9512.net
甜梦文库
当前位置:首页 >> >>

M. P. Appell's function and vector bundles of rank 2 on elliptic curve



arXiv:math/9810084v1 [math.AG] 13 Oct 1998

M. P. APPELL’S FUNCTION AND VECTOR BUNDLES OF RANK 2 ON ELLIPTIC CURVES A. Polishchuk1 Department of Mathematics Harvard University, Cambridge, MA 02138 The purpose of this work is to establish a connection between the function κ(y, x) = κ(y, x, τ ) = exp(πiτ n2 + 2πinx) , n∈Z exp(2πinτ ) ? exp(2πiy )

where y, x, τ ∈ C, Im(τ ) > 0, y ∈ Z + Zτ , and vector bundles of rank 2 on elliptic curves. This function was introduced by M. P. Appell in [2] in order to decompose into simple elements the so called elliptic functions of the third kind (which correspond to meromorphic sections of line bundles on elliptic curves).2 As is well-known Riemann’s theta function arises naturally when considering global sections of line bundles on elliptic curves. We claim that the function κ is connected in a similar way with rank-2 bundles. Namely, the di?erence equation for κ allows to interpret it as a global section of a rank-2 bundle of degree 1 on an elliptic curve. Pursuing this analogy we derive some interesting identities satis?ed by κ similar to the addition formulas for θ. These identities (that were known to Appell) appear as A∞ -constraints in the Fukaya category of an elliptic curve (see [7]). Another identity for function κ turned out to be useful in proving some formulas from Ramanujan’s Notebooks (cf. [4]). However, at the present moment there appears to be no general theory similar to that of theta-identities. The following pair of identities is derived easily from those known to Appell: 1 θ(0)κ(τ /2, 1/2) + θ(1/2)κ((τ + 1)/2, 0) = θ(τ /2)3 . 2 θ(τ /2)3 κ(1/2, τ /2) = θ(1/2)3 κ(τ /2, 1/2) + θ(0)3 κ((τ + 1)/2, 0).
E-mail: apolish@math.harvard.edu The same function was introduced by M. Hermite essentially for the same problem, however, he didn’t publish his results until the appearance of the ?rst part of Appell’s paper. In the second part of this paper Hermite’s method is partially reproduced. Appell actually used the function which di?ers from κ by a theta-function. Our notation is closer to that of Halphen’s book [5].
2 1

1

These identities relating three special values κ(y, x) with y, x points of order 2 on elliptic curve C/Z + τ Z deserve to be considered as analogues of the famous Jacobi identity relating the 4-th powers of values of θ at points of order 2. In fact, all the values κ(x, y ), where y and x are points of order 2 on elliptic curve, except these three can be expressed in terms of theta-function. One may ask whether there is an analogue of modular property for the function κ similar to the functional equation for the theta function. Note that the latter equation re?ects the fact that one can construct a line bundle on the universal elliptic curve (over the analytic moduli stack of elliptic curves with a non-trivial point of order 2) such that θ is its section. Similar, we can construct a vector bundle of rank 2 on the universal elliptic curve. However, it seems that there is no simple formula for the corresponding 1-cocycle of the modular group. So we propose to formulate the modular property of the function κ not in the form of the equation but in the form of divisibility property. More precisely, let us denote κ0 (x, τ ) = exp( We prove that for every element γ = 3πiτ )κ((τ + 1)/2, x, τ ). 4

a b of the group Γ1,2 (this is the standard c d subgroup of SL2 (Z) which preserves the point (τ + 1)/2) the di?erence κ0 ( aτ + b 1 x , ) ? ζ · (cτ + d) · exp(πi( ? 1)x)κ0 (x, τ ) cτ + d cτ + d cτ + d

is divisible by θ(x, τ ) in the ring of holomorphic functions on H × C, where ζ is a root of unity of order 4 which we explicitly determine. Although we concentrate on the case of bundles of degree 1, it seems that almost all facts concerning vector bundles of rank 2 on elliptic curves can be worked out explicitly in terms of the function κ (and elliptic functions). It seems that the case of higher rank bundles is described similarly by various functions derived from κ by taking derivatives and di?erence derivatives (κa1 ? κa2 )/(a1 ? a2 ), etc. In section 3 we show that some natural isomorphisms between rank-2 vector bundles can be written explicitly in terms of κ. As a byproduct we show how to ?nd explicitly (in terms of κ) holomorphic functions φ1 and φ2 on C? such that φ1 (z )θ(z, q 2 ) ? φ2 (z )θ(qz, q 2 ) = 1 (one knows apriori that such functions exist since θ(z, q 2 ) and θ(qz, q 2 ) have no common zeroes). We use both additive and multiplicative notations (i.e. sometimes we consider variables in C and sometimes in C? ). The relation between (some) multiplicative and additive variables is the following: z = exp(2πix), q 1/2 = exp(πiτ ), a = exp(2πiy ). By abuse of notation, for any function in multiplicative variables we will denote
2

