nLab transferred model structure

Redirected from "Kan transfer theorem".
Contents

Context

Model category theory

model category, model \infty -category

Definitions

Morphisms

Universal constructions

Refinements

Producing new model structures

Presentation of (,1)(\infty,1)-categories

Model structures

for \infty-groupoids

for ∞-groupoids

for equivariant \infty-groupoids

for rational \infty-groupoids

for rational equivariant \infty-groupoids

for nn-groupoids

for \infty-groups

for \infty-algebras

general \infty-algebras

specific \infty-algebras

for stable/spectrum objects

for (,1)(\infty,1)-categories

for stable (,1)(\infty,1)-categories

for (,1)(\infty,1)-operads

for (n,r)(n,r)-categories

for (,1)(\infty,1)-sheaves / \infty-stacks

Contents

Idea

For CC a category with the structure of a model category and U:DCU:D\to C a functor having an adjoint, under certain conditions it is possible to transfer the model structure from CC to a model structure on DD by declaring the weak equivalences, and the fibrations (if UU is a right adjoint) or cofibrations (if UU is a left adjoint), in DD to be precisely those morphisms whose images under UU are such in CC. If UU is a right adjoint this is called the right transferred, right induced, or right lifted model structure, and dually in the left case.

Typically this arises in situations where DD consist of the “same” objects as CC but equipped with extra stuff, structure, property, and UU is the corresponding forgetful functor sending objects in DD to their underlying objects in CC. Then a left adjoint FF of UU is the corresponding free functor, while a right adjoint GG is a cofree functor.

Definition

Right transfer

Definition

(right transferred model structure)
Let CC be a model category and

(FU):DUFC (F \dashv U ) \;\colon\; D \underoverset {\underset{U}{\longrightarrow}} {\overset{F}{\longleftarrow}} {\;\;\;\bot\;\;\;} C

a pair of adjoint functors with right adjoint UU. Define a morphism in DD to be

  • a fibration or weak equivalence precisely if its image under UU is so in CC, respectively

    (by adjointness this means that the fibrations form equivalently the class with the right lifting property against the image under FF of the acyclic cofibrations in CC);

  • a cofibration precisely if it has the left lifting property with respect to the fibrations that are also weak equivalences, according to the previous item.

If these classes of maps define a model category structure on DD, then it is called the right transferred model structure (Crans 1995, §3) or sometimes right induced model structure or right lifted model structure from CC along RR.

In the above situation of Def. it is clear what one means by an acyclic fibration in DD, but the notion of acyclic cofibration in DD is, a priori, ambiguous. For the purpose of stating the following Thm. we declare that:

  1. an acyclic fibration in DD is (of course) a morphism that is both a fibration and a weak equivalence, or equivalently whose image under UU is an acyclic fibration in CC.

  2. a morphism is DD is

    1. a cofibration weak equivalence if it is both a cofibration and a weak equivalence in the sense of Def. ;

    2. an anodyne map if it has the left lifting property with respect to all fibrations.

    Of course, if the transferred model structure exists, then these two classes of maps coincide.

Conversely we have:

Theorem

A necessary and sufficient condition for the existence of the right transferred model structure (Def. ) is that:

  1. Every morphism in DD factors as

    1. a cofibration followed by a trivial fibration,

    2. an anodyne map followed by a fibration;

  2. every anodyne map is a weak equivalence.

Proof

Clearly the conditions are necessary. For the converse, the weak equivalences have the 2-out-of-3 property, all the classes of maps are closed under retracts, the lifting properties hold by definition (for the anodyne maps), and we have assumed the factorization properties (for the anodyne maps), so it remains to show that every cofibration weak equivalence is anodyne. This follows by the standard retract argument: if f:ABf \colon A \longrightarrow B is a cofibration weak equivalence, factor it as f=pif = p i with ii anodyne and pp a fibration. Then since anodyne maps are weak equivalences and weak equivalences satisfy 2-out-of-3, pp is an acyclic fibration. Thus ff has the left lifting property against pp, hence ff is a retract of ii and thus also anodyne.

