j. differentialgeometry 82(2009)35-100 TAU-FUNCTIONS ON SPACES OF ABELIAN DIFFERENTIALS AND HIGHER GENUS GENERALIZATIONS OF RAY-SINGER FORMULA Aleksey Kokotov & Dmitry Korotkin Abstract Letw beanAbeliandifferentialonacompactRiemannsurface 2 ofgenusg 1. Then w definesa flatmetricwith conicalsingu- ≥ | | larities and trivial holonomy on the Riemann surface. We obtain anexplicitholomorphicfactorizationformulafortheζ-regularized 2 determinant of the Laplacian in the metric w , generalizing the | | classical Ray-Singer result in g =1. Contents 1. Introduction 36 2. Variational formulas on spaces of Abelian differentials over Riemann surfaces 40 2.1. Coordinates on the spaces of Abelian differentials 40 2.2. Basic holomorphic objects on Riemann surfaces 42 2.3. Variational formulas 48 2.4. Relation to Teichmu¨ller deformation 65 3. Bergman tau-function 67 3.1. Global definition of the Bergman tau-function 69 3.2. Explicit formula for the Bergman tau-function 71 4. Determinants of Laplacians in the metrics w 2 77 | | 4.1. Laplacians on polyhedral surfaces. Basic facts 77 4.2. Determinant of Laplacian 83 4.3. Variation of the resolvent kernel 84 4.4. Variation of the determinant of the Laplacian 87 4.5. Explicit formulas for det∆w2 95 | | References 97 The work of AK was supported by NSERC. The work of DK was partially sup- ported by Concordia Research Chair grant, NSERC, NATEQ and Humboldt foun- dation. We both thank Max-Planck-Institut fu¨r Mathematik at Bonn where the main part ofthiswork was completed forsupport,hospitality andexcellent working conditions. Received 09/13/2005. 35 36 A.KOKOTOV & D.KOROTKIN 1. Introduction The goal of this paper is to give a natural generalization of the Ray- Singer formula for analytic torsion of flat elliptic curves [33] to the case of higher genus. LetAandB betwocomplexnumberssuchthat (B/A) > 0. Taking the quotient of the complex plane C by the latticeℑgenerated by A and B, we obtain an elliptic curve (a Riemann surface of genus one) . Moreover, the holomorphic one-differential dz on C gives rise to aLn Abelian differential w on , so we get a pair (Riemann surface of genus L one, Abelian differential on this surface) and the numbers A,B provide the naturallocal coordinates on the space of such pairs. Inwhat follows we refer to the numbers A,B as moduli. The modulus square w 2 of the Abelian differential w generates a | | smooth flat metric on . Define the determinant of the Laplacian ∆w2 | | L corresponding to this metric via the standard ζ-function regularization: (1.1) det∆w2 = exp ζ (0) , | | {− ∆′ |w|2 } whereζ (s)istheoperatorzeta-function. Nowaslightreformulation ∆|w|2 of the Ray-Singer theorem [33] claims that there holds the equality: det∆w2 (1.2) | | = C η(B/A)4, (B/A)Area( , w 2) | | ℑ L | | where Area( , w 2)= (AB¯), C is a moduli-independentconstant (ac- L | | ℑ tually, C =4) and η is the Dedekind eta-function πiσ η(σ) = exp 1 exp(2πinσ) . 12 − (cid:18) (cid:19)n N Y∈ (cid:0) (cid:1) (Strictly speaking, in [33] det∆ is computed for the Laplacian acting in a line bundle with nontrivial unitary automorphy factors; nevertheless the formula (1.2) is also typically attributed to Ray and Singer. On the otherhand,thisformulais analmostimmediateconsequenceofthefirst Kronecker limit formula, see [24] for detailed discussion.) The main result of this paper is a generalization of the formula (1.2) to the case of Riemann surfaces of genus g > 1. To explain our strategy we first reformulate the Ray-Singer Theorem. For any compact Riemann surface we introduce the prime form L E(P,Q) and the canonical meromorphic bidifferential (1.3) w(P,Q) = d d logE(P,Q) P Q (see [9] or Sect.2.3 below). The bidifferential w(P,Q) has the following local behavior as P Q: → (1.4) 1 1 w(P,Q) = + S (x(P))+o(1) dx(P)dx(Q), (x(P) x(Q))2 6 B (cid:18) − (cid:19) TAU-FUNCTIONS AND HIGHER GENUS RAY-SINGER FORMULA 37 wherex(P)isalocalparameter. ThetermS (x(P))isaprojectivecon- B nection which is called the Bergman projective connection. Let w be an Abelian differential on and, as before, let x(P) be some local parame- L P teron . DenotebyS (x(P))theSchwarzianderivative w, x(P) . w L ThenthedifferenceoftwoprojectiveconnectionsSB SwnisRa(meromoor- − phic) quadratic differential on [37]. Therefore, the ratio (S S )/w B w L − is a (meromorphic) one-differential. In