the corresponding function in additive variables by the same letter (the only section where the additive notation is used is section 4 on modular property of κ). When working with a ?xed elliptic curve Eq = C? /q Z we often omit the variable q in the corresponding elliptic functions and the function κ. Acknowledgment. I’d like to thank B. Berndt and G. Andrews for help with ?nding the references. This work is partially supported by NSF grant. 1. Bundles of rank 2 and degree 1 In this section we always use multiplicative variables. 1.1. Let q ∈ C? be such that |q | < 1, E = Eq = C? /q Z be the corresponding elliptic curve. For every invertible r × r matrix A(z ) of holomorphic functions on C? we de?ne the corresponding holomorphic vector bundle Vr (A) on Eq as follows: Vr (A) = C? × Cr /(z, v ) ? (qz, A(z )v ). Holomorphic sections of Vr (A) correspond to r -tuple of functions v (z ) satisfying the equation v (qz ) = A(z )v (z ). The bundles Vr (A) and Vr (A′ ) are isomorphic if and only if there exists an invertible r × r matrix B (z ) such that A′ (z ) = B (qz )A(z )B (z )?1 . The construction A → V (A) is compatible with tensor products, in particular, if φ is an invertible holomorphic function on C? then Vr (φA) ? V1 (φ) ? Vr (A). 1.2. We denote by L the line bundle V1 (q ?1/2 z ?1 ). The classical theta function θ(z ) = θ(z, q ) =
n∈Z
2 /2

qn

zn

is a section of L (as function in z ). We also denote by Pa the line bundle V1 (a), where a ∈ C? is considered as a constant function on C? . Then we have a canonical isomorphism (1.2.1) Pa ? t? a L ? L.
2 ? t? a L ? ta?1 L ? L .

In particular, this implies that for every a one has (1.2.2)

The classical addition formula for theta-function which we remind below gives a more concrete form of this isomorphism. The basis of global sections of L2 consists of functions ?0 and ?1 de?ned as follows: ?0 (z ) = ?0 (z, q ) = θ(z 2 , q 2 ) =
n∈Z

q n z 2n ,

2

3

?1 (z ) = ?1 (z, q ) = θ

2 1/2 (z 2 , q 2 ) = q (n+1/2) z 2n+1 . 0 n∈Z

Now the addition formula corresponding to (1.2.2) is (1.2.3) 1.3. θ(za)θ(za?1 ) = ?0 (a)?0 (z ) + ?1 (a)?1 (z ). Consider the rank-2 bundle Fa = V2 a 1 ?1/2 ?1 . 0 q z

This bundle is a unique non-trivial extension of L by Pa . Assume that a ∈ q Z . Then H 0 (E, Pa ) = H 1 (E, Pa ) = 0, hence the natural projection H 0 (E, Fa ) → H 0 (E, L) is an isomorphism. This means that there exists a unique global section of Fa projecting to θ(z, q ). This section has form κa (z ) θ (z ) where κa (z ) = κ(a, z ) = κ(a, z, q ) is the unique holomorphic in z function satisfying the equation (1.3.1) κ(a, qz, q ) = aκ(a, z, q ) + θ(z, q ). q n /2 n z κ(a, z, q ) = n n∈Z q ? a converging for all z , since |q | < 1. Using the de?ning equation (1.3.1) one can easily establish the following identities: (1.3.2) (1.3.3) κ(a, z ) = ?a?1 κ(a?1 , qz ?1 ), κ(qa, z ) = q ?1/2 zκ(a, qz ) = q ?1/2 azκ(a, z ) + q ?1/2 zθ(z )
2

This function can be written as the following series

The latter one allows to consider the pair (κ, θ) as a meromorphic section of certain rank-2 bundle on E × E . It is also convenient to consider the function κ ? (a, z ) = θ(?q ?1/2 a)κ(a, z ) which is holomorphic in a and z . Comparing the quasi-periodicity properties of this function in a and z one immediately arrives to the following symmetry identity: (1.3.4) aκ ? (a, z ) = ?q 1/2 z κ ? (?q 1/2 z, ?q ?1/2 a).

