16 lectures on
Gödel’s Incompleteness Theorems
Michaelmas Term 2010
Daniel Isaacson
Faculty of Philosophy
Oxford University
1
CONTENTS 2
11 Provable Σ1 -completeness 88
CONTENTS 4
This unspoken lecture reviews essential results on first-order logic and formal systems
as background to the development of Gödel’s incompleteness theorems.
The notion of logical consequence is the starting point. A rough characterization
of this notion is the following. A sentence φ is a logical consequence of a set of
sentences Γ, symbolized as Γ φ, if and only if φ is true in every interpretation of
the language of Γ ∪ {φ} in which all the sentences of Γ are true. If Γ is empty, Γ φ
reduces to the condition that φ is true in every interpretation of the language of φ,
which is to say, φ is logically valid, symbolized by φ.
Accordingly, we need to make precise the notion of a formal language and the
notion of interpretation of a formal langauge.
5
LECTURE 0 6
(unary, binary, ternary, and so on). A zero-ary function symbol is a constant term,
a zero-ary relation symbol represents a sentence. The shape of a symbol is completely
arbitrary, though there are some conventional choices, e.g. the universal quantifier
is usually written “∀”, but used often to be written “( )” V (the brackets enclosing
the variable of quantification), and is sometimes written “ ” (called a California
quantifier). But the symbols can be anything and in particular, as we shall exploit,
they can be digits also used to generate numerals.
We shall in this course always have full first-order classical (as opposed to intu-
itionistic) predicate logic with identity as our background logic. Quite apart from
the shapes of the symbols, there are choices to be made as to which are our prim-
itive propositional functions and quantifiers. In our background logic we shall take
as our primitives of propositional logic negation, conjunction, disjunction, and im-
plication, for which the symbols will be ∼, ∧, ∨, and ⊃, and both the universal
and existential quantifiers as primitive, written as ∀ and ∃. As you will know from
a previous logic course, we could take as primitive just negation with any one of
conjunction, disjunction, and implication and either the universal or the existential
quantifier, and in our official formal language we shall take as primitive ∼, ⊃, and ∀.
We shall operate with the langauge as if it also contained symbols for conjunction,
disjunction, and the existential quantifiers, but strictly (A ∨ B) will be an abbrevi-
ation for (∼ A ⊃ B), etc. There is also the connective “if and only if”, which we
will symbolize as ≡. Even in our background logic we will take (A ≡ B) to be an
abbreviation for ((A ⊃ B) ∧ (B ⊃ A)).
Stipulating that our background logic is first-order classical logic with identity,
means that we have a primitive symbol for identity, which we will take as the
.
common symbol for identity = (though others are sometimes used, e.g. ≈, ≏, =),
but having a symbol and first-order axioms for identity is not sufficient to make a
logical system first-order logic with identity. Identity is a two place relation which
every object bears to itself and to no other object. The first property is easily
expressed: ∀v1 v1 = v1 , but the second property is not easily expressed, in particular
not expressed by ∀v1 ∀v2 (v1 6= v2 ⊃ v1 6= v2 ) (if v1 is a different object from v2 then
v1 6= v2 ).
terms in the language
formulas in the language
LECTURE 0 7
each numbered formula are the numbers of the Assumptions, if any, on which that
formula depends. The entry in the right-hand column gives the basis on which the
formula in the third column at that line is introduced into the deduction, i.e. either
as an Assumption or by one of the Rules of Inference. An Assumption depends on
itself, so in an application of the rule of Assumption there is one number in the first
column which is the same number as in the second column. If the formula in the
third column of a given line is introduced by a Rule of Inference, the entry in the
fourth column for that line says what Rule of Inference has been used and numbers
of the formulas to which that Rule of Inference has been applied. The Assumptions
on which those formulas depend, as given by the numbers in the first column at
the lines for those formulas, are gathered together as the Assumptions on which
the formula that results from application of that Rule of Inference depends. Four
of the Rules of Inference, ⊃-Introduction, ∼-Introduction, ∨-Elimination, and ∃-
Elimination, discharge an Assumption, so it is possible to arrive at a formula which
depends on no assumptions. Such formulas are, by the Soundness of the Rules of
Inference, logically valid.
∧-Introduction
∧-Elimination1
∧-Elimination2
∨-Introduction1
LECTURE 0 9
∨-Introduction2
assumptions numbering formulas justifications
A (a) G [whatever]
A (b) (F ∨ G) (a) ∨-Introduction1
∨-Elimination
assumptions numbering formulas justifications
A (a) (F ∨ G) [whatever]
{b} (b) F Assumption
B ∪ {b} (c) H [whatever]
{d} (d) G Assumption
C ∪ {d} (e) H [whatever]
A∪B∪C (f) H (a)(c)(e) ∨-Elimination
⊃-Introduction
⊃-Elimination
assumptions numbering formulas justifications
A (a) (F ⊃ G) [whatever]
B (b) F [whatever]
A∪B (c) G (a)(b) ⊃-Elimination
The Introduction and Elimination rules for negation in this system of rules are
not as natural as those for the other logical connectives. A more natural formulation
of logic in natural deduction is to take negation to be defined in terms of a primitive
false sentence, sometimes symbolized as ⊥, by the equivalence ∼ ψ ≡ ψ ⊃⊥. How-
ever, in terms of formalizing informal maths talk it is more natural to take negation
LECTURE 0 10
∼-Introduction
∼-Elimination
∀-Introduction
∀-Elimination
∃-Introduction
∃-Elimination
LECTURE 0 11
=-Introduction
=-Elimination
⊢LDN (F ∨ ∼ F )
(1) (F ∨ ∼ F ) Tautology
Proof.
Note that the Prenex Normal Form Theorem holds only on the assumption that
all domains of interpretation are non-empty. Otherwise we have, for example, that
for v1 not free in G, and G true in the empty domain (e.g. G = ∀v1 v1 = v1 , or
equally G = ∀v1 ∼ v1 = v1 ),
(∀v1 F (v1 ) ⊃ G) is true in the empty domain, since G is true, but ∃v1 (F (v1 ) ⊃ G))
is false, since every existentially quantified statement is false in the empty domain.
Note that prenex normal forms are not in general unique, e.g.
(∀v1 F (v1 ) ⊃ ∀v2 G(v2 ) has as prenex normal form both ∃v1 ∀v2 (F (v1 ) ⊃ G(v2 )) and
∀v2 ∃v1 (F (v1 ) ⊃ G(v2 )).
LECTURE 0 14
15
LECTURE 1 16
power 2 always yields a value, i.e. ∀x∃y(2x = y) is infinitary, and more generally,
quantification over the infinite domain of natural numbers is infinitary. However,
quantification over a bounded, i.e. initial segment of the natural numbers, which is
finite, belongs to finitary mathematics.
Hilbert’s deep insight was to recognize that the formal manipulation of all sym-
bols, not just the symbols for numbers, i.e. numerals and terms built up from
numerals and symbols for arithmetical operations, belongs to finitary mathematics.
In particular,
Hilbert recognized two sorts of finitary statements, general and particular (though
he did not introduce terminology for this distinction). Particular finitary statements
are decided by computations, e.g. 7 × 5 = 35, and 210 = 1024 and truth functional
combinations of them (the truth values of such combinations being computable from
the truth values of the component statements). General finitary statements contain
free variables, and can be thought of as a template for particular finitary state-
ments that result by substitution of numerals for the free variables, for example
x + y = y + x, and (n > 2 ⊃ xn + y n 6= z n ). On the other hand, ∀x∀y x + y = y + x
and ∀n∀x∀y∀z(n > 2 ⊃ xn + y n 6= z n ) are infinitary.
For F (v1 ) a general finitary statement with free variable v1 , bounded quantifica-
tion on the variable v1 , which is finitary, is expressible using (apparently) unbounded
quantification by, in the case of universal quantification, ∀v1 (v1 ≤ t ⊃ F (v1 )), for
t a term in the language of arithmetic, which we abbreviate as (∀v1 ≤ t)F (v1 ),
and in the case of existential quantification, ∃v1 (v1 ≤ t ∧ F (v1 )), which we abbre-
viate (∃v1 ≤ v2 )F (v1 ). For t a numerical term (a numeral or a composition of
arithmetical functions applied to numerals), (∃v1 ≤ t)F (v1 ) and (∀v1 ≤ t)F (v1 ) are
particular finitary statements if v1 is the only free variable in F (v1 ). If t is a free
variable or a composition of arithmetical functions applied to one or more variables,
(∃v1 ≤ t)F (v1 ) and (∀v1 ≤ t)F (v1 ) are general finitary statements.
Hilbert noted that general finitary statements are not closed under negation, i.e.
the negation of a general finitary statement cannot be expressed as a general finitary
statement. For example, Fermat’s Last Theorem is expressible as a general finitary
statement, (n > 2 ⊃ xn + y n 6= z n ), but to say (falsely) that Fermat’s Last Theorem
is false requires existential quantification, ∃n∃x∃y∃z(n > 2 ⊃ xn + y n 6= z n ). On
the other hand, the statement that a specific quadruple of numbers a, b, c, d is a
counterexample, i.e. (a > 2 ∧ an + bn = cn ), is a particular finitary statement.
Statements about particular formal proofs are, as Hilbert recognized, finitary
statements, e.g. that a particular formal proof is or is not a proof of a particular
statement.
LECTURE 1 17
Hilbert missed something about his insight which Gödel realized, namely that
formal proofs can be literally identified with natural numbers, i.e. they could be
taken to be numerical expressions, rather than merely like them. As Gödel put this
point in (1931),
Numbers assigned to formulas of a formal language in this way are called Gödel
numbers.
To carry out the arithmetization of syntax, the system must be able to ”talk”
about numbers, i.e. there must be for each natural number a formal numeral in the
language of the system that denotes that numbers.
Definition 4 For formal languages that have a numeral for each natural number,
we denote by n the numeral for the natural number n.
What Gödel showed is that the property of being (the Gödel number of) a
provable formula is expressible within any system which can express basic arithmetic,
i.e. there is a formula P r(v1 ) with one free variable, in the language of a formal
system for arithmetic, S, such that for every formula X in the language of S, S ⊢ X
if and only if P r(pXq) is true. We shall establish the existence of such a formula for
a particular formal system of arithmetic in Lecture 4. Gödel also showed that for any
formula with one free variable (in particular a formula that expresses the property
of being the Gödel number of an unprovable formula) there is a diagonal sentence,
i.e for formula F (v1 ) there is a sentence D such that the equivalence (D ≡ F (pDq))
is true. We shall establish this result in Lecture 3.
From these results and on the assumption that everything provable in a given
system S is true (about the natural numbers) (a strong assumption, much stronger
than is needed to establish incompleteness, but it is illuminating to consider this
simple case), it is easy to see that for G such that (G ≡∼ P r(pGq)), S 0 G, and also
thereby that G is true, which implies, from the assumption that everything provable
in a system S is true, that S 0∼ G.
LECTURE 1 18
Proof. (1) Suppose that S ⊢ G. (2) Then by (ii), P r(pGq) is true. (3) From
(2) and (i), G is false. (4) From (3) and (iii), S 0 G. (5) Since (4) contradicts (1),
we have by reductio ad absurdum that S 0 G (from assumptions (i), (ii), and (iii)).
(6) From (5) and (ii), P r(pGq) is false. (7) From (6) and (i), G is true.
(8) From (7) and (iii), S 0∼ G. N
so on. On the other had, both addition and multiplication are needed for incom-
pleteness: there is a complete theory of zero, successor, and addition (Presburger
Arithmetic).
(4) The independence of the Gödel sentence from formal arithmetic was unprece-
dented. In the previous hundred years the independence of Euclid’s fifth postulate
from the other postulates of geometry had been established. Gödel’s result differs
from this earlier one in two crucial ways. One was the technique used. The result
concerning the fifth postulate was established by the construction of models. The
Gödel result is purely syntactic (exploiting Hilbert’s insight). The other difference
is even more fundamental. The fifth postulate is neither true nor false, per se. It
is true in Euclidean geometry and false in non-Euclidean geometries. The Gödel
sentence is demonstrably true, though not demonstrable in the system for which it
is constructed.
Proposition 6 For any system S and sentence X, a proof that S 0 X from the
assumption of consistency is best possible.
