module Algebra.Group.Homotopy.BAut where
Deloopings of automorphism groups🔗
Recall that any set generates a group , given by the automorphisms We also have a generic construction of deloopings: special spaces (for a group where the fundamental group recovers For the specific case of deloping automorphism groups, we can give an alternative construction: The type of small types merely equivalent to has a fundamental group of
module _ {ℓ} (T : Type ℓ) where
: Type (lsuc ℓ)
BAut = Σ[ B ∈ Type ℓ ] ∥ T ≃ B ∥
BAut
: BAut
base = T , inc (id , id-equiv) base
The first thing we note is that BAut
is a connected
type, meaning that it only has “a single point”, or, more precisely,
that all of its interesting information is in its (higher) path
spaces:
: (x : BAut) → ∥ x ≡ base ∥
connected = elim! λ b e → inc (p b e) where
connected : ∀ b e → (b , inc e) ≡ base
p _ = EquivJ (λ B e → (B , inc e) ≡ base) refl p
We now turn to proving that
We will define a type family
where a value
codes for an identification
Correspondingly, there are functions to and from these types: The core
of the proof is showing that these functions, encode
and decode
, are inverses.
: BAut → Type ℓ
Code = T ≃ b .fst
Code b
: ∀ b → base ≡ b → Code b
encode = path→equiv (ap fst p)
encode x p
: ∀ b → Code b → base ≡ b
decode (b , eqv) rot = Σ-pathp (ua rot) (is-prop→pathp (λ _ → squash) _ _) decode
Recall that is the type itself, equipped with the identity equivalence. Hence, to code for an identification it suffices to record — which by univalence corresponds to a path We can not directly extract the equivalence from a given point it is “hidden away” under a truncation. But, given an identification we can recover the equivalence by seeing how identifies
: ∀ b (p : base ≡ b) → decode b (encode b p) ≡ p
decode∘encode =
decode∘encode b (λ b p → decode b (encode b p) ≡ p)
J (Σ-prop-square (λ _ → squash) sq)
where
: ua (encode base refl) ≡ refl
sq = ap ua path→equiv-refl ∙ ua-id-equiv sq
Encode
and
decode
are inverses by a direct
application of univalence
.
: ∀ b (p : Code b) → encode b (decode b p) ≡ p
encode∘decode = ua.η _ encode∘decode b p
We now have the core result: Specialising encode
and decode
to the basepoint
, we conclude that loop space
is equivalent to
: (base ≡ base) ≃ (T ≃ T)
Ω¹BAut = Iso→Equiv
Ω¹BAut (encode base , iso (decode base) (encode∘decode base) (decode∘encode base))
We can also characterise the h-level of these connected components.
Intuitively the h-level should be one more than that of the type we’re
delooping, because BAut
“only has
one point” (it’s connected), and as we established right above, the
space of loops of that point is the automorphism group we delooped. The
trouble here is that BAut
has many
points, and while we can pick paths between any two of them, we can not
do so continuously (otherwise BAut
would be a proposition).
This turns out not to matter! Since “being of h-level ” is a proposition, our discontinuous (i.e.: truncated) method of picking paths is just excellent. In the case when is contractible, we can directly get at the underlying equivalences, but for the higher h-levels, we proceed exactly by using connectedness.
: ∀ n → is-hlevel T n → is-hlevel BAut (1 + n)
BAut-is-hlevel (x , f) (y , g) = Σ-prop-path! (sym (ua f') ∙ ua g')
BAut-is-hlevel zero hl where
: ∀ {X} → is-prop (T ≃ X)
extract = ext λ x → ap fst $
extract f g ((f e⁻¹) .snd .is-eqv (hl .centre))
is-contr→is-prop (f .fst x , is-contr→is-prop hl _ _)
(g .fst x , is-contr→is-prop hl _ _)
= ∥-∥-rec extract (λ x → x) f
f' = ∥-∥-rec extract (λ x → x) g
g' (suc n) hl x y =
BAut-is-hlevel {P = λ _ _ → is-hlevel (x ≡ y) (1 + n)}
∥-∥-elim₂ (λ _ _ → is-hlevel-is-prop _)
(λ p q → transport (ap₂ (λ a b → is-hlevel (a ≡ b) (1 + n)) (sym p) (sym q))
(Equiv→is-hlevel (1 + n) Ω¹BAut (≃-is-hlevel (1 + n) hl hl)))
(connected x) (connected y)