ConceptComplete

Sheaf Cohomology Hi(X,F)H^i(X, \mathcal{F})

Sheaf cohomology measures the obstruction to extending local data to global data. It is the central computational tool of algebraic geometry, connecting geometric properties of varieties to algebraic invariants.


Motivation

RemarkWhy cohomology?

The global sections functor Ξ“(X,βˆ’):Sh(X)β†’Ab\Gamma(X, -) : \mathbf{Sh}(X) \to \mathbf{Ab} is left exact but not right exact. Given a short exact sequence of sheaves

0→F′→F→F′′→00 \to \mathcal{F}' \to \mathcal{F} \to \mathcal{F}'' \to 0

we get 0β†’Ξ“(Fβ€²)β†’Ξ“(F)β†’Ξ“(Fβ€²β€²)0 \to \Gamma(\mathcal{F}') \to \Gamma(\mathcal{F}) \to \Gamma(\mathcal{F}'') but the last map need not be surjective. Cohomology measures the failure:

0β†’H0(Fβ€²)β†’H0(F)β†’H0(Fβ€²β€²)β†’H1(Fβ€²)β†’H1(F)β†’β‹―0 \to H^0(\mathcal{F}') \to H^0(\mathcal{F}) \to H^0(\mathcal{F}'') \to H^1(\mathcal{F}') \to H^1(\mathcal{F}) \to \cdots

So H1(Fβ€²)H^1(\mathcal{F}') is the obstruction to lifting global sections of Fβ€²β€²\mathcal{F}'' to F\mathcal{F}.

ExampleLine bundles and HΒΉ

On a complex manifold XX, the exponential sequence 0β†’2Ο€iZβ€Ύβ†’Oβ†’exp⁑Oβˆ—β†’00 \to \underline{2\pi i \mathbb{Z}} \to \mathcal{O} \xrightarrow{\exp} \mathcal{O}^* \to 0 gives:

H1(X,O)β†’H1(X,Oβˆ—)β†’c1H2(X,Z)H^1(X, \mathcal{O}) \to H^1(X, \mathcal{O}^*) \xrightarrow{c_1} H^2(X, \mathbb{Z})

Here H1(X,Oβˆ—)β‰…Pic⁑(X)H^1(X, \mathcal{O}^*) \cong \operatorname{Pic}(X) classifies line bundles. So H1H^1 classifies geometric objects (line bundles, extensions, deformations).


Definition via derived functors

Definition3.4Sheaf cohomology

Let XX be a topological space (or scheme) and F\mathcal{F} a sheaf of abelian groups. The sheaf cohomology groups are the right derived functors of the global sections functor:

Hi(X,F)=RiΞ“(X,F).H^i(X, \mathcal{F}) = R^i \Gamma(X, \mathcal{F}).

Concretely: choose an injective resolution 0→F→I0→I1→⋯0 \to \mathcal{F} \to \mathcal{I}^0 \to \mathcal{I}^1 \to \cdots (which exists since Sh(X)\mathbf{Sh}(X) has enough injectives), apply Γ\Gamma, and take cohomology:

Hi(X,F)=ker⁑(Ξ“(Ii)β†’Ξ“(Ii+1))im⁑(Ξ“(Iiβˆ’1)β†’Ξ“(Ii)).H^i(X, \mathcal{F}) = \frac{\ker(\Gamma(\mathcal{I}^i) \to \Gamma(\mathcal{I}^{i+1}))}{\operatorname{im}(\Gamma(\mathcal{I}^{i-1}) \to \Gamma(\mathcal{I}^i))}.

