The Partition Function Revisited: M. Ram Murty
The Partition Function Revisited: M. Ram Murty
1. Introduction
Almost a century ago, G. H. Hardy and Srinivasa Ramanujan wrote a landmark paper
on the partition function. This paper was significant for two reasons. On the one hand,
it was the first paper to have determined the asymptotic behaviour of the partition
function and on the other, it was significant in that it introduced the circle method into
number theory, which later became an effective tool to study additive questions. A few
years ago, Bruinier and Ono [1] derived an algebraic formula for the partition function
using the theory of harmonic weak Maass forms. In a recent paper [3], the author
and M. Dewar derived the Hardy-Ramanujan asymptotic formula from this algebraic
formula. In this way, we bypass the circle method. In a later paper, [4], we showed that
the asymptotics of the partition function also leads to the asymptotics of coefficients
of the j -function, something which was originally done by Petersson [19] in 1932 and
later Rademacher [20] in 1938 using the circle method. This is particularly interesting
from a historical perspective since our paper shows that the asymptotic behaviour of
the coefficients of the j -function is almost an immediate consequence of the HardyRamanujan paper [9] written in 1918. This paper is a short survey of these recent
developments in the study of the asymptotics of the partition function and related
functions.
Research of the author was partially supported by an NSERC Discovery grant.
261
262
M. Ram Murty
The partition function, denoted p(n), is the number of ways of writing n as a nondecreasing sum of positive integers. Thus, p(1) = 1, p(2) = 2, p(3) = 3 and
p(4) = 5 since
4,
1 + 3,
2 + 2,
1 + 1 + 2,
1+1+1+1+1
are the five partitions of 4. Thus, each partition can be factored uniquely as
1k1 2k2
where the notation symbolizes
n=1
+ 1 + + 1 + 2 + 2 + + 2 +
k1
k2
The partition function is ubiquitous in mathematics. For instance, p(n) is the number
of conjugacy classes of the symmetric group Sn . Since (for any group) the number of
conjugacy classes is equal to the number of irreducible representations, it is natural to
ask if there is a canonical method of constructing all the irreducible representations
from the knowledge of the partitions. Indeed, this is the case and the classical theory
of Young diagrams and Specht modules comprises an aesthetically complete chapter
in the representation theory of the symmetric group. Partitions also play a major role
in the representation theory of G L n . We refer the reader to [7] for further details.
The study of p(n) has a long venerable history and there are many questions
still unanswered about its behaviour. Euler was the first to begin a systematic study
of p(n). From the unique factorization alluded to above, one can write its generating function as
p(n)q n =
(1 q n )1 , |q| < 1.
n=0
n=1
This equation already suggests a modular connection since the famous Dedekind
-function given by
(z) = q
1/24
(1 q n ),
q = e2 i z ,
n=1
(1 q n ) =
n=1
(1)n q
3n 2 +n
2
(1)
n=
n=1
(1 q n )3 =
n=0
(2)
263
24
(n)q n ,
q = e2 i z ,
n=1
and showed that the coefficients (n) are of sufficient arithmetic interest. This motivated his celebrated conjectures regarding the -function and these conjectures had a
pivotal role in the development of 20th century number theory.
In 1972, I. G. MacDonald [13] discovered a fascinating connection between powers
of the -function and properties of affine root systems in the theory of classical Lie
algebras. In fact, MacDonald [13] discovered elegant formulas for the coefficients of
powers of , namely dim g, where g is a finite dimensional simple Lie algebra, and
dim g denotes its dimension. Jacobis formula is then a special case of MacDonalds
identities specialized to the Lie algebra sl2 . In the particular case of the 24-th power of
the -function, MacDonald derives the following curious formula for the Ramanujan
-function:
1
(n) =
(u i u j ),
(3)
1!2!3!4!
i< j
5
i=1
u i = 0,
5
u 2i = 10n.
i=1
Apparently, this was discovered earlier by the physicist Freeman Dyson [5] using
pedestrian methods or so he writes. In his entertaining and inspiring article, Dyson
relates how he had come across this and writes that it is rather surprising that
Ramanujan did not think of it himself. He continues, Pursuing these identities
further by my pedestrian methods, I found that there exists a formula of the same
degree of elegance as (3) for the d-th power of whenever d belongs to the following
sequence of integers:
d = 3, 8, 10, 14, 15, 21, 24, 26, 28, 35, 36, . . .