Remark. That the Gödel sentence G for a system S is not refutable, i.e. ∼ G
is not provable, requires a stronger condition on S than consistency but a condition
much weaker than the soundness of S is sufficient.
symbols f′ , f′′ , f′′′ will denote the functions addition, multiplication, and exponenti-
ation, respectively, which we will write informally as +, ·, and exp or xy in the usual
notation. There are an infinity of variables v′ , v′′ , v′′′ , . . ., which we will usually write
as v1 , v2 , v3 , . . .. If we want to signify a variable without specifying which variable it
is, we will write vi , vj etc or use informal variable letters x, y, z, u, v, w.
The symbol for the propositional connectives negation and implication are ∼
and ⊃. The symbol for the universal quantifier is ∀.
The symbols = and ≤ are for the two-place relations of equality and less than
or equals.
The symbol ♯ will be used to mark breaks between strings of symbols that are
terms and formulas of the language (to be defined in the next lecture) in sequences
of terms and formulas.
An expression in the language is (almost) any finite string of these symbols.
The set of expressions for the language LE is specified by the following recursive
definition.
Remark: It’s for a technical reason (to do with our choice of Gödel numbering)
that we excluded from the class of expressions as here specified strings of more than
one symbol that begin with the symbol ′ .
Definition 7 (notation for expressions and Gödel numbers) En =df the ex-
pression with Gödel number n.
pEq =df the Gödel number of expression E.
LECTURE 1 21
We are used to the idea that numbers are denoted by numerals and that numerals
are not the same thing as numbers. The Roman numerals for the first five non-zero
natural numbers are I, II, III, IV, V, while the Arabic numerals are 1, 2, 3, 4 , 5. The
crucial property of the Arabic numerals is that they are constructed on a place-value
system with a base of 10. That the system of numerals in common use is base 10 is
presumably down to the contingent fact (it could have been otherwise) that human
beings have 10 fingers. Any other number greater or equal to 2 gives a perfectly
good numeral system with that base. Base 2 is used in machine code for computers,
with 0 and 1 represented by current off and current on. The number we write as 15
in base 10 we write as 1111 in base 2 and as 13 in base 12. In the formal language for
arithmetic we shall be using the numerals for numbers use a tally notation, rather
n
z }| {
than place values. The formal numeral for the number n is the expression 0′ . . . ′ ,
i.e. concatenation of the symbol 0 with n-many concatenations of the symbol ′ .
The following function plays a key role in our chosen system of Gödel numbering.
′
0 ( ) f ′ v ∼ ⊃ ∀ = ≤ ♯
1 0 2 3 4 5 6 7 8 9 η ǫ δ
Of course on this assignment not every number is a Gödel number. But we can
effectively tell the ones that are, i.e. we know that if an 8 or a 9 occurs its base 10
notation it must occur within a string of the form 89, 899, 8999, 89999, 899999, and
we know which symbol is coded by counting the number of 9s in that string.
Lecture 2
Remark: Formal variables are expressions of the form v′ , v′′ , v′′′ , . . .. We will
abbreviate the string of symbols consisting of the formal variable symbol v followed
by n subscripts as vn .
Further remark: Smullyan’s formal variables are of the form (v′ ), (v′′ ), (v′′′ ), . . .,
i.e. formal variables as we have defined them enclosed in brackets (p. 15). Unique
readability for terms within expressions does not require enclosing variables within
23
LECTURE 2 24
brackets in this way. The motivation for this unnecessary use of brackets might
be an artefact of what’s simple to write in LaTeX. Evidently Smullyan produces
e.g. the substring v′′ of his variable (v′′ ) by the LaTeX code v_{’’}. If this was
v2 , then the seemingly natural way to write LaTeX code for the concatenation of
v2 with ′ would be $v_{’’}^{\prime}$, but this produces v′′′ , in which the ′ sym-
bol occurs interposed over the string v′′ rather than concatenated at the end of
that string. On the other hand, $(v_{’’})^{\prime}$ produces (v′′ )′ , in which ′
is correctly concatenated at the end. Nonetheless, more complicated LaTeX code
(also with compounded subscript command to lower the apostrophe further so it
looks more convincing as a subscript) produces the required concatenation with vari-
ables not enclosed in brackets, namely $v_{_{’’}}$\hspace{-.12ex}$^{\prime}$,
which compiles to the required concatenation of symbols v′′ ′ .
Notation: For natural number n we write n for the numeral that denotes the
number n, e.g. 7 = 0′′′′′′′ .
2.1.2 Formulas
Definition 15 (Atomic formulas) An atomic formula is any expression of the
form t1 = t2 or of the form t1 ≤ t2 , where t1 and t2 are terms.
[Note that this convention on possible occurrence of free variables other than
those explicitly shown is different from Smullyan’s: “We write F (vi1 , . . . vik ) for any
formula in which vi1 , . . . vik are the only free variables.” (p. 16). Since there are
situations, e.g. in stating the Induction axioms, in which we need to allow for the
possibility of other free variables than those explicitly shown, Smullyan’s convention
has in those situations to be violated, e.g. “F (v1 ) is to be any formula at all (it
may contain free variables other than v1 )” (Smullyan, p. 29). It seems to me more
coherent for the convention to allow other variables, which also then allows us under
the same convention to stipulate in a given situation that there are no other free
variables.]
2.2.2 Truth
Truth for a sentence of LE in the structure of the natural numbers (the intended
interpretation) can be defined by recursion over the recursive generation of the
sentence in the usual way. For none of the results in this course do we require a
formal definition of truth, and I will take it as known informally what it means for
a formula in the language of arithmetic to be true in the structure of the natural
numbers.
2.2.3 Expressibility
Definition 23 (expressibility of relations) A formula F (v1 , . . . , vn ) in LE is said
to express a relation R ⊆ Nn iff for every n-tuple < k1 , . . . , kn > of natural numbers
the sentence F (k 1 , . . . , k n ) is true iff < k1 , . . . , kn >∈ R. In such case the relation
R is said to be expressible in LE .
Lemma 10 For ℓb (n) the length of the base b notation for n, i.e. the number of
digits in the base b notation of n, the two-place relation bℓb (v1 ) = v2 is Arithmetical
Proof. This relation is expressed by the following condition on v1 and v2 .
((v1 = 0∧v2 = b)∨(v1 6= 0∧P owb (v2 )∧v1 < v2 ∧∀v3 ((P owb (v3 )∧v1 < v3 ) ⊃ v2 ≤ v3 ))).
This equivalence is seen as follows: If v1 = 0, ℓ(v1 ) = 1 and b1 = b. The first
disjunct takes care of this case. If v1 6= 0, then the length of v1 (in base b notation) is
the least power of b > v1 , e.g. ℓ10 (935) = 3, and 103 = 1000 is the least power of 10
> 935. [Note that we need to treat these two cases separately, since ℓ(0) = 1, but—
writing µv3 F (v3 ) for the least v3 s.t. F (v3 )—µv3 (10v3 > 0) = 0 since 100 = 1 > 0.]
That v2 is the least power of b greater than v1 is expressed by the two conditions
that bv2 > v1 , and for any v3 such that bv3 > v1 , v2 ≤ v3 .
The above condition is expressible in LE by Lemma 9 and the fact that v1 < v2
is equivalent to (v1 ≤ v2 ∧ ∼ v1 = v2 ). N
Theorem 11 For any number b ≥ 2, the relation v1 ∗b v2 = v3 is Arithmetical.
Proof. The relation v1 ∗b v2 = v3 is expressed by the condition that
v1 · bℓb (v2 ) + v2 = v3 .
For example, 1570 ∗10 365 = 1570365 = 1570000 + 365 = 1570 · 103 + 365 =
1570 · 10ℓ10 (365) + 365.
This condition is equivalent to
∃v4 (bℓb (v2 ) = v4 ∧ ((v1 · v4 ) + v2 ) = v3 ).
By Lemma 10, this relation is Arithmetical. N
Note We could also have defined F [n] as ∃v1 (v1 = n ∧ F (v1 )) since:
Lemma 13 (∀v1 (v1 = n ⊃ F (v1 )) ≡ ∃v1 (v1 = n ∧ F (v1 ))) is logically valid.
Proof. Exercise. N
Our first step in the arithmetization of syntax, i.e. showing that syntactic oper-
ations on expressions of LE can be reflected into arithmetically definable operations
on their Gödel numbers, is to show that the function
s(v1 , v2 ) = p∀v1 (v1 = v2 ⊃ Ev1 )q
is Arithmetical.
The value of the function s(v1 , v2 ) is the Gödel number of a formula logically
equivalent to the substitution of the numeral of the number v2 into the expression
Ev1 when Ev1 is a formula in which v1 occurs free. Note that for some values of v1 ,
Ev1 is a formula in which the variable v1 occurs free, and for other values it is not.
Indeed for some values of v1 , the expression whose Gödel numbers if v1 , i.e. Ev1 ,
will not be a formula, in which case s(v1 , v2 ) is the Gödel number of an expression
which is not a formula—a ‘don’t care’ case. If Ev1 is a formula in which v1 does not
occur free then s(v1 , v2 ) is the Gödel number of a formula that has nothing to do
with substitution of the numeral for v2 into it—another ‘don’t care’ case.
Before we can establish that the two-place function (three-place relation) s(v1 , v2 ) =
v3 is Arithmetical, we must calculate the Gödel numbers of the numerals.
Lemma 14 The Gödel number of the numeral v 2 is 13v2 (in our base-13 assignment
of Gödel numbers ).
v
z }|2 {
Proof. The numeral v2 is the expression 0′ . . . ′ . The symbol 0 is assigned Gödel
number 1, the symbol ′ is assigned the number 0, so that whole expression is, by
v
z }|2 {
concatenation, assigned the number written in base-13 notation as 1 0 . . . 0, which
is 13v2 . N
LECTURE 2 30
Proof. Let k = p∀v1 (v1 =q, a particular number whose base 13 notation—
given our assignment of base 13 digits to the symbols of our language—is 965265η
(or if we use the base 10 Gödel numbering also given in Lecture 1, whose base
10 notation is 899652658999). Given that pv2 q = 13v2 , p⊃q = 8, pEv1 q = v1 ,
p)q = 3, s(v1 , v2 ) = v3 iff ∃v4 (v4 = 13v2 ∧ v3 = k ∗ v4 ∗ 8 ∗ v1 ∗ 3), which by left-
hand association of ∗13 and repeated use of Theorem 11 is Arithmetical, i.e. there
is a formula S(v1 , v2 , v3 ) such that for all numbers m, n, k, S(m, n, k) is true iff
s(m, n) = k. Expressing this argument from Theorem 11 more strictly, the formula
needs to be the following:
Proof. Let D(v1 , v2 ) be the formula that results by substituting v1 for v2 and
v2 for v3 in S(v1 , v2 , v3 ). For all numbers m, n, S(m, m, n) is true iff s(m, m) = n.
By the definition of d(v1 ), s(m, m) = d(m). So D(m, n) is true iff d(m) = n. N
Lecture 3
31
LECTURE 3 32
The proof of the Diagonal Lemma is by a kind of double substitution into the one-
place formula for which a diagonal sentence is being established, first the substitution
of the Arithmetical expression of the diagonal function, and then the substitution
of the numeral for the Gödel number of the formula that results from that first
substitution. Both of these substitutions are quasi-substitutions in this construction.
It might make the idea of the proof more perspicuous if we consider how it goes
with actual substitutions if we have a term s(v1 ) in our language such that s(v1 ) =
pEv1 (v1 )q. (We could have such a language, at the cost of taking more functions as
primitive.)
Theorem 18 (variant diagonal lemma) Given a formula F (v1 ) with one free
variable in a language for arithmetic L that has a term s(v1 ) such that for each
number n, s(n) = pEn (n)q, there is a sentence C in L such that (C ≡ F (pCq)) is
true.
Proof. Consider the formula F (s(v1 )) formed by substituting the term s(v1 ) for
the free occurrences of v1 in F (v1 ). (1) Let k = pF (s(v1 ))q. (2) Let C =df F (s(k)).
(3) Then s(k) = pCq. (4) The numeral s(k) designates the same number as is
designated by the term s(k), i.e. the equation s(k) = s(k) is true, so by substitutivity
of identity, (F (s(k)) ≡ F (s(k))). (5) Hence by (2) and (3), (C ≡ F (pCq)). N
are not specific to decimal notation, even if what is being proved specifically refers
to decimal notation, as in “if the digits of a number in its base 10 notation add up
to 9 then the number is divisible by 9”, which is proved from general properties of
the congruence relation and the fact that 10 ≡ 1 (mod 9).
