Another Relation Between Approaches

to the Schottky Problem

arXiv:alg-geom/9202010 v1 13 Feb 1992

John B. Little

1. Introduction
The recent extensive work on several approaches to the Schottky problem has produced
marked progress on several fronts. At the same time, it has become apparent that there
exist very close connections between the various characterizations of Jacobian varieties
described in Mumford’s classic lectures [M] and the more recent approaches related to the
K.P. equation. Some of the most striking results of this kind are to be found in the papers
[B-D] and [F]. In the first of these, Beauville and Debarre show that for a principally-
polarized abelian variety (p.p.a.v.) (A, Θ) of dimension g, the Andreotti-Mayer condition
dim ΘSing ≥ g − 4 is a consequence of any one of the following hypotheses:
• There are distinct points x, y, z ∈ A such that Θ ∩ Θz ⊂ Θx ∪ Θy ,
• The Kummer variety K(A) has a trisecant line, and
• The theta function θA satisfies the K.P. equation (in the sense described following the
Main Theorem below).
In the second paper cited above, Fay indicates a possible relation between the classical
Schottky-Jung relations and the K.P. hierarchy.
However, until now, the approach via double translation manifolds has seemed to be
quite different from these other approaches to the Schottky problem. The purpose of this
paper is to bring this last approach “into the fold” as it were, and to show precisely how
it relates to the approaches via trisecants and flexes of the Kummer variety, and via the
K.P. equation. Another relation was pointed out in [L], but we believe that the following
theorem gives a much more complete indication of the precise connection.
Main Theorem. Let (A, Θ) be an indecomposable g-dimensional p.p.a.v. over C, and
g
let ψA : A → P2 −1 be the mapping defined by the linear system of second-order theta
functions on A. Assume that the image K(A) = ψA (A) has a “curve of flexes,” or more
precisely that there is an irreducible curve Γ ⊂ A such that for generic p ∈ Γ
g
−1
Γ − p ⊂ VYp = {2t ∈ A|t + Yp ⊂ ψA (ℓ) for some line ℓ ⊂ P2 −1
},

where Yp is the artinian length-3 subscheme Spec(OΓ,p /m3Γ,p ) − p ⊂ (A, 0). Then Θ is a
generalized translation manifold. That is, Θ has a local parametrization of the form:

(1.1) zi = αi (t1 ) + Ai (t2 , . . . , tg−1 ) i = 1, . . . g

1
 It is known that both the hypothesis and the conclusion of this theorem effectively lead
to geometric characterizations of Jacobian varieties among all p.p.a.v. For the hypothesis,
this is work of Gunning and Welters. See for example [W]. Thus, in what follows we
will refer to the hypothesis that K(A) has a “curve of flexes” as the Gunning-Welters
hypothesis. The subsequent paper [A-D] shows that this condition is actually much stronger
than necessary to characterize Jacobians. Indeed, the existence of a jet of VYp of sufficiently
high order at one point is enough. Naturally our proof will not make use of any of these
facts. Instead we will connect the Gunning-Welters hypothesis and the closely related
solutions

∂2
(1.2) u(x, y, t) = 2 2 log θA (xU + yV + tW + z0 )
∂x

(U, V, W ∈ Cg ) of the K.P. equation

(1.3) (3/4)uyy = (ut − (1/4)(6uux + uxxx ))x

to a differential-geometric criterion for the existence of parametrizations of the form (1.1)

for a hypersurface in Cg . The proof of this result occupies §§2,3 of the present paper.
In one sense, our result is somewhat disappointing because it indicates that the ap-
proach to the Schottky problem via translation manifolds can be subsumed in the other
standard approaches. However, even though this is true, this point of view does lead to
one further characterization of Jacobian varieties that may be of interest in its own right.
Namely, (A, Θ) is a Jacobian if and only if its theta function satisfies a certain system of
fifth order PDE (found essentially by eliminating U, V, W from the solution of the K.P.
equation given in (1.2)). The equations are unfortunately extremely complicated, but a
procedure which describes how they may be constructed may be found in [T]. We hope to
return to these equations in a future paper.

