CHAPTER 2.

FIRST ORDER LOGIC

1. Introduction
First order logic is a much richer system than sentential logic. Its interpre-
tations include the usual structures of mathematics, and its sentences enable
us to express many properties of these structures. For example, consider the
following English sentence:
Everything greater than 0 has a square root.
This can be interpreted, for example, in arithmetic on Q, the set of rational
numbers, where it is false, and in arithmetic on R, the set of real numbers,
where it is true. How could we express this in a formal language? The
following is at least a first approximation.
For all x (0 < x → there is some y such that x = y · y).
A formal language that could express this would need symbols for
• quantifiers for all and there exists,
• variables, like x and y, which range over the elements of the set we
are talking about,
• functions, like ·,
• relations, like = and <,
• constants, like 0, which name fixed elements of the domain,
• sentential connectives,
• parentheses.
We will use the symbols ∀ for for all and ∃ for there exists. Thus the
above sentence could be written as

∀x 0 < x → ∃y(x = y · y)
We will normally take formalization one step further. We allow inter-
pretations which interpret <, for example, as any binary relation on the
domain, not just something that “looks” like an order. To emphasize this
we use neutral symbols, like R, F , and c for the relations, functions, and
constants of the language. The only exception is = whose interpretation
is fixed. Thus, fully formalized, the sentence we are looking at would be
written as

∀x R(c, x) → ∃y(x = F (y, y))
An interpretation for this would specify what set of objects is referred to
(the domain of the interpretation), and how to interpret the symbols c, R, F .
One such interpretation has domain Z, the set of all integers, and interprets
c as 2, R(x, y) as “x divides y”, and F as +. Note that the sentence is true
in this interpretation.
Note that this sentence is built from pieces which are not themselves
sentences, for example ∃y(x = F (y, y)). In any interpretation this defines
1
2 CHAPTER 2. FIRST ORDER LOGIC

a subset of the domain, namely the set of all x in the domain such that
x = F (y, y) for some y in the domain. In the interpretation with domain R
with c, R, F interpreted as 0, < .·, this is the set of all non-negative reals. In
the interpretation in the preceding paragraph, this is the set of even integers.
Such an expression will be called a formula, and we will need to know how
to interpret formulas before defining when a sentence is true.

2. Functions, Relations, and Structures

Let A be a non-empty set and let n ≥ 1. An n-ary function f on A assigns
to every n-tuple a1 , . . . , an from A a unique element f (a1 , . . . , an ) ∈ A. Note,
for example, that subtraction is not considered a binary function on N since
3−7 ∈ / N.
If F is an n-ary function on A and G1 , . . . , Gn are each k-ary func-
tions on A then we can define a k-ary function H on A by composition as
H(a1 , . . . , ak ) = F (G1 (a1 , . . . , ak ), . . . , Gn (a1 , . . . , ak )) for all a1 , . . . , ak ∈ A.
For example, from + and · on N we can define (x + y) · (x · y), x · y + y · x, etc.
If the functions G1 , . . . , Gn do not all have the same arity or variable list,
we add dummy variables until they do. Thus we can also obtain x · (y + x),
x · x + y · y, etc.
An n-ary relation R on A picks out a set of n-tuples of elements of A,
those for which the relation holds, written R(a1 , . . . , an ) holds. For all
other n-tuples the relation fails, written R(a1 , . . . , an ) fails. Note that a
unary relation picks out a subset of A. One specific relation of importance
is the binary relation of equality on A , =, which holds between a pair of
elements of A precisely when they are the same.
A structure consists of some non-empty set A, called the domain or uni-
verse of the structure, together with some collection of functions and re-
lations on A and some specific “distinguished” elements of A. We always
assume that = is among the relations, but we allow the cases where there
are no functions and/or no other relations and/or no distinguished elements.
We use A, B etc. to refer to structures. By convention the universe of a
structure, unless otherwise specified, is the coresponding Latin letter.
So the simplest structures consist just of a non-empty set and = on that
set. Most of the structures studied in mathematics can be viewed as struc-
tures in this sense. For example, arithmetic on the integers studies the
structure with universe Z together with +, ·, −, <, =, and perhaps 0 and 1
as distinguished elements. We normally do not mention = since it is always
present, and we frequently write the structure as a tuple. For example the
structure just mentioned could be referred to as
A = (Z, +, ·, −, <, 0, 1).
Since − is definable from + (k − n = l iff l + n = k) we could remove it from
the function list without changing what can be expressed. Similarly, both 0
and 1 are inessential luxuries.
 CHAPTER 2. FIRST ORDER LOGIC 3

3. First Order Languages

Unlike sentential logic, there are many languages for first order logic,
depending on the number and sort of functions, relations, and constants
considered.

