This page intentionally left blank

Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data

Names: Smullyan, Raymond M., author.
Title: A beginner’s further guide to mathematical logic / Raymond Smullyan.
Description: New Jersey : World Scientific, 2016. |
Includes bibliographical references and index.
Identifiers: LCCN 2015033651 | ISBN 9789814730990 (hardcover : alk. paper) |
ISBN 9789814725729 (pbk : alk. paper)
Subjects: LCSH: Logic, Symbolic and mathematical.
Classification: LCC QA9.A1 S619 2016 | DDC 511.3--dc23
LC record available at http://lccn.loc.gov/2015033651

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

On the cover, the three photos from left to right are the logicians
Emil Post, Alan Turing, and Ernst Zermelo.

Copyright © 2017 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.

Typeset by Stallion Press

Email: [email protected]
 Contents

Preface ix

Part I: More on Propositional and First-Order Logic 1

Chapter 1. More on Propositional Logic 3

I. Propositional Logic and the Boolean Algebra of Sets . . . 3

II. An Algebraic Approach . . . . . . . . . . . . . . . . . . . 4
III. Another Completeness Proof . . . . . . . . . . . . . . . . 7
IV. Fidelity to Modus Ponens . . . . . . . . . . . . . . . . . . 13

Chapter 2. More on First-Order Logic 23

I. Magic Sets . . . . . . . . . . . . . . . . . . . . . . . . . . 23
II. Gentzen Sequents and Some Variants . . . . . . . . . . . . 29
III. Craig’s Lemma and an Application . . . . . . . . . . . . . 43
IV. A Unification . . . . . . . . . . . . . . . . . . . . . . . . . 48
V. A Henkin-Style Completeness Proof . . . . . . . . . . . . 54
vi Contents

Part II: Recursion Theory and Metamathematics 67

Chapter 3. Some Special Topics 69

I. A Decision Machine . . . . . . . . . . . . . . . . . . . . . 69
II. Variations on a Theme of Gödel . . . . . . . . . . . . . . . 73
III. R-Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
IV. A Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . 79

Chapter 4. Elementary Formal Systems and Recursive

Enumerability 89

I. More on Elementary Formal Systems . . . . . . . . . . . . 89

II. Recursive Enumerability . . . . . . . . . . . . . . . . . . . 93
III. A Universal System . . . . . . . . . . . . . . . . . . . . . 99

Chapter 5. Some Recursion Theory 113

I. Enumeration and Iteration Theorems . . . . . . . . . . . . 113

II. Recursion Theorems . . . . . . . . . . . . . . . . . . . . . 117

Chapter 6. Doubling Up 133

Chapter 7. Metamathematical Applications 149

I. Simple Systems . . . . . . . . . . . . . . . . . . . . . . . . 149

II. Standard Simple Systems . . . . . . . . . . . . . . . . . . 154

Part III: Elements of Combinatory Logic 171

Chapter 8. Beginning Combinatory Logic 173

 Contents vii

Chapter 9. Combinatorics Galore 185

I. The B-Combinators . . . . . . . . . . . . . . . . . . . . . 185

II. The Permuting Combinators . . . . . . . . . . . . . . . . 187
III. The Q-Family and the Goldfinch, G . . . . . . . . . . . . 190
IV. Combinators Derivable from B,T, M and I
(λ-I Combinators) . . . . . . . . . . . . . . . . . . . . . . 192

Chapter 10. Sages, Oracles and Doublets 205

Chapter 11. Complete and Partial Systems 215

I. The Complete System . . . . . . . . . . . . . . . . . . . . 215

II. Partial Systems of Combinatory Logic . . . . . . . . . . . 224

Chapter 12. Combinators, Recursion and the Undecidable 233

I. Preparation for the Finale . . . . . . . . . . . . . . . . . . 236

II. The Grand Problem . . . . . . . . . . . . . . . . . . . . . 239

Afterword. Where to Go from Here 253

References 261

Index 265
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Preface

This book is a sequel to my Beginner’s Guide to Mathematical Logic

[Smullyan, 2014]. I originally intended both volumes to be a single vol-
ume, but I felt that at my age (now 96), I could pass away at any time,
and I wanted to be sure that I would at least get the basic material out.
The previous volume deals with elements of propositional and first-order
logic, contains a bit on formal systems and recursion, and concludes with
chapters on Gödel’s famous incompleteness theorem, along with related
results.
The present volume begins with a bit more on propositional and first-
order logic, followed by what I would call a “fein” chapter, which simultane-
ously generalizes some results from recursion theory, first-order arithmetic
systems, and what I dub a “decision machine.” Then come four chapters
on formal systems, recursion theory and metamathematical applications in
a general setting. The concluding five chapters are on the beautiful subject
of combinatory logic, which is not only intriguing in its own right, but has
important applications to computer science. Argonne National Laboratory
is especially involved in these applications, and I am proud to say that its
members have found use for some of my results in combinatory logic.
This book does not cover such important subjects as set theory, model
theory, proof theory, and modern developments in recursion theory, but the
reader, after studying this volume, will be amply prepared for the study of
these more advanced topics.
Although this book is written for beginners, there are two chapters −
namely 3 and 8 − that I believe would also be of interest to the expert.
For brevity, all references to the first volume, The Beginner’s Guide to
Mathematical Logic, of this two-volume introduction to mathematical logic
will be given in the remainder of this volume as The Beginner’s Guide
[Smullyan, 2014].

Elka Park
November 2016
 Part I

More on Propositional
and First-Order Logic
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 1

More on Propositional Logic

I. Propositional Logic and the Boolean

Algebra of Sets

Many of you have probably noticed the similarity of the logical connectives
to the Boolean operations on sets. Indeed, for any two sets A and B and
any element x, the element x belongs to the intersection A ∩ B if and only
if x is in A and x is in B. Thus x ∈ (A ∩ B) iff (x ∈ A) ∧ (x ∈ B). Thus the
logical connective ∧ (conjunction) corresponds to the Boolean operation ∩
(intersection). Likewise the logical connective ∨ (disjunction) corresponds
to the Boolean operation ∪ (union), since x ∈ (A ∪ B) iff (x ∈ A) ∨ (x ∈ B)
Also, x ∈ A (x is in the complement of A) if and only if ∼ (x ∈ A) (x is
not in A), so that the logical connective negation corresponds to Boolean
operation of complementation.
Note: As in The Beginner’s Guide [Smullyan, 2014], I will often abbre-
viate the phrase “if and only if” by “iff”, following the valuable suggestion
of Paul Halmos.
We saw in Chapter 1 of The Beginner’s Guide how to verify the cor-
rectness of a Boolean equation by the method of what I called “indexing”.
However, due to the correspondence between the logical connectives and
the Boolean operations on sets, one can also verify Boolean equations by
truth tables. The following single example will illustrate the general idea.
Consider the Boolean equation A ∩ B = A ∪ B. This equation is valid iff
for every element x, the element x is in A ∩ B iff x is in A ∪ B. Thus the
equation A ∩ B = A ∪ B is to the effect that for all x,
4 Part I: More on Propositional and First-Order Logic

x ∈ A ∩ B iff x ∈ A ∪ B. And
x ∈ A ∩ B iff (x ∈ A) ∧ (x ∈ B). Hence,
x ∈ A ∩ B iff ∼((x ∈ A) ∧ (x ∈ B)). Similarly,
x ∈ A ∪ B iff ∼(x ∈ A) ∨ ∼(x ∈ B).

Thus the proposition x ∈ A ∩ B iff x ∈ (A ∪ B) is equivalent to

∼((x ∈ A) ∧ (x ∈ B)) ≡ ∼(x ∈ A) ∨ ∼(x ∈ B).

And that formula is a tautology, for it is an instance of ∼(p ∧ q) ≡ ∼p ∨ ∼q,

as can be seen by replacing (x ∈ A) by p and replacing (x ∈ B) by q.
Similarly, the Boolean equation A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) holds
since p ∧ (q ∨ r) ≡ (p ∧ q) ∨ (p ∧ r) is a tautology.
In general, given a Boolean equation, define its counterpart to be the
result of substituting ∧ for ∩ and ∨ for ∪; ∼t for t (where t is any set
variable or Boolean term), ≡ for =, and propositional variables p, q, r, etc.,
for set variables A, B, C, etc. For example, the counterpart of the Boolean
equation A ∪ B = A ∩ B is the propositional formula ∼p ∨ q ≡ ∼(p ∧ ∼q).
Well, a Boolean equation is valid if and only if its counterpart is a
tautology of propositional logic. And since the formula ∼p ∨ q ≡ ∼(p ∧ ∼q)
is again easily seen to be a tautology, the Boolean equation A ∪ B = A ∩ B
is seen to be valid.
Thus, truth tables can be used to test Boolean equations. Going in the
other direction, the method of “indexing,” described in Chapter 1 of The
Beginner’s Guide [Smullyan, 2014] can be used to verify a statement of
propositional logic, after making the appropriate Boolean equation coun-
terpart to the propositional statement (found by reversing the substitutions
mentioned above).

II. An Algebraic Approach

This approach to propositional logic is particularly interesting and leads to

the subject known as Boolean rings.
For any two natural numbers x and y, if x and y are both even, so is
their product x × y. Thus an even times an even is even. An even times an
 Chapter 1. More on Propositional Logic 5

odd is even. An odd times an even is even, but an odd times an odd is odd.
Let us summarize this in the following table.
x y x×y
E E E
E O E
O E E
O O O

Note the similarity of this table to the truth table for disjunction.
Indeed, if we replace E by T, O by F, and × by ∨, we get the truth table
for disjunction:
x y x∨y
T T T
T F T
F T T
F F F

This is not surprising, since x times y is even if and only if x is even

OR y is even.
However, this might well suggest (as it did to me, many years ago!) that
we might transfer the problem of whether a given formula is a tautology, to
the question of whether or not a certain algebraic expression in variables for
natural numbers is always even (for all possible values of the variables). So
let us now do this, i.e. define a transformation of a formula of propositional
logic into an arithmetic expression (using only + and × and variables for
natural numbers) such that the propositional formula will be a tautology
iff every number corresponding to the numerical value of an instance of the
arithmetic expression (i.e. an instance in which natural numbers replace
the variables) is even.
We let even correspond to truth, odd to falsehood, and multiplication
to disjunction; then we take the letters p, q, r, . . . both as variables for
propositions in logical expressions and as variables for natural numbers in
the corresponding arithmetic expressions. Thus the logical expression p ∨ q
is now transformed to the arithmetic expression p × q, or more simply, pq.
What should we take for negation? Well, just as ∼ transforms truth into
falsehood, and falsehood into truth, we want an operation that transforms
an even number to an odd one, and an odd one to an even one. The obvious
6 Part I: More on Propositional and First-Order Logic

choice is to add 1. Thus we let the logical expression ∼p correspond with

the numerical expression p + 1.
As was seen in Chapter 5 of The Beginner’s Guide [Smullyan, 2014],
once we have ∨ and ∼, we can obtain all the other logical connectives. We
take p ⊃ q (implication) to be ∼p ∨ q, which transforms to (p + 1)q, which
is pq + q.
What about conjunction? Well, we take p ∧ q to be ∼(∼p ∨ ∼q), and
so in the numerical interpretation we take p ∧ q to be ((p + 1)(q + 1)) + 1,
which is (pq + p + q + 1) + 1, which is pq + p + q + 2. But 2 is an even number,
which means that for any number z, the number z + 2 has the same parity
as z (i.e. z + 2 is even if and only if z is even). And so we can drop the
final 2, and take p ∧ q to be pq + p + q.
We could take p ≡ q to be (p ⊃ q) ∧ (q ⊃ p) or we could take p ≡ q to
be ((p ∧ q) ∨ (∼p ∧ ∼q)). But it is much simpler to do the following: Now,
p ≡ q is true iff p and q are either both true or both false. Thus in our
numerical interpretation, we want the number p ≡ q to be even if and only
if p and q are either both even or both odd. And the simple operation that
does the trick is addition: we thus take p ≡ q to be p + q.
We note that p ∨ p is equivalent to p and thus p × p has the same
parity as p. So we can replace p × p (or pp) by p. Also, for any number
p, p + p is always even, and so we can replace p + p with 0 (which is
always even). Moreover, we can eliminate any even number that occurs as
an addend, because eliminating an even addend does not change the parity
of a numerical expression: e.g. p + 0 and p + 4 is of the same parity as p, so
can be replaced by p.
Under the transformations just described, a formula is a tautology in
the usual propositional interpretation, if and only if it is always even in its
numerical interpretation.
Let us consider the tautology (p ⊃ q) ≡∼ (p ∧ ∼q). Well, p ⊃ q corre-
sponds to pq + q. Now, p ∧ ∼q transforms to p(q + 1) + p + (q + 1), which
simplifies to pq+p+p+q+1. We can drop p+p, which is always even, and so
the transformation of p ∧ ∼q reduces to pq + q + 1. Thus ∼(p ∧ ∼q) reduces
to pq + q + 1 + 1, but 1 + 1 is even, and can be dropped, so that ∼(p ∧ ∼q)
reduces to pq + q, which is the same thing that p ⊃ q reduced to. Thus
(p ⊃ q) ≡ ∼(p ∧ ∼q) reduces to (pq + q) + (pq + q), which expression can
be reduced to 0 (since, as mentioned above, the sum of the same two equal
 Chapter 1. More on Propositional Logic 7

numbers is always even). We have thus shown that (p ⊃ q) ≡ ∼(p ∧ ∼q)

is a tautology by showing that the corresponding numerical expression it
reduces to (in this case 0) is always even.
Note: Readers familiar with modern higher algebra know that a
Boolean ring is a ring with the additional properties that x + x = 0 and
x2 = x for any element x of the ring.

III. Another Completeness Proof

We now turn to a completeness proof for propositional logic of an entirely

different nature, and one which has a particularly interesting extension to
first-order logic, which we consider in the next chapter. It is based on the
notion of maximal consistency.
We will consider axiom systems using ∼, ∧, ∨, and ⊃ as undefined con-
nectives, and in which modus ponens (from X and X ⊃ Y , to infer Y )
is the only rule of inference. We will say that a Boolean truth set is a set
S of formulas such that for any pair of formulas X and Y , the following
conditions hold:
(1) (X ∧ Y ) ∈ S iff X ∈ S and Y ∈ S.
(2) (X ∨ Y ) ∈ S iff either X ∈ S or Y ∈ S.
(3) (X ⊃ Y ) ∈ S iff either X ∈
/ S or Y ∈ S.
(4) (∼X) ∈ S iff X ∈ / S (thus, one and only one of X, ∼X is in S).
Notes: [a] Whenever necessary to avoid a possible misreading, given
a formula X and a set S, I will abbreviate the statement that ∼X is a
member of S by (∼X) ∈ S instead of by ∼X ∈ S, as the latter could be
read as “It is not the case that X is in S.” Similarly, I write (∼X) ∈ / S
rather than ∼X ∈ / S to mean that ∼X is not in S, for ∼X ∈ / S might be
read as ∼(X ∈ / S), i.e. that it is not the case that X is not a member of S
(so that X is in S). Outer parentheses around formulas are often omitted
for easier reading, but may be included for greater clarity of meaning; for
instance, parentheses are used to distinguish between (X ∧ Y ) ∈ S from
X ∧ (Y ∈ S), two possible readings of X ∧ Y ∈ S (where X and Y are
variables for sentences, and S is a variable representing a set of sentences).
[b] The reader should also realize that for any statement of the form “X iff
Y ”, the statement is equivalent to its negated form “∼X iff ∼Y ”, by the
8 Part I: More on Propositional and First-Order Logic

very meaning of the term “iff” (which is an abbreviation for “if and only
if”). Thus,
“(∼X) ∈ S iff X ∈ / S”
is equivalent to
“(∼X) ∈
/ S iff X ∈ S”.

If I refer to an assumption or condition in a proof or a problem solution, but

want the reader to take it in its equivalent “negated” form, I often indicate
this fact.
We also recall from The Beginner’s Guide [Smullyan, 2014] the unifying
α, β notation in propositional logic for unsigned formulas:

α α1 α2
β β1 β2
X ∧Y X Y
∼(X ∧ Y ) ∼X ∼Y
∼(X ∨ Y ) ∼X ∼Y
X ∨Y X Y
∼(X ⊃ Y ) X ∼Y
X⊃Y ∼X Y
∼∼X X X

Problem 1. S is a Boolean truth set if and only if for every α and every
formula X the following conditions hold:
(1) α ∈ S iff α1 ∈ S and α2 ∈ S.
(2) Either X ∈ S or (∼X) ∈ S, but not both.

Exercise 1. Prove that S is a Boolean truth set iff for every β and every
formula X the following conditions hold:
(1) β ∈ S iff either β1 ∈ S or β2 ∈ S.
(2) Either X ∈ S or (∼X) ∈ S, but not both.

The definition of a Boolean truth set here can be shown to be equivalent

to the one given for a truth set in Chapter 6 (Propositional Tableaux) of
The Beginner’s Guide [Smullyan, 2014], where we called a set S of propo-
sitional formulas a truth set if it was the set of all formulas true under
some interpretation I of the propositional variables. We also showed there
that the definition was equivalent to the set S satisfying the following three
conditions: for any signed formula X and any α and β:
T0 : Either X or its conjugate X is in S, but not both.
T1 : α is in S if and only if α1 and α2 are both in S.
T2 : β is in S if and only if either β1 is in S or β2 is in S.
 Chapter 1. More on Propositional Logic 9

With these facts in mind, the reader should easily see that the definitions
of (Boolean) truth set in The Beginner’s Guide and this work are equivalent.
Note that once one sees the equivalence between the definitions in the two
books, it is obvious that every Boolean truth set for propositional logic
is satisfiable (since the characterization in The Beginner’s Guide made a
truth set satisfiable by definition).
Indeed, to go a little further (to make connections between very similar
concepts a little clearer), we often define a Boolean valuation as follows.
First, by a valuation (in the context of propositional logic) is meant an
assignment of truth values t and f to all formulas. A valuation v is called
a Boolean valuation if for all formulas X and Y , the following conditions
hold:
B1 : ∼X receives the value t under v if and only if v(X) = f [i.e. when X
is false under v]; and accordingly v(∼X) = f iff v(X) = t.
B2 : X ∧ Y receives the value t under v iff v(X) = v(Y ) = t [i.e. when both
X and Y are true under the valuation v].
B3 : X ∨ Y receives the value t under v iff v(X) = t or v(Y ) = t [i.e.
when either X is true under v or Y is true under v, or both are true
under v].
B4 : X ⊃ Y receives the value t under v iff v(X) = f or v(Y ) = t [i.e. when
either X is false under v or Y is true under v].
The interpretations that we used in The Beginner’s Guide in terms
of assigning truth values to all formulas by assigning truth values to the
propositional variables, and then working up through the structure of the
formulas, using the meanings of the connectives, to obtain a truth value
for every formula, are closely related to the truth values of formulas given
through a Boolean valuation. In fact a given Boolean valuation defines a
unique interpretation yielding the same truth value on all formulas in the
interpretation as in the Boolean valuation: we obtain the interpretation by
looking at the truth values the Boolean valuation assigns to the proposi-
tional variables. And if one is given an interpretation of the propositional
variables, it is clear that the interpretation defines a Boolean valuation
on the formulas, just by the way the interpretation assigns truth values
to all formulas, starting out only with truth values on the propositional
variables.
Because of this tight equivalence between Boolean valuations and inter-
pretations in propositional logic, we will use the words “valuation” and
10 Part I: More on Propositional and First-Order Logic

“interpretation” interchangeably (and will do so later in first-order logic as

well, where a similar tight relationship holds, this being something that was
already discussed in The Beginner’s Guide [Smullyan, 2014]).

Consequence Relations

We consider a relation S ` X between a set S of formulas and a formula X.

We read S ` X as “X is a consequence of S,” or as “S yields X”.
We call the relation ` a consequence relation if for all sets S, S1 and S2
and all formulas X and Y , the following conditions hold:
C1 : If X ∈ S, then S ` X.
C2 : If S1 ` X and S1 ⊆ S2 then S2 ` X.
C3 : If S ` X and S, X ` Y , then S ` Y .
C4 : If S ` X, then F ` X for some finite subset F of S.
In what follows, we assume that ` is a consequence relation.
We write `X to mean ∅ ` X, where ∅ is the empty set. For any n, we
write X1 , . . . , Xn ` Y to mean {X1 , . . . , Xn } ` Y, and for any set S, we
write S, X1 , . . . , Xn to mean S ∪ {X1 , . . . , Xn }.
An important example of a consequence relation is this. For any axiom
system A, take S ` X to mean that X is derivable from the elements of S
together with the axioms of A by use of the inference rules of the system;
in other words, if we take the elements of S and add them as new axioms of
A, then X becomes provable in that enlarged system. Note that under this
relation `, the statement ` X simply says that X is provable in the system
A. It is obvious that this relation ` satisfies conditions C1 , C2 , and C3 .
As for C4 , if X is derivable from S, any derivation uses only finitely many
elements of S.
We now return to consequence relations in general.

Problem 2. Prove that if S ` X and S ` Y , and X, Y ` Z, then S ` Z.

Consistency

In what follows, “set” shall mean a set of formulas.

We shall call a set S inconsistent (with respect to a consequence relation-
ship `) if S ` Z for every formula Z; otherwise, we shall call S consistent.
 Chapter 1. More on Propositional Logic 11

Problem 3. Show that if S is consistent, and S ` X, then S, X is

consistent.

Problem 4. Suppose S is consistent. Show the following:

(a) If S ` X and X ` Y , then S, Y is consistent.
(b) If S ` X and S ` Y and X, Y ` Z, then S, Z is consistent.

Maximal Consistency

We say that S is maximally consistent (with respect to `) if no proper

superset of S is consistent. [By a proper superset of S is meant a set S 0
such that S ⊆ S 0 and S 0 contains one or more elements not in S.]

Problem 5. Show that if M is maximally consistent, then for all formulas

X, Y and Z:
(a) If X ∈ M and X ` Y , then Y ∈ M .
(b) If X ∈ M and Y ∈ M and X, Y ` Z then Z ∈ M .

Boolean Consequence Relations

We shall define a consequence relation ` to be a Boolean consequence rela-

tion if it satisfies the following additional conditions (for all formulas α, X
and Y ):
C5 : (a) α ` α1 ; (b) α ` α2 .
C6 : α1 , α2 ` α.
C7 : X, ∼X ` Y.
C8 : If S, ∼X ` X, then S ` X.
We say that an interpretation satisfies a set S if all elements of S are true
under the interpretation. We shall say that X is a tautological consequence
of S if X is true under all interpretations that satisfy S.

Exercise 2. Define S |= X to mean that X is a tautological consequence

of S. Prove that this relation |= is a Boolean consequence relation. (Hint:
For C4 , use the Corollary to the Compactness Theorem for Propositional
Logic on p. 95 in The Beginner’s Guide [Smullyan, 2014].)
We say a consequence relation ` is tautologically complete if, for every
set S and formula X, whenever X is a tautological consequence of S,
then S ` X.
12 Part I: More on Propositional and First-Order Logic

Problem 6. Suppose ` is tautologically complete. Does it necessarily follow

that ∅ ` X for every tautology X? [We recall that ∅ is the empty set.]
The following, which we aim to prove, will be our main result:
Theorem 1. Every Boolean consequence relation is tautologically complete.
Thus, if ` is a Boolean consequence relation, then:
(1) If X is a tautological consequence of S, then S ` X.
(2) If X is a tautology, then ` X (i.e. ∅ ` X). This means that every system
of propositional logic based on a Boolean consequence relation is also
complete (i.e. that all tautologies are provable within the system).

To prove Theorem 1, we will need the following lemma:

Lemma 1. [Key Lemma] If ` is a Boolean consequence relation and M is

maximally consistent with respect to `, then M is a Boolean truth set.

Problem 7. Prove Lemma 1.

Compactness

We recall from Chapter 4 of The Beginner’s Guide [Smullyan, 2014] that

a property P of sets is called compact if for any set S, S has the property
P if and only if all finite subsets of S have property P . Thus, to say that
consistency is compact is to say that a set S is consistent if and only if all
finite subsets of S are consistent.

Problem 8. [A key fact!] Show that for a Boolean consequence relation `,

consistency (with respect to `) is compact.

Let us also recall the denumerable compactness theorem of Chapter 4

of The Beginner’s Guide, which says that for any compact property P
of subsets of a denumerable set A, any subset of A having property P
can be extended to (i.e. is a subset of) a maximal subset of A having
property P . We have just shown that for a Boolean consequence relation,
consistency is a compact property. Hence, any consistent set is a subset of
a maximally consistent set. This fact, together with the Key Lemma, easily
yields Theorem 1.

Problem 9. Now prove Theorem 1.

 Chapter 1. More on Propositional Logic 13

IV. Fidelity to Modus Ponens

We shall say that a consequence relation ` is faithful to modus ponens if

X, X ⊃ Y ` Y
holds for all X and Y.
In what follows, we shall assume that ` is faithful to modus ponens. We
shall say that a formula X is provable with respect to ` when ` X holds
(i.e. if ∅ ` X). Until further notice, “provable” will mean provable with
respect to `.

Problem 10. Show that if S ` X and S ` X ⊃ Y , then S ` Y.

Problem 11. Show the following:
(a) If X ⊃ Y is provable, and S ` X, then S ` Y.
(b) If X ⊃ (Y ⊃ Z) is provable, and S ` X and S ` Y , then S ` Z.
(c) If X ⊃ Y is provable, then X ` Y.
(d) If X ⊃ (Y ⊃ Z) is provable, then X, Y ` Z.
We now consider an axiom system A for propositional logic in which
modus ponens is the only inference rule. Such a system I will call a standard
system. By a deduction from a set S of formulas is meant a finite sequence
X1 , . . . , Xn of formulas, such that for each term Xi of the sequence, Xi is
either an axiom (of A), or a member of S, or is derivable from two earlier
terms by one application of modus ponens. We call a deduction X1 , . . . , Xn
from S a deduction of X from S if Xn = X. We say that X is deducible
from S to mean that there is a deduction of X from S. We now take S ` X
to be the relation that X is deducible from S. This relation is indeed a
consequence relation.

Problem 12. Prove that ` is a consequence relation.

It is obvious that this relation ` is faithful to modus ponens, since the
sequence X, X ⊃ Y, Y is a deduction of Y from {X, X ⊃ Y }.
We say that the axiom system A has the deduction property if for all
sets S and formulas X and Y, if S, X ` Y, then S ` X ⊃ Y.

Theorem 2. [The Deduction Theorem] A sufficient condition for an axiom

system A to have the deduction property is that the following are provable
14 Part I: More on Propositional and First-Order Logic

in A (for all X, Y and Z) :

A1 : X ⊃ (Y ⊃ X).
A2 : (X ⊃ (Y ⊃ Z)) ⊃ ((X ⊃ Y ) ⊃ (X ⊃ Z)).

For this, we need:

Lemma 2. If A1 and A2 hold, then X ⊃ X is provable.

Until further notice, we assume that A1 and A2 hold.

Problem 13. To prove the above lemma, show that for any formulas X
and Y , the formula (X ⊃ Y ) ⊃ (X ⊃ X) is provable. Then, taking Y ⊃ X
for Y in the formula just proved, conclude the proof of Lemma 2.

Problem 14. In preparation for the proof of the Deduction Theorem, show
the following:
(a) If S ` Y , then S ` X ⊃ Y.
(b) If S ` X ⊃ (Z ⊃ Y ) and S ` X ⊃ Z, then S ` X ⊃ Y.

Problem 15. Prove the Deduction Theorem as follows: Suppose S, X ` Y.

Then there is a deduction Y1 , . . . , Yn from S ∪ {X} in which Y = Yn .
Now consider the sequence X ⊃ Y1 , . . . , X ⊃ Yn . Show that for all i ≤ n,
if S ` X ⊃ Yj for all j < i, then S ` X ⊃ Yi . [Hint: Break the
proof up into the following four cases: Yi is an axiom of A; Yi ∈ S;
Yi = X; Yi comes from two terms of the sequence Y1 , . . . , Yi−1 by modus
ponens.]
It then follows by complete mathematical induction that S ` X ⊃ Yi for
all i ≤ n, and so in particular, S ` X ⊃ Y n , and thus S ` X ⊃ Y. [Hint:
In the original sequence Y1 , . . . , Yn , the term Yi is either an axiom of A, or
is X itself, or comes from two earlier terms by modus ponens. Handle each
case separately.]
We now have all the parts necessary to prove:

Theorem 3. A sufficient condition for a standard axiom system for propo-

sitional logic A to be tautologically complete is that for all X, Y, Z and α,
the following are provable:
A1 : X ⊃ (Y ⊃ X)
A2 : (X ⊃ (Y ⊃ Z)) ⊃ ((X ⊃ Y ) ⊃ (X ⊃ Z))
A3 : (a) α ⊃ α1 and (b) α ⊃ α2
A4 : α1 ⊃ (α2 ⊃ α)
 Chapter 1. More on Propositional Logic 15

A5 : X ⊃ (∼X ⊃ Y )
A6 : (∼X ⊃ X) ⊃ X

Problem 16. Prove Theorem 3 by showing that under the conditions given
in Theorem 3, the relation ` is a Boolean consequence relation and thus
tautologically complete.
One can easily verify that A1 − A6 are all provable in the axiom system
S of Chapter 7 of The Beginner’s Guide [Smullyan, 2014]. [One can make
a tableau proof for each of these formulas and then convert them to proofs
in S by the method explained in that chapter. The completeness proof
for tableaux is not needed for this.] We thus have a second proof of the
completeness of S based on maximal consistency, and thus the theorem:

Theorem 4. The axiom system S of Chapter 7 of The Beginner’s Guide

is tautologically complete.
Maximal consistency and Lindenbaum’s Lemma also play a key role in
alternative completeness proofs for first-order logic, to which we turn in the
next chapter.

Solutions to the Problems of Chapter 1

1. We are to show that S is a Boolean truth set iff the two conditions
of Problem 1 hold. The second condition of Problem 1 is the same as
the fourth condition in the definition of the Boolean truth set, so we
need only consider the first condition of Problem 1. (Remember, when
thinking about “negated” conditions or assumptions, that ∼(X ∧ Y ) is
equivalent to ∼X ∨ ∼Y and ∼(X ∨ Y ) is equivalent to ∼X ∧ ∼Y.)

(a) Suppose S is a Boolean truth set. We must show that in all cases
for α, the first condition of Problem 1 holds, namely that α ∈ S iff
α1 ∈ S and α2 ∈ S.
Case 1. α is of the form X ∧ Y. This is immediate.
Case 2. α is of the form ∼(X ∨ Y ). Then α1 = ∼X and α2 = ∼Y .
Thus we are to show that (∼(X ∨ Y )) ∈ S iff (∼X) ∈ S and
(∼Y ) ∈ S. Well,
• (∼(X ∨ Y )) ∈ S is true iff (X ∨ Y ) ∈
/ S [by (4) in the definition
of a Boolean truth set].
16 Part I: More on Propositional and First-Order Logic

• Then (X ∨ Y ) ∈ / S is true iff X ∈

/ S and Y ∈ / S [by the negated
form of (2) in the definition of a Boolean truth set].
• And X ∈ / S and Y ∈ / S is true iff (∼X) ∈ S and (∼Y ) ∈ S [by
(4) in the definition of a Boolean truth set].
Case 3. α is of the form ∼(X ⊃ Y ). Then α1 = X and α2 = ∼Y .
We are to show that (∼(X ⊃ Y )) ∈ S iff X ∈ S and (∼Y ) ∈ S.
Well,
• (∼(X ⊃ Y )) ∈ S is true iff X ⊃ Y ∈ / S [by (4) in the definition
of a Boolean truth set].
• Then X ⊃ Y ∈ / S is true iff X ∈ S and Y ∈ / S [by the negated
form of (3) in the definition of a Boolean truth set].
• And X ∈ S and Y ∈ / S is true iff X ∈ S and (∼Y ) ∈ S [by (4)
in the definition of a Boolean truth set].
Case 4. α is of the form ∼∼X. Here both α1 and α2 are X, so we must
show that (∼∼X) ∈ S iff X ∈ S. Well, (∼∼X) ∈ S iff (∼X) ∈ / S [by
(4) in the definition of a Boolean truth set], which is true iff X ∈ S
[again by (4) in the definition of a Boolean truth set].
(b) Going in the other direction, suppose the conditions of Problem 1
hold: (i) α ∈ S iff α1 ∈ S and α2 ∈ S, and (ii) for any X, either
X ∈ S or ∼X ∈ S, but not both. We are to show that S is a
Boolean truth set.
Case 1. We are to show that X ∧ Y ∈ S iff X ∈ S and Y ∈ S: Well
X ∧ Y is an α, where X = α1 and Y = α2 and α ∈ S iff α1 ∈ S
and α2 ∈ S [by assumption (i) above]. Thus it is immediate that
X ∧ Y ∈ S iff X ∈ S and Y ∈ S.
Case 2. We are to show that X ∨ Y ∈ S iff X ∈ S or Y ∈ S: Note
that here ∼(X ∨ Y ) is an α, with α1 = ∼X and α2 = ∼Y. Well,
• X ∨ Y ∈ S iff ∼(X ∨ Y ) ∈ / S [by assumption (ii) above].
• ∼(X ∨ Y ) ∈/ S is true iff (∼X) ∈ / S or (∼Y ) ∈
/ S [by the negated
form of assumption (i)].
• Then (∼X) ∈ / S iff X ∈ S and (∼Y ) ∈ / S is true Y ∈ S [by
assumption (ii) above].

Case 3. We are to show that (X ⊃ Y ) ∈ S iff X ∈

/ S or Y ∈ S.
Note that here ∼(X ⊃ Y ) is an α, with α1 = X and α2 = ∼Y.
Well,
• (X ⊃ Y ) ∈ S iff ∼(X ⊃ Y ) ∈
/ S [by assumption (ii)].
 Chapter 1. More on Propositional Logic 17

• Then ∼(X ⊃ Y ) ∈ / S is true iff X ∈

/ S or (∼Y ) ∈
/ S [by the
negated form of assumption (i) above].
• Since it is true that (∼Y ) ∈
/ S iff Y ∈ S [by assumption (ii)
above], we now have X ∈ / S or Y ∈ S, which is what was to be
shown.
Case 4. To show that (∼X) ∈ S iff X ∈ / S: As noted in the
definition of a Boolean truth set, condition (4) of a truth set and
condition (2) of Problem 1 are equivalent.
2. Suppose S ` X and S ` Y and X, Y ` Z. Since X, Y ` Z, then
S, X, Y ` Z [by C2 , since {X, Y } ⊆ S ∪ {X, Y }]. Since S ` Y, then
S, X ` Y [again by C2 ]. Thus S, X ` Y and S, X, Y ` Z. Hence [by
C3 ], S, X ` Z. [To see this more easily, let W = S ∪ {X}.] Then W ` Y
(since this is the same as S, X ` Y ). And W, Y ` Z (since this is the
same as S, X, Y ` Z). Hence W ` Z, by C3 , taking W for S. Then,
since S, X ` Z and S ` X (given), then S ` Z [again by C3 ].
3. Suppose S ` X. We are to show that if S is consistent, so is S, X (i.e.
so is the set S ∪ {X}). We will show the equivalent fact that if S, X is
inconsistent, so is S. Well, suppose S, X is inconsistent. Thus S, X ` Z
for every Z. Since also S ` X (which is given), then S ` Z [by C3 ], and
thus S is also inconsistent.
4. Suppose S is consistent.
(a) Assume S ` X and X ` Y . We are to show that S, Y is consistent.
Well, since X ` Y , then S, X ` Y [by C2 ]. Since also S ` X, we
have S ` Y [by C3 ]. Since S is consistent and S ` Y , then S, Y is
consistent [by Problem 3].
(b) Assume S ` X and S ` Y and X, Y ` Z. We are to show that S, Z
is consistent. Well, since S ` X and S ` Y and X, Y ` Z, it must
also be true that S ` Z [by Problem 2]. Then, since S is consistent
and S ` Z, we have that S, Z is consistent [by Problem 3].
5. Suppose M is maximally consistent. We first note that if M ` X,
then X ∈ M . For suppose M ` X. Since M is consistent, then M, X
is consistent [by Problem 3]. Consequently M, X cannot be a proper
superset of M (no consistent set is). Thus X ∈ M . Now we consider
the two parts of the problem.
(a) Suppose X ∈ M and X ` Y . Since X ∈ M , then M ` X [by C1 ].
Since X ` Y and M ` X, we obtain M ` Y [by C2 ]. Hence, as just
noted, Y ∈ M .
18 Part I: More on Propositional and First-Order Logic

(b) Suppose X ∈ M and Y ∈ M and X, Y ` Z. Then M ` X, M ` Y

[by C1 ] and X, Y ` Z. Thus M ` Z [by Problem 2]. Hence Z ∈ M ,
as noted above.
6. Yes, if ` is tautologically complete, then ∅ ` X for every tauto-
logy X. Here is the proof. Since there are no elements in the empty
set ∅, it is vacuously true that every element of the empty set is true
under all interpretations. (If you don’t believe this, just try to find an
element of the empty set which is not true under all interpretations!)
Thus every interpretation satisfies both ∅ and X, so X is a tautological
consequence of ∅. Then, since ` is tautologically complete, it follows
from the definition of a consequence relation being tautologically com-
plete that ∅ ` X (in fact S ` X holds for every set S [by C2 ]).
7. We are given that ` is a Boolean consequence relation and that M is
maximally consistent with respect to `. By virtue of Problem 1, to show
that M is a Boolean truth set, it suffices to show that the following two
things hold for every α and every X:
(1) α ∈ M if and only if α1 ∈ M and α2 ∈ M .
(2) Either X ∈ M or (∼X) ∈ M , but not both.
Regarding (1): Suppose α ∈ M . Since α ` α1 [by C5 (a)], then α1 ∈ M
[by Problem 5(a)]. Thus if α ∈ M , then α1 ∈ M . Similarly, if
α ∈ M , then α2 ∈ M [by C5 (b)]. Thus if α ∈ M , then α1 ∈ M
and α2 ∈ M .
Conversely, suppose α1 ∈ M and α2 ∈ M . We know that α1 , α2 ` α
[by C6 ]. Hence α ∈ M [by Problem 5(b)].
Thus α ∈ M if and only if α1 ∈ M and α2 ∈ M .
Regarding (2): Since M is consistent, it cannot be that X and ∼X are
both in M , for if they were, then we would have both M ` X and
M ` ∼X, and since X, ∼X ` Z for every Z [by C7 ], we would have
M ` Z [by Problem 2], which would mean that M is inconsistent.
Thus, since M is consistent, it cannot be that X ∈ M and ∼X ∈ M . It
remains to show that either X ∈ M or (∼X) ∈ M . Let me show that
if ∼X is not in M , then X ∈ M .
Well, suppose that ∼X is not in M . Then M, ∼X is not consis-
tent (since M is maximally consistent). Hence M, ∼X ` X (since M,
∼X ` Z for every Z), and therefore M ` X [by C8 ]. Thus X ∈ M
(since M is maximally consistent).
This completes the proof.
 Chapter 1. More on Propositional Logic 19

8. To show that if all finite subsets of a set S are consistent, then so is S,

it suffices to show that if S is inconsistent, then some finite subset of S
is inconsistent. Well, suppose S is inconsistent. Then for any formula
X, we have both S ` X and S ` ∼X (since S ` Z for every Z). Since
S ` X, then F1 ` X for some finite subset F1 of S [by C4 ]. Similarly,
F2 ` ∼X for some finite subset F2 of S. Let F = F1 ∪ F2 . Then F ` X
and F ` ∼X [by C2 ], since F1 ⊆ F and F2 ⊆ F ). Also, X, ∼X ` Z
for all Z [by C7 ]. Thus, for all Z, F ` Z [by Problem 2], and we have
shown that F is inconsistent.
9. We wish to show that every Boolean consequence relation is tautolo-
gically complete. By Problem 8 we know that for any Boolean
consequence relation, consistency with respect to that relation is a
compact property, and so [by the denumerable compactness theo-
rem] every consistent set is a subset of a maximally consistent set,
which must be a Boolean truth set [by the Key Lemma]. Thus
every consistent set is satisfiable (by the discussion after the defi-
nition of a Boolean truth set), and hence every unsatisfiable set is
inconsistent. Now, suppose X is a logical consequence of S. Then
S, ∼X is unsatisfiable, hence inconsistent. Therefore S, ∼X ` X (since
S, ∼X ` Z for every Z), and so S ` X [by C8 ]. This concludes
the proof.
10. Suppose S ` X and S ` X ⊃ Y . It is also true that X, X ⊃ Y ` Y , since
we assume ` is faithful to modus ponens. Hence S ` Y [by Problem 2].
11. Proofs of the four statements:
(a) Suppose X ⊃ Y is provable and S ` X. Since X ⊃ Y is prov-
able (∅ ` X ⊃ Y ), then S ` X ⊃ Y [by C2 ]. Thus S ` X and
S ` X ⊃ Y ; hence S ` Y [by Problem 10].
(b) Suppose X ⊃ (Y ⊃ Z) is provable, and S ` X and S ` Y. Since
S ` X and S ` X ⊃ (Y ⊃ Z) [by C2 ], then S ` Y ⊃ Z [by
Problem 10]. Since also S ` Y , then S ` Z [again by Problem 10].
(c) In 11(a), we take the unit set {X} for S, and thus if X ⊃ Y is
provable and X ` X, then X ` Y. But X ` X does hold, so that if
X ⊃ Y is provable, then X ` Y.
(d) In 11(b), we take the set {X, Y } for S, and thus if X ⊃ (Y ⊃ Z)
is provable, and X, Y ` X, and X, Y ` Y , then X, Y ` Z. But
X, Y ` X and X, Y ` Y both hold, and so if X ⊃ (Y ⊃ Z) is
provable, then X, Y ` Z.
12. We must verify that conditions C1 −C4 hold.
20 Part I: More on Propositional and First-Order Logic

C1 : If X ∈ S, then the unit sequence X (i.e. the sequence whose only

term is X) is a deduction of X from S.
C2 : Suppose S1 ⊆ S2 . Then any deduction Y1 , . . . , Yn from S1 is also a
deduction from S2 .
C3 : If Y1 , . . . , Yn is a deduction of X from S, and Z1 , . . . , Zk is a deduc-
tion of Y from S, X, then the sequence Y1 , . . . , Yn , Z1 , . . . , Zk is a
deduction of Y from S (as is easily verified).
C4 : Any deduction from S uses only finitely many elements X1 , . . . , Xn
of S, hence is also a deduction from the set X1 , . . . , Xn .
13. (a) We assume A1 and A2 hold. In A2 we take X for Z, and obtain
` (X ⊃ (Y ⊃ X)) ⊃ ((X ⊃ Y ) ⊃ (X ⊃ X)).
But by A1 , we have ` X ⊃ (Y ⊃ X). Hence, by modus ponens, we
have
` (X ⊃ Y ) ⊃ (X ⊃ X).
(b) In (X ⊃ Y ) ⊃ (X ⊃ X), taking (Y ⊃ X) for Y , we have
` (X ⊃ (Y ⊃ X)) ⊃ (X ⊃ X).
But by A1 we have that X ⊃ (Y ⊃ X) is provable. So, by modus
ponens, we also have ` X ⊃ X.
14. (a) Suppose S ` Y . Now, by (a) of Problem 11, for any formulas W1
and W2 , if S ` W1 and W1 ⊃ W2 is provable, then S ` W2 . Let us take
Y for W1 and X ⊃ Y for W2 . When we do so, we see that if S ` Y
and Y ⊃ (X ⊃ Y ) is provable, then S ` X ⊃ Y . However, S ` Y [by
assumption] and Y ⊃ (X ⊃ Y ) is provable [by A1 ]; hence S ` X ⊃ Y .
(b) By Problem 11(b), for any formulas W1 , W2 , and W3 , if
W1 ⊃ (W2 ⊃ W3 ) is provable, then if S ` W1 and S ` W2 , it follows
that S ` W3 . We now take X ⊃ (Z ⊃ Y ) for W1 , X ⊃ Z for W2 , and
X ⊃ Y for W3 . Well, the formula W1 ⊃ (W2 ⊃ W3 ) is provable [by A2 ;
it is the formula (X ⊃ (Z ⊃ Y )) ⊃ ((X ⊃ Z) ⊃ (X ⊃ Y ))]. Hence, if
S ` W1 and S ` W2 , then S ` W3 , i.e. we have that if S ` X ⊃ (Z ⊃ Y )
and S ` X ⊃ Z, then S ` X ⊃ Y , which is what was to be proved.
15. We assume that for all j < i, S ` X ⊃ Yj , and we are to prove that
S ` X ⊃ Yi .
Cases 1 and 2. Suppose that either Yi is an axiom of A, or that Yi ∈ S.
If the former, then Yi is provable, so that ` Yi , and so S ` Yi [by C2 ]. If
Yi ∈ S, then again S ` Yi [by C1 ]. Thus, in either case, S ` Yi . Hence
S ` X ⊃ Yi , [by Problem 14(a)].
 Chapter 1. More on Propositional Logic 21

Case 3. Yi = X. Since X ⊃ X is provable [by Lemma 2], then

S ` X ⊃ X [by C2 ], and since Yi = X, then X ⊃ Yi is the formula
X ⊃ X, and so S ` X ⊃ Yi .
Case 4. This is the case where in the original sequence Y1 , . . . , Yn , the
formula Yi came from terms earlier than Yi by modus ponens. Thus
there is a formula Z such that Z and Z ⊃ Yi are earlier terms than Yi .
Hence in the new sequence X ⊃ Y1 , . . . , X ⊃ Yn the formulas X ⊃ Z
and X ⊃ (Z ⊃ Yi ) are earlier terms than X ⊃ Yi , and so, by the
inductive hypothesis, S ` X ⊃ Z and S ` X ⊃ (Z ⊃ Yi ). Then by
Problem 14(b), S ` X ⊃ Y i . This completes the proof.
16. We already know that conditions C1 −C4 hold, and so it remains to
verify conditions C5 −C8 .
Re C5 : (a) Since α ⊃ α1 is provable [by A3 (a)], then α ` α1 [by Prob-
lem 11(c)].
(b) Since α ⊃ α2 is provable [by A3 (b)], then α ` α2 [by Prob-
lem 11(c)].
Re C6 : Since α1 ⊃ (α2 ⊃ α) is provable [by A4 ], then α1 , α2 ` α [by
Problem 11(d)].
Re C7 : Since X ⊃ (∼X ⊃ Y ) is provable [by A5 ], then X, ∼X ` Y [by
Problem 11(d)].
Re C8 : [This is the only part that needs the Deduction Theorem!]
Suppose that S, ∼X ` X Then, by the deduction theorem,
S ` ∼X ⊃ X. But also (∼X ⊃ X) ⊃ X is provable [by A6 ].
Therefore S ` X [by Problem 11(a), taking ∼X ⊃ X for X
and X for Y ].
This concludes the proof that ` is a Boolean consequence relation, and
so by Theorem 1, it follows that all tautologies are provable in the
system (and more generally, that if X is a logical consequence of S,
then S ` X holds).
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 2

More on First-Order Logic

I. Magic Sets

There is another approach to Completeness Theorems, the Skolem–

Löwenheim Theorem and the Regularity Theorem that has an almost
magic-like simplicity. I accordingly dub the central player a magic set.
As in propositional logic, in the context of first-order logic a valuation
is an assignment of truth values t and f to all sentences (with or with-
out parameters). And a valuation v is called a Boolean valuation if for all
sentences (closed formulas) X and Y , the following conditions hold:
B1 : The sentence ∼X receives the value t under v if and only if v(X) = f
[i.e. when X is false under v]; and accordingly v(∼X) = f iff v(X) = t.
B2 : The sentence X ∧ Y receives the value t under v iff v(X) = v(Y ) = t
[i.e. when both X and Y are true under the valuation v].
B3 : The sentence X ∨Y receives the value t under v iff v(X) = t or v(Y ) = t
[i.e. when either X is true under v or Y is true under v].
B4 : The sentence X ⊃ Y receives the value t under v iff v(X) = f or
v (Y ) = t [i.e. when either X is false under v or Y is true under v].

By a first-order valuation (in the denumerable domain of the parame-

ters) is meant a Boolean valuation satisfying the following two additional
conditions:
F1 : The sentence ∀ xϕ(x) is true under v if and only if for every parameter
a, the sentence ϕ(a) is true under v.
F2 : The sentence ∃ xϕ(x) is true under v if and only if the sentence ϕ(a)
is true under v for at least one parameter a.
24 Part I: More on Propositional and First-Order Logic

Let S be a set of sentences of first-order logic. We say that S is truth-

functionally satisfiable, if there is a Boolean valuation in which every
sentence in S is true. Similarly, we say that S is first-order satisfiable
(in first-order contexts, just satisfiable), if there is a first-order valuation
in which every sentence in S is true.
Note: Recall that it was noted in The Beginner’s Guide [Smullyan,
2014] that if we consider the set of all true sentences under a first-order
valuation v, and look at the universe in which the individuals to which
predicates apply as consisting of all the parameters, we can define an inter-
pretation on that universe in which each n-ary predicate is true precisely
on the n-tuples of parameters on which the valuation v says it is true (and
false on the rest of n-tuples of parameters). This is called an interpreta-
tion in the domain of the parameters, and because of this we can say a set
of sentences S is satisfiable in the denumerable domain of the parameters
whenever there is a first-order valuation that satisfies S. Although here we
are not using the concept of an interpretation of first-order logic as defined
in The Beginner’s Guide, it is easy to see that any interpretation of first-
order logic as defined there simultaneously defines a first-order valuation as
defined here.
We leave it to the reader to verify the following:
Lemma. From F1 and F2 it follows that under any first-order valuation,
(1) ∼∃ xϕ(x) is true iff ∼ϕ(a) is true for every parameter a, or,
equivalently, that ∼∃ xϕ(x) and ∀ x∼ϕ(x) are logically equivalent
(i.e. always have the same truth value);
(2) ∼∀ xϕ(x) is true iff ∼ϕ(a) is true for at least one parame-
ter a, or, equivalently, that ∼∀ xϕ(x) and ∃ x∼ϕ(x) are logically
equivalent.

Now, let us again recall the uniform notation of The Beginner’s Guide,
in which we let γ be any formula of the form ∀ xϕ(x) or ∼∃ xϕ(x), and
by γ(a) we meant ϕ(a) or ∼ϕ(a) respectively. Similarly, we let δ be
any formula of the form ∃ xϕ(x) or ∼∀ xϕ(x), and by δ(a) we respec-
tively mean ϕ(a), ∼ϕ(a). Thus, in uniform notation, under any first-order
valuation v:
0
F1 : The sentence γ is true under v if and only if for every parameter a, the
sentence γ(a) is true under v.
0
F2 : The sentence δ is true under v if and only if the sentence δ(a) is true
under v for at least one parameter a.
 Chapter 2. More on First-Order Logic 25

The following proposition will be helpful:

Proposition 1. Suppose that v is a Boolean valuation such that the fol-
lowing two conditions hold (for all sentences γ and δ):
(1) if γ is true under v, then for every parameter a, the sentence γ(a) is
true under v, and
(2) if δ is true under v, then δ(a) is true under v for at least one parameter a.
Then v is a first-order valuation.

Problem 1. Prove Proposition 1.

We say that a set S of sentences of first-order logic is a Boolean truth
set if there is a Boolean valuation v such that S is the set of all sentences
that are true under v. This is equivalent to saying that for all sentences X
and Y , (i) (∼X) ∈ S iff X ∈/ S; (ii) α ∈ S iff α1 ∈ S and α2 ∈ S; (iii) β ∈ S
iff β1 ∈ S or β2 ∈ S.
We call a set of sentences of first-order logic S a first-order truth set if
and only if it is a Boolean truth set and also, for all γ and δ:
(i) γ ∈ S iff γ(a) ∈ S for every parameter a; and
(ii) δ ∈ S iff δ(a) ∈ S for at least one parameter a.
Alternatively, we could say that a set of sentences S is a first-order truth
set if and only if there is a first-order valuation v such that S is the set of
all sentences that are true under v.
Accordingly, Proposition 1 could be restated as follows:

Proposition 10 . Suppose S is a Boolean truth set satisfying the following

two conditions (for all sentences γ and δ):
(1) if γ ∈ S, then for every parameter a, the sentence γ(a) ∈ S, and
(2) if δ ∈ S, then δ(a) ∈ S for at least one parameter a.
Then S is a first-order truth set.
Recall that we say that a (first-order) valuation v satisfies a set of sen-
tences S if all elements of S are true under v. Also, a pure sentence is a
sentence without parameters.
Now for the definition of a magic set. By a magic set M we mean a set
of sentences that satisfies the following two conditions:
M1 : Every Boolean valuation that satisfies M is also a first-order valuation.
M2 : For every finite set S0 of pure sentences, and every finite subset M0 of
M , if S0 is first-order satisfiable, so is S0 ∪ M0 .
26 Part I: More on Propositional and First-Order Logic

We shall soon prove that magic sets exist (which the reader might find
surprising!). But first you will see that one of the important features of
magic sets emerges from the following considerations:
We say that a sentence X is tautologically implied by a set of sentences
S when X is true under all Boolean valuations that satisfy S. We shall call
S a truth-functional basis for first-order logic if for any pure sentence X, the
sentence X is valid if and only if X is tautologically implied by some finite
subset S0 of S (which is equivalent to saying that there are finitely many
elements X1 , . . . , Xn of S such that the sentence (X 1 ∧ . . . ∧ Xn ) ⊃ X is a
tautology).
Note: “Truth-functionally imply” and “tautologically imply” have the
same meaning.

Theorem 1. Every magic set is a truth-functional basis for first-order logic.

Problem 2. Prove Theorem 1. (Hint: Use a fact proved in The Begin-

ner’s Guide [Smullyan, 2014], Chapter 6, Problem 8, namely that if X is
tautologically implied by S, then it is tautologically implied by some finite
subset of S. This fact is an easy corollary of the compactness theorem for
propositional logic, namely that if every finite subset of a denumerable set S
is satisfied by some Boolean valuation, then S is satisfied by some Boolean
valuation.)
Now we turn to the proof that magic sets exist. In Chapter 9 of The
Beginner’s Guide [Smullyan, 2014] we defined a finite set of sentences to
be regular if its elements are each of the form γ ⊃ γ(a) or δ ⊃ δ(a), and
the elements can be placed in a sequence such that for each term of the
sequence, if it is of the form δ ⊃ δ(a), then a does not occur in δ or in
any earlier term of the sequence. We showed that for any pure sentence
X, if X is valid, then there exists a regular set {X1 , . . . , Xn } such that
(X1 ∧ . . . ∧ Xn ) ⊃ X is a tautology. We also showed that for any regular
set R and any set S of pure sentences, if S is (first-order) satisfiable, so is
R ∪ S.
We now define a denumerably infinite set S to be regular if every finite
subset of S is regular. An infinite regular set M will be called complete if
for every γ, the sentence γ ⊃ γ(a) is in M for every parameter a, and if for
every δ, the sentence δ ⊃ δ(a) is in M for at least one parameter a.
 Chapter 2. More on First-Order Logic 27

Theorem 2. Every complete regular set is a magic set.

Problem 3. Prove Theorem 2. [Hint: Proposition 1 will now be helpful.]

Finally, we show that there exists a complete regular set (which com-
pletes the proof that there exists a magic set): We enumerate all δ-sentences
in some denumerable sequence δ1 , δ2 , . . . , δn , . . . . We then consider all our
parameters in some enumeration b1 , b2 , . . . , bn , . . . . We now let a1 be the
first parameter in the enumeration of parameters that does not occur in
δ1 , we let a2 be the first parameter of the enumeration of parameters that
occurs neither in δ1 nor in δ2 and so forth, i.e. for each n we let an be the
first parameter of the enumeration of parameters that occurs in none of the
terms δ1 , δ2 , . . . , δn . We let R1 be the set of all sentences δn ⊃ δn (an ). We
let R2 be the set of all γ ⊃ γ(a), for all γ and every a. Then R1 ∪ R2 is a
complete regular set.

Applications of Magic Sets

The mere existence of magic sets (even those not necessarily regular)
yields particularly simple and elegant proofs of the Skolem–Löwenheim
Theorem and of the first-order compactness theorem as a consequence of
the compactness theorem for propositional logic. We will see these proofs
in a moment.
Note: But first a reminder of something from The Beginner’s Guide
[Smullyan, 2014]. It was noted there that if we consider the set of all true
sentences under a first-order valuation v, and look at the universe in which
the individuals to which predicates apply as consisting of all the parame-
ters, we can define an interpretation in which each n-ary predicate is true
precisely on the n-tuples of parameters on which the valuation v says it
is true (and false on the rest of n-tuples of parameters). This is called a
(first-order) interpretation in the (denumerable) domain of the parameters,
and because of this we can say a set of sentences S is first-order satisfiable
in the domain of the parameters whenever there is a first-order valuation
that satisfies S. Although we are not using the concept of an interpretation
of first-order logic at the moment here, it is easy to see that any interpre-
tation of first-order logic simultaneously defines a first-order valuation over
the denumerable domain of the parameters.
28 Part I: More on Propositional and First-Order Logic

Now suppose S is a denumerable set of pure sentences (no free vari-

ables, no parameters) such that every finite subset of S is first-order sat-
isfiable. Here is an alternative proof that S is first-order satisfiable in the
denumerable domain of the parameters (which simultaneously yields the
compactness theorem for first-order logic, and the Skolem–Löwenheim The-
orem, since if S is first-order satisfiable, then obviously so is every finite
subset of S). The proof we now give uses the compactness theorem for
propositional logic.
Let S be as above and let M be any magic set. Let K be any finite
subset of S ∪ M . We first show that K is first-order satisfiable. Well, K
is the union S0 ∪ M0 of some finite subset S0 of S with some finite subset
M0 of M . Since S0 is first-order satisfiable, so is S0 ∪ M0 [by property M2
of a magic set]. Since S0 ∪ M0 is first-order satisfiable, by definition there
is a first-order valuation v that satisfies S0 ∪ M0 , and v is obviously also
a Boolean valuation [since every first-order valuation is]. This proves that
every finite subset K of S∪M is truth-functionally satisfiable. Hence, by the
compactness theorem for propositional logic, the entire set S ∪ M is truth-
functionally satisfiable. Thus there is a Boolean valuation v that satisfies
S ∪ M , and since v satisfies M , it must be a first-order valuation [by prop-
erty M1 of a magic set]. Thus, by what we said above about the relationship
between the first-order interpretations of The Beginner’s Guide [Smullyan,
2014] and first-order valuations, all sentences of S are true (simultane-
ously satisfiable) in an interpretation over the denumerable domain of the
parameters.
We recall the Regularity Theorem, which is that for every valid pure sen-
tence X, there is a (finite) regular set R which truth-functionally implies X.
(This is a somewhat simplified version of what we proved in The Beginner’s
Guide. We will return to the full version of the Regularity Theorem later in
this chapter.) We proved this using the completeness theorem for tableaux,
namely we showed how, from a closed tableau for ∼X, such a regular set
could be found (we took R to be the set of all γ ⊃ γ(a) such that γ(a) was
inferred from γ by Rule C, together with all sentences δ ⊃ δ(a) such that
δ(a) was inferred from δ by Rule D). A much simpler proof of the Regular-
ity Theorem is easily obtained from the existence of a complete regular set
(which is also a magic set).

Problem 4. How?
 Chapter 2. More on First-Order Logic 29

II. Gentzen Sequents and Some Variants

In Part III of this chapter, we will state and prove an important result
known as Craig’s Interpolation Lemma, which has significant applications
that we will also consider. Craig’s original proof of this result was quite
complicated, and many simpler proofs of this have subsequently been given.
One of these which I will give uses a variant of some axiom systems due to
Gerhard Gentzen [1934, 1935], to which we now turn.

Background

We first need to speak of subformulas. In propositional logic, the notion of

immediate subformula is given explicitly by the following conditions:
I0 : Propositional variables have no immediate subformulas.
I1 : ∼X has X as its only immediate subformula.
I2 : X ∧ Y, X ∨ Y, X ⊃ Y have X and Y as immediate subformulas, and no
other immediate subformulas.
For first-order logic, the notion of immediate subformula is given
explicitly by the following conditions (we recall that an atomic formula
of first-order logic is a formula which contains no logical connectives or
quantifiers):
I00 : Atomic formulas have no immediate subformulas.
I10 : Same as I1 .
I20 : Same as I2 .
I30 : For any parameter a and variable x, ϕ(a) is an immediate subformula
of ∀ xϕ(x) and of ∃ xϕ(x).
Both in propositional logic and first-order logic, the notion of subformula
is implicitly defined by the following conditions:
S1 : If Y is an immediate subformula of X, or is identical to X, then Y is
a subformula of X.
S2 : If Y is an immediate subformula of X, and Z is a subformula of Y, then
Z is a subformula of X.
The above implicit definition can be made explicit as follows: Y is a
subformula of X if and only if either Y = X or there is a finite sequence of
formulas whose first term is X and whose last term is Y and is such that
each term of the sequence other than the first is an immediate subformula
of the preceding term.
30 Part I: More on Propositional and First-Order Logic

Now, G. Gentzen was out to find finitary proofs of the consistency of

various systems (finitary in the sense that the proofs did not involve the use
of infinite sets). For this purpose, he needed a proof procedure that required
that any proof of a formula X used only subformulas of X. Such a proof
procedure is said to obey the subformula principle. Of course, tableaux using
signed formulas obey the subformula principle (and these are closely related
to the Gentzen systems, as we will see). Tableaux using unsigned formulas,
almost, but not quite, obey this principle — they use either subformulas
or negations of them. [For example ∼X is not necessarily a subformula of
∼(X ∨ Y ), but is the negation of one.]
Gentzen took a new symbol “→” and defined a sequent to be an expres-
sion of the form θ → Γ, where each of θ, Γ is a finite, possibly empty,
sequence of formulas. Such an expression is interpreted to mean that if
all terms of θ are true, then at least one term of Γ is true. That is,
under any valuation v, the sequent θ → Γ is true under v iff it is the
case that if all terms of θ are true, then at least one term of Γ is true.
A sequent is called a tautology if it is true under all Boolean valuations,
and valid if it is true under all first-order valuations. If θ is a non-empty
sequence X1 , . . . , Xn and Γ is a non-empty sequence Y1 , . . . , Yk , then
the sequent X1 , . . . , Xn → Y1 , . . . , Yk is logically equivalent to the for-
mula (X1 ∧ . . . ∧ Xn ) ⊃ (Y1 ∨ . . . ∨ Yk ). If θ is the empty sequence,
then θ → Γ is written as ∅ → Γ, where ∅ is the empty set, or more
simply → Γ.
Now, if Γ is a non-empty sequence, then the sequent → Γ (which is
∅ → Γ) is read “if all elements of the empty set are true, then at least one
element of Γ is true.” Well, all elements of the empty set are true (since
there are no elements of the empty set), and so the sequent → Γ is logically
equivalent to the proposition that at least one element of Γ is true. Thus a
sequent → Y1 , . . . , Yk is logically equivalent to the formula Y1 ∨ . . . ∨ Yk .
Now, suppose that the right side Γ of the sequent θ → Γ is empty. The
sequent is then more simply written as θ →. It is interpreted to mean that
if all the terms of the sequence θ are true, then at least one element of the
empty set ∅ is true. Well, it is obviously not the case that some element of
the empty set is true, so that θ → expresses the proposition that it is not the
case that all terms of the sequence θ are true. If θ is a non-empty sequence
X1 , . . . , Xn , then the sequent X1 , . . . , Xn → is logically equivalent to the
formula ∼(X1 ∧ . . . ∧ Xn ).
 Chapter 2. More on First-Order Logic 31

Finally, the → written alone is a sequent, and abbreviates ∅ → ∅, where

∅ is the empty set. What does → by itself mean? It means that if all elements
of the empty set are true, then at least one element of the empty set is true.
Well, all elements of the empty set are vacuously true, but is it obvious that
it is not the case that at least one element set if true. Thus the sequent →
is simply false (under all valuations).
At this point, I must tell you an amusing incident: I was once giv-
ing a lecture on Gentzen sequents to a mathematics group. I started with
Gentzen sequents in which both sides of the arrow were non-empty, and
after having erased, first the left side, and then the right side, I was left
with just the arrow alone on the board. After explaining that it meant
falsehood, the logician William Craig (the inventor of Craig’s Interpolation
Lemma) was in the audience, and with his typical sense of humor, raised
his hand and asked, “And what does it mean if you write nothing on the
board?”
In more modern Gentzen type systems, the sequents are of the form
U → V , where U and V are finite sets of formulas, instead of sequences of
formulas, and this is the course we will take.

Gentzen Type Axiom Systems

We shall first consider a Gentzen type axiom system G0 for propositional

logic. For any finite set S of formulas and any formulas X1 , . . . , Xn , we
abbreviate S ∪ {X1 , . . . , Xn } by S, X1 , . . . , Xn . Thus,
U, X1 , . . . , Xn → V, Y1 , . . . , Yk
abbreviates
U ∪ {X1 , . . . , Xn } → V ∪ {Y1 , . . . , Yk }.

The system G0 for propositional logic has only one axiom scheme and
the eight inference rules below. Here X and Y are formulas of propositional
logic, and U and V are sets of formulas of propositional logic.

