Contents

1 Derivations 2
1.1 Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Illustrating the Difference Between a Valid Argument and a

Proof: A Richard Feynman Story 3

3 A Hilbert-Style Derivation System for First Order Logic 15

3.1 The Logical Axioms . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Proper Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 The Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.4 The Soundness of the Rules . . . . . . . . . . . . . . . . . . . 24
3.5 Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.6 The Compactness Theorem - Statement and Proof . . . . . . 29
3.6.1 The Theorem - Statement . . . . . . . . . . . . . . . . 29
3.6.2 Compactness Theorem - The Proof . . . . . . . . . . . 35

4 Size of models: basic facts 39

4.1 Compactness in Action: Non-Standard Models of Arithmetic . 50

5 Aside for the Mathematicians in Class: Compactness in Non-

Foundational Mathematics 56

1
1 Derivations

1.1 Basics

Two ways for an argument to be good:

semantic (having to do with preserving truth no matter

what the interpretation)

syntactic (pertaining to the relations among the sym-

bols), no necessary reference to interpretations.

In the past lectures, we’ve discussed the semantic idea,

symbolized with the ⊨ - which was that was specified pre-
cisely by formal interpretations (and truth-tables in the
propositional case).

The syntactic sense of “logically good” — which we will

write with the single turnstile “⊢” — captures an impor-
tantly different sense of “good argument” as I’m about
to explain.

2
2 Illustrating the Difference Between a Valid Argument and a
Proof: A Richard Feynman Story

OK, so what does the concept made precise by “⊨” leave

out?

There’s an amusing anecdote about the physicist Richard

Feynman that illustrates the difference in a practical case.

Feynman was not just a great physicist, but exceptionally

fast at jumping to correct conclusions in his head, almost
as if by instinct.

3
One time a PhD candidate was asking the physicist Mur-
ray Gell-Mann about some techniques Feynman had in-
vented for solving problems.

Gell-Mann explained the techniques to the student, and

the student asked “Are these really the techniques Feyn-
man uses?”

Gell-Mann replied: “Well, no, those are the techniques

that Feynman invented and that you should use.

The technique Feynman himself actually uses is:

4
a) Write down the problem.
b) Think really hard. (Gell-Mann presses his fingers to
his temple and makes a “concentrate really hard” face.)
c) Write down the correct answer.”

Any arguments emerging from the a) - c) procedure were

undoubtably valid, in the sense that it wasn’t possible for
the premises to be true and the conclusion false, but that
isn’t what the student was looking for.

The student was looking for ways to make the connection

between problem and solution clear to anyone, not just
people with Feynman’s inspiration.

5
It’s not enough that the premises in principle guarantee
the truth of the conclusion, we want something that will
convince us if we were unconvinced, and find the answer
if there is an answer there to be found.

We sometimes want a proof - a derivation.

It will leave no room for doubt.

How do we do that?

Since Euclid, and no doubt earlier, we have had the idea

that to give a demonstration, you need to take an infer-
ence from A to B, and break it down into a series of very
small steps.

6
So let’s say as a first approximation, that we will need
to refine a little: a proof, or derivation of B from A is a
series of sentences

A, A1, A2, A3, . . . An, B,

where the transition from each sentence to the next is

obvious.

OK - so what does “obvious” mean?

Just like many of the other ideas we study here, “obvi-

ous” is an idea that serves us perfectly well in everyday
life, but as you increase the magnification it starts to get
a little blurry.

7
The philosophy majors in the class will recall trying to
figure out what Réné Descartes meant by “Clear and Dis-
tinct” perception in his Meditations on First Philoso-
phy, which was his effort to work out such an idea of
“sufficiently small step to be obvious”.

All credit to Descartes, who saw and articulated many

deep things well, but as far as the topic presently at hand
is concerned, you don’t really come away from the Med-
itations with a much better idea of “obvious step” than
you went in with.

As usual, the logician’s response to this kind of unclarity

is to look for some precise, specific, clear approximation.

It may not capture every aspect of the informal idea of

“obvious step”, but it will be sharp enough that we can
do exact work with it.

The idea we use is that an inference is obvious in the rel-

evant sense if it turns on an certain aspects of the logical
form of the sentence.

8
Let’s look at a couple of inferences:

If Andrea eats cheese then she’ll get ill.

Andrea ate cheese.
Therefore: Andrea will get ill.

If Alan goes to the store then he’ll buy incense.

