Serre duality on a curve

Intuition [Beginner]

Take a smooth projective curve $X$ — a compact Riemann surface, viewed algebraically — of genus $g$. A line bundle $L$ on $X$ is a way of attaching a one-dimensional vector space to each point so that nearby fibres are smoothly identified. Serre duality says the answer to "what global obstruction prevents $L$ from being captured by its global sections" has a name: it is the dimension of the space of global sections of a different, related bundle.

The related bundle is the canonical-twisted dual of $L$ — formed by combining the bundle of holomorphic 1-forms on $X$ (called the canonical bundle) with the dual of $L$. Serre duality is a precise dictionary: failure-to-be-captured-by-sections of $L$ matches global-sections-vanishing-on-$L$ of the canonical-twisted-by-the-dual.

Two finite-dimensional vector spaces sit at opposite ends of the cohomology of $L$. Serre duality says they are dual to each other, paired by the residue calculus on $X$.

Visual [Beginner]

A schematic of a smooth projective curve $X$ of genus $g$, with a line bundle $L$ depicted by a varying-thickness ribbon along the curve, and the canonical-twisted-dual line bundle drawn as a complementary ribbon. An arrow labelled "residue pairing" connects sections of one ribbon to first-cohomology of the other.

Worked example [Beginner]

Take the Riemann sphere $X={\mathbb{P}}^1$, genus $g=0$. The canonical bundle is $K_X=\mathcal{O}(-2)$. Pick the line bundle $L=\mathcal{O}(-3)$.

Global sections of $L$: polynomials of degree $\le -3$, of which there are none. So the dimension of global sections is $0$.

Global sections of the canonical-twisted dual: this is the line bundle of degree $(-2)-(-3)=1$ on the sphere, namely $\mathcal{O}(1)$. Its global sections are linear polynomials, which form a 2-dimensional space.

Serre duality predicts the first cohomology of $L=\mathcal{O}(-3)$ is dual to that 2-dimensional space, so it is also 2-dimensional. The Riemann-Roch dimension count $\dim H^0-\dim H^1=\mathrm{deg}L+1-g=-3+1-0=-2$ confirms this: $0-2=-2$.

What this tells us: on ${\mathbb{P}}^1$, line bundles of negative degree have no global sections but carry first cohomology equal in dimension to the global sections of the dual-twisted canonical. The two sides of Serre duality are concrete dimension counts, computable for every line bundle in this family.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a smooth projective curve over an algebraically closed field $k$, of genus $g={\dim }_k{H}^0(X,{\omega }_X)$, where ${\omega }_X={\Omega }_{X/k}^1$ is the canonical sheaf: the sheaf of Kähler differentials, locally free of rank one because $X$ is smooth of relative dimension one. Write $K_X$ for this sheaf when emphasising its line-bundle structure; $\mathrm{deg}{K}_X=2g-2$.

For a line bundle (invertible sheaf) $L$ on $X$, the Serre-duality pairing is the composite

$$H^1(X,L){\otimes }_k{H}^0(X,K_X\otimes L^{-1})\stackrel{⌣}{\to }{H}^1(X,K_X)\stackrel{\mathrm{tr}}{\to }k,$$

where the first map is the cup product on Čech cohomology composed with the natural multiplication of sections $L\otimes (K_X\otimes L^{-1})\to K_X$, and $\mathrm{tr}$ is the trace (residue) map sending $H^1(X,K_X)\to k$ via summation of local residues over $X$.

Theorem (Serre duality on a curve). The pairing above is a perfect pairing of finite-dimensional $k$-vector spaces; equivalently, the induced map

$$H^1(X,L)\to H^0(X,K_X\otimes L^{-1}{)}^{\ast }$$

is an isomorphism, natural in $L$.

Equivalent forms.

• Cohomological symmetry: ${\dim }_k{H}^1(X,L)={\dim }_k{H}^0(X,K_X\otimes L^{-1})$.
• Riemann-Roch reformulation: combining Serre duality with 06.04.01, ${\dim }_k{H}^0(X,L)-{\dim }_k{H}^0(X,K_X\otimes L^{-1})=\mathrm{deg}L+1-g$.
• Trace isomorphism: $H^1(X,K_X)\cong k$ canonically, the trace giving the identification.

Counterexamples to common slips.

• $H^1(X,L)$ does not pair with $H^0(X,L^{-1})$ in general; the canonical twist $K_X\otimes L^{-1}$ is essential.
• The non-degeneracy is a statement over $k$, not over the structure sheaf; both sides are finite-dimensional $k$-vector spaces, not ${\mathcal{O}}_X$-modules.
• Degree alone does not determine $L$; two non-isomorphic line bundles of the same degree have isomorphic Euler characteristics but different cohomology dimensions in general (only the difference $h^0-h^1$ depends on degree).

Key theorem with proof [Intermediate+]

Theorem (Serre 1955, Serre duality on a smooth projective curve). Let $X$ be a smooth projective curve of genus $g$ over an algebraically closed field $k$, and let $L$ be a line bundle on $X$. Then the residue pairing

$$H^1(X,L)\times H^0(X,K_X\otimes L^{-1})⟶k$$

is non-degenerate. Equivalently, $H^1(X, L) \cong H^0(X, K_X \otimes L^{-1})^$ canonically.*

Proof. The argument has four steps: define the trace, define the pairing, reduce non-degeneracy to the case $L={\mathcal{O}}_X$, and verify that case via the Hodge decomposition of compact Riemann surfaces.

Step 1 — the trace map. Choose a Čech open cover $\mathfrak{U}=\{U_{\alpha }\}$ of $X$ by affine opens. A class in $H^1(X,K_X)$ is represented by a Čech 1-cocycle $\eta =({\eta }_{\alpha \beta })$ with ${\eta }_{\alpha \beta }\in K_X(U_{\alpha }\cap U_{\beta })$, modulo coboundaries. For each pair $(\alpha ,\beta )$ the section ${\eta }_{\alpha \beta }$ is a meromorphic 1-form on $U_{\alpha }\cap U_{\beta }$ that extends rationally across $X$. The trace is

$$\mathrm{tr}(\eta )=\sum _{p\in X}{\mathrm{Res}}_p(\eta ),$$

where ${\mathrm{Res}}_p(\eta )$ is the residue of any meromorphic 1-form representing the local cohomology class at $p$. Independence of representative follows from the residue theorem: a coboundary ${\eta }_{\alpha \beta }={\omega }_{\beta }-{\omega }_{\alpha }$ for global sections ${\omega }_{\alpha }$ contributes residues that telescope to zero. The trace is well-defined and $k$-linear. The fact that $\mathrm{tr}$ is a non-zero linear functional on the one-dimensional space $H^1(X,K_X)$ uses that $X$ is connected and projective; it makes $\mathrm{tr}$ an isomorphism $H^1(X,K_X)\cong k$.

Step 2 — the pairing. For $\xi \in H^1(X,L)$ represented by a 1-cocycle $({\xi }_{\alpha \beta })$ with ${\xi }_{\alpha \beta }\in L(U_{\alpha \beta })$, and $\sigma \in H^0(X,K_X\otimes L^{-1})$ a global section, the local product $\sigma \cdot {\xi }_{\alpha \beta }$ is a section of $K_X$ over $U_{\alpha \beta }$, and the family $(\sigma \cdot {\xi }_{\alpha \beta })$ is a Čech 1-cocycle in $K_X$. Its cohomology class is independent of the choice of representative cocycle for $\xi$ — this is the Čech-cohomology cup product paired with sheaf multiplication. The pairing is

$$⟨\xi ,\sigma ⟩:=\mathrm{tr}(\sigma \cdot \xi )\in k.$$

Step 3 — reduction to $L={\mathcal{O}}_X$ via twisting. Pick a point $p\in X$ and consider the short exact sequence of sheaves on $X$:

$$0\to L\to L(p)\to L(p){|}_p\to 0,$$

where $L(p)=L\otimes {\mathcal{O}}_X(p)$ and $L(p){|}_p$ is a skyscraper at $p$ with stalk a one-dimensional $k$-vector space. Tensoring with the canonical and dualising in $L$ produces a parallel short exact sequence relating $K_X\otimes L^{-1}$ and $K_X\otimes L^{-1}(-p)$. The pairings on the two sequences are compatible — the cup product and trace are functorial — so non-degeneracy for $L$ is equivalent to non-degeneracy for $L(p)$. By induction on the divisor class of $L$, every line bundle is connected via such single-point twists to ${\mathcal{O}}_X$. Showing the pairing is non-degenerate for ${\mathcal{O}}_X$ implies the result for every $L$.

Step 4 — the case $L={\mathcal{O}}_X$. The pairing becomes

$$H^1(X,{\mathcal{O}}_X)\times H^0(X,K_X)\to H^1(X,K_X)\cong k.$$

In characteristic zero (or via the algebraic-de-Rham comparison in general), the Hodge decomposition $H_{\mathrm{dR}}^1(X)\cong H^1(X,{\mathcal{O}}_X)\oplus H^0(X,K_X)$ holds for compact Riemann surfaces (Hodge 1941), and the cup-product pairing $H_{\mathrm{dR}}^1(X)\times H_{\mathrm{dR}}^1(X)\to H_{\mathrm{dR}}^2(X)=k$ restricts to a perfect pairing between the two summands. The Hodge perfect-pairing reads exactly as the Serre duality pairing for $L={\mathcal{O}}_X$, completing the proof. $\square$

The four-step structure is faithful to Donaldson §11; Hartshorne §IV.1 reorganises the same argument around derived functors and the Yoneda-extension viewpoint, but the underlying content — trace, cup product, twist-reduction, $L={\mathcal{O}}_X$ base case — is identical.

Bridge. The duality proven here builds toward 06.06.06 (Jacobi inversion theorem), where the residue pairing $H^1(X,K_X)\cong k$ controls the period integrals defining the Abel-Jacobi map and the perfect-pairing structure of the period matrix on the Jacobian. The same residue-trace machinery appears again in the theta-divisor calculation behind Riemann's vanishing theorem — the geometric content of "the theta divisor is the image of ${\mathrm{Sym}}^{g-1}(X)$ shifted by the Riemann constant" is a Serre-duality computation at the level of line bundles on $X$. Combined with Riemann-Roch 06.04.01, Serre duality is what converts the Euler-characteristic identity into a sharp two-sided dimension count. Putting these together, the foundational insight is that on a smooth projective curve, the cohomology of every line bundle is controlled by the global sections of its canonical-twisted dual — and that the bridge is the residue calculus on a fixed compact Riemann surface.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not currently formalise the canonical sheaf on a smooth projective curve, the residue trace map $H^1(X;{\omega }_X)\to k$, or the cup-product pairing into $H^1(X;{\omega }_X)$. A proposed signature, in Lean 4 / Mathlib syntax, sketching the target statement:

[object Promise]

The proof depends on names that do not currently exist in Mathlib (the canonical sheaf on a smooth projective curve, Čech cohomology of an invertible sheaf with trace, the cup-product pairing with values in $H^1(X;{\omega }_X)$, the trace isomorphism $H^1(X;{\omega }_X)\cong k$). Each is a candidate Mathlib contribution; until then this unit ships with lean_status: none.

Advanced results [Master]

Serre duality on a curve is the dimension-1 case of a duality theorem in every dimension. The general formulation, due to Serre 1955 for smooth projective varieties and refined by Grothendieck-Hartshorne 1966 to proper morphisms via the dualising complex, states that for a smooth projective variety $X$ of dimension $n$ over a field $k$ and a coherent sheaf $\mathcal{F}$, the trace map $H^n(X,{\omega }_X)\to k$ pairs $H^i(X,\mathcal{F})\otimes H^{n-i}(X,{\mathcal{F}}^{\vee }\otimes {\omega }_X)\to H^n(X,{\omega }_X)\to k$ non-degenerately when $\mathcal{F}$ is locally free. For $n=1$ this is the curve case proved above; for higher $n$ the proof passes through the Koszul complex and the Yoneda construction of Ext.

Vanishing for non-special line bundles. For $L$ a line bundle on a smooth projective curve of genus $g$ with $\mathrm{deg}L>2g-2$, $\mathrm{deg}(K_X\otimes L^{-1})<0$ forces $H^0(X,K_X\otimes L^{-1})=0$, so $H^1(X,L)=0$ by Serre duality. Riemann-Roch then collapses to the explicit formula $\dim H^0(X,L)=\mathrm{deg}L+1-g$, with no first-cohomology correction. This vanishing is the curve case of Kodaira vanishing.

Specialty and Brill-Noether theory. A divisor $D$ on $X$ is special when the corresponding line bundle $\mathcal{O}(D)$ has $\dim H^0(X,K_X\otimes \mathcal{O}(-D))>0$, equivalently $\dim H^1(X,\mathcal{O}(D))>0$. Serre duality identifies the index of speciality $i(D)=\dim H^0(X,K_X\otimes \mathcal{O}(-D))$ as the cohomological obstruction to $\mathcal{O}(D)$ being a complete linear system. The Brill-Noether locus $W_d^r\subset {\mathrm{Pic}}^d(X)$ parameterises line bundles $L$ of degree $d$ with $\dim H^0(X,L)\ge r+1$; Serre duality forces $W_d^r\cong W_{2g-2-d}^{g-d+r-1}$ via $L\mapsto K_X\otimes L^{-1}$, exhibiting a perfect bilateral symmetry on the Picard variety. The Brill-Noether existence theorem (Kempf-Kleiman-Laksov 1971-74) gives $\dim W_d^r\ge \rho =g-(r+1)(g-d+r)$ when this quantity is non-negative; on a generic curve, equality holds and $W_d^r$ is empty when $\rho <0$.

Clifford's inequality. For a special line bundle $L$ on a smooth projective curve, $\dim H^0(X,L)\le \mathrm{deg}(L)/2+1$, with equality iff $L={\mathcal{O}}_X$, $L=K_X$, or $X$ is hyperelliptic and $L$ is a power of the hyperelliptic involution divisor. The proof combines Serre duality with the multiplication-of-sections map $H^0(L)\otimes H^0(K-L)\to H^0(K)$, exhibited in Exercise 6. Clifford's inequality is the curve case of a general dimension bound for special divisors.

Generalisations to higher dimensions. For a smooth projective surface $S$, Serre duality reads $H^i(S,\mathcal{F})\cong H^{2-i}(S,{\mathcal{F}}^{\vee }\otimes K_S{)}^{\ast }$, with the trace map $H^2(S,K_S)\to k$ given by the integration of canonical 2-forms. The Hirzebruch-Riemann-Roch formula and the Noether formula on surfaces are the dimension-2 analogues of the curve Riemann-Roch and use Serre duality at $i=2$. The Koszul complex on a complete intersection of any dimension reduces higher-dimensional Serre duality to the projective space case, where it can be computed directly.

Coherent duality and the dualising complex. Grothendieck-Hartshorne 1966 Residues and Duality generalises Serre duality to proper morphisms $f:X\to Y$ between noetherian schemes. The right-derived functor $f^{!}$ of $f^{\ast }$ produces a dualising complex ${\omega }_X^{\bullet }=f^{!}{\mathcal{O}}_Y$ in the derived category of $X$. For a smooth projective $X$ of dimension $n$ over a field, ${\omega }_X^{\bullet }\cong {\omega }_X[n]$, recovering the Serre version. The general framework powers Verdier duality on real and complex manifolds, $\ell$-adic duality on étale schemes, and the Serre-Grothendieck duality on derived stacks.

Synthesis. Serre duality on a curve is the engine that converts the Riemann-Roch equality $\chi (L)=\mathrm{deg}L+1-g$ into a sharp two-sided dimension count: the cohomological obstruction $\dim H^1(X,L)$ is itself a global-section dimension, computable by inspection on the canonical-twisted-dual side. Read in the opposite direction, the trace map $H^1(X,K_X)\cong k$ is the residue-theorem identification of compact-curve cohomology with the base field, organising every linear-algebraic statement about line bundles on $X$ into a single perfect pairing. The geometric content of "specialty" — the Brill-Noether stratification of $\mathrm{Pic}(X)$ — is exactly the Serre-duality-induced bilateral symmetry $L\mapsto K_X\otimes L^{-1}$ on the Picard variety, with the canonical class $K_X$ acting as the centre of symmetry. Putting these together, the cohomology of every line bundle on a smooth projective curve is determined by the global sections of two related bundles ($L$ itself and $K_X\otimes L^{-1}$) and one numerical invariant ($\mathrm{deg}L$); the Serre duality theorem is the identification of these data with the full cohomology, and the dimension-1 case generalises to the coherent-duality framework that organises the cohomology of every smooth projective variety.

Full proof set [Master]

Lemma (residue theorem on a smooth projective curve). For a meromorphic 1-form $\omega$ on a smooth projective curve $X$ over an algebraically closed field $k$, ${\sum }_{p\in X}{\mathrm{Res}}_p(\omega )=0$.

Proof. On ${\mathbb{P}}^1$ the identity is a direct computation in coordinates: a meromorphic 1-form $\omega =f(z)dz$ with $f$ rational has residues at the finite poles given by partial-fraction expansions, and the residue at $\infty$ in the chart $w=1/z$ is $-{\mathrm{Res}}_{\infty }^{(z)}(f(z)dz/z^2)\cdot$ sign-correction; the sum vanishes by clearing denominators. For a general $X$, choose a finite morphism $f:X\to {\mathbb{P}}^1$ (exists by Riemann-Roch applied to a sufficiently positive line bundle); the trace $f_{\ast }\omega$ is a meromorphic 1-form on ${\mathbb{P}}^1$ whose residues at each point of ${\mathbb{P}}^1$ equal the sum of residues of $\omega$ over the preimage points. Summing over ${\mathbb{P}}^1$ and applying the ${\mathbb{P}}^1$ case completes the argument. $\square$

Lemma (trace isomorphism for ${\omega }_X$). On a smooth projective curve $X$ of genus $g$ over an algebraically closed field $k$, the trace map $\mathrm{tr}:H^1(X,K_X)\to k$ is an isomorphism.

Proof. The Riemann-Roch 06.04.01 computation with $L=K_X$ gives $\chi (K_X)=\mathrm{deg}{K}_X+1-g=(2g-2)+1-g=g-1$. Together with $\dim H^0(X,K_X)=g$, this yields $\dim H^1(X,K_X)=1$. The residue theorem ensures the trace is well-defined on coboundaries (these have residue sums zero by the residue theorem) and non-zero on at least one cocycle (a cocycle representing the cohomology class of $dz/z$ near a chosen point has residue $1$). A non-zero $k$-linear map on a 1-dimensional $k$-vector space is an isomorphism. $\square$

Lemma (cup-product compatibility with twisting). Let $0\to \mathcal{F}\to \mathcal{G}\to \mathcal{H}\to 0$ be a short exact sequence of coherent sheaves on $X$, and let ${\mathcal{F}}^{\prime },{\mathcal{G}}^{\prime },{\mathcal{H}}^{\prime }$ be the corresponding $K_X$-twisted dual sheaves. The Serre-duality pairing for $\mathcal{F}$ and the Serre-duality pairing for $\mathcal{G}$ are connected by the long-exact-sequence connecting maps and Yoneda compatibility.

Proof. Functoriality of Čech cup product across short exact sequences and the naturality of the trace map give the diagram-level statement; both pairings are non-degenerate iff one is, since the long exact sequences in cohomology are dual to each other in the relevant sense. The verification reduces to the case where $\mathcal{H}$ is a skyscraper, treated explicitly in Donaldson §11 and Hartshorne §IV.1. $\square$

Theorem (Serre duality on a curve, full statement). Statement and proof as in the Intermediate-tier Key theorem section.

Proof. The Intermediate-tier proof goes through using the three lemmas above as packaged inputs: trace exists and is non-zero (residue theorem); $H^1(X,K_X)\cong k$ (Lemma 2); pairings compose well with twisting (Lemma 3). The base case $L={\mathcal{O}}_X$ uses the Hodge decomposition of compact Riemann surfaces over $\mathbb{C}$ and the algebraic-de-Rham comparison in positive characteristic; in both settings the pairing $H^1(X,{\mathcal{O}}_X)\times H^0(X,K_X)\to k$ is the Hodge perfect pairing on the rank-$2g$ first cohomology. $\square$

Corollary (vanishing). For $L$ a line bundle on $X$ with $\mathrm{deg}L>2g-2$, $H^1(X,L)=0$.

Proof. $\mathrm{deg}(K_X\otimes L^{-1})=(2g-2)-\mathrm{deg}L<0$. A line bundle of negative degree on a connected smooth projective curve has no global sections (a non-zero global section would be a nowhere-zero rational function with zeros only, of total degree $\le 0$, impossible). Apply Serre duality. $\square$

Corollary (Riemann-Roch reformulation). For every line bundle $L$ on $X$, $\dim H^0(X,L)-\dim H^0(X,K_X\otimes L^{-1})=\mathrm{deg}L+1-g$.

Proof. Riemann-Roch 06.04.01 gives $\dim H^0(L)-\dim H^1(L)=\mathrm{deg}L+1-g$. Substitute $\dim H^1(L)=\dim H^0(K_X\otimes L^{-1})$ from Serre duality. $\square$

Connections [Master]

• Riemann-Roch theorem for compact Riemann surfaces 06.04.01. Serre duality is the second component of Riemann-Roch in its sharpest form. The pair $(\text{Riemann-Roch},\text{Serre duality})$ converts the Euler characteristic $\chi (L)=\mathrm{deg}L+1-g$ into the explicit dimension count $\dim H^0(L)-\dim H^0(K_X\otimes L^{-1})=\mathrm{deg}L+1-g$. Together they are the foundational dimension-counting theorem for line bundles on curves.

• Holomorphic line bundle on a Riemann surface 06.05.02. The objects of Serre duality are line bundles on a compact Riemann surface; the duality pairing is a map between the cohomology of one bundle and the global sections of a second, geometrically determined bundle.

• Holomorphic 1-form 06.06.01. The canonical bundle $K_X$ of holomorphic 1-forms is the load-bearing object on the right side of the Serre-duality pairing. Global sections of $K_X$ give the genus $g$, and the trace map is residue summation of meromorphic 1-forms.

• Period matrix 06.06.02. The period integrals of holomorphic 1-forms over the homology basis of $X$ assemble into a matrix in the Siegel upper half space; the integrality and symmetry conditions on the period matrix (Riemann's bilinear relations) are direct consequences of the residue-pairing structure that powers Serre duality.

• Jacobian variety 06.06.03. The Jacobian $J(X)={\mathbb{C}}^g/\Lambda$ is a complex torus whose tangent space at the origin is identified with $H^1(X,{\mathcal{O}}_X)$; Serre duality identifies this tangent space with $H^0(X,K_X{)}^{\ast }$, giving the dual-of-Hodge identification of the Jacobian.

