On Variables in Mathematics and in Natural Science

Author(s): Karl Menger

Source: The British Journal for the Philosophy of Science , Aug., 1954, Vol. 5, No. 18
(Aug., 1954), pp. 134-142
Published by: Oxford University Press on behalf of The British Society for the
Philosophy of Science

Stable URL: https://www.jstor.org/stable/685170

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend
access to The British Journal for the Philosophy of Science

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 ON VARIABLES IN MATHEMATICS AND IN
NATURAL SCIENCE *

KARL MENGER

IN the development of mathematics and natural science which

began in the seventeenth century, perhaps the most characteristic
concept is that of a variable. But what is a variable ? Russell, wh
investigated all aspects of this notion, called it 'certainly ... one of
the most difficult to understand' and in 190o3 stated that a satisfactory
theory of variables was not to be found in his Principles of Mathematics.
In fifty years this situation has not been improved.
NUMERICAL VARIABLES AND VARIABLES IN THE SENSE OF THE
LOGICIANS. Let C be a class of numbers.' Weierstrass defined a
numerical variable or, more specifically, a variable with the range C as a
symbol standing for any element in C. The letter' x 'in D log x = I/x
(where D denotes the derivative) is a variable of which the range is
the class of all positive numbers. So is the letter' y' in D log y = I/y.
Logic deals with a generalisation of Weierstrass's concept. The
logicians admit any class as the range of a variable; for example, the
letter 'f' in the formula Df2 (o) = 2f(0)'Df(O) is a variable of which
the range is the class of all functions which are differentiable at o.
(Here the number o can, of course, be replaced by any other number.)
Natural science, on the other hand, deals with variables such as gas
pressure and distance travelled by a falling object, which have no
direct relation to Weierstrass's concept.
VARIABLE QUANTITIES. In better agreement with scientific usage,
French analysts have defined a variable as 'une quantit6 qui passe
par une infinit6 de valeurs rtelles, distinctes ou non'. However, this
definition has never been supplemented either by explicit definitions
of' quantit6 ' and 'passer par' or by postulates which implicitly define
these concepts, as postulational geometry defines 'points' and' lines'.
An attempt is here made to fill this gap with a theory 2 of variable

* Received 7. x. 53
1 In this paper, number means finite real number
2 cf. the author's note 'The Ideas of Variable and Function', Proc. Nat. Acad.
Sci., 1953, 39, 956-961, and the second edition of his book, Calculus: A Modern
Approach, Chicago, 1953
134

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 ON VARIABLES IN MATHEMATICS

quantities-briefly, v.q.'s. Whether or not v.q.

complete the French notion, they are to be clearl
variables in the sense of either Weierstrass or the
Let A be any class. By a v.q. with the dom
class of pairs such that (I) in each pair the first m
of A, and the second a number (called a value
element of A is the first member of exactly one
class. The class of all values is called the range
a v.q. with the domain A, and (c, n) a pair bel
is an element of A, and n is a number. By (z), u
pair (c, n') with n' + n; hence n is called the v
Consequently, u is the class of all pairs (o, uo), w
of A. (Thus, in the sense of logic, x is a variable
If K is the class of all circles in a plane, and rK
of the circle K, then the class r of all pairs (K,
domain K. If aK denotes the area of K, the class
is another v.q. with the domain K. The class K
subclasses, each containing all circles which ar
circle of that subclass. If K " K', then re = r
say that variables r and a are invariant with resp
ScIENTIFIC V.Q.'s. Let r be the class of all acts
meter connected with any I mole sample of gas
Iooo. Non-simultaneous acts (even if performed
are regarded as distinct. For each act y belongin
the result of the act, i.e. the observed number o

The class p of all pairs (y, p/) for any y belongin

the domain r, called the pressure of I mole of g
class of volumetric observations made on any I m
at Iooo and if, for any y' belonging to 1', vy' de
y', then the class v of all pairs (y', vy') is a v.q.
called the volume of I mole of gas at iooo. Other
connected with objects dropped from the zero po
If, for any act a by which the scale mark oppos
observed, so denotes the mark observed, then th
(a, sa) is a v.q., called the distance travelled by any f
the class of all pairs in which the first member
timer, set in motion when an object is dropped
1 A v.q. with the domain A in the sense of this definition
a function defined on A. This paper follows the usage of th
the concept of a function to the cases where the domains co
135

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 KARL MENGER

of the scale, and the second member is the result of t

t is a v.q. called the time elapsed since the release of a
The class T of all first members is the domain of t.
element of T (and thus, in the sense of logic, is a va
range T), t is the class of all pairs (7, tr). Any numb
value of t.
A v.q. the range of which consists of only one nu
constant. For instance, while a car is parked, the mi
and the speed are constant v.q.'s.
The domain of any one of these v.q.'s, say ', falls
disjoint subclasses, each containing all acts simultane
on the same sample by different observers. If two su
parallel, p is, at least in a first approximation, invari
to this parallelism. The domains of other v.q.'s (e.
comprise acts of observation on different instrum
thermometers, alcohol thermometers, thermo-coupl
what P. W. Bridgman calls 'paper-and-pencil-operatio
FUNCTIONS. A v.q. the domain of which is a class
called a function. Examples are : (i) the logarithm
number; (2) the reciprocal of any number other t
times the square of any number; (4) a constant fu
with the value 16 for any number. These functions
the classes of all pairs (I) (x, log x) for any positiv
for any x = o; (3) (x, 16 x') for any x; (4) (x, I6
these definitions, x is a numerical variable and may b
other symbol; e.g. the function (i) is the class of all
for any positive y.
A natural symbol for the function (I) is 'log'.
(2), (3), and (4) have no traditional symbols. If n is a
IP we denote the nth power function, i.e. the cl
(x, x") for any x if n > o, and for any x 0 o if n
I" for x (that is, x") can also be written I"x. Fun
We denote (4) by '16', and (3) by '16.P '. Even
function I, the class of all pairs (x, x), in spite of its
portance, has no standard symbol in classical analysi
is reminiscent of the lack of a symbol for the numbe
arithmetic, and results in similar complications. Tra
referred to as the function x" (I, as the function x) a
of unifornmity, every function is named after its v
instance, the function 16 is denoted by ' 16' (a practic
136

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 ON VARIABLES IN MATHEMATICS

confusion with the number 16) and, although n

tables, the function log is referred to as the fu
pars pro toto, since the symbol of the value is lo
symbol of the function itself.
Let f be any function, that is, a class of pair
where no two pairs have equal first and unequa
The domain off (briefly, domf) is the class of
the pairs. The value off for x may be denoted
just as the value of log for x is denoted by log x
If u is a v.q. the range of which is a part of do
element of the domain A of u, then us belongs

off for ur is f(uo)-briefly fuc. The class of

any a belonging to A is a v.q., denoted byfu. It
substitution of u into f In particular, a function
is a part of dom f can be substituted intof; the
fg, having the same domain as g. (N.B. Juxt
(= I1), denotes substitution, not multiplication.
g will be indicated byf'g, for which 16.P and
No v.q.'s but functions 2 permit the substitu
Any function f certainly permits the substitu
I', defined as the class of all pairs (x, x) for any
f.3 Into the function I, every v.q. can be substi
That among the v.q.'s the functions are chara
that they lend themselves to the substitutio
especially of other functions is perhaps most cle
following remark. While we can speak of the lo
of a number and of the logarithm of the pressu
speak of the pressure of the logarithm of a numbe
of the temperature of a gas.
Is w A FUNCTION OF u? If y, and y2 are two
reading belonging to the class r' such that pr1
7,' are two acts of volumetric reading (belo
neously performed on the same gas samples as ,
1 A v.q., in particular, a function may be a class consi
would be difficult to exclude such trivial examples. On
clusion does not seem to impair the significance of the
2 Here, to be quite precise, instead of'functions' w
domains of which contain numbers '.

3 For every element of dom I' (= domf), the function I' assumes the same value
as I. But dom I' may be a proper part ofdom I, in which case I is called an extension
of I, and I' a restriction of I.

I37

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 KARL MENGER

then it is found that vy,' = vyr2'. Consequently, t

of numbers (pr, vy') for any element , of 1 and t
element y' of "', is a function-a fact expressed b
v.q. v is a function of the v.q. p, where the pairin
acts on the same sample is tacitly understood.1 Ina
p and v have been assumed to be invariant with
observations (simultaneously performed on th
different observers), the pair (r, y') may be rep