Alan is going to the store.
Therefore: Alan will buy incense.

If we were to symbolize these in the language of proposi-

tional logic, we get the same thing (using obvious abbre-
viations):

A→I
A
…
I

9
Both of those are logically valid inferences, and in addi-
tion they are pretty basic and immediate. Obvious, even.

We can take the scheme:

A → I, A therefore I

as a “formal inference rule”: it gives permission to make

an inference, based solely on the form of sentences.

Any time you have sentences of the form A → I and A,

you are permitted to write down I

10
We don’t have to even interpret the language with truth-
tables or first-order interpretations to study derivations.

We can just say: any derivation that conforms to this

pattern is correct.

This is crucial if we want a computer to be able to do

this work, for example.

It has to be automatic, require no special insight or un-

derstanding to see an inference is correct.

11
Let’s return to the hand-waving first approximation:

A proof of B from A is a sequence of sentences

A, A1, A2, A3, . . . An, B,

where each transition from a sentence to the next is ob-

vious.

We need to modify the statement so that it can include

more premises than just one.

12
To give us a uniform way to refer to all lines of the deriva-
tion, let’s count the conclusion as the last of the letters
with subscripts.

And we want to use our sharpened idea of “obvious”:

A proof of An from A1, . . . Am is a sequence of sentences

A1, A2, A3, . . . Am, . . . An, where each sentence is
either A1, . . . Am or follows from earlier lines according
to an acceptable formal rule.

13
When we have a derivation of S from premises Γ, we will
write Γ ⊢ S.

As noted above, we call ⊢ the “single turnstile”

The topic of the coming two or three lectures will be to

work out a formal system in first order logic for giving
proofs of formulae from collections of premises.

ii) Study the properties of these systems, in particular

the relationships between ⊢ and ⊨.

14
3 A Hilbert-Style Derivation System for First Order Logic

There are different kinds of derivation systems for FOL,

each of which has its own strengths and weaknesses.

Those of you who have taken Phil 303 will have learned a
system of natural deduction, which is geared to studying
the structure of arguments from hypotheses.

BJB use a more theoretical system called a sequent calcu-

lus, which is useful for studying certain kinds of metathe-
oretic properties of systems, but isn’t all that ideal for a
course like this.

We will use a minimalist approach called a “Hilbert-style”

system that is close to the original model of Euclid: lay
down axioms, and view derivations as drawing conse-
quences from those axioms, sometimes supplemented by
additional assumptions.

I think the Hilbert-style approach is the most straightfor-

ward one to use in connection with the Gödel theorem.

15
3.1 The Logical Axioms

There are just five axioms for the system.

The first three are propositional inferences, not involving

quantifiers at all.

The propositional inferences may seem like a bizarre, ran-

dom collection, but in a few slides you’ll see why they
make sense.

Say that B, C, D are well-formed formulae (wffs, to use

the familiar abbreviation used by Mendelson) of the lan-
guage of FOL.

16
The following are axioms of FOL:

Propositional Axioms:
A1) B → (C → B)
A2) (B → (C → D)) → (B → C) → (B → D)
A3) (∼ C →∼ B) → ((∼ C → B) → B))

Quantifier Axioms
A4) ∀xB(x) → B(t) if t is any term. (See qualification
on the next slide)

A5) ∀x(B → C) → (B → ∀xC) if x doesn’t occur free

in B.

17
Mendelson adds a qualification to A4) using a phrase
“free for x in B(x)”. Rather than explain this new vocab-
ulary which we’ll never use again, I’ll just explain what
the qualification does.

The qualification in A4) is that the term t can’t contain

any free variables that would become bound variables if
t were substituted in B(x).

Here is the sort of thing that condition is meant to rule

out.

Say that t is f (y) and B(x) is the formula ∀yF xy.

If you put f (y) in ∀yF xy you would get ∀yF f (y)y, so

the free variable in f (y) would become bound by the ∀y.
That isn’t supposed to happen.

I mention this “for the record” but there won’t be any

time in the course where we need to know this.

18
3.2 Proper Axioms

In addition to the logical axioms, we will want to give

specific theories of some subject or other.

In particular, in a couple of weeks we’ll lay out some ax-

ioms for formal arithmetic.

When we need to make the distinction, we’ll call these

“proper axioms” to distinguish them from the logical ones.

19
3.3 The Rules

A difference with the classical geometry of Euclid is that

we also have to specify explicitly what formal rules of in-
ference we are going to use.

