「SF-LC」9 ProofObjects

Posted on

“Algorithms are the computational content of proofs.” —Robert Harper

So the book material is designed to be gradually reveal the facts that

Programming and proving in Coq are two sides of the same coin.

e.g.

  • Inductive is useds for both data types and propositions.
  • -> is used for both type of functions and logical implication.

The fundamental idea of Coq is that:

provability in Coq is represented by concrete evidence. When we construct the proof of a basic proposition, we are actually building a tree of evidence, which can be thought of as a data structure.

e.g.

  • implication like A → B, its proof will be an evidence transformer: a recipe for converting evidence for A into evidence for B.

Proving manipulates evidence, much as programs manipuate data.

Curry-Howard Correspondence

deep connection between the world of logic and the world of computation:

propositions             ~  types
proofs / evidence        ~  terms / data values

ev_0 : even 0

  • ev_0 has type even 0
  • ev_0 is a proof of / is evidence for even 0

ev_SS : ∀n, even n -> even (S (S n))

  • takes a nat n and evidence for even n and yields evidence for even (S (S n)).

This is Props as Types.

Proof Objects

Proofs are data! We can see the proof object that results from this proof script.

1
2
3
4
5
6
Print ev_4.
(* ===> ev_4 = ev_SS 2 (ev_SS 0 ev_0) 
             : even 4  *)

Check (ev_SS 2 (ev_SS 0 ev_0)).     (* concrete derivation tree, we r explicitly say the number tho *)
(* ===> even 4 *)

These two ways are the same in principle!

Proof Scripts

Show Proof. will show the partially constructed proof terms / objects.
?Goal is the unification variable. (the hold we need to fill in to complete the proof)

more complicated in branching cases
one hole more subgoal

1
2
3
4
5
6
7
8
9
10
Theorem ev_4'' : even 4.   (*  match? (even 4) *)
Proof.
  Show Proof.              (*  ?Goal  *)
  apply ev_SS.
  Show Proof.              (*  (ev_SS 2 ?Goal)  *)
  apply ev_SS.
  Show Proof.              (*  (ev_SS 2 (ev_SS 0 ?Goal))  *)
  apply ev_0. 
  Show Proof.              (*  ?Goal (ev_SS 2 (ev_SS 0 ev_0))  *)
Qed.

Tactic proofs are useful and convenient, but they are not essential:
in principle, we can always construct the required evidence by hand

Agda doesn’t have tactics built-in. (but also Interactive)

Quantifiers, Implications, Functions

In Coq’s computational universe (where data structures and programs live), to give ->:

  • constructors (introduced by Indutive)
  • functions

in Coq’s logical universe (where we carry out proofs), to give implication:

  • constructors
  • functions!

So instead of writing proof scripts e.g._

1
2
3
4
5
6
7
Theorem ev_plus4 : ∀n, even n → even (4 + n).
Proof.
  intros n H. simpl.
  apply ev_SS.
  apply ev_SS.
  apply H.
Qed.

we can give proof object, which is a function here, directly!

1
2
3
4
5
6
7
8
9
Definition ev_plus4' : ∀n, even n → even (4 + n) :=    (* ∀ is syntax for Pi? *)
  fun (n : nat)    ⇒ 
  fun (H : even n) ⇒
    ev_SS (S (S n)) (ev_SS n H).


Definition ev_plus4'' (n : nat) (H : even n)           (* tricky: implicitly `Pi` when `n` get mentioned?  *)
                    : even (4 + n) :=
  ev_SS (S (S n)) (ev_SS n H).

two interesting facts:

  1. intros x corresponds to λx. (or Pi x.??)
  2. apply corresponds to…not quite function application… but more like filling the hole.
  3. even n mentions the value of 1st argument n. i.e. dependent type!

Recall Ari’s question in “applying theorem as function” e.g. plus_comm
why we can apply value in type-level fun.
becuz of dependent type.

Now we call them dependent type function

→ is degenerated ∀ (Pi)

Notice that both implication (→) and quantification (∀) correspond to functions on evidence.
In fact, they are really the same thing: → is just a shorthand for a degenerate use of ∀ where there is no dependency, i.e., no need to give a name to the type on the left-hand side of the arrow:

1
2
3
4
5
6
7
  ∀(x:nat), nat 
= ∀(_:nat), nat 
= nat → nat

  ∀n, ∀(E : even n), even (n + 2).
= ∀n, ∀(_ : even n), even (n + 2).
= ∀n, even n → even (n + 2).