Proposition 22 v2 ∈ v1 is Arithmetical.
37
LECTURE 4 38
.
1
It was Dedekind who established the first axiomatization of arithmetic, in 1888, which Peano
took over in his publication a year later. Peano cites Dedekind 1888 as the source of his axioms.
It seems to have been Russell who introduced the misnomer Peano Arithmetic.
LECTURE 4 39
However, two formulas within this schema are generated by substitution, namely
F (0) and F (v1′ ), and for ease of arithmetization we want to use quasi-substitution
instead of substitution. We can’t use quasi-substitution directly on F (v1 ) to express
F (v1′ ), since F [v1′ ] would be ∀v1 (v1 = v1′ ⊃ F (v1 )), in which no variable occurs
free, so not equivalent to F (v1′ ). We could change the variable in the auxiliary
quantification, say to v2 if v2 does not occur free in F , i.e. ∀v2 (v2 = v1′ ⊃ F (v2 )),
which is logically equivalent to F (v1′ ). But this involves substitution of the variable
v2 for all free occurrences of v1 in F , which would defeat the purpose of avoiding
substitution. However, we can use a quasi-substitution to obtain from F (v1 ) a
formula logically equivalent to F (vi ), namely ∀v1 (v1 = vi ⊃ F (v1 )) where vi is any
variable that does not occur in F (v1 ). Then we can use quasi-substitution to obtain
a formula without any substitutions that is logically equivalent to F (v1′ ), namely
∀vi (vi = v1′ ⊃ ∀v1 (v1 = vi ⊃ F (v1 )). We abbreviate this formula as F [[v1′ ]]. It’s
easily seen that F [[v1′ ]] is logically equivalent to F (v1′ ). This logical equivalence only
requires that vi does not occur free in F (v1 ), but the sufficient condition that it does
not occur at all in F (v1 ) is easier to express in arithmetized syntax, and that’s the
condition we take.
Definition 35 (the system PAE ) The axioms and rules of inference of PAE are
the following:
A. Logical axioms and rules of inference
Group I propositional logic. All instances of the following schemata:
L1 (F ⊃ (G ⊃ F ))
L2 (F ⊃ (G ⊃ H)) ⊃ ((F ⊃ G) ⊃ (F ⊃ H))
L3 ((∼ F ⊃∼ G) ⊃ (G ⊃ F ))
Group II predicate logic. All instances of the following schemata:
L4 (∀vi (F ⊃ G) ⊃ (∀vi F ⊃ ∀vi G))
L5 (F ⊃ ∀vi F ), provided vi does not occur in F .
L6 ∃vi (vi = t), provided vi does not occur in t.
L7 (vi = t ⊃ (X1 vi X2 ⊃ X1 tX2 )), where X1 and X2 are any expressions such
that X1 vi X2 ) is an atomic formula and t is any term of LE .
Rules of inference
R1 From F and (F ⊃ G), infer G. [Modus Ponens]
R2 From F , infer ∀vi F . [Generalization]
B. Non-logical axioms
Group III axioms specific to each of the primitive non-logical notions of the
language
N1 (v1′ = v2′ ⊃ v1 = v2 )
N2 ∼ 0 = v1′
N3 (v1 + 0) = v1
N4 (v1 + v2′ ) = (v1 + v2 )′
LECTURE 4 40
N5 (v1 · 0) = 0
N6 (v1 · v2′ ) = ((v1 · v2 ) + v1 )
N7 (v1 ≤ 0 ≡ v1 = 0)
N8 (v1 ≤ v2′ ≡ (v1 ≤ v2 ∨ v1 = v2′ ))
N9 (v1 ≤ v2 ∨ v2 ≤ v1 )
N10 v10 = 0′
v′
N11 v12 = (v1v2 · v1 )
Group IV axiom schema of mathematical induction
N12 (F [0] ⊃ (∀v1 (F (v1 ) ⊃ F [[v1′ ]]) ⊃ ∀v1 F (v1 ))),
where, for vi any chosen variable that does not occur in F (v1 ), F [[v1′ ]] is
∀vi (vi = v1′ ⊃ ∀v1 (v1 = vi ⊃ F (v1 ))). Recall that when we write a schematic formula
F (vi ), unless we stipulate otherwise, variables other than vi may occur free in it.
These other free variables are referred to as parameters.
a term, but it’s a delicate question in which languages, i.e. with what primitives,
that bound can or cannot be expressed.
(5) AF (v1 ): Ev1 is an atomic formula, i.e. of the form t1 = t2 or t1 ≤ t2 for t1 , t2
terms.
(∃v2 ≤ v1 )(∃v3 ≤ v1 )(T m(v2 ) ∧ T m(v3 ) ∧ (v1 = v2 ηv3 ∨ v1 = v2 ǫv3 ))
(6) Seqf (v1 ): Ev1 is a formation sequence for formulas, i.e. a finite sequence of
expressions each one of which is either an atomic formula or of the form ∼ E for E
occurring earlier in the sequence or of the form (Ei ⊃ Ej ) for Ei and Ej occurring
earlier in the sequence or of the form ∀vi E for vi any variable and E occurring earlier
in the sequence.
(Seq(v1 )∧(∀v2 ≤ v1 )(v2 ∈ v1 ⊃ (AF (v2 )∨(∃v3 ≤ v1 )(v3 ≺ v2 ∧v2 = 7v3 )∨(∃v3 ≤
v1
v1 )(∃v4 ≤ v1 )(v3 ≺ v2 ∧ v4 ≺ v2 ∧ v2 = 2v3 8v4 3) ∨ (∃v3 ≤ v1 )(∃v4 ≤ v1 )(v3 ≺
v1 v1 v1
v2 ∧ V ar(v4 ) ∧ v2 = 9v4 v3 ))))
(7) F m(v1 ): Ev1 is a formula.
∃v2 (Seqf (v2 ∧ v1 ∈ v2 ))
Note: The remark as at (4) above applies here also. We know by Problem 2
on Problem sheet 1, that we can determine by a finite search whether an expression
is a formula, but it’s a delicate matter to determine in exactly what language of
arithmetic, i.e. with what primitives, this numerical quantifier can be bounded by
a term of the language.
(8) Ax(v1 ): Ev1 is an axiom of PAE . There are seven schemata of logical axioms
L1 − L7 and eleven axioms of arithmetic N1 − N11 plus one axiom schemata of
arithmetic N12 (Induction). We need formulas Li (v1 ) such that Li (v1 ) iff (Ev1 is an
axiom of form Li ) and Ni (v1 ) such that Ni (v1 ) iff (Ev1 is an axiom of form Ni ). The
property of numbers Ax(v1 ) is expressed by (L1 (v1 ) ∨ . . . ∨ L7 (v1 ) ∨ N1 (v1 ) ∨ . . . ∨
N12 (v1 )). Finding N12 (v1 ) is problem 5 on Exercise sheet 2. I will treat a couple of
cases from each of the other groups of axioms.
Logical axioms:
Group I
L1 (v1 ): (∃v2 ≤ v1 )(∃v3 ≤ v1 )(F m(v2 ) ∧ F m(v3 ) ∧ v1 = 2v2 82v3 8v2 33)
L3 (v1 ): (∃v2 ≤ v1 )(∃v3 ≤ v1 )(F m(v2 ) ∧ F m(v3 ) ∧ x = 227v2 87v3 382v3 8v2 33)
Group II
L4 (v1 ): (∃v2 ≤ v1 )(∃v3 ≤ v1 )(∃v4 ≤ v1 )(F m(v2 ) ∧ F m(v3 ) ∧ V ar(v4 ) ∧ v1 =
29v4 2v2 8v3 3829v4 v2 89v4 v3 33)
L7 (v1 ): (∃v2 ≤ v1 )(∃v3 ≤ v1 )(∃v4 ≤ v1 )(∃v5 ≤ v1 )(∃v6 ≤ v1 )(∃v7 ≤ v1 )(V ar(v2 )∧
T m(v3 ) ∧ v6 = v4 v2 v5 ∧ AF (v6 ) ∧ v7 = v4 v3 v5 ∧ v1 = 2v2 ηv3 82v6 8v7 33)
Group III
N1 (v1 ): Ev1 is the axiom N1 , which in primitive notation is
((v′ )′ = (v′′ )′ ⊃ (v′ ) = (v′′ )).
LECTURE 4 43
v1 = 226530η26553082653η265533.
N7 (v1 ): Ev1 is the axiom N7 . To compute the Gödel number of N7 we must
write it in primitive notation. This requires expressing ≡ in terms of ∼ and ⊃,
given by the truth-functional equivalences ((A ≡ B) ≡ (A ⊃ B) ∧ (B ⊃ B)) and
((C ≡ D) ≡∼ (C ⊃∼ D)), which yields ((A ≡ B) ≡∼ ((A ⊃ B) ⊃∼ (B ⊃ B))).
So N7 =∼ (((v′ ≤ 0 ⊃ (v′ ) = 0) ⊃∼ ((v′ = 0 ⊃ (v′ ≤ 0))
v1 = 7222653ǫ182653η138722653η182653ǫ133
(9) P rfP AE (v1 ): Ev1 is a proof in PAE , i.e. a sequence of formulas each one of
which is either an axiom of PAE , or is the result of applying R1 [Modus Ponens]
to two formulas occurring earlier in the sequence, or is the result of applying R2
[Generalization] to a formula occurring earlier in the sequence.
(Seq(v1 ) ∧ (∀v2 ≤ v1 )(v2 ∈ v1 ⊃ (Ax(v2 ) ∨ (∃v3 ≤ v1 )(∃v4 ≤ v1 )(v3 ≺ v2 ∧ v4 ≺
v1 v1
v2 ∧ v4 = 2v3 8v2 3) ∨ (∃v3 ≤ v1 )(∃v4 ≤ v1 )(V ar(v3 ) ∧ v4 ≺ v2 ∧ v2 = 9v3 v4 )))))
v1
(10) P rovP AE (v1 , v2 ): Ev1 is proved by Ev2 .
P rovP AE (v1 , v2 ) ≡ (P rfP AE (v2 ) ∧ v1 ∈ v2 ))
(11) P rP AE (v1 ): Ev1 is provable
∃v2 P rovP AE (v1 , v2 ).
Theorem 24 (Arithmetical proof relation) The two place relation between num-
bers m and n given by the condition that n is the Gödel number of a proof in PAE
of the formula whose Gödel number is m is expressible in LE .
Proof. The formula P rovP AE (v1 , v2 ) in (10) expresses “Ev2 is a proof of Ev1 ”.
This is evident from this formula and (1) - (9). N
Remark: As the construction of the proof predicate for PAE shows, arithmeti-
zation by assignment of digits to symbols and of concatenation of corresponding se-
quences of digits (numbers in the given base notation) to concatenation of sequences
of symbols (expressions) makes the correspondence between formal expressions and
numbers, which Hilbert recognized (see quotation in the first lecture) completely
direct. What Gödel achieved, going beyond Hilbert’s insight, was to show that the
formal syntax of strings of symbols by which a formal system of proof is established
corresponds exactly with arithmetically definable properties of the corresponding
numbers.
LECTURE 4 44
Proof. Having shown that the property of being the Gödel number of a theorem
of PAE is Arithmetical (expressible in LE ), the existence of a true sentence in the
language of arithmetic not provable in PAE follows immediately from Tarski’s theo-
rem, that the set of Gödel numbers of true sentences in the language of arithmetic is
not arithmetical (Exercise sheet 1 problem 4) and the hypothesis that every sentence
provable in PAE is true: By the fact that the Gödel numbers of provable formulas
is arithmetical, we have that this set is not identical with the set of Gödel numbers
of true sentences. If every sentence provable in PAE is true and the set of provable
sentences does not coincide with the set of true sentences, there must exist a true
sentence which is not provable. N
This argument via Tarski’s theorem is highly inefficient since it fails to generate
a particular true sentence that is unprovable in the given system while at the same
time it requires the construction of an arithmetized proof predicate, from which we
can by the Diagonal Lemma, itself used in the proof of Tarski’s theorem, obtain a
particular sentence, namely a diagonal sentence for ∼ P rP AE (v1 ) which we can show,
on the hypothesis that every sentence provable in PAE is true, is unprovable and
(thereby) true, and so whose negation is false and so by the hypothesis unprovable.
We have thus
.
