Let $\pi:E\to X$ be a (smooth) vector bundle, and let $q\in X$. First of all, I have to find a canonical isomorphism $T_{(q,0)}E\cong T_qX\times E_q$. I proceed as follows.

Since $\pi$ is a submersion, we have an exact sequence (to ease the notation I drop the subscript on $d\pi$, it is taken at $(q,0)$) \begin{equation} 0\longrightarrow\ker(d\pi)\overset{\iota}{\longrightarrow}T_{(q,0)}E\overset{d\pi}{\longrightarrow}T_qX\longrightarrow 0 \end{equation} which is split since the zero section $s:X\to E,q\mapsto(q,0)$ yields a map $ds_q$ such that $id_{T_qX}=d\pi\circ ds_q$. Hence the map $\tilde\iota:T_{(q,0)}E\longrightarrow,v\mapsto v-ds_q(d\pi(v))$ gives us a map \begin{equation} \psi:T_{(q,0)}E\longrightarrow T_qX\times\ker(d\pi),v\mapsto(d\pi(v),v-ds_q(d\pi(v))) \end{equation} defines an isomorphism.

However, I am not sure how to construct an isomorphism from $\ker(d\pi)$ to $E_q$. Intuitevely it is clear to me that there is an isomorphism, since the vectors tangent to $\pi^{-1}(q)$ are "vertical" and hence projected down to 0, but I don't know how to construct it.

I need it because next I need to show that under this isomorphism the canonical symplectic form (defined as $\omega^{can}=-d\lambda^{can}$, with $\lambda^{can}_{(q,p)}=p\circ d\pi_{(q,p)}$ for any $(q,p)\in T^\ast X$) agrees with the canonical symplectic form $\omega_{T_qX}$ on $T_qX\times (T^\ast_qX)^\ast$, which is given by $\omega_{T_qX}((a,\phi),(b,\psi))=\psi(a)-\phi(b)$.

This is so far not working for me, and I feel like I'm simply not quite unraveling the definitions correctly and really getting confused at the technicalities. Any help or clarification is much appreciated.

EDIT If you consider the point $(q,0)\in T^\ast X$, why is $\lambda^{can}$ and thus $-d\lambda^{can}$ not trivial at this point? Since you act after everything with the zero map.


1 Answer 1


The isomorphism $ker(d\pi)\to E_q$ is easier to construct in the other direction. Given an element $v\in E_q$, consider $t\mapsto t\cdot v$ for $t\in\mathbb R$. This defines a smooth curve $\mathbb R\to E$, which actually has values in the fiber $E_q$. Taking the derivative at $t=0$ of this curve, one thus obtains a tangent vector in $T_{(q,0)}E$, which by construction lies in $\ker(\pi)$. Now you easily see that this defines a linear injection $E_q\to \ker(\pi)$, which must be an isomorphism since both spaces have the same dimension. (This also works outside the zero section.)

Concerning your Edit: It is true that $\lambda^{can}$ vanishes along the zero section, but this says nothing about its exterior derivative. Compare with the standard coordinate expression $\lambda^{can}=\sum p_idq_i$ vs. $d\lambda^{can}=\sum dp_i\wedge dq_i$ in points where all $p_i$ are zero.


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.