On Advanced Analytic Number Theory
By C.L. Siegel
Tata Institute of Fundamental Research Mumbai, India i
FOREWORD
“Advanced Analytic Number Theory” was first published by the Tata Insti- tute of Fundamental Research in their Lecture Notes series in 1961. It is now being made available in book form with an appendix–an English translation of Siegel’s paper “Berechnung von Zetafunktionen an ganzzahligen Stellen” which appeared in the Nachrichten der Akademie der Wissenschaften in G¨ottingen, Math-Phy. Klasse. (1969), pp. 87-102. We are thankful to Professor Siegel and to the Gottingen¨ Academy for according us permission to translate and publish this important paper. We also thank Professor S. Raghavan, who originally wrote the notes of Profes- sor Siegel’s lectures, for making available a translation of Siegel’s paper.
K. G. Ramanathan ii
PREFACE
During the winter semester 1959/60, I delivered at the Tata Institute of Fundamental Research a series of lectures on Analytic Number Theory. It was may aim to introduce my hearers to some of the important and beautiful ideas which were developed by L. Kronecker and E. Hecke. Mr. S. Raghavan was very careful in taking the notes of these lectures and in preparing the manuscript. I thank him for his help.
Carl Ludwig Siegel Contents
1 Kronecker’s Limit Formulas 1 1 Thefirstlimitformula...... 1 2 The Dedekind η-function ...... 13 3 The second limit formula of Kronecker ...... 20 4 The elliptic theta-function ϑ1(w, z)...... 29 5 TheEpsteinZeta-function ...... 41
2 Applications of Kronecker’s Limit Formulas to Algebraic Number Theory 52 1 Kronecker’s solution of Pell’s equation ...... 52 2 Class number of the absolute class field of Q( √d)(d < 0) . . . 69 3 The Kronecker Limit Formula for real quadratic fields and its applications ...... 75 4 Ray class fields over Q( √d), d < 0...... 89 5 Ray class fields over Q( √D), D > 0...... 102 6 SomeExamples...... 133
3 Modular Functions and Algebraic Number Theory 149 1 Abelian functions and complex multiplications ...... 149 2 Fundamental domain for the Hilbert modular group ...... 165 3 Hilbert modular functions ...... 184 1 EllipticModularForms ...... 220 2 Modular Forms of Hecke ...... 225 3 Examples ...... 231
iii Chapter 1
Kronecker’s Limit Formulas
1 The first limit formula 1 Let s = σ + it be a complex variable, σ and t being real. The Riemann zeta- function ζ(s) is defined for σ > 1 by
∞ s ζ(s) = k− , k=1
s s log k here k− stands for e− , with the real value of log k. The series converges absolutely for σ > 1 and uniformly in every s-half-plane defined by σ 1 + ǫ(ǫ > 0). It follows from a theorem of Weierstrass that the sum-function≥ ζ(s) is a regular function of s for σ > 1. Riemann proved that ζ(s) possesses an analytic continuation into the whole s-plane which is regular except for a simple pole at s = 1 and satisfies the well-known functional equation
s/2 s ((1 s)/2) 1 s π− Γ ζ(s) = π− − Γ − ζ(1 s), 2 2 − Γ(s) being Euler’s gamma-function. We shall now prove Proposition 1. The function ζ(s) can be continued analytically into the half- plane σ > 0 and the continuation is regular for σ > 0, except for a simple pole at s = 1 with residue 1. Further, at s = 1, ζ(s) has the expansion 1 ζ(s) = C + a (s 1) + a (s 1)2 + − s 1 1 − 2 − − 1 Kronecker’s Limit Formulas 2
C being Euler’s constant. We first need to prove the simplest form of a summation formula due to Euler, namely. Lemma. If f (x) is a complex-valued function having a continuous derivative f (x) in the interval 1 x n, then ′ ≤ ≤ 1 n 1 − 1 f (1) + f (n) f ′(x + k) x dx + − 2 2 0 k=1 n n = f (k) f (x)dx. (1) − k=1 1 2 Proof. In face, if a complex-valued function g(x) defined in the interval 0 ≤ x 1, has a continuous derivative g′(x), then we have from integration by parts,≤ 1 1 1 1 g′(x) x = dx = (g(0) + g(1)) g(x)dx. (2) 0 2 2 − 0 (Formula (2) has a simple geometric interpretation in that the right hand side of (2) represents the area of the portion of the (x, y)-plane bounded by the curve y = g(x) and the straight lines y g(0) = (g(1) g(0))x, x = g(0) and x = (g1)). In (2), we now set g(x) = f (x + k−), successively− for k = 1, 2,..., n 1. We then have for k = 1, 2,..., n 1, − − 1 1 1 k+1 f ′(x + k) x dx = ( f (k) + f (k + 1)) f (x)dx. 0 − 2 2 − k Adding up, we obtain formula (1). This is Euler’s summation formula in its simplest form, with the remainder term involving only the first derivative of f (x). It is interesting to notice that if one uses the fact that the Fourier series e2πinx (n = 0, omitted) converges uniformly to x 1/2 in any closed in- − n 2πin − terval (ǫ, 1 ǫ)(0 < ǫ < 1) one obtains from (2) the following useful result, namely, − If f (x) is periodic in x with period 1 and has a continuous derivative, then
1 ∞ 2πinx 2πinx f (x) = e f (x)e− dx. n = 0 −∞ Kronecker’s Limit Formulas 3
Proof of Proposition 1. Let us now specialize f (x) to be the function f (x) = s x log x s x− = e− (log x taking real values for x > 0). The functionx− is evidently 3 continuously differentiable in the interval 1 x n; and applying (1) to the s ≤ ≤ function x− , we get
n 1 1 − s 1 1 1 s s (x + k)− − x dx + (1 + n− ) − − 2 2 k=1 0 n n s s = k− x− dx − k=1 1 1 s n k s 1 n − (if s , 1) = k=1 − −s 1 n 1 − − (3) k− log n(if s = 1). k=1 − Let us now observe that the right-hand side of (3) is an entire function of s.
n σ > s We suppose now that 1. When n tends to infinity 1 x− dx tends to n s 1/(s 1). Further, as n tends to infinity, k− converges to ζ(s). Thus for − k=1 σ > 1, the right hand side of (3) converges to ζ(s) 1/(s 1) as n tends to infinity. − − On the other hand, let the left-hand side of (3) be denoted by ϕn(s). Then ϕ (s) is an entire funciton of s and further, for σ ǫ > 0, n ≥ n ǫ 1 1 ǫ 1 + n− ϕ (s) s k− − + | n | ≤ 2| | 2 k=1
1 ∞ 1 ǫ s k− − + 1. ≤ 2| | k=1
Thus, as n tends to infinity, ϕn(s) converges to a regular function of s in the half-plane σ > 0; this provides the analytic continuation of ζ(s) 1/(s 1) for σ > 0. − − Now the constant a0 in the power-series expansion at s = 1 of 1 ζ(s) = a + a (s 1) + a (s 1)2 + − s 1 0 1 − 2 − − is nothing but lim ϕn(1). In other words, n →∞ 1 1 a0 lim 1 + + + log n n 2 n →∞ − Kronecker’s Limit Formulas 4
= C (Euler’s constant).
Proposition 1 is thus proved. 4 The constant C lies between 0 and 1. It is not known whether it is rational or irrational; very probably, it is irrational. One could determine the constants a1, a2,... also explicitly but this is more complicated. We shall consider now an analogous problem leading to the Kronecker limit formula (Kroneckersche Grenzformle). Instead of the simple linear function x, we consider a positive-definite bi- nary quadratic form Q(u, v) = au2 + 2buv + cv2 in the real variables u and v and with real coefficients a, b, c (we then have a > 0 and ac b2 = d > 0). Associated with Q(u, v), let us define −
∞ s ζQ(s) = (Q(m, n))− , (4) m, n= −∞ where ′ denotes summation over all pairs of integers (m, n), except (0, 0). Q(u , v) being positive-definite, there exists a real number λ > 0, such that Q(u, v) λ(u2 + v2) and it is an immediate consequence that the series (4) converges≥ absolutely for σ > 1 and uniformly in every half-plane defined by σ 1 + ǫ(ǫ > 0). Thus ζQ(s) is a regular function of s, for σ > 1. ≥As in the case of ζ(s) above, we shall obtain an analytic continuation of ζQ(s) regular in the half-plane σ > 1/2, except for a simple pole at s = 1. The constant a0 in the expansion at s = 1 of ζQ(s), viz.
a 1 ζ (s) = − + a + a (s 1) + Q s 1 0 1 − − is given precisely by the Kronecker limit formula. The constant a 1 which is − the residue of ζQ(s) at s = 1 was found by Dirichlet in his investigations on the class-number of quadratic fields. Letus first effect some simplifications. 5 Since Q(u, v) = Q( u, v), we see that − −
∞ s ∞ ∞ s ζQ(s) = 2 (Q(m, 0))− + 2 (Q(m, n))− . (5) m=1 n=1 m = −∞ Further
b 2 b2 Q(u, v) = a u + v + c v2 a − a Kronecker’s Limit Formulas 5
b 2 d b + √ d b √ d = a u + v + v2 = a u + − v u + − − v a a a a = a(u + zv)(u + zv) where arg( √ d) = π/2 and z = (b + √ d)/a = x + iy with y > 0. − − Also, we could, without loss of generality, suppose that d = 1; if Q1(u, v) = √ 2 2 2 2 s/2 (1/ d)(au + 2buv + cv ) = a1u + 2b1uv + c1v , then ζQ1 (s) = d ζQ(s) and a c b2 = 1. The function ds/2 (choosing a fixed branch) is a simple entire 1 1 − 1 function of s and therefore, to study ζQ(s), it is enough to consider ζQ1 (s). 1 1 1 2 Then we have a = y− and Q(u, v) = y− (u + zv)(u + zv) = y− u + zv . Now (5) becomes | |
s ∞ 2s s ∞ ∞ 2s ζ (s) = 2y m− + 2y m + nz − Q | | m=1 n=1 m = −∞ s s ∞ ∞ 2s = 2y ζ(2s) + 2y m + nz − (6) | | n=1 m = −∞ We know from Proposition 1 that ζ(2s) has an analytic continuation in the half-plane σ > 0, regular except for a simple pole at s = 1/2. To obtain an analytic continuation for ζQ(s), we have, therefore, only to investigate the nature of the second term on the right hand side of (6), as a function of s. For this purpose, we need the Poisson summation formula.
Proposition 2. Let f (x) be continuous in ( , ) and let ∞ f (x + m) be −∞ ∞ m= uniformly convergent in 0 x 1. Then −∞ ≤ ≤ ∞ ∞ ∞ 2πikx 2πikξ f (x + m) = e− f (ξ)e dξ. m = k = −∞ −∞ −∞ 6
Proof. The function ϕ(x) = ∞ f (x + m) is continuous in ( , ) and pe- m= −∞ ∞ −∞1 = ϕ ξ 2πikξ ξ riodic in x, of period 1. If ak 0 ( )e d then, by Fej´ er’s´ Theorem, the ∞ 2πikx ∞ (C, 1) sum of ake− is equal to ϕ(x). In particular, if ak converges, k= k= | | then for x in ( −∞, ) −∞ −∞ ∞ 1 ∞ ∞ 2πikx 2πikξ f (x + m) = ϕ(x) = e− ϕ(ξ)e dξ. m = k = 0 −∞ −∞ Kronecker’s Limit Formulas 6
In view of the uniform convergence of ∞ f (x + m)in0 x 1, m= ≤ ≤ −∞ 1 1 ∞ ∞ f (ξ + m)e2πikξdξ = f (ξ + m)e2πikξdξ 0 m = m = 0 −∞ −∞ = ∞ f (ξ)e2πikξdξ. −∞ Thus ∞ ∞ 2πikx ∞ 2πikξ f (x + m) = e− f (ξ)e dξ, (7) m = k = −∞ −∞ −∞ which is the Poisson summation formula. Now we set f (x) = x + iy 2s = z 2s for x in ( , ). Then we see | |− | |− −∞ ∞ ∞ 2s that the series m + z − converges absolutely, uniformly in every interval m= | | N x N, for −∞σ > 1/2. For this purpose, it clearly suffices to consider the interval− ≤ 0≤ x 1, since the series remains unchanged when x is replaces by x + 1. For≤ 0 x≤ 1, ≤ ≤
∞ 2s ∞ 2σ m + z − m + z − | | ≤ | | m = m = −∞ −∞ 2σ 2σ ∞ 2σ 2σ z − + z 1 − + (m− + m− ) ≤ | | | − | m=1
2σ ∞ 2σ 1 < 2y− + 2 m− σ > . 2 m=1 7 Thus, by (7), for σ > 1/2,
∞ 2s ∞ 2πikx ∞ 2s 2πikξ m + z − = e− ξ + iy − e dξ | | | | m = k = −∞ −∞ −∞ ∞ 2πikx ∞ 2 2 s 2πikξ = e− (ξ + y )− e dξ k = −∞ −∞ 1 2s ∞ 2πikx ∞ 2 s 2πikξ = y − e− (1 + ξ )− e dξ, (8) k = −∞ −∞ whenever the series (8) converges. Actually, we shall prove that it converges absolutely for σ > 1/2. For this purpose, let us consider for k > 0, ∞ (1 + −∞ Kronecker’s Limit Formulas 7
2 s 2πikζ ζ )− e dζ as a line integral in the complex ζ-plane; let ζ = ξ + iη. The 2 s function (1 + ζ )− has a logarithmic branch-point at ζ = i and ζ = i. In the complex ζ-plane cut along the η-axis from η = 1 to η = and− from η = 1 to η = , let ABCDEFG denote the contour consisting of∞ the straight line −GA(η = 0−∞, ξ R), the arc AB(ζ = Reiθπ θ π/2), the straight line BC(ξ = 0, 3/2 | η| ≤ R) on the left bank of the cut,≥ the≥ circle ≤ ≤ CDE(ζ = i + (i/2)eiϕ, 0 ϕ 2π), ≤ ≤ the straight line EF on the right bank of the cut (ξ = 0, 3/2 η R) ≤ ≤
Fig. 1
andthe arc FG(ζ = Reiθ, π/2 θ 0). If s lies in a compact set K of the 8 s-plane with σ > 0, then on the arcs≥ AB≥ and FG,
2 s 2πikyζ 2 σ 2πky (1 + ζ )− e 1 + ζ − e− R sin θ | | ≤ | | 2σ 2πky R sin θ = 0(R− e− ), since π arg(1 + ζ2) π on the whole contour. The constants in the O- estimate− depend≤ only on K≤. Thus
π/2 2 s 2πikyζ 2σ 2πky R sin θ (1 + ζ )− e dζ = 0 R− e− Rdθ AB 0 π/2 1 2σ 4kyRθ = 0 R − e− dθ 0 2σ = 0(R− ),
+ ζ2 s 2πikζ ζ and hence AB(1 )− e d tends to zero as R tends to infinity. Similarly + ζ2 s 2πikζ ζ Γ+ FG(1 )− e d also tends to zero as R tends to infinity. If denotes the Kronecker’s Limit Formulas 8 contour BCDEF in the limiting position as R tends to infinity, then for σ > 0, we have, by Cauchy’s Theorem,
∞ 2 s 2πikyζ 2 s 2πikyζ (1 + ζ )− e dζ = (1 + ζ )− e dζ. (9) Γ+ −∞
Let Γ− denote the limiting position of B′C′D′E′F′ (which is the reflection of BCDEF with respect to ξ-axis), as R tends to infinity. We can show similarly that when k < 0,
∞ 2 s 2πikyζ 2 s 2πikyζ (1 + ζ )− e dζ = (1 + ζ )− e dζ. (10) Γ −∞ − We shall now show that the right hand side of (??) defines an entire function of s. In fact, if s = σ+it lies in a compact set K in the s-plane with σ > d(d > + −2 σ 0), then, on Γ we have for a constant c1 depending only on K, 1 + ζ − 2d | | ≤ c1η . Thus for k > 0,
2 s 2πikyζ 2 s 2πkyη (1 + ζ )− e dζ (1 + ζ )− e− dζ Γ+ ≤ | | | | Γ+ 2 σ π t 2πkyη 2d π t 2πkyη 1 + ζ − e | |− dζ c1 η e | |− dζ ≤ Γ+ | | | | ≤ Γ+ | | πky 2d 2πky(η 1 ) πky 2d 2πy(η 1 ) c2e− η e− − 2 dζ c2e− η e− − 2 dζ ≤ Γ+ | | ≤ Γ+ | | πky c e− , (11) ≤ 3 + 2 s 2πkyζ the constants c2 and c3 depending only on K. Hence Γ+ (1 ζ )− e dζ con- 9 verges absolutely and uniformly in K and therefore defines an analytic function of s in any domain contained in K. Since d is an arbitrary positive number, our assertion above is proved. Similar to (11), we have also for k < 0,
2 s 2πikyζ π k y (1 + ζ )− e dζ c3′ e− | | , (12) Γ ≤ − 2 s 2πikyζ and therefore (1 + ζ )− e dζ defines for k < 0, an entire function of s. Γ− In view of (??), (10), (11) and (12), we see that in order to prove the abso- lute convergence of the series (8), for σ > 1/2, we need to prove only that for 2 s σ > 1/2, ∞ (1 + ζ )− dζ exists. But, in fact, for σ > 1/2, −∞ ∞ 2 s ∞ 2 s (1 + ζ )− dζ = 2 (1 + ζ )− dζ 0 −∞ Kronecker’s Limit Formulas 9
∞ s 1 2 = (1 + λ)− λ− 2 dλ (setting ζ = λ) 0 1 1 s 3/2 = − 2 (1 ) − d setting λ = − 1 0 − 1 1 = B , s , 2 − 2 where B(a, b) is Euler’s beta-function. Hence
1 1 2 Γ ∞ 2 s π (s 2 ) (1 + ζ )− dζ = − Γ(s) −∞ Thus, for σ > 1,
1 1 2 Γ ∞ ∞ 2s π (s 2 ) 1 2s 1 2s 2πimx ∞ 2 s 2πimyζ m + z − = − y − + y − ′ e− (1 + ζ )− e dζ, | | Γ(s) m = m = −∞ −∞ −∞ where the accent on Σ indicates the omission of m = 0. Substituting this in (6), 10 we have for σ > 1,
s ζQ(s) = 2y ζ(2s)+ 1 1 2 Γ s ∞ π (s 2 ) 1 2s 1 2s 1 2s 1 2s + 2y − n − y − + n − y − Γ(s) × n=1 ∞ 2πimnx ∞ 2 s 2πimnyζ ′ e− (1 + ζ )− e dζ × m = −∞ −∞ By (??) and (10), therefore, for σ > 1,
1 1 2 Γ s 1 s π (s 2 ) ζ (s) = 2y ζ(2s) + 2y − − ζ(2s 1)+ Q Γ(s) −
1 s ∞ 1 2s ∞ 2πimnx ∞ 2 s 2πimnyζ + 2y − n − e− (1 + ζ )− e dζ; n=1 m = −∞ −∞ i.e.
1 1 2 Γ s 1 s π (s 2 ) 1 s ∞ 1 2s ζ (s) = 2y ζ(2s) + 2y − − ζ(2s 1) + 2y − n − Q Γ(s) − × n=1
∞ 2πimnx 2 s 2πimnyζ e− (1 + ζ )− e dζ+ × Γ+ m=1 Kronecker’s Limit Formulas 10
∞ 2πimnx 2 s 2πimnyζ e (1 + ζ )− e− dζ . (13) Γ m=1 − The double series in (13) converges absolutely, uniformly in every compact subset of the s-plane, in view of the fact that by (11) and (12), it is majorised by ∞ ∞ ∞ 2y1 σ n1 2σ c e πmny + c e πmny , − − 3 − 3′ − n=1 m=1 m=1 whichclearly converges. Since we have seen earlier that the terms of the double 11 series define entire functions of s, it follows by the theorem of Weierstrass that the double series in (13) defines an entire function of s. Formula (13) gives us an analytic continuation of ζQ(s) for σ > 1/2. For, ζ(2s) is regular for σ > 1/2 and from the fact that ζ(s) can be analytically continued into the half-plane σ > 0 such that ζ(s) 1/(s 1) is regular for σ > 0, we see on replacing s by 2s 1 that ζ(2s 1) can− be− continued analytically in the half-plane σ > 1/2 such− that ζ(2s − 1) 1/2(s 1) regular for σ > 1/2 : Γ(s 1 )/Γ(s) is regular for σ > 1/2 and− non-zero− at− s = 1. Hence from − 2 (13), we see that ζQ(s) can be continued analytically in the half-plane σ > 1/2 and its only singularity in this half-plane is a simple pole at s = 1 with residue 1 1 equal to 2 π 2 π 2 1/2 = π. ζ (s) π/( s 1) has therefore a convergent power-series expansion a + Q − − 0 a1(s 1) + at s = 1. We shall now determine the constant a0. Clearly, from (13),−
1 1 2 Γ 1 s π (s 2 ) π a0 = 2yζ(2) + lim 2y − − ζ(2s 1) + s 1 Γ(s) − − s 1 → − ∞ 1 ∞ 2πimnx 2 1 2πimnyζ + 2 e− (1 + ζ )− e dζ+ n + n=1 m=1 Γ ∞ 2πimnx 2 1 2πimnyζ + e (1 + ζ )− e− dζ Γ m=1 − Now from 1 ζ(s) = + C + a (s 1) + s 1 1 − − we have 1 ζ(2s 1) = + C + 2a (s 1) + . − 2(s 1) 1 − − Kronecker’s Limit Formulas 11
Also replacing s by s 1/2 in the formula −
1 1 2s Γ s + Γ(s) = √π 2 − Γ(2s). 2 wehave 12 1 2 2s Γ s Γ(s) = √π2 − Γ(2s 1). − 2 − At s = 1, Γ(s) has the expansion
Γ(s) = 1 + a(s 1) + − and hence Γ2(s) = 1 + 2a(s 1) + − Further at s = 1, Γ(2s 1) = 1 + 2a(s 1) + − − Hence we have at s = 1, by Cauchy multiplication of power-series,
2 2 Γ(2s 1)Γ− (s) = 1 + b(s 1) + − − and thus Γ 1 1 s 1 (s 2 ) 2 2s Γ(2s 1) 2y − π 2 − ζ(2s 1) = 2π(2 √y) − − ζ(2s 1) Γ(s) − Γ2(s) − = 2π(1 + b(s 1)2 + ) − × (1 2 log(2 √y)(s 1) + ) × − − × 1 + C + . × 2(s 1) − In other words, Γ 1 1 s 1 (S 2 ) π lim 2y − π 2 − ζ(2s 1) = 2π(C log(2 √y))/ s 1 Γ(s) − − s 1 − → − 2 1 Now (see Fig. 1), for mn > 0, since (1 + ζ )− has no branch-point at ζ = i or ζ = i, − 2 1 2πimnyζ 2 1 2πimnyζ (1 + ζ )− e dζ = (1 + ζ )− e dζ Γ+ CDE e2πimnyζ = 2πi Residue of at ζ = i 1 + ζ2 Kronecker’s Limit Formulas 12
2πmny = πe− .
Similarly, for mn > 0, 13
e 2πimnyζ (1 + ζ2) 1e 2πimnyζ dζ = 2πi Residue of − at ζ = i − − + 2 Γ− − 1 ζ − 2πmny = πe−
Hence
2 π ∞ 1 ∞ 2πimnx 2πmny a = y + 2π(C log(2 √y)) + 2π e− − + 0 3 − n n=1 m=1
∞ 1 ∞ 2πimnx 2πmny + 2π e − n n=1 m=1 These series converge absolutely and it is practical to sum over n first. Then
2 π ∞ ∞ 1 2πimnz a = y + 2π(C log 2 √y) + 2π e− + 0 3 − n m=1 n=1 ∞ ∞ 1 + 2π e2πimnz n m=1 n=1 2 π ∞ 2πimz = y + 2π(C log(2 √y)) 2π log(1 e− ) 3 − − − − m=1 ∞ 2π log(1 e2πimz) − − m=1
(πi/12)(z z) ∞ 2πimz = 2π(C log(2 √y)) 2π(log e − + log (1 e )+ − − − m=1
∞ 2πimz + log (1 e− )). − m=1 For complex z = x + iy with y > 0, Dedekind defined the η-function,
∞ η(z) = eπiz/12 (1 e2πimz). − m=1
Inthe notation of Dedekind, then, 14 Kronecker’s Limit Formulas 13
a = 2π(C log 2 √y) 2π log η(z)η( z)) 0 − − − = 2π(C log 2 log( √y η(z) 2)). − − | | Thus we have proved Theorem 1. Let z = x + iy, y > 0 and let Q(u, v) = y 1(u + vz) (u + vz). − Then the zeta function ζQ(s) associated with Q(u, v) and defined by ζQ(s) = s ′(Q(m, n))− ,s = σ + it, σ > 1, can be continued analytically into a function m,n of s regular for σ > 1/2 except for a simple pole at s = 1 with residue π and at s = 1, ζQ(s) has the expansion π ζ (s) = 2π(C log 2 log( √y η(z) 2)) + a (s 1) + Q − s 1 − − | | 1 − − This leads us to the interesting first limit formula of Kronecker, viz.
π 2 lim ζQ(s) = 2π(C log 2 log √y η(z) ). s 1 − s 1 − − | | → − We make a few remarks. It is remarkable that the residue of ζQ(s) at s = 1 does not involve a, b, c and this was utilized by Dirichlet in determining the class-number of positive binary quadratic forms. Of course, we had supposed that d = ac b2 = 1; in the general case, the residue of ζ (s) at s = 1 would − Q be π/ √d. This limit formula has several applications; Kronecker himself gave one, namely that of finding solutions of Pell’s (diophantine) equation x2 dy2 = 4, by means of elliptic functions. It has sereval other applications− and it can be generalized in many ways. In the next section, we shall show that as an application of this formula, the transformation-theory of η(z) under the elliptic modular group can be developed.
2 The Dedekind η-function
Let H denote the complex upper halt-plane, namely the set of z = x + iy with y > 0. For z H, Dedekind defined the η-function ∈ ∞ η(z) = eπiz/12 (1 e2πimz). − m=1 15 This infinite product converges absolutely, uniformly in every compact sub- set of H. Thus, as a function of z, η(z) is regular in H. Since none of the factors Kronecker’s Limit Formulas 14 of the convergent infinite product is zero in H, it follows that η(z) , 0, for z in H. If g2 and g3 are the usual constants occurring in Weierstrass’ theory of 12 24 3 2 elliptic functions with the period-pair (1, z), then (2π) η (z) = g2 27g3. The function η(z) is a “modular form of dimension 1/2”. − Let α, β, γ, δ be rational integers such that αδ− βγ = 1. The transformation z z = (αz + β)(γz + δ) 1 takes H onto itself; for,− if z = x + iy , then → ∗ − ∗ ∗ ∗ 1 1 αz + β αz + β y∗ = (z∗ z∗) = 2i − 2i γz + δ − γz + δ
1 2 2 = (z z) γz + δ − = y γz + δ − > 0, (14) 2i − | | | | 1 and clearly z = (δz∗ β)( γz∗+α)− . These transformations are called “modular transformations” and− they− form a group called the “elliptic modular group”. It is known that the elliptic modular group is generated by the simple modular transformations, z z + 1 and z 1/z. In other words, any general modular transformation can→ be obtained by→ iterating − the transformations z z 1 and z 1/z. → ± →We − shall give, in this section, two proofs of the transformation-formula for the behaviour of η(z) under the modular transformation z 1/z. The first proof is a consequence of the Kronecker limit formula proved→in − 1. With z = x + iy H, we associate the positive-definite binary§ quadratic 1 ∈ form Q(u, v) = y− (u + vz)(u + vz). By the Kronecker limit formula for ζQ(s) = ∞ s ′ (Q(m, n))− m,n= −∞ π 2 lim ζQ(s) = 2π(C log 2 log( √y η(z) )). (15) s 1 − s 1 − − | | → − 1 Let z∗ = (αz + β)(γz + δ)− = x∗ + iy∗ be the image of z under a modular trans- formation. Then with z∗, let us associate the positive-definite binary quadratic form 1 Q∗(u, v) = y∗− (u + vz∗)(u + vz∗). Again, by the Kronecker limit formula for 16
∞ = ′ s ζQ∗ (s) (Q∗(m, n))− , m ,n= −∞ π 2 lim ζQ (s) = 2π(C log 2 log y∗ η(z∗) ). (16) s 1 ∗ − s 1 − − | | → − Now, by (??), 1 2 1 2 1 2 y∗− m + nz∗ = y− γz + δ m + n(αz + β)(γz + δ)− | | | | | | Kronecker’s Limit Formulas 15
1 2 = y− m(γz + δ) + n(αz + β) | | 1 2 = y− (mδ + nβ) + (mγ + nα)z . | | Since α, β, γ, δ are integral and αδ βγ = 1, when m, n run over all pairs of rational integers except (0, 0), so do− (mδ + nβ, mγ + nα). Thus, in view of absolute convergence of the series, for σ > 1,
1 2 s ζ (s) = ′(y∗− m + nz∗ )− Q∗ | | m,n 1 2 s = ′(y− (mδ + nβ) + (mγ + nα)z )− | | m,n 1 2 s = ′(y− m + nz )− ; | | m,n i.e. for σ > 1, = δQ(s) ζQ∗ (s)
(ζQ(s) is what is called a “non-analytic modular function of z”).
