(* * We need some basic logic, so I'm stealing this from . See that file * for more details on how these work. *) false : * := forall (A : *), A; false_elim (A : *) (contra : false) : A := contra A; not (A : *) : * := A -> false; not_intro (A : *) (h : A -> false) : not A := h; not_elim (A B : *) (a : A) (na : not A) : B := na a B; and (A B : *) : * := forall (C : *), (A -> B -> C) -> C; and_intro (A B : *) (a : A) (b : B) : and A B := fun (C : *) (H : A -> B -> C) => H a b; and_elim_l (A B : *) (ab : and A B) : A := ab A (fun (a : A) (b : B) => a); and_elim_r (A B : *) (ab : and A B) : B := ab B (fun (a : A) (b : B) => b); or (A B : *) : * := forall (C : *), (A -> C) -> (B -> C) -> C; or_intro_l (A B : *) (a : A) : or A B := fun (C : *) (ha : A -> C) (hb : B -> C) => ha a; or_intro_r (A B : *) (b : B) : or A B := fun (C : *) (ha : A -> C) (hb : B -> C) => hb b; or_elim (A B C : *) (ab : or A B) (ha : A -> C) (hb : B -> C) : C := ab C ha hb; exists (A : *) (P : A -> *) : * := forall (C : *), (forall (x : A), P x -> C) -> C; exists_intro (A : *) (P : A -> *) (a : A) (h : P a) : exists A P := fun (C : *) (g : forall (x : A), P x -> C) => g a h; exists_elim (A B : *) (P : A -> *) (ex_a : exists A P) (h : forall (a : A), P a -> B) : B := ex_a B h; exists_elim_val (A : *) (P : A -> *) (ex_a : exists A P) : A := exists_elim A A P ex_a (fun (a : A) (_ : P a) => a); binop (A : *) := A -> A -> A; -- -------------------------------------------------------------------------------------------------------------- (* * Next we can define equality. This is the same as in . We get a * couple Peano axioms for free as theorems. *) -- implies axiom 5 -- (if `x : nat`, then `y : nat`, since we can only compare objects of the same type) eq (A : *) (x y : A) := forall (P : A -> *), P x -> P y; -- axiom 2 (but as a theorem) eq_refl (A : *) (x : A) : eq A x x := fun (P : A -> *) (Hx : P x) => Hx; -- axiom 3 (but as a theorem) eq_sym (A : *) (x y : A) (Hxy : eq A x y) : eq A y x := fun (P : A -> *) (Hy : P y) => Hxy (fun (z : A) => P z -> P x) (fun (Hx : P x) => Hx) Hy; -- axiom 4 (but as a theorem) eq_trans (A : *) (x y z : A) (Hxy : eq A x y) (Hyz : eq A y z) : eq A x z := fun (P : A -> *) (Hx : P x) => Hyz P (Hxy P Hx); -- This isn't an axiom, but is handy. If `x = y`, then `f x = f y`. eq_cong (A B : *) (x y : A) (f : A -> B) (H : eq A x y) : eq B (f x) (f y) := fun (P : B -> *) (Hfx : P (f x)) => H (fun (a : A) => P (f a)) Hfx; -- -------------------------------------------------------------------------------------------------------------- (* * Now we can define the Peano axioms. Unlike with equality, perga is not * powerful enough to construct the natural numbers (or at least to prove the * Peano axioms as theorems from a definition constructible in perga). However, * working with axioms is extremely common in math. As such, perga has a system * for doing just that, namely the *axiom* system. * * In a definition, rather than give a value for the term, we can give it the * value `axiom`, in which case a type ascription is mandatory. Perga will then * trust our type ascription, and assume going forward that the identifier we * defined is a value of the asserted type. For example, we will use the axiom * system to assert the existence of a type of natural numbers *) nat : * := axiom; (* * As you can imagine, this can be risky. For instance, there's nothing stopping * us from saying * uh_oh : false := axiom; * or stipulating more subtly contradictory axioms. As such, as in mathematics, * axioms should be used with care. * * There's another problem with axioms, namely that perga cannot do any * computations with axioms. The more you can