Recursos
Resumen
You are given four CM sign patterns and one tensor line built from them. Can you tell whether that line is a Hodge class? Read its sign table column by column. That check is the first step of a claimed proof of the rational Hodge conjecture for CM abelian varieties.
Why divisors are not enough
By Lefschetz (1,1), every degree-2 Hodge class is a divisor class, and cup products of divisors stay algebraic. The cohomology of an abelian variety is Λ*H¹, but that still does not make every Hodge class a product of divisors. Weil classes fall outside the divisor algebra, and this already happens on abelian fourfolds. Through two theorems of Milne, algebraicity of CM Hodge classes leads to the Tate conjecture for abelian varieties over finite fields and to the Hodge standard conjecture. The paper narrows the problem step by step. Deligne and André reduce CM Hodge classes to split Weil classes. Balanced tensors are then built from four-label switches. What is left is one line, U(v₁)₁ ⊗ U(v₂)₁ ⊗ U(v₃)_c ⊗ U(v₄)_c.
The column test [1:26]
An odd sign function v satisfies v(cτ) = −v(τ). It makes each embedding line of type (1,0) (sign +) or (0,1) (sign −). Conjugating by g moves the labels to g, g, gc, gc. So in column g the first two signs stay the same and the last two flip. The line has type (2,2) exactly when the four column signs sum to zero. A rational Hodge class needs this at every Galois conjugate, so the full condition is v₁ + v₂ = v₃ + v₄.
- Worked case [1:44]: E = ℚ(ζ₇), c = σ³. With v₁ = (+,+,+), v₂ = (+,−,−), v₃ = (+,+,−), v₄ = (+,−,+), every column gives (2,2) [2:11]. The line is a Hodge class.
- Think-first [2:56]: change v₄ to (+,+,−). The identity column still gives (2,2), but column σ gives (1,3). One good embedding proves nothing.
How the paper makes the line algebraic
- Surface criterion [3:23]: suppose there are Hodge maps into H¹ of a surface S with a nonzero period ∫ h₁(α₁)∪…∪h₄(α₄). Then the image of S is a cycle that pairs nonzero with the line. Four divisor kernels of type (1,1), algebraic by Lefschetz, turn that functional into a vector on the line.
- Theta one-forms [4:09] on a compact quotient of the complex 2-ball (signature (2,1) at one embedding) supply the nonzero period.
- Frobenius slopes [4:47]: a Hecke–Frobenius relation forces slope 0 or 1. Weak admissibility turns these into the pure types (0,1) and (1,0), which are exactly the prescribed CM signs.