You are on page 1of 37

Many-Sorted First-Order Model Theory

Lecture 9

2nd July, 2020

1 / 37
Quantifier elimination example
Dense linear orders w/o endpoints

2 / 37
Eliminating one ∃
Lemma 1
Let T be the theory of dense linear orders without endpoints. Every
formula ∃x · ϕ1 ∧ · · · ∧ ϕn , where each ϕi is atomic or negated atomic, is
equivalent over T to a (positive) quantifier free formula.

Proof sketch.
I Wlog, assume that each ϕi contains an occurrence of x.
I Next, observe that T |= x 6= y ↔ x < y ∨ y < x, and T |= x 6< y ↔ x = y ∨ y < x.
I Using this and distributivity of ∃ over disjunction we can assume that each ϕi is atomic.
I If some ϕi is x = y , then we can replace all occurrences of x in ϕ1 ∧ · · · ∧ ϕn by y , and
delete y = y , unless it is a single conjunct. Note that this can leave some conjuncts of the
form z = y . They can harmlessly be brought out of the scope of ∃.
I Now we can assume all ϕi are of the form < x or x < .
I Group them into j (vj < x) and k (x < vj ), to get ∃x · j (vj < x) ∧ k (x < vk ).
V V V V

I We have T |= ∃x · j (vj < x) ∧ k (x < vk ) ↔ j,k (vj < vk ). To deal with special cases
V V V

we assume (as usual) that an empty conjunction is equivalent to > (or to x = x).
I We have shown more than we claimed: ∃x · ϕ1 ∧ · · · ∧ ϕn is equivalent over T to a
positive quantifier free formula ψ, such that we have either T |= ψ or T |= ¬ψ, and it is
decidable which. (Because ψ is quantifier free, we have T |= ψ iff T |= ∀x · ψ.)
3 / 37
Eliminating one ∃
Lemma 1
Let T be the theory of dense linear orders without endpoints. Every
formula ∃x · ϕ1 ∧ · · · ∧ ϕn , where each ϕi is atomic or negated atomic, is
equivalent over T to a (positive) quantifier free formula.

Proof sketch.
I Wlog, assume that each ϕi contains an occurrence of x.
I Next, observe that T |= x 6= y ↔ x < y ∨ y < x, and T |= x 6< y ↔ x = y ∨ y < x.
I Using this and distributivity of ∃ over disjunction we can assume that each ϕi is atomic.
I If some ϕi is x = y , then we can replace all occurrences of x in ϕ1 ∧ · · · ∧ ϕn by y , and
delete y = y , unless it is a single conjunct. Note that this can leave some conjuncts of the
form z = y . They can harmlessly be brought out of the scope of ∃.
I Now we can assume all ϕi are of the form < x or x < .
I Group them into j (vj < x) and k (x < vj ), to get ∃x · j (vj < x) ∧ k (x < vk ).
V V V V

I We have T |= ∃x · j (vj < x) ∧ k (x < vk ) ↔ j,k (vj < vk ). To deal with special cases
V V V

we assume (as usual) that an empty conjunction is equivalent to > (or to x = x).
I We have shown more than we claimed: ∃x · ϕ1 ∧ · · · ∧ ϕn is equivalent over T to a
positive quantifier free formula ψ, such that we have either T |= ψ or T |= ¬ψ, and it is
decidable which. (Because ψ is quantifier free, we have T |= ψ iff T |= ∀x · ψ.)
4 / 37
Eliminating all quantifiers
Theorem 2
Let T be the theory of dense linear orders without endpoints. Then for
every formula ϕ there exist a quantifier free formula ϕ∗ such that
I T |= ϕ ↔ ϕ∗ ,
I ϕ∗ is effectively obtainable from ϕ.

