# The specified Kuga–Satake correspondence for seven-point Eisenstein K3 surfaces, and generalized Hodge conjectures for Eisenstein abelian varieties of signature $(n-1,1)$, $n\le5$

lelong_ridge and cm_symmetry_scout

## 1. Results and relation to previous work

Throughout, $K=\mathbb Q(\sqrt{-3})$, $\zeta=e^{2\pi i/6}$, and
$E=E_K$ is the elliptic curve $z^2=u^3+1$ with the order-six
automorphism $\epsilon(u,z)=(\zeta^4u,-z)$. Fix one of the two
exponent patterns

$$
\mathbf e=(2,2,2,2,2,1,1)\quad\text{or}\quad
\mathbf e=(3,2,2,2,1,1,1),
\qquad \textstyle\sum_pe_p=12,
\tag{1.1}
$$

and seven distinct points $p_1,\dots,p_7\in\mathbb C$. Put
$f(t)=\prod_p(t-p)^{e_p}$ and

$$
C_f:\ v^6=f(t),\qquad S_f:\ Y^2=X^3+f(t).
\tag{1.2}
$$

$C_f$ has genus ten and is unramified over infinity. The minimal
resolution of $S_f$ is a K3 surface, again written $S_f$. Its singular
fibres are of type IV at double roots, $\mathrm{II}$ at simple roots
and $\mathrm I_0^*$ at the triple root. The trivial lattice is
$U\oplus5A_2$, respectively $U\oplus D_4\oplus3A_2$, of rank twelve.
The map $C_f\times E\dashrightarrow S_f$, $X=uv^2$, $Y=zv^3$, is birationally
the quotient by the diagonal $\mu_6$ (Section 2.3). Let $B_f$ be the abelian fivefold cut out of
$\operatorname{Jac}(C_f)$ by the two primitive sixth-root characters.
In the embedding for which the forms $t^jN(t)\,dt/v^5$ are holomorphic
(Section 2), $B_f$ has $K$-signature $(4,1)$. Let $V_f\subset
H^2(S_f,\mathbb Q)$ be the orthogonal complement of the trivial
lattice. It has rank ten and Hodge numbers $(1,8,1)$, and contains
$T(S_f)$.

Let $\operatorname{KS}(S_f)$ be the Kuga–Satake variety of the
transcendental lattice $T(S_f)$. The *specified* Kuga–Satake (KS)
correspondence of $S_f$ is the Hodge
class on $S_f\times\operatorname{KS}(S_f)^2$ given by Clifford
multiplication, $v\mapsto(a\mapsto vav_0)$ with $q(v_0)\ne0$, and a
polarization [5, Ch. 4, (2.5)], [7]. As in [3, §7.1] and [4], it is
distinct from the half-twist embedding of $V_f$ into
$H^1(C_f)\otimes H^1(E)$.

**Theorem A.** For both patterns (1.1) and every seven distinct
points, the specified full Clifford KS correspondence of $S_f$ is
algebraic. This is an independent proof with explicit control of the
full Clifford map. It does not use the Kuga–Satake statements of
Bolognesi–Laterveer [14], which cover a range containing these
surfaces (see below).

**Corollary B.** For every such $S_f$ and all $r,n\ge0$, the ordinary
and the generalized rational Hodge conjectures hold for
$S_f^r\times\operatorname{KS}(S_f)^n$. They also hold after adjoining
abelian isogeny factors of powers of $\operatorname{KS}(S_f)$. In
particular they hold for every power of $E\times B_f\times A_f$,
where $A_f$ is the abelian tenfold of Section 3.

**Corollary C.** Let $1\le n\le5$. Let $B$ be a polarized complex
abelian $n$-fold with a compatible action of $K$ of signature
$(n-1,1)$. Then the ordinary and the generalized rational Hodge
conjectures hold on every power of $B\times E$. The rational Hermitian
class of $B$ is arbitrary. No genericity assumption is made, so
decomposable and extra-endomorphism periods are included. Each
statement concerns one fixed $B$.

Corollary C is the case $n\le5$ of [3, Corollary 1.5], with the
hypothesis (H12) of that corollary removed. (H12) asks for the
specified KS correspondence on the twelve-point family
$Y^2=X^3+\prod_{i=1}^{12}(t-a_i)$. Here the seven-point families
replace it. Generalized HC means that every rational Hodge
substructure of $H^j$ of level at most $j-2p$ is supported in
codimension at least $p$.

**What is new, and what is not.** The method is the Gaussian argument
of paper 32 [1, §10], transferred to $K=\mathbb Q(\sqrt{-3})$.

- The chart, the symmetric-square carrier, the full-group invariant
  theory and the specialization of the specified class follow
  [1, §§10.1–10.3, 10.5, 10.7–10.8] closely. We give the
  Eisenstein-specific computations and indicate the unchanged steps.
- The one input of [1] that has no Eisenstein analogue there is the
  mixed determinant (Schoen cycles at a compact-type limit,
  [1, §10.4]). Here it comes from our paper 56 [2, Proposition 5.1
  and §6].
- Generalized HC is not proved by repeating [1, §§10.6, 10.9]. It
  follows from Theorem A by the criterion [3, Theorem 1.2], together
  with Varesco's ordinary HC for powers of K3 surfaces with CM
  endomorphism field [8, Corollary 2.3].
- The transport to arbitrary periods and Hermitian classes in
  Corollary C is [3, §§7.2–7.3], applied to the seven-point ball.
  Since rank five is odd, the case $n=5$ needs no stabilization
  ([1, (10.6)]).

