NOTES ON THE HODGE CONJECTURE FOR FERMAT

VARIETIES

GENIVAL DA SILVA JR.

Abstract. We review a combinatoric approach to the Hodge Conjecture for

Fermat Varieties and announce new cases where the conjecture is true.

1. Introduction
The Hodge conjecture is major open problem in Complex Algebraic Geometry
that has been puzzling mathematician for decades now. The modern statement
is the following: Let X be smooth complex projective variety, then the (rational)
cycle class map is surjective:
cl⊗Q : CH p (X) ⊗ Q → H p,p ∩ H 2p (X, Q)
P P
where cl⊗Q ( ai Xi ) = ai [Xi ], ai ∈ Q and [Xi ] is the class of the subvariety Xi .
The case p = 1 is the only case that it is known to hold in general, which follows
from Lefschetz’s theorem on (1, 1)-classes. Special cases have emerged during the
years but all of them were specific for certain classes of varieties. For example,
Abelian varieties of prime dimension, unirational and uniruled fourfolds, hypersur-
faces of degree less than 6, and some others [3].
Using hard Lefschetz theorem, Lefschetz hyperplane theorem and some Hilbert
scheme arguments, we can reduce the Hodge conjecture to the case of an even
dimensional (> 2) variety and primitive middle cohomology classes.
Shioda [5] gave an interesting characterization of the Hodge conjecture for Fermat
varieties, which we now review.

2. Shioda’s work
n
Let Xm ∈ Pn+1 denote the Fermat variety of dimension n and degree m, i.e. the
solution to the equation:
xm m m
0 + x1 + . . . + xn+1 = 0

and µm the group of m-th roots of unity. Let Gnm be quotient of the group
n+2
z }| {
µm × . . . × µm by the subgroup of diagonal elements.
The group Gnm acts naturally on Xmn
by coordinatewise multiplication, moreover,
ˆn n
the character group Gm of Gm can be identified with the group:
Gˆnm = {(a0 , . . . , an+1 )|ai ∈ Zm , a0 + . . . + an+1 = 0}
a
via (ζ0 , . . . , ζn+1 ) 7→ ζ0a0 . . . ζn+1
n+1
, where (ζ0 , . . . , ζn+1 ) ∈ Gnm .
By the previous section, in order to prove the Hodge conjecture, it’s enough
to prove it for primitive classes, therefore in this paper we will focus on primitive
1
2 GENIVAL DA SILVA JR.

cohomology. The action of Gnm extends to the primitive cohomology and makes
i
Hprim n
(Xm i
, Q) and Hprim n
(Xm , C) a Gnm -module. For α ∈ Gˆnm , we set:
n
V (α) = {ξ ∈ Hprim n
(Xm , C)|g ∗ (ξ) = α(g)ξ for all g ∈ Gnm }
Before stating the characterization of Hodge classes we need a few notation. Let
Unm := {α = (a0 , . . . , an+1 ) ∈ Gˆnm |ai 6= 0 for all i}
For α ∈ Unm we set |α| = i <ami > , where < ai > is the representative of ai ∈ Zm
P
between 1 and m − 1. Suppose n = 2p, then we set
Bnm := {α ∈ Unm ||tα| = p + 1 for all t ∈ Z∗m }
2p
Theorem 2.1. [4, 5] Let Hdg p := H p,p ∩ Hprim (X, Q) be the group of primitive
Hodge cycles. Then:
(a) dim V (α)L= 0 or 1, and V (α) 6= 0 ⇐⇒ α ∈ Unm
p
(b) Hdg = α∈Bnm V (α)
n
Now let C(Xm ) denote the subspace of Hdg p which are classes of algebraic
n
cycles. Then C(Xm ) is a Gnm -submodule and by the theorem above there is a
n n
subset Cm ⊂ Bm such that:
M
n
C(Xm )= V (α)
α∈Cn
m

the Hodge conjecture can then be stated as follows:

Conjecture 1 (Hodge Conjecture). For all n, m we have Cnm = Bnm .
By the discussion in the previous section, this is true for n ≤ 2 and all m.
n
The idea to prove this equality for Fermat varieties it to use the fact that Xm
k
’contains’ disjoint unions of Xm with k < n, we then blow that up to find a relation
n
between the cohomologies and to inductively construct algebraic cycles in Xm .
More precisely, we have:
Theorem 2.2. [5] Let n = r + s with r, s ≥ 1. Then there is an isomorphism
r r s s r−1 s−1
f : [Hprim (Xm , C)⊗Hprim (Xm , C)]µm ⊕Hprim r−1
(Xm , C)⊗Hprim s−1
(Xm ∼
, C) −→ n
Hprim n
(Xm , C)
with the following properties:
a) f is Gnm -equivariant
b) f is morphism of Hodge structures of type (0,0) on the first summand and
of type (1, 1) on the second.
c) If n = 2p then f preserves algebraic cycles, moreover if
r−1 r−1 s−1 s−1
Z1 ⊗ Z2 ∈ Hprim (Xm , C) ⊗ Hprim (Xm , C)
then f (Z1 ⊗ Z2 ) = mZ1 ∧ Z2 , where Z1 ∧ Z2 is the algebraic cycle obtained
n
by joining Z1 and Z2 by lines on Xm , when Z1 , Z2 are viewed as cycles in
n
Xm .
In light of this theorem, we introduce the following notation:
Ur,s r s
m = {(β, γ) ∈ Um ×Um |β = (b0 , . . . , br+1 ), γ = (c0 , . . . , cs+1 ), and br+1 +cs+1 = 0}

For (β, γ) ∈ Ur,s

m we define:

β#γ = (b0 , . . . , br , c0 , . . . , cs ) ∈ Ur+s

m
 NOTES ON THE HODGE CONJECTURE FOR FERMAT VARIETIES 3

and for β 0 = (b0 , . . . , br ) ∈ Ur−1

m and γ 0 = (c0 , . . . , cs ) ∈ Us−1
m , we set:

β 0 ∗ γ 0 = (b0 , . . . , br , c0 , . . . , cs ) ∈ Ur+s
m

Using the theorem above we have:

Corollary 2.3. Suppose n = 2p = r + s, where r, s ≥ 1.
a) If r, s are odd and (β 0 , γ 0 ) ∈ Cm
r−1
× Cms−1
then β 0 ∗ γ 0 ∈ Cnm
b) If r, s are even and (β, γ) ∈ (Cm × Cm ) ∩ Ur,s
r s
m then β#γ ∈ Cm
n

n
By the above corollary, the Hodge conjecture can be proven for the Fermat Xm
n
if the following conditions are true for every α ∈ Bm :
(P1) α ∼ β 0 ∗ γ 0 for some (β 0 , γ 0 ) ∈ Br−1 s−1
m × Bm , (r, s odd).
(P2) α ∼ β#γ for some (β, γ) ∈ (Bm × Bm ) ∩ Ur,s
r s
m , (r, s even and positive).
where ∼ means equality up to permutation between factors.
In order to make these conditions more explicit, we introduce the additive semi-
group Mm of non-negative solutions (x1 , . . . , xm−1 ; y) with y > 0, to the following
system of linear equations:
m−1
X
< ti > xi = my for all t ∈ Z∗m
i=1

