A projective algebraic variety over is the solution set in projective space of a system of homogeneous polynomial equations. When it is non-singular it is also a compact complex manifold, so it has a topology and a de Rham cohomology, and the complex structure splits that cohomology by type. The Hodge conjecture asks how much of this topology is itself algebraic: which cohomology classes are the classes of algebraic subvarieties. It is one of the seven Millennium Prize Problems of the Clay Mathematics Institute; the official problem description is Pierre Deligne's The Hodge Conjecture (2000), whose §1 is the source for every definition below.
Timeline.
Fix and write for the homogeneous coordinates (Coord N) and for projective space. All differential geometry is done inside the real Euclidean space of complex matrices (Amb N), into which is embedded by
the orthogonal projection onto the line (proj). The complex structure of at the point is on tangent matrices (Jmap), and a complex scalar acts on tangent vectors by (cmul).
A projective variety (ProjVariety N) is given by a set of homogeneous polynomials; its cone is the set of nonzero common zeros , and its points are . The Zariski tangent space of the cone at is the common kernel of the differentials of all polynomials vanishing on (tangentCone). is non-singular of dimension (IsNonsingular n) when this space has dimension at every point of the cone; this is the Jacobian criterion and forces to be a compact complex manifold of pure dimension (possibly disconnected). The real tangent space is the image of the Zariski tangent space under (tangent). A variety is irreducible when its cone is not covered by two algebraic sets neither of which contains it (IsIrreducible), and its dimension is the minimum, over the cone, of the Zariski tangent dimension, minus one (dim), which for irreducible varieties is the dimension at the generic point; the smooth locus regPts is where that minimum is attained.
A -form on is a smooth -valued alternating -form on (Form N k), considered only through its values on tangent vectors of . Following Deligne (p. 1), (Cohomology X k) is the space of closed -forms modulo exact forms: a form is closed when vanishes on , exact when it agrees on with for a smooth -form . A form has type (HasType p q) when, on tangent vectors of ,
which is Deligne's definition through the -action on (his forms transform by under the contragredient action). is the set of classes with a closed representative (HasTypeClass).
The rational structure is defined by periods: smooth singular -simplices in , integral chains, the standard boundary , cycles, and the period obtained by pulling back to the standard simplex (Chain, boundary, IsCycle, period). A class is rational (IsRational) when its periods over all smooth integral cycles in are rational numbers. By de Rham's theorem this is exactly the image of . A Hodge class (IsHodgeClass p) is a rational class in of type (Deligne, p. 2).
Cycle classes. Integrals over and over a subvariety use the Hausdorff measure of the relevant dimension on , normalised by so that it is the Riemannian volume (volMeasure), and evaluate the integrand on the frame built from a complex orthonormal frame of the tangent space (frame), i.e. the orientation induced by the complex structure. The pairing (pairing) uses the standard shuffle formula for the wedge product (wedgeVal); (integralOver) integrates over the smooth locus of . Following Deligne (pp. 1–2), a class is the class of , (IsCycleClass X Z n p c), when it is Poincaré dual to integration over :
"On a projective non-singular algebraic variety over , any Hodge class is a rational linear combination of classes of algebraic cycles." In the form stated in Lean (HodgeConjecture.hodge_conjecture): for non-singular of dimension , , and every Hodge class ,
hodge_decomposition, Deligne (1), p. 1): every class in is uniquely a sum with .existsUnique_cycleClass, p. 1–2): for an irreducible of codimension there is exactly one class Poincaré dual to integration over .isHodgeClass_of_isCycleClass, p. 2): "the class ... is hence of type ", and it is rational since is an integral cycle.hodge_conjecture_divisors, Remark (iii), p. 2): the Hodge conjecture for , proved by Kodaira and Spencer; Hodge classes in are rational combinations of divisor classes.The result itself. The conjecture is the bridge between the transcendental invariants of a variety (its Hodge structure) and its algebraic cycles. It underlies the theory of motives: Deligne (§5) notes that, granting the algebraicity of the Hodge classes in his Examples 1 and 2, the full conjecture is equivalent to a natural functor from motives to Hodge structures being fully faithful. Consequences include the algebraicity of the Künneth components of the diagonal and the inverse of the hard Lefschetz isomorphism (§4).