Note that this identity appears as the main theorem of [4] where it is applied to derive some of Ramanujan’s formulas. However, it can be found already in Appell’s paper [2] (t.III, p.20) where it is shown to be equivalent to certain identity of Hermite.
4

Finally, the function κ ? (az, z ?1 ) for ?xed a is a section of P?q?1/2 a?1 ? L2 . If we choose a square root a1/2 we can write this as t? L2 . Hence, we can express iq 1/4 a1/2 κ ? (az, z ?1 ) as a linear combination of the corresponding shifts of ?0 and ?1 . Using the vanishing ?0 (iq 1/2 ) = ?1 (i) = 0 we ?nd explicitly the coe?cients and get the following identity: (1.3.5) κ ? (az, z ?1 ) = ?0 (iq 1/4 a1/2 z ) ?1 (iq 1/4 a1/2 z ) κ ? (q ?1/4 a1/2 , q 1/4 a1/2 ) + κ ? (q 1/4 a1/2 , q ?1/4 a1/2 ) 1 / 2 ?0 (i) ?1 (iq )

1.4. The analogue of the property (1.2.1) for rank-2 bundles Fa is the following canonical isomorphism: (1.4.1) Indeed, the left hand side is V while the right hand side is V a b?1 0 b?1 q ?1/2 z ?1 a b?1 1 0 . ?1 = 0 b 0 b?1 q ?1/2 z ?1 a 1 ?1 ?1/2 ?1 0 b q z t? b Fa ? Pb?1 ? Fab .

Now the above isomorphism follows from the equality 1 0 0 b a 1 ?1 ?1/2 ?1 0 b q z

Using (1.2.1) and (1.4.1) we obtain an isomorphism (1.4.2)
? L ? t? b Fa →tb L ? Fab .

Moreover, this isomorphism is compatible with morphisms of both bundles to t? b L ? L. Now the section θ · t? b κa θ · t? bθ of L ? t? b Fa is mapped by (1.4.2) to the section θ · t? b κa ?1 b θ · t? bθ of t? b L ? Fab . On the other hand, the latter bundle has the global section κab · t? bθ . ? θ · tb θ
5

It follows that the di?erence
?1 ? κab · t? b θ ? b θ · tb κa 1/2 is a constant multiple of t? a?1 θ . Considering the value of this di?erence at z = ?q 1/2 and using vanishing θ(?q ) = 0 we get the following identity

(1.4.3)

θ(bz )κ(ab, z ) ? b?1 θ(z )κ(a, bz ) =

θ(?q 1/2 b)θ(a?1 z ) κ(ab, ?q 1/2 ). θ(?q ?1/2 a)

This identity appears in a slightly di?erent form in Halphen’s book [5] (p.481, formula (45) and the next one). Note that κ ? (a, ?q 1/2 ) = aκ ? (a, ?q ?1/2 ) and (1.3.4) gives aκ ? (a, ?q ?1/2 ) = κ ? (1, ?q ?1/2 a). Now using the explicit series for κ(a, z ) we immediately see that θ(?q ?1/2 a) dθ 1 κ ? (1, z ) = lim = q ?1/2 (?q ?1/2 ) = θ(1)θ(?1)θ(q 1/2 ) a→1 1?a da 2 due to Jacobi’s derivative formula. It follows that (1.4.4) κ(a, ?q 1/2 ) = θ(1)θ(?1)θ(q 1/2 ) . 2θ(?q ?1/2 a)

Substituting this into (1.4.3) we get (1.4.5) θ(bz )κ(ab, z ) ? b?1 θ(z )κ(a, bz ) = θ(1)θ(?1)θ(q 1/2 )θ(?q 1/2 b)θ(a?1 z ) . 2θ(?q ?1/2 a)θ(?q ?1/2 ab)

This identity can be considered as a rank-2 analogue of the addition formula (1.2.3). It allows to express κ in terms of κ(q 1/2 , z ) and theta-functions: (1.4.6) κ(a, z ) = q 1/2 a?1 1.5. θ (z ) θ(1)θ(q 1/2 )θ(?a)θ(q ?1/2 z ) 1/2 ?1/2 κ ( q , q az ) + . θ(q ?1/2 az ) 2θ(?q ?1/2 a)θ(q ?1/2 az )

We have a canonical morphism
? 0 0 ? 0 ? ?a,b : H 0 (E, t? b?1 L) ? H (E, tb Fa ) → H (E, tb?1 L ? tb Fa ) ? H (E, L ? Fab ).

Assume that a and ab are not integer powers of q . Then this map is given by ?a,b (t? b?1 θ ? t? t??1 θ · t? b κa b κa ) = b? . ? tb θ btb?1 θ · t? bθ

Using the maps ?a,b one can easily construct a basis in H 0 (E, L ? Fa ). Indeed, assume that a ∈ C? is such that a ∈ q Z and ?a ∈ q Z . Then we claim that H 0 (E, L ? Fa ) is
6

generated by the images of ?a,1 , ??a,?1 and the subspace H 0 (E, L ? Pa ). This follows immediately from the exact sequence (1.5.1) 0 → H 0 (E, L ? Pa ) → H 0 (E, L ? Fa ) → H 0 (E, L2 ) → 0.

