Modular Forms: a First Look
Some functions are so rigidly symmetric that once you know them on a tiny sliver of their domain,
you know them everywhere — the symmetry copies that sliver across the whole plane. A
modular form is the crown jewel of this species. It lives on the
upper half-plane \mathbb{H}=\{z\in\mathbb{C}:\Im(z)>0\},
it transforms in a beautifully controlled way under an infinite group of symmetries, and — almost as
an accident — its Fourier coefficients turn out to be some of the most arithmetically loaded numbers
in all of mathematics.
This is the doorway to the modern theory: the coefficients of modular forms are exactly the numbers
that assemble into deep arithmetic
L-functions,
the objects that Andrew Wiles harnessed to prove Fermat's Last Theorem. Let's meet the symmetry
group first, then the forms themselves.
The modular group and its action on ℍ
The modular group is
\mathrm{SL}(2,\mathbb{Z}) — the 2\times2
integer matrices of determinant 1:
\mathrm{SL}(2,\mathbb{Z})=\left\{\begin{pmatrix} a & b \\ c & d\end{pmatrix} : a,b,c,d\in\mathbb{Z},\ ad-bc=1\right\}.
Each such matrix acts on a point z\in\mathbb{H} as a
Möbius (fractional-linear) transformation:
\gamma\cdot z = \frac{az+b}{cz+d}.
A short computation with ad-bc=1 shows
\Im(\gamma z)=\dfrac{\Im(z)}{|cz+d|^2}>0, so the upper half-plane maps to
itself — the group shuffles \mathbb{H} around without ever leaving it.
The whole infinite group is built from just two generators:
- T=\begin{pmatrix}1&1\\0&1\end{pmatrix}, acting as the horizontal
shift T\!:\,z\mapsto z+1;
- S=\begin{pmatrix}0&-1\\1&0\end{pmatrix}, acting as the inversion
S\!:\,z\mapsto -1/z;
- every element of \mathrm{SL}(2,\mathbb{Z}) is a product of copies of
S and T (and their inverses).
That last bullet is the practical gift: to check that a function respects the whole group,
you only ever need to check it against S and T.
What a modular form is
The word "form" refers to the twist: a modular form doesn't stay equal under the group, it
picks up a precise, predictable factor. Fix an even integer k, the
weight.
A modular form of weight k for
\mathrm{SL}(2,\mathbb{Z}) is a function
f:\mathbb{H}\to\mathbb{C} satisfying all three conditions:
- Holomorphic on \mathbb{H} (complex-differentiable everywhere);
- Modular of weight k:
\displaystyle f\!\left(\frac{az+b}{cz+d}\right)=(cz+d)^k f(z)
for every \begin{pmatrix}a&b\\c&d\end{pmatrix}\in\mathrm{SL}(2,\mathbb{Z});
- Holomorphic at the cusp i\infty: f stays bounded as \Im(z)\to\infty.
The factor (cz+d)^k is called the automorphy factor.
It is the single most important character in the story — set it to 1
(weight 0) and you get an honestly invariant function; keep it,
and you get the rich weighted world where arithmetic hides.
The q-expansion: turning ℍ into a disc
Apply the modularity condition to the generator T. Its matrix has
c=0,\ d=1, so the automorphy factor is (0\cdot z+1)^k=1,
and the rule becomes simply
f(z+1)=f(z).
So every modular form is periodic with period 1. Any such
function can be written as a Fourier series in the variable
q=e^{2\pi i z},
which sends the strip |\Re(z)|\le\tfrac12,\ \Im(z)>0 onto the punctured
unit disc (as \Im(z)\to+\infty, q\to0). The
cusp condition says f extends holomorphically to q=0,
so it has a genuine power series — the q-expansion:
f(z)=\sum_{n\ge0} a_n\,q^n,\qquad q=e^{2\pi i z}.
The coefficients a_n are the arithmetic payload. The constant term
a_0 is the value "at the cusp", f(i\infty)=a_0.
When it vanishes, the form is something special.
A modular form f=\sum_{n\ge0}a_n q^n is a cusp form if it
vanishes at the cusp, i.e. a_0=0, so
f(z)=a_1 q + a_2 q^2 + \cdots \to 0 as \Im(z)\to\infty.
The cusp forms of weight k form a subspace
S_k\subset M_k.
The fundamental domain
Because \mathrm{SL}(2,\mathbb{Z}) shuffles \mathbb{H}
so thoroughly, there is a single tile — a fundamental domain — whose copies under
the group cover the whole upper half-plane exactly once. The standard choice is the region
\mathcal{F}=\left\{z\in\mathbb{H} : |z|\ge 1 \ \text{and}\ |\Re(z)|\le\tfrac12\right\}.
Its left and right edges (\Re(z)=\pm\tfrac12) are glued by
T; the two halves of the bottom arc (|z|=1) are
folded together by S. Step through the picture below: the boundary, the
region itself, its two corner points, and a neighbouring tile.
