Download Modular Forms, Lecture Notes - Mathematics and more Study notes Number Theory in PDF only on Docsity!
Math 259: Introduction to Analytic Number Theory Some more about modular forms
The coefficients of the q-expansions of modular forms and functions
Concerning “Theorem 5 (Hecke)” of [Serre 1973, VII (p.94)]: We use Poincar´e series and estimates on Kloosterman sums to show that the exponent k in an = O(nk) can be improved to k − 14 + o(1). This method, while neither as deep nor as powerful as Deligne’s derivation of the correct k − 12 + o(1), is more generally applicable.
The coefficients of the q-series for j(z) grow much more rapidly. Let us write j(z) = q−^1 + 744 +
n=1 c(n)q
n, adopting the notation in [Serre 1973, p.90].
An easy upper bound follows from the positivity of the c(n):
Lemma. cn exp(4π
n).
Proof : All the c(n) are positive by the formula
j(z) = E 43 /∆ = q−^1 E 43
∏^ ∞
n=
(1 − qn)−^24.
Therefore if we take z = iy then q = e−^2 πy^ > 0 and we find
c(n) = q−n^ · c(n)qn^ < e^2 πny^ j(iy). (1)
If y is bounded above, say y < y 0 , then
j(iy) = j(i/y) = e^2 π/y^ + O(1),
because exp(− 2 π/y) < exp(2πy 0 ) and the series 744 +
n=1 c(n)e − 2 πn/y (^0) for
j(i/y 0 ) − exp(2π/y 0 ) converges. Thus (1) yields c(n) exp(2π(ny + y−^1 )). Taking y = n−^1 /^2 , we deduce c(n) exp(4π
n ), as claimed.
It turns out that this bound exceeds the actual growth of c(n) by a factor of only O(
n ). Asymptotic formulas for c(n) and the coefficients of other modular functions with poles at i∞ can be obtained by the “stationary-phase method”, which for c(n) amounts to estimating the integrand in the Fourier integral
c(n) =
− 1 / 2
j(x + in−^1 /^2 ) dx
near the point x = 0 at which it is maximal. The Hardy-Ramanujan method and Rademacher’s refinement of it, originally applied to the partition function (see below), also yields asymptotic expansions and convergent series respectively for c(n) and the coefficients of similar q-expansions.
Product formulas for modular forms
The product formula for ∆. Concerning “Theorem 6 (Jacobi)” of [Serre 1973, VII (p.95)]: A more direct and arguably simpler way to prove the key
relation G 1 (z) − G(z) = 2πi/z between
G 1 (z) =
n
m
(m + nz)^2
and G(z) =
m
n
(m + nz)^2
is as follows. By the q-expansions, both series converge. Therefore
G 1 (z) − G(z) = lim N →∞
( N
n=−N
m
(m + nz)^2
∑^ N
m=−N
n
(m + nz)^2
Now rewrite the limit as
lim N →∞
[
|m|>N
|n|≤N
|m|≤N
|n|>N
]
(m + nz)^2
and note that 1/(m + nz)^2 = N −^2 /
(m + nz)/N
to recognize the sums as Riemann sums for the double integrals ∫
|r|≥ 1
|s|≤ 1
dr ds (r + sz)^2
|r|≤ 1
|s|≥ 1
dr ds (r + sz)^2
But these integrals are elementary: ∫
|r|≥ 1
|s|≤ 1
dr ds (r + sz)^2
|r|≥ 1
2 dr r^2 − z^2
and similarly (^) ∫
|r|≤ 1
|s|≥ 1
dr ds (r + sz)^2
|r|≤ 1
− 2 dr r^2 − z^2
We could now evaluate the integrals directly using partial fractions, but an even cleaner conclusion is to combine (2) and (3), writing their difference as 2
r=−∞ dr/(r
(^2) − z (^2) ). This is easily evaluated using contour integration. Since
z has positive imaginary part, the unique pole of the integrand in the upper half-plane is at r = z, with residue 1/ 2 z. Therefore the integral equals 2πi/z, and we are done.
