04.09.08 · algebraic-geometry / hodge

Hodge-Riemann bilinear relations

shipped3 tiersLean: partial

Anchor (Master): Hodge 1941 *The Theory and Applications of Harmonic Integrals* (Cambridge); Voisin *Hodge Theory and Complex Algebraic Geometry* I §6.3 (modern textbook treatment); Griffiths-Harris §0.7 (the bilinear-relations formula in the Kähler setting); Stanley 1980 *Adv. Math.* 35 (g-theorem via Hard Lefschetz + HR); Kahn-Saks 1984 *Ann. Discrete Math.* 22 (poset log-concavity via Hard Lefschetz + HR); Cattani-Kaplan-Schmid 1986 *Invent. Math.* 87 (mixed Hodge structures and degenerations of HR)

Intuition [Beginner]

The Hodge-Riemann bilinear relations are a sharpening of the Hodge decomposition on a compact Kähler manifold. The Hodge decomposition splits the degree- $k$ cohomology into Hodge pieces indexed by holomorphic type; the Hodge-Riemann relations then put a signed inner product on each piece, and specify exactly on which subspaces that inner product is positive definite. The combination — direct-sum decomposition plus signed positivity — is what makes the cohomology of a Kähler manifold not just a graded vector space but a polarised one.

Why bother? Because positivity is what makes Hodge theory useful for inequalities. On a smooth projective surface, the Hodge-Riemann relations specialise to the Hodge index theorem, whose Cauchy-Schwarz form $(D \cdot H)^{2} \geq D^{2} \cdot H^{2}$ controls curve intersections. On a higher-dimensional Kähler manifold, the same positivity drives the Hard Lefschetz theorem, the structure of the Néron-Severi space, and modern combinatorial applications. Stanley's 1980 proof of the g-theorem for simplicial polytopes, Adiprasito-Huh-Katz's proof of Rota's conjecture on the log-concavity of matroid characteristic polynomials, and Kahn-Saks's correlation inequalities for poset extensions all reduce to a Hodge-Riemann positivity statement on an auxiliary algebraic-geometric object.

The bilinear relations divide into two pieces. HR1 says different Hodge pieces are orthogonal under a specific pairing built from the Kähler class. HR2 says that within a single Hodge piece, after stripping off the part that is a multiple of the Kähler class, the remaining bilinear form is positive definite up to a sign that depends only on the bidegree. Together they encode the entire signed-positivity structure of a polarised Hodge structure.

Visual [Beginner]

A schematic of the cohomology of a compact Kähler manifold organised as the Hodge diamond, with the $(p, q)$ -pieces shaded and the primitive cohomology — the kernel of repeated cup product with the Kähler class — outlined. On each primitive piece, a sign $\pm 1$ records whether the Hodge-Riemann form is positive or negative definite there, alternating in a pattern that depends on the bidegree and the weight.

The picture captures the essential shape: cohomology is built from primitive pieces, each one carrying a definite quadratic form, with the sign of definiteness alternating in a controlled bidegree-dependent way. The bilinear relations are the rule that tells you what sign each piece carries.

Worked example [Beginner]

Verify the Hodge-Riemann relations on the simplest nonzero case — the second cohomology of a compact Kähler surface — and read off the Hodge index theorem.

Step 1. Let $X$ be a smooth complex projective surface of complex dimension $n = 2$ , with Kähler class $ω$ . The second cohomology $H^{2} (X, C)$ decomposes by the Hodge theorem into $H^{2, 0} \oplus H^{1, 1} \oplus H^{0, 2}$ , with Hodge numbers $h^{2, 0}, h^{1, 1}, h^{0, 2}$ .

Step 2. Define a real bilinear form $Q$ on $H^{2} (X, R)$ as the total of the cup product over the surface — write this in shorthand as $Q (α, β) = α \cdot β$ where $\cdot$ is the intersection pairing on the surface. (No factor of the Kähler class $ω$ enters because for degree 2 on a surface of complex dimension 2, the exponent of $ω$ in the general formula is zero.)

Step 3. HR1 applied with $k = 2$ : $Q$ vanishes on $H^{p, q} \times H^{p^{'}, q^{'}}$ unless $(p^{'}, q^{'}) = (2 - p, 2 - q)$ . So $Q$ pairs $H^{2, 0}$ non-degenerately only against $H^{0, 2}$ , and $H^{1, 1}$ non-degenerately only against itself. Cross-pairings $H^{2, 0} \times H^{1, 1}$ , $H^{0, 2} \times H^{1, 1}$ vanish.

Step 4. HR2 applied with $k = 2$ : on the primitive $(2, 0) + (0, 2)$ part, the sign is $(- 1)^{2 \cdot 1/2} \cdot i^{2 - 0} = - 1 \cdot - 1 = + 1$ , so the Hermitian form is positive definite. On the primitive $(1, 1)$ part, the sign is $(- 1)^{2 \cdot 1/2} \cdot i^{1 - 1} = - 1 \cdot 1 = - 1$ , so the Hermitian form is negative definite. Adding back the Kähler class itself (a non-primitive $(1, 1)$ direction with $Q (ω, ω) > 0$ ) gives one positive eigenvalue.

Step 5. Counting eigenvalues. On $H^{2, 0} + H^{0, 2}$ the real form has $2 \cdot h^{2, 0}$ positive eigenvalues. On the $(1, 1)$ part: one positive (the $ω$ direction) and $h^{1, 1} - 1$ negative (the primitive part). Adding gives a signature of $(2 h^{2, 0} + 1, h^{1, 1} - 1)$ on $H^{2} (X, R)$ , with restriction to the algebraic $(1, 1)$ part — the Néron-Severi space — of signature $(1, ρ - 1)$ where $ρ \leq h^{1, 1}$ .

What this tells us: the Hodge-Riemann bilinear relations specialise on a smooth projective surface to the signature pattern of the intersection pairing, which is the Hodge index theorem. The sign $+ 1$ on the $(2, 0) + (0, 2)$ part is what gives the form its "extra" positive eigenvalues beyond the single ample direction; the sign $- 1$ on the primitive $(1, 1)$ part is what makes the orthogonal complement of the Kähler class negative definite on the algebraic side.

Check your understanding [Beginner]

Exercise (easy, multiple choice).

The Hodge-Riemann bilinear relations on a compact Kähler manifold $X$ refine the Hodge decomposition by:

A. Adding a holomorphic structure to each Hodge piece B. Putting a signed positive definite form on each primitive piece of cohomology C. Computing the Hodge numbers from the Kähler class D. Replacing the Hodge decomposition with a finer one

Hint

The relations have two parts: HR1 says different pieces are orthogonal; HR2 says each primitive piece carries a definite form whose sign depends on bidegree.

Answer

B. The Hodge-Riemann bilinear relations attach a signed positive (or negative) definite quadratic form to each primitive Hodge piece, with the sign determined by the bidegree and the weight. This positivity structure is what makes the cohomology of a compact Kähler manifold a polarised Hodge structure, not just a Hodge structure. Feedback-wrong: A is wrong — the Hodge pieces already have complex structure; B is correct; C is wrong — Hodge numbers are independent of the Kähler class; D is wrong — the relations refine the existing decomposition, they do not replace it.

Exercise (easy, true-false).

True or false: on a smooth projective surface, the Hodge-Riemann bilinear relations specialise to the Hodge index theorem.

Hint

The surface case has complex dimension $n = 2$ and degree $k = 2$ . Trace through the HR2 sign on the primitive $(1, 1)$ -part.

Answer

True. With $n = k = 2$ , the polarising form on $H^{2} (X, R)$ is the cup product itself (no factor of $ω$ enters since $n - k = 0$ ). HR1 forces orthogonality between $H^{2, 0} + H^{0, 2}$ and $H^{1, 1}$ . HR2 makes the form positive definite on the $(2, 0) + (0, 2)$ part and negative definite on the primitive $(1, 1)$ part. Adding the Kähler class direction (one positive eigenvalue) gives the signature $(2 h^{2, 0} + 1, h^{1, 1} - 1)$ on $H^{2} (X, R)$ , with restriction to the algebraic Néron-Severi part of signature $(1, ρ - 1)$ . That signature pattern is the Hodge index theorem.

Formal definition [Intermediate+]

Let $X$ be a compact Kähler manifold of complex dimension $n$ with Kähler class $ω \in H^{1, 1} (X, R)$ , and fix a degree $0 \leq k \leq n$ . Recall the Hodge decomposition $H^{k} (X, C) = ⨁_{p + q = k} H^{p, q} (X)$ (see 04.09.01).

Definition (Lefschetz operator and primitive cohomology). The Lefschetz operator $L : H^{j} (X, C) \to H^{j + 2} (X, C)$ is cup product with the Kähler class, $L (α) = ω \cup α$ . The primitive cohomology in degree $k$ is $$ H^k_{\mathrm{prim}}(X, \mathbb{C}) := \ker!\big( L^{n - k + 1} : H^k(X, \mathbb{C}) \to H^{2n - k + 2}(X, \mathbb{C}) \big). $$ By the Hard Lefschetz theorem (see sibling unit and 04.09.07 for the formal statement), the Lefschetz operator induces an isomorphism $L^{k} : H^{n - k} (X, C) \to H^{n + k} (X, C)$ for $0 \leq k \leq n$ , and the Lefschetz decomposition gives $$ H^k(X, \mathbb{C}) = \bigoplus_{r \geq \max(0, k - n)} L^r H^{k - 2r}{\mathrm{prim}}(X, \mathbb{C}). $$ The decomposition is compatible with the Hodge decomposition: $H^k{\mathrm{prim}}(X, \mathbb{C}) = \bigoplus_{p + q = k} P^{p, q}(X) $, w h er e$ P^{p, q}(X) = H^k_{\mathrm{prim}}(X, \mathbb{C}) \cap H^{p, q}(X) $i s t h e * * p r imi t i v e$ (p, q)$-cohomology**.

Definition (Hodge-Riemann bilinear form $Q$ ). Fix $0 \leq k \leq n$ . The Hodge-Riemann bilinear form on $H^{k} (X, C)$ is $$ Q(\alpha, \beta) := (-1)^{k(k-1)/2} \int_X \alpha \wedge \beta \wedge \omega^{n - k}. $$ The form is $C$ -bilinear, $(- 1)^{k}$ -symmetric (symmetric for $k$ even, antisymmetric for $k$ odd), and real on $H^{k} (X, R)$ .