In general, P → Q is just syntactic sugar for ∀ (_:P), Q.

TaPL also mention this fact for Pi.

Q&A - Slide 15

  1. ∀ n, even n → even (4 + n). (2 + n = S (S n))

Programming with Tactics.

If we can build proofs by giving explicit terms rather than executing tactic scripts,
you may be wondering whether we can build programs using tactics? Yes!

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Definition add1 : nat → nat.
  intro n.
  Show Proof.      
(** 
the goal (proof state):
    
    n : nat
    =======
    nat
    
the response:

    (fun n : nat => ?Goal) 
    
What is really interesting here, is that the premies [n:nat] is actually the arguments!
again, the process of applying tactics is _partial application_
**)

  apply S.
  Show Proof.      
(** 
    (fun n : nat => S ?Goal) 
**)
  apply n. 
Defined.

Print add1.
(* ==> add1 = fun n : nat => S n
            : nat -> nat *)

Notice that we terminate the Definition with a . rather than with := followed by a term.
This tells Coq to enter proof scripting mode (w/o Proof., which did nothing)

Also, we terminate the proof with Defined rather than Qed; this makes the definition transparent so that it can be used in computation like a normally-defined function
(Qed-defined objects are opaque during computation.).

Qed make things unfoldable,
thus add 1 ends with Qed is not computable…
(becuz of not even unfoldable thus computation engine won’t deal with it)

Prof.Mtf: meaning “we don’t care about the details of Proof”

see as well Smart Constructor

This feature is mainly useful for writing functions with dependent types

In Coq - you do as much as ML/Haskell when you can…?
Unlike Agda - you program intensively in dependent type…?

When Extracting to OCaml…Coq did a lot of Object.magic for coercion to bypass OCaml type system. (Coq has maken sure the type safety.)

Logical Connectives as Inductive Types

Inductive definitions are powerful enough to express most of the connectives we have seen so far.
Indeed, only universal quantification (with implication as a special case) is built into Coq;
all the others are defined inductively.
Wow…

CoqI: What’s Coq logic? Forall + Inductive type (+ coinduction), that’s it.

Conjunctions

1
2
3
4
5
6
7
Inductive and (P Q : Prop) : Prop :=
| conj : P → Q → and P Q.

Print prod.
(* ===>
   Inductive prod (X Y : Type) : Type :=
   | pair : X -> Y -> X * Y. *)

similar to prod (product) type… more connections happening here.

This similarity should clarify why destruct and intros patterns can be used on a conjunctive hypothesis.

Similarly, the split tactic actually works for any inductively defined proposition with exactly one constructor
(so here, apply conj, which will match the conclusion and generate two subgoal from assumptions )

A very direct proof:

1
2
3
4
Definition and_comm'_aux P Q (H : P ∧ Q) : Q ∧ P :=
  match H with
  | conj HP HQ ⇒ conj HQ HP
  end.

Disjunction

1
2
3
Inductive or (P Q : Prop) : Prop :=
| or_introl : P → or P Q
| or_intror : Q → or P Q.

this explains why destruct works but split not..

Q&A - Slide 22 + 24

Both Question asked about what’s the type of some expression

1
2
3
4
5
6
7
8
9
10
fun P Q R (H1: and P Q) (H2: and Q R) ⇒
    match (H1,H2) with
    | (conj _ _ HP _, conj _ _ _ HR) ⇒ conj P R HP HR
    end.

fun P Q H ⇒
    match H with
    | or_introl HP ⇒ or_intror Q P HP
    | or_intror HQ ⇒ or_introl Q P HQ
    end.

But if you simply Check on them, you will get errors saying:
Error: The constructor conj (in type and) expects 2 arguments. or
Error: The constructor or_introl (in type or) expects 2 arguments..

Coq Magics, “Implicit” Implicit and Overloading??

So what’s the problem?
Well, Coq did some magics…

1
2
3
4
Print and.
(* ===> *)
Inductive and (A B : Prop) : Prop :=  conj : A -> B -> A /\ B
For conj: Arguments A, B are implicit

constructor conj has implicit type arg w/o using {} in and …

1
2
3
4
5
6
7
8
9
10
11
Inductive or (A B : Prop) : Prop :=
    or_introl : A -> A \/ B | or_intror : B -> A \/ B

For or_introl, when applied to no more than 1 argument:
  Arguments A, B are implicit
For or_introl, when applied to 2 arguments:
  Argument A is implicit
For or_intror, when applied to no more than 1 argument:
  Arguments A, B are implicit
For or_intror, when applied to 2 arguments:
  Argument B is implicit

this is even more bizarre…
constructor or_introl (and or_intror) are overloaded!! (WTF)

And the questions’re still given as if they’re inside the modules we defined our plain version of and & or (w/o any magics), thus we need _ in the positions we instantiate and & or so Coq will infer.