Proof.
I We can assume ϕ is in prenex form, so ϕ = Q1 . . . Qn ψ where ψ is quantifier free.
I Consider the innermost quantifier Qn ; if Qn is ∀, replace it by ¬∃¬. Bring ¬ψ into
disjunctive form, and distribute ∃ over disjunctions.
I Thus, we can assume Qn ψ is (¬)(∃x · ρ1 ∨ · · · ∨ ∃x · ρm ), where each ρi is a conjunction of
atomic and/or negated atomic formulas.
I By Lemma 1 each ∃x · ρi is equivalent over T to a quantifier free σi . Replace them.
I Repeat the procedure n times, to remove all quantifiers and obtain ϕ∗ . The procedure is
clearly effective.

5 / 37
Eliminating all quantifiers
Theorem 2
Let T be the theory of dense linear orders without endpoints. Then for
every formula ϕ there exist a quantifier free formula ϕ∗ such that
I T |= ϕ ↔ ϕ∗ ,
I ϕ∗ is effectively obtainable from ϕ.

Proof.
I We can assume ϕ is in prenex form, so ϕ = Q1 . . . Qn ψ where ψ is quantifier free.
I Consider the innermost quantifier Qn ; if Qn is ∀, replace it by ¬∃¬. Bring ¬ψ into
disjunctive form, and distribute ∃ over disjunctions.
I Thus, we can assume Qn ψ is (¬)(∃x · ρ1 ∨ · · · ∨ ∃x · ρm ), where each ρi is a conjunction of
atomic and/or negated atomic formulas.
I By Lemma 1 each ∃x · ρi is equivalent over T to a quantifier free σi . Replace them.
I Repeat the procedure n times, to remove all quantifiers and obtain ϕ∗ . The procedure is
clearly effective.

6 / 37
Decidabillity
Corollary 3
The theory T of dense linear orders without endpoints is decidable.

Proof.
I Let ϕ be a formula in the signature of T . By Lemma 1 and Theorem 2, ϕ is equivalent
over T to a Boolean combination of formulas ψj (1 ≤ j ≤ n) such that for each j we have
T |= ψj or T |= ¬ψj and it is decidable which.
I Hence the problem T |= ϕ reduces to propositional tautology checking.

Lemma 4
Every formula ϕ is equivalent over T to a disjunction of conjunctions of
atomic formulas. Thus, ϕ(x) is equivalent over T to the statement
x1 ≤ x2 ≤ · · · ≤ xn , for some suitable renumbering of x, where each ≤ is
either < or =.

Proof.
By analysing the proofs of Lemma 1 and Theorem 2.
7 / 37
Decidabillity
Corollary 3
The theory T of dense linear orders without endpoints is decidable.

Proof.
I Let ϕ be a formula in the signature of T . By Lemma 1 and Theorem 2, ϕ is equivalent
over T to a Boolean combination of formulas ψj (1 ≤ j ≤ n) such that for each j we have
T |= ψj or T |= ¬ψj and it is decidable which.
I Hence the problem T |= ϕ reduces to propositional tautology checking.

Lemma 4
Every formula ϕ is equivalent over T to a disjunction of conjunctions of
atomic formulas. Thus, ϕ(x) is equivalent over T to the statement
x1 ≤ x2 ≤ · · · ≤ xn , for some suitable renumbering of x, where each ≤ is
either < or =.

Proof.
By analysing the proofs of Lemma 1 and Theorem 2.
8 / 37
Decidabillity
Corollary 3
The theory T of dense linear orders without endpoints is decidable.

Proof.
I Let ϕ be a formula in the signature of T . By Lemma 1 and Theorem 2, ϕ is equivalent
over T to a Boolean combination of formulas ψj (1 ≤ j ≤ n) such that for each j we have
T |= ψj or T |= ¬ψj and it is decidable which.
I Hence the problem T |= ϕ reduces to propositional tautology checking.

Lemma 4
Every formula ϕ is equivalent over T to a disjunction of conjunctions of
atomic formulas. Thus, ϕ(x) is equivalent over T to the statement
x1 ≤ x2 ≤ · · · ≤ xn , for some suitable renumbering of x, where each ≤ is
either < or =.

Proof.
By analysing the proofs of Lemma 1 and Theorem 2.
9 / 37
Decidabillity
Corollary 3
The theory T of dense linear orders without endpoints is decidable.

Proof.
I Let ϕ be a formula in the signature of T . By Lemma 1 and Theorem 2, ϕ is equivalent
over T to a Boolean combination of formulas ψj (1 ≤ j ≤ n) such that for each j we have
T |= ψj or T |= ¬ψj and it is decidable which.
I Hence the problem T |= ϕ reduces to propositional tautology checking.

Lemma 4
Every formula ϕ is equivalent over T to a disjunction of conjunctions of
atomic formulas. Thus, ϕ(x) is equivalent over T to the statement
x1 ≤ x2 ≤ · · · ≤ xn , for some suitable renumbering of x, where each ≤ is
either < or =.

Proof.
By analysing the proofs of Lemma 1 and Theorem 2.
10 / 37
Completeness
Theorem 5
e
Let A and B be models of T . If A ,→ B, then A ,→ B.

Proof sketch.
I We use Tarski-Vaught criterion. Let ϕ(x, y ) be a formula and let a be a tuple from A.
Wlog, A ≤ B. Assume B |= ∃x · ϕ(x, a). Thus, B |= ϕ(b, a), for some b from B.
I By Lemma 4, this is equivalent to a1 ≤ · · · ≤ ai ≤ b ≤ ai+1 ≤ . . . an , with a renumbered
suitably.
I If b is already in A there is nothing to do, so assume ai < b < ai+1 and b ∈ / A. By density
of A, there is a c ∈ A with ai < c < ai+1 . Therefore, a1 ≤ · · · ≤ ai ≤ c ≤ ai+1 ≤ . . . an
holds in A.
I By Lemma 4, we obtain A |= ϕ(c, a). Hence, A |= ∃x · ϕ(x, a) as required.

Corollary 6
All models of T are elementarily equivalent. Hence, T is complete.

Proof.
Let A and B be dense linear orders without endpoints. We have A ,→ B iff
e e
card(A) ≤ card(B). By Theorem 5, A ,→ B or B ,→ A holds. In either case A ≡ B. 11 / 37
Completeness
Theorem 5
e
Let A and B be models of T . If A ,→ B, then A ,→ B.

Proof sketch.
I We use Tarski-Vaught criterion. Let ϕ(x, y ) be a formula and let a be a tuple from A.
Wlog, A ≤ B. Assume B |= ∃x · ϕ(x, a). Thus, B |= ϕ(b, a), for some b from B.
I By Lemma 4, this is equivalent to a1 ≤ · · · ≤ ai ≤ b ≤ ai+1 ≤ . . . an , with a renumbered
suitably.
I If b is already in A there is nothing to do, so assume ai < b < ai+1 and b ∈ / A. By density
of A, there is a c ∈ A with ai < c < ai+1 . Therefore, a1 ≤ · · · ≤ ai ≤ c ≤ ai+1 ≤ . . . an
holds in A.
I By Lemma 4, we obtain A |= ϕ(c, a). Hence, A |= ∃x · ϕ(x, a) as required.

Corollary 6
All models of T are elementarily equivalent. Hence, T is complete.

Proof.
Let A and B be dense linear orders without endpoints. We have A ,→ B iff
e e
card(A) ≤ card(B). By Theorem 5, A ,→ B or B ,→ A holds. In either case A ≡ B. 12 / 37
Completeness
Theorem 5
e
Let A and B be models of T . If A ,→ B, then A ,→ B.