Since ζQ(s) and ζQ∗ (s) can be analytically continued in the half-plane σ > = 1/2, ζQ(s) ζQ∗ (s) even for σ > 1/2. Hence, from (15) and (16),
2 2 log( √y η(z) ) = log( y η(z∗) ); | | ∗| | i.e. 1 1 η(z) y 4 = η(z∗) y 4 . | | | | By(??), then, we have 17
αz + β η = η(z) γz + δ . (17) γz + δ | || | Let us consider the function η((αz + β)(γz + δ) 1) f (z) = − γz + δη(z) where the branch of √γz + δ is chosen as follows; namely, if γ = 0, then α = δ = 1 and assuming without loss of generality that α = δ = 1, we choose √γz + δ ±= 1. If γ , 0, we might suppose that γ > 0 and then √γz + δ is that branch whose argument always lies between 0 and π for z H. Since η(z) and √γz + δ never vanish in H, (17) means that the funciton f (∈z) which is regular in H is of obsolute value 1. By the maximum modulus principle, f (z) = ǫ, a constant of absolute value 1. We have thus Kronecker’s Limit Formulas 16
Proposition 3. The Dedekind η-function
∞ η(z) = eπiz/12 (1 e2πimz), z H − ∈ m=1
1 satisfies, under a modular transformation z (αz + β) (γz + δ)− the trans- formation formula →
1 η((αz + β)(γz + δ)− ) = ǫ γz + δη(z) with ǫ = ǫ(α,β,γ,δ) and ǫ = 1. | | We shall determine ǫ for the special modular transformations, z z + 1 and z 1/z. → From→ − the very definition of η(z),
η(z + 1) = eπi/12η(z) (18) and so here ǫ = eπi/12. Also if we set z = i in the formula η( 1/z) = ǫ √zη(z), we have η(i) = πi/4 − ǫ √iη(i). Since η(i) , 0, ǫ = 1/ √i = e− . Thusη( 1/z) = e πi/4 √zη(z). We rewrite this as 18 − − 1 z η = η(z), (19) − z i
√z/i being that branch taking the value 1 at z = i. To determine ǫ in the general case, we only observe that any modular transformation is obtained by iteration of the transformations z z 1 and z 1/z. Since every time we apply these transformations we get,→ in± view of → − πi/12 πi/4 th (18) and (19), a factor which is e± or e− , we see that ǫ is a 24 root of unity; ǫ can be determined this way, by a process of “reduction”. The question arises as to whether there exists an explicit expression for ǫ, in terms of α, β, γ and δ. This problem was considered and solved by Dedekind who for this purpose, had to investigate the behaviour of η(z) as z approaches the ‘rational points’ on the real line. This problem has also been considered recently by Rademacher in connection with the so-called ‘Dedekind sums’. We now give a very simple alternative proof of formula (19). Kronecker’s Limit Formulas 17
Fig. 2
Letus consider the integral 19 1 πζ dζ cot πζ cot 8 Ct z ζ where Ct is the contour of the parallelogram in the ζ-plane with vertices at ζ = t, tz, t and tz (t positive and not integral). The integrand− − is a meromorphic function of ζ. It has simple poles at ζ = k 1 πk 1 ± and ζ = kz(k = 1, 2, 3,...) with residues cot and cot πkz respec- ± πk z πk tively, as one can readily notice from the expansion 1 πu cot πu = + πu − 3 valid for 0 < u < 1. Also, the integrand has a pole of the third order at ζ = 0, | 1| with residue 3 (z + 1/z). Let t = h−+ 1/2, h being a positive integer. Then there are no poles of the integrand on Ct and inside Ct there are poles at ζ = 0, k and kz(k = 1, 2,..., h). By Cauchy’s theorem on residues, ± ±
h h 1 ζ dζ 2πi 1 πk 1 1 1 cot πζ cot π = ′ cot + ′ cot πkz z + , 8 C z ζ 8 πk z πk − 3 z h+ 1 k= h k= h 2 − − the accent on Σ indicating the omission of k = 0 from the summation. In other Kronecker’s Limit Formulas 18
words,
1 ζ dζ 1 i h 1 πk cot πζ cot π z + = cot πkz + cot . (20) 8 C z 12 z 2 k z h+ 1 k=1 2 We wish to consider the limit of the relation (20) as h tends to infinity through the sequence of natural numbers. For this purpose, let us first observe that if ζ = ξ + iη, then
eπiζ + e πiζ cot πζ = i − eπiζ e πiζ − − tendsto i if η tends to , and tends to +i, if η tends to . Similarly cot π(ζ/z) 20 tends to− i, when the imaginary∞ part of ζ/z tends to −∞and to i, when the latter tends to − . If O is the origin and ζ tends to infinity∞ along a ray OP with −∞ P on one of the open segments AB, BC, CD, DA of C 1 (see figure), then h+ 2 cot πζ π(ζ/z) tends to the values +1, 1, +1, 1 respectively. Moreover, this process of tending to the respective values− is− uniform, when the ray OP lies between two fixed rays from O, which lie in one of the sectors AOB, BOC, COD or DOA and neither of which coincides with the lines η = 0 or ζ = λz (λ real). This means that the sequence of functions cot π(h + 1/2)ξ cot π(h + 1/2)(ζ/z) (h = 1, 2,...) tends to the limits +1 or 1{ uniformly on any proper } − closed subsegment of a side of C1. Moreover, this sequence of functins is uniformly bounded on C1. This can be seen as follows. Let K j denote the discs ζ j < 1/4, j = 0, 1, 2,.... In view of the periodicity of cot πζ, let us| − confine| ourselves to± the ±strip, 0 ξ 2. We then readily see that in the complement of the set- ≤ ≤ union of the discs K0, K1 and K2 with respect to this vertical strip, the function cot πζ is bounded; for cot πζ tends to πi as η tends to and is bounded away from its poles at ζ = 0, 1 and 2. Indeed therefore,±∞ cot πζ is bounded in the complement of the set union of the discs K , j = 0, 1,... with respect to the j ± entire plane. Since the contours C 1 , for h = 1, 2,... lie in the complement, h+ 2 we have cot πζ α for ζ Ch+ 1 and α independent of h and ζ. By a similar | | ≤ ∈ 2 argument for cot(πζ/z) concerning its poles at kz, we see that for ζ C 1 , h+ 2 h = 1, 2,... cot(πζ/z) β, β independent of±h and ζ. In other words,∈ the sequence of functions| cot| ≤π(h+ 1 )ζ cot π(h+ 1 )(ζ/z) is uniformly bounded on { 2 2 } c1. Now πζ dζ 1 1 ζ dζ cot πζ cot = cot π h + ζ cot π h + , (21) z ζ 2 2 z ζ C + 1 C1 h 2 Kronecker’s Limit Formulas 19 and in view of the foregoing considerations, we can interchange the passage to the limit (as h tends to infinity) and the integration, on the right side of (21). Thus, letting h tend to infinity, we obtain from (20),
πi 1 1 z 1 z 1 dζ z + + − + − 12 z 8 1 − z 1 − z ζ − − i ∞ 1 πk = cot πkz + cot . (22) 2 k z k=1 Now 21 1 z 1 z 1 dζ 1 z z dζ − + − = + . 8 1 − z 1 − z ζ 4 1 1 ζ − − − For z in the ζ-plane cut along the negative ξ-axis, let log z denote the branch that is real on the positive ξ-axis. Then we see
z dζ z dζ = log z and = log z πi. 1 ζ 1 ζ − − Thus from (22),
πi 1 1 πi i ∞ 1 πk z + + log z = cot πkz + cot . (23) 12 z 2 − 2 2 k z k=1 We insert in (23), the expansions
e2πikz ∞ cot πkz = i 1 + 2 = i 1 + 2 e2πimkz − 1 e2πikz − m=1 − and 2πik/z πk e− ∞ 2imπk/z cot = i 1 + 2 = i 1 + 2 e− ; z 1 e 2πik/z − − m=1 then we see that the resulting double series is absolutely convergent. We are therefore justified in summing over k first; we see as a result that the series (23) goes over into
1 η ∞ (1 e2πimz) − z πi 1 log − = log + z + . − (1 e 2πim/z) η(z) 12 z m=1 − − Kronecker’s Limit Formulas 20
Finally 1 η z 1 πi z log − = log z = log , η(z) 2 − 2 i where √z/i is that branch which takes the value 1 at z = i i.e. η( 1/z) = 22 √(z/i)η(z). − We wish to remark that it was only for the sake of simplicity that we took a parallelogram C1 and considered ‘dilatations’ C 1 of C1. It is clear that h+ 2 instead of the parallelogram, we could have taken any other closed contour passing through 1, z, 1, z and through no other poles of the integrand and worked with corresponding− − ‘dilatations’ of the same.
3 The second limit formula of Kronecker
We Shall consider now some problems more general than the ones concerning the analytic continuation of the function ζQ(s) in the half-plane σ > 1/2 and the Kronecker limit formula for ζQ(s). In the first place, we ask whether it is possible to continue ζQ(s) analytically in a larger half-plane; this question shall be postponed for the present. We shall show later ( 5) that ζQ(s) has an analytic continuation in the entire plane which is regular except§ for the simple pole at s = 1 and satisfies a functional equation similar to that of the Riemann ζ-function. ∞ s Instead of ζQ(s) = ′ (Q(m, n))− , one might also consider m,n= −∞
∞ s (Q(m + , n + ν))− m, n= −∞ where and ν are non-zero real numbers which are not both integral and where m and n run independently from to + . Or, instead of the positive-definite binary quadratic form Q(u, v), we−∞ might take∞ a positive-definite quadratic form in more than two variables. Let, in fact, S = (s ), 1 i, j p be the matrix of i j ≤ ≤ a positive-definite quadratic form si j xi x j in p variables. Let us consider 1 i, j p ≤ ≤
∞ = ζS (S ) S i j xi x j x1,..., xp= i, j −∞ x1,..., xp running over all p-tuples of integers, except (0,..., 0). The func- tionζS (S ), known as the Epstein zeta-function, is regular for σ?p/2. It was 23 Kronecker’s Limit Formulas 21
shown by Epstein that ζS (S ) can be continued analytically in the entire s-plane, into a function regular except for a simple pole at S = p/2 and satisfying the functional equation
s 1 (p/2 s) p p π− Γ(S )ζS (S ) = S − 2 π− − Γ s ζS 1 s | | 2 − − 2 − Epstein also obtained for ζS (s), an analogue of the first limit formula of Kro- necker but this naturally involves more complicated functions than η(z). We shall now consider a generalization of ζQ(s) which shall lead us to the second limit formula of Kronecker. namely, let u and v be independent real parameters. Let, as before, Q(m, n) = am2 + 2bmn + cn2 be a positive-definite binary quadratic form and let ac b2 = 1, without loss of generality. We then define, for σ > 1, −
′ 2πi(mu+nv) s ζQ(s, u, v) = e (Q(m, n))− (24) m,n m, n running over all ordered pairs of integers except (0, 0). The series con- verges absolutely for σ > 1 and converges uniformly in every half-plane σ 1 + ǫ(ǫ > 0). Thus ζ (s, u, v) is a regular function of s for σ > 1. If ≥ Q u and v are both integers, then ζQ(s, u, v) = ζQ(s) which has been already con- sidered. We shall, suppose, in the following, that u and v are not both integers. ( ) ∗ Furthermore, on account of the periodicity of ζQ(s, u, v) in u and v, we can clearly suppose that 0 u, v < 1. The condition ( ) then means that at least one of the two conditions≤ 0 < u < 1, 0 < v < 1 is satisfied.∗ We might, without loss of generality, suppose that 0 < u < 1. For, otherwise, if u = 0, then necessarily 0 < v < 1. In this case, let us take the positive-definite binary 2 2 quadratic form Q1(m, n) = cm + 2bmn + an ; we see then that, for σ > 1,
′ 2πi(mv+nu) s ζQ(s, u, v) = e (Q1(m, n))− . m,n And here, since 0 < v < 1, v shall play the role of u in (24). Weshall prove now that ζQ(s, u, v) can be continued analytically into a func- 24 tion regular in the half-plane σ > 1/2 and then determine its value at s = 1; this will lead us to the second limit formula of Kronecker. For σ > 1, by our earlier notation,
s ∞ 2πimu 2s s ∞ 2πinv ∞ 2πimu 2s ζ (s, u, v) = y e m − + y e e m + nz − . (25) Q | | | | m = n = m = −∞ −∞ −∞ Kronecker’s Limit Formulas 22
The first series in (25) converges absolutely for σ > 1/2 and uniformly in every compact subset of this half-plane. Hence it defines a regular function of s for σ > 1/2. The double series in (25) again defines a regular function of s in the half-plane σ > 1. To obtain its analytic continuation in a larger half- plane, we shall carry out the summation of the double series in the following particular way. Namely, we consider the finite partial sums
2πi(mu+nv) 2s ′ e m + nz − . | | n1 n n2 m1 m m2 ≤ ≤ ≤ ≤
We shall show that when m1 and n1 tend to independently and m2 and n2 tend to independently, then these partial sums−∞ which are entire functions of s converge∞ uniformly in every compact subset of the half-plane σ > 1/2. The proof is based on the method of partial summation in Abel’s Theorem. e2πiku Let us define for any integer k, c = . (Recall that 0 < u < 1). k e2πiu 1 2πimu Then e = c + c . Moreover, − m 1 − m 2πiu 1 c 1 e − = α, | m| ≤ | − | where α is independent of m. Now
2πi(mu+nv) 2s ′ e m + nz − | | n1 n n2 m1 m m2 ≤ ≤ ≤ ≤ n2 m2 c + c = e2πinv m 1 − m m + nz 2s n =n1 m =m1 | | n2 m2 2πinv 2s 2s = e c ( m 1 + nz − m + nz − )+ m | − | − | | n =n1 m= m1+1 + c m + nz 2s c m + nz 2s . (26) m2+1 2 − m1 1 − | | − | | Let us also observe that 25
2s 2σ 2σ c + m + nz − α n − y− m2 1| 2 | ≤ | | and 2s 2σ 2σ c m + nz − α n − y− . m1 | 1 | ≤ | | Further 2s 2s m 1 + nz − m + nz − || − | − | | | Kronecker’s Limit Formulas 23
m 2 2 2 s 1 2s (( + nx) + n y )− − ( + nx)d − m 1 − m 2 2 2 (σ+ 1 ) 2 s (( + nx) + n y )− 2 du. ≤ | | m 1 − Thus (26) is majorised by
n2 m2 m n2 2 2 2 (σ+ 1 ) 2σ 2σ 2 s α ′ (( + nx) + n y )− 2 d + 2αy− ′ n − . (27) | | m 1 | | n=n1 m= m1+1 − n=n1
In (27), we sum m from to + instead of summing from m1 + 1 to m2 and then we see that (26) has−∞ the majorant∞
n2 n2 ∞ 2 2 2 (σ+ 1 ) 2σ 2σ 2α s ′ (( + nx) + n y )− 2 d + 2αy− n − . | | | | n=n1 −∞ n =n1 Now 26
∞ 2 2 2 (σ+ 1 ) ∞ 2 2 2 (σ+ 1 ) (( + nx) + n y )− 2 d = ( + n y )− 2 d −∞ −∞ 2σ 2σ ∞ 2 (σ+ 1 ) = n − y− (1 + )− 2 d . (28) | | −∞ Since the integral in (28) converges for σ > 1/2, we see that (26) is majorised n2 2σ by β n − , β depending only on the compact set in which s lies in the n=n1 | | half-plane σ > 1/2 and on α and y. If, in (26), we now carry out the passage of m1 and n1 to and of m2 and n to + independently, then we see that the double series in−∞ (25) summed 2 ∞ ∞ 2σ in this particular manner, is majorised by the series 2β n− . Hence it con- n=1 verges uniformly when s lies in a compact set in the half-plane σ > 1/2. Since each partial sum (26) is an entire function of s, we see, by Weierstrass’ Theo- rem that the double series in (25) converges uniformly to a function which is regular for σ > 1/2 and which provides the necessary analytic continuation in this half-plane. Thus under the assumption that u and v are not both integers, ζQ(s, u, v) has an analytic continuation which is regular for σ > 1/2. We shall now determine its value at s = 1. We wish to make it explicit that hereafter we shall make only the assumption ( ) and not the specific assumption 0 < u < 1. From the convergence∗ proved above, it follows that
z z ∞ e2πimu ζ (1, u, v) = − ′ + Q 2i m 2 m = −∞ | | Kronecker’s Limit Formulas 24
z z ∞ ∞ e2πimu + − ′ e2πinv (29) 2i (m + nz)(m + nz) n = m = −∞ −∞ where the accents on Σ have the usual meaning. e2πimu e2πimu ∞ ∞ 27 In order to sum 2 , we observe that since the series m= m m= m converges uniformly to −∞ 2π(1| /|2 u) in any closed interval. 0 < p u −∞ q < 1, we have − ≤ ≤
u u ∞ e2πimu ∞ e2πimu du = lim ′ du m ǫ 0 m 0 m = → ǫ m = −∞ −∞ u ∞ e2πimu = lim ′ du, ǫ 0 m → m = ǫ −∞ i.e. 1 ∞ e2πimu ∞ e2πimu 2πi(u u2) = lim ′ ′ , − 2πi ǫ 0 m2 − m2 → m = m = −∞ −∞ i.e. ∞ e2πimu π2 2π2(u2 u) = ′ , − − m2 − 3 m = −∞ 2πimu ∞ e since ′ 2 is continuous in u. In other words, m= m −∞ ∞ e2πimu 1 ′ = 2π2 u2 u + . m2 − 6 m = −∞ We shall sum the double series in (29) more generally with z replaced by r, where r H. Leter, we shall set r = z. We remark that with slight changes, − ∈ 2πi(mu+nv) 2s our arguments concerning the series ′ e m + nz − above will also 2πi(mu+nv) | | s go through for the double series ′ e ((m + nz) (m + nr)) . Now for − summing the series
z r ∞ ∞ e2πimu − ′ e2πinv 2i (m + nz)(m + nr) n = m = −∞ −∞ 1 ∞ e2πinv ∞ 1 1 = ′ e2πimu . (30) 2i n m + nr − m + nz n = m = −∞ −∞ we could apply the Poisson summation formula with respect to m to the inner sum on the left side of (30), but we shall adopt a different method here. Kronecker’s Limit Formulas 25
Letus define, for complex z = x + iy, 28
1 ∞ ∞ e2πimu e 2πimu f (z) = + + + − z z + m z m m=1 m=1 − and consider f (z) in the strip 1/2 x 1/2. It is not hard to see that whenever z lies in a compact set− not containing≤ ≤ z = 0 in this strip, the series converges uniformly and hence f (z) is regular in this strip except at z = 0, where it has a simple pole with residue 1. Moreover, one can show that f (z) 1/z is bounded in this strip. Further −
2πiu f (z + 1) = e− f (z). Consider now the function e 2πiuz g(z) = 2πi − ; 1 e 2πiz − − g(z) is again regular in this strip except at z = 0 where it has a simple pole with residue 1. Moreover g(z) = 1/z is bounded in this strip; as a matter of fact, if 0 < u < 1, g(z) tends to 0 if y tends to + or and if u = 0, then g(z) tends ∞ −∞ 2πiu to 0 or 2πi according as y tends to + or . Further g(z + 1) = e− g(z). As a consequence, the function ∞f (z) −∞g(z) is regular and bounded in this 2πiu− strip and since f (z + 1) g(z + 1) = e− ( f (z) g(z)), it is regular in the whole z-plane and bounded, too.− By Liouville’s theorem,− f (z) g(z) = c, a constant. It can be seen that, if u , 0, c = 0 and if u = 0, c = π−i, by using the series expansion for the function π cot πz. In any case, f (nr) f (nz) = g(nr) g(nz). (31) − − In view of the convergence-process of the series (29) described above, we can rewrite the double series (30) as
1 ∞ e2πinv 1 1 ∞ 1 1 1 1 ′ + e2πimu + 2i n nr − nz m + nr m + nr − m + nz − m + hz n = m=1 −∞ − − 2πinv 2πinv 2πiunr 2πiunz 1 ∞ e ∞ e e e = ′ ( f (nr) f (nz)) = π ′ − − , 2i n − n 1 e 2πinr − 1 e 2πinz n = n = − − − − −∞ −∞ (32) by (31). Now, we insert in (32), the expansions 29
e2πimnz, (n > 0) 2πinz 1 = m∞=1 (1 e− )− − 2πmnz − ∞ e− , (n < 0) m=0 Kronecker’s Limit Formulas 26
e 2πimnr, (n > 0) 2πinr 1 = m∞=0 − (1 e− )− 2πimnr − ∞ e , (n < 0) − m=1 valid for z, r H and then the series (32) goes over into − ∈
∞ 1 2πin(v ur) 2πin(v uz) ∞ 2πin(v uz mz) π e − + e− − + (e− − − + n n=1 m=1 + 2πin(v ur mr) + 2πin(v uz+m′z) + 2πin(v ur+mr) e − − e − e− − ) (33) If 0 < u < 1, then this series converges absolutely, since z, r H. Carrying out the summation over n first, we see that this double series− is∈ equal to
∞ 2πi(v uz mz) 2πi(v ur mr) π log (1 e− − − )(1 e − − ) − − − × m=0 ∞ 2πi(v uz+mz) 2πi(v ur+mr) (1 e − )(1 e− − ) . (34) × − − m=1 If u = 0, then necessarily 0 < v < 1 and in this case,
∞ 1 2πinv 2πinv 2πiv 2πiv π (e + e− ) = π log(1 e )(1 e− ). n − − − n=1 Therest of the series (33) is absolutely convergent and summing over n first, 30 we see that the value of the series (33) is given by the expression (34). Let us now set v uz = w, v ur = q. Then replacing z by r in (29), we have finally, for 0 u−, v < 1 and u−and v not simultaneously zero, ≤ + z r ∞ e2πi(mu nv) − 2i (m + nz)(m + nr) m, n= −∞ 2 2 1 ∞ 2πi(w mz) = π i(z r) u u + π log (1 e− − ) − − − 6 − − × m=0 2πi(q mr) ∞ 2πi(w+mz) 2πi(q+mr) (1 e − ) (1 e )(1 e− ) (35) × − × − − m=1 For z H and arbitrary complex w, the elliptic theta-functionϑ1(w, z) is defined by∈ the infinite series,
∞ πi(n+ 1 )2z+2πi(n+ 1 )(w 1 ) ϑ1(w, z) = e 2 2 − 2 . n = −∞ Kronecker’s Limit Formulas 27
It is clear that this series converges absolutely, uniformly when z lies in a com- pact set in H and w in a compact set of the w-plane. Hence ϑ1(w, z) is an analytic function of z and w for z H and complex w. It is proved in the ∈ theory of the elliptic theta-functions that ϑ1(w, z) can also be expressed as an absolutely convergent infinite product, namely,
πi(z/4) πiw πiw ϑ (w, z) = ie (e e− ) 1 − − × ∞ 2πi(w+mz) 2πi(w mz) 2πimz (1 e )(1 e− − )(1 e ). (36) × − − − m=1
We shall identify these two forms of ϑ1(w, z) later on. For the present, we shall consider ϑ1(w, z) as defined by the infinite product. Now a simple rearrangement of the expression (35) gives
z r ∞ e2πi(mu+nv) − ′ 2i (m + nz)(m + nr) m ,n= −∞ 1 1 = π2i(z r) u2 u + + π2i (w q) + (z r) − − − 6 − 6 − − ϑ (w, q)ϑ (q, r) π log 1 1 − − η(z)η( r) − (w q)2 ϑ (w, q)ϑ (q, r) = π2i − π log 1 1 − − z r − η(z)η( r) −2πi(mu+nv) − z r e ϑ1(w, z)ϑ1(q, r) πi((w q)2/(z r)) − ′ = log − e − − . (37) 2πi (m + nz)(m + nr) η(z)η( r) − m,n −
Setting r = z again, we have q = w and then 31
(w w)2 ϑ (w, z)ϑ (w, z) ζ (1, u, v) = π2i − π log 1 1 − Q − z z − η(z) 2 − | | One can see easily from (36) that ϑ (w, z) = ϑ (w, z). Hence 1 1 − ϑ (w, z) ζ (1, u, v) = π2iu2(z r) 2π log 1 , (38) Q − − − η(z) for 0 u, v < 1 and u and v not both zero. We≤ contend that formula (38) is valid for all u and v not both integral. Since ζQ(1, u + 1, v) = ζQ(1, u, v + 1) = ζQ(1, u, v) and w goes over into w + 1 and Kronecker’s Limit Formulas 28 w z respectively under the transformations v v + 1 and u u + 1, it would su−ffice for this purpose to prove that → → ϑ (w + 1, z) ζ (1, u, v) = π2iu2(z r) 2π log 1 Q − − − η(z) and 2 2 ϑ1(w z, z) ζQ(1, u, v) = π i(u + 1) (z r) 2π log − . − − − η(z) The first assertion is an immediate consequence of the fact that ϑ 1(w + 1, z) = ϑ (w, z). To prove the second, we observe that from (36), we have − 1 ϑ (w z, z) (eπi(w z) e πi(w z))(1 e2πiw) 1 = − − − − πiw − πiw 2−πi(w z) ϑ1(w, z) (e e− )(1 e− − ) +2πiw −πiz − = e − and hence 32 ϑ (w z, z) ϑ (w, z) 2π log 1 − log 1 = +π2i (2(w w) (z z) η(z) − η(z) { − − − } = π2i(z z)((u + 1)2 u2). − − − Thus the second assertion is proved and we have
Theorem 2. If u and v are real and not both integral, then the Epstein zeta- function ζQ(s, u, v) defined by (24) for σ > 1, can be continued analytically into a function regular for σ > 1/2 and its value at s = 1 is given by (38).