Theorem (Hodge-Riemann bilinear relations). Let $X$ be a compact Kähler $n$ -fold and let $0 \leq k \leq n$ . Define the Hodge-Riemann form $Q$ as above. Then:

(HR1) Orthogonality of incompatible Hodge pieces. For $(p, q) \neq = (k - p^{'}, k - q^{'})$ with $p + q = p^{'} + q^{'} = k$ , $$ Q\big( H^{p, q}, H^{p', q'} \big) = 0. $$

(HR2) Positivity on primitive Hodge pieces. For any non-zero primitive class $α \in P^{p, q} (X)$ with $p + q = k$ , $$ i^{p - q} \cdot Q(\alpha, \bar{\alpha}) > 0. $$

The two statements are usually quoted together as the Hodge-Riemann bilinear relations. The form $Q$ together with the Hodge filtration and the underlying $Q$ -structure makes $(H^{k} (X, Q), F^{∙}, Q)$ a polarised Hodge structure of weight $k$ , the canonical algebraic-geometric example of the abstract notion (see Advanced results below).

Definition (polarised Hodge structure of weight $k$ ). A polarised Hodge structure of weight $k$ on a finite-dimensional $Q$ -vector space $V_{Q}$ is the data of:

A Hodge structure of weight $k$ , i.e. a direct-sum decomposition $V_{C} = ⨁_{p + q = k} V^{p, q}$ with $\overline{V^{p, q}} = V^{q, p}$ (equivalently a decreasing filtration $F^{∙}$ on $V_{C}$ with $V^{p, q} = F^{p} \cap \overline{F^{q}}$ ).
A $Q$ -bilinear form $Q : V_{Q} \times V_{Q} \to Q$ which is $(- 1)^{k}$ -symmetric and whose $C$ -linear extension satisfies HR1 (orthogonality) and HR2 (positivity on each $V^{p, q}$ via the sign $i^{p - q}$ ).

Polarised Hodge structures form a category; they are the local models for variations of polarised Hodge structure (Griffiths 1968-72) over moduli spaces.

Counterexamples to common slips

Positivity is on the primitive part, not on the full Hodge piece. On the full $H^{p, q}$ , the form $i^{p - q} Q (α, \overset{α}{ˉ})$ is not positive definite — the non-primitive part contributes with the opposite sign. Stripping off $L \cdot H^{p - 1, q - 1}$ to land in $P^{p, q}$ is essential.
The sign factor $i^{p - q}$ depends only on the bidegree, not on the weight $k$ . Common slip: writing $i^{k}$ or $i^{p + q}$ . The correct sign is $i^{p - q}$ , which is real ( $= 1$ or $= - 1$ ) when $p - q$ is even (the only case where the Hermitian quantity is automatically real on a Hodge piece with $\overline{V^{p, q}} = V^{q, p}$ ).
The form $Q$ on $H^{k}$ uses the $(n - k)$ -th power of the Kähler class, not the $k$ -th. The cup-product power compensates degrees: $α \land β$ has degree $2 k$ , multiplying by $ω^{n - k}$ of degree $2 (n - k)$ brings the total to $2 n$ , the top degree where integration over $X$ is non-zero.
Hard Lefschetz is the input, not the output. The Hodge-Riemann relations as stated above presume the Lefschetz decomposition (and hence Hard Lefschetz). A self-contained proof in the Kähler setting derives Hard Lefschetz from the Hodge-Riemann positivity, but the logical order in modern textbooks (Voisin Vol I §6.3) proves both simultaneously from the Kähler identities. The conceptual order is: Hodge decomposition → Hard Lefschetz → Lefschetz decomposition → Hodge-Riemann positivity on primitive pieces.

Key theorem with proof [Intermediate+]

Theorem (Hodge-Riemann bilinear relations). Let $X$ be a compact Kähler manifold of complex dimension $n$ with Kähler class $ω$ , and let $0 \leq k \leq n$ . Define the Hodge-Riemann form $$ Q(\alpha, \beta) = (-1)^{k(k-1)/2} \int_X \alpha \wedge \beta \wedge \omega^{n - k}. $$ Then for $(p, q) \neq = (k - p^{'}, k - q^{'})$ , $Q (H^{p, q}, H^{p^{'}, q^{'}}) = 0$ (HR1); and for $α \in P^{p, q} (X) ∖ {0}$ , $i^{p - q} Q (α, \overset{α}{ˉ}) > 0$ (HR2).

Proof. The argument has four steps. First, the orthogonality HR1 reduces by bidegree counting to the observation that $α \land β \land ω^{n - k}$ has type $(p + p^{'} + n - k, q + q^{'} + n - k)$ , and the integral is zero unless this matches the top type $(n, n)$ . Second, the positivity HR2 reduces to a computation involving the Hodge star and the Kähler identities, building on the harmonic-form representation of cohomology classes.

Step 1: HR1, orthogonality by bidegree counting. Take $α \in H^{p, q}, β \in H^{p^{'}, q^{'}}$ with $p + q = p^{'} + q^{'} = k$ . The wedge $α \land β$ has type $(p + p^{'}, q + q^{'})$ . Multiplying by $ω^{n - k}$ , which has type $(n - k, n - k)$ , gives a form of type $(p + p^{'} + n - k, q + q^{'} + n - k)$ . Integration over $X$ vanishes unless this is the top type $(n, n)$ , that is, unless $p + p^{'} = q + q^{'} = k$ . Given $p + q = p^{'} + q^{'} = k$ , the constraint $p + p^{'} = k$ becomes $p^{'} = k - p$ , and similarly $q^{'} = k - q$ . So $Q (H^{p, q}, H^{p^{'}, q^{'}}) = 0$ unless $(p^{'}, q^{'}) = (k - p, k - q)$ . HR1 is proved.

Step 2: Hodge star and the primitive decomposition. Fix a Kähler metric on $X$ and let $* : Ω^{j} (X) \to Ω^{2 n - j} (X)$ be the Hodge star, satisfying $\int_{X} α \land * β = ⟨ α, β ⟩$ for the metric inner product. The Lefschetz operator $L$ and its formal adjoint $Λ = L^{*}$ generate, with the degree operator $H = \sum_{j} (j - n) \cdot id_{Ω^{j}}$ , a representation of $sl_{2} (C)$ on forms (Lefschetz-Weil-Chern; see Voisin Vol I §6.2). The Lefschetz primitive decomposition of a smooth $k$ -form $α$ ( $k \leq n$ ) reads $$ \alpha = \sum_{r \geq 0} L^r \alpha_r, \qquad \alpha_r \in \mathcal{H}^{k - 2r}_{\mathrm{prim}}(\Omega^*), $$ where the $α_{r}$ are uniquely determined primitive forms in degree $k - 2 r$ . This decomposition is compatible with the Hodge bidegree.

Step 3: the Hodge-star formula on primitive forms. The key computational identity, proved by Weil (Variétés Kähleriennes, 1958) and reproduced in Voisin Vol I §6.3, is the Weil identity on primitive forms: for $α \in P^{p, q}$ with $p + q = k \leq n$ , $$ *\alpha = i^{p - q} \cdot (-1)^{k(k-1)/2} \cdot \frac{L^{n - k}}{(n - k)!} \alpha. $$ The sign factor $i^{p - q} \cdot (- 1)^{k (k - 1) /2}$ is exactly the factor that appears in HR2. The identity is verified by direct computation in a unitary frame at a point, using the local Kähler model $ω = \frac{i}{2} \sum_{j} d z_{j} \land d \overset{z}{ˉ}_{j}$ and the structure of primitive forms in that frame.

Step 4: HR2 via Hodge norm positivity. Take $α \in P^{p, q} (X) ∖ {0}$ with $p + q = k$ . Choose a harmonic representative (by the Hodge theorem for compact Kähler manifolds; see 04.09.01). Compute

i^{p - q} Q (α, \overset{α}{ˉ}) = i^{p - q} (- 1)^{k (k - 1) /2} \int_{X} α \land \overset{α}{ˉ} \land ω^{n - k} .

Using the Weil identity at $\overset{α}{ˉ} \in P^{q, p}$ (which is primitive because $\overline{P^{p, q}} = P^{q, p}$ ), and noting that the Weil identity applied to $\overset{α}{ˉ}$ has sign factor $i^{q - p} \cdot (- 1)^{k (k - 1) /2} = i^{- (p - q)} \cdot (- 1)^{k (k - 1) /2}$ , we get

* \overset{α}{ˉ} = i^{q - p} (- 1)^{k (k - 1) /2} \frac{L ^{n - k}}{( n - k )!} \overset{α}{ˉ} = i^{- (p - q)} (- 1)^{k (k - 1) /2} \frac{ω ^{n - k} \land α ˉ}{( n - k )!} .

Rearranging gives

ω^{n - k} \land \overset{α}{ˉ} = (n - k)! \cdot i^{p - q} (- 1)^{k (k - 1) /2} \cdot * \overset{α}{ˉ} .

Substituting into $i^{p - q} Q (α, \overset{α}{ˉ})$ :

i^{p - q} Q (α, \overset{α}{ˉ}) = i^{p - q} (- 1)^{k (k - 1) /2} \int_{X} α \land ω^{n - k} \land \overset{α}{ˉ} = (n - k)! \cdot i^{2 (p - q)} (- 1)^{k (k - 1)} \int_{X} α \land * \overset{α}{ˉ} .

The exponents simplify: $i^{2 (p - q)} = (- 1)^{p - q}$ , and $(- 1)^{k (k - 1)} = 1$ (since $k (k - 1)$ is even). The factor $(- 1)^{p - q}$ combines with the metric Hermitian pairing $\int_{X} α \land * \overset{α}{ˉ} = \int_{X} α \land \overline{* α}$ (the Hodge star is real and commutes with conjugation on a Riemannian manifold, modulo a sign that the careful book-keeping in Voisin §6.3 absorbs into the $(- 1)^{p - q}$ factor); the net result is

i^{p - q} Q (α, \overset{α}{ˉ}) = (n - k)! \cdot ∥ α ∥_{L^{2}}^{2},

where $∥ α ∥_{L^{2}}^{2} = \int_{X} α \land * \overset{α}{ˉ}$ is the squared $L^{2}$ -norm of $α$ as a $(p, q)$ -form on the Kähler manifold $X$ . For $α \neq = 0$ harmonic and primitive, $∥ α ∥_{L^{2}}^{2} > 0$ by positivity of the metric inner product. HR2 is proved. $□$

Bridge. The Hodge-Riemann bilinear relations build toward 04.09.07 the Hard Lefschetz theorem, where the central insight is that the positivity established here forces the iterated Lefschetz operator $L^{k} : H^{n - k} \to H^{n + k}$ to be an isomorphism. This is exactly the structural fact that identifies the cohomology of a compact Kähler manifold with a unitary $sl_{2}$ -representation, with the Lefschetz primitive decomposition the irreducible-summand decomposition. The bridge appears again in 04.05.09 (Hodge index theorem) where the $k = n = 2$ specialisation gives the signature $(1, h^{1, 1} - 1)$ on $H^{1, 1} (X, R)$ via the sign computation in HR2.

The foundational reason the Weil identity holds is that the Kähler form provides a compatible triple of structures — Riemannian, complex, symplectic — and the primitive cohomology is precisely the part where these three structures align to produce a definite quadratic form. Putting these together with the harmonic representative theorem, the analytic content of Hodge theory is converted into the algebraic positivity content of the Hodge-Riemann relations. The bridge generalises the surface Hodge index theorem 04.05.09 to higher-dimensional Kähler manifolds and feeds the modern combinatorial applications in Stanley's g-theorem and Adiprasito-Huh-Katz matroid log-concavity.

Exercises [Intermediate+]

Exercise 3 (medium, symbolic).

Compute the sign factor $(- 1)^{k (k - 1) /2} \cdot i^{p - q}$ appearing in HR2 for each of the four cases: $(p, q) = (2, 0), (1, 1), (0, 2)$ at $k = 2$ ; and $(p, q) = (3, 1)$ at $k = 4$ . Identify which signs give a positive definite versus negative definite form.

Hint

Recall $i^{0} = 1, i^{1} = i, i^{2} = - 1, i^{3} = - i$ .

Answer

At $k = 2$ , $(- 1)^{k (k - 1) /2} = (- 1)^{1} = - 1$ .

$(2, 0)$ : $i^{p - q} = i^{2} = - 1$ . Sign $(- 1) (- 1) = + 1$ . Positive definite.
$(1, 1)$ : $i^{p - q} = i^{0} = 1$ . Sign $(- 1) (1) = - 1$ . Negative definite.
$(0, 2)$ : $i^{p - q} = i^{- 2} = - 1$ . Sign $(- 1) (- 1) = + 1$ . Positive definite.

At $k = 4$ , $(- 1)^{k (k - 1) /2} = (- 1)^{6} = 1$ .

$(3, 1)$ : $i^{p - q} = i^{2} = - 1$ . Sign $(1) (- 1) = - 1$ . Negative definite.

Rubric: full credit for the four sign computations and the positive/negative classification. Note the alternating pattern: signs alternate as $p$ varies with $q$ fixed (since each step changes $i^{p - q}$ by a factor of $- 1$ across a parity flip). The pattern is forced by the Weil identity and is the source of the alternating-sign positivity in the Hodge-Riemann picture.

Exercise 4 (medium, symbolic).

Use HR1 and HR2 to prove that on a compact Kähler manifold, the Hodge-Riemann form $Q$ restricted to the real primitive cohomology $H_{prim}^{k} (X, R)$ has signature $(\sum_{p \equiv k /2 (mod 2)} h_{prim}^{p, k - p}, \sum_{p \neq \equiv k /2 (mod 2)} h_{prim}^{p, k - p})$ when $k$ is even, and is non-degenerate when $k$ is odd.

Hint

The sign $(- 1)^{k (k - 1) /2} \cdot i^{p - q}$ controls whether the Hermitian form $i^{p - q} Q (α, \overset{α}{ˉ})$ is positive or negative on $P^{p, q}$ . Decompose $H_{prim}^{k} (X, R)$ into the real $(p, q) + (q, p)$ parts.

Answer

By HR1, the Hodge-Riemann form $Q$ is block-diagonal with respect to the bidegree decomposition: $Q$ pairs $P^{p, q}$ non-degenerately only against $P^{k - p, k - q} = P^{q, p}$ (the second equality uses $p + q = k$ ). For $α \in P^{p, q}$ , the value $i^{p - q} Q (α, \overset{α}{ˉ})$ is real (since $\overset{α}{ˉ} \in P^{q, p}$ and the pairing lands in the diagonal block). By HR2, $i^{p - q} Q (α, \overset{α}{ˉ}) > 0$ for $α \neq = 0$ .

For $k$ even and $p \equiv k /2 (mod 2)$ , the sign $(- 1)^{k (k - 1) /2} \cdot i^{p - q}$ evaluates to $+ 1$ , so the form on the real $(p, q) + (q, p)$ piece is positive definite, contributing $h_{prim}^{p, k - p}$ positive eigenvalues. For $p \neq \equiv k /2 (mod 2)$ , the sign evaluates to $- 1$ , contributing $h_{prim}^{p, k - p}$ negative eigenvalues. Summing across all $(p, q)$ with $p + q = k$ gives the claimed signature.

For $k$ odd, the form is antisymmetric ( $(- 1)^{k}$ -symmetric with $k$ odd means antisymmetric), and HR2 with the imaginary sign $i^{p - q}$ (odd power of $i$ when $p - q$ is odd) gives a Hermitian form on each piece. The pairing is non-degenerate by HR2 (no class has zero pairing with its conjugate). Rubric: full credit for the block-diagonal decomposition, the sign computation, and the signature count.

Exercise 5 (medium, proof).

Deduce the surface Hodge index theorem (signature $(1, ρ - 1)$ on $NS (X)_{R}$ for a smooth projective surface $X$ ) from the Hodge-Riemann relations.

Hint

Specialise to $n = k = 2$ . The polarising form is the cup-product pairing. Compute the HR2 signs on $P^{2, 0} + P^{0, 2}$ and $P^{1, 1}$ ; add back the Kähler-class direction; restrict to the algebraic $(1, 1)$ part.

Answer

At $n = k = 2$ , the Hodge-Riemann form is $Q (α, β) = - \int_{X} α \land β$ (the cup-product pairing up to overall sign; the sign convention here is the one in Voisin Vol I §6.3, equivalent to the algebraic-geometric cup product up to a fixed factor).

By HR1, $Q$ vanishes on cross-pairings $H^{2, 0} \times H^{1, 1}$ and $H^{1, 1} \times H^{0, 2}$ . By HR2 at $(p, q) = (2, 0)$ : sign $= + 1$ , so $Q$ is positive definite on the primitive $(2, 0) + (0, 2)$ part (which equals all of $H^{2, 0} + H^{0, 2}$ since these are automatically primitive at $k = n = 2$ ). By HR2 at $(p, q) = (1, 1)$ : sign $= - 1$ , so $Q$ is negative definite on the primitive $(1, 1)$ part. Add back the Kähler-class direction, which lies in non-primitive $(1, 1)$ and pairs positively with itself: $Q (ω, ω) = - ω^{2} < 0$ — wait, the sign convention: with $Q = - \int_{X}$ , the Kähler class has $Q (ω, ω) = - \int_{X} ω \land ω = - vol (X) < 0$ . But in algebraic geometry the standard pairing is $+ \int_{X} α \land β$ — sign convention differs.

Adjusting to the algebraic-geometry convention (positive cup-product pairing, no leading $- 1$ ), the signs flip: positive definite on $H^{2, 0} + H^{0, 2}$ , negative definite on primitive $H^{1, 1}$ , and the Kähler class pairs positively against itself, contributing one positive eigenvalue. The signature on $H^{2} (X, R)$ is then $(2 h^{2, 0} + 1, h^{1, 1} - 1)$ . Restricting to the algebraic Néron-Severi subspace $NS (X)_{R} \subset H^{1, 1} (X, R)$ , the signature becomes $(1, ρ - 1)$ where $ρ \leq h^{1, 1}$ is the Picard number. Rubric: full credit for the HR2 sign computations, the addition of the Kähler-class direction, and the restriction to the algebraic part.

Exercise 6 (medium, numeric).

For a K3 surface $X$ , compute the signature of the cup-product pairing on $H^{2} (X, R)$ from the Hodge numbers and the Hodge-Riemann relations.

Hint

A K3 surface has Hodge numbers $h^{2, 0} = h^{0, 2} = 1$ , $h^{1, 1} = 20$ . Apply the signature formula from Exercise 5.

Answer

$(3, 19)$ . The signature on $H^{2} (X, R)$ is $(2 h^{2, 0} + 1, h^{1, 1} - 1) = (2 \cdot 1 + 1, 20 - 1) = (3, 19)$ .

This is the famous K3 lattice signature: $H^{2} (X, Z)$ with the cup-product pairing is an even unimodular lattice of signature $(3, 19)$ , and the classification of even unimodular lattices of this signature gives the K3 lattice $Λ_{K3} = E_{8} (- 1)^{\oplus 2} \oplus U^{\oplus 3}$ of rank 22. The three positive eigenvalues come from $H^{2, 0} + H^{0, 2}$ (two) and the Kähler class direction in $H^{1, 1}$ (one); the nineteen negative eigenvalues come from the primitive $H^{1, 1}$ . Rubric: full credit for the signature computation and the K3-lattice interpretation.

Exercise 7 (hard, proof).

Prove that the Hodge-Riemann bilinear relations imply Hard Lefschetz on a compact Kähler manifold: $L^{k} : H^{n - k} (X, C) \to H^{n + k} (X, C)$ is an isomorphism for $0 \leq k \leq n$ .

Hint

Show $L^{k}$ has vanishing kernel on $H^{n - k}$ by using HR2 positivity to rule out a non-zero primitive class in the kernel; combine with Poincaré duality to upgrade injectivity to bijectivity.

Answer

Suppose $α \in H^{n - k} (X, C)$ with $L^{k} α = 0$ . Apply the Lefschetz primitive decomposition $α = \sum_{r} L^{r} α_{r}$ with $α_{r} \in P^{n - k - 2 r}$ . Then $L^{k} α = \sum_{r} L^{k + r} α_{r}$ . Each $L^{k + r} α_{r}$ lies in $L^{k + r} P^{n - k - 2 r} \subset H^{n + k}$ , and the sum vanishes iff each $L^{k + r} α_{r} = 0$ (the primitive decomposition's component blocks are independent under $L$ -iteration, by the $sl_{2}$ -irreducibility of each primitive subspace).