There I stopped. I stared for a little while at this queer list of numbers. As I was,
for the time being, a number theorist, they made no sense to me. My mind was so
well compartmentalized that I did not remember that I had met these same numbers
many times in my life as a physicist. If the numbers had appeared in the context of a
problem in physics, I would certainly have recognized them as the dimensions of the
finite dimensional Lie algebras. Since the work of Dyson and MacDonald, there has
been a lot of activity and the relationship of these identities to Lie theory is now well
understood. For instance, we refer the reader to the highly readable article [15] as well
as the excellent Seminaire Bourbaki article by Demazure [2].
Returning to Eulers formula (1), we see that the coefficients of the power series
expansion of
1
p(n)q n
n=0
264
M. Ram Murty
are lacunary and equal to 0 or 1. It is conjectured that the only odd powers of
which are lacunary are the cases of Euler and Jacobi cited above, that is, d = 1, 3. For
positive even values of d, Serre [23] has determined that the complete list is
{2, 4, 6, 8, 10, 14, 26}.
That the coefficients of d are lacunary for d equal to 2, 4, 6, and 8 is due to
Ramanujan and appears in his celebrated paper [21] in which he makes his famous
conjectures about the -function. (Curiously, there is a typo in (104) in the statement
of Ramanujans conjecture on the -function.)
Positive powers of are modular forms either of integral weight or half-integral
weight. In both cases, the growth of the Fourier coefficients is polynomial. By contrast,
the coefficients of the negative powers of are of exponential growth and this is one
of the corollaries of the Hardy-Ramanujan work.
2n/3
,
4n 3
n .
(4)
In their proof, they discovered a new method called the circle method which made
fundamental use of the modular property of the Dedekind -function. We see from the
Hardy-Ramanujan formula that p(n) has exponential growth.
As pointed out by Selberg [22], the circle method has its origins in Ramanujans
first letter to G. H. Hardy written on January 11, 1913 from India. There, he wrote
the coefficient of q n in
(1 2q + 2q 4 2q 9 + 2q 16 )1 ,
is very nearly
sinh n
1
cosh n
.
n
4n
n
Ramanujan seems to have discovered the asymptotic behaviour of the n-th term of a
modular form of weight 1/2. As noted above, the generating function for the partition function is also (apart from a factor of q 1/24 ) a modular form of weight 1/2
and thus, the asymptotic behaviour of p(n) is closely akin to the formula given in
Ramanujans letter. Twenty years after the publication of the Hardy-Ramanujan paper,
Rademacher discovered an explicit formula for p(n):
2
1
sinh
(n
k
3
24
1
d
p(n) =
k Ak (n)
dn
1
2 k=1
n 24
Ak (n) =
265
e i(s(m,k)2mn/k) ,
0m<k,(m,k)=1
and
s(b, c) =
c
j
jb
1
cot
cot
,
4c
c
c
j =1
is the so-called Dedekind sum. Notice that when k = 1, the first term resembles
Ramanujans term in his letter of January 11, 1913.
It seems that Selberg [22] arrived at this formula independently and he wrote,
In the summer of 1937, I had actually myself been studying the paper by Hardy and
Ramanujan and had arrived at Rademachers formula. . . . It always seemed strange to
me that Hardy and Ramanujan did not end up with this formula. . . and I believe firmly
that the responsibility for this rests with Hardy.
n=1
(n)q n ,
(n) =
d,
q = e2 i z ,
d|n
1
= 10 29q 104q 2 +
q
(5)
One can show that F(z) is a weight 2 meromorphic modular form on 0 (6). Let
1 d
1
F(z).
P(z) =
+
2i d z 2 y
This is a weak Maass form of weight zero. It is an eigenfunction of the hyperbolic
Laplacian:
2
2
y 2
+ 2
x 2
y
with eigenvalue 2. Now consider all positive definite integral binary quadratic forms
Q(x, y) = ax 2 + bx y + cy 2 with discriminant b2 4ac = 24n + 1, and 6|a,
266
M. Ram Murty
b 1 (mod 12). The group 0 (6) acts on such forms in the obvious way and we let
Q6n be the finite set of orbit representatives under this action. Let
Q =
b +
1 24n
.
