Introductory Material

1. Logical Metatheory We also need variables in the metalanguage

where the values of the variables are object-langu-
Definition: Logical metatheory is the use of log- age expressions.
ical and mathematical methods to study logic and
Definition: A schematic letter is a metalinguis-
logical systems themselves.
tic variable for object-language expressions, which,
when placed inside a representation of an object lan-
In logical metatheory, rather using a logical system
guage expression, is used as part of a device to men-
to construct a proof about another subject matter
tion all object-language expressions formed by sub-
(or an unknown one), we prove things about logic
stituting its possible value in the position in which it
itself. This might include proving things about
appears.
proving things.
For example, in this course, we will prove that, Whether you know it or not, you’ve seen and used
in first-order predicate logic, we can construct a these already. In your Intro to Logic book’s sheet
derivation or proof for every valid argument, and of deduction rules, modus ponens looked like:
also that every argument we can construct a proof A →B 𝜑→𝜓
for is valid. These results are called, respectively, A 𝜑
completeness and soundness. B or 𝜓
Definition: The object language is the language The cursive or Greek letters are not in the actual
being studied, or the language under discussion. object language; they are metalinguistic schematic
letters, and this rule schema tells you that modus
Definition: The metalanguage is the language ponens covers not just the case Fa → Ga, Fa∴Ga,
used when studying the object language. but any inference of the right form.
The Open Logic Text uses Greek letters for
In this course, the symbolic languages you learned schematic letters whose values are object-language
in previous logic courses are our object languages. formulas; we will be following this convention, and
Our metalanguage is plain English, or actually, adding other kinds of schematic letters and meta-
plain English supplemented by some special sym- language variables as we go.
bols and devices. It is important, however, not to mix up object-
If the metalanguage, we’ll use words like “if”, language expressions and metalanguage ones. We
“and”, “or”, etc., and treat their logic as equivalent cannot write “ 𝜑. . . ”, because “ ” is an object lan-
A A
to the classical rules for “→”, “∧”, “∨”, etc. We as- guage quantifier, and 𝜑 a metalanguage variable.
sume “if . . . then . . . ” obeys modus ponens and modus Instead, we might write “For every formula 𝜑, . . . ”.
tollens, for example. Something like “𝜑 → 𝜓” is a hybrid, belonging
When working in the metalanguage, we will not in neither language cleanly. Some use “Quine cor-
use symbols like “→”, “∧”, “∨”, but will mention ners”, e.g., ⌜𝜑 → 𝜓⌝ to mention not the hybrid, but
them; typically with quotation marks (though often the resulting expression after assigning values to
we get lazy and omit them). The main purpose of 𝜑 and 𝜓 and inserting them appropriately, though
the metalanguage is to discuss the object language, we will use this notation a different way. Hopefully
so we will do this quite a lot. context will make it clear what is meant.

1
2. Set Theory Definition: If 𝑋 and 𝑌 are sets, the union of 𝑋
and 𝑌 , written “𝑋 ∪ 𝑌 ”, is the set that contains ev-
With a very brief exception later in the semester, erything that is a member of either 𝑋 or 𝑌 .
we will not be studying the logic of set theory as an
object language. (Take Philosophy 514, Mathemat- Definition: The intersection of 𝑋 and 𝑌 , written
ical Logic II for that!) We will be using set theory “𝑋 ∩ 𝑌 ”, is the set that contains everything that is a
in the metalanguage, however. Consider all the member of both 𝑋 and 𝑌 .
mathematical notation below to be a supplement Definition: The relative complement or differ-
to English. ence of 𝑋 and 𝑌 , written “𝑋 ⧵ 𝑌 ”, is the set con-
taining all members of 𝑋 that are not members of
Sets 𝑌.
Definition: A set is a collection of entities for which Definition: The empty set or null set, written “ ∅”,
it is determined, for every entity of a given type, that “Λ” or “ { }”, is the set with no members.
the entity either is or is not included in the set.
Definition: If 𝑋 and 𝑌 are sets, then they are dis-
Definition: An urelement is a thing that is not a joint iff they have no members in common, i.e., iff
set. 𝑋 ∩ 𝑌 = ∅.

Definition: An entity 𝐴 is a member of set 𝑋 iff it Ordered n-tuples and relations

is included in that set.
Definition: An ordered n-tuple, written
We write this as: “𝐴 ∈ 𝑋 ”. We write “𝐴 ∉ 𝑋 ” to “⟨𝐴1 , … , 𝐴𝑛 ⟩”, is something somewhat like a set,
mean that 𝐴 is not a member of 𝑋 . except that the elements are given a fixed order, so
Sets are extensional, or determined entirely that ⟨𝐴1 , … , 𝐴𝑛 ⟩ = ⟨𝐵1 , … , 𝐵𝑛 ⟩ iff 𝐴𝑖 = 𝐵𝑖 for all 𝑖
by their members: for sets 𝑋 and 𝑌 , 𝑋 = 𝑌 iff for such that 1 ≤ 𝑖 ≤ 𝑛.
all 𝐴, 𝐴 ∈ 𝑋 iff 𝐴 ∈ 𝑌 .
An ordered 2-tuple, e.g., ⟨𝐴, 𝐵⟩ is also called an
Definition: A singleton or unit set is a set con- ordered pair. An entity is identified with its 1-
taining exactly one member. tuple.
Definition: If 𝑋 and 𝑌 are sets, then the Carte-
“{𝐴}” means the set containing 𝐴 alone. Generally, sian product of 𝑋 and 𝑌 , written “𝑋 × 𝑌 ”, is the
“{𝐴1 , … , 𝐴𝑛 }” means the set containing 𝐴1 , … , 𝐴𝑛 , set of all ordered pairs ⟨𝐴, 𝐵⟩ such that 𝐴 ∈ 𝑋 and
but nothing else. 𝐵 ∈ 𝑌.
The members of sets are not ordered, so from
{𝐴, 𝐵} = {𝐶, 𝐷} one cannot infer that 𝐴 = 𝐶, only Generally, “𝑋 𝑛 ” is used to represent the set of all
that either 𝐴 = 𝐶 or 𝐴 = 𝐷. ordered 𝑛-tuples consisting entirely of members of
Sometimes {𝐴 ∶ … 𝐴 … } is used to name the set 𝑋 . Notice that 𝑋 2 = 𝑋 × 𝑋 .
of all objects 𝐴 such that … 𝐴 … is true of them, but The following definition is philosophically prob-
this notation is potentially dangerous, for reasons lematic, but a common way of speaking in mathe-
we will see later. matics.

Definition: If 𝑋 and 𝑌 are sets, 𝑋 is said to be a
subset of 𝑌 , written “𝑋 ⊆ 𝑌 ”, iff all members of
𝑋 are members of 𝑌 ; and 𝑋 is said to be a proper
subset of 𝑌 , written “𝑋 ⊊ 𝑌 ” or “𝑋 ⊂ 𝑌 ”, iff all
members of 𝑋 are members of 𝑌 , but not all members
of 𝑌 are members of 𝑋 .
Definition: An n-place relation (in extension)
on set 𝑋 is any subset of 𝑋 𝑛 .
A 2-place relation is also called a binary relation.
Binary relations are taken to be of sets of ordered
pairs. A 1-place relation is also called (the extension
of) a property.

2
Definition: If 𝑅 is a binary relation, then the do- Cardinal numbers
main of 𝑅 is the set of all 𝐴 for which there is an 𝐵
Definition: If 𝑋 and 𝑌 are sets, then they are
such that ⟨𝐴, 𝐵⟩ ∈ 𝑅.
equinumerous, written “ 𝑋 ≈ 𝑌 ”, iff there is a one-
Definition: If 𝑅 is a binary relation, the range of one function whose domain is 𝑋 and whose range is
𝑅 is the set of all 𝐵 for which there is an 𝐴 such that 𝑌 .
⟨𝐴, 𝐵⟩ ∈ 𝑅.
Definition: Sets 𝑋 and 𝑌 have the same cardi-
Definition: The field of 𝑅 is the union of the do-
nality or cardinal number if and only if they are
main and range of 𝑅.
equinumerous.
Definition: If 𝑅 is a binary relation, 𝑅 is reflexive
iff ⟨𝐴, 𝐴⟩ ∈ 𝑅 for all 𝐴 in the field of 𝑅. Definition: If 𝑋 and 𝑌 are sets, then the cardinal
number of 𝑋 is said to be smaller than the cardinal
Definition: If 𝑅 is a binary relation, 𝑅 is symmet- number of 𝑌 iff there is a set 𝑍 such that 𝑍 ⊆ 𝑌 and
ric iff for all 𝐴 and 𝐵 in the field of 𝑅, ⟨𝐴, 𝐵⟩ ∈ 𝑅 𝑋 ≈ 𝑍 but there is no set 𝑊 such that 𝑊 ⊆ 𝑋 and
only if ⟨𝐵, 𝐴⟩ ∈ 𝑅. 𝑊 ≈ 𝑌.
Definition: If 𝑅 is a binary relation, 𝑅 is tran-
sitive iff for all 𝐴, 𝐵 and 𝐶 in the field of 𝑅, if Definition: If 𝑋 is a set, then 𝑋 is denumerable
⟨𝐴, 𝐵⟩ ∈ 𝑅 and ⟨𝐵, 𝐶⟩ ∈ 𝑅 then ⟨𝐴, 𝐶⟩ ∈ 𝑅. iff 𝑋 is equinumerous with the set of natural numbers
{0, 1, 2, 3, 4, … , (and so on ad inf.)}.
Definition: A binary relation 𝑅 is an equivalence
relation iff 𝑅 is symmetric, transitive and reflexive. Definition: Aleph null/naught/zero, written
Definition: If 𝑅 is an equivalence relation then, the “ ℵ0 ”, is the cardinal number of any denumerable set.
R-equivalence class on A, written “ [𝐴]𝑅 ”, is the
Definition: If 𝑋 is a set, then 𝑋 is finite iff either
set of all 𝐵 such that ⟨𝐴, 𝐵⟩ ∈ 𝑅.
𝑋 = ∅ or there is some positive integer 𝑛 such that
We sometimes write “𝑅(𝐴, 𝐵)” to mean ⟨𝐴, 𝐵⟩ ∈ 𝑅, 𝑋 is equinumerous with the set {1, … , 𝑛}.
though this can be dangerously object-languagey.
Definition: A set is infinite iff it is not finite.

Functions Definition: A set is countable iff it is either finite

Definition: A function (in extension) is a binary or denumerable.
relation which, for all 𝐴, 𝐵 and 𝐶, if it includes ⟨𝐴, 𝐵⟩
then it does not also contain ⟨𝐴, 𝐶⟩ unless 𝐵 = 𝐶. Homework 1
So if 𝐹 is a function and 𝐴 is in its domain, then Assuming that 𝑋 , 𝑌 and 𝑍 are sets, 𝑅 is a relation,
there is a unique 𝐵 such that ⟨𝐴, 𝐵⟩ ∈ 𝐹 ; this unique and 𝐴 and 𝐵 are any entities, informally verify the
𝐵 is denoted by “𝐹 (𝐴)”. following:
Definition: An n-place function is a func- (1) if 𝑋 ⊆ 𝑌 and 𝑌 ⊆ 𝑍 then 𝑋 ⊆ 𝑍
tion whose domain consists of 𝑛-tuples. Then (2) 𝑌 1∩ ∅ = ∅ and 𝑌 ∪ ∅ = 𝑌
“𝐹 (𝐴1 , … , 𝐴𝑛 )” abbreviates “𝐹 (⟨𝐴1 , … , 𝐴𝑛 ⟩)”. (3) 𝑌 = 𝑌
(4) If 𝑅 is an equivalence relation and 𝐴 and 𝐵 are
Definition: An n-place operation on 𝑋 is a func- in its field, then ([𝐴]𝑅 = [𝐵]𝑅 iff ⟨𝐴, 𝐵⟩ ∈ 𝑅) and
tion whose domain is 𝑋 𝑛 and whose range is a subset (if [𝐴]𝑅 ≠ [𝐵]𝑅 then [𝐴]𝑅 and [𝐵]𝑅 are disjoint).
of 𝑋 . (5) 𝑌 ≈ 𝑌
Definition: If 𝐹 is a function, then 𝐹 is one-one (6) The set of even non-negative integers is denu-
iff for all 𝐴 and 𝐵 in the domain of 𝐹 , 𝐹 (𝐴) = 𝐹 (𝐵) merable.
only if 𝐴 = 𝐵.

3
3. Mathematical Induction constructed out of simpler wffs provided that those
simpler wffs have Φ, then whenever a given natural
Definition: The principle of mathematical in- number 𝑛 has Φ′ then 𝑛 + 1 also has Φ′ . By mathe-
duction states the following: matical induction, all natural numbers have Φ′ , i.e.,
If ( Φ is true of 0), then if (for all natural numbers 𝑛, no matter how many operators a wff contains, it
if Φ is true of 𝑛, then Φ is true of 𝑛 + 1), then Φ is has Φ.
true of all natural numbers. This is invoked by proving:

To use the principle mathematical induction to es- Base step. Φ is true of the simplest well-formed
tablish that something is true of all natural numbers, formulas (wffs) of that language; and
one needs to prove the two antecedents, i.e.:
Induction step. Φ holds of any wffs that are con-
Base step. Φ is true of 0 structed out of simpler wffs provided that
those simpler wffs have Φ.
Induction step. for all natural numbers 𝑛, if Φ is
true of 𝑛, then Φ is true of 𝑛 + 1 Definition: The principle of proof induction:
In a logical system that contains derivations or proofs,
Typically, the induction step is proven by means of if Φ is true of a given step of the proof whenever Φ is
a conditional proof in which it is assumed that Φ true of all previous steps of the proof, then Φ is true
is true of 𝑛, and from this assumption it is shown of all steps of the proof.
that Φ must be true of 𝑛 + 1. In the context of this The principle of proof induction is an obvious corol-
conditional proof, the assumption that Φ is true of lary of the principle of strong induction applied to
𝑛 is called the inductive hypothesis. line numbers.
A related (and equivalent) principle:
Homework 2
Definition: The principle of strong (or complete)
induction states that: Answer all of these we don’t get to in class:
If (for all natural numbers 𝑛, whenever Φ is true of (1) Let Φ be the property a number 𝑥 has just in
all numbers less than 𝑛, Φ is also true of 𝑛) then Φ is case the sum of all numbers leading up to and
true of all natural numbers. including 𝑥 is 𝑥(𝑥+1)
2
. Use the principle of math-
ematical induction to show that Φ is true of all
In this class, we often use corollaries specific to natural numbers.
the study of logical systems. (These are my names; (2) Let Φ be the property a number 𝑥 has just in
in the book’s way of talking these are forms of case it is either 0 or 1 or it is evenly divisible by a
“structural induction”.) prime number greater than 1. Use the principle
Definition: The principle of formula induction of complete induction to show that Φ is true of
states that: all natural numbers.
For a given logical language, if Φ holds of the simplest (3) Let Φ be the property a wff 𝜓 of propositional
well-formed formulas (wffs) of that language, and Φ logic has if and only if has a even number of
holds of any complex wff provided that Φ holds of parentheses. Use the principle of formula induc-
those simpler wffs out of which it is constructed, then tion to show that Φ holds of all wffs of proposi-
Φ holds of all wffs. tional logic.

This is just a version of the above. Let Φ′ be the (4) Consider a logical system for propositional
property a number has if and only if all wffs having logic that has only one inference rule: modus
that number of logical operators have Φ. If Φ is ponens. Use the principle of proof induction to
true of the simplest well-formed formulas, then 0 show that every line of a proof in this system
has Φ . Similarly, if Φ holds of any wffs that are
′ is true if the premises are true.

4
4. Propositional Logic and
Truth Functions Result: Every truth function can be represented
by a propositional formula using ∨, ∧ and ¬ as
I used to do a whole unit on this; this year, we’re its only connectives.
going to skip to metatheory for first-order predicate
logic, with this brief exception.
The book uses these object language symbols:
Proof: Let 𝐺 be an 𝑛-place truth function. For each
¬ negation (alternatives ∼,−) ⟨𝑉1 , … , 𝑉𝑛 ⟩ such that ⟨⟨𝑉1 , … , 𝑉𝑛 ⟩, T⟩ ∈ 𝐺, consider
∧ conjunction (alternatives ⋅, &) the conjunction [¬]p1 ∧ [¬]p2 ∧ … [¬]p𝑛 , where a
∨ (inclusive) disjunction negation is attached to p𝑖 just in case 𝑉𝑖 is F; if 𝑉𝑖 is
→ (material) implication (alternatives ⊃, ⇒) T, use p𝑖 without a negation. Let 𝜑 be the disjunc-
↔ (material) equivalence (alternatives ≡, ⇔) tion of all such conjunctions. (If there are no such
⊥ absurd/falsehood (alternatives ✖, ) conjunctions, this can only be because 𝐺 always
⊤ triviality/truth yields falsity, and we can use the self-contradiction
I assume you are familiar with these. p1 ∧ ¬p1 .) It follows that 𝜑 represents 𝐺. For any
The language also contains atomic proposi- given ⟨⟨𝑉1 , … , 𝑉𝑛 ⟩, 𝑈 ⟩, if 𝑈 is truth, then the corre-
tional letters p1 , p2 , … , p𝑛 , … , q1 , q2 , … , q𝑛 , … . sponding conjunction is true, making the disjunc-
⊥ and ⊤ are not connectives but replacements tion true. If 𝑈 is falsity, then every conjunction in
for statement letters that have a fixed interpretation 𝜑 is false, making 𝜑 false. ❅
as something false, and true, respectively.
To put this back into “English”, take any arbitrary
The connectives can be thought of as represent-
final column of a truth table for, say, a formula with
ing truth-functions. (Negation is one-place; the
three propositional letters.
others two-place.)
p1 p2 p3 “Output”
Definition: An n-place truth function is a func-
T T T T
tion whose domain is ({T(ruth), F(alsity)}) 𝑛 and whose
T T F F
range is a subset of {T, F}.
T F T F
For example, we can think of ∨ as representing: T F F T
F T T T
{⟨⟨T, T⟩, T⟩, ⟨⟨T, F⟩, T⟩, ⟨⟨F, T⟩, T⟩, ⟨⟨F, F⟩, F⟩}
F T F F
𝜑 𝜓 𝜑∨𝜓 F F T F
T T T F F F F
Matching its truth table: T F T We represent this with a formula, 𝜑, by disjoining
F T T the cojunction corresponding to the “true rows”:
F F F
(p1 ∧ p2 ∧ p3 ) ∨ (p1 ∧ ¬p2 ∧ ¬p3 ) ∨ (¬p1 ∧ p2 ∧ p3 )
Definition: A formula 𝜑 with propositional letters
𝑃1 , … , 𝑃𝑛 represents the n-place truth function This formula is true on those rows, and no others.
𝐺 iff for each ⟨⟨𝑉1 , … , 𝑉𝑛 ⟩, 𝑈 ⟩ in 𝐺 if 𝑉𝑖 is the truth We could do the same for any other truth function.
value 𝑃𝑖 for all 𝑖, 1 ≤ 𝑖 ≤ 𝑛, then the formula 𝜑 has
truth value U.
Corollary: A set of connectives/propositional
With this definition you can see that p1 ∨ p2 repre- constants sufficient to define forms equivalent to
sents the function above. ¬, ∨ and ∧ is adequate to form formulas repre-
Because we haven’t formulated the syn- senting any truth function.
tax/semantics precisely, the following metatheo-
retic results are handled a bit informally.

5
Proof: The equivalent forms may be used in place (2) It cannot be that 𝜑 ∗ 𝜓 is true when 𝜑 and
of ¬, ∨ and ∧ in the above proof. ❅ 𝜓 are both true; or else a statement using
only true propositional letters formed using
Let | be the Sheffer stroke, meaning “not both”, it alone can only be true, and negations and
and ↓ be the Peirce/Sheffer dagger, meaning “nei- self-contradictions will be indefinable.
ther. . . nor. . . ”. The truth tables for these operators (3) For similar reasons it cannot be that 𝜑 ∗ 𝜓 is
are then: false when 𝜑 and 𝜓 are false, or any statement
using only false propositional letters formed
𝜑 𝜓 𝜑|𝜓 𝜑 𝜓 𝜑↓𝜓
using it alone can only be false; and negations,
T T F T T F tautologies, etc., will be indefinable.
T F T T F F (4) This means we know at least this much about
F T T F T F its truth table:
F F T F F T
𝜑 𝜓 𝜑∗𝜓
T T F
T F ?
Corollary: The following sets are each individ- F T ?
ually adequate to represent all truth functions: F F T
{¬, ∧}, {¬, ∨}, {¬, →}, {→, ⊥}, {|}, {↓}.
This leaves only four possibilities for the two
unknowns (?’s).
(5) Filling it in T and then F makes 𝜑 ∗ 𝜓 equivalent
Proof: For {¬, ∧}, note that ¬(¬𝜑∧¬𝜓) is equivalent with ¬𝜓, but negation alone can only define two
to 𝜑∨𝜓, so all of ¬, ∨ and ∧ can be represented from truth-functions: negation and double negation;
this set. iterating will just alternate between them.
Similarly, for {¬, ∨}, ¬(¬𝜑 ∨ ¬𝜓) is equivalent to (6) Filling it in F and then T makes 𝜑 ∗ 𝜓 equiva-
𝜑 ∧ 𝜓. lent with ¬𝜑, which is inadequate for the same
For {→, ⊥}, note that 𝜑 → ⊥ is equivalent to ¬𝜑, reason.
and (𝜑 → ⊥) → 𝜓 is equivalent with 𝜑 ∨ 𝜓. (7) Filling it in with T T gives the same function as
For the “not both” Sheffer stroke, |, we have: | and filling it in with F F gives ↓; therefore, ∗
𝜑 | 𝜑 is equivalent to ¬𝜑 must represent the same function that one of
(𝜑 | 𝜑) | (𝜓 | 𝜓) is equivalent to 𝜑 ∨ 𝜓 them does. ❅
(𝜑 | 𝜓) | (𝜑 | 𝜓) is equivalent to 𝜑 ∧ 𝜓
Homework 3
The others are left as homework. ❅
(1) Prove that {¬, →} and {↓} are adequate to rep-
resent all truth functions.
Result: The Sheffer stroke and Peirce/Sheffer (2) Give an example of a 3-place truth functional
dagger are the only binary operators capable operator that by itself is adequate to represent
of representing all truth functions by themselves, all truth functions.
or more exactly, any such binary operator must
represent the same truth function as one or the It would be possible to take a smaller set of propo-
other of them. sitional connectives as “primitive” as regard the
others as “defined” from them. This would even
simplify the deductive systems built on top of them.
The Open Logic Text does not do this, however,
Proof: (1) Suppose ∗ were such a binary operator. and uses ¬, ∧, ∨, →, ↔, ⊥ and ⊤ all as equally basic
parts of its symbolism.

6
Unit 1

Metatheory for First-Order Predicate Logic

1. Syntax The book’s convention on this is not entirely

clear, but it seems to use sans serif (but still ital-
To do metatheory properly, the exact grammar of icized) v 𝑛 for object language variables, and serif
your language needs to be spelled out precisely. I letters schematically.
allow myself some use/mention sloppiness.
Definition: An constant is one of the lowercase
letters early in the alphabet a to d, written with or
Basic building blocks without a numerical subscript.
Definition: A logical connective or logical oper-
E A Examples: a, a2 , b, c124 , etc.
ator is one of the symbols ¬, ∨, ∧, →, ↔, , .
Again, italicized serif 𝑎–𝑑 are schematic.
Definition: A propositional constant is either of
the symbols ⊥ or ⊤. Definition: A predicate is one of the uppercase let-
ters from A to T, written with a numerical superscript
Definition: A variable is one of the lowercase let- ≥ 1, and with or without a numerical subscript, or
ters near the end of the alphabet, v, w, x, y, or z, the identity predicate =2 .
written with or without a numerical subscript:
Examples: A1 , R2 , H4 , F12 , G34 , etc.
Examples: x, x1 , x12 , y, y2 , z, z13 , etc.
The superscript indicates how many terms the pred-
icate takes to form a statement. A predicate with
Important convention: I use the unitalicized sans
a superscript 1 is called a monadic predicate. A
serif letters “x”, “y”, etc., as object language vari-
predicate with a superscript 2 is called a binary or
ables, and italicized serif letters in the same range—
dyadic predicate.
“𝑥”, “𝑦”, “𝑧”—as metalinguistic schematic letters
It is customary to leave these superscripts off
for any object-language variables. Thus, e.g., the
when it is obvious from context what they must be.
schema:
A E.g., “R2 (a, b)” can be written simply “R(a, b)”.
𝑥(𝐹 𝑥 → 𝐺𝑥)
Represents all of: Definition: A function letter is one of the low-
ercase letters from f to l, written with a numerical
A 1
x(F x → G x) 1 superscript ≥ 1, and with or without a numerical
A
y(D1 y → E1 y) subscript.
x3 (L12 x3 → M13 x3 ), etc.
A
Examples: f1 , g2 , h33 , etc.

The difference is subtle, and usually not so impor- The numerical superscript indicates how many ar-
tant to keep straight. After all, object language gument places the function letter has. Again, ob-
variables tend to be interchangeable. vious superscripts can be left off.

7
Constructed expressions If you prefer ( 𝑥) or (𝑥) to 𝑥, that’s fine; I might
A A
not even notice.
Definition: A term of the language is defined re-
As a shorthand convenience, we allow ourselves
cursively as follows:
to omit parentheses with the convention that if
(i) all variables are terms;
things are otherwise ambiguous, the operators ear-
(ii) all constants are terms;
lier on this list are to have wider scopes than those
(iii) if 𝑓 is a function letter with superscript 𝑛, and
later on the list:
𝑡1 , … , 𝑡𝑛 are terms, then 𝑓 (𝑡1 , … , 𝑡𝑛 ) is a term;
(iv) nothing that cannot be constructed by repeated E A
↔, →, ∨, ∧, 𝑥, 𝑥, ¬
applications of the above is a term.
Hence:
Examples: a, x, f(a), g(x, f(y)), etc.
A E
𝑥(𝐹 𝑥 ∧ 𝐺𝑥 → ¬ 𝑦𝐹 𝑦 ∨ 𝐺𝑏)
I use italicized lowercase letters 𝑡, 𝑡1 , 𝑡2 , 𝑢, 𝑢1 , etc.,
schematically for any terms, simple or complex. Should be interpreted to mean:
Unitalicized t or u is never used. A E
𝑥((𝐹 𝑥 ∧ 𝐺𝑥) → ((¬ 𝑦𝐹 𝑦) ∨ 𝐺𝑏))
Definition: A closed term is a term that does not
contain variables (only constants and function let- Repeated operators of the same type associate to
ters). the left, so if we write:

Definition: An atomic formula is a propositional 𝐹 𝑎 ∧ 𝐹 𝑏 ∧ 𝐺𝑐 ∨ 𝐺𝑎 ∨ 𝑅𝑎𝑏

constant or any expression of the form 𝐹 (𝑡1 , … , 𝑡𝑛 )
where 𝐹 is a predicate with superscript 𝑛, and This means:
𝑡1 , … , 𝑡𝑛 are all terms.
((((𝐹 𝑎 ∧ 𝐺𝑏) ∧ 𝐺𝑐) ∨ 𝐺𝑎) ∨ 𝑅𝑎𝑏)
Examples: ⊥, ⊤, F1 (a), R(y, b), F1 (f(x)), R34 (a, b, c),
We also allow other styles [ . . . ] { . . . } to be used if
H4 (x, b, y, g(a, x)), etc.
it helps find the matching parentheses.
Atomic formulas are those that do not have other Definition: A first-order language  is any lan-
formulas as parts. We can leave out parentheses guage that makes use of the above definition of a
after predicates and commas between their terms formula, or modifies it at most by restricting which
if no function letters are used, e.g. “Fa” for “F(a)” constants, function letters and predicates are utilized
and “Rxy” for “R(x, y)”. (provided that it uses at least one predicate letter).

E.g., a language that does not use function letters

Definition: A (well-formed) formula (wff) is re-
still counts as a first-order language. So would a
cursively defined as follows:
language that used no constants and only a single
(i) any atomic formula is a wff;
predicate, say, a set theory whose only predicate
(ii) if 𝜑 is a wff, then ¬𝜑 is a wff;
was for the membership relation.
(iii) if 𝜑 and 𝜓 are wffs, then (𝜑 ∨ 𝜓), (𝜑 ∧ 𝜓), (𝜑 →
𝜓) and (𝜑 ↔ 𝜓) are wffs;
(iv) if 𝜑 is a wff and 𝑥 is a variable, then 𝑥𝜑 and
A Free and bound variables
E
𝑥𝜑 are wffs; A E
Definition: When a quantifier 𝑥 or 𝑥 occurs as
(v) nothing that cannot be constructed by repeated part of a wff 𝜑, the scope of the quantifier is defined
applications of the above is a wff. A E
as the smallest part 𝑥𝜓 or 𝑥𝜓 of 𝜑 such that 𝑥𝜓
A
E
or 𝑥𝜓 is itself a wff.

8
Definition: If 𝑥 is a variable that occurs within a
wff 𝜑, then an occurrence of 𝑥 in 𝜑 is said to be a
bound occurrence iff it occurs in the scope of a
A E
quantifier 𝑥 or 𝑥 (using 𝑥) within 𝜑; otherwise,
the occurrence of 𝑥 is said to be a free occurrence.
Examples: 1. All three occurrences of “x” in
“ x(Fx → Fx)” are bound.
A
2. The (solitary) occurrence of “x” in “Fx → yGy”
A
is free.
Definition: A variable 𝑥 that occurs within a wff
𝜑 is said to a bound variable, or simply bound, iff
there is at least one bound occurrence of 𝑥 in 𝜑.
Definition: A variable 𝑥 that occurs within a wff 𝜑
is said to be a free variable, or simply free, iff there
is at least one free occurrence of 𝑥 in 𝜑.
Notice that “x” is both bound and free in “Fx →
xGx”, because some occurrences are bound and
A
one is free.
Definition: A wff 𝜑 is said to be closed iff 𝜑 con-
tains no free variables; otherwise 𝜑 is said to be open.
If “𝜑” is being used in the metalanguage schemati-
cally for a formula, and 𝑡 for a term free for variable
𝑥 in 𝜑, then the notation “𝜑[𝑡/𝑥]” means the for-
mula that results when all free occurrences of 𝑥 are
replaced by the term 𝑡; this notation is part of the
metalanguage only.
Although less precise, sometimes we shall use
“𝜑(𝑥)” schematically for a formula that is expected
to have 𝑥 free in it; and if we also use “𝜑(𝑡)” in that
context, then what is meant is also what results
by replacing all free occurrences of 𝑥 with 𝑡. Of
course, however, “𝜑” here should not be thought
of as predicate, or even a schematic letter for a
predicate.
Notice in each case that only free variables are
replaced, so if 𝜑(x) is “Gx → x Fx”, then 𝜑(y) is
A
“Gy → x Fx”, not “Gy → y Fy” .
A A
The book uses a symbol ≡ as the metalinguistic
relation of syntactic identity; I avoid this, to reduce
possible confusion for those accustomed to using
this symbol for material equivalence in the object
language.
Homework 4

Definition: A sentence is a closed wff. (1) Determine which occurrences of variables are
bound and which are free in the following wffs:
E A
Definition: If 𝜑 is a wff, 𝑡 is a term and 𝑥 is a vari- x y(Rxy → Syz)
E A
able, then 𝑡 is free for 𝑥 in 𝜑 iff no free occurrence of x( y Ryx ↔ Ryx)
E A E
𝑥 in 𝜑 lies within the scope of some quantifier 𝑦 or
A x¬ y Fyx → z(Fz ∧ Rxz)
E
𝑦 where 𝑦 is a variable occurring in 𝑡 . (2) For each of the formulas 𝜑 on the left, deter-
mine, for each term 𝑡 on the right, which are
Basically, this means that if you substitute 𝑡 for all and which are not free for “x” in 𝜑:
the free occurrences of 𝑥 within 𝜑, you won’t end formulas terms
up inadvertently “binding” any of the variables that Fx y
happen to occur in 𝑡. Fa f(x)
A
yRyx f(x, y)
Examples: E A
y xRyx f(x, a)
1. “a” is free for “x” in “ yGy → Gx”.
A
E
y(Fx ∧ Fy) f(a, y, g(x))
2. “f2 (x, z)” is free for “x” in “ yGy → Gx”.
A
3. “z” is not free for “x” in “ yGy → zRxz”.
A A
4. “f2 (a, z)” is not free for “x” in “ yGy → zRxz”.
E A 2. Semantics
5. “f2 (a, z)” is free for “x” in “ yGy → z xRxz”.
E A A
6. All terms are free for “x” in “ yGy”.
A Interpretations
7. All terms are free for “y” in “ yGy”.
E
Definition: A structure or interpretation M for
a first order language  consists of the following four
things:

9
(1) The specification of some non-empty set, written
|M| to serve as the domain of quantification
for the language.
This set is the sum total of entities the quanti-
fiers are interpreted to “range over”.
(2) A function assigning, for each constant in the
language, some fixed member of |M| for which
it is taken to stand.
For a given constant 𝑐, this member is denoted
in the metalanguage by 𝑐M .
(3) A function assigning, for each predicate 𝐹 with
superscript 𝑛 in the language, some subset of
|M|𝑛 , written 𝐹 M .
That is, the interpretation assigns to each pred-
icate a set of 𝑛-tuples from |M|, its “extension”.
(4) An assignment, for each function letter 𝑓 with
superscript 𝑛 in the language, some 𝑛-place oper-
ation on |M|, written 𝑓 M .
You can take “interpretation” here pretty literally;
an interpretation is basically an assignment of a uni-
verse of discourse to the quantifiers, and an appro-
priate semantic value to each non-logical constant
of the language.
Variable assignments and satisfaction
Definition: A variable assignment, 𝑠, for inter-
pretation M, is a function that maps every variable
of the language of M to a member of |M|.
There can be (and unless M has only a single
member always are) many different variable as-
signments for any given interpretation.
E.g., if |M| is {John, Paul, George, Ringo}, then
there is one that maps every variable to John, one
that maps every variable to Ringo, one that maps
“x” to John and the rest to Ringo, etc.
Suppose we have an open formula, “Gx”, mean-
ing, e.g., “x is a guitarist”. A given interpretation
(structure)—such as the one that makes GM the set
{John, George}—does not say that this is true or
false, because it depends on “x”. But we say the
formula is satisfied when “x” is assigned to George.
Technically, we make satisfaction a relationship
between a variable assignment and a formula (open
or closed). A variable assignment mapping “x” to
10
George satisfies “Gx” on this interpretation, but one
mapping “x” to Ringo will not. The same variable
assignment that satisfies “Gx” might not satisfy “Gy”
if it maps “y” to Ringo.
Definition: The value of term 𝑡 for variable as-
signment 𝑠 for interpretation M, written ValM 𝑠 (𝑡), is
defined recursively as follows:
(1) If 𝑡 is a constant 𝑐, then ValM M
𝑠 (𝑡) is 𝑐 ;
(2) if 𝑡 is a variable 𝑥, then Val𝑠 (𝑡) is 𝑠(𝑥);
M
(3) if 𝑡 is a function term 𝑓 (𝑡1 , … , 𝑡𝑛 ), then ValM 𝑠 (𝑡) is
𝑓 M (ValM 𝑠 (𝑡1 ), … , ValM
𝑠 (𝑡 𝑛 )), i.e., the value of the
function assigned to 𝑓 in the interpretation for
the values of the argument terms as arguments.
Definition: The notion of a variable assignment 𝑠
in interpretation M satisfying a given formula 𝜑,
written M, 𝑠 ⊨ 𝜑, is defined recursively. For a given
variable assignment 𝑠 in interpretation M:
(i) M, 𝑠 ⊨ ⊤, always holds; M, 𝑠 ⊨ ⊥ never holds.
(We can write this as M, 𝑠 ⊭ ⊥.)
I.e., ⊤ and ⊥ are, respectively, satisfied by ev-
ery assignment for every interpretation, and
satisfied by no assignment for any interpre-
tation.
(ii) If 𝜑 is an atomic wff of the form 𝐹 (𝑡1 , … , 𝑡𝑛 ),
then M, 𝑠 ⊨ 𝜑 iff ⟨ValM M
𝑠 (𝑡1 ), … , Val𝑠 (𝑡𝑛 )⟩ ∈
𝐹 M.
I.e., atomic wffs are satisfied when the values
of their terms on the assignment are members
of the extension the interpretation assigns to
the predicate.
(iii) If 𝜑 is of the form ¬𝜓, then M, 𝑠 ⊨ 𝜑 iff M, 𝑠 ⊭
𝜓.
I.e., an assignment satisfies a negation just in
case it does not satisfy what it negates.
(iv) If 𝜑 is of the form 𝜓 ∨𝜃, then M, 𝑠 ⊨ 𝜑 iff either
M, 𝑠 ⊨ 𝜓 or M, 𝑠 ⊨ 𝜃.
I.e., an assignment satisfies a disjunction iff
it satisfies either of the two disjuncts.
(v) If 𝜑 is of the form 𝜓 ∧ 𝜃, then M, 𝑠 ⊨ 𝜑 iff both
M, 𝑠 ⊨ 𝜓 and M, 𝑠 ⊨ 𝜃.
I.e., an assignment satisfies a conjunction iff
it satisfies both conjuncts.
 (vi) If 𝜑 is of the form 𝜓 → 𝜃, then M, 𝑠 ⊨ 𝜑 iff Homework 5
either M, 𝑠 ⊭ 𝜓 or M, 𝑠 ⊨ 𝜃.
I.e., an assignment satisfies a conditional iff (1) Does 𝑠 satisfy “Gx → x Gx”? Explain your
A
it does not satisfy the antecedent, or does answer.
satisfy the consequent. (2) What about 𝑠 ′ , 𝑠 ′′ and 𝑠 ′′′ ? Again, explain your
(vii) If 𝜑 is of the form 𝜓 ↔ 𝜃, then M, 𝑠 ⊨ 𝜑 iff answers.
either both M, 𝑠 ⊨ 𝜓 and M, 𝑠 ⊨ 𝜃 or both Some important results are noted here. We are not
M, 𝑠 ⊭ 𝜓 and M, 𝑠 ⊭ 𝜃. going to bother with proofs for all of them.
I.e., an assignment satisfies a biconditional iff
it satisfies either both or neither side.
Result:
(viii) If 𝜑 is of the form 𝑥𝜓, then M, 𝑠 ⊨ 𝜑 iff every
A
variable assignment 𝑠 ′ differing from 𝑠 by at Modus ponens preserves satisfaction, i.e., for all
most its value for variable 𝑥 (including 𝑠 itself) 𝜑, 𝜓, M and 𝑠 for M, if M, 𝑠 ⊨ 𝜑 and M, 𝑠 ⊨
satisfies 𝜓. 𝜑 → 𝜓 then M, 𝑠 ⊨ 𝜓.