the elliptic case, i.e. when the Riemann surface and the Abelian differential w are obtained from the L lattice mA+nB , this one-differential is holomorphic and admits the { } following explicit expression in the local parameter z (see [8]): S S dlogη(σ) 1 B w (1.5) − = 24πi dz, w − dσ A2 where σ = B/A. Let a,b be a canonical basis of cycles on the elliptic curve , such { } L that the numbers A and B are the corresponding a- and b-periods of the Abelian differential w. Defining (1.6) τ(A,B) := η2(B/A), weseefrom(1.5)thatthefunctionτ issubjecttothesystemofequations ∂logτ 1 S S ∂logτ 1 S S B w B w (1.7) = − , = − . ∂A 12πi w ∂B −12πi w Ib Ia Now the Ray-Singer formula implies that the real-valued expression det∆w2 | | (1.8) Q(A,B) = (B/A)Area( , w 2) ℑ L | | satisfies the same system: ∂logQ 1 S S ∂logQ 1 S S B w B w (1.9) = − , = − . ∂A 12πi w ∂B −12πi w Ib Ia Clearly, if τ(A,B) and Q(A,B) are (respectively) a holomorphic and a real-valued solutions of system (1.7), then Q(A,B) = C τ(A,B)2 with | | some constant factor C. Thus, the Ray-Singer result can be reformu- lated as follows: Theorem 1. 1) The system (1.7) is compatible and has a holo- morphic solution τ. This solution can be found explicitly and is given by (1.6). 2) The variational formulas (1.9) for the determinant of the Lapla- cian ∆w2 hold. | | 3) The expression (1.8) can be represented as the modulus square of a holomorphic function of moduli A,B; this function coincides with the function τ up to a moduli-independent factor. 38 A.KOKOTOV & D.KOROTKIN In what follows we call the function τ (a holomorphic solution to system (1.7)) the Bergman tau-function, due to its close link with the Bergman projective connection. Generalizing thestatement 1 of Theorem 1to higher genus, we define and explicitly compute the Bergman tau-function on different strata of the spaces of Abelian differentials over Riemann surfaces i.e. the g H spaces of pairs ( ,w), where is a compact Riemann surface of genus L L g 1 and w is a holomorphic Abelian differential (i.e. a holomorphic ≥ 1-form) on . In global terms, the “tau-function” is not a function, but L a section of a line bundle over the covering of a stratum of . g H An analog of the Bergman tau-function on spaces of holomorphic differentials was previously defined on Hurwitz spaces (see [14, 15]), i.e. on the spaces of pairs ( ,f), where f is a meromorphic function on L acompactRiemannsurface withfixedmultiplicities ofpolesandzeros L of the differential df. In this case it coincides with the isomonodromic Jimbo-Miwa tau-function for a class of Riemann-Hilbert problems [20, 6], this explains why we use the term “tau-function” also in the context of spaces . g H Generalizing statement 2 of Theorem 1, we introduce the Laplacian ∆w2 corresponding to the flat singular metric w 2. The Laplacian | | | | is acting in the trivial line bundle over . Among other flat metrics L with conical singularities metrics of this form are distinguished by the property that they have trivial holonomy along any closed loop on the Riemann surface. Since Abelian differentials on Riemann surfaces of genus g > 1 do have zeros, the metric w 2 has conical singularities and the Laplacian | | is not essentially self-adjoint. Thus, one has to choose a proper self- adjoint extension: here we deal with the Friedrichs extension. It turns outthatitisstillpossibletodefinethedeterminantofthisLaplacianvia the regularization (1.1). We derive formulas for variations of det∆w2 | | withrespecttonaturalcoordinates onthespaceofAbeliandifferentials. These formulas are direct analogs of system (1.9). Generalizing statement 3 of Theorem 1, we get an explicit formula for the determinant of the Laplacian ∆w2: | | (1.10) det∆w2 = CArea( , w 2) det B τ 2 , | | L | | { ℑ }| | where B is the matrix of b-periods of a Riemann surface of genus g, and the Bergman tau-function τ is expressed through theta-functions and primeforms. Thisformulacanbeconsideredasanaturalgeneralization of the Ray-Singer formula to the higher genus case. Remark 1. The determinants of Laplacians in flat conical metrics firstappearedinworksofstringtheorists(see,e.g., [12]). Anattemptto computesuchdeterminants wasmadein[34]. Theideawastomakeuse ofPolyakov’s formula[31]fortheratioofdeterminantsoftheLaplacians TAU-FUNCTIONS AND HIGHER