module Data.Int.Universal where

Universal property of the integers🔗

We define and prove a type-theoretic universal property of the integers, which characterises them as the initial pointed set equipped with an auto-equivalence, in much the same way that the natural numbers are characterised as being the initial pointed type equipped with an endomorphism.

record Integers (ℤ : Type) : Typeω where
  no-eta-equality
  field
    ℤ-is-set : is-set ℤ
    point  : ℤ
    rotate : ℤ ≃ ℤ

We start by giving $Z$ a mapping-out property: If $X$ is any other type with $p : X$ and $e : X ≃ X,$ then there is a map $Z \to X$ which sends our point (which may as well be called zero) to $p,$ and commutes with our equivalence. Note that commuting with the equivalence implies commuting with its inverse, too.

    map-out       : ∀ {ℓ} {X : Type ℓ} → X → X ≃ X → ℤ → X
    map-out-point : ∀ {ℓ} {X : Type ℓ} (p : X) (r : X ≃ X) → map-out p r point ≡ p
    map-out-rotate
      : ∀ {ℓ} {X : Type ℓ} (p : X) (r : X ≃ X) (i : ℤ)
      → map-out p r (rotate .fst i) ≡ r .fst (map-out p r i)

We obtain a sort of initiality by requiring that map-out be unique among functions with these properties.

    map-out-unique
      : ∀ {ℓ} {X : Type ℓ} (f : ℤ → X) {p : X} {r : X ≃ X}
      → f point ≡ p
      → (∀ x → f (rotate .fst x) ≡ r .fst (f x))
      → ∀ x → f x ≡ map-out p r x

By a standard categorical argument, existence and uniqueness together give us an induction principle for the integers: to construct a section of a type family $P : Z \to Type,$ it is enough to give an element of $P (0)$ and a family of equivalences $P (n) ≃ P (n + 1) .$

  ℤ-η : ∀ z → map-out point rotate z ≡ z
  ℤ-η z = sym (map-out-unique id refl (λ _ → refl) z)

  induction
    : ∀ {ℓ} {P : ℤ → Type ℓ}
    → P point
    → (∀ z → P z ≃ P (rotate .fst z))
    → ∀ z → P z
  induction {P = P} pp pr = section where
    tot : ℤ → Σ ℤ P
    tot = map-out (point , pp) (Σ-ap rotate pr)

    is-section : ∀ z → tot z .fst ≡ map-out point rotate z
    is-section = map-out-unique (fst ∘ tot)
      (ap fst (map-out-point _ _))
      (λ z → ap fst (map-out-rotate _ _ z))

    section : ∀ z → P z
    section z = subst P (is-section z ∙ ℤ-η z) (tot z .snd)

We now prove that the integers are a set of integers. This isn’t as tautological as it sounds, sadly, because our integers were designed to be convenient for algebra, and this specific universal property is rather divorced from algebra. Fortunately, it’s still not too hard, so join me.

open Integers

HIT-Int-integers : Integers HIT.Int
HIT-Int-integers = r where
  module map-out {ℓ} {X : Type ℓ} (l : X ≃ X) where

We start by making a simple observation: Exponentiation commutes with difference, where by exponentiation we mean iterated composition of equivalences. That is: if $n$ is an integer expressed as a formal difference of naturals $x - y,$ then we can compute the power $f^{n} : X ≃ X$ as the difference of equivalences $(f^{y})^{- 1} \circ f^{x} .$

    n-power : Nat → X ≃ X
    n-power zero    = (λ x → x) , id-equiv
    n-power (suc x) = n-power x ∙e l

    private
      lemma : ∀ m n x
        → (n-power n e⁻¹) .fst (n-power m .fst x)
        ≡ (n-power n e⁻¹) .fst (Equiv.from (l) (l .fst (n-power m .fst x)))
      lemma m n x = ap ((n-power n e⁻¹) .fst) (sym (Equiv.η l _))

    go : HIT.Int → X ≃ X
    go (HIT.diff x y) = n-power x ∙e (n-power y e⁻¹)
    go (HIT.quot m n i) = Σ-prop-path!
      {x = n-power m ∙e (n-power n e⁻¹)}
      {y = n-power (suc m) ∙e (n-power (suc n) e⁻¹)}
      (funext (lemma m n)) i

