Progress in Mathematics Volume 177 Series Editors Hyman Bass Joseph Oesterlé Alan WeinsteinChristina Birkenhake Herbert Lange Complex Tori Birkhauser Boston ¢ Basel * BerlinHYSICS LIBRARY Christina Birkenhake Herbert Lange Fakultat fiir Mathematik und Physik Mathematisches Institut Universitit Bayreuth Universitat Erlangen-Niimberg D-95440 Bayreuth, Germany D-91054 Erlangen, Germany Noor YN Come Library of Congress Cataloging-in-Publication Data Birkenhake, Christina. Complex tori / Christina Birkenhake, Herbert Lange. p. _ cm.— (Progress in mathematics ; v. 177) Includes bibliographical references and index. ISBN 0-8176-4103-3 (alk. paper). — ISBN 3-7643-4103-3 (alk. Paper) 1. Complex manifolds, 2. Torus (Geometry) 1. Lange, H. (Herbert), 1943- . I. Title. III. Series; Progress in mathematics (Boston, Mass.) ; Vol. 177. QA613.B455 1999 514.3—de21 99-32326 cIP AMS Subject Classifications: 32J18, 32G20, 14K30, 32M05, 14M17 Printed on acid-free paper. cage © © 1999 Birkhauser Boston Birkhduser All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Birkhauser Boston, c/o Springer-Verlag New York, Inc., 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use of general descriptive names, trade names, trademarks, etc., in this publication, even if the former are not especially identified, is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act, may accordingly be used freely by anyone. ISBN 0-8176-4103-3 SPIN 19901498 ISBN 3-7643-4103-3 Formatted from authors’ files by Te Xniques, Inc., Cambridge, MA. Printed and bound by Hamilton Printing Company, Rensselaer, NY. Printed in the United States of America.Contents Introduction 1 Complex Tori rPOMmNanaRwone = Ne 1 2 3 4 5 N1O® Homomorphisms of Complex Tori. ............. Line Bundles 2... 2. 2 ee ee The Néron-Severi Group .. . The Dual Complex Torus. . . Extensions of Complex Tori Complementary Subtori and Shafarevich Extensions . Simple and Indecomposable Complex Tori ...... The Endomorphism Algebra of a Complex Torus. . . : The Theorem of Oort and Zarhin .............6. The Space of all Complex Tori of Dimensiong ....... jondegenerate Complex Tori Polarizations of Indexk «6.1... 5 ee eee eee Moduli Spaces of Nondegenerate Complex Tori... . . The Rosati Involution ©... 2... 1 eee eee The Dual Polarization ©... 01... eee eee eee Poincaré’s Reducibility Theorem for Nongenerate Complex Tori... 6... ee The Algebraic Dimension... . 2.2.2.2... 00004 Picard Number and Algebraic Dimension of Complex Tori of Dimension2 ... 2.2... 22. eeevi Contents Embeddings into Projective Space 63 1 Kahler Theory of Line Bundles on Complex Tori... .. . 64 2 Harmonic Forms with Values in a Nondegenerate Line Bundle... 1... 2 ee 70 3 Maps Associated to a Nondegenerate Line Bundle . . 74 4 The Family of Abelian Varieties Associated to a Nondegenerate Complex Torus.......--...-... 76 5 Some Properties of the Diffeomeorphism yy_:X + Xy_ . 83 6 The Rational C°-map yry_:X--+Pw ...-..---. 89 Intermediate Jacobians 91 1 Primitive Cohomology of Kahler Manifolds ..... . . 92 2 The Griffiths Intermediate Jacobian. ............ 94 3 Some Properties of the Griffiths Intermediate Jacobian ........... 100 4 The Weil Intermediate Jacobian... . . . - 106 5 The Lazzeri Intermediate Jacobian 110 6 The Abelian Variety Associated to the Griffiths Intermediate Jacobian... 2.2.6... ee ee ee eee 113 Families of Complex Tori 115 1 Complex Tori with Endomorphism Structure. ....... 116 2 The Index of a Complex Torus with Endomorphism Structure 2... eee 119 3 Complex Tori with Endomorphism Structure of Type Ia . 127 4 Complex Tori with Endomorphism Structure of Type Ib . 139 5 Complex Tori with Endomorphism Structure of Type II. . 160 The Parameter Spaces of Complex Tori with Endomorphism Structure 179 1 The Space Hyk - . : . . 180 2 The SpaceK,... . 184 3 The Space Ling 186 4 The Space Mn 189 5 The Space Nine . 192 6 The Space Rinj 195 7 ~~ The Space S, 196 8 TheSpaceJ, ... . 198 9 The Space Upa),y 10 The Space Vpg Moduli Spaces 2071 Moduli Problems ............. 0000000005 2 Homogeneous Spaces with Noncompact Stabilizer . . 3 The Moduli Space of all Complex Tori ........ 4 Moduli Spaces of Nondegenerate Complex Tori... ... . 5 Moduli Spaces of Nondegenerate Complex Tori with Endomorphism Structure. ..........-.-.- 220 The Kronecker Product 225 Anti-Involutions on R-Algebras 229 Complex Structures 233 1 Complex Structures in Spam(R) 6... eee eee 233 2 Complex Structures in O,,,(R) 235 3 Complex Structures in aU,,(H) bee . 236 4 Complex Structures in Um_¢i(H) .........-0-05. 238 References . 241 Glossary of Notations 245 Index 249Introduction A complez torus is a connected compact complex Lie group. Any complex torus is of the form X = C9/A, where A is a lattice in C9. A meromor- phic function on C9, periodic with respect to A, may be considered as a function on X. An abelian variety is a complex torus admitting suf- ficiently many meromorphic functions. ln other words, abelian varieties are exactly the algebraic complex tori. Thus abelian varieties are special complex tori. In fact, a general complex torus of dimension > 2 does not admit any meromorphic function. Whereas abelian varieties are very well investigated — they have been studied for almost two hundred years now — not much is known about arbitrary complex tori, although they are among the simplest complex manifolds. There are very few papers and no book dealing with nonalgebraic complex tori. On the other hand, as the following examples show, these tori frequently occur even if one starts within the category of abelian varieties: (i) Griffiths intermediate Jacobian of an abelian variety of dimension > 3 is a complex torus, which in general is not an abelian variety (see Section 4.2). (ii) Let X, and Xz be abelian varieties. There are uncountably many extensions 0 + X, — X — X2 — 0 as complex tori. Among them there are only countably many for which X is an abelian variety (see Section 1.6). Hence a general extension of abelian varieties is not an abelian variety. Apart from their importance for applications, for example in the theory of algebraic cycles via mtermediate Jacobians, not necessarily algebraicxii Introduction complex tori are interesting for their own sake, and this is the subject of the present book. The central topics are those properties of complex tori which are actually different from the corresponding properties of abelian varieties. The most important such topics are: () (2) (3) (4) (5) An important feature of abelian varieties is that any complex sub- torus admits a complement up to isogeny. This leads to Poincaré’s Reducibility Theorem. For complex tori this is not valid any more. It is easy to construct complex tori admitting only one nontrivial complex subtorus, so one has to distinguish between simple and indecomposable complex tori (see Section 1.7). It is a consequence of Poincaré’s Reducibility Theorem that the endomorphism algebra Endg(X) of an abelian variety X is semi- simple. This is not the case for an arbitrary complex torus X. We will show, however, that the semisimplification Endg(X)s5 of Endg(X) decomposes in the same way as Endg(X) for abelian varieties (see Section 1.8). The positivity of the trace form yields strong restrictions for a finite dimensional Q-algebra to occur as the endomorphism algebra of an abelian variety. The theorem of Oort-Zarhin says that on the other hand, for any finite dimensional Q-algebra A there is a complex torus X with Endg(X) = A. We present a proof, due to Hétte- Scharlau, in Section 1.9. A polarization of an abelian variety X = C?/A is a positive definite hermitian form H on C9, whose imaginary part is integer valued on the lattice A. A complex torus which is not an abelian variety does not, by definition, admit a polarization. But there is a replacement for this: The index of a nondegenerate hermitian form is the number of negative eigenvalues. If, instead of being positive definite, the her- mitian form H is nondegenerate of index k, we call H a polarization of inder k. The pair (X, H) is called a nondegenerate complex torus of index k. \n these terms polarized abelian varieties are nondegen- erate complex tori of index 0. The investigation of nondegenerate complex tori is the subject of Chapter 2. The restriction of a polarization of an abelian variety to a complex subtorus is also a polarization. This is not valid for polarizations of positive index. It is shown in Section 2.5, however, that if (X, H) is a nondegenerate complex torus of index k, and Y is a complex subtorus such that H|Y is nondegenerate, then Y admits a com- plementary complex subtorus Z. Moreover, H|Z is nondegenerate.(6) (7) (8) Introduction xiii This leads to a modification of Poincaré’s Reducibility Theorem: Call a nondegenerate complex torus (X,H) irreducible if X does not admit any nontrivial subtorus Y, such that H|Y is nondegen- erate. Any nondegenerate complex torus is isogenous to a product of irreducible nondegenerate complex tori (see Section 2.5). Abelian varieties are exactly those complex tori which admit a holo- morphic embedding into some projective space. Thus a complex torus X that is not an abelian variety does not admit a holomor- phic projective embedding. However, if X is of dimension g and H a polarization of index k on X, then X admits a differentiable embed- ding into projective space which is holomorphic in g — k variables and antiholomorphic in k variables (see Chapter 3). In fact, choose a line bundle L on X with first Chern class H. Then H*(X, L) is the only nonvanishing cohomology group of L. The vector space H*(X, L) may be considered as the vector space of harmonic forms of type (0,4) with values in L. Choosing a suitable metric of L, these forms define a differentiable map y, : X --» Py on an open dense set of X, holomorphic m g—k variables and antiholomorphic in k variables. An analogue of a theorem of Lefschetz says that yz» is an embedding for n > 3 (see Section 3.6). Let (X = C/A, H) be a nondegenerate complex torus of index k. If di,...,dg denote the elementary divisors of the alternating form Im H on A, the polarization H is called of type (di,..., dg). A symplectic marking of (X, H) is a basis of the lattice A symplec- tic with respect to Im H. We show in Section 7.4 that there ex- ists an analytic fine moduli space H,, for nondegenerate complex tori of index k and type (di, ...,dg) with symplectic marking. Note that Ho and H,,, are just the Siegel upper and lower half space. The group Spo9(Z) acts on Hs in such a way that the quotient Hx /Spog(Z) parametrizes nondegenerate complex tori of index k and type (di,...,dg) (see Section 2.2). Moreover, for k = 0 and g, the quotient H,4/Sp29(Z) is an analytic coarse moduli space for abelian varieties of type (di,...,dg). For 1 < k < 9, Hgx/Spr9(Z) is only a topological coarse moduli space. The fact that Hy ./Spo9(Z) does not admit any complex structure (see Section 7.2) implies that an analytic coarse moduli space for nondegenerate complex tori of index k 4 0, g and type (di,...,d,) does not exist (see Section 7.4). The endomorphism algebra of a simple complex torus X is a skew- field of finite dimension over Q. As we mentioned above Endg(X) is completely arbitrary, whereas for an abelian variety there arexiv (9) (19) Introduction strong restrictions. However, the existence of a polarization H of index k on X gives some restrictions on Endg(X). Similarly as in the case of abelian varieties, the hermitian form H induces an anti- involution on Endg(X), the Rosati-involution. The skew-fields F of finite dimension over Q with anti-involution ’ were classified by Albert. There are roughly three types (see Chapter 5): Type Ia: F is an algebraic number field and ’ is the identity. Type Ib: F is a quaternion algebra over an algebraic number field and ’ is conjugate to the canonical involution on F. Type II: F is of degree d? over an algebraic number field K and ’ restricts to an involution on K. Shimura defined the notion of endomorphism structure of an abelian variety. He constructed, for any skew-field of finite dimension over Q with positive anti-involution, families of abelian varieties with corresponding endomorphism structure. Moreover he showed that every such abelian variety is contained in one of these families. The notion of endomorphism structure immediately generalizes to non- degenerate complex tori of mdex k. In Chapter 5 we construct, for any of the above mentioned skew-fields F' with anti-involution ’, families of nondegenerate complex tori with endomorphism struc- ture in (F,‘) such that any such complex torus is contained in one of these families. Let P denote the parameter space of one of the families of nondegen- erate complex tori just mentioned. In the case of abelian varieties there are three series of irreducible hermitian symmetric spaces of the noncompact types, namely C I, A III, and D III, such that any P is a product of members of these families. In the case of nondegenerate complex tori, P need not be a hermitian symmetric space anymore. Let G denote a real form of a connected complex semisimple Lie group Gc. A flag domain for G is an open G-orbit in a complex flag manifold for Ge. In Chapter 6 we show that there are ten series of flag domains for classical groups such that any parameter space is a product of their members. We shall see that, conversely, for every classical group except GLon4i(R), Oop4iq(R) and O2,41(C) there is a flag domain parametrizing nondegenerate complex tori with endomorphism structure. In Section 7.5 we show that the spaces P are analytic fine moduli spaces for a suitable moduli problem. Let P = H\G be one of the above parameter spaces of nonde- generate complex tori with endomorphism structure. There is anIntroduction xv arithmetic group I’ acting on P such that two points of P represent isomorphic nondegenerate complex tori with endomorphism struc- ture if and only if they differ by an element of I’. In order to see which of the quotient spaces P/T are coarse moduli spaces of the corresponding moduli problem, we have to distinguish two cases: If H is a compact group, P/T is an analytic coarse moduli space for the corresponding moduli functor. If H is noncompact, P/T is only a topological coarse moduli space. Moreover, in this case P/T’ does not admit the structure of a complex analytic space, so there does not exist an analytic coarse moduli space for the corresponding moduli functor (see Section 7.5). As for prerequisites, we freely use the basic language of algebraic geom- etry and complex analysis as outlined for example in [GH]. The first three chapters of our book [CAV] deal with complex tori. In the first sections of Chapter 1 we present the results of these chapters that are needed subsequently without repeating proofs. Most of the main results presented in this book seem to be new. We would, however, like to emphasize the strong influence of the papers [OZ1], [022] of Oort and Zarhin and [Sh] of Shimura. Not new of course are the Theorem of Oort-Zarhin (see Section 1.8) for which we copied the elegant proof of Hétte-Scharlau (see [HS]) and Chapter 4 in which we present the intermediate Jacobians of Griffiths, Weil and Lazzeri as the most important examples for nondegenerate complex tori. Finally we included some well-known elementary results on complex tori and their moduli spaces. Finally a few words about our cross-references. If we refer to Theorem 3.5 in Chapter 6, we write Theorem 6.3.5. If we are within Chapter 6, we omit the numbers of the chapter. Similarly, Section 7.2 means Section 2 of Chapter 7. Proposition A.6 stands for Proposition 6 of Appendix A. Acknowledgments. It is a pleasure to acknowledge the help and support we received from a number of people and institutions. In particular we would like to thank W.-D. Geyer, F. Knop, K.-H. Neeb, and J. Wilhelm for many comments and suggestions. We are very grateful to Mrs. G. Schramm who typed the TEX file. We received financial support from the European Community via the HCM project AGE (contract number ERBCH RXCT 940557).1 Complex Tori A lattice in a complex vector space C? is by definition a discrete subgroup of maximal rank in C®. It is a free abelian group of rank 2g. A compler torus is a quotient X = C9/A with A a lattice in C9. A complex torus is a complex manifold of dimension g. It inherits the structure of a complex Lie group from the vector space C9. In this chapter we study some properties of complex tori without any additional structure. In the first four sections we present the basic notions and results on complex tori. For most of the proofs we refer to [CAV], Chapters 1-3. In particular we introduce the notions of homomorphism, isogeny, rational and analytic representation, the Néron-Severi group and dual complex torus. We quote the Appell-Humbert Theorem and the Riemann-Roch Theorem. In Section 5 the group of extensions Ext(X2, X1) of a complex torus X2 by a complex torus Xj is determined in terms of period matrices. A complement of a subtorus Y of a complex torus X is asubtorus Z of X such that the addition map Y x Z — X is an isogeny. In his book [Sha] Shafarevic gave an example of an extension 0 —- EF, - X + FE, + 0 of elliptic curves such that E, does not admit a complement in X. Oort and Zarhin studied in [OZ2] this type of extensions of elliptic curves and called them Shafarevic extensions. Note that according to Poincaré’s Re- ducibility Theorem, a Shafarevic extension cannot be an abelian variety. In Section 6 we generalize this notion to extensions of arbitrary complex tori. It turns out that for any complex tori X, and X2, an extension e € Ext(X;, X2) is a Shafarevic extension if and only if it is not a torsion element in the group Ext(X, X2). This implies that almost all extensions of X2 by X, are Shafarevic extensions. Another consequence is that one2 1. Complex Tori has to distinguish between simple and indecomposable complex tori. In Section 7 we show that a complex torus X is indecomposable if and only if the endomorphism algebra Endg(X) is a local Q-algebra, and we prove the Krull-Schmidt Theorem for complex tori. In contrast to the case of abelian varieties the endomorphism algebra of a complex torus need not be semisimple. Let Endg(X),. denote the semisimplification of Endg(X). We show in Section 8 that an isogeny X — Xj? x --- x X?