INVESTIGATING ZETA FUNCTIONS AMANDA KNECHT Abstract. In this paper we investigate the Riemann zeta function, ζR (s) = P ∞ −s . The Riemann Hypothesis, which remains unproven after 143 years, n=0 n states that the nontrivial complex roots of ζ have real parts equal to one-half. Our study primarily involves the function ξR used in the functional equation of the Riemann Zeta Function. The roots of ξR correspond exactly to the roots of ζR , and the function satisfies the equation ξR (s) = ξR (s−1). We study the slopes of the level sets < (ξR (s)) = 0 and = (ξR (s)) = 0 in order to find information about the intersection points of the level sets, because these intersection points are the roots of ξR . We then extend our study of the Riemann zeta function to the more general Epstein zeta function, Z (Y, s). The function Z (Y, s) depends on a positive, symmetric n × n real matrix Y and s ∈ C. We define an Epstein ξ function, ξE , as we did for the Riemann Zeta Function and investigate the level sets < (ξE (Y, s)) = 0 and = (ξE (Y, s)) = 0 in order to discover how the matrix Y affects the location of the function’s roots. Contents 1. Introduction 2. The Riemann Zeta Function 3. Consequences of the Slope Conjecture 4. The Epstein Zeta Function References 1 3 13 13 24 1. Introduction For references pertaining to the early history of the Riemann zeta function, we refer to Edwards [Ed] and Titchmarsh [Ti] and to Terras [Te] for background on the Epstein zeta function. In 1859 Bernhard Riemann wrote an 8-page paper, “ On the Number of Primes Less Than a Given Magnitude,” which has made a great impact on modern mathematics . In this paper Riemann studies the Euler product formula: ∞ X Y 1 1 = s 1 n 1 − n=1 p prime ps which is absolutely convergent for all s ∈ C such that formula to define a new function. 1We 1 < (s) > 1. He uses this will use < (s) and = (s) to denote the real and imaginary parts, respectively, of a complex number s. 1 2 AMANDA KNECHT Definition 1. Riemann’s zeta function on the half plane < (s) > 1 is ζ (s) = ∞ X 1 . s n n=1 Riemann analytically continues the zeta function above so that it is absolutely convergent over the entire complex plane except for a simple pole at s = 1. The paper also introduces the functional equation: s 1−s − 1−s − 2s 2 ζR (s) = π Γ ζR (1 − s) . π Γ 2 2 R∞ Here Γ is Euler’s Gamma Function and can be written as Γ (s) = 0 e−t ts−1 dt. By multiplying the functional equation by 21 s (s − 1), Riemann was able to eliminate the poles of the function and create a new function, ξR , defined by: s −s 1 ξR (s) = s (s − 1) Γ π 2 ζR (s) . 2 2 Riemann goes on to show that none of the roots s of ξR (s) = 0 lie on the half plane < (s) > 1. And through a simple investigation, one can see that ξR (s) = ξR (1 − s), so 1 − s is a root if and only if s is a root. Thus, it can easily be shown that there are no roots on the half plane < (s) < 0. Whence, all roots of ξR must lie on the strip 0 ≤ < (s) ≤ 1. He then states, without proof, that for 0 < = (s) < T and T T T < (s) = 21 the number of roots of ξR (s) is approximately 2π log 2π − 2π . And this idea led to the famous Riemann Hypothesis, which states that all the roots of ξR lie on the line < (s) = 21 . It is known that ζR has zeros at every even negative integer, and these zeros are called trivial. The remaining zeros correspond exactly to the zeros of ξR , and so we can restate the Riemann Hypothesis by saying that the nontrivial roots of ζR have real part equal to 12 . The images included in the appendix are used in our research in order to better understand the functions we analize. The first image is a map of the complex plane where each quadrant is assigned a color. When graphing a function, f (z), the original value z is denoted by its location on the graph, and the function value, f (z), is denoted by color. Thus, the image tells us the quadrant each point is mapped to by f. Where two colors meet is a level set <(f (z)) = 0 or =(f (z)) = 0, and when all four colors meet Re(f (z)) = =(f (z)) = 0. One can see from image 2 that the Riemann Hypothesis appears to be true. But this hypothesis pertains only to the Riemann zeta function, which is a specific case of a Selberg Zeta function. The Selberg Zeta function is defined on all manifolds and can be written as X 1 , λs λ6=0 where the λ are the eigenvalues of the Laplacian counted with multiplicity. From studying the geometry of flat tori, Epstein’s zeta function, ζE (Y, s), was created. This more general ζE (Y, s) can be studied much like ζR (s). In this paper, we will investigate ξR , so that we can better understand why the INVESTIGATING ZETA FUNCTIONS 3 Riemann Hypothesis appears to be true. We will then extend our study into the Epstein zeta function in order to understand why the Riemann Hypothesis does not hold for all values of Y inputted into ζE (Y, s). 2. The Riemann Zeta Function Our study of the Riemann Zeta function, ∞ X 1 ζR (s) = , s n n=1 rests primarily on the study of the equations: Z ∞ X u 1 1 1 2 u 2 + 2w = + 4 w − e 4 cosh(uw) (1) ξR e−k πe du. 2 2 16 0 k≥1 (2) ξC 1 +w 4 1 = 2 w2 − Z +4 1 16 ∞ u X e 4 cosh(uw) 0 e−k 2 πeu du. k≥1 We derive equation (2) from the Circle Zeta function in which the λ represent the eigenvalues of the Laplacian of the circle. X 1 (3) ζC (s) = . s λ λ6=0 R∞ Knowing that ΓR(s) = 0 e−u us−1 du, we can change variables letting u = λt to R ∞ −λt 1 s ∞ −λt s−1 −s find Γ (s) = λ 0 e t dt. Then λ = Γ(s) 0 e ts−1 dt. Thus, we can also write the circle zeta function as: ! Z ∞ X 1 ts−1 (4) ζC (s) = e−λt dt. Γ (s) 0 λ6=0 Substituting λ = n2 ), with multiplicity two, into equations (3) and (4 ∞ X 1 = 2ζR (2s) . ζC (s) = 2 2s n n=1 1 ζC (s) = Γ (s) Z 0 ∞ ts−1 2 ∞ X ! −n2 t e dt. n=1 Now we introduce a new function, K (t), based on Jacobi’s theta function [Te]: √ X ∞ X −k2 π 2 π −n2 t K (t) + 1 := 1 + 2 e = √ e t . t k∈Z n=1 √ √ π π K (t) = √ − 1 + K (t) + 1 − √ . t t 4 AMANDA KNECHT Plugging this equation into the circle zeta function: Z ∞ 1 ζC (s) = ts−1 K(t)dt Γ(s) 0 √ Z π 1 π s−1 √ − 1 dt = t Γ(s) 0 t √ Z π π 1 s−1 K (t) + 1 − √ dt t + Γ (s) 0 t ! Z ∞ ∞ X 1 2 + ts−1 2 e−n t dt. Γ (s) π n=1 Looking at the first two integrals above separately, (1) 1 Γ (s) 0 √ π 1 s−1 √ − 1 dt = t Γ (s) t Z π π Z ! 1√ ts− 2 π ts π − |0 s s − 12 1 πs 1 − = f or <(s) > 1/2. Γ (s) s − 12 s (2) 1 Γ (s) 0 √ π t K (t) + 1 − √ dt = t ! Z π X −k2 π2 √π π2 1 √ dt = ts−1 2 e t Let v = Γ (s) 0 t t k≥1 ! √ Z ∞ 2 s−1 X 1 π π −π 2 −k2 v q = 2 e dv 2 Γ(s) π v π2 v k≥1 v ! 1 Z X π (2s− 2 ) ∞ (−s− 21 ) 2 2 e−k v dv. = v Γ (s) π k≥1 s−1 Thus, we write the circle zeta function as, ζC (s) = πs 1 1 − Γ (s) s − 12 s ! 1 Z X π (2s− 2 ) ∞ (−s− 21 ) 2 + v 2 e−k v dv Γ (s) π k≥1 ! Z ∞ ∞ X 1 2 + v s−1 2 e−n v dv. Γ (s) π n=1 INVESTIGATING ZETA FUNCTIONS 5 From the functional equation of ζC we can write, ξC (s) : = π −s Γ (s) ζC (s) . ! Z ∞ X 1 1 1 1 2 = − − 1 + π −( 2 −s) v (( 2 −s)−1) 2 e−k v dv s − s π 2 k∈Z ! Z ∞ ∞ X 2 e−n v dv. v s−1 2 +π −s π n=1 Now we let v = πa to find, Z ∞ X 1 1 1 2 ξC (s) = − + 1 a−1 as + a 2 −s 2 + e−k πa da. s −s 1 2 k≥1 And letting s = ξC 1 4 +w 1 4 + w, = R ∞ −3 P 2 + 1 a 4 (aw + a−w ) 2 k≥1 e−k πa da P R ∞ −3 2 = 2 w21− 1 + 1 a 4 ew log a + e−w log t 2 k≥1 e−k πa da ( ) 16 R ∞ −3 P 2 = 2 w21− 1 + 1 a 4 cosh (w log a) k≥1 e−k πa da. ( ) 16 −1 1 +w 4 − 1 1 −w 4 Finally, we will let u = log a to find that ξC can be written as Z ∞ X u 1 1 2 u 4 cosh (uw) ξC e e−k πe du. +w = + 4 1 2 4 2 w − 16 0 k≥1 It is easy to show that ξC 41 + w = ξC 14 − w , so the function reflects across the line < (z) = 41 . This equation can be related back to the Riemann Zeta Function as follows: ξC (s) = π −s Γ (s) 2ζR (2s) s −s 1 ξR (s) = s (s − 1) Γ π 2 ζR (s) 2 2 s 2ξR (s) π 2 ζR (s) = s (s − 1) Γ 2s 2ζR (2s) = ξC (s) = 4ξR (2s) π s 2s (2s − 1) Γ (s) 4π −s Γ (s) ξR (2s) π s 2s (2s − 1) Γ (s) ξC (s) = 2ξR (2s) s (2s − 1) Letting s = w + 14 : ξC 1 w+ 4 ξR 2w + 12 = 1 w2 − 16 6 AMANDA KNECHT ξR 1 2w + 2 = 1 w − 16 2 ξC 1 w+ 4 We know: Z ∞ X u 1 −k2 πeu 4 cosh (uw) ξC e = + 4 e