(Tuesday, 26 October 2010)
45
LECTURE 5 46
Definition 38 (the system PA) The language L for PA is obtained from the lan-
guage LE for PAE by dropping the condition in the definition of terms for LE , Def-
inition 12, that if t1 and t2 are terms, then so is (t1 f′′′ t2 ), which correspondingly
removes from formulas of the language LE any expressions that contain the expres-
sion f′′′ (without having to make any change to Definition 16). The axioms of PA
are obtained from those of PAE by dropping axioms N10 and N11 (which are not
formulas in the language L of PA).
By simple modification of the construction of an Arithmetical proof predicate
for PAE in Section 4.2, we obtain an Arithmetical proof predicate for PA.
Theorem 28 (proof predicate for PA in LE ) There are formulas P rfPEA (v1 ),
P rovPEA (v1 , v2 ), and P rPEA (v1 ) in the language LE that express the property of being
the Gödel number of a proof sequence for PA, the relation of being the Gödel number
of a formula that occurs in the proof sequence coded by a given number, and the
property of being the Gödel number of a theorem of PA.
Proof. We do this by by modifying the constructions for PAE in Theorem 24 and
Corollary 25. We drop the disjunct corresponding to term formation by the function
expression f′′′ , i.e. v2 = 2v3 4555v4 3, so that Seqt(v1 ) is (Seq(v1 ) ∧ (∀v2 ≤ v1 )(v2 ∈
v1 ⊃ (V ar(v2 ) ∨ N um(v2 ) ∨ (∃v3 ≤ v1 )(v3 ≺ v2 ∧ v2 = v3 0) ∨ (∃v3 ≤ v1 )(∃v4 ≤
v1
v1 )(v3 ≺ v2 ∧ v4 ≺ v2 ∧ (v2 = 2v3 45v4 3 ∨ v2 = 2v3 455v4 3))))). In the formula Ax(v1 )
v1 v1
that expresses “Ev1 is an axiom of PAE ” (p. 39) the disjuncts N10 (v1 ) and N11 (v1 ) are
dropped. With these modifications, the formulas for P rfP AE (v1 ), P rovP AE (v1 , v2 ),
and P rP RE (v1 ) on p. 40 are transformed to corresponding Arithmetical formulas
P rfPEA (v1 ), P rovPEA (v1 , v2 ), and P rPEA (v1 ). N
5.2 Σ0-formulas
Definition 39 (bounded quantifiers) For vi any variable, and n any numeral,
quantification in either of the forms ∀vi (vi ≤ n ⊃ F ) or ∃vi (vi ≤ n ∧ F ) is called
bounded quantification, abbreviated as (∀vi ≤ n)F and (∃vi ≤ n)F , respectively.
Also, for vi any variable, vj any variable such that i 6= j, and F any formula,
quantification in either of the forms ∀vi (vi ≤ vj ⊃ F ) and ∃vi (vi ≤ vj ∧ F ) are called
bounded quantifiers, abbreviated as (∀vi ≤ vj )F and (∃vi ≤ vj )F , respectively.
Remark. The restriction that the variable vj be distinct from the variable vj
when the bound on the quantification is a variable is essential, since ∀vi (vi ≤ vi ⊃ F )
is logically equivalent to ∀vi F , which is unbounded quantification. Also, note that
bounded existential quantifications are, in primitive notation, formulas of the form
∼ ∀vi ∼ (vi ≤ n ∧ F ) and ∼ ∀vi ∼ (vi ≤ vj ∧ F ).
LECTURE 5 47
Corollary 29 For F any Σ0 -formula, vi any variable, and t either a variable dif-
ferent from vi or a numeral, ∃vi (vi ≤ t ∧ F ) is Σ0 .
For reasons that will be apparent shortly, we also label Σ0 -formulas as Π0 and
as ∆0 .
Proof. We show that the relation f (v1 , . . . , vn ) 6= vn+1 is also Σ1 , from which
it follows that the relation f (v1 , . . . , vn ) = vn+1 is ∆1 . By the hypothesis that f is Σ1 ,
there is a Σ0 -formula F (v1 , . . . , vn , vn+1 , vn+2 ) such that ∃vn+2 F (v1 , . . . , vn , vn+1 , vn+2 )
expresses the relation f (v1 , . . . , vn ) = vn+1 . Then, since f is total so that, for every
v1 , . . . , vn , there is vn+1 such that f (v1 , . . . , vn ) = vn+1 , the relation f (v1 , . . . , vn ) 6=
vn+1 is expressed by the condition ∃vn+3 ∃vn+2 (F (v1 , . . . , vn , vn+3 , vn+2 )∧ ∼ vn+3 =
vn+1 ). This condition is equivalently expressed by
∃vn+4 (∃vn+3 ≤ vn+4 )(∃vn+2 ≤ vn+4 )(F (v1 , . . . , vn , vn+3 , vn+2 )∧ ∼ vn+3 = vn+1 )
By Definition 40,
(∃vn+3 ≤ vn+4 )(∃vn+2 ≤ vn+4 )(F (v1 , . . . , vn , vn+3 , vn+2 )∧ ∼ vn+3 = vn+1 )
Proof.
1. ((∃z ≤ y)(x · z = y)∧ ∼ x = 0). Note that by this condition every non-zero
number divides 0 (which reflect the usual practice in number theory that 0 is a
LECTURE 5 50
member of every principal ideal on the natural numbers) and 0 does not divide 0
by the second conjunct, nor any non-zero number by the first conjunct (cf problem
3(a) on Problem sheet 1).
2. (∀z ≤ x)((z | x ∧ z 6= 1) ⊃ p | z).
3. (P owp (y) ∧ y > x ∧ y > 1 ∧ (∀z < y) ∼ (P owp (z) ∧ z > x ∧ z > 1). N
Proof.
x ∗p y = z is Σ0 iff pℓp (y) + y = z iff (∃v1 ≤ z)(v1 = pℓp (y) ∧ ((x · v1 ) + y) = z).
The result follows by part 3 of Lemma 35. N
Lemma 37 The Arithmetical relation xPb y (‘x is part of y’) for base b a prime
number is Σ0 .
Proof.
We need to describe two things (dependently related to each other): a process
whereby a set of ordered pairs of numbers (a1 , b1 ), (a2 , b2 ), . . . , (ar , br ) is coded by a
number k, and a formula K(v1 , v2 , v3 ) whereby, given a code number k, the set of
ordered pairs is decoded.
LECTURE 5 51
Proof. The relation xy = z holds if and only if there is a set of ordered pairs
{(0, 1), (1, x), (2, x2 ), . . . , (y, xy )} and (y, z) is a member of that set. Given the coding
and Σ0 -decoding of finite sets of ordered pairs of numbers by Theorem 38, we can
express this by the formula ∃w(K(y, z, w) ∧ (∀u ≤ w)(∀v ≤ w)(K(u, v, w) ⊃ ((u =
0 ∧ v = 1) ∨ (∃r ≤ w)(∃s ≤ w)(K(r, s, w) ∧ r′ = u ∧ v = x · s))). Since K(v1 , v2 , v3 ) is
Σ0 , this whole formula is Σ1 . Notice that for this result we need the second clause of
Theorem 38, i.e. that for any numbers v1 , v2 , v3 , if K(v1 , v2 , v3 ) holds, then v1 ≤ v3
and v2 ≤ v3 in order to bound the universal quantifiers inside the formula, so that
it is, indeed, Σ1 . N
Proof. The proof is too complicated to give here. See Petr Hájek and Pavel
Pudlák, Metamathematics of First-Order Arithmetic, Springer, Berlin, 1993, pp.
299-303. N
Theorem 41 and its proof are specific to exponentiation and do not generalize
to other functions. On the other hand, the result in Theorem 39 with Corollary 40
generalizes to the following important result:
Proof. Later. N
Lecture 6
6.1 Σ-formulas
We now define a class of formulas each of which is provably equivalent in PA to a
Σ1 -formula, but which are more flexibly expressed than is required in order to be a
Σ1 -formula.
Definition 47 (Σ-formula) . Base: Every Σ0 -formula is a Σ-formula.
Recursion: 1. If F is a Σ-formula, then for any variable vi , the formula ∃vi F is
a Σ-formula.
2. For any Σ-formulas F and G, (F ∨ G) and (F ∧ G) are Σ-formulas.
3. For any Σ0 -formula F and Σ-formula G, (F ⊃ G) is a Σ-formula.
4. If F is a Σ-formula, then for any distinct variables vi and vj , (∃vi ≤ vj )F and
(∀vi ≤ vj )F are Σ-formulas, and for any number n, (∃vi ≤ n)F and (∀vi ≤ n)F are
Σ-formulas.
To show that every Σ-formula is provably equivalent in PA to a Σ1 -formula we
prove the following five Lemmas, corresponding to the five clauses in the definition
of a Σ-formula.
53
LECTURE 6 54
Proof. Any formula F (vi1 , . . . , vik ) is logically equivalent to ∃vik+1 F (vi1 , . . . , vik ),
where vik+1 does not occur free in F (vi1 , . . . , vik ), so by the completeness of first-order
logic in PA, PA ⊢ (F (vi1 , . . . , vik ) ≡ ∃vik+1 F (vi1 , . . . , vik )) N
Lemma 44 If ∃vi F (vi ) is a Σ1 -formula, then ∃vj ∃vi F (vi ) is provably equivalent in
PA to a Σ1 -formula
Lemma 45 Let ∃vi F (vi ) and ∃vj G(vj ) be Σ1 -formulas. Then (∃vi F (vi ) ∨ ∃vj G(vj ))
and (∃vi F (vi ) ∧ ∃vj G(vj )) are each provably equivalent in PA to a Σ1 -formula
Proof.
(a) ((∃vi F (vi ) ∨ ∃vj G(vj )) ≡ ∃vk (F (vk ) ∨ G(vk ))) is logically valid (for vk any
variable substitutable into F and G) (Cf. Problem sheet 1 problem 5(a)), so provable
in PA.
(b) ((∃vi F (vi ) ∧ ∃vj G(vj )) ≡ ∃vi ∃vj (F (vi ) ∧ G(vj ))) is logically valid (on the
assumption, without loss of generality, that vi does not occur free in G and vj does
not occur free in F ). Hence this equivalence is provable in PA. By Lemma 44,
⊢ (∃v1 ∃v2 (F (vi ) ∧ G(vj )) is provably equivalent in PA to a Σ1 formula. N
Lemma 47 For ∃vi R(vi , vj ) a Σ1 -formula, the following formulas are provably equiv-
alent in PA to Σ1 -formulas.
(a) (∃vj ≤ vk )∃vi R(vi , vj )
(b) (∀vj ≤ vk )∃vi R(vi , vj )
and similarly for the formulas in (a) and (b) with a numeral in place of the
variable vk .
Proof.
(a) The formula vj ≤ vk is Σ0 by the definition of Σ0 . Hence by Lemma 43 it is
provably equivalent to a Σ1 formula. Hence by Lemma 45 (vj ≤ vk ∧ ∃vi R(vi , vj )) is
provably equivalent to a Σ1 -formula. Hence by Lemma 44, ∃vj (vj ≤ vk ∧∃vi R(vi , vj )),
which we abbreviate as (∃vj ≤ vk )∃vi R(vi , vj ), is provably equivalent to a Σ1 -
formula. This argument holds also for a numeral n in place of the variable vk .
(b) Exercise (Problem sheet 3 Problem 3(b)). N
LECTURE 6 55
Proof. By induction on the recursive definition of Σ-formulas, the base case and
each induction step established by one of Lemmas 43 - 47. N
A much stronger result than Theorem 49 is true and is needed for what is to
come, namely that the arithmetized proof predicate for PA is Σ1 . The first step is
to show that the two-place formula (P rfP A (v2 ) ∧ v1 ∈ v2 ) is Σ1 . With considerable
effort we can actually establish that it’s Σ0 . We know from Problem 2 on Problem
sheet 1 that whether a string of symbols is a term or a formula is decidable, so we
know that there is a bound on those quantifiers, but giving an explicit formulation
of that bound in the language of PA is hard work which we can avoid. However, for
the overall result that the proof predicate for PA is Σ1 it’s sufficient to show that
(P rfP A (v2 ) ∧ v1 ∈ v2 ) is Σ1 .