Definition 3.1. The symbols of a first order language L are as follows:

(i) some collection of symbols for functions, each of specified arity;
(ii) some collection of symbols for relations, each of specified arity and
including the binary relation symbol =;
(iii) some collection of symbols for constants;
(iv) an infinite set of variables: v0 , v1 , . . . , vn , . . . for all n ∈ N;
(v) the quantifiers ∀ and ∃;
(vi) the sentential connectives: ¬, ∧, ∨, →:
(vii) parentheses and comma: ( , ).

We emphasize that a language may contain no function symbols, no con-

stants, and no relation symbols other than =. We normally use F , G, H for
function symbols, P , Q, R, S for relation symbols, and c, d, e for constants.
In each case, there may also be either sub- or superscripts, for example F ′ ,
R∗ , c1 . Note also that each function symbol and each relation symbol has
a specified arity, although our notation does not show this. A language L
is determined by the function symbols, relation symbols other than =, and
constants that it contains. This set is frequently called the set of non-logical
symbols of L and is written as Lnl .
Just as we used A, B, C to refer to arbitrary atomic sentences Si , we will
x, y, and z (perhaps with sub- or superscripts) to refer to arbitrary variables
vi .

Definition 3.2. A structure for L, or simply an L-structure, is a structure

A which contains an n-ary function F A for every n-ary function symbol F
of L, an n-ary relation RA for every n-ary relation symbol R of L, and a
distinguished element cA for every constant c of L, but no other functions,
relations, or named elements.

For example, if the non-logical symbols of L are a unary function symbol

F and a constant c then an L-structure would be some A = (A, F A , cA )
where A is some non-empty set, F A is some unary function on A, and
cA ∈ A. Examples of such structures include (Z, s, 0) where s(k) = k + 1
for all k ∈ Z, (N+ , f, 1) where f (k) = k2 for all k ∈ N+ , and (Q, g, 12 ) where
g(k) = k2 for all k ∈ Q. N+ = {n ∈ N|n > 0}.
As another example, if the only non-logical symbol of L is a unary relation
symbol P then an L-structure would be some A = (A, P A ) where P A is some
unary relation on, i.e. a subset of, A. If A is finite with n elements then
there are 2n choices for P A . If A is infinite then there are infinitely many,
in fact uncountably many, choices for P A .
4 CHAPTER 2. FIRST ORDER LOGIC

4. Terms, Formulas, and Sentences

In sentential logic, the only meaningful expressions were the sentences.
In first order logic we will also have terms and formulas. Terms will define
functions (or perhaps individual elements) in any structure, and formulas
will define relations in any structure. We first discuss terms via an example.
Consider the language L whose only non-logical symbols are a unary
function symbol F and a constant c. L-structures include A = (Z, s, 0),
B = (N+ , f, 1), and C = (Q, g, 12 ) from the preceding section. c is a term,
since it defines a specific element in any L-structure. Every variable x is a
term, since it varies over the elements of the universe in any structure. F by
itself is not a term, since there is no indication of what element the function
is to be applied to. F (c) is a term, and it defines the element 1 = F A (cA )
in A, 1 = 1 · 1 in B, and 41 = 12 · 21 in C. F (x) is a term since it defines
the function s = F A in A, etc. Similarly F (F (x)) is a term; it defines the
function s(s(x)) = x + 2 in A, the function f (f (x)) = x4 in B, and the
function g(g(x)) = x4 in C.
The formal definition of term is analogous to the definition of sentence in
sentential logic.
Definition 4.1. The terms of L are defined as follows:
(i) every variable and every constant of L is a term of L;
(ii) if t1 , . . . , tn are terms of L and F is an n-ary function symbol of L
then F (t1 , . . . , tn ) is a term of L;
(iii) nothing else is a term of L.
We use T mL for the set of all terms of L.
For example, if F and G are binary functions symbols of L and c and d
are constants of L, then the following are all terms of L:
F (c, c), G(d, c), F (v1 , v4 ), G(v6 , F (v6 , v4 )), F (G(F (c, v1 ), G(d, v2 )), d).
If A is the structure whose universe is N and in which F A = +, GA = ·,
cA = 1, and dA = 2, then F (c, c) and G(d, c) both define 2, F (v1 , v4 ) defines
+ (or, perhaps better, x + y or even x1 + x4 ), and G(v6 , F (v6 , v4 )) defines
y · (y + x) (or perhaps x6 · (x6 + x4 )).
We next discuss formulas via an example.
Consider the language L whose non-logical symbols are the binary relation
symbol R and a constant c. R by itself is not a formula, since it doesn’t
specify which elements are in the relation R. But R(c, c), R(c, v1 ), R(v2 , v2 ),
and R(v2 , v1 ) are formulas, as are c = c, c = v1 , and v2 = v1 . There are called
atomic formulas since they are not built up from other formulas. Starting
with the atomic formulas we build up more complicated formulas using the
connectives and quantifiers. For example ¬R(c, c), (R(v1 , v2 ) → R(v2 , v1 )),
∃v2 R(v2 , v1 ), (¬v1 = c → ∃v2 R(v2 , v1 )) are all formulas.
Definition 4.2. The atomic formulas of L are the expressions of the form
R(t1 , . . . , tn ) where R is an n-ary relation symbol of L and t1 , . . . , tn ∈ T mL .
By convention we customarily write t1 = t2 instead of = (t1 , t2 ).
 CHAPTER 2. FIRST ORDER LOGIC 5

