2022-2023 Inleiding Logica (KI1V13001)

dr. Johannes Korbmacher dr. Colin Caret

Department of Philosophy and Religious Studies

29-09

201/227
Overview

Rehash

7. Soundness and Completeness

7.1 Soundness and Completeness
7.2 The Soundness Theorem
7.3 The Completeness Theorem
7.4 Innite Premiss Sets and Compactness

202/227
Rehash
I Our proof system is the method of tableaux.
¬¬φ φ ∧ ψ ¬(φ ∧ ψ) φ∨ψ ¬(φ ∨ ψ)

φ φ ¬φ ¬ψ φ ψ ¬φ

ψ ¬ψ

¬(φ → ψ) φ→ψ φ↔ψ ¬(φ ↔ ψ)

φ ¬φ ψ φ ¬φ φ ¬φ

¬ψ ψ ¬ψ ¬ψ ψ

I A branch B is closed i p, ¬p ∈ B for some p ∈ P .

I A tableau is closed i every branch is closed.
I Γ ` φ i the complete tableau for Γ ∪ {¬φ} is closed.
I Γ 0 φ i there is at least one open branch.
(
1 if p ∈ B
vB (p) =
0 if p ∈
/B
203/227
7.1 Soundness and Completeness
I (7.1.1) We want to be able to derive the conclusion from the
premises in all and only the valid inferences:

Soundness Theorem. If Γ ` φ, then Γ  φ.

Completeness Theorem. If Γ  φ, then Γ ` φ.
I (7.1.2) Together, the two reduce semantic validity to syntactic
validity.

I (7.1.3) Soundness is a sanity check: only if it follows, we can

infer it.

I (7.1.4) Completeness is a strong property: everything that

follows, you can infer.

204/227
Proof Idea
I (7.1.6) Remember:

Down Preservation. If the formula at the parent node of a rule

is true under a valuation, then at least one
formula on a newly generated child node is true
under the valuation.
Up Preservation. If the formula at a newly generated child
node is true under a valuation, then the formula
at the parent node is true.

I Together, up and down preservation guarantee that when

we're applying the rules, we consider precisely the dierent
ways for a formula to be true.

205/227
7.2 The Soundness Theorem  Proof Idea
I (7.2.1) We're actually going to prove the contrapositive:
I If Γ 2 φ, then Γ 0 φ.
I If Γ 2 φ, then there's a valuation v that makes all the
members of Γ ∪ {¬φ} true.

I That's the initial list.

I By Down Preservation, whenever we apply a rule, we get at

least one branch all of whose members v makes true.

I So, in the complete tableau, there has to be one branch, B,

such that v makes all the formulas on B true.

I But then B cannot be closed. So the tableau is open.

206/227
The Soundness Lemma
I Let B be a branch of a possibly incomplete tableau and v a
valuation:
I (7.2.2) We say that v is faithful to B i JφKv = 1, for all φ ∈ B
p → q, q 0 p
p→q
q
¬p

¬p q

Countermodel: vB (q) = 1, vB (p) = 0.

I (7.2.3) Lemma. Let v be a valuation that is faithful to a
branch B of an incomplete tableau. If a rule is applied to a
formula in the tableau, then v is faithful to at least one branch
B 0 which extends B in the new tableau.

I This is a direct consequence of Down Preservation!

207/227
Example
I v1 with v1 (p) = 1 and v1 (q) = 0
I v2 with v2 (p) = 0 and v2 (q) = 1

p∨q
¬p ∨ ¬q

p q

¬p ¬q ¬p ¬q
7 7

208/227
(7.2.4) Proof of the Soundness Lemma
The proof consists in a one-by-one inspection of the rules.
There are 9 cases. We do: (a) φ→ψ and (b) ¬(φ → ψ).
(a) Let v be faithful to B and φ → ψ ∈ B.
φ→ψ

¬φ ψ
We get two new branches extending B , B1 and B2 .
I We have B1 = B ∪ {¬φ} and B2 = B ∪ {ψ}.
I Since v is faithful to B , we get Jφ → ψKv = 1.
I Since Jφ → ψKv = max(1 − JφKv , JψKv ), we get either
JφKv = 0 or JψKv = 1.
I If JφKv = 0, then J¬φKv = 1 − JφKv = 1. So v is faithful to
B = B ∪ {¬φ}.
1

I If JψKv = 1, then v is faithful to B = B ∪ {ψ}.

2

So, either way, v is faithful to at least one new branch created

by the rule for φ → ψ.
209/227
(7.2.4) Proof of the Soundness Lemma (Cont'd)
(b) Let v be faithful to B and ¬(φ → ψ) ∈ B .
¬(φ → ψ)

¬ψ
We get one new branch B
0 = B ∪ {φ, ¬ψ}.
I Since v is faithful to B , we get J¬(φ → ψ)Kv = 1.
I Since J¬(φ → ψ)Kv = 1 − max(1 − JφKv , JψKv ), we get that
max(1 − JφKv , JψKv ) = 0.
I So JφKv = 1 and JψKv = 0.
I Since J¬ψKv = 1 − JψKv , we can conclude J¬ψKv = 1.
I So v is faithful to B 0 .