Lemma 50 The formula (P rfP A (v2 )∧v1 ∈ v2 ), i.e. (Seq(v2 )∧(∀v3 ≤ v2 )(v3 ∈ v2 ⊃
(Ax(v3 ) ∨ (∃v4 ≤ v2 )(∃v5 ≤ v2 )(v4 ≺ v3 ∧ v5 ≺ v3 ∧ v5 = 2v4 8v2 3) ∨ (∃v4 ≤ v2 )(∃v5 ≤
v2 v2
v2 )(V ar(v4 ) ∧ v5 ≺ v3 ∧ v3 = 9v4 v5 ))) ∧ v1 ∈ v2 ) is equivalent to a Σ1 -formula.
v2
Proof. The key point is that the only place in the construction of P rfP A (v2 ) in
which we used an unbounded quantifier was in T m(v1 ) (Ev1 is a term) as ∃v2 (Seqt(v2 )∧
v1 ∈ v2 ) and F m(v1 ) (Ev1 is a formula) as ∃v2 (Seqf (v2 ∧ v1 ∈ v2 )). These occur
the formula Ax(v1 ) in L1 − L7 and L12 (Induction axioms). In none of these oc-
currences is it in the antecedent of a conditional, and the occurrence of Ax(v1 ) in
(P rfP A (v2 ) ∧ v1 ∈ v2 ) is also not in the antecedent of a conditional. Hence in a
prenex normal form of (P rfP A (v2 ) ∧ v1 ∈ v2 ) these several existential quantifiers
come out as prenex existential quantifiers. By Lemma 44, this formula is Σ1 . N
Corollary 52 {n : P A ⊢ En } is Σ1 .
Proposition 54 {n : PA ⊢ En [n]} is Σ1 .
In order to generalize Theorem 48, and for many other purposes, we require a
Σ0 -pairing function.
Proof. For a given formula, find a prenex normal form for it. By Theorem 56
(3), adjacent like quantifiers can be collapsed to a single quantifier. N
Lecture 7
Σ0-completeness and
Σ1-completeness; weak systems of
arithmetic Q and R (without
induction); Σ0-completeness of
systems R, Q, and PA;
Σ0-soundness and Σ1-soundness
58
LECTURE 7 59
Proof. Half of the condition that S correctly decides every Σ0 -sentence is that
if X is a true Σ0 -sentence then S ⊢ X, i.e. Σ0 -completeness of S.
Conversely, suppose S is Σ0 -complete, and X is any false Σ0 -sentence. Then
∼ X is a true Σ0 -sentence, so by Σ0 -completeness, S ⊢∼ X, as required. N
N2 ∼ v1′ = 0
N3 v1 + 0 = v1
N4 v1 + v2′ = (v1 + v2 )′ .
N5 v1 · 0 = 0
N6 v1 · v2′ = (v1 · v2 ) + v1
N7 (v1 ≤ 0 ≡ v1 = 0)
N8 (v1 ≤ v2′ ≡ (v1 ≤ v2 ∨ v1 = v2′ )).
N9 (v1 ≤ v2 ∨ v2 ≤ v1 ).
The system Q is a variant of one due to Raphael Robinson. We will show that
all true Σ0 -sentences are provable Q and so provable in PA since all the axioms of
Q are axioms of PA. We prove this by proving a yet stronger result, namely that an
even weaker system R, also due to Raphael Robinson, is Σ0 -complete. Instead of
the (finitely) many recursion axioms of PA and Q, it has as axioms infinitely many
instances of computations of addition, multiplication, and inequality, in three axiom
schemata, and two axiom schemata expressing properties of ≤.
Definition 55 (system R) The axioms of R are all sentences sentences and for-
mulas of L generated from natural numbers m and n by the following axiom schemata:
Ω1 m + n = m + n.
Ω2 m · n = m · n.
Ω3 ∼ m = n where m 6= n.
Ω4 (v1 ≤ n ⊃ (v1 = 0 ∨ . . . ∨ v1 = n)).
Ω5 (v1 ≤ n ∨ n ≤ v1 ).
Proof.
′
Lemma 64 For each number k, (R) ⊢ ((v1 = 0 ∨ . . . ∨ v1 = k) ⊃ v1 ≤ k )
Proof. The results follows by ∨-elimination from (n = 0 ∨ n 6= 0). The case for
n = 0 is an instance of Lemma 63, i.e. R ⊢ (v1 = 0 ⊃ v1 ≤ 0). If n 6= 0 then there
is some number k such that n = k + 1. This case follows by ∨-elimination from
′ ′
the disjunction ((v1 = 0 ∨ . . . ∨ v1 = k) ∨ v1 = k ). We have v1 ≤ k from the first
′
disjunct by Lemma 64 and v1 ≤ k from the second disjunct by Lemma 63. So by
′ ′
⊃-Intro, R ⊢ (((v1 = 0 ∨ . . . ∨ v1 = k) ∨ v1 = k ) ⊃ v1 ≤ k ). N
Proposition 68 R is a subsystem of Q.
Proposition 69 Q is Σ0 -complete.
Theorem 70 PA is Σ0 -complete.
Proof. (i) Left to right: We prove the contrapositive. Suppose for every X,
S ⊢ X. Then in particular for any formula Y , S ⊢ Y and S ⊢∼ Y .
(ii) Right to left: We prove the contrapositive. Suppose S is inconsistent, i.e.
there is a formula Y such that S ⊢ Y and X ⊢∼ Y . Then by ∧-introduction (as a
derived rule if not a primitive rule of S), S ⊢ (Y ∧ ∼ Y ). Propositional logic proves
66
LECTURE 8 67
Proof. Let S be a system whose language contains numerals for the natural
numbers and which is sound with respect to truth in arithmetic. Suppose S ⊢
∃wF (w). Then ∃wF (w) is true, i.e. there is a natural number n such that F (n) is
true, which is to say that ∼ F (n) is false. So S 0∼ F (n), which is to say that S is
ω-consistent. N
The converse holds only to a strictly limited extent, as detailed by the following
two propositions, and in general ω-consistency does not imply truth.
Note that in (ii) of the proof of Theorem 78 the hypothesis that S is Σ0 -complete
is not needed. This part of the proof of Theorem 78 is a sharpening of the argument
for Proposition 75.
Proof. By the first half of Theorem 78 and the fact that a Σ1 -sound system is
consistent (since there are sentences it doesn’t prove, namely false Σ1 -sentences).
Or we can prove the contrapositive, as in the proof of Proposition 73, i.e. an
inconsistent theory proves everything so in particular a 1-inconsistency. N
However, Gödel’s original proof made no use of the extra strength of ω-consistency
over 1-consistency, so the proof from 1-consistency is really an improvement in clar-
ity rather than in strength.
Proof. The proof is by vacuous quantification. Let S ⊢ ∃v1 F (v1 ) for F (v1 ) a
Σ0 -sentence (i.e. no free variables). Then S ⊢ ∃v1 ∀v2 F (v1 ), and so by Σ2 -soundness,
∃v1 ∀v2 F (v1 ) is true. Then ∃v1 F (v1 ) is true. N
Proof. (i) Suppose S is 2-consistent and suppose S ⊢ ∃v1 ∀v2 F (v1 , v2 ), where
F (v1 , v2 ) is a Σ0 -formula, and ∃v1 ∀v2 F (v1 , v2 ) is false, which is to say that for
each natural number n, ∃v2 ∼ F (n, v2 ) is a true Σ1 -sentence. Then by the Σ1 -
completeness of every Σ0 -complete theory and predicate logic, for each natural num-
ber n, S ⊢∼ ∀v2 F (n, v2 ). But then S is 2-inconsistent. So by RAA, ∃v1 ∀v2 F (v1 , v2 )
is true.
(ii) Suppose S is Σ2 -sound and suppose S ⊢ ∃v1 ∀v2 F (v1 , v2 ). Then ∃v1 ∀v2 F (v1 , v2 )
is true, so for some number k, ∀v2 F (k, v2 ) is true. Suppose S ⊢∼ ∀v2 F (k, v2 ).
LECTURE 8 70
Proof. This is equivalent to the first half of Theorem 83, by Lemma 84. N
Corollary 86 Theorem 85 gives a variant proof of the second half of Theorem 83,
the First Incompleteness Theorem.
Proof. Suppose G, i.e. ∀v2 ∼ A(a, v2 ), is true. Then every number is not the
code of a proof of Ea [a], which is to say that PA 0 G. But if there is any sentence
that a system doesn’t prove then the system is consistent, which is to say that PA
is consistent. N
There is another argument to show that G is true which is weaker than the ar-
gument for Theorem 85 because it requires a stronger hypothesis, but the argument
itself is of independent interest.
Theorem 88 If a system is Σ0 -complete and does not prove ∃vi F (vi ) for F (vi ) a
Σ0 -formula, then ∃vi F (vi ) is false.
8.4 PA is ω-incomplete
Definition 63 (ω-completeness) A system S in a language containing a numeral
n for each natural number n is ω-complete if for every formula F (v1 ) in the language
of S such that for each natural number n S ⊢ F (n), S ⊢ ∀vi F (vi ) If S is not ω-
complete we say that S is ω-incomplete.
74
LECTURE 9 75
a different sentence from the Gödel sentence which is provably undecidable just on
the hypothesis of consistency of the system.
The Rosser Incompleteness Theorem can be proved from a separation property,
itself of independent interest and which we use also in proving that the diagonal
equivalence in the diagonal lemma is not only true, as we have seen, but also formally
provable.
Proof. Exercise. N
separates A from B in S.
Proof. In order to shorten formulas on the page I shall take A and B to be unary
relations. The proof for A and B as k-ary relations is just a notational variant of
this proof.
i) To show: if n ∈ A, then S ⊢ ∃x(F (n, x) ∧ (∀y ≤ x) ∼ G(n, y)).
LECTURE 9 76
(1) n ∈ A Assumption
(2) there exists k such that S ⊢ F (n, k) (1) and enumeration of A by F (v1 , v2 ) in S
(3) n ∈
/B (1) and the hypothesis that A and B as disjoint
(4) for every m, S ⊢∼ G(n, m) (3) and enumeration of B by G(v1 , v2 ) in S
(5) S ⊢ ((y = 0 ∨ . . . ∨ y = k) ⊃∼ G(n, y)) (4) and substitutivity of =, ∨-elim, ⊃-intro
(6) S ⊢ (y ≤ k ⊃∼ G(n, y)) (5), instance of Ω4 and prop. logic
(7) S ⊢ ∀y(y ≤ k ⊃∼ G(n, y)) (6) by ∀-Intro
(8) S ⊢ (F (n, k) ∧ ∀y(y ≤ k ⊃∼ G(n, y))) (2) (7) ∧-Intro
(9) S ⊢ ∃x(F (n, k) ∧ ∀y(y ≤ k ⊃∼ G(n, y))) (8) ∃-intro
(10) S ⊢ ∃x(F (n, x) ∧ (∀y ≤ x) ∼ G(n, y)) (9) definition of (∀y ≤ x)
(1) n ∈ B Assumption
(2) there exists k such that S ⊢ G(n, k) (1) and enumeration of B by G(v1 , v2 ) in S
(3) n ∈
/A (1) and the hypothesis that A and B as disjoint
(4) for every m, S ⊢∼ F (n, m) (3) and enumeration of A by F (v1 , v2 ) in S
(5) S ⊢ ((y = 0 ∨ . . . ∨ y = k) ⊃∼ F (n, y)) (4) and substitutivity of =, ∨-elim, ⊃-intro
(6) S ⊢ (y ≤ k ⊃∼ F (n, y)) (5), instance of Ω4 and prop. logic
(7) S ⊢ (F (n, y) ⊃∼ y ≤ k) (6) by prop logic (contraposition)
(8) S ⊢ (F (n, y) ⊃ k ≤ y) (7) by Ω5 and prop logic
(9) S ⊢ (F (n, y) ⊃ (k ≤ y ∧ G(n, k))) (2) (8) propositional logic
(10) S ⊢ ∃y(F (n, x) ⊃ (y ≤ x ∧ G(n, y))) (9) ∃-Intro
(11) S ⊢ (F (n, x) ⊃ ∃y(y ≤ x ∧ G(n, y))) (10) predicate logic (anti-prenexing)
(12) S ⊢ ∀x(F (n, x) ⊃ ∃y(y ≤ x ∧ G(n, y))) (11) ∀-Intro
(13) S ⊢ ∀x(F (n, x) ⊃ (∃y ≤ x)G(n, y))) (12) definition of (∃y ≤ x) N
Note that in the proof of the Separation Lemma, the argument for (ii) uses both
Ω4 and Ω5 while the argument for (i) uses just Ω4 .
Proof. By Theorem 93, Lemma 94, and the fact that the sets {n : PA ⊢ En [n]}
and {n : PA ⊢∼ En [n]} are enumerable in PA.