2a
1
P( Q ).
24n 1
6
(6)
QQn
The natural question that arises from this result is if one can derive both the HardyRamanujan formula and the Rademacher formula for p(n) from (6). Since the derivation of (6) uses the theory of weak harmonic Maass forms and not the circle method,
we would then have a new derivation of (4) if we can derive (4) from (6). Indeed, this
is the case as shown in [3]. We proved that for any natural number N ,
(24n1) p(n) =
N
m=1
cm 1
24n1
6m
where the cm s depend on the congruence class of n (mod m) and can be given explicitly (see [3] for the precise formula). For instance,
5
, etc.
c1 (n) = 2 3, c2 (n) = 2(1)n cos
+ cos
12
12
and h(1 24n) is the class number of Q( 1 24n). In [3], we show that both the
Hardy-Ramanujan formula and a formula related to the Rademacher formula can be
deduced.
To elaborate, an integral binary quadratic form
Q(x, y) = ax 2 + bx y + cy 2 ,
a, b, c Z
S L 2(Z),
g :=
(x, y) := Q(x + y, x + y).
(7)
Q
267
z +
.
z :=
z +
For any f : h C, and g S L 2 (Z) given by (7), we define the slash operator
|k by
z +
k
( f |k g)(z) := ( z + ) f
.
z +
It is routine to check that for g S L 2 (Z),
Qg1 = g Q .
We say Q = [a, b, c] is reduced if |b| a c. One can show that a form is reduced
if and only if Q lies in the standard fundamental domain:
We denote by Q1n the set of primitive forms (that is, gcd(a, b, c) = 1) of discriminant
1 24n with Q D . As mentioned before, the group
S L 2 (Z) : 6|
0 (6) =
acts on the set of primitive forms [a, b, c] of discriminant 1 24n with 6|a and
b 1 (mod 12). Let Q6n be the set of equivalence classes of forms for this action.
Gross, Kohnen and Zagier [8] proved there is a bijection between Q1n and Q6n and this
bijection can be given explicitly (see [3] as well as the proposition on page 505 of [8]).
Now, one can write down the q-expansion of P(z):
1
29
5
P(z) = 1
q 1 +
+ 29 +
q +
2 y
y
2 y
Using this q-expansion and standard theory of modular forms, we proceed to derive
the Hardy-Ramanujan formula from the Bruinier-Ono formula. The first step is
to transform the sum in (6) to a sum over Q1n so that the Q lie in the standard
fundamental domain. To do this, we use the explicit bijection provided by the GrossKohnen-Zagier theorem. Indeed, the group 0 (6) has index 12 in S L 2(Z) and we can
choose an explicit set of right coset representatives:
1 0
=
0 1
1 0 1 r
for r = 0, 1
1 ,r =
3
3 1 0 1
(8)
1 1 1 s
1 ,s =
for s = 0, 1, 2
2
2 3 0 1
1 t
0 1
for t = 0, 1, 2, 3, 4, 5.
0,t =
1 0
0 1
268
M. Ram Murty
Table 1. The matrix Q for Q = [a, b, c] Q1n . Triples (a, b, c) which can never
occur are omitted or left blank.
