combinatorial model category


Model category theory

model category



Universal constructions


Producing new model structures

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

Model structures

for \infty-groupoids

for ∞-groupoids

for nn-groupoids

for \infty-groups

for \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



A combinatorial model category is a particularly tractable model category structure. (Notice however that there is also the, closely related, technical notion of a tractable model category).

Being combinatorial means that there is very strong control over the cofibrations in these model structures: there is a set (meaning small set, not proper class) of generating (acyclic) cofibrations, and all objects, in particular the domains and codomains of these cofibrations, are small objects.

So as a slogan we have that

A combinatorial model structure is one that is generated from small data : it is generated from a small set of (acyclic) cofibrations between small objects.

In fact, the combinatoriality condition is a bit stronger than that, as it requires even that every object is small and is the colimit over a small set of generating objects.

There exist large classes of model categories that either are combinatorial or, if not, are Quillen equivalent to ones that are. See the list of examples below.

The relevance of combinatorial model categories is given more abstractly by the result that

Combinatorial simplicial model categories are precisely those model categories that model presentable (∞,1)-categories.

For more see at locally presentable categories - introduction.



(Jeff Smith)

A model category CC is combinatorial if it is



Recall from the discussion at cofibrantly generated model category that this means that CC has a set (not a proper class) II of generating cofibrations and a set JJ of generating trivial cofibrations in that

cof=llp(rlp(I)) cof = llp(rlp(I))
fib=llp(rlp(J)). fib = llp(rlp(J)) \,.

Here fib,cofMor(C)fib, cof \subset Mor(C) is the collection of fibrations and cofibration, respectively, and llp(S),rlp(S)llp(S), rlp(S) is the collection of morphisms satisfying the left or right, respectively, lifting property with respect to a collection of morphisms SS.

Jeff Smith’s theorem, below, gives an equivalent characterization that is usually easier to handle.

Characterization theorems

There are two powerful theorems that characterize combinatorial model categories in terms of data that is often easier to handle:

Smith’s theorem

A central theorem about combinatorial model categories is Jeff Smith’s theorem which establishes the existence of combinatorial model category structures from a small amount of input data.


(Jeff Smith’s theorem)


such that

we have that CC is a combinatorial model category with

  • weak equivalences WW;

  • cofibrations cof(I)cof(I).

Moreover, every combinatorial model category arises in this way.

Here the notation is as described at cofibrantly generated model category, so: inj(I)=rlp(I)inj(I) = rlp(I) and cof(I)=llp(rlp(I))cof(I) = llp(rlp(I)).

This statement was announced by Jeff Smith in 1998 at a conference in Barcelona and appararently first appeared in print as (Beke, theorem 1.7). The above formulation follows (Barwick, prop 2.2).


To prove the first part of the statement, that the given data encodes a combinatorial model category, it is sufficient to find a small set JJ such that

cof(J)=Wcof(I). cof(J) = W \cap cof(I) \,.

With that the statement follows using the small object argument to show the existence of the required factorizations.

To find this small set, we make use of the assumption that the subcategory Arr W(C)Arr(C)Arr_W(C) \subset Arr(C) of weak equivalences and commuting squares in CC between them is an accessible subcategory of the arrow category Arr(C)Arr(C). This means that there is a small set W 0WW_0 \subset W such that every element of WW is a κ\kappa-directed colimit over element in W 0W_0 in Arr W(C)Arr_W(C), for some cardinal number κ\kappa. We set then

J:=W 0cof(I). J := W_0 \cap cof(I) \,.

By the assumption that Arr C(W)Arr_C(W) is an accessibly embedded accessible subcategory of Arr(C)Arr(C) and that CC itself is locally presentable, so that all its objects are small objects, it follows that every morphism

K M W L N \array{ K &\to& M \\ \downarrow && \downarrow^{\mathrlap{\in W}} \\ L &\to& N }

in Arr(C)Arr(C) lifts through one of the components in W 0W_0 of WW (this mechanism is described in detail at small object) as

K P M W 0 W L Q N. \array{ K &\to& P &\to& M \\ \downarrow && \downarrow^{\mathrlap{\in W_0}} && \downarrow^{\mathrlap{\in W}} \\ L &\to& Q &\to& N } \,.

