)
[13] a1 = Position Angle (10, 30, "N")
[14] call List Position Angle (a1)
[15] a1 = Position Angle (20, "N")
[16] call List Position Angle (a1)
[17] a2 = Position Angle (30, 48, 0., "N")
[18] call List Position Angle (a2)

C.14 Problem 6.4.1 : Using a function with the same name in two classes
[ 1] include ’class X.f90’
[ 2] include ’class Y.f90’
[ 3] program main ! modified from Fig. 4.6.2-3F
[ 4] use class Y, Y f => f ! renamed in main
[ 5] implicit none
[ 6] type (X ) :: x, z ; type (Y ) :: y
[ 7] x%a = 22 ! assigns 22 to the a defined in X
[ 8] call X f(x) ! invokes the f() defined in X
[ 9] print *,"x%a = ", x%a ! lists the a defined in X
[10] y%a = 44 ! assigns 44 to the a defined in Y
[11] x%a = 66 ! assigns 66 to the a defined in X
[12] call Y f(y) ! invokes the f() defined in Y
[13] call X f(x) ! invokes the f() defined in X
[14] print *,"y%a = ", y%a ! lists the a defined in X
[15] print *,"x%a = ", x%a ! lists the a defined in X
[16] z%a = y%a ! assign Y a to z in X
[17] print *,"z%a = ", z%a ! lists the a defined in X
[18] end program main ! Running gives:
[19] ! X f() executing ! x%a = 22
[20] ! Y f() executing ! X f() executing
[21] ! y%a = 44 ! x%a = 66
[22] ! z%a = 44

C.15 Problem 6.4.3 : Revising the employee-manager classes

.
The changes are relatively simple. First we add two lines in the Employee class:
interface setData ! a polymorphic member
module procedure setDataE ; end interface

Then we change two other lines:

[ 8] empl = setData ( "Burke", "John", 25.0 )
. . .
[14] mgr = Manager ( "Kovacs", "Jan", 1200.0 ) ! constructor

The generic setData could not also contain setDataM because it has the same argument signature as
setDataE and the compiler would not be able to tell which dynamic binding to select.

c 2001 J.E. Akin 67

68
Appendix D

Companion C++ Examples

D.1 Introduction
It is necessary to be multilingual in computer languages today. Since C++ is often used in the OOP
literature it should be useful to have C++ versions of the same code given earlier in F90. In most cases
these examples have the same variable names and the line numbers are usually very close to each other.
This appendix will allow you to flip from F90 examples in Chapter 4 of the main body of the text to see
similar operations in C++.

[ 1] #include <iostream.h> // system i/o files

[ 2] #include <math.h> // system math files
[ 3] main ()
[ 4] // Examples of simple arithmetic in C++
[ 5] f
[ 6] int Integer Var 1, Integer Var 2; // user inputs
[ 7] int Mult Result, Div Result, Add Result
[ 8] int Sub Result, Mod Result;
[ 9] double Pow Result, Sqrt Result;
[10] cout << "Enter two integers: ";
[11] cin >> Integer Var 1, Integer Var 2;
[12]
[13] Add Result = Integer Var 1 + Integer Var 2;
[14] cout << Integer Var 1 << " + " << Integer Var 2 << " = "
[15] << Add Result << endl;
[16] Sub Result = Integer Var 1 - Integer Var 2 ;
[17] cout << Integer Var 1 << " - " << Integer Var 2 << " = "
[18] << Sub Result << endl;
[19] Mult Result = Integer Var 1 * Integer Var 2 ;
[20] cout << Integer Var 1 << " * " << Integer Var 2 << " = "
[21] << Mult Result << endl;
[22] Div Result = Integer Var 1 / Integer Var 2 ;
[23] cout << Integer Var 1 << " / " << Integer Var 2 << " = "
[24] << Div Result << endl;
[25] Mod Result = Integer Var 1 % Integer Var 2; // remainder
[26] cout << Integer Var 1 << " % " << Integer Var 2 << " = "
[27] << Mod Result << endl;
[28] Pow Result = pow ((double)Integer Var 1, (double)Integer Var 2);
[29] cout << Integer Var 1 << " ˆ " << Integer Var 2 << " = "
[30] << Pow Result << endl;
[31] Sqrt Result = sqrt( (double)Integer Var 1 );
[32] cout << "Square root of " << Integer Var 1 << " is "
[33] << Sqrt Result << endl;
[34] g // end main, Running produces:
[35] // Enter two integers: 25 4
[36] // 25 + 4 = 29
[37] // 25 - 4 = 21
[38] // 25 * 4 = 100
[39] // 25 / 4 = 6, note integer
[41] // 25 % 4 = 1
[42] // 25 ˆ 4 = 390625
[43] // Square root of 25 = 5

Figure D.1: Typical Math and Functions in C++

c 2001 J.E. Akin 69

[ 1] #include <iostream.h> // system i/o files
[ 2] main ()
[ 3] // Examples of a simple loop in C++
[ 4] f
[ 5] int Integer Var;
[ 6]
[ 7] for (Integer Var = 0; Integer Var < 5; Integer Var ++)
[ 8] f
[ 9] cout << "The loop variable is: " << Integer Var << endl;
[10] g // end for
[11]
[12] cout << "The final loop variable is: " << Integer Var << endl;
[13]
[14] g // end main // Running produces:
[15] // The loop variable is: 0
[16] // The loop variable is: 1
[17] // The loop variable is: 2
[18] // The loop variable is: 3
[19] // The loop variable is: 4
[20] // The final loop variable is: 5 <- NOTE

Figure D.2: Typical Looping Concepts in C++

[ 1] #include <iostream.h> // system i/o files

[ 2] main ()
[ 3] // Examples of simple array indexing in C++
[ 4] f
[ 5] int MAX = 5, loopcount;
[ 6] int Integer Array[5] ;
[ 7] // or, int Integer Array[5] = f10, 20, 30, 40, 50 g;
[ 8]
[ 9] Integer Array[0] = 10 ; // C arrays start at zero
[10] Integer Array[1] = 20 ; Integer Array[2] = 30 ;
[11] Integer Array[3] = 40 ; Integer Array[4] = 50 ;
[12]
[13] for ( loopcount = 0; loopcount < MAX; loopcount ++)
[14] cout << "The loop counter is: " << loopcount
[15] << " with an array value of: " << Integer
[16] // end for loop
[17] cout << "The final loop counter is: " << loopcount << endl ;
[18]
[19] g // end main
[20]
[21] // Running produces:
[22] // The loop counter is: 0 with an array value of: 10
[23] // The loop counter is: 1 with an array value of: 20
[24] // The loop counter is: 2 with an array value of: 30
[25] // The loop counter is: 3 with an array value of: 40
[26] // The loop counter is: 4 with an array value of: 50
[27] // The final loop counter is: 5

Figure D.3: Simple Array Indexing in C++

[ 1] #include <iostream.h> // system i/o files

[ 2] main ()
[ 3] // Examples of relational "if" operator, via C++
[ 4] f
[ 5] int Integer Var 1, Integer Var 2; // user inputs
[ 6]
[ 7] cout << "\nEnter two integers: ";
[ 8] cin >> Integer Var 1, Integer Var 2;
[ 9]
[10] if ( Integer Var 1 > Integer Var 2 )
[11] cout << Integer Var 1 << " is greater than " << Integer Var 2;
[12]
[13] if ( Integer Var 1 < Integer Var 2 )
[14] cout << Integer Var 1 << " is less than " << Integer Var 2;
[15]
[16] if ( Integer Var 1 == Integer Var 2 )
[17] cout << Integer Var 1 << " is equal to " << Integer Var 2;
[18]
[19] g // end main
[20]
[21] // Running with 25 and 4 produces: 25 4
[22] // Enter two integers:
[23] // 25 is greater than 4

