You are on page 1of 27

Contents

1 Completeness: The Proof up to Now 2

2 The Completeness proof, Completed 5

3 The case of the Regularity Lemma 14

4 Review: Building a Maximal Consistent in SD Set 19


4.1 What is With the Extra Stage in Getting the List Matching
Numbers and Sentences? . . . . . . . . . . . . . . . . . . . . . 19
4.2 Why do we Need to Assign Every Sentence a Unique Number? 23

5 How to Use the Completeness and Soundness Theorems 26

1
1. Completeness: The Proof up to Now

Our objective is to prove that any for any sentence S of


SL and set of sentences of SL:

SS

Last week we showed that we can prove this if we prove


the following result:

If is consistent in SD then is truth-functionally con-


sistent.

Our strategy to prove that is to take the set and make


it bigger, into a set that is still consistent in SD, but
also so big that if we add just one more sentence, the
result will no longer be consistent in SD.

That is, the set is maximal consistent in SD.

2
How does making bigger like this help us?

To see why, look at the the way that truth-assignments


for complex sentences relate to their parts, and compare
that to the way that the membership of a complex sen-
tence in relates to the membership of its parts in :

a) P is assigned T P is assigned F
b) P&Q is assigned T P is assigned T and Q is assigned T
c) P Q is assigned T P is assigned T or Q is assigned T
d) P Q is assigned T P is assigned F or Q is assigned T
e) P Q is assigned T (P is assigned T and Q is assigned
T ) or (P is assigned F and Q is assigned F)

a) P P
b) P&Q P and Q
c) P Q P or Q
d) P Q P or Q
e) P Q (P and Q ) or (P and
Q )

They are exactly the same, except for the interchange of is


assigned T/is assigned F with / .

3
The main goal of the Tuesday lecture was to prove that
acts as indicated on the map displayed on the pre-
vious slide.

(I called this The Regularity Lemma).

Thanks to this structure, we can dene a truth-assignment


that makes all the sentences in true, if we just stipu-
late that every sentence letter in is assigned T.

The nal installment of the proof of the completeness


theorem is to prove that this truth assignment works.

That is, we need to prove that if you stipulate that every


sentence letter in is T, then every sentence in is
assigned T.

4
2. The Completeness proof, Completed

We are given a maximally consistent . We stipulate


that every sentence letter in is assigned T. (And ev-
ery other sentence letter is assigned F).

Call this truth assignment A.

We want to prove that A makes every sentence in


true.

The proof is by induction: We use the fact that the the-


orem is true for simpler sentences to prove that it is true
for more complex ones.

Base case Every atomic sentence S of SL is true on


A S .

We know the base case is true just by stipulation: That is


how we dened the truth-assignment A on the sentence
letters.

5
Induction Step:

Induction Hypothesis: For every sentence P of SL with


k or fewer connectives, P is true on A P

Must prove, given the induction hypothesis: For every


sentence P of SL with k + 1 connectives, P is true on
A P .

The induction step has several cases, depending on what


P looks like.

6
Case 1: P = Q for some Q

If Q is true on A then Q is false on A because A


is a truth-assignment, and that is how they roll.

Because Q has fewer than k + 1 connectives, the induc-


tion hypothesis applies, and hence:

Q is false on A Q .

So Q

We know from the Regularity Lemma that:


Q Q

So: Q .

7
If Q is false on A then Q is true on A.

Because Q has fewer than k + 1 connectives, the induc-


tion hypothesis applies, and so:

Q is true on A Q .

Hence: Q .

We know by the Regularity Lemma that:

Q Q

Since weve shown Q , this clause of the Regularity


Lemma allows us to conclude that Q .

8
Case 2: P = Q & R for some Q and R

If Q & R is true on A then Q is true on A and R


is true on A by the rules for truth-assignments.

Since both Q and R have fewer than k + 1 connectives,


the induction hypothesis applies, and so:

Q is true on A Q
and R is true on A R

This means we can conclude: Q and R .