(ix) If 𝜑 is of the form 𝑥𝜓, then M, 𝑠 ⊨ 𝜑 iff at

E
least one variable assignment 𝑠 ′ differing from
𝑠 by at most its value for variable 𝑥 (including Proof: Assume both M, 𝑠 ⊨ 𝜑 and M, 𝑠 ⊨ 𝜑 → 𝜓.
𝑠 itself) satisfies 𝜓. Because of the definition of satisfaction for formu-
To understand the last two clauses, let |M| again las of the form 𝜑 → 𝜓, the second assumption means
be {John, Paul, George, Ringo} and let GM be {John, that either M, 𝑠 ⊭ 𝜑 or M, 𝑠 ⊨ 𝜓; the first option is
George}. Consider variable assignment 𝑠 that as- ruled out by the first assumption, so M, 𝑠 ⊨ 𝜓. ❅
signs every variable to John. There are three other
variable assignments that differ from it by at most
what it assigns to “x”: Result: If 𝑠 and 𝑠 ′ are two variable assignments
variable 𝑠 𝑠′ 𝑠 ′′ 𝑠 ′′′ for interpretation M, 𝜑 is a formula, and 𝑠(𝑥) =
v John John John John 𝑠 ′ (𝑥) for every variable 𝑥 that occurs free in 𝜑,
w John John John John then M, 𝑠 ⊨ 𝜑 iff M, 𝑠 ′ ⊨ 𝜑.
x John Paul George Ringo
y John John John John
z John John John John
v1 John John John John
w1 John John John John Corollary: If 𝑠 and 𝑠 ′ are two variable assign-
x1 John John John John ments for interpretation M, 𝜑 is a closed for-
y1 John John John John mula/sentence, then M, 𝑠 ⊨ 𝜑 iff M, 𝑠 ′ ⊨ 𝜑 al-
z1 John John John John ways.
⋮ ⋮ ⋮ ⋮ ⋮
𝑠 satisfies “Gx” because 𝑠’s value for “x”, viz., John, is
a guitarist. But for 𝑠 to satisfy “ x Gx”, this means
A
that not only 𝑠, but also 𝑠 ′ , 𝑠 ′′ and 𝑠 ′′′ all would
have to satisfy “Gx”, but 𝑠 ′ and 𝑠 ′′′ do not. There- Truth, falsity and models
fore not even 𝑠 satisfies “ x Gx”. However, for 𝑠
A
Definition: A wff 𝜑 is true in interpretation M iff
to satisfy “ x Gx”, only one or more of these has
E
for every variable assignment 𝑠 for M it holds that
to satisfy “Gx”; since 𝑠 itself satisfies “Gx”, 𝑠 also M, 𝑠 ⊨ 𝜑; we write this simply as M ⊨ 𝜑.
satisfies “ x Gx”.
E

11
Definition: A wff 𝜑 is false in interpretation M If Γ has a single member, 𝜓, we may write simply
iff M ⊨ ¬𝜑. (I.e., falsity is truth of the negation.) 𝜓 ⊨ 𝜑 and say that 𝜓 logically implies 𝜑; or
for any finite number of members list them as
𝜓1 , … , 𝜓𝑛 ⊨ 𝜑.
Homework 6
Definition: A set of sentences Γ is satisfiable iff
(1) Give an example of a formula 𝜑 and an inter- there is at least one interpretation M such that M ⊨
pretation M where 𝜑 is neither true nor false Γ and unsatisfiable otherwise.
in M, and explain why.
(2) Prove that every sentence 𝜑 is either true or
false in every interpretation M. (You may use Result: If 𝜑 is a sentence, ⊨ 𝜑 iff {¬𝜑} is unsat-
the corollary above.) isfiable.
Definition: If M is an interpretation and Γ a set of
formulas, M is a model of Γ, iff, for all 𝜑 ∈ Γ, M ⊨
𝜑; we abbreviate this by writing M ⊨ Γ.
Proof: (1) First suppose that ⊨ 𝜑 but for reductio
suppose that {¬𝜑} is satisfiable. There is then
By now, we are using the sign ⊨ in a number of
some M such that M ⊨ ¬𝜑. For every 𝑠 in M,
different ways, and there are more to come! Note
M, 𝑠 ⊨ ¬𝜑 and by the definition of satisfaction,
that all uses of this sign are metalinguistic. It cannot
M, 𝑠 ⊭ 𝜑. But this contradicts the fact that
be used in the object language, and certainly not
M ⊨ 𝜑 for every M.
embedded inside an object language statement such
(2) For the other direction, suppose that {¬𝜑} is
as “Fa → (Ga ⊨ Fa)” or something like that.
unsatisfiable. Then there is no M such that
Notice that every interpretation will be a model
M ⊨ ¬𝜑. I.e., for every M there is an 𝑠 for
for some sets of wffs. In fact, the study of formal se-
which M, 𝑠 ⊭ ¬𝜑, and so M, 𝑠 ⊨ 𝜑. Since 𝜑
mantics for artificial languages is sometimes called
is a sentence, the same holds for every other
“model theory”. I tend to use the words “structure”,
assignment 𝑠 in M, and hence M ⊨ 𝜑. Since
“model” and “interpretation” interchangeably.
this is true for every M, it follows that ⊨ 𝜑. ❅

Logical validity and related notions

Definition: A wff 𝜑 is said to be logically true or
(logically) valid iff 𝜑 is true in every interpretation Result: For any sets of sentences Γ and Δ, and
M. This can be abbreviated by writing simply ⊨ 𝜑. every sentence 𝜑, if Γ ⊆ Δ and Γ ⊨ 𝜑 then Δ ⊨ 𝜑.

Just like we can leave off the 𝑠 before the ⊨ when

the formula is satisfied by every assignment in a
given interpretation, we can leave off both the 𝑠 Proof: Assume that Γ ⊆ Δ and Γ ⊨ 𝜑. Let M be
and the M when the formula is satisfied by every any interpretation that makes every member of Δ
assignment in every interpretation. true; since Γ is a subset, this means that M makes
every member of Γ true as well. Because Γ ⊨ 𝜑
Definition: A set of sentences Γ entails formula 𝜑 this means that M ⊨ 𝜑. Since M is arbitrary, this
iff 𝜑 is true in every interpretation that is a model of means that Δ ⊨ 𝜑. ❅
Γ, i.e., for every M, if M ⊨ Γ then M ⊨ 𝜑. We also
write this as Γ ⊨ 𝜑.
Homework 7
We also say that 𝜑 is a logical consequence of Γ iff
Γ entails 𝜑.
(1) Prove that {⊥} is unsatisfiable.

12
(2) Prove that for every set of sentences Γ, Γ ⊨ ⊥ • Sequent calculi (Kind of weird; rules are stated
iff Γ is unsatisfiable. in terms of left and right sequences where each
step means that the truth of all the left formulas
Definition: Two formulas 𝜑 and 𝜓 are logically requires the truth of at least one right formula.)
equivalent iff in every interpretation M, 𝜑 and 𝜓
are satisfied by the same variable assignments. • Tableaux (Tree like structures with branching
rules tracking the semantic rules of the language.)

• Axiomatic systems (We’ll study these in a few

Result: If 𝜑 and 𝜓 are sentences, they are log-
ically equivalent if and only if they are true in
the same interpretations.
Proof: (1) Suppose 𝜑 and 𝜓 are equivalent, and let
𝜑 be true in an interpretation M. Hence every
variable assignment satisfies 𝜑, and because 𝜓 is
equivalent, every variable assignment satisfies
𝜓, and so 𝜓 is true in M. The same is true in
reverse, and so 𝜑 and 𝜓 are true in the same
interpretations.
(2) For the reverse, suppose 𝜑 and 𝜓 are true in
the same interpretations. Let 𝑠 be any variable
assignment in any interpretation M. If 𝑠 satis-
fies 𝜑, all other assignments do as well, because
𝜑 is a sentence. Hence, 𝜑 is true, and so is 𝜓,
and 𝑠 satisfies 𝜓 as well. If 𝑠 does not satisfy 𝜑,
then 𝜑 is not true, and neither is 𝜓. Hence some
variable assignment 𝑠 ′ does not satisfy 𝜓 in M;
because 𝜓 is closed, neither does 𝑠. Therefore, 𝜑
and 𝜓 are satisfied by the same variable assign-
ments in all interpretations, and are logically
equivalent. ❅ rule for it is applied, and can be “checked off”
Tableaux and semantics
Strictly speaking, tableaux or “logic trees” can be
considered a form of deductive system or derivation
system. Indeed, the Open Logic Text distinguishes
four kinds of derivation systems:
• Natural deduction systems (There are probably
what you learned in your introductory classes;
lots of rules with techniques such as indirect and
conditional proofs.)
days; they are the focus of most metatheory.)
If we were doing propositional logic, we could add
“truth tables” to this list. Are truth tables a deduc-
tive system? Technically, yes, but they are designed
to track the semantic rules for truth functions. Be-
cause of this they can be used to demonstrate the
invalidity of an argument, or non-tautologyhood
of a formula.
Tableaux are similar, and the closest we come
to a “decision procedure” for predicate logic. Un-
like natural deduction and axiomatic derivation
systems, they are at least sometimes effective at
helping us find an interpretation that demonstrates
the invalidity of a sentence or argument.
The method:
(1) A tableau takes the form of a downward grow-
ing tree representing a hypothetical interpreta-
tion. At the base are a number of “assumptions”,
which are T or F written before a sentence, rep-
resenting the truth or falsity, respectively, of
the sentence in question in the hypothetical
interpretation.
(2) For each sentence on the branch, the tableau
when the rule is applied. (In the case of true
universals and false existentials, this checking
off is provisional, more on this below.)
(3) Some rules create branches, both of which need
to be “explored”. The branches represent differ-
ent possible ways the hypothetical interpreta-
tion could be.
(4) A branch closes if it makes the same sentence
both true and false. We mark it closed with ⊗.
(5) If the rules are correctly applied to each state-
ment on the branch, no rules reset a true uni-
versal or false existential checkmark, and it
remains open, the open branch describes an

13
 interpretation consistent with the opening as- T𝜑∨𝜓 F𝜑∨𝜓
sumptions. F𝜑
(6) To avoid loops of steps preventing progress, do T𝜑 T𝜓 F𝜓
true existentials and false universals as soon as
possible, but delay re-doing true universals or conjunctions
false existentials afterwards until they are all T𝜑∧𝜓 F𝜑∧𝜓
that is left. T𝜑
(7) It is still possible, however, to get stuck in an T𝜓 F𝜑 F𝜓
infinite loop of steps, which is what prevents
this from being a full decision procedure for
predicate logic. conditionals
T𝜑→𝜓 F𝜑→𝜓
Uses for tableaux: T𝜑
• Testing if a formula is valid by looking for an F𝜑 T𝜓 F𝜓
interpretation in which it is false. biconditionals
• Testing if a set of formulas is satisfiable by look- T𝜑↔𝜓 F𝜑↔𝜓
ing for an interpretation in which they are all
true.
T𝜑 F𝜑 T𝜑 F𝜑
• Testing if a formula is entailed by a set of for-
T𝜓 F𝜓 F𝜓 T𝜓
mulas by looking for an interpretation in which
the members of the set are true, but the alleged universals
entailment is false. (I.e., testing if an argument is
A A
T 𝑥𝜑 F 𝑥𝜑
valid.) T 𝜑[𝑡1 /𝑥] F 𝜑[𝑐/𝑥]
⋮ where 𝑐 is an
Tableaux rules T 𝜑[𝑡𝑛 /𝑥] unused constant
for all closed terms 𝑡𝑛 *
For each category of wff, there is T rule and an F
existentials
rule. What you have is the first line; what you get E E
is below that. Rules must be applied to all open T 𝑥𝜑 F 𝑥𝜑
branches below. T 𝜑[𝑐/𝑥] F 𝜑[𝑡1 /𝑥]
where 𝑐 is an ⋮
propositional constants unused constant F 𝜑[𝑡𝑛 /𝑥]
for all closed terms 𝑡𝑛 *
T⊤ F⊤ T⊥ F⊥
(do nothing) ⊗ ⊗ (do nothing)
* = Do at least one, using the term “a” if there are
other atomics no closed terms yet. Since new closed terms may
T 𝐹 (𝑡1 , … 𝑡𝑛 ) F 𝐹 (𝑡1 , … 𝑡𝑛 ) appear later on the branch, these rules may need to
if opposite above if opposite above be reapplied later, and so no checkmark is “final”.
⊗ ⊗ Example (1):
negations Validity of x(Fa → Gx) → (Fa → x Gx).
A A

T ¬𝜑 F ¬𝜑
F𝜑 T𝜑
disjunctions

14
 fixes what we need to intepret the constants
A A
F x(Fa → Gx) → (Fa → x Gx) ✔
and function letters.
A
T x(Fa → Gx) ✔
E.g., let “a”M = Adele and “b”M = Britney.
A
F Fa → x Gx ✔
T Fa ✔ (3) Include a given object or tuple in the extension
of a predicate just in case the corresponding
A
F x Gx ✔
F Gb ✔ atomic statement appears with T on the branch.
T Fa → Ga So here we have “F”M = {Adele} and “G”M =
T Fa → Gb ✔ {Britney}, since “Fa” and “Gb” are true, but “Fb”
and “Ga” are false.
F Fa ✔ T Gb ✔ (4) The resulting interpretation should make all
⊗ ⊗ the sentences with T in front of them on the
branch true, and those with F in front of them
false.
No possible interpretation can make good on our
initial assumption, and so all must make this sen- In the book you can find a proof of the last claim,
tence true; it is therefore logically valid. and from it, the overall metatheory of tableaux
Example (2): x Fx ∧ x Gx does not entail x(Fx ∧ proving the method sound and complete. In this
E E E
Gx).
course, we will reserve such metatheoretic focus
for axiomatic deductions.
E
T x Fx ∧ x Gx ✔
E The book also annotates tableaux with line num-
E
F x(Fx ∧ Gx) ✔ bers and rule markers; I do not bother.
E
T x Fx ✔ Example (3): An inconclusive tree.
E
T x Gx ✔
T Fa ✔ E A
F x y Rxy → Fa ✔
T Gb ✔ E A
T x y Rxy ✔
F Fa ∧ Ga ✔ F Fa ✔
F Fb ∧ Gb ✔ E
T y Ray ✔
T Rab ✔
E
F Fa ✔ F Ga ✔ T y Rby ✔
⊗ T Rbc ✔
E
F Fb ✔ F Gb ✔ T y Rcy ✔
T Rcd ✔
⊗
⋮
Here we have applied to rule to each line in the
tree, and there are no new closed terms to apply to Each time we add a new constant, it triggers us
the false existential. to redo the true universal, yielding another true
existential, and another new constant, and on it
Describing an interpretation from an open tableau goes. In a case like this, the loop is easy to see, and
branch. it is intuitively clear that the branch will remain
(1) Let |M| have as many elements as there are open. In fact, obviously, it is possible to make this
closed terms on the branch. formula false. Let |M| be the set of natural numbers,
In the example, we have two constants, “a” and let “a”M be zero, let “R”M be the less-than relation
“b”, so let |M| be {Adele, Britney}. and let “F”M be the set of odd numbers.
(2) Assign an item in the domain to each closed But there is no generic “test” that would allow
term; it could be a number, or even the term us to know that branches in general must terminate
itself. Sometimes it’s more fun to pick other ob- in any finite number of steps if they are going to
jects easy to remember from the constants. This close at all. If I have done 10 million steps, is that

15
always enough? Not in general, so this does not
always provide an effective test.
In fact later in the semester, we’ll prove that
there is no effective decision procedure for testing
validity in first-order predicate logic.
Definition: A formula 𝜑 is a theorem of an ax-
iomatic system iff ∅ ⊢ 𝜑, i.e, iff there is a derivation
of it using only axioms and inference rules. We may
write this simply ⊢ 𝜑.

Homework 8 Propositional axioms and MP

Construct tableaux to test the following; for those Classical truth-functional propositional logic can
that are invalid, describe a counterexample inter- be “axiomatized” by taking every formula of one of
pretation. Is it true that . . . the following forms (—there are infinitely many of
(1) ⊨ (Fa ↔ Ga) → (¬Fa ∨ ¬Ga)? each, except (VerumAx)—) as an axiom:
(2) ⊨ x(Fx → y Gy) → x(Fx ∧ Gx)?
E A E
(3) x y Rxy ⊨ x Rxx?
A A A (𝜑 ∧ 𝜓) → 𝜑 (SimpLAx)
(4) ¬ x y Ryx ⊨ x y¬Rxy? (SimpRAx)
A E A E
(𝜑 ∧ 𝜓) → 𝜓
𝜑 → [𝜓 → (𝜑 ∧ 𝜓)] (ConjAx)
3. Axiomatic Deductions 𝜑 → (𝜑 ∨ 𝜓) (AddRAx)
𝜑 → (𝜓 ∨ 𝜑) (AddLAx)
As mentioned above, there are different kinds of (𝜑 → 𝜃) → {(𝜓 → 𝜃) → [(𝜑 ∨ 𝜓) → 𝜃]}
derivations or deduction systems for logic. Ax- (CasesAx)
iomatic deductions were the first to be rigorously
𝜑 → (𝜓 → 𝜑) (TCAx)
used or studied, and the easiest to do metatheory
for, although they can be cumbersome to use. [𝜑 → (𝜓 → 𝜃)] → [(𝜑 → 𝜓) → (𝜑 → 𝜃)]
The chief/defining features are: (CMPAx)
(𝜑 → 𝜓) → [(𝜑 → ¬𝜓) → ¬𝜑] (CMTAx)
• Each deduction can be characterized as a single
ordered series of formulas: 𝜓1 , 𝜓2 , … , 𝜓𝑛 , etc. The ¬𝜑 → (𝜑 → 𝜓) (FAAx)
conclusion is always the last of the series. ⊤ (VerumAx)
• All deductions are direct; there are no special ⊥→𝜑 (FalsumAx)
types of deductions such as “conditional” or “in- (𝜑 → ⊥) → ¬𝜑 (RedAx)
direct”.
¬¬𝜑 → 𝜑 (DNEAx)
• The number of inference rules is kept small, using
axioms in their place where possible. (An “axiom” Propositional logic then requires a single inference
is a line of a deduction that does not need to be rule: modus ponens (MP): if 𝜑𝑗 → 𝜑𝑖 and 𝜑𝑗 already
“justified” by another line.) occur in a derivation, 𝜑𝑖 may follow.
Definition: An axiomatic derivation or proof of Instances of these forms are axioms, but the
a result 𝜑 from a set of formulas Γ is a finite ordered forms themselves can be called axiom schemata.
series of formulas 𝜓1 , 𝜓2 , … , 𝜓𝑛 , where the last item
𝜓𝑛 is 𝜑, and each step 𝜓𝑖 either (a) is an axiom, (b) is ⊢ Fa → Fa
a member of Γ, or (c) follows from previous members 1. Fa → [(Fa → Fa) → Fa] TCAx
of the series by an inference rule. 2. {Fa → [(Fa → Fa) → Fa]} → {[Fa →
(Fa → Fa)] → (Fa → Fa)} CMPAx
Definition: A formula 𝜑 is derivable from a set Γ 3. [Fa → (Fa → Fa)] → (Fa → Fa) 1, 2 MP
iff there exists at least one axiomatic derivation of 𝜑 4. Fa → (Fa → Fa) TCAx
from Γ; this is written Γ ⊢ 𝜑. 5. Fa → Fa 3, 4 MP
If Γ’s members can be listed, we may write simply Usually, in this class we are not interested in par-
𝜓1 , … , 𝜓𝑛 ⊢ 𝜑. ticular theorems in the object language, but forms;

16
we can make the above proof general by replacing This, with the above result 𝜑 ⊢ ¬¬𝜑, mean that if
“Fa” with 𝜑: we ever have a line 𝜑, we can get its double nega-
tion. We could just insert the steps of the derivation
⊢𝜑→𝜑 (“Ref→”)
schema above. So from now on, we will not bother,
1. 𝜑 → [(𝜑 → 𝜑) → 𝜑] TCAx and simply treat “DNI” as if it were an inference
2. {𝜑 → [(𝜑 → 𝜑) → 𝜑]} → {[𝜑 → (𝜑 → 𝜑)] → rule. We call this a derived rule.
(𝜑 → 𝜑)} CMPAx
3. [𝜑 → (𝜑 → 𝜑)] → (𝜑 → 𝜑) 1, 2 MP
4. 𝜑 → (𝜑 → 𝜑) TCAx Corollary: If Γ ⊢ 𝜑1 and . . . and Γ ⊢ 𝜑𝑛 and
5. 𝜑 → 𝜑 3, 4 MP 𝜑1 , … , 𝜑𝑛 ⊢ 𝜓, then Γ ⊢ 𝜓.
This is a derivation schema that shows that every
instance of 𝜑 → 𝜑 is a theorem. We can say that
𝜑 → 𝜓, ¬𝜓 ⊢ ¬𝜑 (“MT”)
𝜑 → 𝜑 is a theorem schema.
Once we have established a theorem schema, 1. 𝜑 → 𝜓 Premise
we can treat it like an axiom schema in subsequent 2. ¬𝜓 Premise
3. ¬𝜓 → (𝜑 → ¬𝜓)
proofs, since it would merely be a matter of replac- TCAx
ing that line of the proof with the corresponding 4. 𝜑 → ¬𝜓 2, 3 MP
steps. We’ll cite the above as “Ref→”. 5. (𝜑 → 𝜓) → ((𝜑 → ¬𝜓) → ¬𝜑) CMTAx
6. (𝜑 → ¬𝜓) → ¬𝜑 1, 5 MP
𝜑 ⊢ ¬¬𝜑 (“DNI”) 7. ¬𝜑 4, 6 MP

1. 𝜑 Premise Again, from now on, we’ll just pretend MT is an

2. 𝜑 → (¬𝜑 → 𝜑) TCAx inference rule just like MP.
3. ¬𝜑 → 𝜑 1, 2 MP
4. (¬𝜑 → 𝜑) → ((¬𝜑 → ¬𝜑) → ¬¬𝜑) CMTAx 𝜑 → 𝜓, 𝜓 → 𝜃 ⊢ 𝜑 → 𝜃 (“HS”)
5. (¬𝜑 → ¬𝜑) → ¬¬𝜑 3, 4 MP 1. 𝜑 → 𝜓 Premise
6. ¬𝜑 → ¬𝜑 Ref→ 2. 𝜓 → 𝜃 Premise
7. ¬¬𝜑 5, 6 MP 3. (𝜓 → 𝜃) → (𝜑 → (𝜓 → 𝜃)) TCAx
Strictly speaking the above does not meet the defi- 4. 𝜑 → (𝜓 → 𝜃) 2, 3 MP
nition of an axiomatic derivation, or even a schema 5. [𝜑 → (𝜓 → 𝜃)] → [(𝜑 → 𝜓) → (𝜑 → 𝜃)]
for one, but it would if we replaced line 6 with the CMPAx
steps from our previous derivation schema, with 6. (𝜑 → 𝜓) → (𝜑 → 𝜃) 4, 5 MP
¬𝜑 everywhere for 𝜑. 7. 𝜑 → 𝜃 1, 6 MP

𝜑 → (𝜓 → 𝜃) ⊢ 𝜓 → (𝜑 → 𝜃) (“Int”)
Result: If Γ ⊢ 𝜑 and 𝜑 ⊢ 𝜓, then Γ ⊢ 𝜓.
1. 𝜑 → (𝜓 → 𝜃) Premise
2. [𝜑 → (𝜓 → 𝜃)] → [(𝜑 → 𝜓) → (𝜑 → 𝜃)]
CMPAx
3. (𝜑 → 𝜓) → (𝜑 → 𝜃) 1, 2 MP
Proof: After the derivation of 𝜑 from Γ, concatenate
4. 𝜓 → (𝜑 → 𝜓) TCAx
the remaining steps of of the derivation 𝜑 ⊢ 𝜓,
5. 𝜓 → (𝜑 → 𝜃) 3, 4 HS
using the 𝜑 from final line of the first derivation
instead of taking 𝜑 as a premise. The result will be Homework 9
a derivation of 𝜓 from Γ. ❅
Give proof schemata for the derived rules for dis-
junctive syllogism (“DS”). Hint: use (CasesAx),
(FAAx) and (Ref→).

17
(1) 𝜑 ∨ 𝜓, ¬𝜑 ⊢ 𝜓
(𝜑 ∨ 𝜓) ∨ 𝜃 ⊢ 𝜑 ∨ (𝜓 ∨ 𝜃) (Assoc)
(2) 𝜑 ∨ 𝜓, ¬𝜓 ⊢ 𝜑
(𝜑 ∧ 𝜓) ∧ 𝜃 ⊢ 𝜑 ∧ (𝜓 ∧ 𝜃) (Assoc)
We’ve already used this simple one, but we might
𝜑 ∧ (𝜓 ∨ 𝜃) ⊢ (𝜑 ∧ 𝜓) ∨ (𝜑 ∧ 𝜃) (Dist)
as well name it: “TC” for “true consequent”.
𝜑 ∨ (𝜓 ∧ 𝜃) ⊢ (𝜑 ∨ 𝜓) ∧ (𝜑 ∨ 𝜃) (Dist)
𝜑⊢𝜓→𝜑 (TC) 𝜑 → 𝜃, 𝜓 → 𝜉, 𝜑 ∨ 𝜓 ⊢ 𝜃 ∨ 𝜉 (CD)
1. 𝜑 Premise 𝜑 → 𝜓, ¬𝜑 → 𝜓 ⊢ 𝜓 (Inev)
2. 𝜑 → (𝜓 → 𝜑) (TCAx) 𝜑 → 𝜓, 𝜃 → 𝜓 ⊢ (𝜑 ∨ 𝜃) → 𝜓 (Cases)
3. 𝜓 → 𝜑 1, 2 MP 𝜑, ¬𝜑 ⊢ ⊥ (Contra)
⊥⊢𝜑 (Explode)
𝜑 → (𝜓 → 𝜃) ⊢ (𝜑 → 𝜓) → (𝜑 → 𝜃) etc.
𝜑 → (𝜓 → 𝜃), 𝜑 → 𝜓 ⊢ 𝜑 → 𝜃 (CMP) In short, we “recapture” all the rules one is used to
having in a natural deduction system.
(Direct from CMPAx and MP once or twice.)
(𝜑 ∧ 𝜓) → 𝜃 ⊢ 𝜑 → (𝜓 → 𝜃) (Exp) Reducing the axiom schemata
1. (𝜑 ∧ 𝜓) → 𝜃 Premise
The book does not add axioms for the biconditional
2. 𝜓 → [(𝜑 ∧ 𝜓) → 𝜃] 1 TC
operator ↔. If we added axiom schemata, most
3. [𝜓 → (𝜑 ∧ 𝜓)] → (𝜓 → 𝜃) 2 CMP
likely we would need:
4. 𝜑 → {[𝜓 → (𝜑 ∧ 𝜓)] → (𝜓 → 𝜃)} 3 TC
5. 𝜑 → [𝜓 → (𝜑 ∧ 𝜓)] ConjAx (𝜑 ↔ 𝜓) → (𝜑 → 𝜓) (EquivLR)
6. 𝜑 → (𝜓 → 𝜃) 4, 5 CMP
(𝜑 ↔ 𝜓) → (𝜓 → 𝜑) (EquivRL)
𝜑 → (𝜓 → 𝜃) ⊢ (𝜑 ∧ 𝜓) → 𝜃 (Imp) (𝜑 → 𝜓) → ((𝜓 → 𝜑) → (𝜑 ↔ 𝜓)) (EquivIn)
1. 𝜑 → (𝜓 → 𝜃) Premise
2. (𝜑 ∧ 𝜓) → 𝜑 SimpLAx Instead of this, however, we could simply eliminate
3. (𝜑 ∧ 𝜓) → (𝜓 → 𝜃) 1, 2 HS ↔ as a primitive part of the notation, and use it
4. (𝜑 ∧ 𝜓) → 𝜓 SimpRAx only as an abbreviation:
5. (𝜑 ∧ 𝜓) → 𝜃 3, 4 CMP
𝜑 ↔ 𝜓 is shorthand for (𝜑 → 𝜓) ∧ (𝜓 → 𝜑)
Other consequences (some easy, some not):
In that case, instances of (EquivLR), (EquivRL), and
𝜑∧𝜓 ⊢𝜑 (Simp) (EquivIn) are just notational variants of instances
𝜑∧𝜓 ⊢𝜓 (Simp) of (SimpLAx), (SimpRAx) and (ConjAx).
𝜑, 𝜓 ⊢ 𝜑 ∧ 𝜓 (Conj) In fact, we could go further; we could use only a
minimum number of primitive signs, say {→, ¬ } or
𝜑⊢𝜑∨𝜓 (Add) even just { | }, and eliminate the axioms governing
𝜑⊢𝜓∨𝜑 (Add) the defined ones. For example, if we took 𝜓 ∨ 𝜑
¬𝜑 ⊢ 𝜑 → 𝜓 (FA) to abbreviate ¬𝜓 → 𝜑, then we wouldn’t need an
¬¬𝜑 ⊢ 𝜑 (DNE) axiom schema for 𝜑 → (𝜓 ∨ 𝜑), as this would just be
an instance of (TCAx), and 𝜓 ∨ 𝜑, ¬𝜓 ⊢ 𝜑 would just
𝜑∨𝜑⊢𝜑 (Taut)
be modus ponens, etc. With only {→, ¬ } as primitive,
𝜑⊢𝜑∧𝜑 (Taut) we can dramatically reduce the axiomatic schemata.
𝜑∧𝜓 ⊢𝜓∧𝜑 (Com) E.g., here’s Frege’s own original axiomatization:
𝜑∨𝜓 ⊢𝜓∨𝜑 (Com)

18
 Where 𝑡 is a closed term, any instance of the
following schemata are considered axioms:
𝜑 → (𝜓 → 𝜑)
(UIAx)
A
[𝜑 → (𝜓 → 𝜃)] → [(𝜑 → 𝜓) → (𝜑 → 𝜃)] 𝑥𝜑 → 𝜑[𝑡/𝑥]
(EGAx)
E
(𝜑 → 𝜓) → (¬𝜓 → ¬𝜑) 𝜑[𝑡/𝑥] → 𝑥𝜑
𝜑 → ¬¬𝜑
Instances of (UIAx) include, e.g., “ x Fx → Fa” and
A
¬¬𝜑 → 𝜑 x(Fx → Gx) → (Fb → Gb). With MP it allows us
A
to “instantiate” variables to constants or function
It’s possible to replace the last three with only: terms. Instances of (EGAx) include “Fa → x Fx”
E
and the like. Hence we get the derived rules (Hard-
(¬𝜑 → ¬𝜓) → [(¬𝜑 → 𝜓) → 𝜑]
egree’s O and I):
A E

And we can go further still, though there are dimin-

(UI)
A
𝑥𝜑(𝑥) ⊢ 𝜑(𝑡)
ishing returns. In 1917, Jean Nicod discovered a
(EG)
E
system for propositional logic using only the Shef- 𝜑(𝑡) ⊢ 𝑥𝜑(𝑥)
fer stroke | as primitive, with the single schema:
We also need rules for “using” existentials and “get-
ting” universals. Different versions of axiomatic
(𝜑 | (𝜓 | 𝜃)) | ((𝜉 | (𝜉 | 𝜉)) |
systems have historically done this different ways.
((𝜁 | 𝜓) | ((𝜑 | 𝜁 ) | (𝜑 | 𝜁 )))) The “old way” involves allowing free variables in
formulas in proofs, and allowing one to “generalize”
The only inference rule is: From 𝜑 | (𝜃 | 𝜓) and 𝜑
on open formulas to arrive at a version with an
infer 𝜓. This suffices for all of propositional logic.
initial quantifier. This way has ceded ground to a
There is something nice about austere systems,
“new way” that uses “dummy constants”, which is
but they can require very alien and complicated
what the Open Logic Text does. This involves the
derivations for what seem like obvious results for
quantifier rules (QR):
“non-primitive” but familiar operators.
I shall assume we have (EquivLR), (EquivRL), QR (two forms):
(EquivIn) and their easy derived rules:
• If constant 𝑎 does not occur in a premise or
𝜑↔𝜓⊢𝜑→𝜓 (BE) 𝜓 then from 𝜓 → 𝜑(𝑎) one may infer 𝜓 →
A
𝜑↔𝜓⊢𝜓→𝜑 (BE) 𝑥𝜑(𝑥).
• If constant 𝑎 does not occur in a premise or
𝜑 → 𝜓, 𝜓 → 𝜑 ⊢ 𝜑 ↔ 𝜓 (BI) 𝜓 then from 𝜑(𝑎) → 𝜓 infer 𝑥𝜑(𝑥) → 𝜓.
E
𝜑 ↔ 𝜓, 𝜑 ⊢ 𝜓 (BMP)
𝜑 ↔ 𝜓, 𝜓 ⊢ 𝜑 (BMP)
Homework 10
𝜑 ↔ 𝜓, ¬𝜑 ⊢ ¬𝜓 (BMT)
𝜑 ↔ 𝜓, ¬𝜓 ⊢ ¬𝜑 (BMT) (a) Show that ⊢ 𝑥(𝜑(𝑥) → ⊥) → ¬ 𝑥𝜑(𝑥).
A E
(b) Suppose that instead of taking the existential
It’s not really important whether we consider these
quantifier 𝑥 as primitive, we simply used the
E
as results of new axioms or just consequences of
notation to abbreviate ¬ 𝑥¬. Prove that we
A
introducing non-primitive notation.
would then not need to take (EGAx) as an ax-
iom schema, because we could prove its in-
Quantifier axioms and rules stances from (UIAx), and that we wouldn’t
need the second form of QR, as we could re-
For predicate logic, we also need axioms/rules for
cover it from the first form.
the quantifiers.

19
(For both, you may use derived rules from proposi- has the property that Γ ⊢ 𝜃 → 𝜓𝑖 . As inductive
tional logic only.) hypothesis, we are allowed to assume that all mem-
bers prior to an arbitrary 𝜓𝑖 have this property. We
The first form of QR allows us to do the work of
universal generalization (or Hardegree’s “univer- must show that given this assumption, 𝜓𝑖 has it too.
sal derivation (UD)”), and the second the work of There are six cases to consider:
existential instantiation (Hardegree’s “ O”). The (1) 𝜓𝑖 is an axiom; then Γ ⊢ 𝜓𝑖 and by TC, Γ ⊢ 𝜃 →
E
𝜓𝑖
latter must wait till after we tackle the “deduction
(2) 𝜓𝑖 is a member of Γ; then Γ ⊢ 𝜓𝑖 and by TC,
theorem”, but for the former, note the following:
Γ ⊢ 𝜃 → 𝜓𝑖 .
(3) 𝜓𝑖 is a member of {𝜃}, i.e., it is 𝜃; in that case,
Γ ⊢ 𝜃 → 𝜓𝑖 by Ref→.
Result (UG): If one obtains a line of the form (4) 𝜓𝑖 follows from previous members of the series,
𝜑(𝑎), where 𝑎 is a constant that does not occur in 𝜓𝑗 and 𝜓𝑗 → 𝜓𝑖 by MP. By the inductive hypoth-
A
a premise, then one can derive 𝑥𝜑(𝑥). esis, Γ ⊢ 𝜃 → 𝜓𝑗 and Γ ⊢ 𝜃 → (𝜓𝑗 → 𝜓𝑖 ). By
CMP, Γ ⊢ 𝜃 → 𝜓𝑖 .
(5) 𝜓𝑖 follows from a previous member of the series
by the first form of QR. Hence 𝜓𝑖 takes the form
Proof: (1) Suppose one obtains 𝜑(𝑎) where 𝑎 is not 𝜉 → 𝑥𝜁 (𝑥), and there is a previous member of
A
in a premise. the 𝜓-series 𝜉 → 𝜁 (𝑎), where 𝑎 does not occur
(2) By TC and MP, we get ⊤ → 𝜑(𝑎). in Γ ∪ {𝜃}, nor in 𝜉. By the inductive hypothesis
(3) By QR, we get ⊤ → 𝑥𝜑(𝑥).
A
Γ ⊢ 𝜃 → (𝜉 → 𝜁 (𝑎)). By Imp, Γ ⊢ (𝜃 ∧ 𝜉) →
(4) By VerumAx and MP we get 𝑥𝜑(𝑥).
A
❅ 𝜁 (𝑎). Because 𝑎 does not occur in any member
of Γ ∪ {𝜃}, it doesn’t occur in 𝜃; hence it also does
From now on, we’ll just call the above steps “UG”, not occur in 𝜃 ∧ 𝜉. Hence, by QR, Γ ⊢ (𝜃 ∧ 𝜉) →
and use it as if it were a derived rule. Note that 𝜑(𝑎) A
𝑥𝜁 (𝑥). By Exp, Γ ⊢ 𝜃 → (𝜉 → 𝑥𝜁 (𝑥)), i.e.,
A
⊢ 𝑥𝜑(𝑥) is generally not the case, however, as 𝑎
A
Γ ⊢ 𝜃 → 𝜓𝑖 .
cannot occur in a premise. (6) 𝜓𝑖 follows a previous member of the series by
the second form of QR. Hence, 𝜓𝑖 takes the form
𝑥𝜉(𝑥) → 𝜁 , and there is a previous member of
E
The deduction theorem
the 𝜓-series 𝜉(𝑎) → 𝜁 , where 𝑎 does not occur
The main thing now keeping this method from be-
in 𝜁 nor in Γ ∪ {𝜃}. By the inductive hypothesis
ing as easy to use as a natural deduction system
Γ ⊢ 𝜃 → (𝜉(𝑎) → 𝜁 ). By Int, Γ ⊢ 𝜉(𝑎) →
is the lack of conditional and indirect (or “reduc-
(𝜃 → 𝜁 ). 𝑎 does not occur in either 𝜃 or 𝜁 , so
tio”) proofs. We’ll prove that these techniques are
by QR, Γ ⊢ 𝑥𝜉(𝑥) → (𝜃 → 𝜁 ). By Int again,
E
not needed, as any apparent derivation using them
Γ ⊢ 𝜃 → ( 𝑥𝜉(𝑥) → 𝜁 ), i.e., Γ ⊢ 𝜃 → 𝜓𝑖 .
E
could be transformed into one that doesn’t.
This covers all possible cases, so by proof induction,
it holds for every member of the 𝜓-series, including
the last one, which is 𝜑. Therefore, Γ ⊢ 𝜃 → 𝜑. ❅
Result (DT): If Γ ∪ {𝜃} ⊢ 𝜑 then Γ ⊢ 𝜃 → 𝜑.
I am very fond of this proof. Essentially, it provides
a recipe for transforming a proof that takes an addi-
tional formula as a hypothesis and turning it into
Proof: Assume that Γ ∪ {𝜃} ⊢ 𝜑 and that the deriva- a proof of the conditional from that hypothesis to
tion of 𝜓 from Γ ∪ {𝜃} is 𝜓1 , … , 𝜓𝑛 . (We assume the the result the original proof reaches with its help.
derivation is written out in full, without derived Suppose we want to show:
rules or cited theorems.) We shall use proof induc- A A A
tion to show that every member of this series 𝜓𝑖 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥𝜑(𝑥) → 𝑥𝜓(𝑥)

20
Let us consider the unabbreviated proof of: 6a. 𝑥𝜑(𝑥) → 𝜑(𝑎) 2a, 5b CMP
A
7a. 𝑥𝜑(𝑥) → 𝜓(𝑎) 4a, 6a CMP
A
A A A
𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝑥𝜑(𝑥) ⊢ 𝑥𝜓(𝑥)
Obviously, this is not the most efficient way to get
1. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) Premise these results. 6a is the same as a line we already
A
2. 𝑥𝜑(𝑥) “Hyp” have, and an axiom to boot! Continuing:
A
3. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) → (𝜑(𝑎) → 𝜓(𝑎)) UIAx 8a. 𝜓(𝑎) → (⊤ → 𝜓(𝑎)) TCAx
A
4. 𝜑(𝑎) → 𝜓(𝑎) 1, 3 MP 8b. 𝑥𝜑(𝑥) → [𝜓(𝑎) → (⊤ → 𝜓(𝑎))] 8a TC
A
5. 𝑥𝜑(𝑥) → 𝜑(𝑎) UIAx 9a. 𝑥𝜑(𝑥) → (⊤ → 𝜓(𝑎)) 7a, 8b CMP
A A
6. 𝜑(𝑎) 2, 5 MP
Line 10 falls into case (5), which is new.
7. 𝜓(𝑎) 4, 6 MP
TCAx 10a. ( 𝑥𝜑(𝑥) ∧ ⊤) → 𝜓(𝑎) 9a Imp
A
8. 𝜓(𝑎) → (⊤ → 𝜓(𝑎))
7, 8 MP 10b. ( 𝑥𝜑(𝑥) ∧ ⊤) → 𝑥𝜓(𝑥) 10a QR
A A
9. ⊤ → 𝜓(𝑎)
9 QR 10c. 𝑥𝜑(𝑥) → (⊤ → 𝑥𝜓(𝑥)) 10b Exp
A A
10. ⊤ → 𝑥𝜓(𝑥)
A
11. ⊤ VerumAx And to finish:
12. 𝑥𝜓(𝑥) 10, 11 MP
A
Line 2 here is a premise, but I’ve labeled it “Hyp” to 11a. ⊤ VerumAx
11b. 11a TC
A
indicate that we can think of it as an extra assump- 𝑥𝜑(𝑥) → ⊤
tion, as if we wanted to do a “conditional” proof for 12a. 𝑥𝜑(𝑥) → 𝑥𝜓(𝑥) 10c, 11b CMP
A A

𝑥𝜑(𝑥) → 𝑥𝜓(𝑥). The recipe provided by the de- Notice the only premise used in the “transformed”
A A
duction theorem turns it into a proof of that. Here proof was at 1a, so this establishes:
Γ is { 𝑥(𝜑(𝑥) → 𝜓(𝑥))} and the 𝜃 is 𝑥𝜑(𝑥). We ap-
A A
ply steps to the original proofs to get 𝑥𝜑(𝑥) → … ,
A A A A
𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥𝜑(𝑥) → 𝑥𝜓(𝑥)
filling in the dots with the original line.
We could even go further and transform the trans-
Start with line 1, which is case (2). formed proof (though this would require expanding
1a. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) Premise all the derived rules first) into one for:
A
1b. 𝑥𝜑(𝑥) → [ 𝑥(𝜑(𝑥) → 𝜓(𝑥))] 1a TC
A A
A A A
⊢ 𝑥(𝜑(𝑥) → 𝜓(𝑥)) → ( 𝑥𝜑(𝑥) → 𝑥𝜓(𝑥))
Notice that 1b here is the original line 1 with
𝑥𝜑(𝑥) → … before it. We want that for all steps.
A
Original line 2 is case (3): Homework 11
2a. 𝑥𝜑(𝑥) → 𝑥𝜑(𝑥) Ref→ On page 17, there is a proof for the derived rule
A A
of modus tollens (MT). Use the recipe provided by
Then case (1): the proof of the deduction theorem to transform
3a. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) → (𝜑(𝑎) → 𝜓(𝑎)) UIAx it, step by step, into a proof of transposition in the
A
3b. 𝑥𝜑(𝑥) → [ 𝑥(𝜑(𝑥) → 𝜓(𝑥)) → form:
A A
(𝜑(𝑎) → 𝜓(𝑎))] 3a TC 𝜑 → 𝜓 ⊢ ¬𝜓 → ¬𝜑 (Trans)
For case (4) we need to rely the inductive hypothe- We have proven that conditional proof is not
sis. needed. Anything you could prove with conditional
proof we can prove without it by transforming the
4a. 𝑥𝜑(𝑥) → (𝜑(𝑎) → 𝜓(𝑎)) 1b, 3b CMP
A
alleged conditional proof into a direct one.
Another case (1): From now on (—well, at least after the above
5a. 𝑥𝜑(𝑥) → 𝜑(𝑎) UIAx homework assignment—) we won’t actually bother
A
5b. 𝑥𝜑(𝑥) → ( 𝑥𝜑(𝑥) → 𝜑(𝑎)) 5a TC with the transformations. If we want to prove a
A A
conditional 𝜃 → 𝜑 from Γ, We’ll just do the proof
Then two more case (4)’s: of 𝜑 with an additional “Hyp”, 𝜃, and then “cite” the