Left transfer

Dually, if CC is a model category and we have an adjunction

(UG):DGUC (U \dashv G ) \;\colon\; D \underoverset { \underset{G}{\longleftarrow} } { \overset{U}{\longrightarrow} } {\;\;\;\bot\;\;\;} C

with UU left adjoint, define a morphism in DD to be

  • a cofibration or weak equivalence precisely if its image under UU is, respectively, in CC.

  • a fibration precisely if it has the right lifting property with respect to the cofibrations that are weak equivalences.

If these classes of maps define a model structure on DD, it is called the left transferred model structure (or sometimes left induced model structure or left lifted model structure) from CC.

The above necessary and sufficient conditions dualize directly.

Existence

For right transfer

Constructing factorizations

The most traditional way to obtain the factorization properties in the right-transferred case (Thm. ) is to assume that CC is cofibrantly generated.

Proposition

In the right-transferred situation (Def. ), suppose that

  1. CC is cofibrantly generated,

  2. the functor FF preserves small objects

    (which is the case, in particular, when UU preserves filtered colimits),

then every morphism in DD factors as a cofibration followed by a trivial fibration, and as an anodyne map followed by a fibration.

Moreover, if every anodyne map is a weak equivalence so that the transferred model structure on DD exists (by Thm. ), then it is itself cofibrantly generated, and for II (resp. JJ) a set of generating cofibrations (resp. acyclic cofibrations) in CC, the image set F(I)F(I) (resp. F(J)F(J)) forms a set of generating cofibrations (resp. acyclic cofibrations) in DD.

Proof

By the hom-isomorphism of the adjoint pair FUF \dashv U, a morphism ff is a fibration (resp. trivial fibration) if and only if it has the right lifting property against the maps in F(I)F(I) (resp. F(J)F(J)). Since FF preserves small objects, the sets F(I)F(I) and F(J)F(J) permit the small object argument in DD, which therefore yields (cofibration, trivial fibration) and (anodyne, fibration) factorizations.

Another way to produce these factorizations is to generate them by a functorial factorization:

Lemma

Suppose F:CD:UF \colon C \rightleftarrows D \colon U is an adjunction, where DD has pushouts, and that

  1. CC has a functorial factorization (L,R)(L,R) with a right weak composition law for factorizations (e.g. if it is algebraic),

  2. the induced functorial factorization on DD permits Garner’s small object argument (e.g. if CC and DD are locally presentable and all functors are accessible).

Then there is an induced (algebraic) weak factorization system on DD whose right class consists precisely of those maps whose UU-image admits an RR-algebra structure.

Proof

Given a morphism f:ABf\colon A\to B in DD, factor it as follows, where the upper commutative square is a pushout:

FUA A F((Uf) L) FE(Uf) Ef F((Uf) R) FUB B\array{ F U A & \longrightarrow & A \\ \mathllap{ {}^{F ((U f)_L)} } \big\downarrow & & \big\downarrow \\ F E (U f) & & E f \\ \mathllap{ {}^{F ((U f)_R)} } \big\downarrow & & \big\downarrow \\ F U B & \longrightarrow & B }

This yields a functorial factorization (L,R)(L',R') on DD. By the universal property of pushout, an RR'-algebra structure on ff is determined by a diagonal lifting in the square

FUA A F((Uf) L) f FE(Uf) F((Uf) R) B \array{ F U A & \longrightarrow & A \\ \mathllap{ {}^{F ((U f)_L)} } \big\downarrow & & \big\downarrow \mathrlap{{}^f} \\ F E (U f) & \underset{F ((U f)_R)}{\longrightarrow} & B }

which corresponds by adjunction to precisely an RR-algebra structure on UfU f. Thus, the right weak composition law for RR-algebras lifts to a right weak composition law for RR'-algebras. Garner’s small object argument applied to (L,R)(L',R') then produces a monad over codcod with a right strong composition law, which is therefore an algebraic weak factorization system whose algebraic right maps are the RR'-algebras, and in particular whose underlying right chlass is as desired those maps whose UU-image is in RR.

