Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Plan Part I Ehrenfeucht-Fraïssé games and the composition method Games in logic Games in logic Games are used to capture “dynamics” of formulas. They offer, understanding of formula constructors in “operational” way. The three settings presented here focus on different constructors and in consequence use very different techniques. Igor Walukiewicz CNRS, LaBRI Bordeaux 1 Ehrenfeucht-Fraïssé games focus on the behaviour of quantifiers. This leads to a notion of type and to the compositional method. Algorithmic-Logical Theory of Infinite Structures October 2007 2 Parity games are used to understand operators defined by fixpoints. This leads to theory of automata on infinite words or trees. 3 Game semantics is used to understand dynamics of propositional constructs. This leads to accurate models of proofs and of programming languages. Question: What are the connections between these three settings? Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 1 / 48 Igor Walukiewicz (LaBRI) Games in logic MSOL and Ehrenfeucht-Fraïssé games Monadic second order logic Monadic second order logic Instead of quantification over elements we have quantification over sets. ∃X .ϕ(X ), ∀X .ϕ(X ) We have the inclusion predicate 2 / 48 We have the inclusion predicate Standard predicates can be “lifted” to sets: succ(X , Y ), X ≤ Y , X ⊆ P ! ≡k (B, Q) ! Important property: (A, P) iff k ! = Tp (B, Q). ! Tp (A, P) θk k-types θ1m θ2m θ1r θ1ε θ2r θ2ε θ3ε θ4r θ3r θ4ε θ5ε θ6ε θ7ε θ8ε θ1ε θ2r θ2ε θ3ε θ4r θ3r θ4ε θ5ε θ6ε 4 / 48 θ7ε " Q) " :Q " ⊆ |A|m } {Tpr (A, P, ! ≡k (B, Q) ! Important property: (A, P) Igor Walukiewicz (LaBRI) iff ! = Tpk (B, Q). ! Tpk (A, P) Games in logic Dagstuhl 2007 Composition theorems Sum The k-type of A + B is determined by (and can be computed from) the k types of A and B. The k-type of A + B is determined by (and can be computed from) the k types of A and B. B P Ordered sum ! The k-type of P θ2m θ1r Dagstuhl 2007 " for k = r · m is Tpk (A, P), Sum A θ1m Games in logic " is the set of qf formulas true in (A, P). Tpε (A, P) Composition theorems θk 3 / 48 " ≡k (B, Q) " iff Duplicator has a winning strategy in the k game on Fact: (A, P) " and (B, Q). " (A, P) If k = r · m then Spoiler chooses one of the structures, say A, and m sets in this structure P1 , . . . , Pm . Duplicator replies by choosing Q1 , . . . , Qm in B and then the " P1 . . . , Pm ) and (B, Q, " Q1 . . . , Qm ). r game is played on (A, P, k Dagstuhl 2007 " and (B, Q) " are k-equivalent, written (A, P) " ≡k (B, Q) " if they Two structures (A, P) satisfy the same formulas of quantifier rank k. If k is empty then Duplicator wins iff the two structures satisfy the same predicates " and Q. " Otherwise Spoiler wins. with respect to P Igor Walukiewicz (LaBRI) Games in logic " n"yn . . . Q " 1"y1 . ϕ("x , "yn , . . . , "y1 ) Q We also have k = (k1 , . . . , kn ), a vector of positive natural numbers. 4 / 48 Igor Walukiewicz (LaBRI) " i "yi stands for the vector of length ki of quantifiers of the same type. where Q X ⊆Y " and (B, Q) " with some distinguished sets of We have two structures (A, P) elements. Dagstuhl 2007 CTL=Chain Logic. A formula of quantifier rank k = (k1 , . . . , kn ) has the form: Standard predicates can be “lifted” to sets: succ(X , Y ), X ≤ Y , X ⊆ P Games in logic Computing MSO-theories of finite sequences and infinite sequences. Quantifier alternation Ehrenfeucht-Fraïssé games for MSOL Igor Walukiewicz (LaBRI) E-F games and the composition theorem for sums (Shelah’s way). E-F games and types Instead of quantification over elements we have quantification over sets. ∃X .