2
2. A Consequence of the Gunning-Welters Hypothesis and the K.P. Equation

We begin by fixing some notation. Let A ∼

= Cg /(Zg + ΩZg ) be a p.p.a.v. and let
X
θA (z, Ω) = exp πi[nt Ωn + 2nt z]
n∈Zg

be its Riemann theta function. We will also make use of the related theta functions with
characteristics
 
ε X
θ (2z, 2Ω) = exp 2πi[(n + ε/2)t Ω(n + ε/2) + 2(n + ε/2)t z]
0 g
n∈Z

where ε is any vector of zeroes and ones in Zg . It is well-known that thes functions form a
basis for the linear system of second-order theta functions on A. We denote by θ~2 (z) the
vector-valued function    
ε
θ~2 (z) = . . . , θ (2z, 2Ω), . . .
0

The mapping ψA in the Main Theorem is the projectivization of θ~2 (z).

Following the work of Welters [W], we interpret the hypothesis of the Main Theorem
as follows. Let p ∈ Γ. The mapping Yp → A is induced by a local homomorphism

OA,0 → C[t]/(t3 )
2
X
f → ∆i (f )ti
i=1

where the ∆i are certain differential operators. There are corresponding translation-
invariant vector fields D1 , D2 on A such that formally
 
X X
exp  Dj tj  ≡ ∆k tk (mod t3 )
j k

The Gunning-Welters hypothesis is that

 
rank θ~2 ∆1 θ~2 ∆2 θ~2 ≤ 2

at all points of 21 (Γ − p). We begin by noting

Proposition 1. ([W]) The artinian scheme Yp (defined by D1 and D2 ) is equal to the
second order jet of VYp at p.
This follows from the fact that for indecomposable p.p.a.v. A, the rank of the 2g ×
(g(g + 1)/2 + 1) matrix
∂2 ~
 
~
θ2 (0) θ2 (0)
∂zi ∂zj

3
is equal to g(g + 1)/2 + 1.
Furthermore, the existence of a third-order jet of Γ at p implies that there exist a
vector field D3 and a scalar d ∈ C such that
3
(2.1) 0 = [D14 − D1 D3 + D22 + d]θ~2 (0)
4
Pg ∂
Pg ∂
Pg ∂
We will write D1 = i=1 Ui ∂z i
, D2 = i=1 Vi ∂z i
, and D3 = i=1 Wi ∂zi . By the
Riemann quadratic theta formula, (2.1) can be rewritten as
2
(2.2) 0 = θxxxx θ − 4θxxx θx + 3θxx + 4θx θt − 4θxt θ + 3θyy θ − 3θy2 + 8dθ 2 ,

where θ = θ(x, y, t) = θA (xU + yV + tW + z0 ). By a direct computation, this equation is

seen to be equivalent to the equation obtained by subsituting (1.2) into (1.3).
The key point for us will be that, under the assumption that Γ is one-dimensional, we
can actually construct a two-dimensional family of relations (2.2) for each z0 ∈ Cg . That
is, there is a 2-parameter family of triples U, V, W which yield relations as in (2.2). This
dimension count may be explained as follows. We have one parameter for the point p ∈ Γ,
and a second for the choice of U ∈ Tp (Γ).
The group of transformations

U −→ λU
V −→ ±(λ2 V + 2αλU )
W −→ λ3 W + 3λ2 α + 3λα2 U
d −→ λ4 d

leaves (2.1) invariant and acts on our set of triples U, V, W . The quotient projects to a
curve Γ′ in the projective space Pg−1 with homogeneous coordinates (U1 , . . . , Ug ). In other
words, Γ′ is the image of the projectivized Gauss mapping on Γ. (The relation between Γ′
and Γ is the same as the relation between the canonical image of a curve C and the W1
subvariety of J(C).)
Now, let us specialize to the case z0 ∈ Θ so that θ(0, 0, 0) = θA (z0 ) = 0. Setting
x = y = t = 0 in (2.2), we have
2
(2.3) 0 = −4θxxx θx + 3θxx + 4θx θt − 3θy2