RemarkKey properties

Sheaf cohomology satisfies:

  1. H0(X,F)=Ξ“(X,F)H^0(X, \mathcal{F}) = \Gamma(X, \mathcal{F}) (global sections).
  2. Long exact sequence: A short exact sequence 0→F′→F→F′′→00 \to \mathcal{F}' \to \mathcal{F} \to \mathcal{F}'' \to 0 yields:

β‹―β†’Hi(Fβ€²)β†’Hi(F)β†’Hi(Fβ€²β€²)β†’Ξ΄Hi+1(Fβ€²)β†’β‹―\cdots \to H^i(\mathcal{F}') \to H^i(\mathcal{F}) \to H^i(\mathcal{F}'') \xrightarrow{\delta} H^{i+1}(\mathcal{F}') \to \cdots

  1. Hi(X,I)=0H^i(X, \mathcal{I}) = 0 for i>0i > 0 if I\mathcal{I} is injective (or flasque, soft, fine, ...).
  2. For a Noetherian scheme and coherent F\mathcal{F}: Hi(X,F)H^i(X, \mathcal{F}) is a finitely generated module (finite-dimensional kk-vector space if XX is projective over kk).
  3. Dimension vanishing: Hi(X,F)=0H^i(X, \mathcal{F}) = 0 for i>dim⁑Xi > \dim X (Grothendieck).

Cohomology of projective space

ExampleH⁰(ℙⁿ, O(d))

The most fundamental computation:

  • H0(Pn,O(d))=k[x0,…,xn]dH^0(\mathbb{P}^n, \mathcal{O}(d)) = k[x_0, \ldots, x_n]_d (degree-dd homogeneous polynomials) for dβ‰₯0d \geq 0
  • H0(Pn,O(d))=0H^0(\mathbb{P}^n, \mathcal{O}(d)) = 0 for d<0d < 0

with dim⁑H0(Pn,O(d))=(n+dd)\dim H^0(\mathbb{P}^n, \mathcal{O}(d)) = \binom{n+d}{d} for dβ‰₯0d \geq 0.

Special cases:

  • H0(Pn,O)=kH^0(\mathbb{P}^n, \mathcal{O}) = k: the only global regular functions are constants.
  • H0(P1,O(2))=kβ‹…x2+kβ‹…xy+kβ‹…y2H^0(\mathbb{P}^1, \mathcal{O}(2)) = k \cdot x^2 + k \cdot xy + k \cdot y^2: 33-dimensional (conics on P1\mathbb{P}^1).
  • H0(P2,O(3))H^0(\mathbb{P}^2, \mathcal{O}(3)): 1010-dimensional (cubics in P2\mathbb{P}^2, i.e., elliptic curves and their degenerations).
ExampleHigher cohomology of O(d) on ℙⁿ

The full computation (Theorem III.5.1 in Hartshorne):

  • H0(Pn,O(d))=k[x0,…,xn]dH^0(\mathbb{P}^n, \mathcal{O}(d)) = k[x_0,\ldots,x_n]_d for i=0i = 0, dβ‰₯0d \geq 0
  • Hi(Pn,O(d))=0H^i(\mathbb{P}^n, \mathcal{O}(d)) = 0 for 0<i<n0 < i < n, all dd
  • Hn(Pn,O(d))=k[x0βˆ’1,…,xnβˆ’1]βˆ’dβˆ’nβˆ’1H^n(\mathbb{P}^n, \mathcal{O}(d)) = k[x_0^{-1},\ldots,x_n^{-1}]_{-d-n-1} for dβ‰€βˆ’(n+1)d \leq -(n+1)
  • Hi(Pn,O(d))=0H^i(\mathbb{P}^n, \mathcal{O}(d)) = 0 otherwise

Key features:

  • Gap: Hi=0H^i = 0 for 0<i<n0 < i < n (all twists). Only H0H^0 and HnH^n can be nonzero.
  • Serre duality: Hn(Pn,O(d))β‰…H0(Pn,O(βˆ’dβˆ’nβˆ’1))∨H^n(\mathbb{P}^n, \mathcal{O}(d)) \cong H^0(\mathbb{P}^n, \mathcal{O}(-d-n-1))^\vee β€” the HnH^n is dual to H0H^0 of the "opposite" twist.
  • Ο‡(Pn,O(d))=(n+dn)\chi(\mathbb{P}^n, \mathcal{O}(d)) = \binom{n+d}{n} (the Hilbert polynomial).
ExampleCohomology on β„™ΒΉ (the simplest case)

On P1\mathbb{P}^1:

| dd | h0(O(d))h^0(\mathcal{O}(d)) | h1(O(d))h^1(\mathcal{O}(d)) | Ο‡(O(d))\chi(\mathcal{O}(d)) | |-----|------------------------|------------------------|------------------------| | β‰€βˆ’2\leq -2 | 00 | βˆ’dβˆ’1-d-1 | d+1d + 1 | | βˆ’1-1 | 00 | 00 | 00 | | 00 | 11 | 00 | 11 | | 11 | 22 | 00 | 22 | | dβ‰₯0d \geq 0 | d+1d+1 | 00 | d+1d+1 |

Note: h0βˆ’h1=d+1=Ο‡h^0 - h^1 = d + 1 = \chi always (Riemann–Roch on P1\mathbb{P}^1), and Serre duality gives h1(O(d))=h0(O(βˆ’dβˆ’2))h^1(\mathcal{O}(d)) = h^0(\mathcal{O}(-d-2)).


Cohomology of curves

ExampleThe genus of a curve

For a smooth projective curve CC of genus gg:

  • H0(C,OC)=kH^0(C, \mathcal{O}_C) = k (connected).
  • H1(C,OC)=kgH^1(C, \mathcal{O}_C) = k^g (gg-dimensional).
  • Hi(C,OC)=0H^i(C, \mathcal{O}_C) = 0 for iβ‰₯2i \geq 2 (dim⁑C=1\dim C = 1).

The genus g=h1(OC)g = h^1(\mathcal{O}_C) is the most basic invariant of a curve. It determines the topology (gg = number of "holes"):

  • g=0g = 0: P1\mathbb{P}^1 (rational curve, sphere).
  • g=1g = 1: elliptic curve (torus).
  • gβ‰₯2g \geq 2: hyperbolic curves (Faltings: finitely many rational points over Q\mathbb{Q}).
ExampleCohomology of line bundles on curves

For a line bundle L\mathcal{L} of degree dd on a genus-gg curve CC:

  • d<0d < 0: H0(C,L)=0H^0(C, \mathcal{L}) = 0 (no nonzero sections of negative degree).
  • d>2gβˆ’2d > 2g - 2: H1(C,L)=0H^1(C, \mathcal{L}) = 0 (by Serre duality + degree argument).
  • d=2gβˆ’2d = 2g - 2: L\mathcal{L} could be the canonical bundle Ο‰C\omega_C; h0(Ο‰C)=gh^0(\omega_C) = g.
  • Riemann–Roch: h0(L)βˆ’h1(L)=dβˆ’g+1h^0(\mathcal{L}) - h^1(\mathcal{L}) = d - g + 1 always.

The "interesting range" is 0≀d≀2gβˆ’20 \leq d \leq 2g - 2, where both H0H^0 and H1H^1 can be nonzero.

ExampleCohomology of an elliptic curve

Let EE be an elliptic curve (g=1g = 1) over kk. Then:

  • h0(OE)=1h^0(\mathcal{O}_E) = 1, h1(OE)=1h^1(\mathcal{O}_E) = 1, so Ο‡(OE)=0\chi(\mathcal{O}_E) = 0.
  • Ο‰Eβ‰…OE\omega_E \cong \mathcal{O}_E (trivial canonical bundle β€” this characterizes elliptic curves among genus-1 curves).
  • For a degree-dd line bundle L\mathcal{L}: h0(L)=dh^0(\mathcal{L}) = d for d>0d > 0, h0(L)=0h^0(\mathcal{L}) = 0 for d<0d < 0, and h0(L)=0h^0(\mathcal{L}) = 0 or 11 for d=0d = 0.

The Riemann–Roch space H0(E,O(nP))H^0(E, \mathcal{O}(nP)) has dimension nn for nβ‰₯1n \geq 1. Taking n=2n = 2: a basis {1,x}\{1, x\}; n=3n = 3: a basis {1,x,y}\{1, x, y\}. The relation in degree 6 gives the Weierstrass equation y2=x3+ax+by^2 = x^3 + ax + b.


Euler characteristic and Hilbert polynomial

Definition3.5Euler characteristic

For a coherent sheaf F\mathcal{F} on a projective scheme XX over kk:

Ο‡(X,F)=βˆ‘i=0dim⁑X(βˆ’1)ihi(X,F)\chi(X, \mathcal{F}) = \sum_{i=0}^{\dim X} (-1)^i h^i(X, \mathcal{F})

where hi=dim⁑kHi(X,F)h^i = \dim_k H^i(X, \mathcal{F}). The Euler characteristic is additive on short exact sequences.

Definition3.6Hilbert polynomial

For a coherent sheaf F\mathcal{F} on Pn\mathbb{P}^n (or a projective scheme with a fixed ample line bundle), the Hilbert polynomial is

PF(d)=Ο‡(F(d))P_\mathcal{F}(d) = \chi(\mathcal{F}(d))

as a function of dd. For d≫0d \gg 0, PF(d)=h0(F(d))P_\mathcal{F}(d) = h^0(\mathcal{F}(d)) (higher cohomology vanishes by Serre). This is a polynomial in dd of degree dim⁑Supp⁑(F)\dim \operatorname{Supp}(\mathcal{F}).

ExampleHilbert polynomials

| Sheaf | P(d)P(d) | Degree | Genus | |-------|--------|--------|-------| | OPn\mathcal{O}_{\mathbb{P}^n} | (n+dn)\binom{n+d}{n} | nn | β€” | | OP1\mathcal{O}_{\mathbb{P}^1} | d+1d + 1 | 11 | 00 | | OC\mathcal{O}_C (CβŠ†P2C \subseteq \mathbb{P}^2, deg ee) | edβˆ’e(eβˆ’3)2ed - \frac{e(e-3)}{2} | 11 | (eβˆ’1)(eβˆ’2)2\frac{(e-1)(e-2)}{2} | | Op\mathcal{O}_p (point) | 11 | 00 | β€” | | OS\mathcal{O}_S (SβŠ†P3S \subseteq \mathbb{P}^3, deg ee) | e2d2+β‹―\frac{e}{2}d^2 + \cdots | 22 | β€” |

The Hilbert polynomial determines the Hilbert scheme: varieties with a given Hilbert polynomial form a projective scheme Hilb⁑P(Pn)\operatorname{Hilb}_P(\mathbb{P}^n).


Vanishing theorems

TheoremGrothendieck vanishing

If XX is a Noetherian topological space of dimension nn, then Hi(X,F)=0H^i(X, \mathcal{F}) = 0 for all i>ni > n and all sheaves F\mathcal{F}.

TheoremSerre vanishing (Theorem B)

Let XX be a projective scheme over a Noetherian ring, O(1)\mathcal{O}(1) a very ample line bundle, and F\mathcal{F} a coherent sheaf. Then there exists d0d_0 such that

Hi(X,F(d))=0forΒ allΒ i>0Β andΒ dβ‰₯d0.H^i(X, \mathcal{F}(d)) = 0 \quad \text{for all } i > 0 \text{ and } d \geq d_0.

Equivalently, for sufficiently large twist, all higher cohomology vanishes.

ExampleApplication of Serre vanishing

On P2\mathbb{P}^2, for any coherent sheaf F\mathcal{F}: Hi(F(d))=0H^i(\mathcal{F}(d)) = 0 for i>0i > 0 and d≫0d \gg 0.