This notion is also sometimes called expressibility (e.g. Elliott Mendelson, In-
troduction to Mathematical Logic, 4th edn, Chapman and Hall, 1997, p. 170), or
numeralwise expressibility (e.g. Stephen Cole Kleene, Introduction to Metamathe-
matics, D. Van Nostrand, 1950, p. 195).
It’s easy to show that strong definability implies weak definability (Proposi-
tion 96). The converse also holds but is considerably more complicated to prove
(Theorem 97).
Proof. To reduce clutter I give the proof for the case of a unary function, which
is also the case of immediate interest since the diagonal function is unary. The
argument for the general case is a notational variant.
Let F (x, y) be a formula that weakly defines f (x) in S, i.e.
(1) If f (a) = b, then S ⊢ F (a, b).
(2) If f (a) 6= b, then S ⊢∼ F (a, b).
Let G(x, y) be the formula (F (x, y)∧∀z(F (x, z) ⊃ y ≤ z)). We show that G(x, y)
strongly defines f (x) in S, i.e. if f (a) = b then S ⊢ (G(a, b) ∧ ∀y(G(a, y) ⊃ y = b)).
We show that S proves both conjuncts, on the assumption that f (a) = b.
(i) To establish the first conjunct, i.e. S ⊢ G(a, b), i.e.
S ⊢ (F (a, b) ∧ ∀z(F (a, z) ⊃ b ≤ z)):
LECTURE 9 79
Proof. (1) If the graph of a total function is Σ1 , the complement of its graph
is also Σ1 ) (Lemma 34 in Lecture 5). (2) The diagonal function is total and its
graph is expressed by a Σ1 -formula, ∃v3 A(v1 , v2 , v3 ), i.e. d(a) = b iff ∃v3 A(a, b, v3 ) is
true. (3) By (1) and (2), the complement of its graph is expressed by a Σ1 -formula
∃v3 B(v1 , v2 , v3 ), i.e. d(a) 6= b iff ∃v3 B(a, b, v3 ) is true. (4) Suppose d(a) = b; then
by (2) ∃v3 A(a, b, v3 ) is true, so there is a number m such that A(a, b, m) is true. (5)
By Σ0 -completeness of S, since S is an extension of R, S ⊢ A(a, b, m). (6) Suppose
d(a) 6= b; then by (2), ∃v3 A(a, b, v3 ) is false, so for every m, ∼ A(a, b, m) is true. (6)
By Σ0 -completeness of S, since S is an extension of R, for each m, S ⊢∼ A(a, b, m).
(7) Hence the d(v1 = v2 is enumerated in S by A(v1 , v2 , v3 ). (8) Similarly, d(v1 6= v2 is
enumerated in S by B(v1 , v2 , v3 ). (9) Hence by the Separation Lemma (Theorem 92,
the formula ∃x(A(v1 , v2 , x) ∧ (∀y ≤ x) ∼ B(v1 , v2 , y)) separates d(v1 ) = v2 in S, i.e.
if d(a) = b, then S ⊢ ∃x(A(a, b, x) ∧ (∀y ≤ x) ∼ B(a, b, y)) and if d(a) 6= b, then
S ⊢∼ ∃x(A(a, b, x)∧(∀y ≤ x) ∼ B(a, b, y)), which is to say that d(v1 ) = v2 is weakly
defined by ∃x(A(v1 , v2 , x)∧(∀y ≤ x) ∼ B(v1 , v2 , y)) in S. (10) Then by Theorem 97,
d(v1 ) = v2 is strongly definable in S. N
Proof. Let F (v1 , v2 ) be the formula that strongly defines f (x) in S. For given
formula G(v1 ) let H(v1 ) =df ∃v2 (F (v1 , v2 )∧G(v2 )). Let n and m be such that f (n) =
m. We establish the provability of the two halves of the required biconditional as
follows:
(i) To show that S ⊢ (G(f (n)) ⊃ H(n)): Since F (v1 , v2 ) strongly defines f (x), by
clause (1) S ⊢ F (n, m). Hence by propositional logic in S, S ⊢ (G(m ⊃ (F (n, m) ∧
G(m)). Then by ∃-introduction in S (note that in inferring ∃v1 A(v1 ) from A(t), v1 is
substituted for some but not necessarily all occurrences of t in A(t)), S ⊢ (G(m) ⊃
∃v2 (F (n, v2 ) ∧ G(v2 ))).
(ii) To show that S ⊢ (H(n) ⊃ G(f (n))): By ∀-elimination from the second
conjunct of the condition for strong definability, S ⊢ (F (n, v2 ) ⊃ v2 = m). Therefore
LECTURE 9 81
by propositional logic in S,
S ⊢ ((F (n, v2 ) ∧ G(v2 )) ⊃ (v2 = m ∧ G(v2 ))). Since ((v2 = m ∧ G(v2 )) ⊃ G(m)) is
logically valid (by substitutivity of identity), S ⊢ ((v2 = m ∧ G(v2 )) ⊃ G(m)), so
by propositional logic in S, ⊢ (F (n, v2 ) ∧ G(v2 )) ⊃ G(m). Hence by ∀-introduction,
S ⊢ ∀v2 (F (n, v2 ) ∧ G(v2 ) ⊃ G(m)), and so by anti-prenexing,
S ⊢ (∃v2 (F (n, v2 ) ∧ G(v2 )) ⊃ G(m)) N
Theorem 100 For S any extension of R, for each formula F (v1 ) in the language of
S, with v1 its only free variable, there is a sentence C such that S ⊢ (C ≡ F (pCq)).
Proof. By Theorem 99, given F (v1 ), there is a formula H(v1 ) such that for
each n, S ⊢ (H(n) ≡ F (d(n))). In particular, for h = pH(v1 )q, S ⊢ (H(h) ≡
F (d(h))). By the construction of d(x), d(h) = pH[h]q. We have noted before that
for any formula F (v1 ), (F (n) ≡ F [n]) is logically valid, so S ⊢ (H(h) ≡ H[h]). So
S ⊢ (H[h] ≡ F (d(h))). Thus taking C =df H[h], we have S ⊢ (C ≡ F (pCq)), as
required. N
Lecture 10
Arithemization of consistency;
provability predicates; Gödel’s
Second Incompleteness Theorem;
Löb’s Theorem; analyzing and
strengthening the First
Incompleteness Theorem
82
LECTURE 10 83
En } and X any sentence in the language of S such that S 0 X, we let ConS , the
formal expression in the language of S of the consistency of S be ∼ P r(pXq).
The notation ConS does not notate relativity to the sentence X, nor relativity
to the chosen Gödel numbering and arithmetization of syntax. I will say something
later about provable invariance of the Gödel sentence and of ConS with respect to
Gödel numbering and arithmetization of syntax. As to relativity of ConS to the
unprovable sentence X, we have the following
Theorem 102 For P r(v1 ) the arithmetical proof predicate for PA constructed for
Theorem 28, and X and Y any sentences in the language of PA,
PA⊢ (P r(pX ⊃ Y q) ⊃ (P r(pXq) ⊃ P r(pY q))).
Proof.
P4 : If S ⊢ (X ⊃ Y ), then by P1 , S ⊢ P (p(X ⊃ Y )). Then by P2 and Modus
ponens, S ⊢ (P (pXq) ⊃ P (pY q)).
P5 : By P4 , S ⊢ (P (pXq) ⊃ P (p(Y ⊃ Z)q)). By P2 , S ⊢ (P (p(Y ⊃ Z)q) ⊃
(P (pY q) ⊃ P (pZq))). Then by propositional logic, S ⊢ (P (pXq) ⊃ (P (pY q) ⊃
P (pZq)).
[Notice that there is no corresponding result for S ⊢ ((X ⊃ Y ) ⊃ Z).]
P6 : If S ⊢ (X ⊃ (P (pXq) ⊃ Y )), then by P5
S ⊢ (P (pXq) ⊃ (P (pP (pXq)q) ⊃ P (pY q))). Then by L2 and Modus ponens,
S ⊢ ((P (pXq) ⊃ P (pP (pXq)q)) ⊃ (P (pXq) ⊃ P (pY q))) Then by P3 and Modus
ponens, S ⊢ (P (pXq) ⊃ P (pY q)). N
LECTURE 10 85
Proof.
Remark 1: Line (5) of this proof of the Second Incompleteness Theorem gives
a formal proof in S of the first half of the First Incompleteness Theorem for S, i.e.
if S is consistent, S 0 G.
Proof. By Proposition 105 and line (7) of the proof of Theorem 104. N
Theorem 107 (Löb’s theorem) Let S be a system in which the Diagonal Lemma
is provable, and let P (v1 ) be a provability predicate for S. For X any formula in the
language of S, if S ⊢ (P (pXq) ⊃ X), then S ⊢ X.
Theorem 110 For P (v1 ) a provability predicate for a consistent theory S, let G be
a sentence in the language of S such that S ⊢ (G ≡∼ P (pGq)), let S ∗ be the system
S ∪{∼ G}, which is consistent, given the consistency of S, let P ∗ (v1 ) be a provability
predicate for S ∗ . Let G∗ be a sentence such that S ∗ ⊢ (G∗ ≡∼ P ∗ (pG∗ q)). Then
S ∗ ⊢∼ G∗ .
Proof. (1) Let X be such that S ⊢∼ X. (2) Then also (by thinning), S ∗ ⊢∼ X.
(3) S ⊢ (X ⊃ (∼ G ⊃ X)), by propositional logic, for any sentence X. Then (4)
S ⊢ (P (pXq) ⊃ P (p(∼ G ⊃ X)q, by property of a provability predicate, and so
by thinning, (5) S ∗ ⊢ (P (pXq) ⊃ P (p(∼ G ⊃ X)q. Then (6) S ∗ ⊢ (P (pXq) ⊃
P ∗ (pXq)) since for all Y , S ∗ ⊢ (P (p(∼ G ⊃ Y )q) ≡ P ∗ (pY q)). By the proof of the
Second Incompleteness Theorem, G is provably equivalent in S to ConS, and so also
by thinning in S ∗ , i.e. (7) S ∗ ⊢ (G ≡∼ P (pXq)), and similarly (8) S ∗ ⊢ (G∗ ≡∼
P ∗ (pXq)). Therefore (9) S ∗ ⊢ (∼ G ⊃∼ G∗ ). Hence (10) S ∗ ⊢∼ G∗ . N
Remark. We know that P A 0∼ GP A . Theorem 110 shows is that this fact can-
not be proved without appealing to a property of PA stronger than it’s consistency.
Proposition 113 Theorem 111 is stronger than Theorem 83, and is best possible.
that S 0 G requires the purely formal hypothesis that S is consistent. The consis-
tency of S implies the truth of G, as noted in Theorem 85. This result holds for
every system for which a Gödel sentence can be constructed. Consequently, whether
or not S 0∼ G is equivalent to whether or not S is sound to the extent that it does
not prove this false sentence. We have seen that the first half of the First Incom-
pleteness Theorem establishes the consistency and unsoundness of S ∪ {∼ G}, if S
is consistent, and we have also seen that S ∪ {∼ GS } ⊢∼ GS∪{∼GS } . Given that S
is consistent, ConS is true. Exactly the amount of soundness of S required for the
second half of the First Incompleteness Theorem is that S does not prove the false
Σ1 -sentence ConS, i.e. S ∪ {ConS } is consistent. We have seen that a theory can
have this much soundness and still not be Σ1 -sound (Theorem 112).
Lecture 11
Provable Σ1-completeness
91
LECTURE 11 92
Remark about the strategy for the proof. The definition of what it is to be a
Σ1 -formula, Definition 41, is explicit (rather than recursive), i.e. a Σ1 formula is
any formula of the form ∃vi F where F is a Σ0 -formula. —— So as in our proof
that R and thereby Q and PA are Σ1 -complete (Propositions 58, 67, and 69, and
Theorem 70), the proof of provable Σ1 -completeness has to go via a proof of provable
Σ0 -completeness.
The definition of Σ0 -formula, Definition 40, is recursive, so the proof of prov-
able Σ0 -completeness must proceed by induction over the recursive definition of
Σ0 -formulas. The sequence of formulas by which a Σ0 -sentence is generated by this
recursion will in general contain free variables. Thus to have a strong enough induc-
tion hypothesis for the inductive argument, we must prove provable Σ0 -completeness
for formulas that may contain free variables.