du. 1 2 w2 − 16 0 k≥1 u P 2 u So letting h (u) = e 4 k≥1 e−k πe , we get: ξR 1 +w 4 1 2w + 2 (5) ξR = 1 2w + 2 1 w2 − 16 " 1 2 2 w − Z 1 16 +4 # ∞ cosh (uw) h (u) du 0 Z ∞ 1 1 2 = +4 w − cosh (uw) h (u) du. 2 16 0 This new function, ξR , reflects across the line <(s) = 1/2. The level sets < ξR 21 + 2w = 0 represent the values of w ∈ C around which < ξR 21+ 2w changes from positive to negative. Whereas, the sets where = ξR 21 + 2w = 0 represent the values of w ∈ C around which = ξR 12 + 2w changes from positive to negative. These level sets occur under two situations. (1) They are found on the critical line, the line < (s) = 21 , when < (w) = 0. Substituting w = 2ıy, y ∈ R, into the functional equation(5 ), we find: Z ∞ u 1 1 1 2 ξR e 4 cosh (ı2uy) h (u) du. + 2ıy = + 4 −4y − 2 2 16 0 Because cosh (ı2uy) = cos (2uy), Z ∞ u 1 1 1 2 ξR + 2ıy = + 4 −4y − e 4 cos (2uy) h (u) du 2 2 16 0 1 which is always real. So = ξR 2 + 2ıy = 0 for all y ∈ R . (2) The general equations satisfied when the ξ function has real or imaginary parts equal to zero can be found by substituting w = x + ıy, x, y ∈ R and x 6= 0, into equation (5): Z ∞ 1 1 1 2 ξR 2 (x + ıy) + = + 4 (x + ıy) − cosh (u (x + ıy)) h (u) du. 2 2 16 0 We know cosh (ux + ıuy) = cosh (ux) cos (uy) + ı sinh (ux) sin (uy) . So 1 ξR 2x + 2ıy + = 2 Z ∞ 1 2 2 4 x −y − cosh (ux) cos (uy) + ı sinh (ux) sin (uy) h (u) du 16 0 Z ∞ 1 cosh (ux) cos (uy) + ı sinh (ux) sin (uy) h (u) du. + + 8ıxy 2 0 INVESTIGATING ZETA FUNCTIONS 7 We can now find the real and imaginary parts of ξR . R∞ 1 1 < ξR cosh (ux) cos (uy) h (u) du + 2x + 2ıy = 4 x2 − y 2 − 16 0 2 R∞ + 21 − 8xy 0 sinh (ux) sin (uy) h (u) du. R∞ 1 1 sinh(ux) sin(uy) h(u)du = ξR + 2x + 2ıy = 4 x2 − y 2 − 16 0 2 R∞ +8xy 0 cosh(ux) cos(uy) h(u)du. These can be studied in order to find where < (ξR ) = = (ξR ) = 0. The first quantities we want to study are the slopes of the equations < (ξR ) = 0 and = (ξR ) = 0. We know that the gradient of a function F (x, y) equals (Fx , Fy ) and is perpendicular to a level set, F (x, y) = C for some constant C. So to find the slope of a level set of the function, we must find a vector perpendicular to the gradient of the set. The vector (−Fy , Fx ) satisfies this need because x (Fx , Fy ) · (−Fy , Fx ) = 0. The slope of this vector, −F , is the slope of F (x, y) = C. Fy In order to find the slopes of < ξ 12 + 2x + 2ıy = 0 and = ξ 21 + 2x + 2ıy = 0, we only need to think of them as functions of x and y and apply what we know from above. The equations for the partial derivatives are: 1 ∂ + 2x + 2ıy = < ξR ∂x 2 Z ∞ 8x cosh(ux) cos(uy)h (u) du 0 Z ∞ 1 2 2 +4 x − y − sinh(ux) cos(uy)uh(u)du 16 0 Z ∞ Z ∞ −8y sinh(ux) sin(uy)h (u) du − 8xy cosh(ux) sin(uy)h (u) du. 0 0 ∂ 1 < ξR + 2x + 2ıy = ∂y 2 Z ∞ Z ∞ 1 2 2 −8y cosh(ux) cos(uy)h (u) du − 4 x − y − cosh(ux) sin(uy)uh(u)du 16 0 0 Z ∞ Z ∞ −8x sinh(ux) sin(uy)h (u) du − 8xy sinh(ux) cos(uy)h (u) du. 0 0 ∂ 1 = ξR + 2x + 2ıy = ∂x 2 Z ∞ Z ∞ 1 2 2 8x sinh(ux) sin(uy)h (u) du + 4 x − y − cosh(ux) sin(uy)uh(u)du 16 0 0 Z ∞ Z ∞ +8y cosh(ux) cos(uy)h (u) du + 8xy sinh(ux) cos(uy)uh (u) du 0 0 8 AMANDA KNECHT ∂ 1 = ξR + 2x + 2ıy = ∂y 2 Z ∞ Z ∞ 1 2 2 −8y sinh(ux) sin(uy)h (u) du + 4 x − y − sinh(ux) cos(uy)uh(u)du 16 0 0 Z ∞ Z ∞ cosh(ux) cos(uy)h (u) du − 8xy cosh(ux) sin(uy)uh (u) du +8x 0 0 With these equations, we can represent the slopes of the functions where the real parts or the imaginary parts of ξR are zero by ∂ − ∂x < ξR 21 + 2x + 2ıy mRe (x + ıy) = ∂ 1 < ξ + 2x + 2ıy R ∂y 2 and ∂ − ∂x = ξR mIm (x + ıy) = ∂ = ξR ∂y 1 2 + 2x + 2ıy . 1 + 2x + 2ıy 2 1 Implicitly we assume that the domain of m is w | < ξ + 2w = 0 and the Re 2 1 domain of mIm is w | = ξ + 2w = 0 . 