On the dimension of totally geodesic submanifolds in the Prym loci

In this paper we give a bound on the dimension of a totally geodesic submanifold of the moduli space of polarised abelian varieties of a given dimension, which is contained in the Prym locus of a (possibly) ramified double cover. This improves the already known bounds. The idea is to adapt the techniques introduced by the authors in collaboration with A. Ghigi and G. P. Pirola for the Torelli map to the case of the Prym maps of (ramified) double covers.


Introduction
The purpose of this paper is to improve the estimates obtained in [4], [5] on the maximal dimension of a germ of a totally geodesic submanifold of the moduli space of polarised abelian varieties of a given dimension, which is contained in the Prym locus of a (possibly) ramified double cover. The idea is to adapt to the Prym case the technique developed in [6] and [9] to give a bound on the maximal dimension of a germ of a totally geodesic submanifold contained in the Torelli locus (see also [11]).
Denote by R g,r the moduli space of isomorphism classes of triples [(C, α, R)] where C is a smooth complex projective curve of genus g, R is a reduced effective divisor of degree 2r on C and α is a line bundle on C such that α 2 = O C (R).
This defines the Prym variety P (C, α, R) which is the connected component containing 0 of the kernel of the norm map N m π : JC→JC. If r > 0, ker N m π is connected. The Prym variety P (C, α, R) is an abelian variety of dimension g − 1 + r endowed with a polarization Ξ, which is induced by restriction of the principal polarisation on JC. The polarisation Ξ is of type δ = (1, . . . , 1, 2, ..., 2 g times ) for r > 0, while if r = 0, 1 it is twice a principal polarisation and we endow P (C, C) with this principal polarisation. Denote by A δ g−1+r is the moduli space of abelian varieties of dimension g − 1 + r with a polarization of type δ. The Prym map P g,r : R g,r → A δ g−1+r is defined as follows: P g,r ([C, α, R]) := [(P (C, α, R), Ξ)].
The map P g,r is generically finite, if and only if dim R g,r ≤ dim A δ g−1+r , and this holds if: either r ≥ 3 and g ≥ 1, or r = 2 and g ≥ 3, r = 1 and g ≥ 5, r = 0 and g ≥ 6. If r = 0 the Prym map is generically injective for g ≥ 7 ( [10], [13]). If r > 0, the works of Marcucci and Pirola [16], Marcucci and Naranjo [15] and Naranjo and Ortega [18] show the generic injectivity in all the cases except for r = 2, g = 3, which was previously studied in [17], [2] and for which the degree of the Prym map is 3. Recently a global Prym-Torelli theorem was proved for all g and r ≥ 6 ( [12] for g = 1, [19] for all g).
In [7] an analogous question to the Coleman Oort conjecture for the Prym maps P g,r for r = 0, 1 was formulated. Namely, the authors asked whether there exist Shimura subvarieties (hence totally geodesic subvarieties) of A δ g−1+r (r = 0, 1) generically contained in the Prym loci.
In [7] and in [8] examples of Shimura curves generically contained in the (possibly ramified) Prym loci have been constructed for low values of g and the computations of such examples suggest that they should not exist if g is sufficiently high.
Assume that dim R g,r ≤ dim A δ g−1+r , so that the differential of the Prym map is generically injective and denote by R 0 g,r the open subset of R g,r where the differential of the Prym map is injective. In this paper we give an estimate on the maximal dimension of a germ of a totally geodesic submanifold generically contained in the Prym loci, passing through a point [(C, α, R)] ∈ R 0 g,r in terms of the gonality k of the curve C. The results are summarised the following Theorem and they improve the estimates given in [4,Theorem 3.2] and [5,Theorem 3.2].
where C is a curve of genus g > 0. Denote by k its gonality and assume that C has no involutions and that g + r ≥ k + 3. Denote by Y a totally geodesic subvariety contained in P g,r (R 0 g,r ) and passing through P g,r ([C, α, R]).
(1) If r > k + 1, then From this, using that the gonality k ≤ [ g+3 2 ], we deduce the following result which improves the estimates obtained in [4,Theorem 3.4] , [5,Theorem 3.4]. Theorem 1.2. (Corollaries 5.3, 6.3). Let Y be a germ of a totally geodesic submanifold of A δ g−1+r which is contained in P g,r (R 0 g,r ), with g ≥ 3. Assume that there exists a point [C, α, R] ∈ Y such that C has no involutions. ( The improvements in the above estimates are obtained using ideas introduced in [9] that allow to use more than one quadric to compute the second fundamental form of the Torelli map. Here we need to adapt this technique to the more complicated case of the Prym maps. In the cases in which the assumptions of this theorem do not hold (i.e. for low values of g and r), we still have the estimates done using only one quadric (Theorem 7.2), as in [4] and [5], that are improved here thanks to a more careful study of the base locus of the linear system M : The result is the following Theorem 1.3. (Theorem 7.2). Let Y be a germ of a totally geodesic submanifold of A δ g−1+r which is contained in P g,r (R 0 g,r ).
Finally, in the case r ≥ 4, in Proposition 7.3 we get a better estimate than the one in Theorem 1.1, if the curve C is hyperelliptic, which is possible since a global Prym-Torelli theorem is proved in [19], [12].
The structure of the paper is the following: In Section 2 we introduce the notation and recall the results on the second fundamental form of the ramified Prym map obtained in [5].
In Section 3 we do a careful study of the linear system M := K C ⊗ α ⊗ F −1 , where F is the g 1 k on the curve C. This technical part is crucial to get the estimates. In Section 4 we explain the technique introduced in [6] and in [9] and adapt it in the case of the Prym-canonical linear system K C ⊗ α, to construct certain quadrics containing the Prym-canonical image of the curve C, where we are able to compute the second fundamental form of the Prym map.
In Section 5 we prove Theorems 1.1, 1.2 for r ≥ 5. In Section 6 we prove Theorems 1.1, 1.2 for r ≤ 4. In Section 7 we do the estimate using only one quadric in the missing cases, that is when r ≤ 4 and for low values of g. Finally we prove Proposition 7.3, in which we allow C to be hyperelliptic if r ≥ 4.

