8

Let $M$ be a manifold and $V$ be an oriented vector bundle. It's well known that if the Euler class of $V$ is non zero, then $V$ can't have a non-vanishing section. The converse is not true, see Vanishing of Euler class , or the answer by John Klein below, given to the first version of this question. So:

Question. Recall that the normal bundle of a smooth orieantable submanifold in $\mathbb R^n$ always has zero Euler class. Does such a normal bundle always has a non-vanishing section? If yes, how to prove this? If no, what is a counterexample?

PS. It turns out that this question is a subquestion of the following much more informed one: Embeddings without nonvanishing normal vector fields

aglearner
  • 13,995

2 Answers2

9

Here's a counter example. Take any embedding of $\mathbb{C} P^2$ in $\mathbb{R}^7$ (such embeddings exist by

Steer, B., On the embedding of projective spaces in Euclidean space, Proc. Lond. Math. Soc., III. Ser. 21, 489-501 (1970). ZBL0206.25501;

the construction is summarised in this answer on MSE).

If such an embedding admits a normal vector field, then by the Compression Theorem of Rourke and Sanderson it is isotopic to an embedding with normal field parallel to the last coordinate of $\mathbb{R}^7=\mathbb{R}^6\times\mathbb{R}$, and then the projection is an immersion $\mathbb{C} P^2\looparrowright \mathbb{R}^6$. Such immersions cannot exist, by András Szűcs' answer here.

Mark Grant
  • 35,004
  • Thanks a lot Mark, that's really cute. May I ask you to comment on the answer of Andras? He writes that if there exists an immersion of $\mathbb CP^2$ to $\mathbb R^6$ then "the normal Euler class has the property, that its square is the normal Pontrjagin class, and that is $−3$ times the signature". I am rusty with Pontriagin classes so can't understand this line... – aglearner Dec 05 '18 at 10:09
  • 2
    @aglearner: Thanks! Andras is using 3 facts about Pontrjagin classes, all of which can be found in Chapter 15 of Milnor and Stasheff: The square of the Euler class of a $2k$ dimensional bundle equals $p_k$; the Pontragin classes satisfy a sum formula $p(E\oplus F) = p(E)p(F)$ modulo $2$-torsion; and $p_1(\mathbb{C}P^2) = 3a^2$, where $a\in H^2(\mathbb{C} P^2)$ is the generator. – Mark Grant Dec 05 '18 at 11:40
  • 2
    (The sum formula implies that $p_1(\nu)=-3a^2$ where $\nu$ is the hypothetical normal bundle which has $\nu\oplus T\mathbb{C} P^2$ a trivial bundle.) – Mark Grant Dec 05 '18 at 11:44
  • Thanks again, Mark, really happy to learn all this! – aglearner Dec 05 '18 at 13:18
6

There are lots of counter-examples.

Here's one: The fibration $\text{SO}(3) \to \text{SO}(4) \to S^3$ splits, since it is the principal bundle of the tangent bundle of $S^3$, and the latter is parallelizable. In particular, $\pi_5(\text{SO}(4)) \cong \pi_5(\text{SO}(3)) \oplus \pi_5(S^3) \cong \Bbb Z/2 \oplus \Bbb Z/2$ .

Let $a$ and $b$ denote the generators for the summands. Then $(a,b)$ determines a rank $4$-vector bundle over $S^6$ with trivial Euler class (since $H^4(S^6) = 0$). But this vector bundle does not have a section, for if it did that would lead to the contradiction that $b=0$.

John Klein
  • 18,574
  • Thank you John. I edited the question, because I do want to know the answer to a part of the original question, which was not covered in the previous question on mathoverflow. – aglearner Dec 04 '18 at 12:24
  • @aglearner, is your question about a submanifold in an arbitrary manifold or a submanifold of $\mathbb{R}^n$? – Deane Yang Dec 04 '18 at 21:05
  • 3
    Dean, my original question was more general, and John immediately gave a counterexample, which illustrates that vanishing of the Euler class doesn't guarantee existence of a non-vanishing section. However, since I really want the answer to my question above (about submanifolds of $\mathbb R^n$), I decided to modify the original question. So now John's answer doesn't answer the current question, but it nicely illustrates it. – aglearner Dec 04 '18 at 21:29