1
a 0 (mod 6)
a 1 (mod 6)
a 2 (mod 6)
a 3 (mod 6)
a 4 (mod 6)
a 5 (mod 6)
1 ,0
2
1 ,2
2
1 ,1
2
1 ,0
2
1 ,2
2
1 ,1
2
1 ,1
3
1 ,0
3
1 ,1
3
1 ,0
3
1 ,1
3
1 ,0
c 0 (mod 6)
c 2 (mod 6)
0,1
0,3
0,4
1 ,0
0,2
1 ,0
0,5
1 ,2
c 0 (mod 6)
c 2 (mod 6)
c 4 (mod 6)
1 ,0
3
1 ,0
3
1 ,0
0,3
0,1
0,5
1 ,1
3
1 ,1
3
1 ,1
6)
6)
6)
6)
1 ,1
1 ,0
0,4
1 ,0
0,1
c 0 (mod 6)
c 4 (mod 6)
0,5
c
c
c
c
0
2
3
5
(mod
(mod
(mod
(mod
0,2
(mod
(mod
(mod
(mod
1 ,2
2
0,1
0,4
1 ,1
2
0,2
0,5
0,4
11
0,0
0,5
0,4
0,3
0,2
0,1
6)
6)
6)
6)
c
c
c
c
0
1
3
4
(mod
(mod
(mod
(mod
(mod
(mod
b (mod 12)
5
7
6)
6)
6)
6)
6)
6)
c
c
c
c
c
c
0
1
2
3
4
5
0,5
0,0
0,0
1 ,1
2
0,3
1 ,1
2
0,0
0,4
0,2
0,0
1 ,2
2
0,3
1 ,2
2
0,3
0,2
0,1
0,0
Here is the bijection between Q1n and Q6n . For each Q Q1n , there exists a unique
coset representative Q from the above list such that the equivalence class of Q Q1
belongs to Q6n . The explicit value of Q is given in Table 1 of [3] which we reproduce
above.
Using the bijection between Q6n and Q1n , the Bruinier-Ono formula now becomes
(24n 1) p(n) =
P( Q Q )
QQ1n
To determine the asymptotics of the right hand side, we exploit the modular
connection. The Maass raising operator Rk acts on complex-valued functions on the
upper half-plane and is given by
+ =i
i
+ .
Rk = 2i
z
y
x
y
y
269
1
R2 F.
4
1
1
P|0 =
R2 F |0 =
R2 (F|2 ).
4
4
Let 6 = e2 i/6 . The Fourier expansion of F at the cusp i is given by (5). The
modular curve X 0 (6) has four cusps given by 0, 1/3, 1/2, i. At the other cusps,
1/3, 1/2 and 0, the expansions are given by
F |2 1 ,r = 2(1)r q 1/2 + 20 (1)r 34q 1/2 +
3
(9)
Since
P|0 Q =
1
R2
R2 F |0 Q =
(F|2 Q ),
4
4
we write this as
R2
R2
(h Q Q q 1/h Q ) +
(F|2 Q h Q Q q 1/h Q ).
4
4
Now, the first term is easily computed to be
hQ
2 i z/h Q
.
1
Q e
2 y
Thus,
R2
hQ
P( Q z) = (P|0 Q )(z) = Q 1
e2 i z/h Q +
(F|2 Q h Q Q q 1/h Q ).
2 y
4
It is not difficult to show that the second term is bounded by an absolute constant
(see [3] for details). Thus, the Bruinier-Ono formula leads to
hQ
e2 i Q /h Q + O(h(1 24n)).
Q 1
(24n 1) p(n) =
2
(
)
Q
1
QQn
270
M. Ram Murty
It is now easy to derive the Hardy-Ramanujan formula from this formula. First, since
the dominant term in each summand is e2 i Q /h Q , we see that for Q = [a, b, c]
this is
m1
[a,b,c] e ib/6m
Q=[a,b,c,]Qn
ah Q =6m
6m
24n1
e 6m + O(h(1 24n)).
24n 1
The dominant term arises when ah Q = 6 and then the exponential term in (10) is
e
2n/3
which begins to agree with the exponential term in the Hardy-Ramanujan asymptotic
formula. The forms Q that give rise to ah Q = 6 are quickly identified and are four in
number:
[1, 1, 6n], [2, 1, 3n], [3, 1, 2n], and [6, 1, n].
For each form, the corresponding root of unity Q is 61 , 65 , 1, 1 respectively (see [3]
for precise details). Since forms corresponding to m 2 in the sum, that is, with
ah Q 12 contribute at most
O(h(1 24n)e
n/6
24n1
e i/6 (61 + 65 + 1 + 1)e 6
= 2 3e 24n1/6 .
Putting everything together, we deduce
p(n)
2n/3
,
4n 3
as n tends to infinity. As indicated in [3], a formula related to the Rademacher-Selberg
explicit formula can also be derived from this analysis.
271
2k
k1 (n)q n ,
Bk
k1 (n) =
d k1 ,
q = e2 i z ,
d|n
n=1
where Bk denotes the k-th Bernoulli number. For k 4, this is a modular form of
weight k for the full modular group S L 2(Z). Recall also that the j -function is given by
j (z) =
Let us write
E 4 (z)3
.