GENUS RAY-SINGER FORMULA 39 corresponding to two smooth conformally equivalent metrics. If one of the metrics in Polyakov’s formula has conical singularity, this formula does not make sense, so one has to choose some kind of regularization of the arising divergent integral. This leads to an alternative defini- tion of the determinant of Laplacian in conical metrics: one may simply take some smooth metric as a reference one and define the determinant of Laplacian in a conical metric through properly regularized Polyakov formula for the pair (the conical metric, the reference metric). Such a way was chosen in [34] (see also [5]) for metrics given by the modulus square of an Abelian differential (which is exactly our case) and metrics given by the modulus square of a meromorphic 1-differential (in this case Laplacians have continuous spectrum and the spectral theory def- inition of their determinants, if possible, must use methods other than the Ray-Singer regularization). In [34] the smooth reference metric is chosen to be the Arakelov metric. Since the determinant of Laplacian in Arakelov metric is known (it was found in [7] and [2], see also [9]), such an approach leads to a heuristic formula for det∆ in a flat conical metric. This result heavily depends on the choice of the regularization procedure. The naive choice of the regularization leads to dependence of det∆ in the conical metric on the smooth reference metric which is obviously unsatisfactory. More sophisticated (and used in [34] and [5]) procedure of regularization eliminates the dependence on the refer- ence metric but provides an expression which behaves as a tensor with respect to local coordinates at the zeros of the differential w and, there- fore, also can not be considered as completely satisfactory. In any case it is unclear whether this heuristic formula for det∆ for conical metrics has something to do with the determinant of Laplacian defined via the spectrum of the operator ∆ in conical metrics. The paper is organized as follows. In Section 2 we derive variational formulas of Rauch type on the spaces of Abelian differentials for ba- sic holomorphic differentials, matrix of b-periods, prime form and other relevant objects. In Section 3 we introduce and compute the Bergman tau-functiononthespaceofAbeliandifferentialsoverRiemannsurfaces. In Section 4 we give a survey of the spectral theory of the Laplacian on surfaces with flat conical metrics (polyhedral surfaces) and derive variational formulas for the determinants of Laplacians in such metrics. Thecomparisonof variational formulas forthetau-functions withvaria- tional formulas for the determinant of Laplacian, together with explicit computation of the tau-functions, leads to the explicit formulas for the determinants. We use our explicit formulas to derive the formulas of Polyakov type, which show how the determinant of Laplacian depends on the choice of a conformal conical metric with trivial holonomy on a fixed Riemann surface. 40 A.KOKOTOV & D.KOROTKIN Acknowledgments. We are grateful to R.Wentworth, S.Zelditch, P. Zograf and, especially, A.Zorich for important discussions. We thank anonymous referee for numerous useful comments and suggestions. 2. Variational formulas on spaces of Abelian differentials over Riemann surfaces 2.1. Coordinates on the spaces of Abelian differentials.The space of holomorphic Abelian differentials over Riemann surfaces g H of genus g is the moduli space of pairs ( ,w), where is a compact L L Riemann surface of genus g > 1, and w is a holomorphic 1-differential on . This space is stratified according to the multiplicities of zeros of L w. The corresponding strata may have several connected components. The classification of these connected components is given in [18]. In particular, the stratum of the space having the highest dimension g H (on this stratum all the zeros of w are simple) is connected. Denote by (k ,...,k ) the stratum of , consisting of differen- g 1 M g H H tials w which have M zeros on of multiplicities (k ,...,k ). Denote 1 M L the zeros of w by P ,...,P ; then the divisor of differential w is given 1 M by (w) = M k P . Let us choose a canonical basis (a ,b ) in the m=1 m m α α homology group H ( ,Z). Cutting the Riemann surface along these 1 P L L cycles we get the fundamental