nLab Whitehead tower in an (infinity,1)-topos



(,1)(\infty,1)-Topos theory

(∞,1)-topos theory

structures in a cohesive (∞,1)-topos

Cohesive \infty-Toposes



If an (∞,1)-topos H\mathbf{H} is locally ∞-connected it has a good notion of geometric homotopy groups of its objects. In terms of these, there is an analog in H\mathbf{H} of the notion of the classical notion of Whitehead tower in the archetypical (,1)(\infty,1)-topos Top:

the Whitehead tower of an object XHX \in \mathbf{H} is the tower of nn-connected homotopy fibers of the canonical morphism into the (∞,1)-topos-theoretic Postnikov tower Π n+1(X)Π n(X)\cdots \to \mathbf{\Pi}_{n+1}(X) \to \mathbf{\Pi}_n(X) \to \cdots of the structured path ∞-groupoid Π(X)\mathbf{\Pi}(X) of XX.

See also the section Universal coverings and geometric Whitehead towers at cohesive (∞,1)-topos.


Let H\mathbf{H} be a locally ∞-connected (∞,1)-topos HΓLConstΠGrpd\mathbf{H} \stackrel{\overset{\Pi}{\to}}{\stackrel{\overset{LConst}{\leftarrow}}{\underset{\Gamma}{\to}}} \infty Grpd. Write

Π:=LConstΠ:HH \mathbf{\Pi} := LConst \circ \Pi : \mathbf{H} \to \mathbf{H}

for the internal fundamental ∞-groupoid in a locally ∞-connected (∞,1)-topos-functor. From the adjunction relation this comes with the canonical natural morphism

XΠ(X). X \to \mathbf{\Pi}(X) \,.

For nn \in \mathbb{N} write

H nτ nH \mathbf{H}_{\leq n} \stackrel{\overset{\tau_{\geq n}}{\leftarrow}}{\overset{}{\hookrightarrow}} \mathbf{H}

for the reflective (∞,1)-subcategory of n-truncated objects of H\mathbf{H} and write τ n\mathbf{\tau}_{\leq n} for the localization

τ n:Hτ nH nH. \mathbf{\tau}_{\leq n} : \mathbf{H} \stackrel{\tau_{\leq n}}{\to} \mathbf{H}_{\leq n} \hookrightarrow \mathbf{H} \,.


Π n:HΠHτ nH \mathbf{\Pi}_n : \mathbf{H} \stackrel{\mathbf{\Pi}}{\to} \mathbf{H} \stackrel{\mathbf{\tau}_{\leq n}}{\to} \mathbf{H}

for the internal homotopy nn-groupoid. For XHX \in \mathbf{H} we have the (∞,1)-Postnikov tower

Π 2(X)Π 1(X)Π 0(X) \cdots \to \mathbf{\Pi}_2(X) \to \mathbf{\Pi}_1(X) \to \mathbf{\Pi}_0(X)

of Π(X)\mathbf{\Pi}(X).


(Whitehead tower)

For XHX \in \mathbf{H}, its Whitehead tower is the sequence of objects

*X (2)X (1)X (0)X * \to \cdots \to X^{(2)} \to X^{(1)} \to X^{(0)} \simeq X

in H\mathbf{H}, where for each nn \in \mathbb{N} the object X (n+1)X^{(n+1)} is the homotopy fiber of the canonical morphism XΠ n+1X \to \mathbf{\Pi}_{n+1}, i.e. the object defined by the pullback diagram

X (n+1) * X Π n+1(X). \array{ X^{(n+1)} &\to& * \\ \downarrow && \downarrow \\ X &\to& \mathbf{\Pi}_{n+1}(X) } \,.

Here the morphisms X (n+1)X (n)X^{(n+1)} \to X^{(n)} are induced from the universality of the pullback:

X (n+1) X (n) Π (n+1)(X) X Π n(X) \array{ X^{(n+1)}&\to&X^{(n)}&& \mathbf{\Pi}_{(n+1)}(X) \\ &\searrow &\downarrow&\nearrow& \downarrow \\ &&X &\to& \mathbf{\Pi}_n(X) }

We have that Π n(X)LConstτ nΠ(X)\mathbf{\Pi}_n(X) \simeq LConst \tau_{\leq n} \Pi(X).