Formulas are now defined analogously to sentences in sentential logic.

Definition 4.3. The formulas of L are defined as follows:
(i) any atomic formula of L is a formula of L;
(ii) if ϕ is a formula of L then so is ¬ϕ;
(iii) if ϕ and ψ are formulas of L then so are (ϕ∧ψ), (ϕ∨ψ), and (ϕ → ψ);
(iv) if ϕ is a formula of L then so are ∀vn ϕ and ∃vn ϕ for all n ∈ N.
We write F mL for the set of formulas of L. The sequence of formulas
showing how a formula ϕ is built up from the atomic formulas is called a
history of ϕ. The formulas in a history of ϕ are called subformulas of ϕ.
In the formula ∃v2 R(v2 , v1 ) the variables v1 and v2 have different roles.
For example consider the structure A whose universe is N and in which R
is interpreted as <. What this formula then says is “there is something less
than v1 ” — this is a property of (the elements interpreting) v1 , but not of
v2 . The variable v1 actually varies over the elements of the universe since it
is not quantified, but v2 does not vary since it is bound by the existential
quantifier. So this formula defines a unary relation (that is, a set), in the
example the set of all k > 0, that is the set of all numbers k ∈ N which
satisfy this property when substituted for the variable v1 . The following
definition makes this distinction precise.
Definition 4.4. An occurrence of a variable x in a formula ϕ is bound
provided the occurrence is in a subformula of ϕ beginning with a quantifier
on x. An occurrence which is not bound is free.
For example, if ϕ = (∃v1 R(v1 , v2 ) ∨ ∃v2 R(v1 , v2 )) then the first two occur-
rences of v1 are bound and the last occurrence is free, but the first occurence
of v2 is free and the last two are bound.
As we will see in the next section, a formula defines a relation of the
(interpretations of the) variables occurring free in the formula. And a for-
mula in which no variable occurs free will be either true or false in any
interpretation.
Definition 4.5. A sentence of L is a formula in which no variable occurs
free.

5. Interpretating Terms, Formulas, and Sentences in

Structures
Let t ∈ T mL and let x1 , . . . , xn list all variables occurring in t. Then t
defines an n-ary function tA in any L-structure A. To emphasize this fact
we will sometimes write t as t(x1 , . . . , xn ). If t contains no variables then t
defines an element tA in any A. If t is a term in the variables x1 , . . . , xn , A is
a structure, and a1 , . . . , an ∈ A then the value of tA (ai , . . . , an ) is obtained
from t by replacing each F by F A , each c by cA , and all occurrences by
x1 , . . . , xn by a1 , . . . , an respectively.
6 CHAPTER 2. FIRST ORDER LOGIC

