I understand that the top Stiefel Whitney class is an obstruction for the tangent bundle of a manifold to have a trivial line sub-bundle. I am looking for a counterexample when removing the word "trivial", i.e: A compact manifold $M$ of dimention n such that $w_n(TM)\neq0$ (or equivalently $\chi(M)$ is odd) and there exists a sub-bundle $\xi\subset TM$ with $\operatorname{rank}(\xi)=1$.

After some thought I understand why one cannot find an example using surfaces, odd-dimensional manifolds or any of $S^n$,$T^n$,$\mathbb{R}P^n$,$\mathbb{C}P^n$. I also understand why manifolds with $H^1(M;\mathbb{Z}_2)=0$ will not work, and why such a manifold (if it exists) cannot be null-cobordant.

My intuition on 4-manifolds (or god forbid anything higher dimensional) is not good enough to know where to look. Any insight as to where one should look for such a creature (or why the hell it should not exist) will be greatly appreciated.


I believe there is no example satisfying all your constraints. If I recall (my memory is a little foggy on this) the result likely goes back to Hopf, and one of his variations on the Poincare-Hopf index theorem. This question might be addressed in the Milnor and Stasheff text. Here is one way to argue the point.

Say the tangent bundle of the manifold $N$ admits a field of lines. Then (at worst) some 2:1 cover of $N$ admits an everywhere non-vanishing vector field. Since the Euler characteristic is the obstruction to such a vector field existing (Poincare-Hopf index theorem) the Euler characteristic of this covering space is zero. But $\chi N$ is a multiple of the euler characteristic of the cover.

i.e. your assumption that the Euler characteristic is odd excludes the possibility of a 1-dimensional sub-bundle.

  • $\begingroup$ Thank you for this simple answer! I feel so silly I missed this. $\endgroup$ Jul 31 '17 at 12:20
  • $\begingroup$ This is the kind of question you'd hope would be answered in most texts that cover Poincare-Hopf but it seems to get lost too often. $\endgroup$ Jul 31 '17 at 14:57
  • $\begingroup$ Definitely! It's quite odd that alot of texts bother proving that if $\chi(M)=0$ then there exists a nonvanishing vector field (which if I recall correctly is abit involved as you need to push all zeros of a vector field to a small ball by a diffeomorphism) but not this simple thing. $\endgroup$ Jul 31 '17 at 20:40
  • $\begingroup$ Can your answer be generalized to a noncompact manifold in the following sense: A manifold $M$ (not necessarily compact) which admits a field of lines must have a non-vanishing vector field? Poincare hopf heavily uses compactness but I can't seem to prove or disprove this. $\endgroup$ Aug 7 '17 at 11:27
  • 1
    $\begingroup$ @AnonymousCoward: That's the idea. I must admit I couldn't immediately find a reference where the details of the argument are written out, but the intuition is clear. If you need a proof for some reason, you could use obstruction theory: the only obstruction to sectioning the sphere tangent bundle of $M^n$ lives in $H^n(M;\pi_{n-1}(S^{n-1}))$, a group which is zero if $M$ is non-compact. $\endgroup$
    – Mark Grant
    Aug 17 '17 at 10:05

Let $\gamma$ be the canonical (real) line bundle over $RP^2$. In other words, $w_1(\gamma)\neq 0$. The total Stiefel-Whitney class of the Whitney sum $\gamma\oplus\gamma$ is $(1+w_1(\gamma))^2=1+w_1(\gamma)^2$. So $w_2(\gamma\oplus\gamma)=w_1(\gamma)^2$ which is nonzero.

  • 3
    $\begingroup$ $\gamma\oplus\gamma$ is not isomorphic to $T\mathbb{R}P^2$ (and neither is $\gamma\oplus1$) as they don't even have the same first Stiefel-Whitney class. In general every real projective space will fail to satisfy my demands as $TS^n$ admits a line sub-bundle only if $n$ is odd, in which case $w_n(\mathbb{R}P^n)=0$ (notice that one can pull back a line sub-bundle of $T\mathbb{R}P^n$ to a line sub-bundle of $TS^n$). $\endgroup$ Jul 31 '17 at 1:54
  • 2
    $\begingroup$ No surfaces will do either, as the only closed surfaces $M$ which admit a line sub-bundle of $TM$ are the torus and the Klein bottle (because such a surface (if not orientable) would have an orientable 2-cover which admits a nowhere vanishing vector field i.e: is a torus), both of which have $\chi(M)=0$ and hence $w_2(TM)=0$. $\endgroup$ Jul 31 '17 at 2:03
  • $\begingroup$ Oh, I missed that you wanted the bundle to be tangent. $\endgroup$ Jul 31 '17 at 2:10

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.