Graduate Texts InMathematies Serge Lang Introduction to Algebraic and Abelian Functions Second Edition 6 Springer-Verlag ae - BOOKS OF RELATED INTEREST BY SERGE LANG Algebra, Third Edition 1987, ISBN 96412-6 U uate Algebra, Second Edition 1990, ISBN 97279-X Complex Analysis, Third Edition 1993, ISBN 9786-0 Real and Functional Analysis, Third Edition 1993, ISBN 94001-4 Algebraic Number Theory, Second Edition 1994, ISBN 942254 Introduction to Complex Hyperbolic Spaces 1987, ISBN 96447-9 OTHER Books By LANG PUBLISHED BY ‘SPRINGER-VERLAG Introduction to Arakelov Theory * Riemann-Roch Algebra (with William Fulton) * ‘Complex Multiplication + Introduction to Modular Forms * Modular Units (with Daniel Kubert) * Fundamentals of Diophantine Geometry * Elliptic Functions * Number ‘Theory I * Cyclotomic Fields I and I « SL(R) Abelian Varieties » Differential and Riemannian Manifolds * Undergraduate Analysis * Elliptic Curves: Diophantine ‘Analysis * Introduction to Linear Algebra * Calculus of Several Variables * First Course se Gah Mail Mathematics * Geometry: A High School Course (with Gene furrow) * Math! Encounters with High lents © The Beauty of Doin, Ma a Ea High School Students * The Beauty of Doing Serge Lang Introduction to Algebraic and Abelian Functions Second Edition Springer-Verlag New York Berlin Heidelberg London Paris Tokyo Hong Kong Barcelona BudapestSerge Lang Depron of Mates Yale University Nv Haven, Conetct 06520 USA Editorial Board J.H. Ewing F.W. Gehring Department of Department of Mathematics Mathematics Indiana University University of Michigan Bloomington, IN 47405 Ann Arbor, MI 48109 USA USA AMS Cassi ions: 14HOJ, 14K2: With 9 illustrations. Library of Congress Cataloging in Publication Data Lang, Serge, 1927- Introduction to algebraic and abelian functions. P.R. Halmos Department of fathematics Santa Clara University Santa Clara, CA 95053 USA (Graduate texts in mathematics; 89) Bibliography: p. 165 Includes index. |. Functions, Algebraic. 2. Functions, Abelian. 1. Title. ML. Series. QA341.L32 1982 515.9°83 82-5733 AACR2 ‘The first edition of Introduction to Alg in 1972 by Addison-Wesley Publishing Co., Inc. © 1972 192 spin . eV New Yorke Alll rights reserved. This work ’ ta ‘Typeset by Interact Printed and bound by 3 Bran Bint, Ara Primed in the United States of America, 2" M- tive Composition Cory 98765432 (Second comected printing, 1995) ISBN 0-367-9074 ebraic and Abelian Functions was published oration, Pleasant Hill, CA, 0-6 Springer-Verlag ISBN 3-540.90710.¢ gee Vetl88 New York Berlin Heidelberg Springer-Verlag Berlin Heidelberg New York Introduction This short book gives an introduction to algebraic and abelian functions, with emphasis on the complex analytic point of view. It could be used for a course or seminar addressed to second year graduate students. ‘The goal is the same as that of the first edition, although I have made a number of additions. I have used the Weil proof of the Riemann-Roch the- orem since it is efficient and acquaints the reader with adeles, which are a very useful tool pervading number theory ‘The proof of the Abel-Jacobi theorem is that given by Artin in a seminar in 1948. As far as I know, the very simple proof for the Jacobi inversion theorem is due to him. The Riemann-Roch theorem and the Abel-Jacobi theorem could form a one semester course. “The Riemann relations which come at the end of the treatment of Jacobi's theorem form a bridge with the second part which deals with abelian functions and theta functions. In May 1949, Weil gave a boost to the basic theory of theta functions in a famous Bourbaki seminar talk, I have followed his exposition of a proof of Poincaré that to each divisor on a complex torus there corresponds a theta function on the universal covering space. However, the correspondence between divisors and theta functions is not needed for the linear theory of theta functions and the projective embedding of the torus when there exists a positive non-degenerate Riemann form. Therefore I have given the proof of existence of a theta function corresponding to a divisor only in the last chapter, so that it does not interfere with the self-contained treat- ment of the linear theory The linear theory gives a good introduction to abelian varieties, in an analytic setting. Algebraic treatments become more accessible to the reader who has gone through the easier proofs over the complex numbers. This includes the duality theory with the Picard, or dual, abelian manifold.TE al Introduction ic analytic facts n Thave included enough material to give sary in the theory of complex multipl ama, or my San) MTecent book on the subject, and have thus tried to make this topic is willing (© accessible at a more elementary level, provided the reader assume some algebraic results. T have also given the example of the Fermat curve, drawing on some recent results of Rohrlich, This curve is both of intrinsic interest, and gives a typical setting forthe general theorems proved in the book. This example illustrates both the theory of periods and the theory of divisor classes. Again this example should make it easier for the reader to read more advanced books and papers listed in the bibliography. New Haven, Connecticut SERGE LANG Contents Chapter [ The Riemann-Roch Theorem $1. Lemmas on Valuations ....... §2. The Riemann-Roch Theorem §3. Remarks on Differential Forms $4. Residues in Power Series Fields §5. The Sum of the Residues §6. The Genus Formula of Hurwitz 87. Examples .....cscssscsessseee §8. Differentials of Second Kind §9. Funetion Fields and Curves $10. Divisor Classes Chapter IL The Fermat Curve $1. The Genus .. §2. Differentials ... ey §3. Rational Images of the Fermat Curve $4. Decomposition of the Divisor Classes Chapter IIL The Riemann Surface §1. Topology and Analytic Structure §2, Integration on the Riemann SurfaceContents viii tr IV The Theorem of Abel-Jacobi «s $1, Abelian Integrals ; : 58 2. Abel’s Theorem soseeessnreneeee 0) veces 6B $5, Jacob's Theorem ..v-veese-eere e §4, Ricmann’s Relations s $5, Duality " Chapter V Periods on the Fermat Curve 2 1, The Logarithm Symbol 5 $2 Pehodsbn the Universal Covering Space 5 §3. Periods on the Fermat Curve srenneeeeeecens 4. Periods on the Related Curves .....000009 2 8 Chapter VI | Linear Theory of Theta Functions $1. Associated Linear Forms . : sees 83 §2. Degenerate Theta Functions ....... ceoseesieeeeeiesenieeees 89 §3. Dimension of the Space of Theta Functions ....... seo » 90 §4. Abelian Functions and Riemann-Roch Theorem on the Torus .... 7 §5. Translations of Theta Functions .........0.ccccccvsseseesereeceenveesesees 10D §6. Projective Embedding .....-.....cc-.0ee ceseceeeeseeeees 104 Chapter VII Homomorphisms and Duality $1. The Complex and Rational Representations. .......... : 10 . enon tnd p-aie Representations. 13 }. Homomorphisms seve : ee 6 . The D ian Manifold .. 118 $6. Relations with Theta Functions 121 $7. The Kummer Pairing ..... 124 §8. Periods and Homology 7 Chapter VIII Riemann i _ i Matrices and Classical Theta Functions » Riemann Matrices §2. The Siegel Upper Half Space 131 $5, ren lete! Upper Half Space a5 tal Theta Functions .. . 138 Contents Chapter IX Involutions and Abelian Manifolds of Quatemion Type 41. Involutions §2. Special Generators was §3. Orders : Eines 148 $4, Lattices and Riemann Forms on C? Determined by Quaternion Algebras 149 §5. Isomorphism Classes 154 Chapter X ‘Theta Functions and Divisors $1. Positive Divisors i ae 21ST §2. Arbitrary Divisors Se 163 §3. Existence of a Riemann Form on an Abelian Variety 163 Bibliography srciseanesiey : 165 Index . “ sevveer 167CHAPTER | The Riemann-Roch Theorem §1. Lemmas on Valuations We recall that a discrete valuation ring o is a principal ideal ring (and there- fore a unique factorization ring) having only one prime. If 1 is a generator of this prime, we call r a local parameter. Every element x + 0 of such a ring can be expressed as a product xery, where r is an integer = 0, and y is a unit. An element of the quotient field K has therefore a similar expression, where r may be an arbitrary integer, which is called the order or value of the element. If r > 0, we say that x has a zero at the valuation, and if r <0, we say that x has a pole. We write r= 0,(x), or vy), or —ord,(x). Let p be the maximal ideal of 0. The map of K which is the canonical map 0 — o/p on 9, and sends an element x € 0 to co, is called the place of the valuation. We shall take for granted a few basic facts concerning valuations, all of which can be found in my Algebra. Especially, if E is a finite extension of K and 0 is a discrete valuation ring in K with maximal ideal p, then there exists a discrete valuation ring © in E, with prime ‘8, such that o=ONK and p=BNK. If w is a prime element of ©, then rD = u*®, and ¢ is called the1. Riemann-Roch The tion index of Cover o (or of Rover). Ife and To are the value groups ion in $ i ™ We say that the pat Sigg) is unramified above (0,2), or that 8 is sere sore "tthe ramification index is equal (0 1, that ise = | unrami we B, Example, Let kbe a field and transcendental over k. Leta Gk. Let o be the set of rational functions flajeld, with f(0, g(0 Ekle] such that g(a) # O js adiscrete valuation ring, whose maximal ideal consists of all such ites such tat f(a) = 0. ‘This is typical situation. In fact re bs algebraically closed (for simplicity), and consider the extension (ayo taine ‘vith one transcendental element x over k, Let 0 be a discrete valuation ring in &(2) containing k. Changing x to I/x if necessary, we may assume that Tp. Thenp MAL] #0, and p 0 k(x] is therefore generated by an irre- ducible polynomial p(x), which must be of degree 1 since we assumed k algebraically closed. Thus p(x) = x ~ a for some a € k. Then it is clear that the canonical map o> ofp induces the map fa) fla on polynomials, and it is then immediate that o consists of all quotients S(2)/g(2) such that g(a) # 0; in other words, we are back in the situation described at the beginning of the example Similarly, let 0 = k{{s]] be the ring of formal power series in one variable Then o is a discrete valuation ring, and its maximal ideal is generated by 1. Every element of the quotient field has a formal series expansion x em ta + ag tat tat + RO tics 4; © k. The place maps x on the value ap if x does not have le. In the applications, we shall study a fi . ly a field K which is a finite extension of A secre extension k(x), where k is algebraically closed, and x is that nn cXet k: Such afield is called a function field in one variable D contining ae the tesidue class field of any discrete valuation ring NB A'S equal to & itself, since we assumed k algebraically closed position 1.1. Ler E be a finite extension OK. Let (D8) be a discrete 1. Lemmas on Valuations a i 3 valuation ring in E above (0,») in K. Su . ippose that E = K(y) where y is the root of a polynomial f(Y) = 0 havi i r the root of poly ing coefficients in 0, leading coeffi- LO) = 0 but f(y) # 0 mod §B. Then 'B is unramified over p. Proof. There exists a constant yp ksuch that y = yy mod $B. By hypothesis, f"(yo) #0 mod $B. Let {y,} be the sequence defined recur- sively by Inst = In = FO FO) ‘Then we leave to the reader the verification that this sequence converges in the completion K, of K, and it is also easy to verify that it converges to the root y since y = yp mod §B but y is not congruent to any other root of f'and 3B. Hence y lies in this completion, so that the completion Ey is embedded in Ky, and therefore $8 is unramified. We also recall some elementary approximation theorems. Chinese Remainder Theorem. Let R be a ring, and let i, . . . , Dube distinct maximal ideals in that ring. Given positive integers 1, . -. s Tn and elements a,... a, ER, there exists x ER satisfying the con- gruences x =a, mod pi for all i. For the proof, cf. Algebra, Chapter Il, §2. This theorem is applied to the integral closure of A(x] in a finite extension. We shall also deal with similar approximations in a slightly different context, namely a field K and a finite set of discrete valuation rings 0), . . ., 9, of K, as follows. Proposition 1.2. [f 0, and 02 are two discrete valuation rings with quotient field K, such that 0, C 03, then 0, hens eth, ee Proof. We shall first prove that if p, and p2 are their maxi ideals, then PoC py. Let y € pz. If y E py, then I/y € 0), whence T/y € P2, a con- tradiction. Hence 2 C Py. Every unit of 0, is @ fortiori a unit of 03. An element y of p, can be written y = a7}!u where wis a unit of 0, and m is an element of order pi. If a is not in pa, it is a unit in 02, a contradiction. Hence 7 is in Pz, and hence so is P = 0:7. This proves P2 = Pi. Finally,1 1nn-Roch Theorem 4 ,. and thus cannot be a unit en His is in 03, and is not in oy, the itis ta ove our proposition assume that our valuation rings 0: (1 + N)are fe ote, a sion relations dain, land hence have no inclusion F K having a zero at © and sts-an element y of K having 4 ; . There exists am Proposition 1.3. * a pole at 0; (j = A sem = 2. Since there ad by induction. Suppose n = 2 eis roo. This willbe proved by induction. Suppose” =X, Ske He no inclusion relation between 0; and 02+ Similarly, we can find z € 0 Now suj ewe “ et z be such that z has a zero at 0; and a pole at o,, Foe eee er atin our eequtemens, because we be schematically zero plus zer0 = zer0, zero plus pole = pole, and the sum of two elements of K having poles of different order again has a pole. ‘A high power of the element y of Proposition 1.3 has a high zero at 0; and a high pole at 0, (j= 2,...,1)- Adding 1 to this high power, and considering 1/(1 + y") we get and ¢ 0», Then z/y bas a zero at 0, and a Corollary. There exists an element 2 of K such that 2 — 1 has a high zero at 0,, and such that z has a high zero at ©, (j a) Denote by ord; the order of an element of K under the discrete valuation associated with o,. We then have the following approximation theorem. Theorem 1.4, Given elements a, .. . , dy of K, and an integer N, there exists an element y © K such that ord\(y — aj) > N. Proof. For each i, use the corollary to get z; close to 1 at 9, and close to O at 0; (j # 1), oF rather at the valuations associated with these valuation Tings. Then 21a +++ + + z,d, has the required property. __ In particular, we can find an element y having given orders at the valua- Mons arising from the 0,, This is used to prove the following inequality. Corollary. Ler E be a finite algebraic extension of K. Let V be the value Sroup of a discrete valuatior n of K, and T; the value groups of a finite number of inequivalent discrete value E i : dations of E €:be the index of in T;. Then Se creme meet Ete De ste: ky. 42. The Riemann-Roch Theorem Proof. Select elements Vinee s Shee Yay 1 She, of E such that yiv (v= 1, . . . , e:) represent distinet cosets of F in T, and have zeroes of high order at the other valuations u, (j # i). We contend that the above elements are linearly independent over K. Suppose we have a relation of linear dependence Deve =0 Say cy, has maximal value in I’, that is, e(cun) = w(c,) all i, v. Divide the equation by cy. Then we may assume that cj, = 1, and that v(c,) = 1. Consider the value of our sum taken at oj. All terms yi1,Ci2iay « - » ve have distinct values because the y’s represent distinct cosets. Hence Cyn + * F Cres Jie) 2 vi(yu). On the other hand, the other terms in our sum have a very small value at by hypothesis. Hence again by that property, we have contradiction, which proves the corollary. §2. The Riemann-Roch Theorem Let k be an algebraically closed field, and let K be a function field in one variable over k (briefly a function field). By this we mean that K is a finite extension of a purely transcendental extension k(x) of k, of transcendence degree 1. We call k the constant field. Elements of K are sometimes called functions. By a prime, or point, of K over k, we shall mean a discrete valuation ring of K containing k (or over k). As we saw in the example of §1, the residue lass field of this ring is then & itself. ‘The set of all such discrete valuation rings (j.e., the set of all points of K) will be called a curve, whose function field is K. We use the letters P, Q for points of the curve, to suggest geometric terminolo; By a divisor (on the curve, or of K over k) we mean an element of the free abelian group generated by the points. ‘Thus a divisor is a formal sum. a= P= > npP where P; are points, and 1; are integers, all but a finite number of which are 0. We call1 Riemann-Roch Theorem Ze the 1 of a, and we call ny . th etree a x4 0, then there ol) > ie der of cat Pi ya finite number of points P such that = 0 forall P. If is not hen ordp(x) = 0 for al i, if.x is constant, pO ee nerd, anal Ge na ody x # 0. Indes one point of k(x) at which x has a 7670, ane one point constant, then there fs or Ty of these points extends t0 only a finite number aan of 3 Deaf isa finite extension of K(x). Hence we can of point is 1 divisor with x, namely () =D meP sand 6 are said to be linearly equivalent if YnpP and 6 = E mipP are divisors, where mp = ofdp(x). Divisor be a — Gis the divisor of a function. If a = we write aZ=b ifand only if mp 2 mp forall P. This clearly defines a (partial) ordering among divisors. We call « positive aa divisor, we denote by L(a) the set of all elements x € K such that (a) 2 —a. If @ is a positive divisor, then L(a) consists of all the functions in K which have poles only in «, with multiplicities at most those of a. It is clear that L(a) is a vector space over the constant field k for any divisor a We let /(a) be its dimension, Our main purpose is to investigate more deeply the dimension [(a) of the vector space L(a) associated with a divisor @ of the curve (we could say of the function field). Let P be a point of V, and 9 its local ring in K. Let p be its maximal ideal. Since k is algebraically closed, o/p is canonically isomorphic to k. We know that o is a valuation ring, belonging to a discrete valuation. Let 1 be a generator of the maximal ideal. Let.x be an element of 0. ‘Then for some constant a in k, we can write x = dy mod p. ‘The function x — ap is in P, and has a zero at 0. We can therefore write x — ay = ty, where yy is in 0. Again by a similar argument we get yy = a, + 1y; with y, € 0, and Ta tat + ye Continuing this procedure, we obtain an expansion of x into a power series, x % + ait + ay? 4s Tris trivial that if each Coefficient a, ‘The quotient field K of 0 can be ember to os then x = 0. 'mbedded in the power series field k((¢)) §2, The Riemann-Roch Theorem as follows. If x is in K, then for some power 1” and hence x can be written » the function rx lies in 0, ay Pet tata eee, If w is another generator of p, then clearly k((1)) = k(( 5 = k((), and our series field depends only on P. We denote it by Kp. Anelement & of Ky ean be writen = 3v_4 4.0" with dy #0. I'm <0, we say that & has a pole of order —m. If m > 0 we say'that & has a zero of order mr and we let im = ord & , Lemma. For any divisor a and any point P, we have Wa+P)s (@) +1, and 1(«) is finite Proof. If a = 0 then I(a) = 1 and L(a) is the constant field because a function without poles is constant. Hence if we prove the stated inequality, it follows that 1(a) is finite for all a. Let m be the multiplicity of P in a. Suppose there exists a function z € L(a + P) but 2 € L(a). Then orde x= — (m+ 1). Let w © L(a + P). Looking at the leading term of the power series ex- pansion at P for w, we see that there exists a constant ¢ such that w — cz has order = —m at P, and hence w €L(a). This proves the inequality, and also the lemma. Let A¥* be the cartesian product of all Kp, taken over all points P. An element of A* can be viewed as an infinite vector = (.. . , , - - .) where & is an element of Kp. The selection of such an element in A* means that random power series has been selected at each point P. Under component- wise addition and multiplication, A* is a ring. It is too big for our purposes, and we shall work with the subring A consisting of all vectors such that &» has no pole at P for all but a finite number of P. This ring A will be called the ting of adeles. Note that our function field K is embedded in A under the mapping KOO EEG. Dy i.e., at the P-component we take x viewed as a power series in Kp, In particular, the constant field k is also embedded in A, which can be viewed as an algebra over & (infinite dimensional).1. Reman Theorem 8 ve, We shall denote by A(a) the subset of 4 Leta be a divisor on our curves Te ody a. Then A(a) is imme iat ord (ene ae set ofall such AC) space orhods oF OA, and iN be take 1 topology consisting of all adel diately seen to be ak as a fundamental SY es a topological ring i spy becomes a topolog! : eBlORBNBGE LC) " 4 whic anetion «such that (x) = — Ais our old vector space L(a), and 9 OK. is immediately seen fo be equal i" ee San, be its degree. ‘The purpose Leta bea divisor 0 “mat deg(a) and 1a) have the same order of of this chapte is precise information on /(a) ~ deg(a).. We shall even. a aia tere are lant depending on our field K’alone such that tually prove thatthe leg(a) + 1g + 8M), Ia) = where 5(a) isa non-negative integer, which is O if deg() is sufficiently large (> 2g - 2). ivi vhich we base further computations ‘We now state a few trivial formulas on which we base f later, ICB and C are two k-subspaces of A, and B DC, then we denote by (B : C) the dimension of the factor space B mod C over k. Proposition 2.1. Let a and b be ovo divisors, Then M(a) > ACO) if and only ifa = 6. If this is the case, then 1. (A(q) : A(b)) = deg(a) - deg(b), and 2. (Aa) : A(O)) = ((A(@) + K) = (ACD) + KY) + ((M@) NK): (AD) KY), Proof. The first assertion is trivial. Formula 1 is easy to prove as follows. Ifa point P appears in a with multiplicity d and in © with multiplicity e, then 4 Be. Ifris.an clement of order | at P in Kp, then the index (t~"Kp : ¢-* Kp) is obviously equal tod — e. The index in formula | is clearly the sum of the finite number of local indices of the above type, as P ranges over all points ina or b. This proves formula 1. As to formula 2, it is an immediate consequence of the elementary homomorphism theorems for vector spaces. and its formal proof will be left as an exercise to the reader, , From Proposition 2.1 we get a fundamental formula: (1) deg(a) ~ deg(b) = (Aca) + K: A(O) + K) + Ia) = 10) for two divisors a and 6 suc! ‘h that a Separate the middle index ‘into two fy know that (A ; A(6) + ) is finite. F ty be 1“ oe “. SAR Then funetion in. Let be the divisor ofits poles, and curve having function field pice the same point Q ofthe rational (2) and the e; are by definition the ramification = 0. For the moment we cannot yet inctions of @ and b, because we do not This will be proved later. 9 indices of the discrete value group in k(y) associated with the point Q, and the extensions of this value group to K. “These extensions correspond to the points P,, We shall now prove that the degree 3 e, of cis equal to [K : k(y)]- We denote [K : k())] by Let 21, ...., 2, be a linear basis of K over k(y). After multiplying each z, with a suitable polynomial in k{y] we may assume that they are integral over ky}, i:€., that no place of K which is finite on k{y] is a pole of any z, All the poles of the z; are therefore among the P, above appearing in ¢ Hence there is an integer jo such that z, © L(joc). Let ube a large positive integer. For any integer s satisfying 0s = a — ply we get therefore y's, EL (uo), and so I(y) = (a — fo + Ln Let N,, be the integer (A(yc) + K: A(Q) + K),soN, = 0. Putting 6 nd = yt in the fundamental formula (1), we get oS «) = Ny tel (ue) — 1 Q ZN. +(u - pot In -1 Dividing (2) by and letting 4 tend to infinity, we get ® e, = a. Taking into account the corollary to Theorem 1.4 we get. ‘Theorem 2.2, Let K be the function field of a curve, and y © K a noncon- stant function. If ¢ is the divisor of poles of y, then deg(c) = [K + k())] Hence the degree of a divisor of a function is equal to 0 (a function has as many zeros as poles). Proof. If we let ¢’ be the divisor of zeros of y then c’ is the divisor of poles of I/y, and [K : K(1/y)] = m also. Corollary. deg(a) is a function of the linear equivalence class of a. ‘A function depending only on linear equivalence will be called a class function. We see that the degree is a class function Returning to (2), we can now write pn = Ny + pn — pon tan —d whence Nu Soon —n +1 and this proves that N,, is uniformly bounded. Hence for large #41. Ricmann-Roch Theorem 10 N, = (Ao + KAO) + K) s a positive integer. is contun, because is always @ POSTIVE IME aay ay Boag of Now define a new function Ol tt rmer by Theorem 2.2 and the latter dest) and fa) chs sae eT ca is a k-isomorphism between L (a) and because the map z > 32 for = L(a - ()). “The fundamental Os (Aa) + K A(b) + KY = ra) ~ (0) 1 formula (1) can be rewritten @ Put 6 =O and a = for two divisors a and & such that a 2 6. (A(uc) + Ks A(Q) + K) = (HO) — (0). ‘This and the result of the preceding paragraph show that r(j40) is uniformly bounded for all large Let b now be any divisor. Take a function z € kL y] having high zeros at all points of & except at those in common with ¢ (i.e., poles of y). Then for some p., (z) + we = 6. Putting a = gc in (3) above, and using the fact that (a) is a class function, we get rb) 5 (uo) and this proves that for an arbitrary divisor 6 the integer r(O) is bounded (The whole thing is of course pure magic.) This already shows that deg(0) and 1(6) have the same order of magnitude. We return to this question later. For the moment, note that if we now keep 6 fixed, and let at vary in (3), then ‘A(a) can be increased so as to include any element of A. On the other hand, is index in that formula is bounded because we have just seen that r(a) is Feed: Hence for some divisor a it reaches its maximum, and for this visor a we must have A = A(a) + K. We state this as a theorem. Zanes 23. Diere exists a divisor a such that A = A(a) + K. This ic elements of K can be viewed as a lattice in A, and that there is a neighborhood A(q) latice covere 41M) which when translated along all pois of this This result allows us to spit the i of (A: A(Q) + K) by 5 0 split the index in (1). We denote the dimension becomes (@). We have just proved that itis finite, and (1) © desta) ~ deg(o) = a(6) or in other words ~ 8(a) + (a) — 1(6) §2. The Riemann-Roch Theorem uw (5) Ma) — deg(a) — (a) = I(b) — deg(b) — (6). This holds fora = 6. However, since two divisors have a sup, (5) holds for any two divisors @ and 6. The genus of K is defined to be that integer g such that Ma) = deg(a) ~ (a) = 1 = g. Itis an invariant of K. Putting a = 0 in this definition, we see that g = 5(0), and hence that g is an integer = 0, g = (A: A(O) + K). Summarizing, we have Theorem 2.4. There exists an integer g = 0 depending only on K such that for any divisor a we have Ma) = deg(a) + 1 = g + (a), where 5(a) By a differential A of K we shall mean a k-linear functional of A which vanishes on some A(a), and also vanishes on K (considered to be embedded in A). The first condition means that A is required to be continuous, when we take the discrete topology on k. Having proved that (A : A(a) + K) is finite, we see that a differential vanishing on A(a) can be viewed as a functional on the factor space A mod'A(a) + K3) and that the set of such differentials is the dual space of our factor space, its dimension over k being therefore 6(a). Note in addition that the differentials form a vector space over K. Indeed, if A is a differential vanishing on A(q), if € is an element of A, and y an element of K, we can define yA by (yA)(€) = A(yé). The functional yA is again a differential, for it clearly vanishes on K, and in addition, it vanishes on A(a + ())). Lrerrnreer) ; We shall call the sets A(a) parallelotopes. We then have the following theorem Theorem 2.5. If Ais a differential, there is a maximal parallelotope \(a) on which A vanishes. Proof. If A vanishes on A(a,) and A(a), and if we put a = sup (1, 2)1. Riemann-Roch Theorem 2 «to prove our theorem it will suffice © proye nee © then A vanishes on AC) Hh fy € L¢a),80(9) an then 3a vanishes that the degree aus contains A(0) because & + (ay) : Si 1S te on A(a +fy)) which vYnA. Hence we'get Tincarly independent over &, ther so are grt t—g + 800) §(0) 2 10) Since (a) = 0, it follows that dega = 5(0) +8 — 1, which proves the desired bound. ‘Theorem 2.6. The differentials form a \-dimensional K-space. Proof, Suppose we have ovo differentials A andj which are linearly independent over K. Suppose x}, ~~» ty and yy, + Yn AFe twWO sets of elements of K which are linearly independent over k. Then the differentials iA, sy 4eAs Yadly «- » Yugb are linearly independent over k, for other. wise we would have a relation Daud + Dd bye =0. Letting x = Eajx,and y = ¥ b,y,, we getx + yu = 0, contradicting the independence of A, over K. Both A and vanish on some parallelotope A\ a (a), for if A vanishes on A(a,) and 2 vanishes on A(a2), we put a inf (a, a), and Al@) = A). Ay) Let 6 be an arbitrary divisor, If y ishes on A(a + (y)) which contai Similarly, yy vanishes on A(q — ginning of our proof, we concl €L(b), so that (y) = —, then yA van- ins A(a ~ 6) because a + (y) 2a — b ) and by definition and the remark at the lude that 8(a ~ 6) = (6). Using Theorem 2.4, we get Ha ~ b) ~ degiay 4 deb) — 14g = 210) = 2 (deg(o) + 1 9 + 3) = 2 deg(b) + 2 ~ 29. $2. The Ri 13 If we take b to be a positive divisor of vei consists of 0 alone, becau: Since deg(a) is constant in the thereby prove the theorem. or of very large degree, then L(a — 6) a function cannot have more zeros than poles. we above inequality, we get a contradiction, and If A is anon-zero differential, then all differentials are of type yA. If A(a) is the maximal parallelotope on which A vanishes, then clearly A(a + (y)) is the maximal parallelotope on which yA vanishes. We get therefore a linear equivalence class of divisors: if we define the divisor (A) associated with A to be a, then the divisor associated with yA is a + (y). This d visor class is alled the canonical class of K, and a divisor in it led a canonical Theorem 2.6 allows us to complete Theorem 2.4 by giving more informa- tion on (a): we can now state the complete Riemann-Roch theorem. Theorem 2.7. Let a be an arbitrary divisor of K. Then Na) = deg(a) + 1 — g + Kea). where ¢ is any divisor of the canonical class. In other words, (a) = Ic — a). Proof. Let ¢ be the divisor which is such that A(c) is the maximal paral- lelotope on which a non-zero differential A vanishes. If 6 is an arbitrary divisor and y € L(b), then we know that yA vanishes on A(c — 6). Con- versely, by Theorem 2.6, any differential vanishing on A(c — 6) is of type 2A for some z € K, and the maximal parallelotope on which A vanishes is (2) + ¢, which must therefore contain A(¢ — 6). This implies that —b, ie. z EL(6). We have therefore proved that 6(¢ — 6) is equal to /(6). The divisor 6 was arbitrary, and hence we can replace it by ¢ — a, thereby proving our theorem, Corollary 1. if ¢ is a canonical divisor, then I(c) = g. Proof. Put a = 0 in the Riemann-Roch theorem. Then Z (a) consists of the constants alone, and so /(a) = 1. Since deg(0) = 0, we get what we want, Corollary 2. The degree of the canonical class is 2g — 2. Proof. Put a = ¢ in the Riemann-Roch theorem, and use Corollary 1.1. Riemann-Roch Theorem, 4 en 8(a) = 0. orollary 3 If deg(a) > 2¢ ~ 2, then (a) = a). Since (a) is equal (0 1 a). : Paine L(c ~ a) = 0 if deg(a) > 28 ~ 2 iction cannot have more zeros §3. Remarks on Differential Forms ‘A derivation D of a ring R is a mapping D: R > R of R into itself which js linear and satisfies the ordinary rule for derivatives, i.c., D(x + y) = Dx + Dy, and D(axy) = xDy + yDx As an example of derivations, consider the polynomial ring A[X] over a field k. For each variable X, the derivative 4/AX taken in the usual manner is a derivation of k{X]. We also get a derivation of the quotient field in the obvious manner, i.e., by defining D(w/v) = (vDu — uDv)/v? We shall work with derivations of a field K. A derivation of K is trivial if Dr = 0 for allx © K. It is trivial over a subfield k of K if Dx = 0 for all x Ek, A derivation is always trivial over the prime field: one sees that D(1) = Di 1) = 2D(1), whence DiI) = 0. We now consider the problem of extending a derivation D on K. Let £ = K(1) be generated by one element. Iff © K[X], we denote by af/ax the Polynomial //AX evaluated at x. Given a derivation D on K, does there exist 8 derivation D* on K(x) coinciding with D on K? If f(X) © KIX} i Polynomial vanishing on x, then MT isa is + then any such D* must satisfy Oo 9 = DFC) = (PU +S (afjaxn*x, t if relation (1) is satisfi Benerators of the ideal in K[X] vanish; Polynomial of this ideal. Thi derivations. ‘The above Sufficient, Lemma 3.1, Ler D Ls Let D be a deri netenson eld of servation Of a field K. Let x be any element in by xin K(x], Then, tu Pe. be 4 Benerator for the ideal determined ement of K(x) satisfyin, ie s 1g the equation ied for every clement in a finite set of is ig ashing on x, then (1) is satisfied by every ‘an Immediate consequence of the rules for necessary condition f; ‘ary condition for the existence of a D* turns out to be = PU) + FC, §3. Remarks on Differential Forms 15, there is one and only one derivation D¥* of K(x) coinciding with D on K, and such that D*x = w Proof. The necessity has been shown above. Conversely, if g(x), h(x) are in K[x], and h(x) # O, one verifies immediately that the mapping D* defined by the formulas Dra a) = 2°C) + g(du *g — gD*h e D*(g/h) = is well defined and is a derivation of K(x). Consider the following special cases. Let D be a given derivation on K. Case 1. x is separable algebraic over K. Let f(X) be the irreducible polynomial satisfied by x over K. Then f'(x) #0. We have 0 = s(x) + fu, whence u = —f°(x)/f'(x). Hence D extends to K(x) uniquely. If D is trivial on K, then D is trivial on K(x). Case 2. x is transcendental over K. Then D extends, and u can be selected arbitrarily in K(x). Case 3. x is purely inseparable over K, so x? — a = 0, with a EK. Then D extends to K(x) if and only if Da = 0. In particular if D is trivial on K, then u can be selected arbitrarily. From these three cases, we see that x is separable algebraic over K if and only if every derivation D of K(x) which is trivial on K is trivial on K(x). Indeed, if x is transcendental, we can always define a derivation trivial on K but not on x, and if x is not separable, but algebraic, then K(x?) # K(x), whence we can find a derivation trivial on K(x?) but not on K(x). The derivations of a field K form a vector space over K if we define zD for z EK by (2D)(x) = 2Dx. Let K be a function field over the algebraically closed constant field k (function field means, as before, function field in one variable). It is an elementary matter to prove that there exists an element x © K such that K is ; separable algebraic over k(x) (cf. Algebra). In particular, a derivation on K 3 is then uniquely determined by its effect on k(x). We denote by ® the K-vector space of derivations D of K over k, (deri- vations of K which are trivial on k). For each z © K, we have a pairing (D, 2) Dzae Roch Theorem, 1. Riema 16 «a, K) into K, Each element: of K gives therefore a K-linear functional into K. : cs. This functional is denoted by dz. We have of the dual space of D, if we These linear functionals form a subspac define ydz by ), ydz) > yDz. Lemma 3.2. If K is a function field (in one variable) over the algebraic ally closed field k, then @ has dimension | over K. An element t & K is ssnch that K over k(t) is separable if and only if dt is a basis of the dual space of @ over K. Proof. If K is separable over k(1), then any derivation on K is determined by its effect ont, If Dr = u, then D = uD,, where Dy is the derivation such that Dj = 1. Thus % has dimension 1 over K, and dr is a basis of the dual space. On the other hand, using cases 2 and 3 of the extension theorem, we see at once that if K is not separable over k(), then dr = 0, and hence cannot be such a basis. The dual space F of & will be called the space of differential forms of Koverk. Any differential form of K can therefore be written as ydx, where K is separable over k(1x). $4. Residues in Power Series Fields The results of this section will be used as lemmas u i S to prove that the sum of the pen eitferetia form in a function field of dimension | is 0... 2) be a power series field, the field of coefficients being arbi t a ing arbit (00 necessarily algebraically closed). If wis an element of that beld which Watten w= at + ast? +++ with a, # 0, then it is clear that Kw) = k((0). De de om element of k((0) can be computed in terms of u or of 1. We eat [order | a local parameter of k((1)). Ae es feta admits @ derivation D, defined in the obvious ately that Dy = 3 epee 8 ah clement of k((0) one verifies immed Afdt. There is also nderve et etivation. We sometimes denote Dirt 80 a derivation D, y defined in the same manner, and the 7 4, Residues in Power Series Fields classical chain rule D,y-D,u = Dyy (or better dy/du- dujdt = dy/dt) holds here because it is a formal result Ify = 2 a,t”, then a; (the coefficient of 1) is called the residue of y with respect to 1, and denoted by res,(9). Proposition 4.1. Let x and y be two elements of k((1)), and let u be another parameter of k((). Then es,(y 2) = rosy PAY aa) ~ S\9 ar) Proof. It clearly suffices to show that for any element y of k((t)) we have res,(3) y dt /du). Since the residue is k-linear as a function of power series, and vanishes on power series which have a zero of high order, it suf- fices to prove our proposition for y = 1 (n being an integer). Furthermore, our result is obviously true under the trivial change of parameter ¢ where a is a non-zero constant. Hence we may assume f = u + aju? + and de/du = 1 + 2a + We have to show that res,(t" dt /du) when n = —1, and 0 otherwise. i When n = 0, the proposition is obvious, because 1" dr/du contains no negative powers of f When n = —1, we have 1 + 2anu + w+ ae + aa 1 du and hence the residue is equal to 1, as desired. : When n < —1, we consider first the case in which the characteristic is 0. In this case, we have a A (AN pet res, (t" difdu) = res, (g G a! )) and this is 0 forn # —1. For arbitrary characteristic, and fixed n < —1, we have form > 1, 1 dt 1+ + du u™(1 + au + Flan ay) ..., a i i i fficients. [tis where F(a, as, « . .) is a formal polynomial with integer coe! ‘ the same for all fields, and contains only a finite number of the coefficients1. Riemann-Roch Theorem 18 tal consequence of the result case, our polynomial ion. is a form ‘aq. Hence the truth of our proposition is a formal cor Paton 1, because we have just seen that in that Flas, ay, -) is identically 0. This proves the propo jon of type yx (with shall ca resi 1, we shall call an exp! ee field) a differential form of that field and the {ferential form is defined to be the residue parameter f of our field ; hat of Proposition 4. Given at field of order write In view of Proposition 4. “x and y in the power series residue res(ydt) of that di taken with respect to any ee or avr han tof POP a power series field k((u)), let 1 be a non-zero clement of that mms 1. After multiplying 1 by a constant if necessary, we can peut but + bw? + = uM + bu + bw +) in k((u)). In fact, one sees ies field k((O) is contained i BS Love ot is exactly equal to m. immediately that the degree of k((u)) over (2) Indeed, by recursion, one can express any element y of &((u)) in the following ‘manner y= fl) + Alu to + fe COU, with f() € k(()). Furthermore, the elements I, u,..., ware linearly independent over k((1)), because our power series field k((u)) has a discrete valuation where u is an element of order 1, and 1 has order m, If we had a relation as above with y = 0, then two terms fi(1)u' and f, (tu! would neces- sarily have the same absolute value with i # j. This obviously cannot be the ease. Hence the degree of k((u)) over k((1)) is equal to m, and is equal to the ramification index of the valuation in k ((1)) having as an element of order 1 with respect to the valuation in k((u)) having u as element of order 1 The following proposition gives the relations between the residues taken in k((w) or in k((1)). By Tr we shall denote the trace from k((u)) to k(() Proposition 4.2. Let k((u)) be a power series field, and let t be a non-zero element of that field, of order m = 1. Let y be an element of k((u)). Then at res, ( a a) = res, (Tr) di). f. We have seen that the powers 1, u"™! form a basis fe Myr eay asis for ver k((), and the trace of an element y of k((u)) can be computed trix representing y on this basis. Multiplying ¢ by a non-zero € not change the validity of the proposition. Hence we may $4. Residues in Power Series F lds 19 fi We bum ge SUL t bw + bu? 4 + with be Ek. One can solve recursively Ww fol) + Au + + So-i(du, where /,(1) are elements of k polynomials in by, by, be written (1), and the coeftici 4 ‘ents of fi(1) are univer + with integer coefficient, that cach Fre Ll) = & Pb’, where each P,,(b) is a pol finite number of Bs \e matrix representing an arbitrary element Yynomial with integer coefficients, involving only a BOD + BDU Ho & Be s(Qur! of k((w)) is therefore of type Goa + Gomi) Get. 00 + * Gan timi(, where Gyu(t) € k((1), and where the coefficients of the G,,(2) are universal polynomials with integer coefficients in the b's and in the coefficients of the (0). This means that our formula, if itis true, is a formal identity having nothing to do with characteristic p, and that our verification can be carried out in characteristic 0. This being the case, we can write = 0", where © =u + cu? ++ is another parameter of the field k((u)). This can be done by taking the binomial expansion for (1 + b\w + - + -)"". In view of Proposition 4.1, it will suffice to prove that res, (» # de) = res,(Tr() do). By linearity, it suffices to prove this for y = v!, ~co oo). If P is rot the point sending x to infinity, but, say the point x = 4, a € k, then x ~ acan be selected as parameter at P, and the residue of a differential form jars the residue of y in its expansion in terms of x — a. The situation is the ‘same as in complex variables. We expand y into partial fractions, y= D culls — bY +f) where f(x) is a polynomial in k[x]. To get resp (yx) we need consider only the coefficient of (x — a)" and hence the sum of the residues taken over all P finite on x is equal to Sy cys Now suppose P is the point at infinity. Then s = 1/x is a local parameter, and dx = —1/t? dt. We must find the coefficient of 1/1 in the expression =yI/t*. It is clear that the residue at 1 of (—1/17)F(1/o) is equal to 0, The other expression can be expanded as follows: AG sf and from this we get # contrition to the residue only from the first term, ‘Which gives precisely —¥ c,1. This proves our theorem in th mia ve pecs 1 case of a purely Next, suppose we have a finite separable algebraic extension K of a purely field F = k(x) of dimension | over the algebra f lgebraically closed Bild Let Q be a point of F, and ¢ a local parameter at @ in F. “Pea point of K lying above Q, and let u be a local parameter at P in §5. The Sum of the Residues 23 K. Under the discrete valuation at P in K exte th i nding that of Q i ordy ¢ = €-ordy uw. The power series field k((u) isa fale oan ae degree ¢ of k((1)). ‘Thus for each Pw tit algebraic extension of k((1)), and the place on K at P + canonical place of the power series field. MW es Let P= 1... 5) be the points of K lyi ( 8 S lying above Q. Let A algebraic closure of k(()). ‘The discrete valuation of Ha) extends vniguely toa valuation of A, which is discrete on every subfield of A finite over k(( » (Proposition 4.3 of $4). Suppose K = F(y) is generated by one element y, satisfying the irreducible polynomial g(Y) with leading coefficient | over F. It splits into imeducible factors over k((1)), say (D 8(Y) = g(¥) +++ g(¥) of degrees dj(j = 1, ... . 1). Let y,be a root of g,(¥’). Then the mapping y+ yy induces an isomorphism of K into A. Two roots of the same g, are conjugate over k((1)), and give rise to conjugate fields. By the uniqueness of the extension of the valuation ring, the induced valuation on K is therefore the same for two such conjugate embeddings. The ramification index relative to this embedding is dj, and we see from (1) that Sd, = n. By Theorem 2.2 of §2 we now conclude that two distinct polynomials g; give rise to two distinct valuations on K, and that s = r, We can therefore identify the fields &(())(9,) with the fields Kp,. For each i = 1, ... , r denote by Tr; the trace from the field Ky, to Fo. Proposition 5.3. The notation being as above, let Tr be the trace from K to F. Then for any y © K, we have Tr(y) = D Tri(y). a Proof. Suppose y is a generator of K over F. If [K : F] =n, then Tr(y) is the coefficient of ¥*~! in the irreducible polynomial g(¥) as above. A similar remark applies to the local traces, and our formula is then obvious from (1). If y is not a generator, let z be a generator. For some constant c Gk, w=y + cz isa generator. The formula being true for czand for w, and both sides of our equation being linear in y, it follows that the equation holds for y, as desired. ‘The next proposition reduces the theorem for an arbitrary function field K to a rational field k(x). = k(x) be a purely Proposition 5.4. Let k be algebraically closed. Let F = k(x) bea transcendental extension of dimension 1, and K a finite algebraic sepa rable extension of F. Let Q be a point of F, and Pi (i= 1, .-- +1)‘och Theorem 1. Riems spove Q. Lety be an cement af K, and ler TF denote the ins of K lying above Q- Face from K to F. Then reso (Tr() ds) = > resp, (y dx). and let u, be a local parameter Proof, Lett be a loeal parameter at Qin h(s)- <0" to Fp. We have inK at P,. Let Tri denote the local trace from S res, (yd =D resn( =Yeu( ) and using Proposition 4.2 of $4, we see that this is equal to 3 wu (t4(-4) a) Since dx/dt is an element of k(x), the trace is homogeneous with respect to this element, and the above expression is equal to B resq (tae a ) = ¥ req (Te(y) do) = resg () Try) de) i : = esq (Tr(y) dx) thereby proving our proposition, ‘Theorem 5.2 now follows immediately, because a differential form can be written ydx, where K is separable algebraic over k(x). Our theorem will allow us to identify differential forms of a function field Kwith the differentials introduced in §2, as k-linear functionals on the ring pak xs which vanish on some A(a) and on K. This is done in the Slowing manner. Let £=(..., &,...) be an adele, Let y a “differential form of K. Then the map “ wows A: ES resp (Gsyde) , map of A into k. Here, of course, i and x as elements of Kp, ‘of our sum are 0, in the expression resp (yt), - Itis also clear that all but a finite number 45. The Sum of the Residues 25 Our k-lincar map Vanishes on some A(a), because the di a e the differential fc s only a finite number of poles. ‘Theorem 5.2 shows that it vanishes on Kh is therefore a differential, and in this way we obtain an embedding of the K-vector space of differential forms into the K-vector space of differsntials, ince both spaces have dimension | over K (the latter by Th this embedding is surjective 'y Theorem 2.6 of §2), Let ydv be a differential form of K. If P is a point of K, we can define the order of ydx at P easily. Indeed, let t be an element of order 1 at P. In th power series field k(()), the element ne dx va isa power series, with a certain order mp independent of the chosen element £. We define mp to be the order of ydx at P, and we let the divisor of ydx be (yd) =D mp. Suppose ordp(ydx) = mp. If ordp(é») = —mp, then ord, (Eydv) = 0, and the residue resp (& ydt) is 0. Hence the differential A vanishes on A(a), where a = (ydx). On the other hand, if A(b) is the maximal parallelotope on which A vanishes, then A(b) D A(a), and 6 2 a. If 6 > a, then for some P, the coefficient of P in b is > mp, and hence the adele (0,0, Ie", 0,.0,. - 2) lies in A(6). One sees immediately from the definitions that resp(e-™~! ydx) # 0, and hence A cannot vanish on A(b). Summarizing we have ‘Theorem 5.3. Let K be a function field of dimension | over the algebra- ically closed constant field k. Each differential form ydx of K gives rise to a differential a € DS rese (Gnd, a and this induces a K-isomorphism of the K-space of differential forms onto1. Ricmann-Roch Theorem 6 ore, the divisors (ydx) and (A) of §2 rentals. Furtherm? the K-space of die 1. The integer (a) is the dimension of the space of differenti Corollary 1. The int _forms w such that () no poles, that is if frst kind if it ha ‘A differemial form cvs said to be of first Kl nd if it (w) 20. 1 forms of first kind is denoted by dfk. For any divisor The space of differenti rential forms «such that («) = —a. Then 4, let Diff (a) be the space of diffe dim Diff (a) = 5(—a). fa = 0 it is clear that Diff (a) contains the space of differentials of first kind. Corollary 2. For any divisisor a > 0 we have 6(—a) = dega— 1 + 8 and dim Diff(a)jatk = deg a — 1 Proof. Since I(—a) = O because a function which has no poles and at least one zero is identically 0, the formulas are special cases of the Riemann-Roch theorem, §6. The Genus Formula of Hurwitz The formula compares the genus of a finite extension, in terms of the ramifica- tion indices. ‘Theorem 6.1. Lek be algebraically closed, and let K be «function field with kas constant field, Let E be a finite separable extension of K of degree 1 oftt fe and gx be the genera of E and K respectively. For each point P of K, and each point Q ofE above P, assume that the ramification index £018 prime to the characteristic of k.' Then Be ~ 2 = neu ~ 2) + Sle =v. @ £7. Examples 2 Proof. I w is any non-zero differential form of K, then we know that ity dares sau 2. Such a form can be writen as ys, with y © Ke We can also view x, y as elements of Ewe can compute the decree ine compare it with that in K to get the formula, as follows. Let p be a cinat A, anu let be a local parameter at P, that is an element of onder Thar P oy K. Iwis a local parameter at Q. then b=ule, where vis a unit at Q. Furthermore, dr =u’ de + eu®'e die. Hence Ordg (yds) = &o*ordy (yd) + (eg — 1), Summing over all Q over P, and then over all P yields the formula, §7. Examples Fields of genus 0. We leave to the reader as an exercise to prove that k(x) itself has genus 0. Conversely, let K be a function field of genus 0 and let P be a point. By the Riemann-Roch theorem, there exists a non-constant function x in L(P), because WPy=1+1-040= and the constants form a I-dimensional subspace of L(P). We contend that K = k(x). Indeed, x has a pole of order 1 at P, and we know that [K : k(.)] is equal to the degree of the divisor of poles, which is 1. Hence we see that K is the field of rational functions in x. Fields of genus 1. Next let ’be a function field of genus 1, and let P again be a point. We have 2g — 2 = 0, so the Riemann-Roch theorem shows that the constant functions are the only elements of L(P). However, since deg(2P) = 2, we have 1QP)=2+1-1=2, so there exists a function x in K which has a pole of order 2 at P, and no other pole. Also IGP) =3+1-1=3, so there exists a function y in K which has a pole of order 3 at P. The seven functions 1, x, x2, x°, ay, 9, y? must be linearly dependent because they all lie in LP) and1. Riemann-Roch Theorem -1=6 MP) = 6 + 1 « coefficient of y? cannot be 0, for Ins re, ge sees 0M the Parity of the Ina cation fan isis Posie es of epee ofthe disor FD (KK) Similarly, 1K kU) «field between K and k(x), and since y is not ‘There is no strictly intermedi in k(2), it follows that k(x, 9) Furthermore, the elation of linear dependence between the above seven functions can be written cy Hay HGH? + CN HON + Ge In characteristic ¢ 2 or 3, we can then make simple transformations of variables, and select x, y so that they satisfy the equation ys4e ax 0 familiar from the theory of elliptic functions. Hyperelliptic fields. Let K = k(x, y) where y satisfies the equation =f, and f(x) is a polynomial of degree », which we may assume has distinct roots. Let us assume that the characteristic of k is ¢ 2. Then the genus of Kis Proof. Letf(x) = Il (x ~ a.) where the elements a, are distinct. Then K Points except the points P, corresponding to those points lying above x = oo. At P, the pose first that m is odd. Let ¢ = 1/x so that has is unramified over k(x) at all =a), and also possibly at ramification index is 2. Sup 88, Dilferentials of Second Kind 29 order I at oo in k(x), We write fo =" T] 0 =a) | Each power series 1 ~ ta, hus a square root in k [Lil], while for n odd, the square root of £" shows that k(x, y) is ramified of order 2 at infinity. “The Hurwitz genus formula yields 220-94 D2-)+Q-N=-44n41 Solving for gy yields gy = (n — 1)/2. If n is even, then the ramification index at infinity is 1 and the Hurwitz formula yields ge = (a — 2)/2. This proves what we wanted §8. Differentials of Second Kind In this section all fields are assumed of characteristic 0. A differential form wis called of the second kind if it has no residues, that is if =0 forall P. It is called of the third kind if its poles have order = 1. ‘The spaces dsk and ddtk of such forms contain the differentials of first kind. ‘The Riemann-Roch theorem immediately shows that the differentials of first kind have dimension g, namely (0) = g, equal to the genus. A differential form is called exact if it is equal to dz for some function z Itis clear that an exact form is of the second kind. We shall be interested in the factor space dsk/exact Theorem 8.1. Assume that K has characteristic 0. Then dim dsk/exact = 2, Proof. We define dsk(a) and dtk(a) just as we defined Diff (a), that is, forms of the prescribed kind whose divisor is = —a. Let Py, . . . , P, be distinct points, and let NV be a positive integer such that (V — I)r > 2g — 2. Ifa differential form is exact, say equal to dz, and lies in dskW > P.),1. Ricmann-Roch Thewecn 0 4 ihe points P,of OndeTS AE MOSLN, then inns poles at 0 inoter words 5 ,), and conversely rEL(N~ Ver” U usw © Para ON 1) EP), dskjexact = Ud a Nas above, anid al Choices of poingy al will sulfice © prove that each Factor the theorem cause if that is the ease, then increas id any further contribution t0 dskjexact where the union is taken Over © To prose the theo rants 2 neers!) ‘This will then also prove: P, be distinct points, and let N be a positive Theorem 8.2. Ler Pie Fore integer such that (N ~ UF skjexact = dsk(N S PAL (WW — IE Pd ; Note that HYp)=UN-DEP)=1, dim dd. ( tecause the only functions = such that d: = O are the constants. On the other hand, also note that dik 9 exact = 0, because a non-constant function z has a pole, and so dz also has a pole. By Riemann-Roch (cf. Corollary 5.7) the dimension of the space of dtk having poles at most atthe points P; modulo the differentials of first kind has dimension o(-EP)—g Putting allthis together, we find, dim dskN S PyidL(N — 1) ¥ p) im Diff © Py/dL((W — 1) 5 p) — dim atk P) (NEP) ~ (HN ~ 1) EP) — 1 =r = 1) 19. Funetion Fields and Curves 3 =2% 24 Mr +1—g- Dr-lege1-@-1 = 2 ‘This proves the theorem, §9. Function Fields and Curves For technical simplicity, we assume again that & has characteristic zero. Let K be a function field in one variable over a field k. ‘This rm K is of transcendence degree 1, and fin K = ACx, 9), with two generators x, y, then we may call (x, ») the generi¢ point of a plane curve, defined by the equation f(X, ¥) = 0, if f is the irreducible polynomial vanishing on (x, y), determined up to a constant factor. A point (a, b) lies on the curve if and only if f(a, 6) = 0. We shall say that the point is simple if Dsf(a, b) + 0 Ifo is a discrete valuation ring of K over k (i.e., containing k) and mits maximal idea al, then 11s principal, and any generator of mt is called a local parameter of © or tt. Assume that the residue class field 0/1 is equal to k, Let ¢: 0 > v/it be the canonical map. If K = k(x, y) and x, y € 0, then we let a = g(x), b = g(y). We see that (a, b) is a point on the curve determined by (x, y). Itz & K, z € 0, we can extend gto all of K by letting (2) = co. We call @ a place of K (over k). We say that the point (a, b) is induced by the place on the curve. weans that ly generated. If we can write Local Uniformization Theorem. Let K be « function field in one variable over (1) Let K = k(x, y) where (x, y) satisfy an irreducible polynomiat SUX, ¥) = O over k. Leta, b © k be such that f(a, b) = 0 but D:f(a, 6) # 0. Then there exists a unique place ¢ of K over k such that (x) = a, @(9) = by and if 0 is the corresponding discrete valuation ring with maximal ideal m, then x ~ a is a generator of w. (2) Conversely, let 0 be a discrete valuation ring of K containing k, with maximal ideal, m1, such that ofmt = k. Letx be a generator of m. Then there exists y © 0 such that K = k(x, y), and such that the point induced by the place on the curve is simple. Proof. To prove (1), we shall prove that any non-zero element g(x, ») E kL, 9)1. Rican Roch Theorem 2 writen in the for™ can be suey) coy BD ca, b) #0. This proves U Fing 9 10a A esca9) with ge Fg iyo. Writ if polyno and that rac, $0 asus polynomial ere A, B are wis uotients Taming of al ir gla, 6) #0, we are oO aor bei) a) gi) E ALY] fr) KIL ea) pla yy = (¥ — Oi Then fib) # 0 since Dsfa. 1 # O Hence g(a, YIP) = Fa. V8). Ie follows that 0x, Pf) — 104, EA = OL = OAK Y for some polynomial Ay. Hence glx, 9) = MAL NSC? e we continue in the same way. We IfAy(a, 6) £0, we are done. If not, we continue in t ‘cannot continue indefinitely, for otherwise, we know that there exists some place off over k inducing the given point, and g(x, y) would have a zero of infinite order atthe discrete valuation ring belonging to that place, which is impossible. Conversely, to prove (2), let K = k(t, 2) where = is integral over kx} Lets = 2... (m2 2) be the conjugates of z over k(x), and extend 0 toa valuation ring © of k(x, 21, Let Fat te HaAT + be the power series expansion of z, with a, € &, and let Pils) = ay t+ + ax" Fori =... mlet ———_————— 59, Funetion Fields and Curves 3 no pole at ©, but ys... , y, have potes at © | + Yh Fe conjugate over k(x). Let f(X, ¥) be the ible polynomial of (x, ») over k. ‘Then a It we take + ‘The eles LOS ¥) = OY = Yate) Furthermore, 44(0) + 0 for some i, otherwise we could factor out some power of X from f(X, ¥). We rewrite f(x, ¥) in the form Fo) = Wbdye yet —yo(Ly 1) (tr- 1) \ yn In the valuation determined by ©, we see thatthe coefficient W = ODI --- Ye cannot have a pole (otherwise divide the two expressions for f(x, ¥) by this coefficient and read the polynomial modulo the maximal ideal m of 0 to get a contradiction). If we denote by a bar the residue cass ofan element of 9 mod mt, then O#FG, Y) = (-1)" WY 5). a =0. Welety = y, and ¥ = 6, Then By definitio Dsf(a, b) as was to be shown. Corollary. Let K be finite over k(x). There is only a finite number of valuations of k(x) over k which are ramified in K Proof. There exist only a finite number of points (a, b) such that 0, fla, b)=0 and Dsf(a, 6) = and there exist only a finite number of valuation rings of K such that x does not lie in the valuation ring (i.e., such that x is at infinity). IK = k(x, y) and f(x, y) = 0 is the irreducible equation for x, y over k, then one calls the set of solutions (a, b) of the equation f(a, b) = O'an affine plane curve, which is a model of the function field. [fall its points are simple, the curve is called non-singular. ‘The totality of all places of K (which are K-valued) is called the set of points on the complete non-singular curveaimee 5 in OO 2: a 1. Riemann-Roch Theorem Bt of course not be a single sion Held K st always cxclogs function fel KT ig Ys excludes od! wih ES in AE A fae neton Bld. si of conse PORE oll K-valued places the complete be taken cae Ct ces 10 call tBE SEE ihe curve or are mode, ad 1 dal EN Ae i defined to be the gems of is nonsingular Moses, the genus Of 3 od elements of Rare in bijection with fnction fed Seta othe st of EK con © alation rings Of the discrete oe ne of Ras the points reference 10 K elements of a io specify the “Then the inclusion FC K gives Suppose that Kis rise 10 8 mapping points of K. THe She constant field k. We speak of gs RK) > RP) v of K associates the ring 0. F of F. This 9 of K assets oran affine model. Indeed, if 0, and (a, bis the ich exc vation ng 0 of Kash eat th sarin a hae ception 2) K where 1, © are in O and the point and if F = K(w Os in corresponding 0 0; nd i ue at Pejuces the values (c, d) on (u, ¢), then we may writ w= ltr 2) where gy ¢: ae rational functions whose denominators donot vanish at (a,b). Then c= ela b) and d= ola, b) §10. Divisor Classes Let Sis be the group of divisors of degree 0, and [Jy the subgroup of divisors of functions. ‘The factor group fy is called the group of divisor classes, Suppose that X is a finite extension of F, and let g: R(K) > R(F) be the associated map onthe curves. ‘Then g induces a homomorphism #4: GK) > €F) $10. Divisor Classes 4s Indeed, ¢, is defined to be @ on points, and is extended by Z-linearity to divisors. Itis an elementary fact of algebra that g, maps divisors of functions to divisors of functions. In fact, if = © K then Naw), el in other words, the image of the divisor of (z) under g is the divisor of the norm. For a proof, ef. Proposition 22 of Chapter | [La 2} The map g on divisors also induces a contravariant map et €F) > EK) as follows, Given a point Q of F, let Py, . . . , P, be the distinct points of K lying above Q, and let e, be the ramification index of P, over Q. ‘Then we define Y eth). XQ) ‘Then @* also maps the divisor of a function z in F to the divisor of that same function, viewed as element of K. ‘This is obvious from the definition of the divisor of a function It is also immediate that 59" = (KF), because if 2 © F then Nyp(2 because in the above notation, "where n= [K: F]. Or, alternatively,his chapter isto givea significant example fr the notions ang Ff this chapter isto give asix ced in the first chapter: ‘The purpose of nd in reaching the Abel-Jacobi as fast as possible can of theorems prov “The reader interest course omit this chapter. $1. The Genus We consider the curve defined by the equation over an algebraically closed field k, and assume that A’ is prime to the characteristic of A. We denote this curve by F (W) and call it the Fermat curve of level N We suppose that N’ = 3 and again let K’ be its function field, K = k(x, 9). We observe that the equation defining the curve is non-singular, so the discrete valuation rings in K are precisely the local rings of points on the ‘cure, including the points with x = oo, arising from the projective equation Wty with z = 0, We have the expression 2. Ditteretials a where the product is taken over all £ jy (N-th roots of unity). Ix is set equal fon W th root of unity, then we obtain a point on the Fermat curve with y = 0, and K is ramified of order N over this point, asi clear from the above equation. Hence IK: kO)] and the ramification index of K over the point x = Cis On the other hand, let ¢ = Ix. Then Vey Re -D, ss aunit in k{L¢l]. Hence x = co (or = 0) is not ramified in A, and there exist N distinct points of F(N) (or K) lying over x ~ so, called the points at infinity in shis section. If we putz = ty, then these N points have coordinates; =o for CE py By the Hurwitz genus formula, if we let g be the genus of F(N), we get: ~2=-2N + D (@ — 1) = -2N + MN = 1) Hence: (N= IW ~ 2) — a Theorem 1.1, The genus of FIN) is g 2. Differentials For integers r, such that 1 Sr, s we let We can also writeMW. The Ferma Cup 38 lie among the points with x = 4, possible poles of he oats with x a pyistinne e expression wesc tha ‘episa pol esi aut zt Pisa pein sch that ww ape sh | s no pole at ws that dv /y"—' has no pole ai dhje™! shows tat tas Brsion Ho xa ener tO ventials of frst kind is given by w,. wing au. A basis for the ait rent stheoresn 2. A basis for thei Is . vouch that +8 = - = Ix en of, Suppose that x(P) = put = Ii Proof. Is EN = Lthen ordp w,, 2 0. This prov from which itis clearthatifr +5 = proves at the differential forms as stated > itis also clear that the differential ed are Ene, re sh et oe ty ot where g is the genus of F(N). an additional structure. The grou serve that these forms have an ack / p ax nets asa group of automorphisms of F(N) by the action (x, > (Gh Shy) where is fixed primitive N-th root of unity. Over the complex numbers, wwe usually take dy = 7°. Then the form w,, (without any restriction on the integers r,s) is an eigenform for the character X,,, such that XeslGs PY = Oe. The linear independence of the differentials of first kind in Theorem 2.1 can therefore also be seen from the fact that they are eigenforms for this Galois group, with distinct characters, When we view wy X py a8 a group of automorphisms of F (N) we shall also write it as G = G(N), and cal it simply the group of natural automor- Phisms of the Fermat curve, or also the Galois group of F (N) over F(I). ‘Theorem 2.2, The forms w,., with 'S55N—1andr +s £0 modN Constitute a basis for dskexact, 43, Rational Lmages of the Fermat Curve 39 Proof. Let 2 = > (x) be the divisor of points above x = © on F(N), taken wit ‘iplici First we note that the space vimien with malig dtk(co)/atk has dimension V — 1 by the Riemann-Roch theorem (Corollary 2 of Theo- rem 5.3, Chapter 1). Checking the order of pole at infinity shows that the forms , with r+s=N and ters are of the third kind, and obviously linearly independent from the differentials of first kind in Theorem 2.1. Since they have the right dimension, they form a basis of dtk(oo)/dtk Given any differential c, it follows that there exists a homogeneous polynomial A(x, ») of degree N — 2 such that @ — h(x, y) is of the second kind. We apply this remark to the forms w,, withr + 5 #0 mod N. We operate with (Z, £) on the above difference, and note that the automorphisms of F (N) preserve the spaces of dsk, Subtracting, we then find that (1 -)o,, with r +s ¢0modN is of second kind, whence w,, is of second kind. Finally, we note that the forms w,, with | = r,s SN —Landr +5 #0 mod N, taken modulo the exact forms are eigenforms for the Galois group uw X My with distinct characters, and hence are linearly independent in dsk/exact. Since the number of such forms is precisely the dimension of dsk/exact, it follows that they form a basis for this factor space, thus proving the theorem. §3. Rational Images of the Fermat Curve Throughout this section, we let 1 =r, sandr +s SN — (r, s) will be called admissible. We shall consider rational images of theUL. The Fermat Curve 40 ‘ function fiekd) following Robrlich [Ro] after ts function Fermat curve (subfields of it Faddeev (Fa 2] We put west and? We let p= pedr, s.N) and M = N/D We write per'D,s =s'D sothat (7, 8's M) Then u, v are related by the equation wl, which, in irreducible form, amounts 10 aw (= wy We let F(r, s) be the “non-singular curve” whose function field is k(u, v), so that we have a map FIN) > Fr. 8), given in terms of coordinates by (> GH, 0) = (0%, ay). Ht € LM) = Z/MZ, we let (t)y be the integer such that S(t SM —-1 and (thy =1modM. IF M = N we omit the subscript M from the notation. If we let K(N) and K(r, s) be the function fields of F(V) and F(r, 5) respectively, then K(N) is Galois over K (r,s). Let G(r, s) be the Galois group. K(N) = k(x, y) | Gr.) G~ by & by 1 Kr, 8) = ku, v) K() = kw 53. Rational Images ofthe Ferma Curve a We note that G(r, s) is the kernel of the character X,,,, and that 0 Crass at K(r, s) ove K(1) is eyclic, of degree Mf. It is in fact a Kummer entensions owe Special ease. Consider the case when N = p is prime 3 and so the intermediate curve is defined by v= ull - w), which is therefore hyperelliptic, The change of variables t=-1 changes this equation to 1 4vr, which is often easier to work with. Let m © Z(M). We say that m is (r, s)-admissible if (mr) and (ms) form an admissible pair, that is 1-5 (mr), (ms) and (mmr) + (ms) = N = 1. ‘Theorem 3.1, A basis of dfk on F(r, s) is given by the forms my for all (r, s)-admissible elements m. Proof. It is clear that xy" lies in the function field of F(r, s), and hence that the forms listed above are of the first kind on F(r, s). Conversely, suppose w is a dfk on F(r, s). Write 0 =D Cy Oye where the sum is over all admissible pairs (g, ). Since w,,, is an eigenform of tay x sy with eigencharacter X,,, and since the factor group (Hw Hu )/ Ker Xs is cyclic, it follows that if c,, # 0 thenHU. The Fermat Curve = (ns), as desired, wr) and = (M18), 3 erm, But then g = (™ for some integer n € 2M) ge that mE ZO. ve and that m is (" sadmissible. Then the of Flr. 9) are equal, for instance becaus ani (ans) are ea) function fields of F(r- Ker X4 = Ket Now aurves in terms of coordinates is given as ance between the curves iM ‘The correspondence follows. Letlsms is (7, s)-admissible. Write a ~ 1 be prime to Af an such that ts residue class mod yy Cr), (sy) = mtr, 3) + NG DD some pair of integers jj. Then we have a commurative diagram: with 50 Si FN) gs F(r, 3) > (denn), (ins) where the bottom arrow is given by (u, 0) (uy, vt" = 0)/), realizing the automorphism of the function field corresponding to the wo models F(r, 5) and F ((mr), (ns)) Two admissible pairs (r, 5) and (g, #) are called equivalent if there exists m © Z(M)* such that g = (mr) and r = (ms). It is clear that inequivalent Pairs correspond to distinct subfields K(r, s) and K(g, #). On the other hand, if (r, 5) and (g, £) are equivalent, then Kir, ) = K@q, 0. , Civen the admissible pair (r, s), and m € Z(M)*, we observe that there i precisely One value of m or —m such that ((mr), (ms)) is admissible. This ollows at once from the fact that for any integer a 3 0 mod N we have (a) + (~a) = N. We shall now is prime = 3.” “PPLY this to the most interesting special case when N = P $4. Decomposition of the Divisor Classes 43 ‘Theorem 3.2. IPN = p is prime 23, then for every admissible pair (r, 8) the curve F(r, 5) has genus (p 1/2, and K(r, s) = K(L, s*) for uniquely determined integer s* such that te pair (1, s*) is adiaisointe Proof. The enus can either be computed directly as we did for the Fé : n ‘ermat curve, or one can use Theorem 3.1, ‘The number of m such that (mr), (ms) is admissible is trivially computed to be (p — 1)/2, using the remark preced- ing the theorem. The statement that K(r, 5) = K(L, s*) is clear Of course, the equivalen instead of (1, s*) we could also have picked a representative in ss of (r, 5) to be (r*, 1). Remark. If we define F(p — ow — Ww, then F(p — 1, 1) has genus 0, since it is also defined by =u u where w = v/u, and (1 ~ w/u is a fractional linear transform of u, so a generator of k(u). The function field of F(p — 1, 1) is therefore equal to E(w). $4. Decomposition of the Divisor Classes Let Fy = F(1, 4) fork = 1,...,p —2and let fi Fp) be the associated rational map. As we have seen in Chapter 1, $10 there is an associated map f., on divisor classes, as well as an inverse map fit. Let & = ‘G(F,) be the group of divisor classes on F;, and let ra f=Oh: 67 OG be the direct sum of the fi. Also let 2 ft: 8 > 6 be the mapU. The Fermat Curye 44 over k = C We shall give the 41. : ‘Theorem he automorphisms of FUY) induced by (x,y) Proof. Let A and ehelre We use the same letters for the induced (G53) and 9) feiisor class group. For any ape 0 of degree 0 on automorphisms oF ions fof the direct image and inverse image give ys F(p) the elemental the formula Tack . = (Aaya). frohya = §, 6 Hence WS yokipl Bae” . id on divisor classes. We need the fac that this is ea diferenials of frst kind. Any general theory will prove ae reader can deduce it for instance from the duality of Theorems Fe ae 6 n Chapter IV. We assume this fact. Then it suffices to prove ate desired relation when the map is applied to the differential forms wo, with pecs ep ~ 1. Since such forms are eigenforms for the Galois group of F(p) over F(1), we see thatthe above relation is equivalent to the relation prof Since r +s Sp — 1 we see that for each pair r, s there exists a unique k satisfying ks = r (mod p). For this value of k, the sum over jis equal to p, andp — p = 0. For the other values of k, the sum over j is equal to 0. ‘This proves the theorem. The theorem gives us a sequence of maps ~ ies divisorele ns tion isp id. In Chapter IV, we shall prove that the group of 'S isomorphic to a complex torus, and it is clear that f, /* @% $4. Decomposition of the Divisor Classes 45 complex analytic homomorphisms, over the compl s ; 1, plex numbers. Frot rem 4.1 we conclude that f* » f has finite kernel, and from the fact tat the dimension of the group of divisor classes is equal to the genus, we see that dim e > dim @,, a ‘Thus f must be surjective, and up to such a homomorphism with finite kernel, we have a decomposition of € into a direct sum of factors corresponding to the curves F; with k = 1, 1p For more general results, ef. Koblitz-Rohrlich [Ko-R].CHAPTER Il The Riemann Surface is chaper isto show how to give a structure of analyig The purpose of I pints on @ curve in the complex numbers, but ye can ples to more genera fields like p-adic numbers.” We ge realy io teed inthe complex case, in order to derive the Abel-Jace i theorem in the next chapter. §1. Topology and Analytic Structure ‘Assume now that kis locally compact. It can be shown that k is the real field, complex field, or a p-adic field. A point P of K is called k-rational ifthe residue class field 0/p of its valuation ring is equal to & itself. The local Uuniformization theorem shows in fact how to interpret primes as non-singular points on plane curves. We let R be the set of all k-rational points of K over 4, and call it the Riemann surface of K over k. We can view the elements of K as functions on R. If P ER, we denote by vp the valuation ring associated with P, and by ily its maximal ideal. If z'E op, then we define 2(P) to be the residue class of z mod mp. ‘Thus 2(P) Ek. If? € Op, then we define 2(P) = co. The elements of k are constant fune- eres aint fet only constant functions. If 2(P) = 0, we say hit and if 2(P) = oo, we say that 2 has a . Foreich x € K. we ae co, we say that z has a pole at P. the compa i {k, co} be the Gauss sphere over k, thats: “Sopse sce btained by adjoining to & a point at infinity. We Et r=q]r, ack 1, Topology and Analytic Strueture i y ‘ @ We embed R in T in the obvious w: ay: An clement P- smbed a oes on the product IP). We topologize R a8 a subspace of I, which amounts to saying thet the topology is the one having the least amount of open sets functions x © K are continuous. ies act soca Wat allie The product is compact, and we contend that R is closed in implies that R is compact losed in. This Proof. Let (ax)scx be in the closure. Let 0 be the set of elements x in K such that a, # oo. Then 0 is a val ition ring, whose corresponding is such that x(Q) = a, for all x. This is easily proved We first note that k C0 because a(P) = a for alla Ek and PER. Letx, y © 0. Then 4x dy # oo. By assumption, there exists P © R such that x(P), y(P), and (& + YP) are arbitrarily close to a,, dy, ays, respectively. For such P, we see that x(P) and y(P) # 20, whence (x + 9)(P) + oo. Hence x + y € 0. Similarly,x ~ y and ay lie in 0, which is therefore a ring. Furthermore, the map x'> a, isa homormophism of o into k, and is the identity on k. This follows from a continuity argument as above. Finally, o is a valuation ring, for suppose x € 0. Then a, = co. Let y =x~!. There exists P € R such that x(P) is close toa, and y(P) is close to a,. Since x(P) is close to infinity, it follows that y(P) is close to 0. Hence a, = 0, $0 y € 0. This proves our assertion. Let P be a point. Let 1 be a generator of the maximal ideal mp, i.e. a lo- cal uniformizing parameter at P. We shall now prove that the map Q-1(Q) gives a topological isomorphism of a neighborhood of P onto a neighborhood of Oink According to the local uniformizing theorem, we can find generators 1, y for K such that the point P is represented by a simple point with coordinates (a, b) ink, and in fact a = 0. Split the polynomial f(0, Y) in the algebraic closure k* of k, Then d is a root of multiplicity 1, so we have £0, Y) =(¥ — YY YE = by The roots of a polynomial are continuous functions of the coefficients. ‘There exists a neighborhood U of 0 in k such that for any element 7 © U, the polynomiabf(7, ¥) has exactly one root in k*, with multiplicity 1, and this root is close to b (in the algebraic closure of k). However, using, for instance, the Newton approximation method, starting with the approximate root b, we can refine b to a root of f(i, Y) in k itself, if we took U sufficiently small, Hence the map Q +> 1(Q) is injective on the set of Q such that (+(2), y(Q)) lies in 4 suitably small neighborhood of (0, 6). Since the topology on R is deter~ mined by the functions in K, and since k is locally compact, we conclude thatIL The Riemann Surface 48 2s a topological isomorphism, ficiently small, the map Q 1@) pone norphisny suffi ; ; a ‘onto a neighborhood of oa isaac onthe F-plane, We observe thy We can choose U Such tt cal uniformization theorem shows tha 1 — 1(Q) is 2 local uniform ative is satisfied by continuity. Indeed, if concerns secon My) £0 for all (a, ") sulliciently close’ tg Daf (a, 6) # 0, ther @, d). | , sng as charts dises on rplanes as above, together with the maps gi oR. proof AP € R, and, waretvo parameters at P, then asa power series expansion in terms of W, SAY and since i algebraic over (1), simple estimates show that this power series is convergent in some neighborhood of the origin. Hence 1Q) = Dany for Q close to 0 in the t-plane, and we see that the functions giving changes of charts are holomorphic. Theorem 1 is valid for any field k which is real, complex, or p-adic. From now on, we shall assume that k = C is the field of complex numbers We have the notion of a meromorphic function on R, that is a quotient of holomorphic functions locally at each point. If f is such a function, we can write it locally around a point P as a power series FQ) =D buoy where ris a parameter at P, and a finite number of negative powers of f may ‘occur. an this way, f may be viewed as embedded in the power series field (9), and thus can be viewed as an adele because it has only a finite humber of poles on R since R is compact. Theorem 1.2. Every meromorphic function on R is in K. Proof. roof, Let L be the field of ‘meromorphic functions on R. If L # K, then Topology and Analytic Structure 81, Topology tic Struct 49 the degree (L: Ke of the factor space of L » cere infinite. We have ° mod K over the complex is (Lt Be = (L + MO): K + AM + (LNAO :K NAO))e ‘The first term on the right is finite as was shown in Weil's proof of the Riemann-Roch theorem, ‘The second is 0 because a function having no pole is a constant (by the maximum modulus principle). Contradiction Theorem 1.3. The Riemann surface is connected. Proof. Let S be a connected component. Let P € S and let z € K bea function having a pole only at P (such a z exists by the Riemann-Roch theorem). Then z is holomorphic on any other component, without pole, hence constant, equal to ¢ on such a component. But z — ¢ has infinitely many zeros, which is impossible since R and § are compact. ‘Theorem 1.4, Let z © K be a non-constant function. The points of K induce points of C(z), and thereby induce a mapping of R onto the z-sphere S., which is a ramified topological covering. The algebraic ramification index ep at a point P of R is the same as the topological index, and the number of sheets of the covering is n =(K:C(2)} Proof. Let P be a point of R and ¢ a parameter at P. Then P induces a point z = a on the z-sphere, and 1° is equal to z — a times a unit in the power series ring C[[¢]]. Since one can extract nth roots in C, we can find a local parameter w at P on R such that uw’ = z ~ a, where a = 2(P). The map u(Q) > u(y = 2(Q) - 4 gives an e to | map of a disc Vp around P onto a dise V, in the z-plane. We shall call such a disc regular for P. We have shown that the topological ramification index is equal to e. As to the number of sheets, all but a finite number of primes of C (2) are unramified in K and, therefore, split completely into n primes of K (by the formula ¥ e; =n). Hence n is the number of sheets. Theorem 1.5. The Riemann surface is triangulable and orientable. Proof. We shall triangulate it in a special way, used later in another theorem. Let z be a non-constant function in K. Consider a triangulation of the z-sphere S, such that every point q of S, ramified in K is a vertex (just addHL. The Riemann Surface 50 ‘A sufticiently high subdivision of the angulation) i operties ¢ following Pre Mertices is ramified, then & jg none of ” i fi S: point, with ramification index en ti such points €0 2 give! thangulation achieves. If Ais a triangle in ‘contained in some regu yual to 1 | in is a triangle with a ram we can now life the wiangulation, umber of vertices on R, and h y to n triangles in ; unio iva point of R ramified above dt and ra parameter at Q on R. We Be ed vertex Q, then AC Vo. First, each vertex lifts t0 a certain of whose vertices is ramified lifts n5,,.we lett be its ramification index amap ty 2a = re has m inverse if we choose ¢ suitably. Each point asi) sym ™ Hence each triangle A with a vertex atq lifts in m ways, 'As to orientability, we can assume that two triangles of our triangulation, none of whose vertices are ramified and having an edge in common, are contained in some regular disc. Secondly, if A has a ramified vertex q, and ‘Ay has an edge in common with A, then A, A, are both contained in V, Now we can orient the triangles on S, such that an edge receives opposite orientation from the two triangles adjacent to it. If A on R covers A, we give B and its edges the same orientation as A. In view of our strengthened conditions, we can lift the orientation, so R is orientable. Let V, E, T be the number of vertices, edges, triangles on the z-sphere. Then denoting by a prime the same objects on R, we get Ban 1 at. On the other hand, let r be the number of primes p of C (z) which are ramified, orp =. Let be the number of primes P of K such that P lies above P v HY 4D my : the Sum being taken over ramified p or infinity, We have 42, Integration on the Rims st =v +S, -n) = -DTe@=y =v —S (er = 1) where this last sum is taken over all primes P of K. Let X, be the Euler characteristic of S, and Xq that of R. We shall prove: Theorem 1.6. Let g be the topological genus of R. Then ' —2n + ¥ (ep ~ 0). Proof. We have X, = By ~ B, + Bz where B, is the i-th Betti number. We have Bo = By = 1. Butalso X,=V —E +T,s0X, =2. Now X = V' — BE" + T’ and by our previous result, this is = AV — > (ep — 1) — ne + nx, — De - D- 2 — 2g, whence But Bj = Bi = | and By = 2g. Hence Xe -2n + > (er - 1) 2g — as desired. Corollary. The algebraic genus is equal to the topological genus. Proof. Both satisfy the same formula. §2. Integration on the Riemann Surface As before, K is a function field over C and R its Riemann surface. san define the notion of meromorphic If U is an open subset of R, we c 0 differential on U, namely an expression of type fx where fis a meromorphicIL The Riemann Surface 52 dx ~ gdvit f/e = dyldx. We We say that J u e 5 and xlies in K. 1 P if, Whenever Fis parameter function on U fal is holomorphic 3 at fferential 1s say that the dif P, and no pole at P. We say that the differential yoint of U. morphic function on U such that imitives differ by a constant, as tion f dx /dt has U if it is so at every P's ‘dx on U is a meror hen two pri at P, then the funet js holomorphic on U ‘A primitive g of f dglde =f. UU is connected, ¢ mW « holomorphic differential ona de V, then « has fis holomorphic on V. Let y be a I-simple: primitive g and * contained in a dise V, let o = fdr on V, let ay=P-@ imitive of @ on V. Then we define [e ‘This number is independent of the choice of Vand g. Indeed, if supp (y) is contained in another disc V;, and g, is a primitive of fdx on V,, then V OV, contains a connected open set W containing y, and g — g, is constant on W. If subd, y = y; + 7: has one more point A, then and let q be a pri (P) — g(Q), an=P-A,dn=A-Q and [eo foe e Ye Hence if yy is contained in a disc V, then i" If y = no; where ois a I-simplex contai i 2 19, i plex contained in some open disc, we define the integral of « over y by li a y by linearity. We get aga a tegral does not change by acubdiciZe Mtearty. We get again that the integ ee is con, Suny |-chain, for r large enough, cach member of the r-th subdivision ~ 0, if some 2, Integration on the Riemann Surface 92 Integra Rictann Surf 3 Cauchy's Theorem. Let w be holomorphic on an o} be a l-eyele on U, homologous t0.0.0n U. Then 2 HEY Jeno 4m where 7 is a 2-chain. For some r, Sd’y has contained in a disc. We have Proof. We have y each one of its simplic Sd'y = Sd'am = Sd’, whence it suffices to prove Cauchy’ s theorem under the assumption that and ‘rare contained in a disc, But then w has a primitive, and the result is trivial Corollary. If U is simply connected, then w has a primitive on U. ‘The pairing or fo gives a bilinear map HAR, Z) x 0,(R) > C, where Q, is the space of differentials of first kind Theorem 2.1, The kernels of this pairing on both sides are 0. In other words, if w is a differential of first kind whose integral along every cycle is 0, then w = 0. Conversely, if y is a cycle such that the integral along y of every dtk is 0, then y is homologous to 0. Proof. As to the first statement, fix a point O on R. Under the hypothesis that w is orthogonal to every cycle, it follows that the association : Po f « o is a holomorphic function on R, whence constant, and therefore that w = 0. The converse is harder to prove and will be done later, Theorem 5.4 of Chapter IV.CHAPTER IV The Theorem of Abel-Jacobi §1. Abelian Integrals i fit is holomorphic every. I be said to be of the first kind i ery. area giemann surface, Such differentials form a vector space over the complex, and by the Riemann-Roch theorem, one sees that the dimension of this space is equal to the genus g of R. From topology, we now take for granted that our Riemann surface can be represented as a polygon with identified sides (Fig. 1). Figure 1 We select a Point O inside the pol} son, differential of first kind (written dh). (which is simply. connected), we define and use it as an origin. Let g bea iven any point P inside the polygon fle) I ° $1. Abelian tntegeals ‘e where Av is any path from O to P lying entirely inside the polygon. Then ¢ valued and holomorphic inside 9 pes - is an arbitrary path on R, not necessarily lying inside the polygon, from O to P, then A~ Art Sma, with suitable integers (and ~ means homologous). Consequently Lon Lerale ‘The numbers alo) =a. =| @ generate an abelian group which will be called the group of periods of g. An integral taken from O to P along any path is well defined modulo periods. The integral over the path A as shown in Fig. 2 Figure 2 is also a period, and we write fe-m@. Consider one of the cycles aj. We can find a simply connected open set U containing a, minus the vertex v. We can now define a holomorphic function f on U as follows. Given any point P on a; — v, we take a path A# lying entirely inside the polygon except for its end point. For any point Q in U we then define f=[e+feWt The Theore "0 of Ay be hi cob $1. The Abelian Imegra 6 the integral fom pio @ being taken om ‘a path fonction /7 er oeorpic OF = path lying entirely in, ri be alee! ‘hich is holomorphic side We can define phic on a, Cincludi & Ou NE the ve ; ert ri a [re ‘ 7V. Periods on the Fermat Curve 4 denote the group of S-units by oughout. We ‘or simply a unit if S is fixed tht ae or simply se the surface obtains R, and let xd by deleting the points of S from pur ace of R’ fee then fs p is holomorphic exp logfop =S°P etG = Aut(,p) be the group of covering automorphisms, and let € G ‘The function 2+ (log fo paz) — (log f« P=) is constant on U. Indeed, s0 (log f«p) « is another choice of a branch of the logarithm, so it ders by a constant since U is connected. For any z € U, we have [doa sen . [4 log fe p(oz) — log f= pt where the integral from z to a7 is taken along any curve (independent of the curve since U is simply connected), and the integral over y is taken on the projection of that curve into the Rieman surface. Therefore there exists an integer L(f, -) such that for all 2, log fc p(az) ~ log fo p(z) = 2miL(f, 0). Lemma 1.1. The symbol L(f, 0) is bimultiplicative in f and 0. Proof. The formulas Lp, )= LU, 0) + Lig, «) and Lif, oo’) = L(f, 0) + Lf @) ‘are trivially proved from the definitions, and the fact that the integral of dlog f » p bet it does: a ia ro ts Of U does not depend on the choice of paths ‘Thus we get a pairing §2, Periods on the Universal Covering Space 5 E x Aw(U, py where AutU, py namely s the factor group of G by its commutator group, G*=G/G Let m be a positive integer, and let f'E E be a unit. Then fp has an m-th root on U, given by tl exp 5, lB =p) If g is any m-th root (any two differ by an m-th root of unity), then for any o EG we get son 1 mi ore! = exp—™ LU, a) where ¢ = e*™"", and hence ot = pling §2. Periods on the Universal Covering Space Let 2 = (wr, |, @,) be a basis for the dfk on R. Again let p:U>R be the universal covering space as in the preceding section. For each oEG=AMU,p) and wEU ‘Then the homology class of p « ly in let 1, be a path in U from u to ou P ‘zis another point, then H,(R’, Z) is independent of the choice of u. Indeed, if we have a “quadrilateral” in U as shown on the figure. Figure 5Y, Periods on the Fermat Curve 6 Its projection in R’ looks like: LP po ” ma Figure 6 ‘We have for any dfk a LuerLee Lr But the imtegral around the quadrilateral of p*« is equal 10 7240, because YU is simply connected. Furthermore, frepete Ie: so the independence of the homology class of ply follows by duality be- tween cycles and differential forms. The association ar class of ply in H(R', Z) gives an isomorphism GIG ~ HR’, Z), where G* is the commutator subgroup. Since we have a surjective homomor- phism A(R’, Z) > H(R, Z)> 0, induced by the inclusion R' C R, we can represent an element of H,(R, Z) by a cycle p<, for some o € G. ‘The period lattice of R is spanned by the vectors [0= [ a= [pro for any choice of u EU, where o ranges over representatives of G/G* 43, Periods on the F mat Conve 1 §3. Periods on the Fermat Curve ‘We apply the considerations of the first two sections to the Fe: rN defined by cetions to the Fermat curve F(N) Note that (1) is isomorphic to projective L-space P', b vr je space P', by the function x. IF we delete the points {0, 1}, we obtain an affine curve (of genus 0). We let ALU C~{0, 1} be the universal covering space. (The choice of symbol A is not accidental arises because itis the classical symbol for the classical function generating the modular function field of level 2-see later in the book.) We let = AU, A). We take the set of points at infinity on F(V) to be the inverse image of 0, 1 on F(1). This is therefore the set of points OLD COD, We, 0), where and € =e, The above coordinates are the projective coordinates (x, y, 2) for the projective equation As before, we let F(N)' be the open affine curve obtained by deleting the points of infinity as above from F(N). It corresponds to an intermediate covering between U and F(1)', and we let (N) = Aut(U/FO)’). ‘The relevant maps are shown in the following diagram. uV. Periods on the Fermat Cury B We let gexep am 9 ¥and Fare N-th roots of Aand | — respectively on U. In the presery so ¥ and Fare N- cease, let us abbreviate La, a) = ao) and LU = A, @) = BO). ‘Then by §1, for « € ® we have chy Feoage and From these relations, we see that such that (0) ain) = set of 7 Blo) = 0 (mod N) ranted from elementary topotogy that @ is a free group on We take for hema It follows that two generators, say A and B. O(N) = unique normal subgroup of ® such that @/P(V) ~= Z(NY°,andan isomorphism is given by a (a(o), B(a)) mod N. It follows also that OW, Thus (WV) is the smallest normal subgroup containing ABA ~'B~', Ay, By; and therefore that (WV) is also generated by all elements generated by A®, BY and YABA"'B™! 7, with y E ; AY and BY But the elements yABA~!B~' y~! are precisely those of the form OpABA™'B™' pa, with o € @(V) and p ranging over a set of right 4 it coset re sentatives of P(N) in, Hence we have proved: " presen Lemma 3.1, A set of generators for ®(N) mod (NY is given by 2 AN BN ABIABA" BABY! with Os j,k SN - 1 43. Periods on the Fermat Curve cr) For the applications, we have wo dtrmine two ge precisely. We selet areal point with Oe yn 1 eeng se Ae Bmore Above Zp, 6 Al). We pick a closed having Winding number 1 around O and lit ito U, erie sche yaa 8 at the point 1 he ot a 8 x TS oy Figure 7 7 ‘Then the lifted path ends at a point Au for son u me uniquely dete! We denote this lifted path by I, 50 that homgtope ty te path C1 1 shown in Fig Ta). Similarly, we construct I 0 that A is the ath shown in Fig. 7(b el cd such that 2 path shown in Fig (b). ‘Then B is uniquely determined such that A » ly is C= (-%, 1, and has winding number 1 around 1. We obtain: L(A, A) L(A, B) = 0, 1, L(l- A, A) L(A, B) = ‘We know from Chapter II that the dfk's are of the form Isrs and r+ssN- a a with y opto, = L297 80% pta,, We recall the definition of the beta function: Bla, b) = f iL = ode. lo ‘Theorem 3.2 (Rohrlich). The period lattice of F(N) relative to the basis {w,,} is spanned by the vectors vn -0rl§)9 G forsj,ksN-1a V. erids om the Ferma Cary 0 vgure of $2 with the representatives Proof, We foltow that the sence ose Y Ive fe get of /@* that we found above: “saa =o f phe l “This reduces the computation of [re we cases for represematives of UN )/@INy" 1 thE CE go =ABA'B. a=A’, ao =B 1p), Lex’ = 2. Then First let o = ABA” B Lets wy Nace’) [oor where (he Bde bye Ae ta)! ely the integral sign is the analytic continuation tial form under a he real segment between Oand 1. ‘Then Al, of the real principal value on is homotopic to the sum of four pieces as shown. The circles have radius ¢, tending to 0, so in the limit the final integral can be taken to be that over the segment, properly oriented, with the appropriate analytic continuations Aer the first tum around 0, the differential form changes by a factor of o7'C =. After the tum around 1, it changes by a factor of ¢*, and thus met £e,.. After the final turn around 0, it becomes £* ,,.- There fore: [i oe@aCle ree ey [lcm = nena 44. Periods on the Related Curvey 81 change of variables in the integ 8e of variables in the integral onthe right immediately gives -a-e0 ~ ery (S i). as desired Ifo = A® of o = B® then a simih lar argument shows that pan feet YY dc because the value is equal to the Beta integeal times le Cte age ‘This proves the theorem §4, Periods on the Related Curves In Chapter II, §3 we had defined the curves F(r, s) by letting us and ov =a"y! ‘Then F(r, s) is the non-singular curve whose function field is Cl, &. Let A, B be the generators of @ which we selected previously, having winding numbers 0 and | respectively around 0 and 1. We let (1, 8) = Aut(U/F(r, 8), where F(r, 5)’ is the affine curve obtained by deleting the points at infinity from F(r, s). We have a Galois diagram: U 1 FIN)’ k(x, y) L | P(r, 8) DN) Fors) kG xy), L Fay kG y")\V. Periods om the Fermat Curve 2 at, BM and A*B~* where sthen (r,s) is generated by AN). A (, BM and wen oma Da Bed rs. ND Theorem 4.1 Rohrlich). The period latice of Fir, 8) és generated by the vectors: drm) _ sm) N 1 (erm - Xt - nya — 1, components of the above vector are indexed where 1s j,k Further. Proof. The last statement follows from (2). Since L is R-bilinear, it follows that £ is real valued, so our theorem is proved. . Theorem 1.2. Ler Sty, y) = E(ix, y), Then S is symmetric. Also the form H(, y) = Elix, y) + E(x, y) is hermitian, so $ is the real part of H Proof. We expand the value for $(x, y) — S(y, 2) in terms of L. We find at once that S(x,y) — S(y, a) = HEC, y) — Elex, 99] Since the left-hand side is real and the right-hand side is pure imaginary, they must both be 0, Hence S is symmetric, and also E(x, y) = El, iy). One proves that iH (x, Ys iH(x, y HG, ) A(x, y) = AQ, 9 ‘as was to be shown H(ix, y) = by direct computation. ‘Thus H is hermitian,Vi. Linear Theory of Theta Funetions 86 ent is a trivial theta valent if their quotient is @ 0 i inde 8 i function in an equiva ing a unique thet in finde @ ermined by additonal normalizing ‘Two theta functions are C: function. We are interested ence class (up to a constant factor) ditions. - eosuppose we have two solutions Ly Li (0 the eq uations: LQ.) = EG») Lx, 9) = Ly(e, 9) — LQ 9 = EC »” - sc and C-linear in x, so C-bilinear. If we change F Then L ~ Lis symmet adratic form, then the associated L changes by a et) where q is a quad! by e” ramet C bilinear form. On the other hand, we have E(x, y) [H(x, 9) - HO, 9) Therefore BAe” Lit, y) differing from L by a C-bilinear symmetric form. is a possible solution, Changing F by a unique factor 7" (up to a constant) makes it so that the associated L satisfies For a further normalization, we still can multiply the function by the exponential of a C-linear term. Define Kw) = JW) ~3 20,0) We shall find a C-linear function A such that Im A = Im K mod Z. From (1) we get 4 Kw +o) KW) + Ko) + SEU, » (mod Z) hs shows that im Kis adlitve on D, wih ales in C72. Since Dis frees } to a C-valued function which is additive on D, and then extend it to an R-linear function on V. Let g =m K, and let gu, Associated Linear Forms 87 AQ) = BG) + igi, Then Mix) = g(—2) + ig(ix) —8O) + ig(ix) = IAD, so that A is C-linear, and K — A is real valued We may now state our desired normalization, We shall say that a theta function is normalized if it satisfies the cond: atisfies the condi- tions’ 1 Ni L(x, y) = 5 HG 9) N 2. The function K on V is real valued. Iris clear that the normalized theta functions form a subgroup of all theta functions. For a normalized theta function, the basic elation takes the form F(x + w) = F(a) exp (2 Ey Hx) + i Hw, w) + xw}) If we define a map : D > C, (complex numbers of absolute value 1) by nko wu) then the basic property (4) for K can be written wu + 0) = Yayoye*", One sometimes calls the associated quadratic character. The previous discussion proves: Theorem 1.3. /n any equivalence class of theta functions, there exists a normalized theta function, unique up to a non-zero constant factor Proof. We have shown existence. As to uniqueness, We ne uses noted the uniqueness of the quadratic factor, For the Une fein Im K = O then the C-linear function A in the previous discussion 18" shows the uniqueness.Vi Linear Theory of Theta Functions re than that, because they give Us an somorpl ae ns as follows. Let Th(L, J) as is Bren spaces of theta functions a eT J a Te space Pac tta functions of type (Lx J). On the th ea ay he hermitin form associated ‘with , and if y is a functio * ation, let iotute value 1, defined by the above equation, let Than (HW) 1 with L = 3, and In fact, the arguments prove Mme = space of theta functions which are normalized, peer ‘Theorem 1.4. Given (L, J) there exists @ unique (uP 10 constant factor) trivial theta function g such that the map Feeak gives an isomorphism TH, D> Maras # “The next theorem gives additional information in the case of entire theta functions. ‘Theorem 1.5. Suppose that F is a normalized entire theta function. Then (i) There exists a number C > 0 such that Fel sce? — for all EV. (ii) The associated hermitian form H is positive (not necessarily defi- nite) } Proof. As to the first inequality, let 8(2) = F(z)e(-(m/ HG, 2), where e(w) = Then for u € D, Ble + u) = g(zjelMecarrn!, ae spent a perches oh real. Hence |g(z + u)| = |g(2)|. This means + cx it niall rhe ic, and continuous, so that it is bounded. Our Now suppose that H (zo, reese esi 8 St some complex sumber 2. The fune- and from our inequality we see at once that it tends 42, Degenerate Theta Functions 89 a Hence it is ‘ to 0 at infinity itis constant, and this cons Wie, 2) <0 forall sufficiently close to zy, we constant must be 0. Since ME orbood of zy, whence F is identically 0, which is ‘peal one wle. is proves ouFtheoren 2, Degenerate Theta Functions We shall see that the theory of theta functions factors WAV on which the hermitian form is trivial cto through the subspace Spsisting of all z such that #7(z, 2) = 0. Then Vy isa com nek subspace, sailed the kernel of H, or null space of 1. We note that fH es tie fe, w) = Oforallw EV. Indeed, for real rwehave then Hw + 12, w + 2) = Aw, w) + 2 Re H(z, w). Letting 1 be large shows that Re H(2, ww) = 0. Similarly, th ; ei ing i instead of. This proves bm hat Vy ta Vector space, and that the weaker condition H(z, 2) = 0 proves the stronger one as stated “Thus H induces a hermitian form on V/V, and the above property shows that on V/Vi1, H is definite, that is, it has no null space. If is positive, then the induced form on V/Vy is positive definite ‘Theorem 2.1. Let F be a normalized entire theta function on the complex space V, with respect to the latice D, and let H be its associated hermisian form. Let Vy be the kernel of H. Then (i) The image of the lattice D in V/Vy is discrete ii) The values of F depend only on the cosets of Vu. Proof. Let zy © Vy and let x V. By assumption, for any complex 2, We have H(x + 220, % + 220) = H(x *)- ‘The estimate in Theorem 1.5 shows that |F(x + 220)| 5 Ce a Hes, ») z, whence constant, and yu, be R. For Hence F(x + 229) is an entire, bounded function of therefore equal to F(a). This proves (ii). As for Tet ay + + elements of D whose residue classes i, (mod Vy) generate V/V over all x € V sufficiently close to 0, we haveVi Linear Theory of Theta Functions lec, w)| <4 s Fis close to 0, we ifvisan element of D such that € is ¢ i me ae Sarat i, because Ete, ie ae These A ve a 1) ey and hence H(o, 2) = 0 for al enc . for aS a athe mage of Din VIN CS spe Fe 2.1 we may view F asa theta function of V/Vn with respect ones ‘D mod Vy. Its associated hermitian form on V/Vj is that induced tote a postive definite. Ths isthe case which Will Be further analyzed yH, §3. Dimension of the Space of Theta Functions ‘Again we consider a complex vector space V of din De By a Riemann form E on V with respect t0 D- bilinear form vension n, with a lattice we mean a real-valued E:VxV>R, satisfying the following conditions: RF 1. The form E is alternating. RF 2. It takes integral values on D x D RF 3. The form (x, y) > Elir, 9) és symmetric positive. If this last symmetric form is also positive definite, then we say that is a non-degenerate Riemann form. If we select a basis for D over Z, then itis also a basis for Vover R. The matrix representing a non-0, if y #0. a . Hence y = Oby the non-degeneracy condition. Hence x = 0. This proves ‘Observe that the lemma shows that if V' is ess e,ithen the space generated over R by View ai, ‘and that this is an R-direct sum. ssion of the Space of Theta Functions 3. im 93 Let B: V x VC be symmetric and C-bitinear, ana + nd let AVoe be Celinear. Ibis clear that the space of emi RE somorphic to the space of entire theta fun Tideed, if G isa trivial theta function of type fag a quadratic form and a linear form), the theta functions of type (L, J) Rt ORE UL + B,J +4) ) (formed by ex, ntia . nthe association renenti FGF . ism between the t ° gives an isomorphism 1 WO Spaces, its inverse being gi snultiplication with G~", which is of type (-B, A), Hence ne given by dimension of the space © to determine the ThiL, J) of entire theta functions of type (L, J), we shall be able to adjust our L, J to be most convenient. ‘The dimension is now given by the following thegrem, Theorem 3.1 (Frobenius), Let (L, J) be a non-degenerate type with respect to (V, D). Then the entire theta functions on V with respect to D having this type, together with 0, form a complex vector space of dimen- sion equal to the pfaffian of E with respect to D. Proof. We first observe that L is symmetric on {e,, . . . , é,) (notation as above), and hence on the space V’ generated by ¢,,.. . , e, over R. Conse- quently, there is a symmetric C-bilinear form B on V such that L — B is 0 on [e1, ¢,], whence 0 on V'. Similarly, there is a C-linear form A on Vsuch that J — Ais O one, . . . , é,. By the above remarks, this reduces our proof to the case when (*) L(, e) = 0 for all z € V, and J(e) = 0 forall j = 1,..- 5m In this case the conditions defining a theta function are particularly simple, and our theorem can be formulated as follows: pect 10 (V, D). Theorem 3.2. Let (L, J) be a non-degenerate type with resp Assume that e1, 1) B+ + D lew Cal is a Frobenius decomposition, and that (*) holds. Let ¢ = (uy), ant ZE Viet {2,,. . . , 24} be the coordinates of z with respect1o the C-basis: fe... ,e,). Then the space of entire theta functions of pe GisVi Linear Theory of Theta Functions 4 precisely he space of emtre functions satisfying the conditions FG +6) =F@) F(z + 5) = FOE +O)» where e(w) =e, and this space has dimension dy» ++ dy. Where dy = Eley &). two conditions are precisely those defining theta Tattice, and (L, J). We must now construct theta ditions with respect to the e,, we have periodic~ function has a Fourier expansion Sawer, Proof. Ivis clear that the functions with respect to our functions, In view of the con ity, and hence any such theta Fi where the sum is taken over all r © ZS, We shall also write (7, z) instead of the dot product r-z. The second system of conditions now imposes relations among the coefficients a(r), namely alr = de) = alrelr te) ~ 4). where again e(w) = e?"". The values a(r) can be fixed arbitrarily for Osn nde! ‘Suppose that we started with a type (L, J) for which the alternating form Eis degenerate. We are still interested in the dimension of the space of theta functions of this type. We may assume that it is normalized. Then the elements of Th(L, J) induce theta functions on the factor torus V/D, where ¥ = V/Vq, and Vy is the null space of the associated hermitian form. Fur- thermore dim Th(L, J) = dim Th(L, J), where L, J is the induced type on V/D. Now E is non-degenerate, and the pfaffian of E will be called the reduced pfaffian of £, We then obtain: Theorem 3.4. The dimension of Th(L, J) is equal to the reduced pfaffian of the associated alternating form E. “ me Proof. This follows from the theorem, which jimensi er Dis fen , Which gives us the dimension in the 4 ee 2 {et (Ly J) be @ non-degenerate type with respect to (V, D). ¥y happen that there is a bigger lattice D’ for which (L, J) is also @ pondegencrate type with respect to (V, D' Y are limit ee. ie eh to (V, D'). However, such D’ are limited. [Abelian Functions and Riemann-Roch " Roch Theorem on the Torus ” we ae+Y by ms in terms of a Frobenius basis of V. ‘Then Eley bd, and Ele.) = aa, ‘These values have to be integers, Hence a, of values mod Z, and there are only a finit b,can take on only a finite number ite number of possible lattices D’ For each we D', the factor group D'/D is finite equation (1) of 81, there are only a finite number of exter equation (1) imber of exten: By the functional sions of J to maps, Theorem 3.5. Let (L, J) be a non-degenerate type with respect (V, D) Then all theta functions of this type except possibly those ing ina finite union of subspaces of dimension < p§(E) are not theta functions with respect to a lattice strictly larger than D. Proof. If F is a theta function of type (L, J’) with respect to (V, D’), then the pfaffian of E with respect to Dis equal to a (D':D) 0- For any non-degenerate Positive divisor X with Riemann form E, we have Lexar" ph) fimension Proof. This is obvious from Theorem 3-1 because the desired di 's given as the pfaffian of ro + °° * + TaberVic Linear Theory of Theta Functions 100 Corollary 1. The function field of V/D has transc cendence degree at most n= dimeV. abelian functions fi, . . - »fn which are ‘can write them all over a common (after factoring out Proof. Suppose that there exist abel algebraically independent over C. We e the denominator @, which we may assume non-degenera the null space of V/D if necessary). Write G = 4/6 Then the monomials OO, Nt a lie in £(66), whose dimension is given by Theorem 4.1 and is equal to pil). However, the above monomials are linearly independent, and there are m+n m of them, which grows like x". Hence m 0. ‘The number of such monomials is 45, Translations of Theta Functions 101 which is asymptotic to "fn! 35% — oo. Since the dimen mension of (rx, + sy’) Pf (rE + sEy) = r* pf(E) + lower terms, ifllows that ifs > ! (Es, then there exis fence among these monomials.” Such a relatioy for g, with coefficients in CLA, fi) of degree agers cation select an abelian function of maximal degree over the fel CR” ENE then by the primitive clement theorem of elementary eld then. ‘tis awe function must generate the whole field of abelian functions thet en cour theorem. » thereby proving, relation of linear depen- §5. Translations of Theta Functions We now consider the type associated with translations of theta functions. Let a EV. If @isa theta function, we let 8 be the function such that 6,(2) = Ox — a), and call 8, the translation of @ by a. It is obviously a theta function. The definition is made such that if X is the divisor of 6, then X, is the divisor of 6,. Lemma 5.1. (i) If is of type (L, J) then 8, és of type (L, J ~ La), where L,(u) = L(a, w) (ii) If 8is normalized, the type of the normalized theta function in the same equivalence class as , is (L, J — E,) where E,(u) = E(a, w). Proof. We have = exp 2i[L(¢ — a, w) + IW) a is the from whit ion is obvious, Note that the bilinear form L ist ich our first assertion is ot vd hermitian form H is the same for any translation of 8, whence the associate same, - : For the second assertion, if we multiply @ by the trivial theta Function exp {om if Hy, al}ere - car THeory of Theta Functions 102 La «in sean apie af HS Sy ten 2 i a thea a SIO oe peng on Kw) - Ea. © sonmalized theta functions. Wil G90, een H = HI if andl only if 1 be entire 5 Let BO, respective sel oman ff, is (L. J Ly), it syists. Since the type of such a exist : ueness of @ normalized se : goer ‘L’, whence H = foe roast V, be the null space fal ifn, Converse. SUPPOSE TA sn V/V = V by 82. and StH. Then both @ and & induce wwe have the formulas 66+ ors, wy + HEH + KUL: A(x) WF ete ee, + FLO + KOO ve) Note that K’ — K (mod Z) is a with the same L, where €() = €2" homomorphism by (4), $1, and that eink aR ‘Therefore there exists an element depend only on the class 7 of w mod Vi @ € V such that the character e2mikwi-Aual ig given by elk’ — Kw) = eo" ‘We now use the functional equation for 6 and @. Let f = 6'/8,. Then the hypothesis that 4, & are normalized (i.c. that 1 L(x, ») = 3 HC 9D) implies at once that f(x + u) = f(x) forall uw in D, or in other words that f is abelian. This proves the desired property. In terms of divisors, we can express Theorem 5.2 as follows. Let, ty ; Aa Big ee pote cso. Then Ex = Ey if and only if there exists If @ is a theta function, we let Cl function, (8) denote its class modulo the group ever by the vil ta functions andthe bean fanctons. The aco" vtranstations of Theta Functions gs. Trams 103 | of theta functions modulo trivial ones, gr0uP "es, and abelian ones, will be called wepicard group of (VD). oF of VID pwo theta Functions are called algebraicall seston ofthe other (modu wiv ang abelian hes ft wegen 5.2 we see tat this condition isequivae eections). tn light ‘ed Riemann forms (or hermitian forms) eeu ‘toe fea de val. Furthermore, applies also to divisors, via representative ay be said to be algebrai they and only if it satisfies ether one of the ao equivalent 10 equiv conditions y= ¥ = Yo for some positive divisor ¥ and some a € V. tie associ ie potion of algebraic equivalen Meus functions. Thus a divisor alent he associated Riemann form is equal 10 0. tris obvious that the map Go: a> CU(B,/8) or CUX, —X) isa homomorphism of the torus ¥/D into the Picard group, and in fact i h 5 fi the subgroup of elements algebraically equivalent to 270, which is denoted by Pico(¥/D). We shall be interested in its kernel. ‘Theorem 5.3. Let 8 be an entire non-degenerate theta function with Riemann form E. Then the kernel of gis finite in V/D. and is represented by those elements a EV such that E(a, «) € Z for all u € D. We have dy dy)? order of Ker gy = pflE where d), .. . « d, are the elementary divisors of E. Proof. We have seen in Lemma 5.1 that if (L, J) is the type of @, which wwe may assume to be normalized, then the type of the normalized theta function in the equivalence class of 6, is (L, J ~ E,)-_ Suppose that is in the kernel of yp, “Then 6,/0 is equivalent to an abelian function. ‘The type of a normalized theta function in its equivalence class is then (0, -E,) by the above (depending on Lemma 5.1), and it is also (0, 0) sinee an abelian function is normalized. This implies that E,(u) = Ela, EZ forall w € D, But E is Z-valued on D x D. Hence it follows tet if we express a as a linear combination of a basis {ti, « - » tne} for D with real coefficients, then in fact these coefficients are rational and have bounded denominators. ‘This means precisely that the kernel ‘of gis finite in V/D- Using a Froteniue basis, one sees in fact that the denominalos aT the elementary divisors di, . + » di—— =a VI. Linear Theory of Theta Functions 104 , then a fortiori itis equivalent to the al theta function. for tis equivalent to aos 8 ei ee ate sation APES OVE the constant abelian statement on, normalized, with associa ‘Theores «pean entre theta function, normal sed a ra aoe an emhe nll space of He Then Vin + Dis of fie tian form n ahs in the kernel of Ge iV: € non-de Indeed, esults at once from the non-dege! c 5 we Proof This rests 91 O peta function on V = V/V, With tespect to 0 oe iy. Ie, ~ 6, and d, is the normalized theta function cnuvalent oe a. tha ti ts of type (LJ — 6 py Lerume 3 i » Since 6/9 sag tancion, tfallowsthat E(a,D)C Z, But E(a.D) ~ ED), Janae ie ered of @ inthe rons ¥/D. We can then apply Theorem usd 5.3 to conclude the proof. §6. Projective Embedding Let 4 be an entire theta function, and let {4), 1, + Om} be a basis for (4). Then we may view this basis as giving a map Fx (B00), + BaD) of V{D into projective space P”, defined at all those points x where not all 6 vanish simultaneously, and called the map induced by the linear system of We shall see that if the torus has a non-degenerate Riemann form, then there exists such that the above map gives a complex analytic embedding ofthe torus into projective space Instead ofa basis for £(6,), we could just as well take a set of generators forthe vector space $£(4,). We would obtain a map defined at precisely the same points. Itis also useful to remark that the map is defined at x if and only if there exists some @ in (4) such that @(x) # 0. If @is written as a linear combination of basis elements, the condition that 6(x) = Ois equivalent to the Condition that the image of the point lie in a certain hyperplane Let 8be an entire theta function, Let X be the set of its zeros, that is, the set a Points x such that 6(x) = 0. We may view X as a subset of V, or as eee Yo. because it is clear that X is invariant under translations by isclear that eet by X the set ofall elements —x, with x € X. It an the Set of zeros of the theta function 6, that is, the function b 2) Also, for any point a € V, the translation Xs with x € X is the set of zeros of @, (this is the we did, by the formula 6,(z) = @(z — a)). The f such zero sets X, U + = - U Xq is the set of 22r05 ason for defining 6, as union of a finite number of 6, Projective Embedding os of the product 0, * * + 8, of the corresponding ther, jt is not the whole space. ‘Pondling theta functions, In Particular, ‘Theorem 6.1 (Lefschetz), Let 6 be an enti Then the map of ire non-degenerate theta tion: The mith "ap Of VID into projective space indueed byte ane system L(0') is everywhere defined, and is an analytic neue jee eat into projective space. istic embedding of VD Proof. For any points a, b & V, we observe that the function Ox — a(x — BOK + a + by lies in £(0°). To see that the map is well define siven a point x, there exist a, b eh tah he oleae fs ah coats 0, But tis is trivial: We first find a such that x — a does nor In ig ee of zeros of 8, and then find b such that tx ~ B)6(x +a +B) 40. Ineah case this amounts to finding a point not lying ina finite union of cots or san of theta functions. Next we prove thatthe map is injective. In other words, given x, y € ¥ if vand y have the same image in projective space, then x and y differ by a latice point. That the image of x and y isthe same means that there exote a complex number y # 0 such that forall, 2, and all F of the same type as 6, we have Fw — F(x — DF +2 +b) = FO —2F(y — OF(y +2 +). By Theorem 3.5 we can select F in $£(@) such that F is not a theta function ith respect to any lattice strictly larger than D. Let D’ = D + Zo, where x = y. We shall prove that v is of finite order modulo D, and that F is a theta function with respect to D’. It then follows that » € D, as desired. Given any point zo, we can find 6 such that F(x — b)F(x + 20 + B)F(y — DF(y + 20+ 1) #0, and hence such that this inequality holds in a neighborhood of z.. This means that in the neighborhood of zy there is a holomorphic function go having no zero, and such that F(x = 2) = F(y — Dao) in the neighborhood of that point. It is then clear that such functions go are analytic continuations of each other, and therefore that there exists an entire function g without zeros such that for all 2 we have F(x — 2) = F(y -— 28@.-V1, Linear Theory of Theta Functions 106, is formula can be wien im te form With o =x —) pte +) = FM. eros. The theta relation for F now sho ion h without zeros. T id forsome enti re ton, ofthe Form that A is in fact & (ey = Ces ore vis in the kernel of gy and is therefore of iol ‘D by Theorem 5.3. We can easily determine 4 ae “yor Thar Ais of type (0, L.) if (Ly J) is the type of Nw — L(c, a) € Z forall w € D. But 2a) — Lev, W) = Aa) ~ bets 0) ~ BO, 0). where A is C-linear. se) Liew) is real valued because 2, L are Relinear in z, and the tenent a > toate V over R. But A and L are also C-linear, and consequently we must have Mz) = LG, ©). It follows that i a theta function with respect to D’, and therefore concludes the proof. ‘There remains to see that the differential of our projective mapping does not vanish at any point, whence our map is an embedding since V/D is ‘compact. Select any point x, and a function G in (6°) such that G(x) # 0. Any F € (6°) gives rise 10 a possible projective coordinate, such that the corresponding affine coordinate when we dehomogenize with respect to Gis, F/G. To see that the differential of our mapping does not vanish at x, it suffices to prove that given a vector v © V, © # 0, there exists one such function F such that (F/G) # 0. ‘We may take a basis for V such that v = (1,0,... , 0). We have acon = SME) = Feo, * Saypote tha ppose that for every F £06), we have d(F/G)(x)v = 0. ‘Then HG), _ dG Fa) Ga) go, Projective Embedding 107 fora fixed number a, and every F such that FQ) 4 9 system, we have In our coordinate are =F oy, EO), a 1 Foy FO a Let 1 a @=1 4, 19 = WO, wherever defined. We select for F the function F O(: — a)O(z + a + b)0(2 — by, with a, b arbitrary. Then we have f(x —a) + fx -b) + fix+atb =a for all a, b outside an exceptional set where the denominators on the left vanish. Now we consider the function of 2 given by fe = 2) + fe —b) +f +2 +d), which is constant, and differentiate with respect to each variable z;. We get x Sw tz +b). From the right-hand side, we see that these partial derivatives are constant for some open set of z, whence it follows that in an open set of z where f is defined, we have ol es tos beet BF A(z) dz, with constants a, .. . @yy B. Let (2) = bayzh + agzizn to + Mazi 2n + BE ‘Then the first partial derivative of the function a(2)e-* = 6)VE. Linear Theory of Theta Function, everywhere. ‘This means that & (which insome open se nL variables. But 0) is an entire to 6) depends mand we have & contradiction of Theorem 5,3 ident abelian functio, vebraically independent a ions on a algebraic ar ann form. sist nal ° renerate Rie Corollary. There Car des torus VID having a” ve can find abelian functions fi, p y + stihe point, that is, map a neighborhood int on an ese func omen (eve tnalytically independen') retes the proof of the hypothesis we needed in finitely generated extension of a purely Proof: For any point on the fon above corollary comp! $4 nas that the function field ve eanendenl xeon pen define At be an abelian manifla Fhe owns a projective embedding of A. The above theorem proves i ‘Theorem 6.2. 1f(V, D) admits a non-degenerate Riemann form, then VjD is an abelian manifold. “The converse can also be proved (see Chapter X, §3). We do not need it for the rest of this book. Consequently for our purposes, it is more convenient to take as definition that 4 is an abelian manifold if and only if there exists a non-degenerate Riemann form on (V, D) Appendix: The 1-dimensional Case 1k may be useful to the reader if we recall here briefly the situation in dimension 1. To deal with theta functions in this case, we need only use one basic theta function, having a simple zero at all the lattice points. Let D=[w,, w:]. Because of the one-variable situation, we can write down the function, wamely the Weierstrass sigma function, w= I (1-3) o0[E +32) } ti or which 0 has been deleted. Taking the logarithmic Known as the Weierstrass zeta function, namely where D’ derivative yi penx The Fedimensional Cae, 109 S| ‘Taking the derivative once more yields minus th ; ‘ 1e We nv and one sees easily that it is periodic. Hence for any wna’® © function, any period w we get s@ + w) = (2) + mw), where 8 Z-inear in cw. Integrating, and exponentistng that " it follows therefore oz + w) = Wu) exp [n1w(: V1 oe + 2 az), with a suitable function Y(w). Note here that L(z, @) = ylw)z is C-linear in 2, and Z-linear in «, as it should. Thus we see that ois a theta function. It is easy to prove that Wo) = 1 if «2 isa period, W(w) = —1 if 2 is not a period Finally, to get a theta function associated with a divisor, we just take translations of @, and products, for example, T]eG-aa™, it with suitable multiplicities. ‘Thus the theorem to be proved in Chapter IX becomes essentially trivial in dimension 1, because of the Weierstrass product ‘expression for entire functions. . Consider the linear system of divisors 2 —3(0), where 0 is the origin, In that system, we have the Weierstrass ? -function - 92) = £18) And its derivative 2'(z), The general theorem asserts that the coordinates (1,9 @,8'@) ae the affine coordinates of a projective embedding of the torus. Only the Point at infinity (corresponding to the origin) is missing from this representa tion. For more details, look up any book on elliptic functions (e.g. mine).CHAPTER VII Homomorphisms and Duality 1s the elementary theory of homomorphisms and endo. i lan aed First we relate the rational and complex representations to a purely algebraic representation on the points of finite onder, Then we prove the complete reducibility theorem of Poincaré, show. ing that an abelian manifold admits a product decomposition into simple ones, up to isogeny. Finally, we deal with the duality which arises from the nondegenerate hermitian form, and show how the dual manifold corresponds to divisor classes of divisors algebraically equivalent to 0, The duality includes an essentially algebraic pairing between points and such divisors, and a formula inthe last section relates this algebraic pairing with the analytic data and the Riemann form, §1. The Complex and Rational Representations Let D, D’ be two lattices in V, V" respectively. Any complex analytic homomorphism do: VID —> V'/D! can be lifted to a C.) 1a " i lia commutatigg * CTineae map A: V—> V" making the following diagram a v—wy Vib — vip 1. The Complex and Rational Representations um -qnis follows easily by writing a power series ex srigin, and seeing that the additivity property ¢ i Stuegree 1 are equal 10 0 in this power serigy "Th ecu are ‘is C-linear. The global assertion follows, since for goad feat O» Our map ‘large integer N such that x/N is near 0, "Then "MYX © V we ean find (org) =o whence Ay = A (mod D'). Of course one c: standpoint that the homomorphism Ay lifts t of VD. by wae We shall usually use the same symbol A for the map on V and imap on V/D. The ring of complex analytic maps.” mS inaiced "an also see this from the general ‘0 the universal covering space V A: VD > Vp, i.e, of V/D into itself, will be denoted by End(V/D). By the above remarks this ring is represented as a subring ofthe linear endomorphism of V. and eg call this the analytic representation of End(V/D). Its tensor product with the rational numbers Q will be denoted by End(VjD)q. If k © End(V)D) (or End(V/D)q)), and if we wish to distinguish its induced linear map on V, then wé denote the latter by V(A) or Ay, or Ac. Let V/D be a complex torus as before, and let D=[y.. 5 Wan. ‘We now take V = C", that is, we fix a C-basis for V, so that we have complex coordinates for elements of V, and we let e, . . . , ¢ be the unit vectors, viewed as column vectors, so that . 1 0 0 0 (ey. ed = Ce Similarly, we view uy, . . . , Havas column veet6rs, so that (Mi, «+ 5 Han) is ann x 2n matrix of complex numbers. : 5 Let A € End(C'/D). Then A has a representation by ann x n complex matrix C(A) with respect to the basis {e, . . » és}, and we have 1 eC). (Qe de,) = Cr -VIL. Homomorphisms and Dua vality 112 rtiplication is to be understood formally. ‘That is, if ix multip ‘The matrix mul ca) = © then = ee) ( Deven det 7 be understood in the same way. On the I representation Q(A), because A many (ene tions below are (0 similar multiplicat So ave the rational other hand, we also hi D into itself. ‘Thus Ohta ee Atta) = te HOOD, trix. Let U be the mx 2a ma ; is a 2n x 2n rational matrix. Le 2a mati where Q(A) is & wy Hy that iS consisting of the components of U = (m, ‘Then (Ay, oo 6s Altay) = (Cte oo CV CCA and also (Ayes ‘Therefore obtain the relation CMU = UQ(A). Theorem 1.1. The matrix (0) ‘is non-sir " ‘singular (where the bar denotes complex conjugate). Pr itt the above matrix. Suppose M is singular. Then thete iu ‘ector Z (2n-tuple) such that MZ = 0 (we view Z as # ox col Haale inn Yeston). This implies that UZ = 0 and Uz = 0, so that also UZ = 0- 42, Rational and p-adic Representations Hence poth Z + Zand (Z ~ a implies a relation of linear dependence over R heme en 8 0. This, wich is impossible, and proves our assertion, en the column of U, ‘Theorem 1.2. The rational representation direct sum of the complex representation and its conjugen sate. t A> QA) is equivalent 0 the j Proof. By Theorem 1.1, the matrix | e o is invertible, and by the above relation, we also get th ju relation, namely get the complex conjugate COU = TQ). ca _0 \(u) _ (uv ( 0 aia) (e) = (Zou. This proves our theorem. ‘Therefore §2. Rational and p-adic Representations Let p be a prime number. ‘The points of period p’ on V/D constitute the subgroup (/D), 1D. Let 7,(V/D) be the set of all infinite vectors (ay, ay as, «+ +) Such that a, © (V/D), and pa, = ar. Then T}(V/D) is & group under componentwise addition, and is called the Tate group. If = is a p-adicVi. Homenorphisms and Duy, jon of 2 00 Tp(V/D) as follows. We soy 1 operation of 2 08 TH(V/ 5 Select integer, We ee i aethat m= 5, (mod p- a ee ‘a We define an ordinary imteBe* Tependent of the choice of lefine This isi r= ma. 2 (ay + an operation ofZ 00 7,(V/D). which therefore becomes tad clearly is without corsion. nsion 2n over Zp, if 0 is the compley Itis clear that We 8 ‘a module over Zp» Tv /0) is free of dimen Actually. v dimension of V- 1 May be vectors in 7 wed as follows. Letm, as (VID) This is easily proved a aan ons ate linearly independent over ie whose frst components re linearly independent over Z,; for if we vectors ee not Trae independence, we could assume that not all the ta ns are divisible by p. and hence the projection of this relation on the frat component would contradict the hypothesis mace on the ay, Naat pe show that the x; form a basis of Z,(V/D) over Zp. We prove this by av inductive argument. Suppose that we can write every element w of 7,(A) as a linear combination xy 41+ nt (mod p’T,(V/D)), «) with integers 2, € Z. Letw = (bi, 4 brs bry» +»). By definition, we have for the first r + 1 components EAM iy ee ey irs) + + Fae(A2r ty 2 reer) =, + Bp, Brrr) + (Os. - «0, Crs) for some ¢,,) € (V/D),.1. By the very choice of the vectors x, there exist integers di, ... , doy such that Cost =P" igus #6 + day P anasto Tf we replace 21,... , za, by 2) + dip"... 2a, + d-p", we see that We have extended the congruence (1) fi ire igruence (1) from r to r + 1 We define the p-s ; et at deciors the p-adic counterpart of the space V by letting V,(V/D) be the This gives us what we (a, a, 43, ..) such that as € (VD), for some r, and PAs 1 for a |. Projecting (on the first component gives an exact eater all r > 0. Proj 8 TAVD) > VyvID) > (vppy" > 0, onal and p-adic Representations 2. lis ere (¥/D)" isthe union of (V/D), fo apo {eco points of P-Power order on the torus (Pye ee we have a natural isomorphism "Bers r, that is, the 1 =; DID ~ Dip" pr DID = Dip'b obtained by multiplication by p’, Taking the inv isomorphism ‘ese limit over r, we get an 2 T,(V/D) ~ Z, @, D. If A € End(V/D), then we can represent A as a Z,-end 7,(V/D) by the obvious action endomorphism of Ty Ary day. « .) = (Aa, dag, . . .), and, similarly, we get the representation of V,(A) on V,(V/D). This represen- tation of End (V/D) on V,(V/D) is called the p-adic representation. Theorem 2.1. The p-adic representation of End (A) is equivalent to the rational representation. Proof. Obvious from the boxed isomorphism (2) above. The p-adic representations on points of finite order were first introduced by Deuring and Weil, apparently more or less simultaneously in 1940, 1941. Deuring used them extensively in his paper “Die Typen der Multiplikatoren- Tinge elliptischer Funktionenkérper,” Abh. Math. Sem. Hamb. 1941, pp. 197-272, and his previous paper on the theory of correspondences in the same year. Weil mentions them in his paper, “Sur les fonctions algébriques a corps de constantes fini,” C. R. Acad. Sci. Paris 210 (1940), pp. 592-594, and develops them considerably in his book on abelian varieties. For applications to complex multiplication, cf. the book by Shimura and Taniyama ea ; inthe bibliography. Tate notice around 1957 that by taking the inverse NOt ie., the infinite vectors as defined at the beginning of this section, ons Nr 2et actually a module over Z, (or Q,), Which gives @ more natural Way describe the representation.Vi. Homomomhisns and Daygy 116 §3. Homomorphisms and B another complex torus, py up of complex analytic homomorph ‘additive group of compl phisms Hom, B) we mesh spe general properties of such homomomphism, of ino. Weshallstimplex torus a5 above. By a sublorus we mean y Le VD be 8 coteanples linear subspace OF VD’ = DA) Yr torus V0", where V8 D' isa lattice in V’ Let A= VID be a comple 10 Let A, B be complex toruses and let fave be a complex analytic homomorphism. Then Im f is a complex subtorus a i of B. and B = W/C, where V, W are finite dimensional Proof. Let A = V/D and B = W 1c. | 7 aie ver the complex numbers, and D, C are lattices. ‘Then f lifts toa C-linear map fivow, s0,f(V) is a C-linear subspace of W, and f(D) CC. Since the image f(V/D) is compact, it follows that f(D) is a lattice in f(V), thus proving that f(A) is a complex subtorus, Similarly, the kernel of fcontains a subtorus as a subgroup of finite index Also observe that iff: A —> B is any homomorphism (complex analytic, of course) and f #0 then Im f has dimension > 0, and for any integer N > Owe have Af #0. ‘Thus it follows that the natural map Hom(4, 8) > Hom(A, B) ® Q = Hom(A, B)g is injectives. By an isogeny f:A>B We mean a (complex analytic) homor Bh te gd ne finite yest) Romomorphim which is surjective an , 'geny, and Nis a positive integer such that every Se has period dividing W, then Ay (the group of points of ns Kerf, and consequently there exists a homomorphism BBA such that gf = BLS Nd. Note that Ay m (ZV) where n = dim A: 44, Complete Reducibility of Poincaré 17 Furthermore, if fis an isogeny, then it h; rig where g of =N-id. We denote us fs an inverse this inverse i Hen tes namely in Hom(B, A)g by fas §4. Complete Reducibility of Poincaré Call a complex torus A simple if it has no comy son 0. Then any some element P< Ena) mea Sie i oe ‘quently has an inverse in End(A)g. Therefore End(A)g Iris false that a subtorus usually has complementary subtorus. We now use the existence of Riemann forms to prove whatever is true in genes called Poincaré’s complete reducibility theorem . Let A’ = V'/0' be a subtorus of A = V/D. If (V, D) has a Riemann form E, then the restriction of E to (V’, D’) is obviously a Riemann form, which is non-degenerate if E is non-degenerate. This is clear from the positive definiteness of the associated hermitian form. Theorem 4.1. Let A’ = V‘/D' be asubtorus of A = V/D, and assume that (V, D) has a non-degenerate Riemann form. Then there exists a subtorus Av = V"/D" such that A=A'+A" and A’ OA" is finite Proof. Let V" be the orthogonal complement of V’ with respect to the positive definite hermitian form A associated with the Riemann form on (V, D). By the Gram-Schmidt orthogonalization process already used in Chapter VI, §3, we can see easily that the orthogonal complement D" of D’ in D (which is discrete) has complementary rank, namely 2n — rank of D', Hence it is a lattice in V". The sum D' + D" is then of finite index in D, and the theorem follows. Note that in the orthogonalization process, we can solve first with rational numbers, and then multiply by appropriate positive integers to clear denominators. This gives rise to the finite index. In Theorem 4.1, it is clear that the sum map Ax ATA i is isogenous is an isogeny. Theorem 4.1 implies that any abelian manifold is isoge toa productVil. Homomomphises and Dusty 118 Ax Ag x A Ae is simple. ness of simple modules in basic algebra, The unidaie for semisimple modules in by ies Ke _ sa argument OQ has a direct product decomposition into matics Coie gogenous {0 i uct Sing IFA genous 8 Pr BxBx XB sme abelian manifold repeated r times, then of the si End(A) © Q ~ Mat, (End(B) ® Q), ere the ight nnd side is the ring of - x matrices, whose components are w in End(B) ® Q. §5. The Dual Abelian Manifold ‘Again let V be an n-dimensional vector space over C, a lattice D, and suppose that (V, D) admits a non-degenerate Riemann form, Wee shall then call the complex analytic torus V/D an abelian manifold. When V/D is embedded in a projective space, as an algebraic subvariety, its image in projective space is then called an abelian variety. We shall continue, however, to deal withthe abelian manifold and theta functions. | follow Weil [7] inthis section. We denote by V* the complex antidual space of V. It consists of the inctionals, that is, of the maps BVOC which are R-linear and satisfy &(x) —i€(x) for all x © V. We denote the Value of € at an element x of V by x). a ba & *)J8 a bilinear map from V*_x V into C. Cina Ths th naar alve at is ED) is a functional, ie is functionals. Note that VF fs a Merely the complex conjugates of te Space in order to make a arias Space over C, (We take the antidt a pris) in map analytic later, instead of anti vite ©" He £5 uniquely determined by its imaginary part. Indeed, 45, The Dual Abelian Manifold 49 f= p+ in, where p= Re €and = Im & Then Ey ~i€(x) means that HG) = pli) and Gn) = pop Conversely, given an R-linear map ws V—> tional € = p + iM, by letting p be the functi above relation. We obtain an R-bilinear map R, We can define an antifunc- ‘on given in terms of ux by the (> Im &, x), which is non-degenerate by the above remark (that x determin es 8). Th for, from elementary algebra, we conclude that the set of element ae ev such that Im (u*, u) € Z for all w € D, is a lattice D* in v= Lemma 5.1. Ler C, be the group of complex numbers of absolute value 1. For each & V*, let X; be the element of Hom(D, C,) defined by erating) Xe(u) Then the map &+-> X¢is an isomorphism between V*/D* and Hom(D, C;). Proof. Our map > X; is clearly a homomorphism. Suppose that Xe) = 1 forall wu ED. This means that € D*, and conversely. Hence our map is injective on the factor group V*/D*. To see surjectivity, suppose given a homomorphism D— RYZ. Since D is free, we can lift this homomorphism to an additive ‘map 4: D—> R, and such a map can then be extended by R-linearity to an Relinear map V > R, since D spans V over R. As we remarked above, #2 is the imaginary part of a complex functional , thereby proving our lemma. Let E be a non-degenerate Riemann form on (V, D) and let H be its associated hermitian form, so that E is the imaginary part of H, and H(x, y) = Elix, y) + EG, y)- There exists a unique C-linear map ovo Ve Such that ' fVIL. Homomorphisis and Dug ality 120 Hx, 9) = (#el» » : 4 recalling that E(u, ¥) lies ‘aking imaginary parts, and is sa home forall» C aot intfaps D into D¥, that, induces a hoy ve : Lior Jrophism, also denoted bY Pes ge: VID > V*/D" ‘Then itis clear that 1 is non-degenerate morphism. In this case we can Suppose that E ismon-degenerate- and that the map ge: V—> V* IS transport E to V* by defining EX(gela), Ge(9)) = EO Ys or equivalently EXE, 1) = Elge! , ge") Note that ge maps D onto a sublattice of D*, and hence some integral multiple of ge maps a sublattice of D onto D*, Among other things, we have proved ‘Theorem 5.2. Suppose that E is a non-degenerate Riemann form on (V, D) and let E* be its transport to V*, There exists a positive integer m such that mE* is a non-degenerate Riemann form on (V*, D*). We now see that if V/D is an abelian manifold, so is V*/D*. Let A € End(V/D). We can define as usual its transpose on End(V*/D*) by the condition (AE, x) = E, Ax). From this and a non-degenerate Riemann form £ on (V, D) we can define an involution on End(V/D) by the condition N= AE = ge! Age be Jeers in immediate consequence of the next theorem that the association is indeed an involution, that is an anti-automorphism of period 2. Theorem 5,3, Ler E be a sociated hermitian form with respect to H, that is non-degenerate Riemann form on (V, D), with H. Let A = Af, Then A’ is the adjoint of Hx, y) = H(x, Xy), xyeV Therefore WAN’) > ig, #0. 46, Relations with Theta Functions 121 proof. ‘Trivial computation as follows, HONS 9) = HUGE" AGEL, ») = (Age, y) = (el, Ay) = Ha, ay). ‘The assertion about the trace comes from standard elemey nt Indeed, AA’ is self-adjoint, and so can be diagonalized in the linear algebra on V. Furthermore, AA’ is positive as an operator on V. so all cae elements are positive, so their sum (the trace) is > 0," liagonal Note that the trace in the theorem is the trac representation, Since the rational representation is a direct sum of the cone plex and its conjugate, it follows thatthe positivity statement also applies to the rational representation. Furthermore, since the rational representation is defined over Q, the trace is Q-valued on End(A). € with respect to the complex §6. Relations with Theta Functions ‘A normalized theta function is said to be algebraically equivalent to 0 if its associated hermitian form H is 0. By (4) of Chapter VI, §1, this means that for such a theta function F, the functional equation takes the shape F(x + u) = F(xbr(u), where YJ : D — C; isa character of D, uniquely determined by F, and which we shall call the associated character of F. In previous notation, 20K) Yew) = It depends only on the linear equivalence class of F. Conversely, two normalized theta functions which are algebraically equivalent to 0 and have the same associated character are linearly equivalent. The group of theta functions modulo equivalence, and modulo tis sub- stoup of those linearly equivalent to 0 was defined to be the Picard group (of divisor classes). The subgroup of those algebraically equivalent to = We denoted by Picg(V/D). We deal only with the latter, and so call it group for short. ‘ hia ‘The associated character F > Up above induces an inject phism ive homomor- Pico(V/D) > Hom(D, C1). We shall see in a moment that it is an isomorphism.Vil. Homenortisns aa Day ality 2 5.1, that if 0 is normalized VI, Lemma 5.1, de zed of ty We pve in Csr ih is mormalized and inthe same gg PE “a thea mee iE). Also recall the map ‘has type (+ “ gaia C1AI9, sta function in the class Of 8, /0 is algebraee and ote tha the normalize theta func isebraicaliy alent 100. inear equivalence class of 8, coal er associated with he Finer ea 19s there. fore equal to gee ints heta function with nonede | Let @ be an entire theta function wit legenerate incon ‘on Then we have a diagram which commutes with fac ie tor -1 vip —> veDe ee) cD x Pico(V/D) —> Hom(D, Cc) The right vertical map X is the isomorphism induced by E+» X, of Lemma 5.1. The bottom map associates with each element of Picg its associated character, and is an isomorphism. In particular, the homomor. phism gs has finite kernel, and Pi¢s(VD) is isomorphic to V#/D*. Proof. The theorem is obvious by putting together what we know. Start with an element x € V. Its image ge(x) is characterized by the condition H(c, w) = (oa), u). The comesponding characteris gi atin imag, teem er is given by exponentiating imaginary parts, that Nps wes e2aticay I We have seen that goi the associated going around the other way, haracter of the normalized theta function in the class of 6/8 is 5 This makes the (— ss Sine gis since =) commutativity clear. follows that » and the right vertical map is an isomorphism, it Nett we een Tyee is surjective, so an isomorphism as desited- Phism, Let Possible versions of the transpose of a homomor AS VD. vip 46, Relations with ‘Theta Functions 123, be whomomorphism. For our purposes here, et us denote Ae: VIDS vere the transpose, such that for z © Vand £ © V* we have (ang, = (€ dz). We use the transpose notation ‘A for the induced map + Pico(V/D) — Pies(V\/D,) which sends a divisor on V to its inverse image on V.. In terms of theta functions, if F is a theta function representing a divisor on VD then FA represents the image of this divisor under ‘A. ‘Theorem 6.2. The following diagram is commutative: vps —+ vps Pico(V/D) —> Pico(Vi/D,). The vertical maps are the natural isomorphisms, which make correspond ERO where m6 and Fe(z + u) = Fy(c)ue(w). Welw) Proof. From the equalities Feo AG + my) = FeAzi) reali) = Fe(dzy + Aur) = Feder) Ue (A) We conclude that Ure alts) = Ur (Am): Then og) = emia een) = Y(t) ivity is clear, Therefore F,.g has associated character arg, and the commutativityVIL. Homomoephisms a Duy 124 sire non-degenerate tet faNctION, BIVINg rise yyy an ent Now let @ be normophis™ ue sees VD —> Pico VP) Ye may define an involution End(V/D)o. We may ¢ Let A € End(V/Dho Ape git o'Ae Go here ‘his the transpose on the Picard grou, where A is 1 the Riemann form associated with 8. Then the typ Theorem 6.3, Le! involutions Aj and Aq are equell Proof. This is an immediate consequence ofthe preceding (Wo theorems, and of the diagram: vps % \e | | v/D ve VD P Picy(V/D) => Piew(V/D) and the two end triangles have sign — The central square is commutat which cancels. Let X be a divisor on V/D. We denote by “€(X) the set of all divisors ¥ on V/D for which there exist two positive integers m, m’ such that mX is algebraically equivalent to Y. We call €(X) a polarization of V/D if €() Contains the positive divisor of an entire non-degenerate theta function; or ‘equivalently, if some positive multiple of a divisor is a hyperplane section in some projective embedding. Suppose that X is the divisor of the entire theta function 6, and is non-degenerate. Then it is clear that the involution depends only on the class €(X), and therefore A; is also denoted by Ag, where i the polarization @ = €(X). ‘Theorem 6.3 gives an analytic description « involution defined purely algebraically by the formula Am dM. §7. The Kummer Pairing We shall now di cy : aa “pe scribe the p-adic version of the duality between V/D and ils 4. The Kummer Pairing 1s Let € € V* be such that NE D* for application later, we shall consider the ea he moment, itis notationally convenient y {eeall the Wo isomorphisms Some positive inte s ger N. (In the ios when WV isa power of p, but for ‘Work with arbitrary integers.) We Picu(V/D} Hom (D, €,) = veipe ‘therefore, with & we can associate a normal tire) fe, Well defined up to multipl ally equivalent to 0, and satisfying lized theta function (of course course not ion by an abelian function, algebraic Le +H) = fa) = flajemines, forx € Vand « € D. Given a finite number of values of u € D, we ean always find x sufficiently general in V such that the terms in the above equation are defined. ‘The condition NE © D* implies that fY is an abelian function. Ifa E Vand Na € D, it follows that there exists an N-th root of unity ey(é, a) such that Fe(Nx + Na) = ey(&, a) f(a). Itis clear that this root of unity does not depend on the choice of fin a linear equivalent class, nor on the choice of & (mod D*) and a (mod D), Therefore the association (&, a) > ev(E, a) gives a well-defined map (V*/D®y x (VIDy > Bay i it itis tivi fied that Where pry is the group of N-th roots of unity, and it is trivially veri this map is Z-blinear, ie. is paring, It is equally easily verted tha the kernels on both sides are 0, i.e. that the map puts (V*/D*)v and (V/D)y in perfect duality ‘There is a special case of £ which is of interest. Suppose that 6 is a aa f nondegenerate entire theta function, with associated Riemann form E. be v. then there is a unique normalized theta function fin ele ‘sociated character is ~ Ej. We suppose that No = 0 (mod), YD. Then (tod D), in other words, b and a represent points of order in 1/0 Ute fhe Foot of unity ey corresponding to this special chee can formVIL, Homomorphisns and Dat 126 sbvious from the definitions, and Theorem 6.1. asis ¢ version. Instead of N we use p* to the p-adic version P™ in the on. Then | (V/D)y really means (V/D),», and similarly for ‘Wen we pass from p10 p*"', it is easily seen that the to the inverse limit, that is, that if Now let us pa ing discussi (vepD*) and Hy ; pairing is consistent with respe peteeD? and pa GD, then even a = ex PE Pad Therefore we can express our pairing in terms of the Tate group, and we obiain a Z,-bilinear map 7, (V*/D*) x T,(V/D) > T,(w), which make the two Z,-modules on the left Z,,-dual to each other. Note that T, (a) is formed of vectors Gis bo Sin) where {is a p".th root of unity such that {&., = Gy. ‘Thus T,(y) is a I-dimensional space over Z,, which cannot be identified with Z, unless we have selected a basis, ,__ Remark. The advantage of the root of unity ey (é, a) and the p-adic pairing isthat they hold algebraically. When we embed fC abelian torus in projective nae - rar comles Points A¢ of an abelien variety A, which may the zeros of ao Ove" afield k. “As we shall see in the next chapter, see tat he defen non i V correspond to a divisor on A. One can then involving only the deg ne: 2 cortesPOnds to a purely algebraic definition of unity then le iether C88 Cotesponding to £ on A. The resulting roots extensions of k (this e complex numbers, containing k, but generate possibly over Q, and N si sficen ; ways the case if we take & to be finitely generated the definite advantage tog ne, THER these roots of unity, and Tp), have Galois group of the aay can be used as representation spaces for the Which takes essential ase C084 Of k, thus giving an additional tool Sunt of more arithmetic aspects of the situation. It 5. Periods and Homology 127 jsthen absurd in such a context to identi dperates trivially on Zp the algebraic theory and applications . pooks of Lang, Mumford, Shimura, and Weil the reader is referred to the To(W) with 2, since a Galois group s8. Periods and Homology LetA = V/D be a complex torus, and complex coordinates 21, . 2. , 2, We first observe that the holomorphic |-forms complex numbers by Pick an isomorphism of V with C*, with on A are spanned over the dz, de, Proof. The coordinates 21, .. . , z, define a complex analytic ch; every point of A. Any holomorphi¢ I-form has an expression Si@)dzy + +++ + fule)dey at every point, where the functions fi, . .. f, are holomorphic, and the functions which form the coefficient of dz, at every point are then analytic continuations of each other on the torus. Hence they are holomorphic on the torus, i.e. periodic, so constant. Hence dz, ... , dz, form a basis of the L-forms over C, as was to be shown. Denote the vector space of holomorphic I-forms on A by 11,(4) = 0. We have a pairing HA, Z) x > C given by mare This induces a homormophism H(A, Z) > 9 jis map is injective, and that its im of (,/Hj(A, Z) with A itself. al vector space, and that it has into the dual space. We shall prove that th image is a lattice, giving rise to an isomorphis! First we note that , can be viewed as a re dimension 2n over R. Indeed, the forms ay, 6 5 vy Gay ide «<> + idenVI. Hon rmphioms and Duaiy 128 BR, as one sees immediately since if We wriyg are linearly independent O¥67 a t Me then dens diye oo Een Mn rndent real coordinates: weal eto every cycle in H(A, 2), then the map cular 10 rl action on A, which is therefore constant, and there. ‘This implies that in the above pairing, the is a free abelian group on 2n generatoy ) is 0. Since H,(A, Z) is a free al ' a tris mage in has real rank = 21, it follows that this real rank be equal 2a, and therefore that (A, Z) may be viewed as a latice in O,, ‘There is also a natural isomorphism are indepet If a is perpen defines a holomorphic fun fore equal 10 0. Hence w = 0. D—> AVA, 2), obtained as follows. ‘The group D may be viewed as the fundamental group of the covering piVoV/D=A Ifw € D and |, isa path from 0 to w in V, then p « J, is a cycle on A. ‘The association > class of p «ly in HVA, Z) ives rise to the above isomorphism ‘Thus we obtain a natural embedding Po: D—> HA, Z) > A, whose image in (, will be den p i 1 wted by DB. nthe other hand, let: © V. Let 4, be any path from 0 to z. Then po isa path in A, whi f ° to the functions! "N° MY Wiew as an element of fy, that is, giving rise 43, Periods and Homology 129 pei Voy, ‘theorem 8.1. The map py is a C-linear isomo "phism, inducing ani phism Phism, inducing an isomor- pr VID > O/B = Ona, D. Proof. For each j = 1,.... n we have “de a. “Thus it is clear that the map is C-linear, and we have alread " ly seen above the induces an injective map on D. This proves the theorem, since D generates Vover R, and its image generates © over R (being a latice. Now let {a, + Wa} be a basis of Oy over C. Let A be the abelian group of periods, that is the group of elements Ju... 00, forall y € HA, 2, ‘Theorem 8.2. The map pol CET) establishes a complex analytic isomorphism of A = V/D with C*/A, send- ing Don A Proof. The map is well defined, from V into C'/A. The above duality immediately implies that the kernel is D. The map is locally surjective in a neighborhood of the origin because d: , dz, are local analytic coordi- nates, and it is surjective by additivity, and the fact that given z € C", the clement 2/N is close to the origin for N a large positive integer. Since the map is obviously analytic, the theorem follows. Theorem 8.3. Let f:V/D> VD bea complex analytic map such that f(0) = 0. Then fis a homomorphism. the induced linear map on back, We further get the anf w= frre. Thus we have extended the map p toa map Proof. Let Q,(f): Q(V'/D') > 4(V/D) be holomorphic differential forms, obtained by pull dual map (linear)Vt Homomorphihs nd Day fy: fur) > (V2. fed by the change of variables formu that We haye g ly verifies and itis ommutatie diagram: an (yp) —> "72" a» IF =f m — vi v f Hence fis linear, as was to be shown CHAPTER VIII Riemann Matrices and Classical Theta Functions jemann Matrices sl. Let H be a positive definite hermitian form on C*. We may write H in terms of its real and imaginary parts as H(u, v) = E(iu, v) + iE(u, 0) where E is alternating and real valued. By convention, H is linear in the first variable, and anti-linear in the second. Let A be a lattice in C*. We call H (or E) a Riemann form for the pair (C*, A) if EW, )EZ for uv€Aa. Note that the form (1, v) > E(iu, v) is symmetric positive definite, Let W = 0m... Wa) bea basis for A over R. We view w, (j= 1, . « « » 2n) as column vectors in C*, so that W is ann x 2n matrix, Elements of C* may then be written in the form 4 Wx with xER™, e , 2. There and elements of the lattice may be written in this form with x € 2°. : ‘is @ unique alternating matrix P such that 3i, Rismann Matiees and Clsicl Theta Fapgion is the matrix re Pony repre. Of Rtn fa ans that “Alternating” ting form with senting the alternat if P = (py) then El respect to the standard 22h, matrix such that Let C be the 2 x 2a re iw = WC. Then E(iWx, Wy) = EWWCx, Wy) = W'CPy. It follows that ‘CP is symmetric positive, and we also have 'CP = -PC. The following two lemmas express in terms of the relevant matrices two conditions characterizing a Riemann form, namely the facts that the form E(iWx, Wy) is symmetric and positive definite. Lemma 1.1. The symmetry of ‘CP is equivalent with the condition We-""W = 0. Proof. This symmetry is equi lent with the condition Ww om _ -() Pw, W) = (7) Pccw, WM, Using once more the definiti " ion iW = WC ing vatri: - cations, yells the dec and performing the matrix multipl conclusion. Lemma 1.2. The matrix associated with H is M= 21H Wy 5 0, And $0 for u, © € C* we haye Hw, 0) = uM Proof, Let M be the matrix as indicated above It ig clear that M is Matrices Riemann ® 133, permitian, Thus it will suffice t9 show that Hu, w uM i, Let i, = ily be / times the unit n xm matsix. Then ( 0\(w w 0 =) W, -(§) & since H(i, «) = E(u, «), we put u = Wr to obtain: wy i, 0\(W 0 “g) = “ew, Whew, Wy-p(W) "(in 0 \ (wi = = CW, how, v(t) ( 7 ra) (6 ~i,) \W) * ‘The terms on the far left and far right are (u, 'Z) and (u, 7) respectively We then carry out the matrix multiplication for the middle product which ig equal to W) ow aml [(i) ew 7m] We use Lemma 1.1, and we use the fact that 0 xX\'_/o i) yo} ~\x' 0 with the appropriate matrices X, Y to find that the above expression for H(u, w) is equal to H(u, «) =~ XPCx = = 'xP itu P Wy — UWP W tu = itu PW)", This concludes the proof. Now let W = (a, ws) where «, «2 are square matrices, let ob a Co 0) and suppose that P = J. Performing the matrix multiplication, we see that relations of Lemmas 1.1 and 1.2 can be rewritten as: RR 1. ww, — « ‘ws = 0, o equivalently WI'W = 0.—— ‘ut, Riemann Matrices and Classical Theta Fun 134 * mas)! > 0, or equivatentty AWI'W > 0. RR 2. 12's ~ 1 " ions. Taking an inverse and a ¢o, jemann relations. Taking ie rose ale eo ne event ih conjugate RR 2’. flan — 'H) > 0. 2) of complex matrices satisfy set of all pairs (0, «2) OF # X 1 ct sine Let be the se oF a PL as usual, We let SP(R2) be the subgroup gf Jann relation c the Rey consisting ofall matrices Mf such that MM 2» is closed under transpose. ; Tae the Siegel upper half space, consisting of all matrices r € Mat,(C) (n x complex matrices) which are symmetric and whose imaginary partis symmetric positive definite, Lemma 1.3. Let (a, @:) € B (Ig E GLC) then g(w, w2) ER. (i) FM E Spz,(R) then (w,, w2)M ER (iil) The matrices a, «2 are invertible, (iv) We have ws'o, € Gy. Proof. The first two assertions are immediate from the definitions of 2, Spin, and the Riemann relations. As for (iii), suppose there exists a vector © EC" such that ‘vw, = 0, Then ‘v(w'@: — 16 whence 0 = 0 by RR 2. Hence w;' exists Furthermore J € Sp,,(R), 80 (-ey @) = (wy, 2) ER b 7 y (i } and therefore w;" exists, thus proving (ii). Finally (wz',, 1,) ER by (i) and therefe concludes the i aan WW) follows from the Riemann relations RR 2. This 1m light of Lemma 1.3, Relations reads if = zand w; = |,, then the second Riemann sp. Te Siegel Upper Hall Space asst ‘Thus let diy + dy be positive integers su and put dy | dy ne BO \ = aiagid, ° Given the Riemann form E integral valued onthe lat i fife s+ 0 We} of A such that wei Reecesits si Bow}, moes= (9, ‘). ‘Then EW'x, Wy) ='xJsy for x,y ER™ Let 10 v=() 3) and W=W'U-! so W! = WU. ‘Then matrix multiplication shows that E(Wx, Wy) = ‘xJy and A = WZ = WUZ = (wy, 052". This kind of changes of coordinates reduces the situation to the preceding one, when all the elementary divisors are equal to 1. §2. The Siegel Upper Half Space A polarization of a complex torus (for this form. Thus a polarized complex torus is a triple (V Complex dimensional space V, a lattice A, and a Riemann section) is a choice of Riemann ple (V, A, E) consisting of form E, We sayi Risa MCE a Case They py 136 «ation ifthe elementary divisors d, ‘“ arization ifthe eleme! FB is gS canis is pC rn axis ae all equal (© We shall now yin! that 10.2 Froben on, and obtain a classification of such gy with respect (0.8 ois rite for simpli OF folds UP mamratzed (SieBe) UP isms. If we have tattice A, s0 the form E is repres write triple ng them as a quotieny ion of by parame modulo the’ of = EOUP Of aun 1 per lf pe ism of V with C*, and a (0 he fs can be writen aS al 1X 2M Matrix W, ang P with espeet to this basis, then we gp at its elements cal ented by the matrix =(C%, W. P) Ww! fe matrix P representing E is of the form (8,3 ‘This corresponds to having chosen the elementary divisors dy, .. ., Suppose that th where 1y isthe identity am maliX. Frobenius basis forthe lattice, and to all being equal to 1. ‘To each matrix z in 9, we may associate the corresponding n x 2r matrix (1) which we may view as a matrix W of the components of a basis for the latice as atthe beginning of the preceding section Lemma 2.1. Every isomorphism class of principally polarized abelian ‘manifold comains a representative (C", (2,1), J) with z ED,, for which the columns of (2,1,) form a Frobenius ba. ibrar Let W = (a, ws) be a matrix satisfying the Riemann relations, nd Was lumns form a basis forthe period latice of the abelian manifold If @ = \,, then the Ri re re 1e property le iemann rel equivalent wi y iene, lations are equivalent with the property Since licati ‘multiplication by «wz! on Cm is a linear isomorphism, we see thit Lemma 2.1 follows immediately from Lemma 1 3. To get uniqueness ‘uniqueness we hi 5 ‘Monon Ae betes, HHO" Out by an appropriate grou of B= set of at Pais (cx, 2) satisfying the Riemann relations; oo. The Siege! Ure Hal Space 137 subgroup of GLy, sps(R) +(R) consisting ofall matics y such i yi ys J. ren Sp2x(R) is closed under transpose. Write We have wired Pandonty fad be= I, and a'b, ed are symmetric When 1 = 1, this means that y lies in SLy We shall see that the association 2 (az + bez +d)" = ye) defines an operation of Sp2,(R) on ,. We have by Lemma 1.3 (i) ab «, w(’ ) (cb + d)-\Ga + ) ES, (ca +c, 2+ dER Hence by (iv). z so taking transposes yields (laz + fo\('bz +d) E Ge Since Sp;,(R) is closed under transposes, it follows that y(z) € Sus At is then immediately verified that this defines an operation of Spz-(R) on Oy. Suppose that (w{, «$) is another basis for the lattice A. Then there is a matrix M © GL,,(Z) such that (wi, 03) = (wr, @:)'M, and conversely. The form E with matrix J with respect 0 the basis (@, 3) has the matrix MIM basis has the same With respect to the basis (w}, «3). Therefore the new tnateix for the Riemann form if and only if Af € Spa We letvat, Riemaan Matrices and Classical tha, ine on, 138 _ , qut(s) = Gin) 51 RD = Span(Z) re ices of a basis forthe latice, ang mate + and a es we have made. We define fina affect the choi ian manifolds. Any to b th fe polarized abi Is Any tose valence cla88 i ty.v) and (C% e lattice preserving the matrix J , basis of the lattice the asa dif by 2 omphism on C* (multiplication by an ig Pe, ftw Hee rnmng the second Pall ofthe latice veering ge vai eto. ‘Theorem 2.2. The map s+ isomorphism class of (C", (2, ln), J) induces a bijection I\$.> §3. Fundamental Theta Functions This seetion follows Shimura [Sh]. Let A bea lattice in C*, We recall that a theta function on C" with respect to A is an entire function f satisfying the condition Su +1) = fwe(Ae, Dw) 2m foru EC’! € A, and e(2) = e**, Here wis an arbitrary function, and A 'sCinear in, R-linear in J. A theta function is called normalized if there is a hermitian positive form H such that 1 fut = sone( 1 (u + ; é ‘))¥o and if yh Athearn He value equal to 1. We let E be the imaginary part of, i and atemating. We assume that His C-linear ints Variable, anti.) . tos tae °8inits second variable. It follows from the above su 49. Fundamental Theta Funetions 139 Wd +P) = weOucrye(! wow re( EU, ) We denote by THU, Us A) the vector sp : Uhsying the above conditions witht, yuna oy funetions. This notation is better adapted tothe, onl tpefor eee mnind than the previous notation in Chapter Vj. Pons We now have in Lot z € and for r,s Re define 1 : -y ell: Olu, 2.7 8) ZG (M20 + m) $F & miu 4 ») ‘The positive definiteness of the imaginary part of . zi tial term goes to zero like e~ sures that the exponen for some c > 0, whence we ag . a have absol convergence, uniform for u in any compact set of C*. We shall au lute three simple transformation formulas, mw stats Th 1. For a, b © R" we have blu 2.7 +4, b+ 8) 1 = (5 ‘aza + ‘alu + $4 0 + 2a +b,2,7,5), ‘his follows directly from the above definition, expanding out the exponent and collecting terms. Th 2. For a, b € Z" we have Ou, 2, r+ a, s+ b) = e('rb) Hu, 2, 7,5). Again this follows directly by observing that the sum expressing the left hand side can first be changed by replacing m + a by min the sum over m, and then the simple term e('rb) comes out as a factor. Combining these two formulas yields: Th 3. If'a, b © Z" then Ou + 2a + by 2,18) = e(- ja ae trees ‘0 213. - e(- Mage au) Baw, 23) where X_,,(a, b) = e(—'sa + rb) is a character. “=. run. Risiann Matices and Clase They py tions 1, 8) (as function of 4) is noy i non theta function so Normal. *8 10 Obtain, lw sarin n THEN (OU ) is 4 wivial theta Function, ayg 2) is quadrats so Qu owe) - QHD = (« + se -7, 2p isnot holomorphic in =. Define asin Shimura lt, 2.13) = (QU 2) OU 2 F.8D Note that Q(u » the hermitian form of the last section, defined by We shall find again Hu, 0) = 2i'ue ~ 3) Nove that puting (= za + bwith a, b E Z" we find 1 ilus tat) = (ua g)at (ut gluta: = 2 ‘Th4, For a, b EZ" we have outa bens) f(t L 1, semehitd (Laleet ))a(Lad)tastedh This suggests that we define ea +B) = ej ‘) and Vapa(ea +) =e ‘Thus = UX ar Where X,.., is the character defined by (ea + b) = e('rb — ‘sa, Let [z,1,) be the lattice of, fons, we hon os 20 +B with a, bE Z*, Then with be that Th 4 can be stated as follows. gp: Fandarental iets Fanetions ry The functions OU 3) He in TW BX, esa et eu character a Now let X be any cl ter on C", viewed as R2" jn the form a8 RE if we express elements a+b with ab eRe, We observe that the map 0+ ox gives an isomorphism THH, y, A) => Th, gx, A), On the other hand, define ce of entire functions satisfying 1, = e(- 5 (az ~ ‘a. for a,b EZ". ‘Then certainly the functions @(u, =, r, s)X,...»(u) lie in this space. Thus we have isomorphisms Th(z) = The. Yes [25 Ia)) = THs, Yorsy L2s Ua. Each one is obtained by multiplication with an appropriate function. Theorem 3.1. Let L = [z, 8], where 5 = diag(d, . . . , d,) with O1, and hence must have degree 2. Furthermore, Py divides the characteristic polynomial of the absolutely irreducible 2-dimen- sional representation of E®Q, so P. is equal to this characteristic polyno ‘ial. In particular, the norm and trace are those of the minimal polynomial. We let the factorization bedns and Abs MAGUS of Quang 1x, bol op X? = (aX + ne(a), pa == = 2 ™ and zed nr(a) = aa’, a= fa € & then we let a / is an involution of Q, that is yrem 1.1. The map a> & isan 1 Q. thar j om vomorpism of ode” 2 au ar. It suffices to prove the is obviously linear Propeny Proof. The map # (apy = Bra’ 7 ce, norm are the ordina matrices, and the trace, norm are nary trace ang when a Bae Scalzi, it suffices to prove the relation whea ne tei have algebraically invertible. But then independent coefficients, and in particular ag a! = a"!det(a), and the relation is obvious. This proves the theorem ‘We note that if F is a subfield of Q over k, then F is a quadratic extension of k and the involution a a’ induces the non-trivial automorphism of F if Fis separable over k. Indeed, the formula a’ = a~'nr(@) shows that a’ also lis inthis subfield, and a, a are the roots of the minimal polynomial of a over k. ‘The involution «+> a will be called the canonical involution. Proposition 1.2. Every inner automorphism of Q commutes with the involution. Proof. Immediate from the fact that aa’ is an element of k, and s0 commutes wit all elements of Q. Theorem 1.3. Let F bea separable quadratic extension of k contained it Let @:F~ O be aklinear embedding which is nor the identity. Thet there exists an inner automorphism of Q which induces on F. In par Muar, every automorphism of Q is inner. Prog , We shall need the remark that there is an isomorphism 280 —> End,(Q,,) gu ivotions 145, Qe, denotes Q viewed as a vector space whe S Ane 4 over k, The i . sfollows. An element ¥ @,@6, gi isomorphism is et as 3B. Bives rise to an endomorphion ect xD aixpi. is association is a homomorphism of Q@Q i sis association isa phism of QO ino End , Hemically Ze, and hence is injective singe Q@Yie spe yet ideals other than O and the whole algebra). I ig an isomerate ose jimensions of the domain and range are equal. eonete slice te Now we may view Q as a vector space over F in two ways i patural way, using the multiplication in Q. Second, by the action oe @9> e@x foraerxreg ‘The dimension of Q over F, either way, is equal to 2. in So there is an iso- morphism A:Q > Q such that A(ax) = g(a)A(x), and A in particular can be viewed as an element of End, (Q,,). Thus by the first remark, we have A=> a BB, and in particular, Aa) D aaB; fora € F. ‘The relation A(ax) = g(a)A(x) shows that DY aarp} = gla) asp} forallx €Q, ‘0 DY aap =D eas. Me i. Let a; for some i, lence aja = g(a)a; for all a € F, and all i eons @ #0. We claim that @ is invertible in Q. This Since a & F, we have