(z)
j (z) =
1
c(n)q n ,
+
q
q = e2 i z .
n=0
e4 n
c(n)
.
2n 3/4
It would be of interest if one could derive a Bruiner-Ono type formula for c(n) and
thus obtain an exact formula for these coefficients. They are of substantial interest
since they are related to the representation theory of the Monster group, as was discovered by McKay and Borcherds. Initially, Dewar and I were looking for such a formula, but soon realized that the asymptotic beahviour of c(n) can be deduced from
the asymptotics of p(n). Here is a sketch of how this is done. For the details, we refer
the reader to [4].
Let Mk! be the space of weight k weakly holomorphic modular forms on S L 2 (Z).
These are meromorphic modular forms whose only poles (if any) are at i. If f Mk!
has ordi f = m < 0, then
f =
[k/12]+m
b j E k+12(m j ) j m ,
j =0
for some b j C, b0
= 0. This is because f m is an ordinary (holomorphic) modular
form of weight k + m and one has an explicit basis in terms of E k s and powers of .
Using the asymptotics of p(n), we proved in [4] the following:
272
M. Ram Murty
f (n)q n ,
q = e2 i z .
n=m
Then, as n ,
f (m) n k/21/4 4 nm
f (n) i k
e
,
m
2n
i=
1.
This theorem will be used to prove the following theorem of Petersson [19] and
Rademacher [20]:
Theorem 2. The j -function is weakly holomorphic of weight zero with ordi j = 1
so that
1
c(n)
e4 n ,
2n 3/4
which is the Petersson-Rademacher formula.
The proof makes use of Laplaces saddle point method. In the course of our studies,
we derived the following theorem which is of independent interest and is really a
theorem concerning power series.
Theorem 3. Suppose
f (z) =
f (n)q n ,
n=0
g(z) =
g (n)q n ,
n=0
where
f (n) c f n e A
g (n) cg n e B
f g (n)q n ,
n=0
we have
f g (n) c f cg 2 2
A2+1 B 2+1
++3/4
( A2 +B 2 )n
n
e
,
2
2
++5/4
(A + B )
as n .
The idea of the proof of this theorem is easy to explain. Clearly,
f g (n) =
n
r=0
f (r)g (n r).
(11)
273
n
r (n r) e A
r +B nr
r=0
n
G(r/n)e
n F(r/n)
r=0
where
and
G(x) = x (1 x)
F(x) = A x + B 1 x,
both being continuous functions on (0, 1). The function F(x) is easily seen to be an
increasing function on (0, c) and a decreasing function on (c, 1), where
c=
A2
A2
.
+ B2
Note that F(c) = A2 + B 2. It is evident that the major contribution to the sum will
come from values of j such that j/n is close to c. By the continuity of G, for any
> 0, there is a > 0 such that for |x c| < 2
(1 )G(c) < G(x) < (1 + )G(c)
so that our sum (11) in question can be split into three parts:
S0 + S1 + S2
according as r (c )n; (c )n < r < (c + )n; r (c + )n, respectively. The
first and last sums are easily shown to be negligible. Indeed,
[(c)n]
S0 = O n ||+||
e n F(r/n)
r=0
= O(n
||||+1
n F(c)
= o(n ++3/4 e
n F(c)
(12)
)
).
Similarly,
S2 = O(n ||+||+1 e
= o(n
++3/4
n F(c+)
n F(c)
).
(13)
274
M. Ram Murty
Thus, the main contribution comes from S1 . For n sufficiently large, we see that
(1 )2 c f cg n + T < S1 < (1 + )2 c f cg n + T ,
where
T =
[(c+)n]
G(r/n)e
n F(r/n)
r=[(c)n]
[(c+)n]
r=[(c)n]
1 n F(r/n)
e
.
n
It remains to determine the asymptotics of R, which resembles a Riemann sum for the
integral
c+
e n F(x) d x.
c
This intuition is accurate and the latter integral can be studied using Laplaces saddle
point method (sometimes called the method of stationary phase in the literature). The
precise details of the transition from our sum to the integral are given in Lemma 4 of
our paper [4]. For this exposition, it suffices to indicate the general philosopy of the
method, which can be applied to study the asymptotic behaviour of
b
et F(x) d x,
a
1
2
e n F(c)
e n 2 F (c)(xc) d x.
c
n 1/4 (x
n F(c) 1/4
n 1/4
n 1/4
e2 F
(c)v 2
dv
275
keeping in mind that F (c) < 0, so that convergence is assured. We finally conclude
that
2
++3/4
n F(c)
n
e
,
S1 c f cg G(c)
|F (c)|
from which our result follows. This completes our sketch of the proof of Theorem 3.