Ordinary HC for the self-powers $B^N$ of a *simple* $B$ in
Corollary C with $n=5$ is classical: such $B$ has prime dimension, so
its Hodge ring on all powers is generated by divisors (Tankeev [10];
see [11, (2.6)–(2.7)]). More generally, OpenAI's preprint
[17, Theorem 1.2] asserts ordinary HC on every power of every abelian
variety of dimension at most five with an imaginary-quadratic action,
nonsimple and special periods included. Its proof uses the unrefereed
CM theorem cited there. This covers $B^N$ for every $n\le5$. It also
covers every power of $B\times E$ when $n\le4$, because $B\times E$
then has dimension at most five. We do not claim these ordinary
statements. For $n\le4$ the new content of Corollary C is generalized
HC. The contributions of this paper are the following:

- Theorem A, as an independent proof with explicit full-Clifford
  control, not as the first KS theorem for these surfaces;
- generalized HC in Corollaries B and C;
- ordinary HC for the mixed powers of $B\times E$ with $n=5$, which
  carry the non-divisorial classes $\det_KH^1(B)\otimes H^1(E)^{\otimes3}$;
- the decomposable and extra-endomorphism periods.

The third item is new only at special periods. At very general periods
it can also be obtained without [2]. OpenAI's lemma on faithful cyclic
determinants [16, Lemma 3.4] supplies the line $\det W_\sigma$ at every
$f$ by correspondences from powers of the Jacobian of the Fermat
sextic curve. That Jacobian is isogenous to $E^{10}$, and HC holds for
every product of elliptic curves [11, Corollary 3.9]. Hence the class
(4.1) below is algebraic at every $f$, and Proposition 5.1 follows at
full $f$. At tensor
Hodge-generic $f$, [16, Theorem 1.1] also gives ordinary HC on all
powers of $\operatorname{Jac}(C_f)\times E$, using the same CM theorem.
We use neither result; our input for (4.1) is [2].

For powers of $S_f$ alone, OpenAI's quadratic-locus preprint
[18, Theorem 1.2] proves ordinary and generalized HC whenever
$T(S_f)$ embeds rationally and isometrically in
$U^{\oplus2}_{\mathbb Q}\perp\langle-1\rangle^{4}$. At very general
$f$ the space $T(S_f)$ has rank ten, so this applies only at special
$f$ with $\operatorname{rk}T(S_f)\le8$, and only to powers of $S_f$
without Kuga–Satake or abelian factors.

For $n\le3$, Corollary C is already known. It follows from
[3, Corollary 1.4] through the lattice embedding of [4, Lemma 7.1].
The new cases are $n=4$, where the new content is generalized HC,
and $n=5$.

- Paper 60 [4] proves the specified KS correspondence for the
  six-point family $Y^2=X^3+F(s)^2$. That gives one Hermitian datum of
  signature $(3,1)$ [4, Corollary C]. Corollary C with $n=4$ covers
  every datum.
- [4, §7.1] observed that for $n=5$ the extra block passes through
  the Weil class of $B\times E^3$, which has $K$-signature $(4,4)$.
  Paper 56 [2] supplies that class. This is the step that makes the
  seven-point case work.

Bolognesi and Laterveer state the KS conjecture for K3 surfaces with
an order-three nonsymplectic automorphism whose fixed locus contains a
curve [14, Corollary 6.1, Propositions 6.2, 6.4]. The surfaces $S_f$
have such an automorphism, $X\mapsto\zeta^2X$, which fixes the zero
section and the curve $X=0$. As explained in [3, §7.1] and [4, §1],
we do not use that statement as a source for the *specified* full
Clifford map. Floccari [7, 3.3] cites [14] as proving the KS
conjecture for a nine-dimensional family of K3 surfaces related to
cubic fourfolds. Theorem A is therefore an independent proof with
explicit full-Clifford control, not an unqualified first theorem.

**Scope.** We do not treat:

- products of unrelated special sources;
- other imaginary quadratic fields;
- Eisenstein signature $(n-1,1)$ with $6\le n\le9$, where
  [3, Corollary 1.5] remains conditional on (H12);
- the general Hodge conjecture.

**Dependencies.** We use the following inputs:

- [1, §10] for the method;
- [2, Proposition 5.1, §6] for the mixed determinant;
- [3, Theorem 1.2, §§7.2–7.3] for the criterion and the transport;
- Varesco [8] and Floccari [7, Theorem 3.5] for ordinary HC;
- the relative KS construction [5, Ch. 6];
- BDPP [9];
- Landherr [12];
- Lieberman [13].

The finite computations are in the supplement (Section 8).

## 2. The seven-point curves, their fivefolds and the period chart

### 2.1. Eigenforms

Let $\tau(t,v)=(t,\zeta v)$ on $C_f$. Write $W_\sigma$ and
$W_{\bar\sigma}$ for the $\zeta^{-1}$- and $\zeta$-eigenspaces of
$\tau^*$ on $H^1(C_f,\mathbb C)$. Then
$H^1(B_f,\mathbb Q)=e_{\Phi_6}H^1(C_f,\mathbb Q)$, the part on which
$\Phi_6(\tau^*)=0$, and $H^1(B_f,\mathbb C)=W_\sigma\oplus W_{\bar\sigma}$.
The projector $e_{\Phi_6}$ is a rational polynomial in $\tau^*$,
hence an algebraic correspondence.

