nLab homotopy pullback

Context

$(\infty,1)$-Limits and colimits

limits and colimits

model category

for ∞-groupoids

Contents

Idea

A homotopy pullback is a special kind of homotopy limit: the appropriate notion of pullback in the context of homotopy theory. Homotopy pullbacks model the quasi-category pullbacks in the (infinity,1)-category that is presented by a given homotopical category. Since pullbacks are also called fiber products, homotopy pullbacks are also called homotopy fiber products.

A homotopy pushout is the dual concept.

For more details see homotopy limit.

In the context of homotopy type theory a homotopy pullback is the interpretation of the space of solutions to an equation.

Definition

In category theory

As with all homotopy limits, there is both a local and a global notion of homotopy pullback.

The global definition says that the homotopy pullback of a cospan $X \to Z \leftarrow Y$ in a category with weak equivalences $C$ is its image under the right derived functor of the base change functor $pb: C^{\to \leftarrow} \to C$.

The local definition says that the homotopy pullback of $X \to Z \leftarrow Y$ in a category with a notion of homotopy consists of a square

$\array{ P & \to& Y \\ \downarrow && \downarrow \\ X &\to& Z }$

that commutes up to homotopy, and such that for any other square

$\array{ T & \to& Y \\ \downarrow && \downarrow \\ X &\to& Z }$

that commutes up to homotopy, there exists a morphism $T\to P$, unique up to homotopy, making the two triangles commute up to homotopy, and similarly for homotopies and higher homotopies. In other words, there is an equivalence

$Map(T,P) \simeq HoSq(T,X\to Z\leftarrow Y)$

between the space of maps $T\to P$ and the space of homotopy commutative squares with vertex $T$.

In good situations, such as when $X,Y,Z$ are fibrant in a model category, the two constructions agree up to weak equivalence.

Note that in both cases, there is a canonical map from the actual pullback $X\times_Z Y$ to the homotopy pullback $X\times_Z^h Y$. In the global case this comes by the definition of a derived functor; in the local case it comes because a commutative square is, in particular, a homotopy commutative one.

In homotopy type theory

In homotopy type theory the homotopy pullback of a term of function type

$f : A \to C$

along a term of function type

$g : B \to C$

is given formally by precisely the same formula that would also define the ordinary fiber product of functions of sets, namely by

$\left\{ a : A, b : B \;\; | \;\; f(a) = g(b) \right\} \,.$

Spelled out, this is the dependent sum over the dependent identity type over the evaluation of $f$ and $g$.

What in classical logic is interpreted as the set of pairs $(a,b)$ such that $f(a)$ and $g(b)$ are equal here becomes the restriction of a mapping cocylinder.

Formal proof that this is the homotopy pullback in homotopy type theory is in (Brunerie). Proof in the categorical semantics of homotopy type theory is below.

Concrete constructions

We discuss various concrete constructions by ordinary pullbacks and ordinary limits such that under some sufficient conditions these compute homotopy pullbacks, up to weak equivalence.

General

Proposition

Let $A \to C \leftarrow B$ be a diagram in some model category. Then sufficient conditions for the ordinary (1-categorical) pullback $A \times_C B$ to present the homotopy pullback of the diagram are

Both statements are classical. They are reviewed for instance as Lurie, prop. A.2.4.4. The proof of the second statement is spelled out here.

Remark

Notice that a fibrant resolution of the diagram in the injective model structure on functors has both morphisms be a fibration. So the first point above says that (in the special case of pullbacks) something weaker than this is sufficient for computing the homotopy limit of the diagram.

This can be explained in model-categorical terms by the fact that the category of cospans also has a Reedy model structure in which the fibrant objects are precisely those considered in the first point above, and that homotopy limits can equally well be computed using this model structure (specifically, the adjunction $Const \dashv Lim$ is Quillen with respect to it).

Due to prop. 1 one typically computes homotopy pullbacks of a diagram by first forming a resolution of one of the two morphisms by a fibration and then forming an ordinary pullback.

If all objects involved are already fibrant, then such a resolution is provided by the factorization lemma. This says that a fibrant resoltuion of $B \to C$ is given by the total composite vertical morphism in

$\array{ C^I \times_C B &\to& B \\ \downarrow && \downarrow \\ C^I &\to& C \\ \downarrow \\ C } \,,$

where $C \stackrel{\simeq}{\to} C^I \to C \times C$ is a path object for the fibrant object $C$ (For instance, when $C$ is a closed monoidal homotopical category then this can be taken to be the internal hom with an interval object $I$.)

