Sheaf Cohomology in Algebraic Geometry¶
Job: coherent cohomology \(H^i(X,\mathcal{F})\) via Čech and derived functors — Serre duality and classical Riemann–Roch as working tools on curves/projective space.
Core¶
Coherent and quasi-coherent sheaves¶
On a scheme \(X\): an \(\mathcal{O}_X\)-module \(\mathcal{F}\) is quasi-coherent if \(X\) has an affine cover \(U_i=\operatorname{Spec} A_i\) with \(\mathcal{F}|_{U_i}\cong\widetilde{M_i}\) for \(A_i\)-modules \(M_i\). Coherent (under locally Noetherian hyp.): locally \(\widetilde{M}\) with \(M\) f.g., and kernels of maps between finite free sheaves behave well (equiv. f.p. on affines when \(A\) Noetherian).
| Class | Affine \(\operatorname{Spec} A\) | Stability |
|---|---|---|
| \(\operatorname{QCoh}\) | all \(A\)-modules | kernels, cokernels, extensions; \(f_*\) for qcqs |
| \(\operatorname{Coh}\) | f.g. modules (\(A\) Noeth.) | same under proper pushforward (Grothendieck) |
| Locally free rank \(r\) | projective f.g. rank \(r\) | vector bundles |
Serre’s \(\mathcal{O}(n)\). On \(\mathbb{P}^n_A\), twisting sheaves \(\mathcal{O}(n)\); every coherent \(\mathcal{F}\) fits \(\bigoplus_i\mathcal{O}(n_i)^{\oplus m_i}\twoheadrightarrow\mathcal{F}\) for \(n_i\ll 0\) (or dual formulation with surjections from \(\mathcal{O}(-N)\)).
Global sections and derived functors¶
\(\Gamma(X,-):\operatorname{Ab}(X)\to\mathbf{Ab}\) (sheaves of abelian groups) is left exact. Sheaf cohomology: $$ H^i(X,\mathcal{F}) := R^i\Gamma(X,\mathcal{F}), $$ computed via injective (or flasque / Godement) resolutions. Long exact sequence from \(0\to\mathcal{F}'\to\mathcal{F}\to\mathcal{F}''\to 0\).
Vanishing. If \(X\) is Noetherian of dimension \(d\) (topological), \(H^i(X,\mathcal{F})=0\) for \(i>d\) (Grothendieck). For \(X\) affine, \(H^i(X,\mathcal{F})=0\) for \(i>0\) and \(\mathcal{F}\) quasi-coherent (Serre).
Čech cohomology¶
For open cover \(\mathfrak{U}=\{U_i\}\) and sheaf \(\mathcal{F}\), Čech complex \(\check{C}^\bullet(\mathfrak{U},\mathcal{F})\) with $$ \check{C}^p = \prod_{i_0<\cdots<i_p}\mathcal{F}(U_{i_0}\cap\cdots\cap U_{i_p}), $$ differential alternating restrictions. \(\check{H}^p(\mathfrak{U},\mathcal{F})=H^p(\check{C}^\bullet)\).
Theorem (comparison). For \(X\) separated, \(\mathcal{F}\) quasi-coherent, and \(\mathfrak{U}\) affine open cover, \(\check{H}^p(\mathfrak{U},\mathcal{F})\cong H^p(X,\mathcal{F})\). On \(\mathbb{P}^n\), standard affine cover \(D_+(x_i)\) computes all coherent cohomology in practice.
| Object | Computation |
|---|---|
| \(H^0(\mathbb{P}^n_k,\mathcal{O}(m))\) | homogeneous polynomials degree \(m\) (\(m\ge 0\)); \(0\) if \(m<0\) |
| \(H^n(\mathbb{P}^n_k,\mathcal{O}(-n-1))\) | \(k\) (generator \(x_0^{-1}\cdots x_n^{-1}\) in Čech) |
| \(H^i(\mathbb{P}^n,\mathcal{O}(m))\) | vanishes unless \(i=0\) or \(i=n\) (Serre) |
Higher direct images¶
For \(f:X\to Y\), \(R^if_*\mathcal{F}\) sheafifies \(U\mapsto H^i(f^{-1}U,\mathcal{F})\). Leray spectral sequence \(E_2^{p,q}=H^p(Y,R^qf_*\mathcal{F})\Rightarrow H^{p+q}(X,\mathcal{F})\). Proper + coherent \(\Rightarrow\) \(R^if_*\) coherent (Noetherian hyp.).
Serre duality (pointer)¶
Theorem (Serre duality, smooth projective). \(X/k\) smooth projective pure dim \(n\), \(\omega_X=\det\Omega_{X/k}\). For coherent \(\mathcal{F}\), $$ H^i(X,\mathcal{F})^\vee \cong \operatorname{Ext}^{n-i}_X(\mathcal{F},\omega_X), $$ and if \(\mathcal{F}\) locally free, \(\operatorname{Ext}^{n-i}(\mathcal{F},\omega_X)\cong H^{n-i}(X,\mathcal{F}^\vee\otimes\omega_X)\). In particular \(H^n(X,\omega_X)\cong k\) (trace).
Curves. Genus \(g=\dim H^1(X,\mathcal{O}_X)=\dim H^0(X,\omega_X)\). Duality: \(H^1(L)^\vee\cong H^0(\omega\otimes L^{-1})\) for line bundles \(L\).
Grothendieck duality upgrades this to derived categories and proper morphisms — derived categories.
Riemann–Roch (classical)¶
Theorem (RR for curves). \(X\) smooth projective curve genus \(g\) over \(k\) alg. closed, \(D\) divisor (or line bundle \(L=\mathcal{O}(D)\)): $$ \dim H^0(X,L) - \dim H^1(X,L) = \deg L + 1 - g. $$ With Serre duality: \(\ell(D)-\ell(K-D)=\deg D+1-g\).
| Special case | Conclusion |
|---|---|
| \(\deg L>2g-2\) | \(H^1(L)=0\); \(\dim H^0=\deg L+1-g\) |
| \(L=\mathcal{O}\) | \(\chi(\mathcal{O})=1-g\) |
| Canonical \(K\) | \(\deg K=2g-2\); \(\ell(K)=g\) |
Hirzebruch–RR / GRR (pointer). For higher dimension, \(\chi(\mathcal{F})=\int_X \operatorname{ch}(\mathcal{F})\operatorname{td}(T_X)\); Grothendieck–RR in \(K\)-theory for proper morphisms.
Euler characteristic. \(\chi(\mathcal{F})=\sum_i (-1)^i\dim H^i(X,\mathcal{F})\); additive in exact sequences; constant in flat proper families (upper semicontinuity of \(h^i\), constancy of \(\chi\)).
See it¶
Canonical embedding of non-hyperelliptic curves. For \(g\ge 3\) non-hyperelliptic smooth projective curve, \(|K|\) is very ample and embeds \(X\hookrightarrow\mathbb{P}^{g-1}\) as a degree \(2g-2\) curve. RR + duality compute \(\dim|K|=g-1\) and that \(H^0(K)\) separates points/tangents precisely when no \(g^1_2\) (hyperelliptic linear system). Petri’s theorem and moduli of curves use this cohomological package constantly.
Cohomology and base change. Flat proper \(f:X\to S\), coherent \(\mathcal{F}\) flat over \(S\): for \(s\in S\), the map \(R^if_*\mathcal{F}\otimes\kappa(s)\to H^i(X_s,\mathcal{F}_s)\) is iso under cohomological flatness / vanishing of adjacent \(R^{i\pm1}\) — failure modes organize degeneration of Hodge numbers in families.
Check¶
Compute \(H^1(\mathbb{P}^1_k,\mathcal{O})\) and \(H^1(\mathbb{P}^1_k,\mathcal{O}(-2))\) via Čech on \(\{U_0,U_1\}\).
\(U_i=D_+(x_i)\cong\mathbb{A}^1\), \(U_0\cap U_1=\operatorname{Spec} k[t,t^{-1}]\). \(\check{C}^0=\mathcal{O}(U_0)\times\mathcal{O}(U_1)\), \(\check{C}^1=\mathcal{O}(U_0\cap U_1)\). For \(\mathcal{O}\): sections polynomials in \(t\) and in \(t^{-1}\); difference on overlap — cokernel of \((f,g)\mapsto f-g\) is \(0\) (Laurent with only nonnegative/nonpositive parts glue). For \(\mathcal{O}(-2)\): transition multiplies by \(t^{2}\) (conventions); \(H^1\cong k\cdot t^{-1}\) one-dimensional, dual to \(H^0(\mathcal{O})\cong k\) by Serre.
Why does RR not determine \(\dim H^0\) alone for intermediate degrees?
Only \(\chi(L)=\ell(D)-\ell(K-D)\) is topological/degree-theoretic. Both terms can be positive (special divisors, Brill–Noether); Clifford’s theorem bounds \(\ell(D)\le\deg D/2+1\) for special \(D\).
Pitfalls¶
- Čech depends on cover until comparison theorems apply; bad covers give wrong groups.
- Coherent \(\neq\) quasi-coherent: infinite products of structure sheaves, \(j_!\mathcal{O}_U\) on non-closed opens often fail coherence.
- GAGA: for proper \(X/\mathbb{C}\), analytic and algebraic coherent cohomology agree; open varieties need care.
- Base field not alg. closed: dimensions after extension; geometric genus vs arithmetic genus.
Refs¶
- Hartshorne, Algebraic Geometry (III)
- Mumford, Abelian Varieties (RR applications)
- The Stacks Project, Cohomology of Schemes
- Huybrechts, Fourier-Mukai Transforms in Algebraic Geometry (derived upgrade)
Next¶
Foundations · Schemes · AG index · Homological algebra · Derived categories · Algebraic topology · Prismatic cohomology — the p-adic cohomology theory unifying crystalline / de Rham / étale