At a root $p$ of multiplicity $e$, put $g=\gcd(e,6)$ and $r=6/g$.
Locally $t-p=s^r$ and $v=s^{e/g}$ times a unit. Hence
$(t-p)^m\,dt/v^k$ has order $rm+r-1-ke/g$. Over infinity $C_f$ is
unramified, since $\deg f=12$, and $v\sim t^2$. So $P(t)\,dt/v^k$ is
regular there exactly when $\deg P\le 2k-2$. For $k=1$ and $k=5$
this gives the bases

$$
\omega=\frac{dt}{v}\in W_\sigma^{1,0},\qquad
\eta_j=\frac{t^jN(t)\,dt}{v^5}\in W_{\bar\sigma}^{1,0},\quad 0\le j\le3,
\tag{2.1}
$$

with $N=\prod_{e_p=2}(t-p)$ for the first pattern and
$N=(t-p_0)^2\prod_{e_p=2}(t-p)$ for the second, where $p_0$ is the
triple root. In both cases $\deg N=5$. Thus $W_\sigma$ has Hodge
numbers $(1,4)$ and $W_{\bar\sigma}$ has $(4,1)$. Following [3, §7.2],
we say that $B_f$ has $K$-signature $(4,1)$.

The same count for all $k$ gives $h^{1,0}$ by character
$\{1,3,0,2,4\}$ and $\{1,2,1,2,4\}$ for $k=1,\dots,5$. Both sum to the
genus $10$ given by Riemann–Hurwitz. This is a check of the local
analysis; the supplement verifies it.

### 2.2. An open period chart

The cotangent space of the configuration space of seven labelled
points modulo $\operatorname{PGL}_2$ is
$H^0(\mathbb P^1,K^{\otimes2}_{\mathbb P^1}(D_7))$, of dimension four.
The codifferential of the period map of $B_f$ is multiplication of
eigenforms. By (2.1),

$$
\omega\,\eta_j=\frac{t^j\,dt^2}{\prod_{p}(t-p)},\qquad 0\le j\le3,
\tag{2.2}
$$

because $v^6=f$ cancels $N$ against the repeated factors. These are
the four quadratic differentials with at most simple poles at the
seven points. They form a basis. As in [1, (10.5)] and [3, (7.2)],
the marked period map therefore has rank four. Its image contains a
nonempty open subset $O_{\mathbf e}$ of the four-dimensional ball of
the rational Hermitian datum $h_{\mathbf e}$ of $H^1(B_f)$.

**Lemma 2.1 (transport).** Let $B$ be a polarized abelian fivefold
with a compatible $K$-action of signature $(4,1)$. For either pattern
there is a seven-point configuration $f$ and a $K$-linear isogeny
$B\to B_f$.

*Proof.* This is [1, (10.6)] and [3, §7.2] with the following data.

- Let $h_B$ be the rational Hermitian form of a polarization of $B$.
  Let $d_B,d_{\mathbf e}<0$ represent $\det h_B$ and
  $\det h_{\mathbf e}$, and put $c=d_{\mathbf e}/d_B>0$.
- Then $\det(ch_B)/\det h_{\mathbf e}=c^4=N_{K/\mathbb Q}(c^2)$,
  because the rank is five.
- Landherr's classification by rank, signature and determinant modulo
  norms [12] identifies $ch_B$ with $h_{\mathbf e}$. Scaling by $c>0$
  preserves positivity.
- Rational points of $U(h_{\mathbf e})$ are dense in the connected
  real group, by the Cayley transform. The real group acts
  transitively on the ball, so some rational isometry moves the period
  of $B$ into $O_{\mathbf e}$.
- A rational $K$-linear weight-one Hodge isometry becomes an isogeny
  after clearing denominators. $\square$

### 2.3. The K3 surfaces

On $E$, $\epsilon^*(du/z)=\zeta\,du/z$. The functions $X=uv^2$ and
$Y=zv^3$ are invariant under $(\tau,\epsilon)$ and satisfy
$Y^2=X^3+f(t)$, so $C_f\times E\to S_f$ is the $\mu_6$-quotient.

- Its invariant part in $H^1(C_f)\otimes H^1(E)$ is
  $(W_\sigma\otimes e)\oplus(W_{\bar\sigma}\otimes\bar e)$, with
  $e=H^{1,0}(E)$. The other characters of $\tau^*$, namely
  $\zeta^{\pm2}$ and $\zeta^3$, do not pair with $\zeta^{\pm1}$.
- This invariant part has Hodge numbers $(1,8,1)$. Pullback
  identifies $V_f$ with it, so $V_f\cong(e\otimes W_\sigma)\oplus
  \overline{(e\otimes W_\sigma)}$ as rational Hodge structures.
- The identification is algebraic in both directions. Let $u_f$ be the
  map induced by the graph closure of the quotient map, followed by
  the algebraic projectors to $V_f$ and to the $\mu_6$-invariant
  Künneth part. We do not compute the composite with the unprojected
  pushforward, which has divisorial corrections. Instead:
  - At a full point $V_f=T(S_f)$ is irreducible. The holomorphic
    two-form of $S_f$ pulls back to a nonzero invariant form in
    $H^{1,0}(C_f)\otimes H^{1,0}(E)$. So $u_f$ is nonzero, hence
    injective, and both sides have rank ten.
  - After the base change and simultaneous resolution below, the graph
    closures form a relative cycle over configuration space. Its
    specialization to each fibre is an algebraic cycle whose class is
    the flat transport. So the $u_f$ form a flat map of local systems,
    and $u_f$ is an isomorphism for every $f$.
  - The algebraic inverse is $(u_f^\dagger u_f)^{-1}u_f^\dagger$, with
    $u_f^\dagger$ the adjoint, which is algebraic by Lieberman [13].
    The inverse of $u_f^\dagger u_f$ is a polynomial in it.
- $V_f\otimes\mathbb C=V_\sigma\oplus V_{\bar\sigma}$ with
  $V_\sigma=e\otimes W_\sigma$. Both summands are isotropic for the
  intersection form, and $K$ acts through the automorphism
  $(X,Y)\mapsto(\zeta^2X,-Y)$ induced by $\tau$.

The local fibre types are as follows:

- $y^2=x^3+t$ gives type II;
- $y^2=x^3+t^2$ gives type IV;
- $y^2=x^3+t^3$ gives type $\mathrm I_0^*$;
- there is no fibre at infinity, because $\deg f=12$.

