There are some classical results stating sufficient conditions on a manifold $\Sigma$ such that every homotopy equivalence $\Pi(\Sigma) \stackrel{\simeq}{\longrightarrow} \Pi(\Sigma)$ is homotopic to a diffeomorphism $\Pi(\Sigma \stackrel{\simeq}{\longrightarrow} \Sigma)$ (here "$\Pi$" denotes the functor from (simplicial) smooth manifolds to homotopy types, i.e. to infinity-groupoids).

What is known about this when the homotopy equivalences are required to respect the tangent bundle structure, up to homotopy?

More specifically, I am wondering about the following:

Since $\Sigma$ is a manifold (of dimension $n$, say), so $\Pi(\Sigma)$ is canonically equipped with a map $\Pi(\tau_\Sigma) : \Pi(\Sigma) \to B GL(n) \simeq B O(n)$. One may hence consider the automorphism group $\mathrm{Aut}_{/B O(n)}(\Pi(\Sigma))$ whose elements are homotopy auto-equivalences of $\Pi(\Sigma)$ that are required to fit into a diagram

$$ \array{ \Pi(\Sigma) && \stackrel{\simeq}{\longrightarrow} && \Pi(\Sigma) \\ & \searrow &\Leftarrow_{\simeq}& \swarrow \\ && B O(n) } \,. $$

This is naturally a topological group, whose homotopy type reflects homotopies between such maps over $B O(n)$.

How close is this to the homotopy type of the diffeomorphism group? I.e. how close is

$$ B \mathrm{Diff}(\Sigma) $$


$$ B \mathrm{Aut}_{/B O(n)}(\Pi(\Sigma)) $$


That's what I am wondering. For readers tolerant of more stacky language I would like to add the following thought, which may or may not have some bearing on this:

Write $\mathbf{B}GL(n)$ for the smooth stack which is the homotopy quotient of the point by the action of the smooth group $GL(n)$. Then the smooth tangent bundle of $\Sigma$ is equivalently encoded in a morphism of smooth stacks

$$ \tau_\Sigma : \Sigma \longrightarrow \mathbf{B}GL(n) \,. $$

Since $\Pi(\mathbf{B}GL(n)) \simeq B GL(n) \simeq B O(n)$ this is such that $\Pi(\tau_\Sigma)$ is the map on bare homotopy types considered above.

Now stack homomorphisms in the higher slice topos over $\mathbf{B}GL(n)$ are equivalently local diffeomorphisms

$$ \mathrm{LocalDiffeos}(\Sigma) \simeq \mathrm{SmoothStacks}_{/\mathbf{B}GL(n)}(\Sigma, \Sigma) \,. $$

Hence when considering the diffeomorphism group of $\Sigma$ we may equivalently think of the slice automorphism group of $\tau_\Sigma$ over $\mathbf{B}GL(n)$.

From this perspective the question is to which extent one may "take $\Pi$ into the internalized slice automorphism group" to pass from

$$ \Pi(\mathbf{Aut}_{/\mathbf{B}GL(n)}(\Sigma)) = \Pi (\mathrm{Diff}(\Sigma)) $$


$$ \mathrm{Aut}_{/\Pi(\mathbf{B}GL(n))}(\Pi(\Sigma)) = \mathrm{Aut}_{/B O(n)}(\Pi(\Sigma)) \,. $$


I wanted to say I think this is a great question, though phrasing things in terms of stacks might scare off some of the people who can best answer this question. I think in general understanding the relationship between these two spaces is a very interesting and very hard problem.

Here is an example which shows that they can definitely be different.

3-dimensional Lens spaces $L(p,q)$ have trivial tangent bundles, so the group you write $Aut_{/BO(3)}(\Pi(\Sigma))$ will just be the homotopy automorphism group of the homotopy type $\Pi(\Sigma)$. So if they are homotopy equivalent, then they are homotopy equivalent over $BO(3)$.

Now we look at the classification of lens spaces (see wikipedia for example) and we see that $L(7,1)$ and $L(7,2)$ are lens spaces which are homotopy equivalent but which are not diffeomorphic. So for these manifolds we have:

$$BAut_{/BO(3)}(\Pi(L(7,1)) \simeq BAut_{/BO(3)}(\Pi(L(7,2))$$

Now what about their diffeomorphism groups? Well we can look at arXiv:math/0411016 and we see that as spaces:

$$Diff(L(7,1)) \simeq S^1 \times S^3 \sqcup S^1 \times S^3 $$

$$Diff(L(7,2)) \simeq S^1 \times S^1 \sqcup S^1 \times S^1 $$

which implies that $BDiff(L(7,1)) \not\simeq BDiff(L(7,2))$.

So this shows that for at least one of these spaces the classifying space of the diffeomorphism group differs from the classifying space of the tangential homotopy automorphism group.

  • $\begingroup$ Thanks a lot for taking the time to compile this excellent argument. That saves me from walking down a dead-end. $\endgroup$ – Urs Schreiber Oct 24 '14 at 9:02
  • 5
    $\begingroup$ Against my better judgement I will use the notation from the question. Just because an $n$-manifold $M$ has trivial tangent bundle doesn't mean that $Aut_{/BO(n)}(\Pi(M)) \to Aut(\Pi(M))$ is an equivalence. But that doesn't affect Chris' argument, as $L(7,1)$ and $L(7,2)$ are still homotopy equivalent over $BO(3)$, so $Aut_{/BO(3)}(\Pi(L(7,1))) \simeq Aut_{/BO(3)}(\Pi(L(7,2)))$. $\endgroup$ – Oscar Randal-Williams Oct 24 '14 at 14:02
  • $\begingroup$ Let me see if I understand the difference. If I look at $Aut_{/BO(n)}(pt)$, then I should get $O(n)$ and not a point. Is that right? $\endgroup$ – Chris Schommer-Pries Oct 24 '14 at 15:05
  • $\begingroup$ That's right (you have to make $* \to BO(n)$ fibrant). $\endgroup$ – Oscar Randal-Williams Oct 24 '14 at 15:19
  • $\begingroup$ If you allow, I have a followup question here: mathoverflow.net/q/185441/381 $\endgroup$ – Urs Schreiber Oct 26 '14 at 17:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.