DETERMINANTS OF LAPLACIANS by Marek Kierlanczyk M.S. Warsaw University (1979) Submitted to the Department of Mathematics in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY IN MATHEMATICS at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY Marek Kierlanczyk 1986 The author hereby grants to M.I.T. permission to reproduce and to distribute copies of this thesis document in whole or in part. SIGNATURE OF AUTHOR: i Department of Mathematics June 10, 1986 CERTIFIED BY Isadore M. Singer Thesis Supervisor ACCEPTED BY: A U Nesmith C. Ankeny Chairman, Departmental Graduate Committee MA SSACHUSE TT S NSTITUTc OF TECHNOLOGY .FP 96 1986 IRARI 3 ARCHIVES -2- DETERMINANTS OF LAPLACIANS by Marek Kierlanczyk Submitted to the Department of Mathematics on June 10, 1986, in partial fulfillment of the Requirements for the Degree of Doctor of Philosophy in Mathematics, ABSTRACT This thesis deals with determinants of Laplacian and the square of Dirac operators. For tori they are computed in terms of theta functions, and for compact Riemann surfaces of genus at least two in terms of Selberg zeta functions. are presented. Applications to conformal geometry Finally, we discuss some obstructions to the existence of parallel spinors and symplectic structures. Thesis Supervisor: Isadore M. Singer Title: John D. MacArthur Professor of Mathematics -3- TABLE OF CONTENTS Page INTRODUCTION ................... 5 CHAPTER 1 DETERMINANTS OF LAPLACIAN AND DIRAC ON THE TORUS......................................... 7 CHAPTER 2 DETERMINANTS OF LAPLACIAN AND DIRAC ON THE RIEMANN SURFACE............................ .23 CHAPTER 3 SOME ASPECTS OF CONFORMAL GEOMETRY ............... 31 CHAPTER 4 PARALLEL SPINORS AND SYMPLECTIC STRUCTURES........ 40 -4- ACKNOWLEDGEMENTS I wish to warmly thank my thesis advisor, Professor Isadore M. Singer, for his teaching, guidance, and encouragement. Thanks also go to the Mathematics Department of the Massachusetts Institute of Technology and the University of California at Berkeley. Finally, I would like to thank my wife, Ki, for her moral support through the turbulent times of my studies. -5- INTRODUCTION Determinants of elliptic operators, especially of a and Dirac ' operators on compact Riemann surfaces, have attracted wide attention among mathematicians and physicists. The increasing evidence that string theories can provide models that unify gravitation and matter into a finite quantum theory generated a lot of interesting mathematical questions. In this thesis, we will give the answer to a few of them. In Chapter 1 we derive formulas for determinants of Laplacian and 2 on two-dimensional tori in terms of theta functions. General- izations to the n-dimensional case are discussed. In Chapter 2 we derive the formulas for the determinant of the Laplacian on compact Riemann surfaces in terms of the Selberg zeta function and for det( 2 + a) as well. In Chapter 3 we analyze a new conformal invariant discovered recently by Parker-Rosenberg. is not a conformal invariant. We also show that the determinant of 2 -6- In Chapter 4 we prove that there are no parallel spinors on Riemann surfaces of genus greater than one. We also analyze some consequences of Thurston's conjecture about the existence of a symplectic structure on almost complex manifolds with a nonnilpotent second cohomology class. -7- CHAPTER 1 DETERMINANTS OF LAPLACIAN AND DIRAC ON THE TORUS Let W be a compact complex analytic manifold without boundary, complex dimension N. complex (p,q) If D 'q(W,L) of denotes the space of C forms on W with values in a flat holomorphic vector bundle L, then the exterior differential d splits: d = d' + d", where Dp+1,q d':D 'q + d":DP' +- D Pq+ Since the notation 3 is often used for the operator which we write as d", the system of complexes which it usually called the a-complex. p, Let Z L We are using the notation of [1]. denote the kernel of d" in D '9(WL). cohomology groups H '9(WL) , defines for each choice of p is where 0 = Z 'q/d"Dp 'q The Dolbeault (WL) are isomorphic to is the sheaf of germs of forms of type (p,O) with holomorphic coefficients. -8- X be a finite dimensional unitary representation of the Now let fundamental group 7 (W), bundle. Let Dp'q(W,X) and let L(X) be the associated complex vector = D 'q(WL(X)). If W is the simply connected covering manifold of W, with 7 l acting as deck transformations, a form f in D(W,X) may be identified with a vector valued form f on W which satisfies foy X(Y)f, yeff . Suppose W has a Hermitian metric. Since X is unitary, the associated duality operator * satisfies *:Dp'q(WX *D-q,N-p(WX), and determines the inner product (f,g) in D P). Since the vector bundle