The modular form η of weight one-half. Jacobi’s product formula
∆(z) = q
∏^ ∞
q=
(1 − qn)^24 (4)
suggests that we consider the 24th root
η(z) = q^1 /^24
∏^ ∞
q=
(1 − qn) (5)
of ∆ as a “modular form of weight 1/2” for the full modular group PSL 2 (Z). (By q^1 /^24 we mean the branch e^2 πiz/^24 of the 24th root of q.) That is, for any g =
(a b c d
∈ SL 2 (Z) we have
η
( (^) az + b cz + d
= g (cz + d)^1 /^2 η(z) (6)
to deduce that the two functions are proportional, and compare leading terms to conclude that they are equal.
[Further examples of this approach to proving identities between q-series can be found in the Exercises. In general θχ(z/qi) is a modular form of weight 1/2 for some congruence subgroup of PSL 2 (Z) depending on q. The pentagonal-number theorem, as well as some of the identities in the Exercises, can also be obtained as special cases of Jacobi’s triple-product identity
∑^ ∞
m=−∞
umqm
2
∏^ ∞
n=
(1 + uq^2 n−^1 )(1 + u−^1 q^2 n−^1 )(1 − q^2 n).
For example, (7) is obtained by substituting (q^3 /^2 , −q^1 /^2 ) for (q, u).]
A remarkable consequence of (7) is Euler’s recurrence for the partition function p(r). This is the number of ways to write an integer r ≥ 0 as the sum of positive integers up to rearrangement of the summands, or equivalently as
n=1 ann (with an being the number of summands that equal n). We readily obtain the generating function
∑^ ∞
r=
p(r)qr^ =
∏^ ∞
n=
(1 + qn^ + q^2 n^ + q^3 n^ +.. .) =
∏^ ∞
n=
(1 − qn)−^1
= 1 + q + 2q^2 + 3q^3 + 5q^4 + 7q^5 + 11q^6 + 15q^7 + · · ·.
By (7) this yields
m=−∞(−1)
mp(r − 1 2 m(3m^ + 1)) = 0 for all integers^ m >^ 0. (The sum is finite because p(r) = 0 for r < 0.) Therefore
p(n) = p(n − 1) + p(n − 2) − p(n − 5) − p(n − 7) + p(n − 12) + − − + · · ·.
Weighted theta functions
Suppose Γ is a lattice in the n-dimensional real inner product space V , and Γ′^ the dual lattice {x ∈ V : ∀y ∈ Γ, (x, y) ∈ Z}. We have seen already that the Poisson summation formula applied to the function exp(−πt|x|^2 ) yields a functional equation connecting the theta functions of Γ and Γ′, which Serre [1973, VII, Prop.16, p.107] writes in the form
ΘΓ(t) = t−n/^2 v−^1 ΘΓ′ (t−^1 )
with v = Vol(V /Γ) the covolume and
ΘΓ(t) :=
x∈Γ
exp(−πt|x|^2 ).
Using the functional equation, Serre then shows that when Γ is a unimodular even lattice the function θΓ(z) := ΘΓ(z/i) is a modular form of weight n/2 for PSL 2 (Z).
This can be generalized in several directions. For instance, if we require only that (x, y) ∈ Q for all x, y ∈ Γ then θΓ is still a modular form of weight n/2 for
some congruence subgroup of PSL 2 (Z). (See the Exercises.) We shall pursue a different generalization, applying Poisson to functions on V more general than exp(−πt|x|^2 ). We begin by introducing these functions.
Harmonic polynomials. Let f : V →C be a function with
V |f^ (x)|^ dx^ <^ ∞. Recall that we defined the Fourier transform fˆ : V →C by
fˆ (y) =
x∈V
e^2 πi(x,y)f (x) dx.
We review some familiar properties of the Fourier transform. We shall apply these only to functions in the Schwartz space S, consisting of infinitely differ- entiable functions on V each of whose derivatives is O((1 + |x|)−N^ ) for all N. We thus assume that f ∈ S, in which case justification of all these properties is straightforward.