* induces an isomorphism Endg(X)ss ~ Mn, (Endg(X1)s0) x +++ x My, (Endg(X;).s); which may be considered as an analogue of Poincaré’s Reducibility The- orem. In Section 9 we give the proof of Hétte-Scharlau of the theorem of Oort-Zarhin saying that any algebra of finite type over Q occurs as the endomorphism algebra of a complex torus. Finally in Section 10 we con- struct a parameter space for all complex tori. It will be used in Chapter 7 to construct the moduli space of all complex tori. 1 Homomorphisms of Complex Tori A compler torus X of dimension g is by definition a connected compact complex Lie group of dimension g. Let V := ToX be the tangent space of X at the unit element 0 € X and +: V — X the exponential map. According to [CAV] Lemma 1.1.1, is surjective and its kernel A is a lattice in V. Hence there is an isomorphism X2V/A. Conversely, given a finite dimensional complex vector space V and a lat- tice A C V, acting on V by translations, the quotient V/A is a complex torus. Complex tori will always be considered as such quotients. In order to describe a complex torus X = V/A choose bases e1,...,€y of V and A1,...; Avg of the lattice A. Writing A; = Va Ajie; the matrix Ay «+ Atay = : i € M(g x 2g,C) Aad s+ Agag is called a period matrix for X. The period matrix determines the complex torus X completely: X ~ C9/T1Z9. Certainly this representation depends on the choice of the bases. A matrix TI € M(g x 29,C) is the period matrix for some complex torus X if and only if the square matrix (#) € Mzg(C) is nonsingular (see [CAV] Proposition 1.1.2).1. Homomorphisms of Complex Tori 3 Proposition 1.1. Let X be a complex torus of dimension g. Then X admits a period matrix of the form (r,1,) with r € M,(C) such that det(Im7) # 0. Conversely every such matrix is the period matrix of a complex torus. Proof. Let Tl € M(g x 2g,C) be any period matrix of X. Then TI is of rank g. Hence there are indices 1 < i; < +++ < ig < 2g such that the corresponding minor IT;,,._,, € M,(C) is invertible. Changing if necessary the basis of the lattice IIZ”9 (i.e., permuting the columns of IT properly) we may assume that the minor IT941,..,29 is invertible. Now changing the basis of C? by Tj, 29 the new petiod matrix is of the form (r, 1). By what we have said above rT .1g\_ T-F 1g \ _poyg 0 # det ( 7 ) = det ( o L ) = (2%)% det(Imr). The converse implication is obvious. a Remark 1.2. A period matrix II € M(g x 2g,C) may be interpreted as an R-linear isomorphism II : R?? — C9. This isomorphism induces a complex structure J on R* : J is the R-linear automorphism of R?? fitting into the commutative diagram rR» =, o Ji Lil, Re 4, ce. Conversely, any complex structure J on R”? defines a uniquely determined complex torus: Let (7%, J) denote the complex vector space associated to J. Then X3 = (R’, J)/29 is a complex torus of dimension g. The matrix of any isomorphism R79 5 C9 is a period matrix for Xj. For a more detailed discussion of the rela- tionship between IT and J see Section 3. Let X’ = V’/A' be a second complex torus. A homomorphism f : X X’ is a holomorphic map, compatible with the group structures. The translation by an element to € X is defined to be the holomorphic map try : X + X, r++ 4429. Any holomorphic map is the composition of a homomorphism and a translation: for a holomorphic map h : X — X' there is a unique homomorphism f : X — X’ such that h=trof4 1. Complex Tori (see [CAV] Proposition 1.2.1). The homomorphism f lifts to a uniquely determined C-linear map F : V > V’. This gives an injective homomor- phism of abelian groups ~ Pa: Hom(X, X') + Home(V,V’) , fre F, called the analytic representation of Hom(X, X’). Its restriction F|A is Z-linear and determines F and f completely. Thus we get an injective homomorphism pr: Hom(X, X") + Hom,(A,A"), f 4 FIA, called the rational representation of Hom(X, X’). The extensions of pg and p, to Homg(X, X') := Hom(X, X') @z Q are denoted by the same letters. If X = X’, then p, and p, are represen- tations of the rings End(X) := Hom(X, X) and Endg(X) := End(X) @z Q. Suppose II € M(g x 2g,C) and II’ € M(q’ x 2g’,C) are period matrices for X and X'. With respect to the chosen bases, p,(f) can be considered as a matrix in M(g’ x g,C) and p,(f) can be considered as an element of M(2g’ x 2g,Z). The matrices pa(f) and p,(f) fit into the following commutative diagram Zs Lo o(f) 4 t pal) zy & ee, i.e., by the equation pa( f)I = W'p,(f). (1) Conversely, any two matrices A € M(g' x g,C) and R € M(2g! x 2g, Z) satisfying ATI = IIR define a homomorphism X > X’. Remark 1.3. For period matrices II = (7,1,) and I’ = (7’,1y) as in Proposition 1.1. equation (1) is equivalent to ager ( 24) =o. For the proofs of the following two propositions we refer to [CAV] Section 1.2.2. Line Bundles 5 Proposition 1.4. The extension of the rational representation p, of Endg(X) to Endg(X) @g C is equivalent to the direct sum of the an- alytic representation and its complex conjugate: Pr @ ide ~ pa B Py. Proposition 1.5. Suppose f : X — X' is a homomorphism of complex tori. a) The image imf is a complex subtorus of X'. b) The kernel ker f is a closed subgroup of X. The connected component (ker fo containing 0 is a complex subtorus of X of finite index in ker f. A surjective homomorphism f : X — X' with finite kernel is called an isogeny. Define the exponent e = e(f) of the isogeny f to be the exponent of the finite group ker f, i.e., the smallest positive integer n such that n-z =O for all x € ker f. Proposition 1.6. For any isogeny f : X — X' of exponent e there exists an isogeny g : X' + X, unique up to isomorphisms, such that gf = e idx and fg = eidy. For the proof see [CAV] Proposition 1.2.6. oO Hence isogenies define an equivalence relation on the set of complex tori. Two complex tori are called isogenous if there is an isogeny between them. An element of End(X) is an isogeny if and only if it is invertible in Endg(X). For a homomorphism f : X — X’ we define the degree to be _f ord(kerf) if kerf is finite, deg f: { 0 otherwise. If moreover f is an endomorphism of X, then deg f = det p,(f). (2) Example 1.7. For any integer n # 0 the homomorphism nx : X > X, 2 ++ nz is an isogeny of degree 79. Its kernel X, & (Z/nZ)*" is called the group of n-division points of X. 2 Line Bundles Let X = V/A be a complex torus. In this section we describe the group Pic(X) = H}(O%) of holomorphic line bundles on X.6 1. Complex Tori The exponential sequence 0 + Z— Ox — O} — 1 yields the long exact cohomology sequence 1 HK, Z) > (Ox) + H(O%) HPUX.) The image c,(L) € H?(X,Z) of a line bundle L € H1(O%) is called the first Chern class of L. Define the Néron-Severi group NS(X) to be the image of c, : H'(O%) — H?(X,Z). There are canonical isomorphisms of NS(X) with a) the group of hermitian forms H : V x V — C whose imaginary part takes integer values on the lattice A, and b) the group of alternating forms E € Alt?(A,Z) C Alt?(V,R) such that E(i-,-) is symmetric. (See [CAV] Proposition 2.1.6 and Lemma 2.1.7). Define Pic?(X) to be the kernel of c, : H!(O%) — NS(X). For further details on the Néron- Severi group see Section 3 below. Recall that two line bundles Z, and Ly on X are called analytically equivalent if there is a connected complex analytic space T, a ine bundle £ on X x T and points t,t, € T such that LIX x {ti} ~ L; for i = 1,2. It is easy to see that Pic?