L(X) is flat, the exterior differential d has the formal adjoint -*d* = '+ V" with 6' = - ' 6" = -d* Let A +DP~lp + D q be the corresponding Laplacian pq A p,q -q =-(6@@dl + d"": -9- Let HP)9(W,X) be the space of harmonic forms in that is, satisfying Af = 0. (X), The Hodge theorem states each harmonic form is closed (and coclosed), and that the resulting inclusion map of H pq(WX) into the Dolbeault group HP q(WL) is an isomorphism onto. When the metric is KMhler, w = is closed, 2 h that is, when the 2-form dzi A dzk jk then Apq = -(1/2)(6d + d6), and the Dolbeault groups may be identified with subgroups of the cohomology of W with coefficients in L(X). In this case, one has A - 3 2 A where A is the Hodge Laplacian dd* + d*d. A generates a semi-group of compact operators exp(tA As t on the Hilbert space completion of D p,(W,X). operators approach the projections P HPq (W,X). of D '9(W,X) + + c, ), t > 0, these onto the sub-space The zeta function associated with the Laplacian A is defined by tA p~ (S,X) = (S) 0 t itr(e q - Pq )dt = Z X0(-x )-s X <0 n n for Re s large, the sum running over the non-zero eigenvalues Xn of A p~q. (s,X) extends to a meromorphic function in the s-plane, which is analytic at s = 0. dtr (A-s) ds d ds -s j Here -s is -10- Formally, at s = 0, this would be Z log X= log det A = log det A. Hence, we define (S) s=0 Consider the 2-dimensional torus, T = C/F, where r is the lattice generated by 1 and T, ImT > 0. Laplacian A5. p = q = Consider the corresponding We assume that our representation X is trivial and 0. Theorem 1.1 det A7 = (IMT)2 1 rI(t) 4 where II li 2 TrnT (1 - e2Wn =RelT n=1 Proof The eigenvalues of A on D*(W,l) are in,n 42 (I 2 2 I-~ - n) Applying the Poisson summation formula we obtain -lmit + nI2 4tme tr(et m,n -11- Now we have CO c(s) = 1 f ts- -tA 1tr(e- )dt 0 _ InT 1 + 1f IMT 4rF(s) F(s) 47(s-1) -I+n12 4t s-2 e dt 0 00 + where E = fts-i (tr et)dt sum over all (m,n) 0 (0,0). When Re s < 0, we apply the expression for tr(e-t ) in the last term on the right to obtain - lInt -mt+n1 2 4t ts-2 e ' IT = 4TrF(s) dt = 0 F(1-s) 4U F ' j s) 4 ImT + ni 1-s 2* Consider E(T,u) (- IMT ImT+nI u , Re u > 1 This function can be continued analytically into a function of u regular for Re u > 1/2 except for a simple pole at u=1 with the residue T. E(Tu) At u=1 it has an expansion: = + 2R(Y - log2 - log((ImT) |T1(T)1 )) + a (u-1) + It is called the first limit formula of Kronecker. -12- Rewriting C(s) further we have = C (S) lInT IT (I 41 F(l-s) r- 1-s E (T,1-s) (ImT) lint Si1 = ( ) [S.-2F'(l)s 2 [s-2F1(1)s+...] =( -) 1 [- +..] [- 2. + 2Tr(y-log(2(Imt) z 2 (t)I ))+...] + 2 (Y-log(2(Imt) 2 2 1 = [1 + 2log(2(ImT)2 If(T) 2) s + ... ] -(n)s as F'(l) = -y Therefore 1 V(0) =-[log( 2 + 2log(2(ImT) I) 2 and finally det A- = exp(-C'(0)) = (Im2T)2 1(T)14 which ends the proof. Observe that the obtained expression is not SL(2,Z) invariant; that is not surprising as the eigenvalues X do not have this property also. Notice that all of the essential ingredients: the Poisson summation formula, the Kronecker limit formula, and the explicit formula for eigenvalues of the Laplacian have n-dimensional analogues. Although we are considering the Hodge Laplacian dd* + d*d acting on smooth functions, our result extends to forms. The identity -13- I....d x A (f dx ) dx (A f)dx 0 p 1 P 11 i 1 shows that the p-forms spectrum is the same as the function spectrum, each repeated (n) times. p determines det A Therefore the det A completely. p 0 Now we will gather some information which will be useful in deriving a formula for the determinant of Laplacian on an n-dimensional torus T = Rn/L, L = AZn, The metric structure of Rn A C GL(n;R). projects to T such that vol(T) = Idet Al. The set L = {a C Rn n t V a E L} is the dual lattice of L; L = (A-i) Zn. 2 ~2 are 42 Hal a a C Z, The eigenvalues of T for a arbitrary in L where IIIIis the Euclidean norm. The Poisson summation formula states -a trke -tA )L det A n 2 (41 t) e aEZ tA)a 4t n so ?