Let x 1 ,... , xn be orthonormal coordinate functions on V , so x = (x 1 ,... , xn). Differentiating the integral for fˆ with respect to xj , we find that the the j-th
partial derivative of fˆ equals 2πi times the Fourier transform of xj f (x). Con- versely, by integrating by parts with respect to xj we find that the Fourier transform of the j-th partial derivative fj equals − 2 πiyj fˆ. This immediately yields the observation that if f ∈ S then fˆ ∈ S as well. The Laplacian Lf of f is defined by Lf =
∑n j=1 ∂ (^2) f /∂x 2 j , and does not depend on the orthonormal coordinate system. We see that the Fourier transform of Lf equals − 4 π^2 |y|^2 fˆ , and that the Fourier transform of |x|^2 f equals −(4π^2 )−^1 L fˆ.
We know that, for any t ∈ C with Re(t) > 0, the function f (x) = exp(−πt|x|^2 ) has Fourier transform fˆ (y) = t−n/^2 exp(−πt−^1 |y|^2 ). By induction we can de- scribe the Fourier transforms of functions of the form P (x) exp(−πt|x|^2 ), where P is any polynomial in n variables.
Lemma. For any polynomial P on V , the function P (x) exp(−πt|x|^2 ) on V has Fourier transform Q(y) exp(−πt−^1 |y|^2 ) for some polynomial Q depending on P and t. Moreover, if P is of degree d then so is Q, and the d-th homogeneous part of Q is (i/t)dt−n/^2 times the d-th homogeneous part of P.
Proof : By linearity it is enough to prove this when P is a monomial
∏n j=1 x
dj j. In that case P (x) exp(−πt|x|^2 ) factors as
∏n j=1 x
dj j e −πtx^2 j (^) , so its Fourier transform
is the product of the Fourier transforms of the functions xd j je−πtx
(^2) j
. Let fd (y),
then, be the Fourier transform of xde−πtx 2
. We know that f 0 (y) = t−^1 /^2 e−πt − (^1) y 2
and fd+1(y) = (2πi)−^1 f (^) d′(y). It follows inductively that fd (y) = Qd(y)e−πt
− (^1) y 2
for some polynomial Qd of degree d and leading coefficient t−d−^ (^12)
. Indeed Q 0 = t−^1 /^2 and Qd+1 = iQd/t + Q′ d/ 2 πi. The assertion of the Lemma thus holds
for P (x) =
∏n j=1 x
dj j , when^ Q(y) =^
∏n j=1 Qd(yj^ ), and therefore by linearity for all P.
Now suppose P is homogeneous of degree d. In general Q will not be homoge- neous, except for d = 0 and d = 1. We shall show that the space of polynomi- als P for which Q is homogeneous, and thus equal to (i/t)dt−n/^2 , coincides with
through our derivation of (10) in reverse, with the roles of x and y switched and with t replaced with t−^1. We conclude that ∇Q(y) · y = d · Q(y), and thus that Q is homogeneous of degree d. By our earlier Lemma, Q thus equals (i/t)dt−n/^2 P.
The harmonic polynomials of degree d appear also as an irreducible representa- tion of the orthogonal group O(V ); their restriction to the unit sphere of V are the “spherical harmonics” of degree d, which constitute an eigenspace for the Laplacian on that sphere. They may also be defined as the space of homoge- neous polynomials of degree d whose restriction to the unit sphere is orthogonal to all the polynomials of degree < d.
If P is a harmonic polynomial, we may apply the Poisson summation formula to P (x) exp(−πt|x|^2 ) to find the following generalization of Serre’s Prop.16:
Proposition. Let Γ be a lattice in the n-dimensional real inner-product space V , with dual lattice Γ′^ and covolume v. For any harmonic polynomial P of degree d on V , define the weighted theta function ΘΓ,P by
ΘΓ,P (t) :=
x∈Γ
P (x) exp(−πt|x|^2 ) (Re(t) > 0).
Then ΘΓ,P (t) = (i/t)dt−n/^2 v−^1 ΘΓ′,P (t−^1 )
for all t of positive real part.
Note that this Proposition is nontrivial only for even d, because ΘΓ,P = 0 identically for P of odd degree (why?). In particular we have:
Corollary. If moreover Γ is an even unimodular lattice then
θΓ,P (z) := ΘΓ,P (z/i) =
x∈Γ
P (x)q (^12) |x|^2
is a modular form of weight (n/2) + d for PSL 2 (Z).
