Torelli Locus: Jacobians in Moduli Space

• The Torelli locus is the moduli subvariety in A₍g₎ comprised of principally polarized Jacobians, uniquely determined by the Torelli map.
• Its local geometry is elucidated via the second fundamental form and explicit rank estimates, revealing isotropic conditions and restrictions on totally geodesic subvarieties.
• The locus interfaces with arithmetic and degeneration phenomena, addressing non-nodal extensions, Shimura subvarieties, and unlikely intersections in both characteristic zero and positive characteristic.

The Torelli locus is the subvariety of the moduli space $A_g$ of principally polarized abelian varieties that is realized by Jacobians of curves. If $M_g$ denotes the moduli space of smooth projective curves of genus $g$, the Torelli map

$j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$

or, in alternate notation, $\tau_g\colon M_g\to A_g$, identifies curves with their principally polarized Jacobians. Several conventions coexist: the open Torelli locus is the image $T_g^\circ=j(M_g)$, while the closed Torelli locus $T_g$ is either its closure in $A_g$ or, equivalently in the compact-type setting, the image of $M_g^{ct}$, the moduli of stable curves of compact type (Moonen et al., 2011, Pries, 31 Aug 2025). The subject lies at the intersection of the Torelli theorem, the Schottky problem, the geometry of special subvarieties of $A_g$, and the arithmetic of Jacobians in positive characteristic (Ghigi, 2018).

1. Moduli-theoretic framework and basic geometry

For $M_g$0, the basic dimensions are

$M_g$1

Hence the expected codimension of the open Torelli locus is

$M_g$2

This has the familiar consequence that $M_g$3 for $M_g$4, while for $M_g$5 the Torelli locus is a proper subvariety (Moonen et al., 2011, Pries, 31 Aug 2025). In this sense, the Torelli locus is the moduli-theoretic answer to the question “which principally polarized abelian varieties are Jacobians?”

Torelli’s theorem gives injectivity on geometric points: a smooth curve is determined by its principally polarized Jacobian (Moonen et al., 2011, Pries, 31 Aug 2025). At the same time, the geometry of the map is subtler than pointwise injectivity. Outside the hyperelliptic locus, the Torelli map is an immersion in characteristic $M_g$6; along the hyperelliptic locus, ramification phenomena occur, and in positive characteristic the infinitesimal behavior depends sharply on the characteristic (Moonen et al., 2011, Landesman, 2019).

The compact-type boundary is intrinsic to the closed Torelli locus. If $M_g$7 is a stable curve of compact type, then $M_g$8 is again an abelian scheme with a canonical principal polarization, so the Torelli map extends from $M_g$9 to $g$0 (Moonen et al., 2011). The closed Torelli locus therefore contains Jacobians of singular stable curves of compact type, and its boundary consists of decomposable principally polarized abelian varieties coming from products of Jacobians of the irreducible components (Moonen et al., 2011, Pries, 31 Aug 2025). This boundary behavior already indicates that $g$1 is not merely the open Jacobian locus with a compactification attached ad hoc; it is a moduli-theoretically meaningful closure.

2. Local geometry and the second fundamental form

A central local invariant of the Torelli locus is the second fundamental form of the Torelli immersion into $g$2 equipped with the Siegel metric. For a non-hyperelliptic curve $g$3, one has

$g$4

and

$g$5

Under these identifications, the transpose of the differential of the Torelli map is the multiplication map

$g$6

with kernel

$g$7

the space of quadrics containing the canonical image of $g$8 (Ghigi, 2018). Thus the normal directions to the Torelli locus are canonically identified with quadrics in the canonical ideal.

Dualizing the second fundamental form yields a map

$g$9

which is the basic local differential-geometric object attached to the Torelli immersion (Ghigi, 2018). A key structural result is that, for $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$0 and diagonal $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$1,

$j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$2

and there exists a canonical anti-invariant section

$j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$3

such that

$j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$4

This realizes the second fundamental form as literal multiplication by $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$5 on $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$6 (Ghigi, 2018). The construction is Hodge-theoretic, while its restriction to the diagonal is closely related to the second Gaussian map.