21
deduction theorem (“DT”) as thereby establishing Proof: Assume Γ ∪ {𝜑} ⊢ ⊥. By the deduction theo-
that Γ ⊢ 𝜃 → 𝜑 is true as well. rem, Γ ⊢ 𝜑 → ⊥. By RedAx and MP, Γ ⊢ ¬𝜑. ❅
Sometimes it helps to write proofs using the
sign ⊢ to keep track of what premises/hypotheses From now on we allow ourselves to use the reductio
a line depends on, e.g.: technique; notice moreover that 𝜑, ¬𝜑 ⊢ ⊥ follows
easily from FAAx.
(CQ) Finally the deduction theorem demonstrates the
A E
⊢ 𝑥¬𝜑(𝑥) → ¬ 𝑥𝜑(𝑥)
relationship between the second form of QR and the
1. Hyp
A A
𝑥¬𝜑(𝑥) ⊢ 𝑥¬𝜑(𝑥) “natural deduction” rule that allows you to instanti-
2. 1 UI
A
𝑥¬𝜑(𝑥) ⊢ ¬𝜑(𝑎) ate an existential to a “new” or “unused” constant
3. 2 FA
A
𝑥¬𝜑(𝑥) ⊢ 𝜑(𝑎) → ⊥ (“EI” or “ O/ -Elim”). Note that 𝑥𝜑(𝑥) ⊢ 𝜑(𝑎) is
E E E
4. 3 QR
A E
𝑥¬𝜑(𝑥) ⊢ 𝑥𝜑(𝑥) → ⊥ not in general the case, but what is the case is this:
5. RedAx
E E
⊢ ( 𝑥𝜑(𝑥) → ⊥) → ¬ 𝑥𝜑(𝑥)
6. 4, 5 MP
A E
𝑥¬𝜑(𝑥) ⊢ ¬ 𝑥𝜑(𝑥)
7. 6 DT
A E
⊢ 𝑥¬𝜑(𝑥) → ¬ 𝑥𝜑(𝑥) E
Result (EI): If Γ ⊢ 𝑥𝜑(𝑥) and Γ ∪ {𝜑(𝑎)} ⊢ 𝜓,
It is important for QR and rules/methods based on where 𝑎 does not occur in Γ, or 𝜓, then Γ ⊢ 𝜓.
it, like UG, that a certain constant not appear in
the premises/hypotheses. This applies to those that
have not already been discharged. Consider:
Proof: Suppose Γ ⊢ 𝑥𝜑(𝑥) and Γ ∪ {𝜑(𝑎)} ⊢ 𝜓 and
E
A A
𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥(¬𝜓(𝑥) → ¬𝜑(𝑥)) that 𝑎 does not occur in Γ or 𝜓. By the deduction
theorem, Γ ⊢ 𝜑(𝑎) → 𝜓. By QR, Γ ⊢ 𝑥𝜑(𝑥) → 𝜓,
E
1. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥(𝜑(𝑥) → 𝜓(𝑥)) Premise and by MP, Γ ⊢ 𝜓.
A A
❅
2. ¬𝜓(𝑎) ⊢ ¬𝜓(𝑎) Hyp
3. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝜑(𝑎) → 𝜓(𝑎) 1 UI
A
4. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), ¬𝜓(𝑎) ⊢ ¬𝜑(𝑎) 2, 3 MT For example:
A
5. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ ¬𝜓(𝑎) → ¬𝜑(𝑎) 4 DT
A
A E E
𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝑥𝜑(𝑥) ⊢ 𝑥𝜓(𝑥)
6. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥(¬𝜓(𝑥) → ¬𝜑(𝑥)) 5 UG
A A

1. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥(𝜑(𝑥) → 𝜓(𝑥)) Premise

A A
Here we could not have applied UG at line 4
2. 𝑥𝜑(𝑥) ⊢ 𝑥𝜑(𝑥) Premise
E E
to get 𝑥¬𝜑(𝑥), because 𝑎 occurs in one of the
A
premises/hypotheses that line depends on. How- 3. 𝜑(𝑎) ⊢ 𝜑(𝑎) Hyp
4. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝜑(𝑎) → 𝜓(𝑎) 1 UI
A
ever, it is OK to apply it to line 5 to get line 6, be-
5. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝜑(𝑎) ⊢ 𝜓(𝑎) 3, 4 MP
A
cause we have “discharged” that hypotheses with
6. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝜑(𝑎) ⊢ 𝑥𝜓(𝑥) 5 EG
A E
DT. It is important to be careful about assump-
7. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝜑(𝑎) → 𝑥𝜓(𝑥) 6 DT
A E
tions/premises used for DT and QR/UG.
8. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥𝜑(𝑥) → 𝑥𝜓(𝑥) 7 QR
A E E
9. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝑥𝜑(𝑥) ⊢ 𝑥𝜓(𝑥) 2, 8 MP
A E E
Corollaries of the deduction theorem
Written up like this, we never actually use the ex-
The deduction theorem also underwrites the indi- istential line (2) until the end, and we simply take
rect proof or reductio-style proof technique. 𝜑(𝑎), naturally thought of as the “instance” of (2), as
a new hypothesis at line (3). It is harmless, though
not technically correct, to write this as follows:
Result (RAA): If Γ ∪ {𝜑} ⊢ ⊥, then Γ ⊢ ¬𝜑.
1. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝑥(𝜑(𝑥) → 𝜓(𝑥)) Premise
A A
2. 𝑥𝜑(𝑥) ⊢ 𝑥𝜑(𝑥) Premise
E E
3. 𝑥𝜑(𝑥) ⊢∗ 𝜑(𝑎) 2 EI
E
4. 𝑥(𝜑(𝑥) → 𝜓(𝑥)) ⊢ 𝜑(𝑎) → 𝜓(𝑎) 1 UI
A

22
5. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝑥𝜑(𝑥) ⊢∗ 𝜓(𝑎) 3, 4 MP 𝑥, and since M, 𝑠 ⊨ 𝑥𝜑(𝑥) it must be that
A E A
6. 𝑥(𝜑(𝑥) → 𝜓(𝑥)), 𝑥𝜑(𝑥) ⊢ 𝑥𝜓(𝑥) 5 EG M, 𝑠 ′ ⊨ 𝜑(𝑥). This is a contradiction, and there-
A E E
fore every instance of 𝑥𝜑(𝑥) → 𝜑(𝑡) must be
A
Here I have marked the turnstiles at lines 3 and 5
with asterisks, since the ⊢ relation doesn’t actually valid after all.
(3) Every instance of 𝜑(𝑡) → 𝑥𝜑(𝑥) is logically
E
hold here. However, once the “dummy constant”
disappears, the trick applies and the result does valid. Again suppose for reductio that some
formula of this form were not valid. Then
hold. (Even in your natural deduction systems, you
there is some interpretation M and variable
were not allowed to end a proof with the dummy
assignment 𝑠 where M, 𝑠 ⊭ 𝜑(𝑡) → 𝑥𝜑(𝑥). By
E
constant as a part of the conclusion.) Hence, I will
allow you from here to pretend as if there is a rule the definition of satisfaction, M, 𝑠 ⊨ 𝜑(𝑡) and
M, 𝑠 ⊭ 𝑥𝜑(𝑥). Let 𝑠 ′ be the variable assign-
E
“EI” (or “ -Elim/ -Out”), since derivations using
E E
ment just like 𝑠 except ValM 𝑠 ′ (𝑥) = 𝑡 . Since
M
them can always be transformed into one using the
M, 𝑠 ⊨ 𝜑(𝑡) it follows that M, 𝑠 ⊨ 𝜑(𝑥). But
′
“instance” as a hypothesis, with QR, instead.
then there is an assignment different from 𝑠
By this point we have probably recaptured
at most its value for 𝑥 that satisfies 𝜑(𝑥), and
pretty much all the logical rules you’re used to us-
hence M, 𝑠 ⊨ 𝑥𝜑(𝑥), which is a contradiction.
E
ing, or could easily get the rest. But we want some-
Hence every instance of 𝜑(𝑡) → 𝑥𝜑(𝑥) is valid
E
thing stronger: to prove that every logically valid
after all.
argument has a corresponding axiomatic deriva-
(4) Cases (1)–(3) exhaust the axioms in question;
tion. We also want the reverse. This brings us to
hence, all logical axioms are logically valid. ❅
our main metatheoretic results.

4. Bridging ⊢ and ⊨
Result (Soundness): If Γ ⊢ 𝜑 then Γ ⊨ 𝜑.
Soundness and consistency

Result: The axioms of propositional and quan- Proof: Assume Γ ⊢ 𝜑. We need to show that every
tifier logic are all logically valid. interpretation that is a model of Γ also makes 𝜑
true.
Let the derivation of 𝜑 from Γ be the series
𝜓1 , … , 𝜓𝑛 . We will show by proof induction that
every 𝜓𝑖 in this series is such that every interpreta-
Proof: (1) The propositional axioms are all truth-
tion that is a model of Γ is also a model of it. We are
table tautologies; we here skip over the actual
entitled to assume that we have already shown this
tables. The semantic rules track the truth table
to be true for every member of the 𝜓-series prior
rules, so these are all valid.
to 𝜓𝑖 .
(2) Every instance of 𝑥𝜑(𝑥) → 𝜑(𝑡), where 𝑡 is
A
There are five cases to consider.
a closed term is logically valid. Suppose for
(1) 𝜓𝑖 is an axiom. By our earlier result, 𝜓𝑖 is logi-
reductio that this is not the case for some for-
cally valid, and hence true in every interpreta-
mula of this form. Then there is an interpreta-
tion, including all those making the members
tion M and variable assignment 𝑠 such that
of Γ true.
M, 𝑠 ⊭ 𝑥𝜑(𝑥) → 𝜑(𝑡). But then M, 𝑠 ⊨
A
(2) 𝜓𝑖 is a member of Γ; it is therefore made true
𝑥𝜑(𝑥) and M, 𝑠 ⊭ 𝜑(𝑡). Let 𝑠 ′ be the variable
A
by all interpretations making every member of
assignment just like 𝑠 except ValM 𝑠 ′ (𝑥) = 𝑡 .
M
Γ true, since it is one.
Since M, 𝑠 ⊭ 𝜑(𝑡) it must be that M, 𝑠 ′ ⊭ 𝜑(𝑥).
(3) 𝜓𝑖 follows from previous members of the series
But 𝑠 ′ differs from 𝑠 by at most its value for
by MP. Then there are previous members of

23
 the series 𝜓𝑗 and 𝜓𝑗 → 𝜓𝑖 . By the inductive It is a model of Γ as well, since Γ does not
hypothesis, every interpretation making the contain 𝑎. So it makes 𝜁 (𝑎) → 𝜉 true, and
members of Γ true make these true. Let M be M′ , 𝑠 ⊨ 𝜁 (𝑎) → 𝜉. But 𝜉 does not contain 𝑎, and
such an interpretation. Since it makes 𝜓𝑗 and since M, 𝑠 ⊭ 𝜉 it must also be that M′ , 𝑠 ⊭ 𝜉.
𝜓𝑗 → 𝜓𝑖 true, it must also make 𝜓𝑖 true by the Then it must be that M′ , 𝑠 ⊭ 𝜁 (𝑎). But 𝜁 (𝑎)
semantics of →. The same is true for any model is just like 𝜁 (𝑥) except containing 𝑎 where
of Γ. 𝜁 (𝑥) contains 𝑥 free, and since M, 𝑠 ′ ⊨ 𝜁 (𝑥)
(4) 𝜓𝑖 follows from a previous member of the series it follows that M′ , 𝑠 ⊨ 𝜁 (𝑎), which contradicts
by the first form of QR. Then it takes the form our earlier result. Hence, the assumption that
𝜉 → 𝑥𝜁 (𝑥) and there is a previous member 𝑥𝜁 (𝑥) → 𝜉 is not true in some model M of
A E
𝜉 → 𝜁 (𝑎), where 𝑎 does not occur in Γ. By the Γ must be mistaken, and 𝑥𝜁 (𝑥) → 𝜉, i.e., 𝜓𝑖 ,
E
inductive hypothesis, every model of Γ makes must be true in all such models.
𝜉 → 𝜁 (𝑎) true. Suppose for reductio that some These five cases exhaust the possible steps of the
model M of Γ does not make 𝜉 → 𝑥𝜁 (𝑥) true. derivation, and so it is true for all steps 𝜓𝑖 in the
A
Then for some 𝑠, M, 𝑠 ⊨ 𝜉 and M, 𝑠 ⊭ 𝑥𝜁 (𝑥). 𝜓-series that Γ ⊨ 𝜓𝑖 . 𝜑 is the last member of this
A
This means that for some 𝑠 ′ differing from 𝑠 series, and so Γ ⊨ 𝜑. ❅
only for its value for 𝑥 is such that M, 𝑠 ′ ⊭ 𝜁 (𝑥).
Now consider the interpretation M′ differing
from M only in that 𝑎M = ValM 𝑠 ′ (𝑥). M is a
′
Corollary: All theorems of pure logic are logi-
model of Γ, and so M must be as well, since
′
cally valid, i.e., if ⊢ 𝜑 then ⊨ 𝜑.
𝑎 does not occur in Γ. Since |M| = |M′ |, the
variable assignment is also an assignment for
M′ . Since M′ is a model of Γ, by the induc-
Proof: If ⊢ 𝜑 then ∅ ⊢ 𝜑, and so by the above
tive hypothesis, it makes 𝜉 → 𝜁 (𝑎) true, and
∅ ⊨ 𝜑. But every interpretation is a model of ∅
therefore M′ , 𝑠 ⊨ 𝜉 → 𝜁 (𝑎). But 𝜉 does not con-
trivially, and so every interpretation makes 𝜑 true,
tain 𝑎, and so since M, 𝑠 ⊨ 𝜉 it is also true that
i.e., ⊨ 𝜑. ❅
M′ , 𝑠 ⊨ 𝜉 and hence M′ , 𝑠 ⊨ 𝜁 (𝑎). But since 𝑠 ′
differs from 𝑠 only in assigning to 𝑥 the same Definition: A set of sentences Γ is inconsistent iff
thing as M′ generally assigns to 𝑎, and 𝜁 (𝑎) Γ ⊢ ⊥.
differs from 𝜁 (𝑥) by containing 𝑎 where 𝜁 (𝑥)
contains 𝑥 free, and M, 𝑠 ′ ⊭ 𝜁 (𝑥) it follows that
M′ , 𝑠 ⊭ 𝜁 (𝑎). This is a contradiction. Hence, Corollary (Consistency): (a) If Γ is inconsis-
M must make 𝜉 → 𝑥𝜁 (𝑥) (i.e., 𝜓𝑖 ) true after
A
tent, then Γ is unsatisfiable, and (b) ⊬ ⊥.
all, and the same follows for any other model
of Γ. Hence 𝜓𝑖 is true for every model of Γ.
(5) 𝜓𝑖 follows from a previous member of the se-
ries by the second form of QR. 𝜓𝑖 takes the form Proof: Suppose Γ ⊢ ⊥; it follows that Γ ⊨ ⊥, and
E
𝑥𝜁 (𝑥) → 𝜉, and there is a previous wff in the then every model of Γ makes ⊥ true. If Γ were
𝜓-series of the form 𝜁 (𝑎) → 𝜉, where 𝑎 is in satisfiable, some interpretation would make ⊥ true,
neither Γ nor 𝜉. By the inductive hypothesis, which is impossible. Similarly, if ⊢ ⊥, then ⊥ would
𝜁 (𝑎) → 𝜉 is true in every model of Γ. Suppose be logically valid, and true in every interpretation,
for reductio that 𝑥𝜁 (𝑥) → 𝜉 is not true in some
E which is impossible. ❅
model M of Γ. Then there is some 𝑠 such that
M, 𝑠 ⊨ 𝑥𝜁 (𝑥) and M, 𝑠 ⊭ 𝜉. There is some
E Homework 12
𝑠 ′ differing from 𝑠 only in what it assigns to 𝑥 Prove (a) Γ ⊢ ¬𝜑 iff Γ ∪ {𝜑} is inconsistent; and (b)
such that M, 𝑠 ′ ⊨ 𝜁 (𝑥). Again consider inter- Γ ∪ {𝜑} and Γ ∪ {¬𝜑} are both inconsistent iff Γ is
pretation M′ just like M except 𝑎M = ValM
′
𝑠 ′ (𝑥). inconsistent.

24
Lemmas for completeness • Each function letter is assigned ⟨3, 𝑛, 𝑖, 𝑙⟩,
where 𝑛 is the number of its superscript, 𝑖
The opposite direction from soundness is complete-
the number of its subscript (0 if none), and 𝑙
ness; this is the more difficult direction, and we’ll
is 1–7 depending on the letter “f”–“l”.
need some definitions and lemmas first. Complete-
• Each predicate is assigned ⟨4, 𝑛, 𝑖, 𝑙⟩, where 𝑛
ness for first-order predicate logic was first shown
is the number of its superscript, 𝑖 the num-
by Kurt Gödel in 1930, but the strategy we’re using
ber of its subscript (0 if none), and 𝑙 is 1–20
derives from Leon Henkin.
depending on the letter “A”–“T”.
Definition: If ⟨𝑘1 , … , 𝑘𝑛 ⟩ is an 𝑛-tuple of natural (2) Each simple sign then has a “symbol code”, i.e.,
numbers, its code is the product of the first 𝑛 primes the code for the 𝑛-tuple we have assigned it.
raised respectively to the powers of (𝑘1 + 1), … , (𝑘𝑛 + E.g., the “symbol code” for the predicate A1 is
1). 25 ⋅ 32 ⋅ 51 ⋅ 72 .
(3) For each string of simple symbols, we assign
Example: The code for ⟨4, 10, 3, 0⟩ is 2 ⋅ 3 ⋅ 5 ⋅ 7 ,
5 11 4 1 it a unique number, called its Gödel number,
or 24,800,580,000. which is the code for the sequence of symbol
codes for the symbols making it up in order.
The reason we add one to the exponents is to ensure E.g., for the wff “A1 (a)” the Gödel number is
2(2 ⋅3 ⋅5 ⋅7 )+1 ⋅3(2 ⋅3 )+1 ⋅5(2 ⋅3 ⋅5 )+1 ⋅7(2 ⋅3 )+1 , which
5 2 1 2 1 12 3 1 2 1 13
different results when the sequence has 0’s at the
end, since, e.g., 24 ⋅ 310 ⋅ 53 ⋅ 70 = 24 ⋅ 310 ⋅ 53 but is some stupidly huge number for such a simple
2 ⋅ 3 ⋅ 5 ⋅ 7 ≠ 2 ⋅ 3 ⋅ 5 . No two sequences
5 11 4 1 5 11 4
sentence.
will have the same code this way, as numbers with (4) For each natural number 𝑛, there is an 𝑛th low-
different prime factorizations are always different. est number that is the Gödel number of a sen-
tence of . The function mapping each 𝑛 to the
sentence that is the one with the 𝑛th lowest is
Result: For any first-order language , the set a 1–1 function between natural numbers and
of its sentences is denumerable, i.e., we can gen- sentences of . Hence, the set of sentences of
erate a one–one correspondene between the set of  is denumerable. ❅
natural numbers and the set of its sentences.
We’ll be doing a lot more with Gödel numbers in
our next unit, unsurprisingly. In this unit, it plays
only a very minor role in our next result.

Proof: (1) We begin by assigning a different 𝑛-tuple
to each simple sign in the language as follows:
• The logical operators, constants and punctu-
ation marks on the list ⊥ ⊤ ¬ ∨ ∧ → ↔
EA
( ) , are respectively assigned ⟨0, 1⟩ through
⟨0, 13⟩.
• Each variable is assigned ⟨1, 𝑖, 𝑙⟩, where 𝑖 is
its subscript if it has one (0 otherwise) and 𝑙
is 1, 2, 3, 4, or 5, depending on whether the
variable uses letter “v”, “w”, “x”, “y” or “z”.
• Each constant is assigned ⟨2, 𝑖, 𝑙⟩ where 𝑖 is
its subscript if it has one (0 otherwise), and
𝑙 is 1, 2, 3, 4, or 5 depending on whether the
letter used is “a”, “b”, “c”, “d” or “e”.
Definition: A set of sentences Γ is complete iff for
every closed wff 𝜑, either 𝜑 ∈ Γ or ¬𝜑 ∈ Γ.
The above is not to be confused with “complete”
=
in the sense of capturing every semantically valid
sentence.
Definition: A set of sentences Γ is universal iff
for every wff 𝜑(𝑥) that contains at most 𝑥 free, if it
is the case for all closed terms 𝑡 that 𝜑(𝑡) ∈ Γ, then
A
𝑥𝜑(𝑥) ∈ Γ.
The book talks about “saturated” sets instead of
universal sets, which is a distinct but related notion.
My proof is subtly different, and somewhat more
streamlined than theirs, in the interest of saving
time.

25

Result (Lindenbaum Extension Lemma):
If Γ is a consistent set of sentences, then there
is a set of sentences Δ such that: (a) Γ ⊆ Δ, (b)
Δ is consistent, (c) Δ is complete, and (d) Δ is
universal.
Proof: (1) Assume that Γ is a consistent set of sen-
tences.
(2) We need an infinite list of constants that do
not occur in Γ, adding new constants to the
language if need be. I will use e, e1 , … , e𝑛 , … .
(You will note that “e” was not included in the
definition of a constant on p. 7.)
(3) By the denumerability of the set of sentences
of the language, we can arrange them in an
infinite sequence:
𝜑1 , 𝜑2 , 𝜑3 , … , etc.
Making use of this sequence, let us recursively
define an infinite sequence of sets:
Γ0 , Γ1 , Γ2 , … , etc.
As follows:
a) Let Γ0 = Γ.
b) We define Γ𝑛+1 in terms of Γ𝑛 in one of the
following three ways:
(i) if Γ𝑛 ∪ {𝜑𝑛+1 } is consistent, then let
Γ𝑛+1 = Γ𝑛 ∪ {𝜑𝑛+1 };
(ii) if Γ𝑛 ∪ {𝜑𝑛+1 } is inconsistent and 𝜑𝑛+1
does not take the form 𝑥𝜓(𝑥), then
A
let Γ𝑛+1 = Γ𝑛 ∪ {¬𝜑𝑛+1 };
(iii) if Γ𝑛 ∪ {𝜑𝑛+1 } is inconsistent and 𝜑𝑛+1
does take the form 𝑥𝜓(𝑥), then let
A
Γ𝑛+1 = Γ𝑛 ∪ {¬𝜑𝑛+1 } ∪ {¬𝜓(e𝑚 )}, where
“e𝑚 ” is the first e-constant that does
not occur in Γ𝑛 .
(4) Let Δ be the union of all of the members of the
Γ-sequence (i.e., Γ0 ∪ Γ1 ∪ Γ2 ∪ … etc.)
(5) Obviously, Γ ⊆ Δ. This establishes part (a) of
the consequent of the Lemma.
(6) Every member of the Γ-sequence is consistent.
We prove this by mathematical induction.
26
Base step: Γ0 is Γ, and it is consistent ex hy-
pothesi.
Induction step: Suppose Γ𝑛 is consistent. It
follows that Γ𝑛+1 is consistent by a proof by
cases:
Case (i): Γ𝑛+1 = Γ𝑛 ∪ {𝜑𝑛+1 } and Γ𝑛 ∪ {𝜑𝑛+1 } is
consistent, so Γ𝑛+1 is consistent.
Case (ii): Γ𝑛+1 = Γ𝑛 ∪ {¬𝜑𝑛+1 }, and Γ𝑛 ∪ {𝜑𝑛+1 }
is inconsistent.
• Hence Γ𝑛 ∪ {𝜑𝑛+1 } ⊢ ⊥.
• By the inductive hypothesis and
homework 12, Γ𝑛 ∪{𝜑𝑛+1 } and Γ𝑛 ∪
{¬𝜑𝑛+1 } are not both inconsistent.
• Hence, Γ ∪ {¬𝜑𝑛+1 }, i.e., Γ𝑛+1 , is
consistent.
Case (iii): Γ𝑛+1 = Γ𝑛 ∪ {¬𝜑𝑛+1 } ∪ {¬𝜓(e𝑚 )}, Γ𝑛 ∪
{𝜑𝑛+1 } is inconsistent and 𝜑𝑛+1 takes
the form 𝑥𝜓(𝑥).
A
• Γ𝑛 ∪ {𝜑𝑛+1 } ⊢ ⊥.
• By RAA, Γ𝑛 ⊢ ¬𝜑𝑛+1 .
• Suppose for reductio that Γ𝑛+1 is
inconsistent.
• So Γ𝑛 ∪ {¬𝜑𝑛+1 } ∪ {¬𝜓(e𝑚 )} ⊢ ⊥.
• By DT, Γ𝑛 ∪ {¬𝜓(e𝑚 )} ⊢ ¬𝜑𝑛+1 →
⊥.
• By MP, Γ𝑛 ∪ {¬𝜓(e𝑚 )} ⊢ ⊥.
• By RAA and DNE, Γ𝑛 ⊢ 𝜓(e𝑚 ).
• e𝑚 is not included in Γ𝑛 . Hence,
by UG, Γ𝑛 ⊢ 𝑥𝜓(𝑥), or in other
A
words, Γ𝑛 ⊢ 𝜑𝑛+1 .
• We already have Γ𝑛 ⊢ ¬𝜑𝑛+1 , so
Γ𝑛 ⊢ ⊥.
• So Γ𝑛 is inconsistent, which con-
tradicts the inductive hypothesis.
• Hence Γ𝑛+1 is consistent.
(7) It follows from (6) that Δ is consistent.
a) Note that the Γ-sequence is constantly ex-
panding: For all 𝑗 and 𝑘 such that 𝑗 < 𝑘,
Γ𝑗 ⊆ Γ𝑘 . Crudely, Δ can be thought of as
the upper limit of the expansion.
b) So every finite subset of Δ is a subset of
some Γ𝑖 for some suitably large 𝑖.
c) However, every derivation from Δ has only
a finite number of steps, and hence only
makes use of a finite subset of Δ.
d) If Δ ⊢ ⊥, for some suitably large 𝑖, it would
have to be that Γ𝑖 ⊢ ⊥.
 e) This is impossible because all the members Hence, the Lemma is established. ❅
of the Γ-sequence are consistent by (6).
f) Hence, Δ is consistent. The Lindenbaum extension lemma is another one
g) This establishes part (b) of the consequent of my favorites: any consistent story can be filled
of the Lemma. in to be a complete story of the way things are.
(8) Δ is obviously complete as well. We now show that if you have a complete story
a) All sentences are members of the sequence about the way things are, you can turn it into a
𝜑1 , 𝜑2 , … , etc. model.
b) For each 𝜑𝑖 , either it or its negation is a
Definition: An interpretation M is a denumer-
member of Γ𝑖 , and Γ𝑖 ⊆ Δ.
able interpretation iff its domain of quantification
c) So for all closed wffs 𝜑1 , 𝜑2 , … , etc., either
it or its negation is included in Δ. |M| is denumerable.
d) This establishes part (c) of the consequent
of the Lemma.
(9) Finally, Δ is also universal. Result: If Δ is a consistent and complete set of
a) We show this by reductio. Suppose other- sentences, then Δ is closed under entailment, i.e.,
wise, i.e., suppose that there is a wff 𝜓(𝑥) for every sentence 𝜑, if Δ ⊢ 𝜑 then 𝜑 ∈ Δ.
that contains at most 𝑥 free, such that for
all closed terms 𝑡, 𝜓(𝑡) ∈ Δ, but 𝑥𝜓(𝑥) ∉ Δ.
A
b) 𝑥𝜓(𝑥) is closed, so because Δ is complete,
A
it must be that ¬ 𝑥𝜓(𝑥) ∈ Δ.
A
Proof: Assume otherwise, i.e., Δ ⊢ 𝜑 but 𝜑 ∉ Δ.
c) Because 𝑥𝜓(𝑥) is closed, it also follows
A
Then, because Δ is complete, ¬𝜑 ∈ Δ, and so Δ ⊢
that 𝑥𝜓(𝑥) is a member of the 𝜑-sequence,
A
¬𝜑. But then Δ ⊢ ⊥, which is impossible, as Δ is
i.e., 𝑥𝜓(𝑥) is 𝜑𝑛+1 for some number 𝑛.
A
consistent. ❅
d) Obviously, however, since 𝑥𝜓(𝑥) ∉ Δ, it
A
follows that Γ𝑛+1 is not obtained from Γ𝑛
using case (i).
Result (The Term Model Lemma): If Δ is a
e) Nor was it obtained using case (ii), since
consistent, complete, and universal set of sen-
𝜑𝑛+1 is of the form 𝑥𝜓(𝑥).
A
tences, then there is at least one denumerable
f) This leaves case (iii), so Γ𝑛+1 is Γ𝑛 ∪{¬𝜑𝑛+1 }∪
interpretation that is a model for Δ.
{¬𝜓(e𝑚 )} for some 𝑚.
g) Hence, ¬𝜓(e𝑚 ) ∈ Γ𝑛+1 and so ¬𝜓(e𝑚 ) ∈ Δ.
h) But by our assumption, it holds for all
closed terms 𝑡 that 𝜓(𝑡) ∈ Δ.
i) The constant “e𝑚 ” is a closed term, so Proof: (1) Assume that Δ is a consistent, complete,
𝜓(e𝑚 ) ∈ Δ. and universal set of closed wffs. (It is thus
j) Hence, both Δ ⊢ ¬𝜓(e𝑚 ) and Δ ⊢ 𝜓(e𝑚 ), closed under entailment.) We can then de-
and so Δ ⊢ ⊥. scribe a denumerable interpretation M, called
k) However, this is impossible, because we the “term model for Δ” using the following pro-
have already shown Δ to be consistent. cedure.
l) Our supposition has been shown to be im- (2) Essentially, we’ll let all the closed terms of the
possible, hence Δ is universal. language stand for themselves. (Another possi-
m) This establishes part (d) of the consequent ble way of constructing a model would be to
of the Lemma. let each closed term stand for its Gödel num-
(10) By suitably defining Δ, we have shown each of ber. However, let us proceed using the former
parts (a)–(d) of the consequent of the Lemma method.)
on the basis of the assumption of its antecedent.

27
(3) Let the domain of quantification |M| of M be Case (a): 𝜑 takes the form ¬𝜓, where 𝜓 has
the set of closed terms of the language. Note one fewer connective than 𝜑.
that there are denumerably many closed terms, • By the inductive hypothesis, 𝜓 ∈ Δ
so M is a denumerable interpretation. iff M ⊨ 𝜓.
(4) For each constant 𝑐, let 𝑐 M be 𝑐 itself. • Because Δ is consistent, if 𝜑 ∈ Δ,
(5) For each function letter 𝑓 𝑛 , let 𝑓 𝑛M be that 𝑛- then 𝜓 ∉ Δ.
place operation on |M| which includes all or- • Because Δ is complete, if 𝜓 ∉ Δ,
dered pairs of the form then 𝜑 ∈ Δ.
• So 𝜓 ∉ Δ iff 𝜑 ∈ Δ.
⟨⟨𝑡1 , … , 𝑡𝑛 ⟩, 𝑓 𝑛 (𝑡1 , … , 𝑡𝑛 )⟩ • Hence 𝜑 ∈ Δ iff M ⊭ 𝜓.
• Since 𝜓 is closed, M ⊨ ¬𝜓 iff M ⊭
i.e., the operation that has the closed term 𝜓.
𝑓 𝑛 (𝑡1 , … , 𝑡𝑛 ) as value for ⟨𝑡1 , … , 𝑡𝑛 ⟩ as argument. • Hence, 𝜑 ∈ Δ iff M ⊨ ¬𝜓, i.e., 𝜑 ∈
(6) For each predicate 𝐹 𝑛 , let 𝐹 𝑛M be that subset of Δ iff M ⊨ 𝜑.
|M|𝑛 that includes the 𝑛-tuple ⟨𝑡1 , … , 𝑡𝑛 ⟩ iff the Case (b): 𝜑 takes the form 𝜓 ∨ 𝜃, where 𝜓 and
atomic wff 𝐹 𝑛 (𝑡1 , … , 𝑡𝑛 ) is included in Δ. 𝜃 have fewer connectives. First, if
(7) We must now prove that M is a model for Δ, i.e., 𝜑 ∈ Δ then M ⊨ 𝜑:
that for all sentences 𝜑, if 𝜑 ∈ Δ, then M ⊨ 𝜑. • Assume 𝜑 ∈ Δ, i.e., 𝜓 ∨ 𝜃 ∈ Δ.
We will actually prove something stronger, i.e., • It must be that either 𝜓 ∈ Δ or
that for all sentences 𝜑, 𝜑 ∈ Δ iff M ⊨ 𝜑. 𝜃 ∈ Δ. Otherwise, because Δ
We prove this by formula induction. is complete, we would have both
Base step: 𝜑 is a atomic. ¬𝜓 ∈ Δ and ¬𝜃 ∈ Δ. But then by
• The cases of ⊥ and ⊤ are trivial. Because DS, Δ ⊢ 𝜃 and then Δ ⊢ ⊥, which
Δ is consistent, ⊥ ∉ Δ and M ⊭ ⊥, and is impossible.
similarly ⊤ ∈ Δ and M ⊨ ⊤. • By the inductive hypothesis, either
• Focus on the cases where 𝜑 takes the M ⊨ 𝜓 or M ⊨ 𝜃. Either way,
form 𝐹 𝑛 (𝑡1 , … , 𝑡𝑛 ) where 𝑡1 , … , 𝑡𝑛 are closed M ⊨ 𝜓 ∨ 𝜃, i.e., M ⊨ 𝜑.
terms. Likewise if M ⊨ 𝜑 then 𝜑 ∈ Δ.
• Because closed terms contain no variables, • Assume M ⊨ 𝜑, i.e., M ⊨ 𝜓 ∨ 𝜃.
all variable assignments in M will asso- • By the semantics for ∨, Either
ciate each 𝑡𝑖 with itself. M ⊨ 𝜓 or M ⊨ 𝜃.
• So by the definition of satisfaction, all vari- • Hence, by the inductive hypothe-
able assignments in M will satisfy 𝜑 iff sis, either 𝜓 ∈ Δ or 𝜃 ∈ Δ. Either
⟨𝑡1 , … , 𝑡𝑛 ⟩ ∈ 𝐹 𝑛M . way, Δ ⊢ 𝜓 ∨ 𝜃 by Add.
• By the definition of truth, M ⊨ 𝜑 iff • By closure, 𝜓 ∨ 𝜃 ∈ Δ, i.e., 𝜑 ∈ Δ.
⟨𝑡1 , … , 𝑡𝑛 ⟩ ∈ 𝐹 𝑛M . Putting these two results together,
• By our characterization of M under (6) we get that 𝜑 ∈ Δ iff M ⊨ 𝜑.
above, ⟨𝑡1 , … , 𝑡𝑛 ⟩ ∈ 𝐹 𝑛M iff 𝐹 (𝑡1 , … , 𝑡𝑛 ) ∈ Δ. Case (c): 𝜑 takes the form 𝜓 ∧ 𝜃, where 𝜓 and
• So 𝐹 (𝑡1 , … , 𝑡𝑛 ) ∈ Δ iff M ⊨ 𝜑, i.e., 𝜑 ∈ Δ iff 𝜃 have fewer connectives. We can
M ⊨ 𝜑. show that 𝜑 ∈ Δ iff M ⊨ 𝜑. (This is
Induction step: Assume as inductive hypoth- homework.)
esis that it holds for all sentences 𝜓 with fewer Case (d): 𝜑 takes the form 𝜓 → 𝜃, where 𝜓
connectives than 𝜑, that 𝜓 ∈ Δ iff M ⊨ 𝜓. We and 𝜃 are sentences with fewer con-
will then show that it holds for 𝜑 that 𝜑 ∈ Δ iff nectives. First we prove that if 𝜑 ∈ Δ
M ⊨ 𝜑 as well. then M ⊨ 𝜑.
This proceeds by a proof by cases on the make- • Suppose 𝜑 ∈ Δ.
up of 𝜑. • Since Δ is complete and consistent,

28
 𝜓 ∈ Δ or ¬𝜓 ∈ Δ, but not both, and
likewise with 𝜃.
• However, because 𝜓 → 𝜃 ∈ Δ and
Δ is consistent, it cannot be that
both 𝜓 ∈ Δ and ¬𝜃 ∈ Δ, or else MP
would give us Δ ⊢ 𝜃 and Δ ⊢ ⊥.
• It follows that either 𝜓 ∉ Δ or 𝜃 ∈
Δ.
• By the inductive hypothesis, either
M ⊭ 𝜓 or M ⊨ 𝜃.
• By the semantics for conditionals,
it follows either way that M ⊨
𝜓 → 𝜃, i.e., M ⊨ 𝜑.
Now we prove that if M ⊨ 𝜑 then
𝜑 ∈ Δ.
• Suppose M ⊨ 𝜑, i.e., M ⊨ 𝜓 →
𝜃.
• Because 𝜓 and 𝜃 are closed, ei-
ther M ⊭ 𝜓 or M ⊨ 𝜃.
• By the inductive hypothesis,
𝜓 ∉ Δ or 𝜃 ∈ Δ.
• If 𝜓 ∉ Δ, then, because Δ is com-
plete, ¬𝜓 ∈ Δ. Then Δ ⊢ 𝜓 → 𝜃
by FA, and by closure, 𝜓 → 𝜃 ∈
Δ.
• If 𝜃 ∈ Δ, then by TC, Δ ⊢ 𝜓 → 𝜃,
and by closure, 𝜓 → 𝜃 ∈ Δ.
• So either way, 𝜓 → 𝜃 ∈ Δ, i.e.,
𝜑 ∈ Δ.
Case (e): 𝜑 takes the form 𝜓 ↔ 𝜃, where 𝜓 and
𝜃 have fewer connectives. We can
show that 𝜑 ∈ Δ iff M ⊨ 𝜑. (More
homework.)
Case (f): 𝜑 takes the form 𝑥𝜓(𝑥), where 𝜓(𝑥)
A lows that, for all closed terms 𝑡,
contains fewer connectives.
First we prove that if 𝜑 ∈ Δ then
M ⊨ 𝜑.
Case (g): 𝜑 takes the form 𝑥𝜓(𝑥), where 𝜓(𝑥)
• Suppose 𝜑 ∈ Δ, i.e., 𝑥𝜓(𝑥) ∈ Δ.
A
• Because 𝜓(𝑥) contains at most 𝑥
free, for all closed terms 𝑡, 𝜓(𝑡) is
a sentence.
• Because Δ is complete, for all
closed terms 𝑡, either 𝜓(𝑡) ∈ Δ or
¬𝜓(𝑡) ∈ Δ.
• However, since Δ is consistent, it
must be that for all closed terms 𝑡
29
, 𝜓(𝑡) ∈ Δ.
• By the inductive hypothesis, for
all closed terms 𝑡, M ⊨ 𝜓(𝑡).
• Because the domain of quantifi-
cation for M is the set of closed
terms, and every closed term is
interpreted as standing for itself,
a variable assignment 𝑠 will sat-
isfy 𝜓(𝑥) iff 𝜓(𝑡) is true where 𝑡 is
ValM𝑠 (𝑥).
• Because 𝜓(𝑡) is true in M for all
closed terms 𝑡, all variable assign-
ments of M will satisfy 𝜓(𝑥), and
will also satisfy 𝑥𝜓(𝑥).
A
• Hence, M ⊨ 𝑥𝜓(𝑥), i.e., M ⊨ 𝜑.
A
We now prove that if M ⊨ 𝜑 then
𝜑 ∈ Δ.
• Suppose M ⊨ 𝜑, i.e., all variable
assigments satisfy 𝑥𝜓(𝑥).
A
• Hence, all variable assignments
satisfy 𝜓(𝑥), regardless of what en-
tity in the domain gets assigned to
𝑥.
• Because the domain of quantifi-
cation for M is the set of closed
terms, and every closed term is
interpreted as standing for itself,
variable assignment 𝑠 will satisfy
𝜓(𝑥) iff 𝜓(𝑡) is true for that closed
term 𝑡 which is ValM 𝑠 (𝑥).
• So, for all closed terms 𝑡, M ⊨
𝜓(𝑡).
• By the inductive hypothesis, it fol-
𝜓(𝑡) ∈ Δ.
• Because Δ is universal, it follows
that 𝑥𝜓(𝑥) ∈ Δ, i.e., 𝜑 ∈ Δ.
A
E
contains fewer connectives.
First we prove that if 𝜑 ∈ Δ then
M ⊨ 𝜑.
• Suppose 𝜑 ∈ Δ, i.e., 𝑥𝜓(𝑥) ∈ Δ.
E
• It cannot be that ¬𝜓(𝑡) ∈ Δ for all
closed terms 𝑡. For if this were
true, then, because Δ is univer-
sal, 𝑥¬𝜓(𝑥) ∈ Δ, hence Δ ⊢
A
 ¬ 𝑥𝜓(𝑥) by CQ and MP, and then
E
Δ ⊢ ⊥, which is impossible. Result (Satisfiability): If Γ is a consistent set
• Hence for some closed term 𝑡, of sentences, then there is a denumerable inter-
¬𝜓(𝑡) ∉ Δ and therefore 𝜓(𝑡) ∈ Δ. pretation that is a model of Γ.
• By the inductive hypothesis, M ⊨
𝜓(𝑡). Hence for any variable as-
signment 𝑠, M, 𝑠 ⊨ 𝑥𝜓(𝑥), since
E
for the 𝑠 ′ differing from 𝑠 in assign- Proof: Assume Γ is consistent. By the Lindenbaum
ing 𝑡 to 𝑥, M, 𝑠 ′ ⊨ 𝜓(𝑥). Extension Lemma, there is a consistent, complete
• Therefore M ⊨ 𝑥𝜓(𝑥), i.e., M ⊨
E and universal set Δ, of which it is a subset. By the
𝜑. Term Model lemma, there is a denumerable inter-
Now we prove that if M ⊨ 𝜑 then pretation M which is a model for Δ. This means
𝜑 ∈ Δ. M makes every sentence in Δ true, and therefore
• Suppose M ⊨ 𝜑, i.e., M ⊨ also makes every sentence in Γ true. ❅
E
𝑥𝜓(𝑥).
• For any assigment 𝑠, there is one
differing from it 𝑠 ′ at its assign- Corollary (Löwenheim-Skolem Theorem):
ment to 𝑥 that satisfies 𝜓(𝑥). Be- If a set Γ of sentences is satisfiable, then it has a
cause the domain of quantifica- denumerable interpretation as a model.
tion is the set of closed terms,
ValM𝑠 ′ (𝑥) = 𝑡 for some closed term.
Clearly, M, 𝑠 ′ ⊢ 𝜓(𝑡) as well, and
because 𝜓(𝑡) is closed, M ⊨ 𝜓(𝑡). Proof: Suppose Γ is satisfiable. By our Consistency
• By the inductive hypothesis, result earlier, it must be consistent. Therefore, by
𝜓(𝑡) ∈ Δ. Satisfiability, it has a denumerable interpretation
• By EG, Δ ⊢ 𝑥𝜓(𝑥), and by clo-
E
as a model. ❅
sure, 𝑥𝜓(𝑥) ∈ Δ, i.e., 𝜑 ∈ Δ
E
(8) By induction, regardless of 𝜑’s length, 𝜑 ∈ Δ iffThe suprising thing about this result is that it shows
M ⊨ 𝜑. So M is a model for Δ. This establishes that no matter what kind of interpretation a theory
the Lemma. (collection of sentences) is intended to have, if it
❅
has any kind of interpretation making it true, it has
one of denumerable size. This is very counterinu-
Homework 13 tive if the “intended” models quantify over larger
Fill in the argument for cases (c) and (e) above; infinities, such as a theory for the mathematics of
compare cases (a), (b) and (d) for the general lines real numbers.
of argument. For case (e), you may cite the ↔-
“axioms” given on p. 18, or the derived rules that
follow from them (p. 19) as needed. Result (Completeness): If Γ ⊨ 𝜑 then Γ ⊢ 𝜑.