Furthermore, if we now add the extra condition that U 6= 0 is tangent to Θ at z0 , so that

θx (0, 0, 0) = 0, then (2.3) reduces to
2
0 = θxx − θy2

at x = y = t = 0, or
 2 !2
X X
(2.4) 0= θij (z0 )Ui Uj  − θi (z0 )Vi ,
i,j i

4
 2
where we have written θi = ∂θ ∂ θA
∂zi and θij = ∂zi ∂zj .
A

It is interesting to note that it is precisely this same special case of the K.P. equation
that was used by Beauville and Debarre in [B-D] to link several different characterizations
of Jacobians. We will use (2.4) in a different way in what follows. The condition that
U ∈ Tz0 (Θ) is that
X
(2.5) θi (z0 )Ui = 0
i

As p varies on the curve Γ ⊂ A, the point with homogeneous coordinates (U1 , . . . , Ug ) varies
on the curve Γ′ ⊂ Pg−1 . The equation (2.5) defines the hyperplane PTz0 (Θ) ⊂ Pg−1 , which
will meet the curve Γ′ in a finite number of points q1 (z0 ), . . . , qn (z0 ). If z0 varies in a small
open set on Θ, we can express the coordinates of any one of these intersection points, say
q1 , as analytic functions of any convenient set of local coordinates on Θ. For simplicity,
we will assume that U1 (z0 ) 6= 0 for all z0 in our small open set on Θ.
Since dim(Γ′ ) = 1, we may also assume (by renumbering if necessary) that after we
divide by U1 to obtain affine coordinates, each of τ2 = U2 /U1 , . . . , τg = Ug /U1 can be
expressed in terms of τ2 on Γ′ . We have completed the preliminary constructions to prove
Proposition 2. Let (A, Θ) be an indecomposable p.p.a.v. satisfying the Gunning-Welters
hypothesis. Then there exists an analytic function τ2 on an open subset X ⊂ Θ and analytic
functions
Pg τ3 (τ2 ), . . . , τg (τ2 ), and σ1 (τ2 ), . . . , σg (τ2 ) such that for all z0 ∈ X,
(i) Pi=1 θ(z0 )τi = 0, P (by convention, here and in the following equation we put τ1 = 1),
(ii) i,j θij (z0 )τi τj + i θi (z0 )σi = 0, and
dτi
(iii) there is an analytic function λ = λ(τ2 ) such that σi = λ · dτ 2
for i = 1, . . . , g.
Proof. (i) is the condition that (U1 , . . . , Ug ) ∈ Tz0 (Θ), which holds by the argument given
before the statement of the Proposition. From (2.4) and (i), we can obtain relations such
as (ii) for any σi = ±Vi /U12 + µUi , where µ is an arbitary function of τ2 (independent of
i). By Proposition 1, at each point of Γ, U and V span the osculating plane to Γ. Hence,
for the local parameter τ2 on Γ, we have dU dτ2
i
= aVi + bUi for some functions a, b of τ2 . We
can take
1 dτi
σi =
aU dτ
 1 2 
1 dUi Ui dU1
= −
aU12 dτ2 aU13 dτ2
 
1 b 1
= 2 Vi + − Ui
U1 a aU13
Then (ii) and (iii) follow from (2.4) and (i). △

We remark that since (2.4) factors as

   
X X X X
0= θij (z0 )Ui Uj + θi (z0 )Vi  ·  θij (z0 )Ui Uj − θi (z0 )Vi 
i,j i i,j i

5
there are usually at least two systems of functions τi , σj satisfying the conditions of Propo-
sition 2. For if τi , σj give one solution of the equations (i) and (ii), then −τi , −σj give
another. This is a consequence of the symmetry of the theta divisor.