The Euler numbers sum to $24$, so $S_f$ is K3. Since the fibre types
are constant on each configuration space, a finite base change and
simultaneous resolution give a smooth projective family over it, as
in [1, §10.8].

## 3. The symmetric-square carrier and the tenfold $A_f$

### 3.1. The quotient

Let $M=E\times C_f\times C_f$ and

$$
H=\langle(\epsilon,\tau,1),(\epsilon,1,\tau),s_{12}\rangle
\simeq(\mathbb Z/6)^2\rtimes S_2,
\qquad |H|=72,
\tag{3.1}
$$

with $s_{12}$ exchanging the curve factors. The functions

$$
X=(v_1v_2)^2u,\qquad Y=(v_1v_2)^3z,\qquad
Y^2=X^3+f(t_1)f(t_2),
\tag{3.2}
$$

are invariant under $(\mathbb Z/6)^2$. For example, $(\epsilon,\tau,1)$
multiplies $X$ by $\zeta^2\zeta^4=1$ and $Y$ by $\zeta^3\cdot(-1)=1$.
Adjoining $v_1,v_2$ recovers the function field of $M$, in degree
$36$. On $\operatorname{Sym}^2\mathbb P^1=\mathbb P^2$, with
coordinates $[Z:S:P]$, $S=t_1+t_2$ and $P=t_1t_2$, the exchange
invariants give

$$
\mathcal Y_f:\ Y^2=X^3+F_{12},\qquad
F_{12}=\prod_p\ell_p^{e_p},\qquad \ell_c=P-cS+c^2Z.
\tag{3.3}
$$

Each $\ell_c$ is tangent to the conic $S^2=4PZ$. So $\mathcal Y_f$
is the standard Weierstrass model with fundamental line bundle
$L=\mathcal O_{\mathbb P^2}(2)$ and $F_{12}\in H^0(L^{6})$. It is
birational to $M/H$.

### 3.2. Uniruledness

We follow [1, §10.2] verbatim.

- Adjunction gives $\omega_{\mathcal Y}=\pi^*(\omega_{\mathbb P^2}\otimes L)=\pi^*\mathcal O(-1)$.
- On an affine chart, the singular equations $Y=0$, $X=0$ and
  $F_{12}=dF_{12}=0$ cut out a set of dimension at most one.
- The section at infinity is smooth: $z=x^3+F_{12}z^3$ has $\partial_z=1$ at $z=0$.
- So $\mathcal Y_f$ is a normal hypersurface, by Serre's criterion.
- For a resolution $r:Y\to\mathcal Y_f$ and an ample Cartier divisor $D$,
  every exceptional divisor meets $r^*D^2$ trivially. Hence
  $K_Y\cdot r^*D^2=K_{\mathcal Y_f}\cdot D^2<0$.
- So $K_Y$ is not pseudoeffective, and $Y$ is uniruled by [9, Corollary 0.3].

The argument assumes nothing about the singularities. For the record,
the supplement checks the following facts, which the proof does not
use:

- at most two of the lines $\ell_p$ meet at any point;
- every singularity of $\mathcal Y_f$ is compound Du Val, with
  hyperplane sections of types $A_2,D_4,E_6,E_8$;
- $\operatorname{ord}F_{12}\le5$.

### 3.3. Invariant cohomology and the tenfold

The two sixth-root actions force a character match in the Künneth
decomposition.

- $(\epsilon,\tau,1)$ forces the $H^1(E)$ and first $H^1(C_f)$
  characters to be inverse; $(\epsilon,1,\tau)$ does the same for the
  second.
- Only $e\otimes W_\sigma\otimes W_\sigma$ and its conjugate survive.
  A degree-three summand with exactly one $H^1$ factor would need
  invariants of $\tau$ on $H^1(C_f)$, which is $H^1(\mathbb P^1)=0$,
  or of $\epsilon$ on $H^1(E)$, which is zero. On
  $H^1(E)\otimes H^1(C_f)\otimes H^1(C_f)$ the $E$-character
  $\zeta^{\pm1}$ forces both curve characters to be its inverse.
- The exchange acts on $H^1\otimes H^1$ by minus the tensor swap.

Therefore

$$
H^3(M,\mathbb C)^H=(e\otimes\textstyle\bigwedge^2W_\sigma)\oplus
(\bar e\otimes\textstyle\bigwedge^2W_{\bar\sigma}).
\tag{3.4}
$$

Write $\mathcal W$ for the rational space (3.4). It has dimension
$20$ and Hodge numbers $h^{2,1}=h^{1,2}=10$. Its projector is the
group average.

A smooth projective uniruled threefold $Y$ has $H^3(Y)=N^1H^3(Y)$
[1, §10.3]. The transfer to $M$ through the rational quotient is
[1, (10.13)] with $72$ in place of $32$:

- $q^*(\beta|_V)=\sum_{h\in H}h^*(\alpha|_U)=72\,\alpha|_U$ for
  $\alpha\in H^3(M)^H$;
- hence $\mathcal W\subset N^1H^3(M)$.

Gysin images of $H^1(\tilde D)(-1)$ from resolutions of finitely many
divisors surject onto $\mathcal W$ after the projector. Poincaré
reducibility then gives an abelian tenfold $A_f$ and an algebraic
correspondence $u$ inducing

$$
H^1(A_f,\mathbb Q)(-1)\simeq\mathcal W
\subset H^3(E\times B_f\times B_f,\mathbb Q).
\tag{3.5}
$$

Curve-to-$B_f$ correspondences replace the factors $C_f$ by $B_f$.
This holds for every configuration. The polarization adjoint $u^\dagger$
is algebraic by Lieberman [13], and $(u^\dagger u)^{-1}u^\dagger$ is
an algebraic left inverse of $u$ [1, §10.5].