Proof sketch.
I We use Tarski-Vaught criterion. Let ϕ(x, y ) be a formula and let a be a tuple from A.
Wlog, A ≤ B. Assume B |= ∃x · ϕ(x, a). Thus, B |= ϕ(b, a), for some b from B.
I By Lemma 4, this is equivalent to a1 ≤ · · · ≤ ai ≤ b ≤ ai+1 ≤ . . . an , with a renumbered
suitably.
I If b is already in A there is nothing to do, so assume ai < b < ai+1 and b ∈ / A. By density
of A, there is a c ∈ A with ai < c < ai+1 . Therefore, a1 ≤ · · · ≤ ai ≤ c ≤ ai+1 ≤ . . . an
holds in A.
I By Lemma 4, we obtain A |= ϕ(c, a). Hence, A |= ∃x · ϕ(x, a) as required.

Corollary 6
All models of T are elementarily equivalent. Hence, T is complete.

Proof.
Let A and B be dense linear orders without endpoints. We have A ,→ B iff
e e
card(A) ≤ card(B). By Theorem 5, A ,→ B or B ,→ A holds. In either case A ≡ B. 13 / 37
Completeness
Theorem 5
e
Let A and B be models of T . If A ,→ B, then A ,→ B.

Proof sketch.
I We use Tarski-Vaught criterion. Let ϕ(x, y ) be a formula and let a be a tuple from A.
Wlog, A ≤ B. Assume B |= ∃x · ϕ(x, a). Thus, B |= ϕ(b, a), for some b from B.
I By Lemma 4, this is equivalent to a1 ≤ · · · ≤ ai ≤ b ≤ ai+1 ≤ . . . an , with a renumbered
suitably.
I If b is already in A there is nothing to do, so assume ai < b < ai+1 and b ∈ / A. By density
of A, there is a c ∈ A with ai < c < ai+1 . Therefore, a1 ≤ · · · ≤ ai ≤ c ≤ ai+1 ≤ . . . an
holds in A.
I By Lemma 4, we obtain A |= ϕ(c, a). Hence, A |= ∃x · ϕ(x, a) as required.

Corollary 6
All models of T are elementarily equivalent. Hence, T is complete.

Proof.
Let A and B be dense linear orders without endpoints. We have A ,→ B iff
e e
card(A) ≤ card(B). By Theorem 5, A ,→ B or B ,→ A holds. In either case A ≡ B. 14 / 37
Quantifier elimination in general

15 / 37
Quantifier elimination
Definition 7
Let Σ be a signature an K a class of Σ-structures. A set Φ of formulas is
called an elimination set for K, if for every formula ϕ(x) of Σ there is a
formula ϕ∗ (x) such that
I ϕ∗ (x) is a Boolean combination of formulas in Φ,
I For every A ∈ K and every tuple a from A we have A |= ϕ(a) iff
A |= ϕ∗ (a).
If there is an elimination set Φ for K, such that all formulas in Φ are
quantifier free, then we say that K admits quantifier elimination. If
K = Mod(T ) for some theory T , we also say that T admits quantifier
elimination.

Lemma 8
If every formula ∃x · α1 ∧ · · · ∧ αn , where each αi is atomic or negated
atomic, is equivalent over T to a quantifier free formula, then T admits
quantifier elimination.
16 / 37
Quantifier elimination
Definition 7
Let Σ be a signature an K a class of Σ-structures. A set Φ of formulas is
called an elimination set for K, if for every formula ϕ(x) of Σ there is a
formula ϕ∗ (x) such that
I ϕ∗ (x) is a Boolean combination of formulas in Φ,
I For every A ∈ K and every tuple a from A we have A |= ϕ(a) iff
A |= ϕ∗ (a).
If there is an elimination set Φ for K, such that all formulas in Φ are
quantifier free, then we say that K admits quantifier elimination. If
K = Mod(T ) for some theory T , we also say that T admits quantifier
elimination.

Lemma 8
If every formula ∃x · α1 ∧ · · · ∧ αn , where each αi is atomic or negated
atomic, is equivalent over T to a quantifier free formula, then T admits
quantifier elimination.
17 / 37
Dense linear orders again
Exercise 1
Let K be the class of dense linear orders. Define Φ to be
I ∃x · ∀y · x = y ∨ x < y
I ∃x · ∀y · x = y ∨ y < x
I ∀y · x = y ∨ x < y
I ∀y · x = y ∨ y < x
I x <y

Prove that Φ is an elimination set for K.

Exercise 2
Compare K with dense linear orders without endpoints. Let A, B ∈ K.
e
I Does A ,→ B imply A ,→ B?
I Are A and B elementarily equivalent?

Exercise 3
Repeat Exercise 2 for K0 – the class of dense linear orders with endpoints.
18 / 37
Dense linear orders again
Exercise 1
Let K be the class of dense linear orders. Define Φ to be
I ∃x · ∀y · x = y ∨ x < y
I ∃x · ∀y · x = y ∨ y < x
I ∀y · x = y ∨ x < y
I ∀y · x = y ∨ y < x
I x <y

Prove that Φ is an elimination set for K.

Exercise 2
Compare K with dense linear orders without endpoints. Let A, B ∈ K.
e
I Does A ,→ B imply A ,→ B?
I Are A and B elementarily equivalent?

Exercise 3
Repeat Exercise 2 for K0 – the class of dense linear orders with endpoints.
19 / 37
Dense linear orders again
Exercise 1
Let K be the class of dense linear orders. Define Φ to be
I ∃x · ∀y · x = y ∨ x < y
I ∃x · ∀y · x = y ∨ y < x
I ∀y · x = y ∨ x < y
I ∀y · x = y ∨ y < x
I x <y

Prove that Φ is an elimination set for K.

Exercise 2
Compare K with dense linear orders without endpoints. Let A, B ∈ K.
e
I Does A ,→ B imply A ,→ B?
I Are A and B elementarily equivalent?

Exercise 3
Repeat Exercise 2 for K0 – the class of dense linear orders with endpoints.
20 / 37
Model completeness
Definition 9
A theory T is model complete if every embedding between models of T is
elementary. That is, for every A, B ∈ Mod(T ) we have that A ,→ B
e
implies A ,→ B.

Theorem 10
If a theory T admits quantifier elimination, then T is model complete.

Proof.
I Let A, B be models of T such that (wlog) A ≤ B. Let ϕ(x) be a formula, and let a be a
tuple from A.
I By quantifier elimination, we have
(?) T |= ϕ(x) ↔ ϕ∗ (x, y ), for some quantifier-free ϕ∗ .
I Then, ϕ∗ is both Π1 and Σ1 , so it is preserved by substructures and extensions (by
Loś-Tarski), so we have A |= ϕ∗ (a, y ) if and only if B |= ϕ∗ (a, y ).
I Combining this with (?), we get A |= ϕ(a) if and only if B |= ϕ(a), as required.

21 / 37
Model completeness
Definition 9
A theory T is model complete if every embedding between models of T is
elementary. That is, for every A, B ∈ Mod(T ) we have that A ,→ B
e
implies A ,→ B.

Theorem 10
If a theory T admits quantifier elimination, then T is model complete.

Proof.
I Let A, B be models of T such that (wlog) A ≤ B. Let ϕ(x) be a formula, and let a be a
tuple from A.
I By quantifier elimination, we have
(?) T |= ϕ(x) ↔ ϕ∗ (x, y ), for some quantifier-free ϕ∗ .
I Then, ϕ∗ is both Π1 and Σ1 , so it is preserved by substructures and extensions (by
Loś-Tarski), so we have A |= ϕ∗ (a, y ) if and only if B |= ϕ∗ (a, y ).
I Combining this with (?), we get A |= ϕ(a) if and only if B |= ϕ(a), as required.

22 / 37
Model completeness
Definition 9
A theory T is model complete if every embedding between models of T is
elementary. That is, for every A, B ∈ Mod(T ) we have that A ,→ B
e
implies A ,→ B.

Theorem 10
If a theory T admits quantifier elimination, then T is model complete.

Proof.
I Let A, B be models of T such that (wlog) A ≤ B. Let ϕ(x) be a formula, and let a be a
tuple from A.
I By quantifier elimination, we have
(?) T |= ϕ(x) ↔ ϕ∗ (x, y ), for some quantifier-free ϕ∗ .
I Then, ϕ∗ is both Π1 and Σ1 , so it is preserved by substructures and extensions (by
Loś-Tarski), so we have A |= ϕ∗ (a, y ) if and only if B |= ϕ∗ (a, y ).
I Combining this with (?), we get A |= ϕ(a) if and only if B |= ϕ(a), as required.

23 / 37
Some consequences of model completeness
Definition 11
A model A of a theory T is called algebraically prime if A embeds into
every model of T .

Example 12
Let T be the theory of fields of characteristic p (for a prime p). Then
GF (p) is a (unique) algebraically prime model of T .

Theorem 13
Let T be a model complete theory. If T has an algebraically prime model,
then T is complete.

Proof.
I Let A be an algebraically prime model of T , and let B, C be arbitrary models of T .
e e
I By model completeness of T we have A ,→ B and A ,→ C.
I It follows that B ≡ C (exercise).
I Thus, all models of T are elementarily equivalent, so T is complete.
24 / 37
Some consequences of model completeness
Definition 11
A model A of a theory T is called algebraically prime if A embeds into
every model of T .

Example 12
Let T be the theory of fields of characteristic p (for a prime p). Then
GF (p) is a (unique) algebraically prime model of T .

Theorem 13
Let T be a model complete theory. If T has an algebraically prime model,
then T is complete.

Proof.
I Let A be an algebraically prime model of T , and let B, C be arbitrary models of T .
e e
I By model completeness of T we have A ,→ B and A ,→ C.
I It follows that B ≡ C (exercise).
I Thus, all models of T are elementarily equivalent, so T is complete.
25 / 37
Some consequences of model completeness
Definition 11
A model A of a theory T is called algebraically prime if A embeds into
every model of T .

Example 12
Let T be the theory of fields of characteristic p (for a prime p). Then
GF (p) is a (unique) algebraically prime model of T .

Theorem 13
Let T be a model complete theory. If T has an algebraically prime model,
then T is complete.

Proof.
I Let A be an algebraically prime model of T , and let B, C be arbitrary models of T .
e e
I By model completeness of T we have A ,→ B and A ,→ C.
I It follows that B ≡ C (exercise).
I Thus, all models of T are elementarily equivalent, so T is complete.
26 / 37
Some consequences of model completeness
Definition 11
A model A of a theory T is called algebraically prime if A embeds into
every model of T .

Example 12
Let T be the theory of fields of characteristic p (for a prime p). Then
GF (p) is a (unique) algebraically prime model of T .

Theorem 13
Let T be a model complete theory. If T has an algebraically prime model,
then T is complete.

Proof.
I Let A be an algebraically prime model of T , and let B, C be arbitrary models of T .
e e
I By model completeness of T we have A ,→ B and A ,→ C.
I It follows that B ≡ C (exercise).
I Thus, all models of T are elementarily equivalent, so T is complete.
27 / 37
Some consequences of model completeness

Theorem 14
Let T be a model complete theory. Then T is a Π2 theory.

Proof.
I We will show that T is closed under unions of chains, and use Theorem 10 of Lecture 6
(Chang-Loś-Suszko).
I Let (Ai : i < κ) be a chain of models of T , and let A = i<κ Ai .
S

I By model completeneness, it is an elementary chain, so by Lemma 7 of Lecture 6, A is an


elementary extension of each Ai .
I So, A |= T and hence T is closed under unions of chains.

Example 15
The converse of Theorem 13 does not hold. Consider the theory of dense
linear orders and Exercise 2.

28 / 37
Some consequences of model completeness

Theorem 14
Let T be a model complete theory. Then T is a Π2 theory.

Proof.
I We will show that T is closed under unions of chains, and use Theorem 10 of Lecture 6
(Chang-Loś-Suszko).
I Let (Ai : i < κ) be a chain of models of T , and let A = i<κ Ai .
S

I By model completeneness, it is an elementary chain, so by Lemma 7 of Lecture 6, A is an


elementary extension of each Ai .
I So, A |= T and hence T is closed under unions of chains.

Example 15
The converse of Theorem 13 does not hold. Consider the theory of dense
linear orders and Exercise 2.

29 / 37
Some consequences of model completeness

Theorem 14
Let T be a model complete theory. Then T is a Π2 theory.

Proof.
I We will show that T is closed under unions of chains, and use Theorem 10 of Lecture 6
(Chang-Loś-Suszko).
I Let (Ai : i < κ) be a chain of models of T , and let A = i<κ Ai .
S

I By model completeneness, it is an elementary chain, so by Lemma 7 of Lecture 6, A is an


elementary extension of each Ai .
I So, A |= T and hence T is closed under unions of chains.

Example 15
The converse of Theorem 13 does not hold. Consider the theory of dense
linear orders and Exercise 2.

30 / 37
Connections to decidability
Theorem 16
Let T be a complete, recursively axiomatisable theory. Then T is
decidable.

Proof.
I Since T is recursively axiomatisable, we can recursively enumerate the consequences of T .
I Since T is complete, for any formula ϕ, either ϕ or ¬ϕ will appear in the enumeration.

Theorem 17
Let T be a theory with a reduction set Φ. Assume that the following hold:
I For every formula ϕ, there is an algorithm for constructing a Boolean
combination of formulas from Φ, such that the resulting formula ϕ∗ is
equivalent to ϕ over T .
I For every formula ϕ ∈ Φ it is decidable whether T |= ϕ.
Then, T is decidable.
31 / 37
Connections to decidability
Theorem 16
Let T be a complete, recursively axiomatisable theory. Then T is
decidable.

Proof.
I Since T is recursively axiomatisable, we can recursively enumerate the consequences of T .
I Since T is complete, for any formula ϕ, either ϕ or ¬ϕ will appear in the enumeration.

Theorem 17
Let T be a theory with a reduction set Φ. Assume that the following hold:
I For every formula ϕ, there is an algorithm for constructing a Boolean
combination of formulas from Φ, such that the resulting formula ϕ∗ is
equivalent to ϕ over T .
I For every formula ϕ ∈ Φ it is decidable whether T |= ϕ.
Then, T is decidable.
32 / 37
Connections to decidability
Theorem 16
Let T be a complete, recursively axiomatisable theory. Then T is
decidable.

Proof.
I Since T is recursively axiomatisable, we can recursively enumerate the consequences of T .
I Since T is complete, for any formula ϕ, either ϕ or ¬ϕ will appear in the enumeration.

Theorem 17
Let T be a theory with a reduction set Φ. Assume that the following hold:
I For every formula ϕ, there is an algorithm for constructing a Boolean
combination of formulas from Φ, such that the resulting formula ϕ∗ is
equivalent to ϕ over T .
I For every formula ϕ ∈ Φ it is decidable whether T |= ϕ.
Then, T is decidable.
33 / 37
Theories admitting quantifier elimination
The following classes/theories admit quantifier elimination:
I Dense linear orders without endpoints (Tarski).
I Presburger arithmetic (Presburger). Equivalent to Peano arithmetic
without multiplication. It is complete and decidable.
I Algebraically closed fields (Tarski). Fields satisfying
I ∀x1 . . . xn · ∃y · y n + x1 y n−1 + · · · + xn−1 y + xn = 0, for every n ∈ N
I Ordered real closed fields (Tarski). Fields with ordering relation
compatible with the operations, satisfying
I ∀x1 . . . xn : x12 + · · · + xn2 6= −1, for every n ∈ N
I ∀x · ∃y · x = y 2 ∨ −x = y 2
I ∀x1 . . . xn ∃y : y n + x1 y n−1 + · · · + xn−1 y + xn = 0, for every odd n.