Again this generalizes to other lattices: if Γ ⊂ V is a lattice such that (x, x) ∈ Q for all x ∈ Γ, and x 0 is any vector in Γ ⊗ Q, then
θΓ,P,x 0 (z) :=
x∈Γ+x 0
P (x)eπi(x,x)z
is a modular form of weight (n/2)+d for some congruence subgroup of PSL 2 (Z). Unless 2x 0 ∈ Γ, this form is usually nontrivial even for odd d. We have in effect already seen this for d = n = 1 when we introduced the modified theta functions ϑχ(u) prove the functional equation for L(s, χ) when χ(−1) = −1.
One useful application of this construction is the construction of “spherical r-designs”. These are finite subsets S of the unit sphere in V that are very well distributed in the following sense: for any polynomial P of degree at most r, the average of P over the unit sphere equals |S|−^1
x∈S P^ (x). It is known that this is equivalent to the condition that
x∈S P^ (x) = 0 for all nonconstant harmonic
polynomials P of degree at most r. (See the Exercises.) For example, if n = 2 then the vertices of a regular N -gon constitute a spherical (N − 1)-design.
Proposition. Let Γ is a self-dual unimodular lattice in an n-dimensional real inner product space. Assume that Γ contains no nonzero vector x with 6 |x|^2 ≤ (n/2) + r. Then, for each ρ > 0 such that Γ has vectors of length ρ, the set {x/ρ : x ∈ Γ, |x| = ρ} is a spherical r-design.
Proof : Let P be a nonconstant harmonic polynomial of degree d ≤ r. We have seen that θΓ,P is a modular form of weight (n/2) + d. We claim that this form
is identically zero. The Proposition will follow because the qρ
(^2) / 2 coefficient of θΓ,P is ρd^ times the sum of P (x/ρ) over lattice vectors x of length ρ.
Since d > 0, the constant coefficient of θΓ,P vanishes. By hypothesis all the other
terms P (x)q(x,x)/^2 in the sum for θΓ,P have exponent greater than ((n/2)+r)/6. But a cusp form of weight w that vanishes to order > w/6 at the cusp must be identically zero.
It is known that r cannot exceed 11 for any Γ, and that if r = 11 then 24|n and n is bounded above. Examples are known only for n = 24 (the celebrated Leech lattice) and n = 48. (See for instance [CS 1993].) In addition to providing good spherical codes for numerical integration, this Proposition and its generaliza- tions to other kinds of lattices has been used to investigate the existence and uniqueness of certain lattices, notably in the four-page proof [Conway 1969] that the Leech lattice is the unique even unimodular lattice in R^24 with no vectors of length
Exercises
Concerning modular groups and their rings of modular forms:
- i) Prove Serre’s assertion (bottom of p.111) that S and T 2 generate an index- subgroup of PSL 2 (Z) by identifying this subgroup with {
(a b c d
∈ PSL 2 (Z) : 2|b}. ii) Find similar pairs of generators for Γ 0 (N ) (N = 2, 3 , 4), where
Γ 0 (N ) :=
a b c d
∈ PSL 2 (Z) : N |c
In particular, show that Γ 0 (4) is a free group on two generators. iii) What are the groups generated by T and Sd : z ↔ − 1 /dz for d = 2, 3 , 4?
- For each of the groups in the previous Exercise, determine the ring of modular forms and the ideal of cusp forms in that ring.
[There are only finitely many congruence groups that can be treated in this way; can you find any others? See [Takeuchi 1977] for more information. The group generated by T and Sd arises naturally in the analysis of theta functions of lattices Γ with Γ′^ ∼= d−^1 /^2 Γ, sometimes called “isodual lattices”.]
Some more identities and results involving one-dimensional theta functions:
- Prove that
∑∑
m,m′∈Z
mm′(m + m′)χ 3 (m)qm
(^2) +mm′+m′^2 = 12 η(9z)^8
[
= 12q^3
∏^ ∞
n=
(1 − q^9 n)^8
]
Concerning harmonic polynomials etc.:
- Complete the proof that if P (x) is harmonic then P (x) exp(−πt|x|^2 ) has Fourier transform (i/t)dt−n/^2 P (y) exp(−πt−^1 |y|^2 ).