Formalizing it. Nothing in this circle of ideas exists in Mathlib: there is no de Rham or singular cohomology of manifolds, no Hodge theory, no cycle class map. The definitions of this mission build the objects from scratch on Mathlib's calculus (ContinuousAlternatingMap, extDeriv), multivariate polynomials, and Hausdorff measure, in a form that is elementary enough to audit. The milestones are all classical theorems whose formalization requires developing harmonic theory or its substitutes (for 1), Poincaré duality and Lelong's theorem on the integration current (for 2 and 3), and the exponential sequence with Kodaira–Spencer's argument or an equivalent (for 4). None has a machine-checked proof anywhere. The goal fixes in Lean the exact statement the Clay prize asks for; two earlier entries on this platform under the name hodge_conjecture stated an equality of two constant one-point types and have been retired.
Every known approach to the conjecture has to construct algebraic cycles, and there is no general method for doing so. The tools that prove the known cases do not extend: the Lefschetz argument uses the exponential sequence , which has no analogue for ; the Lefschetz pencil induction of Griffiths and Zucker requires not only the conjecture for hyperplane sections but that all of the intermediate Jacobian come from algebraic cycles, which fails in general (Deligne, §3). The natural first idea, to show a Hodge class is algebraic by deforming it, only yields Deligne's result with Cattani and Kaplan that the locus of Hodge classes in a family is a countable union of algebraic subvarieties, not algebraicity of the classes.
EuclideanSpace ℂ (Fin (N+1) × Fin (N+1)), a real inner product space whose Euclidean metric is the one used for the Hausdorff measure and for orthonormal frames. Varieties may be disconnected; non-singularity forces pure dimension.∞ is , not analytic. Derivatives are Mathlib's total functions; the smoothness hypotheses make them honest wherever they are used.μH[d] is un-normalised; volMeasure N d multiplies it by ( for ), which is the constant making Hausdorff measure equal to Lebesgue measure on -dimensional Euclidean subspaces. Without this factor the identity would be off by a power of and the conjecture would change.Classical.epsilon; on the smooth locus such frames exist and the value is independent of the choice.IsCycleClass is a predicate, so if no class were Poincaré dual to the goal would be false, not vacuous; milestone 2 asserts the witness exists and is unique. Rationality quantifies over all smooth cycles in , so it cannot be met by a degenerate choice of representative.namespace HodgeConjecture
theorem hodge_conjecture (N n p : ℕ) (hp : p ≤ n) (X : ProjVariety N) (hX : X.IsNonsingular n)
(c : X.Cohomology (2 * p)) (hc : X.IsHodgeClass p c) :
∃ (m : ℕ) (Z : Fin m → ProjVariety N) (a : Fin m → ℚ) (cl : Fin m → X.Cohomology (2 * p)),
(∀ i, (Z i).cone ⊆ X.cone ∧ (Z i).IsIrreducible ∧ (Z i).dim = n - p ∧
X.IsCycleClass (Z i) n p (cl i)) ∧
c = ∑ i, ((a i : ℚ) : ℂ) • cl i := by sorry
end HodgeConjectureThis is the Hodge conjecture exactly as stated in Deligne's Clay problem description (p. 2): "On a projective non-singular algebraic variety over , any Hodge class is a rational linear combination of classes of algebraic cycles."
Let be a projective variety that is non-singular of dimension , let , and let be a Hodge class: a class with rational periods on all smooth integral cycles (so ) that is represented by a closed form of type . Then there exist finitely many irreducible algebraic subvarieties of dimension (codimension ), rational numbers , and classes , each Poincaré dual to integration over the corresponding , such that
The conjecture asserts that the transcendental condition "rational of type " characterises the classes coming from algebraic geometry. It is known for , and (Lefschetz , Kodaira–Spencer), hence for , and is open in general, even for the diagonal of . A proof would settle the Millennium Prize Problem.
Formalization Note is given by homogeneous polynomial equations and realised in the space of matrices via ; non-singularity is the Jacobian criterion (Zariski tangent spaces of the cone of dimension ), which allows to be disconnected but forces pure dimension . Cohomology is de Rham cohomology of ambient forms restricted to ; the rational structure is defined by periods over smooth cycles; is characterised by for all closed -forms , with integrals against the correctly normalised Hausdorff measure and the orientation induced by the complex structure. The family may be empty (), which only produces .
No submissions yet on this mission's goal. Log in and be the first to prove it.