A homotopy-commuting diagram

X (n) * X Π n(X) \array{ X^{(n)} &\to& {*} \\ \downarrow && \downarrow \\ X &\to& \mathbf{\Pi}_n(X) }

in H\mathbf{H} corresponds by the adjunction relation to diagram

Π(X (n)) * Π(X) Π n(X) \array{ \Pi(X^{(n)}) &\to& {*} \\ \downarrow && \downarrow \\ \Pi(X) &\to& {\Pi}_n(X) }

in ∞Grpd. This being universal means that Π(X (n))\Pi(X^{(n)}) is nn-connected, and universal with that property as an object over Π(X)\Pi(X).



For *XH* \to X \in \mathbf{H} a pointed object and nn \in \mathbb{N}, n1n \geq 1 , define the object B nπ n(X)\mathbf{B}^n \mathbf{\pi}_n(X) to be the homotopy fiber of Π n(X)Π n1(X)\mathbf{\Pi}_n(X) \to \mathbf{\Pi}_{n-1}(X), so that we have a fibration sequence

B nπ n(X)Π n(X)Π n1(X). \mathbf{B}^n \mathbf{\pi}_n(X) \to \mathbf{\Pi}_n(X) \to \mathbf{\Pi}_{n-1}(X) \,.

With Π(X)GrpdTop\Pi(X) \in \infty Grpd \simeq Top the underlying topological space of XHX \in \mathbf{H} (its geometric realization) we have that

Bπ n(X)LConst nπ n(X), \mathbf{B} \mathbf{\pi}_n(X) \simeq LConst \mathcal{B}^n \pi_n(X) \,,

where nπ n(X)\mathcal{B}^n \pi_n(X) denotes the homotopy fiber of Π n(X)Π (n1)(X)\Pi_n(X) \to \Pi_{(n-1)}(X) in ∞Grpd.



This follows from τ nLConstΠ(X)LConstτ nΠ(X)\mathbf{\tau}_{\leq n} LConst \Pi(X) \simeq LConst \tau_{\leq n} \Pi(X). This, in turn, can for instance be checked in terms of the model structure on simplicial presheaves, using that τ\tau is objectwise the coskeleton operation. More on that at Postnikov tower in an (∞,1)-category.


For each n1n \geq 1 we have a fibration sequence

X (n)X (n1)B nπ n(X). X^{(n)} \to X^{(n-1)} \to \mathbf{B}^n \mathbf{\pi}_n(X) \,.

Regard the diagram

X (n) * X (n1) B nπ n(X) * X Π n(X) Π (n1)(X). \array{ X^{(n)} &\to& * \\ \downarrow && \downarrow \\ X^{(n-1)} & \to & \mathbf{B}^n \mathbf{\pi}_n(X) &\to& * \\ \downarrow && \downarrow && \downarrow \\ X &\to& \mathbf{\Pi}_n(X) &\to& \mathbf{\Pi}_{(n-1)}(X) } \,.

Here the right square is the defining (,1)(\infty,1)-pullback diagram of B nπ n(X)\mathbf{B}^n \mathbf{\pi}_n(X) from above. Take also the left bottom square to be a homotopy pullback. Then from the pasting rules of pullbacks it follows that the composite bottom rectangle is also a pullback, which identifies the object X (n1)X^{(n-1)} on the left as indicated.

Similarly, form now the top square as a pullback. Then by the composition law of pullbacks we find that the composite vertical rectangle is a pullback, which identifies the top left object as X (n)X^{(n)}.


For \infty-groupoids

In the archetypical (,1)(\infty,1)-topos H=\mathbf{H} = ∞Grpd the functors Π\Pi and Π\mathbf{\Pi} are the identity and so Π 1(X)Π 0(X)\cdots \to \mathbf{\Pi}_1(X) \to \mathbf{\Pi}_0(X) is just the standard Postnikov tower X 1X 0\cdots \to X_1 \to X_0 .

If we use the model structure on simplicial sets in order to present Grpd\infty Grpd, then the Postnikov tower may be realized by the simplicial coskeleton operation

cosk 2(X)cosk 1(X). \cdots \to \mathbf{cosk}_2(X) \to \mathbf{cosk}_1(X) \,.

If XX is connected, the homotopy pullback

X (n+1) * X X (n) \array{ X^{(n+1)} &\to& * \\ \downarrow && \downarrow \\ X &\to& X_{(n)} }

defining the Whitehead tower of an \infty-groupoid incarnated as a Kan complex XX may be computed as an ordinary pullback using a fibrant replacement diagram, such as thatr replacing the point by the decalage simplicial set, hence as the ordinary pullback

X (n+1) Deccosk nX X cosk nX \array{ X^{(n+1)} &\to& Dec \mathbf{cosk}_n X \\ \downarrow && \downarrow \\ X &\to& \mathbf{cosk}_n X }

in sSet.

