The Hodge Conjecture states that every Hodge class of a non singular projective variety over $\mathbf{C}$ is a rational linear combination of cohomology classes of algebraic cycles: Even though I'm able to understand what it says, and at first glance I do find it a very nice assertion, I cannot grasp yet why it is so relevant as to be considered one of the biggest open problems in algebraic geometry. Which are its implications?

Going a bit further, what about the Tate conjecture?


2 Answers 2


Let $K$ be one of the following fields: the complex numbers, a finite field, a number field (and we could amalgamate the last two into the more general case of a field finitely generated overs its prime subfield).

In each case we can consider the category of smooth projective varieties over $K$, with morphisms being correspondences [added: modulo the relation of cohomological equivalence; see David Speyer's comment below for an elaboration on this point]. (I am really thinking of the category of pure motives, but there is no particular need to invoke that word.)

In each case we also have a natural abelian (in fact Tannakian) category in play: in the complex case, the category of pure Hodge structures, and in the other cases, the category of $\ell$-adic representations of $G_K$ (the absolute Galois group of $K$) (for some prime $\ell$, prime to the characteristic of $K$ in the case when $K$ is a finite field).

Now taking cohomology gives a functor from the category of smooth projective varieties to this latter category (via Hodge theory in the complex case, and the theory of etale cohomology in the other cases). The Hodge conjecture (in the complex case) and the Tate conjecture (in the other cases) then says that this functor is fully faithful.

The consequence (if the conjecture is true) is that we can construct correspondences between varieties simply by making computations in some much more linear category.

Let me give an example of how this can be applied in number theory:

Fix two primes $p$ and $q$, and let $\mathcal O_D$ be a maximal order in the quaternion algebra $D$ over $\mathbb Q$ ramified at $p$ and $q$, and split everywhere else (including at infinity). Let $\mathcal O_D^1$ denote the multiplicative group of norm one elements in $\mathcal O_D$.

Since $D \otimes_{\mathbb Q} \mathbb R \cong M_2(\mathbb R)$, we may regard $\mathcal O_D^1$ as a discrete subgroup of $SL_2(\mathbb R)$, and form the quotient $X := \mathcal O_D^1\backslash \mathcal H$ (where $\mathcal H$ is the upper half-plane).

We may also consider the usual congruence subgroup $\Gamma_0(pq)$ consisting of matrices in $SL_2(\mathbb Z)$ that are upper triangular modulo $pq$, and form $X_0(pq)$, the compacitifcation of $\Gamma_0(pq)\backslash \mathcal H$.

Now the theory of modular and automorphic forms and their associated Galois representations show that $X$ and $X_0(pq)$ are both naturally curves over $\mathbb Q$, and that there is an embedding of Galois representations $H^1(X) \to H^1(X_0(pq))$. Thus the Tate conjecture predicts that there is a correspondence between $X$ and $X_0(pq)$ inducing this embedding. Passing to Hodge structure, we would then find that the periods of holomorphic one-forms on $X$ should be among the periods of holomorphic one-forms on $X_0(pq)$.

Now the theory of $L$-functions shows that the periods of holomorphic one-forms on $X_0(pq)$ in certain cases compute special values of $L$-functions attached to modular forms on $\Gamma_0(pq)$. So putting this altogether, we would find that (given the Tate conjecture) we could compute special values of $L$-functions for certain modular forms by taking period integrals on the curve $X$. In certain respects $X$ is better behaved than $X_0(pq)$, and so this is an important technique in investigating the arithmetic of $L$-functions.

Now it happens that in this case the Tate conjectures is a theorem (of Faltings), and so the above argument is actually correct and complete.

But there are infinitely many other analogous situations in the theory of Shimura varieties where the Tate conjecture is not yet known, but where one would like to know it.

There are also similar arguments where one uses the Hodge conjecture to move from information about periods to information about Galois representations and arithmetic.

I should also say that there are lots of partial results along the above lines in these cases; there are many inventive techniques that people have introduced for getting around the Hodge and Tate conjectures. (Deligne's use of absolute Hodge cycles, applications of theta corresondences by Shimura, Harris, Kudla, and others, ... .) But the Hodge and Tate conjectures stand as fundamental guiding principles telling us what should be true, and which we would dearly like to see proved.