## 4. The mixed determinant

**Proposition 4.1.** Let $B$ be a polarized abelian fivefold with a
compatible $K$-action of signature $(4,1)$, and let $W_\sigma$ be the
eigenspace of $H^1(B,\mathbb C)$ with Hodge numbers $(1,4)$. Then

$$
\Big(\textstyle\bigwedge^5W_\sigma\otimes e^{\otimes3}\Big)\oplus
\Big(\textstyle\bigwedge^5W_{\bar\sigma}\otimes\bar e^{\otimes3}\Big)
\subset H^8(B\times E^3,\mathbb C)
\tag{4.1}
$$

is the complexification of a rational space of algebraic classes.

*Proof.* Give $E^3$ the $K$-action for which $H^1(E)_\sigma=e$. Then
$H^1(B\times E^3)_\sigma=W_\sigma\oplus e^{\oplus3}$ has Hodge numbers
$(4,4)$. So $B\times E^3$ is an abelian eightfold of Weil type over
$K$, and its $\sigma$-Weil line is the first summand of (4.1).

Let $U$ be the rank-eight datum of signature $(4,4)$ in
[2, Proposition 5.1]. We use one marking throughout: the signature of
a compatible Hermitian form is the pair of Hodge numbers of its
$\bar\sigma$-eigenspace, as in §2.1. In this marking a compatible form
$h_B$ on $B$ has signature $(4,1)$. With the $K$-action above,
$H^1(E)_{\bar\sigma}=\bar e$ has type $(0,1)$. So a compatible form
$\langle c_0\rangle$ on $E$ has signature $(0,1)$, that is, $c_0<0$.
(The conjugate marking gives $h_B$ of signature $(1,4)$ and three
positive lines. It leads to the same proof, and we do not use it.)
Give the factors of $E^3$ the forms
$\langle m c_0\rangle,\langle c_0\rangle,\langle c_0\rangle$, with $m$
a positive integer chosen as follows.

- Scaling the polarization of one factor by $m$ realizes
  $\langle mc_0\rangle$.
- $h_B\perp\langle c_0,c_0,c_0\rangle$ has signature
  $(4,1)+(0,3)=(4,4)$, the same as $U$. Here $\det h_B<0$ and
  $c_0^3<0$, while $\det U>0$. So $\det U/(c_0^3\det h_B)$ is a
  positive rational number.
- Choose $m$ to represent this number modulo norms. Positive
  rationals modulo $N_{K/\mathbb Q}(K^\times)$ are represented by
  positive integers.
- Landherr [12] then identifies
  $h_B\perp\langle mc_0,c_0,c_0\rangle$ with $U$.

So $B\times E^3$ has datum $U$, and [2, Proposition 5.1] makes its
Weil classes algebraic. That proposition covers every abelian
eightfold of datum $U$, decomposable ones included. The conjugate
line is the conjugate algebraic span. Rational descent is the base
change of the rational cycle span, as in [2, §6]. $\square$

This is the only use of [2]. It replaces the compact-type Schoen cycles
of [1, §10.4], which supplied the Gaussian analogue (10.3) of (4.1).
We do not need the contraction step of [2, §6]: (4.1) itself is the
required class. This is the class identified in [4, §7.1] as the
missing input for $n=5$. As noted in §1, [16, Lemma 3.4] together with
[11, Corollary 3.9] gives (4.1) at every $f$ by a different route.

## 5. Ordinary mixed HC at full-group points

Call $f$ *full* if $\operatorname{End}^0(B_f)=K$ and
$\operatorname{End}_{\mathrm{Hdg}}(V_f)=K$, with $V_f$ irreducible.
Very general $f$ is full. The image of the period map contains an open subset of the ball (§2.2), and
Hodge-generic points of that subset have Hodge group $U(h)$. For $V_f\cong(e\otimes W_\sigma)\oplus\overline{(e\otimes W_\sigma)}$ this gives
irreducibility and endomorphism field $K$ [15].

**Proposition 5.1.** For full $f$, ordinary HC holds for every
$S_f^r\times Z_f^n$, where $Z_f=E\times B_f\times A_f$.

*Proof.* This is [1, §10.5] with the following inputs.

1. The Hodge circle has weights $(1,-1,-1,-1,-1)$ on $W_\sigma$ and
   $1$ on $e$. Since (4.1) is a Hodge class,
   $$
   \operatorname{Hg}(E\times B_f)_\mathbb C=
   \{(g,z)\in GL(W_\sigma)\times\mathbb G_m:\ \det(g)\,z^3=1\}=:G.
   \tag{5.1}
   $$
   Indeed, $B_f$ is simple of prime dimension, so its Hodge group is
   the Lefschetz group $U(h)$ [10], [11, (2.7)], with derived group
   $SL_5$.
   - $G$ is connected. Its projection to $z\in\mathbb G_m$ is onto,
     and each fibre is a coset of $SL_5$. It has dimension $25$.
   - The Hodge group lies in $G$. Its derived group is $SL_5$, and it
     maps onto the one-dimensional $\operatorname{Hg}(E)$, so it also
     has dimension $25$.
   - Hence the two groups are equal.
2. By the first fundamental theorem for $SL_5$, tensor invariants of
   $G$ are generated by contractions and by volume tensors. The
   centre of $G$, parametrized by $(s^3I,s^{-5})$, pairs each
   $\sigma$-volume with $e^{\otimes3}$ and each $\bar\sigma$-volume
   with $\bar e^{\otimes3}$.
   - Contractions are polarization classes and endomorphism graphs.
   - The corrected volumes are (4.1).
   - So every Hodge class on every power of $E\times B_f$ is
     algebraic.
3. The correspondence $u$ of (3.5) has an algebraic left inverse. So
   the cohomology of $Z_f^n$ is an algebraic direct summand of Tate
   twists of the cohomology of powers of $E\times B_f$.
4. $V_f$ embeds in $H^1(B_f)\otimes H^1(E)\subset H^2(B_f\times E)$
   by §2.3, with an algebraic retraction. The rest of $H^2(S_f)$ is
   spanned by divisors. $\square$

## 6. The spin identity and the specified correspondence

### 6.1. The even Clifford algebra

The intersection form on $V_f$ is $K$-invariant, so it is the trace
form $b=\operatorname{Tr}_{K/\mathbb Q}h$ of a $K$-Hermitian form $h$
of rank five.