This multiplication description is effective. It leads to rank estimates for $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$7 when $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$8 has small rank, and those rank estimates translate into restrictions on linear subspaces isotropic for the second fundamental form. In particular, if $j\colon M_g\to A_g,\qquad [C]\mapsto [J(C)]$9 is a totally geodesic germ generically contained in the Torelli locus, then its tangent space must be isotropic for $\tau_g\colon M_g\to A_g$0, and this yields dimension bounds. One refinement gives

$\tau_g\colon M_g\to A_g$1

improving earlier bounds derived from gonality and from a single rank estimate for $\tau_g\colon M_g\to A_g$2 (Frediani et al., 2019). The same method also produces a sharper hyperelliptic bound,

$\tau_g\colon M_g\to A_g$3

for totally geodesic subvarieties generically contained in the hyperelliptic Torelli locus (Frediani et al., 2019).

A recurrent heuristic in this circle of work is that the Torelli locus should be “very curved” inside $\tau_g\colon M_g\to A_g$4, not in the sense of an explicit sectional-curvature formula, but in the sense that its second fundamental form is highly nontrivial and severely obstructs large totally geodesic pieces (Ghigi, 2018).

3. Totally geodesic subvarieties, Hodge loci, and Shimura geometry

The ambient space $\tau_g\colon M_g\to A_g$5 is locally symmetric: analytically it is a quotient of Siegel space, a Hermitian symmetric domain. Within this setting, Hodge loci are geometrically rigid. An elementary argument shows that Hodge loci in $\tau_g\colon M_g\to A_g$6 are totally geodesic, and hence special or Shimura subvarieties are totally geodesic as well (Ghigi, 2018). This ties the differential geometry of the Torelli locus directly to the arithmetic geometry of special subvarieties.

The resulting global problem is usually framed by the Coleman–Oort philosophy: for large genus, there should be no positive-dimensional special subvariety $\tau_g\colon M_g\to A_g$7 with

$\tau_g\colon M_g\to A_g$8

In the form stated by Moonen–Oort, this is expected for large $\tau_g\colon M_g\to A_g$9, in any case $T_g^\circ=j(M_g)$0 (Moonen et al., 2011). The Torelli locus itself is not special for $T_g^\circ=j(M_g)$1, and no special subvariety contains it (Moonen et al., 2011). This is a basic structural point: the Torelli locus is defined by a moduli problem of curves, but it is too generic from the Mumford–Tate viewpoint to be a Shimura subvariety.

Low genus nevertheless furnishes many counterexamples to naive forms of Coleman–Oort. Families of cyclic covers of $T_g^\circ=j(M_g)$2, and more generally Galois covers, produce special subvarieties inside the Torelli locus when a numerical condition equating family dimension with the dimension of $T_g^\circ=j(M_g)$3 holds. For covers of $T_g^\circ=j(M_g)$4, exactly $T_g^\circ=j(M_g)$5 data $T_g^\circ=j(M_g)$6 with $T_g^\circ=j(M_g)$7 occur for $T_g^\circ=j(M_g)$8, giving $T_g^\circ=j(M_g)$9 distinct Shimura subvarieties, all in genera $T_g$0 and none in genera $T_g$1 (Frediani et al., 2014). For Galois coverings of elliptic curves, there are exactly $T_g$2 positive-dimensional families satisfying the analogous sufficient condition for $T_g$3, all with $T_g$4; only $T_g$5 give genuinely new Shimura subvarieties, the others reproducing examples already obtained from covers of $T_g$6 (Frediani et al., 2015).

These examples can be analyzed at the symmetric-space level. The low-genus Shimura subvarieties arising from Galois coverings are uniformized by explicit Hermitian symmetric domains such as $T_g$7, $T_g$8, and products of discs, and in the elliptic-cover cases the known fibrations by totally geodesic subvarieties correspond to actual product decompositions of the uniformizing symmetric spaces (Tamborini, 2020). At the same time, not every totally geodesic subvariety in the Jacobian locus is arithmetic: low-genus examples include totally geodesic curves that are not Shimura, even though every Shimura variety is totally geodesic (Ghigi, 2018).

The landscape changed again with explicit genus-$T_g$9 examples of non-PEL type. Two families of genus $A_g$0 curves, one hyperelliptic and one non-hyperelliptic, produce Shimura curves in $A_g$1 of Mumford type, analytically $A_g$2 and $A_g$3, and these are realized by actual Jacobians throughout a dense open of the family (Bouchet et al., 30 Sep 2025). This shows that special subvarieties in the Torelli locus are not confined to the classical PEL/Galois-cover framework.