210/227
The Soundness Theorem
(7.2.5) Theorem (Soundness). If Γ ` φ, then Γ  φ.
Proof:

I Suppose that Γ 2 φ.
I So, there's a v that makes all the members of Γ ∪ {¬φ} true.

I That's the initial list of our tableau.

I Every time we apply a rule, v remains faithful to at least one

branch.

I So, in the complete tableau, there's a branch B that v is

faithful to.

I Suppose that B is closed, i.e. p, ¬p ∈ B .

I It follows that v (p) = 1 and v (p) = 0. Contradiction.

I So B is open, meaning Γ 0 φ.

211/227
Tableau Verication 1
(7.2.6) Theorem. If the tableau algorithm gives the answer that a
set is unsatisable, then the set is unsatisable.
Proof:

I Suppose that the algorithm says Γ is unsatisable.

I So, the tableau for Γ is closed.

I Suppose, for contradiction, that Γ is satisable via v.

I Using the soundness lemma, we can infer that there's a branch
B in the tableau for Γ which is open (as in soundness).

I Hence the tableau for Γ is open. Contradiction.

I So Γ is unsatisable.

212/227
7.3 The Completeness Theorem  Proof Idea
I (7.3.1) We'll prove once more, the contrapositive:
I If Γ 0 φ, then Γ 2 φ.
I The proof idea is simple:
I If Γ 0 φ, then the complete tableau for Γ ∪ {¬φ} is open via
open branch B .
I We get the associated valuation vB for B :

1 if p ∈ B
(
vB (p) =
0 if p ∈/ B
I Using Up Preservation, we can inductively argue that vB
makes all B 's true.
I But then vB satises Γ ∪ {¬φ}.
I Which gives us Γ 2 φ.
I The devil is in the detail.

213/227
The Completeness Lemma
(7.3.2) Lemma (Completeness Lemma). Let B be an open branch
of a complete tableau. Then vB is faithful to B.
Proof: The proof is a (rather complicated) induction.
I We note that vB is faithful to B i for all φ ∈ L:
1. if φ ∈ B , then JφKvB = 1, and
2. if ¬φ ∈ B , then JφKvB = 0.
I This we prove by induction.
(i) Base case. We need to show that for all p ∈ P :
1. If p ∈ B , then JpKvB = 1
2. If ¬p ∈ B , then JpKvB = 0.
I 1. is immediate from the denition of vB .
I For 2. note that if ¬p ∈ B , then p ∈
/ B since B is open. So, by
denition, vB (p) = 0. Since JpKvB = vB (p), we get our desired
claim.

214/227
The Completeness Lemma (Cont'd)
(ii) Induction steps.
(a) We need to prove that if φ enjoys the property, then ¬φ enjoys
the property. From the induction hypothesis
1. If φ ∈ B , then JφKvB = 1.
2. If ¬φ ∈ B , then JφKvB = 0.
We derive:
1'. If ¬φ ∈ B , then J¬φKvB = 1.
2'. If ¬¬φ ∈ B , then J¬φKvB = 0.
I For 1', suppose that ¬φ ∈ B . By 2., we have JφKvB = 0. But
J¬φKvB = 1 − JφKvB and so J¬φKvB = 1.
I For 2', assume ¬¬φ ∈ B . Since B is an open branch of a
complete tableau, every rule that can be applied has been:
¬¬φ

φ
So, φ ∈ B . But then, by 1., we get JφKvB = 1. And since
J¬φKvB = 1 − JφKvB , we get J¬φKvB = 0, as desired.

215/227
The Completeness Lemma (Cont'd)
(ii) Induction steps.
(b) We do the case for φ ∧ ψ .
We have two pairs of induction hypotheses:
1φ . if φ ∈ B , then JφKvB = 1, and
2φ . if ¬φ ∈ B , then JφKvB = 0.
And
1ψ . if ψ ∈ B , then JψKvB = 1, and
2ψ . if ¬ψ ∈ B , then JψKvB = 0.
What we need to prove are:
1φ∧ψ . if φ ∧ ψ ∈ B , then Jφ ∧ ψKvB = 1, and
2φ∧ψ . if ¬(φ ∧ ψ) ∈ B , then Jφ ∧ ψKvB = 0.

216/227
The Completeness Lemma (Cont'd)
I Suppose that φ ∧ ψ ∈ B. Since B is an open branch of a
complete tableau, every rule that can be applied has been
applied:

φ∧ψ

ψ
I So, we can conclude that both φ, ψ ∈ B .
I But by 1φ . and 1ψ ., this means that JφKvB = 1 and JψKvB = 1.
I Since Jφ ∧ ψKvB = min(JφKvB , JψKvB ), we get Jφ ∧ ψKvB = 1,
as desired.

