04.03.05 · algebraic-geometry / cohomology

Serre's vanishing and finiteness theorems

shipped3 tiersLean: none

Anchor (Master): Serre 1955 *Faisceaux Algébriques Cohérents* (Annals of Math. 61); Hartshorne §III.5; Grothendieck-Dieudonné EGA III §2; Mumford *Lectures on Curves on an Algebraic Surface* §14 (Castelnuovo-Mumford regularity)

Intuition [Beginner]

A coherent sheaf on a projective variety carries finitely many independent local conditions, packaged as a finite-dimensional family of sections, modulo finite relations. Serre's two theorems take this finiteness on the input side and turn it into finiteness on the output side. The output is the cohomology, which records the obstruction to gluing local sections into global ones. Serre's finiteness theorem says that obstruction is itself finite-dimensional in every cohomological degree. Serre's vanishing theorem says that if you twist the sheaf by a sufficiently large positive integer, the obstruction disappears entirely above degree zero.

The intuition behind the vanishing: twisting a coherent sheaf by $O (n)$ for large $n$ buys you many global sections. Once the global sections are large enough, the local-to-global gluing condition is solved by a single global section, and the higher cohomology degrees record nothing. The intuition behind the finiteness: a coherent sheaf is built from finitely many generators with finitely many relations, and the cohomology of such a finitely-built object cannot be any larger than the cohomology of its building blocks, which are line bundles whose cohomology we already counted.

The pay-off is that every cohomology calculation on a projective variety reduces to a finite linear-algebra problem over the base ring. The Hilbert polynomial, the Euler characteristic, Riemann-Roch, the Hilbert scheme — every one of them rests on Serre's two theorems as the foundation that makes the calculations finite-dimensional and asymptotically vanishing.

Visual [Beginner]

A horizontal axis labelled by the twist $n$ and a vertical axis labelled by cohomology dimension. For each cohomological degree $i \geq 1$ , a curve that is non-zero for small $n$ and drops to zero at some threshold $n_{0} (F, i)$ . For $i = 0$ , a curve that grows polynomially in $n$ and stays positive. A vertical dashed line marks the largest threshold across all $i \geq 1$ , namely $n_{0} (F) = max_{i} n_{0} (F, i)$ , beyond which all higher cohomology is zero.

$A schematic dimension diagram for the cohomology of a coherent sheaf twisted by $\mathcal{O}(n)$, with higher cohomology curves dropping to zero past a threshold $n_0$ and the global-sections curve growing polynomially.$

Worked example [Beginner]

Take $X = P^{2}$ and let $C \subseteq X$ be a smooth plane curve of degree $e = 3$ (an elliptic curve embedded in the projective plane). The ideal sheaf $I_{C}$ records the polynomials vanishing on $C$ , and the twist $I_{C} (d)$ records degree- $d$ polynomials that vanish on $C$ . Compute the threshold $n_{0}$ at which $H^{1} (P^{2}, I_{C} (d))$ vanishes.

Step 1. The short exact sequence of sheaves on $P^{2}$ is $0 \to I_{C} (d) \to O (d) \to O_{C} (d) \to 0$ , and the ideal sheaf identifies with $O (d - 3)$ because $C$ is cut out by a single degree- $3$ polynomial. So $I_{C} (d) = O (d - 3)$ .

Step 2. The dimension of $H^{1} (P^{2}, O (m))$ equals zero for every $m \in Z$ (intermediate vanishing on $P^{2}$ in cohomological degree one).

Step 3. Putting them together: $H^{1} (P^{2}, I_{C} (d)) = H^{1} (P^{2}, O (d - 3)) = 0$ for every $d$ . The threshold is $n_{0} = - \infty$ in this example because the ideal sheaf of a hypersurface is itself a line bundle.

What this tells us. For an ideal sheaf cut out by a single equation, the twisting trick immediately reduces to the dimension table on $P^{N}$ , and the threshold $n_{0}$ is read directly off that table. For more general coherent sheaves the same idea works inductively: build the sheaf from line bundles by short exact sequences, propagate the table through each long exact sequence, and the threshold for the whole sheaf is the largest of the thresholds of the building blocks.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Fix a noetherian ring $A$ and a noetherian projective scheme $X$ over $A$ . By definition, $X$ is projective over $A$ when there is a closed immersion $i : X ↪ P_{A}^{N}$ for some $N$ , and the sheaf $O_{X} (1) := i^{*} O_{P_{A}^{N}} (1)$ is a very ample line bundle on $X$ relative to $Spec A$ . For any $O_{X}$ -module $F$ and integer $n$ , write $F (n) := F \otimes_{O_{X}} O_{X} (n)$ for the $n$ -th twist.

A sheaf $F$ is coherent on $X$ when it is finitely generated as an $O_{X}$ -module and has finitely generated relations: locally on every affine open $U = Spec B \subseteq X$ , the restriction $F ∣_{U}$ is the sheafification of a finitely presented $B$ -module. On a noetherian scheme, coherence is equivalent to quasi-coherence plus finite generation locally; the noetherian hypothesis makes the relations automatically finite.

Serre's finiteness theorem. Let $A$ be noetherian, $X$ a projective scheme over $A$ , and $F$ a coherent $O_{X}$ -module. Then for every $i \geq 0$ , the sheaf cohomology $H^{i} (X, F)$ is a finitely generated $A$ -module. Moreover, $H^{i} (X, F) = 0$ for $i > dim X$ .

The latter assertion is Grothendieck's vanishing theorem: on a noetherian topological space of Krull dimension $n$ , sheaf cohomology vanishes above degree $n$ for every abelian sheaf. On a projective scheme of dimension $dim X = n$ , this caps the range of cohomological degrees at $0 \leq i \leq n$ .

Serre's vanishing theorem. Under the same hypotheses, there exists an integer $n_{0} = n_{0} (F, O_{X} (1))$ such that for every $n \geq n_{0}$ and every $i > 0$ ,

$H^{i} (X, F (n)) = 0.$

The threshold $n_{0}$ depends on $F$ and on the choice of very ample $O_{X} (1)$ , but not on $i$ — once $n$ is at least $n_{0}$ , every higher cohomology vanishes simultaneously.

A useful packaging of the vanishing assertion is via Castelnuovo-Mumford regularity: $F$ is $m$ -regular with respect to $O_{X} (1)$ when $H^{i} (X, F (m - i)) = 0$ for every $i > 0$ . The regularity $reg (F)$ is the smallest $m$ for which $F$ is $m$ -regular. Once $F$ is $m$ -regular, it is $m^{'}$ -regular for every $m^{'} \geq m$ ; in particular Serre vanishing gives an explicit choice $n_{0} = reg (F)$ that simultaneously kills every $H^{i}$ for $i > 0$ and $n \geq n_{0}$ .

Counterexamples to common slips

The threshold $n_{0}$ depends on the choice of very ample $O_{X} (1)$ . Replacing $O_{X} (1)$ by $O_{X} (2)$ (still very ample, by Veronese) shifts every threshold; the existence of $n_{0}$ is invariant, the integer is not.
Both theorems require coherence of $F$ . A non-coherent quasi-coherent sheaf, e.g. an infinite direct sum of line bundles, can have infinite-dimensional cohomology and no asymptotic vanishing.
The finiteness assertion needs $A$ noetherian. Without noetherian hypotheses on the base, a coherent sheaf can fail to have finitely generated cohomology even on a projective variety.
The vanishing assertion needs the ample-twist direction $n \to + \infty$ . Twisting by $O (- n)$ for $n \to \infty$ produces non-vanishing high cohomology, not vanishing low cohomology — this is the dual range, governed by Serre duality.