Completeness and compactness

We now get a result key for completeness. In fact, Proof: (1) Suppose Γ ⊨ 𝜑.
some books label this result itself “completeness”. (2) Suppose for reductio that Γ ∪ {¬𝜑} is consistent.
(3) Then, by Satisfiability, there is model M for
Γ ∪ {¬𝜑}, and so M ⊨ ¬𝜑.
(4) But since M is a model for Γ and Γ ⊨ 𝜑, it
follows that M ⊨ 𝜑.

30
(5) Every variable assignment 𝑠 for M both satis- (5) The members of Δ were taken from Γ, so Δ ⊆ Γ.
fies and does not satisfy 𝜑, which is impossible.(6) Clearly, Δ ⊢ 𝜑, and so, by Soundness, Δ ⊨ 𝜑.
(6) Hence, Γ ∪ {¬𝜑} is inconsistent. For (b), left-to-right, note if Γ is satisfiable, then
(7) This means that Γ ∪ {¬𝜑} ⊢ ⊥. some interpretation M makes every member of Γ
(8) By the deduction theorem Γ ⊢ ¬𝜑 → ⊥. true; this interpretation also makes every member
(9) By RedAx and DNE, Γ ⊢ 𝜑. ❅ of any subset Δ of Γ true, and so each such set Δ is
satisfiable.
For (b), right-to-left:
(1) Suppose Γ is finitely satisfiable.
Corollary (⊢/⊨ Bridge): Γ ⊢ 𝜑 iff Γ ⊨ 𝜑 (2) Suppose for reductio that Γ is inconsistent.
(3) Then Γ ⊢ ⊥, and by soundness, Γ ⊨ ⊥.
(4) But then, by (a), there is a finite subset Δ of Γ,
Δ ⊨ ⊥.
Proof: Follows directly from Soundness and Com- (5) This means Δ is unsatisfiable, which is impos-
pleteness. ❅ sible, as Γ is finitely satisfiable.
(6) Hence Γ is consistent, and by Satisfiability, Γ is
satisfiable. ❅
Corollary: ⊢ 𝜑 iff ⊨ 𝜑
Compactness is also seen as a surprising result. The
book also gives direct proofs of compactness that
do not go “through” completeness.
Proof: By the above, ∅ ⊢ 𝜑 iff ∅ ⊨ 𝜑. By definition, A related but relatively obvious result I’m
⊢ 𝜑 iff ∅ ⊢ 𝜑, and clearly ⊨ 𝜑 iff ∅ ⊨ 𝜑, as all throwing in just because it will be useful later:
interpretations are models of ∅ trivially. ❅

Definition: A set of sentences Γ is finitely satis- Result (Logical Conditionalization): Γ ⊢

fiable iff every finite subset Δ of Γ is satisfiable. 𝜑 iff there is a finite number of members of Γ,
𝜓1 , … , 𝜓𝑛 such that ⊢ (𝜓1 ∧ … ∧ 𝜓𝑛 ) → 𝜑.

Corollary (Compactness): (a) Γ ⊨ 𝜑 iff there

is a finite subset Δ of Γ such that Δ ⊨ 𝜑, and (b)
Γ is satisfiable iff it is finitely satisfiable. Proof: (1) Suppose Γ ⊢ 𝜑; only a finite number of
members of Γ can be used in the proof; let them
be 𝜓1 , … , 𝜓𝑛 .
(2) Hence 𝜓1 , … , 𝜓𝑛 ⊢ 𝜑
Proof: For the right-to-left part of (a), notice that (3) By successive applications of the deduction the-
every interpretation making every member of Γ orem, ⊢ 𝜓1 → (𝜓2 → … (𝜓𝑛 → 𝜑)), and by Imp,
true also makes every member of Δ true, and so ⊢ (𝜓1 ∧ … ∧ 𝜓𝑛 ) → 𝜑.
must make 𝜑 true. (4) For the other direction, assume ⊢ (𝜓1 ∧ … ∧
The left-to-right direction is more interesting: 𝜓𝑛 ) → 𝜑, where each 𝜓𝑖 ∈ Γ.
(1) Suppose Γ ⊨ 𝜑. (5) By Conj and MP, Γ ⊢ 𝜑. ❅
(2) By completeness, Γ ⊢ 𝜑.
(3) This means there is an axiomatic derivation of
𝜑 from Γ. This derivation is of finite length, and Homework 14
can use only finitely many members from Γ. Suppose that 𝜑 is a contingent sentence without
(4) Let Δ be the members used in the derivation. quantifiers. For example, 𝜑 might be “Fa → Ga”, or
For reasons given above, Δ is finite. “(R(a, f(a)) ∧ T(b, a)) ↔ (T(b, a) ↔ R(a, f(a)))”. Now

31
suppose we form a schema by replacing each dis- Definition: A sentence is an identity axiom iff it
tinct atomic formula in 𝜑 with a distinct schematic is Ref=Ax or has the form LLAx below, where 𝜑 does
letter. For our examples, we’d get, respectively, not contain any constants:
“𝜓1 → 𝜓2 ” and “(𝜓1 ∧ 𝜓2 ) ↔ (𝜓2 ↔ 𝜓1 )”. Let Γ be
(Ref=Ax)
A
the set of all instances of the resulting schema. (E.g., xx=x
if 𝜑 is “Fa → Ga”, then Γ is the set of all condition- 𝑥 𝑦(𝑥 = 𝑦 → (𝜑 → 𝜑[𝑦/𝑥])) (LLAx)
A A A A
𝑧1 … 𝑧𝑛
als.) Assuming only that 𝜑 is contingent (i.e., not a
tautology or self-contradiction, truth-table-wise), (Here 𝑧1 … 𝑧𝑛 are all the other free variables in 𝜑 be-
A A
show that Γ is inconsistent. sides 𝑥 and 𝑦; if there are none, then the 𝑧1 … 𝑧𝑛
part is omitted.)

5. Identity Logic Definition: Let 𝐼 be the set of identity axioms for

first-order language .
Syntax and semantics
Definition: A sentence 𝜑 is identity-derivable from
To capture the laws of identity logic, we just need Γ iff Γ ∪ 𝐼 ⊢ 𝜑. We can abbreviate this as Γ ⊢= 𝜑.

to treat “=2 ”, which we included in our definition
of predicates, in a special way. (So far, it’s just been ⊢= 𝑡 = 𝑡 (Ref=)
a predicate that an interpretation could interpret
1. x x = x Ref=Ax
A
any way it sees fit.)
We write 𝑡1 = 𝑡2 as an alternative to =2 (𝑡1 , 𝑡2 ), 2. 𝑡 = 𝑡 1 UI
and 𝑡1 ≠ 𝑡2 as an alternative to ¬= (𝑡1 , 𝑡2 ).
2

⊢= 𝑥 𝑦(𝑥 = 𝑦 → 𝑦 = 𝑥) (Sym=T)
A A
Definition: An interpretation M is identity-
M
normal iff =2 is the identity relation for the do-
main of M, i.e., the set of ordered pairs ⟨𝐴, 𝐴⟩ for 1. 𝑧 𝑥 𝑦(𝑥 = 𝑦 → (𝑥 = 𝑧 → 𝑦 = 𝑧)) LLAx
A A A

every 𝐴 where 𝐴 ∈ |M|. 2. 𝑎 = 𝑏 → (𝑎 = 𝑎 → 𝑏 = 𝑎) 1 UI×3

3. 𝑎 = 𝑎 → (𝑎 = 𝑏 → 𝑏 = 𝑎) 2 Int
Definition: A sentence 𝜑 is identity-valid iff for 4. 𝑎=𝑎 Ref=
every identity-normal interpretation M, M ⊨ 𝜑. I 5. 𝑎 = 𝑏 → 𝑏 = 𝑎 4, 5 MP
6. 𝑥 𝑦(𝑥 = 𝑦 → 𝑦 = 𝑥) 6 UG×2
A A
abbreviate this ⊨= 𝜑.

Definition: A sentence 𝜑 is an identity-logic- 𝑡1 = 𝑡2 ⊢= 𝑡2 = 𝑡1 (Sym=)

consequence of Γ iff for every identity-normal- By UI×2 on Sym=T and MP.
interpretation M, if M ⊨ Γ then M ⊨ 𝜑. I ab-
breviate this Γ ⊨= 𝜑. ⊢= 𝑥 𝑦 𝑧(𝑥 = 𝑦 → (𝑦 = 𝑧 → 𝑥 = 𝑧))
A A A
(Trans=T)
The book usually means ⊨= by ⊨; we’ll switch to 1. 𝑧 𝑥 𝑦(𝑥 = 𝑦 → (𝑥 ≠ 𝑧 → 𝑦 ≠ 𝑧))
A A A
LLAx
that later too. 2. 𝑎 = 𝑏 → (𝑎 ≠ 𝑐 → 𝑏 ≠ 𝑐) 1 UI×3
3. 𝑎 = 𝑏 → (𝑏 = 𝑐 → 𝑎 = 𝑐) 2 SL
4. 3 UG×3
A A A
Identity axioms and derived rules 𝑥 𝑦 𝑧(𝑥 = 𝑦 → (𝑦 = 𝑧 → 𝑥 = 𝑧))

The book also axiomatizes identity logic slightly 𝑡1 = 𝑡2 , 𝑡2 = 𝑡3 ⊢= 𝑡1 = 𝑡3 (Trans=)

differently, but this way makes it slightly easier to
apply our earlier results here. Follows from Trans=T, UI×3 and MP×2.

𝑡1 = 𝑡2 , 𝜑(𝑡1 ) ⊢= 𝜑(𝑡2 ) (LL)

𝑡1 = 𝑡2 , 𝜑(𝑡2 ) ⊢= 𝜑(𝑡1 )

32
The first follows by LLAx, UI as needed and MP×2. (4) Hence, for any identity-normal interpretation,
Note that if 𝜑(𝑥) contains constants we wish to re- M, all identity axioms are true, which means
main, we can use the instance of LLAx with free they are all identity-valid. ❅
variables 𝑧1 , … , 𝑧𝑛 in place of them, and start by in-
stantiating to those constants. The second follows
from the first and Sym=.
=
In identity logic we can define numerically def- Result (Identity soundness): If Γ ⊢ 𝜑 then
inite quantifiers: here 𝑦 is the first variable not Γ ⊨= 𝜑.
occurring in 𝜑(𝑥).

is shorthand for ¬ 𝑥𝜑(𝑥) Proof: (1) Assume Γ ⊢= 𝜑, i.e., Γ ∪ 𝐼 ⊢ 𝜑.

E E
0 𝑥𝜑(𝑥)
E
𝑛+1 𝑥𝜑(𝑥) is shorthand for (2) By our more generic Soundness result for first-
E E
𝑥[𝜑(𝑥) ∧ 𝑛 𝑦(𝑦 ≠ 𝑥 ∧ 𝜑(𝑦))] order logic, Γ ∪ 𝐼 ⊨ 𝜑.
(3) Let M be an arbitrary identity-normal model.
Because all members of 𝐼 are identity-valid,
So in particular: M ⊨ 𝐼 . If M ⊨ Γ, then, M ⊨ Γ ∪ 𝐼 , and thus
M ⊨ 𝜑.
abbreviates (4) In other words, 𝜑 is true in all identity-normal
E E E
1 𝑥𝜑(𝑥) 𝑥[𝜑(𝑥) ∧ ¬ 𝑦(𝑦 ≠ 𝑥 ∧ 𝜑(𝑦))]
models of Γ, and hence Γ ⊨= 𝜑. ❅
Homework 15
Show that ⊢= 1 𝑥𝜑(𝑥) ↔ 𝑥 𝑦(𝜑(𝑦) ↔ 𝑦 = 𝑥).
E A E

Corollary: If ⊢= 𝜑 then ⊨= 𝜑.
Identity logic metatheory

Result: Every member of 𝐼 is identity-valid.

Proof: (1) Let M be an identity-normal interpreta-
tion.
(2) For Ref=Ax, note for any given assignment
𝑠, whatever ValM 𝑠 (x) is, ⟨Val𝑠 (x), Val𝑠 (x)⟩ ∈
M M
M
=2 , and so M, 𝑠 ⊨ x = x, and therefore
M ⊨ x x = x.
A
(3) Suppose for reductio that for some instance of
LLAx, M ⊭ 𝑧1 … 𝑧𝑛 𝑥 𝑦(𝑥 = 𝑦 → (𝜑 →
A A A A
𝜑[𝑦/𝑥])). Then there would have to be an vari-
able assignment 𝑠 where M, 𝑠 ⊭ 𝑥 = 𝑦 → (𝜑 →
𝜑[𝑦/𝑥]). Then M, 𝑠 ⊨ 𝑥 = 𝑦 and M, 𝑠 ⊨ 𝜑 but
M, 𝑠 ⊭ 𝜑[𝑦/𝑥]. However, if M, 𝑠 ⊨ 𝑥 = 𝑦, then
2M
⟨ValM M
and so, because M
𝑠 (𝑥), Val𝑠 (𝑦)⟩ ∈ =
is identity-normal, Val𝑠 (𝑥) = ValM
M
𝑠 (𝑦). But
then since M, 𝑠 ⊨ 𝜑 it must be that M, 𝑠 ⊨
𝜑[𝑦/𝑥], which contradicts our earlier result.
Result (Factoring lemma): For any interpre-
tation (identity-normal or otherwise) M, if M ⊨
𝐼 , then there is an interpretation M′ which is
identity-normal such that, for all 𝜑, M ⊨ 𝜑 iff
M′ ⊨ 𝜑.
Proof: Let M be a model of 𝐼 . By our more general
soundness result, all 𝜑 such that 𝐼 ⊢ 𝜑 are true in
M, so all 𝜑 such that ⊢= 𝜑 are true in M.
Let 𝐸 be the relation M assigns to =2 (also
M
known as =2 ). 𝐸 is not guaranteed to be the iden-
tity relation, but it must have these properties:
• 𝐸 is reflexive, since M ⊨ x x = x.
A
• 𝐸 is symmetric, since M ⊨ 𝑥 𝑦(𝑥 = 𝑦 → 𝑦 =
A A
𝑥).
• 𝐸 is transitive, since M ⊨ 𝑥 𝑦 𝑧(𝑥 = 𝑦 →
A A A
(𝑦 = 𝑧 → 𝑥 = 𝑧)).

33
• Therefore, 𝐸 is an equivalence relation.
• Moreover, if ⟨𝐴1 , 𝐴2 ⟩ ∈ 𝐸 for any two members
of the domain 𝐴1 and 𝐴2 , then 𝐴1 and 𝐴2 are
“indiscernible” according to M, that is, they are
members of precisely the same extensions of pre-
cisely the same predicates in precisely the same
positions, since, e.g.:
M ⊨ 𝑥 𝑦(𝑥 = 𝑦 → (𝐹 1 𝑥 → 𝐹 1 𝑦))
A A
M ⊨ 𝑧 𝑥 𝑦(𝑥 = 𝑦 → (𝑅2 𝑥𝑧 → 𝑅2 𝑦𝑧))
A A A
M ⊨ 𝑧 𝑥 𝑦(𝑥 = 𝑦 → (𝑅2 𝑧𝑥 → 𝑅2 𝑧𝑦))
A A A
and so on.
Our goal is to construct an identity-normal inter-
pretation which renders all and only the same sen-
tences true. We do this by swapping out the ele-
ments of the original interpretation with their 𝐸-
equivalence classes.
More precisely let M′ be the interpretation
characterized as follows:
• |M′ | is set of 𝐸-equivalence classes formed on
members of |M|, i.e., the set of all [𝐴]𝐸 where
𝐴 ∈ |M|.
• For each constant 𝑐, 𝑐M = [𝑐M ]𝐸 .
′
• For each function letter 𝑓 𝑛 , 𝑓 𝑛M is the
′
set of all pairs ⟨⟨[𝐴1 ]𝐸 , … , [𝐴𝑛 ]𝐸 ⟩, [𝐵]𝐸 ⟩ where
⟨⟨𝐴1 , … , 𝐴𝑛 ⟩, 𝐵⟩ ∈ 𝑓 𝑛M .
• For each predicate 𝐹 𝑛 , 𝐹 𝑛M is the set of all
′
𝑛-tuples ⟨[𝐴1 ]𝐸 , … , [𝐴𝑛 ]𝐸 ⟩ where ⟨𝐴1 , … , 𝐴𝑛 ⟩ ∈
𝐹 𝑛M .
First, note that M′ is identity-normal. The pred-
icate =2 is assigned all ordered pairs ⟨[𝐴1 ]𝐸 , [𝐴2 ]𝐸 ⟩
such that ⟨𝐴1 , 𝐴2 ⟩ ∈ 𝐸. Since 𝐸 is an equivalence
relation, this means that [𝐴1 ]𝐸 and [𝐴2 ]𝐸 must be
the same classes (per your first homework).
We now must show that for all sentences M ⊨
𝜑 iff M′ ⊨ 𝜑. For each variable assignment 𝑠 of
M, there is a corresponding one 𝑠 # such that for
any variable 𝑥, ValM
′
𝑠 # (𝑥) = [Val𝑠 (𝑥)]𝐸 . It follows
M
inductively from our characterization of M′ ’s as-
signment to constants and function letters that
′
ValM𝑠 # (𝑡) = [Val𝑠 (𝑡)]𝐸 for all terms 𝑡.
M
We will show by formula induction that it holds
of every formula 𝜑, whether open or closed, M, 𝑠 ⊨
𝜑 iff M′ , 𝑠 # ⊨ 𝜑 for all such pairs 𝑠 and 𝑠 # .
Base step. Suppose 𝜑 is atomic. If 𝜑 is
⊤ then M, 𝑠 ⊨ 𝜑 and M′ , 𝑠 # ⊨ 𝜑. If 𝜑
34
is ⊥, then neither of these is true. Other-
wise, 𝜑 takes the form 𝐹 𝑛 (𝑡1 , … , 𝑡𝑛 ). If M, 𝑠 ⊨
𝜑, then ⟨ValM 𝑠 (𝑡1 ), … , ValM
𝑠 (𝑡𝑛 )⟩ ∈ 𝐹
𝑛M
and so
M′ M′
and so M , 𝑠 ⊨ 𝜑.
𝑛M ′ ′ #
⟨Val𝑠# (𝑡1 ), … Val𝑠# (𝑡𝑛 )⟩ ∈ 𝐹
If not, then not.
Induction step. Suppose for all formulas 𝜓 with
fewer connectives than 𝜑, that for every pair of
assignments 𝑠 and 𝑠 # it holds that M, 𝑠 ⊨ 𝜓 iff
M′ , 𝑠 # ⊨ 𝜓. We must show the same is true for
𝜑. There are three kinds of cases to consider.
(1) 𝜑 is a molecular statement, and takes one of the
forms ¬𝜓, 𝜓 ∨ 𝜃, 𝜓 ∧ 𝜑, 𝜓 → 𝜃, or 𝜓 ↔ 𝜃. By the
inductive hypothesis, M, 𝑠 ⊨ 𝜓 iff M′ , 𝑠 # ⊨ 𝜓
and M, 𝑠 ⊨ 𝜃 iff M′ , 𝑠 # ⊨ 𝜃 for any 𝑠, 𝑠 # . But
the satisfaction or not of a molecular statement
depends only on the satisfaction (or not) of its
parts. Hence, M, 𝑠 ⊨ 𝜑 iff M′ , 𝑠 # ⊨ 𝜑, for any
pair 𝑠, 𝑠 # .
(2) 𝜑 takes the form 𝑥𝜓. For some 𝑠, suppose
A
M, 𝑠 ⊨ 𝜑, which means that for every 𝑠 ′ dif-
ferent from 𝑠 at what is assigned to 𝑥 is such
that M, 𝑠 ′ ⊨ 𝜓. Every assignment 𝑠 ′# different
from 𝑠 # in M′ by what gets assigned to 𝑥 is
paired up with such an 𝑠 ′ in M and so by the in-
ductive hypothesis, M′ , 𝑠 ′# ⊨ 𝜓, and therefore
M′ , 𝑠 # ⊨ 𝑥𝜓, i.e., M′ , 𝑠 # ⊨ 𝜑. Suppose instead
A
that M, 𝑠 ⊭ 𝜑, then for some 𝑠 ′ , M, 𝑠 ′ ⊭ 𝜓 and
for the corresponding 𝑠 ′# we have M′ , 𝑠 ′# ⊭ 𝜓
and thus M′ , 𝑠 # ⊭ 𝜑. So M, 𝑠 ⊨ 𝜑 iff M′ , 𝑠 # ⊨ 𝜑.
(3) 𝜑 takes the form 𝑥𝜓. For some 𝑠, suppose
E
M, 𝑠 ⊨ 𝜑, which means that for some 𝑠 ′ differ-
ent from 𝑠 at what is assigned to 𝑥 is such that
M, 𝑠 ′ ⊨ 𝜓. 𝑠 ′ is paired up with some 𝑠 ′# in M′ ,
and by the inductive hypothesis, M′ , 𝑠 ′# ⊨ 𝜓,
and therefore M′ , 𝑠 # ⊨ 𝑥𝜓, i.e., M′ , 𝑠 # ⊨ 𝜑.
E
Suppose instead that M, 𝑠 ⊭ 𝜑, then for every
𝑠 ′ differing from 𝑠 by what it assigns to 𝑥 is such
that M, 𝑠 ′ ⊭ 𝜓 and, by the inductive hypothe-
sis, for each 𝑠 ′# differing from 𝑠 # in this way we
have M′ , 𝑠 ′# ⊭ 𝜓 and thus M′ , 𝑠 # ⊭ 𝜑. So again
M, 𝑠 ⊨ 𝜑 iff M′ , 𝑠 # ⊨ 𝜑.
Hence regardless of the complexity of 𝜑, for each
pair of assignments 𝑠 and 𝑠 # , M, 𝑠 ⊨ 𝜑 iff M′ , 𝑠 # ⊨ 𝜑.
Such pairs exhaust the variable assignments in M
and M′ , so M ⊨ 𝜑 iff M′ ⊨ 𝜑. ❅
 Result (Identity completeness): If Γ ⊨= 𝜑
then Γ ⊢= 𝜑.

Homework 16
Prove the identity completeness result above. A
hint: the proof is similar to the more generic com-
pleteness proof on p. 30, except with Γ ∪ 𝐼 ∪ {¬𝜑}
playing the role of Γ ∪ {¬𝜑}, and at some point the
Factoring lemma is needed.

Corollary: Γ ⊢= 𝜑 iff Γ ⊨= 𝜑

Proof: Immediate from identity soundness and

identity completeness. ❅

From here on out, we are mainly interested in logi-

cal systems built on identity logic, and so we will
begin using ⊢ to mean ⊢= and ⊨ to mean ⊨= , unless
otherwise noted.

Homework 17
Let schmatisfaction be the relation that holds
between an intepretation, a variable assignment
and a formula 𝜑 that is just like satisfaction except
that it treats ⊤ like ⊥, i.e., no M and 𝑠 schmatisfy
⊤, by definition. From this notion we can define
schmuth and schmlogical schmalidity, etc., just like
truth and logical validity, etc., are defined from
regular satisfaction. Using these notions, and proof-
induction, prove that (VerumAx) is independent of
the other axioms of first-order logic, i.e., that one
cannot prove it from instances of the other axiom
schemata and MP and QR alone.

35
Unit 2

Metamathematics and the Gödel Results

1. Theories in General Definition: A theory T is axiomatized by a set Γ

iff T is the closure of Γ.
In this unit, we take logical systems for mathemat-
ics as our object languages. Definition: A theory T is an extension of T′ iff
T′ ⊆ T.
Definition: A set of sentences Γ is closed (or more
specifically “closed under entailment”) iff for all sen- Note that because sets have extensional identity
tences 𝜑 in a given first-order language, if Γ ⊢ 𝜑, conditions, the same theory might be axiomatized
then 𝜑 ∈ Γ. by different sets of axioms, if those axioms allow
you to derive all and only the same results.
Definition: A theory is a closed set of sentences.
Definition: The closure of a set of sentences Γ is
the set of all sentences that can be derived from Γ, i.e., Result: If T is the closure of Γ (or axiomatized
{𝜑 ∶ Γ ⊢ 𝜑}. by Γ), then for all 𝜑, Γ ⊢ 𝜑 iff T ⊢ 𝜑.

A closed set or theory is one containing all its logi-

cal consequences. Any set that isn’t a theory has a
closure which is a theory. Homework 18
There are two ways that theories are typically
Prove the result above.
specified: either by providing a specification of a
This result allows us to talk more or less inter-
set of “proper” or “non-logical” axioms of which
changeably about a set of axioms and the “theory”
the theory is the “closure”, or by providing an in-
which is its closure.
terpretation where the theory is everything true
in that interpretation. One thing we might be in-
terested in is connecting the two: can the same 2. Theories for Arithmetic
theory be equivalently specified both semantically
and deductively? We are concerned here only with theories where
By convention, we shall start using boldface the intended language and semantics deal only with
abbreviations as constant names of theories, espe- natural numbers 0, 1, 2, 3, … , their properties and
cially where the same abbreviation, non-bold face, relations. The set of natural numbers is (in mathe-
is a set of sentences of which it is the closure. For matics) often called N.
example, with might use A as the name of the clo-
sure of some set we’re calling 𝐴. (We’ll continue
The language of arithmetic
to use uppercase Greek letters such as Γ and Δ as
variables in the metalanguage for sets of sentences, Definition: The language of arithmetic, 𝐴 , is
including theories, but also T as a variable for an the first-order language making use of the following
arbitrary theory.) vocabulary:

36
(1) There is only one constant, a, but instead of “ a”,
by convention we write “ 0”.
Result: TA is a theory.
(2) There are two predicates, both dyadic, A2 and =2 ,
but by convention, rather than writing “A2 (𝑡1 , 𝑡2 )”
and “=2 (𝑡1 , 𝑡2 )” we write 𝑡1 < 𝑡2 and 𝑡1 = 𝑡2 .
(3) There are three function letters, one monadic and
Proof: (1) Let 𝜑 be an arbitrary sentence such that
two dyadic, f1 , f2 and f12 , but by convention, rather
TA ⊢ 𝜑.
than writing f1 (𝑡), f2 (𝑡1 , 𝑡2 ) and f12 (𝑡1 , 𝑡2 ), we write
(2) By the soundness of first-order logic with iden-
(𝑡)′ , (𝑡1 + 𝑡2 ), and (𝑡1 × 𝑡2 ), and omit the paren-
tity, TA ⊨ 𝜑. This means that 𝜑 is true in every
theses when there is no ambiguity.
model of TA.
We shall for the moment consider logical systems (3) By the definition of TA, N makes every mem-
using this more austere language; one in which ber of TA true, and so N is a model of TA.
there are no other predicates, constants or function (4) Hence, 𝜑 is true in N.
signs (at least taken as primitive). We still use the (5) Thus, 𝜑 ∈ TA by the definition of TA.
full range of variables. “Dummy” constants are also (6) Since 𝜑 was arbitrary, this shows that TA is
allowed in UG and EI proofs, though not axioms. closed under entailment, and therefore, is a the-
ory. ❅

Semantics for 𝐴 The question then arises if there is a different way

These vocabulary items in 𝐴 have obvious in- of specifying TA, one that appeals to a limited set
tended interpretations. of axioms. Obviously, TA is trivially axiomatized
by itself, but this is not very useful. If the axioms
Definition: The standard interpretation or one is allowed to use are “everything true”, proofs
standard model of arithmetic, N, is the interpreta- using this set of axioms do not serve very well as
tion defined as follows. an independent means for establishing what is true
(1) The domain of quantification |N| for N is N. and what isn’t. Is there a finite, or at least easily and
(2) The constant “ 0” is interpreted as standing for independently determinable, list of axioms from
zero, i.e., 0N = zero. which all other truths of arithmetic follow, i.e., of
(3) The predicate “=2 ” is interpreted as standing for which TA is the closure?
identity, i.e., =2N is {⟨𝑛, 𝑛⟩ ∶ 𝑛 ∈ N}. (This
makes N identity-normal.) Peano and Robinson arithmetics
(4) The predicate “<” is interpreted as standing for
the less-than relation on N, i.e., <N is {⟨𝑚, 𝑛⟩ ∶ At the end of the 19th and beginning of the 20th cen-
𝑚 ∈ N, 𝑛 ∈ N and 𝑚 is less than 𝑛}. tury, prior to Gödel’s results, it was widely thought
(5) The functor “ ′ ” is interpreted as standing for the that all of arithmetic could be derived from the
successor function on N, i.e., ′N is {⟨𝑚, 𝑛⟩ ∶ 𝑚 ∈ following assumptions:
N, 𝑛 ∈ N and 𝑛 is one more than 𝑚}.
Definition: Stated in English, the Peano-
(6) The functor “ +” is interpreted as standing for the
Dedekind axioms are the following principles:
addition operation on N, i.e., +N is {⟨⟨𝑚, 𝑛⟩, 𝑠⟩ ∶
(P1) Zero is a natural number.
𝑚, 𝑛, 𝑠 ∈ N and 𝑠 is the sum of 𝑚 and 𝑛}.
(P2) Every natural number has a successor which is
(7) The functor “ ×” is interpreted as standing for
also a natural number.
multiplication, i.e., ×N is {⟨⟨𝑚, 𝑛⟩, 𝑠⟩ ∶ 𝑚, 𝑛, 𝑠 ∈
(P3) Zero is not the successor of any natural number.
N and 𝑠 is the product of 𝑚 and 𝑛}.
(P4) No two natural numbers have the same succes-
Definition: Let true arithmetic or TA be the set sor.
of sentences that are true in N, i.e., {𝜑 ∶ N ⊨ 𝜑}. (P5) If something is both true of zero, and true of the
successor of a number whenever it is true of that

37
 number, then it is true of all natural numbers The work of (P1) is basically captured by having “0”
(i.e., the principle of mathematical induction). as a constant assigned to a member of the domain
of the quantification, and needn’t be expressed by a
These were first considered in the context in which separate axiom. Similarly, the work of (P2) is done
it was presupposed that one could make use of set- more or less just by having “′ ” in the language.
theoretic definitions of things like multiplication,
addition, less-than, etc. Dropping this assumption, Definition: Let the set Q be the set containing Q1
and refactoring for modern symbolic logic, we can through Q8 .
generate the following set of axioms more or less
doing the work of (P1)–(P4): Definition: Robinson arithmetic, or Q, is the
theory which is the closure of Q.
′ ′
(Q1 )
A A
x y(x = y → x = y)
Robinson arithmetic does not have any way
′
(Q2 )
A
x0≠x
to capture mathematical induction as an object-
x(x ≠ 0 → y x = y′ ) (Q3 ) language notion, as thus is weaker than the system
A E

(Q4 ) intended by Peano and Dedekind. It grew to be

A
xx+0=x
′ ′
(Q5 ) something that was studied only later, as an object
A A
x y x + y = (x + y)
(Q6 ) of comparison with Peano arithmetic. As we shall
A
xx×0=0
see as we proceed, however, is it is already very
x y x × y′ = (x × y) + x (Q7 )
A A
powerful and interesting on its own.
x y(x < y ↔ z z′ + x = y) (Q8 )
A A E
To capture (P5), we can consider this schema—
(PA9 )—for mathematical induction, where 𝑦1 , … , 𝑦𝑛
We might gloss these in English this way:
are the free variables of 𝜑(𝑥) apart from 𝑥:
Q1 . No two numbers have the same successor (P4).
Q2 . Zero is not the successor of any number (P3). 𝑦1 … 𝑦𝑛 ([𝜑(0) ∧ 𝑥(𝜑(𝑥) → 𝜑(𝑥 ′ ))] → 𝑥𝜑(𝑥))
A A A A
Q3 . Every number other than zero is the succes-
sor of some number. (This partly captures our Definition: Let PA be the set containing Q1
intention only to quantify over natural num- through Q8 and every instance of the induction
bers.) schema PA9 .
Q4 . Any number plus zero is itself.
Q5 . Any number plus the successor of another Definition: (First-order) Peano arithmetic, or
number is the successor of the sum of the first PA, is the theory which is the closure of set PA.
number and the second. (Together, this and the
previous axiom provide an implicit definition Technically, the set PA of axioms for PA has in-
of addition, which is needed when taking it as finitely many members, as PA9 is not a single axiom,
primitive.) but a schema which has infinitely many instances.
Q6 . Any number multiplied by zero is zero. In a higher-order logic, the schematic letter 𝜑 might
Q7 . The product of one number and the successor be replaced by an object-language variable, and PA
of another is equal to the sum of their product might be axiomatized in a finite way.
and the first. (Again, this and the previous one Since Q is a subset of PA, clearly Q is a subset
implicitly define multiplication.) of PA, and so PA is an extension of Q.
Q8 . One natural number is less than another if
and only if there is a non-zero number which
sums with the first to yield the second. (This Result: Theories Q, PA and TA are all consis-
basically provides a complete definition of tent.
less-than for natural numbers; indeed, one
might have taken 𝑡1 < 𝑡2 to simply abbreviate
𝑧 𝑧 ′ + 𝑡1 = 𝑡2 instead.)
E

38
Proof: The members of all three theories are true Most proofs in PA of universally quantified sen-
in the standard interpretation of arithmetic. For tences make use of MI.
TA this is true by definition. For Q and PA, note
that the members of Q and PA are all true, and by Homework 19
Prove: PA ⊢ 𝑥 𝑦(𝑥 × 𝑦 = 0 → 𝑥 = 0 ∨ 𝑦 = 0).
A A
soundness, so are their entailments. Hence, these
theories are satisfiable, and therefore consistent.❅ You can use (0+) above.