Remarkably, we will only need two of them. For any wffs

of FOL B and C;

Modus Ponens: B, B → C ⊢ C

Generalization:: B(x) ⊢ ∀xB(x)

The generalization rule might seem a bit odd: doesn’t

that mean that you can generalize from B(x) being true
of one thing to B(x) being true of everything?

In fact, the rule isn’t saying that, since x here is a vari-

able, not a name. So you can see the rule as saying
roughly that if you know B(x) is true / satisfied with-
out knowing what x is, then you can assert that B(x) is
true of everything.

20
A quick explanation:

To keep the treatment of derivations and metatheory

streamlined, I’ve slightly modified the BJB system to be
closer to Mendelson, in ways that won’t make a difference
to us going forward.

These changes are only going to be in use for the next

few lectures.

In BJB you reason with formulae that are sentences -

they have no free variables.

In Mendelson you can use formulae with free variables.

This simplifies the generalization rule — if you are only

reasoning with sentences then you need to put a bunch
of restrictions on how it is used, because you have names
instead of variables.

Those of you who have taken 303 with me know what i

mean.

21
I’d rather avoid those complications right now, as they
won’t matter for what we will be doing in the second half
of the course..

So for the next few lectures, we’ll reason with wffs, and
though I will occasionally speak of wffs being true, I mean
“true if sentences, satisfied if formulae that are not sen-
tences”.

22
I want to pause over the generalization rule for a moment,
because it illustrates a crucial point about the difference
between → and ⊢.

→ is a symbol in the formal language FOL - it serves to

connect formulae of the language of FOL to make other
formulae of the language of F OL.

⊢ is not a symbol in the formal language of FOL. It is a

symbol we are using to reason about FOL.

As one case where they diverge, note that:

B(x) ⊢ ∀xB(x) — that is just the rule of Generalization.

But B(x) → ∀xB(x) is not a logical truth.

23
3.4 The Soundness of the Rules

Now we have two different ideas of good argument rep-

resented by ⊢ and ⊨.

What is the relationship between the two?

Is it possible that for some set of formulae Γ and a for-

mula S, Γ ⊢ S but Γ ̸⊨ S?

This would be a problem for the rules, because it would

mean that you could use the rules to go from true premises
to a false conclusion.

That’s as serious a vice as logical rules can have.

24
In fact, the system determined by the rules and axioms
I’ve given you is sound in the sense that:

If Γ is a set of sentences of the language of FOL and S is

a sentence of the language of FOL, then:

Γ⊢S⇒Γ⊨S

A more informal way to put this is that if the premises

in Γ are true in an interpretation, then anything you de-
rive from them using this system will also be true in that
interpretation.

I’ll leave the proof of the soundness theorem for later.

25
3.5 Completeness

What about the other direction?

Do we know that:

Γ⊨S⇒Γ⊢S?

If this were not true, it would not be as major a defect as

unsoundness would be.

But it wouldn’t be ideal, because there would be some

arguments that would be valid in the sense of ⊨ where
our derivation system couldn’t derive the conclusion from
the premises.

26
But I’m happy to tell you that we do know that

Γ⊨S⇒Γ⊢S

That is, our derivation system for FOL is complete

The proof of completeness is a bit tricky, so I’ll leave it

until Monday.

27
So putting the two sides together, we have that:

Γ⊨S⇒Γ⊢S

I’m putting off the proofs of soundness and completeness

so that we can put this equivalence to use first, to see
some of the things we can derive from it.

One particularly deep result is the next thing on the

agenda: The compactness theorem

28
3.6 The Compactness Theorem - Statement and Proof

3.6.1 The Theorem - Statement

The compactness theorem is among those results that get

more surprising the more you know about it.

Like watching a sport you know a lot about with some-

one else who doesn’t, there is an irresistible urge to shout:
look you should really be amazed by this!

Some terminology: I’ll say that a set of sentences Γ has

a model if there is an interpretation that makes every
sentence in Γ true.

29
Compactness theorem (form 1): If a set of sentences Γ
has no model, then some finite subset of Γ has no model.

Compactness theorem (form 1 contrapositive): If every

finite subset of a set of sentences Γ has a model, then Γ
has a model.

Compactness theorem (form 2): If for some set Γ and

sentence S we have Γ ⊨ S, then there is some finite
subset Γ′ ⊆ Γ such that Γ′ ⊨ S.