In the special case that KLK \to L is in II, we can refine this to a factoring through J=Wcof(I)J = W \cap cof(I) as follows:

using the small object argument factor the canonical morphism L KPQL \coprod_K P \to Q as L KPcell(I)Rinj(J)QL \coprod_K P \stackrel{\in cell(I)}{\to} R \stackrel{\in inj(J)}{\to}Q. Then LQL \to Q lifts to LRL \to R and we obtain the factorization

K P M I W 0cof(I) W L R N \array{ K &\to& P &\to& M \\ \downarrow^{\mathrlap{\in I}} && \downarrow^{\mathrlap{\in W_0 \cap cof(I)}} && \downarrow^{\mathrlap{\in W}} \\ L &\to& R &\to& N }

of the original square from an element in II to an element in WW through an element in J=Wcof(I)J = W \cap cof(I). (In Beke, following Jeff Smith, this is called the solution set condition: W 0cof(I)W_0 \cap cof(I) is “a solution set for WW at II”).

By following now precisely the small object argument with the only difference that one factors all the squares over which one takes a colimit in that argument through elements in JJ as above, it follows now that every morphism AWBA \stackrel{\in W}{\to} B in WW may be factored as

Acell(J)Cinj(I)B. A \stackrel{\in cell(J)}{\to} C \stackrel{\in inj(I)}{\to} B \,.

This we use to show that every morphism fcof(I)Wf \in cof(I) \cap W is in cof(J)cof(J):

since fWf \in W we may factor ff as above and since fcof(I)f \in cof(I) we obtain a lift σ\sigma in

A cell(J) C f σ inj(I) B = B. \array{ A &\stackrel{\in cell(J)}{\to}& C \\ \downarrow^{f} &{}^\sigma \nearrow& \downarrow^{\mathrlap{\in inj(I)}} \\ B &\stackrel{=}{\to}& B } \,.

Rearranging this it becomes a retract diagram in Arr(C)Arr(C)

A = A = A f cell(J) f B σ C inj(I) B \array{ A &\stackrel{=}{\to}& A &\stackrel{=}{\to}& A \\ \downarrow^f && \downarrow^{\mathrlap{\in cell(J)}} && \downarrow^f \\ B &\stackrel{\sigma}{\to}& C & \stackrel{\in inj(I)}{\to} & B }

which shows that ff is a retract of an element in cell(J)cof(J)cell(J) \subset cof(J), hence itself in cof(J)cof(J).

And the converse statement is immediate: by definition Jcof(I)WJ \subset cof(I) \cap W and cof(J)cof(J) is the saturation of JJ under the operation of forming retracts of transfinite compositions of pushouts of elements of JJ, under which cof(I)Wcof(I) \cap W is assumed to be closed.

In total we have indeed cof(J)=cof(I)Wcof(J) = cof(I) \cap W which shows that the II and WW given determine a combinatorial model category.

To see the converse, that every combinatorial model structure arises this way, it is sufficient to show that for every combinatorial model category the category Arr W(C)Arr(C)Arr_W(C) \subset Arr(C) is an accessible category.

For applications of this theorem, the following auxiliary statements are useful.


For CC a combinatorial model category, the full subcategory inclusion

Mor(C) WMor(C) Mor(C)_W \hookrightarrow Mor(C)

of the arrow category on the weak equivalences is an accessible inclusion of an accessible category.

This is due to Smith. A proof appears as (Dugger, 7.4). See also (Barwick, prop. 1.10).



F:Mor(A)Mor(B) F : Mor(A) \to Mor(B)

be an accessible functor between arrow categories. Let BB be equipped with weak equivalences WW such that the full subcategory inclusion

Mor(W)Mor(B) Mor(W) \hookrightarrow Mor(B)

on the weak equivalences is an accessible embedding of an accessible category. Then so is the full subcategory of Mor(A)Mor(A) on the pre-images F 1(W)F^{-1}(W) in AA.