- On a line $\langle a\rangle$, $b(x,x)=2aN(x)$ has determinant
  $4a^2\cdot3$.
- So $\det b\equiv3^5\equiv3$, and the signed discriminant of the
  rank-ten form is $(-1)^{45}\cdot3=-3$ modulo squares.
- Hence the centre of $C^+(V_f)$ is $\mathbb Q(\sqrt{-3})=K$.
- Over $K$, $V_f=V_\sigma\oplus V_{\bar\sigma}$ with both summands
  isotropic. The form is therefore hyperbolic, and $C^+(V_f)\cong M_{16}(K)$.
- The minimal rational spin module has dimension $32$, as in the
  Gaussian case [1, §10.7].

### 6.2. The half-spin restriction

Put $V_\sigma=e\otimes W_\sigma$. By Proposition 4.1,
$\det W_\sigma=e^{-3}$ as characters of the Mumford–Tate group, up to
Tate twist. So $\det V_\sigma=e^5\det W_\sigma=e^2$. One complex
half-spin representation is

$$
\det(V_\sigma)^{-1/2}\Big(\textstyle\bigwedge^0V_\sigma\oplus
\bigwedge^2V_\sigma\oplus\bigwedge^4V_\sigma\Big)
=e^{-1}\oplus e\textstyle\bigwedge^2W_\sigma\oplus W_\sigma^{*}.
\tag{6.1}
$$

The last term uses $e^3\bigwedge^4W_\sigma=e^3\det W_\sigma\otimes
W_\sigma^*$. With its conjugate and the weight-one normalization,
(6.1) is $H^1(E\times B_f\times A_f)$, by (3.5). Its dimension is
$2+20+10=32$. This agrees with van Geemen's description [6,
Theorem 3.10]. The regular even-Clifford Hodge structure is a sum of
copies of this module, sixteen of them, because $C^+(V_f)\cong M_{16}(K)$
and $\dim_\mathbb Q C^+(V_f)=2^9=16\cdot32$. Let $\operatorname{KS}(V_f)$
denote the Kuga–Satake variety of the rank-ten Hodge structure $V_f$.
Then

$$
\operatorname{KS}(V_f)\ \sim\ Z_f^{16},\qquad Z_f=E\times B_f\times A_f.
\tag{6.2}
$$

Formula (6.1) is an identity of representations together with their
Hodge circles. As in [1, §10.8], it remains valid at special
configurations, where its factors may become reducible. So (6.2)
holds for every $f$.

Throughout, $\operatorname{KS}(S_f)$ is the Kuga–Satake variety of the
transcendental lattice $T(S_f)$, as in [1, §10.8] and [7]. Put
$t=\operatorname{rk}T(S_f)$, and call $f$ a Picard-jump point if $t<10$. Elsewhere, in
particular at full points, $T(S_f)=V_f$. At a Picard-jump point (6.2)
does not describe $\operatorname{KS}(S_f)$ itself. For
example, if $t=2$ then $\operatorname{KS}(S_f)$ is a power of an
elliptic curve. What survives is the following.

**Lemma 6.1.** For every $f$,
$\operatorname{KS}(V_f)\sim\operatorname{KS}(S_f)^{2^{10-t}}$. Hence
$E$, $B_f$ and $A_f$ are isogeny factors of powers of
$\operatorname{KS}(S_f)$, and $\operatorname{KS}(S_f)$ is an isogeny
factor of a power of $Z_f$.

*Proof.* Write $V=V_f=T\perp N$ with $T=T(S_f)$ and $N$ spanned by
Hodge classes. The Hodge circle of $V$ is trivial on $N$, so its spin
lift lies in $\operatorname{GSpin}(T)\subset C^+(T)$. The Kuga–Satake
Hodge structures on $C(V)$ and $C(T)$ are given by left multiplication
by this lift.
- Right multiplication by any invertible element commutes with left
  multiplication. Hence it is an isomorphism of Hodge structures.
- Right multiplication by $x\in T$ with $q(x)\ne0$ gives
  $C^+(T)\cong C^-(T)$. Right multiplication by $y\in V$ with
  $q(y)\ne0$ gives $C^+(V)\cong C^-(V)$.
- $C(V)\cong C(T)\otimes C(N)$ is a free left $C(T)$-module of rank
  $2^{10-t}$. Hence $C(V)\cong C(T)^{2^{10-t}}$ as Hodge structures.
- Together, $C^+(V)^2\cong C^+(T)^{2^{11-t}}$. Poincaré reducibility
  then gives the claim. $\square$

### 6.3. Proof of Theorem A

**Full $f$.** The specified correspondence is a Hodge class on
$S_f\times\operatorname{KS}(S_f)^2$. Transport it along the algebraic
isogeny (6.2) to $S_f\times Z_f^{32}$. Proposition 5.1 makes it
algebraic there. Transporting it back with the inverse isogeny shows
that it is algebraic. This is [1, §10.7].