To show that this computation respects the quotient, we must calculate that $(f^{y})^{- 1} \circ f^{- 1} f \circ f^{x}$ is $(f^{y})^{- 1} \circ f^{x},$ which follows almost immediately from the properties of equivalences, cancelling the $f^{- 1} f$ critical pair in the middle.

  r : Integers HIT.Int
  r .ℤ-is-set = hlevel 2
  r .point = 0
  r .rotate = HIT.sucℤ , HIT.sucℤ-is-equiv
  r .map-out point rot int = map-out.go rot int .fst point
  r .map-out-point p _ = refl
  r .map-out-rotate p rot i = map-out.map-suc rot i _

Using elimination by sign, we can divide the proof of uniqueness to the case where $n$ is a positive natural number, where $n$ is a negated natural number, and when $n$ is zero. The case $n = 0$ is one of the assumptions, the other cases follow by induction (on naturals).

  r .map-out-unique {X = X} f {point} {rot} path htpy =
    HIT.Int-elim-by-sign (λ z → f z ≡ r .map-out _ _ z) unique-pos unique-neg path
    where abstract
      unique-pos : ∀ k → f (HIT.diff k 0) ≡ map-out.n-power rot k .fst point
      unique-pos zero = path
      unique-pos (suc k) = htpy (HIT.diff k 0) ∙ ap (rot .fst) (unique-pos k)

      unique-neg : ∀ k → f (HIT.diff 0 k) ≡ Equiv.from (map-out.n-power rot k) point
      unique-neg zero = path
      unique-neg (suc k) =
           sym (Equiv.η rot _)
        ·· ap (Equiv.from rot) (
               sym (htpy (HIT.diff 0 (suc k)))
            ·· ap f (sym (HIT.quot 0 k))
            ·· unique-neg k)
        ·· sym (map-out.negatives⁻¹ rot k _)

Inductive integers are integers🔗

In the 1Lab, we have another implementation of the integers, in addition to the ones defined by quotient, which we have already characterised as satisfying the universal property, above. These are the inductive integers: defined as a particular binary coproduct of natural numbers. To avoid the problem of having “two zeroes”, one of the summands is tagged “negative successor,” rather than “successor”, so that negsuc 0 indicates the number $- 1 .$

We have already proven that the inductive integers have a successor equivalence: What we now do is prove this equivalence is universal.

Int-integers : Integers Ind.Int
Int-integers = r where
  module map-out {ℓ} {X : Type ℓ} (l : X ≃ X) where
    pos : Nat → X ≃ X
    pos zero    = _ , id-equiv
    pos (suc x) = pos x ∙e l

    neg : Nat → X ≃ X
    neg zero    = l e⁻¹
    neg (suc x) = neg x ∙e (l e⁻¹)

    to : Ind.Int → X ≃ X
    to (Ind.pos x) = pos x
    to (Ind.negsuc x) = neg x

  r : Integers Ind.Int
  r .ℤ-is-set = Discrete→is-set Ind.Discrete-Int
  r .point = Ind.pos 0
  r .rotate = Ind.suc-equiv
  r .map-out p e i = map-out.to e i .fst p
  r .map-out-point p _ = refl
  r .map-out-rotate p e = go where
    go : ∀ x → r .map-out p e (r .rotate .fst x)
             ≡ e .fst (r .map-out p e x)
    go (Ind.pos x)          = refl
    go (Ind.negsuc zero)    = sym (Equiv.ε e _)
    go (Ind.negsuc (suc x)) = sym (Equiv.ε e _)
  r .map-out-unique f {p} {rot} fz fr = go where
    pos : ∀ n → f (Ind.pos n) ≡ map-out.pos rot n .fst p
    pos zero = fz
    pos (suc n) = fr (Ind.pos n) ∙ ap (rot .fst) (pos n)

    map-pred : ∀ n → f (Ind.predℤ n) ≡ Equiv.from rot (f n)
    map-pred n = sym (Equiv.η rot _)
              ·· ap (Equiv.from rot) (sym (fr _))
              ·· ap (Equiv.from rot ∘ f) (Ind.suc-predℤ n)

    neg : ∀ n → f (Ind.negsuc n) ≡ map-out.neg rot n .fst p
    neg zero = map-pred (Ind.pos 0) ∙ ap (Equiv.from rot) fz
    neg (suc n) = map-pred (Ind.negsuc n) ∙ ap (Equiv.from rot) (neg n)

    go : ∀ i → f i ≡ r .map-out _ _ i
    go (Ind.pos x) = pos x
    go (Ind.negsuc x) = neg x