Figure D.4: Typical Relational Operators in C++

c 2001 J.E. Akin 70

[ 1] #include <iostream.h>
[ 2] main ()
[ 3] // Illustrate a simple if-else logic in C++
[ 4] f
[ 5] int Integer Var;
[ 6]
[ 7] cout << "Enter an integer: ";
[ 8] cin >> Integer Var;
[ 9]
[10] if ( Integer Var > 5 && Integer Var < 10 )
[11] f
[12] cout << Integer Var << " is greater than 5 and less than 10"
[13] << endl; g
[14] else
[15] f
[16] cout << Integer Var << " is not greater than 5 and less than 10"
[17] << endl; g // end of range of input
[18]
[19] g // end program main
[20]
[21] // Running with 3 gives: 3 is not greater than 5 and less than 10
[22] // Running with 8 gives: 8 is greater than 5 and less than 10

Figure D.5: Typical If-Else Uses in C++

c 2001 J.E. Akin 71

[ 1] #include <iostream.h>
[ 2] main ()
[ 3] // Examples of Logical operators in C++
[ 4] f
[ 5] int Logic Var 1, Logic Var 2;
[ 6]
[ 7] cout << "Enter logical value of A (1 or 0): ";
[ 8] cin >> Logic Var 1;
[ 9]
[10] cout << "Enter logical value of B (1 or 0): ";
[11] cin >> Logic Var 2;
[12]
[13] cout << "NOT A is " << !Logic Var 1 << endl;
[14]
[15] if ( Logic Var 1 && Logic Var 2 )
[16] f
[17] cout << "A ANDed with B is true " << endl;
[18] g
[19] else
[20] f
[21] cout << "A ANDed with B is false " << endl;
[22] g // end if for AND
[23]
[24] if ( Logic Var 1 || Logic Var 2 )
[25] f
[26] cout << "A ORed with B is true " << endl;
[27] g
[28] else
[29] f
[30] cout << "A ORed with B is false " << endl;
[31] g // end if for OR
[32]
[33] if ( Logic Var 1 == Logic Var 2 )
[34] f
[35] cout << "A EQiValent with B is true " << endl;
[36] g
[37] else
[38] f
[39] cout << "A EQiValent with B is false " << endl;
[40] g // end if for EQV
[41]
[42] if ( Logic Var 1 != Logic Var 2 )
[43] f
[44] cout << "A Not EQiValent with B is true " << endl;
[45] g
[46] else
[47] f
[48] cout << "A Not EQiValent with B is false " << endl;
[49] g // end if for NEQV
[50]
[51] g // end main
[52] // Running with 1 and 0 produces:
[53] // Enter logical value of A (1 or 0): 1
[54] // Enter logical value of B (1 or 0): 0
[55] // NOT A is 0
[56] // A ANDed with B is false
[57] // A ORed with B is true
[58] // A EQiValent with B is false
[59] // A Not EQiValent with B is true

Figure D.6: Typical Logical Operators in C++

c 2001 J.E. Akin 72

[ 1] // Program to find the maximum of a set of integers
[ 2] #include <iostream.h>
[ 3] #include <stdlib.h> // for exit
[ 4] #define ARRAYLENGTH 100
[ 5] long integers[ARRAYLENGTH];
[ 6]
[ 7] // Function interface prototype
[ 8] long maxint(long [], long);
[ 9]
[10] // Main routine
[11]
[12] main() f // Read in the number of integers
[13] long i, n;
[14]
[15] cout << "Find maximum; type n: "; cin >> n;
[16] if ( n > ARRAYLENGTH || n < 0 ) f
[17] cout << "Value you typed is too large or negative." << endl;
[18] exit(1);
[19] g // end if
[20]
[21] for (i = 0; i < n; i++) f // Read in the user’s integers
[22] cout << "Integer " << (i+1) << ": "; cin >> integers[i]; cout
[23] << endl; g // end for
[24] cout << "Maximum: ", cout << maxint(integers, n); cout << endl;
[25] g // end main
[26]
[27] // Find the maximum of an array of integers
[28] long maxint(long input[], long input length) f
[29] long i, max;
[30]
[31] for (max = input[0], i = 1; i < input length; i++) f
[32] if ( input[i] > max ) f
[33] max = input[i]; g // end if
[34] g // end for
[35] return(max);
[36] g // end maxint // produces this result
[37] // Find maximum; type n: 4
[38] // Integer 1: 9
[39] // Integer 2: 6
[40] // Integer 3: 4
[41] // Integer 4: -99
[42] // Maximum: 9

Figure D.7: Search for Largest Value in C++

c 2001 J.E. Akin 73

[ 1] #include <iostream.h>
[ 2]
[ 3] // declare the interface prototypes
[ 4] void Change ( int& Input Val);
[ 5] void No Change ( int Input Val);
[ 6]
[ 7] main ()
[ 8] // illustrate passing by reference and by value in C++
[ 9] f
[10] int Input Val;
[11]
[12] cout << "Enter an integer: ";
[13] cin >> Input Val;
[14] cout << "Input value was " << Input Val << endl;
[15]
[16] // pass by value
[17] No Change ( Input Val ); // Use but do not change
[18] cout << "After No Change it is " << Input Val << endl;
[19]
[20] // pass by reference
[21] Change ( Input Val ); // Use and change
[22] cout << "After Change it is " << Input Val << endl;
[23] g
[24]
[25] void Change (int& Value)
[26] f
[27] // changes Value in calling code IF passed by reference
[28] Value = 100;
[29] cout << "Inside Change it is set to " << Value << endl;
[30] g
[31]
[32] void No Change (int Value)
[33] f
[34] // does not change Value in calling code IF passed by value
[35] Value = 100;
[36] cout << "Inside No Change it is set to " << Value << endl;
[37] g
[38] // Running gives:
[39] // Enter an integer: 12
[40] // Input value was 12
[41] // Inside No Change it is set to 100
[42] // After No Change it is 12
[43] // Inside Change it is set to 100
[44] // After Change it is 100

Figure D.8: Passing Arguments by Reference and by Value in C++

c 2001 J.E. Akin 74

[ 1] #include <iostream.h>
[ 2] main ()
[ 3] // Compare two character strings in C++
[ 4] // Concatenate two character strings together
[ 5] f
[ 6] char String1[40];
[ 7] char String2[20];
[ 8] int length;
[ 9]
[10] cout << "Enter first string (20 char max):";
[11] cin >> String1;
[12]
[13] cout << "Enter second string (20 char max):";
[14] cin >> String2;
[15]
[16] // Compare
[17] if ( !strcmp(String1, String2) ) f
[18] cout << "They are the same." << endl;
[19] g
[20] else f
[21] cout << "They are different." << endl;
[22] g // end if the same
[23]
[24] // Concatenate
[25] strcat(String1, String2) ; // add onto String1
[26]
[27] cout << "The combined string is: " << String1 << endl;
[28] length = strlen( String1 );
[29] cout << "The combined length is: " << length << endl;
[30] length = strlen( String1 );
[31]
[32] g // end main
[33] // Running with "red" and "bird" produces:
[34] // Enter first string (20 char max): red
[35] // Enter second string (20 char max): bird
[36] // They are different.
[37] // The combined string is: redbird
[38] // The combined length is: 7
[39] // But, "the red" and "bird" gives unexpected results

Figure D.9: Using Two Strings in C++

[ 1] #include <iostream.h>
[ 2] #include <stdlib.h>
[ 3] #include <math.h> // system math files
[ 4]
[ 5] main()
[ 6] // Convert a character string to an integer in C++
[ 7] f
[ 8] char Age Char[5];
[ 9] int age;
[10]
[11] cout << "Enter your age: ";
[12] cin >> Age Char;
[13]
[14] // convert with intrinsic function
[15] age = atoi(Age Char);
[16]
[17] cout << "Your integer age is " << age << endl;
[18] cout << "Your hexadecimal age is " << hex << age << endl;
[19] cout << "Your octal age is " << oct << age << endl;
[20]
[21] g // end of main
[22]
[23] // Running gives:
[24] // Enter your age: 45
[25] // Your integer age is 45.
[26] // Your hexadecimal age is 2d.
[27] // Your octal age is 55.