Existential Quantification

To give evidence for an existential quantifier, we package a witness x together with a proof that x satisfies the property P:

1
2
3
4
5
6
7
Inductive ex {A : Type} (P : A → Prop) : Prop :=
| ex_intro : ∀x : A, P x → ex P.

Check ex.                    (* ===> *) : (?A -> Prop) -> Prop 
Check even.                  (* ===> *) : nat -> Prop  (* ?A := nat  *)
Check ex even.               (* ===> *) : Prop 
Check ex (fun n => even n)   (* ===> *) : Prop     (* same *)

one interesting fact is, outside of our module, the built-in Coq behaves differently (magically):

1
2
3
Check ev.                    (* ===> *) : ∀ (A : Type), (A -> Prop) -> Prop
Check even.                  (* ===> *) : nat -> Prop  (* A := nat  *)
Check ex (fun n => even n)   (* ===> *) : ∃ (n : nat) , even n : Prop  (* WAT !? *)

A example of explicit proof object (that inhabit this type):

1
2
Definition some_nat_is_even : ∃n, even n :=
  ex_intro even 4 (ev_SS 2 (ev_SS 0 ev_0)).

the ex_intro take even first then 4…not sure why the order becomes this… 

1
Check (ex_intro).            (* ===> *) : forall (P : ?A -> Prop) (x : ?A), P x -> ex P

To prove ex P, given a witness x and a proof of P x. This desugar to ∃ x, P x

  • the P here, is getting applied when we define prop ∃ x, P x.
  • but the x is not mentioned in type constructor…so it’s a existential type.
    • I don’t know why languages (including Haskell) use forall for existential tho.

exists tactic = applying ex_intro

True and False

1
2
3
4
5
Inductive True : Prop :=
  | I : True.

(* with 0 constructors, no way of presenting evidence for False *)
Inductive False : Prop := .

Equality

1
2
3
4
5
6
Inductive eq {X:Type} : X → X → Prop :=
| eq_refl : ∀x, eq x x.

Notation "x == y" := (eq x y)
                    (at level 70, no associativity)
                    : type_scope.

given a set X, it defines a family of propositions “x is equal to y,”, indexed by pairs of values (x and y) from X.

Can we also use it to construct evidence that 1 + 1 = 2?
Yes, we can. Indeed, it is the very same piece of evidence!

The reason is that Coq treats as “the same” any two terms that are convertible according to a simple set of computation rules.

nothing in the unification engine but we relies on the reduction engine.

Q: how much is it willing to do?
Mtf: just run them! (since Coq is total!)

1
2
3
4
Lemma four: 2 + 2 == 1 + 3.
Proof.
  apply eq_refl.
Qed.

The reflexivity tactic is essentially just shorthand for apply eq_refl.

Slide Q & A

  • (4) has to be applicable thing, i.e. lambda, or “property” in the notion!

In terms of provability of reflexivity

1
2
(fun n => S (S n)) = (fun n => 2 + n)          (* reflexivity *)
(fun n => S (S n)) = (fun n => n + 2)          (* rewrite add_com *)

Inversion, Again

We’ve seen inversion used with both equality hypotheses and hypotheses about inductively defined propositions. Now that we’ve seen that these are actually the same thing

In general, the inversion tactic…

  1. take hypo H whose type P is inductively defined
  2. for each constructor C in P
    1. generate new subgoal (assume H was built with C)
    2. add the arguments (i.e. evidences of premises) of C as extra hypo (to the context of subgoal)
    3. (apply constructor theorem), match the conclusion of C, calculates a set of equalities (some extra restrictions)
    4. adds these equalities
    5. if there is contradiction, discriminate, solve subgoal.

Q

Q: Can we write + in a communitive way?
A: I don’t believe so.

Ground truth

  • provided by direct observation (instead of inference)

Ground term

  • that does not contain any free variables.

Groundness

  • 根基性?

Weird Axiomness might break the soundness of generated code in OCaml…