(Beke, prop. 1.18).


By general properties of accessible categories (see there) the full inverse image along an accessible functor of a full accessible subcategory is again accessible.

Dugger’s theorem

The following theorem is precisely the model-category theory version of the statement that every locally presentable (∞,1)-category arises as the localization of an (∞,1)-category of (∞,1)-presheaves.


(Dan Dugger)

Every combinatorial model category CC is Quillen equivalent to a left Bousfield localization L SSPSh(K) projL_S SPSh(K)_{proj} of the global projective model structure on simplicial presheaves SPSh(K) projSPSh(K)_{proj} on a small category KK

L SSPSh(K) proj QuillenC. L_S SPSh(K)_{proj} \stackrel{\simeq_{Quillen}}{\to} C \,.

This is (Dugger, theorem 1.1) building on results in (DuggerUniversalHomotopy).


The proof proceeds (the way Dugger presents it, at least) in roughly three steps:

  1. Use that [C op,sSet Quillen] proj[C^{op}, sSet_{Quillen}]_{proj} is in some precise sense the homotopy- free cocompletion of CC. This means that every functor γ:CB\gamma : C \to B from a plain category CC to a model category BB factors in an essentially unique way through the Yoneda embedding j:C[C op,sSet]j : C \to [C^{op},sSet] by a Quillen adjunction

    (γ^R):BRγ^[C op,sSet Quillen] proj. (\hat \gamma \dashv R) : B \stackrel{\overset{\hat \gamma}{\leftarrow}} {\underset{R}{\to}} [C^{op}, sSet_{Quillen}]_{proj} \,.

    The detailed definitions and detailed proof of this are discussed at (∞,1)-category of (∞,1)-presheaves.

  2. For a given combinatorial model category BB, choose C:=B λ cofC := B_\lambda^{cof} the full subcategory on a small set (guaranteed to exist since BB is locally presentable) of cofibrant λ\lambda-compact objects, for some regular cardinal λ\lambda, and show that the induced Quillen adjunction

    BRi^[(B λ cof) op,sSet] proj B \stackrel{\overset{\hat i}{\leftarrow}}{\underset{R}{\hookrightarrow}} [(B_\lambda^{cof})^{op}, sSet]_{proj}

    induced by the above statement from the inclusion i:B λ cofBi : B_\lambda^{cof} \hookrightarrow B exhibits BB as a homotopy-reflective subcategory in that the derived counit i^QRId \hat i \circ Q \circ R \stackrel{\simeq}{\to} Id (QQ some cofibrant replacement functor) is a natural weak equivalence on fibrant objects (recall from adjoint functor the characterization of adjoints to full and faithful functors).

  3. Define SS to be the set of morphisms in [(C λ cof) op,sSet][(C_\lambda^{cof})^{op}, sSet] that the left derived functor i^Q\hat i \circ Q of i^\hat i (here QQ is some cofibrant replacement functor) sends to weak equivalences in BB. Then form the left Bousfield localization L S[(C λ cof) op,sSet] projL_S [(C_\lambda^{cof})^{op},sSet]_{proj} with respect to this set of morphisms and prove that this is Quillen equivalent to BB.

Carrying this program through requires the following intermediate results.

First recall from the discussion at (∞,1)-category of (∞,1)-presheaves that to produce the Quilen adjunction (i^R)(\hat i \dashv R) from ii, we are to choose a cofibrant resolution functor

I:C[Δ,B] I : C \to [\Delta,B]

of i:C=B λ cofBi : C= B_\lambda^{cof} \to B.

The adjunct of this is a functor I˜:C×ΔB\tilde I : C \times \Delta \to B. For each object bBb \in B write (C×Δb)(C \times \Delta \downarrow b) for the slice category induced by this functor.