(s) reads CO detA n S)= (47)2 det A n 7 TT 4 r() f acZ- 0 s- n_ t 2 -attA)a 4t e 0 F(-n - s) 2 (S) n-s (at (AtA)a) - dt -14- Let S be a n x n matrix of a positive definite real quadratic form and let p be a complex variable with Re p > n/2. Then Epstein's zeta function is defined by Z (S,p)=! n 2 I , (ta Sa) aeZn-O where the sum is over all column vectors with integral coordinates, not all of which are zero. Define n k (S) n n n lim m. {Z (SP) n 2 I'( )IdetSI 2 1 12 ~ - n 2 2 This is just the constant term in the Laurent expansion of the Epstein zeta function at p = n/2. We will introduce some notation following A. Terras [Z]* n = n, + n 2 , with 1 < ni < n-l. (n) S(n) [T(nP) 0 0 S'2 12 1 1 and T I n2) (n2 _ t Let n Then S can be represented as t 22] -01 2 Let - t. *) I am grateful to P. Sarnak for this reference. L_ 1 -15- =t Define z{b} = 1 1 bQ + i t 2 (S -l [b]) 2 , n Y - log 2 - for n even, r=1 c n = < 1 -(n-3) , (2r+l) y -E for n odd r=0 Then k(S) = Zn-_1 2p n 2 + Idet S 2 (-) 2 2 1 X{c n - log(t2 II (1 - exp{27iz{b}})I 2 bEZn-1-0,b(mod+l) This is an n-dimensional generalization of the Kronecker first limit formula. forts In the 2-dimensional case s = s s12 s2 -q ] 2 2 1 q t 0 0 1 0 s2 1 2 12 > 1 q ,andqs /s 2 12 ' -16- Then 1 lim {Z2 (S,p) - r(det 5) p-i p+1 TT = 1 {X - log 2 - log(t 2 TI(z)I 2 )} (det S)2 1 for z = q + i t 11 22 s2 = -1 s2 (S12 + i(det S? To obtain the expression for the determinant of the Laplacian on the n-dimensional torus, one uses Terras' functional equation [2] for the Epstein zeta function: 1 -. Pp)Z p 2 - I(P)Z (S P) = (detS) 2i 1 ~-2n F( 1 -p) Z (S _-11 ) -P) together with the generalizations of Kronecker's limit formula included on the previous page. -17- We assume the reader knows about the Dirac operator so we discuss it only briefly. Let X be a compact, orthonormal frames E is oriented riemannian manifold. a principal SO-bundle. The bundle of Suppose E lifts to give a principal Spin-bundle E; then X is a spin manifold and we can define via the spin representation a vector bundle V = E x S, the E lifts bundle of spinors. differ by a Z 2 1-cocycle, to E iff w2 (x) = 0 and any two liftings so the number of inequivalent liftings (the number of spin structures) is # H (X,Z2). V splits into V+ : C (V) - V C (V). In even dimensions, so one can define the desired Dirac operator In particular we get a decomposition of harmonic and negative H_ harmonic spinors H into the space of positive H spinors. Let X be a complex manifold. It turns out that X is a spin manifold when the canonical line bundle K has a holomorphic square root (W 2 c1 ()mod 2). Furthermore, the set of spin structures on X is correspondence with the [L:L 2 = K,L a holomorphic line bundle]. Choosing a spin structure means choosing an L, same as :C (AOeven d L) + in 1-1 so jf:V C (AO,odd 3 L). manifold we have H+ ~ Heven(X,(L)), H~ + V turns out to be the For a compact KAhler Hodd(X,0(L)). -18- Consider now a 1-dimensional torus T and the square of the Dirac operator on T. There are four of them, spin structures. The eigenvalues of these operators are -41T2 m,n 1 m + 4I2 2 where (e1 ,E2 ), corresponding to the underlying 1 2 2 - t(n + f cE) 2 i = 0,1 classifies the spin structures on T (which can be thought as Z2 characters of the fundamental group). See Friedrich [3] where the case of Tn is also to be found. A straightforward analysis, using Epstein's functional equation, leads to a formula for the determinant which amounts to the Kronecker's second limit formula. Theorem 1.4 1 det 2 det where (EE2) 1 1 2 e4 1 2 ak 2 - 1 =Ie2 (0,0) and 01 is the theta function 2ni(IkIT - akw) Tri(w + 1) C (w-n(T) 1 6 e 1 sign (k + ) k 2 - (l-e 2 Ik + 11 2 -19- Proof We will follow closely Ray-Singer [1]. Consider first ?(s) ' (s) - the zeta function for the Laplacian on the space D0 (W,X) of C sections of L(X). It is easy to see that the eigenvalues of W on D*(W,X) are m, = m,n - 42 (ImT)2 1 u+m - T(v+n)2 and the eigenfunctions corresponding to ,n are 2Vi Pmn(z) = exp Im(z(u + m - T(v + n))}. We have, of course, x tr(e = ) l t e m,n We can also write the heat kernel explicitly as Pt(z,z') = MnX(mT + n)pt (z,z' + mT + n), where p (zz') 1 t t4 -Iz-z'1 2/4t e is the euclidean heat kernel on C. tr(e t) = IMT Z M,n 4nt From this we obtain -|mt+n 2/4te 2Ti(imu+nv) i.e., the Poisson summation formula again. = -20- The above yields the analytic continuation of the zeta function. For Re s large we can write c(s) = Zm,n (-X m,n )-s 00 s-1 tr(e tA)dt 1ft 1 (s) IMT +IMT 4n(s-1) 4ur s + s 2 2 in+n >0 f 2Tri(mu+nv) 1s-2 f t0 e 0 2-m+n12/t etdt ts- tr(e tA)dt. r~)1 The right side defines a meromorphic function in the s-plane, which is the desired continuation; note that vanishes at s = 0. When Re s < 0, we can insert the second expression for tr(e ) in the last term on the right to obtain 2 _ 2Tri(mu+nv) f _4IMT 411F(s) IMT 40 2 2 m +n >0 F(l-s) F(s) ts-2e-Imt+n1 /4tdt 0 27i(mu+nv)( m2 2>0 4 1-s Imt + n 12 Since the series does not converge absolutely when s explain how to evaluate '(0). 