ϕ(X ), ∀X .ϕ(X ) X ⊆Y Monadic second order logic. [Thomas, “Ehrenfeucht Games, the Composition Method, and the Monadic Theory of Ordinal Words.” Structures in Logic and Computer Science, LNCS 1261] Dagstuhl 2007 MSOL and Ehrenfeucht-Fraïssé games Ehrenfeucht-Fraïssé games focus on the behaviour of quantifiers. This leads to a notion of type and to the compositional method. In consequence we study how to cut structures so that from the theory of parts one can obtain the theory of the whole. The other application is to obtain normal forms of formulas. i∈ω 5 / 48 Ai is determined by r-type of the sequence (typek (A0 ), typek (A1 ), . . . ); where k and r have the same length. (This sequence is over the alphabet {Qτ : τ ∈ Types k }. Tpk (A) Tpk (B) θ8ε Tpk (A) Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 6 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 7 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 7 / 48 Part Ib Computing the theory of finite sequences Computing the theory of "ω, ≤$ The theory of finite sequences Consider structures An = ({1, . . . , n}, ≤). The base step: Tpm (ω) Let Tpkfin = {Tpk (An ) : n ∈ N} (these are all possible k-thypes of finite sequences) We compute by hand the theory Tpm (ω) that is: {Tpε (ω, P) : P ∈ P(ω)m } k Computing Tp (Fin(m)) Deciding logics using compositional theorems Observation Tpk (A1 ) A k-type of a structure tells what k-formulas are true in the structure. k k " where each Qi is either ∅ or ω. If we can compute Tpk (ω) then we can Tpk (ω, Q) k Tp (A2 ) = Tp (A1 ) + Tp (A1 ) First, we will calculate possible types of finite sequences. .. . Then, we calculate the types for infinite sequences. k Tp (An+1 ) = Tpk (An ) + Tpk (A1 ) = Tpk (Aj ) We take Tpkfin "n = i=1 Remark Induction step for k · m for j ≤ n " :P " ∈ P(ω)m } Tpk·m (ω) = {Tpk (ω, P) " can be presented as τ + Each Tpk (ω, P) k Tp (Ai ). Computing rest is ∅. Tpk·m fin gives us all possible k-types of {({1, . . . , n}, ≤, P1 , . . . , Pm ) : for n ∈ N}. Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 8 / 48 Ramsey argument on types Tpkfin (m) = {τ : ∃σ ∈ Tpk·m fin , τ ∈ σ} Games in logic " can be presented as τ + We show that each Tpk (ω, P) " Consider A = (ω, ≤, P) ! ω " :P " ∈ P(ω)m } Tpk·m (ω) = {Tpk (ω, P) σ for τ, σ ∈ Tpkfin (m). " can be presented as τ + Each Tpk (ω, P) Computing rest is ∅. Each interval (i, j) has its own k-theory. By Ramsey Theorem there is an infinite S ⊆ N and σ ∈ Tpkfin (m) such that for all i, j ∈ S, FA (i, j) = σ So Tpk (A) can be presented as τ + ! ω σ; where τ = F (1, i) and i = min(S). In consequence, it is enough to compute all possible: ! τ + ω σ for τ, σ ∈ Tpkfin (m). ! i∈ω Dagstuhl 2007 11 / 48 We use first-order logic over predicates x ≤ y, left(x), right(x). We will not need vectorial ranks. So we write Tpk (A) for the k-type of A, where k ∈ N. k From trees to paths: exp (t, v) ω σ for τ, σ ∈ Tpkfin (m). 10 / 48 Understanding logics on trees FO=CTL∗ over finite binary trees. Variants of this argument work for infinite trees, or unranked trees, etc. Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 12 / 48 (λ(w1 ), r, Tp (v1 ↓)) Remark Introducing parity games via model-checking. By Kamp theorem over sequences FOL=LTL, so we have an equivalent LTL formula α (with predicates from Σ × {l, r} × Tpk ) k (λ(w4 ), l, Tp (w4 ↓)) 13 / 48 Parity games are used to understand operators defined by fixpoints. Such operators talk not about a model but about an unfolding of the model. Their semantics is captured by infinite plays and parity condition. For every Tpk (expk (t, v)) there is a first order formula defining sequences with this property. (λ(w3 ), r, Tpk (v3 ↓)) Dagstuhl 2007 Parity games and model checking On infinite trees it is not the case as in FOL we cannot single out an infinite path. By composition theorem it suffices to know {Tpk (expk (t, v)) : v ∈ t}. Recall that expk (t, v) are the sequences over Σ × {l, r} × Tpk . (λ(w2 ), l, Tpk (v2 ↓)) Games in logic The expressive powers of FO and CTL∗ on finite binary trees are the same We want to express Tpk+1 (t). k Igor Walukiewicz (LaBRI) Part II Proof: We express FO-types in CTL∗ exp(t, w4 ) = Parity games ≡ model-checking of the mu-calculus. Where we can go from here: changing winning conditions, or changing the way games are played. By induction every type in Tpk is expressible by a CTL∗ formula. Theorem We convert α to a CTL∗ formula α # by replacing predicates from Σ × {l, r} × Tpk (t) by an appropriate CTL∗ formulas. The k + 1-type of a finite binary tree t is determined by the set Igor Walukiewicz (LaBRI) Dagstuhl 2007 Observation Theorem (Hafer & Thomas) v3 ! (Tpk (A0 ), Tpk (A1 ), . . . ); where k and r have the same length. (This sequence is over the alphabet {Qτ : τ ∈ Types k }. First-order theory of finite binary trees w4 Games in logic Part Ic Ai is determined by r-type of the sequence FO=CTL∗ on finite binary trees w3 Igor Walukiewicz (LaBRI) Composition theorems for trees permit to talk about properties of trees in terms of its paths. A composition theorem for trees v2 σ for τ, σ ∈ Tpkfin (m). " where each Qi is either ∅ or ω. If we can compute Tpk (ω) then we can Tpk (ω, Q) Games in logic v1 ω " where only one Qi = ω and the σ reduces to computing Tpr (ω, Q) " where only one Qi = ω and the σ reduces to computing Tpr (ω, Q) ω Ordered sum ! The k-type of We have FA : N2 → Tpkfin (m). w2 9 / 48 ω ! Induction step for k · m " :P " ∈ P(ω)m } Tpk·m (ω) = {Tpk (ω, P) w1 Dagstuhl 2007 Computing the theory of "ω, ≤$ (cont.) We want to compute: Igor Walukiewicz (LaBRI) Igor Walukiewicz (LaBRI) ! {Tpk (expk (t, v)) : v ∈ t} Games in logic We write CTL∗ formula of the form Dagstuhl 2007 14 / 48 Igor Walukiewicz (LaBRI) $ i∈I % Eα #i ∧ A( Games in logic i∈I )α #i expressing Tpk+1 (t). Dagstuhl 2007 15 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 16 / 48 Verification as a game Example Example Verification (Model Checking) a Eve wins from the initial position of G(M, ψ) iff M ! ψ s ! α∧β s!β s!α s ! (a)α s ! [a]α t!α t!α s!β Igor Walukiewicz (LaBRI) Dagstuhl 2007 s4 ! P Example a s2 b b s6 Game s0 ! (a)[b]P s1 ! [b]P s4 ! P b s6 s0 ! (a)[b]P s2 ! [b]P s5 ! P Igor Walukiewicz (LaBRI) s1 ! [b]P s4 ! P s6 ! P Games in logic Dagstuhl 2007 18 / 48 s2 ! [b]P s5 ! P Igor Walukiewicz (LaBRI) s6 ! P Games in logic Game rules: reachability Game rules: safety Reachability: "·$∗ P Safety: [·]ω P Dagstuhl 2007 18 / 48 ! (a)[b]P a s1 s5 s5 ? s0 b s4 s6 b Game s1 ! [b]P 17 / 48 s2 b s4 b s0 ! (a)[b]P (s, t) ∈ Ra Games in logic s5 b ! (a)[b]P a s1 Game Game rules s ! α∨β a s2 b s4 Construct a game G(M, ψ) of two players: Adam and Eve. Fix the rules in such a way that ? s0 ! (a)[b]P a s1 Reformulation s!α ? s0 Given a transition system M and a property ψ, check if M ! ψ s0 s1 s2 s0 s1 s2 (·)∗ P (·)∗ P (·)∗ P [·]ω P [·]ω P [·]ω P P ∨ (·)(·)∗ P P ∨ (·)(·)∗ P P ∨ (·)(·)∗ P P ∧ [·][·]ω P P ∧ [·][·]ω P P ∧ [·][·]ω P Who wins? Who wins? Eve wins if the game ends. Eve wins if the game continues forever or ends because there is no successor. s2 ! [b]P s5 ! P Igor Walukiewicz (LaBRI) s6 ! P Games in logic Dagstuhl 2007 18 / 48 Different games for different proprieties M, s ! α modal logic G(M, s, α) finite duration games reachability Igor Walukiewicz (LaBRI) Games in logic 19 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 Games Games Definition (Game G = "VE , VA , R, λ : V → C , Acc ⊆ C ω $) Definition (Game G = "VE , VA , R, λ : V → C , Acc ⊆ C ω $) c e b d c e b d 20 / 48 safety a termination Dagstuhl 2007 a f f Definition (Winning a play) non-termination Eve wins a play v0 v1 . . . iff the sequence is in Acc. Definition (Winning position) A strategy for Eve is σE : V ∗ × VE → V . A strategy is winning from a given position iff all the plays starting in this position and respecting the strategy are winning. A position is winning if there is a winning strategy from it. Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 21 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 22 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 22 / 48 What kind of winning conditions The parity condition Part IIb Properties Definition (Parity condition: Ω : V → {0, . . . , d}) "v ∈ Acc reachability safety iff lim inf Ω(vn ) is even n→∞ Other conditions in terms of parity condition etc. Almost always states from F ⊆ V . Ω : V → {1, 2} such that Ω(v) = 2 iff v ∈ F . reachability: Acc =(sequences passing through a position from F ), safety: Acc =(sequences of elements of F ), Reachability for F . Arrange so that each state from F is winning. repeated reachability: Acc =(sequences with infinitely many elements from F ). ultimately safe: Acc =(almost all elements from F ). Igor Walukiewicz (LaBRI) Games in logic Parity games ≡ µ-calculus model checking Infinitely often states from F ⊆ V . Ω : V → {0, 1} such that Ω(v) = 0 iff v ∈ F . Winning conditions Safety for F . Ω(v) = 0 for v ∈ F and arrange so that all states not in F are loosing. Dagstuhl 2007 23 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 24 / 48 Igor Walukiewicz (LaBRI) The mu-calculus Games for the mu-calculus Game rules Syntax Setup Game rules We are given a transition system M and a formula α0 . P | ¬P | X | α | α ∨ β | α ∧ β | (a)α | [a]α | µX .α | νX .α Given M = (V , {Ea }a∈Act , P M , . . .) and Val : Var → P(V ) we define [[α]]M Val ⊆ P(V ). s ! α∨β s!α [[X ]]M Val =Val(X ) s!α Game rules M [[P]]M Val =P s ! α∧β s!β s!α s!β # # # M [[(a)α]]M Val ={v : ∃v . Ea (v, v ) ∧ v ∈ [[α]]Val } [[µX .α(X )]]M Val = & {S ⊆ V : [[α(S)]]M Val ⊆ S} In s ! P Eve wins iff s ∈ P M Notation: M, s ! α for s ∈ [[α]]M Val , where Val will be clear from the context. Games in logic Dagstuhl 2007 26 / 48 Example: Reachability s1 Igor Walukiewicz (LaBRI) s!α s!α s0 Games in logic Dagstuhl 2007 29 / 48 s ! α∧β s!β s!α s!α In s ! ¬P 25 / 48 s ! [a]α s ! (a)α s!β In s ! P Eve wins iff s ∈ P M s!α (s, t) ∈ RaM Eve wins iff s ∈ P M (s, t) ∈ RaM Eve wins iff s /∈ P M s ! µX .α(X ) s ! νX .α(X ) s ! α(µX .α(X )) s ! α(νX .α(X )) Dagstuhl 2007 27 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 28 / 48 Dagstuhl 2007 29 / 48 Example: Reachability s1 Reachability: "·$∗ P ≡ µX .P ∨ "·$X s0 s1 α ≡ µX .P ∨ (·)X α ≡ µX .P ∨ (·)X α P ∨ (·)α P ∨ (·)α Igor Walukiewicz (LaBRI) Dagstuhl 2007 These two rules may be the source of infinite plays. Games in logic Reachability: "·$∗ P ≡ µX .P ∨ "·$X s2 α ≡ µX .P ∨ (·)X Igor Walukiewicz (LaBRI) s ! [a]α Example: Reachability Reachability: "·$∗ P ≡ µX .P ∨ "·$X s0 In s ! ¬P s ! (a)α What to do with µX .α(X ) and νX .α(X )? We will give a characterization of the semantics in terms of games Igor Walukiewicz (LaBRI) s ! α∨β We define a game G(M, α0 ) where Eve wins from (s0 ! α0 ) iff M, s0 ! α0 . Semantics Games in logic Games in logic s2 Dagstuhl 2007 29 / 48 Igor Walukiewicz (LaBRI) Games in logic s2 Example: Reachability Example: Reachability Reachability: "·$∗ P ≡ µX .P ∨ "·$X s0 s1 s2 α ≡ µX .P ∨ (·)X α α P ∨ (·)α P ∨ (·)α Igor Walukiewicz (LaBRI) Reachability: "·$∗ P ≡ µX .P ∨ "·$X Games in logic Dagstuhl 2007 29 / 48 Example: Reachability s0 s1 s2 α ≡ µX .P ∨ (·)X α α P ∨ (·)α P ∨ (·)α Eve wins if the game ends in P. µX .α(X ) = s1 s2 s0 s1 α ≡ µX .P ∨ (·)X α α α ≡ µX .P ∨ (·)X α α P ∨ (·)α P ∨ (·)α P ∨ (·)α P ∨ (·)α P ∨ (·)α P ∨ (·)α Igor Walukiewicz (LaBRI) µX1 Games in logic P ∨ (·)α " τ ∈Ord Eve wins if the game ends in P. µX .α(X ) = Dagstuhl 2007 29 / 48 s2 β ≡ νX .P ∧ [·]X β β P ∧ [·]β P ∧ [·]β P ∧ [·]β 2 Part IIc Other program logics (with fixpoint definable operators) can be handled in the same way. This approach permits also to handle satisfiability. It also explains algorithmics of verification nicely. This is especially useful for verification of infinite structures. ν2 X .µ3 Y (P ∧ (·)X ) ∨ (·)Y Games in logic Dagstuhl 2007 30 / 48 Mean pay-off game: G = "VE , VA , R, w : (VE ∪ VA ) → N$ Theorem (Ehrenfeucht & Mycielski) n→inf For Adam it is lim sup. n 1' n Outcome of v0 , v1 , . . . is i=1 (1 − δ) here 0 < δ < 1 is a discount factor. !∞ i=0 For every finite discounted payoff game the value exists in every vertex and is given as a unique solution of the set of equations: xv = (1 − δ)w(v) + i δ w(vi ) if v ∈ VE if v ∈ VA When δ → 1 then VδZP (v) → V EM (v). Adam has a strategy from v to have an outcome ≤ Vv . Games in logic max(v,u)∈R δxu min(v,u)∈R δxu Theorem (Zwick & Paterson) Eve has a strategy from v to have an outcome ≥ Vv , and Igor Walukiewicz (LaBRI) ( There are optimal positional strategies. Value of the game in a vertex v is a number Vv such that: 32 / 48 31 / 48 Theorem (Zwick and Paterson) Value of the game Dagstuhl 2007 Dagstuhl 2007 Every vertex of a mean payoff game has a value. Moreover the two players have positional optimal strategies. w(vi ). Discounted payoff game G = "VE , VA , R, w : (VE ∪ VA ) → R$ Games in logic Games in logic Results on payoff games lim inf Igor Walukiewicz (LaBRI) Igor Walukiewicz (LaBRI) Other kinds of winning conditions Outcome for Eve of a play v0 , v1 , . . . is: Different ways of winning and playing games. 29 / 48 Remarks Example Igor Walukiewicz (LaBRI) Dagstuhl 2007 There is a µ-calculus formula which is true exactly in positions where Eve wins. Eve wins if the smallest priority appearing infinitely often is even. 29 / 48 Games in logic Game can be represented as a transition system. The winning condition is the parity condition Dagstuhl 2007 µτ X .α(X ) Game solving ⇒ MC if β is a subformula of α then β has bigger rank than α. Eve wins if the game continues for ever ends in P. τ ∈Ord The problem M, s0 ! α0 is reduced to deciding if Eve wins from the position (s0 ! α0 ) in the game G(M, α0 ). 4 ··· ν’s have even ranks, µ1 Y .ν2 X .(P ∧ (·)X ) ∨ (·)Y " MC ⇒ game solving α(X1 , X2 , . . . ) νX4 3 Igor Walukiewicz (LaBRI) s2 Model checking ≡ game solving µ’s have odd ranks, µτ X .α(X ) s1 Games in logic µX3 νX2 1 s0 Igor Walukiewicz (LaBRI) Reachability: "·$∗ P ≡ µX .P ∨ "·$X s0 Defining winning conditions Reachability: "·$∗ P ≡ µX .P ∨ "·$X Safety: [·]ω P ≡ νX .P ∧ [·]X Example: Reachability Dagstuhl 2007 33 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 34 / 48 Modifying rules for playing games Deterministic Probabilistic Perfect information stochastic games Perfect information stochastic games, cont. Definition None of the players may be sure to win Apart from positions for Eve and Adam there are randomized positions where a successor is chosen according to a probability distribution. 1/2 Turn based 2/3 bx 1/3 5 0 1/3 2/3 ay Games in logic Dagstuhl 2007 35 / 48 Concurrent games Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 36 / 48 Part III Game semantics Example (d,d) hide (r,d), (d,t Dagstuhl 2007 Adam’s moves: delay,throw,out α1 ∨ α2 2 β α1 2 β Proofs as games. home α1 ∧ α2 2 β α2 2 β α1 2 β α 2 β 1 ∨ β2 Game semantics: an example. Brief summary of the results, and (potential) applications. (o,o) α 2 β1 α 2 β2 α 2 β1 What one can prove with these rules? There exists randomized strategies, but they may require infinite memory. Not much. These rules characterize ∨ and ∧ in the free lattice. Games in logic Dagstuhl 2007 38 / 48 Satisfiability cont. Different ways of proving Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 39 / 48 Satisfiability cont. Different ways of proving α1 ∧ α2 + α1 ∨ α2 α1 ∧ α2 + α1 ∨ α2 α1 ∧ α2 2 α1 ∨ α2 α1 ∧ α2 2α1 ∨ α2 α1 2 α1 α2 2α2 α1 2 α1 α2 2α2 α1 ∧ α2 2 α1 2 α →(α →α) qo qp ao ap α1 ∧ α2 2α2 Things we cannot prove (α1 ∨ α2 ) ∧ β 2 (α1 ∧ β) ∨ (α2 ∧ β) The normal solution is to go to sets of formulas on both sides. α1 ∧ α2 2 α1 ∨α2 qo qp ao Games in logic Dagstuhl 2007 40 / 48 Dagstuhl 2007 42 / 48 2 α →(α →α) qo qp ao ap Proofs as programs Strategies instead of proofs Igor Walukiewicz (LaBRI) Games in logic Strategies instead of proofs α1 ∧ α2 2α1 ∨ α2 α1 ∧ α2 2α2 α 2 β2 Proofs as strategies α1 ∧ α2 2 α1 ∨ α2 α1 ∧ α2 2 α1 Igor Walukiewicz (LaBRI) α2 2 β α 2 β1 ∨ β2 Observation [de Alfaro, Henzinger] Igor Walukiewicz (LaBRI) 37 / 48 Game rules Game semantics is used to understand dynamics of propositional constructs. It extracts computational content from proofs, abstracting from irrelevant detail. This leads to accurate models of proofs and programming languages. As for E-F types, it is not the winning the matters, but the ways to play. Eve’s moves: delay,run,out ) Games in logic We have examined s ! α now we look at α + β Two players choose their moves concurrently. Their joint choice determines the successor. wet Igor Walukiewicz (LaBRI) Satisfiability in propositional logic Definition ) (r,t 3 In a finite game each state has a value and each player has an positional pure and optimal strategy. Adam wins in this game. by by Igor Walukiewicz (LaBRI) 4 Theorem (de Alfaro &, Majumdar, Chetterjee & Jurdzinski & Henzinger, Zielonka) 3 4 1/3 ay 5 Eve wins with the probability 2/3 and Adam with the probability 1/3. ax bx Concurrent 0 Example 1/2 ax 2/3 Dagstuhl 2007 41 / 48 ap Igor Walukiewicz (LaBRI) λx.λy.y : N →N →N qo qp ao ap α1 ∧ α2 2 α1 ∨α2 qo qp ao ap Games in logic Dagstuhl 2007 41 / 48 Igor Walukiewicz (LaBRI) λx.λy.x : N →N →N qo qp ao ap Games in logic More examples Program semantics as a strategy Program semantics as a strategy f :N →N 2 if (f (5) = 6) then 7 else 0: N qo qp qo 5p 6o 7p f :N →N 2 if (f (5) = 6) then 7 else 0: N qo qp qo 5p 6o 7p Remarks f :N →N 2 if (f (5) = 6) then 7 else 0: N qo qp qo 5p 1o 0p Different strategies correspond to different programs. The semantics is not about provability. The semantics is not about provability. Semantics of a program is, roughly, a set of words: all plays permitted by the strategy. Semantics of a program is, roughly, a set of words: all plays permitted by the strategy. Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 43 / 48 Idealized Algol b ::= com | exp | var Games in logic Γ 2 skip, Ω : com i ∈ {0, . . . , max} Γ 2 i : exp Γ, x : τ 2 M : τ # Γ 2 λx.M : τ → τ # Γ 2 M : τ → τ# Γ 2 N : τ Γ 2 MN : τ # Γ 2 M : exp Γ 2 M0 , M1 : b Γ 2 ifzero MM0 M1 : b qp qp qo qo qp 5p no qo qp 5p Dagstuhl 2007 np 1p m0 no mo (m + n)p 43 / 48 Igor Walukiewicz (LaBRI) Games in logic mp Dagstuhl 2007 44 / 48 Reasoning about programs using game semantics Two terms are observationally equivalent, M ≈ N , if they behave the same in all contexts C [M ] ⇓ skip iff C [N ] ⇓ skip Γ, x : τ 2 x : τ Reasoning about program equivalence Quantification over all program contexts C [·] makes this notion hard to work with. If so, we can check program equivalence by comparing languages for the two automata. Theorem (Abramsky & McCusker) Γ 2 M : com Γ 2 N : b Γ 2 M; N : b Construct the strategy representing a given program (typically using finite approximations for data domains). Sometimes this strategy can be represented by a finite automaton, or by a deterministic pushdown automaton. The notion is very compelling as it captures well semantical indistinguishability. The game semantics of IA is fully abstract, which means that for all M and N : [[M ]] = [[N ]] iff M ≈ N . Γ 2 M : var Γ 2 N : exp Γ 2 M := N : com Results For Idealized Algol we have quite good understanding for which subclasses the equivalence problem is decidable. Remarks For programs of type N → N the semantics will be the input/output relation. The setting scales up to higher-order programs with free variables. Iteration and recursion Γ 2 M : exp Γ 2 N : com Γ 2 while M do N : com f :N → N 2 f (f (5)) : N qo Observational equivalence τ ::= b | τ → τ Rules The semantics is compositional. f : N → N 2 if (f (5) = 6) then 7 else 0: N Γ, x : τ 2 M : τ Γ 2 µx τ .M : τ Games in logic Dagstuhl 2007 45 / 48 Conclusions In E-F games, the semantics of a formula in a structure is represented by a type: a tree describing all potential extensions. In parity games, the semantics of a fixpoint formula is represented as an infinite unfolding of a formula. Winning conditions on infinite plays are needed to handle fixpoints of different types. In game semantics, formula, or program, is represented by a set of plays it admits. This set can be sometimes regular or context-free. Questions Are there connections between these settings? Are there nontrivial results that can be transferred from one setting to another? Igor Walukiewicz (LaBRI) Igor Walukiewicz (LaBRI) Nested calls qo Game semantics Types Igor Walukiewicz (LaBRI) f :N →N 2 f (5) + f (1) :N qo Remarks Different strategies correspond to different programs. Γ 2 M : var Γ 2!M : exp Calling a function twice Games in logic Dagstuhl 2007 48 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 46 / 48 Igor Walukiewicz (LaBRI) Games in logic Dagstuhl 2007 47 / 48