Monadic second order logic as a model companion Sam van Gool - - PowerPoint PPT Presentation

monadic second order logic as a model companion
SMART_READER_LITE
LIVE PREVIEW

Monadic second order logic as a model companion Sam van Gool - - PowerPoint PPT Presentation

Monadic second order logic as a model companion Sam van Gool samvangool@me.com University of Utrecht (until 31 Aug) / Paris (starting 1 Sept) Logic Colloquium, Prague 13 August 2019 Automata and logic: example A programming problem: given


slide-1
SLIDE 1

Monadic second order logic as a model companion

Sam van Gool

samvangool@me.com

University of Utrecht (until 31 Aug) / Paris (starting 1 Sept)

Logic Colloquium, Prague 13 August 2019

slide-2
SLIDE 2

Automata and logic: example

◮ A programming problem: given a natural number in binary,

w ∈ {0, 1}∗, determine if w is congruent to 1 modulo 3.

1 / 9

slide-3
SLIDE 3

Automata and logic: example

◮ A programming problem: given a natural number in binary,

w ∈ {0, 1}∗, determine if w is congruent to 1 modulo 3.

◮ Solution 1: a (deterministic) automaton A: q0 q1 q2 1 1 1

Answer yes iff A accepts w.

1 / 9

slide-4
SLIDE 4

Automata and logic: example

◮ A programming problem: given a natural number in binary,

w ∈ {0, 1}∗, determine if w is congruent to 1 modulo 3.

◮ Solution 1: a (deterministic) automaton A: q0 q1 q2 1 1 1

Answer yes iff A accepts w.

◮ Solution 2: a monadic second order formula ϕ(W0, W1):

∃Q0∃Q1∃Q2(Q0(first) ∧ Q1(last) ∧ Partition(Q0, Q1, Q2)∧ ∀x([W0(x) ∧ Q0(x) → Q0(Sx)] ∧ [W1(x) ∧ Q0(x) → Q1(Sx)] ∧ . . . )) Answer yes iff w = (W0, W1) makes ϕ true.

1 / 9

slide-5
SLIDE 5

Regular languages

Regular languages over a finite alphabet Σ are subsets L ⊆ Σω which are ...

◮ recognizable by a finite automaton;

  • r, equivalently,

◮ definable by a formula of S1S,

the monadic second order logic of one successor. Büchi 1960

2 / 9

slide-6
SLIDE 6

A model complete theory

A functional language L : Boolean algebra operations (⊥, ∪, −), two unary functions, X and F, and a constant I.

3 / 9

slide-7
SLIDE 7

A model complete theory

A functional language L : Boolean algebra operations (⊥, ∪, −), two unary functions, X and F, and a constant I. The Boolean algebra P(ω) is an L-structure with:

◮ Xa := {t ∈ ω | t + 1 ∈ a}, ◮ Fa := {t ∈ ω | ∃t′ ≥ t : t′ ∈ a}, ◮ I := {0}.

3 / 9

slide-8
SLIDE 8

A model complete theory

A functional language L : Boolean algebra operations (⊥, ∪, −), two unary functions, X and F, and a constant I. The Boolean algebra P(ω) is an L-structure with:

◮ Xa := {t ∈ ω | t + 1 ∈ a}, ◮ Fa := {t ∈ ω | ∃t′ ≥ t : t′ ∈ a}, ◮ I := {0}.

Theorem

The first order L-theory of P(ω) is model complete.

A theory T ∗ is model complete iff every formula is T ∗-equivalent to an existential formula.

Ghilardi, G. JSL 2017

3 / 9

slide-9
SLIDE 9

Proving model completeness with automata

L-theory of P(ω)

4 / 9

slide-10
SLIDE 10

Proving model completeness with automata

L-theory of P(ω) S1S

“standard translation”

4 / 9

slide-11
SLIDE 11

Proving model completeness with automata

L-theory of P(ω) S1S Word automaton

“standard translation” Büchi’s Theorem

4 / 9

slide-12
SLIDE 12

Proving model completeness with automata

L-theory of P(ω) S1S Word automaton

“standard translation” Büchi’s Theorem existential L-description

4 / 9

slide-13
SLIDE 13

Proving model completeness with automata

L-theory of P(ω) S1S Word automaton

“standard translation” Büchi’s Theorem existential L-description existential L-description

4 / 9

slide-14
SLIDE 14

An existential L-description of a word automaton

◮ Let A = (Q, Σ, δ, q0, F) be a word automaton over a finite

alphabet Σ, i.e., a function δ: Q × Σ → P(Q), an initial state q0 ∈ Q and a subset F ⊆ Q of final states.