Axiom Scheme of the system G0

U, X → V, X
Rules of the system G0
32 Part I: More on Propositional and First-Order Logic

Conjunction
U, X, Y → V
C1 :
U, X∧Y → V

U → V, X U → V, Y
C2 :
U → V, X∧Y

Disjunction
U → V, X, Y
D1 :
U → V, X∨Y
U, X → V U, Y → V
D2 :
U, X∨Y → V

Implication
U, X → V, Y
I1 :
U → V, X⊃Y

U → V, X U, Y → V
I2 :
U, X⊃Y →V

Negation
U, X → V
N1 :
U → V, ∼X

U → V, X
N2 :
U, ∼X → V

The inference rules introduce the logical connectives. There are two rules
for each connective — one to introduce the connective on the left side of the
arrow, the other to introduce the connective on the right side of the arrow.
As an example of how the rules are applied, the conjunction rule C1 is that
from the sequent U, X, Y → V one can infer the sequent U, X ∧ Y → V . As
for Rule C2 , it says that from the two sequents U → V, X and U → V, Y
one can infer the sequent U → V, X ∧ Y.
It should be obvious that the axioms of G0 are tautologies, and we
leave it to the reader to verify that the inference rules preserve tautologies,
in other words, that in any application of an inference rule, the conclu-
sion is truth-functionally implied by the premises. It therefore follows (by
 Chapter 2. More on First-Order Logic 33

mathematical induction) that we have: All provable sequents of G0 are

tautologies; thus the system G0 is correct.
We will soon prove that the system G0 is complete, i.e. that all tautolo-
gies are provable in G0 .

Problem 5. Show that the sequent U, X → V is logically equivalent to

U → V, ∼X and that the sequent U → V, X is logically equivalent to
U, ∼X → V .

The System G0 in Uniform Notation

Let S be a non-empty set {T X1 , . . . , T Xn , F Y1 , . . . , F Yk } of signed for-

mulas. By |S| we shall mean the sequent X 1 , . . . , Xn → Y1 , . . . , Yk . We
call |S| the corresponding sequent of S. We leave it to the reader to verify
the important fact that S is unsatisfiable if and only if its corresponding
sequent |S| is a tautology (and, for later use when we get to Gentzen sys-
tems for first-order logic), that if the elements of S are signed formulas of
first-order logic, then S is unsatisfiable iff |S| is valid.
Of course, if S is a set {F Y1 , . . . , F Yk } in which there are no formulas
signed T , we take |S| to be the sequent → Y1 , . . . , Yk , and if S is a set
{T X1 , . . . , T Xn } in which there are no formulas signed F , we take |S| to
be the sequent X1 , . . . , Xn →.
More generally, for any set of signed formulas, if U is the set of formulas
X such that T X ∈ S, and V is the set of formulas Y such that F Y ∈ S,
then |S| is the sequent U → V .
The system G0 in uniform notation is the following:
Axioms
(where S is a set of signed formulas)

|S, T X, F X|

Inference Rules
|S, α1 , α2 |
A: (α, α1 , and α2 are signed formulas)
|S, α|

|S, β1 | |S, β2 |
B: (β, β1 , and β2 are signed formulas)
|S, β|
34 Part I: More on Propositional and First-Order Logic

Problem 6. Verify that the above uniform system really is the

system G0 .
Proofs in Gentzen type systems are usually in tree form, unlike the linear
form of proofs in the axiom systems previously considered. Only now, unlike
the trees of tableaux, the tree must grow upwards if the user wishes to prove
a statement by analyzing it from the outside in (as must take place by the
tableau rules): the origin is at the bottom and the end points are at the top.
The origin is the sequent to be proved, and the end points are the axioms
used in the proof. As an example, here is a proof in tree form of the sequent
p ⊃ q → ∼q ⊃ ∼p.
(1) p → q, p (2) p, q → q
(3) p, p ⊃ q → q [from (1) and (2), by Rule I2 ]
(4) p ⊃ q → q, ∼p [from (3), by Rule N1 ]
(5) p ⊃ q, ∼q → ∼p [from (4), by Rule N2 ]
(6) p ⊃ q → ∼q ⊃ ∼p [from (5), by Rule I1 ].

Completeness of G0

We now wish to show that all tautologies are provable in the Gentzen system
G0 for propositional logic.
By a tableau proof of a sequent X1 , . . . , Xn → Y1 , . . . , Yk is meant a
closed tableau proof showing that the set {T X1 , . . . , T Xn , F Y1 , . . . , F Yk }
is unsatisfiable (if the tableau is started with all these formulas, the tableau
can be extended by the tableau rules for propositional logic in such a way that
every branch closes). We will now show how, from a tableau proof of a sequent,
we can obtain a proof of the sequent in the Gentzen system G0 . To this end, it
will be useful to first introduce, as an intermediary, another type of tableau
that I will now call a block tableau (this was called a modified block tableau
in my book First-Order Logic [Smullyan, 1968, 1995]). The analytic tableaux,
which we have studied so far, are variants of the tableaux of Evert Beth [1959],
whereas the block tableaux, to which we now turn, are variants of the tableaux
of J. Hintikka [1955]. Like the tableaux of Hintikka, the points of the tree in a
block tableau proof (this time again with the origin at the top, as in analytic
tableaux) are not single formulas, but are finite sets of formulas, which we call
the blocks of the tableaux. And what we do at any stage of the construction
depends solely on the end points of the tree (rather than on the branch as a
whole, as in analytic tableaux).
 Chapter 2. More on First-Order Logic 35

We will first consider block tableaux for propositional logic. By a block

tableau for a finite set K of signed formulas of propositional logic, we mean
a tree constructed by placing the set K at the origin, and then continuing
according to the following rules (here, for a set S and formulas X, Y , we
abbreviate the sets S ∪ {X} and S ∪ {X} ∪ {Y } by {S, X} or {S, X, Y }
respectively):

Block Tableau Inference Rules for Propositional Logic

A0 :
{S, α}

{S, α1 , α2 }
0
B:
{ S, β }

{S, β1 } {S, β2 }
In words, these rules are:
0
A : To any end point {S, α} on the tree, we subjoin {S, α1 , α2 } as sole
successor.
B 0 : To any end point {S, β} on the tree, we simultaneously subjoin {S, β1 }
as left successor and {S, β2 } as right successor.

We call a block tableau closed if each end point contains some element
and its conjugate.
Changing from an analytic tableau to a block tableau is a relatively
simple matter. To begin with, in constructing an analytic tableau, when
using Rule A, and inferring α1 from α, subjoin α2 immediately after it.
That is, use Rule A in the form:
α

α1
α2
Now, to start a block tableau for a sequent X1 , . . . , Xn → Y1 , . . . , Yk ,
put the set {T X1 , . . . , T Xn , F Y1 , . . . , F Yk } at the origin, for we wish to
36 Part I: More on Propositional and First-Order Logic

prove that set of formulas is unsatisfiable. This concludes stage 0 of the

construction of the block tableau corresponding to the analytic tableau
that begins as follows:
T X1
·
·
·
T Xn
F Y1
·
·
·
F Yk
Then, for each n, after completion of stage n for the two tableaux, when
we use Rule A (Rule B) in the analytic tableau, we use Rule A0 (Rule B 0 ,
respectively) in the block tableau, and that concludes stage n + 1 of the
constructions. We continue this until both tableaux close (which must hap-
pen at some stage, if the set of formulas {T X1 , . . . , T Xn , F Y1 , . . . , F Yk }
is unsatisfiable, i.e. the sequent is a tautology). Since we know that the
tableau method for propositional logic is complete, this construction shows
that the block tableau method for propositional logic is also complete.
As an example, here is a closed analytic tableau for the sequent
p ⊃ q, q ⊃ r → p ⊃ r:
 Chapter 2. More on First-Order Logic 37

Here is the corresponding block tableau:

To convert a block tableau to a proof in the Gentzen system G0 , just replace

each block
{T X1 , . . . , T Xn , F Y1 , . . . , F Yk }

by the sequent X 1 , . . . , Xn → Y1 , . . . , Yk , and the resulting tree, when

turned upside down, is a proof in the Gentzen system.
For example, from the above closed block tableau, we get the following
proof in the Gentzen system:

Remarks. It is not surprising that this conversion of a block tableau to

a proof in the Gentzen system works, because the block tableau rules are
essentially the rules of the Gentzen system G0 turned upside down. That
38 Part I: More on Propositional and First-Order Logic

is, in Rule A0 of block tableaux

{ S, α }

{ S, α1 , α2 }
if we replace the premise and conclusion by the corresponding sequents, we
get the rule
| S, α |

| S, α1 , α2 |
which is Rule A of the system G0 turned upside down. Likewise, if in Rule
B 0 of block tableaux
{ S, β }

{S, β1 } {S, β2 }
we replace the premise and conclusions by the corresponding sequents, we
get the rule
| S, β |

|S, β1 | |S, β2 |
which is Rule B of the system G0 turned upside down.

A Gentzen Type System G1 for First-Order Logic

For the Gentzen system G1 , we take the axioms and inference rules of
the system G0 (only now applied to formulas of first-order logic instead of
formulas of propositional logic), and add the following inference rules:

∀1 : U, ϕ(a) → V ∀2 : U → V, ϕ(a)
U, ∀ xϕ(X) → V U → V, ∀ xϕ(X)∗
∗
Providing a does not occur
in the conclusion sequent.
 Chapter 2. More on First-Order Logic 39

∃1 : U → V, ϕ(a) ∃2 : U, ϕ(a) → V
U → V, ∃ xϕ(x) U, ∃ xϕ(x) → V ∗
∗
Providing a does not occur
in the conclusion sequent.

One proves the completeness of G1 from the completeness of first-order

block tableau in essentially the same manner as in propositional logic. One
first makes a block tableau, whose rules are those of G0 (with sequents
replaced by sets of signed formulas and turned upside down) together with:
C: {S, γ}

{S, γ, γ(a)}

D: {S, δ}∗

{S, δ, δ(a)}
∗
Providing a does not occur
in the antecedent sequent.

One then converts a closed block tableau into a proof in the Gentzen
system G1 as before, i.e. one replaces each block {S} by the sequent |S|
and turns the resulting tree upside down.

The System GG

In some of the inference rules of the systems G0 and G1 (specifically the

negation and implication rules), one transfers a formula from one side of
the arrow to the other (or at least incorporates it into some formula on the
other side). We need a modification of the system in which this does not
happen. This modification, which we will call GG, has a certain symmetry
with respect to the two sides of the arrow, and was called a symmetric
Gentzen system in Smullyan [1968, 1995].
Here is the system GG, which is admirably suited to uniform notation
(since we are dealing with sequents here, all formulas refer to unsigned
sentences):
40 Part I: More on Propositional and First-Order Logic

Axiom Schemes of GG
U, X → V, X
U, X, ∼X → V
U → V, X, ∼X
Rules of GG
U, α1 , α2 → V U → V, β1 , β2
(A)
U, α → V U → V, β
U, β 1 → V U, β 2 → V U → V, α1 U → V, α2
(B)
U, β → V U → V, α
U, γ(a) → V U → V, δ(a)
(C)
U, γ → V U → V, δ
U, δ(a) → V U → V, γ(a)
(D)
U, δ → V ∗ U → V, γ ∗
∗
Providing a does not occur
in the conclusion sequent.

The Completeness of GG

We need another variant of analytic tableaux to prove the completeness of

GG. In some of the rules of analytic tableaux, a conclusion of a rule has a
different sign than that of the premise (which reflects the fact that in the
corresponding Gentzen system, a sentence might change from one side of
the arrow to the other). We now need an analytic tableau system in which
this does not happen.
By an altered tableau we mean one constructed according to the following
rules:
T ∼∼X F ∼∼X

TX FX
TX∧Y FX∧Y

TX
TY
FX FY
 Chapter 2. More on First-Order Logic 41

TX∨Y FX∨Y

FX
FY
TX TY
TX ⊃Y FX ⊃Y

F ∼X
FY
T ∼X TY

T ∼ (X ∧ Y ) F ∼ (X ∧ Y )

F ∼X
F ∼Y
T ∼X T ∼Y

T ∼ (X ∨ Y ) F ∼ (X ∨ Y )

T ∼X
T ∼Y
F ∼X F ∼Y

T ∼ (X ⊃ Y ) F ∼ (X ⊃ Y )

TX
T ∼Y
FX F ∼Y

T ∀ xϕ(x) F ∃ xϕ(x)

T ϕ(a) F ϕ(a)

T ∃ xϕ(x)∗ F ∀ xϕ(x)∗

T ϕ(a) F ϕ(a)
T ∼ ∃ xϕ(x) F ∼ ∀ xϕ(x)

T ∼ ϕ(a) F ∼ ϕ(a)
42 Part I: More on Propositional and First-Order Logic

T ∼ ∃ xϕ(x) F ∼ ∀ xϕ(x)

T ∼ ϕ(a) F ∼ ϕ(a)

T ∼ ∀ xϕ(x)∗ F ∼ ∃ xϕ(x)∗

T ∼ ϕ(a) F ∼ ϕ(a)

In the quantification rules (above and below), the asterisk “∗ ” means

that the parameter in the conclusion must not appear in the premise.
Here are the rules of the system altered tableaux in uniform notation
(where the α, β, γ and δ are of course unsigned formulas):

For altered tableaux, a set of signed formulas is called closed if it contains

either some T X and F X or some T X and T ∼X or some F X and F ∼X.
A branch is called closed if the set of sentences on the branch is a closed
set, and an altered tableau is called closed if every branch is closed.
 Chapter 2. More on First-Order Logic 43

We will need to know that the altered tableau method is both correct
and complete, i.e. that there is a closed altered tableau for a set S if and only
if S is unsatisfiable. To this end, we can save half the labor by the following
notation, which might aptly be dubbed a super-unifying notation.
We let A be any signed sentence of the form T α or F β. If it is the
former, we let A1 = T α1 and let A2 = T α2 ; if the latter, we let A1 = F β1
and let A2 = F β2 . We let B be any sentence of the form T β or F α. If
it is the former, we let B1 , B2 be T β1 and T β2 respectively; if the latter,
we let B1 , B2 be F α1 and F α2 , respectively. We let C be any sentence
of the form T γ or F δ, and we take C(a) to be T γ(a), F δ(a), respec-
tively. We let D be any sentence of the form T δ or F γ, and we take
D(a) to be T δ(a), F γ(a), respectively. With this notation, the altered
tableau rules for first-order logic are now most succinctly expressed as
follows:
A B C D∗

A1 B1 B2 C(a) D(a)
A2
Problem 7. Prove that the altered tableau system is correct and complete,
and then finish the proof of the completeness of the Gentzen system GG.

III. Craig’s Lemma and an Application

Craig’s Interpolation Lemma

A sentence Z of first-order logic is called an interpolant for a sentence X ⊃ Y

if X ⊃ Z and Z ⊃ Y are both valid and every predicate and parameter of
Z occurs in both X and Y . Craig’s celebrated Interpolation Lemma (often
simply called Craig’s Lemma) is that if X ⊃ Y is valid, then there is an
interpolant for it, provided that Y alone is not valid and X alone is not
unsatisfiable [Craig, 1957].
If we allow t’s and f ’s to be part of our formal language, and define
“formula” accordingly, then the proviso above is unnecessary, because if
X ⊃ Y is valid and Y alone is valid, then t is an interpolant for X ⊃ Y ,
44 Part I: More on Propositional and First-Order Logic

or if X alone is unsatisfiable, then f is an interpolant for X ⊃ Y (as the

reader can verify). This is the course we shall take.
Formulas that involve t or f will temporarily be called non-standard
formulas. As shown in The Beginner’s Guide [Smullyan, 2014], any
non-standard formula is logically equivalent to either a standard formula
or to t or to f .
There is a corresponding interpolation lemma for propositional logic.
For propositional implication X ⊃ Y , a propositional formula Z is called
an interpolant for X ⊃ Y if X ⊃ Z and Z ⊃ Y , and if every propositional
variable in Z occurs in both X and Y . Craig’s Lemma for proposition logic
is that if X ⊃ Y is a tautology, it has an interpolant.
Returning to first-order logic, a formula Z is called an interpolant for a
sequent U → V , if U → Z and Z → V are both valid, and if every predicate
and parameter of Z occurs in at least one element of U and at least one
element of V . Obviously, Z is an interpolant for a sequent

X1 , . . . , Xn → Y1 , . . . , Yk

if and only if Z is an interpolant for the formula

(X1 ∧ . . . ∧ Xn ) ⊃ (Y1 ∨ . . . ∨ Yk ),

so that the existence of interpolants for all valid sequents is equivalent to

the existence of interpolants for all valid sentences X ⊃ Y . Thus we will
consider Craig’s Lemma for First-Order Logic in the equivalent form that
there exist interpolants for all valid sequents.
We will prove Craig’s Lemma for the Gentzen system GG.

Proof of Craig’s Interpolation Lemma

for the Gentzen System GG

It suffices to show that there is an interpolant for each of the axioms of

GG, and that for each of the inference rules of GG, if there is an inter-
polant for the premise, or interpolants for the two premises, then there in
an interpolant for the conclusion.

Exercise. Prove this for the axioms and for Rules A and B (which don’t
involve the quantifiers).
 Chapter 2. More on First-Order Logic 45

Proving the claim of the exercise really proves Craig’s Interpolation

Lemma for propositional logic. Now let’s turn to Rules C and D, which do
involve the quantifiers.
The proof for Rule D is relatively simple. Suppose that X is an inter-
polant for U, δ(a) → V, and that a does not occur in U, δ or V. Then X
must also be an interpolant for U, δ → V for the following reasons:
Since X is an interpolant of U, δ(a) → V , then every parameter of
X occurs in V (as well as in U, δ(a)). But since a does not occur in V
(by hypothesis), then a cannot occur in X. From this, it follows that all
parameters of X occur in U , δ (as well as in V ).
To see this, let b be any parameter that occurs in X. We know that
b 6= a. We also know that b occurs in V (all parameters of X do). It
remains to show that b occurs in U , δ. If b occurs in U , we are done. If not,
then b must occur in δ(a) (since b occurs in U, δ(a)). But the parameters
of δ are those of δ(a) other than a. Since b occurs in δ(a) and b 6= a, then
b must occur in δ. This proves that all parameters of X occur in U, δ (as
well as in V ). Also all predicates of X occur in U, δ and in V since they
all occur in U, δ(a) and in V , and the predicates of δ are the same as those
of δ(a).
Finally, since U, δ(a) → X is valid (as well as X → V ), and a
does not occur in U, δ or in X, then U, δ → X is valid (in fact it fol-
lows from U, δ(a) → X by Rule D). Thus X is an interpolant for the
sequent U, δ → V .
The proof that if X is an interpolant for U → V, γ(a), and a does not
occur in U or in V, γ(a), then X is an interpolant for U → V, γ is similar
to the preceding proof, and is left to the reader. This takes care of Rule D.
The proof for Rule C is much trickier! Suppose that X is an interpolant
for U, γ(a) → V . Then U, γ(a) → X and X → V are of course valid, but X
may fail to be an interpolant for U, γ → V because the parameter a might
occur in X, but not in U, γ. If a does not occur in X, or if a does occur in
U, γ, there is no problem. The critical case is one in which a does occur in
X, but not in U, γ. For this case, we find another interpolant for U, γ → V
as follows:
We take some variable x that does not occur in X and we let ϕ(x) be
the result of replacing every occurrence of a in X by x. We note that ϕ(a)
is X itself. We will show that ∀ xϕ(x) is an interpolant for U, γ → V .
46 Part I: More on Propositional and First-Order Logic

Let γ 0 = ∀ xϕ(x). Then γ 0 (a) = ϕ(a), which is X. We already know that

U, γ → γ 0 (a) is valid (because it is U, γ → X), and since a does not occur
in U, γ, and obviously not in γ 0 (which is ∀ xϕ(x)), then U, γ → γ 0 is valid
(it is in fact derivable from U, γ → γ 0 (a) by Rule D of the system GG).
Also γ 0 (a) → V (which is X → V ) is valid by hypothesis. Hence γ 0 → V
is valid (by Rule C, taking U to be the empty set). Thus U, γ → γ 0 and
γ 0 → V are both valid, and all predicates and parameters of γ 0 are both
in U, γ and in V (since a does not occur in γ 0 ). Thus ∀ xϕ(x), which is γ 0 ,
is an interpolant for U, γ → V . This completes the proof for the γ rule of
Rule C.
As to the δ rule of Rule C, a similar proof, which we leave to the reader,
reveals the following: Suppose X is an interpolant for U → V, δ(a). Then if
a does not occur in X, or a does occur in V, δ, then X is an interpolant for
U → V, δ. But if a does occur in X but not in V, δ, then an interpolant for
U → V, δ is ∃ xϕ(x), where ϕ(x) is the result of replacing every occurrence
of a in X by a variable x that does not occur in X.
This concludes the proof of Craig’s Interpolation Lemma.

Beth’s Definability Theorem

One important application of Craig’s Interpolation Lemma is that it pro-

vides a neat proof of an important result of the Dutch logician E. Beth
[1959]. While discussing Beth’s Definability Theorem, I will switch to
the language of first-order interpretations that we used frequently in
The Beginner’s Guide [Smullyan, 2014] rather than that of first-order val-
uations, with which we have spent more time here.
We consider a finite set S of sentences without parameters and involv-
ing only predicates P, P1 , . . . Pn of degree 1. For any sentence X whose
predicates are among P, P1 , . . . , Pn , as usual we write S ` X to mean that
X is a logical consequence of S (i.e. is true in all interpretations that sat-
isfy S). We say that P is explicitly definable from P1 , . . . , Pn with respect
to S iff there is a formula ϕ(x) whose predicates are among P1 , . . . , Pn
(and hence do not involve P ) such that S ` ∀ x(P x ≡ ϕ(X)). Such a for-
mula ϕ(x) is called an explicit definition of P from P1 , . . . , Pn with respect
to S.
There is another type of definability called implicit definability, which
consists of the following: We take a new predicate P 0 , distinct from any of
 Chapter 2. More on First-Order Logic 47

P, P1 , . . . , Pn and which does not occur in any element of S, and we let S 0

be the result of replacing P by P 0 in every sentence in S. Then P is said
to be implicitly definable from P1 , . . . , Pn with respect to S iff
S ∪ S 0 ` ∀ x(P x ≡ P 0 x).
When this is true, we also say that the sentences of S implicitly define the
predicate P .
We remark that this condition is equivalent to the condition that for
any two interpretations that satisfy S, if they agree on each of P1 , . . . , Pn ,
then they agree on P . (Two interpretations I1 and I2 are said to agree on
a predicate Q iff they assign the same value to Q.) This equivalence is not
necessary for the proof of Beth’s theorem, but is of independent interest.
A proof of this can be found in my book Logical Labyrinths [Smullyan, 2008,
2009, pp. 294--295].
It is relatively easy to show that if P is explicitly definable from
P1 , . . . , Pn with respect to S, then it is implicitly so definable.
Problem 8. Show that if P is explicitly definable from P1 , . . . , Pn with
respect to S, then it is implicitly so definable.

Theorem B. [Beth’s Definability Theorem] If P is implicitly definable from

P1 , . . . , Pn with respect to S, then it is explicitly so definable.

This theorem is far from obvious, but Craig’s Lemma provides a neat
and elegant proof of it.
Suppose that P is implicitly definable from P1 , . . . , Pn with respect to
S. Let P 0 and S 0 be as previously defined, and assume we have
S ∪ S 0 ` ∀ x(P x ≡ P 0 x).
Let X be the conjunction of the elements of S (the order does not matter),
and let X 0 be the conjunction of the elements of S 0 . We thus have
(X ∧ X 0 ) ⊃ ∀ x(P x ≡ P 0 x).
Since the above sentence is valid, then for any parameter a the sentence
(X ∧ X 0 ) ⊃ (P a ≡ P 0 a) is valid. [We recall that S is a set of sentences
without parameters because that is one of the conditions on both explicit
and implicit definitions. Thus the sentence X ∧ X 0 has no parameters.]
Hence by propositional logic, the sentence (X ∧ X 0 ) ⊃ (P a ⊃ P 0 a), is valid.
Thus, again by propositional logic, so is the following sentence:
(X ∧ P a) ⊃ (X 0 ⊃ P 0 a). (Verify this!)
48 Part I: More on Propositional and First-Order Logic

In this sentence, P does not occur in X 0 ⊃ P 0 a and P 0 does not occur

in X ∧ P a. By Craig’s Lemma, there is an interpolant Z for the sentence
(X ∧ P a) ⊃ (X 0 ⊃ P 0 a). All predicates and parameters of Z occur in both
X ∧ P a and in X 0 ⊃ P 0 a. Hence neither P nor P 0 can occur in Z; in fact all
the predicates of Z are among P1 , . . . , Pn . Also, Z contains no parameters
other than a, since X and X 0 contain no parameters, as noted above. Let x
be a variable that does not occur in Z, and let ϕ(x) be the result of substi-
tuting x for all occurrences of a in Z. But ϕ(a) = Z, so ϕ(a) is an interpolant
for (X ∧ P a) ⊃ (X 0 ⊃ P 0 a). Hence the following two sentences are valid:
(1) (X ∧ P a) ⊃ ϕ(a).
(2) ϕ(a) ⊃ (X 0 ⊃ P 0 a).

From (1), we get the valid sentence:

(10 ) X ⊃ (P a ⊃ ϕ(a)).

From (2), we get the valid sentence:

(20 ) X 0 ⊃ (ϕ(a) ⊃ P 0 (a)).

Since (20 ) is valid, so is its notational variant:

(200 ) X ⊃ (ϕ(a) ⊃ P (a)).

[If a sentence is valid, it remains valid if we replace any predicate by a new

predicate.]
By (10 ) and (200 ) we get the valid sentence:
X ⊃ (P a ≡ ϕ(a)).
Since a does not occur in X, the sentence X ⊃ ∀ x(P x ≡ ϕ(x)) is valid.
Since X is the conjunction of the elements of S, it then follows that
S ` ∀ x(P x ≡ ϕ(x)), and so the formula ϕ(x) explicitly defines P from
P1 , . . . , Pn with respect to S.
This concludes the proof of Beth’s Definability Theorem.
Note: The fact that S was assumed finite was not really necessary for
the proof. A compactness argument can be used to modify the proof for an
infinite set.

IV. A Unification

Let us review some of the main results that we have proved about first-order
logic in The Beginner’s Guide [Smullyan, 2014] and in this chapter.
 Chapter 2. More on First-Order Logic 49

T1 : The Completeness Theorem for Analytic Tableaux. Every valid formula

is provable by the tableau method.
T2 : The Skolem–Löwenheim Theorem. For any set S of formulas, if S is
satisfiable, then it is satisfiable in a denumerable domain.
T3 : The Compactness Theorem. For any infinite set S of formulas, if every
finite subset of S is satisfiable, so is S.
T4 : The Completeness of a First-Order Axiom System.
T5 : The Regularity Theorem. For every valid pure sentence X, there is a
(finite) regular set R which truth-functionally implies X.
T6 : Craig’s Interpolation Lemma. If the sentence X ⊃ Y (of propositional
or first-order logic) is valid, there is a sentence Z such that X ⊃ Z
and Z ⊃ Y are both valid and, for propositional logic, every propo-
sitional variable occurring in Z occurs in both X and Y , while, for
first-order logic, every predicate and parameter of Z occurs in both
X and Y .

In my paper “A Unifying Principle in Quantification Theory” [Smullyan,

1963], I stated and proved a result that yields all of T1 − T6 as special cases.
To this we now turn. What follows is slightly different from the original
formulation of 1963.
Consider any property Φ of sentences. A set of sentences that has
property Φ will be called Φ-consistent (i.e. “consistent” with having the
property Φ).
We are now going to be especially interested in certain sets of sen-
tences that have what we will call an analytic consistency property, and
we will use the Greek symbol Γ to signal that a property of sets is
an analytic consistency property. Thus we define a set S of sentences
to be Γ-consistent — which means the same as saying that the set S
has the analytic consistency property Γ — if the following conditions
hold:
Γ0 : S contains no element and its negation (or no element and its conjugate,
if we are considering signed formulas).
Γ1 : For any α in S, the sets S ∪ {α1 } and S ∪ {α2 } are both Γ-consistent
(and hence so is S ∪ {α1 , α2 }).
Γ2 : For any β in S, at least one of the sets S ∪ {β1 } and S ∪ {β2 } is
Γ-consistent.
Γ3 : For any γ in S and any parameter a, the set S ∪ {γ(a)} is Γ-consistent.
Γ4 : For any δ in S, the set S ∪ {δ(a)} is Γ-consistent for any parameter a
that does not occur in any element of S.
50 Part I: More on Propositional and First-Order Logic

Theorem U. [The Unification Theorem] For any analytic consistency prop-

erty Γ, if S is Γ-consistent, then S is satisfiable in the domain of the
parameters.
One proof of Theorem U is quite similar to the completeness proof for
tableaux: Given a Γ-consistent set S, it easily follows that no tableau for
S can close (verify this!). Hence a systematic tableau for S will contain an
open branch that includes all elements of S, and is a Hintikka set, and thus
satisfiable in the domain of the parameters.

Discussion. We proved Theorem U using analytic tableaux, although

the theorem makes no reference to tableaux. My 1963 proof used neither
tableaux nor König’s lemma, and is as follows: We shall consider now only
the case that S is denumerable (the construction for a finite set S is even
simpler, and the modification is left to the reader). We shall also assume
that only finitely many parameters are found in the formulas of S (if not,
we can always introduce a new denumerable group of parameters, and use
them along with the parameters in the formulas of S, in the construction
ahead).
In what follows, consistent will mean Γ-consistent. Given a consis-
tent denumerable set of sentences S, the idea is to define an infinite
sequence of formulas whose set of terms is a Hintikka set that contains
all the elements of S. The sequence is defined in stages: at each stage
we have a finite segment (X1 , . . . , Xk ) of our desired infinite sequence,
and at the next stage, we extend that sequence to a larger finite sequence
(X1 , . . . , Xk , . . . , Xk+m ).
First some definitions and notations: Let θ be a finite sequence
(X1 , . . . , Xk ). We shall say that θ is consistent with S when the set
S ∪{X1 , . . . , Xk } is consistent. For any formulas Y1 , . . . , Ym , by the expres-
sion θ, Y1 , . . . , Ym is meant the sequence (X1 , . . . , Xk , Y1 , . . . , Ym ).
Given a consistent denumerable set S, let s1 , . . . , sn , . . . be an enumer-
ation of S. We let θ1 be the unit sequence (s1 ). This sequence is obviously
consistent with S, since s1 ∈ S. This concludes the first stage.
Now suppose we have completed the nth stage and have on hand a
sequence (X1 , . . . , Xk ), where k ≥ n. We call this sequence θn , and we
assume it to be consistent with S. We then extend θn to a finite sequence
θn+1 in a manner depending on the nature of Xn :
(a) If Xn is an α, we take θn+1 to be θn , α1 , α2 , sn+1 .
 Chapter 2. More on First-Order Logic 51

(b) If Xn is an β, then either θn , β1 or θn , β2 is consistent with S. If the

former we take θn+1 to be θn , β1 , sn+1 ; if the latter, we take θn+1 to
be θn , β2 , sn+1 .
(c) If Xn is a γ, we let a be the first parameter for which γ(a) is not a
term of θn , and we take θn+1 to be θn , γ(a), γ, sn+1 .
(d) If Xn is a δ, we take θn+1 to be θn , δ(a), sn+1 , where a is a parameter
new to S and new to θn .
This concludes stage n + 1 of the sequence.
Since for each n the sequence θn is consistent with S, then no formula
and its conjugate (or negation, if we are dealing with unsigned formulas)
are terms of θn . Hence no formula and its conjugate (or negation) are terms
of θ. Also it is obvious from the construction that if α is a term of θ, so are
α1 and α2 ; and if β is a term of θ, so is either β1 or β2 ; and if γ is a term of
θ, then so is γ(a) for every parameter a (since we encounter γ denumerably
many times); and if δ is a term of θ, then so is δ(a) for some parameter a.
Thus the set of terms of θ is a Hintikka set and includes all elements of S,
and so S is denumerably satisfiable in the domain of the parameters.
This construction makes no explicit reference to trees, although θ is in
fact the leftmost open branch of a systematic tableau for S. This proof does
not use König’s Lemma.

Applications: We will provide a sketch of the application of Γ-consistency

in proofs of the important results listed above. The details will be left to
the reader.

T1 : (Completeness of the tableau method).

Call a set S tableau-consistent if there exists no closed tableau for S. It
is easily seen that tableau-consistency is an analytic consistency property.
Hence by the Unification Theorem, every tableau-consistent set is satisfi-
able. Therefore, if a set S is unsatisfiable, there must be a closed tableau for
S. Now, if a formula X is valid, then the set {∼X} is unsatisfiable, so that
there is a closed tableau for ∼X, which means that X is tableau provable.

T2 : (The Skolem-Löwenheim Theorem).

Satisfiability itself is easily seen to be an analytic consistency property.
Hence by Theorem U, every satisfiable set is satisfiable in the denumerable
domain of the parameters.
52 Part I: More on Propositional and First-Order Logic

T3 : (The Compactness Theorem).

Call a set S F -consistent if all finite subsets of S are satisfiable.
F -consistency is easily seen to be an analytic consistency property. Hence,
by Theorem U, if all finite subsets of S are satisfiable, then S is satisfiable.

T4 : (Completeness of an axiom system for first-order logic).

Give an axiom system for first-order logic, call a set S consistent (with
respect to the system) if no element of S is refutable, and no finite set
{X 1 , . . . , Xn } of elements of S are such that the conjunction X1 ∧ . . . ∧ Xn
is refutable. For the axiom system given in The Beginner’s Guide [Smullyan,
2014], or for any other standard system, this consistency is easily shown to
be an analytic consistency property. Hence every consistent set is satisfiable
and every unsatisfiable set is inconsistent. If now X is valid, then ∼X is
unsatisfiable, and hence inconsistent. This means that ∼X is refutable, i.e.
(∼∼X) is provable, and so is X. Thus every valid formula is provable, and
the system is complete.

T5 : (The Regularity Theorem).

Let us review some definitions and facts: We recall that a (finite) regular
set R means a set whose members can be arranged in a sequence
Q1 ⊃ Q1 (a1 ), . . . , Qn ⊃ Qn (an ),
where for each i ≤ n, Qi is either a γ or a δ, and if Qi is a δ, the parameter
ai does not occur in δ, nor in any earlier term of the sequence. Such a
parameter (occurring in a Qi (ai ) where Qi is a δ in a regular set R) is
called a critical parameter of R.
We recall that a set S is said to tautologically imply a formula X
if X is true in all Boolean valuations that satisfy S. For S a finite set
{X1 , . . . , Xn }, this condition is equivalent to the condition that the sen-
tence (X 1 ∧ . . . ∧ Xn ) ⊃ X is a tautology (or, in the case that n = 1, that
X1 ⊃ X is a tautology). A finite set {X1 , . . . , Xn } is said to be truth-
functionally unsatisfiable if the sentence ∼(X 1 ∧ . . . ∧ Xn ) is a tautology.
The Regularity Theorem is that for every valid sentence X, there is
a regular set R that tautologically implies X, and which is such that no
critical parameter of R occurs in X.
 Chapter 2. More on First-Order Logic 53

For any finite set S (of sentences), an associate of S is a regular set R

such that R ∪ S is truth-functionally unsatisfiable, and no critical param-
eter of R occurs in any element of S. Now the Regularity Theorem is
equivalent to the proposition that every unsatisfiable (finite) set S has an
associate.

Problem 9. Show that the Regularity Theorem is equivalent to the propo-

sition that every unsatisfiable (finite) set S has an associate.

Let us now call a finite set S A-consistent if S has no associate. It can

be shown that A-consistency is an analytic consistency property. Hence by
Theorem U, for any finite set S, if S has no associate, then S is satisfiable.
Hence if S is unsatisfiable, then S has an associate, which proves the
Regularity Theorem [by Problem 8].

T6 : (Craig’s Interpolation Lemma).

By a partition S1 |S2 of a set S we mean a pair of subsets S1 and S2 of S
such that every element of S is in either S1 or S2 , but not in both; in other
words S1 ∪ S2 = S and S1 ∩ S2 = ∅.

By a partition interpolant between S1 and S2 we mean a sentence Z

such that all predicates and parameters of Z occur in both S1 and S2 , and
S1 ∪ {∼Z} and S2 ∪ {Z} are both unsatisfiable. Now, call a set S Craig-
consistent if there is a partition S1 |S2 of S such that there is no partition
interpolant between S1 and S2 . Well, Craig-consistency can be shown to
be an analytic consistency property. Hence by Theorem U, every Craig-
consistent set is satisfiable. Therefore if S is unsatisfiable, then for every
partition S1 |S2 of S, there is a partition interpolant between S1 and S2 .
Now, suppose X ⊃ Y is valid. Then the set {X, ∼Y } is unsatisfi-
able. Consider the partition {X}|{∼Y }. Then there is a partition inter-
polant Z between {X} and {∼Y }. Thus {X, ∼Z} (which is {X} ∪ {∼Z})
and {Z, ∼Y } (which is {∼Y } ∪ {Z}) are both unsatisfiable, hence
X ⊃ Z and Z ⊃ Y are both valid, and of course every predicate
and parameter of Z are in both X and Y . Thus Z is an interpolant
of X ⊃ Y .

Grand Exercise. Verify that the six properties which I claimed to be

analytic consistency properties really are so.
54 Part I: More on Propositional and First-Order Logic

V. A Henkin-Style Completeness Proof

Adolf Lindenbaum (a Polish logician born in 1904 and killed by the

Nazis in 1941), proved in the 1920s that every consistent set could
be extended to a maximally consistent set, a result that afterwards became
known as Lindenbaum’s Lemma. Leon Henkin [1949] published a proof
of the completeness of various axiom systems for first-order logic by a
method which is completely different from the ones we have so far con-
sidered, and which incorporates the method Lindenbaum used to prove his
lemma.
Henkin’s method combines maximal consistency with another condition:
A set S of sentences is called E-complete (existentially complete) if for any
element δ in S, there is at least one parameter a such that δ(a) ∈ S. Henkin
showed that every consistent set S can be extended to a set M that is both
maximally consistent and E-complete, and that every such set M is a (first-
order) truth set. We will see now that Henkin’s method applies to a more
general situation.
Recall from Section IV that we are using the phrase “the set S of sen-
tences is Γ-consistent” to mean that S has the property Γ and that Γ is
an analytic consistency property. We define a property ∆ of sentences to
be a synthetic consistency property, if it is an analytic consistency property
and satisfies the additional condition that for any set S having property
∆, and any sentence X, either S ∪ {X} has the property ∆, or S ∪ {∼X}
has the property ∆. We will call a set ∆-consistent if it has a particular
synthetic consistency property ∆. So in what follows “Γ” will be used to
designate any analytic consistency property, and “∆” will be used to desig-
nate any synthetic consistency property (but remember that any synthetic
consistency property is also an analytic consistency property).
A property P of sets is said to be of finite character if for any set S, S
has the property P if and only if all finite subsets of S have property P .
Henkin’s completeness proof for standard axiom systems for first-order
logic goes through for any synthetic consistency property ∆ of finite
character. In this more general form its statement is the following: for any
synthetic consistency property ∆ of finite character, any ∆-consistent set S
can be extended to a maximally ∆-consistent set M , and this can be done
in such a way that the set M will be a first-order truth set.
 Chapter 2. More on First-Order Logic 55

We will show a suitable modification of Henkin’s proof for ∆-consistent

properties (synthetic consistency properties) of finite character yields the
following Henkin-style proof for Γ-consistent properties of finite character,
i.e. for any analytic consistency property of finite character).
So, consider any analytic consistency property Γ that is also of finite
character. We will show that if S is Γ-consistent, then S can be extended to
a set M which is both maximally Γ-consistent and E-complete, and that the
set M obtained in this way is a Hintikka set, and hence satisfiable (so that
M ’s subset S is also satisfiable).

Problem 10. Show that if M is both maximally Γ-consistent and

E-complete, then M is a Hintikka set.
Note: As previously remarked, for any synthetic consistency property ∆
of finite character, if M is both maximally ∆-consistent and E-complete,
then M is not only a Hintikka set, but even a truth set. Indeed, more
generally, any Hintikka set S having the property that for every X, either
X ∈ S or (∼X) ∈ S, must be a truth set. We leave the proof of this as an
exercise for the reader.
Now, given an analytic consistency property Γ of finite character, how
do we actually go about extending a given Γ-consistent set S of finite char-
acter to a set M which is both maximally Γ-consistent and E-complete?
(As in our proof of Theorem U above, we will assume that the number of
parameters in S is finite, or that we have an additional denumerable num-
ber of parameters available to us, none of which occur in any formula of S,
i.e. parameters that we can just add to the parameters in S if necessary.)
Here is Lindenbaum’s method of extending a consistent set S to a
maximally consistent set M (assuming consistency to be of finite charac-
ter), which we will extend from the conventional meaning of consistent
to our meaning of Γ-consistent (i.e. analytically consistent): We first
arrange all the sentences of first-order logic in some infinite sequence
X1 , X2 . . . , Xn , Xn+1 , . . . . Now consider a Γ-consistent set S. We construct
an infinite sequence of sets S0 , S1 , . . . , Sn , Sn+1 , . . . as follows: We take
S0 to be S. Then, for every positive n, assuming Sn already defined, we
take Sn+1 to be Sn if Sn ∪ {Xn } is not consistent; while if Sn ∪ {Xn } is
consistent, and if Xn is not some δ, then we take Sn+1 to be Sn ∪ {Xn }.
But if Sn ∪ {Xn } is consistent and Xn is some δ, then Sn ∪ {δ, δ(a)} is
also consistent for any parameter a new to Sn ∪ {δ} (since we are dealing
56 Part I: More on Propositional and First-Order Logic

with an analytic consistency property). In this case we take Sn+1 to be

Sn ∪ {δ, δ(a)}, where a is a parameter new to Sn and to δ. The set M
that results from taking the union of all the sets S0 , S1 , . . . , Sn , Sn+1 , . . .
is both maximally consistent and E-complete. For it is easily seen to be E-
complete by construction (realize that at every stage of the construction,
there will always be a new parameter available). And the union M of all
the consistent sets Sn is consistent because we are assuming our consistency
property to be of finite character. And it is maximally consistent because
any formula not already in it was kept out of M because adding it would
have made a finite subset of M inconsistent.
Immediately after proving that every consistent set S can be extended to
a set M that is both maximally consistent and E-complete, Henkin stated
as a corollary the completeness of first-order logic, that is, that every valid
formula was provable. But he did use a well-known property of the axiom
systems he was considering that does not apply to the slightly different
formalization of first-order logic in this volume (his systems included the
constants t and f , which we did discuss in the context of propositional logic
in The Beginner’s Guide [Smullyan, 2014], and which can also be employed
in first-order logic).
Now, we might try to show the completeness of first-order logic simi-
larly to the way Henkin did by first showing that ordinary consistency is an
analytic consistency property (a common definition of an individual sen-
tence X being consistent would be that there exists a formula Z such that
X ⊃ Z is not provable). But far easier for us is to just remind ourselves
that in Section IV of this chapter (the Unification section), we showed that
both the first-order tableau method of proof and the standard first-order
axiomatic systems of proof are complete by defining the appropriate ana-
lytic consistency property for each of them.
We now see that there are basically two types of completeness proofs
directly applicable to the usual formulations of First-Order Logic. One is
along the lines of Lindenbaum and Henkin, which extends a consistent set
directly to a truth set, and for this we need the full force of synthetic
consistency properties. The second type of completeness proof (which is
along the lines of Gödel, Herbrand, Gentzen, Beth, and Hintikka) extends
a consistent set, not directly to a truth set, but to a Hintikka set, which
can be further extended to a truth set. And for the second type of proof,
we do not need synthetic consistency, but only analytic consistency.
 Chapter 2. More on First-Order Logic 57

Solutions to the Problems of Chapter 2

1. We are given that v is a Boolean valuation.

(1) We are given that if γ is true under v, then so is γ(a) for every
parameter a. In addition, we must prove the converse, i.e. that if
γ(a) is true under v for every parameter a, then γ is true under v.
We shall prove the equivalent proposition that if γ is false under v,
then so is γ(a) for at least one parameter a.
We first consider the case that γ is of the form ∀ xϕ(x). Now
suppose that ∀ xϕ(x) is false under v. Then ∼∀ xϕ(x) is true under
v [since v is a Boolean valuation]. Hence by (2) of the lemma, there
is at least one parameter a such that ∼ϕ(a) is true under v. Hence
ϕ(a) is false under v, which is what was to be shown [since if γ is
∀ xϕ(x), γ(a) is ϕ(a)].
Now consider the case that γ is of the form ∼∃ xϕ(x). Well,
suppose that ∼∃ xϕ(x) is false under v. This means that ∃ xϕ(x) is
true under v [since v is a Boolean valuation]. Then by F2 , ϕ(a) is
true for at least one parameter a. But then ∼ϕ(a) is false for that
parameter. And that is what we needed to prove, since when the γ
formula is of the form ∼∃ xϕ(x), γ(a) is ∼ϕ(a).
(2) We are given that if δ is true under v, then so is δ(a) for at least
one parameter a. Again we must prove the converse, i.e. that if
δ(a) is true for at least one parameter a, so is δ. We will prove the
equivalent proposition that if δ is false, then so is δ(a) for every
parameter a.
We first consider the case that δ is of the form ∃ xϕ(x). So sup-
pose ∃ xϕ(x) is false. Then ∼∃ xϕ(x) is true under v [since v is a
Boolean valuation]. Then by (1) of the Lemma, ∼ϕ(a) is true for
every parameter a. Hence ϕ(a) is false for every parameter a, which
was what had to be proved [since ϕ(a) is δ(a) when δ is of the form
suppose ∃ xϕ(x)].
Now consider the case that δ is of the form ∼∀ xϕ(x). Well, sup-
pose ∼∀ xϕ(x) is false. Then ∀ xϕ(x) is true [since v is a Boolean
valuation]. Then by F1 , ϕ(a) is true for every parameter a. This
means that, ∼ϕ(a) is false for every parameter a [since v is a
Boolean valuation], which is what had to be proved [since when
δ is of the form ∼∀ xϕ(x), δ(a) is of the form ∼ϕ(a)].
58 Part I: More on Propositional and First-Order Logic

2. We must show that every magic set M is a truth-functional basis for

first-order logic. Thus we must show that if M is a magic set and
X is a pure sentence, then X is valid if and only if X is tautolog-
ically implied by some finite subset M0 of M . We assume M is a
magic set.
(a) Suppose that X is a valid pure sentence. Then X is true under all
first-order valuations, and thus also true under all Boolean valu-
ations that satisfy M (because all such valuations are also first-
order valuations by condition M1 in the definition of a magic set).
Thus X is tautologically implied by M [by the definition of tauto-
logical implication], and hence by some finite subset M0 of M
[by Problem 8 of Chapter 6 of The Beginner’s Guide [Smullyan,
2014]].
(b) Conversely, suppose X is a pure sentence that is tautologically
implied by some finite subset M0 of M . Then the set M0 ∪ {∼X} is
not truth-functionally satisfiable (i.e. not satisfied by any Boolean
valuation), so that M0 ∪ {∼X} is not first-order satisfiable (since
every first-order valuation is also a Boolean valuation). If ∼X were
first-order satisfiable, then M0 ∪ {∼X} would also be first-order
satisfiable [by M2 ], but since it isn’t, then ∼X is not first-order
satisfiable, which means that X is valid.
This completes the proof.
3. We must show that if M is a complete regular set of sentences, then M
is a magic set. So let M be a complete regular set of sentences.
Proof of M 1 : We must show that any Boolean valuation that satisfies
M is also a first-order valuation. So suppose v is a Boolean valuation
that satisfies M . By Proposition 1, to show that v is a first-order valu-
ation, it suffices to show that if γ is true under v, so is γ(a) for every
parameter a, and that if δ is true under v, then so is δ(a) for at least
one parameter a. Well, suppose γ is true under v. Since M is complete,
γ ⊃ γ(a) is in M for every parameter a. Thus γ ⊃ γ(a) is true under
v (since every member of M is true under v). Then γ(a) must also be
true under v (since v is a Boolean valuation).
Now suppose δ is true under v. Since M is complete, δ ⊃ δ(a) is in
M for some parameter a. Thus δ ⊃ δ(a) if true under v (since every
member of M is true under v). Then δ(a) must also be true under v
(since v is a Boolean valuation).
 Chapter 2. More on First-Order Logic 59

Proof of M 2 : We must show that for every finite set S of pure sen-
tences, and every finite subset M0 of M , if S is first-order satisfiable,
so is S ∪ M0 .
We first show that for any finite set S of sentences (possibly with
parameters), if S is (first-order) satisfiable, then:
(1) S ∪ {γ ⊃ γ(a)} is satisfiable, for any γ and any a;
(2) S ∪ {δ ⊃ δ(a)} is satisfiable, for any δ and any a that does not
occur in δ or in any element of S.
Now (1) is obvious, since γ ⊃ γ(a) is valid, hence true under all
interpretations.
As for (2), let I be an interpretation of all predicates and parameters
of S in some non-empty universe U which is such that all elements
of S are true under I. Extend I to an interpretation I1 by assigning
values in U to all predicates and parameters of δ which do not occur
in S. Now let a be any parameter which occurs neither in δ nor in
any element of S. We now wish to extend I1 to an interpretation I2
in which δ(a) has a truth value — that is, we must assign a value k
of U to the parameter a and we wish to do so in a way that makes
δ ⊃ δ(a) true. Well, if δ is false under I1 , then any k in U will do, since
δ ⊃ δ(a) will automatically be true. On the other hand, if δ is true
under I1 , then there must be at least one element k in U such that δ(a)
is true when k is assigned to a. Thus, we assign such an element k to
a, and then, since this assignment makes δ(a) true, δ ⊃ δ(a) must also
be true.
This proves (1) and (2). Finally, consider a finite subset M0 of M .
Then M0 is a finite regular set. Arrange M0 in some regular sequence
Q1 ⊃ Q1 (a1 ), . . . , Qn ⊃ Qn (an ), where for each i ≤ n, Qi is either a
γ or a δ, and if Qi is a δ, the parameter ai does not occur in δ, nor in
any earlier term of the sequence.
Now let S be any satisfiable finite set of pure sentences. Let
S1 = S ∪ {Q1 ⊃ Q1 (a1 )};

And for each i < n, let

Si+1 = Si ∪ {Qi+1 ⊃ Qi+1 (ai+1 )}.
By (1) and (2) above, S1 is satisfiable, and for each i < n, if Si is
satisfiable, so is Si+1 . It then follows by mathematical induction that
60 Part I: More on Propositional and First-Order Logic

each of the sets S1 , S2 , . . . , Sn is satisfiable. Thus S ∪ M0 , which is Sn ,

is satisfiable.
4. Let M be a complete regular set of sentences. By Theorem 2, it is also a
magic set, so that by Theorem 1, it is a truth-functional basis for first-
order logic. Thus, for any valid pure sentence X there is a finite subset
R of M which truth-functionally implies X. The set R is obviously
regular.
5. For any set S of formulas, let Con S be the conjunction of the elements
of S (the order does not matter) if S is non-empty, or t if S is empty
(in which case all elements of S are true). In either case, under any
interpretation, Con S is true if and only if all elements of S are true.
Let Dis S be the disjunction of all the elements of S if S is non-empty,
or f if S is empty. In either case, under any interpretation, Dis S is
true iff at least one element of S is true. A sequent U → V is logically
equivalent to the simple formula Con U ⊃ Dis V .
We are to show that U, X → V is logically equivalent to U → V, ∼X,
or what is the same thing, that Con (U, X) ⊃ Dis V is equivalent to
Con U ⊃ Dis (V, ∼X). [The proof that U → V, X is logically equiva-
lent to U, ∼X → V is similar.]
Well, under any interpretation (Boolean valuation), X is either true
or false.
Case 1. Let X be true. Then Con (U, X) is logically equiva-
lent to Con (U ), so that Con (U, X) ⊃ Dis V is logically equiv-
alent to Con (U ) ⊃ Dis V . But also, since X is true, ∼X is
false, so that Dis (V, ∼X) is logically equivalent to Dis (V ). Thus
Con U ⊃ Dis(V, ∼X) is also equivalent to Con U ⊃ Dis (V ). So in
this case, the sequents U, X → V and U → V, ∼X are both equivalent
to U → V , hence equivalent to each other.
Case 2. Let X be false. Then Con (U, X) is false, hence
Con (U, X) ⊃ Dis V is true. But also, since X is false, then ∼X is
true, so that Dis (V, ∼X) is true, and therefore Con U ⊃ Dis (V, ∼X)
is true. Thus in this case the sequents U, X → V and U → V, ∼X are
both true, hence equivalent to each other.
6. S is a set T X1 , . . . , T Xn , F Y1 , . . . , F Yk of signed formulas. We let U
be the set X1 , . . . , Xn and let V be the set {Y1 , . . . , Yk }.
 Chapter 2. More on First-Order Logic 61

First, let us consider the axioms {S, T X, F X} of the uniform

system. The set {S, T X, F X} is the set
{T X1 , . . . , T Xn , F Y1 , . . . , F Yk , T X, F X}
and so |S, T X, F X| is the sequent
X1 , . . . , Xn , X → Y1 , . . . , Yk , X.
This is U, X → V, X, the axiom of G0 .
As for the inference rules, we must consider the various α and
β cases separately. For example, consider the case that α is of the
form F X ⊃ Y . Then α1 = T X and α2 = F Y . Then |S, α1 , α2 | is
|T X1 , . . . , T Xn , F Y1 , . . . , F Yk , T X, F Y |, which is the sequent
X1 , . . . , Xn , X → Y1 , . . . , Yk , Y,
which is U, X → V, Y , and so for the case α = F X ⊃ Y , Rule A is
Rule I1 of the system G0 .
We leave it to the reader to verify the other cases, specifically that if
α is T X ∧ Y, F X ∨ Y, T ∼X, F ∼X, then Rule A is respectively Rule
C1 , D1 , N2 , N1 of the system G0 , and if β is F X∧Y, T X∨Y, T X ⊃ Y ,
then Rule B is respectively Rule C2 , D2 , I2 .
7. First for the correctness of the altered tableau method. We are to show
that if there is a closed altered tableau for a set, then the set is really
unsatisfiable.
For any satisfiable finite set S of signed formulas, the following facts
hold:
(1) If A ∈ S then S ∪ {A1 } and S ∪ {A2 } are both satisfiable.
(2) If B ∈ S then either S ∪ {B1 } or S ∪ {B2 } is satisfiable.
(3) If C ∈ S then for every parameter a the set S ∪{C(a)} is satisfiable.
(4) If D ∈ S then for every parameter a new to S, the set S ∪ {D(a)}
is satisfiable.
These facts can be proved in the same manner as in The Beginner’s
Guide [Smullyan, 2014], replacing α, β, γ, δ by A, B, C, D respectively.
It then follows that if S is satisfiable, at no stage can an altered tableau
for S close. Hence if there is a closed altered tableau for S, then S is
unsatisfiable. Thus the altered tableau method is correct.
62 Part I: More on Propositional and First-Order Logic

As for completeness, one can make a systematic altered tableau

for a set in essentially the same manner as for analytic tableaux (in
the instructions given in The Beginner’s Guide [Smullyan, 2014], just
replace α, β, γ, δ by A, B, C, D respectively). If the systematic altered
tableau runs on forever without closing, then it contains at least one
infinite open branch, and the set S of the sentences on the branch
satisfies the following conditions:
(0) For no X in S is it the case that both T X and F X are both in S,
or that T X and T ∼X are both in S or that F X and F ∼X are
both in S.
(1) If A ∈ S, so are A1 and A2 .
(2) If B ∈ S, then B1 ∈ S or B2 ∈ S.
(3) If C ∈ S, then so is C(a) for every parameter a.
(4) If D ∈ S, then so is D(a) for at least one parameter a.
Such a set S is very much like a Hintikka set, and the proof of
its satisfiability is very close to that for Hintikka set. About the only
significant difference is in the assignment of truth values to the atomic
sentences. Now, for any atomic unsigned sentence X, if T X ∈ S or
F ∼X ∈ S, give X the value truth (in which case T X and F ∼X are
then true). If either T ∼X ∈ S or F X ∈ S, then give X the value
false (in which case T ∼X and F X are then true). [No ambiguity can
result, because if T X ∈ S or F ∼X ∈ S, then neither T ∼X ∈ S nor
F X ∈ S can hold, and if T ∼X ∈ S or F X ∈ S, then neither T X ∈ S
nor F ∼X ∈ S can hold.] If none of T X, T ∼X, F X, F ∼X occur in S,
then give X the value truth or falsity at will — say truth. Under this
valuation, all elements of S can be shown to be true by mathematical
induction on degrees, much as in the proof of Hintikka’s Lemma. Thus
S is satisfiable.
We now see that if the systematic altered tableau for S does not close,
then S is satisfiable, hence if S is not satisfiable, then any systematic
altered tableau for S must close. This completes the completeness proof
for the altered tableau method.
To complete the proof of the completeness of the Gentzen system
GG, we now need altered block tableaux for first-order logic, whose
rules in uniform notation are the following:
 Chapter 2. More on First-Order Logic 63

∗
Providing a is not in the premise.

We recall that if S is the set {T X1 , . . . , T Xn , F Y1 , . . . , F Yk }, by

|S| we mean the sequent X 1 , . . . , Xn → Y1 , . . . , Yk .
For any block tableau rule of the form
S1

S2
by the counterpart of the rule we mean
|S2 |

|S1 |
(i.e. “from the sequent |S2 |, to infer the sequent |S1 |”). For any block
tableau rule of the form
S1

S2 S3
by its counterpart is meant the rule
|S2 | |S3 |

|S1 |
64 Part I: More on Propositional and First-Order Logic

(i.e. “from the sequents |S2 | and |S3 |, to infer the sequent |S 1 |”). Well,
the counterparts of the eight block tableau rules are precisely the eight
rules of the system GG. To see this, let U be the set of unsigned sen-
tences X such that T X ∈ S and let V be the set of sentences Y such
that F Y ∈ S. Consider, for example, the block tableau rule
{S, T α}

{S, T α1 , T α2 }
The counterpart of this rule is
|S, T α1 , T α2 |

|S, T α|
but this is just
U, α1 , α2 → V

U, α → V
which is a rule of GG.
Another example: The counterpart of the rule is

{S, F α}

{S, F α1 } {S, F α2 }

|S, F α1 | |S, F α2 |

|S, F α|
which is
U → V, α1 U → V, α2

U → V, α
which is a rule of GG.
The reader can verify the cases of the other six rules.
Now if T is a closed altered block tableau for a set of signed formulas
whose counterpart is the sequent U → V , each end point S is closed,
and the sequence |S| is an axiom of GG:
 Chapter 2. More on First-Order Logic 65

(i) |S, T X, F X| is U, X → V, X;
(ii) |S, T X, T ∼X| is U, X, ∼X → V ;
(iii) |S, F X, F ∼X| is U → V, X, ∼X .
And since the counterparts of the block rules are rules of GG, it follows
that if we replace each point S of the tree T by the sequent |S| and
turn the tree upside down, we have a proof of the sequent U → V in
the system GG. This concludes the proof of the completeness of GG.
8. It is understood that P 0 is a predicate distinct from each of
P, P1 , . . . , Pn , and that P 0 occurs in no element of S, and that S 0
is the result of substituting P 0 for P in every element of S.
Now suppose that P is explicitly definable from P1 , . . . , Pn with
respect to S. Let ϕ(x) be a formula whose predicates are among
P1 , . . . , Pn but do not involve P and which is such that
S ` ∀ x(P x ≡ ϕ(X)).
Then also, of course, S ` ∀ x(P 0 x ≡ ϕ(X)). Hence,
0

S ∪ S 0 ` ∀ x(P x ≡ ϕ(X)) ∧ ∀ x(P 0 x ≡ ϕ(X)),

which implies
S ∪ S 0 ` ∀ x((P x ≡ ϕ(X)) ∧ (P 0 x ≡ ϕ(X))).
Therefore S ∪ S 0 ` ∀ x(P x ≡ P 0 x) is logically valid, as the reader
can verify (say, with a tableau). Thus P 0 is implicitly definable from
P1 , . . . , Pn with respect to S.
9. (a) Suppose every unsatisfiable set has an associate. Consider any valid
sentence X. Then the unit set {∼X} is unsatisfiable, hence has
an associate R. Since R ∪ {∼X} is not truth-functionaly satisfi-
able, then R tautologically implies X [since X must be clearly
true in any Boolean valuation satisfying R]. Moreover, X con-
tains no critical parameter of R, which proves the Regularity
Theorem.
(b) Conversely, assume the Regularity Theorem. Let S be a finite set
{X1 , . . . , Xn } of sentences which is unsatisfiable. Then the sen-
tence ∼(X1 ∧ . . . ∧ Xn ) is valid. Hence by the assumed Regular-
ity Theorem, there is a regular set R which tautologically implies
∼(X1 ∧ . . . ∧ Xn ) and which is such that no critical parameter
of R occurs in any element of ∼(X1 ∧ . . . ∧ Xn ), and thus no
critical parameter of R occurs in any element of S. Since R tau-
tologically implies ∼(X1 ∧ . . . ∧ Xn ), then R ∪ {X1 , . . . , Xn } is
66 Part I: More on Propositional and First-Order Logic

truth-functionally unsatisfiable, i.e. R ∪ S is truth-functionally

unsatisfiable. Thus R is an associate of S.
10. We are given that M is maximally Γ-consistent and E-complete, and we
are to show that M is a Hintikka set. In what follows, “consistent” will
mean Γ-consistent. We shall say that a sentence X is consistent with M
if the set M ∪ {X} is consistent. Since M is maximally consistent, then
if X is consistent with M , then X must be a member of M (otherwise
M would be only a proper subset of the larger consistent set M ∪ {X},
contrary to M ’s maximality).
H0 : No formula and its conjugate (or formula and its negation,
if unsigned formulas are in questions) can be in the set M ,
because M is consistent, and that is one of the conditions on (Γ)
consistency.
H1 : Suppose α ∈ M . Then of course α is consistent with M . Thus
α1 and α2 are both consistent with M (since Γ is an analytic
consistency property). Therefore α1 and α2 are members of M .
This proves that if α ∈ M , then α1 ∈ M and α2 ∈ M .
H2 : Suppose β ∈ M . Then β is consistent with M and thus either β1 is
consistent with M or β2 is consistent with M , and therefore either
β1 ∈ M or β2 ∈ M .
H3 : Suppose γ ∈ M . Then γ is consistent with M , so that γ(a) is
consistent with M for every parameter a. Therefore γ(a) ∈ M for
every parameter a.
H4 : Suppose δ ∈ M . Then δ(a) ∈ M for at least one parameter a by
the assumption that M is E-complete.
By H1 −H4 , M is a Hintikka set.
 Part II

Recursion Theory
and Metamathematics
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 3

Some Special Topics

The items of this chapter represent reconstructions and generalizations of

various results in recursion theory, undecidability and incompleteness.
First for a preliminary matter, which I should have addressed much earlier:
By a pair (x, y) [more often called an ordered pair], is meant a set whose
elements are x and y, together with a designation which specifies that x is
the first element and y the second. In contrast, the unordered pair {x, y} [note
the curly brackets instead of parentheses] is just the set whose members are x
and y, without any specification that one of the two elements comes first and
the other second. The set {x, y} is the same as the set {y, x}, but the pair (x, y)
is different from the pair (y, x) [except in the case when x = y].
Throughout this chapter, the word number shall mean positive integer,
and we will consider an operation that assigns to each pair (x, y) of num-
bers a number denoted xy, or sometimes x ∗ y. We will assume the operator is
such that xy (or x ∗ y) uniquely determines the x and the y — that is, the only
way that x ∗ y can be the same as z ∗v is when x = z and y = v. And we assume
that every number n is assigned to some pair x ∗ y, for some x and some y. We
generally assume also that since the function ∗ is 1-1 and onto from the set of
ordered pairs of positive integers to the set of positive integers, we can use its
inverse to go from a given a number n to the pair (x, y) such that n = x ∗ y.

I. A Decision Machine

We consider a calculating machine with infinitely many registers R1 , R2 , . . . ,

Rn , . . . . We call n the index of Rn . Each register is, so to speak, in charge
70 Part II: Recursion Theory and Metamathematics

of a certain property of numbers. To find out whether a given number n

has a certain property, one goes to the register in charge of that property
and feeds in the number. The machine then goes into operation, and one
of three things happens:
(1) The machine eventually halts and flashes a signal — say a green light —
signifying that the number does have the property, in which case we
say that the register affirms n.
(2) The machine eventually halts and flashes a different signal — say a
red light — to indicate that the number doesn’t have the property, in
which case we say that the register denies n.
(3) The machine runs on forever, without ever being able to determine
whether or not the number does have the property, in which case we
say that n stumps the register, or that the register is stumped by n.
We assume we are given a machine M that satisfies the following two
conditions:
M1 : With each register R in the calculating machine is associated another
register R0 in the calculating machine called the opposer of R, which
affirms those numbers which R denies, and denies those numbers which
R affirms (and hence is stumped by those and only those numbers
which stump R).
M2 : With each register R in the calculating machine is associated another
register R# in the calculating machine called the diagonalizer of R,
such that for any number x, the register R# affirms x iff R affirms xx
and denies x iff R denies xx. (I.e. R# affirms x iff R affirms x ∗ x and
R# denies x iff R denies x ∗ x, for our special function ∗ which has the
properties listed above for it; we will write xy for x ∗ y from now on
in this chapter.)
Recall our denumerable list of registers R1 , R2 , . . . , Rn , . . . . For any
number n, we let n0 be the index of (Rn )0 , and let n# be the index of
(Rn )# . Thus Rn0 = (Rn )0 and Rn# = (Rn )# .
We will say that two registers Ra and Rb behave the same way towards
a number n when they either both affirm n, or both deny n or are both
stumped by n. We call two registers similar if they behave the same way
towards every number. It is easy to see that for any register R, the registers
R0# and R#0 are similar, for each denies x iff R affirms xx, and each affirms
#
x iff R denies xx. (R#0 means the same as (R0 ) and R#0 means the same as
 Chapter 3. Some Special Topics 71

(R# )0 , but we will always eliminate the parentheses if the resulting meaning
is clear.)

Universal Registers

We shall call a register U universal if, for all numbers x and y, the register
U affirms xy iff Rx affirms y.
The propositions that follow will be proved later on in a more general
framework. Meanwhile, some of you might like to try proving them on your
own.

Proposition 1.1. Any universal register can be stumped (by some number
or other).

Proposition 1.2. If at least one of the registers is universal, then some

register is stumped by its own index (i.e. for some number n, the register
Rn is stumped by n).

Contra-Universal Registers

We shall call a register V contra-universal if for all numbers x and y, the

register V affirms xy iff Rx denies y.

Proposition 1.3. Any contra-universal register V can be stumped.

Proposition 1.4. If some register is universal, then some register is

stumped by the index of its opposer. If some register is contra-universal,
then some register is stumped by its own index.

Creative Registers

We shall call a register C creative if, for every register R, there is at least
one number n such that C affirms n iff R affirms n.

Proposition 1.5. Any creative register can be stumped.

Proposition 1.6. Every universal register is creative, and every contra-

universal register is creative.
72 Part II: Recursion Theory and Metamathematics

Note: Propositions 1.5 and 1.6 provide additional and particularly sim-
ple proofs of Propositions 1.1 and 1.3.
For the rest of this section, we let A be the set of all numbers xy such
that Rx affirms y, and let B be the set of all numbers xy such that Rx
denies y.

Fixed Points

We call a number xy a fixed point of a register R if R affirms xy iff xy ∈ A

(i.e. Rx affirms y) and R denies xy iff xy ∈ B (i.e. Rx denies y).

Proposition 1.7. Every register has a fixed point.

Proposition 1.8. Suppose that U is universal and V is contra-universal.

Then:
(a) Any fixed point of U 0 will stump U .
(b) Any fixed point of V will stump V .

Proposition 1.9. Suppose R is a register that affirms all numbers in A and

doesn’t affirm any number in B. Then R can be stumped. [This proposition
is particularly interesting.]

A Halting Problem

Let us say that a register R halts at a number n if R either affirms n or

denies n. Call a register R a stump detector if R halts at those and only
those numbers xy such that Rx is stumped by y.

Proposition 1.10. No register is a stump detector.

Domination

We say that n dominates m if Rn affirms all numbers affirmed by Rm .

Proposition 1.11. Suppose U is universal and V is contra-universal and

that R affirms all numbers affirmed by U and denies all numbers affirmed
by V . Then R can be stumped.
This proposition can be restated in terms of domination as follows:
Suppose U is universal and V is contra-universal and there is a register R
 Chapter 3. Some Special Topics 73

such that R dominates U and R0 dominates V . Then there exists an x such

that R is stumped by x.

Affirmation Sets

By the affirmation set of a register R we shall mean the set of all numbers
affirmed by R. We shall call a set of numbers S an affirmation set if it is
the affirmation set of some register.

Proposition 1.12. If one of the registers is universal, then there exists an

affirmation set whose complement is not an affirmation set.

II. Variations on a Theme of Gödel

Some of the items here are repetitions of results in The Beginner’s Guide
[Smullyan, 2014]. The reason for these repetitions will be apparent later on.
Again, the propositions that follow will be proved later, in a more general
setting.
We consider a mathematical system S in which we have an infinite
sequence H1 , H2 , . . . , Hk , . . . , of expressions called predicates, and to each
predicate H and each number n is assigned an expression denoted H(n),
called a sentence. [Informally, we think of H as the name of some property
of numbers, and of the sentence H(n) as expressing the proposition that
the number n has the property named by H.]
We assume that every sentence is H(n) for some H and some n. We
arrange all the sentences in a sequence S1 , S2 , . . . , Sk , . . . in such a manner
that for numbers x and y the sentence Sx∗y is the sentence Hx (y). We call
n the index of the predicate Hn , and we call k the index of the sentence Sk .
Thus x ∗ y is the index of the sentence Hx (y).

Provable, Refutable and Undecidable Sentences

Some of the sentences are called provable and some are called refutable, and
we assume the system is consistent, by which we mean that no sentence
is both provable and refutable. A sentence is called decidable if it is either
provable or refutable, and it is called undecidable if it is neither provable nor
74 Part II: Recursion Theory and Metamathematics

refutable. The system is called complete if every sentence is decidable, and

is called incomplete otherwise. We let P be the set of all provable sentences
and R be the set of all refutable sentences.
For any set W of sentences, we let W0 be the set of all indices of the
elements of W . Then P0 is the set of all n such that Sn is provable, and R0
is the set of all n such that Sn is refutable.
We shall call two sentences X and Y equivalent if when one is provable,
so is the other, and when one is refutable, so is the other (and hence, when
one is undecidable, so is the other); in other words, they are either both
provable, both refutable, or both undecidable.
We are given that the system S obeys two conditions:
(1) With each predicate H is associated a predicate H 0 called the negation
of H such that for every number n the sentence H 0 (n) is provable iff
H(n) is refutable, and is refutable iff H(n) is provable.
(2) With each predicate H is associated a predicate H # called the diagonal-
izer of H such that for each number n the sentence H # (n) is equivalent
to the sentence H(n ∗ n).

Set Representation

We shall say that a predicate H represents the set of all numbers n such
that H(n) is provable. Thus to say that H represents a number set A is to
say that for all numbers n, the sentence H(n) is provable iff n ∈ A. We call
the set A representable if some predicate represents it.

Provability and Refutability Predicates

We shall call H a provability predicate if it represents the set P0 . Thus the

statement that H is a provability predicate is equivalent to the condition
that for every n the sentence H(n) is provable iff Sn is provable.
We shall call K a refutability predicate if it represents the set R0 .
Thus the statement that K is a refutability predicate is equivalent to
the condition that for every n the sentence K(n) is provable iff Sn is
refutable.

Proposition 2.1. If H is a provability predicate, then H(n) is undecidable

for some n.
 Chapter 3. Some Special Topics 75

Proposition 2.2. If one of the predicates is a provability predicate, then

there is some n such that Hn (n) is undecidable.

Proposition 2.3. If K is a refutability predicate, then K(n) is undecidable

for some n.
Recall that n0 is the index of the negation of Hn , so that (Hn0 ) = (Hn )0 .

Proposition 2.4. If some predicate is a provability predicate, then there is

some n such that Hn (n0 ) is undecidable. If some predicate is a refutability
predicate, then there is some n such that Hn (n) is undecidable.

Creative Predicates

We shall call a predicate K creative if for every predicate H there is at least

one number n such that H(n) is provable iff K(n) is provable.

Proposition 2.5. If K is creative, then K(n) is undecidable for some n.

Proposition 2.6. Every provability predicate is creative, and every refut-

ability predicate is creative.

Fixed Points

We shall call a number n a fixed point of a predicate H if Sn is equivalent

to H(n). [This is the same as saying that a number xy is a fixed point of a
predicate H if Hx (y) is equivalent to H(xy).]

Proposition 2.7. Every predicate has a fixed point.

Proposition 2.8. Suppose that H is a provability predicate and that K is

a refutability predicate. Then:
(a) If n is a fixed point of H 0 , then H(n) is undecidable.
(b) If n is a fixed point of K, then K(n) is undecidable.

Proposition 2.9. Suppose H is a predicate such that H(n) is provable

for any n in the set P0 , and H(n) is not provable for any n in R0 . Then
H(n) is undecidable for some n. [This is related to Rosser’s incompleteness
proof.]

Proposition 2.10. There cannot be a predicate H such that for all numbers
n the sentence H(n) is decidable iff Sn is undecidable.
76 Part II: Recursion Theory and Metamathematics

Domination

We will say that a predicate Hb dominates a predicate Ha if Hb (n) is prov-

able for every n for which Ha (n) is provable.

Proposition 2.11. If H dominates some provability predicate, and H 0

dominates some refutability predicate, then H(n) is undecidable for some n.

Proposition 2.12. If one of the predicates is a provability predicate,

then there exists a representable number set whose complement is not
representable.

III. R-Systems

The systems of I and those of II are really the same, only in different dress.
The common underlying theme will be fully explained in the next section
of this chapter. Meanwhile, we will consider yet another embodiment of the
theme, which has direct applications to recursion theory.
We consider a denumerable collection of number sets, which we will call
“R-sets”, and an enumeration A1 , A2 , . . . , An , . . . of them in some order,
and another enumeration B1 , B2 , . . . , Bn , . . . of them in some other order,
such that for any disjoint pair (S1 , S2 ) of R-sets, there is some n such
that S1 = An and S2 = Bn . We shall call such a pair of enumerations an
R-system if the following three conditions hold:
R0 : For every n, the set An is disjoint from Bn i.e., An ∩ Bn = ∅.
R1 : Associated with each number n is a number n0 such that An0 = Bn
and Bn0 = An [thus the ordered pair (An0 , Bn0 ) is the pair (Bn , An )].
R2 : Associated with each n is a number n# such that for every number x,
(1) x ∈ An# iff xx ∈ An .
(2) x ∈ Bn# iff xx ∈ Bn .
Note: The collection of all recursively enumerable sets can be arranged
in a pair of sequences A1 , A2 , . . . , An , . . . and B1 , B2 , . . . , Bn , . . . such that
conditions R0 , R1 and R2 all hold (for a recursive function f (x, y) = x ∗ y).
Thus all the results that follow are generalizations of results in recursion
theory.
 Chapter 3. Some Special Topics 77

Universal and Contra-Universal Sets

We call an R-set Aw universal if for all numbers x and y the number

xy ∈ Aw iff y ∈ Ax . We call an R-set Av contra-universal if for all numbers
x and y the number xy ∈ Av iff y ∈ Bx .

Proposition 3.1. If some R-set Aw is universal, then Aw is not the com-

plement of Bw (i.e. some number x lies outside both Aw and Bw ).

Proposition 3.2. If some R-set Aw is universal, then there is some n such

that n itself lies outside both An and Bn .

Proposition 3.3. If the R-set Av is contra-universal, then Av is not the

complement of Bv (i.e. some number x lies outside both Av and Bv ).

Proposition 3.4. If some R-set Aw is universal, then there is a number

n such that n0 lies outside both An and Bn . If some R-set Av is contra-
universal, then there is a number n such that n lies outside both An and Bn .

Creative R-Sets

We shall call an R-set Ac creative if for every R-set Ax there is a least one
number n such that n ∈ Ac iff n ∈ Ax .

Proposition 3.5. If Ac is creative, then Ac is not the complement of Bc .

Proposition 3.6. Any universal or contra-universal R-set is creative.

Fixed Points

We now let A be the set of all numbers xy such that y ∈ Ax , and let B be
the set of all numbers xy such that y ∈ B x .
We shall call a number x a fixed point of the pair (An , Bn ) if x ∈ An
iff x ∈ A and x ∈ Bn iff x ∈ B. [An equivalent definition is that a number
xy is a fixed point of a pair (An , Bn ) if xy ∈ An iff y ∈ Ax and xy ∈ Bn iff
y ∈ Bx .]

Proposition 3.7. Every pair (An , Bn ) has a fixed point.

78 Part II: Recursion Theory and Metamathematics

Proposition 3.8. (a) If Aw is a universal R-set, then any fixed point

of (Aw0 , Bw0 ) = (Bw , Aw ) lies outside both Aw and Bw . (b) If Av is a
contra-universal R-set, then any fixed point of (Av , Bv ) lies outside both Av
and Bv .

Proposition 3.9. If A ⊆ Ah and Ah is disjoint from B, then some number

lies outside both Ah and Bh .

Proposition 3.10. There is no number n such that for all numbers x and
y, xy ∈ An ∪ Bn iff y ∈
/ Ax ∪ Bx .

R-Separable R-Sets

We shall say that an R-set S1 is R-separable from an R-set S2 if there

is some R-set S such that S1 is a subset of S and S2 is a subset of the
complement S of S.

Domination

We say that n dominates m when Am ⊆ An , i.e. when Am is a sub-

set of An .

Proposition 3.11. If Aw is universal and Av is contra-universal, then Aw

is not R-separable from Av . [This generalizes an important result in recur-
sion theory, namely that there exist two disjoint recursively enumerable
sets that are not recursively separable.]
Note. Proposition 3.11 can be rewritten in terms of domination, but
the rewritten version is very long. It is simpler just to say that the solution
to Problem 11 in the synthesis section (whose statement is expressed in
terms of the concept of domination) can be used to prove Proposition 3.11
(by contradiction), as the reader will be able to see in the solutions to the
problems of this chapter.

Proposition 3.12. If some R-set is universal, then there is an R-set whose

complement is not an R-set. [This generalizes the important fact that
there is a recursively enumerable set whose complement is not recursively
enumerable.]
 Chapter 3. Some Special Topics 79

IV. A Synthesis

The Synthesis System

As already mentioned, the discussions and problems of sections I, II and

III are really all the same, only dressed differently. In each of the three, we
have numerical relations A(x, y) and B(x, y) such that the following three
conditions hold:
D0 : A(x, y) and B(x, y) cannot both hold simultaneously for any ordered
pair of numbers (x, y).
D1 : Associated with each number x is a number x0 such that for all
numbers y:
(1) A(x0 , y) iff B(x, y);
(2) B(x0 , y) iff A(x, y).
D2 : Corresponding to each number x is a number x# such that for all y:
(1) A(x# , y) iff A(x, y ∗ y);
(2) B(x# , y) iff B(x, y ∗ y).
• For the decision machine M of I, A(x, y) is the relation “Rx affirms
y”, and B(x, y) is the relation “Rx denies y”.
• For the mathematical system S of II, A(x, y) is the relation “Hx (y)
is provable”, and B(x, y) is the relation “Hx (y) is refutable”.
• For the mathematical R-systems of III, A(x, y) is the relation
“y ∈ Ax ”, and B(x, y) is the relation “y ∈ Bx ”.
We shall call the pair A(x, y) and B(x, y) a duo if conditions D0 , D1
and D2 hold. Any result proved about duos in general is simultaneously
applicable to the items of sections I, II, and III. Thus, in the problems that
follow, for each n ≤ 12, the solution to Problem n simultaneously proves
Propositions 1n, 2n, and 3n of sections I, II, and III.
In this section we will again abbreviate x ∗ y by xy and we will omit the
word “section” in front of I, II, and III. We let A be the set of all numbers
xy such that A(x, y) is true, and we let B be the set of all numbers xy such
that B(x, y) is true. It will be valuable to know that not only are A(x, y)
and B(x, y) disjoint [for every pair (x, y)], but A and B are also disjoint
(contain no common element n). This is because every n is xy for some
80 Part II: Recursion Theory and Metamathematics

x and y, and it is given that A(x, y) and B(x, y) don’t both hold, which
means that xy cannot be in both A and B, and thus n cannot be in both
A and B.
• For the decision machine M of I, A is the set of all numbers xy such that
Rx affirms y, and B is the set of all numbers xy such that Rx denies y.
• For the mathematical system S of II, A is the set P0 (which is the set of
all numbers xy such that Hx (y) is provable), and B is the set R0 (the set
of all numbers xy such that Hx (y) is refutable).
• For the mathematical R-systems of III, A is the set of all numbers xy
such that y ∈ Ax , and B is the set of all numbers xy such that y ∈ Bx .

Undecidable Numbers

We shall call a number n undecidable if n is in neither A nor B. Thus xy

is undecidable if neither A(x, y) nor B(x, y) holds. If the number n is not
undecidable, we call it decidable.
• For the decision machine M of I, n = xy being undecidable means that
Rx is stumped by y.
• For the mathematical system S of II, n = xy being undecidable
means that the sentence Hx (y) is undecidable (i.e. neither provable nor
refutable).
• For the mathematical R-systems of III, n = xy being undecidable means
that y lies outside both Ax and Bx .
The following is an important fact: Since A is disjoint from B, it follows
that to prove that a number n is undecidable, it suffices to show that n ∈ A
iff n ∈ B, because this would mean that n is either in both A and B, or in
neither one. Since it cannot be in both, it must be in neither one, i.e. it is
undecidable.

Universal Numbers

We shall call a number w universal if for all x it is the case that wx ∈ A iff
x ∈ A, or, which is the same thing, for all x and y, the relation A(w, xy)
holds iff A(x, y) holds.
• For the decision machine M of I, to say that Rw is a universal register
is to say that for all x and y, Rw affirms xy iff Rx affirms y.
• For the mathematical system S of II, to say that Hw is a provability
predicate (a “universal predicate for provability”) is to say that for all x
 Chapter 3. Some Special Topics 81

and y, Hw (xy) is provable iff Sxy is provable (which is the same as saying
Hx (y) is provable).
• For the mathematical R-systems of III, to say that w is universal is to
say that Aw is a universal set, i.e. that xy ∈ Aw iff y ∈ Ax .

Contra-Universal Numbers

We call a number v contra-universal if vx ∈ A iff x ∈ B, or, what is the

same thing, for all x and y, A(v, xy) iff B(x, y).
• For the decision machine M of I, to say that Rv is a contra-universal
register is to say that for all x and y, Rv affirms xy iff Rx denies y.
• For the mathematical system S of II, to say that Hv is a refutability
predicate (a “contra-universal predicate”) is to say that for all x and
y, Hv (xy) is provable iff Sxy is refutable (which is the same as saying
that Hx (y) is refutable or that ∼H x (y) is provable).
• For the mathematical R-systems of III, to say that v is contra-universal
is to say that, for all x and y, xy ∈ Av iff y ∈ Bx .

Creative Numbers

We call a number c creative if for every number n there is at least one

number x such that cx ∈ A iff nx ∈ A.
• For the decision machine M of I, to say that a register Rc is creative is
to say that for every register Rn , there is at least one number x such
that Rc affirms x iff Rn affirms x.
• For the mathematical system S of II, to say the predicate Hc is creative
is to say that for every predicate Rn , there is at least one number x such
that Hc (x) is provable iff Hn (x) is provable.
• For the mathematical R-systems of III, to say that c is creative is to
say that, for every number n, there is at least one number x such that
x ∈ Ac iff x ∈ An .

Fixed Points

We let “x is a fixed point of n” mean that nx ∈ A iff x ∈ A and nx ∈ B iff

x ∈ B.
• For the decision machine M of I, to say that a number x ∗ y is a fixed
point of a register R is to say that R affirms x ∗ y iff x ∗ y ∈ A (i.e. Rx
affirms y) and R denies x ∗ y iff x ∗ y ∈ B (i.e. Rx denies y).
82 Part II: Recursion Theory and Metamathematics

• For the mathematical system S of II, to say that xy is a fixed point of

the predicate Hn is to say that Hn (xy) is equivalent to Sxy , i.e. both
these sentences are provable, or both are refutable, or both are neither.
[This is the same as saying that Hn (xy) is equivalent to Hx (y).]
• For the mathematical R-systems of III, to say that a number x is a fixed
point of a pair (An , Bn ) if x ∈ An iff x ∈ A and x ∈ Bn iff x ∈ B.

Problem 1. Prove that if w is universal, then there is at least one number

x such that wx is undecidable.

Problem 2. Prove that if some number is universal, then there is at least

one number x such that xx is undecidable.

Problem 3. Prove that if v is contra-universal, then there is at least one

number x such that vx is undecidable.

Problem 4. Prove that if some number is universal, then there is some

number x such that xx0 is undecidable. And prove that if some num-
ber is contra-universal, then there is some number x such that xx is
undecidable.

Problem 5. Prove that if c is creative, then there is at least one number

x such that cx is undecidable.

Problem 6. Prove that every universal number is creative, and that every
contra-universal number is creative.

Problem 7. Prove that every number has a fixed point.

Problem 8. Suppose w is universal and v is contra-universal. Prove:

(a) If x is any fixed point of w0 , then x is undecidable, and so is wx.
(b) If x is any fixed point of v, then x is undecidable, and so is vx.

Problem 9. Suppose h is a number such that, for all numbers x:

(a) If x ∈ A, then hx ∈ A.
(b) If x ∈ B, then hx ∈
/ A.
Prove that hx is undecidable for some x.
Why is this a strengthening of the result of Problem 1?

Problem 10. Prove that there is no number n such that for all numbers
x the number nx is decidable if and only if x is undecidable.
 Chapter 3. Some Special Topics 83

Domination

Let us say that a number n dominates a number m if, for every number x,
if mx ∈ A, then nx ∈ A.
• For the decision machine M of I, to say that the register Rn domi-
nates the register Rm is to say that Rn affirms all numbers affirmed
by Rm .
• For the mathematical system S of II, to say that the predicate Hn domi-
nates the predicate Hm is to say that that for all x, if Hm (x) is provable,
so is Hn (x).
• For the mathematical R-systems of III, to say that An dominates Am is
to say that Am ⊆ An .

Problem 11. Show that if n dominates some universal number w and

n0 dominates some contra-universal number v, then nx is undecidable for
some x.

Set Representation

We will say that the number n represents the set S if, for all x, nx ∈ A iff
x ∈ S.
• For the decision machines of I, to say that a register R represents a set
S is just to say that the set S is the affirmation set of the register R.
• For the mathematical systems S of II, to say that predicate H represents
a set S is just to say that S is the set of all numbers n such that H(n)
is provable.
• For the mathematical R-systems of III, to say that the R-set An repre-
sents a set S is just to say that y ∈ An iff y ∈ S, i.e., An = S.

Problem 12. Prove that if some number is universal, then there is a rep-
resentable set whose complement is not representable.

Solutions to the Problems of Chapter 3

1. Suppose w is universal. Then for any number x, w(w#0 x) ∈ A holds

iff w#0 x ∈ A, which is true iff w# x ∈ B, which is true iff w(xx) ∈ B.
Thus:
84 Part II: Recursion Theory and Metamathematics

(1) w(w#0 x) ∈ A iff w(xx) ∈ B.

If we take w#0 for x in (1), we obtain:

(2) w(w#0 w#0 ) ∈ A iff w(w#0 w#0 ) ∈ B.
But A and B are disjoint, so w(w#0 w#0 ) can be in neither A nor B,
# #
which shows that w(w#0 w#0 ) is undecidable; w(w0 w0 ) is also unde-
cidable, as the reader can verify.
2. Suppose w is universal. Then, by D1 , for any number x, w#0 x ∈ B
holds iff w# x ∈ A, which is true iff w(xx) ∈ A, which is true iff xx ∈ A.
Thus w#0 x ∈ B iff xx ∈ A. We take w#0 for x and obtain w#0 w#0 ∈ B
iff w#0 w#0 ∈ A. Hence w#0 w#0 is undecidable.
3. Suppose v is contra-universal. Then for any number x, v(v # x) ∈ A
is true iff v # x ∈ B, which is true iff v(xx) ∈ B. Thus v(v # x) ∈ A iff
v(xx) ∈ B. We take v # for x and obtain v(v # v # ) ∈ A iff v(v # v # ) ∈ B.
Thus v(v # v # ) is undecidable.
4. Suppose w is universal. We have seen in the solution to Problem 2
that w#0 w#0 is undecidable. Therefore w# w#0 is undecidable. [For any
numbers x and y, x0 y is undecidable iff xy is undecidable. (Why?)] Thus
nn0 is undecidable for n = w# .
Next, suppose that v is contra-universal. For any number x, we know
that v # x ∈ A iff v(xx) ∈ A. We take v # for x and obtain v # v # ∈ A
iff v(v # v # ) ∈ A, which in turn is the case iff v # v # ∈ B (since v is
contra-universal). Thus v # v # ∈ A iff v # v # ∈ B, so that v # v # must
be undecidable.
5. Suppose c is creative. Then for any x there is some n such that cn ∈ A
iff xn ∈ A. Taking c0 for x, we obtain that for some n, cn ∈ A iff
0
c0 n ∈ A. But c n ∈ A iff cn ∈ B. Thus cn ∈ A iff cn ∈ B, so that cn is
undecidable.
6. (a) Suppose w is universal. For any number x, w(x# x# ) ∈ A is true
iff x# x# ∈ A, which is true iff x(x# x# ) ∈ A. Thus wn ∈ A iff
xn ∈ A, for n = x# x# .
(b) Suppose v is contra-universal. For any x, v(x#0 x#0 ) ∈ A is true
iff x#0 x#0 ∈ B, which is true iff x# x#0 ∈ A, which is true iff
x(x#0 x#0 ) ∈ A . Thus vn ∈ A iff xn ∈ A, for n = x#0 x#0 .
7. For any number n, n# n# ∈ A holds iff n(n# n# ) ∈ A, and n# n# ∈ B
holds iff n(n# n# ) ∈ B. Hence n# n# is a fixed point of n.
 Chapter 3. Some Special Topics 85

8. (a) Suppose w is universal and n is a fixed point of w0 . Then n ∈ A

iff wn ∈ A is true (since w is universal), which is true iff w0 n ∈ B
(by D1 ), which is true iff n ∈ B (since n is a fixed point of w0 ).
Thus n ∈ A iff n ∈ B, so that n is undecidable. Since n is a fixed
point of w0 , then w0 n is also undecidable (why?), and hence wn is
undecidable.
(b) Suppose v is contra-universal and n is a fixed point of v. Then
n ∈ A iff vn ∈ A is true (since n is a fixed point of v). But vn ∈ A
iff n ∈ B (since v is contra-universal). Thus n ∈ A iff n ∈ B, and is
thus undecidable. Since n is a fixed point of v, it is also true that
vn ∈ B iff n ∈ B. Since n can be in neither A nor B, it follows
from what we just saw that vn can be in neither A nor B either,
and is thus also undecidable.
# #
Remarks. We know that w0 w0 is a fixed point of w0 (since by the
solution to Problem 7, x# x# is a fixed point of x, for any x). But it is
also true that w#0 w#0 is a fixed point of w0 (in fact x#0 x#0 is a fixed
point of x0 , for any x, as the reader can verify). Thus (a) alone provides
another proof that w(w#0 w#0 ) is undecidable.
9. We are given that for all x:
If x ∈ A, then hx ∈ A.
If x ∈ B, then hx ∈
/ A.
#0
Let k be the number h . We first show that kk is undecidable.
(a) Suppose kk ∈ A. Then h(kk) ∈ A [by (a)]. Thus h# k ∈ A so that
h#0 k ∈ B. And thus kk ∈ B [since k = h#0 ]. Thus if kk ∈ A it is
also true that kk ∈ B, but kk cannot be in both A and B, and so
kk ∈/ A.
(b) Suppose kk ∈ B. Then h(kk) ∈ / A [by (b)]. Hence h#0 k ∈
/ B, which
#0
means that kk ∈ / B [since k = h ]. Thus if kk ∈ B, then kk ∈ / B,
which is a contradiction. Therefore kk ∈
/ B.
Thus kk ∈/ A and kk ∈ / B, and so kk is undecidable. Thus h#0 h#0 is
undecidable, hence so is h# h#0 , and therefore so is h(h#0 h#0 ).
Note: The above proof is from scratch. A swifter proof is possible
using the fact that every number has a fixed point. We leave it to the
reader to show that if n is any fixed point of h, then hn is undecidable.
Next, why is this result a strengthening of the result of Problem 1?
The answer is that if h is universal, then it obviously satisfies conditions
(a) and (b).
86 Part II: Recursion Theory and Metamathematics

10. If a number n exists, it must have a fixed point, by Problem 7. But if

x is a fixed point of n, then (by the definition of a fixed point) nx ∈ A
iff x ∈ A and nx ∈ B iff x ∈ B. Clearly, for such an x, it cannot be
that nx if decidable iff x is undecidable.
11. This follows easily from Problem 9: Suppose h dominates w and h0
dominates v, where w is universal and v is contra-universal.
If x ∈ A, then wx ∈ A, so that hx ∈ A [since h dominates w]. Thus,
(1) If x ∈ A, then hx ∈ A.
Next, suppose x ∈ B. Then vx ∈ A, so that h0 x ∈ A, and then
hx ∈ B, and finally hx ∈
/ A. Thus,
(2) If x ∈ B, then hx ∈
/ A.
Thus, by (1) and (2), h satisfies the hypotheses of Problem 9, and so
hx is undecidable for some x.
Because the relationship between the solution to Problem 11 and
Propositions 11 and 35 are a little more complex than some of the
others, we will give the proofs of these propositions (using the solution
to Problem 11) here. First let us review what “n dominates m” means
for the first three sections of this chapter:

I: Rn affirms all numbers affirmed by Rm .

II: For all x, if Hm (x) is provable, so is Hn (x).
III: Am ⊆ An .

First let us consider Proposition 11: Suppose U is universal and V

is contra-universal and that R affirms all numbers affirmed by U and
denies all numbers affirmed by V . Then R can be stumped.
In the text we gave a restatement of this claim in terms of domination:
Suppose U is universal and V is contra-universal and there is a
register R such that R dominates U and R0 dominates V . Then
there exists an x such that R is stumped by x.
It is easy to see that the restatement is equivalent to the original
statement, for to say that R denies all numbers affirmed by V is equiv-
alent to saying that R0 affirms all numbers affirmed by V . To bring this
restatement closer to the expression of Problem 11 in the synthesis, all
we have to do is the following: We let n be the index of R, w be the
index of U , and v be the index of V ; then the restatement of Proposi-
tion 11 is equivalent to: If Rn dominates the universal register Rw and
Rn0 dominates the contra-universal register Rv , then the register Rn
 Chapter 3. Some Special Topics 87

can be stumped by some x. And this is the statement of Problem 11,

reinterpreted in the vocabulary of the concepts of I.
The application to Proposition 23 is obvious.
Finally, let us consider Proposition 35: If Aw is universal and Av is
contra-universal, then Aw is not R-separable from Av .
To show that Aw is not R-separable from Av , suppose that S1 and S2
are disjoint R-sets such that Aw ⊆ S 1 and Av ⊆ S 2 . We must show that
some number lies outside of both S1 and S2 . Now, for some number n,
S1 = An and S2 = Bn . Then S1 = An and S2 = An0 . So Aw ⊆ An and
Av ⊆ An0 , which means that n dominates w and n0 dominates v, and
so, by the solution to Problem 11, there is some number x such that
nx is undecidable, i.e x ∈ / An and x ∈/ B n . Thus x ∈
/ An and x ∈
/ An0 ,
i.e. x ∈
/ S1 and x ∈/ S 2 , which is what was to be shown.
12. We first show that no number x can represent the complement A of A.
To show this it suffices to demonstrate that, given any number x there
is at least one number y such that it is not the case that xy ∈ A iff
y ∈ A, or, what is the same thing, that there is at least one number y
such that it is the case that xy ∈ A iff y ∈ A. Well, any fixed point y
of x is such a number. Thus the set A is not representable.
Now, if there is a universal number w, then w represents A (why?),
so that A is then a representable set whose complement A is not
representable.
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 4

Elementary Formal Systems

and Recursive Enumerability

I. More on Elementary Formal Systems

Let us review the definition of an elementary formal system and some of

the basic properties of such systems established in The Beginner’s Guide
[Smullyan, 2014].
By an alphabet K is meant a finite sequence of elements called symbols.
Any finite sequence of these symbols is called a string or expression in K.
By an elementary formal system (E) over K is meant a collection of the
following items:
(1) The alphabet K.
(2) Another alphabet V of symbols called variables. We will usually use
the letters x, y, z, w with or without subscripts, as variables.
(3) Still another alphabet of symbols called predicates, each of which is
assigned a positive integer called the degree of the predicate. We usually
use capital letters for predicates.
(4) Two more symbols called the punctuation sign (usually a comma) and
the implication sign (usually “→”).
(5) A finite sequence of expressions, all of which are formulas (formulas
will be defined below).
By a term is meant any string composed of symbols of K and variables.
A term without variables is called a constant term. By an atomic formula
90 Part II: Recursion Theory and Metamathematics

is meant an expression P t, where P is a predicate of degree 1 and t is a

term, or an expression Rt1 , . . . , tn , where R is a predicate of degree n and
t1 , . . . , tn are terms.
By a formula is meant an atomic formula, or an expression of the form
F1 → F2 . . . → Fn , where F1 , F2 , . . . , Fn are atomic formulas. This con-
cludes the definition of a formula.
By a sentence is meant a formula with no variables.
By an instance of a formula is meant the result of substituting constant
terms for all variables in the formula.
An elementary formal system (E) over K is thus formed by distinguish-
ing a particular finite subset of the formulas, which are called the axiom
schemes of the system. The set of all instances of all the axiom schemes is
called the set of axioms of the system.
A sentence is called provable in the elementary formal system (E) if its
being so is a consequence of the following two conditions:
(1) Every axiom of the system is provable.
(2) For every atomic sentence X and every sentence Y , if X and X → Y
are provable, so is Y.
More, explicitly, by a proof in the system is meant a finite sequence of
sentences such that each member Y of the sequence is either an axiom of
the system, or there is an atomic sentence X such that X and X → Y
are earlier members of the sequence. A sentence is then called provable
in the system if it is an element in the sequence of some proof in the
system.
Elementary formal systems are designed to formalize the nature of a
mechanical procedure, and the provable sentences of a given elementary
formal system that have the form of some predicate P of degree n fol-
lowed by an n-tuple of elements of the alphabet K are just those that
can be generated by the given elementary formal system (which must
of course include the predicate P in at least one of its axiom schemes
if such a sentence is to be provable). And when such a term P k1 . . . kn
can be generated by the system, we say that it is true that the n-tuple
k1 , . . . , kn stands in the relation P , or we may say that P k1 . . . kn is true.
Thus we identify provability and truth when it comes to elementary formal
systems.
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 91

Representability

We say that a predicate P of degree n represents a relation R of degree n

of n-tuples of strings of K if for all constant terms x1 , . . . , xn , the sentence
P x1 , . . . , xn is provable in (E) iff the relation Rx1 , . . . , xn holds. For n = 1,
a predicate P of degree (1) represents the set of all constant terms x such
that P x is provable in (E). A set or relation is said to be representable in
(E) if some predicate represents it, and a relation or set is called formally
representable if it is representable in some elementary formal system.

Some Basic Closure Properties

For any n ≥ 2 we regard a relation of degree n on a set S as being a
set of n-tuples (x1 , . . . , xn ) of members of S. We consider subsets of S to
be special cases of relations on S, namely relations of one argument. We
now wish to establish some basic closure properties of the collection of all
relations (and sets) representable over K.
First, let us review some more facts discussed in The Beginner’s Guide
[Smullyan, 2014]. Consider two elementary formal systems (E1 ) and (E2 )
over a common alphabet K. We shall call the two systems independent if
they contain no common predicates. By (E1 ) ∪ (E2 ) we shall mean the EFS
(elementary formal system) whose axioms are those of (E1 ) together with
those of (E2 ). Obviously every provable sentence of (E1 ) and of (E2 ) is
provable in (E1 ) ∪ (E2 ). And if (E1 ) and (E2 ) are independent, then any
sentence provable in (E1 ) ∪ (E2 ) is clearly provable in (E1 ) or (E2 ) alone,
and hence the representable relations of (E1 ) ∪ (E2 ) are the representable
relations in (E1 ) alone, together with those representable in (E2 ) alone.
Similarly, if (E1 ), (E2 ), . . . , (En ) are mutually independent (no two
having a common predicate), then the representable relations of
(E1 ) ∪ . . . ∪ (En ), are the result of joining together those of each of
(E1 ), . . . , (En ) alone.
Now suppose R1 , . . . , Rn are each formally representable over K. Since
we are assuming an unlimited stack of predicates at our disposal, we can
represent R1 , . . . , Rn in mutually independent systems (E1 ), (E2 ), . . . , (En ),
and so we have:

Proposition 1. If R1 , . . . , Rn are each formally representable over K, then

they can all be represented in a common EFS over K.
92 Part II: Recursion Theory and Metamathematics

Now for our closure properties:

(a) Unions and Intersections. For any two relations R1 (x1 , . . . , xn ),
R2 (x1 , . . . , xn ) of the same degree, by their union R1 ∪ R2 — also
written R1 (x1 , . . . , xn ) ∨ R2 (x1 , . . . , xn ) — is meant the set of all
n-tuples (x1 , . . . , xn ) that are in either R1 or R2 or in both. Thus
(R1 ∪ R2 )(x1 , . . . , xn ) iff R1 (x1 , . . . , xn ) ∨ R2 (x1 , . . . , xn ). By the inter-
section R1 ∩ R2 — also written R1 (x1 , . . . , xn ) ∧ R2 (x1 , . . . , xn ) — is
meant the relation that holds for (x1 , . . . , xn ) iff
R1 (x1 , . . . , xn ) ∧ R2 (x1 , . . . , xn ) :
(b) Existential Quantifiers. For any relation R(x1 , . . . , xn , y) of degree two
or higher, by its existential quantification ∃ yR(x1 , . . . , xn , y) — also
abbreviated ∃R — is meant the set of all n-tuples (x1 , . . . , xn ) such
that R(x1 , . . . , xn , y) holds for at least one y:
(c) Explicit Transformation. Let n be any positive integer and consider n
variables x1 , . . . , xn . Let R be a relation of degree k and let α1 , . . . , αk
each be either one of the n variables x1 , . . . , xn or a constant term (i.e.
a string of symbols of K). By the relation λx1 , . . . , xn R(α1 , . . . , αk )
is meant the set of all n-tuples (a1 , . . . , an ) of strings of K such that
R(b1 , . . . , bk ) holds, where each bi is defined as follows:
(1) If αi is one of the variables xj , then bi = aj .
(2) If αi is a constant c, then bi = c.
[For example, for n = 3 and constants c and d, λx1 , x2 , x3 R(x3 , c, x2 , x2 , d)
is the set of all triples (a1 , a2 , a3 ) such that the relation R(a3 , c, a2 , a2 , d)
holds.]
In this situation, we say that the n-ary relation
λx1 , . . . , xn R(α1 , . . . , αk )

is explicitly definable from the k-ary relation R and that the operation which
assigns λx1 , . . . , xn R(α1 , . . . , αk ) to the relation R is an explicit transfor-
mation.

Problem 1. Show that the collection of relations formally representable

over K is closed under unions, intersections, existential quantification and
explicit transformations. In other words, prove the following:
(a) If R1 and R2 are of the same degree and formally representable over
K, then R1 ∪ R2 and R1 ∩ R2 are formally representable over K and
of the same degree as R1 and R2 .
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 93

(b) If the relation R(x1 , . . . , xn , y) is a relation of degree n + 1(n greater

than or equal to one) that is formally representable over K, then the
relation ∃ yR(x1 , . . . , xn , y) is a relation of degree n that is formally
representable over K.
(c) If the relation R(x1 , . . . , xk ) is a relation of degree k that is formally
representable over K, then λx1 , . . . , xn R(α1 , . . . , αk ) is a relation of
degree n that is formally representable over K (here each αi is one of
the variables x1 , . . . , xn or a constant).
We recall that a set or relation W is said to be solvable over K if W
and its complement W are both formally representable over K.
Problem 2. Prove that the collection of all sets and relations which are
solvable over K is closed under union, intersection, complementation and
explicit transformation.

II. Recursive Enumerability

We shall now identify the positive integers with the dyadic numerals which
represent them (as defined in The Beginner’s Guide [Smullyan, 2014]). We
consider the two-sign alphabet {1, 2}, which we call “D”, and define a
dyadic system to be an elementary formal system over the alphabet D. We
define a numerical relation to be recursively enumerable if it is representable
in some dyadic system. (In The Beginner’s Guide we called these relations
dyadically representable.) It turns out that the recursively enumerable sets
and relations are the same as the Σ1 -relations — and also the same as those
representable in Peano Arithmetic.
We call a set R recursive if it and its complement R are both recursively
enumerable. Thus a set is recursive if it is solvable in some dyadic system.
In The Beginner’s Guide, we showed that the relations Sxy (y is the
successor of x), x < y, x ≤ y, x = y, x 6= y, x + y = z, x × y = z and xy = z
are all recursively enumerable (in fact, recursive).
We also know (taking D as our alphabet K) that the collection of recur-
sively enumerable relations is closed under unions, intersections, explicit
transformations, and existential quantifications. We now need some further
closure properties.
94 Part II: Recursion Theory and Metamathematics

Finite Quantifications (Bounded Quantifiers)

For any numerical relation R(x1 , . . . , xn , y), by the relation

(∀ z ≤ y)R(x1 , . . . , xn , z)

[sometimes abbreviated by ∀F R and called the finite universal quan-

tification of R] is meant the set of all (n + 1)-tuples (x1 , . . . , xn , y)
such that R(x1 , . . . , xn , z) holds for every z that is less than or
equal to y. Thus (∀z ≤ y)R(x1 , . . . , xn , z) holds iff R(x1 , . . . , xn , 1),
R(x1 , . . . , xn , 2), . . . , R(x1 , . . . , xn , y) all hold. (∀ z ≤ y)R(x1 , . . . , xn , z) is
equivalent to the formula
∀z(z ≤ y ⊃ R(x1 , . . . , xn , z)).

By the relation (∃z ≤ y)R(x1 , . . . , xn , z) [sometimes abbreviated by

∃F R and called the finite existential quantification of R] is meant the set of
all (n + 1)-tuples (x1 , . . . , xn , y) such that there is at least one z ≤ y such
that R(x1 , . . . , xn , z) holds. Thus (∃z ≤ y)R(x1 , . . . , xn , z) is the disjunction
of the relations R(x1 , . . . , xn , 1), R(x1 , . . . , xn , 2), . . . , R(x1 , . . . , xn , y). The
formula (∃z ≤ y)R(x1 , . . . , xn , z) is equivalent to the formula
∃z(z ≤ y ∧ R(x1 , . . . , xn , z)).

The quantifiers (∀x ≤ c) and (∃x ≤ c) are called bounded quantifiers.

Problem 3. Show the following:
(a) If R(x1 , . . . , xn , y) is recursively enumerable, so are ∀F R and ∃F R.
(b) If R(x1 , . . . , xn , y) is recursive, so are ∀F R and ∃F R.

Before continuing, let us recall the definitions of Σ0 and Σ1 formulas

of elementary arithmetic. In The Beginner’s Guide [Smullyan, 2014] we
defined a Σ0 formula and a Σ0 relation as follows: By an atomic Σ0 formula
we mean any formula of one of the forms c1 = c2 or c1 ≤ c2 , where c1 and
c2 are both terms. We then define the class of Σ0 formulas of Elementary
Arithmetic by the following inductive scheme:
(1) Every atomic Σ0 formula is a Σ0 formula.
(2) For any Σ0 formulas F and G, the formulas ∼ F, F ∧ G, F ∨ G, and
F ⊃ G are Σ0 formulas.
(3) For any Σ0 formula F , and any variable x, and any c which is either a
Peano numeral or a variable distinct from x, the expressions (∀x ≤ c)F
and (∃x ≤ c)F are Σ0 formulas.
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 95

No formula is a Σ0 formula unless its being so is a consequence of (1),

(2) and (3) above.
Thus a Σ0 formula is a formula of Elementary Arithmetic in which all
quantifiers are bounded. A set or relation is called a Σ0 set or relation (or
just “Σ0 ”) if it is expressed by a Σ0 formula.
A relation is called Σ1 if it is expressed by a formula of the form
∃ zR(x1 , . . . , xn , z), where the relation R(x1 , . . . , xn , z) is Σ0 .
We noted in The Beginner’s Guide [Smullyan, 2014] that all Σ0 and Σ1
relations and sets are arithmetic, because the formulas that express them
are just a subset of all the formulas of Elementary Arithmetic that can be
used to express number sets and relations. Moreover, we discussed the fact
that, given any Σ0 sentence (a Σ0 formula with no free variables), we can
effectively decide whether it is true or false.
By Problem 2, taking D for the alphabet K, the collection of all recur-
sive sets and relations is closed under unions, intersections, complements,
and explicit transformations. It is also closed under finite universal and
existential quantifications, as we have just seen. This collection also con-
tains the relations x + y = z and x × y = z. From all these facts, we have
the following proposition:

Proposition 2. Every Σ0 -relation is recursive.

We also know by Problem 1, taking D for the alphabet K, that the
existential quantification of a recursively enumerable relation is recursively
enumerable. Now, every Σ1 -relation is the existential quantification of a
Σ0 -relation, hence of a recursively enumerable (in fact recursive) relation,
and is therefore recursively enumerable. This proves that:

Proposition 3. All Σ1 -relations are recursively enumerable.

We now wish to prove that, conversely, all recursively enumerable rela-
tions are Σ1 , and hence that being recursively enumerable is the same as
being Σ1 !
First, for some very useful facts about Σ1 -relations:

Problem 4. Suppose R(x1 , . . . , xn , y) is Σ1 . Show that the relation

∃yR(x1 , . . . , xn , y) [as a relation between x1 , . . . , xn ] is Σ1 .
For any function f (x) (from numbers to numbers), we say that the
function f (x) is Σ1 if the relation f (x) = y is Σ1 .
96 Part II: Recursion Theory and Metamathematics

Proposition 4. If A is a Σ1 set and f (x) is a Σ1 -function, and A0 is the

set of all x such that f (x) ∈ A, then A0 is Σ1 . [A0 is what is called the
domain of the function f (x) when the range of the function is restricted to
the set A.]

Problem 5. Prove the above proposition.

To prove that every recursively enumerable relation is Σ1 , we will illus-
trate the proof for recursively enumerable sets. We will show that every
recursively enumerable set A is Σ1 . The proof for relations is a modifica-
tion we leave to the reader.
We use the dyadic Gödel numbering of The Beginner’s Guide [Smullyan,
2014]: For any ordered alphabet K, viz. {k1 , k2 , . . . , kn }, we assign the Gödel
number 12 to k1 , 122 to k2 , and a 1 followed by n 2’s to kn . And then, for
any compound expression, we take its Gödel number to be the result of
replacing each symbol of K by its Gödel number. We use g to denote the
Gödel number function, e.g. the Gödel number of g(k3 k1 k2 ) = 122212122.
In The Beginner’s Guide, we showed that if a set S of strings of symbols
of K is formally representable over K, then the set S0 of Gödel numbers of
the members of S is Σ1 . Of course this also holds when K is the two-sign
alphabet {1, 2}. Thus if A is recursively enumerable, then the set A0 of the
Gödel numbers of the members of A (in dyadic notation) is Σ1 . Our job
now is to get from the fact that a set of Gödel numbers A0 is Σ1 to the
fact that A itself is Σ1 . Now, A is the set of all x such that g(x) ∈ A0 , and
so [by Proposition 4 above] it suffices to show that the relation g(x) = y
is Σ1 . To this end, we use the following lemma, which was proved in The
Beginner’s Guide:

Lemma K. There exists a Σ0 relation K(x, y, z) such that the following

two conditions hold:
(1) For any finite set (a1 , b1 ), . . . , (an , bn ) of ordered pairs of numbers
(positive integers), there is a number z (called a code number of the
set) such that for all x and y, K(x, y, z) holds if and only if (x, y) is
one of the pairs (a1 , b1 ), . . . , (an , bn ).
(2) For all x, y, z if K(x, y, z) holds, then x < z and y < z

Problem 6. Using Lemma K, prove the relation g(x) = y is Σ1 .

Extending this results to all relations gives:
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 97

Proposition 5. Every recursively enumerable relation is Σ1 and every Σ1

relation is recursively enumerable.

A Recursive Pairing Function δ(x, y)

A function f which assigns to each pair (x, y) of numbers a number f (x, y)

is said to be 1-1 (which is read as “one-to-one”) if for every number z
there is at most one pair (x, y) such that f (x, y) = z. In other words, if
f (x, y) = f (z, w), then x = z and y = w. The function f is said to be
onto the set N of numbers, if every number z is f (x, y) for some pair (x, y).
Thus f is 1-1 onto N iff every number z is f (x, y) for one and only one
pair (x, y).
We now want a recursive function δ(x, y) which is 1-1 onto N . [Note
that such a function provides an enumeration of all ordered pairs of pos-
itive integers.] There are many ways to construct such a function. Here
is one way: We start with (1, 1) followed by the three pairs whose high-
est number is 2, in the order (1, 2), (2, 2), (2, 1), followed by the five pairs
whose highest number is 3, in the order (1, 3), (2, 3), (3, 3), (3, 1), (3, 2), fol-
lowed by . . . (1, n), (2, n), . . . , (n, n), (n, 1), (n, 2), . . . , (n, n − 1), etc. We
then take δ(x, y) to be the position that (x, y) takes in this order-
ing, which is the number of pairs which precede (x, y) plus 1. For
example,

δ(1, 1) = 1, δ(1, 2) = 2, δ(2, 2) = 3, δ(2, 1) = 4, δ(1, 3) = 5,

δ(2, 3) = 6, δ(3, 3) = 7, δ(3, 1) = 8, δ(3, 2) = 9, δ(1, 4) = 10.

In fact, the following hold:

2
(a) If m ≤ n, then δ(m, n) = (n − 1) + m.
2
(b) If n < m, then δ(m, n) = (n − 1) + m + n.

To see that (a) is true, we first note that for any n, the pairs that
preceded (1, n) in the ordering are those in which x and y are both less
2
than n, and there are (n − 1) such pairs [since there are n − 1 numbers
2
(positive integers) that are less than n]. Thus δ(1, n) = (n − 1) + 1. Hence
2 2 2
δ(2, n) = (n − 1) + 2, δ(3, n) = (n − 1) + 3, . . . , δ(n, n) = (n − 1) + n.
2
Thus, for any m ≤ n, δ(m, n) = (n − 1) + m. This proves (a).
98 Part II: Recursion Theory and Metamathematics

2 2
As for (b), since δ(m, m) = (m − 1) +m, then δ(m, 1) = (m − 1) +m+1;
2 2
δ(m, 2) = (m − 1) + m + 2, . . . , δ(m, m − 1) = (m − 1) + m + (m − 1).
2
Thus for any n ≤ m, δ(m, n) = (m − 1) + m + n.
Problem 7. Prove that the function δ(x, y) is recursive, in fact Σ0 .

The Inverse Functions K and L

Each number z is δ(x, y) for just one pair (x, y). We let K(z) [to often be
abbreviated Kz] be x, and let L(z) [or Lz] be y. Thus Kδ(x, y) = x and
Lδ(x, y) = y.

Problem 8. (a) Prove that δ(Kz, Lz) = z. (b) Prove that the functions
K(x) and L(x) are Σ0 , and hence recursive.

The n-tupling Functions δn (x1 , . . . , xn )

For each n ≥ 2 we define the function δn (x1 , . . . , xn ) by the following induc-

tive scheme:

δ2 (x1 , x2 ) = δ(x1 , x2 ),
δ3 (x1 , x2 , x3 ) = δ(δ2 (x1 , x2 ), x3 ),
..
.
δn+1 (x1 , . . . , xn+1 ) = δ(δn (x1 , . . . , xn ), xn+1 ).

An obvious induction argument shows that for each n ≥ 2 the func-

tion δn (x1 , . . . , xn ) is a 1-1 recursive function onto the set N of positive
integers.

Problem 9. Prove the following:

(a) For any recursively enumerable set A and n ≥ 2, the relation
δn (x1 , . . . , xn ) ∈ A is recursively enumerable.
(b) For any recursively enumerable relation R(x1 , . . . , xn ), the set A
of all numbers δn (x1 , . . . , xn ) such that R(x1 , . . . , xn ) is recursively
enumerable.
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 99

The Inverse Functions Kin

For each n ≥ 2 and 1 ≤ i ≤ n, we define Kin (x) as follows:

x = δn (x1 , . . . , xn ) for just one n-tuple (x1 , . . . , xn ), and we take Kin (x)
to be xi . Thus Kin (δn (x1 , . . . , xn )) = xi .

Problem 10. Prove the following:

(a) δn (K1n (x), K2n (x), . . . , Knn (x)) = x.
(b) K12 (x) = Kx and K22 (x) = Lx.
(c) If x = δn+1 (x1 , . . . , xn+1 ), then Kx = δn (x1 , . . . , xn ) and Lx = xn+1 .
(d) If i ≤ n then Kin+1 (x) = Kin (Kx). If i = n + 1, then Kin+1 (x) = Lx.
(e) For any recursively enumerable relation R(x1 , . . . , xn , y1 , . . . , ym ),
there is a recursively enumerable relation M (x, y) such that
for all x1 , . . . , xn , y1 , . . . , ym , we have R(x1 , . . . , xn , y1 , . . . , ym ) iff
M (δn (x1 , . . . , xn ), δm (y1 , . . . , ym )).

III. A Universal System

We now wish to construct a “universal” system (U ), in which we can express

all propositions of the form that such and such a number is in such and
such a recursively enumerable number set, or that such and such an n-tuple
of numbers stands in such and such a recursively enumerable relation on
n-tuples of numbers. (We recall that we are using the word “number” to
mean “positive integer”.)
In preparation for the construction of the system (U ), we need to “tran-
scribe” all dyadic systems (elementary formal systems over the alphabet
{1, 2}) into a single finite alphabet. We now define a transcribed dyadic
system — to be abbreviated “TDS” — to be a system like a dyadic system,
except that instead of taking our variables and predicates to be individ-
ual symbols, we take two signs, v and the “accent sign” 0 and we define a
transcribed variable — to be abbreviated “TS variable” — to be any string
vαv, where α is a string of accents. Similarly, we define a transcribed pred-
icate — to be abbreviated “TS predicate” — to be a string of p’s followed
by a string of accents; the number of p’s is to indicate the degree of the
predicate. A transcribed dyadic system is then like a dyadic system, except
100 Part II: Recursion Theory and Metamathematics

that we use transcribed variables instead of single symbols for variables,

and transcribed predicates instead of individual symbols for predicates. We
thus have a single alphabet K7 , namely the symbols from which all TS
variables and predicates, and eventually formulas, are constructed:
0
v p 1 2 , →

Note that the symbols 1 and 2, from which we will construct our dyadic
numerals for the positive integers, correspond to the alphabet K in the
definition of an elementary formal system given in The Beginner’s Guide
[Smullyan, 2014] and earlier here.
It is obvious that representability in an arbitrary dyadic system is equiv-
alent to representability in a transcribed dyadic system.
We define “TS term”, “atomic TS formula”, “TS formula”, “TS sen-
tence” as we did “term”, “atomic formula”, “formula”, “sentence”, only
using TS variables and TS predicates instead of variables and predicates,
respectively.
We now construct the universal system (U ) in the following manner:
We first extend the alphabet K7 to the alphabet K8 by adding the symbol
“∧”. Then, by a base we mean a string of the form X1 , . . . , Xk (or a single
formula X, if k = 1). Here X and X1 , . . . , Xk are all TS formulas. These
TS formulas are called the components of the base X1 ∧ . . . ∧ Xk , or of the
base X. Then we extend the alphabet K8 to the alphabet K9 by adding
the symbol “`”, and we define a sentence of (U ) to be an expression of the
form B ` X, where B is a base and X is a TS sentence (a TS formula with
no variables). We call the sentence B ` X true if X is provable in that TDS
whose axiom schemes are the components of B. Thus X1 ∧ . . . ∧ Xk ` X is
true iff X is provable in that TDS whose axiom schemes are X1 , . . . , Xk .
By a predicate H of (U ) [not to be confused with a TS predicate], we
mean an expression of the form B ` P , where B is a base and P is a
transcribed predicate. We define the degree of B ` P to be the degree of P
(i.e. the number of occurrences of p in P ). A predicate H of the form B ` P
of (U ) of degree n is said to represent the set of all n-tuples (a1 , . . . , an ) of
numbers such that the sentence B ` P a1 , . . . , an is true. Thus a predicate
X1 ∧ . . . ∧ Xk ` P of degree n represents (in U ) the set of all n-tuples
(a1 , . . . , an ) such that P a1 , . . . , an is provable in that TDS whose axiom
schemes are X1 , . . . , Xk ; thus it represents in (U ) the same relation as P
represents in the TDS whose axiom schemes are X1 , . . . , Xk . Thus a relation
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 101

or set is representable in (U ) if and only if it is recursively enumerable. In

this sense, (U ) is called a universal system for all recursively enumerable
relations.
We shall let T be the set of true sentences of (U ), and let T0 be the set
of Gödel numbers of the members of T (using the dyadic Gödel numbering).
We now wish to show that the set T is formally representable in an
elementary formal system over the alphabet K9 , and that the set T0 is
recursively enumerable. Our entire development of recursion theory is based
on this crucial fact!
We are to construct an elementary formal system W over K9 in which
T is represented. To do this, we must first note that the implication sign
of W must be different from the implication sign of transcribed systems.
We will use “→” for the implication sign of W, and use “imp” for the
implication sign of transcribed systems. Similarly, we will use the ordinary
comma for the punctuation sign of W and use “com” for the punctuation
sign of transcribed systems. For variables of W, (not to be confused with
transcribed variables), we will use the letters x, y, z, w, with or without sub-
scripts. Predicates (not to be confused with predicates of (U ) or transcribed
predicates) are to be introduced as needed.
Problem 11. Construct W by successively adding predicates and axiom
schemes to represent the following properties and relations:
N (x) : x is a number (dyadic numeral)

Acc(x) : x is a string of accents

V ar(x) : x is a transcribed variable

dv (xy) : x and y are distinct TS variables

P (x) : x is a transcribed predicate

t(x) : x is a transcribed term

F0 (x) : x is a transcribed atomic formula

F (x) : x is a transcribed formula

S0 (x) : x is a transcribed atomic sentence

102 Part II: Recursion Theory and Metamathematics

Sub(x, y, z, w) : x is a string compounded from TS variables, numerals, TS

predicates, com and imp, and y is a TS variable, and w is the result of
substituting z for every occurrence of y in x

pt(x, y) : x is a TS formula and y is the result of substituting numerals for

some (but not necessarily all) TS variable of x (such a formula y is called
a partial instance of x)

in(x, y) : x is a TS formula and y is the result of substituting numerals for

all variables of x (such a y is an instance of x)

B(x) : x is a base

C(x, y) : x is a component of the base y

T (x) : x is a true sentence of (U )

Since T0 is formally representable, then T0 is Σ1 , hence recursively
enumerable.

The Recursive Unsolvability of (U )

We now wish to show that the complement T0 of T0 is not recursively

enumerable, and thus that the set T0 is recursively enumerable, but not
recursive.
For any set A of numbers, define a sentence X of (U ) to be a Gödel
sentence for A if either X is true and its Gödel number X0 is in A, or X is
false and its Gödel number is not in A. In other words, a sentence X of (U )
is a Gödel sentence for A if X is true iff X0 ∈ A. There obviously cannot be
a Gödel sentence for T 0 , for such a sentence would be true iff it were not
true (i.e. iff its Gödel number was not in T0 ), which is impossible. Thus, to
show that T0 is not recursively enumerable, it suffices to show that for any
recursively enumerable set, there is a Gödel sentence.
To this end, for any string X of symbols of K7 , we define its norm
to be XX0 , where X0 is the Gödel number of X. Any dyadic numeral n
itself has a Gödel number n0 , hence a norm nn0 . We recall that our dyadic
Gödel numbering has the advantage of being an isomorphism with respect
to concatenation, i.e. if n is the Gödel number of X and m is the Gödel
number of Y , then nm is the Gödel number of XY . Therefore, if n is the
Gödel number of X, then nn0 is the Gödel number of Xn (the norm of X).
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 103

Thus the Gödel number of the norm of X is the norm of the Gödel number
of X.
For any number x, we let normx be the norm of x.
Problem 12. Show that the relation normx = y is recursively enu-
merable.
For any set A of numbers we let A# be the set of all numbers whose
norm is in A.

Lemma. If A is recursively enumerable, so is A# .

Problem 13. Prove the above lemma.

Problem 14. Using the above lemma, now prove that if A is recursively
enumerable, then there is a Gödel sentence for A.
As we have seen, there cannot be a Gödel sentence for T0 , and thus by
Problem 14, the set T0 cannot be recursively enumerable. And so the set
T0 is recursively enumerable, but not recursive.

Solutions to the Problems of Chapter 4

1. (a) We have already proved this for when R1 and R2 are sets (i.e.
relations of degree 1), but the proof for relations of degree n is
essentially the same: Suppose R1 and R2 are relations of degree n
and that both are formally representable over K. There they are
representable in a common elementary formal system over K. In
this system let P1 represent R1 and let P2 represent R2 . Then take
a new predicate Q and add the following two axioms:

P1 x1 , . . . , xn → Qx1 , . . . , xn
P2 x1 , . . . , xn → Qx1 , . . . , xn .

Then Q represents R1 ∪R2 .

To represent R1 ∩ R2 , instead of the above two axioms, add the

following single axiom:

P1 x1 , . . . , xn → P2 x1 , . . . , xn → Qx1 , . . . , xn .

Then Q will represent R1 ∩R2 .