(0+)
A A
PA ⊢ 𝑥 𝑦(𝑥 + 𝑦 = 0 → 𝑥 = 0 ∧ 𝑦 = 0) (+0)
A
PA ⊢ 𝑥 0 + 𝑥 = 𝑥

1. 𝑏 + 0 = 0 Hyp 1. 0 + 0 = 0 Q4 ,UI
2. 𝑏 + 0 = 𝑏 Q4 , UI 2. 0 + 𝑏 = 𝑏 Hyp
3. 𝑏 = 0 1, 2 LL 3. 0 + 𝑏′ = (0 + 𝑏)′ Q5 , UI
4. 0 = 0 Ref= 4. 0 + 𝑏′ = 𝑏′ 2, 3 LL
5. 𝑏 = 0 ∧ 0 = 0 3, 4 Conj 5. 0 + 𝑏 = 𝑏 → 0 + 𝑏′ = 𝑏′ 2–4 DT
6. 𝑥(0 + 𝑥 = 𝑥 → 0 + 𝑥 ′ = 𝑥 ′ ) 4 UG
A
6. 𝑏 + 0 = 0 → 𝑏 = 0 ∧ 0 = 0 1–5 CP/DT
7. 𝑥 0 + 𝑥 = 𝑥 1, 6 MI
A
7. 𝑏 + 𝑐 = 0 → 𝑏 = 0 ∧ 𝑐 = 0 Hyp
8. 𝑏 + 𝑐 = 0
′
Hyp
9. 𝑏 + 𝑐 = (𝑏 + 𝑐)
′ ′
Q5 , UI PA ⊢ 𝑥 𝑦 𝑥 ′ + 𝑦 = (𝑥 + 𝑦)′ (AltQ5 )
A A
10. (𝑏 + 𝑐)′ = 0 8, 9 LL
1. 𝑏′ + 0 = 𝑏′ Q4 , UI
11. (𝑏 + 𝑐)′ ≠ 0 Q2 , Sym=T, UI, MT
2. 𝑏 + 0 = 𝑏 Q4 , UI
12. ⊥ 10, 11 Contra
3. 𝑏′ = 𝑏′ Ref=
13. 𝑏 = 0 ∧ 𝑐 = 0
′
12 Explode
4. (𝑏 + 0) = 𝑏
′ ′
2, 3 LL
14. 𝑏 + 𝑐 = 0 → 𝑏 = 0 ∧ 𝑐 = 0
′ ′
8–13 CP/DT
5. 𝑏 + 0 = (𝑏 + 0)
′ ′
1, 4 LL
15. (𝑏 + 𝑐 = 0 → 𝑏 = 0 ∧ 𝑐 = 0) →
6. 𝑏′ + 𝑐 = (𝑏 + 𝑐)′ Hyp
(𝑏 + 𝑐′ = 0 → 𝑏 = 0 ∧ 𝑐′ = 0) 7–14 CP/DT
7. 𝑏 + 𝑐 = (𝑏 + 𝑐)
′ ′ ′ ′
Q5 , UI
16. 𝑦[(𝑏 + 𝑦 = 0 → 𝑏 = 0 ∧ 𝑦 = 0) →
A
8. 𝑏′ + 𝑐 ′ = (𝑏 + 𝑐)′′ 6, 7 LL
(𝑏 + 𝑦 ′ = 0 → 𝑏 = 0 ∧ 𝑦 ′ = 0)] 15 UG
9. 𝑏 + 𝑐′ = (𝑏 + 𝑐)′ Q5 , UI
17. 𝑥 [([𝑥 + 0 = 0 → 𝑥 = 0 ∧ 0 = 0] ∧
A
A 10. 𝑏 + 𝑐 = (𝑏 + 𝑐 )
′ ′ ′ ′
8, 9 LL
𝑦[(𝑥 + 𝑦 = 0 → 𝑥 = 0 ∧ 𝑦 = 0) →
11. 𝑏′ + 𝑐 = (𝑏 + 𝑐)′ → 𝑏′ + 𝑐′ = (𝑏 + 𝑐′ )′ 6–10 DT
(𝑥 + 𝑦 ′ = 0 → 𝑥 = 0 ∧ 𝑦 ′ = 0)]) →
12. 𝑦[𝑏′ + 𝑦 = (𝑏 + 𝑦)′ → 𝑏′ + 𝑦 ′ = (𝑏 + 𝑦 ′ )′ ] 11 UG
A
PA9
A
𝑦(𝑥 + 𝑦 = 0 → 𝑥 = 0 ∧ 𝑦 = 0)]
13. 𝑦 𝑏′ + 𝑦 = (𝑏 + 𝑦)′ 5, 12 MI
A
18. ([𝑏 + 0 = 0 → 𝑏 = 0 ∧ 0 = 0] ∧ 14. 𝑥 𝑦 𝑥 + 𝑦 = (𝑥 + 𝑦)
A A ′ ′
13 UG
A
𝑦[(𝑏 + 𝑦 = 0 → 𝑏 = 0 ∧ 𝑦 = 0) →
(𝑏 + 𝑦 ′ = 0 → 𝑏 = 0 ∧ 𝑦 ′ = 0)]) →
(Com+)
A A
A
𝑦(𝑏 + 𝑦 = 0 → 𝑏 = 0 ∧ 𝑦 = 0) 17 UI PA ⊢ 𝑥 𝑦 𝑥 + 𝑦 = 𝑦 + 𝑥
19. 𝑦(𝑏 + 𝑦 = 0 → 𝑏 = 0 ∧ 𝑦 = 0) 6, 16, 18 MP×2
A
1. 𝑏 + 0 = 𝑏 Q4 , UI
20. 𝑥 𝑦(𝑥 + 𝑦 = 0 → 𝑥 = 0 ∧ 𝑦 = 0) 19 UG
A A
2. 0 + 𝑏 = 𝑏 +0, UI
So formulated, we can think of lines 1–6 as the base 3. 𝑏 + 0 = 0 + 𝑏 1, 2 LL
step and lines 7–17 as the induction step, where line 4. 𝑏 + 𝑐 = 𝑐 + 𝑏 Hyp
7 is the inductive hypothesis. Actually, this example 5. 𝑏 + 𝑐 = (𝑏 + 𝑐)
′ ′
Q5 , UI
is a bit unusual in that line 7 is never actually used, 6. 𝑏 + 𝑐′ = (𝑐 + 𝑏)′ 3, 4 LL
and so we might have shortened things by getting 7. 𝑐′ + 𝑏 = (𝑐 + 𝑏)′ AltQ5 , UI
line 14 with FA and line 15 by TC, but the structure 8. 𝑏 + 𝑐 = 𝑐 + 𝑏
′ ′
6, 7 LL
of the induction would be less clear then. 9. 𝑏 + 𝑐 = 𝑐 + 𝑏 → 𝑏 + 𝑐 ′ = 𝑐′ + 𝑏 4–8 DT
10. 𝑦(𝑏 + 𝑦 = 𝑦 + 𝑏 → 𝑏 + 𝑦 ′ = 𝑦 ′ + 𝑏) 9 UG
A
We might as well make this a derived rule:
11. 𝑦 𝑏 + 𝑦 = 𝑦 + 𝑏 3, 10 MI
A
PA, 𝜑(0), 𝑥(𝜑(𝑥) → 𝜑(𝑥 ′ )) ⊢ 𝑥𝜑(𝑥) (MI) 12. 𝑥 𝑦 𝑥 + 𝑦 = 𝑦 + 𝑥 10 UG
A A A A

39
3. A Naïve Attempt to Go Set theoretic definitions
(𝑡 ∩ 𝑢) for {𝑥 ∶ 𝑥 ∈ 𝑡 ∧ 𝑥 ∈ 𝑢}
More Primitive: System F (𝑡 ∪ 𝑢) for {𝑥 ∶ 𝑥 ∈ 𝑡 ∨ 𝑥 ∈ 𝑢}
Do we really need to take all of 0, ′ , +, ×, < as (𝑡 ⧵ 𝑢) for {𝑥 ∶ 𝑥 ∈ 𝑡 ∧ 𝑥 ∉ 𝑢}
primitive? Cardinal numbers would seem to be 𝑡 for {𝑥 ∶ 𝑥 ∉ 𝑡}
(𝑡 ⊆ 𝑢) for 𝑥(𝑥 ∈ 𝑡 → 𝑥 ∈ 𝑢)
A
what sets or classes have in common when they
are the “same size”: can we define a number as the V for {𝑥 ∶ 𝑥 = 𝑥}
class of all classes of a certain size, perhaps, in set ∅ for {𝑥 ∶ 𝑥 ≠ 𝑥}
theory. Here is a tempting try. {𝑡} for {𝑥 ∶ 𝑥 = 𝑡}
Set theory can be formulated using only the {𝑡, 𝑢} for {𝑥 ∶ 𝑥 = 𝑡 ∨ 𝑥 = 𝑢}
predicate ∈, but it is often more convenient to in- ⟨𝑡, 𝑢⟩ for {{𝑡}, {𝑡, 𝑢}}
(𝑡 × 𝑢) for {𝑥 ∶ 𝑦 𝑧(𝑥 = ⟨𝑦, 𝑧⟩ ∧ 𝑦 ∈ 𝑡 ∧ 𝑧 ∈ 𝑢)}
E E
troduce the notation {𝑥 ∶ 𝜑(𝑥)} as well:
Dom(𝑡) for {𝑥 ∶ 𝑦(⟨𝑥, 𝑦⟩ ∈ 𝑡)}
E
Rng(𝑡) for {𝑥 ∶ 𝑦(⟨𝑦, 𝑥⟩ ∈ 𝑡)}
E
Syntax Fld(𝑡) for (Dom(𝑡) ∪ Rng(𝑡))
Inv(𝑡) for {𝑥 ∶ 𝑦 𝑧(𝑥 = ⟨𝑦, 𝑧⟩ ∧ ⟨𝑧, 𝑦⟩ ∈ 𝑡)}
E E
1. We add to the syntax of predicate logic the fol-
Fnct(𝑡) for 𝑥 𝑦 𝑧(⟨𝑥, 𝑦⟩ ∈ 𝑡 ∧
A A A
lowing subnective, which yields a term for any
variable 𝑥 and wff 𝜑(𝑥). ⟨𝑥, 𝑧⟩ ∈ 𝑡 → 𝑦 = 𝑧)
{𝑥 ∶ 𝜑(𝑥)} One-one(𝑡) for (Fnct(𝑡) ∧ Fnct(Inv(𝑡)))
All occurrences of 𝑥 here are bound.
2. We also choose a two-place predicate letter E2 to Mathematical definitions
(𝑡 ≈ 𝑢) for 𝑥(One-one(𝑥) ∧ Dom(𝑥) = 𝑡 ∧
E
use for the membership relation. An expression
of the form (𝑡 ∈ 𝑢) is shorthand for E2 (𝑡, 𝑢), and Rng(𝑥) = 𝑢)
(𝑡 ∉ 𝑢) is shorthand for ¬E2 (𝑡, 𝑢).; we also use Card(𝑡) for {𝑥 ∶ 𝑥 ≈ 𝑡}
the identity predicate. 0 for Card(∅)
𝑡 ′ for {𝑥 ∶ 𝑦(𝑦 ∈ 𝑥 ∧ 𝑥 ⧵ {𝑦} ∈ 𝑡)}
E
1 for 0′
Axioms 2 for 1′
System F is the theory which is the closure of the set 3 for 2′ . . . and so on for other numerals
containing all specified instances of the following N for
A A
schemata, (NC) and (Ext) (and their universal clo- {𝑥 ∶ 𝑦(0 ∈ 𝑦 ∧ 𝑧(𝑧 ∈ 𝑦 → 𝑧 ′ ∈ 𝑦) → 𝑥 ∈ 𝑦)}
Fin(𝑡) for 𝑥(𝑥 ∈ N ∧ 𝑡 ∈ 𝑥)
E
sures, if 𝜑 and 𝜓 contain additional free variables):
Infin(𝑡) for ¬Fin(𝑡)
A
𝑥(𝜑(𝑥) ↔ 𝑥 ∈ {𝑦 ∶ 𝜑(𝑦)}), (NC) Denum(𝑡) for (𝑡 ≈ N)
Ctbl(𝑡) for (Fin(𝑡) ∨ Denum(𝑡))
for all cases in which the variable 𝑦 is free for 𝑥 in (𝑡 ≤ 𝑢) for 𝑥 𝑦 𝑧(𝑥 ∈ 𝑡 ∧ 𝑦 ∈ 𝑢 ∧ 𝑧 ⊆ 𝑦 ∧ 𝑥 ≈ 𝑧)
E E E
𝜑(𝑥). (𝑡 < 𝑢) for ((𝑡 ≤ 𝑢) ∧ ¬(𝑢 ≤ 𝑡))
A §(𝑡) for {𝑥 ∶ 𝑥 ∈ N ∧ 𝑥 < 𝑡}
𝑥(𝑥 ∈ {𝑦 ∶ 𝜑(𝑦)} ↔ 𝑥 ∈ {𝑧 ∶ 𝜓(𝑧)}) →
(𝑡 + 𝑢) for Card((§(𝑡) × {0}) ∪ (§(𝑢) × {1}))
{𝑦 ∶ 𝜑(𝑦)} = {𝑧 ∶ 𝜓(𝑧)} (Ext) (𝑡 ⋅ 𝑢) for Card(§(𝑡) × §(𝑢))
where {𝑦 ∶ 𝜑(𝑦)} and {𝑧 ∶ 𝜓(𝑧)} do not contain 𝑥
free. Results
With these definitions in place, one can derive the
Some intuitive definitions Peano-Dedekind axioms as theorems in the follow-
(Let 𝑥, 𝑦 and 𝑧 below be the first variables not oc- ing forms:
curring in 𝑡 and 𝑢.) (P1) F ⊢ 0 ∈ N

40
(P2) Homework 20
A
F ⊢ 𝑥(𝑥 ∈ N → 𝑥 ′ ∈ N)
(P3)
A ′
F ⊢ 𝑥(𝑥 ∈ N → 0 ≠ 𝑥 ) Without using Russell’s paradox or other contra-
(P4)
A
F ⊢ 𝑥(𝑥 ∈ N ∧ 𝑦 ∈ N → (𝑥 ′ = 𝑦 ′ → 𝑥 = 𝑦)) diction, prove F ⊢ 𝑥 𝑦({𝑥} = {𝑦} → 𝑥 = 𝑦).
A A
(P5)
A
F ⊢ 𝜑(0) ∧ 𝑥(𝜑(𝑥) → 𝜑(𝑥 ′ )) →
A Let us return to our consideration of Q and PA as
𝑥(𝑥 ∈ N → 𝜑(𝑥))
presented in the Open Logic Text, however.
As well as analogues of PA’s other axioms:
A E
F⊢ 𝑥(𝑥 ∈ N → (𝑥 ≠ 0 → 𝑦 𝑥 = 𝑦 ′ ))
F⊢
A
𝑥(𝑥 ∈ N → 𝑥 + 0 = 𝑥)
4. Numerals and Q versus PA
A A
F⊢ 𝑥 𝑦(𝑥 ∈ N ∧ 𝑦 ∈ N → 𝑥 + 𝑦 ′ = (𝑥 + 𝑦)′ ) Numerals are the canonical signs for numbers in a
A
F⊢ 𝑥 𝑥 ⋅0=0 given language or communication system. In the
A A
F⊢ 𝑥 𝑦(𝑥 ∈ N ∧ 𝑦 ∈ N → (𝑥 ⋅ 𝑦 ′ ) = ((𝑥 ⋅ 𝑦) + 𝑥)) decimal system used in mathematics and everyday
A A
F⊢ 𝑥 𝑦(𝑥 ∈ N ∧ 𝑦 ∈ N → life, a numeral would be a series of digits like “112”
E
(𝑥 < 𝑦 ↔ 𝑧(𝑧 ∈ N ∧ 𝑧 ′ + 𝑥 = 𝑦))) or “504”, without function signs, etc.
Also, we have, e.g.: In language 𝐴 we can give this definition in-
E
F ⊢ ( 1 𝑥𝜑(𝑥)) ↔ ({𝑥 ∶ 𝜑(𝑥)} ∈ 1) stead:
E
F ⊢ ( 2 𝑥𝜑(𝑥)) ↔ ({𝑥 ∶ 𝜑(𝑥)} ∈ 2) Definition: In 𝐴 , the numeral for number 𝑛 is
E
F ⊢ ( 3 𝑥𝜑(𝑥)) ↔ ({𝑥 ∶ 𝜑(𝑥)} ∈ 3) the constant “ 0” followed by 𝑛 successor function
And so on. signs.

The numeral for zero (0) is “0”.

Disaster The numeral for one (1) is “0′ ”.
The system F, unfortunately, is inconsistent due to The numeral for two (2) is “0 ”.
′′

Russell’s paradox: The numeral for three (3) is “0′′′ ”, and so on.
Convention: Let 𝑛 be an abbreviation for the nu-
F ⊢ {𝑦 ∶ 𝑦 ∉ 𝑦} ∉ {𝑦 ∶ 𝑦 ∉ 𝑦} ↔ meral for number 𝑛 in language 𝐴 .
{𝑦 ∶ 𝑦 ∉ 𝑦} ∈ {𝑦 ∶ 𝑦 ∉ 𝑦} This notation 𝑛 can be a bit confusing: what oc-
curs under the bar is a metalanguage expression
Proof: Direct from (NC) and universal instantiation.
for a number; we can use any mathematical nota-
Whence both F ⊢ {𝑦 ∶ 𝑦 ∉ 𝑦} ∈ {𝑦 ∶ 𝑦 ∉ 𝑦}, and
tion of regular mathematics there. But what the
F ⊢ {𝑦 ∶ 𝑦 ∉ 𝑦} ∉ {𝑦 ∶ 𝑦 ∉ 𝑦}.
whole notation 𝑛 including the overbar represents
Hence F ⊢ 𝜑 for all wffs 𝜑, making the system
is always an object-language numeral of the form
entirely unsuitable for mathematics. We have both
“0′′…(𝑛 times)…′ ”.
F ⊢ 1 + 1 = 2 and F ⊢ 1 + 1 = 3!
(And obviously, this is not meant as the set-
There have been many modifications in various
complement notation we used in F.)
set theories and higher-order logics which do gen-
Generally, expected results involving numerals
erate consistent theories that can “interpret” Peano
only (and not quantified variables) are obtainable
arithmetic, but none have the same intuitive appeal.
even in the weaker system Q, e.g.:
When I teach Mathematical Logic II, this is a main
theme. Q⊢1+1=2

1. 0′ + 0′ = (0′ + 0)′ Q4 UI
2. 0′ + 0 = 0′ Q3 UI
3. 0′ + 0′ = 0′′ 1, 2 LL
4. 1 + 1 = 2 3 def. 𝑛
Indeed, we can be more general:

41
 Proof: (1) By Num+ above, we have both:
Result (Num+): For any natural numbers 𝑛 Q⊢𝑛+𝑚=𝑛+𝑚
and 𝑚, Q ⊢ 𝑛 + 𝑚 = 𝑛 + 𝑚.
Q⊢𝑚+𝑛=𝑚+𝑛

(2) By the fact of commutativity of addition we

accept in the metalanguage, the numeral for
Proof: Notice the result does not follow by Ref= 𝑛 + 𝑚 is the same numeral as that for 𝑚 + 𝑛,
or anything similar. On the left side of the above, and so by symmetry and transitivity of identity,
we have the object-language “+” sign; on the right, Q ⊢ 𝑛 + 𝑚 = 𝑚 + 𝑛. ❅
we have the one from the metalanguage, and the
resulting numeral is the numeral for the sum. Of course, all theorems of Q are also theorems of
Although Q does not have object-language in- PA, and so the results above also hold for PA.
duction principles, we can still do induction in the It is interesting to contrast NumCom+ with our
metalanguage on the numbers used in Q’s numer- earlier result of Com+ for PA. NumCom+ holds for
als. We will prove this by induction on 𝑚. both Q and PA, but Com+ is not a theorem of Q.
(1) Let 𝑛 be an arbitrary natural number.
(2) When 𝑚 is zero, the numeral for 𝑛 + 𝑚 (𝑛 + 0) is (Com+)
A A
PA ⊢ 𝑥 𝑦 𝑥 + 𝑦 = 𝑦 + 𝑥
just the numeral for 𝑛, and we have Q ⊢ 𝑛+0 =
𝑛 + 0 by Q4 . The difference of course is that one uses object-
(3) As inductive hypothesis suppose for a given 𝑚 language variables, and one uses metalanguage
that we have Q ⊢ 𝑛 + 𝑚 = 𝑛 + 𝑚. We must now variable for object-language numerals, which are
show that Q ⊢ 𝑛 + 𝑚 + 1 = 𝑛 + (𝑚 + 1). not variables but closed terms.
(4) By Q5 and UI, Q ⊢ 𝑛 + 𝑚′ = (𝑛 + 𝑚)′ . This difference is typical of the difference be-
(5) The numeral for 𝑚 + 1 is 0 followed by 𝑚 + 1 tween PA and Q. In Q, one can typically prove all
many successor signs, which is the same as 𝑚′ , the instances of the universally quantified theorems
and so the above result is also Q ⊢ 𝑛 + 𝑚 + 1 = of PA obtained by induction by doing a “forced
(𝑛 + 𝑚)′ . march” through all the instances leading up to a
(6) By the inductive hypothesis and LL, Q ⊢ 𝑛 + given instance. But one cannot obtain that general
𝑚 + 1 = 𝑛 + 𝑚′ . result itself. Quite often, however, getting all the
(7) By the mathematical fact of associativity of ad- instances is enough to be useful.
dition we accept in the metalanguage, 𝑛+(𝑚+1) There are a few universal statements that can
′ be proven in Q alone, but only those that do not
is (𝑛 + 𝑚) + 1, the numeral for which is 𝑛 + 𝑚 .
(8) Hence our earlier result is the same as Q ⊢ require induction in the object language, e.g.:
𝑛 + 𝑚 + 1 = 𝑛 + (𝑚 + 1).
(≮0)
A
(9) Therefore, by mathematical induction, Q ⊢ 𝑛 + Q ⊢ 𝑥 ¬𝑥 < 0
𝑚 = 𝑛 + 𝑚 is true for any 𝑚, and also for any 𝑛, 1. 𝑏 < 0 Hyp
since 𝑛 was arbitrary. ❅ 2. z z′ + 𝑏 = 0
E
1, Q8 UI, BMP
3. 𝑐 + 𝑏 = 0
′
2 EI
4. 𝑏 = 0 Hyp
5. 𝑐′ + 0 = 0 3, 4 LL
Result (NumCom+): For any natural num- 6. 𝑐′ + 0 = 𝑐′ Q4 , UI
bers 𝑛 and 𝑚, Q ⊢ 𝑛 + 𝑚 = 𝑚 + 𝑛. 7. 0 = 𝑐 ′
5, 6 LL
8. 0 ≠ 𝑐 ′
Q2 , UI
9. ⊥ 7, 8 Contra
10. 𝑏 ≠ 0 4–9 RAA
11. y 𝑏 = y 10, Q3 , UI, MP
E ′

42
12. 𝑏 = 𝑑 ′ 11 EI (5) But Q ⊢ 0 ≠ 0′…however many…′ by Q2 and UI.
13. 𝑐′ + 𝑑 ′ = 0 3, 12 LL (6) Hence we get ⊥, and so by RAA, Q ⊢ 𝑛 ≠ 𝑚.
14. 𝑐′ + 𝑑 ′ = (𝑐′ + 𝑑)′ Q5 , UI For part (b):
15. 0 = (𝑐′ + 𝑑)′ 13, 14 LL (1) Suppose that Q ⊢ 𝑛 = 𝑚 but for reductio sup-
16. 0 ≠ (𝑐′ + 𝑑)′ Q2 , UI pose that 𝑛 and 𝑚 are different.
17. ⊥ 15, 16 Contra (2) By part (a), Q ⊢ 𝑛 ≠ 𝑚, and so Q ⊢ ⊥.
18. ¬𝑏 < 0 1–17 RAA (3) This impossible, since Q is consistent.
19. 𝑥 ¬𝑥 < 0 8 UG (4) Therefore it must be 𝑛 and 𝑚 are the same. ❅
A

The above also holds for PA, with the same proof.
Result (Num×): For any natural numbers 𝑛
and 𝑚, Q ⊢ 𝑛 × 𝑚 = 𝑛 × 𝑚.
Result (NumAltQ𝟓 ): For every natural num-
A
ber 𝑛, Q ⊢ 𝑥 𝑥 ′ + 𝑛 = (𝑥 + 𝑛)′ .

Homework 21
(a) Prove Num×. You may use induction in
the metalanguage (only) as well as appeal to
Proof: We prove this by induction on 𝑛 in the
metalanguage. For the base step, we show that
Q ⊢ 𝑥 𝑥 ′ + 0 = (𝑥 + 0)′ :
A

Num+. 1. 𝑏 + 0 = 𝑏 Q4 , UI
(b) Using induction (again, in the metalanguage) 2. 𝑏′ + 0 = 𝑏′ Q4 , UI
on the number of function signs they contain, 3. 𝑏 = 𝑏
′ ′
Ref=
prove that for all closed terms 𝑡 of the language 4. (𝑏 + 0) = 𝑏
′ ′
1, 3 LL
of arithmetic, 𝐴 , there is a numeral 𝑛 such 5. 𝑏′ + 0 = (𝑏 + 0)′ 2, 4 LL
that Q ⊢ 𝑡 = 𝑛. 6. 𝑥 𝑥 + 0 = (𝑥 + 0) 5 UG
A ′ ′

For the induction step, we can assume that Q ⊢

𝑥 𝑥 ′ + 𝑛 = (𝑥 + 𝑛)′ , and must show that Q ⊢
A
Result (Num=): For any natural numbers 𝑛 A
𝑥 𝑥 ′ + 𝑛 + 1 = (𝑥 + 𝑛 + 1)′ :
and 𝑚, (a) if 𝑛 is different from 𝑚, then Q ⊢
1. 𝑥 𝑥 ′ + 𝑛 = (𝑥 + 𝑛)′ Inductive hypothesis
A
𝑛 ≠ 𝑚; and (b) if Q ⊢ 𝑛 = 𝑚 then 𝑛 and 𝑚 are
in fact the same number. 2. 𝑏 + 𝑛 = (𝑏 + 𝑛)
′ ′
1 UI
3. 𝑏 + 𝑛′ = (𝑏 + 𝑛)′ Q5 , UI
4. 𝑏′ + 𝑛 = 𝑏 + 𝑛′ 2, 3 LL
5. (𝑏 + 𝑛) = (𝑏 + 𝑛)
′ ′ ′ ′
Ref=
Proof: For part (a): 6. (𝑏 ′
+ 𝑛) ′
= (𝑏 + 𝑛 ′ ′
) 4, 5 LL
(1) Suppose that 𝑛 and 𝑚 are different; then one 7. 𝑏′ + 𝑛′ = (𝑏 + 𝑛) Q5 , UI
′ ′ ′ ′

is larger than the other, and 𝑛 and 𝑚 contain a 8. 𝑏 + 𝑛 = (𝑏 + 𝑛 ′ ′

) 6, 7 LL
different number of “′ ” signs. 9. 𝑏 + 𝑛 + 1 = (𝑏 + 𝑛 + 1)
′ ′
Def. 𝑛 + 1
(2) In the object language, assume for RAA that 10. 𝑥 𝑥 + 𝑛 + 1 = (𝑥 + 𝑛 + 1) 9 UG
A ′ ′

𝑛 = 𝑚.
(3) The above means that we have 0′…𝑛 times…′ =
0′…𝑚 times…′ , where one side has more “′ ”s than
the other. Result (Spread): For every natural number 𝑛,
A ′
(4) By either 𝑛 or 𝑚 instances of Q1 and MP, de- Q ⊢ 𝑥(𝑥 < 𝑛 ↔ 𝑥 = 0 ∨ … ∨ 𝑥 = 𝑛).
pending on which is smaller, we get either
0′…(𝑛−𝑚 times)…′ = 0 or ⊢ 0 = 0′…(𝑚−𝑛 times)…′ .

43
Proof: Induction again. Base case: 19. 𝑑 ′ = 0′ 18, Ref=, LL
1. 𝑏 < 0 ′
Hyp 20. 𝑏 = 0 ′
10, 19 LL
2. z z′ + 𝑏 = 0′ 1, Q8 , UI, BMP 21. 𝑏 = 0 ∨ 𝑏 = 0′ ∨ . . . ∨ 𝑏 = 𝑛′ 20 Add
E
3. 𝑐 + 𝑏 = 0
′ ′
2 EI 22. 𝑑 = 0 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 ′
18–21 CP
4. 𝑏 ≠ 0 Hyp 23. 𝑑 = 0′ Hyp
5. y 𝑏 = y Q3 , UI, MP 24. 𝑑 = 0 18, Ref=, LL
E ′ ′ ′′

6. 𝑏 = 𝑑 ′
4 EI 25. 𝑏 = 0 ′′
10, 24 LL
7. 𝑐′ + 𝑑 ′ = 0′ 3, 6 LL 26. 𝑏 = 0 ∨ 𝑏 = 0′ ∨ 𝑏 = 0′′ ∨ . . . ∨ 𝑏 = 𝑛′ 25 Add
8. 𝑐 + 𝑑 = (𝑐 + 𝑑)
′ ′ ′ ′
Q5 , UI 27. 𝑑 = 0 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛
′ ′
23–26 CP
9. (𝑐′ + 𝑑)′ = 0′ 7, 8 LL ⋮
10. 𝑐′ + 𝑑 = 0 9, Q1 , UI, MP 28. 𝑑 = 𝑛 Hyp
11. z z + 𝑑 = 0 10 EG 29. 𝑑 = 𝑛 28, Ref=, LL
E ′ ′ ′

12. 𝑑 < 0 11, Q8 , UI, BMP 30. 𝑏 = 𝑛′ 10, 29 LL

13. ¬ 𝑑 < 0 ≮0, UI 31. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 ′
30 Add
14. ⊥ 12, 13 Contra 32. 𝑑 = 𝑛 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′ 28–31 CP
15. 𝑏 = 0 4–14 RAA 33. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′ 17, 22, 27, . . . , 32 Cases
16. 𝑏 < 0 → 𝑏 = 0
′
1–15 CP 34. 𝑏 ≠ 0 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′ 8–33 CP
17. 𝑏 = 0 Hyp 35. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′ 7, 34 Inev
18. 0′ + 0 = 0′ Q4 , UI 36. 𝑏 < 𝑛′′ → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′ 2–35 CP
19. z z + 0 = 0 18 EG 37. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 Hyp
E ′ ′ ′

20. 0 < 0′ 19, Q8 , UI, BMP 38. 𝑏 = 0 Hyp

21. 𝑏 < 0 ′
17, 20 LL 39. 𝑛 + 0 = 𝑛
′′ ′′
Q4 , UI
22. 𝑏 = 0 → 𝑏 < 0′ 17–21 CP 40. z z′ + 0 = 𝑛′′ 39 EG
E
23. 𝑏 < 0′ ↔ 𝑏 = 0 16, 22 BI 41. z z′ + 𝑏 = 𝑛′′ 38, 40 LL
E
24. 𝑥(𝑥 < 0 ↔ 𝑥 = 0) 23 UG 42. 𝑏 < 𝑛 41, Q8 , UI, BMP
A ′ ′′

43. 𝑏 = 0 → 𝑏 < 𝑛′′ 38–42 CP

For the induction step, we assume Q ⊢ 𝑥(𝑥 <
A
44. 𝑏 = 0 ′
Hyp
𝑛′ ↔ 𝑥 = 0 ∨ … ∨ 𝑥 = 𝑛) and need to show that
A ′
Q ⊢ 𝑥(𝑥 < 𝑛 + 1 ↔ 𝑥 = 0 ∨ … ∨ 𝑥 = 𝑛 + 1). 45. 𝑛 ′
+ 0 ′
= 𝑛 ′′
Num+
Ind. hyp. 46. z z′ + 0 = 𝑛′′ 45 EG
′ E ′ ′′
1. 𝑥(𝑥 < 𝑛′ ↔ 𝑥 = 0 ∨ … ∨ 𝑥 = 𝑛)
A
47. z z + 𝑏 = 𝑛 44, 46 LL
E
2. 𝑏 < 𝑛′′ Hyp
48. 𝑏 < 𝑛 ′′
47, Q 8 UI, BMP
,
3. z z′ + 𝑏 = 𝑛′′ 2, Q8 , UI, BMP
E
49. 𝑏 = 0 → 𝑏 < 𝑛
′ ′′
44–48 CP
4. 𝑐′ + 𝑏 = 𝑛′′ 3 EI
5. 𝑏 = 0 Hyp ⋮
5 Add 50. 𝑏′= 𝑛 ′ Hyp
′
6. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 ′

5–6 CP 51. 0 + ′𝑛 =′ 𝑛 ′′ Num+

′′
7. 𝑏 = 0 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛′
52. z z + 𝑛 = 𝑛 51 EG
E
8. 𝑏 ≠ 0 Hyp
53. z z + 𝑏 = 𝑛 50, 52 LL
′ E ′′
9. y 𝑏 = y′ 3, Q3 , UI, BMP
E
54. 𝑏 < 𝑛 ′′
53, Q8 , UI, BMP
10. 𝑏 = 𝑑 ′ 9 EI
11. 𝑐′ + 𝑑 ′ = 𝑛′′ 4, 10 LL 55. 𝑏 = 𝑛 ′
→ 𝑏 < 𝑛 ′′
50–54 CP
Q5 , UI 56. 𝑏 < 𝑛 37, 43, 49, . . . , 56 Cases
′′
12. 𝑐′ + 𝑑 ′ = (𝑐′ + 𝑑)′
13. (𝑐′ + 𝑑)′ = 𝑛′′ 11, 12 LL 57. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 ′
→ 𝑏 < 𝑛′′ 37–56 CP
58. 𝑏 < 𝑛 ↔ 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛
′′ ′
36, 57 BI
14. 𝑐′ + 𝑑 = 𝑛′ 13, Q1 , UI, MP
59. 𝑥(𝑥 < 𝑛 ↔ 𝑥 = 0 ∨ . . . ∨ 𝑥 = 𝑛 ) 58 UG
A ′′ ′
15. z z′ + 𝑑 = 𝑛′ 14 EG
E
′
60.
A
𝑥(𝑥 < 𝑛 + 1 ↔ 𝑥 = 0 ∨ . . . ∨ 𝑥 = 𝑛 + 1)
16. 𝑑 < 𝑛′ 15, Q8 , UI, BMP
17. 𝑑 = 0 ∨ . . . ∨ 𝑑 = 𝑛 1, 16, UI, BMP 59 Def. 𝑛 + 1
18. 𝑑 = 0 Hyp
(Spread) also holds for PA, with the same proof.

44
 21. 𝑐 = 𝑑 ′ 20 EI
22. 𝑑 + 𝑛 = 𝑏
′′
12, 21 LL
Result (Trichotomy): For every number 𝑛,
A 23. 𝑑 ′′ + 𝑛 = (𝑑 ′ + 𝑛)′ NumAltQ5 , UI
Q ⊢ 𝑦(𝑦 < 𝑛 ∨ 𝑛 < 𝑦 ∨ 𝑦 = 𝑛).
24. (𝑑 ′ + 𝑛)′ = 𝑏 22, 23 LL
25. 𝑑 ′ + 𝑛′ = (𝑑 ′ + 𝑛)′ Q5 , UI
26. 𝑑 + 𝑛 = 𝑏
′ ′
24, 25 LL
27. z z + 𝑛 = 𝑏 26 EG
E ′ ′
Proof: Induction in the metalanguage again. For 28. 𝑛′ < 𝑏 27, Q8 , UI, BMP
the base: 29. 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 28 Add
1. 𝑏 = 0 Hyp
30. 𝑐 ≠ 0 → 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 19–29 CP
2. 𝑏 < 0 ∨ 0 < 𝑏 ∨ 𝑏 = 0 1 Add31. 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 18, 30 Inev
3. 𝑏 = 0 → 𝑏 < 0 ∨ 0 < 𝑏 ∨ 𝑏 = 0 1–2 CP32. 𝑛 < 𝑏 → 𝑏 < 𝑛 ∨ 𝑛 < 𝑏 ∨ 𝑏 = 𝑛
′ ′ ′
10–31 CP
4. 𝑏 ≠ 0 Hyp
33. 𝑏 = 𝑛 Hyp
5. y 𝑏 = y 4, Q3 , UI, MP
E ′
34. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 33 Add
6. 𝑏 = 𝑐 ′ 5 EI
35. 𝑏 < 𝑛′ 2, 34 MP
7. 𝑐 + 0 = 𝑐
′ ′
Q4 , UI
36. 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 35 Add
8. 𝑐′ + 0 = 𝑏 6, 7 LL
37. 𝑏 = 𝑛 → 𝑏 < 𝑛 ∨ 𝑛 < 𝑏 ∨ 𝑏 = 𝑛
′ ′ ′
33–36 CP
9. z z′ + 0 = 𝑏 8 EG
E
38. 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 1, 9, 32, 37 Cases
10. 0 < 𝑏 9, Q8 , UI, BMP39. 𝑦(𝑦 < 𝑛 + 1 ∨ 𝑛 + 1 < 𝑦 ∨ 𝑦 = 𝑛 + 1)
A
11. 𝑏 < 0 ∨ 0 < 𝑏 ∨ 𝑏 = 0 10 Add 38, UG, Def. 𝑛 + 1
12. 𝑏 ≠ 0 → 𝑏 < 0 ∨ 0 < 𝑏 ∨ 𝑏 = 0 4–11 CP
13. 𝑏 < 0 ∨ 0 < 𝑏 ∨ 𝑏 = 0 3, 12 Inev In PA, we get not only (Trichotomy) but the object-
14. 𝑦(𝑦 < 0 ∨ 0 < 𝑦 ∨ 𝑦 = 0) 13 UG language quantified version:
A
A A
For the induction step, we assume Q ⊢ 𝑦(𝑦 <
A PA ⊢ 𝑥 𝑦(𝑦 < 𝑥 ∨ 𝑥 < 𝑦 ∨ 𝑦 = 𝑥)
𝑛 ∨ 𝑛 < 𝑦 ∨ 𝑦 = 𝑛) and proceed to show Q ⊢ by doing the induction in the object language in-
A
𝑦(𝑦 < 𝑛 + 1 ∨ 𝑛 + 1 < 𝑦 ∨ 𝑦 = 𝑛 + 1) stead.
1. 𝑏 < 𝑛 ∨ 𝑛 < 𝑏 ∨ 𝑏 = 𝑛 Ind. hyp., UI
2. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 → 𝑏 < 𝑛 ′
Spread, UI, BE
3. 𝑏 < 𝑛 → 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 − 1 Spread, UI, BE
5. Segue: Remaining
4. 𝑏 < 𝑛 Hyp Questions
5. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 − 1 3, 4 MP
6. 𝑏 = 0 ∨ . . . ∨ 𝑏 = 𝑛 − 1 ∨ 𝑏 = 𝑛 5 Add • Q obviously is not, but is PA sufficient to estab-
7. 𝑏 < 𝑛 ′
2, 6 MP lish everything true in the standard interpreta-
8. 𝑏 < 𝑛 ∨ 𝑛 < 𝑏 ∨ 𝑏 = 𝑛
′ ′ ′
7 Add tion, i.e., is PA the same as TA?
9. 𝑏 < 𝑛 → 𝑏 < 𝑛 ∨ 𝑛 < 𝑏 ∨ 𝑏 = 𝑛
′ ′ ′
4–8 CP
• Related to this, for every (closed) sentence 𝜑 in
10. 𝑛 < 𝑏 Hyp
𝐴 , is it the case that either PA ⊢ 𝜑 or PA ⊢ ¬𝜑,
11. z z′ + 𝑛 = 𝑏 10, Q8 , UI, BMP
E
as it is for TA?
12. 𝑐′ + 𝑛 = 𝑏 11 EI
13. 𝑐 = 0 Hyp • As noted before, TA “cheats” as—at least as we
14. 0′ + 𝑛 = 𝑛′ Num+ have described it—there is no independently ef-
15. 𝑐 + 𝑛 = 𝑛
′ ′
13, 14 LL fective way to decide whether or not something
16. 𝑏 = 𝑛 ′
12, 15 LL is allowed as an axiom.
17. 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 16 Add
• If the set PA doesn’t do it, is there any set that
18. 𝑐 = 0 → 𝑏 < 𝑛′ ∨ 𝑛′ < 𝑏 ∨ 𝑏 = 𝑛′ 13–17 CP
“axiomatizes” TA, where there is an effective pro-
19. 𝑐 ≠ 0 Hyp
cedure for deciding what is included and what
20. y 𝑐 = y′ 19, Q3 , UI, MP
E
isn’t?

45
• Is there an effective procedure for deciding what that take any natural numbers as arguments and
is a member of TA, or for that matter the (closed) only return natural numbers as values.
theories PA or Q (or for that matter, the closure Concepts applicable to numbers can often be
of ∅ under logical consequence, i.e., all logical extended to other things if there are ways of attach-
truths)? ing numbers to them, such as attaching a Gödel
numbering of formulas.
• What are the limits to which mathematical prop- For the moment, however, we are not consider-
erties, relations and functions can be represented ing any particular object language, and the notation
in a system such as Q or PA? What does it even in this section should just be taken as that of regular
mean for it to “represent” them anyway? mathematics, as employable in the metalanguage.
The following three sets of functions have been
Providing answers to these questions requires get- proven to be the same set:
ting serious about what we mean by an “effective
way” of deciding something. We now segue into • The set of (general) recursive functions on natural
talking about at least one attempt to make this no- numbers.
tion more precise: the theory of recursive functions. • The set of functions on natural numbers repre-
sentable in the 𝜆-calculus.
• The set of functions on natural numbers com-
6. Recursive Functions putable by a Turing machine.

Background Membership in any three of these is a tempting anal-

ysis of what it means for a function to be “effectively
Informally, we could say that a property or rela- decidable”. The fact that these are all provably the
tion is “effectively decidable” if there is a purely same arguably lends support to the following:
mechanical or rote way of determining whether
or not it holds for a given thing or given relata, Definition: The Church-Turing thesis is the the-
something that could be calculated or computed sis that a function is effectively computable if and
without any special ingenuity or creativity or room only if it is (general) recursive/𝜆-definable/Turing
for debate, using a process guaranteed to terminate computable.
after a finite number of steps.
Similarly, we could say that a function is “ef- Here we focus on the first notion.
fectively computable” if there is such a means for
determining its value for a given argument or ar- The basic/initial functions
guments. To bridge these:
Definition: The zero function is the constant
Definition: The characteristic function of n- function whose value for any natural number is 0,
place relation 𝑅, written 𝜒𝑅 , is the 𝑛-place function written:
that yields 1, if 𝑅 holds between its arguments, and zero(𝑥) = 0.
yields 0, otherwise.
{ Definition: The successor function is the func-
1, if 𝑅(𝑥0 , … , 𝑥𝑛−1 ) holds, tion is the function whose value for any natural num-
𝜒𝑅 (𝑥0 , … , 𝑥𝑛−1 ) = ber is the next natural number, written:
0, otherwise.
succ(𝑥) = one more than 𝑥.
It seems natural to think that a relation is effectively
decidable iff its characteristic function is effectively Note we do not write “𝑥 ′ ” here, as we are working
computable. in the regular mathematics of the metalanguage,
For the moment we are going to restrict our not the object language of Q or PA, and we don’t
focus to operations on natural numbers: functions want to confuse them.

46
Definition: The 𝑛-place projection function for Or, if 𝑘 is a constant natural number, and 𝑔 is a two-
position 𝑖 (where 0 ≤ 𝑖 ≤ (𝑛 − 1)) is the function place function, then the function defined by primitive
that takes 𝑛 arguments, and returns the 𝑖th argument recursion from 𝑘 and 𝑔 is the one-place function ℎ
(counting from 0), unchanged, written: such that:

𝑃𝑖𝑛 (𝑥0 , … , 𝑥𝑛−1 ) = 𝑥𝑖 . ℎ(0) = 𝑘

For example 𝑃13 (3, 6, 12) = 6, 𝑃01 (55) = 55, etc.
There are different project functions for different
values of 𝑛 and 𝑖.
Together, these functions are called the initial
or basic functions. Clearly, there seems to be an
“effective procedure” for calculating their values for
any given argument(s).
ℎ(𝑦 + 1) = 𝑔(𝑦, ℎ(𝑦)).
Recursive definitions are to definitions what induc-
tion is to proving things. We define its value for 0 as
the (last) argument, then this gives us a means for
determining the next value, and so on, where we
can “work our way up” to the value for any given
natural number.

Definition: If 𝑓 is (𝑛 + 1)-place function which

Methods of defining functions is such that for every set of natural numbers
𝑥0 , … , 𝑥𝑛−1 , there is always at least one number 𝑦
What ways of defining new functions in terms of
where 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = 0, then the function de-
others preserve “effective calculability”? The most
fined by minimization on 𝑓 is the 𝑛-place function
obvious involves feeding the values of one function
ℎ such that:
into another function as arguments. This is called
composition or substitution:
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = the least 𝑦 such that
Definition: If 𝑓 is an 𝑚-place function, and 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = 0
𝑔0 , … , 𝑔𝑚−1 are each 𝑛-place functions, then the com-
position of 𝑓 with 𝑔0 , … , 𝑔𝑚−1 is the 𝑛-place function This method is also called “unbounded search” or
ℎ such that: the “choice of least rule”.
This one is a bit more involved. If we know how
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = to calculate 𝑓 , we can calculate ℎ for arguments
𝑓 (𝑔0 (𝑥0 , … , 𝑥𝑛−1 ), … , 𝑔𝑚−1 (𝑥0 , … , 𝑥𝑛−1 )). 𝑥0 , … , 𝑥𝑛−1 by calculating 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 0) and then
𝑓 (𝑥0 , … , 𝑥𝑛−1 , 1) and so on until we find the first one
that returns 0, which is then the value of ℎ for those
In the simplest case (𝑛 = 𝑚 = 1) we have ℎ(𝑥) = arguments.
𝑓 (𝑔(𝑥)), which is sometimes also written (𝑓 ◦ 𝑔)(𝑥).
If we know how to calculate 𝑓 ’s and the 𝑔-
functions’ values, then to calculate ℎ, we merely The set of (primitive) recursive
calculate the 𝑔’s for the same arguments, and then functions
use those values as arguments to 𝑓 .
Definition: The set of primitive recursive func-
Definition: If 𝑓 is an 𝑛-place function, and 𝑔 is an tions is the smallest set of functions containing the
(𝑛 + 2)-place function, then the function defined by zero function, the successor function, the projection
primitive recursion from 𝑓 and 𝑔 is the (𝑛 + 1)- functions, and is closed under composition and prim-
place function function ℎ such that: itive recursion.