How to formulate this result for Σ0 -formulas with free variables requires some
thought. It is not that P A ⊢ (X ⊃ P r(pXq)) for all Σ1 -formulas, including ones
with free variables. For example,
variable v1 , i.e. s(x, y) = p∀v1 (v1 = y ⊃ Ex )q, to allow substitution on any specified
variable, i.e. s(x, y, z) = p∀vz (vz = y ⊃ Ex )q
Proposition 117 For the function s(x, y, z) = p∀vz (vz = y ⊃ Ex )q there is a
Σ1 -formula S(x, y, z, w) in the language of PA such that for all natural numbers
n1 , n2 , n3 , n4 , S(n1 , n2 , n3 , n4 ) is true if and only if s(n1 , n2 , n3 ) = n4 .
Proof.
p∀vz (vz = y ⊃ Ex )q =
. . 5} η ∗ 13y ∗ 8x3.
. . 5} 26 |5 .{z
p∀v′ . . .′ (v′ . . .′ = y ⊃ Ex )q = 96 5| .{z
|{z} |{z} z z
z z
The function f (z) = 5| .{z
. . 5} is generated by the following primitive recursion:
z
f (0) = 5
f (n + 1) = 5 · 13n+1 + f (n), so by generalization of Theorem 39, f (x) = y is Σ1 ,
so the relation s(x, y, z) = z is Σ1 . N
Definition 70 (arithmetized proof predicate with free variables) For P r(v1 )
a formula in the language of PA that expresses {n : P A ⊢ En } and F (vk1 , . . . , vkm )
any formula in the language of PA with exactly the free variables shown, and k =
max{k1 , . . . , km }, P r[pF (vk1 , . . . , vkm )q](vk1 , . . . , vkm ) =df
∃vk+1 . . . ∃vk+m ((S(pF (vk1 , . . . , vkm )q, vk1 , k1 , vk+1 ) ∧ S(vk+1 , vk2 , k2 , vk+2 ) ∧ . . .
∧ S(vk+m−1 , vkm , km , vk+m )) ∧ P r(vk+m )).
Note that F (vk1 , . . . , vkm ) and P r[pF (vk1 , . . . , vkm )q](vk1 , . . . , vkm ) have the same
free variables.
Proof. For F with no free variables, consider P r[pF q] with one vacuous quan-
tifier, e.g. ∀v1 (S(pF q, v1 , 0′ , v2 ) ⊃ P r(v2 )). etc. N
Proposition 119 For any natural numbers a1 , . . . am , P r[pF (vk1 , . . . , vkm )q](a1 , . . . , am )
is true if and only if P r(pF (a1 , . . . , am )q) is true.
Proof. By the definition of P r[pF q], and provable equivalence of F (a) and F [a].
N
Theorem 120 (P1∗ = P1 generalized to allow free variables) For any formula
F (vk1 , . . . , vkm ) in the language of PA, if PA ⊢ F (vk1 , . . . , vkm ), then
PA ⊢ P r[pF (vk1 , . . . , vkm )q](vk1 , . . . , vkm ).
Proof. The proof is by induction on the number of free variables in F (vk1 , . . . , vkm ).
It consists of arithmetizing the following argument, which for simplicity I formu-
late for the case in which F has just one free variable.
If PA⊢ F (vk ), then by logic in PA, PA⊢ ∀vk F (vk ). Then by P1 , PA ⊢ P r(p∀vk F (vk )q).
For each natural number n, (∀vk F (vk ) ⊃ F [n]) is logically valid, so PA⊢ (∀vk F (vk ) ⊃
F [n]). Then by P1 , PA ⊢ P r(p(∀vk F (vk ) ⊃ F [n])q). Then by P2 , PA ⊢ (P r(p∀vk F (vk )q) ⊃
P r(pF [n]q)). Then by Modus pones, PA ⊢ P r(pF [n]q). We need to formalize this
argument in PA. N
Theorem 121 (P2∗ = P2 generalized to allow free variables) For any formu-
las F (vk1 , . . . , vkm ) and G(vr1 , . . . , vrs ) in the language of PA,
PA⊢ (P [(F ⊃ G)](vk1 , . . . , vkm , vr1 , . . . vrs ) ⊃ (P [F ](vk1 , . . . , vkm ) ⊃ P [G](vr1 , . . . vrs ))).
Proof. Exercise. N
Lemma 122 For all formulas F (vj1 , . . . , vjm ) and G(vk1 , . . . , vkn ), P A ⊢ (P [pF q](vj1 , . . . , vjm ) ⊃
(P [pGq](vk1 , . . . , vkn ) ⊃ P [p(F ∧ G)q])(vj1 , . . . , vjm , vk1 , . . . , vkn ))
Proof Exercise. N
LECTURE 11 95
Definition 72 For F (vi ) a formula with free variable vi and t any term, F (vi /t)
is the result of substituting the term t for all occurrences of the variable vi in the
formula F (vi ).
Lemma 123 Let F (vi ) be a formula with free-variable vi . Let vj be a variable and
t a term both free for vi in F (vi ). Then the following equivalence is logically valid:
Proof. From the definition of P r[pF (vi )](vi ), substitutivity of =, and logical
equivalence of F (vi ) and F [vi ]. N
Lemma 124 Let F (vi ) be a formula with free-variable vi and let t(vj ) be a term
free for vi in F (vi with variable vj distinct from vi . Then the following equivalence
is logically valid:
Proof. From the definition of P r[pF (vi )](vi ), substitutivity of =, and logical
equivalence of F (vi ) and F [vi ]. N
Our proof in Lecture 7 that PA is Σ0 -complete went by way of proving the very
strong result that the extremely weak system R is Σo -complete and then showing
that R is a subsystem of PA. Proving Σ0 -completeness requires heavy use of math-
ematical induction, so that proof cannot be formalized in R or Q. It could be
formalized in PA, but that would be a very roundabout way of proving the provable
Σ0 -completeness of PA in PA. Rather what we want to do is formalize in PA a di-
rect proof of the Σ0 -completeness of PA. I will illustrate an informal direct proof of
the Σ0 -completeness of PA that we will be formalizing in PA, by going through an
informal proof for the case of an atomic Σ0 -formula of the form v1 + v2 = v3 .
Induction step:
We have to show, within PA, that there exists a derivation of
(∀v3 (v1 +v2 = v3 ⊃ P r[v 1 + v2 = v3 q](v1 , v2 , v3 )) ⊃ ∀v3 (v1 +v2′ = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2′ , v3 ))).
(11) P1∗
(13) ((P r[pv1 + v2 = v4 q](v1 , v2 , v4 ) ∧ P r[pv1 + v2′ = (v1 + v2 )′ q](v1 , v2 )) ⊃
P r[pv1 + v2′ = v4′ q](v1 , v2 , v4 )) (12) P2∗
(14) (P r[pv1 + v2 = v4 q](v1 , v2 , v4 ) ≡ P r[pv1 + v2 = v3 q](v1 , v2 , v4 ))
Lemma 123
(15) (P r[pv1 + v2 = v4 q](v1 , v2 , v4 ) ≡ P r[pv1 + v2 = v3 q](v1 , v2′ , v4′ ))
′ ′
Lemma 124
′
(16) ((P r[pv1 + v2 = v3 q](v1 , v2 , v4 ) ∧ P r[pv1 + v2 = (v1 + v2 )′ q](v1 , v2 )) ⊃
P r[pv1 + v2 = v3 q](v1 , v2′ , v4′ )) (13)(14)(15) propositional logic
′
(17) P r[pv1 + v2 = (v1 + v2 )′ q](v1 , v2 ) (3) P1∗
(1)(2)(6) (18) P r[pv1 + v2 = v3 q](v1 , v2′ , v4′ ) (16)(10)(17) ⊃-elim
′
(1)(2)(6) (19) P r[pv1 + v2 = v3 q](v1 , v2 , v3 ) (18)(6) subst =
′
(1)(2) (20) P r[pv1 + v2 = v3 q](v1 , v2 , v3 ) (6)(19) ∃-elimination
(1) (21) (v1 + v2′ = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2′ , v3 )) (2)(20) ⊃-intro
LECTURE 11 98
(1) (22) ∀v3 (v1 + v2′ = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2′ , v3 )) (21) ∀-intro (v3 not free in (1))
(23) (∀v3 (v1 + v2 = v3 ⊃ P r[v 1 + v2 = v3 q](v1 , v2 , v3 )) ⊃
∀v3 (v1 + v2′ = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2′ , v3 ))) (1)(22) ⊃-intro
(iii) F is (∀v1 ≤ v2 )G(v1 ), for G(v1 ) a Σ0 -formula. This means that v2 is free in
F . To simplify notation we shall take it that no other variables are free in F , i.e. v1
is the only free variable in G(v1 ). We need to give a proof in PA of (F ⊃ P r[F ]),
i.e.
((∀v1 ≤ v2 )G(v1 ) ⊃ ∀v3 (S(p(∀v1 ≤ v2 )G(v1 )q, v2 , 0′′ , v3 ) ⊃ P (v3 )).
The proof is by induction on the variable v2 occurring free in the formula (F ⊃
P r[F ]).
Base case: We need to prove that ((∀v1 ≤ 0)G(v1 ) ⊃ ∀v3 (S(p(∀v1 ≤ 0)G(v1 )q, 0, 0′′ , v3 ) ⊃
P (v3 ))
To simplify notation I shall write v1 as x and v2 as y.
Step (9) in the above derivation calls for comment. In deriving (9) from (8)
∀-Intro is applied to (8) and the status of (8), as an induction hypothesis, is that of
assumption. If the variable in (8) were the variable of induction this ∀-Intro would
be illegitimate. However, the induction of which (8) is an induction hypothesis is
over formulas, so the assumption is about the formula G(x), and not about x, i.e.
x is a free variable in (8) to which ∀-Intro may be applied. N
∀yF (y)
101
LECTURE 12 102
Definition 74 PAω ⊢ X if and only if there is a tree of finite height with formulas
in the language of PA at each node and with X at the bottom node, such that for
each formula at a node without predecessor is a theorem of PA, and such that each
node is either a theorem of PA or there is one node directly above the node at which
X occurs such that for Y the formula at the predecessor node, PA ⊢ (Y ⊃ X), or
X is the result of an application of the ω-rule. By a tree consisting of a single node
(height 0), if PA ⊢ X, then PAω ⊢ X.
For G the Gödel sentence for S, PAω ⊢ G. A derivation using the ω-rule has
infinitely many premisses and hence is an infinite object, unlike a formal proof, or
our usual idea of a informal proof, so derivation of G by the ω-rule cannot be said
to constitute a proof of G. Indeed derivability from the axioms of PA by the ω-rule
is tantamount to truth.
The completeness of PAω with respect to truth is entirely to do with the ω-rule,
and essentially nothing to do with the axioms for arithmetic of PA, as shown by the
fact that the corresponding system for R, i.e. Rω =df R + ω-rule, is also complete
with respect to truth.
Proof. The only facts about PA used in the proof of Proposition 128 are that
PA is Σ0 -complete and that PA ⊢ (F (m) ⊃ ∃vi F (vi )). Both these properties are
also facts about R. N
There are constructive versions of the ω-rule, based on the fact that we can
state in a single sentence that all numerical instances of a given formula F (v1 ) are
provable, and by the arithmetization of syntax such single sentences can be expressed
LECTURE 12 103
in the language of arithmetic, namely as ∀v1 P r[pF (v1 )q](v1 ), where P r[pF (v1 )q](v1 )
is defined by Definition 70 in Lecture 11. We can then give finite expression to an
ω-rule by the sentence:
Definition 75 For F (v1 ) a formula in the language of PA with one free variable, a
Uniform Reflection Principle is any sentence of the form
(∀v1 P r[pF (v1 )q](v1 ) ⊃ ∀v1 F (v1 )).
Proof. The following derivation shows the existence of a formal proof in PA.
Theorem 133 For F (v1 ) any Π1 -formula with one free variable,
PA ∪{ConP A } ⊢ (∀v1 P r[pF (v1 )q](v1 ) ⊃ ∀v1 F (v1 )).
Proof.
(1) (16) ∀v1 F (v1 ) (15) ∀-intro (v1 not free in (1)
(17) (∀v1 P r[pF (v1 )q](v1 ) ⊃ ∀vi F (v1 ) (1) (16) ⊃-intro
This proof covers the case where F (v1 ) is Σ0 rather than Π0 , in which case the
justification at line (3) is Theorem 126 rather than Corollary 127. N
Definition 76 For theories S1 and S2 formulated in the same language (or such that
the language of S2 is an extension of the language of S1 or such that the language
of S1 can interpreted in the language of S2 , but we shall not be concerned with these
more general cases), S2 is an extension of S1 if for each formula X in the language
of S1 , if S1 ⊢ X, then S2 ⊢ X (or S2 ⊢ X ∗ where X ∗ is the translation of X into
the language of S2 ).
example, that we had found, for Fermat’s great theorem, a proof in which the [in-
finitary] logical function ǫ was used. We could then make a finitary proof out of it
in the following way.”
Leaving aside the question, in what minimal system can the consistency of PA
be proved, which is beyond the scope of this course (the answer is, very roughly,
constructive principles of abstract mathematics, rather than finitary principles of
concrete mathematics), a precise working out of the argument Hilbert sketched
requires that the proof of Theorem 133 be carried out in finitary mathematics.
Hilbert never formulated clearly what he meant by finitary mathematics, i.e. he
never gave a formal system of finitary mathematics, and I won’t enter here the
debate over what formal system should be taken to capture the intended notion of
finitary arithmetic. Rather, I will address the question, in how weak a subsystem of
PA can the argument for Theorem 133 be carried out?
The key point is that such a system must be strong enough to prove provable Σ0 -
completeness of PA. Hilbert did not explicitly formulate provable Σ0 -completeness,
but it is implicit in his argument, and implicitly he takes it to be a fact of finitary
mathematics. “Let us assume that numerals p, a, b, c (p > 2) satisfying Fermat’s
equation ap +bp = cp are given; then we could also obtain this equation as a provable
formula by giving the form of a proof to the procedure by which we ascertain that
the numbers ap + bp and cp coincide.”
Proving provable Σ0 -completeness requires mathematical induction, so we may
take the question to be, how much induction, measured by complexity in the arith-
metical hierarchy of formulas to which induction must be applied, is needed for this
proof?
The minimum is Σ1 -induction, i.e. axioms N12 for Σ1 -formulas, as in the step of
the proof in Lecture 11 in which we proved that
((∀v1 ≤ v2 )G ⊃ ∀v3 (S(p(∀v1 ≤ v2 )Gq, v2 , 0′′ , v3 ) ⊃ P (v3 )) by induction on the free
variable v2 . However, we also used Π2 -induction, in our proof in PA that (v1 +
v2 = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2 , v3 )), since we proved this by induction on the
formula ∀v3 (v1 + v2 = v3 ⊃ P r[pv1 + v2 = v3 q](v1 , v2 , v3 )). It might be that there
is some clever way to reconstruct that proof so that the universal quantification of
the induction formula is not needed. In any case, from what has been established,
we have the following theorem corresponding to Hilbert’s argument claiming that
infinitary mathematics is conservative over finitary mathematics with respect to
Π1 -theorems.
107
LECTURE 13 108
The axioms and inference rules of GL arise by abstraction from the arithmetized
proof predicate for PA. Conversely, the axioms and inference rules, and hence all
theorems of GL, translate into theorems of PA. This result means that GL is sound
with respect to interpretation in PA.
There is also a completeness theorem, due to Robert Solovay, for GL with respect
to interpretation in PA.
LECTURE 13 110
Theorem 136 (Solovay completeness theorem for GL) For X any sentence
in the language of GL, if for every interpretation ∗ (as in Definition 82), P A ⊢ X ∗ ,
then GL ⊢ X; or equivalently, for X any sentence in the language of GL, if GL 0 X,
then there is an interpretation ∗ such that P A 0 X ∗ .
Remarks. In proofs that depend on this theorem I will say “by propositional
logic in GL” or just “by propositional logic”, rather than citing Theorem 139. By
the compactness theorem for truth functional logic, the above result holds for Y a
truth functional consequence of any set of provable formulas, not just finite sets of
formulas, but we have not need for this generalization.
Proof. (i) Both ((X ∧ Y ) ⊃ X) and ((X ∧ Y ) ⊃ Y ) are tautologies and hence
axioms of GL. By Lemma 138, GL ⊢ (2(X ∧Y ) ⊃ 2X) and GL ⊢ (2(X ∧Y ) ⊃ 2Y ).
Then by propositional logic in GL, GL ⊢ (2(X ∧ Y ) ⊃ (2X ∧ 2Y ))
(ii) Since (X ⊃ (Y ⊃ (X ∧ Y ))) is a tautology, GL ⊢ (X ⊃ (Y ⊃ (X ∧ Y ))).
Hence by Lemma 138, GL ⊢ (2X ⊃ 2(Y ⊃ (X ∧ Y ))). By A2, GL ⊢ (2(Y ⊃
(X ∧ Y ) ⊃ (2Y ⊃ 2(X ∧ Y )))). Then by propositional logic in GL, GL ⊢ (2X ⊃
(2Y ⊃ 2(X ∧ Y ))). The formulas ((2X ∧ 2Y ) ⊃ 2X) and ((2X ∧ 2Y ) ⊃ 2Y )
are tautologies and hence axioms of GL, so by propositional logic in GL, GL ⊢
((2X ∧ 2Y ) ⊃ 2(X ∧ Y )). N
Theorem 145 (substitution on more than one sentence letter) For F any
formula with sentence letters pk1 , . . . , pkn , and for pairs of formulas Ai , Bi , i =
1, . . . , n, such that GL ⊢ (Ai ≡ Bi ), GL ⊢ (F (A1 , . . . , An ) ≡ F (B1 , . . . , Bn )), where
F (A1 , . . . , An ) is the result of substituting Ai for pki in F , and F (B1 , . . . , Bn ) is the
result of substituting Bi for pki in F .
LECTURE 13 113
Proof. Exactly the same proof structure as for Theorem 144, with just reformu-
lation of the induction hypothesis so that it’s for multiple substitutions, establishes
this result. N
Theorem 144 immediately establishes that GL is closed under substitution of
provable equivalents.
Proof. The proof is by induction over the recursion that generates the formula
F.
If F = p, what is to be proved is GL ⊢ (2(A ≡ B) ⊃ 2(A ≡ B)), which is a
tautology and hence provable in GL.
If F = q for p 6= q, what is to be proved is GL ⊢ (2(A ≡ B) ⊃ 2(q ≡ q)). Since
(q ≡ q) is a tautology, GL ⊢ (q ≡ q), so by R2 , GL ⊢ 2(q ≡ q). The result follows
by propositional logic in GL.
LECTURE 13 114
Lemma 150 For each formula X in the language of GL, the formulas 2X, 2 ⊡ X,
and ⊡2X are provably equivalent in GL.
LECTURE 13 115
Proof. The proof is by induction over the recursion that generates the formula
F.
If F = p, what is to be proved is GL ⊢ (⊡(A ≡ B) ⊃ (A ≡ B)), which holds by
Definition 84 and propositional logic in GL.
If F = q for p 6= q, what is to be proved is GL ⊢ (⊡(A ≡ B) ⊃ (q ≡ q)). For any
formula H, (H ⊃ (q ⊃ q)) is a tautology and hence an axiom of GL, so in particular
GL ⊢ (⊡(A ≡ B) ⊃ (q ≡ q)).
If F =⊥, what is to be proved is GL ⊢ (⊡(A ≡ B) ⊃ (⊥≡⊥)). The argument is
as for the preceding case.
If F = (G ⊃ H), we have as induction hypotheses, GL ⊢ (⊡(A ≡ B) ⊃ (Gp (A) ≡
Gp (B))), and GL ⊢ (⊡(A ≡ B) ⊃ (Hp (A) ≡ Hp (B))). By Definition 83(4), what
is to be proved is GL ⊢ (⊡(A ≡ B) ⊃ ((Gp (A) ⊃ Hp (A)) ≡ (Gp (B) ⊃ Hp (B)))),
which follows from the induction hypotheses by propositional logic in GL.
LECTURE 13 116
Corollary 155 (variant proof of Theorem 148) For all formulas A, B, and F
and propositional variable p, GL ⊢ (2(A ≡ B) ⊃ 2(Fp (A) ≡ Fp (B))).
Theorem 156 (Theorem 154 generalized to multiple substitutions) Let F (pi1 , . . . , pim )
be a formula with sentence letters pi1 , . . . , pim . Then
GL ⊢ (⊡(A1 ≡ B1 ) ∧ . . . ∧ ⊡(Am ≡ Bm )) ⊃ (F (A1 , . . . , Am ) ≡ F (B1 , . . . , Bm ))).
Proof. Exercise. N
Lecture 14
Examples of sentences in which the sentence letter p is modalized: (1) 2p, (2)
∼ 2p, (3) 2 ∼ p, (4) ∼ 2 ∼ p, (5) (2p ⊃ q) (6) 2(2p ⊃ p), (7) (2(2p ⊃ p) ⊃ 2p),
(8) 2(p ≡ (2p ⊃ q)), (9) (2(2p ⊃ q)∧ ∼ 2p), (10) ⊥, (11) (⊥⊃⊥), (12) 2 ⊥, (13)
q (where q 6= p), (14) 2q (in these latter cases p is modalized in the sentence since p
does not occur in the sentence and so every occurrence of p in the sentence is within
the scope of 2).
Examples of sentences not modalized in p: (1) p, (2) (p ⊃⊥), (3) (2p ⊃ p), (4)
(p ≡ (2p ⊃ q).
117
LECTURE 14 118
and sentences 2C1 (p), . . . , 2Cn (p) such that X(p) = D(2C1 (p), . . . , 2Cn (p)) (i.e.
the result of substituting each sentence 2Ci (p) for all occurrences of the variable pi
in the sentence D(pk1 , . . . , pkn )) is called a decomposition of X with respect to p;
D(pk1 , . . . , pkn ) is called a decomposition sentence for X, and 2C1 (p), . . . , 2Cn (p)
are called components of X.
We shall see that every sentence modalized in pi has a fixed point with respect
to pi . We establish this result first for the simplest case of a sentence modalized in
pi , i.e. of the form 2Y (pi ).
Lemma 160 For pi any propositional variable and Y (pi ) any sentence in the lan-
guage of GL in which pi occurs, the result of substituting (⊥⊃⊥) for each occurrence
of pi in 2Y (pi ) is a fixed point for 2Y (pi ).
Proof. In this proof I abbreviate (⊥⊃⊥) as ⊤, and write 2Y (⊤) for the result
of substituting (⊥⊃⊥) for each occurrence of pi in 2Y (pi ). The following derivation
is a proof in GL.
Before proving the fixed point theorem itself we need another lemma and for
that lemma we need a definition:
is solvable (in GL) iff there are sentences F1 , . . . , Fm not containing pk1 , . . . , pkm
such that for 1 ≤ i ≤ m,
is solvable.
Now by Lemma 160 applied to the sentence that results from 2Cm+1 (pk1 , . . . , pkm , pkm+1 )
by substituting Fi (pkm+1 ) for pki for each i such that 1 ≤ i ≤ m, i.e.
2Cm+1 (F1 (pkm+1 ), . . . , Fm (pkm+1 ), pkm+1 ), there is a sentence Fm+1 such that
(2) GL ⊢ Fm+1 ≡ 2Cm+1 (F1 (Fm+1 ), . . . , Fm (Fm+1 ), Fm+1 )).
By substitution of Fm+1 into the provable equivalences (1), the sentences
F1 (Fm+1 ), . . . , Fm (Fm+1 ), Fm+1
solve the set of equivalences. N
Theorem 162 (Fixed Point Theorem) For every sentence X in the language of
GL modalized in pr , there is a sentence F containing only sentence letters that occur
in X and not containing pr such that GL ⊢ (F ≡ Xpr (F )).
[1] George Boolos, The Logic of Provability, Cambridge University Press, 1993.
[3] David Hilbert, “On the infinite” (1926); English translation in Jean van Hei-
jenoort (ed), (1927), p. 471.
[4] David Hilbert, “Die Grundlagen der Mathematik”, Abhandlungen aus dem
mathematischen Seminar der Hamburgeshcen Universität 6 (1928), English
translation by Stephan Bauer-Mengelberg, ‘The foundations of mathematics”
in Jean van Heijenoort (ed.) From Frege to Gödel; A Source Book in Mathe-
matical Logic 1879-1931, Harvard University Press, 1967.
[6] Craig Smorynski, “The incompleteness theorems”, Jon Barwise (ed.), Handbook
of Mathematical Logic, Horth-Holland Publishing Company, 1977, pp. 821-865.
124