If XX is a pointed \infty-groupoid then B nπ n(X)\mathbf{B}^n \mathbf{\pi}_n(X) is the Eilenberg-Mac Lane space K(π n(X),n)K(\pi_n(X),n).

For topological \infty-groupoids

It is often useful to think of topological spaces as embedded into the (∞,1)-topos of topological ∞-groupoids, i.e. the (∞,1)-category of (∞,1)-sheaves/∞-stacks on (a small version of) the site of “all” topological spaces.

H=Sh (,1)(Top). \mathbf{H} = Sh_{(\infty,1)}(Top) \,.

The universal covering spaces

For XX a topological space regarded as a representable object in Sh (,1)(Top)Sh_{(\infty,1)}(Top), we have that Π 1(X)\mathbf{\Pi}_1(X) is the fundamental groupoid of XX, regarded as a topological ∞-groupoid in the obvious way.

Then the homotopy fiber X (1)X^{(1)} of the constant path inclusion XΠ 1(X)X \to \mathbf{\Pi}_1(X) is the universal covering space of XX, as described in detail there.

The higher covering spaces

The analog of this statement for the higher items X (n)X^{(n)} with n>1n \gt 1 in the Whitehead tower of a topological spaces regarded in the (,1)(\infty,1)-topos of topological groupoids has been studied in

The following reviews some central ideas of this.

Non-traditional approach in TopTop

The following is some semblance of current research, except the stuff about the string 2-group. All errors and stupid ideas are mine - David Roberts

For locally nice spaces (say locally contractible), it is desirable to have a functorial construction of the Whitehead tower. A shadow of a low-dimensional case of this can be seen in the construction of the string 2-group given by Baez-Crans-Schreiber-Stevenson. Since finite-dimensional Lie groups have non-trivial third homotopy group, it is not possible to form the 3-connected cover in the category fin.dim.LieGrpfin.dim.LieGrp, like it is possible to take the 0-, and 1-connected covers. While most people give up the smoothness and make do with the topological group String GString_G, The BCSS construction leaps out of that category into that of strict (Frechet) Lie 2-groups. The construction is also functorial.

David Roberts: As an aside, I know of two classes of spaces with explicitly constructed (i.e. not via co-killing homotopy groups) 2-connected covers: the 2-sphere and its quotients by Z/n\mathbf{Z}/n (lens spaces), and the loop group ΩG\Omega G of a compact, simple, simply-connected Lie group GG (the latter is the level-1 central extension by U(1)U(1), the first needs no introduction). Does anybody know of any other nn-connected covers (other than, say, the Hopf fibrations) that are canonically given?

However, without one could demand a conceptually similar approach to the nn-connected cover of a general (locally nice) space or smooth manifold. For n=2n=2, and taking only topological spaces for consideration, this is contained in my thesis work. The rough result is that one gets a 2-connected topological groupoid X (2)X^{(2)} equipped with a map to XX that factors through the universal covering space X (1)X^{(1)}. This is functorial, and generalises to higher connected covers (at least heuristically - I don’t have nn-categorical superpowers).

But the general idea is that one would get an (n1)(n-1)-groupoid X (n)X^{(n)} over XX which is nn-connected and such that the map to XX factors through the (n2)(n-2)-groupoid X (n1)X^{(n-1)}. The map to XX should induce isomorphisms on homotopy groups π i\pi_i for i>ni\gt n, as in the usual Whitehead tower.

If we want to consider arbitrary nn and retain some sort of local triviality on our connected covers we cannot get away from the assumption that the space in question is locally contractible. An assumption of this sort appears in

  • Bertrand Toen, Vers une interpretation Galoisienne de la theorie de l’homotopie, Cahiers de Top. et Geom. Diff. Cat., Vol. XLIII-4 (2002), 257-312.

in the context of locally constant \infty-stacks and their monodromy (I haven’t got this article at the moment, and I suspect they may be (,1)(\infty,1)-stacks, but I’m not 100 percent certain -DMR). Technically speaking, we don’t need local contractibility but the existence of a basis of open sets UXU \hookrightarrow X such that this inclusion map is null-homotopic. But I will continue to call this local contractibilty, for lack of a better term (if there is such a term, I’d like to know -DMR).