2 0 2 and the fact that θ2 and t? ?1 θ generate H (E, L ). It follows that the following 0 vectors form a basis of H (E, L ? Fa ):

v0 (a) =

θ(?z ) · κ(?a, ?z ) θ(a?1 z ) θ(z ) · κ(a, z ) . , v?1 (a) = , v1 (a) = 2 ?θ(?z )2 0 θ (z )

One can write the map ?a,b expicitly in terms of this basis using (1.4.6). Namely, one has b?1 · ?a,b (t? b?1 θ ? where λb = νa,b θ(q 1/2 b?1 )θ(q 1/2 b) , θ(q 1/2 )2 t? b κa ) = λb · v1 (ab) ? λ?b · v?1 (ab) + (νa,b ? νa,?b ) · v0 (ab) t? bθ

θ(1)θ(q 1/2 b)θ(?q 1/2 b)θ(?q 1/2 b?1 )θ(b)θ(ab) . = 2θ(q 1/2 )θ(?q 1/2 a?1 )θ(q ?1/2 ab)θ(?ab2 ) 2. Various formulas

2.1. Let us consider special values of κ ? (a, z ) for a, z ∈ {±1, ±q 1/2 }. It is easy to see that all of them except for κ ? (q 1/2 , ?1), κ ? (?q 1/2 , 1) and κ ? (?1, q 1/2 ) can be expressed in terms of values of the theta function. We have already seen how to compute κ(a, ?q 1/2 ). For example, (2.1.1) 1 κ(?q 1/2 , ?q 1/2 ) = θ(?1)θ(q 1/2 ), 2 1 κ(?1, ?q 1/2 ) = θ(1)θ(?1). 2

(2.1.2)

The latter identity appears as entry 18 in [3], p.152. Also using (1.3.2) and (1.3.4) one can easily deduce that (2.1.3) 1 κ(?1, ±1) = θ(±1), 2 1 κ(±q 1/2 , q 1/2 ) = θ(q 1/2 ), 2
7

(2.1.4)

while κ(q 1/2 , 1) = κ(?q 1/2 , ?1) = 0 as one can see immediately from the formulas κ(q
1/2

q n /2+n , z) = (z ?n ? z n+1 ), n+1/2 1 ? q n≥0 q n /2+n (z ?n + z n+1 ). , z) = n+1/2 1 + q n≥0
2

2

κ(?q

1/2

Note that three exceptional values correspond to pairs (a, z ) which are stable under involution (a, z ) → (?q 1/2 z, ?q ?1/2 a) (modulo q Z ). 2.2. Using the above information about the special values of κ one can derive from (1.4.6) the following identities (2.2.1) 2κ(q 1/2 b?1 , q 1/2 b) = θ(q ?1/2 b) + θ(1)θ(b) θ(?q 1/2 b). θ(?b)

(2.2.2) κ(q
1/2 ?1 2 ?1 ?1 θ(q 1/2 )θ(q ?1/2 b)θ(?q ?1/2 b) (q )2 ∞ (?q )∞ (?b)∞ (?qb )∞ (b)∞ (qb )∞ b , b) = = . 2θ(?b) (q 1/2 b)∞ (q 1/2 b?1 )∞

(2.2.3)

θ(?z )κ(a, z ) + θ(z )κ(?a, ?z ) =

θ(q 1/2 )2 θ(1)θ(?1)θ(?a?1 z ) . 2θ(q 1/2 a?1 )θ(?q 1/2 a?1 )

Here is another identity which follows from (1.4.6): (2.2.4) θ(q 1/2 )2 θ(?b)κ(q 1/2 b?1 , ?b) = θ(q 1/2 b?1 )2 θ(?1)κ(q 1/2 , ?1) + θ(?q 1/2 b?1 )2 θ(1)κ(?q 1/2 , 1) 2.3. The special values κ(q 1/2 , ?1), κ(?q 1/2 , 1) and κ(?1, q 1/2 ), which we were not able to express in terms of theta-functions, satisfy the following two identities: (2.3.1) 1 θ(1)κ(q 1/2 , ?1) + θ(?1)κ(?q 1/2 , 1) = θ(q 1/2 )3 . 2 θ(q 1/2 )3 κ(?1, q 1/2 ) = θ(?1)3 κ(q 1/2 , ?1) + θ(1)3 κ(?q 1/2 , 1).