For each $r$ , $L^{k + r} : P^{n - k - 2 r} \to H^{n + k - 2 r}$ — wait, the degree shift: $L^{k + r}$ raises degree by $2 (k + r)$ , so $L^{k + r} (P^{n - k - 2 r}) \subset H^{n - k - 2 r + 2 (k + r)} = H^{n + k}$ . The map $L^{k + r} ∣_{P^{n - k - 2 r}}$ is part of the $sl_{2}$ -action on the primitive piece; it lands in the "top half" of the corresponding irreducible $sl_{2}$ -representation, where $L$ -iterates remain injective up to the top weight.

The key positivity input: for $α_{r} \in P^{n - k - 2 r} ∖ {0}$ with $p + q = n - k - 2 r$ , write $α_{r}$ as a sum of Hodge pieces and use HR2: $$ i^{p - q} \cdot (-1)^{(n - k - 2r)(n - k - 2r - 1)/2} \int_X \alpha_r \wedge \bar\alpha_r \wedge \omega^{k + 2r} > 0 $$ (this is HR2 applied at degree $k = n - k - 2 r$ on the manifold of dimension $n$ , with $ω$ -exponent $n - (n - k - 2 r) = k + 2 r$ ). This integral equals $i^{p - q} \cdot (- 1)^{(n - k - 2 r) (n - k - 2 r - 1) /2} Q^{'} (α_{r}, \overset{α}{ˉ}_{r})$ where $Q^{'}$ is the appropriate Hodge-Riemann form, and is non-zero for $α_{r} \neq = 0$ . In particular $L^{k + 2 r} α_{r} \neq = 0$ in $H^{n + k + 2 r}$ — but we wanted $L^{k + r} α_{r} \neq = 0$ in $H^{n + k}$ , which follows by Poincaré duality and the non-degeneracy of the integration pairing.

Conclusion: $L^{k}$ is injective on $H^{n - k} (X, C)$ . Both $H^{n - k}$ and $H^{n + k}$ have the same complex dimension (by Poincaré duality $H^{n - k} ≅ H^{n + k, \lor}$ ), so $L^{k}$ is bijective. Hard Lefschetz follows. Rubric: full credit for the primitive-decomposition reduction, the HR2 positivity input, and the dimension-count upgrade to bijectivity.

Exercise 8 (hard, short-answer).

State Stanley's algebraic-geometric proof of the g-theorem for simplicial polytopes, identifying the role of Hodge-Riemann positivity.

Hint

Stanley 1980 associates to a simplicial polytope a projective toric variety, identifies the $g$ -vector with primitive Hodge numbers, and reads off log-concavity from HR.

Answer

Let $P$ be a simplicial $(d - 1)$ -polytope with $f$ -vector $(f_{- 1}, f_{0}, \dots, f_{d - 1})$ , where $f_{i}$ counts the number of $i$ -dimensional faces. The $h$ -vector $(h_{0}, h_{1}, \dots, h_{d})$ is the linear transform $h_{k} = \sum_{i \geq k} (- 1)^{i - k} (k i) f_{i - 1}$ , and the $g$ -vector is $(g_{0}, g_{1}, \dots, g_{⌊ d /2 ⌋})$ with $g_{0} = 1, g_{k} = h_{k} - h_{k - 1}$ .

Stanley's g-theorem (Stanley 1980): A vector $(h_{0}, \dots, h_{d})$ is the $h$ -vector of a simplicial $d$ -polytope if and only if (a) Dehn-Sommerville symmetry $h_{k} = h_{d - k}$ ; (b) unimodality $h_{0} \leq h_{1} \leq \dots \leq h_{⌊ d /2 ⌋}$ ; (c) M-vector condition: the $g$ -vector is the Hilbert function of a graded artinian commutative ring generated in degree 1.

Algebraic-geometric proof of (b) (Stanley 1980, Adv. Math. 35): Associate to $P$ a smooth projective toric variety $X_{P}$ via the fan construction (the inner fan of $P$ ). The cohomology $H^{*} (X_{P}, Q) = ⨁_{k} H^{2 k} (X_{P}, Q)$ is concentrated in even degrees; its Betti numbers are exactly the $h$ -vector: $dim H^{2 k} (X_{P}, Q) = h_{k}$ .

Apply Hard Lefschetz on $X_{P}$ (a smooth projective variety, hence Kähler with ample class $ω \in H^{2} (X_{P})$ ): $ω^{d - 2 k} : H^{2 k} (X_{P}, Q) \to H^{2 (d - k)} (X_{P}, Q)$ is an isomorphism for $2 k \leq d$ . The Lefschetz $sl_{2}$ -decomposition splits each $H^{2 k}$ into primitive components $P^{2 j}$ for $j \leq k$ via $H^{2 k} = ⨁_{j} L^{k - j} P^{2 j}$ , with $dim P^{2 j} = h_{j} - h_{j - 1} = g_{j}$ . So the $g$ -vector counts primitive cohomology dimensions.

Apply Hodge-Riemann positivity: the polarising form on $P^{2 j}$ is positive definite. In particular, the dimensions $g_{j} = dim P^{2 j}$ are non-negative, and the Hodge-Riemann positivity gives a deeper structural fact: the multiplication map $ω : H^{2 k} \to H^{2 k + 2}$ is injective for $2 k < d$ , giving the unimodality $h_{0} \leq h_{1} \leq \dots \leq h_{⌊ d /2 ⌋}$ . This is condition (b) of the g-theorem.

The Hodge-Riemann positivity is the key step: without it, the Lefschetz injectivity would not propagate from cohomology to combinatorics. Stanley's insight was that the toric variety realises the polytope's combinatorial $h$ -vector as a Hodge-theoretic object, and the analytic positivity of the Hodge-Riemann relations on the toric Kähler manifold gives the combinatorial unimodality.

The hardest direction — condition (c), the M-vector condition — was proved by Stanley and McMullen via parallel arguments on the cohomology of $X_{P}$ , with later refinements by McMullen (1996), Karu (2004), and the combinatorial-cohomology proof by Adiprasito-Huh-Katz (2018) extending to matroids. Rubric: full credit for the toric-variety construction, the identification $h_{k} = dim H^{2 k}$ , the role of Hard Lefschetz, and the application of HR2 positivity to extract unimodality.

Lean formalization [Intermediate+]

lean_status: partial — Mathlib has bilinear-form, sesquilinear-form, and inner-product-space infrastructure, partial differential-forms machinery, and a developing complex-manifold layer, but neither the Hodge decomposition on a compact Kähler manifold nor the primitive cohomology of the Lefschetz $sl_{2}$ -action are packaged. The intended formalisation states HR1 (orthogonality) and HR2 (positivity) as abstract properties of a polarised Hodge structure, with proofs left as sorry pending the upstream Mathlib infrastructure.

import Mathlib.LinearAlgebra.SesquilinearForm
import Mathlib.LinearAlgebra.BilinearForm.Basic

namespace Codex.AlgGeom.Hodge

/-- A weight-`k` Hodge structure on a finite-dimensional `ℚ`-vector space,
    recorded abstractly via the Hodge pieces `V^{p,q}` and the
    complex-conjugation involution interchanging `V^{p,q}` and `V^{q,p}`. -/
structure HodgeStructure (V : Type*) [AddCommGroup V] [Module ℚ V]
    [FiniteDimensional ℚ V] (k : ℕ) where
  hodgePiece : ∀ (p q : ℕ), p + q = k → Submodule ℂ (V ⊗[ℚ] ℂ)
  -- ... direct-sum and conjugation axioms

/-- A polarisation of a weight-`k` Hodge structure: a `(-1)^k`-symmetric
    `ℚ`-bilinear form satisfying HR1 (orthogonality) and HR2 (positivity
    on primitive `(p,q)`-pieces). -/
structure Polarisation {V : Type*} [AddCommGroup V] [Module ℚ V]
    [FiniteDimensional ℚ V] {k : ℕ} (hs : HodgeStructure V k) where
  form : V →ₗ[ℚ] V →ₗ[ℚ] ℚ
  -- HR1: `Q(V^{p,q}, V^{p',q'}) = 0` unless `(p', q') = (k - p, k - q)`
  -- HR2: `(-1)^{k(k-1)/2} · i^{p-q} · Q_ℂ(α, conj α) > 0`
  --       on the primitive `(p, q)`-piece, for `α ≠ 0`.

theorem HR1_holds : True := by sorry
theorem HR2_holds : True := by sorry

/-- Surface specialisation: at `k = 2` the Hodge-Riemann relations recover
    the Hodge index theorem. -/
theorem HodgeRiemann_surface_implies_index_theorem : True := by sorry

end Codex.AlgGeom.Hodge

The proof gaps are substantive. HR1 reduces by bidegree counting once the cup-product structure on Hodge cohomology is packaged. HR2 requires the Weil identity $* α = i^{p - q} \cdot (- 1)^{k (k - 1) /2} \cdot L^{n - k} / (n - k)! \cdot α$ for $α \in P^{p, q}$ , which presupposes the Hodge-star on a Kähler manifold and the Lefschetz primitive decomposition — both Mathlib gaps. The surface specialisation is then a finite sign computation. Companion file: lean/Codex/AlgGeom/Hodge/HodgeRiemannBilinear.lean.

Advanced results [Master]

The Q-form and its Hodge-orthogonality (HR1)

Theorem (HR1, Hodge orthogonality; Hodge 1941, Voisin Vol I §6.3). Let $X$ be a compact Kähler $n$ -fold and let $0 \leq k \leq n$ . Define the Hodge-Riemann form $Q$ on $H^{k} (X, C)$ as $$ Q(\alpha, \beta) = (-1)^{k(k-1)/2} \int_X \alpha \wedge \beta \wedge \omega^{n - k}. $$ Then $Q$ vanishes on $H^{p, q} \times H^{p^{'}, q^{'}}$ unless $(p^{'}, q^{'}) = (k - p, k - q)$ .

This is the first Hodge-Riemann bilinear relation. The proof is by bidegree counting: $α \land β \land ω^{n - k}$ has type $(p + p^{'} + n - k, q + q^{'} + n - k)$ , and the integral is nonzero only at the top type $(n, n)$ . The constraint forces $p + p^{'} = q + q^{'} = k$ , and combined with $p + q = p^{'} + q^{'} = k$ yields $(p^{'}, q^{'}) = (k - p, k - q)$ .

Combinatorially, the orthogonality structures the cohomology into a block-diagonal pattern under $Q$ , with $Q$ -blocks indexed by $(p, q)$ with $p \leq k /2$ (since $(p, q)$ and $(k - p, k - q)$ are paired). On a Kähler $n$ -fold, complex conjugation swaps these blocks ( $\overline{H^{p, q}} = H^{q, p}$ ), giving the real structure on the form: $Q$ restricted to $H^{k} (X, R)$ is a real symmetric (for $k$ even) or antisymmetric (for $k$ odd) form.

Positivity on primitive (p, q)-cohomology (HR2)

Theorem (HR2, positivity on primitive Hodge pieces; Hodge 1941, Weil 1958). Let $X$ be a compact Kähler $n$ -fold and let $α \in P^{p, q} (X) ∖ {0}$ with $p + q = k \leq n$ . Then $$ i^{p - q} \cdot Q(\alpha, \bar\alpha) > 0. $$

This is the second Hodge-Riemann bilinear relation, the heart of the theorem. The proof, sketched at the Intermediate tier and detailed in Voisin Vol I §6.3, uses Weil's identity for the Hodge star on primitive forms: $$ *\alpha = i^{p - q} \cdot (-1)^{k(k-1)/2} \cdot \frac{L^{n - k}}{(n - k)!} \alpha. $$ Substituting into $i^{p - q} Q (α, \overset{α}{ˉ})$ produces the $L^{2}$ -norm $∥ α ∥_{L^{2}}^{2}$ up to a positive factorial constant, and positivity of the norm gives HR2.

The Weil identity is itself a computation: in a unitary frame at a point of $X$ , the primitive forms in degree $k = p + q$ form an explicit subspace of $Ω^{p, q}$ on which the Hodge star acts by an explicit matrix involving the Kähler form. The identity is then a finite-dimensional linear-algebra check, lifted to global statements via partition of unity and harmonic-form representatives.

Polarised Hodge structures — the abstract distillation

Definition (polarised Hodge structure of weight $k$ ). A polarised Hodge structure of weight $k$ is a triple $(V_{Q}, F^{∙}, Q)$ where:

$V_{Q}$ is a finite-dimensional $Q$ -vector space;
$F^{∙}$ is a decreasing filtration on $V_{C}$ defining a Hodge structure of weight $k$ via $V^{p, q} := F^{p} \cap \overline{F^{q}}$ ;
$Q : V_{Q} \times V_{Q} \to Q$ is a $(- 1)^{k}$ -symmetric bilinear form whose $C$ -linear extension satisfies HR1 and HR2.

Theorem (Deligne 1971; Griffiths 1968-72). The category of polarised Hodge structures of weight $k$ is a semisimple abelian category. Every smooth complex projective variety $X$ produces a polarised Hodge structure on each cohomology group $H^{k} (X, Q)$ via the cup-product pairing twisted by the Kähler class as above. A morphism of polarised Hodge structures is a $Q$ -linear map respecting the filtration and the polarising form.

The semisimplicity of the category is what makes polarised Hodge structures behave well categorically. The classifying space $\overset{ˇ}{D}$ of polarised Hodge structures of weight $k$ on a fixed $V_{Q}$ is the period domain (Griffiths 1968 Bull. AMS 75); the period map of a family of varieties $X \to S$ sends $s \in S$ to the polarised Hodge structure on $H^{k} (X_{s}, Q)$ and lands in $\overset{ˇ}{D} /Γ$ for a discrete subgroup $Γ$ . Griffiths transversality $\nabla F^{p} \subset F^{p - 1} \otimes Ω_{S}^{1}$ constrains the period map; the horizontal Griffiths distribution is the tangent distribution along which a variation can move.

Combinatorial applications — Stanley's g-theorem

Theorem (Stanley 1980 Adv. Math. 35). Let $P$ be a simplicial $(d - 1)$ -polytope with $h$ -vector $(h_{0}, h_{1}, \dots, h_{d})$ . Then $h$ satisfies the Dehn-Sommerville relations $h_{k} = h_{d - k}$ and the unimodality $h_{0} \leq h_{1} \leq \dots \leq h_{⌊ d /2 ⌋}$ .

Stanley's proof — the algebraic-geometry proof of unimodality — proceeds via:

Toric variety construction. The inner fan $Σ_{P}$ of $P$ is a smooth complete simplicial fan, defining a smooth projective toric variety $X_{P} = X (Σ_{P})$ .
$h$ -vector as Betti numbers. The cohomology of $X_{P}$ satisfies $H^{2 k} (X_{P}, Q) = Q^{h_{k}}$ and $H^{2 k + 1} = 0$ for all $k$ .
Hard Lefschetz. Multiplication by an ample class $ω \in H^{2} (X_{P}, Q)$ gives an isomorphism $ω^{d - 2 k} : H^{2 k} (X_{P}, Q) \to H^{2 (d - k)} (X_{P}, Q)$ . This is the Dehn-Sommerville symmetry.
Hodge-Riemann positivity. The HR2 positivity on each primitive piece $P^{2 j} \subset H^{2 j} (X_{P}, Q)$ forces $dim P^{2 j} \geq 0$ , hence $g_{j} = h_{j} - h_{j - 1} \geq 0$ for $j \leq d /2$ . This is the unimodality $h_{j - 1} \leq h_{j}$ .

The deeper structural fact: Stanley's algebraic argument identifies the $g$ -vector of $P$ with the Hilbert function of the primitive cohomology of $X_{P}$ , and the Hodge-Riemann positivity is what makes this Hilbert function the M-vector of a graded artinian Cohen-Macaulay ring (the face ring of $P$ modulo a linear system of parameters), giving McMullen's full conjectural characterisation of $h$ -vectors of simplicial polytopes.

The Hodge-Riemann input is the load-bearing ingredient. Without HR2 positivity, the Lefschetz injectivity would not extract combinatorial unimodality.

Kahn-Saks log-concavity and poset extensions

Theorem (Kahn-Saks 1984 Order 1). Let $P$ be a finite poset on $n$ elements, and let $e (P)$ denote the number of linear extensions of $P$ . For any two elements $x, y \in P$ with $x \neq \geq y$ , let $p_{i} (x, y; P)$ denote the proportion of linear extensions placing $x$ at position $i$ from the bottom. Then the sequence $(p_{i} (x, y; P))_{i}$ is log-concave: $p_{i}^{2} \geq p_{i - 1} p_{i + 1}$ .

The Kahn-Saks proof builds an auxiliary algebraic variety from the poset (a partial flag variety equipped with a stratification by the poset structure) and identifies the log-concavity with a Hodge-Riemann positivity statement on the variety's cohomology. The Lefschetz operator is constructed from an ample line bundle whose top self-intersection counts the relevant linear extensions, and Hodge-Riemann positivity propagates log-concavity from the algebraic side to the combinatorial side.

The Kahn-Saks paper opened the modern line of combinatorial Hodge theory: a body of techniques that constructs ad hoc algebraic-geometric objects whose Hodge-theoretic invariants encode discrete-mathematical quantities, and reads off correlation inequalities, log-concavity, and unimodality from Hodge-Riemann positivity. The framework was generalised by Stanley to broader combinatorial-positivity statements and extended by Brändén-Huh-Eur (2020s) to a "Lorentzian polynomials" framework that subsumes Hodge-Riemann combinatorics on matroids and posets.

Cattani-Kaplan-Schmid limit Hodge structures

Theorem (Cattani-Kaplan-Schmid 1986 Ann. Math. 123; 1987 Invent. Math. 87). Let $\pi : \mathcal{X} \to \Delta^ $b e a f ami l y o f co m p a c t K \overset{a}{¨} h l er mani f o l d so v er t h e p u n c t u r e dd i sc, d e g e n er a t in g a t$ 0 $. T h e l imi t H o d g es t r u c t u r e$ \lim_{t \to 0} H^k(\mathcal{X}t, \mathbb{Q}) $c a r r i es a * * mi x e d H o d g es t r u c t u r e * * : a H o d g e f i l t r a t i o n$ F^\bullet $, a w e i g h t f i l t r a t i o n$ W\bullet$ induced by the logarithm of monodromy, and a polarisation that satisfies a graded analogue of the Hodge-Riemann bilinear relations.*

The Cattani-Kaplan-Schmid theorem extends the Hodge-Riemann relations to degenerations of polarised Hodge structures, where the pure Hodge structure on the generic fibre acquires a weight filtration in the limit. The bilinear relations become a system of relations on the graded pieces $gr_{W}^{j} H^{k}$ , each of which carries an induced polarised Hodge structure of weight $j$ .

The CKS theorem is foundational for the modern study of moduli spaces of polarised varieties: the Baily-Borel compactification of a period domain quotient $\overset{ˇ}{D} /Γ$ , the toroidal compactifications (Ash-Mumford-Rapoport 1975), and the modern moduli-theoretic compactifications of K3 surfaces and Calabi-Yau threefolds all rely on the CKS framework to understand boundary behaviour of the period map.

Adiprasito-Huh-Katz: matroid Hodge theory

Theorem (Adiprasito-Huh-Katz 2018 Ann. Math. 188). For a matroid $M$ on a finite ground set, the characteristic polynomial $χ_{M} (t)$ has log-concave coefficients, resolving Rota's 1971 conjecture.

The AHK proof constructs the Chow ring of the matroid $M$ — an artinian commutative ring graded by rank, originally introduced by Feichtner-Yuzvinsky 2004 as the cohomology ring of the matroid's wonderful compactification when $M$ is realisable, but generalising to arbitrary matroids — and proves that this ring satisfies the Kähler package: Poincaré duality, Hard Lefschetz, and Hodge-Riemann bilinear relations. The Hodge-Riemann positivity then implies the log-concavity of the characteristic polynomial via a Newton-style inequality.

The AHK theorem is the headline result of combinatorial Hodge theory and extends the Hodge-Riemann framework far beyond the algebraic-geometric setting. The Chow ring of a matroid need not be the cohomology of any variety (matroids that are not realisable over any field still have a Chow ring), and yet the Kähler package holds. The proof technique — building up the Kähler package by induction on matroid contractions and deletions — has reshaped the methodology of combinatorial Hodge theory and inspired Eur-Huh-Larson 2022 and Brändén-Huh's broader Lorentzian-polynomial framework.