30
Aside:

The name “compactness” is shared with a related prop-

erty in real analysis called the compactness property (some-
times the Heine-Borel property if we are speaking about
entire spaces):

(By “real analysis” I mean calculus and the developments

arising out of it that pertain to the real numbers and func-
tions of them.)

A set S ⊆ R is said to be compact if given any covering

C of S with a family of open sets, there is a finite subset
of C (a finite subcovering) that also covers S.

The Heine-Borel Theorem says that any closed, bounded

subset of Rn is compact.

31
There is a profound connection between the topological
fact about spaces and the logical fact about consequences
of sets of sentences.

For those who know what this means:

It is in fact possible to prove compactness for proposi-

tional logic directly from the Tychonoff Theorem on the
compactness of the product of compact topological spaces
and then extend that to first-order logic.

32
The way we will prove compactness today is to rely on a
theorem that I’ll prove Monday called the Completeness
Theorem (for First Order Logic), that pertains to the re-
lations between ⊢ and ⊨.

With completeness in hand, the proof of compactness is

quite simple, almost an afterthought.
In fact, compactness is a deeper fact just about semantic
entailment, that can be proven without appeal to com-
pleteness, and it is a strength of the BJB approach that
it displays this depth by proving compactness first, with
completeness proven subsequently.

Such a direct approach to compactness was first, I be-

lieve, worked out by Bruno Poizat. See his A Course in
Model Theory, Springer 2000

Poizat’s method is different from BJB’s, and employs

a very cool, somewhat topological, construction known
as the ultraproduct construction, which Dhruv may be
telling you about in a few weeks.

33
One consequence of the Compactness theorem:

Any set of sentences with arbitrarily large finite models

must also have an infinite model.

(This is a question on the problem set.)

34
3.6.2 Compactness Theorem - The Proof

I mentioned that in BJB there is a direct, and very pro-

found argument for the compactness theorem, and then
the completeness theorem is proven as a consequence of
the compactness theorem.

From the point of view of the proper arrangement of

ideas, I think that is the right way to go – compactness
is really a more fundamental fact about FOL than com-
pleteness is.

(I will admit that it is hard to specify precisely what I

mean by that last sentence.... But one point is that com-
pleteness holds or doesn’t depending on the derivation
system you use, while compactness is provable without
reference to any derivation system.)

However, from the point of view of a logic teacher, it’s eas-

ier to approach things the other way around: prove com-
pleteness first and then derive compactness from that.

On this approach, the compactness theorem falls out of

the completeness and soundness theorems as almost an
afterthought.

35
Here is the reasoning for form 2 of compactness from com-
pleteness and soundness, We want to show:

Theorem: If for some set Γ and sentence S we have

Γ ⊨ S, then there is some finite subset Γ′ ⊆ Γ such that
Γ′ ⊨ S.

Say we have a set Γ and sentence S of FOL such that:

Γ⊨S

By the completeness theorem, Γ ⊢ S.

36
That is, there is a derivation of S whose only premises
are sentences of Γ.

Since every derivation is finite, there must be a finite sub-

set Γ′ ⊆ Γ such that Γ′ consists of exactly the sentences
of Γ used in the derivation of S.

So there is a finite subset Γ′ ⊆ Γ such that Γ′ ⊢ S.

By the soundness theorem, Γ′ ⊨ S

So there is a finite subset Γ′ ⊆ Γ such that Γ′ ⊨ S.

37
More briefly:

Γ ⊨ S ⇔ Γ ⊢ S (Completeness and Soundness)

⇔ Γ′ ⊢ S for some finite Γ′ ⊆ Γ (Derivations are

finite)
⇔ Γ′ ⊨ S for some finite Γ′ ⊆ Γ (Completeness
and Soundness).

This may make it seem as if compactness is just a tech-

nicality, an accident arising from a possibly arbitrary re-
striction on derivations: if infinitely many premises were
allowed the middle equivalence wouldn’t be so automatic.

38
4 Size of models: basic facts

There is a subtle interaction between sets of first order

formulae and the models of those formulae.

Choosing the right first order formulae can force a model

to be a certain way, but there are other features of mod-
els that no first order formulae will secure.

The first question you can ask about is size: how large
or small a model can be.

We measure size according to the size of the domain.

A model “has size n” for some natural number n, “is

finite”, “is denumerable”, “is non-denumerable” etc. ac-
cording as its domain “has size n” for some natural num-
ber n, “is finite”, “is denumerable”, “is non-denumerable”
etc.