Second fundamental form of the Prym map
In this section we recall the results contained in [5,Section 2]. Denote by C a smooth complex projective curve of genus g, let R be a reduced divisor of degree 2r on C and α a line bundle on C such that α 2 = O C (R). To such data we can associate a double cover π :C → C branched on R. The Prym variety P (C, α, R) associated with this data is the polarised abelian variety given by the connected component containing the origin of kernel of the norm map N m π : JC → JC. For r > 0 the kernel of the norm map is connected. For r > 1 the polarisation Ξ is the restriction to P (C, α, R) of the principal polarisation ΘC on the jacobian ofC. For r = 0, 1 the polarisation ΘC |P (C,α,R) is twice a principal polarisation Ξ and we endow P (C, α, R) with this principal polarisation.
The Prym map P g,r : R g,r → A δ g−1+r , is the map that associates to a point [(C, α, R)] ∈ R g,r the isomorphism class of its Prym variety P (C, α, R) with the polarisation Ξ.
The dual of the differential of the Prym map P g,r at a generic point [(C, α, R)] is given by the multiplication map (1) (dP g,r ) * : S 2 H 0 (C, K C ⊗ α)→H 0 (C, K 2 C (R)) and it is surjective at the generic point if either r ≥ 3 and g ≥ 1, or r = 2 and g ≥ 3, r = 1 and g ≥ 5, r = 0 and g ≥ 6 (see [14]).
In these cases we denote by R 0 g,r the non empty open subset of R g,r where the Prym map P g,r is an immersion. Notice that if r ≥ 3, a global Torelli theorem is proved in [19] (and in [12] for g = 1), hence R 0 g,r = R g,r . Let us consider the (orbifold) tangent bundle exact sequence of the Prym map We endow A δ g−1+r with the orbifold metric induced by the symmetric metric on the Siegel space H g−1+r . The dual of the associated second fundamental form with respect to the metric connection of the above exact sequence is a map In [5] a description of this second fundamental form is given in terms of the second fundamental form of the Torelli map of the covering curvesC. Denote by U the open subset of R 0 g,r where there is a universal familyf :C → U and the differential of the modular map U →Mg, At the point b 0 := [(C, α, R)] ∈ U corresponding to the 2 : 1 cover π :C→C, the space P * g,r Ω 1 , and the dual of the exact sequence (2) at the point b 0 becomes The dual of the second fundamental form of the Prym map at the point b 0 is a map (4) ρ P : ) Denote by I 2 (KC) the kernel of the multiplication map S 2 H 0 (KC )→H 0 (2KC ) and by I 2 (KC ) + its invariant subspace by the action of the involution onC. Observe that we have the inclusion