In particular, therefore, the factorizations of any accessible model structure right-lift along any adjunction between locally presentable model categories, whether or not they are cofibrantly generated. But the cofibrantly generated case can also be subsumed, by taking (L,R)(L,R) to be the one-step factorization produced by the generating left maps.

Left-lifting is generally rather trickier. But by invoking fancier categorical machinery, one can show more generally:

Lemma

Suppose given a functor U:DCU \colon D\to C between locally presentable categories and an accessible weak factorization system (L,R)(L,R) on CC.

  1. If UU has a left adjoint FF, then (L,R)(L,R) right-lifts along UU to an accessible wfs on DD.

  2. If UU has a right adjoint GG, then (L,R)(L,R) left-lifts along UU to an accessible wfs on DD.

Moreover, if (L,R)(L,R) is cofibrantly generated, so are the lifted wfs.

Proof

See HKRS15 and GKR18 for the lifting of accessible wfs. Cofibrant generation of a right-lifted wfs follows as in Theorem above, while in the left-lifted case it is Remark 3.8 of Makkai & Rosický 2014.

This sort of result seems to be necessary for the case of left-lifting; the simpler cofibrantly-generated argument does not dualize as directly.

Verifying acyclicity

The “acyclicity condition” in Thm. , i.e. that anodyne maps are weak equivalences, is usually the most difficult to check. Sometimes useful is the observation that for right-lifting in the cofibrantly generated case, it suffices to show that sequential colimits of pushouts of images under FF of the generating acyclic cofibrations in CC (i.e. forming F(J)F(J)-cell complexes) yield weak equivalences in DD. This is because the small object actually factors any map as such an F(J)F(J)-cell complex followed by a fibration; hence by the retract argument every anodyne map is a retract of an F(J)F(J)-cell complex.

Another useful sufficient condition is the following, going back roughly to section II.4 of (Quillen).

Proposition

In the situation of right transfer (Def. ), where U:DCU \colon D\to C is a right adjoint, suppose that

  1. DD has a fibrant replacement functor (in fact it suffices for individual objects and morphisms to have fibrant replacements, while functoriality is not necessary),

  2. DD has path objects for fibrant objects, i.e. a factorization of the diagonal Δ:AA×A\Delta : A \to A \times A as a weak equivalence followed by a fibration Δ:AP(A)fibA×A\Delta \colon A \stackrel{\simeq}{\to} P(A) \stackrel{fib}{\to} A \times A.

Then any anodyne map in DD is a weak equivalence. Thus, if the two factorizations exist in DD, then the transferred model structure exists.

Proof

The notation in the following is mostly that of Rezk 02, Lemma 7.6.
Let RR be a fibrant replacement functor, with natural weak equivalence j:IdRj \colon Id \to R, suppose f:XYf \colon X\to Y is anodyne, and consider the following commuting diagram:

Both squares are pullbacks, defining the object PRfP R f and the morphism ii making the following square commute:

The morphism π\pi is the composite PRfRX×RYRXP R f\to R X \times R Y \to R X at the top, which is a pullback of PRYRYP R Y \to R Y; but the latter is a fibration (as the composite PRYRY×RYRYP R Y \to R Y \times R Y \to R Y) and a weak equivalence (as a retraction for RYPRYR Y \to P R Y), so it and hence also π\pi are acyclic fibrations. Moreover, since πi=j X\pi i = j_X, by 2-out-of-3 ii is also a weak equivalence.

Now the projection PRfPYP R f\to P Y is a fibration, so since ff is anodyne there is a lift gg in the second square as shown. Since ii and j Yj_Y are weak equivalences, by 2-out-of-6 it follows that ff is a weak equivalence.