(X) is the group of line bundles on X which are analytically equivalent to 0 (see [CAV] Proposition 2.5.3). On the other hand the Néron-Severi group is the group of analytic equivalence classes of line bundles on X and (1) reduces to the short exact sequence 0 - Pic®(X) — Pic(X) % NS(X) = 0. The Appell-Humbert Theorem gives an interpretation of this exact se- quence. In order to state the theorem we need some notation: Let H : V x V —C be an element of NS(X) considered as a hermitian form on V. Denote by C; = {z € C||z| = 1} the circle group. A semicharacter for H is a map x: A — C, satisfying the functional equation X(A + 1) = x(A)x(werin FO) for all A, u € A. Given a semicharacter x for H define the canonical factor agzay 2 AX VC, az, v) 2 x(A)eFO FRO It is easy to see that Ora) Hox) = E+ Ha, x1x2)+2. Line Bundles 7 Hence multiplication defines a group structure on the set P(A) of canoni- cal factors aqz,,). By definition the semicharacters for the hermitian form 0 € NS(X) are the characters in Hom(A,C)) and aco.) = x for x € Hom(A, C,). Hence x + a(o,x) yields a canonical inclusion of Hom(A, C;) into the goup P(A). Moreover there is a canonical map P(A) — Pic(X) defined as follows: acy.) € P(A) induces an action of A on the trivial vector bundle V x C 4V Ax(V x€)>VxC, (A,(v,#)) 6 (ut AaGz)(A, 2) - t)- The quotient L(H,x) = V xC/A (2) is a holomorphic line bundle on X, the image of acz,,) in Pic(X). Appell-Humbert Theorem 2.1. The homomorphism P(A) > Pic(X) induces an isomorphism of exact sequences O- Hom(A,C:) + P(A) + NX) 40 | I= l= 0+ Pic(X) = Pic(X) 4 NS(X) 30. For a proof we refer to [CAV] Section 2.2 or [M], Section 1.2. a The construction (2) of the line bundle L = L(H,x) yields also a description of its holomorphic sections: given a section s € H°(X, L) its pull back via « : V — X is a holomorphic function f on V satisfying the functional equation F(v +A) = a¢e739(As 2) F(v) (3) for all v € V and \ € A. Conversely any holomorphic function on V satisfying (3) comes from a global section of L in this way. These functions are called canonical theta functions for the factor az.) (respectively for L). In the sequel we often consider the global sections of L as canonical theta functions. Let L = L(H,x) bea tine bundle on X and dim X = g. By the elemen- tary divisor theorem, the lattice A admits a basis with respect to which the alternating form F = ImH € Alt?(A, Z) is given by the matrix ( 4 p ) € M;,(Z) where D = diag(d;,...,d;,0...,0) with positive integers d|do|---|d, and t = rkH. The numbers d),...,d; are uniquely determined by L. The8 1, Complex Tori matrix D as well as the vector (d,,..., d¢,0..., 0) is called the type of the line bundle L. The radical of H is the group A(L)o = {v € V| H(v, V) = 0}. The group . K(L)o := A(L)o/A(L)o NA is a complex subtorus of X of dimension g —t. Theorem 2.2. Let L = L(H,x) be a line bundle on a complex torus X of dimension g whose first Chern class H has r positive and s negative eigenvalues. Then the dimension of H%(L) is Me die drgs = 8 SS g—-7 and L|K(L)o is trivial if 0 otherwise AN(L) = where (d1,..-;dr42,0...,0) as the type of L. (For the proof see (cav] Theorem 3.5.5). Notice that L|K(L)o is triv- ial if and only if the semicharacter x of L is constant = 1 on the lattice A(D)on A. a As a direct consequence of Theorem 2.2 we get for the Euler-Poincaré characteristic x(L) = )78_9(—1)”h"(L) of L: Analytic Riemann-Roch Theorem 2.3. Let L be a line bundle of type (d,...,dg) on X whose first Chern class H admits s negative eigenvalues. Then x() = (-1)fdy dy. In particular x(L) = 0 if and only if dy = 0, that is if the first Chern class H is degenerate. The geometric version of the Riemann-Roch The- orem expresses the Euler-Poincaré characteristic of L in terms of the self intersection number of L. Geometric Riemann-Roch Theorem 2.4. For any line bundle L on X For a proof see [CAV] Section 3.6. Q3. The Néron-Severi Group 9 3 The Néron-Severi Group Let X = V/A be a complex torus of dimension g. Recall the Néron-Severi group NS(X) = Pic(X)/Pic®(X) of X. Proposition 3.1. The Néron-Severi group NS(X) is a free abelian group of rank p(X) < g?. The rank p(X) of NS(X) is called the Picard number of X . Proof. The Néron-Severi group NS(X) = ¢;(H'(Ox)) is a subgroup of H?(X, Z) which is free abelian according to [CAV] Corollary 1.3.3. On the other hand it is well known that the image of c; in H?(X,C) = H?°(X)@ H*(X)® H°?(X) is contained in H™(X) which is of dimension g? (see [CAV] Theorem 1.4.1). a Proposition 3.2. The Picard number of a complex torus is invariant under isogenies. Proof. Suppose f : X — Y is an isogeny. Obviously the pull back homo- morphism f* : NS(Y) + NS(X) is injective. It is also surjective since f is invertible in Home(X,Y). a The following example shows that the upper bound for the Picard num- ber is sharp. For this, recall from Section 2 the following description of the Néron-Severi group NS(X)={H:VxV—>C|H hermitianand ImH(A x A) c Z}. Example 3.3. Let E be the elliptic curve C/(i, 1)Z? and X = E%. Then TI = (i1,,1,) is a period matrix for X. A hermitian matrix H € M,(C) is an element of NS(X) if and only if Im(*THT) = Im( thy) = ( 2Rch ft) € Mag(Z)- Hence NS(X) = {H € M,(Z+iZ)| H hermitian} which is a free abelian group of rank g? . Let H be a hermitian form in NS(X). Its imaginary part Im H deter- mines H uniquely. As mentioned in the previous section this gives another description of the Néron-Severi group of X: E alternating, E(Ax A) CZ, \ NS(X) = {e: VxV—R) ona BG.) symmetric10 1. Complex Tori According to Proposition 1.1 there are bases for V and A such that the period matrix for X is of the form II = (7,1,) with r € M,(C) such that detImr # 0. With respect to these bases the alternating forms E c NS(X) are given by alternating matrices in Mo,(Z). The follow- ing proposition allows us to compute the Néron-Severi group in many cases. Proposition 3.4. With respect to the chosen bases _ A B A and C alternating and ws(x) = {( -B c) € M2) A~Br+'r'B-+ ‘rr =0 }. If g = 2 write A = (°,3),B = (27),C = (2.§) andz = (71 3). Then (453) € NS(X) if and only ifa,...,f € Z and a+ erm — brig + fT — dt + edet7 = 0. For abelian surfaces this relation is due to Humbert (see [Hu]). Notice that for arbitrary dimension g the equation defining NS(X) can also be written in the following form ran( 4 )(2,)=0 For the proof we need the following lemma. The period matrix IT = (7, 1,) of X induces a complex structure J, on R¥ fitting into the commutative diagram Re 4g) ce Je | Jia, Re 9) (see Remark 1.2). Writing 7 = z+ iy, an immediate computation shows: -1 1 Lemma 3.5. J, = ( y a y _1 ) . ~y- ry re —ry Remark 3.6. For an arbitrary period matrix II = (u,v) of X the square matrix v is not necessarily invertible. But (7) is invertible and the com- plex structure Jy corresponding to II is n=e(B) (8 4) (8). Proof of Proposition 3.4. An alternating form EF: V x V — R given by an alternating matrix M = (4, 2) € Mz9(Z) is an element of NS(X)3. The Néron-Severi Group i if and only if E(i-,-) is symmetric. In terms of coordinates this means that ‘J,M = *MJ,. But tug, = ( TAy e+ Byt+ Baye —Ay?+ Bay? \ _ (a1 a2 7" \ By te+Cy+Cay 2 ‘By +Cay? J \ a3 a4)” On the other hand taking real and imaginary parts, equation A — Br + *r ‘B+ *rCr = 0 is equivalent to A= Br-'r'B+ *yCy- 'rCr (1) and 'y'B— ty'Cz = By — ‘xCy. (2) Hence it suffices to show that ‘MJ, = (@} 9?) is symmetric if and only if equations (1) and (2) are valid. It is easy to see that a4 is symmetric if and only if (2) holds. Suppose (8&2) is symmetric. We have to show that (1) holds. But Ay") — Bry" = -ag = —*ag = —'2(*y?B + ty? £0) — ty'C = —za4— ty'C or equivalently A = Br- '‘rayyt tyCy Br —'z'B— 'xCr+ 'yCy. So (1) holds. Suppose conversely (1) and (2) hold. Reading the previous computation upside down, (1) and (2) imply that a2 = ‘ag. Hence it remains to show that a) is symmetric. But a = (-Ay)+ Bry ")z + By = agz+ By = tagr + By = (tnty "B+ ty'C + ‘ty! 2'C)a + By = ‘rlagr t+ ty'Cr + By which is symmetric by (2). oO One can also consider Proposition 3.4 from a different point of view: Let I:= (4, “) denote the standard alternating form on R®. According to Proposition 1.1 any 7 € M,(C) with detImz # 0 yields a complex torus X, = (R®, J,)/Z?. One may ask, when is the standard alternating form I an element of the Néron-Severi group of X,?12-1. Complex Tori "E Corollary 3.7. I € NS(X,) if and only if r is symmetric. Proof. The equation for NS(X;) in Proposition 3.4 applied to the alter- nating form J reduces to —r + *r =0. o A square matrix A = (ai;) € My(C) is called backsymmetric if aj; = Qn41-jn4i-i for all 1 < i,j 2 the situation is different. The following example shows that in every dimension > 2 there are complex tori admitting only line bundles which are analytically equivalent to zero. Remark 3.11. Let + € M,(C) such that the real and imaginary parts of the entries of 7 are algebraically independent over Q. Then det(Im r) 4 0, implying that (7,1,) is the period matrix of a complex torus X of dimension g. On the other hand the equation for NS(X) in Proposition 3.4 has no nontrivial solutions, whence NS(X) = 0. In other words the Picard number p(X) is 0.14 1. Complex Tori : 4 The Dual Complex Torus a Let X = V/A be a complex torus of dimension g. Consider the