104 Part II: Recursion Theory and Metamathematics

(b) Let P represent the relation R(x1 , . . . , xn , y) in (E). Take a new

predicate Q of degree n and add the following axiom scheme:
P x1 , . . . , xn , y → Qx1 , . . . , xn .
Then Q represents the relation
∃ yR(x1 , . . . , xn , y) in (E).
(c) Let P represent the relation R(x1 , . . . , xk ) in (E). Take a new pred-
icate Q and add the following axiom scheme:
P α1 , . . . , αk → Qx1 , . . . , xn .
Then Q represents the relation
λx1 , . . . , xn R(α1 , . . . , αk ) in (E).
2. Complementation is trivial. As for unions and intersections, suppose
that W1 and W2 are of the same degree and are both solvable over K.
Then W1 , W2 , W1 and W2 are all formally representable over K. We
already know from Problem 1 that W1 ∪ W2 and W1 ∩ W2 are for-
mally representable over K. It remains to show that the complements
of W1 ∪ W2 and W1 ∩ W2 are formally representable over K. But these
complements are W1 ∪ W2 and W1 ∩ W2 , which we can use our knowl-
edge of Boolean sets to see to be equal to W1 ∩ W2 and W1 ∪ W2 ,
respectively. Since these latter sets are the intersections and unions of
sets we know to be formally representable over K, we now see that
W1 ∪ W2 and W1 ∩ W2 are formally representable over K, and thus
that the union or intersection of two sets that are solvable over K is
solvable over K.
Now assume that the R(x1 , . . . , xk ) is solvable over K. Then its
complement, which we will denote by R(x1 , . . . , xk ) is also solv-
able over K, since the two are complements of each other. We
have seen that the explicit transformations λx1 , . . . , xn R(α1 , . . . , αk )
and λx1 , . . . , xn R(α1 , . . . , αk ), where each αi is one of the variables
x1 , . . . , xn or a constant, are also formally representable over K. But
each of these is the complement of the other, so we see that the col-
lection of sets and relations solvable over K is closed under explicit
transformation as well.
3. (a) Suppose that the relation R(x1 , . . . , xn , y) is recursively enumer-
able. Let D be a dyadic system in which R is represented by the
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 105

predicate P , x < y is represented by E, and the relation “the suc-

cessor of x is y” is represented by S.
Let W (x1 , . . . , xn , y) be the relation (∀z ≤ y)R(x1 , . . . , xn , z).
To represent the relation W , take a new predicate Q and add the
following two axiom schemes:
P x1 , . . . , xn , 1 → Qx1 , . . . , xn , 1
Qx1 , . . . , xn , z → Sz, z 0 → Ez, y → Qx1 , . . . , xn , z 0 .
Then Q represents W .
As for the relation (∃z ≤ y)R(x1 , . . . , xn , z) (as a rela-
tion between x1 , . . . , xn , y), it is equivalent to the relation
∃z(z ≤ y ∧ R(x1 , . . . , xn , z). Let S(x1 , . . . , xn , y, z) be the relation
z ≤ y∧R(x1 , . . . , xn , z). It must be recursively enumerable, because
it is the intersection S1 ∩ S2 of the relations S1 and S2 where
S1 = λx1 , . . . , xn , y, z (z ≤ y)
and
S2 = λx1 , . . . , xn , y, z R(x1 , . . . , xn , z).
Since S1 and S2 are respectively definable from the recursively
enumerable relations z ≤ y and R(x1 , . . . , xn , z) by explicit trans-
formation, they are recursively enumerable, and so is their intersec-
tion S. Thus so is the existential quantification ∃zS(x1 , . . . , xn , z),
which is the relation (∃z ≤ y)R(x1 , . . . , xn , z).
(b) Suppose R is recursive. Thus R and its complement R are both
recursively enumerable. Since R is recursively enumerable, so are
∀F (R) and ∃F (R) [by (a)]. Since R is recursively enumerable, so
are ∀F (R) and ∃F (R) [again by (a)]. But as the reader can ver-
ify, ∀F (R) and ∃F (R) are respectively the complements of ∃F (R)
and ∀F (R), and so the complements of ∃F (R) and ∀F (R) are both
recursively enumerable, which means that ∃F (R) and ∀F (R) are
both recursive.
4. Let us first note that for any relation R(x1 , . . . , xn , y, z) the following
conditions are equivalent:
(1) ∃y∃zR(x1 , . . . , xn , y, z)
(2) ∃w(∃y ≤ w)(∃z ≤ w)R(x1 , . . . , xn , y, z)
It is obvious that (2) implies (1). Now, suppose that (1) holds.
Then there are numbers y and z such that R(x1 , . . . , xn , y, z). Let
106 Part II: Recursion Theory and Metamathematics

w be the maximum of y and z. Then y ≤ w and z ≤ w, and so

(∃y ≤ w)(∃z ≤ w)R(x1 , . . . , xn , y, z) holds for such a number w. Hence
(2) holds.
Now suppose R(x1 , . . . , xn , y) is Σ1 . Then there is a Σ0 rela-
tion S(x1 , . . . , xn , y, z) such that R(x1 , . . . , xn , y) is equivalent
to ∃zS(x1 , . . . , xn , y, z). Thus ∃yR(x1 , . . . , xn , y) is equivalent to
∃y∃zS(x1 , . . . , xn , y, z), which, as shown above, is Σ1 [since the rela-
tion (∃y ≤ w)(∃z ≤ w)S(x1 , . . . , xn , y, z) is Σ0 ].
5. x ∈ A0 iff f (x) ∈ A, which is true iff ∃y(f (x) = y ∧ y ∈ A). We first
show that the relation f (x) = y ∧ y ∈ A is Σ1 .
Well, since the relation f (x) = y is Σ1 , then there is a Σ0 relation
S(x, y, z) such that f (x) = y iff ∃zS(x, y, z).
Since the set A is Σ1 , there is a Σ0 relation R(x, y) such that x ∈ A
iff ∃wR(x, w). Thus f (x) = y ∧ y ∈ A, iff
∃zS(x, y, z) ∧ ∃wR(x, w),
which is true iff ∃z∃w(S(x, y, z) ∧ R(x, w)). Since the condition
S(x, y, z) ∧ R(x, w) is Σ0 , then ∃w(S(x, y, z)R(x, w), is Σ1 , hence so
is the condition ∃z∃w(S(x, y, z)R(x, w)) [by Problem 4].
Thus the condition f (x) = y ∧ y ∈ A is Σ1 , and therefore so is
∃y(f (x) = y ∧ y ∈ A) [again by Problem 4]. Thus A0 is Σ1 .
6. Our goal is to prove that the relation g(x) = y is Σ1 . First let R1 (x, y)
be the Σ0 relation (x = 1 ∧ y = 12) ∨ (x = 2 ∧ y = 122) Thus R1 (x, y)
holds iff x is the single digit 1 or the single digit 2 and g(x) = y.
Let R2 (x, y, x1 , y1 ) be the Σ0 relation
(x = x1 1 ∧ y = y1 12) ∨ (x = x1 2 ∧ y = y1 122).
Call a set S of ordered pairs of numbers special, if for every pair (x, y) in
S, either R1 (x, y) or (∃x1 ≤ x)(∃y1 ≤ y)((x1 , y1 ) ∈ S ∧ R2 (x, y, x1 , y1 )).
It is easily seen by mathematical induction on the length of x (i.e. the
number of occurrences of 1 or 2) that if S is special, then for every pair
(x, y) in S, it must be that g(x) = y.
Next we must show that conversely, if g(x) = y, then (x, y) belongs
to some special set.
Well, the x in g(x) = y must be of the form x = d1 d2 . . . dn , where
each di is the digit 1 or the digit 2. For each i ≤ n, we let gi be the
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 107

dyadic Gödel number of di (thus, if di = 1, then gi = 12 and if di = 2,

then gi = 122). Now let Sx be the set
{(d1 , g1 ), (d1 d2 , g1 g2 ), . . . , (d1 d2 . . . dn , g1 g2 . . . gn )} .
The set Sx is obviously special, and contains the pair (x, g(x)) [which
is the pair (d1 d2 . . . dn , g1 g2 . . . gn )]. Thus if g(x) = y, then (x, y) ∈ Sx
and so (x, y) belongs to a special set. Thus we have:
(1) g(x) = y iff (x, y) belongs to a special set.
We now use Lemma K. For any number z, we let Sz be the set of all
pairs (x, y) such that K(x, y, z) holds. Thus (x, y) ∈ Sz iff K(x, y, z).
Since (x, y) ∈ Sz iff K(x, y, z), then to say that Sz is special is to say
that for all x and y, K(x, y, z) implies that the following Σ0 condition —
call it A(x, y, z) — holds:
R1 (x, y) ∨ (∃x1 ≤ x)(∃y1 ≤ y)(K(x1 , y1 , z) ∧ R2 (x, y, x1 , y1 )).
Thus Sz is special iff the following condition holds:
C(z): ∀x∀y(K(x, y, z) ⊃ A(x, y, z)).
However, C(z) is equivalent to the following Σ0 condition:
P (z): (∀x ≤ z)(∀y ≤ z)(K(x, y, z) ⊃ A(x, y, z)).
To see the equivalence, it is obvious that C(z) implies P (z). To see the
converse, suppose P (z) holds. We must show that C(z) holds. Let x
and y be arbitrary numbers. We need to show that if K(x, y, z) holds,
then so does A(x, y, z). If K(x, y, z) holds, then it must also be true
that x ≤ z and y ≤ z [by the second condition in Lemma K], and
therefore A(x, y, z) holds [by the assumption P (z)]. Thus, for all x and
y, K(x, y, z) ⊃ A(x, y, z), and so C(z) holds.
This proves that C(z) is equivalent to P (z), and hence Sz is special
iff P (z) holds.
Finally, g(x) = y iff (x, y) belongs to some special set, which is true
iff
∃z((x, y) ∈ Sz ∧ Sz is special),
which is true iff
∃z(K(x, y, z) ∧ P (z)).
Thus the relation g(x) = y is Σ1 (since K(x, y, z) ∧ P (z) is Σ0 ), which
completes the proof that every recursively enumerable set A is Σ1 .
108 Part II: Recursion Theory and Metamathematics

7. δ(x, y) = z iff
2 2
(x ≤ y ∧ z = (y − 1) + x) ∨ (y < x ∧ z = (x − 1) + x + y).
2
The condition z = (y − 1) + x can be written

(∃w ≤ z)(y = w + 1 ∧ z = w2 + x).

2
The condition z = (x − 1) + x + y can be written

(∃w ≤ z)(x = w + 1 ∧ z = w2 + x + y).

As the formula expressing δ(x, y) = z can be rewritten as a

Σ0 formula, the function δ is Σ0 , and consequently recursive [by
Proposition 2].
8. (a) z = δ(x, y) for some x and y. Then Kz = x and Lz = y, so that
δ(x, y) = δ(Kz, Lz). Thus z [which is δ(x, y)] equals δ(Kz, Lz).
(b) K(z) = x iff (∃y ≤ z)(δ(x, y) = z) and L(z) = y iff
(∃x ≤ z)(δ(x, y) = z). (It is not difficult to verify that if δ(x, y) = z,
both x ≤ z and y ≤ z.)
9. Conceptually, we know the following:
(a) δn (x1 , . . . , xn ) ∈ A iff ∃y(δn (x1 , . . . , xn ) = y ∧ y ∈ A).
(b) x ∈ A iff ∃x1 . . . ∃xn (x = δn (x1 , . . . , xn ) ∧ R(x1 , . . . , xn )).
To see that this understanding solves the problem, we can amplify as
follows:
(a) When we say the set A is recursively enumerable, what we mean
is that there’s a relation of degree 1, R(x), in the form of a Σ1
formula F1 with one free variable x (and one existential quanti-
fier) such that the set of numbers in A is the set of all numbers
x for which F1 (x) holds. Also, since δn (x1 , . . . , xn ) = y has been
shown to be recursive, there is a Σ0 formula F2 containing the free
variables x1 , . . . , xn , y which holds for an n + 1-tuple of numbers
iff the equation δn (x1 , . . . , xn ) = y is true (when the n + 1 num-
bers are “plugged in” appropriately). So δn (x1 , . . . , xn ) ∈ A can be
expressed by the formula

∃y(F 2 (x1 , . . . , xn , y) ∧ F1 (y)).

This can be seen to be a Σ1 formula, for the single unbound quan-

tifier in F1 can be brought to the front in a way the reader should
be familiar with by now.
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 109

(b) Since the relation R(x1 , . . . , xn ) is recursively enumerable, there

is a Σ1 formula F1 with the free variables x1 , . . . , xn such that
F1 (x1 , . . . , xn ) holds iff R(x1 , . . . , xn ) is true. Assuming F2 as in
part (a) of this problem, x ∈ A can be expressed by the formula
∃x1 . . . ∃xn (F2 (x1 , . . . , xn , x) ∧ F1 (x1 , . . . , xn )).
10. (a) For a given x, since δn is 1-1 and onto, x is equal to δn (x1 , . . . , xn )
for just one n-tuple x1 , . . . , xn . Moreover for each i ≤ n, Kin (x) is
defined to be xi . So
δn (K1n (x), K2n (x), . . . , Knn (x)) = δn (x1 , . . . , xn ) = x.
(b) Since x = δ2 (Kx, Lx) [by the definition of δ2 and Problem 8], we have
that K12 (x) = K12 (δ2 (Kx, Lx)) = Kx and K22 (δ2 (Kx, Lx)) = Lx.
(c) Suppose x = δn+1 (x1 , . . . , xn+1 ). Then [by the definition of
δn+1 ], x = δ(δ n (x1 , . . . , xn ), xn+1 ). Thus Kx = δn (x1 , . . . , xn )
and Lx = xn+1 .
(d) Let x = δn+1 (x1 , . . . , xn , xn+1 ) for some x1 , . . . , xn+1 . Then by
definition
Kin+1 (x) = xi .
Now suppose i ≤ n. Then Kx = δn (x1 , . . . , xn ) [by (c)], so that
Kin (Kx) = Kin (δn (x1 , . . . , xn ) = xi .
But also Kin+1 (x) = xi . So Kin+1 (x) = Kin (Kx) [for i ≤ n].
n+1
Now, Kn+1 (x) = xn+1 = Lx [by (c)].
(e) Take M (x, y) = R(K1n (x), . . . , Knn (x)K1m (y), . . . , Km m
(y)). Then
R(x1 , . . . , xn , y1 , . . . , ym ) iff M (δn (x1 , . . . , xn ), δm (y1 , . . . , ym )).
11. N (x): x is a number
N1
N2
N x → N y → Ny
Acc(x): x is a string of accents
Acc 0
Acc x → Acc x0
V ar(x): x is a TS variable
Acc x → V ar vx
110 Part II: Recursion Theory and Metamathematics

dv(x, y): x and y are distinct TS variable

Acc x → Acc y → dv vxv, vxyv
dv x, y → dv y, x
P (x): x is a TS predicate
Acc x → P px
P x → P px
t(x): x is a TS term
N x → tx
V ar x → t x
t x → t y → t xy
F0 (x): x is a TS atomic formula
Acc x → t y → F0 pxy
F0 x → t y → F0 pxcomy
F (x): x is a TS formula
F0 x → F x
F0 x → F y → F ximpy
S0 (x): x is a TS atomic sentence
Acc x → N y → S0 pxy
S0 x → N y → S0 pxcomy
S(x): x is a TS sentence
S0 x → S y
S0 x → S y → S ximpy
Sub(x, y, z): x is a string, y a TS variable, z a numeral, and. . .
N x → V ar y → N z → Sub x, y, z, x
V ar x → N z → Sub x, x, z, z
dv x, y → N z → Sub x, y, z, x
P x → V ar y → N z → Sub x, y, z, x
V ar y → N z → Sub com, y, z, com
V ar y → N z → Sub imp, y, z, imp
Sub x, y, z, w → Sub x1 , y, z, w1 → Sub xx1 , y, z, ww1
pt(x, y): x is a TS atomic formula and y is a partial instance of x
F x → Sub x, y, z, w → pt x, w
pt x, y → pt y, z → pt x, z
 Chapter 4. Elementary Formal Systems and Recursive Enumerability 111

in(x, y): x is a TS atomic formula and y is an instance of x

pt x, y → S y → in x, y

B(x): x is a TS base

F x→Bx
Bx → F x → B x ∧ y

C(x, y): x is a component of the base and y

F x → C x, x
C x, y → F z → C x, y ∧ z
C x, y → F z → C x, z ∧ y

T (x): x is a true TS sentence

C x, y → in x, x1 → T y ` x1
T y ` x → T y ` ximpz → S0 x → T y ` z
12. Let D be a dyadic system with predicate G and the following axiom
schemes:
G 1, 12
G 2, 122
G x → G y → G xy

Then G represents the relation “the Gödel number of x is y”. Now add
a predicate M and the axiom scheme:

G x, y → M x, xy.

Then M represents the relation “norm(x) = y”.

13. Suppose A is recursively enumerable. Let D be a dyadic system in which
“A” represents the set A and “M ” represents the relation “the norm
of x is y”. Add a predicate “B” and the axiom scheme

M x, y → A y → B x.

Then B represents the set A# .

Remark. Without using elementary formal systems or dyadic system,

one can directly show that if A is Σ1 , so is A# , and hence that if A is
recursively enumerable, so is A# . As a good exercise, show that if A is
Σ1 , so is A# .
112 Part II: Recursion Theory and Metamathematics

14. The following beautiful proof is a modification of Gödel’s famous diag-

onal argument in his incompleteness theorem proof, and does for recur-
sion theory what Gödel did for mathematical systems.
Suppose A is recursively enumerable. Then so is A# (by the Lemma).
Let H be a predicate of (U ) that represents A# . Then for every num-
ber n,
Hn is true iff n ∈ A# , which if true iff nn0 ∈ A.
Thus Hn is true iff nn0 ∈ A.
We take for n the Gödel number h of H, and obtain that Hh is true
iff hh0 ∈ A.
However, hh0 is the Gödel number of Hh! Thus Hh is true iff its
Gödel number is in A, and if thus a Gödel sentence for A.
 Chapter 5

Some Recursion Theory

I. Enumeration and Iteration Theorems

We now turn to two basic theorems of recursion theory — the enumeration

theorem, due to Emil Post [1943; 1944] and to Stephen Kleene [1952], and
the iteration theorem, due to Kleene [1952] and independently discovered
by Martin Davis [1958, 1982]. Once we have established these two results,
the universal system (U ) of the last chapter will have served its purpose,
and will no longer be needed.

The Enumeration Theorem

We wish to arrange all recursively enumerable sets in an infinite sequence

ω1 , ω2 , . . . , ωn , . . . (allowing repetitions) in such a way that the relation “ωx
contains y” is a recursively enumerable relation between x and y.
We recall that for any number (dyadic numeral) y, we are letting y0
denote the dyadic Gödel number of y. For any x and y we now define
r(x, y) to be xy0 [x followed by y0 ].
We use the letter T to denote the set of true sentences of the universal
system (U ) and use T0 to denote the set of Gödel numbers of the sentences
in T . We now define ωn to be the set of all x such that r(n, x) ∈ T0 . Thus
x ∈ ωn iff r(n, x) ∈ T0 .

Problem 1. (a) Prove that ωn is recursively enumerable. (b) Prove that

for every recursively enumerable set A, there is some n such that A = ω n .
114 Part II: Recursion Theory and Metamathematics

We shall call n an index of a set A if A = ω n . We have just shown that

a set A is recursively enumerable if and only if it has an index, and so we
have succeeded in arranging all recursively enumerable sets in an infinite
sequence ω1 , ω2 , . . . , ωn , . . . in such a manner that the relation ωx contains
y (which is the relation r(x, y) ∈ T0 ) is recursively enumerable.
We next wish to show that for each n ≥ 2, we can enumerate
all recursively enumerable relations of degree n in an infinite sequence
R1n , R2n , . . . , Rin , . . . such that the relation Rxn (y1 , . . . yn ) is a recursively
enumerable relation among the variables x, y1 , . . . , yn . For this, it will be
useful to use the indexing of recursively enumerable sets that we already
have, along with the pairing function δ(x, y) and the n-tupling functions
δn (x1 , . . . , xn ). We simply define Rin (x1 , . . . , xn ) to be δn (x1 , . . . , xn ) ∈ ωi .
Since ωi is recursively enumerable and the function δn (x1 , . . . , xn ) is also
recursively enumerable, so is Rin (x1 , . . . , xn ) [by (a) of Problem 9 of
Chapter 4]. Also, for any recursively enumerable relation R(x1 , . . . , xn ), the
set A of numbers δn (x1 , . . . , xn ) such that R(x1 , . . . , xn ) holds is recursively
enumerable [by (b) of Problem 9 of Chapter 4]. Thus the set A has some
index i so that R(x1 , . . . , xn ) holds iff δn (x1 , . . . , xn ) ∈ ωi , which is true iff
Rin (x1 , . . . , xn ). Thus R = Rin . Thus every recursively enumerable relation
has an index. [We also write Rn1 for x ∈ ω n , since we are regarding number
sets as relations of degree 1.]
We have thus proved:

Theorem E. [The Enumeration Theorem] For each n, the set of all recur-
sively enumerable relations of degree n can be enumerated in an infinite
sequence R1n , R2n , . . . , Rxn , . . . such that the relation Rxn (y1 , . . . yn ), as a rela-
tion among x, y 1 , . . . , yn , is recursively enumerable.
For each n we let U n+1 (x, y1 , . . . yn ) be the relation Rxn (y1 , . . . , yn ) [as
a relation among x, y 1 , . . . , yn ]. These relations are referred to as universal
relations: U 2 is the universal relation for all recursively enumerable sets, U 3
is the universal relation for all recursively enumerable relations of degree 2,
and so forth.
We sometimes write Rx (y1 , . . . , yn ) instead of Rxn (y1 , . . . , yn ), since
the number of arguments makes the superscript n clear. And we write
U (x, y1 , . . . , yn ) instead of U n+1 (x, y1 , . . . , yn ).
 Chapter 5. Some Recursion Theory 115

Iteration Theorems

The second basic tool of recursion theory is the iteration theorem, which
we will shortly state and prove in both a simple and more general form. To
fully appreciate the significance and power of these theorems, the reader
should first try solving the following exercises:

Exercise 1. We know that for any two recursively enumerable sets A

and B, the union A ∪ B is recursively enumerable. Moreover, the pro-
cess of finding this union is effective, in the sense that there is a recur-
sive function ϕ(i, j) such that if i is an index of A and j is an index of
B, then ϕ(i, j) is an index of A ∪ B — in other words there is a recur-
sive function ϕ(i, j) such that for all numbers i and j, ωϕ(i,j) = ωi ∪ ωj .
Prove this.
∨
For any relation R(x, y), by its inverse is meant the relation R satisfying
∨
the condition R (x, y) iff R(x, y).

Exercise 2. Show that there is a recursive function t(i) such that for
every number i the relation Rt(i) (x, y) is the inverse of the relation Ri (x, y),
i.e. Rt(i) (x, y) iff Ri (y, x). [This can be paraphrased: Given an index of a
recursively enumerable relation of two arguments, one can effectively find
an index of its inverse.]
For any function f (x) and number set A, by f −1 (A) is meant the set of
all numbers n such that f (n) ∈ A. Thus n ∈ f −1 (A) iff f (n) ∈ A.

Exercise 3. Prove that for any recursive function f (x), there is a recursive
function ϕ(i) such that for every number i, it is the case that
ωϕ(i) = f −1 (ωi ).

The reader who has solved the above exercises has probably had to go
back to the universal system (U ) each time. The iteration theorem, to which
we now turn, will free us from this once and for all!

Theorem 1. [Simple Iteration Theorem] For any recursively enumerable

relation R(x, y) there is a recursive function t(i) such that for all numbers
x and i, x ∈ ωt(i) iff R(x, i).
116 Part II: Recursion Theory and Metamathematics

Problem 2. Prove Theorem 1. [Hint: The recursively enumerable relation

R(x, y) is represented in the universal system (U ) by some predicate H.
Consider the Gödel number h of H, as well as the Gödel number c of the
punctuation symbol com.]

Theorem 2. [Iteration Theorem] For any recursively enumerable relation

R(x1 , . . . , xm , y1 , . . . , yn ) there is a recursive function ϕ(i1 , . . . , in ) such that
for all x1 , . . . , xm , i1 , . . . , in ,

R(x1 , . . . , xm , i1 , . . . , in ) iff Rϕ(i1 ,...,in ) (x1 , . . . , xm ).

One way to prove Theorem 2 is to use Theorem 1 and the fact that
for any recursively enumerable relation R(x1 , . . . , xm , y1 , . . . , yn ) there is a
recursively enumerable relation M (x, y) such that R(x1 , . . . , xm , y1 , . . . , yn )
iff M (δm (x1 , . . . , xm ), δn (y1 , . . . , yn )) [by Problem 10 (e) of Chapter 4].

Problem 3. Prove Theorem 2.

Let us now see how the iteration theorem provides simple and elegant
solutions to the three recently given exercises.
In Exercise 1, let R(x, y1 , y2 ) be the recursively enumerable relation

x ∈ ωy1 ∨ x ∈ ωy2 .

Then, by the iteration theorem, there is a recursive function ϕ(i1 , i2 ) such

that for all numbers x, i1 , i2 ,

x ∈ ωϕ(i1 ,i2 ) iff R(x, i1 , i2 ), which is true iff x ∈ ωi1 ∨ x ∈ ωi2 .

Therefore, ωϕ(i1 ,i2 ) = ωi1 ∪ ωi2 .

For Exercise 2, let R(x1 , x2 , y) be the recursively enumerable relation
Ry (x2 , x1 ). It is recursively enumerable because it is explicitly defined
from the recursively enumerable relation U (x1 , x2 , y). Then by the sim-
ple iteration theorem there is a recursive function ϕ(i) such that for all x
and i, Rϕ(i) (x1 , x2 ) iff R(x1 , x2 , i), which is true iff Ri (x2 , x1 ).
For Exercise 3, given a recursive function f (x), consider the recursively
enumerable relation R(x, y) – viz. f (x) ∈ ωy [it is recursively enumerable
since it is equivalent to ∃z(f (x) = z ∧ z ∈ wy )]. By the simple itera-
tion theorem there is a recursive function t(i) such that for all x and i,
x ∈ ωt(i) iff R(x, i) which is true iff f (x) ∈ ωi . Thus ωt(i) = f −1 (ω i ).
More serious applications of the iteration theorem will soon follow.
 Chapter 5. Some Recursion Theory 117

Maximal Indexing

A denumerable sequence A1 , A2 , . . . , An , . . . of all recursively enumerable

sets is called a recursively enumerable indexing if the relation x ∈ Ay is
recursively enumerable. It is called a maximal indexing if for every recur-
sively enumerable indexing B1 , B2 , . . . , Bn , . . . there is a recursive function
f (i) such that for all i the condition Af (i) = Bi holds. [Informally, this
means that given a B-index of a recursively enumerable set, we can effec-
tively find an A-index of the set.]
Another application of the iteration theorem yields:

Theorem 3. The enumeration ω1 , ω2 , . . . , ωn , . . . given by the enumeration

theorem in maximal.

Problem 4. Prove Theorem 3.

The following theorem is easily obtained from the iteration theorem,
combined with the enumeration theorem.

Theorem 4. [Uniform Iteration Theorem] For any numbers m and n, there

is a recursive function ϕ(i, i1 , . . . , in ) [sometimes written Snm (i, i1 , . . . , in )]
such that for all x1 , . . . , xm , i1 , . . . , in ,
Rϕ(i,i1 ,...in ) (x1 , . . . , xm ) iff Ri (x1 , . . . , xm , i1 , . . . , in ).

Problem 5. Prove Theorem 4.

II. Recursion Theorems

Recursion theorems, which can be stated in many forms, and are sometimes
called fixed point theorems, have many applications to recursion theory and
metamathematics. To illustrate their rather startling nature, consider the
following mathematical “believe it or nots.”
Which of the following propositions, if true, would surprise you?
1. There is a number n such that ωn = ωn+1 .
2. For any recursive function f (x), there is a number n such that
ωf (n) = ωn .
3. There is a number n such that ωn contains n as its only member, i.e.
ωn = {n}.
118 Part II: Recursion Theory and Metamathematics

4. For any recursive function f (x) there is a number n such that

ωn = {f (n)}.
5. There is a number n such that for every number x, x ∈ ωn iff
n ∈ ωx .

Well, believe it or not, all the above statements are true! They are all
special cases of the theorem that follows:
Consider a recursively enumerable relation R(x, y). Now consider a num-
ber n and the set x: R(x, n), i.e. the set of all numbers x such that R(x, n)
holds. This set is recursively enumerable, hence is ωn∗ for some n∗ . Given
the relation R, what n∗ is depends on what n is. Wouldn’t it be curious if
n could be chosen such that n∗ is n itself? Well, this is indeed so:

Theorem 5. [Weak Recursion Theorem] For any recursively enumerable

relation R(x, y), there is a number n such that ωn = {x : R(x, n)} [in other
words, for all x, we have n ∈ ωn iff R(x, n)].
I will give the beginning of the proof, and leave it as a problem for the
reader to finish it.
The relation Ry (x, y), as a relation between x and y is recursively enu-
merable. Hence, by the simple iteration theorem there is a recursive function
t(y) such that for all x and y, x ∈ ωt(y) iff Ry (x, y).
Now, take any recursively enumerable relation R(x, y). Then the rela-
tion R(x, t(y)) is also a recursively enumerable relation between x and y,
hence has some index m. Thus Rm (x, y) iff R(x, t(y)). Now comes the great
trick: . . . .

Problem 6. Finish the proof.

As a consequence of Theorem 5, we have:

Theorem 6. For any recursive function f (x) there is a number n such that
ωf (n) = ωn .

Problem 7. (a) Prove Theorem 6. (b) Now prove the other four “believe
it or nots”.

Discussion. We have derived Theorem 6 as a corollary of Theorem 5. One

can alternatively first prove Theorem 6 directly, and then obtain Theorem
5 as a corollary of Theorem 6 as follows: Assume Theorem 6. Now consider
a recursively enumerable relation R(x, y). By the iteration theorem, there
 Chapter 5. Some Recursion Theory 119

is a recursive function f (x) such that for all x and y,

x ∈ ωf (y) iff R(x, y).
But by Theorem 6, there is a number n such that ωf (n) = ωn . Thus for all
x, x ∈ ωn iff x ∈ ωf (n) , which is true iff R(x, n). Thus x ∈ ωn iff R(x, n).

Unsolvable Problems and Rice’s Theorem

Some of the aforementioned believe it or nots may seem to be somewhat

frivolous applications of the weak recursion theorem. We now turn to a far
more significant application.
Let us call a property P (n) solvable if the set of all n having the property
is a recursive set. For example, suppose P (n) is the property that ωn is a
finite set. To ask whether this property is solvable is to ask whether the set
of all indices of all finite sets is recursive.
Here are some typical properties P (n) that have come up in the
literature:
(1) ωn is an infinite set.
(2) ωn is empty.
(3) ωn is recursive.
(4) ωn = ω1 .
(5) ωn contains 1.
(6) The relation Rn (x, y) is single-valued, i.e. for each x there is at most
one y such that Rn (x, y).
(7) The relation Rn (x, y) is functional, i.e. for each x there is exactly one
y such that Rn (x, y).
(8) All numbers are in the domain of Rn (x, y), i.e. for each (positive integer)
x there is at least one y such that Rn (x, y).
(9) All numbers are in the range of Rn (x, y), i.e. for each (positive integer)
y there is a least one x such that Rn (x, y).

For each of these nine properties, we can ask if it is solvable. These are
typical questions of recursion theory that have been answered by individual
methods. Rice’s theorem, to which we now turn, answers a host of such
questions in one fell swoop.
A number set A is called extensional if it contains with any index of a
recursively enumerable set all other indices of it as well — in other words,
for every i and j, if i ∈ A and ωi = ωj , then j ∈ A.
120 Part II: Recursion Theory and Metamathematics

Theorem R. [Rice’s Theorem] The only recursive extensional sets are

N and ∅.
We will soon see how Rice’s theorem [Rice, 1953] is relevant to the nine
properties mentioned above, but first let us go to its proof. We first need
the following:

Lemma. For any recursive set A other than N and ∅, there is a recursive
function f (x) such that for all n, n ∈ A iff f (n) ∈
/ A.

Problem 8. (a) Prove the above lemma. [Hint: If A is neither N nor ∅,

then there is at least one number a ∈ A and at least one number b ∈ / A.
Use these two numbers a and b to define the function f (x). (b) Now prove
Rice’s Theorem. [Hint: Use the function f (x) of the Lemma and then apply
Theorem 6 to the function f (x).]

Applications
Let us consider the first of the aforementioned nine properties P (n), namely
that ωn is infinite. Obviously, if ωi is infinite and ωi = ωj , then ωj is also
infinite. Thus the set of the indices of all infinite sets is extensional. Also,
there is at least one n such that ωn is infinite and at least one n such that
ωn is not infinite. Hence by Rice’s Theorem, the set of the indices of all
infinite sets is not recursive.
The same argument applies to all the other eight cases, so that none of
them are solvable!

The Strong Recursion Theorem

By the weak recursion theorem, for any number i, there is a number n such
that ωn = {x : Ri (x, n)}. Given the number i, can we effectively find such
an n as a recursive function of i? That is, is there a recursive function ϕ(i)
such that for every number i, ωϕ(i) = {x : Ri (x, ϕ(i))}? The answer is yes,
and this is Myhill’s famous fixed point theorem, which is a special case of
the following:

Theorem 7. [Strong Recursion Theorem] For any recursively enumerable

relation M (x, y, z), there is a recursive function ϕ(i) such that for all i,
ωϕ(i) = {x : M (x, i, ϕ(i))}.
 Chapter 5. Some Recursion Theory 121

We will give two proofs of this important theorem. The first is along
traditional lines, and requires two applications of the iteration theorem.
The second proof is along different lines, and requires only one application
of the iteration theorem.
Here is the beginning of the first proof (to be finished by the reader in
Problem 9(a)).

As in the proof of the weak recursion theorem, by the (simple)

iteration theorem, for every y, there is a recursive function d(y) such
that ωd(y) = {x : Ry (x, y)}. Now, given a recursively enumerable relation
M (x, y, z), consider the recursively enumerable relation M (x, z, d(y)) as a
relation among x, y and z. By another application of the iteration theorem,
there is a recursive function t(i) such that for all i, x and y, Rt(i) (x, y) iff
M (x, i, d(y)). Then . . . .
For the second proof (to be finished by the reader – Problem 9 (b)
below), let S(x, y, z) be the recursively enumerable relation Rz (x, y, z). By
the iteration theorem, there is a recursive function t(y, z) such that for all
y and z, ωt(y,z) = {x : S(x, y, z)}, and thus ωt(y,z) = {x : Rz (x, y, z)}.
Now, given a recursively enumerable relation M (x, y, z), the relation
M (x, y, t(y, z)) is recursively enumerable, hence has an index h. Thus
Rh (x, y, z) iff M (x, y, t(y, z)). Then . . . .

Problem 9. (a) Finish the first proof. (b) Finish the second proof.

As a special case of Theorem 7, we have:

Theorem 8. [Myhill’s Fixed Point Theorem] There is a recursive function

ϕ(y) such that for all y, ωϕ(y) = {x : Ry (x, ϕ(y))}.

Problem 10. Why is this a special case?

Creative, Productive, Generated

and Universal Sets

To say that a set A is not recursively enumerable is to say that given any
recursively enumerable subset ωi of A there is a number n in A but not
in ωi — a witness, so to speak that ωi is not the whole of A. A set A is
called productive if there is a recursive function ϕ(x) — called a productive
122 Part II: Recursion Theory and Metamathematics

function for A — such that for every number i, if ωi is a subset of A, then

ϕ(i) ∈ A − ωi [i.e. ϕ(i) is in A, but not in ωi ].
A set A is called co-productive if its complement A is productive. This
is equivalent to the condition that there is a recursive function ϕ(x) —
called a co-productive function for A — such that for every number i, if ωi
is disjoint from A, then ϕ(i) lies outside A and outside ωi .
A set A is called creative if it is co-productive and recursively enumer-
able. Thus a creative set A is one which is recursively enumerable, and is
such that there is a recursive function ϕ(i) — called a creative function for
A — such that for any number i, if ωi is disjoint from A, then ϕ(i) lies
outside both A and ωi (it is a witness to the fact that A is not recursively
enumerable).

Post’s Sets C and K

A simple example of a creative set is Post’s set C, the set of all numbers x
such that x ∈ ωx . Thus for any number i, i ∈ C iff i ∈ ωi . If ωi is disjoint
from C, then i is outside both ωi and C, and therefore the identity function
I(x) is a creative function for C.
A less immediate example of a creative set is the set K of all numbers
∂(x, y) such that x ∈ ωy . The set K was also introduced by Post [1943] and
is sometimes called the complete set. The proof that K is creative is less
immediate than the proof for C, and involves use of the iteration theorem,
as we will see.

Complete Productivity and Creativity

It is also correct to say that a set A is not recursively enumerable iff for
every recursively enumerable set ωi , whether a subset of A or not, there is
a number j such that j ∈ A iff j 6= ωi — thus attesting to the fact that
A 6= ωi . By a completely productive function for A is meant a recursive
function ϕ(x) such that for every number i, ϕ(i) ∈ A iff ϕ(i) ∈/ ωi . A set A
is called completely productive if there is a completely productive function
for A.
A set A is called completely creative if A is recursively enumerable and
there is a completely productive function ϕ(x) for the complement A of A;
 Chapter 5. Some Recursion Theory 123

and such a function is called a completely creative function for A. Thus a

completely creative function for A is a recursive function ϕ(x) such that
for every number i, ϕ(i) ∈ A iff ϕ(i) ∈ ωi .
Post’s set C of all x such that x ∈ ωx is obviously not only creative,
but completely creative, since the identity function is a completely creative
function for C. Therefore a completely creative set exists. Post apparently
did not capitalize on the fact that C is not only creative, but completely
creative. I hazard a guess that if he had capitalized on this fact, then many
results in recursion theory would have come to light sooner.
We will see later that in fact any productive set is completely productive,
and hence that any creative set is completely creative. This important result
is due to John Myhill [1955], and will be seen to be a consequence of the
strong recursion theorem.

Generative Sets
Many important results about completely creative sets do not use the fact
that the set is recursively enumerable, and so we define a set A to be gen-
erative if there is a recursive function ϕ(x) — called a generative function
for A — such that for all numbers i, ϕ(i) ∈ A iff ϕ(i) ∈ ω i . Thus a gener-
ative set is like a completely creative set except that it is not necessarily
recursively enumerable. A set A is generative iff its complement A is com-
pletely productive. Our emphasis will be on generative sets.

Many-One Reducibility

By a (many-one) reduction of a set A to a set B is meant a recursive function

f (x) such that A = f −1 (B); i.e. for all numbers x, x ∈ A iff f (x) ∈ B. A set
A is said to be (many-one) reducible to B if there is a (many-one) reduction
of A to B. There are other important types of reducibility that occur in the
literature of recursion theory, but throughout this chapter, reducible will
mean many-one reducible.

Theorem 9. If A is reducible to B and A is generative, then B is

generative.
It will be useful to state and prove Theorem 9 in a more specific form:
For any recursively enumerable relation R(x, y), by the iteration theorem
124 Part II: Recursion Theory and Metamathematics

there is a recursive function t(y) such that ωt(i) = {x : R(x, i)}. Such a
function t(y) will be called an iterative function for the relation R(x, y).
We will now state and prove Theorem 9 in the more specific form:

Theorem 9∗ . If f (x) reduces A to B and ϕ(x) is a generative function

for A, and t(y) is an iterative function for the relation f (x) ∈ ωy , then
f (ϕ(t(x))) is a generative function for B.

Problem 11. Prove Theorem 9∗ .

Universal Sets
A set A is called universal if every recursively enumerable set is reducible
to A.
Post’s complete set K is obviously universal.

Problem 12. Why?

Theorem 10. Every universal set is generative.

Problem 13. Prove Theorem 10. [Hint: Use Theorem 9.]

Uniformly Universal Sets

For any function f (x, y) of two arguments, and for any number i, by fi
is meant the function which assigns to each x the number f (x, i). Thus
fi (x) = f (x, i).
A set A is called uniformly universal under a recursive function f (x, y)
if for all i and x, x ∈ ωi iff f (x, i) ∈ A; in other words, iff fi reduces ωi to
A. And we say that A is uniformly universal if it is universal under some
recursive function f (x, y).

Theorem 11. Every universal set is uniformly universal.

Problem 14. Prove Theorem 11. [Hint: Use the fact that Post’s set K is
not only universal, but is also uniformly universal. Why?]

Exercise. Suppose A is uniformly universal under f (x, y) and that t(y) is

an iterative function for the relation f (x, x) ∈ ωy . Show that the function
f (t(x), t(x)) is a generative function for A.
 Chapter 5. Some Recursion Theory 125

Remarks. The above, together with the fact that every universal set is
uniformly universal, yields another proof that every universal set is gen-
erative, and unlike our previous proof of this fact, make no appeal to the
existence of a previously constructed recursively enumerable generative set
(for instance, Post’s set C). Actually, the two proofs can be seen to yield
two different generative functions for Post’s set K.
Having shown that every universal set is generative, we now wish to
show that, conversely, every generative set is universal, and thus that a set
is generative if and only if it is universal.
To begin with, it is pretty obvious that if B is a generative set under
ϕ(x) (i.e. if ϕ(x) is a generative function for B), then, for every i:
(1) If ωi = N, then ϕ(i) ∈ B [N is the set of positive integers];
(2) If ωi = ∅, then ϕ(i) ∈/ B [∅ is the empty set].

Problem 15. Why is this so?

Problem 16. Prove that for every recursively enumerable set A, there is
a recursive function t(y) such that for every number i:
(1) If i ∈ A, then ωt(i) = N;
(2) If i ∈
/ A, then ωt(i) = ∅. [Hint: Define M (x, y) iff y ∈ A. The relation
M (x, y) is recursively enumerable, hence by the iteration theorem there
is a recursive function t(y) such that for all i, ωt(i) = {x : M (x, i)}.]

Theorem 12. Every generative set is universal.

Problem 17. Prove Theorem 12. [Hint: This follows fairly easily from
Problems 15 and 16.]
We next wish to show that every co-productive set is generative. Then,
by Theorem 12, we get Myhill’s result that every co-productive set (and
hence every creative set) is universal. [We recall that a set is creative iff it
is both co-productive and recursively enumerable.]

Weak Co-productivity

We recall that a set A is said to be co-productive under a recursive function

ϕ(x) if for every number i such that ωi is disjoint from A, the number ϕ(i)
lies outside both A and ωi . This obviously implies the following weaker
condition:
126 Part II: Recursion Theory and Metamathematics

C1 : For every i such that ωi is disjoint from A and such that ωi contains
at most one member, the number ϕ(i) lies outside both A and ωi .

Amazingly enough, this weaker condition on a recursive function is

enough to ensure that A is generative (and hence universal).
This weaker condition C1 implies the following even weaker condition:

C2 : For every number i,

(1) If ωi = ∅. Then ϕ(i) ∈
/ A;
(2) If ωi = {ϕ(i)}, then ϕ(i) ∈ A.

Problem 18. Why does C1 imply C2 ?

We will say that A is weakly co-productive under ϕ(x) if ϕ(x) is recursive
and condition C2 holds.

Theorem 13. [After Myhill] If A is co-productive — or even weakly

co-productive — then A is generative.
To prove Theorem 13, we need the following lemma:

Lemma. For any recursive function ϕ(x) there is a recursive function t(y)
such that for all y,
ωt(y) = ωy ∩ {ϕ(t(y))} .

Problem 19. Prove the above lemma. [Hint: Let M (x, y, z) be the recur-
sively enumerable relation x ∈ ωy ∧x = ϕ(z). Then use the strong recursion
theorem, Theorem 7.]

Problem 20. Now prove Theorem 13. [Hint: Suppose ϕ(x) is a weakly
co-productive function for A. Let t(y) be a recursive function related to
ϕ(x) as in the above lemma. Then show that the function ϕ(t(y)) is a
generative function for A.]
From Theorems 13 and 12 we have:

Theorem 14. [Myhill] If A is productive, or even weakly productive, then

A is universal.

Discussion

One can go directly from co-productivity (or even weak co-productivity) to

universality as did Myhill. Briefly, the argument is the following: Suppose
 Chapter 5. Some Recursion Theory 127

B is weakly co-productive under ϕ(x) and that A is a set that we wish

to reduce to B. Applying the strong recursion theorem to the recursively
enumerable relation y ∈ A ∧ x = ϕ(z) [as a relation among x, y and z],
we have a recursive function t(y) such that for all y, if y ∈ A, then
ωt(y) = {ϕ(y)}, and if y ∈
/ A, then ωt(y) = ∅. Then the function ϕ(t(x)) can
be seen to reduce A to B. The reader should find it a profitable exercise to
fill in the details.

Recursive Isomorphism

A set A is said to be recursively isomorphic to a set B if there is a recursive

1-1 function from A onto B. Such a function ϕ(x) is called a recursive
isomorphism from A to B, and A is said to be recursively isomorphic to B
under ϕ(x), if ϕ(x) is a recursive isomorphism from A to B.
If A is recursively isomorphic to B under ϕ(x), then obviously B is
recursively isomorphic to A under the inverse function ϕ−1 (x). [We recall
that ϕ−1 (x) = y iff ϕ(y) = x.]

Exercise. Suppose A is recursively isomorphic to B. Prove that if A is

recursively enumerable, (recursive, productive, generative, creative, univer-
sal) then B is respectively recursively enumerable (recursive, productive,
generative, creative, universal).
Myhill [1955] proved the celebrated result that any two creative sets are
recursively isomorphic.

Bonus Question. Is it possible for a universal set to be recursive?

Solutions to the Problems of Chapter 5

1. (a) We showed in the last chapter that the relation g(x) = y (i.e.
x0 = y) is recursively enumerable. Hence the relation r(x, y) = z is
recursively enumerable, since it is equivalent to

∃ w(g(y) = w ∧ xw = z).

Therefore, for any number n, the relation r(n, x) = y [as a rela-

tion between x and y] is recursively enumerable. And so for any
recursively enumerable set A, the condition r(n, x) ∈ A [as a
property of x] is recursively enumerable, since it is equivalent to
128 Part II: Recursion Theory and Metamathematics

∃ y(r(n, x) = y ∧ y ∈ A). In particular, since T0 is recursively enu-

merable, the condition ∃ y(r(n, x) = y ∧ y ∈ T0 ) is recursively enu-
merable, and this is the condition x ∈ ωn . Thus ωn is recursively
enumerable.
(b) Every recursively enumerable set A is represented in (U ) by some
predicate H. Thus for every number x, x ∈ A iff Hx ∈ T . We let h
be the Gödel number of H, and so for all x, Hx ∈ T iff r(h, x) ∈ T0 ,
which is true iff x ∈ ωh . Thus A = ωh .
2. The relation R(x, y) is represented in (U ) by some predicate H, and so
for all numbers i and x, R(i, x) iff Hicomx ∈ T , the latter of which is
true iff hi0 cx0 ∈ T0 , where h, i0 , c, x0 are the respective Gödel numbers
of H, i, com, x (in dyadic notation). Thus R(i, x) iff r(hi0 c, x) ∈ T0 ,
which is true iff x ∈ ωhi0 c . We thus take t(y) to be hy0 c, and so
t(i) = hi0 c, so that now R(i, x) iff x ∈ ωt(i) .
3. Given a recursively enumerable relation R(x1 , . . . , xm , y1 , . . . , yn ), we
use the recursively enumerable relation M (x, y), which yields, for all
numbers x1 , . . . , xm , i1 , . . . , in ,
(1) R(x1 , . . . , xm , i1 , . . . , in ) iff M (δm (x1 , . . . , xm ), δn (i1 , . . . , in )).
We now apply Theorem 1 to the relation M (x, y), and so there is a
recursive function t(i) such that for every x and i we have x ∈ ωt(i)
iff M (x, i). We now take δm (x1 , . . . , xm ) for x and δn (i1 , . . . , in ) for
i, and obtain
(2) δm (x1 , . . . , xm) ∈ ωt(δn (i1 ,...,in )) iff M (δm (x1 , . . . , xm ), δn (i1 , . . . , in )).
By (1) and (2) we have:
(3) R(x1 , . . . , xm , i1 , . . . , in ) iff δm (x1 , . . . , xm ) ∈ ωt(δn (i1 ,...,in )) which
is true iff Rt(δn (i1 ,...,in )) (x1 , . . . , xm ).
We thus take ϕ(i1 , . . . , in ) to be t(δn (i1 , . . . , in )) and obtain
R(x1 , . . . , xm , i1 , . . . , in ) iff Rϕ(i1 ,...,in ) (x1 , . . . , xm ).
4. We will show that our recursively enumerable enumeration
ω1 , ω2 , . . . , ωn , . . . is maximal. Given a recursively enumerable enumera-
tion B1 , B2 , . . . , Bn , . . ., let R(x, y) be the recursively enumerable rela-
tion x ∈ By . By the simple iteration theorem, there is a recursive
function f (x) such that for all x and i, we have x ∈ ωf (i) iff R(x, i),
which is true iff x ∈ Bi . Thus x ∈ ωf (i) iff x ∈ Bi , and so ωf (i) = Bi .
5. Define S(x1 , . . . , xm , y, y1 , . . . , yn ) iff Ry (x1 , . . . , xm , y1 , . . . , yn ). The
relation S is recursively enumerable (why?), and so by the
iteration theorem, there is a recursive function ϕ(i, i1 , . . . , in )
 Chapter 5. Some Recursion Theory 129

such that for all x1 , . . . , xm , i, i1 , . . . , in : Rϕ(i,i1 ,...,in ) (x1 , . . . , xm ) iff

S(x1 , . . . , xm , i, i1 , . . . , in ) iff Ri (x1 , . . . , xm , i1 , . . . , in ).
6. The great trick is to take m for y, and we thus obtain Rm (x, m) iff
R(x, t(m)). But we also have Rm (x, m) iff x ∈ ωt(m) ! Thus x ∈ ωt(m)
iff R(x, t(m)), and so x ∈ ωn iff R(x, n), where n is the number t(m).
[Clever, huh?]
7. (a) Given a recursive function f (x) let R(x, y) be the relation
x ∈ ωf (y) . It is recursively enumerable [since it is equivalent to
∃ z(f (y) = z ∧ x ∈ ωz )], so that by Theorem 5, there is some n such
that for all x, x ∈ ωn iff R(x, n). But R(x, n) ≡ x ∈ ωf (n) , and so,
for all x, x ∈ ωn iff x ∈ ωf (n) . Thus ωn = ωf (n) .
(b) Believe it or not:
#1 is but a special case of (a), namely when f (x) = x + 1.
#2 is just Theorem 6!
#3 is but the special case of #4 in which f (x) = x. And so we
must show #4: For f (x) a recursive function, the relation x = f (y)
is recursively enumerable. Hence by Theorem 5, there is some n
such that for all x, x ∈ ωn iff x = f (n). This means that f (n) is
the one and only member of ωn , and so ωn = {f (n)}.
As for #5, let R(x, y) be the relation y ∈ ω x . This relation is recur-
sively enumerable, and so there is a number n such that for all x,
x ∈ ωn iff R(x, n), but R(x, n) iff n ∈ ωx , and so x ∈ ωn iff x ∈ ωx .
8. (a) We are given that A is recursive and that A 6= N and A 6= ∅. Since
A 6= ∅, then there is at least one number a such that a ∈ A, and
since A 6= N there is at least one number b such that b ∈ / A.
Define f (x) to be b if x ∈ A and to be a if x ∈ / A. Since A
is recursive, then A and its complement A are both recursively
enumerable. Hence the function f (x) is recursive, since the rela-
tion f (x) = y is equivalent to the recursively enumerable relation
(x ∈ A ∧ y = b) ∨ (x ∈ A ∧ y = a). If x ∈ A, then f (x) ∈ / A [since
f (x) = b], and if x ∈
/ A, then f (x) ∈ A [since f (x) = a]. Thus x ∈ A
iff f (x) ∈
/ A.
(b) Let A be a recursive set that is neither N nor ∅. Take f (x) as in the
lemma. By Theorem 6, there is some n such that ωn = ωf (n) . Thus n
and f (n) are indices of the same recursively enumerable set, yet one
of them is in A and the other isn’t. Therefore A is not extensional.
130 Part II: Recursion Theory and Metamathematics

9. (a) If we take t(i) for y in the statement Rt(i) (x, y) iff M (x, i, d(y))
we obtain Rt(i) (x, t(i)) iff M (x, i, d(t(i))). But it is also true
that ωd(t(i)) = {x : Rt(i) (x, t(i))} [this is true because, for any n,
ωd(n) = {x : Rn (x, n)}]. Thus ωd(t(i)) = {x : M (x, i, d(t(i))). We
now take ϕ(i) to be d(t(i)), yielding ωϕ(i) = {x : M (x, i, ϕ(i))}.
(b) As noted, Rh (x, y, z) iff M (x, y, t(y, z)). We take h for z and so
Rh (x, y, h) iff M (x, y, t(y, h)). But since ωt(y,h) = {x : Rh (x, y, h)},
it follows that ωt(y,h) = {x : M (x, y, (t(y, h)))}. We thus take ϕ(y)
to be t(y, h), yielding ωϕ(y) = {x : M (x, y, ϕ(y))}.
10. This is the case that M (x, y, z) is the relation Ry (x, z).
11. Since f reduces A to B, then for all x, f (x) ∈ B iff x ∈ A. For any
number i we take ϕ(t(i)) for x and we obtain:
(1) f (ϕ(t(i))) ∈ B iff ϕ(t(i)) ∈ A.
Also, ϕ(t(i)) ∈ A iff ϕ(t(i)) ∈ ωt(i) [since ϕ(x) is a generative
function for A], so we have that:
(2) f (ϕ(t(i))) ∈ B iff ϕ(t(i)) ∈ ωt(i) .
Next, since t(y) is an iterative function for the relation f (x) ∈ ωy ,
then for all x and i we have x ∈ ωt(i) iff f (x) ∈ ω i . We now take
ϕ(t(i)) for x and obtain:
(3) ϕ(t(i)) ∈ ωt(i) iff f (ϕ(t(i))) ∈ ωi .
By (2) and (3), we see that f (ϕ(t(i))) ∈ B iff f (ϕ(t(i))) ∈ ωi , and
thus f (ϕ(t(x))) is a generative function for B.
12. For any number i, x ∈ ωi iff for all x, δ(x, i) ∈ K. Let ϕi be the
function that assigns to each x the number δ(x, i). Thus ϕi (x) = δ(x, i).
Consequently, x ∈ ωi iff ϕi (x) ∈ K, and so ϕi reduces ωi to K.
13. Suppose A is universal. Post’s set C is recursively enumerable and gen-
erative. Since C is recursively enumerable and A is universal, then C
is reducible to A. Then, since C is generative, so is A [by Theorem 9].
14. Post’s set K is obviously uniformly universal under the recursive func-
tion δ(x, y). Now, suppose A is universal. Then K is reducible to A
under some recursive function f (x). Then A must be uniformly uni-
versal under the function f (δ(x, y)), because for any numbers i and x,
x ∈ ωi iff δ(x, i) ∈ K, which is true iff f (δ(x, i)) ∈ A . Thus x ∈ ωi iff
f (δ(x, i)) ∈ A .

15. We are given that ϕ(x) is a generative function for B. Thus for every
number i,
(a) ϕ(i) ∈ ωi iff ϕ(i) ∈ B.
 Chapter 5. Some Recursion Theory 131

(1) Suppose ωi = N. Then by (a), ϕ(i) ∈ N iff ϕ(i) ∈ B, but ϕ(i) is in

N, so ϕ(i) ∈ B.
(2) Suppose ωi = ∅. Then by (a), ϕ(i) ∈ ∅ iff ϕ(i) ∈ B, but ϕ(i) ∈/ ∅,
so ϕ(i) ∈
/ B.

16. For all x, we have M (x, y) iff y ∈ A. We are letting t(y) be an iterative
function for the relation M (x, y), i.e. t is a recursive function such that
for every x and i: x ∈ ωt(i) iff M (x, i), which is true iff i ∈ A. Thus
x ∈ ωt(i) iff i ∈ A.
(1) If i ∈ A, then for every x, x ∈ ωt(i) , which means that ωt(i) = N.
(2) If i ∈
/ A, then for every x, x ∈ / ωt(i) , which means that ωt(i) is the
empty set ∅.
17. Suppose B is a generative set. We wish to show that B is universal.
Let ϕ(x) be a generative function for B. Consider any recursively enu-
merable set A which we wish to reduce to B. Let t(y) be a recursive
function related to A as in Problem 16. We assert that the function
ϕ(t(x)) reduces A to B.
(1) Suppose i ∈ A. Then ωt(i) = N [by Problem 16(1)]. Hence
ωϕ(t(i)) ∈ B [by Problem 15(a), taking t(i) for i].
(2) Suppose i ∈ / A. Then ωt(i) = ∅ [by Problem 16(2)]. Hence
ωϕ(t(i)) ∈
/ B [by Problem 15(b), taking t(i) for i].
Thus if i ∈ A, then ϕ(t(i)) ∈ B, and if i ∈
/ A, then ϕ(t(i)) ∈
/ B, or, what
is the same thing, if ϕ(t(i)) ∈ B then i ∈ A. Thus i ∈ A iff ϕ(t(i)) ∈ B,
and so the function ϕ(t(x)) reduces A to B.
18. Assume condition C1 .
(1) Suppose ωi = ∅. Obviously ∅ is disjoint from A, and so by C1 , the
number ϕ(i) ∈
/ A [also, ϕ(i) ∈/ ωi ].
(2) Suppose ωi = {ϕ(i)}. If ωi were disjoint from A, then by C1 , we
would have ϕ(i) ∈
/ ωi , and thus ϕ(i) ∈
/ ϕ(i)}, which is absurd. Thus
ϕ(i) ∈ A.

19. Define M (x, y, z) to be x ∈ ωy ∧ x = ϕ(z). By the strong recursion

theorem (Theorem 7), there is a recursive function t(y) such that
ωt(y) = {x : M (x, y, (t(y)))}
= {x : x ∈ ωy ∧ x = ϕ(t(y))},
= {x : x ∈ ωy } ∩ {x : x = ϕ(t(y))},
= ωy ∩ {ϕ(t(y))}.
132 Part II: Recursion Theory and Metamathematics

20. We want to show that for any number y, ϕ(t(y)) ∈ ωy iff ϕ(t(y)) ∈ A.
(1) Suppose ϕ(t(y)) ∈ ωy . Then ωy ∩ {ϕ(t(y))} = {ϕ(t(y))}. But also
ωy ∩ {ϕ(t(y))} = ωt(y) [by the Lemma]. Then ϕ(t(y)) ∈ A [by
condition C2 , taking t(y) for i, since A is weakly co-productive
under ϕ(x)]. Thus if ϕ(t(y)) ∈ ωy , then ϕ(t(y)) ∈ A.
(2) Suppose ϕ(t(y)) ∈ / ωy . Then ωy ∩ {ϕ(t(y))} = ∅. Again by the
Lemma, ωy ∩ {ϕ(t(y))} = ωt(y) . Thus ωt(y) = ∅. Again by condition
C2 , taking t(y) for i, ϕ(t(y)) ∈ / A. Thus if ϕ(t(y)) ∈ / ωy , then
ϕ(t(y)) ∈
/ A, or, equivalently if ϕ(t(y)) ∈ A, then ϕ(t(y)) ∈ ωy , and
thus ϕ(t(y)) is a generative function for A.

Answer to the Bonus Question. Of course not! A universal set is genera-

tive and a generative set A cannot be recursive, for if it were, its complement
A would be recursively enumerable, hence there would be a number n such
that n ∈ A iff n ∈ A, which is impossible.
 Chapter 6

Doubling Up

For my intended applications to the mathematical systems of the next

chapter, we will need double analogues of earlier results.

Double Recursion Theorems

In preparation for the double recursion theorems we will consider here, we

note that by the iteration theorem, there is a recursive function d such that
for all y and z, ωd(y,z) = {x : Ry (x, y, z)}. This function will play a key
role, and I shall call it the double diagonal function.

Theorem 1. [Weak Double Recursion Theorem] For any two recursively

enumerable relations M1 (x, y, z) and M2 (x, y, z), there are numbers a and
b such that:
(1) ωa = {x : M1 (x, a, b)};
(2) ωb = {x : M2 (x, a, b)}.

Problem 1. Prove Theorem 1. [Hint: Use the double diagonal function d

and consider the relations M1 (x, d(y, z), d(z, y)) and M2 (x, d(z, y), d(y, z)).]

The Strong Double Recursion Theorem

We now wish to show that in Theorem 1, the numbers a and b can be found
as recursive function of the indices i and j of the relations M1 (x, y, z) and
M2 (x, y, z). This is Theorem 3 below, which will soon be seen to be a
consequence of:
134 Part II: Recursion Theory and Metamathematics

Theorem 2. [Strong Double Recursion Theorem] For any recursively en-

umerable relations M1 (x, y1 , y2 , z1 , z2 ) and M2 (x, y1 , y2 , z1 , z2 ), there are
recursive functions ϕ1 (i, j) and ϕ2 (i, j) such that for all numbers i and j:
(1) ωϕ1 (i,j) = {x : M1 (x, i, j, ϕ1 (i, j), ϕ2 (i, j))};
(2) ωϕ2 (i,j) = {x : M2 (x, i, j, ϕ1 (i, j), ϕ2 (i, j))}.

Problem 2. Prove Theorem 2. [Hint: We again use the double diagonal

function d. By the iteration theorem there are recursive functions t1 (i, j)
and t2 (i, j) such that for all numbers i, j, x, y, z:
(1) Rt1 (i,j) (x, y, z) iff M1 (x, i, j, d(y, z), d(z, y));
(2) Rt2 (i,j) (x, y, z) iff M2 (x, i, j, d(z, y), d(y, z)).]

As an obvious corollary of Theorem 2 we have:

Theorem 3. For any two recursively enumerable relations M1 (x, y, z1 , z2 )

and M2 (x, y, z1 , z2 ), there are recursive functions ϕ1 (i, j) and ϕ2 (i, j) such
that for all numbers i and j:
(1) ωϕ1 (i,j) = {x : M1 (x, i, ϕ1 (i, j), ϕ2 (i, j))};
(2) ωϕ2 (i,j) = {x : M2 (x, j, ϕ1 (i, j), ϕ2 (i, j))}.

A special case of Theorem 3 is:

Theorem 4. [Strong Double Myhill Theorem] There are recursive functions

ϕ1 (i, j) and ϕ2 (i, j) such that for all numbers i and j:
(1) ωϕ1 (i,j) = {x : Ri (x, ϕ1 (i, j), ϕ2 (i, j))};
(2) ωϕ2 (i,j) = {x : Rj (x, ϕ1 (i, j), ϕ2 (i, j))}.

Problem 3. (a) Why is Theorem 3 a corollary of Theorem 2? (b) Why is

Theorem 4 a special case of Theorem 3?

Discussion

The proof we have given of the strong double recursion theorem involved
three applications of the iteration theorem. Actually, the theorem can be
proved using only one application of the iteration theorem, and that can
be done in more than one way.
Several workers in the field of recursion theory have found a way of
deriving the weak double recursion theorem as a consequence of the strong
(single) recursion theorem, and I have subsequently found a way of deriving
the strong double recursion theorem as a consequence of the strong (single)
recursion theorem.
 Chapter 6. Doubling Up 135

All this can be found in a more in-depth study of recursion theorems

in my book Recursion Theory for Metamathematics [Smullyan, 1993], in
which several other variants of the recursion theorem are also treated.

More Doubling Up

We shall say that an ordered pair (A1 , A2 ) is reducible to the ordered pair
(B1 , B2 ) under a recursive function f (x) if f (x) simultaneously reduces A1
to B1 and A2 to B2 . This means that A1 = f −1 (B1 ) and A2 = f −1 (B2 ); in
other words, it means that for every number x:
(1) If x ∈ A1 , then f (x) ∈ B1 ;
(2) If x ∈ A2 , then f (x) ∈ B2 ;
(3) If x ∈/ A1 ∪ A2 , then f (x) ∈
/ B1 ∪ B2 .
If (1) and (2) above both hold, without (3) necessarily holding, we say
that (A1 , A2 ) is semi-reducible to the ordered pair (B1 , B2 ) under the recur-
sive function f (x).
We shall call a pair (U1 , U2 ) doubly universal if every disjoint pair (A, B)
of recursively enumerable sets is reducible to (U1 , U2 ). We shall call a pair
(U1 , U2 ) semi-doubly universal if every disjoint pair (A, B) of recursively
enumerable sets is semi-reducible to (U1 , U2 ).
We shall call a disjoint pair (A, B) doubly generative under a recursive
function ϕ(x, y) if, for every disjoint pair (ωi , ωj ) of recursively enumerable
sets, the following holds:
(1) ϕ(i, j) ∈ A iff ϕ(i, j) ∈ ωi ;
(2) ϕ(i, j) ∈ B iff ϕ(i, j) ∈ ωj .
In this chapter we will prove that a pair (A, B) is doubly generative iff it
is doubly universal.
We shall call a disjoint pair (A, B) doubly co-productive under a recur-
sive function g(x, y) if for every disjoint pair of recursively enumerable sets
(ωi , ωj ), if ωi is disjoint from A and ωj is disjoint from B, then the number
g(i, j) is outside all four sets ωi , ωj , A and B.
It is obvious that if (A, B) is doubly generative, then (A, B) is doubly
co-productive. The converse is also true, but this is not trivial. The proof
requires something like the strong double recursion theorem.
All this is closely connected to the topic of effective inseparability, to
which we now turn.
136 Part II: Recursion Theory and Metamathematics

Effective Inseparability

First we must discuss a very useful principle: Consider two recursively enu-
merable sets A and B. They are both Σ1 and so there are Σ0 relations
R1 (x, y) and R2 (x, y) such that for all x,
(1) x ∈ A iff ∃yR1 (x, y);
(2) x ∈ B iff ∃yR2 (x, y).
Now define the sets A0 and B 0 by the following conditions:
x ∈ A0 iff ∃ y(R1 (x, y) ∧ (∀z ≤ y)(∼R2 (x, y)).
x ∈ B 0 iff ∃ y(R2 (x, y) ∧ (∀z ≤ y)(∼R1 (x, y)).
The sets A and B 0 are both recursively enumerable and disjoint.
0

Problem 4. Why are the sets A0 and B 0 disjoint?

We thus have:

Separation Principle: For any two recursively enumerable sets A and

B, there are disjoint recursively enumerable sets A0 and B 0 such that
A − B ⊆ A0 and B − A ⊆ B 0 .
A disjoint pair of sets (A, B) is said to be recursively inseparable if there
is no recursive superset of A disjoint from B, or, what is the same thing,
if there is no recursive superset of B disjoint from A (these are the same
because if C is a recursive superset of B disjoint from A, its complement C
is a recursive superset of A disjoint from B). A pair of sets (A, B) is said to
be recursively separable if the pair is not recursively inseparable (i.e. there
does exist a recursive superset of A disjoint from B or a recursive superset of
B disjoint from A). Clearly a pair of sets (A, B) that is recursively separable
must be disjoint.
Now, to say that (A, B) is recursively inseparable is equivalent to saying
that for any disjoint recursively enumerable supersets ωi and ωj of A and
B respectively, there is a number n outside both ωi and ωj [i.e. ωi is not
the complement of ωj ].

Problem 5. Why are the characterizations of recursive inseparability in

the two preceding paragraphs equivalent?
A disjoint pair (A, B) is called effectively inseparable if there is a recur-
sive function ζ(x, y) [the Greek letter ζ is read as “zeta”] — called an
 Chapter 6. Doubling Up 137

effective inseparability function for (A, B) — such that for any recursively
enumerable disjoint supersets ωi and ωj of A and B respectively, the num-
ber ζ(i, j) lies outside both ωi and ωj [ζ(i, j) is a witness, so to speak, that
ωi is not the complement of ωj ].
We shall call a disjoint pair (A, B) completely effectively inseparable if
there is a recursive function ζ(x, y) — called a completely effective inseparabil-
ity function for (A, B) — such that for any recursively enumerable supersets
ωi and ωj of A and B respectively, whether they are disjoint or not, the num-
ber ζ(i, j) is either in both ωi and ωj or in neither, in other words ζ(x, y) ∈ ωi
iff ζ(x, y) ∈ ωj . [If ωi and ωj happen to be disjoint, then ζ(x, y) is obvi-
ously outside both ωi and ωj , and so any completely effective inseparability
function for (A, B) is also an effective inseparability function for (A, B).]
We will later show that the non-trivial result that if (A, B) is effectively
inseparable, and A and B are both recursively enumerable, then (A, B) is
completely effectively inseparable. The only proof of this that I know of
uses the strong double recursion theorem.