We are finally led to the following formula: viz. for all u and v not both integral,
z z e2πi(mu+nv) (w w)2 ϑ (w, z) − ′ = π2i − 2π log 1 . 2i m + nz 2 − z z − η(z) m,n | | − This is the second limit formula of Kronecker. We may rewrite it as
2πi(mu+nv) 2 z z ′ e ϑ (v uz, z) 2 − = log 1 − eπizu . (39) 2πi m + nz 2 η(z) − | | If u and v are both integral, then v uz is a zero of ϑ1(w , z) and then both sides are infinite. − Kronecker’s Limit Formulas 29
4 The elliptic theta-function ϑ1(w, z)
For z = x + iy H and arbitrary complex w, the elliptic theta-function ϑ (w, z) ∈ 1 was defined in 3, by the infinite product (36). ϑ1(w, z) is a regular function of z and w, for§z in H and arbitrary complex w. We shall now develop the transformation-theory of ϑ1(w, z) under modular transformations, as a conse- quence of the second limit formula of Kronecker. Letu and v be arbitrary real numbers which are not simultaneously integral. 33 Let z z = (αz + β)(γz + δ) 1 = x + iy be a modular transformation. Then → ∗ − ∗ ∗ u∗ = δu + γv and v∗ = βu + αv are again real numbers, which are not both integral. For σ > 1, we have
2πi(mu∗+nv∗) 2πi((mδ+nβ)u+(mγ+nα)v) s ∞ e s e y∗ = y ′ m + nz 2s (mδ + nβ) + (mγ + nα)z 2s m, n= ∗ m,n −∞ | | | | e2πi(mu+nv) = ys ′ . m + nz 2s m,n | | From 3, we know that both sides, as functions of s, have analytic continua- tions regula§ in the half-plane σ > 1/2 and hence the equality of the two sides is valid even for σ > 1/2. In particular, for s = 1,
e2πi(mu∗+nv∗) e2πi(mu+nv) y∗ ′ = y ′ . (40) m + nz 2 m + nz 2 m,n | ∗| m,n | | By Kronecker’s second limit formula,
2 2 2 (w∗ w∗) ϑ1(w∗, z∗) (w w) ϑ1(w, z) π i − 2π log = π∗i − 2π log − z∗ z∗ − η(z∗) − z z − η(z) − − where w = v uz and w = v u z = w/(γz + δ). Therefore, − ∗ ∗ − ∗ ∗ 2 2 ϑ (w , z ) ϑ (w, z) πi (ww) πi (w w∗) log 1 ∗ ∗ log 1 = ∗ − η(z ) − η(z) 2 z z − 2 z z∗ ∗ − ∗ − 2 2 π iγ w w 2 = = log e(πiγw )/(γz+δ) 2 γz + δ − γz + δ Thus w αz+β ϑ1 γz+δ , γz+δ ϑ (w, z) 2 = 1 e(πiγw )/(γz+δ) ω, (41) αz+β η(z) η γz+δ where ω = 1 and ω might depend on every one of w, z, α, β, γ and δ. Butfrom 34 | | Kronecker’s Limit Formulas 30
(41), we observe that ω is a regular function of z in H and w, except possibly when w is of the form m + nz with integral m and n. But since ω = 1, we see, by applying the maximum modulus principle to ω as a function| of| w and of z, that ω is independent of w and z. Hence ω = ω(α,β,γ,δ). Now from the fact that
ϑ (w, z) = 2πη3(z)w + 1 we note that ϑ1′ (0, z), the value of the derivative of ϑ1(w, z) with respect to w at w = 0, is equal to 2πη3(z). Differentiating the relation (41) with respect to w at w = 0, we get αz+β 1 ϑ1′ 0, γz+δ ϑ′ (0, z) = 1 ω, γz + δ αz+β η(z) η γz+δ i.e. 1 2 αz + β 2 (γz + δ)− η = ωη (z). γz + δ We have thus rediscovered the formula αz + β η = ǫ γz + δη(z) γz + δ with ǫ being a 24th root of unity depending only on α, β, γ and δ and ω = ǫ2. Rewriting (41), we obtain Proposition 4. The elliptic theta-function has the transformation formula
w αz + β 3 (πiγw2)/(γz+δ) ϑ1 , = ǫ γz + δe ϑ1(w, z) γz + δ γz + δ 1 3 th under a modular substitution z (αz + β)(γz + δ)− , ǫ being an 8 root of unity depending only on α, β, γ and→ δ. From the transformation-theory of η(z) we know that corresponding to the modular transformations z z+1 and z 1/z, ǫ3 has respectively the values πi/4 3πi/4 → → − e and e− . Hence
πi/4 ϑ1(w, z + 1) = e ϑ1(w, z) (42) w 1 πiw2/z 1 z ϑ1 , = e ϑ1(w, z). z − z i i Once again using the fact that the modular transformations z z + 1 and 35 → Kronecker’s Limit Formulas 31 z 1/z generate the elliptic modular group, we can determine ǫ3, with the help→ of − (42), by a process of ‘reduction’. Hermite found an explicit expression for ǫ3, involving the well-known Gaussian sums. We now wish to make another remark which is of function-theoretic signif- icance and which is again, a consequence of Kronecker’s second limit formula. Let us define for z H and real u and v (not both integral), the function ∈ ϑ (v uz, z) 2 f (z, u, v) = log 1 − eπizu η(z) choosing a fixed branch of the logarithm in H. Then from (40) and (39),
ϑ1(v∗ u∗z∗, z∗) 2 ϑ1(v uz, z) 2 log − eπiz∗u∗ = log − eπizu . η(z∗) η(z) Hence αz + β f ,δu + γv, βu + αv = f (z, u, v) + iλ, γz + δ where λ is real and might depend on z, u, v, α, β, γ and δ. But since λ is a regular function of z, whose imaginary part is zero in H, λ is independent of z. Thus αz + β f ,δu + γv, βu + αv = f (z, u, v) + iλ(u, v,α,β,γ,δ). (43) γz + δ Let now q > 1, be a fixed integer and let u and v be proper rational fractions with reduced denominator q i.e. u = a/q, v = a/q, (a, b, q) = 1 and 0 u, v < 1. Since αδ βγ = 1, we have again. ≤ − (δa + γb, βa + αb, q) = 1.
ϑ1(v uz, z) 2 Nowfrom (39), we see that log − eπizu is invariant under the 36 η(z) transformations u u + 1 and v v + 1. Hence f (z , u, v) picks up a purely imaginary additive→ constant under→ these transformations. If now, (δa + γb)/q ≡ (a∗/q)( mod 1) and (βa + αb)/q (b∗/q)( mod 1) where a∗/q, b∗/q are proper rational fractions with reduced≡ denominator q, then
αz + β δa + γb βa + γb αz + β a∗ b∗ f , , = f , , + iλ′ γz + δ q q γz + δ q q where λ′ is real and depends on a, b, α, β, γ and δ. Writing f (z, a/q, b/q) as fa,b(z), we get from (43), αz + β fa ,b = fa,b(z) + iλ∗(a, b,α,β,γ,δ) (44) ∗ ∗ γz + δ Kronecker’s Limit Formulas 32
λ∗(a, b,α,β,γ,δ) being a real constant depending on a, b, α, β, γ and δ. The number of reduced fractions a/q, b/q with reduced denominator q is given by v(q) = q2 (1 1/p2)p running through all prime divisors of q. Corre- p q − | sponding to each pair a/q, b/q, we have a function fa,b(z) and the relation (44) asserts that when we apply a modular transformation on z, these functions are permuted among themselves, except for a purely imaginary additive constant. 1 α β 1 0 The modular transformations z to(αz + β)(γz + δ)− for which γ δ 0 1 ( mod q), form a subgroup M (q) of the elliptic modular group called the≡prin- cipal congruence subgroup of level (stufe) q. The group M (q) has index qv(q) in the elliptic modular group and acts dis- continuously on H. We can construct in the usual way, a fundamental domain for M (q) in H. This can be made into a compact Riemann surface by identify- ing the points on the boundary which are ‘equivalent’ under M (q) and ‘adding the cusps’. Complex-valued functions f (z) which are meromorphic in H and invariant under the modular transformations in M (q) and have at most a pole in the ‘local uniformizer’ at the ‘cusps’ of the fundamental domain are called modular functions of level q. They form an algebraic function field of one variable (over the field of complex numbers) which coincides with the field of meromorphic functions on the Riemann surface. Inthe case when the transformation z (αz + β)(γz + δ) 1 is in M (q), then 37 → − referring to (44), we have a∗ = a, b∗ = b and
αz + β fa,b = fa,b(z) + iλ∗(a, b,α,β,γ,δ). (45) γz + δ
The function fa,b(z) is regular everywhere in H. To examine its singularities (in the local uniformizers) at the ‘cusps’ on the Riemann surface, it suffices in virtue of (44) to consider the singularity of fa,b(z) at the ‘cusp at infinity’. It can be seen that at the ‘cusp at infinity’, fa,b(z) has the singularity given by πiz(a2q 2 aq 1+1/6) log e − − − . In view of (45), then fa,b(z) is an abelian integral on the Riemann surface associated with M (q) and the quantities λ∗(a, b,α,β,γ,δ) are ‘periods’. Since the expression u2 u + 1/6 does not vanish for rational u, − fa,b(z) is an abelian integral strictly of the third kind. It might be an interesting problem to determine the ‘periods’, λ∗(a, b,α,β,γ,δ). We had defined the function ϑ1(w, z) by the infinite product (36); now, we shall identify it with the infinite series expansion by which it is usually defined. For this purpose, we define for z H and complex w, ∈
∞ πi(n+ 1 )2z+2πi(n+ 1 )(w 1 ) f (w, z) = e 2 2 − 2 . (46) n = −∞ Kronecker’s Limit Formulas 33
Clearly f (w, z) is a regular function of w and z. Moreover, it is easy to see that
f (w + 1, z) = f (w, z). (47) − Also replacing n by n + 1 on the right-hand side of (46), we have
∞ πiz(n+1+ 1 )2+2πi(n+1+ 1 )(w 1 ) f (w, z) = e 2 2 − 2 n = −∞ = e2πiw+πiz f (w + z, z), − i.e. 2πiw πiz f (w + z, z) = e− − f (w, z). (48) − We know already that ϑ (w + 1, z) = ϑ (w, z) 1 − 1 and 2πiw πiz ϑ (w + z, z) = e− − ϑ (w, z). 1 − 1 f (w, z) Hence,for fixed z H, the function h(w, z) = is a doubly periodic 38 ∈ ϑ1(w, z) function of w with independent periods 1 and z and is regular in w except possibly when w = m + nz with integral m and n, since ϑ1(w, z) has simple zeros at these points. But f (w, z) also has zeros at these points, since, replacing n by n 1 on the right-hand side of (46), we have f ( w, z) = f (w, z) and hence−f (0−, z) = 0 and from (47) and (48), f (m + nz, z) =− 0 for integral− m and n. Thus h(w, z) is a doubly periodic entire function of w and by Liouville’s theorem, it is independent of w, i.e. h(w, z) = h(z). The function h(z) is a regular function of z in H and to determine its be- haviour under the modular transformations, it suffices to consider its behaviour 1 under the transformations z z + 1 and z z− . We can show easily that h(z + 1) = h(z). Moreover, by→ (42), → −
w 1 πiw2/z 1 z ϑ1 , = e ϑ1(w, z). z − z i i Let us assume, for the present, that we have proved the formula
w 1 2 1 z f , = eπiw .z f (w, z). (49) z − z i i It will then follow that h( 1/z) = h(z). Hence h(z) is a modular function and it is indeed regular everywhere− in H. Kronecker’s Limit Formulas 34
Now, a fundamental domain of the elliptic modular group in H is given by the set of z = x + iy H for which z 1 and 1/2 x 1/2. Let z tend to infinity in this fundamental∈ domain.| | Then ≥ it is− easy to≤ see≤ that the functions πiz/4 πiz/4 πiw πiw ϑ1(w, z)e− and f (w, z)e− tend to the same limit i(e e− ). Hence h(z) 1 tends to zero as z tends to infinity in the fundamental− − domain. But then,− being regular in H and invariant under modular transformations, it attains its maximum at some point in H. By the maximum modulus principle, h(z) 1 is a constant and since h(z) 1 tends to zero as z tends to infinity, h(z) = 1.− In − other words, ϑ1(w, z) = f (w, z). To complete the proof, all we need to do is to prove (49). We may first rewrite f (w, z) as
πi(w 1 )2/z ∞ πiz(n+ 1 +w/z 1/2z)2 f (w, z) = e− − 2 e 2 − n = −∞ 39 πξx2 ∞ Let ξ > 0 and let g(x) = e− , for < x < . Then the series g(x+ −∞ ∞ n= n) converges uniformly in the interval 0 x 1 and by the Poisson summation −∞ formula (7), ≤ ≤
∞ πξ(x+n)2 ∞ 2πinx ∞ πξx2 2πinx e− = e e− − dx, (50) n = n = −∞ −∞ −∞ (provided the series on the right-hand side converges). Now
πn2/ξ ∞ πξx2 2πinx e− ∞ π(x+in/ √ξ)2 e− − dx = e− dx, ( ξ > 0). √ξ −∞ −∞ By the Cauchy integral theorem, one can show that
∞ π(x+in/ √ξ)2 ∞ πx2 e− dx = e− dx = γ(say). −∞ −∞ It is now obvious that the series on the right-hand side of (50) converges abso- lutely. Hence
∞ πξ(x+n)2 γ ∞ πn2/ξ+2πinx e− = e− . (51) n = √ξ n = −∞ −∞ Setting ξ = 1 and x = 0 in (51), we have immediately γ = 1. Both sides of (51) represent, for complex x, analytic functions of x and since they are equal for real x, formula (51) is true even for complex x. Moreover, both sides of (51) represent analytic functions of ξ, for complex ξ with Re ξ > 0. Since they Kronecker’s Limit Formulas 35 coincide for real ξ > 0, we see again that (51) is valid for all complex ξ with Re ξ > 0. We have thus the theta-transformation formula
∞ πξ(n+v)2 1 ∞ πn2/ξ+2πinv e− = e− (52) n = √ξ n = −∞ −∞ valid for all complex v and complex ξ with Re ξ > 0, √ξ denoting that branch which is positive for real ξ > 0. Settingξ = iz and v = (1/2) + (w/z) (1/2z) in (53), we have 40 − − πi(w 1 )2/z e− − 2 ∞ πin2/z+2πin((1/2)+(w/z) (1/2z)) f (w, z) = e− − . z n= i −∞ Now
πin2/z+2πin((1/2)+(w/z) (1/2z)) πi(w 1 )2/z e− − − − 2 πi(n+ 1 )2/z+2πi(n+ 1 )(w/z 1/2) πiw2/z+πi/2 = e− 2 2 − − .
Thus ie πiw2/z w 1 f (w, z) = − f , , z z − z i which is precisely (49). Thus
Proposition 5. For the elliptic theta-function ϑ1(w, z) defined by the infinite product (36), we have the series-expansion
∞ πi(n+ 1 )2z+2πi(n+ 1 )(w 1 ) ϑ1(w, z) = e 2 2 − 2 . n = −∞ Using the method employed above, we shall also prove Proposition 6. The Dedekind η-function η(z) has the infinite series expansion
πiz/12 ∞ λ πiz(3λ2 λ) η(z) = e ( 1) e − − λ = −∞ Proof. Let us define for z H, ∈
πiz/12 ∞ πiz(3λ2 λ) f (z) = e ( 1)∗e − . − λ = −∞ Kronecker’s Limit Formulas 36
The function f (z) is regular in H and moreover, it is clear that f (z + 1) = f (z)eπi/12. Again,
πiz/12 ∞ 2πiz(λ 1 (1 1/z))2 πiz(1 1/z)2/12 f (z) = e e − 6 − − − λ = −∞ πi/12z πi/6 ∞ 3πiz(λ 1 (1 1/z))2 = e− e e − 6 − λ = −∞ πi/12z πi/6 i ∞ πiλ2/3z πiλ(1 1/z)/3 = e− e e− − − , 3z λ = −∞ using formula (52). Here √i/3z denotes that branch which is positive for z = i. 41 ∞ πi/3zλ2 πiλ/3(1 1/z) Now we split up e− − − as g0(z) + g1(z) + g2(z), where for λ= k = 0, 1, 2, −∞ ∞ πi(3 +k)2/3z πi(3 +k)(1 1/z)/3 gk(z) = e− − − . n = −∞ Clearly,
∞ πi(3 2 )/z g (z) = ( 1) e− − , 0 − = −∞ πi/3 g1(z) = e− g0(z),
∞ πi(9 2+12 +4)/3z πi(3 +2)(1 1/z)/3 g2(z) = e− − − = −∞ 2πi/3z 2πi/3 ∞ 3πi ( +1)/z = e− − ( 1) e− . − = −∞ Now ∞ 3πi ( +1)/z ∞ 3πi ( +1)/z ( 1) e− = ( 1) e− , − − − = = −∞ −∞ on replacing by 1 , and hence g (z) = 0. Thus − − 2
πi/12z i πi/6 πi/6 ∞ πi(3 2 )/z f (z) = e− (e + e− ) ( 1) e− − 3z − = −∞ i 1 = f . z − z Kronecker’s Limit Formulas 37
Asa consequence, the function h(z) = f (z)/n(z) is invariant under the transfor- 42 mations z z + 1 and z 1/z and hence under all modular transformations. Moreover,→ since η(z) does→ not − vanish anywhere in H, h(z) is regular in H; fur- ther, as z tends to infinity in the fundamental domain, h(z) 1 tends to zero. − Now by the same argument as for ϑ1(w, z) above, we can conclude that h(z) = 1 and the proposition is proved.
The integers (1/2)(3λ2 λ) for λ = 1, 2,... are the so-called ‘pentagonal numbers’. − We now give another interesting application of Kronecker’s second limit formula. Let us consider, once again, formula (37). The left-hand side may be looked upon as a trigonometric series in u and v; it is true, of course, that u and v are not entirely independent. We shall now see how, using the infinite series expansion of ϑ1(w, z), Kronecker showed that the function
(πi(w q)2)/(z r) ϑ (w, z)ϑ (q, r)e − − 1 1 − has a valid Fourier expansion in u and v and derived a beautiful formula from (37). First,
∞ πiz(n+ 1 )2+2πi(n+ 1 )(w 1 ), ϑ1(w, z) = e 2 2 − 2 n = −∞ ∞ πir(m+ 1 )2+2πi(m+ 1 )(q 1 ) ϑ (q, r) = e− 2 2 − 2 , 1 − m = −∞ replacing m by m 1, we have − −
∞ πir(m+ 1 )2 2πi(m+ 1 )(q 1 ) ϑ (q, r) = e− 2 − 2 − 2 . 1 − m = −∞ Hence
ϑ (w, z)ϑ (q, r) 1 1 − ∞ ∞ πi z(n+ 1 )2+2(n+ 1 )(w 1 ) r(m+ 1 )2 2(m+ 1 )(q 1 ) = e { 2 2 − 2 − 2 − 2 − 2 }. m = n = −∞ −∞ This double series converges absolutely and replacing n by n + m in the inner sum, we have Kronecker’s Limit Formulas 38
ϑ (w, z)ϑ (q, r) 1 1 − ∞ ∞ πi(z r)(m+ 1 )2+2πi(m+ 1 )(w+nz q)+πin2z+2πin(w 1 ) = e − 2 2 − − 2 m = n = −∞ −∞ ∞ n πin2z+2πinw πi(w+nz q)2/(z r) ∞ πi(z r)(m+ 1 +(w+nz q)/(z r))2 = ( 1) e − − − e − 2 − − − n = m = −∞ −∞ Applying the theta-transformation formula to the inner sum, we have 43
πi(w q)2/(z r) ϑ (w, z)ϑ (q, r)e − − 1 1 − i ∞ n πin2z+2πinw πi(n2z2+2nz(w q))/(x r) = ( 1) e − − − z r − × n = − −∞ ∞ m πim2/(z r) 2πim(w q+nz)/(z r) ( 1) e− − − − − × − m = −∞ i mn+m+n πi(m+nz)(m+nr)/(z r)+2πi((m+nz)q (m+nr)w)/(z r) = ( 1) e− − − − z r − − m,n √i/z r being that branch which assumes the value 1 for z r = i. Let− − i Q(ξ,η) = (ξ + ηz)(ξ + ηr) = aξ2 + bξη + cηn, z r − where i i(z + r) izr a = , b = and c = . z r z r z r − − − Then
πi(w q)2/(z r) ϑ (w, z)ϑ (q, r)e − − 1 1 − i mn+m+n πQ(m,n)+2πi((m+nz)q (m+nr)w)/(z r) = ( 1) e− − − z r − − m,n Formula (37) now becomes
i mn+m+n πQ(m,n)+2πi(mu+nv) + 1 e2πi(mu nv) z r m,n( 1) e− ′ = log − − −2π Q(m, n) η(z)η( r) m,n − It is remarkable that the quadratic form Q(m, n) emerges undisturbed on the 44 Kronecker’s Limit Formulas 39 right-hand side; the right-hand side is free from z and r, except for the fac- tors √i/(z r) and η(z)η( r). To eliminate z and r therefrom, we differentiate − πi(w q)2/(−z r) ϑ1(w, z)ϑ1(q, r)e − − once with respect to w and q at w = 0, q = 0, and then we get −
i mn+m+n πQ(m,n) ϑ′ (0, z)ϑ′ (0, r) = ( 1) e− × 1 1 − z r − − m,n 2πi(m + nz) 2πi(m + nr) − , z r z r − − i.e. 3 3 3 i (m+1)(n+1) πQ(m,n) η (z)η ( r) = ( 1) Q(m, n)e− , − z r − − m,n i.e. 1 3 i (m+1)(n+1) πQ(m,n) η(z)η( r) = ( 1) Q(m, n)e− − z r − − m,n the branch of the cube root on the right-hand side being determined by the condition that it is real and positive when r = z, since η(z)η( z) > 0. We have then the following formula due to Kronecker, namely−
2πi(mu+nv) mn+m+n πQ(m,n)+2πi(mu+nv) 1 e ( 1) e− ′ m,n = log − 1 . (53) −2π Q(m, n) 3 m,n ( 1)(m+1)(n+1)Q(m, n)e πQ(m,n) m,n − − It is remarkable that, eventually, on the right-hand side of (53), z and r do not appear explicitly and only the quadratic form Q(m, n) appears on the right. The quadratic form Q(ξ, n) = dξ2 +bξη+cη2 introduced above is a complex quadratic form with discriminant
(r + z)2 + 4rz b2 4ac = − = 1. − (z r)2 − − Moreover,its real part is positive-definite for real ξ and η. For proving this, 45 i we have to show that for (ξ, n) , (0, 0), Re (ξ + zη) (ξ + rη) > 0, z r 1 − , i.e. Im z r (ξ + zn)(ξ + rη) > 0. If η = 0, this is trivial. If η 0, we may take η = −1− without loss of generality and then we have only to show that
z r 1 1 Im − > 0, i.e., Im > 0. (ξ + z)(ξ + r) ξ + r − ξ z − Kronecker’s Limit Formulas 40
But this is obvious, since z, r H and ξ being real, ξ + z, (ξ + r) and hence, 1/(ξ + z), 1/(ξ + r) are all− in H∈. − − Conversely, if Q(ξ,η) = aξ2 + bξη + cη2 is a complex quadratic form with real part positive-definite for real ξ and η and discriminant 1 and if Re a > 0, i − it can be written as (ξ +ηz)(ξ +ηr) for some z, r H. In fact, if Re a > 0, z r − ∈ Q(ξ, η) = a(ξ + ηz)(ξ−+ ηr), where z, r are the roots of the quadratic equation aλ2 bλ + c = 0. Since the real part of Q(ξ,η) is positive definite, z and r have− both got to be complex. Now the discriminant of Q(ξ,η) is 1 and hence a2(z r)2 = 1, i.e. a = i/(z r). We may take a = i/(z r) and then− z r H. − − i ± − − − ∈ Now Q(ξ, η) = (ξ + ηz)(ξ + ηr). We have only to show that z, r H. z r − ∈ − i Consider the expression Q(ξ, 1) = (ξ + z)(ξ + r). If both z and r are in H z r then as ξ varies from to + , the− argument of Q(ξ, 1) decreases by 2π and if both z and r are−∞ in H then∞ as ξ varies from to + , the argument of Q(ξ, 1)− increases− by 2π. But neither case is possible,−∞ since∞ Re Q(ξ, 1) > 0 and so Q(ξ, 1) lies always in the right half-plane for real ξ. Hence either z, r or z, r H. But, again, since z r H we see that z, r H. − − When∈ we subject the quadratic− ∈ form Q(ξ,η) to− the∈ unimodular transfor- mation (ξ, η) (αξ + βη, γξ + δη) where α, β, γ, δ are integers such that αδ βγ = →1, then Q(ξ,η) goes over into the quadratic form Q (ξ, η) = − ± ∗ Q(αξ + βη,γξ + δη). The quadratic forms Q∗(ξ,η) and Q(ξ, n) are said to be equivalent. Q∗(ξ, η) again has discriminant 1 and its real part is positive- definite for real ξ and η. − Let u∗ = αu + γv and v∗ = βu + δv. Then we assert that the right-hand side of (53) is invariant if we replace Q(m, n), u and v respectively byQ∗(m, n), 46 u∗ and v∗. This is easy to prove, for the series on the right-hand side converge absolutely and when m, n run over all integers independently, then so do m∗ = (m +1)(n +1) αm + βn, n∗ = γm + δn. And all we need to verify is that ( 1) ∗ ∗ = ( 1)(m+1)(n+1). This is very simple to prove; for ( 1)(m+1)(n+1)−= +1 unless m − − and n are both even; and m and n are both even if and only if m∗ and n∗ are both even. The invariance of the left hand side of (53) under the transformation Q(m, n), u, v to Q∗(m, n), u∗, v∗ could be also directly proved by observing that for e2πi(mu+nv) Re s > 1, ′ is invariant under this transformation and hence, by (Q)(m, n)s analytic continuation, the invariance holds good even for s = 1. Kronecker’s Limit Formulas 41
5 The Epstein Zeta-function
We shall consider here a generalization of the function ζQ(s, u, v) introduced in 3. Let Q be the matrix of a positive-definite quadratic form Q(x1,..., xn) in § x1 the n variables x ,..., x (n 2). Further let x denote the n-rowed column . . 1 n ≥ . xn A homogeneous polynomial P(x) = P(x1,..., xn) of degree g in x1,..., xn is called a spherical function (Kugel-funktion) of order g with respect to Q(x1,..., xn), if it satisfies the differential equation
∂2P(x) q∗ = 0, i j ∂x ∂x 1 i, j n i j ≤ ≤ 1 where the matrix (qi∗ j) = Q− . Let, for a matrix A, A′ denote its transpose; we shall abbreviate A′QA as Q[A], for an n-rowed matrix A. By an isotropic vector of Q, we mean a com- plex column w satisfying Q[w] = 0. It is easily verified that if w is an isotropic g vector of Q, then the polynomial (x′Qw) is a spherical function of order g with respect to Q(x1,..., xn). Moreover, it is known thatif P(x) is a spherical 47 M P g function of order g with respect to Q(x1,..., xn), then (x) = (x′Qwm) , m=1 w1,..., wM being isotropic vectors of Q. Let u, v be two arbitrary n-rowed real columns and g, a non-negative in- teger. Further, let P(x) be a spherical function of order g with respect to Q(x1,..., xn). We define, for σ > n/2, the zeta-function P(m + v) ζ(s, u, v, Q, P) = e2πim′u (Q[m + v])s+g/2 m +v,0 where m runs over all n-rowed integral columns such that m + v , 0 (0 being the n-rowed zero-column). By the remark on P(x) above, ζ(s, u, v, Q, P)isa linear combination of the series of the form g ((m + v)′Qw) e2πim′u , (Q[m + v])s+g/2 m +v,0 w being an isotropic vector of Q. These series have been investigated in de- tail by Epstein; in special cases as when g = 0 and Q is diagonal, Lerch has considered them independently and has derived for them a functional equation which he refers to as the “generalized Malmsten-Lipschitz relation.” Kronecker’s Limit Formulas 42