2 ∂ 1 1 + 2w = 0 is negative if < ξ + 2x + 2ıy and The slope of < ξ R 2 ∂x 2 1 ∂ 1 < ξ + 2x + 2ıy have the same sign, and likewise for the slope of = ξ + 2w = R 2 ∂y 2 0. Image 3, which is a graph of ξE , helps us visualize these slopes. From analyzing these and other graphs, we were able to develop the following conjectures. Conjecture 1. (Slope Conjecture) For = (w) > 0, mRe 12 + 2w and mIm 12 + 2w are positive when < (w) < 0 and negative when < (w) > 0. All analytic functionssatisfy the equality f (−w) = f (−w), and thus our function f (w) = ξR 21 + 2w satisfies f (−w) = f (w). Thus, by symmetry it suffices to prove just one of the cases for < (w). From this conjecture, we can deduce that as < (w) approaches zero the slopes must also approach zero or become undefined. If we wish to prove that mRe 21 + 2w approaches zero as Re (w) approaches zero, we must show that as < (w) approaches zero, 1 ∂ < ξ + 2x + 2ıy approaches zero faster than the partial derivative with R 2 ∂x respect to y approaches zero. Otherwise, the slope becomes undefined. Conjecture 2. The partial derivative with respect to x of < ξ 21 + 2x + 2ıy = 0 approaches zero faster than the partial derivative with respect to y as x approaches zero. Another way to look at the slopes of these level sets is to notice that two vectors are perpendicular when one vector’s slope is the negative reciprocal of the other 1 1 vector’s. So for all w ∈ C satisfying < ξR 2w + 2 = = ξR 2w + 2 = 0, the equation mRe (w)mIm (w) = −1 must also be satisfied. We know from before that this occurs when <(w) = 0. But the Slope Conjecture implies that this never occurs outside of the critical line because the slopes have the same sign. So for all INVESTIGATING ZETA FUNCTIONS 9 points w ∈ C − 0 satisfying either < ξ 12 + 2w = 0 or Im ξ 12 + 2w = 0, mRe (w)mIm (w) is nonnegative. But in order to prove that the level sets are never perpendicular outside of the critical strip, we only need to show a weaker case, the product of the slopes,mRe (w)mIm (w), for a fixed w is always greater than −1. Let Z ∞ A = cosh(ux) cos(uy)h(u)du, 0 Z ∞ sinh(ux) cos(uy)uh(u)du, B = 0 Z ∞ C = sinh(ux) sin(uy)h(u)du, 0 Z ∞ cosh(ux) sin(uy)h(u)du. D = 0 Plugging these into the equations for the real and imaginary parts of ξR and the partial derivatives we find 1 1 < ξR + 2x + 2ıy = 4 x2 − y 2 − 16 A + 12 − 8xyC 2 1 1 = ξR + 2x + 2ıy = 4 x2 − y 2 − 16 C + 8xyA 2 1 ∂ 1 + 2x + 2ıy = 8xA + 4 x2 − y 2 − 16 < ξR B − 8yC − 8xyD ∂x 2 ∂ 1 1 < ξR + 2x + 2ıy = −8xC − 4 x2 − y 2 − 16 D − 8yA − 8xyB ∂y 2 ∂ 1 1 = ξR + 2x + 2ıy = 8xC + 4 x2 − y 2 − 16 D + 8yA + 8xyB ∂x 2 ∂ 1 1 = ξR + 2x + 2ıy = 8xA + 4 x2 − y 2 − 16 B − 8yC − 8xyD. ∂y 2 (x2 −y2 − 161 )A 1 Now for x, y ∈ R such that < ξR 2x + 2ıy + 12 = 0, C = + 16xy . 2xy For x, y ∈ C such that = ξR 2x + 2ıy + 21 = 0, C = x2−2xyA . Whence, we ( −y2 − 161 ) can rewrite the slopes of the level sets we have been studying in terms of A, B, and D. 10 AMANDA KNECHT Conjecture 3. (Weaker statement that implies the Slope Conjecture) Define A, B and D as before and write 1 −8x(A−Dy)−4(x2 −y 2 − 16 )[B−A/x]+ 2x1 F (x, y) = 1 1 2 2 −8y(A+By)−4x(x −y − 16 )(A/y+D)− 2y 2 G(x, y) = 1 −16x2 yA−8y (x2 −y 2 − 16 )(A+Bx)−4D(x2 −y2 − 161 ) 2 1 −16xy 2 A+8x(x2 −y 2 − 16 )(A−Dy)+4B (x2 −y2 − 161 ) . Then F (x, y) · G(x, y) > −1 if x 6= 0 and w = x + ıy is either in the domain of mRe 21 + 2w or in the domain of mIm 12 + 2w . Now, instead of only thinking about the derivatives withd respect to x 1and d 1 y, we can also analyze the vectors dw < ξR 2w + 2 and dw = ξR 2w + 2 . Some simple observations make our study of these vectors easier. First, we know (z) (z) that < (f (z)) = f (z)+f and =(f (z)) = f (z)+f for any function f (z). Now 2 2 when f (z) is analytic, ∂ < (f ∂z ∂ ∂z f (z) = 0. Thus we find that for analytic functions ∂ d (z) and ∂z = (f (z)) = −ı f (z) . We know from before that, 2 dz Z ∞ 1 1 1 ξR 2w + cosh(uw) h (u) du. = + 4 w2 − 2 2 16 0 (z)) = 1 ∂ f 2 ∂z Thus, Z ∞ Z ∞ ∂ 1 1 2 ξR 2w + = 8w cosh(uw) h(u)du+4 w − sinh(uw) uh(u)du. ∂w 2 16 0 0 Applying the formulae above to ξR , we find: Z ∞ Z ∞ 1 ∂ 1 2 cosh(uw) h(u)du+2 w − sinh(uw) uh(u)du. < ξR 2w + = 4w ∂w 2 16 0 0 Z ∞ Z ∞ ∂ 1 1 2 = ξR 2w + sinh uw uh (u) du. = −4ıw cosh uw h (u) du−2ı w − ∂w 2 16 0 0 Now in order to understand these partial derivatives, we returned to analyzing the graphs