Figure D.10: Converting a String to an Integer with C++

c 2001 J.E. Akin 75

[ 1] #include <iostream.h>
[ 2]
[ 3] // Define structures and components in C++
[ 4]
[ 5] struct Person // define a person structure type
[ 6] f
[ 7] char Name[20];
[ 8] int Age;
[ 9] g;
[10]
[11] struct Who Where // use person type in a new structure
[12] f
[13] struct Person Guest;
[14] char Address[40];
[15] g;
[16]
[17] // Fill a record of the Who Where type components
[18] main ()
[19] f
[20] struct Who Where Record;
[21]
[22] cout << "Enter your name: ";
[23] cin >> Record.Guest.Name;
[24]
[25] cout << "Enter your city: ";
[26] cin >> Record.Address;
[27]
[28] cout << "Enter your age: ";
[29] cin >> Record.Guest.Age;
[30]
[31] cout << "Hello " << Record.Guest.Age << " year old "
[32] << Record.Guest.Name << " in " << Record.Address << endl;
[33] g
[34] // Running with input: Sammy, Houston, 104 gives
[35] // Hello 104 year old Sammy in Houston
[36] //
[37] // But try: Sammy Owl, Houston, 104 for a bug

Figure D.11: Using Multiple Structures in C++

c 2001 J.E. Akin 76

Appendix E

Glossary of Object Oriented Terms

abstract class: A class primarily intended to define an instance, but can not be instantiated without
additional methods.
abstract data type: An abstraction that describes a set of items in terms of a hidden data structure and
operations on that structure.
abstraction: A mental facility that permits one to view problems with varying degrees of detail depend-
ing on the current context of the problem.
accessor: A public member subprogram that provides query access to a private data member.
actor: An object that initiates behavior in other objects, but cannot be acted upon itself.
agent: An object that can both initiate behavior in other objects, as well as be operated upon by other
objects.
ADT: Abstract data type.
AKO: A Kind Of. The inheritance relationship between classes and their superclasses.
allocatable array: A named array having the ability to dynamically obtain memory. Only when space
has been allocated for it does it have a shape and may it be referenced or defined.
argument: A value, variable, or expression that provides input to a subprogram.
array: An ordered collection that is indexed.
array constructor: A means of creating a part of an array by a single statement.
array overflow: An attempt to access an array element with a subscript outside the array size bounds.
array pointer: A pointer whose target is an array, or an array section.
array section: A subobject that is an array and is not a defined type component.
assertion: A programming means to cope with errors and exceptions.
assignment operator: The equal symbol, “=”, which may be overloaded by a user.
assignment statement: A statement of the form “variable = expression”.
association: Host association, name association, pointer association, or storage association.
attribute: A property of a variable that may be specified in a type declaration statement.
automatic array: An explicit-shape array in a procedure, which is not a dummy argument, some or all
of whose bounds are provided when the procedure is invoked.

c 2001 J.E. Akin 77

base class: A previously defined class whose public members can be inherited by another class. (Also
called a super class.)
behavior sharing: A form of polymorphism, when multiple entities have the same generic interface.
This is achieved by inheritance or operator overloading.
binary operator: An operator that takes two operands.
bintree: A tree structure where each node has two child nodes.
browser: A tool to find all occurrences of a variable, object, or component in a source code.
call-by-reference: A language mechanism that supplies an argument to a procedure by passing the ad-
dress of the argument rather than its value. If it is modified, the new value will also take effect
outside of the procedure.
call-by-value: A language mechanism that supplies an argument to a procedure by passing a copy of
its data value. If it is modified, the new value will not take effect outside of the procedure that
modifies it.
class: An abstraction of an object that specifies the static and behavioral characteristics of it, including
their public and private nature. A class is an ADT with a constructor template from which object
instances are created.
class attribute: An attribute whose value is common to a class of objects rather than a value peculiar to
each instance of the class.
class descriptor: An object representing a class, containing a list of its attributes and methods as well
as the values of any class attributes.
class diagram: A diagram depicting classes, their internal structure and operations, and the fixed rela-
tionships between them.
class inheritance: Defining a new derived class in terms of one or more base classes.
client: A software component that users services from another supplier class.
concrete class: A class having no abstract operations and can be instantiated.
compiler: Software that translates a high-level language into machine language.
component: A data member of a defined type within a class declaration
constructor: An operation, by a class member function, that initializes a newly created instance of a
class. (See default and intrinsic constructor.)
constructor operations: Methods which create and initialize the state of an object.
container class: A class whose instances are container objects. Examples include sets, arrays, and
stacks.
container object: An object that stores a collection of other objects and provides operations to access or
iterate over them.
control variable: The variable which controls the number of loop executions.
data abstraction: The ability to create new data types, together with associated operators, and to hide
the internal structure and operations from the user, thus allowing the new data type to be used in a
fashion analogous to intrinsic data types.
data hiding: The concept that some variables and/or operations in a module may not be accessible to a
user of that module; a key element of data abstraction.

c 2001 J.E. Akin 78

data member: A public data attribute, or instance variable, in a class declaration.

data type: A named category of data that is characterized by a set of values. together with a way to
denote these values and a collection of operations that interpret and manipulate the values. For an
intrinsic type, the set of data values depends on the values of the type parameters.

deallocation statement: A statement which releases dynamic memory that has been previously allo-
cated to an allocatable array or a pointer.

debugger software: A program that allows one to execute a program in segments up to selected break-
points, and to observe the program variables.

debugging: The process of detecting, locating, and correcting errors in software.

declaration statement: A statement which specifies the type and, optionally, attributes of one or more
variables or constants.

default constructor: A class member function with no arguments that assigns default initial values to
all data members in a newly created instance of a class.

defined operator: An operator that is not an intrinsic operator and is defined by a subprogram that is
associated with a generic identifier.

deque: A container that supports inserts or removals from either end of a queue.

dereferencing: The interpretation of a pointer as the target to which it is pointing.

derived attribute: An attribute that is determined from other attributes.

derived class: A class whose declaration indicates that it is to inherit the public members of a previously
defined base class.

derived type: A user defined data type with components, each of which is either of intrinsic type or of
another derived type.

destructor: An operation that cleans up an existing instance of a class that is no longer needed.

destructor operations: Methods which destroy objects and reclaim their dynamic memory.

domain: The set over which a function or relation is defined.

dummy argument: An argument in a procedure definition which will be associated with the actual
(reference or value) argument when the procedure is invoked.

dummy array: A dummy argument that is an array.

dummy pointer: A dummy argument that is a pointer.

dummy procedure: A dummy argument that is specified or referenced as a procedure.

dynamic binding: The allocation of storage at run time rather than compile time, or the run time asso-
ciation of an object and one of its generic operations..

edit descriptor: An item in an input/output format which specifies the conversion between internal and
external forms.

encapsulation: A modeling and implementation technique (information hiding) that separates the exter-
nal aspects of an object from the internal, implementation details of the object.

exception: An unexpected error condition causing an interruption to the normal flow of program control.

c 2001 J.E. Akin 79

explicit interface: For a procedure referenced in a scoping unit, the property of being an internal proce-
dure, a module procedure, an external procedure that has an interface (prototype) block, a recursive
procedure reference in its own scoping unit, or a dummy procedure that has an interface block.
explicit shape array: A named array that is declared with explicit bounds.

external file: A sequence of records that exists in a medium external to the program.

external procedure: A procedure that is defined by an external subprogram.