The series introduced above converge absolutely for σ > n/2, uniformly in every half-plane σ n/2+ǫ(ǫ > 0), as can be seen from the fact that they have a majorant of the form≥
n g mi + vi i=1 | | c n σ+g/2 m +v,0 2 mi + vi i=1 | | for a constant c depending only on Q, w and g. Hence they represent analytic functions of s for σ > n/2. Thus ζ(s, u, v, Q, P) is an analytic function of s for σ > n/2. We shall study the analytic continuation and the functional equation of ζ(s, u, v, Q, P). The proof will be based on Riemann’s method of obtaining the 48 functional equation of the ζ-function using the theta-transformation formula. First we obtain the following generalization of (52). Proposition 7. If Q is the matrix of a positive-definite quadratic form in m variables with real coefficients and v an n-rowed complex column, then
πq[m+v] 1 πQ 1 m +2πm v e− = Q − 2 e− − | | ′ , (54) | | m m
1 where, on both sides m runs over all n-rowed integral columns and Q 2 is the positive square root of Q , the determinant of Q. | | | | Proof. We shall assume formula (54) proved for Q and v of at most n 1 rows and uphold it for n. For n = 1, the formula has been proved already. − Now, it is well known that 1 P q P 0 E P− q A = = 1 (55) q′ r 0′ r P− [q] 0′ 1 − where P is (n 1)-rowed and symmetric and E, the (n 1)-rowed identity − − = 1 = k = u matrix. Setting λ r P− [q] and writing m 1 and v w with k and u of n 1 rows, we have, in− view of (55), − 2 1 πQ[m+v] πλ(l+w) πP[k+u+P− q(l+w)] e− = E− e− , m 1 k m, k running over all n-rowed and (n 1)-rowed integral columns respectively and l over all integers. By induction hypothesis,−
1 1 1 1 πP[k+u+P− q(l+w)] πP− [k]+2πik′(u+P− q(l+w)) e− = P − 2 e− | | k k Kronecker’s Limit Formulas 43
Thus
1 1 2 1 (l+w) πQ[m+v] πP πλ(l+w) +2πik′ P− q e− = P − 2 e− − [k] + 2πik′u e− | | m k l
Now, by (52), 49
∞ 2 1 πλ(l+w) +2πik′ P q(l+w) e− − l = −∞ 2 1 w ∞ 2 1 πλw +2πik′ P q πλl 2πil(k′ P q+iλw) = e− − e− − − l = −∞ 1 2 1 w ∞ 1 1 2 πλw +2πik′ P− q πλ− (l k′ P− q iλω) = λ− 2 e− e− − − l = −∞ 1 ∞ 1 1 2 πλ− (l k′ P− q) +2πilw = λ− 2 e− − l = −∞ 1 1 1 1 2 But Q = P λ and Q [m] = P [k] + λ ( k′P q + 1) and hence | | | | − − − − −
1 1 πQ[m+v] πQ m +2πi(k′u+lw) e− = Q − 2 e− − | | | | m k,l and (54) is proved. Formula (54) can also be upheld by using the Poisson summation formula in n variables. Further, it can be shown to be valid even for complex symmetric Q whose real part is positive (i.e. the matrix of a positive-definite quadratic 1 form). For, Q− again has positive real part and the absolute convergence of the series on both sides is ensured. Moreover, considered as functions of the n(n + 1)/2 independent elements of Q, they represent analytic functions and since they are equal for all real positive Q, we see, by analytic continuation, that they are equal for all complex symmetric Q with positive real part. Let u be another arbitrary complex n-rowed column. Replacing v by v 1 − iQ− u in (54), we obtain
πQ[m+v iQ 1u] 1 πQ 1[m]+2πim (v iQ 1u) e− − − = Q − 2 e− − ′ − − , | | m m i.e. πQ[m+v]+2πim u 1 πQ 1[m u]+2πi(m u) v e− ′ = Q − 2 e− − − − ′ . (56) | | m m Kronecker’s Limit Formulas 44
Let w be an isotropic vector of Q and λ, an arbitrary complex number. Replac- ingv by v + λw in (56), we have 50
πQ[m+v] 2πλ(m+v) Qw+2πim u e− − ′ ′ m 1 πQ 1[m u]+2πi(m u) v+2πiλ(m u) w = Q − 2 e− − − − ′ − ′ . | | m
Both sides represent analytic functions of λ and differentiating them both, g times with respect to λ at λ = 0, we get
πQ[m+v]+2πim u g e− ′ ((m + v)′Qw) m
2πiu′v e− 1 πQ 1[m u]+2πim v g = Q − 2 e− − − ′ ((m u)′w) . (57) ig | | − m
Associated with a spherical function P(x) with respect to Q[x], let us de- 1 fine P∗(x) = P(Q− x); P∗(x) is a spherical function of order g with respect m 1 P P P g to Q− [x]. Moreover ∗∗(x) = (x). Further, if (x) = (x′Qwi) , then i=1 m P g ∗(x) = (x′wi) . Thus, if we set i=1 πQ[m+v]+2πim u f (Q, u, v, P) = e− ′ P(m + v), m where m runs over all n-rowed integral columns, then we see at once from (57) that f (Q, u, v, P) satisfies the functional equation
g 2πiu v 1 1 i e ′ f (Q, u, v, P) = Q − 2 f (Q− , v, u, P∗). | | − If now we replace Q by xQ for x > 0, then P(x) goes into xgP(x) and P∗(x) remains unchanged and we have
g 2πiu v n/2 g 1 1 1 i e ′ f (xQ, u, v, P) = x− − Q − 2 f (x− Q− , v, u, P∗). (58) | | − Thus we have Proposition 8. Let Q be an m-rowed complex symmetric matrix with positive real part, P(x) a spherical function of order g with respect to Q and P (x) ∗ − Kronecker’s Limit Formulas 45
1 P(Q− x). Let u and v be two arbitrary m-rowed complex columns and x > 0. Then
2πiu v g πxQ[m+v]+2πim u e ′ i e− ′ P(m + v) m n/2 g 1 πx 1 Q 1[m u]+πim v = x− − Q − 2 e− − − − ′ P∗(m u). | | − m 51 To study the analytic continuation of ζ(s, u, v, Q, P), we obtain an integral representation of the same, by using the well-known formula due to Euler, namely, for t > 0 and Re s > 0,
s s ∞ s πtx dx π− Γ(s)t− = x e− . 0 x For σ > n/2, we then have
(s+g/2) g π− Γ s + ζ(s, u, v, Q, P) 2 2πim u ∞ s+g/2 πxQ[m+v] dx = e ′ P(m + v) x e− x m +v,0 0
∞ s+g/2 πxQ[m+v]+2πim u dx = x e− ′ P(m + v) , (59) 0 x m +v,0 in view of the absolute convergence of the series, uniform for σ n/2 + ǫ(ǫ > 0). ≥ Let now, for an n-rowed real column x,
0, if x is not integral, ρ(x, g) = 1, if x is integral and g = 0, 0, if x is integral and g > 0. Then
πxQ[m+v]+2πim u 2πiu v e− ′ P(m + v) = f (xQ, u, v, P) e− ′ ρ(v, g). − m +v,0 From (59), we see, as a consequence, that
(s+g/2) g π− Γ s + ζ(s, u, v, Q, P) 2 Kronecker’s Limit Formulas 46
∞ s+g/2 πxQ[m+v]+2πim u dx = x e− ′ P(m + v) + 1 x m +v,0 1 1 s+g/2 dx 2πiuv s+g/2 dx + x f (xQ, u, v, P) ρ(v, g)e− ′ x . 0 x − 0 x In view of (58), 52 1 dx xs+g/2 f (xQ, u, v, P) 0 x 1 1 Q 2 dx = − xs g/2 n/2 f (x 1Q 1, v, u, P ) . g| 2|πiu v − − − − ∗ i e ′ 0 − x Again, since
1 1 f (x− Q− , v, u, P∗) − πx 1 Q 1[m u]+2πim v 2πiu v = e− − − − ′ P∗(m u) + ρ(u, g)e ′ , − m u,0 − we have 1 s g/2 n/2 1 1 dx x − − f (x− Q− , v, u, P∗) 0 − x 1 s g/2 n/2 πx 1 Q 1[m u]+2πim v dx = x − − e− − − − ′ P∗(m u) + − x 0 m u,0 − 1 2πiu v s (g/2) (n/2) dx + ρ(u, g)e ′ x − − 0 x
∞ (n/2) s+(g/2) πxQ 1[m u]+2πim v dx = x − e− − − ′ P∗(m u) + − x 1 m u,0 − 2πiu v ρ(u, g)e ′ + g n . s − 2 − 2 Thus, for σ > n/2,
(s+g/2) g π− Γ s + ζ(s, u, v, Q, P) 2 1 2πiu v Q 2 ρ(u, g) ρ(v, g)e− ′ = | |− + g n g ig s − s + − 2 − 2 2 Kronecker’s Limit Formulas 47
∞ s+g/2 πxQ[m+v]+2πim u dx + x e− ′ P(m + v) + x 1 m +v,0 1 Q − 2 ∞ 1 dx + (n/2) s+(g/2) πxQ− [m u]+2πim′v P . | 2|πiu v x − e− − ∗(m u) ige ′ × − x 1 m u,0 − (60) 53 One verifies easily that the functions within the square brackets on the πs+g/2 right-hand side of (60) are entire functions of s. Since is also an en- Γ(s + g/2) tire function of s, formula (60) gives the analytic continuation of ζ(s, u, v, Q, P) into the whole s-plane. The only possible singularities arise from those of
1 s+g/2 2πiu′v π Q − 2 ρ(u, g) ρ(v, g)e− | | g g g n g Γ s + i s − s + − 2 − 2 2 2 Now πs+g/2 πs+g/2 = g g Γ s + g s + Γ s + + 1 2 2 2 is an entire function of s and so s = g/2 cannot be a singularity of ζ(s, u, v, Q, P). Moreover, if g > 0 or if u is not− integral, then ρ(u, g) = 0 and hence in these cases, s = (g + n)/2 can not be a pole of ζ(s, u, v, Q, P). If g = 0, and u is integral, ζ(s, u, v, Q, P) has a simple pole at s = n/2 with residue n/2 1 π Q − 2 /Γ(n/2). Moreover, from (60) it is easy to verify that ζ(s, u, v, Q, P) satisfies| | the functional equation
sΓ g π− s + ζ(s, u, v, Q, P) 2 2πiu′v e− 1 (n/2 s) n g n 1 = Q − 2 π− − Γ s + ζ s, v, u, Q− , P∗ (61) ig | | 2 − 2 2 − − Wehave then finally. 54 Theorem 3. The function ζ(s, u, v, Q, P) has an analytic continuation into the whole s-plane, which is an entire function of s if either g > 0 or if g = 0 and u is not integral. If g = 0 and u is integral, then ζ(s, u, v, Q, P) is meromorphic in the entire s-plane with the only singularity at s = n/2 where it has a simple n/2 1 pole with the residue π /( Q 2 Γ(n/2)). In all cases, ζ(s, u, v, Q, P) satisfies the functional equation (61)|. | Kronecker’s Limit Formulas 48
1 2 Let now n = 2 and Q(x1, x2) = y− x1 + x2z , in our earlier notation. x1 m1 u2 | v1 | z z Let x = x , m = m , u = u and v = v . Both −1 and −1 are isotropic 2 2 − 1 2 g vectors of Q [x] but we shall take the spherical function P (x) = ( i(x1 + x2z)) , corresponding to the former. Then by the above, the function −
s+g/2 g y 2πi(m u m u ) (m1 + v1 + z(m2 + v2)) ζ(s, u, v, Q, P) = e 1 2− 2 1 ig m + v + z(m + v ) 2s+g m +v,0 | 1 1 2 2 | satisfies the functional equation
s g π− Γ s + ζ(s, u, v, Q, P) 2 2πiu′v e− (1 s) g 1 = π− − Γ 1 s + ζ(1 s, v, u, Q− , P∗). ig − 2 − − 1 1 2 Now Q− [x] has the simple form y− x1z x2 and moreover, P∗(x) = ( x1z+ g | − | − x2) . Thus for σ > 1,
1 ζ(s, v, u, Q− , P∗) − g ( (m1 u2)z + m2 + u1) = ys+g/2 e2πi(m1v1+m2v2) − − z(m u ) + m + u 2s+g m u,0 1 2 2 1 − | − − | g ((m1 + u1) + z(m2 + u2)) = s+g/2 2πi(m1v2 m2v1) . y e − 2s+g (m1 + u1) + z(m2 + u2) m1 + u1 ,0 (m2) (u2) | |
u1 v1 Let us now define for u∗ = , v∗ = , the function u2 v2 g (m1 + v1 + z(m2 + v2)) ζ(s, u , v , z, g) = ys e2πi(m1u2 m2u1) , ∗ ∗ − + + + 2s+g + , m1 v1 z(m2 v2) m v∗ 0 | |
g g/2 forσ > 1. It is clear that ζ(s, u∗, v∗, z, g) = i y− ζ(s, u, v, Q, P) and from 55 g/2 1 (??) we see that ζ(s, v∗, u∗, z, g) = y− ζ(s, v, u, Q− , P∗). If now we defined s − ϕ(s, u∗, v∗, z, g) = π− Γ(s + g/2)ζ(s, u∗, v∗, z, g), we deduce from above that ϕ(s, u∗, v∗, z, g) satisfies the nice functional equation
2πi(u v u v ) ϕ(s, u∗, v∗, z, g) = e 1 2− 2 1 ϕ(1 s, v∗, u∗, z, g). −
In the case g > 0, ζ(s, u∗, v∗, z, g) is an entire function of s and one can ask for analogues of Kronecker’s second limit formula even here. For even g, one gets by applying the Poisson summation formula, a limit formula which is connected with elliptic functions. For odd g, the limit formulas which one gets Kronecker’s Limit Formulas 49 are more complicated and involve Bessel functions. If we take the particular case g = 1, there occurs in the work of Hecke, a limit formula (as s tends to 1/2) which has an interesting connection with the theory of complex multiplication. For n > 2, Epstein has obtained for u = 0, v = 0 and g = 0, an analogue of the first limit formula of Kronecker. This formula involves more complicated functions that the Dedekind η-function. Perhaps, in general for n > 2, one cannot expect to get a limit formula which would involve analytic functions of several complex variables. Bibliography
[1] S. Bochner: Vorlesungen ¨uber Fouriersche Integrale, Chelsea, New York, (1948). [2] R. Dedekind: Erl¨auterungen zu den Fragmenten XXVIII, Collected Works of Bernhard Riemann, Dover, New York, (1953), 466-478. [3] P. Epstein: Zur Theorie allgemeiner Zeta funktionen, I, II, Math, Ann. 56 (1903), 615-644, ibid, 63 (1907), 205-216. [4] E. Hecke: Bestimmung der Perioden gewisser Integrale durch die Theorie der Klassenkorper, Werke. Gottingen (1959), 505-524. [5] F. Klein: Uber die elliptischen Normalkurven der n-ten Ordnung, Gesamm. Math. Abhandlungen. Bd. III. Berlin. (1923). 198-254.
[6] Kronecker:Zur Theorie der Clliptischen Funktionen, Werke. Vol. 4. 347- 56 495, vol. 5, 1-132, Leipzig, (1929). [7] M. Lerch: Zakladove´ Theorie Malmstenovskych rad, Rozpravy ˇceske akademie Ciˇsare Frantiska Josefa, Bd. 1, No. 27, (1892), 525-592. [8] C. Meyer: Die Berechnung der Klassenzahl abelscher K¨orper ¨uber quadratischen Zahlk¨orpern, Akademie Verlag, Berlin, 1957. [9] H. Rademacher: Zur Theorie der Modulfunktionen, J. f¨ur die reine u. angew. Math., vol. 167 (1931), 312-336. Collected papers. [10] B. Schoenesberg: Das Verhalten von mehrfachen Thetareihen bei Modul- substitutionen, Math. Ann. 116 (1939), 511-523. 1 τ [11] C. L. Siegel: A simple proof of η = η(τ) , Mathematika 1 −η i (1954), 4. Ges Abhand.
50 Bibliography 51
[12] E. T. Whittaker and G. N. Watson: Course of Modern Analysis, Cam- bridge, (1946). Chapter 2
Applications of Kronecker’s Limit Formulas to Algebraic Number Theory
1 Kronecker’s solution of Pell’s equation 57 Let K be an algebraic number field of degree n over Q, the field of rational numbers. Let, for any ideal a in K, N(a) denote its norm. Further let s = σ + it be a complex variable. Then for σ > 1, we define after Dedekind, the zeta function s ζK(s) = (N(a))− , a where the summation is over all non-zero integral ideals of K. More generally, with a character χ of the ideal class group of K, we asso- ciate the L-series
s s 1 LK(s, χ) = χ(a)(N(a))− = (1 χ(p)(N(p))− )− , a p − for σ > 1. The product on the right runs over all prime ideals p of K. We have h such series associated with all the h characters of the ideal class group of K. It has been proved by Hecke that these functions LK(s, χ) can be continued analytically into the whole plane and they satisfy a functional equation for s 1 s. → −
52 Applications to Algebraic Number Theory 53
Now, χ being an ideal class character, we may write
s LK(s, χ) = χ(A) (N(a))− , A a A ∈ s A running over all the ideal classes of K. If we define ζ(s, A) = (N(a))− , a A then it can be shown that lim(s 1)ζ(s, A) = κ, a positive constant ∈ depending 58 s 1 − only upon the field K and→not upon the ideal class A. (For example, in the case of an imaginary quadratic field of discriminant d, κ = 2/w √ d, w being the number of roots of unity in K and κ = 2 log ǫ/ √d in the case− of a real quadratic field of discriminant d, where ǫ is the fundamental unit and ǫ > 1). Since ζK(s) = ζ(s, A), we have A
lim(s 1)ζK(s) = lim(s 1)ζ(s, A) = κ h s 1 − s 1 − → A → where h denotes the class number of K. From now on, we shall be concerned only with quadratic fields of discrim- inant d. We have then, on direct computation,
ζK(s) = ζ(s)Ld(s), (63) where ζ(s) is the Riemann zeta-function and Ld(s) is defined as follows:
∞ d s Ld(s) = n− , (σ > 1) n n=1
d being the Legendre-Jacobi-Kronecker symbol. n Then
lim(s 1)ζK(s) = lim(s 1)ζ(s)Ld(s) = lim Ld(s) = Ld(1), s 1 − s 1 − s 1 → → → since Ld(s) converges in the half-plane σ > 0 and lim(s 1)ζ(s) = 1. s 1 − We obtain therefore, →
∞ d 1 Ld(1) = = κ h. (64) n n n=1 Applications to Algebraic Number Theory 54
For the determination of the left-hand side, we consider more generally, for any character χ(, 1) of the ideal class group of K, χ(a) lim LK(s, χ) = LK(1, χ) = . s 1 N(a) → a
Now, 59 κ ζ(s, A) = + ρ(A) + , s 1 − and so
LK(s, χ) = χ(A)ζ(s, A) A κ = χ(A) + ρ(A) + s 1 A − = χ(A)ρ(A) + terms involving higher powers of (s 1), − A since, χ being , 1, χ(A) = 0. On taking the limit as s 1, we have A →
LK(1, χ) = χ(A)ρ(A). (65) A
Therefore, the problem of determination of LK(1, χ) has been reduced to that of ρ(A). We now study LK(s, χ) for a special class of characters, the so-called genus characters (due to Gauss) to be defined below. We call a discriminant, a prime discriminant, if it is divisible by only one prime. In that case, d = p if d is odd, or if d is even, d = 4 or 8. ± − ± Proposition 9. Every discriminant d can be written uniquely as a product of prime discriminants.
Proof. It is known that if d is odd, then d = p1 ... pk where p1,..., pk are mutually distinct odd primes. Let P = p according± as p 1( mod 4); 1 ± 1 1 ≡ ± then P1 is a prime discriminant and d/P1 is again an odd discriminant. Using induction on the number k of prime factors of d (odd!) it can be shown that d/P1 = P2,..., Pk where Pi are odd prime discriminants. If d is even, then we know that
(a) d = 8p ,..., p , if d/4 2( mod 4), ± 2 k ≡ Applications to Algebraic Number Theory 55
(b) d = 4p ,..., p , if d/4 3( mod 4), ± 2 k ≡ p2,..., pk being odd primes. Now we may write d = P1d1 where P1 is chosen to be 4, +8 or 8 such that d is an odd discriminant. From the above it is 60 − − 1 clear that d = P1P2 ... Pk where P1,..., Pk are all prime discriminants. This decomposition is seen to be unique. We may now introduce the genus characters. Let d1 be the product of any of the factors P1,..., Pk of d. Then d1 is again a discriminant and d d. Let d = d d ; d is also a discriminant. Indentifying 1| 1 2 2 the decompositions d = d1d2 and d = d2d1, we note that the number of such k 1 decompositions (including the trivial one, d = 1 d) is 2 − where k is the number of different prime factors of d. For any such decomposition d = d1d2 of d and for any prime ideal p not dividing d, we define d1 χ(p) = χd (p) = , 1 N(p) d where 1 is the Legendre-Jacobi-Kronecker symbol. We shall show that N(p)
d2 d1 χd (p) = = . 2 N(p) N(p)
In face, since p d, p belongs to one of the following types. (a) pp′ = (p), × 2 N(p) = p or (b) p = (p), N(p) = p . If pp′ = (p), N(p) = p then d d d = 1 = 1 2 p N(p) N(p)
d d which means that either both are +1 or both are 1. i.e. 1 = 2 . If − N(p) N(P) 2 d p = (p), N(p) = p then p = 1; but − d d d 1 2 = = 1, N(p) N(p) p2
d d so that we have again 1 = 2 . N(p) N(p) d d When p d, one of the symbols 1 , 2 is zero, and the other non- | N(p) N(p) zero, we take χ(p) to be the non-zero value. In any case, χ(p) = 1. This ± definition can then be extended to all ideals a of K as follows: If a = pkql,..., 61 Applications to Algebraic Number Theory 56 we define χ(a) = (χ(p))k(χ(q))l,..., so that
χ(ab) = χ(a)χ(b) for any two ideals a, b of K. By a genus character, we mean a character χ defined as above, correspond- k 1 ing to any one of the 2 − different decompositions of d as product of two mutually coprime discriminants. We shall see later that the genus characters form an abelian group of order k 1 2 − . Now we shall obtain an interesting consequence of our definition of the character χ, with regard to the associated L-series. Consider the L-series, for s = σ + it, σ > 1
s s 1 LK(s, χ) = χ(a)(N(a))− = (1 χ(p)(N(p))− )− a p − s 1 = (1 χ(p)(N(p))− )− . − p p (p) | The prime ideals p of K are distributed as follows: d (a) p = (p), = 1, N(p) = p2, p − d (b) pp′ = (p), = +1, N(p) = p, p d (c) p2 = (p), = 0, N(p) = p. p d d d d d In case (a), = 1 = 1 2 implies that one of 1 , 2 is +1 and the p − p p p p other 1. So − s 1 2s 1 (1 χ(p)(N(p))− )− = (1 p− )− − − p (p) | s 1 s 1 = (1 p− )− (1 + p− )− − 1 1 d1 s − d2 s − = 1 p− 1 p− − p − p
In case (b), 62 Applications to Algebraic Number Theory 57
1 1 s 1 d1 s − d1 s − (1 χ(p)(N(p))− )− = 1 p− 1 p− − − p − p p (p) | 1 1 d1 s − d2 s − = 1 p− 1 p− , − p − p d d d d d since 1 = = 1 2 implies that 1 = 2 . In case (c) p d implies p p p p p | that p d1 or p d2. We shall assume without loss of generality that p d2. Then d | | | 2 = 0 and p
1 1 s 1 d1 s − d2 s − (1 χ(p)(N(p))− )− = 1 p− 1 p− . − − p × − p p (p) | From all cases, we obtain
s 1 L (s, χ) = (1 χ(p)(N(p))− )− K − p p (p) | 1 1 d1 s − d2 s − = 1 p− 1 p− − p − p p
= Ld1 (s)Ld2 (s). We have therefore, Theorem 4 (Kronecker). For a genus character χ of a K corresponding to the decomposition d = d1d2, we have,
LK(s, χ) = Ld1 (s)Ld2 (s). (66) Let us consider the trivial decomposition d = 1 d or d = 1 and d = d. 1 2 Then Ld1 (s) = ζ(s) and Ld2 (s) = Ld(s). In other words, (66) reduces to (63). If d , 1, ζ (s) = ζ(s)L (s), from (63). But L (s) can then be continued 1 Q( √d1) d1 d1 analytically into the whole plane and it is an entire function. Two ideals a and b are said to be equivalent in the “narrow sense” if a = b(γ) with N(γ) > 0. (For α K, N(α) denotes its norm over Q.) ∈ This definition coincides with the usual definition of equivalence in the case 63 of an imaginary quadratic field since the norm of every element is positive. In the case of a real quadratic field, if there exists a unit ǫ in the field with N(ǫ) = 1, then both the equivalence concepts are the same, as can easily be seen. − Otherwise, if h0 denotes the number of classes under the narrow equivalence and h, the number of classes under the usual equivalence, we have h0 = 2h. Applications to Algebraic Number Theory 58
Proposition 10. For a genus character χ, χ(a) = χ(b), if a and b are equivalent in the narrow sense. Proof. We need only to prove that χ((α)) = 1 for α integral and N(α) > 0.
d1 (a) Let (α, d1) = (1). We have to show that = 1. Since N(α) > 0, N((α)) d this is the same as proving that 1 = 1. N(α)
d1 (1) Suppose d1 is odd. Then = 1. If d is even, d = 4m (say). Then 4 α = x + y √m with x, y rational integers. If not, 2α = x′ + y′ √m with x′ and y′ rational integers. In any case, since (d1/4) = 1, it is d d sufficient to prove that 1 = 1. i.e. 1 = 1 for two N(2α) a2 mb2 2 2 −2 2 rational integers a, b with a mb > 0 and (a mb , d1) = 1, in both cases. Now d d implies− that d m so that from− the periodicity 1| 1| d1 of the Jacobi symbol, this is the same as = 1 for (a, d1) = 1, a2 which is obvious.
(2) Suppose d1 is even. Then d1 = (even discriminant) (odd dis- criminant) and we need consider only the even part, since× we have already disposed of the odd part in 1). We have then three possibil- ities d = 4, +8 and 8. 1 − − (i) d1 = 4. Then d = ( 4)( m) so that m is again a discrimi- nant and− 1( mod 4).− − − Consider≡ now d 4 1 = − ; N(α) x2 my2 − (x2 my2) 1( mod 4) since both x, y cannot be even or odd, − ≡ for (α, d1) = (1). From the periodicity of the Jacobi symbol, 64 follows then that 4 4 − = − = 1. x2 my2 1 − (ii) d1 = +8. Here d = 8 m/2; m/2 is then an odd discriminant and 1( mod 4) or equivalently,× m 2( mod 8). Now,≡ ≡ d 8 1 = ; N(α) x2 my2 − Applications to Algebraic Number Theory 59
(x2 my2) 1( mod 8) so that − ≡ ± 8 8 8 1 = = 1 or = = = 1. x2 my2 1 7 7 − If d = 8, m 2( mod 8) and x2 my2 1, 3( mod 8) 1 − ≡ − − ≡ d1 8 1 and = − = = 1. So, if (α, d1) = (1), χ((α)) = N(α) 3 3 1. If(α, d1) , (1) and (α, d2) = (1), we can apply the same arguments for d2 instead of d1 and prove χ((α)) = 1.