Concretely for OC\mathcal{O}_C where CC is a degree-3 curve: H1(OC(d))=0H^1(\mathcal{O}_C(d)) = 0 for dβ‰₯1d \geq 1 (by Serre duality: H1(OC(d))β‰…H0(Ο‰C(βˆ’d))∨=H0(OC(βˆ’d))∨=0H^1(\mathcal{O}_C(d)) \cong H^0(\omega_C(-d))^\vee = H^0(\mathcal{O}_C(-d))^\vee = 0 for d>0d > 0).

TheoremKodaira vanishing

Let XX be a smooth projective variety over a field of characteristic 00, and L\mathcal{L} an ample line bundle. Then:

Hi(X,Ο‰XβŠ—L)=0forΒ i>0.H^i(X, \omega_X \otimes \mathcal{L}) = 0 \quad \text{for } i > 0.

Equivalently, Hi(X,Lβˆ’1)=0H^i(X, \mathcal{L}^{-1}) = 0 for i<dim⁑Xi < \dim X (by Serre duality).

ExampleApplication of Kodaira vanishing

On Pn\mathbb{P}^n with L=O(d)\mathcal{L} = \mathcal{O}(d), d>0d > 0: Ο‰PnβŠ—O(d)=O(dβˆ’nβˆ’1)\omega_{\mathbb{P}^n} \otimes \mathcal{O}(d) = \mathcal{O}(d - n - 1). Kodaira gives Hi(O(dβˆ’nβˆ’1))=0H^i(\mathcal{O}(d-n-1)) = 0 for i>0i > 0, i.e., Hi(O(m))=0H^i(\mathcal{O}(m)) = 0 for i>0i > 0 and m>βˆ’(n+1)m > -(n+1). This reproduces part of the computation of Hβˆ—(Pn,O(d))H^*(\mathbb{P}^n, \mathcal{O}(d)).

Warning: Kodaira vanishing fails in characteristic p>0p > 0 (Raynaud's counterexample).


Higher direct images

Definition3.7Higher direct images

For a morphism f:X→Yf: X \to Y and a sheaf F\mathcal{F} on XX, the higher direct images are

Rifβˆ—FR^i f_* \mathcal{F}

the sheafification of U↦Hi(fβˆ’1(U),F)U \mapsto H^i(f^{-1}(U), \mathcal{F}). These are quasi-coherent if ff is quasi-compact, quasi-separated and F\mathcal{F} is quasi-coherent; coherent if ff is proper and F\mathcal{F} is coherent.

ExampleCohomology and base change

For a flat projective morphism f:Xβ†’Tf: X \to T and a coherent sheaf F\mathcal{F} flat over TT, the function t↦hi(Xt,Ft)t \mapsto h^i(X_t, \mathcal{F}_t) is upper semicontinuous. It is locally constant if and only if Rifβˆ—FR^i f_* \mathcal{F} is locally free.

Example: a family of elliptic curves f:Eβ†’Tf: \mathcal{E} \to T. Then R1fβˆ—OER^1 f_* \mathcal{O}_\mathcal{E} is a line bundle on TT (the Hodge bundle), since h1(OEt)=1h^1(\mathcal{O}_{E_t}) = 1 for all tt.


Summary

RemarkCohomology roadmap
InvariantDefinitionGeometric meaning
h0(F)h^0(\mathcal{F})dim⁑Γ(X,F)\dim \Gamma(X, \mathcal{F})Global sections (functions, forms, ...)
h1(OX)h^1(\mathcal{O}_X)dim⁑H1\dim H^1Genus (curves), irregularity (surfaces)
h0(Ο‰X)h^0(\omega_X)pgp_g, geometric genusTop holomorphic forms
Ο‡(OX)\chi(\mathcal{O}_X)βˆ‘(βˆ’1)ihi\sum (-1)^i h^iArithmetic genus
P(d)=Ο‡(O(d))P(d) = \chi(\mathcal{O}(d))Hilbert polynomialDegree, dimension, genus in one package
H1(X,Oβˆ—)H^1(X, \mathcal{O}^*)Pic⁑(X)\operatorname{Pic}(X)Line bundles / divisor classes