define within perga natively, the * better, as computations done without axioms can be utilized by perga. For * example, in , we define the natural numbers as Church * numerals entirely within perga. There, the proof that `1 + 1 = 2` is just * `eq_refl`, since they reduce to the same thing. Here, `1 + 1 = 2` will * require a proof, since perga is unable to do computations with things defined * as axioms. * * With these warnings in place, the Peano axioms are proven to be consistent, * so we should be fine. I'm formalizing the second order axioms given in the * wikipedia article on the Peano axioms * (https://en.wikipedia.org/wiki/Peano_axioms). *) -- axiom 1: 0 is a natural number zero : nat := axiom; -- axiom 6: For every `n`, `S n` is a natural number. suc (n : nat) : nat := axiom; -- axiom 7: If `S n = S m`, then `n = m`. suc_inj : forall (n m : nat), eq nat (suc n) (suc m) -> eq nat n m := axiom; -- axiom 8: No successor of any natural number is zero. suc_nonzero : forall (n : nat), not (eq nat (suc n) zero) := axiom; -- axiom 9: Induction! For any proposition φ on natural numbers, if φ(0) holds, -- and if for every natural number n, φ(n) ⇒ φ(S n), then φ holds for all n. nat_ind : forall (φ : nat -> *), φ zero -> (forall (n : nat), φ n -> φ (suc n)) -> forall (n : nat), φ n := axiom; -- -------------------------------------------------------------------------------------------------------------- (* * Now that we have stipulated these axioms, we are free to use them to make * definitions, prove theorems, etc. * * Our first theorem, as a warm up, is to prove that every natural number is * either 0 or the successor of another natural number. * * First, we will make a bunch of abbreviations, since these terms get really * long and complicated really quickly. *) -- First, the predecessor of `n` is `m` if `n = suc m`. pred (n m : nat) : * := eq nat n (suc m); -- Our claim is a disjunction, whose first option is that `n = 0`. szc_l (n : nat) := eq nat n zero; -- The second option is that `n` has a predecessor. szc_r (n : nat) := exists nat (fun (m : nat) => pred n m); -- So the claim we are trying to prove is that either one of the above options -- holds for every `n`. szc (n : nat) := or (szc_l n) (szc_r n); -- And here's our proof! suc_or_zero : forall (n : nat), szc n := -- We will prove this by induction. nat_ind szc -- For the base case, the first option holds, i.e. 0 = 0 (or_intro_l (szc_l zero) (szc_r zero) (eq_refl nat zero)) -- For the inductive case, suppose the theorem holds for `n`. (fun (n : nat) (_ : szc n) => -- Then the right option holds for `suc n`, since `suc n` is the -- successor of `n` or_intro_r (szc_l (suc n)) (szc_r (suc n)) (exists_intro nat (pred (suc n)) n (eq_refl nat (suc n)))); (* * Now we would like to define addition and prove lots of lovely theorems about * addition, but first we need a mechanism for even defining addition in the * first place. The way we will go about this is proving that we can define * functions recursively. In particular, we will prove that for any `A : *`, any * `fzero : A`, and any `fsuc : nat -> A -> A`, there is a unique function * `f : nat -> A` satisfying the following equations: * (1) f zero = fzero * (2) f (suc x) = fsuc x (f x) *) -- Here is equation (1) rec_cond_zero (A : *) (fzero : A) (fsuc : nat -> A -> A) (f : nat -> A) := eq A (f zero) fzero; -- And here's equation (2) rec_cond_suc (A : *) (fzero : A) (fsuc : nat -> A -> A) (f : nat -> A) := forall (x : nat) (y : A), eq A (f x) y -> eq A (f (suc x)) (fsuc x y); (* * Our proof strategy is to first define a relation `R` on `nat` such that * `R x y` only holds if `f x = y` for every function `f : nat -> A` satisfying * equations (1) and (2). Then we will prove that `R` is actually total, and use * that to define our recursive function. Then we will show that all function * satisfying equations (1) and (2) have the same outputs, and that this * function actually satisfies the equations, meaning it is uniquely defined. *) -- Here's the relation. rec_rel (A : *) (fzero : A) (fsuc : nat -> A -> A) (n : nat) (x : A) : * := forall (f : nat -> A), rec_cond_zero A fzero fsuc f -> rec_cond_suc A fzero fsuc f -> eq A (f n) x; -- Little lemma showing that `R zero fzero`. rec_rel_zero (A : *) (fzero : A) (fsuc : nat -> A -> A) : rec_rel A fzero fsuc zero fzero := fun (f : nat -> A) (Hzero : rec_cond_zero A fzero fsuc f) (Hsuc : rec_cond_suc A fzero fsuc f) => Hzero; -- Likewise, we show that if `R x y`, then `R (suc x) (fsuc x y)`. rec_rel_suc (A : *) (fzero : A) (fsuc : nat -> A -> A) (x : nat) (y : A) (H : rec_rel A fzero fsuc x y) : rec_rel A fzero fsuc (suc x) (fsuc x y) := fun (f : nat -> A) (Hzero : rec_cond_zero A fzero fsuc f) (Hsuc : rec_cond_suc A fzero fsuc f) => Hsuc x y (H f Hzero Hsuc); -- This is an abbreviation for the statement that `R` is defined on `x`. rec_rel_def (A : *) (fzero : A) (fsuc : nat -> A -> A) : nat -> * := fun (x : nat) => exists A (rec_rel A fzero fsuc x); -- Finally, we prove that `R` is total. rec_rel_total (A : *) (fzero : A) (fsuc : nat -> A -> A) : forall (x : nat), rec_rel_def A fzero fsuc x := -- We prove this by induction. nat_ind (rec_rel_def A fzero fsuc) -- For our base case, since `R zero fzero` by the earlier lemma, `R` is defined on `zero`. (exists_intro A (rec_rel A fzero fsuc zero) fzero (rec_rel_zero A fzero fsuc)) -- For the inductive case, let `x : nat` and suppose `R` is defined on `x`. (fun (x : nat) (IH : rec_rel_def A fzero fsuc x) => -- Then let `y` be such that `R x y`. exists_elim A (exists A (rec_rel A fzero fsuc (suc x))) (rec_rel A fzero fsuc x) IH (fun (y : A) (H : rec_rel A fzero fsuc x y) => -- Then we claim `R (suc x) (fsuc x y)`. exists_intro A (rec_rel A fzero fsuc (suc x)) (fsuc x y) -- Indeed this follows by the earlier lemma. (rec_rel_suc A fzero fsuc x y H))); -- With this, we can define a function `nat -> A`. rec_def (A : *) (fzero : A) (fsuc : nat -> A -> A) (x : nat) : A := exists_elim A A (rec_rel A fzero fsuc x) (rec_rel_total A fzero fsuc x) (fun (y : A) (_ : rec_rel A fzero fsuc x y) => y); -- Now we can prove that for any two functions `f g : nat -> A` satisfying -- equations (1) and (2), we must have that `f x = g x` for all `x : nat`. rec_def_unique (A : *) (fzero : A) (fsuc : nat -> A -> A) (f g : nat -> A) (Fzero : rec_cond_zero A fzero fsuc f) (Fsuc : rec_cond_suc A fzero fsuc f) (Gzero : rec_cond_zero A fzero fsuc g) (Gsuc : rec_cond_suc A fzero fsuc g) : forall (x : nat), eq A (f x) (g x) := -- We prove this by induction. nat_ind (fun (x : nat) => eq A (f x) (g x)) -- For the base case, we have -- f zero = fzero (by (1) for f) -- = g zero (by (1) for g) (eq_trans A (f zero) fzero (g zero) Fzero (eq_sym A (g zero) fzero Gzero)) -- Now suppose `f x = g x`. We want to show that `f (suc x) = g (suc x)`. (fun (x : nat) (IH : eq A (f x) (g x)) => -- Well, we have -- f (suc x) = fsuc x (f x) (by (2) for f) -- = fsuc x (g x) (by the inductive hypothesis) -- = g (suc x) (by (2) for g) (eq_trans A (f (suc x)) (fsuc x (f x)) (g (suc x)) (Fsuc x (f x) (eq_refl A (f x))) (eq_trans A (fsuc x (f x)) (fsuc x (g x)) (g (suc x)) (eq_cong A A (f x) (g x) (fun (y : A) => fsuc x y) IH) (eq_sym A (g (suc x)) (fsuc x (g x)) (Gsuc x (g x) (eq_refl A (g x))))))); -- Almost there! We just need to show that `rec_def` actually satisfies the -- equations.