Summary: algebraic cycles are very rich objects, which straddle two worlds, the world of period integrals and the world of Galois representations. Thus, if the Hodge and Tate conjectures are true, we know that there are profound connections between those two worlds: we can pass information from one to the other through the medium of algebraic cycles. If we had these conjectures available, it would be an incredible enrichment of our understanding of these worlds; as it is, people expend a lot of effort to find ways to pass between the two worlds in the way the Hodge and Tate conjectures predict should be possible.

Another example, added following the request of the OP to provide an example just involving complex geometry: If $X$ is a K3 surface, then from the Hodge structure on the primitive part of $H^2(X,\mathbb C)$, one can construct an abelian variety, the so-called Kuga--Satake abelian variety associated to $X$. The construction is made in terms of Hodge structures. One expects that in fact there should be a link between $X$ and its associated Kuga--Satake abelian variety provided by some correspondence, but this is not known in general. It would be implied by the Hodge conjecture. (This constructions is discussed in this answer and in the accompanying comments.)

  • 23
    $\begingroup$ This is fantastically instructive. Thank you for it. $\endgroup$ Feb 3, 2011 at 17:36
  • 6
    $\begingroup$ This is a superb answer. Let me point out one thing you have glossed over, because it tripped me up when I was trying to follow this: Let $X$ and $Y$ be varieties, and let $Z_1$ and $Z_2$ be two subvarieties of $X \times Y$. Then $[Z_1]$ and $[Z_2]$ are the same map in the category of correspondences if $Z_1$ and $Z_2$ are equivalent cycles on $X \times Y$. For example, the different inclusions of a point into $\mathbb{P}^1$ are all the same correspondence. If you miss this point, you will think that $H^*$ is obviously nonfaithful. $\endgroup$ Feb 3, 2011 at 19:01
  • 3
    $\begingroup$ Dear David, Thanks for this; I should have said that the morphisms are correspondences modulo homological equivalence. I may make an edit to this effect, since the potential for confusion is severe! Best wishes, Matthew $\endgroup$
    – Emerton
    Feb 3, 2011 at 19:09
  • 2
    $\begingroup$ I'm wondering, do we know an explicit correspondence between $X$ and $X_0(pq)$ inducing the embedding on the $H^1$'s ? $\endgroup$
    – Alex
    Feb 4, 2011 at 19:20
  • 1
    $\begingroup$ Can't edit comments... I meant over $\mathbb{Q}$ or a number field. I'm aware of some results on the geometric realization of the Jacquet-Langlands correspondence, but it always seems to come from the reductions mod $p$ of the Shimura varieties, and that would only give maps between the motives over a finite field, and we would like to have maps between the motives over $\mathbb{Q}$. $\endgroup$
    – Alex
    Feb 4, 2011 at 19:26

Here are three points, and you'd have to care about at least one of them, I think.

(1) A (co)homology class is better understood if it is represented geometrically in some way.

This point really belongs to geometry, and to a stubborn resistance to eliminating "elements" from co(homology) groups. In other words you might not believe this if the programmatic view of algebraic topology, in terms of category theory and functors, has your complete assent.

(2) Algebraic geometry of codimension 2 or more is much harder than in codimension 1.

Hyperplane sections only get you so far, in other words.

(3) The Hodge conjecture is harder, the more special the variety is.

This point indicates a connection at least to the attitude in number theory, where diophantine considerations are most interesting for the "least random" sets of equations. Or to the allied view that the Hodge conjecture predicts "hidden secrets", algebraic cycles that can be seen to be present from abstract considerations that are pretty much "invariant theory", much more easily than they can be constructed.


  • 1
    $\begingroup$ What do you mean when you say "elements"? $\endgroup$
    – S. Carnahan
    Feb 3, 2011 at 15:51
  • 1
    $\begingroup$ Just a turn of phrase: not every abelian group has to be seen in terms of abelian categories, which is where homological algebra tends towards. A subvariety may give you an actual element of a cohomology group (up to all sorts of caveats). $\endgroup$ Feb 3, 2011 at 16:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.