of ξR . This time we concentrated on the how the lines < ξR 2w + 21 = 1 0 and = ξR 2w + = 0 change as the point w changes. We know that 2 ∂ ∂z = 1 2 ∂ ∂x ∂ − ı ∂y . Thus both ∂ ∂x and ∂ ∂y ∂ ∂z ∂ if ∂z are negative if and only if is in ∂ ∂ the second quadrant. Likewise, both ∂x and ∂y are positive if and only is in the fourth quadrant. We can use this information and the analysis of the graphs to make the following conjecture. Conjecture 4. (Equivalent to the Slope Conjecture) For < (w) > 0 and = (w) > 0: ∂ • If < ξR 2w + 21 = 0, then the vector ∂w < ξR 2w + 12 is in the second quadrant when = ξR 2w + 21 > 0 and in the fourth quadrant when = ξR 2w + 12 < 0. 1 ∂ • If = ξR 2w + 2 = 0, then the vector ∂w = ξR 2w + 12 is in the second quadrant when < ξR 2w + 21 < 0 and in the fourth quadrant when < ξR 2w + 12 > 0. INVESTIGATING ZETA FUNCTIONS 11 Now that we have discovered where these level sets intersect, we will turn our attention to the roots created at the intersection of the level sets <(f (z)) = 0 and =(f (z)) = 0 of an analytic function f (z). Images 4 and 5 let us visualize roots of an analytic function. These images suggest the following lemmas. Lemma 1. If the gradient of an analytic function, f (z), is not equal to zero at z0 , then the lines < (f (z)) = < (f (z0 )) and = (f (z)) = = (f (z0 )) intersect perpendicularly. Proof: We know that a function f (x, y) = u (x, y) + ıv (x, y) is analytic if it satisfies the Cauchy Riemann equations: ux = vy uy = −vx . We see that the gradients of u (x, y) and v (x, y) are (ux , uy ) and (vx , vy ), respectively. Because we are assuming our function f (x, y) is analytic, we use the Cauchy Riemann equations to notice that the gradient of v (x, y) is also (−uy , ux ) . And we can represent the vectors tangent to the level sets by u⊥ (z0 ) = (−uy (z0 ) , ux (z0 )) and v⊥ (z0 ) = (ux (z0 ) , uy (z0 )) . Now we see that when the gradient is non-zero, they form perpendicular angles because (−uy (z0 ) , ux (z0 )) · (ux (z0 ) , uy (z0 )) = 0. Lemma 2. Roots of analytic functions are simple if and only if the gradients of the real and imaginary parts of the function are not zero vectors. Proof: Let f (z) be an analytic function with a root at z0 . First suppose z0 is a simple root. Then we can write f (z) = (z − z0 ) g (z) where g (z0 ) is not equal to zero. Letting z = x + ıy, z0 = x0 + ıy0 , and g (z) = g1 (z) + ıg2 (z) we find: f (z) = [(x − x0 ) g1 (z) − (y − y0 ) g2 (z)] + ı[(x − x0 ) g2 (z) + (y − y0 ) g1 (z)]. Taking partial derivatives of the real and imaginary parts gives us: ∇< = (x − x0 ) g1x (z) − (y − y0 ) g2x (z) + g1 (z) , (x − x0 ) g1y (z) − (y − y0 ) g2y (z) − g2 (z) . ∇= = (x − x0 ) g2x (z) + (y − y0 ) g1x (z) + g2 (z) , (x − x0 ) g2y (z) + (y − y0 ) g1y (z) + g1 (z) . So , ∇Re (z0 ) = (g1 (z0 ) , −g2 (z0 )) and ∇Im (z0 ) = (g2 (z0 ) , g1 (z0 )) . And g (z0 ) 6= 0 implies that g1 (z0 ) and g2 (z0 ) are never simultaneously zero. Therefore the gradients are not zero vectors. Now suppose the gradients are not zero vectors. We can write f (z) = (z − z0 ) g (z) as before, but this time (z − z0 ) might divide g (z). Just as before, ∇Re (z0 ) = (g1 (z0 ) , −g2 (z0 )) and ∇Im (z0 ) = (g2 (z0 ) , g1 (z0 )) . Because g (z0 ) = g1 (z0 ) + ıg2 (z0 ) and the gradients are not zero vectors, g (z0 ) is not equal to zero. Thus, (z − z0 ) does not divide g (z), so z0 is a simple root. Lemma 3. The mth -order roots of an analytic function occur only when the level sets, < (f (z)) = 0 and = (f (z)) = 0, of the function intersect to create 4m angles π measuring 2m radians each. 12 AMANDA KNECHT Proof: Suppose f (z) is analytic at z0 and has an mth -order root at z0 . Then f (z0 ) = f 0 (z0 ) = f 00 (z0 ) = . . . = f (m−1) (z0 ) = 0 but f (m) (z0 ) 6= 0. 2 f (m+1) (z0 ) f (m+2) (z0 ) Letting g (z) = (m+1)f (m) (z ) (z − z0 ) + (m+2)(m+1)f (m) (z ) (z − z0 ) + . . ., 0 0 f (z) = (z − z0 )m f (m)(z0 ) (1 + g (z)) , which is valid in a neighborhood of z0 . m! Now let θ0 = lim arg (z − z0 ) z→z0 φ0 = lim arg (f (z)) z→z0 Then, f (m) (z0 ) (1 + g (z)) . φ0 = lim arg (z − z0 ) z→z0 m! And because arg (xyz) = arg (x) + arg (y) + arg (z) for any x, z, y ∈ C, 1 + g (z) m (m) φ0= lim arg (z − z0 ) + arg f (z0 ) + arg . z→z0 m! g(z) → 0 as z → z0 , so φ0→ m arg (z − z0 ) + arg f (m) (z0 ) as z → z0 . m We will now let α denote the angle between two arcs C1 and C2 passing through z0 and let β denote the angle between the arcs Γ1 = f (C1 ) and Γ2 = f (C2 ) . Then α = θ1 − θ2 and β = φ1 − φ2 . And so, β = mθ1 − arg f (m) (z0 ) − mθ2 + arg f (m) (z0 ) . = m (θ1 − θ2 ) . = mα. Letting the arc C1 be an arc in the level set < (f (z)) = 0 and C2 be an arc in the level set = (f (z)) = 0 the function f maps C1 and C2 to the real and imaginary axes, respectively. We know that the angle between the real and imaginary axes . on the complex plane is π2 +2kπ radians. So β = mα = π2 +2kπ implies α = π+4kπ 2m Thus the angle between the level sets < (f (z)) = 0 and = (f (z)) = 0 is a multiple π of 2m radians. Now let c = f (m)(z0 ) m! 6= 0 and write m p m f (z) = (z − z0 ) c (1 + g (z)) = (z − z0 ) c + 1 + g(z) . m Define a new function h(z) by p h(z) = (z − z0 ) m c (1 + g (z)). p Observe that since g(z0 ) = 0, m c + 1 + g(z) is well √ defined in a neighborhood of z0 . Because g(z) → 0 as z → z0 , h0 (z0 ) = m c 6= 0. Thus by the Inverse Function Theorem, h−1 (z) exists in a neighborhood of z = z0 . Now we know kπ the angle between the curves C1 = z|arg(z) = 2m and C2 = {z|arg(z) = 0} kπ −1 is 2m radians. Because h preserves angles, the angle between h−1 (C1 ) and kπ h−1 (C1 ) is 2m radians. Knowing f (z) = (h(z))m , the angle between f (h−1 (C1 )) = INVESTIGATING ZETA FUNCTIONS 13 z|arg(z) = kπ and f (h−1 (C2 )) = {z|arg(z) = 0} is kπ radians. Since the images 2 2 of these curves are the real and imaginary axes for every k ∈ Z, 4m angles are created. 3. Consequences of the Slope Conjecture The proof of the following statements rests entirely on the previously stated lemmas and the proof of the Slope Conjecture. Consequence 1. The only non-trivial roots of the Riemann Zeta Function, ζR can be found on its critical line, < (w) = 0. 1 2 + 2w , Proof: If roots did exist outside the critical line, they would be the result of the level sets of the real and imaginary parts of ξ intersecting. But the Slope Conjecture tells us that on either side of the line <(z) = 1/2 the slopes of the level sets < ξR 21 + 2w = 0 and = ξR 21 + 2w = 0 have the same sign. Thus, outside of this line the level sets do not intersect. Therefore, the only simple roots of ξR 12 + 2w occur when < (w) = 0. We know that 1 1 1 − 12 −2w ξR + 2w = π Γ + 2w ζR + 2w . 2 2 2 And thus ζR 1 + 2w 2 = ξR 1 1 2 π − 2 −2w Γ + 2w 1 2 + 2w . 1 Realizing that π − 2 −2w is never zero and that Γ only has simple roots at w = −1 −3 −5 , 4 , 4 , . . ., where we know the trivial zeros of ζ lie, we conclude that ζR 21 + 2w 4 has non-trivial roots exactly when ξR 21 + 2w = 0. And thus, ζR 21 + 2w = 0 only when < (w) = 0. Consequence 2. All of the roots of the Riemann Zeta Function are simple. Proof: We know from the Slope Conjecture that the only roots of ζR are on the line <(z) = 1/2. Now if these slopes were not simple, then for <(w) > 0 either the slope mRe ( 21 + 2w) or mIm ( 21 + 2w) would have to approach a number greater than or equal to 1 as w → 0. But this would contradict the Slope Conjecture; thus all roots are simple. 4. The Epstein Zeta Function The Epstein Zeta function is a generalization of the Riemann Zeta function and is defined as follows, 14 AMANDA KNECHT Definition 2. [Te] Epstein’s zeta function of Y , an n × n, positive symmetric matrix, and s ∈ C, with < (s) > n/2 is Z (Y, s) = X Y [v]−s . v∈Zn −0 Here v is a column vector and Y [v] =t vY v with t v = transpose of v. Thus P Y [v] = ni,j=1 yij vi vj . In the special case where n = 1 and Y = 1, we find that Z (1, s) = ζR (2s) . The general Epstein zeta function has an analytic continuation similar to that of ζR (s). Terras [Te] gives the following theorem for the analytic