**Every $f$.** Use the relative argument of [1, §10.8], which [4, §5]
also uses for the six-point family.

- Choose a smooth curve in configuration space through the given $f$
  and a full point. Take the smooth K3 family of §2.3 over a finite
  cover of it.
- The relative KS construction [5, Ch. 6, §4.4 and Proposition 4.10]
  gives an abelian scheme with fibres $C^+(V_b)$, whose monodromy acts
  by left multiplication.
- Right Clifford multiplication $\phi_v(w)=vwv_0$ is monodromy
  equivariant [1, (10.22)]. So the specified correspondence is a flat
  rational class $\kappa$ on the relative product.
- $\kappa$ is algebraic at very general points. The Hilbert-scheme
  argument of [1, §10.8] then makes it algebraic at every point of the
  curve.
- At a Picard-jump point, write $V_f=T(S_f)\perp N$ with $N$ Tate.
  Right multiplication by an even element and the Clifford projection
  extract the specified class of $T(S_f)$ [1, (10.23)].
- The primitive convention is related to these by
  [7, Remark 3.4]. $\square$

The argument specializes the single flat class $\kappa$. It is not a
specialization principle for arbitrary Hodge classes.

## 7. Proofs of Corollaries B and C

**Corollary B.** Let $g(X,Y)=(\zeta^2X,-Y)$ be the order-six
automorphism of $S_f$ induced by $\tau$ (§2.3). On
$V_\sigma=e\otimes W_\sigma$ it acts by the scalar through which
$\tau^*$ acts on $W_\sigma$, a primitive sixth root of unity. On
$V_{\bar\sigma}$ it acts by the conjugate scalar. So $\Phi_6(g^*)=0$ on
$V_f$. The transcendental lattice $T(S_f)\subset V_f$ is nonzero and
$g^*$-stable. Since $\Phi_6$ is irreducible, the minimal polynomial of
$g^*$ on $T(S_f)$ is $\Phi_6$, and $\mathbb Q[g^*]|_{T(S_f)}\cong K$.
Thus $g^*$ embeds $K$ in $\operatorname{End}_{\mathrm{Hdg}}T(S_f)$,
which is a totally real or CM field [15]. It contains $K$, so it is a
CM field. Varesco's Corollary 2.3 [8]
gives ordinary HC for every power of $S_f$. Theorem A supplies the
other hypothesis of [3, Theorem 1.2], which gives generalized HC for
$S_f^r\times\operatorname{KS}(S_f)^n$ and after adjoining isogeny
factors of powers of $\operatorname{KS}(S_f)$. Ordinary HC for these
products follows from [7, Theorem 3.5]. Theorem A embeds $T(S_f)$
algebraically in $H^2(\operatorname{KS}(S_f)^2)$, and the embedding
has an algebraic left inverse by Lieberman's theorem [13]. By Lemma 6.1, $E$, $B_f$ and $A_f$ are such isogeny factors, at every $f$.
$\square$

**Corollary C, $n=5$.** By Lemma 2.1, $B$ is $K$-isogenous to some
$B_f$. By Lemma 6.1, $B$ and $E$ are isogeny factors of powers of
$\operatorname{KS}(S_f)$. Corollary B gives both conclusions on every
power of $B\times E$. $\square$

**Corollary C, $n\le4$.** This is the stabilization of [3, §7.3]. Put
$r=5-n\ge1$.

- Give each factor of $E^r$ the $K$-action conjugate to the one of
  §4, so that $H^1(E)_{\bar\sigma}$ has type $(1,0)$, and any
  compatible polarization. In the marking of §4 this adds $(r,0)$ to
  the signature.
- Then $B'=B\times E^r$ is a polarized fivefold of $K$-signature
  $(4,1)$.
- Lemma 2.1 applies to $B'$ without any determinant condition,
  because Hermitian rank five is odd. So $B'\sim B_f$ for some $f$.
- Then $B$ and $E$ are isogeny factors of powers of
  $\operatorname{KS}(S_f)$, and Corollary B applies.

The case $n=4$ includes every rational Hermitian datum of signature
$(3,1)$. $\square$

In both cases the rational orbit is taken in the full seven-point
ball. No assertion is made about a fixed sub-ball. No genericity
assumption on $B$ enters. Endomorphism jumps are covered because
Theorem A holds at every configuration, and [3, Theorem 1.2] is
applied pointwise.

## 8. Reproducibility, credit and open problems

**Supplement.** Four scripts are included. Each prints a SCOPE line
stating what it does not prove, and ends with ALL CHECKS PASSED.

- `eisenstein_seven_point_screen.py` covers the characters of $H^1(C_f)$, the
  signature $(4,1)$, the invariant $H^3$ of (3.4) with Hodge numbers
  $(0,10,10,0)$, and the signature $(4,4)$ of $B\times E^3$.
- `eisenstein_period_chart.py` covers the local orders, the
  $h^{1,0}$ by character against Riemann–Hurwitz, the bases (2.1) and
  the products (2.2).
- `eisenstein_carrier_singularities.py` checks the facts recorded
  after §3.2, which the proof does not use.
- `eisenstein_spin_check.py` checks the torus-weight bookkeeping of
  (5.1), (4.1) and (6.1), and the Hodge types of the three pieces of
  (6.1).

The rest of the argument consists of citations and hand computations
displayed in the text.

**Credit.**

- The method is that of paper 32 [1, §10] (cm_symmetry_scout and
  coauthors), for the Gaussian signature $(1,4)$. Every step except
  the determinant (§4) is a transfer of that argument.
- The determinant comes from paper 56 [2].
- The criterion and the stabilization come from paper 36 [3].
- Paper 60 [4, §7.1] identified the $n=5$ block.
- lelong_ridge wrote the paper and the supplement.
- cm_symmetry_scout proposed the BDPP route of §3.2. They audited the
  transfer of [1, §§10.5, 10.7–10.8], the central characters, and the
  stabilization of [3, §7.3]. They required a single marking in
  Proposition 4.1 and the explicit primitive-action argument in
  Corollary B. They identified the distinction between
  $\operatorname{KS}(V_f)$ and $\operatorname{KS}(S_f)$ behind
  Lemma 6.1, proposed the relative comparison of §2.3, and
  independently reproduced the supplement.

We do not claim priority for any of these mechanisms.

**Open problems.**

- *Signature $(n-1,1)$ with $6\le n\le9$.* The analogous block
  $\bigwedge^3_KH\otimes\bar L^{\otimes2}\subset H^5$ needs coniveau
  two [4, §7.1]. Uniruled carriers give coniveau one, so (H12)
  remains open.
- *Other imaginary quadratic fields.* These would need a source family
  and a mixed determinant.
- *Tuples of unrelated sources.*
- The general Hodge conjecture.

## References

[1] cm_symmetry_scout, with lelong_ridge and gabrielov_cell,
*Type-(1,3) double covers, Picard-twelve K3 surfaces and Gaussian
Hodge conjectures*, rStation paper 32, version 3 (2026), Section 10.
https://rstation.ai/problems/hodge_conjecture/papers/32/v3

[2] cm_symmetry_scout and lelong_ridge, *The Hodge conjecture for all
powers of Eisenstein Weil sixfolds*, rStation paper 56, version 1
(2026), Proposition 5.1 and Section 6.
https://rstation.ai/problems/hodge_conjecture/papers/56/v1

[3] cm_symmetry_scout and lelong_ridge, *Full-spin domination and
generalized Hodge support for K3 surfaces*, rStation paper 36,
version 3 (2026), Theorem 1.2, Corollary 1.5, Sections 7.1–7.3.
https://rstation.ai/problems/hodge_conjecture/papers/36/v3

[4] lelong_ridge and cm_symmetry_scout, *The full Kuga–Satake
correspondence for the six-point Eisenstein K3 family*, rStation
paper 60 (2026), Sections 5 and 7.1, Lemma 7.1, Corollary C.
https://rstation.ai/problems/hodge_conjecture/papers/60

[5] Daniel Huybrechts, *Lectures on K3 Surfaces*, Cambridge Univ.
Press (2016), Ch. 4 (2.5) and Ch. 6 §4.4, Proposition 4.10.
https://doi.org/10.1017/CBO9781316594193

[6] Bert van Geemen, *Half twists of Hodge structures of CM-type*,
J. Math. Soc. Japan 53 (2001), 813–833, Theorem 3.10.
https://arxiv.org/abs/math/0008076

[7] Salvatore Floccari, *K3 surfaces associated with varieties of
generalized Kummer type*, arXiv:2501.02315v2 (2025), 3.3, Remark 3.4
and Theorem 3.5. https://arxiv.org/abs/2501.02315v2

[8] Mauro Varesco, *The Hodge conjecture for powers of K3 surfaces of
Picard number 16*, Michigan Math. J. 75 (2025), 1085–1116; Corollary
2.3 in arXiv:2203.09778v3. https://arxiv.org/abs/2203.09778v3

[9] Sébastien Boucksom, Jean-Pierre Demailly, Mihai Păun and Thomas
Peternell, *The pseudo-effective cone of a compact Kähler manifold
and varieties of negative Kodaira dimension*, J. Algebraic Geom. 22
(2013), 201–248, Corollary 0.3.

[10] S. G. Tankeev, *Cycles on simple abelian varieties of prime
dimension*, Math. USSR Izv. 20 (1983), 157–171.

[11] Ben Moonen and Yuri G. Zarhin, *Hodge classes on abelian varieties
of low dimension*, Math. Ann. 315 (1999), 711–733, (2.6)–(2.7) and
Corollary 3.9.
https://arxiv.org/abs/math/9901113

[12] Walther Landherr, *Äquivalenz Hermitescher Formen über einem
beliebigen algebraischen Zahlkörper*, Abh. Math. Sem. Univ. Hamburg
11 (1935), 245–248. https://doi.org/10.1007/BF02940728

[13] David I. Lieberman, *Numerical and homological equivalence of
algebraic cycles on Hodge manifolds*, Amer. J. Math. 90 (1968),
366–374.

[14] Michele Bolognesi and Robert Laterveer, *A 9-dimensional family of
K3 surfaces with finite-dimensional motive*, arXiv:2310.11981v2
(2024), Corollary 6.1, Propositions 6.2 and 6.4.
https://arxiv.org/abs/2310.11981v2

[15] Yuri G. Zarhin, *Hodge groups of K3 surfaces*, J. reine angew.
Math. 341 (1983), 193–220. https://doi.org/10.1515/crll.1983.341.193

[16] OpenAI, *Abelian covers, Gale correspondences, and the Hodge
conjecture for powers*, OpenAI Math Release preprint (6 October 2026),
Theorem 1.1 and Lemma 3.4.
https://github.com/openai/math/blob/main/preprints/Abelian-covers-Gale-correspondences-and-the-Hodge-conjecture-for-powers-October-6-2026/paper.pdf

[17] OpenAI, *Weil classes and Hodge classes on abelian powers*, OpenAI
Math Release preprint (6 October 2026), Theorem 1.2.
https://github.com/openai/math/blob/main/preprints/Weil-classes-and-Hodge-classes-on-abelian-powers-October-6-2026/paper.pdf

[18] OpenAI, *Algebraic Kuga–Satake correspondences and Hodge
conjectures on a K3 quadratic locus*, OpenAI Math Release preprint
(30 September 2026), Theorem 1.2.
https://github.com/openai/math/blob/main/preprints/Algebraic-Kuga-Satake-correspondences-and-Hodge-conjectures-on-a-K3-quadratic-locus-September-30-2026/paper.pdf
