Logic and Proof

Logic and Proof
Release 0.1
Jeremy Avigad, Robert Y. Lewis, and Floris van Doorn
Jan 27, 2018

CONTENTS
1 Introduction 1
1.1 Mathematical Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Symbolic Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Interactive Theorem Proving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 The Semantic Point of View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Goals Summarized . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 About this Textbook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Propositional Logic 7
2.1 A Puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 A Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Rules of Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 The Language of Propositional Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Natural Deduction for Propositional Logic 17

3.1 Derivations in Natural Deduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Forward and Backward Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 Reasoning by Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Some Logical Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4 Propositional Logic in Lean 25

4.1 Expressions for Propositions and Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 More commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3 Building Natural Deduction Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4 Forward Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 Definitions and Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.6 Additional Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5 Classical Reasoning 41
5.1 Proof by Contradiction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Some Classical Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6 Semantics of Propositional Logic 47

6.1 Truth Values and Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6.2 Truth Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.3 Soundness and Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
i
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7 First Order Logic 53

7.1 Functions, Predicates, and Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.2 The Universal Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.3 The Existential Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
7.4 Relativization and Sorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.5 Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
8 Natural Deduction for First Order Logic 61

8.1 Rules of Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.2 The Universal Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
8.3 The Existential Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8.4 Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
8.5 Counterexamples and Relativized Quantifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
9 First Order Logic in Lean 69

9.1 Functions, Predicates, and Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.2 Using the Universal Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
9.3 Using the Existential Quantifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
9.4 Equality and calculational proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
9.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
10 Semantics of First Order Logic 85

10.1 Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
10.2 Truth in a Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
10.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
10.4 Validity and Logical Consequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.5 Soundness and Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
10.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
11 Sets 93
11.1 Elementary Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
11.2 Calculations with Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
11.3 Indexed Families of Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
11.4 Cartesian Product and Power Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
11.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
12 Sets in Lean 105

12.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
12.2 Some Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
12.3 Indexed Families . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
12.4 Power Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
12.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
13 Relations 117
13.1 Order Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
13.2 More on Orderings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
13.3 Equivalence Relations and Equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
13.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
14 Relations in Lean 123

14.1 Order Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
ii
14.2 Orderings on Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
14.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
15 Functions 129
15.1 The Function Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
15.2 Injective, Surjective, and Bijective Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
15.3 Functions and Subsets of the Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
15.4 Functions and Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
15.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
16 Functions in Lean 137

16.1 Functions and Symbolic Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
16.2 Second- and Higher-Order Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
16.3 Functions in Lean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
16.4 Defining the Inverse Classically . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
16.5 Functions and Sets in Lean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
16.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
17 The Natural Numbers and Induction 149

17.1 The Principle of Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
17.2 Variants of Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
17.3 Recursive Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
17.4 Defining Arithmetic Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
17.5 Arithmetic on the Natural Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
17.6 The Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
17.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
18 The Natural Numbers and Induction in Lean 165

18.1 Induction and Recursion in Lean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
18.2 Defining the Arithmetic Operations in Lean . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
18.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
19 Elementary Number Theory 171

19.1 The Quotient-Remainder Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
19.2 Divisibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
19.3 Prime Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
19.4 Modular Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
19.5 Properties of Squares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
19.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
20 Combinatorics 181
20.1 Finite Sets and Cardinality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
20.2 Counting Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
20.3 Ordered Selections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
20.4 Combinations and Binomial Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
20.5 The Inclusion-Exclusion Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
20.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
21 The Real Numbers 191

21.1 The Number Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
21.2 Quotient Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
21.3 Constructing the Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
21.4 The Completeness of the Real Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
21.5 An Alternative Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
21.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
iii
22 The Infinite 199
22.1 Equinumerosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
22.2 Countably Infinite Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
22.3 Cantor’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
22.4 An Alternative Definition of Finiteness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
22.5 The Cantor-Bernstein Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
22.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
23 Axiomatic Foundations 207

23.1 Basic Axioms for Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
23.2 The Axiom of Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
23.3 The Remaining Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
23.4 Type Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
23.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
iv
CHAPTER
ONE
INTRODUCTION
1.1 Mathematical Proof
Although there is written evidence of mathematical activity in Egypt as early as 3000 BC, many scholars
locate the birth of mathematics proper in ancient Greece around the sixth century BC, when deductive proof
was first introduced. Aristotle credited Thales of Miletus with recognizing the importance of not just what
we know but how we know it, and finding grounds for knowledge in the deductive method. Around 300
BC, Euclid codified a deductive approach to geometry in his treatise, the Elements. Through the centuries,
Euclid’s axiomatic style was held as a paradigm of rigorous argumentation, not just in mathematics, but in
philosophy and the sciences as well.
Here is an example of an ordinary proof, in contemporary mathematical language. It establishes a fact that
was known to the Pythagoreans.
√
Theorem. 2 is irrational, which is to say, it cannot be expressed as a fraction a/b, where a and b are
integers.
√
Proof. Suppose 2 = a/b for some pair of integers a and b. By removing any common factors,√ we can
assume a/b is in lowest terms, so that a and b have no factor in common. Then we have a = 2b, and
squaring both sides, we have a2 = 2b2 .
The last equation implies that a2 is even, and since the square of an odd number is odd, a itself must be
even as well. We therefore have a = 2c for some integer c. Substituting this into the equation a2 = 2b2 , we
have 4c2 = 2b2 , and hence 2c2 = b2 . This means that b2 is even, and so b is even as well.
The fact that a and
√ b are both even contradicts the fact that a and b have no common factor. So the original
assumption that 2 = a/b is false.
In the next example, we focus on the natural numbers,
N = {0, 1, 2, . . .}
A natural number n greater than or equal to 2 is said to be composite if it can be written as a product
n = m · k where neither m nor k is equal to 1, and prime otherwise. Notice that if n = m · k witnesses the
fact that n is composite, then m and k are both smaller than n. Notice also that, by convention, 0 and 1
are considered neither prime nor composite.
Theorem. Every natural number greater than equal to 2 can be written as a product of primes.
1
Logic and Proof, Release 0.1
Proof. We proceed by induction on n. Let n be any natural number greater than 2. If n is prime, we are
done; we can consider n itself as a product with one term. Otherwise, n is composite, and we can write
n = m · k where m and k are smaller than n and greater than 1. By the inductive hypothesis, each of m and
k can be written as a product of primes, say m = p1 · p2 · . . . · pu and k = q1 · q2 · . . . · qv . But then we have
n = m · k = p1 · p2 · . . . · pu · q1 · q2 · . . . · qv ,
a product of primes, as required.
Later, we will see that more is true: every natural number greater than 2 can be written as a product of
primes in a unique way, a fact known as the fundamental theorem of arithmetic.
The first goal of this course is to teach you to write clear, readable mathematical proofs. We will do this by
considering a number of examples, but also by taking a reflective point of view: we will carefully study the
components of mathematical language and the structure of mathematical proofs, in order to gain a better
understanding of how they work.
1.2 Symbolic Logic
Towards understanding how proofs work, it will be helpful to study a subject known as “symbolic logic,”
which provides an idealized model of mathematical language and proof. In the Prior Analytics, the ancient
Greek philosopher set out to analyze patterns of reasoning, and developed the theory of the syllogism. Here
is one instance of a syllogism:
Every man is an animal.

Every animal is mortal.
Therefore every man is mortal.
Aristotle observed that the correctness of this inference has nothing to do with the truth or falsity of the
individual statements, but, rather, the general pattern:
Every A is B.
Every B is C.
Therefore every A is C.
We can substitute various properties for A, B, and C; try substituting the properties of being a fish, being a
unicorn, being a swimming creature, being a mythical creature, etc. The various statements that result may
come out true or false, but all the instantiations will have the following crucial feature: if the two hypotheses
come out true, then the conclusion comes out true as well. We express this by saying that the inference is
valid.
Although the patterns of language addressed by Aristotle’s theory of reasoning are limited, we have him to
thank for a crucial insight: we can classify valid patterns of inference by their logical form, while abstracting
away specific content. It is this fundamental observation that underlies the entire field of symbolic logic.
2 Chapter 1. Introduction
In the seventeenth century, Leibniz proposed the design of a characteristica universalis, a universal symbolic
language in which one would express any assertion in a precise way, and a calculus ratiocinatur, a “calculus
of thought” which would express the precise rules of reasoning. Leibniz himself took some steps to develop
such a language and calculus, but much greater strides were made in the nineteenth century, through the
work of Boole, Frege, Peirce, Schroeder, and others. Early in the twentieth century, these efforts blossomed
into the field of mathematical logic.
If you consider the examples of proofs in the last section, you will notice that some terms and rules of
inference are specific to the subject matter at hand, having to do with numbers, and the properties of being
prime, composite, even, odd, and so on. But there are other terms and rules of inference that are not domain
specific, such as those related to the words “every,” “some,” “and,” and “if … then.” The goal of symbolic
logic is to identify these core elements of reasoning and argumentation and explain how they work, as well
as to explain how more domain-specific notions are introduced and used.
To that end, we will introduce symbols for key logical notions, including the following:
• A → B, “if A then B“
• A ∧ B, “A and B“
• A ∨ B, “A or B“
• ¬A, “not A“
• ∀x A, “for every x, A“
• ∃x A, “for some x, A“
We will then provide a formal proof system that will let us establish, deductively, that certain entailments
between such statements are valid.
The proof system we will use is a version of natural deduction, a type of proof system introduced by Gerhard
Gentzen in the 1930’s to model informal styles of argument. In this system, the fundamental unit of judgment
is the assertion that an assertion, A, follows from a finite set of hypotheses, Γ. This is written as Γ ⊢ A. If
Γ and ∆ are two finite sets of hypotheses, we will write Γ, ∆ for the union of these two sets, that is, the set
consisting of all the hypotheses in each. With these conventions, the rule for the conjunction symbol can be
expressed as follows:
Γ ⊢ A ∆ ⊢ B
Γ, ∆ ⊢ A ∧ B
This should be interpreted as saying: assuming A follows from the hypotheses Γ, and B follows from the
hypotheses ∆, A ∧ B follows from the hypotheses in both Γ and ∆.
We will see that one can write such proofs more compactly leaving the hypotheses implicit, so that the rule
above is expressed as follows:
A B
A∧B
In this format, a snippet of the first proof in the previous section might be rendered as follows:
∀x (¬even(x) → ¬even(x2 ))
¬even(b) ¬even(b) → ¬even(b2 ))
¬even(b2 ) even(b2 )
⊥
even(b)
The complexity of such proofs can quickly grow out of hand, and complete proofs of even elementary
mathematical facts can become quite long. Such systems are not designed for writing serious mathematics.
1.2. Symbolic Logic 3

Rather, they provide idealized models of mathematical inference, and insofar as they capture something of
the structure of an informal proof, they enable us to study the properties of mathematical reasoning.
The second goal of this course is to help you understand natural deduction, as an example of a formal
deductive system.
1.3 Interactive Theorem Proving
Early work in mathematical logic aimed to show that ordinary mathematical arguments could be modeled
in symbolic calculi, at least in principle. As noted above, complexity issues limit the range of what can be
accomplished in practice; even elementary mathematical arguments require long derivations that are hard
to write and hard to read, and do little to promote understanding of the underlying mathematics.
Since the end of the twentieth century, however, the advent of computational proof assistants has begun
to make complete formalization feasible. Working interactively with theorem proving software, users can
construct formal derivations of complex theorems that can be stored and checked by computer. Automated
methods can be used to fill in small gaps by hand, verify long calculations axiomatically, or fill in long
chains of inferences deterministically. The reach of automation is currently fairly limited, however. The
strategy used in interactive theorem proving is to ask users to provide just enough information for the
system to be able to construct and check a formal derivation. This typically involves writing proofs in a sort
of “programming language” that is designed with that purpose in mind. For example, here is a short proof
in the Lean theorem prover:
section
variables (P Q : Prop)
theorem my_theorem : P ∧ Q → Q ∧ P :=
assume h : P ∧ Q,
have P, from and.left h,
have Q, from and.right h,
show Q ∧ P, from and.intro ‹Q› ‹P›
end
If you are reading the present text in online form, you will find a button underneath the formal “proof script”
that says “try it!” Pressing the button copies the proof to an editor window at right, and runs a version of
Lean inside your browser to process the proof, turn it into an axiomatic derivation, and verify its correctness.
You can experiment by varying the text in the editor and pressing the “play” button to see the result.
Proofs in Lean can access a library of prior mathematical results, all verified down to axiomatic foundations.
A goal of the field of interactive theorem proving is to reach the point where any contemporary theorem can
be verified in this way. For example, here is a formal proof that the square root of two is irrational, following
the model of the informal proof presented above:
import data.nat.prime
open nat
theorem sqrt_two_irrational {a b : N} (co : gcd a b = 1) :

a^2 ̸= 2 * b^2 :=
assume h : a^2 = 2 * b^2,
have 2 | a^2,
by simp [h],
have 2 | a,
from dvd_of_prime_of_dvd_pow prime_two this,
exists.elim this $
assume (c : nat) (aeq : a = 2 * c),
have 2 * (2 * c^2) = 2 * b^2,

by simp [eq.symm h, aeq]; simp [pow_succ],
have 2 * c^2 = b^2,
from eq_of_mul_eq_mul_left dec_trivial this,
have 2 | b^2,
by simp [eq.symm this],
have 2 | b,
from dvd_of_prime_of_dvd_pow prime_two this,
have 2 | gcd a b,
from dvd_gcd ‹2 | a› ‹2 | b›,
have 2 | (1 : N),
by simp * at *,
show false, from absurd ‹2 | 1› dec_trivial
The third goal of this course is to teach you to write elementary proofs in Lean. The facts that we will ask
you to prove in Lean will be more elementary than the informal proofs we will ask you to write, but our
intent is that formal proofs will model and clarify the informal proof strategies we will teach you.
1.4 The Semantic Point of View
As we have presented the subject here, the goal of symbolic logic is to specify a language and rules of
inference that enable us to get at the truth in a reliable way. The idea is that the symbols we choose denote
objects and concepts that have a fixed meaning, and the rules of inference we adopt enable us to draw true
conclusions from true hypotheses.
One can adopt another view of logic, however, as a system where some symbols have a fixed meaning, such
as the symbols for “and,” “or,” and “not,” and others have a meaning that is taken to vary. For example, the
expression P ∧ (Q ∨ R), read “P and either Q or R,” may be true or false depending on the basic assertions
that P , Q, and R stand for. More precisely, the truth of the compound expression depends only on whether
the component symbols denote expressions that are true or false. For example, if P , Q, and R stand for
“seven is prime,” “seven is even,” and “seven is odd,” respectively, then the expression is true. If we replace
“seven” by “six,” the statement is false. More generally, the expression comes out true whenever P is true
and at least one of Q and R is true, and false otherwise.
From this perspective, logic is not so much a language for asserting truth, but a language for describing
possible states of affairs. In other words, logic provides a specification language, with expressions that can
be true or false depending on how we interpret the symbols that are allowed to vary. For example, if we
fix the meaning of the basic predicates, the statement “there is a red block between two blue blocks” may
be true or false of a given “world” of blocks, and we can take the expression to describe the set of worlds
in which it is true. Such a view of logic is important in computer science, where we use logical expressions
to select entries from a database matching certain criteria, to specify properties of hardware and software
systems, or to specify constraints that we would like a constraint solver to satisfy.
There are important connections between the syntactic / deductive point of view on the one hand, and
the semantic / model-theoretic point of view on the other. We will explore some of these along the way.
For example, we will see that it is possible to view the “valid” assertions as those that are true under all
possible interpretations of the non-fixed symbols, and the “valid” inferences as those that maintain truth in
all possible states and affairs. From this point of view, a deductive system should only allow us to derive
valid assertions and entailments, a property known as soundness. If a deductive system is strong enough to
allow us to verify all valid assertions and entailments, it is said to be complete.
The fourth goal of course is to convey the semantic view of logic, and understand how logical expressions
can be used to specify states of affairs.
1.4. The Semantic Point of View 5

1.5 Goals Summarized
To summarize, these are the goals of this course:

• to teach you to write clear, “literate,” mathematical proofs
• to introduce you to symbolic logic and the formal modeling of deductive proof
• to introduce you to interactive theorem proving
• to teach you to understand how to use logic as a precise language for making claims about systems of
objects and the relationships between them, and specifying certain states of affairs.
Let us take a moment to consider the relationship between some of these goals. It is important not to confuse
the first three. We are dealing with three kinds of mathematical language: ordinary mathematical language,
the symbolic representations of mathematical logic, and computational implementations in interactive proof
assistants. These are very different things!
Symbolic logic is not meant to replace ordinary mathematical language, and you should not use symbols like
∧ and ∨ in ordinary mathematical proofs any more than you would use them in place of the words “and”
and “or” in letters home to your parents. Natural languages provide nuances of expression that can convey
levels of meaning and understanding that go beyond pattern matching to verify correctness. At the same
time, modeling mathematical language with symbolic expressions provides a level of precision that makes it
possible to turn mathematical language itself into an object of study. Each has its place, and we hope to get
you to appreciate the value of each without confusing the two.
The proof languages used by interactive theorem provers lie somewhere between the two extremes. On the one
hand, they have to be specified with enough precision for a computer to process them and act appropriately;
on the other hand, they aim to capture some of the higher-level nuances and features of informal language in
a way that enables us to write more complex arguments and proofs. Rooted in symbolic logic and designed
with ordinary mathematical language in mind, they aim to bridge the gap between the two.
1.6 About this Textbook
Both this online textbook and the Lean theorem prover it invokes are new and ongoing projects, and in
places they are still rough. Please bear with us! Your feedback will be quite helpful.
CHAPTER
TWO
PROPOSITIONAL LOGIC
2.1 A Puzzle
The following puzzle, titled “Malice and Alice,” is from George J. Summers’ Logical Deduction Puzzles.
Alice, Alice’s husband, their son, their daughter, and Alice’s brother were involved in a murder. One of the
five killed one of the other four. The following facts refer to the five people mentioned:
1. A man and a woman were together in a bar at the time of the murder.
2. The victim and the killer were together on a beach at the time of the murder.
3. One of Alice’s two children was alone at the time of the murder.
4. Alice and her husband were not together at the time of the murder.
5. The victim’s twin was not the killer.
6. The killer was younger than the victim.
Which one of the five was the victim?
Take some time to try to work out a solution. (You should assume that the victim’s twin is one of the five
people mentioned.) Summers’ book offers the following hint: “First find the locations of two pairs of people
at the time of the murder, and then determine who the killer and the victim were so that no condition is
contradicted.”
2.2 A Solution
If you have worked on the puzzle, you may have noticed a few things. First, it is helpful to draw a diagram,
and to be systematic about searching for an answer. The number of characters, locations, and attributes
is finite, so that there are only finitely many possible “states of affairs” that need to be considered. The
numbers are also small enough so that systematic search through all the possibilities, though tedious, will
eventually get you to the right answer. This is a special feature of logic puzzles like this; you would not
expect to show, for example, that every even number greater than two can be written as a sum of primes by
running through all the possibilities.
Another thing that you may have noticed is that the question seems to presuppose that there is a unique
answer to the question, which is to say, of all the states of affairs that meet the list of conditions, there
is only one person who can possibly be the killer. A priori, without that assumption, there is a difference
between finding some person who could have been the victim, and show that that person had to be the
7
victim. In other words, there is a difference between exhibiting some state of affairs that meets the criteria,
and demonstrating conclusively that no other solution is possible.
The published solution in the book not only produces a state of affairs that meets the criterion, but at the
same time proves that this is the only one that does so. It is quoted below, in full.
From (1), (2), and (3), the roles of the five people were as follows: Man and Woman in the bar, Killer and
Victim on the beach, and Child alone.
Then, from (4), either Alice’s husband was in the bar and Alice was on the beach, or Alice was in the bar
and Alice’s husband was on the beach.
If Alice’s husband was in the bar, the woman he was with was his daughter, the child who was alone was his
son, and Alice and her brother were on the beach. Then either Alice or her brother was the victim; so the
other was the killer. But, from (5), the victim had a twin, and this twin was innocent. Since Alice and her
brother could only be twins to each other, this situation is impossible. Therefore Alice’s husband was not
in the bar.
So Alice was in the bar. If Alice was in the bar, she was with her brother or her son.
If Alice was with her brother, her husband was on the beach with one of the two children. From (5), the
victim could not be her husband, because none of the others could be his twin; so the killer was her husband
and the victim was the child he was with. But this situation is impossible, because it contradicts (6).
Therefore, Alice was not with her brother in the bar.
So Alice was with her son in the bar. Then the child who was alone was her daughter. Therefore, Alice’s
husband was with Alice’s brother on the beach. From previous reasoning, the victim could not be Alice’s
husband. But the victim could be Alice’s brother because Alice could be his twin.
So Alice’s brother was the victim and Alice’s husband was the killer.
This argument relies on some “extralogical” elements, for example, that a father cannot be younger than his
child, and that a parent and his or her child cannot be twins. But the argument also involves a number of
common logical terms and associated patterns of inference. In the next section, we will focus on some of the
key logical terms occurring in the argument above, words like “and,” “or,” “not,” and “if … then.”
Our goal is to give an account of the patterns of inference that govern the use of those terms. To that
end, using the methods of symbolic logic, we will introduce variables A, B, C, … to stand for fundamental
statements, or propositions, and symbols ∧, ∨, ¬, and → to stand for “and,” “or,” “not,” and “if … then
… ,” respectively. Doing so will let us focus on the way that compound statements are built up from basic
ones using the logical terms, while abstracting away from the specific content. We will also adopt a stylized
notation for representing inferences as rules: the inscription
A B
C
indicates that statement C is a logical consequence of A and B.
2.3 Rules of Inference
2.3.1 Implication
The first pattern of inference we will discuss, involving the “if … then …” construct, can be hard to discern.
Its use is largely implicit in the solution above. The inference in the fourth paragraph, spelled out in greater
detail, runs as follows:
8 Chapter 2. Propositional Logic

If Alice was in the bar, Alice was with her brother or her son.
Alice was in the bar.
Alice was with her brother or son.
This rule is sometimes known as modus ponens, or “implication elimination,” since it tells us how to use an
implication in an argument. As a rule, it is expressed as follows:
A→B A →E
B
Read this as saying that if you have a proof of A → B, possibly from some hypotheses, and a proof of A,
possibly from hypotheses, then combining these yields a proof of B, from the hypotheses in both subproofs.
The rule for deriving an “if … then” statement is more subtle. Consider the beginning of the third paragraph,
which argues that if Alice’s husband was in the bar, then Alice or her brother was the victim. Abstracting
away some of the details, the argument has the following form:
Suppose Alice’s husband was in the bar.

Then …
Then …
Then Alice or her brother was the victim.
Thus, if Alice’s husband was in the bar, then Alice or her brother was the victim.
This is a form of hypothetical reasoning. On the supposition that A holds, we argue that B holds as well.
If we are successful, we have shown that A implies B, without supposing A. In other words, the temporary
assumption that A holds is “canceled” by making it explicit in the conclusion.
1
A
..
.
B 1 →I
A→B
The hypothesis is given the label 1; when the introduction rule is applied, the label 1 indicates the rele-
vant hypothesis. The line over the hypothesis indicates that the assumption has been “canceled” by the
introduction rule.
2.3.2 Conjunction
As was the case for implication, other logical connectives are generally characterized by their introduction
and elimination rules. An introduction rule shows how to establish a claim involving the connective, while
an elimination rule shows how to use such a statement that contains the connective to derive others.
Let us consider, for example, the case of conjunction, that is, the word “and.” Informally, we establish a
conjunction by establishing each conjunct. For example, informally we might argue:
Alice’s brother was the victim.
2.3. Rules of Inference 9

Alice’s husband was the killer.

Therefore Alice’s brother was the victim and Alice’s husband was the killer.
The inference seems almost too obvious to state explicitly, since the word “and” simply combines the two
assertions into one. Informal proofs often downplay the distinction. In symbolic logic, the rule reads as
follows:
A B ∧I
A∧B
The two elimination rules allow us to extract the two components:
Alice’s husband was in the bar and Alice was on the beach.
So Alice’s husband was in the bar.
Or:
Alice’s husband was in the bar and Alice was on the beach.
So Alice was on the beach.
In symbols, these patterns are rendered as follows:

A∧B ∧El
A∧B ∧Er
A B
Here the l and r stand for “left” and “right”.
2.3.3 Negation and Falsity
In logical terms, showing “not A” amounts to showing that A leads to a contradiction. For example:

…
This situation is impossible.
Therefore Alice’s husband was not in the bar.
This is another form of hypothetical reasoning, similar to that used in establishing an “if … then” statement:
we temporarily assume A, show that leads to a contradiction, and conclude that “not A” holds. In symbols,
the rule reads as follows:
1
A
..
.
⊥ 1 ¬I
¬A

The elimination rule is dual to these. It expresses that if we have both “A” and “not A,” then we have a
contradiction. This pattern is illustrated in the informal argument below, which is implicit in the fourth
paragraph of the solution to “Malice and Alice.”
The killer was Alice’s husband and the victim was the child he was with.
So the killer was not younger than his victim.
But according to (6), the killer was younger than his victim.
In symbolic logic, the rule of inference is expressed as follows:

¬A A ¬E
⊥
Notice also that in the symbolic framework, we have introduced a new symbol, ⊥. It corresponds to natural
language phrases like “this is a contradiction” or “this is impossible.”
What are the rules governing ⊥? In the proof system we will introduce in the next chapter, there is
no introduction rule; “false” is false, and there should be no way to prove it, other than extract it from
contradictory hypotheses. On the other hand, the system provides a rule that allows us to conclude anything
from a contradiction:
⊥ ⊥E
A
The elimination rule also has the fancy Latin name, ex falso sequitur quodlibet, which means “anything you
want follows from falsity.”
This elimination rule is harder to motivate from a natural language perspective, but, nonetheless, it is needed
to capture common patterns of inference. One way to understand it is this. Consider the following statement:
For every natural number n, if n is prime and greater than 2, then n is odd.
We would like to say that this is a true statement. But if it is true, then it is true of any particular number
n. Taking n = 2, we have the statement:
If 2 is prime and greater than 2, then 2 is odd.
In this conditional statement, both the antecedent and succedent are false. The fact that we are committed
to saying that this statement is true shows that we should be able to prove, one way or another, that the
statement 2 is odd follows from the false statement that 2 is prime and greater than 2. The ex falso neatly
encapsulates this sort of inference.
Notice that if we define ¬A to be A → ⊥, then the rules for negation introduction and elimination are
nothing more than implication introduction and elimination, respectively. We can think of ¬A expressed
colorfully by saying “if A is true, then pigs have wings,” where “pigs have wings” is stands for ⊥.
Having introduced a symbol for “false,” it is only fair to introduce a symbol for “true.” In contrast to “false,”
“true” has no elimination rule, only an introduction rule:
⊤
Put simply, “true” is true.

2.3.4 Disjunction
The introduction rules for disjunction, otherwise known as “or,” are straightforward. For example, the claim
that condition (3) is met in the proposed solution can be justified as follows:
Alice’s daughter was alone at the time of the murder.

Therefore, either Alice’s daughter was alone at the time of the murder, or Alice’s son was alone at the time
of the murder.
In symbolic terms, the two introduction rules are as follows:

A ∨Il
B ∨Ir
A∨B A∨B
Here, again, the l and r stand for “left” and “right”.
The disjunction elimination rule is trickier, but it represents a natural form of case-based hypothetical
reasoning. The instances that occur in the solution to “Malice and Alice” are all special cases of this rule, so
it will be helpful to make up a new example to illustrate the general phenomenon. Suppose, in the argument
above, we had established that either Alice’s brother or her son was in the bar, and we wanted to argue for
the conclusion that her husband was on the beach. One option is to argue by cases: first, consider the case
that her brother was in the bar, and argue for the conclusion on the basis of that assumption; then consider
the case that her son was in the bar, and argue for the same conclusion, this time on the basis of the second
assumption. Since the two cases are exhaustive, if we know that the conclusion holds in each case, we know
that it holds outright. The pattern looks something like this:
Either Alice’s brother was in the bar, or Alice’s son was in the bar.
Suppose, in the first case, that her brother was in the bar. Then … Therefore, her husband was on the beach.
On the other hand, suppose her son was in the bar. In that case, … Therefore, in this case also, her husband
was on the beach.
Either way, we have established that her husband was on the beach.
In symbols, this pattern is expressed as follows:

1 1
A B
.. ..
. .
A∨B C C 1 ∨E
C
What makes this pattern confusing is that it requires two instances of nested hypothetical reasoning: in the
first block of parentheses, we temporarily assume A, and in the second block, we temporarily assume B.
When the dust settles, we have established C outright.
There is another pattern of reasoning that is commonly used with “or,” as in the following example:
Either Alice’s husband was in the bar, or Alice was in the bar.
Alice’s husband was not in the bar.
So Alice was in the bar.
In symbols, we would render this rule as follows:

A∨B ¬A
B
We will see in the next chapter that it is possible to derive this rule from the others. As a result, we will not
take this to be a fundamental rule of inference in our system.
2.3.5 If and only if
In mathematical arguments, it is common to say of two statements, A and B, that “A holds if and only if B
holds.” This assertion is sometimes abbreviated “A iff B,” and means simply that A implies B and B implies
A. It is not essential that we introduce a new symbol into our logical language to model this connective,
since the statement can be expressed, as we just did, in terms of “implies” and “and.” But notice that the
length of the expression doubles because A and B are each repeated. The logical abbreviation is therefore
convenient, as well as natural.
The conditions of “Malice and Alice” imply that Alice is in the bar if and only if Alice’s husband is on the
beach. Such a statement is established by arguing for each implication in turn:
I claim that Alice is in the bar if and only if Alice’s husband is on the beach.
To see this, first suppose that Alice is in the bar.
Then …
Hence Alice’s husband is on the beach.
Conversely, suppose Alice’s husband is on the beach.
Then …
Hence Alice is in the bar.
Notice that with this example, we have varied the form of presentation, stating the conclusion first, rather
than at the end of the argument. This kind of “signposting” is common in informal arguments, in that is
helps guide the reader’s expectations and foreshadow where the argument is going. The fact that formal
systems of deduction do not generally model such nuances marks a difference between formal and informal
arguments, a topic we will return to below.
The introduction is modeled in natural deduction as follows:
1 1
A B
.. ..
. .
B A 1 ↔I
A↔B
The elimination rules for iff are unexciting. In informal language, here is the “left” rule:
Alice is in the bar if and only if Alice’s husband is on the beach.

Alice is in the bar.
Hence, Alice’s husband is on the beach.
The “right” rule simply runs in the opposite direction.

Alice is in the bar if and only if Alice’s husband is on the beach.

Alice’s husband is on the beach.
Hence, Alice is in the bar.
Rendered in natural deduction, the rules are as follows:

A↔B A ↔El
A↔B B ↔Er
B A
2.3.6 Proof by Contradiction
We saw an example of an informal argument that implicitly uses the introduction rule for negation:

…
Therefore Alice’s husband was not in the bar.
Consider the following argument:
Suppose Alice’s husband was not on the beach.

…
Therefore Alice’s husband was on the beach.
At first glance, you might think this argument follows the same pattern as the one before. But a closer look
should reveal a difference: in the first argument, a negation is introduced into the conclusion, whereas in the
second, it is eliminated from the hypothesis. Using negation introduction to close the second argument would
yield the conclusion “It is not the case that Alice’s husband was not on the beach.” The rule of inference
that replaces the conclusion with the positive statement that Alice’s husband was on the beach is called a
proof by contradiction. (It also has a fancy name, reductio ad absurdum, “reduction to an absurdity.”)
It may be hard to see the difference between the two rules, because we commonly take the statement “Alice’s
husband was not not on the beach” to be a roundabout and borderline ungrammatical way of saying that
Alice’s husband was on the beach. Indeed, the rule is equivalent to adding an axiom that says that for every
statement A, “not not A” is equivalent to A.
There is a style of doing mathematics known as “constructive mathematics” that denies the equivalence of
“not not A” and A. Constructive arguments tend to have much better computational interpretations; a proof
that something is true should provide explicit evidence that the statement is true, rather than evidence that
it can’t possibly be false. We will discuss constructive reasoning in a later chapter. Nonetheless, proof by
contradiction is used extensively in contemporary mathematics, and so, in the meanwhile, we will use proof
by contradiction freely as one of our basic rules.
In natural deduction, proof by contradiction is expressed by the following pattern:

1
¬A
..
.
⊥ RAA,1
A
The assumption ¬A is canceled at the final inference.
2.4 The Language of Propositional Logic
The language of propositional logic starts with symbols A, B, C, … which are intended to range over basic
assertions, or propositions, which can be true or false. Compound expressions are built up using parentheses
and the logical symbols introduced in the last section. For example,
((A ∧ ¬B) → ¬(C ∨ D))
is an example of a propositional formula.

When writing expressions in symbolic logic, we will adopt an order of operations which allows us to drop
superfluous parentheses. When parsing an expression:
• negation binds most tightly
• then conjunctions and disjunctions, from right to left
• and finally implications and bi-implications.
So, for example, the expression ¬A ∨ B → C ∧ D is understood as ((¬A) ∨ B) → (C ∧ D)
For example, suppose we assign the following variables:
• A: Alice’s husband was in the bar
• B: Alice was on the beach
• C: Alice was in the bar
• D: Alice’s husband was on the beach
Then the statement “either Alice’s husband was in the bar and Alice was on the beach, or Alice was in the
bar and Alice’s husband was on the beach would be rendered as
(A ∧ B) ∨ (C ∧ D)
Sometimes the appropriate translation is not so straightforward, however. Because natural language is more
flexible and nuanced, a degree of abstraction and regimentation is needed to carry out the translation.
Sometimes different translations are arguably reasonable. In happy situations, alternative translations will
be logically equivalent, in the sense that one can derive each from the other using purely logical rules. In
less happy situations, the translations will not be equivalent, in which case the original statement is simply
ambiguous, from a logical point of view. In cases like that, choosing a symbolic representation helps clarify
the intended meaning.
Consider, for example, a statement like “Alice was with her son on the beach, but her husband was alone.”
We might choose variables as follows:
• A: Alice was on the beach
• B: Alice’s son was on the beach
• C: Alice’s husband was alone
2.4. The Language of Propositional Logic 15

In that case, we might represent the statement in symbols as A ∧ B ∧ C. Using the word “with” may seem to
connote more than the fact that Alice and her son were both on the beach; for example, it seems to connote
that they aware of each others’ presence, interacting, etc. Similarly, although we have translated the word
“but” and “and,” the word “but” also convey information; in this case, it seems to emphasize a contrast,
while in other situations, it can be used to assert a fact that is contrary to expectations. In both cases, then,
the logical rendering models certain features of the original sentence while abstracting others.
2.5 Exercises
1. Here is another (gruesome) logic puzzle by George J. Summers, called “Murder in the Family.”
Murder occurred one evening in the home of a father and mother and their son and daughter.
One member of the family murdered another member, the third member witnessed the crime,
and the fourth member was an accessory after the fact.
(a) The accessory and the witness were of opposite sex.
(b) The oldest member and the witness were of opposite sex.
(c) The youngest member and the victim were of opposite sex.
(d) The accessory was older than the victim.
(e) The father was the oldest member.
(f) The murderer was not the youngest member.
Which of the four—father, mother, son, or daughter—was the murderer?
Solve this puzzle, and write a clear argument to establish that your answer is correct.
2. Using the mnemonic F (Father), M (Mother), D (Daughter), S (Son), M u (Murderer), V (Victim),
W (Witness), A (Accessory), O (Oldest), Y (Youngest), we can define propositional variables like
F M (Father is the Murderer), DV (Daughter is the Victim), F O (Father is Oldest), V Y (Victim is
Youngest), etc. Notice that only the son or daughter can be the youngest, and only the mother or
father can be the oldest.
With these conventions, the first clue can be represented
((F A ∨ SA) → (M W ∨ DW )) ∧ ((M A ∨ DA) → (F W ∨ SW )),
in other words, if the father or son was the accessory, then the mother or daughter was the witness,
and vice-versa. Represent the other five clues in a similar manner.
Representing the fourth clue is tricky. Try to write down a formula that describes all the possibilities
that are not ruled out by the information.
3. Consider the following three hypotheses:
• Alan likes kangaroos, and either Betty likes frogs or Carl likes hamsters.
• If Betty likes frogs, then Alan doesn’t like kangaroos.
• If Carl likes hamsters, then Betty likes frogs.
Write a clear argument to show that these three hypotheses are contradictory.

CHAPTER
THREE
NATURAL DEDUCTION FOR PROPOSITIONAL LOGIC
Reflecting on the arguments in the previous chapter, we see that, intuitively speaking, some inferences are
valid and some are not. For example, if, in a chain of reasoning, we had established “A and B,” it would
seem perfectly reasonable to conclude B. If we had established A, B, and “If A and B then C,” it would
be reasonable to conclude C. On the other hand, if we had established “A or B,” we would not be justified
concluding B without further information.
The task of symbolic logic is to develop a precise mathematical theory that explains which inferences are
valid and why. There are two general approaches to spelling out the notion of validity. In this chapter,
we will consider the deductive approach: an inference is valid if it can be justified by fundamental rules of
reasoning that reflect the meaning of the logical terms involved. In Chapter 6 we will consider the “semantic”
approach: an inference is valid if it is an instance of a pattern that always yields a true conclusion from true
hypotheses.
3.1 Derivations in Natural Deduction
We have seen that the language of propositional logic allows us to build up expressions from propositional
variables A, B, C, . . . using propositional connectives like →, ∧, ∨, and ¬. We will now consider a formal
deductive system that we can use to prove propositional formulas. There are a number of such systems on
offer; the one will use is called natural deduction, designed by Gerhard Gentzen in the 1930’s.
In natural deduction, every proof is a proof from hypotheses. In other words, in any proof, there is a finite
set of hypotheses {B, C, . . .} and a conclusion A, and what the proof shows is that A follows from B, C, . . ..
Like formulas, proofs are built by putting together smaller proofs, according to the rules. For instance, the
way to read the and-introduction rule,
A B
A∧B
is as follows: if you have a proof P1 of A from some hypotheses, and you have a proof P2 of B from some
hypotheses, then you can put them together using this rule to obtain a proof of A ∧ B, which uses all the
hypotheses in P1 together with all the hypotheses in P2 . For example, this is a proof of (A ∧ B) ∧ (A ∧ C)
from three hypotheses, A, B, and C:
A B A C
A∧B A∧C
(A ∧ B) ∧ (A ∧ C)
One thing that makes natural deduction confusing is that when you put together proofs in this way, hy-
potheses can be eliminated, or, as we will say, canceled. For example, we can apply the implies-introduction
rule to the last proof, and obtain the following proof of B → (A ∧ B) ∧ (A ∧ C) from only two hypotheses,
A and C:
17
1
A B A C
A∧B A∧C
(A ∧ B) ∧ (A ∧ C)
1
B → (A ∧ B) ∧ (A ∧ C)
Here, we have used the label 1 to indicate the place where the hypothesis B was canceled. Any label will
do, though we will tend to use numbers for that purpose.
We can continue to cancel the hypothesis A:
2 1 2
A B AC
A∧B A∧C
(A ∧ B) ∧ (A ∧ C)
1
B → (A ∧ B) ∧ (A ∧ C)
2
A → (B → (A ∧ B) ∧ (A ∧ C))
The result is a proof using only the hypothesis C. We can continue to cancel that hypothesis as well:
2 1 2 3
A B A C
A∧B A∧C
(A ∧ B) ∧ (A ∧ C)
1
B → (A ∧ B) ∧ (A ∧ C)
2
A → (B → (A ∧ B) ∧ (A ∧ C))
3
C → (A → (B → (A ∧ B) ∧ (A ∧ C)))
The resulting proof uses no hypothesis at all. In other words, it establishes the conclusion outright.
Notice that in the second step, we canceled two “copies” of the hypothesis A. In natural deduction, we can
choose which hypotheses to cancel; we could have canceled either one, and left the other hypothesis open. In
fact, we can also carry out the implication-introduction rule and cancel zero hypotheses. For example, the
following is a short proof of A → B from the hypothesis B:
B
A→B
In this proof, zero copies of A are canceled.
Also notice that although we are using letters like A, B, and C as propositional variables, in the proofs
above we can replace them by any propositional formula. For example, we can replace A by the formula
(D ∨ E) everywhere, and still have correct proofs. In some presentations of logic, different letters are used
for to stand for propositional variables and arbitrary propositional formulas, but we will continue to blur the
distinction. You can think of A, B, and C as standing for propositional variables or formulas, as you prefer.
If you think of them as propositional variables, just keep in mind that in any rule or proof, you can replace
every variable by a different formula, and still have a valid rule or proof.
Finally, notice also that in these examples, we have assumed a special rule as the starting point for building
proofs. It is called the assumption rule, and it looks like this:
A
What it means is that at any point we are free to simply assume a formula, A. The single formula A
constitutes a one-line proof, and the way to read this proof is as follows: assuming A, we have proved A.
The remaining rules of inference were given in the last chapter, and we summarize them here.
Implication:
1
A
.. A→B A
. →E
B
B 1 →I
A→B
18 Chapter 3. Natural Deduction for Propositional Logic

Conjunction:
A B ∧I
A∧B ∧El
A∧B ∧Er
A∧B A B
Negation:
1
A
.. ¬A A
. ¬E
⊥
⊥ 1 ¬I
¬A
Disjunction:
1 1
A B
A B .. ..
∨Il ∨Ir . .
A∨B A∨B
A∨B C C 1 ∨E
C
Truth and falsity:
⊥ ⊥E ⊤I
A ⊤
Bi-implication:
1 1
A B
.. .. A↔B A ↔E A↔B B
. . l ↔Er
B A
B A 1 ↔I
A↔B
Reductio ad absurdum (proof by contradiction):
1
¬A
..
.
⊥ 1 RAA
A
3.2 Examples
Let us consider some more examples of natural deduction proofs. In each case, you should think about
what the formulas say and which rule of inference is invoked at each step. Also pay close attention to which
hypotheses are canceled at each stage. If you look at any node of the tree, what has been established at that
point is that the claim follows from all the hypotheses above it that haven’t been canceled yet.
The following is a proof of A → C from A → B and B → C:
1
A A→B
B B→C
C 1
A→C
Intuitively, the formula
(A → B) ∧ (B → C) → (A → C)
“internalizes” the conclusion of the previous proof. The ∧ symbol is used to combine hypotheses, and the
→ symbol is used to express that the right-hand side is a consequence of the left. Here is a proof of that
formula:
3.2. Examples 19
2
1 (A → B) ∧ (B → C) 2
A A→B (A → B) ∧ (B → C)
B B→C
C 1
A→C 2
(A → B) ∧ (B → C) → (A → C)
The next proof shows that if a conclusion, C, follows from A and B, then it follows from their conjunction.
1
2
A∧B
A → (B → C) A A∧B
1
B→C B
C 1
A∧B →C 2
(A → (B → C)) → (A ∧ B → C)
The conclusion of the next proof can be interpreted as saying that if it is not the case that one of A or B is
true, then they are both false. It illustrates the use of the rules for negation.
1 2
3
A 3
B
¬(A ∨ B) A∨B ¬(A ∨ B) A∨B
⊥ 1
⊥ 2
¬A ¬B
¬A ∧ ¬B 3
¬(A ∨ B) → ¬A ∧ ¬B
Finally, the next two examples illustrate the use of the ex falso rule. The first is a derivation of an arbitrary
formula B from ¬A and A:
¬A A
⊥
B
The second shows that B follows from A and ¬A ∨ B:
1
¬A A
⊥ 1
¬A ∨ B B B 1
B
In some proof systems, these rules are taken to be part of the system. But we do not need to that with our
system: these two examples show that the rules can be derived from our other rules.
3.3 Forward and Backward Reasoning
Natural deduction is supposed to represent an idealized model of the patterns of reasoning and argumentation
we use, for example, when working with logic puzzles as in the last chapter. There are obvious differences:
we describe natural deduction proofs with symbols and two-dimensional diagrams, whereas our informal
arguments are written with words and paragraphs. It is worthwhile to reflect on what is captured by the
model. Natural deduction is supposed to clarify the form and structure of our logical arguments, describe
the appropriate means of justifying a conclusion, and explain the sense in which the rules we use are valid.
Constructing natural deduction proofs can be confusing, but it is helpful to think about why it is confusing.
We could, for example, decide that natural deduction is not a good model for logical reasoning. Or we
might come to the conclusion that the features of natural deduction that make it confusing tell us something
interesting about ordinary arguments.
In the “official” description, natural deduction proofs are constructed by putting smaller proofs together to
obtain bigger ones. To prove A ∧ B → B ∧ A, we start with the hypothesis A ∧ B. Then we construct,

separately, the following two proofs:

A∧B A∧B
B A
Then we use these two proofs to construct the following one:
A∧B A∧B
B A
B∧A
Finally, we apply the implies-introduction rule to this proof to cancel the hypothesis and obtain the desired
conclusion:
1 1
A∧B A∧B
B A
B∧A 1
A∧B →B∧A
The process is similar to what happens in an informal argument, where we start with some hypotheses, and
work forward towards a conclusion.
Suppose Susan is tall and John is happy.

Then, in particular, John is happy.
Also, Susan is tall.
So John is happy and Susan is tall.
Therefore we have shown that if Susan is tall and John is happy, then John is happy and Susan is tall.
However, when we read natural deduction proofs, we often read them backwards. First, we look at the
bottom to see what is being proved. Then we consider the rule that is used to prove it, and see what
premises the rule demands. Then we look to see how those claims are proved, and so on. Similarly, when
we construct a natural deduction proof, we typically work backwards as well: we start with the claim we are
trying to prove, put that at the bottom, and look for rules to apply.
At times that process breaks down. Suppose we are left with a goal that is a single propositional variable,
A. There are no introduction rules that can be applied, so, unless A is a hypothesis, it has to come from
an elimination rule. But that underspecifies the problem: perhaps the A comes from applying the and
elimination rule to A ∧ B, or from applying the or elimination rule to C and C → A. At that point, we look
to the hypotheses, and start working forwards. If, for example, our hypotheses are C and C → A ∧ B, we
would then work forward to obtain A ∧ B and A.
There is thus a general heuristic for proving theorems in natural deduction:
1. Start by working backwards from the conclusion, using the introduction rules. For example, if you are
trying to prove a statement of the form A → B, add A to your list of hypotheses and try to derive B.
If you are trying to prove a statement of the form A ∧ B, use the and-introduction rule to reduce your
task to proving A, and then proving B.
2. When you have run out things to do in the first step, use elimination rules to work forwards. If you
have hypotheses A → B and A, apply modus ponens to derive B. If you have a hypothesis A ∨ B,
use-or elimination to split on cases, considering A in one case and B in the other.
In Chapter 5 we will add one more element to this list: if all else fails, try a proof by contradiction.
The tension between forward and backward reasoning is found in informal arguments as well, in mathematics
and elsewhere. When we prove a theorem, we typically reason forward, using assumptions, hypotheses,
definitions, and background knowledge. But we also keep the goal in mind, and that helps us make sense of
the forward steps.
3.3. Forward and Backward Reasoning 21

When we turn to interactive theorem proving, we will see that Lean has mechanisms to support both forward
and backward reasoning. These form a bridge between informal styles of argumentation and the natural
deduction model, and thereby provide a clearer picture of what is going on.
Another confusing feature of natural deduction proofs is that every hypothesis has a scope, which is to say,
there are only certain points in the proof where an assumption is available for use. Of course, this is also a
feature of informal mathematical arguments. Suppose a paragraph begins “Let x be any number less than
100,” argues that x has at most five prime factors, and concludes “thus we have shown that every number
less than 100 has at most five factors.” The reference “x”, and the assumption that it is less than 100, is
only active within the scope of the paragraph. If the next paragraph begins with the phrase “Now suppose
x is any number greater than 100,” then, of course, the assumption that x is less than 100 no longer applies.
In natural deduction, a hypothesis is available from the point where it is assumed until the point where it
is canceled. We will see that interactive theorem proving languages also have mechanisms to determine the
scope of references and hypotheses, and that these, too, shed light on scoping issues in informal mathematics.
3.4 Reasoning by Cases
The rule for eliminating a disjunction is confusing, but we can make sense of it with an example. Consider
the following informal argument:
George is either at home or on campus.

If he is at home, he is studying.
If he is on campus, he is with his friends.
Therefore, George is either studying or with his friends.
Let A be the statement that George is at home, let B be the statement that George is on campus, let C be
the statement that George is studying, and let D be the statement the George is with his friends. Then the
argument above has the following pattern: from A ∨ B, A → C, and B → D, conclude C ∨ D. In natural
deduction, we can not get away with drawing this conclusion in a single step, but it does not take too much
work to flesh it out into a proper proof. Informally, we have to argue as follows.
Georges is either at home or on campus.

Case 1: Suppose he is at home. We know that if he is at home, then he is studying. So, in this
case, he is studying. Therefore, in this case, he is either studying or with his friends.
Case 2: Suppose he is on campus. We know that if he is on campus, then he is with his friends.
So, in this case, he is with his friends. Therefore, in this case, he is either studying or with his
friends.
Either way, George is either studying or with his friends.
The natural deduction proof looks as follows:

1 1
A→C A B→D B
C D
A∨B C ∨D C ∨D 1
C ∨D

You should think about how the structure of this proof reflects the informal case-based argument above it.
For another example, here is a proof of A ∧ (B ∨ C) → (A ∧ B) ∨ (A ∧ C):
2 2
A ∧ (B ∨ C) A ∧ (B ∨ C)
1 1
2 A B A C
A ∧ (B ∨ C) A∧B A∧C
B∨C (A ∧ B) ∨ (A ∧ C) (A ∧ B) ∨ (A ∧ C)
1
(A ∧ B) ∨ (A ∧ C)
2
(A ∧ (B ∨ C)) → ((A ∧ B) ∨ (A ∧ C))
3.5 Some Logical Identities
Two propositional formulas, A and B, are said to be logically equivalent if A ↔ B is provable. Logical
equivalences are similar to identities like x + y = y + x that occur in algebra. In particular, one can show
that if two formulas are equivalent, then one can substitute one for the other in any formula, and the results
will also be equivalent. (Some proof systems take this to be a basic rule, and interactive theorem provers
can accommodate it, but we will not take it to be a fundamental rule of natural deduction.)
For reference, the following list contains some commonly used propositional equivalences, along with some
noteworthy formulas. Think about why, intuitively, these formulas should be true.
1. Commutativity of ∧: A ∧ B ↔ B ∧ A
2. Commutativity of ∨: A ∨ B ↔ B ∨ A
3. Associativity of ∧: (A ∧ B) ∧ C ↔ A ∧ (B ∧ C)
4. Associativity of ∨ (A ∨ B) ∨ C ↔ A ∨ (B ∨ C)
5. Distributivity of ∧ over ∨: A ∧ (B ∨ C) ↔ (A ∧ B) ∨ (A ∧ C)
6. Distributivity of ∨ over ∧: A ∨ (B ∧ C) ↔ (A ∨ B) ∧ (A ∨ C)
7. (A → (B → C)) ↔ (A ∧ B → C).
8. (A → B) → ((B → C) → (A → C))
9. ((A ∨ B) → C) ↔ (A → C) ∧ (B → C)
10. ¬(A ∨ B) ↔ ¬A ∧ ¬B
11. ¬(A ∧ B) ↔ ¬A ∨ ¬B
12. ¬(A ∧ ¬A)
13. ¬(A → B) ↔ A ∧ ¬B
14. ¬A → (A → B)
15. (¬A ∨ B) ↔ (A → B)
16. A ∨ ⊥ ↔ A
17. A ∧ ⊥ ↔ ⊥
18. A ∨ ¬A
19. ¬(A ↔ ¬A)
20. (A → B) ↔ (¬B → ¬A)
21. (A → C ∨ D) → ((A → C) ∨ (A → D))
3.5. Some Logical Identities 23

22. (((A → B) → A) → A)
All of these can be derived in natural deduction using the fundamental rules listed in Section 3.1. But some
of them require the use of the reductio ad absurdum rule, or proof by contradiction, which we have not yet
discussed in detail. We will discuss the use of this rule, and other patterns of classical logic, in the Chapter
5.
3.6 Exercises
When constructing proofs in natural deduction, use only the list of rules given in Section 3.1.
1. Give a natural deduction proof of ¬(A ∧ B) → (A → ¬B).
2. Give a natural deduction proof of (A → C) ∧ (B → ¬C) → ¬(A ∧ B).
3. Give a natural deduction proof of (A ∧ B) → ((A → C) → ¬(B → ¬C)).
4. Take another look at Exercise 3 in the last chapter. Using propositional variables A, B, and C for
“Alan likes kangaroos,” “Betty likes frogs” and “Carl likes hamsters,” respectively, express the three
hypotheses in the previous problem as symbolic formulas, and then derive a contradiction from them
in natural deduction.
5. Give a natural deduction proof of A ∨ B → B ∨ A.
6. Give a natural deduction proof of ¬A ∧ ¬B → ¬(A ∨ B)
7. Give a natural deduction proof of ¬(A ∧ B) from ¬A ∨ ¬B. (You do not need to use proof by
contradiction.)
8. Give a natural deduction proof of ¬(A ↔ ¬A).
9. Give a natural deduction proof of (¬A ↔ ¬B) from hypothesis A ↔ B.