0, let us We will assume v ' 0 (mod 1); if v = 0 we would have u 0 0 (mod 1), since the character is not trivial, and the interchange of u and -v is the same as replacing T by -1/T. Then following Siegel, one can establish the uniform convergence by summing first over n. Write -21- r(-s) c(s) =IT +Z 2lTinv n e2Timu zo e2 n=-M M-YO 4 )1-s n 4 2rinv( 1S lMT + ni Introduce the Dirichlet kernel A n and set bn = erinv sin TI(n+1)v sin Tv linT + n1 2s-2 Then 1 e 2minv zo n=--we I - -n(A -n lmT+n2-2s = z' < A)bl n-inb~ .co An (bn+l - bn ) 1 Z' = sin rv -O 1 < = sinnv b n+l -bi n 2 (mI)2-2s* We see that if v 0 0 (mod 1), the series for C(s) converges uniformly for Re s < 1/2, and we may evaluate C'(0) by this method of summation as: =2 Im T 'Tr n1 n 2 co2Tnv (*) + --- 1 e 2TrM 2 imu c =-o 2ninv 2lmImTl 2 emT + n The first series above is standard, and the first term on the right side of (*) is given by TrImT(2v 2 -2v 1 +), 0 < v <. -22- To compute the second series, define e-2n(imlImTlx+kl+imReT(x+k)) F(x) = Z k=-00 F is periodic and Lipschitz; hence its Fourier series converges at each point to the value of F. The Fourier coefficients are given by f1e-2 inxF(x)dx = f 0 e-2ff(lmImTlxl+i(mReT+n)x)dx ImIImT 1 |ImT + n 2 So the second term on the right in (*) becomes 1 M O = -COC 2nimu Im, log o -27(ImIImTlv+kl+imReT(v+k)) k=--W - e2Tri( Ik1T-Ck(u-Tv)) 1 2 The proof is completed by noting that eigenvalues for the Dirac operator on T with underlying spin structure (e the special twisted case. Namely, , 2) are identical to instead of considering a general non-trivial character given by X(mT+n) = e2-i(um+vn) 0 < u,v < 1, take u Remark: 1 e2 1 2 el. It gives the formula of Theorem 1.4. There is no analogue of the second Kronecker limit formula in n-dimension so the extension of the Theorem 1.2 for h2 seems to be difficult (see the footnote on page 39). -23- CHAPTER 2 DETERMINANTS OF LAPLACIAN FOR COMPACT RIEMANN SURFACES We remind the reader that every compact Riemann surface M of genus at least two can be realized as H/F where r is a discrete subgroup of SL2(R) (more precisely in SL2 (R)/+I) defined up to a conjugacy and acting on H without fixed points. There exists a 6g - 6 dimensional family of nonisomorphic Riemann surfaces of genus g. In terms of a factor representation H/T there is a 6g - 6 dimensional family of nonconjugate, discrete subgroups of SL 2 (R), each of them is isomorphic to the fundamental group of the Riemann surface of genus g. For compact Rieman surface H/P each element of r is of the form z + e 2P z , with real fixed points zo,z an element of F primitive if it and p = p is not a power in F. union of disjoint conjugacy classes. PY = > 0. r is the P , if Y,Y' are conjugate and simultaneously primitive or not primitive. We call -24- Selberg introduced the zeta function: -2p Y(6+k) Z (6,x) = H H {y} k=0 ) det (1-X(y) e which is well defined by the above product for Re 6 > 1. It can be extended to an entire function with zeros at the non-positive integers and also at the points 1 + i 2 Laplacian. - -X- 4 for each eigenvalue X of the Z satisfies the following functional equation: 6-1/2 Zp (1-6 ,X) = Zr(6,X) exp(-4n(g-1) f 0 r tanh Let F(T), F(T') be discrete subgroups of SL(2,R) H/F(T) and H/F(T') have the same genus g > 1. Tr dr) such that Consider the standard -1 curvature metric on those Riemann surfaces induced from H. Let X : Fr(T) +* U(l), X' : r(T') nontrivial representations. S) f + U(l) be Following [1], consider t s-1 e tA(T) dt if X is nontrivial; 0 and (s F(S) Remember that Co,0 = C0,1. ts-1 (tr etA(T) -l)dt, if X is trivial. 0 Notice then in both cases 00? (S)Th c ) r rAeIdr = f S 1 (tr e A(T) -tr e tA(v ))dt 0 ts The Selberg trace formula applied to the heat kernel on D 0 (H/]F(tE),X) states -25- tr(e ( t0 ) = (2g-2) f + r2)t dr + r tanh (Trr) e 0 2 (*) t 1 Y sinh(kp Y) tr(Xk I + I {Y} k=1 k 4 t 2 P - where the sum is taken over conjugacy classes of primitive elements of F. Notice that the first term on the right-hand side of the trace formula is the same for H/F(T) and H/F(t') because our Riemann surfaces have the same genus (or equivalently, via the Gauss-Bonnet theorem, they have the same volume). Let's call the second term on the right-hand side of (*) (t). Fx We have IF I < Ce -at-(b/t) with a,b > 0 and Is) f = -AV,(S) CO (S) (t) ts-1(F .C(t) dt 0() - F r~)0 When Re s < 1, we can write CO ts-1t = ___0 (6(6-1))-se-6 (6-l)t(26-1)d6 1 Using this and interchanging integrations CO (S (6)-G (6(6-1))-s(G f1XT 1'(s)'(l-s) ) where G) Gx ( 6 ) = Xj~(~d (26-1) 6 ( 6 l1)t F fe 0~- (t,d = dlo log-rd r MX (6))d6 -26- So we conclude det A(T) det A(T) that is, Zet (,X) - P(-) lX) is independent of the complex structure and det AX(T) = Constant ZF(T)(1,X) The same technique also applies if both representations X, X' are trivial. Then the Selberg zeta function has a simple root at 1 as opposed to the twisted case. detLA(T) = constant * Therefore, d F Z 161 |() - Our constants are universal for all Riemann surfaces of