(2.3.2)

The ?rst formula is obtained by substituting b = z = ?1, a = ?q 1/2 into (1.4.5), while (2.3.2) is the specialization of (2.2.4) for b = ?q 1/2 . We can rewrite (2.3.1) as
8

follows q n +n ? q n /2+n ? 2 k k 2 /2 ? + ( = ( ? 1) q ) ( q k /2 ) ? (?1)n n+1/2 1 ? q n+1/2 n≥0 1 + q n≥0 k ∈Z k ∈Z 2?
? ?
2

?

?

2

?

q

(n2 +n)/2

n≥0

?3 ?

.

The LHS can be further rewritten as (?1)n q k
n≥0,k ∈n+2Z
2 /2+n2 /2+n

(

1 1 + )+ n +1 / 2 1?q 1 + q n+1/2 ( 1 1 ? )= n +1 / 2 1?q 1 + q n+1/2
2 2

(?1)n q k
n≥0,k ∈n+1+2Z

2 /2+n2 /2+n

2

(?1)n
n≥0,l∈Z

n?2l+1) /2+n /2+2n+1/2 q (n+2l) /2+n /2+n nq + 2 . ( ? 1) 1 ? q 2n+1 1 ? q 2n+1 n≥0,l∈Z

2

2

Simplifying we get the identity
? ?

q

(n2 +n)/2

n≥0

?3 ?

=

(?1)
n≥0,l∈Z

nq

n2 +n?2nl+2l2

(1 + q 2l+1 )
2 +n)/2

1 ? q 2n+1 q (n )3 , namely

Note that G. Andrews in [1] has obtained another formula for (
? ?

q
n≥0

(n2 +n)/2

?3 ?

=
n≥0,2n≥j ≥0

q 2n

2 +2n?j (j +1)/2

(1 + q 2n+1 )

1 ? q 2n+1

where all the coe?cients in the RHS are positive (which in particular proves that every number is a sum of three triangular numbers). It would be interesting to see directly why the right hand sides in two formulas are equal. 3. Isomorphisms between bundles In this section we show that some isomorphisms between rank-2 bundles can be written explicitly using the function κ. 3.1. The ?rst isomorphism we are interested in is an isomorphism of Fa (see section 1) and the unique non-trivial extension of det Fa ? Pa ? L by O. The latter extension is realized by the rank-2 bundle
′ Fa = V2

1 1 . ?1/2 0 q az ?1
9

′ To give an isomorphism Fa ? Fa we have to ?nd a 2 × 2 matrix of holomorphic functions B (z ) with the property

(3.1.1)

1 1 a 1 B (z )? 1 . ?1/2 ?1/2 ?1 = B (qz ) 0 q az ?1 0 q z

′ Since B should send global sections of Fa to global sections of Fa we can choose B in the form κ (z ) f (z ) . B= a θ (z ) g (z )

Now it is easy to see that (3.1.1) implies g (qz ) = a?1 g (z ) ? a?1 θ(z ), hence, g (z ) = ?a?1 κa?1 (z ). To ?nd f (z ) we note that the determinant of B should be q -periodic, hence, a constant function. Therefore, we have f (z )θ(z ) = c ? a?1 κa (z )κa?1 (z ) for some constant c. Substituting z = ?q 1/2 we ?nd using (1.4.4) that c = ca = a?1 κa (?q 1/2 )κa?1 (?q 1/2 ) = and the matrix B is equal to κa (z ) θ (z )
ca ?a?1 κa (z )κa?1 (z ) θ (z ) ?a?1 κa?1 (z )

θ(1)2 θ(?1)2 θ(q 1/2 )2 , 4aθ(?q ?1/2 a)θ(?q 1/2 a)

.

′ 3.2. Now we want to compare bundles Fa ? Fa with push-forwards of line bundles ? 2Z of degree 1 on Eq2 = C /q under the natural isogeny π : Eq2 → Eq . It is su?cient ′ ′ ′ to do this for F1 since Fa is obtained from F1 by translation. We claim that there is an isomorphism ′ F1 ? π? L′

where L′ is a line bundle on Eq2 with multiplicator ?q ?1/2 z ?1 , i.e. L′ = V1q (?q ?1/2 z ?1 ), where V1q is the construction of section 1 with q replaced by q 2 . Indeed, the unique section of L′ (which is given by θ(?q ?1/2 z, q 2 )) induces the global non-vanishing section of π? L′ , so π? L′ ?ts into exact sequence 0 → O → π? L′ → L → 0.
10
2 2