(b) If (α, d1) , (1) and (α, d2) , (1), we decompose (α) as follows: (α) = p1p2 ... plq where pi d and (q, d) = (1). We choose in the narrow class of 1 | p1− , an integral ideal q1 with (d, q1) = (1). Then for the element α1 with (α ) = p q , N(α ) > 0, χ((α )) = 1, since (α , d )or(α , d ) = (1). 1 1 1 1 1 1 1 1 2 Similarly for p2, construct q2 with (d, q2) = 1 and for α2 with (α2) = p2q2 and N(α2) > 0, χ((α2)) = 1 and so on. Finally, we obtain
2 2 2 (αα1α2 ...) = p1p2 ... pl p where (b, d) = (1). But p d imply that p2 = (p ) with p d. Therefore (αα α ...) = i| i i i| 1 2 (p1 p2 ...)b so that b is a principal ideal = (ρ) (say). Now N(αα1α2 ...) = 2 2 2 p1 p2 ... p1N(ρ) and χ((αα1 ...αl)) = χ((ρ)) = 1 since N(ρ) > 0 and (ρ, d) = (1). But χ((α1)) = 1,...χ((αl)) = 1 so that χ((α)) = 1. The proof is now 65 complete.
We have just proved that χ(a) depends only on the narrow class of a, say A χ is therefore a character of the ideal class group in the “narrow sense”. Then, s LK(s, χ) = χ(A) (N(a))− = χ(A)ζ(s, A) (67) A a A A ∈ where A runs over all the ideal classes in the ‘narrow sense’. Using a method of Hecke, we shall now give an alternative proof of the fact that χ((α)) = 1 for principal ideals (α) with N(α) > 0 and (α, d) , (1). Denote by χ0 that character of the ideal class group in the narrow sense, such that χ0(a) = χ(a) for all ideals a in whose prime factor decomposition, only prime ideals p which do not divide d, occur. We need only to prove that χ(a) = χ0(a) for all ideals a. Applications to Algebraic Number Theory 60
s s Let fχ0 (s) = χ0(a)(N(a))− and fχ(s) = χ(a)(N(a))− . Then a a s fχ (s) (1 χ(p)(N(p)) ) 0 = − − , f (s) (1 χ (p)(N(p)) s) χ p d 0 − | − since the remaining factors cancel out. We shall show that the product on the right is = 1, or fχ0 (s) = fχ(s).
We know that fχ(s) = Ld1 (s) = Ld2 (s) from (66) and fχ0 (s) = χ0(A). A ζ(s, A) from (67). Both have functional equations of the same type, so that if we denote by R(s), the quotient, the functional equation is simply R(s) = R(1 s). − We shall now arrive at a contradiction, by supposing that χ , χ0. For, then, there exists a prime ideal p with p d and χ(p) = 1, χ0(p) = 1. log( 1) + 2kπi | ± ∓ Choose s = ∓ (k, such that s , 0). This means that p s = − log p − 1. Consider the product ∓ (1 χ(p)(N(p)) s) R(s) = − − . (1 χ (p)(N(p)) s) p d 0 − | − The above value of s is a zero of the denominator and it cannot be cancelled by any factor in the numerator, for that would mean
log( 1) + 2kπi log( 1) + 2lπi ∓ = ± , log p log q or in other words, λ = log p/ log q is rational; i.e. qλ = p holds for p, q 66 primes and λ rational. This is not possible since p , q. By the same argument, the zeros and poles of R(s) cannot cancel with those on the other side of the equation R(s) = R(1 s). But this is a contradiction. In other words, χ(a) = − χ0(a) for all ideals a of K. Tow ideals a and b are in the same genus, if χ(a) = χ(b) for all genus characters χ defined as above, associated with the decompositions of d. The ideals a for which χ(a) = 1 for all genus characters χ, constitute the principal genus. Then the narrow classes A for which χ(A) = 1 are the classes in the principal genus. If H denotes the group of narrow classes and G, the subgroup of classes lying in the principal genus, the quotient group G = H/G k 1 gives the group of different genera. We claim that the order of G is 2 − , k being the number of different prime factors of d. For, suppose we have proved k 1 that the genus characters form a group of order 2 − . From the definition of a genus and from the theory of character groups of abelian groups, we know that Applications to Algebraic Number Theory 61 this group is the character group of G. By means of the isomorphism between k 1 G and its character group, it would then follow that the order of G is also 2 − . If now remains for us to prove
k 1 Proposition 11. The genus characters form an abelian group of order 2 − , where k is the number of distinct prime factors of d. Proof. Now, χ(a) = 1 implies that χ2 = 1 for all genus characters χ. So, 1 ± the character χ− exists and = χ. We shall prove that the genus characters and closed under multiplication and that they are all different for different decom- positions of d.
Let d = d1d2 = d1∗d2∗. Set d1 = qu1 and d1∗ = qu1∗ so that q = (d1, d1∗) and (u1, u1∗) = 1. If d = u u then d d = q2d . Also for the characters χ and χ associ- 3 1 1∗ 1 1∗ 3 d1 d1∗ ated with the two decompositions d d and d d of d, we have χ χ = χ and 1 2 1∗ 2∗ d1 d1∗ d3 d3 d, since both d1 d and u1∗ d and they are coprime. Hence χd3 is again a genus character.| Now, if |d , d ,| we have to show that χ , χ or χ χ , χ (the 1 1∗ d1 d1∗ d1 d1∗ 1 identity character); or in other words, we need only to prove that for a proper 67 decomposition d = d3d3∗, the associated character χ is not equal to χ1. We have, from (66), for σ > 1,
L (s, χ) = L (s)L (s) = χ(A)ζ(s, A). K d3 d3∗ A
Comparing the residue at s = 1 in the two forms for LK(s, χ), we have χ(A) = A 0, which implies that χ , χ1. Thus the genus characters are different for different decompositions of d k 1 and they form a group. Since there are 2 − different decompositions of d, this k 1 group of genus characters is of order 2 − . A genus character is a narrow class character of order 2. Conversely, we can show that every narrow class character of order 2 is a genus character. For this, we need the notion of an ambiguous (narrow) ideal class. A narrow ideal class A satisfying A = A′ (the conjugate class of A in K) or equivalently A2 = E (the principal narrow class) is called an ambiguous ideal class. The ambiguous classes clearly form a group. Proposition 12. The group of ambiguous ideal classes in Q( √d) is of order k 1 2 − , k being the number of distinct prime factors of d.
Proof. We shall pick out one ambiguous ideal b (i.e. such that b = b′, the conjugate ideal) from each ambiguous ideal class A and show that there are k 1 2 − such inequivalent ideals. Applications to Algebraic Number Theory 62
Let a be any ideal in the ambiguous ideal class A. Since a a′, we may 1 ∼ take aa′− = (λ) with N(λ) = 1 and λ totally positive (in symbols; λ > 0). (For d < 0, the condition “totally positive” implies no restriction). Let ρ = 1 + λ. Then ρ , 0 since λ , 1. Further λ = (1 + λ)/(1 + λ′) = ρ/ρ′. If we define b = a/(ρ), the b = b , for− b/b = a/a (ρ )/(ρ) = (1). Further b A, since ρ > 0 ′ ′ ′ ′ ∈ (i.e. ρ > 0, ρ′ > 0). We may take b to be integral without loss of generality. We call b primitive, if the greatest rational integer r dividing b is 1. For any ambiguous integral ideal b we have b = rb1 with r, the greatest integer dividing b and b primitive. If b A, b A. 1 ∈ 1 ∈ We shall now prove that any primitive integral ambiguous ideal b is of the 68 form pλ1 ,..., pλk with λ = 0 or 1 and p ϑ, (p , p ), where ϑ is the different of 1 k i i| i j √ 1 s 1 a K = Q( d) over Q. Let b = q1 ,..., qs . Now b = b′ implies that q1 ,..., qs = 1 s 1 t q′1 ,..., q′s . We have then q1 = q′t or 1 = t and q1 = qt′. We assert then 1 t 1 t = 1, for if not, q1 = qt′ implies that q1′ = qt and the factor q1 qt = (q1q1′ ) is a rational ideal , (1). This is a contradiction to the hypothesis that b is primitive.
Hence t = 1 and q1 = q1′ . The same argument applies to all prime ideals qi. 2 Since qi = (pi) with prime numbers pi, the exponents i are either 0 or 1. for otherwise they bring in rational ideals which are excluded by the primitive λ1 λk nature of b. Therefore, we have b = p1 ,..., pk with pi ϑ and λi = 0 or 1. k 1 | Now, we shall show that there are exactly 2 − inequivalent (in the narrow sense) such ideals. For the same, it is enough to prove that there is only one l1 lk non-trivial relation of the form p1 ,..., pk (1) in the narrow sense, for, then, k ∼ k 1 it would imply that among the 2 such ideals, there are 2 − inequivelent ones, which is what we require. We shall first prove uniqueness, namely that if there is one non-trivial re- lation, then it is uniquely determined. Later, we show the existence of a non- trivial relation.
l1 lk , (a) Let p1 ,..., pk = (ρ) with ρ > 0 and li > 0or(ρ) (1). Then the set i (l1,..., lk) is uniquely determined.
For, (ρ) = (ρ′) implies that ρ = η ρ′ with a totally positive unit η. The group of totally positive units in K being cyclic, denote by ǫ, the generator of this group. (Note that, for d < 0, the condition “totally positive” imposes no restrictions). Since (ρ) = (ρǫn), when ρ is replaces by ρǫn, η goes over to ηǫ2n. Choosing n suitably, we may assume without loss of generality η = 1 or ǫ. We shall see that (ρ) , (1) implies that η , 1. For, if η = 1, then ρ = ρ′ = a natural number. Now (N(ρ)) = 2 2l1 2lk l1 lk , (ρ ) = p1 ,..., pk = (p1 ,..., pk ), if N(pi) = pi, pi p j and ρ being a Applications to Algebraic Number Theory 63
natural number, this is possible only if all li = 0, in which case (ρ) = (1). But that is a contradiction to the hypothesis (ρ) , (1). We are therefore left with the case η = ǫ. Then set = 1 η = 1 ǫ , 0. − − We have = η ′. Denoting by δ = √d, (ρ/ δ)′ = ρ/ δ = r = a rational number−, 0. Hence (ρ) = ( δ)(r) and the decomposition 1 1 ( δ) = (ρ) = pl1 ,..., plk r r 1 k
is uniquely determined or in other words, the set (l1,..., lk) is uniquely 69 fixed.
(b) We shall now prove the existence of a non-trivial relation pl1 ,..., plk 1 k ∼ (1)(li = 0 or 1) in the narrow sense. Consider = 1 ǫ. Then δ is in general not primitive. Now choose a rational number− r suitably so that r δ is primitive and denote it by ρ. Then r δ = ρ = ǫρ′. Now if d < 0, N( δ) > 0 clearly and if d > 0, 2 N( δ) = dǫ ′ > 0, again. Hence N(ρ) > 0. We may assume that ρ > 0 and since ǫ is not a square, ρ is not a unit, so that (ρ) , (1). Since (ρ) is primitive, (ρ) , rational ideal. Now (ρ) = (ρ′), ρ > 0 and (ρ) is primitive l1 lk so that p1 ,..., pk = (ρ) (1) in the narrow sense. Further this relation is non-trivial as we have∼ just shown. Proposition 12 is thus completely proved. Now, the group of narrow class characters χ with χ2 = 1 is again of k 1 order 2 − , since it is isomorphic to the group of ambiguous (narrow) ideal classes in K. But the genus characters from a subgroup (of order k 1 2 2 − ) of the group of narrow class characters χ with χ = 1. Hence every narrow class character of order 2 is a genus character. Or, equivalently every real narrow class character is a genus character. For any genus character χ, χ(i2) = 1 for every ideal i. In other words, i2 is in the principal genus. Or, for any two ideals a and b with a bi2 (in the narrow sense), we have χ(a) = χ(b) for every genus character∼ χ. That the converse is also true is shown by Theorem 5 (Gauss). If two ideals a and b are in the same genus, there exists an ideal i such that a bi2, in the narrow sense. In particular, if a is in the principal genus, a i2 ∼for an ideal i. ∼ Proof. Two narrow classes A and B are by definition, in the same genus if and only if χ(A) = χ(B) for all genus characters χ or by the foregoing, for all narrow class characters χ with χ2 = 1. But from the duality theory of subgroups of Applications to Algebraic Number Theory 64
1 abelian groups and their character groups, we deduce immediately that AB− = C2 for a narrow ideal class C. This is precisely the assertion above.
Remarks. (1) The notion of genus in the theory of binary quadratic forms 70 is the same as above, if one carries over the definition by means of the correspondence between ideals and quadratic forms.
(2) Hilbert generalized the notion of genus to arbitrary algebraic number fields and used it for higher reciprocity laws and class field theory.
We now come back to formulas (66) and (67) (d < 0). If d = d1d2
Ld1 (s)Ld2 (s) = χ(A)ζ(s, A), A
the summation running over all (narrow) ideal classes A in K = Q( √d). 1 Consider any ideal b A− . For a A, ab = (γ) with N(γ) > 0 we have then ∈ ∈ s (N(b)) s ζ(s, A) = (N(γ))− , w b γ,0 | where w denotes the number of units in K = Q( √d). Let [α, β] be an integral basis of b. Then we may write b = [α, β] = (α)[1, β/α] so that β/α = z = x + iy with y > 0, i.e. we may suppose that b = [1, z]. Then, for γ b, if γ = m + nz with m, n integers , (0, 0), N(γ) = m + nz 2. We obtain∈ | | s N([1, z]) 2s ζ(s, A) = ′ m + nz − w | | m,n
1 z Now, abs. = z′ z = 2y = N(b) √ d , so that 1 z′ | − | | | s 1 2y 2s ζ(s, A) = ′ m + nz − w √ d | | | | m,n s 1 1 2 2 − s ′ 2s = y m + nz − . w √ d √ d m,n| | | | | | From Kronecker’s first limit formula, we obtain the following expansion:
s 2s 1 2 y ′ m + nz − = π + 2C 2 log 2 2log( √y η(z) ) + . | | s 1 − − | | m,n − Applications to Algebraic Number Theory 65
Further 71 s 1 2 − (s 1) log 2/ √ d 2 = e − | | = 1 + (s 1) log + . √ d − √ d | | | | On multiplication, we obtain 2π 1 ζ(s, A) = + 2C log d 2log( N([1, z]) η(z) 2)+ w √ d s 1 − | | − | | | | − + terms with higher powers of (s 1) . − We now set F(A) = N([1, z]) η(z) 2. ( ) | | ∗ Clearly, F(A) depends only upon the class A. In case d < 4, w = 2, so that − 2π Ld (s)Ld (s) = − χ(A) log F(A)+ 1 2 √ d | | A + terms with higher powers of (s 1). − Taking s = 1, we have 2π Ld (1)Ld (1) = − χ(A) log F(A). 1 2 √ d | | A Let us assume without loss of generality, d1 > 0, d2 < 0. Then, by Dirich- let’s class number formula for a quadratic field, we have
2h1 log ǫ Ld1 (1) = √d1 whre ǫ is the fundamental unit and h1 the class-number of Q( √d1) and
2πh2 Ld (1) = , 2 w √ d | 2| h2 denoting the class-number and w, the number of roots of unity in Q( √d2). 72 On taking the product of these two, we obtain 2h h 1 2 log ǫ = χ(A) log F(A). w − A Summing up, we have the following. Let d = d d < 4 be the discrimi- 1 2 − nant of an imaginary quadratic field over Q and let d1 > 0, d2 < 0 be again dis- criminants of quadratic fields over Q with class-numbers h1, h2, respectively. Let w be the number of roots of unity in Q( √d2). Then one has Applications to Algebraic Number Theory 66
Theorem 6. For the fundamental unit ǫ of Q( √d1), we have the formula
2h h /w χ(A) ǫ 1 2 = (F(A))− , (68) A where A runs over all the ideal classes in Q( √d) and F(A) has the meaning given in ( ) above. ∗ If w = 2, the exponent of ǫ in (68) is a positive integer. 2 2 Now ǫ satisfies Pell’s diophantine equation x d1y = 1. And F(A) involves the values of η(z) . So we have a solution of− Pell’s equation± by means of “elliptic functions”.| This| was found by Kronecker in 1863. An Example. We shall take d = 20, d1 = 5, d2 = 4 so that d = d1d2. We know that h = h = 1, w = 4. The− class number of Q−( √ 20) is 2 and for the 1 2 − two ideal classes, say A1, A2, we can choose as representatives the ideals
1 + √ 5 b = [1, √ 5] = (1) and b = 1, − . 1 − 2 2 Now N(b1) = 1 and N(b2) = 1/2. From the definition of F(A), we have
2 2 1 1 + √ 5 F(A1) = η( √ 5) and F(A2) = η − . | − | √2 2 If χ(, 1) be the genus character associated with the above decompos ition of d, 5 then necessarily χ(A1) = 1 and χ(A2) = = 1. Formula (68) now takes the 2 − form 4 1 + √ 5 η − 1 2 ǫ = . (69) 2 η( √ 5) 4 | − | From the product expansion of the η-function, we obtain 73
π √5/12 2 4 6 η( √ 5) = e− (1 q )(1 q )(1 q ) ... , − | − − − | 1 + √ 5 where q = e π √5 and for η − , the following expansion: − 2 1 + √ 5 π √5/24 2 3 4 η − = e− (1 + q)(1 q )(1 + q )(1 q ) ... . 2 | − − | Applications to Algebraic Number Theory 67
On substituting these expansions in (69), we have 1 ǫ = eπ √5/6(1 + q)4(1 + q3)4(1 + q5)4 .... 2
π √5/3 3 3 9 Now, e > 10 so that q < 10− or q < 10− . Hence the product on the right converges rapidly and if we replace the infinite product by 1, we have
1 π √5/6 2 ǫ = e + O(10− )− 2 i.e. 1 + √5 1 π √5/6 2 ǫ = e (upto an error of the order of 10− ). 2 ∼ 2 It is to be noted that in the expression for ǫ as an infinite product, if one cuts off at any finite stage, the resulting number on the right is always transcen- dental, but the limiting value ǫ on the left is algebraic. We consider now the trivial decomposition
ζd(s) = ζ(s)Ld(s) = ζ(s, A). A On substituting the expansion of ζ(s, A) on the right side in powers of (s 1), we have − 2πh 1 2 ζ (s) = + 2C log d log F(A) + . (70) d s 1 − | | − h w √ d A | | − ∞ d s On the other hand, ζ(s) = 1/(s 1) + C + and Ld(s) = n− = 74 − n=1 n L(1) + (s 1)L (1) + . From (70), we have now the equation, − ′ 2πh 1 2 + 2C log d log F(A) + s 1 − | | − h w √ d A | | − 1 = + C + (L(1) + (s 1)L′(1) + ). s 1 − − On comparing coefficients of 1/(s 1) and constant terms on both sides, we obtain − 2πh L(1) = , w √ d | | (71) 2πh 2 L′(1) = C log d log F(A) , w √ d − | | − h A | | Applications to Algebraic Number Theory 68 or we have an explicit expression for L (1) 2 ′ = C log d log F(A). (71) L(1) − | | − h ′ A We shall obtain now for the left side, an infinite series expansion. Let us consider the infinite product
1 d s − L(s) = 1 p− . − p p this product converges absolutely for σ > 1, and on taking logarithmic deriva- tives, we obtain
1 L′(s) d s − d s = 1 p− p− log p L(s) − − p p p d log p p = . (72) − d p ps − p
On passing to the limit as s 1, if the series on the right converges at 75 → s = 1, then by an analogue of Abel’s theorem, it is equal to L′(1)/L(1). But it is rather difficult to prove. One may proceed as follows: d Let π+(X) = p : p prime X and = 1 and π (X) = p : p prime { ≤ p } − { d X and = 1 . Then π+(X) π (X) tends to less rapidly than π(X) ≤ p − } − − ∞ as X . If one could show that this function has the estimate 0(X/(log X)′) where→r ∞> 2, then the convergence of the series on the right side of (72) at s = 1, can be proved. For this estimate, one requires a generalization of the proof of prime number theorem for arithmetical series. We shall now compute L′(1)/L(1) for some values of d. Examples. d = 4. The quadratic field Q( √d) = Q( √ 1) with discriminant 4 has class number− 1. Also, − − d = 1 if n 1 (mod 4) and = 1 if n 3 (mod 4). n ≡ − ≡ Therefore, s s s L(s) = 1− 3− + 5− − − Applications to Algebraic Number Theory 69 and 1 1 1 L(1) = 1− 3− + 5− − − the so-called Leibnitz series. From Dirichlet’s class number formula, we have
2π π 1 1 1 L(1) = = = 1− 3− + 5− . 4.2 4 − −
More interesting is the series for L′(1)/L(1): L (1) log 3 log 5 log 7 log 11 log 13 ′ = + + L(1) 4 − 4 8 12 − 12 − = (C log 4 4 log η(i)). − − Now π/12 ∞ 2πn η(i) = e− (1 e− ). − n=1
2π ∞ nπ We have 6− < 1/400 and hence e− is rapidly convergent. In other words 76 n=1 ∞ 2nπ 2 4 (1 e− ) is absolutely convergent and upto an error of 10− , it can be n=1 − replaced by 1. We obtain consequently
L (1) π ′ C log 4 + . L(1) ∼ − 3 2 Class number of the absolute class field of Q( √d)(d < 0).