Kleene’s Construction

There are several ways to construct a completely effectively inseparable pair

of recursively enumerable sets. The following way is due to Kleene [1952].
We shall call a recursive function k(x, y) a Kleene function for a disjoint
pair (A1 , A2 ) if, for all numbers i and j,
(1) If k(i, j) ∈ ωi − ωj , then k(i, j) ∈ A2 ;
(2) If k(i, j) ∈ ωj − ωi , then k(i, j) ∈ A1 .

Proposition 1. If k(x, y) is a Kleene function for a disjoint pair (A1 , A2 ),

then (A1 , A2 ) is completely effectively inseparable under k(x, y).

Problem 6. Prove Proposition 1.

We shall call a disjoint pair (A1 , A2 ) a Kleene pair if there is a Kleene
function for (A1 , A2 ).
By Proposition 1, any Kleene pair is completely effectively inseparable.

Theorem 5. There exists a Kleene pair (K1 , K2 ) of recursively enumerable

sets.
138 Part II: Recursion Theory and Metamathematics

Problem 7. Prove Theorem 5. [Hint: Use the pairing function δ(x, y). Let
A1 be the set of all numbers δ(i, j) such that δ(i, j) ∈ ωj and let A2 be
the set of all numbers δ(i, j) such that δ(i, j) ∈ ωi . Show that δ(x, y) is a
Kleene function for the pair (A1 − A2 , A2 − A1 ). Then use the separation
principle to get the recursively enumerable sets K1 and K2 .]
From Theorem 5 and Proposition 1, we have:

Theorem 6. There is a completely effectively inseparable pair of recursively

enumerable sets, to wit, the Kleene pair (K1 , K2 ).

Proposition 2. If (A1 , A2 ) is a Kleene pair and if (A1 , A2 ) is semi-

reducible to (B1 , B2 ) and B1 is disjoint from B2 , then (B1 , B2 ) is a
Kleene pair.

Problem 8. Prove Proposition 2. [Hint: Let f (x) be a function that

semi-reduces (A1 , A2 ) to (B1 , B2 ), and let k(x, y) be a Kleene function
for (A1 , A2 ). By the iteration theorem, there is a recursive function t(y)
such that ωt(y) = {x : f (x) ∈ ωy }. Show that ωt(y) = f −1 (ωy ). From the
functions k(x, y), f (x), t(y) construct a Kleene function for (B1 , B2 ).]

Theorem 7. Every disjoint semi-doubly-universal pair is completely effec-

tively inseparable.

Problem 9. Why is Theorem 7 true?

Doubly Generative Pairs

We recall that a disjoint pair (A, B) is called doubly generative if there

is a recursive function ϕ(x, y) — called a doubly generative function
for (A, B) — such that, for every disjoint pair (ωi , ωj ), the following
holds:
(1) ϕ(i, j) ∈ A iff ϕ(i, j) ∈ ωi ;
(2) ϕ(i, j) ∈ B iff ϕ(i, j) ∈ ωj .

Problem 10. Prove that if (A, B) is called doubly generative, then A and
B are each generative sets.
The next theorem is vital:

Theorem 8. If (A, B) is completely effectively inseparable and A and B

are both recursively enumerable, then (A, B) is doubly generative.
 Chapter 6. Doubling Up 139

Problem 11. Prove Theorem 8. [Hint: (A, B) is completely effectively

inseparable under some recursive function ζ(x, y). By the iteration theorem,
there are recursive functions t1 (n), t2 (n) such that for all x and n:
(1) ωt1 (n) = {x : x ∈ ωn ∨ x ∈ A};
(2) ωt2 (n) = {x : x ∈ ωn ∨ x ∈ B};
Let ϕ(i, j) = ζ(t1 (j), t2 (i)). Show that (A, B) is doubly generative under
the function ϕ(i, j).]
Since the Kleene pair (K1 , K2 ) is completely effectively inseparable and
K1 and K2 are both recursively enumerable, it follows from Theorem 8
that the pair (K1 , K2 ) is doubly generative. Thus there exists a doubly
generative pair of recursively enumerable sets.

Doubly Generative, Doubly Universal

Suppose (A1 , A2 ) is a disjoint pair which is doubly generative under the

function g(x, y). Then, for all numbers i and j, the following three condi-
tions hold:
C1 : If ωi = N and ωj = ∅, then g(i, j) ∈ A1 ;
C2 : If ωi = ∅ and ωj = N, then g(i, j) ∈ A2 ;
C3 : If ωi = ∅ and ωj = ∅, then g(i, j) ∈
/ A1 ∪ A2 .

Problem 12. Prove this.

Problem 13. Suppose (A1 , A2 ) is a disjoint pair and that g(x, y) is a

recursive function such that conditions C1 and C2 above hold. Prove that for
any disjoint pair (B1 , B2 ) of recursively enumerable sets there are recursive
functions t1 (x), t2 (x) such that:
(1) (B1 , B2 ) is semi-reducible to (A1 , A2 ) under g(t1 (x), t2 (x));
(2) If condition C3 also holds, then (B1 , B2 ) is reducible to (A1 , A2 ) under
g(t1 (x), t2 (x)). [Hint: Use the result of Problem 16 of the previous
chapter.]
From Problems 12 and 13 immediately follows:

Theorem 9. Every doubly generative pair is doubly universal.

We now see that for a disjoint pair (A, B), being completely effectively
inseparable, being doubly generative, and being doubly universal, are all
one and the same thing. Now to complete the picture.
140 Part II: Recursion Theory and Metamathematics

Double Co-productivity

We recall that a disjoint pair (A, B) is defined to be doubly co-productive

under a recursive function g(x, y) if for all i and j, if ωi is disjoint from ωj ,
and ωi is disjoint from A and ωj is disjoint from B, then g(i, j) lies outside
all the four sets ωi , ωj , A and B.
We claim that if (A, B) is doubly co-productive under a recursive
function g(x, y), then, for all numbers i and j, the following conditions
hold:
D0 : If ωi = ωj = ∅, then g(i, j) ∈
/ A ∪ B;
D1 : If ωi = {g(i, j)} and ωj = ∅, then g(i, j) ∈ A;
D2 : If ωi = ∅ and ωj = {g(i, j)}, then g(i, j) ∈ B.

Problem 14. Prove that if (A, B) is doubly co-productive under the recur-
sive function g(x, y), then conditions D0 , D1 and D2 hold.
We will say that (A, B) is called weakly doubly co-productive under the
recursive function g(x, y) if conditions D0 , D1 and D2 hold.
We now wish to prove:

Theorem 10. If (A, B) is doubly co-productive, or even weakly doubly co-

productive, under a recursive function g(x, y), then (A, B) is doubly gener-
ative (and hence also doubly universal).
For the proof of Theorem 10 we need:

Lemma. For any recursive function g(x, y), there are recursive functions
t1 (i, j), t2 (i, j) such that for all numbers all i and j,
(1) ωt1 (i,j) = ωi ∩ {g(t1 (i, j), t2 (i, j))};
(2) ωt2 (i,j) = ωj ∩ {g(t1 (i, j), t2 (i, j))}.

To prove this, we will need the strong double recursion theorem, or

rather its corollary, Theorem 3.

Problem 15. Prove the above Lemma. [Hint: Let M1 (x, y, z1 , z2 ) be

the recursively enumerable relation x ∈ ωy ∧ x = g(z1 , z2 ), and let
M2 (x, y, z1 , z2 ) be the same relation. Then use Theorem 3.]

Problem 16. Now prove Theorem 10. [Hint: Suppose (A, B) is weakly dou-
bly co-productive under g(x, y). By the lemma, there are recursive functions
t1 (x, y), t2 (x, y) such that for all i and j,
 Chapter 6. Doubling Up 141

(1) ωt1 (i,j) = ωi ∩ {g(t1 (y1, y2 ), t2 (y1, y2 ))};

(2) ωt2 (i,j) = ωj ∩ {g(t1 (y1, y2 ), t2 (y1, y2 ))}.
Let ϕ(x, y) = g(t1 (x, y)t2 (x, y)), and show that (A, B) is doubly generative
under ϕ(x, y).]
Next we wish to prove:

Theorem 11. If (A, B) is effectively inseparable and A and B are both

recursively enumerable, then (A, B) is doubly generative (and hence doubly
universal).

Problem 17. Prove Theorem 11. [Hint: (a) First prove that for any recur-
sively enumerable set A there is a recursive function t(i) such that for all
i, ωt(i) = ωi ∪ A. (b) Now suppose that (A, B) is effectively inseparable
under g(x, y) and that A and B are both recursively enumerable sets. Let
t1 (i), t2 (i) be recursive functions such that for all i, ωt1 (i) = ωi ∪ A and
ωt2 (i) = ωi ∪ B. Let g(x, y) = ζ(t1 (y), t2 (x)) and show that (A, B) is doubly
generative under the function g(x, y).]

Solutions to the Problems of Chapter 6

1. Let m be an index of the relation M1 (x, d(y, z), d(z, y)), and let n be
an index of M2 (x, d(z, y), d(y, z)). Then:
(1) Rm (x, y, z) iff M1 (x, d(y, z), d(z, y)),
(2) Rn (x, y, z) iff M2 (x, d(z, y), d(y, z)).
In (1), take m for y and n for z, and in (2), take n for y and m for z.
As a result, we have:
(10 ) Rm (x, m, n) iff M1 (x, d(m, n), d(n, m)),
(20 ) Rn (x, n, m) iff M2 (x, d(m, n), d(n, m)).
Now, ωd(m,n) = {x : Rm (x, m, n)} = {x : M1 (x, d(m, n), d(n, m))} [by
(10 )]; and
ωd(n,m) = {x : Rn (x, n, m)} = {x : M2 (x, d(m, n), d(n, m))} [by (20 )].
We thus take a = d(m, n) and b = d(n, m), and have:
ωa = {x : M1 (x, a, b)} and ωb = {x : M2 (x, a, b)}.
2. We define:
ϕ1 (i, j) = d(t1 (i, j), t2 (i, j)) and
ϕ2 (i, j) = d(t2 (i, j), t1 (i, j)),
142 Part II: Recursion Theory and Metamathematics

where t1 (i, j) and t2 (i, j) are the functions mentioned in the hint and
d(x, y) is the double diagonal function. We will see that these two func-
tions work.
By the definitions of t1 and t2 , we know that for all numbers
i, j, x, y, z,
(1) Rt1 (i,j) (x, y, z) iff M1 (x, i, j, d(y, z), d(z, y));
(2) Rt2 (i,j) (x, y, z) iff M2 (x, i, j, d(z, y), d(y, z)).
In (1), we take t1 (i, j) for y and t2 (i, j) for z, and in (2), we take t2 (i, j)
for y and t1 (i, j) for z, obtaining:
(10 ) Rt1 (i,j) (x, t1 (i, j), t2 (i, j)) iff M1 (x, i, j, ϕ1 (i1 , i2 ), ϕ2 (i1 , i2 ));
(20 ) Rt2 (i,j) (x, t2 (i, j), t1 (i, j)) iff M2 (x, i, j, ϕ1 (i1 , i2 ), ϕ2 (i1 , i2 )).
Also:
(3) ωϕ1 (i,j) = ωd(t1(i,j), t2(i,j) ) = {x : Rt1 (i,j) (x, t1(i,j), t2(i,j) )};
(4) ωϕ2 (i,j) = ωd(t2(i,j), t1(i,j) ) = {x : Rt2 (i,j) (x, t2(i,j), t1(i,j) )}.

From (3) and (10 ), we have ωϕ1 (i,j) = {x : M1 (i, j, ϕ1 (i, j), ϕ2 (i, j))}.
From (4) and (20 ), we have ωϕ2 (i,j) = {x : M2 (i, j, ϕ1 (i, j), ϕ2 (i, j))}.
3. (a) Given the recursively enumerable relations M1 (x, y, z1 , z2 ) and
M2 (x, y, z1 , z2 ) define two new relations M10 (x, y1 , y2 , z1 , z2 ) iff
M1 (x, y1 , z1 , z2 ) and M20 (x, y1 , y2 , z1 , z2 ) iff M2 (x, y2 , z1 , z2 ). Then
apply Theorem 2 to the relations M10 and M20 .
(b) This follows from Theorem 3 by taking M1 (x, y, z1 , z2 ) and
M2 (x, y, z1 , z2 ) to both be the relation Ry (x, z1 , z2 ).
4. If x ∈ A0 and x ∈ B 0 , we get the following contradiction: There must
be numbers n and m such that:
(1) R1 (x, n).
(2) R2 (x, m).
(3) If R2 (x, m), then m > n [since (∀ z ≤ n) ∼R2 (x, z)].
(4) If R1 (x, n), then n > m [since (∀ z ≤ m) ∼R1 (x, z)].
From (1) and (4) it follows that n > m. From (2) and (3) it follows
that it is also true that m > n, which is impossible.
5. We are to show that the following two characterizations of the recursive
inseparability of the disjoint pair of sets (A, B) are equivalent:
C1 : There is no recursive superset of A disjoint from B.
C2 : For any disjoint recursively enumerable supersets ωi of A and ωj
of B, ωi cannot be the complement of ωj .
 Chapter 6. Doubling Up 143

To show that C1 implies C2 , assume C1 . Now suppose ωi , ωj are disjoint

supersets of A and B respectively. If ωi were the complement of ωj then
ωi would be a recursive superset of A disjoint from B, contrary to C1 .
Therefore ωi is not the complement of ωj , which proves C2 .
Conversely, suppose C2 holds. Let A0 be any superset of A disjoint
from B. We are to show that A0 cannot be recursive and hence C1
holds. Well, if A0 were a recursive set ωi , its complement ωi would be a
recursively enumerable superset ωj of B, disjoint from A, which would
be contrary to C2 . Thus A0 is not recursive.
6. We are given that k(x, y) is a recursive function such that for all
i and j,
(1) k(i, j) ∈ ωi − ωj implies k(i, j) ∈ A2 ;
(2) k(i, j) ∈ ωj − ωi implies k(i, j) ∈ A1 .
Now consider any recursively enumerable superset ωi of A1 and
any recursively enumerable superset ωj of A2 . We are to show that
k(i, j) ∈ ωi iff k(i, j) ∈ ωj [and thus that (A1 , A2 ) is completely effec-
tively inseparable under k(x, y)].
If k(i, j) ∈ ωi − ωj , then k(i, j) ∈ A2 [by (1)], and hence k(i, j) ∈ ωj
[since A2 ⊆ ωj ], but no number in ωi − ωj could possibly be in ωj , and
therefore k(i, j) ∈/ ωi − ωj . By a symmetric argument, k(i, j) ∈/ ωj − ωi .
Consequently, k(i, j) ∈ ωi iff k(i, j) ∈ ωj .
This completes the proof.
7. Since δ(i, j) ∈ ωj iff δ(i, j) ∈ A1 and δ(i, j) ∈ ωi iff δ(i, j) ∈ A2 ,
then δ(i, j) ∈ ωj − ωi iff δ(i, j) ∈ A1 − A2 and δ(i, j) ∈ ωi − ωj iff
δ(i, j) ∈ A2 −A1 . Thus δ(x, y) is a Kleene function for (A1 −A2 , A2 −A1 ).
By the separation principle, there are recursively enumerable sets K1
and K2 such that A1 − A2 ⊆ K1 and A2 − A1 ⊆ K2 . Since δ(x, y)
is a Kleene function for (A1 − A2 , A2 − A1 ), and A1 − A2 ⊆ K1 and
A2 − A1 ⊆ K2 , it follows that δ(x, y) is a Kleene function for the
recursively enumerable pair (K1 , K2 ). [In general, it is easily verified
that if ζ(x, y) is a Kleene function for a pair (B1 , B2 ), and if B10 , B20 are
disjoint supersets of B1 B2 respectively, then ζ(x, y) must be a Kleene
function for (B10 , B20 ).]
8. As mentioned in the hint: Let f (x) be the function that semi-reduces
(A1 , A2 ) to (B 1 , B2 ), and let t(y) be a recursive function given by the
iteration theorem such that for all x and y, x ∈ ωt(y) iff f (x) ∈ ωy [thus
ωt(y) = {x : f (x) ∈ ωy }]. Also f (x) ∈ ωy iff x ∈ f −1 (ωy ) and so x ∈ ωt(y)
144 Part II: Recursion Theory and Metamathematics

iff x ∈ f −1 (ωy ). Let k be a Kleene function for the (disjoint) Kleene pair
(A1 A2 ). We now let k 0 (x, y) be the function f (k(t(x), t(y))), and we will
show that k 0 (x, y) is a Kleene function for the pair (B1 , B2 ). [At this
point, the reader might like to try showing this without reading further.]
First we show that if k 0 (i, j) ∈ ωj − ωi , then k 0 (i, j) ∈ B1 .
Well, suppose k 0 (i, j) ∈ ωj −ωi . Thus f (k(t(i), t(j))) ∈ ωj −ωi , and so
k(t(i), t(j)) ∈ f −1 (ωj − ωi ).
But
f −1 (ωj − ωi ) = f −1 (ωj ) − f −1 (ωi ) = ωt(j) − ωt(i)
[since f −1 (ωj ) = ωt(j) and f −1 (ωi ) = ωt(i) , as we have shown]. Thus
k(t(i), t(j)) ∈ ωt(j) − ωt(i) ,
so that k(t(i), t(j)) ∈ A1 [since k(x, y) is a Kleene function for (A1 , A2 )],
and therefore f (k(t(i), t(j))) ∈ B1 [since (A1 , A2 ) is semi-reducible to
(B1 , B2 ) under f (x)]. Thus k 0 (i, j) ∈ B1 .
This proves that if k 0 (i, j) ∈ ωj − ωi , then k 0 (i, j) ∈ B1 .
By a similar argument, if k 0 (i, j) ∈ ωi − ωj , then k 0 (i, j) ∈ B2 . Thus
0
k (x, y) is a Kleene function for (B1 , B2 ).
9. If (A, B) is semi-doubly universal, then by definition the Kleene pair
(K1 , K2 ) is semi-reducible to (A, B). Hence [by Proposition 2] (A, B) is
in turn a Kleene pair, and therefore completely effectively inseparable
[by Proposition 1].
10. Suppose ϕ(x, y) is a doubly generative function for (A, B). Let c be
any index of the empty set, and thus for every number i, the set ωi is
disjoint from ωc . Thus ϕ(x, c) [as a function of x] is a generative func-
tion for A, because for every number i, ωi is disjoint from ωc , so that
ϕ(i, c) ∈ A iff ϕ(i, c) ∈ ωi [by (1)]. Similarly, ϕ(c, x) [as a function of x]
is a doubly generative function for the set B.
11. The set {x : x ∈ ωn ∨ x ∈ A} is simply ωn ∪ A, and the set
{x : x ∈ ωn ∨ x ∈ B} is simply ωn ∪ B. Using the t1 and t2 recom-
mended in the hint, we have:
(1) ωt1 (n) = ωn ∪ A;
(2) ωt2 (n) = ωn ∪ B.
Thus A and ωn are both subsets of ωt1 (n) for all n, and B and ωn are
both subsets of ωt2 (n) for all n. Also as in the hint, we take
ϕ(x, y) = ζ(t1 (y), t2 (x)).
 Chapter 6. Doubling Up 145

Given two numbers i and j, let k = ϕ(i, j). Thus k = ζ(t1 (j), t2 (i)).
Since (A, B) is completely effectively inseparable under ζ(x, y) and
A ⊆ ωt1 (j) and B ⊆ ωt2 (i) [remember, A ⊆ ωt1 (n) and B ⊆ ωt2 (n)
for all n], and ζ(x, y) is a completely effective inseparability function
for (A, B), then
(a) k ∈ ωt1 (j) iff k ∈ ωt2 (i) .
Now, consider any disjoint pair (ωi , ωj ). We are to show that k ∈ A iff
k ∈ ωi and k ∈ B iff k ∈ ωj , and thus that (A, B) is doubly generative
under ϕ(x, y).
Suppose that k ∈ A. Then k ∈ ωt1 (j) [since A ⊆ ωt1 (n) for every n].
Hence k ∈ ωt2 (i) [by (a)]. Thus k ∈ ωi ∪ B [since ωt2 (i) = ωi ∪ B]. But
k ∈
/ B, since B is disjoint from A. Thus k ∈ ωi . This proves that if
k ∈ A then k ∈ ωi .
Conversely, suppose that k ∈ ωi . Then k ∈ ωt2 (i) [since ωi ⊆ ωt2 (i) ].
Thus k ∈ ωt1 (j) [by (a)], and since ωt1 (j) = ωj ∪ A , then k ∈ ωj ∪ A.
But since ωj is disjoint from ωi we have the k ∈/ ωj , and so k ∈ A. Thus
if k ∈ ωi , then k ∈ A.
This proves that k ∈ A iff k ∈ ωi . The proof that k ∈ B iff k ∈ ωj is
similar.
12. We are given that (A1 , A2 ) is doubly generative under g(x, y).
(a) Suppose that ωi = N and ωj = ∅. Of course ωi is disjoint from ωj , so
that g(i, j) ∈ N iff g(i, j) ∈ A1 [and also g(i, j) ∈ ∅ iff g(i, j) ∈ A2 ],
but since g(i, j) ∈ N, it follows that g(i, j) ∈ A1 .
(b) Similarly, if ωi = ∅ and ωj = N, then g(i, j) ∈ A2 .
(c) If ωi = ∅ and ωj = ∅, since ∅ is obviously disjoint from ∅, then
g(i, j) ∈ A1 iff g(i, j) ∈ ∅, and g(i, j) ∈ ∅ iff g(i, j) ∈ A2 . But since
g(i, j) ∈
/ ∅, then g(i, j) ∈/ A1 and g(i, j) ∈
/ A2 .
13. By Problem 16 of the last chapter, for any recursively enumerable set
B there is a recursive function t(y) such that for all numbers i, if i ∈ B
then ωt(i) = N, and if i ∈
/ B, then ωt(i) = ∅.
Now, let (B1 , B2 ) be any disjoint pair of recursively enumer-
able sets. There are recursive functions t1 (i), t2 (i) such that for any
number i,
(1) If i ∈ B1 , then ωt1 (i) = N;
(2) If i ∈
/ B1 , then ωt1 (i) = ∅;
146 Part II: Recursion Theory and Metamathematics

(3) If i ∈ B2 , then ωt2 (i) = N;

(4) If i ∈
/ B2 , then ωt2 (i) = ∅.
Now assume that conditions C1 and C2 hold between the function
g(x, y) and the pair (A1 , A2 ).
(a) Suppose i ∈ B1 . Then ωt1 (i) = N [by (1)]. Since B2 is disjoint from
B1 , then i ∈ / B2 , and hence ωt2 (i) = ∅ [by (4)], and thus the pair
(ωt1 (i) , ωt2 (i) ) is (N, ∅), and so g(t1 (i), t2 (i)) ∈ A1 . This proves that
i ∈ B1 implies g(t1 (i), t2 (i)) ∈ A1 .
(b) The proof that i ∈ B2 implies g(t1 (i), t2 (i)) ∈ A2 is similar.
Thus (B1 , B2 ) is semi-reducible to (A1 , A2 ) under the function
g(t1 (x), t2 (x)).
(c) Suppose that condition C3 also holds. Now suppose that i ∈ / B1
and i ∈ / B2 . Then by (2) and (4), ωt1 (i) = ∅ and ωt2 (i) = ∅, and
therefore g(t1 (i), t2 (i)) ∈ / A1 and g(t1 (i), t2 (i)) ∈/ A2 [by C3 ]. Thus
g(t1 (x), t2 (x)) reduces (B1 , B2 ) to (A1 , A2 ).
14. We are given that (A, B) is doubly co-productive under the recursive
function g(x, y).
D0 : This is obvious!
D1 : Suppose ωi = {g(i, j)} and ωj = ∅. We are to show that
g(i, j) ∈ A. Consider the following conditions:
(1) ωi is disjoint from ωj ;
(2) ωj is disjoint from B;
(3) ωi is disjoint from A.
If these three conditions all hold, then g(i, j) would be outside all four
sets A, B, ωi , ωj [by the definition of double co-productivity]. Hence
g(i, j) would be outside ωi , hence outside {g(i, j)}, which is not pos-
sible. Therefore it cannot be that all three of these conditions hold,
but (1) and (2) do hold, therefore (3) doesn’t hold, which means that
ωi is not disjoint from A. Thus g(i, j) is not disjoint from A, so that
g(i, j) ∈ A This proves D1 .
D2 : The proof of D2 is symmetric to that of D1 .
15. By Theorem 3 there are recursive functions t1 (i, j) and t2 (i, j) such
that for all numbers i and j,
(1) ωt1 (i,j) = {x : M1 (x, i, t1 (i, j), t2 (i, j))};
(2) ωt2 (i,j) = {x : M2 (x, j, t1 (i, j), t2 (i, j))}.
 Chapter 6. Doubling Up 147

Thus,
(10 ) ωt1 (i,j) = {x : x ∈ ωi ∧ x = g(t1 (i, j), t2 (i, j))}
= ωi ∩ { g(t1 (i, j), t2 (i, j))};
(20 ) ωt2 (i,j) = {x : x ∈ ωj ∧ x = g(t1 (i, j), t2 (i, j))}
= ωj ∩ {g(t1 (i, j), t2 (i, j))}.
16. Consider any disjoint pair (ωi , ωj ). Let ϕ(i, j) = g(t1 (i, j), t2 (i, j)),
where t1 (y1, y2 ), t2 (y1, y2 ) are as given by the lemma. Then by the
lemma,
(a) ωt1 (i,j) = ωi ∩ {ϕ(i, j)};
(b) ωt2 (i,j) = ωj ∩ {ϕ(i, j)}.
We first wish to show:
(1) If ϕ(i, j) ∈ ωi , then ϕ(i, j) ∈ A;
(2) If ϕ(i, j) ∈ ωj , then ϕ(i, j) ∈ B.
To show (1), suppose that ϕ(i, j) ∈ ωi . Then ωi ∩ {ϕ(i, j)} = {ϕ(i, j)}.
Hence, by (a), ωt1 (i,j) is the set {ϕ(i, j)}.
Since ϕ(i, j) ∈ ωi and ωi is disjoint from ωj , then ϕ(i, j) ∈ / ωj ,
and therefore ωj ∩ {ϕ(i, j)} = ∅. Hence, by (b), ωt2 (i,j) = ∅. Thus
ωt1 (i,j) is the unit set {ϕ(i, j)} and ωt2 (i,j) is the empty set ∅, and so
by condition D1 of the definition of “weakly doubly co-productive,”
g(t1 (i, j), t2 (i, j)) ∈ A. Thus ϕ(i, j) ∈ A, which proves (1).
The proof of (2) is similar (using D2 instead of D1 ).
It remains to prove the converses of (1) and (2). To this end, we first
establish:
(3) If ϕ(i, j) ∈ A ∪ B, then ϕ(i, j) ∈ ωi ∪ ωj .
We will prove the equivalent propositions that if ϕ(i, j) ∈
/ ωi ∪ ωj , then
ϕ(i, j) ∈
/ A ∪ B.
Well, suppose ϕ(i, j) ∈ / ωi ∪ ωj . Then ϕ(i, j) ∈ / ωi and ϕ(i, j) ∈
/ ωj .
Since ϕ(i, j) ∈/ ωi , then ωi ∩ {ϕ(i, j)} = ∅. Then by (a), ωt1 (i,j) = ∅.
Similarly, since ϕ(i, j) ∈/ ωj , then ωt2 (i,j) = ∅. Hence by condition D0
of the definition of “weakly doubly productive,”
g(t1 (i, j), t2 (i, j)) ∈
/ A ∪ B.
Thus ϕ(i, j) ∈/ A ∪ B. This proves that if ϕ(i, j) ∈ / ωi ∪ ωj , then
ϕ(i, j) ∈
/ A ∪ B, and therefore if ϕ(i, j) ∈ A ∪ B, then ϕ(i, j) ∈ ωi ∪ ωj ,
which is (3).
148 Part II: Recursion Theory and Metamathematics

Now, to prove the converse of (1), suppose ϕ(i, j) ∈ A. Then

ϕ(i, j) ∈ A ∪ B, so that ϕ(i, j) ∈ ωi ∪ ωj . Since ϕ(i, j) ∈ A, then
ϕ(i, j) ∈/ B [because B is disjoint from A]. Therefore ϕ(i, j) ∈ / ωj
[because if ϕ(i, j) ∈ ωj , then by (2) ϕ(i, j) would be in B, which it
isn’t]. Since ϕ(i, j) ∈ ωi ∪ ωj but ϕ(i, j) ∈ / ωj , then ϕ(i, j) ∈ ωi . Thus
if ϕ(i, j) ∈ A, then ϕ(i, j) ∈ ωi , which is the converse of (1). The proof
of the converse of (2) is similar.
17. (a) Suppose that A is recursively enumerable. Let M (x, y) be the
recursively enumerable relation x ∈ ωy ∨ x ∈ A. By the itera-
tion theorem, there is a recursive function t(i) such that for all i,
ωt(i) = {x : M (x, i)}. Thus
ωt(i) = {x : x ∈ ωi ∨ x ∈ A} = ωi ∪ A.
(b) Suppose (A, B) is effectively inseparable under ζ(x, y) and A and
B are both recursively enumerable. Now let (ωi , ωj ) be a disjoint
pair such that ωi is disjoint from A and ωj is disjoint from B. Since
ωi is disjoint from both ωj and A, then ωi is disjoint from ωj ∪ A.
Since B is disjoint from both ωj and A, then B is disjoint from
ωj ∪ A. Thus ωi and B are both disjoint from ωj ∪ A, so that ωi ∪ B
is disjoint from ωj ∪ A.
Thus, ωj ∪A and ωi ∪B are disjoint supersets of A and B respec-
tively. By (a), there are recursive functions t1 and t2 such that
ωt1 (j) = ωj ∪ A and ωt2 (i) = ωi ∪ B. Therefore ωt1 (j) , ωt2 (i) are dis-
joint supersets of A and B respectively. Because the pair (A, B) is
assumed to be effectively inseparable under ζ(x, y), ζ(t1 (j), t2 (i))
[which we define g(i, j) to be] lies outside both ωt1 (j) and ωt2 (i) .
But ωt1 (j) = ωj ∪ A and ωt2 (i) = ωi ∪ B, which means g(i, j) must
of course lie outside both ωi and ωj (and A and B as well). Thus
we see that (A, B) is doubly generative under the function g(x, y).
 Chapter 7

Metamathematical Applications

In The Beginner’s Guide [Smullyan, 2014] we studied the system of Peano

Arithmetic. Some of the results regarding that system, as well as those
regarding other formal systems, depend on the use of the logical connectives
and quantifiers, but not all do. Those that do not can be neatly generalized
to more abstract systems called representation systems, which I introduced
in the Theory of Formal Systems [Smullyan, 1961]. In this chapter, I gen-
eralize further using systems I call simple systems, in terms of which it is
shown that many results of Peano Arithmetic and other systems are essen-
tially results of recursion theory in a new dress.
In what follows, number shall mean natural number, and function shall
mean function whose arguments and values are natural numbers. Any argu-
ment of a function ranges over the whole set of natural numbers.

I. Simple Systems

By a simple system S I shall mean a pair (P, R) of disjoint sets, together

with a function neg(x) and a function r(x, y) satisfying condition C1 below.
For any numbers h and n, we abbreviate neg(h) by h0 , and r(h, n) by
h(n), or sometimes by rh (n).
C1 : For any numbers h and n, h0 (n) ∈ P iff h(n) ∈ R, and h0 (n) ∈ R iff
h(n) ∈ P .
150 Part II: Recursion Theory and Metamathematics

Intended Applications

With any first-order system F of arithmetic we associate a simple system

S as follows: We arrange all the sentences of F in an infinite sequence
S1 , S2 , . . . , Sn , . . . according to the magnitude of their Gödel numbers.
We refer to n as the index of the sentence Sn . We take one particu-
lar variable x and arrange all formulas with x as their only free vari-
able in an infinite sequence ϕ1 (x), ϕ2 (x), . . . , ϕn (x), . . . according to the
magnitude of their Gödel numbers. We refer to n as the index of ϕn (x).
We arrange all formulas with the two variables x and y as their only
free variables in an infinite sequence ψ1 (x, y), ψ2 (x, y), . . . , ψn (x, y), . . . ,
according to the magnitudes of their Gödel numbers, and for any num-
bers n, x and y, we define s(n, x, y) to be the index of the sentence
ψn (x̄, ȳ).
For the associated simple system S of F, we take P to be the set of
indices of the provable sentences, and take R to be the set of indices
of the refutable sentences. For any numbers h and n, we take r(h, n)
to be the index of the sentence ϕh (n̄) [i.e. the sentence resulting from
substituting the number n̄ (the name of n) for all free occurrences of
x in ϕh (x)]. We take h0 [which is the abbreviation of neg(h)] to be the
index of the formula ∼ϕh (x) Of course ϕh0 (x) is provable (refutable)
iff ϕh (x) is refutable (provable). And so the items (P, R), neg(h), and
r(x, y) constitute a simple system. Consequently, whatever we prove for
simple systems in general will hold for Peano Arithmetic and other
systems.
Returning to simple systems in general, in accordance with intended
applications, we might call members of P provable numbers, and members
of R refutable numbers. We will call a number decidable if it is in either P
or R, and call it undecidable otherwise. We call S complete if every number
is decidable, otherwise incomplete.
We call two numbers equivalent if whenever either one is in P , so is the
other, and if either one is in R, so is the other.
We will refer to the function r(x, y) as the representation function for
sets of the system S, and we say that a number h represents the set A of
all numbers n such that h(n) ∈ P . Thus, to say that h represents A is to
say that for all n, h(n) ∈ P iff n ∈ A. We call A representable in S if some
number h represents it.
 Chapter 7. Metamathematical Applications 151

We say that h completely represents A iff h represents A and h0 repre-

sents the complement A of A.
We say that h defines A if, for all numbers n:
(1) n ∈ A implies h(n) ∈ P ;
(2) n ∈ A implies h(n) ∈ R.

We say that A is definable (completely representable) if some number h

defines (completely represents) A.
We say that h contra-represents A if for all n the number h(n) ∈ R iff
n ∈ A.

Problem 1. (a) Show that h defines A iff h completely represents A.

(b) Show that h contra-represents A iff h0 represents A, so that A is contra-
representable iff it is representable.
For any function f (x) and set A, by f −1 (A) is meant the set of all n
such that f (n) ∈ A. Thus n ∈ f −1 (A) iff f (n) ∈ A.
Recall that rh (n) is r(h, n) and that rh (n) = h(n), where r(x, y) is the
representation function for sets.

Problem 2. Show that h represents A iff A = rh−1 (P ). Show also that h

contra-represents A iff A = rh−1 (R).

Problem 3. Prove that if some set A is represented in S and its complement

A is not, then S is incomplete.
For any set A, by A* we shall mean the set of all numbers h such that
h(h) ∈ A.

Problem 4. Prove that neither the complement P ∗ of P ∗ , nor the comple-

ment R∗ of R∗ , is representable in S.
By the diagonalization of a number h we shall mean the number h(h).
By the skew diagonalization of h we shall mean the number h(h0 ).
It follow from Problems 3 and 4 that if either P ∗ or R∗ is repre-
sentable in S, then S is incomplete. The following problem yields a sharper
result.

Problem 5. Prove the following:

(a) If h represents R∗ , then its diagonalization h(h) is undecidable.
152 Part II: Recursion Theory and Metamathematics

(b) If h represents P ∗ , then its skew-diagonalization h(h0 ) is undecidable.

The following problem yields still stronger results.

Problem 6. Prove the following:

(a) If h represents some superset of R∗ disjoint from P ∗ , then h(h) is
undecidable.
(b) If h represents some superset of P ∗ disjoint from R∗ , then h(h0 ) is
undecidable.
Why are these results stronger than those of Problem 5?

Discussion

Gödel’s original incompleteness proof essentially involved representing the

set P ∗ . [In the Theory of Formal Systems [Smullyan, 1961], I pointed out
that incompleteness also follows if R∗ is representable.] To represent P ∗
Gödel had to assume that the system was ω-consistent. It was John Barkley
Rosser [1936] who was able to eliminate the hypothesis of ω-consistency by
representing, not R∗ , but some superset of R∗ disjoint from P ∗ .

The Diagonal Function d(x)

We define d(x) to be r(x, x), which is x(x). We note that for any set A the
set A∗ is d−1 (A).

Problem 7. Prove that for any sets A and B:

(a) If A ⊆ B, then f −1 (A) ⊆ f −1 (B).
(b) If A is disjoint from B, then f −1 (A) is disjoint from f −1 (B).
(c) f −1 (A) is the complement of f −1 (A).
In particular, taking d(x) for f (x),
(a1 ) If A ⊆ B, then A∗ ⊆ B ∗ .
(a2 ) If A is disjoint from B, then A∗ is disjoint from B ∗ .
∗
(a3 ) A = A∗ .

Admissible Functions

We shall call a function f (x) admissible if for every number h there

is a number k such that for every n the number k(n) is equivalent
to h(f (n)).
 Chapter 7. Metamathematical Applications 153

Separation

Given a disjoint pair (A, B), we say that a number h separates A from B
(in S) if h represents a superset of A and contra-represents some superset
of B.
We say that h exactly separates A from B if h both represents A itself
and contra-represents B.
The following problem is key.

Problem 8. Suppose f (x) is admissible. Prove the following:

(a) If A is representable, so is f −1 (A). If A is contra-representable, so is
f −1 (A).
(b) If a disjoint pair (A, B) is exactly separable in S, then so is the pair
(f −1 (A), f −1 (B)).

Normal Systems

We shall call S a normal system if the diagonal function d(x) is admissible.

This means that for every number h there is a number — which we shall
denote h# — such that for all n the numbers h# (n) is equivalent to h(d(n)).

Problem 9. Suppose S is normal. If A is representable in S, does it neces-

sarily follow that A∗ is representable in S?

Problem 10. Show that for any normal system, neither P nor R is
representable.

Fixed Points

By a fixed point of h is meant a number n such that h(n) is equivalent

to n.

Problem 11. Prove that if S is normal, then every number h has a fixed
point.

Problem 12. Prove the following:

(a) If h represents R, then any fixed point of h is undecidable.
(b) If h represents P , then any fixed point of h0 is undecidable.

Exercise. Show the following strange facts:

154 Part II: Recursion Theory and Metamathematics

(1) If h represents some superset of R disjoint from P , then h(h) is

undecidable.
(2) If h represents some superset of P disjoint from R, then h(h0 ) is
undecidable.
We will later have important use for the following:

Problem 13. Show that no superset A of R∗ , with A disjoint from P ∗ ,

can be definable in S. [Hint: Show that if R∗ ⊆ A, and h defines A
in S, then h is in both A and P ∗ (and hence that A is not disjoint
from P ∗ ).]

Exercise. Show that if P ∗ ⊆ A, and h defines A in S, then h0 is in both A

and R∗ (and hence that A is not disjoint from R∗ ).

II. Standard Simple Systems

Now recursion theory enters the picture. We shall call a simple system S a
standard simple system if the sets P and R are recursively enumerable, and
the functions neg(x), r(x, y) and s(x, y, z) are recursive. In what follows, S
will be assumed to be a standard simple system. [Recall that the function
s(n, x, y) is the index of the sentence ψn (x̄, ȳ).]

Problem 14. Show that for any set A,

(1) If A is recursively enumerable, so is A∗ .
(2) If A is recursive, so is A∗ .
More generally, show that for any recursive function f (x),
(10 ) If A is recursively enumerable, so is f −1 (A).
(20 ) If A is recursive, so is f −1 (A).

Problem 15. Show that every set representable in S is recursively enumer-

able, and that every set definable in S is recursive.

Problem 16. Show the following:

(a) If some recursively enumerable but non-recursive set is representable
in S, then S is incomplete.
(b) If all recursively enumerable sets are representable in S, then S is
incomplete.
 Chapter 7. Metamathematical Applications 155

Question: Suppose we are given, not that all recursively enumerable sets
are representable in S, but only that all recursive sets are representable
in S. Is that enough to guarantee that S is incomplete? The reader might
like to try answering this on his or her own. The answer will appear in this
text later on.

Undecidability and Incompleteness

The word “undecidable” has two very different meanings in the literature,
which might sometimes cause confusions. On the one hand, a sentence of
a system is called undecidable in the system if it is neither provable nor
refutable in the system. On the other hand, the system as a whole is
called undecidable if the set of provable sentence is not recursively solv-
able (or: under an effective Gödel numbering, the set of Gödel numbers of
the provable sentences is not recursive).
We shall thus call a simple system S undecidable if the set P is not
recursive.
The two senses of undecidable, though different, have the following
important link:

Theorem 1. If S is undecidable (but standard), then S is incomplete (some

number is undecidable in S, i.e. outside both P and R).

Problem 17. Prove Theorem 1.

It was known prior to 1961 that if all recursively enumerable sets are rep-
resentable in a system, then the system is undecidable. In the Theory of
Formal Systems [Smullyan, 1961] I proved the stronger fact that if all recur-
sive sets are representable in a system, the system is undecidable. My proof
goes over to simple systems.

Theorem 2. If all recursive sets are representable in S, then S is

undecidable.

Problem 18. Prove Theorem 2. [Hint: Use the fact already proved that
if P is recursive, so is P ∗ (Problem 14).]

Remark. For a standard system S, if all recursive sets are representable

in S, then S is undecidable by Theorem 2, hence incomplete by Theorem 1.
156 Part II: Recursion Theory and Metamathematics

Thus if all recursive sets are representable in a standard system S, then S

is incomplete, which answers the question raised following Problem 16.

Recursive Inseparability

Problem 19. If the pair (P, R) is recursively inseparable, does it necessarily

follow that S is undecidable?

Problem 20. Show that for any recursive function f (x) and any disjoint
pair (A, B), if the pair (f −1 (A), f −1 (B)) is recursively inseparable, then
the pair (A, B) is also recursively inseparable. Conclude that if (P ∗ , R∗ ) is
recursively inseparable, so is (P, R).
We next wish to prove:

Theorem 3. If every recursive set is definable in S then the pair (P ∗ , R∗ )

is recursively inseparable, and so is the pair (P, R).

Problem 21. Prove Theorem 3. [Hint: Use Problem 13.]

Recall that, given a disjoint pair (A, B), we say that a number h separates
A from B (in S) if h represents a superset of A and contra-represents some
superset of B. And we say that h exactly separates A from B if h both
represents A and contra-represents B.
We next wish to prove the following theorem, which has the interesting
corollaries that follow it:

Theorem 4. If some disjoint pair (A1 , A2 ) is recursively inseparable in S,

but separable in S, then the pair (P, R) is recursively inseparable.

Corollary 1. If some recursively inseparable pair is separable in S, then S

is undecidable.

Corollary 2. If some recursively inseparable pair is separable in S, then S

is incomplete (assuming S is standard).

Note: Corollary 2 is essentially Kleene’s symmetric form of Gödel’s theo-

rem [Kleene, 1952]. Kleene proved this for a particular pair of recursively
inseparable sets, but the proof works for any recursively inseparable pair of
recursively enumerable sets.
 Chapter 7. Metamathematical Applications 157

Problem 22. Prove Theorem 4.

Rosser Systems

We shall call S a Rosser system if for any two recursively enumerable sets
A and B, there is a number h that separates A − B from B − A. [If A and
B are disjoint, then of course such an h also separates A from B.]
We shall call a Rosser system S an exact Rosser system, if for any two
disjoint recursively enumerable sets A and B, there is a number h that
exactly separates A from B.
Gödel’s proof of the incompleteness of systems like Peano Arithmetic
involved representing a certain recursively enumerable set, but the only
way of representing recursively enumerable sets that was known at that
time involved the hypothesis of ω-consistency. However, Andrzej Ehren-
feucht and Solomon Feferman [1960] found a way of representing all recur-
sively enumerable sets in Peano Arithmetic, without having to assume
ω-consistency! In terms of simple systems, what they showed was that if
S is a Rosser system in which all recursive functions of one argument are
admissible, then all recursively enumerable sets are representable in S. Later
came a stronger result, which, for simple systems, is:

Theorem 5. [After Putnam, Smullyan, 1961] If S is a Rosser system in

which all recursive functions of one argument are admissible, then S is an
exact Rosser system.
Theorem 5 is an obvious consequence of the following two propositions:

Proposition 1. If S is a Rosser system, then some doubly universal pair

is exactly separable in S.

Proposition 2. If some doubly universal pair is exactly separable in S and

if all recursive functions of one argument are admissible, then every disjoint
pair of recursively enumerable sets is exactly separable in S.

Problem 23. Prove Propositions 1 and 2. [Hint: For Proposition 1, show

that if the Kleene pair (K1 , K2 ) (or any other effectively inseparable pair of
recursively enumerable sets) is separable in S, then some doubly universal
pair is exactly separable in S.]
158 Part II: Recursion Theory and Metamathematics

Effective Rosser Systems

We shall call S an effective Rosser system if there is a recursive func-

tion Π(x, y) such that for any numbers i and j, Π(i, j) separates
(ωi − ωj ) from (ωj − ωi ). Such a function Π(x, y) will be called a Rosser
function.
Effective Rosser systems have some very interesting properties, as we
will see. For one thing, we will soon see that every effective Rosser system
is an exact Rosser system (a result that is incomparable in strength with
Theorem 5).
The following proposition is key for this, as well as for some further
results.

Proposition 3. If S is an effective Rosser system, then for any recursively

enumerable relations R1 (x, y), R2 (x, y), there is a number h such that for
any n:
(1) R1 (n, h) ∧ ∼ R2 (n, h) implies h(n) ∈ P ;
(2) R2 (n, h) ∧ ∼ R1 (n, h) implies h(n) ∈ R.

Problem 24. Prove Proposition 3. [Hint: Apply the weak double recursion
theorem to the relations R1 (x, Π(y, z)), R2 (x, Π(y, z)), where Π(x, y) is a
Rosser function for S.]
Now we can prove:

Theorem 6. If S is an effective Rosser system, then S is an exact Rosser

system.

Problem 25. Prove Theorem 6. [Hint: Let R1 (x, y) be the relation

x ∈ A ∨ y(x) ∈ R, and let R2 (x, y) be the relation x ∈ B ∨ y(x) ∈ P . Recall
that y(x) = r(y, x). Apply Proposition 3 to R1 (x, y) and R2 (x, y).]

Rosser Fixed-Point Property

We say that S has the Rosser fixed point property if for any two recursive
functions f1 (x) and f2 (x), there is a number h such that for all n for which
ωf1 (n) is disjoint from ωf2 (n) :
(1) h(n) ∈ ωf1 (n) implies h(n) ∈ P ;
(2) h(n) ∈ ωf2 (n) implies h(n) ∈ R.
 Chapter 7. Metamathematical Applications 159

Theorem 7. Every effective Rosser system has the Rosser fixed point
property.

Problem 26. Prove Theorem 7. [Hint: Given recursive functions f1 (x) and
f2 (x), apply Proposition 3 to the relations y(x) ∈ ωf1 (x) and y(x) ∈ ωf2 (x) .]

Uniform Incompleteness

By the principal part of a simple system S we mean the pair (P, R).
We call a simple system S0 an extension of a simple system S if P ⊆ P 0
and R ⊆ R0 , where (P 0 , R0 ) is the principal part of S0 , and the representation
function r(x, y) and the function neg (x) are the same in both S and S0 .
Now consider an infinite sequence S = S1 , S2 , . . . , Sn , . . ., where
each Sn+1 is an extension of Sn . Such a sequence is called an effective
sequence, or a recursively enumerable sequence, if the relations x ∈ Py
and x ∈ Ry are both recursively enumerable. The system S is called
uniformly incompletable if, for any recursively enumerable sequence S1 ,
S2 , . . . , Sn ,. . . , of extensions of S, there is a number h such that, for
every n, the number h(n) is an undecidable number of Sn (i.e. h(n) is
outside Pn ∪ Rn ).
The following is one of the main results.

Theorem 8. Every effective Rosser system is uniformly incompletable.

Problem 27. Prove Theorem 8. [Hint: Given a recursively enumerable

sequence of extensions of S, show that there are recursive functions f1 (x)
and f2 (x) such that for all n the set ωf1 (n) is Rn and ωf2 (n) is Pn . Then
use the Rosser fixed point property.]

***
Before leaving this chapter, I should say a little more about the rela-
tion of simple systems to the first-order arithmetical systems in standard
formalization of Tarski [1953].
Consider such a first-order system F and its associated simple system
S, as defined much earlier in this chapter.
F is called a Rosser system, more specifically a Rosser system for sets,
iff its associated simple system S is a Rosser system, as defined here, which
160 Part II: Recursion Theory and Metamathematics

is equivalent to the condition that for any pair of recursively enumerable

sets A and B there is a formula ϕh (x) such that for all n, if n ∈ A − B,
then ϕh (n̄) is provable (in F), and if n ∈ B − A, then ϕh (n̄) is refutable in
F. [In terms of the associated simple system S, if n ∈ A − B, then h(n) ∈ P
and if n ∈ B − A, then h(n) ∈ R.]
The system F is called an exact Rosser system (for sets) if for any pair
of disjoint recursively enumerable sets A and B, there is a predicate H such
that for all n, n ∈ A iff H(n) is provable, and n ∈ B iff H(n) is refutable.
In the literature, a function f (x) is said to be definable if there is a
formula ψ(x, y) such that for all n and m:
(1) If f (n) = m, then ψ(n, m) is provable;
(2) If f (n) 6= m, then ψ(n, m) is refutable;
(3) The sentence ∀x∀y((ψ(n, x) ∧ ψ(n, y)) ⊃ x = y) is provable.
As far as I know, the term admissible function does not appear in the
literature, except in my own writings. However, if a formula F (x, y) defines
a function f (x), then for any formula H(x), if we take G(x) to be the
formula ∃y(F (x, y) ∧ H(x)), then it can be verified that for every n, not
only is it the case that G(n) is equivalent to H(f (n)), but the very sentence
G(n) ≡ H(f (n)) is provable in the system. For details, see Theorem 2,
Chapter 8 of Gödel’s Incompleteness Theorems [Smullyan, 1992], in which
I use the term “strongly definable” for what I am calling “definable” here.
Thus every definable function f (x) is indeed admissible. The
Ehrenfeucht–Feferman Theorem, as stated, is that any consistent Rosser
system in which all recursive functions f (x) are definable is an exact Rosser
system.
In the context of a first-order system F, a formula ψ(x, y) is said to
separate a relation R1 (x, y) from a relation R2 (x, y) if for all n and m,
(1) R1 (n, m) implies that ψ(n, m) is provable.
(2) R2 (n, m) implies that ψ(n, m) is refutable.

F is called a Rosser system for binary relations if for any two recursively
enumerable relations R1 (x, y), R2 (x, y), there is a formula ψ(x, y) that sep-
arates the relation R1 (x, y)∧∼R2 (x, y) from R2 (x, y)∧∼R1 (x, y). Sheperd-
son’s Theorem is that every consistent Rosser system for binary relations is
an exact Rosser system for sets. This appears to be incomparable in strength
with the Putnam–Smullyan Theorem, which is that every consistent Rosser
system in which all recursive functions of one argument are definable is an
exact Rosser system for sets.
 Chapter 7. Metamathematical Applications 161

The proof of Shepherdson’s Theorem (which can be found in

[Shepherdson, 1961]) is not very different from my proof that every effective
Rosser system is an exact Rosser system. Indeed, this theorem and its proof
borrowed heavily from Shepherdson’s Theorem and proof.

Solutions to the Problems of Chapter 7

1. (a) It is obvious that if h completely represents A, then h defines A.

To show the converse, suppose h defines A. Then for all n,
(1) n ∈ A implies h(n) ∈ P ;
(2) n ∈ A implies h(n) ∈ R.
We are to show the converses of (1) and (2).
Suppose h(n) ∈ P . Then h(n) ∈/ R (since R is disjoint from P ).
Thus by (2) n ∈ A doesn’t hold, so that n ∈ A. The proof that
h(n) ∈ R implies that n ∈ A is similar.
(b) h contra-represents A iff for every n, n ∈ A holds iff h ∈ R,
which is true iff h0 (n) ∈ P , which is true iff h0 represents A.
Conversely, suppose h0 represents A. Then for all n, n ∈ A
iff h0 (n) ∈ P , which is true iff h(n) ∈ R. And so h contra-
represents A.

2. We are to show that h represents A iff A = rh−1 (P ). We first recall that

rh (n) = h(n), since rh (n) = r(h, n) = h(n).
(1) Suppose h represents A. Then n ∈ A iff h(n) ∈ P , which is true iff
rh (n) ∈ P , which is true iff n ∈ rh−1 (P ). Thus n ∈ A iff n ∈ rh−1 (P ).
Since this is true for all n, A = rh−1 (P ).
(2) Conversely suppose A = rh−1 (P ). Then n ∈ A iff n ∈ rh−1 (P ). But
n ∈ rh−1 (P ) is true iff rh (n) ∈ P , which is true iff h(n) ∈ P . Thus
n ∈ A iff h(n) ∈ P . Since this is true for all n, h represents A.
The proof that h contra-represents A iff A = rh−1 (R) is similar.

3. We prove the equivalent proposition that if S is complete, then the

complement of any representable set is representable.
Well, suppose S is complete, and that h represents A. We will show
that h0 represents A. Consider any number n.
(1) Suppose n ∈ A. Then n ∈/ A. Hence h(n) ∈/ P [since h represents A].
Hence h(n) ∈ R [since S is complete]. Thus h0 (n) ∈ P . Thus n ∈ A
implies h0 (n) ∈ P .
162 Part II: Recursion Theory and Metamathematics

(2) Conversely, suppose h0 (n) ∈ P . Then [by condition C1 ] h(n) ∈ R,

so that h(n) ∈
/ P . Thus n ∈
/ A [since h represents A], so that n ∈ A.
Thus h0 (n) ∈ P implies that n ∈ A.
By (1) and (2), h0 represents A.
4. If some number h represents P ∗ , we can reach contradiction as follows:
h(h) ∈ P iff h ∈ P ∗ [since h represents P ∗ , for all n we have n ∈ P ∗ iff
/ P ∗ , which is true iff h(h) ∈
h(n) ∈ P ]. But h ∈ P ∗ iff h ∈ / P . Thus we
get the contradiction that h(n) ∈ P iff h(n) ∈ / P , so that no number h
represents P ∗ .
By a similar argument (using R instead of P ), the set R∗
cannot be contra-representable. Thus R∗ cannot be representable
[by Problem 1(b)].
5. (a) Suppose h represents R∗ . Then, for all n, h(n) ∈ P iff n ∈ R∗ ,
which is true iff n(n) ∈ R. Taking h for n, we have h(h) ∈ P iff
h(h) ∈ R. Since h(h) cannot be in both P and R, h(h) must be
outside both of them, and is thus undecidable.
(b) Suppose h represents P ∗ . Then, for all n, h(n) ∈ P iff n ∈ P ∗ . Thus
h(h0 ) ∈ P iff h0 ∈ P ∗ , and h0 ∈ P ∗ iff h0 (h0 ) ∈ P , which is true iff
h(h0 ) ∈ R. Thus h(h0 ) ∈ P iff h(h0 ) ∈ R, and since P is disjoint
from R, h(h0 ) is outside both P and R, and is thus undecidable.
6. (a) Suppose h represents some superset A of R∗ that is disjoint from
P ∗ . We must show that h(h) is undecidable.
Well, if h ∈ P ∗ , then h(h) ∈ P , and h ∈ A [since h represents A],
contrary to the fact that P ∗ is disjoint from A. Thus h ∈ / P ∗ , since
assuming so led to a contradiction, and so h(h) ∈ / P.
If h ∈ A, then h(h) ∈ P [since h represents A]. Thus h ∈ P ∗ ,
again contrary to the fact that A is disjoint from P ∗ . Thus h ∈
/ A,
∗
since assuming so led to a contradiction. Then h ∈ / R [since
R∗ ⊆ A], and so h(h) ∈/ R.
Thus h(h) ∈/ P and h(h) ∈ / R so that h(h) is undecidable.
(b) Suppose h represents some superset A of P ∗ disjoint from R∗ . We
must show that h(h0 ) is undecidable.
Well, if h0 ∈ A, then h(h0 ) ∈ P [since h represents A]. Hence
h0 (h0 ) ∈ R [by condition C1 ], and so h0 ∈ R∗ , contrary to the fact
that R∗ is disjoint from A. Therefore h0 ∈ / A, since assuming so led
to a contradiction. Then h0 ∈/ P ∗ [since P ∗ ⊆ A], and so h0 (h0 ) ∈
/ P,
and therefore h(h0 ) ∈/ R.
 Chapter 7. Metamathematical Applications 163

If h0 ∈ R∗ , then h0 (h0 ) ∈ R, so that h(h0 ) ∈ P , so that h0 ∈ A

[which is represented by h], which is again contrary to the fact that
R∗ is disjoint from A. Hence h0 ∈ / R∗ , since assuming so led to a
0 0
contradiction. Then h (h ) ∈ / R, and therefore h(h0 ) ∈
/ P.
0 0
Thus h(h ) ∈ / R and h(h ) ∈ / P , and therefore h(h0 ) is
undecidable.
The reason why these results are stronger than those of
Problem 5 is this: To begin with, since P is disjoint from R, then P ∗
is disjoint from R∗ (why?). Therefore, if h represents R∗ , it auto-
matically represents some superset of R∗ disjoint from P ∗ , namely
R∗ itself. [Any set is both a subset and a superset of itself.] Simi-
larly, if h represents P ∗ , then it represents a superset of P ∗ disjoint
from R∗ , namely P ∗ itself. Thus the results of Problem 5 are but
special cases of those of Problem 6.

7. (a) Suppose A ⊆ B. For any n, if n ∈ f −1 (A), then f (n) ∈ A. Thus

f (n) ∈ B [since A ⊆ B], and hence n ∈ f −1 (B). Thus n ∈ f −1 (A)
implies n ∈ f −1 (B), and so f −1 (A) ⊆ f −1 (B).
(b) Suppose A is disjoint from B. If n ∈ f −1 (A), then f (n) ∈ A. Thus
f (n) ∈/ B. Thus n ∈ f −1 (A) implies that n ∈ / f −1 (B), and thus
−1 −1
f (A) is disjoint from f (B).
(c) n ∈ f −1 (A) iff f (n) ∈ A, which is true iff f (n) ∈
/ A, which is true
/ f −1 (A), which is true iff n ∈ f −1 (A). Thus n ∈ f −1 (A) iff
iff n ∈
n ∈ f −1 (A), and since this holds for every n, we have
f −1 (A) = f −1 (A).
The statements (a1 ), (a2 ), and (a3 ) clearly follow from statements (a),
(b), and (c) when f = d(x), because d−1 (A) is the set A∗ for any set A.

8. Suppose f (x) is admissible. Given h, let k be such that k(n) is equiva-

lent to h(f (n)) for all n.
(a) We assert that if h represents a set A, then k represents f −1 (A),
and that if h contra-represents A, then k contra-represents f −1 (A).
To prove the former, suppose h represents A. Then n ∈ f −1 (A)
iff f (n) ∈ A, which is true iff h(f (n)) ∈ P [since h represents A],
which is true iff k(n) ∈ P . Thus, for all n, n ∈ f −1 (A) iff k(n) ∈ P ,
and thus k represents f −1 (A).
The proof that if h contra-represents A, then k contra-represents
−1
f (A) is similar (just replace “P ” by “R” and “represents” by
“contra-represents”).
164 Part II: Recursion Theory and Metamathematics

(b) Suppose h exactly separates the set A from the set B. Then h
represents A and contra-represents B. Then, by the solution to
part (a), k represents f −1 (A) and contra-represents f −1 (B), and
so k exactly separates f −1 (A) from f −1 (B).

9. Of course it follows! If S is normal, the diagonal function d(x) is admis-

sible. Hence if A is representable, so is d−1 (A) [by Problem 8], and
d−1 (A) is the set A∗ .
∗
10. For a normal system, if P were representable, then P would be
∗
representable [by Problem 9]. But P = P ∗ [by Problem 7], so that
P ∗ would be representable, which is contrary to Problem 4. A similar
argument show that R is not representable.
11. Suppose S is normal. For any numbers h and n, h# (n) is equivalent to
h(d(n)), so that h# (h# ) is equivalent to h(d(h# )). But d(h# ) = h# (h# )
and so h# (h# ) is equivalent to h(h# (h# )), which means that h# (h# )
is a fixed point of h.
12. (a) Suppose h represents R and that n is some fixed point of h [i.e. n is
equivalent to h(n)]. Since h represents R, then n ∈ R iff h(n) ∈ P .
But h(n) ∈ P iff n ∈ P [since n is a fixed point of h]. Thus n ∈ R
iff n ∈ P . But R and P are assumed to be disjoint, so the fixed
point n of h can belong to neither P nor R, which means that n
must be undecidable.
(b) Suppose h represents P and that n is some fixed point of h0 . Since
h represents P , then n ∈ P iff h(n) ∈ P . But h(n) ∈ P iff h0 (n) ∈ R
[by condition C1 ]. And h0 (n) ∈ R iff n ∈ R [since n is a fixed point
of h0 ]. Thus n ∈ P iff n ∈ R, and so again the arbitrary fixed point
n of h0 is undecidable.
13. Suppose R∗ ⊆ A and h defines A in S. We will show that A cannot be
disjoint from P ∗ . Suppose h ∈
/ A. Then h(h) ∈ R [since h defines A].
Hence h ∈ R∗ , which would mean h ∈ A [since R∗ ⊆ A], which is
impossible. Thus it cannot be that h ∈ / A, which implies h ∈ A. It then
further follows that h(h) ∈ P [since h defines, and hence represents, A].
Therefore h ∈ P ∗ . Thus h ∈ A and h ∈ P ∗ , so that A is not disjoint
from P ∗ .
14. We will first consider the general case.
(10 ) Let f (x) be a recursive function and suppose the set A is recur-
sively enumerable. For any number x, x ∈ f −1 (A) iff f (x) ∈ A
 Chapter 7. Metamathematical Applications 165

which is true iff ∃y(y = f (x) ∧ y ∈ A). Thus f −1 (A) is recursively

enumerable.
(20 ) If A is also recursive, then A and its complement A are both recur-
sively enumerable, so that f −1 (A) and f −1 (A) are both recursively
enumerable, as just shown. However, f −1 (A) is the complement of
f −1 (A), so that f −1 (A) is recursive.
To apply the general case to (1) and (2): Since the function
r(x, y) is recursive, so is the diagonal function d(x) [which is
r(x, x)], and since A∗ = d−1 (A), it follows from (10 ) and (20 ) that
if A is recursively enumerable (recursive), so is A∗ .
15. We must show that (a) every set representable in S is recursively enu-
merable, and that (b) every set definable in S is recursive.
(a) Suppose h represents A. Then for all n, n ∈ A iff r(h, n) ∈ P .
Let f (x) = r(h, x) Since the function r(x, y) is recursive, so is the
function f (x). Thus for all, n ∈ A iff f (n) ∈ P , and so A = f −1 (P ).
Since P is recursively enumerable, and f (x) is a recursive function,
then f −1 (P ) is recursively enumerable [by Problem 14], and thus
A is recursively enumerable.
(b) If A is completely representable [which is equivalent to A being
definable, by Problem 1(a)], then A and A are both representable,
hence both recursively enumerable, and so A is then recursive.
16. (a) Suppose that A is a recursively enumerable set that is not recursive,
and that A is representable in S. Since A is not recursive, its com-
plement A is not recursively enumerable, hence not representable
[since only recursively enumerable sets are representable in S, by
Problem 15]. Thus A is a representable set whose complement is
not representable. Hence S is incomplete [by Problem 3].
(b) This follows from (a) and the fact that there exists a recursively
enumerable set that is not recursive, such as Post’s complete set K
[if all recursively enumerable sets were representable in S, K would
also have to be recursively enumerable].
17. Suppose S is undecidable and effective. Then the set P is not recursive,
and so its complement P is not recursively enumerable. Yet the set
R is recursively enumerable [since S is effective], and so R is not the
complement of P . But R is disjoint from P , which means that some
n must be outside both P and R, and is thus undecidable. Thus S is
incomplete.
166 Part II: Recursion Theory and Metamathematics

18. Suppose all recursive sets are representable in S. If P were recursive,

then P ∗ would be recursive [by Problem 14]. Thus P ∗ would be recur-
sive, hence representable in S, which would be contrary to Problem 4.
Therefore P is not recursive, and S is thus undecidable [by the definition
of an undecidable system].
19. Of course it does! If a disjoint pair of sets (A, B) is recursively insep-
arable, then by definition no recursive superset of A can be disjoint
from B. Thus A cannot be recursive [for it were, then A itself would
be a recursive superset of A disjoint from B]. Similarly, B cannot be
recursive.
Thus, if (P, R) were recursively inseparable, then P would not be
recursive, so that S would be undecidable.
20. We are to show that if f (x) is a recursive function and (A, B)
is any disjoint pair, if (f −1 (A), f −1 (B)) is disjoint and recursively
inseparable, so is (A, B). We will prove the equivalent proposi-
tion that if a disjoint pair (A, B) is recursively separable, so is
(f −1 (A), f −1 (B)).
Well, suppose that A is disjoint from B and that A is recursively
separable from B. Then there is a recursive superset A0 of A disjoint
from B. Hence f −1 (A) is disjoint from f −1 (B) [by Problem 7(b)]. And
f −1 (A) is recursive [by Problem 14]. Thus f −1 (A0 ) is a recursive super-
set of f −1 (A) that is disjoint from f −1 (B), and so f −1 (A) is recursively
separable from f −1 (B).
Since (P ∗ , R∗ ) = (d−1 (P )d−1 (R)) and d(x) is a recursive function,
it follows from what was just shown that if (P ∗ R∗ ) is recursively insep-
arable, so is (P, R).
21. Suppose all recursive sets are definable in S. We will show that (P ∗ , R∗ )
is recursively inseparable, from which we will be able to conclude that
so is the pair (P, R) [by Problem 20].
Let A be any superset of R∗ disjoint from P ∗ . By Problem 13, A can-
not be definable in S. If A were recursive, then it would be definable in
S [since the problem assumes this]. Thus A cannot be recursive. Con-
sequently there is no recursive superset of R∗ disjoint from P ∗ , which
means that (P ∗ , R∗ ) is recursively inseparable.
22. Suppose (A1 , A2 ) is disjoint and recursively inseparable, and that h
separates A1 from A2 in S. Then h represents some superset B1 of
 Chapter 7. Metamathematical Applications 167

A1 and contra-represents some superset B2 of A2 . In this situation B1

and B2 must be disjoint [because n ∈ B1 iff h(n) ∈ P and n ∈ B2
iff h(n) ∈ R, but P and R are disjoint]. Since (A1 , A2 ) is recursively
inseparable, and A1 ⊆ B1 and A2 ⊆ B2 , then the pair (B1 , B2 ) is
recursively inseparable (verify this!). Now, by Problem 2, B1 = rh−1 (P )
and B2 = rh−1 (R), where r(x, y) is the representation function for sets.
Thus the pair (rh−1 (P ), rh−1 (R)) is recursively inseparable, and since the
function rh (x) is recursive [because r(x, y) is], then the pair (P, R) is
recursively inseparable [by Problem 20].
Now, Corollary 1 follows because if (P, R) is recursively inseparable,
then of course P is not recursive, so that S is undecidable.
Corollary 2 then follows from Corollary 1 and Theorem 1 (if S is
effective but undecidable, then S is incomplete).
23. (a) To prove Proposition 1, suppose K1 is separable from K2 in S and
let h effect the separation. Then h represents some superset A1
of K1 and contra-represents some superset A2 of K2 . Since P is
disjoint from R, then A1 is disjoint from A2 . Of course, h exactly
separates A1 from A2 .
Since (K1 , K2 ) is effectively inseparable, so is the pair (A1 , A2 )
[verify this!]. Since S is an effective system, then A1 and A2 are
recursively enumerable sets. Thus (A1 , A2 ) is an effectively insepa-
rable pair of recursively enumerable sets, hence is doubly universal
[by Theorem 11 of Chapter 6]. Thus h exactly separates the doubly
universal pair (A1 , A2 ).
(b) To prove Proposition 2, suppose that some doubly universal pair
(U1 , U2 ) is exactly separable in S and that all recursive functions of
one argument are admissible in S. Let (A, B) be a disjoint pair of
recursively enumerable sets that we wish to exactly separate in S.
Since (U1 , U2 ) is doubly universal, then there is a recursive func-
tion f (x) such that A = f −1 (U1 ) and B = f −1 (U2 ). By hypothesis,
f (x) is admissible (all recursive functions of one argument are).
Thus, by Problem 8, f −1 (U1 ) is exactly separable from f −1 (U2 ),
and thus A is exactly separable from B.