Then by prop. 1 we have

Corollary

If in $A \stackrel{f}{\to} C \stackrel{g}{\leftarrow} B$ all three objects are fibrant objects, then the homotopy pullback of this diagram is presented by the ordinary limit in

$\array{ A \times_{C}^h B &\to&\to& \to& B \\ \downarrow &\searrow&&& \downarrow \\ \downarrow & & C^I &\stackrel{d_1}{\to}& C \\ \downarrow & & \downarrow^{d_0} \\ A & \to & C } \,.$

or, which is the same up to isomorphism, as the ordinary pullback

$\array{ A\times_C^h B & \to & C^I \\ \downarrow && \downarrow \\ A\times B & \stackrel{(f,g)}{\to} & C\times C } \,.$

The canonical morphism $A \times_{C} B \to A \times_C^h B$ here is induced by the canonical section $s : C \to C^I$ into the path space object, hence by the commutativity of the diagram

$\array{ A\times_C B &\to& C & \stackrel{s}{\to} & C^I \\ \downarrow &&&{}_{\mathllap{\Delta}}\searrow& \downarrow \\ A\times B && \to && C\times C } \,.$

The homotopy pullback constructed in this way is an example of a strict homotopy limit, as mentioned at homotopy limit. In such a case, one can say that an arbitrary homotopy-commutative square

$\array{ W & \to& Y \\ \downarrow && \downarrow \\ X &\to& Z }$

is a homotopy pullback square if the induced morphism from $W$ to the strict homotopy pullback is a weak equivalence.

A useful class of example of this is implied by the following:

Proposition

Let $\mathcal{C}$ be a category of simplicial presheaves over some site equipped with a local projective model structure on simplicial presheaves with respect to that site.

Then an ordinary pullback of $A \to C \leftarrow B$ in $\mathcal{C}$ is a homotopy pullback already when one of the two morphisms is an objectwise Kan fibration.

Proof

The global projective model structure on simplicial presheaves is right proper. So by prop. 1 the ordinary pullback in question presents the homotopy pullback in the global structure. By the discussion at homotopy limit and Bousfield localization of model categories, this presents the (∞,1)-pullback of the diagram of (∞,1)-presheaves, and the fibrant replacement of that pullback in the local model structure presents the (∞,1)-sheafification of this (∞,1)-presheaf. This is (essentially by definition, see (∞,1)-topos) a left exact (∞,1)-functor and hence preserves finite (∞,1)-limits.

In homotopy type theory

If we unwind the categorical semantics of the above definition

$A \times_C^h B \simeq \{ a : A, b : B | (f(a) = g(b)) \}$

of the homotopy pullback in homotopy type theory, we re-obtain the above prescription for how to construct homotopy pullbacks.

So let the ambient category be a suitable type-theoretic model category.

Example

The type $a : A, b : B \vdash (f(a) = g(b))$ is obtained by substitution from the identity type of $C$. By the discussion there, the categorical semantics of substitution is given by pullback of the fibrations that interpret the dependent types, and so this is interpreted as the pullback $[a : A, b : B \vdash (f(a) = g(b))] \coloneqq (f,g)^* C^I$ of the path space object of $C$:

$\array{ [a : A, b : B \vdash (f(a) = g(b))] &\to& [Id C] \\ \downarrow && \downarrow \\ [a : A , b : B] &\stackrel{(f,g)}{\to}& [c_1 : C , c_2 : C] } \;\; = \;\; \array{ (f,g)^* C^I &\to& C^I \\ \downarrow && \downarrow \\ A \times B & \stackrel{(f,g)}{\to} & C \times C } \,.$

Forming the dependent sum over $a : A, b : B$ is simply interpreted as regarding the resulting object $(f,g)^* C^I$ as an object in $\mathcal{C} \simeq \mathcal{C}_{/*}$ instead of as an object in the slice category $\mathcal{C}_{/ A \times B}$.

Since by assumption on the categorical interpretation of a type, all objects here are fibrant, this coincides with the expression of the homotopy pullback from corollary 1 above.

Example

Specifically, let $f \colon A \longrightarrow B$ be a function, then the categorical semantics for the expression

$\underset{b \colon B}{\sum} fib(f,b) = \underset{b \colon B}{\sum} \underset{a \colon A}{\sum} (f(a) = b)$

is the canonical fibration replacement of $f$ as it appears notably in the factorization lemma

$\array{ A \times_B B^I &\longrightarrow& B^I \\ \downarrow && \downarrow \\ A \times B &\stackrel{(f, Id)}{\longrightarrow}& B \times B } \,.$

Properties

Homotopy fiber characterization

In plain homotopy types:

Proposition
$\array{ A &\stackrel{f}{\longrightarrow}& B \\ \downarrow && \downarrow \\ C &\stackrel{g}{\longrightarrow}& D }$

is a homotopy pullback diagram precisely if it induces a weak equivalence on all homotopy fibers

$\array{ hfib(f) &\longrightarrow& A &\stackrel{f}{\longrightarrow}& B \\ \downarrow^{\mathrlap{\simeq}} && \downarrow && \downarrow \\ hfib(g) &\longrightarrow& C &\stackrel{g}{\longrightarrow}& D }$

e.g. (CPS, 5.2)

Examples

Fiber sequences

Of particular interest are consecutive homotopy pullbacks of point inclusions. These give rise to fiber sequences and loop space objects.

References

General

See the references at homotopy limit .

A fairly comprehensive resource is the appendix of