a given genus g. Recently, one of them has been expressed by D'Hoker-Phong in terms of the genus, Euler constant, fr and the derivative of the Riemann zeta function at -1. See [4]. We remark that the above formulas look almost identical for Laplacian acting on 1 and 2-forms. The point is that for any eigenvalue for Laplacian on 1-forms, one can find, by the Hodge theorem, an eigenform of the form df,*df, where f is an eigenfunction corresponding to X. So the spectrum for 1-forms is twice the spectrum on functions, and the spectrum for 2-forms is just identical to the latter. -27- Finally, one should remark that the formulas for the determinant of the Laplacian (acting on functions) can be almost automatically extended from the two-dimensional hyperbolic manifolds to the compact space forms of symmetric spaces of rank one. All the necessary tools, the Selberg trace formula and the Selberg zeta function, have been established for such spaces by Gangolli [5] and really nothing else is needed. Now we will discuss some recent results of D'Hoker-Phong [6] on the determinant of the Dirac operator on Riemann surfaces. Let Tn denote the underlying space of a complex line bundle {f(z)dzn} for a compact Riemann surface M with a fixed hermitian metric of constant curvature -1. If we fix a spin structure among the 2 2g possible ones, we may also consider n = (odd integer)/2, and view T1/2 as the space of spinors and Tn as the spaces of spinor-tensor fields. derivative V sends Tn into Tn e (T1 accordingly as Vz n Tn V Tn 0 + =V n + Vz, n z with 'iT), and it V": TX'+ z (that is V = 3 9 T1 T n- The covariant can be decomposed Tne TTf + on sections of Tn). The spaces of L2-sections of Tare Hilbert spaces, since spinor-tensor fields can be paired at each point of M, and then integrated over M using ds2. With this pairing, (V1)+ = - V VZ is the Dirac operator, and the natural covariant Laplacians on T7" are ZV 'M A+ = -V n n Z - n z n z n -28- In local isothermal coordinates z, we can write ds 2 = 2g - dz dz, n zz Vzf = gZZ n zz z f dzdz g - (g )nf*g. zz zz M <fig> Tn Our M = H/F, F is a discrete subgroup of SL(2,R)/{+1}, all of whose elements are hyperbolic. SL(2,R) Let P be the subgroup of containing -I which projects to P, and let Yi,. .,Y2g fixed set of generators for P. be a A spin structure v on M corresponds to a choice of multipliers v(Y) E {+l} on y E r which is multiplicative and satisfies v(-I) = -1. Such a choice is determined by the values of v on the generators Y and there are 22g of them. D'Hoker-Phong [61 developed trace formulas for such operators which yield Theorem 2.1 For arbitrary half-integer n -t(A Tr (e - n n(n+l)) - nn n(= It) e + In(t) -29- where p2t 2 t In(t) I V(Y) = I Y primitive p=l 2 1 n 4( 4(rt)2 2 s 4t e and In W e (2n-2m-1) e (n-m)(n-m-l)t = -2(2-2g) O<m<n- 1 -t/ -b 2/4t Co 4 - 47(2-2g) 2 (47t3/ f db 0 b e ch (n-[n])b sh b/2 They also introduced the two Selberg zeta functions Zn(s) = H Y prim. (1-v(X) 2n e -(p+s)l) H p=0 for n=0, Let N- denote the number of zero modes of the Laplacian n A±-. n Note that N 1/2 = N- -1/2 corresponds to the number of zero modes of the Dirac operator. For g > 3, it not only depends on the spin structure but also on the conformal class of the metric. D'Hoker-Phong computed determinants of A- in terms of values of n Selberg zeta functions at half integer points except this Al case.*) 2 Notice that by considering avoid those difficulties. A1/2 + a with a large, we can The determinant of such operators can be computed by the same technique as applied in [6] for A+, n > 1/2. n - One uses the identity v f xV- x dx = 2 ( K (2/Vi ), > 0, X > 0 0 *) Peter Sarnak's very recent preprint "Determinants of Laplacians" completely settles this problem. -30- Also Tr(e-t(a = 1/2 + a2) -z Tr e-t(A1/ 1/2 1/2 1/( (t) + 1 1/ e 2 + 1+ (a2 _ ) Finally we state Theorem 2.2 det (A1/2 + a2 = constant - Z1/2(a + 1/2) The proof is similar to that of Theorem 2.1. We will finish this section by making an obvious remark, following from the trace formula, that the length spectrum determines the spectrum of the Dirac operator. converse is true. It is not clear to the author if the -31- CHAPTER 3 SOME ASPECTS OF CONFORMAL GEOMETRY Consider the conformal Laplacian L = d*d + 4n-) acting on functions on a Riemannian manifold Mn with scalar curvature s. is a conformally invariant operator. to construct new conformal invariants. It Parker, Rosenberg [7] used 0 In particular they proved: For odd dimensional manifolds det 0 is a conformal Theorem 3.1 invariant. We offer a different proof of this theorem as Corollary 2. Here det E = (-1) e if ker U = {0} and there are v negative eigenvalues (counted with multiplicity). if ker 0 i {0} 0 Note that det E is nonzero on a conformal class admitting a metric of positive scalar curvature. Namely, where c is a positive constant. in this metric E = A + c, Since A is nonnegative, the operator on the right-hand side does not have a kernel, justifying the above remark. For the sake of completeness, (M ,g) be a C R its we state the Yamabe Problem. Let compact Riemannian manifold of dimension n > 3, scalar curvature. The problem is: does there exist a metric g', -32- conformal to g, constant? such that the scalar curvature R' of the metric is The positive answer has been recently proved by R. Schoen. Therefore, in this metric l = A + constant , where both sides act on functions. Let B be a 0th order differential operator (or multiplication by a matrix function B(x)). Let k(BIA) be the coefficients of the asymptotic expansion for Tr(B e-tA ) when t +0; + so that k(BA)t-k Tr(B e-tA) (*) We have Sk(BIA) f tr(B(x) Tk(xIA))dv(x) = where Tk (xlA)t-k e-tA (x) From Shvarz [8] we have Theorem 3.2 Suppose that for an elliptic, positive operator A depending on the parameter T, there is a positive elliptic operator T and 0th order differential operator B satisfying the relation Tr e 'T=O = t i-Tr(B e-tT) Then d - - log det A(T) IT= 4(BIT) - Tr(B P(T)) where P denotes a projection on the kernel. -33- Proof Using (*) we can represent ?(SJA) for large Re s > 0 in the form a (**) s - = 6 k F'(s) k>O sk0 6 s-k Tr(P(A)) + f ts- 1 Tr(et + f ts- 1 p(t)dt - 6 P(A))dt} where p(t) = Tr e-tA akt-k , a k>0 = (Dk(11A) , (ak = 0 for the finite number of subscripts k defined by the order of A and the dimension of the manifold), and 6 is an arbitrary positive number. The right-hand side of (**) is an analytic function for all Re s > 0 (except poles). Because C(sIA) is defined by analytic continuation from the domain Re s >> 0 then for small Re s , (sjA) is also defined by (**). Then from the definition of the regularized determinant det A = exp C(sIA)Is= S 0 (-L it follows, that ak_ - log det A k>0 6k 6 + f t~ - TrP(A)) + + Y(a k o CO p(t)dt + 0 Here Y = F'(l), the Euler constant. f t- 1 Tr(e-tA -P(A))dt 6 -34- Taking into account that d = dT -k T=0 = -k (k(BIT) and the assumptions of the theorem, we have 0, log det A - = (k(BJT)6-k + 6 d + f (Tr B e - -tT f (Tr(Be -tT))dt - 6 k>O dt k(BIT)t-k )dt - Sdt k>O Finally, taking into account the asymptotic expansion of Tr B e-tT for t + 0 and the relation lim t+O Tr B e-tT = Tr B P(T) we complete the proof of the theorem. The previous theorem has the following application: Theorem 3.3 Let A(T) be the family of nonnegative selfadjoint elliptic operators such that dA(T) = L(T)A(T) + A(T)R(T) , dTc where L(T), R(T) are differential operators. log det A(T) dT Then = (D (L(T) + R(T)IA(T)) 0 - Tr((L(T) + R(T))PA(T)) -35- Proof We have d -tA(T) edT -t Tr = dTr e dT -tA(T) tA(T))e + Tr (R e-tA(T) ( Tr(L e-tA(T) t Applying the previous theorem we finish the proof. = d*d + For the conformal Laplacian LI (n-) we have n-2 n+2 -- 7=;- 09 Here g (x)v(x), function on M. d du uh(x) -e where P(x) , u E [0,1] and h is a C Then _ Pg n+2 h 4 g + 2 . 4 g h V Applying Theorem 3.3 to Og without zero eigenvalues we obtain Corollary 1 6 log det Ig = 0(h ) . Taking into account that for n odd %0(BIA) = 0, we have Corollary 2 For n odd we have 6 log detL = 0 , -36- This result was first discovered by Parker-Rosenberg [7] in a slightly different setting. anomaly effect. coefficient For n even, however, we encounter the conformal For instance, for n=4, computing the Seeley (h0Ih ), we obtain Corollary 3 6 log det W f h(x)(R f472 = - RRR R6 + As) dvol Proof For B of the zero order we obviously have k(BIA) = f tr(B(x) ' k(xlA)) dvol Also for the operator A + k(x), where k(x) is an arbitrary smooth function,we have < x lexp(-t(A+k))Ix> = 1 + t(-s n 1 k) + - (4nt)2 + t2 [- ( - k)2 -R 26s 180sv Here Rp6 R + 180 R v R P6 + 14-k)]+... 6 5 stands for the curvature tensor, RP=%O6 Ricci tensor, and s = Rfrtg c u for the scalar curvature . P o h -37- In order to derive the latter expression in four dimensions one should write g v(x) according to the powers of the curvature tensor and its covariant derivatives expressed in the geodesic coordinates. Finally, take k(x) = =1 s(x). At the end of this chapter we will present some results on the conformal aspects of the square of Dirac operators. The following two facts can be found in Hitchin [9]. Theorem 3.4 For g < 3, the dimension of the space of harmonic spinors is independent of the metric. For g ' 3, this is no longer the case. Theorem 3.5 The dimension of the space of harmonic spinors on a two-dimensional riemannian manifold