Lemma (Dugger, prop. 4.2)

For every fibrant object bBb \in B we have that the homotopy colimit hocolim(C×Δb)B)hocolim (C \times \Delta \downarrow b) \to B) is weakly equivalent to i^QR(b)\hat i \circ Q\circ R (b).

Corollary (Dugger, cor. 4.4) The induced Quillen adjunction

B[C op,sSet] B \stackrel{\leftarrow}{\to} [C^{op}, sSet]

is a homotopy-reflective embedding precisely if the canonical morphisms

hocolim(C×Δb)b hocolim (C \times \Delta \downarrow b) \to b

are weak equivalences for every fibrant object bBb \in B.

Notice that the theorem just mentions plain combinatorial model categories, not simplicial model categories. But of course by basic facts of enriched category theory Funct(C op,SSet)Funct(C^{op}, SSet) is an SSet-enriched category and its projective global model structure on functors Func(C op,SSet) projFunc(C^{op}, SSet)_{proj} is compatibly a simplicial model category, as are all its Bousfield localizations. (See model structure on simplicial presheaves for more details.) Therefore an immediate but very useful corollary of the above statement is


Every combinatorial model category is Quillen equivalent to one which is

Tractable combinatorial model categories

A combinatorial model category is a tractable model category precisely if the set II of generating cofibrations can be chosen such that all elements have a cofibrant object as domain.

A left proper combinatorial model category precisely if the set JJ of generating trivial cofibrations can be chosen with cofibrant domain.

This are corollaries 2.7 and 2..8 in Bar.


Homotopy colimits


In a combinatorial model category, for every sufficiently large regular cardinal κ\kappa the following holds:

See also at filtered homotopy colimit.


This appears as proposition 7.3 in Dug00, reproduced for instance as prop. 2.5 in Bar.

The point is to choose κ\kappa such that all domains and codomains of the generating cofibrations are κ\kappa-compact object. This is possible since by assumption that CC is a locally presentable category all its objects are small objects, hence each a λ\lambda-compact object for some cardinal λ\lambda. Take κ\kappa to be the maximum of these.

Let F,G:JCF, G : J \to C be κ\kappa-filtered diagrams in CC and FGF \to G a natural transformation that is degreewise a weak equivalence. Using the functorial factorization provided by the small object argument this may be factored as FHGF \to H \to G where the first transformation is objectwise an acyclic cofibration and the second objectwise an acyclic fibration, and by functoriality of the factorization this sits over a factorization

lim Flim Hlim G. \lim_\to F \stackrel{\simeq}{\hookrightarrow} \lim_\to H \stackrel{}{\to}\lim_\to G \,.

It remains to show that the second morphism is a weak equivalence. But by our factorization and by 2-out-of-3 applied to our componentwise weak equivalences, we have that all its components H(j)G(j)H(j) \to G(j) are acyclic fibrations.

At small object it is described in detail how κ\kappa-smallness of an object XX implies that morphisms from XX into a κ\kappa-filtered colimit lift to some component of the colimit

H(j1) H(j) H(j+1) f^ X f lim H. \array{ \cdots&\to&H(j-1) &\to& H(j) &\to& H(j+1) &\to& \cdots \\ &&&{}^{\mathllap{\exists \hat f}}\nearrow&\downarrow & \swarrow \\ &&X& \stackrel{\forall f}{\to} &\lim_\to H } \,.

So given a diagram

X lim H I Y lim G \array{ X &\to& \lim_\to H \\ \downarrow^{\mathrlap{\in I}} && \downarrow \\ Y &\to& \lim_\to G }

we are guaranteed, by the κ\kappa-smallness of XX and YY that we established above, a lift

X H(j) lim H I rlp(I) Y G(j) lim G \array{ X &\to& H(j) &\to& \lim_\to H \\ \downarrow^{\mathrlap{\in I}} && \downarrow^{\in \mathrlap{\in rlp(I)}} && \downarrow \\ Y &\to& G(j) &\to& \lim_\to G }

into some component at jJj \in J and hence a lift

X H(j) lim H I rlp(I) Y G(j) lim G. \array{ X &\to& H(j) &\to& \lim_\to H \\ \downarrow^{\mathrlap{\in I}} & \nearrow & \downarrow^{\in \mathrlap{\in rlp(I)}} && \downarrow \\ Y &\to& G(j) &\to& \lim_\to G } \,.