C-vector space 2 := Home(V,C) of C-antilinear forms ¢: V — C. The canonical R-bilinear form (, ):8xVOR, (2,v) = Imé(v) is nondegenerate. Hence R= {@€8| (¢,A) CZ} is a lattice in Q and the quotient = H/R is a complex torus of dimension g, the dual complex torus. Identifying V with the space of C-antilinear forms 2 — C by double anti-duality, we get Re=x. Next we compute a period matrix of the dual complex torus R: Suppose (7, 1g) is a period matrix for the complex torus X, and J, the associated complex structure on R?9. So X = (R”, J,)/2; in particular we identify the complex vector spaces V = (R9, J,). Let Jt denote the adjoint of J, with respect to the standard alternating form I= (4, “ ). Consider the complex vector space (R”, Js"). The nonde- generate sesquilinear form S: (R*, Jt") x (RB, J) 2 C, S(e,m) = T(r, m) + iI (€,m) is C-linear in the first and C-antilinear in the second argument. So S identifies the complex vector space (R79, Jt~ *) with the vector space 2 = Home((R9, J,),C) and hence = (RY, ITY 2%. Proposition 4.1. If (7,1,) is a period matriz for X, then (*7,1y) is a period matriz for the dual complex torus X.4, The Dual Complex Torus 15 Proof. It remains to show that Js = J, since the complex structure determines the complex torus. But using Lemma 3.5 a _ ty-Lty tyal Jt =I T= 1,1 = ( aty tgtytte cine =J, O The following proposition gives a geometric interpretation of the dual complex torus. Proposition 4.2. The canonical homomorphism 21 — Hom(A,C,), Lr+ ei’) induces an isomorphism X — Pic?(X). (See [CAV] Proposition 2.4.1.) This shows that the group Pic?(X) admits the structure of a complex torus in a natural way. Let X, = V,/A, denote complex tori for v = 1,2 and f : X; > Xz, a homomorphism. The (anti-) dual map pa(f)* : 22421, 2+ &0pa(f) induces a homomorphism of complex tori £:%, 3%, the dual homomorphism. So “” is an exact (see [CAV], Proposition 2.4.2) contravariant functor from the category of complex tori into itself. The following lemma is easy to check: Lemma 4.3. Let f : X; —» Xq be a homomorphism of compler tori a) palf) = *palf)- b) p.(f) = *p-(F)- As a consequence we get Corollary 4.4. If f : X; — Xz is an isogeny of complex tori, the dual homomorphism f : Xz -+ X, is also an isogeny and deg f = deg f. Let L be a line bundle on X. For any point x € X the line bundle tiL ®@ L~ has first Chern class zero. Identifying X= Pic? (X ), according to Proposition 4.2 we get a map on: X 9X, ro ttbel which is a homomorphism according to the Theorem of the Square (see [CAV] 2.3.3). The map ¢z has the following properties. Proposition 4.5. a) The analytic representation of , is ox :V > 9, ve H(v,-) where H = e,(L). b) br = $4 under the identification X = X.16 1, Complex Tori c) The map NS(X) > Hom(X,X), L ++ $1 is a homomorphism of groups. d) For any homomorphism f : Y + X of complex tori the following diagram commutes xX & fT Lf y “4 P, For the proof see [CAV], Lemma 2.4.5 and Corollary 2.4.6. Accord- ing to [CAV], Lemma 2.4.7 the kernel K(L) of $1 satisfies the following properties Proposition 4.6. a) K(L@P)=K{(L) for any P € Pic®(X). b) K(L)=X if and only if L € Pic®(X). The propositions show that the homomorphism ¢, as well as its kernel K(L) depend only on the class of L in NS(X). Moreover we have the following criterion for a homomorphism of X into X to come from a line bundle im this way (see [CAV] Theorem 2.5.5). Proposition 4.7. Suppose X = V/A and f : X > X is a homomor- phism of complez tori. Then the following statements are equivalent: i) f = $1 for some line bundle L € Pic(X), ti) The map V x V > C, (v,w) & pa(f)(v)(w) is hermitian. A Poincaré bundle for X is a holomorphic line bundle P on X x X satisfying i) P\Xx{P}~P forevery Pe X = Pic(X), ii) P\{0} x X is trivial. According to [CAV] Theorem 2.5.1 there exists a Poincaré bundle on X x X, uniquely determined up to isomorphisms. It satisfies the following universal property (see [CAV], Proposition 2.5.2). Proposition 4.8. For any normal complez analytic space T and any line bundle L on X xT satisfying i) LIX x {t}e Pic(X) for every teT, ii) L{O}xT is trivial, there is a unique holomorphic map y:T + X such that Le (id x p)*P.5. Extensions of Complex Tori 7 5 Extensions of Complex Tori Let X; and X, be complex tori. An ertension of X2 by Xj is an exact sequence of complex tori e: 09K 5X SX, 0. Let e’ : 0 — X, > X’ + X, — 0 be a second extension of X2 by X,. The extensions e and e’ are isomorphic if there is an isomorphism {:X — X’ fitting into the following commutative diagram 00> X + X + X, +0 foot Il @) 09 XY > X > X% 0. Let Ext(X2, X,) denote the set of isomorphism classes of extensions of X_ by X}. Lemma 5.1. a) A homomorphism h : X; + X2 of complex tori induces @ map At : Ext(Xp, X1) > Ext(X, X). 4) A homomorphism g : X, + X{ of complex tori induces a map g. : Ext(X2,X1) > Ext(Xa, Xj). c) Ext is a functor, contravariant in the first and covariant in the second argument. Proof. a) Let e € Ext(X2,X,) be given as above. Define X’ to be the connected component of the kernel of the homomorphism (p, —A) : X x Xi Xa, (z,y) + p(x) — Ay). Diagram chasing gives the following commutative diagram with exact rows which defines h*e € Ext(X}, X1): hte: O04 Xi > X’ + Xi 0 (oe a e: 0-7 X > X % X, 70. It is easy to see that this definition does not depend on the choice of the representing exact sequence. ok b) Dualizing the exact sequence e gives an element é € Ext(X1, X2). Define axe = (9°6)". The proof of c) is immediate. a18 1. Complex Tori The set Ext(X2,X,) is endowed with a composition law defined as follows: Let e and e’ be extensions in Ext(X2,X,). The product e x e' € Ext(X2 x X2,X) x X1) is represented by the exact seqence 0 + X1 xX, > X x X' 3 X, x X, 7 0. IF A: Xp > Xz x Xo is the diagonal map and uw: X, x X12 X the addition map, the sum e + e’ is defined to be the image of e x e’ under the composition _ Ext(X x Xo, Xi x Xi) As Ext (Xo, Xi x Xi) A Ext(X2,.%i) ete :=p,A"(e xe’). Proposition 5.2. Ext(X2,X1) is an abelian group with respect to the addition (e,e') + e+e. We leave the proof to the reader. o Next we describe the group Ext(X2,X1) in terms of period matrices. Fix period matrices Il, € M(g, x 2g,,C), v = 1,2 for X; and X2 respectively. Given an extension e : 0 — X, ~ X — X, — 0, there is a period matrix of X of the form (4 ¢,) with o € M(g, x 2g2,€). Conversely it is obvious that for any a € M(gi x 2g2,C) the matrix (% g,) is a period matrix defining an extension of X by X1. Lemma 5.3. Let 0,0’ € M(g, x 2g2,€). The period matrices ( ,?,) and (" %,) define isomorphic extensions of Xz by X, if and only if o' =o +IhM + Allg with some matrices A € M(gi X g2,C), and M € M(2q; x 2go,Z). Proof. This follows immediately from equation (1) in Section 1 which here has the form 1, A To \_fth & log, —M o 0 ly 0 m)~\o0 mh 0 Im )* Proposition 5.4. Suppose the period matrices ("} {,) and (%} 5) define extensions e and e! in Ext(X2,X1). The period matrix (BF on fi 2) defines the extension e +e’ € Ext(X2, Xi). Proof. As above denote X = Cato/(%} 2 )Z?m+2m and X! = Cn+m/ (3 2 )z?n+2n, The complex torus X x X’, defined by the extension e xe’, is given by the period matrix TI 0 o 0 0m 0 a 0 0 Ih oO 0 0 O IL5. Extensions of Complex Tori 19 On the other hand, if A*(e x e’) : 0 4 X1 x Xi 3 Y > Xq > 0, the complex torus Y is given by a period matrix of the form Th 0 a 0 Th 6B }. 00 kh The homomorphism e x e' — A*(e x e’) corresponds to a homomorphism Y — X x X' of complex tori, given by the equation 1, 0 A) th 0 a 0 In Aa (42) 0 0 Ip oon 0 0 T; 0 o 0 ly 0 M _{ 0 Mh 0 a! 0 lan Me =| 0 0m 0 0 0 Ip 0 0 0 Ib 0 0 Ip with A; € M(g, x g2,C) and M; € M(2g1 x 2go, Z). Comparing both sides gives a = o64+1\M — Al, B= o'+ThM.— Aglly. Hence according to Lemma 5.3 Tm og 0 Th of 00, is also a period matrix for Y, respectively A*(e x e’). Denoting e +e’: 0 X; — Z > X2 — 0, a period matrix for Z is of the form (™ Th ). The homomorphism p, : A*(e x e’) +e +e! defines a homomorphism Y — Z which is given by the equation Th 0 (3 * ft) 0 Ih o 9 8 tas \ 0 0 th -{mhr lin lan M “\.0 Tb 0 0 le with A € M(g, x go,C) and M € M(2q, x 2g2,Z). This yields r = o+o'—TM + Ally. Again Lemma 5.3 shows that (1% 742") is a period matrix for Z, respectively e + e'. aq20 1, Complex Tori As a consequence the assignment o ++ C#+9/(" f)Z?%+2% induces a surjective homomorphism of abelian groups wo = ¥mnm : M(g1 x 292,C) > Ext (X2, Xi). According to Lemma 5.3 its kernel is ker p = II, M(2g1 x 2g2,Z) + M(gr x 92, C)T- Certainly this homomorphism depends on the choice of the period matri- ces I]; and IIz. The following proposition shows that the homomorphism » is bifunctorial Proposition 5.5. a) For a homomorphism h : X} — X2 the following diagram commutes M(g, x 292, €) "23? Bxt(X2,.