ℎ(𝑥0 , … , 𝑥𝑛−1 , 0) = 𝑓 (𝑥0 , … , 𝑥𝑛−1 ), and Or put another way, a function is primitive recur-
ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦 + 1) = sive iff:
𝑔(𝑥0 , … , 𝑥𝑛−1 , 𝑦, ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦)), (1) it is one of the initial functions (zero, succ, 𝑃𝑖𝑛 );
or

47
(2) it can be defined by a finite series of steps of
recursion and/or composition from functions
already shown to be primitive recursive.
Definition: The set of (general) recursive func-
tions is the smallest set of functions containing the
zero function, the successor function, the projection
functions and is closed under composition, primitive
recursion and minimization.
A function is general recursive if it can be defined
from other recursive functions, starting with the
initial functions, using a finite series of steps of the
methods of composition, primitive recursion and
minimization.
All primitive recursive functions are also gen-
eral recursive; the reverse is not true, but we shall
not bother with the proofs of this in this course.
Definition: Relation 𝑅 is (primitive) recursive
relation iff its characteristic function is (primitive)
recursive. (A property can be considered a 1-place
relation, this also defines the notion of “(primitive)
recursive property.”)
Some convenience results
Result (dummy arguments): If the 𝑛-place
function 𝑓 is (primitive) recursive, then
so is the (𝑛 + 1)-place function ℎ, whose
value, ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑥𝑛 ), is always simply
𝑓 (𝑥0 , … , 𝑥𝑛−1 ), so that the last argument to ℎ is
always simply ignored.
Proof: Function ℎ can be defined by composition
using 𝑓 and the projection functions:
ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑥𝑛 ) = 𝑓 (𝑃0𝑛+1 (𝑥0 , … , 𝑥𝑛 ),
𝑛+1
… , 𝑃𝑛−1 (𝑥0 , … , 𝑥𝑛 )).
Result (permuting arguments): If 𝑛-place
function 𝑓 is (primitive) recursive, then
so is the 𝑛-place function ℎ whose value,
ℎ(… , 𝑥𝑖 , … , 𝑥𝑗 , … ), is always 𝑓 (… , 𝑥𝑗 , … , 𝑥𝑖 , … ).
Proof: Again, using composition and projection:
ℎ(… , 𝑥𝑖 , … , 𝑥𝑗 , … ) = 𝑓 (… , 𝑃𝑗𝑛 (𝑥0 , … , 𝑥𝑛−1 ),
… , 𝑃𝑖𝑛 (𝑥0 , … , 𝑥𝑛−1 ), … ). ❅
Result (identifying arguments): If the (𝑛 +
1)-place function 𝑓 is (primitive) recursive,
then so is 𝑛-place function ℎ whose value,
ℎ(𝑥0 , … , 𝑥𝑛−1 ) is always 𝑓 (𝑥0 , 𝑥0 , … , 𝑥𝑛−1 ).
Proof: Our method is similar to the above:
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = 𝑓 (𝑃0𝑛 (𝑥0 , … , 𝑥𝑛−1 ),
𝑃0𝑛 (𝑥0 , … , 𝑥𝑛−1 ), … , 𝑃𝑛−1
𝑛
(𝑥0 , … , 𝑥𝑛−1 )). ❅
Result (all polyadic zero functions): For
any 𝑛, the 𝑛-place zero function zero𝑛 is
primitive recursive.
Proof: This follows by composition, since:
zero𝑛 (𝑥0 , … , 𝑥𝑛−1 ) = zero(𝑃0𝑛 (𝑥0 , … , 𝑥𝑛−1 )) ❅
Result (all constant functions): For any 𝑛
and 𝑘, the 𝑛-place constant function const𝑛𝑘 , the
value of which for any 𝑛-arguments, is always 𝑘
❅
regardless of what the arguments are, is primitive
recursive.
48
Proof: This can be proven by induction on 𝑘. For (i) For any 𝑛 > 2, the minimum of 𝑛 arguments,
𝑘 = 0, the 𝑛-place constant function is the same as because each such function can be defined by
the 𝑛-place zero function. For the rest, the 𝑛-place substitution using the previous one:
constant function whose value is always 𝑘 + 1 can min(𝑥0 , … , 𝑥𝑛−1 , 𝑥𝑛 ) =
be defined by composition since: min(min(𝑥0 , … , 𝑥𝑛−1 ), 𝑥𝑛 )
(j) Maximum of 2 (or more) arguments:
const𝑛𝑘+1 (𝑥0 , … , 𝑥𝑛−1 ) = succ(const𝑛𝑘 (𝑥0 , … , 𝑥𝑛−1 )).
max(𝑥, 𝑦) = 𝑦 + (𝑥 ∸ 𝑦)
❅
max(𝑥0 , … , 𝑥𝑛 ) = max(max(𝑥0 , … , 𝑥𝑛−1 ), 𝑥𝑛 )
The upshot of the above results is that we need not (k) Characteristic function of not being zero:
be fussy when giving definitions using composi- 𝜒NotZero (𝑥) = 𝑥 ∸ pred(𝑥)
tion, recursion, etc., about the order and number (l) Characteristic function of being zero:
of arguments, and about whether or not we use 𝜒IsZero (𝑥) = 1 ∸ 𝜒NotZero (𝑥)
a constant value where a function is technically (m) Remainder upon division of 𝑦 by 𝑥: rem(𝑥, 𝑦).
needed. Recursion:
rem(𝑥, 0) = 0
rem(𝑥, 𝑦 + 1) =
Examples succ(rem(𝑥, 𝑦))×𝜒NotZero (|𝑥 −succ(rem(𝑥, 𝑦))|)
(n) Quotient upon division of 𝑦 by 𝑥: d(𝑥, 𝑦).
(Rounded down.) Recursion:
Result: The functions below are primitive re-
d(𝑥, 0) = 0
cursive.
d(𝑥, 𝑦 + 1) =
d(𝑥, 𝑦) + 𝜒IsZero (|𝑥 − succ(rem(𝑥, 𝑦))|)
Definition: The graph of 𝑛-place function 𝑓 , writ-
(a) Addition: 𝑥 + 𝑦. Definable by recursion:
ten 𝛾𝑓 , is the (𝑛 + 1)-place relation that holds between
𝑥 + 0 = 𝑃01 (𝑥) = 𝑥
𝑥0 , … , 𝑥𝑛−1 , 𝑥𝑛 iff 𝑓 (𝑥0 , … , 𝑥𝑛−1 ) = 𝑥𝑛 .
𝑥 + (𝑦 + 1) = succ(𝑥 + 𝑦)
(b) Multiplication: 𝑥 × 𝑦. Recursion again:
𝑥 × 0 = zero(𝑥) = 0
𝑥 × (𝑦 + 1) = (𝑥 × 𝑦) + 𝑥 Result: If 𝑓 is (primitive) recursive function,
(c) 𝑥 to the power of 𝑦: 𝑥 𝑦 . Recursion: then its graph, 𝛾𝑓 , is a recursive relation.
𝑥 0 = const11 (𝑥) = 1
𝑥 𝑦+1 = (𝑥 𝑦 ) × 𝑥
(d) Predecessor: pred(𝑥). Recursion:
Proof: The characteristic function 𝜒𝛾𝑓 of 𝛾𝑓 can be
pred(0) = 0
defined in terms of 𝑓 as follows:
pred(𝑦 + 1) = 𝑃02 (𝑦, pred(𝑦)) = 𝑦
(e) Subtract-as-much-as-you-can: 𝑥 ∸ 𝑦. Recur- 𝜒𝛾𝑓 = 𝜒IsZero (|𝑥𝑛 − 𝑓 (𝑥0 , … , 𝑥𝑛−1 )|). ❅
sion:
𝑥 ∸0=𝑥
Homework 22
𝑥 ∸ (𝑦 + 1) = pred(𝑥 ∸ 𝑦)
(f) Absolute difference: |𝑥 − 𝑦| . Composition: Show that if 𝛾𝑓 is a recursive relation then 𝑓 is
|𝑥 − 𝑦| = (𝑥 ∸ 𝑦) + (𝑦 ∸ 𝑥) recursive function. Hint: use minimization.
(g) Factorial: 𝑥! Recursion:
0! = 1 Bounded sums and products
(𝑦 + 1)! = 𝑦! × (𝑦 + 1)
(h) Minimum of 2 arguments: min(𝑥, 𝑦). Compo- The following notation:
sition: ∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧)
min(𝑥, 𝑦) = 𝑥 ∸ (𝑥 ∸ 𝑦) 𝑧<𝑦

49
stands for the (𝑛 + 1)-place bounded sum function Recursive properties and relations
ℎ, whose value for 𝑥0 , … , 𝑥𝑛−1 , 𝑦 as arguments is the
sum of all the values of 𝑓 for 𝑥0 , … , 𝑥𝑛−1 , 0 through
𝑥0 , … , 𝑥𝑛−1 , 𝑦 − 1. Result: The equality, less-than-or-equal-to and
evenly-dividing-into relations are primitive re-
cursive.
Result: If 𝑓 is (primitive) recursive, then so is
the bounded sum function ℎ, as explained above.

Proof:

Proof: The function ℎ can be defined by recursion 𝜒= (𝑥, 𝑦) = 𝜒IsZero (|𝑥 − 𝑦|),
as follows: 𝜒≤ (𝑥, 𝑦) = 𝜒IsZero (𝑥 ∸ 𝑦),
ℎ(𝑥0 , … , 𝑥𝑛−1 , 0) = 0 𝜒| (𝑥, 𝑦) = 𝜒IsZero (rem(𝑥, 𝑦)).
ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦 + 1) = ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦) +
❅
𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) ❅
Similar results follow for the form: Definition: The negation of a relation R, viz.,
∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧) “not-𝑅”, is the relation that holds of 𝑥0 , … , 𝑥𝑛−1 iff 𝑅
𝑧≤𝑦 does not hold of 𝑥0 , … , 𝑥𝑛−1 .
This is definable by composition, since: Definition: The conjunction of relations R and
∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧) = ∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧) S, written “𝑅-and-𝑆”, is the relation that holds of
𝑧≤𝑦 𝑧<𝑦+1 𝑥0 , … , 𝑥𝑛−1 iff 𝑅 and 𝑆 both hold of 𝑥0 , … , 𝑥𝑛−1 .
Similarly for doubly bounded sums: Definition: The disjunction of two relations R
and S, written “𝑅-or-𝑆”, is the relation that holds for
∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧) =
𝑦<𝑧<𝑣
𝑥0 , … , 𝑥𝑛−1 iff either 𝑅 or 𝑆 hold of 𝑥0 , … , 𝑥𝑛−1 .
∑ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧 + 𝑦 + 1)
𝑧<pred(𝑣∸𝑦) (And so on for other propositional connectives.)
The following notation:
∏ 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑧) Result (propositional connectives):
𝑧<𝑦 If 𝑅 and 𝑆 are (primitive) recursive relations, then
stands for the (𝑛+1)-place bounded product func- so are their negations, conjunctions, disjunctions,
tion ℎ, whose value for 𝑥0 , … , 𝑥𝑛−1 , 𝑦 as arguments and so on.
is the product of all the values of 𝑓 for 𝑥0 , … , 𝑥𝑛−1 , 0
through 𝑥0 , … , 𝑥𝑛−1 , 𝑦 − 1.
Proof: By definition, if 𝑅 and 𝑆 are (primitive) re-
Result: If 𝑓 is (primitive) recursive, so is the cursive, their characteristic functions 𝜒𝑅 and 𝜒𝑆 are
bonded product function ℎ, as defined above. (primitive) recursive, in which case the character-
istic functions of their negations and disjunctions
can be defined as follows:
𝜒not-𝑅 (𝑥0 , … , 𝑥𝑛−1 ) = 1 ∸ 𝜒𝑅 (𝑥0 , … , 𝑥𝑛−1 )
Homework 23
𝜒𝑅-and-𝑆 (𝑥0 , … , 𝑥𝑛−1 ) =
Prove the result above. 𝜒𝑅 (𝑥0 , … , 𝑥𝑛−1 ) × 𝜒𝑆 (𝑥0 , … , 𝑥𝑛−1 )
Similar results follow for bounded products for 𝑧 ≤ Other propositional operations on relations can be
𝑦, as well as doubly bounded products (𝑦 < 𝑧 < 𝑣). defined in terms of conjunction and negation. ❅

50
Example: 𝜒< (𝑥, 𝑦) = 𝜒≤ (𝑥, 𝑦) × (1 ∸ 𝜒= (𝑥, 𝑦)).
I use the notation: Result (unbounded search operator): If 𝑅
𝑧 is a recursive relation such that for every
∀ 𝑅(𝑥 , … , 𝑥
𝑧<𝑦
0 𝑛−1 , 𝑧) set of numbers 𝑥0 , … , 𝑥𝑛−1 , there is 𝑦 such
that 𝑥0 , … , 𝑥𝑛−1 , 𝑦 holds for some 𝑦, then
in the metalanguage(!) to stand for the relation (𝜇𝑦)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑦) is a recursive function.
𝑄 that holds for 𝑥0 , … , 𝑥𝑛−1 , 𝑦 iff the relation 𝑅
holds for all arguments 𝑥0 , … , 𝑥𝑛−1 , 𝑧 for every
𝑧 < 𝑦. (The Open Logic Text writes instead
“( 𝑧 < 𝑦)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑧)”, but I find this too close Proof: Suppose 𝑅 is such a relation. By our earlier
A
to the notation used in the object language, and result, so is not-𝑅, and its characteristic function
potentially confusing.) 𝜒not-𝑅 will return 0 for some 𝑦 for every 𝑥0 , … , 𝑥𝑛−1 .
Hence, we can define (𝜇𝑦)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑦) using
minimization as yielding the least 𝑦 such that
Result (bounded quantification): If 𝑅 is a 𝜒not-𝑅 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = 0. ❅
(primitive) recursive number theoretic relation,
Note that if we use the “unbounded” search op-
so is the universally quantified relation 𝑄 as an-
erator “(𝜇𝑥)” when defining functions, the results
notated above.
are only guaranteed to be recursive, not primitive
recursive, even if the relation used in the definition
is primitive recursive.
Proof: The characteristic function for 𝑄 can be de- However, if we can define the function using a
fined in terms of the characteristic function for 𝑅 bounded search operator instead, the result will be
by substitution as follows: primitive recursive if the relation is.

𝜒𝑄 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = ∏ 𝜒𝑅 (𝑥0 , … , 𝑥𝑛−1 , 𝑧). ❅

𝑧<𝑦 Result (bounded search operator): If 𝑅 is a
(primitive) recursive relation, then the function:
Homework 24 (𝜇𝑦 < 𝑧)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑦)
Prove that 𝑧
whose value for 𝑥0 , … , 𝑥𝑛−1 , 𝑧 as arguments, is
∃ 𝑅(𝑥 , … , 𝑥
𝑧<𝑦
0 𝑛−1 , 𝑧)
the least 𝑦 less than 𝑧 such that 𝑅(𝑥0 , … , 𝑥𝑛 , 𝑦)
(where this means what you expect) is (primitive) holds, and whose value is 𝑧 if there is no such 𝑦,
recursive if 𝑅 is. is a (primitive) recursive function

Similar results follow for bounded quantifiers using

≤, doubly bounded quantifiers, etc.
The notation: Proof: The function can be defined using the char-
acteristic function of not-𝑅 by composition as fol-
(𝜇𝑦)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑦)
lows:
represents the 𝑛-place function whose value for
𝑥0 , … , 𝑥𝑛−1 as arguments returns the least 𝑦 such (𝜇𝑦 < 𝑧)𝑅(𝑥0 , … , 𝑥𝑛−1 , 𝑦) =
that the (𝑛 + 1)-place relation 𝑅 holds between
𝑥0 , … , 𝑥𝑛−1 and 𝑦. ∑ ∏ 𝜒not-𝑅 (𝑥0 , … , 𝑥𝑛−1 , 𝑤) . ❅
𝑦<𝑧 ( 𝑤≤𝑦 )

51
As 𝑧 increases, the bounded product will keep
adding 1 to the bounded sum so long as 𝑅 does
Result: The functions, properties and relations
not hold for any 𝑥0 , … , 𝑥𝑛−1 , 𝑤 where 𝑤 ≤ 𝑧. As
below are primitive recursive.
soon as a 𝑧 is reached for which 𝑅 does hold for
some 𝑥0 , … , 𝑥𝑛−1 , 𝑤 where 𝑤 ≤ 𝑧, the bounded prod-
uct will stop adding to the bounded sum, and so
the result will identical to the least such 𝑧.
Notice the above proof does not make use of (a) Being prime:
minimization, and so something defined in this way prime(𝑥) = (2 ≤ 𝑥)-and-
𝑦
will be primitive recursive assuming 𝑅 is. ∀ (if 𝑦|𝑥 then 𝑦 = 1 or 𝑦 = 𝑥)
𝑦≤𝑥
(b) Lowest prime above 𝑥:
Result (conditional definitions): nextPrime(𝑥) =
If 𝑅0 , … , 𝑅𝑚−1 are exclusive (primitive) recursive (𝜇𝑦 ≤ 𝑥! + 1)((𝑥 < 𝑦)-and-prime(𝑦))
th
relations, and 𝑓0 , … , 𝑓𝑚 are (primitive) recursive (c) The 𝑥 prime, starting from 0. (Recursion.)
functions, and function ℎ is such that 𝑝0 = 2
𝑝𝑦+1 = nextPrime(𝑝𝑦 )
⎧
⎪
⎪ 𝑓0 (𝑥0 , … , 𝑥𝑛−1 ), if 𝑅0 (𝑥0 , … , 𝑥𝑛−1 ), (d) The length of the sequence coded by 𝑥:
⎪ ⎧
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = ⎨ ⋮ ⎪
⎪ 0, if 𝑥 = 0 or 𝑥 = 1
⎪
⎪ ⎪
⎩𝑓𝑚 (𝑥0 , … , 𝑥𝑛−1 ), otherwise.
⎪ len(𝑥) = ⎨1 + (𝜇𝑦 < 𝑥)(𝑝𝑦 |𝑥 and not-(𝑝𝑦+1 |𝑥)),
⎪
⎪
then ℎ is also (primitive) recursive. ⎪
⎩ otherwise.
(e) The code of the sequence appending number
𝑦 to the sequence { encoded by 𝑥:
2𝑦+1 , if 𝑥 = 0 or 𝑥 = 1,
Proof: append(𝑥, 𝑦) =
𝑥 × (𝑝len(𝑥) )𝑦+1 , otherwise.
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = (f) For any 𝑛 ≥ 1, the code of 𝑥1 , … , 𝑥𝑛 :
(𝑓0 (𝑥0 , … , 𝑥𝑛−1 ) × 𝜒𝑅0 (𝑥0 , … , 𝑥𝑛−1 )) + … + code1 (𝑥1 ) = append(0, 𝑥1 )
code𝑛+1 (𝑥1 , … , 𝑥𝑛 , 𝑥𝑛+1 ) =
(𝑓𝑚 (𝑥0 , … , 𝑥𝑛−1 ) × 𝜒not-(𝑅0 -or-. . . -or-𝑅𝑚−1 ) (𝑥0 , … , 𝑥𝑛−1 )).
append(code𝑛 (𝑥1 , … , 𝑥𝑛 ), 𝑥𝑛+1 )
❅
(g) The 𝑦 th{member of the series encoded by 𝑥:
From here on out, we often simply give direct defini- 0, if len(𝑥) < 𝑦,
(𝑥)𝑦 =
tions of properties and relations in terms of notions (𝜇𝑧 < 𝑥) not-(𝑝𝑦𝑧+2 |𝑥), otherwise.
already shown to be (primitive) recursive, rather
(h) The code of the sequence resulting from con-
than defining their characteristic functions. The
catenating the first 𝑧 elements of the sequence
transitions are usually not difficult.
encoded by 𝑦 to the sequence encoded by 𝑥:
hconcat(𝑥, 𝑦, 0) = 𝑥
Coding functions hconcat(𝑥, 𝑦, 𝑧 + 1) =
append(hconcat(𝑥, 𝑦, 𝑧), (𝑦)𝑧 )
Recall that the “code” for a given sequence of num-
bers 𝑘0 , … , 𝑘𝑛−1 is the product of the first 𝑛 primes (i) The code of the sequence which concatenates
raised to the power one greater than each of these the full sequence encoded by 𝑦 to the end of
numbers: sequence encoded by 𝑥:
𝑥 ⌢ 𝑦 = hconcat(𝑥, 𝑦, len(𝑦))
2𝑘0 +1 × 3𝑘1 +1 × … × (𝑝𝑛−1 )𝑘𝑛−1 +1 .

52
Course-of-values recursion This function cannot be defined by the recursion
rule in a simple way, because its value for 𝑦 + 1
Because a single number can be used to encode a
depends not only on its value for 𝑦 but also on its
finite sequence of numbers, it is possible to define a
value for 𝑦 − 1.
function whose value for 𝑥 as argument encodes the
However, fib# can be defined by recursion as
sequence (“course”) of values of another function
follows:
for all arguments leading up to and including 𝑥.
If 𝑓 is a (𝑛 + 1)-place operation on natural num- fib#(0) = 0
bers, then the notation “𝑓 #” is used for the (𝑛 + 1)- ⎧
⎪
⎪ 22 , if 𝑦 = 0,
place number-theoretic function whose value for ⎪
⎪
𝑥0 , … , 𝑥𝑛−1 , 𝑦 is the number that encodes the series ⎪22 ⋅ 32 , if 𝑦 = 1,
fib#(𝑦 + 1) = ⎨
of values for 𝑓 for all sets of arguments starting ⎪
⎪ append(fib#(𝑦), (fib#(𝑦))𝑦∸2 +
⎪
from 𝑥0 , … , 𝑥𝑛−1 , 0 and ending with 𝑥0 , … , 𝑥𝑛−1 , 𝑦 −1. ⎪
⎪ (fib#(𝑦))𝑦∸1 ), otherwise.
⎩
Since fib# is primitive recursive, and fib can be
obtained from it by composition:
Result: A function 𝑓 is (primitive) recursive iff
𝑓 # is (primitive) recursive. fib(𝑥) = (fib#(𝑥 + 1))𝑥
The function fib is also primitive recursive.
Similar results follow for other functions one would
Proof: We prove this in both directions. want to define in a similar sort of way. Generally,
(a) Suppose that 𝑓 has already been shown to be we can say that if a function 𝑓 is obtained from
(primitive) recursive. One can then obtain 𝑓 # (primitive) recursive functions by course-of-values
by composition as follows: recursion, then 𝑓 is (primitive) recursive itself.
Course-of-values recursion is to simple recur-
𝑓 #(𝑥0 , … , 𝑥𝑛−1 , 𝑦) = ∏(𝑝𝑧 )𝑓 (𝑥0 ,…𝑥𝑛−1 ,𝑧)+1 sion what strong induction is to weak induction.
𝑧<𝑦
Homework 25
(b) On the other hand, suppose that 𝑓 # has already Show that the function 𝑔, described below, is prim-
been shown to be (primitive) recursive. One itive recursive, by giving an explicit definition of
can then obtain 𝑓 as follows: 𝑔#:
𝑓 (𝑥0 , … 𝑥𝑛−1 , 𝑦) = (𝑓 #(𝑥1 , … , 𝑥𝑛 , 𝑦 + 1))𝑦 ❅ 𝑔(0) = 2 and 𝑔(1) = 4 and
Sometimes it is easier to define 𝑓 # recursively than 𝑔(𝑥 + 2) = (3 × 𝑔(𝑥 + 1)) − ((2 × 𝑔(𝑥)) + 1)
it is to define 𝑓 , especially, a function whose value From now on, we will not fuss about recursive def-
for a given number depends not only upon its value initions always only appealing to the value for the
for the previous number, but upon more than one or previous argument, but also allow such definitions
even all of its prior values. Such functions are said that appeal to the value for any lesser argument.
to be obtained by course-of-values recursion,
rather than simple recursion.
7. Representing Recursive
Example: Consider fib(𝑥) whose value for any 𝑥 is
the 𝑥th item in the Fibonacci sequence, which Functions in Q/PA
adds the previous two members to get the next
(staring with 1, 1): Definitions
We now return to our discussion of formal systems.
1, 1, 2, 3, 5, 8, 13, 21, 34, 55, … , and so on.
Our eventual goal to prove that the functions that

53
can be represented in systems Q and PA are pre- and so Q ⊢ 𝑘 = 𝑘 ∧ 0 = 𝑚 by two appeals to Ref=
cisely the recursive functions. and conjoining. Moreover, Q ⊢ y(𝑘 = 𝑘 ∧ 0 =
A
First we need to make the notion of repre- y → 𝑚 = y), i.e., Q ⊢ y(𝑘 = 𝑘 ∧ 0 = y → 0 = y),
A
sentability precise. by generalizing on the tautology 𝑘 = 𝑘 ∧ 0 = 𝑏 →
0 = 𝑏. ❅
Definition: If T is a theory, and 𝑓 is an 𝑛-place
operation on the natural numbers, then 𝑓 is repre-
sentable in T iff there an open formula with 𝑛+1 free
Result: The successor function is representable
variables, 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦), such that, for any set of nat-
in Q.
ural numbers 𝑘1 , … , 𝑘𝑛 , 𝑚, where 𝑓 (𝑘1 , … , 𝑘𝑛 ) = 𝑚,
we have:
(a) T ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 , 𝑚); and
(b) T ⊢ 𝑦(𝜑(𝑘1 , … , 𝑘𝑛 , 𝑦) → 𝑚 = 𝑦).
A
Proof: Let 𝜑(x, y) be y = x′ . Suppose that succ(𝑘) =
𝑚. This means that 𝑚 is one more than 𝑘 and the
Note 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦) need not take the form 𝑦 = numeral 𝑚 is 𝑘 ′ . Hence Q ⊢ 𝑚 = 𝑘 ′ by Ref=, and
𝑓 (𝑥1 , … , 𝑥𝑛 ) for some object-language function let- Q ⊢ y(y = 𝑘 ′ → y = 𝑚) by generalizing on a
A
ter 𝑓 (though it could). tautology. ❅
Definition: If T is a theory, and 𝑅 is an 𝑛-place
relation between natural numbers, then 𝑅 is repre- Homework 26
sentable in T iff there an open formula with 𝑛 free Prove that addition and multiplication are repre-
variables, 𝜑(𝑥1 , … , 𝑥𝑛 ), such that, for any set of nat- sentable in Q.
ural numbers 𝑘1 , … , 𝑘𝑛 , . . . :
(a) If 𝑅(𝑘1 , … , 𝑘𝑛 ) then T ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 ); and
(b) If not-𝑅(𝑘1 , … , 𝑘𝑛 ), then T ⊢ ¬𝜑(𝑘1 , … , 𝑘𝑛 ). Result: For any 𝑛 and 𝑖, the projection function
𝑃𝑖𝑛 is representable in Q.

Result: Every function and relation repre-

sentable in Q is also representable in PA.
Proof: Let 𝜑 be the formula x = x ∧ x1 = x1 ∧ … ∧
x𝑛−1 = x𝑛−1 ∧ x𝑖 = y. Now suppose 𝑃𝑖𝑛 (𝑘0 , … , 𝑘𝑛−1 ) =
𝑚. This means that 𝑚 = 𝑘𝑖 , and so they have the
same numeral. Hence we have Q ⊢ 𝑘0 = 𝑘0 ∧
Proof: Since every theorem of Q is also a theorem
… ∧ 𝑘𝑛−1 = 𝑘𝑛−1 ∧ 𝑘𝑖 = 𝑚 by Ref= and conjoining.
of PA, the results needed to obtain that a given
Similarly, Q ⊢ y(𝑘0 = 𝑘0 ∧ … ∧ 𝑘𝑛−1 = 𝑘𝑛−1 ∧ 𝑘𝑖 =
A
function or relation is represented in Q by a given
y → 𝑚 = y) by generalizing on a tautology. ❅
formula are also obtainable in PA for the same
formula. ❅
Representable relations
The initial functions
Result: The equality relation is representable
in Q.
Result: The zero function is representable in Q.

Proof: Consider the wff x1 = x2 . If 𝑘1 is equal to 𝑘2 ,

Proof: Let 𝜑(x, y) be the formula x = x ∧ 0 = y. they have the same numeral, and so Q ⊢ 𝑘1 = 𝑘2 by
Suppose zero(𝑘) = 𝑚. Then 𝑚 is zero, and 𝑚 is “0”, Ref=. If 𝑘1 is not 𝑘2 , then by Num=, Q ⊢ 𝑘1 ≠ 𝑘2 . ❅

54
 We already have Q ⊢ 𝜓(𝑘1 , … , 𝑘𝑛 ), and
so we can eliminate the second disjunct.
Result: The less-than relation is representable
Thus we get 𝑏 = 1, i.e., 𝑏 = 𝑚, and so by
in Q.
symmetry of identity 𝑚 = 𝑏. Discharging
and generalizing Q ⊢ 𝑦((𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑦 =
A
1) ∨ (¬𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑦 = 0) → 𝑚 = 𝑦).
(6) The proof for the second case is just like the
Proof: Consider the wff x1 < x2 . Suppose 𝑘1 is less first, with the roles of the two conjuncts of
than 𝑘2 . Then there is some non-zero number 𝑗 𝜑(𝑥1 , … , 𝑥𝑛 ) reversed. ❅
such that 𝑗 + 𝑘1 = 𝑘2 . By Num+, Q ⊢ 𝑗 + 𝑘1 = 𝑘2 .
Since 𝑗 is not zero, the numeral for 𝑗 is of the form
′
𝑗 − 1 , and so by EG, Q8 and BMP, Q ⊢ 𝑘1 < 𝑘2 .
Suppose that 𝑘1 is not less than 𝑘2 . Either 𝑘2 is 0,
Result: If the characteristic function 𝜒𝑅 of rela-
or it isn’t. If it is, then we have Q ⊢ ¬𝑘1 < 𝑘2 from
tion 𝑅 is representable in Q, then so is 𝑅.
(≮0). Suppose 𝑘2 is greater than 0. We know that 𝑘1
is distinct from every number less than 𝑘2 , so, by
Num=:

Q ⊢ 𝑘1 ≠ 0
Homework 27
Q ⊢ 𝑘1 ≠ 1
Prove the result above. You may use the fact
⋮ (from Q2 ) that Q ⊢ 0 ≠ 1. Hint: if 𝜒𝑅 is repre-
Q ⊢ 𝑘1 ≠ 𝑘2 − 1 sented by 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦), then 𝑅 is represented by
𝜑(𝑥1 , … , 𝑥𝑛 , 1); show this.
Therefore, Q ⊢ ¬(𝑘1 = 0 ∨ 𝑘1 = 1 ∨ … ∨ 𝑘1 = 𝑘2 − 1),
and so by (Spread) and BMT, Q ⊢ ¬𝑘1 < 𝑘2 . ❅
Composition and minimization

Result: If relation 𝑅 is representable in Q, then Result: The set of functions representable in Q

so is its characteristic function 𝜒𝑅 . (or any other theory built on identity logic) is
closed under composition, i.e., if 𝑓 , 𝑔0 , … , 𝑔𝑚−1
are each representable in Q, then the function ℎ
such that:
Proof: (1) Suppose 𝑅 is representable in Q by
𝜓(𝑥1 , … , 𝑥𝑛 ). ℎ(𝑥0 , … , 𝑥𝑛−1 ) =
(2) Let 𝜑(𝑥1 , … , 𝑥𝑛 ) be (𝜓(𝑥1 , … , 𝑥𝑛 ) ∧ 𝑦 = 1) ∨ 𝑓 (𝑔0 (𝑥0 , … , 𝑥𝑛−1 ), … , 𝑔𝑚−1 (𝑥0 , … , 𝑥𝑛−1 )).
(¬𝜓(𝑥1 , … , 𝑥𝑛 ) ∧ 𝑦 = 0).
(3) Suppose 𝜒𝑅 (𝑘1 , … , 𝑘𝑛 ) = 𝑚. Then either is also representable in Q.
𝑅(𝑘1 , … , 𝑘𝑛 ) and 𝑚 = 1 or not-𝑅(𝑘1 , … , 𝑘𝑛 ) and
𝑚 = 0.
(4) Consider the first case. Then because
𝜓(𝑥1 , … , 𝑥𝑛 ) represents 𝑅, we have Q ⊢ Proof: (1) Suppose that 𝑓 , 𝑔 , … , 𝑔
0 𝑚−1 are all rep-
𝜓(𝑘1 , … , 𝑘𝑛 ); and we have Q ⊢ 𝑚 = 1 by resentable in Q. Let the wffs that repre-
Ref=. From these by Conj and Add we get Q ⊢ sent them be, respectively 𝜉(𝑥1 , … , 𝑥𝑚 , 𝑦) and
(𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑚 = 1) ∨ (¬𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑚 = 0). 𝜓0 (𝑥1 , … , 𝑥𝑛 , 𝑦), … , 𝜓𝑚−1 (𝑥1 , … , 𝑥𝑛 , 𝑦).
(5) Assume in the object language that
(𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑏 = 1) ∨ (¬𝜓(𝑘1 , … , 𝑘𝑛 ) ∧ 𝑏 = 0).

55
 (2) Now consider the wff below, which I’ll abbrevi- (12) Hence Q ⊢ 𝑦(𝜑(𝑘1 , … , 𝑘𝑛 , 𝑦) → 𝑙 = 𝑦) by UG.
A
ate 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦): (13) Together (5) and (12) mean that ℎ is repre-
sentable by 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦). ❅
E E E
𝑧 𝑧1 … 𝑧𝑚−1 [𝜓0 (𝑥1 , … , 𝑥𝑛 , 𝑧) ∧
𝜓1 (𝑥1 , … , 𝑥𝑛 , 𝑧1 ) ∧ … ∧ 𝜓𝑚−1 (𝑥1 , … , 𝑥𝑛 , 𝑧𝑚−1 ) ∧
𝜉(𝑧, 𝑧1 , … , 𝑧𝑚−1 , 𝑦)].
Result: The set of functions representable in
(3) Suppose now that ℎ(𝑘1 , … , 𝑘𝑛 ) = 𝑙, and let Q is closed under minimization, i.e., if 𝑓 is a
the values of 𝑔0 , … , 𝑔𝑚−1 for 𝑘1 , … , 𝑘𝑛 as argu- function for which there is always a 𝑦 such that
ments be 𝑗0 , … , 𝑗𝑚−1 respectively; this means 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = 0 for any given natural num-
that 𝑓 (𝑗0 , … , 𝑗𝑚−1 ) = 𝑙. bers 𝑥0 , … , 𝑥𝑛−1 , and 𝑓 is representable in Q, then
(4) Because these formulas represent these func- so is the function ℎ where:
tions, we have:
ℎ(𝑥0 , … , 𝑥𝑛−1 ) = the least 𝑦 such that
Q ⊢ 𝜓0 (𝑘1 , … , 𝑘𝑛 , 𝑗0 ) 𝑓 (𝑥0 , … , 𝑥𝑛−1 , 𝑦) = 0
Q ⊢ 𝜓1 (𝑘1 , … , 𝑘𝑛 , 𝑗1 )
is also representable in Q.
⋮
Q ⊢ 𝜓𝑚−1 (𝑘1 , … , 𝑘𝑛 , 𝑗𝑚−1 )
Q ⊢ 𝜉(𝑗0 , … , 𝑗𝑚−1 , 𝑙)
Proof: (1) Suppose 𝑓 is such a function, and
(5) By Conj, and EG, Q ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 , 𝑙). suppose 𝑓 is represented by the wff
(6) In the object language, assume 𝜑(𝑘1 , … , 𝑘𝑛 , 𝑏) 𝜓(𝑥1 , … , 𝑥𝑛 , 𝑥𝑛+1 , 𝑦).
for arbitrary 𝑏. (2) By definition, if 𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 ) = 𝑚, then
(7) By EI and Simp, for some constants 𝑐, 𝑐1 , … , 𝑐𝑚−1 ,
we get: Q ⊢ 𝜓(𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 , 𝑚)
A
Q ⊢ 𝑦(𝜓(𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 , 𝑦) → 𝑚 = 𝑦).
𝜓0 (𝑘1 , … , 𝑘𝑛 , 𝑐)
𝜓1 (𝑘1 , … , 𝑘𝑛 , 𝑐1 ) (3) Let 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦) be the wff:
A
⋮ 𝜓(𝑥1 , … , 𝑥𝑛 , 𝑦, 0)∧ 𝑧(𝑧 < 𝑦 → ¬𝜓(𝑥1 , … , 𝑥𝑛 , 𝑧, 0))
𝜓𝑚−1 (𝑘1 , … , 𝑘𝑛 , 𝑐𝑚−1 )
𝜉(𝑐, 𝑐1 , … , 𝑐𝑚−1 , 𝑏) We will show that this wff represents ℎ.
(4) Suppose that ℎ(𝑘1 , … , 𝑘𝑛 ) = 𝑚; we need to show
(8) By the definition of representation, we also that:
have:
Q ⊢ 𝜓(𝑘1 , … , 𝑘𝑛 , 𝑚, 0) ∧ (4a)
A
𝜓0 (𝑘1 , … , 𝑘𝑛 , 𝑐) → 𝑗0 = 𝑐 𝑧(𝑧 < 𝑚 → ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑧, 0))
(4b)
A
𝜓1 (𝑘1 , … , 𝑘𝑛 , 𝑐1 ) → 𝑗1 = 𝑐1 Q ⊢ 𝑦[𝜓(𝑘1 , … , 𝑘𝑛 , 𝑦, 0) ∧
⋮ A
𝑧(𝑧 < 𝑦 → ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑧, 0)) →
𝜓𝑚−1 (𝑘1 , … , 𝑘𝑛 , 𝑐𝑚−1 ) → 𝑗𝑚−1 = 𝑐𝑚−1 𝑚 = 𝑦]

(9) Putting the results above together, by LL, we (5) Since ℎ(𝑘1 , … , 𝑘𝑛 ) = 𝑚, we know that:
get 𝜉(𝑗0 , … , 𝑗𝑚−1 , 𝑏). (5a) 𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑚) = 0 and
(10) We have 𝑦(𝜉(𝑗0 , … , 𝑗𝑚−1 , 𝑦) → 𝑙 = 𝑦), and so (5b) For all 𝑗 < 𝑚, 𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑗) ≠ 0.
A
we get 𝑙 = 𝑏. (6) By (2) and (5a), we have the first conjunct of
(11) Hence Q ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 , 𝑏) → 𝑙 = 𝑏 by DT. (4a).