Gonality
Let [C, α, R] ∈ R 0 g,r , where C is a curve of genus g > 0. Denote by k its gonality.
Hence h 0 (M ) ≥ 2, since g + r ≥ k + 3. The Clifford index of C is either k − 2 (computed by F ) or k − 3. Denote by B the base locus of M and set b := deg(B). The line bundle M has degree equal to 2g − 2 + r − k.
(3) If r = k+2, and D is an effective divisor of degree s such that h 0 (M (−D)) = h 0 (M ) − 1 = n, then s ≤ 2. Moreover s = 2 if and only if either C is hyperelliptic and F ∼ = O C (D), or |M | gives a birational map that contracts D.
In the case s = 2, either C is hyperelliptic and F ∼ = O C (D), or |M | gives a birational map to its image that contracts D.
Corollary 3.2. If g ≥ 3, C is not hyperelliptic, g + r ≥ k + 3, and r > k + 1, then the morphism given by |M | is birational onto its image.
Proof. The proof follows from Proposition 3.1.
Let us now study the morphism given by |M − B|. Assume that g + r ≥ k + 4.
Proof. By Riemann Roch we have Corollary 3.6. If g + r ≥ k + 4, r ≤ k + 1, b + r ≥ 5, then s 1 ≤ 2, hence the morphism induced by |M (−B)| is either birational on its image, or it has degree 2 on its image.
Proof. By Proposition 3.5 we have We want to study the linear system |M (−B − D 1 )| and give sufficient conditions ensuring that the map associated with |M (−B − D 1 )| has at most degree 2 on its image. Assume that g + r ≥ k + 5. We have the following Proposition 3.7. Assume g + r ≥ k + 5, b + r ≤ 4, r ≤ k + 1. With the above notation, assuming s 1 ≥ 3, let D 2 be an effective divisor of degree s 2 > 0 such that if and only if Corollary 3.8. If g + r ≥ k + 5, b + r = 4, r ≤ k + 1, then the morphism induced by |M (−B)| is either birational or it has degree 2 on its image.
Proof. By Proposition 3.7 we know that s 2 ≤ 2, hence the morphism induced by |M (−B − D 1 )| is either birational, or it has degree 2 on its image, hence the same holds for the morphism induced by |M (−B)|. Now we repeat the strategy explained above for the cases b + r ≤ 3. Set D l an effective divisor of maximal degree s l , l = 1, ..., 5, such that h 0 (M (−B − D 1 − ... − D l )) = n + 1 − l ≥ 2. This holds if g + r ≥ k + 3 + l. Proposition 3.9. Assume that g + r ≥ k + 4 + m, b + r ≤ 5 − m, r ≤ k + 1, m ≥ 1. Assume s 1 , ..., s m ≥ 3, let D m+1 be an effective divisor of degree s m+1 > 0 such that if and only if

Quadrics and estimate on the second fundamental form
In this section we explain how to construct certain quadrics contained in I 2 (K C ⊗ α) on which we are able to compute the second fundamental form of the Prym map ρ P . Here we adapt the techniques introduced in [6] and [9] for the Torelli map to the case of the Prym maps.
4.1. The quadrics and the second fundamental form. With the above notation, assume that the morphism f induced by |M (−B)| is birational on its image.
and fix now and for all two indipendent sections x, y ∈ H 0 (F ). Take a section t ∈ H 0 (M (−B)) and consider the associated divisor D(t) ∈ |M (−B)|. Assume that t is such that: (1) D(t) ∩ (Z ∪ B) = ∅, where Z is the ramification divisor of the map φ x,y : . Consider the quadric Q s = xt ⊙ ys − xs ⊙ yt Clearly Q s ∈ I 2 (K C ⊗ α), ∀s ∈ W . Denote by π :C→C the 2 : 1 cover given by α, and by ψ : C→P 1 the morphism induced by s, t . Then clearly {p 1 , ..., where the last equality follows from (5).
The following result has been proved in [5, (3.2),(3. 3)], with the variant that all the quadrics Q s , s ∈ W , are taken into account.

