nLab
fibration

Context

Homotopy theory

Category theory

There are a number of different types of morphism bearing the name fibration, which are all connected to each other at least by a zigzag of relationships.

Contents

Classical homotopy theory

In classical homotopy theory, a fibration p:EBp:E\to B is a continuous function between topological spaces that has a certain lifting property. The most basic property is that given a point eEe\in E and a path [0,1]B[0,1] \to B in BB starting at p(e)p(e), the path can be lifted to a path in EE starting at ee.

One generally also assumes the lifting of additional structures (including “higher homotopies”) in BB which, in particular, imply that the path lifting is unique up to homotopy. Different choices of what can be lifted give rise to different notions of fibration, for example:

  • In a Hurewicz fibration, all sorts of homotopies can be lifted.

  • In a Serre fibration, topological nn-cells can be lifted for all nn.

  • In a Dold fibration or “halb-fibration,” all homotopies can be lifted, but the lifting only has to agree with the given initial map up to vertical homotopy. A Hurewicz fibration is a Dold fibration where the vertical homotopy is stationary.

All three of these definitions give rise to a long exact sequence of homotopy groups. In fact, the exact sequence would follow from only requiring up-to-homotopy lifting for cubes. There doesn’t seem to be a name for this sort of map, but there is the following:

Abstract homotopy theory

Inspired by the role of fibrations in algebraic topology, part of the structure of a model category or a category of fibrant objects is a class of maps called “fibrations,” which also possess a lifting property relating them to the rest of the structure (cofibrations and weak equivalences). Examples include:

Fibrations have many good properties in homotopy theory. For example, under some extra assumptions pullback of a fibration is already a homotopy pullback (see there for details). Sufficient conditions are that in addition all three objects involved are fibrant objects or that the model category is a right proper model category.

Generally, every morphism can be replaced by a weakly equivalent fibration, which gives one way to compute a homotopy pullback by comuting an ordinary pullback of a fibrant enough weakly quivalent cospan diagram.

Transports and classifying spaces

There is a classical thorem that covering spaces p:EBp:E\to B of a locally connected space BB (which have unique path lifting) are equivalent to functors Π(B)Set\Pi(B)\to Set from the fundamental groupoid of BB to Set (hence to permutation representations of the fundamental group ).

The functor corresponding to p:EBp:E\to B takes a point bBb\in B to its fiber p 1(b)p^{-1}(b), and a path α\alpha from bb to bb' to the function p 1(b)p 1(b)p^{-1}(b) \to p^{-1}(b') defined by “the endpoint of the lift of α\alpha.”

Generalizing this massively, arbitrary topological fibrations p:EBp:E\to B correspond to functors from the fundamental \infty-groupoid of BB to the (,1)(\infty,1)-category of spaces, in an analogous way. Points are sent to fibers, paths to the endpoints of liftings, homotopies between paths to the result of lifting such homotopies, and so on. This functor is sometimes called the parallel transport corresponding to the fibration. There are a number of ways to make this precise, some of which make sense in classical homotopy theory (e.g. actions by a loop space) and some of which require higher categorical machinery. A discussion is at higher parallel transport in the section Flat ∞-parallel transport in Top.

More generally, one can consider fibrations in which the fibers are equipped with some extra structure, such as being a vector space or having an action by a group. This corresponds to restricting the codomain of the transport to some category of spaces with structure and structure-preserving maps. The most common version of this is a bundle with some structure group GG, in which case the transport lands in BG\mathbf{B}G, the delooping groupoid of GG with a singleobject (thought of as a generic GG-torsor) and GG as its endomorphisms. In such cases the word “parallel” is often added in front of “transport.”

If we replace the topological groupoid BG\mathbf{B}G its classifying space, G\mathcal{B}G – which is the geometric realization of simplicial topological spaces of the nerve of BG\mathbf{B}G – then passing to π 0\pi_0 recovers the classical fact that “classifying spaces classify”: there is a bijection between GG-bundles over a space XX and homotopy classes of maps XGX\to \mathcal{B}G. This bijection is realized by pulling back to XX the “universal GG-bundle” GG\mathcal{E}G \to \mathcal{B}G over the space G\mathcal{B}G. There are also classifying spaces for more general types of fibrations, constructed from the relevant subcategories of TopTop.

Fibrations in category theory

It is common in category theory to consider the objects of a slice category C/XC/X as “objects of CC varying over XX.” For example, an object AXA\to X of Set/XSet/X can be identified with an XX-indexed family {A x} xX\{A_x\}_{x\in X} of sets, where A xA_x is the fiber of AXA\to X over xXx\in X. Likewise, if XX is a topological space, we can regard an object of Top/XTop/X as a family of spaces (the fibers) “varying continuously” over XX. But as we have seen, in the topological case, in order to make this varying into a “functor” XTopX\to Top we need the map to be a fibration.

In category theory, there is analogous notion of when a functor p:EBp:E\to B is a fibration or fibered category or Grothendieck fibration or Cartesian fibration, which is exactly what is needed to make the assignment bp 1(b)b \mapsto p^{-1}(b) into a pseudo-functor BCatB\to Cat. (Actually, there are two such notions, since a functor on a non-groupoid can be either covariant or contravariant.) Thus, in this case Cat is the analogue of the “classifying space”, and there is a “universal Cartesian fibrationCat *CatCat_* \to Cat of which every other fibration is a pullback (modulo size issues).

Categorical fibrations also have a “lifting” property, but the liftings must satisfy an extra “universality” condition. For this reason, they are not the fibrations in any model structure on Cat. However, fibrations of this sort between groupoids are the same as isofibrations, and thus are the fibrations in the folk model structure on Grpd. See Grothendieck fibration for more details.

A discrete fibration is one in which we use Set instead of Cat as the classifying space.

Fibrations in higher category theory

Both forms of 2-categorical dualization are commonly encountered in the context of fibrations. Moreover, the distinction between the two is appreciable. Cofibrations? play a role dual to that of fibrations in homotopy theory, notably in the axioms for a model category. In this context, cofibrations have an entirely different geometric flavor from fibrations. On the other hand, opfibrations are the same as fibrations homotopically because of the invertibility of 2-cells. The duality between fibrations and opfibrations is more visible in category theory, where it’s the same duality as one finds between limits and colimits, for example. Confusingly, some (much?) of the older categorical (and algebro-geometric?) literature uses “cofibration” to mean “opfibration”.

An object such that the unique morphism to the terminal object is a fibration (in abstract homotopy theory) is called a fibrant object.

A replacement of a morphism be a weakly equivalent fibration is also called a resolution.

Revised on June 21, 2012 23:20:14 by Urs Schreiber (89.204.137.104)