5
$\begingroup$

Let $E$, $M$ be smooth finite dimensional, Hausdorff and second-countable manifolds. Let $\pi:E \longrightarrow M$ be a surjective submersion.

$\pi$ is locally trivial if $\forall p\in M$, $\exists U \ni p$ open neighborhood such that there is a diffeomorphism: $$ \phi:\pi^{-1}(U)\longrightarrow U \times \pi^{-1}(p)$$ $$ \text{proj}_2\circ \phi=\pi $$

I read that if $\pi$ is proper then is locally trivial (Ehresmann theorem). But not every locally trivial submersion is proper (vector bundles for example).

Is there any sufficient and necessary condition for a surjective submersion to be locally trivial?

$\endgroup$
1
  • 3
    $\begingroup$ Here is a fun sufficient criterion: if the fibers of dimension $\neq 4$ then it suffices that the submersion is locally trivial as a topological manifold bundle for it to be locally trivial as a smooth manifold bundle. This is Essay II of Kirby-Siebenmann. $\endgroup$
    – skupers
    Commented Nov 10, 2020 at 22:38

2 Answers 2

3
$\begingroup$

A connection for a (surjective) submersion $\pi\colon E\to M$ is a complementary subbundle $\mathcal H E\subset TE$ of the vertical bundle $\mathcal VE=ker (D\pi).$ A connection locally defines a parallel transport in $E$ along curves $\gamma\colon I\to M$, by lifting to a horizontal curve, i.e., $$\hat\gamma\colon \tilde I\to E;\; \hat\gamma'(t)\in\mathcal HE_{\hat\gamma(t)}.$$ As opposed to the case of linear connections on vector bundles (or principal connections on principal bundles), the parallel transport does not always exist globally on the interval of definition $T$ for $\gamma$, but only exists on relatively open sub-intervals $\tilde I\subset I.$ A connection is called complete if every curve $\gamma$ admits a global horizontal lift. I think the following observation can be attributed to Ehresmann: A surjective submersion is locally trivial if and only if it admits for all $p\in M$ an open neighbourhood $U$ of $p$ and a connection $\mathcal HE$ which is complete whence restricted to $U\subset M$. In fact, you can use the parallel transport to construct diffeomorpisms $\phi$ as wanted.

$\endgroup$
3
  • $\begingroup$ Thanks! By the word fibration do you mean surjective submersion? In topology it means something different. I find this observation very beautiful. In Michor's book, there's a proof that all the fiber bundles (locally trivial surjective submersions) have complete connections. I think we can recover a complete connection from these complete locally connections. I was thinking if there is on another criteria similar than $\pi$ to be proper, but weaker. $\endgroup$ Commented Nov 11, 2020 at 9:51
  • $\begingroup$ Yes, I meant surjective submersion. I would also guess that one can recover a complete connections by patching locally complete connections, but I haven't looked at the details. I will have a look into Michor's book. The one proof of Ehresmann theorem I am aware of uses connections, and the simple observation that any connection on a proper surjective submersion is complete. $\endgroup$
    – Sebastian
    Commented Nov 11, 2020 at 10:10
  • 1
    $\begingroup$ The proof in Michor's book (just like the other known proofs in the literature) has a critical gap as pointed out by Matias del Hoyo in Complete connections on fiber bundles, Indagationes Mathematicae, Volume 27, Issue 4, 2016, Pages 985-990. He provides an alternative proof for this result (the result is true). The issue with the existing proofs is that convex combinations of complete Ehresmann connections or fibred Riemannian metrics need not be complete. Del Hoyo gives counterexamples for these assertions. His alternative construction is well worth a detailed read! $\endgroup$ Commented Oct 10, 2023 at 23:52
3
$\begingroup$

The following differential-topological characterization is Theorem B in Meigniez, Gaël. Submersions, fibrations and bundles. Trans. Amer. Math. Soc. 354 (2002), no. 9, 3771--3787. It is currently available in the website of the author: http://web.univ-ubs.fr/lmba/meigniez/docu/preprints/sfb.pdf (the page is http://web.univ-ubs.fr/lmba/meigniez/docu/travaux.html, in case the link to the pdf changes at some point). I do not really know the proof, as I only skimmed through the paper some years ago, but I remember that the paper looked very interesting to me.

Let the dimension of $M$ be $m$ and let the dimension of $E$ be $n+m$.

Theorem A surjective smooth submersion $\pi:E\to M$ is a (locally trivial fibre) bundle if and only if it admits an exhaustive, isotopy invariant, $(m−1)$-fibred family of vertical domains.

To understand the statement, we need some definitions (taken from Section II.1 of the same reference):

  • A vertical domain is an $n$-dimensional compact submanifold of a fibre, $X\subset E_p$, with a smooth boundary.

Let $VE(X,E)$ denote the space of vertical embeddings of $X$ into $E$ (so, those with image in a fibre), with the topology of smooth uniform convergence. Let $VE^0(X,E)$ denote the connected component of $VE(X,E)$ containing the original inclusion $X\to E_p$.

Let $VD=\coprod_{p\in M} VD_p$ be a family of vertical domains (Note that each $VD_p$ is itself a collection of vertical domains, but all in the same fibre $E_p$). It is called:

  • exhaustive if every compact subset of every fibre is contained in some $X\in VD$;

  • isotopy invariant if for every $X\in VD$ and every $\phi \in VE^0(X,E)$ we have $\phi(X)\in VD$;

  • $r-$fibred if, for any two domains $X,X′\in VD_p$ such that $X\subset Int(X′)$, the restriction map

$$\rho_{X,X′}:VE^0(X′,E)\to VE^0(X,E)$$ is an $r$-fibration (i.e., has the homotopy lifting property for polytopes of dimension at most $r$).

In the same paper there is discussion of some other sufficient conditions for local triviality, see II.1,applications, and also on conditions for a surjective submersion to be a fibration, i.e., satisfying homotopy lifting properties (part I of the paper). For example, the following easier (i.e. with conditions possibly easier to verify) characterization is Corollary 28 in the paper:

Corollary (dim-1 base principle for bundles)

a) A surjective submersion $\pi:E→M=R^m$ is a bundle if and only if it is a bundle over each straight line in the base parallel to one of the axes.

b) A surjective submersion $\pi:E→M$ is a bundle if and only if for every smooth path $\gamma:[0,1]→M$, the pullback $\gamma^*\pi:\gamma^*E→[0,1]$ is a bundle.

c) The theorem is still true if we change "$(m−1)$-fibred" to "$0$-fibred".

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .