The Hodge Conjecture

⚡ Key Takeaways

Statement: On a smooth complex projective variety, every rational Hodge class (a rational cohomology class of type $(p,p)$ ) is a rational linear combination of the cohomology classes of algebraic subvarieties. Result in words: the topological/Hodge-theoretic constraint of being type $(p,p)$ is sufficient to guarantee a class comes from algebraic geometry. Where it applies: smooth projective varieties over $\mathbb{C}$ (and conjecturally over arbitrary fields via algebraic cycles). Who/when: posed by W. V. D. Hodge in his 1950 ICM address; one of the seven Clay Millennium Problems (2000), prize $1{,}000{,}000.

💡 The Core Insight

Cohomology classes coming from algebraic subvarieties are forced — by the existence of a holomorphic structure compatible with integration over cycles — to be of pure Hodge type $(p,p)$ . This is a necessary condition, easily proved. The Hodge Conjecture asserts this necessary condition is also sufficient: nothing accidental lives in $(p,p)$ . In effect it claims that the Hodge structure on cohomology "knows" exactly which classes are realized by polynomials, and there is no purely transcendental obstruction beyond Hodge type.

DefinitionHodge Decomposition

Let $X$ be a compact Kähler manifold (e.g. a smooth complex projective variety) of complex dimension $n$ . For each integer $k$ with $0 \le k \le 2n$ , the complex singular cohomology decomposes as

1

\quad H^k(X,\mathbb{C}) \;=\; \bigoplus_{p+q=k} H^{p,q}(X), \qquad \overline{H^{p,q}(X)} = H^{q,p}(X),

Derivation hint: From Hodge theory each de Rham class has a unique harmonic representative; the Laplacian commutes with the $(p,q)$ -decomposition of forms on a Kähler manifold via the Kähler identities $\Delta_d = 2\Delta_{\bar\partial}$ , splitting harmonic $k$ -forms by type. where $H^{p,q}(X) = H^q(X,\Omega_X^p)$ is the Dolbeault group of classes representable by harmonic forms with $p$ holomorphic and $q$ antiholomorphic differentials. The integers $h^{p,q} = \dim_\mathbb{C} H^{p,q}$ are the Hodge numbers.

DefinitionHodge Classes

Fix $0 \le p \le n$ . The space of rational Hodge classes of degree $2p$ is

2

\quad \mathrm{Hdg}^p(X) \;=\; H^{2p}(X,\mathbb{Q}) \,\cap\, H^{p,p}(X) \;=\; \big\{\, \alpha \in H^{2p}(X,\mathbb{Q}) : \alpha \otimes \mathbb{C} \in H^{p,p}(X)\,\big\}.

Derivation hint: Embed $H^{2p}(X,\mathbb{Q})\hookrightarrow H^{2p}(X,\mathbb{C})=\bigoplus_{a+b=2p}H^{a,b}$ and intersect the rational lattice with the single summand $H^{p,p}$ ; rationality plus type $(p,p)$ are the defining conditions. A class is a Hodge class iff it is rational and unchanged under the Hodge symmetry, i.e. lies in the "middle" diagonal piece.

DefinitionThe Cycle Class Map

Let $Z^p(X)$ be the free abelian group on irreducible closed subvarieties of codimension $p$ . There is a cycle class map sending a codimension- $p$ subvariety $Z$ to its fundamental cohomology class:

3

\quad \mathrm{cl}\colon Z^p(X)\otimes \mathbb{Q} \;\longrightarrow\; H^{2p}(X,\mathbb{Q}), \qquad [Z] \longmapsto \mathrm{cl}([Z]),

Derivation hint: A smooth subvariety $Z$ carries an oriented real $2(n-p)$ -dimensional homology class; Poincaré duality turns it into a degree- $2p$ cohomology class. For singular $Z$ resolve or use Borel–Moore homology / currents (integration over $Z$ defines a closed current of bidegree $(p,p)$ ). The image consists of algebraic classes, and by construction lands inside $\mathrm{Hdg}^p(X)$ .

TheoremThe Hodge Conjecture

Let $X$ be a smooth projective variety over $\mathbb{C}$ . Then the cycle class map is surjective onto Hodge classes after tensoring with $\mathbb{Q}$ :

4

\quad \mathrm{cl}_\mathbb{Q}\colon Z^p(X)\otimes \mathbb{Q} \;\twoheadrightarrow\; \mathrm{Hdg}^p(X) \;=\; H^{2p}(X,\mathbb{Q})\cap H^{p,p}(X).