For example, suppose F and G are binary function symbols of L, c is a

constant of L, and A is an L-structure with universe N and interpreting F
by +, G by ·, and c by 1. Let t = G(F (v2 , c), G(v2 , F (c, c))). Then tA is a
unary function and for any k ∈ N we have tA (k) = (k + 1) · (k · (1 + 1)) or
simply (k + 1) · k · 2.
With the same L and A, consider the term t = F (G(v3 , v3 ), G(v3 , v4 )).
Then tA is a binary function and tA (k, l) = k · k + k · l for every k, l ∈ N.
Let ϕ ∈ F mL and let x1 , . . . , xn list all of the variables which occur free
in ϕ. Then ϕ defines an n-ary relation ϕA in any L-structure A. We will
frequently write ϕ as ϕ(x1 , . . . , xn ) in this case. The following definition of
satisfaction shows both how ϕA is defined and how the truth of a sentence
in A is defined.
Definition 5.1. Let A be an L-structure. We define the relation a1 , . . . , an
satisfies ϕ in A for ϕ(x1 , . . . , xn ) ∈ F mL and a1 , . . . , an ∈ A as follows:
(i) If ϕ is R(t1 , . . . , tk ) then a1 , . . . , an satisfies ϕ in A iff
RA (tA A
1 (a1 , . . . , an ), . . . , tk (a1 , . . . , an )) holds;
if ϕ is t1 = t2 then a1 , . . . , an satisfies ϕ in A provided tA 1 (a1 , . . . , an ) =
A
t2 (a1 , . . . , an ).
(ii) If ϕ is ¬ψ then a1 , . . . , an satisfies ϕ in A iff a1 , . . . , an does not
satisfy ψ in A.
(iii) The cases ϕ = (ψ ∗ θ) for a binary connective ∗ are also defined just
as in sentential logic.
(iv) If ϕ is ∀yψ for some formula ψ(x1 , . . . , xn , y) then a1 , . . . , an satisfies
ϕ in A iff a1 , . . . , an , b satisfies ψ in A for every b ∈ A; the case
ϕ = ∃yψ is handled analogously.
Notation: We write A |= ϕ(a1 , . . . , an ) if a1 , . . . , an satisfies ϕ in A. If ϕ
is a sentence we write A |= ϕ, also read A models ϕ, if the empty sequence
satisfies ϕ in A.
The relation ϕA can now be defined as follows:
ϕA (a1 , . . . , an ) holds iff A |= ϕ(a1 , . . . , an ).
WARNING: ϕ(a1 , . . . , an ) is not a formula of L for n ≥ 1 since a1 , . . . , an
are not symbols of L.
We look at some examples where Lnl = {R} for a binary relation symbol
R. Consider first the structure A = (N, <). No k ∈ N satisfies R(v1 , v1 )
since k < k always fails. Therefore every k satisfies ¬R(v1 , v1 ) and hence
A |= ∀v1 ¬R(v1 , v1 ).
k satisfies ∃yR(y, x) in A iff there is some l ∈ N such that A |= R(l, k),
i.e. l < k, which holds iff k > 0. Therefore the formula ¬∃yR(y, x) defines
the set {0}.
k1 , k2 satisfies ∃y(R(v1 , y) ∧ R(y, v2 )) iff there is some l ∈ N with A |=
R(k1 , l) ∧ R(l, k2 ), that is, iff k1 + 1 < k2 . Therefore ¬∃y(R(v1 , y) ∧ R(y, v2 ))
defines the relation which holds between k1 , k2 iff k1 + 1 = k2 , i.e. k2 is the
immediate successor of k1 .
 CHAPTER 2. FIRST ORDER LOGIC 7

The reader should consider these examples also for the structures B =
(N, ≤), C = (Z, <), D = (Q, <), and E = (N, >). In particular, show that
{0} is also definable in B and E, but not by the same formula as in A.

6. Translations from English

In this section we give some examples of translating mathematical English
into sentences of an appropriate first order language. We start with some
examples on quantifiers.
Let L have unary relations symbols O and P as its only non-logical sym-
bols. Let A be the L-structure with universe N and which interprets O as
the set of odd numbers and P by the set of prime numbers. We explain how
to correctly translate some sentences about A.
(1) “All odd numbers are prime”. We can only quantify over the entire
universe of the structure, so this has to be translated by saying “For every
number, if it is odd then it is prime”. We thus obtain ∀x(O(x) → P (x)).
(2) “Some odd numbers are prime”. This states that some numbers are
both odd and prime, so the correct translation is ∃x(O(x) ∧ P (x)).
(3) “No odd numbers are prime”. We don’t have a separate quantifier for
“there is no x”. This statement is simply the negation of “Some odd number
is prime”, and so its translation is ¬∃x(O(x) ∧ P (x)). This is logically
equivalent (see section 7) to the sentence ∀x(O(x) → ¬P (x)), but this is
not the most natural translation.
More generally, most basic mathematical statements about functions and
sets can be translated easily, after a little practice. For example, If F is a
unary function symbol the statement “F is one-to-one” becomes ∀x∀y(F (x) =
F (y) → x = y). If P is a unary realtion symbol then the statement “P con-
tains at least two elements” becomes ∃x∃y(P (x) ∧ P (y) ∧ ¬x = y). How
would you say “P contains exactly two elements”?