CHAPTER
FOUR
PROPOSITIONAL LOGIC IN LEAN
In this chapter, you will learn how to write proofs in Lean. We will start with a purely mechanical translation
that will enable you to represent any natural deduction proof in Lean. We will see, however, that such a
style of writing proofs is not very intuitive, nor does it yield very readable proofs. It also does not scale well.
We will then consider some mechanisms that Lean offers that support a more forward-directed style of
argumentation. Since these proofs look more like informal proofs but can be directly translated to natural
deduction, they will help us understand the relationship between the two.
4.1 Expressions for Propositions and Proofs
At its core, Lean is what is known as a type checker. This means that we can write expressions and ask
the system to check that they are well formed, and also ask the system to tell us what type of object they
denote. Try this:
variables A B C : Prop
#check A ∧ ¬ B → C
In the online version of this text, you can press the “try it!” button to copy the example to the editor
window, press the “play” button, and then hover over the markers on the left to read the messages.
In the example, we declare three variables ranging over propositions, and ask Lean to check the expression
A ∧ ¬ B → C. The output of the #check command is A ∧ ¬ B → C : Prop, which asserts that A ∧ ¬
B → C is of type Prop. In Lean, every well-formed expression has a type.
The logical connectives are rendered in unicode. The following chart shows you how you can type these
symbols in the editor, and also provides ascii equivalents, for the purists among you.
Unicode Ascii Emacs

true
false
¬ not \not, \neg
∧ / \and
∨ / \or
→ -> \to, \r, \imp
↔ <-> \iff, \lr
∀ forall \all
∃ exists \ex
λ fun \lam, \fun
̸ = ~= \ne
25
So far, we have only talked about the first seven items on the list. We will discuss the quantifiers, lambda,
and equality later. Try typing some expressions and checking them on your own. You should try changing
one of the variables in the example above to D, or inserting a nonsense symbol into the expression, and take
a look at the error message that Lean returns.
In addition to declaring variables, if P is any expression of type Prop, we can declare the hypothesis that P
is true:
variables A B : Prop
variable h : A ∧ ¬ B
#check h
Formally, what is going on is that any proposition can be viewed as a type, namely, the type of proofs of that
proposition. A hypothesis, or premise, is just a variable of that type. Building proofs is then a matter of
writing down expressions of the write type. For example, if P is any expression of type A ∧ B, then and.left
P is an expression of type A, and and.right P is an expression of type B. In other words, if P is a proof of A
∧ B, and and.left P is a name for the proof you get by applying the left elimination rule for and:
..
.
P
..
.
A∧B
A
Similarly, and.right P is the proof of B you get by applying the right elimination rule. So, continuing the
example above, we can write
#check and.left h
#check and.right h
The two expressions represent, respectively, these two proofs:

h h
A ∧ ¬B A ∧ ¬B
A ¬B
Notice that in this way of representing natural deduction proofs, there are no “free floating” hypotheses.
Every hypothesis has a label. In Lean, we will typically use expressions like h, h1, h2, … to label hypotheses,
but you can use any identifier you want.
If h1 is a proof of A and h2 is a proof of B, then and.intro h1 h2 is a proof of A ∧ B. So we can continue
the example above:
#check and.intro (and.right h) (and.left h)
This corresponds to the following proof:

h h
A ∧ ¬B A ∧ ¬B
¬B A
¬B ∧ A
What about implication? The elimination rule is easy: if P1 is a proof of A → B and P2 is a proof of A then
P1 P2 is a proof of B. Notice that we do not even need to name the rule: you just write P1 followed by P2 ,
26 Chapter 4. Propositional Logic in Lean

as though you are applying the first to the second. If P1 and P2 are compound expressions, put parentheses
around them to make it clear where each one begins and ends.
variables A B C D : Prop
variable h1 : A → (B → C)
variable h2 : D → A
variable h3 : D
variable h4 : B
#check h2 h3
#check h1 (h2 h3)
#check (h1 (h2 h3)) h4
Lean adopts the convention that applications associate to the left, so that an expression h1 h2 h3 is inter-
preted as (h1 h2) h3. Implications associate to the right, so that A → B → C is interpreted as A → (B →
C). This may seem funny, but it is a convenient way to represent implications that take multiple hypotheses,
since an expression A → B → C → D → E means that E follows from A, B, C, and D. So the example above
could be written as follows:
variable h1 : A → B → C
variable h2 : D → A
variable h3 : D
variable h4 : B
#check h2 h3
#check h1 (h2 h3)
#check h1 (h2 h3) h4
Notice that parentheses are still needed in the expression h1 (h2 h3).
The implication introduction rule is the tricky one, because it can cancel a hypothesis. In terms of Lean
expressions, the rule translates as follows. Suppose A and B have type Prop, and, assuming h is the premise
that A holds, P is proof of B, possibly involving h. Then the expression assume h : A, P is a proof of A →
B. For example, we can construct a proof of A → A ∧ A as follows:
variable A : Prop
#check (assume h : A, and.intro h h)
Notice that we no longer have to declare A as a premise. The word assume makes the premise local to the
expression in parentheses, and after the assumption is made, we can refer to h. Given the assumption h :
A, and.intro h h is a proof of A ∧ A, and so the expression assume h : A, and.intro h h is a proof of
A → A ∧ A. In this case, we could leave out the parentheses because the expression is unambiguous:
variable A : Prop
#check assume h : A, and.intro h h
Above, we proved ¬ B ∧ A from the premise A ∧ ¬ B. We can instead obtain a proof of A ∧ ¬ B → ¬ B

∧ A as follows:
#check (assume h : A ∧ ¬ B, and.intro (and.right h) (and.left h))
All we did was move the premise into a local assume.
4.1. Expressions for Propositions and Proofs 27

(By the way, the assume command is just alternative syntax for the lambda symbol, so we could also have
written this:
#check (λ h : A ∧ ¬ B, and.intro (and.right h) (and.left h))
You will learn more about the lambda symbol later.)
4.2 More commands
Let us introduce a new Lean command, example. This command tells Lean that you are about to prove a
theorem, or, more generally, write down an expression of the given type. It should then be followed by the
proof or expression itself.
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h : A ∧ ¬ B,
and.intro (and.right h) (and.left h)
When given this command, Lean checks the expression after the := and makes sure it has the right type. If
so, it accepts the expression as a valid proof. If not, it raises and error.
Because the example command provides information as to the type of the expression that follows (in this
case, the proposition being proved), it sometimes enables us to omit other information. For example, we can
leave off the type of the assumption:
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h,
Because Lean knows we are trying to prove an implication with premise A ∧ ¬ B, it can infer that when we
write assume h, the identifier h labels the assumption A ∧ ¬ B.
We can also go in the other direction, and provide the system with more information, with the word show.
If A is a proposition and P is a proof, the expression “show A, from P” means the same thing as P alone,
but it signals the intention that P is a proof of A. When Lean checks this expression, it confirms that P really
is a proof of A, before parsing the expression surrounding it. So, in our example, we could also write:
example : A ∧ ¬ B → ¬ B ∧ A :=
show ¬ B ∧ A, from and.intro (and.right h) (and.left h)
We could even annotate the smaller expressions and.right h and and.left h, as follows:
example : A ∧ ¬ B → ¬ B ∧ A :=
show ¬ B ∧ A, from and.intro
(show ¬ B, from and.right h)
(show A, from and.left h)

This is a good place to mention that Lean generally ignores whitespace, like indentation and returns. We
could have written the entire example on a single line. In general, we will adopt conventions for line breaks
and indentation that shows the structure of a proof and makes it easier to read.
Although in the examples above the show commands were not necessary, there are a number of good reasons
to use it. First, and perhaps most importantly, it makes the proofs easier for us humans to read. Second, it
makes the proofs easier to write: if you make a mistake in a proof, it is easier for Lean to figure out where
you went wrong and provide a meaningful error message if you make your intentions clear. Finally, proving
information in the show clause often makes it possible for you to omit information in other places, since Lean
can infer that information from your stated intentions.
There are notational variants. Rather than declare variables and premises beforehand, you can also present
them as “arguments” to the example, followed by a colon:
example (A B : Prop) : A ∧ ¬ B → ¬ B ∧ A :=
show ¬ B ∧ A, from and.intro (and.right h) (and.left h)
There are two more tricks that can help you write proofs in Lean. The first is using sorry, which is a magical
term in Lean which provides a proof of anything at all. It is also known as “cheating.” But cheating can
help you construct legitimate proofs incrementally: if Lean accepts a proof with sorry’s, the parts of the
proof you have written so far have passed Lean’s checks for correctness. All you need to do is replace each
sorry with a real proof to complete the task.
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h, sorry
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h, and.intro sorry sorry
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h, and.intro (and.right h) sorry
example : A ∧ ¬ B → ¬ B ∧ A :=
assume h, and.intro (and.right h) (and.left h)
The second trick is the use of placeholders, represented by the underscore symbol. When you write an
underscore in an expression, you are asking the system to try to fill in the value for you. This falls short of
calling full-blown automation to prove a theorem; rather, you are asking Lean to infer the value from the
context. If you use an underscore where a proof should be, Lean typically will not fill in the proof, but it will
give you an error message that tells you what is missing. This will help you write proof terms incrementally,
in a backward-driven fashion. In the example above, try replacing each sorry by an underscore, _, and take
a look at the resulting error messages. In each case, the error tells you what needs to be filled in, and the
variables and hypotheses that are available to you at that stage.
One more tip: if you want to delimit the scope of variables or premises introduced with the variables
command, put them in a block that begins with the word section and ends with the word end. We will use
this mechanism below.
4.3 Building Natural Deduction Proofs
In this section, we describe a mechanical translation from natural deduction proofs, by giving a translation
for each natural deduction rule. We have already seen some of the correspondences, but we repeat them all
here, for completeness.
4.3. Building Natural Deduction Proofs 29

4.3.1 Implication
We have already explained that implication introduction is implemented with assume, and implication
elimination is written as application.
example : A → B :=
assume h : A,
show B, from sorry
section
variable h1 : A → B
variable h2 : A
example : B := h1 h2
end
Since every example begins by declaring the necessary propositional variables, we will henceforth suppress
that declaration in the text.
4.3.2 Conjunction
We have already seen that and introduction is implemented with and.intro, and the elimination rules are
and.left and and.right.
section
variables (h1 : A) (h2 : B)
example : A ∧ B := and.intro h1 h2
end
section
variable h : A ∧ B
example : A := and.left h
example : B := and.right h
end
4.3.3 Disjunction
The or introduction rules are given by or.inl and or.inr.
section
variable h : A
example : A ∨ B := or.inl h
end
section
variable h : B
example : A ∨ B := or.inr h
end

The elimination rule is the tricky one. To prove C from A ∨ B, you need three arguments: a proof h of A ∨
B, a proof of C from A, and a proof of C from B. Using line breaks and indentation to highlight the structure
as a proof by cases, we can write it with the following form:
section
variable h : A ∨ B
example : C :=
or.elim h
(assume h1 : A,
show C, from sorry)
(assume h1 : B,
show C, from sorry)
end
Notice that we can reuse the label h1 in each branch, since, conceptually, the two branches are disjoint.
4.3.4 Negation
Internally, negation ¬ A is defined by A → false, which you can think of as saying that A implies something
impossible. The rules for negation are therefore similar to the rules for implication. To prove ¬ A, assume
A and derive a contradiction.
section
example : ¬ A :=
assume h : A,
show false, from sorry
end
If you have proved a negation ¬ A, you can get a contradiction by applying it to a proof of A.
section
variable h1 : ¬ A
variable h2 : A
example : false := h1 h2
end
4.3.5 Truth and falsity
The ex falso rule is called false.elim:
section
variable h : false
example : A := false.elim h
end
There isn’t much to say about true beyond the fact that it is trivially true:
example : true := trivial

4.3.6 Bi-implication
The introduction rule for “if and only if” is iff.intro.

example : A ↔ B :=
iff.intro
(assume h : A,
show B, from sorry)
(assume h : B,
show A, from sorry)
As usual, we have chosen indentation to make the structure clear. Notice that the same label, h, can be
used on both branches, with a different meaning in each, because the scope of an assume is limited to the
expression in which it appears.
The elimination rules are iff.elim_left and iff.elim_right:
section
variable h1 : A ↔ B
variable h2 : A
example : B := iff.elim_left h1 h2
end
section
variable h1 : A ↔ B
variable h2 : B
example : A := iff.elim_right h1 h2
end
Lean recognizes the abbreviation iff.mp for iff.and_elim_left, where “mp” stands for “modus ponens”.
Similarly, you can use iff.mpr, for “modus ponens reverse”, instead of iff.and_elim_right.
4.3.7 Reductio ad absurdum (proof by contradiction)
Finally, there is the rule for proof by contradiction, which we will discuss in greater detail in Chapter 5. It
is included for completeness here.
The rule is called by_contradiction. It has one argument, which is a proof of false from ¬ A. To use the
rule, you have to ask Lean to allow classical reasoning, by writing open classical. You can do this at the
beginning of the file, or any time before using it. It you say open classical in a section, it will remain in
scope for that section.
section
open classical
example : A :=
by_contradiction
(assume h : ¬ A,
show false, from sorry)
end
4.3.8 Examples
In the last chapter, we constructed the following proof A → C from A → B and B → C:

1
A A→B
B B→C
C 1
A→C
We can model this in Lean as follows:
variable h2 : B → C
example : A → C :=
assume h : A,
show C, from h2 (h1 h)
Notice that the hypotheses in the natural deduction proof that are not canceled are declared as variables in
the Lean version.
We also constructed the following proof:
1
2
A∧B
A → (B → C) A A∧B
1
B→C B
C 1
A∧B →C 2
(A → (B → C)) → (A ∧ B → C)
Here is how it is written in Lean:
example (A B C : Prop) : (A → (B → C)) → (A ∧ B → C) :=
assume h1 : A → (B → C),
assume h2 : A ∧ B,
show C, from h1 (and.left h2) (and.right h2)
This works because and.left h2 is a proof of A, and and.right h2 is a proof of B.

Finally, we constructed the following proof of A ∧ (B ∨ C) → (A ∧ B) ∨ (A ∧ C):
2 2
A ∧ (B ∨ C) A ∧ (B ∨ C)
1 1
2 A B A C
A ∧ (B ∨ C) A∧B A∧C
B∨C (A ∧ B) ∨ (A ∧ C) (A ∧ B) ∨ (A ∧ C)
1
(A ∧ B) ∨ (A ∧ C)
2
(A ∧ (B ∨ C)) → ((A ∧ B) ∨ (A ∧ C))
Here is a version in Lean:
example (A B C : Prop) : A ∧ (B ∨ C) → (A ∧ B) ∨ (A ∧ C) :=
assume h1 : A ∧ (B ∨ C),
or.elim (and.right h1)
(assume h2 : B,
show (A ∧ B) ∨ (A ∧ C),
from or.inl (and.intro (and.left h1) h2))
(assume h2 : C,
show (A ∧ B) ∨ (A ∧ C),
from or.inr (and.intro (and.left h1) h2))
In fact, bearing in mind that assume is alternative syntax for the symbol λ, and that Lean can often infer
the type of an assumption, we can make the proof remarkably brief:

λ h1, or.elim (and.right h1)
(λ h2, or.inl (and.intro (and.left h1) h2))
(λ h2, or.inr (and.intro (and.left h1) h2))
The proof is cryptic, though. Using such a style makes proofs hard to write, read, understand, maintain,
and debug. In the next section we will describe a remarkably simple device that makes it much easier to
understand what is going on.
4.4 Forward Reasoning
Lean supports forward reasoning by allowing you to write proofs using the have command.
variable h2 : B → C
example : A → C :=
assume h : A,
have h3 : B, from h1 h,
show C, from h2 h3
Writing a proof with have h : A, from P, ... h ... has the same effect as writing ... P .... This
have command checks that P is a proof of A, and then give you the label h to use in place of P. Thus the last
line of the previous proof can be thought of as abbreviating show C, from h2 (h1 h), since h3 abbreviates
h1 h. Such abbreviations can make a big difference, especially when the proof P is very long.
There are a number of advantages to using have. For one thing, it makes the proof more readable: the
example above states B explicitly as an auxiliary goal. It can also save repetition: h3 can be used repeatedly
after it is introduced, without duplicating the proof. Finally, it makes it easier to construct and debug the
proof: stating B as an auxiliary goal makes it easier for Lean to deliver an informative error message when
the goal is not properly met.
In the last section, we considered the following proof:

show C, from h1 (and.left h2) (and.right h2)
Using have, it can be written more perspicuously as follows:

have h3 : A, from and.left h2,
have h4 : B, from and.right h2,
show C, from h1 h3 h4
We can be even more verbose, and add another line:



have h5 : B → C, from h1 h3,
show C, from h5 h4
Adding more information doesn’t always make a proof more readable; when the individual expressions are
small and easy enough to understand, spelling them out in detail can introduce clutter. As you learn to use
Lean, you will have to develop your own style, and use your judgment to decide which steps to make explicit.
Here is how some of the basic inferences look, when expanded with have. In the and-introduction rule, it is
a matter showing each conjunct first, and then putting them together:
example (A B : Prop) : A ∧ B → B ∧ A :=
show B ∧ A, from and.intro h3 h2
Compare that with this version, which instead states first that we will use the and.intro rule, and then
makes the two resulting goals explicit:
show B ∧ A, from
and.intro
(show B, from and.right h1)
(show A, from and.left h1)
Once again, at issue is only readability. Lean does just fine with the following short version:
λ h, and.intro (and.right h) (and.left h)
When using the or-elimination rule, it is often clearest to state the relevant disjunction explicitly:
example (A B C : Prop) : C :=
have h : A ∨ B, from sorry,
show C, from or.elim h
(assume h1 : A,
show C, from sorry)
(assume h2 : B,
show C, from sorry)
Here is a have-structured presentation of an example from the previous section:
assume h1 : A ∧ (B ∨ C),
have h3 : B ∨ C, from and.right h1,
show (A ∧ B) ∨ (A ∧ C), from
or.elim h3
(assume h4 : B,
have h5 : A ∧ B, from and.intro h2 h4,
show (A ∧ B) ∨ (A ∧ C), from or.inl h5)
(assume h4 : C,
have h5 : A ∧ C, from and.intro h2 h4,
show (A ∧ B) ∨ (A ∧ C), from or.inr h5)
4.4. Forward Reasoning 35

4.5 Definitions and Theorems
Lean allows us to name definitions and theorems for later use. For example, here is a definition of a new
“connective”:
def triple_and (A B C : Prop) : Prop :=

A ∧ (B ∧ C)
As with the example command, it does not matter whether the arguments A, B, and C are declared before-
hand with the variables command, or with the definition itself. We can then apply the definition to any
expressions:
variables D E F G : Prop
#check triple_and (D ∨ E) (¬ F → G) (¬ D)
Later, we will see more interesting examples of definitions, like the following function from natural numbers
to natural numbers, which doubles its input:
def double (n : N) : N := n + n
What is more interesting right now is that Lean also allows us to name theorems, and use them later, as
rules of inference. For example, consider the following theorem:
theorem and_commute (A B : Prop) : A ∧ B → B ∧ A :=

Once we have defined it, we can use it freely:
variables C D E : Prop
variable h1 : C ∧ ¬ D
variable h2 : ¬ D ∧ C → E
example : E := h2 (and_commute C (¬ D) h1)
It is annoying in this example that we have to give the arguments C and ¬ D explicitly, because they are
implicit in h1. In fact, Lean allows us to tell this to Lean in the definition of and_commute:
theorem and_commute {A B : Prop} : A ∧ B → B ∧ A :=

here the squiggly braces indicate that the arguments A and B are implicit, which is to say, Lean should infer
them from the context when the theorem is used. We can then write the following instead:
variables C D E : Prop
variable h1 : C ∧ ¬ D
variable h2 : ¬ D ∧ C → E
example : E := h2 (and_commute h1)
Indeed, Lean’s library has a theorem, and_comm, defined in exactly this way.
By the way, we could avoid the assume step in the proof of and_comm by making the hypothesis into an
argument:
theorem and_commute {A B : Prop} (h : A ∧ B) : B ∧ A :=


The two definitions yield the same result.

Definitions and theorems are important in mathematics; they allow us to build up complex theories from
fundamental principles. Lean also accepts the word lemma instead of theorem.
What is interesting is that in interactive theorem proving, we can even define familiar patterns of inference.
For example, all of the following inferences were mentioned in the last chapter:
namespace hide
variables {A B : Prop}
theorem or_resolve_left (h1 : A ∨ B) (h2 : ¬ A) : B :=

or.elim h1
(assume h3 : A, show B, from false.elim (h2 h3))
(assume h3 : B, show B, from h3)
theorem or_resolve_right (h1 : A ∨ B) (h2 : ¬ B) : A :=

or.elim h1
(assume h3 : A, show A, from h3)
(assume h3 : B, show A, from false.elim (h2 h3))
theorem absurd (h1 : ¬ A) (h2 : A) : B :=

false.elim (h1 h2)
end hide
In fact, Lean’s library defines or.resolve_left, or.resolve_right, and absurd. We used the namespace
command to avoid naming conflicts, which would have raised an error.
When we ask you to prove basic facts from propositional logic in Lean, as with propositional logic, our goal
is to have you learn how to use Lean’s primitives. As a result, for those exercises, you should not use facts
from the library. As we move towards real mathematics, however, you can use facts from the library more
freely.
4.6 Additional Syntax
In this section, we describe some extra syntactic features of Lean, for power users. The syntactic gadgets
are often convenient, and sometimes make proofs look prettier.
For one thing, you can use subscripted numbers with a backslash. For example, you can write h1 by typing
h\1. The labels are irrelevant to Lean, so the difference is only cosmetic.
Another feature is that you can omit the label in an assume statement, providing an “anonymous” hypothesis.
You can then refer back to the last anonymous assumption using the keyword this:
example : A → A ∨ B :=
assume : A,
show A ∨ B, from or.inl this
Alternatively, you can refer back to unlabeled assumptions by putting them in French quotes:
example : A → B → A ∧ B :=
assume : A,
assume : B,
show A ∧ B, from and.intro ‹A› ‹B›
4.6. Additional Syntax 37

You can also use the word have without giving a label, and refer back to them using the same conventions.
Here is an example that uses these features:
theorem my_theorem {A B C : Prop} :
A ∧ (B ∨ C) → (A ∧ B) ∨ (A ∧ C) :=
assume h : A ∧ (B ∨ C),
have A, from and.left h,
have B ∨ C, from and.right h,
show (A ∧ B) ∨ (A ∧ C), from
or.elim ‹B ∨ C›
(assume : B,
have A ∧ B, from and.intro ‹A› ‹B›,
show (A ∧ B) ∨ (A ∧ C), from or.inl this)
(assume : C,
have A ∧ C, from and.intro ‹A› ‹C›,
show (A ∧ B) ∨ (A ∧ C), from or.inr this)
Another trick is that you can write h.left and h.right instead of and.left h and and.right h whenever
h is a conjunction, and you can write ⟨h1, h2⟩ instead of and.intro h1 h2 whenever Lean can figure out
that a conjunction is what you are trying to prove. With these conventions, you can write the following:
assume h : A ∧ B,
show B ∧ A, from ⟨h.right, h.left⟩
This is nothing more than shorthand for the following:

assume h : A ∧ B,
show B ∧ A, from and.intro (and.right h) (and.left h)
Even more concisely, you can write this:

assume h, ⟨h.right, h.left⟩
You can even take apart a conjunction with an assume, so that this works:
assume ⟨h1 , h2 ⟩, ⟨h2 , h1 ⟩
Similarly, if h is a biconditional, you can write h.mp and h.mpr instead of iff.mp h and iff.mpr h, and
you can write ⟨h1, h2⟩ instead of iff.intro h1 h2. As a result, Lean understands these proofs:
example (A B : Prop) : B ∧ (A ↔ B) → A :=
assume ⟨hB, hAB⟩,
hAB.mpr hB
example (A B : Prop) : A ∧ B ↔ B ∧ A :=
⟨assume ⟨h1 , h2 ⟩, ⟨h2 , h1 ⟩, assume ⟨h1 , h2 ⟩, ⟨h2 , h1 ⟩⟩
Finally, you can add comments to your proofs in two ways. First, any text after a double-dash -- until
the end of a line is ignored by the Lean processor. Second, any text between /- and -/ denotes a block
comment, and is also ignored. You can nest block comments.
/- T his is a block comment.
It can fill multiple lines. -/

example (A : Prop) : A → A :=
assume : A, -- assume the antecedent
show A, from this -- use it to establish the conclusion
4.7 Exercises
Prove the following in Lean:
example : A ∧ (A → B) → B :=
sorry
example : A → ¬ (¬ A ∧ B) :=
sorry
example : ¬ (A ∧ B) → (A → ¬ B) :=
sorry
example (h1 : A ∨ B) (h2 : A → C) (h3 : B → D) : C ∨ D :=

sorry
example (h : ¬ A ∧ ¬ B) : ¬ (A ∨ B) :=
sorry
example : ¬ (A ↔ ¬ A) :=
sorry
4.7. Exercises 39

CHAPTER
FIVE
CLASSICAL REASONING
If we take all the rules of propositional logic we have seen so far and exclude reductio ad absurdum, or proof
by contradiction, we have what is known as intuitionistic logic. In intuitionistic logic, it is possible to view
proofs in computational terms: a proof of A ∧ B is a proof of A paired with a proof of B, a proof of A → B
is a procedure which transforms evidence for A into evidence for B, and a proof of A ∨ B is a proof of one or
the other, tagged so that we know which is the case. The ex falso rule makes sense only because we expect
that there is no proof of falsity; it is like the empty data type.
Proof by contradiction does not fit it well with this world view: from a proof of a contradiction from ¬A,
we are supposed to magically produce a proof of A. We will see that with proof by contradiction, we can
prove the law of the excluded middle, A ∨ ¬A. From a computational perspective, this would say that we
can ways decide whether or not A is true.
Classical reasoning does introduce a number of principles into logic, however, that can be used to simplify
reasoning. In this chapter, we will consider these principles, and see how they follow from the basic rules.
5.1 Proof by Contradiction
Remember that in natural deduction, proof by contradiction is expressed by the following pattern:
1
¬A
..
.
⊥ 1
A
The assumption ¬A is canceled at the final inference.
In Lean, the inference is named by_contradiction, and since it is a classical rule, we have to use the
command open classical before it is available. Once we do so, the pattern of inference is expressed as
follows:
open classical
variable (A : Prop)
example : A :=
by_contradiction
(assume h : ¬ A,
show false, from sorry)
One of them most important consequences of this rule is the law of the excluded middle. In mathematical
arguments, one often splits a proof into two cases, assuming first A and then ¬A. Using the elimination rule
41
for disjunction, this is equivalent to using A ∨ ¬A, a classical principle known as the law of the excluded
middle. Here is a proof of this, in natural deduction, using a proof by contradiction:
1
2
A
¬(A ∨ ¬A) A ∨ ¬A
⊥ 1
¬A 1
A ∨ ¬A ¬(A ∨ ¬A)
⊥ 1
A ∨ ¬A
Here is the same proof rendered in Lean:
open classical
variable (A : Prop)
example : A ∨ ¬ A :=
by_contradiction
(assume h1 : ¬ (A ∨ ¬ A),
have h2 : ¬ A, from
assume h3 : A,
have h4 : A ∨ ¬ A, from or.inl h3,
show false, from h1 h4,
have h5 : A ∨ ¬ A, from or.inr h2,
show false, from h1 h5)
The principle is known as the law of the excluded middle because it says that a proposition A is either true
or false; there is no middle ground. As a result, the theorem is named em in the Lean library. For any
proposition A, em A denotes a proof of A ∨ ¬ A, and you are free to use it any time classical is open:
open classical
example (A : Prop) : A ∨ ¬ A :=
or.elim (em A)
(assume : A, or.inl this)
(assume : ¬ A, or.inr this)
Or even more simply:
open classical
example (A : Prop) : A ∨ ¬ A :=
em A
In fact, we can go in the other direction, and use the law of the excluded middle to justify proof by contra-
diction. You are asked to do this in the exercises.
Proof by contradiction is also equivalent to the principle ¬¬A ↔ A. The implication from right to left holds
intuitionistically; the other implication is classical, and is known as double-negation elimination. Here is a
proof in natural deduction:
2 1 1 2
¬¬A ¬A ¬A A
⊥ 1
⊥ 1
A ¬¬A 2
¬¬A ↔ A
And here is the corresponding proof in Lean:
42 Chapter 5. Classical Reasoning

open classical
example (A : Prop) : ¬ ¬ A ↔ A :=
iff.intro
(assume h1 : ¬ ¬ A,
show A, from by_contradiction
(assume h2 : ¬ A,
show false, from h1 h2))
(assume h1 : A,
show ¬ ¬ A, from assume h2 : ¬ A, h2 h1)
In the next section, we will derive a number of classical rules and equivalences. These are tricky to prove.
In general, to use classical reasoning in natural deduction, we need to extend the general heuristic presented
in Section 3.3 as follows:
1. First, work backwards from the conclusion, using the introduction rules.
2. When you have run out things to do in the first step, use elimination rules to work forwards.
3. If all else fails, use a proof by contradiction.
Sometimes a proof by contradiction is necessary, but when it isn’t, it can be less informative by a direct
proof. Suppose, for example, we want to prove A ∧ B ∧ C → D. In a direct proof, we assume A, B, and C,
and work towards D. Along the way, we will derive other consequences of A, B, and C, and these may be
useful in other contexts. If we use proof by contradition, on the other hand, we assume A, B, C, and ¬D,
and try to prove ⊥. In that case, we are working in an inconsistent context; any auxiliary results we may
obtain that way are subsumed by the fact that we ultimately ⊥ is a consequence of the hypotheses.
5.2 Some Classical Principles
We have already seen that A ∨ ¬A and ¬¬A ↔ A are two important theorems of classical propositional
logic. In this section we will provide some more theorems, rules, and equivalences. Some will be proved here,
but most will be left to you in the exercises. In ordinary mathematics, these are generally used without
comment. It is nice to know, however, that they can all be justified using the basic rules of classical natural
deduction.
If A → B is any implication, the assertion ¬B → ¬A is known as the contrapositive. Every implication
implies its contrapositive, and the other direction is true classically:
1
¬B → ¬A ¬B 2
¬A A
⊥ 1
B 2
A→B
Here is another example. Intuitively, asserting “if A then B” is equivalent to saying that it cannot be the
case that A is true and B is false. Classical reasoning is needed to get us from the second statement to the
first.
2 1
3
A ¬B
¬(A ∧ ¬B) A ∧ ¬B
⊥ 1
B 2
A→B 3
¬(A ∧ ¬B) → (A → B)
Here are the same proofs, rendered in Lean:
5.2. Some Classical Principles 43

open classical
variables (A B : Prop)
example (h : ¬ B → ¬ A) : A → B :=
assume h1 : A,
show B, from
by_contradiction
(assume h2 : ¬ B,
have h3 : ¬ A, from h h2,
show false, from h3 h1)
example (h : ¬ (A ∧ ¬ B)) : A → B :=
assume : A,
show B, from
by_contradiction
(assume : ¬ B,
have A ∧ ¬ B, from and.intro ‹A› this,
show false, from h this)
Notice that in the second example, we used an anonymous assume and an anonymous have. We used the
brackets \f< and \f> to write ‹A›, referring back to the first assumption. The first use of the word this
refers back to the assumption ¬ B, while the second one refers back to the have.
Knowing that we can prove the law of the excluded middle, it is convenient to use it in classical proofs. Here
is an example, with a proof of (A → B) ∨ (B → A):
1 2
¬B B
1
⊥
B A 2
A→B B→A
B ∨ ¬B (A → B) ∨ (B → A) (A → B) ∨ (B → A)
1
(A → B) ∨ (B → A)
Here is the corresponding proof in Lean:
open classical
variables (A B : Prop)
example : (A → B) ∨ (B → A) :=
or.elim (em B)
(assume h : B,
have A → B, from
assume : A,
show B, from h,
show (A → B) ∨ (B → A),
from or.inl this)
(assume h : ¬ B,
have B → A, from
assume : B,
have false, from h this,
show A, from false.elim this,
show (A → B) ∨ (B → A),
from or.inr this)

Using classical reasoning, implication can be rewritten in terms of disjunction and negation:
(A → B) ↔ ¬A ∨ B
The forward direction requires classical reasoning.

The following equivalences are known as De Morgan’s laws:
¬(A ∨ B) ↔ ¬A ∧ ¬B
¬(A ∧ B) ↔ ¬A ∨ ¬B
The forward direction of the second of these requires classical reasoning.

Using these identities, we can always push negations down to propositional variables. For example, we have
¬(¬A ∧ B → C) ↔ ¬(¬(¬A ∧ B) ∨ C)
↔ ¬¬(¬A ∧ B) ∧ ¬C
↔ ¬A ∧ B ∧ ¬C
A formula built up from ∧, ∨, and ¬ in which negations only occur at variables is said to be in negation
normal form.
In fact, using distributivity laws, one can go on to ensure that all the disjunctions are on the outside, so
that the formulas is a big or of and’s of propositional variables and negated propositional variables. Such a
formula is said to be in disjunctive normal form. Alternatively, all the and’s can be brought to the outside.
Such a formula is said to be in conjunctive normal form. An exercise below, however, shows that putting
formulas in disjunctive or conjunctive normal form can make them much longer.
5.3 Exercises
1. Show how to derive the proof-by-contradiction rule from the law of the excluded middle, using the
other rules of natural deduction. In other words, assume you have a proof of ⊥ from ¬A. Using A ∨ ¬A
as a hypothesis, but without using the rule RAA, show how you can go on to derive A.
2. Give a natural deduction proof of ¬(A ∧ B) from ¬A ∨ ¬B. (You do not need to use proof by
contradiction.)
3. Construct a natural deduction proof of ¬A ∨ ¬B from ¬(A ∧ B). You can do it as follows:
(a) First, prove ¬B, and hence ¬A ∨ ¬B, from ¬(A ∧ B) and A.
(b) Use this to construct a proof of ¬A, and hence ¬A ∨ ¬B, from ¬(A ∧ B) and ¬(¬A ∨ ¬B).
(c) Use this to construct a proof of a contradiction from ¬(A ∧ B) and ¬(¬A ∨ ¬B).
(d) Using proof by contradiction, this gives you a proof of ¬A ∨ ¬B from ¬(A ∧ B).
4. Give a natural deduction proof of ¬A ∨ B from A → B. You may use the law of the excluded middle.
5. Put (A ∨ B) ∧ (C ∨ D) ∧ (E ∨ F ) in disjunctive normal form, that is, write it as a big “or” of “and“‘s.
6. Prove ¬ (A ∧ B) → ¬ A ∨ ¬ B by replacing the sorry’s below by proofs.
5.3. Exercises 45
open classical
variables {A B C : Prop}
-- Prove ¬ (A ∧ B) → ¬ A ∨ ¬ B by replacing the sorry's below

-- by proofs.
lemma step1 (h1 : ¬ (A ∧ B)) (h2 : A) : ¬ A ∨ ¬ B :=

have ¬ B, from sorry,
show ¬ A ∨ ¬ B, from or.inr this
lemma step2 (h1 : ¬ (A ∧ B)) (h2 : ¬ (¬ A ∨ ¬ B)) : false :=

have ¬ A, from
assume : A,
have ¬ A ∨ ¬ B, from step1 h1 ‹A›,
show false, from h2 this,
show false, from sorry
theorem step3 (h : ¬ (A ∧ B)) : ¬ A ∨ ¬ B :=

by_contradiction
(assume h' : ¬ (¬ A ∨ ¬ B),
show false, from step2 h h')
7. Also do these:
open classical
variables {A B C : Prop}
example (h : ¬ B → ¬ A) : A → B :=
sorry
example (h : A → B) : ¬ A ∨ B :=
sorry

CHAPTER
SIX
SEMANTICS OF PROPOSITIONAL LOGIC
Classically, we think of propositional variables as ranging over statements that can be true or false. And,
intuitively, we think of a proof system as telling us what propositional formulas have to be true, no matter
what the variables stand for. For example, the fact that we can prove C from the hypotheses A, B, and
A ∧ B → C seems to tell us that whenever the hypotheses are true, then C has to be true as well.
Making sense of this involves stepping outside the system and giving an account of truth — more precisely,
the conditions under which a propositional formula is true. This is one of the things that symbolic logic was
designed to do, and the task belongs to the realm of semantics. Formulas and formal proofs are syntactic
notions, which is to say, they are represented by symbols and symbolic structures. Truth is a semantic
notion, in that it ascribes a type of meaning to certain formulas.
Syntactically, we were able to ask and answer questions like the following:
• Given a set of hypotheses, Γ, and a formula, A, can we derive A from Γ?
• What formulas can be derived from Γ?
• What hypotheses are needed to derive A?
The questions we consider semantically are different:
• Given an assignment of truth values to the propositional variables occurring in a formula A, is A true
or false?
• Is there any truth assignment that makes A true?
• Which are the truth assignments that make A true?
In this chapter, we will not provide a fully rigorous mathematical treatment of syntax and semantics. That
subject matter is appropriate to a more advanced and focused course on mathematical logic. But we will
discuss semantic issues in enough detail to give you a good sense of what it means to think semantically, as
well as a sense of how to make pragmatic use of semantic notions.
6.1 Truth Values and Assignments
The first notion we will need is that of a truth value. We have already seen two, namely, “true” and “false.”
We will use the symbols T and F to represent these in informal mathematics. These are the values that ⊤
and ⊥ are intended to denote in natural deduction, and true and false are intended to denote in Lean.
In this text, we will adopt a “classical” notion of truth, following our discussion in Section 5. This can be
understood in various ways, but, concretely, it comes down to this: we will assume that any proposition
is either true or false (but, of course, not both). This conception of truth is what underlies the law of the
excluded middle, A ∨ ¬A. Semantically, we read this sentence as saying “either A is true, or ¬A is true.”
Since, in our semantic interpretation, ¬A is true exactly when A is false, the law of the excluded middle says
that A is either true or false.
47
The next notion we will need is that of a truth assignment, which is simply a function that assigns a truth
value to each element of a propositional variables. In this section, we will distinguish between propositional
variables and arbitrary formulas by using letters P, Q, R, . . . for the former and A, B, C, . . . for the latter.
For example, the function v defined by
• v(P ) := T
• v(Q) := F
• v(R) := F
• v(S) := T
is a truth assignment for the set of variables {P, Q, R, S}.
Intuitively, a truth assignment describes a possible “state of the world.” Going back to the Malice and Alice
puzzle, let’s suppose the following letters are shorthand for the statements:
• P := Alice’s brother was the victim
• Q := Alice was the killer
• R := Alice was in the bar
In the world described by the solution to the puzzle, the first and third statements are true, and the second
is false. So our truth assignment gives the value T to P and R, and the value F to Q.
Once we have a truth assignment v to a set of propositional variables, we can extend it to a valuation function
v̄, which assigns a value of true or false to every propositional formula that depends only on these variables.
The function v̄ is defined recursively, which is to say, formulas are evaluated from the bottom up, so that
value assigned to a compound formula is determined by the values assigned to its components. Formally,
the function is defined as follows:
• v̄(⊤) = T
• v̄(⊥) = F
• v̄(ℓ) = v(ℓ), where ℓ is any propositional variable.
• v̄(¬A) = T if v̄(A) is F, and vice versa.
• v̄(A ∧ B) = T if v̄(A) and v̄(B) are both T, and F otherwise.
• v̄(A ∨ B) = T if at least one of v̄(A) and v̄(B) is T; otherwise F.
• v̄(A → B) = T if either v̄(B) is T or v̄(A) is F, and F otherwise. (Equivalently, v̄(A → B) = F if
v̄(A) is T and v̄(B) is F, and T otherwise.)
The rules for conjunction and disjunction are easy to understand. “A and B” is true exactly when A and B
are both true; “A or B” is true when at least one of A or B is true.
Understanding the rule for implication is trickier. People are often surprised to hear that any if-then
statement with a false hypothesis is supposed to be true. The statement “if I have two heads, then circles
are squares” may sound like it ought to be false, but by our reckoning, it comes out true. To make sense of
this, think about the difference between the two sentences:
• “If I have two heads, then circles are squares.”
• “If I had two heads, then circles would be squares.”
The second sentence is an example of a counterfactual implication. It asserts something about how the
world might change, if things were other than they actually are. Philosophers have studied counterfactuals
for centuries, but mathematical logic is concerned with the first sentence, a material implication. The
material implication asserts something about the way the world is right now, rather than the way it might
48 Chapter 6. Semantics of Propositional Logic

have been. Since it is false that I have two heads, the statement “if I have two heads, then circles are squares”
is true.
Why do we evaluate material implication in this way? Once again, let us consider the true sentence “every
natural number that is prime and greater than two is odd.” We can interpret this sentence as saying that
all of the (infinitely many) sentences in this list are true:
• if 0 is prime and greater than 2, then 0 is odd
• …
The first sentence on this list is a lot like our “two heads” example, since both the hypothesis and the
conclusion are false. But since it is an instance of a statement that is true in general, we are committed
to assigning it the value T. The second sentence is a different: the hypothesis is still false, but here the
conclusion is true. Together, these tell us that whenever the hypothesis is false, the conditional statement
should be true. The fourth sentence has a true hypothesis and a true conclusion. So from the second and
fourth sentences, we see that whenever the conclusion is true, the conditional should be true as well. Finally,
it seems clear that the sentence “if 3 is prime and greater than 2, then 3 is even” should not be true. This
pattern, where the hypothesis is true and the conclusion is false, is the only one for which the conditional
will be false.
Let us motivate the semantics for material implication another way, using the deductive rules described in
the last chapter. Notice that, if B is true, we can prove A → B without any assumptions about A.
B
A→B
This follows from the proper reading of the implication introduction rule: given B, one can always infer
A → B, and then cancel an assumption A, if there is one. If A was never used in the proof, the conclusion
is simply weaker than it needs to be. This inference is validated in Lean:
variable hB : B
assume hA : A,
show B, from hB
Similarly, if A is false, we can prove A → B without any assumptions about B:

1
¬A A
⊥ 1
A→B
In Lean:
variable hnA : ¬ A
assume hA : A,
show B, from false.elim (hnA hA)
Finally, if A is true and B is false, we can prove ¬(A → B):
6.1. Truth Values and Assignments 49

1
A→B A
¬B B
⊥ 1
¬(A → B)
Once again, in Lean:
variable hA : A
variable hnB : ¬B
example : ¬ (A → B) :=
assume h : A → B,
have hB : B, from h hA,
show false, from hnB hB
Now that we have defined the truth of any formula relative to a truth assignment, we can answer our first
semantic question: given an assignment v of truth values to the propositional variables occurring in some
formula φ, how do we determine whether or not φ is true? This amounts to evaluating v̄(φ), and the
recursive definition of φ gives a recipe: we evaluate the expressions occurring in φ from the bottom up,
starting with the propositional variables, and using the evaluation of an expression’s components to evaluate
the expression itself. For example, suppose our truth assignment v makes A and B true and C false. To
evaluate (B → C) ∨ (A ∧ B) under v, note that the expression B → C comes out false and the expression
A ∧ B comes out true. Since a disjunction “false or true” is true, the entire formula is true.
We can also go in the other direction: given a formula, we can attempt to find a truth assignment that will
make it true (or false). In fact, we can use Lean to evaluate formulas for us. In the example that follows,
you can assign any set of values to the proposition symbols A, B, C, D, and E. When you run Lean on this
input, the output of the eval statement is the value of the expression.
-- Define your truth assignment here
def A := tt
def B := ff
def C := tt
def D := tt
def E := ff
def test (p : Prop) [decidable p] : string :=

if p then "true" else "false"
#eval test ((A ∧ B) ∨ ¬ C)

#eval test (A → D)
#eval test (C → (D ∨ ¬E))
#eval test (¬(A ∧ B ∧ C ∧ D))
Try varying the truth assignments, to see what happens. You can add your own formulas to the end of the
input, and evaluate them as well. Try to find truth assignments that make each of the formulas tested above
evaluate to true. For an extra challenge, try finding a single truth assignment that makes them all true at
the same time.
6.2 Truth Tables
The second and third semantic questions we asked are a little trickier than the first. Given a formula A,
is there any truth assignment that makes A true? If so, which truth assignments make A true? Instead
of considering one particular truth assignment, these questions ask us to quantify over all possible truth
assignments.

Of course, the number of possible truth assignments depends on the number of propositional letters we’re
considering. Since each letter has two possible values, n letters will produce 2n possible truth assignments.
This number grows very quickly, so we’ll mostly look at smaller formulas here.
We’ll use something called a truth table to figure out when, if ever, a formula is true. On the left hand side
of the truth table, we’ll put all of the possible truth assignments for the present propositional letters. On
the right hand side, we’ll put the truth value of the entire formula under the corresponding assignment.
To begin with, truth tables can be used to concisely summarize the semantics of our logical connectives:
A B A∧B A B A∨B A B A→B
T T T T T T T T T
T F F T F T T F F
F T F F T T F T T
F F F F F F F F T
We will leave it to you to write the table for ¬A, as an easy exercise.
For compound formulas, the style is much the same. Sometimes it can be helpful to include intermediate
columns with the truth values of subformulas:
A B C A→B B→C (A → B) ∨ (B → C)
T T T T T T
T T F T F T
T F T F T T
T F F F T T
F T T T T T
F T F T F T
F F T T T T
F F F T T T
By writing out the truth table for a formula, we can glance at the rows and see which truth assignments
make the formula true. If all the entries in the final column are T, as in the above example, the formula is
said to be valid.
6.3 Soundness and Completeness
Fix a deductive system, such as natural deduction. A propositional formula is said to be provable if there
is a formal proof of it in the system. A propositional formula is said to be a tautology, or valid, if it is true
under any truth assignment. Provability is a syntactic notion, insofar as it asserts the existence of a syntactic
object, namely, a proof. Validity is a semantic notion, insofar as it has to do with truth assignments and
valuations. But, intuitively, these notions should coincide: both express the idea that a formula A has to
be true, or is necessarily true, and one would expect a good proof system to enable us to derive the valid
formulas.
Because of the way we have chosen our inference rules and defined the notion of a valuation, this intuition
holds true. The statement that every provable formula is valid is known as soundness, and the statement
that we can prove every valid formula is known as completeness.
These notions extend to provability from hypotheses. If Γ is a set of propositional formulas and A is a
propositional formula, then A is said to be a logical consequence of Γ if, given any truth assignment that
makes every formula in Γ true, A is true as well. In this extended setting, soundness says that if A is provable
from Γ, then A is a logical consequence of Γ. Completeness runs the other way: if A is a logical consequence
of Γ, it is provable from Γ.
Notice that with the rules of natural deduction, a formula A is provable from a set of hypotheses
{B1 , B2 , . . . , Bn } if and only if the formula B1 ∧ B2 ∧ · · · ∧ Bn → A is provable outright, that is, from
6.3. Soundness and Completeness 51

no hypotheses. So, at least for finite sets of formulas Γ, the two statements of soundness and completeness
are equivalent.
Proving soundness and completeness belongs to the realm of metatheory, since it requires us to reason about
our methods of reasoning. This is not a central focus of this book: we are more concerned with using
logic and the notion of truth than with establishing their properties. But the notions of soundness and
completeness play an important role in helping us understand the nature of the logical notions, and so we
will try to provide some hints here as to why these properties hold for propositional logic.
Proving soundness is easier than proving completeness. We wish to show that whenever A is provable from
a set of hypotheses, Γ, then A is a logical consequence of Γ. In a later chapter, we will consider proofs by
induction, which allows us to establish a property holds of a general collection of objects by showing that
it holds of some “simple” ones and is preserved under the passage to objects that are more complex. In
the case of natural deduction, it is enough to show that soundness holds of the most basic proofs — using
the assumption rule — and that it is preserved under each rule of inference. The base case is easy: the
assumption rule says that A is provable from hypothesis A, and clearly every truth assignment that makes
A true makes A true. The inductive steps are not much harder; it involves checking that the rules we have
chosen mesh with the semantic notions. For example, suppose the last rule is the and introduction rule. In
that case, we have a proof of A from some hypotheses Γ, and a proof of B from some hypotheses ∆, and
we combine these to form a proof of A ∧ B from the hypotheses in Γ ∪ ∆, that is, the hypotheses in both.
Inductively, we can assume that A is a logical consequence of Γ and that B is a logical consequence of ∆.
Let v be any truth assignment that makes every formula in Γ ∪ ∆ true. Then by the inductive hypothesis,
we have that it makes A true, and B true as well. By the definition of the valuation function, v̄(A ∧ B) = T,
as required.
Proving completeness is harder. It suffices to show that if A is any tautology, then A is provable. One
strategy is to show that natural deduction can simulate the method of truth tables. For example, suppose
A is build up from propositional variables B and C. Then in natural deduction, we should be able to prove
(B ∧ C) ∨ (B ∧ ¬C) ∨ (¬B ∧ C) ∨ (¬B ∧ ¬C),
with one disjunct for each line of the truth table. Then, we should be able to use each disjunct to “evaluate”
each expression occurring in A, proving it true or false in accordance with its valuation, until we have a
proof of A itself.
A nicer way to proceed is to express the rules of natural deduction in a way that allows us to work backwards
from A in search of a proof. In other words, first, we give a procedure for constructing a derivation of A by
working backwards from A. Then we argue that if the procedure fails, then, at the point where it fails, we
can find a truth assignment that makes A false. As a result, if every truth assignment makes A true, the
procedure returns a proof of A.
6.4 Exercises
1. Show that A → B, ¬A ∨ B, and ¬(A ∧ ¬B) are logically equivalent, by writing out the truth table and
showing that they have the same values for all truth assignments.
2. Write out the truth table for (A → B) ∧ (B ∧ C → A).
3. Show that A → B and ¬B → ¬A are equivalent, by writing out the truth tables and showing that
they have the same values for all truth assignments.
4. Does the following entailment hold?
{A → B ∨ C, ¬B → ¬C} |= A → B
Justify your answer by writing out the truth table (sorry, it is long). Indicate clearly the rows where
both hypotheses come out true.

CHAPTER
SEVEN
FIRST ORDER LOGIC
Propositional logic provides a good start at describing the general principles of logical reasoning, but it
does not go far enough. Some of the limitations are apparent even in the “Malice and Alice” example from
Chapter 2. Propositional logic does not give us the means to express a general principle that tells us that if
Alice is with her son on the beach, then her son is with Alice; the general fact that no child is younger than
his or her parent; or the general fact that if someone is alone, they are not with someone else. To express
principles like these, we need a way to talk about objects and individuals, as well as their properties and
the relationships between them. These are exactly what is provided by a more expressive logical framework
known as first-order logic, which will be the topic of the next few chapters.
7.1 Functions, Predicates, and Relations
Consider some ordinary statements about the natural numbers:

• Every natural number is even or odd, but not both.
• A natural number is even if and only if it is divisible by two.
• If some natural number, x, is even, then so is x2 .
• A natural number x is even if and only if x + 1 is odd.
• Any prime number that is greater than 2 is odd.
• For any three natural numbers x, y, and z, if x divides y and y divides z, then x divides z.
These statements are true, but we generally do not think of them as logically valid: they depend on as-
sumptions about the natural numbers, the meaning of the terms “even” and “odd,” and so on. But once we
accept the first statement, for example, it seems to be a logical consequence that the number of stairs in the
White House is either even or odd, and, in particular, if it is not even, it is odd. To make sense of inferences
like these, we need a logical system that can deal with objects, their properties, and relations between them.
Rather than fix a single language once and for all, first-order logic allows us to specify the symbols we wish
to use for any given domain of interest. In this section, we will use the following running example:
• the domain of interest is the natural numbers, N.
• there are objects, 0, 1, 2, 3, ….
• there are functions, addition and multiplication, as well as the square function, on this domain.
• there are predicates on this domain, “even,” “odd,” and “prime.”
• there are relations between elements of this domain, “equal,” “less than”, and “divides.”
53
For our logical language, we will choose symbols 1, 2, 3, add, mul, square, even, odd, prime, lt, and so on, to
denote these things. We will also have variables x, y, and z ranging over the natural numbers. Note all of
the following.
• Functions can take any number of arguments: if x and y are natural numbers, it makes sense to write
mul(x, y) and square(x). so mul takes two arguments, and square takes only one.
• Predicates and relations can also be understood in these terms. The predicates even(x) and prime(x)
take one argument, while the binary relations divides(x, y) and lt(x, y) take two arguments.
• Functions are different from predicates! A function takes one or more arguments, and returns a value.
A predicate takes one or more arguments, and is either true or false. We can think of predicates as
returning propositions, rather than values.
• In fact, we can think of the constant symbols 1, 2, 3, . . . as special sorts of function symbols that take
zero arguments. Analogously, we can consider the predicates that take zero arguments to be the
constant logical values, ⊤ and ⊥.
• In ordinary mathematics, we often use “infix” notation for binary functions and relations. For example,
we usually write x × y or x · y instead of mul(x, y), and we write x < y instead of lt(x, y). We will use
these conventions when writing proofs in natural deduction, and they are supported in Lean as well.
• We will treat the equality relation, x = y, as a special binary relation that is included in every first-order
language.
First-order logic allows us to build complex expressions out of the basic ones. Starting with the variables
and constants, we can use the function symbols to build up compound expressions like these:
x + y + z, (x + 1) × y × y, square(x + y × z)
Such expressions are called “terms.” Intuitively, they name objects in the intended domain of discourse.
Now, using the predicates and relation symbols, we can make assertions about these expressions:
even(x + y + z), prime((x + 1) × y × y), square(x + y × z) = w, x+y <z
Even more interestingly, we can use propositional connectives to build compound expressions like these:
• even(x + y + z) ∧ prime((x + 1) × y × y)
• ¬(square(x + y × z) = w) ∨ x + y < z
• x < y ∧ even(x) ∧ even(y) → x + 1 < y
The second one, for example, asserts that either (x+yz)2 is not equal to w, or x+y is less than z. Remember,
these are expressions in symbolic logic; in ordinary mathematics, we would express the notions using words
like “is even” and “if and only if,” as we did above. We will use notation like this whenever we are in the
realm of symbolic logic, for example, when we write proofs in natural deduction. Expressions like these are
called formulas. In contrast to terms, which name things, formulas say things; in other words, they make
assertions about objects in the domain of discourse.
7.2 The Universal Quantifier
What makes first-order logic powerful is that it allows us to make general assertions using quantifiers. The
universal quantifier ∀ followed by a variable x is meant to represent the phrase “for every x.” In other words,
it asserts that every value of x has the property that follows it. Using the universal quantifier, the examples
with which we began this previous section can be expressed as follows:
• ∀x ((even(x) ∨ odd(x)) ∧ ¬(even(x) ∧ ¬odd(x)))
54 Chapter 7. First Order Logic

• ∀x (even(x) ↔ 2 | x)
• ∀x (even(x) → even(x2 ))
• ∀x (even(x) ↔ odd(x + 1))
• ∀x (prime(x) ∧ x > 2 → odd(x))
• ∀x ∀y ∀z(x | y ∧ y | z → x | z)
It is common to combine multiple quantifiers of the same kind, and write, for example, ∀x, y, z (x | y ∧ y |
z → x | z) in the last expression.
Here are some notes on syntax:
• In symbolic logic, the universal quantifier is usually taken to bind tightly. For example, ∀x P ∨ Q is
interpreted as (∀x P ) ∨ Q, and we would write ∀x (P ∨ Q) to extend the scope.
• Be careful, however. In other contexts, especially in computer science, people often give quantifiers
the widest scope possible. This is the case with Lean. For example, ∀ x, P ∨ Q is interpreted as ∀
x, (P ∨ Q), and we would write (∀ x, P) ∨ Q to limit the scope.
• After the quantifier ∀x, the variable x is bound. For example, the expression ∀x (even(x) ∨ odd(x)) is
expresses that every number is even or odd. Notice that the variable x does not appear anywhere in the
informal statement. The statement is not about x at all; rather x is a dummy variable, a placeholder
that stands for the “thing” referred to within a phrase that beings with the words “every thing.” We
think of the expression ∀x (even(x) ∨ odd(x)) as being the same as the expression ∀y (even(y) ∨ odd(y)).
Lean treats these expressions as the same as well.
• In Lean, the expression ∀ x y z, x | y → y | z → x | z is interpreted as ∀ x y z, x | y → (y
| z → x | z), with parentheses associated to the right. The part of the expression after the universal
quantifier can therefore be interpreted as saying “given that x divides y and that y divides z, x divides
z.” The expression is logically equivalent to ∀ x y z, x | y ∧ y | z → x | z, but we will see that,
in Lean, it is often convenient to express facts like this as an iterated implication.
A variable that is not bound is called free. Notice that formulas in first-order logic say things about their
free variables. For example, in the interpretation we have in mind, the formula ∀y (x ≤ y) says that x is
less than or equal to every natural number. The formula ∀z (x ≤ z) says exactly the same thing; we can
always rename a bound variable, as long as we pick a name that does not clash with another name that is
already in use. On the other hand, the formula ∀y(w ≤ y) says that w is less than or equal to every natural
number. This is an entirely different statement: it says something about w, rather than x. So renaming a
free variable changes the meaning of a formula.
Notice also that some formulas, like ∀x, y (x ≤ y ∨ y ≤ x), have no free variables at all. Such a formula is
called a sentence, because it makes an outright assertion, a statement that is either true or false about the
intended interpretation. In Chapter 10 we will make the notion of an “intended interpretation” precise, and
explain what it means to be “true in an interpretation.” For now, the idea that formulas say things about
about object in an intended interpretation should motivate the rules for reasoning with such expressions.
In Chapter 1 we proved that the square root of two is irrational. One way to construe the statement is as
follows:
For every pair of integers, a and b, if b ̸= 0, it is not the case that a2 = 2b2 .
The advantage of this formulation is that we can restrict our attention to the integers, without having to
consider the larger domain of rationals. In symbolic logic, assuming our intended domain of discourse is the
integers, we would express this theorem using the universal quantifier:
∀a, b b ̸= 0 → ¬(a2 = 2b2 ).
How do we prove such a theorem? Informally, we would use such a pattern:
7.2. The Universal Quantifier 55

Let a and b be arbitrary integers, suppose b ̸= 0, and suppose a2 = 2b2 .

…
Contradiction.
What we are really doing is proving that the universal statement holds, by showing that it holds of “arbitrary”
values a and b. In natural deduction, the proof would look something like this:
1 2
b ̸= 0 a2 = 2 × b2
.. ..
. .
⊥ 2
¬(a2 = 2 × b2 )
1
b ̸= 0 → ¬(a2 = 2 × b2 )
∀b (b ̸= 0 → ¬(a2 = 2 × b2 ))
∀a ∀b (b ̸= 0 → ¬(a2 = 2 × b2 ))
Notice that after the hypotheses are canceled, we have proved b ̸= 0 → ¬(a2 = 2 × b2 ) without making any
assumptions about a and b; at this stage in the proof, they are “arbitrary,” justifying the application of the
universal quantifiers in the next two rules.
This example motivates the following rule in natural deduction:
A(x)
∀x A(x)
provided x is not free in any uncanceled hypothesis. Here A(x) stands for any formula that (potentially)
mentions x. Also remember that if y is any “fresh” variable that does not occur in A, we are thinking of
∀x vA(x) as being the same as ∀y A(y).
What about the elimination rule? Suppose we know that every number is even or odd. Then, in an ordinary
proof, we are free to assert “a is even or a is odd,” or “a2 is even or a2 is odd.” In terms of symbolic logic,
this amounts to the following inference: from ∀x (even(x) ∨ odd(x)), we can conclude even(t) ∨ odd(t) for
any term t. This motivates the elimination rule for the universal quantifier:
∀xA(x)
A(t)
where t is an arbitrary term.
In a sense, this feels like the elimination rule for implication; we might read the hypothesis as saying “if
x is any thing, then x is even or odd.” The conclusion is obtained by applying it to the fact that n is a
thing. Note that, in general, we could replace n by any term in the language, like n(m + 5) + 2. Similarly,
the introduction rule feels like the introduction rule for implication. If we want to show that everything has
a certain property, we temporarily let x denote an arbitrary thing, and then show that it has the relevant
property.
7.3 The Existential Quantifier
Dual to the universal quantifier is the existential quantifier, ∃, which is used to express assertions such as
“some number is even,” or, “between any two even numbers there is an odd number.”
The following statements about the natural numbers assert the existence of some natural number:
• There exists an odd composite number. (Remember that a natural number is composite if it is greater
than 1 and not prime.)