Key theorem with proof [Intermediate+]

Theorem (Serre 1955; Hartshorne III.5.2). Let $A$ be a noetherian ring, $X$ a projective scheme over $A$ with very ample line bundle $O_{X} (1)$ , and $F$ a coherent $O_{X}$ -module. Then:

(Finiteness.) For every $i \geq 0$ , the cohomology $H^{i} (X, F)$ is a finitely generated $A$ -module.
(Vanishing.) There exists an integer $n_{0} = n_{0} (F)$ such that $H^{i} (X, F (n)) = 0$ for every $n \geq n_{0}$ and every $i > 0$ .

Proof. The argument runs in three ingredients: reduction to projective space, explicit cohomology of line bundles on projective space, and induction on cohomological degree via short exact sequences.

Ingredient 1: reduction to $X = P_{A}^{N}$ . The closed immersion $i : X ↪ P_{A}^{N}$ has the property that $i_{*} F$ is coherent on $P_{A}^{N}$ (closed-immersion pushforward of a coherent sheaf is coherent), and $i_{*}$ is exact on the category of quasi-coherent sheaves with the higher direct images $R^{p} i_{*} F = 0$ for $p \geq 1$ (closed immersions are affine). The Leray spectral sequence for $i$ degenerates on the row $p = 0$ , giving $H^{q} (X, F) ≅ H^{q} (P_{A}^{N}, i_{*} F)$ for every $q$ . The twist $F (n) = F \otimes_{O_{X}} i^{*} O_{P_{A}^{N}} (n)$ pushforwards to $i_{*} F (n) = i_{*} F \otimes_{O_{P_{A}^{N}}} O_{P_{A}^{N}} (n) = (i_{*} F) (n)$ by the projection formula. So both finiteness of $H^{i} (X, F)$ and the asymptotic vanishing of $H^{i} (X, F (n))$ follow from the corresponding statements for the coherent sheaf $i_{*} F$ on $P_{A}^{N}$ . Replace $X$ by $P_{A}^{N}$ and $F$ by $i_{*} F$ for the remainder of the proof.

Ingredient 2: explicit cohomology of line bundles on $P_{A}^{N}$ . The cohomology $H^{p} (P_{A}^{N}, O (d))$ is computed in 04.03.04. The dimension table over a field generalises to a graded $A$ -module description: $H^{0} (P_{A}^{N}, O (d)) = A [x_{0}, \dots, x_{N}]_{d}$ , the degree- $d$ piece of the polynomial ring, free of rank $(N N + d)$ over $A$ for $d \geq 0$ and zero otherwise; $H^{N} (P_{A}^{N}, O (d))$ is the $A$ -module spanned by Laurent monomials $x_{0}^{a_{0}} \dots x_{N}^{a_{N}}$ with all $a_{i} < 0$ and $\sum a_{i} = d$ , free of rank $(N - d - 1)$ over $A$ for $d \leq - N - 1$ and zero otherwise; and $H^{p} (P_{A}^{N}, O (d)) = 0$ for $0 < p < N$ and for $p > N$ . Each of these is a finitely generated $A$ -module — even free, with explicit rank — so finiteness holds for $F = O (d)$ directly. The vanishing assertion for $O (d) (n) = O (d + n)$ kicks in at $n_{0} = - d$ for the top cohomology $H^{N}$ (since $d + n \geq - N$ kills it for $N \geq 1$ , and for $N = 0$ the line-bundle case is degenerate and reduces to the affine vanishing on $Spec A$ ), and the middle and high degrees are zero for every $n$ . So both statements hold for $F = O (d)$ .

The same conclusion extends to finite direct sums: a sum $⨁_{j = 1}^{k} O (d_{j})$ has cohomology $⨁_{j = 1}^{k} H^{p} (P_{A}^{N}, O (d_{j}))$ , a finite direct sum of finitely generated $A$ -modules, with vanishing threshold $n_{0} = max_{j} (- d_{j}) + 1$ .

Ingredient 3: descending induction via short exact sequences. For an arbitrary coherent $F$ on $P_{A}^{N}$ , Serre's theorem A (also established in FAC 1955) gives a surjection $O (- d)^{\oplus k} ↠ F$ for some $d, k \geq 0$ . Concretely: $F$ has finitely many global sections after sufficient twisting (this is Serre's theorem A), and a finite generating set produces a surjection from a free coherent sheaf of the form $O (- d)^{\oplus k}$ . Let $K$ be the kernel:

$0 \to K \to O (- d)^{\oplus k} \to F \to 0.$

The kernel $K$ is coherent because the category of coherent sheaves on a noetherian scheme is closed under kernels. The long exact sequence in cohomology reads, for every $n \in Z$ ,

$\dots \to H^{i - 1} (F (n)) \to H^{i} (K (n)) \to H^{i} (O (n - d))^{\oplus k} \to H^{i} (F (n)) \to H^{i + 1} (K (n)) \to \dots .$

(Here we abbreviate $H^{i} (-) = H^{i} (P_{A}^{N}, -)$ and use exactness of the twist functor.)

Finiteness, by descending induction on $i$ . The base case $i > N$ uses Grothendieck vanishing: every $H^{i}$ vanishes for $i > N = dim P_{A}^{N}$ , so finiteness reduces to the finiteness of the zero $A$ -module. For $i \leq N$ , suppose finiteness has been established for every coherent sheaf on $P_{A}^{N}$ in cohomological degrees $> i$ . The long exact sequence gives a four-term sequence

$H^{i} (O (n - d))^{\oplus k} \to H^{i} (F (n)) \to H^{i + 1} (K (n)) \to H^{i + 1} (O (n - d))^{\oplus k} .$

Take $n = 0$ . The outer two terms are finitely generated (the line-bundle case gives finite generation directly; the inductive hypothesis gives finite generation of $H^{i + 1} (K)$ ). The middle term $H^{i} (F)$ sits between two finitely generated modules over a noetherian ring $A$ , hence is itself finitely generated. The induction closes.

Vanishing, by descending induction on $i$ . The base case $i > N$ is Grothendieck vanishing, no twist needed. For $1 \leq i \leq N$ , suppose Serre vanishing has been established for every coherent sheaf on $P_{A}^{N}$ in degrees $> i$ , with thresholds depending on the sheaf. Apply this to $K$ : there exists $n_{1}$ such that $H^{i + 1} (K (n)) = 0$ for $n \geq n_{1}$ . Apply the line-bundle case to $O (- d)$ : there exists $n_{2}$ such that $H^{i} (O (n - d)) = 0$ for $n \geq n_{2}$ (specifically, $n_{2} = d - N + 1$ from the dimension table). The four-term sequence becomes, for $n \geq max (n_{1}, n_{2})$ ,

$0 \to H^{i} (F (n)) \to 0,$

so $H^{i} (F (n)) = 0$ in this range. Set $n_{0} (F, i) := max (n_{1}, n_{2})$ and let $n_{0} (F) := max_{1 \leq i \leq N} n_{0} (F, i)$ . For $n \geq n_{0} (F)$ , every $H^{i} (F (n))$ vanishes for $1 \leq i \leq N$ , and Grothendieck vanishing handles $i > N$ . The induction closes.