7. Validity, Satisfiability, and Logical Consequence

We define validity and satisfiability of sentences just as we did for senten-
tial logic.
Definition 7.1. A sentence θ of L is logically true, or valid, iff A |= θ for
every L-structure A. We write |= θ to mean that θ is valid.
Definition 7.2. A sentence θ of L is satisfiable iff A |= θ for some L-
structure A.
The following important fact is proved exactly like the corresponding
result for sentential logic (see Lemma 4.1 in Chapter 1).
Lemma 7.1. The sentence θ is satisfiable iff ¬θ is not valid; θ is valid iff
¬θ is not satisfiable.
Unfortunately there is no method like truth tables to check the satisfi-
ability or validity of a sentence. Sometimes in checking whether or not a
8 CHAPTER 2. FIRST ORDER LOGIC

sentence is valid it helps to suppose that there is some A falsifying it and

work backwards to see what A must look like.
Example. Decide whether or not θ defined as ∀x(P (x) → Q(x)) →
(∀xP (x) → ∀xQ(x)) is valid. The definition of satisfaction yields the fol-
lowing equivalences:
A 6|= θ iff
A |= ∀x(P (x) → Q(x)) and A 6|= ∀xP (x) → ∀xQ(x) iff
A |= ∀x(P (x) → Q(x)), A |= ∀xP (x), and A 6|= ∀xQ(x) iff
A |= P (a) → Q(a) for all a ∈ A, A |= P (a) for all a ∈ A, and A |= ¬Q(b0 )
for some b0 ∈ A iff
(either A |= ¬P (a) or A |= Q(a)) for all a ∈ A, A |= P (a) for all a ∈ A,
and A |= ¬Q(b0 ) for some b0 ∈ A.
But it is impossible to satisfy this last condition, so there can be no such A
and so θ is valid.
Example Decide whether or not θ defined as ∃x(P (x) → Q(x)) → (∃xP (x) →
∃xQ(x)) is valid. We obtain the following sequence of equivalences:
A 6|= θ iff
A |= ∃x(P (x) → Q(x)), A |= ∃xP (x), and A 6|= ∃xQ(x) iff
A |= P (a0 ) → Q(a0 ) for some a0 ∈ A, A |= P (a1 ) for some a1 ∈ A, and
A |= ¬Q(b) for all b ∈ A.
But the conditions in this last line can be satisfied, for example by defining
A = {0, 1}, P A = {1}, and QA = ∅. Thus θ is not valid, since it is false on
the structure A we just defined.
It is convenient to define validity also for formulas with free variables. Let
ϕ ∈ F mL have free variables x1 , . . . , xn , and let A be an L-structure. We
say that ϕ is true on A, notation A |= ϕ, provided that A |= ∀x1 , . . . , ∀xn ϕ.
Note that, unlike sentences, usually neither ϕ nor ¬ϕ is true on A.
Definition 7.3. A formula ϕ of L is valid iff A |= ϕ for every L-structure
A. We write |= ϕ to mean that ϕ is valid.
In the first example above we showed that
|= ∀x(P (x) → Q(x)) → (∀xP (x) → ∀xQ(x)).
A similar argument shows that for any formulas ϕ and ψ we have
|= ∀x(ϕ → ψ) → (∀xϕ → ∀xψ).
Satisfiability of sets of sentences and logical consequence are also defined
just as in sentential logic.
Definition 7.4. A set Σ of sentences of L is satisfiable iff there is some A
such that A |= Σ (meaning A |= σ for every σ ∈ Σ).
Definition 7.5. Let Σ be a set of sentences of L and let ϕ ∈ F mL . Then
ϕ is a logical consequence of Σ, written Σ |= ϕ, iff A |= ϕ for every A |= Σ.
We also obtain the following just as in sentential logic.
Lemma 7.2. Assume that Σ ⊆ SnL and θ ∈ SnL . Then Σ |= θ iff (Σ ∪
{¬θ}) is not satisfiable.
 CHAPTER 2. FIRST ORDER LOGIC 9

We also have the expected notion of equivalence of formulas.

Definition 7.6. Formulas ϕ and ψ are equivalent, written ϕ ≡ ψ, iff |= ϕ →
ψ and |= ψ → ϕ.

For example, just as in sentential logic, we see the following:

(ϕ ∨ ψ) ≡ (¬ϕ → ψ) and
(ϕ ∧ ψ) ≡ ¬(ϕ → ¬ψ).
Furthermore, using the definition of satisfaction, we also obtain:
∃xϕ ≡ ¬∀x¬ϕ.
We thus obtain the following result:

Theorem 7.1. For any ϕ ∈ F mL there is some ϕ∗ ∈ F mL such that ϕ ≡ ϕ∗

and ϕ∗ is built using just ¬, → and ∀.