• Every natural number greater than one has a prime divisor.

• For every n, if n has a prime divisor smaller than n, then n is composite.
These statements can be expressed in first-order logic using the existential quantifier as follows:
• ∃n (odd(n) ∧ composite(n))
• ∀n (n > 1 → ∃p (prime(p) ∧ p | n))
• ∀n ((∃p (p | n ∧ prime(p) ∧ p < n)) → composite(n))
After we write ∃n, the variable n is bound in the formula, just as for the universal quantifier. So the formulas
∃n composite(n) and ∃m composite(m) are considered the same.
How do we prove such existential statements? Suppose we want to prove that there exists an odd composite
number. To do this, we just present a candidate, and show that the candidate satisfies the required properties.
For example, we could choose 15, and then show that 15 is odd and that 15 is composite. Of course, there’s
nothing special about 15, and we could have proven it also using a different number, like 9 or 35. The choice
of candidate does not matter, as long as it has the required property.
In a natural deduction proof this would look like this:
..
.
odd(15) ∧ composite(15)
∃n(odd(n) ∧ composite(n))
This illustrates the introduction rule for the existential quantifier:
A(t)
∃xA(x)
where t is any term. So to prove an existential formula, we just have to give one particular term for which
we can prove that formula. Such term is called a witness for the formula.
What about the elimination rule? Suppose that we know that n is some natural number and we know that
there exists a prime p such that p < n and p | n. How can we use this to prove that n is composite? We can
reason as follows:
Let p be any prime such that p < n and p | n.
…
Therefore, n is composite.
First, we assume that there is some p which satisfies the properties p is prime, p < n and p | n, and then we
reason about that p. As with case-based reasoning using “or,” the assumption is only temporary: if we can
show that n is composite from that assumption, that we have essentially shown that n is composite assuming
the existence of such a p. Notice that in this pattern of reasoning, p should be “arbitrary.” In other words,
we should not have assumed anything about p beforehand, we should not make any additional assumptions
about p along the way, and the conclusion should not mention p. Only then does it makes sense to say that
the conclusion follows from the “mere” existence of a p with the assumed properties.
In natural deduction, the elimination rule is expressed as follows:
1
A(y)
..
.
∃xA(x) B
1
B
Here we require that y is not free in B, and that the only uncanceled hypotheses where y occurs freely are
the hypotheses A(y) that are canceled when you apply this rule. Formally, this is what it means to say that
7.3. The Existential Quantifier 57

y is “arbitrary.” As was the case for or elimination and implication introduction, you can use the hypothesis
A(y) multiple times in the proof of B, and cancel all of them at once.
Intuitively, the rule says that you can prove B from the assumption ∃xA(x) by assuming A(y) for a fresh
variable y, and concluding, in any number of steps, that B follows. You should compare this rule to the
rule for or elimination, which is somewhat analogous. In the following example, we show that if A(x) always
implies ¬B(x), then there cannot be an x for which both A(x) and B(x) holds.
7.4 Relativization and Sorts
In first-order logic as we have presented it, there is one intended “universe” of objects of discourse, and the
universal and existential quantifiers range over that universe. For example, we could design a language to
talk about people living in a certain town, with a relation loves(x, y) to express that x loves y. In such a
language, we might express the statement that “everyone loves someone” by writing ∀x ∃y loves(x, y).
You should keep in mind that, at this stage, loves is just a symbol. We have designed the language with
a certain interpretation in mind, but one could also interpret the language as making statements about
the natural numbers, where loves(x, y) means that x is less than or equal to y. In that interpretation, the
sentence
∀x, y, z (loves(x, y) ∧ loves(y, z) → loves(x, z))
is true, though in the original interpretation it makes an implausible claim about the nature of love triangles.
In Chapter 10, we will spell out the notion that the deductive rules of first-order logic enable us to determine
the statements that are true in all interpretations, just as the rules of propositional logic enable us to
determine the statements that are true under all truth assignments.
Returning to the original example, suppose we want to represent the statement that, in our town, all the
women are strong and all the men are good looking. We could do that with the following two sentences:
• ∀x (woman(x) → strong(x))
• ∀x (man(x) → good-looking(x))
These are instances of relativization. The universal quantifier ranges over all the people in the town, but
this device gives us a way of using implication to restrict the scope of our statements to men and women,
respectively. The trick also comes into play when we render “every prime number greater than two is odd”:
∀x (prime(x) ∧ x > 2 → odd(x)).
We could also read this more literally as saying “for every number x, if x is prime and x is greater than to
2, then x is odd,” but it is natural to read it as a restricted quantifier.
It is also possible to relativize the existential quantifier to say things like “some woman is strong” and “some
man is good-looking.” These are expressed as follows:
• ∃x (woman(x) ∧ strong(x))
• ∃x (man(x) ∧ good-looking(x))
Notice that although we used implication to relativize the universal quantifier, here we need to use conjunction
instead of implication. The expression ∃x (woman(x) → strong(x)) says that there is something with the
property that if it is a woman, then it is strong. Classically this is equivalent to saying that there is something
which is either not a woman or is strong, which is a funny thing to say.
Now, suppose we are studying geometry, and we want to express the fact that given any two distinct points
p and q and any two lines L and M , if L and M both pass through p and q, then they have to be the same.

(In other words, there is at most one line between two distinct points.) One option is to design a first-order
logic where the intended universe is big enough to include both points and lines, and use relativization:
∀p, q, L, M (point(p) ∧ point(q) ∧ line(L) ∧ line(M )

∧on(p, L) ∧ on(q, L) ∧ on(p, M ) ∧ on(q, M ) → L = M )
But dealing with such predicates is tedious, and there is a mild extension of first-order logic, called many-
sorted first-order logic, which builds in some of the bookkeeping. In many-sorted logic, one can have different
sorts of objects — such as points and lines — and a separate stock of variables and quantifiers ranging
over each. Moreover, the specification of function symbols and predicate symbols indicates what sorts of
arguments they expect, and, in the case of function symbols, what sort of argument they return. For example,
we might choose to have a sort with variables p, q, r, . . . ranging over points, a sort with variables L, M, N, . . .
ranging over lines, and a relation on(p, L) relating the two. Then the assertion above is rendered more simply
as follows:
∀p, q, L, M (on(p, L) ∧ on(q, L) ∧ on(p, M ) ∧ on(q, M ) → L = M )
7.5 Equality
In symbolic logic, we use the expression s = t to express the fact that s and t are “equal” or “identical.” The
equality symbol is meant to model what we mean when we say, for example, “Alice’s brother is the victim,”
or “2 + 2 = 4.” We are asserting that two different descriptions refer to the same object. Because the notion
of identity can be applied to virtually any domain of objects, it is viewed as falling under the province of
logic.
Talk of “equality” or “identity” raises messy philosophical questions, however. Am I the same person I was
three days ago? Are the two copies of Huckleberry Finn sitting on my shelf the same book, or two different
books? Using symbolic logic to model identity presupposes that we have in mind a certain way of carving up
and interpreting the world. We assume that our terms refer to distinct entities, and writing s = t asserts that
the two expressions refer to the same thing. Axiomatically, we assume that equality satisfies the following
three properties:
• reflexivity: t = t, for any term t
• symmetry: if s = t, then t = s
• transitivity: if r = s and s = t, then r = t.
These properties are not enough to characterize equality, however. If two expressions denote the same thing,
then we should be able to substitute one for any other in any expression. It is convenient to adopt the
following convention: if r is any term, we may write r(x) to indicate that the variable x may occur in r.
Then, if s is another term, we can thereafter write r(s) to denote the result of replacing s for x in r. The
substitution rule for terms thus reads as follows: if s = t, then r(s) = r(t).
We already adopted a similar convention for formulas: if we introduce a formula as A(x), then A(t) denotes
the result of substituting t for x in A. With this in mind, we can write the rules for equality as follows:
s=t r=s s=t
t=t t=s r=t
s=t s=t P (s)
r(s) = r(t) P (t)
In the next chapter, you will learn how to use them.
Using equality, we can define even more quantifiers.
• We can express “there are at least two elements x such that A(x) holds” as ∃x ∃y (x ̸= y ∧A(x)∧A(y)).
7.5. Equality 59
• We can express “there are at most two elements x such that A(x) holds” as ∀x ∀y ∀z (A(x) ∧ A(y) ∧
A(z) → x = y ∨ y = z ∨ x = z). This states that if we have three elements a for which A(a) holds,
then two of them must be equal.
• We can express “there are exactly two elements x such that A(x) holds” as the conjunction of the
above two statements.
As an exercise, write out in first order logic the statements that there are at least, at most, and exactly three
elements x such that A(x) holds.
In logic, the expression ∃!x A(x) is used to express the fact that there is a unique x satisfying A(x), which
is to say, there is exactly one such x. As above, this can be expressed as follows:
∃x A(x) ∧ ∀y ∀y ′ (A(y) ∧ A(y ′ ) → y = y ′ )
The first conjunct says that there is at least one object satisfying A, and the second conjunct says that there
is at most one. The same thing can be expressed more concisely as follows:
∃x (A(x) ∧ ∀y (A(y) → y = x))
You should think about why this second expression works. In the next chapter we will see that, using the
rules of natural deduction, we can prove that these two expressions are equivalent.
7.6 Exercises
1. A perfect number is a number that is equal to the sum of its proper divisors, that is, the numbers that
divide it, other than itself. For example, 6 is perfect, because 6 = 1 + 2 + 3.
Using a language with variables ranging over the natural numbers and suitable functions and predicates,
write down first-order sentences asserting the following. Use a predicate perfect to express that a
number is perfect.
(a) 28 is perfect.
(b) There are no perfect numbers between 100 and 200.
(c) There are (at least) two perfect numbers between 200 and 10,000. (Express this by saying that
there are perfect numbers x and y between 200 and 10,000, with the property that x ̸= y.)
(d) Every perfect number is even.
(e) For every number, there is a perfect number that is larger than it. (This is one way to express
the statement that there are infinitely many perfect numbers.)
Here, the phrase “between a and b” is meant to include a and b.
By the way, we do not know whether the last two statements are true. They are open questions.
2. Using a language with variables ranging over people, and predicates trusts(x, y), politician(x),
knows(x, y), and related-to(x, y), and rich(x), write down first-order sentences asserting the follow-
ing:
(a) Nobody trusts a politician.
(b) Anyone who trusts a politician is crazy.
(c) Everyone knows someone who is related to a politician.
(d) Everyone who is rich is either a politician or knows a politician.
In each case, some interpretation may be involved. Notice that writing down a logical expression is
one way of helping to clarify the meaning.

CHAPTER
EIGHT
NATURAL DEDUCTION FOR FIRST ORDER LOGIC
8.1 Rules of Inference
In the last chapter, we discussed the language of first-order logic, and the rules that govern their use. We
summarize them here:
The universal quantifier:
A(x) ∀x A(x)
∀I ∀E
∀y A(y) A(t)
In the introduction rule, x should not be free in any uncanceled hypothesis. In the elimination rule, t can
be any term that does not clash with any of the bound variables in A.
The existential quantifier:
1
A(y)
A(t) ..
∃I .
∃xA(x)
∃xA(x) B
1 ∃E
B
In the introduction rule, t cant be any term that does not clash with any of the bound variables in A. In
the elimination rule, y should not be free in B or any uncanceled hypothesis.
Equality:
refl s=t symm r=s s=t trans
t=t t=s r=t
s=t s=t P (s)
subst subst
r(s) = r(t) P (t)
Strictly speaking, only refl and the second substitution rule are necessary. The others can be derived from
them.
8.2 The Universal Quantifier
The following example of a proof in natural deduction shows that if, for every x, A(x) holds, and for every
x, B(x) holds, then for every x, they both hold:
61
1 2
∀x A(x) ∀x B(x)
A(y) B(y)
A(y) ∧ B(y)
∀y (A(y) ∧ B(y))
2
∀x B(x) → ∀y (A(y) ∧ B(y))
1
∀x A(x) → (∀x B(x) → ∀y (A(y) ∧ B(y)))
Notice that neither of the assumptions 1 or 2 mention y, so that y is really “arbitrary” at the point where
the universal quantifiers are introduced.
Here is another example:
1
∀x A(x)
A(y)
A(y) ∨ B(y)
∀x (A(x) ∨ B(x))
1
∀x A(x) → ∀x (A(x) ∨ B(x))
As an exercise, try proving the following:
∀x (A(x) → B(x)) → (∀x A(x) → ∀x B(x)).
Here is a more challenging exercise. Suppose I tell you that, in a town, there is a (male) barber that shaves
all and only the men who do not shave themselves. You can show that this is a contradiction, arguing
informally, as follows:
By the assumption, the barber shaves himself if and only if he does not shave himself. Call this
statement (*).
Suppose the barber shaves himself. By (*), this implies that he does not shave himself, a contra-
diction. So, the barber does not shave himself.
But using (*) again, this implies that the barber shaves himself, which contradicts the fact we
just showed, namely, that the barber does not shave himself.
Try to turn this into a formal argument in natural deduction.
Let us return to the example of the natural numbers, to see how deductive notions play out there. Suppose
we have defined even and odd in such a way that we can prove:
• ∀n (¬even(n) → odd(n))
• ∀n (odd(n) → ¬even(n))
Then we can go on to derive ∀n (even(n) ∨ odd(n)) as follows:
∀n (¬even(n) → odd(n))
1
¬even(n) → odd(n) ¬even(n)
1
even(n) odd(n)
even(n) ∨ ¬even(n) even(n) ∨ odd(n) even(n) ∨ odd(n)
1
even(n) ∨ odd(n)
∀n (even(n) ∨ odd(n))
We can also prove and ∀n ¬(even(n) ∧ odd(n)):
62 Chapter 8. Natural Deduction for First Order Logic

H
even(n) ∧ odd(n)
H
odd(n) → ¬even(n) odd(n) even(n) ∧ odd(n)
¬even(n) even(n)
⊥ H
¬(even(n) ∧ odd(n))
∀n ¬(even(n) ∧ odd(n))
As we move from modeling basic rules of inference to modeling actual mathematical proofs, we will tend to
shift focus from natural deduction to formal proofs in Lean. Natural deduction has its uses: as a model of
logical reasoning, it provides us with a convenient means to study metatheoretic properties such as soundness
and completeness. For working within the system, however, proof languages like Lean’s tend to scale better,
and produce more readable proofs.
8.3 The Existential Quantifier
Remember that the intuition behind the elimination rule for the existential quantifier is that if we know
∃x A(x), we can temporarily reason about an arbitrary element y satisfying A(y) in order to prove a
conclusion that doesn’t depend on y. Here is an example of how it can be used. The next proof says that if
we know there is something satisfying both A and B, then we know, in particular, that there is something
satisfying A.
2
A(y) ∧ B(y)
A(y)
1
∃x(A(x) ∧ B(x)) ∃xA(x)
2
∃xA(x)
1
∃x(A(x) ∧ B(x)) → ∃xA(x)
The following proof shows that if there is something satsifying either A or B, then either there is something
satisfying A, or there is something satisfying B.
3 3
A(y) B(y)
∃x A(x) ∃x B(x)
2
A(y) ∨ B(y) ∃x A(x) ∨ ∃x B(x) ∃x A(x) ∨ ∃x B(x)
1 3
∃x (A(x) ∨ B(x)) ∃x A(x) ∨ ∃x B(x)
2
∃x A(x) ∨ ∃x B(x)
1
∃x (A(x) ∨ B(x)) → ∃x A(x) ∨ ∃x B(x))
The following example is more involved:
1 3
∀x (A(x) → ¬B(x)) A(x) ∧ B(x)
3
A(x) → ¬B(x) A(x) A(x) ∧ B(x)
¬B(x) B(x)
2
∃x (A(x) ∧ B(x)) ⊥
3
⊥ 2
¬∃x (A(x) ∧ B(x))
1
∀x (A(x) → ¬B(x)) → ¬∃x (A(x) ∧ B(x))
In this proof, the existential elimination rule (the line labeled 3) is used to cancel two hypotheses at the same
time. Note that when this rule is applied, the hypothesis ∀x (A(x) → ¬B(x)) has not yet been canceled.
8.3. The Existential Quantifier 63

So we have to make sure that this formula doesn’t contain the variable x freely. But this is o.k., since this
hypothesis contains x only as a bound variable.
Another example is that if x does not occur in P , then ∃x P is equivalent to P :
1 2 1
∃x P P 2
P
P ∃x P 1
∃x P ↔ P
This short but tricky, so let us go through it carefully. On the left, we assume ∃x P to conclude P . We
assume P , and now we can immediately cancel this assumption by existential elimination, since x does not
occur in P , so it doesn’t occur freely in any assumption or in the conclusion. On the right we use existential
introduction to conclude ∃x P from P .
8.4 Equality
Recall the natural deduction rules for equality:

s=t r=s s=t
t=t t=s r=t
s=t s=t P (s)
r(s) = r(t) P (t)
Keep in mind that we have implicitly fixed some first-order language, and r, s, and t are any terms in
that language. Recall also that we have adopted the practice of using functional notation with terms. For
example, if we think of r(x) as the term (x + y) × (z + 0) in the language of arithmetic, then r(0) is the term
(0 + y) × (z + 0) and r(u + v) is ((u + v) + y) × (z + 0). So one example of the first inference on the second
line is this:
u+v =0
((u + v) + y) × (z + 0) = (0 + y) × (z + 0)
The second axiom on that line is similar, except now P (x) stands for any formula, as in the following
inference:
u+v =0 x + (u + v) < y
x+0<y
Notice that we have written the reflexivity axiom, t = t, as a rule with no premises. If you use it in a proof,
it does not count as a hypothesis; it is built into the logic.
In fact, we can think of the first inference on the second line as a special case of the first. Consider, for
example, the formula ((u + v) + y) × (z + 0) = (x + y) × (z + 0). If we plug u + v in for x, we get an instance
of reflexivity. If we plug in 0, we get the conclusion of the first example above. The following is therefore a
derivation of the first inference, using only reflexivity and the second substitution rule above:
u+v =0 ((u + v) + y) × (z + 0) = ((u + v) + y) × (z + 0)

((u + v) + y) × (z + 0) = (0 + y) × (z + 0)
Roughly speaking, we are replacing the second instance of u + v in an instance of reflexivity with 0 to get
the conclusion we want.
Equality rules let us carry out calculations in symbolic logic. This typically amounts to using the equality
rules we have already discussed, together with a list of general identities. For example, the following identities
hold for any real numbers x, y, and z:
• commutativity of addition: x + y = y + x
• associativity of addition: (x + y) + z = x + (y + z)

• additive identity: x + 0 = 0 + x = x
• additive inverse: −x + x = x + −x = 0
• multiplicative identity: x · 1 = 1 · x = x
• commutativity of multiplication: x · y = y · x
• associativity of multiplication: (x · y) · z = x · (y · z)
• distributivity: x · (y + z) = x · y + x · z, (x + y) · z = x · z + y · z
You should imagine that there are implicit universal quantifiers in front of each statement, asserting that
the statement holds for any values of x, y, and z. Note that x, y, and z can, in particular, be integers
or rational numbers as well. Calculations involving real numbers, rational numbers, or integers generally
involve identities like this.
The strategy is to use the elimination rule for the universal quantifier to instantiate general identities, use
symmetry, if necessary, to orient an equation in the right direction, and then using the substitution rule for
equality to change something in a previous result. For example, here is a natural deduction proof of a simple
identity, ∀x, y, z ((x + y) + z = (x + z) + y), using only commutativity and associativity of addition. We have
taken the liberty of using a brief name to denote the relevant identities, and combining multiple instances
of the universal quantifier introduction and elimination rules into a single step.
assoc comm assoc
(x + z) + y = x + (z + y) y+z =z+y (x + y) + z = x + (y + z)
x + (z + y) = (x + z) + y (x + y) + z = x + (z + y)
(x + y) + z = (x + z) + y
∀x, y, z ((x + y) + z = (x + z) + y)
There is generally nothing interesting to be learned from carrying out such calculations in natural deduction,
but you should try one or two examples to get the hang of it, and then take pleasure in knowing that it is
possible.
8.5 Counterexamples and Relativized Quantifiers
Consider the statement:

Every prime number is odd.
In first-order logic, we could formulate this as ∀p (prime(p) → odd(p)). This statement is false, because there
is a prime number which is even, namely the number 2. This is called a counterexample to the statement.
More generally, given a formula ∀x A(x), a counterexample is a value t such that ¬A(t) holds. Such
a counterexample shows that the original formula is false, because we have the following equivalence:
¬∀x A(x) ↔ ∃x ¬A(x). So if we find a value t such that ¬A(t) holds, then by the existential intro-
duction rule we can conclude that ∃x ¬A(x), and then by the above equivalence we have ¬∀x A(x). Here is
a proof of the equivalence:
5
¬A(x)
4
¬(∃x ¬A(x)) ∃x ¬A(x) 2
⊥ 5 ∀x A(x)
3
A(x) ¬A(y) A(y)
1 1
¬∀x A(x) ∀x A(x) ∃x ¬A(x) ⊥
3
⊥ 4
⊥ 2
∃x ¬A(x) ¬∀x A(x)
1
¬∀x A(x) ↔ ∃x ¬A(x)
8.5. Counterexamples and Relativized Quantifiers 65

One remark about the proof: at the step marked by 4 we cannot use the existential introduction rule,
because at that point our only assumption is ¬∀x A(x), and from that assumption we cannot prove ¬A(t)
for a particular term t. So we use a proof by contradiction there.
As an exercise, prove the “dual” equivalence yourself: ¬∃x A(x) ↔ ∀x ¬A(x). This can be done without
using proof by contradiction.
In Chapter 7 we saw examples of how to use relativization to restrict the scope of a universal quantifier.
Suppose we want to say “every prime number is greater than 1”. In first order logic this can be written as
∀n(prime(n) → n > 1). The reason is that the original statement is equivalent to the statement “for every
natural number, if it is prime, then it is greater than 1”. Similarly, suppose we want to say “there exists a
prime number greater than 100.” This is equivalent to saying “there exists a natural number which is prime
and greater than 100,” which can be expressed as ∃n (prime(n) ∧ n > 100).
As an exercise you can prove the above results about negations of quantifiers also for relativized quantifiers.
Specifically, prove the following statements:
• ¬∃x (A(x) ∧ B(x)) ↔ ∀x (A(x) → ¬B(x));
• ¬∀x (A(x) → B(x)) ↔ ∃x(A(x) ∧ ¬B(x))
For reference, here is a list of valid sentences involving quantifiers:
• ∀x A ↔ A if x is not free in A
• ∃x A ↔ A if x is not free in A
• ∀x (A(x) ∧ B(x)) ↔ ∀x A(x) ∧ ∀x B(x)
• ∃x (A(x) ∧ B) ↔ ∃ xA(x) ∧ B if x is not free in B
• ∃x (A(x) ∨ B(x)) ↔ ∃ xA(x) ∨ ∃ xB(x)
• ∀x (A(x) ∨ B) ↔ ∀x A(x) ∨ B if x is not free in B
• ∀x (A(x) → B) ↔ (∃x A(x) → B) if x is not free in B
• ∃x (A(x) → B) ↔ (∀x A(x) → B) if x is not free in B
• ∀x (A → B(x)) ↔ (A → ∀x B(x)) if x is not free in A
• ∃x (A(x) → B) ↔ (A(x) → ∃ xB) if x is not free in B
• ∃x A(x) ↔ ¬∀x ¬A(x)
• ∀x A(x) ↔ ¬∃x ¬A(x)
• ¬∃x A(x) ↔ ∀x ¬A(x)
• ¬∀x A(x) ↔ ∃x ¬A(x)
All of these can be derived in natural deduction. The last two allow us to push negations inwards, so we
can continue to put first-order formulas in negation normal form. Other rules allow us to bring quantifiers
to the front of any formula, though, in general, there will be multiple ways of doing this. For example, the
formula
∀x A(x) → ∃y ∀z B(y, z)
is equivalent to both
∃x, y ∀z (A(x) → B(y, z))
and
∃y ∀z ∃x (A(x) → B(y, z)).
A formula with all the quantifiers in front is said to be in prenex form.

8.6 Exercises
1. Give a natural deduction proof of
∀x (A(x) → B(x)) → (∀x A(x) → ∀x B(x)).
2. Give a natural deduction proof of ∀x B(x) from hypotheses ∀x (A(x) ∨ B(x)) and ∀y ¬A(y).
3. From hypotheses ∀x (even(x) ∨ odd(x)) and ∀x (odd(x) → even(s(x))) give a natural deduction proof
∀x (even(x) ∨ even(s(x))). (It might help to think of s(x) as the function defined by s(x) = x + 1.)
4. Give a natural deduction proof of ∃x A(x) ∨ ∃x B(x) → ∃x (A(x) ∨ B(x)).
5. Give a natural deduction proof of ∃x (A(x) ∧ C(x)) from the assumptions ∃x (A(x) ∧ B(x)) and
∀x (A(x) ∧ B(x) → C(x)).
6. Prove some of the other equivalences in the last section.
7. Consider some of the various ways of expressing “nobody trusts a politician” in first-order logic:
• ∀x (politician(x) → ∀y (¬trusts(y, x)))
• ∀x, y (politician(x) → ¬trusts(y, x))
• ¬∃x, y (politician(x) ∧ trusts(y, x))
• ∀x, y (trusts(y, x) → ¬politician(x))
They are all logically equivalent. Show this for the second and the fourth, by giving natural deduction
proofs of each from the other. (As a shortcut, in the ∀ introduction and elimination rules, you can
introduce / eliminate both variables in one step.)
8. Formalize the following statements, and give a natural deduction proof in which the first three state-
ments appear as (uncancelled) hypotheses, and the last line is the conclusion:
• Every young and healthy person likes baseball.
• Every active person is healthy.
• Someone is young and active.
• Therefore, someone likes baseball.
Use Y (x) for “is young,” H(x) for “is healthy,” A(x) for “is active,” and B(x) for “likes baseball.”
9. Give a natural deduction proof of ∀x, y, z (x = z → (y = z → x = y)) using the equality rules in
Section 8.4.
10. Give a natural deduction proof of ∀x, y (x = y → y = x) using only these two hypotheses (and none of
the new equality rules):
• ∀x (x = x)
• ∀u, v, w (u = w → (v = w → u = v))
(Hint: Choose instantiations of u, v, and w carefully. You can instantiate all the universal quantifiers
in one step, as on the last homework assignment.)
11. Give a natural deduction proof of ¬∃x (A(x) ∧ B(x)) ↔ ∀x (A(x) → ¬B(x))
12. Give a natural deduction proof of ¬∀x (A(x) → B(x)) ↔ ∃x (A(x) ∧ ¬B(x))
13. Remember that both the following express ∃!x A(x), that is, the statement that there is a unique x
satisfying A(x):
• ∃x (A(x) ∧ ∀y (A(y) → y = x))
8.6. Exercises 67
• ∃x A(x) ∧ ∀y ∀y ′ (A(y) ∧ A(y ′ ) → y = y ′ )

Do the following:
• Give a natural deduction proof of the second, assuming the first as a hypothesis.
• Give a natural deduction proof of the first, asssuming the second as a hypothesis.
(Warning: these are long.)

CHAPTER
NINE
FIRST ORDER LOGIC IN LEAN
9.1 Functions, Predicates, and Relations
In the last chapter, we discussed the language of first-order logic. We will see in the course of this book that
Lean’s built-in logic is much more expressive; but it includes first-order logic, which is to say, anything that
can be expressed (and proved) in first-order logic can be expressed (and proved) in Lean.
Lean is based on a foundational framework called type theory, in which every variable is assumed to range
elements of some type. You can think of a type as being a “universe,” or a “domain of discourse,” in the
sense of first-order logic.
For example, suppose we want to work with a first-order language with one constant symbol, one unary
function symbol, one binary function symbol, one unary relation symbol, and one binary relation symbol.
We can declare a new type U (for “universe”) and the relevant symbols as follows:
constant U : Type
constant c : U
constant f : U → U
constant g : U → U → U
constant P : U → Prop
constant R : U → U → Prop
We can then use them as follows:
variables x y : U
#check c
#check f c
#check g x y
#check g x (f c)
#check P (g x (f c))
#check R x y
The #check command tells us that the first four expressions have type U, and that the last two have type
Prop. Roughly, this means that the first four expressions correspond to terms of first-order logic, and that
the last two correspond to formulas.
Note all the following:
• A unary function is represented as an object of type U → U and a binary function is represented as
an object of type U → U → U, using the same notation as for implication between propositions.
69
• We write, for example, f x to denote the result of applying f to x, and g x y to denote the result of
applying g to x and y, again just as we did when using modus ponens for first-order logic. Parentheses
are needed in the expression g x (f c) to ensure that f c is parsed as a single argument.
• A unary predicate is presented as an object of type U → Prop and a binary function is represented
as an object of type U → U → Prop. You can think of a binary relation R as being a function that
assumes two arguments in the universe, U, and returns a proposition.
• We write P x to denote the assertion that P holds of x, and R x y to denote that R holds of x and y.
You may reasonably wonder what difference there is between a constant and a variable in Lean. The following
declarations also work:
variable U : Type
variable c : U
variable f : U → U
variable g : U → U → U
variable P : U → Prop
variable R : U → U → Prop
variables x y : U
#check c
#check f c
#check g x y
#check g x (f c)
#check P (g x (f c))
#check R x y
Although the examples function in much the same way, the constant and variable commands do very
different things. The constant command declares a new object, axiomatically, and adds it to the list of
objects Lean knows about. In contrast, when it is first executed, the variable command does not create
anything. Rather, it tells Lean that whenever we enter an expression using the corresponding identifier, it
should create a temporary variable of the corresponding type.
Many types are already declared in Lean’s standard library. For example, there is a type written nat or N,
that denotes the natural numbers:
#check nat
#check N
You can enter the unicode N with \nat or \N. The two expressions mean the same thing.
Using this built-in type, we can model the language of arithmetic, as described in the last chapter, as follows:
namespace hide
constant mul : N → N → N
constant add : N → N → N
constant square : N → N
constant even : N → Prop
constant odd : N → Prop
constant prime : N → Prop
constant divides : N → N → Prop
constant lt : N → N → Prop
constant zero : N
constant one : N
70 Chapter 9. First Order Logic in Lean

end hide
We have used the namespace command to avoid conflicts with identifiers that are already declared in the
Lean library. We can again use the #check command to try them out:
variables w x y z : N
#check mul x y
#check add x y
#check square x
#check even x
In fact, all of the functions, predicates, and relations discussed here, except for the “square” function and
“prime,” are defined in the core Lean library. They become available to us when we put the commands
import data.nat and open nat at the top of a file in Lean.
constant square : N → N
variables w x y z : N
#check even (x + y + z) ∧ prime ((x + 1) * y * y)

#check ¬ (square (x + y * z) = w) ∨ x + y < z
#check x < y ∧ even x ∧ even y → x + 1 < y
Here, we declare the constants square and prime axiomatically, but refer to the other operations and
predicates in the Lean library. In this book, we will often proceed in this way, telling you explicitly what
facts from the library you should use for exercises.
Again, note the following aspects of syntax:
• In contrast to ordinary mathematical notation, in Lean, functions are applied without parentheses or
commas. For example, we write square x and add x y instead of square(x) and add(x, y).
• The same holds for predicates and relations: we write even x and lt x y instead of even(x) and
lt(x, y), as one might do in symbolic logic.
• The notation add : N → N → N indicates that addition assumes two arguments, both natural num-
bers, and returns a natural number.
• Similarly, the notation divides : N → N → Prop indicates that divides is a binary relation, which
assumes two natural numbers as arguments and forms a proposition. In other words, divides x y
expresses the assertion that x divides y.
Lean can help us distinguish between terms and formulas. If we #check the expression x + y + 1 in Lean,
we are told it has type N, which is to say, it denotes a natural number. If we #check the expression even
(x + y + 1), we are told that it has type Prop, which is to say, it expresses a proposition.
In Chapter 7 we considered many-sorted logic, where one can have multiple universes. For example, we
might want to use first-order logic for geometry, with quantifiers ranging over points and lines. In Lean, we
can model this as by introducing a new type for each sort:
variables Point Line : Type

variable lies_on : Point → Line → Prop
We can then express that two distinct points determine a line as follows:
9.1. Functions, Predicates, and Relations 71

#check ∀ (p q : Point) (L M : Line),

p ̸= q → lies_on p L → lies_on q L → lies_on p M → lies_on q M → L = M
Notice that we have followed the convention of using iterated implication rather than conjunction in the
antecedent. In fact, Lean is smart enough to infer what sorts of objects p, q, L, and M are from the fact that
they are used with the relation on, so we could have written, more simply, this:
#check ∀ p q L M, p ̸= q → lies_on p L → lies_on q L → lies_on p M → lies_on q M → L = M
9.2 Using the Universal Quantifier
In Lean, you can enter the universal quantifier by writing \all. The motivating examples from Section 7.1
are rendered as follows:

constant odd : N → Prop
#check ∀ x, (even x ∨ odd x) ∧ ¬ (even x ∧ odd x)

#check ∀ x, even x ↔ 2 | x
#check ∀ x, even x → even (x^2)
#check ∀ x, even x ↔ odd (x + 1)
#check ∀ x, prime x ∧ x > 2 → odd x
#check ∀ x y z, x | y → y | z → x | z
Remember that Lean expects a comma after the universal quantifier, and gives it the widest scope possible.
For example, ∀ x, P ∨ Q is interpreted as ∀ x, (P ∨ Q), and we would write (∀ x, P) ∨ Q to limit the
scope. If you prefer, you can use the plain ascii expression forall instead of the unicode ∀.
In Lean, then, the pattern for proving a universal statement is rendered as follows:
variable U : Type
example : ∀ x, P x :=
assume x,
show P x, from sorry
Read assume x as “fix an arbitrary value x of U.” Since we are allowed to rename bound variables at will,
we can equivalently write either of the following:
variable U : Type
example : ∀ y, P y :=
assume x,
show P x, from sorry
example : ∀ x, P x :=
assume y,
show P y, from sorry
This constitutes the introduction rule for the universal quantifier. It is very similar to the introduction
rule for implication: instead of using assume to temporarily introduce an assumption, we use assume to

temporarily introduce a new object, y. (In fact, assume and assume are both alternate syntax for a single
internal construct in Lean, which can also be denoted by λ.)
The elimination rule is, similarly, implemented as follows:
variable U : Type
variable h : ∀ x, P x
variable a : U
example : P a :=
show P a, from h a
Observe the notation: P a is obtained by “applying” the hypothesis h to a. Once again, note the similarity
to the elimination rule for implication.
Here is an example of how it is used:
variable U : Type
variables A B : U → Prop
example (h1 : ∀ x, A x → B x) (h2 : ∀ x, A x) : ∀ x, B x :=

assume y,
have h3 : A y, from h2 y,
have h4 : A y → B y, from h1 y,
show B y, from h4 h3
Here is an even shorter version of the same proof, where we avoid using have:
example (h1 : ∀ x, A x → B x) (h2 : ∀ x, A x) : ∀ x, B x :=

assume y,
show B y, from h1 y (h2 y)
You should talk through the steps, here. Applying h1 to y yields a proof of A y → B y, which we then
apply to h2 y, which is a proof of A y. The result is the proof of B y that we are after.
In the last chapter, we considered the following proof in natural deduction:
1 2
∀x A(x) ∀x B(x)
A(y) B(y)
A(y) ∧ B(y)
∀y (A(y) ∧ B(y))
2
∀x B(x) → ∀y (A(y) ∧ B(y))
1
∀x A(x) → (∀x B(x) → ∀y (A(y) ∧ B(y)))
Here is the same proof rendered in Lean:
variable U : Type
example : (∀ x, A x) → (∀ x, B x) → (∀ x, A x ∧ B x) :=
assume hA : ∀ x, A x,
assume hB : ∀ x, B x,
assume y,
have Ay : A y, from hA y,
have By : B y, from hB y,
show A y ∧ B y, from and.intro Ay By
9.2. Using the Universal Quantifier 73

Here is an alternative version, using the “anonymous” versions of have:
variable U : Type
example : (∀ x, A x) → (∀ x, B x) → (∀ x, A x ∧ B x) :=
assume hA : ∀ x, A x,
assume hB : ∀ x, B x,
assume y,
have A y, from hA y,
have B y, from hB y,
show A y ∧ B y, from and.intro ‹A y› ‹B y›
The exercises below ask you to prove the barber paradox, which was discussed in the last chapter. You
can do that using only propositional reasoning and the rules for the universal quantifer that we have just
discussed.
9.3 Using the Existential Quantifier
In Lean, you can type the existential quantifier, ∃, by writing \ex. If you prefer you can use the ascii
equivalent, exists. The introduction rule is exists.intro and requires two arguments: a term, and a
proof that that term satisfies the required property.
variable U : Type
example (y : U) (h : P y) : ∃ x, P x :=
exists.intro y h
The elimination rule for the existential quantifier is given by exists.elim. It follows the form of the natural
deduction rule: if we know ∃x, P x and we are trying to prove Q, it suffices to introduce a new variable, y,
and prove Q under the assumption that P y holds.
variable U : Type
variable Q : Prop
example (h1 : ∃ x, P x) (h2 : ∀ x, P x → Q) : Q :=

exists.elim h1
(assume (y : U) (h : P y),
have h3 : P y → Q, from h2 y,
show Q, from h3 h)
The following example uses both the introduction and the elimination rules for the existential quantifier.
variable U : Type
example : (∃ x, A x ∧ B x) → ∃ x, A x :=
assume h1 : ∃ x, A x ∧ B x,
exists.elim h1
(assume y (h2 : A y ∧ B y),
have h3 : A y, from and.left h2,
show ∃ x, A x, from exists.intro y h3)

Notice the parentheses in the hypothesis; if we left them out, everything after the first ∃ x would be included
in the scope of that quantifier. From the hypothesis, we obtain a y that satisfies A y ∧ B y, and hence A y
in particular. So y is enough to witness the conclusion.
It is sometimes annoying to enclose the proof after an exists.elim in parenthesis, as we did here with the
assume ... show block. To avoid that, we can use a bit of syntax from the programming world, and use a
dollar sign instead. In Lean, an expression f $ t means the same thing as f (t), with the advantage that
we do not have to remember to close the parenthesis. With this gadget, we can write the proof above as
follows:
variable U : Type
example : (∃ x, A x ∧ B x) → ∃ x, A x :=
exists.elim h1 $
assume y (h2 : A y ∧ B y),
have h3 : A y, from and.left h2,
show ∃ x, A x, from exists.intro y h3
The following example is more involved:
example : (∃ x, A x ∨ B x) → (∃ x, A x) ∨ (∃ x, B x) :=
assume h1 : ∃ x, A x ∨ B x,
exists.elim h1 $
assume y (h2 : A y ∨ B y),
or.elim h2
(assume h3 : A y,
have h4 : ∃ x, A x, from exists.intro y h3,
show (∃ x, A x) ∨ (∃ x, B x), from or.inl h4)
(assume h3 : B y,
have h4 : ∃ x, B x, from exists.intro y h3,
show (∃ x, A x) ∨ (∃ x, B x), from or.inr h4)
Note again the placement of parentheses in the statement.

In the last chapter, we considered the following natural deduction proof:
1 3
∀x (A(x) → ¬B(x)) A(x) ∧ B(x)
3
A(x) → ¬B(x) A(x) A(x) ∧ B(x)
¬B(x) B(x)
2
∃x (A(x) ∧ B(x)) ⊥
3
⊥ 2
¬∃x (A(x) ∧ B(x))
1
∀x (A(x) → ¬B(x)) → ¬∃x (A(x) ∧ B(x))
Here is a proof of the same implication in Lean:
variable U : Type
example : (∀ x, A x → ¬ B x) → ¬ ∃ x, A x ∧ B x :=
assume h1 : ∀ x, A x → ¬ B x,
exists.elim h2 $
assume x (h3 : A x ∧ B x),
have h4 : A x, from and.left h3,
9.3. Using the Existential Quantifier 75

have h5 : B x, from and.right h3,

have h6 : ¬ B x, from h1 x h4,
show false, from h6 h5
Here, we use exists.elim to introduce a value x satisfying A x ∧ B x. The name is arbitrary; we could
just as well have used z:
example : (∀ x, A x → ¬ B x) → ¬ ∃ x, A x ∧ B x :=
assume h1 : ∀ x, A x → ¬ B x,
exists.elim h2 $
assume z (h3 : A z ∧ B z),
have h4 : A z, from and.left h3,
have h5 : B z, from and.right h3,
have h6 : ¬ B z, from h1 z h4,
Here is another example of the exists-elimination rule:
variable U : Type
variable u : U
variable P : Prop
example : (∃x : U, P) ↔ P :=
iff.intro
(assume h1 : ∃x, P,
exists.elim h1 $
assume x (h2 : P),
h2)
(assume h1 : P,
exists.intro u h1)
It is subtle: the proof does not go through if we do not declare a variable u of type U, even though u does not
appear in the statement of the theorem. The semantics of first-order logic, discussed in the next chapter,
presuppose that the universe is nonempty. In Lean, however, it is possible for a type to be empty, and so
the proof above depends on the fact that there is an element u in U.
These features are all illustrated in the following example:
variable U : Type
variables P R : U → Prop
variable Q : Prop
example (h1 : ∃x, P x ∧ R x) (h2 : ∀x, P x → R x → Q) : Q :=

let ⟨y, hPy, hRy⟩ := h1 in
show Q, from h2 y hPy hRy
9.4 Equality and calculational proofs
In Lean, reflexivity, symmetry, and transitivity are called eq.refl, eq.symm, and eq.trans, and the second
substitution rule is called eq.subst. Their uses are illustrated below.
variable A : Type
variables x y z : A

variable P : A → Prop
example : x = x :=
show x = x, from eq.refl x
example : y = x :=
have h : x = y, from sorry,
show y = x, from eq.symm h
example : x = z :=
have h1 : x = y, from sorry,
have h2 : y = z, from sorry,
show x = z, from eq.trans h1 h2
example : P y :=
have h1 : x = y, from sorry,
have h2 : P x, from sorry,
show P y, from eq.subst h1 h2
The rule eq.refl above assumes x as an argument, because there is no hypothesis to infer it from. All the
other rules assume their premises as arguments. Here is an example of equational reasoning:
variables (A : Type) (x y z : A)
example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
have h3 : x = y, from eq.symm h1,
show x = z, from eq.trans h3 h2
This proof can be written more concisely:
example : y = x → y = z → x = z :=
assume h1 h2, eq.trans (eq.symm h1) h2
Because calculation is so important in mathematics, however, Lean provides more efficient ways of carrying
them out. One is the rewrite tactic. Typing begin and end in a Lean proof puts Lean into “tactic mode,”
which means that Lean then expects a list of instructions. The command rewrite then uses identities to
change the goal. For example, the previous proof could be written as follows:
example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
show x = z,
begin
rewrite ←h1,
apply h2
end
The rewrite tactic can be abbreviated rw.

The first command changes the goal x = z to y = z; the minus sign before h1 tells Lean to use the equation
in the reverse direction. After that, we can finish the goal by applying h2.
An alternative is to rewrite the goal using h1 and h2, which reduces the goal to x = x. When that happens,
rewrite automatically applies reflexivity.
9.4. Equality and calculational proofs 77

example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
show x = z,
begin
rw ←h1,
rw h2
end
In fact, a sequence of rewrites can be combined, using square brackets:

example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
show x = z,
begin
rw [←h1, h2]
end
And when you reduce a proof to a single tactic, you can use by instead of begin ... end.
example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
show x = z, by rw [←h1, h2]
We will see in the coming chapters that in ordinary mathematical proofs, one commonly carries out calcu-
lations in a format like this:
t1 = t2
. . . = t3
. . . = t4
. . . = t5
Lean has a mechanism to model calculational proofs like this. Whenever a proof of an equation is expected,
you can provide a proof using the identifier calc, following by a chain of equalities and justification, in the
following form:
calc
e1 = e2 : justification 1
... = e3 : justification 2
The chain can go on as long as needed. Each justification is the name of the assumption or theorem that is
used. For example, the previous proof could be written as follows:
example : y = x → y = z → x = z :=
assume h1 : y = x,
assume h2 : y = z,
calc
x = y : eq.symm h1
... = z : h2
As usual, the syntax is finicky; notice that there are no commas in the calc expression, and the colons and
dots need to be entered exactly in that form. All that varies are the expressions e1, e2, e3, ... and the
justifications themselves.

The calc environment is most powerful when used in conjunction with rewrite, since we can then rewrite
expressions with facts from the library. For example, Lean’s library has a number of basic identities for the
integers, such as these:
variables x y z : int
example : x + 0 = x :=
add_zero x
example : 0 + x = x :=
zero_add x
example : (x + y) + z = x + (y + z) :=
add_assoc x y z
example : x + y = y + x :=
add_comm x y
example : (x * y) * z = x * (y * z) :=
mul_assoc x y z
example : x * y = y * x :=
mul_comm x y
example : x * (y + z) = x * y + x * z :=
left_distrib x y z
example : (x + y) * z = x * z + y * z :=
right_distrib x y z
You can also write the type of integers as Z, entered with either \Z or \int. Notice that, for example,
add_comm is the theorem ∀ x y, x + y = y + x. So to instantiate it to s + t = t + s, you write add_comm
s t. Using these axioms, here is the calculation above rendered in Lean, as a theorem about the integers:
example (x y z : int) : (x + y) + z = (x + z) + y :=
calc
(x + y) + z = x + (y + z) : add_assoc x y z
... = x + (z + y) : eq.subst (add_comm y z) rfl
... = (x + z) + y : eq.symm (add_assoc x z y)
Using rewrite is more efficient, though at times we have to provide information to specify where the rules
are used:
calc
(x + y) + z = x + (y + z) : by rw add_assoc
... = x + (z + y) : by rw [add_comm y z]
... = (x + z) + y : by rw add_assoc
In that case, we can use a single rewrite:
by rw [add_assoc, add_comm y z, add_assoc]
If you #check the proof before the sequence of rewrites is sufficient, the error message will display the
remaining goal.
9.4. Equality and calculational proofs 79

variables a b d c : int
example : (a + b) * (c + d) = a * c + b * c + a * d + b * d :=
calc
(a + b) * (c + d) = (a + b) * c + (a + b) * d : by rw left_distrib
... = (a * c + b * c) + (a + b) * d : by rw right_distrib
... = (a * c + b * c) + (a * d + b * d) : by rw right_distrib
... = a * c + b * c + a * d + b * d : by rw ←add_assoc
Once again, we can get by with a shorter proof:
example : (a + b) * (c + d) = a * c + b * c + a * d + b * d :=
by rw [left_distrib, right_distrib, right_distrib, ←add_assoc]
9.5 Exercises
1. Fill in the sorry.
section
variable A : Type
variable f : A → A
variable h : ∀ x, P x → P (f x)
-- Show the following:

example : ∀ y, P y → P (f (f y)) :=
sorry
end
section
variable U : Type
example : (∀ x, A x ∧ B x) → ∀ x, A x :=
sorry
end
section
variable U : Type
variables A B C : U → Prop
variable h1 : ∀ x, A x ∨ B x
variable h2 : ∀ x, A x → C x
variable h3 : ∀ x, B x → C x
example : ∀ x, C x :=
sorry
end
4. Fill in the sorry’s below, to prove the barber paradox.

open classical -- not needed, but you can use it
-- This is an exercise from Chapter 4. Use it as an axiom here.

axiom not_iff_not_self (P : Prop) : ¬ (P ↔ ¬ P)
example (Q : Prop) : ¬ (Q ↔ ¬ Q) :=
not_iff_not_self Q
section
variable Person : Type
variable shaves : Person → Person → Prop
variable barber : Person
variable h : ∀ x, shaves barber x ↔ ¬ shaves x x
-- Show the following:

example : false :=
sorry
end

section
variable U : Type
example : (∃ x, A x) → ∃ x, A x ∨ B x :=
sorry
end

section
variable U : Type
variable h1 : ∀ x, A x → B x
variable h2 : ∃ x, A x
example : ∃ x, B x :=
sorry
end

variable U : Type
example (h1 : ∃ x, A x ∧ B x) (h2 : ∀ x, B x → C x) :

∃ x, A x ∧ C x :=
sorry
8. Complete these proofs.

variable U : Type
example : (¬ ∃ x, A x) → ∀ x, ¬ A x :=
sorry
9.5. Exercises 81
example : (∀ x, ¬ A x) → ¬ ∃ x, A x :=
sorry

variable U : Type
variables R : U → U → Prop
example : (∃ x, ∀ y, R x y) → ∀ y, ∃ x, R x y :=
sorry
10. The following exercise shows that in the presence of reflexivity, the rules for symmetry and transitivity
are equivalent to a single rule.
theorem foo {A : Type} {a b c : A} : a = b → c = b → a = c :=
sorry
-- notice that you can now use foo as a rule. The curly braces mean that
-- you do not have to give A, a, b, or c
section
variable A : Type
variables a b c : A
example (h1 : a = b) (h2 : c = b) : a = c :=

foo h1 h2
end
section
variable {A : Type}
variables {a b c : A}
-- replace the sorry with a proof, using foo and rfl, without using eq.symm.
theorem my_symm (h : b = a) : a = b :=
sorry
-- now use foo, rfl, and my_symm to prove transitivity

theorem my_trans (h1 : a = b) (h2 : b = c) : a = c :=
sorry
end
11. Replace each “sorry” below by the correct axiom from the list.
-- these are the axioms for a commutative ring
#check @add_assoc
#check @add_comm
#check @add_zero
#check @zero_add
#check @mul_assoc
#check @mul_comm
#check @mul_one
#check @one_mul
#check @left_distrib
#check @right_distrib
#check @add_left_neg
#check @add_right_neg

#check @sub_eq_add_neg
variables x y z : int
theorem t1 : x - x = 0 :=
calc
x - x = x + -x : by rw sub_eq_add_neg
... = 0 : by rw add_right_neg
theorem t2 (h : x + y = x + z) : y = z :=
calc
y = 0 + y : by rw zero_add
... = (-x + x) + y : by rw add_left_neg
... = -x + (x + y) : by rw add_assoc
... = -x + (x + z) : by rw h
... = (-x + x) + z : by rw add_assoc
... = 0 + z : by rw add_left_neg
... = z : by rw zero_add
theorem t3 (h : x + y = z + y) : x = z :=
calc
x = x + 0 : sorry
... = x + (y + -y) : sorry
... = (x + y) + -y : sorry
... = (z + y) + -y : by rw h
... = z + (y + -y) : sorry
... = z + 0 : sorry
... = z : sorry
theorem t4 (h : x + y = 0) : x = -y :=
calc
x = x + 0 : by rw add_zero
... = x + (y + -y) : by rw add_right_neg
... = (x + y) + -y : by rw add_assoc
... = 0 + -y : by rw h
... = -y : by rw zero_add
theorem t5 : x * 0 = 0 :=
have h1 : x * 0 + x * 0 = x * 0 + 0, from
calc
x * 0 + x * 0 = x * (0 + 0) : sorry
... = x * 0 : sorry
... = x * 0 + 0 : sorry,
show x * 0 = 0, from t2 _ _ _ h1
theorem t6 : x * (-y) = -(x * y) :=

have h1 : x * (-y) + x * y = 0, from
calc
x * (-y) + x * y = x * (-y + y) : sorry
... = x * 0 : sorry
... = 0 : by rw t5 x,
show x * (-y) = -(x * y), from t4 _ _ h1
theorem t7 : x + x = 2 * x :=
calc
x + x = 1 * x + 1 * x : by rw one_mul
... = (1 + 1) * x : sorry
... = 2 * x : rfl
9.5. Exercises 83

CHAPTER
TEN
SEMANTICS OF FIRST ORDER LOGIC
In Chapter 6, we emphasized a distinction between the syntax and the semantics of propositional logic.
Syntactic questions have to do with the formal structure of formulas and the conditions under which different
types of formulas can be derived. Semantic questions, on the other hand, concern the truth of a formula
relative to some truth assignment.
As you might expect, we can make a similar distinction in the setting of first order logic. The previous two
chapters have focused on syntax, but some semantic ideas have slipped in. Recall the running example with
domain of interest N, constant symbols 0, 1, 2, 3, function symbols add and mul, and predicate symbols
even, prime, equals, lt, etc. We know that the sentence ∀y lt(0, y) is true in this example, if lt is interpreted
as the less-than relation on the natural numbers. But if we consider the domain Z instead of N, that same
formula becomes false. The sentence ∀y lt(0, y) is also false if we consider the domain N, but (somewhat
perversely) interpret the predicate lt(x, y) as the relation “x is greater than y” on the natural numbers.
This indicates that the truth or falsity or a first order sentence can depend on how we interpret the quantifiers
and basic relations of the language. But some formulas are true under any interpretation: for instance,
∀y (lt(0, y) → lt(0, y)) is true of under all the interpretations considered in the last paragraph, and, indeed,
under any interpretation we choose. A sentence like this is said to be valid; this is the analogue of a tautology
in propositional logic, which is true under every possible truth assignment.
We can broaden the analogy: a “model” in first order logic is the analogue of a truth assignment in propo-
sitional logic. In the propositional case, choosing a truth assignment allowed us to assign truth values to all
formulas of the language; now, choosing an model will allow us to assign truth values to all sentences of a
first order language. The aim of the next section is to make this notion more precise.
10.1 Interpretations
The symbols of the language in our running example – 0, 1, add, prime, and so on – have very suggestive
names. When we interpret sentences of this language over the domain N, for example, it is clear for which
elements of the domain prime “should” be true, and for which it “should” be false. But let us consider a
first order language that has only two unary predicate symbols fancy and tall. If we take our domain to be
N, is the sentence ∀x (fancy(x) → tall(x)) true or false?
The answer, of course, is that we don’t have enough information to say. There’s no obvious meaning to the
predicates fancy or tall, at least not when we apply them to natural numbers. To make sense of the sentence,
we need to know which numbers are fancy and which ones are tall. Perhaps multiples of 10 are fancy, and
even numbers are tall; in this case, the formula is true, since every multiple of 10 is even. Perhaps prime
numbers are fancy and odd numbers are tall; then the formula is false, since 2 is fancy but not tall.
We call each of these descriptions an interpretation of the predicate symbols fancy and tall in the domain
N. Formally, an interpretation of a unary predicate P in a domain D is the set of elements of D for which
P is true. For an example, the standard interpretation of prime in N that we used above was just the set of
prime natural numbers.
85
We can interpret constant, function, and relation symbols in a similar way. An interpretation of constant
symbol c in domain D is an element of D. An interpretation of a function symbol f with arity n is a function
that maps n elements of D to another element of D. An interpretation of a relation symbol R with arity n
is the set of n tuples of elements of D for which R is true.
It is important to emphasize the difference between a syntactic predicate symbol (or function symbol, or
constant symbol) and the semantic predicate (or function, or object) to which it is interpreted. The former
is a symbol, relates to other symbols, and has no meaning on its own until we specify an interpretation.
Strictly speaking, it makes no sense to write prime(3), where prime is a predicate symbol and 3 is a natural
number, since the argument to prime is supposed to be a syntactic term. Sometimes we may obscure this
distinction, as above when we specified a language with constant symbols 0, 1, and 2. But there is still a
fundamental distinction between the objects of the domain and the symbols we use to represent them.
Sometimes, when we interpret a language in a particular domain, it is useful to implicitly introduce new
constant symbols into the language to denote elements of this domain. Specifically, for each element a of the
domain, we introduce a constant symbol ā, which is interpreted as a. Then the expression prime(3̄) does
make sense. Interpreting the predicate symbol prime in the natural way, this expression will evaluate to
true. We think of 3̄ as a linguistic “name” that represents the natural number 3, in the same way that the
phrase “Bob Dylan” is a name that represents the flesh-and-blood folk singer.
10.2 Truth in a Model
Fix a first-order language. Suppose we have chosen a domain D to interpret the language, along with an
interpretation in D of each of the symbols of that language. We will call this structure — the domain D,
paired with the interpretation — a model for the language. A model for a first-order language is directly
analogous to a truth assignment for propositional logic, because it provides all the information we need to
determine the truth value of each sentence in the language.
The procedure for evaluating the truth of a sentence based on a model works the way you think it should, but
the formal description is subtle. Recall the difference between terms and assertions that we made earlier in
Chapter 4. Terms, like a, x + y, or f (c), are meant to represent objects. A term does not have a truth value,
since (for example) it makes no sense to ask whether 3 is true or false. Assertions, like P (a), R(x, f (y)), or
a + b > a ∧ prime(c), apply predicate or relation symbols to terms to produce statements that could be true
or false.
The interpretation of a term in a model is an element of the domain of that model. The model directly
specifies how to interpret constant symbols. To interpret a term f (t) created by applying a function symbol
to another term, we interpret the term t, and then apply the interpretation of f to this term. (This process
makes sense, since the interpretation of f is a function on the domain.) This generalizes to functions of
higher arity in the obvious way. We will not yet interpret terms that include free variables like x and y, since
these terms do not pick out unique elements of the domain. (The variable x could potentially refer to any
object.)
For example, suppose we have a language with two constant symbols, a and b, a unary function symbol f ,
and a binary function symbol g. Let M be the model with domain N, where a and b are interpreted as 3
and 5, respectively, f (x) is interpreted as the function which maps any natural number n to n2 , and g is the
addition function. Then the term g(f (a), b) denotes the natural number 32 + 5 = 14.
Similarly, the interpretation of an assertion is a value T or F. For the sake of brevity, we will introduce new
notation here: if A is an assertion and M is a model of the language of A, we write M |= A to mean that A
evaluates to T in M, and M ̸|= A to mean that A evaluates to F. (You can read the symbol |= as “satisfies”
or “validates.”)
To interpret a predicate or relation applied to some terms, we first interpret those terms, and then see if
the interpretation of the relation symbol is true of those objects. To continue with the example, suppose
our language also has a relation symbol R, and we extend M to interpret R as the greater-than-or-equal-to
86 Chapter 10. Semantics of First Order Logic

relation. Then we have M ̸|= R(a, b), since 3 is not greater than 5, but M |= R(g(f (a)), b), since 14 is
greater than 5.
Interpreting expressions using the logical connectives ∧, ∨, →, and ¬ works exactly as it did in the proposi-
tional setting. M |= A ∧ B exactly when M |= A and M |= B, and so on.
We still need to explain how to interpret existential and universal expressions. We saw that ∃x A intuitively
meant that there was some element of the domain that would make A true, when we “replaced” the variable
x with that element. To make this a bit more precise, we say that M |= ∃xA exactly when there is an
element a in the domain of M such that, when we interpret x as a, then M |= A. To continue the example
above, we have M |= ∃x (R(x, b)), since when we interpret x as 6 we have M |= R(x, b).
More concisely, we can say that M |= ∃x A when there is an a in the domain of M such that M |= A[ā/x].
The notation A[ā/x] indicates that every occurrence of x in A has been replaced by the symbol ā.
Finally, remember that ∀x A meant that A was true for all possible values of x. We make this precise by
saying that M |= ∀x A exactly when for every element a in the domain of M, interpreting x as a gives
that M |= A. Alternatively, we can say that M |= ∀x A when for every a in the domain of M, we have
M |= A[ā/x]. In our example above, M ̸|= ∀x (R(x, b)), since when we interpret x as 2 we do not have
M |= R(x, b).
These rules allow us to determine the truth value of any sentence in a model. (Remember, a sentence is
a formula with no free variables.) There are some subtleties: for instance, we’ve implicitly assumed that
our formula doesn’t quantify over the same variable twice, as in ∀x ∃x A. But for the most part, the
interpretation process tells us to “read” a formula as talking directly about objects in the domain.
10.3 Examples
Take a simple language with no constant symbols, one relation symbol ≤, and one binary function symbol +.
Our model M will have domain N, and the symbols will be interpreted as the standard less-than-or-equal-to
relation and addition function.
Think about the following questions before you read the answers below. Remember, our domain is N, not Z
or any other number system.
1. Is it true that M |= ∃x (x ≤ x)? What about M |= ∀x (x ≤ x)?
2. Similarly, what about M |= ∃x (x + x ≤ x)? M |= ∀x (x + x ≤ x)?
3. Do the sentences ∃x ∀y (x ≤ y) and ∀x ∃y (x ≤ y) mean the same thing? Are they true or false?
4. Can you think of a formula A in this language, with one free variable x, such that M |= ∀x A but
M ̸|= ∃x A?
These questions indicate a subtle, and often tricky, interplay between the universal and existential quantifiers.
Once you’ve thought about them a bit, read the answers:
1. Both of these statements are true. For the former, we can (for example) interpret x as the natural
number 0. Then, M |= x ≤ x, so the existential is true. For the latter, pick an arbitrary natural
number n; it is still the case that when we interpret x as n, we have M |= x ≤ x.
2. The first statement is true, since we can interpret x as 0. The second statement, though, is false. When
we interpret x as 1 (or, in fact, as any natural number besides 0), we see that M ̸|= x + x ≤ x.
3. These sentences do not mean the same thing, although in the specified model, both are true. The first
expresses that some natural number is less than or equal to every natural number. This is true: 0 is
less than or equal to every natural number. The second sentence says that for every natural number,
there is another natural number at least as big. Again, this is true: every natural number a is less
10.3. Examples 87
than or equal to a. If we took our domain to be Z instead of N, the first sentence would be false, while
the second would still be true.
4. The situation described here is impossible in our model. If M |= ∀xA, then M |= A[0̄/x], which
implies that M |= ∃xA. The only time this situation can happen is when the domain of our model is
empty.
Now consider a different language with constant symbol 2, predicate symbols prime and odd, and binary
relation <, interpreted in the natural way over domain N. The sentence ∀x ((2 < x ∧ prime(x)) → odd(x))
expresses the fact that every prime number bigger than 2 is odd. It is an example of relativization, discussed
in Section 7.4. We can now see semantically how relativization works. This sentence is true in our model
if, for every natural number n, interpreting x as n makes the sentence true. If we interpret x as 0, 1, or 2,
or as any non-prime number, the hypothesis of the implication is false, and thus (2 < x ∧ prime(x)) is true.
Otherwise, if we interpret x as a prime number bigger than 2, both the hypothesis and conclusion of the
implication are true, and (2 < x ∧ prime(x)) is again true. Thus the universal statement holds. It was an
example like this that partially motivated our semantics for implication back in Chapter 3; any other choice
would make relativization impossible.
For the next example, we will consider models that are given by a rectangular grid of “dots.” Each dot has
a color (red, blue, or green) and a size (small or large). We use the letter R to represent a large red dot and
r to represent a small red dot, and similarly for G, g, B, b.
The logical language we use to describe our dot world has predicates red, green, blue, small and large,
which are interpreted in the obvious ways. The relation adj(x, y) is true if the dots referred to by x and
y are touching, not on a diagonal. The relations same-color(x, y), same-size(x, y), same-row(x, y), and
same-column(x, y) are also self-explanatory. The relation left-of(x, y) is true if the dot referred to by x is left
of the dot referred to by y, regardless of what rows the dots are in. The interpretations of right-of, above,
and below are similar.
Consider the following sentences:
1. ∀x (green(x) ∨ blue(x))
2. ∃x, y (adj(x, y) ∧ green(x) ∧ green(y))
3. ∃x ((∃z right-of(z, x)) ∧ (∀y (left-of(x, y) → blue(y) ∨ small(y))))
4. ∀x (large(x) → ∃y (small(y) ∧ adj(x, y)))
5. ∀x (green(x) → ∃y (same-row(x, y) ∧ blue(y)))
6. ∀x, y (same-row(x, y) ∧ same-column(x, y) → x = y)
7. ∃x ∀y (adj(x, y) → ¬same-size(x, y))
8. ∀x ∃y (adj(x, y) ∧ same-color(x, y))
9. ∃y ∀x (adj(x, y) ∧ same-color(x, y))
10. ∃x (blue(x) ∧ ∃y (green(y) ∧ above(x, y)))
We can evaluate them in this particular model:
R r g b
R b G b
B B B b
There they have the following truth values:

1. false
2. true

3. true
4. false
5. true
6. true
7. false
8. true
9. false
10. false
For each sentence, see if you can find a model that makes the sentence true, and another that makes it false.
For an extra challenge, try to make all of the sentences true simultaneously. Notice that you can use any
number of rows and any number of columns.
10.4 Validity and Logical Consequence
We have seen that whether a formula is true or false often depends on the model we choose. Some formulas,
though, are true in every possible model. An example we saw earlier was ∀y (lt(0, y) → lt(0, y)). Why is this
sentence valid? Suppose M is an arbitrary model of the language, and suppose a is an arbitrary element
of the domain of M. Either M |= lt(0, ā) or M |= ¬lt(0, ā). In either case, the propositional semantics of
implication guarantee that M |= lt(0, ā) → lt(0, ā). We often write |= A to mean that A is a valid.
In the propositional setting, there is an easy method to figure out if a formula is a tautology or not. Writing
the truth table and checking for any rows ending with F is algorithmic, and we know from the beginning
exactly how large the truth table will be. Unfortunately, we cannot do the same for first-order formulas.
Any language has infinitely many models, so a “first-order” truth table would be infinitely long. To make
matters worse, even checking whether a formula is true in a single model can be a non-algorithmic task. To
decide whether a universal statement like ∀x P (x) is true in a model with an infinite domain, we might have
to check whether P is true of infinitely many elements.
This is not to say that we can never figure out if a first-order sentence is a tautology. For example, we have
argued that ∀y (lt(0, y) → lt(0, y)) was one. It is just a more difficult question than for propositional logic.
As was the case with propositional logic, we can extend the notion of validity to a notion of logical conse-
quence. Fix a first-order language, L. Suppose Γ is a set of sentences in L, and A is a sentence of L. We will
say that A is a logical consequence of Γ if every model of Γ is a model of A. This is one way of spelling out
that A is a “necessary consequence” of A: under any interpretation, if the hypotheses in Γ come out true, A
is true as well.
10.5 Soundness and Completeness
In propositional logic, we saw a close connection between the provable formulas and the tautologies – specif-
ically, a formula is provable if and only if it is a tautology. More generally, we say that a formula A is a
logical consequence of a set of hypotheses, Γ, if and only if there is a natural deduction proof of A from Γ.
It turns out that the analogous statements hold for first order logic.
The “soundness” direction — the fact that if A is provable from Γ then A is true in any model of Γ — holds
for reasons that are similar to the reasons it holds in the propositional case. Specifically, the proof proceeds
by showing that each rule of natural deduction preserves the truth in a model.
10.4. Validity and Logical Consequence 89

The completeness theorem for first order logic was first proved by Kurt Gödel in his 1929 dissertation.
Another, simpler proof was later provided by Leon Henkin.
Theorem. If a formula A is a logical consequence of a set of sentences Γ, then A is provable from Γ.
Compared to the version for propositional logic, the first order completeness theorem is harder to prove. We
will not go into too much detail here, but will indicate some of the main ideas. A set of sentences is said
to be consistent if you cannot prove a contradiction from those hypotheses. Most of the work in Henkin’s
proof is done by the following “model existence” theorem:
Theorem. Every consistent set of sentences has a model.
From this theorem, it is easy to deduce the completeness theorem. Suppose there is no proof of A from Γ.
Then the set Γ ∪ {¬A} is consistent. (If we could prove ⊥ from Γ ∪ {¬A}, then by the reductio ad absurdum
rule we could prove A from Γ.) By the model existence theorem, that means that there is a model M of
Γ ∪ {¬A}. But this is a model of Γ that is not a model of A, which means that A is not a logical consequence
of Γ.
The proof of the model existence theorem is intricate. Somehow, from a consistent set of sentences, one has
to build a model. The strategy is to build the model out of syntactic entities, in other words, to use terms
in an expanded language as the elements of the domain.
The moral here is much the same as it was for propositional logic. Because we have developed our syntactic
rules with a certain semantics in mind, the two exhibit different sides of the same coin: the provable sentences
are exactly the ones that are true in all models, and the sentences that are provable from a set of hypotheses
are exactly the ones that are true in all models of those hypotheses.
We therefore have another way to answer the question posed in the previous section. To show that a sentence
is valid, there is no need to check its truth in every possible model. Rather, it suffices to produce a proof.
10.6 Exercises
1. In a first-order language with a binary relation, R(x, y), consider the following sentences:
• ∃x ∀y R(x, y)
• ∃y ∀x R(x, y)
• ∀x, y (R(x, y) ∧ x ̸= y → ∃z (R(x, z) ∧ R(z, y) ∧ x ̸= z ∧ y ̸= z))
For each of the following structures, determine whether of each of those sentences is true or false.
• the structure (N, ≤), that is, the interpretation in the natural numbers where R is ≤
• the structure (Z, ≤)
• the structure (Q, ≤)
• the structure (N, |), that is, the interpretation in the natural numbers where R is the “divides”
relation
• the structure (P (N), ⊆), that is, the interpretation where variables range over sets of natural
numbers, where R is interpreted as the subset relation.

2. Create a 4 x 4 “dots” world that makes all of the following sentences true:
• ∀x (green(x) ∨ blue(x))
• ∃x, y (adj(x, y) ∧ green(x) ∧ green(y))
• ∃x (∃z right-of(z, x) ∧ ∀y (left-of(x, y) → blue(y) ∨ small(y)))
• ∀x (large(x) → ∃y (small(y) ∧ adj(x, y)))
• ∀x (green(x) → ∃y (same-row(x, y) ∧ blue(y)))
• ∀x, y (same-row(x, y) ∧ same-column(x, y) → x = y)
• ∃x ∀y (adj(x, y) → ¬same-size(x, y))
• ∀x ∃y (adj(x, y) ∧ same-color(x, y))
• ∃y ∀x (adj(x, y) → same-color(x, y))
• ∃x (blue(x) ∧ ∃y (green(y) ∧ above(x, y)))
3. Fix a first-order language L, and let A and B be any two sentences in L. Remember that ⊨ A means
that A is valid. Unpacking the definition, show that if ⊨ A ∧ B, then ⊨ A and ⊨ B.
4. Give a concrete example to show that ⊨ A ∨ B does not necessarily imply ⊨ A or ⊨ B. In other words,
pick a language L and choose particular sentences A and B such that A ∨ B is valid, but neither A nor
B is valid.
10.6. Exercises 91

CHAPTER
ELEVEN
SETS
We have come to a turning point in this textbook. We will henceforth abandon natural deduction, for
the most part, and focus on ordinary mathematical proofs. We will continue to think about how informal
mathematics can be represented in symbolic terms, and how the rules of natural deduction play out in the
informal setting. But the emphasis will be on writing ordinary mathematical arguments, not designing proof
trees. Lean will continue to serve as a bridge between the informal and formal realms.
In this chapter, we consider a notion that has come to play a fundamental role in mathematical reasoning,
namely, that of a “set.”
11.1 Elementary Set Theory
In a publication in the journal Mathematische Annalen in 1895, the German mathematician Georg Cantor
presented the following characterization of the notion of a set (or Menge, in his terminology):
By a set we mean any collection M of determinate, distinct objects (called the elements of M) of
our intuition or thought into a whole.
Since then, the notion of a set has been used to unify a wide range of abstractions and constructions.
Axiomatic set theory, which we will discuss in a later chapter, provides a foundation for mathematics in
which everything can be viewed as a set.
On a broad construal, any collection can be a set; for example, we can consider the set whose elements
are Ringo Star, the number 7, and the set whose only member is the Empire State Building. With such a
broad notion of set we have to be careful: Russell’s paradox has us consider the set S of all sets that are not
elements of themselves, which leads to a contradiction when we ask whether S is an element of itself. (Try
it!) The axioms of set theory tell us what sets exist, and have been carefully designed to avoid paradoxical
sets like that of the Russell paradox.
In practice, mathematicians are not so freewheeling in their use of sets. Typically, one fixes a domain such as
the natural numbers, and consider subsets of that domain. In other words, we consider sets of numbers, sets
of points, sets of lines, and so on, rather than arbitrary “sets.” In this text, we will adopt this convention:
when we talk about sets, we are always implicitly talking about sets of elements of some domain.
Given a set A of objects in some domain and an object x, we write x ∈ A to say that x is an element of
A. Cantor’s characterization suggests that whenever we have some property, P , of a domain, we can form
the set of elements that have that property. This is denoted using “set-builder notation” as {x | P (x)}. For
example, we can consider all the following sets of natural numbers:
• {n | n is even}
• {n | n is prime}
• {n | n is prime and greater than 2}
93
• {n | n can be written as a sum of squares}

• {n | n is equal to 1, 2, or 3}
This last set is written more simply {1, 2, 3}. If the domain is not clear from the context, we can specify it
by writing it explicitly, for example, in the expression {n ∈ N | n is even}.
Using set-builder notation, we can define a number of common sets and operations. The empty set, ∅, is the
set with no elements:
∅ = {x | false}
Dually, we can define the universal set, U, to be the set consisting of every element of the domain:
U = {x | true}
Given two sets A and B, we define their union to be the set of elements in either one:
A ∪ B = {x | x ∈ A or x ∈ B}
And we define their intersection to be the set of elements of both:
A ∩ B = {x | x ∈ A and x ∈ B}
We define the complement of a set of A to be the set of elements that are not in A:
A = {x | x ∈
/ A}
We define the set difference of two sets A and B to be the set of elements in A but not B:
A \ B = {x | x ∈ A and x ∈
/ B}
Two sets are said to be equal if they have exactly the same elements. If A and B are sets, A is said to be
a subset of B, written A ⊆ B, if every element of A is an element of B. Notice that A is equal to B if and
only if A is a subset of B and B is a subset of A.
Notice also that just everything we have said about sets so far is readily representable in symbolic logic. We
can render the defining properties of the basic sets and constructors as follows:
∀x (x ∈ ∅ ↔ ⊥)
∀x (x ∈ U ↔ ⊤)
∀x (x ∈ A ∪ B ↔ x ∈ A ∨ x ∈ B)
∀x (x ∈ A ∩ B ↔ x ∈ A ∧ x ∈ B)
∀x (x ∈ A ↔ x ∈
/ A)
∀x (x ∈ A \ B ↔ x ∈ A ∧ x ∈
/ B)
The assertion that A is a subset of B can be written ∀x (x ∈ A → x ∈ B), and the assertion that A is equal
to be can be written ∀x (x ∈ A ↔ x ∈ B). These are all universal statements, that is, statements with
universal quantifiers in front, followed by basic assertions and propositional connectives. What this means
is that reasoning about sets formally often amounts to using nothing more than the rules for the universal
quantifier together with the rules for propositional logic.
Logicians sometimes describe ordinary mathematical proofs as informal, in contrast to the formal proofs in
natural deduction. When writing informal proofs, the focus is on readability. Here is an example.
Theorem. Let A, B, and C denote sets of elements of some domain, U. Then A∩(B ∪C) = (A∩B)∪(A∩C).
94 Chapter 11. Sets

Proof. Let x be arbitrary, and suppose x is in A ∩ (B ∪ C). Then x is in A, and either x is in B or x is in

C. In the first case, x is in A and B, and hence in A ∩ B. In the second case, x is in A and C, and hence
A ∩ C. Either way, we have that x is in (A ∩ B) ∪ (A ∩ C).
Conversely, suppose x is in (A ∩ B) ∪ (A ∩ C). There are now two cases.
First, suppose x is in A ∩ B. Then x is in both A and B. Since x is in B, it is also in B ∪ C, and so x is in
A ∩ (B ∪ C).
The second case is similar: suppose x is in A ∩ C. Then x is in both A and C, and so also in B ∪ C. Hence,
in this case also, x is in A ∩ (B ∪ C), as required.
Notice that this proof does not look anything like a proof in symbolic logic. For one thing, ordinary proofs
tend to favor words over symbols. Of course, mathematics uses symbols all the time, but not in place of
words like “and” and “not”; you will rarely, if ever, see the symbols ∧ and ¬ in a mathematics textbook,
unless it is a textbook specifically about logic.
Similarly, the structure of an informal proof is conveyed with ordinary paragraphs and punctuation. Don’t
rely on pictorial diagrams, line breaks, and indentation to convey the structure of a proof. Rather, you
should rely on literary devices like signposting and foreshadowing. It is often helpful to present an outline
of a proof or the key ideas before delving into the details, and the introductory sentence of a paragraph can
help guide a reader’s expectations, just as it does in an expository essay.
Nonetheless, you should be able to see elements of natural deduction implicitly in the proof above. In formal
terms, the theorem is equivalent to the assertion
∀x (x ∈ A ∩ (B ∪ C) ↔ x ∈ (A ∩ B) ∪ (A ∩ C)),
and the proof proceeds accordingly. The phrase “let x be arbitrary” is code for the ∀ introduction rule, and
the form of the rest of the proof is a ↔ introduction. Saying that x is in A ∩ (B ∪ C) is implicitly an “and,”
and the argument uses ∧ elimination to get x ∈ A and x ∈ B ∪ C. Saying x ∈ B ∪ C is implicitly an “or,”
and the proof then splits on cases, depending on whether x ∈ B or x ∈ C.
Modulo the unfolding of definition of intersection and union in terms of “and” and “or,” the “only if” direction
of the previous proof could be represented in natural deduction like this:
1 1
y ∈ A ∩ (B ∪ C) y ∈ A ∩ (B ∪ C)
2 2
y∈A y∈B y∈A y∈C
1
y ∈ A ∩ (B ∪ C) y ∈A∩B y ∈A∩C
y ∈B∪C y ∈ (A ∩ B) ∪ (A ∩ C) y ∈ (A ∩ B) ∪ (A ∩ C)
2 ..
y ∈ (A ∩ B) ∪ (A ∩ C) .
1
y ∈ A ∩ (B ∪ C) ↔ y ∈ (A ∩ B) ∪ (A ∩ C)
∀x (x ∈ A ∩ (B ∪ C) ↔ x ∈ (A ∩ B) ∪ (A ∩ C))
In the next chapter, we will see that this logical structure is made manifest in Lean. But writing long proofs
in natural deduction is not the most effective to communicate the mathematical ideas. So our goal here is
to teach you to think in terms of natural deduction rules, but express the steps in ordinary English.
Here is another example.
Theorem. (A \ B) \ C = A \ (B ∪ C).
Proof. Let x be arbitrary, and suppose x is in (A \ B) \ C. Then x is in A \ B but not C, and hence it is
in A but not in B or C. This means that x is in A but not B ∪ C, and so in A \ (B ∪ C).
11.1. Elementary Set Theory 95

Conversely, suppose x is in A \ (B ∪ C). Then x is in A, but not in B ∪ C. In particular, x is in neither B

nor C, because otherwise it would be in B ∪ C. So x is in A \ B, and hence in (A \ B) \ C.
Perhaps the biggest difference between informal proofs and formal proofs is the level of detail. Informal
proofs will often skip over details that are taken to be “straightforward” or “obvious,” devoting more effort
to spelling out inferences that are novel of unexpected.
Writing a good proof is like writing a good essay. To convince your readers that the conclusion is correct,
you have to get them to understand the argument, without overwhelming them with unnecessary details.
It helps to have a specific audience in mind. Try speaking the argument aloud to friends, roommates, and
family members; if their eyes glaze over, it is unreasonable to expect anonymous readers to do better.
One of the best ways to learn to write good proofs is to read good proofs, and pay attention to the style of
writing. Pick an example of a textbook that you find especially clear and engaging, and think about what
makes it so.
Natural deduction and formal verification can help you understand the components that make a proof correct,
but you will have to develop an intuitive feel for what makes a proof easy and enjoyable to read.
11.2 Calculations with Sets
Calculation is a central to mathematics, and mathematical proofs often involve carrying out a sequence of
calculations. Indeed, a calculation can be viewed as a proof in and of itself that two expressions describe the
same entity.
In high school algebra, students are often asked to prove identities like the following:
n(n+1) (n+1)(n+2)
Proposition. 2 + (n + 1) = 2 , for every natural number n.
In some places, students are asked to write proofs like this:
Proof.
n(n + 1) (n + 1)(n + 2)
+ (n + 1) =?
2 2
n2 + n 2n + 2 n2 + 3n + 2
+ =?
2 2 2
n2 + n + 2n + 2 n2 + 3n + 2
=?
2 2
n2 + 3n + 2 n2 + 3n + 2
=
2 2
Mathematicians generally cringe when they see this. Don’t do it! It looks like an instance of forward
reasoning, where we start with a complex identity and end up proving x = x. Of course, what is really
meant is that each line follows from the next. There is a way of expressing this, with the phrase “it suffices
to show.” The following presentation comes closer to mathematical vernacular:
96 Chapter 11. Sets

Proof. We want to show

n(n + 1) (n + 1)(n + 2)
+ (n + 1) = .
2 2
To do that, it suffices to show
n2 + n 2n + 2 n2 + 3n + 2
+ = .
2 2 2
For that, it suffices to show
n2 + n + 2n + 2 n2 + 3n + 2
= .
2 2
But this last equation is clearly true.
The narrative doesn’t flow well, however. Sometimes there are good reasons to work backwards in a proof,
but in this case it is easy to present the proof in a more forward-directed manner. Here is one example:
Proof. Calculating on the left-hand side, we have
n(n + 1) n2 + n 2n + 2
+ (n + 1) = +
2 2 2
n2 + n + 2n + 2
=
2
n2 + 3n + 2
= .
2
On the right-hand side, we also have
(n + 1)(n + 2) n2 + 3n + 2
= .
2 2
n(n+1) n2 +3n+2
So 2 + (n + 1) = 2 , as required.
Mathematicians often use the abbreviations “LHS” and “RHS” for “left-hand side” and “right-hand side,”
respectively, in situations like this. In fact, here we can easily write the proof as a single forward-directed
calculation:
Proof.
n(n + 1) n2 + n 2n + 2
+ (n + 1) = +
2 2 2
n2 + n + 2n + 2
=
2
n2 + 3n + 2
=
2
(n + 1)(n + 2)
= .
2
11.2. Calculations with Sets 97

Such a proof is clear, compact, and easy to read. The main challenge to the reader is to figure out what
justifies each subsequent step. Mathematicians sometimes annotate such a calculation with additional infor-
mation, or add a few words of explanation in the text before and/or after. But the ideal situation is to carry
out the calculation is small enough steps so that each step is straightforward, and needs to no explanation.
(And, once again, what counts as “straightforward” will vary depending on who is reading the proof.)
We have said that two sets are equal if they have the same elements. In the previous section, we proved that
two sets are equal by reasoning about the elements of each, but we can often be more efficient. Assuming
A, B, and C are subsets of some domain U, the following identities hold:
• A∪A=U
• A∩A=∅
• A=A
• A∪A=A
• A∩A=A
• A∪∅=A
• A∩∅=∅
• A∪U =U
• A∩U =A
• A∪B =B∪A
• A∩B =B∩A
• (A ∪ B) ∪ C = A ∪ (B ∪ C)
• (A ∩ B) ∩ C = A ∩ (B ∩ C)
• A∩B =A∪B
• A∪B =A∩B
• A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
• A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)
• A ∩ (A ∪ B) = A
• A ∪ (A ∩ B) = A
This allows us to prove further identities by calculating. Here is an example.
Theorem. Let A and B be subsets of some domain U. Then (A ∩ B) ∪ B = A ∪ B.

Proof.
(A ∩ B) ∪ B = (A ∪ B) ∩ (B ∪ B)
= (A ∪ B) ∩ U
= A ∪ B.
Here is another example.
98 Chapter 11. Sets

Theorem. Let A and B be subsets of some domain U. Then (A \ B) ∪ (B \ A) = (A ∪ B) \ (A ∩ B).

Proof.
(A \ B) ∪ (B \ A) = (A ∩ B) ∪ (B ∩ A)
= ((A ∩ B) ∪ B) ∩ ((A ∩ B) ∪ A)
= ((A ∪ B) ∩ (B ∪ B)) ∩ ((A ∪ A) ∩ (B ∪ A))
= ((A ∪ B) ∩ U) ∩ (U ∩ B ∩ A)
= (A ∪ B) ∩ (A ∩ B)
= (A ∪ B) \ (A ∩ B)
Classically, you may have noticed that propositions, under logical equivalence, satisfy identities similar to
sets. That is no coincidence; both are instances of boolean algebras. Here are the identities above translated
to the language of a boolean algebra:
• A ∨ ¬A = ⊤
• A ∧ ¬A = ⊥
• ¬¬A = A
• A∨A=A
• A∧A=A
• A∨⊥=A
• A∧⊥=⊥
• A∨⊤=⊤
• A∧⊤=A
• A∨B =B∨A
• A∧B =B∧A
• (A ∨ B) ∨ C = A ∨ (B ∨ C)
• (A ∧ B) ∧ C = A ∧ (B ∧ C)
• ¬A ∧ B = ¬A ∨ ¬B
• ¬A ∨ B = ¬A ∧ ¬B
• A ∧ (B ∨ C) = (A ∧ B) ∨ (A ∧ C)
• A ∨ (B ∧ C) = (A ∨ B) ∧ (A ∨ C)
• A ∧ (A ∨ B) = A
• A ∨ (A ∧ B) = A
Translated to propositions, the first theorem above is as follows:
Theorem. Let A and B be elements of a boolean algebra. Then (A ∧ ¬B) ∨ B = B.

Proof.
(A ∧ ¬B) ∨ B = (A ∨ B) ∧ (¬B ∨ B)
= (A ∨ B) ∧ ⊤
= (A ∨ B).
11.2. Calculations with Sets 99

11.3 Indexed Families of Sets
If I is a set, we will sometimes wish to consider a family (Ai )i∈I of sets indexed by elements of I. For
example, we might be interested in a sequence
A0 , A 1 , A 2 , . . .
of sets indexed by the natural numbers. The concept is best illustrated by some examples.
• For each natural number n, we can define the set An to be the set of people alive today that are of age
n. For each age we have the corresponding set. Someone of age 20 is an element of the set A20 , while
a newborn baby is an element of A0 . The set A200 is empty. This family (An )n∈N is a is a family of
sets indexed by the natural numbers.
• For every real number r we can define Br to be the set of positive real numbers larger than r, so
Br = {x ∈ R | x > r and x > 0}. Then (Br )r∈R is a family of sets indexed by the real numbers.
• For every natural number n we can define Cn = {k ∈ N | k is a divisor of n} as the set of divisors of n.
Given a family (Ai )i∈I of sets indexed by I, we can form its union:
∪
Ai = {x | x ∈ Ai for some i ∈ I}
i∈I
We can also form the intersection of a family of sets:

∩
Ai = {x | x ∈ Ai for every i ∈ I}
i∈I
∪ ∩
So an element x is in i∈I Ai if and only if x is in Ai for some i in I, and x is in i∈I Ai if and only if x is in
Ai for every i in I. These operations are represented in symbolic logic by the existential and the universal
quantifiers. We have:
∪
• ∀x (x ∈ i∈I Ai ↔ ∃i ∈ I (x ∈ Ai ))
∩
• ∀x (x ∈ i∈I Ai ↔ ∀i ∈ I (x ∈ Ai ))
Returning∪to the examples above, we can compute the∩union and intersection ∪ of each family. For the first
example, n∈N An is the set ∩ of all living people, and A
n∈N n = ∅. Also,
∪ r∈R Br = R>0 ∩, the set of all
positive real numbers, and r∈R Br = ∅. For the last example, we have n∈N Cn = N and n∈N Cn = {1},
since 1 is a divisor of every natural number.
Suppose that I contains just two elements, say I = {c, d}. Let (Ai )i∈I be a family of sets indexed by
I. Because I has two elements, this family consists of just the two sets Ac and Ad . Then the union and
intersection of this family are just the union and intersection of the two sets:
∪
Ai = Ac ∪ Ad
i∈I
∩
Ai = Ac ∩ Ad .
i∈I
This means that the union and intersection of two sets are just a special case of the union and intersection
of a family of sets.
We also have equalities for unions and intersections of families of sets. Here are a few of them:
100 Chapter 11. Sets

∪ ∪
• A∩ i∈I Bi = i∈I (A ∩ Bi )
∩ ∩
• A ∪ i∈I Bi = i∈I (A ∪ Bi )
∩ ∪
• i∈I Ai = i∈I Ai
∪ ∩
• i∈I Ai = i∈I Ai
∪ ∪ ∪ ∪
• i∈I j∈J Ai,j = j∈J i∈I Ai,j
∩ ∩ ∩ ∩
• i∈I j∈J Ai,j = j∈J i∈I Ai,j
In the last two lines, Ai,j is indexed by two sets I and J. This means that for every i ∈ I and j ∈ J we have
a set Ai,j . For the first four equalities, try to figure out what the rule means if the index set I contains two
elements.
∪
Let’s
∪ prove the first identity. Notice how the logical forms of the assertions x ∈ A ∩ i∈I Bi and x ∈
i∈I (A ∩ Bi ) dictate the structure of the proof.
Theorem. Let A be any subset of some domain U , and let (Bi )i∈I be a family of subsets of U indexed by
I. Then
∪ ∪
A∩ Bi = (A ∩ Bi )
i∈I i∈I
∪
Proof. Suppose
∪ x is in A ∩ i∈I Bi . Then x is in A and x is in Bj for some j ∈ I. So x is in A ∩ Bj , and
hence in i∈I (A ∩ Bi ).
∪
Conversely, suppose ∪ x is in i∈I (A ∩ Bi ). Then,
∪ for some j in I, x is in A ∩ Bj . Hence x is in A, and since
x is in Bj , it is in i∈I Bi . Hence x is in A ∩ i∈I Bi , as required.
11.4 Cartesian Product and Power Set
The ordered pair of two objects a and b is denoted (a, b). We say that a is the first component and b is
the second component of the pair. Two pairs are only equal if the first component are equal and the second
components are equal. In symbols, (a, b) = (c, d) if and only if a = c and b = d.
Given two sets A and B, we define the cartesian product A × B of these two sets as the set of all pairs where
the first component is an element in A and the second component is an element in B. In set-builder notation
this means
A × B = {(a, b) | a ∈ A and b ∈ B}.
Note that if A and B are subsets of a particular domain U, the set A × B need not be a subset of the same
domain. However, it will be a subset of U × U .
Some axiomatic foundations take the notion of a pair to be primitive. In axiomatic set theory, it is common
to define an ordered pair to be a particular set, namely
(a, b) = {{a}, {a, b}}.
Notice that if a = b, this set has only one element:
(a, a) = {{a}, {a, a}} = {{a}, {a}} = {{a}}.
11.4. Cartesian Product and Power Set 101

The following theorem shows that this definition is reasonable.
Theorem. Using the definition of ordered pairs above, we have (a, b) = (c, d) if and only if a = c and b = d.
Proof. If a = c and b = d then clearly (a, b) = (c, d). For the other direction, suppose that (a, b) = (c, d),
which means
{{a}, {a, b}} = {{c}, {c, d}} .
| {z } | {z }
L R
Suppose first that a = b. Then L = {{a}}. This means that {c} = {a} and {c, d} = {a}, from which we
conclude that c = a and d = a = b.
Now suppose that a ̸= b. If {c} = {a, b} then we conclude that a and b are both equal to c, contradicting
a ̸= b. Since {c} ∈ L, {c} must be equal to {a}, which means that a = c. We know that {a, b} ∈ R, and
since we know {a, b} ̸= {c}, we conclude {a, b} = {c, d}. This means that b ∈ {c, d}, since b ̸= a = c, we
conclude that b = d.
Hence in both cases we conclude that a = c and b = d, proving the theorem.
Using ordered pairs we can define the ordered triple (a, b, c) to be (a, (b, c)). Then we can prove that
(a, b, c) = (d, e, f ) if and only if a = d, b = e and c = f , which you are asked to do in the exercises. We can
also define ordered n-tuples, which are sequence of n objects, in a similar way.
Given a set A we can define the power set P(A) to be the set of all subsets of A. In set-builder notation we
can write this as
P(A) = {B | B ⊆ A}.
If A is a subset of U, P(A) may not be a subset U, but it is always a subset of P(U).
11.5 Exercises
1. Prove the following theorem: Let A, B, and C be sets of elements of some domain. Then A ∪ (B ∩ C) =
(A ∪ B) ∩ (A ∪ C). (Henceforth, if we don’t specify natural deduction or Lean, ‘‘prove’’ and ‘‘show’’
mean give an ordinary mathematical proof, using ordinary mathematical language rather than symbolic
logic.)
2. Prove the following theorem: Let A and B be sets of elements of some domain. Then A \ B = A ∪ B.
3. Two sets A and B are said to be disjoint if they have no element in common. Show that if A and B
are disjoint, C ⊆ A, and D ⊆ B, then C and D are disjoint.
4. Let A and B be sets. Show (A \ B) ∪ (B \ A) = (A ∪ B) \ (A ∩ B), by showing that both sides have
the same elements.
5. Let A, B, and C be subsets of some domain U. Give a calculational proof of the identity A \ (B ∪ C) =
(A \ B) \ C, using the identities above. Also use the fact that, in general, C \ D = C ∩ D.
6. Similarly, give a calculational proof of (A \ B) ∪ (A ∩ B) = A.
7. Give calculational proofs of the following:
• A \ B = A \ (A ∩ B)
• A \ B = (A ∪ B) \ B

• (A ∩ B) \ C = (A \ C) ∩ B
8. Prove that if (Ai,j )i∈I,j∈J is a family indexed by two sets I and J, then
∪∩ ∩∪
Ai,j ⊆ Ai,j .
i∈I j∈J j∈J i∈I
Also, find a family (Ai,j )i∈I,j∈J where the reverse inclusion does not hold.
9. Prove using calculational reasoning that
( )  
∪ ∪ ∪
Ai ∩ Bj  = (Ai ∩ Bj ).
i∈I j∈J i∈I
j∈J
∪ ∪ ∪
The notation i∈I (Ai ∩ Bj ) means i∈I j∈J (Ai ∩ Bj ).
j∈J
10. Using the definition (a, b, c) = (a, (b, c)), show that (a, b, c) = (d, e, f ) if and only if a = d, b = e and
c = f.
11. Prove that A × (B ∪ C) = (A × B) ∪ (A × C)
12. Prove that (A ∩ B) × (C ∩ D) = (A × C) ∩ (B × D). Find an expression for (A ∪ B) × (C ∪ D) consisting
of unions of cartesian products, and prove that your expression is correct.
13. Prove that that A ⊆ B if and only if P(A) ⊆ P(B).
11.5. Exercises 103


CHAPTER
TWELVE
SETS IN LEAN
In the last chapter, we noted that although in axiomatic set theory one consider sets of disparate objects, it is
more common in mathematics to consider subsets of some fixed domain, U. This is the way sets are handled
in Lean. For any data type U, Lean gives us a new data type, set U, consisting of the sets of elements of
U. Thus, for example, we can reason about sets of natural numbers, or sets of integers, or sets of pairs of
natural numbers.
12.1 Basics
Given A : set U and x : U, we can write x ∈ A to state that x is a member of the set A. The character ∈
can be typed using \in.
import data.set
open set
variable {U : Type}
variables A B C : set U
variable x : U
#check x ∈ A
#check A ∪ B
#check B \ C
#check C ∩ A
#check -C
#check ∅ ⊆ A
#check B ⊆ univ
You can type the symbols ⊆, ∅, ∪, ∩, \ as \subeq \empty, \un, \i, and \\, respectively. We have made
the type variable U implicit, because it can typically be inferred from context. The universal set is denoted
univ, and set complementation is denoted with a negation symbol. Basic set-theoretic notions like these are
defined in Lean’s core library, but additional theorems and notation are available in an auxiliary library that
we have loaded with the command import data.set, which has to appear at the beginning of a file. The
command open set lets us refer to a theorem named set.mem_union as mem_union.
The following pattern can be used to show that A is a subset of B:
example : A ⊆ B :=
assume x,
assume h : x ∈ A,
show x ∈ B, from sorry
And the following pattern be used to show that A and B are equal:
105
example : A = B :=
eq_of_subset_of_subset
(assume x,
assume h : x ∈ A,
show x ∈ B, from sorry)
(assume x,
assume h : x ∈ B,
show x ∈ A, from sorry)
Alternatively, we can use the following pattern:

example : A = B :=
ext (assume x, iff.intro
(assume h : x ∈ A,
show x ∈ B, from sorry)
(assume h : x ∈ B,
show x ∈ A, from sorry))
Here, ext is short for “extensionality.” In symbolic terms, it is the following fact:
∀x (x ∈ A ↔ x ∈ B) → A = B.
This reduces proving A = B to proving ∀x (x ∈ A ↔ x ∈ B), which we can do using ∀ and ↔ introduction.
Moreover, Lean supports the following nifty feature: the defining rules for union, intersection and other
operations on sets are considered to hold “definitionally.” This means that the expressions x ∈ A ∩ B and
x ∈ A ∧ x ∈ B mean the same thing to Lean. This is the same for the other constructions on sets; for
example x ∈ A \ B and x ∈ A ∧ ¬ (x ∈ B) mean the same thing to Lean. You can also write x ∈ / B for
¬ (x ∈ B), where ∈ / is written using \notin. For the other set constructions, the defining equivalences in
the last chapter hold definitionally. The following example illustrates these features.
example : ∀ x, x ∈ A → x ∈ B → x ∈ A ∩ B :=
assume x,
assume : x ∈ A,
assume : x ∈ B,
show x ∈ A ∩ B, from and.intro ‹x ∈ A› ‹x ∈ B›
example : A ⊆ A ∪ B :=
assume x,
assume : x ∈ A,
show x ∈ A ∪ B, from or.inl this
example : ∅ ⊆ A :=
assume x,
assume : x ∈ ∅,
show x ∈ A, from false.elim ‹x ∈ (∅ : set U)›
Remember from Section 4.5 that we can use assume without a label, and refer back to hypotheses using
French quotes, entered with \f< and \f>. We have used this feature in the previous example. Without that
feature, we could have written the examples above as follows:
assume x,
assume h1 : x ∈ A,
assume h2 : x ∈ B,
show x ∈ A ∩ B, from and.intro h1 h2
106 Chapter 12. Sets in Lean

assume x,
assume h : x ∈ A,
show x ∈ A ∪ B, from or.inl h
assume x,
assume : x ∈ ∅,
show x ∈ A, from false.elim ‹x ∈ (∅ : set U)›
Below, and in the chapters that follow, we will begin to use assume and have command without labels, but
you should feel free to adopt whatever style you prefer.
Notice also that in the last example, we had to annotate the empty set by writing (∅ : set U) to tell Lean
which empty set we mean. Lean can often infer information like this from the context (for example, from
the fact that we are trying to show x ∈ A, where A has type set U), but in this case, it needs a bit more
help.
Alternatively, we can use versions theorems in the Lean library that are designed specifically for use with
sets:
assume x,
assume : x ∈ A,
assume : x ∈ B,
show x ∈ A ∩ B, from mem_inter ‹x ∈ A› ‹x ∈ B›
assume x,
assume h : x ∈ A,
show x ∈ A ∪ B, from mem_union_left B h
assume x,
assume : x ∈ ∅,
show x ∈ A, from absurd this (not_mem_empty x)
Remember that absurd can be used to prove any fact from two contradictory hypotheses h1 : P and h2 :
¬ P. Here the not_mem_empty x is the fact x ∈
/ ∅. You can see the statements of the theorems using the
#check command in Lean:
#check @mem_inter
#check @mem_of_mem_inter_left
#check @mem_of_mem_inter_right
#check @mem_union_left
#check @mem_union_right
#check @mem_or_mem_of_mem_union
#check @not_mem_empty
Here, the @ symbol in Lean prevents it from trying to fill in implicit arguments automatically, and instead
display the full statement of the theorem.
The fact that Lean can identify sets with their logical definitions makes it easy to prove inclusions between
sets:
example : A \ B ⊆ A :=
assume x,
assume : x ∈ A \ B,
show x ∈ A, from and.left this
12.1. Basics 107

example : A \ B ⊆ -B :=
assume x,
have x ∈
/ B, from and.right this,
show x ∈ -B, from this
Once again, we can use versions of the theorems designed specifically for sets:
example : A \ B ⊆ A :=
assume x,
show x ∈ A, from mem_of_mem_diff this
example : A \ B ⊆ -B :=
assume x,
have x ∈
/ B, from not_mem_of_mem_diff this,
show x ∈ -B, from this
Once again, the fact that Lean has to unfold definitions means that it can be confused at times. For example,
in the proof below, if you replace the last line by sorry, Lean has trouble figuring out that you want it to
unfold the subset symbol:
variable {U : Type}
variables A B : set U
example : A ∩ B ⊆ B ∩ A :=
assume x,
assume h : x ∈ A ∩ B,
have h1 : x ∈ A, from and.left h,
have h2 : x ∈ B, from and.right h,
and.intro h2 h1
One workaround is to use the show command; in general, providing Lean with such additional information
is often helpful. Another is to workaround is to give the theorem a name, which prompts Lean to use a
slightly different method of processing the proof, fixing the problem as a lucky side effect.
example : A ∩ B ⊆ B ∩ A :=
assume x,
show x ∈ B ∩ A, from sorry
theorem my_example : A ∩ B ⊆ B ∩ A :=
assume x,
sorry
12.2 Some Identities
Here is the proof of the first identity that we proved informally in the previous chapter:

example : A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) :=
(assume x,
assume h : x ∈ A ∩ (B ∪ C),
have x ∈ A, from and.left h,
have x ∈ B ∪ C, from and.right h,
or.elim (‹x ∈ B ∪ C›)
(assume : x ∈ B,
have x ∈ A ∩ B, from and.intro ‹x ∈ A› ‹x ∈ B›,
show x ∈ (A ∩ B) ∪ (A ∩ C), from or.inl this)
(assume : x ∈ C,
have x ∈ A ∩ C, from and.intro ‹x ∈ A› ‹x ∈ C›,
show x ∈ (A ∩ B) ∪ (A ∩ C), from or.inr this))
(assume x,
assume : x ∈ (A ∩ B) ∪ (A ∩ C),
or.elim this
(assume h : x ∈ A ∩ B,
have x ∈ B, from and.right h,
have x ∈ B ∪ C, from or.inl this,
show x ∈ A ∩ (B ∪ C), from and.intro ‹x ∈ A› this)
(assume h : x ∈ A ∩ C,
have x ∈ C, from and.right h,
have x ∈ B ∪ C, from or.inr this,
show x ∈ A ∩ (B ∪ C), from and.intro ‹x ∈ A› this))
Notice that it is considerably longer than the informal proof in the last chapter, because we have spelled out
every last detail. Unfortunately, this does not necessarily make it more readable. Keep in mind that you can
always write long proofs incrementally, using sorry. You can also break up long proofs into smaller pieces:
theorem inter_union_subset : A ∩ (B ∪ C) ⊆ (A ∩ B) ∪ (A ∩ C) :=
assume x,
assume h : x ∈ A ∩ (B ∪ C),
have x ∈ B ∪ C, from and.right h,
or.elim (‹x ∈ B ∪ C›)
(assume : x ∈ B,
have x ∈ A ∩ B, from and.intro ‹x ∈ A› ‹x ∈ B›,
show x ∈ (A ∩ B) ∪ (A ∩ C), from or.inl this)
(assume : x ∈ C,
have x ∈ A ∩ C, from and.intro ‹x ∈ A› ‹x ∈ C›,
show x ∈ (A ∩ B) ∪ (A ∩ C), from or.inr this)
theorem inter_union_inter_subset :
(A ∩ B) ∪ (A ∩ C) ⊆ A ∩ (B ∪ C) :=
assume x,
assume : x ∈ (A ∩ B) ∪ (A ∩ C),
or.elim this
(assume h : x ∈ A ∩ B,
have x ∈ B, from and.right h,
have x ∈ B ∪ C, from or.inl this,
(assume h : x ∈ A ∩ C,
have x ∈ C, from and.right h,
have x ∈ B ∪ C, from or.inr this,
12.2. Some Identities 109

example : A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) :=
(inter_union_subset A B C)
(inter_union_inter_subset A B C)
Notice that the two theorems depend on the variables A, B, and C, which have to be supplied as arguments
when they are applied. They also depend on the underlying type, U, but because the variable U was marked
implicit, Lean figures it out from the context.
In the last chapter, we showed (A ∩ B) ∪ B = B. Here is the corresponding proof in Lean:
example : (A ∩ -B) ∪ B = A ∪ B :=
calc
(A ∩ -B) ∪ B = (A ∪ B) ∩ (-B ∪ B) : by rw union_distrib_right
... = (A ∪ B) ∩ univ : by rw compl_union_self
... = A ∪ B : by rw inter_univ
Translated to propositions, the theorem above states that for every pair of elements A and B in a Boolean
algebra, (A ∧ ¬B) ∨ B = B. Lean allows us to do calculations on propositions as though they are elements
of a Boolean algebra, with equality replaced by ↔.
example : (A ∧ ¬ B) ∨ B ↔ A ∨ B :=
calc
(A ∧ ¬ B) ∨ B ↔ (A ∨ B) ∧ (¬ B ∨ B) : by rw and_or_distrib_right
... ↔ (A ∨ B) ∧ true : by rw not_or_self
... ↔ (A ∨ B) : by rw and_true
12.3 Indexed Families

∩
Remember that∪ if (Ai )i∈I is a family of sets indexed by I, then i∈I Ai denotes the intersection of all the
sets Ai , and i∈I Ai denotes their union. In Lean, we can specify that A is a family of sets by writing A :
I → set U where I is a Type. In other words, a family of sets is really a function which for each element i
of type I returns a set A i. We can then define the union and intersection as follows:
variables {I U : Type}
def Union (A : I → set U) : set U := { x | ∃ i : I, x ∈ A i }

def Inter (A : I → set U) : set U := { x | ∀ i : I, x ∈ A i }
section
variables (x : U) (A : I → set U)
example (h : x ∈ Union A) : ∃ i, x ∈ A i := h
example (h : x ∈ Inter A) : ∀ i, x ∈ A i := h
end
The examples show that Lean can unfold the definitions so that x ∈ Inter A can be treated as ∀ i, x ∈
A i and x ∈ Union A can be treated as ∃ i, x ∈ A i. To refresh your memory as to how to work with the
universal and existential quantifiers in Lean, see Chapters 9. We can then define notation for the indexed
union and intersection:

∪
notation `∩` binders `, ` r:(scoped f, Union f) := r
notation ` ` binders `, ` r:(scoped f, Inter f) := r
variables (A : I → set U) (x : U)
∪
example (h : x ∈ ∩ i, A i) : ∃ i, x ∈ A i := h
example (h : x ∈ i, A i) : ∀ i, x ∈ A i := h
∩ ∪ ∪ ∩
You can type and with \I and \Un, respectively. As with quantifiers, the notation i, A i and
i, A i bind∩ the variable i in the expression, and the scope extends as widely as possible. For example, if
you write i, A i ∪ B, Lean assumes that the ith element of the sequence is A i ∪ B. If you want to
restrict the scope more narrowly, use parentheses.
The good news is that Lean’s library does define indexed union and intersection, with this notation, and the
definitions are made available with import data.set. The bad news is that it uses a different definition, so
that x ∈ Inter A and x ∈ Union A are not definitionally equal to ∀ i, x ∈ A i and ∃ i, x ∈ A i, as
above. The good news is that Lean at least knows that they are equivalent:
import data.set
open set
variables {A B : I → set U}
∪
theorem exists_of_mem_Union {x : U} (h : x ∈ i, A i) :
∃ i, x ∈ A i :=
by simp * at *
theorem∪ mem_Union_of_exists {x : U} (h : ∃ i, x ∈ A i) :
x ∈ i, A i :=
by simp * at *
∩
theorem forall_of_mem_Inter {x : U} (h : x ∈ i, A i) :
∀ i, x ∈ A i :=
by simp * at *
theorem∩ mem_Inter_of_forall {x : U} (h : ∀ i, x ∈ A i) :
x ∈ i, A i :=
by simp * at *
The command simp * at * calls upon Lean’s automation to carry out the proofs. Here is an example of
how these can be used:
∩ ∩ ∩
example : ( i, A i ∩ B i) = ( i, A i) ∩ ( i, B i) :=
ext $
assume x : U,
iff.intro ∩
(assume h : x ∈ i, A i ∩ B i,
have h1 : ∀ i, x ∈ A i ∩ B i, from forall_of_mem_Inter h,
have h2 : ∀ i, x ∈ A i, from assume i, and.left (h1 i),
have h3 : ∀ i, ∩x ∈ B i, from assume i, and.right (h1 i),
have h4 : x ∈ ∩ i, A i, from mem_Inter_of_forall h2,
have h5 : x ∈ i, B i, from mem_Inter_of_forall h3,
and.intro h4 h5)
∩ ∩
(assume h : x ∈ ( i, A i) ∩ ( i, B i),
have h1 : ∀ i, x ∈ A i,
from forall_of_mem_Inter (and.left h),
have h2 : ∀ i, x ∈ B i,
12.3. Indexed Families 111

from forall_of_mem_Inter (and.right h),

have h3 : ∀ i, x ∈ A i ∩ B i,
from assume
∩ i, and.intro (h1 i) (h2 i),
show x ∈ i, A i ∩ B i, from mem_Inter_of_forall h3)
Even better, we can prove introduction and elimination rules for intersection and union:
import data.set
open set
variables {A : I → set U}
∩
theorem Inter.intro {x : U} (h : ∀ i, x ∈ A i) : x ∈ i, A i :=
by simp; assumption
@[elab_simple] ∩
theorem Inter.elim {x : U} (h : x ∈ i, A i) (i : I) : x ∈ A i :=
by simp at h; apply h
theorem∪ Union.intro {x : U} (i : I) (h : x ∈ A i) :
x ∈ i, A i :=
by {simp, existsi i, exact h}
theorem Union.elim
∪ {b : Prop} {x : U}
(h1 : x ∈ i, A i) (h2 : ∀ (i : I), x ∈ A i → b) : b :=
by {simp at h1 , cases h1 with i h, exact h2 i h}
Don’t worry about what the proofs mean. What is important is how they can be used, which is what the
following patterns illustrate:
∩
example (x : U) : x ∈ i, A i :=
Inter.intro $
assume i,
show x ∈ A i, from sorry
∩
example (x : U) (i : I) (h : x ∈ i, A i) : x ∈ A i :=
Inter.elim h i
∪
example (x : U) (i : I) (h : x ∈ A i) : x ∈ i, A i :=
Union.intro i h
∪
example (C : Prop) (x : U) (h : x ∈ i, A i) : C :=
Union.elim h $
assume i,
assume h : x ∈ A i,
show C, from sorry
Remember that the dollar sign saves us the trouble of having to put parentheses around the rest of the proof.
Notice that with Inter.intro and Inter.elim, proofs using indexed intersections looks just like proofs using
the universal quantifier. Similarly, Union.intro and Union.elim mirror the introduction and elimination
rules for the existential quantifier. The following example provides one direction of an equivalence proved
above:
variables (A : I → set U) (B : I → set U) (C : set U)

∩ ∩ ∩
example : ( i, A i ∩ B i) ⊆ ( i, A i) ∩ ( i, B i) :=
assume x : U, ∩
assume h : x ∈∩ i, A i ∩ B i,
have h1 : x ∈ i, A i, from
Inter.intro $
assume i : I,
have h2 : x ∈ A i ∩ B i, from Inter.elim h i,
show x ∈ A∩i, from and.left h2,
have h2 : x ∈ i, B i, from
Inter.intro $
assume i : I,
have h2 : x ∈ A i ∩ B i, from Inter.elim h i,
show x ∩∈ B i, from and.right
∩ h2,
show x ∈ ( i, A i) ∩ ( i, B i), from and.intro h1 h2
You are asked to prove the other direction in the exercises below. Here is an example that shows how to use
the introduction and elimination rules for indexed union:
∪ ∪
example : ( i, C ∩ A i) ⊆ C ∩ ( i, A i) :=
assume x, ∪
assume h : x ∈ i, C ∩ A i,
Union.elim h $
assume i,
assume h1 : x ∈ C ∩ A i,
have h2 : x ∈ C, from and.left h1,
have h3 : x ∈ A∪ i, from and.right h1,
have h4 : x ∈ ∪ i, A i, from Union.intro i h3,
show x ∈ C ∩ i, A i, from and.intro h2 h4
Once again, we ask you to prove the other direction in the exercises below.
Sometimes we want to work with families (Ai,j )i∈I,j∈J indexed by two variables. This is also easy to manage
in Lean: if we declare A : I → J → set U, then given i : I and j : J, we have that A i j : set U.
(You should interpret the expression I → J → set U as I → (J → set U), so that A i has type J →
set U, and then A i j has type set U.) Here is an example of a proof involving a such a doubly-indexed
family:
section
variables {I J U : Type}
variables (A : I → J → set U)
∪ ∩ ∩ ∪
example : ( i, j, A i j) ⊆ ( j, i, A i j) :=
assume x, ∪ ∩
assume h : x ∈ i, j, A i j,
Union.elim h $
assume i, ∩
∈
assume h1 ∩: x ∪ j, A i j,
show x ∈ j, i, A i j, from
Inter.intro $
assume j,
have h2 : x ∈ A i j, from Inter.elim h1 j,
Union.intro i h2
end
12.3. Indexed Families 113

12.4 Power Sets
We can also define the power set in Lean:

variable {U : Type}
def powerset (A : set U) : set (set U) := {B : set U | B ⊆ A}
example (A B : set U) (h : B ∈ powerset A) : B ⊆ A :=

h
As the example shows, B ∈ powerset A is then definitionally the same as B ⊆ A.

