nLab strict factorization system

Contents

Idea

A strict factorization system is like an orthogonal factorization system, but the factorizations are specified uniquely on the nose, rather than merely up to isomorphism.

Definition

A strict factorization system on a category CC comprises wide subcategories EE and MM of CC such that every morphism in CC factors uniquely (not just uniquely up to unique isomorphism) as mem e for eEe \in E and mMm \in M.

Note that a strict factorization system is not necessarily an orthogonal factorization system, since EE and MM may not contain the isomorphisms in CC. However, for every strict factorization system, there is a unique orthogonal factorization system containing it, given by closing EE under postcomposition by isomorphisms and closing MM under precomposition by isomorphisms (see §2.1 of Grandis (2000)).

Examples

  • Every category has two “trivial” strict factorization systems in which EE, respectively MM, consists of the identity morphisms only. The corresponding orthogonal factorization systems are those with one class consisting of the isomorphisms.

  • Part of the structure of a (non-generalized) Reedy category is a strict factorization system.

  • The category Set, as defined in a material set theory, has a strict factorization system consisting of the surjections and the inclusions of subsets. Its associated orthogonal factorization system consists of surjections and injections. Other categories built out of structured sets have similar strict factorization systems.

  • Each opfibration f:ABf \colon A \rightarrow B equipped with a splitting determines a strict factorisation system on AA whose classes of morphisms are the chosen opcartesian lifts and the vertical morphisms (those morphisms sent to identities by ff).

As algebras of the factorization 2-monad

Grandis, following Korostenski & Tholen, it is shown in detail how strict factorization systems are strict algebras for the 2-monad () (-)^\downarrow.

Such a 2-monad is induced by the comonoid structure of Cat\downarrow \in \mathbf{Cat}. Its unit η X:XX \eta_X : X \to X^\downarrow sends an object xXx \in X to its identity map, while its multiplication μ X:X ×X \mu_X : X^{\downarrow \times \downarrow} \to X^\downarrow sends a square (u,v):fg(u,v):f \to g in XX to its diagonal vf=guvf=gu. The 2-monad (() ,η,μ)((-)^\downarrow, \eta, \mu) is called by Grandis factorization monad.

Notably, every X X^\downarrow is equipped with a strict factorization system, given by top trivial and bottom trivial squares:

In fact, X X^\downarrow equipped with such strict fs is the free strict fs on XX. This can be formulated as follows:

Proposition

The category X X^\downarrow with the aforementioned strict fs is universal in the following sense: for each other category YY equipped with a strict fs and functor F:XYF:X \to Y, there is a functor G:X YG:X^\downarrow \to Y, factoring FF through η X\eta_X and that strictly preserves factorization systems.

Proof

Define G(f)G(f) to be the image of the factorization of FfFf.

Remark

The above also holds weakly for ortoghonal fss, so that GG now preserves the fs only up to unique functorial isomorphism.

We thus displayed η\eta as a universal arrow for () (-)^\downarrow, showcasing its nature as a left 2-adjoint to the forgetful functor U:FsCatCatU:\mathbf{FsCat} \to \mathbf{Cat}. Grandis goes on to show this adjunction is in fact 2-monadic, exhibiting () (-)^\downarrow-algebras as strict factorization systems.

A strict algebra of () (-)^\downarrow is a category XX equipped with a functor t:X Xt:X^\downarrow \to X such that t.η x=1 Xt.\eta_x = 1_X and t.t =t.μ Xt . t^\downarrow = t.\mu_X.

Let’s unpack the correspondence here.

The functor tt is easy to conceptualize: it sends a morphism f:abf:a \to b to its image t(f)t(f), this being the middle object in its (for now, only conjectural) factorization. Indeed, we can recover the desired strict factorization systems by looking at the image of the free factorization system on X X^\downarrow. Define a map in XX to be left (LL) if it equals t(1,f)t(1,f) for some f:abXf:a \to b \in X, and right (RR) if it equals t(f,1)t(f,1).

Now to show this pair (L,R)(L,R) is a strict fs we need to prove every morphism factors uniquely. By unitality a=t(η X(a))a = t(\eta_X(a)), so that given f:abf:a \to b in XX, there are left maps at(f)a \to t(f) and right maps t(f)bt(f) \to b given by the image of the factorization of η X(f)\eta_X(f) in X X^\downarrow. These maps must compose to ff, again by unitality, so (L,R)(L,R)-factorizations exists. Uniqueness follows by noticing f=t(h,1)t(1,k)f = t(h,1)t(1,k) implies f=h=kf=h=k.

Notice we didn’t use associativity, and in fact it can be proven from unitality alone. Let (u,v):fg(u,v):f \to g be an object in X ×X^{\downarrow \times \downarrow}, i.e. a commutative square in XX. Then t.t (u,v)t.t^\downarrow(u,v) is the object obtained by factoring t(u,v)t(u,v) (corresponding to the horizontal factorization in the square below) while t.μ X(u,v)t.\mu_X(u,v) is the object obtained by factoring vfvf (corresponds to the diagonal factorization below):

The morphisms in the horizontal factorization can be precomposed with t(1,f)t(1,f) and postcomposed t(g,1)t(g,1) to get a different factorization of vfvf. By uniqueness, we must have t.t (u,v)=t.μ X(u,v)t.t^\downarrow(u,v)=t.\mu_X(u,v).

Remark

A similar story holds for orthogonal factorization systems, which are equivalent to normal (i.e. strictly unital) pseudoalgebras of the factorization monad.

As distributive laws in a bicategory

One reason strict factorization systems are of interest is that they can be identified with distributive laws in the bicategory of spans, as shown by Rosebrugh & Wood (2002).

Ordinary orthogonal factorization systems can be similarly characterized by:

References

Strict factorization systems were defined in:

See also:

Last revised on July 4, 2024 at 15:09:22. See the history of this page for a list of all contributions to it.