8. Properties of Validity and Logical Consequence

In this section we identify some classes of valid formulas and some prop-
erties of logical consequence which will, in particular, be used in defining
our system of deduction.
Tautologies. Let θ be a sentence of sentential logic which is a tautology,
for example, A → (B → A). Let θ ∗ result by replacing each atomic sentence
in θ by a formula of L, for example replacing A by P (x) and replacing B by
Q(x), so θ ∗ is P (x) → (Q(x) → P (x)). Then θ ∗ is a formula of L and |= θ ∗ .
We will also call the formula θ ∗ a tautology.
Universal Quantification. (a) Assume that Σ |= ϕ, where Σ ⊆ SnL
and ϕ ∈ F mL . Then for any variable x we have Σ |= ∀xϕ. If x does not
occur free in ϕ this is clear, since in this case ϕ ≡ ∀xϕ. If x does occur free
in ϕ we obtain this since A |= ϕ means that A |= ∀x1 . . . ∀xn ϕ and x is some
xi .
(b) If x does not occur free in ϕ then |= (ϕ → ∀xϕ) since, as remarked
above, ϕ ≡ ∀xϕ under this condition. The assumption that x does not occur
free in ϕ is essential since, for example, (P (x) → ∀xP (x)) is not valid.
Modus Ponens. Assume that Σ |= ϕ and Σ |= (ϕ → ψ). Then Σ |= ψ.
Let x1 , . . . , xn list all variables occurring free in either ϕ or ψ, so we can
write ϕ(x1 , . . . , xn ) and ψ(x1 , . . . , xn ). Let A |= Σ. Then A |= ϕ(a1 , . . . , an )
and A |= ϕ(a1 , . . . , an ) → ψ(a1 , . . . , an ) for all a1 , . . . , an ∈ A. So by the
definition of satisfaction we must have A |= ψ(a1 , . . . , an ) for all a1 , . . . , an ∈
A, and so A |= ψ.
Substitution. Suppose that ϕ has only x free, so we can write ϕ as ϕ(x).
If t is any term, then the formula obtained by replacing all free occurrences
of x in ϕ by t will be written as ϕ(t). We “expect” that (∀xϕ(x) → ϕ(t))
should be valid since, intuitively, ∀xϕ says that every element of the universe
satisfies ϕ and ϕ(t) says that the element named by t satisfies ϕ. If t is
a constant c then this is exactly correct, since A |= ∀xϕ(x) implies that
A |= ϕ(a) for all a ∈ A, in particular for a = cA .
10 CHAPTER 2. FIRST ORDER LOGIC

But if t is a variable y (or contains a variable) this can fail — let ϕ(x) be
∃yR(x, y); then ϕ(y) is ∃yR(y, y) and it is easy to see that (∀x∃yR(x, y) →
∃yR(y, y)) is not valid. But this is the only difficulty — if no new occurrence
of y in ϕ(y) is bound (for example if y does not occur bound in ϕ(x)) then
|= (∀xϕ(x) → ϕ(y)). More generally, |= (∀xϕ(x) → ϕ(t)) provided no new
occurrence in ϕ(t) of a variable in t is bound.
Finally, if ϕ also has z1 , . . . , zk free, so ϕ can be written as ϕ(x, z1 , . . . , zk ),
then we have |= (∀xϕ(x, z1 , . . . , zk ) → ϕ(t, z1 , . . . , zk )) under the same cir-
cumstances.

9. A Formal Proof System

In this section we define a deductive system such that the deducible for-
mulas are precisely the valid formulas — this is the Completeness Theorem,
proved in the next Chapter. As with sentential logic we first specify a set
of formulas called the logical axioms, and the only rule we use is modus
ponens. A formula is deducible iff it is obtained from the axioms by a finite
number of applications of modus ponens. We restrict to formulas using only
¬, →, and ∀. There is no loss in doing so since we know by Theorem 7.1
that every formula is logically equivalent to such a formula.
Definition 9.1. The set ΛL of (logical) axioms of L consists of all formulas
of the following forms:
(1) ∀y1 . . . ∀yn ϕ where ϕ is a tautology,

(2) ∀y1 . . . ∀yn ∀xϕ(x, . . . ) → ϕ(t, . . . ) where no new occurrence in
ϕ(t, . . . ) of a variable in t is bound,

(3) ∀y1 . . . ∀yn ∀x(ϕ → ψ) → (∀xϕ → ∀xψ) ,
(4) ∀y1 . . . ∀yn (ϕ → ∀xϕ) where x does not occur free in ϕ,
(5) (axioms for equality): x = x,
(x = y → y = x),
x = y → (y = z → x = z) ,

x1 = y1 → x2 = y2 → . . . (xm = ym → (R(x1 , . . . , xm ) → R(y1 , . . . , ym ))
for each m-ary relation symbol R, and a similar axiom for functions.
Lemma 9.1. If ϕ ∈ ΛL then |= ϕ.
Definition 9.2. A (logical) deduction is a finite sequence ϕ1 , . . . , ϕn of for-
mulas of L such that for every i ≤ n one of the following holds:
(i) ϕi ∈ ΛL ,
(ii) there are j, k < i such that ϕi follows from ϕj and ϕk by modus
ponens, that is, ϕk = (ϕj → ϕi ).
Definition 9.3. A formula ϕ is (logically) deducible, notation ⊢ ϕ, iff there
is a deduction whose last formula is ϕ.
As in sentential logic, every axiom is deducible by a deduction of length
one, and any non-trivial deduction has length at least three. Here is an
example, where L contains a unary relation symbol P :
1. ∀x¬P (x) → ¬P (x) – axiom (2)
 CHAPTER 2. FIRST ORDER LOGIC 11