We apply this to derive asymptotics of the multipartition function (sometimes called
the colored partition function), pk (n) which is given by the generating function
pk (n)q n =
n=0
(1 q n )k .
n=1
k = k/4 3/4,
pk (n) ck n k e Ak
Ak = 2k/3.
as n tends to infinity.
Proof. We proceed by induction and use Theorem 3. For k = 1, this is the HardyRamanujan formula. Applying Theorem 3 we easily find the recursions:
k+1 = k 1/4
Ak+1 = A2k + 2 2 /3
from which the formulas for k and Ak easily follow. The only complicated term is
the constant term ck+1 for which we have the recurrence:
2 +1
Ak k ( 2/3)1
ck
ck+1 = 2 2
4 3 ( A2k + 2 2 /3)k/42/4
n=1
(1 q n )24k =
n=0
p24k (n)q n .
276
M. Ram Murty
1
p24k (n) (k/n)6k+1/4 e4 kn .
2n
f (n)q n ,
f (n) 0,
q = e2 i z .
n=0
Suppose that
f (n) c f e A
/n ,
c f , A, > 0.
Let E k be the normalized Eisenstein series of weight k for the full modular group.
Then, for
f E k (z) =
f Ek (n)q n ,
n=0
we have
c f eA n
f Ek (n) k/2
n
4i
A
k
,
i=
1.
2k
k1 (n)q n ,
Bk
n=1
is the Eisenstein series of weight k for the full modular group with
k1 (n) =
d k1 ,
d|n
k
to recall that i Bk < 0 where i = 1. In other words, the coefficients of the
q-expansion of g = i k (E k 1) are all positive. Then,
f E k = f + i k f g,
so that
f Ek (n) = f (n) + i k f g (n).
Since the asymptotics of the first term on the right hand side is given, it suffices to
determine the asymptotics of the second. As before,
f g (n) =
n
r=0
f (n r)g (r).
277
Because of the exponential growth of f (n), the bulk of the contribution to this sum
comes from terms r with r < n with < 1. Thus, putting H(x) = (1 x) for
0 < x < 1, our sum becomes
c f n
g (r)H(r/n)e A n 1r/n .
r< N
g (r)e Ar/2
r<n
It is not difficult to extend the sum to infinity and calculate the negligible error term
that emerges. Our sum then becomes
c f eA
n
g (r)e Ar/2
r=1
c f eA n
k/2 (4/A)k ,
n
from which the main result of our theorem follows. There are of course, several technical details to be considered and the reader can refer to [4] for them. But we hope
this discussion highlights the essential strategy.
We remark that to study c(n), we first apply Theorem 3 to derive the asymptotics
of Fourier coefficients of 1/ which is essentially 1/24 . We then apply Theorem 5
to E 43 / which is the j -function. From this, Theorem 2 is easily deduced.
5. Concluding remarks
Clearly, the methods outlined in this paper have potential applications to other problems, most notably to those where the cumbersome circle method had been previously
applied. The study of the asymptotics of the partition function is nearly a 100 years
old and as there is an extensive literature, one can re-examine much of the literature
to see where our methods are applicable.