FIFO: First in, first out storage; a queue.

friend: A method, in C++, which is allowed privileged access to the private implementation of another
object.

function body: A block of statements that manipulate parameters to accomplish the subprogram’s pur-
pose.

function definition: Program unit that associates with a subprogram name a return type, a list of argu-
ments, and a sequence of statements that manipulate the arguments to accomplish the subprogram’s
purpose
function header: A line of code at the beginning of a function definition; includes the argument list, and
the function return variable name.

generic function: A function which can be called with different types of arguments.
generic identifier: A lexical token that appears in an INTERFACE statement and is associated with all
the procedures in the interface block.

generic interface block: A form of interface block which is used to define a generic name for a set of
procedures.
generic name: A name used to identify two or more procedures, the required one being determined by
the types of the non-optional arguments in the procedure invocation.

generic operator: An operator which can be invoked with different types of operands.
Has-A: A relationship in which the derived class has a property of the base class.

hashing technique: A technique used to create a hash table, in which the array element where an item
is to be stored is determined by converting some item feature into an integer in the range of the size
of the table.
heap: A region of memory used for data structures dynamically allocated and deallocated by a program.

host: The program unit containing a lower (hosted) internal procedure.

host association: Data, and variables automatically available to an internal procedure from its host.

information hiding: The principle that the state and implementation of an object should be private to
that object and only accessible via its public interface.

inheritance: The relationship between classes whereby one class inherits part or all of the public de-
scription of another base class, and instances inherit all the properties and methods of the classes
which they contain.

instance: A individual example of a class invoked via a class constructor.

instance diagram: A drawing showing the instance connection between two objects along with the num-
ber or range of mapping that may occur.

c 2001 J.E. Akin 80

instantiation: The process of creating (giving a value to) instances from classes.

intent: An attribute of a dummy argument that which indicates whether it may be used to transfer data
into the procedure, out of the procedure, or both.

interaction diagram: A diagram that shows the flow of requests, or messages between objects.

interface: The set of all signatures (public methods) defined for an object.

internal file: A character string that is used to transfer and/or convert data from one internal storage
mode to a different internal storage mode.

internal procedure: A procedure contained within another program unit, or class, and which can only
be invoked from within that program unit, or class.

internal subprogram: A subprogram contained in a main program or another subprogram.

intrinsic constructor: A class member function with the same name as the class which receives initial
values of all the data members as arguments.

Is-A: A relationship in which the derived class is a variation of the base class.

iterator: A method that permits all parts of a data structure to be visited.

keyword: A programming language word already defined and reserved for a single special purpose.

LIFO: Last in, first out storage; a stack.

link: The process of combining compiled program units to form an executable program.

linked list: A data structure in which each element identifies its predecessor and/or successor by some
form of pointer.

linker: Software that combines object files to create an executable machine language program.

list: An ordered collection that is not indexed.

map: An indexed collection that may be ordered.

matrix: A rank-two array.

member data: Variables declared as components of a defined type and encapsulated in a class.

member function: Subprograms encapsulated as members of a class.

method: A class member function encapsulated with its class data members.

method resolution: The process of matching a generic operation on an object to the unique method
appropriate to the object’s class.

message: A request, from another object, for an object to carry out one of its operations.

message passing: The philosophy that objects only interact by sending messages to each other that re-
quest some operations to be performed.

module: A program unit which allows other program units to access variables, derived type definitions,
classes and procedures declared within it by USE association.

module procedure: A procedure which is contained within a module, and usually used to define generic
interfaces, and/or to overload or define operators.

nested: Placement of a control structure inside another control structure.

c 2001 J.E. Akin 81

object: A concept, or thing with crisp boundaries and meanings for the problem at hand; an instance of
a class.

object diagram: A graphical representation of an object model showing relationships, attributes, and
operations.

object-oriented (OO): A software development strategy that organizes software as a collection of ob-
jects that contain both data structure and behavior. (Abbreviated OO.)

object-oriented programming (OOP): Object-oriented programs are object-based, class-based, sup-

port inheritance between classes and base classes and allow objects to send and receive messages.

object-oriented programming language: A language that supports objects (encapsulating identity,

data, and operations), method resolution, and inheritance.
octree: A tree structure where each node has eight child nodes.

OO (acronym): Object-oriented.

operand: An expression or variable that precedes or succeeds an operator.

operation: Manipulation of an object’s data by its member function when it receives a request.
operator overloading: A special case of polymorphism; attaching more than one meaning to the same
operator symbol. ‘Overloading’ is also sometimes used to indicate using the same name for differ-
ent objects.
overflow: An error condition arising from an attempt to store a number which is too large for the storage
location specified; typically caused by an attempt to divide by zero.

overloading: Using the same name for multiple functions or operators in a single scope.

overriding: The ability to change the definition of an inherited method or attribute in a subclass.
parameterized classes: A template for creating real classes that may differ in well-defined ways as
specified by parameters at the time of creation. The parameters are often data types or classes, but
may include other attributes, such as the size of a collection. (Also called generic classes.)
pass-by-reference: Method of passing an argument that permits the function to refer to the memory
holding the original copy of the argument

pass-by-value: Method of passing an argument that evaluates the argument and stores this value in the
corresponding formal argument, so the function has its own copy of the argument value
pointer: A single data object which stands for another (a “target”), which may be a compound object
such as an array, or defined type.

pointer array: An array which is declared with the pointer attribute. Its shape and size may not be
determined until they are created for the array by means of a memory allocation statement.

pointer assignment statement: A statement of the form “pointer-name ) target”.

polymorphism: The ability of an function/operator, with one name, to refer to arguments, or return
types, of different classes at run time.
post-condition: Specifies what must be true after the execution of an operation.

pre-condition: Specifies the condition(s) that must be true before an operation can be executed.

private: That part of an class, methods or attributes, which may not be accessed by other classes, only
by instances of that class.

c 2001 J.E. Akin 82

protected: (Referring to an attribute or operation of a class in C++) accessible by methods of any de-
scendent of the current class.

prototype: A statement declaring a function’s return type, name, and list of argument types.

pseudocode: A language of structured English statements used in designing a step-by-step approach to

solving a problem.

public: That part of an object, methods or attributes, which may be accessed by other objects, and thus
constitutes its interface.

quadtree: A tree structure where each tree node has four child nodes.

query operation: An operation that returns a value without modifying any objects.

rank: Number of subscripted variables an array has. A scalar has rank zero, a vector has rank one, a
matrix has rank two.

scope: That part of an executable program within which a lexical token (name) has a single interpreta-
tion.

section: Part of an array.

sequential: A kind of file in which each record is written (read) after the previously written (read) record.

server: An object that can only be operated upon by other objects.

service: A class member function encapsulated with its class data members.

shape: The rank of an array and the extent of each of its subscripts. Often stored in a rank-one array.

side effect: A change in a variable’s value as a result of using it as an operand, or argument.

signature: The combination of a subprogram’s (operator’s) name and its argument (operand) types.
Does not include function result types.

size: The total number of elements in an array.

stack: Region of memory used for allocation of function data areas; allocation of variables on the stack
occurs automatically when a block is entered, and deallocation occurs when the block is exited

stride: The increment used in a subscript triplet.

strong typing: The property of a programming language such that the type of each variable must be
declared.

structure component: The part of a data object of derived type corresponding to a component of its
type.

sub-object: A portion of a data object that may be referenced or defined independently of other portions.
It may be an array element, an array section, a structure component, or a substring.

subprogram: A function or subroutine subprogram.

subprogram header: A block of code at the beginning of a subprogram definition; includes the name,
and the argument list, if any.

subscript triplet: A method of specifying an array section by means of the initial and final subscript
integer values and an optional stride (or increment).

super class: A class from which another class inherits. (See base class.)

c 2001 J.E. Akin 83

supplier: Software component that implements a new class with services to be used by a client software
component.
target: The data object pointed to by a pointer, or reference variable.
template: An abstract recipe with parameters for producing concrete code for class definitions or sub-
program definitions.
thread: The basic entity to which the operating system allocates CPU time.
tree: A form of linked list in which each node points to at least two other nodes, thus defining a dynamic
data structure.
unary operator: An operator which has only one operand.
undefined: A data object which does not have a defined value.
underflow: An error condition where a number is too close to zero to be distinguished from zero in the
floating-point representation being used.
utility function: A private subprogram that can only be used within its defining class.
vector: A rank-one array. An array with one subscript.
vector subscript: A method of specifying an array section by means of a vector containing the subscripts
of the elements of the parent array that are to constitute the array section.
virtual function: A genetic function, with a specific return type, extended later for each new argument
type.
void subprogram: A C++ subprogram with an empty argument list and/or a subroutine with no returned
argument.
work array: A temporary array used for the storage of intermediate results during processing.