polygon (the fundamental polygon is L not simply-connected unless all basic cycles pass through one point). Inside of we choose M 1 paths l wbhich connect the zero P with m 1 L − other zeros P of w, m = 2,...,M. The set of paths a ,b ,l gives m α α m a basis inbthe relative homology group H ( ;(w),Z). Then the local 1 L coordinates on (k ,...,k ) can be chosen as follows [19]: g 1 M H (2.1) A := w , B := w , z := w , α α m Iaα Ibα Zlm α = 1,...,g; m = 2,...,M. The area of the surface in the metric L w 2 can be expressed in terms of these coordinates as follows: | | g Vol( ) = A B¯ . α α L −ℑ α=1 X Ifallzerosofwaresimple,wehaveM = 2g 2; therefore,thedimension − of the highest stratum (1,...,1) equals 4g 3. g H − P The Abelian integral z(P) = w provides a local coordinate in a P1 neighborhood of any point P except the zeros P ,...,P . In a 1 M ∈RL neighborhood of P the local coordinate can be chosen to be (z(P) m − z )1/(km+1). Thelatterlocalcoordinateisoftencalledthe distinguished m local parameter. TAU-FUNCTIONS AND HIGHER GENUS RAY-SINGER FORMULA 41 The following construction helps to visualize these coordinates in the case of the highest stratum H (1,...,1). g A 1 B 1 A2 B2 z 3 z 4 A 3 z 2 B 3 0 Figure 1. Representation of a generic point of the stra- tum (1,1,1,1) bygluingthreetorialongcutsconnect- 3 H ing zeros of w. Consider g parallelograms Π ,...,Π in the complex plane with co- 1 g ordinate z having the sides (A ,B ), ..., (A ,B ). Provide these par- 1 1 g g allelograms with a system of cuts [0,z ], [z ,z ], ..., [z ,z ] 2 3 4 2g 3 2g 2 − − (each cut should be repeated on two different parallelograms). Identi- fying the opposite sides of the parallelograms and glueing the obtained g tori along the cuts we get a compact Riemann surface of genus g. L (See figure 1 for the case g = 3). Moreover, the differential dz on the complex plane gives rise to a holomorphic differential w on which has L 2g 2 zeros at the ends of the cuts. Thus, we get a point ( ,w) from − L (1,...,1). Itcan beshownthat any generic pointof (1,...,1) can g g H H beobtained viathisconstruction; moresophisticated glueingisrequired to represent points of other strata, or non generic points of the stratum (1,...,1). g H 42 A.KOKOTOV & D.KOROTKIN The assertion about genericity follows from the theorem of Masur and Veech ([21], [38], see also [19]) stating the ergodicity of the natu- ral SL(2,R)-action on connected components of strata of the space of (normalized) Abelian differentials. Namely, denote by (1,...,1) the Hg′ set of pairs ( ,w) from (1,...,1) such that w 2 = 1. Let a pair g L H | | ( ,w) from (1,...,1) be obtained via the aboLve construction. Then L Hg′ R under the action of A SL(2,R) it goes to the pair ( ,w ) which 1 1 ∈ L is obtained by gluing the parallelograms A(Π ),...,A(Π ) along the 1 g cuts [0,Az ], ..., [Az ,Az ], where the group SL(2,R) acts on 2 2g 3 2g 2 − − z-plane as follows a b A := : z (a z+b z)+i(c z +d z). c d 7→ ℜ ℑ ℜ ℑ (cid:18) (cid:19) Thus, the set of pairs ( ,w) from (1,...,1) which can be glued from L Hg′ tori is invariant with respect to ergodic SL(2,R)-action. Moreover, by varying of local coordinates in a small open neighbourhood of a given pair ( ,w) which is glued from tori, we get a small domain of positive L measure containing pairs ( ,w) which can be glued from tori. Acting on this smalldomain by theLSL(2,R) group,we get asetof fullmeasure in the stratum (1,...,1). Hg′ To shorten the notations it is convenient to consider the coordinates A , B , z altogether. Namely, in the sequel we shall denote them α α m { } by ζ , k = 1,...,2g+M 1, where k − (2.2) ζ := A , ζ := B , α = 1,...,g , ζ := z , α α g+α α 2g+m m+1 where m = 1,...,M 1. − Let us also introduce corresponding cycles s , k = 1,...,2g+M 1, k − as follows: (2.3) s = b , s = a , α= 1,...,g ; α α g+α α − the cycle s , m = 1,...,M 1 is defined to be the small circle with 2g+m − positive orientation around the point P . m+1 Now we are going to prove variational formulas (analogs of classi- cal Rauch’s formulas), which describe dependence of basic holomorphic objects on Riemann surfaces (the normalized holomorphic differentials, the matrix of b-periods, the canonical meromorphic bidifferential, the Bergman projective connection, the prime form, etc.) on coordinates (2.1) on the spaces (k ,...,k ). We start from description of the g 1 M H objects we shall need in the sequel. 2.2. Basic holomorphic objects on Riemann surfaces. Denoteby v (P)thebasisofholomorphic1-formson normalizedby v = δ . α L aα β αβ P For a basepoint P we define the Abel map (P) = v from the 0 Aα P0H α Riemann surface to its Jacobian. L R The matrix of b-periods of the surface is given by B := v . L αβ bα β H TAU-FUNCTIONS AND HIGHER GENUS RAY-SINGER FORMULA 43 Recall also the definition and properties of the prime form E, canon- ical meromorphic bidifferential w and Bergman projective connection S . B The prime form E(P,Q) (see [8, 9]) is an antisymmetric 1/2-diffe- − rential with respect to both P and Q. Let Θ[ ](z) be the genus g theta- ∗ function corresponding to the matrix of b-periods B with some odd half-integer characteristic [ ]. Introduce the holomorphic differential g ∗ q(P) = Θ[ ] (0)v (P). All zeros of this differential are double α=1 ∗zα α and one can define the prime form on by P L Θ[ ]( (P) (Q)) (2.4) E(P,Q) = ∗ A −A ; q(P) q(Q) this expression is independentof tphe choipce of the oddcharacteristic [ ]. ∗ The prime form has the following properties (see [9], p.4): Under tracing of Q along the cycle a the prime form remains α • invariant; under the tracing along b it gains the factor α Q (2.5) exp( πiB 2πi v ) . αα α − − ZP On the diagonal Q P the prime form has first order zero (and • → no other zeros or poles) with the following asymptotics: (2.6) E(x(P),x(Q)) dx(P) dx(Q) = (x(Q) x(P)) − 1 1 S (x(P)p)(x(Q) px(P))2 +O(x(Q) x(P))3 , B × − 12 − − (cid:18) (cid:19) where the subleading term S is called the Bergman projective B connection and x(P) is an arbitrary local parameter. We recall that an arbitrary projective connection S transforms under a change of the local coordinate y x as follows: → 2 dx (2.7) S(y) = S(x) + x,y dy { } (cid:18) (cid:19) where 2 x 3 x ′′′ ′′ x,y = { } x − 2 x ′ (cid:18) ′(cid:19) is the Schwarzian derivative. It is easy to verify that the term S in B (2.6) indeed transforms as (2.7) under a change of the local coordinate. Difference of two projective connections is a quadratic differential on . L Thecanonical meromorphicbidifferential w(P,Q) is definedby (1.3): w(P,Q) = ∂ ∂ logE(P,Q). P Q It is symmetric: w(P,Q) = w(Q,P) and has all vanishing a-periods with respect to both P and Q; the only singularity of w(P,Q) is the second order pole on the diagonal P = Q with biresidue 1. The sub- leading term in expansion of w(P,Q) around diagonal is equal to S /6 B 44 A.KOKOTOV & D.KOROTKIN (1.4). Theb-periodsof w(P,Q) with respectto any of its arguments are given by the basic holomorphic differentials: w(P, ) = 2πiv (P). bα · α The prime form can be expressed as follows in terms of the bidiffer- H ential w(P,Q) ([9], p.3): (2.8) E2(P,Q)dx(P)dy(Q) = Q0 Q lim (x(P ) x(P))(y(Q) y(Q ))exp w( , ) , 0 0 P0→P,Q0→Q − − (cid:18)−ZP0 ZP · · (cid:19) where x and y are any local parameters near P and Q , respectively. 0 0 Remark 2. Let us comment on the formula (1.3) for w(P,Q). Since E(P,Q) is a 1/2 differential with respect to P and Q, this formula − should be understood as w(P,Q) = ∂ ∂ logE(P,Q) dx(P) dy(Q) , P Q { } where x and y are arbitrary local parampeters. Dpue to the presence of the operator ∂ ∂ , this expression is independent of the choice of these P Q local parameters; therefore it can be written in a shorter form (1.3), see [8, 27]. In the same way we shall understand the formula for the normalized (all a-periods vanish) differential of the third kind with poles at points P and Q and residues 1 and 1, respectively (see [27], vol.2, p.212), − which is extensively used below: E(R,P) (2.9) W (R) = ∂ log . P,Q R E(R,Q) This expression should be rigorously understood as E(R,P) dx(P) (2.10) W (R) = ∂ log , P,Q R E(R,Q) dy(Q) p where x and y are arbitrary local coordinatpes; independence of (2.10) of the choice of these local coordinates justifies writing it in the short form (2.9). Denote by S (x(P)) the projective connection given by the Schwarz- w P ian derivative w, x(P) , where x is a local parameter on . L The next obnject we shalloneed is the vector of Riemann constants: R 1 1 g x (2.11) KP = + B v v , α 2 2 αα− β α β=1,β=αIaβ (cid:18) ZP (cid:19) X6 wheretheinterior integralistakenalongapathwhichdoesnotintersect ∂ . L b
Description: