Lecture Notes in Mathematics Edited by A. Dold and B. Eckmann 437 David Masser Elliptic Functions and Transcendence Springer-Verlag Berlin - Heidelberg - New York 1975Dr. D. W. Masser Dept. of Mathematics University of Nottingham University Park Nottingham NG7 2RD/England Library of Congress Cataloging in Publication Data Messer, David ¥A2Man, 1928- iptie functions’ ani transcend (Lecture notes tn nathenaties 5 £37) Et 2, fasbere, tr Leotiire notes dn 510" Be B15! 3537 Ten BzIE5 AMS Subject Classifications (1970): 10D25, 10F35, 38A25 ISBN 3-540-07136-9 Springer-Verlag Berlin - Heidelberg - New York ISBN 0-387-07136-9 Springer-Verlag New York « Heidelberg « Berlin This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically those of translation, reprinting, re-use of illustrations, broadcasting, reproduction by photo- copying machine or similar means, and storage in data banks. Under § 54 of the German Copyright Law where copies are made for other than private use, a fee is payable to the publisher, the amount of the fee to be determined by agreement with the publisher. © by Springer-Verlag Berlin - Heidelberg 1978. Printed in Germany. Offsetdruck: Julius Beltz, Hemsbach/Bergstr.Introduction Chapter I. A transcendence measure... --- 2 ee ee ee ee ee 7 Gets introduction) ge 1.2. Preliminary lemmas. . 6... eee et 33 Procol of @hectem fe Chapter II. vanishing of linear forms without complex multiplication 16 2 le Introduct§on) dl 2.2. Preliminary lemmas. 2... ee ee ee ee ee ee 16 2.3. The main lemma. 2. eee ee ee ee 2.4. The auxiliary function. «6.1 ee ee ee 26 @ 5: Procol of Theorem (0 hl Chapter III. vanishing of linear forms with complex multiplication . 36 Ole introduction ee C8 3.2. The upper bound... ee ee ee ee 26 ies proof cf Thcor-m@ Itrss 6 Chapter IV. An effective proof of a theorem of Coates. ....... 44 Chapter V. A lower bound for non-vanishing linear forms. ...... 49 5.1. Introduction. «26. 6 ee eet et ee 5.2, The auxiliary function. . 2... ee ee ee 5 5.3. Proof of Theorem IV... 1 eee ee ee ee 6 Chapter VI. Lemmas on elliptic functions with complex multiplication 63 6.1. Introduction. 6.6 ee ee 6.2. Multiplication formulae... ee ee ee 3 6.3. Estimates for algebraic points... -.--- ee ee ee OB Chapter VII. Linear forms in algebraic point: Gal, introduction, gee cl 2) cour lemmas es ss 7.3. A simplification. ©... ee ee 86 Ge4. the auxiliary function eee oe 7.5. The Wronskian 6. 6 eee ee ee e108 (63 (thescace 0-92), postscript aIv Appendix I A non-analytic modular function. . - - - eee ee ee ee NB Appendix IZ Zeros of polynomials in several variables... . +++ e+e e+ + +123 Appendix IIT A transcendence theorem for algebraic points... .... +++ + + +132 Appendix IV Rational points on curves of genus one with complex multiplication. .137 Rete eeINTRODUCTION These Notes are concerned with some new transcendence properties, or more exactly linear independence properties, of certain numbers associated with elliptic functions. The purpose of this general introduction is to describe the results and provide a historical context for them. Essentially we treat special periodic functions f(z) of a single complex variable z. The methods of transcen- dence theory are most likely to succeed when these functions have an algebraic addition theorem, Such functions are rather strictly delineated by an old theorem of Weierstrass, which says that up to algebraic dependence f(z) must be one of the following. First, £(z) can be the exponential function e“*, and we take a as an algebraic number so that the differential equation of f(z) is defined over the field Aof algebraic numbers, Then it is a classical result of Lindemann that the fundamental period 2ni/a of f(z) is a transcendental number. The second possibility is that f(z) is a Weierstrass elliptic function P(z) satisfying the differential equation (Pz)? = 4(P(2))* = G22) - oa for complex numbers gz, gs with g2? 4 27937. Once again we take g2, g; to be algebraic numbers (and this assump- tion will be maintained throughout these Notes). This is a doubly periodic function, and so we may choose a fundamental pair of periods wi, w2 with the imaginary partvI of w2/w: positive. We also consider the associated Weier- strass zeta function ¢(z) defined by ¢'(z) = -P(z) and normalized additively to be an odd function of z. This possesses an adequate addition theorem, and there exist quasi-periods n, and n, corresponding to w; and w, such that c(2 tw) = el2) tng = 1,2). Siegel was the first to investigate the arithmetical properties of w; and wz; he proved in [26] that they cannot both be algebraic numbers. A few years later Schneider in his fundamental researches improved on this result by obtaining the transcendence not only of w; and w2 but also of n, and nz. An account of this can be found in his book [25]. The exponential case above is naturally included by considering the five numbers wi, w2, 1, 2 and 2ni together; thus we see that each is transcendental. ‘The most general theorem extending this statement would assert the algebraic independence of these five numbers, but there are two reasons why this cannot yet be proved. First, the general techniques are not available, and secondly, the Legendre relation wana ~ wing = 2nd shows that such a theorem would be false. So we turn to the more fruitful question of linear independence are the above five numbers together with 1 linearly independent over the field A? The answer to this involves a splitting of cases. We consider the set of complex numbers } # 0 such that F(z) and F(z) are algebraically dependent functions. By the addition theorem, this set together with 1 = 0 forms a field F. 1f F#Q the field of rational numbers we say that P(z) has complexVIL multiplication; this is the exception rather than the rule and then F must be a complex quadratic extension K of Q. Otherwise F= Q means that #(z) is without complex multi- plication and in this case an affirmative answer to the above question had been conjectured. Partial answers were first obtained by Schneider in {25}, who proved that 2ni/wi, ni/wi, and awi+fn; are trans~ cendental for non-zero algebraic numbers a and 8. In addition he showed that w/w; is transcendental under the necessary condition that P(z) has no complex multiplication; for otherwise w;/w; lies in K. These results were consid- erably generalized by Baker with the aid of his many- variable techniques. In two papers [2], [3] he proved the transcendence of any non-zero linear combination of w1, w2, 1a and ng with algebraic coefficients, and shortly afterwards Coates [12] obtained the extended result for the five numbers w:, 02, M1, N2 and 2ni, In Chapter IZ of these Notes we prove that these five numbers are them- selves linearly independent when /°(z) has no complex multiplication; in conjunction with the result of Coates this establishes the linear independence of 1, wir w2y Nir na and 2ni over A. In other words, the vector space V spanned over A by these six numbers is of maximal dimension six. The proof attempts to imitate the proof of the theorem of Coates mentioned above. The main difficulty arises from the periodicity of the auxiliary function ¢; this means that the zeros of ¢ on the real diagonal do not carry as much information as they normally would. Therefore we are compelled to extrapolate further to deduce that ¢ is small on a large part of the whole complex diagonal. AfterVIII using a device to replace ¢ by a simpler function ¥, we may regard ¥ as a polynomial, and as such it inherits the smallness of ¢ on a large subset 4 of €°, To show that 4 is well-distributed in a rather weak sense is a matter of diophantine approximation (this appears in a slightly dis- guised form in the proof); we then require arguments essentially in one complex variable to see that this im- possibly restricts ¥. The results of Schneider mentioned above on the transcendence of 2ni/wi, ni/w: and w2/w; can be regarded as first steps towards the theorem of Chapter II, the next (and most recent) step was taken by Coates who showed in [13] that w,, w, and 2ni are linearly independent over Awhen P(z) has no complex multiplication, ‘The new idea in the proof of this result involves an appeal to the deep and extensive theory of Serre on division points of Pz). However, for historical reasons the theory of transcendence has always laid emphasis on the concept of effectiveness; it requires, for example, that all constants occurring in the proofs should in principle be capable of explicit evaluation. Serre's theory was not constructed for transcendence purposes and sometimes fails to satisfy such criteria, Therefore it is of some interest to have a more elementary proof of Coates’ result which does have an effective structure. We provide such a proof in Chapter Iv. So far we have not discussed what happens when f(z) has complex multiplication. Since w2/u, is an algebraic number it is clearly untrue that the six numbers 1, 01, wa, Mis N2 and 2ni are linearly independent over A. In Chapter III we use a very simple argument to prove theIx slightly unexpected fact that there is another essentially distinct linear relation between the six numbers. This leaves four candidates for a basis of V, namely, 1, wir mi and 2ni, and in the same chapter we show that these four numbers are indeed linearly independent over A. These results imply that the dimension of Vv is four in the case of complex multiplication. After proving theorems of this type we consider a secondary objective of transcendence theory. ‘This is to derive quantitative refinements in the form of measures; these tend to be more important for applications to other areas of number theory. In the situation outlined so far these refinements are of the following kind. If the co- efficients ao, aiy a2, 81, B2 and y denote algebraic numbers not all zero, our results show that the expression A= ao + ayy + Gem, + Bim, + Beng + y-2nd does not vanish (under additional conditions depending on complex multiplication). ‘The problem is then to find a positive lower bound for the absolute value of A in terms of the degrees and heights of the coefficients. We shall give two examples of such estimates. In Chapter I we treat the case ap = 8, = 82 = y = 0 when there is no complex multiplication; this is equivalent to a transcendence measure for w2/w;. Feldman (16) has obtained a measure for a more general class of numbers but the lower bound turns out to be too weak for the application we have to make in Chapter II, Although our proof is just a modification of Feldman's argument*, it is important to # Since I wrote these Notes, Feldman has published a paper in which he carries out this modification himself. This appears in Acta Arithmetica, 24, 477-489 and the re- sult it contains is slightly sharper than mine.give the details: first because of the crucial improve- ment in the lower bound, and secondly because all Feldman explicitly obtains in his paper is a measure of irration- ality, which would be trivial for our particular number w;/w; (since it is not real). In our other example we consider the general form A when a) #0. Previously Baker [4] had given an estimate in the case 6, = 8; = y = 0 of the form JA] > ¢ exp(-(1og HX), where H is the maximum of the heights of a», a: and az, © > © depends only on their degrees and the numbers w; and 2, and x is a large absolute constant. In Chapter V we show that for the more general form x can be taken as any number greater than 1 provided only a» # 0; more precisely JA] > © exp(- 10g H(1og log #)7**), where H is now the maximum of the heights of a9, a1, dar 81, Ba and y, and C > 0 depends only on their degrees, the numbers w; and w2, and the arbitrarily chosen c > 0. Here the dependence on H is quite near best possible, for stan- dard arguments show that the absolute value of A can frequently be smaller than H~ for some positive absolute constant x, This completes the outline of our study of the periods associated with P(z). The remaining chapters investigate the arithmetical nature of a more general class of numbers, the algebraic points of F(z). TheseXI are defined as complex numbers u such that either u is a pole of P(z) or P(u) is an algebraic number, The analogous definition for the exponential function e” gives just the logarithms of algebraic numbers, and the theorem of Baker states that such numbers are linearly independent over Aif they are linearly independent over Q@. We cannot enunciate the corresponding result for elliptic functions until we recall the definition above of the field Fof complex multiplication; this is either @ or a complex quadratic extension K of @. Then it had been conjectured that algebraic points of P(z) are linearly independent over Aif they are linearly independent over F. Chapters VI and VII of these Notes are devoted to a proof of this conjecture when F #@. In (25] Schneider obtained the first relevant result by proving the conjec- ture for two algebraic points and unrestricted F. But for three or more algebraic points it seems necessary to take F¥ Qin order that the techniques of Baker can be used, The advantage of complex multiplication is that the zeros of the auxiliary function can be taken on the diagonal corresponding to a part of the lattice of inte- gers of K. The extrapolation procedure operates on this wider range of zeros without the necessity of introducing division points, which would have involved algebraic number fields of intractably high degree. To obtain a final contradiction we use stronger versions of some of the ideas of Chapter II, but the proof is further compli- cated by the unruly behaviour of elliptic functions under differentiation. In fact we obtain a proof by induction on the number of algebraic points - not of the original conjecture but of a certain refinement, This gives aXII Positive lower bound for the absolute value of the linear form A= aqui +... + agUn where u;, ... ,Un are algebraic points linearly indepen- dent over Kand a1, «+. a, are algebraic numbers not all zero. More exactly, we show that for ¢ > O and a positive integer d there exists an effectively computable positive constant C = C(e,d,W1, +++/Un,S2r93) Such that [a] > ce™ whenever the degrees and heights of ai, 1%, do not exceed d and H respectively. This compares favourably with the estimate of Feldman in (16] for n = 2 and un- restricted F which quantifies the result of Schneider mentioned above. Such is the major part of these Notes; there remain the appendices. In the first of these we present a longer but more enlightening proof of a lemma appearing in Chapter III; it also gives a slightly better result. In Appendix II we consider two lemmas with a certain independent inter- est which were proved in Chapter II and Chapter VII. These concern the zeros of polynomials in several complex variables, and we use various methods to show that best possible estimates can sometimes be achieved. Appendix III is an outline of how the proof in Chapter VII may be modified to yield a further transcen- dence result for algebraic points. The original proof gives a criterion for the possible vanishing of A; we now show that if A does not vanish it is a transcendental number. Together with the earlier result this implies that the numbers 1, u,, ... ,U, are linearly independent over Aif ui, ... ,Uy are linearly independent over K.XIII Finally Appendix IV deals with an application of the theorem of Chapter VII. If C is a curve of genus 1 defined over Q, it is known that the magnitudes of rational points on C do not increase too rapidly in terms of their denominators. This does not appear to be explicitly stated in the literature, but it is implicit in the ideas of Siegel. It also follows froma result of Coates [11] giving a lower bound for the absolute value of A when iy see s%_ are rational integers; but the proof of this result uses the Thue-Siegel-Roth theorem and consequently shares the non-effective features of Siegel's work, When Chas complex multiplication our theorem allows the estimate for rational points to be improved and made effective in terms of the Mordell-weil group. The research in these Notes was done at Trinity College, Cambridge with the support of a Science Research Council grant. I wish to thank Dr. J. Coates for generously making available to me some preliminary studies relating to the algebraic points problem of Chapter VII, In particular the credit for the main idea of Lemma 7.7 must be divided between him and my research supervisor Dr. A. Baker. But ultimately the work owes its existence to the advice and encouragement of Dr, Baker who first suggested the topic of the title as a possible field of research.NOTATION For a meromorphic function £(21, ... ;2n) of the complex variables 21, ... ,2n we write (B/B2_ PME (Zay vee Bade (Z1_¢ eee 42q) = (8/82,)™ and for non-negative integers 2,m we denote by P(z,£,m) the m-th derivative of (P(z))*. Ifa is an algebraic number we define its size as the maximum of the absolute values of its conjugates and we say that an integer a > 0 is a denominator for a if aa is an algebraic inte- ger. The height H of a is the maximum of the absolute values of the relatively prime integer coefficients in the minimal polynomial of a, Thus a has a denominator not exceeding i given by the leading coefficient of this polynomial. Also if d is the degree of a, the inequality < dH is easily verified (see [1], p.206) and clearly the same estimate is valid for the size of a. For a real number x we denote by [x] the greatest integer not exceeding x. Finally we use c, c1, to signify positive con- stants depending on various parameters, and we adopt the convention that the constants appearing in the proof of a lemma are allowed to depend only on those parameters appearing in its enunciation.CHAPTER ONE 1.1 Introduction In this chapter we shall establish a transcendence measure for the ratio t = w2/w, when P has no complex multiplication. Since the measure turns out to be stronger than that implied by the more general result [16] of Feldman, it has some independent interest; however, it will also be an essential component of the proof of Theorem II. Theorem I For any ¢ > 0 we have = a] > C exp(-(log #)7**) ay for all algebraic numbers a of height H, where C > 0 depends only on t, € and the degree of a. 1.2 Preliminary lemmas In this section we collect together some elementary lemmas needed for the proof of Theorem I. The first two deal with simple properties of Plz). Lemma_1.1 There is a closed disc {) in the complex plane with centre at z = & such that(4) Plwiz), Plwez) are regular in D (44) If z, 2! lie in D El eilz - 2'| < |Plwyz) - Plwyz")| < c2| for i = 1,2, where c,, cg depend only on w and wz. Proof Since " (kw,) #0, we can find a closed disc ) centrea at z = & in which P(w,z) and Plw.z) are regular and simple (see for example [14] p.260). Then the continuous function (Blogz) - Pluyz"))/(z - 2") (242) og PY (uy 2) (z = 2") does not vanish on the compact set ox J) and thus its absolute o(z,2") value possesses a positive lower bound and a finite upper bound in this region. Lemma_1.2 For any positive integer 2, the m-th derivative of (P(z))* can be expressed in the form Piz rym) = Yule, tt ,e",m,2) (Plz) ) © (PH (2) E(B (2) where the summation is over all non-negative integers t, t', eh 2t + 3t! + 4t" = m+ 28 and U(t,t',t",m,£) denotes a rational integer with absolute value at most micy” for some absolute constant c3. Proof See Lemma 2 of [3]. Lemma 1.3 For an integer L > 1 and complex numbers p(A) (0 < A < L) L let o(z) = J pare. ioSuppose 09, «+. ,0, are distinct complex numbers with _ ly - mI oo: aoe In| 1 2 6. Then for all we have PAD] € (eyS, m | I 4 S/6F oe 16 (op) | for some absolute constant c,. Proof We write and use the interpolation formula of Lagrange e o(z) = ae oy P(Z)/(P' (ay) (z - a) to estimate the coefficients of ¢(z), noting that [P'(o,)| > 8% ana also that the coefficients of P(z)/(z ~ on) do not exceed (cs$)” in absolute value. Now we let & be the set of points of the complex plane that are congruent to a point of ) modulo the periods of P(wiz) ana the periods of P(w2z). The next lemma exhibits tmany'! points of &. Lemma_1.4 There is a constant c.g depending only on w; and wz such that for any integer L > c.g the following is true. There are at least L+l distinct integers ro, --- /r, with absolute values at most c7L such that z(i,m) =k + QL? + ryt (2) lies in & for all integers 2,m with 0 < a,m< L. Proof Let ¢ be a constant depending only on wi and w2 so largethat the estimates below are valid. When A, B, C are non- negative integers not exceeding T > c, the points Aw,? + Bwywe + Cw2* lie in a square centred at the origin with side at most cpT. This square may be divided into at most cy(cT)* smaller disjoint squares of side c‘j; hence if uTS < cg (L + 1)c?T? < T? (3) at least one smaller square contains at least L+2 of these points. Then the L+l differences between a fixed one and the others have absolute values at most 2c! and are of the form = Pr? + Gyirwe + Tye? (o< ms Lb). The resulting value of z(2,m) given by (2) is clearly con- gruent to % + 2L7? modulo the periods of (wz). Also since yt = dy ~ Padi /o2 + bg/wive we see that z(2,m) is congruent to & + £L? + 6,/uiW2 modulo the periods of P(w,z). Thus if L > ¢ the point z(2,m) lies in & for o < a,m< Le Finally if rq for m #n it would follow that Ip? + quiwz| < |@| + [6n] < 4e7 where Pp = Pm - Pny d = dm - Gn. Since p,q are not both zero and w/w; is not real this is impossible for large enough c. Therefore rj, ... ,rz are distinct and from (3) Izq| < 2P < cob. This completes the proof.Lemma 1.5 There is a constant cig depending only on w; and we with the following property. For an integer L > 1 and complex numbers p(Ai,Az) (0 AiyAz < L) let * se " o(z) = FY pay ede) (Pwr 2) (Pw, z) Mo Meo and for r % 0 let M(r) be the maximum modulus of $(z) at points z of © with |z| c. From that lemma there are L+1 distinct integers ro, -.. sr, of absolute value at most cj;3L such that the points z(iym) = & + QL? + ryt (0 < a.m <1) lie in&. Hence |z(i,m)| < cb and if M= M(cyL) we have Jo(z(2,m))] < Me. (4) oo aQ2) = I P(r ede) (y (2) (5) where ~ y(2) = Pw, + 0 LL?) it follows that o(z(Lym)) : - Ya) (x (m)) (6) Ase where x(m) = (wz z(%,m)).We proceed to use Lemma 1.3 to derive upper bounds first for laQ2)] and then for |p(i,42)|. From Lemma 1.1 it is clear oo ee [x(m)| < ci , |x(m) - x(n)| > cisé, where B= [mq - r,)t - M] and % is the pole of [(w2z) nearest to (rq - r,)t- If Q2 = q + pwi/w2 for integers p,q we must have |p|,|q| < cish and so € = Jwiuz|7*|pur? + quiwr - (tq - ry)w2* | > erry provided cy) is large enough. Therefore from (6), (4) and Lemma 1.3 we have for all Az laQ2)] < G/)cebm. Finally it is easy to see that if 2 #m ly(@)] < cis, ly) - y(m)| > bse , whence from (5) we obtain for all hi, IPOr Az) | < (L/u) EM, We conclude this section with two simple but useful algebraic lemmas. Lemma_1.6 Let a1, «++ ,a, be algebraic numbers of degrees at most d and heights at most H > 2, and write L= Qa, ..- sa))- Then there is an integral basis of L over @ consisting of numbers wi, s+. ,Wy Of size at most H&, Furthermore if 4 is an algebraic integer of L with size s thenA= mw, +... + mywy for rational integers m, ... ,my with absolute values at most H&s. Here cz; and cz: depend only on n and d. Proof Let N {L:@] and denote by a1, .-. Oy the embeddings of Linto€. For integers , ... ,2, we write a= ta toe + tay and 8, = 2, Ala - ay) aQci M > 0, and let uj, (1 € i ¢M, 1 c. Here c is sufficiently large for the validity of the subsequent estimates. On setting 6 = 6/100, k= [(log H)?**] , B= [x*} the proof proceeds by a sequence of lemmas. Lemma 1.8 There are rational integers p(\i1,42), not all zero, with absolute values at most H*, such that the function oz) = F FB pOysdad (Plu 2)) (Pla) AZo Aso satisfies 10, (a) < exp(-K¥*'%) (8) for all integers m with O k. Thus if £ = [F:Q]and wi, ... swe is an integral basis chosen according to Lemma 1.6 we may write pakQ(ha,Az-m) = niwi + +. + ngewe where nj denotes a rational integer with absolute value at most H*K, It follows that if the f equations Lo: Y oY pir =0 (lei £) Aso Aso hold for all integers m with 0 < m< k we shall have A, = 0. These are M = (k + 1)£ equations in N = (L + 1)? unknowns P(Ar+Az)- and since N > kMe > aM we may use Lemma 1.7 to find a non-trivial solution in rational integers with absolute values at most Nimax Ing] < Hoe. It remains to verify the inequalities (8). But it is clear that w:™%,(%) is the expression A, with a replaced by" tT in Q(Ai,Az mm); since 6 < 1/200 we have 3 + 1008 > (2 + 406) (% + 186) and hence from (7) < exp (-k%*% In - Thus for 0 < p< k* we see that Mens a) | = aM] « ep |t - a] < expi-k and if m< k* the coefficients of a“ in Q(A1,A2,m) do not exceed x" in absolute value (cf. (9)); whence for all integers m with 0 < m< k* we have 1 8 Jui 6,4) — Ag] < exp (-k™ de (10) The assertions (8) now follow. Lemma 1.9 Suppose Z > 6 and let $(z) = P(z)0(z), where Piz) = i A, (ye = ayy and 2; runs through all periods of ®(z) with |9;| < jwi|Z (4 = 1,2). Then $(z) is regular for |z| < Z, and for any z with |z| < 4Z and any non-negative integer m we have le(z)| < Hoek gowkZ* , |p czy] < mmzent2” Proof The first inequality is clear from Lemma 1.8 and lemma 1 of [3], while the second is a consequence of the Cauchy integral . P(z) dz “ond jy Te = yet P(t) taken over the positively oriented unit circle with centre at z=12 Lemma_1.10 105 For all integers m with O< m< k"” we have ies158 1, (k) |< exp(-k . Proof Since $(z) has period 1 we see from Lemma 1.8 that for all integers s,m with lesche= {k*?] , oc mek qu) the inequality Jo,(s + k)] < exp(-K'"%* is valid, We set Z = 10h and denote by $(z) the corresponding function defined in Lemma 1.9. Further we write Fiz) = Tl (2-8 - W)* Therefore for integers s,m in the ranges (11) we have g ‘im’ tats = FL (R)r(s + Way (5 + 0) The right side of this may be estimated by means of Lemma 1.93 since LZ? < cyhk'® < Kk (12) we find that cig Par lon(s + k)] < k ince |®.(k) | < exp(-k ye (13) Let ¢ be any number in the region Q of Lemma 1.1. We proceed to derive an upper bound for |¢(z¢)| using the formula oh (2 f (z) az a (s+) He + ah | qin - oh D, eee rime) (ua) FC) Tz c)F (2) Z me § where C is the positively described circle |z| = 5hi = f 2s Tims) = | “(= yFt) % + and C, is the positively described circle with centre at13 s+ and radius that of D. Note that since this radius can- not exceed %, the circles C, do not intersect each other. For z on C we have from Lemma 1.9 Jo(z) | < Hos 2 Geb 2" while for each linear factor of F(z) Jz-s-%| > 2]¢- 5-4] and from this last inequality IP(c) /F(z)| < 2*, Also, since |F(z)| > 1 for z on Cy, it is clear that |z(m,s)| < ch whence from the estimates (12), (13) and log H< REF ce! |E(t)| < (anh < et we deduce that fo(c) | < exp(-cishk) « Now the absolute value of each linear factor of P(t) exceeds 2 with at most cyL exceptions, and since — is in } we have the lower bound cy for these exceptions. Thus |P(¢)| > 1 and we conclude that J@(c)| < exp(-eishk) We now use Cauchy's integral sa = BE | PELE taken over the positively described boundary of 9 to deduce that if 0 < mc ky = [kl] J@n(%)| < exp(-cyehk) . Therefore from (10) JAn| < exp (-cishk) «4 On the other hand A, is an algebraic number of degree at most cao, and the estimates of Lemma 1.8 show its size to be at most HS ; thus if Ag #0 we obtain the lower bound [Am] > We. Since ky logH < hkt-* this implies A, = 0, and the inequality of the lemma now follows from (10). Lemma 1.11 et Joey < et, For all z in& with < we have Proof From the preceding lemma we see that Jon(s + | < exp(-K¥t's® y for all integers s,m with léesem = (kK) , oc mek. (as) We define the function P(z) as before with h replaced by hi, and since a Bh? < cahikyk® < k* (6) we deduce that (13) holds for the range (15). From the periodicity of ¢(z) it clearly suffices to prove the lemma for complex numbers ¢ in & with ts Isl < 1 [Re(t)| < &e We write k, F(z) = NM (2-8 - 4)" and use (14) when C is the positively described circle |z| = 5h, and the sum runs over integers s,m in the ranges (15).15 The additional restriction on ¢ ensures that |I(m,s)| < - and from (16) we eventually find that Ie(c)| < exp(-cashiki). As before |P(¢)| > 1, and therefore Joc] < ek* which completes the proof of the lemma We now prove Theorem I as follows. From Lemma 1.5 we see that for all 41, A; IpOi,2)| < (L/weat where » is the minimum of the absolute value of the linear form A = Aw? + Buyw, + Cw2? for integers A, B, C, not all zero, with absolute values at most cab. But Juv?al = Ja + Bt + cr?| 2 Jo] - exp(-K?7 _ where o =A + Ba + Ca’. Now o # 0, since if a ¢ 2 by supposition the height of a is exactly H > cab. Thus |o| > (LH)S* , which implies u > He, and finally [POrsAad| < Heel ek oat He and the functions F(A,u,z) are defined by F(A,0,2) = F(A,Z) (0 c; there are integers A, B, C with max({A],|[B|,|C]) = H such that ioe etles®) (i) < |Aw,? + Bwoyw2 + Cw2?| < x7,\ (44) 9s” < H< c2x? (ili) |C] > cake Proof This is another application of the Box Principle resembling the proof of Lemma 1.4. When D, E, F are non- negative integers not exceeding T > c (where as before c is sufficiently large) the points Dw,;* + Ew;w, + Fw? lie in a square centred at the origin with side at most c.T. This square may be divided into at most csT?X? smaller disjoint squares of side 4x; hence if yr? < csT?x? < T? at least one of these squares contains at least two such points and so their difference 6 = Aw? + Burwy + Cor? has absolute value at most X7. Also H = max({A|,|Bl,|C]) < 2T < cx, and this proves the right sides of (i) and (ii). For the left sides we denote by a, a' the roots of the equation A + Bx + Cx? = 0 so that 6 = Cw? (tT - a) (tT - a"). Then from Theorem I with « = 4 it follows that lel > coe ti > etignt provided that X > c. Also since |6| < X7? we may assume without loss of generality that A See xn te and now Theorem I with ¢ = } shows that H > e( are19 remains to prove (iii). But since tT is not real we have Aw)? + Bwrw2| > camax({A],|B]) and the truth of the assertion is now apparent. For the proof of Theorem II we shall require a lemma concerning the values taken by polynomials at points which are well-distributed in a certain sense. Lemma 1.3 provides such a result for polynomials in one variable, and we now present a more specialized version for three variables. In €? we define the absolute value of z = (z1,22,23) bY = fei]? + |za]? + zs]? and we call z a real point if z1, z2, zs are real. We denote by B the unit ball < 1, and we let B, be the set of real points of B. Lemma 2.3 3 For an integer L > 1 let 4 be a subset of © containing a point within (21L?)"! of every point of %,, and for complex numbers p(Ai,AzrAs) (0 € AreAarAas € L) write Lok o£ oz) LF YF pOaedardsdar adeas. Aso Aso Age Then for all Aiy Aa, As we have IpOisdarda) | < (eipbiee" sup jo(z)|. Zink Proof We first note the result of Markov (23] that if f(x) is a polynomial of degree at most L then ' _ . 17 882, |£' (x) | n2z). For an integer L > 1 and complex numbers p(A9/A1+Az) (0 € AorAask2 < L) let ‘ ‘ ¢ Ae A, a, oz) = J fF plroeAaeAa) C£ (zd) (Plwi zd)" (Plw2z))* oso Ago and for r > 0 let M(r) denote the maximum modulus of ¢(z) at points of & with |z| < r. Then for all Ag, Airy Az we have IpQordrsd2) | < (ew L) oo M(cay exp( (LogL) *) where Cy, C13 and cy depend only on 61, 82, wi and we.Proof Without loss of generality we may suppose that L > c for a sufficiently large constant c and also 82 #0. We use Lemma 2.2 to find integers A, B, C, not all zero, and integers D, E, F, not all zero with max(|A|,|B],|C|) < L™, max(|D|,|E],|F]) < exp((logL) *) such that if we write 6 = Aw? + Buyw2 + Cw2* » © = Duy? + Ewiug + For? then l6) exp(-(ogL) 5), whence < exp((logL) *) and so Jz(eorriera)| < exp((logL) *) .23 It follows that for all ro, riy r2 |o(z(rorrisr2))| < M = M(exp((logL)*)) , and we shall base our arguments on these inequalities. For an integer r with absolute value at most L? we write z(r) = % + 6x/wiweb? and x(x) = Plwsz(r)) (i = 1,2) and we first verify the inequalities [Plwgz(ro,ri,t2)) — xy (ry) < Le? (i = 1,2). (23) In fact for i = 1 the left side of this vanishes, by (21). Also from the definition of m we have J@xi/wiL? + mé/w; - Or2/iL7| = dur? | {m- (re - r1)/|6]L7] < cas 16] so that from (22) and Lemma 1.1 we deduce that the left side of (23) for i = 2 is at most eis ([6| + 2le]) < L*. Next we examine the behaviour of f(z) at the points z(ro/¥i,t2), and for brevity we write " gy (2) = clwyz) - nyz, (4 = 1,2) whence £(z) = Bigi(z) + B2g2(z)- Then on using the Legendre relation wan, ~ ing = 2nd we see from (21) that gi(z(ro,ri,r2)) + 2mwirgF/w, = gi (z(ri)) - 2mimC/w,. (24) Similarly if = 2(r1) + (mo + roe) /wiwe24 we find from (22) that G2(zlro,risr2)) + 2wixrpD/w2 = g2(z') - 2wimA/we. (25) Since the points z', z(ri) lie ind, the corresponding values of g; and gz are bounded, and hence from (24) and (25) \€(z(to,ra,r2)) - oh] < cy (2 + |mal + |[mc]) < exp((logh)®) (26) where Q = -2n4(BiF/w1 + B2D/w2). It is now necessary to derive a lower bound for |A| that justifies treating the right side of (26) as an error term. For brevity write 8 = -8:/82. Then 8t = D/F + w2h/2niB2F ana 8+ = -E/F + €/wywaF + w1A/20iB2F and these imply |rm(Bt)| < cw JAl/JF] , |tm(B+t)| < cos (1 + [A|)/|F fee) By hypothesis @ and t are not complex conjugates, and since t is not real, this means that not both of the left sides of (27) can vanish. Therefore [a] > calF] - 2 and a lower bound for |F| is provided by (ii) and (iii) of Lemma 2.2; we find that |F] > exp((1ogL)*) whence Ja] > exp((logL)”). (28) The inequalities (23) and (26) express the fact that the values of the functions £(z), P (wiz) and Plw2z) at thepoints z(ro,ri1,r2) are in a sense distributed independently as ro, r; and rz; vary. By a simple change of variables we can achieve the situation of Lemma 2.3. Let be = L590, ay = Pkwy) » by = Ow, PChwy)/wrw2b® (4 = 1,2) and define the polynomial pou OY Me , A YY pOdoeda eda) (boxe) (artbixi)" (aztbex2)*s Azo eo From the power series expansion of (wiz) about z = % we have Q(X Xi 7x2) = [xq (eq) - ay - byry/L*| < ca L® (i = 1,2) and hence the point (¢,£1,£2) of C defined by bobo = £(z(rovrisra)) + ay + byEy = Plwyz(rorrisre2)) (i = 1,2) satisfies, from (23), (26) and (28) [Eg - ry/L3] < cy L™ (i = 0,1,2). Now if (nisnz-n3) is any point of % there are integers ro, Yi, Y2 with absolute values at most L? such that Ing - ry/L?| < 1/2u3 (4 = 0,1,2). It follows that within 2L of each point of B, there is a point (6 ,61,/62) with JQ(EorE1,E2)| = [Plz(roerasra))| < My and therefore if q(A»,A1,42) denotes the coefficient of xotx pic! in Q(xo.%1/%2) we see from Lemma 2.3 that laQordaydad| < Lom for all dor Air Ave Finally since L P(AorAiyA2) = (-a2) ba q(Aorarha)28 we conclude that IpQordasda)| < Loe, and this completes the proof of the lemma. 2.4 The Au: iary Function We commence the proof of Theorem II by constructing the auxiliary function. We have seen that in [12] Coates proved that a non-zero linear combination of the numbers Wir W2r Nir N2 and 2mi is transcendental. We shall show that the equation 41 + aawe + Bini + Bane + ¥.2mi = O (29) is impossible if at least one of the algebraic numbers a1, a2, Bir Bay Y is non-zero. Accordingly we assume that (29) holds with algebraic numbers a,, a2, 81, 82 and y, and from the result of Coates [13] establishing the impossibility of (29) with 8, = 62 = 0 it is clear that we may suppose B2 # O. Further, as in [12], we suppose that %g2, %gs and the co- efficients in (29) are algebraic integers, and as usual this involves no loss of generality. We shall eventually derive a contradiction, and this will prove Theorem II. For a large integer k we set b= ikl, n= (k%] and we write (21422423) P(AgrArsAzeAs) (E (21 ¢22-Z3)) (Plwiz1)) (Plw2z2) er, where £(Z1,Z2-Z3) = Giwizy + G2weZ2 + Bio (wizi) + B2b(wet2) + 2wiyzsar and the coefficients p(Ao,Ai1,Az2,A3) are yet to be determined. Formally this is the auxiliary function used by Coates [12], but his assumption is that the right side of (29) is an algebraic number a, ¥ 0. However, the vanishing of a» does not invalidate his arguments except in the concluding para- graphs. Thus denoting by c,ci, ... positive constants depending only on a1, a2, 81, B2r Yr W1 and w2 we see that for k > ¢ the following lemmas hold. Lemma 2.5 There exist rational integers p(Xo+A1/AzvA3), not all zero, with absolute values at most k'™k, such that cam, (Sth sth sth) = 0 for all integers s with 1 < s < h and all non-negative integers m, mz, m3 with m, + mg + m3 ¢ ke Proof See Lemma 7 of [12]. Lemma 2.6 Suppose that Z 2 6 and let O(ZryZ2e%s) = O(Zr-2Z20%5) yt Te (wg - ayy where 2, runs over all poles of F(z) with |a;] < Jw4|Z- Then $(z1,/22,23) is regular in the disc |z,| ¢ Z (i = 1,2,3), and for any z with < 4% and for any non-negative integers mM, M2, m3 with m) +m, + my € k we have [mmm (2-2 2)| < kMRZOEZ Proof See Lemma 8 of [12].28 Lemma 2.7 Let Q, S, 2 be numbers with 1 < Q < S < Z-1, and let Mi, Mz, M3 be non-negative integers with m; + m2 +m: < k. Suppose that q, r, s are integers with q even, (r-q) = 1, and l 2) (s' + r'/q') - (s + x/q)| for each factor of F(z); the number of sets s, r, q occuring here is at least c,S;Q,2, and so @ > 20:2" |e(gt + r¥/qt) |. Hence on noting the estimates oye 2 ee te 6 oe cee hea we deduce that Jy(st + xt/qt) |< 2-cexSs@e* on the other hand, the hypotheses of Lemma 2.7 are satisfied with Q Or, , S = Sra and primed letters through- out; since y(s' + xr'/q') is the left side of (31) multiplied by a non-zero factor we have the lower bound : [w(st # rt/qt) | > Seoq OKO31 But kO,,¢ < cr ht and therefore this is inconsistent with the upper bound. This contradiction proves the lemma. For the next lemma we write K = exp((log k)™) . Lemma 2.9 For all points z of & with absolute value at most K and all integers m with 0 < m< L we have le em(zzz| K? certainly satisfies I+1< (log k)*. Therefore if we define the function $(z1,22/23) with 2 = 10S; and if for any integer u with O < u < L we write V(2) = 9 oy (2122) we can deduce the inequalities (32) from Lemma 2.8. As in that lemma, the maximum modulus principle gives the estimate [weer | < 2reuk Se2F gg K* Finally wie) = Poo ylGrb ro) where Pe it A ws > oy and the product is over all poles 24 with || < |w4|Z (i = 1,2). Clearly there are at most cy periods 2, such that |wyc - 2;| < 2, and since ¢ lies in & we have32 wit - &:| > ci, for these periods. Hence |P| > 1 and the lemma follows. 2.5 Proof of Theorem II We define the functions Aa i. FOQs) = POs 21s22,23) = 2 DL pOorhasha sa) (E(x ,22928)) awe MFO deo (P(wizid (Plw2z2) so that @(Za¢Za423) = 2 F (Ag) e27EAazs We assume that not all the _ F(A3) vanish identically and we shall deduce a contradiction. Suppose ro, ..+ stm are the distinct integers between 0 and L such that F(rp) is not identically zero; thus 0 < M < L and Oz pz20%s) = i F(rpetts We are going to use Lemma 2.1 to construct a new function ¥ in which the exponential terms are absent; to this end we define the functions GUY) = GO, ¥-21 122/29) (0 < uv sm by G(u,0) = F(zp) o ry) yo = 2nd os } and W(z1,Z2,23) is the Wronskian of the functions (0), ... ,@(M) with respect to z;, where . Y ni ter: @(v) = O(v,21,22025) = JY (Qnizy) (rye, pao We proceed to establish an upper bound for |w(z,z,z)| when z satisfies the conditions of the lemma, The first column of the determinant defining W(z,z,z) consists of the terms %o,0,n (22,2) and from Lemma 2.9 these are at most a sn absolute value. To estimate the other terms we observe that if z is a complex number with |z3 - z| = 1 [Ovezvz,29)] < Koek and we have the integral (recall the integrand is regular in Za) (/aes¥ (v2.2.2) = Be | ee) as, TE es taken over the positively described unit circle centred at z. Thus the other entries in the Wronskian determinant are at most Kes'K” in absolute value, and it follows that |wiz,z,2)| « KAEKY gk 5 gi Finally since IRp > 2, le] < eth K we deduce the inequality of the lemma immediately from (33). It is easy to verify by induction on v thatLoL ogy m Gury) = (ma YE YE ave edn Aa em) (£ (215225250) AZO MeO mao : : (Pwr zi)" (Plw2z2)) 7, where @ = ahordadarm) = (2) 5" Y Aq QoL) fe Oo-mL padre arte) ifm < dy and q is zero otherwise. Thus we see that the coefficients q are integers of the field Q(y) whose sizes are at most k“**. It follows that the function ¥ may be written in the form 2122/23) = 5 (Vo rVarVa) (E£ (21/22 423))" (Plwizid)” (P(w2z2))" where N = L(L + 1) and the coefficients r(vo,v1,V2) are algebraic integers of Q(y) with sizes at most cw*. We now observe that in view of (29) £(2,2,2) = 81 (e(wiz) — niz) + B2(e(wez) ~ naz) (34) and further Biw, + B2w2 # O since 8; # O and w2/w: is trans- cendental. Therefore from Lemma 2.4 and Lemma 2.10 it follows that Ix (vorvirva)| «Noe ce for all ve, vi and v2. If x(vo,Visv2) # 0 this would contra- dict the obvious lower bound cw” ; hence r(vorvirv2) = 0 for all vo, Vi, V2 and ¥(21,22,23) 18 identically zero. From (33) the Wronskian W(z;,Z2,z3) vanishes identically and so by the basic property of Wronskians there exist meromorphic functions H(v) = H(v,Z1,Z2) (O < vy < M) independent of z3 and not all zero, such that Hv) ev) = 0.35 Thus we have * LT BQ)F(ry)e7 hs = 0 Dp where . v E(u) = E(we2ir22) = J (2nir,) (vy) (35) Since the exponential function ten algebraic function, this implies that E(y)F(x,) = 0 for all y and therefore E(u) vanishes identically for all . But the determinant of the equations (35) is of Vandermonde type and plainly non-zero; it follows that H(v) = 0 for all v. In view of the assumption made at the beginning of this section, this contradiction implies that F(A) vanishes identically for all 43. But, using (34) again, the function F(\3,z,/2,z) is of the type considered in Lemma 2.4. A rather trivial application of this lemma shows that all the coefficients p(\y,/A1,;Azeds) must vanish, and this contradiction finally establishes Theorem II. As a postscript we remark that this last use of Lemma 2.4 is a little unnecessary, for it is easy to prove directly the algebraic independence of the functions £(z1,22/23)+ Plwizi) and Plw2z2). (See, e.g., [18], p-111.)CHAPTER THREE 3.1 Introduction In this chapter we assume that F(z) has complex multiplication over the complex quadratic field K, ana our principal object is to prove the following theorem. Theorem III The dimension of v over A is four and the numbers 1, Wir Ni and 2ni are a basis for V. 3.2 The Upper Bound We show in this section,among other things, that the dimension of V is at most four. Since the ratio Tt = w2/w, lies in K, there exist coprime integers A,B,C # 0 such that A+ Brt+cCr? =0, and the relation tw; - wz = 0 implies that the dimension of V is at most five. To proceed further we prove the following lemma. Lemma_3.1 There is an algebraic number « of the field K(ga,gs) such that Ani ~ Ctn2 = Kwa. (36)37 Proof We define « by (36) and consider the function £(z) = -Ag(Cz) + Cte(Ctz) + CtKz. (37) Then we have £(z + wi) - £(z) “ACh; + C?tn2 + CtTKwi = 0 and, since eee we also have £(z + wz) - £(z) = -ACn2 + CT(-Ani ~ Bn2) + Crkw2 = 0. Therefore f(z) is a doubly periodic function with the same periods as P(z), and it follows that f(z) is a rational function of ((z) and f(z). Let @ be an arbitrary mono- morphism of Kig2,g3,«) fixing the field F = Kiga,g3), and let £%(z) be the function obtained by applying this to the Laurent expansion of f(z) about the origin, Since o acting in this way leaves f(z) and P'(z) fixed, we deduce that £7(z) is also a rational function of P(z) and P(z). Hence the function £(z) - £7 (2) = Ct(K ~ K%)z is an elliptic function, which is only possible if x = K Since o is arbitrary, this proves that x lies in F. The preceeding lemma implies that the numbers 1, w1, ni and 2ni certainly span V; hence Theorem III will follow from the linear independence of these numbers over A. This will be proved in the next section after we have derived an additional property of the number x.Lemma _3.2 The algebraic number x vanishes if and only if one of the invariants g2, gs vanishes. Becoe For any complex number z in the upper half plane Im(z) > O we write q = e?"¥ and define the function E,(z) =1- 24 } ng™(1- qq")? =1- 24 iy o(n)q? eel where o(n) denotes the sum of the divisors of n. This is a well-known function with modular properties, and it satisfies the functional equation (e.g., (24]) E2(=1/z) = 27E2(z) + 6z/ni « (38) Furthermore we have B2(t) = 3wyni/n?. From this together with the Legendre relation or its equi- valent (38) we obtain the formula kK = <0? (B + 2C1) (1) /3wi? where (2) = Ex(z) - 3/(nImz). Therefore the vanishing of « implies that $(t) = 0, and we proceed to find all the zeros of $(z) in the upper half plane. From (38) it is easily seen that $(z) has period 1 and o(-1/z) = 276(z), (39) so that the set of zeros is mapped into itself by the modular group I, Hence we may confine our search to a fundamental region R for I, and we take this as the region defined bymkhesc 0; by the remarks above we know this to be at t = If s = -k we obtain the equation 1-24 J (a o(mpe?™ - 3/nt = 0. (40) aft The derivative with respect to t of the left side of (40) is40 Bre? (-1 +f (-1) "no (nye ?™M*) 4 aynt? (4) and the sum from n = 2 to © can be estimated as before. The same upper bound 3 is obtained, and so (41) exceeds -288ne?"* /5 + 3/nt? = (2881/5t?) (15/2880? - t?e 2"), But the maximum of t*e?"* for t » 4¥F occurs at t = kv3 itself; the inequality 15/2881? > 3y/4 is easily checked, and it follows that the left side of (40) is monotonic. We have already found the solution t = 4/3, and this completes the proof of the lemma. 3.3. Proof of Theorem IIT We complete the derionstration of this theorem by proving the linear independence of the four numbers 1, wi, ni and 2mi. From the transcendence of w;/n and w;/ni and the result of Coates in [12] it will suffice to show that the relation aw: + 8ni + ¥2nd = 0 (42) is impossible with gz, 493, a, 8 #0, y #O all algebraic integers. Accordingly we proceed to deduce a contradiction from the hypothesis that (42) holds. For a large number k we set b= [k] , n= [xt and £(zi,Z2) = awiz, + BO(wiz1) + ¥2nize F then we write eo FE A 2m Aaa, O(ziez2) = J J J pOrardreda) (£(21,22))*(Plwizr) errs , EO Azo hsso where the coefficients p(\o,A1,A2) are chosen by the following lemma.a Lemma_3.3 There exist rational integers p(Ao,A1,42), not all zero, with absolute values at most k™™* , such that am, (Stk Stk) = 0 for all integers s with 1 < s ¢ h and all non-negative integers my, mz with m + mz ¢ ky Ezooe Since the parameters are the same as those of the aux- iliary function of Chapter II, the estimates in the proof of Lemma 2.5 are valid a fortiori; hence that proof remains virtually unchanged in the present circumstances. We now need an analogue of Lemma 2.9, and to state it we redefine K as k’?, Lemma_3.4 Let z be a complex number congruent to a point of D modulo the periods of P(wiz) whose absolute value does not exceed K. Then for all integers m with 0 < m< L we have 19m (2-2) | < ae Exot The conditions on z are enough to ensure that the extrapolation techniques of Chapter II are applicable. Con- sequently the proof closely follows the proofs of Lemma 2.8 and Lemma 2.9, except that the range of values of J in Lemma 2.8 is now unnecessarily large. We now construct the Wronskian with respect to z2 as42 in section 2.5; the corresponding function ¥(z1,22) takes the form oe ¥ Vezieza) = FY elverva) (E(za,z2)) (Pwd), where N = L(L + 1) and the coefficients r(vo,vi) are algebraic integers of Q(y) with size at most c\*. In view of the remark at the end of section 2.5 the vanishing of these coefficients will prove the Theorem since the functions £(z1,22) ana f(w:z1) are plainly algebraically independent. To show that r(vo,v1) = 0 for all vo, vi we follow the proof of Lemma 2.10 and conclude that for all z in the range of Lemma 3.4 we have [¥(z,z)| < ef’, We use this inequality at the points z= 2(f,m) =k + LN? + mt (0 < &,m< N). If we fix £ and write svi) = J rlvorvi) (y¥(R))” wee where y(t) = PCr + win?) it is clear that 4 J s(vi) (xtm)”™ ° ¥(z (2m) ,2(2,m)) where x(m) = £(z(£,m) ,z(2,m)). From (42) and the Legendre relation we see that x(m) = x(O) ~ 2nimB/wi, and hence from Lemma 1.3 Is(vi)| « eX (O 20@Ip(z')| whence ly(z| < 2e@8T < Finally since z' lies in & we have |P(z',z')| > 1 as usual, and the lemma follows. We proceed to construct the Wronskian by writing ‘ L J pidorAavAa) (Plwrza) " (Plwez2) )* Aveo Jee (0 < Ae < L)aT and we assume the existence of an integer M » O such that precisely M+ 1 of the functions F(4,) are not identically zero. Denoting these by F(re), ».+ /F(ry), we define for Ocvem Ov) = O(vezi,22) = x, (x, cata)” (£ (2122) PE (EA), and we use Lemma 2.1 to calculate the Wronskian W(zi,Z2) of (0), ... ,O(M) with respect to z,. After some reduction we obtain the result m(ea et) W(ZiyZ2) = (a2W27) M (eq ~ ru) (£ (21622) U (zr 022) d>op where " a or Ulzi,Z2) = det Glurv)s bro ocpven and - - > RP yom Stu) = 6ovtae = ELD (fh) (aad Plmerdr edad (Plorzi) Plaza yharv) On using Lemma 1.2 to simplify the last term in the defini~ tion of G(u,v) we see that Ulzrez2) = J a(Plwrza)Y" (Plw2z2))* CP waza)” (Pwaz2))™ where q = q(vi,V2/V2',V2") and the summation is over all non-negative integers vi, v2, v2", V2" at most N = kL. Furthermore there is a positive integer a , independent of Viv Var Va'y V2" , such that a and aq are algebraic integers of the field Qigz/gs,az) with size at most cst” Since P(w2z2) is an even function of zz, it follows that the function V(zZi¢22) = Uz, ,%2)U(Z.,~Z2) may be written in the form48 vizisza) = fe (ps.02) (Ptwsz1))? (Plaza) where R= k#, and there exists a positive integer b such that b and br(p:,02) are algebraic integers of the above field with size at most cot. Suppose now that z is any complex number of & with [z| < k°. We use Lemma 4.1 to estimate |W(z,z)| by its first column, and we find that |wiz,z)| < e*” whence Iviz,z)| < ek, Lemma 1.5 now provides an upper bound for the coefficients weg k)4 r(pir02)1 the quantity y clearly exceeds e“%"" by the arguments of Lemma 2.2 and hence Ix (orsp2)| < e® (0 < pire2 < R)- Therefore r(p1,92) = 0 for all pi, P2, and so W(z:,z2) is identically zero. The argument now proceeds on lines established in section 2.5, and the proof of the result is easily completed.CHAPTER FIVE 5.1 Introduction In order to obtain quantitative refinements of Theorem II and Theorem III, it is natural to consider the linear form A = Go + Gi: + GoW2 + Bini + Bane + Y.2mi where ao, ai, G2, 61, 82 and y are algebraic numbers not all zero. The problem is then to find an effective lower bound for |A| in terms of the maximum d of the degrees and the maximum H of the heights of the coefficients. Partial results for the case 6; = 62 = y = 0 occur in a paper (4) of Baker and in our Theorem I; these take the form [a] > Cexp(-(log #)“) (43) for some absolute constant k > 1 and some number C > 0 depending only on wi, ®2 and a. In view of the work of the preceding chapters the general problem may now be solved; but unless y = 0 the actual lower bounds obtained for the case ao = O turn out to be much weaker than (43). However, if ao #0, a stronger result than (43) can be derived, and the purpose of this chapter is to prove the following theorem.Theorem Iv Let ao, a1, G2, 81, Bz and y be algebraic numbers with degrees at most d and heights at most H, and suppose a» # 0. Then for any © > O there is a constant C > 0 depending only on wi, w2, a and e€ such that |A| > Cexp(-log H(log log H)7**). In fact if a2 = 8; = 82 = 0 then 7+e may be replaced by 4+ and in this way we obtain quite a sharp transcendence measure for the number t+w where w is any period of (z). 5.2 The Auxiliary Function For the proof of Theorem IV we assume algebraic numbers ao #0, a1, az, Bi, Be and y exist with degrees at most a and heights at most H » 3 such that for some « withO c for some sufficiently large c. It is easy to see that we may take kg2, %g3 to be algebraic integers without loss of generality. Let 6 = €/30, and define k = [log H(log 10g #79], 2 = logk, he= [2178] bo = [277] , Dany = bo = [ke] , by = [kee]. We write £(Z1,Z2-Z3) = ayWiZ, + Ggw2zZ2 + Bi (wizi) + Bet (wee2) + Y-2mizs and _ @ziezarzs) = | FPF pthordupdaeds) (Elza /z2,za)) lero Aro Ato Ayo (P(wizr))* (Plw2z2) MerMasze where the coefficients p(Ao,A1,42,43) are yet to be determined.51 Apart from the change in parameters, this is the same auxiliary function as that appearing in Coates’ paper [12]. Hence from the calculations of [12] we see that for non- negative integers mi, mz, m3 we(2ndy> Foes 7 r mee P(Xor +++ sAs)PoPiPaPay mmm, (216%2 023) = ws where = (™ = (Ms) Meas gami baz, Pye (Bt) Ploeg deemeruy) 1 Ps = G ars ertthszs for i = 1,2 and on TT -T,) Po = J, J, wiuevaveta (tel)... (raya th) v8 gh (£ (21122423) %™ « In the formula for Py, the numbers Ur. = Ur(Tr Tr tietil yeti” wirdode Ug = Ug (t2,te' sterte! rte” Herth) are rational integers with absolute values at most yi!c,/"" pa and zie, respectively, while = ay - BRITT RY v a (4 = 1,2) with P=Plyzy) » B= Plwyzy) » P= Pwyzy). Finally the summation q, extends over all non-negative integers Gio (ido Cho ey ae t1 4 Qty" + 2t; + Sti" + 4ti" = yi + Aor ci tie «45 and J, over all non-negative integers t2, tz‘, ta, ta‘, ta” with ta + 2t2! + 2t2 + 3t2' + 4te" = ur + Tie t+! 6 tie