I can't see a good approach to the third part of the following problem:

Let $f: M \to M$ be a smooth map of a compact oriented manifold into itself. Denote by $H^q(f)$ the induced map on the cohomology $H^q(M)$. The Lefschetz number of f is defined to be

$$L(f) = \sum_q (-1)^q \text{trace } H^q(f)$$

Let $\Gamma$ be the graph of $f$ in $M\times M$

  1. Show that $\int_\Delta \eta_\Gamma = L(f)$
  2. Show that if $f$ has no fixed points, then $L(f)$ is zero.
  3. At a fixed point $P$ of $f$ the derivative $Df_p$ is an endomorphism of the tangent space $T_pM$. We define the multiplicity of the fixed point $P$ to be

$$ \sigma_P = \text{sign} \, \det(Df_p - I)$$

Show that if the graph $\Gamma$ is transversal to the diagonal $\Delta$ in $M \times M$, then

$$L(f) = \sum_P \sigma_P$$

where $P$ ranges over the fixed points of $f$. Here $\eta_S$ denotes the Poincaré dual of the submanifold $S$.

I'd be glad, if someone could help me out. Best would be only a hint and not an entire solution.


I guess one might somehow use that a form representing the Poincaré dual of a submanifold $S$ can be chosen to have support in an arbitrarily small neighborhood of $S$ to turn the problem into a local problem of computing integrals in neighborhoods of the fixed points of $f$, noting that

$$\int_\Delta \eta_\Gamma = \int_{M\times M} \eta_\Gamma\wedge \eta_\Delta$$

and $\eta_\Gamma\wedge \eta_\Delta$ would then be nonzero only in a arbitrarily small neighborhood of the intersections of $\Delta$ and $\Gamma$, i.e. the set of fixed points of $f$, so that one could work in local coordinates there. However, what I would find particularly strange is how -- if we could compute local integrals as suggested above -- they would turn out to be so nicely integervalued? But I don't get far trying to push this idea further...

I also actually don't really know what $\eta_\Gamma$ looks like. I've tried figuring out some expression for it, but without luck.

As usual, many thanks for your always helpful suggestions!

Best regards,



I think, I've got it now. One can use the fact that $η_\Delta$ is just the Thom class of the normal bundle of $\Delta$ and by transversality, this implies that the normal bundle of $\Delta$ is the tangent bundle of $\Gamma$ at the intersection points and vice versa. So integration boils down to integration along the fiber.

This gives that the integrals around the intesection points have values $\pm 1$, depending on $sign(\det(Df_p - I))$. (Using the idea already mentioned after the question)


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.