2. (∀x¬P (x) → ¬P (x)) → (P (x) → ¬∀x¬P (x)) – axiom (1) where the
tautology is (A → ¬B) → (B → ¬A)
3. (P (x) → ¬∀x¬P (x))
This establishes that ⊢ (P (x) → ¬∀x¬P (x)), which may look more fa-
miliar as ⊢ (P (x) → ∃xP (x)).
We can show that every deducible formula is valid exactly as in sentential
logic (see Theorem 8.1 in Chapter 1).
Theorem 9.1. If ⊢ ϕ then |= ϕ.
We also use deductions from hypotheses.
Definition 9.4. Let Γ ⊆ F mL . Then a deduction from Γ is a finite sequence
ϕ1 , . . . , ϕn of formulas of A such that for every i ≤ n one of the following
holds:
(i) ϕi ∈ (Γ ∪ ΛL ),
(ii) there are j, k < i such that ϕi follows from ϕj and ϕk by modus
ponens.
We say that ϕ is deducible from Γ, notation Γ ⊢ ϕ, iff there is a deduction
from Γ whose last formula is ϕ.
We have the following fact, corresponding to the rule of modus ponens.
Lemma 9.2. Let Γ ⊆ F mL and let ϕ, ψ ∈ F mL . Assume that Γ ⊢ ϕ and
Γ ⊢ (ϕ → ψ). Then Γ ⊢ ψ.
We also have soundness for deductions from sets of sentences.
Theorem 9.2. Let Σ ⊆ SnL and let ϕ ∈ F mL . If Σ ⊢ ϕ then Σ |= ϕ.

10. The Deduction Theorem and Other Derived Rules

We prove several “derived rules” which will help in showing formulas are
deducible. The first of these is the Deduction Theorem, which is stated and
proved just as it was for sentential logic.
Theorem 10.1. (The Deduction Theorem) Assume that (Γ ∪ {ϕ}) ⊢ ψ.
Then Γ ⊢ (ϕ → ψ).
Just as in sentential logic, we use the Deduction Theorem to reduce the
problem of showing that Γ ⊢ (ϕ → ψ) to the simpler problem of showing
that (Γ ∪ {ϕ}) ⊢ ψ.
The various forms of Generalization are similarly used to simplify the
problem of showing that Γ ⊢ ∀xϕ.
Theorem 10.2. (Generalization) Assume that the variable x does not occur
free in any formula in Γ. Assume that Γ ⊢ ϕ. Then Γ ⊢ ∀xϕ.
Proof. (Outline) Let ϕ1 . . . . , ϕn be a deduction from Γ of ϕ. We show that
Γ ⊢ ∀xϕi for all i ≤ n by induction on i. If ϕi ∈ ΛL then also ∀xϕi ∈ ΛL . If
ϕi ∈ Γ then by hypothesis x cannot occur free in ϕi , so (ϕi → ∀xϕi ) ∈ ΛL
12 CHAPTER 2. FIRST ORDER LOGIC

and thus Γ ⊢ ∀xϕi by modus ponens. The inductive step uses a logical
axiom of form (3). 

Note that the hypothesis is automatically satisfied if Γ = ∅ or Γ is a set

of sentences.
We note without proof the following variation on Generalization.
Theorem 10.3. (Generalization on Constants) Let Γ ⊆ F mL and let ϕ(x) ∈
F mL . Assume that the constant c does not occur in any formula in Γ and
does not occur in ϕ(x). Assume that Γ ⊢ ϕ(c). Then Γ ⊢ ∀xϕ(x).
There are several special results useful for showing that negations are
deducible. They are easily proved by referring to the appropriate tautology.
Note that the converses also hold, but this is usually not useful in deductions.
Lemma 10.1. (a) If Γ ⊢ ϕ then Γ ⊢ ¬¬ϕ.
(b) If Γ ⊢ ϕ and Γ ⊢ ¬ψ then Γ ⊢ ¬(ϕ → ψ).
(c) (Contraposition) If (Γ ∪ {ϕ}) ⊢ ψ then (Γ ∪ {¬ψ}) ⊢ ¬ϕ.
Consistent sets are defined as in sentential logic and have the same prop-
erties with the same proofs, which we therefore omit.
Definition 10.1. A set Γ of formulas is consistent iff there is no formula ϕ
such that Γ ⊢ ϕ and Γ ⊢ ¬ϕ. Γ is inconsistent if it is not consistent.
Lemma 10.2. If Γ is inconsistent then Γ ⊢ ψ for all formulas ψ.
Theorem 10.4. (Proof by Contradiction) Γ ⊢ ϕ if (and only if ) (Γ ∪ {¬ϕ})
is inconsistent. Γ ⊢ ¬ϕ if (and only if ) (Γ ∪ {ϕ}) is inconsistent.
We illustrate the use of these rules in the following example.
Example. Show that ⊢ ∃x∀yR(x, y) → ∀y∃xR(x, y).
Eliminating the existential quantifier we see that what we need to do is to
show that ⊢ ¬∀x¬∀yR(x, y) → ∀y¬∀x¬R(x, y).
By the Deduction Theorem it suffices to show ¬∀x¬∀yR(x, y) ⊢ ∀y¬∀x¬R(x, y).
By Generalization, since y is not free on the left of ⊢, it suffices to show
¬∀x¬∀yR(x, y) ⊢ ¬∀x¬R(x, y).
By Contraposition it suffices to show ∀x¬R(x, y) ⊢ ∀x¬∀yR(x, y).
By Generalization, since x is not free on the left of ⊢, it suffices to show
∀x¬R(x, y) ⊢ ¬∀yR(x, y).
By Proof by Contradiction it suffices to show that Γ = {∀x¬R(x, y), ∀yR(x, y)}
is inconsistent, which is shown by the following deduction from Γ.
1. ∀x¬R(x, y) — hypothesis
2. ∀yR(x, y) — hypothesis
3. ∀x¬R(x, y) → ¬R(x, y) — axiom (2)
4. ∀yR(x, y) → R(x, y) — axiom (2)
5. ¬R(x, y) — MP on lines 1, 3
6. R(x, y) — MP on lines 2, 4
 CHAPTER 2. FIRST ORDER LOGIC 13

11. Expressability
What can we say about a structure using sentences of first order logic?
What can we define in a structure using formulas of first order logic? We
give some examples for the language L whose only non-logical symbol is a
binary relation symbol R.
An important class of L-structures is the class of linear orders. This
includes the familiar structures (N, <), (Z, <), (Q <, ), and (R, <). The
class of linear orders can be axiomatized by the set Σl.o. containing the
following L-sentences:
∀x¬R(x, x),

∀x∀y∀z (R(x, y) ∧ R(y, z)) → R(x,
z) ,
∀x∀y R(x, y) ∨ x = y ∨ R(y, x) .
Other properties of linear orders follow
from these three axioms. For exam-
ple, Σl.o. |= ∀x∀y R(x, y) → ¬R(y, x) .
This definition of linear order gives what are sometimes called strict linear
orders – non-strict linear orders include (N, ≤), etc.
Linear orders may differ on other sentences of L — for example, any
two of (N, <), (Z, <), and (Q, <) can be distinguished by a sentence of L.
However, (Q, <) and (R, <) both satisfy precisely the same sentences of L.
In fact, if we define Σd.l.o. to be Σl.o. together with the sentences

∀x∃yR(x, y),
∀y∃xR(x, y), and ∀x∀y R(x, y) → ∃z(R(x, z) ∧ R(x, y)) then we have the
following remarkable result, whose proof is beyond the scope of these notes.
Theorem 11.1. For every θ ∈ SnL either Σd.l.o. |= θ or Σd.l.o. |= ¬θ.
Corollary 11.1. Let A and B be L-structures and assume that they both
model Σd.l.o. . Then for every θ ∈ SnL we have A |= θ iff B |= θ.
If we add to Σl.o. sentences saying “there is a first element, every element
has an immediate successor, and every element except the first has an imme-
diate predecessor” we obtain the set Σn.l.o. which axiomatizes the sentences
true on (N, <).
Theorem 11.2. For every θ ∈ SnL we have (N, <) |= θ iff Σn.l.o. |= θ.
Of course there are other structures which are also models of Σn.l.o., for
example (N r {0}, <) and ({k ∈ Z : k ≤ 0}, >). More importantly, there are
also models A of Σn.l.o. which contain “infinite” elements, that is, elements
a such that {b ∈ A : RA (b, a) holds} is infinite. This will be a consequence
of the Completeness Theorem proved in the next Chapter.