The Regularity Lemma tells us:


P&Q P and Q

So it follows that Q & R .

9
If Q & R is false on A then either Q is false on A
or R is false on A by the rules for truth-assignments.

Since both Q and R have fewer than k + 1 connectives,


the induction hypothesis applies, and so:

Q is true on A Q and
R is true on A R

So either Q or R .

The Regularity Lemma tells us:


P&Q P and Q

So we can conclude: Q & R , which is what we


needed to show.

10
Case 4: P = Q R for some Q and R.

If Q R is true on A then Q is false on A or R is


true on A.

Since both Q and R have fewer than k + 1 connectives,


the induction hypothesis applies, and so Q or
R .

By the Regularity Lemma, P Q P or


Q , so:

Q R .

11
If Q R is false on A then Q is true on A and R
is false on A.

Since both Q and R have fewer than k + 1 connectives,


the induction hypothesis applies, and so Q and
R .

By the Regularity Lemma, P Q P or


Q , so:

Q R .

12
Cases 3 and 5 are questions on the problem set.

This completes the proof that if is maximal consis-


tent, then the truth assignment making every sentence
letter in true and every sentence letter not in false
makes every sentence in true.

Since we have that is truth-functionally con-


sistent, because there is a truth assignment making every
sentence in true.

That is, we have proven that if is consistent in SD, then


is truth-functionally consistent.

As we proved at the outset of last weeks rst lecture, it


follows from this that if  P then P

13
3. The case of the Regularity Lemma

I received a couple of requests to say a bit more about


the proof of the Regularity Lemma, so Ill go over the
case that I skipped over Tuesdays lecture.

We want to show:

P Q P or Q

This is a biconditional, so we need to prove both the


and the directions.

14
Say that P Q

Either P or P .

If P , we are done, so assume P

If P , then along with the fact that P Q


we can produce a derivation of Q using premises from :

1 PQ Assumption
2 P Assumption
3 Q 1,2 E

Hence Q, and so by Jove, Q .

Hence, either P or Q

15
Say that P or Q

Consider each of the two possibilities in turn:

i) P

Since is maximal consistent in SD, this means that for


some sentence R of SD, {P} R and {P} R

Let {A1; A2; . . . ; An} be all the premises from used


to derive R and R from {P}

The following (minus line numbers and some steps) is a


derivation of P Q from {A1; A2; . . . ; An} with deriva-
tions of R and R indicated by vertical dots. (Next
slide)

16
A1 Assumption
A2 Assumption
Assumption
Assumption
An Assumption
P A/ I
Q A/ I
..
.
R
R
Q E
PQ I

So P Q and so by Jove, P Q .

17
The other possibility is:

ii) Q

In that case, we can derive P Q quite easily from :

1 Q Assumption
2 P A/ I
3 Q 1R
4 PQ I

So P Q and so by Jove, P Q .

So whichever of P or Q holds, we have that


P Q .

This proves the second direction, and so proves the bi-


conditional.

18
4. Review: Building a Maximal Consistent in SD Set
4.1. What is With the Extra Stage in Getting the List Matching
Numbers and Sentences?

(This is relevant to the rst question on problem set 7a.)

The textbook ( p. 244 - 255) presents a procedure for


assigning numbers to sentences of SL so that every num-
ber is assigned a unique sentence, and every sentence is
assigned a unique number.

(I discuss this procedure in lecture 17, slides 9 - 15.)


That is, we can specify an (innite) list:
1 S1
2 S2
3 S3
4 S4
5 S5
.... ..
.. ....
n Sn
.... ..
.. ....

With every natural number occurring exactly once on


the right-hand side, and every sentence of SL occurring
at exactly one place as one of the sentences Si.