56
 (7) Either 𝑚 is 0 or it is not. Suppose it 0. Then Gödel’s 𝛽-function
by (≮0), the antecedent of the second conjunct
Consider the function defined as follows:
of (4a) is provably false for every 𝑧, and so the
conditional is always true, and we can prove 𝛽(𝑥1 , 𝑥2 , 𝑥3 ) = rem(1 + ((𝑥3 + 1) × 𝑥2 ), 𝑥1 )
the conjunct in Q.
(8) Suppose instead that 𝑚 is not 0. Then 𝑚 is Surprisingly, for any series of natural numbers with
′
𝑚−1. 𝑛 + 1 members
(9) By (2) and (5b), for all numbers 𝑗 < 𝑚 we have
a result of the form below, where 𝑟 is some 𝑘0 , 𝑘1 , 𝑘2 , … , 𝑘𝑛
number other than 0.
A one can find two fixed natural numbers 𝑏 and 𝑐
Q ⊢ 𝑦(𝜓(𝑘1 , … , 𝑘𝑛 , 𝑗, 𝑦) → 𝑟 = 𝑦) such that, for any 𝑖 ≤ 𝑛, 𝛽(𝑏, 𝑐, 𝑖) = 𝑘𝑖 .
(10) Instantiating 𝑦 to 0, and using the fact that To see this, first, let 𝑐 be
Q ⊢ 𝑟 ≠ 0 whenever 𝑟 is not 0 (Num=), by (max(𝑛, 𝑘0 , 𝑘1 , 𝑘2 , … , 𝑘𝑛 ))!
MT, we get a result of the form below for every
𝑗 < 𝑚: Next, consider the following sequence:
Q ⊢ ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑗, 0).
𝑢0 , 𝑢1 , 𝑢2 , … , 𝑢𝑛
′
(11) In the object language, suppose 𝑏 < 𝑚 − 1 . By
where each 𝑢𝑖 = 1 + ((𝑖 + 1) × 𝑐), for all 𝑖 ≤ 𝑛.
(Spread), we get:
No two members of the 𝑢-series have a factor
𝑏=0∨𝑏=1∨…∨𝑏=𝑚−1 (2.1) in common other than 1. (Tedious arithmetic.)
(12) By (10) and (11) and a big proof by cases, we It follows from this and a known principle of
get ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑏, 0). modular arithmetic, the Chinese remainder theorem,
(13) Discharging and generalizing, we have: that there is at least one number 𝑏 such that the
A ′
remainder upon division of 𝑏 by 𝑢𝑖 is always 𝑘𝑖 for
Q ⊢ 𝑧(𝑧 < 𝑚 − 1 → ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑧, 0)) every 𝑖 from 0 to 𝑛. (More tedious arithmetic.)
which by (8) is the second conjunct of (4a). Therefore, because each 𝑢𝑖 is 1 + ((𝑖 + 1) × 𝑐), it
(14) Thus whether 𝑚 is or is not 0, (4a) is true. follows that rem(1 + ((𝑖 + 1) × 𝑐), 𝑏) = 𝑘𝑖 , which is
(15) To show (4b), assume in the object language for to say that 𝛽(𝑏, 𝑐, 𝑖) = 𝑘𝑖 .
arbitrary 𝑏 that: Example: Suppose that our 𝑘-sequence is simply:
𝜓(𝑘1 , … , 𝑘𝑛 , 𝑏, 0) ∧ 1, 2, 1
A
𝑧(𝑧 < 𝑏 → ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑧, 0))
Then 𝑛 = 2 and max(2, 1, 2, 1) is also 2, and finally,
(16) By (Trichotomy), we have: 𝑐 = 2!, which is also 2.
𝑏<𝑚∨𝑚<𝑏∨𝑏=𝑚 Then, the 𝑢-series is:
(17) But by (4a), 𝑏 < 𝑚 → ¬𝜓(𝑘1 , … , 𝑘𝑛 , 𝑏, 0), and so 3, 5, 7
by MT with the first conjunct of (15), ¬𝑏 < 𝑚.
(18) By a similar argument using the first conjunct (These numbers share no common factor.)
of (4a) and the second conjunct of (15), we get It follows by the Chinese remainder theo-
¬𝑚 < 𝑏. rem that there at least one number 𝑏 such that
(19) By DS on (16), we get 𝑏 = 𝑚 and by Sym=, rem(3, 𝑏) = 1, and rem(5, 𝑏) = 2, and rem(7, 𝑏) = 1.
𝑚 = 𝑏. (In this case, 𝑏 could be 22, or 127, etc.)
(20) Dicharging the assumption at (15), and apply- For this sequence, for all 0 ≤ 𝑖 ≤ 2,
ing UG, we get (4b). ❅
𝛽(22, 2, 𝑖) = 𝑘𝑖

57
The upshot of all this is that it provides yet another
method of talking indirectly about sequences of
Result: The set of functions representable in Q
numbers. Each sequence corresponds to a 𝑏 and 𝑐,
is closed under primitive recursion, i.e., if 𝑓 and
and its elements can be “retrieved” by 𝛽.
𝑔 are representable in Q, then so is the function
ℎ, where either:

Result: The function 𝛽 is representable in Q. ℎ(𝑥0 , … , 𝑥𝑛−1 , 0) = 𝑓 (𝑥0 , … , 𝑥𝑛−1 ), and

ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦 + 1) =
𝑔(𝑥0 , … , 𝑥𝑛−1 , 𝑦, ℎ(𝑥0 , … , 𝑥𝑛−1 , 𝑦)),
Sketch of proof: We use the wff, which we shall later or
abbreviate as 𝜑𝛽 (𝑥1 , 𝑥2 , 𝑥3 , 𝑦): ℎ(0) = 𝑘
ℎ(𝑦 + 1) = 𝑔(𝑦, ℎ(𝑦)).
E
𝑧[𝑥1 = ((1 + ((𝑥3 + 1) × 𝑥2 )) × 𝑧) + 𝑦 ∧
𝑦 < 1 + ((𝑥3 + 1) × 𝑥2 )]

This says that 𝑦 is the remainder when 𝑥1 is divided Sketch of proof: A fuller proof is given in the text; we
by 1 + ((𝑥3 + 1) × 𝑥2 ) because it says that when 𝑦 is shall content ourselves with an rough explanation
added to a multiple of 1 + ((𝑥3 + 1) × 𝑥2 ) you get 𝑥1 , and an example, as the full proof is rather intricate.
and that 𝑦 is less than that number, so the multiple Suppose ℎ is definable recursively from 𝑓 and 𝑔
in question must be the greatest possible less than (or 𝑘 and 𝑔) in the way indicated, and suppose that
or equal to it. ℎ(𝑘1 , … , 𝑘𝑛 , 𝑗) = 𝑚. There is then a sequence of val-
We shall not bother giving the full proof that ues 𝑣0 , … , 𝑣𝑗 of ℎ for the same first 𝑛 arguments and
this meets the conditions for representability, but all the final arguments leading up to and including
we have already shown that all the functions and 𝑗:
relations involved (+, ×, =, <) meet those condi-
tions. 𝑣0 = ℎ(𝑘1 , … , 𝑘𝑛 , 0) = 𝑓 (𝑘1 , … , 𝑘𝑛 ) (or 𝑘)
We use the power of the 𝛽 function to represent 𝑣1 = ℎ(𝑘1 , … , 𝑘𝑛 , 1) = 𝑔(𝑘1 , … , 𝑘𝑛 , 0, 𝑣0 )
sequences of numbers to capture recursive defini- ⋮ ⋮
tions in Q and PA. Note that 𝑏 and 𝑐 are numbers 𝑣𝑗 = ℎ(𝑘1 , … , 𝑘𝑛 , 𝑗) = 𝑚 = 𝑔(𝑘1 , … , 𝑘𝑛 , 𝑗 − 1, 𝑣𝑗−1 )
representing a given sequence, then 𝜑𝛽 (𝑏, 𝑐, 𝑖, 𝑘)
says that 𝑘 is the 𝑖th member of this sequence, e.g.: For example, suppose ℎ(𝑥, 𝑦) = 𝑥 𝑦 . In the case
ℎ(2, 3) = 8, we have:
Q ⊢ 𝜑𝛽 (22, 2, 0, 1)
Q ⊢ 𝜑𝛽 (22, 2, 1, 2) 𝑣0 = ℎ(2, 0) =1
Q ⊢ 𝜑𝛽 (22, 2, 2, 1) 𝑣1 = ℎ(2, 1) = 𝑣0 × 2
𝑣2 = ℎ(2, 2) = 𝑣1 × 2
And a consequence of this, such as:
𝑣3 = ℎ(2, 3) = 8 = 𝑣2 × 2
E E
𝑥 𝑦[𝜑𝛽 (𝑥, 𝑦, 0, 1) ∧ 𝜑𝛽 (𝑥, 𝑦, 1, 2) ∧ 𝜑𝛽 (𝑥, 𝑦, 2, 1)]
We will use the object-language representation of
Gödel’s 𝛽-function in effect to claim that there is a
is true just in case there exists a sequence whose
sequence
first three elements are 1, 2 and 1.
𝑣0 , … , 𝑣𝑗
having the appropriate characteristics.
Representing recursion in Q
Suppose that 𝑓 is represented in the object
language by 𝜑𝑓 (𝑥1 , 𝑥2 , … , 𝑥𝑛 , 𝑦) (or in the simpler

58
case, we use 𝑦 = 𝑘), and suppose 𝑔 is repre- is represented by the following wff:
sented by 𝜑𝑔 (𝑥1 , … , 𝑥𝑛 , 𝑥𝑛+1 , 𝑥𝑛+2 , 𝑦). Then the wff
𝜑ℎ (𝑥1 , … , 𝑥𝑛 , 𝑥𝑛+1 , 𝑦) we can use to represent ℎ is: E E E
𝑣 𝑤( 𝑦2 [𝜑𝛽 (𝑣, 𝑤, 0, 𝑦2 ) ∧ 𝑦2 = 1] ∧
E E E 𝜑𝛽 (𝑣, 𝑤, 𝑥2 , 𝑦) ∧
𝑣 𝑤( 𝑦2 [𝜑𝛽 (𝑣, 𝑤, 0, 𝑦2 ) ∧ 𝜑𝑓 (𝑥1 , … , 𝑥𝑛 , 𝑦2 )] ∧ A E E
𝑧 [𝑧 < 𝑥2 → 𝑦3 𝑦4 [𝜑𝛽 (𝑣, 𝑤, 𝑧, 𝑦3 ) ∧
𝜑𝛽 (𝑣, 𝑤, 𝑥𝑛+1 , 𝑦) ∧
A E E 𝜑𝛽 (𝑣, 𝑤, 𝑧 ′ , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 𝑥1 ]])
𝑧 [𝑧 < 𝑥𝑛+1 → 𝑦3 𝑦4 [𝜑𝛽 (𝑣, 𝑤, 𝑧, 𝑦3 ) ∧
𝜑𝛽 (𝑣, 𝑤, 𝑧 ′ , 𝑦4 ) ∧ 𝜑𝑔 (𝑥1 , … , 𝑥𝑛 , 𝑧, 𝑦3 , 𝑦4 )]]) This says that there is a sequence of natural num-
bers with 𝑥2 + 1 members, the first of which is 1,
Ugly! What on Earth does this say?! the last of which is 𝑦, and each one relates to the
previous one by being its product when multiplied
• What we want it to say is that 𝑦 is the value
by 𝑥1 . With some thought, it is clear that this is the
of the recursive function ℎ for 𝑥1 , … , 𝑥𝑛 , 𝑥𝑛+1 as
case if and only if 𝑦 = 𝑥1𝑥2 .
arguments. Does it?
Completing the full proof of the meta-
• Remember that the 𝛽 function is used to talk
theorem would involve establishing that if
indirectly about finite sequences. Because each
ℎ(𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 ) = 𝑚, then:
finite sequence corresponds to a fixed 𝑏 and 𝑐
such that 𝛽(𝑏, 𝑐, 𝑖) is always the 𝑖th member of the Q ⊢ 𝜑ℎ (𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 , 𝑚)
sequence, quantification over sequences can in A
effect be done by quantifying over two numbers. Q ⊢ 𝑦[𝜑ℎ (𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 , 𝑦) → 𝑚 = 𝑦]
• The existential quantification over 𝑣 and 𝑤 at the
This is where the proof gets very involved for the
start of this wff in effect says “there is a sequence
general case.
such that . . . ”.
However, consider the example that 13 = 1.
• Given that 𝜑𝛽 (…) represents the 𝛽 function, and
In that case, the 𝑣-sequence is 10 , 11 , 12 , 13 , which is
𝜑𝑓 (…) represents 𝑓 , the first conjunct on the in-
1, 1, 1, 1. The 𝛽-function encoding for this sequence
side says that there is a 𝑦2 at the start (0-spot) of
uses 𝑏 = 1 and 𝑐 = 6. We have:
the sequence, and it’s the value of 𝑓 for 𝑥1 , … 𝑥𝑛 .
This basically says how the sequence of values Q ⊢ 𝜑𝛽 (1, 6, 0, 1)
begins.
Q ⊢ 𝜑𝛽 (1, 6, 1, 1)
• Next, it says that 𝑦 is at the 𝑥𝑛+1 -spot of the se-
quence of values, which is to be expected if 𝑦 is Q ⊢ 𝜑𝛽 (1, 6, 2, 1)
the value of ℎ when ℎ’s last argument is 𝑥𝑛+1 . Q ⊢ 𝜑𝛽 (1, 6, 3, 1)
• Lastly, given that 𝜑𝑔 (…) represents 𝑔, it says that Q⊢1=1
for each previous spot in the sequence (each 𝑧- Q⊢1=1×1
spot, where 𝑧 is below the argument we’re cur-
rently considering), the member of the sequence from the fact that 𝜑𝛽 (𝑥1 , 𝑥2 , 𝑥3 , 𝑦) represents 𝛽 and
at the next spot (𝑦4 ) is obtained from the member other results we have established.
at the 𝑧-spot (𝑦3 ) in the appropriate way from the From the first and fifth, we get:
𝑔 function.
Let us consider the example of 𝑥1𝑥2 .
E
Q ⊢ 𝑦2 [𝜑𝛽 (1, 6, 0, 𝑦2 ) ∧ 𝑦2 = 1]
Its definition uses 𝑘 = 1 (this plays the role of
𝑓 ) and multiplication (this plays the role of 𝑔). The pairs of first and second, second and third,
Making some minor simplifications, these are and third and fourth, along with the last, give us:
represented in Q by the wffs 𝑦 = 1 and 𝑦 = 𝑥1 × 𝑥2 .
According to the above recipe, the function 𝑥1𝑥2

59
 E E
Q ⊢ 𝑦3 𝑦4 [𝜑𝛽 (1, 6, 0, 𝑦3 ) ∧
𝜑𝛽 (1, 6, 0′ , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 1] Corollary: All (primitive and general) recur-
E E sive relations are representable in Q and PA.
Q ⊢ 𝑦3 𝑦4 [𝜑𝛽 (1, 6, 1, 𝑦3 ) ∧
′
𝜑𝛽 (1, 6, 1 , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 1]
E E
Q ⊢ 𝑦3 𝑦4 [𝜑𝛽 (1, 6, 2, 𝑦3 ) ∧
′ Proof: Suppose 𝑅 is a recursive relation. This
𝜑𝛽 (1, 6, 2 , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 1]
means that its characteristic function 𝜒𝑅 is a re-
The numerals 0, 1, 2 exhaust the numerals for 𝑧 < 3, cursive function. By the result above, 𝜒𝑅 can be
and so, by (Spread) and a proof by cases: represented in Q. By what you proved in home-
work 27, it follows that 𝑅 can be represented in Q
A E E
𝑧 [𝑧 < 3 → 𝑦3 𝑦4 [𝜑𝛽 (1, 6, 𝑧, 𝑦3 ) ∧ (and thus PA). ❅
𝜑𝛽 (1, 6, 𝑧 ′ , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 1]]
We shall later prove that the converse is also true:
all functions/relations representable in Q (and even
Putting some of these results together, and existen- PA) are recursive.
tially generalizing from 1 and 6, we get:
Homework 28
Show that if 𝑓 is a recursive function, then there is
E E E
Q ⊢ 𝑣 𝑤( 𝑦2 [𝜑𝛽 (𝑣, 𝑤, 0, 𝑦2 ) ∧ 𝑦2 = 1] ∧
a wff 𝜑(𝑥1 , … , 𝑥𝑛 , 𝑦) in the language 𝐴 such that,
𝜑𝛽 (𝑣, 𝑤, 3, 1) ∧
A E E for all natural numbers 𝑘1 , … , 𝑘𝑛 and 𝑚, it holds that
𝑧 [𝑧 < 3 → 𝑦3 𝑦4 [𝜑𝛽 (𝑣, 𝑤, 𝑧, 𝑦3 ) ∧ N ⊨ 𝜑(𝑘1 , … , 𝑘𝑛 , 𝑚) iff 𝑓 (𝑘1 , … , 𝑘𝑛 ) = 𝑚.
′
𝜑𝛽 (𝑣, 𝑤, 𝑧 , 𝑦4 ) ∧ 𝑦4 = 𝑦3 × 1]]) (Hint: appeal to the fact that Q ⊆ TA.)

Which is our way of saying that 13 = 1. The pattern 8. Arithmetization of Syntax

of this example generalizes to all our functions we
could define by the primitive recursion method. System ,
Result: All (primitive and general) recursive
functions are representable in Q and PA.
Proof: By results achieved so far, we know that the
set of functions representable in Q contains the ini-
tial functions (the zero function, the successor func-
tion, and projection functions) and is closed under
composition, primitive recursion and minimization.
Since the set of (general) recursive functions is the
smallest such set, all its members are contained
within. This includes the primitive recursive func-
tions, which are a subset of the general recursive
functions. Moreover, all those representable in Q
are also representable in PA. ❅
Thought of as broadly as possible, a syntax is just
a series of rules for what counts as an “allowed”
string of basic symbols, and a “deduction system”
just a set of rules for which allowed strings, and
manipulations between strings, are “favored” or
“endorsed”. This would even include:
Definition: The basic symbols of , are “ □” and
“ #”.
Definition: An allowed string of , is any finite
string of the basic symbols that begins with “ □”.
Examples: “□”, “□□”, “□##□”, are allowed strings,
but “#□□#” is not.
Definition: An allowed string follows by another
allowed string by the rule add-circle iff it is just like
the other string, except with “ #” appended at the
end.

60
Example: “□##” follows by add-circle from “□#”. says that □##### is favored, since its Schmödel
number (32) has the property of being a power of 2
Definition: The set of favored strings of , is the iff it is either 1, or is only divisible by 1 and even
smallest set containing “ □” and closed under add- numbers.
circle. Of course, by means of Gödel numbering, sys-
tems such as Q and PA can even, at least, partially,
We can think of , as a “deduction system” whose act as their own metatheories, and by so doing, even
“theorems” or “favored strings” include exactly: “say” things about themselves in a sense.
□
□#
□##
Gödel numbering
□### We begin by assigning a different number to every
□####, etc. simple sign in the language of first-order logic; see
We could do “metatheory” for ,, and it would be page 25 for details.
pretty easy to do. In fact, its metatheory can be Definition: The Gödel number of a string of
reduced to simple mathematics if we arithmetize symbols of the language of first-order logic is the
its syntax, or convert our talk of strings to talk of number that encodes (using the first however-many
numbers. primes raised to one more than each number) the
series of the numbers for the simple symbols in order.
Definition: The Schmödel number of an allowed
string of , is the number which is represented in Example: The Gödel number of string “A1 (a)” is
binary notation when “ □” is interpreted as the digit 2(2 ⋅3 ⋅5 ⋅7 )+1 ⋅ 3(2 ⋅3 )+1 ⋅ 5(2 ⋅3 ⋅5 )+1 ⋅ 7(2 ⋅3 )+1 , since the
5 2 1 2 1 12 3 1 2 1 13

“ 1” and “#” is interpreted as “ 0”. number of simple symbol A1 is 25 ⋅ 32 ⋅ 51 ⋅ 72 , the

number of “(” is “21 ⋅ 312 ”, and so on.
Examples: The Schmödel number of “□##” is 4
(100 in binary), and the Schmödel number of Warning: Be sure to differentiate 1the 12 number for
“□□#□#” is 26 (11010 in binary). the simple symbol, “(”, which is 2 ⋅ 3 , from the
Gödel number of the single-character-long string
“(”, which is 2(2 ⋅3 )+1 .
1 12

We shall use the notation:

Result: An allowed string of , is favored iff its
# 𝜑#
Schmödel number is a power of 2.
in the metalanguage for the Gödel number of string
𝜑, so that, e.g.:
# A1 (a)# 5 2 1 2 1 12 )+1
= 2(2 ⋅3 ⋅5 ⋅7 )+1 ⋅ 3(2 ⋅3
3 1 2 1 13 )+1
⋅ 5(2 ⋅3 ⋅5 )+1 ⋅ 7(2 ⋅3
Proof: Each favored string is derived from □
(Schmödel number 1) by adding some number of Since whatever string 𝜑 is, # 𝜑# is just a particular
circles at the end, which is the same as multiplying
constant number, we can make use of this notation
its Schmödel number by 2 (just like adding a 0 at the
in recursive definitions.
end in decimal notation is the same as multiplying Constructed expressions such as (complex)
by 10). ❅
terms and formulas are just kinds of strings, and so
we may use this notation for their Gödel numbers.
Indeed, the “arithmetization” of system , means We can extend the notion as such:
that we can do its metatheory in systems like Q
and PA. Arguably: Definition: The Gödel number of a sequence of
strings 𝜑0 , … , 𝜑𝑛−1 is the number that encodes the
E A E
32 = 1∨ 𝑥[ 𝑦 𝑥 ×𝑦 = 32 → (𝑥 = 1∨ 𝑦 2×𝑦 = 𝑥)] sequence of Gödel numbers of the strings in the series,
in order.

61
Example: Our first example derivation: (b) 𝑥 is the Gödel number of a string of just a
constant: 𝑦 𝑧
Fa → [(Fa → Fa) → Fa]
{Fa → [(Fa → Fa) → Fa]}
Const(𝑥) iff ∃ ∃
𝑦<𝑥 1≤𝑧≤5
𝑥 = code(code(2, 𝑦, 𝑧))
Or for 𝐴 :
→ {[Fa → (Fa → Fa)] → (Fa → Fa)}
Const𝐴 (𝑥) iff 𝑥 = # 0#
[Fa → (Fa → Fa)] → (Fa → Fa)
Fa → (Fa → Fa) (c) 𝑥 is the Gödel number a 𝑦-place function letter
𝑧 𝑤
Fa → Fa FL(𝑥, 𝑦) iff ∃ ∃
𝑧<𝑥 1≤𝑤≤7
𝑥 = code(code(3, 𝑦, 𝑧, 𝑤))

Is a sequence of strings having the Gödel number:
# Fa→[(Fa→Fa)→Fa]# +1
2 ×
# {Fa→[(Fa→Fa)→Fa]}→{[Fa→(Fa→Fa)]→(Fa→Fa)}# +1
3 ×
FL𝐴 (𝑥, 𝑦) iff (𝑦 = 1 and 𝑥 = # ′ # ) or
(𝑦 = 2 and (𝑥 = # +# or 𝑥 = # ×# ))
(d) 𝑥 is the Gödel number of a 𝑦-place predicate:
Pred(𝑥, 𝑦) iff (𝑦 = 2 and 𝑥 = # =# ) or
𝑧 𝑤

5
# [Fa→(Fa→Fa)]→(Fa→Fa)# +1
×
∃ ∃
𝑧<𝑥 1≤𝑤≤10
𝑥 = code(code(4, 𝑦, 𝑧, 𝑤))

# Fa→(Fa→Fa)# +1 Pred𝐴 (𝑥, 𝑦) iff 𝑦 = 2 and (𝑥 = # =# or 𝑥 =

7 × # <# )
# Fa→Fa# +1
11 (e) flatten(𝑥) = the Gödel number for the string
obtained by concatenating each item in the
A single finite number, albeit an enormous one.
sequence of strings having Gödel number 𝑥
with # ,# in between each item.
Warning: Just as we must differentiate between the
number of a simple symbol and the Gödel number (f) sequenceBound(𝑥, 𝑦) = the greatest possible
of a single-symbol string of just it, we must dif- number a code for a sequence of length 𝑦
ferentiate between the Gödel number of a string whose elements are no greater than 𝑥 could be
from that of a single-string-long sequence of strings (g) 𝑥 is the Gödel number of a term:
containing just it, i.e., the difference between Term(𝑥) iff Var(𝑥) or Const(𝑥) or
𝑦 𝑧 𝑤
# Fa → [(Fa → Fa) → Fa]# and 2# Fa→[(Fa→Fa)→Fa]# +1 .
∃∃ ∃
[FL(𝑦, 𝑧) and
𝑦<𝑥 𝑧<𝑥 𝑤<sequenceBound(𝑥,𝑧)

Recursive syntax 𝑥 = 𝑦 ⌢ # (# ⌢ flatten(𝑤) ⌢ # )# and

len(𝑤) = 𝑧 and
𝑣 𝑥1
∀ ∀ (if 𝑥
0≤𝑣<𝑧 𝑥1 <𝑥
1 = (𝑤)𝑣 then Term(𝑥1 ))]
Result: The following functions, properties and
(This appears to define Term(𝑥) in terms of
relations relating to the syntax of first-order logic
itself, but notice 𝑥1 is bound below 𝑥, and so
are primitive recursive.
this is allowed by course of values recursion.)
We can define Term𝐴 similarly, with Const𝐴
and FL𝐴 , etc.
(h) num(𝑥): the Gödel number of the numeral for
We sometimes bother giving the formal definitions
𝑥, defined recursively:
showing them to be primitive recursive, sometimes
num(0) = # 0#
won’t bother, as it gets tedious. More details can
num(𝑥 + 1) = # f1 (# ⌢ num(𝑥) ⌢ # )#
be found in the book.
(Recall that 𝑡 ′ is shorthand for f1 (𝑡).)
(a) 𝑥 is the Gödel number of a string of just a
(i) 𝑥 is the Gödel number of an atomic formula.
variable: 𝑦
𝑧 Atom(𝑥) iff 𝑥 = # ⊥# or 𝑥 = # ⊤# or
Var(𝑥) iff ∃ ∃
𝑦<𝑥 1≤𝑧≤6
𝑥 = code(code(1, 𝑦, 𝑧))

62
 𝑦 𝑧 𝑤
number of the result of replacing the numeral
∃∃ ∃
𝑦<𝑥 𝑧<𝑥 𝑤<sequenceBound(𝑥,𝑧)
[Pred(𝑦, 𝑧) and for 𝑦 for all free occurrences of the (object-
language) variable 𝑥 in the formula with Gödel
𝑥 = 𝑦 ⌢ # (# ⌢ flatten(𝑤) ⌢ # )# and
number 𝑦:
len(𝑤) = 𝑧 and
𝑣 𝑥1 diag(𝑦) = Subst(𝑦, num(𝑦), # 𝑥 # )
∀ ∀ (if 𝑥
0≤𝑣<𝑧 𝑥1 <𝑥
1 = (𝑤)𝑣 then Term(𝑥1 ))]
(Atom𝐴 (𝑥) similarly.) Derivations and recursion
(j) 𝑥 is the the Gödel number of a well-formed
formula:
Frm(𝑥) iff Atom(𝑥) or Result: The following properties and relations
𝑦 are recursive.
∃ (Frm(𝑦) and 𝑥 = #¬# ⌢ 𝑦) or
𝑦<𝑥
𝑦 𝑧 𝑤
∃ ∃ ∃ (Frm(𝑦) and Frm(𝑧) and
𝑦<𝑥 𝑧<𝑥 𝑤<𝑥
(a) MP(𝑥, 𝑦): 𝑥 is the Gödel number of a sequence
(𝑤 = # →# or 𝑤 = # ↔# or
where the 𝑦 th element follows by modus ponens
𝑤 = # ∨# or 𝑤 = # ∧# ) and
from previous elements of the sequence.
𝑥 = # (# ⌢ 𝑦 ⌢ 𝑤 ⌢ 𝑧 ⌢ # )# ) or 𝑧 𝑤
𝑦 𝑧 𝑤 MP(𝑥, 𝑦) iff ∃ ∃(
∃∃∃
𝑦<𝑥 𝑧<𝑥 𝑤<𝑥 (
Frm(𝑦) and Var(𝑧) and 𝑧<𝑦 𝑤<𝑦
# (#
(𝑥)𝑧 = ⌢ (𝑥)𝑤 ⌢ # →# ⌢ (𝑥)𝑦 ⌢ # )# )
(𝑤 = # # or 𝑤 = # # ) and
A E
(b) QR(𝑥, 𝑦): 𝑥 is the Gödel number of a sequence
𝑥 = # 𝑤# ⌢ # 𝑧# ⌢ # 𝑦 # ) where the 𝑦 th element follows by the quantifier
We can get Frm𝐴 similarly. rules (QR) from a previous element.
(k) FreeOcc(𝑥, 𝑦, 𝑧): the formula with Gödel num- (c) 𝑥 is the Gödel number of an instance of the
ber 𝑥 contains a free occurrence of the variable axiom schema (SimpLAx):
with Gödel number 𝑦 in position 𝑧. SimpLAx(𝑥) iff
(l) Sent(𝑥): 𝑥 is the Gödel number of a sentence 𝑦 𝑧
(i.e., formula without free variables) ∃ ∃ (Sent(𝑦) and Sent(𝑧) and
𝑦<𝑥 𝑧<𝑥
(And Sent𝐴 (𝑥) similarly from Frm𝐴 .)
𝑥 = # ((# ⌢ 𝑦 ⌢ # ∧# ⌢ 𝑧 ⌢ # ) →# ⌢ 𝑦 ⌢ # )# )
(m) Subst(𝑥, 𝑦, 𝑧): the Gödel number of the for-
(And so on for all the other logical axiom
mula resulting by replacing all free occur-
schemata.)
rences of the variable with Gödel number 𝑧
(d) IsAx(𝑥): 𝑥 is the Gödel number of an instance
with the term with Gödel number 𝑦.
of any of the axiom schemata of first-order
(n) UnivClosure(𝑥): the Gödel number of the for-
logic with identity (including, e.g., Ref=Ax,
mula resulting from prepending a universal
LLAx).
quantifier of the form 𝑦 to the formula with
A
(e) Deriv(𝑥): 𝑥 is the Gödel number of a sequence
Gödel number 𝑥 for every variable it contains
constituting a purely logical derivation (i.e.,
free.
one without premises or non-logical axioms):
(o) Neg(𝑥): the Gödel number of the negation of 𝑦
the formula with Gödel number 𝑥 Deriv(𝑥) iff ∀
𝑦<len(𝑥)
(IsAx((𝑥)𝑦 ) or
(p) Cond(𝑥, 𝑦): the Gödel number of the condi-
tional whose antecedent is the conjunction of MP(𝑥, 𝑦) or QR(𝑥, 𝑦))
all wffs whose Gödel numbers are in the se-
quence encoded by 𝑥 and whose consequent Definition: Γ is a (primitive) recursive set of
is the wff whose Gödel number is 𝑦. sentences iff the property of being the Gödel number
(q) The (evil) diagonalization function: the Gödel of a member of Γ is (primitive) recursive.

63
Definition: A theory is recursively axiomatiz-
able (or sometimes, simply, axiomatizable) iff it is
the closure of a recursive set (or equivalently, axiom-
atized by a recursive set).
Result: Q is recursively axiomatizable.
Result: If Γ is a (primitive) recursive set of sen-
tences, then the relation Prf Γ (𝑥, 𝑦) that holds be-
tween 𝑥 and 𝑦 when 𝑥 is the Gödel number of
purely logical derivation of a formula of the form
(𝜓1 ∧ . . . ∧ 𝜓𝑛 ) → 𝜑, where each of 𝜓𝑖 ∈ Γ and
𝑦 is the Gödel number of 𝜑, is also (primitive)
recursive.

Proof: Q is the closure of the set of axioms Q1 –Q8 ,

and the property of being the Gödel number of one
of these is simply the disjunction of being each one
of them, which is primitive recursive:
IsQAx(𝑥) iff
𝑥 = # x y(x′ = y′ → x = y)# or …
A A
or 𝑥 = # x y(x < y ↔ z z′ + x = y)#
A A E
❅ 𝑤
Result: PA is recursively axiomatizable.
Proof: PA is the closure of the set containing Q1 –
Q8 and every instance of PA9 . Hence 𝑥 is the Gödel
number of a member of this set iff IsQAx(𝑥) or:
𝑦 𝑧 𝑣 sive relations.
∃ ∃ ∃ (Frm(𝑧) and Var(𝑣) and
𝑦≤𝑥 𝑧<𝑦 𝑣<𝑥
Proof: Let 𝑅Γ be the (primitive) recursive property
a number has iff it is the Gödel number of a member
of Γ. We can define Prf Γ as follows:
𝑧
Prf Γ (𝑥, 𝑦) iff Deriv(𝑥) and ∃
𝑧<sequenceBound(𝑥,𝑥)[
(𝑥)len(𝑥)−1 = Cond(𝑧, 𝑦) and ∀
𝑤<len(𝑧)
𝑅Γ ((𝑧)𝑤 )]
This says that 𝑥 is the Gödel number of a purely
logical derivation, the last line of which is a condi-
tional with an antecedent that is a conjunction of
formulas all of which come from Γ, and the conse-
quent of which has Gödel number 𝑦. ❅
Corollary: Prf Q and Prf PA are primitive recur-

𝑦 = # ((# ⌢ Subst(𝑧, # 0# , 𝑣) ⌢ # ∧ # ⌢ 𝑣 ⌢
A

# (# ⌢ 𝑧 ⌢ # →# ⌢ Subst(𝑧, # f1 (# ⌢ 𝑣 ⌢ # )# , 𝑣)
# )) # # )#
Result: Prf Γ (𝑥, # 𝜑# ) holds for some 𝑥 iff Γ ⊢ 𝜑.
and
A
⌢ → ⌢𝑣⌢𝑧⌢
𝑥 = UnivClosure(𝑦))
and so is primitive recursive. ❅ Proof: Follows directly from the definition of Prf Γ
and the Logical Conditionalization corollary of
Compactness listed on page 31. ❅
Result: If Γ is a recursive set of sentences, so
is any set Γ ∪ {𝜑1 , … , 𝜑𝑛 } obtained by adding a Representable functions are recursive
finite number of sentences to Γ.

Result: Every function representable in Q (or

PA, or any other recursively axiomatizable the-
ory of which Num= holds) is recursive.
Homework 29
Prove the result above.

64
Proof: (1) Let 𝑓 be a function representable in Q.
By definition, there is a wff 𝜑𝑓 (𝑥1 , … , 𝑥𝑛 , 𝑦) such
that if 𝑓 (𝑘1 , … , 𝑘𝑛 ) = 𝑚, then:
Q ⊢ 𝜑𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑚)
A definition given for 𝑔.
Q ⊢ 𝑦(𝜑𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑦) → 𝑚 = 𝑦)
(2) Let 𝐴 be the (𝑛 + 1)-place function which
yields as value the Gödel number of the for-
mula that results by replacing the free variables
𝑥1 , … , 𝑥𝑛 , 𝑦 in 𝜑𝑓 (𝑥1 , … , 𝑥𝑛 , 𝑦) by the numerals
for its arguments, i.e.:
• The same holds for any other arguments
to 𝑓 and 𝑔, and so 𝑓 and 𝑔 are the same
function.
(8) Hence 𝑓 can be recursively defined with the
❅
Corollary: In any consistent, recursively ax-
iomatizable extension of Q, every representable
function is recursive.

𝐴(𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 ) = # 𝜑𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 )#

(3) 𝐴 is primitive recursive, as it can be defined as
follows:
Homework 30
𝐴(𝑘1 , … , 𝑘𝑛 , 𝑘𝑛+1 ) = Subst(Subst(… Prove the corollary above, with the crucial piece
being that any extension of Q in which Num= did
(Subst(# 𝜑𝑓 (𝑥1 , … , 𝑥𝑛 , 𝑦)# , num(𝑘1 ), # 𝑥1 # ), … ),
not hold would be inconsistent. (Num= is stated on
num(𝑘𝑛 ), # 𝑥𝑛 # ), num(𝑘𝑛+1 ), # 𝑦 # ) page 43.)
(4) Now consider the relation 𝑅 such that:
𝑅(𝑘1 , … , 𝑘𝑛 , 𝑠) is true iff 𝑠 encodes a pair of num-
bers 𝑗1 and 𝑗2 , where Prf Q (𝑗1 , # 𝜑𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑗2 )# ), Corollary: The set of functions representable
or roughly, where 𝑗1 is the Gödel number of an in Q (or PA) is precisely the set of (general) re-
object language proof that 𝑓 (𝑘1 , … , 𝑘𝑛 ) = 𝑗2 . cursive functions.
(5) 𝑅 is primitive recursive, as:
𝑧1 𝑧2
𝑅(𝑘1 , … , 𝑘𝑛 , 𝑠) iff ∃ ∃ [𝑠 = code(𝑧 , 𝑧 ) and
𝑧1 <𝑠 𝑧2 <𝑠
1 2

Prf Q (𝑧1 , 𝐴(𝑘1 , … , 𝑘𝑛 , 𝑧2 ))] Corollary: The set of relations representable in

Q (or PA) is precisely the set of (general) recur-
(6) We can now give a recursive definition of func-
sive relations.
tion 𝑔 so:
𝑔(𝑘1 , … , 𝑘𝑛 ) = ((𝜇𝑠)𝑅(𝑘1 , … , 𝑘𝑛 , 𝑠))1
(7) 𝑔 is none other than 𝑓 .
• To see this, suppose 𝑓 (𝑘1 , … , 𝑘𝑛 ) = 𝑚 and 9. Diagonalization
𝑔(𝑘1 , … , 𝑘𝑛 ) = 𝑗. This means that 𝑗 is some
number for which there is a proof estab- Preliminaries
lishing:
Some notation:
Q ⊢ 𝜑𝑓 (𝑘1 , … , 𝑘𝑛 , 𝑗) ⌜𝜑⌝ is shorthand for # 𝜑# .
• By this and (1), we get Q ⊢ 𝑚 = 𝑗, and I.e., ⌜𝜑⌝ is the object language numeral for the
so, by Num=, 𝑚 and 𝑗 are in fact the same Gödel number of 𝜑.
number.

65
Example: ⌜A1 (a)⌝ is the string consisting of

A
2(2 ⋅3 ⋅5 ⋅7 )+1 ⋅ 3(2 ⋅3 )+1 ⋅ 5(2 ⋅3 ⋅5 )+1 ⋅ 7(2 ⋅3 )+1 succes-
5 2 1 2 1 12 3 1 2 1 13

𝑦(𝜃diag (𝑥, 𝑦) → 𝜑1 (𝑦)) ∶

sor function signs applied to “0”. Don’t try writing
that out in full.

𝜑3 (𝑥) ∶

𝜑2 (𝑥) ∶

𝜑1 (𝑥) ∶
⋮

⋮
For our next result, consider an (infinitely large)
chart with each row representing an open formula
with one free variable, 𝑥, and each column rep-
resenting the Gödel number of an open formula,
arranged in the same order.

A
𝑦(𝜃diag (⌜𝜑1 (𝑥)⌝, … ( 𝑦(𝜃diag (⌜𝜑2 (𝑥)⌝, …
Where they meet, we put the sentence in which

𝜑3 (⌜𝜑1 (𝑥)⌝)

𝜑2 (⌜𝜑1 (𝑥)⌝)

𝜑1 (⌜𝜑1 (𝑥)⌝)
the open formula of the row is asserted of the nu-

#𝜑
meral for the number for the column.

