module Homotopy.Space.Circle where

Spaces: The circle🔗

The first example of nontrivial space one typically encounters when studying synthetic homotopy theory is the circle: it is, in a sense, the perfect space to start with, having exactly one nontrivial path space, which is a free group, and the simplest nontrivial free group at that: the integers.

data  : Type where
  base : 
  loop : base  base

Diagramatically, we can picture the circle as being the \infty-groupoid generated by the following diagram:

In type theory with K, the circle is exactly the same type as . However, with univalence, it can be shown that the circle has two different paths:

_ = 
möbius :   Type
möbius base = Bool
möbius (loop i) = ua (not , not-is-equiv) i

When pattern matching on the circle, we are asked to provide a basepoint b and a path l : b ≡ b, as can be seen in the definition above. To make it clearer, we can also define a recursion principle:

S¹-rec :  {} {A : Type } (b : A) (l : b  b)    A
S¹-rec b l base     = b
S¹-rec b l (loop i) = l i

We call the map möbius a double cover of the circle, since the fibre at each point is a discrete space with two elements. It has an action by the fundamental group of the circle, which has the effect of negating the “parity” of the path. In fact, we can use möbius to show that loop is not refl:

parity : base  base  Bool
parity l = subst möbius l true

_ : parity refl  true
_ = refl

_ : parity loop  false
_ = refl

refl≠loop : ¬ refl  loop
refl≠loop path = true≠false (ap parity path)

Fundamental group🔗

We now calculate the loop space of the circle, relative to an arbitrary implementation of the integers: any type that satisfies their type-theoretic universal property. We call this a HoTT take: the integers are the homotopy-initial space with a point and an automorphism. What we’d like to do is prove that the loop space of the circle is also an implementation of the integers, but it’s non-trivial to show that it is a set directly.

module S¹Path {} (univ : Integers ) where
  open Integers univ

We start by defining a type family that we’ll later find out is the universal cover of the circle. For now, it serves a humbler purpose: A value in Cover(x)\mathrm{Cover}(x) gives a concrete representation to a path basex\mathrm{base} \equiv x in the circle. We want to show that (basebase)Z(\mathrm{base} \equiv \mathrm{base}) \simeq \mathbb{Z}, so we set the fibre over the basepoint to be the integers:

  Cover :   Type
  Cover base     = 
  Cover (loop i) = ua rotate i

We can define a function from paths basebase\mathrm{base} \equiv \mathrm{base} to integers — or, more precisely, a function from basex\mathrm{base} \equiv x to Cover(x)\mathrm{Cover}(x). Note that the path basex\mathrm{base} \equiv x induces a path ZCover(x)\mathbb{Z} \equiv \mathrm{Cover}(x), and since Z\mathbb{Z} is equipped with a choice of point, then is Cover(x)\mathrm{Cover}(x).

  encode :  x  base  x  Cover x
  encode x p = subst Cover p point

Let us now define the inverse function: one from integer to paths. By the mapping-out property of the integers, we must give an equivalence from (basebase)(\mathrm{base} \equiv \mathrm{base}) to itself. Since (basebase)(\mathrm{base} \equiv \mathrm{base}) is a group, any element ee induces such an equivalence, by postcomposition xxex \mapsto x \cdot e. We take ee to be the generating non-trivial path, e=loope = \mathrm{loop}.

  post-loop : (base  base)  (base  base)
  post-loop = Iso→Equiv $
    (_∙ loop) , iso (_∙ sym loop)
       p  sym (∙-assoc p _ _) ·· ap (p ∙_) (∙-invl loop) ·· ∙-idr p)
       p  sym (∙-assoc p _ _) ·· ap (p ∙_) (∙-invr loop) ·· ∙-idr p)

  loopⁿ :   base  base
  loopⁿ n = map-out refl post-loop n

To prove that the map nloopnn \mapsto \mathrm{loop}^n is an equivalence, we shall need to extend it to a map defined fibrewise on the cover. We shall do so cubically, i.e., by directly pattern-matching on the argument xx.

When we’re at the base point, we already know what we want to do: we have a function loop:Z(basebase)\mathrm{loop}^{-} : \mathbb{Z} \to (\mathrm{base} \equiv \mathrm{base}) already, so we can use that. For the other case, we must provide a path looploop\mathrm{loop}^{-} \equiv \mathrm{loop}^{-} laying over the loop, which we can picture as the boundary of a square. Namely,

  decode :  x  Cover x  base  x
  decode base = loopⁿ

We can provide such a square as sketching out an open cube whose missing face has the boundary above. Here’s such a cube: the missing face has a dotted boundary.

  decode (loop i) code j = hcomp ( i   j) λ where
    k (k = i0)  square-loopⁿ (unglue ( i) code) i j
    k (i = i0)  loopⁿ (Equiv.η rotate code k) j
    k (i = i1)  loopⁿ code j
    k (j = i0)  base
    k (j = i1)  loop i

We understand this as a particularly annoying commutative diagram. For example, the front face expresses the equation loopn1loop=loopn\mathrm{loop}^{n-1}\mathrm{loop} = \mathrm{loop}^{n}. The proof is now straightforward to wrap up:

  encode-decode :  x (p : base  x)  decode x (encode x p)  p
  encode-decode _ = J  x p  decode x (encode x p)  p) $
    ap loopⁿ (transport-refl point)  map-out-point _ _

  encode-loopⁿ : (n : )  encode base (loopⁿ n)  n
  encode-loopⁿ n = p  sym (map-out-unique id refl  _  refl) n) where
    p : encode base (loopⁿ n)  map-out point rotate n
    p = map-out-unique (encode base  loopⁿ)
      (ap (encode base) (map-out-point _ _)  transport-refl point)
       x  ap (encode base) {y = loopⁿ x  loop} (map-out-rotate _ _ _)
          ·· subst-∙ Cover (loopⁿ x) loop point
          ·· uaβ rotate (subst Cover (loopⁿ x) point))

  ΩS¹≃integers : (base  base)  
  ΩS¹≃integers = Iso→Equiv $
    encode base , iso loopⁿ encode-loopⁿ (encode-decode base)

open S¹Path Int-integers public