continuation. Theorem 1. The Analytic Continuation of Epstein’s Zeta Function Let R∞ γ (s, x) be the incomplete gamma function defined by γ (s, x) = x us−1 e−u du, and set −s Z G (s, x) = x γ (s, x) = ∞ u(s−1) e−xu du. 1 Then Epstein’s zeta function can be analytically continued to all s ∈ C with its only pole a simple one at s = n/2. Here |Y | denotes the determinant of Y . The analytic continuation comes from the incomplete gamma expansion: Λ(Y, s) = π −s Γ(s)Z(Y, s) P |−1/2 1 − 2s + 21 v∈Zn −0 G (s, πY [v]) + |Y |−1/2 G = |Y2s−n n 2 − s, πY −1 [v] . Thus Epstein’s Zeta function satisfies the functional equation Λ(Y, s) = |Y |−1/2 Λ(Y −1 , n/2 − s). Now for a 2 × 2 matrix Y, we define a new function ξE (Y, S) that is defined on the entire complex plane and shares its roots with the nontrivial roots of Z(Y, s). To do this we merely alter the function Λ(Y, s), which is defined in Theorem 4.1. But before defining our new function, we first simplify our symmetric 2 × 2 matrix. Say Y = cM where M is a symmetric matrix and c is a nonzero conP −s stant. Then Z(Y, s) = Z(cM, s) = = c−s Z(M, s), and c−s v∈Zn −0 (cM [v]) INVESTIGATING ZETA FUNCTIONS is never zero. So given a matrix Y ,let c = 1 |Y |1/n and then M = 15 1 Y |Y | and 1 n |Y | = 1. Thus we see that every symmetric matrix can be written ||Y |1/n | as a product of a nonzero constant and a matrix with determinant 1, and every |M | = Epstein Zeta function can be written as the product c−s Z(M, s) with |M | = 1. So we will restrict our study to matrices with determinant 1. Earlier discov t 0 eries of nontrivial zeros of Z(Y, s), [Te], when Y is of the form and 0 1/t t = 10, 100, 1000 . . . prompts us to simplify even further to matrices of this form. 1/t 0 −1 Now with this form of matrix we see that because Y = , 0 t X X X X 1 1 2 Y −1 [v]. a + tb2 = Y [v] = ta2 + b2 = t t v∈Zn −0 v∈Zn −0 a,b∈Z a,b∈Z not both zero not both zero Finally we can define our new function ξE (Y, s) = = s(2s − 2)Λ (1 − s) X = 1 + s (s − 1) (G (s, πY [v]) + G (1 − s, πY [v])). v∈Z2 −0 Letting s = 1 2 + x + ıy since the line of symmetry is <(z) = 1/2 for a 2 × 2 matrix 1 ξE (Y, + x + ıy) = 1+ 2 X 1 1 1 2 2 −1 − + x − y + 2ıxy G + x + ıy, πY [v] + G − x − ıy, πY [v] . 4 2 2 2 v∈Z −0 Because our column vector has two elements which are not both zero, say a, b ∈ Z, we can define a new function f (a, b) = πY [v] = π ta2 + 1t b2 and rewrite t 0 1 ξE , 2 + x + ıy as 0 1/t 1 t 0 ξE , + x + ıy = 0 1/t 2 Z ∞ 1 X 1 1 2 2 = 1 + − + x − y + 2ıxy e−uf (a,b) u− 2 +x+ıy + u− 2 −x−ıy du 4 1 a,b∈Z not both zero 1 = 1 + − + x2 − y 2 + 2ıxy 4 X a,b∈Z not both zero Z 1 ∞ 2 e−uf (a,b) √ cosh ((x + ıy)(ln u)) du. u 16 AMANDA KNECHT Knowing that cosh(x + ıy) = cos(y) cosh(x) + ı sin(y) sinh(x), we write the real and imaginary parts of ξE 1 t 0 < ξE , + x + ıy = 0 1/t 2 Z ∞ X 2 1 2 2 e−uf (a,b) √ cos(y ln u) cosh(x ln u)du = 1+ − +x −y 4 u 1 a,b∈Z not both zero −2xy 1 a,b∈Z not both zero t 0 0 1/t X = ξE = 2xy ∞ Z X 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du. u 1 = , + x + ıy 2 Z ∞ 2 e−uf (a,b) √ cos(y ln u) cosh(x ln u)du u 1 a,b∈Z not both zero 1 + − + x2 − y 2 4 Z X a,b∈Z not both zero 1 ∞ 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du. u Again we take the partial derivatives of these functions. ∂ 1 t 0 < ξE , + x + ıy = 0 1/t ∂x 2 Z ∞ X 1 2 2 2 = − +x −y e−uf (a,b) √ cos(y ln u) sinh(x ln u)(ln u)du 4 u 1 a,b∈Z X +2x Z e 1 a,b∈Z not both zero X −2y ∞ Z a,b∈Z not both zero −2xy not both zero ∞ −uf (a,b) X a,b∈Z not both zero 1 Z 1 2 √ cos(y ln u) cosh(x ln u)du u 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du u ∞ 2 e−uf (a,b) √ sin(y ln u) cosh(x ln u)(ln u)du. u INVESTIGATING ZETA FUNCTIONS 17 ∂ 1 t 0 < ξE , + x + ıy = 0 1/t ∂y 2 Z ∞ X 2 1 2 2 e−uf (a,b) √ sin(y ln u) cosh(x ln u)(ln u)du = − − +x −y 4 u 1 a,b∈Z not both zero X −2y Z −2x ∞ 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du u Z a,b∈Z not both zero X −2xy 2 e−uf (a,b) √ cos(y ln u) cosh(x ln u)du u 1 a,b∈Z not both zero X ∞ 1 ∞ Z 1 a,b∈Z not