1 (𝑥)
⋮

⋮
Suppose 𝜑1 (𝑥) means “𝑥 is even”. Then as we

#
go across the first row, each value is the formula
claiming that the number at the top is even.
The numbers along the “diagonal” of the chart

A
are special. These are underlined. The first one is

𝜑3 (⌜𝜑2 (𝑥)⌝)

𝜑2 (⌜𝜑2 (𝑥)⌝)

𝜑1 (⌜𝜑2 (𝑥)⌝)
the Gödel number of the claim that the Gödel num-

#𝜑
2 (𝑥)
ber of “𝑥 is even” is even. The “evil” diagonalization
⋮

⋮
function, diag, introduced on p. 63, is the function

#
from the numbers at the top to the number of the
underlined formula in their columns.
But diag is a recursive function, and is itself
A
𝑦(𝜃diag (⌜𝜑3 (𝑥)⌝, … ⋯

representable, by 𝜃diag (𝑥, 𝑦), say. 𝜑3 (⌜𝜑3 (𝑥)⌝)

𝜑2 (⌜𝜑3 (𝑥)⌝)

𝜑1 (⌜𝜑3 (𝑥)⌝)
For each open formula 𝜑𝑖 (𝑥), there is, some-

#𝜑
3 (𝑥)
where else on the list, the open sentence asserting
⋮

that every value of diagonalization applied to 𝑥 sat-

#
isfies 𝜑𝑖 (𝑥), i.e., 𝑦(𝜃diag (𝑥, 𝑦) → 𝜑𝑖 (𝑦)). A formula
A
in its row claims that the underlined number in its
own column satisfies 𝜑𝑖 (𝑥), e.g., in the first cell there
⋯ 𝜑3 (⌜ 𝑦(𝜃diag (𝑥, 𝑦) → 𝜑1 (𝑦))⌝) ⋯

⋯ 𝜑2 (⌜ 𝑦(𝜃diag (𝑥, 𝑦) → 𝜑1 (𝑦))⌝) ⋯

⋯ 𝜑1 (⌜ 𝑦(𝜃diag (𝑥, 𝑦) → 𝜑1 (𝑦))⌝) ⋯

⋯
is the (true!) claim that # 𝜑1 (⌜𝜑1 (𝑥)⌝)# is even.
Since this second open formula has its own row,
#
A

A
𝑦(𝜃diag (⌜ 𝑦(𝜃diag … ⌝, …

A
𝑦(𝜃diag (𝑥, 𝑦) → 𝜑1 (𝑦))#

there will eventually come some column where the

diagonal reaches this row. The sentence in that
spot asserts that the Gödel number of the sentence
A
⋮

in the underlined spot in that column is even: i.e.,

that its own Gödel number is even!
This special square is called a “fixed point”.
Fixed points make self-reference possible, and no-
tice that there will be one not only for 𝜑1 (𝑥), but
for any other open formula.
⋯

66
The fixed point theorem 5. 𝜑(𝑏) 1, 4 LL
6. 𝜃diag (⌜𝛼(𝑥)⌝, 𝑏) → 𝜑(𝑏) 2–5 DT
7. 𝑦(𝜃diag (⌜𝛼(𝑥)⌝, 𝑦) → 𝜑(𝑦)) 6 UG
A

Result (fixed point theorem): For any the- 8. 𝛼(⌜𝛼(𝑥)⌝) 7 Def. 𝛼(𝑥)
ory T in which all recursive functions are rep- 9. 𝜓 8 Def. 𝜓
resentable, it holds that for every formula 𝜑(𝑥) 10. 𝜑(⌜𝜓⌝) → 𝜓 1–9 DT
with one free variable, there is a wff 𝜓 such that: (9) By (7) and (8) and BI, T ⊢ 𝜓 ↔ 𝜑(⌜𝜓⌝).
❅
T ⊢ 𝜓 ↔ 𝜑(⌜𝜓⌝)

Corollary: The fixed point theorem applies to

Q, PA, TA, and any other theory extending Q.
Proof: (1) Suppose T is such a theory, and 𝜑(𝑥) an
arbitrary open formula.
(2) T represents the recursive function diag; hence
there is a wff 𝜃diag (𝑥, 𝑦), where if diag(𝑘) = 𝑚, Proof: All recursive functions are represented in Q,
then: and therefore would also be in any theory contain-
ing it, PA and TA included. ❅
T ⊢ 𝜃diag (𝑘, 𝑚)
A
T ⊢ 𝑦(𝜃diag (𝑘, 𝑦) → 𝑚 = 𝑦)
10. Gödel’s Results and
(3) Consider the wff 𝑦(𝜃diag (𝑥, 𝑦) → 𝜑(𝑦)), and
A
abbreviate it as 𝛼(𝑥). Corollaries
(4) Let 𝜓 be the sentence 𝛼(⌜𝛼(𝑥)⌝). (Note that
this does not contain 𝑥 free, as ⌜𝛼(𝑥)⌝ is just 0 Expressing Prf in the object language
followed by many “′ ”s.)
(5) Given the way that 𝜓 is constructed:
Result (Prf ⊢): In any theory T, which is a re-
# # #
diag( 𝛼(𝑥) ) = 𝜓 .# cursively axiomatizable extension of Q, and Γ
is the (or a) recursive set of which T is the clo-
(6) By (2) and (5) we have: sure, there is a formula PrfΓ (𝑥, 𝑦) such that, if
Prf Γ (𝑘, 𝑚), then:
T ⊢ 𝜃diag (⌜𝛼(𝑥)⌝, ⌜𝜓⌝) (6a)
T ⊢ 𝑦(𝜃diag (⌜𝛼(𝑥)⌝, 𝑦) → ⌜𝜓⌝ = 𝑦) (6b) Q (and T) ⊢ PrfΓ (𝑘, 𝑚)
A

(7) First we show T ⊢ 𝜓 → 𝜑(⌜𝜓⌝). And if not-Prf Γ (𝑘, 𝑚)

1. 𝜓 Hyp
2. 𝛼(⌜𝛼(𝑥)⌝) 1 Def. 𝜓 Q (and T) ⊢ ¬PrfΓ (𝑘, 𝑚)
3. 𝑦(𝜃diag (⌜𝛼(𝑥)⌝, 𝑦) → 𝜑(𝑦)) 2 Def. 𝛼(𝑥)
A
(Here we use sans serif Prf to differentiate the
4. 𝜃diag (⌜𝛼(𝑥)⌝, ⌜𝜓⌝) → 𝜑(⌜𝜓⌝) 3 UI
object-language expression.)
5. 𝜑(⌜𝜓⌝) 4, (6a) above, MP
6. 𝜓 → 𝜑(⌜𝜓⌝) 1–5 DT
(8) Next we show T ⊢ 𝜑(⌜𝜓⌝) → 𝜓:
1. 𝜑(⌜𝜓⌝) Hyp Proof: Since Γ is a recursive set, Prf Γ is a recursive
2. 𝜃diag (⌜𝛼(𝑥)⌝, 𝑏) Hyp relation. Since Q (and also T itself) represents all
3. 𝜃diag (⌜𝛼(𝑥)⌝, 𝑏) → ⌜𝜓⌝ = 𝑏 (6b), UI recursive relations, there must be some formula
4. ⌜𝜓⌝ = 𝑏 2, 4 MP meeting the conditions above. ❅

67
Where T is a recursively axiomated theory, we shall
conventionally use PrfT interchangeably with some
one the formulas PrfΓ , where Γ is recursive and ax-
iomatizes T. E.g., PrfQ (bold Q) for PrfQ (regular Q)
and PrfPA for PrfPA . Strictly speaking, Prf Γ and Prf Δ
may be different relations even when both Γ and Δ
axiomatize the same theory T, and so PrfΓ and PrfΔ
are different wffs, but in general it is unimportant
which we pick.
Let us also use the notation:
TA ⊢ ¬ 𝑦¬𝜑(𝑦). But then for every 𝑛, N ⊨ ¬𝜑(𝑛).
A
Since the numerals exhaust the domain of quan-
tification for N, it would hold that N ⊨ 𝑦¬𝜑(𝑦),
A
and thus TA ⊢ 𝑦¬𝜑(𝑦). But this would make TA
A
inconsistent, which we know not to be true. ❅
Corollary: Q, PA, and TA are 𝜔-consistent.

ProvT (𝑥) abbreviates 𝑦 PrfT (𝑦, 𝑥)

E
Two senses of completeness
We can think of this as a “provability” predicate:
there exists some proof of the formula with Gödel On page 25, we gave this definition:
number 𝑥 in (one of the axiomatizations of) T. Definition: A set of sentences Γ is complete iff for
every closed wff 𝜑, either 𝜑 ∈ Γ or ¬𝜑 ∈ Γ.
𝜔-consistency We can apply this to theories.
Definition: A set of formulas Γ is 𝜔-consistent Definition: A closed wff 𝜑 is an undecidable sen-
iff for every formula 𝜑(𝑦) with one free variable, if tence of T iff both T ⊬ 𝜑 and T ⊬ ¬𝜑.
E
Γ ⊢ ¬𝜑(𝑛) for every number 𝑛, then Γ ⊬ 𝑦𝜑(𝑦).

Result: If T is a theory, then it is complete iff

Result: If a theory T is 𝜔-consistent, then it is it has no undecidable sentences.
consistent.

Proof: By the definition of a theory, T is closed

Proof: Suppose T is 𝜔-consistent but suppose for under entailment, and so 𝜑 ∈ T iff T ⊢ 𝜑. ❅
reductio that it is inconsistent. Then every sentence This notion of completeness is due to Emil Post, and
can be proven in T, and so for every open sentence for me it is tempting to write it as completenessPost ,
it would be true that T ⊢ ¬𝜑(𝑛) for every number since there is another sense of completeness used
𝑛, and also that T ⊢ 𝑦𝜑(𝑦), contradicting its 𝜔- in logic: the sense of capturing everything valid
E
consistency. ❅ according to the semantics at play, sometimes
called “semantic completeness”, or if we like, “⊨-
completeness”.
Result: If T ⊆ TA, then T is 𝜔-consistent. These aren’t the same notion. In some contexts,
they’re barely related. When we proved that first-
order logic with identity is complete in the sense
that if 𝜑 is identity-valid, then ⊢= 𝜑, we certainly
Proof: Suppose that T ⊆ TA but suppose for reduc- weren’t proving, or attempting to prove that either
tio that T is not 𝜔-consistent. There is then some ⊢= 𝜑 or ⊢= ¬𝜑 for every 𝜑. If 𝜑 is contingent,
formula such that T ⊢ ¬𝜑(𝑛) for every number neither is, nor should be, true.
𝑛, and also that T ⊢ 𝑦𝜑(𝑦), and therefore that However, if the semantics at play is truth in
E
TA ⊢ ¬𝜑(𝑛) for every number 𝑛, and also that just one intended interpretation, such as truth-in-N,
TA ⊢ 𝑦𝜑(𝑦). From the latter and simple logic, there is more of a connection.
E

68

Result: If for all 𝜑 such that N ⊨ 𝜑, it holds
that T ⊢ 𝜑, then T is complete. (Or if 𝜑 is N-⊨-
complete then it is completePost .)
Gödel’s first incompleteness
theorem
Suppose a sentence  says of itself that it’s not
provable. If we could prove it, what we would prove
is false. If we can’t prove it, there is something true
we cannot prove. Frustrating.

Proof: Suppose T proves all sentences true in the

standard interpretation. For every sentence 𝜑, ei- Result: If theory T is any 𝜔-consistent, recur-
ther it or its negation is true in the standard inter- sively axiomatizable extension of Q, then T has
pretation. Hence, either T ⊢ 𝜑 or T ⊢ ¬𝜑. ❅ at least one undecidable sentence, T , called “the
Gödel sentence for T”.

Corollary: TA is complete.
Proof: (1) Suppose T is such a theory.
(2) (Prf ⊢) applies, and so we have a wff PrfΓ (𝑥, 𝑦),
in terms of which we can define ProvT (𝑥).
Proof: TA meets the condition in the previous re- (3) Consider the wff ¬ProvT (𝑥). This is a wff with
sult by definition. ❅ only one free variable. By the fixed point theo-
rem, there is a wff T such that:
In the other direction, if a theory is completePost ,
and N is a model for it, then it is N-⊨-complete as T ⊢ T ↔ ¬ProvT (⌜T ⌝)
well. ( says of itself, “I’m not provable in T!”)
T

(4) First, we show that T ⊬ T .

Result: If a theory T ⊆ TA is complete, then
for all 𝜑 such that N ⊨ 𝜑, it holds that T ⊢ 𝜑.
Proof: Suppose T is such a theory and suppose N ⊨
𝜑. Since T is complete, either T ⊢ 𝜑 or T ⊢ ¬𝜑.
But suppose T ⊢ ¬𝜑 for reductio. Since T ⊆ TA,
it follows that TA ⊢ ¬𝜑 and therefore N ⊨ ¬𝜑.
However, it cannot be that both N ⊨ 𝜑 and N ⊨ ¬𝜑.
Hence it must be that T ⊢ 𝜑. ❅ • Hence our assumption that T ⊢ T is impos-
So in this context, there is a kind of convergence (5) By (4), we know that there is no number 𝑛
between the two senses. But this is not the case in
every context, because in other contexts, we may be
interested in more than just a single interpretation.
In effect, Gödel showed that PA is incomplete
in both senses. How? Well, speak of the devil, it’s
finally time for . . . (drumroll please) . . .
• Suppose for reductio that T ⊢ T .
• Then Prf T (𝑘, # T # ) holds for some 𝑘.
• By (Prf ⊢), T ⊢ PrfT (𝑘, ⌜T ⌝).
• By EG, T ⊢ 𝑦 PrfT (𝑦, ⌜T ⌝).
E
• This is the same as T ⊢ ProvT (⌜T ⌝).
• But by (3) and our assumption, T ⊢
¬ProvT (⌜T ⌝).
• Hence T is inconsistent.
• However, T must be consistent, as it is 𝜔-
consistent.
sible.
for which Prf T (𝑛, # T # ) holds. Hence for every
number 𝑛, not-Prf T (𝑛, # T # ) instead. By (Prf ⊢),
we have, for every number 𝑛:
T ⊢ ¬PrfT (𝑛, ⌜T ⌝)
(6) Since T is 𝜔-consistent, from (5), it follows that
E
T ⊬ 𝑦 PrfT (𝑦, ⌜T ⌝)

69
(7) But this means that T ⊬ ¬T . Proof: Every extension of PA is also an extension
• Suppose otherwise, i.e., that T ⊢ ¬T . of Q, so the first corollary above applies. ❅

• By (3), T ⊢ ProvT (⌜T ⌝).

• But this is T ⊢ 𝑦 PrfT (𝑦, ⌜T ⌝), which
E
Corollary: TA is not recursively axiomatiz-
we have already shown impossible. ❅ able.
(8) By (4) and (7), we get that T is an undecidable
sentence of T.

Proof: TA is an 𝜔-consistent extension of Q. If it

were recursively axiomatizable, it would be incom-
Corollary: Every recursively axiomatizable, 𝜔- plete. However, it is complete. ❅
consistent extension of Q is incomplete.

Corollary: PA is not TA.

Proof: By the first incompleteness theorem, every

such theory has at least one undecidable sentence.❅
Proof: PA is recursively axiomatizable whereas TA
is not. (One is complete, the other is not, and so
Corollary: PA is incomplete. on . . . ) ❅

Homework 31
Proof: PA is a recursively axiomatizable, 𝜔-consis- Let TheoremPA be the property a number has just
tent extension of Q, and so the previous corollary in case it is the Gödel number of a wff 𝜑 such that
applies. ❅ PA ⊢ 𝜑. Prove that despite what one may assume,
TheoremPA is not representable in PA by the wff
ProvPA (𝑥) as defined earlier, at least using the tech-
Corollary: There are sentences true in the stan- nical definition of “representable”.
dard interpretation N not provable in PA.
Homework 32
Show that there are infinitely many non-equivalent
interpretations that are models of PA. (interpreta-
tions are equivalent when they make precisely the
Proof: This follows from our earlier result that all
same set of sentences true.) Hint: consider the infi-
N-⊨-complete theories are completePost . Indeed,
nite series:
it is clear that PA is itself such a sentence, as it
asserts of itself that it is not provable in PA, and is PA , PA+ , PA++ , PA+++ , …
correct about that. ❅
Where this is the series of Gödel sentences, starting
with that of PA (as axiomatized by PA), then that
Corollary: Every 𝜔-consistent, recursively ax- of the theory PA+ adding PA as an axiom, then
iomatizable extension of PA is incomplete. that of the theory PA++ got by adding PA+ as an
axiom, and so on. Consider also the Satisfiability
result and homework 12 from our previous unit.

70
The Gödel-Rosser theorem
The idea of a consistent but non-𝜔-consistent arith-
metical theory seems weird. It’s sort of like that
friend who says they want to get together, only
not on Monday, and not on Tuesday, and not on (7) Let’s show that T ⊬ T .
Wednesday, and not next Monday either, and so on.
Still, one might be tempted to endorse such a
theory if it were a route to completeness. But J. B.
Rosser is a spoilsport.
Result: If theory T is a consistent, recursively
axiomatizable extension of Q, then T has at least
one undecidable sentence, T , called “the Rosser
sentence for T”.
Proof: (1) Suppose T is such a theory.
(2) (Prf ⊢) applies, so we can again use object lan-
guage formula PrfT (𝑥, 𝑦).
(3) The function Neg(𝑘), whose value for 𝑘 as ar-
gument, is the Gödel number of the negation
of the formula with Gödel number 𝑘, is primi-
tive recursive, and thus representable in T. Let
neg(𝑥, 𝑦) represent it, so, for every wff 𝜑:
T ⊢ neg(⌜𝜑⌝, ⌜¬𝜑⌝)
A
T ⊢ 𝑦(neg(⌜𝜑⌝, 𝑦) → ⌜¬𝜑⌝ = 𝑦)
(4) Consider the wff, hereafter abbreviated 𝜌(𝑥):
A A
𝑦 [PrfT (𝑦, 𝑥) → 𝑧 (neg(𝑥, 𝑧) →
E
𝑤(𝑤 < 𝑦 ∧ PrfT (𝑤, 𝑧)))]
(5) The fixed point theorem applies, and so there
is a wff T such that:
T ⊢ T ↔ 𝜌(⌜T ⌝)
Given how we’ve defined 𝜑(𝑥), T says of itself
that for every proof of it, there’s a proof with a
smaller Gödel number of its negation (or “all”
its negations). Or even more crudely, T says
“if I’m provable, so is my negation!”
71
(6) By applying (3) to T itself, we have:
T ⊢ neg(⌜T ⌝, ⌜¬T ⌝)
A
T ⊢ 𝑦(neg(⌜T ⌝, 𝑦) → ⌜¬T ⌝ = 𝑦)
• Suppose for reductio that T ⊢ T .
• From (5), we get T ⊢ 𝜌(⌜T ⌝), i.e., T ⊢
A A
𝑦 [PrfT (𝑦, ⌜T ⌝) → 𝑧 (neg(⌜T ⌝, 𝑧) →
E
𝑤(𝑤 < 𝑦 ∧ PrfT (𝑤, 𝑧)))]
• Prf T (𝑘, # T # ) holds for some number 𝑘, and
by (Prf ⊢), T ⊢ PrfT (𝑘, ⌜T ⌝).
• Together, we get T ⊢ 𝑧 (neg(⌜T ⌝, 𝑧) →
A
𝑤(𝑤 < 𝑘 ∧ PrfT (𝑤, 𝑧))).
E
• With (6), we then have T ⊢ 𝑤(𝑤 <
E
𝑘 ∧ PrfT (𝑤, ⌜¬T ⌝)); by simple logic, T ⊢
A
¬ 𝑤(𝑤 < 𝑘 → ¬PrfT (𝑤, ⌜¬T ⌝)).
• T consistent and so it must be that T ⊬ ¬T .
• But then for every number 𝑛, and a fortiori,
for every number less than 𝑘, by (Prf ⊢):
T ⊢ ¬PrfT (𝑛, ⌜¬T ⌝)
• In the object language, assume that 𝑏 < 𝑘.
By (Spread), 𝑏 = 0 ∨ … ∨ 𝑏 = 𝑘 − 1. By
a big proof by cases ¬PrfT (𝑏, ⌜¬T ⌝). Dis-
charging and generalizing, T ⊢ 𝑤(𝑤 < 𝑘 →
A
¬PrfT (𝑤, ⌜¬T ⌝)).
• But then T ⊢ ⊥, which is impossible.
(8) Step (7) ensures us that for every number 𝑛,
not-Prf T (𝑛, # T # ), and so, T ⊢ ¬PrfT (𝑛, ⌜T ⌝)
by (Prf ⊢).
(9) Now we show T ⊬ ¬T .
• Suppose for reductio that T ⊢ ¬T .
• Then for some 𝑗, Prf T (𝑗, # ¬T # ), and so, by
(Prf ⊢), T ⊢ PrfT (𝑗, ⌜¬T ⌝).
• Suppose in the object language that
PrfT (𝑏, ⌜T ⌝).
• By (Trichotomy), we have 𝑏 < 𝑗 ∨𝑗 < 𝑏∨𝑏 = 𝑗.
• We can rule out 𝑏 = 𝑗, because, by (8), we
have T ⊢ ¬PrfT (𝑗, ⌜T ⌝), which contradicts
our supposition.
• We can also rule out 𝑏 < 𝑗. For suppose 𝑏 < 𝑗.
Then, by (Spread), we have 𝑏 = 0 ∨ … ∨ 𝑏 =
𝑗 − 1, and each disjunct, with (8), leads to a
contradiction with our supposition.
 • Hence 𝑗 < 𝑏. So 𝑗 < 𝑏 ∧ PrfT (𝑗, ⌜¬T ⌝), and
then, by EG, 𝑤(𝑤 < 𝑏 ∧ PrfT (𝑤, ⌜¬T ⌝)). Result: The derivability conditions hold for
E
• Make another supposition in the object lan- ProvPA (𝑥) for PA, i.e., for all sentences 𝜑 and
guage: neg(⌜T ⌝, 𝑐). Then, by (6), ⌜¬T ⌝ = 𝜓:
𝑐, and we get 𝑤(𝑤 < 𝑏 ∧ PrfT (𝑤, 𝑐)) by (HB1) If PA ⊢ 𝜑 then PA ⊢ ProvPA (⌜𝜑⌝).
E
LL. Discharging and generalizing, we get: (HB2) PA ⊢ ProvPA (⌜𝜑 → 𝜓⌝) →
𝑧 (neg(⌜T ⌝, 𝑧) → 𝑤(𝑤 < 𝑏 ∧ PrfT (𝑤, 𝑧))).
A E
(ProvPA (⌜𝜑⌝) → ProvPA (⌜𝜓⌝))
• Discharging and generalizing the earlier (HB3) PA ⊢ ProvPA (⌜𝜑⌝) →
assumption PrfT (𝑏, ⌜T ⌝), we get T ⊢ ProvPA (⌜ProvPA (⌜𝜑⌝)⌝)
𝜌(⌜T ⌝).
• But then, by (5), we get T ⊢ T , which would
make T inconsistent, which is impossible.
(10) By (7) and (9), T is an undecidable sentence Sad excuse: While proving (HB1) for ProvPA (𝑥) can
of T. ❅ be done using results we have studied so far, es-
tablishing (HB2) and (HB3) are too complicated to
be worth our time. They involve digging in to the
exact definition of PrfPA (𝑥, 𝑦), which comes from
the recursive definition of the relation Prf PA from
Corollary: Every consistent, recursively axiom- the initial functions, and how this translates into
atizable extension of PA is incomplete. the object language with many nested uses of 𝜑𝛽 ,
which is a nightmare. Moreover, our main interest
is using these to help demonstrate what PA can-
not prove, and clearly, if we were worried that the
above did not hold, this would not improve our
Proof: Every extension of PA is also an extension
estimation of what PA can do.
of Q, so the Gödel-Rosser theorem applies, and it
has at least one undecidable sentence. ❅ Homework 33
Prove that (HB1) holds for ProvPA (𝑥). Hint: use
The Hilbert-Bernays derivability (Prf ⊢).
conditions
Definition: The formula ProvT (𝑥) for theory T Gödel’s second incompleteness
meets the Hilbert-Bernays derivability condi- theorem
tions iff the following are true for all sentences 𝜑
and 𝜓: PA is consistent, but PA cannot prove its own con-
sistency.
(HB1) If T ⊢ 𝜑 then T ⊢ ProvT (⌜𝜑⌝).
(HB2) T ⊢ ProvT (⌜𝜑 → 𝜓⌝) → For any theory T for which we have a prov-
(ProvT (⌜𝜑⌝) → ProvT (⌜𝜓⌝)) ability predicate ProvT (𝑥), let ConT be the wff
(HB3) T ⊢ ProvT (⌜𝜑⌝) → ProvT (⌜ProvT (⌜𝜑⌝)⌝) ¬ProvT (⌜⊥⌝).

If you are familiar with modal logic, you may notice

a strong similarity between these and the princi- Result: PA ⊬ ConPA
ples/rules of modal system S4.

Proof: (1) Recall the Gödel sentence for PA; we

have, PA ⊢ PA ↔ ¬ProvPA (⌜PA ⌝).

72
 (2) PA ⊢ PA → ¬ProvPA (⌜PA ⌝) by (1) and BE.
(3) PA ⊢ PA → (ProvPA (⌜PA ⌝) → ⊥) by (2) and
some easy logic.
(4) PA ⊢ ProvPA (⌜PA → (ProvPA (⌜PA ⌝) → ⊥)⌝)
by (3) and (HB1).
(5) PA ⊢ ProvPA (⌜PA ⌝) →
ProvPA (⌜ProvPA (⌜PA ⌝) → ⊥⌝)
by (4), (HB2) and MP.
(6) PA ⊢ ProvPA (⌜ProvPA (⌜PA ⌝) → ⊥⌝) →
[ProvPA (⌜ProvPA (⌜PA ⌝)⌝) → ProvPA (⌜⊥⌝)]
by (HB2).
(7) PA ⊢ ProvPA (⌜PA ⌝) →
[ProvPA (⌜ProvPA (⌜PA ⌝)⌝) → ProvPA (⌜⊥⌝)]
by (5) and (6) and HS.
(8) PA ⊢ ProvPA (⌜PA ⌝) →
ProvPA (⌜ProvPA (⌜PA ⌝)⌝) by (HB3).
(9) PA ⊢ ProvPA (⌜PA ⌝) → ProvPA (⌜⊥⌝)
by (7) and (8) and CMP.
(10) PA ⊢ ¬ProvPA (⌜⊥⌝) → ¬ProvPA (⌜PA ⌝)
by (9) and Trans.
(11) PA ⊢ ¬ProvPA (⌜PA ⌝) → PA by (1) and BE.
(12) PA ⊢ ¬ProvPA (⌜⊥⌝) → PA
by (10) and (11) and HS.
(13) ¬ProvPA (⌜⊥⌝) is ConPA , and so if it were true
that PA ⊢ ConPA , by MP, it would be true that
PA ⊢ PA , but we know it is not.
(14) Hence it must be that PA ⊬ ConPA . ❅ fixed point theorem tells us that there is a wff
Actually, it was discovered later that not only can (4) T ⊢ 𝜓 → (ProvT (⌜𝜓⌝) → 𝜑) by (3) and BE.
PA not prove of itself that it cannot prove a con- (5) T ⊢ ProvT (⌜𝜓 → (ProvT (⌜𝜓⌝) → 𝜑)⌝)
tradiction, it can never prove that it cannot prove
something, even if it cannot prove it.
Löb’s theorem
You may be familiar with the paradox variously
called Curry’s paradox or Löb’s paradox or the (8) T ⊢ ProvT (⌜𝜓⌝) →
Santa Claus paradox, which is another conse-
quence of self-reference run amok.
Let 𝑆 be the self-referential sentence:
“If 𝑆 is true, Santa Claus exists.”
We can use this to “prove” that Santa Claus exists.
1. 𝑆 is true. (Assumption)
2. If 𝑆 is true, Santa Claus exists. (1, Meaning of 𝑆) (12) T ⊢ 𝜓 by (3) and (11) and BMP.
3. Santa Claus exists. (1, 2 MP)
4. If 𝑆 is true, Santa Claus exists. (1–3 CP)
5. 𝑆 is true. (4, Meaning of 𝑆)
6. Santa Claus exists. (4, 5 MP)
Of course, the choice of “Santa Claus exists” here is
arbitrary. We could have used any other sentence
instead.
As we know, the fixed point theorem makes
self-reference possible in Q and PA, etc. We cannot
quite recreate Curry’s paradox, because we don’t
have a truth predicate. (We’ll prove we don’t soon.)
But if we have a provability predicate with certain
characteristics, we can wreak havoc in a similar
way, by forming sentences that say of themselves
if they’re provable, then something else.
Result: If T is an extension of Q for which
ProvT (𝑥) meets the Hilbert-Bernays derivabil-
ity conditions, then for any sentence 𝜑 if T ⊢
ProvT (⌜𝜑⌝) → 𝜑 then T ⊢ 𝜑.
Proof: (1) Let T be such a theory, and 𝜑 any wff.
(2) Suppose that T ⊢ ProvT (⌜𝜑⌝) → 𝜑.
(3) Consider the open sentence ProvT (𝑥) → 𝜑. The
𝜓 such that, T ⊢ 𝜓 ↔ (ProvT (⌜𝜓⌝) → 𝜑).
by (4) and (HB1).
(6) T ⊢ ProvT (⌜𝜓⌝) → ProvT (⌜ProvT (⌜𝜓⌝) → 𝜑⌝)
by (5), (HB2) and MP.
(7) T ⊢ ProvT (⌜ProvT (⌜𝜓⌝) → 𝜑⌝) →
(ProvT (⌜ProvT (⌜𝜓⌝)⌝) → ProvT (⌜𝜑⌝))
by (HB2).
(ProvT (⌜ProvT (⌜𝜓⌝)⌝) → ProvT (⌜𝜑⌝))
by (6) and (7) and HS.
(9) T ⊢ ProvT (⌜𝜓⌝) → ProvT (⌜ProvT (⌜𝜓⌝)⌝)
by (HB3).
(10) T ⊢ ProvT (⌜𝜓⌝) → ProvT (⌜𝜑⌝)
by (8) and (9) and CMP.
(11) T ⊢ Prov T (⌜𝜓⌝) → 𝜑 by (2) and (10) and HS.
(13) T ⊢ ProvT (⌜𝜓⌝) by (12) and (HB1).
(14) T ⊢ 𝜑 by (11) and (13) and MP. ❅

73
Löb’s theorem is extremely weird. Shouldn’t it be 11. Arithmetical Definability
true for every sentence, even a false one, that if it
were provable, then it’d be true? How could this
and Truth
enough to show that it is in fact true? Definition: A 𝑛-place relation 𝑅 between natural
numbers is arithmetically definable iff there is a
wff 𝜑(𝑥1 , … , 𝑥𝑛 ) of 𝐴 such that 𝑅(𝑘1 , … , 𝑘𝑛 ) iff N ⊨
Corollary: For any sentence 𝜑 of PA, if PA ⊢
𝜑(𝑘1 , … , 𝑘𝑛 ).
ProvPA (⌜𝜑⌝) → 𝜑 then PA ⊢ 𝜑.
Recall that we can consider properties to be 1-place
relations, so a property of numbers is arithmetically
Proof: PA meets the conditions for Löb’s theorem definable iff there is a 𝜑(𝑥) where 𝑘 has the property
to apply. ❅ iff 𝜑(𝑘) is true in N. We restrict ourselves to 𝐴
here, so 𝜑(𝑥) must use only =, <, +, ×, ′ , 0, and
logical symbols.
Corollary: There is no sentence 𝜑 such that
PA ⊢ ¬ProvPA (⌜𝜑⌝).
Result: If 𝑅 is representable in any theory T
where T ⊆ TA, then 𝑅 is arithmetically defin-
Proof: (1) Suppose PA ⊢ ¬ProvPA (⌜𝜑⌝) for reduc- able.
tio.
(2) By FA, PA ⊢ ProvPA (⌜𝜑⌝) → 𝜑.
(3) By the previous corollary, PA ⊢ 𝜑.
(4) By (HB1), PA ⊢ ProvPA (⌜𝜑⌝). Proof: (1) Suppose 𝑅 is representable in such a the-
(5) Hence PA ⊢ ⊥, which we know to be impossi- ory T.
ble. ❅ (2) Then there is a wff 𝜑(𝑥1 , … , 𝑥𝑛 ), where
a) if 𝑅(𝑘1 , … , 𝑘𝑛 ), then T ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 ); b) if
Note that we did not use the second incompleteness not-𝑅(𝑘1 , … , 𝑘𝑛 ) then T ⊢ ¬𝜑(𝑘1 , … , 𝑘𝑛 )
theorem in proving Löb’s theorem, and so we might (3) Suppose 𝑅 holds. Then T ⊢ 𝜑(𝑘 , … , 𝑘 ), and
1 𝑛
have got it instead directly from this corollary. since T ⊆ TA, TA ⊢ 𝜑(𝑘1 , … , 𝑘𝑛 ), and so N ⊨
In a sense, these results show that there are lim- 𝜑(𝑘1 , … , 𝑘𝑛 ).
its to the extent that PA captures its own metathe- (4) Suppose N ⊨ 𝜑(𝑘 , … , 𝑘 ) and for reductio sup-
1 𝑛
ory, even though its language can often express pose not-𝑅(𝑘1 , … , 𝑘𝑛 ). Then T ⊢ ¬𝜑(𝑘1 , … , 𝑘𝑛 ).
metatheoretic results about itself that it cannot But by reasoning similar to the previous direc-
prove. tion, this means that N ⊨ ¬𝜑(𝑘1 , … , 𝑘𝑛 ), which
Homework 34 is impossible. ❅
Consider the Henkin sentence, PA for PA, which
is like the “opposite” of the Gödel sentence: rather
than asserting its own unprovability, it asserts its
own provability. Answer these questions: Corollary: Every recursive relation is arith-
(a) What instance of the fixed point theorem gives metically definable.
us PA ?
(b) Is PA an undecidable sentence of PA like
PA ?
(c) If not, PA ⊢ PA or PA ⊢ ¬PA ? Proof: Every recursive relation is representable in
(d) Is PA true in the standard interpretation? Is Q and Q ⊆ TA. ❅
there any way to tell?

74
However, not all arithmetically definable relations
are representable in Q and PA. For example, the
property a number has just in case it is the Gödel
number of a theorem of PA is arithmetically defin-
able by ProvPA (𝑥), but it is not representable in PA
by that wff, nor any other. (We’ll prove that in the
next section.)
Definition: By extension, we say that property Φ of
sentences of a first-order language is arithmetically
definable iff the property of being the Gödel number
of a sentence having the property is arithmetically
definable in the above sense, i.e., if there is a wff 𝜑(𝑥)
of 𝐴 such that 𝜓 has Φ iff N ⊨ 𝜑(⌜𝜓⌝).
Result (Tarski’s theorem): The property of
truth in N is not arithmetically definable.
Proof: (1) Suppose for reductio that truth in N is
arithmetically definable.
(2) By the definition of arithmetical definability,
there is a wff which we could write as true(𝑥)
such that for all wffs 𝜓, N ⊨ 𝜓 iff N ⊨
true(⌜𝜓⌝).
(3) If we apply the fixed point theorem to ¬true(𝑥),
we get:
TA ⊢  ↔ ¬true(⌜⌝)
(An arithmetized version of the liar sentence of
the Epimenides/Liar paradox.)
(4) It cannot be that N ⊨ . For suppose oth-
erwise. Then, by (2), N ⊨ true(⌜⌝), which
would mean that TA ⊢ true(⌜⌝). Then, by (3),
TA ⊢ ¬, and hence both N ⊨  and N ⊨ ¬,
which is impossible.
(5) Since  is closed, from (4), N ⊨ ¬. But then,
TA ⊢ ¬, and by (3), TA ⊢ true(⌜⌝). This
would mean N ⊨ true(⌜⌝), and by (2), N ⊨ ,
and again,  and its negation are both true in
N. This is impossible
(6) Hence, our assumption at (1) must be mistaken,
which means truth in N is not arithmetically
definable after all. ❅ decidable.
75
Note that truth in N is, set-theoretically, the same
“property” as being provable in TA. We have seen
that, e.g., ProvPA (𝑥) doesn’t “represent” provabil-
ity in PA, even though it semantically means that
property. For TA, not only is there not an open wff
ProvTA that “represents” provability in TA (= truth)
there isn’t even a wff in the language of 𝐴 that
means that at all.
A question to ponder: what are the conse-
quences for the philosophy of mathematics that
arithmetical truth is not itself something that can
be studied in a purely arithmetical way?
Corollary: Truth in N is not a recursive prop-
erty.
Proof: All recursive properties are arithmetically
definable, but it is not. ❅
(Actually, we were in a position to prove this earlier.
The set of true sentences of arithmetic is the theory
TA. We proved earlier that TA is not recursively
axiomatizable. But trivially, TA axiomatizes itself.
Hence, TA cannot be a recursive set.)
12. Undecidability
Another corollary of Tarski’s theorem
Corollary: If the Church-Turing thesis is true,
there is no effective decision procedure for decid-
ing whether or not a sentence of the language of
arithmetic is true, as standardly interpreted.
Proof: According to the Church-Turing thesis, a
property is effectively decidable iff it is recursive.
Since the property of being (the Gödel number of) a
truth of arithmetic (being such that N ⊨ 𝜑 for a sen-
tence 𝜑 of 𝐴 ) is not recursive, it is not effectively
❅
But it gets worse. Not only is there no recursive • The same questions can be asked about Q or even
decision procedure for deciding truth, there is not barebones logic and the pattern of answers are
even a recursive procedure for deciding what is the same.
provable or not in a given consistent theory com-
patible with arithmetic, whether it be TA, or even The undecidability of arithmetical
PA, Q, or pure predicate logic!
theories
Definition: A theory T is recursively decidable
(or just decidable) iff it is a recursive set of sentences.
Otherwise, it is recursively undecidable. Result (recursive undecidability): If theory
T is a consistent extension of Q then T is recur-
Recall that theories are sets closed under entail-
sively undecidable.
ment. To say that a theory is undecidable is to say
that there is no recursive decision procedure for de-
termining what is entailed by a certain set of axioms.
A couple warnings: Proof: (1) Suppose that T is such a theory, and sup-
• Don’t confuse recursive decidability with recur- pose for reductio that T is recursively decidable.
sive axiomatizability. A theory may be the closure (2) This means the property of being the Gödel
of a recursive axiom set, without itself being a number of a member of T is a recursive prop-
recursive set. erty.
• Don’t confuse the issue of whether a theory is (3) Since T is a theory, 𝜑 ∈ T iff T ⊢ 𝜑.
recursively decidable with the issue of whether or (4) T is an extension of Q, and so it represents all
not it has undecidable sentences. The use of the recursive properties.
same word here is accidental. TA for example, is (5) There is therefore a wff 𝜏(𝑥), such that, for any
recursively undecidable even though it does not wff 𝜑:
have any undecidable sentences. (a) If T ⊢ 𝜑 then T ⊢ 𝜏(⌜𝜑⌝); and
(b) If T ⊬ 𝜑 then T ⊢ ¬𝜏(⌜𝜑⌝).
(6) Apply the fixed point theorem to ¬𝜏(𝑥) to get
Deciding proofs versus deciding
a Gödel-like sentence:
theorems
T ⊢  ↔ ¬𝜏(⌜⌝)
• Question 1: Is there an recursive method for de-
(7) It cannot be that T ⊢ , since then, by (5)-(a)
termining what is, and what is not, provable in a
T ⊢ 𝜏(⌜⌝), and by (6), T ⊢ ¬, which would
theory like PA?
make T inconsistent, which we have assumed
• Deriv, and Prf PA are recursive relations: if I give
is not the case.
you a alleged proof, there is an effective proce-
(8) But (7) means that T ⊬ , and so, by (5)-(b),
dure to check whether the proof is correct or
T ⊢ ¬𝜏(⌜⌝). But then by (6), T ⊢ . This we
incorrect.
have already shown to be impossible.
• Because of this, Prf PA is representable as
(9) Hence our assumption at (1) that T is recur-
PrfPA (𝑥, 𝑦) in PA itself.
sively decidable must be mistaken. ❅
• Using PrfPA (𝑥, 𝑦) we can define the wff ProvPA (𝑥).
• In a homework exercise, you showed that the
property of being the Gödel number of a member
of PA is not representable by this. Corollary: Q and PA are recursively undecid-
• Question 2: Can this property be represented in able.
PA at all?
• Since PA represents all and only recursive rela-
tions, these questions stand or fall together, and Proof: PA is an extension of Q, and Q is trivially
the answers to both are no. an extension of itself, and both are consistent. ❅

76
Finite extensions and decidability
Result (finite extension lemma): If theory
T is axiomatized by set Γ, and theory T∗ is
axiomatized by the set Γ ∪ {𝜑1 , … , 𝜑𝑛 }, adding
a finite number of sentences to Γ, then if T is
recursively decidable, so is T∗ .
Proof: Q is a finite extension of FOL𝐴 , as it is ax-
iomatized by the set Q = ∅ ∪ {Q1 , … , Q8 }. Hence, if
FOL𝐴 were recursively decidable, Q would be as
well, but we know Q is not. ❅
Corollary: FOL is recursively undecidable.

Proof: Suppose for reductio that FOL is recursively

Proof: (1) Suppose T and T∗ are such theories.
decidable. Then the property 𝑅FOL of being the
(2) Since T is axiomatized by Γ, we have that for
Gödel number of a member of FOL is as well. The
all 𝜓, T ⊢ 𝜓 iff Γ ⊢ 𝜓.
property of being the Gödel number of a sentence
(3) Similarly, for all 𝜓, T∗ ⊢ 𝜓 iff Γ ∪ {𝜑1 , … , 𝜑𝑛 } ⊢
of the language of arithmetic, Sent𝐴 , is primitive
𝜓.
recursive. The conjunction of any two recursive
(4) By DT and Imp in one direction, and Conj and
properties is recursive. But the conjunction of 𝑅FOL
MP in the other, it is clear that Γ ∪ {𝜑1 , … , 𝜑𝑛 } ⊢
and Sent𝐴 is equivalent to the property of being
𝜓 iff Γ ⊢ (𝜑1 ∧ … ∧ 𝜑𝑛 ) → 𝜓.
the Gödel number of a member of FOL𝐴 , which
(5) Hence, for all 𝜓, T∗ ⊢ 𝜓 iff T ⊢ (𝜑1 ∧ … ∧ 𝜑𝑛 ) →
we have just shown, is not recursive. Hence, FOL
𝜓.
is recursively undecidable. ❅
(6) Let 𝑅T be the property of being the Gödel num-
ber of a member of T.
(7) We can define the property 𝑅T∗ of being the
Gödel number of a member of T∗ this way: Corollary: The set of logically valid sentences
is not a recursive set of sentences.
𝑅T∗ (𝑥) iff 𝑅T (# ((𝜑1 ∧ … ∧ 𝜑𝑛 ) →# ⌢ 𝑥 ⌢ # )# )
(8) If 𝑅T is a recursive property, then this is a re-
cursive definition and 𝑅T∗ is recursive as well.
(9) In other words, if T is recursively decidable, so
is T∗ . Proof: By the soundness and completeness results
of our first unit, this set is precisely FOL. ❅

Logic itself is undecidable

Corollary: If the Church-Turing thesis is true,
These results are due to Alonzo Church. there is no effective decision procedure for de-
Definition: FOL𝐴 is the theory which is the clo- termining whether or not a formula is logically
sure of ∅ in the language of arithmetic 𝐴 . valid.

Definition: FOL is the theory which is the closure

of ∅ in a language using the full range of constants,
function letters and predicates.
Result: FOL is recursively undecidable.
𝐴
Proof: If the Church-Turing thesis were true, then
if there were such a procedure, it would have to
be recursive, but we know the set of logically valid
formulas is not a recursive set. ❅

77
 Corollary: If the Church-Turing thesis is true,
there is no effective decision procedure for deter-
mining in general whether or not an argument
is logically valid.

Proof: If there were such a procedure for determin-

ing whether or not Γ ⊨ 𝜑 in general, it would apply
also to the case ∅ ⊨ 𝜑, but this is the same as deter-
mining whether or not 𝜑 is valid. ❅

Homework 35
Prove that if the Church-Turing thesis is true, there
is no effective procedure for determining whether
or not two sentences 𝜑 and 𝜓 are logically equiv-
alent. (This is the bane of my attempt to have a
computer check my students’ Phil 110 translation
exercises.)

Homework 36
Very briefly (no more than a paragraph), describe
what you take to be the philosophical or practi-
cal importance of the results in this unit, such as
Gödel’s results, Tarski’s theorem, recursive unde-
cidability, etc.

78