19
The assignment takes place in three steps: First you as-
sign a (two digit) number to every symbol in the vo-
cabulary of SL (where you count the individual digits of
subscripts as separate symbols):

f () :
10
11
12 &
13
14
15 (
16 )
20 0
21 1
.... ..
.. ....
29 9
30 A
31 B
.... ..
.. ....
55 Z

20
Then we associate each sentence with the string of num-
bers corresponding to its symbols (including the digits of
subscripts as symbols):

If S is the string 1234 . . . n then the first number


associated with S is:

f (1)f (2)f (3)f (4) . . . f (n)

For example, the sentence A11 (B2&C) is associated


with the string:

f (A)f (1)f (1)f ()f (()f (B)f (2)f (&)f (C)f ())

Which is 30212111153122123216

Each string is assigned a unique number, and in fact if


you are given the number of a string, there is a simple
mechanical procedure to nd out what the string is.

21
The reason that I put first number in boldface is that
this assignment of numbers to sentences doesnt nish the
job we need to do.

It assigns a unique number to every sentence of SL, but


there are numbers that dont get assigned any sentence
of SL.

We need the list to be complete: Every number must


be assigned a unique sentence.

That is why we need to add one more step:

The numbers assigned to sentences in the rst order can


be arranged in a list according to size, with the smallest
number rst, the next smallest second, the next smallest
3rd, etc.

Then list each sentence according to its place in this list.


The sentence with the smallest number gets 1, the sen-
tence with the second smallest gets 2, etc.

That is the final, official list.

22
4.2. Why do we Need to Assign Every Sentence a Unique Num-
ber?

Say you start out with a set that is consistent in SD.

In the textbook (p. 255 - 57) and in lecture 18 p. 24 - 31


we see how to extend this to a maximal consistent in SD
set , with .

Remember the procedure: Dene an innite sequence of


sets in stages, then take the union of all of them:

Stage 0: Let 0 =
{
n1 {Sn } if n1 {Sn } is consistent in SD
Stage n: Let n =
n1 if n1 {Sn } is not consistent in SD

Let be the union of all the ns.

In other words, is the set consisting of the original


set , plus every sentence that was added at one of the
stages specied above.

23
The reason that we need to have the list of sentences is
that we use it to dene the stages. Each stage considers
exactly one sentence, indexed by the number assigned to
the sentence.

Why do we need to do things this way? Why not just


put in all the extra sentences at once?

The reason is that there are going to be several dier-


ent maximal consistent in SD sets extending , and they
wont be compatible.

For example, say we start with = {A, A B, C&D}.


This is consistent in SD. We can make it bigger by one
sentence by adding E, or by adding E, and in each
case, we will still have a set that is consistent in SD:

{A, A B, C&D, E} and {A, A B, C&D, E} are


both consistent in SD

But of course, {A, A B, C&D, E, E} is not consis-


tent in SD. You can add either of E or E, but not both.

24
This also means that adding one of these sentences will
force you to exclude the other.

That is why we have to build the set in ordered stages,


and consider the sentences one at a time, in a xed order.

That is why we need to have all the sentences listed, in-


dexed by the natural numbers.

25
5. How to Use the Completeness and Soundness The-
orems

The completeness and soundness theorems are interest-


ing in their own right, but also they provide useful tools
for proving things about derivability in SD and truth-
functional entailment.

Some facts are easier to prove if we consider them in


terms of derivations, others if we consider them in terms
of truth-functional entailment.

Questions 2 and 3 on problem set 7a are illustrations of


this.

It is a challenge to show that there is no derivation of a


contradiction from {A}, but the problem is much more
straightforward if you reframe it in terms of truth-functional
entailment.

The soundness theorem allows you to do that.

26
Proving the compactness theorem for truth-functional en-
tailment is conceptually challenging, but proving it for
derivations in SD is simple.

So reformulate the compactness theorem as a statement


about derivations in SD using the completeness theorem.

Prove the corresponding fact about derivations in SD: If


S then there is a nite such that S.

Then use the soundness theorem to translate this fact


back into a statement about truth-functional entailment.

The whole thing should take at most 5 or 6 lines.

27