On the other hand, it is easy to check that π? L′ is a simple bundle, hence the above ′ sequence doesn’t split and π? L′ is isomorphic to F1 . To write this isomorphism explicitly note that 0 ?q ?1/2 z ?1 . π? L′ = V2 1 0 Thus, we are looking for a matrix C (z ) such that (3.2.1) Let C (z ) = 1 1 0 ?q ?1/2 z ?1 C (z )? 1 . ?1/2 ?1 = C (qz ) 1 0 0 q z c11 (z ) c12 (z ) . Then from (3.2.1) we deduce that the vector c21 (z ) c22 (z ) c22 (z ) ?c21 (z ) is a global section of π? L′ . Hence, we can choose C with c21 (z ) = θ(?q ?1/2 z, q 2 ) and c22 (z ) = ?c21 (qz ). As before (3.2.1) implies that det(C ) is a (non-zero) constant, i.e. (3.2.2) c11 (z )c21 (qz ) + c12 (z )c21 (z ) = c for some constant c. Another consequence of (3.2.1) is that c11 (z ) = c12 (qz ) ? c21 (z ). Substituting this in (3.2.2) we get c12 (qz )c21 (qz ) = ?c12 (z )c21 (z ) + c + c21 (z )c21 (qz ). In other words, if we denote φ(z ) = c12 (z )c21 (z ) ? c/2 then we have φ(qz ) = ?φ(z ) + c21 (z )c21 (qz ). Note that c21 (z )c21 (qz ) = λ · θ(?z, q ) θ(1, q 2 )θ(q, q 2) . θ(q 1/2 , q ) It follows that φ(z ) = λ · κ?1 (?z, q ), i.e. c c12 (z )c21 (z ) = + λ · κ?1 (?z, q ). 2 ?1/2 Substituting z = q (zero of c21 ) we obtain λ= θ(1, q )θ(?1, q ) c = ?λ · κ?1 (?q ?1/2 , q ) = λ · κ(?1, ?q 1/2 , q ) = λ · 2 2 (in the last equality we used (2.1.2)). Similarly, we ?nd c c11 (z )c21 (qz ) = ? λ · κ?1 (?z, q ). 2
11

where

So ?nally the matrix C is equal to λ·
θ (1,q )θ (?1,q )/2?κ?1 (?z,q ) θ (?q 1/2 z,q 2 ) ?1/2 2

λ·

θ(?q

z, q )

θ (1,q )θ (?1,q )/2+κ?1 (?z,q ) θ (?q ?1/2 z,q 2 ) 1/2 2

?θ(?q

z, q )

.

Note that since det C is a non-zero constant as a byproduct we can ?nd explicitly holomorphic functions φ1 , φ2 on C? with the property φ1 (z )θ(z, q 2 ) ? φ2 (z )θ(qz, q 2 ) = 1. 4. Modular property Throughout this section we use additive notation. 4.1. Let us recall the modular interpretation of the functional equation for thetafunction. For this we have to consider the stack A+ 1 of elliptic curves (over C) equipped with a non-trivial point of order 2. It is well-known that A+ 1 is the quotient of the upper-half plane H by the action of the subgroup Γ1,2 ? SL2 (Z) consisting a b with det γ = 1 and ac ≡ bd ≡ 0 mod (2). The universal of matrices γ = c d elliptic curve E → A+ 1 is the quotient of C × H by the natural action of the semi-direct product of Z2 with Γ1,2 , where the action of Z2 is given by (x, τ ) → (x + m + nτ, τ ) while the above matrix in Γ1,2 acts by γ : (x, τ ) → ( x aτ + b , γ (τ ) = ). cτ + d cτ + d

Let π : C × H → E be the natural projection. Below we will deal with holomorphic vector bundles V on E (of ranks 1 and 2) such that π ? V is trivial. Fixing such a trivialization gives a 1-cocycle of G = Z2 ?Γ1,2 with values in the group of holomorphic functions C × H → GLr (C), where the action of G on C × H is as above, r = rk V . Conversely, given such a 1-cocycle c(g )(x, τ ), where g ∈ G we can construct a vector bundle on E as the quotient of C × H × Cr by the action of G which sends ((x, τ ), v ) to (g (x, τ ), c(g )(x, τ )v ). Thus, the set of isomorphism classes of holomorphic bundles on E with trivial pull-back to C × H can be identi?ed with H 1 (G, GLr (O(C × H))). Since G is a semi-direct product of Z2 and Γ1,2 , any 1-cocycle of G is uniquely determined by its restrictions to Z2 and Γ1,2 (these restricted cocycles should be compatible via the action of Γ1,2 on Z2 ). For example, θ(x, τ ) is a global section of the line bundle L on E given by a cocycle whose restriction to Z2 is (m, n) → exp(?πin2 τ ? 2πinx),
12

and the restriction to Γ1,2 is (4.1.1) γ= cx2 a b → ζ (γ ) · (cτ + d)1/2 exp(πi ) c d cτ + d

where ζ (γ ) is the 8-th root of unity appearing in the functional equation for θ (the latter equation expresses the fact that θ is a section of L). Note that the cocycle (4.1.1) factors into a product of two cocycles of Γ1,2 , the ?rst is with values in O? (H) ? O? (C × H): γ → cθ (γ ) = ζ (γ ) · (cτ + d)1/2 , and the second is given by the exponential factor. Actually, cθ is a coboundary since θ(0, τ ) is an invertible function on H and cθ (γ ) = θ(0, γ (τ )) . θ(0, τ )

We can’t factor cθ further due to indeterminacy in a choice of a square root of (cτ + d). 2 However, c2 θ does factor as the product of a character γ → ζ (γ ) of Γ1,2 with values in roots of unity of order 4, and a cocycle γ → (cτ + d) which corresponds to the line bundle ω ?1 on A+ 1 (where ω is the restriction of the relative canonical bundle ωE /A+ to the zero section). In particular, since c2 θ is a coboundary we obtain that 1 + ω is isomorphic to the line bundle on A1 associated with the character ζ (γ )2 . The explicit formula for ζ (γ ) (see e.g. [6]) implies that ζ (γ )2 =
? ?(?1)(d?1)/2 , ?exp(? πic ),
2

d odd, c odd

4.2. Now we are going to consider a rank-2 bundle on E which is an extension of L by certain line bundle whose restriction to every elliptic curve is of 2-torsion. More presicely, let L be a restriction of L to a particular elliptic curve E . Then we have L ? O(η ) where η ∈ E is the point of order 2 corresponding to (τ + 1)/2 (this ?1 isomorphism is induced by the theta-function). Now let Pη = O(η ? e) ? ωE be the 2-torsion line bundle corresponding to η , trivialized at zero e ∈ E . Then we have
?1 2 ? H 1 (L?1 ? Pη ) ? H 1 (O(?e) ? ωE ) ? H 0 (ω E ).

It follows that there is a canonical non-splitting extension
2 0 → Pη ? ω E → V → L → 0.

In other words, there should exist a universal extension on E 0 → Pη ? ω 2 → V → L → 0
13

where Pη is de?ned in the same way as Pη with η being the divisor x = (τ + 1)/2 mod (Z + Zτ ), e being the zero section, and ωE replaced by ω . The pull-back of Pη to C × H is trivial and the corresponding 1-cocycle of G is given by (m, n) → exp(πin(τ + 1)), 1 a b → exp(πi( ? 1)x). c d cτ + d Below we construct a bundle V on E ?tting into above exact sequence together with a trivialization of π ? V such that the pair (κ, θ) de?nes a global section of V . Note that one has a similar rank-2 bundle on the stack of elliptic curves (without a choice of a point of order 2), which is an extension of O(e) by ω . However, we will stick to the case of A+ 1 in order to stress the analogy with the usual functional equation for θ (which takes place on this stack). 4.3. Recall that for an elliptic curve E = E τ = C/(Z + Zτ ) we have de?ned in section 1 the rank-2 bundle on E τ V τ = Fexp(πi(τ +1)) = V2 exp(πi(τ + 1)) 1 0 exp(?πiτ ? 2πix)

τ which is an extension of L = Lτ by Pη = Pη . Now for an element

γ=

a b ∈ Γ1,2 c d

we have the corresponding isomorphism fγ : E τ → E γ (τ ) : x → x cτ + d

induced by the action of Γ1,2 on C × H. The 1-cocycle of G corresponding to L (resp. ? γ (τ ) ? γ (τ ) τ Pη ) evaluated at γ induces an isomorphism fγ L ? Lτ (resp. fγ Pη ? Pη ). Using these isomorphisms we consider the pull-back by fγ of the extension
γ (τ ) 0 → Pη → V γ (τ ) → Lγ (τ ) → 0 τ as an extension of Lτ by Pη . The class of the latter extension di?ers from the class of the extension given by V τ by a non-zero constant. Hence, there exists a unique isomorphism ? γ (τ ) V τ →fγ V

which induces the identity map on Lτ and a a non-zero constant multiple of the τ identity on Pη .
14

