1 / 19

Proofs and Programs

Proofs and Programs. Wei Hu 11/01/2007. Outline. Motivation Theory Lambda calculus Curry-Howard Isomorphism Dependent types Practice Coq. Motivation - Why Learn Coq. Helps understand PL theory better Good for CS615 (sadly, not quals) Coq is becoming popular

christmas
Télécharger la présentation

Proofs and Programs

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Proofs and Programs Wei Hu 11/01/2007

  2. Outline • Motivation • Theory • Lambda calculus • Curry-Howard Isomorphism • Dependent types • Practice • Coq Wei Hu

  3. Motivation - Why Learn Coq • Helps understand PL theory better • Good for CS615 (sadly, not quals) • Coq is becoming popular • People rethinking trusted software • Functional programming is gaining attention • Has been used to • Prove mathematical theorems • Specify hardware • Recently, papers being published for • Operating systems • Compilers Wei Hu

  4. Coq is an Interactive Theorem Prover • Two types of theorem provers • Automated TP • Simplify, CVC, … • Interactive TP(aka, Proof Assistants) • Coq, PVS, ACL2, HOL, Isabelle, Twelf, NuPRL, Agda • Coq’s highlights • Higher-order intuitionistic logic • Higher-order type theory • Embedded functional programming language • Goal transformation through tactics • Mainly works interactively, with limited support of automation to discharge trivial propositions Wei Hu

  5. Coq as a Programming Language • ML-like Fixpoint is_even (n:nat) : bool := match n with | 0 => true | 1 => false | S (S n') => is_even n’ end. Eval compute in is_even 3. Let’s try it out! • Restrictions • No side effects • No non-terminating programs (to avoid inconsistency) Wei Hu

  6. Program Extraction Fixpoint is_even (n:nat) : bool := match n with | 0 => true | 1 => false | S (S n') => is_even n’ end. Coq let recis_even = function | O -> True | S n0 -> (match n0 with | O -> False | S n' -> is_even n') OCaml is_even n = case n of O -> True S n0 -> (case n0 of O -> False S n' -> is_even n') Haskell Wei Hu

  7. Nothing Is Built-in! Inductive bool : Set := | true : bool | false : bool. Inductive nat : Set := | O : nat | S : nat -> nat. Coq (theories/Init/Datatypes.v) type bool = | True | False type nat = | O | S of nat OCaml data Bool = True | False data Nat = O | S Nat Haskell Wei Hu

  8. Simply Typed Lambda Calculus • Type • Base types: nat, bool, … • Function types: nat->nat, nat->bool, … • Term • Variables: x, y, z • Abstractions: x: T1.t2 • Applications: M N • Environment () • Gives typing for variables Wei Hu

  9. Typing Rules x : T   |- t1 : T11 -> T12 , x: T1 |- t2 : T2  |- t2 : T11 (Var) (Abs) (App)  |- x: T1.t2 : T1 -> T2  |- x : T  |- t1t2 : T12 The typing rules work equally well in logic! Wei Hu

  10. Curry-Howard Isomorphism • Propositions = Types • Proofs (inhabit Props) = Programs (inhabit Sets) • Proof tactics = Program constructs • So what? • Proof checking now becomes type checking (can be undecidable if non-termination is allowed) • Programs and proofs are organically unified • Why is it hard to prove something? • Unlike general programming, we are doing type-level programming • In other words, we are constructing programs for a given type • Tactics help with the construction Wei Hu

  11. C-H at Work Section example1. Hypothesis A B : Prop. Hypothesis C : B -> A. Lemma Var : B -> A. exact C. Qed. Lemma Abs : A -> A. intros. exact H. Qed. Lemma App : A -> (A -> B) -> B. intros. apply H0. exact H. Qed. x : T   |- t1 : T11 -> T12 , x: T1 |- t2 : T2  |- t2 : T11  |- t1t2 : T12  |- x : T  |- x: T1.t2 : T1 -> T2 function or implication? Wei Hu

  12. Type-centric View • -> is primitive, others not • theories/Init/Logic.v Inductive True : Prop := I : True. Inductive False : Prop :=. Definition not (A:Prop) := A -> False. Inductive and (A B:Prop) : Prop := conj : A -> B -> A /\ B. Inductive or (A B:Prop) : Prop := | or_introl : A -> A \/ B | or_intror : B -> A \/ B. BHK Interpretation (Brouwer-Heyting-Kolmogorov) Wei Hu

  13. Intuitionistic (Constructive) Logic vs. Classical Logic • A set of equivalent axioms missing in IL • Law of excluded middle: P ∨ ~P • Double negation rule: ~~P → P • Peirce’s Law: ((P→Q)→P)→P • The correspondence between call/cc and Pierce’s Law • See Coq FAQ #30 “What axioms can be safely added to Coq?” • CL takes a denotational view • Assigns to every variable a value (true or false) • Builds truth tables for primitive operations /\ ∨ ~ • (P -> Q) ≡ (~P ∨ Q ) Wei Hu

  14. Polymorphism and Universal Quantification • Parametric polymorphism • You have been using it all the time! • Polymorphic functions • Definition id := fun (X : Set) (x : X) => x. • Check id. id : forall X : Set, X -> X. • Polymorphic data types (e.g., lists) • Universal quantification • Lemma refl : forall P : Prop, P -> P. intros. apply H. Qed. • Print refl. refl = fun (P : Prop) (H : P) => H : forall P : Prop, P -> P Wei Hu

  15. Dependent Types • Type systems in ML-like languages are just not expressive enough • We want to specify propositions like ∀ n:nat, n ≤ n The result type (n ≤ n) depends on arguments (n) • We generalize arrows (A -> B, or λx:A. B) to dependent products (∏x:A. B) where B is dependent on x. • In Coq: • forall n:nat, n<=n • forall mn: nat, m + n = n + m Wei Hu

  16. Dependent Types • Compatibility with non-dependent product • The type of a function that builds a pair: forall (A B:Set) (a:A) (b:B), A*B forall A B : Set, A -> B -> A * B • Defining existential • ∀ x, (P x -> ∃ x, P x) • Dependent sums • Dependent types in programming • Epigram (http://www.e-pig.org/) • Ynot (http://www.eecs.harvard.edu/~greg/ynot.html) • Concoqtion (http://www.metaocaml.org/concoqtion/) • Omega (http://web.cecs.pdx.edu/~sheard/) • ATS (http://www.cs.bu.edu/~hwxi/ATS/) Wei Hu

  17. Polymorphism vs. Dependent Types • Polymorphism • Terms can take types as arguments • Dependent types • Types can take terms as arguments • Inductive types and predicates • Generalization of conventional user-defined data types • We are talking about parameterized data types (e.g. vector n) and propositions (e.g. m <= n) Wei Hu

  18. Conclusions • We talked about type theory • Curry-Howard correspondence is cool • Dependent types matter* • We did not cover common proof techniques • Proof by induction • Use of tactics and tacticals * T. Altenkirch, C. McBride, and J. McKinna, “Why Dependent Types Matter” Wei Hu

  19. How to Learn Coq? • Resources (for beginners) • http://www.cs.virginia.edu/~wh5a/coq.html • Hardest parts (IMO) • Design/Formalization • Combination of tactics • Tools are getting better, but still hard • What’s the next breakthrough? Wei Hu

More Related