3. Generalized Translation Manifolds

In this section, we will show that the conclusion of Proposition 2 of §2 implies that
the theta-divisor Θ is a generalized translation manifold (see the Main Theorem in §1 and
[L]). The following proof is inspired by a similar discussion in [T].
Proposition 3. Let H ⊂ Cg be an analytic hypersurface, defined by the equation
f (z1 , . . . , zg ) = 0. Assume that there exists an analytic function τ2 = τ2 (z1 , . . . , zg ) on
H, and analytic functions τ1 = 1, τi(τ2 ) for i = 3, . . . , g, and σj (τ2 ) for j = 1, . . . , g
satisfying
P
(i) Pi fi (z)τi = 0, P
(ii) i,j fij (z)τi τj + i fi (z)σi = 0 and
dτi
(iii) there exists an analytic function λ = λ(τ2 ) such that σi = λ dτ 2
for i = 1, . . . , g.
If H is not developable (that is, if the rank of the Gauss map on H is generically g −1),
then there exists an analytic parametrization

(3.1) zi = αi (t1 ) + Ai (t2 , . . . , tg−1 ) i = 1, . . . , g

for H. Conversely, the existence of a parametrization (3.1) for H implies the existence of
τi and σj satisfying (i), (ii), and (iii).
Proof. The converse is easily seen by substituting (3.1) into the equation of H and
differentiating twice with respect to t1 . From
∂
0= f (α(t1 ) + A(t2 , . . . , tg−1 ))
∂t1
we obtain X
0= fi (z)α′i (t1 )
i

for all z ∈ H. Differentiating again,

∂2
0= f (α(t1 ) + A(t2 , . . . , tg−1 ))
∂t21
which yields X X
0= fij (z)α′i (t1 )α′j (t1 ) + fi (z)α′′i (t1 )
i,j i

Hence, by renumbering the coordinates if necessary, we can take τ2 = α′2 /α′1 and τi = α′i /α′1
will all be functions of τ2 . Then following the same idea as in the proof of Proposition 2,
letting
1 dτi 1
σi = ′ = ′ 3 (α′1 α′′i − α′i α′′1 )
α1 dt1 (α1 )

6
we get a system of functions satisfying (i), (ii), and (iii).
For the direct implication, suppose that a system of functions τi , σj exists satisfying
(i), (ii), and (iii). Consider any one of the submanifolds K ⊂ H defined by setting τ2 = t0
(constant). Let z0 = (z10 , . . . , zg0 ) be an arbitrary point on K, and construct the integral
curve α = α(z1 ) of the system of ODE

dz2 dzg
= τ2 , . . . , = τg
dz1 dz1

with initial condition z0 in Cg . By condition (i), along α

X dαi X
(3.2) fi (α) = fi (α)τi = 0
i
dz1 i

Hence α lies on H (but not on K or any of the other submanifolds τ2 = constant).

Differentiating (3.2) with respect to z1 , along α we have
X X dτi dτ2
(3.3) 0= fij (α)τi τj + fi (α)
i,j i
dτ2 dz1

Subtract (3.3) from (ii) to obtain

 
X dτi dτ2
0= fi (α) σi −
i,j
dτ2 dz1

dτi
By hypothesis (iii), we can write σi = λ dτ 2
so this last equation becomes
! 
X dτi dτ2
(3.4) 0= fi (α) λ−
i
dτ2 dz1

The first factor cannot be zero under our hypotheses P since if it were, H would be de-
velopable. Indeed from (3.3), we would have i,j fi,j (α)τi τj = 0 and this implies that
the rank of the projectivized Gauss map is ≤ g − 2 everywhere on X ⊂ H, as fol-
lows. We express the submanifold K locally as the intersection of f (z1 , . . . , zg ) = 0 and
zg = h(z1 , . . . , zg−1 ). Then for constant τ2 , differentiating (i) with respect to z1 , . . . , zg−1
on K we obtain
X X ∂h
(3.5) fij (z0 )τi = − fig (z0 )τi j = 1, . . . g − 1
i i
∂zj
P P
Since j τj [ i fij (z0 )τi ] = 0 by (3.3), substituting from (3.5) we obtain
" # g−1