The two corners are geometrically special: \rho=e^{2\pi i/3}=-\tfrac12+\tfrac{\sqrt3}{2}i
(and its mirror \rho+1=e^{\pi i/3}) is fixed by an order-3
symmetry, while i is fixed by the order-2 element
S. Those fixed points are exactly why the valence formula (below) carries
fractions \tfrac12 and \tfrac13.
Worked example — verifying the transformation on S and T
Suppose f is periodic (so f(z+1)=f(z), the
T-condition already met) and satisfies the single extra equation
f\!\left(-\tfrac1z\right)=z^k f(z).
Claim: then f is modular of weight
k for the entire group. Why? For the matrix
S=\begin{pmatrix}0&-1\\1&0\end{pmatrix} we read off
a=0,b=-1,c=1,d=0, so
\dfrac{az+b}{cz+d}=\dfrac{-1}{z} and the automorphy factor is
(cz+d)^k=z^k — exactly the equation above. For
T=\begin{pmatrix}1&1\\0&1\end{pmatrix} we have c=0,d=1,
factor 1, giving f(z+1)=f(z). Since
S and T generate the group, and the
automorphy factors multiply consistently under composition (the cocycle relation),
the two checks propagate to every matrix. Two equations pin down infinitely many.
Write j(\gamma,z)=cz+d for the automorphy factor of
\gamma=\begin{pmatrix}a&b\\c&d\end{pmatrix}. A direct computation shows
it satisfies the cocycle identity
j(\gamma_1\gamma_2,\,z)=j(\gamma_1,\,\gamma_2 z)\cdot j(\gamma_2,\,z).
This is precisely what makes the weight-k condition
composable: if f transforms correctly under
\gamma_1 and under \gamma_2, the factors
(cz+d)^k stack up exactly right so it transforms correctly under the
product \gamma_1\gamma_2. Verify a generating set and you've verified
the whole group for free — no infinite checking required.
The headline examples
Eisenstein series E_k
The most direct way to manufacture a weight-k form is to average
the automorphy factor over the group — the Eisenstein series. For even
k\ge4 (needed for convergence), the normalised version has the clean
q-expansion
E_k(z)=1-\frac{2k}{B_k}\sum_{n\ge1}\sigma_{k-1}(n)\,q^n,
where \sigma_{k-1}(n)=\sum_{d\mid n}d^{k-1} is the divisor-power sum and
B_k is a Bernoulli number. The two workhorses are
E_4(z)=1+240\sum_{n\ge1}\sigma_3(n)q^n=1+240q+2160q^2+6720q^3+\cdots,
E_6(z)=1-504\sum_{n\ge1}\sigma_5(n)q^n=1-504q-16632q^2-\cdots.
Worked example — the q^2 coefficient of E_4.
We need 240\,\sigma_3(2). The divisors of 2 are
1 and 2, so
\sigma_3(2)=1^3+2^3=1+8=9, giving
240\times9=2160. That the coefficients of a symmetric analytic function are
literally sums of cubes of divisors is the first taste of why modular forms encode
arithmetic.
The discriminant \Delta and Ramanujan's \tau
The first cusp form appears at weight 12. It is the
modular discriminant
\Delta(z)=q\prod_{n\ge1}(1-q^n)^{24}=\sum_{n\ge1}\tau(n)\,q^n = q-24q^2+252q^3-1472q^4+\cdots,
a weight-12 cusp form (note a_0=0, and
a_1=1). Its coefficients define Ramanujan's tau function
\tau(n). It also has the tidy expression
\Delta=\dfrac{E_4^{\,3}-E_6^{\,2}}{1728}.
The j-invariant
Divide a weight-12 form by another and the automorphy factors cancel,
leaving a weight-0 function. The famous one is the
j-invariant
j(z)=1728\,\frac{E_4^{\,3}}{\Delta}=\frac1q+744+196884\,q+21493760\,q^2+\cdots,
which is genuinely invariant: j(\gamma z)=j(z) for all
\gamma. It has a pole at the cusp (the 1/q
term), so it is a modular function, not a modular form — the weight-0
world drops the holomorphy-at-the-cusp requirement.
The q-coefficient of j is
196884. The smallest non-trivial irreducible representation of the
Monster group — the largest sporadic finite simple group — has dimension
196883. In 1978 John McKay noticed
196884=196883+1, an observation so improbable it was nicknamed
"monstrous moonshine." It turned out to be no coincidence at all: the whole
q-expansion of j encodes the Monster's
representation theory, a link proved by Richard Borcherds and worth a Fields Medal. Modular forms
reach into places nobody expected.
The space M_k is finite-dimensional
Here is the structural miracle that makes the theory usable. For each weight the modular forms form a
vector space M_k — and it is finite-dimensional, usually
tiny. A weight-k form is pinned down by only a handful of numbers.