• Abel-Jacobi map 06.06.04. The Abel-Jacobi map $A:{\mathrm{Div}}^0(X)\to J(X)$ uses the residue pairing for its definition; the kernel of $A$ is the principal-divisor subgroup, characterised cohomologically through Serre duality.

• Theta function 06.06.05. The zero locus of the Riemann theta function on $J(X)$ is the theta divisor $\Theta$, identified with ${\mathrm{Sym}}^{g-1}(X)$ via Abel-Jacobi shifted by the Riemann constant; Riemann's vanishing theorem, characterising this divisor, is a Serre-duality computation on line bundles of degree $g-1$.

• Jacobi inversion theorem 06.06.06 (pending). The proof that ${\mathrm{Sym}}^g(X)\to J(X)$ is surjective and birational uses Serre duality on line bundles of degree $g$ to count sections, identifying the generic fibre of the Abel-Jacobi map with a single point.

• Sheaf cohomology 04.03.01. Serre duality is a theorem in coherent sheaf cohomology; the Čech construction of $H^1$ and the cup-product pairing live entirely in this framework.

• Canonical sheaf and Riemann-Roch theorem for curves 04.04.01. The algebraic version of the curve case, equivalent by GAGA. The canonical sheaf ${\omega }_X$ on a smooth projective curve plays the same role on both sides of the equivalence.

Historical & philosophical context [Master]

Jean-Pierre Serre proved the duality theorem in 1955 in Un théorème de dualité ^{[Serre 1955]} (Comment. Math. Helv. 29, 9-26), generalising classical residue-pairing identities for Riemann surfaces (Riemann 1857; Roch 1865; Klein-Poincaré school) to a coherent-sheaf statement valid for every smooth projective variety over an algebraically closed field. The paper appeared the same year as Serre's Faisceaux algébriques cohérents (Annals of Mathematics 61) ^{[Serre 1955]}, the foundational text on coherent sheaves; the duality paper used the framework of FAC to lift the Riemann-surface residue calculus into a clean cohomological identity on every smooth projective variety.

The classical curve case had been understood implicitly since Riemann's 1857 Theorie der Abelschen Functionen and Roch's 1865 supplement: the index of speciality $i(D)$ of a divisor $D$ on a Riemann surface, defined as the dimension of holomorphic 1-forms vanishing along $D$, is the cohomological obstruction to $\mathcal{O}(D)$ having the expected number of sections. The identification of $i(D)$ with $\dim H^0(K_X-D)=\dim H^1(\mathcal{O}(D))$ as a perfect pairing was implicit in Riemann-Roch but not formalised; Serre's 1955 paper gave it the modern cohomological form.

Alexander Grothendieck and Robin Hartshorne extended Serre duality to a relative theorem for proper morphisms in the 1966 Residues and Duality (Springer LNM 20) ^{[Grothendieck-Hartshorne 1966]}, introducing the dualising complex ${\omega }_X^{\bullet }$ in the derived category and the right-derived functor $f^{!}$. The framework subsumes Serre's smooth-projective case, Verdier duality on locally compact spaces, and the $\ell$-adic duality on étale schemes; it is the prototype of every "six-functor formalism" since.

Hartshorne's Algebraic Geometry (1977) §III.7 and §IV.1 ^{[Hartshorne IV.1]} presents the Serre-duality theorem on curves as a direct application of the Yoneda construction of Ext, giving a uniform proof for all smooth projective varieties via the Koszul complex. Donaldson's Riemann Surfaces (Oxford GTM 22, 2011) §11 ^{[Donaldson Riemann Surfaces]} gives the complex-analytic proof via Hodge decomposition and integrates it with the Riemann-Roch proof in the same chapter; this is the closest match for the present unit's proof structure.

Bibliography [Master]

[object Promise]