%) -pr(h) 4 dar. Yam M(gi x 293,C) 3" Ext(X3, X41). 4) For a homomorphism g : X; + X{ the following diagram commutes M(aqr x 292, C) “23 Ext(X2,X1) po(g) | to on M(gl x 2g2,C) “25? Ext(Xp, X1). Proof. For an extension e € Ext(X2,X1) choose matrices ¢ € M(q x 2g2,C) with Yn,,n,(¢) =e and o! € M(g: x 2g5,C) with on,,1,(0") = h*e. According to diagram (2) in the proof of Lemma 5.1 the matrices o and o’ are related by the equation 1, A Th o \_/fth o ly, M J p(t) ) 0 mJ-“Lo m)\ 0 pm with A € M(g, x 95,C) and M € M(2g; x 295, Z). Comparing both sides we get o' =ap,(h) + IM — AI. Now Lemma 5.3 implies ~(op,-(h)) = (0) = h*e which implies a). The proof of b) follows by the same arguments. Oo In terms of period matrices the previous proposition says that if (" 7, ) is a period matrix defining an extension e :0 - X; — X > X20 then (% 7") is a period matrix for the extension he and (Hee) . 9 m 2 is a period matrix for g,e. Another consequence is5. Extensions of Complex Tori 21 Corollary 5.6. For e € Ext(X2, Xi) andneZ Nx, € = NX.e = Ne. The homomorphism y : M(g; x 2g2,C) — Ext(X2, Xi) depends on the choice of the period matrices II, and II. In order to get a better descrip- tion of we normalize the period matrices: According to Proposition 1.1 we may choose bases in such a way that the period matrices are of the form Th =(mi,1,,) and Tp = (72, 1g2) with 7, € M,,(C), detIm7; 4 0. Using this setup we have Proposition 5.7. The embedding M(g. x g2,C) + M(gi x 2g2,C), a+ (a,0), together with p induces an isomorphism of abelian groups M(gx X g2,C)/ (tis Vq)) M(2g1 X 2g2, Z) (12) > Ext(X1, Xo). Proof. Suppose o = (01,02) € M(g: x 2g2,C) with o; € M(gi x g2,C) corresponds to the extension e = #(o) € Ext(X2, Xi). By Lemma 5.3 the matrix o ~ oll = (01 ~ 0972, 0) corresponds to the same extension e. This shows that every extension in Ext(X2, Xi) can be represented by a matrix o = (a,0) with a € M(qi x g2,C). Hence we get a homomorphism of abelian groups M(9; x go, C) > Ext(X2, Xi), which is certainly surjective. According to Lemma 5.3 the matrices a and a/ € M(g: x g2,C) define the same extension, if and only if (a—a’,0) M, M2 m1 (44 Me ) + AT (71. My + Mg + Ata, 1M2 + My + A) for some A € M(g, x g2,C) and M = (48 Mi?) © M(2g: x 292, Z). Setting A=-—mM, — Mg we get a—a! = 1M, — Mot + M3 — Mata M, —M \ (1 1 = (rt) ( MoM, ) ( tn ) € (ri, Ly) M (292 22,2) ( 2 ) . This proves the assertion. Oo22 1. Complex Tori 6 Complementary Subtori and Shafarevich Extensions Let X; be a subtorus of a complex torus X. A complement of X in X isa subtorus Y of X such that the additon map » : Xi x Y — X is an isogeny. In this case X; is called a factor of X. If X is an abelian variety the Poincaré Reducibility Theorem ({CAV] 5.3.7) implies that every subtorus of X is a factor of X. We will see that for arbitrary complex tori this is not the case. The following proposition gives a criterion for a complex subtorus to admit a complement. Proposition 6.1. Let X; be a subtorus of a complex torus X, let X2 = X/X, ande :0 4 Xi + X 24 X_ — 0 the associated extension. The following conditions are equivalent: a) X, is a factor of X, b) X is isogenous to X, x Xe, ¢) there is a homomorphism g : Z — X of complex tori such that the composition pg: Z — X, is an isogeny, d) there exists a homomorphism f : Xz; — X and a positive integer n such that pf =nx,, e) there is a positive integer n such that ne = 0 in Ext(X2, Xi), {) using the notation of Proposition 5.7: if a € M(g; X g2,C) defines the extension e € Ext(X»2, Xi), then @ € (71, 1g.) M(2gr X 292, Q)( 12). Proof. a) b) is trivial. a) =>d) Let e: ¥Y<+X be a complement of X;. The composition pu : Y — X; is an isogeny. According to Proposition 1.6 there is an isogeny h: X,— Y with puh = nx, for some positive integer n. Now f := th satisfies condition d). d) > c) Define Z := im{f : X, + X} and let g : Z + X be the natural embedding. c)=>b) Consider the homomorphism X, x Z + X, (21,2) > 1, —g(z). Its kernel is {(g(z), z)|z € ker pg} which is finite by assumption. Hence Xx Z ~— X is an isogeny. Now b) follows from the fact that Z is isogenous to X2. d) =e) Let f : X, + X be a homomorphism with pf = nx,. Pulling back the extension e with nx, we get the commutative diagram nye: 07 X > X' + X 30 lb et bane e: On-n X% ~ X > X& 0.7. Simple and Indecomposable Complex Tori 23 By the universal property of the fibre product, f defines a splitting X_ > X', ie, nk,e = ne = 0. The converse implication follows from the same diagram. The equivalence e) + f) is an immediate consequence of Proposition 5.7. Oo Remark 6.2. Let X, and X2 be abelian varieties (for the definition see Section 2.1) and e: 0 X, + X > X2 — 0 be an extension. A direct consequence of the previous proposition is: X is an abelian variety if and only if the extension e is a torsion element in Ext(Xo, Xi). An extension 0 > X, + X — X, — 0 of complex tori is called a Sha- farevich extension if the subtorus X; of X does not admit a complement in X. This notion was introduced by Oort and Zarhin in [OZ2] in the case of elliptic curves X, and X2. The first example of such an extension was given by Shafarevich in [Sha] pp. 351-352. Corollary 6.3. For any compler tori X, and X2 of positive dimension almost all extensions are Shafarevich extensions. Proof. By Proposition 5.7 there are uncountably many extensions of X2 by X,. But according to Proposition 6.1 only countably many are not Shafarevich extensions. BC 7 Simple and Indecomposable Complex Tori Let X be a complex torus of dimension g. Recall from Section 1 the endomorphism ring End(X). The Q-algebra Endg(X) is called the endo- morphism algebra of X. Proposition 7.1. Endg(X) is an algebra of dimension < 4g over Q. Proof. The rational representation induces an embedding of Endg(X) into Mr,(Q). a A nonzero complex torus X is called simple if 0 and X are the only com- plex subtori of X. The Poincaré Reducibility Theorem ((CAV] Theorem 5.3.7) says that any abelian variety is isogenous to a product of simple abelian varieties. This statement is not valid for arbitrary complex tori. Example 7.2. Let e:0— Xi: X % X, — 0 be a Shafarevich exten- sion with simple complex tori X, and Xz. Then X; is the only nontrivial complex subtorus of X, in particular X is not isogenous to a product of24° 1. Complex Tori nonzero complex tori. In fact, suppose Y # 0 is a subtorus of X. Since the torus X is simple, the image p(Y) is either 0 or X>. In the first case Y = Xj, X; being simple. In the second case Y M X cannot be finite, since e is a Shafarevich extension. Hence Y MN X, = Xi which implies Y=X. A nonzero complex torus X is called indecomposable, if X is not isoge- nous to a product Y x Z of nonzero complex tori Y and Z. Certainly any simple complex torus is indecomposable. X simple = X_ indecomposable. Example 7.2 shows that the converse implication is not valid. Proposition 7.3. For a nonzero complex torus X the following condi- tions are equivalent: a) X is indecomposable, b) Any endomorphism of X is either an isogeny or nilpotent, c) Endg(X) is a local Q-algebra. Proof. a) > b) Suppose f € End(X) is not an isogeny. The complex subtori Z, :=im f* C X form a descending chain XIDM9°--DZ 9D Zn D-:. We have to show that Z, = 0 for n >> 0. For this it suffices to show that Znii & Zn, if Zn # 0. If this is not the case, the restriction f|Z, : Zn — Zn+1 is an isogeny and hence the group ker{f|Zn} = (kerf*t")NZ, is finite. This implies that Z, is a complement of the complex subtorus (kerf"*")q in X, contradicting the assumption that X is indecomposable. b) + a) Suppose X is isogenous to a product Y x Z of nonzero com- plex tori Y and Z. Then Endg(X) ~ Endg(Y x Z). But the composition Y¥xZ%Y4Y x Z is neither an isogeny nor nilpotent. For the equiva- lence b) <> c) notice that the unique maximal ideal of Endg(X) consists just of the nilpotent elements. a Lemma 7.4. Let X be an indecomposable complez torus and (py, ..., Pr © End(X). If Di, vi is an isogeny, at least one of the yp; 's is an isogeny. Proof. It suffices to prove the lemma for r = 2. By Proposition 1.6 there is an isogeny ~ € End(X) such that yy. + ~y2 = nx for some positive integer n. Hence wy; and yy. commute. If both wy, and ype are nilpo- tent, so is their sum nx, a contradiction. ao8. The Endomorphism Algebra of a Complex Torus 25 Theorem 7.5. Given a complex torus X #0, there is an isogeny X — XP x + x XPe (1) with indecomposable pairwise nonisogenous complez tori X, and positive integers n,. The complex tori X, and the integers n, are uniquely deter- mined up to tsogenies and permutations. Proof. The existence of the decomposition (1) is obvious. We have to prove the uniqueness statement. Suppose X = X,x--+-x X, and Y = Y, x--+x ¥, with indecomposable complex tori X; and Y;, and f : X — Y is an isogeny. Let p, : KX + X, and q, : Y — Y, denote the natural projections and u, : X, + X and j, : ¥, + Y the natural embeddings. According to Proposition 1.6 there is an isogeny g : Y + X such that fg = ny and gf = nx. The endomorphisms y, := P193.duft1 € End(X) satisfy : : Ve=pg9 (35.0) fa =pigfa =n -piy =nx,. Bel w= By Lemma 7.4 one of the y, 's is an isogeny. We may assume that 9 is an isogeny. Then gift; : X; — % is an isogeny, X; and Y, being indecomposable. Moreover, the following diagram commutes: o- X, 3 X 3 Xx--xX, 30 ajal oo o|f Lh o- ¥% BY s ¥gx-xY 30. By diagram chasing f, is an isogeny. Repeating this process gives the assertion. o Remark 7.6. In more sophisticated terms, the category with complex tori as objects and elements of Homa(-,-) as morphisms is an exact cate- gory, and Theorem 7.5 is the Krull-Schmidt Theorem for exact categories with bi-chain condition. (See [At]). 