(vYi, y) where Y, is equivalent to r, and y is

Similarly, the class of all pairs (rK, aK) for any cir
a fact expressed by saying that a is a function of r,
i with itself being tacitly understood.
Let u and w be any two v.q.'s, say, the classes of
for any a belonging to A, and (P, wP) for any P
w a function of u? No attempt will here be ma
question absolutely. It will, however, be answer
one-to-one mapping of a part of A on a part of B,
of pairs (ac, ) such that each element of A (resp. B
second) member of at most one pair belonging

w is a function
belong to 17 and ofu relative
if uoc = ura2,tothen
U, provided
also wfi1 =that,
wf,. if
In (ac,
thisP1) and (c2, 2)
case,
the class of all pairs (uc, wf) for any pair (a, f) belonging to H is a
function.
It may be that w is a function of u relative to some mappings but
not to others. Suppose two oscillators of equal period have a phase
difference of 1/4 of a period. Relative to pairs of simultaneous obser-
vations of the positions of the oscillators, neither position is a function
of the other. Relative to pairs of observations differing in time by
1/4 of a period, either position is a function of the other. Relative to
certain other pairs, one position is a function of the other but not
vice versa.

WHAT FUNCTION OF U IS W? Boyle not only recognised the cla

of all pairs (py. vy') as a function, but, if proper units are chosen
identified it with a simple known function, I-1. For he found tha
v ' = I/pr for each y belonging to 1 ; hence his function is the cla
of all 2 pairs (p, I/ply). For falling objects, Galileo implicitly intro
1 cf. N. R. Campbell, Physics. The Elements, Cambridge, 1920o, p. 329 if.
2 The domain of Boyle's function, being the range of the variable quantity
consists of positive numbers only; the function really is a restriction of I-V in
sense of footnote 3 on p. I37.
138

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 ON VARIABLES IN MATHEMATICS

duced the class of pairs (r, c) each consisting (I)

a timer, set in motion when an object is dro
taneous act a of observing the scale mark oppo
If t and so are the numerical results in seconds a
that is to say, the class of pairs (tr, sc) = (tT, I6
the function 16.1. x (Here, in the sense of logic
the range T).
Traditionally, physical laws are expressed
v = Iup, s = I6t2 or, more generally, w = fu.
to elements of the domains, and the formulae c
the sense of logic. A law in this form equates a
be called a denominate function, just as a statement
equates some property of an object and a denom
is the length of some rod, R = 7j ft. and gener
like properties of objects. Such results can be f
pure numbers : (R in ft.) = 74 and (B in A)
s,, and w, denote the functions connecting v w
with u, the physical laws can be formulated in term
vP = I-', st = 6.P, and w. =f The variable
(corresponding to the units, ft. and A) are called
dependent. Incidentally, there is a more far-rea
the functional connection of v.q.'s and the num
properties, with substitution of functions repla
numbers. For instance, relative to proper mapp
wi, = g, v, = h, and w, = fimplyf= gh, just
A) = e, and (C in A)= c imply c= d'e; and
B)= I.
Traditionally, the functions v,, s,, and w, are denoted by v(p),
s(t), and w(u). A comparison of the last three symbols with the
symbols 'log x ' and 'f(x) ' strongly suggests (I) that v, s, and w are
functions such as log and f; (2) that x is of the same nature as p, t,
and u. But neither is the case : (I) v and s are scientific v.q.'s (cer-
tainly not functions) and w is any v.q.; (2) x is a numerical variable
and not a v.q. such as p, t, and u. In the formula D log x = I/x, the
numerical variable, the range of which is the class of all positive
numbers, can be replaced (a) by any number (say 3) belonging to the
range : D log 3 = I/3 ; (b) by any other numerical variable (say y)

1 More precisely, instead of' identical with 'we should say 'a restriction of'
(in the sense of footnote 3 on p. I37).
K I39

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 KARL MENGER

having the same range: D log y = I/y. The nat

toto caelo different: (a) t in s(t) cannot be repla
symbol s(3) does not reveal which distance is mean
time is 3 or that when the velocity c is 3 or that
scientific v.q. is 3. Actually, physicists avoid th
write s(t = 3), s(c = 3), or whatever the situati
(b) t in s(t) cannot be replaced by c or any letter sy
other than the time.
VARIABLE QUANTITIES AND FUNCTIONS IN PURE
ANALYSIS. In analysis, derivatives and integrals
defined by well-known limiting processes. In theore
= cos x, the numerical variable x can be complet
the lack of a traditional symbol for the power functio
made it impossible to write D log = I-' an
just as D sin = cos. In the formula cos x dx =
asa

'x dx' adds nothin

can, without a1
am
In fa
able
calculus?.
But how can results about cos or f (which contain no
numerical variables) be applied to a force p, considering that it is
necessary to distinguish between impulse (the integral of p with
regard to the time t) and work (the integral of p with regard to the
distance s)? The answer is as follows: Since integration is defined
for functions, and p is a scientific v.q. and not a function, Jp is a
meaningless symbol; but the functions connecting p with other
v.q.'s such as t and s can be integrated: p, (traditionally written

p dt) is an impulse, whereas p, (or p ds) is work. For an

elastic force, connected with the time by p, = cos, the impulse is

a a=S
-= cos = sin b - sin a. If p is connected with s by a Hooke
1 cf. the Introduction and Chapters VIII and IX of Calculus: A Modern Approach,
quoted in footnote 2 on p. 134
140

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 ON VARIABLES IN MATHEMATICS

function p, = - I, then the work is p

Similarly, if v is the volume of an expand
is a scientific v.q. and not a function), 'D
but the function va connecting v with th
derivative Dvi; and certainly, if r is the rad
= 4r.P. To ask for the square root or the log o
But it is possible to compute the square root
necting the city with a square foot (its area in sq
of the number connecting the city to an inhabi

In Dv, and Jp, the (independent) scientific v.q

n/2

as important as v and p. But in D sin x = cos x and cos x dx= I,

the numerical variable x is not as important as the function sin and

cos. In fact, in these formulae x is completely superfluous. The
tacit identification of independent v.q.'s in denominate functions and
numerical variables in the formulae of calculus is a century-old mis-
understanding which can be clearly traced to the three principal
shortconmings of traditional analysis: (I) the equivocal use of the
term 'variable ', sometimes in the sense of Weierstrass, sometimes in
the sense of a scientific v.q., and sometimes in the sense of a general
v.q.; (2) the unfortunate notation which suggests a parallelism
between p(t) and f(x), and conceals the true analogies between p(t)
andf, as well as between p(t = 3) and f(3); (3) the lack of a symbol
for the nth power function, which has necessitated the use of numer-
ical variables in formulae such as D(I6 x2) = 32 x.
Moreover, little attention has been paid to the fact that differential
calculus deals with (I) a pure operator, D (a symbol introduced by
Arbogast, whereas Lagrange used a dash, and Heaviside the letter
'p') ; (2) an operator here called denominate (which Leibniz denoted
by 'd/dx ', Cauchy by 'Di'). Newton's dot, inasmuch as it is used
for differentiation with respect to the time, is implicitly denominate.
The pure operator associates a function with a function, whereas the
denominate operator associates a v.q. with a pair of v.q.'s. For
instance, ds/dt (= S) and dv/dp are scientific v.q.'s (the velocity and
the rate of expansion with respect to pressure) and so are the impulse

from a on, and the work from c on, which may be denoted by Jp dt
I4I

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms
 KARL MENGER

and ds. In general, dw/du and w du denote v.q

change w with u, and the cumulation of w with u). T

nected with u by the functions (dw/du), = Dw, and (J w du), = wS.

From the theorem of pure calculus (I) D sin = cos it follows that
w, = sin implies Dw, = cos or, in denominate form : (2) If u and w
are v.q.'s such that w = sin u, then dw/du = cos u. Without reference
to w, (2) assumes the form : (3) If u is any v.q., such as time or pres-
sure, then d sin u/du = cos u. (N.B. (3) must not be confused with the
statement : Ifx is any number, such as I or rr/2, then D sin x= cos x.)
Formulation (I) follows the usage of the physicist who writes v= Ip
rather than vy' = I/pr. Expressing Theorem (i) in the forms (2)
and (3) is comparable to expressing the arithmetical result (i') 15= 35
in the forms (2') If A and B are two objects such that B = i 5A, then
B = 3'5A or (3') If A is any object, such as a penny or an apple,
then IA = 3"5A. Arithmeticians, except when instructing very
young children, confine their statements to the realm of pure numbers
without in any way jeopardising the applicability of their results.
They never stoop to references to objects (not even to objects of the
most general kind). It seems likely that, once the notion of' variable'
is fully elucidated, analysts will rise above the level of references to
v.q.'s (even to v.q.'s of the most general kind) without in any way
jeopardising the applicability of their results. They will confine
their statements to the realm of pure functions, such as sine, 16.12, and
f The introduction of a symbol for the identity function makes it
possible to dispense with numerical variables and to express theorems

in formulae such as D log = I-, D(f) = 2 .f. DJ exp. = e - I.

These formulae can be directly applied to functions connecting
variable quantities.
Illinois Institute of Technology
Chicago

1 In writing this paper the author has greatly profited from suggestions of Dr
James K. Senior.

142

This content downloaded from

84.226.222.49 on Sun, 14 Jan 2024 15:56:30 +00:00
All use subject to https://about.jstor.org/terms