Derivation hint: This is the conjecture itself, not a theorem; equivalently every rational $(p,p)$ class equals $\sum_i a_i\,\mathrm{cl}([Z_i])$ with $a_i\in\mathbb{Q}$ and $Z_i$ irreducible codimension- $p$ subvarieties.

⚠️ Common mistake

The original 1950 formulation used integral coefficients ("every integral $(p,p)$ class is algebraic"). Atiyah and Hirzebruch disproved this in 1962 using torsion phenomena detected by topological K-theory. The correct, surviving statement is the rational one above. Even the "rational modulo torsion" version requires care: there exist non-torsion integral Hodge classes that are not algebraic (Kollár's examples), but their rational multiples can still be algebraic, consistent with the rational conjecture.

ProofAlgebraic classes are of type

(p,p)

We prove the easy half: $\mathrm{im}(\mathrm{cl}_\mathbb{Q}) \subseteq \mathrm{Hdg}^p(X)$ .

Step 1. Let $Z\subset X$ be an irreducible subvariety of codimension $p$ . Its class $\mathrm{cl}([Z])\in H^{2p}(X,\mathbb{Q})$ is by definition rational; we must show it lies in $H^{p,p}$ .

Step 2. Represent $\mathrm{cl}([Z])$ as a current: for any smooth closed $2(n-p)$ -form $\omega$ ,

5

\quad \int_X \mathrm{cl}([Z])\wedge \omega \;=\; \int_{Z_{\mathrm{reg}}} \omega,

Derivation hint: The integration current $[Z]$ pairs a test form with its restriction to the smooth locus $Z_{\mathrm{reg}}$ ; this is well-defined since $Z_{\mathrm{sing}}$ has real codimension $\ge 2$ in $Z$ and contributes nothing.

Step 3. Decompose $\omega = \sum_{a+b=2(n-p)} \omega^{a,b}$ into types. On the complex $(n-p)$ -dimensional manifold $Z_{\mathrm{reg}}$ , a form $\omega^{a,b}$ restricts to zero unless $a \le n-p$ and $b \le n-p$ ; combined with $a+b = 2(n-p)$ this forces $a=b=n-p$ . So only the $(n-p,n-p)$ part survives integration.

Step 4. Hence the current $[Z]$ annihilates all types except its dual type. By the Hodge decomposition of currents, $\mathrm{cl}([Z])$ pairs nontrivially only with $H^{n-p,n-p}$ , forcing $\mathrm{cl}([Z])\in H^{p,p}(X)$ .

Step 5. Being both rational (Step 1) and of type $(p,p)$ (Step 4), $\mathrm{cl}([Z])\in \mathrm{Hdg}^p(X)$ . By $\mathbb{Q}$ -linearity the entire image lies in $\mathrm{Hdg}^p(X)$ . $\blacksquare$ The converse — that every element of $\mathrm{Hdg}^p(X)$ arises this way — is the unproven Hodge Conjecture.

∎

Figure

The Hodge diamond and the diagonal where Hodge classes live

mathematician · William Vallance Douglas Hodge | 1903–1975

A Scottish geometer, Hodge developed the theory of harmonic integrals in the 1930s, proving that the cohomology of a compact Riemannian manifold is represented by harmonic forms, and that on a Kähler manifold these split by type. His 1941 book The Theory and Applications of Harmonic Integrals founded modern Hodge theory. In his 1950 Cambridge ICM address he posed the conjecture bearing his name, seeking to characterize algebraic cycles cohomologically. Hodge held the Lowndean Chair at Cambridge, was knighted in 1959, and received the Royal Society's Copley Medal. His insistence that analytic data (harmonic forms) controls algebraic geometry remains a guiding theme of the field.

🧠 Intuition

Think of cohomology as a way of counting "holes" of a space, refined into directions by how holomorphic they are. Algebraic subvarieties — the zero loci of polynomials — are rigid, complex-analytic objects; when you integrate a differential form over one, only the perfectly balanced $(p,p)$ part can survive, because the subvariety has equal "holomorphic" and "antiholomorphic" dimensions. So algebraicity forces balance. The conjecture is the bold reverse claim: if a rational class is perfectly balanced, then it must secretly be supported on polynomial loci. The mental picture is a sieve: the Hodge decomposition sieves cohomology by type, and the conjecture says the sieve's $(p,p)$ -slot catches exactly the algebraic classes — no transcendental impostors slip through. The shock is that this should be true at all, since cohomology is a soft topological gadget while algebraic cycles are hard geometric ones; the conjecture asserts a hidden equivalence between flexibility and rigidity.

ExampleDivisors on a surface — the Lefschetz

(1,1)

theorem

Given: $X$ a smooth projective surface ( $n=2$ ), and the case $p=1$ . We want to identify $\mathrm{Hdg}^1(X)=H^2(X,\mathbb{Q})\cap H^{1,1}(X)$ .

Steps. 1. The exponential sequence $0\to\mathbb{Z}\to\mathcal{O}_X\xrightarrow{\exp}\mathcal{O}_X^*\to 0$ gives a connecting map $H^1(X,\mathcal{O}_X^*)=\mathrm{Pic}(X)\to H^2(X,\mathbb{Z})$ , the first Chern class $c_1$ . 2. The image of $c_1$ lands in the kernel of $H^2(X,\mathbb{Z})\to H^2(X,\mathcal{O}_X)=H^{0,2}$ , i.e. exactly the integral $(1,1)$ classes. 3. Lefschetz's theorem (1924): every integral class in $H^2(X,\mathbb{Z})\cap H^{1,1}$ is $c_1$ of a line bundle, hence the class of a divisor (a difference of curves). 4. Tensoring with $\mathbb{Q}$ , every rational $(1,1)$ class is a $\mathbb{Q}$ -combination of curve classes — codimension $p=1$ subvarieties.

Answer: $\boxed{\mathrm{Hdg}^1(X)=\mathrm{im}(\mathrm{cl}_\mathbb{Q})\ \text{for}\ p=1,\ \text{any smooth projective }X.}$ This is the only general case fully proven; it holds for all dimensions $n$ , not just surfaces.

ExampleAbelian fourfold — a numerical Hodge class count

Given: $A=V/\Lambda$ a complex abelian variety of dimension $g=4$ , with $H^k(A,\mathbb{Q})=\wedge^k H^1(A,\mathbb{Q})$ and $\dim H^1=2g=8$ . Consider middle cohomology $k=4$ , codimension $p=2$ .

Steps. 1. Total dimension: $\dim_\mathbb{Q} H^4(A,\mathbb{Q})=\binom{8}{4}=70$ . 2. Hodge numbers: $h^{p,q}=\binom{g}{p}\binom{g}{q}=\binom{4}{p}\binom{4}{q}$ . For $p+q=4$ : $h^{4,0}=\binom{4}{4}\binom{4}{0}=1$ , $h^{3,1}=\binom{4}{3}\binom{4}{1}=16$ , $h^{2,2}=\binom{4}{2}\binom{4}{2}=36$ , $h^{1,3}=16$ , $h^{0,4}=1$ . Check: $1+16+36+16+1=70$ . ✓ 3. The space $H^{2,2}$ has complex dimension $36$ ; Hodge classes are the rational classes inside it, $\mathrm{Hdg}^2(A)\subseteq H^{2,2}$ . 4. For a generic abelian fourfold the only Hodge classes come from products of divisor classes (intersections of polarizations), forming a small subspace. But special abelian fourfolds (e.g. of Weil type) can carry "exotic" Hodge classes of dimension up to $4$ not generated by divisors.

Answer: $\boxed{h^{2,2}=36,\ \text{and Weil-type } A \text{ may have a } 4\text{-dimensional space of exotic Hodge classes whose algebraicity is unknown in general.}}$ These exotic classes are the cutting edge: they are the simplest cases where the Hodge Conjecture remains open.

⚠️ Common mistake

"Algebraic cycles inject into cohomology, so $\mathrm{cl}_\mathbb{Q}$ being surjective means it's an isomorphism." False. The cycle class map has an enormous kernel: cycles homologically equivalent to zero (e.g. those in the kernel of the Abel–Jacobi map) map to $0$ . The conjecture is purely about surjectivity onto Hodge classes, not injectivity. The kernel is studied by Chow groups, Bloch–Beilinson filtrations, and is itself vast and mysterious.