Note that this condition also dualizes straightforwardly to the left-transferred case.

For left transfer

Proposition

Given a pair of adjoint functors 𝒟RL𝒞 \mathcal{D} \underoverset {\underset{R}{\longleftarrow}} {\overset{L}{\longrightarrow}} {\;\;\bot\;\;} \mathcal{C} such that:

  1. 𝒞\mathcal{C} and 𝒟\mathcal{D} are locally presentable categories,

  2. 𝒞\mathcal{C} is equipped with the structure of a cofibrantly generated model category (hence a combinatorial model category) with classes of (co-)/fibrations and weak equivalences Cof,Fib,WMor(𝒞)Cof, Fib, W \,\subset\, Mor(\mathcal{C}),

  3. RLP(L 1Cof)L 1(W)RLP\big( L^{-1} Cof \big) \,\subset\, L^{-1}(W) (i.e. co-anodyne maps are weak equivalences),

then the left-transferred model category structure on 𝒟\mathcal{D} exists (i.e. with cofibrations L 1(Cof)L^{-1}(Cof) and weak equivalences L 1(W)L^{-1}(W)) and is itself cofibrantly generated.

Proof

As observed in the proof of BHKKRS 2015, Thm. 2.23, the existence of the required cofibrantly generated factorization systems follows by Makkai & Rosický 2014, Rem. 3.8 with the comment below Def. 2.3 there, which in turn invokes Lurie 2009, §A.1.5.12. In particular, from inside the proof of Makkai & Rosický 2014, Thm. 3.2 one has that a small set of generating (acyclic) cofibrations of the left transferred model structure is given by those between κ \kappa -presentable objects, for some un-countable regular cardinal κ\kappa.

Properties

General

Lemma

If CC carries the structure of a right proper model category, then also a right-transferred model structure on DD (Def. ) is right proper.

Proof

Let

A× CB f B Fib A C \array{ A \times_C B &\stackrel{f}{\to}& B \\ \downarrow && \downarrow^{\mathrlap{\in Fib}} \\ A &\stackrel{\simeq}{\to}& C }

be a pullback diagram in DD, with the bottom morphism a weak equivalence and the right morphism a fibration. We need to show that then also the top morphism ff is a weak equivalence. By definition of transfer, this is equivalent to U(f)U(f) being a weak equivalence in CC.

Since UU is a right adjoint it preserves pullbacks, so that also

U(A× CB) U(f) U(B) Fib U(A) U(C) \array{ U(A \times_{C} B) &\stackrel{U(f)}{\to}& U(B) \\ \downarrow && \downarrow^{\mathrlap{\in Fib}} \\ U(A) &\stackrel{\simeq}{\to}& U(C) }

is a pullback diagram in CC. Since by definition of the transferred model strucure this is still the pullback of a weak equivalence along a fibration, and since CC is assumed to be right proper, it follows that U(f)U(f) is a weak equivalence in CC, hence that ff is a weak equivalence in DD.

Enrichment

Often the underlying model category CC is an enriched model category over some monoidal model category SS and one wishes to transfer also the model enrichment.

Proposition

Assume the adjunction

(FU):DUFC (F \dashv U )\; : \; D \stackrel{\overset{F}{\leftarrow}}{\underset{U}{\to}} C

satisfies the conditions of the above proposition for right transfer, so that the model structure on CC is transferred to DD. Consider the case that CC is moreover an SS-enriched model category and that DD can be equipped with the structure of a SS-enriched category that is also SS-powered and copowered.

Assume now that the SS-powering of DD is taken by UU to the SS-powering of CC, in that U(d (s 1s 2))=U(d) (s 1s 2)U(d^{(s_1 \to s_2)}) = U(d)^{(s_1 \to s_2)}.

Then the transferred model structure and the SS-enrichment on DD are compatible and make DD an SS-enriched model category.