- Prove that the Laplacian is a surjective linear operator on the vector space C[V ] of polynomials on V. Conclude that the space of harmonic polynomials of degree d on an n-dimensional real inner-product space is the difference
(n+d− 1 d
n+d− 3 d− 2
between the dimensions of homogeneous polynomials of degrees d and d − 2.
- Show that the only homogeneous polynomial P (x) such that |x|^2 P (x) is harmonic is the zero polynomial. Deduce that every homogeneous polynomial P (x) of degree d can be written uniquely as
∑bd/ 2 c j=0 |x|
2 j (^) Pj (x) with the Pj
being harmonic polynomials of degree d − 2 j. Conclude that the vector space of polynomial functions of degree at most d on the unit sphere in V is the direct sum of the spaces of spherical harmonics of degree d′^ with d′^ ≤ d, and thus that a finite set S of unit vectors is an r-design if and only if
x∈S P^ (x) = 0 for all nonconstant harmonic polynomials P of degree at most r.
- Prove that, given the dimension of V , there exists for each d = 0, 1 , 2 ,... a unique monic Pd ∈ R[x] of degree d such that |x|dPd(x 1 /|x|) is a harmonic polynomial. [This Pd is proportional to a Gegenbauer orthogonal polynomial, and the spherical harmonic Pd(x 1 ) on the unit sphere of V is called a “zonal spherical harmonic”. These have many applications in mathematics, including those of the next Exercise.]
12.√ i) Let Γ be an even unimodular lattice in R^32 with no vectors of length
- Use the modularity of θΓ to prove that Γ has 146880 vectors of the mini- mal nonzero length 2. Use theta functions weighted by the polynomials of the previous Exercise to prove that if v 0 is one of those vectors then Nk vectors v of length 2 such that (v, v 0 ) = k, where N 0 = 80910, N± 1 = 31744, N± 2 = 240, N± 4 = 1, and Nk = 0 for all k /∈ { 0 , ± 1 , ± 2 , ± 4 }. Obtain similar results for the distribution of (v, v 1 ) where v 1 is a fixed lattice vector of length
6 and v ∈ Γ has length 2, or vice versa. ii) Can you obtain similar conditions on the 39600 minimal vectors of an even unimodular lattice in R^40 with no vectors of length
iii) If Γ is an even unimodular lattice in R^48 whose shortest nonzero vectors have length
6, or in R^72 whose shortest nonzero vectors have length
8, how much combinatorial data can you obtain about the configurations of short vectors in Γ?
[It is known that there are several such “extremal” lattices Γ in dimension 32, and several in dimension 48. Their classification is not yet complete, but all must satisfy these combinatorial constraints. There are literally millions of
inequivalent such lattices in dimension 40. It is not yet known whether there exists an extremal lattice in dimension 72. See again [CS 1993].]
- Prove that a set of N unit vectors in a 2-dimensional real inner-product space is a spherical (N − 1)-design^3 if and only if it comprises the vertices of a regular N -gon.
References
[BB 1987] Borwein, J.M., Borwein, P.B.: Pi and the AGM: a study in ana- lytic number theory and computational complexity. New York: Wiley, 1987. [QA241.B774]
[Conway 1969] Conway, J.H.: A characterisation of Leech’s lattice. Inventiones Math. 7 (1969), 137–142.
[CS 1993] Conway, J.H., Sloane, N.J.A.: Sphere Packings, Lattices and Groups. New York: Springer 1993.
[Elkies 2000] Elkies, N.D.: Lattices, Linear Codes, and Invariants, Notices Amer. Math. Soc. 47 (2000), 1238–1245 and 1382–1391.
[Serre 1973] Serre, J.-P.: A Course in Arithmetic. New York: Springer, 1973.
[Takeuchi 1977] Takeuchi, K: Commensurability classes of arithmetic triangle groups, J. Fac. Sci. Univ. Tokyo 24 (1977), 201–212.
(^3) One might say “circular (N − 1)-design” in this setting...