CounterexampleThe integral Hodge Conjecture fails

Atiyah and Hirzebruch (1962) constructed smooth projective varieties carrying torsion classes $\alpha\in H^{2p}(X,\mathbb{Z})\cap H^{p,p}$ that are not algebraic — they are obstructed by nontrivial differentials in the Atiyah–Hirzebruch spectral sequence relating singular cohomology to topological K-theory, an obstruction invisible rationally. Concretely, a Steenrod-square type operation kills algebraicity. Kollár (1992) gave non-torsion counterexamples: certain hypersurfaces where a generator of $H^4(X,\mathbb{Z})$ is not algebraic, but a multiple of it is. Moral: the conjecture survives only after $\otimes\,\mathbb{Q}$ . The denominators are essential.

⚠️ Common mistake

"It is enough to prove it for the middle-dimensional cohomology." While the hard Lefschetz theorem and the Lefschetz hyperplane theorem reduce some cases, the conjecture for all $p$ on all smooth projective varieties does not formally reduce to a single degree. However, by a result using the standard conjectures' machinery and the hard Lefschetz isomorphism, the Hodge Conjecture for varieties of dimension $\le 2n$ in degree $\le n$ would have consequences in higher degrees — but no clean full reduction exists unconditionally.

connection

The Hodge Conjecture sits at the crossroads of several deep theories. Via Hodge Theory it is a statement about the $(p,p)$ pieces of the Hodge decomposition. Through Algebraic Cycles and Chow Groups it concerns the image of the cycle class map. Its arithmetic analogue is the Tate Conjecture, replacing Hodge structures with Galois representations on étale cohomology, and the two are expected to be deeply parallel. Grothendieck's Standard Conjectures on Algebraic Cycles would imply key consequences (e.g. that the Künneth components of the diagonal are algebraic). It is the motivating problem for the theory of Motives: Hodge classes should correspond to morphisms in a category of pure motives. The conjecture also governs the Mumford–Tate Group, whose invariants are exactly the Hodge classes.

🔬 Where this appears

⏳ History

- 1924 — Solomon Lefschetz proves the

(1,1)

theorem: integral

(1,1)

classes on a smooth projective variety are classes of divisors, the first and still principal case.

- 1930s–1941 — W. V. D. Hodge develops harmonic integral theory, proving the Hodge decomposition on Kähler manifolds; his book appears in 1941.

- 1950 — Hodge poses the conjecture in his ICM plenary address in Cambridge, in integral form.

- 1962 — Michael Atiyah and Friedrich Hirzebruch disprove the integral Hodge Conjecture using topological K-theory and torsion obstructions; the rational version becomes the standard formulation.

- 1969 — Alexander Grothendieck publishes "Hodge's general conjecture is false for trivial reasons," refining the generalized version (Hodge–Grothendieck conjecture) and stressing coniveau filtrations.

- 1977 — Steven Zucker and others prove cases for varieties with large automorphism or special structure; partial results accumulate for abelian varieties.

- 1982 — Pierre Deligne proves Hodge classes on abelian varieties are absolutely Hodge, a deep partial result strongly supporting the conjecture in that setting.

- 1992 — János Kollár constructs non-torsion counterexamples to the integral conjecture, sharpening the necessity of rational coefficients.

- 2000 — The Clay Mathematics Institute names the Hodge Conjecture one of the seven Millennium Prize Problems, with a

1{,}000{,}000

award (official statement by Pierre Deligne).

open-problem

The conjecture is open in essentially all nontrivial cases beyond $p=1$ . Key frontiers and difficulties:

Abelian varieties of Weil type provide the simplest Hodge classes not obviously algebraic; these remain unproven in general and are the natural testing ground.
The generalized Hodge Conjecture (Hodge–Grothendieck) concerns the largest sub-Hodge structure of given coniveau supported on a subvariety; Grothendieck corrected Hodge's first attempt, and even this refined version is wide open.
Relation to the Tate Conjecture: over number fields, the Hodge and Tate conjectures together would imply a tight correspondence; neither is known, though Deligne's absolute Hodge result links them conditionally.
No known general construction produces the required cycles: even when a Hodge class is known to exist, exhibiting the algebraic cycle realizing it is typically impossible with current methods. The conjecture is "existential" with no algorithmic content.
Variational Hodge Conjecture: if a Hodge class is algebraic at one point of a family, is it algebraic at nearby points where it stays Hodge? This local-to-deformation version is also open and is implied by (but possibly weaker than) the full conjecture.

Formal Statement

Why the Necessary Direction Holds

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World (Mathematical) Applications

Historical Development

Subtleties & Limits

See Also