varies with the choice of a metric. but Theorem 3.6 The dimension of the space of harmonic spinors is a conformal invariant. -38- In order to discover other conformal invariants related to the Dirac operator, let's consider the zeta function of D2 CO L(t) -t e = , i=o where eigenvalues are counted with their multiplicities. The following fact is a consequence of the general theory of elliptic operators and the evaluation of the universal constant in particular cases like a sphere. Theorem 3.7 IU) (4Trt) t+0+ 2. E d.tj j=o a where do = dim S . vol(X) d dim S = - d2 12 S fs(x) dx X Here S stands for the spin bundle. There is also an expression for the coefficient d 2 but it will not be needed for our 2-dimensional case. Note that in such a case the dimension of the positive spinors is one and 1 1(0)f (- 1 s(x))dv 2-(2-2g) 48 g-d e dim ker D 2 _ - 1-dimker D - k 2 2 -39- We will state it as: Theorem 3.8 The value at zero of the zeta function of the Dirac square operator is a conformal invariant. Now consider the metric (1+e)g together with the metric g. = (1+E)D2 (l+E)g and We have D2 g as a consequence d dE v' =0 (0) D 2(ln:)d (l+E)g + V D2 g =n(1+d) 20) 6 0D E=Q + 2 ( g D2 (0) g This number is not equal to zero except in very special cases. We will state it as Theorem 3.9 The determinant of D2 is not a conformal invariant. I.M. Singer suggested one should consider the difference of determinants related to two different spin structures rather than a single determinant. Remember that spin structures can be thought as Z2 characters of the fundamental group, with the so we expect an analogy torsion (for the twisted case) on a Riemann surface. Compare"Theta Functions, Modular Invariance, and Strings", L. Alvarez-Gaum6, G. Moore, C. Vafa, HUTP-86/A017. -40- CHAPTER 4 PARALLEL SPINORS AND SYMPLECTIC STRUCTURES We have already seen in the previous chapter that the dimension of the harmonic spinors space was important when computing the determinants of the Dirac operator. We include below a few results relevant to this issue. If X is a spin manifold with a zero scalar curvature, then every harmonic spinor is parallel. The following two theorems (due to Hitchin [9,10]) show that parallel spinors are not very common. Theorem 4.1. Let X be a compact simply connected spin manifold which admits a parallel spinsor. Then if dim X is even (resp.odd), + X (resp. + X x S1) is a KAhler manifold with a vanishing Ricci tensor. There are no known examples of odd dimensions. Theorem 4.2. On a compact 4-dimensional manifold, the existence of a non-trivial parallel spinor field implies that the manifold is either the flat torus or a K3 surface. -41- Now we are ready to state: Theorem 4.3 A Riemann surface M of genus 0 or genus at least two does not admit a nonzero parallel spinor. Proof ) We obviously have c1 (M) = c1 (K) and c( is a nonzero parallel spinor. c (M) 12 . Suppose s Then s+, s_ are also parallel and at least one of them must be nontrivial. vanishing section of /~Kso = If s+, 1 0, then s+ is a never by the Gauss-Bonnet theorem c1 (/K ) = 0. 0,1 If s_, 1 0 1 0, then s_ is nonvanishing section of A . Chern class of this bundle is + c(M) + 2 K, and the first Therefore, again c1 (M) = 0 so M is a torus. We will close this chapter with some remarks related to zero Ricci and scalar curvatures in 4-dimensions. I thank I.M. Singer for improving my originally weaker version of this theorem. -42- Let M 3 CP . be any nonsingular algebraic surface of degree 4 in Then M 4 is simply connnected, has signature -16 and Euler characteristic 24. Its intersection form has the form 0 E + E + 3 8 8 1 1 0 where E 8 2 1 1 2 1 1 2 1 1 2 1 1 2 1 1 2 1 1 1 2 1 and all other entries are 0. 2 Its second Betti number is 22. As every diagonal entry is even, M is a spin manifold (those two facts are equivalent for simply connected 4 manifolds (Milnor)). A K3 surface is a complex surface with the first Betti and first Chern class zero. Kodaira's theorem states that all K3 surfaces are diffeomorphic to a quartic surface in CP 3 . Moreover, for a K3 surface the admitting of a riemannian metric of zero scalar curvature is equivalent to the admitting of a Ricci-flat Kghler structure (see [10]). Yau's proof of the Calabi's conjecture yields the existence of a Ricci flat metric on the K3 surface (the simple-connectivity of a K3 surface guarantees that it does not admit any flat metric). Therefore M carries a metric with a zero scalar curvature. Consider M = M4 # M4. a nonzero A genus. It is a simply connected spin manifold with -43- Theorem 4.4 M does not admit a metric with zero scalar curvature. Since the scalar curvature is the trace of the Ricci tensor, M does not carry a metric with zero Ricci curvature. Proof If M admits a metric of a nonnegative scalar curvature, then by a theorem of Kazdan and Warner, it admits a metric of scalar curvature identically zero. But then by the Lichnerowicz vanishing theorem there exists a parallel spinor,**) so by Hitchin's theorem +M is a Khler manifold with a vanishing Ricci tensor. But according to the Atiyah-Singer index theorem:*) index a (+M) = 1 1 X(+M) + I(+m= = (2X(M ) 2) + - 4 2T(M4)) =1 l- + 8 / Z Therefore +M is not a complex Kghler manifold. One should also mention, that (2k+l)M = M4#M4#...