217/227
The Completeness Lemma (Cont'd)
I Next, suppose that ¬(φ ∧ ψ) ∈ B . Again, all rules that can be
applied have been applied:

¬(φ ∧ ψ)

¬φ ¬ψ
I So, we can conclude that either (∗) ¬φ ∈ B or (∗∗) ¬ψ ∈ B .
(∗) I If ¬φ ∈ B , from 2φ ., we get JφK = 0.
I Since Jφ ∧ ψKvB = min(JφKvB , JψKvB ), this means we get
Jφ ∧ ψKvB = 0.
(∗∗) I If ¬ψ ∈ B , from 2ψ ., we get JφK = 0.
I So, we get Jφ ∧ ψKvB = 0.
I So either way, Jφ ∧ ψKvB = 0, as desired.

218/227
The Completeness Theorem
(7.3.3) Theorem (Completeness). If Γ  φ, then Γ ` φ.
Proof:

I Assume that Γ 0 φ.
I By denition, this means that there's an open branch, B, in
the complete tableau for Γ ∪ {¬φ}.
I Consider the associated valuation vB .
I By the completeness lemma vB makes all the members of B
true.

I Since Γ ∪ {¬φ} ⊆ B , it follows that vB satises Γ ∪ {¬φ}.

I So Γ 2 φ, as desired.

219/227
Tableau Verication 2
(7.3.4) Theorem. If the tableau algorithm gives the answer that a
set is satisable, then the set is satisable.
Proof:

I Suppose that the algorithm says Γ is satisable.

I So, the tableau for Γ is open.

I We can infer that there's a branch B in the tableau for Γ

which is open.

I By the completeness lemma vB makes all the members of B

true..

I Since Γ ⊆ B, we can infer that vB satises Γ as desired.

(7.3.5) Theorem (Decidability). Propositional logic is decidable,

i.e. there exists an algorithm which after nitely many steps
correctly determines whether a given inference (with nitely any
premises) is valid.

220/227
7.4 Innite Premiss Sets and Compactness
I (7.4.1) For the formal treatment of mathematics, we
sometimes need innite premise sets:
I The theory of natural numbers cannot be reduced to nitely
many axioms.
I (7.4.2) We'll go innite, but we assume there is a way of
writing Γ as the set {φi : i ∈ I }, where I ⊆ N+ .
I (7.4.3) Examples:
(a) {p, ¬p, ¬¬p, . . .} (or more precisely the smallest set X such
that p ∈ X and if φ ∈ X then ¬φ ∈ X ).
(b) P = {pi : i ∈ N}
(c) {¬p2i , p2i+1 : i ∈ N} which contains ¬pi for each even i and pi
for each odd i ∈ N.
(d) Let v be any valuation. Then the set Tv = {φ : JφKv = 1} is
always innite!
I (7.4.4) Γφ works just like before.

221/227
Truth-Tables Don't Work

I (7.4.5) But truth-tables no longer work:

I With nite Γ, we can use:

φ1 , . . . , φn  ψ i  (φ1 ∧ . . . ∧ φn ) → ψ

I But with innite Γ, there is no theorem:

φ1 , φ2 , . . .  ψ i  (φ1 ∧ φ2 ∧ . . .) → ψ
| {z }
???

222/227
(7.4.6) Innite Tableaux
I Tableaux, however, work.

I To make the tableau for {φ1 , φ2 , . . . , ¬ψ} write down ¬ψ as

the initial list and do its tableau.

I Then, add φ1 to the initial list, and extend the tableau for ¬ψ
to the tableau for {φ1 , ¬φ}.
I Then add φ2 , and extend the tableau to the one for
{φ1 , φ2 , ¬φ}.
I And so on.

I Γ`φ i the complete tableau is closed, Γ0φ i the complete

(?) tableau is open.

I Note that a complete tableau (which is open) can be innite!

223/227
Example
p, ¬¬p, ¬¬¬¬p, . . . 0 q
.
.
.
¬¬¬¬p
¬¬p
p
¬q

¬¬p

.
.
.

224/227
Compactness
But if a tableau closes, it closes after nitely many steps (at some
point):
Theorem (Compactness). Let Γ be an innite set of formulas. If
Γ ` φ, then there exists a nite set Σ⊆Γ such that Σ ` φ.
Proof:

I Let Γ = {ψi : i ∈ I } for some I ⊆ N+ .

I Assume that Γ ` φ.
I This means that the tableau closes at some point.

I So there is a smallest i such that ψj is in the tableau for all

j ≤ i.
I We've actually made the tableau for {¬φ} ∪ {ψj : j ≤ i}.
I So {ψj : j ≤ i} ` φ.
I But clearly {ψj : j ≤ i} is nite and {ψj : j ≤ i} ⊆ Γ.

225/227
Core Ideas (Lecture Version)
I Soundness and completeness reduce validity to syntax.

I The soundness theorem is a sanity check.

I The completeness theorem is a surprising mathematical fact.

I For tableaux, soundness relies on the fact that in every rule, if

the upper formula is true, then at least one of the lower
formulas is true (`down preservation').

I For tableaux, completeness relies on the fact that if one of the

lower formulas in a rule is true, so is the upper formula (`up
preservation').

I Truth tables don't work for innitary premise sets but tableaux
do.

I Compactness tells us that if there's a proof, we can always nd

a nitary one.

226/227
Thanks!

227/227