We shall now apply the method outlined in 57 to determine the class number § of the absolute class field of K0 = Q( √d), with d < 0. But, for the time being, however, let K0 be an arbitrary algebraic number field. The absolute class field K/K0 is, by definition, the largest field K containing K0, which is both abelian and unramified over K0 (i.e. with relative discriminant over K0 equal to (1)). The absolute class field, first defined by Hilbert, is also known as the Hilbert class field; its existence and uniqueness were proved by Furtwangler.¨ It can be shown that there are only finitely many abelian unramified extensions of K0 and all these are contained in one abelian extension and this is the maximal one. It has further the property that the Galois group G(K/K0) is isomorphic to the narrow ideal class group of K0. Applications to Algebraic Number Theory 70
Consider now any abelian unramified extension K1 of K0. Then G(K1/K0) is isomorphic to a factor group of G(K/K0), i.e. to a factor group of the narrow ideal class group of K0. The character group of G(K1/K0) is a subgroup of the group of h0 ideal class characters in the narrow sense. K
K1
K0
Q
We have then for the zeta function, 77
ζK1 (s) = L(s, χ), χ the product on the right running over all characters in this subgroup. We shall now derive an expression for the quotient H/h of the class num- bers H of K and h of K0 in the case K0 = Q( √d) with d < 0, by using the above product formula for K1 = K and comparing the residue at s = 1 on both sides. We have indeed ζ (s) = ′ L(s, χ) ζ (s), K1 d χ the product ′ running over all characters χ , 1. On comparing the residues at s = 1 on both sides, we obtain
2r1 (2π)r2 R 2πh ′ H = L(1, χ). (73) W √ D w √ d | | | | Here r1 and r2 denote the number of real and distinct complex conjugates of the field K1. The regulator of K1 is denoted by R and is defined as follows: By g0 g1 gr a theorem of Dirichlet, every unit ǫ of K1 is of the form ǫ = ǫ0 ǫ1 ...ǫr with rational integers gi, ǫ0 being a root of unity. Then ǫ1,...,ǫr (with r = r1 +r2 1) (1) (−r1 are called fundamental units. Define integers ek as follows. If K1 ,..., K )1 (r1+1) (r1+r2) are the real conjugates of K1 and K1 ,..., K1 the complex ones, then we define 1 if k r1 ek = ≤ 2 if r1 < K r. ≤ Applications to Algebraic Number Theory 71
(k) (k) Then R = det(ek log el )(l, k = 1 to r); here ǫl denoting the conjugates of ǫl in the usual order. | | The regulator R is then a real number , 0 and without loss of generality, one may assume R to be positive. It can then be shown that R is independent of the choice of the fundamental units. In the case of the rational number field and an imaginary quadratic field, R is by definition, equal to 1. Further in (73), D denotes the absolute discriminant of K1 and W, the number of roots of unity in K1. We shall now take K1 = K, i.e. the absolute class field. Then K is totally 78 imaginary of degree 2h, i.e. r1 = 0, r2 = h; since K is unramified over K0, D = dh. We shall further assume that d < 4. Then (73) may be rewritten as − (2π)hRH πh 2π = ′ χ(A) log F(A) W d h/2 √ d √ d − | | | | χ | | A or 2RH = h ′ χ(A) log F(A) . (74) W − χ A With every element A of a finite group A ,..., A , we associate an inde- k { 1 h} terminate uA (k = 1,..., h). Then the determinant u 1 is called the group k Ak− Al determinant and was first introduced by Dedekind| and| extensively used by Frobenius. In the case of an abelian group we have then a decomposition of this group determinant, as follows:
uA 1 A = χ(A)uA = uA χ(A)uA . k− l χ A A χ,1 A Let us suppose Ah = E (the identity); we can show by an elementary transfor- mation that
uA 1 A = uA uA 1 A uA 1 (k, l = 1 to h 1). k− l k− t − k− − A Then, from the above, we deduce that
uA 1 A uA 1 = χ(A)uA . k− t − k− χ,1 A Taking u = log F(A), we obtain A − Applications to Algebraic Number Theory 72
1 F(A− ) χ(A) log F(A) = det log k − F(A 1A ) χ,1 A k− l 2 N(bA ) η(zA ) = det log k | k | =∆(say). 2 N(b 1 ) η(Z 1 ) Ak At− Ak At− | | From 74, we gather that 79 2H ∆ = . (75) W h R In other words, ∆/R is a rational number. We shall now discuss the (h 1)- rowed determinant ∆. − Now, bh is a principal ideal = (β ) (say) with β K . Then N(bh = β 2 A 1 l l 0 A 1 l l− ∈ l− | | h/2 so that N(bA 1 ) = βl . Define l− | | 24h η (zE) ρl = (l = 1 to h 1). 12 24h β η z 1 − l Al− We shall now prove that ρl are all units in the absolute class field K. For 1 the same, we first see that ρl depends only upon the class Al− . If b is replaced by b(λ) with λ K , then ∈ 0 24h η (1, zE) ρl = ρl 12h 12h 24h → (β1λ )λ η 1, z 1 − Al− 24 12 since, in the “homogeneous” notation, η (λω1,λω2) = λ− η(ω1,ω2). 1 Now, we pick out a prime ideal p in the class Al− with the property that pp′ = (p), p , p′; such prime ideals always exist in each class (, E) as a consequence of Dirichlet’s theorem. We shall show that the ρl defined with respect to p are units in the absolute class field K. p = B ,..., B p B = Let ( ) 1 r be the decomposition of in K. If we denote degK0 i ν, then r v = h. The ideal pv is principal and equal to (β) (say). Let [ω ,ω ] 1 2 be an integral base of K0 and [ω1∗,ω2∗] an integral base of p. Then we have (ω1∗ω2∗) = (ω1ω2)Pv where Pv is a 2-rowed square matrix of rational integers and Pp = p. Denoting| | by 24 η ((ω ,ω )Pp) ϕ ω = 12 1 2 Pp ( ) p 24 η ((ω1,ω2))
(with ω = ω2/ω1), we know from the theory of complex multiplication that 80 Applications to Algebraic Number Theory 73
ϕ (ω) K and the principal ideal Pp ∈ 12 ϕPp (ω) = (p′ ) (meaning the extended ideal in K). On taking the vth power on both sides,
v 12v v 12 12 ϕPp (ω) = (p′ ) = (p′ ) = (β ) or in other words,
v 12 ϕPp (ω) = β ǫ with ǫ, a unit in K. Writing out explicitly, we have, finally, since β 2 = Npv = pv, | | 12 24v β η ((ω ω )Pp) 1 2 = ǫ, . 24v a unit in K η ((ω1ω2)) The same also holds for a suitable unit ǫ, also with h instead of v since v h. (See references (2) Deuring, specially pp. 32-33, (4) Fricke, (5) Fueter). | If σ : α(h) α(k)(k = 1 to h) denotes the automorphism of K/K cor- → 0 responding to the class Ak under the isomorphism between G(K/K0) and the ideal class group of K0, it can be shown that
12h 2 (k) N(bAk ) η(zAk ) ρl = | | (k = 1 to h 1). | | 2 − N(bA A 1 ) η(zA A 1 ) k l− | k l− | We may then rewrite (75) as
(k) h 1 h 2 h det(log ρl ) 12 − h − 2 H = W | | (k, l = 1 to h 1), (76) det(log ǫ(k) ) − | l | (k) (k) where ǫ1,...,ǫh 1 are a system of fundamental units and ǫ , ρ denote the − l l conjugates of ǫl and ρl(l = 1 to h 1) respectively, the conjugates being chosen in the manner indicated above. Since− the number on the left is strictly positive, the (h 1) units ρ1,...,ρh 1 are independent, i.e. they have no relation between − − them. Hence the group generated by these units is of finite index in the whole 81 unit group and the index is precisely given by the integer on the left side of (76). We have thus proved Applications to Algebraic Number Theory 74
Theorem 7. Let K be the absolute class field of an imaginary quadratic field K0 = Q( √d), d < 4, H, h the class numbers of K and K0 respectively, R the regulator of K and− W the number of roots of unity in K. Then for H we have the formula H W ∆ = h 2 R where ∆ = (log ρ(k) 1/12h) ,l,k = 1, 2,..., h 1 and ρ(k) are conjugates of h 1 | | l | | − l − independent units ρ1,...,ρh 1 in K, as found above. − Example. We shall take d = 5, i.e. K = Q( √ 5). − 0 − The class number h of K0 is 2. The absolute class field K is then a bi- quadratic field and is given by K ( √ 1) = Q( √ 5, √ 1). Further, 0 − − − Q( √5, √ 1) = K −
2 Q( √ 5) = K − 0
2 Q
W = 4 and the fundamental unit is ǫ = (1 + √5)/2. We take A1, A2(= E) 1 to be the two ideal classes of K0 and the unit ǫ1 = ǫ− . The bases for the two √ √ ideal representatives are given by bA1 = [1, (1 + 5)/2] and bA2 = [1, 5]. (Refer to the example on page 71. Then from the above,− we have −
12 2 η( √ 5) 4 ρ = | − | 1 4 | | 1+ √ 5 η − 2 and from (69), we have then ρ = ǫ 12. From (76), we gather then that | 1| | 1| log( ρ ) 12.22 H = 4 | 1| = 4.12 or H = 1. log( ǫ ) | 1| 82 We shall consider, later, class numbers of more general relative abelian extensions (which are not necessarily unramified) of imaginary quadratic fields by using Kronecker’s second limit formula. Applications to Algebraic Number Theory 75
3 The Kronecker Limit Formula for real quadratic fields and its applications
In the last section, we applied Kronecker’s first limit formula to determine the constant term in the expansion at s = 1 of the zeta-function ζ(s, A) associated with an ideal class A of an imaginary quadratic field over Q. In what follows, we shall consider first, a similar problem for a real quadratic field K0 = Q( √d), d > 0. This problem which is more difficult, was solved by Hecke in his paper, “Uber¨ die Kroneckersche Grenzformel fur¨ reelle quadratische Korper¨ und die Klassenzahl relativ-abelscher Korper”.¨ We shall, however, follow a method slightly different from Hecke’s. We start from the series
∞ s 2s f (z, s) = y ′ m + nz − , −| | m ,n= −∞ with z = x + iy, y > 0 and s = σ + it, σ > 1. It is easy to verify that f (z, s) is a ‘non-analytic modular function’, i.e. for a modular transformation z z∗ = (αz + β)/(γz + δ), we have f (z∗, s) = f (z, s). →We now make a remark which will not be used later, but which is interesting in itself, namely, f (z, s) satisfies the partial differential equation
y2∆ f = s(s 1) f (77) − where ∆ = (∂2/∂x2) + (∂2/∂y2) is the Laplace operator. For proving this, let us first observe that if F(z, w) is a complex-valued function, twice differentiable in z and w and if (z, w) (z∗, w∗) where z∗(αz + 1 1 → β)(γz + δ)− , w∗ = (αw + β)(γw + δ)− with α, β, γ, δ real and αδ βγ = 0, then 83 it can be shown directly by computation that −
2 2 2 ∂ (F(z, w)) 2 ∂ F(z∗, w∗) (z w) = (z∗ w∗) . − ∂z∂w − ∂z∗∂w∗ Setting w = z and αδ βγ > 0, we see that y2∆ is an operator invariant under the transformation z z−. Now consider the function ys. It satisfies the equation → ∗ y2∆(ys) = s(s 1)ys. − If we use the invariance property of y2∆, then we see that
2 s s y ∆(y∗ ) = s(s 1)y∗ . − Applications to Algebraic Number Theory 76
Since y = (αδ βγ)y γz + δ 2, we have ∗ − | |− 2 s 2s s 2s y ∆(y γz + δ − ) = s(s 1)y γz + δ − | | − | | where γ, δ are real numbers not both zero. It is now an immediate consequence that f (z, s) satisfies the equation (77). s 2s 2πi(mu+nv) In fact, it also follows that the series y ′ m + nz − e satisfies the (m,n) | | differential equation (77). We shall now derive some properties of a function F which satisfies the equation (77). Suppose the function F itself can be written in the form,
F = (F 1)/(s 1) + F0 + F1(s 1) + ; − − − then on setting s(s 1)F = (s 1)2F + (s 1)F in (77) = and by comparing coefficients, we obtain− the recurrence− − relation−
2 y ∆Fn = Fn 1 + Fn 2, n = 0, 1, 2,..., − − where F 2 = 0. Now,− consider the function f (z, s). From Kronecker’s first limit formula, 2 F 1 = π and from the above recurrence relation, we obtain y ∆F0 = π. If we − 2 write F0 = π log y + G, then y ∆G = 0. In other words, G = F0 + π log y is a potential function.− This is also explained by the fact that
G = 2π(C log 2 log η(z) 2). − − | | 84 s 2s 2πi(mu+nv) For the function F(z, s) = y ′m,n m + nz − e , from Kronecker’s 2| | second limit formula, F 1 = 0 so that y ∆F0 = 0, i.e. F0 is a potential function, − which can also be seen directly from the expression for F0. These remarks stand in connection with the work of Maass and Selberg on Harmonic analysis. α β Now, let γ δ be the matrix of a hyperbolic substitution having two real fixed points ω , ω′ (ω , ω′), both being finite. We shall assume without loss of generality, that ω <ω. Set u = (z ω)(z ω ) 1. Then u = (z ω)(z ω ) 1. ′ − − ′ − ∗ ∗ − ∗ − ′ − The transformation z z∗ corresponds to the transformation u u∗ = λu with λ, a positive real constant→ , 1. We can also assume without loss→ of generality that λ > 1 (otherwise, we may take the inverse substitution). If we introduce a v new variable v,by defining u = λ , then the transformation u u∗ = λu goes over to v v = v + 1. → → ∗ Applications to Algebraic Number Theory 77
Consider a non-analytic function f of z, invariant under the substitution z z = (αz + β)(γz + δ) 1. Then f considered as a function of v, has period → ∗ − 1. If v = v1 + iv2, then f (z) = g(v1, v2) has period 1 in v1. log u Now u = λv1 or v = | | so that e2πiv1 = u 2πi/ log λ. If g(v , v ) possesses | | 1 log λ | | 1 2 a valid Fourier expansion with respect to v1, it is of the form
∞ 2πinv1 f (z) = g(v1, v2) = cn(v2)e n = −∞ ∞ u 2πin/ log λ = c∗ u (78) n u | | n = −∞ log u/u since v = . In general, the Fourier coefficients c in (78), are not 2 2i log λ n∗ constants but if u has a constant argument, i.e. if u/u is a constant, then cn∗ are constants. We shall compute the Fourier coefficients cn∗ of f (z, s) in the case when ω, ω′ come from a real quadratic field K0 = Q( √d) with discriminant d > 0. It is interesting to see that upto certain factors, the Fourier coefficients are just 85 Hecke’s “zeta-functions with Gr¨ossencharacters” ssociated with K0. Suppose ǫ is a non-trivual (, 1) unit in K . Let [ω, 1] be an integral basis ± 0 of an ideal b in K0. Without loss of generality, we may suppose that ω>ω′ (otherwise ω has this property!). Now (ǫ)−b = b so that ǫω = αω + β, ǫ = γω + δ with rational integers α, β, γ, δ. We may also write
α β ω ω ω ω ǫ 0 ′ = ′ γ δ 1 1 1 1 0 ǫ′ It follows then that αδ βγ = ǫǫ = 1. In the case αδ βγ = 1, define u − ′ ± − ωu + ω′ αz + β ωu∗ + ω′ such that z = or if z∗ = , then z∗ = ; in other words, u + 1 γz + δ u∗ + 1 z ω z ω u = − ′ and u = ǫ2u. If αδ βγ = N(ǫ) = 1, define u = − ′ and ω z ∗ − − ω z −αz + β − if z = , u = ǫ2u. Since z z is a hyperbolic transformation, the ∗ γz + δ ∗ − → ∗ points ω′, z, z∗, ω lie on the same segment of a circle in the case N(ǫ) = 1. We may assume without loss of generality, that z∗ lies to the right of z, or otherwise, 2 we can take the reciprocal substitution. We have then ǫ > 1, since ǫ>ǫ′, as a consequence of our assumption that z∗ lies to the right of z ǫ itself might be positive or negative. If ǫ is negative, we take instead of α, β, γ, δ, the integers Applications to Algebraic Number Theory 78
α, β, γ, δ so that we secure finally that ǫ > 1. In the second case, when −N(ǫ)−= −1, if− z lies on the small segment ω λω (see figure), z would lie on − ′ ∗ the big segment ω′ ω which is the complement of the small segment obtained by reflecting the original one on ω′ω. If we take z on the semicricle ω′vω on ω′ω as diameter, then z and z∗ would lie on the same segment. So, in this case, under the same argument, ǫ > 1.
In either case, ǫ > 1 and consequently ǫ′ < 1 so that ǫ ǫ′ > 0. But 86 ǫ ǫ = γ(ω ω ) implies that γ > 0. − − ′ − ′ Now, if z lies on ω′vω, u is purely imaginary with arg u = π/2. On setting 2 u = u′i, we obtain the transformation u′ u′∗ = ǫ u′; u′ is real and positive. Hereafter we need not distinguish between→ the two cases. 2v If we define v such that u = ǫ (writing u instead of u′) then u u∗ s 2s → becomes v v + 1. The function f (z, s) = y ′m,n m + nz − , is invariant → 1 | | under the modular substitution z (αz + β)(γz + δ)− and hence as a function of v, it has period 1. It has a valid→ Fourier expansion in v of the form f (z, s) = ∞ 2πikv ake . Now k= −∞ 1 2πikv ak = f (z, s)e− dv 0 ǫ2 1 πik/ log ǫ du = f (z, s)u− . 2 log ǫ 1 u On changing the variable from u to z, the series f (z, s) being uniformly convergent on the corresponding segment of ω′vω, we may interchange inte- gration and summation and obtain
ǫ2 s 1 y πik/ log ǫ du ak = u− . 2 log ǫ m + nz 2s u m,n 1 | | Applications to Algebraic Number Theory 79
ωui + ω From the relation z = ′ it follows that ui + 1 u(ω ω ) (m + nω)ui + (m + nω ) y = − ′ and m + nz = ′ u2 + 1 ui + 1 Setting β = m + nω then β b since m and n are both rational integers and ω ω = N(b) √d. Then our∈ integral becomes − ′ ǫ2 s 1 s s/2 u πik/ log ǫ du ak = (N(b)) d u− . 2 log ǫ 2 2 + 2 u b β,0 1 β u β′ | To bring the right hand-side to proper shape again, we use an idea of Hecke. We shall now get rid of β in the integrand by introducing U = u β/β′ . The 87 integral then reduces to | |
β πik/ log ǫ (βǫ/β′ǫ′) us (πik/ log ǫ) du β s | | − , N( ) − 2 s β′ | | β/β (u + 1) u | ′| on writing u instead of U. For two elements β, γ, , 0, (β) = (γ) implies that γ = βǫn, n = 0, 1, 2,... with ǫ > 1, being the fundamental unit in Q( √d). Then± ± ± n n n γ βǫ γ = βǫ , γ′ = β′ǫ′ and = n . | | | | | | | | γ′ β′ǫ′ We have therefore, on taking into account γ = βǫn of γ = βǫn, − (βǫ/β′ǫ′) s (πik/ log ǫ) | | u − du (u2 + 1)s u b β,0 β/β′ | | | (βǫn+1/β ǫ n+1) ∞ ′ ′ us (πik/ log ǫ) du = | | − . 2 2 s n n (u + 1) u b (β),0 n = (βǫ /β′ǫ′ ) | −∞ | | 2n 2(n+1) Now since ǫ > 1, the intervals ( β/β′) ǫ , β/β′ ǫ ) fill out the half- line (0, ) exactly once, without gaps| and overlaps,| | as n tends to on one side and∞+ on the other side. Hence we have −∞ ∞ (βǫn+1/β ǫ n+1) ∞ ′ ′ us (πik/ log ǫ) du us (πik/ log ǫ) du | | − = ∞ − , 2 2 s 2 2 s n n (u + 1) u (u + 1) u n = (βǫ /β′ǫ′ ) 0 −∞ | | and s πik s πik Γ Γ + us (πik/ log ǫ) du 2 − 2 log ǫ 2 2 log ǫ ∞ − = 2 2 s 0 (u + 1) u Γ(s) Applications to Algebraic Number Theory 80
Therefore, we have
1 2πikv ak = f (z, s)e− dv 0 s πik s πik Γ Γ + 1 2 − 2 log ǫ 2 2 log ǫ = (N(b))sds/2 2 log ǫ Γ(s) × πik/ log ǫ β s N(β) − , × β | | b (β),0 ′ | since the expression under the summation Σ does not change for associated 88 elements β and γ. Upto a product of Γ-factors, the Fourier coefficients ak are Dirichlet series. For k = 0,
2 s 1 Γ 1 2 s/2 s a = f (z, s)dv = d (N(a))− 0 2 log ǫ Γ(s) 0 a A s ∈ Γ2 1 = 2 ds/2ζ(s, A), 2 log ǫ Γ(s)
1 where A denotes the ideal class of b− in the wide sense. If k , 0, we define for σ > 1,
s ζ(s, χ, A) = χ(a)(N(a))− a A ∈ 1 s = (χ(b))− χ((β))(N(a))− ; a A ab=∈(β) The function ζ(s, χ, A) is called the zeta-function of the class A and associated with the Grossencharacter¨ χ, which is defined as follows: The character χ is defined on principal ideals (β) of K0 as β πik/ log ǫ χ((β)) = β′ (χ((β)) is independent of the generator β by definition). Then χ is extended to all ideals i as follows: If ik = (v), define χ(i) as a hth root of χ((v)) so that v πik/ log ǫ χ(ih) = χ((v)) = v′ Applications to Algebraic Number Theory 81
It is clear that χ is multiplicative. 89 We have now s χ(b)ζ(s, χ, A) = χ((β))(N(a))− a A ab=(β∈),(0) s s = (N(b)) χ((β)) N(β) − , | | b ( β),(0) | so that we may rewrite the expression for ak as follows:
s πik s πik Γ Γ + 2 log ǫ s/2 2 − 2 log ǫ 2 2 log ǫ a = χ(b)ζ(s, χ, A). k d Γ(s) One can make some applications from the nature of the Fourier coefficients ak. We know that the function f (z, s) satisfies the following functional equa- tion, viz. s (1 s) π− Γ(s) f (z, s) = π− − Γ(1 s) f (z, 1 s). − − One can then show that from the analytic continuation of f (z, s) it follows that the Fourier coefficients ak as functions of s have also analytic continua- tions into the whole s-plane and satisfy a functional equation similar to the above. In the particular case, when k = 0, it follows that
s s/2 2 s (1 s) (1 s)/2 2 1 s π− d Γ ζ(s, A) = π− − d − Γ − ζ(1 s, A). 2 2 −
Hence, for the zeta-function ζk0 (s) = ζ(s, A), we have A
s s/2 2 s (1 s) (1 s)/2 2 1 s π− d Γ ζK0 (s) = π− − d − Γ − ζK0 (1 s). 2 2 − This was discovered first, by Hecke, who also introduced the zeta-functions with Grossencharacters¨ and derived a functional equation for the same. We shall now use Kronecker’s first limit formula to study the behaviour of 90 ζ(s, A) at s = 1. From Kronecker’s first limit formula, we have π f (z, s) = + 2π(C log 2 log √y η(z) 2 + ). s 1 − − | | − Applications to Algebraic Number Theory 82
It can be shown that the series on the right can be integrated term by term with respect to v (after transforming z into v) in the interval (0, 1), i.e. s Γ2 1 1 f (z, s)dv = 2 ds/2ζ(s, A) 0 2 log ǫ Γ(s) π 1 = + 2π(C log 2) 2π log( √y η(z) 2)dv + . s 1 − − | | − 0 It follows therefore that the function on the left has a pole at s = 1 with residue π and the constant term in the expansion is provided by the integral which cannot in general be computed. It is to be noted here that in the case of the imaginary quadratic field, we had only the integrand on the right side and the argument z was an element of the field, but here z is a (complex) variable. One cannot get rid of the integral even if one uses the series expansion for the integrand. The above formula was found by Hecke and is the Kronecker limit for- mula for a real quadratic field. One can also consider ak for k , 0 and obtain a similar formula. Changing this integral to a contour integral, we shall later obtain an ana- logue of Kronecker’s solution of Pell’s equation in terms of elliptic functions. We have now, for real v, 1 v+1 a0 = f (z, s)dv = f (z, s)dv 0 v s Γ2 1 = 2 ds/2ζ(s, A). 2 log ǫ Γ(s) Using the Legendre formula 91
s s + 1 1 s Γ Γ = √π2 − Γ(s), 2 2 we obtain s + 1 s + 1 Γ2 Γ2 Γ(s) Γ(s) 2 2 = = . s s s + 1 π22(1 s)Γ(s) Γ2 Γ2 Γ2 − 2 2 2 But on the other hand, s + 1 Γ2 2 = 1 + terms in (s 1)2, Γ(s) − Applications to Algebraic Number Theory 83 so that we have Γ(s) v+1 = s/2 ζ(s, A) 2 log ǫd− s f (z, s)dv Γ2 v 2 v+1 2 log ǫ 2 1 (s 1) 2 = (2 d− 2 ) − (1 + terms in (s 1) ) f (z, s)dv. π √d − v
2 1 (s 1) (s 1)(2 log 2 2 log √4 d) By setting (2 d− 2 ) − = e − − and expanding it in powers of (s 1), and applying Kronecker’s first limit formula for f (z, s), we obtain finally,−
2 log ǫ 4 ζ(s, A) = (1 + (s 1)(2 log 2 2 log √d) + ) π √d − − ×
1 4 + 2C (2 log 2 2 log √d) × s 1 − − − −v+1 4 2 log( √y√d η(z) 2)dv + , − v | | i.e. v+1 2 log ǫ 1 4 ζ(s, A) = + 2C 2 log( √y√d η(z) 2)dv + . (79) √d s 1 − v | | − 2 log ǫ From this, we deduce that ζ(s, A) has a pole at s = 1 with residue which 92 √d is independent of the ideal class A. This result was first discovered by Dirichlet. It would be nice if one could compute the integral and express it in terms of an analytic function, but it looks impossible. We shall only simplify it to a certain extent by converting it into a contour integral. z ω From the substitution ui = − ′ follows that ω z − z ω y(ω ω ) u = Im − ′ = − ′ ω z (z ω)(z ω) − − − and similarly 1 y(ω ω′) u− = − . (z ω )(z ω ) − ′ − ′ On multiplying the two, we have y2 1 = . (80) (z ω)(z ω ) (z ω)(z ω ) − − ′ − − ′ ω ω ω ω − ′ − ′ Applications to Algebraic Number Theory 84
Consider now the expression
(z ω)(z ω ) F(z) = − − ′ = az2 + bz + c. (81) N(b)
We then claim that a, b, c are rational with (a, b, c) = 1 and b2 4ac = d. From Kronecker’s generalization of Gauss’s theorem on the content− of a product of polynomials with algebraic numbers as coefficients, we have for the product (ξ ωη)(ξ ω η), (1, ω)(1, ω ) = (1, (ω + ω ),ωω ) or in other words, − − ′ − − ′ − ′ ′ bb′ = (1,λ, ) where λ and are rational, i.e.
1 λ , , = 1. N(b) N(b) N(b) 1 λ But, from our definition of F(z), a = , b = , c = so that N(b) N(b) N(b) (a, b, c) = 1. The discriminant of F(z) is
(ω + ω )2 4ωω (ω ω )2 b2 4ac = ′ − ′ = − ′ = d. − N(b)2 N(b)2 1 Further a = > 0. We can then show that the class of the quadratic form 93 N(b) F(z) is uniquely determined by the ideal class A. 4 From (80) it follows that y2d = F(z) F(z) or equivalently, √y√d = 4 F(z) F(z). Now, for the integral in (79), we have
v+1 z∗ 4 log ǫ 4 dz − log( √u√d η(z) 2)dv = 2 log( 4 F(z)η(z) 2) , | | | | F(z) √d v z since, by definition of v, √d dz dv = − 2 log ǫ F(z) and the transformation v v + 1 corresponds to z z∗ on the segment of the orthogonal circle. On the→ right side, z may be taken→ arbitrarily on the segment. Thus, let A be an ideal class, in the wide sense, of K0 = Q( √d), d > 0 and b an ideal in A with an integral basis [1,ω], ω>ω′. Corresponding to b, let α β be defined as on p. 85 and let for z h, z = (αz + β)(γz + δ) 1. Let, γ δ ∈ ∗ − further, ǫ > 1 be the fundamental unit in K0. Then we have Applications to Algebraic Number Theory 85
Theorem 8 (Hecke). For the zeta-function ζ(s, A) associated with the class A, we have the limit formula
2 log ǫ 1 4C log ǫ z∗ dz lim ζ(s, A) = + 2 log( 4 F(z)η(z) 2) , s 1 − √d s 1 √d z | | F(z) → − where the integration is over a segment zz∗ of the semicircle on ω′ω as diameter and F(z) is defined by (81).
Now, the question arises whether one could transform the integral in such a 94 way that it is independent of the choice of z and also the path of integration. It is possible to do that, by the method of Herglotz who reduced this to an integral involving simpler functins than η(z). If N(ǫ) = 1, then the transformation v v + 1 is equivalent to z z∗ = αz + β → → with α β , a modular matrix. γz + δ γ δ If N(ǫ) = 1, on replacing ǫ by ǫ2, N(ǫ2) = 1 and the transformation − α z + β v v + 1 goes over to the transformation z ′ ′ with α′ β′ a modular γ′ δ′ → → γ′z + δ′ matrix. Further, because of the periodicity of f (z, s),
v+2 v+1 f (z, s)dv = 2 f (z, s)dv v v so that we may assume without loss of generality that N(ǫ) = 1. Now, the behaviour of √4 F(z) η(z) under a modular substitution can be studied. We know that η(z∗) = ρ √γz + δη(z) where ρ is a 24th root of unity. Further, (z ω)(z ω′) (z∗ ω)(z∗ ω′) = − − − − (γz + δ)2 if one uses the fact that ω = ω∗ and γω + δ = ǫ. From these two, it follows that
√4 F(z) 4 F(z ) = κ ∗ + √γz δ with κ, a 4th root of unity. Therefore
4 4 F(z∗)η(z∗) = ρκ F(z)η(z). (82) (We emphasize here that (82) holds only for a hyperbolic substitution and a power of the same, since we have made essential use of the fact that ω and ω′ Applications to Algebraic Number Theory 86 are fixed points of the substitution in obtaining the transformation formula for F(z).) Consider the function √4 F(z)η(z). It is an analytic function having no ze- ros in the upper half plane so that on choosing a single valued branch of 95 log(√4 F(z)η(z)), we have a regular function in the upper z-half plane. Let us denote log √4 F(z)η(z) by g(v). Then (82) implies that g(v+1) g(v) = 2πiλ (say) with λ rational. One can then express λ by means of the so-called− Dedekind sums. Setting g(v) 2πiλv = h(v), we have h(v + 1) = h(v) possesses − 2πiv ∞ 2πinv a Fourier development in e , or in other words, h(v) = cne , where n= −∞ v+1 = cn are constants, since h(v) is an analytic function of v and c0 v h(v)dv. Here v lies in a certain strip enclosing the real axis. + = v 1 Now, consider the complex integral c0 v h(v)dv. Here v is, in general, complex and the path of integration may be any curve between v and v+1 lying in the strip in which the Fourier expansion is vaid. We have
√d z∗ dz v+1 log( 4 F(z)η(z) 2) = 2 Re (g(v))dv (v real) −2 log ǫ | | F(z) z v v+1 = 2 Re h(v)dv v
= 2 Re(c0).
