7

For each $i\geq 0$, describe $H^i(X,\mathcal{O}_X)$ where $X:=\text{Spec}(k[x,y, z])\setminus\{(x,y,z)\}$.

The first definition of cohomology I've learned involves injective resolutions, which I have no idea how to apply here.

I've read some authors who claimed that Cech cohomology is often useful to compute sheaf cohomology in real life, so I decided to take that road.

If $A:=k[x,y,z]$ and $Y:=\text{Spec}(A)$, I've thought about using the affine open cover $U_x:=Y\setminus V(x)$, $U_y:=Y\setminus V(y)$ and $U_z:=Y\setminus V(z)$.

That way, $\mathcal{O}_X(U_x)=A_x$, $\mathcal{O}_X(U_y)=A_y$ and $\mathcal{O}_X(U_z)=A_z$ and consequently: $$C^0:=C^0(\mathcal{O}_X)=A_x\times A_y\times A_z$$ $$C^1=A_{xy}\times A_{xz}\times A_{yz}$$ $$C^2=A_{xyz}$$

I've checked that $\left(\frac{a}{x^n},\frac{b}{y^m},\frac{c}{z^\ell}\right)\in\ker(d^0)\Leftrightarrow\frac{a}{x^n}=\frac{b}{x^m}=\frac{c}{z^\ell}\in k[x,y,z]$, so $\ker(d^0)\simeq A$ and $H^0(X,\mathcal{O}_X)=A$.

Now, determining $H^1(X,\mathcal{O}_X)=\ker(d^1)/\text{im}(d^0)$ and $H^2(X,\mathcal{O}_X)=C^1/\text{im}(d^1)$ is considerably more complicated and made me wonder whether or not this is the best option.

I've also tried the simpler case $\text{Spec}(k[x,y])\setminus\{(x,y)\}$, and even then I found hard to describe $H^1(X,\mathcal{O}_X)$ explicitly.

This looks like a standard problem, so I can't help but wonder if there isn't a simpler approach which could work even for the general case $\text{Spec}(k[x_1,...,x_n])\setminus\{(x_1,...,x_n)\}$.

rmdmc89
  • 9,939
  • 5
    Your strategy ought to work and would be the usual way to attack this at the level of being relatively new to cohomology. – KReiser Jun 22 '20 at 00:27
  • @KReiser, is that a good idea also for the general case $\text{Spec}(k[x_1,...,x_n])\setminus{(x_1,...,x_n)}$? – rmdmc89 Jun 22 '20 at 01:07
  • 2
    That's what I intended to say. If I remember this computation correctly, there's no extra trouble with using this same strategy for any $n$. (I'm not saying that this is going to be a one-liner or anything, just that the essential ideas of the proof don't change. I'm also not saying that this is "easy", but you can just sit down and slog through it, and then you'll know the result forever and not need to reprove it, unless someone asks you on MSE.) – KReiser Jun 22 '20 at 02:33
  • 1
    @rmdmc89 non-affine open subset of affine space are pretty pathological. I don't know if you should expect a "trick" for calculating the cohomology. – hunter Jun 22 '20 at 20:57
  • @hunter, I didn't expect a trick exactly, but maybe a more suiting method. Since all I know is injective resolutions vs. Cech cohomology, I thought I should ask – rmdmc89 Jun 22 '20 at 21:04
  • @rmdmc89 makes sense! I think Cech is the way to go. – hunter Jun 22 '20 at 21:43

1 Answers1

2

Here is the general result:
Let $X=\operatorname {Spec}k[X_1,\cdots, X_n]\setminus \{\langle X_1,\cdots,X_n\rangle\}=\mathbb A^n_k\setminus \{O\}$. Then:

  1. For $n=1$:
    $ H^0(X,\mathcal O_X)=k[X_1,X_1^{-1}]$
    $H^i(X,\mathcal O_X)=0 $ for $i\gt 0$
  2. For $n\gt 1$:
    $H^0(X,\mathcal O_X)=k[X_1,\cdots, X_n]$,
    $ H^i(X,\mathcal O_X)=0$ for $i\neq 0,n-1$,
    $ H^{n-1}(X,\mathcal O_X)=\bigoplus_{r_1\cdot\cdots,r_n\gt 0} k\cdot X_1^{-r_1} X_2^{-r_2} \cdots X_n^{-r_n}.$

The proof is by Čech cohomology. A small part can be found here.
Remarks
a) Remember that if a topological space has a covering $\mathcal U$ consisting of $n$ open subsets, then for any sheaf $\mathcal F$ of abelian groups, we have $\check H^i(\mathcal U,\mathcal F)=0$ as soon as $i\geq n$ (because Čech cohomology can be computed by alternating cochains).
b) Remember also that for $n=1$ the variety $X$ is affine, so that $\mathcal O_X$ is acyclic (as is any coherent sheaf).

  • I have to say, after knowing the final result, the problem turned out to be WAY easier. Now that I look at it, I realize that it would have helped to think about it geometrically first. Making the analogy with deRahm cohomology for manifolds for example, it would be expected to find that $H^i=0$ for $i=1,...,n-2$ and $H^{n-1}\neq 0$. Is that kind of thinking reasonable or was it just a nice coincidence? – rmdmc89 Jun 25 '20 at 18:14
  • Did you have any "beforehand" insight for this problem? Geometrical suggestion or whatever else? Thank you! – rmdmc89 Jun 25 '20 at 18:26
  • 1
    @mdmc89 I had no insight : I was taught Čech cohomology long ago by my wonderful, brilliant friend Otto Forster who showed me how that powerful technique solves many problems. And, yes, the analogy with a noncompact manifold having top dimension cohomology equal to zero is a good heuristic guide. – Georges Elencwajg Jun 28 '20 at 08:18