In fact, powerset is defined in Lean in exactly this way, and is available to you when you import data.set
and open set. Here is an example of how it is used:
#check powerset A
example : A ∈ powerset (A ∪ B) :=
assume x,
assume : x ∈ A,
show x ∈ A ∪ B, from or.inl ‹x ∈ A›
In essence, the example proves A ⊆ A ∪ B. In the exercises below, we ask you to prove, formally, that for
every A B : set U, we have powerset A ⊆ powerset B
12.5 Exercises
1. Fill in the sorry’s.

section
variable U : Type
example : ∀ x, x ∈ A ∩ C → x ∈ A ∪ B :=
sorry
example : ∀ x, x ∈ -(A ∪ B) → x ∈ -A :=
sorry
end

import data.set
open set
section
variable {U : Type}
example (A B C : set U) : ∀ x, x ∈ A ∩ C → x ∈ A ∪ B :=
sorry
example (A B : set U) : ∀ x, x ∈ -(A ∪ B) → x ∈ -A :=

sorry
/- defining "disjoint" -/

def disj (A B : set U) : Prop := ∀ {|x|}, x ∈ A → x ∈ B → false
example (A B : set U) (h : ∀ x, ¬ (x ∈ A ∧ x ∈ B)) :

disj A B :=
assume x,
assume h1 : x ∈ A,
assume h2 : x ∈ B,
have h3 : x ∈ A ∧ x ∈ B, from and.intro h1 h2,
show false, from h x h3
-- notice that we do not have to mention x when applying

-- h : disj A B
example (A B : set U) (h1 : disj A B) (x : U)
(h2 : x ∈ A) (h3 : x ∈ B) :
false :=
h1 h2 h3
-- the same is true of ⊆

example (A B : set U) (x : U) (h : A ⊆ B) (h1 : x ∈ A) :
x ∈ B :=
h h1
example (A B C D : set U) (h1 : disj A B) (h2 : C ⊆ A)

(h3 : D ⊆ B) :
disj C D :=
sorry
end
3. Prove the following facts about indexed unions and intersections, using the theorems Inter.intro,
Inter.elim, Union.intro, and Union.elim listed above.
∩ ∩ ∩
example : ( i, A i) ∩ ( i, B i) ⊆ ( i, A i ∩ B i) :=
sorry
∪ ∪
example : C ∩ ( i, A i) ⊆ i, C ∩ A i :=
sorry
4. Prove the following fact about power sets. You can use the theorems subset.trans and subset.refl
variable {U : Type}
-- For this exercise these two facts are useful

example (h1 : A ⊆ B) (h2 : B ⊆ C) : A ⊆ C :=
subset.trans h1 h2
example : A ⊆ A :=
subset.refl A
example (h : A ⊆ B) : powerset A ⊆ powerset B :=

sorry
example (h : powerset A ⊆ powerset B) : A ⊆ B :=

sorry
12.5. Exercises 115


CHAPTER
THIRTEEN
RELATIONS
In Chapter 7 we discussed the notion of a relation symbol in first-order logic, and in Chapter 10 we saw
how to interpret such a symbol in a model. In mathematics, we are generally interested in different sorts of
relationships between mathematical objects, and so the notion of a relation is ubiquitous. In this chapter,
we will consider some common kinds of relations.
In some axiomatic foundations, the notion of a relation is taken to be primitive, but in axiomatic set theory,
a relation is taken to be a set of tuples of the corresponding arity. For example, we can take a binary relation
on A to be a subset of A × A, where R(a, b) means that (a, b) ∈ R. The foundational definition is generally
irrelevant to everyday mathematical practice; what is important is simply that we can write expressions like
R(a, b), and that they are true or false, depending on the values of a and b. In mathematics, we often use
infix notation, writing aRb instead of R(a, b).
13.1 Order Relations
We will start with a class of important binary relations in mathematics, namely, partial orders.
Definition. A binary relation ≤ on a domain A is a partial order if it has the following three properties:
• reflexivity: a ≤ a, for every a in A
• transitivity: if a ≤ b and b ≤ c, then a ≤ c, for every a, b, and c in A
• antisymmetry: if a ≤ b and b ≤ a then a = b, for every a and b in A.
Notice the compact way of introducing the symbol ≤ in the statement of the definition, and the fact that
≤ is written as an infix symbol. Notice also that even though the relation is written with the symbol ≤, it
is the only symbol occurring in the definition; mathematical practice favors natural language to describe its
properties.
You now know enough, however, to recognize the universal quantifiers that are present in the three clauses.
In symbolic logic, we would write them as follows:
• ∀a (a ≤ a)
• ∀a, b, c (a ≤ b ∧ b ≤ c → a ≤ c)
• ∀a, b (a ≤ b ∧ b ≤ a → a = b)
Here the variables a, b, and c implicitly range over the domain A.
The use of the symbol ≤ is meant to be suggestive, and, indeed, the following are all examples of partial
orders:
117
• ≤ on the natural numbers

• ≤ on the integers
• ≤ on the rational numbers
• ≤ on the real numbers
But keep in mind that ≤ is only a symbol; it can have unexpected interpretations as well. For example, all
of the following are also partial orders:
• ≥ on the natural numbers
• ≥ on the integers
• ≥ on the rational numbers
• ≥ on the real numbers
These are not fully representative of the class of partial orders, in that they all have an additional property:
Definition. A partial order ≤ on a domain A is a total order (also called a linear order) if it also has the
following property:
• for every a and b in A, either a ≤ b or b ≤ a.
You can check these these are two examples of partial orders that are not total orders:
• the divides relation, x | y, on the integers
• the subset relation, x ⊆ y, on sets of elements of some domain A
On the integers, we also have the strict order relation, <, which is not a partial order, since it is not reflexive.
It is, rather, an instance of a strict partial order:
Definition. A binary relation < on a domain A is a strict partial order if it satisfies the following:
• irreflexivity: a ≮ a for every a in A.
• transitivity: a < b and b < c implies a < c, for every a, b, and c in A.
A strict partial order is a strict total order (or strict linear order) if, in addition, we have the following
property:
• trichotomy: a < b, a = b, or a > b for every a and b in A.
Here, b ≮ a means, of course, that it is not the case that a < b, and a > b is alternative notation for b < a.
To distinguish an ordinary partial order from a strict one, an ordinary partial order is sometimes called a
weak partial order.
Proposition. A strict partial order < on A is asymmetric: for every a and b, a < b implies b ≮ a.
Proof. Suppose a < b and b < a. Then, by transitivity, a < a, contradicting irreflexivity.
On the integers, there are precise relationships between < and ≤: x ≤ y if and only if x < y or x = y, and
x < y if and only if x ≤ y and x ̸= y. This illustrates a more general phenomenon.
118 Chapter 13. Relations

Theorem. Suppose ≤ is a partial order on a domain A. Define a < b to mean that a ≤ b and a ̸= b. Then
< is a strict partial order. Moreover, if ≤ is total, so is <.
Theorem. Suppose < is a strict partial order on a domain A. Define a ≤ b to mean a < b or a = b. Then
≤ is a partial order. Moreover, if < is total, so is ≤.
We will prove the first here, and leave the second as an exercise. This proof is a nice illustration of how
universal quantification, equality, and propositional reasoning are combined in a mathematical argument.
Proof. Suppose ≤ is a partial order on A, and < be defined as in the statement of the theorem. Irreflexivity
is immediate, since a < a implies a ̸= a, which is a contradiction.
To show transitivity, suppose a < b and b < c. Then we have a ≤ b, b ≤ c, a ̸= b, and b ̸= c. By the
transitivity of ≤, we have a ≤ c. To show a < c, we only have to show a ̸= c. So suppose a = c. then, from
the hypotheses, we have c < b and b < c, violating asymmetry. So a ̸= c, as required.
To establish the last claim in the theorem, suppose ≤ is total, and let a and b be any elements of A. We
need to show that a < b, a = b, or a > b. If a = b, we are done, so we can assume a ̸= b. Since ≤ is total,
we have a ≤ b or a ≤ b. Since a ̸= b, in the first case we have a < b, and in the second case, we have a > b.
13.2 More on Orderings
Let ≤ be a partial order on a domain, A, and let < be the associated strict order, as defined in the last
section. It is possible to show that if we go in the other direction, and define ≤′ to be the partial order
associated to <, then ≤ and ≤′ are the same, which is to say, for every a and b in A, a ≤ b if and only if
a ≤′ b. So we can think of every partial order as really being a pair, consisting of a weak partial order and
an associated strict one. In other words, we can assume that x < y holds if and only if x ≤ y and x ̸= y,
and we can assume x ≤ y holds if and only if x < y or x = y.
We will henceforth adopt this convention. Given a partial order ≤ and the associated strict order <, we
leave it to you to show that if x ≤ y and y < z, then x < z, and, similarly, if x < y and y ≤ z, then x < z.
Consider the natural numbers with the less-than-or-equal relation. It has a least element, 0. We can express
the fact that 0 is the least element in at least two ways:
• 0 is less than or equal to every natural number.
• There is no natural number that is less than 0.
In symbolic logic, we could formalize these statements as follows:
• ∀x (0 ≤ x)
• ∀x (x ≮ 0)
Using the existential quantifier, we could render the second statement more faithfully as follows:
• ¬∃x (x < 0)
13.2. More on Orderings 119

Notice that this more faithful statement is equivalent to the original, using deMorgan’s laws for quantifiers.
Are the two statements above equivalent? Say an element y is minimum for a partial order if it is less than
or equal to any other element; this is, if it takes the place of 0 in the first statement. Say that an element
y is minimal for a partial order if no element is less than it; that is, if it takes the place of 0 in the second
statement. Two facts are immediate.
Theorem. Any minimum element is minimal.

Proof. Suppose x is minimum for ≤. We need to show that x is minimal, that is, for every y, it is not the
case that y < x. Suppose y < x. Since x is minimum, we have x ≤ y. From y < x and x ≤ y, we have y < y,
contradicting the irreflexivity of <.
Theorem. If a partial order ≤ has a minimum element, it is unique.
Proof. Suppose x1 and x2 are both minimum. Then x1 ≤ x2 and x2 ≤ x1 . By antisymmetry, x1 = x2 .
Notice that we have interpreted the second theorem as the statement that if x1 and x2 are both minimum,
then x1 = x2 . Indeed, this is exactly what we mean when we say that something is “unique.” When a partial
order has a minimum element x, uniqueness is what justifies calling x the minimum element. Such an x is
also called the least element or the smallest element, and the terms are generally interchangeable.
The converse to the second theorem – that is, the statement that every minimal element is minimum – is
false. As an example, consider the nonempty subsets of the set {1, 2} with the subset relation. In other
words, consider the collection of sets {1}, {2}, and {1, 2}, where {1} ⊆ {1, 2}, {2} ⊆ {1, 2}, and, of course,
every element is a subset of itself. Then you can check that {1} and {2} are each minimal, but neither is
minimum. (One can also exhibit such a partial order by drawing a diagram, with dots labeled a, b, c, etc.,
and upwards edges between elements to indicate that one is less than or equal to the other.)
Notice that the statement “a minimal element of a partial order is not necessarily minimum” makes an
“existential” assertion: it says that there is a partial order ≤, and an element x of the domain, such that x
is minimal but not minimum. For a fixed partial order ≤, we can express the assertion that such an x exists
as follows:
∃x (∀y (y ≮ x) ∧ ∀y (x ≤ y)).
The assertion that there exists a domain A, and a partial order ≤ on that domain A, is more dramatic: it
is a “higher order” existential assertion. But symbolic logic provides us with the means to make assertions
like these as well, as we will see later on.
We can consider other properties of orders. An order is said to be dense if between any two distinct elements,
there is another element. More precisely, an order is dense if, whenever x < y, there is an element z satisfying
x < z and z < y. For example, the rational numbers are dense with the usual ≤ ordering, but not the integers.
Saying that an order is dense is another example of an implicit use of existential quantification.
13.3 Equivalence Relations and Equality
In ordinary mathematical language, an equivalence relation is defined as follows.
Definition. A binary relation ≡ on some domain A is said to be an equivalence relation if it is reflexive,

symmetric, and transitive. In other words, ≡ is an equivalent relation if it satisfies these three properties:
• reflexivity: a ≡ a, for every a in A.

• symmetry: if a ≡ b, then b ≡ a, for every a and b in A.

• transitivity: if a ≡ b and b ≡ c, then a ≡ c, for every a, b, and c in A.
We leave it to you to think about how you could write these statements in first-order logic. (Note the
similarity to the rules for a partial order.) We will also leave you with an exercise: by a careful choice of
how to instantiate the quantifiers, you can actually prove the three properties above from the following two:
• ∀a (a ≡ a)
• ∀a, b, c (a ≡ b ∧ c ≡ b → a ≡ c)
Try to verify this using natural deduction or Lean.
These three properties alone are not strong enough to characterize equality. You should check that the
following informal examples are all instances of equivalence relations:
• the relation on days on the calendar, given by “x and y fall on the same day of the week”
• the relation on people currently alive on the planet, given by “x and y have the same age”
• the relation on people currently alive on the planet, given by “x and y have the same birthday”
• the relation on cities in the United States, given by “x and y are in the same state”
Here are two common mathematical examples:
• the relation on lines in a plane, given by “x and y are parallel”
• for any fixed natural number m ≥ 0, the relation on natural numbers, given by “x is congruent to y
modulo m” (see Chapter 19).
Here, we say that x is congruent to y modulo m if they leave the same remainder when divided by m. Soon,
you will be able to prove rigorously that this is equivalent to saying that x − y is divisible by m.
Consider the equivalence relation on citizens of the United States, given by “x and y have the same age.”
There are some properties that respect that equivalence. For example, suppose I tell you that John and
Susan have the same age, and I also tell you that John is old enough to vote. Then you can rightly infer
that Susan is old enough to vote. On the other hand, if I tell you nothing more than the facts that John
and Susan have the same age and John lives in South Dakota, you cannot infer that Susan lives in South
Dakota. This little example illustrates what is special about the equality relation: if two things are equal,
then they have exactly the same properties.
Let A be a set and let ≡ be an equivalence relation on A. There is an important mathematical construction
known as forming the quotient of A under the equivalence relation. For every element a in A, let [a] be
the set of elements {c | c ≡ a}, that is, the set of elements of A that are equivalent to a. We call [a] the
equivalence class of A. The set A/≡, the quotient of A by ≡, is defined to be the set {[a] : a ∈ A}, that is,
the set of all the equivalence classes of elements in A. The exercises below as you to show that if [a] and [b]
are elements of such a quotient, then [a] = [b] if and only if a ≡ b.
The motivation is as follows. Equivalence tries to capture a weak notion of equality: if two elements of A
are equivalent, they are not necessarily the same, but they are similar in some way. Equivalence classes
collect similar objects together, essentially glomming them into new objects. Thus A/≡ is a version of the
set A where similar elements have been compressed into a single element. For example, given the equivalence
relation ≡ of congruence modulo 5 on the integers, N/≡ is the set {[0], [1], [2], [3], [4]}, where, for example,
[0] is the set of all multiples of 5.
13.3. Equivalence Relations and Equality 121

13.4 Exercises
1. Suppose < is a strict partial order on a domain A, and define a ≤ b to mean that a < b or a = b.
• Show that ≤ is a partial order.
• Show that if < is moreover a strict total order, then ≤ is a total order.
(Above we proved the analogous theorem going in the other direction.)
2. Suppose < is a strict partial order on a domain A. (In other words, it is transitive and asymmetric.)
Suppose that ≤ is defined so that a ≤ b if and only if a < b or a = b. We saw in class that ≤ is a
partial order on a domain A, i.e.~it is reflexive, transitive, and antisymmetric.
Prove that for every a and b in A, we have a < b iff a ≤ b and a ̸= b, using the facts above.
3. An ordered graph is a collection of vertices (points), along with a collection of arrows between vertices.
For each pair of vertices, there is at most one arrow between them: in other words, every pair of
vertices is either unconnected, or one vertex is “directed” toward the other. Note that it is possible to
have an arrow from a vertex to itself.
Define a relation ≤ on the set of vertices, such that for two vertices a and b, a ≤ b means that there is
an arrow from a pointing to b.
On an arbitrary graph, is ≤ a partial order, a strict partial order, a total order, a strict total order, or
none of the above? If possible, give examples of graphs where ≤ fails to have these properties.
4. Let ≡ be an equivalence relation on a set A. For every element a in A, let [a] be the equivalence class
of a: that is, the set of elements {c | c ≡ a}. Show that for every a and b, [a] = [b] if and only if a ≡ b.
(Hints and notes:
• Remember that since you are proving an ‘‘if and only if’’ statement, there are two directions to
prove.
• Since that [a] and [b] are sets, [a] = [b] means that for every element c, c is in [a] if and only if c
is in [b].
• By definition, an element c is in [a] if and only if c ≡ a. In particular, a is in [a].)
5. Let the relation ∼ on the natural numbers B be defined as follows: if n is even, then n ∼ n + 1, and
if n is odd, then n ∼ n − 1. Furthermore, for every n, n ∼ n. Show that ∼ is an equivalence relation.
What is the equivalence class of the number 5? Describe the set of equivalence classes {[n] | n ∈ N}.
6. Show that the relation on lines in the plane, given by “l1 and l2 are parallel,” is an equivalence
relation. What is the equivalence class of the x-axis? Describe the set of equivalence classes {[l] |
l is a line in the plane}.
7. A binary relation ≤ on a domain A is said to be a preorder it is is reflexive and transitive. This is weaker
than saying it is a partial order; we have removed the requirement that the relation is asymmetric. An
example is the ordering on people currently alive on the planet defined by setting x ≤ y if and only if
x ‘s birth date is earlier than y ‘s. Asymmetry fails, because different people can be born on the same
day. But, prove that the following theorem holds:
Theorem. Let ≤ be a preorder on a domain A. Define the relation ≡, where x ≡ y holds if and only
if x ≤ y and y ≤ x. Then ≡ is an equivalence relation on A.

CHAPTER
FOURTEEN
RELATIONS IN LEAN
In the last chapter, we noted that set theorists think of a binary relation R on a set A as a set of ordered
pairs, so that R(a, b) really means (a, b) ∈ R. An alternative is to think of R as a function which, when
applied to a and B, returns the proposition that R(a, b) holds. This is the viewpoint adopted by Lean: a
binary relation on a type A is a function A → A → Prop. Remember that the arrows associate to the right,
so A → A → Prop really means A → (A → Prop). So, given a : A, R a is a predicate (the property of
being related to A), and given a b : A, R a b is a proposition.
14.1 Order Relations
With first-order logic, we can say what it means for a relation to be reflexive, symmetric, transitive, anti-
symmetric, and so on:
namespace hide
variable {A : Type}
def reflexive (R : A → A → Prop) : Prop :=

∀ x, R x x
def symmetric (R : A → A → Prop) : Prop :=

∀ x y, R x y → R y x
def transitive (R : A → A → Prop) : Prop :=

∀ x y z, R x y → R y z → R x z
def antisymmetric (R : A → A → Prop) : Prop :=

∀ x y, R x y → R y x → x = y
end hide
We can then use the notions freely. Notice that Lean will unfold the definitions when necessary, for example,
treating reflexive R as ∀ x, R x x.
variable R : A → A → Prop
example (h : reflexive R) (x : A) : R x x := h x
example (h : symmetric R) (x y : A) (h1 : R x y) : R y x :=

h x y h1
example (h : transitive R) (x y z : A) (h1 : R x y) (h2 : R y z) :

R x z :=
123
h x y z h1 h2
example (h : antisymmetric R) (x y : A) (h1 : R x y)

(h2 : R y x) :
x = y :=
h x y h1 h2
In the command variable {A : Type}, we put curly braces around A to indicate that it is an implicit
argument, which is to say, you do not have to write it explicitly; Lean can infer it from the argument R.
That is why we can write reflexive R rather than reflexive A R: Lean knows that R is a binary relation
on A, so it can infer that we mean reflexivity for binary relations on A.
Given h : transitive R, h1 : R x y, and h2 : R y z, it is annoying to have to write h x y z h1 h2
to prove R x z. After all, Lean should be able to infer that we are talking about transitivity at x, y, and z,
from the fact that h1 is R x y and h2 is R y z. Indeed, we can replace that information by underscores:
example (h : transitive R) (x y z : A) (h1 : R x y)

(h2 : R y z) :
R x z :=
h _ _ _ h1 h2
But typing underscores is annoying, too. The best solution is to declare the arguments x y z to a transitivity
hypothesis to be implicit as well:
variable {A : Type}
def transitive (R : A → A → Prop) : Prop :=

∀ {x y z}, R x y → R y z → R x z
example (h : transitive R) (x y z : A) (h1 : R x y) (h2 : R y z) :

R x z :=
h h1 h2
In fact, the notions reflexive, symmetric, transitive, and antisymmetric are defined in Lean’s core
library in exactly this way, so we are free to use them without defining them. (That is why we put our
temporary definitions of in a namespace hide; that means that the full name of our version of reflexive is
hide.reflexive, which, therefore, doesn’t conflict with the one defined in the library.‘‘
In Section 13.1 we showed that a strict partial order — that is, a binary relation that is transitive and
irreflexive — is also asymmetric. Here is a proof of that fact in Lean.
variable A : Type
example (h1 : irreflexive R) (h2 : transitive R) :

∀ x y, R x y → ¬ R y x :=
assume x y,
assume h3 : R x y,
assume h4 : R y x,
have h5 : R x x, from h2 h3 h4,
have h6 : ¬ R x x, from h1 x,
124 Chapter 14. Relations in Lean

In mathematics, it is common to use infix notation and a symbol like ≤ to denote a partial order. Lean
supports this practice:
section
parameter A : Type
parameter R : A → A → Prop
local infix ≤ := R
example (h1 : irreflexive R) (h2 : transitive R) :

∀ x y, x ≤ y → ¬ y ≤ x :=
assume x y,
assume h3 : x ≤ y,
assume h4 : y ≤ x,
have h5 : x ≤ x, from h2 h3 h4,
have h6 : ¬ x ≤ x, from h1 x,
end
The parameter and parameters commands are similar to the variable and variables commands, except
that parameters are fixed within a section. In other words, if you prove a theorem about R in the section
above, you cannot apply that theorem to another relation, S, without closing the section. Since the parameter
R is fixed, Lean allows us to define notation for R to be used locally in the section.
In the example below, having fixed a partial order, R, we define the corresponding strict partial order and
prove that it is, indeed, a strict order.
section
parameters {A : Type} (R : A → A → Prop)
parameter (reflR : reflexive R)
parameter (transR : transitive R)
parameter (antisymmR : ∀ {a b : A}, R a b → R b a → a = b)
local infix ≤ := R
definition R' (a b : A) : Prop := a ≤ b ∧ a ̸= b
local infix < := R'
theorem irrefl (a : A) : ¬ a < a :=

assume : a < a,
have a ̸= a, from and.right this,
have a = a, from rfl,
show false, from ‹a ̸= a› ‹a = a›
theorem trans {a b c : A} (h1 : a < b) (h2 : b < c) : a < c :=

have a ≤ b, from and.left h1 ,
have a ̸= b, from and.right h1 ,
have b ≤ c, from and.left h2 ,
have b ̸= c, from and.right h2 ,
have a ≤ c, from transR ‹a ≤ b› ‹b ≤ c›,
have a ̸= c, from
assume : a = c,
have c ≤ b, from eq.subst ‹a = c› ‹a ≤ b›,
have b = c, from antisymmR ‹b ≤ c› ‹c ≤ b›,
show false, from ‹b ̸= c› ‹b = c›,
show a < c, from and.intro ‹a ≤ c› ‹a ̸= c›
end
14.1. Order Relations 125

Notice that we have used suggestive names reflR, transR, antisymmR instead of h1, h2, h3 to help remember
which hypothesis is which. The proof also uses anonymous have and assume, referring back to them with
the French quotes, \f< anf \f>. Remember also that eq.subst ‹a = c› ‹a ≤ b› is a proof of the fact
that amounts for substituting c for a in a ≤ b. You can also use the equivalent notation
Notice that we have used the command open eq.ops to avail ourselves of the extra notation for equality
proofs, so that the expression ‹a = c› ▶ ‹a ≤ b›, where the triangle is written \t.
In Section Section 13.1, we also noted that you can define a (weak) partial order from a strict one. We ask
you to do this formally in the exercises below.
Here is one more example. Suppose R is a binary relation on a type A, and we define S x y to mean that
both R x y and R y x holds. Below we show that the resulting relation is reflexive and symmetric.
section
parameter A : Type
parameter R : A → A → Prop
variable h1 : transitive R
variable h2 : reflexive R
def S (x y : A) := R x y ∧ R y x
example : reflexive S :=
assume x,
have R x x, from h2 x,
show S x x, from and.intro this this
example : symmetric S :=
assume x y,
assume h : S x y,
have h1 : R x y, from h.left,
have h2 : R y x, from h.right,
show S y x, from ⟨h.right, h.left⟩
end
In the exercises below, we ask you to show that S is transitive as well.

In the first example, we use the anonymous assume and have, and then refer back to the have with the
keyword this. In the second example, we abbreviate and.left h and and.right h as h.left and h.right,
respectively. We also abbreviate and.intro h.right h.left with an anonymous constructor, writing ⟨h.
right, h.left⟩. Lean figures out that we are trying to prove a conjunction, and figures out that and.intro
is the relevant introduction principle. You can type the corner brackets with \< and \>, respectively.
14.2 Orderings on Numbers
Conveniently, Lean has the normal orderings on the natural numbers, integers, and so on defined already.
open nat
variables n m : N
#check 0 ≤ n
#check n < n + 1

example : 0 ≤ n := zero_le n
example : n < n + 1 := lt_succ_self n
example (h : n + 1 ≤ m) : n < m + 1 :=
have h1 : n < n + 1, from lt_succ_self n,
have h2 : n < m, from lt_of_lt_of_le h1 h,
have h3 : m < m + 1, from lt_succ_self m,
show n < m + 1, from lt.trans h2 h3
There are many theorems in Lean that are useful for proving facts about inequality relations. We list some
common ones here.
variables (A : Type) [partial_order A]

variables a b c : A
#check (le_trans : a ≤ b → b ≤ c → a ≤ c)
#check (lt_trans : a < b → b < c → a < c)
#check (lt_of_lt_of_le : a < b → b ≤ c → a < c)
#check (lt_of_le_of_lt : a ≤ b → b < c → a < c)
#check (le_of_lt : a < b → a ≤ b)
Here the declaration at the top says that A has the structure of a partial order. There are also properties
that are specific to some domains, like the natural numbers:
variable n : N
#check (nat.zero_le : ∀ n : N, 0 ≤ n)
#check (nat.lt_succ_self : ∀ n : N, n < n + 1)
#check (nat.le_succ : ∀ n : N, n ≤ n + 1)
14.3 Exercises
1. Replace the sorry commands in the following proofs to show that we can create a partial order R' out
of a strict partial order R.
section
parameters {A : Type} {R : A → A → Prop}
parameter (irreflR : irreflexive R)
local infix < := R
def R' (a b : A) : Prop := R a b ∨ a = b

local infix ≤ := R'
theorem reflR' (a : A) : a ≤ a := sorry
theorem transR' {a b c : A} (h1 : a ≤ b) (h2 : b ≤ c):

a ≤ c :=
sorry
theorem antisymmR' {a b : A} (h1 : a ≤ b) (h2 : b ≤ a) :

a = b :=
sorry
14.3. Exercises 127

end
2. Replace the sorry by a proof.
section
parameters {A : Type} {R : A → A → Prop}
parameter (reflR : reflexive R)
def S (a b : A) : Prop := R a b ∧ R b a
example : transitive S :=
sorry
end
3. Only one of the following two theorems is provable. Figure out which one is true, and replace the
sorry command with a complete proof.
section
parameters {A : Type} {a b c : A} {R : A → A → Prop}
parameter (Rab : R a b)
parameter (Rbc : R b c)
parameter (nRac : ¬ R a c)
-- Prove one of the following two theorems:
theorem R_is_strict_partial_order :
irreflexive R ∧ transitive R :=
sorry
theorem R_is_not_strict_partial_order :
¬(irreflexive R ∧ transitive R) :=
sorry
end
4. Complete the following proof.
open nat
example : 1 ≤ 4 :=
sorry

CHAPTER
FIFTEEN
FUNCTIONS
In the late nineteenth century, developments in a number of branches of mathematics pushed towards a
uniform treatment of sets, functions, and relations. We have already considered sets and relations. In this
chapter, we consider functions and their properties.
A function, f , is ordinary understood as a mapping from a domain X to another domain Y . In set-theoretic
foundations, X and Y are arbitrary sets. We have seen that in a type-based system like Lean, it is natural to
distinguish between types and subsets of a type. In other words, we can consider a type X of elements, and
a set A of elements of that type. Thus, in the type-theoretic formulation, it is natural to consider functions
between types X and Y, and consider their behavior with respect to subsets of X and Y.
In everyday mathematics, however, set-theoretic language is common, and most mathematicians think of
a function as a map between sets. When discussing functions from a mathematical standpoint, therefore,
we will also adopt this language, and later switch to the type-theoretic representation when we talk about
formalization in Lean.
15.1 The Function Concept
If X and Y are any sets, we write f : X → Y to express the fact that f is a function from X to Y . This
means that f assigns a value f (x) in Y to every element x of X. The set X is called the domain of f , and
the set Y is called the codomain. (Some authors use the word “range” for the codomain, but today it is
more common to use the word “range” for what we call the image of A below. We will avoid the ambiguity
by avoiding the word range altogether.)
The simplest way to define a function is to give its value at every x with an explicit expression. For example,
we can write any of the following:
• Let f : N → N be the function defined by f (n) = n + 1.
• Let g : R → R be the function defined by g(x) = x2 .
• Let h : N → N be the function defined by h(n) = n2 .
• Let k : N → {0, 1} be the function defined by
{
0 if n is even
k(n) =
1 if n is odd.
The ability to define functions using an explicit expression raises the foundational question as to what counts
as legitimate “expression.” For the moment, let us set that question aside, and simply note that modern
mathematics is comfortable with all kinds of exotic definitions. For example, we can define a function
f : R → {0, 1} by
{
0 if x is rational
f (x) =
1 if x is irrational.
129
This is at odds with a view of functions as objects that are computable in some sense. It is not at all clear
what it means to be presented with a real number as input, let alone whether it is possible to determine,
algorithmically, whether such a number is rational or not. We will return to such issues in a later chapter.
Notice that the choice of the variables x and n in the definitions above are arbitrary. They are bound
variables in that the functions being defined do not depend on x or n. The values remain the same under
renaming, just as the truth values of “for every x, P (x)” and “for every y, P (y)” are the same. Given
an expression e(x) that depends on the variable x, logicians often use the notation λx e(x) to denote the
function that maps x to e(x). This is called “lambda notation,” for the obvious reason, and it is often quite
handy. Instead of saying “let f be the function defined by f (x) = x + 1,” we can say “let f = λ x(x + 1).”
This is not common mathematical notation, and it is best to avoid it unless you are talking to logicians or
computer scientists. We will see, however, that lambda notation is built in to Lean.
For any set X, we can define a function iX (x) by the equation iX (x) = x. This function is called the identity
function. More interestingly, let f : X → Y and g : Y → Z. We can define a new function k : X → Z by
k(x) = g(f (x)). The function k is called the composition of f and g or f composed with g and it is written
g ◦f . The order is somewhat confusing; you just have to keep in mind that to evaluate the expression g(f (x))
you first evaluate f on input x, and then evaluate g.
We think of two functions f, g : X → Y as being equal, or the same function, when for they have the same
values on every input; in other words, for every x in X, f (x) = g(x). For example, if f, g : R → R are defined
by f (x) = x + 1 and g(x) = 1 + x, then f = g. Notice that the statement that two functions are equal is a
universal statement (that is, for the form “for every x, …”).
Proposition. For every f : X → Y , f ◦ iX = f and iY ◦ f = f .

Proof. Let x be any element of X. Then (f ◦ iX )(x) = f (iX (x)) = f (x), and (iY ◦ f )(x) = iY (f (x)) = x.
Suppose f : X → Y and g : Y → X satisfy g ◦ f = iX . Remember that this means that g(f (x)) = x for
every x in X. In that case, g is said to be a left inverse to f , and f is said to be a right inverse to g. Here
are some examples:
• Define f, g : R → R by f (x) = x + 1 and g(x) = x − 1. Then g is both a left and a right inverse to f ,
and vice-versa.
• Write R≥0 √ ≥0
√ 2reals. Define f : R → R by f (x) = x , and define g : R → R
to denote the nonnegative 2 ≥0
by g(x) = x. Then f (g(x)) = ( x) = x for every x in the √ domain of g, so f is a left inverse to g,

and g is a right inverse to f . On the other hand, g(f (x)) = x2 = |x|, which is not the same as x
when x is negative. So g is not a left inverse to f , and f is not a right inverse to g.
The following fact is not at all obvious, even though the proof is short:
Proposition. Suppose f : X → Y has a left inverse, h, and a right inverse k. Then h = k.

Proof. Let y be any element in Y . The idea is to compute h(f (k(y)) in two different ways. Since h is a left
inverse to f , we have h(f (k(y))) = k(y). On the other hand, since k is a right inverse to f , f (k(y)) = y, and
so h(f (k(y)) = h(y). So k(y) = h(y).
If g is both a right and left inverse to f , we say that g is simply the inverse of f . A function f may have
more than one left or right inverse (we leave it to you to cook up examples), but it can have at most one
inverse.
130 Chapter 15. Functions

Proposition. Suppose g1 , g2 : Y → X are both inverses to f . Then g1 = g2 .

Proof. This follows from the previous proposition, since (say) g1 is a left inverse to f , and g2 is a right
inverse.
When f has an inverse, g, this justifies calling g the inverse to f , and writing f −1 to denote g. Notice that if
f −1 is an inverse to f , then f is an inverse to f −1 . So if f has an inverse, then so does f −1 , and (f −1 )−1 = f .
For any set A, clearly we have i−1 X = iX .
Proposition. Suppose f : X → Y and g : Y → Z. If h : Y → X is a left inverse to f and k : Z → Y is a

left inverse to g, then h ◦ k is a left inverse to g ◦ f .
Proof. For every x in X,
(h ◦ k) ◦ (g ◦ f )(x) = h(k(g(f (x)))) = h(f (x)) = x.
Corollary. The previous proposition holds with “left” replaced by “right.”

Proof. Switch the role of f with h and g with k in the previous proposition.
Corollary. If f : X → Y and g : Y → Z both have inverses, then (f ◦ g)−1 = g −1 ◦ f −1 .
15.2 Injective, Surjective, and Bijective Functions
A function f : X → Y is said to be injective, or an injection, or one-one, if given any x1 and x2 in A, if

f (x1 ) = f (x2 ), then x1 = x2 . Notice that the conclusion is equivalent to its contrapositive: if x1 ̸= x2 , then
f (x1 ) ̸= f (x2 ). So f is injective if it maps distinct element of X to distinct elements of Y .
A function f : X → Y is said to be surjective, or a surjection, or onto, if for every element y of Y , there is
an x in X such that f (x) = y. In other words, f is surjective if every element in the codomain is the value
of f at some element in the domain.
A function f : X → Y is said to be bijective, or a bijection, or a one-to-one correspondence, if it is both
injective and surjective. Intuitively, if there is a bijection between X and Y , then X and Y have the same
size, since f makes each element of X correspond to exactly one element of Y and vice-versa. For example,
it makes sense to interpret the statement that there were four Beatles as the statement that there is a
bijection between the set {1, 2, 3, 4} and the set {John, Paul, George, Ringo}. If we claimed that there were
five Beatles, as evidenced by the function f which assigns 1 to John, 2 to Paul, 3 to George, 4 to Ringo, and
5 to John, you should object that we double-counted John — that is, f is not injective. If we claimed there
were only three Beatles, as evidenced by the function f which assigns 1 to John, 2 to Paul, and 3 to George,
you should object that we left out poor Ringo — that is, f is not surjective.
The next two propositions show that these notions can be cast in terms of the existence of inverses.
Proposition. Let f : X → Y .
• If f has a left inverse, then f is injective.
• If f has a right inverse, then f is surjective.
• If f has an inverse, then it is f bijective.
15.2. Injective, Surjective, and Bijective Functions 131

Proof. For the first claim, suppose f has a left inverse g, and suppose f (x1 ) = f (x2 ). Then g(f (x1 )) =
g(f (x2 )), and so x1 = x2 .
For the second claim, suppose f has a right inverse h. Let y be any element of Y , and let x = g(y). Then
f (x) = f (g(y)) = y.
The third claim follows from the first two.
The following proposition is more interesting, because it requires us to define new functions, given hypotheses
on f .
Proposition. Let f : X → Y .
• If X is nonempty and f is injective, then f has a left inverse.
• If f is surjective, then f has a right inverse.
• If f if bijective, then it has an inverse.
Proof. For the first claim, let x̂ be any element of X, and suppose f is injective. Define g : Y → X by setting
g(y) equal to any x such that f (x) = y, if there is one, and x̂ otherwise. Now, suppose g(f (x)) = x′ . By the
definition of g, x′ has to have the property that f (x) = f (x′ ). Since f is injective, x = x′ , so g(f (x)) = x.
For the second claim, because f is surjective, we know that for every y in Y there is any x such that f (x) = y.
Define h : B → A by again setting h(y) equal to any such x. (In contrast to the previous paragraph, here we
know that such an x exists, but it might not be unique.) Then, by the definition of h, we have f (h(y)) = y.
Notice that the definition of g in the first part of the proof requires the function to “decide” whether there
is an x in X such that f (x) = y. There is nothing mathematically dubious about this definition, but in
many situations, this cannot be done algorithmically; in other words, g might not be computable from the
data. More interestingly, the definition of h in the second part of the proof requires the function to “choose”
a suitable value of x from among potentially many candidates. We will see in Section 23.3 that this is a
version of the axiom of choice. In the early twentieth century, the use of the axiom of choice in mathematics
was hotly debated, but today it is commonplace.
Using these equivalences and the results in the previous section, we can prove the following:
Proposition. Let f : X → B and g : Y → Z.

• If f and g are injective, then so is g ◦ f .
• If f and g are surjective, then so is g ◦ f .
Proof. If f and g are injective, then they have left inverses h and k, respectively, in which case h ◦ k is a
left inverse to g ◦ f . The second statement is proved similarly.
We can prove these two statements, however, without mentioning inverses at all. We leave that to you as an
exercise.
Notice that the expression f (n) = 2n can be used to define infinitely many functions with domain N, such
as:
• a function f : N → N
• a function f : N → R

• a function f : N → {n | n is even}
Only the third one is surjective. Thus a specification of the function’s codomain as well as the domain is
essential to making sense of whether a function is surjective.
15.3 Functions and Subsets of the Domain
Suppose f is a function from X to Y . We may wish to reason about the behavior of f on some subset A
of X. For example, we can say that f is injective on A if for every x1 and x2 in A, if f (x1 ) = f (x2 ), then
x1 = x2 .
If f is a function from X to Y and A is a subset of X, we write f [A] to denote the image of f on A, defined
by
f [A] = {y ∈ Y | y = f (x) for some x in A}.
In words, f [A] is the set of elements of Y that are “hit” by elements of A under the mapping f . Notice
that there is an implicit existential quantifier here, so that reasoning about images invariables involves the
corresponding rules.
Proposition. Suppose f : X → Y , and A is a subset of X. Then for any x in A, f (x) is in f [A].

Proof. By definition, f (x) is in f [A] if and only if there is some x′ in A such that f (x′ ) = f (x). But that
holds for x′ = x.
Proposition. Suppose f : X → Y and g : Y → Z. Let A be a subset of X. Then
(g ◦ f )[A] = g[f [A]].
Proof. Suppose z is in (g ◦ f )[A]. Then for some x ∈ A, z = (g ◦ f )(x) = g(f (x)). By the previous
proposition, f (x) is in f [A]. Again by the previous proposition, g(f (x)) is in g[f [A]].
Conversely, suppose z is in g[f [A]]. Then there is a y in f [A] such that f (y) = z, and since y is in f [D],
there is an x in A such that f (x) = y. But then (g ◦ f )(x) = g(f (x)) = g(y) = z, so z is in (g ◦ f )[A].
Notice that if f is a function from X to Y , then f is surjective if and only if f [X] = Y . So the previous
proposition is a generalization of the fact that the composition of surjective functions is surjective.
Suppose f is a function from X to Y , and A is a subset of X. We can view f as a function from A to Y ,
by simply ignoring the behavior of f on elements outside of A. Properly speaking, this is another function,
denoted f ↾ A and called “the restriction of f to A.” In other words, given f : X → Y and A ⊆ X,
f ↾ A : A → Y is the function defined by (f ↾ A)(x) = x for every x in A. Notice that now “f is injective on
A” means simply that the restriction of f to A is injective.
There is another important operation on functions, known as the preimage. If f : X → Y and B ⊆ Y , then
the preimage of B under f , denoted f −1 [B], is defined by
f −1 [B] = {x ∈ X | f (x) ∈ B},
that is, the set of elements of X that get mapped into B. Notice that this makes sense even if f does not
have an inverse; for a given y in B, there may be no x‘s with the property f (x) ∈ B, or there may be many.
If f has an inverse, f −1 , then for every y in B there is exactly one x ∈ X with the property f (x) ∈ B,
in which case, f −1 [B] means the same thing whether you interpret it as the image of B under f −1 or the
preimage of B under f .
15.3. Functions and Subsets of the Domain 133

Proposition. Suppose f : X → Y and g : Y → Z. Let C be a subset of Z. Then
(g ◦ f )−1 [C] = f −1 [g −1 [C]].
Proof. For any y in C, y is in (g ◦ f )−1 [C] if and only if g(f (y)) is in C. This, in turn, happens if and only
if f (y) is in g −1 [C], which in turn happens if and only if y is in f −1 [g −1 [C]].
Here we give a long list of facts properties of images and preimages. Here, f denotes an arbitrary function
from X to Y , A, A1 , A2 , . . . denote arbitrary subsets of X, and B, B1 , B2 , . . . denote arbitrary subsets of Y .
• A ⊆ f −1 [f [A]], and if f is injective, A = f −1 [f [A]].
• f [f −1 [B]] ⊆ B, and if f is surjective, B = f [f −1 [B]].
• If A1 ⊆ A2 , then f [A1 ] ⊆ f [A2 ].
• If B1 ⊆ B2 , then f −1 [B1 ] ⊆ f −1 [B2 ].
• f [A1 ∪ A2 ] = f [A1 ] ∪ f [A2 ].
• f −1 [B1 ∪ B2 ] = f −1 [B1 ] ∪ f −1 [B2 ].
• f [A1 ∩ A2 ] ⊆ f [A1 ] ∩ f [A2 ], and if f is injective, f [A1 ∩ A2 ] = f [A1 ] ∩ f [A2 ].
• f −1 [B1 ∩ B2 ] = f −1 [B1 ] ∩ f −1 [B2 ].
• f [A] \ f [B] ⊆ f [A \ B].
• f −1 [A] \ f −1 [B] ⊆ f [A \ B].
• f [A] ∩ B = f [A ∩ f −1 [B]].
• f [A] ∪ B ⊇ f [A ∪ f −1 [B]].
• A ∩ f −1 [B] ⊆ f −1 [f [A] ∩ B].
• A ∪ f −1 [B] ⊆ f −1 [f [A] ∪ B].
Proving identities like this is typically a matter of unfolding definitions and using basic logical inferences.
Here is an example.
Proposition. Let X and Y be sets, f : X → Y , A ⊆ X, and B ⊆ Y . Then f [A] ∩ B = f [A ∩ f −1 [B]].

Proof. Suppose y ∈ f [A] ∩ B. Then y ∈ B, and for some x ∈ A, f (x) = y. But this means that x is in
f −1 [B], and so x ∈ A ∩ f −1 [B]. Since f (x) = y, we have y ∈ f [A ∩ f −1 [B]], as needed.
Conversely, suppose y ∈ f [A ∩ f −1 [B]]. Then for some x ∈ A ∩ f −1 [B], we have f (x) = y. For this x, have
x ∈ A and f (x) ∈ B. Since f (x) = y, we have y ∈ B, and since x ∈ A, we also have y ∈ f [A], as required.

15.4 Functions and Relations
A binary relation R(x, y) on A and B is functional if for every x in A there exists a unique y in B such that
R(x, y). If R is a functional relation, we can define a function fR : X → B by setting fR (x) to be equal to
the unique y in B such that R(x, y). Conversely, it is not hard to see that if f : X → B is any function, the
relation Rf (x, y) defined by f (x) = y is a functional relation. The relation Rf (x, y) is known as the graph
of f .
It is not hard to check that functions and relations travel in pairs: if f is the function associated with a
functional relation R, then R is the functional relation associated the function f , and vice-versa. In set-
theoretic foundations, a function is often defined to be a functional relation. Conversely, we have seen that
in type-theoretic foundations like the one adopted by Lean, relations are often defined to be certain types
of functions. We will discuss these matters later on, and in the meanwhile only remark that in everyday
mathematical practice, the foundational details are not so important; what is important is simply that every
function has a graph, and that any functional relation can be used to define a corresponding function.
So far, we have been focusing on functions that take a single argument. We can also consider functions
f (x, y) or g(x, y, z) that take multiple arguments. For example, the addition function f (x, y) = x + y on the
integers takes two integers and returns an integer. Remember, we can consider binary functions, ternary
functions, and so on, and the number of arguments to a function is called its “arity.” One easy way to make
sense of functions with multiple arguments is to think of them as unary functions from a cartesian product.
We can think of a function f which takes two arguments, one in A and one in B, and returns an argument
in C as a unary function from A × B to C, whereby f (a, b) abbreviates f ((a, b)). We have seen that in
dependent type theory (and in Lean) it is more convenient to think of such a function f as a function which
takes an element of A and returns a function from B → C, so that f (a, b) abbreviates (f (a))(b). Such a
function f maps A to B → C, where B → C is the set of functions from B to C.
We will return to these different ways of modeling functions of higher arity later on, when we consider
set-theoretic and type-theoretic foundations. One again, we remark that in ordinary mathematics, the
foundational details to not matter much. The two choices above are inter-translatable, and sanction the
same principles for reasoning about functions informally.
In mathematics, we often also consider the notion of a partial function from X to Y , which is really a
function from some subset of X to Y . The fact that f is a partial function from X to Y is sometimes written
f : X ↛ Y , which should be interpreted as saying that f : A → Y for some subset A of Y . Intuitively, we
think of f as a function from X → Y which is simply “undefined” at some of its inputs; for example, we can
think of f : R ↛ R defined by f (x) = 1/x, which is undefined at x = 0, so that in reality f : R \ {0} → R.
The set A is sometimes called the domain of f , in which case, there is no good name for X; others continue
to call X the domain, and refer to A as the domain of definition. To indicate that a function f is defined at
x, that is, that x is in the domain of definition of f , we sometimes write f (x) ↓. If f and g are two partial
functions from X to Y , we write f (x) ≃ g(x) to mean that either f and g are both defined at x and have the
same value, or are both undefined at x. Notions of injectivity, surjectivity, and composition are extended to
partial functions, generally as you would expect them to be.
In terms of relations, a partial function f corresponds to a relation Rf (x, y) such that for every x there is at
most one y such that Rf (x, y) holds. Mathematicians also sometimes consider multifunctions from X to Y ,
which correspond to relations Rf (x, y) such that for every x in X, there is at least one y such that Rf (x, y)
holds. There may be many such y; you can think of these as functions which have more than one input
value. If you think about it for a moment, you will see that a partial multifunction is essentially nothing
more than an arbitrary relation.
15.5 Exercises
1. Let f be any function from X to Y , and let g be any function from Y to Z.
15.4. Functions and Relations 135

• Show that if g ◦ f is injective, then f is injective.

• Give an example of functions f and g as above, such that that g ◦ f is injective, but g is not
injective.
• Show that if g ◦ f is injective and f is surjective, then g is injective.
2. Let f and g be as in the last problem. Suppose g ◦ f is surjective.
• Is f necessarily surjective? Either prove that it is, or give a counterexample.
• Is g necessarily surjective? Either prove that it is, or give a counterexample.
3. A function f from R to R is said to be strictly increasing if whenever x1 < x2 , f (x1 ) < f (x2 ).
• Show that if f : R → R is strictly increasing, then it is injective (and hence it has a left inverse).
• Show that if f : R → R is strictly increasing, and g is a right inverse to f , then g is strictly
increasing.
4. Let f : X → Y be any function, and let A and B be subsets of X. Show that f [A ∪ B] = f [A] ∪ f [B].
5. Let f : X → Y be any function, and let A and B be any subsets of X. Show f [A] \ f [B] ⊆ f [A \ B].
6. Define notions of composition and inverse for binary relations that generalize the notions for functions.

CHAPTER
SIXTEEN
FUNCTIONS IN LEAN
16.1 Functions and Symbolic Logic
Let us now consider functions in formal terms. Even though we have avoided the use of quantifiers and
logical symbols in the definitions in the last chapter, by now you should be seeing them lurking beneath the
surface. That fact that two functions f, g : X → Y are equal if and only if they take the same values at
every input can be expressed as follows:
∀x ∈ X (f (x) = g(x)) ↔ f = g
This principle is a known as function extensionality, analogous to the principle of extensionality for sets,
discussed in Section 12.1. Recall that the notation ∀x ∈ X P (x) abbreviates ∀x (x ∈ X → P (x)), and
∃x ∈ X P (x) abbreviates ∃x (x ∈ X ∧ P (x)), thereby relativizing the quantifiers to A.
We can avoid set-theoretic notation if we assume we are working in a logical formalism with basic types for
X and Y , so that we can specify that x ranges over X. In that case, we will write instead
∀x : X (f (x) = g(x) ↔ f = g)
to indicate that the quantification is over X. Henceforth, we will assume that all our variables range over
some type, though we will sometimes omit the types in the quantifiers when they can be inferred from
context.
The function f is injective if it satisfies
∀x1 , x2 : X (f (x1 ) = f (x2 ) → x1 = x2 ),
and f is surjective if
∀y : Y ∃x : X f (x) = y.
If f : X → Y and g : Y → X, g is a left inverse to f if
∀x : X g(f (x)) = a.
Notice that this is a universal statement, and it is equivalent to the statement that f is a right inverse to g.
Remember that in logic it is common to use lambda notation to define functions. We can denote the identity
function by λx x, or perhaps λx : X x to emphasize that the domain of the function is X. If f : X → Y and
g : Y → Z, we can define the composition g ◦ f by g ◦ f = λx : X g(f (x)).
Also remember that if P (x) is any predicate, then in first order logic we can assert that there exists a unique
x satisfying P (x), written ∃!x P (x), with the conjunction of the following two statements:
• ∃x P (x)
137
• ∀x1 , x2 (P (x1 ) ∧ P (x2 ) → x1 = y1 )

Equivalently, we can write
∃(P (x) ∧ ∀x′ (P (x′ ) → x′ = x)).
Assuming ∃!x P (x), the following two statements are equivalent:

• ∃x (P (x) ∧ Q(x))
• ∀x (P (x) → Q(x))
and both can be taken to assert that “the x satisfying P also satisfies Q.”
A binary relation R on X and Y is functional if it satisfies
∀x ∃!y R(x, y).
In that case, a logician might use iota notation,
f (x) = ιy R(x, y)
to define f (x) to be equal to the unique y satisfying R(x, y). If R satisfies the weaker property
∀x ∃y R(x, y),
a logician might use the Hilbert epsilon to define a function
f (x) = εy R(x, y)
to “choose” a value of y satisfying R(x, y). As we have noted above, this is an implicit use of the axiom of
choice.
16.2 Second- and Higher-Order Logic
In contrast to first-order logic, where we start with a fixed stock of function and relation symbols, the topics
we have been considering in the last few chapters encourage us to consider a more expressive language with
variables ranging over functions and relations as well. For example, saying that a function f : X → Y has a
left-inverse implicitly involves a quantifying over functions,
∃g ∀x g(f (x)) = x.
The theorem that asserts that if any function f from X to Y is injective then it has a left-inverse can be
expressed as follows:
∀x1 , x2 (f (x1 ) = f (x2 ) → x1 = x2 ) → ∃g ∀x g(f (x)) = x.
Similarly, saying that two sets X and Y have a one-to-one correspondence asserts the existence of a function
f : X → Y as well as an inverse to f . For another example, in Section 15.4 we asserted that every functional
relation gives rise to a corresponding function, and vice-versa.
What makes these statements interesting is that they involve quantification, both existential and universal,
over functions and relations. This takes us outside the realm of first-order logic. One option is to develop a
theory in the language of first-order logic in which the universe contains functions, and relations as objects;
we will see later that this is what axiomatic set theory does. An alternative is to extend first-order logic to
involve new kinds of quantifiers and variables, to range over functions and relations. This is what higher-order
logic does.
138 Chapter 16. Functions in Lean

There are various ways to go about this. In view of the relationship between functions and relations described
above, one can take relations as basic, and define functions in terms of them, or vice-versa. The following
formulation of higher-order logic, due to the logician Alonzo Church, follows the latter approach. It is
sometimes known as simple type theory.
Start with some basic types, X, Y, Z, . . . and a special type, Prop, of propositions. Add the following two
rules to build new types:
• If U and V are types, so is U × V .
• If U and V are types, so is U → V .
The first intended to denote the type of ordered pairs (u, v), where u is in U and v is in V . The second is
intended to denote the type of functions from U to V . Simple type theory now adds the following means of
forming expressions:
• If u is of type U and v is of type V , (u, v) is of type v.
• If p is of type U × V , then (p)1 is of type U and (p)2 if of type V . (These are intended to denote the
first and second element of the pair p.)
• If x is a variable of type U , and v is any expression of type V , then λx v is of type U → V .
• If f is of type U → V and u is of type U , f (u) is of type V .
In addition, simple type theory provides all the means we have in first-order logic — boolean connectives,
quantifiers, and equality — to build propositions.
A function f (x, y) which takes elements of X and Y to a type Z is viewed as an object of type X × Y → Z.
Similarly, a binary relation R(x, y) on X and Y is viewed as an object of type X × Y → Prop. What makes
higher-order logic “higher order” is that we can iterate the function type operation indefinitely. For example,
if N is the type of natural numbers, N → N denotes the type of functions from the natural numbers to the
natural numbers, and (N → N) → N denotes the type of functions F (f ) which take a function as argument,
and returns a natural number.
We have not specified the syntax and rules of higher-order logic very carefully. This is done in a number of
more advanced logic textbooks. The fragment of higher-order logic which allows only functions and relations
on the basic types (without iterating these constructions) is known as second-order logic.
These notions should seem familiar; we have been using these constructions, with similar notation, in Lean.
Indeed, Lean’s logic is an even more elaborate and expressive system of logic, which fully subsumes all the
notions of higher-order logic we have discussed here.
16.3 Functions in Lean
The fact that the notions we have been discussing have such a straightforward logical form means that it
is easy to define them in Lean. The main difference between the formal representation in Lean and the
informal representation above is that, in Lean, we distinguish between a type X and a subset A : set X of
that type.
In Lean’s library, composition and identity are defined as follows:
variables {X Y Z : Type}
def comp (f : Y → Z) (g : X → Y) : X → Z :=
λx, f (g x)
infixr ` ◦ ` := comp
16.3. Functions in Lean 139

def id (x : X) : X :=
x
Ordinarily, we use funext (for “function extensionality”) to prove that two functions are equal.
example (f g : X → Y) (h : ∀ x, f x = g x) : f = g :=
funext h
But Lean can prove some basic identities by simply unfolding definitions and simplifying expressions, using
reflexivity.
lemma left_id (f : X → Y) : id ◦ f = f := rfl
lemma right_id (f : X → Y) : f ◦ id = f := rfl
theorem comp.assoc (f : Z → W) (g : Y → Z) (h : X → Y) :
(f ◦ g) ◦ h = f ◦ (g ◦ h) := rfl
theorem comp.left_id (f : X → Y) : id ◦ f = f := rfl
theorem comp.right_id (f : X → Y) : f ◦ id = f := rfl
We can define what it means for f to be injective, surjective, or bijective:

def injective (f : X → Y) : Prop := ∀ {|x1 x2 |}, f x1 = f x2 → x1 = x2
def surjective (f : X → Y) : Prop := ∀ y, ∃ x, f x = y
def bijective (f : X → Y) := injective f ∧ surjective f
Marking the variables x1 and x2 implicit in the definition of injective means that we do not have to write
them as often. Specifically, given h : injective f, and h1 : f x1 = f x2 , we write h h1 rather than h
x1 x2 h1 to show x1 = x2 .
We can then prove that the identity function is bijective:
theorem injective_id : injective (@id X) :=
assume x1 x2 ,
assume H : id x1 = id x2 ,
show x1 = x2 , from H
theorem surjective_id : surjective (@id X) :=

assume y,
show ∃ x, id x = y, from exists.intro y rfl
theorem bijective_id : bijective (@id X) :=

and.intro injective_id surjective_id
More interestingly, we can prove that the composition of injective functions is injective, and so on.
theorem injective_comp {g : Y → Z} {f : X → Y}
(Hg : injective g) (Hf : injective f) :
injective (g ◦ f) :=
assume x1 x2 ,
assume : (g ◦ f) x1 = (g ◦ f) x2 ,
have f x1 = f x2 , from Hg this,
show x1 = x2 , from Hf this

theorem surjective_comp {g : Y → Z} {f : X → Y}
(hg : surjective g) (hf : surjective f) :
surjective (g ◦ f) :=
assume z,
exists.elim (hg z) $
assume y (hy : g y = z),
exists.elim (hf y) $
assume x (hx : f x = y),
have g (f x) = z, from eq.subst (eq.symm hx) hy,
show ∃ x, g (f x) = z, from exists.intro x this
theorem bijective_comp {g : Y → Z} {f : X → Y}
(hg : bijective g) (hf : bijective f) :
bijective (g ◦ f) :=
have ginj : injective g, from hg.left,
have gsurj : surjective g, from hg.right,
have finj : injective f, from hf.left,
have fsurj : surjective f, from hf.right,
and.intro (injective_comp ginj finj)
(surjective_comp gsurj fsurj)
The notions of left and right inverse are defined in the expected way.
-- g is a left inverse to f
def left_inverse (g : Y → X) (f : X → Y) : Prop := ∀ x, g (f x) = x
-- g is a right inverse to f
def right_inverse (g : Y → X) (f : X → Y) : Prop := left_inverse f g
In particular, composing with a left or right inverse yields the identity.
def id_of_left_inverse {g : Y → X} {f : X → Y} : left_inverse g f → g ◦ f = id :=

assume H, funext H
def id_of_right_inverse {g : Y → X} {f : X → Y} : right_inverse g f → f ◦ g = id :=

assume H, funext H
Notice that we need to use funext to show the equality of functions.

The following shows that if a function has a left inverse, then it is injective, and if it has a right inverse,
then it is surjective.
theorem injective_of_left_inverse {g : Y → X} {f : X → Y} :
left_inverse g f → injective f :=
assume h, assume x1 x2 , assume feq,
calc x1 = g (f x1 ) : by rw h
... = g (f x2 ) : by rw feq
... = x2 : by rw h
theorem surjective_of_right_inverse {g : Y → X} {f : X → Y} :
right_inverse g f → surjective f :=
assume h, assume y,
let x : X := g y in
have f x = y, from calc
f x = (f (g y)) : rfl
... = y : by rw [h y],
show ∃ x, f x = y, from exists.intro x this
16.3. Functions in Lean 141

16.4 Defining the Inverse Classically
All the theorems listed in the previous section are found in the Lean library, and are available to you when
you open the function namespace with open function:
open function
#check comp
#check left_inverse
#check has_right_inverse
Defining inverse functions, however, requires classical reasoning, which we get by opening the classical
namespace:
open classical
section
variables A B : Type
variable R : A → B → Prop
example : (∀ x, ∃ y, R x y) → ∃ f : A → B, ∀ x, R x (f x) :=
axiom_of_choice
example (h : ∃ x, P x) : P (some h) :=
some_spec h
end
The axiom of choice tells us that if, for every x : X, there is a y : Y satisfying R x y, then there is a
function f : X → Y which, for every x chooses such a y. In Lean, this “axiom” is proved using a classical
construction, the some function (sometimes called “the indefinite description operator”) which, given that
there is some x satisfying P x, returns such an x. With these constructions, the inverse function is defined
as follows:
open classical function

local attribute [instance] prop_decidable
variables {X Y : Type}
noncomputable def inverse (f : X → Y) (default : X) : Y → X :=

λ y, if h : ∃ x, f x = y then some h else default
Lean requires us to acknowledge that the definition is not computational, since, first, it may not be algorith-
mically possible to decide whether or not condition H holds, and even if it does, it may not be algorithmically
possible to find a suitable value of x.
Below, the proposition inverse_of_exists asserts that inverse meets its specification, and the subsequent
theorem shows that if f is injective, then the inverse function really is a left inverse.
theorem inverse_of_exists (f : X → Y) (default : X) (y : Y)

(h : ∃ x, f x = y) :
f (inverse f default y) = y :=
have h1 : inverse f default y = some h, from dif_pos h,
have h2 : f (some h) = y, from some_spec h,
eq.subst (eq.symm h1) h2
theorem is_left_inverse_of_injective (f : X → Y) (default : X)

(injf : injective f) :
left_inverse (inverse f default) f :=
let finv := (inverse f default) in
assume x,
have h1 : ∃ x', f x' = f x, from exists.intro x rfl,
have h2 : f (finv (f x)) = f x, from inverse_of_exists f default (f x) h1,
show finv (f x) = x, from injf h2
16.5 Functions and Sets in Lean
In Section 7.4 we saw how to represent relativized universal and existential quantifiers when formalizing
phrases like “every prime number greater than two is odd” and “some prime number is even.” In a similar
way, we can relativize statements to sets. In symbolic logic, the expression ∃x ∈ A P (x) abbreviates
∃x (x ∈ A ∧ P (x)), and ∀x ∈ A P (x) abbreviates ∀x (x ∈ A → P (x)).
Lean also defines notation for relativized quantifiers:
variables (X : Type) (A : set X) (P : X → Prop)
#check ∀ x ∈ A, P x
#check ∃ x ∈ A, P x
Here is an example of how to use the bounded universal quantifier:
example (h : ∀ x ∈ A, P x) (x : X) (h1 : x ∈ A) : P x := h x h1
Using bounded quantifiers, we can talk about the behavior of functions on particular sets:
import data.set
open set function
variables (A : set X) (B : set Y)
def maps_to (f : X → Y) (A : set X) (B : set Y) := ∀ x ∈ A, f x ∈ B
def inj_on (f : X → Y) (A : set X) := ∀ (x1 ∈ A) (x2 ∈ A), f x1 = f x2 → x1 = x2
def surj_on (f : X → Y) (A : set X) (B : set Y) := B ⊆ f '' A
The expression maps_to f A B asserts that f maps elements of the set A to the set B, and the expression
inj_on f A asserts that f is injective on A. The expression surj_on f A B asserts that, viewed as a function
defined on elements of A, the function f is surjective onto the set B. Here are examples of how they can be
used:
variables (f : X → Y) (A : set X) (B : set Y)
example (h : maps_to f A B) (x : X) (h1 : x ∈ A) : f x ∈ B := h x h1
example (h : inj_on f A) (x1 x2 : X) (h1 : x1 ∈ A) (h2 : x2 ∈ A)

(h3 : f x1 = f x2 ) : x1 = x2 :=
h x1 h1 x2 h2 h3
example (h : surj_on f A B) (y : Y) (h1 : y ∈ B) : ∃ x, x ∈ A ∧ f x = y :=

h h1
16.5. Functions and Sets in Lean 143

Actually, it is slightly more convenient to mark the variables in maps_to and inj_on as implicit:
def maps_to (f : X → Y) (A : set X) (B : set Y) := ∀ {x}, x ∈ A → f x ∈ B
def inj_on (f : X → Y) (A : set X) := ∀ {x1 x2 }, x1 ∈ A → x2 ∈ A → f x1 = f x2 → x1 = x2
In that case, we can leave out some arguments:
variables (f : X → Y) (A : set X) (B : set Y)
example (h : maps_to f A B) (x : X) (h1 : x ∈ A) : f x ∈ B := h h1
example (h : inj_on f A) (x1 x2 : X) (h1 : x1 ∈ A) (h2 : x2 ∈ A)

(h3 : f x1 = f x2 ) : x1 = x2 :=
h h1 h2 h3
In the examples below, we’ll use the versions with implicit arguments.
The expression surj_on f A B asserts that, viewed as a function defined on elements of A, the function f is
surjective onto the set B:
With these notions in hand, we can prove that the composition of injective functions is injective. The proof
is similar to the one above, though now we have to be more careful to relativize claims to A and B:
theorem inj_on_comp (fAB : maps_to f A B) (hg : inj_on g B) (hf: inj_on f A) :

inj_on (g ◦ f) A :=
assume x1 x2 : X,
assume x1A : x1 ∈ A,
assume x2A : x2 ∈ A,
have fx1B : f x1 ∈ B, from fAB x1A,
have fx2B : f x2 ∈ B, from fAB x2A,
assume h1 : g (f x1) = g (f x2),
have h2 : f x1 = f x2, from hg fx1B fx2B h1,
show x1 = x2, from hf x1A x2A h2
We can similarly prove that the composition of surjective functions is surjective:
theorem surj_on_comp (hg : surj_on g B C) (hf: surj_on f A B) :

surj_on (g ◦ f) A C :=
assume z,
assume zc : z ∈ C,
exists.elim (hg zc) $
assume y (h1 : y ∈ B ∧ g y = z),
exists.elim (hf (and.left h1)) $
assume x (h2 : x ∈ A ∧ f x = y),
show ∃x, x ∈ A ∧ g (f x) = z, from
exists.intro x
(and.intro
(and.left h2)
(calc
g (f x) = g y : by rw and.right h2
... = z : by rw and.right h1))
The following shows that the image of a union is the union of images:
theorem image_union : f '' (A1 ∪ A2 ) =f '' A1 ∪ f '' A2 :=

ext (assume y, iff.intro
(assume h : y ∈ image f (A1 ∪ A2 ),
exists.elim h $

assume x h1,
have xA1 A2 : x ∈ A1 ∪ A2 , from h1.left,
have fxy : f x = y, from h1.right,
or.elim xA1 A2
(assume xA1 , or.inl (mem_image xA1 fxy))
(assume xA2 , or.inr (mem_image xA2 fxy)))
(assume h : y ∈ image f A1 ∪ image f A2 ,
or.elim h
(assume yifA1 : y ∈ image f A1 ,
exists.elim yifA1 $
assume x h1,
have xA1 : x ∈ A1 , from h1.left,
mem_image (or.inl xA1 ) fxy)
(assume yifA2 : y ∈ image f A2 ,
exists.elim yifA2 $
assume x h1,
have xA2 : x ∈ A2 , from h1.left,
mem_image (or.inr xA2 ) fxy)))
16.6 Exercises
1. Fill in the sorry’s in the last three proofs below.
open function int algebra
def f (x : Z) : Z := x + 3
def g (x : Z) : Z := -x
def h (x : Z) : Z := 2 * x + 3
example : injective f :=
assume x1 x2,
assume h1 : x1 + 3 = x2 + 3, -- Lean knows this is the same as f x1 = f x2
show x1 = x2, from eq_of_add_eq_add_right h1
example : surjective f :=
assume y,
have h1 : f (y - 3) = y, from calc
f (y - 3) = (y - 3) + 3 : rfl
... = y : by rw sub_add_cancel,
show ∃ x, f x = y, from exists.intro (y - 3) h1
example (x y : Z) (h : 2 * x = 2 * y) : x = y :=
have h1 : 2 ̸= (0 : Z), from dec_trivial, -- this tells Lean to figure it out itself
show x = y, from eq_of_mul_eq_mul_left h1 h
example (x : Z) : -(-x) = x := neg_neg x
example (A B : Type) (u : A → B) (v : B → A) (h : left_inverse u v) :

∀ x, u (v x) = x :=
h
example (A B : Type) (u : A → B) (v : B → A) (h : left_inverse u v) :

right_inverse v u :=
16.6. Exercises 145

-- fill in the sorry's in the following proofs
example : injective h :=
sorry
example : surjective g :=
sorry
example (A B : Type) (u : A → B) (v1 : B → A) (v2 : B → A)

(h1 : left_inverse v1 u) (h2 : right_inverse v2 u) : v1 = v2 :=
funext
(assume x,
calc
v1 x = v1 (u (v2 x)) : sorry
... = v2 x : sorry)
2. Fill in the sorry in the proof below.
import data.set
open function set
variable f : X → Y
variables A B : set X
example : f '' (A ∪ B) = f '' A ∪ f '' B :=

(assume y,
assume h1 : y ∈ f '' (A ∪ B),
exists.elim h1 $
assume x h,
have h2 : x ∈ A ∪ B, from h.left,
have h3 : f x = y, from h.right,
or.elim h2
(assume h4 : x ∈ A,
have h5 : y ∈ f '' A, from mem_image h4 h3,
show y ∈ f '' A ∪ f '' B, from or.inl h5)
(assume h4 : x ∈ B,
have h5 : y ∈ f '' B, from mem_image h4 h3,
show y ∈ f '' A ∪ f '' B, from or.inr h5))
(assume y,
assume h2 : y ∈ f '' A ∪ f '' B,
or.elim h2
(assume h3 : y ∈ f '' A,
exists.elim h3 $
assume x h,
have h4 : x ∈ A, from h.left,
have h6 : x ∈ A ∪ B, from or.inl h4,
show y ∈ f '' (A ∪ B), from mem_image h6 h5)
(assume h3 : y ∈ f '' B,
exists.elim h3 $
assume x h,
have h4 : x ∈ B, from h.left,
have h6 : x ∈ A ∪ B, from or.inr h4,

show y ∈ f '' (A ∪ B), from mem_image h6 h5))
-- remember, x ∈ A ∩ B is the same as x ∈ A ∧ x ∈ B

example (x : X) (h1 : x ∈ A) (h2 : x ∈ B) : x ∈ A ∩ B :=
and.intro h1 h2
example (x : X) (h1 : x ∈ A ∩ B) : x ∈ A :=
and.left h1
-- Fill in the proof below.

-- (It should take about 8 lines.)
example : f '' (A ∩ B) ⊆ f '' A ∩ f '' B :=

assume y,
assume h1 : y ∈ f '' (A ∩ B),
show y ∈ f '' A ∩ f '' B, from sorry
16.6. Exercises 147


CHAPTER
SEVENTEEN
THE NATURAL NUMBERS AND INDUCTION
This chapter marks a transition from the abstract to the concrete. Viewing the mathematical universe in
terms of sets, relations, and functions gives us useful ways of thinking about mathematical objects and
structures and the relationships between them. At some point, however, we need to start thinking about
particular mathematical objects and structures, and the natural numbers are a good place to start. The
nineteenth century mathematician Leopold Kronecker once proclaimed “God created the whole numbers;
everything else is the work of man.” By this he meant that the natural numbers (and the integers, which we
will also discuss below) are a fundamental component of the mathematical universe, and that many other
objects and structures of interest can be constructed from these.
In this chapter, we will consider the natural numbers and the basic principles that govern them. In Chapter
18 we will see that even basic operations like addition and multiplication can be defined using means described
here, and their properties derived from these basic principles. Our presentation in this chapter will remain
informal, however. In Chapter 19, we will see how these principles play out in number theory, one of the
oldest and most venerable branches of mathematics.
17.1 The Principle of Induction
The set of natural numbers is the set
N = {0, 1, 2, 3, . . .}.
In the past, opinions have differed as to whether the set of natural numbers should start with 0 or 1, but
these days most mathematicians take them to start with 0. Logicians often call the function s(n) = n + 1 the
successor function, since it maps each natural number, n, to the one that follows it. What makes the natural
numbers special is that they are generated by the number zero and the successor function, which is to say,
the only way to construct a natural number is to start with 0 and apply the successor function finitely many
times. From a foundational standpoint, we are in danger of running into a circularity here, because it is not
clear how we can explain what it means to apply a function “finitely many times” without talking about the
natural numbers themselves. But the following principle, known as the principle of induction, describes this
essential property of the natural numbers in a non-circular way.
Principle of Induction. Let P be any property of natural numbers. Suppose P holds of zero, and whenever
P holds of a natural number n, then it holds of its successor, n + 1. Then P holds of every natural number.
This reflects the image of the natural numbers as being generated by zero and the successor operation: by
covering the zero and successor cases, we take care of all the natural numbers.
The principle of induction provides a recipe for proving that every natural number has a certain property:
to show that P holds of every natural number, show that it holds of 0, and show that whenever it holds of
149
some number n, it holds of n + 1. This form of proof is called a proof by induction. The first required task is
called the base case, and the second required task is called the induction step. The induction step requires
temporarily fixing a natural number n, assuming that P holds of n, and then showing that P holds of n + 1.
In this context, the assumption that P holds of n is called the inductive hypothesis.
You can visualize proof by induction as a method of knocking down an infinite stream of dominoes, all at
once. We set the mechanism in place and knock down domino 0 (the base case), and every domino knocks
down the next domino (the induction step). So domino 0 knocks down domino 1; that knocks down domino
2, and so on.
Here is an example of a proof by induction.
Theorem. For every natural number n,
1 + 2 + . . . + 2n = 2n+1 − 1.
Proof. We prove this by induction on n. In the base case, when n = 0, we have 1 = 20+1 − 1, as required.
For the induction step, fix n, and assume
1 + 2 + . . . + 2n = 2n+1 − 1
(the induction hypothesis). We need to show that this same claim holds with n replaced by n + 1. But this
is just a calculation:
1 + 2 + . . . + 2n+1 = (1 + 2 + . . . + 2n ) + 2n+1
= 2n+1 − 1 + 2n+1
= 2 · 2n+1 − 1
= 2n+2 − 1.
In the notation of first-order logic, if we write P (n) to mean that P holds of n, we could express the principle
of induction as follows:
P (0) ∧ ∀n (P (n) → P (n + 1)) → ∀n P (n).
But notice that the principle of induction says that the axiom holds for every property P , which means that
we should properly use a universal quantifier for that, too:
∀P (P (0) ∧ ∀n (P (n) → P (n + 1)) → ∀n P (n)).
Quantifying over properties takes us out of the realm of first-order logic; induction is therefore a second-order
principle.
The pattern for a proof by induction is expressed even more naturally by the following natural deduction
rule:
1
P (n)
..
.
P (0) P (n + 1)
∀n P (n)
150 Chapter 17. The Natural Numbers and Induction

You should think about how some of the proofs in this chapter could be represented formally using natural
deduction.
For another example of a proof by induction, let us derive a formula that, given any finite set S, determines
the number of subsets of S. For example, there are four subsets of the two-element set {1, 2}, namely ∅, {1},
{2}, and {1, 2}. You should convince yourself that there are eight subsets of the set {1, 2, 3}. The following
theorem establishes the general pattern.
Theorem. For any finite set S, if S has n elements, then there are 2n subsets of S.
Proof. We use induction on n. In the base case, there is only one set with 0 elements, the empty set, and
there is exactly one subset of the empty set, as required.
In the inductive case, suppose S has n + 1 elements. Let a be any element of S, and let S ′ be the set
containing the remaining n elements. In order to count the subsets of S, we divide them into two groups.
First, we consider the subsets of S that don’t contain a. These are exactly the subsets of S ′ , and by the
inductive hypothesis, there are 2n of those.
Next we consider the subsets of S that do contain a. Each of these is obtained by choosing a subset of S ′
and adding a. Since there are 2n subsets of S ′ , there are 2n subsets of S that contain a.
Taken together, then, there are 2n + 2n = 2n+1 subsets of S, as required.
We have seen that there is a correspondence between properties of a domain and subsets of a domain. For
every property P of natural numbers, we can consider the set S of natural numbers with that property, and
for every set of natural numbers, we can consider the property of being in that set. For example, we can
talk about the property of being even, or talk about the set of even numbers. Under this correspondence,
the principle of induction can be cast as follows:
Principle of Induction. Let S be any set of natural numbers that contains 0 and is closed under the
successor operation. Then S = N.
Here, saying that S is “closed under the successor operation” means that whenever a number n is in S, so
is n + 1.
17.2 Variants of Induction
In this section, we will consider variations on the principle of induction that are often useful. It is important
to recognize that each of these can be justified using the principle of induction as stated in the last section,
so they need not be taken as fundamental.
The first one is no great shakes: instead of starting from 0, we can start from any natural number, m.
Principle of Induction from a Starting Point. Let P be any property of natural numbers, and let m
be any natural number. Suppose P holds of m, and whenever P holds of a natural number n greater than
or equal to m, then it holds of its successor, n + 1. Then P holds of every natural number greater than or
equal to m.
17.2. Variants of Induction 151

Assuming the hypotheses of this last principle, if we let P ′ (n) be the property “P holds of m + n,” we can
prove that P ′ holds of every n by the ordinary principle of induction. But this means that P holds of every
number greater than or equal to m.
Here is one example of a proof using this variant of induction.
Theorem. For every natural number n ≥ 5, 2n > n2 .

Proof. By induction on n. When n = 5, we have 2n = 32 > 25 = n2 , as required.
For the induction step, suppose n ≥ 5 and 2n > n2 . Since n is greater than or equal to 5, we have
2n + 1 ≤ 3n ≤ n2 , and so
(n + 1)2 = n2 + 2n + 1
≤ n2 + n2
< 2n + 2n
= 2n+1 .
For another example, let us derive a formula for the sum total of the angles in a convex polygon. A polygon
is said to be convex if every line between two vertices stays inside the polygon. We will accept without
proof the visually obvious fact that one can subdivide any convex polygon with more than three sides into a
triangle and a convex polygon with one fewer side, namely, by closing off any two consecutive sides to form
a triangle. We will also accept, without proof, the basic geometric fact that the sum of the angles of any
triangle is 180 degrees.
Theorem. For any n ≥ 3, the sum of the angles of any convex n-gon is 180(n − 2).
Proof. In the base case, when n = 3, this reduces to the statement that the sum of the angles in any triangle
is 180 degrees.
For the induction step, suppose n ≥ 3, and let P be a convex (n + 1)-gon. Divide P into a triangle and an
n-gon. By the inductive hypotheses, the sum of the angles of the n-gon is 180(n − 2) degrees, and the sum
of the angles of the triangle is 180 degrees. The measures of these angles taken together make up the sum
of the measures of the angles of P , for a total of 180(n − 2) + 180 = 180(n − 1) degrees.
For our second example, we will consider the principle of complete induction, also sometimes known as total
induction.
Principle of Complete Induction. Let P be any property that satisfies the following: for any natural
number n, whenever P holds of every number less than n, it also holds of n. Then P holds of every natural
number.
Notice that there is no need to break out a special case for zero: for any property P , P holds of all the
natural numbers less than zero, for the trivial reason that there aren’t any! So, in particular, any such
property automatically holds of zero.
Notice also that if such a property P holds of every number less than n, then it also holds of every number
less than n + 1 (why?). So, for such a P , the ordinary principle of induction implies that for every natural

number n, P holds of every natural number less than n. But this is just a roundabout way of saying that P
holds of every natural number. In other words, we have justified the principle of complete induction using
ordinary induction.
To use the principle of complete induction we merely have to let n be any natural number and show that
P holds of n, assuming that it holds of every smaller number. Compare this to the ordinary principle of
induction, which requires us to show P (n + 1) assuming only P (n). The following example of the use of this
principle is taken verbatim from the introduction to this book:
Theorem. Every natural number greater than or equal to 2 can be written as a product of primes.
Proof. We proceed by induction on n. Let n be any natural number greater than 2. If n is prime, we are
done; we can consider n itself as a product with one factor. Otherwise, n is composite, and we can write
n = m · k where m and k are smaller than n and greater than 1. By the inductive hypothesis, each of m and
k can be written as a product of primes, say
m = p1 · p2 · . . . · pu
and
k = q1 · q2 · . . . · qv .
But then we have
n = m · k = p1 · p2 · . . . · pu · q1 · q2 · . . . · qv ,
a product of primes, as required.
Finally, we will consider another formulation of induction, known as the least element principle.
The Least Element Principle. Suppose P is some property of natural numbers, and suppose P holds of
some n. Then there is a smallest value of n for which P holds.
In fact, using classical reasoning, this is equivalent to the principle of complete induction. To see this,
consider the contrapositive of the statement above: “if there is no smallest value for which P holds, then P
doesn’t hold of any natural number.” Let Q(n) be the property P does not hold of n. Saying that there is
no smallest value for which P holds means that, for every n, if P holds at n, then it holds of some number
smaller than n; and this is equivalent to saying that, for every n, if Q doesn’t hold at n, then there is a
smaller value for which Q doesn’t hold. And that is equivalent to saying that if Q holds for every number
less than n, it holds for n as well. Similarly, saying that P doesn’t hold of any natural number is equivalent
to saying that Q holds of every natural number. In other words, replacing the least element principle by
its contrapositive, and replacing P by “not Q,” we have the principle of complete induction. Since every
statement is equivalent to its contrapositive, and every predicate has its negated version, the two principles
are the same.
It is not surprising, then, that the least element principle can be used in much the same way as the principle
of complete induction. Here, for example, is a formulation of the previous proof in these terms. Notice that
it is phrased as a proof by contradiction.
Theorem. Every natural number greater than equal to 2 can be written as a product of primes.
17.2. Variants of Induction 153

Proof. Suppose, to the contrary, there some natural number greater than or equal to 2 cannot be written
as a product of primes. By the least element principle, there is a smallest such element; call it n. Then n is
not prime, and since it is greater than or equal to 2, it must be composite. Hence we can write n = m · k
where m and k are smaller than n and greater than 1. By the assumption on n, each of m and k can be
written as a product of primes, say
m = p1 · p2 · . . . · pu
and
k = q1 · q2 · . . . · qv .
But then we have
n = m · k = p1 · p2 · . . . · pu · q1 · q2 · . . . · qv ,
a product of primes, contradicting the fact that n cannot be written as a product of primes.
Theorem. Every natural number is interesting.

Proof. Suppose, to the contrary, some natural number is uninteresting. Then there is a smallest one, n. In
other words, n is the smallest uninteresting number. But that is really interesting! Contradiction.
17.3 Recursive Definitions
Suppose I tell you that I have a function f : N → N in mind, satisfying the following properties:
f (0) = 1
f (n + 1) = 2 · f (n)
What can you infer about f ? Try calculating a few values:
f (1) = f (0 + 1) = 2 · f (0) = 2
f (2) = f (1 + 1) = 2 · f (1) = 4
f (3) = f (2 + 1) = 2 · f (2) = 8
It soon becomes apparent that for every n, f (n) = 2n .

What is more interesting is that the two conditions above specify all the values of f , which is to say, there
is exactly one function meeting the specification above. In fact, it does not matter that f takes values in
the natural numbers; it could take values in any other domain. All that is needed is a value of f (0) and a
way to compute the value of f (n + 1) in terms of n and f (n). This is what the principle of definition by
recursion asserts:
Principle of Definition by Recursion. Let A be any set, and suppose a is in A, and g : N × A → A.

Then there is a unique function f satisfying the following two clauses:
f (0) = a
f (n + 1) = g(n, f (n)).

The principle of recursive definition makes two claims at once: first, that there is a function f satisfying the
clauses above, and, second, that any two functions f1 and f2 satisfying those clauses are equal, which is to
say, they have the same values for every input. In the example with which we began this section, A is just
N and g(n, f (n)) = 2 · f (n).
In some axiomatic frameworks, the principle of recursive definition can be justified using the principle of
induction. In others, the principle of induction can be viewed as a special case of the principle of recursive
definition. For now, we will simply take both to be fundamental properties of the natural numbers.
As another example of a recursive definition, consider the function g : N → N defined recursively by the
following clauses:
g(0) = 1
g(n + 1) = (n + 1) · g(n)
Try calculating the first few values. Unwrapping the definition, we see that g(n) = 1 · 2 · 3 · . . . · (n − 1) · n
for every n; indeed, definition by recursion is usually the proper way to make expressions using “…” precise.
The value g(n) is read “n factorial,” and written n!.
Indeed, summation notation
∑
f (i) = f (0) + f (1) + . . . + f (n − 1)
i<n
and product notation

∏
f (i) = f (0) · f (1) · · · · · f (n − 1)
i<n
∑
can also be made precise using recursive definitions. For example, the function k(n) = i<n f (i) can be
defined recursively as follows:
k(0) = 0
k(n + 1) = k(n) + f (n)
Induction and recursion are complementary principles, and typically the way to prove something about a
recursively defined function is to use the principle of induction. For example, the following theorem provides
a formulas for the sum 1 + 2 + . . . + n, in terms of n.
∑
Theorem. For every n, i<n+1 i = n(n + 1)/2.
Proof. In the base case, when n = 0, both sides are equal to 0.
In the inductive step, we have
( )
∑ ∑
i= i + (n + 1)
i<n+2 i<n+1
= n(n + 1)/2 + n + 1
n2 + n 2n + 2
= +
2 2
n2 + 3n + 2
=
2
(n + 1)(n + 2)
= .
2
17.3. Recursive Definitions 155

There are just as many variations on the principle of recursive definition as there are on the principle of
induction. For example, in analogy to the principle of complete induction, we can specify a value of f (n)
in terms of the values that f takes at all inputs smaller than n. When n ≥ 2, for example, the following
definition specifies the value of a function fib(n) in terms of its two predecessors:
fib(0) = 0
fib(1) = 1
fib(n + 2) = fib(n + 1) + fib(n).
Calculating the values of fib on 0, 1, 2, . . . we obtain
0, 1, 1, 2, 3, 5, 8, 13, 21, . . .
Here, after the second number, each successive number is the sum of the two values preceding it. This
is known as the Fibonacci sequence, and the corresponding numbers are known as the Fibonacci numbers.
An ordinary mathematical presentation would write Fn instead of fib(n) and specify the sequence with the
following equations:
F0 = 0, F1 = 1, Fn+2 = Fn+1 + Fn
But you can now recognize such a specification as an implicit appeal to the principle of definition by recursion.
We ask you to prove some facts about the Fibonacci sequence in the exercises below.
17.4 Defining Arithmetic Operations
In fact, we can even use the principle of recursive definition to define the most basic operations on the natural
numbers and show that they have the properties we expect them to have. From a foundational standpoint,
we can characterize the natural numbers as a set, N, with a distinguished element 0 and a function, succ(m),
which, for every natural number m, returns its successor. These satisfy the following:
• 0 ̸= succ(m) for any m in N.
• For every m and n in N, if m ̸= n, then succ(m) ̸= succ(n). In other words, succ is injective.
• If A is any subset of N with the property that 0 is in A and whenever n is in A then succ(n) is in A,
then A = N.
The last clause can be reformulated as the principle of induction:
Suppose P (n) is any property of natural numbers, such that P holds of 0, and for every n, P (n)
implies P (succ(n)). Then every P holds of every natural number.
Remember that this principle can be used to justify the principle of definition by recursion:
Let A be any set, a be any element of A, and let g(n, m) be any function from N × A to A. Then
there is a unique function f : N → A satisfying the following two clauses:
• f (0) = a
• f (succ(n)) = g(n, f (n)) for every n in N .
We can use the principle of recursive definition to define addition with the following two clauses:
m+0=m
m + succ(n) = succ(m + n)

Note that we are fixing m, and viewing this as a function of n. If we write 1 = succ(0), 2 = succ(1), and so
on, it is easy to prove n + 1 = succ(n) from the definition of addition.
We can proceed to define multiplication using the following two clauses:
m·0=0
m · succ(n) = m · n + m
We can also define a predecessor function by
pred(0) = 0
pred(succ(n)) = n,
and truncated subtraction by
m−̇0 = 0
m−̇(succ(n)) = pred(m−̇n).
With these definitions and the induction principle, one can prove all the following identities:
• n ̸= 0 implies succ(pred(n)) = n
• 0+n=n
• succ(m) + n = succ(m + n)
• (m + n) + k = m + (n + k)
• m+n=n+m
• m(n + k) = mn + mk
• 0·n=0
• 1·n=x
• (mn)k = m(nk)
• mn = nm
We will do the first five here, and leave the remaining ones as exercises.
Proposition. For every natural number n, if n ̸= 0 then succ(pred(n)) = n.

Proof. By induction on n. We have ruled out the case where n is 0, so we only need to show that the claim
holds for succ(n). But in that case, we have succ(pred(succ(n)) = succ(n) by the second defining clause of
the predecessor function.
Proposition. For every n, 0 + n = n.
Proof. By induction on n. We have 0 + 0 = 0 by the first defining clause for addition. And assuming
0 + n = n, we have 0 + succ(n) = succ(0 + n) = n, using the second defining clause for addition.
Proposition. For every m and n, succ(m) + n = succ(m + n).
Proof. Fix m and use induction on n. Then n = 0, we have succ(m) + 0 = succ(m) = succ(m + 0), using
the first defining clause for addition. Assuming the claim holds for n, we have
succ(m) + succ(n) = succ(succ(m) + n)

= succ(succ(m + n))
= succ(m + succ(n)),
17.4. Defining Arithmetic Operations 157

using the inductive hypothesis and the second defining clause for addition.
Proposition. For every m, n, and k, (m + n) + k = m + (n + k).
Proof. By induction on k. The case where k = 0 is easy, and in the induction step we have
(m + n) + succ(k) = succ((m + n) + k)
= succ(m + (n + k))
= m + succ(n + k)
= m + (n + succ(k)))
using the inductive hypothesis and the definition of addition.

Proposition. For every pair of natural numbers m and n, m + n = n + m.
Proof. By induction on n. The base case is easy using the second proposition above. In the inductive step,
we have
m + succ(n) = succ(m + n)
= succ(n + m)
= succ(n) + m
using the third proposition above.
17.5 Arithmetic on the Natural Numbers
Continuing as in the last section, we can establish all the basic properties of the natural numbers that play
a role in day-to-day mathematics. We summarize the main ones here:
m+n=n+m (commutativity of addition)

m + (n + k) = (m + n) + k (associativity of addition)
n + 0 = n (0 is a neutral element for addition)
n·m=m·n (commutativity of multiplication)
m · (n · k) = (m · n) · k (associativity of multiplication)
n · 1 = n (1 is an neutral element for multiplication)
n · (m + k) = n · m + n · k (distributivity)
n·0=0 (0 is an absorbing element for multiplication)
In an ordinary mathematical argument or calculation, they can be used without explicit justification. We
also have the following properties:
• n + 1 ̸= 0;
• if n + k = m + k then n = m;
• if n · k = m · k and k ̸= 0 then n = m.
We can define m ≤ n, “m is less than or equal to n,” to mean that there exists a k such that m + k = n. If
we do that, it is not hard to show that the less-than-or-equal-to relation satisfies all the following properties,
for every n, m, and k:
• n ≤ n (reflexivity);
• if n ≤ m and m ≤ k then n ≤ k (transitivity);

• if n ≤ m and m ≤ n then n = m (antisymmetry);

• for all n and m, either n ≤ m or m ≤ n is true (totality);
• if n ≤ m then n + k ≤ m + k;
• if n + k ≤ m + k then n ≤ m;
• if n ≤ m then nk ≤ mk;
• if m ≥ n then m = n or m ≥ n + 1;
• 0 ≤ n.
Remember from Chapter 13 that the first four items assert that ≤ is a linear order. Note that when we write
m ≥ n, we mean n ≤ m.
As usual, then, we can define m < n to mean that m ≤ n and m ̸= n. In that case, we have that m ≤ n
holds if and only if m < n or m = n.
Proposition. For every m, m + 1 ̸≤ 0.

Proof. Otherwise, we would have (m + 1) + k = (m + k) + 1 = 0 for some k.
In particular, taking m = 0, we have 1 ̸≤ 0.
Proposition. We have m < n iff and only if m + 1 ≤ n.

Proof. Suppose m < n. Then m ≤ n and m ̸= n. So there is a k such that m + k = n, and since m ̸= n,
we have k ̸= 0. Then k = u + 1 for some u, which means we have m + (u + 1) = m + 1 + u = n, so m ≤ n,
as required.
In the other direction, suppose m + 1 ≤ n. Then m ≤ n. We also have m ̸= n, since if m = n, we would
have m + 1 ≤ m + 0 and hence 1 ≤ 0, a contradiction.
In a similar way, we can show that m < n if and only if m ≤ n and m ̸= n. In fact, we can demonstrate all
of the following from these properties and the properties of ≤:
• n < n is never true (irreflexivity);
• if n < m and m < k then n < k (transitivity);
• for all n and m, either n < m, n = m or m < n is true (trichotomy);
• if n < m then n + k < m + k;
• if k > 0 and n < m then nk < mk;
• if m > n then m = n + 1 or m > n + 1;
• for all n, n = 0 or n > 0.
The first three items mean that < is a strict linear order, and the properties above means that ≤ is the
associated linear order, in the sense described in Section 13.1.
Proof. We will prove some of these properties using the previous characterization of the less-than relation.
17.5. Arithmetic on the Natural Numbers 159

The first property is straightforward: we know n ≤ n + 1, and if we had n + 1 ≤ n, we should have n = n + 1,

a contradiction.
For the second property, assume n < m and m < k. Then n + 1 ≤ m ≤ m + 1 ≤ k, which implies n < k.
For the third, we know that either n ≤ m or m ≤ n. If m = n, we are done, and otherwise we have either
n < m or m < n.
For the fourth, if n + 1 ≤ m, we have n + 1 + k = (n + k) + 1 ≤ m + k, as required.
For the fifth, suppose k > 0, which is to say, k ≥ 1. If n < m, then n + 1 ≤ m, and so nk + 1 ≤ nk + k ≤ mk.
But this implies nk < mk, as required.
The rest of the remaining proofs are left as an exercise to the reader.
Here are some additional properties of < and ≤:

• n < m and m < n cannot both hold (asymmetry);
• n + 1 > n;
• if n < m and m ≤ k then n < k;
• if n ≤ m and m < k then n < k;
• if m > n then m ≥ n + 1;
• if m ≥ n then m + 1 > n;
• if n + k < m + k then n < m;
• if nk < mk then k > 0 and n < m.
These can be proved from the ones above. Moreover, the collection of principles we have just seen can be
used to justify basic facts about the natural numbers, which are again typically taken for granted in informal
mathematical arguments.
Proposition. If n and m are natural numbers such that n + m = 0, then n = m = 0.

Proof. We first prove that m = 0. We know that m = 0 or m > 0. Suppose that m > 0. Then
n + m > n + 0 = n. Since n ≥ 0, we conclude that n + m > 0, which contradicts the fact that n + m = 0.
Since m > 0 leads to a contradiction, we must have m = 0.
Now we can easily conclude that n = 0, since n = n + 0 = n + m = 0. Hence n = m = 0.
Proposition. If n is a natural number such that n < 3, then n = 0, n = 1 or n = 2.
Proof. In this proof we repeatedly use the property that if m > n then m = n + 1 or m > n + 1. Since
2 + 1 = 3 > n, we conclude that either 2 + 1 = n + 1 or 2 + 1 > n + 1. In the first case we conclude n = 2,
and we are done. In the second case we conclude 2 > n, which implies that either 2 = n + 1, or 2 > n + 1.
In the first case, we conclude n = 1, and we are done. In the second case, we conclude 1 > n, and appeal
one last time to the general principle presented above to conclude that either 1 = n + 1 or 1 > n + 1. In the
first case, we conclude n = 0, and we are once again done. In the second case, we conclude that 0 > n. This
leads to a contradiction, since now 0 > n ≥ 0, hence 0 > 0, which contradicts the irreflexivity of >.

17.6 The Integers
The natural numbers are designed for counting discrete quantities, but they suffer an annoying drawback:
it is possible to subtract n from m if n is less than or equal to m, but not if m is greater than n. The set of
integers, Z, extends the natural numbers with negative values, to make it possible to carry out subtraction
in full:
Z = {. . . , −3, −2, −1, 0, 1, 2, 3, . . .}
We will see in a later chapter that the integers can be extended to the rational numbers, the real numbers, and
the complex numbers, each of which serves useful purposes. For dealing with discrete quantities, however,
the integers will get us pretty far.
You can think of the integers as consisting of two copies of the natural numbers, a positive one and a negative
one, sharing a common zero. Conversely, once we have the integers, you can think of the natural numbers
as consisting of the nonnegative integers, that is, the integers that are greater than or equal to 0. Most
mathematicians blur the distinction between the two, though we will see that in Lean, for example, the
natural numbers and the integers represent two different data types.
Most of the properties of the natural numbers that were enumerated in the last section hold of the integers
as well, but not all. For example, it is no longer the case that n + 1 ̸= 0 for every n, since the claim is false
for n = −1. For another example, it is not the case that every integer is either equal to 0 or greater than 0,
since this fails to hold of the negative integers.
The key property that the integers enjoy, which sets them apart from the natural numbers, is that for every
integer n there is a value −n with the property that n + (−n) = 0. The value −n is called the negation of
n. We define subtraction n − m to be n + (−m). For any integer n, we also define the absolute value of n,
written |n|, to be n if n ≥ 0, and −n otherwise.
We can no longer use proof by induction on the integers, because induction does not cover the negative
numbers. But we can use induction to show that a property holds of every nonnegative integer, for example.
Moreover, we know that every negative integer is the negation of a positive one.As a result, proofs involving
the integers often break down into two cases, where one case covers the nonnegative integers, and the other
case covers the negative ones.
17.7 Exercises
1. Write the principle of complete induction using the notation of symbolic logic. Also write the least
element principle this way, and use logical manipulations to show that the two are equivalent.
2. Show that for every n, 02 + 12 + 22 + . . . n2 = 16 n(n + 1)(n + 2).
3. Show that for every n, 03 + 13 + . . . + n3 = 14 n2 (n + 1)2 .
4. Given the definition of the Fibonacci numbers in Section 17.3, prove Cassini’s identity: for every n,
2
Fn+1 − Fn+2 Fn = (−1)n . Hint: in the induction step, write Fn+2
2
as Fn+2 (Fn+1 + Fn ).
∑
5. Prove i<n F2i+1 = F2n .
6. Prove the following two identities:
2
• F2n+1 = Fn+1 + Fn2
2
• F2n+2 = Fn+2 − Fn2
Hint: use induction on n, and prove them both at once. In the induction step, expand F2n+3 =
F2n+2 + F2n+1 , and similarly for F2n+4 . Proving the second equation is especially tricky. Use the
inductive hypothesis and the first identity to simplify the left-hand side, and repeatedly unfold the
17.6. The Integers 161

Fibonacci number with the highest index and simplify the equation you need to prove. (When you
have worked out a solution, write a clear equational proof, calculating in the ‘‘forwards’’ direction.)
7. Prove that every natural number can be written as a sum of distinct powers of 2. For this problem,
1 = 20 is counted as power of 2.
8. Let V be a non-empty set of integers such that the following two properties hold:
• if x, y ∈ V , then x − y ∈ V
• if x ∈ V , then every multiple of x is an element of V
Prove that there is some d ∈ V , such that V is equal to the set of multiples of d. Hint: use the least
element principle.
9. Give an informal but detailed proof that for every natural number n, 1 · n = n, using a proof by
induction, the definition of multiplication, and the theorems proved in Section 17.4.
10. Show that multiplication distributes over addition. In other words, prove that for natural numbers m,
n, and k, m(n + k) = mn + mk. You should use the definitions of addition and multiplication and
facts proved in Section 17.4 (but nothing more).
11. Prove the multiplication is associative, in the same way. You can use any of the facts proved in Section
17.4 and the previous exercise.
12. Prove that multiplication is commutative.
13. Prove (mn )k = mnk .
14. Following the example in Section 17.5, prove that if n is a natural number and n < 5, then n is one of
the values 0, 1, 2, 3, or 4.
15. Prove that if n and m are natural numbers and nm = 1, then n = m = 1, using only properties listed
in Section 17.5.
This is tricky. First show that n and m are greater than 0, and hence greater than or equal to 1. Then
show that if either one of them is greater than 1, then nm > 1.
16. Prove any of the other claims in Section 17.5 that were stated without proof.
17. Prove the following properties of negation and subtraction on the integers, using only the properties
of negation and subtraction given in Section 17.6.
• If n + m = 0 then m = −n;
• −0 = 0;
• If −n = −m then n = m;
• m + (n − m) = n;
• −(n + m) = −n − m;
• If m < n then n − m > 0;
• If m < n then −m > −n;
• n · (−m) = −nm;
• n(m − k) = nm − nk;
• If n < m then n − k < m − k.
18. Suppose you have an infinite chessboard with a natural number written in each square. The value in
each square is the average of the values of the four neighboring squares. Prove that all the values on
the chessboard are equal.

19. Prove that every natural number can be written as a sum of distinct non-consecutive Fibonacci numbers.
For example, 22 = 1 + 3 + 5 + 13 is not allowed, since 3 and 5 are consecutive Fibonacci numbers, but
22 = 1 + 21 is allowed.
17.7. Exercises 163


CHAPTER
EIGHTEEN
THE NATURAL NUMBERS AND INDUCTION IN LEAN
18.1 Induction and Recursion in Lean
Internally, in Lean, the natural numbers are defined as a type generated inductively from an axiomatically
declared zero and succ operation:
inductive nat : Type

| zero : nat
| succ : nat → nat
If you click the button that copies this text into the editor in the online version of this textbook, you will
see that we wrap it with the phrases namespace hide and end hide. This puts the definition into a new
“namespace,” so that the identifiers that are defined are hide.nat, hide.nat.zero and hide.nat.succ, to
avoid conflicting with the one that is in the Lean library. Below, we will do that in a number of places where
our examples duplicate objects defined in the library. The unicode symbol N, entered with \N or \nat, is a
synonym for nat.
Declaring nat as an inductively defined type means that we can define functions by recursion, and prove
theorems by induction. For example, these are the first two recursive definitions presented in the last chapter:
open nat
def two_pow : N → N
| 0 := 1
| (succ n) := 2 * two_pow n
def fact : N → N
| 0 := 1
| (succ n) := (succ n) * fact n
Addition and numerals are defined in such a way that Lean recognizes succ n and n + 1 as essentially the
same, so we could instead write these definitions as follows:
def two_pow : N → N
| 0 := 1
| (n + 1) := 2 * two_pow n
def fact : N → N
| 0 := 1
| (n + 1) := (n + 1) * fact n
If we wanted to define the function m^n, we would do that by fixing m, and writing doing the recursion on
the second argument:
165
def pow : N → N → N
| m 0 := 1
| m (n + 1) := m * pow m n
Lean is also smart enough to interpret more complicated forms of recursion, like this one:
def fib : N → N
| 0 := 0
| 1 := 1
| (n + 2) := fib (n + 1) + fib n
In addition to defining functions by recursion, we can prove theorems by induction. In Lean, each clause of
a recursive definition results in a new identity. For example, the two clauses in the definition of pow above
give rise to the following two theorems:
theorem pow_zero (n : N) : pow n 0 = 1 := rfl

theorem pow_succ (m n : N) : pow m (n+1) = m * pow m n := rfl
Notice that we could alternatively have used (pow m n) * m in the second clause of the definition of pow.
Of course, we can prove that the two definitions are equivalent using the commutativity of multiplication,
but, using a proof by induction, we can also prove it using only the associativity of multiplication, and
the properties 1 * m = m and m * 1 = m. This is useful, because the power function is also often used in
situations where multiplication is not commutative, such as with matrix multiplication. The theorem can
be proved in Lean as follows:
theorem pow_succ' (m n : N) : pow m (succ n) = (pow m n) * m :=

nat.rec_on n
(show pow m (succ 0) = pow m 0 * m, from calc
pow m (succ 0) = m * pow m 0 : by rw pow_succ
... = m * 1 : by rw pow_zero
... = m : by rw mul_one
... = 1 * m : by rw one_mul
... = pow m 0 * m : by rw pow_zero)
(assume n,
assume ih : pow m (succ n) = pow m n * m,
show pow m (succ (succ n)) = pow m (succ n) * m, from calc
pow m (succ (succ n)) = m * (pow m (succ n)) : by rw pow_succ
... = m * (pow m n * m) : by rw ih
... = (m * pow m n) * m : by rw mul_assoc
... = pow m (succ n) * m : by rw pow_succ)
This is a typical proof by induction in Lean. It begins with the phrase nat.induction_on n, and is followed
by the base case and the inductive hypothesis. The proof can be shortened using rewrite and simp:
theorem pow_succ' (m n : N) : pow m (succ n) = (pow m n) * m :=

nat.rec_on n
(show pow m (succ 0) = pow m 0 * m,
by rw [pow_succ, pow_zero, mul_one, one_mul])
(assume n,
assume ih : pow m (succ n) = pow m n * m,
show pow m (succ (succ n)) = pow m (succ n) * m,
by simp [pow_succ, ih])
Remember that you can write a rewrite proof incrementally, checking the error messages to make sure
things are working so far, and to see how far Lean got.
In any case, the power function is already defined in the Lean library as pow_nat. (It is defined generically
166 Chapter 18. The Natural Numbers and Induction in Lean

for any type that has a multiplication; the nat in pow_nat refers to the fact that the exponent is a natural
number.) The definition is essentially the one above, and the theorems above are also there:
import algebra.group_power
local infix ` ^ ` := pow_nat
#check @pow_nat
#check @pow_zero
#check @pow_succ
#check @pow_succ'
variables m n : N
#check m^n
As another example of a proof by induction, here is a proof of the identity m^(n + k) = m^n * m^k.
theorem pow_add (m n k : N) : m^(n + k) = m^n * m^k :=
nat.rec_on k
(show m^(n + 0) = m^n * m^0, from calc
m^(n + 0) = m^n : by rw add_zero
... = m^n * 1 : by rw mul_one
... = m^n * m^0 : by rw pow_zero)
(assume k,
assume ih : m^(n + k) = m^n * m^k,
show m^(n + succ k) = m^n * m^(succ k), from calc
m^(n + succ k) = m^(succ (n + k)) : by rw nat.add_succ
... = m^(n + k) * m : by rw pow_succ'
... = m^n * m^k * m : by rw ih
... = m^n * (m^k * m) : by rw mul_assoc
... = m^n * m^(succ k) : by rw pow_succ')
Notice the same pattern. This time, we do induction on k, and the base case and inductive step are routine.
Once again, with a bit of cleverness, we can shorten the proof with rewrite:
theorem pow_add (m n k : N) : m^(n + k) = m^n * m^k :=
nat.rec_on k
(show m^(n + 0) = m^n * m^0,
by rewrite [add_zero, pow_zero, mul_one])
(assume k,
assume ih : m^(n + k) = m^n * m^k,
show m^(n + succ k) = m^n * m^(succ k),
by rewrite [nat.add_succ, pow_succ', ih, mul_assoc, pow_succ'])
You should not hesitate to use calc, however, to make the proofs more explicit. Remember that you can
also use calc and rewrite together, using calc to structure the calculational proof, and using rewrite to
fill in each justification step.
18.2 Defining the Arithmetic Operations in Lean
In fact, addition and multiplication are defined in Lean essentially as described in Section 17.4. The defining
equations for addition hold by reflexivity, but they are also named add_zero and add_succ:
open nat
variables m n : N
18.2. Defining the Arithmetic Operations in Lean 167

example : m + 0 = m := add_zero m
example : m + succ n = succ (m + n) := add_succ m n
Similarly, we have the defining equations for the predecessor function and multiplication:
open nat
#check @pred_zero
#check @pred_succ
#check @mul_zero
#check @mul_succ
Here are the five propositions proved in Section 17.4.
theorem succ_pred (n : N) : n ̸= 0 → succ (pred n) = n :=

nat.rec_on n
(assume H : 0 ̸= 0,
show succ (pred 0) = 0, from absurd rfl H)
(assume n,
assume ih,
assume H : succ n ̸= 0,
show succ (pred (succ n)) = succ n,
by rewrite pred_succ)
theorem zero_add (n : nat) : 0 + n = n :=

nat.rec_on n
(show 0 + 0 = 0, from rfl)
(assume n,
assume ih : 0 + n = n,
show 0 + succ n = succ n, from
calc
0 + succ n = succ (0 + n) : rfl
... = succ n : by rw ih)
theorem succ_add (m n : nat) : succ m + n = succ (m + n) :=

nat.rec_on n
(show succ m + 0 = succ (m + 0), from rfl)
(assume n,
assume ih : succ m + n = succ (m + n),
show succ m + succ n = succ (m + succ n), from
calc
succ m + succ n = succ (succ m + n) : rfl
... = succ (succ (m + n)) : by rw ih
... = succ (m + succ n) : rfl)
theorem add_assoc (m n k : nat) : m + n + k = m + (n + k) :=

nat.rec_on k
(show m + n + 0 = m + (n + 0), by rw [add_zero, add_zero])
(assume k,
assume ih : m + n + k = m + (n + k),
show m + n + succ k = m + (n + (succ k)), from calc
m + n + succ k = succ (m + n + k) : by rw add_succ
... = succ (m + (n + k)) : by rw ih
... = m + (n + succ k) : by rw add_succ)
theorem add_comm (m n : nat) : m + n = n + m :=

nat.rec_on n

(show m + 0 = 0 + m, by rewrite [add_zero, zero_add])

(assume n,
assume ih : m + n = n + m,
show m + succ n = succ n + m, from calc
m + succ n = succ (m + n) : by rw add_succ
... = succ (n + m) : by rw ih
... = succ n + m : by rw succ_add)
18.3 Exercises
1. Consider the theorems in Section 17.4 and Section 17.5. Formalize as many of them as you can in
Lean.
18.3. Exercises 169


CHAPTER
NINETEEN
ELEMENTARY NUMBER THEORY
In the last two chapters, we saw that the natural numbers are characterized by the fact that they support
proof by induction and definition by recursion. Moreover, with these components, we can actually define +,
×, and < in a suitable axiomatic foundation, and prove that they have the relevant properties. In Section
17.6 we also discussed the integers, which include negative numbers and support the operation of subtraction.
The natural numbers and the integers are the central components of number theory, a branch of mathematics
dating back to the ancients. In this chapter, we will discuss some of the rudiments of this subject.
19.1 The Quotient-Remainder Theorem
A key property of the integers that we will use here is the quotient-remainder theorem:
Theorem. Let n and m be integers with m > 0. Then there are integers q and r satisfying n = mq + r and
0 ≤ r < m.
Proof. First we prove this in the case where n is a natural number, in which case use complete induction
on n. Let n be any natural number. If n < m, then we can take q = 0 and r = n, and we indeed have
n = mq + r and 0 ≤ r < m. Otherwise, we have n ≥ m. In this case n − m is a natural number smaller than
n. By induction hypothesis, we know that we can find q ′ and r′ such that n − m = mq ′ + r′ and 0 ≤ r′ < m.
Then we can choose q = q ′ + 1 and r = r′ , and we obtain n = mq + r and 0 ≤ r < m, as desired.
If n is negative, then −(n + 1) is a natural number, hence we can use the previous part for −(n + 1) to obtain
q ′ and r′ such that −(n + 1) = mq ′ + r′ and 0 ≤ r′ < m. Now let q = −(q ′ + 1) and r = m − r′ − 1. Then
we can compute
mq + r = −m(q ′ + 1) + m − r′ − 1
= −(mq ′ + r′ ) − m + m − 1
= −(−(n + 1)) − 1
=n+1−1
= n.
Also, since r′ ≥ 0 we have r < m and since r′ < m we have r ≥ 0. This completes the proof.
Intuitively, q is the integer quotient when you divide n by m and r is the remainder. Remember that using
the word “the” presupposes that there are unique values meeting that description. That is, in fact, the case:
171
Proposition. If n and m are as above, n = mq + r and n = mq ′ + r′ with both r and r′ less than m, then
q = q ′ and r = r′ .
Proof. By assumption, we have mq + r = mq ′ + r′ . It suffices to show that q = q ′ , because then mq = mq ′ ,
and hence r = r′ .
Suppose q ̸= q ′ . Then either q < q ′ or q ′ < q. Suppose without loss of generality that q < q ′ . (The other case
is symmetric.) Then mq < mq ′ , so we can subtract mq from both sides of the equality mq + r = mq ′ + r′ to
obtain
r = mq ′ + r′ − mq = m(q − q ′ ) + r′ .
But since q ′ < q, we have q − q ′ ≥ 1, which means
m(q − q ′ ) + r′ ≥ m + r′ ≥ m,
which contradicts the fact that r < m.
19.2 Divisibility
We can define divisibility on the integers as follows.
Definition. Given two integers m and n, we say that m is a divisor of n, written m | n, if there exists some
integer k such that m · k = n. We also say that n is divisible by m or that m divides n. We write m ∤ n to
say that m is not a divisor of n.
We can now prove the following:
Theorem. The relation | is reflexive and transitive. Also, if n | m and m | n, then m = ±n. This means
that restricted to the natural numbers, this relation is a partial order.
Proof. Reflexivity is immediate, because n · 1 = n, hence n | n.
For transitivity, suppose m | n and n | r. Then there are k, ℓ such that m · k = n and n · ℓ = r. Now we
compute
m · (k · ℓ) = (m · k) · ℓ
=n·ℓ
= r.
Suppose that n and m are integers such that n | m and m | n. Then there exist k and ℓ such that n · k = m
and m · ℓ = n. We distinguish two cases. If n = 0, then we have m = n · k = 0 = n, so we are done. If n ̸= 0,
then we use the the equations to get n · k · ℓ = m · ℓ = n, and we can cancel n on both sides to get k · ℓ = 1.
We conclude that k = ℓ = ±1, hence we get m = n · k = ±n.
Note that this means that if n and m are both natural numbers, then n = m, which means that | is
antisymmetric, and hence a partial order, on the natural numbers.
See Exercise 1 for some basic properties of divisibility.
172 Chapter 19. Elementary Number Theory

An integer is even if it is divisible by 2, in other words, n is even if 2 | n. An integer is odd if it is not even.
Of course, odd numbers are of the form 2k + 1 for some k, and we can prove this now.
Theorem. If n is an odd integer, then n = 2k + 1 for some integer k.