c 2001 J.E. Akin 84

Appendix F

Subject Index

In the index the F90/95 intrinsic attributes, functions, subroutines, statements, etc. are shown in upper-
case letters even though Fortran is not case sensitive. The page numbers are cited with the chapter (or
appendix) number followed by a period, followed by the pages in that chapter separated by commas.
Topics that occur frequently are only cited at their first few uses.

c 2001 J.E. Akin 0

 array-valued function 4.31
A edit descriptor 4.34,35 array-valued literal constant
ABS intrinsic 4.24 B.1 ASCII
abstract data type 2.5,7 character 2.1 4.33,35
access restriction 1.23 2.5 3.1 collating sequence 4.32
ACCESS specifier assembly language 1.18
accessor A.1 ASSIGN statement B.16,17
accuracy of real arithmetic assignment operator 1.11,12
ACHAR intrinsic 4.33 assignment statement A.1 B.17
ACTION specifier ASSOCIATED intrinsic 4.45
actual argument assumed-length dummy argument
actual array argument assumed-shape array
actual pointer argument assumed-size array
ADVANCE specifier 3.8 4.49,50 asterisk
allocatable array 5.3 format specifier 2.3,4,5
ALLOCATABLE attribute 5.3 B.12 default input unit
ALLOCATE default output unit
statement 5.3 B.13 attribute
status 4.29,30 B.13 ALLOCATABLE B.16
ALLOCATED intrinsic B.1 DIMENSION B.16
allocation statement B.4 EXTERNAL
allocation status INTENT B.16
alphabetic sorting INTRINSIC
alternate RETURN statement B.18 KIND B.16
ampersand object 1.22, 2.4, A.1
analysis OPTIONAL B.16
.AND. 3.17 4.16,17,48 PARAMETER B.16
angle class POINTER B.16
APOSTROPHE PRIVATE B.16
apostrophe edit descriptor PUBLIC B.16
argument SAVE B.16
actual SEQUENCE
association TARGET B.16
dummy attributes 1.23
function 4.23 automatic array 4.48 5.3
list automatic character length
presence
subroutine 4.23 B edit descriptor 4.36
arguments 1.15 back substitution
arithmetic expression back-up B.5
arithmetic IF statement BACKSPACE statement 4.29 B.17
arithmetic operators 4.4 base class 7.1 A.2
arithmetic unit behaviour 1.23
array binary digit 4.29
allocatable binary file 5.5
allocation binary number 4.36
assumed-shape binary operator 4.29
assumed-size 4.31 binary tree
automatic 5.2,3 A.1 bit intrinsic functions 4.29 B.5
bounds blank character 4.33
conformable BLANK specifier
constant BLOCK DATA statement B.16
constructor 4.24 5.4,10,14 B.4 block IF construct
deferred-shape block WHERE construct
dimensions B.4,7 BN edit descriptor
dummy argument 5.3 Boolean 2.1
element bottom-up design 1.4
explicit-shape bounds
extent 5.1 array B.5
extraction 5.9 lower
flip 5.9 scalar B.5
initialization 5.2,4,5 upper
input bubble sort 4.52
inquiry B.4 bug 1.8
mask 5.10,11 BZ edit descriptor
name in an I/O list
output C language 1.1
overflow A.1 C++ language 1.1,11,15,21 2.2
packing B.4 4.4,5,8,9,10,14,17,19,22
pointer 4.23,27,29,35,38,39,40,41
rank 4.31,53 4.45,47 5.2,4,9,24,25
reduction B.5 call by reference 4.31
reshape 5.13,14 call by value 4.31
shape 5.1 CALL statement
shifts 5.14 CASE construct
size 5.1 CASE DEFAULT statement 4.17,18
subscripts 5.2 case expression
unknown size 4.31 case selector 4.17,18
array element order CASE statement 4.17,18
array of pointers character
array processing intrinsics 5.6,7,11 argument
array section array
array specification assignment
array variable constant 3.8
array-valued derived-type component control

c 2001 J.E. Akin 1

 data I/O data abstraction 1.23
expression data hiding 3.1
length DATA statement B.16,19
pointer 4.45 data member 2.9
substring data structure
character edit descriptor 3.8 defining 4.39
character set initializing 4.39
default 4.32 interpretation 4.40
Fortran nested 4.38
CHARACTER statement 4.3 B.13 data types 2.1
CHARACTER type 2.1 4.31 Date class 3.5
characteristics DEALLOCATE
dummy argument statement 5.3 B.13
result variable status 4.29
chemical element 2.4,7 deallocation
Circle class 3.2,4,19 debugging 1.19,20
class decimal exponent range
base decimal precision
defined default
derived 4.38 accessibility
hierarchies 3.2 character set
classes 1.18,23 2.8 3.1 constructor 1.23
CLOSE statement 4.29 input unit 5.5
closing a file kind 4.3
CMPLX intrinsic 5.7 output unit 5.5
collating sequence precision
colon edit descriptor private accessibility
colon operator 4.7,25 5.8 public accessibility
column extraction 5.9 deferred-shape array
comment defined operation 4.31
fixed source DELIM specifier B.16
free source delimiter
! statement dereferencing 4.8
comments 1.1,6 4.1 derived class 7.1
COMMON block derived type
COMMON block name argument
COMMON statement 4.27 B.16 component
comparing character strings 4.32 constant
comparison of two real definition in a module
compiler 1.19 3.6 destructor 3.2
COMPLEX statement 4.3 B.13,21 dimension
COMPLEX type 2.1, B.6 attribute
component DIMENSION statement 4.25 B.19
derived type 2.4 direct access
composition READ statement
computed GO TO B.16,21 WRITE statement
concatenation DIRECT specifier B.16
operator DO
condition abort 4.10,20
end-of-file 4.29 construct 4.10
end-of-record 4.29 cycle
error forever 4.10
conditionals 1.6,7,14 4.13 loop 2.9
conformable arrays named 4.9,20
connectivity 5.12 nested 4.19
constant termination
character until 4.10,20
derived type variable B.16
integer DO statement 2.9 4.9 B.13
literal DO WHILE statement 4.9,16,20 B.19
named documentation 1.21
real DOT PRODUCT intrinsic 1.14
constant expression DOUBLE PRECISION attribute 2.1 4.3
constructor DOUBLE PRECISION statement B.16,17
default 1.23 double precision 2.3
intrinsic 2.5 3.2 doubly linked list 8.15
manual 2.8 3.7 Drill class 6.1
structure dummy argument
constructors 1.23 3.2 dummy array argument 5.3
containers 8.1 dummy pointer argument
CONTAINS statement 2.9 3.1 4.25,43 dynamic binding
continuation 1.11 B.18 dynamic character 4.31
CONTINUE statement B.16 dynamic data structure
control characters 4.32,35 dynamically allocated array 5.3
conversion constants dynamically allocated memory
copies B.6
count-controlled DO 1.13 4.11 E edit descriptor
counting B.6 edit descriptor
CPU time 4.28 A 4.34,35
curve-fitting 4.49,50 B 4.36
CYCLE BN
named 4.20 BZ
statement 4.9 B.13 D B.20
E B.20
D edit descriptor EN

