2
$\begingroup$

Let $F\stackrel i \to E\stackrel \pi\to S^1$ a fiber bundle over the circle $S^1$. There is a long exact sequence sequence in cohomology, called Wang: $$\dots\to H^k(E)\stackrel {i^*}\to H^k(F)\stackrel {f^*-I}\to H^k(F)\stackrel \delta\to H^{k+1}(E)\to \dots ,$$ where $I$ is the identity and $f^*$ is induced by the monodromy of the bundle.

In the case of a bundle over $S^n$, $n>1$, the long exact sequence above follows from the Serre spectral sequence (see Wikipedia). However, using the presented argument, one obtains: $$\dots\to H_q(E)\to E^1_{1,q-1}\stackrel d\to E^1_{0,q-1}\to H_{q-1}(E)\to\dots~.$$ Thus in order to conclude the Wang sequence one must identify the first page and the boundary map of the Serre spectral sequence. I hoped to avoid this.

Another way of proving the result was mentioned in this mathoverflow answer.
The strategy is to apply Mayer-Vietoris to the bundle. As I didn't do an argument like this before, I did not understand how to use monodromy.

I would appreciate if anyone could point me in the right direction in either of the two arguments.

$\endgroup$
  • 1
    $\begingroup$ On behalf of @Gary Kennedy: An elementary argument is outlined in Lemma 8.4 of Milnor's book Singular Points of Complex Hypersurfaces. $\endgroup$ – dantopa Apr 17 at 20:58
6
$\begingroup$

Here is the Mayer-Vietoris argument :

Let $N,W,S,E$ be the north, west, south and east pole on the circle. Let $U=S^1\setminus\{E\}$ and $V=S^1\setminus \{W\}$. This is an open covering of $S^1$ such that $U$ and $V$ are contractible so that $\pi^{-1}(U)$ is homeomorphic to $F\times U$ and even homotopicaly equivalent to $F\times\{N\}$. Similarly $\pi^{-1}(V)\simeq F\times\{S\}$ and $\pi^{-1}(U\cap V)\simeq F\times\{N\}\cup F\times\{S\}$.

Now write the Mayer-Vietoris exact sequence associated to the covering of the total space $E$ by $\pi^{-1}(U)$ and $\pi^{-1}(V)$. This is : $$...\rightarrow H^i(E)\rightarrow H^i(\pi^{-1}(U))\oplus H^i(\pi^{-1}(V))\rightarrow H^i(\pi^{-1}(U \cap V))\rightarrow H^{i+1}(E)\rightarrow ...$$ Using the above homotopy equivalence, this is : $$...\rightarrow H^i(E)\rightarrow H^i(F\times\{N\})\oplus H^i(F\times\{S\})\rightarrow H^i(F\times\{N\})\oplus H^i(F\times\{S\})\rightarrow H^{i+1}(E)\rightarrow ...$$ But what are the maps between these four copies of $H^i(F)$ ? So this enough to give the maps $H^i(F\times\{N\})\to H^i(F\times\{N\})\oplus H^i(F\times\{S\})$ and $H^i(F\times\{S\})\to H^i(F\times\{N\})\oplus H^i(F\times\{S\})$.

For the first one, remember that this map comes from the restriction $H^i(\pi^{-1}(U))\to H^i(\pi^{-1}(U\cap V))$. So this the identity on $H^i(F\times\{N\})$ and a map $u:H^i(F\times\{N\})\to H^i(F\times\{S\})$ which intuitively means "cohomology class lying on $F\times\{N\}$ that have been moved to $F\times\{S\}$ along the path from the north pole to the south avoiding the east (because $U=S^1\setminus\{E\}$). You can prove this more rigorously by writing down all the diagrams with all the maps, including the homotopy equivalence. I will not do it here.

Similarly the second map $H^i(F\times\{S\})\to H^i(F\times\{N\})\oplus H^i(F\times\{S\})$ is the identity on the second component and a map $v:H^i(F\times\{S\})\to H^i(F\times\{N\})$ which intuitively moves the cohomology classes on $F\times\{S\}$ to $F\times\{N\}$ along the path from the south pole to the north which avoids the west.

So the map in the long exact sequence looks like $$H^i(F\times\{N\})\oplus H^i(F\times\{S\})\xrightarrow{\begin{pmatrix}1&v\\u&1 \end{pmatrix}} H^i(F\times\{N\})\oplus H^i(F\times\{S\})$$

The point is that there is an extra copy $H^i(F)$ on each side here and we would like to remove it. So let us have a look at its kernel and its cokernel :

  • the kernel is the set of couple of classes $(x,y)$ such that $x+v(y)=0$ and $u(x)+y=0$. So this is the same as the classes $(x,-u(x))$ where $x=vu(x)$. But $vu$ is exactly the monodromy : we take a class and move it along a path going from the north to the south pole avoiding the east, and then from the south to the north avoiding the west, so a whole loop. Hence the kernel is $\ker(f^*-I)$
  • similarly, the cokernel is the set of couple $(x,y)$ modulo the relation $(x,y)=0$ iff $(x,y)=(a,u(a))+(v(b),b)$. But the map $(x,y)\mapsto x-v(y)$ induces an isomorphism $\operatorname{coker}\begin{pmatrix}1&v\\u&a\end{pmatrix}\simeq \operatorname{coker}(f^*-I)$.

Thus, you can replace the map by $f^*-I:H^i(F\times\{N\})\to H^i(F\times\{N\})$. Of course $H^i(F\times\{N\})$ is just $H^i(F)$ and you get the Wang sequence.

Note that to be more precise, one need to work at the level of complexes. But the argument is the same.

$\endgroup$
  • $\begingroup$ Thank you for your detailled and clear answer! $\endgroup$ – klirk Nov 17 '18 at 17:10

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.