The computation of the integral on the left side therefore reduces to the de- termination of c0, which in turn is independent of v and the path of integration. The trick for computing the integral is to convert it into an infinite integral by letting z and z∗ α/γ. Expanding g(v) in an infinite series, one can express this→ ∞infinite integral→ as a definite integral of an elementary function. (See G. Herglotz.) We shall now obtain an analogue of Kronecker’s solution of Pell’s equation in this case. Let us define
z∗ dz g(A) = 2 log( 4 F(z)η(z) 2) . (83) | | F(z) z Then, associated with a character χ of the ideal class group (in the wide sense), we have L(s, χ) = χ(A)ζ(s, A), A the summation running over all wide classes A. 96 Applications to Algebraic Number Theory 87
From (79), we obtain, for χ , 1, the expansion,
L(s, χ) = χ(A)g(A) + terms involving (s 1). (84) − A Before proceeding to derive the analogue of Kronecker’s solution of Pell’s equation from the above, we shall examine under what conditions, a genus character, which is a character of the narrow class group of Q( √d), is also a character of the wide class group. Let χ be a genus character, defined as follows: If d = d1d2, then for all ideals a with (a, d1) = (1), χ is defined by d1 χ(a) = . (We may suppose without loss of generality that d1 is odd). N(a) Now, in general, χ is not a character of the wide class group. It will be so, if χ((α)) = 1 for all integers α with (α, d1) = (1). For elements α with N(α) > 0, this is true by the definition of χ. If N(α) < 0, then d d χ((α)) = 1 = 1 = 1 N((α)) N(α) − for all α, if it is true for one such α. d Take α = 1 + √d so that N(α) = 1 d < 0. We need examine only 1 , − d 1 − d1 being odd. Then
d d 1 1 +1 if d1 > 0 1 = − = − = d 1 d1 d1 1 if d1 < 0. − − Since d is positive, either both d1, d2 are positive or both are negative. In the former case, χ continues to be a character of the wide class group and in the latter case, it is no longer a character of the wide class group. Case (i). Let us assume, that both d1 and d2 are positive. Then the genus character χ as defined above, is also a wide class character. We have a decomposition of L(s, χ) due to Kronecker, from (66) as follows:
L(s, χ) = Ld1 (s)Ld2 (s) and on taking the values on both sides at s = 1, we obtain 97
L(1, χ) = Ld1 (1)Ld2 (1). From (84), we have now L(1, χ) = χ(A)g(A), summation running over A all wide classes A. Further 2 log ǫ1h1 Ld1 (1) = √d1 Applications to Algebraic Number Theory 88
if ǫ1 denotes the fundamental unit of Q( √d1) and h1, the wide class number of
Q( √d1) and similarly for Ld2 (1). Thus as an analogue of Kronecker’s solution of Pell’s equation as derived in (68), we have the following Proposition 13. Let χ be a genus character of Q( √d), d > 0, corresponding to the decomposition d = d1d2(d1 > 0, d2 > 0) of d. Let h1, h2 be the class numbers and ǫ1, ǫ2 the fundamental units of Q( √d1), Q( √d2) respectively. Then 4h1h2 log ǫ1 log ǫ2 = √d χ(A)g(A), (85) A where A runs over all the ideal classes of Q( √d) in the wide sense and g(A) is defined by (83). The expression on the right side of (85) is not very simple. It is not known whether the number on the left side is rational or irrational. It is probable that the number on the left side is a complicated transcendental number, so that one cannot expect a simple value on the right side. Case (ii). Suppose d1 and d2 are both negative. Then again from the de- composition formula (66), we have
2 2πh1 2πh2 4π h1h2 L(1, χ) = Ld (1)Ld (1) = = 1 2 w √ d w √ d w w √d 1 | 1| 2 | 2| 1 2 where h1, h2 denote the wide class numbers of Q( √d1) and Q( √d2), and w1, w2 the number of roots of unity in Q( √d1) and Q( √d2) respectively. We shall see in 5 that 98 § π2 L(1, χ) = χ(B)G(B), √d B
B running over all narrow classes of Q( √d) and G(B) being numbers depend- ing only on B. We shall further prove that G(B) are rational numbers which can be realized in terms of periods of certain abelian integrals of the third kind. We then have, as an analogue of Kronecker’s solution of Pell’s equation, the following: 4h h 1 2 = χ(B)G(B). w w 1 2 B
Example. Consider the field Q( √10) with discriminant d = 40 = 5.8; d1 = 5 and d2 = 8. Applications to Algebraic Number Theory 89
The class number of Q( √10) is 2. Since the fundamental unit ǫ = 3 + √10 has norm 1, the narrow classes are the same as wide classes. We can now choose the− two ideal class representatives as follows
1 1 (1) = b1 = [1, √10] and b2 = 1, √10 with N(b2) = . 2 2
The class numbers h1 and h2 of Q( √5) and Q( √8) are both 1 and the funda- 1 + √5 mental units are respectively and 1 + √2. The only character χ , 1 2 has the property that χ(b1) = 1 and χ(b2) = 1, so that we have from (85), the following: −
1 + √5 2 log log(1 + √2) = √10(g(E) g(A)). 2 − We shall make a remark for more general applications. Consider the abso- lute class field of Q( √d) with d > 0. For computing the class number of this field, one can proceed in the same way, as in the case of an imaginary quadratic field. For the same, one requires the computation of L(1, χ) for arbitrary narrow class characters. This will be done in 5, for a special type of characters. § 4 Ray class fields over Q( √d), d < 0 99 Let K be an algebraic number field of degree n over Q, the field of rational numbers. Let r1 and 2r2 be the number of real and complex conjugates of K (1) (r1) respectively, so that r1 + 2r2 = n. Let K ,..., K be the real conjugates of K and K(r1+1),..., K(n) be the complex conjugates of K. Let, for α K, α(i) K(i), i = 1, 2,..., n, denote the conjugates of α. Further let f be a given∈ ∈ α1 α2 integral ideal in K. Two numbers γ1 = , γ2 = with α1, β1, α2, β2 integral β1 β2 in K and with β1β2 coprime to f are (multiplicatively) congruent modulo f (in symbols, γ1 γ2( mod ∗f) if α1β2 α2β1( mod f)). If γ1, γ2 are integers in K, this is the≡ usual congruence modulo≡ f. b Let us consider non-zero fractional ideals a of the form a = where b and c c are integral ideals in K coprime to f. These fractional ideals a form, under the usual multiplication of ideals, an abelian group which we shall denote by Gf. Let Gf be the subgroup of Gf consisting of all principal ideals (α) for which α > 0 (i.e. α(i) > 0, for i = 1, 2,..., r ) and α 1( mod f). It is clear that if 1 ≡ ∗ (γ) Gf and γ 1( mod f), then γ might be written as α/β where α and β ∈ ≡ ∗ Applications to Algebraic Number Theory 90 are integers in K satisfying the conditions, α > 0, β > 0, α 1( mod f) and β 1( mod f). ≡ ≡ The quotient group Gf/Gf is a finite abelian group called the ray class group modulo f; its elements are called ray classes modulo f. Two ideals a and b in Gf are equivalent modulo f(a b mod f) if they lie in the same ray class modulo ∼ f, i.e. a = (γ)b with (γ) Gf. ∈ If r1 = 0 and f = (1), equivalence modulo f is the usual equivalence in the wide sense and the ray class group modulo f is the class group of K in the wide sense. If r1 = 2, r2 = 0 (i.e. K is a real quadratic field) and f = (1), then the equivalence modulo f is the equivalence in the narrow sense and the ray class group is the class group of K in the narrow sense. Let χ be a character of the group of ray classes modulo f. Then associated with χ, we define for σ > 1, the L-series s L(s, χ) = χ(a)(N(a))− a,0 where N(a) is the norm of a in K and the summation is extended over all inte- gral ideals a coprime to f. It is clear that L(s, χ) is a regular function of s for 100 σ > 1. Moreover, due to the multiplicative character of χ, we have for σ > 1, an Euler-product decomposition for L(s, χ), namely
s 1 L(s, χ) = (1 χ(p)(N(p))− )− − p f | where the infinite product is extended over all prime ideals p coprime to f. Hecke has shown that L(s, χ) can be continued analytically as a meromor- phic function of s in the whole s-plane and that when χ is a “proper” character, there is a functional equation relating L(s, χ) with L(1 s, χ), where χ is the conjugate character. If χ , 1 (the principal character), then− L(s, χ) is an entire function of s. If χ = 1 and f = (1), then L(s, χ) is the Dedekind zeta function of K. We are interested in determining the value at s = 1 of L(s, χ), in the case when K is a real or imaginary quadratic number field and χ is not the principal character. For this purpose, we need to apply Kronecker’s second limit for- mula. Later, we shall use this for the determination of the class number of the “ray class field” of K, corresponding to the ideal f. First we shall investigate the structure of a ray class character χ. Let α and α2 β(, 0) be integers coprime to f such that α = β( mod f). Then γ = satisfies β2 2 2 γ = 1( mod ∗f) and γ > 0 so that χ((γ)) = 1 i.e. χ((α )) = χ((β )). In other words, χ((α)) = χ((β)) or χ((α/β)) = 1. ± ± Applications to Algebraic Number Theory 91
Let L be the multiplicative group of λ , 0 in K. Corresponding to a ray class character χ modulo f, we define a character v of L as follows. For λ L, we find an integer α K such that α 1( mod f) and αλ > 0 and define∈ v(λ) = χ((α)). It is first∈ clear that v(λ)≡ is well-defined; for, if v(λ) = χ((β)) for another integer β K satisfying the same conditions, then we see at once α ∈ α that 1( mod f) and > 0 and so χ((α)) = χ((β)). It is easily verified β ≡ ∗ β that v(λ ) = v(λ)v( ). Moreover, v(λ) = 1; for, λ2 > 0 and by definition, ± v(λ2) = χ((α)) for α satisfying α 1( mod f) and α > 0 so that (v(λ))2 = 101 v(λ2) = χ((α)) = +1. We shall call v≡(λ), a character of signature. Consider the subgroup B of L consisting of λ > 0. Clearly v(λ) = 1 for all λ B. Thus v(λ) may be regarded as a character of the quotient group L/B. ∈ Now L/B is an abelian group of order 2r1 exactly, since one can find λ L for (i) ∈ which the real conjugates λ , i = 1, 2,..., r1, have arbitrarily prescribed signs. Also, there are 2r1 distinct characters u of L/B defined by
r1 λ(i) gi u(λ) = , gi = 0 or 1,λ L. λ(i) ∈ i=1 | | There can indeed be no more that 2r1 characters of L/B and hence v(λ) coin- cides with one of these characters u, of L/B. We do not however assert that every character u of L/B is realizable from a ray class character modulo f, in the manner described above. Let now γ K such that γ 1( mod ∗f) and (γ) Gf. By definition, v(γ) = χ((δ)) for∈ an integer δ such≡ that δ 1( mod ∈f) and δγ > 0. But χ((δ)) = χ((γ)) and hence v(γ) = χ((γ)). Let≡ α and β(, 0) be two integers coprime to f such that α = β( mod f). Then γ = α/β 1( mod f) and ≡ ∗ χ((α)) v(α) = χ((γ)) = v(γ) = . χ((β)) v(β) Thus for integral α(, 0) coprime to f, the ratio χ((α))/v(α) depends only on the residue class of α modulo f and is in fact, a character of the group G(f) of prime residue classes modulo f. We may denote it by χ(α). Thus
Proposition 14. Any character χ((α)) of the ray class group modulo f may be written in the form χ((α)) = v(α)χ(α) where v(α) is a character of signature and χ(α) is a character of the group G(f). Applications to Algebraic Number Theory 92
Before proceeding further, we give a simple illustration of the above. Let us take K to be Q, the field of rational numbers and f to be the principal ideal 102 ( d ) in the ring of rational integers, d being the discriminant of a quadratic field over| | Q. For an integer m coprime to d, we define d m for d > 0 m m χ((m)) = |d| | | for d < 0 m | | d where is the Legendre-Jacobi-Kronecker symbol. The group Gf now m consists| of| fractional ideals (m/n) where m and n are rational integers coprime to d. We extend χ to the ideals (m/n) in Gf by setting χ((m/n)) = χ((m))/χ((n)). Now it is known that if m n( mod d) and mn is coprime to d, then ≡ d d = if d > 0 m n | | | | and d m d n = if d < 0 m m n n | | | | | | | | Thus d χ(m) = for d > 0 m | | and d m χ(m) = for d < 0 m m | | | | m are characters of G( d ). Moreover v(m) = is clearly a character of signa- | | m ture. It is easily verified that χ((m/n)) is a ray| | class character modulo ( d ) and that χ((m)) = v(m)χ(m). | | s Now L(s, χ) = a(a)(N(a))− , where A runs over all the ideal classes A in the wide sense and a over all the non-zero integral ideals in A, which are 1 coprime to f. In the class A− , we can choose an integral ideal bA coprime to f and abA = (β) where β is an integer divisible by bA and coprime to f. 103 Conversely, and principal ideal (β) divisible by bA and coprime to f is of the form abA, where a is an integral ideal in A coprime to f. Moreover, χ(bA)χ(a) = χ((β)) and N(b ) N(a) = N(β) , where N(β) is the norm of β. Thus A | | s s L(s, χ) = χ(b )(N(b )) χ((β)) N(β) − A A | | A b (β) A| Applications to Algebraic Number Theory 93
where the inner summation is over all principal ideals (β) divisible by bA and 1 coprime to f. Since A− runs over all classes in the wide sense when A does so, we may assume that bA is an integral ideal in A coprime to f. Moreover, in view of Proposition 14, we have for σ > 1,
s s L(s, χ) = χ(b )(N(b )) v(β)χ(β) N(β) − (86) A A | | A b (β) A| where now A runs over all the ideal classes of K in the wide sense, bA is a fixed integral ideal in A coprime to f and the inner sum is over all principal ideals (β) divisible by bA and coprime to f. We may extend χ(β) to all residue classes modulo f by setting χ(α) = 0 for α not coprime to f. Thus we may regard the inner sum in (86) as extended over all principal ideals (β) divisible by bA. In order to render the series in (86) suitable for the application of Kronecker’s limit formula, we have to replace χ(β) by an exponential of the form e2πi(mu+nv) occurring in the limit formula. We shall, in the sequel, express χ(β) as an exponential sum by using an idea due to Lagrange, which is as follows. Let x1,..., xn be n distinct roots of a polynomial f (x) of degree n, with th coefficients in a field K0 containing all the n roots of unity. Let, further, the field M = K0(x1,..., xn) be an abelian extension of K0, with galois group H. σi Let us assume moreover that if σ1,...,σn H, then xi = x1 i = 1, 2,..., n. ∈ n Now, if χ is a character of H, then let us define yχ = χ(σi)xi. It is clear that i=1 y M and χ ∈ n n σ j σiσ j σiσ j yχ = χ(σi)x1 = χ(σ j) χ(σiσ j)x1 = χ(σ j)yχ. i=1 i=1 Hence yn K . Now, if χ denotes the conjugate character of χ and σ H, then χ ∈ 0 ∈ n n σiσ σ yχ = χ(σiσ)x1 = χ(σ) χ(σi)xi . i=1 i=1
n , = 1 σ If yχ 0, then χ(σ) y−χ χ(σi)xi . We shall use a similar method to express 104 i=1 χ(β) as an exponential sum whose terms involve β in the exponent. First we need the following facts concerning the different of an algebraic number field M of finite degree over Q. Let for α M, S (α) denote the ∈ trace of α. Let a be an ideal (not necessarily integral) in M and let a∗ be the “complementary” ideal to a, namely the set of λ M, for which S (λα) is a rational integer for all α a. It is known that aa is∈ independent of a and in fact ∈ ∗ Applications to Algebraic Number Theory 94
1 aa∗ = (1)∗ = ϑ− , where ϑ is the different of M. Further clearly (a∗)∗ = a and N(ϑ) = d , where d is the discriminant of M. These facts can be verified in a | | simple way if M is a quadratic field over Q, with discriminant d. Let [α1, α2] be an integral basis of a. The ideal a is the set of λ M for which S (λα ) ∗ ∈ 1 and S (λα2) are both rational integers. If α1′ , α2′ denote the conjugates of α1 α2′ α1′ and α2 respectively, then it is easliy verifed that , − α α α α α α α α 1 2′ − 2 1′ 1 2′ − 2 1′ √ is an integral basis of a∗. Now α1α2′ α2α1′ = N(a) d and this means that − ± 1 a∗ = (1/N(a) √d)a′. Since (N(a)) = aa′, we see that a∗ = a− (1/ √d) i.e. aa∗ = 1 (1/ √d) = ϑ− . Moreover N(ϑ) = d . | | 1 1 Let us consider now the ideal a = f− ϑ− in K; clearly a∗ = f. Let us choose 1 1 in the class of fϑ, an integral ideal q coprime to f. Then qf− ϑ− = (γ) for γ K 1 ∈ i.e. (γ)ϑ = qf− has exact denominator f. If K were a quadratic field and f, a principal ideal, then fϑ is principal and we may take q = (1). We shall consider γ fixed this way once for all, in the sequel. Let us observe that if λ f, S (λγ) is a rational integer. ∈ With a view to express χ(β) as an exponential sum, we now define the sum
T = χ(λ)e2πiS (λγ) (87) λ mod f where λ runs over a full system of representatives of residue classes modulo f; for f = (1), clearly T = 1. We see that T is defined independently of the 105 choice of representatives λ, for, if runs over another system of representatives modulo f, then λ ( mod f) in some order and in this case, χ(λ) = χ( ) and e2πiS (λγ) = e2πiS ( γ≡) since S ((λ )γ) is a rational integer. Moreover, in this sum, λ may be supposed to run only− over representatives of elements of G(f), since χ(α) = 0, for α not coprime to f. The sums of this type were first investigated by Hecke; similar sums for the case of the rational number field have been studied by Gauss, Dirichlet and for complex characters χ, by Hasse. Let α be an integer in K coprime to f. Then αλ runs over a complete set of prime residue classes modulo f when λλ does so. Thus
T = χ(αλ)e2πiS (αλγ) λ mod f = χ(α) χ(λ)e2πiS (αλγ), λ mod f i.e. Tχ(α) = χ(λ)e2πiS (αλγ). (88) λ mod f Applications to Algebraic Number Theory 95
We shall presently see that (88) is true even for α not coprime to f and that T , 0, when χ is a so-called proper character of G(f). Let g be a proper divisor of f and ψ, a character of G(g). We may extend ψ into a character χ of G(f) by setting χ(λ) = ψ(λ) for λ coprime to f and χ(λ) = 0, otherwise. We say that a character χ of G(f) is proper, if it is not derivable in this way from a character of G(g) for a proper divisor g of f. A ray class character χ modulo f is said to be proper if the associated char- acter χ of G(f) is proper; χ is then said to have f as conductor. Let χ be a ray class character modulo f and let g be a proper divisor of f. Moreover let for integers α, β coprime to f such that α β( mod g) and ≡ αβ > 0, χ(α) = χ(β). We can associate with χ, a ray class character χ0 modulo g as follows. If p is a prime ideal coprime to f, define χ0(p) = χ(p). If p is a prime ideal coprime to g but not to f, we can find a number α such that α 1( ≡ mod ∗g), α > 0 and (α)p is coprime to f. We then set χ0(p) = χ((α)p). We see that χ0 is well-defined for all prime ideals coprime to g and we extend χ0 106 multiplicatively to all ideals in the ray classes modulo g. Clearly, χ0 is a ray class character modulo G and for integral α coprime to g, χ0((α)) = χ0(α)v(α), where χ0(α) is the associated character of G(g) and v(α) is the same signature character as the one associated with χ. We now say the ray class character χ modulo f having the property described above with respect to g, has g as conductor, if the associated χ0 has g as conductor. Let from now on, χ be a proper ray class character modulo f. Since χ(α) = 0 for α not coprime to f, all we need to prove (88) for proper χ and for α not coprime to f is to show that the right hand side of (88) is zero. Let b be the greatest common divisor of (α) and 1 f and let g = fb− . Since χ is proper, it is not derivable from any character of G(g). In other words, there exist integers λ and coprime to f such that λ ( mod g) and χ(λ) , χ( ). Otherwise, if for all integers λ and coprime≡ to f and for which λ ( mod g) it is true that χ(λ) = χ( ), then we can define ≡ a character χ0 of G(g) such that for λ coprime to f, χ(λ) = χ0(λ) which is a contradiction. Thus we can find integers and ν coprime to f such that v( mod g) and χ( ) , χ(v). Now clearly ≡
χ( ) χ(λ)e2πiS (αλγ ) 2πiS (αλγ) λ mod f χ(λ)e = (89) χ(v) χ(λ)e2πiS (αλγν). λ mod f λ mod f Further since α αν( mod f), S (αλγ ( ν)) is a rational integer and hence ≡ − χ(λ)e2πiS (αλγ ) = χ(λ)e2πiS (αλγν). λ mod f λ mod f Applications to Algebraic Number Theory 96
But since χ( ) , χ(ν), we see from (89), that χ(λ)e2πiS (αλγ) = 0. λ mod f Hence (88) is true for all integral α. We now proceed to prove the T , 0. In fact, using (89), we have
2πiS (αγ) TT = T χ(α)e− α mod f 2πiS (α(λ 1)γ) = χ(λ) e − (90) λ mod f α mod f Now if α runs over a complete system of representatives of residue classes 107 modulo f, so does α + ν for every integer ν and hence
e2πiS ( αγ) = e2πiS ( γν) e2πiS ( αγ) α mod f α mod f Thus, if there exists at least one integer ν such that S ( γν) is not a rational integer, e2πiS ( αγ) = 0. Now S ( γν) is a rational integer for all integers ν α mod f 1 1 if and only if γ ϑ i.e. if and only if ( )qf− is integral i.e. f. Thus ∈ − ∈ 0, if < f, e2πiS ( αγ) = N(f), if f. α mod f ∈ From this and from (90), we have thenT 2 = N(f), i.e. T = √N(f). The determination of the exact value of T/ T | is| of the same order| | of difficulty as the corresponding problem for “generalized| | Gauss sums” considered by Hasse. We have, finally, for proper χ modulo f, as a consequence of (88),
1 2πiS (λβγ) χ(β) = T − χ(λ)e (91) λ mod f Let us notice that β appears in the exponent in the sum on the right-hand side of (??). We now insert the value of χ(β) as given by (??) in the series on the right hand side of (86). In view of the absolute convergence of the series for σ > 1 and in view of the fact that the sum in (93) is a finite sum, we are allowed to rearrange the terms as we like. We then obtain for σ > 1 and for a ray class character χ modulo f with f as conductor, 1 L(s, χ) = χ(λ) χ(b )(N(b ))s T A A × λ mod f A Applications to Algebraic Number Theory 97
2πiS (λβγ) s v(β)e N(β) − . (92) × | | b (β) A| In (92), λ runs over a full system of representatives of prime residue classes 108 modulo f, A over representatives of the ideal classes of K in the wide sense and the inner sum is over all principal non-zero ideals (β) divisible by bA. We shall use (92) to determine the value of L(s, χ) at s = 1 when K is a real or imaginary quadratic field over Q. In this section, we shall consider only the case when K is an imaginary quadratic field over Q, with discriminant D < 0. Here v(β) = 1 identically, since there is no nontrivial character of signature. Moreover, we could assume f , (1), for otherwise, L(s, χ) precisely the Dedekind zeta function of K. Let, therefore, f , (1) and let w and wf denote respectively the number of all roots of unity and of roots of unity ǫ satisfying ǫ 1( mod f). From (92), we have for σ > 1, ≡ 1 L(s, χ) = χ(λ) χ(b )(N(b ))s w T A A × λ mod f A 2πiS (λβγ) s e (N(β))− (93) × b β,0 A|
where the inner summation is over all β , 0 in bA. We now contend that as λ runs over a full system of representatives of the prime residue classes modulo f and bA over a complete set of representatives (integral and coprime to f) of the classes in the wide sense, then (λ)bA covers exactly w/wf times, a complete system of representatives of the ray classes modulo f. In fact, let ǫ1,...,ǫv(v = w/wf) be a complete system of roots of unity in K, incongruent modulo f. The corresponding residue classes modulo f form a subgroup E(f) of G(f). Let λi, i = 1,..., r be a set of integers whose residue classes modulo f constitute a full system of representatives of the cosets of G(f) modulo E(f). Then it is easily verified that when λ runs over the elements λ1,...,λr and bA over the representatives of the classes in the wide sense, (λ)bA covers exactly once a full system of representatives of the ray classes modulo f. Our assertion above is an immediate consequence. Thus, we have from (93), for σ > 1,
1 s 2πiS (βγ) s L(s, χ) = χ(bB)(N(bB)) e (N(β))− , (94) Twf B b β,0 B|
where B runs over the ray classes modulo f and bB is a fixed integral ideal in B 109 coprime to f. Applications to Algebraic Number Theory 98
Let [β1, β2] be an integral basis of bB; we can assume, without loss of gen- β2 erality that = zB = xB + iyB with yB > 0. Then if β bB, β = mβ1 + nβ2 for β1 ∈ rational integers m, n and N(β) = N(β ) m + nz 2. Moreover, 1 × | B| N(b ) D = β β β β = 2y N(β ). B | | | 1 2 − 2 1| B 1 Thus
s 2πiS (βγ) s (N(bB)) e (N(β))− b β,0 B| 2y s B ′ 2πiS ((mβ1+nβ2)γ) 2s = e m + nzB − , √ D | | | | m,n where, on the right hand side, the summation is over all ordered pairs of rational integers (m, n) not equal to (0, 0). Let us set now uB = S (β1γ) and vB = S (β2γ) and let f be the smallest positive rational integer divisible by f. Then in view of the fact that (γ)ϑ has exact denominator f and bB is coprime to f, it follows that uB and vB are rational numbers with the reduced common denominator f . Since f , (1), uB and vB are not simultaneously integral. We then have
s 2πiS (βγ) s (N(bB)) e (N(β))− b β,0 B| 2 s s ′ 2πi(muB+nvB) 2s = y e m + nzB − . (95) √ D B | | | | m,n We know that the function of s defined for σ > 1 by the infinite series on the right hand side of (95) has an analytic continuation which is an entire function of s. Its value at s = 1 is given precisely by Kronecker’s second limit formula. Indeed, by (39), we have
2 ′ ϑ (v u Z , z ) 2 2πi(muB+nvB) 2 1 B B B B πiu zB yB e m + nzB − = π = π log − e B . m,n | | − − η(z ) B πiuBvB Inserting the factor e− of absolute value 1 on the right hand side, we have 110
2πi(mu +nv ) 2 y ′ e B B m + nz − B | B| m,n 2 ϑ1(vB uBzB, zB) πiu (u z v ) = π log − e B B B− B . (96) − η(zB) Applications to Algebraic Number Theory 99
Now we define for real numbers u, v not simultaneously integral and z H, the function ∈ πiu(uz v) ϑ1(v uz, z) ϕ(v, u, z) = e − − . η(z)
In each ray class B modulo f, we had chosen a fixed integral ideal bB with β2 integral basis [β1, β2] and defined uB = S (β1γ), vB = S (β2γ) and zB = . β1 Now, the left hand side of (95) depends only on the ray class B. Hence from (95) and (96), it is clear that log ϕ(v , u , z ) 2 depends only on B and not on | B B B | the special choice of bB or β1, β2. From (94), (95) and (96) we can now deduce Theorem 9. If χ is a proper ray class character modulo an integral ideal f , (1) of Q( √D), D < 0, the associated L(s, χ) can be continued analytically into an entire function of s and its value at s = 1 is given by
2π 2 L(1, χ) = χ(bB) log ϕ(vB, uB, zB) , (97) −Twf √ D | | | | B where B runs over all the ray classes modulo f and T is the sum defined by (87). We shall need (97) later for the determination of the class number of the ‘ray class field’ of Q( √D). By the ray class field (modulo f) of an algebraic number field k, we mean the relative abelina extension K0 of k, with Galois group isomorphic to the group of ray classes (modulo f) in k such that the prime divisors of f are the only prime ideals which are ramified in K0. We are now interested first in determining the nature of the numbers ϕ(vB, uB, zB). For this purpose, we observe that ϕ(v, u, z) is a regular function of 111 z in H and we shall study its behaviour when z is subjected to modular transfor- mations and the real variables v and u undergo certain linear transformations. In fact, using the transformation formula for ϑ1(w, z) and η(z) proved earlier and the definition of ϕ(v, u, z), one easily verifies the following formulae, viz.
πiu ϕ(v + i, u, z) = e− ϕ(v, u, z), − ϕ(v, u + 1, z) = eπivϕ(v, u, z), − (98) ϕ(v + u, u, z + 1) = eπi/6ϕ(v, u, z), 1 πi/2 ϕ( u, v, z− ) = e− ϕ(v, u, z). − − Now, corresponding to a modular transformation z z = (az + b) (cz + d) 1 → ∗ − we define v∗ = av + bu and u∗ = cv + du. The last two transformation formulae Applications to Algebraic Number Theory 100 for ϕ(v, u, z) above, merely mean that corresponding to the elementary modular transformations z z = z + 1 or z z = z 1, we have → ∗ → ∗ − −
ϕ(v∗, u∗, z∗) = ρϕ(v, u, z), where ρ is a 12th root of unity. Since these two transformations generate the modular group, we see that for any modular transformation z z , → ∗
ϕ(v∗, u∗, z∗) = ρϕ(v, u, z) (98)′ where ρ = ρ(a, b, c, d) is a 12th root of unity which can be determined explicitly. Let (v, u) be a pair of rational numbers with reduced common denominator f > 1. Let us define for z H, ∈ Φ(v, u, z) = ϕ12 f (v, u, z).
Then as a consequence of the above formulae for ϕ(v, u, z), we see that Φ(v + 1, u, z) = Φ(v, u, z), Φ(v, u + 1, z) = Φ(v, u, z) and Φ(v∗, u∗, z∗) = Φ(v, u, z). If z z = (az + b)(cz + d) 1 is a modular transformation of level f , then v v → ∗ − ∗ − and u∗ u are rational integers and in view of the periodicity of Φ(v, u, z) in v and u, we− see that
Φ(v, u, z) = Φ(v∗, u∗, z∗) = Φ(v, u, z∗).
112 Let (vi, ui)i = 1, 2,..., q run over all the paits of rational numbers lying between 0 and 1 and having reduced common denominator f . Then corre- sponding to each pair (vi, ui), we have a function Φi(z) = Φ(vi, ui, z) which, as seen above, is invariant under modular transformations of level f . Moreover, 1 if z z∗ = (az + b)(cz + d)− is an arbitrary modular transformation, then for some→i, v v ( mod 1) and u u ( mod 1) so that in view of the periodicity ∗j ≡ i ∗j ≡ i of Φ(vi, ui, z) in vi and ui we see that
Φi(z∗) = Φ(vi, ui, z∗) = Φ(v∗j, u∗j, z∗) = Φ(v j, u j, z) = Φ j(z).