24. Suppose Π(x, y) is a Rosser function for S. Given recursively enumer-

able relations R1 (x, y), R2 (x, y), by the weak double recursion theorem
(Theorem 1 of Chapter 6), taking M1 (x, y, z) to be R1 (x, Π(y, z))
168 Part II: Recursion Theory and Metamathematics

and M2 (x, y, z) to be R2 (x, Π(y, z)), there are numbers a and b

such that:
ωa = {x : R1 (x, Π(a, b))};
ωb = {x : M2 (x, Π(a, b))}.
We let h be the number Π(a, b). Then ωa = {x : R1 (x, h)} and
ωb = {x : R2 (x, h)}. Thus ωa − ωb = {x : R1 (x, h) ∧ ∼R2 (x, h)} and
ωb − ωa = {x : R2 (x, h) ∧ ∼R1 (x, h)}. Since Π(x, y) is a Rosser function,
then h, which is Π(a, b), separates ωa − ωb from ωb − ωa , and thus
separates {x : R1 (x, h)∧ ∼ R2 (x, h)} from {x : R2 (x, h) ∧ ∼R1 (x, h)}.
25. Let A and B be disjoint recursively enumerable sets. Let R1 (x, y)
be the relation x ∈ A ∨ y(x) ∈ R and let R2 (x, y) be the relation
x ∈ B ∨ y(x) ∈ P . Applying Proposition 3 to the pair (R1 , R2 ), there
is a number h such that for all n,
R1 (n, h) ∧ ∼R2 (n, h) implies h(n) ∈ P and
R2 (n, h) ∧ ∼R1 (n, h) implies h(n) ∈ R. Thus
(1) [(n ∈ A ∨ h(n) ∈ R) ∧ ∼(n ∈ B ∨ h(n) ∈ P )] ⊃ h(n) ∈ P ;
(2) [(n ∈ B ∨ h(n) ∈ P ) ∧ ∼(n ∈ A ∨ h(n) ∈ R)] ⊃ h(n) ∈ R.
Also,
(3) ∼(n ∈ A ∧ n ∈ B) [since A is disjoint from B];
(4) ∼(h(n) ∈ P ∧ h(n) ∈ R) [since P is disjoint from R].
From (1), (2), (3) and (4) it follows by propositional logic that n ∈ A
iff h(n) ∈ P , and n ∈ B iff h(n) ∈ R.
To see this, let us use the following abbreviations:
p1 for n ∈ A; p2 for n ∈ B;
q1 for h(n) ∈ P ; q2 for h(n) ∈ R.
Thus we have:
(1) [(p1 ∨ q2 ) ∧ ∼(p2 ∨ q1 )] ⊃ q1 ;
(2) [(p2 ∨ q1 ) ∧ ∼(p1 ∨ q2 )] ⊃ q2 ;
(3) ∼(p1 ∧ p2 );
(4) ∼(q1 ∧ q2 ).
We are to infer that p1 ≡ q1 and p2 ≡ q2 .
(a) First we show that p1 ⊃ q1 . Well, suppose p1 . Then p1 ∨ q2 is true.
Thus (1) is equivalent to
(10 ) ∼(p2 ∨ q1 ) ⊃ q1 .
 Chapter 7. Metamathematical Applications 169

Since p1 is true (by assumption), then p2 is false [by (3)], so that

p2 ∨ q1 is equivalent to q1 , so that (10 ) is equivalent to
(20 ) ∼q1 ⊃ q1 .
From this q1 follows, which proves that p1 ⊃ q1 .
(b) For the converse, suppose q1 is true. Then p2 ∨ q1 is true, and so
(2) is equivalent
(20 ) ∼(p1 ∨ q2 ) ⊃ q2 .
Since q1 is true (by assumption), then q2 is false [by (4)], so that
p1 ∨ q2 is equivalent to p1 , so that (20 ) is equivalent to
(200 ) ∼p1 ⊃ q2 .
Since q2 is false, it follows from (200 ) that p1 is true. This proves
that q1 ⊃ p1 . Thus p1 ≡ q1 .
The proof that p2 ≡ q2 is similar (just interchange p1 with p2
and q1 with q2 ).
Note: This proof is a modification of a proof of a result of John
Shepherdson.
26. Suppose S is an effective Rosser system. Given two recursive func-
tions f1 (x) and f2 (x) let R1 (x, y) be the relation y(x) ∈ ωf1 (x) and
let R2 (x, y) be the relation y(x) ∈ ωf2 (x) .
Applying Proposition 3 to these two relations, there is a number h
such that for every n,
(1) h(n) ∈ ωf1 (n) ∧ ∼(h(n) ∈ ωf2 (n) ) implies h(n) ∈ P ;
(2) h(n) ∈ ωf2 (n) ∧ ∼(h(n) ∈ ωf1 (n) ) implies h(n) ∈ R.
If ωf1 (n) is disjoint from ωf2 (n) , then (1) and (2) are respectively equiv-
alent to:
(10 ) h(n) ∈ ωf1 (n) implies h(n) ∈ P ;
(20 ) h(n) ∈ ωf2 (n) implies h(n) ∈ R.
Thus S has the Rosser fixed point property.
27. Consider a recursively enumerable sequence S = S1 , S2 , . . . , Sn , . . ., of
extensions of S. The relation x ∈ Ry is recursively enumerable. Hence by
the iteration theorem, there is a recursive function f1 (x) such that for
all n, we have ωf1 (n) = {x : x ∈ Rn }. Thus ωf1 (n) = Rn . Similarly, there
is a recursive function f2 (x) such that, for all n, we have ωf1 (n) = Pn .
Now suppose that S is effectively a Rosser system. Then S has the
Rosser fixed point property, since for every n the sets ωf1 (x) and ωf2 (x)
170 Part II: Recursion Theory and Metamathematics

are disjoint. So by the Rosser fixed point property, there is a number h

such that for all n:
(1) h(n) ∈ ωf1 (n) implies h(n) ∈ P ;
(2) h(n) ∈ ωf2 (n) implies h(n) ∈ R.
Thus,
(10 ) h(n) ∈ Rn implies h(n) ∈ P ;
(20 ) h(n) ∈ Pn implies h(n) ∈ R.
Since P ⊆ Pn and R ⊆ Rn , then by (10 ) and (20 ), we have:
(10 ) h(n) ∈ Rn implies h(n) ∈ Pn ;
(20 ) h(n) ∈ Pn implies h(n) ∈ Rn .
Since Pn is disjoint from Rn , it must be the case that h(n) ∈
/ P n and
/ Rn . Thus S is uniformly incompletable.
h(n) ∈
 Part III

Elements of Combinatory Logic

b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 8

Beginning Combinatory Logic

The subject of combinatory logic, initiated by Moses Schönfinkel [1924],

is a most elegant one, as the reader will see. It plays a significant role in
computer science and provides yet another approach to recursion theory.
In this book we provide only the beginnings of the subject, hoping to whet
the reader’s appetite for more.
By an applicative system is meant a set C together with an operation that
assigns to each element x and each element y (in that order) an element
denoted (xy). In combinatory logic we often omit parentheses, with the
understanding that parentheses are to be restored to the left: xyz abbrevi-
ates ((xy)z), not (x(yz)). Also xyzw abbreviates (((xy)z)w), etc. Note that
this is not the same as the common set of rules for eliminating parentheses
in school mathematics and even in many college mathematics courses, so
it takes some getting used to. But have courage, for it will become natural
fairly quickly. (But always remember to put outside parentheses around
an expression when you replace a variable by a compound expression; for
instance, replacing the variable x with SI in the expression S(Kx)K would
result in S(K(SI))K.) Also note that, in general, in applicative systems xy
is not the same as yx, nor is x(yz) the same as (xy)z (i.e. the commuta-
tive and associative rules of elementary arithmetic do not generally apply
here).
The elements of C are called combinators. It is to be understood that C
contains more than one element.
174 Part III: Elements of Combinatory Logic

Fixed Points

An element y is called a fixed point of an element x if xy = y. Fixed points

play a key role in combinatory theory.

The Composition Condition

For any elements A, B and C, we say that C combines A with B if

Cx = A(Bx) for every element x. We say that the composition condition
holds if for every A and B, there is some C that combines A with B.

Introducing the Combinator M, the Mocker

(the Mockingbird)

A combinator M will be called a mocker, or a duplicator, if

M x = xx
for every element x. Here is our first interesting result:

Theorem 1. The following two conditions are sufficient to ensure that

every element A has a fixed point:
C1 : The composition condition holds.
C2 : There is a mocker M .

Problem 1. Prove Theorem 1.

Note: Theorem 1 states a basic fact of combinatorial logic. The solution,
though quite brief, is very ingenious, and derives ultimately from the work
of Gödel. It is related to many of the fixed point results in several areas in
mathematical logic.
The solution given in the proof of Theorem 1 reveals more information
than is given in the statement of the theorem. This extra information is
important, and we record it as follows:

Theorem 1*. If C combines x with M, then CC is a fixed point of x.

Egocentric Elements

We shall call an element x egocentric if it is a fixed point of itself, i.e. if

xx = x.
 Chapter 8. Beginning Combinatory Logic 175

Problem 2. Prove that if conditions C1 and C2 of the hypothesis of

Theorem 1 hold, then at least one element is egocentric.

Agreeable Elements

We shall say that two elements A and B agree on an element x if Ax = Bx.

We shall call an element A agreeable if for every element B, there is at least
one element x on which A and B agree. Thus A is agreeable if, for every
B, Ax = Bx for at least one x.

Problem 3. (a) [A Variant of Theorem 1] Suppose we are given condition C1

of the hypothesis of Theorem 1 [i.e. for every A and B, there is some C that
combines A with B], but instead of being given that there is a mocker M , we
are given that at least one element A is agreeable. Prove that every element has
a fixed point. (b) Theorem 1 is really but a special case of (a). Why?

Problem 4. As an exercise, show that if the composition condition C1

holds, and if C combines A with B, then if C is agreeable, so is A.

Problem 5. Again, assuming the composition condition C1 , show that for

any elements A, B and C, there is an element E such that Ex = A(B(Cx))
for every x.

Compatible Elements

We will say that an element A is compatible with B if there are elements x

and y such that Ax = y and By = x. When such a pair of elements (x, y)
exists, it is called a cross point of the pair (A, B).

Problem 6. Prove that if conditions C1 and C2 of Theorem 1 hold, then any

two elements A and B are compatible, i.e. any two elements have a cross point.

Smug Elements

We shall call an element smug if it is compatible with itself.

Problem 7. If an element has a fixed point, is it necessarily smug?

Problem 8. Prove that if the composition condition C1 holds and if there

is also at least one smug element, then there is at least one element that
has a fixed point.
176 Part III: Elements of Combinatory Logic

Fixation

We shall say that the element A is fixated on B if Ax = B for every

element x.

Problem 9. Is it possible for an element A to be fixated on more than one

element?

Problem 10. Which, if either, of the following two statements are true?
(a) If y is a fixed point of x, then x is fixated on y.
(b) If x is fixated on y, then y is a fixed point of x.

Narcissistic Elements

We shall call an element A narcissistic if it is fixated on itself, i.e. if Ax = A

for every x.

Problem 11. Which, if either, of the following statements is true?

(a) Every narcissistic element is egocentric.
(b) Ever egocentric element is narcissistic.

Problem 12. Determine whether or not the following statement is true: If

A is narcissistic, then for all elements x and y, Ax = Ay.

Problem 13. If A is narcissistic, does it necessarily follow that for all x

and y, Axy = A?

Problem 14. Narcissism is contagious! If A is narcissistic, then so is Ax

for every x. Prove this.

Introducing the Combinator K, the Kestrel

We shall call a combinator K a cancellator, or a kestrel, if

Kxy = x
for every x and y. This K plays a key role in combinatory logic, as we shall
see.
Note: Why the alternative word “kestrel”? Well, in my book To Mock
a Mockingbird [Smullyan, 1985], I used talking birds as combinators. If one
 Chapter 8. Beginning Combinatory Logic 177

calls out the name of a bird y to a bird x the bird x calls back the name
of some bird, and this bird I call xy. Thus xy is the bird named by x upon
hearing the name of y. I gave bird names to the standard combinators
of the literature, usually names whose first letter is the first letter of the
combinator in the literature. For example, I call the bird representing the
standard combinator K by the name kestrel. The standard combinators
B, C, W and T , which we will soon come across, I respectively called the
bluebird, cardinal, warbler and thrush. I called the mocker M a mockingbird,
because in my bird model, M x = xx means that M ’s response to x is the
same as x’s response to x.
To my surprise, I recently found out that many of my bird names have
been adopted in the more recent combinatory literature! For example, there
is a book by Reginald Braithwaite entitled Kestrels, Quirky Birds and Hope-
less Egocentricity [2013]. In light of this, I feel justified in sometimes using
bird names for various combinators.
In what follows, K is a kestrel.

Problem 15. Prove that any fixed point of K is egocentric.

Problem 16. Prove that if Kx is egocentric, then x is a fixed point of K.

Problem 17. Prove that if Kx is a fixed point of K, then x is a fixed point

of K.

Problem 18. In general, it is not the case that if Ax = Ay, then x must
necessarily be y, but it is the case if A is a kestrel. Prove that if Kx = Ky,
x must necessarily be y.
The principle expressed in Problem 18 (that if Kx = Ky, x must nec-
essarily be y) is important and will be called the cancellation law.

Problem 19. Prove that for all x, Kx 6= K.

Problem 20. (a) Prove that no kestrel can be egocentric. (b) Prove that
any fixed point of a kestrel is narcissistic.
Next I want to show that if C contains a kestrel, then C contains
infinitely many elements (recall that C always contains more than one
element).

Problem 21. Consider the set {K, KK, KKK, . . .}. Is this set infinite?
178 Part III: Elements of Combinatory Logic

Problem 22. What about the set {K, KK, K(KK), K(K(KK)), . . .}, i.e.
the set {K1 , K2 , . . . , Kn , . . .} where K1 = K and Kn+1 = K(Kn ) for all n.
Is this set infinite?

Introducing the Combinator L, the Lark

We shall call a combinator L a lark if for all x and y, the following condition
holds:
Lxy = x(yy).
The lark was not a standard combinator in the literature prior to the pub-
lication of my To Mock a Mockingbird, but it is so now. Several papers
have been written about it. The lark plays a key role in my treatment of
combinatory logic.
One nice thing about the lark is that without assuming conditions C1
and C2 , the mere existence of a lark ensures that every element has a fixed
point.

Problem 23. Prove this.

Problem 24. Prove that no element can be both a lark and a kestrel.

Problem 25. Prove that if a lark L is narcissistic, then L must be a fixed

point of every element x.

Problem 26. Prove that no kestrel can be a fixed point of a lark.

However, it is possible for a lark to be a fixed point of a kestrel.

Problem 27. Show that if a lark L is a fixed point of a kestrel, then L

must be a fixed point of every element.
An interesting thing about larks is that the existence of a lark L, without
any other information, is enough to guarantee the existence of an egocentric
element. In fact, we can write down an expression using only the letter L
and parentheses that names an egocentric element. The shortest expression
I have been able to find uses twelve occurrences of the letter L. Is there a
shorter one? I raised this as an open problem in To Mock a Mockingbird,
and I don’t know if this problem has been solved yet.

Problem 28. Given a lark L, prove that there is an egocentric element,

and then write down the name of such an element, using only the letter L
and parentheses.
 Chapter 8. Beginning Combinatory Logic 179

Discussion

Given two expressions involving only L and parentheses, is there a purely

mechanical procedure to decide whether or not they name the same ele-
ment? I posed this as an open problem in To Mock a Mockingbird, and it
has been solved affirmatively by at least three people — Richard Statman
in his paper “The Word Problem for Smullyan’s combinatorial lark is decid-
able [1989], and by M. Sprenger and M. Wymann-Böni in “How to decide
the lark” [1993].

Introducing the Combinator I,

the Identity Combinator

By an identity combinator is meant an element I satisfying the condition

that Ix = x for every x.
Despite its apparently lowly status, the identity operator I will later
be seen to play an important role. Meanwhile, here are some trivial facts
about I:
(1) I is agreeable if and only if every element has a fixed point.
(2) If every pair of elements is compatible, then I is agreeable.
(3) I, though egocentric, cannot be narcissistic.

Problem 29. Prove these three facts. [It is not necessary to assume the
composition principle, but it is necessary to use the fact that C contains
more than one element.]

Problem 30. Prove that if C contains a lark L and an identity element I,

then C contains a mocker M .

Sages

An element θ will be called a sage if its existence not only guarantees that
every element has a fixed point, but θ is also such that when applied to any
element x, it yields a fixed point of x. Thus, θ is a sage, if, for every x, θx
is a fixed point of x, i.e., x(θx) = θx for every element x.
In the classic literature of combinatory logic, sages are called fixed point
combinators. I introduced the word sage in my book To Mock a Mockingbird,
and the term sage has subsequently become standard in the literature, so
I will use the term sage here. Much study has been devoted to sages, and
180 Part III: Elements of Combinatory Logic

in a later chapter, I will show many ways of deriving a sage from various
other combinators. For now, let me just state that if C contains a lark L
and a mocker M , and if the composition law holds, them C contains a sage.

Problem 30. Prove this.

Note: It follows from the last problem that if C contains a lark L and
an identity combinator I, and obeys the composition condition, the C con-
tains a sage, because in that case C must also contain a mocker M by
Problem 30.

Solutions to the Problems of Chapter 8

1. Let C be an element that combines A with M . Thus, for every element

x, we have Cx = A(M x). But M x = xx, and so Cx = A(xx) for
every x. Now take C for x, and we see that CC = A(CC) which shows
us that CC is a fixed point of A. [Clever, eh?]
2. From conditions C1 and C2 , there is a mocker M . Moreover, every
element has a fixed point [by Theorem 1]. Thus M itself has a fixed
point E. Thus M E = E. Also M E = EE [since M is a mocker], and
so E = EE, which means that E is egocentric. Thus any fixed point of
M is egocentric.
Remark. Since E is egocentric and M E = E, it follows that M E is
egocentric. Doesn’t the world “ME” tell its own tale?
3. (a) Suppose A is agreeable. Given any element x, by condition C1 there
is some element C that combines x with A. Thus Cy = x(Ay) for
every element y. Since A is agreeable, then there is some y ∗ such
that Cy ∗ = Ay ∗ , and hence it follows from Cy ∗ = x(Ay ∗ ) that
Ay ∗ = x(Ay ∗ ), and so Ay ∗ is a fixed point of x.
(b) Theorem 1 is a special case of (a), since a duplicator M is obviously
agreeable: M agrees with every element x on some element, namely
on x [since M x = xx].
4. We are given that the composition condition C1 holds and that C com-
bines A with B and that C is agreeable. We are to show that A is
agreeable. Well, consider any element D. We are to show that A agrees
with D on some element. Well, let E be an element that combines D
with B. Now, C agrees with E on some element F , since C is agreeable.
We will show that A agrees with D on BF .
 Chapter 8. Beginning Combinatory Logic 181

Since C agrees with E on F , then CF = EF . Also EF = D(BF ),

since E combines D with B. Thus CF = EF = D(BF ), and so
CF = D(BF ). But also CF = A(BF ), since C combines A with B,
and so A(BF ) = D(BF ). Thus A agrees with D on BF .
5. We assume that the composition condition C1 holds. Given A, B and
C, let D combine B with C. Thus for any x, Dx = B(Cx). Now let E
combine A with D. Then Ex = A(Dx) = A(B(Cx)).
6. For any elements A and B, we are given that there is some C that
combines A with B. Thus for all y, we have Cy = A(By). We are also
given that there is a mocker M , and so C has some fixed point y. Thus
y = Cy and Cy = A(By), yielding y = A(By). Now take x to be By,
so that y = Ax and x = By. Thus Ax = y and By = x.
7. Of course it is: Suppose x is a fixed point of A, so that Ax = x. Now take
y to be x itself. Then Ax = y and Ay = x, which means that A is smug.
8. We are given that the composition condition C1 holds and that there is
a smug element A. Thus there are elements x and y such that Ax = y
and Ay = x. Since Ax = y, we can substitute Ax for y in the equation
Ay = x, yielding A(Ax) = x. Now let B be an element that combines A
with itself. Thus Bx = A(Ax), and since A(Ax) = x, we have Bx = x.
Thus B has the fixed point x.
9. Of course not! If A is fixed on x and A is fixed on y, then for any
element z we see that Az = x and Az = y so that x = y.
10. It is obviously (b) that is true. Suppose x is fixated on y. Then xz = y
for every z. Hence, taking y for z, we see that xy = y, so that y is a
fixed point of x.
11. It is obviously (a) that is true. Suppose A is narcissistic. Then Ax = A
for every x, so of course AA = A.
12. As given, the statement is obviously true: If A is narcissistic, then
Ax = A and Ay = A. Thus Ax and Ay are both equal to A, so that
Ax = Ay.
13. Yes, it does: If A is narcissistic, then Ax = A, so that Axy = Ay. But
also Ay = A, yielding Axy = A.
14. Suppose A is narcissistic. Then by the last problem Axy = A. Thus
(Ax)y = A for every y, which means that Ax is narcissistic.
15. Suppose Kx = x (i.e that x is a fixed point of K). Then Kxx = xx.
But also Kxx = x (since Kxy = x for every y). Thus Kxx is
182 Part III: Elements of Combinatory Logic

equal to both xx and to x, so that xx = x, which means that x is

egocentric.
16. Suppose Kx is egocentric. Thus Kx(Kx) = Kx. Also Kx(Kx) = x
(since Kxy = x for every y). Thus Kx(Kx) is equal to both Kx and
x, so that Kx = x.
17. Suppose Kx is a fixed point of K. Then Kx is egocentric by Problem 15.
Hence x is a fixed point of K by Problem 16.
18. Suppose Kx = Ky. Now, Kxz = x for every z. Since Kx = Ky, it
follows that Kyz = x for every z. But also Kyz = y. Hence x = y.
19. We will show that if Kx = K, then for any elements y and z, it must
be the case that y = z, which is contrary to the assumed condition that
C contains more than one element.
Well, suppose Kx = K. Then for any y and z, Kxy = Ky and
Kxz = Kz. But Kxy = x and Kxz = x, so that Ky = Kz. Then by
the cancellation law (see Problem 18), y = z.
20. (a) This is immediate from the last problem: Since Kx 6= K for all x,
then KK 6= K, so that K is not egocentric.
(b) Suppose KA = A (i.e. A is a fixed point of K). Then Ax = KAx
for all x. But also KAx = A for all x. Thus Ax = A for all x, which
means that A is narcissistic.
21. Let U1 = K; U2 = KK; U3 = KKK, etc. We show that the set
{U1 , U2 , U3 , . . . , Un , . . .} is far from infinite. It contains only K and
KK. We see this as follows:
If Un = K, then Un+2 = KKK, which is K (because K is a kestrel).
Thus if Un = K, then Un+2 = K, and, since U1 = K, then Un = K for
every odd number n.
Also, if Un = KK then Un+1 = KKK which is K (again because K
is a kestrel). Thus, if Un = KK then Un+2 = KK and since U2 = KK
it follows that Un = KK for every even n. Therefore we see that, for
all n, Un = K or Un = KK.
22. This is a very different story! Let K1 = K, K2 = KK; K3 = K(KK)
[which is K(K2 )], etc. Thus for each n, we have Kn+1 = K(Kn ). The
set {K1 , K2 , . . . , Kn , . . .} is indeed infinite, as we will see.

We will show that for any numbers n and m, if n 6= m, then

Kn 6= Km . To show this, it suffices to show that if n < m, then
 Chapter 8. Beginning Combinatory Logic 183

Kn 6= Km (because if n 6= m, then either n < m or m < n). Thus

it suffices to show that for any positive integers n and b, Kn 6= Kn+b .
By Problem 19, K 6= K(Kb ); in fact K 6= Kx for any x. Since
K = K1 and K(Kb ) = Kb+1 , we have:

(1) K1 6= K1+b .

Next, by the cancellation law (Problem 18), if Kx = Ky, then x = y.

Equivalently, if x 6= y, then Kx 6= Ky. Thus if Kn 6= Km , then
KK n 6= KK m , and so we have:

(2) If Kn 6= Km , then Kn+1 6= Km+1 .

Now, by (1), we have K1 6= K1+b . Hence by repeated applications of (2),

we see that K2 6= K2+b , K3 6= K3+b , . . . , Kn 6= Kn+b . This concludes
the proof.
23. We consider a lark L. For any x and y, (Lx)y = x(yy). Now, taking Lx
for y, we see that (Lx)(Lx) = x((Lx)(Lx)). Thus (Lx)(Lx) is a fixed
point of x! Note that (Lx)(Lx) is the same element as Lx(Lx) by how
we define the use of parentheses (or rather the lack thereof).
24. Consider a kestrel K. For any x and y, Kxy = x. Taking K for both x
and y, we have:
(1) KKK = K.

If K were a lark, then Kxy = x(yy) for all x and y, so that for x = K
and y = K, we would have:

(2) KKK = K(KK).

For (1) and (2) it would follow that K = K(KK), violating the fact
that for no x can K = Kx (by Problem 19). Thus K cannot be a lark.
25. Suppose that a lark L is narcissistic. Then Lxy = L for every x and y
by the solution to Problem 13. Taking Lx for y, we obtain Lx(Lx) = L.
But Lx(Lx) is a fixed point of x [by the solution to Problem 23]. Thus
L is a fixed point of x.
26. Suppose that LK = K (i.e. that K is a fixed point of L). Then
LKK = KK. But LKK = K(KK) (since L is a lark), so that
K(KK) = KK. Then by the cancellation law, taking KK for x and K
for y, we see that KK = K, which means that K is egocentric, which
is contrary to Problem 20. Therefore K cannot be a fixed point of L.
184 Part III: Elements of Combinatory Logic

27. If the lark L is a fixed point of the kestrel K, then KL = L. So, for any
x, KLx = Lx. But KLx = L, since K is a kestrel. So we have Lx = L
for all x, and L is narcissistic. Thus L is a fixed point of every element
x by Problem 25.
28. We first show that if x is any fixed point of LL, then xx must be
egocentric.
Well, suppose that LLx = x. Now LLx = L(xx) [since L is a lark].
Thus x = L(xx). Therefore xx = L(xx)x, which is xx(xx). Thus xx is
egocentric.
We can actually compute a fixed point of LL, in terms of just L:
As we saw in the solution of Problem 23, for any x, a fixed point
of x is Lx(Lx). Taking LL for x, we see that a fixed point of LL is
L(LL)(L(LL)), and therefore an egocentric element is L(LL)(L(LL))
repeated, which is:
L(LL)(L(LL))(L(LL)(L(LL))).
29. (1) Suppose I is agreeable. Then for any x there is some y such that
Iy = xy. But Iy = y, so that y = xy, and thus y is a fixed point of x.
Conversely, suppose that every element x has a fixed point y. Then
xy = y. But y = Iy, so that xy = Iy, and thus I agrees with x on y.
(2) Suppose that every pair of elements is compatible. Now consider
any element A. Then A is compatible with I. Thus there are
elements x and y such that Ax = y and Iy = x. Since Iy = x,
then y = x (since y = Iy). Thus Ax = y and y = x and it follows
that Ax = x, which means that x is a fixed point of A. Thus every
element has a fixed point, and consequently I is agreeable by (1).
(3) Since Ix = x for every x, it follows that II = I, which means that I
is egocentric. If I were narcissistic, that would mean that for every
x, Ix = I. But also Ix = x. Hence if I were narcissistic, then we
would have x = I for every element x, contrary to the fact that C
has more than one element.
30. Take M to be LI. Well, LIx = I(xx) = xx. Thus LI is a mocker.
31. We will see that any element θ that combines M with L must be a sage.
We already know that Lx(Lx) is a fixed point of x (by the solution
to Problem 23). Now, Lx(Lx) = M (Lx), and consequently M (Lx) is
a fixed point of x. Now, suppose that θ combines M with L. Then
θx = M (Lx), and so θx is a fixed point of x. Thus θ is a sage.
 Chapter 9

Combinatorics Galore

We now discuss some of the most prominent combinators of the literature,

as well as some which I introduced in To Mock a Mockingbird.

I. The B-Combinators

Introducing the B Combinator, the Bluebird

The B-combinator is a combinator satisfying the following condition (for

all x, y and z):
Bxyz = x(yz).
This combinator is also known as the bluebird, a name I introduced in
To Mock a Mockingbird. I shall sometimes use that name. This combinator
is one of the basic combinators.

Problem 1. Suppose C contains a bluebird B and a mocker M . Then the

composition condition must hold (why?). Hence, by Theorem 1 of the last
chapter, every element x has a fixed point. Moreover, one can write down
an expression for a fixed point of x, using just the symbols B, M and x
(and parentheses, of course). Write down such an expression.

Problem 2. Show that if B and M are present (as members of C), then
some element of C is egocentric; in fact, write down the name of an egocen-
tric element in terms of B and M .

Problem 3. Now write down a narcissistic element in terms of B, M and

K (kestrel).
186 Part III: Elements of Combinatory Logic

Some Derivatives of B
∧
From B alone, one can derive combinators D, B1 , E, B2 , D1 B + , D2 , E ,
which are defined by the following conditions:
(a) Dxyzw = xy(zw), the dove
(b) B1 xyzw = x(yzw)
(c) Exyzwv = xy(zwv), the eagle
(d) B2 xyzwv = x(yzwv)
(e) D1 xyzwv = xyz(wv)
(f) B + xyzw = x(y(zw))
(g) D2 xyzwv = x(yz)(wv)
∧
(h) E xy1 y2 y3 z1 z2 z3 = x(y1 y2 y3 )(z1 z2 z3 ).
All these combinators, including B, are called compositors, since they
serve to introduce parentheses. The standard ones are the bluebird B and
D, which I call the dove. We shall also have good use for E, which I call
the eagle.

Problem 4. (1) Derive these combinators from B. [Hints: Do the following

in order:
(a) Express D in terms of B.
(b) Express B1 in terms of B and D (which can then be reduced to an
expression in terms of B alone).
(c) Express E in terms of B and B1 .
(d) Express B2 in terms of B and E.
(e) Express D1 in terms of B and B1 , or, alternatively, in terms of B
and D.
(f) Express B + in terms of B and D1 .
(g) Interestingly, D2 can be expressed in terms of D alone!
∧
(h) Express E in terms of E alone!]

(2) Using mathematical induction, show that for each positive integer n,
there is a combinator Bn , derivable from B, satisfying the condition

Bn xy1 . . . yn+2 = x(y1 . . . yn+2 ), n ≥ 0.

Note that by this definition, the bluebird B also has the name B0 , and B1
and B2 are the same as defined above (where different variables were used
in the definition).
 Chapter 9. Combinatorics Galore 187

II. The Permuting Combinators

We now come to an interesting family of combinators known as permuters.

Introducing the Combinator T, the Thrush

The simplest permuting combinator is T , which is defined by the

following:
T xy = yx.
This permuter T is standard. In To Mock a Mockingbird, I called it the
thrush.
Two elements x and y are said to commute if xy = yx.

Problem 5. Prove that from the thrush T and the lark L one can derive
a combinator A that commutes with every element x.

Introducing the Combinator R, the Robin

In To Mock a Mockingbird, I introduced a combinator R, which I termed a

robin. It was defined by the condition:
Rxyz = yzx.

Problem 6. Show that a robin R can be derived from B and T .

Introducing the Combinator C, the Cardinal

The combinator C is defined by the condition:

Cxyz = xzy.
This C is standard in the literature, and of basic importance. I called
it the cardinal in To Mock a Mockingbird. It is derivable form B and T ,
as was discovered by Alonzo Church [1941]. Church’s construction was
quite tricky! It used eight letters, and I doubt that it can be done with
fewer.
Having derived R from B and T , it is quite simple to get C.

Problem 7. How can one derive C from R?

188 Part III: Elements of Combinatory Logic

Problem 8. The solution of Problem 7, when reduced to B and T has nine

letters. It can easily be shortened to eight, thus yielding Church’s expression
for C. Do this reduction.

Problem 9. Can a thrush T be derived from C and the identity combi-

nator I?

Problem 10. We have seen that C is derivable from R alone. Can one
derive R from C alone?
Note: Taking BBT for R, and RRR for C, a useful fact to observe is
that for any x:
Cx = B(T x)R
because Cx = RRRx = RxR = BBT xR = B(T x)R.

Introducing the Combinator F, the Finch

A finch F is defined by the following condition:

F xyz = zyx.
A finch F can be derived from B and T in several ways. It can be derived
from B and R, or from B and C, or from T and the eagle E.

Problem 11. (a) It is easiest to derive F from all three of B, R and C. How?
[Then, of course, F is derivable from B and R, or from B and C, since R and C
and inter-derivable.] (b) Show how to derive F from B and E.

Introducing the Combinator V, the Vireo

A vireo is a combinator V satisfying the following condition:

V xyz = zxy.
The combinator V has a sort of opposite effect to that of the finch F . V is
derivable from B and T . It is easiest to derive it from C and F .

Problem 12. Derive V from C and F .

Problem 13. We have seen that V is derivable from C and F . It is also

true that F is derivable from C and V . How?

Problem 14. As a curiosity, show that the identity combinator I is deriv-

able from R and K (kestrel).
 Chapter 9. Combinatorics Galore 189

Some Relations

Given a permuting combinator A satisfying the condition Axyz = abc,

where abc is some permutation of xyz, by A∗ is meant the combinator
satisfying the condition:
A∗ wxyz = wabc.
Thus C ∗ , R∗ , F ∗ , V ∗ are combinators defined by the following conditions:
C ∗ wxyz = wxzy,
R∗ wxyz = wyzx,
F ∗ wxyz = wzyx,
V ∗ wxyz = wzxy.

Each of these combinators is derivable from B and T . It is easiest to

derive them from combinators already derived from B and T .

Problem 15. Derive:

(a) C ∗ from B and C;
(b) R∗ from B and C;
(c) F ∗ from B, C ∗ and R∗ ;
(d) V ∗ from C ∗ and F ∗ .
Consider again a permutation combinator A satisfying the condition
Axyz = abc, where abc is a permutation of xyz. We let A∗∗ be the
combinator defined by the condition
A∗∗ vwxyz = vwabc.
Thus C ∗∗, R∗∗, F ∗∗, V ∗∗ are combinators defined by the following
conditions:
C ∗∗ vwxyz = vwxzy,
R∗∗ vwxyz = vwyzx,
F ∗∗ vwxyz = vwzyx,
V ∗∗ vwxyz = vwzxy.

Problem 16. Show that C ∗∗ , R∗∗ , F ∗∗ , V ∗∗ are all derivable from B

and T .
The problem is far simpler than it might appear. One can handle all
four cases in one fell swoop!
190 Part III: Elements of Combinatory Logic

Vireos Revisited

We have derived V from C and F , obtaining an expression, which, when

reduced to B and T , has 16 letters. It can alternatively be derived from C ∗
and T , yielding an expression, which, when reduced to B and T , has only
ten letters. How?

Problem 17. Derive V from C ∗ and T .

Problem 18. We have not yet considered a combinator A satisfying the

condition Axyz = yxz. Is there one derivable from B and T ?

III. The Q-Family and the Goldfinch, G

We now come to a family of combinators, all derivable from B and T , that

both permute and introduce parentheses. The first member is Q defined by
the condition:
Qxyz = y(xz).

I termed Q the queer bird in To Mock a Mockingbird.

Problem 19. Derive Q from B and T . [Hint: There is a two-letter

expression involving B and one other combinator already derived from B
and T that works!]

Q1 and Q2

The two combinators Q1 and Q2 are defined by the following conditions:

Q1 xyz = x(zy),
Q2 xyz = y(zx).

In To Mock a Mockingbird, the combinators Q1 and Q2 were respectively

called the quixotic and the quizzied birds.

Problem 20. Derive Q1 and Q2 from B and T , or from any combinators

already derived from them.
There is an old Chinese proverb that says that if a cardinal C is present,
then you cannot have a quixotic bird without a quizzied bird, nor a quizzied
 Chapter 9. Combinatorics Galore 191

bird without a quixotic bird, or if there isn’t such a Chinese proverb, then
there should be!

Problem 21. What is the sense behind such a proverb?

The Birds Q3 (Quirky), Q4 (Quacky),

Q5 (Quintessential), Q6 (Quivering)

The quirky bird, Q3 , the quacky bird, Q4 , the quintessential bird, Q5 , and
the quivering bird, Q6 , are defined by the following conditions:
Q3 xyz = z(xy),
Q4 xyz = z(yx),
Q5 xyzw = z(xyw),
Q6 xyzw = w(xyz).

Problem 22. Derive these from combinators derivable from B and T .

There is, or should be, another Chinese proverb which says that if a
cardinal is present, then you cannot have a quirky bird without a quacky
bird, nor a quacky bird without a quirky bird.

Problem 23. What is the sense behind that one?

Problem 24. Show that Q4 can be derived from T and Q1 .

Problem 25. We have seen that Q is derivable from B and T . But also,
B is derivable from Q and T . Prove this. [It is not obvious.]

Problem 26. One can derive C from Q and T more easily than from B and
T , in fact, with an expression of only four letters. How can this be done?

The Combinator G, the Goldfinch

In To Mock a Mockingbird I introduced a combinator G, called the goldfinch,

which I found quite useful. It is defined by the condition:
Gxyzw = xw(yz).
I do not know if this combinator was previously known or not.

Problem 27. Derive G from B and T , or from combinators already derived

from B and T .
192 Part III: Elements of Combinatory Logic

IV. Combinators Derivable from B, T, M

and I (λ-I Combinators)

Combinators derivable from B, T, M and I form an extremely important

class, whose significance will be discussed in a later chapter. These combi-
nators are known as λ–I combinators.
A useful combinator M2 is defined by the condition
M2 xy = xy(xy).
It is easily derivable from M and B.

Problem 28. Do so.

The Combinator W, the Warbler and

the Combinator W 0 , the Converse Warbler

A standard combinator is W , which is defined by the condition

W xy = xyy.
W should not be confused with the lark L: Lxy = x(yy).
Alonzo Church showed how to derive W from B, T, M and I. His deriva-
tion was both bizarre and ingenious. His expression for W involved 24 let-
ters and 14 pairs of parentheses! Fairly soon afterwards, J. B. Rosser found
an expression for W in terms of just B, T and M that included only ten
letters.
I called W the Warbler in To Mock a Mockingbird. Before deriving it
from B, T and M , it will prove convenient to first consider another combi-
nator W 0 , which might be called a converse Warbler. W 0 is defined by the
condition
W 0 xy = yxx.

Problem 29. There are two interesting ways of deriving W 0 from B, T

and M . One way is to first derive W 0 from M2 and the robin R, and
then reduce that expression to B and T , which should yield an expres-
sion of five letters. Another way is to first derive W 0 from B, T and M2 ,
which when reduced to B, T and M yields a different expression, one
that also contains five letters. The reader should try to find both these
expressions!
 Chapter 9. Combinatorics Galore 193

Problem 30. One can easily derive W from W 0 and the cardinal C, getting
an expression which, when reduced to B, T and M , will have 13 letters.
However, it can be further reduced to an expression having only ten letters
in two different ways, depending on the choice for W . Do these things.

Problem 31. M is obviously derivable from W and T , and also from W

and I. Also, I is derivable from W and K. Find these derivations.

Problem 32. Some useful relatives of W are W1 , W2 , W3 , defined by the

following conditions:
W1 xyz = xyzz,
W2 xyzw = xyzww,
W3 xyzwv = xyzwvv.

Show that these are all derivable from B, T and M , in fact from W
and B.

The Combinator H, the Hummingbird

I have found good use for a combinator H (dubbed the hummingbird in

To Mock a Mockingbird). The combinator H is defined by the condition

Hxyz = xyzy.

Problem 33. Show that H is derivable from B, C and W , and thus from
B, M and T .

Problem 34. One can also derive W from H and R, or, more neatly, from
C, H and R. How? [Hint: First derive W 0 .]

Larks Revisited

We recall the lark L defined by the condition Lxy = x(yy). The lark can
be derived from B, T and M in many ways.

Problem 35.
(a) Show that L is derivable from B, R and M or from B, C and M , and
then reduce the latter expression to one in terms of B, M and T .
(b) Show that L is derivable from B and W . This fact is rather
important.
194 Part III: Elements of Combinatory Logic

(c) My favorite derivation of L is from M and the queer combinator Q. It

is also the simplest. When reduced to B, M and T , we get the same
expression as in (a). Try this!

The Combinator S, the Starling

One of the most important combinators in the literature of Combinatory

Logic is the combinator S, which is defined by the following definition:
Sxyz = xz(yz).
I called S the starling in To Mock a Mockingbird. One reason why S is so
important is that from S and K one can derive all possible combinators, in
a sense I will make precise in a later chapter.
The starling is derivable from B, M and T , and more easily from B, C
and W . The standard expression for S in terms of B, C and W has seven
letters, but in To Mock a Mockingbird, I found another expression having
only six letters. For this I used the goldfinch G, which we recall is defined
by the condition Gxyzw = xw(yz).

Problem 36. Derive S from B, W and G, and then reduce the resulting
expression to B, C and W . [Hint: Use W2 , which we learned from the
solution to Problem 32 can be expressed as B(BW ).]
We recall that the hummingbird H was defined by the condition
Hxyz = xyzy.

Problem 37. Derive H from S and R.

Problem 38. Show that both W and M are derivable from S and T .

Problem 39. Express a warbler W in terms of S and C.

Note: We now see that the class of combinators derivable from B, C
and W is the same as the class of combinators derivable from B, C and S,
since S is derivable from B, C and W [by Problem 36], and W is derivable
from S and C [by Problem 39].

The Order of a Combinator

By a combinator of order one is meant a combinator A such that Ax can

be expressed in terms of x alone (a single variable and no combinators).
 Chapter 9. Combinatorics Galore 195

For example, M is of order 1, since M x = xx, and the expression xx does

not involve the letter M . Another example is the identity combinator I,
since Ix = x. Another is W L, since W Lx = Lxx = x(xx).
By a combinator of order two is meant one whose defining condition
involves just two variables (and no combinators), as is the case for W and L.
In general, for any positive integer n, by a combinator of order n is meant
one whose defining condition involves n variables. The permuting combina-
tors C, R, F and V are of order three, as are B and Q.
It is not true that every combinator has an order, for T I cannot be of
any order n, since T Ix1 , . . . , xn = x1 Ix2 , . . . , xn , which cannot be reduced
any further (and so we cannot get rid of the combinator I on the right side
of this equation). On the other hand IT has an order, since it is equal to
T , which is of order two.
A useful fact to note is that for any combinators A1 and A2 , the com-
binator RA1 A2 is equal to CA2 A1 , since RA1 A2 x = A2 xA1 , and also
CA2 A1 x = A2 xA1 .

The P Group of Combinators

Problem 40. We shall have good use for some of the combinators P, P1 , P2
and P3 satisfying the following conditions:
(a) P xyz = z(xyz),
(b) P1 xyz = y(xxz),
(c) P2 xyz = x(yyz),
(d) P3 xyz = y(xzy).
These can be derived from B, M and T , in fact from B, Q and W (and
hence from B, C and W ). How?

Problem 41. Show that M is derivable from P and I.

The Φ Group of Combinators

We will later need combinators Φ, Φ2 , Φ3 , Φ4 , defined by the following

conditions:
Φxyzw = x(yw)(zw),
Φ2 xyzw1 w2 = x(yw1 w2 )(zw1 w2 ),
196 Part III: Elements of Combinatory Logic

Φ3 xyzw1 w2 w3 = x(yw1 w2 w3 )(zw1 w2 w3 ),

Φ4 xyzw1 w2 w3 w4 = x(yw1 w2 w3 w4 )(zw1 w2 w3 w4 ).

Problem 42. (a) Derive Φ from B and S. (b) Show by induction that for
each n, there is a combinator Φn derivable from B and S satisfying the
equation:
Φn xyzw1 . . . wn = x(yw1 . . . wn )(zw1 . . . wn ).

Solutions to the Problems of Chapter 9

1. To begin with, let us note that if B is present, then the composi-

tion condition must hold, since for any elements x and y, an ele-
ment that combines x with y is Bxy [since (Bxy)z is Bxyz, which
is x(yz)].
We recall from the solution of Problem 1 of the last chapter that if C
is any element that combines x with M , then CC is a fixed point of x
(we stated this result as Theorem 1∗ ). Well, BxM combines x with M ,
and so BxM (BxM ) is a fixed point of x. But BxM (BxM ) is also equal
to M (BxM ), by the definition of M [as applied to the first M in the
expansion of M (BxM )]. So M (BxM ) is also a fixed point of x (for every
x!). We will use this result several times in this chapter and the next.
2. Since B is present, the composition condition holds. Then by the solu-
tion of Problem 2 of the last chapter, any fixed point of M is egocen-
tric. Now, M (BM M ) is a fixed point of M [by the previous problem
of this chapter, taking M for x]; hence M (BM M ) is egocentric. Let
us double-check: To reduce clutter, let A be BM M . We are to show
that M A is egocentric. Well, M A = M (BM M ) = BM M (BM M ). But
BM M (BM M ) = M (M (BM M )), since B is a bluebird. Then we have
M (M (BM M )) = M (M A) = M A(M A). Thus M A = M A(M A), and
M A is egocentric.
3. By Problem 20 (b) of the last chapter, any fixed point of a kestrel K is
narcissistic. Since M (BKM ) is a fixed point of K [by the solution to
Problem 1], it must be narcissistic [as the reader can check directly, by
showing that M (BKM )x = M (BKM )].
 Chapter 9. Combinatorics Galore 197

4. (1) (a) We take D to be BB. Let us check that this works:

BBxy = B(xy). Hence BBxyz = B(xy)z. Then
BBxyzw = B(xy)zw = xy(zw).
Thus Dxyzw = xy(zw).
(b) We take B1 to be DB, which can then be expressed in terms of
B alone, since D can be so expressed. We can see this definition
works as follows:
B1 xyz = DBxyz = Bx(yz).
Thus,
B1 xyzw = Bx(yz)w = x((yz)w) = x(yzw).
Since D = BB, then DB = BBB. Thus in terms of B alone
B1 = BBB.
(c) We take E to be BB1 , which in terms of B alone is B(BBB).
We can see this definition works as follows:
Exy = BB1 xy = B1 (xy). Thus
Exyzwv = B1 (xy)zwv = xy(zwv).
(d) We take B2 to be EB (which, in terms of B, is B(BBB)B).
Now,
B2 xyzw = EBxyzw = Bx(yzw).
Hence
B2 xyzwv = EBxyzwv = Bx(yzw)v = x(yzwv).
(e) We can take D1 to be B1 B, or we can take D1 to be BD.
Actually B1 B is the same as BD because
B1 B = BBBB = B(BB) = BD.
It will be quicker to take D1 to be B1 B. Then
D1 xyz = B1 Bxyz = B(xyz).
Hence,
D1 xyzwv = B1 Bxyzwv = B(xyz)wv = xyz(wv).
(f) We take B + to be D1 B. Thus,
B + xyzw = D1 Bxyzw = Bxy(zw) = x(y(zw)).
198 Part III: Elements of Combinatory Logic

(g) We take D2 to be DD. Then

D2 xyzwv = DDxyzwv = Dx(yz)wv = x(yz)(wv).
∧
(h) We take E to be EE. Then
∧
E xy1 y2 y3 z1 z2 z3 = EExy1 y2 y3 z1 z2 z3 = Ex(y1 y2 y3 )z 1 z2 z3
= x(y1 y2 y3 )(z 1 z2 z3 ).
(2) We will show by mathematical induction that we can define all the
combinators
Bn xy1 . . . yn+2 = x(y1 . . . yn+2 ), n ≥ 0,
in terms of the combinator B. Well, we have already noted that
B0 is B itself. Now assume that Bn can be defined in terms of the
combinator B. Then we claim that Bn+1 = BBBn :
Bn+1 xy1 y2 . . . yn+2 yn+3 = BBBn xy1 y2 . . . yn+2 yn+3
= B(Bn x)y1 y2 . . . yn+2 yn+3 = (Bn x)(y1 y2 ) . . . yn+2 yn+3
= x((y1 y2 ) . . . yn+2 yn+3 ) = x(y1 y2 . . . yn+2 yn+3 ).

5. Actually, any fixed point A of T will commute with every element

x. For suppose T A = A. Then for every element x, we see that
Ax = T Ax = xA. Thus A commutes with x.
6. Take R to be BBT . Then
Rxyz = BBT xyz = B(T x)yz = T x(yz) = yzx.
7. Actually C is derivable from R alone! Take C to be RRR. Then:
Cxyz = RRRxyz = RxRyz = Ryxz = xzy.
8. In terms of B and T , C = BBT (BBT )(BBT ), which has nine letters.
It is equal to B(T (BBT ))(BBT ), which is Church’s expression.
9. Easily! Take T to be CI. Then T xy = CIxy = Iyx = yx.
10. Yes. Take R to be CC. Then Rxyz = CCxyz = Cyxz = yzx.
11. (a) BCR is a finch, because
BCRxyz = C(Rx)yz = (Rx)zy = Rxzy = zyx.
In terms of B and R, F = B(RRR)R. In terms of B and C,
F = BC(CC).
 Chapter 9. Combinatorics Galore 199

(b) ET T ET is a finch, because

ET T ET xyz = T T (ET x)yz = T x(T yz) = T yzx = zyx.
We might note that in terms of B and T , the expression BCR for the
finch has 12 letters (since C has 8 and R has 3), while using the expres-
sion B(RRR)R for the finch gives us 13 letters in terms of B and T ;
using the expression BC(CC) for the finch, we get 25 letters in terms
of B and T .
As for F = ET T ET , as it stands, when E is replaced by B(BBB),
we get an expression of 11 letters, but the expression can be reduced
to only 8 letters in terms of B and T , as we will now see:
ET T = B(BBB)T T = BBB(T T ) = B(B(T T )).
Hence,
ET T ET = B(B(T T ))ET = B(T T )(ET ) = B(T T )(B(BBB)T ).
Thus, expressing F in terms of E and T , and then reducing that expres-
sion to one in terms of B and T , yields the shortest expression for the
finch F .
12. Take V to be CF . Then CF xyz = F yxz = zxy.
Note: When reduced to B and T , the expression CF has 16 letters
(since C and F each have 8). This 16-letter expression cannot be
reduced further. However, one can get an expression for V in terms
of B and T which has only ten letters, as we will later see.
13. CVxyz = Vyxz = zyx. Thus CV is a finch, F .
14. For any element A, the element RAK must be an identity combina-
tor, for RAKx = KxA = x. In particular, RKK and RRK are both
identity combinators.
15. Here are the derivations requested for C ∗ , R∗ , F ∗ and V ∗ :
(a) Take C ∗ to be BC: BCwxyz = C(wx)yz = wxzy.
(b) We have just derived C ∗ from B and C, and R∗ can in fact
be derived from C ∗ alone. Thus take R∗ to be C ∗ C ∗ . Well,
C ∗ C ∗ wxy = C ∗ wyx. Hence
C ∗ C ∗ wxyz = C ∗ wyxz = wyzx = R∗ wxyz.
So, in terms of B and C, R∗ = BC(BC).
(c) Take F ∗ to be BC ∗ R∗ . Then,
BC ∗ R∗ wxyz = C ∗ (R∗ w)xyz = R∗ wxzy = wzyx.
200 Part III: Elements of Combinatory Logic

(d) Take V ∗ to be C ∗ F ∗ . Well, C ∗ F ∗ wxyz = F ∗ wyxz = wzxy.

16. Here is the secret! Let A be any of the four combinators C, R, F, V . We
can take A∗∗ to be BA∗ . Here is why:
Axyz = abc, for some permutation abc of xyz.
Then A∗ wxyz = wabc and
A∗∗ vwxyz = BA∗ vwxyz = A∗ (vw)xyz = (vw)abc = vwabc.
17. Take V to be C ∗ T . Well,
C ∗ T xyz = T xzy = zxy.
Thus C ∗ T is a vireo.
18. Of course! The combinator T itself is such a combinator, since
T xy = yx, so that T xyz = yxz.
19. The expression CB works:
CBxyz = Byxz = y(xz) = Qxyz.
We thus take Q to be CB.
20. Take Q1 to be C ∗ B. Well, C ∗ Bxyz = Bxzy = x(zy). Take Q2 to be
R∗ B. Now, R∗ Bxyz = Byzx = y(zx).
21. The fact is that Q2 is derivable from C and Q1 , and Q1 is derivable
from C and Q2 :
CQ1 xyz = Q1 yxz = y(zx) = Q2 xyz;
CQ2 xyz = Q2 yxz = x(zy) = Q1 xyz.
22. Q3 : Take Q3 to be V ∗ B. Well, V ∗ Bxyz = Bzxy = z(xy). Alterna-
tively, we can take Q3 to be BC(CB) or QQC. In terms of B and
T , we could simply take Q3 to be BT .
Q4 : We could take Q4 to be F ∗ B: We see that
F ∗ Bxyz = Bzyx = z(yx).
Alternately, we could take Q4 to be
CQ3 : CQ3 xyz = Q3 yxz = z(yx).
Q5 : Take Q5 to be BQ: Q3 : BQxyzw = Q(xy)zw = z(xyw).
Q6 : Take Q6 to be B(BC)Q5 . Then
Q6 xyzw = B(BC)Q5 xyzw = (BC)(Q5 x)yzw
= C(Q5 xy)zw = Q5 xywz = w(xyz).
 Chapter 9. Combinatorics Galore 201

23. Q4 is derivable from C and Q3 , and Q3 is derivable from C and Q4 :

CQ3 xyz = Q3 yxz = z(yx) = Q4 xyz.
CQ4 xyz = Q4 yxz = z(xy) = Q3 xyz.
24. We can express Q4 as Q1 T :
Q1 T xyz = T (yx)z = z(yx) = Q4 xyz.
25. The bluebird B can be seen to be the same as QT (QQ) as follows:
QT (QQ)xyz = QQ(T x)yz = T x(Qy)z = Qyxz = x(yz) = Bxyz.
26. The cardinal C can be expressed as QQ(QT ):
QQ(QT )xyz = QT (Qx)yz = Qx(T y)z = T y(xz) = xzy = Cxyz.
27. Take G to be BBC. Then
BBCxyzw = B(Cx)yzw = Cx(yz)w = xw(yz).
28. Obviously, take M2 to be BM :
BM xy = M (xy) = xy(xy) = M2 xy.
29. For the first method, we take W 0 to be M2 R. Then
M2 Rxy = Rx(Rx)y = Rxyx = yxx = W 0 xy.
Reducing to B, T and M , we take BM for M2 and BBT for R, obtain-
ing the expression BM (BBT ).
As for the second way to derive W 0 , we take W 0 to be B(M 2 B)T .
In this case,
B(M2 B)T xy = M2 B(T x)y = B(T x)(B(T x))y
= T x(B(T x)y) = B(T x)yx = T x(yx) = yxx = W 0 xy.
Taking BM for M2 , this version of W 0 reduces to B(BM B)T . Thus
we have two ways of expressing W 0 from B, T and M : BM (BBT ) and
B(BM B)T .
30. CW 0 is a warbler, since CW 0 xy = W 0 yx = xyy = W xy.
We now use the fact previously noted that for any x, Cx = B(T x)R
(this was shown right after the statement of Problem 10 in this chap-
ter), and so CW 0 = B(T W 0 )R. Thus B(T W 0 )R is a warbler. Hence
B(T W 0 )BBT is a warbler (since R = BBT ). As seen in the last prob-
lem, we can take W 0 to be either BM (BBT ) or B(BM B)T , and so we
can take W to be either
B(T (BM (BBT )))BBT or B(T (B(BM B)T ))BBT.
The latter is Rosser’s expression for W .
202 Part III: Elements of Combinatory Logic

31. W T x = T xx = xx = M x.
W Ix = Ixx = xx = M x
W Kx = Kxx = x = Ix.
32. We will take W1 to be BW , W2 to be BW1 , W3 to be BW 2 . Then
BW xyz = W (xy)z = xyzz = W1 xyz.
BW1 xyzw = W1 (xy)zw = (xy)zww = xyzww = W2 xyzw.
BW2 xyzwv = W2 (xy)zwv = (xy)zwvv = xyzwvv = W3 xyzwv.
Note: In general, if for all n we take Wn+1 to be BW n , then, for every n,
Wn xy1 . . . yn z = xy1 . . . yn zz
as can be shown by mathematical induction.
33. We take H to be W ∗ C ∗ , which is BW (BC). Then
Hxy = W ∗ C ∗ xy = C ∗ xyy.
Thus,
C ∗ xyyz = xyzy = Hxyz.
34. First we take W 0 to be HR. Thus HRxy = Rxyx = yxx = W 0 xy. Then
take W to be CW 0 . This is a warbler: CW 0 xy = W 0 yx = xyy = W xy.
Thus C(HR) is a warbler, and since C is RRR, the warbler is also
derivable from H and R alone.
35. (a) RMB is a lark, since RM Bxy = BxM y = x(M y) = x(yy) = Lxy.
Also, so is CBM (as the reader can verify). To express RM B in
terms of B, M and T , we take BBT for R in RM B, so that
RM B = BBT M B = B(T M )B.
Thus B(T M )B is a lark.
(b) BW B is also a lark, since
BW Bxy = W (Bx)y = Bxyy = x(yy) = Lxy.
(c) QM is a lark, since QM xy = x(M y) = x(yy) = Lxy. Reducing
this to B, M and T , we see that QM = CBM [since Q = CB by
Problem 19]. Also, for any x and y, it is a fact that Cxy = Ryx,
because [by Problem 7] Cxy = RRRxy = RxRy = Ryx. Thus
Cxy = Ryx. In particular, CBM = RM B, so that QM = RM B,
which is the expression we obtained for L in (a), and which reduced
to B(T M )B.
 Chapter 9. Combinatorics Galore 203

36. Recall that W2 was defined by the condition W2 xyzw = xyzww in the
statement of Problem 32 and that in the solution to Problem 32 we
saw W2 to be B(BW ), i.e. defined in terms of B and W . We can take
the starling S to be W2 G, since W2 Gxyz = Gxyzz = xz(yz) = Sxyz.
Since we saw in the solution to Problem 27 that G = BBC, we also see
that S is B(BW )(BBC) in terms of to B, C and W .
37. SR is a hummingbird, since SRxyz = Ry(xy)z = xyzy = Hxyz.
38. (a) ST is a warbler, since ST xy = T y(xy) = xyy = W xy.
(b) ST T is a mocker, since ST T x = T x(T x) = T xx = xx = M x.
Another way to see this is based on our having seen that W T in a
mocker [in the solution to Problem 31], and since we can take ST
for W [because ST xy = T y(xy) = xyy = W xy], again we see that
STT is a mocker.
39. By Problem 34, C(HR) is a warbler W . By the solution to Problem 37,
SR is a hummingbird (H). Thus C(SRR) is a warbler. But by the solu-
tion to Problem 10, CC is a robin R, and so, for W , we can take
C(S(CC)(CC)).
40. (a) Take P to be W2 Q5 . Then W2 Q5 xyz = Q5 xyzz = z(xyz) = P xyz.
(b) Take P1 to be LQ. Then LQxyz = Q(xx)yz = y(xxz) = P1 xyz.
(c) Take P2 to be CP1 . Then CP1 xyz = P1 yxz = x(yyz) = P2 xyz.
(d) Take P3 to be BCP . Then
BCP xyz = C(P x)yz = P xzy = y(xzy) = P3 xyz.
41. Take M to be P II. Then P IIx = x(IIx) = x(Ix) = xx = M x.
42. (a) Take Φ to be B(BS)B. Then
B(BS)Bxyzw = BS(Bx)yzw = S(Bxy)zw
= Bxyw(zw) = x(yw)(zw) = Φxyzw.
(b) We take Φ1 = Φ. Now, suppose Φn is defined and behaves as shown
in the equation Φn xyzw1 . . . wn = x(yw1 . . . wn )(zw1 . . . wn ). We
take Φn+1 to be BΦn Φ. Then
BΦn Φxyzw1 . . . wn wn+1 = Φn (Φx)yzw1 . . . wn wn+1
= (Φx)(yw1 . . . wn )(zw1 . . . wn )wn+1
= x(yw1 . . . wn+1 )(zw1 . . . wn+1 )
= Φn+1 xyzw1 . . . wn+1 .
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 10

Sages, Oracles and Doublets

Sages Again

We recall that by a sage is meant a combinator θ such that θx is a fixed

point of x for every x; in other words, x(θx) = θx.
Several papers in the literature are devoted to the construction of sages.
In the chapter before last, we did not construct any sages. We merely
proved that a sage exists, providing that the mocker M exists and that the
composition condition holds. We now turn to several ways of constructing
sages.

Problem 1. Construct a sage from B, M and the robin R, recalling that

Rxyz = yzx. [Hint: Recall from the solution to Problem 1 in Chapter 9
that M (BxM ) is a fixed point of x.]

Problem 2. Now derive a sage from B, M and the cardinal C.

Problem 3. A simpler construction of a sage uses M, B and the lark L.

One such construction provides simple alternative solutions to the last two
problems. Can you find it?

Problem 4. Derive a sage from B, M and W .

It is a bit tougher to derive a sage from B, W and C. We will consider
some ways of doing so.

Problem 5. Let us first derive a sage from B, L and the queer bird Q
[Qxyz = y(xz)].
206 Part III: Elements of Combinatory Logic

Problem 6. Using the result of the last problem, now write down an expres-
sion for a sage in terms of B, W and C.

Problem 7. A particularly neat and simple construction of a sage is from

Q, M and L. Can you find it?

Problem 8. Actually, a sage can be constructed from Q and M alone.

How?
We recall the starling S satisfying the condition Sxyz = xz(yz).

Problem 9. Construct a sage from L and S.

Problem 10. Now show that a sage can be constructed from B, W

and S. This can be done in several ways, one of which uses only five let-
ters. (The sage presented in the set of solutions with 5 letters is due to
Curry.)

Problem 11. Construct a sage from M and P . [P xyz = z(xyz)].

Problem 12. Construct a sage from P and W .

Problem 13. Construct a sage from P and I.

The Turing Combinator

In 1937, the great Alan Turing discovered a remarkable combinator U ,

which will soon surprise you! It is defined by the condition:
U xy = y(xxy).
It is ultimately derivable from B, T and M in several ways (it is also deriv-
able from S, L and I).

Problem 14. Derive U from W1 and P1 .

Problem 15. Derive U from W and P .

Problem 16. Derive U from P and M .

Problem 17. Derive U from S, I and L.

Now for the surprise!

Problem 18. There is a sage that is derivable from U alone! How?

 Chapter 10. Sages, Oracles and Doublets 207

The Oracle O

A truly remarkable combinator O is defined by the condition:

Oxy = y(xy).
You will soon see why O is so remarkable! In To Mock a Mockingbird,
I dubbed it the owl, but I now prefer the grander name oracle, which is
befitting it.

Problem 19. Derive O from Q and W .

Problem 20. Derive O from B, C and W .

Problem 21. Derive O from S and I.

Problem 22. Derive O from P and I.

Problem 23. Derive a sage from O, B and M .

Problem 24. Derive a sage from O and L.

Problem 25. Derive a Turing combinator U from O, B and M .

Problem 26. Derive a Turing combinator from O and L.

Now, why is the oracle O so remarkable? Well, in the solutions to
Problems 23 and 24, we have seen two sages constructed from O and other
combinators, and in both cases the sages happened to be fixed points of O.
Was that a mere coincidence? No! The first remarkable fact about the oracle
O is that all of its fixed points are sages.

Problem 27. Prove that all fixed points of O are sages.

Now for the second remarkable fact about O. The converse of the state-
ment of Problem 27 holds, i.e., all sages are fixed points of O. Thus the
class of all fixed points of O is the same as the class of all sages!

Problem 28. Prove that every sage is a fixed point of O.

Double Sages

A pair (y1 , y2 ) is called a double fixed point of a pair (x1 , x2 ) if x1 y1 y2 = y1

and x2 y1 y2 = y2 . We say the system has the weak double fixed point property
208 Part III: Elements of Combinatory Logic

if every pair (x1 , x2 ) has a double fixed point. We say that the system had
the strong double fixed point property if there is a pair (θ1 , θ2 ), called double
sage pair — or, more briefly, a double sage — such that for every x and y,
the pair (θ1 xy, θ2 xy) is a double fixed point of (x, y), i.e.
x(θ1 xy)(θ2 xy) = θ1 xy
and y(θ1 xy)(θ2 xy) = θ2 xy.
There are many ways to construct double sages from the combinators B, M
and T , several of which can be found in Chapter 17 of my book Diagonal-
ization and Self-Reference [Smullyan, 1994], and one of which we shall now
consider. Others, which I will leave as exercises, have solutions which can
be found in the book just mentioned.

The Nice Combinator N

My favorite construction of a double sage uses a combinator N satisfying

the following condition:
N zxy = z(N xxy)(N yxy).
The existence of such an N obviously implies that the system has the
weak double fixed point property.

Problem 29. Why?

Before considering the construction of a double sage, let us see how a
nice combinator N can be ultimately derived from B, T and M . To this
end, we will use a combinator n defined by the condition:
nwzxy = z(wxxy)(wyxy).

Problem 30.
(a) Derive n from B, C and W or from combinators derivable from them.
[It is easiest to derive n directly from C, W, W ∗ , V ∗ and Φ3 .]
(b) Now show that any fixed point of n (such as M (Ln)) is a nice
combinator.

Problem 31. Show that from N, C, W and W1 one can construct combi-
nators θ1 and θ2 such that (θ1 , θ2 ) is a double sage.
 Chapter 10. Sages, Oracles and Doublets 209

Double Oracles

We will call a pair (A1 , A2 ) a double oracle if every double fixed point of
(A1 , A2 ) is a double sage. It turns out that every double sage is a double
fixed point of any double oracle (A1 , A2 ).
To my delight, double oracles exist, if B, T and M are present.
Let O1 and O2 be defined by the following conditions:
O1 zwxy = x(zxy)(wxy),
O2 zwxy = y(zxy)(wxy).

Problem 32. Show that (O1 , O2 ) is a double oracle.

Solutions to the Problems of Chapter 10

1. Since BxM (BxM ) = M (BxM ) is a fixed point of x (by the solution

to Problem 1 of Chapter 9), we want to find a combinator θ such that
θx = M (BxM ). Well, by the definition of R, BxM = RM Bx, and so
M (BxM ) = M (RM Bx). Also, by the definition of B,
M (RM Bx) = BM (RM B)x.
We thus take θ to be BM (RM B).
2. Let us now use a fact stated at the end of the last chapter, namely
that for any combinators A1 and A2 , the combinator RA1 A2 is
equal to CA2 A1 . In particular, RM B is equal to CBM , so that
RM Bx = CBM x. Then, since M (RM Bx) is a fixed point of x, so
is M (CBM x). Also, M (CBM x) = BM (CBM )x. We thus now take θ
to be BM (CBM ).
3. We now use the fact that Lx(Lx) is a fixed point of x. Thus so is
M (Lx), which is also BM Lx. Thus BM L is a sage.
We proved in the last chapter (in the solution to Problem 35) that
RM B is a lark, so that CBM is also a lark, since it is equal to RM B.
We can replace the L in BM L with either RM B or CBM , showing
that both BM (RM B) and BM (CBM ) are sages.
210 Part III: Elements of Combinatory Logic

This shows us alternative ways of solving Problems 1 and 2.

4. As we saw in the solution of Problem 35 of the last chapter, BW B is
also a lark. Thus we can also replace the L In BM L by BW B, showing
us that BM (BW B) is also a sage.
5. We again use the fact that Lx(Lx) is a fixed point of x. Now,

W (QL(QL))x = QL(QL)xx = QL(Lx)x = Lx(Lx).

Thus W (QL(QL)) is a sage.

6. We have just proved that W (QL(QL)) is a sage. We can take CB
for Q (by the solution to Problem 19 in Chapter 9), thus getting the
expression W (CBL(CBL)), which can be shortened to W (B(CBL)L).
Then, by the solution of Problem 35 of Chapter 9, we can take BW B
for L, thus getting the sage

W (B(CB(BW B))(BW B)).

We will later find another solution to this problem.
7. Again we use the fact that Lx(Lx), and hence M (Lx), is a fixed point
of x. But M (Lx) = QLM x. Thus QLM is a sage.
8. In the expression QLM of the previous problem, we can take QM for L
(by Problem 35 of Chapter 9), thus obtaining Q(QM )M as another
sage.
9. Lx(Lx) is a fixed point of x. But SLLx = Lx(Lx). Hence SLL is a
sage!
10. Here is one way to get a sage from S, W and B: Since SLL is a sage
(as shown in the solution to the last problem), we can take BW B
for L in SLL, obtaining S(BW B(BW B)), which can be shortened to
S(W (B(BW B))), which has six letters. However, there is a more clever
and more economical way: By the definition of W , SLL = W SL. Now
replace L by BW B in W SL, and we have the sage W S(BW B), one
letter shorter! This is Curry’s expression for a sage.
Note that since S is derivable from B, C and W (by Problem 36 of
Chapter 9), so is W S(BW B), thus getting a third way of deriving a
sage from B, C and W .
11. M (P M ) is a sage: M (P M )x = P M (P M )x = x(M (P M )x).
12. W P (W P ) is a sage: W P (W P )x = P (W P )(W P )x = x((W P )(W P )x).
 Chapter 10. Sages, Oracles and Doublets 211

13. P II(P (P II)) is a sage, since M (P M ) is a sage by the solution to

Problem 11 above, and M can be expressed by P II, as seen in the
solution to Problem 41 of Chapter 9.
14. W1 P1 is a Turing combinator: W1 P1 xy = P1 xyy = y(xxy) = U xy.
15. Another Turing combinator is W P , since
W P xy = P xxy = y(xxy) = U xy.
16. Another Turing combinator is P M , since
P M xy = y(M xy) = y(xxy) = U xy.
17. Another Turing combinator is L(SI), since
L(SI)xy = SI(xx)y = Iy(xxy) = y(xxy) = U xy.
18. In the equation U xy = y(xxy), just substitute U for x, and we have
U U y = y(U U y). Thus U U is a sage.
19. QQW is an oracle, since QQW xy = W (Qx)y = Qxyy = y(xy) = Oxy.
20. We take CB for Q in QQW from the previous problem, which shows
us that CB(CB)W is an oracle.
21. SI is an oracle, since SIxy = Iy(xy) = y(xy) = Oxy.
22. P I is an oracle, since P Ixy = y(Ixy) = y(xy) = Oxy.
23. See the solution to Problem 25 below.
24. See the solution to Problem 26 below.
25. BOM is a Turing combinator, since
BOM xy = O(M x)y = y(M xy) = y(xxy) = U xy.
Now, since BOM is a Turing combinator, and if U is any Turning
combinator, then U U is a sage (by the solution to Problem 18), it
follows that BOM (BOM ) is a sage (which can also be verified directly),
which solves Problem 23.
Please note that BOM (BOM ) is also a fixed point of O. Indeed (as
noted in the solution to Problem 1 of Chapter 9 and recalled in the
solution to Problem 1 of this chapter), BxM (BxM ) is a fixed point of
x, for any x.
26. LO is a Turing combinator, since LOxy = O(xx)y = y(xxy) = U xy.
Now, since LO is a Turing combinator, then LO(LO) is a sage, which
solves Problem 24. [Now we can say, “Lo! Lo! A sage!!”] Again LO(LO)
is not only a sage, but also a fixed point of O!
212 Part III: Elements of Combinatory Logic

27. Suppose A is a fixed point of O. Then OA = A. Hence

Ax = OAx = x(Ax).
Since Ax = x(Ax), A is a sage.
28. Consider any sage θ. For any x, we have θx = x(θx). Also Oθx = x(θx).
Thus θx = O(θx) (they are both equal to x(θx)). Since θx = Oθx for
every term x, it follows that θ = Oθ, which means that θ is a fixed
point of O!
29. In the equation N zxy = z(N xxy)(N yxy) substitute x for z to obtain:
(1) N xxy = x(N xxy)(N yxy).
Next, in the equation N zxy = z(N xxy)(N yxy) substitute y for z to
obtain:
(2) N yxy = y(N xxy)(N yxy).
By (1) and (2), the pair (N xxy, N yxy) is a double fixed point of (x, y).
30. (a) We take n to be C(V ∗ Φ3 W (W2 C)). We leave it to the reader to
verify that nwzxy = z(wxxy)(wyxy).
(b) Let N be any fixed point of n. Then N = nN . Thus
N zxy = nN zxy = z(N xxy)(N yxy).
31. Take θ1 to be W N and θ2 to be W1 (CN ) Then
(a) θ1 xy = W N xy = N xxy, so that θ1 xy = N xxy.
(b) θ2 xy = W1 (CN )xy = CN xyy = N yxy, so that θ2 xy = N yxy.
Now,
(1) N xxy = x(N xxy)N (yxy) and
(2) N yxy = y(N xxy)N (yxy).
Since N xxy = θ1 xy and N yxy = θ2 xy, then, by (1) and (2), we have:
(10 ) θ1 xy = x(θ1 xy)(θ2 xy) and
(20 ) θ2 xy = y(θ1 xy)(θ2 xy).
Thus (θ1 , θ2 ) is a double sage.
32. Suppose (A1 , A2 ) is a double fixed point of (O1 , O2 ). Then
O1 A1 A2 = A1
and O2 A1 A2 = A2 . Hence:
A1 xy = O1 A1 A2 xy = x(A1 xy)(A2 xy),
A2 xy = O2 A1 A2 xy = y(A1 xy)(A2 xy).
 Chapter 10. Sages, Oracles and Doublets 213

Thus (A1 , A2 ) is a double sage, and (O1 , O2 ) is a double oracle, as we

were asked to show.
Before proving that every double sage is a double fixed point of
(O1 , O2 ), let us note that if A1 xy = A2 xy for all x and y, then A1 = A2 ,
because from A1 xy = A2 xy for all x and y it follows that A1 x = A2 x
for all x, and hence that A1 = A2 .
Now, suppose that (θ1 , θ2 ) is a double sage. Thus:
(1) θ1 xy = x(θ1 xy)(θ2 xy),
(2) θ2 xy = y(θ1 xy)(θ2 xy).
Also,
(10 ) O1 θ1 θ2 xy = x(θ1 xy)(θ2 xy),
(20 ) O2 θ1 θ2 xy = y(θ1 xy)(θ2 xy).
Therefore, θ1 xy = O1 θ1 θ2 xy and θ2 xy = O2 θ1 θ2 xy.
Then θ1 = O1 θ1 θ2 and θ2 = O2 θ1 θ2 , which means that (θ1 , θ2 ) is a
double fixed point of (O1 , O2 ).
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 11

Complete and Partial Systems

I. The Complete System

From the two combinators S and K, all possible combinators are derivable!
Let me explain just what I mean by this.
In the formal language of combinatory logic, we have a denumerable
list x1 , x2 , . . . , xn , . . . of symbols called variables, and some symbols called
constants, each of which is the name of an element of the applicative system.
The notion of term is recursively defined by the following conditions:
(1) Every variable and constant is a term.
(2) If t1 and t2 are terms, so is (t1 t2 ).
It is understood that no expression is a term unless its being so is a
consequence of conditions (1) and (2) above.

Problem 1. Give a more explicit definition of what it means for an

expression to be a term.
We continue to abbreviate terms by removing parentheses if no ambi-
guity can result, using the afore-mentioned convention that parentheses are
to be restored to the left, e.g. xyz is to be read ((xy)z) etc.
For a moment, let us recall in general all the combinators you have
been introduced to in Chapters 8 through 10. You learned each combina-
tor’s meaning from what we have called its “defining equation” (its defin-
ing condition). Most of those defining equations for combinators are of the
form Ax1 x2 . . . xn = t(x1 x2 , . . . , xn ), where x1 x2 , . . . , xn are the variables
that occur in the term t and the term t is built only from these variables
216 Part III: Elements of Combinatory Logic

(and parentheses; we will now [mostly] stop mentioning those necessary

parentheses, which are always there whenever a term is not a single vari-
able or a single constant, even if we suppress writing some of them for
easier reading). For instance, you were introduced to the robin by its defin-
ing equation Rxyz = yzx = ((yz)x) and to the cardinal by its defining
equation: Cxyz = xzy = ((xz)y). However, the defining equation of the
kestrel, Kxy = x is different in that not all the variables that we wish to
have as arguments of the combinator K occur in the term t (which is x in
this case), although all the variables in t occur in the list of arguments for
K, the combinator being defined.
Combinators A of the first type, in which the variables of the term and
the combinator being defined are identical, are called non-cancellative com-
binators (or λ-I combinators). We will discuss this subgroup of combinators
further in Section II. Combinators A of the second type, in which not all
the variables in the arguments of the combinator being defined occur in the
term t are called cancellative combinators, because A effectively cancels out
those of the variables that do not occur in t. For example, the bluebird has
the defining equation Bxyz = x(yz) in which the variables are identical on
both sides of the equation, so it is non-cancellative. But in Kxy = x and
(KI)xy = y, one of the two arguments on the left is missing in the term
on the right, so the two combinators K and KI are cancellative (these are
the only cancellative operators you have seen so far).
Thus the general form of a defining equation for a combinator is
Ax1 x2 . . . xn = t, where t is constructed only from variables and the vari-
ables of t are among x1 , x2 , . . . , xn . I.e. although t must contain at least
one variable in order to be a term, and every variable in t is one of the
variables x1 , x2 , . . . , xn , not all the variables that occur in x1 , x2 , . . . , xn
need to occur in t (e.g. as in the case of the kestrel).
When t is a term constructed only from variables that are among
x1 , x2 , . . . , xn , and our applicative system C contains an element A (no ele-
ment of C contains variables) such that Ax1 x2 . . . xn = t holds, then we
say that A is a combinator of the applicative system C and also say that the
combinator A with arguments x1 , x2 , . . ., xn realizes the term t. The equation
Ax1 x2 . . . xn = t is called the defining equation of the combinator A.
It turns out that all constants in any applicative system are combinators,
because in applicative systems the meaning of constants is always specified
by a defining equation. And the systems are also always defined so that
every term constructed from constants alone is assumed to be a combinator
 Chapter 11. Complete and Partial Systems 217

of the system. Thus the combinators and elements of the system are one
and the same. But the constants may be a subset of the set of elements.
So now you know what I mean by “every possible combinator” when
I said that every possible combinator is derivable from S and K. I was
talking about any combinator that can be specified by a defining equation
of the above form!
Thus the claim that every combinator in our applicative system C can
be derived from (constants) S and K just says that for every combinator
A specified by a defining equation Ax1 x2 . . . xn = t (where t is built only
from variables, and those variables are among x1 , x2 , . . . , xn ), we can find
a term AS,K constructed only from S and K (and no variables at all!) such
that we can replace the A in its defining equation by AS,K and get a true
equation (true because of the meanings of S and K): AS,K x1 x2 . . . xn = t.
Consequently, if we assume our applicative system C includes the ele-
ments (combinators) S and K, and we can prove our claim, then every
defining equation you have seen that has defined one of the combinators
you were introduced to, specifies a combinator that can be derived only
from S and K, so that the only constants we need in our system to obtain
all the combinators you have seen (and infinitely many more) are S and K.
But why should a specification of the ordered arguments of a combina-
tor and a term built only from variables included in those arguments be
enough to specify any combinator? Well, first of all, as you have just seen,
such a specification includes an infinite number of combinators, including
all you have studied, as well as all those that can be defined from any term
built from variables, and an ordered list of argument variables that includes
the variables of the term. The claim that these are “all the combinators”
essentially says that the essence of combinators is hidden in those terms
built from variables and the argument list for the combinator. Here’s one
way to think about that: Since applicative systems are to be imagined as
being concerned with the order of “applications” of some sort of things to
other things of the same sort (say, applications involving functions, algo-
rithms, processes, etc.; you will see some quite varied interpretations of
applications in the next chapter), the structure of parentheses in complex
terms involving variables can be seen as defining an order on the applica-
tions of whatever is filled in for the variables at some point in time. For
instance, consider the order of applications defined by the parentheses of
(((rs)((tu)v))((wx)y)). [It certainly has nothing to do with the alphabetical
order of the names of the variables.] In any case, you can see that, if we
218 Part III: Elements of Combinatory Logic

can prove what we have claimed, then in any applicative system including
at least S and K as constants, there is a combinator A for which

Arstuvwxy = (((rs)((tu)v))((wx)y)),

as well as a (cancellative) combinator A0 for which

A0 rstuvwxy = ((wu)r)(((vx)u)s).

Combinatorial Completeness

Now we wish to prove our principal claim, namely that given any ordered list
of variables to be taken as arguments for a combinator, and any term t built
only from variables included in the list of arguments, there is a combinator
with the given arguments (in the order specified in the list) that is derivable
from just S and K (no variables) which realizes the term t.
To begin with, the identity combinator I is derivable from S and K:
just take I to be SKK, for SKKx = Kx(Kx) = x = Ix. Thus it suffices
to show that every combinator is derivable from S, K and I. This is what
we will now do.
Let us first mention the following induction principle, which we will call
the term induction principle: Consider a set S whose members are variables
or constants or both, and let S + be the set of all terms built from elements
of S (and parentheses). To show that all terms in S + have a certain property
P , it suffices to show that all elements of S have the property, and that for
any terms t1 and t2 of S + , if t1 and t2 have property P , then so does (t1 t2 ).
This principle is as self-evident as mathematical induction, and can in fact
be derived from mathematical induction on the number n of occurrences of
elements of S in the term.
Let us call W a simple term if it is built only from variables and S, K,
and I. In the algorithm we are about to present, we will assume we will
be starting with a simple term W built only from variables, but we will
successively turn it into a simple term built from S, K, and I and variables.
In the end, we will have built a term T from W that is built only of S,
K, and I (by having eliminated all the variables in a specified ordered list
of arguments for a combinator T , a list that includes all the variables in
W). And if x1 , . . . , xn is the list of variables specified as arguments for
the combinator, it will be the case that T x1 . . . xn = W, where the W
 Chapter 11. Complete and Partial Systems 219

in question is the original one containing only variables (all of which are
included among the x1 , . . . , xn ).
Now, consider any simple term W and consider a variable x, which may
or may not be in W. We shall call a term Wx an x-eliminate of W if the
following three conditions hold:
(a) The variable x does not occur in Wx .
(b) All variables in Wx are in W.
(c) The equation Wx x = W holds.
We wish to first show that every simple term W has an x-eliminate Wx
that is derivable from S, K and I and variables in W excluding the variable
x. We define Wx by the following inductive rules:
(1) If W is x itself, we take Wx to be I.
(2) If W is any term in which x does not occur, we take Wx to be KW.
(3) If W is a compound term WV in which x occurs, we take (WV)x to be
SWx Vx .
Note that for any possible simple term, one and only one of the three rules
will apply. When rule (3) is applied, it will require finding the x-eliminates
of the two parts of the compound term.
When we called the scheme of rules for x-elimination we just introduced
inductive, we meant that if we have a simple term from which we wish to
eliminate the variable x we operate repeatedly on the components of our
term working from outside inwards looking for x-eliminates of the compo-
nents of the term, until we have eliminated x from all the subterms and
finally the whole term.
Let us consider an example. Let’s see if we can derive from S, K and I
a combinator A of one argument x with the defining equation Ax = (xx).
(Thus we are checking whether the mocker/mockingbird is derivable from
S, K and I.) We want to see if eliminating x, the only variable in the term
(and the only variable in the list of required arguments for the combina-
tor) will give us a term in S, K and I. Well, (xx) is a compound term in
which x occurs, so rule (3) tell us that the x-eliminate of (xx) consists of
the starling S placed in front of the x-eliminates of the two parts of the
compound term, which are both x. By rule (1), the x-eliminate of x is I.
So the x-eliminate of (xx) is SII. Let us check that the three conditions
which an x-eliminate must satisfy actually hold here: (a) The variable x
does not occur in SII; (b) All variables in SII are in (xx) [this holds
vacuously, since there are no variables in SII]; (c) SIIx = Ix(Ix) = xx.
220 Part III: Elements of Combinatory Logic

(Of course xx is just (xx) with its outer parentheses suppressed for easier
reading.)
Now let us prove that for any simple term, the method of x-eliminates
we have described does result in a term that satisfies the conditions on an
x-eliminate.

Proposition 1. Wx is an x-eliminate of W. In other words:

(1) The variable x does not occur in Wx .
(2) All variables of Wx occur in W.
(3) The equation Wx x = W holds.
This proposition is proved by term induction.

Problem 2. (i) Show that if W is x, then Wx is an x-eliminate of W.

(ii) Show that if W is any term in which x does not occur, then Wx is
an x-eliminate of W.
(iii) Show that if Wx and Vx are x-eliminates of W and V respectively,
so is (WV)x (i.e. so is SWx Vx ).
It then follows by term induction that the term Wx is an x-eliminate of
W for every simple term W.

Problem 3. Since I is an x-eliminate of x, it follows that an x-eliminate of

Wx is SWx I. However, if x does not occur in W, there is a much simpler
x-eliminate of Wx, namely W itself! Prove this.
Now the reader might already realize that if the term t from which
we wish to “eliminate” variables contains a number of variables, we must
successively “eliminate” each of the variables that occur in the term, if we
wish to arrive at a variable-free term (i.e. the desired combinator). Indeed,
we must also “eliminate” all the variables that do not occur in our original
term, but which occur in the ordered list of argument variables for the
combinator we are seeking. Moreover, we must do our variable eliminations
in a particular order. To that process we now turn.
For two variables x and y, by Wx,y is meant (Wx )y .
We shall call a term T an x, y eliminate of W if the following three
conditions hold:
(1) Neither of the variables x, y occurs in T.
(2) All variables of T occur in W.
(3) The equation Txy = W holds.
 Chapter 11. Complete and Partial Systems 221

Problem 4. Which, if either, of the following statements is true?

(1) Wx,y is an x, y eliminate of W.
(2) Wx,y is an y, x eliminate of W.
We now take Wx1 x2 ... xn as an abbreviation of (. . . (Wx1 )x2 ) . . .)xn ).
We define a term T to be an x1 , x2 , . . . , xn eliminate of W if the following
three conditions hold:
(1) None of the variables x1 , x2 , . . . , xn occur in T.
(2) All the variables in T occur in W.
(3) The equation Tx1 x2 . . . xn = W holds.
From the result of Problem 4, and mathematical induction, we have:

Proposition 2. Wxn xn−1 ... x1 is an x1 , x2 , . . . , xn eliminate of W.

Now let W be a term built only from variables among x1 , x2 , . . . , xn .
Suppose we want a combinator A with arguments x1 x2 . . . xn derivable
from only S, K and I (using no variables) and satisfying the condi-
tion Ax1 . . . xn = W. Well, we take A to be Wxn xn−1 ... x1 , which is an
x1 , x2 , . . . , xn eliminate of W.
All variables of A occur in W, hence are among x1 , x2 , . . . , xn , yet none
of the variables x1 , x2 , . . . , xn are in A, which means that A contains no
variables at all. Since A has been formed by a variable-elimination scheme
which introduces only the combinators S, K, and I, A must be built only
from S, K, and I. And the condition Ax1 x2 . . . xn = W holds.
Now we define a class of combinators to be combinatorially com-
plete when, for every simple term t and every list of possible argu-
ments x1 , x2 , . . . , xn for a combinator, if every variable in the term t
occurs in x1 , x2 , . . . , xn , there is a combinator A in the class such that
Ax1 x2 . . . xn = t (i.e. when, for any such term t, there is a combinator A
with arguments x1 , x2 , . . . , xn that realizes t).
Thus we have proved:

Theorem. The class of combinators derivable from S and K is combina-

torially complete.
We would like to point out that, given a term t in S, K and I (that is,
including terms in which variables are liberally strewn throughout) and an
ordered list of variables x1 , x2 , . . . , xn including all the the variables in t,
the method of eliminates described here can also be used, if one wishes, to
222 Part III: Elements of Combinatory Logic

transform any such term t into an expression Wx1 x2 . . . xn in which W is

a combinator built only with S, K and I. (To see this one only has to look
once more through the method and the proof that the method works to see
that it applies as well in this situation.)
Now let us consider an example of how to find a cancellative combinator
with more than one variable using the method of eliminates: Suppose we
want a combinator A satisfying the condition Axy = y. Thus A has the
two arguments x, y and the term we want to use to define A does not
include x. Well, we want A to be a y, x eliminate of y, i.e. an x-eliminate
of a y-eliminate of y. Thus we must first find a y-eliminate of y, and that is
simply I. We then want an x-eliminate of I, which is KI. Thus our solution
should be KI. Let us check: KIx = I, so that KIxy = Iy = y.
Now for a permuting combinator: How can we derive the thrush T
from S, K and I? [T xy = yx]. We have to find an x-eliminate of a y-
eliminate of yx. Well, a y-eliminate of y is I, and a y-eliminate of x is
Kx, and so a y-eliminate of yx is SI(Kx). We now need an x-eliminate
of SI(Kx) = (SI)(Kx). Well, an x-eliminate of SI is K(SI), and an x-
eliminate of Kx is simply K (by Problem 3), and so an x-eliminate of
SI(Kx) is S(K(SI))K. Thus S(K(SI))K should be our solution. Let us
check:
S(K(SI))Kx = K(SI)x(Kx) = SI(Kx).
Then S(K(SI))Kxy = SI(Kx)y = Iy(Kxy) = y(Kxy) = yx.
It would be a good exercise for the reader to try to derive various
combinators for M, W, L, B, C from S, K and I, using the procedure of
eliminates. The procedure can be easily programed for a computer. How-
ever, it should be pointed out that the procedure, surefire as it is, can be
very tedious and often leads to much longer expressions than can be found
by using some cleverness and ingenuity.
There is an interesting table on Internet called “Combinator Birds” that
gives an expression in S and K alone (no I) for all the combinators you have
encountered in this book and more. This table can currently be found at
angelfire.com/tx4/cus/combinator/birds.html. For instance, there you will
see how exceedingly long are the derivations from S and K alone of the finch
and the hummingbird. In the first column of the table it is immediately after
the Greek letter λ (and before the period) that the creators of the table
list the arguments of the combinator in question, since those arguments
cannot be determined simply from the constant-free term used to define the
 Chapter 11. Complete and Partial Systems 223

combinator. For the goldfinch, the first column entry in the goldfinch row,
λabcd·ad(bc), tells us that the combinator referred to in this row of the table
has the arguments a, b, c and d, and the defining equation is Gabcd = ad(bc).
The next two column entries for the row tells us that the common symbol for
this combinator is G and that it is commonly called the goldfinch. The next
column entry on the row tells us that this combinator can be derived from
B, C and W as BBC. And then the final column entry for the goldfinch
row tells us the goldfinch can be derived from S and K alone by the term
((S(K((S(KS))K)))((S((S(K((S(KS))K)))S))(KK)).

A Fixed Point Problem

There are two results of combinatory logic known as the First and Second
Fixed Point Theorems. The second one will be dealt with in the next chap-
ter. The first will be considered now. To motivate this, the reader should
try the following exercise:

Exercise. (a) Find a combinator Γ that satisfies the condition

Γxy = x(Γy). (b) Find a combinator Γ that satisfies the condition
Γxy = Γyx.
The solutions of (a) and (b) are special cases of the First Fixed Point
Theorem.
In what follows t(x, x1 , . . . , xn ) is a term in which x, x1 , . . . , xn are
precisely the variables that occur in the term. Recall that all realizable
terms are built up only from variables (and parentheses).

Theorem F 1 . [First Fixed Point Theorem] For any realizable term

t(x, x1 , . . . , xn ) there is a combinator Γ satisfying the condition
Γx1 . . . xn = t(Γ, x1 , . . . , xn ).

Remarks. This theorem may be a bit startling, since the satisfying condi-
tion for Γ involves Γ itself! This is related to self-reference or recursion, as
will be seen in the next chapter.

Problem 5.
(1) Prove Theorem F1 . [There are two different proofs of this. One involves
a fixed point of a combinator that realizes the term t(x, x1 , . . . , xn ). The
other uses a combinator that realizes the term t((xx), x1 , . . . , xn ).]
224 Part III: Elements of Combinatory Logic

(2) Now find solutions to (a) and (b) of the exercise stated before
Theorem F1 .

Doubling Up

Theorem F1 has the following double analogue:

Theorem F 1 F 1 . [First Double Fixed Point Theorem] For any two real-
izable terms t1 (x, x1 , . . . , xn ) and t2 (x, x1 , . . . , xn ) there are combinators
Γ1 , Γ2 such that:
(1) Γ1 x1 . . . xn = t1 (Γ2 , x1 , . . . , xn ).
(2) Γ2 x1 . . . xn = t2 (Γ1 , x1 , . . . , xn ).
Problem 6. Prove Theorem F1 F1 .

II. Partial Systems of Combinatory Logic

λ-I Combinators

We will now consider combinators A defined by some condition

Ax1 . . . xn = t, where t is a term all of whose variables are among the vari-
ables x1 . . . xn . We recall that if all of the variables x1 . . . xn occur in t, then
A is called a λ-I combinator, or a non-cancellative combinator, while all
other combinators (i.e. those for which one or more of the variables x1 . . . xn
do not occur in t) are called cancellative combinators (e.g. K and KI), since
they cancel out those of the variables x1 . . . xn that do not occur in t.
For any class C of combinators, a set S of elements of C is called a
basis for C if all elements of C are derivable from elements of S. We have
already shown that the class of all combinators has a finite basis, namely
the two elements S and K. We now wish to show that the four combinators
S, B, C and I form a basis for the class of all λ-I combinators (and hence
that B, M, T and I also form a basis, since S, B, C and I are themselves
derivable from B, M, T and I).
To this end, let us define a nice term to be a term built from the symbols
S, B, C and I and variables. If a nice term has no variables, then it is a
combinator, and will accordingly be called a nice combinator. Thus a nice
combinator is a combinator built only from S, B, C and I.
 Chapter 11. Complete and Partial Systems 225

We define an x-eliminate of a term as before and we must now show

that if W is any nice term, then W has a nice x-eliminate, provided that
x actually occurs in W! Well, for this purpose we must redefine our set of
recursive rules to produce the appropriate x-eliminate Wx .
1. If W = x, we take Wx to be I. [Thus xx = I.]
2. For a compound term WV in which x occurs, it occurs either in W or
in V or in both.
(a) If x occurs both in W and V, we take (WV)x to be SWx Vx .
(b) If x occurs in W but not in V, we take (WV)x to be CWx V.
(c) If x occurs in V but not in W, then:
(c1 ) if V consists of x alone, we take (WV)x to be W. [Thus
(Wx)x = W, if x does not occur in W.]
(c2 ) if V 6= x (but x occurs in V and not in W), we take (WV)x to
be BWVx .
Note that this set of rules, unlike the previous set of rules for x-
elimination, contains no rule for eliminating x when the original term you
wish to eliminate x from is a term in which x doesn’t occur. But this
causes no problem in our current non-cancellative combinator situation. If
our original term contains no occurrence of x, we would have no reason to
even start off on an x-elimination, because we only eliminate variables from
the list of arguments for the desired combinator, and if the original term
contains no x, there would be no x in the argument list, so we wouldn’t
try to do x-elimination at all. So we are only concerned with doing an x-
elimination either on an x standing alone, which is covered by Rule (1), or
on a compound term WV in which x occurs in either W or V or both. If we
do find after eliminating x from such a compound term (and we are told
by the rules as to what to do when x doesn’t occur in one of the terms)
that there are no longer any x’s in the compound term that result, we just
know that we have the final x-eliminate for the term (or subterm) we are
working on.

Proposition 3. For any nice term W in which x occurs, the term Wx is

a nice x-eliminate of W.

Problem 7. Prove Proposition 3.

We leave it to the reader to show that for a nice term W, if vari-
ables x and y occur in W, then (Wx )y is a nice y, x eliminate of W,
and more generally that if x1 , . . . , xn occur in W, then Wxn xn−1 ... x1 is
226 Part III: Elements of Combinatory Logic

a nice x1 , . . . , xn eliminate of W, and so if x1 , . . . , xn are all the variables

of W, then Wxn xn−1 ... x1 is a nice combinator A satisfying the condition
Ax1 . . . xn = W. Since A is nice, then it is also a λ-I combinator (since
S, B, C and I are all λ-I combinators, and for any two λ-I combinators
A1 A2 is again a λ-I combinator).
Notice that this time we are proving Proposition 3 for every term built
up from S, B, C, I and variables (we said any nice term). I pointed out
earlier we also could have done this for terms built from S and K and vari-
ables with the earlier method of eliminates and the same proofs. But since
terms built up only from variables are also terms built up from S, B, C, I
and possibly variables, we have the following immediate consequence of
Proposition 3:

Corollary. For every constant-free term t and every list of possible argu-
ments x1 , x2 , . . . , xn for a combinator, if the variables occurring in the term
t are x1 , x2 , . . . , xn too, there is a combinator A in the class of variable-free
terms built from S, B, C and I such that Ax1 x2 . . . xn = t (i.e. there is a
non-cancellative combinator A that realizes the term t).
This proves that S, B, C and I form a basis for the class of all λ-I
combinators.
It is known that no proper subset of {S, B, C, I} is a basis for the class
of λ-I combinators, but J. B. Rosser found a curious two-element basis for
the class of λ-I combinators, namely I and a combinator J defined by the
condition:
Jxyzw = xy(xwz).
How Rosser found that is a mystery to me! If the reader is inter-
ested, a derivation of B, T and M from J and I can be found in
To Mock a Mockingbird, and probably in other books on combinatory
logic.

B, T, I Combinators

The class of all combinators derivable from B, T and I has been studied by
Rosser [1936] in connection with certain logical systems in which duplica-
tive combinators like M, W, S, J have no place. In To Mock a Mockingbird
I showed that these three combinators can be replaced by only two, namely
 Chapter 11. Complete and Partial Systems 227

I and the combinator G (the goldfinch), which we recall is defined by the

condition:
Gxyzw = xw(yz).
Whether this discovery was new or not, I honestly don’t know. Now let us
derive B and T from G and I.

Problem 8.
(a) First derive from G and I the combinator Q3 (which, we recall, is
defined as satisfying the condition Q3 xyz = z(xy)).
(b) Then derive the cardinal C from G and I. [Cxyz = xzy.]
(c) From C and I, we can obtain T . [Alternatively, we can get T from Q3
and I].
(d) From C we can get the robin R [Rxyz = yzx], and then from C, R and
Q3 , we can get the queer combinator Q [Qxyz = y(xz)], and then from
Q and C we can obtain B.

Solutions to the Problems of Chapter 11

1. t is a term if and only if there is a finite sequence t1 , . . . , tn = t such that

for each i ≤ n, either ti is a variable or constant, or there are numbers
j1 and j2 both less than i such that ti = (tj1 , tj2 ).
2. We will show that no matter which rule applies to our simple term W,
then Wx will be an x-eliminate of W.
(i) Let us first consider the case that W is x itself. Then, by rule (1),
Wx = I. We must thus show that I is an x-eliminate of W. Well, since
there are no variables in the term I, then of course the variable x does
not occur in I, and it is vacuously true that all variables in I (of course
there are none) occur in W. Also, Ix = x, and so I is an x-eliminate
of W.
(ii) Now we will show that if W is any term in which x does not occur,
then, by rule (2) KW is an x-eliminate of W. Clearly, x does not occur
in KW. Secondly, the variables of KW are variables of W, so that all
variables of KW are variables of W. Thirdly, KWx = W. Thus KW is
indeed an x-eliminate of W.
(iii) Finally, we consider the composite term WV in which x does
occur, so that rule (3) applies. Suppose that Wx is an x-eliminate of W,
and that Vx is an x-eliminate of V. We must show that SWx Vx is an
228 Part III: Elements of Combinatory Logic

x-eliminate of WV. Let us check the three conditions to see if SWx Vx is

an x-eliminate of WV:
(a) Since x does not occur in Wx (since Wx is an x-eliminate of W) and
x does not occur in Vx ), and x does not occur in S, it follows that
x does not occur in SWx Vx .
(b) Every variable y in SWx Vx must occur in WV, because y either
occurs in Wx or Vx or both, which means it occurs in W or V or
both. So it occurs in the compound term WV.
(c) Lastly, we must show that SWx Vx x = WV. Well,
SWx Vx x = Wx x(Vx x).
Since Wx x = W and Vx x = V, then Wx x(Vx x) = WV. Thus
SWx Vx x = WV.
By (a), (b), and (c), SWx Vx is an x-eliminate of WV (assuming Wx is
an x-eliminate of W and assuming Vx is an x-eliminate of V).
3. (a) The variable x does not occur in W (by hypothesis).
(b) All variables of W obviously occur in Wx.
(c) Obviously Wx = Wx.
4. It is clear that the statement (2) is true. Here is why:
(a) The variable y does not occur in (Wx )y . The variable x does not
occur in Wx , hence not in (Wx )y , since all the variables in (Wx )y
occur in Wx . Thus neither x nor y occurs in Wx,y .
(b) All variables of Wx occur in W, and all variables in (Wx )y occur in
Wx . Hence all variables of Wx,y occur in W.
(c) (Wx )y y = Wx , so that (Wx )y yx = Wx x = W.
By (a), (b), and (c), (Wx )y is an y, x eliminate of W.
5. For the first proof, let A be a combinator that realizes the term t. Thus
Axx1 . . . xn = t(x, x1 , . . . , xn ).
We let Γ be a fixed point of A, so that
Γx1 . . . xn = AΓx1 . . . xn = t(Γ, x1 , . . . , xn ).
Thus Γx1 . . . xn = t(Γ, x1 , . . . , xn ).
For the second proof, suppose A realizes the term t(xx, x1 , . . . , xn ),
i.e. for all x, x1 , . . . , xn
Axx1 . . . xn = t(xx, x1 , . . . , xn ).
 Chapter 11. Complete and Partial Systems 229

If we take A for x, we obtain

AAx1 . . . xn = t(AA, x1 , . . . , xn ).

We thus take AA to be Γ!

Note to the reader: I believe that virtually all fixed points and recur-
sion theorems in recursion theory and in combinatory logic, as well as
the construction of Gödel self-referential sentences, are simply elaborate
variations on this trick of the above second proof. In all cases, one has
an equation containing Ax of the left side of an equal sign, and xx on
the right side. Taking A for x, we end up with AA on both sides.
Now here is the solution of the Exercise:
(a) Since Qzxy = x(zy), we take Γ to be a fixed point of Q, so that

Γxy = QΓxy = x(Γy).

Thus Γxy = x(Γy).

(b) Since Czxy = zyx, we take Γ to be a fixed point of C, so that

Γxy = CΓxy = Γyx.

Thus Γxy = Γyx.

6. Again, there are two different ways of proving this, and both are of
interest.
For the first proof, we recall that by a cross point of a pair of compati-
ble points (A, B) is meant a pair of elements (x, y) such that Ax = y and
By = x. Since we are working with a complete system, the hypothesis
of Problem 6, Chapter 8 (that conditions C1 and C2 both hold) is met,
so that every pair of terms has a cross point (in the language of that
problem, any two combinators are compatible).
Now let A1 realize t1 and let A2 realize t2 . Thus
(3) A1 xx1 . . . xn = t1 (x, x1 , . . . , xn ).
(4) A2 xx1 . . . xn = t2 (x, x1 , . . . , xn ).
We now let (Γ1 , Γ2 ) be a cross point of (A2 , A1 ) [not of (A1 , A2 ) but of
(A2 , A1 )]. Thus, A2 Γ1 = Γ2 and A1 Γ2 = Γ1 . Then
(10 ) Γ1 x1 . . . xn = A1 Γ2 x1 . . . xn = t1 (Γ2 , x1 , . . . , xn ).
(20 ) Γ2 x1 . . . xn = A2 Γ1 x1 . . . xn = t2 (Γ1 , x1 , . . . , xn ).
230 Part III: Elements of Combinatory Logic

For the second proof, given terms t1 (x, x1 , . . . , xn ) and t2 (x, x1 , . . . , xn ),

let A1 and A2 be combinators satisfying the following conditions:
(1) A1 xyx1 . . . xn = t1 (xyx, x1 , . . . , xn ).
(2) A2 xyx1 . . . xn = t2 (xyx, x1 , . . . , xn ).
Then,
(10 ) A1 A2 A1 x1 . . . xn = t1 (A2 A1 A2 , x1 , . . . , xn ).
(20 ) A2 A1 A2 x1 . . . xn = t2 (A1 A2 A1 , x1 , . . . , xn ).
We thus take Γ1 = A1 A2 A1 and Γ2 = A2 A1 A2 .
7. We are given that W is a nice term and that x occurs in W. At each
stage of the construction of W, no new constant was introduced other
than S, B, C or I, so that Wx is certainly nice.
We leave it to the reader, using term induction, to prove that x does
not occur in Wx and that all variables in Wx occur in W. It remains to
show that Wx x = W. We do this by induction.
1 If W = x, then Wx is I and Wx x = Ix = IW = W. So Wx x = W.
It remains to show the statement for when we have a compound
term WV.
2(a) Suppose that x is in both W and V, and that Wx x = W and
Vx x = V. We are to show that (WV)x x = WV. Well,
(WV)x = SWx Vx ,
so that
(WV)x x = SWx Vx x = Wx x(Vx x) = WV.
2(b) For the case when x occurs in W but not in V, we assume by the
induction hypothesis that Wx x = W. Now, in this case,
(WV)x = CWx V,
so that
(WV)x x = (CWx V)x = Wx xV = WV.
2(c) Now for the case that x is in V but not in W.
(c1 ) Suppose V = x. Then (WV)x = W, so that
(WV)x x = Wx = WV.
(c2 ) In the case that Vx 6= x (and when x does not occur in W), we
assume that Vx x = V and we must show that (WV)x x = WV.
Well, in this case (WV)x = BWVx , so that
(WV)x x = BWVx x = W(Vx x) = WV.
This concludes the proof.
 Chapter 11. Complete and Partial Systems 231

8. (a) We take Q3 to be GI. Then GIxyz = Iz(xy) = z(xy) = Q3 xyz.

(b) We take C to be GGII. Then GGIIx = Gx(II) = GxI, so that
GGIIxyz = GxIyz = xz(Iy) = xzy = Cxyz.
(c) We can take T to be CI. Then CIxy = Iyx = yx = T xy.
Alternatively, we can take T to be Q3 I. Then
Q3 Ixy = y(Ix) = yx = T xy.
(d) We recall that CC is the robin R, since
CCxyz = Cyxz = yzx = Rxyz.
We now take Q to be GRQ3 . Then
GRQ3 xyz = Ry(Q3 x)z = Q3 xzy = y(xz) = Qxyz.
Finally, we take B to be CQ. Then
CQxyz = Qyxz = x(yz) = Bxyz.
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Chapter 12

Combinators, Recursion
and the Undecidable

In this chapter, we work with the complete system: all combinators

derivable from S and K. We shall now see how combinatory logic is related
to propositional logic, recursion theory, and the undecidable.

Logical Combinators

We now want two combinators t and f to stand for truth and falsity respec-
tively. Several workable choices are possible, and the one we take is due to
Henk Barendregt [1985]. We take t to be the kestrel K, and for f we take
the combinator KI, so that in what follows t is an abbreviation for K,
while f is an abbreviation for KI. We shall call t and f propositional com-
binators. We note that for any combinators x and y, whether propositional
combinators or not, txy = x and f xy = y, which turns out to be a technical
advantage.
There are only two propositional combinators t and f (i.e. K and KI).
We now use the letters p, q, r, s as standing for arbitrary propositional com-
binators, rather than propositions, and we call p true if p = t, and false if
p = f . We define the negation ∼ p of p by the usual conditions: ∼t = f
and ∼f = t; we define the conjunction p ∧ q by the truth-table conditions:
t ∧ t = t, t ∧ f = f, f ∧ t = f , f ∧ f = f . We define the disjunction p ∨ q,
the conditional p ⊃ q, and the bi-conditional p ≡ q similarly. We now want
combinators that can realize these operations.

Problem 1. Find combinators N (negation combinator), c (conjunction

combinator), d (disjunction combinator), i (implication combinator), and e
234 Part III: Elements of Combinatory Logic

(equivalence or bi-conditional combinator), satisfying the following condi-

tions:
(a) N p = ∼ p,
(b) cpq = p ∧ q,
(c) dpq = p ∨ q,
(d) ipq = p ⊃ q,
(e) epq = p ≡ q.
We now see that by combining the above combinators in various ways,
we have combinators that can do propositional logic, in the sense that they
can compute compound truth tables.

Arithmetic Combinators

We shall now see how the arithmetic of the natural numbers can be embed-
ded in combinatory logic.
Each natural number n is represented by a combinator denoted n (not
to be confused with the Peano numeral designating n). There are several
schemes of doing this, one of which is due to Church [1941]. The scheme
we will use is again due to Henk Barendregt [1985]. Here the vireo V plays
a major role [Vxyz = zxy]. We let σ (read as “sigma”) be the successor
combinator Vf (which is V(KI)). We take 0 to be the identity combinator
I, take 1 to be σ0; take 2 to be σ1, . . . , take n + 1 = σn, . . . . The first thing
that needs to be shown is that the combinators 0, 1, 2, . . . , are all distinct.

Problem 2. Show that if m 6= n, then m 6= n. [Hint: First show that for

every n, 0 6= n+ , where n+ is an abbreviation for n + 1. Then show that if
n+ = m+ , then n = m. Finally, show, for all m and n and all positive k,
that n 6= n + k.]
The combinators 0, 1, 2, . . . are called numerical combinators. A combi-
nator A is called an arithmetical combinator of type 1 if, for every n, there
is some m such that An = m. We call A an arithmetical combinator of type
2 if, for every n and m, there is some number p such that An m = p. In
general, A is called an arithmetical combinator of type n if, for all numbers
x1 , . . . , xn , there is a number y such that Ax1 . . . xn = y.
We now wish to show that there is an addition combinator, which we
will denote ⊕, such that ⊕m n = m + n, and a multiplication combinator,
 Chapter 12. Combinators, Recursion and the Undecidable 235

which we will denote ⊗, such that ⊗m n = m × n, and an exponential

combinator Egsuch that Egm n = mn .
We need some items in preparation for all this:
For any positive number n, by its predecessor is meant the number
n − 1. Thus, for any number n, the predecessor of n+ is n. We now need
a combinator P that calculates predecessors, in the sense that P n+ = n
for all n.

Problem 3. Find such a combinator P .

The Zero Tester

We have vital need of a combinator Z, called the zero tester, such that
Z0 = t and Zn = f for any positive n.

Problem 4. Find such a combinator Z.

Problem 5. Does there exist a combinator A such that A0xy = x, but

Anxy = y, for n positive?
We now consider some recursion properties.

Problem 6. Prove that for any combinator A and any number k, there is
a combinator Γ satisfying the following two conditions:
C1 : Γ0 = k.
C2 : Γn+ = A(Γn).

Hints:
1. C2 is equivalent to the following condition: For any positive n,
Γn = A(Γ(P n)).
2. Use the zero tester Z as in Problem 5.
3. Use the fixed point theorem.

Problem 7. Now show that there are combinators ⊕, ⊗ and Egsatisfying

the following conditions:
(1) ⊕ m n = m + n.
(2) ⊗ m n = m × n.
(3) Egm n = mn .
236 Part III: Elements of Combinatory Logic

The Recursion Property for Combinatory Logic, in its most general

form, is that for any terms f (y1 , . . . , yn ) and g(x, z, y 1 , . . . , yn ), there is
a combinator Γ such that for all numerals x, y1 , . . . , yn , the following two
conditions hold:
(1) Γ0y1 . . . yn = f (y1 , . . . , yn ).
(2) Γx y1 . . . yn = g((P x)(Γ(P x)y1 . . . yn ), y1 , . . . , yn ).

A combinator Γ that does this exists by the fixed point theorem, and is
defined by the condition:
Γxy1 . . . yn = Zxf (y1 , . . . , yn )g((P x), (Γ(P x)y1 . . . yn ), y1 , . . . , yn ).

This brilliant idea was due to Alan Turing!

I. Preparation for the Finale

Property Combinators

By a property combinator is meant a combinator such that for every number

n, Γn = t or Γn = f .
A set A of (natural) numbers is said to be (combinatorially) computable
if there is a property combinator Γ such that Γn = t for every n in A,
and Γn = f for every n that is not in A. Such a combinator Γ is said to
compute A.
We might remark that the important thing about combinatorial com-
putability is that a set A is combinatorially computable if and only if it is
recursive.

Problem 8. Show that the set E of even numbers is computable.

[Hint: The property of being even is the one and only property sat-
isfying the following conditions: (1) 0 is even; (2) for positive n, n is
even if and only if its predecessor is not even. Now use the fixed point
theorem.]

Problem 9. Suppose Γ computes A. Does it follow that N Γ computes the

complement A of A? [N is the negation combinator Vft.]
 Chapter 12. Combinators, Recursion and the Undecidable 237

Relational Combinators

By a relational combinator of degree n is meant a combinator A such that

for all numbers k1 , . . . , kn , either Ak1 . . . kn = t or Ak1 . . . kn = f , and
such a combinator A is said to compute the set of all n-tuples (k 1 , . . . , kn )
such that Ak1 . . . kn = t. Thus for any relation R(x1 , . . . , xn ), we say that A
computes R to mean that for all n-tuples (k 1 , . . . , kn ), if R(k 1 , . . . , kn ) holds,
then Ak1 . . . kn = t, and if R(k 1 , . . . , kn ) doesn’t hold, then Ak1 . . . kn = f .

Problem 10. Show that there is a relational combinator g that computes

the relations x > y (x is greater than y). [Hint: This relation is uniquely
determined by the following conditions:
1. If x = 0, then x > y is false.
2. If x 6= 0 and y = 0, then x > y is true.
If x 6= 0 and y 6= 0, then x > y is true if and only if x − 1 > y − 1 is
true.]

Functional Combinators

Consider a function f (x) of one argument, i.e. an operation that assigns to

each number n a number devoted by f (n). A combinator A will be said to real-
ize the function f (x) if, for every number n, the condition An = f (n) holds.

Problem 11. Show that if functions f (x) and g(x) are both realizable, so
is the function f (g(x)), i.e. the operation that assigns to each number n
the number f (g(n)).

Problem 12. Suppose R(x, y) is computable and that f (x) and g(x) are
both realizable. Show that the following are computable:
(a) The relation R(f (x), y).
(b) The relation R(x, g(y)).
(c) The relation R(f (x), g(y)).

The Minimization Principle

Consider a relational combinator A of degree 2 such that for every number

n, there is at least one number m such that An m = t. Such a combinator
238 Part III: Elements of Combinatory Logic

is sometimes called regular. If A is regular, then, for every n, there must

be a smallest number k such that Ank = t. The minimization principle
is that for every regular combinator A, there is a combinator A0 , called a
minimizer of A, such that for every number n, A0 n = k, where k is the
smallest number for which Ank = t.
We seek to prove the minimization principle. The following preliminary
problem will be most helpful:

Problem 13. Show that for any regular relational combinator A of

degree 2, there is a combinator A1 such that for all numbers n and m,
(1) If An m = t, then A1 n m = m.
(2) If An m = f , then A1 n m = A1 nm + 1.

Problem 14. Now prove the Minimization Principle: for every regular
relational combinator A of degree 2, there is a combinator A0 , called a
minimizer of A, such that for every number n, A0 n = k, where k is the
smallest number for which Ank = t. [Hint: Use the A1 of Problem 13 and
the cardinal C [Cxyz = xzy].]

The Length Measurer

By the length of a number n we shall mean the number of digits in n when

n is in ordinary base 10 notation. The numbers from 0 to 9 have length 1;
those from 10 to 99 have length 2, those from 100 to 999 have length 3, etc.
Actually the length of n is the smallest number k such that 10k > n.
We now want a combinator L0 that measures the length of any number,
that is, we want such that L0 n = k, where k is the length of n.

Problem 15. Prove that there is a length measurer L0 .

Concatenation to the Base 10

For any numbers n and m, by n ∗ m we shall now mean the number

which, when written in ordinary base 10 notation, consists of n (in base
10 notation) followed by m (in base 10 notation). For example
47 ∗ 386 = 47386.

We now want a combinator ∗gthat realizes this operation of concatena-

tion ∗.
 Chapter 12. Combinators, Recursion and the Undecidable 239

Problem 16. Show that there is a combinator ∗gsuch that for every n and
m, ∗gn m = n ∗ m. [Hint: x ∗ y = (x × 10k ) + y, where k is the length of y.
For example,
26 ∗ 587 = 26000 + 587 = (26 × 103 ) + 587,
and 3 is the length of 587.]

II. The Grand Problem

By an S, K term we shall mean a term without variables, built from just

the constants S and K. We use letters X, Y, Z to stand for unspecified S,
K terms. In what follows “term” will mean S, K term.
Of course, distinct terms might designate the same combinator. For
example, KKK designates the same combinator as KKI. For both desig-
nate the kestrel K, or as we say the equation KKK = KKI holds, or is
true. An equation is true if and only if its being so is a consequence of the
given conditions (which can be called the axioms for combinatory logic) —
SXY Z = XZ(Y Z) and KXY = X — and by using the inference rules of
identity, namely X = X and if X = Y , then XZ = Y Z and ZX = ZY ;
also, the additional rules of equality: if X = Y then Y = X; if X = Y and
Y = Z then X = Z.
The grand question now is this: Given two terms X and Y , is there
a systematic way of deciding whether or not the equation X = Y holds?
This can be reduced to a question of deciding whether a given number
belongs to a certain set of numbers. This translation uses the device of
Gödel numbering. How this translation is effected is the first thing we shall
now consider.
All sentences (frequently called equations here) are built from the fol-
lowing five symbols:
S K ( ) =
1 2 3 4 5
In this section all such sentences will be equations of the form
S, K term = S, K term.

Under each symbol I have written its Gödel number. The Gödel num-
ber of a compound expression (a complex term, or a sentence) is the
number obtained by replacing S by 1, K by 2, . . . , = by 5, and reading
240 Part III: Elements of Combinatory Logic

the result as a number in base 10. For example, the Gödel number of
KS(= is 2135).
We let T be the set of true sentences, and let T0 be the set of the Gödel
numbers of the true sentences. The question now is whether the set T0 is
combinatorially computable. Is there a combinator Γ such that for every n,
Γn = t if n ∈ T0 , and Γn = f if n ∈
/ T0 ?
The question is of staggering importance, since any formal mathemat-
ical question can be reduced to whether a certain number belongs to T0 .
The question is thus equivalent to the question of whether there can be a
universal computer that can settle all formal mathematical questions.
As the reader has doubtless guessed, the answer is no, which we now
set out to prove.

Gödel Numerals

Let me first point out that there are two different schemes for constructing
terms, which involve two different ways of introducing parentheses. One
scheme consists of building terms according to the following two rules:
(1) S and K are terms.
(2) If X and Y are terms, so is (XY ).
The second scheme replaces (2) by:
(20 ) If X and Y are terms, so is (X)(Y ).
We shall continue to use the first scheme.
Now, for any number n, the numeral n, like any other term, has a Gödel
number. We let n# be the Gödel number of the numeral n. For example, 0
is the combinator I, which in terms of S and K is ((SK)K), whose Gödel
number is 3312424. Thus 0# = 3312424.
As for 1# , it is the Gödel number of the numeral 1, and 1 = (σ0), where
σ is the combinator Vf , and Vf , when reduced to S and K, is a horribly
long expression, and consequently has an unpleasantly long Gödel number,
which I will abbreviate by the letter “s”. Thus, in what follows, s is the
Gödel number of (the term whose abbreviation is) σ, and so 1# is the Gödel
number of (σ0), which is 3 ∗ s ∗ 0# ∗ 4.
Next, 2# uses the Gödel number of (σ1), and the Gödel number of that
term is 3 ∗ s ∗ 1# ∗ 4. Similarly 3# = 3 ∗ s ∗ 2# ∗ 4. And so on . . . .
 Chapter 12. Combinators, Recursion and the Undecidable 241

We now need to show that the function f (n) = n# is combinatorially

realizable, i.e. that there is a combinator δ such that δn = n# holds (for
every n).

Problem 17. Show that there is such a combinator δ.

Normalization

For any expression X, by dXe is meant the numeral that designates the
Gödel number of X; we might call dXe the Gödel numeral of X. By the
norm of X is meant XdXe, i.e. X followed by the Gödel numeral of X. If n
is the Gödel number of X, then n# is the Gödel number of dXe, so n ∗ n#
is the Gödel number of XdXe, i.e. of the norm of X.
We now want a combinator ∆, called a normalizer, such that
∆n = n ∗ n# holds, for all n.

Problem 18. Exhibit such a combinator ∆.

Problem 19. Which, if either, of the following statements is true?

(a) ∆dXe = XdXe
(b) ∆dXe = dXdXee
The normalizer can do some amazing things, as you will see!

The Second Fixed Point Principle

We say that a term X designates a number n if the equation X = n

holds. Obviously one term that designates n is the numeral n, but there
are many others. For example, for n = 8, the terms ⊕4 4, ⊕5 3, ⊕2 6 all
designate 8. Indeed there are infinitely many terms that designate n, e.g.
In, I(In), I(I(In)), etc.
We call a term an arithmetic term if it designates some number. Every
numeral is an arithmetic term, but not every arithmetic term is a numeral.
It is impossible for any numeral to designate its own Gödel number,
because the Gödel number of n is larger than n (n# > n). However, this does
not mean that no arithmetic term can designate its own Gödel number. In
fact, there is an arithmetic term that designates its own Gödel number! Also
there is one that designates twice its Gödel number; one that designates
242 Part III: Elements of Combinatory Logic

eight times its Gödel number plus 15 — indeed, for any realizable function
f (x), there is an arithmetic term X which designates f (n), where n is
the Gödel number of X! This is immediate from the second fixed point
principle, which is that for any combinator A, there is a term X such that
AdXe = X holds. We now wish to prove this principle.

Problem 20. Prove the Second Fixed Point Theorem: show that for any
combinator A, there is a term X such that AdXe = X holds.

Problem 21. Prove that for any realizable function f (x) there is an (arith-
metic term) X that designates f (n), where n is the Gödel number of X.
[Hint: Use the second fixed point theorem.]

Representability

We shall say that a combinator Γ represents a number set A if for every

number n, the equation Γn = t holds iff n ∈ A. Thus Γ represents the set
of all n such that Γn = t holds. (Note that the combinatorial computability
of a number set A by a combinator Γ that we worked with earlier requires
more than representability, for in the case of combinatorial computability
if n ∈
/ A then it must be the case that Γn = f , while for representability, it
is only necessary that Γn 6= t.)

Problem 22. Suppose Γ computes A. Which, if either, of the following

statements are true?
(1) A is representable.
(2) The complement A of A is representable.
Recall that for any function f (x) and set A, by f −1 (A) is meant the set
of all numbers n such that f (n) ∈ A. Thus n ∈ f −1 (A) iff f (n) ∈ A.

Problem 23. Prove that if the function f (x) is realizable, and if A is

representable, then f −1 (A) is representable.

Gödel Sentences

We shall call a sentence (equation) X a Gödel sentence for a number set

A if X is true if and only if the Gödel number of X is in A. We now
aim to prove that if A is representable, then there is a Gödel sentence
for A.
 Chapter 12. Combinators, Recursion and the Undecidable 243

Problem 24. Suppose A is representable. Prove that the set of all n such
that (n ∗ 52) ∈ A is representable. [The significance of n ∗ 52 is that if n is
the Gödel number of X, then n ∗ 52 is the Gödel number of X = t.]

Problem 25. Now prove that for any representable set A, there is a Gödel
sentence for A.
Now we have all the key pieces to prove that the set T0 of Gödel numbers
of the true sentences is not computable.

Problem 26. Prove that T0 is not computable.

Discussion

We have just seen that the complement of the set T0 is not representable.
What about the set T0 ? Is it representable? Yes, it is. As suggested earlier,
combinatory logic can be formalized. The following is an axiom system for
the full combinatory logic:

Symbols: S, K, (, ), =

Terms: As previously defined recursively

Sentences (or Equations): Expressions of the form X = Y , where X and Y

are terms.

Axiom Schemes: For any terms X, Y, Z:

(1) SXY Z = XZ(Y Z)
(2) KXY = X
(3) X = X.

Inference Rules
R1 : From X =Y to infer Y = X.
R2 : From X =Y to infer XZ = Y Z.
R3 : From X =Y to infer ZX = ZY .
R4 : From X =Y and Y = Z to infer X = Z.

It is relatively easy to construct an elementary formal system in which

one can successively represent the set of terms, sentences, and provable
sentences. Thus the set T is formally representable, from which it easily
244 Part III: Elements of Combinatory Logic

follows that the set T0 is recursively enumerable. And it is well known that
recursive enumerability is the same thing as representability in combinatory
logic. Thus the set T0 is indeed representable.
The set T0 is another example of a recursively enumerable set that is
not recursive.
As previously mentioned, any formal mathematical question can be
reduced to a question of whether a certain number is in T0 , that is, one
can associate with each formal mathematical question a number n such
that n ∈ T0 iff the question has an affirmative answer. Since there is no
purely mechanical method of determining which numbers are in T0 , there
is then no purely mechanical method of determining which formal math-
ematical statements are true and which are false. Mathematics requires
intelligence and ingenuity. Any attempt to fully mechanize mathematics is
doomed to failure. In the prophetic words of Emil Post [1944], “Mathe-
matics is and must remain essentially creative.” Or, in the witty words of
Paul Rosenbloom: “Man can never eliminate the necessity of using his own
intelligence, regardless of how cleverly he tries.”

Solutions to the Problems of Chapter 12

1. (a) We need a negation combinator N satisfying the condition

N x = xf t. Well, we can take N to be V f t, where V is the
vireo [V xyz = zxy]. Thus N x = V f tx = xf t as desired. Then
N t = tf t = f (since t = K, or because txy = x, for all x and y),
and N f = f f t = t (since f = KI, or because f xy = y, for all x
and y). Thus N p =∼ p, where p is either t or f .
(b) This time we want the conjunction combinator c to be such that
cxy = xyf . We take c to be Rf , where R is the Robin [Rxyz = yzx].
Thus cxy = Rf xy = xyf , as desired. Then (using the fact that for
all x and y, txy = x and f xy = f ):
(1) ctt = ttf = t,
(2) ctf = tf f = f ,
(3) cf t = f tf = f ,
(4) cf f = f f f = f .
(c) Now we want the disjunction combinator d to be such that
dxy = xty.
 Chapter 12. Combinators, Recursion and the Undecidable 245

For this purpose, we take d to be T t, where T is the thrush

[T xy = yx]. Then dxy = xty (verify!). And dpq = p ∨ q (verify
all four cases!).
(d) Take the implication combinator i to be Rt, where R is the robin
[Rxyz = yzx]. Then ixy = xyt, because ixy = Rtxy = xyt. Then
ipq = p ⊃ q (verify!).
(e) We want the equivalence combinator e to be such that
exy = xy(N y).
For this we may take e to be CSN:
CSN xy = SxN y = xy(N y).
Then epq has the same values as p ≡ q. Verify, recalling that we
have shown above that N t = f , that Nf = t, and that, for all x
and y, txy = x and f xy = f .
2. Suppose 0 = n+ for some n. Thus I = σn = V f n. Consequently,
IK = V f nK = Kf n = f . Thus IK = KI, so that K = KI contrary
to Problem 19 of Chapter 8.
Next, we must show that if n 6= m, then n+ 6= m+ . To do so, we
instead show the equivalent fact that if n+ = m+ then n = m.
Well, suppose that n+ = m+ . Then σn = σm, or V f n = V f m, so
that V f nf = V f mf . Applying V here, we see that f f n = f f m, which
implies that n = m [again since it is always true that f xy = y]. This
proves that if n 6= m, then n + 1 6= m + 1.
Now, for any positive k, we saw at the beginning of this solution
that 0 6= k. After what we have just proved, we now see as well that
it follows from 0 6= k that 1 6= 1 + k, 2 6= 2 + k, . . . , n 6= n + k, . . . .
This proves that if m 6= n, then m 6= n (because if m 6= n then for some
positive k, m = n + k or n = m + k).
3. Since n+ = σn, we want a combinator P such that P (σn) = n. We
take P to be T f , where T is the thrush [T xy = yx]. Then
P (σn) = T f (σn) = σnf = Vf nf = f f n = n.
Voila!
4. We take Z to be T t, where T is the thrush [T xy = yx]. Recall that
0 = I and σ = Vf .
(1) Z0 = T t0 = 0t = It = t. Thus Z0 = t.
246 Part III: Elements of Combinatory Logic

(2) Zn+ = T tn+ = n+ t = σnt = Vf nt = tf n = f. Thus Zn+ = f .

5. Yes, the zero tester Z is such an A.
Z0xy = txy = x.
Zn+ xy = f xy = y.
6. Conditions C1 and C2 are respectively equivalent to:
C10 : Γx = k if x = 0.
C20 : Γx = A(Γ(P x)) if x = n, for n positive
Now, by Problem 5, Zxk(A(Γ(P x))) is k if x = 0, and is A(Γ(P x)) if
x = n, for n positive. We thus want a combinator Γ which satisfies the
condition:
Γx = Zxk(A(Γ(P x))).
Well, such a combinator Γ exists by the fixed point theorem. Specifi-
cally, we can take Γ to be a fixed point of a combinator θ satisfying the
condition: θyx = Zxk(A(y(P x))). Then Γx = θΓx = Zxk(A(Γ(P x))).
7. (1) The addition combinator ⊕ is uniquely determined by the following
two conditions:
a. n + 0 = n.
b. n + m+ = (n + m)+ .
We therefore want a combinator ⊕ such that:
a0 . ⊕n0 = n.
b0 . ⊕nm+ = σ(⊕n m).
We thus want ⊕ to be such that, for any n and m, whether 0 or
positive, the following holds: ⊕n m = Zm n(σ(⊕(n(P m)))).
Such a combinator exists by the fixed point theorem.
(2) The multiplication operation × is uniquely determined by the fol-
lowing two conditions:
a. n × 0 = 0.
b. n × m+ = (n × m) + n.
Thus the combinator ⊗ that we want, and which exists by the fixed
point theorem, is defined by the condition:
⊗n m = Zm0(⊕(⊗n(P m))n).
(3) The exponential operation is uniquely determined by the following
two conditions:
a. n0 = 1.
+
b. nm = nm × n.
 Chapter 12. Combinators, Recursion and the Undecidable 247

Thus the combinator Egwe want, and which exists by the fixed point
theorem, is defined by the condition Egm n = Zm1(⊗( Egn(P m))n).
8. By virtue of conditions (1) and (2) of the hint, we want a combinator
Γ to satisfy the condition:
Γx = Zxt(N (Γ(P x))).
Such a combinator can be found by the fixed point theorem.
9. No, it does not follow that N Γ computes A. Suppose n ∈ A. Then
Γn = t. However, it does not follow that N Γn = f ; what does follow is
that N (Γn) = f . Thus it is not N Γ that computes A but rather BN Γ
[Bxyz = x(yz)]. For BN Γn = N (Γn) = f when Γn = t (i.e. when
n ∈ A) and BN Γn = N (Γn) = t when Γn = f (i.e. when n ∈ A).
10. We want a combinator g such that for all numbers x and y the following
conditions hold:
(1) If Zx = t, then gx y = f .
(2) If Zx = f , then
(a) If Zy = t, then gx y = t.
(b) If Zy = f , then gx y = g(P x)(P y).
We thus want g to satisfy the following:
gxy = Zxf (Zyt(g(P x)(P y))).
Such a g exists by the fixed point theorem.
11. Suppose A1 realizes f (x) and A2 realizes g(x). Then the functions
f (g(x)) is realized by BA1 A2 , since
BA1 A2 n = A1 (A2 n) = A1 g(n) = f (g(n)).
12. Let Γ compute the relation R(x, y), and let A1 realize f (x) and A2
realize g(x). Then:
(a) BΓA1 computes the relation R(f (x), y), because:
BΓA1 n m = Γ(A1 n)m = Γf (n)m,
which is t if R(f (n), m) holds, and is f if R(f (n), m) doesn’t hold.
Thus BΓA1 computes R(f (x), y).
(b) This is trickier! A combinator that computes the relation R(x, g(y))
is BCDΓA2 , where C is the cardinal [Cxyz = xzy] and D is
the combinator defined by Dxyzw = xy(zw). We learned that
D = BB.
Well,
BCDΓA2 n m = C(DΓ)A2 n m = DΓnA2 m = Γn(A2 m) = Γng(m),
248 Part III: Elements of Combinatory Logic

which is t if R(n, g(m)) holds, and is f if R(n, g(m)) doesn’t hold.

Thus BCDΓA2 computes R(x, g(y)).
(c) Let S(x, y) be the relation R(x, g(y)), which is computed by
BCDΓA2 . Then by (a) S(f x), y), which is R(f (x), g(y)) is com-
puted by B(BCDΓA2 )A1 .
Alternatively, let δ(x, y) be the relation R(f (x), y). It is com-
puted by BΓA1 , according to (a). Then R(f (x), g(y)) is δ(x, g(y)),
hence is computed by BCD(BΓA1 )A2 , according to (b).
13. Given a regular relational combinator A of degree 2, by the fixed point
theorem, there is a combinator A1 satisfying the condition:
A1 xy = (Axy)y(A1 x(σy)).
Thus, for any numbers n and m, A1 n m = (An m)m(A1 nm + 1).
If An m = t then A1 n m = tm(A1 nm + 1) = m.
If An m = f then A1 n m = f m(A1 nm + 1) = A1 nm + 1.
14. We take A0 to be CA1 0, where C is the cardinal and A1 is the combi-
nator of Problem 13. Then, for any number n, A0 n = CA1 0n = A1 n0.
Now let k be the smallest number such that Ank = t. We illustrate
the proof for k = 3 (the reader should have no trouble in generalizing
the proof for arbitrary k). Thus it must be the case that An0 = f ;
An1 = f ; An2 = f ; An3 = t.
Now, A0 n = A1 n0. Since An0 = f , then A1 n0 = A1 n1. Since An1 = f ,
then A1 n1 = A1 n2. Since An2 = f , then A1 n2 = A1 n3. Since An3 = t,
then A1 n3 = 3. Thus A0 n =, A1 n0 = A1 n1 = A1 n2 = A1 n3 = 3. So
A0 n = 3.
15. Let R(x, y) be the relation 10y > x. We must first find a combinator A
that computes the relation 10y > x.
We let R1 (x, y) be the relation R(y, x), i.e. the relation 10x > y. If
we find a combinator A1 that computes the relation R1 (x, y), then CA1
will compute the relation R(x, y) (why?).
To compute the relation 10x > y, we know that g computes the
relation x > y and that Eg 10 realizes the function 10x . Hence by
Problem 11, the relation 10x > y is computed by Bg( Eg 10). Thus
C(Bg( Eg10)) computes the relation 10y > x. We thus take the length
measurer L0 to be a minimizer of C(Bg( Eg10)).
16. Take ∗gsuch that ∗gxy = ⊕(⊗x( Eg10(L0 y))y.
 Chapter 12. Combinators, Recursion and the Undecidable 249

17. Let A be a combinator satisfying the equation An = 3 ∗ s ∗ n ∗ 4.

[Specifically, we can take A to be B(C ∗g 4)( ∗g 3 ∗ s), where B is
the bluebird and C is the cardinal, as the reader can verify.] Then
# #
An# = (n + 1) [since An# = 3 ∗ s ∗ n# ∗ 4 = (n + 1) ].
Then, by the Recursion Property, or more directly by Problem 6,
there is a combinator δ such that:
(1) δ0 = 0# .
(2) δn + 1 = A(δn).
By mathematical induction it follows that δn = n# , by the following
reasoning:
(a) δ0 = 0# , by (1).
(b) Now suppose n is such that δn = n# .
#
We must show that δn + 1 = (n + 1) . Well, by (2), δn + 1 = A(δn).
And, of course, δn = n# , by the induction hypothesis. Therefore,

A(δn) = An# = (n + 1)# .

And so

δn + 1 = (n + 1)# .

This completes the induction.

18. Take ∆ to be S ∗gδ, where S is the starling [Sxyz = xz(yz)]. Then,
recalling that δn = n# we have:

∆n = S ∗gδn = ∗gn(δn) = ∗gnn# = n ∗ n# .

Note: In To Mock a Mockingbird, I took ∆ to be W (DC ∗gδ), which

also works, which should not be surprising, since for any x, y and z,
Sxyz = W (DCxy)z, as the reader can verify.
19. It is (b) that is true. Let n be the Gödel number of X. Then n = dXe,
so that ∆n = ∆dXe, and since ∆n = n ∗ n# , we have:
(1) ∆dXe = n ∗ n# .
Since n is the Gödel number of X, then n# , which we first defined to be
the Gödel number of n, must also be the Gödel number of dXe, since
n = dXe. Thus n ∗ n# is the Gödel number of XdXe, and so:
(2) n ∗ n# = dXdXee
(c) By combining (1) and (2), we see (b), i.e. that ∆dXe = dXdXee.
250 Part III: Elements of Combinatory Logic

20. We take X to be BA∆dBA∆e where ∆ is the normalizer. By the def-

inition of the bluebird B, we have BA∆dBA∆e = A(∆dBA∆e). Con-
sequently,
(1) X = A(∆dBA∆e) But also:
(2) A(∆dBA∆e) = AdXe.
Here is the reason for (2): By Problem 19, ∆dY e = dY dY ee for any Y .
Hence
∆dBA∆e = dBA∆dBA∆ee,

which is dXe. Since ∆dBA∆e = dXe, then A(∆dBA∆e) = AdXe,

which proves (2).
By (1) and (2), we have X = AdXe and so AdXe = X.
21. Suppose the function f (x) is realizable, and let A realize f (x). By the
second fixed point principle just proved, there is a term X such that
X = AdXe. Let n be the Gödel number of X, so that n = dXe, and
therefore AdXe = An and so X = An. Also An = f (n), since A realizes
f (x), which means that X designates f (n), where n is the Gödel number
of X.
22. Both (1) and (2) are true: Suppose Γ computes A. This means that for
every n:
(a) n ∈ A implies that Γn = t;
(b) n ∈
/ A implies that Γn = f .
(1) We must show that A is representable. Well, it is Γ itself that
represents A. To see this, it suffices to see that the converse of
(a) holds, i.e. that if Γn = t, then n ∈ A. So suppose Γn = t. If n
were not in A, then by (b) we would have that Γn = f , and hence
that t = f , which cannot be. Thus n ∈ A.
(2) Now we must show that A is representable. Since Γ computes A,
then BN Γ computes the complement A of A [by the solution of
Problem 9]. Thus BN Γ represents A [by (1)].
23. Suppose that Γ1 realizes f (x) and that Γ2 represents A. We let
Γ = BΓ2 Γ1 [Bxyz = x(yz)], and we show that Γ represents f −1 (A).
Well,
Γn = BΓ2 Γ1 n = Γ2 (Γ1 n) = Γ2 f (n).
Thus Γn = Γ2 f (n), so that Γn = t iff Γ2 f (n) = t, which is true iff
f (n) ∈ A [since Γ2 represents A], which is true iff n ∈ f −1 (A). Thus
Γn = t iff n ∈ f −1 (A), which means that Γ represents f −1 (A).
 Chapter 12. Combinators, Recursion and the Undecidable 251

24. This is but a special case of Problem 23: Let f (x) = x ∗ 52. By Problem
16, f (x) is realizable by the combinator ∗gx52. If A is representable,
so is f −1 (A) [by Problem 23], and f −1 (A) is the set of all n such that
f (n) ∈ A, i.e. the set of all n such that (n ∗ 52) ∈ A.
25. Suppose A is representable. Let A0 be the set of all n such that the
number n ∗ 52 is in A. Then A0 is representable [by Problem 24]. Let
Γ represent the set A0 . By the fixed point theorem, there is some X
such that ΓdXe = X. We will show that the sentence X = t is a Gödel
sentence for A.
Let n be the Gödel number of X. Then n = dXe. Hence Γn = ΓdXe,
and therefore ΓdXe = t iff Γn = t, which is true iff n ∈ A0 [since
Γ represents A0 ]. And n ∈ A0 iff (n ∗ 52) ∈ A. Thus ΓdXe = t iff
(n ∗ 52) ∈ A. But n ∗ 52 is the Gödel number of the sentence X = t.
Thus the sentence X = t is a Gödel sentence for A.
26. There cannot be a Gödel sentence for the complement of T0 , for such a
sentence X would be true if and only if its Gödel number n was not in
T0 , which means that X would be true iff its Gödel number was not the
Gödel number of a true sentence, and this is impossible. [Looked at it
another way, every sentence is a Gödel sentence for the set T0 , and no
sentence can be a Gödel sentence for both a set A and its complement
(why?). Hence no sentence can be a Gödel sentence for the complement
of T0 .]
Since there is no Gödel sentence for the complement of T0 , then the
complement of T0 is not representable [by Problem 25], and therefore
the set T0 is not computable [by Problem 22].
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Afterword
Where to Go from Here

This book and its predecessor have only scratched the surface of the
collection of topics subsumed under the name of “Mathematical Logic.”
We have done only the beginnings of the fields of recursion theory and
combinatory logic, and have done virtually nothing in model theory, proof
theory, and other logics such as modal logic, intuitionistic logic, relevance
logic, and others.
I strongly suggest that you next turn your attention to set theory and
the continuum problem. Let me tell you a wee bit about this, hoping to
whet your appetite.
One purpose of the subject known as axiomatic set theory is to develop
all mathematics out of the notions of logic (the logical connectives and the
quantifiers) together with the notion of an element being a member of a
set of elements. We have used the symbol “∈” for “is a member of” and
“x ∈ A” for “x is a member of the set A”.
A pioneer in the development of axiomatic set theory was Gottlob Frege
[1893]. His system had, in addition to axioms of first-order logic, just one
axiom of set theory, namely that for any property P , there is a unique
set consisting of those and only those things that have the property P .
Such a set is written {x : P (x)}, which is read “the set of all x’s having the
property P .
This principle of Frege is sometimes referred to as the abstraction prin-
ciple, or the unlimited abstraction principle. One can define the identity
relation x = y in terms of set inclusion as “x and y belong to the same sets”
[i.e. ∀ z(x ∈ z ≡ y ∈ z)]. Frege’s abstraction principle has the marvelous
254 Afterword

advantage of allowing us to obtain just about all the sets necessary for
mathematics, for example, the following:
P1 : The empty set ∅, which is {x : ∼ (x = x)}.
P2 : For any two elements a and b, the set {a, b} whose members are just a
and b. This set is {x : x = a ∨ x = b}.
P3 : For any set a, the power set P(a), i.e. {y : y ⊆ a}, thus

{y : ∀ z(z ∈ y ⊃ z ∈ a)}.

P4 : For any set a, the union ∪ a, i.e. the set of all elements that are members
of an element of a, thus ∪ a = {x : ∃ y(y ∈ a ∧ x ∈ y)}.

Despite the useful things Frege’s system can do, it has an extremely
serious drawback: it is inconsistent! This was pointed out to Frege in a
letter by Bertrand Russell [1902], who observed that according to Frege’s
abstraction principle, there would exist the set A of all sets which are not
members of themselves (A = {x : x ∈ / x}). Thus, for any set x, one would
have x ∈ A iff x ∈/ x. Taking A for x, we would have A ∈ A iff A ∈ / A,
which is a contradiction! Thus Frege’s system is inconsistent.
Frege was broken-hearted over Russell’s discovery, and felt that his
whole life’s work had been in vain. Here he was wrong: His work was sal-
vaged by Zermelo and others, and is really the basis for Zermelo–Fraenkel
set theory (frequently abbreviated as ZF), which is one of the most signif-
icant systems of set theorem in existence, the other of similar rank being
the system in Principia Mathematica of Whitehead and Russell [1910].
What Zermelo [1908] did was to replace Frege’s abstraction principle
by the following, known as the limited abstraction principle, which is that
for any property P of sets and any set a, there exists the set of all the
elements of a having property P . Thus for any set a, there exists the set
{x : P (x) ∧ x ∈ a}. This principle appears to be free of any possible contra-
diction. However, Zermelo had to take the existence of the sets ∅, {a, b},
P(a), and ∪ a as separate axioms. He also took an additional axiom known
as the axiom of infinity, which we will discuss later.
On the basis of the Zermelo axioms on hand, one can now derive the
positive integers. First, let us note that the set {a, b} is well-defined even if
a and b are the same element, and then {a, a} is simply the set {a}. Well,
Zermelo took 0 to be the empty set ∅; 1 to be the set whose only element is
0; 2 to be the set whose only element is 1, and so forth. Thus the Zermelo
 Where to Go From Here 255

natural numbers are ∅, {∅}, {{∅}}, etc. Thus for each natural number n,
n + 1 is {n}.
Later, Von Neumann [1923] proceeded differently. In his scheme, each
natural number is the set of all lesser natural numbers. Thus 0 is ∅ and 1 is
{∅} as with Zermelo, but 2 is now not {1}, but rather the set {0, 1}; 3 is the
set {0, 1, 2}; . . . , n + 1 is the set {0, 1, . . . , n}. This is the scheme now gen-
erally adopted, since it generalizes in the infinite case to the important sets
called ordinals, which we will soon discuss. Note that in Zermelo’s scheme,
each natural number other than 0 contains just one element, whereas in
Von Neumann’s scheme, each number n contains exactly n elements.
Now that we have the natural numbers, how do we define what it means
to be a natural number? That is, we want a formula N (x) using just the
logical connectives and quantifiers and the single predicate symbol ∈ such
that for any set a, the sentence N (a) is true if and only if a is a natural
number. Well, for any set a, we define a+ as a ∪ {a}, i.e. the set whose
elements are those of a, together with a itself. Thus

1 = 0+ = {0}, 2 = 1+ = 1 ∪ {1}; 3 = 2+ = 2 ∪ {2}

(which is {0, 1, 2}), etc.

We now define a set a to be inductive if 0 ∈ a, and if, for all sets b, if

b ∈ a, then b+ ∈ a. We then define a set x to be a natural number if
it belongs to all inductive sets. The principle of mathematical induction
follows immediately from the very definition of a natural number. Indeed,
the five Peano Postulates follow from the definition of “natural number”.
However, we don’t yet have any guarantee that there is such a thing as
the set of all natural numbers. Zermelo had to take this as an additional
axiom, and this is known as the axiom of infinity. The set of all natural
numbers is denoted “ω”.
Let us recall the Peano Postulates:

1. 0 is a natural number.
2. If n is a natural number, so is n+ .
3. There is no natural number such that n+ = 0.
4. If n+ = m+ , then n = m.
5. [Principle of Mathematical Induction] Every inductive set contains all
the natural numbers.
256 Afterword

All these five postulates are easy consequences of the definition of nat-
ural number. The axiom of infinity is not necessary to prove the Peano
Postulates! This fact appears to not be completely well known. I recall that
when I once announced these facts at a lecture, a very eminent logician
later told me that he was quite shocked!
Now for the ordinals: First we define a set x to be transitive if it con-
tains with each of its elements y all elements of y as well; in other words,
each element of x is a subset of x. We recall that for any set a, we are
taking a+ to be the set a ∪ {a}, and a+ is called the successor of a.
Roughly speaking, the ordinals are those sets, starting with ∅, which can
be obtained by taking the successor of any ordinal already obtained, and
by taking any transitive set of ordinals already obtained. That is, we want
to define ordinals such that 0 is an ordinal, every successor of an ordi-
nal is an ordinal, and every transitive set of ordinals is an ordinal. One
definition that works is this: Define x to be an ordinal if x is transitive
and every transitive proper subset of x is a member of x. [We recall that
by a proper subset of x is meant a subset of x that is not the whole of
x.] It can be seen from this definition that ∅ is an ordinal, the succes-
sor of any ordinal is an ordinal, and any transitive set of ordinals is an
ordinal.
It is obvious that every natural number is an ordinal (use mathematical
induction). Also, the set ω of all natural numbers is transitive (verify!), and
so ω is an ordinal. Once we have ω, we have the ordinal ω + , also denoted
ω + 1. Successively, we have ω + 2 (ω ++ ), ω + 3, . . . , ω + n, . . . . However, we
do not yet have a set which contains all these ordinals, and the existence
of such a set cannot be derived from Zermelo’s axioms. But then Abraham
Fraenkel [1958] added another axiom known as the axiom of substitution,
also called the axiom of replacement, which does ensure the existence of
such a set, and this set is denoted ω × 2 or ω · 2.
Roughly speaking, the axiom of replacement is that given any operation
F which assigns to each set x a set F (x), and given any set a, there is a set
[denoted F −1 (a)] consisting of all, and only, those sets x such that F (x) ∈ a.
Well, there is the set denoted ω · 2 of all elements ω + n, where n is any
natural number. Then we have the ordinals
ω · 2 + 1, ω · 2 + 2, ω · 2 + 3, . . . , ω · 2 + n, . . . , ω · 3.
 Where to Go From Here 257

After that finally come ω · 4, ω · 5, . . . , ω · ω, . . . ω ω , . . .. We can keep

going.
The ordinals so far considered are all denumerable. But there exist
non-denumerable ordinals as well. Indeed, with the axiom of substitu-
tion/replacement (of Zermelo–Fraenkel set theory), for every set x, there is
an ordinal α of the same size as x (i.e. α can be put into 1-1 correspondence
with x).
An ordinal is called a successor ordinal if it is α+ [α ∪ {α}] for some
ordinal α. Ordinals other than 0 and successor ordinals are called limit
ordinals. All natural numbers other than 0 are successor ordinals. The first
limit ordinal is ω. The ordinals ω + 1, ω + 2, . . . , ω + n are all successor
ordinals, and ω · 2 (the set of all of them) is the next limit ordinal. The
ordinals ω · 3, ω · 4, . . . , ω · n are all limit ordinals, and so is ω · ω. Actually,
an infinite ordinal is a successor ordinal if and only if it is of the form α + n,
for some limit ordinal α and some positive integer n.
An ordinal α is said to be less than an ordinal β (in symbols, α < β) if
α ∈ β. Thus each ordinal is the collection of all lesser ordinals.

Rank

Given a scheme that assigns to each ordinal α a set denoted Sα , for any
ordinal λ, by Uα<λ Sα is meant the union of all the sets Sα , where α < λ.
By an important result known as the transfinite recursion theorem, one can
assign to each ordinal α a set Rα such that for every ordinal α and every
limit ordinal λ, the following three conditions hold:
(1) R0 = ∅.
(2) Rα+ = P(Rα ) [the set of all subsets of Rα ].
(3) Rλ = Uα<λ Rα .
A set is said to have rank if it is a member of some Rα , and by the rank
of such a set x is meant the least ordinal α such that x ∈ Rα .
Do all sets have rank? This is so iff the ∈ relation is well founded, i.e.
iff there exists no infinite sequence x1 , x2 , . . . xn , . . . such that for each n,
the set xn+1 is a member of xn . This condition is taken as another axiom
of set theory, and is known as the axiom of foundation. It rules out sets
without rank.
258 Afterword

The Axiom of Choice and the Generalized

Continuum Hypothesis

The Axiom of Choice (frequently abbreviated AC) in one form is that

for any non-empty set a of non-empty sets, there is a function C (a so-
called choice function for a) that assigns to each member x of a an element
of x. [The function, so to speak, chooses one element from each of the
elements of a.]
An equivalent form of AC is that for any non-empty set a of non-empty
sets, there is a set b whose members consist of just one member of each of
the members of a.
What is the status of AC? Most working mathematicians of the world
accept it as being true. There are a few, though, who don’t. What is its
status with respect to ZF (Zermelo–Fraenkel Set Theory)? Well, Gödel
[1940] showed that it is consistent with ZF (assuming that ZF is itself
consistent, which we will continue to assume). Thus AC is not refutable in
ZF. Sometime later, Paul Cohen [1963, 1964] proved that the negation of
AC is consistent with ZF. Thus AC is undecidable in ZF.
We recall Cantor’s theorem, that for any set a, its power set P(a) is
larger than a. Cantor conjectured that for an infinite set there is no set
b intermediate in size between a and P(a) [that is, that there is no set b
which is larger than a but smaller than P(a)], and this result is known as
the Generalized Continuum Hypothesis (frequently abbreviated as GCH).
What is its status? This is far more puzzling! Unlike the case of the axiom
of choice, which most logicians believe is true, most mathematicians, as well
as most logicians specializing in set theory, don’t have the slightest idea as
to whether GCH is true or false. Gödel conjectured that it is false, despite
the fact that he proved it consistent with the axioms of ZF. His proof of the
consistency of GCH with ZF is surely one of the most remarkable things
in mathematical logic! Equally remarkable is the proof of Paul Cohen that
the negation of GCH is consistent with ZF, even when AC is added to the
axioms of ZF. Thus GCH is undecidable in ZF.
Gödel’s proof of the consistency of AC and GCH used his notion of
constructible sets, to which we now turn.
 Where to Go From Here 259

Constructible Sets

Consider a sentence X whose constants are all in a set a. X is said to be

true over a if it is true when the quantifiers are interpreted as ranging over
all elements of a, i.e. it is true when “∀x” is read as “for all x in a” and
“∃x” is read as “for some x in a”. A formula ϕ(x) whose constants are
all in a is said to define over a the set of all elements k such that ϕ(k) is
true over a, and a subset b of a is called definable over a (more completely,
first-order definable over a) if it is defined over a by some formula whose
constants are all in a.
We let F(a) be the set of all sets definable over a.
Gödel introduced the notion of constructible sets as follows: Again by
the transfinite recursion theorem, one can assign to each ordinal α a set
Mα such that for every ordinal α and every limit ordinal λ, the following
three conditions hold:
(1) M0 = ∅.
(2) Mα+ = F (Mα ).
(3) Mλ = Uα<λ Mα .

The definition of the Mα ’s differs from that of the Rα ’s only in condition

(2): Whereas Rα+1 consists of all subsets of Rα , the set Mα+1 consists of
only those subsets of Mα that are definable over Mα .
A set is called constructible if it is a member of some Mα .
Are all sets constructible? This is a grand unsolved problem! Gödel con-
jectured that this is not the case, even though he proved that it was consis-
tent with the axioms of ZF. The significance of the constructible sets is that
the constructability of all sets implies the generalized continuum hypothesis!
Moreover, this implication is provable in ZF. Thus if we add to the axioms of
ZF the axiom that all sets are constructible (which is known as the axiom of
constructability), the continuum hypothesis (as well as the axiom of choice)
becomes provable. And since, as Gödel showed, the axiom of constructabil-
ity is consistent with ZF, so is the generalized continuum hypothesis (as well
as the axiom of choice). This is how Gödel showed the consistency of GCH
with ZF.
260 Afterword

Some years later, Paul Cohen [1963, 1964] showed, by a completely

different method, that the negation of GCH is consistent with ZF, even
with the addition of AC. Thus GCH is undecidable in ZF.
Despite the undecidability of GCH in ZF, the question remains as to
whether or not it is really true. To many so-called “formalists”, the question
has no meaning. They say that is provable in some axiom systems and
disprovable in others. To the so-called “Platonist” like Gödel, this position
is most unsatisfactory. I would liken it to the following: Suppose a bridge
is built and on the next day an army is to march over it. Will the bridge
hold or not? It does no good to say, “Well, in some axiom systems, it can
be proved that the bridge will hold, and in others, it can be proved that it
won’t.” We want to know whether it will really hold or not. And likewise
with the generalized continuum hypothesis. It is not enough to say it is
undecidable in ZF. The fact remains that for every infinite set a, either
there is a set b intermediate in size between a and P(a) or there isn’t, and
we want to know which. As already, indicated, Gödel conjectured that there
is such a set b for some infinite set a, and that when more is known about
sets, we will see that Cantor’s conjecture is wrong. [Realize that for any
non-empty finite set a, there are always sets b with more elements than a
and less elements than P(a).]
The whole study of the independence of GCH in ZF is fascinating
beyond belief, and the reader who studies this is in for a real treat!
 References

Barendregt, Henk, The Lambda Calculus — Its Syntax and Semantics, Studies in
Logic and the Foundations of Mathematics Vol. 103, North-Holland, 1985.
Beth, Evert, The Foundations of Mathematics. A Study in the Philosophy of
Science, North-Holland, 1959.
Braithwaite, Reginald, Kestrels, Quirky Birds and Hopeless Egocentricity, ebook
published by http://leanpub.com, 2013.
Church, Alonzo, The Calculi of Lambda Conversion, Princeton Univ. Press, 1941.
Cohen, Paul J., “The independence of the continuum hypothesis,” Proceedings of
the National Academy of Sciences of the United States of America 50(6):
1143–1148, 1963.
Cohen, Paul J., “The independence of the continuum hypothesis, II,” Proceedings
of the National Academy of Sciences of the United States of America 51(1):
105–110, 1964.
Craig, William, “Three uses of the Herbrand–Gentzen theorem in relating model
theory and proof theory,” The Journal of Symbolic Logic 22(3): 269–285,
1957.
Davis, Martin, Computability and Unsolvability, McGraw-Hill, 1958; Dover, 1982.
Ehrenfeucht, Andrzej and Feferman, Solomon, “Representability of recursively
enumerable sets in formal theories,” Arch. Fur Math. Logick und Grundla-
genforschung 5: 37–41, 1960.
Fraenkel, Abraham Bar-Hillel, Yehoshua and Levy Azriel, Foundations of Set
Theory, North-Holland, 1973 (originally published in 1958). (Fraenkel’s final
word on ZF and ZFC, according to Wikipedia.)
Frege, Gottlob, Grundgesetze der Arithmetic, Verlag Hermann Pohle, Vol. I/II,
1893. Partial translation of volume I, The Basic Laws of Arithmetic, by
M. Furth, Univ. of California Press, 1964.
Gentzen, Gerhard, “Untersuchungen über das logische Schliessen I,” Mathema-
tische Zeitschrift 39(2): 176–210, 1934.
Gentzen, Gerhard, “Untersuchungen über das logische Schliessen II,” Mathema-
tische Zeitschrift 39(3): 405–431, 1935.
262 References

Gödel, Kurt, The Consistency of the Axiom of Choice and of the Generalized
Continuum Hypothesis with the Axioms of Set Theory, Princeton Univ.
Press, 1940.
Henkin, Leon, “The completeness of the first-order functional calculus,” J. Sym-
bolic Logic, 14: 159–166, 1949.
Hintikka, Jaakko, “Form and content in quantification theory,” Acta Philosophica
Fennica, 8: 7–55. 1955.
Kleene, Stephen Cole, “Recursive predicates and quantifiers,” Trans. Amer. Math.
Soc. 53: 41–73, 1943.
Kleene, Stephen Cole, Introduction to Metamathematics, North-Holland, 1952.
Myhill, John, “Creative sets,” Z. Math. Logik Grundlagen Math. 1: 97–108, 1955.
Post, Emil Leon, “Formal reductions of the general combinatorial decision prob-
lem,” American Journal of Mathematics 65: 197–215, 1943.
Post, Emil Leon, “Recursively enumerable sets of positive integers and their deci-
sion problems,” Bull. Amer. Math. Soc. 50(5): 284–316, 1944.
Putnam, Hilary and Smullyan, Raymond, “Exact separation of recursively
enumerable sets within theories,” Journal of the American Mathematical
Society, 11(4): 574–577, 1960.
Rice, H. Gordon, “Classes of recursively enumerable sets and their decision prob-
lems,” Trans. Amer. Math. Soc. 74(2): 358–366, 1953.
Rosser, John Barkley, “Extensions of some theorems of Gödel and Church,” Jour-
nal of Symbolic Logic, 1(3): 87–91, 1936.
Russell, Bertrand, “Letter to Frege,” 1902. This very interesting letter can be
found (translated from the German) in Van Heigenoort, Jean, From Frege
to Gödel, A Source Book in Mathematical Logic, 1879–1931, Harvard Univ.
Press, 1967. It is also available online at a Harvard website: http://isites.
harvard.edu/fs/docs/icb.topic1219929.files/FregeRussellCorr.pdf
Schönfinkel, Moses, “Über die Bausteine der mathematischen Logik”, Mathema-
tische Annalen 92, 1924; translated as “On the building blocks of mathe-
matical logic” and included in Van Heigenoort, Jean, From Frege to Gödel,
A Source Book in Mathematical Logic, 1879–1931, Harvard Univ. Press,
1967.
Shepherdson, John, “Representability of recursively enumerable sets in formal
theories,” Archiv für Mathematische Logik und Grundlagenforschung, 119–
127, 1961.
Smullyan, Raymond, Theory of Formal Systems, Princeton Univ. Press, 1961.
Smullyan, Raymond, “A unifying principle in quantification theory”, Proceedings
of the National Academy of Sciences, 49(6): 828–832, 1963.
Smullyan, Raymond, To Mock a Mockingbird, Alfred A. Knopf, 1985.
Smullyan, Raymond, Gödel’s Incompleteness Theorems, Oxford Univ. Press, 1992.
Smullyan, Raymond, Recursion Theory for Metamathematics, Oxford, 1993.
Smullyan, Raymond, Diagonalization and Self-Reference, Oxford Science
Publications, 1994.
Smullyan, Raymond, First-Order Logic, Springer-Verlag, 1968; Dover, 1995.
Smullyan, Raymond, Logical Labyrinths, CRC Press, 2008; A. K. Peters, Ltd.
2009.
 References 263

Smullyan, Raymond, The Beginner’s Guide to Mathematical Logic, Dover, 2014.

Sprenger, M. and Wymann-Böni, M., “How to decide the lark,” Theoretical Com-
puter Science, 110: 419–432, 1993.
Statman, Richard, “The word problem for Smullyan’s lark combinator is decid-
able,” Journal of Symbolic Computation, 7(2): 103–112, 1989.
Tarski, Alfred, Undecidable Theories, North-Holland, 1953.
Von Neumann, John, “On the introduction of transfinite numbers,” in Jean van
Heijenoort, From Frege to Gödel: A Source Book in Mathematical Logic,
1879–1931 (3rd ed.), Harvard Univ. Press, 1923, pp. 346–354 (English
translation of von Neumann 1923), 1973.
Whitehead, Alfred North and Russell, Bertrand, Principia Mathematica,
Cambridge Univ. Press, 1910, Vol. 1. Reprinted by Rough Draft Printing,
2011.
Zermelo, Ernst, “Untersuchungen über die Grundlagen der Mengenlehre I,”
Mathematische Annalen 65(2): 261–281, 1908. English translation: Hei-
jenoort, Jean van (1967), “Investigations in the foundations of set theory,”
From Frege to Gödel: A Source Book in Mathematical Logic, 1879–1931,
Source Books in the History of the Sciences, Harvard Univ. Press, 1967,
pp. 199–215.
b2530   International Strategic Relations and China’s National Security: World at the Crossroads

This page intentionally left blank

 Index

1-1 (one-to-one) function, 97 admissible function in a simple

δ(x, y), recursive pairing function, 97 system, 152
δn (x1 , . . . , xn ), recursive n-tupling affirmation of a number by a register,
function, 98 70
Γ-consistent set of sentences, 49–56 affirmation set of a register, 73
λ-I combinators, 192–196 agreeable element, in combinatory
ωn , 113 logic, 175
Φ, Φ2 , Φ3 , Φ4 combinators, 195, 196 algebraic approach to verifying
Φn combinator, 196 tautologies, 4–7
alphabet of an elementary formal
σ, successor combinator, 234
system, 89, 90
Σ0 formula of elementary arithmetic,
altered tableau method, 40–43
94, 95
analytic consistency property,
Σ0 set or relation, 95
49–56
Σ1 formula of elementary arithmetic, analytic tableau, 34–36, 40, 49–56
95
applications of magic sets, 27, 28
Σ1 set or relation, 93, 95 applications of Rice’s Theorem, 120
applications of simple systems, 150
∼AC is consistent with ZF (Cohen), arithmetic combinators, 234, 235
258 arithmetic set or relation, 95
A∗ , C ∗ , R∗ , F ∗ and V ∗ combinators, arithmetic term of combinatory logic,
189 241
A∗ number set in a simple system, arithmetical combinator of type n,
151 234
A∗∗ , C ∗∗ , R∗∗ , F ∗∗ and V ∗∗ associate regular set for a finite set of
combinators, 189 sentences, 53
abstraction principle, 253 atomic formula of an elementary
AC (Axiom of Choice), 258–260 formal system, 89, 90
AC is consistent with ZF (Gödel), 258 Axiom of Choice (AC), 258–260
AC is undecidable in ZF (Gödel and axiom of constructability, 259
Cohen), 258 axiom of foundation, 257
addition combinator, 234, 235 axiom of infinity, 254, 256
266 Index

axiom of replacement, 256 cancellator (kestrel) combinator K,

axiom of substitution, 256 176
axiomatic set theory, 253–260 Cantor’s theorem, 258
axioms (and axiom schemes) of an cardinal combinator C, 187
elementary formal system, 90 Church, Alonzo, 187, 188, 192, 234
axioms and rules of inference for closure properties on representable
combinatory logic, 239, 243 relations in elementary formal
axioms of set theory, 253–256, systems, 91, 92
258–260 closure under existential
quantification of representable
B combinator (bluebird), 185 relations in elementary formal
B, M, T , and I are a basis for all λ-I systems, 92
combinators, 224–226 closure under explicit transformation
B, M , T and I derivable combinators of representable relations in
(λ-I combinators), 192–196 elementary formal systems, 92
B, T and I combinators, 226, 227 closure under unions and intersections
B1 , B2 , Bn , and B + combinators, 186 of representable relations in
Barendregt, Henk, 233, 234 elementary formal systems, 92
base in the universal system (U ), 100 co-productive function for a set, 122
basis for a class of combinators, 224 co-productive set, 122
Beth’s Definability Theorem, 46–48 Cohen, Paul, 258, 260
Beth, Evert, 34, 46, 48, 56 combinator birds table on Internet,
block tableau, 34–39 with derivations from S and K,
bluebird combinator B, 185 222, 223
Boolean consequence relation, 11–15 combinator B (bluebird), 185
Boolean equations, verification by combinator C (cardinal), 187
truth tables, 3, 4 combinator D (dove), 186
Boolean ring, 4, 7 ∧
Boolean truth set for first-order logic, combinator E , 186
25 combinator E (eagle), 186
Boolean truth set for propositional combinator F (finch), 188
logic, 7 combinator for addition, 234
Boolean valuation for first-order combinator for exponentiation, 235
logic, 23 combinator for multiplication, 234
Boolean valuation for propositional combinator G (goldfinch), 191
logic, 9 combinator H (hummingbird), 193
bounded quantifiers, 94 combinator I (identity), 179
Braithwaite, Reginald, 177 combinator J (Rosser), 226
combinator K (kestrel, cancellator),
C and K (Post sets), 122 176
C combinator (cardinal), 187 combinator L (lark), 178
C1 , condition that a simple system combinator M (mockingbird, mocker,
must satisfy, 149 duplicator), 174
cancellation law, in combinatory combinator M2 , 192
logic, 177 combinator n, 208
cancellative combinator, 216 combinator N (nice combinator), 208
 Index 267

combinator O (oracle, owl), 207 compact property of sets, 12

combinator Φn , 196 Compactness Theorem
combinator Q (queer bird), 190 (denumerable), 12, 49, 52
combinator Q1 (quixotic bird), 190 compatible pair of elements, in
combinator Q2 (quizzied bird), 190 combinatory logic, 175
combinator Q3 (quirky bird), 191 complete infinite regular set of
combinator Q4 (quacky bird), 191 sentences, 26
combinator Q5 (quintessential bird), complete mathematical system S, 74
191 complete representation of a number
combinator Q6 (quivering bird), 191 set by a number in a simple
combinator R (robin), 187 system, 151
combinator S (starling), 194 complete set K, 122
combinator σ (successor), 234 complete simple system, 150
combinator T (thrush), 187 complete system for combinatory
combinator U (Turing combinator), logic, 215–223
206 completely creative function for a set,
combinator V (vireo), 188 123
combinator W (warbler), 192 completely creative set, 122
combinator W 0 (converse warbler), completely effective inseparability of
192 two sets, 137, 139
combinatorial completeness of a completely effectively inseparable pair
system of combinatory logic, of sets, 137–139
215–223 completely productive function for a
combinatorially complete class of set, 122
combinators, 215–223 completely productive set, 122
combinatorially computable number completeness of a first-order axiom
set, 236 system, 49, 52
combinatorics, 171–251 completeness of the altered tableau
combinators, 173 method, 43
combinators A∗ , C ∗ , R∗ , F ∗ and V ∗ , completeness of the analytic tableau
189 method, 49, 51
combinators A∗∗ , C ∗∗ , R∗∗ , F ∗∗ and completeness of the block tableau
V ∗∗ , 189 method for propositional logic, 36,
combinators B, T and I, 226, 227 37
combinators B1 , B2 , Bn , and B + , 186 completeness of the Gentzen axiom
combinators D1 and D2 , 186 system for first-order logic G1 , 39
combinators Φ, Φ2 , Φ3 , Φ4 , 195, 196 completeness of the Gentzen system
combinators P, P1 , P2 , P3 , 195 G0 , 34–38
combinators W1 , W2 , W2 , 193 completeness of the (symmetric)
combinators derivable from B, M , T Gentzen system for first-order logic
and I (λ-I combinators), 192–196 GG, 40–43
combinators derived from the B completeness proof for propositional
combinator alone, 186 logic based on maximal
combinators for t and f and for the consistency, 7–15
logical connectives, 233, 234 components of a base in the universal
combinatory logic, 171–251 system (U ), 100
268 Index

composition condition for a creative function for a set, 122

combinatory logic, 174 creative number in a synthesis
compositor combinator, 186 system, 81
computation of a number set by a creative predicate in a mathematical
property combinator, 236 system S, 75
concatenation in base 10 combinator, creative R-set, 77
238 creative register, 71
consequence relation, 10–15 creative set, 122
consistency of the Axiom of Choice critical parameter of a regular set,
(AC) with ZF (shown by Gödel), 52
258 cross point of a pair of compatible
consistency of the Generalized elements, in combinatory logic, 175
Continuum Hypothesis (GCH) with Curry’s sage, 206
ZF (shown by Gödel), 258, 259
consistency of the negation of the D combinator (dove), 186
Axiom of Choice (AC) with ZF D1 and D2 combinators, 186
(shown by Cohen), 258 Davis, Martin, 113
consistency of the negation of the decidable number in a synthesis
Generalized Continuum Hypothesis system, 80
(GCH) with ZF (shown by Cohen), decidable number of a simple system,
258 150
consistent/inconsistent set of decidable sentence in a mathematical
formulas, 10 system S, 73
consistent mathematical system S, decidable simple system, 155
73 decision (calculating) machine, 69–73
constant term of an elementary deduction in an axiom system, 13
formal system, 89 deduction property of an axiom
constructible set, 258, 259 system, 13
construction of the universal system Deduction Theorem, 13, 14
(U ), 99–101 definable function, 160
continuum hypothesis, 258–260 defining equation (defining condition)
contra-representation of a number set of an element of a combinatory
by a number in a simple system, system, 215
151 definition of a number set by a
contra-universal number in a number in a simple system, 151
synthesis system, 81 degree of a predicate of an
contra-universal R-set, 77 elementary formal system, 89
contra-universal register, 71 denial of a number by a register, 70
converse warbler combinator W 0 , 192 derivatives of the combinator B
correctness of the altered tableau alone, 186
method, 43 designation of a number by an S, K
correctness of the Gentzen system G0 term, 241
for propositional logic, 32, 33 diagonal function d(x) in a simple
Craig’s Interpolation Lemma (Craig’s system, 152
Lemma), 43–46, 49, 53 diagonalization of a number in a
Craig, William, 29, 31, 43–46, 49, 53 simple system, 151
 Index 269

diagonalizer H # of a sentence H in a effective method of finding a set,

mathematical system S, 74 115
diagonalizer R# of a register R, 70 effective Rosser system (as a simple
domination of one number by another system), 158, 159, 161
in a synthesis system, 83 egocentric element, in combinatory
domination of one predicate by logic, 174
another in a mathematical system Ehrenfeucht, Andrzej, 157, 160
S, 76 elementary formal systems and
domination of one R-set by another, recursive enumerability, 89–103
78 enumeration and iteration theorems,
domination of one register over 113–117
another, 72 Enumeration Theorem, 113, 114
double diagonal function, 133, 134 equivalence of Craig’s Interpolation
double fixed point of a pair of Lemma for first-order logic and for
elements, 207, 208 the Gentzen system GG, 44
double oracle, 209 equivalence of provability and truth
double recursion theorems, 133–135 in elementary formal systems, 90
double sage (double sage pair), 207, equivalence of representability in an
208 arbitrary dyadic system and
doubly co-productive ordered pair of representability in a transcribed
sets, 135, 140 dyadic system, 100
equivalence of valuations and
doubly generative ordered pair of
interpretations, 9, 10, 27
sets, 135, 138–141
equivalent numbers of a simple
doubly universal ordered pair of sets,
system, 150
135, 139–141
equivalent sentences in a
dove combinator D, 186
mathematical system S, 74
duplicator (mockingbird, mocker)
exact Rosser (first-order) system for
combinator M , 174
sets, 160
dyadic (elementary formal) system,
exact Rosser system (as a simple
93
system), 157–161
dyadic numeral, 93
exact separation of a disjoint pair of
dyadically representable set or
number sets in a simple system,
relation, 93
153
∧ explicit definition of a predicate with
E combinator, 186 respect to a set of sentences, 46, 47
E combinator (eagle), 186 explicit definition of a relation from
E-complete set of sentences, 54–56 another relation in an elementary
eagle combinator E, 186 formal system, 92
effective (and recursively enumerable) explicit transformation of a relation
sequence of simple systems, each of into another relation in an
which is an extension of the elementary formal system, 92
preceding system, 159 exponential combinator, 235
effective inseparability of two sets, extension of a simple system, 159
136–139, 141 extensional number set, 119
270 Index

F combinator (finch), 188 G combinator (goldfinch), 191

Feferman, Solomon, 157, 160 G0 , 31–38
fidelity (faithfulness) to modus G0 in uniform notation, 33
ponens, 13 G1 , 38, 39
finch combinator F , 188 GCH (Generalized Continuum
finitary proof, 30 Hypothesis), 258–260
finite character property, 54–56 GCH is consistent with ZF (Gödel),
finite existential quantification of a 258, 259
relation R, 94 GCH is undecidable in Z (Gödel and
finite quantifications of relations, 94 Cohen), 258
finite universal quantification of a Generalized Continuum Hypothesis
relation R, 94 (GCH), 258–260
First Double Fixed Point Theorem generative function for a set, 123
for combinators, 224 generative set, 123
First Fixed Point Theorem for Gentzen axiom system (symmetric)
combinators, 223 for first-order logic GG, 39–46
first-order logic, 23–66 Gentzen axiom system for first-order
first-order satisfiable sentence, 24 logic G1 , 38, 39
first-order truth set, 25 Gentzen axiom system for
first-order valuation, 23 propositional logic G0 , 31–38
fixation of one element on another, in Gentzen sequent, 29–46
combinatory logic, 175 Gentzen type axiom systems, 31–46
fixed point combinator, 179, 205–208 Gentzen, Gerhard, 29–46, 56
fixed point of a number in a simple GG, 39–46
system, 153 Gödel, Kurt, 73–76, 96, 102, 103, 156,
fixed point of a number in a synthesis 240–242, 258, 259
system, 81 Gödel numbering using dyadic
fixed point of a pair of R-sets, 77 numerals, 96
fixed point of a predicate in a Gödel numeral of an S, K term of
mathematical system S, 75 combinatory logic, 241
fixed point of a register, 72 Gödel sentence (equation of
fixed point of an element, in combinatorial logic) of a number
combinatory logic, 174 set, 242
fixed point theorems (recursion Gödel sentence of the universal
theorems) in recursion theory, system U for a number set A, 102
117–121 goldfinch combinator G, 191
formally representable set or relation, grand problem: Is there a systematic
91 way to decide whether two S, K
formula of an elementary formal combinators are equal?, 239–243
system, 89, 90
Fraenkel, Abraham, 254, 256–260 H combinator (hummingbird), 193
Frege, Gotttlob, 253, 254 H 0 , negation of a sentence H in a
functional combinator, 237 mathematical system S, 74
H # , diagonalizer of a sentence H in a
∼GCH is consistent with ZF mathematical system S, 74
(Cohen), 258, 260 halting of a decision machine, 72
 Index 271

Henkin, Leon, 54–56 interpretation of first-order sentences

Henkin-style completeness proof for in the domain of the parameters,
first-order logic, 54–56 24
Hintikka set, 50, 51, 55, 56 inverse functions K and L, 98
Hintikka, Jaakko, 34, 50, 51, 55, 56 inverse functions Kin , 99
hummingbird combinator H, 193 inverse of a binary relation, 115
iteration and enumeration theorems,
I combinator (identity), 179 113–117
identity combinator I, 179 Iteration Theorem, 116
implication sign of an elementary iterative function for a relation
formal system, 89 R(x, y), 124
implicit definition of a predicate with
respect to a set of sentences, 46, 47 J and I are a basis for all λ-I
combinators, 226
incomplete simple system, 150, 155
J combinator (Rosser), 226
inconsistency of Frege’s system of set
theory, 254
K combinator (kestrel, cancellator),
independent elementary formal
176
systems, 91
kestrel (cancellator) combinator K,
index of a predicate in a
176
mathematical system S, 73
Kestrels, Quirky Birds and Hopeless
index of a recursively enumerable set, Egocentricity, 177
114
Kleene function for a disjoint pair of
index of a register, 69 sets, 137, 138
index of a sentence in a mathematical Kleene pair of disjoint sets, 137–139
system S, 73 Kleene’s construction of a completely
index of a sentence in a simple effectively inseparable pair of
system, 150 recursively enumerable sets, 137,
indexing method for verification of 138
tautologies by way of Boolean Kleene’s symmetric form of Gödel’s
equations, 3, 4 Theorem, 156
inductive set, 255 Kleene, Stephen, 113, 137–139, 156,
instance of a formula of an 157
elementary formal system, 90
interpolant for a first-order sentence L combinator (lark), 178
X ⊃ Y , 43 lark combinator L, 178
interpolant for a first-order sequent Lemma K, 96
X → Y , 44 length measurer combinator, for a
interpolant for a propositional base 10 number, 238
sequent X → Y , 44 less than relation on ordinals,
Interpolation Lemma (Craig’s 257
Interpolation Lemma), 43–46, 49, limit ordinal, 257
53 limited abstraction principle, 254
interpretation, 8–11, 24, 27, 28, 46, Lindenbaum’s method of extending a
47 consistent set S to a maximally
272 Index

consistent set M when consistency n combinator, 208

is of finite character, 54–56 N combinator (nice combinator),
Lindenbaum, Adolf, 15, 54–56 208
logical (propositional) combinators, n-tupling function δn (x1 , . . . , xn ), 98
233, 234 narcissistic element, in combinatory
logical connectives and Boolean logic, 176
equations on sets, correspondence natural number (numerical)
between, 3, 4 combinators, 234
neg(x), requisite function of a simple
M combinator (mockingbird, mocker, system, 149
duplicator), 174 negated form of an iff statement, 7, 8
M2 combinator, 192 negation H 0 of a sentence H in a
magic set, 23–28 mathematical system S, 74
many-one reducible set, 123 negation of Generalized Continuum
many-one reduction of one set to Hypothesis (GCH) is consistent
another, 123 with ZF (Cohen), 258, 260
mathematical system S (variations on nice combinator N , 208
a theme of Gödel), 73–76 non-cancellative (λ-I) combinator,
maximal indexing (recursively 216
enumerable) of all recursively non-denumerable ordinal, 257
enumerable sets, 117
non-standard formula, 44
maximally consistent set of formulas,
norm of a string of symbols of the
11, 12
alphabet of the universal system U ,
mechanical procedure, 90
102
metamathematical applications,
norm of an S, K term, 241
149–161
normal simple system, 153
method of eliminates required to
normalizer of a natural number
obtain a combinator derived from
combinator, 241
S and K, 219
method of eliminates required to numerical (natural number)
obtain a combinator derived from combinators, 234, 235
S, B, C and I, 225 numerical relation A(x, y) in a
minimization principle, 237, synthesis system, 79
238 numerical relation B(x, y) in a
minimizer of a regular relational synthesis system, 79
combinator, 238
mocker (mockingbird, duplicator) O combinator (oracle owl), 207
combinator M , 174 one-to-one (1-1) function, 97
mockingbird (mocker, duplicator) onto function, 97
combinator M , 174 opposer R0 of a register, 70
modus ponens, 7 oracle (owl) combinator O, 207
multiplication combinator, 234, 235 order of a combinator, 194
Myhill’s Fixed Point Theorem, 120, order of elimination of variables in
121 the method of eliminates, 221
Myhill, John, 120, 121, 123, 125–127, ordinal, 255–257
134 owl (oracle) combinator O, 207
 Index 273

P, P1 , P2 , P3 combinators, 195 proof of Craig’s Interpolation Lemma

(P, R), pair of disjoint number sets, for the Gentzen first-order system
making up a simple system, 149 GG, 44–46
parentheses conventions for logical proof of the Compactness Theorem
formulas including set by use of the Unification Theorem,
inclusion/exclusion, 7 52
partial system for λ-I proof of the Compactness Theorem
(non-cancellative) combinators, for first-order logic using magic
224–226 sets, 28
partial systems for combinatory logic,
proof of the completeness of an axiom
224–227
system for first-order logic by use
partition interpolant between two
of the Unification Theorem, 52
sets, 53
proof of the completeness of
partition of a set, 53
propositional logic based on
Peano Postulates, 255
maximal consistency, 7–15
permuting combinator (permuter),
187–191 proof of the completeness of standard
Platonists versus formalists on the axiom systems for first-order logic
undecidability of the Generalized by Henkin’s method, 54–56
Continuum Hypothesis, 260 proof of the completeness of the
positive integers, as defined in set method of analytic tableaux by use
theory, 254, 255 of the Unification Theorem, 51
Post’s sets C and K, 122 proof of the efficacy of the method of
Post, Emil, 113, 122–125, 244 eliminates required to obtain a
predecessor combinator, 235 combinator derived from S and K,
predicate of a mathematical system 220, 221
S, 73 proof of the efficacy of the method of
predicate of an elementary formal eliminates required to obtain a
system, 89 combinator derived from S, B, C
predicate of the universal system U , and I, 225, 226
100 proof of the Regularity Theorem by
principal part (P, R) of a simple use of the Unification Theorem, 52,
system, 159 53
Principia Mathematica, 254
proof of the Regularity Theorem
productive function for a set, 121,
using magic sets, 28
122
proof of the Skolem–Löwenheim
productive set, 121
Theorem by use of the Unification
proof of a sentence of an elementary
Theorem, 51
formal system, 90
proof of a sequent in Gentzen type proof of the Skolem–Löwenheim
system, 34 Theorem using magic sets, 28
proof of Beth’s Definability Theorem proof of the Unification Theorem by
by using Craig’s Lemma, 47, 48 means other than tableaux or
proof of Craig’s Interpolation Lemma König’s Lemma, 50, 51
by use of the Unification Theorem, proof of the Unification Theorem
53 using tableaux, 50
274 Index

proof that the set of Gödel numbers R combinator (robin), 187

of the true sentences of U is not r(h, n), in the simple system
recursive, 102, 103 associated with a first-order system
proof that the set of true sentences T of arithmetic, 150
of the universal system U is r(x, y), for recursion theory, 113
formally representable and that the r(x, y), requisite (representation)
set T0 is recursively enumerable, function of a simple system, 149,
101, 102 150
property combinator, 236 R-separable pair of R-sets, 78
propositional logic, 3–22, 35–38 R-set, 76
propositional logic combinators, 233, R-system, 76–78
234
R0 , opposer of a register R, 70
provability predicate in a
R# , diagonalizer of a register R, 70
mathematical system S, 74
rank of a set, 257
provable number of a simple system,
150 realization of a function by a
combinator, 237
provable sentence of a mathematical
system S, 73 realization of a term by a combinator,
provable sentence of an elementary 216
formal system, 90 Recursion Property for Combinatory
punctuation sign of an elementary Logic, 236
formal system, 89 recursion theorems (fixed point
pure sentence of first-order logic, 25 theorems) in recursion theory,
Putnam, Hillary, 157, 160 117–121
Putnam–Smullyan Theorem, 160 recursion theory, 67–170
recursive inseparability of two number
Q combinator (queer bird), 190 sets in a simple system, 156
Q1 combinator (quixotic bird), 190 recursive inseparability of two sets,
Q2 combinator (quizzied bird), 136, 156
190 recursive isomorphism between two
Q3 combinator (quirky bird), 191 sets, 127
Q4 combinator (quacky bird), 191 recursive pairing function δ(x, y), 97
Q5 combinator (quintessential bird), recursive set, 93
191 recursive unsolvability of the
Q6 combinator (quivering bird), universal system U , 102, 103
191 recursively enumerable (effective)
quacky bird combinator Q4 , 191 sequence of simple systems, each of
queer bird combinator Q, 190 which is an extension of the
quintessential bird combinator Q5 , preceding system, 159
191 recursively enumerable indexing of all
quirky bird combinator Q3 , 191 recursively enumerable sets, 117
quivering bird combinator Q6 , recursively enumerable set or
191 relation, 93–103
quixotic bird combinator Q1 , 190 recursively separable pair of sets, 136,
quizzied bird combinator Q2 , 190 156
 Index 275

reducibility of one ordered pair of sets Rosser system (as a simple system),
to another ordered pair of sets, 135, 157–161
139 Rosser (first-order) system for binary
reducible set, 123 relations, 160
reduction of one set to another, 123 Rosser (first-order) system for sets,
refutability predicate in a 159, 160
mathematical system S, 74 Rosser fixed point property in a
refutable number of a simple system, simple Rossor system, 158
150 Rosser function in an effective Rosser
refutable sentence of a mathematical (simple) system, 158
system S, 73 Rosser system (as a simple system),
register of a decision (calculating) 157–161
machine, 69 Rosser, J. Barkley, 75, 152, 157–161,
regular relational combinator, 237, 192, 226
238 Russell, Bertrand, 254
regular set of sentences of first-order
logic, 26 S combinator (starling), 194
Regularity Theorem, 23, 28, 49, 52,
s(n, x, y), in the simple system
53
associated with a first-order system
relational combinator, 237 of arithmetic, 150
representability in a transcribed
S and K as constants from which all
dyadic system, 100
other combinators can be derived,
representability in an arbitrary
215–223
dyadic system, 93
S, B, C, and I are a basis for all λ-I
representability in the universal
combinators, 224–226
system U , 100
S, K term, 239
representation function r(x, y) for
number sets with respect to a sage constructions, 205–208
simple system, 149, 150 sage element, in a combinatory logic,
representation of a natural number 179
by a combinator, 234 satisfaction of a set of sentences by a
representation of a number set by a valuation, 25
combinator, 242 satisfiability of first-order sentences in
representation of a set by a number the domain of the parameters, 24
in a simple system, 150 Schöfinkel, Moses, 173
representation of a set by a number Second Fixed Point Theorem, for
in a synthesis system, 83 combinators, 242
representation of a set in a semi-doubly universal ordered pair of
mathematical system S, 74 sets, 135, 138
representation of a set or relation by semi-reducibility of one ordered pair
a predicate in an elementary formal of sets to another ordered pair of
system, 91 sets, 135, 138, 139
Rice’s Theorem, 119, 120 sentence of a mathematical system S,
Rice, H. G., 119, 120 73
robin combinator R, 187 sentence of an elementary formal
Rosenbloom, Paul, 244 system, 90
276 Index

sentence of the universal system U , stumping of a machine by a number,

100 70
separability of a pair of R-sets, 78 subformula principle, 30
separation of a disjoint pair of successor combinator σ, 234
number sets in a simple system, 153 successor ordinal, 256, 257
separation of two binary relations by successor set, 256
a formula in a first-order system, symmetric Gentzen axiom system for
160 first-order logic GG, 39–46
Separation Principle, 136, 138 synthesis of results from three fields
sequent (Gentzen), 29–46 (special topics), 69–83
sequent corresponding to a set of Synthesis System, 79–83
signed formulas, 33 synthetic consistency property, 54–56
set A in synthesis system, 79
set B in synthesis system, 79 T combinator (thrush), 187
set operations, similarity with logical tableau proof of a sequent, 34
connectives, 3, 4 Tarski, Alfred, 159
Shepherdson’s Theorem, 161 tautological consequence of a set of
Shepherdson, John, 161 formulas, 11, 12
similar registers, 70
tautological implication of a sentence
Simple Iteration Theorem, 115, 116
by a set of sentences, 26
simple system, 149–161
tautological sequent, 30
simple system associated with a
tautologically complete consequence
system of first-order arithmetic, 150
relation, 11, 12
simple term, in an S, K combinatorial
term induction principle, 218
system, 218
term of a combinatory system, 215
skew diagonalization of a number in a
simple system, 151 term of an elementary formal system,
Skolem–Löwenheim Theorem, 23, 27, 89
28, 49, 51 thrush combinator T , 187
smug element, in a combinatory logic, To Mock a Mockingbird, 175–226
175 transcribed dyadic system (TDS), 99,
solvable property P , 119 100
solvable set or relation, 93 transcribed predicate (TS predicate),
Sprenger, M., 179 99
standard axiom system for transcribed variable (TS variable), 99
propositional logic, 13 transitive rescursion theorem, 257
standard simple system, 154–161 transitive set, 256
starling combinator S, 194 true sentence of an elementary formal
Statman, Richard, 179 system, 90
strong double fixed point property, true sentence of the universal system
208 U , 100
Strong Double Myhill Theorem, 134 truth set in first-order logic, 25
Strong Double Recursion Theorem, truth table verification of Boolean
133–135, 137, 140 equations, 3, 4
Strong Recursion Theorem, 120, 121 truth-functional basis for first-order
stump detector register, 72 logic, 26
 Index 277

truth-functional implication of a unsolvable problems and Rice’s

sentence by a set of sentences, 26 Theorem, 119, 120
truth-functionally satisfiable, of
first-order sentences, 24 V combinator (vireo), 188
TS predicate (transcribed predicate), valid sequent, 30
99 valuation, 9, 10, 23–27
TS variable (transcribed variable), 99 valuation for first-order logic, 23
Turing combinator U , 206 variable of an elementary formal
Turing, Alan, 206, 207, 236 system, 89
variations on a theme of Gödel
U combinator (Turing combinator), (a mathematical system S), 73–76
206 vireo combinator V , 188
U system, universal system for Von Neumann’s positive integers as
recursively enumerable sets and sets, 255
relations, 99–103 Von Neumann, John, 255
undecidability of the Axiom of Choice
(AC) with ZF (shown by Gödel W combinator (warbler), 192
and Cohen), 258 W 0 combinator (converse warbler),
undecidability of the Generalized 192
Continuum Hypothesis (GCH) with W1 , W2 , W3 combinators, 193
ZF (shown by Gödel and Cohen), warbler combinator W , 192
258, 260 weak double fixed point property, 207
undecidable number of a simple Weak Double Recursion Theorem,
system, 150 133, 134
undecidable number of a synthesis Weak Recursion Theorem, 118
system, 80 weakly co-productive set, 125, 126
undecidable sentence in a weakly doubly co-productive ordered
mathematical system S, 73 pair of sets, 140
undecidable simple system, 155 well-founded  relation, 257
Unification Theorem, 48–56 Whitehead, Alfred North, 254
uniform incompleteness (for a simple word problem for Smullyan’s
system), 159 combinatorial lark is decidable,
Uniform Iteration Theorem, 117 179
uniformly incompletable simple Wymann-Böni, M., 179
system, 159
uniformly universal set, 124 x-eliminate of a combinatorial term,
universal number in a synthesis 219
system, 80 x1 , . . . , xn eliminate of a
universal R-set, 77 combinatorial term, 221
universal register, 71
universal relation, 114 Zermelo’s positive integers as sets,
universal set, 124 254, 255
universal system U for recursively Zermelo, Ernst, 254–260
enumerable sets and relations, zero tester combinator, 235
99–103 ZF, Zermelo–Fraenkel set theory,
unlimited abstraction principle, 253 254–260