Mumford Type Abelian Fourfolds

• Abelian fourfolds of Mumford type are polarized varieties constructed from a totally real cubic field and a quaternion algebra, forming a unique non-PEL Shimura family.
• They possess a generic endomorphism ring equal to Z and exhibit exceptional Hodge tensors along with a distinguished K3-type substructure in their cohomology.
• Explicit Jacobian examples in the Torelli locus and detailed reduction theory underscore their significance in modern arithmetic geometry and moduli studies.

Abelian fourfolds of Mumford type are polarized abelian fourfolds lying on the one-dimensional Shimura family arising from Mumford’s construction from a totally real cubic field and a quaternion algebra split at exactly one real place. In the modern formulation, they are special points and generic fibers of a non-PEL Shimura curve of Hodge type in $\mathcal A_4$, distinguished by a small Mumford–Tate group, generic endomorphism ring $\mathbf Z$, and additional Hodge tensors not forced by polarization (Bouchet et al., 30 Sep 2025). Subsequent work has described their tensor and reduction structures, their CM specializations, their relation to K3-type Hodge theory, and explicit realizations by Jacobians of genus $4$ curves (Zhu, 2018).

1. Classical construction and moduli interpretation

Mumford’s construction starts with a totally real cubic field $K$, a quaternion algebra $B/K$, $B$ split at exactly one real place and non-split at the other two, and

$\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$

One then defines

$G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$

and the corestriction gives a representation

$\rho:G\to \GL_{8,\Q},$

which lands in $\GSp(8,\Q)$. With the Hodge map

$\mathbf Z$0

in

$\mathbf Z$1

the pair $\mathbf Z$2 is a Shimura datum, and the corresponding Shimura curve parametrizes polarized abelian $\mathbf Z$3-folds (Bouchet et al., 30 Sep 2025).

An equivalent representation-theoretic description writes the group as

$\mathbf Z$4

for a quaternion division algebra $\mathbf Z$5 with $\mathbf Z$6 of degree $\mathbf Z$7; in the fourfold case $\mathbf Z$8. Over $\mathbf Z$9,

$4$0

and the defining $4$1-dimensional Hodge representation satisfies

$4$2

The resulting one-dimensional Shimura varieties are Mumford curves, and in the fourfold case a generic fiber is a polarized abelian fourfold whose generic Hodge group is $4$3 (Xia, 2013).

This construction is of Hodge type rather than PEL type. Its specialness is encoded in the Hodge structure and Mumford–Tate representation, not in generic extra endomorphisms. That distinction is central throughout the later theory (Bouchet et al., 30 Sep 2025).

2. Hodge-theoretic signature and exceptional cycles

A defining feature of abelian fourfolds of Mumford type is that generically

$4$4

while the Hodge structure still carries additional Hodge tensors. In the explicit Torelli-locus realizations, this is stated as the presence of “additional Hodge tensors beyond those forced by polarization,” even though the generic endomorphism algebra remains trivial (Bouchet et al., 30 Sep 2025).

More concretely, for an abelian fourfold $4$5 of Mumford type there are exceptional Hodge classes

$4$6

which cannot be expressed as linear combinations of products of divisor classes. The Hodge classes in $4$7 are generated by divisor classes together with $4$8 (Bouchet et al., 30 Sep 2025). This is the standard Hodge-theoretic hallmark of the Mumford locus in $4$9.

On the weight-$K$0 side, Galluzzi showed that $K$1 of a Mumford fourfold contains a $K$2-dimensional Hodge substructure of K3 type with Hodge numbers

$K$3

Zhu identifies this subspace as

$K$4

where $K$5 and $K$6 is the Killing form on $K$7. For a maximal order $K$8, the canonical lattice

$K$9

has discriminant

$B/K$0

Every K3 surface arising from this transcendental Hodge structure admits an elliptic fibration with section, and the Mordell–Weil torsion is trivial except in a specific exceptional case where it can be $B/K$1 (Zhu, 2018).

These results clarify that the “exceptional” character of Mumford fourfolds is visible simultaneously in $B/K$2, in codimension-$B/K$3 Hodge classes, and in a distinguished K3-type summand of $B/K$4.