#M4 2k+1 times has another interesting property. It has nonzero second and fourth *) See Peter Gilkey, "The index theorem and the heat equation," Publish or Perish, Inc., 1974. **) One can also use Theorem 4.2 as was pointed out to the author by I.M. Singer. -44- Betti numbers. Moreover, it is an almost complex manifold which does not carry a complex structure (Van de Ven). A symplectic manifold is a manifold of dimension 2k with a closed k 2-form a such that a is nonsingular. 2k If M is a closed symplectic manifold, then the cohomology class of a is nontrivial, and all its powers up to k are nonzero. M also has an almost complex structure associated with a, up to a homotopy. The existence of such an element of H 2(M 2kR) is a necessary condition for the compact manifold to admit a symplectic structure. Now let us recall Thurston's Conjecture: Every closed 2k manifold which has an almost complex structure T and a real second cohomology class a such that ak 0 has a symplectic structure. See his paper "Some Simple Examples of Symplectic Manifolds," Proc. AMS, Vol. 55, (2), 1976,pp. 467-468. Weinstein's Question: Are there any compact simply connected symplectic manifolds with no underlying Kdhler structure? (Compare his article, "Fat Bundles and Symplectic Manifolds," Adv. Math, 37, 1980.) *) See also Dusa McDuff, "Examples of simply-connected symplectic non-Kghlerian manifolds", J. Diff. Geom. 20, 1984. -45- We will combine those facts into Theorem 4.5: If Thurston's Conjecture is true, then Weinstein's Question has a positive answer. Consider (2k+l)M , k > 0. Proof: It satisfies all the assumptions of Thurston's conjecture which would imply that it is a symplectic manifold. At the same time it is non-KAhler and simply connected. M Question: has an as the nonsingular hypersurface in CP 3 induced symplectic structure from CP . Does the Darboux theorem and convexity considerations (fwl + (1-f)w 2 ) yield the existence of a symplectic structure on (2k+l)M 4 ? Let Q denote the naturally oriented underlying differentiable manifold of the product of a Riemann surface of genus 2 and one of genus 0. Finally, for integers l,m,n > 0, 1+m+n > 0, be a connected sum of 1 copies of +CP 2, let Wlm$n m copies of -CP , and n copies of Q. Also let A = CP2 # 2(SIxS ), B = (S xS2) # 2(S1xS ). Following Van de Ven ("On the Chern numbers of certain complex and almost complex manifolds," PNAS, number of 1,m,n; W1,M,n, Vol. 155, 1966), for some infinite A and B have almost complex structure but no complex structures which implies they have no K~hler structures. -46- Since homology groups of a connected sum of oriented manifolds are easy to express in terms of homology groups of summands; Wi,m,n, A,B have nonzero second and fourth Betti numbers. So we can always find the element a in the second real homology class such that a2 t 0. If Thurston's Conjecture is true, then all those manifolds have a symplectic structure. So we have found new examples of compact symplectic manifolds which admit no Kghler structures. -47- BIBLIOGRAPHY [1] D.B. Ray, I.M. Singer, "Analytic Torsion for Complex Manifolds," Ann. of Math. (2) 98, 1973, 154-177. [2] A. Terras, "Bessel Series Expansions of the Epstein Zeta Function and the Functional Equation," Trans. of the AMS, Vol. 183, September 1973. [3] Th. Friedrich, Zur Abhgngigkeit des Dirac-Operator von der Spin-Struktur, Coll.Math 47, 1984, p. 61. [4] E. D'Hoker, D. Phong, "Multiloop Amplitudes for the Bosonic Polyakov String," preprint. [5] R. Gangolli, "Zeta Function of Selberg's Type for Compact Space J. Math., 21, 1977, Forms of Symmetric Spaces of Rank One," Ill 403-423. [6] E. D'Hoker, D. Phong, "On Determinants of Laplacians on Riemann Surfaces," preprint. [7] T. Parker, preprint. [8] A.S. Shvarz, "Elliptic Operators in Quantum Field Theory," Modern Problems in Mathematics," Vol. 17, (in Russian). [9] N. [10] Hitchin, S. Rosenberg, "Invariants of Conformal Laplacians," "Harmonic Spinors," Adv. in Math, 14, 1974, 1-44. , "Compact Four-Dimensional Einstein Manifolds," J. Diff. Geom., 9, 1974, 435-441.