4.2.
The zero locus of the quadrics and the estimate. For an element z ∈ V , write z = d i=1 z i v i and denote by [z] := [z 1 , ..., z d ] ∈ P d−1 ∼ = P(V ). Consider the locus First we have the following Proof. Notice that by the uniform position (see e.g. [1, Ch.3]), we know that for all i ∈ {1, ..., n}, there exists exactly a section s i ∈ W such that s i (p j ) = 0, ∀j ∈ {1, ..., n}, j = i, s i (p i ) = 0, s i (p k ) = 0, ∀k > n, hence by Theorem 4.1 we get We need to estimate the dimension of a linear space Π ⊂ Z. Denote by T the linear subspace of V corresponding to Π.
We have the inclusion Since β(s n ) has rank d − n + 1, dim(h(V )) = d − n + 1 and h : T −→ V is injective, we get: In this section we apply Proposition 4.3 in the case b + r ≥ 5. We have the following where C is a curve of genus g > 0. Denote by k its gonality and assume that C has no involutions and that g + r ≥ k + 3. Denote by Y a totally geodesic subvariety contained in P g,r (R 0 g,r ) and passing through P g,r ([C, α, R]).
(1) If r > k + 1, then (2) If g + r ≥ k + 4, r ≤ k + 1, b + r ≥ 5, then Proof. Under the above assumptions in both cases, by Corollaries 3.2, 3.6, the linear system |M (−B)| induces a birational map on its image (in the first case B = ∅). Let S be the tangent space of Y at at P g,r ([C, α, R]). Let V be as above, then T = S ∩ V is a linear subspace where all the quadrics vanish, so by Proposition 4.3 we get: In case (1) . Then we conclude, since b ≤ 5, by Corollary 3.4.

Remark 5.2.
In the second case of Theorem 5.1, since b + r ≥ 5, we can apply Lemma 3.3 and we have the following cases: Let Y be a germ of a totally geodesic submanifold of A δ g−1+r which is contained in P g,r (R 0 g,r ), with g ≥ 3. Assume that there exists a point [C, α, R] ∈ Y such that C has no involutions.
6. The case b + r ≤ 4 Let us now consider the case b + r ≤ 4. We have the following Theorem 6.1. Let [C, α, R] ∈ R 0 g,r , where C is a curve of genus g > 0, denote by k its gonality. Assume that C has no involutions. Denote by Y a totally geodesic subvariety contained in P g,r (R 0 g,r ) and passing through P g,r ([C, α, R]). If g + r ≥ k + 4 + m, r ≤ k + 1, b + r = 5 − m, m ≥ 1, then Proof. By Corollaries 3.8, 3.10, the linear system |M (−B)| is birational on its image. Let S be the tangent space of Y at at P g,r ([C, α, R]). Let V be as above, then T = S ∩ V is a linear subspace where all the quadrics vanish, so by Proposition 4.3 we get: Corollary 6.2. Let Y be a germ of a totally geodesic submanifold of A δ g−1+r which is contained in P g,r (R 0 g,0 ), with g ≥ 4m, r = 5 − m, m ≥ 1. Assume that there exists a point [C, α, R] ∈ Y such that C has no involutions.Then dim Y ≤ 2g − m + 5 if g is even, dim Y ≤ 2g − m + 6 if g is odd.
In conclusion we have the following estimates: Corollary 6.3. Let Y be a germ of a totally geodesic submanifold of A δ g−1+r which is contained in P g,r (R 0 g,0 ), with r ≤ 4. Assume that there exists a point [C, α, R] ∈ Y such that C has no involutions.

Estimate with one quadric
In the cases in which the assumptions of corollaries 5.3, 6.2 do not hold, we still have the estimates obtained in [4] and [5] using only one quadric, which can be improved by Lemma 3.3 and Corollary 3.4 in the case r ≤ k + 1.