X X ∂h 
0= fig (z0 )τi τg − τj
i j=1
∂zj

7
The second factor here is not zero since α is not tangent to K. Hence the first factor must
be zero. This together with (3.5) shows that the projectivized Gauss map on H has rank
≤ g − 2 at z0 . Since z0 was general on K, as z0 and τ2 vary, we see that H would be
developable if the first factor in (3.4) were zero.
Thus, from (3.4) λ = dτ dz1 , and the ODE defining the curve α now take the form
2

dz1 1 dz2 τ dzg τg

= , = ,..., =
dτ2 λ dτ2 λ dτ2 λ

Integrating from τ2 = t0 to τ2 = t1 we obtain a parametrization for α as follows.

t1
τi
Z
(3.6) zi = zi0 + dτ2
t0 λ

(i = 1, . . . , g). The most important feature of these parametric equations is that the inte-
grand in the second term is a function of τ2 alone. As a consequence, the integral curves
α starting from all points z0 ∈ K are parallel translates of one another in the ambient
Cg . This also implies that the submanifolds K are parallel translates of one another in
Cg . Thus if we substitute any parametrization zi0 = Ai (t2 , . . . , tg−1 ) for K into (3.6), we
obtain a parametrization of the form (3.1). The hypersurface generated in this way must
coincide with H, since as we noted above each integral curve α is contained in H. △

Combining Propositions 2 and 3 completes the proof of the Main Theorem from §1,
since the theta divisor of an indecomposable p.p.a.v. is always non-developable.
In [L], we showed that a p.p.a.v. whose theta divisor has two distinct parametrizations
of the form (3.1) satisfying certain general position hypotheses is the Jacobian of a non-
hyperelliptic curve. It seems highly likely to us that these general position hypotheses are
satisfied automatically whenever a p.p.a.v. has two parametrizations of the form (3.1), but
we do not have a proof. The idea would be to analyze the “degenerate” generalized double
translation manifolds where these general position hypotheses do not hold, and show that
they have geometric properties incompatible with those of theta divisors, e.g. rulings, etc.
The remark following Proposition 2 shows in addition that for theta divisors the
existence of one parametrization (3.1) usually implies the existence of a second. It is
reasonable to conjecture that the only time this fails is for hyperelliptic Jacobians, where
the extra symmetry of the W1 subvariety causes the two generically distinct solutions of
(2.5) to coincide.
We conclude by mentioning once again that sections 17-19 of Tschebotarow’s paper
[T] contain a discussion of a systematic procedure for eliminating the τi , σi , λ from the
hypotheses of Proposition 3. The result is a simultaneous system of two fifth order PDE
on the defining equation of the hypersurface which are satisfied if and only if the hypersur-
face is a generalized translation manifold. Applied to theta functions, these PDE should
characterize the theta functions of Jacobians.

8
 References

[A-D] Arbarello, E. and DeConcini, C. On a set of equations characterizing Riemann matri-

ces, Annals of Math. 120 (1984), 119-141.
[B-D] Beauville, A. and Debarre, O. Une relation entre deux approches du probleme de
Schottky, Inv. Math. 86 (1986), 195-207.
[F] Fay, J. Schottky relations on 12 (C − C), in Proceedings of 1987 Bowdoin Summer
Research Institute on Theta Functions, Proc. of Symposia in Pure Math. 49 (1989),
485-502.
[L] Little, J. Translation manifolds and the Schottky problem, in Proceedings of 1987
Bowdoin Summer Research Institute on Theta Functions, Proc. of Symposia in Pure
Math. 49 (1989), 517-530.
[M] Mumford, D. Curves and their Jacobians Ann Arbor: University of Michigan Press,
1975.
[T] Tschebotarow, N. Über Flächen welche Imprimitivitätssysteme in Bezug auf eine
gegebene kontinuirliche Transformationsgruppe enthalten, Receuil Math. (Sbornik)
34 (1927), 149-204.
[W] Welters, G. On flexes of the Kummer variety, Indagationes Math. 45 (1983), 501-520.

John B. Little
College of the Holy Cross
Worcester, MA 01610
U.S.A.