In other words, the functions Φi(z) are permuted among themselves by an arbi- trary modular transformation. Now, Φi(z) is regular in H and invariant under modular transformations 2πiz/ f of level f . Moreover, Φi(z) has in the local uniformizer e at infinity, a power-series expansion with at most a finite number of negative powers. Since Φi(z∗) = Φ j(z) for some j, we see that Φi(z) has at most a pole in the local uni- formizers at the ‘parabolic cusps’ of the corresponding fundamental domain. Applications to Algebraic Number Theory 101
Thus Φi(z) is a modular function of level f . Since a modular transformation merely permutes the functions Φi(z), we see that any elementary symmetric function of the functions Φi(z) is a modular function. From the theory of com- plex multiplications, it can be shown by considering the expansions ‘at infinity’ of the functions Φi(z), that Φi(z) satisfies a polynomial equation whose coef- ficients are polynomials, with rational integral coefficients, in j(z) (the elliptic modular invariant). If z lies in an imaginary quadratic field K over Q, it is known that for z < Q, j(z) is an algebraic integer. Since Φi(z) depends inte- grally on j(z), it follows that for such z, Φi(z) is an algebraic integer. Actually, from the theory of complex multiplication, one may show that for z K(z < Q), ∈ Φi(z) is an algebraic integer in the ray class field modulo f over K. In particular, 12 f the numbers ϕ (vB, uB, zB) corresponding to the ray classes B modulo f in K, are algebraic integers in the ray class field modulo f, over K. Let now K0 be the ray class field modulo f over K(= Q( √D), D < 0). Let ∆, R, W and g be respectively the discriminant of K0 relative to Q, the regulator 113 of K0, the number of roots of unity in K0 and the order of the Galois group of K0 over K. Let H and h denote respectively the class numbers of K0 and K. Itis clear that K0 has no real conjugates over Q and has 2g complex conjugates over Q. Moreover ∆ = D gN(ϑ), where N(ϑ) is the norm in K/Q of the relative | | | | discriminant ϑ of K0 over K. From class field theory, we know that
ζ (s) = ζ (s) L(s, χ ) ( ) K0 K 0 ∗ χ 0,1 where χ runs over all the non-principal ray class characters modulo f and if fχ is the conductor of χ, then χ0 is the proper ray class character modulo fχ associated with χ. It is known that ϑ = fχ. Multiplying both sides of ( ) χ ∗ above by s 1 and letting s tend to 1, we have − (2π)g H R 2πh = L(1, χ0). W √ ∆ w √ D | | | | χ,1 But, from (97), we have
2π 2 L(1, χ0) = χ0(bB0 ) log ϕ vB0 , uB0 , zB0 −T0wf √ D χ | | B0 where 2πiS (λγ0) T0 = T0(χ0) = χ0(λ)e , λ mod fχ Applications to Algebraic Number Theory 102
γ K is chosen such that (γ √d) has exact denominator f , B runs over 0 ∈ 0 χ 0 all the ray classes modulo fχ, bB0 is a fixed integral ideal in B0 coprime to fχ and wfχ is the number of roots of unity congruent to 1 modulo fχ. Witn N(fχ) g denoting the norm in K over Q, of the ideal fχ, we have ( N(fχ)) D = ∆ . It χ | | | | is now easy to deduce Theorem 10. For the class number H of the ray class field modulo f over K = Q( √D), D < 0, we have the formula
H W N(Fχ) 2 = χ (b ) log ϕ v , u , z 0 B0 B0 B0 B0 h w R − T0wfχ χ,1 B0 When f is a rational integral ideal in K, Fueter has derived a ‘similar’ for- 114 mula for the class number of the ray class field modulo f or the ‘ring class field’ modulo f over K, by using a “generalization” of Kronecker’s first limit formula. Unlike in the case of the absolute class field over K, it is not possible, in general, (for f , (1)) to write the product occurring in the formula for H/h χ,1 above, as a (g 1)-rowed determinant whose elements are of the form log η(k) − | i | where η(k), i = 1, 2,..., g 1 are conjugates of independent units η lying in i − i K0.
5 Ray class fields over Q( √D), D > 0.
Let K be a real quadratic field over Q, with discriminant D > 0. Let f be a given integral ideal in K and χ(, 1), a proper character of the group of ray classes modulo f in K. As in the case of the imaginary quadratic field, we shall first determine the value at s = 1 of the L-series L(s, χ) associated with χ and use this later to determine the class number of special abelian extensions of K. We start from formula (92) proved earlier, namely, for σ > 1,
1 L(s, χ) = T − χ(λ) χ(b ) A × λ mod f A s 2πiS (λβγ) s (N(b )) v(β)e N(β) − , × A | | b (β),(0) A| where A runs over all the ideal-classes of K in the wide sense, bA is a fixed integral ideal in A, chosen once for all and coprime to f and the inner sum is extended over all non-zero principal ideals (β) divisible by bA. Further, χ(λ) and v(λ) are respectively the character of G(f) and the character of signature Applications to Algebraic Number Theory 103 associated with the ray class character χ as in Proposition 14. Besides, γ is a q number in K such that (γ √C) has exact denominator f i.e. (γ) = where f( √D) (q, f) = (1); the number γ will be fixed throughout this section. Let, for α K, α′ denote its conjugate. We may, in the above, suppose ∈ β2 that [β1, β2] is a fixed integral basis of bA such that if ω = , then, without β1 loss of generality, ω>ω . We have, then, (ω ω) N(β ) = N(b ) √D. 115 ′ − | 1 | A Further, let uA = S (λβ1γ) and vA = S (λβ2γ); then, if β = mβ1 + nβ2, we have e2πiS (λβγ) = e2πi(muA+nvA). The character of signature v(λ) associated with the ray class character χ may, in general, be one of the following, namely, for λ , 0 in K, (i) v(λ) = 1 N(λ) (ii) v(λ) = N(λ) | | λ (iii) v(λ) = λ | | λ (iv) v(λ) = ′ . λ | ′| In what follows, we shall be concerned only with such ray class characters χ, for which the corresponding v(λ) is defined by (i) or (ii). We shall not deal with the characters χ, for which either (iii) or (iv) occurs, the determination of L(1, χ) being quite complicated in these cases. Let us first suppose that the character of signature v(λ) associated with χ is given by v(λ) = 1 for all λ , 0 in K. In other words, for an integer α in K, coprime to f, χ((α)) = χ(α). We may, moreover, suppose that f , (1), since otherwise, χ is a character of the wide class group and L(s, χ) is just the corresponding Dedekind zeta function of K. Let z = x + iy be a complex variable with y > 0. Corresponding to an ideal class A of K in the wide sense and an integer λ coprime to f, we define for σ > 1, the function
s 2πi(mu +nv ) 2s g(z, s,λ, u , v ) = y ′ e A A m + nz − A A | | m,n where the summation is over all pairs of rational integers m, n not simultane- ously zero. As a function of z, g(z, s,λ, uA, vA) is not regular but as a function of s, it is clearly regular for σ > 1. In fact, since uA and vA are not both in- tegral, we know that g(z, s,λ, uA, vA) has an analytic continuation which is an Applications to Algebraic Number Theory 104
entire function of s. For the sake of brevity, we shall denote it by gA(z,λ); this does not mean that it is a function of z and λ depending only on the class A. It does depend on the choice of the ideal bA, the integral basis [β1, β2], the residue 116 class of λ modulo f and further on γ. But we have taken bA, β1, β2 and γ to be fixed. In order to find L(1, χ), we shall employ the idea of Hecke already referred 2πiS (λβγ) s to on p. 86 ( 3, Chapter II) and express the series e N(β) − as § b (β),(0) | | A| an integral over a suitable path in H, with the integrand involving gA(z,λ). We shall denote the group of all units in K by Γ and the subgroup of units congruent to 1 modulo f, by Γf. Moreover, let Γf∗ be the subgroup of Γf con- sisting of ρ such that ρ > 0. The group Γf∗ is infinite cyclic and let ǫ be the generator of Γ , which is greater than 1. We see that ǫ 1( mod f), N(ǫ) = 1, f∗ ≡ ǫ>ǫ′ > 0. Since [ǫβ1, ǫβ2] is again an integral basis of bA, ǫβ2 = aβ2+bβ1, ǫβ1 = cβ2+ dβ1 where a, b, c, d are rational integers such that ad bc = N(ǫ) = 1. Further, − 1 if ω = β2/β1, then ǫω = aω + b, ǫ = cω + d so that ω = (aω + b)(cω + d)− . We also have the same thing true for ω′, namely, ǫω′ = aω′ + b, ǫ′ = cω′ + d, 1 so that we have ω′ = (aω′ + b) (cω′ + d)− . Thus the modular transformation z z = (az+b) (cz+d) 1 is hyperbolic, with ω, ω as fixed points and it leaves → ∗ − ′ fixed the semicircle on ω′ω as diameter. If now we introduce the substitution 1 1 z = (ωpi + ω′)(pi + 1)− (or equivalently, po = (z ω′)(ω z)− we see that when z( H) lies on this semicircle, p is real and positive.− As− a matter of fact, ∈ when z( H) describes the semicircle form ω′ to ω, p runs over all positive real ∈ 1 numbers from 0 to . Let p∗i = (z∗ ω′) (ω z∗)− ; it is easy to verify that 2 ∞ − − p∗ = ǫ p. This means that p∗ > p, since ǫ > 1. Consequently, whenever z lies on the semicircle on ω′ω as diameter, z∗ again lies on the semicircle and always to the right of z. Consider now
z 0∗ dp F(A,λ, s) = g(z, s,λ, uA, vA) , z0 p the integral being extended over the arc of this semicircle, from a fixed point 1 z0 H to z0∗ = (az0 + b)(cz0 + d)− . In view of the uniform convergence of the ∈ s 2πi(mu +nv ) 2s series y ′ e A A m + nz on this zrc we have, for σ > 1, m,n | |− z0∗ z0∗ dp 2πiS (λβγ) s 2s dp g (z,λ) = ′ e y m + nz − , A P | | P z0 m,n z0 where β = mβ1 + nβ2. Setting = m + nω = β/β1, we verify easily that 117 Applications to Algebraic Number Theory 105 m + nz 2 = ( 2 p2 + 2)(p2 + 1) 1 and y = p(ω ω )(p2 + 1) 1. Thus | | ′ − − ′ −
z0∗ p0∗ s 2s dp s s 2 2 2 s dp y m + nz − = (ω ω′) P ( p + ′ )− z0 | | p − p0 p / p (ω ω )s ′ 0∗ ps dp = ′ | | ′ − s 2 s p p N( ) / p (p + 1) p → | | | ′| 0 s 2 (ββ′ /β′β1) p0ǫ s N(b ) √D | 1 | p dp = A 2 s N(β) | (ββ /β β ) p (p + 1) p | | | 2′ ′ 1 | 0 We have, therefore,
(ββ /β β ) ǫ2 p 1′ ′ 1 0 ps dp ,λ, = b √ s 2πiS (λβγ) β s | | F(A s) (N( A) D) e N( ) − 2 s | | (ββ′ /β β1) p0 (p + 1) p b A β,0 | 1 ′ | | (99) We use once again the trick employed by Hecke (p. 86). For a fixed integer
β K, all the integers in K which are associated with β with respect to Γf∗ are ∈ k l of the form βǫ± , k = 0, 1, 2,.... Keeping β fixed, if we replace β by βǫ in the integral on the right hand side of (99), then the integral goes over into
2l+2 (ββ1′ /β′β1) ǫ p0 | | s 2 s dp p (p + 1)− . (ββ /β β ) ǫ2l p p | 1′ ′ 1 | 0 2k 2k+2 Now since ǫ > 1, the intervals ( ββ1′ β′β1 p0ǫ , ββ1′ β′β1 p0ǫ ) cover the entire interval (0, ) without gaps| and| overlaps,| | as k| runs| over all rational integers from ∞to + . Moreover, since ǫ 1( mod f) and N(ǫ) = 1, we have e2πiS (λβǫkγ−∞) = e2πiS∞(λβγ) and N(βǫk) = N≡(β) . Thus the total contribution to the sum on the right hand side| of (99)| from| all| integers associated with a
fixed integer β with respect to Γf∗, is given by
e2πiS (λβγ) ps dp e2πiS (λβγ) Γ2(s/2) ∞ = N(β) s (p2 + 1)s p N(β) s 2Γ(s) | | 0 | | As a consequence, 118
s 2 (N(bA) √D) Γ (s/2) 2πiS (λβγ) s F(A,λ, s) = e N(β) − (100) 2Γ(s) | | b β(Γ ) A| f∗ where, on the right hand side, the summation is over a complete set of integers
β , 0 in bA, which are not associated with respect to Γf∗. Applications to Algebraic Number Theory 106
Let ρ denote a complete set of units incongruent modulo f and let ǫ be { i} { j} a complete set of representatives of the cosets of Γf modulo Γf∗. It is clear that ρ ǫ runs over a full set of representatives of the cosets of Γ modulo Γ . Thus { i j} f∗ 2 Γ (s/2) s 2πiS (λρ ǫ βγ) s F(A,λ, s) = (N(b ) √D) e i j N(β) − (101) 2Γ(s) A | | ρi,ǫ j b (β) A| where, on the right hand side, the inner sum is over all non-zero principal ideals (β) divisible by bA. By the signature of an element α , 0 in K, we mean the pair (α/ α , α / α ). | | ′ | ′| Now we can certainly find a set l of integers l such that l 1( mod f) and the set ǫ consists precisely{ of} 4 elements with the 4 possible≡ different { j l} signatures. Let λk be a set of integers λk constituting a complete system of representatives of{ the} cosets of G(f) modulo E(f). Here E(f) is the group of prime residue classes mod f, containing at least one unit in K. The set λkρi is seen to be a complete set of representatives of the prime residue classes{ } modulo f. It is easy to prove
Proposition 15. The set of ideals (λk l)bA serves as a full system of repre- sentatives of the ray classes modulo{ f. }
Proof. Obviously it suffices to show that (λk l) runs over a complete set of representatives of the ray classes modulo f{lying in} the principal wide class. In fact, if (α) Gf, then α ρiλk( mod ∗f) for some ρi and λk and moreover, 119 ∈ ≡ α we can find ǫ j l such that > 0. Since ǫ j l 1( mod f), we have λkρiǫ j l ≡ (α) (λ ρ ǫ ) = (λ modulo f. It is easy to verify that no two elements of ∼ k l i j k l the set (λk l) are equivalent modulo f. From{ (101),} we then have for σ > 1
χ(λk l)χ(bA)F(A,λk l, s)
λ k, l,A Γ2(s/2) = χ(λ )χ(b )(N(b ) √D)s 2Γ(s) k l A A × λ k, l,A 2πiS (λ ρ ǫ βγ) s e k l i j N(β) − × | | b A (β) ρi|,ǫ j Γ2(s/2) = χ(λ )χ(b )(N(β ) √D)s 2Γ(s) k A A × λ k,ρi,A 2πiS (λ ρ ǫ βγ) s χ( ) e k l i j N(β) − . (102) × l | | ǫ j, l b (β) A| Applications to Algebraic Number Theory 107
1( mod f), e2πiS (λk lρiǫ jβγ) = e2πiS (λkρiβγ) and since 1( mod f), χ( ) = 1. ≡ l ≡ l Hence the sum in (102) is independent of ǫ j, l and we have
χ(λk l)χ(bA)F(A,λk l, s)
λ k, l,A Γ2(s/2) = 4 Ds/2 χ(λ ρ )χ(b )(N(b ))s 2Γ(s) k i A A × λ k,ρi,A 2πiS (λ ρ βγ) s e k i N(β) − . | | b (β),(0) A|
Let (λk l)bA lie in the ray class B modulo f. Denoting (λk l)bA by bB, we know that bB runs over a full system of representatives of the ray classes modulo f. Also, if α1 = λk lβ1, α2 = λk lβ2, then [α1, α2] is an integral basis 120 of bB and we may now denote the function
s 2πiS ((mα +nα )γ) 2s g (z,λ ) = y ′ e 1 2 m + nz − A k l | | m,n
z0∗ dp by gB(z, s) and gB(z, s) by F(B, s). z0 p The use of the notation F(B, s) is justified for, from (101), we see that F(B, s) depends only on the ray class B modulo f (and on γ) and not on the particular integral ideal bB chosen in B. For, if we replace bB by ( )bB where > 0, 1( mod f) and ( )b is an integral ideal coprime to f, then it is ≡ ∗ B easy to show that S (λk lρiǫ j βγ) S (λk lρiǫ jβγ) is a rational integer and in addition, (N(b ))s N(β) s again depends− only on the ideal class of b . A | |− A Now χ((λk l))χ(bA) = χ(bB). Moreover, λkρi runs over a complete set of representatives λ of the prime residue classes{ modulo} f. As a consequence, we have Γ2(s/2)Ds/2 χ(b )F(B, s) = 2 χ(λ) B Γ(s) × B λ mod f s s χ(b )(N(b )) (λβγ) N(β) − × A A ∗∗∗∗∗∗ | | A b (β) A| where B runs over all ray classes moudlo f ******** a complete system of prime residue classes modulo f and A ***** ideal classes in the wide sense. In other words, we have for σ **********,
1 s/2 χ(b )F(B, s) = (s/2)(Γ(s))− D T L(s, χ). (103) B ∗∗∗∗∗ B Applications to Algebraic Number Theory 108
The function gB(z, s) is an Epstein zeta-function of the type of ζ(s, u, v, z, 0) discussed earlier in 5, Chapter I. Using the simple functional equation for the Epstein zeta function§ ζ(s, u, v, z, 0), we shall now derive a functional equation for L(s, χ). Proposition 16. If χ is a proper ray class character modulo f(, (1)) whose associated character of signature v(λ) 1, then L(s, χ) is an entire func- tion of s satisfying the following functional≡ equation, namely, if ξ(s, χ) = s 2 s/2 π− Γ (s/2)(DN(F)) L(s, χ), then χ(q) ξ(s, χ) = ξ(1 s, χ), T/ √N(f) − 121 Remark. The functional equation ( ) for ****** generalization to arbitrary al- gebraic number **** derived by Hecke∗ by using the generalized **** formula’ (for algebraic number ****** same, it is interesting to deduce the same **** by using the functional equation satisfied by ζ(s, u, v, z, 0). Proof. From (92), we have
s 2 s/2 s 2 s/2 s π− Γ (s/2)D T L(s, χ) = π− Γ (s/2)D χ(λ) χ(bA)(N(bA)) λ A 2πiS (λβγ) s e N(β) − × | | b (β),0 A| π sΓ2(s/2)Ds/2 = − χ(λ) χ(b )(N(b ))s e(f) A A × λ A
× b β(Γ ) A| f∗
vA where e(f) is the index of Γ∗ in Γ. Now, if we set u∗ = − , then, in the notation f uA of 5, Chapter I, we have g (z, s,λ, u , v ) = ζ(s, u , 0, z , 0) and by (100) § A A A ∗ s 2 s/2 2 π− Γ (s/2)D T L(s, χ) = χ(λ) χ(b ) e(f) A × λ A
z0∗ s dp π− Γ(s)ζ(s, u∗, 0, z, 0) . × z0 p
By the method of analytic continuation of ζ(s, u∗, 0, z, 0) discussed earlier in 5, Chapter I, we have §
z0∗ s dp π− Γ(s)ζ(s, u∗, 0, z, 0) p Applications to Algebraic Number Theory 109
z∗ 0 dp ∞ 1 2 dt = ′ πiy− m+nz +2πi(muA+nvA) s e− | | t . z p 0 t 0 m,n 1 ∞ ... + ∞ ... 122 The inner integral 0 may be split up as 0 1 . Now we may use πy 1t m+nz 2+2πi(mu +nv ) the theta-transformation formula for the series e− − | | A A and m,n further make use of the inequality y 1 m + nz 2 c(m2 + n2) uniformly on the − | | ≥ arc from z0 to z0∗, for a constant c independent of m and n. If, in addition, we keep in mind that χ is not the principal character, we can show that L(s, χ) is an entire function of s.
Using now the functional equation of ζ(s, u∗, 0, z, 0), we have
s 2 s/2 2 π− Γ (s/2)D T L(s, χ) = χ(λ) χ(b ) e(f) A × λ A
z0∗ (1 s) dp π− − Γ(1 s)ζ(1 s, 0, u∗, z, 0) . × z0 − − p
1 s 2(1 s) For σ < 0, ζ(1 s, 0, u∗, z, 0) = y − m vA + (n + uA)z − − and by the − m,n | − | same arguments as above, we can show that
∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗ 2 1 s 1 s (1 s) = Γ − (N(bA) √D) − N( ) − − 2 | | =β+ν β b ∈ ′A √ where, on the right hand side, = β1′ S (λβ2γ) + β2′ S (λβ1γ) = DλγN(bA) and runs over a complete set of numbers− of the form β + ν with−β b , such ∈ ′A that they are not mutually associated with respect to Γf∗. Let δ be an integer in K such that af = (δ) with an integral ideal a coprime to f. Then, for σ < 0, we obtain from above that
(1 s) s 2 s/2 π− − (1 s)/2 2 π− Γ (s/2)D T L(s, χ) = D − Γ ((1 s)/2) e(f) − × 1 s 1 s χ(b )(N(b )) − (N(af)) − χ(λ) × A A × A λ (1 s) N(β) − − , × | | β νδ( mod ab′ f) ≡ A β,0,ab′ β(Γ ) A| f∗ Applications to Algebraic Number Theory 110 where, in the inner sum on the right hand side, β runs over a complete set of 123 integers in ab′A which are not mutually associated with respect to Γf∗ and which satisfy β νδ( mod ab′ f). If β νδ mod ab′ f, then ≡ A ≡ A χ(β) = χ(νδ) = χ( γ √DλδN(b )) = χ((N(b )γ √Dλδ)) − A A √ = χ(bA)χ(b′A)χ((γ Dδ))χ(λ) = χ(bA)χ(b′A)χ(aq)χ(λ).
Hence χ(λ) = χ(β)χ(bAb′Aaq). Further if λ runs over representatives of prime residue classes modulo f and β runs independently over a complete set of inte- gers in ab′ congruent modulo ab′ f to √DλγδN(b ) and not mutually associ- A A − A ated with respect to Γf∗, then β runs over a complete set of integers in ab′A not mutually associated with respect to Γf∗ and satisfying ((β), ab′Af) = ab′A, If we 1 note, in addition, that χ(β) = 0 for those β in ab′A for which (β)(ab′A)− is not coprime to f, we can show finally that
(1 s) χ(λ) N(β) − − = χ(b b′ aq) | | A A × ab′ β(Γ ) A| 1∗ β νδ( mod ab′ F) ≡ A (1 s) χ(β) N(β) − − . × | | ab′ β(Γ ) A| f∗ β,0 Thus, for σ < 0, π (1 s) π sΓ2 / s/2 , χ = − − (1 s)/2Γ2 / χ q − (s 2)D T L(s ) s 1 D − ((1 s) 2) ( ) (N(f)) − − × 1 s χ(ab′ )N((ab′ )) − × A A × A (1 s) χ(β) N(β) − − | | ab′ (β),(0) A| χ(q) = π (1 s) (1 s)/2Γ2 / , χ , − − D − ((1 s) 2) s 1 L(1 s ) − (N(f)) − − since ab′A again runs over a system of representatives of the ideal classes simi- lar to bA. Since L(s, χ) is an entire function of s, the above functional relation is 124 s s s/2 clearly valid for all s. We have now only to set ξ(s, χ) = π− Γ (s/2)(DN(f)) L(s, χ), to see that the equation ( ) is true. ∗ Let now [α1, α2] be an integral basis for the ideal bB in the ray class B and let ω = α2/α1, uB = S (α1γ), vB = S (α2γ). By Kronecker’s second limit formula, we have g (z, s) = π log ϕ(v , u , z) 2 B − | B B | Applications to Algebraic Number Theory 111
+ terms involving higher powers of (s 1) .... − Letting s tend to 1, we have from (103),
z 1 0∗ dp L(1, χ) = χ(b ) log ϕ(v , u , z) 2 . − B | B B | p 2T √D B z0 Now it is easy to verify that
dp (ω ω ) √D = − ′ dz = dz, p −(z ω)(z ω ) − F (z) − − ′ B where √D 2 F (z) = (z ω)(z ω′) = a z + b z + c ( ) B ω ω − − 1 1 1 ∗ − ′ has the property that a1, b1, c1 are rational integers with the greatest common divisor 1, a > 0 and b2 4a c = D. Thus we have 1 1 − 1 1 Theorem 11. For a proper ray class character modulo f , (1) in Q( √D), D > 0, with the associated v(λ) 1, the value of L(s, χ) at s = 1 is given by ≡ z 1 0∗ dz L(1, χ) = χ(b ) log ϕ(v , u , z) 2 , 2T B | B B | F (z) B z0 B where B runs over all the ray classes modulo f and F (z) is given by ( ) above. B ∗ As remarked earlier, the terms of the sum on the right hand side depend only on γ (fixed!) and on the ray class B and not on the choice of the ideal bB in B. We observe further that it does not seem to be possible to simplify this 125 formula any further, in an elementary way. One might try to follow the method of Herglotz to deal with the integrals but then the invariance properties of the integrand are lost in the process. We proceed to discuss the case when the character of signature v(λ) asso- ciated with the given ray class character χ is of type (ii), i.e. for integral α(, 0) N(α) coprime to f, we have χ((α)) = χ(α) . In this case, we shall now see N(α) that the value at s = 1 of L(s, χ) can be| determined| in terms of elementary functions, using the first or second limit formula of Kronecker, according as f = (1) or f , (1). The fact that the character of signature v(λ) associated with χ is defined by N(λ) v(λ) = for λ , 0 (104) N(λ) | | Applications to Algebraic Number Theory 112 implies automatically a condition on f. For instance, if ρ is a unit congruent to N(ρ) 1 modulo, f, then = χ(ρ)v(d) = χ((ρ)) = 1. In other words, every unit N(ρ) congruent to 1 modulo| f| should necessarily have norm 1. To determine L(1, χ), we start as before from formula (92), namely, for σ > 1,
1 s L(s, χ) = T − χ(λ) χ(b )(N(b )) A A × λ mod f A
2πiS (λβγ) N(β) s e N(β) − . × N(β) | | b (β),(0) A| | | We shall try to express the inner sum on the right hand side as an integral, as before. We shall follow the same notation as in the case treated above. ∂ 1 ∂ i ∂ Let = ; then in view of the uniform convergence of the ∂z 2 ∂x − 2 ∂y series defining the function gA(z, s,λ, uA, vA) (denoted by gA(z,λ) for brevity) for σ > 1, we have
∂gA(z,λ) 2πi(mu +nv ) ∂ s 2s = ′ e A A y m + nz − ∂z ∂ { | | } m,n z s s 1 2πi(mu +nv ) (2s 2) 2 = y − ′ e A A m + nz − − (m + nz)− . (105) 2i | | m,n 126 z ∂g (z,λ) Let us now consider the integral 0∗ A dz, extended over the same z0 ∂z arc of the semi-circle in H as considered earlier. Due to the uniform conver- gence of the series (105) on this arc, we have
z∗ z∗ s 1 0 ∂gA(z,λ) s 0 y − dz = ′ e2πi(muA+nvA) dz ∂z 2i m + nz 2s 2(m + nz)2 z0 m,n z0 | | − s = Ds/2(N(b ))s 2 A × p0∗ s 1 2πiS (λβγ) s p − dp e N(β1) − . × | | ( 2 p2 + 2)s 1( pi + )2 b β,0 p0 ′ − ′ A| Effecting the substitution p / p, we see that → | ′ | p 0∗ ps 1 − dp 2 2 2 s 1 2 p0 ( p + ′ ) − ( pi + ′) Applications to Algebraic Number Theory 113