Quantifier elimination for industry


The theory of ordered real closed fields has a lot of industrial applications.
The main focus (as far as I know) is on efficient algorithms for quantifier
elimination. Ask Prof. Hirokazu Anai.
34 / 37
Theories admitting quantifier elimination
The following classes/theories admit quantifier elimination:
I Dense linear orders without endpoints (Tarski).
I Presburger arithmetic (Presburger). Equivalent to Peano arithmetic
without multiplication. It is complete and decidable.
I Algebraically closed fields (Tarski). Fields satisfying
I ∀x1 . . . xn · ∃y · y n + x1 y n−1 + · · · + xn−1 y + xn = 0, for every n ∈ N
I Ordered real closed fields (Tarski). Fields with ordering relation
compatible with the operations, satisfying
I ∀x1 . . . xn : x12 + · · · + xn2 6= −1, for every n ∈ N
I ∀x · ∃y · x = y 2 ∨ −x = y 2
I ∀x1 . . . xn ∃y : y n + x1 y n−1 + · · · + xn−1 y + xn = 0, for every odd n.

Quantifier elimination for industry


The theory of ordered real closed fields has a lot of industrial applications.
The main focus (as far as I know) is on efficient algorithms for quantifier
elimination. Ask Prof. Hirokazu Anai.
35 / 37
A weak version of Hilbert’s Nullstellensatz
Theorem 18 (Hilbert)
Let F be an algebraically closed field, and E a finite system of equations
and inequations over F in variables x. Suppose E has a solution in some
field G extending F. Then, E has a solution in F.

Proof.
I The assumeption that G extends F, means (wlog) F ≤ G.
I Let G be the algebraic closure of G. Then, we have F ≤ G ≤ G.
I Now, E = {p1 (x) = 0, . . . , pn (x) = 0, q1 (x) 6= 0, . . . , qm (x) 6= 0} for some polynomials
p1 , . . . , pn , q1 , . . . , qm with coefficients from F.
I Let ψ(x, c) be the conjunction of all formulas in E , with c being the coefficients.
I Then, E has a solution in G means G |= ∃x · ψ(x, c). This is an existential formula, so by
Loś-Tarski G |= ∃x · ψ(x, c).
I The theory of algebraically closed fields is model complete (see previous slide).
I Thus, the embedding F ≤ G is elementary, that is, F  G.
I So, F |= ∃x · ψ(x, c) as required.

36 / 37
A weak version of Hilbert’s Nullstellensatz
Theorem 18 (Hilbert)
Let F be an algebraically closed field, and E a finite system of equations
and inequations over F in variables x. Suppose E has a solution in some
field G extending F. Then, E has a solution in F.

Proof.
I The assumeption that G extends F, means (wlog) F ≤ G.
I Let G be the algebraic closure of G. Then, we have F ≤ G ≤ G.
I Now, E = {p1 (x) = 0, . . . , pn (x) = 0, q1 (x) 6= 0, . . . , qm (x) 6= 0} for some polynomials
p1 , . . . , pn , q1 , . . . , qm with coefficients from F.
I Let ψ(x, c) be the conjunction of all formulas in E , with c being the coefficients.
I Then, E has a solution in G means G |= ∃x · ψ(x, c). This is an existential formula, so by
Loś-Tarski G |= ∃x · ψ(x, c).
I The theory of algebraically closed fields is model complete (see previous slide).
I Thus, the embedding F ≤ G is elementary, that is, F  G.
I So, F |= ∃x · ψ(x, c) as required.

37 / 37

You might also like