8 The Endomorphism Algebra of a Complex Torus In this section we study the endomorphism algebra of a complex torus X. Recall that the radical of a finite dimensional Q-algebra is the union of its nilpotent left ideals. The algebra is called semisimple if its radical26 1. Complex Tori is zero. According to [CAV] Corollary 5.3.8 the endomorphism algebra of an abelian variety is semisimple. The following example shows that for arbitrary complex tori this is not the case. Example 8.1. Let Y be a simple complex torus and e:0-+Y + X 4% Y — 0 a Shafarevich extension. According to Example 7.2 the complex torus X is indecomposable, and by Proposition 7.3 Endg(X) is a local algebra of finite dimension over Q. So its radical coincides with its maxi- mal ideal. But ep is a nonzero element of the maximal ideal, so Endg(X) is not a semisimple algebra. The radical rad(Endg(X)) of Endg(X) is a two sided ideal. The semi- simplification of Endg(X) is the quotient algebra Endg(X)as ‘= Endg(X) /rad(Endg(X)). By Proposition 7.3 a complex torus X is indecomposable if and only if Endg(X)ss is a skew field. Recall that a factor of a complex torus X is a subtorus of X admitting a complement. If X and Y are abelian vari- eties admitting no isogenous factors, then Hom(Y, X) = 0. The followmg example shows that this need not be the case for complex tori. Example 8.2. Let 0+ Y + X — Z +0 be a Shafarevich extension of simple complex tori Y and Z. Then X is indecomposable (see Example 7.2). In particular Y and X admit no isogenous nontrivial factors. But Hom(Y, X) # 0 since it contains 1: Y 4 X. Let X and Y be complex tori. The set Ixy = {5 aif; | a; € Hom(Y, X), 6; € Hom(X, »| finite is a two-sided ideal in End(X). The above mentioned property of abelian varieties implies that Ixy = 0 for abelian varieties X and Y with no isogenous factors. For arbitrary complex tori we still have the following property Proposition 8.3. Let X and Y be complex tori such that no factor of X is isogenous to a factor of Y. Then Ig8X =0. Proof. Write dim X = g. It suffices to show that a,(,-+-a,(,; = 0 for arbitrary a; € Hom(Y,X) and 6, € Hom(X,Y). For this define8. The Endomorphism Algebra of a Complex Torus 27 Z, = im(a;6;+--a48,) C X. By definition a;4; : Z;-1 — Z; is surjective. Hence it suffices to show that dim Z;_, > dim Z; as long as dim Z;_, > 0. Suppose dim Z,_; = dim Z, > 0 for some v. Then a, : Z,-1 — Z, is an isogeny and the following diagram commutes = Zr 4 xX al \ abv LavB-orB . Y ay Z Now Z,-; is a factor in X with complement Z := ker(a,@,---aifi)o since Z,_1NZ = ker(a,G,| Z,-1) is finite and dim Z,_; + dimZ = X. On the other hand the lower triangle implies that Z,1 is isogenous to a factor of Y. This contradicts the assumption. a Theorem 8.4. Suppose f : X — Xf! x---x X is an isogeny with pair- wise nonisogenous indecomposable comple tori X, and positive integers ny. The isogeny f induces an isomorphism Endg(X)ss ¥ Mn, (Endg(X1)ss) X +++ x Mp,(Endg(X;) ss). Proof. Step I: Let X and Y be complex tori such that no factor of X is isogenous to a factor of Y. There is a canonical isomorphism Endg(X x ¥)s5 & Endg(X) ss x Endg(¥)ss- To see this note that there is a canonical isomorphism _ ff B\|aeEndg(X), 6 €Homa(¥, xX) Endg(X x ¥) = {( 7 6 ) 1 € Homg(X,¥), 6 € Endg(¥) }. It suffices to show that {(28) |B € Homa(¥, X), 7 € Homa(X,¥)} € rad(Endg(X x ¥)). But 0B)2 _ (By 0 70) =(% we with By € Ixy and 7G € Iy,x. According to Proposition 8.3 the endo- morphisms (7 and ¥f are nilpotent, which implies the assertion. Step II: t. Endg(X)ss RE Endg(Xf" x --- x XP")as Endg(XP*)ex x «++ x Endo(X?*)es Mr, (Endg(X1)) so % ++ X Mng(Endg(X-))ss My, (Endg(X1)ss) X ++: X Mn, (Endg(X;)es) + iS iS iS28 1. Complex Tori Here the isomorphism (1) is induced by f, (2) is the canonical isomor- phism of Step I, (3) is induced by the canonical isomorphisms Endg(X7") ~ M,,,(Endg(X;)). For the canonical isomorphism (4) notice that M,,(R).s ~ M,,{R,.) for any finite dimensional Q-algebra R (see Mec Theorem 6.15). a Notice that in the statement of the theorem one cannot replace Endg(:)es by Endg(-). A counterexample is given by Example 7.2. 9 The Theorem of Oort and Zarhin There are strong restrictions for an algebra to be the endomorphism al- gebra of an abelian variety (see [CAV], Chapter 5). For complex tori the situation is completely different: In fact Oort and Zarhin proved the fol- lowing theorem in {OZ1]: Theorem 9.1. For every finite dimensional Q-algebra A there exists a complex torus X such that A= Endg(X). Here we follow the proof of Hétte and Scharlau [HS]. Lemma 9.2. Suppose X; and X2 are complex tori with Hom(X,, X2) = Hom(X2, X1) = 0. Then for every extension e : 0 + X; + X — Xz — 0 there is a canonical isomorphism End(X) ~ {(g,h) € End(X;) x End(X2)|g.e = he}. Proof. Let f € End(X). By the assumption that Hom(X;, X;) = 0 for i # j the endomorphism f defines uniquely determined endomorphisma g and h such that the following diagram commutes 0o- X > X +X > 0 lo lr La 0- %m ~> X > & > 0.9, The Theorem of Oort and Zarhin 29 Certainly the map f + (9, h) is injective. The extensions g,e and h*e are given by the second and third row of the following commutative diagram e: Om X ~ X +- & > 0 * ot J ll ge: O-+ X > X + X + 0 o ll le ll hte: 0 + X > X” +> X > 0 ll J La e: O-+- X% ~ X +— &% - 0. Now g,e = h*e if and only if there exists an isomorphism a : X’ + X” fitting into the diagram. Obviously this is the case if and only if there exists an f : X —> X inducing g on X; and h on X2. Qa Lemma 9.3, For every finite dimensional Q-algebra A there exists a pos- itive integer n and matrices y,z € M,(Z) such that Ax {ze M,(Q)|zy = yz, rz = zz}. Proof. Consider the regular representation A + M,(Q), r = rkgA. This gives Q” an A-module structure in a natural way. Then A’ = End,4(Q’) coincides with the centralizer and A” = End4/(Q") with the double cen- tralizer of A in M,(Q). Now Theorem 8, Section XVII, 6 of [L] applies and says that the natural map A — A” is an isomorphism. Hence if t),..., tm are generators for A’ over Q, then A={s €M,(Q)|tis = st; for i=1,...,m}. (In the special case of a semisimple algebra A this follows from the usual double centralizer theorem (see e.g. J], I], Theorem 4.10). Without loss of generality we may assume that t; € M,(Z). Set n =r(m+ 2) and 01 4 #O 0... 0 y=[ i fel : “ed " : 0... 0 0 ... 1, 0 Now an immediate computation gives the assertion. a Proof of Theorem 9.1. Let FE, and Ey be nonisogenous elliptic curves without complex multiplication (for the notion of complex multiplication30 1. Complex Tori see [CAV] Section 9.6). In particular Endg(E,;) = Q and Hom (E), Ey) = Hom (E2,E,) = 0. According to Proposition 5.7 Extg(E2,A\) := Ext (Eg, E,) @z Q is a Q-vector space of infinite dimension. Choose 1, €2, €3 € Ext(Ea, E,) whose images in Extg(E2, E;) are linearly independent. Suppose A is a subalgebra of M,(Q) as in Lemma 9.3. Obviously we have Exta(£f, Hf) © My(Bxtg (Ex, Ey))- Using Proposition 5.5 and Corollary 5.6 one easily sees that the Q- algebras Endg(E?) ~ M,(Q), respectively Endg(E?) = Mn(Q) act from the left, respectively from the right, on Extg(Z7, Ef) by matrix multipli- cation. Let y and z € M,(Z) be the matrices of Lemma 9.3 defining A. Consider the extension 03 BE? 3X E30 defined by the element e = y(1n ® €1) + 2(1n ® €2) + 1n @ eg. According to Lemma 9.2 Endo(X) Ie {(9,h) € Endg(Et) x Endo(E3)|g.e = h*e} {(9,h) € M,(Q) x M,(Q)|ge = eh} {(g,h) € Ma(Q) x Mn(Q)lgy = yh, gz = zh, g = h} = A. lle i For equation * use (1, @ e,)h = h(1, @ e,) and the linear independence of 1, €g and e3. a Remark 9.4. The extension 0 + E? + X — E} — 0 occurring in the proof of Theorem 9.1 is always a Shafarevich extension. This is a direct consequence of Proposition 6.1 a) > e): The image of e € Ext(Ey, E;) in Extg(E2, E;) is nonzero. In particular X is not an abelian variety. 