Equivalently, one can say that there exists a unique 1-cocycle of G with values in GL2 (O(C × H)) such that its restriction to Z2 is (4.3.1) (m, n) → exp(πin(τ + 1)) 1 2 0 exp(?πin τ ? 2πinx)

and its restriction to Γ1,2 has form (4.3.2) γ= kγ (τ ) · exp(πi( cτ1 ? 1)x) φγ (x, τ ) a b +d → cx2 c d ) 0 cθ (γ ) · exp(πi cτ +d

for some holomorphic function φγ (x, τ ) and some invertible holomorphic function kγ (τ ). Moreover, it follows from de?nition of κ that κ((τ + 1)/2, x, τ ) θ(x, τ ) is a section of the bundle V de?ned by the above cocycle. This allows us to ?nd functions kγ and φγ explicitly. Namely, for any γ ∈ Γ1,2 we should have (4.3.3) κ((γ (τ ) + 1)/2, 1 x , γ (τ )) = kγ (τ ) · exp(πi( ? 1)x) · κ((τ + 1)/2, x, τ ) + φγ (x, τ )θ(x, τ ). cτ + d cτ + d

To ?nd kγ let us substitute x = (τ + 1)/2 in this equation. Note that the equation (1.3.1) together with the vanishing θ((τ + 1)/2, τ ) = 0 implies that κ((τ + 1)/2, x + m + nτ, τ ) = exp(πin(τ + 1))κ((τ + 1)/2, x, τ ). Using this together with the formulas 1 = a ? cγ (τ ), cτ + d τ = ?b + dγ (τ ) cτ + d we get from (4.3.3) d?c?1 (γ (τ ) + 1))κ((γ (τ ) + 1)/2, (γ (τ ) + 1)/2, γ (τ )) = 2 πi kγ (τ ) · exp( (a ? b ? 1 + (d ? c)γ (τ ) ? τ )) · κ((τ + 1)/2, (τ + 1)/2, τ ), 2 exp(πi i.e. kγ = exp( πi κ((γ (τ ) + 1)/2, (γ (τ ) + 1)/2, γ (τ )) (d ? a + b ? c ? γ (τ ) + τ )) · . 2 κ((τ + 1)/2, (τ + 1)/2, τ )
15

Now using the formula (2.1.1) for κ((τ +1)/2, (τ +1)/2, τ ) and the functional equation for the theta-function we obtain πi 3πi kγ = exp( (d ? a + b ? c) + (τ ? γ (τ ))) · ζ (γ )2 · χ(γ ) · (cτ + d) 2 4 where χ is the character of Γ1,2 with values in 4th roots of unity de?ned as follows: χ(γ ) =
? ?(?1)a/2 exp( πi (ab + cd)),
4 ?(?1)c/2 exp( πi (ab 4

+ cd)),

a even, c even

It is easy to see that (?1)(d?a+b?c)/2 = χ(γ )2 ζ (γ )4, so we can write ?nally 3πi (τ ? γ (τ ))) · ζ (γ )?2 · χ?1 (γ ) · (cτ + d) 4 Now the equation (4.3.3) gives the explicit formula for φγ . Summarizing we can say that the existence of the rank-2 bundle V on the universal elliptic curve E is equivalent to the following divisibility property of κ. Let us denote 3πiτ )κ((τ + 1)/2, x, τ ). κ0 (x, τ ) = exp( 4 Then for every γ ∈ Γ1,2 the di?erence (4.3.4) kγ = exp( x 1 , γ (τ )) ? ζ (γ )?2 · χ?1 (γ ) · (cτ + d) · exp(πi( ? 1)x) · κ0 (x, τ ) cτ + d cτ + d is divisible by θ(x, τ ). κ0 ( References
1. G. E. Andrews, The ?fth and seventh order mock theta functions. Trans. AMS 293 (1986), 113–134. 2. M. P. Appell, Sur le fonctions doublement periodique de troisieme espece, Annales scienti?ques ? de l’Ecole Normale Sup? erieure, 3e s? erie, t.I, p.135, t.II, p.9, t.III, p.9 (1884–1886). 3. B. C. Berndt, Ramanujan’s Notebooks, Part IV. Springer-Verlag, 1994. 4. R. Evans, Generalized Lambert series, in Analytic number theory, Proceedings of a conference in honor of Heini Halberstam, ed. B. Berndt et al., vol.1, 357–370. Birkhauser, 1996. 5. G.-H. Halphen, Trait? e des fonctions elliptiques, I. Paris, 1886. 6. D. Mumford, Tata lectures on theta I, Birkh¨ auser, 1983. 7. A. Polishchuk, Massey and Fukaya products on elliptic curves, preprint math.AG/9803017.

16



更多相关文章:
更多相关标签:

All rights reserved Powered by 甜梦文库 9512.net

copyright ©right 2010-2021。
甜梦文库内容来自网络,如有侵犯请联系客服。zhit325@126.com|网站地图