Also, define Mm (y) as those solutions where y is fixed. Note that by Gordan’s
lemma, Mm is finitely generated.
Definition 2.4. An element a ∈ Mm is called decomposable if a = c + d for
some c, d ∈ Mm , otherwise it’s called indecomposable. An element is called
quasi-decomposable if a + b = c + d for some a ∈ Mm (1) and c, d ∈ Mm with
c, d 6= a.
With this notation we can identify elements of Bnm with elements of Mm using
the map:
n n
{} : α = (a0 , . . . , an+1 ) ∈ Bnm 7→ {α} = (x1 (α), . . . , xm−1 (α), + 1) ∈ Mm ( + 1)
2 2
where xk (α) is the number os i’s such that < ai >= k.
Note that α satisfies (P 1) above if and only if {α} is decomposable. If α satisfies
(P 2) then {α} is quasi-decomposable. Conversely, if the latter is true then α
satisfies (P 1) or (P 2). So it makes sense to introduce the following conditions:
(Pmn
) Every indecomposable elemets of Mm (y) with 3 ≤ y ≥ n2 + 1, if any, is
quasi-decomposable.
(Pm ) Every indecomposable elemets of Mm (y) with y ≥ 3 is quasi-decomposable.
By the results above we conclude:
Theorem 2.5. [5] If condition (Pm ) is satisfied, then the Hodge conjecture is true
n n n
for Xm for any n. If (Pm ) is satisfied then the Hodge conjecture is true for Xm .
For m prime or m = 4, Mm is generated by Mm (1) which gives:
n
Theorem 2.6. [4, 5] If m is prime or m = 4, the Hodge conjecture is true for Xm
for all n.
Shioda manually verified condition (Pm ) for m ≤ 20 and concluded:
n
Theorem 2.7. [5] If m ≤ 20, the Hodge conjecture is true for Xm for all n.
4 GENIVAL DA SILVA JR.

Starting at m = 21 the number of indecomposables and the length of elements

of Mm are very large so it’s hard to verify (Pm ) by hand for unknown cases, unless
m = p2 a square of a prime. In the latter case, condition (Pm ) is not always
true, it’s false for m = 25 for example. However, Aoki[1] explicitly constructed the
algebraic cycles that generates each V (α), such cycles are called standard cycles.
Theorem 2.8. [1] If m = p2 , the Hodge conjecture is true for Xm
n
for all n, even
though condition (Pm ) may be false.

3. New cases of the Hodge conjecture

A natural question is whether or not the Hodge conjecture can always be proved
using condition (Pm ). As described above, there are false negatives, i.e. (Pm ) is
false but the Hodge conjecture is still true.
This is due to the fact that there are cycles not coming from the induced struc-
ture, see [1]. The next obvious question is then for which values of m, if any, the
condition (Pm ) is false, besides m = p2 . This would say that there are cycles, not of
type standard as in [1], such that they too do not come from the induced structure.
Technically they are candidates for a counter-example to the Hodge conjecture.
We used SAGE math to answer that question and more generally to investigate
when conditin (Pm ) is true. All the code used in this section can be found in [2].
In the case of Fermat fourfolds, we computed first all the indecomposable ele-
ments with length ≥ 3, because the (2, 2) cycles have length exactly 3. So the idea
was to find values of m for which there were none of them.
Theorem 3.1. If m ≤ 100 is odd and not divisible by 3, then the Hodge conjecture
4
is true for all Fermat fourfolds Xm .
This is a strong evidence that the Hodge conjecture should be true for fourfolds
4
Xm where m is odd and 3 - m. It also suggests the structure of B4m , namely, if
3 | m then there are indecomposables elements of length 3.
The following corollary is immediate by [5]:
Corollary 3.2. If mi ≤ 100 are odd integers and not divisible by 3, then the Hodge
4 4
conjecture is true for arbitrary products of Fermat fourfolds Xm 1
× . . . Xm k
.
We slightly extended Shioda’s work by verifying condition (Pm ) for m = 21, 27:
n n
Theorem 3.3. The Hodge conjecture is true for Fermats X21 and X27 .
An interesting case is m = 33, where condition (Pm ) is false, because we’ve
explicitly founded a cycle that is not quasi-decomposable and is not of standard
type either.
Proposition 3.4. Condition P33 is false. More precisely, the cycle
(0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 3)
is not quasi-decomposable in M33 .
This proposition confirms that starting at n = 4, there are cycles not coming
from the induced structure. Therefore, we can not prove the Hodge conjecture only
using this approach. One thing that can be done is to find explicitly the algebraic
cycles whose class project non trivially to V (α) for each α ∈ Bnm , see [1].
 NOTES ON THE HODGE CONJECTURE FOR FERMAT VARIETIES 5

In the particular case where m = 3d and 3 - d, as above, we have a candidate.

Consider the following elementary symmetric polynomials in x= (x0 , . . . , x5 ):
p1 (x) := x0 + x1 + x2 + x3 + x4 + x5 = 0
(3.1) p2 (x) := x0 x1 + x0 x2 + . . . x4 x5 = 0
p3 (x) := x0 x1 x2 + . . . x3 x4 x5 = 0
Recall the Newton identity:
(3.2) x30 + x31 + x32 + x33 + x34 + x35 = p1 (x)3 − 3p1 (x)p2 (x) + 3p3 (x)3
Set xd = (xd0 , . . . , xd5 ), then:
(3.3) xm m m m m m d 3 d d d 3
0 + x1 + x2 + x3 + x4 + x5 = p1 (x ) − 3p1 (x )p2 (x ) + 3p3 (x )

Let W denotes the following variety in P5 :

(3.4) p1 (xd ) = p2 (xd ) = p3 (xd ) = 0
4
By construction, W ⊂ Xm is a subvariety of codimension 2, so [W ] ∈ Hdg 2 (Xm
4
)
.
Question 1. Can [W ] project non trivially in V (α) for every α ∈ B4m which is not
quasi-decomposable and not of standard type?
If the answer is yes, then we would have a positive answer to the Hodge conjecture
in this case.
We know by Schur’s lemma that the number of indecomposable elements is
n
finite. Given m ∈ Z+ , in order to prove the Hodge conjecture for Xm and any n,
it’s enough to prove for all Xm n
, n ≤ n0 , where n0 = 2(m0 − 1) and m0 is the largest
length of all the indecomposable elements in Mm .
Let Im be set of indecomposable elements of Mm . Define φ : Z+ → Z+ by the
rule
(3.5) φ(m) = {max y | (x1 , . . . xm−1 , y) ∈ Im }
The following is then immediate:
n
Proposition 3.5. If the Hodge conjecture is true for Xm , for all n ≤ φ(m), then
n
it’s true for Xm and any n.
Therefore, for Fermat varieties of degree m, we don’t need to check the Hodge
conjecture in every dimension. It’s enough to prove the result for dimension up to
φ(m).
A natural question that arises is then what is the explicit expression of the
function φ(m). For m prime or m = 4, we know already that φ(m) = 1. Also, by
[1], we know that for p > 2 prime φ(p2 ) = p+1
2 . Here’s a table with the a few values
of φ(m):

m φ(m) m φ(m) m φ(m) m φ(m)

20 5 26 7 32 9 38 11
21 3 27 5 33 5 39 5
22 7 28 7 34 5 40 17
23 1 29 1 35 8 41 1
24 9 30 9 36 13 42 11
25 3 31 1 37 1 43 1
6 GENIVAL DA SILVA JR.

Based on the values above and the ones already computed, we believe the following
is true:
pk−1 +1
Conjecture 2. For p > 2 prime, we have φ(pk ) = 2 , and φ(2l ) = 2l−2 + 1
for l > 2.
Computing φ(m) for m < 48 gives the following:

17

m+3
13 y= 3
φ(m)

9
7
5
3
1
5 10 15 20 25 30 35 40 45
m

Question 2. It seems that φ(m) ≤ f (m) for some linear function f . Can f be
described explicitly?
For m ≥ 48, computations become more and more time consuming, even for the
computer, and specially if m has a lot of prime powers in its prime decomposition.
But the results obtained here give us a glimpse about the structure of Mm and
consequently, the Hodge conjecture.

References
1. N. Aoki, Some new algebraic cycles on Fermat varieties. J. Math. Soc. Japan 39 (1987), no.
3, 385–396.
2. G. da Silva Jr, https://sites.google.com/view/genival-da-silva-jr/in%C3%ADcio (Ac-
cessed December 25, 2020).
3. J. Lewis, A survey of the Hodge Conjecture, CRM Monograph, Ser. 10, AMS, Providence, RI,
1999.
4. Z. Ran, Cycles on Fermat hypersurfaces. Compositio Mathematica, Tome 42 (1980) no. 1, pp.
121-142.
5. T. Shioda,The Hodge conjecture for Fermat varieties. Math. Ann. 245, 175–184 (1979).