4. The hyperelliptic Torelli locus

The hyperelliptic locus occupies a distinguished position because the full Torelli map fails infinitesimal injectivity precisely at hyperelliptic curves for $A_g$4, but the restricted map behaves better. If $A_g$5 denotes the Torelli map restricted to the hyperelliptic locus, then $A_g$6 is a radimmersion over $A_g$7; it is an immersion over $A_g$8, and for $A_g$9 it is an immersion in every characteristic (Landesman, 2019). In characteristic $M_g^{ct}$0, however, the behavior changes qualitatively: for $M_g^{ct}$1, the restricted map is generically inseparable and at every geometric point

$M_g^{ct}$2

Thus the hyperelliptic Torelli locus contributes genuinely inseparable and nonreduced behavior in characteristic $M_g^{ct}$3 (Landesman, 2019).

The deformation-theoretic mechanism is explicit. The tangent map to Torelli is dual to the multiplication map

$M_g^{ct}$4

and on the hyperelliptic locus the tangent space is cut out by the image of a map

$M_g^{ct}$5

In characteristic different from $M_g^{ct}$6, $M_g^{ct}$7 contributes directions missing from the Torelli multiplication image, so the restricted map is immersive; in characteristic $M_g^{ct}$8, $M_g^{ct}$9 lies inside the Torelli image, producing the uniform $A_g$0 dimensional kernel (Landesman, 2019).

The hyperelliptic Torelli locus also admits its own Coleman–Oort-type nonexistence theorem. For $A_g$1, there is no positive-dimensional Shimura subvariety contained generically in the hyperelliptic Torelli locus (Chen et al., 2015). The argument combines reduction to the case of simple Shimura varieties via Hecke translates, codimension-$A_g$2 boundary in the Baily–Borel compactification, the identification of the singular part of the hyperelliptic Torelli locus with the decomposable locus, and the affineness of the open hyperelliptic locus (Chen et al., 2015). This complements the differential-geometric result that for $A_g$3 there are no totally geodesic curves contained generically in the hyperelliptic Torelli locus (Lu et al., 2014).

A common misconception is that hyperelliptic Jacobians are exactly where Torelli “breaks down.” Infinitesimally this is true for the full map $A_g$4, but after restriction to $A_g$5 the map is still immersive away from characteristic $A_g$6 (Landesman, 2019).

5. Boundary geometry, compactifications, and degeneration of Jacobians

The boundary behavior of the Torelli locus is governed by compact type. On $A_g$7, the divisor $A_g$8 parametrizes irreducible nodal curves, while

$A_g$9

parametrizes stable curves of compact type (Pries, 31 Aug 2025). For a compact-type nodal curve $M_g$00, the Jacobian decomposes as

$M_g$01

which explains why the boundary of the closed Torelli locus is the decomposable locus (Pries, 31 Aug 2025). By contrast, self-gluing produces generalized Jacobians fitting into an extension by a torus, so full injectivity of Torelli fails on the stable boundary (Moonen et al., 2011, Pries, 31 Aug 2025).

The extension problem beyond the nodal boundary is subtler. Alexeev extended the Torelli map to the Deligne–Mumford compactification using stable semiabelic pairs, while Alexeev–Brunyate showed that the Torelli map does not extend over the cuspidal locus in Schubert’s pseudostable compactification (Han et al., 2024). A recent extension theorem identifies a genuinely non-nodal locus where extension still works. For curves with rational $M_g$02-fold points

$M_g$03

the Torelli map extends over the normalization of the compactification of stable separating fold-like curves,

$M_g$04

and more generally over the normalized quasi-separating fold-like locus in any Smyth compactification,

$M_g$05

(Han et al., 2024). The decisive combinatorial input is the Caporaso–Viviani criterion describing when stable curves have the same image under the compactified Torelli map.

This shows that non-nodal singularities do not uniformly obstruct extension. Cusps do obstruct it, but certain seminormal, quasi-separating fold-like singularities still admit canonical stable semiabelic limits (Han et al., 2024). A plausible implication is that the compactified Torelli locus is geometrically controlled by a finer boundary combinatorics than “nodal versus non-nodal.”

6. Arithmetic, positive characteristic, and unlikely intersections

In positive characteristic, the Torelli locus becomes an arithmetic-geometric testing ground for which $M_g$06-divisible groups occur for Jacobians. The basic question is: given a symmetric Newton polygon $M_g$07, does the Newton stratum $M_g$08 meet the open Torelli locus $M_g$09? This is equivalent to asking whether $M_g$10 occurs as the Newton polygon of a smooth curve of genus $M_g$11 (Pries, 31 Aug 2025).

The relevant ambient strata have explicit dimensions. For a symmetric Newton polygon $M_g$12, if $M_g$13 denotes the count of integral lattice points on or above $M_g$14 in the standard way, then

$M_g$15

and for the supersingular polygon $M_g$16,

$M_g$17

(Pries, 31 Aug 2025). The codimension comparison with $M_g$18 yields a useful “unlikely intersection” heuristic: for $M_g$19, supersingular points in the open Torelli locus lie in a Newton stratum whose codimension exceeds $M_g$20 (Pries, 31 Aug 2025).

Concrete existence results nevertheless occur. A Torelli-locus argument using compact-type boundary points shows that for every prime $M_g$21 there exists a smooth supersingular curve of genus $M_g$22; equivalently, $M_g$23 for every $M_g$24 (Pries, 31 Aug 2025). More generally, every irreducible component of the $M_g$25-rank $M_g$26 stratum in $M_g$27 has dimension

$M_g$28

so smooth curves of every genus $M_g$29 and every $M_g$30-rank $M_g$31 exist (Pries, 31 Aug 2025). The closed Torelli locus, especially its compact-type boundary, is central in these arguments because products of lower-genus Jacobians provide explicit boundary points in prescribed Newton strata.

A distinct arithmetic direction concerns unlikely intersections on Hodge-generic curves inside the Torelli locus. If $M_g$32 is a smooth irreducible Hodge-generic curve defined over $M_g$33, and if $M_g$34 intersects the $M_g$35-dimensional stratum of the Baily–Borel boundary of $M_g$36, then for odd $M_g$37 there are only finitely many points $M_g$38 for which the Jacobian $M_g$39 is non-simple (Papas, 2022). This is a Zilber–Pink-type finiteness statement deduced from polynomial height bounds for exceptional points in one-parameter geometric variations of Hodge structure, obtained by extending André’s G-functions method to odd weight (Papas, 2022).

These results show two complementary arithmetic faces of the Torelli locus. In positive characteristic it records which $M_g$40-divisible groups arise from curves; over $M_g$41 it supports finiteness statements for atypical endomorphism phenomena along Hodge-generic one-parameter families (Pries, 31 Aug 2025, Papas, 2022).

7. Torelli space, the hyperelliptic branch locus, and topology upstairs

A finer version of the Torelli locus appears upstairs in Torelli space. If $M_g$42 is Teichmüller space modulo the Torelli group, the period map

$M_g$43

is a $M_g$44-fold branched cover onto its image, and the branch locus is the hyperelliptic locus $M_g$45 in Torelli space (Brendle et al., 2012). Each component has fundamental group

$M_g$46

the hyperelliptic Torelli group (Brendle et al., 2012).

The group $M_g$47 admits an explicit geometric generating set: $M_g$48 This gives a topological interpretation of the hyperelliptic branch locus of the period map, and adding compact-type hyperelliptic degenerations kills precisely these loops, so each component of the partial compactification $M_g$49 becomes simply connected (Brendle et al., 2012).

The topology of the closure is nevertheless large. For $M_g$50, if $M_g$51 is a component of the hyperelliptic locus in compact-type Torelli space, then

$M_g$52

is infinite-dimensional, and therefore $M_g$53 does not have the homotopy type of a finite CW complex (Kordek, 2015). In genus $M_g$54, a component is the zero locus of an even theta-null in $M_g$55, the reducible boundary is a simple normal crossings divisor, and one has

$M_g$56

while

$M_g$57

is free abelian (Kordek, 2015). This gives a concrete analytic model for a component of the hyperelliptic Torelli locus upstairs.

These topological results refine the usual algebro-geometric picture. The Torelli locus is not only a subvariety of $M_g$58; in Torelli space it acquires branch behavior, monodromy, and infinite boundary topology that are invisible on coarse moduli (Brendle et al., 2012, Kordek, 2015).