c 2001 J.E. Akin 2

 ES 3.4 B.20 external procedure 4.47
F B.20 EXTERNAL statement B.13
G B.20
I 4.36,38 B.20 F edit descriptorB.20
L Fibonacci number
O 4.36 B.20 ADT 2.7
P class 2.8
S file
SP access
SS connection
T B.20 creation
TL B.20 disconnection
TR B.20 existence
X B.20 external
Z 4.36 B.20 inquiry
/ B.20 internal
: position
ELEMENTAL prefix FILE specifier
ELSE IF statement 4.9,16 fill in B.7
ELSE statement 3.16 4.9 fixed source form
ELSE WHERE statement 4.9 floating-point numbers
embedded format flow control 1.13 4.1,9
Employee class 7.5,9,12,15 FMT specifier
EN edit descriptor FORALL construct
encapsulation 3.1 FORM specifier
END DO statement 2.9 format
END FUNCTION statement embedded
END IF statement 3.16 4.9 list-directed
END INTERFACE statement 4.37 user input
END MODULE statement 2.3 FORMAT statement 3.4 B.14
END PROGRAM statement 2.3 4.2 formatted file
END SELECT statement formatted I/O statement
END statement formatted record
END SUBROUTINE statement FORMATTED specifier
end-of-file condition 4.29 Fortran Character Set
end-of-record condition 4.29 fraction 4.43,44
end-of-transmission 4.32 free source form
END= 4.30 function
ENDFILE statement 4.29 B.20 elemental
ENTRY statement B.13,20 length
EOSHIFT intrinsic 5.11,14 name
.EQ., see == pure
equality of two reals reference
EQUIVALENCE statement B.16,20 result
.EQV. 4.17 type
EM specifier with no arguments
error function actual argument
checking for function dummy argument
compilation FUNCTION statement 2.9 4.22
condition functions 1.6,15 4.22
execution
I/O G edit descriptor
logical Game of Life 1.4,9,16,20 4.23,25
semantic gather 5.12,14,15
>
syntactic Gaussian elimination
ES edit descriptor .GE., see =
exception 4.29,30 generic
exception descriptor defined operator
IOSTAT 4.13,29 function 3.2,4
STAT 4.29 5.3 identifier
executable statement interface 3.4,7,15
execution error interface block
existance B.7 name 4.31
EXISTS specifier 4.30 operator
EXIT procedure 4.31
named 4.20 geometry module
statement 4.9,25 global variables 1.16 4.27
explicit Global Position class 6.7
interface 4.31 GO TO statement 4.9,16,19 B.14
loops 4.9 Great Arc class 6.7
explicit-shape array greatest common divisor 3.16 4.49
exponent range .GT., see >
exponential fit 4.50
exponential format hash table
expression hexadecimal number 4.36
arithmetic host association
constant host program unit
evaluation host scoping unit
in an output list
mixed-mode expression I edit descriptor 4.36,38 B.20
expressions 1.12 4.1 IF
extending an operator construct 3.16 4.14
extent 5.1 named 4.18
EXTERNAL attribute nested 4.14
external file 4.13,37,47 IF statement 3.5 4.15 B.14

c 2001 J.E. Akin 3

 IF ELSE 3.16 4.14 DEALLOCATE 5.3
imaginary part 4.8 DIGITS
implicit declaration 4.3 DOT PRODUCT 1.14 5.6,7,11
implicit interface 4.25 EOSHIFT 5.11,14
implicit loop 4.13,24 5.5 EPSILON 5.7
IMPLICIT NONE statement 4.3,28 B.14 EXP 4.8 5.7
IMPLICIT statement 4.3 FLOAT
INCLUDE line 3.6 FLOOR 4.8 5.8
INDEX intrinsic 4.33,36 HUGE 4.7
index array IACHAR 4.33,35
index bounds IAND 4.29
infinite loop 4.10 IBCLR 4.29
information hiding 3.1 IBITS 4.29
inheritance 3.1,10 4.27 7.1 IBSET 4.29
initial statement ICHAR 4.33
initial value IEOR 4.29
initialization expression IMAG 4.8
input device INDEX 4.32
input editing INT 5.7,8
input list IOR 4.29
input record ISHFT 4.29
input statement ISHFTC 4.29
inquiry B.7 KIND 2.3
list-directed 4.13 LBOUND
input unit LEN 4.33,36
I/O statement LEN TRIM 4.33
INQUIRE statement 4.30 B.14 LGE 4.33
inquire-by-file 4.30 LGT 4.33
inquire-by-output-list LLE 4.33
inquire-by-unit 4.30 LLT 4.33
instance LOG 4.8 5.7
INT intrinsic 5.7,8 LOG10 4.8 5.7
integer LOGICAL 2.1
argument MATXUL 5.5,6,7,11
constant MAXLOC 4.24 5.7,11
division MAXVAL 4.24 5.7,11
expression MERGE 5.11
kind MINLOC 4.24 5.7,11
literal constant MINVAL 4.24 5.7,11
numbers MOD 4.6,8
pointer 4.45 MODULO 3.16 4.8
INTEGER type 2.1,9 MVBITS 4.29
INTENT attribute 3.4 4.9,23 NINT 4.8 5.7,8
INTENT statement B.14 NOT 4.29
interface 1.2,32 3.2 4.30 NULL
INTERFACE ASSIGNMENT state- PACK 5.11
ment 3.16 4.44 PRESENT
B.14 PRODUCT 5.7,11
interface block 3.16 6.4 RANDOM NUMBER 5.7
interface body 4.31 RANDOM SEED 5.7
INTERFACE OPERATOR statement 3.16 4.44 REAL 5.7
5.12,13 B.14 REPEAT 4.33 5.11
INTERFACE statement 3.4 4.37 B.14 RESHAPE 5.4,7,11
internal file 4.35 SCAN 4.33
internal procedure 4.24,28 SELECTED INT KIND 2.1
internal variable SELECTED REAL KIND 2.1,3
INTRINSIC attribute B.22 SHAPE 5.7
intrinsic constructor 4.43 SIGN 4.24 5.7
intrinsic data type 4.4 SIN 1.13 4.8 5.7
intrinsic procedures and calls SINH 4.8 5.7
ABS 4.8,24 5.7 SIZE 4.23 5.7,13
ACHAR 4.33,35 SPREAD 5.11
ACOS 4.8 5.7 SQRT 4.6,8 5.7
ADJUSTL SUM 1.14 4.23 5.6,11
AIMAG 5.7 TAN 5.7
AINT 4.8 5.7,8 TANH 4.8 5.8
ALL 5.7,11 TINY 5.8
ALLOCATE 5.3 TRANSFER 4.29 5.11
ALLOCATED 5.3 TRANSPOSE 5.6,8,11
ANINT 5.7,8 TRIM 4.33
ANY 5.7,11 UBOUND
ASSOCIATED 4.45,46 UNPACK 5.11
ASIN 4.8 5.7 VERIFY 4.33
ATAN 4.8 5.7 WHERE 4.25
ATAN2 1.15 4.8 5.7 INTRINSIC statement B.14,22
BIT SIZE 4.29 inverse of a matrix 4.48
BTEST 4.29 IOLENGTH specifier B.14
CEILING 4.8 5.7,8 IOSTAT specifier B.13
CHAR 4.32 iteration count
CMPLX 5.7 iterative methods
CONJG 4.8 5.7 iterator
COS 4.8 5.7
COSH 4.8 5.7
COUNT 5.7,11 keyword
CSHIFT 5.11,14 argument