Thereby lim Hlim G\lim_\to H \to \lim_\to G is in rlp(I)Wrlp(I) \subset W.

Bousfield localization

Combinatorial model categories, like cellular model categories have a good theory of Bousfield localizations, at least if in addition they are left proper. See Bousfield localization of model categories for more on this.


Basic examples

Basic examples are

Cisinski model structures

More generally, every Cisinski model structure is combinatorial.

Derived examples

Further classes of examples are obtained from such basic examples by localizing presheaf categories with values in these:

From cofibrantly generated model categories

Not every cofibrantly generated model category is also a combinatorial model category.

For instance:


Top with the standard model structure on topological spaces is cofibrantly generated, but not combinatorial. But it is Quillen equivalent to a combinatorial model structure, namely to the standard model structure on simplicial sets (see homotopy hypothesis).

One might therefore ask which cofibrantly generated model categories are Quillen equivalent to combinatorial ones. It turns out that if one assumes the large-cardinal hypothesis Vopěnka's principle, then every cofibrantly generated model category is Quillen equivalent to a combinatorial one. In fact, if we slightly generalize the notion of “cofibrantly generated,” this statement is equivalent to Vopěnka’s principle. For a discussion of this see

  • J. Rosicky, Are all cofibrantly generated model categories combinatorial? (ps)

Although Vopěnka’s principle cannot be proven from ZFC, and in fact is fairly strong as large cardinal hypotheses go, this means that looking for cofibrantly generated model categories that are not Quillen equivalent to combinatorial ones is probably a waste of time. Certainly, all known cofibrantly generated model categories are Quillen equivalent to simplicial ones, usually in a fairly natural way.

Simplicial combinatorial model categories

Those combinatorial model categories that are at the same time simplicial model categories are precisely those that present presentable (∞,1)-categories. See combinatorial simplicial model category.

Locally presentable categories: Cocomplete possibly-large categories generated under filtered colimits by small generators under small relations. Equivalently, accessible localizations of free cocompletions. Accessible categories omit the cocompleteness requirement; toposes add the requirement of a left exact localization.

(n,r)-categoriestoposeslocally presentableloc finitely preslocalization theoremfree cocompletionaccessible
(0,1)-category theorylocalessuplatticealgebraic latticesPorst’s theorempowersetposet
category theorytoposeslocally presentable categorieslocally finitely presentable categoriesAdámek-Rosický’s theorempresheaf categoryaccessible categories
model category theorymodel toposescombinatorial model categoriesDugger’s theoremglobal model structures on simplicial presheavesn/a
(∞,1)-topos theory(∞,1)-toposeslocally presentable (∞,1)-categoriesSimpson’s theorem(∞,1)-presheaf (∞,1)-categoriesaccessible (∞,1)-categories


Much of the theory of combinatorial model categories goes back to Jeff Smith. Apparently Smith will eventually present a book on this subject. To date, however, his ideas and results appear reproduced in articles of other authors.

After Jeff Smith presented his recognition theorem at a conference in Barcelona, its first appearance in a publication is apparently lemma 1.8 in

The very definition of combinatorial model categories appears also for instance as definition A.2.6.1 in

or definition 1.3 in

Smith’s theorem appears as proposition (Lurie, A.2.6.8) and as (Barwick, prop. 1.7).

Dugger’s theorem is in

based on results in

See also

  • Zhen Lin Low, The heart of a combinatorial model category, Theory and Applications of Categories, Vol. 31, 2016, No. 2, pp 31-62 (arXiv:1402.6659)

Revised on November 19, 2016 09:24:54 by Urs Schreiber (