10 The Space of all Complex Tori of Dimension g A family of complex tori is by definition a flat holomorphic map of com- plex manifolds wT: X 3B such that for every b € B the fibre r-1(b) is a complex torus. In this section we construct a family of complex tori which contains all complex tori of dimension g.10. The Space of all Complex Tori of Dimension g 31 Let Gr"(C%) denote the Grassmannian variety of n-dimensional quo- tient vector spaces CN —+ V. Choosing a basis of V, an element z € Gr"(C¥) is represented by a matrix II € M(n x N,C) of rank n. Con- versely every II € M(n x N,C) of rank n defines an element of Gr*(C™). Two matrices II and II’ represent the same element of Gr*(C) if and only if there is an a € GL,(C) such that all = Il’. If there is no ambigu- ity, we do not distinguish notationally between the elements z of Gr"(C*) and their representatives II € M(n x N,C). Note that for z; € Gr*(C%) and 2, € Gr’-"(C¥) the notation z A zz, # 0 means that the subvector spaces ker z; and ker z, of C™ intersect transversally. In terms of the representing matrices this is the case if and only if det (73) # 0. More- over for z € Gr"(C”) denote by Z the complex conjugate quotient vector space, given by the complex conjngate matrix IT in M(n x N,C). Consider the Grassmannian variety Gr9(C). Notice that an element z:C% — V of Gr9(C”%) restricts to an (R-linear) isomorphism R77 + V if and only if zAZ #0. Denote by B, the following open submanifold of the Grassmannian Gr9(C?#) By = {2 € Gr?(C*) |z AZ AO}. Every element z : C77 — V of By defines a g-dimensional complex torus X, = V/2(Z). Conversely, any complex torus X of dimension g is of the form X = C2/T1Z* with a period matrix I € M(g x 2g,C). Since det (#) 4 0, the period matrix II represents an element z € B, such that X = X,. In order to construct a family of complex tori over B, with fibres X, we prove first Lemma 10.1. B, ~ GL,(C)\GL2,(R), where GL,(C) embeds into Glng(R) via A+iBH (457). Proof. We show that the map GLg(R) + M(g x 29,C), M+ Ty = (1g, t1y)M induces a bijection GL,(C)\GL2,(R) > B,. To see this note that for every square matrix M = (7) € M2,(R) with z,y € M(g x 2g,R) we have ly tly \yy_( tiv \_ ( Un 1, -il, ~\a-iy) \lw/*32 1, Complex Tori Hence M € GLz,(R) if and only if Ty represents an element of B,. Moreover for a= A+iB € GL,(C) we have te )(4 BY y (49) (Tv), 1, -,)\B A oa) \ Tw This implies the assertion. 0 According to the lemma the quotient GL,(C)\GL2,(R) inherits the structure of a complex manifold. Consider the group actions GL,(C) x (C? x GLn(R)) > C% x GLn,(R), (a,(v,M)) 4 (av, (f 72)M) where a = A+iB, and (C2 x GLa,(R)) x Z* + C2 x GLag(R), ((v,M),A) > (u + Tye(A), M) where as above IIyq = (1,,%1,)M. The quotient %, = GLg(C)\(C! x GLa, (R))/2™ is a complex manifold since GL,(C)\GLn,(R) is a complex manifold and the action of Z?9 is free. The projection onto the second factor induces a holomorphic map 1: AX, > By. By construction the fibres are m~!(z) = X,. Combining everything we have Theorem 10.2. 7: % — B, is a family of complez tori of dimension g such that n—*(z) = X, for every z € By. In particular the family contains every complez torus of dimension g. The following proposition shows that a given complex torus X occurs many times as a fibre of the family m : 4, > By. Proposition 10.3. Suppose 2, z2 € By. The complex tori X,, and X,, are isomorphic if and only if 2 = 2M for some M € GL.,(Z). Proof. Choose representatives 1, and II; of z and 2. Then X,, = C2 /T1;Z*4. Suppose there is an isomorphism y : X,, > X,,. Its rational representation M € GL,,(Z) and analytic representation a € GL,(C) are related by the equation all; = T,M.10. The Space of all Complex Tori of Dimension g 33 Since all, and Il, define the same element z, of By, this gives 2, = 22M. The converse implication is obvious. The arithmetic subgroup GL2.(Z) of GLa.(IR) acts on the space B, from the right in a natural way. The quotient B,/GL2.(Z) parametrizes the set of all isomorphism classes of complex tori of dimension g. In Chapter 7 we will see that By/GL2,(Z) is no complex analytic space; does not satisfy even the separation axiom Tj. So it is no moduli space for complex tori in the usual sense. For a more precise statement see Section 7.3. cho2 Nondegenerate Complex Tori To any smooth projective curve C one can associate an abelian variety, the Jacobian variety J(C). In [W2] Weil showed that, more generally, to any smooth projective variety M of dimension n and any p < n, one can associate an abelian variety, the p-th intermediate Jacobian of M. It has, however, the disadvantage that it does not depend holomorphically on M in general. It was Griffiths’ idea to modify the definition in such a way that the new intermediate Jacobian J%(M) varies holomorphically with M. It is a complex torus, but in general not an abelian variety. It admits, however, a class of line bundles whose first Chern class is a nondegenerate hermitian form. This is a special case of the following situation: Let X be a complex torus of dimension g and H € NS(X) a nondegenerate hermitian form. Suppose k denotes the index of H, that is, the number of negative eigenvalues of H. We call such a hermitian form a polarization of index k. (Note that in [G] H is called a k-convex polarization). If H is a polarization of index k on a complex torus X, we call the pair (X, H) a nondegenerate complex torus of index k. In view of the definition of a pseudo-Riemannian manifold [He] one might be tempted to call (X, H) a pseudo-abelian or semi-abelian variety, but these notions have already a different meaning. Note that a nondegenerate complex torus of index 0 is a polarized abelian variety. It is the aim of this chapter to derive the main properties of nondegenerate complex tori of index k. Section 1 contains the definition, some examples and first properties of polarizations of index k. In Section 2 we construct paramieter spaces for nondegenerate complex tori of index k. They will be used in Chapter 7 for the construction of moduli spaces. In Section 3 we introduce the Rosati-36 2. Nondegenerate Complex Tori involution on Endg(X) for any nondegenerate complex torus (X, H) and show that the Néron-Severi group tensorized with Q is canonically isomor- phic to the symmetric part of Endg(X). Section 4 contains the definition of the dual polarization. If (X, H) is a nondegenerate complex torus and Y a subtorus of X, the restriction of H to Y is not necessarily non- degenerate any more. However, if it is, then Y admits a complementary subtorus in X. This leads to a version of Poincaré’s Reducibility Theorem for nondegenerate complex tori, which we prove in Section 5: Any non- degenerate complex torus is isogenous to a product of irreducible ones, that is nondegenerate complex tori which do not admit any nontrivial nondegenerate subtorus. An important invariant of a complex torus X is the transcendence degree of the field of meromorphic functions of X, called the algebraic dimension a(X) of X. It is introduced in Section 6. Moreover, we show that for every complex torus X there is a universal maximal quotient abelian variety of dimension a(X). Finally, in Section 7, we work out the possibilities for the pairs (Picard number, algebraic dimension) for complex tori of dimension 2. 1 Polarizations of Index k Let X = V/A be a complex torus of dimension g and L a kne bundle on X. Its first Chern class H = c(L) is a hermitian form on V whose imaginary part Im H takes integral values on the lattice A. The ine bun- dle L is called nondegenerate if the hermitian form H is nondegenerate. Suppose L is nondegenerate and its first Chern class H admits exactly k negative eigenvalues. According to Theorem 1.2.2 the only nonvanishing cohomology group of L is H*(L). The number k is called the indez of the line bundle L as well as the index of the hermitian form H and denoted by ind(L) respectively ind(H). Notice that the index is defined only for nondegenerate line bundles respectively hermitian forms. Recall that a polarization of a complex torus is the first Chern class of a positive definite ine bundle on X. (An abelian variety is by definition a complex torus admitting a polarization.) We define more generally a polarization of index k on X to be the first Chern class of a Hine bundle of index k. Hence a polarization of index 0 is a polarization in the usual sense. By abuse of notation we sometimes consider the line bundle itself as a polarization. Recall from Section 1.2 that the type of a nondegen- erate line bundle L is the vector (d1,...,d) of elementary divisors of Im H. The type of a polarization is by definition the type of the defining line bundle. A polarization of index k is called principal, if it is of type (,..- 1.