Proof. By the quotient-remainder theorem, we can write n = 2k+r for some integers k and r with 0 ≤ r < 2.
The last condition means that r = 0 or r = 1. In the first case, we have n = 2k, hence 2 | n, contradicting
that n is odd. So we have r = 1, which means that n = 2k + 1.
Theorem. Every sequence of k consecutive numbers contains a number divisible by k.
Proof. Denote the largest element of the sequence by n. This means that the sequence is n − (k − 1), . . . , n −
1, n. By the quotient-remainder theorem, we have n = qk + r for some integers q and r with 0 ≤ r < k.
From these inequalities we conclude that n − r is in our sequence, and n − r = qk, hence divisible by k.
Definition. Given two integers m and n such that either m ̸= 0 or n ̸= 0, we define the greatest common
divisor gcd(m, n) of m and n to be the largest integer d which is both a divisor of m and n, that is d | m
and d | n.
This largest integer exists, because there is at least one common divisor, but only finitely many. There is
at least one, since 1 is a common divisor of any two integers, and there are finitely many, since a nonzero
number has only finitely many divisors.
If n = m = 0, then we define gcd(0, 0) = 0.
The greatest common divisor of two numbers is always a natural number, since 1 is always a common divisor
of two numbers. As an example, let us compute the greatest common divisor of 6 and 28. The positive
divisors of 6 are {1, 2, 3, 6} and the positive divisors of 28 are {1, 2, 4, 7, 14, 28}. The largest number in both
these sets is 2, which is the greatest common divisor of 6 and 28.
However, computing the greatest common divisor of two numbers by listing all the divisors of both numbers
is a lot of work, so we will now consider a method to compute the greatest common divisor more efficiently.
Lemma. For all integers n, m and k we have gcd(n, m) = gcd(m, n − km).

Proof. Let d = gcd(n, m) and r = n − km. If n = m = 0, then d = 0 = gcd(m, r), and we’re done.
In the other case we first show that the set of common divisors of n and m is the same as the set of the
common divisors of m and r. To see this, let d′ | m and d′ | n. Then also d′ | n − km by Exercise 1 below.
Hence d′ is a common divisor of m and r. On the other hand, if d′ is a divisor of m and r, then d′ | r + km,
hence d′ | n, hence d′ is a common divisor of n and m.
Since the sets of common divisors are the same, the largest element in each set is also the same, hence
gcd(n, m) = gcd(m, n − km).
Lemma. For all integers n we have gcd(n, 0) = |n|.
Proof. Every number is a divisor of 0, hence the greatest common divisor of n and 0 is just the greatest
divisor of n, which is the absolute value of n.
These two lemmas give us a quick way to compute the greatest common divisor of two numbers. This is
called the Euclidean Algorithm. Suppose we want to compute gcd(a, b).
• We let r0 = a and r1 = b.
19.2. Divisibility 173

• Given rn and rn+1 we compute rn+2 as the remainder of of rn when divided by rn+1 .
• Once rn = 0, we stop, and gcd(a, b) = |rn−1 |.
This works, because by the lemmas above, we have gcd(rk , rk+1 ) = gcd(rk+1 , rk+2 ), since rk+2 = rk − qrk+1
for some q. Hence if rn = 0 we have
gcd(a, b) = gcd(r0 , r1 ) = gcd(rn−1 , rn ) = gcd(rn−1 , 0) = |rn−1 |.
For example, suppose we want to compute the greatest common divisor of 1311 and 5757. We compute the
following remainders:
5757 = 4 × 1311 + 513

1311 = 2 × 513 + 285
513 = 1 × 285 + 228
285 = 1 × 228 + 57
228 = 4 × 57 + 0.
Hence gcd(1311, 5757) = 57. This is much quicker than computing all the divisors of both 1311 and 5757.
Here is an important result about greatest common divisors. It is only called a “lemma” for historical reasons.
Theorem (Bézout’s Lemma). Let s and t be integers. Then there are integers a and b such that as + bt =
gcd(s, t).
Proof. We compute gcd(s, t) by the Euclidean Algorithm given above, and during the algorithm we get
the intermediate values r0 , r1 , . . . , rn where rn = 0. Now by induction on k we prove that we can write
rk = ak s + bk t for some integers ak and bk . Indeed: r0 = 1 · s + 0 · t and r1 = 0 · s + 1 · t. Now if we assume
that rk = ak s + bk t and rk+1 = ak+1 s + bk+1 t, we know that rk+2 = rk − q · rk+1 , where q is the quotient of
rk when divided by rk+1 . These equations together give
rk+2 = (ak − qak+1 )s + (bk − qbk+1 )t
This completes the induction. In particular, rn−1 = an−1 s + bn−1 t, and since gcd(s, t) = ±rn−1 we can write
gcd(s, t) as as + bt for some a and b.
Corollary. If c is any common divisor of n and m, then c | gcd(n, m).

Proof. By Bézout’s Lemma, there are a and b such that gcd(n, m) = an + bm. Since c divides both n and
m, c divides an + bm by Exercise 1 below, and hence also gcd(n, m).
Of special interest are pairs of integers which have no divisors in common, except 1 and −1.
Definition. Two integers n and m are coprime if gcd(n, m) = 1.
Proposition. Let n, m and k be integers such that n and k are coprime. If k | nm then k | m.
Proof. By Bézout’s Lemma, there are a and b such that an+bk = 1. Multiplying by m gives anm+bkm = m
Since k divides nm, k divides the left-hand side of the equation, hence k | m.

19.3 Prime Numbers
In this section we consider properties of prime numbers.
Definition. An integer p ≥ 2 is called prime if the only positive divisors of p are 1 and p. An integer n ≥ 2
which is not prime is called composite.
An equivalent definition of a prime number is a positive number with exactly 2 positive divisors.
Recall from Chapter 17 that every natural number greater than 1 can be written as the product of primes.
In particular, ever natural number greater than 1 is divisible by some prime number.
We now prove some other properties about prime numbers.
Theorem. There are infinitely many primes.

Proof. Suppose for the sake of contradiction that there are only finitely many primes p1 , p2 , . . . , pk . Let
n = p1 × p2 × · · · × pk . Since n is divisible by pi for all i ≤ k we know that n + 1 is not divisible by pi for
any i. However, we assumed that these are all primes, contradicting the fact that every number is divisible
by a prime number.
Lemma. If n is an integer and p is a prime number, then either n and p are coprime or p | n.
Proof. Let d = gcd(n, p). Since d is a positive divisor of p, either d = 1 or d = p. In the first case, n and p
are coprime by definition, and in the second case we have p | n.
Proposition. If n and m are integers and p is a prime number such that p | nm then either p | n or p | m.
Proof. Suppose that p ∤ n. By the previous lemma, this means that p and n are coprime. From this we can
conclude that p | m.
The last result in this section captures that the primes are the “building blocks” of the positive integers for
multiplication: all other integers can be written as a product of primes in an essentially unique way.
Theorem (Fundamental Theorem of Arithmetic). Let n > 0 be an integer. Then there are primes p1 , . . . , pk
such that n = p1 ×· · ·×pk . Moreover, these primes are unique up to reordering. That means that if there are
prime numbers q1 , . . . , qℓ such that q1 × · · · × qℓ = n, then the qi are a reordering of the pi . To be completely
precise, this means that there is a bijection σ : {1, . . . , k} → {1, . . . , k} such that qi = pσ(i) .
Remark. 1 can be written as the product of zero prime numbers. The empty product is defined to be 1.
Proof. We have already seen that every number can be written as the product of primes, so we only need
to prove the uniqueness up to reordering. Suppose this is not true, and by the least element principle,
let n be the smallest positive integers such that n can be written as the product of primes in two ways:
n = p1 × · · · × pk = q1 × · · · × qℓ .
Since 1 can be written as product of primes only as empty product, we have n > 1, hence k ≥ 1. Since pk is
prime, we must have pk | qj for some j ≤ ℓ. By swapping qj and qℓ , we may assume that j = ℓ. Since qℓ is
also prime, we have pk = qℓ .
Now we have p1 × · · · × pk−1 = q1 × · · · × qℓ−1 . This product is smaller than n, but can be written as product
of primes in two different ways. But we assumed n was the smallest such number. Contradiction!
19.3. Prime Numbers 175

19.4 Modular Arithmetic
In the discussion of equivalence relations in Section 13.3 we considered the example of the relation of modular
equivalence on the integers. This is sometimes thought of as “clock arithmetic.” Suppose you have a 12-hour
clock without a minute hand, so it only has an hour hand which can point to the hours 12, 1, 2, 3, 4, 5, 6,
7, 8, 9, 10, 11 and then it wraps to 12 again. We can do arithmetic with this clock.
• If the hand currently points to 10, then 5 hours later it will point to 3.
• If the hand points to 7, then 23 hours before that, it pointed to 8.
• If the hand points to 9, and we work for a 8 hours, then when we are done the hand will point to 5. If
we worked twice as long, starting at 9, the hand will point to 1.
We want to write these statements using mathematical notation, so that we can reason about them more
easily. We cannot write 10 + 5 = 3 for the first expression, because that would be false, so instead we use the
notation 10 + 5 ≡ 3 (mod 12). The notation (mod 12) indicates that we forget about multiples of 12, and
we use the “congruence” symbol with three horizontal lines to remind us that these values are not exactly
equal, but only equal up to multiples of 12. The other two lines can be formulated as 7 − 23 ≡ 8 (mod 12)
and 9 + 2 · 8 ≡ 1 (mod 12).
Here are some more examples:
• 6 + 7 ≡ 1 (mod 12)
• 6 · 7 ≡ 42 ≡ 6 (mod 12)
• 7 · 5 ≡ 35 ≡ −1 (mod 12)
The last example shows that we can use negative numbers as well.
We now give a precise definition.
Definition. For integers a, b and n we say that a and b are congruent modulo n if n | a − b. This is written
a ≡ b (mod n). The number n is called the modulus.
Typically we only use this definition when the modulus n is positive.
Theorem. Congruence modulo n is an equivalence relation.

Proof. We have to show that congruence modulo n is reflexive, symmetric and transitive.
It is reflexive, because a − a = 0, so n | a − a, and hence a ≡ a (mod n).
To show that it is symmetric, suppose that a ≡ b (mod n). Then by definition, n | a − b. So n | (−1) · (a − b),
which means that n | b − a. This means by definition that b ≡ a (mod n).
To show that it is transitive, suppose that a ≡ b (mod n) and b ≡ c (mod n). Then we have n | a − b and
n | b − c. Hence we have n | (a − b) + (b − c) which means that n | a − c. So a ≡ c (mod n).
This theorem justifies the “chaining” notation we used above when we wrote 7 · 5 ≡ 35 ≡ −1 (mod 12).
Since congruence modulo 12 is transitive, we can now actually conclude that 7 · 5 ≡ −1 (mod 12).
Theorem. Suppose that a ≡ b (mod n) and c ≡ d (mod n). Then a + c ≡ b + d (mod n) and a · c ≡ b · d
(mod n).

Moreover, if a ≡ b (mod n) then ak ≡ bk (mod n) for all natural numbers k.

Proof. We know that n | a − b and n | c − d. For the first statement, we can calculate that (a + c) − (b + d) =
(a − b) + (c − d), so we can conclude that n | (a + c) − (b + d) hence that a + c ≡ b + d (mod n).
For the second statement, we want to show that n | a · c − b · d. We can factor a · c − b · d = (a − b) · c + b · (c − d).
Now n divides both summands on the right, hence n divides a · c − b · d, which means that a · c ≡ b · d (mod n).
The last statement follows by induction on k. If k = 0, then 1 ≡ 1 (mod n), and for the induction step,
suppose that ak ≡ bk (mod n), then we have ak+1 = a · ak ≡ b · bk = bk+1 (mod n)
This theorem is useful for carrying out computations modulo n. Here are some examples.
• Suppose we want to compute 77 · 123 modulo 12. We know that 77 ≡ 5 (mod 12) and 123 ≡ 3
(mod 12), so 77 · 123 ≡ 5 · 3 ≡ 15 ≡ 3 (mod 12)
• Suppose we want to compute 99 · 998 modulo 10. We know that 99 ≡ −1 (mod 10) and 998 ≡ −2
(mod 10), hence 99 · 998 ≡ (−1) · (−2) ≡ 2 (mod 10).
• Suppose we want to know the last digit of 101101 . Notice that the last digit of a number n is congruent
to n modulo 10, so we can just compute 101101 ≡ 1101 ≡ 1 (mod 10). So the last digit of 101101 is 1.
Warning. You cannot do all computations you might expect with modular arithmetic:
• You are not allowed to divide congruent numbers in modular arithmetic. For example 12 ≡ 16 (mod 4),
but we are not allowed to divide both sides of the equation by 2, because 6 ̸≡ 8 (mod 4).
• You are not allowed to compute in exponents with modular arithmetic. For example 8 ≡ 3 (mod 5),
but 28 ̸≡ 23 (mod 5). To see this: 28 = 256 ≡ 1 (mod 5), but 23 = 8 ≡ 3 (mod 5).
Recall the quotient-remainder theorem: if n > 0, then any integer a can be expressed as a = nq + r, where
0 ≤ r < n. In the language of modular arithmetic this means that a ≡ r (mod n). So if n > 0, then every
integer is congruent to a number between 0 and n − 1 (inclusive). So there “are only n different numbers”
when working modulo n. This can be used to prove many statements about the natural numbers.
Proposition. For every integer k, k 2 + 1 is not divisible by 3.

Proof. Translating this problem to modular arithmetic, we have to show that k 2 + 1 ̸≡ 0 (mod 3) or in
other words that k 2 ̸≡ 2 (mod 3) for all k. By the quotient-remainder theorem, we know that k is either
congruent to 0, 1 or 2, modulo 3. In the first case, k 2 ≡ 02 ≡ 0 (mod 3). In the second case, k 2 ≡ 12 ≡ 1
(mod 3), and in the last case we have k 2 ≡ 22 ≡ 4 ≡ 1 (mod 3). In all of those cases, k 2 ̸≡ 2 (mod 3). So
k 2 + 1 is never divisible by 3.
Proposition. For all integers a and b, a2 + b2 − 3 is not divisible by 4.

Proof. We first compute the squares modulo 4. We compute
02 ≡ 0 (mod 4)
12 ≡ 1 (mod 4)
22 ≡ 0 (mod 4)
32 ≡ 1 (mod 4)
Since every number is congruent to 0, 1, 2 or 3 modulo 4, we know that every square is congruent to 0 or 1
modulo 4. This means that there are only four possibilities for a2 + b2 (mod 4). It can be congruent to 0 + 0,
1 + 0, 0 + 1 or 0 + 0. In all those cases, a2 + b2 ̸≡ 3 (mod 4) Hence 4 ∤ a2 + b2 − 3, proving the proposition.
19.4. Modular Arithmetic 177

Recall that we warned you about dividing in modular arithmetic. This doesn’t always work, but often it
does. For example, suppose we want to solve 2n ≡ 1 (mod 5). We cannot solve this by saying that n ≡ 12
(mod 5), because we cannot work with fractions in modular arithmetic. However, we can still solve it by
multiplying both sides with 3. Then we get 6n ≡ 3 (mod 5), and since 6 ≡ 1 (mod 5) we get n ≡ 3 (mod 5).
So instead of dividing by 2 we could multiply by 3 to get the answer. The reason this worked is because
2 × 3 ≡ 1 (mod 5).
Definition. Let n and a be integers. A multiplicative inverse of a modulo n is an integer b such that ab ≡ 1
(mod n).
For example, 3 is a multiplicative inverse of 5 modulo 7, since 3 × 5 ≡ 1 (mod 7). But 2 has no multiplicative
inverse modulo 6. Indeed, suppose that 2b ≡ 1 (mod 6), then 6 | 2b − 1. However, 2b − 1 is odd, and cannot
be divisible by an even number. We can use multiplicative inverses to solve equations. If we want to solve
ax ≡ c (mod n) for x and we know that b is a multiplicative inverse of a, the solution is x ≡ bc (mod n)
which we can see by multiplying both sides by b.
Lemma Let n and a be integers. a has at most one multiplicative inverse modulo n. That is, if b and b′ are
both multiplicative inverses of a modulo n, then b ≡ b′ (mod n).
Proof. Suppose that ab ≡ 1 ≡ ab′ (mod n). Then we can compute bab′ in two ways: b ≡ b(ab′ ) = (ba)b′ ≡ b′
(mod n).
Proposition. Let n and a be integers. a has a multiplicative inverse modulo n if and only if n and a are
coprime.
Proof. Suppose b is a multiplicative inverse of a modulo n. Then n | ab − 1. Let d = gcd(a, b). Since d | n
we have d | ab − 1. But since d is a divisor of ab, we have d | ab − (ab − 1) = 1. Since d ≥ 0 we have d = 1.
Hence n and a are coprime.
On the other hand, suppose that n and a are coprime. By Bézout’s Lemma we know that there are integers
b and c such that cn + ba = gcd(n, a) = 1. We can rewrite this to ab − 1 = (−c)n, hence n | ab − 1, which
means by definition ab ≡ 1 (mod n). This means that b is a multiplicative inverse of a modulo n.
Note that if p is a prime number and a is a integer not divisible by p, then a and p are coprime, hence a has
a multiplicative inverse.
19.5 Properties of Squares
Mathematicians from ancient times have been interested in the question as to which integers can be written
as a sum of two squares. For example, we can write 2 = 11 + 11 , 5 = 22 + 12 , 13 = 32 + 22 . If we make a
sufficiently long list of these, an interesting pattern emerges: if two numbers can be written as a sum of two
squares, then so can their product. For example, 10 = 5 · 2, and we can write 10 = 32 + 12 . Or 65 = 13 · 5,
and we can write 65 = 82 + 12 .
At first, one might wonder whether this is just a coincidence. The following provides a proof of the fact that
it is not.

Theorem. Let x and y be any two integers. If x and y are both sums of two squares, then so is xy.
Proof. Suppose x = a2 + b2 , and suppose y = c2 + d2 . I claim that
xy = (ac − bd)2 + (ad + bc)2 .
To show this, notice that on the one hand we have
xy = (a2 + b2 )(c2 + d2 ) = a2 c2 + a2 d2 + b2 c2 + b2 d2 .
On the other hand, we have
(ac − bd)2 + (ad + bc)2 = (a2 c2 − 2abcd + b2 d2 ) + (a2 d2 + 2abcd + b2 c2 )

= a2 c2 + b2 d2 + a2 d2 + b2 c2 .
Up to the order of summands, the two right-hand sides are the same.
√
We will now prove that 2 is not a fraction of two integers.
√
Theorem. There are no integers a and b such that ab = 2.
√
Proof. Suppose that ab = 2 for some integers a and b. By canceling common factors, we may assume that
2
a and b are coprime. By squaring both sides, we get ab2 = 2, and multiplying both sides by b2 gives a2 = 2b2 .
Since 2b2 is even, we know that a2 is even, and since odd squares are odd, we conclude that a is even. Hence
we can write a = 2c for some integer c. This means that (2c)2 = 2b2 , hence 2c2 = b2 . The same reasoning
shows that b is even. But we assumed that a and b are coprime, which contradicts the fact that they are
both even.
√
Hence there are no integers a and b such that ab = 2.
19.6 Exercises
1. Prove the following properties about divisibility (for any integers a, b and c):
• if a | b and a | c then a | b + c and a | b − c;
• if a | b then a | bc;
• a | 0;
• if 0 | a then a = 0;
• if a ̸= 0 then the statements b | c and ab | ac are equivalent;
• if a | b and b ̸= 0 then |a| ≤ |b|.
2. Prove that for any integer n, n2 leaves a remainder of 0 or 1 when you divide it by 4. Conclude that
n2 + 2 is never divisible by 4.
3. Prove that if n is odd, n2 − 1 is divisible by 8.
4. Prove that if m and n are odd, then m2 + n2 is even but not divisible by 4.
5. Say that two integers “have the same parity” if they are both even or both odd. Prove that if m and
n are any two integers, then m + n and m − n have the same parity.
19.6. Exercises 179

6. Write 11160 as product of primes.

7. List all the divisors of 42 and 198, and find the greatest common divisor by looking at the largest
number in both lists. Also compute the greatest common divisor of the numbers by the Euclidean
Algorithm.
8. Compute gcd(15, 55), gcd(12345, 54321) and gcd(−77, 110)
9. Show by induction on n that for every pair of integers x and y, x − y divides xn − y n . (Hint: in the
induction step, write xn+1 − y n+1 as xn (x − y) + xn y − y n+1 .)
10. Compute 212 (mod 13). Use this to compute 21212004 (mod 13).
11. Find the last digit of 9999 . Can you also find the last two digits of this number?
12. Prove that 5022 − 2250 is divisible by 7.
13. Check whether the following multiplicative inverses exist, and if so, find them.
• The multiplicative inverse of 5 modulo 7;
• The multiplicative inverse of −2 modulo 9.
14. Find all integers x such that 75x ≡ 45 (mod 8).
15. Show that for every integer n the number n4 is congruent to 0 or 1 modulo 5. Hint: to simplify the
computation, use that 44 ≡ (−1)4 (mod 5).
16. Prove that the equation n4 + m4 = k 4 + 3 has no solutions in the integers. (Hint: use the previous
exercise.)
17. Suppose p is a prime number such that p ∤ k. Show that if kn ≡ km (mod p) then n ≡ m (mod p).
18. Let n, m and c be given integers. Use Bézout’s Lemma to prove that the equation an + bm = c has a
solution for integers a and b if and only if gcd(n, m) | c.
19. Suppose that a | n and a | n and let d = gcd(n, m). Prove that gcd( na , m d
a ) = a . Conclude that for any
n m
two integers n and m with greatest common divisor d the numbers d and d are coprime.

CHAPTER
TWENTY
COMBINATORICS
Combinatorics is the art of counting without counting. It is a fundamental mathematical task to determine
how many things there are in a given collection, and when the collection is large, it can be tedious or
infeasible to count the elements individually. Moreover, when the collection is described in terms of a
changing parameter (say, a natural number, n), we would like a formula that tells us how the number of
objects depends on that parameter. In this chapter we will set up a foundation for achieving this goal, and
learn some of the tricks of the trade.
20.1 Finite Sets and Cardinality
It will be helpful, for every natural number n, to have a canonical set of elements of size n. To that end, we
will choose the set
[n] = {m | m < n} = {0, 1, . . . , n − 1}.
We used the same notation, [n], to describe equivalence classes with respect to an equivalence relation, but
hopefully our intended meaning will always be clear from the context.
A set A of elements is said to be finite if there is a bijection from [n] to A for some n. In that case, we would
like to say that A has n elements, or that the set A has cardinality n, and write |A| = n. But to do so, we
need to know that when A is finite, there is a unique n with the property above.
Suppose there are bijections from both [m] and [n] to A. Composing the first bijection with the inverse of
the second, we get a bijection from [m] to [n]. It seems intuitively clear that this implies m = n, but our
goal is to prove this from the fundamental properties of sets, functions, and the natural numbers.
So suppose, for the sake of contradiction, m ̸= n. Without loss of generality, we can assume m > n (why?).
In particular, there is an injective function f from [m] to [n]. Since m > n, m ≥ n + 1, and so we can restrict
f to get an injective function from [n + 1] to [n]. The next theorem shows that this cannot happen.
Theorem. For any natural number n, there is no injective function from [n + 1] to [n].
Proof. By induction on n. The theorem is clear when n = 0, because [1] = {0} and [0] = ∅. If f were an
injective function from [1] to [0], we would have f (0) ∈ ∅, which is impossible.
So suppose the claim is true for n, and suppose f is an injective function from [n + 2] to [n + 1]. We consider
two cases.
In the first case, suppose n is not in the image of f . Then f maps [n + 2] to [n], and restricting the domain,
we have an injective function from [n + 1] to [n], contradicting the inductive hypothesis.
In the second case, there is some m < n + 2 such that f (m) = n. The idea is to alter f slightly to get an
injective function from [n + 1] to [n], again contradicting the inductive hypothesis. If m = n + 1, which is
181
to say it is the last element of [n + 2] that is mapped to the last element of [n + 1], we can just restrict f to
[n + 1]. The fact that f was injective implies that all the elements in [n + 1] are mapped to n.
Otherwise, define f ′ : [n + 1] → [n] by
{
′ f (i) if i ̸= m
f (i) =
f (n + 1) if i = m.
In other words, we map m to the value that n + 1 was mapped to. Since f is injective, f (n + 1) ̸= f (m),
and so f (n + 1) < n, as required. It is not hard to check that f ′ is injective, so we have the contradiction
we were after.
This theorem is known as the “pigeonhole principle.” It implies that if n + 1 pigeons inhabit n holes, then at
least one hole has more than one pigeon. The principle implies that for every finite set A, there is a unique
n such that there is a bijection from [n] to A, and we can define the cardinality of A to be that n.
20.2 Counting Principles
Here is a basic counting principle.
Theorem. Let A and B be disjoint finite sets. Then |A ∪ B| = |A| + |B|.

Proof. Suppose f : [m] → A and g : [n] → B are bijections. Define h : [m + n] → A ∪ B by
{
f (i) if i < m
h(i) =
g(i − m) if m ≤ i < m + n
To see that h is surjective, note that every k in A ∪ B can be written as either k = f (i) for some i ∈ [m] or
k = g(j) for some j ∈ [n]. In the first case, k = f (i) = h(i), and in the second case, k = g(j) = h(m + j).
It is not hard to show that h is also injective. Suppose h(i) = h(j). If h(i) is in A, then it is not in the range
of g, and so we must have h(i) = f (i) and h(j) = f (j). Then f (i) = f (j), the injectivity of f implies that
i = j. If h(i) is instead in B, the argument it similar.
The proof only spells out our basic intuitions: if you want to list all of the elements of A ∪ B, you can list
all the elements of A and then all the elements of B. And if A and B have no elements in common, then
to count the elements of A ∪ B, you can count the elements of A and then continue counting the elements
of B. Once you are comfortable translating the intuitive argument into a precise mathematical proof (and
mathematicians generally are), you can use the more intuitive descriptions (and mathematicians generally
do).
Here is another basic counting principle:
Theorem. Let A and B be finite sets. Then |A × B| = |A| · |B|.
Notice that this time we are counting the number of ordered pairs (a, b) with a ∈ A and b ∈ B. The exercises
ask you to give a detailed proof of this theorem. There are at least two ways to go about it. The first is to
start with bijections f : [m] → A and g : [n] → B and describe an explicit bijection h : [m · n] → A × B. The
182 Chapter 20. Combinatorics

second is to fix m, say, and use induction on n and the previous counting principle. Notice that if U and V
are any sets and w is not in V , we have
U × (V ∪ {w}) = (U × V ) ∪ (U × {w}),
and the two sets in this union are disjoint.

∪ ∩
Just as we have notions of union∏ i∈I Ai and intersection i∈I Ai for indexed families of sets, it is useful to
have a notion of a product i∈I Ai . We can think of an element a of this product as a function ∏ which, for
each element i ∈ I, returns an element ai ∈ Ai . For example, when I = {1, 2, 3}, an element of i∈I Ai is
just a triple a1 , a2 , a3 with a1 ∈ A1 , a2 ∈ A2 , and a3 ∈ A3 . This is essentially the same as A1 × A2 × A3 , up
to the fiddly details as to whether we represent a triple as a function or with iterated pairing (a1 , (a2 , a3 )).
Theorem. Let I be a finite index set, and let (Ai )i∈I be a family of finite sets. Then:
∪ ∑
• If each pair of sets Ai , Aj are disjoint, then | i∈I Ai | = i∈I |Ai |.
∏ ∏
• | i∈I Ai | = i∈I |Ai |.
Proof. By induction on |I|, using the previous counting principles.
We can already use these principles to carry out basic calculations.
Example. The dessert menu at a restaurant has four flavors of ice cream, two kinds of cake, and three
kinds of pie. How many dessert choices are there?
Solution. 4 + 2 + 3 = 9, the cardinality of the union of the three disjoint sets.
Example. The menu at a diner has 6 choices of appetizers, 7 choices of entrée, and 5 choices of dessert.
How many choices of three-course dinners are there?
Solution. A three-course dinner is a triple consisting of an appetizer, an entrée, and a dessert. There are
therefore 6 · 7 · 5 = 210 options.
A special case of the previous counting principles arises when all the
∪ sets have the same size. If I has
cardinality k and each Ai has cardinality n, then the cardinality of i∈I Ai is k · n if the sets are pairwise
∏
disjoint, and the cardinality of i∈I Ai is nk .
Example. A deck of playing cards has four suits (diamonds, hearts, spades, and clubs) and 13 cards in each
suit, for a total of 4 · 13 = 52.
Example. A binary string of length n is a sequence of n many 0’s and 1’s. We can think of this as an
element of
∏
{0, 1}n = {0, 1},
i<n
so there are 2n many binary strings of length n.
There is another counting principle that is almost too obvious to mention: if A is a finite set and there is a
bijection between A and B, then B is also finite, and |A| = |B|.
20.2. Counting Principles 183

Example. Consider the power set of [n], that is, the collection of all subsets of {0, 1, 2, . . . , n − 1}. There
is a one-to-one correspondence between subsets and binary strings of length n, where element i of the string
is 1 if i is in the set and 0 otherwise. As a result, we have |P([n])| = 2n .
20.3 Ordered Selections
Let S be a finite set, which we will think of as being a set of options, such as items on a menu or books that
can be selected from a shelf. We now turn to a family of problems in combinatorics that involves making
repeated selections from that set of options. In each case, there are finitely many selections, and the order
counts: there is a first choice, a second one, a third one, and so on.
In the first variant of the problem, you are allowed to repeat a choice. For example, if you are choosing 3
flavors from a list of 31 ice cream flavors, you can choose “chocolate, vanilla, chocolate.” This is known as
ordered selection with repetition. If you are making k choices from among n options in S, such a selection is
essentially a tuple (a0 , a1 , . . . , ak−1 ), where each ai is one of the n elements in S. In other words, the set of
ways of making k selections from S with repetition is the set S k , and we have seen in the last section that
if S has cardinality n, the set S k has cardinality nk .
Theorem. Let S be a set of n elements. Then the number of ways of making k selections from S with
repetition allowed is nk .
Example. How many three-letter strings (like “xyz,” “qqa,” …) can be formed using the twenty-six letters
of the alphabet?
Solution. We have to make three selections from a set of 26 elements, for a total of 263 = 17, 576 possibilities.
Suppose instead we wish to make k ordered selections, but we are not allowed to repeat ourselves. This
would arise, from example, if a museum had 26 paintings in its storeroom, and has to select three of them
to put on display, ordered from left to right along a wall. There are 26 choices for the first position. Once
we have made that choice, 25 remain for the second position, and then 24 remain for the third. So it seems
clear that there are 26 · 25 · 24 arrangements overall.
Let us try to frame the problem in mathematical terms. We can think of an ordered selection of k elements
from a set S without repetition as being an injective function f from [k] to S. The element f (0) is the first
choice; f (1) is the second choice, which has to be distinct from f (0); f (2) is the third choice, which has to
be distinct from f (0) and f (1); and so on.
Theorem. Let A and B be finite sets, with |A| = k and |B| = n, and k ≤ n. The number of injective
functions from A to B is n · (n − 1) · . . . · (n − k + 1).
Proof. Using induction on k, we will show that for every A, B, and n ≥ k, the claim holds. When k = 0
there is only one injective function, namely the function with empty domain. Suppose A has cardinality
k + 1, let a0 be any element of A. Then any injective function from A to B can be obtained by choosing an
element b0 for the image of a0 , and then choosing an injective function from A \ {a0 } to B \ {b0 }. There are
n choices of b0 , and since |A \ {a0 }| = n − 1 and |B \ {b0 }| = k − 1, there are (n − 1) · . . . · (n − k + 1) choices
of the injective function, by the inductive hypothesis.

Theorem. Let S be a finite set, with |S| = n. Then the number of ways of making k selections from S
without repetition allowed is n · (n − 1) · . . . · (n − k + 1).
Proof. This is just a restatement of the previous theorem, where A = [k] and B = S.
If A is a finite set, a bijection f from A to A is also called a permutation of A. The previous theorem shows
that if |A| = n then the number of permutations of A is n · (n − 1) · . . . · 1. This quantity comes up so often
that it has a name, n factorial, and a special notation, n!. If we think of the elements of A listed in some
order, a permutation of A is essentially an ordered selection of n elements from A without repetition: we
choose where to map the first element, then the second element, and so on. It is a useful convention to take
0! to be equal to 1.
The more general case where we are choosing only k elements from a set A is called a k-permutation of
A. The theorem above says that the number of k-permutations of an n-element set is equal to n!/(n − k)!,
because if you expand the numerator and denominator into products and cancel, you get exactly the n ·
(n − 1) · . . . · (n − k + 1). This number is often denoted P (n, k) or Pkn , or some similar variant. So we have
P (n, k) = n!/(n − k)!. Notice that the expression on the right side of the equality provides an efficient way
of writing the value of P (n, k), but an inefficient way of calculating it.
20.4 Combinations and Binomial Coefficients
In the last section, we calculated the number of ways in which a museum could arrange three paintings along
a wall, chosen from among 26 paintings in its storeroom. By the final observation in the previous section,
we can write this number as 26!/23!.
Suppose now we want to calculate the number of ways that a museum can choose three paintings from
its storeroom to put on display, where we do not care about the order. In other words, if a, b, and c are
paintings, we do not want to distinguish between choosing a then b then c and choosing c then b then a.
When we were arranging paintings along all wall, it made sense to consider these two different arrangements,
but if we only care about the set of elements we end up with at the end, the order that we choose them does
not matter.
The problem is that each set of three paintings will be counted multiple times. In fact, each one will be
counted six times: there are 3! = 6 permutations of the set {a, b, c}, for example. So to count the number of
26!
outcomes we simply need to divide by 6. In other words, the number we want is 3!·23! .
There is nothing special about the numbers 26 and 3. The same formula holds for what we will call unordered
selections of k elements from a set of n elements, or k-combinations from an n-element set. Our goal is
once again to describe the situation in precise mathematical terms, at which point we will be able to state
the formula as a theorem.
In fact, describing the situation in more mathematical terms is quite easy to do. If S is a set of n elements,
an unordered selection of k elements from S is just a subset of S that has cardinality k.
Theorem. Let S be any set with cardinality n, and let k ≤ n. Then the number of subsets of S of cardinality
n!
k is k!(n−k)! .
Proof. Let U be the set of unordered selections of k elements from S, let V be the set of permutations of [k],
and let W be the set of ordered selections of k elements from S. There is a bijection between U × V and W ,
as follows. Suppose we assign to every k-element subset {a0 , . . . , ak−1 } of S some way of listing the elements,
as shown. Then given any such set and any permutation f of [k], we get an ordered the ordered selection
(af (0) , af (1) , . . . , af (k−1) ). Any ordered selection arises from such a subset and a suitable permutation, so the
20.4. Combinations and Binomial Coefficients 185

mapping is surjective. And a different set or a different permutation results in a different ordered selection,
so the mapping is injective.
By the counting principles, we have
P (n, k) = |W | = |U × V | = |U | · |V | = |U | · k!,
so we have |U | = P (n, k)/k! = n!

k!(n−k)! .
Example. Someone is going on vacation and wants to choose three outfits from ten in their closet to pack
in their suitcase. How many choices do they have?
10! 10·9·8
Solution. 3!7! = 3·2·1 = 120.
The number of unordered selections of k elements from a set ( )of size n, or, equivalently, the number of k-
combinations from an n-element set, is typically denoted by nk( , )C(n, k), Ckn , or something similar. We will
use the first notation, because it is most common. Notice that n0 = 1 for every n; this makes sense, because
there is exactly one subset of any n-element set of cardinality 0.
Here is one important property of this function.
( n) ( )
Theorem. For every n and k ≤ n, we have k = n
n−k .
Proof. This is an easy calculation:
n! n!
= .
(n − k)!(n − (n − k))! (n − k)!k!
But it is also easy to see from the combinatorial interpretation: choosing k outfits from n to take on vacation
is the same task as choosing n − k outfits to leave home.
Here is another important property.
Theorem. For every n and k, if k + 1 ≤ n, then

( ) ( ) ( )
n+1 n n
= + .
k+1 k+1 k
Proof. One way to understand this theorem is in terms of the combinatorial interpretation. Suppose you
want to choose k + 1 outfits out of n + 1. Set aside one outfit, say, the
( nblue
) one. Then you have two choices:
you can either choose k + 1 outfits from the remaining ones, with k+1 possibilities; or you can take the
blue one, and choose k outfits from the remaining ones.
The theorem can also be proved by direct calculation. We can express the left-hand side of the equation as
follows:
( )
n+1 (n + 1)!
=
k+1 (k + 1)!((n + 1) − (k + 1))!
(n + 1)!
=
(k + 1)!(n − k)!

Similarly, we can simplify the right-hand side:

( ) ( )
n n n! n!
+ = +
k+1 k (k + 1)!(n − (k + 1))! k!(n − k)!
n!(n − k) (k + 1)n!
= +
(k + 1)!(n − k − 1)!(n − k) (k + 1)k!(n − k)!
n!(n − k) (k + 1)n!
= +
(k + 1)!(n − k)! (k + 1)!(n − k)!
n!(n − k + k + 1)
=
(k + 1)!(n − k)!
n!(n + 1)
=
(k + 1)!(n − k)!
(n + 1)!
=
(k + 1)!(n − k)!
Thus the left-hand side and the right-hand side are equal.
( ) ( )
For every n, we know n0 = nn = 1. The previous theorem then gives a recipe to compute all the binomial
( ) ( )
coefficients: once we have determine nk for some n and every k ≤ n, we can determine the values of n+1
k
for every k ≤ n + 1 using the recipe above. The results can be displayed graphically in what is known as
Pascal’s triangle:
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
( )
Specifically, if we start counting at 0, the kth element of the nth row is equal to nk .
( n) n
There is also a connection
(n) between k and the polynomials (a + b) , namely, that the kth coefficient of
n
(a + b) is exactly k . For example, we have
(a + b)4 = a4 + 4a3 b + 6a2 b2 + 4ab3 + b4 .

( n)
For that reason, the values k are often called binomial coefficients, and the statement that
∑ (n)
(a + b)n = an−k bk
k
k≤n
is known as the binomial theorem.

There are a couple of ways of seeing why this theorem holds. One is to expand the polynomial,
(a + b)n = (a + b)(a + b) · · · (a + b)
and notice that the coefficient of the term an−k bk is equal to the number of ways of taking the summand
) ( n n)− k(npositions.
b in exactly k positions, and a in the( remaining ) Another way to prove the result is to use
induction on n, and use the identity n+1
k+1 = k+1 + k . The details are left as an exercise.
Finally, we have considered ordered selections with and without repetitions, and unordered selections without
repetitions. What about unordered selections with repetitions? In other words, given a set S with n elements,
20.4. Combinations and Binomial Coefficients 187

we would like to know how many ways there are of making k choices, where we can choose elements of S
repeatedly, but we only care about the number of times each element was chosen, and not the order. We
have the following:
(n+k−1)
The number of unordered selections of k elements from an n-element set, with repetition, is k .
A proof of this is outlined in the exercises.
20.5 The Inclusion-Exclusion Principle
Let A and B be any two subsets of some domain, U . Then A = A \ B ∪ (A ∩ B), and the two sets in the
union are disjoint, so we have |A| = |A \ B| + |A ∩ B|. This means |A \ B| = |A| − |A ∩ B|. Intuitively, this
makes sense: we can count the elements of A \ B by counting the elements in A, and then subtracting the
number of elements that are in both A and B.
Similarly, we have A ∪ B = A ∪ (B \ A), and the two sets on the right-hand side of this equation are disjoint,
so we have
|A ∪ B| = |A| + |B \ A| = |A| + |B| − |A ∩ B|.
If we draw a Venn diagram, this makes sense: to count the elements in A ∪ B, we can add the number of
elements in A to the number of elements in B, but then we have to subtract the number of elements of both.
What happen when there are three sets? To compute |A ∪ B ∪ C|, we can start by adding the number of
elements in each, and then subtracting the number of elements of |A ∩ B|, |A ∩ C|, and |B ∩ C|, each of which
have been double-counted. But thinking about the Venn diagram should help us realize that then we have
over-corrected: each element of A ∩ B ∩ C was counted three times in the original sum, and the subtracted
three times. So we need to add them back in:
|A ∪ B ∪ C| = |A| + |B| + |C| − |A ∩ B| − |A ∩ C| − |B ∩ C| + |A ∩ B ∩ C|.
This generalizes to any number of sets. To state the general result, suppose ∩ the sets are numbered
A0 , . . . , An−1 . For each nonempty subset I of {0, . . . , n − 1}, consider i∈I Ai . If |I| is odd (that is,
equal to 1, 3, 5, …) we want to add the cardinality of the intersection; if it is even we want to subtract it.
This recipe is expressed compactly by the following formula:

∪ ∑ ∩
|I|
Ai = (−1) Ai

i<n ∅̸=I⊆[n] i∈I
You are invited to try proving this as an exercise, if you are ambitious. The following example illustrates its
use:
Example. Among a group of college Freshmen, 30 are taking Logic, 25 are taking History, and 20 are taking
French. Moreover, 11 are taking Logic and History, 10 are taking Logic and French, 7 are taking History
and French, and 3 are taking all three. How many students are taking at least one of the three classes?
Solution. Letting L, H, and F denote the sets of students taking Logic, History, and French, respectively,
we have
|L ∪ H ∪ F | = 30 + 25 + 20 − 11 − 10 − 7 + 3 = 50.

20.6 Exercises
1. Suppose that, at a party, every two people either know each other or don’t. In other words, “x knows
y” is symmetric. Also, let us ignore the complex question of whether we always know ourselves by
restricting attention to the relation between distinct people; in other words, for this problem, take “x
knows y” to be irreflexive as well.
Use the pigeonhole principle (and an additional insight) to show that there must be two people who
know exactly the same number of people.
2. Show that in any set of n + 1 integers, two of them are equivalent modulo n.
3. Spell out in detail a proof of the second counting principle in Section 20.2.
4. An ice cream parlor has 31 flavors of ice cream.
(a) Determine how many three-flavor ice-cream cones are possible, if we care about the order and
repetitions are allowed. (So choosing chocolate-chocolate-vanilla scoops, from bottom to top, is
different from choosing chocolate-vanilla-chocolate.)
(b) Determine how many three flavor ice-cream cones are possible, if we care about the order, but
repetitions are not allowed.
(c) Determine how many three flavor ice-cream cones are possible, if we don’t care about the order,
but repetitions are not allowed.
5. A club of 10 people has to elect a president, vice president, and secretary. How many possibilities are
there:
(a) if no person can hold more than one office?
(b) if anyone can hold any number of those offices?
(c) if anyone can hold up to two offices?
(d) if the president cannot hold another office, but the vice president and secretary may or may not
be the same person?
6. How many 7 digit phone numbers are there, if any 7 digits can be used? How many are there is the
first digit cannot be 0?
7. In a class of 20 kindergarten students, two are twins. How many ways are there of lining up the
students, so that the twins are standing together?
8. A woman has 8 murder mysteries sitting on her shelf, and wants to take three of them on a vacation.
How many ways can she do this?
9. In poker, a “full house” is a hand with three of one rank and two of another (for example, three kings
and two fives). Determine the number of full houses in poker.
10. We saw in Section 20.4 that
( ) ( ) ( )
n+1 n n
= + .
k+1 k+1 k
Replacing k + 1 by k, whenever 1 ≤ k ≤ n, we have

( ) ( ) ( )
n+1 n n
= + .
k k k−1
( n)
Use this to show, by induction on n, that for every k ≤ n, that if S is any set of n elements, k is the
number of subsets of S with k elements.
20.6. Exercises 189

11. How many distinct arrangements are there of the letters in the word MISSISSIPPI?
(Hint: this is tricky. First, suppose all the S’s, I’s, and P’s were painted different colors. Then determine
how many distinct arrangements of the letters there would be. In the absence of distinguishing colors,
determine how many times each configuration appeared in the first count, and divide by that number.
12. Prove the inclusion exclusion principle.
13. Use the inclusion exclusion principle to determine the number of integers less than 100 that are divisible
by 2, 3, or 5.
( )
14. Show that the number of unordered selections of k elements from an n-element set is n+k−1 k .
Hint: consider [n]. We need to choose some number i0 of 0’s, some number i1 of 1’s, and so on, so that
i0 + i1 + . . . + in−1 = k. Suppose we assign to each such tuple a the following binary sequence: we write
down i0 0’s, then a 1, then i1 0′ s, then a 1, then i2 0’s, and so on. The result is a binary sequence of
length n + k − 1 with exactly k 1’s, and such binary sequence arises from a unique tuple in this way.

CHAPTER
TWENTYONE
THE REAL NUMBERS
21.1 The Number Systems
We have already come across some of the fundamental number systems: the natural numbers, N, the integers,
Z, and the rationals, Q. In a sense, each subsequent element of the list was designed to remedy defects in
the previous system: we can subtract any integer from any other integer and end up with another integer,
and we can divide any rational number by a nonzero rational number and end up with a rational number.
The integers satisfy all of the following properties:
• Addition is associative and commutative.
• There is an additive identity, 0, and every element x has an additive inverse, −x.
• Multiplication is associative and commutative.
• There is a multiplicative identity, 1.
• Multiplication distributes over addition: for every x, y, and z, we have x(y + z) = xy + xz.
• The ordering ≤ is a total order.
• For any elements x, y, and z, if x ≤ y then x + z ≤ y + z.
• For any elements x and y, if 0 ≤ x and 0 ≤ y then 0 ≤ xy.
The first five clauses say that with ×, +, 0, and 1, the integers form a commutative ring, and the last three
say that together with ≤, the structure is an ordered ring. The natural numbers lack additive inverses, so
they satisfy a slightly weaker set of axioms that make them an ordered semiring. On the other hand, the
rational numbers also form an ordered ring, satisfying the following additional property:
• Every nonzero element has a multiplicative inverse, x−1 .
This makes them an instance of an ordered field.
It is worth knowing that once we have the natural numbers, it is possible to construct the integers and
rational numbers, using set-theoretic constructions you have already seen. For example, we can take an
integer to be a pair (i, n) of natural numbers where i is either 0 or 1, with the intention that (0, n) represents
the positive integer n, and (1, n) represents the negative integer −(n + 1). (We use −(n + 1) instead of −n to
avoid having two representations of 0.) With this definition, the integers are simply {0, 1} × N. We can then
go on to define the operations of addition and multiplication, the additive inverse, and the order relation,
and prove they have the desired properties.
This construction has the side effect that the natural numbers themselves are not integers; for example, we
have to distinguish between the natural number 2, and the integer 2. In Lean, for example, this is the case.
In ordinary mathematics, it is common to think of the natural numbers as a subset of the integers. Once we
construct the integers, however, we can throw away the old version of the natural numbers, and afterwards
identify the natural numbers as nonnegative integers.
191
We can do the same for the rationals, defining them to be the set of pairs (a, b) in Z × N, where either a = 0
and b = 1, or b > 0 and a and b have no common divisor (other than 1 and −1). The idea is that (a, b)
represents a/b. With this definition, the rationals are really a subset of Z × N, and we can then define all
the operations accordingly.
In the next section, we will define a more sophisticated approach, one which will scale to a definition of the
real numbers. And in a later chapter, we will show how to construct the natural numbers from the axioms
of set theory. This shows that we can construct all the number systems from the bottom up.
But first, let us pause for a moment to consider why the real numbers are needed. We have seen that 2 has
no rational square root. This means, in a sense, that there is a “gap” in the rationals: the are rationals whose
squares are arbitrarily close to 2, but there is no rational x with the property
√ that x2 = 2. But it seems
intuitively clear that there should be some number with that property: 2 is the length of the diagonal of
a square with side length 1. Similarly, π, the area of a circle with radius 1, is missing from the rationals.
These are the kinds of defects that the real numbers are designed to repair.
√
You may be used to thinking of real numbers as (potentially) infinite decimals: for example, 2 =
1.41421356 . . . and π = 3.14159265 . . .. A central goal of this chapter is to make the “…” precise. The
idea is that we can take an infinite decimal to represent a sequence of rational approximations. For example,√
we can approximate the square root of 2 with the sequence 1, 1.4, 1.41, 1.414, . . .. We would like to define 2
to be the “limit” of that sequence, but we have seen that the sequence does not have a limit in the rationals.
So we have to construct new objects, the real numbers, to serve that purpose.
In fact, we will define the real numbers, more or less, to be such sequences of rational approximations. But
we will have to deal with the fact that, for example, there are lots of ways of approximating the square root
of two. For example, we can just as well approach it from above, 2, 1.5, 1.42, . . ., or by oscillating above and
below. The next section will show us how to “glue” all these sequences together and treat them as a single
object.
21.2 Quotient Constructions
Let A be any set, and let ≡ be any equivalence relation on A. Recall from Section 13.3 that we can assign to
every element a of A the equivalence class [a], where b ∈ [a] means b ≡ a. This assignment has the property
that for every a and b, a ≡ b if and only if [a] = [b].
Given any set A and equivalence relation ≡, define A/≡ to be the set {[a] | a ∈ A} of equivalence classes of
A modulo ≡. This set is called “A modulo ≡,” or the quotient of A by ≡. You can think of this as the set
A where equivalent elements are “glued together” to make a coarser set.
For example, if we consider the integers Z with ≡ denoting equivalence modulo 5 (as in Section 19.4), then
Z/≡ is just {[0], [1], [2], [3], [4]}. We can define addition on Z/≡ by [a] + [b] = [a + b]. For this definition
to make sense, it is important to know that the right-hand side does not depend on which representatives
of [a] and [b] we choose. In other words, we need to know that whenever [a] = [a′ ] and [b] = [b′ ], then
[a + b] = [a′ + b′ ]. This, in turn, is equivalent to saying that if a ≡ a′ and b ≡ b′ , then a + b ≡ a′ + b′ .
In other words, we require that the operation of addition respects the equivalence relation, and we saw in
Section 19.4 that this is in fact the case.
This general strategy for transferring a function defined on a set to a function defined on a quotient of that
set is given by the following theorem.
Theorem. Let A and B be any sets, let ≡ be any equivalence relation defined on A, and let f : A → B.
Suppose f respects the equivalence relation, which is to say, for every a and a′ in A, if a ≡ a′ , then
f (a) = f (a′ ). Then there is a unique function f¯ : A/≡ → B, defined by f¯([a]) = f (a) for every a in A.
192 Chapter 21. The Real Numbers

Proof. We have defined the value of f¯ on an equivalence class x by writing x = [a], and setting f¯(x) = f (a).
In other words, we say that f¯(x) = y if and only if there is an a such that x = [a], and f (a) = y. What is
dubious about the definition is that, a priori, it might depend on how we express x in that form; in other
words, we need to show that there is a unique y meeting this description. Specifically, we need to know that
if x = [a] = [a′ ], then f (a) = f (a′ ). But since [a] = [a′ ] is equivalent to a ≡ a′ , this amounts to saying that
f respects the equivalence relation, which is exactly what we have assumed.
Mathematicians often “define” f¯ by the equation f¯([a]) = f (a), and then express the proof above as a proof
that “f¯ is well defined.” This is confusing. What they really mean is what the theorem says, namely, that
there is a unique function meeting that description.
To construct the integers, start with N × N. Think of the pair of natural numbers (m, n) as representing
m − n, where the subtraction takes place in the integers (which we haven’t constructed yet!). For example,
both (2, 5) and (6, 9) represent the integer −3. Intuitively, the pairs (m, n) and (m′ , n′ ) will represent the
same integer when m − n = m′ − n′ , but we cannot say this yet, because we have not yet defined the
appropriate notion of subtraction. But the equation is equivalent to m + n′ = m′ + n, and this makes sense
with addition on the natural numbers.
Definition. Define the relation ≡ on N × N by (m, n) ≡ (m′ , n′ ) if and only if m + n′ = m′ + n.

Proposition. ≡ is an equivalence relation.
Proof. For reflexivity, it is clear that (m, n) ≡ (m, n), since m + n = m + n.
For symmetry, suppose (m, n) ≡ (m′ , n′ ). This means m + n′ = m′ + n. But the symmetry of equality
implies (m′ , n′ ) ≡ (m, n), as required.
For transitivity, suppose (m, n) ≡ (m′ , n′ ), and (m′ , n′ ) = (m′′ , n′′ ). Then we have m + n′ = m′ + n and
m′ + n′′ = n′ + m′′ . Adding these equations, we get
m + n′ + m′ + n′′ = m′ + n + n′ + m′′ .
Subtracting m′ + n′ from both sides, we get m + n′′ = n + m′′ , which is equivalent to (m′ , n′ ) = (m′′ , n′′ ),
as required.
We can now define the integers to be N × N/≡. How should we define addition? If [(m, n)] represents m − n,
and [(u, v)] represents u − v, then [(m, n)] + [(u, v)] should represent (m + u) − (n + v). Thus, it makes sense
to define [(m, n)] + [(u, v)] to be [(m + u) − (n + v)]. For this to work, we need to know that the operation
which sends (m, n) and (u, v) to (m + u, n + v) respects the equivalence relation.
Proposition. If (m, n) ≡ (m′ , n′ ) and (u, v) ≡ (u′ , v ′ ), then (m + u, n + v) ≡ (m′ + u′ , n′ + v ′ ).