Construction using topological nn-groupoids

Consider the fundamental nn-groupoid Π n(X)\Pi_n(X) of the locally contractible space XX (as a Trimble n-groupoid, say), or at least for now its underlying globular set. We can take the compact open-topology on the set of kk-morphisms for k<nk \lt n. As the space is locally contracible, in particular semi-locally nn-connected, the space Hom(S n1,X)Hom(S^{n-1},X) is semi-locally simply-connected (I have a fragment of a paper saying this is true for the 'absolute case' - that is, locally nn-connected implies the mapping space locally simply connected, but I expect it to be true for the relative case -DMR). In particular, we can take the fundamental groupoid Π 1(Hom(S n1,X))\Pi_1(Hom(S^{n-1},X)), which has a topology given in the usual way?. The arrow space of this fundamental groupoid is then non other than the space of nn-arrows of Π n(X)\Pi_n(X). It needs to be checked that the nn compositions # k\#_k, k=0,,n1k=0,\ldots,n-1 are continuous, as well as a bunch of other stuff, but I think this should follow from (unique) lifting theorems for covering spaces.

The object space of Π n(X)\Pi_n(X) is just XX, and so there is an inclusion XΠ n(X)X \hookrightarrow \Pi_n(X), and it is this that replaces the Postnikov section in the Whitehead construction outlined above. The topological fundamental nn-groupoid, even though it contains apparently more homotopical information than the untopologised fundamental nn-groupoid Π n(X) δ\Pi_n(X)^\delta (δ=\delta=discrete topology), I posit that under the assumptions on XX, the inclusion Π n(X) δΠ n(X)\Pi_n(X)^\delta \to \Pi_n(X) has an ana-n-functor pseudoinverse (taking the Grothendieck pretopology of open covers should be enough). On passing to the homotopy colimit this span should become a span weak homotopy equivalences, and so we can consider the topologised and the untopologised to be different representatives of the nn-type of XX.

Given a basepoint xXx\in X, we can form the tangent nn-groupoid T xΠ n(X)T_x \Pi_n(X), which is equivalent to the trivial nn-groupoid ** (even as a topological nn-groupoid), and gives us what should be in any sensible definition a fibration T xΠ n(X)Π n(X)T_x\Pi_n(X) \to \Pi_n(X). Pull back this fibration to XX, and call the resulting thing X (n)X^{(n)}. It is fairly easy to see that X (n)X^{(n)} is a topological (n1)(n-1)-groupoid over XX. This then should be the nn-connected cover of XX. For n=1n=1 this is precisely the classical construction of the universal covering space of a pointed space. For n=2n=2 this is treated in

  • D.M. Roberts, Fundamental bigroupoids and 2-covering spaces, PhD thesis, available here

and the two-dimensional homotopical tools developed there can be used to show that X (2)X^{(2)} is 2-connected.

A word is probably in order about the notion of kk-connectedness for topological nn-groupoids. This has its usual meaning, once homotopy groups π i\pi_i have been defined. The reader should be warned that these have nothing to do with the groups π 0Eq(1 . . 1 x)/\pi_0 Eq(1_{._{._{1_x}}})/\sim obtained from considering the autoequivalence ni+1n-i+1-group of the identity (i1)(i-1)-arrow on the identity (i2)(i-2)-arrow on … on the object xx. These should be defined in such a way as to agree with the homotopy colimit hocolimXhocolim X of XX considered as a truncated simplicial space. In particular, π 1\pi_1 of a topological groupoid is not the group of automorphisms of the basepoint, but a quotient of the set of generalised paths, an idea going back to Haefliger (for an intro see the early parts of chapter 2 of the above thesis, available at HomePage, or the preprint H. Colman, On the 1-homotopy type of Lie groupoids, arXiv:math/0612257).

There are other interpretations of X (n)X^{(n)}:

  • The (0-)source fibre of Π n(X)\Pi_n(X), which is:
  • The pullback of (s 0,t 0):Hom Π n(X)X×X(s_0,t_0):Hom_{\Pi_n(X)} \to X\times X along the inclusion {x}×XX×X\{x\}\times X \to X\times X, where Hom Π n(X)Hom_{\Pi_n(X)} is the (internalised) hom-(n1)(n-1)-groupoid familiar from the definition of a Trimble nn-groupoid.
  • The 'vertical fundamental (n1)(n-1)-groupoid' of PXXPX \to X, the path fibration.

The last can be thought of as a families version of the usual fundamental (n1)(n-1)-groupoid: take vertical paths, vertical homotopies between paths etc.

Last revised on September 24, 2018 at 15:41:25. See the history of this page for a list of all contributions to it.