The two statements are proved simultaneously: finiteness gives the tools for the vanishing induction (the inductive hypothesis is "finiteness in degrees $> i$ "), and vanishing in turn supplies the thresholds. $□$

The theorem is the engine of computational algebraic geometry on projective schemes. It promotes the dimension table on $P^{N}$ to a finiteness-and-vanishing pattern for every coherent sheaf, and the explicit threshold $n_{0}$ — Castelnuovo-Mumford regularity in the master tier — turns the asymptotic statement into a concrete bound for any specific calculation.

Bridge. The construction here builds toward 04.04.01 (Riemann-Roch theorem for curves), where finiteness of $H^{0}$ and $H^{1}$ for a line bundle on a smooth projective curve is exactly Serre finiteness, and the dimension formula $χ (L) = de g L + 1 - g$ encodes the Euler-characteristic side of the finiteness statement; the proof of Riemann-Roch routes through Serre vanishing for the asymptotic shape of $dim H^{0} (L)$ . The same construction appears again in 04.05.05 (ample line bundles), where the very-ample / ample correspondence with closed embeddings runs through global sections of $O (d)$ on $P^{N}$ , and Serre vanishing is the assertion that ampleness of $L$ is equivalent to $H^{i} (X, F \otimes L^{n}) = 0$ for $i > 0$ and $n ≫ 0$ for every coherent $F$ — the cohomological characterisation of ampleness. Putting these together, the foundational insight is that projective schemes have a well-behaved cohomology theory because each coherent sheaf is built from finitely many copies of line bundles whose cohomology is already explicit, and twisting controls the threshold past which the higher cohomology vanishes. The unit also couples to 04.03.04 (cohomology of line bundles on projective space), which supplies the dimension table that drives the entire reduction, and the same finiteness-and-vanishing pattern reappears in 06.04.04 (Serre duality on a curve) where the dual statement provides the basis-by-basis pairing that makes finite-dimensionality concrete.

Exercises [Intermediate+]

Exercise 3 (medium, numeric).

Let $C \subseteq P^{2}$ be a smooth plane curve of degree $e = 4$ (a quartic). Use the short exact sequence $0 \to O (d - 4) \to O (d) \to O_{C} (d) \to 0$ to compute $dim_{k} H^{0} (C, O_{C} (1))$ over an algebraically closed field $k$ .

Hint

Take the long exact sequence in cohomology with $d = 1$ and use the dimension table on $P^{2}$ .

Answer

The long exact sequence for $d = 1$ reads

$0 \to H^{0} (O (- 3)) \to H^{0} (O (1)) \to H^{0} (O_{C} (1)) \to H^{1} (O (- 3)) \to \dots .$

The dimension table gives $H^{0} (O (- 3)) = 0$ (negative degree), $H^{0} (O (1)) = k^{3}$ (degree- $1$ polynomials in $x_{0}, x_{1}, x_{2}$ ), and $H^{1} (O (- 3)) = 0$ (intermediate vanishing on $P^{2}$ ). So $H^{0} (C, O_{C} (1)) = k^{3}$ , of dimension $3$ . Geometrically, the linear system $∣ O_{C} (1) ∣$ on a quartic curve is the restriction of the complete linear system of lines in $P^{2}$ . Rubric: full credit for the dimension $3$ via the long exact sequence; partial credit for the dimension only.

Exercise 4 (medium, symbolic).

Show that for a coherent sheaf $F$ on $P_{A}^{N}$ and the surjection $O (- d)^{\oplus k} ↠ F$ from Serre's theorem A, the kernel $K$ is also coherent and satisfies the same theorem A: there exists $d^{'}, k^{'}$ with $O (- d^{'})^{\oplus k^{'}} ↠ K$ .

Hint

Coherence is closed under kernels on a noetherian scheme. Theorem A applies to every coherent sheaf, including $K$ .

Answer

The category of coherent sheaves on a noetherian scheme is an abelian subcategory of quasi-coherent sheaves, closed under kernels and cokernels. The kernel $K$ of a map of coherent sheaves $O (- d)^{\oplus k} \to F$ is coherent, since: locally, the map of $O_{X}$ -modules corresponds to a map of finitely presented modules over the local affine coordinate ring, and the kernel of such a map is finitely presented when the ring is noetherian. So $K$ is coherent on $P_{A}^{N}$ .

Serre's theorem A applies to every coherent sheaf on $P_{A}^{N}$ : for $K$ , there is a $d^{'} ≫ 0$ such that $K (d^{'})$ is generated by global sections, and finitely many of these generators produce a surjection $O_{P_{A}^{N}}^{\oplus k^{'}} ↠ K (d^{'})$ . Twisting by $- d^{'}$ gives $O (- d^{'})^{\oplus k^{'}} ↠ K$ , which is the desired statement. Iterating gives a free resolution $\dots \to O (- d_{2})^{\oplus k_{2}} \to O (- d_{1})^{\oplus k_{1}} \to O (- d_{0})^{\oplus k_{0}} \to F \to 0$ with all terms direct sums of line bundles, finite at each stage; on $P_{A}^{N}$ the resolution can be chosen of length at most $N + 1$ (Hilbert syzygy theorem). Rubric: full credit for the coherence-of-kernel argument and the application of theorem A; partial credit for the coherence assertion only.

Exercise 5 (medium, short-answer).

Take $X = P^{1}$ and let $p \in X$ be a closed point. Let $O_{p}$ be the skyscraper sheaf at $p$ . Compute $H^{i} (P^{1}, O_{p} (n))$ for every $i$ and $n$ .

Hint

A skyscraper sheaf has the same underlying stalk under any twist (the twist by $O (n)$ does not change a sheaf supported at a point), and its higher cohomology vanishes by direct computation on the standard cover.

Answer

The skyscraper sheaf $O_{p}$ has stalk $k$ at $p$ and zero elsewhere. The twist $O_{p} (n) = O_{p} \otimes O (n)$ is again a skyscraper sheaf at $p$ with stalk $k$ (the line bundle $O (n)$ is locally free of rank one, so tensoring with a skyscraper at $p$ produces a skyscraper at $p$ with the same stalk). On the standard two-set cover ${D_{+} (x_{0}), D_{+} (x_{1})}$ of $P^{1}$ , the point $p$ lies in at least one piece (every closed point is in some $D_{+} (x_{i})$ for the standard cover); the Čech complex of $O_{p} (n)$ has $\overset{ˇ}{C}^{0} = k \oplus 0$ (or $0 \oplus k$ , depending on which piece contains $p$ ) and $\overset{ˇ}{C}^{1} = 0$ (the overlap is the punctured line, which does not contain $p$ ). So $H^{0} = k$ and $H^{i} = 0$ for $i \geq 1$ . The cohomology of a skyscraper is concentrated in degree zero, independent of the twist. Rubric: full credit for $H^{0} = k$ and $H^{i} = 0$ for $i \geq 1$ , with the support-on-a-point justification.

Exercise 6 (medium, symbolic).

Prove that on a noetherian projective scheme $X$ over a noetherian ring $A$ , the Euler characteristic $χ (X, F (n)) := i \sum (- 1)^{i} rank_{A} H^{i} (X, F (n))$ is a polynomial in $n$ of degree at most $dim X$ .

Hint

Use the closed-embedding reduction to $P_{A}^{N}$ and the resolution by line bundles. The Euler characteristic is additive on short exact sequences, and the dimension table on $P_{A}^{N}$ gives a polynomial.

Answer

By the closed-embedding reduction, $χ (X, F (n)) = χ (P_{A}^{N}, i_{*} F (n))$ . By Serre's theorem A (Exercise 4), the coherent sheaf $i_{*} F$ on $P_{A}^{N}$ admits a finite resolution $E_{∙} \to i_{*} F \to 0$ with $E_{j} = ⨁_{l} O_{P_{A}^{N}} (d_{j, l})$ a finite direct sum of line bundles, of length at most $N + 1$ (Hilbert syzygy). Twisting commutes with this resolution: $E_{∙} (n) \to i_{*} F (n) \to 0$ remains exact for every $n$ . The Euler characteristic is additive on exact sequences: $χ (P_{A}^{N}, i_{*} F (n)) = \sum_{j} (- 1)^{j} χ (P_{A}^{N}, E_{j} (n))$ . Each summand is itself a sum of $χ (P_{A}^{N}, O (d_{j, l} + n))$ , and the dimension table gives $χ (P_{A}^{N}, O (m)) = (N N + m)$ as a polynomial in $m$ of degree $N$ . So $χ (P_{A}^{N}, i_{*} F (n))$ is a finite alternating sum of polynomials of degree $N$ in $n$ , hence a polynomial in $n$ of degree at most $N$ . The degree is bounded by $dim X \leq N$ because the Hilbert polynomial of $i_{*} F$ has degree equal to $dim supp (i_{*} F) = dim X$ , with the cancellation in the alternating sum dropping the degree from $N$ to $dim X$ . The conclusion: $χ (X, F (n))$ is a polynomial in $n$ of degree at most $dim X$ , the Hilbert polynomial of $F$ relative to $O_{X} (1)$ . Rubric: full credit for the resolution-and-additivity argument with the degree bound; partial credit for the resolution argument without the degree refinement.

Exercise 7 (hard, short-answer).

Prove the cohomological characterisation of ampleness: a line bundle $L$ on a noetherian projective scheme $X$ over a noetherian ring $A$ is ample if and only if for every coherent sheaf $F$ there is an integer $n_{0} = n_{0} (F)$ such that $H^{i} (X, F \otimes L^{n}) = 0$ for every $i > 0$ and $n \geq n_{0}$ .

Hint

Forward direction: ampleness gives that $L^{m}$ is very ample for some $m ≫ 0$ , reducing to Serre vanishing. Reverse direction: the vanishing condition on $O (- d)$ -twists for $d ≫ 0$ produces global generation of $L^{n}$ for $n ≫ 0$ , hence very ampleness for some power.

Answer

Forward direction. If $L$ is ample, then $L^{m}$ is very ample for some $m \geq 1$ (definition of ampleness). The very ample line bundle $L^{m}$ provides a closed embedding $X ↪ P_{A}^{N}$ with $O_{X} (1) = L^{m}$ . Serre vanishing applied to $F$ gives $n_{0}^{'} = n_{0} (F)$ such that $H^{i} (X, F \otimes L^{m \cdot n}) = 0$ for $n \geq n_{0}^{'}$ and $i > 0$ . To handle non-multiples of $m$ , apply the same theorem to $F \otimes L^{j}$ for each $j \in {0, 1, \dots, m - 1}$ , getting thresholds $n_{0} (j)$ . Set $n_{0} (F) := m \cdot max_{j} n_{0} (j) + m$ . For $n \geq n_{0} (F)$ , write $n = m q + j$ with $0 \leq j < m$ and $q \geq n_{0} (j)$ , then $F \otimes L^{n} = (F \otimes L^{j}) \otimes L^{m q}$ , and Serre vanishing on the very ample $L^{m}$ kills the higher cohomology.

Reverse direction. Suppose the vanishing condition holds. Apply it with $F = O_{X}$ and a fixed coherent ideal sheaf $I$ corresponding to a closed point $p \in X$ : there is an $n_{1}$ such that $H^{1} (X, I \otimes L^{n}) = 0$ for $n \geq n_{1}$ . The short exact sequence $0 \to I \otimes L^{n} \to L^{n} \to L^{n} ∣_{{p}} \to 0$ gives a surjection $H^{0} (X, L^{n}) ↠ L^{n} ∣_{{p}}$ , so $L^{n}$ is generated by global sections at $p$ . Varying $p$ and using the noetherian quasi-compact hypothesis, $L^{n}$ is globally generated for $n \geq n_{2}$ . Applying the same argument to ideal sheaves of pairs ${p, q}$ and to first-order infinitesimal neighbourhoods $Spec (O_{p} / m_{p}^{2})$ gives, for $n \geq n_{3}$ , that $L^{n}$ separates points and tangent vectors. So $L^{n}$ is very ample for $n \geq n_{3}$ , and $L$ is ample by definition. Rubric: full credit for both directions with the global-generation / separation-of-points arguments; partial credit for one direction.

Exercise 8 (hard, short-answer).

Define the Castelnuovo-Mumford regularity $reg (F)$ of a coherent sheaf $F$ on $P_{A}^{N}$ as the smallest integer $m$ for which $H^{i} (P_{A}^{N}, F (m - i)) = 0$ for every $i > 0$ . Prove that $reg (F)$ is finite, and show that if $F$ is $m$ -regular, then $F$ is $m^{'}$ -regular for every $m^{'} \geq m$ .

Hint

Finiteness follows from Serre vanishing applied at each cohomological degree separately. The " $m$ -regular implies $(m + 1)$ -regular" assertion uses the long exact sequence for a generic hyperplane section.

Answer

Finiteness. Serre vanishing gives, for each $i \in {1, \dots, N}$ , a threshold $n_{i} (F)$ such that $H^{i} (P_{A}^{N}, F (n)) = 0$ for $n \geq n_{i}$ . Set $m_{0} := max_{i} (n_{i} + i)$ ; then for every $i \in {1, \dots, N}$ and $m \geq m_{0}$ , the twist $m - i \geq n_{i}$ , so $H^{i} (F (m - i)) = 0$ . The smallest such $m$ is finite, hence $reg (F) < \infty$ .

Monotone propagation. Suppose $F$ is $m$ -regular. Choose a generic hyperplane $H \subseteq P_{A}^{N}$ such that the natural map $F (- 1) \to F$ given by multiplication by the equation of $H$ is injective on the relevant cohomology (this uses the noetherian + generic-hyperplane hypothesis). The short exact sequence $0 \to F (- 1) \to F \to F ∣_{H} \to 0$ twisted by $O (m + 1 - i)$ gives the long exact sequence

$\dots \to H^{i - 1} (F ∣_{H} (m + 1 - i)) \to H^{i} (F (m - i)) \to H^{i} (F (m + 1 - i)) \to H^{i} (F ∣_{H} (m + 1 - i)) \to \dots .$

The hypothesis $reg (F) \leq m$ gives $H^{i} (F (m - i)) = 0$ for $i \geq 1$ . Inductively, $F ∣_{H}$ is $m$ -regular on the hyperplane $H ≅ P_{A}^{N - 1}$ (the regularity passes to hyperplane sections by the same long exact sequence argument), so $H^{i - 1} (F ∣_{H} (m + 1 - i)) = 0$ for $i \geq 1$ . Combining, $H^{i} (F (m + 1 - i)) = 0$ for $i \geq 1$ , so $F$ is $(m + 1)$ -regular. Iterating gives $m^{'}$ -regularity for every $m^{'} \geq m$ .

Conclusion: $reg (F)$ is finite and $F$ is $m^{'}$ -regular for every $m^{'} \geq reg (F)$ , so the regularity index gives an explicit threshold for Serre vanishing: $n_{0} (F) = reg (F)$ kills every higher cohomology of every twist $n \geq n_{0}$ . Rubric: full credit for finiteness and monotonicity; partial credit for finiteness alone with a complete justification.

Lean formalization [Intermediate+]

lean_status: none — Mathlib has the categorical sheaf-cohomology apparatus and projective space, but no packaged statement of Serre's vanishing-and-finiteness theorems on a projective scheme. The intended formalisation reads schematically:

[object Promise]

The proof gap is substantive on multiple fronts. The closed-embedding reduction requires the higher-direct-image vanishing for closed immersions and the projection formula. The line-bundle base case requires the explicit dimension table from 04.03.04 in its $A$ -module form. The descending induction requires Serre's theorem A (every coherent sheaf on $P_{A}^{N}$ admits a surjection from $O (- d)^{\oplus k}$ ) and the long-exact-sequence apparatus on cohomology. None of these are implemented in Mathlib as named pieces of a Serre-FAC arc. The Castelnuovo-Mumford regularity $reg (F)$ providing an explicit threshold is a separate formalisation target.

Advanced results [Master]

Theorem (Serre 1955, FAC §65; Hartshorne III.5.2, full statement). Let $A$ be a noetherian ring, $X$ a projective scheme over $A$ with very ample line bundle $O_{X} (1)$ , and $F$ a coherent $O_{X}$ -module. Then:

(Finiteness.) $H^{i} (X, F)$ is a finitely generated $A$ -module for every $i \geq 0$ , and $H^{i} (X, F) = 0$ for $i > dim X$ .
(Vanishing.) There exists $n_{0} = n_{0} (F) \in Z$ such that $H^{i} (X, F (n)) = 0$ for every $n \geq n_{0}$ and every $i > 0$ .

The proof, recorded above, runs through three ingredients: closed-embedding reduction to $X = P_{A}^{N}$ , the explicit dimension table for line bundles on projective space 04.03.04, and descending induction via short exact sequences from Serre's theorem A.

Castelnuovo-Mumford regularity. A coherent sheaf $F$ on $P_{A}^{N}$ is $m$ -regular (with respect to $O (1)$ ) when $H^{i} (P_{A}^{N}, F (m - i)) = 0$ for every $i \geq 1$ . The regularity index $reg (F)$ is the smallest such $m$ .

Theorem (Mumford, Lectures on Curves on an Algebraic Surface §14). If $F$ is $m$ -regular, then:

$F$ is $m^{'}$ -regular for every $m^{'} \geq m$ ;
$F (m)$ is generated by global sections;
the multiplication map $H^{0} (P_{A}^{N}, F (m)) \otimes H^{0} (P_{A}^{N}, O (d)) \to H^{0} (P_{A}^{N}, F (m + d))$ is surjective for every $d \geq 0$ .

The regularity index is therefore a quantitative refinement of Serre vanishing: $n_{0} (F) = reg (F)$ is the explicit threshold, and the regularity controls global generation and the multiplication structure on graded modules associated to $F$ . For an ideal sheaf $I_{Y} \subseteq O_{P^{N}}$ of a smooth projective subscheme $Y \subseteq P^{N}$ with Hilbert polynomial $P$ , $reg (I_{Y})$ is bounded above by an explicit polynomial in the degrees of the components of $P$ (Bayer-Stillman 1987, Bayer-Mumford 1993).

Relative Serre theorems (Grothendieck, EGA III). The absolute statements have a relative-morphism generalisation. For a proper morphism of noetherian schemes $f : X \to S$ and a coherent $O_{X}$ -module $F$ , the higher direct images $R^{i} f_{*} F$ are coherent $O_{S}$ -modules for every $i$ . If in addition $f$ is projective with a relative very ample $O_{X} (1)$ , there is an integer $n_{0}$ such that $R^{i} f_{*} F (n) = 0$ for every $i > 0$ and $n \geq n_{0}$ . The absolute case is recovered with $S = Spec A$ and $X$ projective over $A$ , with $R^{i} f_{*} F$ corresponding to $H^{i} (X, F)$ as an $A$ -module.

The relative finiteness theorem is the cornerstone of coherent base change: the locus on $S$ where the function $s \mapsto dim H^{i} (X_{s}, F_{s})$ jumps is closed, and $dim H^{i} (X_{s}, F_{s})$ is upper semi-continuous in $s$ . The Cohomology and Base Change theorem of Grothendieck (EGA III §7) provides the technical infrastructure for moduli problems, where the dimension of cohomology at fibres encodes the moduli-theoretic structure.

Serre duality as the dual statement. On a smooth projective scheme $X$ of dimension $n$ over an algebraically closed field $k$ , with canonical sheaf $ω_{X}$ , Serre duality reads $H^{i} (X, F) ≅ H^{n - i} (X, F^{\lor} \otimes ω_{X})^{*}$ for every locally free coherent sheaf $F$ . The duality identifies the "low cohomology" of $F$ with the "high cohomology" of $F^{\lor} \otimes ω_{X}$ , and Serre vanishing for one side translates into a vanishing assertion for the other under the dual hypothesis. Concretely, $L$ ample on $X$ gives Serre vanishing $H^{i} (X, F \otimes L^{n}) = 0$ for $i > 0$ and $n ≫ 0$ ; under Serre duality this translates to $H^{n - i} (X, F^{\lor} \otimes ω_{X} \otimes L^{- n}) = 0$ for $n - i < n$ and $n ≫ 0$ , recovering the dual non-vanishing pattern in the negative-twist range.

Kodaira vanishing as a strengthening. For a smooth projective variety $X$ over $C$ and an ample line bundle $L$ , the Kodaira vanishing theorem asserts $H^{i} (X, ω_{X} \otimes L) = 0 for every i > 0.$ This is stronger than Serre vanishing because the threshold is $n = 1$ — a single twist of $ω_{X}$ by an ample bundle suffices. Serre vanishing only gives a threshold $n_{0} ≫ 0$ depending on $ω_{X}$ and $L$ . The proof of Kodaira vanishing uses Hodge theory and the Lefschetz hyperplane theorem; over fields of positive characteristic the statement can fail (Raynaud 1978), so Kodaira's theorem is a characteristic-zero refinement of Serre vanishing.

Kawamata-Viehweg vanishing. A further refinement: for a smooth projective variety $X$ over $C$ and a $Q$ -divisor $Δ$ such that $L - Δ$ is nef and big, $H^{i} (X, ω_{X} \otimes L) = 0$ for $i > 0$ . The Kawamata-Viehweg theorem unlocks the Minimal Model Program by producing vanishing in situations where neither Serre nor Kodaira applies (the $Q$ -divisor case where $L$ is not strictly ample but only nef and big).

Serre's affineness criterion. A complementary cohomological statement: a noetherian scheme $X$ is affine if and only if $H^{i} (X, F) = 0$ for every coherent (equivalently, every quasi-coherent) sheaf $F$ and every $i \geq 1$ . The forward direction is Serre's affine vanishing theorem. The reverse uses the cohomological vanishing to construct enough functions to embed $X$ into affine space.

Synthesis. Serre's vanishing and finiteness theorems are the foundational finiteness statement of cohomological algebraic geometry: every coherent sheaf on a projective scheme has finite-dimensional cohomology that vanishes asymptotically under twisting. The proof reduces general coherent sheaves to line bundles via short exact sequences, and the dimension table for line bundles on projective space 04.03.04 supplies the explicit base case. Castelnuovo-Mumford regularity refines the asymptotic threshold to an explicit integer that controls global generation and the multiplication structure on graded modules. Grothendieck's relative finiteness in EGA III generalises the absolute statement to a proper morphism of noetherian schemes, providing the cohomological infrastructure for moduli problems and base change. Serre duality identifies the low cohomology of $F$ with the high cohomology of $F^{\lor} \otimes ω_{X}$ , turning Serre vanishing into a vanishing statement on each side of the duality. Kodaira vanishing strengthens the asymptotic vanishing to a one-twist statement on smooth projective varieties over $C$ , and Kawamata-Viehweg further extends to $Q$ -divisors. Putting these together, the foundational reason every projective scheme inherits a computable cohomology theory is the closed embedding $X ↪ P^{N}$ and the Hilbert syzygy resolution: a coherent sheaf admits a finite-length resolution by sums of line bundles, the dimension table on $P^{N}$ propagates through each stage, and the threshold $n_{0}$ is read off as the maximum across the building blocks. The same finiteness-and-vanishing pattern reappears in 04.04.01 (Riemann-Roch for curves) and in 06.04.04 (Serre duality on a curve), where the dual statement provides the basis-by-basis pairing that makes finite-dimensionality concrete.

Full proof set [Master]

Theorem (Serre 1955, FAC §65; Hartshorne III.5.2), full proof. Recall the setup: $A$ a noetherian ring, $X$ a projective scheme over $A$ with very ample $O_{X} (1)$ pulled back from a closed immersion $i : X ↪ P_{A}^{N}$ , $F$ a coherent $O_{X}$ -module.

Reduction to $P_{A}^{N}$ . For a closed immersion $i$ of locally noetherian schemes, the pushforward functor $i_{*} : QCoh (X) \to QCoh (P_{A}^{N})$ is exact, and the higher direct images $R^{p} i_{*} G = 0$ for $p \geq 1$ and every quasi-coherent $G$ . (Proof: $i$ is an affine morphism — locally, the inclusion of a closed subscheme into $Spec B$ is the inclusion of $Spec (B / I)$ , which is affine. Affine pushforwards have vanishing higher direct images on quasi-coherent sheaves.) The Leray spectral sequence $E_{2}^{p, q} = H^{p} (P_{A}^{N}, R^{q} i_{*} F) ⟹ H^{p + q} (X, F)$ degenerates on the column $q = 0$ , giving the isomorphism $H^{p} (X, F) ≅ H^{p} (P_{A}^{N}, i_{*} F)$ . The projection formula identifies $i_{*} (F (n)) = (i_{*} F) (n)$ . Replace $X$ by $P_{A}^{N}$ and $F$ by $i_{*} F$ .

Base case: $F = O (d)$ for some $d \in Z$ . The cohomology $H^{p} (P_{A}^{N}, O (d))$ is computed directly via Čech on the standard cover ${D_{+} (x_{i})}_{i = 0}^{N}$ 04.03.04. The graded $A$ -module description: $H^{0} (P_{A}^{N}, O (d)) = A [x_{0}, \dots, x_{N}]_{d}$ , free of rank $(N N + d)$ over $A$ for $d \geq 0$ and zero otherwise; $H^{N} (P_{A}^{N}, O (d))$ has a basis of Laurent monomials $x_{0}^{a_{0}} \dots x_{N}^{a_{N}}$ with all $a_{i} < 0$ and $\sum a_{i} = d$ , free of rank $(N - d - 1)$ over $A$ for $d \leq - N - 1$ and zero otherwise; $H^{p} (P_{A}^{N}, O (d)) = 0$ for $0 < p < N$ and for $p > N$ . Each $A$ -module is finitely generated (free, with explicit rank), and the vanishing thresholds for the higher cohomology are: $n_{0} = - d$ in degree $N$ , since $O (d) (n) = O (d + n)$ has vanishing $H^{N}$ for $d + n \geq - N$ ; in middle and high degrees, the cohomology is zero for every twist. Both finiteness and vanishing are direct.

Inductive step on cohomological degree. By Serre's theorem A (FAC §59), every coherent $F$ on $P_{A}^{N}$ admits a surjection $E_{0} := O (- d)^{\oplus k} ↠ F$ for some $d, k$ . The kernel $K$ is coherent (closure of coherent sheaves under kernels on a noetherian scheme), giving a short exact sequence $0 \to K \to E_{0} \to F \to 0$ . The long exact sequence in cohomology, twisted by $O (n)$ , reads $\dots \to H^{i - 1} (F (n)) \to H^{i} (K (n)) \to H^{i} (E_{0} (n)) \to H^{i} (F (n)) \to H^{i + 1} (K (n)) \to H^{i + 1} (E_{0} (n)) \to \dots$ .

Finiteness, descending induction on $i$ . The Grothendieck vanishing $H^{i} = 0$ for $i > N$ provides the base case (zero is finitely generated). For the inductive step at degree $i \leq N$ , the four-term sequence at $n = 0$ is $H^{i} (E_{0}) \to H^{i} (F) \to H^{i + 1} (K) \to H^{i + 1} (E_{0})$ . The two outer terms are finitely generated $A$ -modules — the leftmost from the line-bundle base case, the rightmost from the inductive hypothesis applied to $K$ in degree $i + 1$ . Apply the snake lemma / 5-lemma kernel-cokernel inheritance between four-term exact sequences over a noetherian ring: if the two outer terms are finitely generated, the cokernel of the left map is finitely generated, and the kernel of the right map (which equals the image of the middle map by exactness) is finitely generated. The middle term $H^{i} (F)$ fits into a short exact sequence $0 \to im (H^{i} (E_{0}) \to H^{i} (F)) \to H^{i} (F) \to ker (H^{i + 1} (K) \to H^{i + 1} (E_{0})) \to 0$ , both ends finitely generated, hence $H^{i} (F)$ finitely generated. The induction closes.

(The descending induction is a standard device for cohomology: the highest cohomological degree is handled directly by Grothendieck vanishing, and each lower degree is reduced to two adjacent degrees via the long exact sequence.)