c 2001 J.E. Akin 4

 KIND intrinsic 2.2,3 masks 4.13 5.10,12,25 B.1,2,3,9
kind massively parallel computer
default mathematical constants 2.3
inquiry Matlab 4.4,5,7,8,9,10,12,14
selector 4.22,23,27 5.2,4,5,6,7
type B.8,13 5.9,9,24,25
kind type parameter matrix
of an expression addition 5.18,24
column 5.11,16
L edit descriptor diagonal 5.17
label 4.16 factorization 5.21
latitude 6.7 inverse 5.12,20,24
LBOUND intrinsic
<
multiplication 5.12,19,24
.LE., see = operations 5.12
leading blanks partition 5.17
least squares fit 4.49,52 row 5.16
LEN intrinsic 4.33 shifts 5.14
LEN TRIM intrinsic 4.34 square 5.16
length symmetric 5.17
of a character argument transpose 5.12,17
of a character variable 4.35 mean 4.23
specification memory
lexical comparison intrinsic 4.33 allocation
LGE intrinsic 4.33 deallocation
LGT intrinsic 4.33 leak 8.9
library function 1.19 message
line mixed kind expressions
continuation 1.11 B.18 mixed-mode expression
maximum length of model number
multiple statements on 4.48 bit
linked list integer
circular real
double 8.15 modular design 1.5,6
pointer 4.45,47 modular program development 1.2
single 8.10 module 1.18 2.3 3.1 4.27,28
list-directed module procedure
data value termination on input MODULE PROCEDURE state-
format specifier ment 3.4,16 5.12,13 7.1
formatting MODULE statement 2.9 7.1 B.14
input module variable 2.3 4.28
output multiple inheritance 7.1
PRINT statement
READ statement name length
literal constant NAME specifier
array-valued named
LLE intrinsic 4.33 DO construct 4.9
LLT intrinsic 4.33 IF construct 4.18
local variable CASE construct
location in an array B.8 SELECT construct 4.18
logical named constant
expression 1.13 NAMED specifier
function NAMELIST statement B.16,22
literal constant .NE., see /=
value negative iteration count
variable negative subscript value
logical IF statement .NEQV. 4.17
LOGICAL intrinsic nested
logical operator 4.17 data structures 4.38
LOGICAL statement B.14 DO loops 4.13
LOGICAL type 2.1 4.17 B.8 IF blocks 4.14
longitude 6.7 implied loops 4.12
loop scoping unit
abort 4.19 Newton-Raphson method 1.13,25
cycle 4.19 NEXTREC specifier
counter 4.10,11 NML specifier B.23
implied 4.12,24 node
indexed 1.13 4.11,48 non-advancing I/O 3.18
infinite 4.10 non-advancing READ statement
named 4.9 non-counting DO loop
nested 4.12,13 non-default
post-test 4.10,20 character
pre-test 4.10,20 character set
variable 1.13 complex number
loops 1.6,7,13 4.12 integer
loss of precision kind
lower bound logical
<
lower case letters 4.36 real
.LT., see .NOT. 4.17,49
NULL 4.46
main program null character 4.33
maintainability NULLIFY statement 4.46 B.14,22
Manager class 3.1 7.5,8,10,12,13,15 NUMBER specifier
mantissa number B.9
many-one array section numerical sorting
masked array assignment

c 2001 J.E. Akin 5

 O edit descriptor 4.36 PRIVATE qualifier
object 3.1 PRIVATE statement 2.9 B.14
object-oriented procedure
analysis 1.21,23 argument
design 1.21,24 interface
languages 1.21,24 intrinsic
programming 1.1,21 3.1 6.1 libraries
obsolescent statements 2.1,2 B.17 size
octal number 4.36 PRODUCT intrinsic
ONLY qualifier 4.27 7.1 B.25 Professor class 7.3
OPEN statement B.14,25 program
OPENED specifier 4.30 design 1.3,11
operator errors
definition 4.43 name
overloading 3.14 4.43 structure
operators testing 1.18
arithmetic PROGRAM statement 2.3 4.2 B.14
binary program unit
concatenation external routine
unary function
optional argument 2.9 3.7 internal routine
OPTIONAL attribute 2.9 3.7 4.30,31 main
OPTIONAL statement B.14,22 module
.OR. 3.7 4.49 subroutine
order of evaluation projectile 4.49
effect of parentheses prototype 1.6,23 4.30
output pseudocode 1.5,8,9,14,16 4.10
output device PUBLIC attribute
output editing public attributes 2.6
output format public member
output list PUBLIC statement 2.9 B.14
expression in PURE prefix
output statement
list-directed quadratic equation 1.8
overflow
overloading 3.14,16 4.43 random access
random number
P edit descriptor range
PACK intrinsic procedure range of a DO loop
PAD specifier B.16 rank
padding 5.14 rational number
parallel computers 3.19 arithmetic 3.15
PARAMETER attribute 2.3 class 3.14
PARAMETER statement B.14,22 derived type 3.16
parameterized constants READ specifier
parameterized data types READ statement 2.4 4.12,29 B.14
parameterized real variables read-only file
parameterized variables READWRITE specifier
parentheses to set order real
pass by reference 4.7,31 argument
pass by value 4.8,31 arithmetic operation
PAUSE statement B.16,22 constant
peripheral device DO variable
Person class 3.6 4.47 exponential literal constant
physical constants expression
pointer literal constant
allocation number
array part (of a complex number)
assignment 4.45 variable
association status REAL intrinsic
component of derived type REAL statement 2.3 B.14
deallocation REAL type 2.1
dummy argument REC specifier B.16
input and output RECL specifier B.16
inquiry B.10 Rectangle class 3.2,3,4,19
linked list recursion
nullification recursive
return value 4.31 algorithms 4.49
target 4.45 bisection method
variable data structure
POINTER attribute 4.31,45 function B.15,21,24
POINTER statement B.14 procedure
pointer-valued function process
polymorphism 3.1 7.1 reference
POSITION specifier subroutine B.15,24,25
post-condition RECURSIVE qualifier 3.16 B.21,24
power law fit 4.50 reduction B.11
precedence orders 4.5 relational expression
precision 2.2 relational operator 4.4
pre-condition 3.7 4.30 rename modifier 7.1,4 B.25
PRESENT intrinsic renaming of module entities
PRINT statement repeat count
PRIVATE attribute repeatable edit descriptor
private components 2.6 repeated format
prive members 2.6 RESHAPE intrinsic

c 2001 J.E. Akin 6

 reshaping an array B.11 NONE B.16
restricting access to module NULL B.16
restrictions on a logical IF NUMBER
restrictions on a DO loop OLD B.16
result length OPENED B.21
RESULT specification 2.9 3.4 B.13 PAD B.16
result variable POSITION B.16
return from a procedure QUOTE B.16
RETURN statement 4.9 B.15,16 READ B.16
returns 1.15 READWRITE B.16
reverse order B.11 REC B.16,23,25
REWIND statement 4.29 B.15,24 RECL B.16
root REPLACE B.16
roots of a quadratic equation REWIND B.16
round-off error SEARCH B.16
row extraction 5.9 SEQUENTIAL B.16
SIZE B.16
S edit descriptor STAT B.13
SAVE attribute STATUS B.16
SAVE statement B.15,24 UNFORMATTED B.16
in a module 4.27 UNIT B.14,16
scalar UNKNOWN B.16
conformable with an array WRITE B.16
scalar product of two vectors ZERO B.16
scalar variable SS edit descriptor
scale factor statement entity
scatter 5.12,14 statement function B.16
scope statement label 4.16
scoping unit statement order
scratch file statements 1.1,2,11
SELECT CASE statement 4.9,17,18 B.15 STATUS specifier
SELECTED INT KIND intrinsic STOP statement 3.7 4.9 5.3 B.15
SELECTED REAL KIND intrinsic storage unit
semantic error character
SEQUENCE attribute B.24 numeric
sequential access string 2.1
sequential file strong typing 4.3,28
sequential I/O statement structure
SEQUENTIAL specifier components 4.38
shape constructor 4.43
side effects 8.9 structured programming 1.15
SHAPE intrinsic procedure Student class 3.7,11
shifts B.11 subprograms 4.21
simultaneous linear equations subprogram interface 3.16 4.31 6.4
singly linked list 8.10 subroutine
size actual argument
SIZE intrinsic 5.13 arguments
SIZE specifier call
solution of linear equations dummy argument
sorting, bubble interface
source form library
fixed form name
free form result
SP edit descriptor SUBROUTINE statement 2.9 4.22 B.15
space character subscript 2.9 5.1
SPACING intrinsic B.17 subscript expression
sparse matrix subscript triplet 4.7 5.2,8
Sparse Vector class substring
specification SUM intrinsic 1.14
specification expression swap 9.1
specification statement syntactic error 1.1,19
specifier SYSTEM CLOCK call 4.28
ACCESS B.16
ACTION tabs 4.50
ADVANCE B.16,23 T edit descriptor
APPEND B.16 target B.11
ASIS B.16 TARGET attribute 4.45
BLANK B.16 TARGET statement B.15
DELIM B.16 template 9.1
DIRECT B.16 TL edit descriptor
END B.16,22 top-down design
EOR B.16 TR edit descriptor
ERR B.13 TRANSPOSE intrinsic 5.4
EXIST B.21 TRIM intrinsic 4.33
FILE B.21 trailing blanks
FMT B.14,16,23 tree-structured data 4.45 8.1
FORM B.16 truncation error
FORMATTED B.16 type conversion intrinsic
IOLENGTH B.14,21 TYPE declaration statement 2.4,9
IOSTAT B.13,16,21 type parameter
NAME B.14,21 TYPE statement 2.4,9 B.15
NAMED
NEXTREC UBOUND intrinsic
NEW B.16 unary operator
NML B.23,25 undefined array