39
A particular denumerable model that we’ll be seeing a lot
of is the standard model of formal arithmetic, with the
natural numbers as the domain and the canonical inter-
pretations of 0, +, × and ′.

There are other models of formal arithmetic too, and the

non-standard models of arithmetic have some remarkable
properties.

40
Once we start to work with infinite models, we will lose
some control over how large we can make the model be.

But if we restrict attention for the moment to finite mod-

els, there is more we can do.

Examples:

Here is a sentence that is true in all and only the models

that have at least n members:

∃x1∃x2 . . . ∃xn−1∃xn(x1 ̸= x2 & x1 ̸= x3 & . . . &xn−1 ̸=

xn )

Where the sentence includes a conjunct xi ̸= xj for every

1 ≤ i, j ≤ n, i ̸= j

41
So for example,

∃x1∃x2∃x3(x1 ̸= x2 & x1 ̸= x3 & x2 ̸= x3)

requires that the domain contain at least 3 objects in or-

der to be true.

Note that the inequalities are necessary: If you say ∃x1∃x2

[sentence] there is nothing in the quantifiers alone to en-
sure that the two objects have to be distinct.

42
BJB give a different example, that also works.

∀x1 ∀x2 . . . ∀xn−1 ∃xn (x1 ̸= xn & x2 ̸= xn & . . . &xn−1 ̸= xn )

It may take a moment to see that this actually does what

it is supposed to do. See if you can figure out why

∀x1∀x2∃x3(x1 ̸= x3 & x2 ̸= x3)

can only be true in interpretations with a domain con-

taining at least 3 objects.

43
Let’s look at the case for n = 3.

∀x1∀x2∃x3(x1 ̸= x3 & x2 ̸= x3)

Your first thought might be: Can’t that be satisfied by a

domain with two objects {a, b}? Just let x1 and x2 be a,
and then x3 can be b.

But that won’t work, since we are dealing with the uni-
versal quantifier out front: so if we have two objects a
and b in the model, then we can pick x1 to be a and x2
to be b and our model will have to satisfy:

∃x3(a ̸= x3 & b ̸= x3), which will require there to be a

third object.

44
To get a sentence that is true in all and only the mod-
els that have at most n objects you need only negate a
sentence that is true in models with at least n+1 objects.

For example, negating the BJB sentence gives:

∃x1 ∃x2 . . . ∃xn−1 ∀xn (x1 = xn ∨ x2 = xn ∨ . . . ∨ xn−1 = xn )

which is a sentence true in the models with at most n − 1

objects.

A related simple trick, when you have a sentence that

has n quantified variables x1, x2, . . . xn is to conjoin a
clause (remaining in the scope of the quantifiers bind-
ing x1, x2, . . . xn):

∀xn+1(x1 = xn+1 ∨ x2 = xn+1 ∨ . . . ∨ xn = xn+1).

This will ensure that the objects picked out by the first
variables are the only ones in the domain.

45
I’ll follow BJB and write:

In for a sentence true in all and only models with at least

n objects, and write:

Jn for a sentence true in all and only models with at most

n objects.

A sentence true in exactly the models with n objects is

of course:

Kn = I n & J n

46
The set {I1, I2, . . . , In, . . .} has only infinite models, since
a model of it has at least n objects for arbitrarily large
n.

Note that this set has the feature that although the whole
set is true in only infinite models, any finite subset is true
in at least one finite model (and in fact infinitely many
of them).

(This will be useful for one of the problems on the prob-

lem set.)

47
There are finite sets of sentences that have only infinite
models.

(And, of course, given a finite set S of sentences with just

infinite models, we can form a single sentence with just
infinite models by taking the conjunction of the sentences
in S.)

Here is a simple example. These three sentences taken

together have only infinite models:

(1) ∀x∃yR(x, y)
(2) ∀x∀y ∼(Rxy&Ryx)
(3) ∀x∀y∀z((Rxy&Ryz) → Rxz).

These sentences have at least one familiar model: ⟨N, <⟩,

the natural numbers with “less than”.

48
How do we show that

(1) ∀x∃yR(x, y)
(2) ∀x∀y ∼(Rxy&Ryx)
(3) ∀x∀y∀z((Rxy&Ryz) → Rxz).

has only infinite models?

Say that M is a model of these sentences. To see why

you need infinitely many objects, begin with just one ob-
ject o1 ∈ |M|.