Proof. The first equivalence means m + n′ = m′ + n, and the second means u + v ′ = u′ + v. Adding
the two equations, we get (m + u) + (n′ + v ′ ) ≡ (m′ + u′ ) + (n + v), which is exactly the same as saying
(m + u, n + v) ≡ (m′ + u′ , n′ + v ′ ).
Every natural number n can be represented by the integer [(n, 0)], and, in particular, 0 is represented by
[(0, 0)]. Moreover, if [(m, n)] is any integer, we can define its negation to be [(n, m)], since [(m, n)]+[(n, m)] =
[(m + n, n + m)] = [(0, 0)], since (m + n, n + m) ≡ (0, 0). In short, we have “invented” the negative numbers!
We could go on this way to define multiplication and the ordering on the integers, and prove that they have
the desired properties. We could also carry out a similar construction for the rational numbers. Here, we
21.2. Quotient Constructions 193

would start with the set Z × Z>0 , where Z>0 denotes the strictly positive integers. The idea, of course, is
that (a, b) represents (a/b). With that in mind, it makes sense to define (a, b) ≡ (c, d) if ad = bc. We could
go on to define addition, multiplication, and the ordering there, too. The details are tedious, however, and
not very illuminating. So we turn, instead, to a construction of the real numbers.
21.3 Constructing the Real Numbers
The problem we face is that the sequence 1, 1.4, 1.41, 1.414, 1.4142, . . . of rational numbers seems to approach
a value that would be the square root of 2, but there is no rational number that can play that role. The next
definition captures the notion that this sequence of numbers “seems to approach a value,” without referring
to a value that it is approaching.
Definition. A sequence of rational numbers (qi )i∈N is Cauchy if for every rational number ε > 0, there is
some natural number N ∈ N such that for all i, j ≥ N , we have that |qi − qj | < ε.
Roughly speaking, a Cauchy sequence is one where the elements become arbitrarily close, not just to their
successors but to all following elements. It is common in mathematics to use ε to represent a quantity that
is intended to denote something small; you should read the phrase “for every ε > 0” as saying “no matter
how small ε is.” So a sequence is Cauchy if, for any ε > 0, no matter how small, there is some point N ,
beyond which the elements stay within a distance of ε of one another.
Cauchy sequences can be used to describe these gaps in the rationals, but, as noted above, many Cauchy
sequences can be used to describe the same gap. At this stage, it is slightly misleading to say that they
“approach the same point,” since there is no rational point that they approach; a more precise statement is
that the sequences eventually become arbitrarily close.
Definition. Two Cauchy sequences p = (pi )i∈N and q = (qi )i∈N are equivalent if for every rational number
ε > 0, there is some natural number N ∈ N such that for all i ≥ N , we have that |pi − qi | < ε. We will write
p ≡ q to express that p is equivalent toq.
Proposition. ≡ is an equivalence relation on Cauchy sequences.
Proof. Reflexivity and symmetry are easy, so let us prove transitivity. Suppose (pi ) ≡ (qi ) and (qi ) ≡ (ri ).
We want to show that the sequence (qi ) is equivalent to (ri ). So, given any ε > 0, choose N0 large enough such
that for every i ≥ N0 , |pi − qi | < ε/2. Choose another number, N1 , so that for every i ≥ N1 , |qi − ri | < ε/2.
Let N = max(N0 , N1 ). Then for every i ≥ N , we have
|pi − ri | = |(pi − qi ) + (qi − ri )| < |pi − qi | + |qi − ri | ≤ ε/2 + ε/2 = ε,
as required.
Notice that the proof uses the triangle inequality, which states for any rational numbers a and b, |a + b| ≤
|a| + |b|. If we define |a| to be the maximum of a and −a, the triangle inequality in fact holds for any ordered
ring:
Theorem. Let a and b be elements of any ordered ring. Then |a + b| ≤ |a| + |b|.

Proof. By the definition of absolute value, it suffices to show that a + b ≤ |a| + |b| and −(a + b) ≤ |a| + |b|.
The first claim follows from the fact that a ≤ |a| and b ≤ |b|. For the second claim, we similarly have
−a ≤ |a| and −b ≤ |b|, so −(a + b) = −a + −b ≤ |a| + |b|.
In the theorem above, if we let a = x − y and b = y − z, we get |x − z| ≤ |x − y| + |y − z|. The fact that |x − y|
represents the distance between x and y on the number line explains the name: for any three “points” x, y,
and z, the distance from x to z can’t be any greater than the distance from x to y plus the distance from y
to z.
We now let A be the set of Cauchy sequences of rationals, and define the real numbers, R, to be A/≡. In
other words, the real numbers are the set of Cauchy sequence of rationals, modulo the equivalence relation
we just defined.
Having the set R by itself is not enough: we also would like to know how to add, subtract, multiply, and
divide real numbers. As with the integers, we need to define operations on the underlying set, and then
show that they respect the equivalence relation. For example, we will say how to add Cauchy sequences of
rationals, and then show that if p1 ≡ p2 and q1 ≡ q2 , then p1 + q1 ≡ p2 + q2 . We can then lift this definition
to R by defining [p] + [q] to be [p + q].
Luckily, it is easy to define addition, subtraction, and multiplication on Cauchy sequences. If p = (pi )i∈N and
q = (qi )i∈N are Cauchy sequences, let p + q = (pi + qi )i∈N , and similarly for subtraction and multiplication.
It is trickier to show that these sequences are Cauchy themselves, and to show that the operations have the
appropriate algebraic properties. We ask you to prove some of these properties in the exercises.
We can identify each rational number q with the constant Cauchy sequence q, q, q, . . ., so the real numbers
include all the rationals. The next step is to abstract away the details of the particular construction we have
chosen, so that henceforth we can work with the real numbers abstractly, and no longer think of them as
given by equivalence classes of Cauchy sequences of rationals.
21.4 The Completeness of the Real Numbers
We constructed the real numbers to fill in the gaps in the rationals. How do we know that we have got them
all? Perhaps we need to construct even more numbers, using Cauchy sequences of reals? The next theorem
tells us that, on the contrary, there is no need to extend the reals any further in this way.
Definition. Let r be a real number. A sequence (ri )i∈N of real numbers converges to r if, for every ε > 0,
there is an N such that for every i ≥ N , |ri − r| < ε.
Definition. A sequence (ri )i∈N converges if it converges to some r.
Theorem. Every Cauchy sequence of real numbers converges.
The statement of the theorem is often expressed by saying that the real numbers are complete. Roughly, it
says that everywhere you look for a real number, you are bound to find one. Here is a similar principle.
Definition. An element u ∈ R is said to be an upper bound to a subset S ⊆ R if everything in S is less than

or equal to u. S is said to be bounded if there is an upper bound to S. An element u is said to be a least
upper bound to S if it is an upper bound to S, and nothing smaller than u is an upper bound to S.
Theorem. Let S be a bounded, nonempty subset of R. Then S has a least upper bound.
21.4. The Completeness of the Real Numbers 195

The rational numbers do not have this property: if we set S = {x ∈ Q | x2 < 2}, then the rational number
2 is an upper bound for S, but S has no least upper bound in Q.
It is a fundamental theorem that the real numbers are characterized exactly by the property that they are
a complete ordered field, such that every real number r is less than or equal to some natural number N .
Any two models that meet these requirements must behave in exactly the same way, at least insofar as the
constants 0 and 1, the operations + and ∗, and the relation ≤ are concerned. This fact is extremely powerful
because it allows us to avoid thinking about the Cauchy sequence construction in normal mathematics.
Once we have shown that our construction meets these requirements, we can take R to be “the” unique
complete totally ordered field and ignore any implementation details. We are also free to implement R in
any way we choose, and as long as it meets this interface, and as long as they do not refer to the underlying
representations, any theorems we prove about the reals will hold equally well for all constructions.
21.5 An Alternative Construction
Many sources use an alternative construction of the reals, taking them instead to be Dedekind cuts. A
Dedekind cut is an ordered pair (A, B) of sets of rational numbers with the following properties:
• Every rational number q is in either A or B.
• Each a ∈ A is less than every b ∈ B.
• There is no greatest element of A.
• A and B are both nonempty.
The first two properties show why we call this pair a “cut.” The set A contains all of the rational numbers
to the left of some mark on the number line, and B all of the points to the right. The third property tells us
something about what happens exactly at that mark. But there are two possibilities: either B has a least
element, or it doesn’t. Picturing the situation where A has no greatest element and B has no least element
may be tricky, but consider the example A = {x ∈ Q | x2 < 2} and B = {x ∈ Q | x2 > 2}. There is no
rational number q such that q 2 = 2, but there are rational numbers on either side that are arbitrarily close;
thus neither A nor B contains an endpoint.
We can define R to be the set of Dedekind cuts. A Dedekind cut (A, B) corresponds to a rational number q
if q is the least element of B, and to an irrational number if B has no least element. It is straightforward to
define addition on R:
(A1 , B1 ) + (A2 , B2 ) = ({a1 + a2 | a1 ∈ A1 , a2 ∈ A2 }, {b1 + b2 | b1 ∈ B1 , b2 ∈ B2 })
Some authors prefer this construction to the Cauchy sequence construction because it avoids taking the quo-
tient of a set, and thus removes the complication of showing that arithmetic operations respect equivalence.
Others prefer Cauchy sequences since they provide a clearer notion of approximation: if a real number r is
given by a Cauchy sequence (qi )i∈N , then an arbitrarily close rational approximation of r is given by qN for
a sufficiently large N .
For most mathematicians most of the time, though, the difference is immaterial. Both constructions create
complete linear ordered fields, and in a certain sense, they create the same complete linear ordered field.
Strictly speaking, the set of Cauchy reals is not equal to the set of Dedekind reals, since one consists of
equivalence classes of rational Cauchy sequences and one consists of pairs of sets of rationals. But there is
a bijection between the two sets that preserves the field properties. That is, there is a bijection f from the
Cauchy reals to the Dedekind reals such that
• f (0) = 0
• f (1) = 1
• f (x + y) = f (x) + f (y)

• f (x · y) = f (x) · f (y)
• f (−x) = −f (x)
• f (x−1 ) = f (x)−1
• f (x) ≤ f (y) ⇐⇒ x ≤ y.
We say that the two constructions are isomorphic, and that the function f is an isomorphism. Since we
often only care about the real numbers in regard to their status as a complete ordered field, and the two
constructions are indistinguishable as ordered fields, it makes no difference which construction is used.
21.6 Exercises
1. Show that addition for the integers, as defined in Section 21.2, is commutative and associative.
2. Show from the construction of the integers in Section 21.2 that a + 0 = a for every integer a.
3. Define subtraction for the integers by a − b = a + (−b), and show that a − b + b = a for every pair of
integers a and b.
4. Define multiplication for the integers, by first defining it on the underlying representation and then
showing that the operation respects the equivalence relation.
5. Show that every Cauchy sequence is bounded: that is, if (qi )i∈N is Cauchy, there is some rational M
such that |qi | ≤ M for all i. Hint: try letting ε = 1.
6. Let p = (pi )i∈N and q = (qi )i∈N be Cauchy sequences. Define p + q = (pi + qi )i∈N and pq = (pi qi )i∈N .
(a) Show that p + q is Cauchy. That is, for arbitrary ε > 0, show that there exists an N such that
for all i, j ≥ N , |(pi + qi ) − (pj + qj )| < ε.
(b) Show that pq is Cauchy. In addition to the triangle inequality, you will find the previous exercise
useful.
7. These two parts show that addition of Cauchy sequences respects equivalence.
(a) Show that if p, p′ , q are Cauchy sequences and p ≡ p′ , then p + q ≡ p′ + q.
(b) Using the first part of this problem, show that if p, p′ , q, q ′ are Cauchy sequences, p ≡ p′ , and
q ≡ q ′ , then p+q ≡ p′ +q ′ . You can use the fact that addition on the real numbers is commutative.
8. Show that if (A1 , B1 ) and (A2 , B2 ) are Dedekind cuts, then (A1 , B1 ) + (A2 , B2 ) is also a Dedekind cut.
21.6. Exercises 197


CHAPTER
TWENTYTWO
THE INFINITE
22.1 Equinumerosity
Remember that in Chapter 20 we defined, for each natural number n, the set [n] = {0, 1, . . . , n − 1}. We
then said that a set A is finite if there is a bijection between A and [n] for some n. A set is said to be infinite
if it is not finite.
If A and B are two finite sets, then they have the same cardinality if and only if there is a bijection between
them. It turns out that the same notion of “having the same cardinality” makes sense even if A and B are
not finite.
Definition. Two sets A and B are said to be equinumerous, written A ≈ B, if there is a bijection between
them. Equivalently, we say that A and B have the same cardinality.
At this stage, saying that A and B have the same cardinality may sound strange, because it is not clear
that there is any object, “the cardinality of A,” that they both “have.” It turns out that, in set-theoretic
foundations, there are certain objects — generalizations of the natural numbers — that one can use to
measure the size of an infinite set. There are known as the “cardinal numbers” or “cardinals.” But they
are irrelevant to our purposes here: for the rest of this chapter, when we say that A and B have the same
cardinality, we mean neither more nor less than the fact that there is a bijection between them.
The following theorem says, essentially, the equinumerosity is an equivalence relation. (The caveat is that
so far we have spoke only of relations between sets, and the collection of all sets is not itself a set.)
Proposition. Let A, B, and C be any sets.

• A ≈ A.
• If A ≈ B, then B ≈ A.
• If A ≈ B and B ≈ C then A ≈ C.
The proof is left as an exercise.
199
22.2 Countably Infinite Sets
The set of natural numbers, N, is a prototypical example of an infinite set. To see that it is infinite, suppose,
on the other hand, that it is finite. This means that there is a bijection f between N and [n] for some natural
number n. We can restrict to the subset [n + 1] of N, and thereby obtain an injective map from [n + 1] to
[n]. But this violates the pigeonhole principle, proved in Chapter 20.
Definition. A set A is said to be countably infinite if it is equinumerous with N. A set A is said to be

countable if it is finite or countably infinite.
Since the identity map id(x) = x is a bijection on any set, every set is equinumerous with itself, and thus N
itself is countably infinite.
The term “countably infinite” is meant to be evocative. Suppose A is a countable set. By definition, there
is a bijection f : N → A. So A has a “first” element f (0), a “second” element f (1), a “third” element f (2),
and so on. Since f is a bijection, for every element a of A, a is the nth element enumerated in this way, for
a unique value of n. That is, each element of A is “counted” at some finite stage.
With this definition in hand, it is natural to wonder which of our favorite sets are countable. Is the set of
integers Z countable? How about the set of rationals Q, or the set of reals R? At this point, you should
reflect on the logical form of the statement “A is countable,” and think about what is required to show that
a set A does or does not have this property.
Theorem. The set of integers, Z, is countable.

Proof. We need to show that there exists a bijection between N and Z. Define f : N → Z as follows:
{
n/2 if n is even
f (n) =
−(n + 1)/2 if n is odd
We claim that f is a bijection. To see that it is injective, suppose f (m) = f (n). If f (m) (and hence also
f (n)) is nonnegative, then m and n are even, in which case m/2 = n/2 implies m = n. Otherwise, m and n
are odd, and again −(m + 1)/2 = −(n + 1)/2 implies m = n.
To see that f is surjective, suppose a is any integer. If a is nonnegative, then a = f (2a). If a is strictly
negative, then 2a − 1 is also strictly negative, and hence −(2a − 1) is an odd natural number. In that case,
it is not hard to check that a = f (−(2a − 1)).
We will now build up an arsenal of theorems that we can use to show that various sets are countable.
Theorem. A set A is countable if and only if A is empty or there is a surjective function f : N → A.

Proof. For the forward direction, suppose A is countable. Then it is either finite or countably infinite. If A
is countably infinite, there is a bijection from N to A, and we are done. Suppose, then, that A is finite. If A
is empty, we are done. Otherwise, for some n, there is a bijection f : [n] → A, with n ≥ 1. Define a function
g : N → A as follows:
{
f (i) if i < n
g(i) =
f (0) otherwise
200 Chapter 22. The Infinite

In other words, g enumerates the elements of A by using f first, and then repeating the element f (0). Clearly
f is surjective, as required.
In the other direction, if A is finite, then it is countable, and we are done. So suppose A is not finite. Then
it is not empty, and so there is a surjective function f : N → A. We need to turn f into a bijective function.
The problem is that f may not be injective, which is to say, elements in A may be enumerated more than
once. The solution is to define a function, g, which eliminates all the duplicates. The idea is that g should
enumerate the elements f (0), f (1), f (2), . . ., but skip over the ones that have already been enumerated.
To be precise, the function g is defined recursively as follows: g(0) = f (0), and for every i, g(i + 1) = f (j),
where j is the least natural number such that f (j) is not among {g(0), g(1), g(2), . . . , g(i)}. The assumption
that A is infinite and f is surjective guarantees that some such j always exists.
We only need to check that g is a bijection. By definition, for every i, g(i + 1) is different from g(0), . . . , g(i).
This implies that g is injective. But we can also show by induction that for every i, {g(0), . . . , g(i)} ⊇
{f (0), . . . , f (i)}. Since f is surjective, g is too.
In a manner similar to the way we proved that the integers are countable, we can prove the following:
Theorem. If A and B are countably infinite, then so is A ∪ B.

Proof. Suppose f : N → A and g : N → B are surjective. Then we can define a function h : N → A ∪ B:
{
f (n/2) if n is even
h(n) =
g((n − 1)/2) if n is odd
It is not hard to show that h is surjective.
Intuitively, if A = {f (0), f (1), f (2), . . .} and B = {g(0), g(1), g(2), . . .}, then we can enumerate A ∪ B as
{f (0), g(0), f (1), g(1), f (2), g(2), . . .}.
The next two theorems are also helpful. The first says that to show that a set B is countable, it is enough
to “cover” it with a surjective function from a countable set. The second says that to show that a set A is
countable, then it is enough to embed it in a countable set.
Theorem. If A is countable and f : A → B is surjective, then B is countable.

Proof. If A is countable, then there is a surjective function g : N → A, and f ◦ g is a surjective function
from N → B.
Theorem. If B is countable and f : A → B is injective, then A is countable.
Proof. Assuming f : A → B is injective, it has a left inverse, g : B → A. Since g has a right inverse, f , we
know that g is surjective, and we can apply the previous theorem.
Corollary. If B is countable and A ⊆ B, then A is countable.
Proof. The function f : A → B defined by f (x) = x is injective.
Remember that N × N is the set of ordered pairs (i, j) where i and j are natural numbers.
Theorem. N × N is countable.
22.2. Countably Infinite Sets 201

Proof. Enumerate the elements as follows:
(0, 0), (1, 0), (0, 1), (2, 0), (1, 1), (1, 2), (3, 0), (2, 1), (1, 2), (0, 3), . . .
If you think of the pairs as coordinates in the x-y plane, the pairs are enumerated along diagonals: first the
diagonal with pairs whose elements sum to 0, then the diagonal with pairs whose elements sum to 1, and so
on. This is often called a “dovetailing” argument, because if you imagine drawing a line that weaves back
and forth through the pairs enumerated this ways, it will be analogous to the a carpenter’s practice of using
a dovetail to join two pieces of wood. (And that term, in turn, comes from the similarity to a dove’s tail.)
As far as proofs go, the informal description above and the associated diagram are perfectly compelling. It
is possible to describe a bijection between N × N explicitly, however, in algebraic terms. You are asked to
do this in the exercises.
The previous theorem has a number of interesting consequences.
Theorem. If A and B are countable, then so is A × B.

Proof. If p is any element of N × N, write p0 and p1 to denote the two components. Let f : N → N × N be a
surjection, as guaranteed by the previous theorem. Suppose g : N → A and h : N → B be surjective. Then
the function k(i) = (g(f (i)0 ), h(f (i)1 )) is a surjective function from N to A × B.
Theorem. The set of rational numbers, Q, is countable.
Proof. By the previous theorem, we know that Z × Z is countable. Define f : Z × Z → Q by
{
i/j if j ̸= 0
f (i, j) =
0 otherwise
Since every element of Q can be written as i/j for some i and j in Z, f is surjective.
Theorem. Suppose that A is countable. For each n, the set An is countable.
Proof. Remember that we can identify the set of n-tuples of elements from A with A × . . . × A, where there
are n copies of A in the product. The result follows using induction on n.
Theorem. Let (A∪ i )i∈N be a family of sets indexed by the natural numbers, and suppose that each Ai is
countable. Then i Ai is countable.
Proof. Suppose for each i, fi is a∪surjective function from N to Ai . Then the function g(i, j) = fi (j) is a
surjective function from N × N to i Ai .
Theorem. Suppose that A is countable. Then the set of finite sequences of elements of A is countable.
∪
Proof. The set of finite sequences of elements of A is equal to i Ai , and we can apply the previous two
theorems.
Notice that the set of all alphanumeric characters and punctuation (say, represented as the set of all ASCII
characters) is finite. Together with the last theorem, this implies that there are only countably many
sentences in the English language (and, indeed, any language in which sentences are represented by finite
sequences of symbols, chosen from any countable stock).
At this stage, it might seem as though everything is countable. In the next section, we will see that this is
not the case: the set of real numbers, R, is not countable, and if A is any set (finite or infinite), the powerset
of A, P(A), is not equinumerous with A.

22.3 Cantor’s Theorem
A set A is uncountable if it is not countable. Our goal is to prove the following theorem, due to Georg
Cantor.
Theorem. The set of real numbers is uncountable.

Proof. Remember that [0, 1] denotes the closed interval {r ∈ R | 0 ≤ r ≤ 1}. It suffices to show that there
is no surjective function f : N → [0, 1], since if R were countable, [0, 1] would be countable too.
Recall that every real number r ∈ [0, 1] has a decimal expansion ∑∞of the form r = 0.r0 r1 r2 r3 r4 . . ., where each
ri is a digit in {0, 1, . . . , 9}. More formally, we can write r = i=0 10
ri
i for each r ∈ R with 0 ≤ r ≤ 1.
(Notice that 1 can be written 0.9999 . . .. In general every other rational number in [0, 1] will have two
representations of this form; for example, 0.5 = 0.5000 . . . = 0.49999 . . .. For concreteness, for these numbers
we can choose the representation that ends with zeros.)
As a result, we can write
• f (0) = 0.r00 r10 r20 r30 r40 . . .
• f (1) = 0.r01 r11 r21 r31 r41 . . .
• f (2) = 0.r02 r12 r22 r32 r42 . . .
• f (3) = 0.r03 r13 r23 r33 r43 . . .
• f (4) = 0.r04 r14 r24 r34 r44 . . .
• …
(We use superscripts, ri , to denote the digits of f (i). The superscripts do not mean the “ith power.”)
Our goal is to show that f is not surjective. To that end, define a new sequence of digits (ri )i∈N by
{
̸ 7
7 if rii =
ri =
3 otherwise.
The define the real number r = 0.r0 r1 r2 r3 . . .. Then, for each i, r differs from f (i) in the ith digit. But this
means that for every i, f (i) ̸= r. Since r is not in the range of f , and hence f is not surjective. Since f was
arbitrary, there is no surjective function from N to [0, 1].
(We chose the digits 3 and 7 only to avoid 0 and 9, to avoid the case where, for example, f (0) = 0.5000 . . .
and r = 0.4999 . . .. Since there are no zeros or nines in r, since the ith digit of r differs from f (i), it really
is a different real number.)
This remarkable proof is known as a “diagonalization argument.” We are trying to construct a real number
with a certain property, namely, that it is not in the range of f . We make a table of digits, in which the rows
represent infinitely many constraints we have to satisfy (namely, that for each i, f (i) ̸= r), and the columns
represent opportunities to satisfy that constraint (namely, by choosing the ith digit of r appropriately).
The complete the construction by stepping along the diagonal, using the ith opportunity to satisfy the ith
constraint. This technique is used often in logic and computability theory.
The following provides another example of an uncountable set.
Theorem. The power set of the natural numbers, P(N), is uncountable.
22.3. Cantor’s Theorem 203

Proof. Let f : N → P(N) be any function. Once again, our goal is to show that f is not surjective. Let S
be the set of natural numbers, defined as follows:
S = {n ∈ N | n ∈
/ f (n)}
In words, for every natural number, n, n is in S if and only if it is not in f (n). Then clearly for every n,
f (n) ̸= S. So f is not surjective.
We can also view this as a diagonalization argument: draw a table with rows and columns indexed by the
natural numbers, where the entry in the ith row and jth column is “yes” if j is an element of f (i), and “no”
otherwise. The set S is constructed by switching “yes” and “no” entries along the diagonal.
In fact, exactly the same argument yields the following:
Theorem. For every set A, there is no surjective function from A to P(A).

Proof. As above, if f is any function from A to P(A), the set S = {a ∈ A | a ∈
/ f (a)} is not in the range of
f.
This shows that there is an endless hierarchy of infinities. For example, in the sequence N, P(N), P(P(N)), . . .,
there is an injective function mapping each set into the next, but no surjective function. The union of all
those sets is even larger still, and then we can take the power set of that, and so on. Set theorists are still
today investigating the structure within this hierarchy.
22.4 An Alternative Definition of Finiteness
One thing that distinguishes the infinite from the finite is that an infinite set can have the same size as a
proper subset of itself. For example, the natural numbers, the set of even numbers, and the set of perfect
squares are all equinumerous, even though the latter two are strictly contained among the natural numbers.
In the nineteenth century, the mathematician Richard Dedekind used this curious property to define what
it means to be finite. We can show that his definition is equivalent to ours, but the proof requires the axiom
of choice.
Definition. A set is A Dedekind infinite if A is equinumerous with a proper subset of itself, and Dedekind
finite otherwise.
Theorem. A set is Dedekind infinite if and only it is infinite.
Proof. Suppose A is Dedekind infinite. We need to show it is not finite; suppose, to the contrary, it is
bijective with [n] for some n. Composing bijections, we have that [n] is bijective with a proper subset of
itself. This means that there is an injective function f from [n] to a proper subset of n. Modifying f , we
can get an injective function from [n] into [n − 1], contradicting the pigeonhole principle.
Suppose, on the other hand, that A is infinite. We need to show that there is an injective function f from A
to a proper subset of itself (because then f is a bijection between A and the range of f ). Choose a sequence
of distinct element a0 , a1 , a2 , . . . of A. Let f map each ai to ai+1 , but leave every other element of A fixed.
Then f is injective, but a0 is not in the range of f , as required.

22.5 The Cantor-Bernstein Theorem
Saying that A and B are equinumerous means, intuitively, that A and B have the same side. There is also
a natural way of saying that A is not larger than B:
Definition. For two sets A and B, we say the cardinality of A is less than or equal to the cardinality of B,
written A ⪯ B, when there is an injection f : A → B.
As an exercise, we ask you to show that ⪯ is a preorder, which is to say, it is reflexive and transitive. Here
is a natural question: does A ⪯ B and B ⪯ A imply A ≈ B? In other words, assuming there are injective
functions f : A → B and g : B → A, is there necessarily a bijection from A to B?
The answer is “yes,” but the proof is tricky. The result is known as the Cantor-Bernstein Theorem, and we
state it without proof.
Theorem. For any sets A and B, if A ⪯ B and B ⪯ A, then A ≈ B.
22.6 Exercises
1. Show that equinumerosity is reflexive, symmetric, and transitive.

2. Show that the function f (x) = x/(1 − x) is a bijection between the interval [0, 1) and R≥0 .
3. Show that the g(x) = x/(1 − |x|) gives a bijection between (−1, 1) and R.
(i+j)(i+j+1)
4. Define a function J : N × N → N by J(i, j) = 2 + i. This goal of this problem is to show
that J is a bijection from N × N to N.
(a) Draw a picture indicating which pairs are sent to 0, 1, 2, . . ..
(b) Let n = i + j. Show that J(i, j) is equal the number of pairs (u, v) such that either u + v < n, or
u + v = n and u < i. (Use the fact that 1 + 2 + . . . + n = n(n + 1)/2.)
(c) Conclude that J is surjective: to find i and j such that J(i, j) = k, it suffices to find the largest
n such that n(n + 1)/2 ≤ k, let i = k − n(n + 1)/2, and let j = n − i.
(d) Conclude that J is injective: if J(i, j) = J(i′ , j ′ ), let n = i + j and n′ = i′ + j ′ . Argue that n = n′ ,
and so i = i′ and j = j ′ .
5. Let S be the set of functions from N to {0, 1}. Use a diagonal argument to show that S is uncountable.
(Notice that you can think of a function f : N → {0, 1} as an infinite sequence of 0’s and 1’s, given by
f (0), f (1), f (2), . . .. So, given a function F (n) which, for each natural number n, returns an infinite
sequence of 0’s and 1’s, you need to find a sequence that is not in the image of F .)
6. If f and g are functions from N to N, say that g eventually dominates f if there is some n such that
for every m ≥ n, g(m) > f (m). In other words, from some point on, g is bigger than f .
Show that if f0 , f1 , f2 , . . . is any sequence of functions from N to N, indexed by the natural numbers,
then there is a function g that eventually dominates each fi . (Hint: construct g so that for each i,
g(n) > fi (n) for every n ≥ i.)
7. Show that the relation ⪯ defined in Section 22.5 is reflexive and transitive.
22.5. The Cantor-Bernstein Theorem 205


CHAPTER
TWENTYTHREE
AXIOMATIC FOUNDATIONS
In this final chapter, our story comes full circle. We started our journey with symbolic logic, using the
propositional connectives to model logical terms like “and,” “or,” “not,” and “implies.” To that we added
the quantifiers and function and relation symbols of first-order logic. From there, we moved to sets, functions,
and relations, which are ubiquitous in modern mathematics; the natural numbers and induction; and then
topics such as number theory, combinatorics, the real numbers, and the infinite. Here we return to symbolic
logic, and see how it can be used to provide a formal foundation for all of mathematics.
Specifically, we will consider an axiomatic framework known as Zermelo-Fraenkel set theory, which was
introduced early in the twentieth century. In the set-theoretic view of mathematics, every mathematical
object is a set. The axioms assert the existence of sets with various properties. From the collection of all
sets, we carve out the usual inhabitants of the mathematical universe, not just the various number systems
we have considered, but also pairs, finite sequences, relations, functions, and so on. This provides us with
an idealized foundation for everything we have done since Chapter 11.
At the end of this chapter, we will briefly describe another axiomatic framework, dependent type theory,
which is the one used by Lean. We will see that it provides an alternative perspective on mathematical
objects and constructions, but one which is nonetheless inter-interpretable with the set-theoretic point of
view.
23.1 Basic Axioms for Sets
The axioms of set theory are expressed in first-order logic, for a language with a single binary relation symbol,
∈. We think of the entire mathematical universe as consisting of nothing but sets; if x and y are sets, we
can express that x is an element of y by writing x ∈ y. The first axiom says that two sets are equal if and
only if they have the same elements:
Extensionality: ∀x, y (x = y ↔ ∀z(z ∈ x ↔ z ∈ y))
The next axiom tells us that there is at least one interesting set in the universe, namely, the set with no
element:
Empty set: ∃x ∀y y ∈
/x
Here, if course, x ∈/ y abbreviates ¬(x ∈ y). By the axiom of extensionality, the set asserted to exist by this
axiom is unique: in other words, if x1 and x2 each has no elements, then, vacuously, any element is in one if
and only if it is in the other, so x1 = x2 . This justifies using the word the in the phrase the empty set. Given
this fact, it should seem harmless to introduce a new symbol, ∅, to denote the set matching that description.
Indeed, one can show that this is case: in a precise sense, such expansions to a first-order language can be
viewed as nothing more than a convenient manner of expression, and statements in the bigger language can
be translated to the original language in a way that justifies all the expected inferences. We will not go into
207
the details here, and, rather, take this fact for granted. Using the new symbol, the empty set axiom tells us
the empty set satisfies the following property:
∀y y ∈
/∅
The third axiom tells us that given two sets x and y, we can form a new set z whose elements are exactly x
and y:
Pairing: ∀x, y ∃z ∀w (w ∈ z ↔ w = x ∨ w = y)
There is a stealth usage of this axiom lurking nearby. The axiom does not require that x and y are different,
so, for example, we can take them both to be the empty set. This tells us that the set {∅}, whose only
element is the empty set, exists. More generally, the axiom tells us that for any x, we have the set {x}
whose only element is x, and for any x and y, we have {x, y}, as described above. Once again, the axiom of
extensionality tells us that the sets meeting these descriptions are unique, so it is fair to use the corresponding
notation. We are now off and running! We now have all of the following sets, and more:
∅, {∅}, {{∅}}, {∅, {∅}}, {{{∅}}}, . . .
Still, we can never form a set with more than two elements in this way. To that end, it would be reasonable
to add an axiom∪that asserts for every x and y, the set x ∪ y exists. But we can do better. Remember that
if
∪ x is any set, x denotes the union of all the sets in x. In other words, for any set z, z is an element of
x if and only if z is in w for some set w in x. The following axiom asserts that this set exists:
Union: ∀x ∃y ∀z (z ∈ y ↔ ∃w (w ∈ x ∧ z ∈ w))
∪
Once again, this justifies the use of the notation.
∪ We get the ordinary binary union using this axiom
together with pairing, since we have x ∪ y = {x, y}.
At this stage, it will be useful to invoke some additional notation that was first introduced in our informal
presentation of sets. If A is any first-order formula in the language of set theory, ∀x ∈ y A abbreviates
∀x (x ∈ y → A) and ∃x ∈ y A abbreviates ∃x (x ∈ y ∧ A), relativizing the quantifiers as described in Section
7.4. The expression x ⊆ y abbreviates ∀z ∈ x (z ∈ y), as you would expect.
The next axiom asserts that for every set x, the power set, P(x) exists.
Power Set: ∀x ∃y ∀z (z ∈ y ↔ z ⊆ x)
We have begun to populate the universe with basic set constructions. It is the next axiom, however, that
gives set theory its remarkable flexibility. Properly speaking, it is not a single axiom, but a schema, an
infinite family of axioms given by a single template. The schema is meant to justify set-builder notation
{w | . . .} that was ubiquitous in Chapter 11. The first question we need to address is what we are allowed to
write in place of the “….” In our informal presentation of set theory, we said that one can define a set using
any property, but that only begs the question here as to what counts as a “property.” Axiomatic set theory
provides a simple but powerful answer: we can use any first-order formula in the language of set theory.
Another concern centers around Russell’s paradox, as discussed in Section 11.1. Any theory that allows us
to define the set {w | w ∈
/ w} is inconsistent, since if we call this set z, we can show z ∈ z if and only if z ∈
/ z,
which is a contradiction. Once again, set theory offers a simple and elegant solution: for any formula A(z)
and set y, we can instead form the set {w ∈ y | A(w)}, consisting of the elements of y that satisfy A. In
other words, we have to first use the other axioms of set theory to form a set y that is big enough to include
all the elements that we want to consider, and then use the formula A to pick out the ones we want.
The axiom schema we want is called separation, because we use it to separate the elements we want from
those in a bigger collection.
Separation: ∀x1 , x2 , . . . , xn , y ∃z ∀w (w ∈ z ↔ w ∈ y ∧ A(w, x1 , x2 , . . . , xn ))
208 Chapter 23. Axiomatic Foundations

Here, A can be any formula, and the list of variables x1 , . . . , xn that are shown indicate that the formula A
can have some parameters, in which case the set we form depends on this values. For example, in ordinary
mathematics, given a number m we can form the set {n ∈ N | prime(n) ∧ n > m}. In this example, the
description involves m and n, and the set so defined depends on m.
We could use the separation axiom to simplify the previous axioms. For example, as long as we know that
any set x exists, we can define the empty set as {y ∈ x | ⊥}. Similarly, in the pairing axiom, it is enough to
assert that there is a set that contains x and y as element, because then we can use separation to carve out
the set whose elements are exactly x and y.
These are only the first six axioms of set theory; we have four more to go. But these axioms alone provide
a foundation for reasoning about sets, relations, and functions, as we did in Chapter 11, Chapter 13, and
Chapter 15. For example, we have already defined the union operation, and we can define set intersection
x ∩ y as {z ∈ x ∪ y | z ∈ x ∧ z ∈ y}. We cannot define arbitrary set complements; for example, the
exercises as you to show that in set theory we can prove that there is no set that contains all sets, and so the
complement of the empty set does not exist. But given any two sets x and y, we can define their difference
x \ y as {z ∈ x | z ∈/ y}. The exercises below ask you to show that we can also define indexed unions and
intersections, once we have developed the notion of a function.
We would like to define a binary relation between two sets x and y to be a subset of x × y, but we first have
to define the cartesian product x × y. Remember that in Section 11.4 we defined the ordered pair (u, v) to
be the set {{u}, {u, v}}. As a result, we can use the separation axiom to define
x × y = {z ∈ . . . | ∃u ∈ x ∃v ∈ y (z = (u, v))}
provided we can prove the existence of a set big enough to fill the “….” In the exercises below, we ask you to
show that the set P(P(x ∪ y)) contains all the relevant ordered pairs. A binary relation r on x and y is then
just a subset of x × y, where we interpret r(u, v) as (u, v) ∈ r. We can think of ordered triples from the sets
x, y, z as elements of x × (y × z) and so on. This gives us ternary relations, four-place relations, and so on.
Now we can say that a function f : x → y is really a binary relation satisfying ∀u ∈ x ∃!v ∈ y f (u, v), and
we write f (u) = v when v is the unique element satisfying f (u, v). A function f taking arguments from sets
x, y, and z and returning an element of w can be interpreted as a function f : x × y × z → w, and so on.
With sets, relations, and functions, we have the basic infrastructure we need to do mathematics. All we are
missing at this point are some interesting sets and structures to work with. For example, it would be nice to
have a set of natural numbers, N, with all the properties we expect it to have. So let us turn to that next.
23.2 The Axiom of Infinity
With the axioms we have so far, we can form lots of finite sets, starting with ∅ and iterating pairing, union,
powerset, and separation constructions. This will give us sets like
∅, {∅}, {{∅}}, {∅, {∅}}, {{{∅}}}, . . .
But none of the axioms so far preclude the possibility that there are any more sets than that. In particular,
none of the axioms gives us an infinite set. So we need a further axiom to tell us that such a set exists.
Remember that in Chapter 17 we characterized the natural numbers as a set with a distinguished element,
0, and an injective, operation succ, satisfying the principles of induction and recursive definition. In set
theory, everything is a set, so if we want to represent the natural numbers in that framework, we need to
identify them with particular sets. There is a natural choice for 0, namely, the empty set, ∅. For a successor
operation, we will use the function succ defined by succ(x) = x ∪ {x}. The choice is a bit of a hack; the
best justification for the definition is that it works. With this definition, the first few natural numbers are
as follows:
0 = ∅, 1 = {∅}, 2 = {∅, {∅}}, 3 = {∅, {∅}, {∅, {∅}}}, . . .
23.2. The Axiom of Infinity 209

It is more perspicuous to write them as follows:
0 = ∅, 1 = {0}, 2 = {0, 1}, 3 = {0, 1, 2}, 4 = {0, 1, 2, 3}, . . .
In general, n + 1 is represented by the set {0, 1, . . . , n}, in which case, m ∈ n is the same as m < n. This is
just an incidental property of our encoding, but it is a rather charming one.
Recall from Chapter 17 that we can characterize the set of natural numbers as follows:
• There is an element 0 ∈ N and there is an injective function succ : N → N, with the additional property
that succ(x) ̸= 0 for any x in N.
• The set N satisfies the principle of induction: if x is a subset of N that contains 0 and is closed under
succ (that is, whenever z is in N, so is succ), then x = N.
We have already settled on the definitions of 0 and succ, but we don’t yet have any set that contains the
first and is closed under applying the second. The axiom of infinity asserts precisely that there exists such
a set.
Infinity: ∃x (∅ ∈ x ∧ ∀y (y ∈ x → y ∪ {y} ∈ x))
Say a set x is inductive if it satisfies the property after the existential quantifier, namely, that it contains
the empty set and is closed under our successor operation. Notice that the set of natural numbers, which
we are still trying to define formally, has this property. The axiom of infinity asserts the existence of some
inductive set, but not necessarily the natural numbers themselves; an inductive set can have other things in
it as well. In a sense, the principle of induction says that the natural numbers is the smallest inductive set.
So we need a way to separate that set from the one asserted to exist by the axiom of infinity.
Let x be any inductive set, as asserted to exist by the axiom of infinity. Let
∩
y = {z ⊆ x | x is inductive}
Here z ⊆ x can also be written z ∈ P(x), so the inside set exists using the separation axiom. According to
this definition, y is the intersection of every inductive set, so an element w is in x if and only if w is in every
inductive set. We claim that y itself is inductive. First, we have ∅ ∈ y, since the empty set is an element of
every inductive set. Next, suppose w is in y. Then w is in every inductive set. But since every inductive set
is closed under successor, succ(w) is in every inductive set. So succ(w) is in the intersection of all inductive
sets — which is y!
The more interesting point is that y also satisfies the principle of induction. To see this, suppose u contains
the empty set and is closed under succ. Then u is inductive, and since y is the intersection of all inductive
sets, we have y ⊆ u. Since we assumed u ⊆ y, we have u = y, which is what we want.
To summarize, then, we have proved the existence of a set that contains 0 and is closed under a successor
operation and satisfies the induction axiom. Moreover, there is only one such set: if y1 and y2 both have
this property, then so does y1 ∩ y2 , and by the induction principle, this intersection has to be equal to both
y1 and y2 , in which case y1 and y2 are equal. It then makes sense to call the unique set with these properties
the natural numbers, and denote it by the symbol N.
There is only one piece of the puzzle missing. It is clear from the definition that 0 is not the successor of
any number, but it is not clear that the successor function is injective. We can prove that by first noticing
that the natural numbers, as we have defined them, have a peculiar property: if z is a natural number, y is
an element of z, and x is an element of y, then x is an element of z. This says exactly that the ∈ relation is
transitive on natural numbers, which is not surprising, since we have noted that ∈ on the natural numbers,
under our representation, coincides with <. To prove this claim formally, say that a set z is transitive if it
has the property just mentioned, namely, that every element of an element of z is an element of z. This is
equivalent to saying that for every y ∈ z, we have y ⊆ z.

Lemma. Every natural number is transitive.

Proof. By induction on the natural numbers. Clearly, ∅ is transitive. Suppose x is transitive, and suppose
y ∈ succ(x) and z ∈ y. Since succ(x) = x ∪ {x}, we have y ∈ x or y ∈ {x}. If y ∈ x, then by the inductive
hypothesis, we have z ∈ x, and hence z ∈ succ(x). Otherwise, we have y ∈ {x}, and so y = x. In that case,
again we have z ∈ x, and hence z ∈ succ(x).
The next lemma shows that, on transitive sets, union acts like the predecessor operation.
∪
Lemma. If x is transitive, then succ(x) = x.
∪ ∪
Proof. Suppose y is in succ(x) = (x ∪ {x}). Then either y ∈ z for some z ∈ x, or y ∈ x. In the first case,
also have y ∈ x, since x is transitive.
∪
Conversely, suppose y is in x. Then y is in succ(x), since we have x ∈ succ(x).
Theorem. succ is injective on N.
∪ Suppose x∪and y are in N, and succ(x) = succ(y). Then x and y are both transitive, and we have
Proof.
x = succ(x) = succ(y) = y.
With that, we are off and running. Although we will not present the details here, using the principle of
induction we can justify the principle of recursive definition. We can then go on to define the basic operations
of arithmetic and derive their properties, as done in Chapter 17. We can go on to define the integers, the
rational numbers, and the real numbers, as described in Chapter Chapter 21, and to develop subjects like
number theory and combinatorics, as described in Chapters Chapter 19 and Chapter 20. In fact, it seems
that any reasonable branch of mathematics can be developed formally on the basis of axiomatic set theory.
There are pitfalls, for example, having to do with large collections: for example, just as it is inconsistent to
postulate the existence of a set of all sets, in the same way, there is no collection of all partial orders, or all
groups. So when interpreting some mathematical claims, care has to be taken in some cases to restrict to
sufficiently large collections of such objects. But this rarely amounts to more than careful bookkeeping, and
it is a remarkable fact that, for the most part, the axioms of set theory are flexible and powerful enough to
justify most ordinary mathematical constructions.
23.3 The Remaining Axioms
The seven axioms we have seen are quite powerful, and suffice to represent large portions of mathematics.
We discuss the remaining axioms of Zermelo-Fraenkel set theory here.
So far, none of the axioms we have seen rule out the possibility that a set x can be an element of itself, that
is, that we can have x ∈ x. The following axiom precludes that:
Foundation ∀x (∃y y ∈ x → ∃y (y ∈ x ∧ ¬∃z (z ∈ x ∧ z ∈ y)))
The axiom says that if x is a nonempty set, there is an element y of x with the property that no element of
y is again an element of x. This implies we cannot have a descending chain of sets, each one an element of
the one before:
x1 ∋ x2 ∋ x3 ∋ . . .
If we apply the axiom of foundation to the set {x1 , x2 , x3 , . . .}, we find that some element xi does not contain
any others, which is only possible if the sequence has terminated with xi . In other words, the axiom implies
23.3. The Remaining Axioms 211

(and is in fact equivalent to) the statement that the elementhood relation is well founded, which explains
the name.
The axioms listed in the previous section tell a story of how sets come to be: we start with the empty set,
and keep applying constructions like power set, union, and separation, to build more sets. Set theorists often
imagine the hierarchy of sets as forming a big V, with the empty set at the bottom and a set at any higher
level comprising, as its elements, sets that appear in levels below. In a precise sense (which we will not spell
out here), the axiom of foundation says that every set arises in such a way.
Now consider the following sequence of sets:
∅, P(∅), P(P(∅)), P(P(P(∅))), . . .
It is consistent with all the axioms we have seen so far that every set in the mathematical universe is an
element of one of these. That still gives us a lot of sets, but, since we have described that sequence, we can
just as well imagine a set that contains all of them:
{∅, P(∅), P(P(∅)), P(P(P(∅))), . . .}
The following axiom implies the existence of such a set.
Regularity: ∀x, y1 , . . . , yn (∀z ∈ x ∃!w A(z, w, y1 , . . . , yn ) →

∃u ∀z ∃w (w ∈ u ∧ A(z, w, y1 , . . . , yn )))
Like the axiom of separation, this axiom is really a schema, which is to say, a separate axiom for each formula
A. Here, too, the variables y1 , y2 , . . . , yn are free variables that can occur in A. To understand the axiom,
it is easiest to think of them as parameters that are fixed in the background, and then ignore them. The
axioms says that if, for every z in x, there is a unique w satisfying A(z, w), then there is a single set, u, that
is big enough to contain a w for every such z. In other words, if you think of A as a function whose domain
is x, the axiom asserts that there is a set big enough to include its range. In the example above, x is the
natural numbers, and A(z, w) says that w is the z-fold iterate of the power set of the empty set.
The nine axioms we have listed so far comprise what is known as Zermelo-Fraenkel Set Theory. There is
on additional axiom, the axiom of choice, which is usually listed separately for historical reasons: it was
once considered controversial, and in the early days, mathematicians considered it important to keep track
of whether the axiom was actually used in a proof. There are many equivalent formulations, but this one is
one of the most straightforward:
∪
Choice: ∀x (∅ ∈/ x → ∃f : x → x ∀y ∈ x f (y) ∈ y)
The axiom says that for any collection x of nonempty sets, there is a function f that selects an element from
each one. We used this axiom, informally, in Section 15.2 to show that every surjective function has a right
inverse. In fact, this last statement can be shown to be equivalent to the axiom of choice on the basis of the
other axioms.
To summarize, then, the axioms of Zermelo-Fraenkel Set Theory with the axiom of choice are as follows:
1. Extensionality:
∀x, y (x = y ↔ ∀z(z ∈ x ↔ z ∈ y))
2. Empty set:
∃x ∀y y ∈
/x

3. Pairing:
∀x, y ∃z ∀w (w ∈ z ↔ w = x ∨ w = y)
4. Union:
∀x ∃y ∀z (z ∈ y ↔ ∃w (w ∈ x ∧ z ∈ w))
5. Power set:
∀x ∃y ∀z (z ∈ y ↔ z ⊆ y
6. Separation:
∀x1 , x2 , . . . , xn , y ∃z ∀w (w ∈ z ↔ w ∈ y ∧ A(w, x1 , x2 , . . . , xn ))
7. Infinity:
∃x (∅ ∈ x ∧ ∀y (y ∈ x → y ∪ {y} ∈ x)
8. Foundation:
∀x (∃y y ∈ x → ∃y (y ∈ x ∧ ¬∃z (z ∈ x ∧ z ∈ y)))
9. Regularity:
∀x, y1 , . . . , yn (∀z ∈ x ∃!w A(z, w, y1 , . . . , yn ) →

∃u ∀z ∃w (w ∈ u ∧ A(z, w, y1 , . . . , yn )))
10. Choice:
∪
∀x (∅ ∈
/ x → ∃f : x → x ∀y ∈ x f (y) ∈ y)
23.4 Type Theory
As a foundation for mathematics, Zermelo-Fraenkel set theory is appealing. The underlying logic, first-order
logic, provides the basic logical framework for quantifiers and the logical connectives. On top of that, the
theory describes a single, intuitively natural concept, that of a set of elements. The axioms are plausible
23.4. Type Theory 213

eminently reasonable. It is remarkable that virtually all of modern mathematics can be reduced to such
simple terms.
There are other foundations on offer, however. These tend to be largely inter-interpretable with set theory.
After all, set-theoretic language is now ubiquitous in everyday mathematics, so any reasonable foundation
should be able to make sense of such language. On the other hand, we have already noted that set-theory
is remarkably expressive and robust, and so it should not be surprising that other foundational approaches
can often be understood in set-theoretic terms.
This is, in particular, true of dependent type theory, which is the basis of the Lean theorem prover. The
syntax of type theory is more complicated than that of set theory. In set theory, there is only one kind of
object; officially, everything is a set. In contrast, in type theory, every well-formed expression in Lean has a
type, and there is a rich vocabulary of defining types.
In fact, Lean is based on a version of an axiomatic framework known as the Calculus of Inductive Construc-
tions, which provides all of the following:
• A hierarchy of type universes, Type 0, Type 1, Type 2, … and a special type Prop. The expression
Type abbreviates Type 0, and saying T : Type can be interpreted as saying that T is a datatype. The
type Prop is the type of propositions.
• Dependent function types Π x : A, B x. An element f of this type is a function which maps any
element a of type A to and element f a of type B a. The fact that the type of the output depends on
the type of the input is what makes the function “dependent.” In the case where the output type does
not depend on the input, we have the usual function type A → B.
• Inductive types, like the natural numbers, specified by its constructors, like zero and successor. Each
such type comes with principles of induction and recursion.
These constructions account for both the underlying logic of assertions (that is, the propositions) as well as
the objects of the universe, which are elements of the ordinary types.
It is straightforward to interpret type theory in set theory, since we can view each type as a set. The
type universes are simply large collections of sets, and dependent function types and inductive types can be
explained in terms of set-theoretic constructions. We can view Prop as the set {⊤, ⊥} of truth values, just
as we did when we described truth-table semantics for propositional logic.
Given this last fact, why not just use set theory instead of type theory for interactive theorem proving?
Some interactive theorem provers do just that. But type theory has some advantages:
• The fact that the rules for forming expressions are so rigid makes it easier for the system to recognize
typographical errors and provide useful feedback. In type theory, if f has type N → N it can be
applied only to a natural number, and a theorem prover can flag an error if the argument has the
wrong type. In set theory, anything can be applied to anything, whether or not doing so really makes
sense.
• Again, because the rules for forming expressions are so rigid, the system can infer useful information
from the components of an expression, whereas set theory would require us to make such information
explicit. For example, with f as above, a theorem prover can infer that a variable x in f x should have
type N, and that the resulting expression again has type N. In set theory, x ∈ N has to be stated as
an explicit hypothesis, and f (x) ∈ N is then a theorem.
• By encoding propositions as certain kinds of types, we can use the same language for defining mathe-
matical objects and writing mathematical proofs. For example, we can apply a function to an argument
in the same way we apply a theorem to some hypotheses.
• Expressions in a sufficiently pure part of dependent type theory have a computational interpretation,
so, for example, the logical framework tells us how to evaluate the factorial function, given its definition.
In set theory, the computational interpretation is specified independently, after the fact.

These facts hark back to the separation of concerns that we raised in Chapter 1: different axiomatic founda-
tions provide different idealized descriptions of mathematical activity, and can be designed to serve different
purposes. If you want a clean, simple theory that accounts for the vast majority of mathematical proof,
set theory is hard to beat. If you are looking for a foundation that makes computation central or takes
the notion of a function rather than a set as basic, various flavors of type theory have their charms. For
interactive theorem proving, pragmatic issues regarding implementation and usability come into play. What
is important to recognize is that what all these idealized descriptions have in common is that they are all
designed to model important aspects of mathematical language and proof. Our goal here has been to help
you reflect on those features of mathematical language and proof that give mathematics its special character,
and to help you better understand how they work.
23.5 Exercises
1. Use an argument similar Russell’s paradox to show that there is no “set of all sets,” that is, there is
no set that contains every other set as an element.
2. Suppose∩x is a nonempty set, say, containing an element y. Use∩the axiom of separation to show that
the set y exists. (Remember that something is an element of y if it is an element of every element
of y.
3. Justify the claim in Section 23.1 that every element of x × y is an element of P(P(x ∪ y)).
∪
4. Given a set x and a function A : x → y, use the axioms of set theory to prove the existence of i∈x A(i).
23.5. Exercises 215

Logic and Proof

Uploaded by

Copyright:

Available Formats

Logic and Proof

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Logic and Proof

Uploaded by

Copyright:

Available Formats

What are the main topics covered in the textbook?

What are the main topics covered in the textbook?

How is logic formally represented and reasoned about in the textbook?

How is logic formally represented and reasoned about in the textbook?

Logic and Proof

Jeremy Avigad, Robert Y. Lewis, and Floris van Doorn

Jan 27, 2018

3 Natural Deduction for Propositional Logic 17

4 Propositional Logic in Lean 25

6 Semantics of Propositional Logic 47

7 First Order Logic 53

8 Natural Deduction for First Order Logic 61

9 First Order Logic in Lean 69

10 Semantics of First Order Logic 85

12 Sets in Lean 105

14 Relations in Lean 123

16 Functions in Lean 137

17 The Natural Numbers and Induction 149

18 The Natural Numbers and Induction in Lean 165

19 Elementary Number Theory 171

21 The Real Numbers 191

23 Axiomatic Foundations 207

1.1 Mathematical Proof

In the next example, we focus on the natural numbers,

a product of primes, as required.

1.2 Symbolic Logic

Every man is an animal.

1.2. Symbolic Logic 3

1.3 Interactive Theorem Proving

theorem sqrt_two_irrational {a b : N} (co : gcd a b = 1) :

have 2 * (2 * c^2) = 2 * b^2,

1.4 The Semantic Point of View

1.4. The Semantic Point of View 5

1.5 Goals Summarized

To summarize, these are the goals of this course:

1.6 About this Textbook

2.3 Rules of Inference

8 Chapter 2. Propositional Logic

Suppose Alice’s husband was in the bar.

Alice’s brother was the victim.

2.3. Rules of Inference 9

Alice’s husband was the killer.

In symbols, these patterns are rendered as follows:

2.3.3 Negation and Falsity

Suppose Alice’s husband was in the bar.

10 Chapter 2. Propositional Logic

In symbolic logic, the rule of inference is expressed as follows:

If 2 is prime and greater than 2, then 2 is odd.

2.3. Rules of Inference 11

Alice’s daughter was alone at the time of the murder.

In symbolic terms, the two introduction rules are as follows:

In symbols, this pattern is expressed as follows:

In symbols, we would render this rule as follows:

12 Chapter 2. Propositional Logic

2.3.5 If and only if

Alice is in the bar if and only if Alice’s husband is on the beach.

The “right” rule simply runs in the opposite direction.

2.3. Rules of Inference 13

Alice is in the bar if and only if Alice’s husband is on the beach.

Rendered in natural deduction, the rules are as follows: