04.05.09 · algebraic-geometry / divisors

Hodge index theorem

shipped3 tiersLean: none

Anchor (Master): Hartshorne §V.1.9; Beauville *Complex Algebraic Surfaces* §IV.2 + §VIII (signature applications); Griffiths-Harris §0.7 + §1.2 (Hodge-Riemann bilinear relations and the Lefschetz decomposition); Voisin *Hodge Theory and Complex Algebraic Geometry* I §6 (Hodge index as a corollary of the Hodge-Riemann relations)

Intuition [Beginner]

The Hodge index theorem is a single statement about the shape of the intersection pairing on a smooth projective surface. The intersection pairing assigns an integer to every pair of divisor classes on the surface, and the theorem says this pairing has exactly one "positive direction" and many "negative directions". If you pick any divisor class with strictly positive self-intersection — for example, an ample class such as a hyperplane section — then every divisor class perpendicular to it under the pairing has self-intersection less than or equal to zero. There is one positive axis, and the rest of the space is negative.

Why bother? Because this signature pattern controls a lot of surface geometry. On a curve there is only one intersection number, the degree, and signs are tracked by a single dimension. On a surface the Picard group has finite rank $ρ$ — the Picard number — and the intersection pairing is a symmetric bilinear form of signature $(1, ρ - 1)$ . That one positive direction is the "size" of the surface in the projective embedding, and the negative directions encode every other constraint on how curves can sit on the surface. The theorem reduces a question about all pairs of divisors to a question about a single distinguished positive axis.

The same signature pattern reappears under different names across mathematics: Minkowski space in special relativity has signature $(1, 3)$ , and the algebraic-geometry signature $(1, ρ - 1)$ is a Picard-group analogue. The Hodge index theorem is what makes "Lorentzian" intuition transfer to the lattice of curves on a surface — one timelike direction, many spacelike directions, and a Cauchy-Schwarz inequality controlling how they interact.

Visual [Beginner]

A schematic of a smooth projective surface with two divisor classes drawn on it: an ample class $H$ singled out as the positive axis, and an arbitrary divisor class $D$ decomposed into a piece parallel to $H$ and a piece perpendicular to $H$ . The perpendicular piece carries non-positive self-intersection by the theorem, and the parallel piece carries the entire positive contribution. A second panel shows the resulting signature pattern as a "light cone" in the real Néron-Severi space: one positive direction (the ample axis) and many negative directions (the orthogonal hyperplane).

$A schematic of the Néron-Severi space of a smooth projective surface as a Lorentzian-style space, with the ample divisor class as the unique positive direction and the orthogonal hyperplane as the negative subspace, illustrating the signature $(1, \rho - 1)$ of the intersection pairing.$

The picture captures the essential shape: the real Néron-Severi space of a smooth projective surface is a real vector space of dimension $ρ$ equipped with a symmetric bilinear form whose signature is $(1, ρ - 1)$ . The positive axis is anchored by any ample class. The rest of the space is negative semidefinite once you mod out by the positive axis.

Worked example [Beginner]

Verify the signature pattern $(1, ρ - 1)$ on three surfaces — $P^{2}$ , the quadric $P^{1} \times P^{1}$ , and a blow-up of $P^{2}$ at a point — by writing the intersection form as a matrix and reading off the signature.

Step 1. The projective plane $P^{2}$ . The Picard group is generated by the hyperplane class $L$ with $L^{2} = 1$ , so the Picard number is $ρ = 1$ . The real intersection space $NS (P^{2})_{R}$ is one-dimensional, and the intersection form is the $1 \times 1$ matrix $(1)$ , which has signature $(1, 0)$ . The Hodge index pattern $(1, ρ - 1) = (1, 0)$ matches.

Step 2. The quadric surface $P^{1} \times P^{1}$ . The Picard group is generated by the two ruling classes $h$ and $f$ with $h^{2} = f^{2} = 0$ and $h \cdot f = 1$ , so $ρ = 2$ . The intersection matrix in the basis $(h, f)$ is $$ M = \begin{pmatrix} 0 & 1 \ 1 & 0 \end{pmatrix}. $$ The eigenvalues of $M$ are $+ 1$ and $- 1$ , giving signature $(1, 1)$ . The Hodge index pattern $(1, ρ - 1) = (1, 1)$ matches. The ample class $h + f$ (which is the diagonal hyperplane class on the quadric) has self-intersection $(h + f)^{2} = 2 h \cdot f = 2 > 0$ , the unique positive direction.

Step 3. The blow-up $P^{2} = Bl_{p} P^{2}$ of the projective plane at a point. The Picard group is generated by the pullback hyperplane class $π^{*} L$ and the exceptional divisor $E$ , with $(π^{*} L)^{2} = 1$ , $E^{2} = - 1$ , and $π^{*} L \cdot E = 0$ . So $ρ = 2$ and the intersection matrix in the basis $(π^{*} L, E)$ is

M = (10 0 - 1) .

This is diagonal with eigenvalues $+ 1$ and $- 1$ , giving signature $(1, 1)$ , and the Hodge index pattern $(1, ρ - 1) = (1, 1)$ matches. The ample class $π^{*} L - ϵ E$ for small $ϵ > 0$ has positive self-intersection $1 - ϵ^{2} > 0$ , picking out the positive direction.

Step 4. Read off the consequence. On each of the three surfaces, the intersection pairing has exactly one positive direction once diagonalised over the reals. Any divisor class $D$ perpendicular to the ample class therefore has $D^{2} \leq 0$ . On $P^{2}$ this is vacuous (every class is a multiple of $L$ , the perpendicular subspace is zero). On $P^{1} \times P^{1}$ the orthogonal complement of $h + f$ is the line spanned by $h - f$ , with $(h - f)^{2} = - 2 h \cdot f = - 2$ , the predicted negative. On the blow-up, the orthogonal complement of $π^{*} L$ is the line spanned by $E$ , with $E^{2} = - 1$ , again the predicted negative.

What this tells us: the Hodge index pattern $(1, ρ - 1)$ is read off the intersection matrix as the difference between the number of positive eigenvalues and the number of negative eigenvalues. On every smooth projective surface, the unique positive eigenvalue is anchored by any ample class. The remaining negative eigenvalues encode the "compactness" of the Néron-Severi lattice: every divisor class perpendicular to the ample direction satisfies $D^{2} \leq 0$ , with strict inequality when $D$ is non-zero in the Néron-Severi group.

Check your understanding [Beginner]

Exercise (easy, multiple choice).

The Hodge index theorem for a smooth projective surface $X$ states that the intersection pairing on the real Néron-Severi space $NS (X)_{R}$ has which signature?

A. $(0, ρ)$ — entirely negative
B. $(ρ, 0)$ — entirely positive
C. $(1, ρ - 1)$ — one positive direction, the rest negative
D. $(ρ /2, ρ /2)$ — equal numbers of positive and negative

Hint

The theorem singles out one preferred positive direction (anchored by an ample class) and asserts that the orthogonal complement is negative semidefinite.

Answer

C. The Hodge index theorem says the intersection pairing on the real Néron-Severi space $NS (X)_{R}$ has signature $(1, ρ - 1)$ , where $ρ = ρ (X)$ is the Picard number. The unique positive eigenvalue is anchored by any ample class, and every divisor class orthogonal to the ample direction satisfies $D^{2} \leq 0$ , with strict inequality when $D$ is non-zero modulo numerical equivalence.

Exercise (easy, true-false).

True or false: on a smooth projective surface, any two ample divisor classes have strictly positive intersection.

Hint

Both ample classes lie in the positive cone defined by the Hodge index theorem. The Hodge-index inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ applied to two ample classes gives a positivity statement on their pairing.

Answer

True. If $H_{1}$ and $H_{2}$ are ample divisor classes, both have strictly positive self-intersection, $H_{i}^{2} > 0$ . The Hodge index inequality (the Cauchy-Schwarz form of the theorem) reads $(H_{1} \cdot H_{2})^{2} \geq H_{1}^{2} \cdot H_{2}^{2} > 0$ , hence $H_{1} \cdot H_{2} \neq = 0$ . Both $H_{1}$ and $H_{2}$ lie in the same connected component of the positive cone (the ample cone, which is connected by convexity), so $H_{1} \cdot H_{2} > 0$ rather than $< 0$ . Geometrically, the intersection number of two ample classes counts a transverse intersection of two very ample divisors after passing to high powers, which is a non-negative integer; positivity rules out zero.

Formal definition [Intermediate+]

Let $k$ be an algebraically closed field, let $X$ be a smooth projective surface over $k$ , and let $NS (X) = Pic (X) / Pic^{0} (X)$ be the Néron-Severi group of $X$ — the quotient of the Picard group by the connected component of the identity, equivalently the group of divisor classes modulo algebraic equivalence. The Néron-Severi theorem of Severi (proved in full generality by André Néron) states that $NS (X)$ is a finitely generated abelian group; its rank $ρ (X) = ρ$ is the Picard number of $X$ . Write $NS (X)_{R} = NS (X) \otimes_{Z} R$ , a real vector space of dimension $ρ$ , and write $\cdot$ for the symmetric bilinear intersection pairing $NS (X) \times NS (X) \to Z$ (see 04.05.06), extended by linearity to a symmetric bilinear form on $NS (X)_{R}$ .

Definition (Hodge index theorem, signature form). The Hodge index theorem for the smooth projective surface $X$ asserts that the symmetric bilinear intersection pairing on $NS (X)_{R}$ has signature $(1, ρ - 1)$ , where $ρ$ is the Picard number. Equivalently, after diagonalisation over $R$ , the intersection form has exactly one positive eigenvalue and $ρ - 1$ negative eigenvalues.

Definition (Hodge index inequality form). Let $H \in NS (X)_{R}$ be a divisor class with $H^{2} > 0$ — for instance, an ample class. The Hodge index theorem in inequality form asserts that for every $D \in NS (X)_{R}$ , $$ (D \cdot H)^2 \geq D^2 \cdot H^2, $$ with equality if and only if $D$ is a real multiple of $H$ in $NS (X)_{R}$ .

Definition (Hodge index theorem, orthogonal-complement form). Let $H \in NS (X)_{R}$ be a class with $H^{2} > 0$ . Let $H^{⊥} = {D \in NS (X)_{R} : D \cdot H = 0}$ be its orthogonal complement under the intersection pairing. The Hodge index theorem in orthogonal-complement form asserts that the restriction of the intersection pairing to $H^{⊥}$ is negative definite. Equivalently, for every non-zero $D \in H^{⊥}$ , $$ D^2 < 0. $$

Definition (Néron-Severi group versus numerical equivalence). Two divisor classes $D, D^{'} \in Pic (X)$ are numerically equivalent, written $D \equiv D^{'}$ , if $D \cdot C = D^{'} \cdot C$ for every curve $C \subset X$ . The quotient $Num (X) = Pic (X) / \equiv$ is a finitely generated free abelian group; its rank equals the Picard number $ρ$ . For most surfaces $Num (X) = NS (X) / (torsion)$ , and the intersection pairing factors through $Num (X)$ to give a non-degenerate pairing $Num (X) \times Num (X) \to Z$ . The Hodge index theorem is most cleanly stated on $Num (X)_{R}$ , where the pairing is non-degenerate by construction.

Counterexamples to common slips

The signature is $(1, ρ - 1)$ on the real Néron-Severi space, not on the lattice over $Z$ . Over $Z$ the intersection lattice can be non-degenerate, even unimodular, but the relevant notion of signature is real-analytic and comes from diagonalising over $R$ .
The inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ requires $H^{2} > 0$ . For a class $H$ with $H^{2} = 0$ (an isotropic class — a ruling on $P^{1} \times P^{1}$ , or a fibre class on a ruled surface), the inequality has the opposite sense or fails to be informative, and the orthogonal-complement form must be replaced by a careful analysis of the radical of the pairing.
The orthogonal-complement form requires the ambient class $H$ to have strictly positive self-intersection. An ample class has this property; a nef class with $H^{2} = 0$ does not. Replacing "ample" with "nef" without checking $H^{2} > 0$ produces a false statement, and the standard proof breaks down at the strictly-positive step.
The Picard number $ρ$ is bounded above by the topological invariant $h^{1, 1} (X)$ for a smooth complex projective surface — the Lefschetz $(1, 1)$ -theorem identifies $NS (X)$ with the integral classes in $H^{1, 1} (X, Z)$ . The Hodge index theorem in the cohomological version asserts the cup-product pairing on $H^{1, 1} (X, R)$ has signature $(1, h^{1, 1} - 1)$ , and the surface-Néron-Severi version is recovered by restricting to algebraic classes.

Key theorem with proof [Intermediate+]

Theorem (Hodge index theorem; Hartshorne V Theorem 1.9). Let $X$ be a smooth projective surface over an algebraically closed field $k$ . The symmetric bilinear intersection pairing on $NS (X)_{R}$ has signature $(1, ρ - 1)$ , where $ρ = rank NS (X)$ is the Picard number. Equivalently, if $H$ is any divisor class with $H^{2} > 0$ , then for every $D \in NS (X)_{R}$ , $$ (D \cdot H)^2 \geq D^2 \cdot H^2, $$ with equality if and only if $D$ is a real multiple of $H$ .

Proof. The argument has four steps. First, reduce the signature statement to the inequality statement. Second, decompose an arbitrary divisor class along the $H$ -axis. Third, use Riemann-Roch and Serre vanishing to show the orthogonal complement is negative definite. Fourth, derive the inequality form from the orthogonal decomposition.

Step 1: equivalence of the formulations. Suppose the signature of the intersection pairing on $NS (X)_{R}$ is $(1, ρ - 1)$ . Pick a class $H$ with $H^{2} > 0$ ; it spans the positive eigenspace. Decompose an arbitrary $D \in NS (X)_{R}$ as $D = α H + D^{'}$ with $D^{'} \in H^{⊥}$ . Then $D \cdot H = α H^{2}$ , $D^{2} = α^{2} H^{2} + (D^{'})^{2}$ . Compute $$ (D \cdot H)^2 - D^2 \cdot H^2 = \alpha^2 (H^2)^2 - \big(\alpha^2 H^2 + (D')^2\big) H^2 = - (D')^2 \cdot H^2. $$ Since $(D^{'})^{2} \leq 0$ on $H^{⊥}$ (negative semidefiniteness) and $H^{2} > 0$ , this difference is $\geq 0$ , giving $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ . Equality holds when $(D^{'})^{2} = 0$ on $H^{⊥}$ , which by negative definiteness of $H^{⊥}$ forces $D^{'} = 0$ , that is, $D = α H$ a real multiple of $H$ . Conversely, the inequality with equality characterising real multiples of $H$ forces the orthogonal complement to be negative definite, and hence the signature to be $(1, ρ - 1)$ .

Step 2: existence of a positive class. A smooth projective surface admits an ample divisor class — pull back $O (1)$ along any closed embedding $X ↪ P^{N}$ and read off a very ample $H \in NS (X)$ . Such an $H$ has $H^{2} > 0$ because the self-intersection of a very ample divisor is the degree of $X$ in the projective embedding, a positive integer. So at least one direction in $NS (X)_{R}$ is positive, and the signature has at least one $+ 1$ eigenvalue.

Step 3: orthogonal complement is negative semidefinite. Fix $H$ ample with $H^{2} > 0$ . Take a class $D \in H^{⊥}$ , that is, $D \cdot H = 0$ . The argument shows $D^{2} \leq 0$ . The key input is the surface Riemann-Roch identity (see 04.05.08), $$ \chi(\mathcal{O}_X(nD)) = \chi(\mathcal{O}_X) + \tfrac{1}{2},nD \cdot (nD - K_X) = \chi(\mathcal{O}_X) + \tfrac{n^2}{2} D^2 - \tfrac{n}{2} D \cdot K_X, $$ valid for every integer $n$ .

Suppose for contradiction that $D^{2} > 0$ . The leading-order asymptotic in $n$ of the right side is $\frac{n ^{2}}{2} D^{2} \to + \infty$ as $n \to \infty$ , hence $χ (O_{X} (n D)) \to + \infty$ . By the bound $χ (O_{X} (n D)) \leq h^{0} (n D) + h^{2} (n D)$ and Serre duality $h^{2} (n D) = h^{0} (K_{X} - n D)$ , at least one of $h^{0} (n D)$ or $h^{0} (K_{X} - n D)$ is unbounded as $n \to + \infty$ .

Since $H$ is ample and $D \cdot H = 0$ , the divisor $n D$ has $n D \cdot H = 0$ as well. An effective $E$ on $X$ pairs strictly positively with every ample class, because the intersection number of an ample class with any irreducible curve is positive (the Nakai-Moishezon criterion for ampleness) and a non-zero effective divisor is a non-negative integer combination of irreducible curves with at least one strictly positive coefficient. Hence $E \cdot H > 0$ for every non-zero effective $E$ . So if $h^{0} (n D) > 0$ for some $n > 0$ , the effective representative $E \in ∣ n D ∣$ would have $E \cdot H = n D \cdot H = 0$ , contradicting $E \cdot H > 0$ . Therefore $h^{0} (n D) = 0$ for all $n > 0$ .

The same argument applied to $K_{X} - n D$ for large $n$ shows $h^{0} (K_{X} - n D) = 0$ eventually, since $(K_{X} - n D) \cdot H = K_{X} \cdot H$ is a fixed integer independent of $n$ , while $K_{X} - n D$ becomes anti-ample at large $n$ along the $H$ -pairing and cannot represent a non-zero effective class once $n$ exceeds $K_{X} \cdot H / (H \cdot H^{'})$ for any positive $H^{'}$ paired with it. Combining the two vanishings, $χ (O_{X} (n D)) = - h^{1} (n D)$ for large $n$ , forcing $h^{1} (n D)$ to grow asymptotically like $\frac{n ^{2}}{2} D^{2}$ , which contradicts the polynomial bound $h^{1} (n D) = O (n)$ from Serre vanishing applied to a twist by an ample class. The conclusion: $D^{2} \leq 0$ for every $D \in H^{⊥}$ .

Step 4: negative definiteness on the orthogonal complement. Suppose $D \in H^{⊥}$ with $D^{2} = 0$ . The argument shows $D = 0$ in $NS (X)_{R}$ , that is, $D$ is numerically zero. By the Hodge index inequality applied at $D$ and $H$ , $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ , that is, $0 \geq 0$ , holds with equality; by the equivalence in Step 1, equality forces $D$ to be a real multiple of $H$ . Since $D \cdot H = 0$ and $H^{2} > 0$ , the multiple must be zero, so $D = 0$ in $NS (X)_{R}$ . Hence the intersection pairing on $H^{⊥}$ is negative definite, and the signature of the pairing on $NS (X)_{R}$ is $(1, ρ - 1)$ .

Combining Steps 1–4 produces the signature form and the inequality form simultaneously. $□$

Bridge. The Hodge index theorem builds toward the lattice-theoretic study of algebraic surfaces, where the central insight is that the Néron-Severi group of every smooth projective surface is a Lorentzian lattice of signature $(1, ρ - 1)$ — one timelike direction, $ρ - 1$ spacelike directions — and this single structural fact organises the geometry of curves on the surface, the structure of the ample cone, and the cohomological invariants that classify the surface. This bridge appears again in 04.09.01 (Hodge decomposition), where the cup-product pairing on $H^{1, 1} (X, R)$ for a complex projective surface inherits signature $(1, h^{1, 1} - 1)$ from the Hodge-Riemann bilinear relations, and the algebraic-geometric Hodge index theorem on $NS (X)$ is the integral-lattice shadow of the analytic Hodge-Riemann statement on primitive cohomology.

The foundational reason is that intersection numbers on a surface compute cup products in middle-degree cohomology, and signature is a cup-product invariant. Putting these together, the §V.1 framework of Hartshorne — intersection pairing, adjunction, Riemann-Roch on surfaces, Hodge index — assembles the four pillars of surface theory, and the bridge identifies the Hodge index signature with the Hodge-Riemann signature on $H^{1, 1}$ , which in turn is dual to the Lefschetz decomposition on primitive cohomology. The Hodge index theorem is what makes the inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ the algebraic Cauchy-Schwarz, and identifies the ample cone with one of the two connected components of the positive cone ${D \in NS (X)_{R} : D^{2} > 0}$ , a fact that generalises to the cone theorem in higher-dimensional birational geometry.

Exercises [Intermediate+]

Exercise 1 (easy, numeric).

On $P^{1} \times P^{1}$ with rulings $h, f$ ( $h^{2} = f^{2} = 0$ , $h \cdot f = 1$ ), compute the self-intersection of the class $2 h - 3 f$ , and verify the Hodge index inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ at $D = 2 h - 3 f$ and $H = h + f$ .

Hint

Expand $(2 h - 3 f)^{2}$ and $(2 h - 3 f) \cdot (h + f)$ using $h^{2} = f^{2} = 0$ and $h \cdot f = 1$ . The class $h + f$ is ample on $P^{1} \times P^{1}$ with $(h + f)^{2} = 2$ .

Answer

$D^{2} = - 12$ , $D \cdot H = - 1$ , inequality reads $1 \geq - 24$ , satisfied. $D^{2} = (2 h - 3 f)^{2} = 4 h^{2} - 12 h \cdot f + 9 f^{2} = 0 - 12 + 0 = - 12$ . $D \cdot H = (2 h - 3 f) \cdot (h + f) = 2 h^{2} + 2 h \cdot f - 3 f \cdot h - 3 f^{2} = 0 + 2 - 3 - 0 = - 1$ . $(D \cdot H)^{2} = 1$ . $D^{2} \cdot H^{2} = - 12 \cdot 2 = - 24$ . The inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ reads $1 \geq - 24$ , which holds with substantial slack. The Hodge index theorem predicts the inequality and the negative $D^{2}$ — the class $D$ is not a real multiple of $H$ , so strict inequality is expected. Rubric: full credit for $D^{2} = - 12$ and the verification of the inequality.

Exercise 3 (medium, symbolic).

On a smooth projective surface $X$ with ample divisor $H$ , prove that for any two divisor classes $D_{1}, D_{2} \in NS (X)_{R}$ with $D_{i} \cdot H = 0$ for $i = 1, 2$ , the Cauchy-Schwarz inequality $(D_{1} \cdot D_{2})^{2} \leq D_{1}^{2} \cdot D_{2}^{2}$ holds on the negative-definite subspace $H^{⊥}$ , with the sign of the inequality reversed from the Lorentzian Hodge index form.

Hint

On the negative-definite subspace $H^{⊥}$ , the intersection form $- (\cdot)$ is positive definite, so the standard Cauchy-Schwarz inequality applies to $- (\cdot)$ .

Answer

The intersection pairing restricted to $H^{⊥}$ is negative definite by the Hodge index theorem (Step 4 of the proof). The bilinear form $⟨ D_{1}, D_{2} ⟩ = - D_{1} \cdot D_{2}$ on $H^{⊥}$ is therefore positive definite. The standard Cauchy-Schwarz inequality for a positive definite form reads $$ \langle D_1, D_2 \rangle^2 \leq \langle D_1, D_1 \rangle \cdot \langle D_2, D_2 \rangle. $$ Substituting $⟨ D_{i}, D_{j} ⟩ = - D_{i} \cdot D_{j}$ , $$ (-D_1 \cdot D_2)^2 \leq (-D_1^2) \cdot (-D_2^2), $$ that is, $(D_{1} \cdot D_{2})^{2} \leq D_{1}^{2} \cdot D_{2}^{2}$ . Equality holds if and only if $D_{1}$ and $D_{2}$ are linearly dependent in $H^{⊥}$ . Rubric: full credit for the sign flip and the recognition that the inequality goes the opposite way to the Lorentzian Hodge index inequality on the full space.

Exercise 5 (medium, short-answer).

Let $X$ be a smooth projective surface and let $H, H^{'}$ be two ample classes. Show that $H \cdot H^{'} > 0$ .

Hint

Apply the Hodge index inequality $(H^{'} \cdot H)^{2} \geq (H^{'})^{2} \cdot H^{2}$ together with the connectedness of the ample cone.

Answer

Both $H$ and $H^{'}$ satisfy $H^{2} > 0$ and $(H^{'})^{2} > 0$ (ample divisors have strictly positive self-intersection by the Nakai-Moishezon criterion or by self-intersection of a hyperplane section). The Hodge index inequality reads $(H \cdot H^{'})^{2} \geq H^{2} \cdot (H^{'})^{2} > 0$ , hence $H \cdot H^{'} \neq = 0$ . The ample cone ${cH + c^{'} H^{'} : c, c^{'} > 0}$ is path-connected (any two ample classes are joined by a straight-line path of ample classes by convexity of the ample cone, which follows from the convex-combination preservation of positivity under the Nakai-Moishezon criterion). The intersection number $H \cdot H^{'}$ is continuous in $H, H^{'}$ along this path and never zero, so it has constant sign. Evaluating at the endpoint $H^{'} = H$ gives $H \cdot H = H^{2} > 0$ , so $H \cdot H^{'} > 0$ for every pair of ample classes. Rubric: full credit for the Hodge index inequality, the non-vanishing of $H \cdot H^{'}$ , and the connectedness argument fixing the sign.

Exercise 6 (medium, numeric).

Let $X$ be a K3 surface containing two smooth rational curves $C_{1}$ and $C_{2}$ — that is, two smooth rational curves with $C_{i}^{2} = - 2$ by the adjunction-derived K3 identity (see 04.05.07). Use the Hodge index inequality to bound $∣ C_{1} \cdot C_{2} ∣$ in terms of an ample divisor $H$ on $X$ with $H^{2} = d > 0$ .

Hint

Apply the Hodge index inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ to the class $D = a C_{1} + b C_{2}$ for various coefficients $a, b$ , and use $C_{i}^{2} = - 2$ together with the constraint $D \cdot H = a (C_{1} \cdot H) + b (C_{2} \cdot H)$ .

Answer

Set $a_{i} = C_{i} \cdot H \geq 0$ (intersection of an ample class with an irreducible curve is non-negative). Consider $D = a_{2} C_{1} - a_{1} C_{2}$ , which satisfies $D \cdot H = a_{2} (C_{1} \cdot H) - a_{1} (C_{2} \cdot H) = a_{2} a_{1} - a_{1} a_{2} = 0$ , hence $D \in H^{⊥}$ . By the Hodge index theorem, $D^{2} \leq 0$ , with equality iff $D = 0$ in $NS (X)_{R}$ .

Expand: $D^{2} = a_{2}^{2} C_{1}^{2} - 2 a_{1} a_{2} (C_{1} \cdot C_{2}) + a_{1}^{2} C_{2}^{2} = - 2 a_{2}^{2} - 2 a_{1} a_{2} (C_{1} \cdot C_{2}) - 2 a_{1}^{2}$ . The constraint $D^{2} \leq 0$ becomes $$ -2 a_2^2 - 2 a_1 a_2 (C_1 \cdot C_2) - 2 a_1^2 \leq 0, $$ that is, $a_{1} a_{2} (C_{1} \cdot C_{2}) \geq - (a_{1}^{2} + a_{2}^{2})$ . By AM-GM, $a_{1}^{2} + a_{2}^{2} \geq 2 a_{1} a_{2}$ , so $a_{1} a_{2} (C_{1} \cdot C_{2}) \geq - 2 a_{1} a_{2}$ , and (assuming $a_{1}, a_{2} > 0$ ) $C_{1} \cdot C_{2} \geq - 2$ .

A complementary upper bound comes from the rank- $2$ sublattice $Λ \subset NS (X)$ spanned by $C_{1}$ and $C_{2}$ , whose Gram matrix has determinant $C_{1}^{2} \cdot C_{2}^{2} - (C_{1} \cdot C_{2})^{2} = 4 - (C_{1} \cdot C_{2})^{2}$ . The Hodge index pattern forces $Λ$ to have signature compatible with $(1, ρ - 1)$ on the ambient lattice. The sharp form: $Λ$ is negative definite iff its Gram determinant is positive and its diagonal entries are negative, equivalent to $∣ C_{1} \cdot C_{2} ∣ < 2$ ; $Λ$ is degenerate at $∣ C_{1} \cdot C_{2} ∣ = 2$ ; and $Λ$ becomes indefinite for $∣ C_{1} \cdot C_{2} ∣ > 2$ . Rubric: full credit for the orthogonal-class construction $D = a_{2} C_{1} - a_{1} C_{2}$ and the lower bound $C_{1} \cdot C_{2} \geq - 2$ from $D^{2} \leq 0$ .

Exercise 7 (hard, short-answer).

Prove the orthogonal-complement form of the Hodge index theorem from the inequality form: given the inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ for all $D$ and $H^{2} > 0$ , show that the restriction of the intersection pairing to $H^{⊥}$ is negative definite.

Hint

Take $D \in H^{⊥}$ , so $D \cdot H = 0$ . The inequality becomes $0 \geq D^{2} \cdot H^{2}$ , which with $H^{2} > 0$ forces $D^{2} \leq 0$ . For the strictness, apply the inequality to $α H + D$ and use the equality characterisation.

Answer

For $D \in H^{⊥}$ , $D \cdot H = 0$ by definition, so the inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ becomes $0 \geq D^{2} \cdot H^{2}$ . Since $H^{2} > 0$ , this forces $D^{2} \leq 0$ , giving negative semidefiniteness on $H^{⊥}$ .

For strict negative definiteness, suppose $D \in H^{⊥}$ has $D^{2} = 0$ . The argument shows $D = 0$ . Consider the class $α H + D$ for arbitrary real $α$ . By the inequality, $((α H + D) \cdot H)^{2} \geq (α H + D)^{2} \cdot H^{2}$ . Compute $(α H + D) \cdot H = α H^{2}$ and $(α H + D)^{2} = α^{2} H^{2} + 2 α D \cdot H + D^{2} = α^{2} H^{2} + 0 + 0 = α^{2} H^{2}$ . The inequality becomes $α^{2} (H^{2})^{2} \geq α^{2} H^{2} \cdot H^{2}$ , that is, $α^{2} (H^{2})^{2} \geq α^{2} (H^{2})^{2}$ , which holds with equality. By the equality characterisation of the inequality, $α H + D$ is a real multiple of $H$ for every $α$ , in particular for $α = 0$ : $D$ is a real multiple of $H$ . Since $D \cdot H = 0$ and $H^{2} > 0$ , the multiple is zero, so $D = 0$ in $NS (X)_{R}$ . Rubric: full credit for the semidefinite argument and the definiteness step using the equality case.

Exercise 8 (hard, short-answer).

Use the Hodge index theorem to prove that on a smooth projective surface $X$ , the ample cone is one of the two connected components of the open positive cone ${D \in NS (X)_{R} : D^{2} > 0}$ .

Hint

The positive cone ${D : D^{2} > 0}$ in a real vector space with a symmetric bilinear form of signature $(1, ρ - 1)$ has exactly two connected components, opposite under the negation $D \mapsto - D$ . The ample cone is convex and contained in one such component.

Answer

By the Hodge index theorem, the intersection pairing on $NS (X)_{R}$ has signature $(1, ρ - 1)$ . In a real vector space of dimension $ρ$ with a symmetric bilinear form of signature $(1, ρ - 1)$ , the open positive cone ${D : D^{2} > 0}$ has exactly two connected components: the upper component ${D : D^{2} > 0, D \cdot H > 0}$ and the lower component ${D : D^{2} > 0, D \cdot H < 0}$ , where $H$ is any class with $H^{2} > 0$ . The two components are opposite under $D \mapsto - D$ .

The ample cone $Amp (X) \subset NS (X)_{R}$ is open and convex, and every ample $H$ satisfies $H^{2} > 0$ (the self-intersection of a hyperplane section in a projective embedding is the positive degree of the embedded surface). So the ample cone is contained in the positive cone. By convexity, the ample cone is path-connected. Two ample classes $H_{1}, H_{2}$ satisfy $H_{1} \cdot H_{2} > 0$ (Exercise 5), so the line segment ${(1 - t) H_{1} + t H_{2} : t \in [0, 1]}$ has positive self-intersection at every $t$ (the quadratic in $t$ is non-negative and positive at the endpoints, hence positive throughout by the discriminant computation). The ample cone therefore lies entirely in one connected component of the positive cone. The closure of the ample cone is the nef cone, which is the closure of one of the two components; the open ample cone is the interior. Rubric: full credit for the signature-based decomposition of the positive cone into two components, the convexity of the ample cone, and the identification of the ample cone with one of the two components.

Lean formalization [Intermediate+]

Mathlib has the real-bilinear-form signature infrastructure and partial divisor / Picard-group machinery on a smooth projective scheme, but no named Hodge index theorem on a smooth projective surface. The intended formalisation reads schematically:

import Mathlib.AlgebraicGeometry.Divisor.Basic
import Mathlib.AlgebraicGeometry.Picard
import Mathlib.LinearAlgebra.QuadraticForm.Real
import Mathlib.LinearAlgebra.BilinearForm.Basic

variable {k : Type*} [Field k] [IsAlgClosed k]
variable (X : Scheme) [IsSmooth X k] [IsProjective X k] (hd : X.dimension = 2)

/-- The Néron-Severi group of a smooth projective surface,
as the Picard group modulo the connected component of the identity. -/
noncomputable def NeronSeveri : Type :=
  Quotient (Pic.modPicZero X)

/-- The intersection pairing extended to the real Néron-Severi space. -/
noncomputable def intersectionBilinearReal :
    LinearMap.BilinForm ℝ (NeronSeveri X ⊗[ℤ] ℝ) :=
  (intersectionPairing X).map (Int.cast : ℤ → ℝ)

/-- Hodge index theorem, signature form. -/
theorem hodge_index_signature :
    (intersectionBilinearReal X).signature = (1, picardNumber X - 1) := by
  -- Step 1: ample class provides one positive direction
  -- Step 2: Riemann-Roch + Serre vanishing yields negative semidefinite orthogonal complement
  -- Step 3: equality case forces numerical triviality
  sorry

/-- Hodge index theorem, inequality form. -/
theorem hodge_index_inequality
    (H D : NeronSeveri X ⊗[ℤ] ℝ) (hH : H ⬝ H > 0) :
    (D ⬝ H) ^ 2 ≥ (D ⬝ D) * (H ⬝ H) := by
  -- equivalent to signature form via orthogonal decomposition
  sorry

The proof gap is substantive. Mathlib needs five pieces wired together: the Néron-Severi group as the quotient of the Picard group by the connected component of the identity (Mathlib has the Picard scheme but the quotient is not packaged as the Néron-Severi group), the symmetric bilinear intersection pairing on $NS (X) \otimes R$ (a corollary of the surface intersection pairing, which itself is a Mathlib gap), the existence of an ample class with strictly positive self-intersection (from Nakai-Moishezon), the surface Riemann-Roch identity (a Mathlib gap), and the Serre-vanishing input for the asymptotic growth of $h^{0} (n D)$ on the orthogonal complement of an ample class. The signature form is the natural first formalisation target, and the inequality form follows by the orthogonal-decomposition argument in Step 1 of the proof.

Advanced results [Master]

Theorem (Hodge index theorem, cohomological version; Griffiths-Harris §1.2 + Voisin Ch. 6). Let $X$ be a smooth complex projective surface. The cup-product pairing $$ H^{1, 1}(X, \mathbb{R}) \times H^{1, 1}(X, \mathbb{R}) \to H^{2, 2}(X, \mathbb{R}) \cong \mathbb{R} $$ has signature $(1, h^{1, 1} (X) - 1)$ , where $h^{1, 1} (X) = dim_{C} H^{1, 1} (X)$ is the Hodge number. The algebraic-geometric Hodge index theorem on $NS (X)_{R}$ is the restriction of the cup-product pairing on $H^{1, 1} (X, R)$ to the integral classes — the image of the Lefschetz $(1, 1)$ map $NS (X) \otimes R ↪ H^{1, 1} (X, R)$ .

The cohomological version follows from the Hodge-Riemann bilinear relations on the primitive cohomology of a Kähler manifold. For a smooth complex projective surface, the second cohomology $H^{2} (X, R)$ decomposes into the Lefschetz primitive part $H_{prim}^{2} (X, R)$ and the multiple-of-the-class-of-the-hyperplane part, and the Hodge-Riemann relations assert that the cup-product pairing is positive definite on the $(0, 2) + (2, 0)$ part of the primitive cohomology and negative definite on the $(1, 1)$ part of the primitive cohomology. Adding back the class of the hyperplane section, which carries one positive eigenvalue, produces the signature $(1, h^{1, 1} - 1)$ on $H^{1, 1} (X, R)$ .

Theorem (Hodge-Riemann bilinear relations, surface case; Hodge 1941). Let $X$ be a smooth complex projective surface with Kähler class $ω$ and let $H_{prim}^{2} (X, R) = ker (ω ⌣ - : H^{2} (X, R) \to H^{4} (X, R))$ be the primitive cohomology. The cup-product pairing satisfies:

On $H^{2, 0} \oplus H^{0, 2}$ (the primitive $(2, 0) + (0, 2)$ part): positive definite.
On $H_{prim}^{1, 1}$ (the primitive $(1, 1)$ part): negative definite. Adding back the Kähler class itself produces one positive eigenvalue, so the full signature on $H^{1, 1} (X, R)$ is $(1, h^{1, 1} - 1)$ .

The Hodge-Riemann relations are the foundational identity of Hodge theory. On a Kähler manifold of complex dimension $n$ the primitive cohomology in degree $k$ carries a definite bilinear form $(α, β) \mapsto (- 1)^{k (k - 1) /2} \cdot i^{p - q} \int_{X} α \land \overline{β} \land ω^{n - k}$ , definite of alternating sign on each Hodge $(p, q)$ -piece by the Hodge-Riemann formula. The surface case ( $n = 2$ , $k = 2$ ) is the original Hodge index theorem; the higher-dimensional generalisation requires the Lefschetz primitive decomposition.

Theorem (uniqueness of the ample class up to positive scaling on Picard number 1). Let $X$ be a smooth projective surface with Picard number $ρ (X) = 1$ . The Néron-Severi group $NS (X)$ is generated by a single class, and the ample cone is the positive ray generated by an ample divisor.

This follows from the Hodge index pattern $(1, ρ - 1) = (1, 0)$ at $ρ = 1$ : the Néron-Severi space is one-dimensional, and the unique positive eigenvalue is anchored by the generator. The ample classes form the positive half-line; the anti-ample classes the negative half-line. Examples: $P^{2}$ , a smooth quadric surface in $P^{3}$ with rank- $1$ Picard (over a general field, the smooth quadric has $ρ = 2$ over $\overset{ˉ}{k}$ but $ρ = 1$ over $k$ if the quadric has no $k$ -rational ruling), and a smooth cubic surface over an algebraically closed field has $ρ = 7$ , so Picard number $1$ is a restrictive condition.

Theorem (Néron-Severi lattice of a K3 surface; Beauville §VIII). Let $X$ be a complex projective K3 surface (a smooth simply connected surface with $ω_{X} ≅ O_{X}$ ). The integral cohomology lattice $H^{2} (X, Z)$ with cup product is the K3 lattice $Λ = E_{8} (- 1)^{\oplus 2} \oplus U^{\oplus 3}$ , an even unimodular lattice of signature $(3, 19)$ and rank $22$ . The Néron-Severi sublattice $NS (X) \subset H^{2} (X, Z)$ is a primitive sublattice; its signature is $(1, ρ - 1)$ by the Hodge index theorem on $X$ . The Picard number $ρ (X)$ ranges from $1$ (for the very general algebraic K3) to $20$ (for the maximal-Picard K3 surfaces classified by Šafarevič and Šafarevič-Pjateckiĭ).

The K3 lattice picture is the foundational example of the lattice-theoretic study of surfaces. The Hodge index theorem provides the signature constraint $(1, ρ - 1)$ on the Néron-Severi sublattice; the Torelli theorem for K3 surfaces (Pjateckiĭ-Šafarevič) inverts this to recover the K3 surface from the Hodge structure on the Néron-Severi lattice, making the lattice-theoretic invariants a complete classification of polarised K3 surfaces.

Theorem (signature criterion for ampleness; Nakai-Moishezon plus Hodge index). Let $X$ be a smooth projective surface and let $D \in NS (X)$ be a divisor class. Then $D$ is ample if and only if (a) $D^{2} > 0$ , (b) $D \cdot C > 0$ for every irreducible curve $C \subset X$ , and (c) $D$ lies in the same connected component of the positive cone as some ample class.

The Nakai-Moishezon criterion is the standard characterisation of ampleness on a surface, and the Hodge index theorem is what makes condition (c) precise. Without the signature theorem, the positive cone could in principle be path-connected, and conditions (a) and (b) would suffice. The Hodge index pattern $(1, ρ - 1)$ implies the positive cone has two components, and one must specify which component the ample cone lies in.

Theorem (Reider's criterion; Reider 1988). Let $X$ be a smooth projective surface and let $L$ be a nef line bundle with $L^{2} \geq 5$ . Then $K_{X} + L$ is base-point-free unless there is an effective divisor $E$ with $L \cdot E = 0$ and $E^{2} = - 1$ , or $L \cdot E = 1$ and $E^{2} = 0$ .

Reider's criterion is the surface analogue of the Bombieri pluricanonical-mapping theorem and uses the Hodge index theorem indirectly: the case analysis of small $L \cdot E$ values and small $E^{2}$ values is constrained by the signature pattern, and the exceptional cases are exactly the classes that lie close to the boundary of the positive cone. The proof technique (Bogomolov-style instability of a rank- $2$ vector bundle) uses the Hodge index inequality on the Chern classes of the destabilising sub-bundle.

Synthesis. The Hodge index theorem is the foundational signature statement on the Néron-Severi lattice of a smooth projective surface, and the central insight is that the algebraic Cauchy-Schwarz inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ for $H^{2} > 0$ is the surface manifestation of a universal signature pattern that organises curve geometry, ample-cone structure, and lattice theory on a single algebraic surface through the identity $sig (NS (X)_{R}, \cdot) = (1, ρ - 1)$ . Three apparently distinct constructions — the surface Riemann-Roch identity combined with Serre vanishing, the Hodge-Riemann bilinear relations on primitive cohomology in the complex-analytic setting, and the Nakai-Moishezon characterisation of the ample cone — fit into one signature statement. Putting these together, the Hodge index theorem is what makes the ample cone one of the two connected components of the positive cone, the foundational reason that any two ample classes pair to a strictly positive integer, the central insight that identifies the Néron-Severi lattice with a Lorentzian lattice of signature $(1, ρ - 1)$ , and the dual statement that the orthogonal complement of any positive-self-intersection class is negative definite. This bridge appears again in 04.09.01 (Hodge decomposition), where the cohomological version of the index theorem follows from the Hodge-Riemann bilinear relations on primitive cohomology, and the algebraic statement on $NS (X)$ is identified with the integral-lattice shadow of the analytic statement on $H^{1, 1} (X, R)$ .

The Hodge index theorem also generalises in two directions. To higher dimensions, the Hodge-Riemann bilinear relations on the primitive cohomology of a Kähler manifold of complex dimension $n$ produce a signature pattern on $H^{p, p} (X, R)$ that, for $X$ a smooth complex projective threefold, is fundamental to the structure theory of the Néron-Severi space of divisor classes and the Mori cone of curve classes. The cone theorem of Mori, the cone of curves dual to the nef cone of divisors, and the structure of the Mori program for threefold birational geometry are all anchored by signature constraints generalising the surface Hodge index theorem. To lattice theory, the integral version — the Hodge index theorem on $NS (X)$ as a Lorentzian lattice — feeds the modern study of automorphism groups of surfaces (the Torelli theorem for K3 surfaces; the Cremona group acting on the Picard lattice of a rational surface; the Conway-Sloane lattice machinery applied to Picard lattices of surfaces of low Picard number). The Hodge index pattern $(1, ρ - 1)$ identifies $NS (X)_{R}$ with a Lorentzian space; the discrete automorphism group of the lattice acts as a discrete subgroup of $O (1, ρ - 1)$ , and the geometry of the Néron-Severi lattice becomes a geometry of hyperbolic space.

The synthesis is structural: every signature-driven classification result for smooth projective surfaces — the structure of the ample cone, the Hodge-Riemann relations on primitive cohomology, the Néron-Severi lattice classification of K3 surfaces, the Reider criterion, the Bogomolov-Miyaoka-Yau inequality — is a corollary of Hodge index together with surface Riemann-Roch and adjunction. Hodge index supplies the lattice signature; Riemann-Roch supplies the Euler-characteristic data; adjunction supplies the canonical-class identities. Together they constitute the foundational analytic input to the Castelnuovo-Beauville-Bombieri-Kodaira classification of surfaces.

Full proof set [Master]

Proposition (positive cone has two components on a surface with $ρ \geq 1$ ). Let $V$ be a real vector space of dimension $ρ \geq 1$ equipped with a symmetric bilinear form of signature $(1, ρ - 1)$ . The open positive cone $P = {v \in V : v \cdot v > 0}$ has exactly two path-connected components, swapped by the negation $v \mapsto - v$ .

Proof. Pick an orthogonal basis $(e_{0}, e_{1}, \dots, e_{ρ - 1})$ for $V$ with $e_{0} \cdot e_{0} = 1$ and $e_{i} \cdot e_{i} = - 1$ for $i \geq 1$ . Write $v = x_{0} e_{0} + \sum_{i \geq 1} x_{i} e_{i}$ . Then $v \cdot v = x_{0}^{2} - \sum_{i \geq 1} x_{i}^{2}$ . The condition $v \cdot v > 0$ becomes $x_{0}^{2} > \sum_{i \geq 1} x_{i}^{2}$ , that is, $∣ x_{0} ∣ > ∥ x_{\geq 1} ∥_{2}$ . This is the standard description of the open positive cone of a Lorentzian inner product: two opposing components, ${x_{0} > ∥ x_{\geq 1} ∥_{2}}$ and ${x_{0} < - ∥ x_{\geq 1} ∥_{2}}$ , each homeomorphic to an open Euclidean half-space. Negation $v \mapsto - v$ swaps them. Path-connectedness within each component is via the straight-line segment, since the open cone is convex within each component (a sum of two future-pointing causal vectors is future-pointing causal). $□$

Proposition (intersection pairing is non-degenerate on $Num (X)_{R}$ ). Let $X$ be a smooth projective surface and let $Num (X)$ be the Picard group modulo numerical equivalence. The intersection pairing on $Num (X)_{R}$ is non-degenerate.

Proof. Let $D \in Num (X)_{R}$ with $D \cdot D^{'} = 0$ for every $D^{'} \in Num (X)_{R}$ . The map $D^{'} \mapsto D \cdot D^{'}$ is identically zero, so in particular $D \cdot C = 0$ for every curve $C \subset X$ (since curve classes generate $Num (X)_{R}$ — every divisor class is a difference of effective classes by the Picard-group decomposition for surfaces). By the definition of numerical equivalence, $D \equiv 0$ , that is, $D = 0$ in $Num (X)$ . Hence the pairing is non-degenerate on $Num (X)_{R}$ . $□$

Theorem (Hodge index theorem on a smooth projective surface), proof. Given in full at the Intermediate tier: pick an ample $H \in NS (X)_{R}$ with $H^{2} > 0$ from the projective embedding. Decompose $NS (X)_{R} = R \cdot H \oplus H^{⊥}$ . Show $(D^{'})^{2} < 0$ for every non-zero $D^{'} \in H^{⊥}$ by combining surface Riemann-Roch, Serre vanishing, and the contradiction with positive ample-pairing on effective divisors. The orthogonal-complement form gives the signature $(1, ρ - 1)$ ; the inequality form follows from the decomposition by direct computation. $□$

Proposition (Hodge index inequality, equivalent algebraic form). Let $X$ be a smooth projective surface and let $H, D \in NS (X)_{R}$ with $H^{2} > 0$ . The inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ holds, with equality if and only if $D$ is a real multiple of $H$ .

Proof. Decompose $D = α H + D^{'}$ with $α = (D \cdot H) / H^{2} \in R$ and $D^{'} = D - α H \in H^{⊥}$ (this is the orthogonal projection of $D$ onto the line $R \cdot H$ ). Compute: $$ D^2 = \alpha^2 H^2 + 2 \alpha (D' \cdot H) + (D')^2 = \alpha^2 H^2 + 0 + (D')^2, $$ since $D^{'} \cdot H = 0$ by construction. Therefore $$ D^2 \cdot H^2 = \alpha^2 (H^2)^2 + (D')^2 \cdot H^2. $$ Also $D \cdot H = α H^{2}$ , so $(D \cdot H)^{2} = α^{2} (H^{2})^{2}$ . Subtracting, $$ (D \cdot H)^2 - D^2 \cdot H^2 = -(D')^2 \cdot H^2. $$ By the orthogonal-complement form of the Hodge index theorem, $(D^{'})^{2} \leq 0$ on $H^{⊥}$ , and $H^{2} > 0$ . Therefore $(D \cdot H)^{2} - D^{2} \cdot H^{2} \geq 0$ , that is, $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ . Equality holds if and only if $(D^{'})^{2} = 0$ , which by negative definiteness on $H^{⊥}$ forces $D^{'} = 0$ , that is, $D = α H$ is a real multiple of $H$ . $□$

Theorem (cohomological Hodge index theorem on a smooth complex projective surface), stated without proof here — full proof in Voisin Hodge Theory and Complex Algebraic Geometry I, §6.3 ^[pending]. The cup-product pairing on $H^{1, 1} (X, R)$ for a smooth complex projective surface has signature $(1, h^{1, 1} (X) - 1)$ , and the algebraic-geometric Hodge index theorem on $NS (X)_{R}$ is the restriction of this cup-product pairing to the integral classes. The proof uses the Hodge-Riemann bilinear relations on primitive cohomology in the surface case ( $n = 2$ , $k = 2$ ), the Lefschetz $(1, 1)$ -theorem identifying $NS (X) \otimes R$ with the integral classes in $H^{1, 1} (X, R)$ , and a compatibility of the cup product on $H^{2} (X, R)$ with the algebraic intersection pairing on $NS (X)$ via the cycle-class map.

Theorem (Hodge-Riemann bilinear relations, surface case), stated without proof here — full proof in Griffiths-Harris Principles of Algebraic Geometry §0.7 + §1.2 ^[pending]. On a smooth complex projective surface $X$ with Kähler class $ω$ , the cup-product pairing is positive definite on the primitive $(2, 0) + (0, 2)$ part of $H^{2} (X, R)$ and negative definite on the primitive $(1, 1)$ part. The proof is by the Hodge-Riemann formula for the primitive intersection pairing on a Kähler manifold, with the surface case appearing as the simplest non-degenerate dimension ( $n = 2$ , $k = 2$ , primitive cohomology is the orthogonal complement of the Kähler class). The Hodge index theorem on $NS (X)$ is the integral shadow.

Connections [Master]

Intersection pairing on a surface 04.05.06. The Hodge index theorem is a structural statement about the intersection pairing, asserting that its real extension to $NS (X)_{R}$ has signature $(1, ρ - 1)$ . Without the intersection pairing, the signature statement has no setting; with it, the Hodge index theorem identifies $NS (X)_{R}$ as a Lorentzian space and the ample cone as one connected component of the positive cone. The intersection pairing supplies the symmetric bilinear form; Hodge index supplies its signature.
Adjunction formula 04.05.07. Adjunction and Hodge index are two of the four foundational identities on a smooth projective surface, both anchored to the intersection pairing. Adjunction computes genera of embedded curves through $2 g - 2 = K_{X} \cdot C + C^{2}$ ; Hodge index constrains the global signature of the pairing. The two theorems combine in the K3 surface analysis: the inequality $C^{2} \geq - 2$ for smooth curves on a K3 (from adjunction) and the negative-definiteness of $H^{⊥}$ for an ample $H$ on a K3 (from Hodge index) together produce the lattice structure of the K3 Néron-Severi group.
Riemann-Roch theorem for surfaces 04.05.08. The proof of the Hodge index theorem uses the surface Riemann-Roch identity $χ (O_{X} (D)) = χ (O_{X}) + \frac{1}{2} D \cdot (D - K_{X})$ as the asymptotic input: for a class $D$ in the orthogonal complement of an ample $H$ with $D^{2} > 0$ , surface Riemann-Roch forces unbounded growth of $χ (O_{X} (n D))$ , which contradicts the ample-pairing constraint. Without surface Riemann-Roch, the proof of the orthogonal-complement form breaks; with it, the proof is a one-step asymptotic argument.
Picard group 04.05.02. The Hodge index theorem is most cleanly stated on the Néron-Severi group $NS (X) = Pic (X) / Pic^{0} (X)$ , the quotient of the Picard group by the connected component of the identity. The Picard scheme provides the structure of $Pic^{0} (X)$ as an abelian variety; the Néron-Severi theorem identifies $NS (X)$ as the finitely generated rank- $ρ$ component. Hodge index supplies the signature of the intersection pairing on $NS (X)_{R}$ .
Hodge decomposition 04.09.01. The cohomological version of the Hodge index theorem identifies the algebraic-geometric statement on $NS (X)$ as the integral-lattice shadow of an analytic statement on $H^{1, 1} (X, R)$ . The Hodge-Riemann bilinear relations supply the analytic input; the cycle-class map $NS (X) \otimes R ↪ H^{1, 1} (X, R)$ identifies the algebraic and analytic versions. Hodge decomposition is what makes the cohomological signature pattern $(1, h^{1, 1} - 1)$ precise.
Kodaira vanishing 04.09.02. Kodaira vanishing and the Hodge index theorem together constrain the cohomology of an ample line bundle on a smooth projective surface. Kodaira vanishing kills $h^{1}$ and $h^{2}$ for an ample $L$ twisted by $ω_{X}$ ; Hodge index gates the structure of the orthogonal complement of $L$ in the Néron-Severi group. Both are anchored by the analytic theory of Kähler manifolds and the algebraic theory of positivity on surfaces.
Ample line bundle 04.05.05. The ample cone on a smooth projective surface is one of the two connected components of the positive cone ${D : D^{2} > 0}$ by the Hodge index theorem (Exercise 8). Without Hodge index, the ample cone could in principle be more complicated; with it, the ample cone is identified with a single connected component, and ampleness becomes a signature-driven condition on the divisor class. The Nakai-Moishezon criterion provides the algebraic test; Hodge index provides the geometric setting.
Riemann-Roch theorem for curves 04.04.01. The proof of Hodge index uses curve Riemann-Roch indirectly through the surface Riemann-Roch identity, which itself uses adjunction on a smooth curve embedded in the surface and curve Riemann-Roch on that embedded curve. The chain Hodge index $\leftarrow$ surface Riemann-Roch $\leftarrow$ adjunction $\leftarrow$ curve Riemann-Roch identifies curve theory as the foundational input to the signature theorem on the surface.
Hard Lefschetz theorem 04.09.07. Hard Lefschetz generalises the surface signature pattern to compact Kähler manifolds of arbitrary dimension. The iterated Lefschetz operator $L^{k} : H^{n - k} (X, C) \to H^{n + k} (X, C)$ is an isomorphism, and the induced Lefschetz $sl_{2}$ -decomposition organises the cohomology into primitive subspaces. The surface Hodge index signature $(1, ρ - 1)$ is the algebraic shadow of Hard Lefschetz on degree-2 cohomology: the unique positive direction is the Kähler class and the negative directions are the primitive $(1, 1)$ -classes orthogonal to it.
Hodge-Riemann bilinear relations 04.09.08. The Hodge-Riemann relations are the $n$ -dimensional generalisation of the Hodge index theorem. On a compact Kähler manifold the cup-product pairing twisted by $ω^{n - k}$ is positive definite on each primitive Hodge piece $H_{prim}^{p, q}$ with an explicit sign $i^{p - q} (- 1)^{k (k - 1) /2}$ . The surface Hodge index theorem on $NS (X)$ is the integral-lattice shadow of the Hodge-Riemann relations specialised to $n = 2$ , $k = 2$ , and the cohomological version $(1, h^{1, 1} - 1)$ on $H^{1, 1} (X, R)$ follows from Hodge-Riemann on the primitive $(1, 1)$ -component plus the Kähler-class positive direction.
Signature of a $4 k$ -manifold and the intersection form 03.06.19. The smooth $4$ -manifold signature theorem of 03.06.19 is the differential-topology extension of the surface Hodge index pattern: a closed oriented $4$ -manifold has an integer unimodular intersection form on $H^{2} (M; Z)$ , and the signature pattern $(b_{+}, b_{-})$ is the topological analogue of the surface $(1, ρ - 1)$ Hodge index signature. On a smooth projective surface the two coincide via the cycle-class map $NS (X) ↪ H^{2} (X; Z)$ , identifying the algebraic surface signature theorem as the projective specialisation of the four-manifold signature. Anchor phrase: surface signature = Hodge index; $4$ -manifold extension is 03.06.19.