For instance, in 1954, Meinardus [14] obtained using the saddle point method the
asymptotic formula for the general coefficient of the power series expansion of general
products of the form
(1 q n )an ,
n=1
278
M. Ram Murty
where an are real and non-negative numbers. He obtained a general asymptotic formula of the type
Cn exp(n /+1 (1 + 1/)( A( + 1) ( + 1))1/(+1) )
where is the abscissa of convergence of the Dirichlet series
an
n=1
ns
A is its residue at s = . C, are explicit constants and denotes the Riemann zeta
function. As an application of his method, he derives the asymptotic formula for the
generalized partition function pa,b (n), whose generating function is given by
pa,b (n)q =
n
n=0
(1 q an+b )1 .
n=1
Ae
2n/3
,
n
but he could not determine the constant A by his method. This was done later by
Newman [18]. With respect to the elementary study of pa,b (n), there are several readable accounts, a recent one being the article [11]. It would be interesting to re-examine
all of these partition questions in the light of the theory of weak Maass forms.
Finally, this being a paper to commemorate the legacy of Srinivasa Ramanujan,
it seems fitting to look back from our vantage point of the 21st century and see his
contributions in the mathematical landscape of the 20th century. Certainly, the remarkable feature of his work is its interconnectedness with other parts of mathematics and
other branches of science, most notably physics. Indeed, the relationship of his work to
Lie groups and representation theory is illuminating and this is highlighted through the
Langlands program (see for example, [16] as well as [17]). A further avenue to explore
in this context are the recent works connecting the Rogers-Ramanujan identities with
representation theory of Kac-Moody Lie algebras (see [12]). Surely, what now we
see is a shadow of what must come.
Acknowledgements
I would like to thank the organizers of the Legacy of Srinivasa Ramanujan
conference held at the University of Delhi from December 17 to 22, 2012 for their
279
kind invitation and hospitality. I also thank Michael Dewar, Sanoli Gun, Kumar Murty,
Ken Ono, Mike Roth, Purusottam Rath and the referee for their comments on an
earlier draft of this paper.
References
[1] J. H. Bruinier and K. Ono, Algebraic formulas for the coefficients of half-integral
weight forms, to appear in Advances in Mathematics.
[2] M. Demazure, Identites de MacDonald, Seminaire Bourbaki, 197576, no. 483,
191201.
[3] M. Dewar and M. Ram Murty, A derivation of the Hardy-Ramanujan formula from
an arithmetic formula, Proceedings of the American Math. Society, 141 (2013)
19031911.
[4] M. Dewar and M. Ram Murty, An asymptotic formula for the coefficients of j (z),
International Journal of Number Theory, 9(3) (2013) 112.
[5] F. J. Dyson, Missed Opportunities, Bulletin of the American Math. Society, 78 (1972)
635652.
[6] P. Erdos, On an elementary proof of some asymptotic formulas in the theory of partitions, Annals of Math., 43 (1942) 437450.
[7] W. Fulton and J. Harris, Representation Theory, A First Course, Springer (1991).
[8] B. Gross, W. Kohnen and D. Zagier, Heegner points and derivatives of L-series, II,
Math. Annalen, 278 (1987) 497562.
[9] G. H. Hardy and S. Ramanujan, Asymptotic formulae in combinatory analysis, Proc.
London Math. Soc., 17(2) (1918) 75115.
[10] A. E. Ingham, A Tauberian theorem for partitions, Annals of Math., 42(2) (1941)
10751090.
[11] D. M. Kane, An elementary derivation of the asymptotics of partition functions,
Ramanujan Journal, 11 (2006) 4966.
[12] J. Lepowsky and R. L. Wilson, A new family of algebras underlying the RogersRamanujan identities and generalizations, Proc. Nat. Acad. Sci. U.S.A., 78 (1981)
no. 12, part 1, 72547258.
[13] I. G. MacDonald, Affine root systems and Dedekinds -function, Inventiones Math.,
15 (1972) 91143.
[14] G. Meinardus, Asymptotische Aussagen u ber Partitionen, Math. Z., 59 (1954) 388398.
[15] A. Milas, Virasoro algebra, Dedekind -function and specialized MacDonald identities,
Transformation Groups, 9(3) (2004) 273288.
[16] V. Kumar Murty, Ramanujan and Harish-Chandra, Math. Intelligencer, 15 (1993) no. 2,
3339.
[17] M. Ram Murty and V. Kumar Murty, The Mathematical Legacy of Srinivasa
Ramanujan, Springer, New Delhi (2013).
[18] D. Newman, The evaluation of the constant in the formula for the number of partitions
of n, American Journal of Mathematics, 73 (1951) 599601.