By (1), there has to be an object y such that Ro1y. By 2),

y must be different from x, so there is an o2 ∈ |M|, o1 ̸=
o2 so that M ⊨ Ro1o2.

Again by (1), there must be an o3 ∈ |M|, with M ⊨

Ro2o3. By (2), o2 ̸= o3.
By (3), M ⊨ (Ro1o2&Ro2o3) → Ro1o3, so M ⊨ Ro1o3.

So by (2) again, o1 ̸= o3. So we need at least three ob-

jects in |M|.…

We can keep going this way, to show that there must be

more than n objects in the domain, for every n.

49
4.1 Compactness in Action: Non-Standard Models of Arithmetic

Compactness can be a powerful tool for proving the ex-

istence of models with interesting properties.

I’ll concentrate on one example: Showing that there are

non-standard models of arithmetic.

Next week, we’ll start looking at theories of arithmetic

formulated in the language of arithmetic, which is intro-
duced in BJB example 9.1 p. 103.

The non-logical vocabulary is the constant 0, the sym-

bol for less-than <, the successor function ′, and the two
place function symbols for addition and multiplication:
+, ×

(Recall that the symbols are written in boldface to dis-

tinguish the case where you are talking about the symbol
from the case where you are talking about the object, re-
lation or function.)

50
Alternative notation: Sometimes in the context of the
language of arithmetic a convention is used of writing a
bar over the symbol to indicate you are talking about the
symbol and not what the symbol denotes:

¯ +̄, ×
0̄, ′̄, <, ¯

This has the advantage that it is easier to write on a

blackboard, and it often works better on slides too.

I’ll sometimes use this convention instead of boldface to

distinguish a sign from what it signifies.

51
Next week, we’ll take as our base theory of arithmetic
the quite rudimentary theory called Q, determined by
the axioms set down on BJB p.208.

(Q is the theory of all the sentences that can be derived

from those axioms.)

They say that 0 is not a successor, that successor is one-

one, give recursion equations for plus and times, and a
few other simple things like that.

Q is true in the standard model of arithmetic, whose do-

main is N, and which interprets <, ′, + as the successor
function, the less-than relation, the addition function and
the multiplication function respectively.

Are there other models of Q?

As we’ll see, the answer is: yes, in fact there are lots of
them.

52
Perhaps the easiest way to show there must be non-
standard models of arithmetic is to use compactness.

The argument I’m about to give doesn’t depend on the

fact that you are asking about Q in particular.

It will work for any first-order theory in the language of

arithmetic that is true in the standard model.

Add a new symbol a, and take the theory:

Q ∪ {0 < a, 0′ < a, 0′′ < a, . . .}

In any interpretation making all the sentences in the big-

ger set true, there will be an object a that is larger than
every natural number.

53
Take any finite subset

F ⊆ Q ∪ {0 < a, 0′ < a, 0′′ < a, . . .}.

Given any such F , there will be a greatest n such that

0n < a ∈ F . Say that the greatest such n is k.

F will be true in the standard model of arithmetic aug-

mented with an interpretation of a as k + 1, since every
sentence of Q is true in the standard model, and every
instance of 0n < a ∈ F is true if n < k + 1

54
Since every finite subset of

Q ∪ {0 < a, 0′ < a, 0′′ < a, . . .}

has a model, that means that by compactness, the whole

theory

Q ∪ {0 < a, 0′ < a, 0′′ < a, . . .}

has a model Nnonstandard.

Since Nnonstandard is a model of the larger theory, it is also

a model of Q

But since Nnonstandard contains an element that is greater

than every natural number, it must be a non-standard
model.

55
5 Aside for the Mathematicians in Class: Compactness in Non-
Foundational Mathematics

First-order compactness can be useful for non-foundational

mathematics as well, for relating finite and infinite cases.

It shows up especially in cases where there are system-

atic interconnections between finite versions and infinite
versions of a theory, such as the theory of (finite) fields
in characteristic p and (infinite) fields of characteristic 0.

As one example: A first-order sentence ϕ is true in every

algebraically closed field of characteristic 0 if and only if
there is a prime p > 0 such that ϕ holds in every field of
characteristic p̃ > p.

I’ve posted a paper in the Optional Readings File in

Canvas called “Ramsey’s Theorem and Compactness” by
Marcia Groszek which explores some applications of com-
pactness that relate finite and infinite combinatorics.

56