Historical & philosophical context [Master]

The Hodge index theorem on a smooth projective surface was first proved synthetically by Beniamino Segre in the late 1930s and early 1940s in the Italian-school tradition of intersection-theoretic surface geometry — Sulle curve algebriche di una superficie regolare (Rendiconti del Seminario Matematico di Torino, 1937) ^[pending] and subsequent papers — as a structural statement about the signature of the intersection pairing on the rational Néron-Severi lattice of a smooth complex algebraic surface. Segre's proof used the algebraic surface Riemann-Roch identity together with Serre-style vanishing arguments on the asymptotic growth of cohomology of an ample twist, in the spirit later codified by Hartshorne and Beauville.

The cohomological version — the signature $(1, h^{1, 1} - 1)$ of the cup-product pairing on $H^{1, 1} (X, R)$ for a smooth complex projective surface — was proved by W. V. D. Hodge in The Theory and Applications of Harmonic Integrals (Cambridge University Press 1941) ^[pending] as a corollary of the Hodge-Riemann bilinear relations on primitive cohomology of a Kähler manifold. Hodge's relations are the foundational identity of Hodge theory, asserting that the cup-product pairing on the primitive part of $H^{k} (X, R)$ for a compact Kähler manifold of complex dimension $n$ is definite of alternating sign on each Hodge $(p, q)$ -piece. The surface case ( $n = 2$ , $k = 2$ ) is the original Hodge index theorem; Segre's algebraic statement on $NS (X)$ is the integral-lattice shadow of Hodge's analytic statement on $H^{1, 1} (X, R)$ .

The modern scheme-theoretic framing was assembled by Oscar Zariski, André Weil, and Jean-Pierre Serre in the 1950s and 1960s, with Robin Hartshorne's Algebraic Geometry (Springer GTM 52, 1977) §V.1 Theorem 1.9 providing the canonical English-language statement and proof on a smooth projective surface over an algebraically closed field of arbitrary characteristic. Hartshorne's proof reduces the signature statement to the inequality $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ for $H^{2} > 0$ , then derives the inequality from surface Riemann-Roch together with an asymptotic argument on $χ (O_{X} (n D))$ for large $n$ , with Serre vanishing supplying the cohomology bound. Arnaud Beauville's Complex Algebraic Surfaces (Cambridge LMS Student Texts 34, 1996) §IV.2 ^[pending] gives the modern textbook treatment, and Claire Voisin's Hodge Theory and Complex Algebraic Geometry I (Cambridge SAM 76, 2002) §6.3 ^[pending] supplies the cohomological derivation from the Hodge-Riemann relations.

The Hodge index theorem entered the modern study of surfaces through the work of Šafarevič and Pjateckiĭ-Šafarevič on the Torelli theorem for K3 surfaces in the 1970s, which inverted the Hodge-index-constrained lattice data of a K3 surface to recover the surface itself, and through Miyaoka and Yau's 1977 proof of the Bogomolov-Miyaoka-Yau inequality $K_{X}^{2} \leq 9 χ (O_{X})$ for surfaces of general type, which uses the Hodge index inequality on the Chern classes of the cotangent bundle. The theorem also feeds the Mori program for threefold birational geometry, where the cone theorem and the structure of the Mori cone of curves are signature-constrained by the higher-dimensional analogue of Hodge index on the Néron-Severi space.

Bibliography [Master]

@book{HartshorneAG,
  author    = {Hartshorne, Robin},
  title     = {Algebraic Geometry},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {52},
  year      = {1977}
}

@book{BeauvilleSurfaces,
  author    = {Beauville, Arnaud},
  title     = {Complex Algebraic Surfaces},
  publisher = {Cambridge University Press},
  series    = {London Mathematical Society Student Texts},
  volume    = {34},
  edition   = {2},
  year      = {1996}
}

@book{GriffithsHarris,
  author    = {Griffiths, Phillip and Harris, Joseph},
  title     = {Principles of Algebraic Geometry},
  publisher = {John Wiley \& Sons},
  series    = {Wiley Classics Library},
  year      = {1994}
}

@book{HodgeHarmonic,
  author    = {Hodge, W. V. D.},
  title     = {The Theory and Applications of Harmonic Integrals},
  publisher = {Cambridge University Press},
  year      = {1941}
}

@article{Segre1937,
  author  = {Segre, Beniamino},
  title   = {Sulle curve algebriche di una superficie regolare},
  journal = {Rendiconti del Seminario Matematico di Torino},
  year    = {1937},
  pages   = {1--35}
}

@book{VoisinHodge,
  author    = {Voisin, Claire},
  title     = {Hodge Theory and Complex Algebraic Geometry I},
  publisher = {Cambridge University Press},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {76},
  year      = {2002}
}

@article{Reider1988,
  author  = {Reider, Igor},
  title   = {Vector bundles of rank $2$ and linear systems on algebraic surfaces},
  journal = {Annals of Mathematics},
  volume  = {127},
  year    = {1988},
  pages   = {309--316}
}

@article{Miyaoka1977,
  author  = {Miyaoka, Yoichi},
  title   = {On the {C}hern numbers of surfaces of general type},
  journal = {Inventiones Mathematicae},
  volume  = {42},
  year    = {1977},
  pages   = {225--237}
}

@article{Yau1977,
  author  = {Yau, Shing-Tung},
  title   = {Calabi's conjecture and some new results in algebraic geometry},
  journal = {Proceedings of the National Academy of Sciences USA},
  volume  = {74},
  year    = {1977},
  pages   = {1798--1799}
}

@article{PjateckijSafarevic,
  author  = {Pjateckiĭ-Šapiro, I. I. and Šafarevič, I. R.},
  title   = {A {T}orelli theorem for algebraic surfaces of type {K}3},
  journal = {Izvestiya Akademii Nauk SSSR, Seriya Matematicheskaya},
  volume  = {35},
  year    = {1971},
  pages   = {530--572}
}

@book{LazarsfeldPositivity,
  author    = {Lazarsfeld, Robert},
  title     = {Positivity in Algebraic Geometry I, II},
  publisher = {Springer-Verlag},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete},
  volume    = {48--49},
  year      = {2004}
}

Prerequisites

04.05.06
04.05.07
04.05.08
04.05.02
04.04.01

Tier anchors

beginner: Beauville *Complex Algebraic Surfaces* §IV.2 informal — signature of the intersection form on the Néron-Severi lattice
intermediate: Hartshorne *Algebraic Geometry* §V.1 Theorem 1.9 (Hodge index theorem); Beauville §IV.2
master: Hartshorne §V.1.9; Beauville *Complex Algebraic Surfaces* §IV.2 + §VIII (signature applications); Griffiths-Harris §0.7 + §1.2 (Hodge-Riemann bilinear relations and the Lefschetz decomposition); Voisin *Hodge Theory and Complex Algebraic Geometry* I §6 (Hodge index as a corollary of the Hodge-Riemann relations)

References

TODO_REF
Hartshorne — Algebraic Geometry · §V.1, Theorem 1.9 (Hodge index theorem); Remark 1.9.1 (signature consequences); §V.1 (Néron-Severi group)
TODO_REF
Beauville — Complex Algebraic Surfaces · Ch. IV §2 (Hodge index theorem on a surface, statement and proof); Ch. VIII (signature applications to K3 and Enriques surfaces)
TODO_REF
Griffiths-Harris — Principles of Algebraic Geometry · §0.7 (Hodge-Riemann bilinear relations on a Kähler manifold); §1.2 (Lefschetz decomposition and the primitive cohomology pairing on a complex projective surface)
TODO_REF
Voisin — Hodge Theory and Complex Algebraic Geometry I · Ch. 6 §3 (Hodge-Riemann bilinear relations); Ch. 6 §3.2 (Hodge index theorem as the surface case of the Hodge-Riemann relations on primitive cohomology)
TODO_REF
Hodge — The Theory and Applications of Harmonic Integrals · Cambridge University Press 1941; the originator-text for the Hodge-Riemann bilinear relations from which the Hodge index theorem on a surface is the degree-two specialisation
TODO_REF
Segre — Sulle curve algebriche di una superficie regolare · Rendiconti del Seminario Matematico di Torino (1937) and subsequent papers in the 1940s; the Italian-school index theorem for curves on a surface, predating Hodge's harmonic-integral framing and constituting the algebraic-geometric statement of what is now the Hodge index theorem on a surface

Reviewer

TBD

Estimated time

beginner: 18m
intermediate: 45m
master: 80m