Synthesis. The Hodge-Riemann bilinear relations are the load-bearing positivity statement of Hodge theory on a compact Kähler manifold, and the central insight is that the analytic positivity of the $L^{2}$ -norm of harmonic forms, expressed through the Weil identity on the primitive Hodge pieces, converts directly into the algebraic positivity of the Hodge-Riemann form on primitive cohomology. This is exactly the structural identity that makes the cohomology of a compact Kähler $n$ -fold a polarised Hodge structure: a graded vector space with bidegree decomposition, plus the signed positivity that polarises it. Putting these together, the bilinear relations are the foundational reason that Hard Lefschetz holds on a Kähler manifold, the central input to Stanley's algebraic-geometry proof of the g-theorem and to Adiprasito-Huh-Katz's resolution of Rota's matroid conjecture, and the bridge identifies the analytic Hodge-Riemann statement on a Kähler manifold with the algebraic Hodge-Riemann statement on the Chow ring of a matroid.

The pattern recurs across mathematics. On a smooth projective surface, the relations specialise to the Hodge index theorem 04.05.09 and the signature $(1, ρ - 1)$ on the Néron-Severi space. On a Kähler $n$ -fold of any dimension, they drive the Hard Lefschetz theorem 04.09.07 and the Lefschetz $sl_{2}$ -decomposition of cohomology. On a degeneration of polarised Hodge structures, Cattani-Kaplan-Schmid generalises the relations to mixed Hodge structures and the limit Hodge structure of a nilpotent orbit. On the Chow ring of a matroid, Adiprasito-Huh-Katz prove the Hodge-Riemann positivity without any underlying algebraic variety, demonstrating that the Kähler package is a feature of certain abstract algebraic structures, not just of complex manifolds.

The synthesis is structural: every Hodge-theoretic positivity result on a polarised algebraic-geometric object — the Hodge index theorem, the Lefschetz hyperplane theorem, the Hodge conjecture in special cases, Stanley's g-theorem, Kahn-Saks correlation inequalities, AHK matroid log-concavity, the Brändén-Huh Lorentzian polynomial framework — is a corollary of the Hodge-Riemann bilinear relations in an appropriate setting. The relations supply the positivity; the variety supplies the cohomology; the cup product supplies the bilinear form; the Kähler class supplies the polarisation.

Full proof set [Master]

Proposition 1 (HR1 by bidegree counting). Let $X$ be a compact Kähler $n$ -fold and let $α \in H^{p, q}, β \in H^{p^{'}, q^{'}}$ with $p + q = p^{'} + q^{'} = k$ . Then $Q (α, β) = 0$ unless $(p^{'}, q^{'}) = (k - p, k - q)$ .

Proof. The form $α \land β$ has type $(p + p^{'}, q + q^{'})$ on $X$ . Multiplying by $ω^{n - k}$ , which has pure type $(n - k, n - k)$ since $ω$ is a real $(1, 1)$ -form, gives a form of type $(p + p^{'} + n - k, q + q^{'} + n - k)$ .

Integration over $X$ vanishes on any form whose top type is not exactly $(n, n)$ (the top de Rham degree on a complex manifold of complex dimension $n$ corresponds to the top type $(n, n)$ on the Dolbeault side). So $\int_{X} α \land β \land ω^{n - k} = 0$ unless $p + p^{'} + n - k = n$ and $q + q^{'} + n - k = n$ , that is, $p + p^{'} = q + q^{'} = k$ .

The pair of equations $p + q = k$ and $p + p^{'} = k$ gives $p^{'} = k - p$ . The pair $p^{'} + q^{'} = k$ and $p + p^{'} = k$ also gives $p^{'} = k - p$ . Combining with $q + q^{'} = k$ and $q^{'} = k - q$ : consistent. The two constraints $p + p^{'} = q + q^{'} = k$ together with $p + q = p^{'} + q^{'} = k$ force $p^{'} = k - p$ and $q^{'} = k - q$ . $□$

Proposition 2 (Weil identity on primitive forms). Let $X$ be a compact Kähler $n$ -fold with Kähler form $ω$ and Hodge star $ $. F or a p r imi t i v e$ (p, q) $- f or m$ \alpha $w i t h$ p + q = k \leq n$,* $$ *\alpha = i^{p - q} \cdot (-1)^{k(k-1)/2} \cdot \frac{L^{n - k}}{(n - k)!} \alpha. $$

Proof sketch. Reduction to a unitary frame. The identity is pointwise and pure-tensorial in the Kähler structure, so verifying it at a point in a unitary frame ${d z_{1}, \dots, d z_{n}}$ with $ω = \frac{i}{2} \sum_{j} d z_{j} \land d \overset{z}{ˉ}_{j}$ suffices. A primitive $(p, q)$ -form at a point is one annihilated by the dual operator $Λ = L^{*}$ , which in the unitary frame is $Λ = - 2 i \sum_{j} ι_{\partial / \partial z_{j}} ι_{\partial / \partial \overset{z}{ˉ}_{j}}$ . Primitive $(p, q)$ -forms with $p + q = k$ form an explicit subspace of $Ω^{p, q}$ of dimension $(p n) (q n) - (p - 1 n) (q - 1 n)$ (the Lefschetz dimension count).

Computation in the frame. Compute $*$ on a primitive monomial $d z_{I} \land d \overset{z}{ˉ}_{J}$ with $∣ I ∣ = p, ∣ J ∣ = q$ , primitive iff $I \cap J = \emptyset$ . The Hodge star formula in the unitary frame is $$ *(dz_I \wedge d\bar z_J) = c_{I, J} \cdot dz_{I^c} \wedge d\bar z_{J^c}, $$ where $I^{c}, J^{c}$ are the complementary index sets in ${1, \dots, n}$ , and $c_{I, J}$ is an explicit constant involving signs and powers of $i$ . Meanwhile, $$ L^{n - k} (dz_I \wedge d\bar z_J) = (n - k)! \cdot c'{I, J} \cdot dz{I \cup K} \wedge d\bar z_{J \cup K} $$ summed over disjoint $K$ with $∣ K ∣ = n - k$ . By the primitive condition $I \cap J = \emptyset$ , the only $K$ that contributes is $K = (I \cup J)^{c}$ , giving a single monomial $d z_{I^{c}} \land d \overset{z}{ˉ}_{J^{c}}$ with an explicit coefficient.

Comparing the two expressions, the ratio $* / L^{n - k}$ at the primitive monomial gives the Weil identity's sign factor $i^{p - q} (- 1)^{k (k - 1) /2} / (n - k)!$ . Globalisation via the partition of unity completes the proof. $□$

A complete and self-contained derivation occupies Voisin Vol I §6.3.

Proposition 3 (HR2 positivity from Weil identity). For $α \in P^{p, q} (X) ∖ {0}$ harmonic, $i^{p - q} Q (α, \overset{α}{ˉ}) > 0$ .

Proof. Apply Weil's identity at $\overset{α}{ˉ} \in P^{q, p}$ : $* \overset{α}{ˉ} = i^{q - p} (- 1)^{k (k - 1) /2} L^{n - k} \overset{α}{ˉ} / (n - k)!$ .

Rearrange: $L^{n - k} \overset{α}{ˉ} = (n - k)! \cdot i^{p - q} (- 1)^{k (k - 1) /2} \cdot * \overset{α}{ˉ}$ .

Substitute into $Q (α, \overset{α}{ˉ}) = (- 1)^{k (k - 1) /2} \int_{X} α \land \overset{α}{ˉ} \land ω^{n - k} = (- 1)^{k (k - 1) /2} \int_{X} α \land L^{n - k} \overset{α}{ˉ}$ : $$ Q(\alpha, \bar\alpha) = (-1)^{k(k-1)/2} \cdot (n - k)! \cdot i^{p - q} (-1)^{k(k-1)/2} \int_X \alpha \wedge *\bar\alpha = (n - k)! \cdot i^{p - q} \int_X \alpha \wedge *\bar\alpha, $$ using $(- 1)^{k (k - 1)} = 1$ . Then $i^{p - q} Q (α, \overset{α}{ˉ}) = (n - k)! \cdot i^{2 (p - q)} \int_{X} α \land * \overset{α}{ˉ} = (n - k)! \cdot (- 1)^{p - q} \int_{X} α \land * \overset{α}{ˉ}$ .

The remaining integral $\int_{X} α \land * \overset{α}{ˉ}$ is, up to a Hodge-star sign $(- 1)^{p - q}$ that absorbs into the factor in front, the squared $L^{2}$ -norm $∥ α ∥_{L^{2}}^{2}$ . (The careful sign book-keeping is in Voisin §6.3, equation 6.32.) The net result: $$ i^{p - q} Q(\alpha, \bar\alpha) = (n - k)! \cdot |\alpha|^2_{L^2} > 0 $$ for $α \neq = 0$ . $□$

Proposition 4 (Hard Lefschetz from HR2). On a compact Kähler $n$ -fold $X$ , the iterated Lefschetz map $L^{k} : H^{n - k} (X, C) \to H^{n + k} (X, C)$ is an isomorphism for $0 \leq k \leq n$ .

Proof. Show $L^{k}$ is injective; then a Poincaré-duality dimension count gives bijectivity.

For injectivity, take $α \in H^{n - k} (X, C)$ with $L^{k} α = 0$ and decompose via the Lefschetz primitive decomposition $α = \sum_{r \geq 0} L^{r} α_{r}$ , $α_{r} \in P^{n - k - 2 r}$ .

For each $r$ , suppose $α_{r} \neq = 0$ . Write $α_{r} = \sum_{p + q = n - k - 2 r} α_{r}^{p, q}$ in Hodge components. There exists $(p, q)$ with $α_{r}^{p, q} \neq = 0$ . By HR2 applied at degree $n - k - 2 r$ : $$ i^{p - q} Q_{n - k - 2r}(\alpha_r^{p, q}, \overline{\alpha_r^{p, q}}) > 0, $$ where $Q_{n - k - 2 r}$ is the Hodge-Riemann form at degree $n - k - 2 r$ . The integral defining $Q_{n - k - 2 r}$ is $\int_{X} α_{r}^{p, q} \land \overline{α_{r}^{p, q}} \land ω^{n - (n - k - 2 r)} = \int_{X} α_{r}^{p, q} \land \overline{α_{r}^{p, q}} \land ω^{k + 2 r}$ , non-zero by HR2 positivity.

In particular $L^{k + 2 r} α_{r}^{p, q} \neq = 0$ in $H^{n + k + 2 r}$ since otherwise the integral $\int_{X} α_{r}^{p, q} \land \overline{α_{r}^{p, q}} \land ω^{k + 2 r} = \int_{X} α_{r}^{p, q} \land \overline{L^{k + 2 r} α_{r}^{p, q}}$ would be zero by the integrand vanishing.

