Tom de Jong, 22 May 2025
An anonymous reviewer of our TYPES abstract [1] suggested that some of our
results could be generalized to weak factorization systems. Here we consider a
factorization system based on embeddings and maps whose fibers are all injective
types. We are also thinking about the connections to *algebraic* weak
factorization systems.
[1] Tom de Jong and MartΔ±Μn HΓΆtzel EscardΓ³.
"Examples and counter-examples of injective types in univalent mathematics".
Abstract for the 31st International Conference on Types for Proofs and
Programs (TYPES). 2025.
url: https://msp.cis.strath.ac.uk/types2025/abstracts/TYPES2025_paper6.pdf.
\begin{code}
{-# OPTIONS --safe --without-K #-}
open import UF.FunExt
module InjectiveTypes.WeakFactorizationSystem
(fe : FunExt)
where
open import InjectiveTypes.Blackboard fe
open import MLTT.Spartan
open import UF.Base
open import UF.Embeddings
open import UF.Equiv
open import UF.EquivalenceExamples
open import UF.IdEmbedding
open import UF.Univalence
\end{code}
We define a fiberwise algebraically injective map to be one whose fibers are all
algebraically injective types.
NB. It may be that fiberwise flabbiness is more convenient to work with.
\begin{code}
fiberwise-ainjective : {A : π€ Μ } {B : π₯ Μ } β (A β B) β (π¦ π£ : Universe)
β ((π¦ β π£) βΊ β π€ β π₯) Μ
fiberwise-ainjective f π¦ π£ =
each-fiber-of f (Ξ» F β ainjective-type F π¦ π£)
\end{code}
Any map can be factored as an embedding followed by a fiberwise algebraically
injective map.
\begin{code}
embedding-fiberwise-ainjective-factorization
: {A : π€ Μ } {B : π₯ Μ }
β is-univalent (π€ β π₯)
β (f : A β B)
β Ξ£ X κ (π€ β π₯)βΊ Μ ,
Ξ£ l κ (A β X) ,
Ξ£ r κ (X β B) , (f οΌ r β l)
Γ is-embedding l
Γ fiberwise-ainjective r (π€ β π₯) (π€ β π₯)
embedding-fiberwise-ainjective-factorization {π€} {π₯} {A} {B} ua f =
X , l , r , refl , l-is-embedding ua , r-fiberwise-ainjective ua
where
X : (π€ β π₯) βΊ Μ
X = Ξ£ b κ B , (fiber f b β (π€ β π₯) Μ )
ΞΉ : (Y : π€' Μ ) β Y β (Y β π€' Μ )
ΞΉ Y = Id
ΞΉ-is-embedding : is-univalent π€' β (Y : π€' Μ ) β is-embedding (ΞΉ Y)
ΞΉ-is-embedding ua _ = UA-Id-embedding ua fe
l : A β X
l = NatΞ£ (Ξ» b β ΞΉ (fiber f b)) β β domain-is-total-fiber f β
l-is-embedding : is-univalent (π€ β π₯) β is-embedding l
l-is-embedding ua =
β-is-embedding
(equivs-are-embeddings' (domain-is-total-fiber f))
(NatΞ£-is-embedding
(fiber f)
(Ξ» b β fiber f b β π€ β π₯ Μ )
(Ξ» b β ΞΉ (fiber f b))
(Ξ» b β ΞΉ-is-embedding ua (fiber f b)))
r : X β B
r = prβ
r-fiberwise-ainjective : is-univalent (π€ β π₯)
β fiberwise-ainjective r (π€ β π₯) (π€ β π₯)
r-fiberwise-ainjective ua b =
equiv-to-ainjective
(fiber r b)
(fiber f b β π€ β π₯ Μ )
(power-of-ainjective (universes-are-ainjective-Ξ ' ua))
(prβ-fiber-equiv b)
\end{code}
We have (specified) diagonal lifts of embeddings against fiberwise algebraically
injective maps.
We consider a commutative square with j an embedding and r fiberwise
algebraically injective and we look to find diagonal filler: a map e : Y β D
making the resulting triangles commute.
f
X ------> D
| ^ |
| / |
j | βe / | r
| / |
| / |
v / v
Y ------> E
g
Note that in our case we have a designated filler.
\begin{code}
module lifting-problem
{X : π€ Μ } {Y : π₯ Μ } {D : π¦ Μ } {E : π£ Μ }
(j : X β Y) (f : X β D) (r : D β E) (g : Y β E)
(j-is-embedding : is-embedding j)
(r-fiberwise-ainjective : fiberwise-ainjective r (π€ β π₯) π£')
(comm-sq : r β f βΌ g β j)
where
lifting-problem-has-a-solution : Ξ£ eΒ κ (Y β D) , (e β j βΌ f) Γ (r β e βΌ g)
lifting-problem-has-a-solution = e , e-upper-triangle , e-lower-triangle
where
module _ (y : Y) where
fΜ
: fiber j y β fiber r (g y)
fΜ
(x , e) = (f x , (r (f x) οΌβ¨ comm-sq x β©
g (j x) οΌβ¨ ap g e β©
g y β))
π : Ξ£ e κ fiber r (g y) , ((p : fiber j y) β e οΌ fΜ
p)
π = ainjective-types-are-aflabby
(fiber r (g y))
(r-fiberwise-ainjective (g y))
(fiber j y)
(j-is-embedding y)
fΜ
eΚΈ = prβ π
eΚΈ-extends : (p : fiber j y) β eΚΈ οΌ fΜ
p
eΚΈ-extends = prβ π
e : Y β D
e y = prβ (eΚΈ y)
e-lower-triangle : r β e βΌ g
e-lower-triangle y = prβ (eΚΈ y)
e-upper-triangle : e β j βΌ f
e-upper-triangle x = ap prβ I
where
I : eΚΈ (j x) οΌ fΜ
(j x) (x , refl)
I = eΚΈ-extends (j x) (x , refl)
\end{code}
In the above it is important to work with *algebraically* injective types: if we
instead had assumed that each fiber of r is only injective, then we would have
gotten for each y : Y an unspecified element of D only, which, in the absence of
choice, fails to produce a function.
Finally, since propositions and injective types are closed under retracts and
because retractions of maps induce retractions of their fibers [HoTTBook,
Lemma 4.7.3], the embeddings and the fiberwise injective maps are closed under
retracts, so that by the "retract argument" [Rie14, Lemma 11.2.3], the
embeddings and fiberwise injective maps indeed give rise to a weak factorization
system.
TODO. Formalize this and, as a preliminary, retracts of maps.
[Rie14] Emily Riehl. Categorical Homotopy Theory.
New Mathematical Monographs 24.
Cambridge University Press, 2014.
doi: 10.1017/ CBO9781107261457.
url: https://math.jhu.edu/~eriehl/cathtpy.pdf.