read

I have been of a couple of days over the Christmas and New Year. But now it is time to start my 100 days of Fibonacci project again.

Today I chose Coq and I decided to look at proving properties about programs. The key idea in this post is to prove that two different implementations of the Fibonacci function indeed are equivalent. Concretely I have chosen the direct recursive and the accumulated recursive implementations as the subjects.

These implementations are the same as in the first article. This is done deliberately as they are idiomatic to functional programming. Furthermore, they showcase quite well why this has a value. The directly recursive implementation is easy to understand and stays close to both the definition of Fibonacci and the common understanding of the function. On the other hand, the recursive function with the accumulator is harder to understand but provides a significant speedup.

In this article, I show that the rest of the program can be indifferent to which implementation is in use. This is done by proving theorems, which is considerably more time consuming that providing tests. The effort returns in favor the strongest guarantee we can have for the property, a mathematical proof.

Day 7 - Coq!

As mentioned the two implementations are the directly recursive and the recursive with accumulation. These are both implementations we have seen before. But we use them again as they provide a good body for showcasing proving program equivalence. Next we have the two implementations.

Fixpoint fib_direct (n : nat) : nat :=
  match n with
    | 0    => 0
    | S n' => match n' with
                | 0 => 1
                | S n'' => fib_direct n' + fib_direct n''
              end
  end.

Above is the directly recursive implementation and next is the accumulated recursive implementation.

Fixpoint fib_accumulator (n : nat) (a : nat) (b : nat) : nat :=
  match n with
    | 0      => b
    | (S n') => fib_accumulator n' (a+b) a
  end.
(* An alias to scrape away unnecessary information *)
Definition fib_acc (n : nat) := fib_accumulator n 1 0.

The results can be calculated by the Compute command. It is hereafter visible in the *goals* window (Assuming you use the proof assistant).

Compute fib_direct 10.
Compute fib_acc 10.

A couple of things look a bit different. The function signature looks kind of weird. Furthermore, we are doing some kind of match stuff with something that looks a bit like numbers.

In the function signature, we use the Fixpoint to denote a recursive function. This is because Coq encodes recursion in its type and implement what’s called iso-recursion.

For integers, Coq does not use the regular atomic data types we know from ex. Java. Instead, it has (and only has) inductive data types. In these Peano naturals has been implemented. This means that zero is represented as O, one as S O, two as S (S O) and so forth (this does not implement integers but only naturals which is just fine for Fibonacci).

Coq automatically expands syntactic decimal numbers to its internal representation. Hence 5 is just syntactic sugar for S S S S S O. This is the same representation I implemented in Prolog earlier on.

As usual, the code is available on Github.

It is necessary to be able to reduce data to atomic pieces to do the proving we want to do. This is to create a total dependency between the type and its value. Imagine Java’s int type. This type holds, at least, different values without the compiler knowing which. In that case, it is not possible to reason about the return value of a function, which we need for the next section.

Coq - proof assistant

Coq is not meant as a general-purpose programming language. In fact, it is a proof assistant. This means that it is not possible to interact with the surrounding world. It is not possible to read command line arguments or invoke system calls. On the other hand, Coq can prove properties.

In the above we had two different implementations of the Fibonacci function. One might wonder whether these functions always return the same value. That is, does fib_direct 10 and fib_acc 10 yield the same result? This can easily be checked by running the code. The problem is when we want to check for all values of n. Is it true that fib_direct n = fib_acc n? This we can actually prove in Coq, and so I have done.

The first step I took to prove equivalence of the implementations was to characterize the Fibonacci function.

Definition specification_of_fibonacci (f : nat -> nat) :=
  f 0 = 0
  /\
  f 1 = 1
  /\
  forall n : nat,
    f (S (S n)) = f (S n) + f n.

This is, in fact, a proposition. Is is true if we have a function that behaves as the specification. This specification is very close to the directly recursive implementation.

The next thing is to prove that this definition is unambiguous. This means that f will behave identical independent to the actual implementation. In Coq this looks like following (The whole code is available on Github).

Lemma there_is_only_one_fib :
  forall f1 f2 : nat -> nat,
    specification_of_fibonacci f1 ->
    specification_of_fibonacci f2 ->
    forall n : nat,
      f1 n = f2 n.
Proof.
  unfold specification_of_fibonacci.
  intros f1 f2.
  intros [ bc_f1_0 [ bc_f1_1 ic_f1 ] ]. (* Destruct the conjunctive clauses *)
  intros [ bc_f2_0 [ bc_f2_1 ic_f2 ] ].
   
  (* strengthening the induction hypothesis *)
  assert (H_fib : forall m : nat,
                    f1 m = f2 m /\ f1 (S m) = f2 (S m)).
  intro m.
  ...
  apply hf.
Qed.

What happens above is that we specify a lemma. Given two functions, that both satisfies the specification of the Fibonacci function, they yield the same result for all values of n.

After the specification of the lemma, we provide an actual proof. this is a series of commands that generates a function which in turn satisfies the type of the lemma. This process is somewhat out of scope, and I will look into this in another post.

After this, all we need to do is to prove that both the implementations indeed satisfies the specification. Hereafter we can use the property that there is only one Fibonacci function to prove the equivalence.

Lemma fib_direct_satisfies_specification :
  specification_of_fibonacci fib_direct.
Proof.
...
Qed.

Lemma fib_acc_satisfies_specification :
  specification_of_fibonacci fib_acc.
Proof.
...
Qed.

And the proof that they are equivalent

Theorem fib_direct_is_equivalent_to_fib_acc :
  forall n: nat,
    fib_acc n = fib_direct n.
Proof.
  intros n.
  rewrite (there_is_only_one_fib fib_acc 
                                 fib_direct 
                                 fib_acc_satisfies_specification
                                 fib_direct_satisfies_specification).
  reflexivity.
Qed.

The signature of there_is_only_one_fib was given above. We specialize it with the arguments so we can rewrite our goal. The goal is then rewritten with forall n, fib_acc n = fib_direct n. This yields a goal which is trivially true. Namely fib_direct n = fib_direct n.

We are now sure that for any n we can provide to the directly recursive implementation of Fibonacci, the other implementation yields the same result.

Conclusion

In this post, I implemented the Fibonacci function in the same way as in the first article. These are idiomatic to functional programming and as such it makes sense to keep using them here.

They key concept introduced here is the notion of proving propositions. I proved that the two implementations indeed behave identically. We have scratched the surface when it comes to the underlying theory. But already saw the fruits of the efforts: The direct implementation is slow but easy to understand. The other implementation is faster but slightly more complicated. We have used this to prove that they always return the same value for the same input.

Blog Logo

Mads Buch


Published

Image

Mads Buch

Technical articles

Back to Overview