5 / 9

slide-15
SLIDE 15

An existential L-description of a word automaton

◮ Let A = (Q, Σ, δ, q0, F) be a word automaton over a finite

alphabet Σ, i.e., a function δ: Q × Σ → P(Q), an initial state q0 ∈ Q and a subset F ⊆ Q of final states.

◮ Write Σ = {0, . . . , s}, Q = {0, . . . , m}, q0 = 0. ◮ A word W : ω → Σ is a partition (W0, . . . , Ws) of ω; Wj = W −1(j).

5 / 9

slide-16
SLIDE 16

An existential L-description of a word automaton

◮ Let A = (Q, Σ, δ, q0, F) be a word automaton over a finite

alphabet Σ, i.e., a function δ: Q × Σ → P(Q), an initial state q0 ∈ Q and a subset F ⊆ Q of final states.

◮ Write Σ = {0, . . . , s}, Q = {0, . . . , m}, q0 = 0. ◮ A word W : ω → Σ is a partition (W0, . . . , Ws) of ω; Wj = W −1(j).

Key Observation. The automaton A accepts a word W : ω → Σ iff P(ω), [wi → Wi] | = α(w0, . . . , ws), where α is the ∃ L-formula:

5 / 9

slide-17
SLIDE 17

An existential L-description of a word automaton

◮ Let A = (Q, Σ, δ, q0, F) be a word automaton over a finite

alphabet Σ, i.e., a function δ: Q × Σ → P(Q), an initial state q0 ∈ Q and a subset F ⊆ Q of final states.

◮ Write Σ = {0, . . . , s}, Q = {0, . . . , m}, q0 = 0. ◮ A word W : ω → Σ is a partition (W0, . . . , Ws) of ω; Wj = W −1(j).

Key Observation. The automaton A accepts a word W : ω → Σ iff P(ω), [wi → Wi] | = α(w0, . . . , ws), where α is the ∃ L-formula: ∃q0, . . . , qm(“the qi partition ω” ∧

  • 0≤i≤m

0≤j≤s

 qi ∩ wj ⊆

  • k∈δ(i,j)

Xqk  

5 / 9

slide-18
SLIDE 18

An existential L-description of a word automaton

◮ Let A = (Q, Σ, δ, q0, F) be a word automaton over a finite

alphabet Σ, i.e., a function δ: Q × Σ → P(Q), an initial state q0 ∈ Q and a subset F ⊆ Q of final states.

◮ Write Σ = {0, . . . , s}, Q = {0, . . . , m}, q0 = 0. ◮ A word W : ω → Σ is a partition (W0, . . . , Ws) of ω; Wj = W −1(j).

Key Observation. The automaton A accepts a word W : ω → Σ iff P(ω), [wi → Wi] | = α(w0, . . . , ws), where α is the ∃ L-formula: ∃q0, . . . , qm(“the qi partition ω” ∧

  • 0≤i≤m

0≤j≤s

 qi ∩ wj ⊆

  • k∈δ(i,j)

Xqk   ∧ I ⊆ q0 ∧ F

  • i∈F qi
  • = ⊤).

5 / 9

slide-19
SLIDE 19

The theory is a model companion

A theory T ∗ is a model companion of a theory T iff T ∗ is model complete, and T and T ∗ have the same universal consequences.

Theorem

The L-theory of P(ω) is the model companion of the theory of L-structures axiomatized by the following universal sentences:

6 / 9

slide-20
SLIDE 20

The theory is a model companion

A theory T ∗ is a model companion of a theory T iff T ∗ is model complete, and T and T ∗ have the same universal consequences.

Theorem

The L-theory of P(ω) is the model companion of the theory of L-structures axiomatized by the following universal sentences:

◮ equations for Boolean algebras; ◮ X is a Boolean homomorphism; ◮ Fa is the least fixed point of x → a ∨ Xx; ◮ I is an atom which is below Fa for any a = ⊥, and XI = ⊥.

Ghilardi, G. JSL 2017

6 / 9

slide-21
SLIDE 21

Binary trees

The full binary tree is 2∗, finite sequences of 0’s and 1’s.

7 / 9

slide-22
SLIDE 22

Binary trees

The full binary tree is 2∗, finite sequences of 0’s and 1’s. To obtain a model complete theory, more complex operators are needed to express acceptance by a tree automaton.

7 / 9

slide-23
SLIDE 23

Binary trees

The full binary tree is 2∗, finite sequences of 0’s and 1’s. To obtain a model complete theory, more complex operators are needed to express acceptance by a tree automaton. A functional language L2 : Boolean algebra operations (⊥, ∪, −), constant I, unary operations X0, X1, binary operations EU and AF.

7 / 9

slide-24
SLIDE 24

Binary trees

The full binary tree is 2∗, finite sequences of 0’s and 1’s. To obtain a model complete theory, more complex operators are needed to express acceptance by a tree automaton. A functional language L2 : Boolean algebra operations (⊥, ∪, −), constant I, unary operations X0, X1, binary operations EU and AF. The Boolean algebra P(2∗) is an L2-structure with

◮ I := {ǫ}, ◮ Xia := {t ∈ ω | t · i ∈ a} for i = 0, 1, ◮ t ∈ EU(a, b) iff there Exists a path t = t0, . . . , tn such that, for

i < n, ti ∈ a, and (Until) tn ∈ b,

7 / 9

slide-25
SLIDE 25

Binary trees

The full binary tree is 2∗, finite sequences of 0’s and 1’s. To obtain a model complete theory, more complex operators are needed to express acceptance by a tree automaton. A functional language L2 : Boolean algebra operations (⊥, ∪, −), constant I, unary operations X0, X1, binary operations EU and AF. The Boolean algebra P(2∗) is an L2-structure with

◮ I := {ǫ}, ◮ Xia := {t ∈ ω | t · i ∈ a} for i = 0, 1, ◮ t ∈ EU(a, b) iff there Exists a path t = t0, . . . , tn such that, for

i < n, ti ∈ a, and (Until) tn ∈ b,

◮ t ∈ AF(a, −b) iff for All infinite paths t = t0, t1, . . . there is a

(Future) ti ∈ a, provided that tj ∈ b for infinitely many j.

7 / 9

slide-26
SLIDE 26

Model companion for binary trees

Theorem

The L2-theory of P(2∗) is model complete, and is in fact the model companion of an L2-theory with a finite universal axiomatization. Ghilardi, G. LICS 2016

8 / 9

slide-27
SLIDE 27

Model companion for binary trees

Theorem

The L2-theory of P(2∗) is model complete, and is in fact the model companion of an L2-theory with a finite universal axiomatization. Ghilardi, G. LICS 2016

◮ Proving model completeness crucially uses tree automata

  • riginally developed for deciding S2S (Rabin 1969).

◮ We obtain an analogous result for ‘bisimulation-invariant’

MSO, i.e., the modal µ-calculus (Janin-Walukiewicz 1996).

8 / 9

slide-28
SLIDE 28

Ongoing work and questions

◮ Ongoing work: extending these results to general trees; this

requires an infinite language that can count successors.

◮ Where do L-structures and L2-structures fit in model theory?

◮ Context: model companions also exist for Heyting algebras and

certain modal algebras; but the methods are different.

◮ Can automata methods be useful for proving the model

completeness of other theories (especially if they have a ‘computation’ flavor)?

9 / 9

slide-29
SLIDE 29

Monadic second order logic as a model companion

Sam van Gool

samvangool@me.com

University of Utrecht (until 31 Aug) / Paris (starting 1 Sept)

Logic Colloquium, Prague 13 August 2019