Vanishing, descending induction on $i$ . For $i > N$ , the cohomology vanishes for every twist (Grothendieck), so the threshold is $n_{0} = - \infty$ in this range. For $1 \leq i \leq N$ , suppose vanishing has been established for every coherent sheaf in degrees $i + 1, i + 2, \dots, N$ with thresholds $n_{0} (G, j)$ depending on the sheaf $G$ . Apply the inductive hypothesis to $K$ in degree $i + 1$ : there is $n_{1} = n_{0} (K, i + 1)$ such that $H^{i + 1} (K (n)) = 0$ for $n \geq n_{1}$ . Apply the line-bundle base case to $O (n - d)$ : $H^{i} (O (n - d)) = 0$ for $n \geq d - N + 1$ if $i = N$ , or for every $n$ if $0 < i < N$ . Set $n_{2} = d - N + 1$ if $i = N$ and $n_{2} = - \infty$ otherwise. The four-term sequence $H^{i} (E_{0} (n)) \to H^{i} (F (n)) \to H^{i + 1} (K (n))$ becomes $0 \to H^{i} (F (n)) \to 0$ for $n \geq max (n_{1}, n_{2})$ , hence $H^{i} (F (n)) = 0$ in this range. Set $n_{0} (F, i) := max (n_{1}, n_{2})$ and $n_{0} (F) := max_{1 \leq i \leq N} n_{0} (F, i)$ . For $n \geq n_{0} (F)$ , every $H^{i} (F (n))$ vanishes for $1 \leq i \leq N$ , and Grothendieck handles $i > N$ . The induction closes simultaneously for all coherent $F$ on $P_{A}^{N}$ . $□$

Theorem (Serre theorem A, FAC §59). Let $A$ be noetherian and $F$ a coherent sheaf on $P_{A}^{N}$ . There exists $d_{0} = d_{0} (F)$ such that for every $d \geq d_{0}$ , $F (d)$ is generated by finitely many global sections.

Proof of theorem A. Cover $P_{A}^{N}$ by the standard affines ${D_{+} (x_{i})}_{i = 0}^{N}$ . On each $D_{+} (x_{i}) ≅ Spec A [y_{0}^{(i)}, \dots, y_{N}^{(i)}]$ (with $y_{j}^{(i)} = x_{j} / x_{i}$ , removing the index $j = i$ ), the restriction $F ∣_{D_{+} (x_{i})}$ corresponds to a finitely generated $A [y_{*}^{(i)}]$ -module $M_{i}$ . Each $M_{i}$ has a finite generating set ${m_{i, k}}_{k = 1}^{r_{i}}$ . Each generator $m_{i, k}$ lifts to a global section of $F (d_{i, k})$ for some $d_{i, k} \geq 0$ via the standard graded-module-to-sheaf correspondence — explicitly: $m_{i, k}$ as an element of $M_{i} = F (D_{+} (x_{i}))$ admits a description by a Laurent polynomial in $x_{0}, \dots, x_{N}$ with denominator a power of $x_{i}$ , and multiplying by $x_{i}^{d_{i, k}}$ for some $d_{i, k}$ produces a global section of $F (d_{i, k})$ that restricts to $x_{i}^{d_{i, k}} m_{i, k}$ on $D_{+} (x_{i})$ . Setting $d_{0} := max_{i, k} d_{i, k}$ , the global sections of $F (d_{0})$ generate the local module on each $D_{+} (x_{i})$ (after multiplication by an appropriate power of $x_{i}$ , which is a unit on $D_{+} (x_{i})$ ). For $d \geq d_{0}$ , the same global sections, twisted further, continue to generate. So $F (d)$ is globally generated for $d \geq d_{0}$ , proving theorem A. $□$

Theorem (cohomological characterisation of ampleness — Hartshorne III.5.3, full proof). Let $X$ be a noetherian projective scheme over a noetherian ring $A$ and $L$ a line bundle on $X$ . Then $L$ is ample if and only if for every coherent sheaf $F$ there is $n_{0} (F)$ such that $H^{i} (X, F \otimes L^{n}) = 0$ for $i > 0$ and $n \geq n_{0} (F)$ .

The two directions are recorded in Exercise 7. Forward: ampleness gives $L^{m}$ very ample for some $m \geq 1$ , and Serre vanishing on the very ample bundle $L^{m}$ produces the threshold for multiples of $m$ , with the residue range handled by twisting by $L^{j}$ for each $j \in {0, \dots, m - 1}$ . Reverse: the vanishing condition on ideal sheaves of points and infinitesimal neighbourhoods gives global generation, separation of points, and separation of tangent vectors for $L^{n}$ at $n ≫ 0$ , hence very ampleness for some power, hence ampleness of $L$ . The reverse direction uses the noetherian-quasi-compactness of $X$ to upgrade "global generation at each closed point" to "uniform global generation across a single threshold". $□$

Theorem (Castelnuovo-Mumford regularity, Mumford 1966). Let $F$ be a coherent sheaf on $P_{A}^{N}$ . The regularity index $reg (F) := min {m : F is m -regular}$ is finite, and if $F$ is $m$ -regular then $F$ is $m^{'}$ -regular for every $m^{'} \geq m$ . Moreover, $F (m)$ is generated by global sections for $m \geq reg (F)$ , and the multiplication map $H^{0} (F (m)) \otimes H^{0} (O (d)) \to H^{0} (F (m + d))$ is surjective for $m \geq reg (F)$ and $d \geq 0$ .

The finiteness and monotonicity arguments are recorded in Exercise 8: finiteness follows from Serre vanishing applied at each cohomological degree separately; monotonicity uses a generic hyperplane section and the long exact sequence $0 \to F (- 1) \to F \to F ∣_{H} \to 0$ .

The global-generation assertion: for $m \geq reg (F)$ , the long exact sequence for the closed point $p \in P_{A}^{N}$ is $\dots \to H^{0} (F (m)) \to F (m) ∣_{p} \to H^{1} (I_{p} \otimes F (m)) \to \dots$ and the right-hand $H^{1}$ vanishes because the ideal sheaf $I_{p} \otimes F (m)$ has regularity bounded by $reg (F) + 1$ (a fact used in the iterative regularity bound). So $H^{0} (F (m)) ↠ F (m) ∣_{p}$ , giving global generation at $p$ ; varying $p$ gives global generation everywhere.

The surjectivity of the multiplication map: the analogous long exact sequence applied to the multiplication map $F (m) \otimes O (d) \to F (m + d)$ on global sections, combined with the regularity of $F$ , gives surjectivity for $d \geq 0$ . The detailed bookkeeping is in Mumford §14. $□$

Connections [Master]

Cohomology of line bundles on projective space 04.03.04. The dimension table for $H^{p} (P_{A}^{N}, O (d))$ is the explicit base case that drives the entire reduction: Serre's vanishing-and-finiteness theorems for coherent sheaves on a projective scheme are the descending-induction extension of the line-bundle table to general coherent sheaves, propagated through short exact sequences. Without the closed-form table, the theorems would be existence assertions; with the table, they become quantitative.
Čech cohomology of sheaves on schemes 04.03.03. The reduction to $P_{A}^{N}$ uses Cartan's comparison theorem to compute cohomology via Čech on the standard cover, and the line-bundle base case is itself a direct Čech computation. The descending induction on cohomological degree is invisible in the Čech picture — it appears only after the long exact sequence in derived-functor cohomology — so the Čech machinery sets up the calculation, and the abstract long exact sequence drives the inductive argument.
Riemann-Roch theorem for curves 04.04.01. Riemann-Roch $χ (C, L) = de g L + 1 - g$ for a line bundle $L$ on a smooth projective curve is the explicit dimension formula that follows from Serre vanishing in the curve case. Once $de g L ≫ 0$ , $H^{1} (C, L) = 0$ by Serre vanishing, and $χ$ collapses to $dim H^{0}$ , which is computed explicitly from the genus and degree. The Riemann-Roch formula is the dimension-count consequence of finiteness, with Serre vanishing ensuring the formula extrapolates to the asymptotic regime where one cohomology group is forced to zero.
Ample line bundles and projective embeddings 04.05.05. The cohomological characterisation of ampleness (Hartshorne III.5.3) gives an alternative definition of ampleness purely in terms of vanishing: $L$ is ample if and only if Serre vanishing holds for every coherent $F$ with respect to $L^{n}$ . This is the bridge between the geometric definition (some power gives a closed embedding) and the cohomological one (vanishing of higher cohomology asymptotically), and it is the entry point for Kodaira's vanishing theorem and the Minimal Model Program.
Serre duality 04.03.06 pending (planned). The dual statement to Serre vanishing on a smooth projective scheme: $H^{i} (X, F)$ pairs perfectly with $H^{n - i} (X, F^{\lor} \otimes ω_{X})$ . Vanishing of one side is vanishing of the other, and finite-dimensionality of one side is finite-dimensionality of the other. The duality identifies the asymptotic vanishing pattern of low cohomology under positive twists with the asymptotic vanishing pattern of high cohomology under negative twists.
Sheaf cohomology survey 06.04.07. The Riemann-surface counterpart of Serre vanishing is recorded as part of the standard cohomology table on a compact Riemann surface of genus $g$ : for a line bundle $L$ of degree $\geq 2 g - 1$ , $H^{1} (X, L) = 0$ , the curve case of Serre vanishing. The four-pictures-agree mantra of the survey extends Serre's theorems through Cartan-Leray, Dolbeault, harmonic, and derived-functor pictures, with each picture carrying its own version of the finiteness-and-vanishing statement.
Hilbert scheme. The relative finiteness theorem of Grothendieck (EGA III) is the foundational input for the construction of the Hilbert scheme: the dimension function $s \mapsto dim H^{i} (X_{s}, F_{s})$ on a flat family is upper semi-continuous and the locus of constant Hilbert polynomial cuts out a representable functor. Without Serre's finiteness theorem there would be no Hilbert polynomial; without the relative version there would be no Hilbert scheme.

Historical & philosophical context [Master]

The vanishing and finiteness theorems originate with Jean-Pierre Serre's 1955 paper Faisceaux Algébriques Cohérents (Annals of Mathematics 61, 197-278), the founding paper of sheaf-cohomological algebraic geometry. Serre introduced the framework of coherent sheaves on a complex algebraic variety, defined cohomology via Čech on a finite affine cover, and proved theorems A (global generation under sufficient twisting), B (vanishing of higher cohomology on affines), and the consolidated finiteness-and-vanishing theorem on projective varieties as the foundational results of the new theory. The same paper proved Serre duality on smooth projective varieties; the projective-space dimension table served as the model calculation against which the general theorems were tested.

Robin Hartshorne's Algebraic Geometry (Springer GTM 52, 1977) codified the calculations as Theorems III.5.2 and III.5.3, with the closed-embedding reduction to $P_{A}^{N}$ and the descending induction via short exact sequences becoming the canonical pedagogical version. Hartshorne's framing — coherent sheaves on noetherian schemes, very ample line bundles, twisting by $O_{X} (1)$ , Serre's theorem A as the primitive surjection from a finite direct sum of line bundles — is the form learned by every contemporary algebraic geometer. The proof structure is unchanged from Serre 1955 but presented in scheme-theoretic language rather than the original analytic-coherent-sheaf framework.

Alexander Grothendieck's Éléments de géométrie algébrique III (with Jean Dieudonné, Publ. Math. IHÉS 11 and 17, 1961-63) established the relative version: for a proper morphism $f : X \to S$ of noetherian schemes and a coherent $F$ , the higher direct images $R^{i} f_{*} F$ are coherent. The relative finiteness theorem is the foundational input for moduli problems and base change, and it generalises Serre's absolute statement by replacing $S = Spec A$ with an arbitrary noetherian base. Grothendieck's framework allows the projective hypothesis on $X$ to be relaxed to properness, with the very ample line bundle replaced by relative ampleness.

David Mumford's Lectures on Curves on an Algebraic Surface (Annals of Math. Studies 59, 1966) introduced the regularity index in §14 — Mumford attributes the concept to a 1893 paper of Castelnuovo on linear systems, hence the name Castelnuovo-Mumford regularity. The regularity index gives an explicit threshold for Serre vanishing: $reg (F)$ is the smallest $m$ for which $F$ is $m$ -regular, and once $F$ is $m$ -regular, $F (m)$ is globally generated and the multiplication map on graded modules is surjective. Bayer-Stillman 1987 and Bayer-Mumford 1993 produced explicit polynomial bounds for $reg (I_{Y})$ for ideal sheaves of subschemes with given Hilbert polynomial.

Kunihiko Kodaira proved the vanishing theorem $H^{i} (X, ω_{X} \otimes L) = 0$ for $i > 0$ and $L$ ample on a smooth projective complex variety in 1953 (On a differential-geometric method in the theory of analytic stacks, Proc. Nat. Acad. Sci. USA 39, 1268-1273). Kodaira's proof uses Hodge theory; the theorem fails in positive characteristic (Raynaud 1978), so it is a characteristic-zero strengthening of Serre's asymptotic vanishing. Yujiro Kawamata and Eckart Viehweg independently extended the theorem in 1982 to nef-and-big $Q$ -divisors, providing the cohomological vanishing theorem that drives the Minimal Model Program.

[object Promise]

Bibliography [Master]

[object Promise]

Prerequisites

04.03.03
04.03.04

Tier anchors

beginner: Hartshorne §III.5 informal — twisting by $\mathcal{O}(n)$ for large $n$ kills higher cohomology
intermediate: Hartshorne §III.5 + Theorems III.5.2 and III.5.3
master: Serre 1955 *Faisceaux Algébriques Cohérents* (Annals of Math. 61); Hartshorne §III.5; Grothendieck-Dieudonné EGA III §2; Mumford *Lectures on Curves on an Algebraic Surface* §14 (Castelnuovo-Mumford regularity)

References

TODO_REF
Hartshorne — Algebraic Geometry · §III.5 Theorems 5.2 and 5.3 (Serre vanishing and finiteness for projective coherent sheaves)
TODO_REF
Serre — Faisceaux Algébriques Cohérents · Annals of Math. 61 (1955) 197-278, originator: Theorems A, B and the vanishing/finiteness theorems on projective varieties
TODO_REF
Grothendieck-Dieudonné — EGA III · Publ. Math. IHÉS 11, 17 (1961-63), §2 cohomologie des faisceaux cohérents (relative finiteness and vanishing for proper morphisms)
TODO_REF
Mumford — Lectures on Curves on an Algebraic Surface · Annals of Math. Studies 59 (1966), §14 Castelnuovo-Mumford regularity
TODO_REF
Liu — Algebraic Geometry and Arithmetic Curves · §5.2 Serre's theorems on cohomology of projective schemes

Reviewer

TBD

Estimated time

beginner: 18m
intermediate: 45m
master: 75m