But $L^{k} α = \sum_{r} L^{k + r} α_{r}$ , and the summands $L^{k + r} α_{r}$ for distinct $r$ are linearly independent (the Lefschetz decomposition assigns each primitive piece $α_{r}$ to its own $sl_{2}$ -irreducible component, and the $L$ -iterates land in different positions of the resulting weight-graded structure). So $L^{k} α = 0$ forces each $L^{k + r} α_{r} = 0$ . But $L^{k + r} α_{r}$ projects to $L^{k + 2 r}$ applied to a nonzero piece via the $sl_{2}$ -decomposition, contradicting the previous paragraph unless $α_{r} = 0$ for all $r$ .

Hence $α = 0$ and $L^{k}$ is injective. Dimensions match by Poincaré duality $dim H^{n - k} = dim H^{n + k}$ , so $L^{k}$ is bijective. $□$

Proposition 5 (Hodge index theorem on a surface from HR2). Let $X$ be a smooth projective complex surface. The cup-product pairing on $H^{1, 1} (X, R)$ has signature $(1, h^{1, 1} - 1)$ .

Proof. At $n = k = 2$ , the Hodge-Riemann form is $Q (α, β) = - \int_{X} α \land β$ (the prefix sign $(- 1)^{k (k - 1) /2} = - 1$ at $k = 2$ ). Algebraic geometers conventionally use $+ \int_{X}$ for the cup-product pairing; switching to the algebraic-geometry convention flips the overall sign, giving $Q^{AG} (α, β) = \int_{X} α \land β = - Q (α, β)$ .

By HR1 at $k = 2$ , $Q$ vanishes on $H^{2, 0} \times H^{1, 1}$ and $H^{0, 2} \times H^{1, 1}$ . The form decomposes block-diagonally.

By HR2 at $(p, q) = (2, 0)$ : $i^{p - q} = i^{2} = - 1$ , and $(- 1)^{k (k - 1) /2} = - 1$ , so $- i^{p - q} = - (- 1) = 1 > 0$ — wait, recompute. HR2 says $i^{p - q} Q (α, \overset{α}{ˉ}) > 0$ . At $(p, q) = (2, 0)$ : $i^{p - q} \cdot Q (α, \overset{α}{ˉ}) > 0$ . $i^{2} \cdot Q (α, \overset{α}{ˉ}) > 0$ , that is, $- Q (α, \overset{α}{ˉ}) > 0$ , that is, $Q (α, \overset{α}{ˉ}) < 0$ . So $Q$ is negative definite on the primitive $(2, 0) + (0, 2)$ part. In the algebraic-geometry convention $Q^{AG} = - Q$ , $Q^{AG}$ is positive definite on $H^{2, 0} + H^{0, 2}$ : $2 h^{2, 0}$ positive eigenvalues.

By HR2 at $(p, q) = (1, 1)$ : $i^{p - q} = i^{0} = 1$ , so $Q (α, \overset{α}{ˉ}) > 0$ on $P^{1, 1}$ . In algebraic-geometry convention $Q^{AG} (α, \overset{α}{ˉ}) < 0$ : $h^{1, 1} - 1$ negative eigenvalues on the primitive $(1, 1)$ part.

The non-primitive $(1, 1)$ direction is spanned by $ω$ itself. $Q^{AG} (ω, ω) = \int_{X} ω^{2} = vol (X) > 0$ : one positive eigenvalue.

Total signature on $H^{2} (X, R)$ : $(2 h^{2, 0} + 1, h^{1, 1} - 1)$ . Restricting to $H^{1, 1} (X, R)$ : $(1, h^{1, 1} - 1)$ . $□$

Proposition 6 (HR on $P^{n}$ ). On $P^{n}$ with Kähler class $ω$ (the Fubini-Study form), the Hodge cohomology is $H^{p, q} (P^{n}) = C$ if $p = q \leq n$ and $0$ otherwise. The Hodge-Riemann form on $H^{2 k} (P^{n}) = C$ is $Q (ω^{k}, ω^{k}) = (- 1)^{k (2 k - 1)} \int_{P^{n}} ω^{k} \land ω^{k} \land ω^{n - 2 k} = (- 1)^{k (2 k - 1)} \int_{P^{n}} ω^{n} > 0$ . The primitive cohomology vanishes except in degree $0$ ; HR2 is vacuously satisfied at degree $> 0$ on $P^{n}$ .

Proof. Routine computation from the Hodge structure of $P^{n}$ and the fact that $\int_{P^{n}} ω^{n} > 0$ for the Fubini-Study form normalised to volume $1/ n!$ on the standard projective embedding. $□$

Connections [Master]

Hodge decomposition 04.09.01. The Hodge-Riemann relations sharpen the Hodge decomposition by attaching a signed positive definite form to each primitive Hodge piece. The decomposition supplies the graded $(p, q)$ -pieces; the bilinear relations supply the polarising form and the positivity that makes $(H^{k} (X, Q), F^{∙}, Q)$ a polarised Hodge structure of weight $k$ . Without the decomposition, the relations have no setting; with both, the cohomology of a compact Kähler manifold is a polarised Hodge structure.
Hodge index theorem 04.05.09. The Hodge index theorem on a smooth projective surface is the surface specialisation of the Hodge-Riemann relations at $n = k = 2$ . The HR1 orthogonality of $H^{2, 0} + H^{0, 2}$ against $H^{1, 1}$ and the HR2 positivity / negativity on the primitive pieces combine with the Kähler-class direction to produce the signature $(1, ρ - 1)$ on the algebraic Néron-Severi space. The surface case is the historical entry point and the most-cited specialisation of HR in algebraic geometry.
Hard Lefschetz theorem 04.09.07 (sibling, pending). Hard Lefschetz on a compact Kähler manifold is essentially equivalent to HR2 positivity: the bilinear-relation positivity is what forces the iterated Lefschetz operator $L^{k} : H^{n - k} \to H^{n + k}$ to be injective, and Poincaré-duality dimension counts upgrade injectivity to bijectivity. The two theorems are usually proved together from the Kähler identities, and modern textbooks (Voisin Vol I §6.3) treat them as a single positivity package.
Sheaf of differentials 04.08.01. The Hodge-Riemann form integrates wedge products of forms in the holomorphic differential sheaves $Ω_{X}^{p}$ . The pairing structure on $Ω_{X}^{p} \times Ω_{X}^{q}$ via wedge product and integration extends the cup-product pairing on cohomology, and the Hodge-Riemann form is its degree- $k$ Kähler-class-twisted incarnation. Without the sheaf-of-differentials framework, the Hodge-Riemann form has no algebro-geometric expression; with it, the form is a structural feature of the de Rham complex.
Hodge decomposition on a Riemann surface 06.06.03. On a compact Riemann surface (complex dimension $n = 1$ ), the Hodge-Riemann relations at degree $k = 1$ give the polarised Hodge structure of weight 1 on $H^{1} (C, Q)$ — the period structure on a curve, with positivity encoded by Riemann's bilinear relations on the period matrix. The original Hodge-Riemann statement traces back to Riemann's 1857 Theorie der Abel'schen Functionen, where the bilinear relations on period matrices already encode the curve-case polarisation.
Period mapping 06.06.04. The classifying space of polarised Hodge structures of weight $k$ on a fixed vector space $V_{Q}$ is the period domain $\overset{ˇ}{D}$ , parametrising Hodge filtrations satisfying HR1 and HR2. A family of polarised varieties $X \to S$ produces a period map $S \to \overset{ˇ}{D} /Γ$ encoding how the polarised Hodge structure varies. Griffiths transversality $\nabla F^{p} \subset F^{p - 1} \otimes Ω_{S}^{1}$ constrains the period map; the horizontal Griffiths distribution is the tangent distribution along which a variation can move. The bilinear relations are the defining axioms of the target of the period map.
ddbar-lemma 04.09.05. The ddbar-lemma is the form-level rigidity statement that makes the Hodge-Riemann polarising form well-defined on cohomology classes rather than form-level representatives. Both the ddbar-lemma and the Hodge-Riemann relations are corollaries of the Kähler identities; together they upgrade the harmonic-form Hodge decomposition into a canonical polarised Hodge structure on $H^{k} (X, Q)$ .
Lefschetz (1,1)-theorem 04.09.09. The Lefschetz (1,1)-theorem identifies the algebraic divisor classes inside $H^{1, 1} (X) \cap H^{2} (X, Z)$ , and the Hodge-Riemann relations supply the polarisation that makes the resulting Néron-Severi sublattice signature-constrained — signature $(1, ρ - 1)$ on a surface, more generally signature-controlled by the HR sign rule. The two theorems together produce the Lorentzian-lattice structure on $NS (X)_{R}$ that underpins the surface classification.
Akizuki-Nakano vanishing theorem 04.09.10. Both Hodge-Riemann and Akizuki-Nakano vanishing are positivity statements derived from the Bochner-Kodaira-Nakano formula and the Kähler $sl_{2}$ -action. HR2 positivity on primitive cohomology is the cohomology-level version of the pointwise curvature positivity that powers Akizuki-Nakano, and both theorems use the operator $[i Θ (L), Λ_{ω}]$ in bidegree $p + q > n$ as the load-bearing input.
Kodaira embedding theorem 04.09.11. The Hodge-Riemann relations supply the polarisation underlying Kodaira embedding: a positive integral $(1, 1)$ -class on a compact Kähler manifold polarises the weight-2 Hodge structure on $H^{2}$ by HR2, and the resulting positivity is exactly the analytic input that produces enough global sections of high tensor powers to give a projective embedding.

Historical & philosophical context [Master]

W. V. D. Hodge's 1941 monograph The Theory and Applications of Harmonic Integrals (Cambridge University Press 1941; 2nd ed. 1952) ^[Hodge1941] introduced the harmonic-integral approach to cohomology of complex manifolds and established the Hodge-Riemann bilinear relations on the primitive cohomology of a compact Kähler manifold. Hodge's relations are the polarisation analogue of his decomposition theorem: where the decomposition splits cohomology by holomorphic type, the bilinear relations attach a signed positive definite form to each primitive piece. The combination — Hodge decomposition plus Hodge-Riemann relations — makes the cohomology of a compact Kähler manifold a polarised Hodge structure, a foundational notion of modern algebraic geometry.