3. CM special points and the Shioda comparison

A major clarification in the CM theory is that not every CM abelian fourfold with exceptional codimension-$B/K$5 Hodge classes is of Mumford type. The guiding example is Shioda’s Jacobian

$B/K$6

a $B/K$7-dimensional CM abelian variety with codimension-$B/K$8 Hodge classes not generated by divisors. Although this behavior resembles Mumford’s examples, the paper “Constructing a CM Mumford fourfold from Shioda’s fourfold” proves: $B/K$9 Equivalently, $B$0 is not parameterized by any Shimura curve that parameterizes Mumford constructions (Zhu, 2018).

The decisive invariant is the Weil-type signature for the $B$1-action on $B$2. For a CM abelian fourfold of Weil type one must have

$B$3

For Shioda’s fourfold, however,

$B$4

Since CM Mumford fourfolds are of Weil type, this excludes $B$5 from Mumford’s construction (Zhu, 2018).

The same paper gives a second obstruction. In

$B$6

the only $B$7-dimensional Hodge substructure has Hodge numbers

$B$8

whereas for a Mumford fourfold there is a Hodge substructure of K3 type with Hodge numbers

$B$9

Thus the Mumford–Tate representations are inequivalent (Zhu, 2018).

At the same time, the paper is constructive. By “flipping one of the embeddings $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$0 to its complex conjugate” and twisting the polarization by a unit in

$\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$1

it constructs a genuine CM Mumford fourfold $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$2 with multiplication by $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$3. In this sense the Shioda fourfold is a near miss: not itself of Mumford type, but raw material from which a CM Mumford fourfold can be built (Zhu, 2018).

4. Explicit Jacobians in the Torelli locus

For more than fifty years Mumford’s construction was known abstractly, but no explicit Jacobian examples were available. This changed with the paper “Mumford-type Shimura curves contained in the Torelli locus,” which gave two explicit $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$4-parameter families of genus $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$5 curves over $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$6 whose Jacobians are Shimura families of Mumford type (Bouchet et al., 30 Sep 2025).

The first family is hyperelliptic: $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$7 Its analytification is

$\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$8

The second family is non-hyperelliptic and canonically embedded: $\Cor_{K/\Q}(B)\cong \Mat_{8\times 8}(\Q).$9 and

$G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$0

Both families define Shimura curves inside $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$1, are contained in the Torelli locus, and are not of PEL type (Bouchet et al., 30 Sep 2025).

The proof combines semistable reduction at $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$2, extension over the stacky compactifications $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$3 and $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$4, equality in the Arakelov inequality, arithmetic triangle groups, and Moonen’s classification of Shimura curves in $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$5. The outcome is that the generic fiber is simple of dimension $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$6 and lies in the corestriction/Mumford case. These are the first explicit examples over $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$7 of abelian varieties with

$G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$8

and additional Hodge tensors, and they answer Gross’s question whether abelian fourfolds of Mumford type can be Jacobians of curves (Bouchet et al., 30 Sep 2025).

5. Reduction theory and arithmetic geometry

The reduction theory of Mumford fourfolds has two complementary directions: structural descriptions of good reduction in characteristic $G'(\Q)=\{x\in B^\times \mid xx^\ast=1\}, \qquad G(\Q)=\Q^\times\cdot \{x\in B^\times\mid xx^\ast=1\}\subset B^\times,$9, and density statements for ordinary, simple, supersingular, or basic reduction.

