Say we have hypersurface $D \subset \mathbb{P}^2$. I ran into the following sequence of isomorphisms, justified only with "by Poincare duality formulated algebro-geometrically": $$ H^2(\mathbb{P}^2-D,\mathbb{Q}) \cong H^3_D(\mathbb{P}^2,\mathbb{Q}) \cong H_1(D,\mathbb{Q}) $$ I looked up cohomology with supports, and I could only find something like $H^k_c(X) \cong H^{n-k}(X)^*$. Could you help me with either a reference or an explenation of why the above hold? Thank you.

  • $\begingroup$ This is usually called a purity isomorphism. The way I see it with its connection to Poincaré duality is through the six functors formalism and Verdier duality. Are you familiar with it ? $\endgroup$ – Roland Apr 11 '18 at 16:00
  • $\begingroup$ Actually you don't need the six functors formalism, nor Verdier duality. Just localization sequences in cohomology with compact support, and/or Borel Moore homology. Can you say what tools you have at your disposal so that I can adapt my answer ? $\endgroup$ – Roland Apr 11 '18 at 16:27
  • $\begingroup$ @Roland I don't have much more at my disposal then the sheaf cohomology that can be found in Hartshorne. $\endgroup$ – baltazar Apr 11 '18 at 16:29
  • $\begingroup$ Ok, I have remove mine. I will try several answers with different tools then. $\endgroup$ – Roland Apr 11 '18 at 16:32

There is two isomorphisms in your question, the second one is indeed a form of Poincaré duality/purity isomorphism. The first one is on the contrary specific to the situation. I will explain both of them.

First let us talk about the Poincaré duality. Every homology/cohomology groups will be with rational coefficients, so I won't write them.

If $Z\subset X$ is a closed subset and $U$ is its complement, then there are localization long exact sequences in cohomology with compact support, with Borel-Moore homology and cohomology : $$ ...\to H^i_c(U)\to H^i_c(X)\to H^i_c(Z)\to H^{i+1}_c(U)\to...$$ $$ ...\to H^{BM}_{d-i}(Z) \to H^{BM}_{d-i}(X) \to H^{BM}_{d-i}(U)\to H^{BM}_{d-i-1}(Z)\to...$$ $$ ...\to H^i_Z(X)\to H^i(X)\to H^i(U)\to H^{i+1}_Z(X)\to...$$

A version of Poincaré duality says that for an oriented manifold of (topological) dimension $d$, then $H^i=H^{BM}_{d-i}=(H_c^{d-i})^*$. In the case where $X,U,Z$ are smooth, then the second exact sequence is the dual of the first.

But we also have an identification between the second and the third : there is a relative Poincaré duality $H^i_Z(X)=H^{BM}_{d-i}(Z)$. Together with this relative duality, the second and the third exact sequence are the same.

This relative Poincaré duality gives in your specific case, you get $H_1(D)=H^3_D(\mathbb{P}^2)$.

So here is a "derived" variant of the previous isomorphism, and what is usually called the purity isomorphism.

Let $i:Z\rightarrow X$ and $j:U\rightarrow X$ be the closed and open immersion. Let $f:X\to pt$ be the canonical morphism. Then, in the derived category of abelian groups, there are isomorphisms : $$ H^i(X)=Hom_D(\mathbb{Q},Rf_*f^*\mathbb{Q}[i])$$ $$H^{BM}_{d-i}(X)=Hom_D(\mathbb{Q},Rf_*Rf^!\mathbb{Q}[i-d])$$ $$ H^i_c(X)=Hom_D(\mathbb{Q},Rf_!f^*\mathbb{Q}[i])$$

And the long exact sequence above comes from the distinguished triangles $$ i_*Ri^!\to 1\to Rj_*j^*\overset{+1}\to$$ $$ j_!j^*\to 1\to i_*i^*\overset{+1}\to$$ Now the version of Poincaré-Verdier duality states that for an oriented manifold of dimension $d$, we have an isomorphism of functor $Rf^!=f^*[d]$. This gives in particular the isomorphism $H^i=H^{BM}_{d-i}$.

If $X$ and $Z$ are smooth and $Z$ of codimension $c$, then $$i^!\mathbb{Q}_X[d]=i^!f^!\mathbb{Q}=(fi)^!\mathbb{Q}=\mathbb{Q}_Z[d-c]$$ Hence $i^!\mathbb{Q}=\mathbb{Q}[-c]$. This holds isomorphism is called a purity isomorphism in algebraic geometry.

Now let us have a look at the first isomorphism $H^2(\mathbb{P}^2-D)\simeq H^3_D(\mathbb{P}^2)$. We will apply the long exact sequences with $X=\mathbb{P}^2$ and $Z=D$ a smooth hypersurface. Since they are proper, $H^i=H^i_c$ and $H^{BM}_i=H_i$. The localization sequence in (say Borel Moore homology) is then :

$$ ...\to H_2(D)\to H_2(\mathbb{P}^2)\to H_2^{BM}(\mathbb{P}-D)\to H_1(D)\to H_1(\mathbb{P}^2)\to... $$ But now, by Poincaré dualiy (with $d=4$) this is : $$ ...\to H^2_D(\mathbb{P}^2)\to H^2(\mathbb{P}^2)\to H^2(\mathbb{P}-D)\to H^3_D(\mathbb{P}^2)\to H^3(\mathbb{P}^2)\to... $$

Now, since we are working with rational coefficient, $[D]\in H_2(D)$ is sent to a generator of $H_2(\mathbb{P}^2)$ so the map $H_2(D)\to H_2(\mathbb{P}^2)$ is onto, and since $H_1(\mathbb{P}^2)=0$, we have the isomorphism $H^2(\mathbb{P}^2-D)\simeq H_1(D)\simeq H^*_D(\mathbb{P}^2)$

  • $\begingroup$ Thank you for the detailed reply! Do you happen to have some references for these things? I haven't dealt with derived categories before. I easily found a definition but in this context I get a bit lost. $\endgroup$ – baltazar Apr 12 '18 at 10:52

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.