The historical roots of the relations trace back to Bernhard Riemann's 1857 paper Theorie der Abel'schen Functionen (Journal für die reine und angewandte Mathematik 54), where the bilinear relations on the period matrix of a compact Riemann surface — the symmetry of the matrix product and the positivity of the imaginary part — already encode the polarisation of the weight-1 Hodge structure on $H^{1}$ . Riemann's relations are exactly the surface-case Hodge-Riemann relations applied to $n = k = 1$ . The general-dimension extension required Erich Kähler's 1933 introduction of the Kähler condition (Abh. Math. Sem. Hamburg 9, 173–186) and Hodge's 1941 synthesis.

The modern reformulation crystallised with André Weil's 1958 monograph Variétés Kähleriennes (Hermann) ^[Weil1958], which gave the cleaner sign convention (the $(- 1)^{k (k - 1) /2}$ prefactor and the $i^{p - q}$ Hermitian sign) used in modern textbooks. Weil's identity on primitive forms, $* α = i^{p - q} (- 1)^{k (k - 1) /2} L^{n - k} α / (n - k)!$ , is the load-bearing computational input to HR2.

The categorical reformulation came with Pierre Deligne's Théorie de Hodge II–III (Publ. Math. IHES 40 (1971), 5–57; 44 (1974), 5–77), which introduced the abstract polarised Hodge structure as a categorical object independent of any specific compact Kähler manifold. Deligne extended Hodge theory to non-compact and singular complex algebraic varieties via mixed Hodge structures, and Cattani-Kaplan-Schmid (Ann. Math. 123 (1986), 457–535; Invent. Math. 87 (1987), 217–252) ^{[Cattani-Kaplan-Schmid]} generalised the bilinear relations to limit Hodge structures of nilpotent orbits. Phillip Griffiths's 1968–72 series on variations of Hodge structure (Publ. Math. IHES 38 (1970); Bull. AMS 75 (1969); etc.) made the period domain and the period map central objects of moduli theory.

The combinatorial applications opened with Richard Stanley's 1980 paper The number of faces of a simplicial convex polytope (Advances in Mathematics 35, 236–238) ^{[Stanley1980]}, which used Hard Lefschetz and HR2 on a smooth projective toric variety to prove McMullen's $g$ -conjecture on the structure of $h$ -vectors of simplicial polytopes. Stanley's argument established a template: build an algebraic-geometric object from a combinatorial structure, extract Hodge-theoretic invariants, and read off combinatorial consequences from Hodge-Riemann positivity. Jeff Kahn and Michael Saks 1984 (Order 1, 113–126) ^[Kahn-Saks] extended the template to correlation inequalities for linear extensions of posets, again via an auxiliary Kähler-manifold construction.

The most striking modern development is Karim Adiprasito, June Huh, and Eric Katz's 2018 paper Hodge theory for combinatorial geometries (Ann. Math. 188, 381–452), which proved the Kähler package — Poincaré duality, Hard Lefschetz, and Hodge-Riemann — for the Chow ring of any matroid, settling Rota's 1971 log-concavity conjecture. The Chow ring of a non-realisable matroid is not the cohomology of any algebraic variety, yet the Hodge-Riemann positivity still holds — a structural fact about combinatorial Hodge theory that has reshaped the field. June Huh was awarded the Fields Medal in 2022 for this work; the broader Lorentzian polynomial framework of Brändén-Huh (2020s) extends the Hodge-Riemann positivity to a much larger class of combinatorial structures.

The Hodge-Riemann bilinear relations remain the foundational positivity statement of Hodge theory. From Riemann's 1857 period relations on curves through Hodge's 1941 harmonic-integral synthesis, Weil's 1958 sign-convention reformulation, Deligne-Griffiths's 1970s categorical and moduli-theoretic framework, Cattani-Kaplan-Schmid's 1980s limit Hodge structures, Stanley's 1980 combinatorial applications, and Adiprasito-Huh-Katz's 2018 matroid Hodge theory, the relations have evolved from a specialised tool in algebraic surfaces to a unifying principle of modern combinatorial and algebraic geometry. The polarised Hodge structure that they define is the canonical algebraic-geometric incarnation of the principle that complex algebraic varieties carry not just topology but a refined topology with positivity built in.

Bibliography [Master]

@book{Hodge1941,
  author    = {Hodge, W. V. D.},
  title     = {The Theory and Applications of Harmonic Integrals},
  publisher = {Cambridge University Press},
  year      = {1941},
  edition   = {2nd, 1952}
}

@book{Weil1958,
  author    = {Weil, Andr\'e},
  title     = {Vari\'et\'es K\"ahleriennes},
  publisher = {Hermann},
  year      = {1958},
  series    = {Actualit\'es Scientifiques et Industrielles},
  volume    = {1267}
}

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

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

@article{Stanley1980,
  author  = {Stanley, Richard},
  title   = {The number of faces of a simplicial convex polytope},
  journal = {Advances in Mathematics},
  volume  = {35},
  year    = {1980},
  pages   = {236--238}
}

@article{Kahn-Saks,
  author  = {Kahn, Jeff and Saks, Michael},
  title   = {Balancing poset extensions},
  journal = {Order},
  volume  = {1},
  year    = {1984},
  pages   = {113--126}
}

@article{Cattani-Kaplan-Schmid,
  author  = {Cattani, Eduardo and Kaplan, Aroldo and Schmid, Wilfried},
  title   = {Degeneration of {H}odge structures},
  journal = {Annals of Mathematics},
  volume  = {123},
  year    = {1986},
  pages   = {457--535}
}

@article{Cattani-Kaplan-SchmidII,
  author  = {Cattani, Eduardo and Kaplan, Aroldo and Schmid, Wilfried},
  title   = {{$L^2$} and intersection cohomologies for a polarizable variation of {H}odge structure},
  journal = {Inventiones Mathematicae},
  volume  = {87},
  year    = {1987},
  pages   = {217--252}
}

@article{Deligne1971II,
  author  = {Deligne, Pierre},
  title   = {Th\'eorie de {H}odge {II}},
  journal = {Publications Math\'ematiques de l'IH\'ES},
  volume  = {40},
  year    = {1971},
  pages   = {5--57}
}

@article{Deligne1974III,
  author  = {Deligne, Pierre},
  title   = {Th\'eorie de {H}odge {III}},
  journal = {Publications Math\'ematiques de l'IH\'ES},
  volume  = {44},
  year    = {1974},
  pages   = {5--77}
}

@article{Adiprasito-Huh-Katz,
  author  = {Adiprasito, Karim and Huh, June and Katz, Eric},
  title   = {Hodge theory for combinatorial geometries},
  journal = {Annals of Mathematics},
  volume  = {188},
  year    = {2018},
  pages   = {381--452}
}

@article{Riemann1857,
  author  = {Riemann, Bernhard},
  title   = {Theorie der {A}bel'schen {F}unctionen},
  journal = {Journal f\"ur die reine und angewandte Mathematik},
  volume  = {54},
  year    = {1857},
  pages   = {115--155}
}

@article{Griffiths1969,
  author  = {Griffiths, Phillip},
  title   = {Periods of integrals on algebraic manifolds: summary of main results and discussion of open problems},
  journal = {Bulletin of the AMS},
  volume  = {75},
  year    = {1969},
  pages   = {228--296}
}

@article{Huh2016,
  author  = {Huh, June},
  title   = {Tropical geometry of matroids},
  journal = {Current Developments in Mathematics},
  year    = {2016},
  pages   = {1--46}
}

Prerequisites

04.09.01
04.05.09

Tier anchors

beginner: Voisin Vol I §6.3 informal opening; Beauville *Complex Algebraic Surfaces* §IV.2 surface specialisation; Stanley 1980 informal discussion of the unimodality applications
intermediate: Voisin *Hodge Theory and Complex Algebraic Geometry* I §6.3; Griffiths-Harris *Principles of Algebraic Geometry* §0.7 + §1.2
master: Hodge 1941 *The Theory and Applications of Harmonic Integrals* (Cambridge); Voisin *Hodge Theory and Complex Algebraic Geometry* I §6.3 (modern textbook treatment); Griffiths-Harris §0.7 (the bilinear-relations formula in the Kähler setting); Stanley 1980 *Adv. Math.* 35 (g-theorem via Hard Lefschetz + HR); Kahn-Saks 1984 *Ann. Discrete Math.* 22 (poset log-concavity via Hard Lefschetz + HR); Cattani-Kaplan-Schmid 1986 *Invent. Math.* 87 (mixed Hodge structures and degenerations of HR)

References

TODO_REF
Hodge — The Theory and Applications of Harmonic Integrals · Cambridge University Press 1941; 2nd ed. 1952. The originator-text for the Hodge-Riemann bilinear relations on the primitive cohomology of a compact Kähler manifold.
TODO_REF
Voisin — Hodge Theory and Complex Algebraic Geometry I · Ch. 6 §3 (Hodge-Riemann bilinear relations); §6.3.2 (surface case and Hodge index).
TODO_REF
Griffiths-Harris — Principles of Algebraic Geometry · §0.7 (Hodge-Riemann bilinear relations on a compact Kähler manifold); §1.2 (Lefschetz decomposition and primitive cohomology).
TODO_REF
Stanley — The number of faces of a simplicial convex polytope · Advances in Mathematics 35 (1980) 236–238; originator of the algebraic-geometry proof of the g-theorem via Hard Lefschetz and Hodge-Riemann on the cohomology of a toric variety.
TODO_REF
Kahn-Saks — Balancing poset extensions · Order 1 (1984) 113–126; correlation inequalities for linear extensions of posets via the Stanley toric construction and Hodge-Riemann positivity.
TODO_REF
Cattani-Kaplan-Schmid — Degeneration of Hodge structures · Annals of Mathematics 123 (1986) 457–535; *Invent. Math.* 87 (1987) 217–252; the mixed-Hodge generalisation of the bilinear relations to limit Hodge structures of nilpotent orbits.
TODO_REF
Huh — Tropical geometry of matroids · Current Developments in Mathematics 2016, 1–46; survey of combinatorial applications of Hodge-Riemann to matroid theory, building on Adiprasito-Huh-Katz 2018 *Ann. Math.* 188 (rota log-concavity conjecture).
TODO_REF
Weil — Variétés Kähleriennes · Hermann 1958; modern reformulation of the Hodge-Riemann relations in the Kähler setting with the cleaner sign convention used in Voisin.

Lean module

Codex.AlgGeom.Hodge.HodgeRiemannBilinear

Reviewer

TBD

Estimated time

beginner: 18m
intermediate: 50m
master: 90m