c 2001 J.E. Akin 7

 undefined pointer status
underflow
unformatted
file
I/O statements
record
UNFORMATTED specifier
unit number
unit specifier
until construct
upper bound
upper case letters 4.36
US Military Standard 4.29
USE association
USE statement 2.3 3.2,4,16 4.27
7.1 B.15
variable
character
declaration
initial value
internal
local
name 4.2
variables 1.12 4.1
Vector class 5.25 B.12
vector subscript 5.12
WHERE construct 5.10 B.25
WHERE statement 3.4 4.9 5.9 B.14,25
while loop 4.20,222
WRITE specifier 3.16 B.14
WRITE statement
X edit descriptor
Z edit descriptor
zero-sized array
! comment
continuation marker
namelist data initiator
( )
implied loop bounds
subscript bounds
(/ /) array constructor 5.2
** exponentiation 4.8
+ overloaded 3.16
- overloaded
/
edit descriptor
list-directed data terminator
namelist data terminator
value separator
// concatination 4.32
/=
not equal
overloaded
:
edit descriptor
subscript triplet
:: attribute terminator 2.3
; statement terminator
<
less than
<=
overloaded
less than or equal to
overloaded
=
assignment
overloaded 3.16
==
equal to
>=
overloaded 3.16
greater than or equal to
=>
overloaded
>
rename option
greater than
overloaded
character in a name

c 2001 J.E. Akin 8

Appendix G

Program Index

c 2001 J.E. Akin 1

 Int deg min 6.11,13,16
add Rational 3.16 Int deg min sec 6.11,13,16
Add to Q 8.5,7,8 invert 3.17,18
Angle 6.11,12,16 is equal to 3.17
array indexing 4.11 Is Q Empty 8.5,7,8
assign 4.44 Is Q Full 8.5,7,8
Change 4.32 is Stack Empty 8.3,4
check basis D.12 is Stack Full 8.3,4
circle area 3.4 is S L empty 8.12,13
class Angle 6.12 less than Object 8.14
class Circle 3.4,19 linear fit 4.51
class Date 3.7,10,136.17 list 3.17
class Drill 6.5 list 4.44
class Employee 7.5,6,8,9,11,12 D.21 List Angle 6.11,12
class Fibonacci Numbers 2.9 List Great Arc 6.10,15
class Global Position 6.14 List Positions 6.10,14,16
class Great Arc 6.15 D.21 List Position Angle 6.11,13
class Manager 7.8,9,10,11,12,14 List Pt to Pt 6.10,15,16
class Object 8.13,14,15 logical operators 4.17
class Person 3.9,10,12,13 6.17 lsq fit 4.52
class Position Angle 6.12 D.21 make Person 3.9,10,13
class Professor 7.3 make Professor 7.3
class Queue 8.6 make Rational 3.17
class Rational 3.16,18 make Rectangle 3.5
class Rectangle 3.4,19 make Stack 8.3,4
class Sparse Vector D.14 make Student 3.12
class Stack 8.3 Manager 7.8,9,13
class Student 3.12,13 Math constants 2.3
class Vector D.7 maximum 4.26
clip 4.24 maxint 4.26
clip an array 4.24 mean 4.23
compare strings 4.34 mult Fraction 4.43,44
Conversion Constants D.3 mult Rational 3.17,18
convert 3.16 next generation 4.25
copy Rational 3.16,18 No Change 4.32
create a type 2.4 Object 8.4,13,14,18
Create Q 8.5,7,8 Ops example 5.13
Date 3.7 out 6.4,5
Decimal min 6.11,12 passing arguments 4.32
Decimal sec 6.11,12,16 pay 7.9,10
Default Angle 6.11,12 payE 7.5,6,12
define structures 4.42 payM 7.8,9,13
delete Rational 3.16,18 Person 3.9,10
derived class name 7.1 Physical Constants D.4
destroy D L List 8.16 pop from Stack 8.3,4
doubly linked list 8.16 print 7.3
Drill 6.3,4,5,6 PrintPay 7.11,12,14
D L insert before 8.17,18 PrintPayEmployee 7.11,12
D L new 8.16,18 PrintPayManager 7.11,13
equal Fraction 4.44 print Date 3.7
equal integer 3.15 print DOB 3.9
equal to Object 8.14 print DOD 3.9
exception 4.30 print DOM 3.12
exceptions 4.30 print D L list 8.17,18
exception status 4.30 print GPA 3.12,13
Fibonacci 2.9 print name 3.9,10,13
Fraction 4.44 print Nationality 3.9
Fractions 4.44 print sex 3.9
game of life 4.25 print S L list 8.12,13
gcd 3.16 4.54 pt expression 4.46
geometry 3.4 push on Stack 8.3,4
getEmployee 7.8,13 Rational 3.17,18
getName 7.9,12 readData 4.37 D.6
getNameE 7.5,6,12 read Date 3.7
getNameM 7.8,9,13 Read Position Angle 6.11,13
getRate 7.5,12 rectangle area 3.4
get Arc 6.10,15 reduce 3.17
Get Capacity of Q 8.5,7,8 relational if 4.15
get Denominator 3.16 Remove from Q 8.5,7,8
Get Front of Q 8.5,7,8 setData 7.9,11
get Latitude 6.10,14 setDataE 7.5,6,12,14
Get Length of Q 8.5,7,8 setDataM 7.8,9,13,14
get Longitude 6.10,14 setSalaried 7.8,9,10,11,13,14
get mr rate 6.3,5,6 set Date 3.7
get Numerator 3.17 set DOB 3.9,10,13
Get Obj at Ptr 8.17,18 set DOD 3.9,10
get person 3.12,13 set DOM 3.12,13
Get Ptr to Obj 8.17,18 set Latitude 6.10,14
get torque 6.3,5,6 set Lat and Long at 6.10,14,16
Global Position 6.10,14,16 set Longitude 6.10,14
Great Arc 6.10,15,16 simple arithmetic 4.6
hello world 4.2,53 simple if else 4.16
in 6.4,5 simple loop 4.11
inputCount 4.37 D.6 singly linked list 8.11
Int deg 6.11,12,16 spy 4.25

c 2001 J.E. Akin 2

 string to integer 4.36
Student 3.12
swap chemical element 9.2
swap integer 9.1
swap library 9.2
swap objects 9.1
S L delete 8.11,13
S L insert 8.12,13
S L new 8.12,13
test Arc 6.16
test Drill 6.6
test D L L 8.18
test matrix 4.48
test Professor 7.4
test Queue 8.8
test Stack 8.4
test S L L 8.13
test Vector D.11
tic 4.28
tic toc 4.28
toc 4.28
to Decimal Degrees 6.11,13,16
to lower 4.36,37
to Radians 6.11,13,16
to upper 4.36,37 D.5
up down 4.37
watch D.5

c 2001 J.E. Akin 3