Proof

By the axioms of enriched model category one sufficient condition to be checked is that for sts \to t any cofibration in SS and for XYX \to Y any fibration in DD, we have that the induced morphism

X tX s× Y sY t X^t \to X^s \times_{Y^s} Y^{t}

is a fibration, which is a weak equivalence if at least one of the two input morphisms is. By the induced model structure, this is checked by applying UU. But by assumption UU commutes with the powering, and since UU is a right adjoint it commutes with taking the pullback, so that under UU the morphism is

U(X) tU(X) s× U(Y) sU(Y) t U(X)^t \to U(X)^s \times_{U(Y)^s} U(Y)^{t}

which is the morphism induced from U(X)U(Y)U(X) \to U(Y). That this is indeed an (acyclic) fibration follows now from the fact that CC is an SS-enriched model category.

Examples

Example

The right-transfer along an adjoint equivalence LRL \dashv R always exists, and here also the cofibrations are given by the preimage under RR.

This is straightforward to check directly, even without the above existence theorems (e.g. math.se:a/2220718).

Example

The projective model structure on functors M DM^D is right-transferred from the product model structure on M obDM^{ob D}. Dually, the injective model structure on functors, when it exists, is left-transferred from M obDM^{ob D}.

Example

The model structure on algebraic fibrant objects is right-transferred from the underlying model category by forgetting the choice of fillers.

Example

If TT is an accessible strict 2-monad on a locally finitely presentable 2-category KK. then the category TAlg sT Alg_s of strict TT-algebras admits a right-transferred model structure from the 2-trivial model structure on KK. The acyclicity condition is proved by using pseudolimits of arrows for path objects, and the (cofibration, trivial fibration) factorization is constructed using a version of the construction of Lemma (since the 2-trivial model structure on KK may not be cofibrantly generated).

Example

A counterexample is provided as Example 3.7 of (Goerss & Schemmerhorn 2007). Let kk be a field of characteristic 2 and consider the adjunction

(SU):CGA kUSCh *k (S \dashv U )\; : \; CGA_k \stackrel{\overset{S}{\leftarrow}}{\underset{U}{\to}} Ch_*k

of the symmetric algebra functor and the forgetful functor between graded commutative DGAs and chain complexes. One sees that SS does not preserve the weak equivalence between 0 and the complex with one copy of kk in degrees nn and n1n-1. Since all chain complexes are cofibrant this means that (SU)(S \dashv U ) cannot be upgraded to a Quillen adjunction.

Example

A more category-theoretic counterexample is that the Reedy model structure on directed graphs internal to Cat does not right-transfer to the category DblCat? of double categories along the free-forgetful adjunction. One of the generatic trivial cofibrations for the Reedy model structure “transports” a horizontal arrow (in double-category terminology) along two vertical isomorphisms, which is a levelwise equivalence. But upon pushing out the free double category generated by this along a map into some other double category, new composite horizontal arrows may be generated in the codomain that are not isomorphic to the image of anything in the domain; thus the pushout is no longer a levelwise equivalence. So, the acyclicity condition fails.

References

The explicit notion of right transfer is due to:

and for the case of cofibrantly generated model categories due to:

The existence result of right transfer due to Crans 1995, Thm. 3.3 and Dwyer, Hirschhorn & Kan 1997, Thm. 91 (“Crans-Kan transfer theorem”) is reviewed/reworked, with variations, in:

These arguments all go back (cf. Goerss & Shemmerhorn 2007, p. 20) to:

The above proof of acyclicity in the presence of path objects is taken from:

Discussion of left transfer (left-induced model structures):

Beware that HKRS15 contains an error, corrected by:

Technical lemmas regarding the underlying cellular categories:

The right-transferred model structure on algebras for a 2-monad is from

Last revised on April 28, 2023 at 20:03:29. See the history of this page for a list of all contributions to it.