both zero 2 e−uf (a,b) √ cos(y ln u) sinh(x ln u)(ln u)du. u 1 ∂ t 0 = ξE , + x + ıy = 0 1/t ∂x 2 Z ∞ X 1 2 2 2 = − +x −y e−uf (a,b) √ sin(y ln u) cosh(x ln u)(ln u)du 4 u 1 a,b∈Z X +2y e 1 a,b∈Z not both zero X +2x Z a,b∈Z not both zero +2xy not both zero ∞ −uf (a,b) Z X a,b∈Z not both zero ∞ 1 Z 1 2 √ cos(y ln u) cosh(x ln u)du u 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du u ∞ 2 e−uf (a,b) √ cos(y ln u) sinh(x ln u)(ln u)du. u 18 AMANDA KNECHT ∂ 1 t 0 = ξE , + x + ıy = 0 1/t ∂y 2 Z ∞ X 2 1 2 2 e−uf (a,b) √ cos(y ln u) sinh(x ln u)(ln u)du = − +x −y 4 u 1 a,b∈Z Z X +2x e 1 a,b∈Z not both zero a,b∈Z not both zero 1 Z X −2xy ∞ Z X −2y not both zero ∞ −uf (a,b) a,b∈Z not both zero 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du u ∞ 1 2 √ cos(y ln u) cosh(x ln u)du u 2 e−uf (a,b) √ sin(y ln u) cosh(x ln u)(ln u)du. u As before we can define new functions A, B, C and D. Z ∞ X 2 e−uf (a,b) √ cos(y ln u) sinh(x ln u)du(ln u) A = u 1 a,b∈Z not both zero B = X a,b∈Z not both zero C = X a,b∈Z not both zero D = X a,b∈Z not both zero Z ∞ 2 e−uf (a,b) √ cos(y ln u) cosh(x ln u)du u ∞ 2 e−uf (a,b) √ sin(y ln u) sinh(x ln u)du u ∞ 2 e−uf (a,b) √ sin(y ln u) cosh(x ln u)(ln u)du. u 1 Z 1 Z 1 With these new functions we can more easily express the partial derivatives and show that ξE ∂ < ξE ∂x ∂ < ξE ∂y ∂ = ξE ∂x ∂ = ξE ∂y is an analytic function. 1 t 0 , + x + ıy = − 14 + x2 − y 2 A + 2xB − 2yC − 2xyD 0 1/t 2 1 t 0 , + x + ıy = −2xyA − 2yB − 2xC − − 14 + x2 − y 2 D 0 1/t 2 1 t 0 , + x + ıy = 2xyA + 2yB + 2xC + − 41 + x2 − y 2 D 0 1/t 2 1 t 0 , + x + ıy = − 14 + x2 − y 2 A + 2xB − 2yC − 2xyD. 0 1/t 2 Now we turn our study from an investigation of the change of t 0 1 ξE , 2 + x + ıy as x and y change to a study of how the function 0 1/t INVESTIGATING ZETA FUNCTIONS 19 changes with respect to t by taking the derivative with respect to t of ξE . ∂ 1 t 0 ξE , + x + ıy = 0 1/t ∂t 2 1 2 2 = π − + x − y + 2ıxy · 4 Z ∞ 1 X 1 b2 2 −ue−uf (a,b) u− 2 +x+ıy + u− 2 −x−ıy du a − 2 t 1 a,b∈Z not both zero 1 2 2 = π − + x − y + 2ıxy · 4 Z ∞ X 2 b2 2 −ue−uf (a,b) √ cosh((1 + x) ln u)du. a − 2 t u 1 a,b∈Z not both zero This expression can be simplified to ∂ 1 t 0 = ξE , + x + ıy 0 1/t ∂t 2 X 1 2 2 = −π − + x − y + 2ıxy 4 a,b∈Z b2 a − 2 t 2 · not both zero b2 1 b2 3 2 2 + x + ıy, π ta + + G − − x − ıy, π ta + . G 2 t 2 t Similarly to the investigation of ξR , we studied the images of the graphs for different values of t, which are included in the appendix. The numerical data produced by these graphs suggests that when t is sufficiently large the analogue of the Slope Conjecture fails. However the conjecture does seem to be valid when 0 < t < 7. Also the analogue of the Riemann Hypothesis fails for larger t. The non-trivial roots of ξE we encountered are results of the slopes of the level sets < (ξE (s)) = 0 and = (ξE (s)) = 0 being both positive and negative for <(s) > 1/2. The first five images were created by the program Fractal Explorer written by Peter Stone. The first image is the four color graph of the complex plane. Points colored blue have positive real and imaginary parts, while red points have negative real and positive imaginary parts. Green and yellow are similar. When this program graphs a function f (z), the location of the point is the value z and 24 AMANDA KNECHT the color represents the quadrant f maps z to. The last images, those graphing ξE , were created by a graphing applet written by Ken Richardson. For these images, purple represents the first, yellow the second, blue the third, and red the fourth quadrant. References [Ed] [Pa] [Te] [Ti] H.M. Edwards. Riemann’s zeta function. Dover Publications, Mineola, New York, 1974. S.J. Patterson. An introduction to the theory of the Riemann zeta-function. Cambridge University Press, Cambridge, 1989. Audrey Terras. Harmonic analysis on symmetric spaces and applications.I. SpiringerVerlang, New York, 1985. E.C. Titchmarsh. The theory of the Riemann zeta-function. Oxford University Press, London, 1951.