For a Mumford curve $\rho:G\to \GL_{8,\Q},$0, Xia gives a characteristic-$\rho:G\to \GL_{8,\Q},$1 description of good reduction in the generically ordinary case. In the fourfold case $\rho:G\to \GL_{8,\Q},$2, for infinitely many primes $\rho:G\to \GL_{8,\Q},$3 there exists a model $\rho:G\to \GL_{8,\Q},$4 such that the $\rho:G\to \GL_{8,\Q},$5-divisible group decomposes as

$\rho:G\to \GL_{8,\Q},$6

where $\rho:G\to \GL_{8,\Q},$7 is a versally deformed height $\rho:G\to \GL_{8,\Q},$8 Barsotti–Tate group and $\rho:G\to \GL_{8,\Q},$9 is étale of height $\GSp(8,\Q)$0. On crystalline cohomology,

$\GSp(8,\Q)$1

with $\GSp(8,\Q)$2 a rank-$\GSp(8,\Q)$3 Dieudonné crystal and $\GSp(8,\Q)$4 a rank-$\GSp(8,\Q)$5 unit-root crystal. The ordinary locus is characterized by the condition $\GSp(8,\Q)$6, where $\GSp(8,\Q)$7 for the relevant reflex-field prime (Xia, 2013).

A different arithmetic perspective is provided by the theory of generalized Mumford type. An abelian variety is weakly of Mumford type of rank $\GSp(8,\Q)$8 if, for a density one set of primes $\GSp(8,\Q)$9,

$\mathbf Z$00

with $\mathbf Z$01 acting through the $\mathbf Z$02-th external tensor power of the standard representation. The classical fourfold case is $\mathbf Z$03, so

$\mathbf Z$04

and the derived group is of type $\mathbf Z$05. In this setting, places of good ordinary reduction have Dirichlet density $\mathbf Z$06, and the reduction is absolutely simple away from a set of places of Dirichlet density zero (Thakur, 2016).

Recent work establishes an Elkies-type theorem for some actual Mumford fourfolds. If $\mathbf Z$07 is a totally real cubic number field with

$\mathbf Z$08

$\mathbf Z$09 is unramified at all finite places and exactly one real place, and $\mathbf Z$10 is an abelian fourfold in the one-dimensional family defined in Mumford’s paper by $\mathbf Z$11 with field of moduli $\mathbf Z$12, then $\mathbf Z$13 has supersingular reduction at infinitely many primes (Chen, 10 Nov 2025). The proof passes through a Hodge-type Shimura datum

$\mathbf Z$14

CM cycles, integral models, local deformation theory, and real-geodesic analysis on the Shimura curve.

6. Generalizations and neighboring fourfold classes

The term “Mumford type” has a precise scope, and several nearby fourfold families should be distinguished from it. The unitary PEL Shimura curve attached to the genus $\mathbf Z$15 family

$\mathbf Z$16

is explicitly described as a compact Shimura curve of abelian fourfolds with $\mathbf Z$17-action and prescribed signature; it is “not Mumford’s original abelian fourfolds of Mumford type,” but a “closely related Shimura-curve family of special abelian fourfolds of unitary PEL type” (Li et al., 7 Nov 2025). Likewise, the Picard modular fourfold parametrizing abelian fourfolds of Weil type for $\mathbf Z$18 is a different special family, governed by $\mathbf Z$19, even though it also exhibits exceptional Hodge classes and K3-type phenomena (Geemen et al., 2013).

Another adjacent theory is the study of de Rham–Betti groups for simple type IV abelian fourfolds. For simple CM abelian fourfolds, simple fourfolds with quartic CM endomorphism field, and simple Weil-type abelian fourfolds, the de Rham–Betti group equals the Mumford–Tate group: $\mathbf Z$20 The anti-Weil type case remains open. These results are structurally relevant to the Mumford-type landscape, but they do not directly classify classical Mumford fourfolds (Ji, 2 Nov 2025).

The modern picture is therefore stratified. Classical abelian fourfolds of Mumford type form a non-PEL Shimura-curve family with generic endomorphism ring $\mathbf Z$21 and additional Hodge tensors. CM special points of that family are of Weil type and can be accessed explicitly through the Shioda comparison. Explicit Jacobian models now exist in the Torelli locus. Their weight-$\mathbf Z$22 theory contains a canonical K3-type summand. Their reduction theory is controlled by both tensor decompositions of $\mathbf Z$23-divisible groups and minimal-rank monodromy. Around this core sit several neighboring families—unitary PEL, Picard modular, Weil type, and type IV—which clarify by contrast what is specific to the Mumford fourfold phenomenon.