module Data.Nat.Divisible where

Divisibility🔗

In the natural numbers, divisibility1 is the property expressing that a given number can be expressed as a multiple of another, and we write aba | b2 when there exists some kk such that b=kab = ka.

Note the use of an existential quantifier, which is a bit annoying: For divisibility to truly be a property, we must work under a truncation, since otherwise there would be N\mathbb{N}-many proofs that 000 | 0, since for any nn, we have 0=n00 = n0. To avoid this sticky situation, we define divisibility with a single step of indirection:

__ : Nat  Nat  Type
zero  ∣ y = y ≡ zero
suc x ∣ y = fibre (_* suc x) y

infix 5 __

In this way, we break the pathological case of 000 | 0 by decreeing it to be the (contractible) type 0=00 = 0. Every other natural number is already handled, because the function “(1+x)(1 + x) *” is injective. With this indirection, we can prove that divisibility is a mere property:

∣-is-prop :  x y  is-prop (x ∣ y)
∣-is-prop zero y n k = prop!
∣-is-prop (suc x) y (n , p) (n' , q) = Σ-prop-path! (*-suc-inj x n n' (p ∙ sym q))

instance
  H-Level-∣ :  {x y} {n}  H-Level (x ∣ y) (suc n)
  H-Level-∣ = prop-instance (∣-is-prop _ _)

The type xyx | y is, in fact, the propositional truncation of (x)1(y)(* x)^{-1}(y) — and it is logically equivalent to that type, too!

∣-is-truncation :  {x y}  (x ∣ y) ≃ ∥ fibre (_* x) y ∥
∣-is-truncation {zero} {y} =
  prop-ext!
     p  inc (y , *-zeror y ∙ sym p))
    (∥-∥-rec! (λ{ (x , p)  sym p ∙ *-zeror x }))
∣-is-truncation {suc x} {y} = Equiv.to is-prop≃equiv∥-∥ (∣-is-prop (suc x) y)

∣→fibre :  {x y}  x ∣ y  fibre (_* x) y
∣→fibre {zero} wit  = 0 , sym wit
∣→fibre {suc x} wit = wit

fibre→∣ :  {x y}  fibre (_* x) y  x ∣ y
fibre→∣ f = Equiv.from ∣-is-truncation (inc f)

divides :  {x y} (q : Nat)  q * x ≡ y  x ∣ y
divides x p = fibre→∣ (x , p)

As an ordering🔗

The notion of divisibility equips the type N\mathbb{N} with yet another partial order, i.e., the relation xyx | y is reflexive, transitive, and antisymmetric. We can establish the former two directly, but antisymmetry will take a bit of working up to:

∣-refl :  {x}  x ∣ x
∣-refl = divides 1 (*-onel _)

∣-trans :  {x y z}  x ∣ y  y ∣ z  x ∣ z
∣-trans {zero} {zero} p q = q
∣-trans {zero} {suc y} p q = absurd (suc≠zero p)
∣-trans {suc x} {zero} p q = 0 , sym q
∣-trans {suc x} {suc y} {z} (k , p) (k' , q) = k' * k , (
  k' * k * suc x   ≡⟨ *-associative k' k (suc x)
  k' * (k * suc x) ≡⟨ ap (k' *_) p ⟩
  k' * suc y       ≡⟨ q ⟩
  z                ∎)

We note in passing that any number divides zero, and one divides every number.

∣-zero :  {x}  x ∣ 0
∣-zero = divides 0 refl

∣-one :  {x}  1 ∣ x
∣-one {x} = divides x (*-oner x)

A more important lemma is that if xx divides a non-zero number yy, then xyx \le y: the divisors of any positive yy are smaller than it. Zero is a sticking point here since, again, any number divides zero!

m∣sn→m≤sn :  {x y}  x ∣ suc y  x ≤ suc y
m∣sn→m≤sn {x} {y} p with ∣→fibre p
... | zero  , p = absurd (zero≠suc p)
... | suc k , p = difference→≤ (k * x) p

This will let us establish the antisymmetry we were looking for:

∣-antisym :  {x y}  x ∣ y  y ∣ x  x ≡ y
∣-antisym {zero}  {y}     p q = sym p
∣-antisym {suc x} {zero}  p q = absurd (suc≠zero q)
∣-antisym {suc x} {suc y} p q = ≤-antisym (m∣sn→m≤sn p) (m∣sn→m≤sn q)

Elementary properties🔗

Since divisibility is the “is-multiple-of” relation, we would certainly expect a number to divide its multiples. Fortunately, this is the case:

∣-*l :  {x y}  x ∣ x * y
∣-*l {y = y} = fibre→∣ (y , *-commutative y _)

∣-*r :  {x y}  x ∣ y * x
∣-*r {y = y} = fibre→∣ (y , refl)

  1. Not to be confused with a proper division algorithm.↩︎

  2. read “a divides b”↩︎