For even k\ge0 (and M_k=0 for odd or negative
k),
- \dim M_k=\left\lfloor \tfrac{k}{12}\right\rfloor if
k\equiv 2 \pmod{12}, and
\left\lfloor \tfrac{k}{12}\right\rfloor+1 otherwise;
- consequently \dim M_0=\dim M_4=\dim M_6=\dim M_8=\dim M_{10}=1, while
\dim M_{12}=2 (the first weight with a cusp form).
And they all knit together into one object. Multiplying a weight-k form by
a weight-\ell form gives a weight-(k+\ell) form,
so the direct sum M_\bullet=\bigoplus_k M_k is a graded ring — and it is as
simple as a ring can be:
The graded ring of modular forms for \mathrm{SL}(2,\mathbb{Z}) is the free
polynomial ring
\bigoplus_{k} M_k=\mathbb{C}[E_4,E_6],
with E_4 and E_6 algebraically independent. Every
modular form is a unique polynomial in E_4 and E_6.
Worked example. Since \dim M_8=1 and
E_4^{\,2} is a weight-8 form with constant term
1, we must have E_8=E_4^{\,2} outright.
Comparing q-coefficients gives the non-obvious identity
\sigma_7(n)=\sigma_3(n)+120\sum_{m=1}^{n-1}\sigma_3(m)\sigma_3(n-m) — an
arithmetic convolution formula that falls out purely because a two-dimensional-looking space is
actually one-dimensional.
The valence formula — counting the zeros
Why is M_k so small? Because a modular form has almost no room to have
zeros. Counting them (weighted by the corner symmetries) gives a rigid constraint.
For a non-zero modular form f of weight k, the
orders of vanishing satisfy
\operatorname{ord}_\infty(f)+\tfrac12\operatorname{ord}_i(f)+\tfrac13\operatorname{ord}_\rho(f)+\!\!\sum_{P\ne i,\rho}\!\!\operatorname{ord}_P(f)=\frac{k}{12},
the sum running over points P of the fundamental domain. The fractions
\tfrac12,\tfrac13 come from the extra symmetry at
i and \rho.
Worked example. Take \Delta, weight
12, so the right-hand side is 1. Since
\Delta=q\prod(1-q^n)^{24} and the product is non-zero on
\mathbb{H}, its only zero is a simple one at the cusp
(\operatorname{ord}_\infty=1) — accounting for the entire budget of
1. Hence \Delta is non-vanishing on all of
\mathbb{H}, which is exactly why we're allowed to divide by it to build
j.
Three subtleties trip up every newcomer:
-
The weight is even. For the full group
\mathrm{SL}(2,\mathbb{Z}), the matrix
-I=\begin{pmatrix}-1&0\\0&-1\end{pmatrix} acts trivially on
\mathbb{H} but its automorphy factor is
(-1)^k. Modularity forces f=(-1)^k f, so for
odd k the only form is f\equiv0. There are
no odd-weight forms for the full modular group — only even k.
-
The factor (cz+d)^k is not optional. Drop it (set
k=0) and you demand a genuinely invariant holomorphic bounded
function — but those are just constants (a holomorphic function invariant under the group, bounded
on the compact quotient, must be constant). All the interesting content lives in the weighted
transformation. Weight 0 with a pole allowed gives modular
functions like j, not forms.
-
Holomorphy at the cusp is part of the definition. A function can be holomorphic on
\mathbb{H} and transform correctly yet blow up as
\Im(z)\to\infty — that is not a modular form. You must also
demand a well-behaved (no negative powers of q) expansion at
i\infty. This is what separates the forms E_k,\Delta
from a function like j, whose 1/q disqualifies
it as a form.
In 1916 Ramanujan stared at \tau(n) and conjectured two astonishing
things. First, that it is multiplicative:
\tau(mn)=\tau(m)\tau(n) whenever
\gcd(m,n)=1 — a hint that \Delta is a
simultaneous eigenfunction of a family of operators (the
Hecke operators).
Second, the sharp size bound |\tau(p)|\le 2p^{11/2} for primes
p. Mordell proved the multiplicativity soon after — but the size bound
resisted for fifty-eight years, until Pierre Deligne derived it in 1974 as a corollary of his
proof of the Weil conjectures, work that won him a Fields Medal. A single coefficient of a single
modular form sat at the frontier of algebraic geometry for over half a century.
Where this is heading
We built modular forms as symmetric functions on \mathbb{H}, but their real
power is arithmetic. Because \Delta (and the Eisenstein series) are
eigenfunctions of the Hecke operators, their coefficients are multiplicative, and that
multiplicativity is exactly what lets you package them into an Euler product — an
L-function
of the form L(f,s)=\sum_{n\ge1}a_n n^{-s}=\prod_p(\cdots)^{-1}. These
L-functions have analytic continuations and functional equations mirroring the
Riemann zeta function,
and they are the objects at the heart of the modern Langlands program and the
general theory of L-functions.
A "first look" at a symmetric function on the upper half-plane turns out to be a first look at the
deepest structure in number theory.