HOW TO DISCOVER THE ROGERS–RAMUNUJAN IDENTITIES

arXiv:1609.05325v1 [math.CA] 17 Sep 2016

GAURAV BHATNAGAR

Abstract. We examine a method to conjecture two very famous identities

that were conjectured by Ramanujan, and later found to be known to Rogers.

1. Introduction
The so-called Rogers–Ramunujan identities were sent by Ramanujan to Hardy
nearly 100 years ago. In the next few years, the identities were circulated amongst
mathematicians, but nobody, including Ramanujan, was able to prove them. Then
one day, while riffling through old back copies of a journal, Ramanujan himself
discovered them in an obscure paper written in 1894 by the English mathematician
Rogers. This spurred both Rogers and Ramanujan to provide simpler proofs of the
identities, that were published in 1919.
The identities are:
q q4 q9
1+ + + + ···
(1 − q) (1 − q)(1 − q 2 ) (1 − q)(1 − q 2 )(1 − q 3 )
1 1
= × ;
(1 − q)(1 − q 6 )(1 − q 11 )(1 − q 16 ) · · · (1 − q 4 )(1 − q 9 )(1 − q 14 ) · · ·
(1)

and,
q2 q6 q 12
1+ + + + ···
(1 − q) (1 − q)(1 − q 2 ) (1 − q)(1 − q 2 )(1 − q 3 )
1 1
= × .
(1 − q 2 )(1 − q 7 )(1 − q 12 )(1 − q 17 ) · · · (1 − q 3 )(1 − q 8 )(1 − q 13 ) · · ·
(2)

About these Hardy [6, p. xxxiv] remarked:

It would be difficult to find more beautiful formulae than the “Rogers–
Ramunujan” identities . . .
The purpose of this article is to introduce you to the Rogers–Ramunujan identi-
ties, by discussing an approach to discover them. When you see that they appear
from a very simple generalization of the simplest possible infinite continued frac-
tion, that in turn is related to the celebrated Fibonacci sequence, perhaps you may
begin to agree with Hardy’s opinion of these formulas.

Date: January 18, 2012.

1
2 GAURAV BHATNAGAR

2. On infinite series and products

Examine the first of the Rogers–Ramunujan identities (1). On the left hand side,
there is an infinite sum of terms such as
2
qk
.
(1 − q)(1 − q 2 ) · · · (1 − q k )
On the other side, there is an infinite product, that has terms such as
1
1 − qm
where m is either of the form 5k + 1 or 5k + 4, for k = 0, 1, 2, . . . . Our first goal is
to understand how to assign meaning to such infinite sums and products.
To that end, let us examine the simplest such identity, namely the geometric
series:
1
1 + q + q2 + q3 + · · · = , where |q| < 1 (3)
1−q
Here the left hand side is an infinite sum of terms such as q k , and the right hand
side consists of the single factor of the form 1/(1 − q m ), with m = 1.
Recall the proof of this formula. We first prove the finite version, the sum of the
first n terms of a geometric progression:
1 − qn
1 + q + q 2 + · · · + q n−1 = .
1−q
Let
Sn = 1 + q + q 2 + · · · + q n−1 .
Multiply both sides by (1 − q) to obtain:
Sn (1 − q) = (1 − q) + (q − q 2 ) + (q 2 − q 3 ) + · · · + (q n−1 − q n )
= 1 − qn
1 − qn
=⇒ Sn = .
1−q
To go to the infinite series, we first write the sum Sn as
1 qn
Sn = −
1−q 1−q
and then take the limit as n → ∞. Note that if q is a real number between −1 and
1, then q n → 0 as n → ∞. Thus we can say that
1
lim Sn = .
n→∞ 1−q
For those of you who are familiar with complex numbers, we may take q to be a
complex number that satisfies |q| < 1. In either case, we can say that the infinite
series on the LHS of (3) converges to the expression on the RHS. This completes
the proof of the formula for the sum of the geometric series.
To summarize, one way to make sense of the infinite series is to first consider
finite sums and then take limits. There is a similar approach for infinite products
which we will not get into.
 HOW TO DISCOVER THE ROGERS–RAMUNUJAN IDENTITIES 3

There is another approach to infinite series and products that can help us avoid
questions of convergence. In this formal approach, we consider a formal power
series, an expression of the form
X∞
ak q k := a0 + a1 q + a2 q 2 + a3 q 3 + · · · ,
k=0
where we have used the Σ notation for writing series in shorthand. We say that
the coefficient of q k is ak . We regard this infinite series as an ‘infinite degree
polynomial’ which we can add and subtract (and even multiply) term-wise, just by
extending the rules of algebra of polynomials. Two series are considered equal, if
for each k, the coefficient of q k is the same in both the series. Here we regard q as
an indeterminate, and use it as a symbol without giving it any value!
In the formal approach, we can use the geometric series (3) to expand each term
of the form 1/(1 − q m ) into the infinite series
1 + q m + q 2m + · · ·
and then multiply them together, to write sums and products as a formal power
series. Of course, we can only multiply a finite number of terms at one time.
For example, the first few terms of the LHS of the first Rogers–Ramunujan
identity (1) can be written as
1 + q(1 + q + q 2 + q 3 + q 4 + · · · ) + q 4 (1 + q + q 2 + · · · )(1 + q 2 + q 4 + q 6 + · · · ) + · · · .
We can see that the first few terms are
1 + q + q 2 + q 3 + 2q 4 + · · · .
Try this approach to find the first few terms of the RHS of the identity (1) and check
that they match. (I heartily recommend that you calculate more terms on both
sides, using a computer algebra package such as Maxima, Maple or Mathematica.
Maxima is freely available on the Internet. Download and use!)
To summarize, in the formal approach to understand infinite identities, we regard
both sides to be formal power series, without regard to whether the series converge.
We regard the identity to be true if for each k, the coefficients of q k match on both
sides. But to be able to calculate any coefficient, the computation should involve
only a finite number of algebraic operations.
For the purpose of this article, we ignore issues of convergence altogether, and
only consider these identities at a formal level.

3. The Simplest Continued Fraction

There is one more interesting infinite object that we will briefly examine before
we begin our study of Ramanujan’s formulas. Consider the continued fraction
1
1+ .
1
1+
1
1+
1 + ···
To analyze this continued fraction, we truncate the continued fraction at various
places and compute the values of the resulting (finite) continued fractions. These
truncated fractions are called convergents.
1= 1 = 11 ;
4 GAURAV BHATNAGAR

1
1+ = 2 = 21 ;
1
1 1
1+ = 1+ 2 = 23 ;
1
1+
1

1 1
1+ = 1+ 3/2 = 53 ;
1
1+
1
1+
1

1 1
1+ = 1+ 5/3 = 85 .
1
1+
1
1+
1
1+
1
The convergents are:
1 2 3 5 8
, , , , ,....
1 1 2 3 5
Note that to calculate a particular fraction, we use the value of the previous con-
vergent, so the nth convergent wn satisfies the following recurrence relation:
1
wn = 1 + .
wn−1
Further, note that the denominator of the nth convergent is the numerator of the
previous convergent. This suggests the substitution
Fn
wn = .
Fn−1
Substituting this value in the recurrence relation yields:
Fn 1
=1+ ,
Fn−1 Fn−1 /Fn−2
or
Fn = Fn−1 + Fn−2
after clearing out the denominators.
Let F0 = 1 and F1 = 1. Then we generate the rest of the sequence using this
rule to obtain
1, 1, 2, 3, 5, 8, 13, 21, . . .
The convergents are easy to determine too, by taking ratios of successive numbers
in this sequence. This sequence, known as the Fibonacci Sequence, starred
prominently in the novel (and movie) The da Vinci Code.
I suggest that you compute ratios Fn /Fn−1 for n = 1, 2, 3, . . . and convince
yourself that the ratios converge. Use a computer. Can you figure out the exact
value of the limit? The limit (to which the continued fraction converges) is called
the golden mean.
If you find continued fractions attractive, then you will love this Olds [7] book.
Continued fractions have appeared earlier in Resonance: Shirali [8] gives a clear
 HOW TO DISCOVER THE ROGERS–RAMUNUJAN IDENTITIES 5

introduction, and gives examples of such fractions related to the number e; Sury
[9] explains their appearance in the context of special functions.

4. The Rogers–Ramanujan Continued Fraction

We are now ready to dive into Ramanujan’s world. The first step is to generalize
the continued fraction by adding an additional parameter q to it.
The q-generalization of
1 + 1 + 1 + ··· + 1 = n
is the familiar sum of the Geometric Progression
1 − qn
1 + q + q 2 + · · · + q n−1 = .
1−q
Similarly, one would like to generalize the continued fraction
1
1+
1
1+
1
1+
1+ ···
to
q
1+ . (4)
q2
1+
q3
1+
1 + ···
To gain some intuition, truncate the continued fraction at different places and
evaluate the resulting (finite) fractions.
1 = 1;
q
1+ = 1 + q;
1
q q 1 + q + q2
1+ = 1+ = ;
q2 1 + q2 1 + q2
1+
1

q q(1 + q 3 ) 1 + q + q2 + q3 + q4
1+ 2
= 1+ 2 3
= .
q 1+q +q 1 + q2 + q3
1+
q3
1+
1
At each step, we would like to use the previous calculation, just as in §3. For
example, to compute
q
1+
q2
1+
q3
1+
1
we would like to use
q
1+ .
q2
1+
1
6 GAURAV BHATNAGAR

But the powers of q are not quite right. However, if we have an additional parameter
z
zq
1+ ,
zq 2
1+
1
then we can adjust the powers of q, by replacing z by zq.
So consider:
zq
c(z, q) = 1 + . (5)
zq 2
1+
zq 3
1+
1 + ···
That gives us a recurrence relation:
zq
c(z, q) = 1 + . (6)
c(zq, q)
The continued fraction c(z, q) is called the Rogers–Ramunujan continued fraction.

5. The sum side

The sum sides of the Rogers–Ramunujan identities appear as we solve the recur-
rence relation (6).
To gain further insight, we again turn to computation. The first few fractions
are as follows:
c0 (z, q) = 1 = 1;
zq
c1 (z, q) = 1+ = 1 + zq;
1
zq zq 1 + zq + zq 2
c2 (z, q) = 1+ = 1 + = ;
zq 2 1 + zq 2 1 + zq 2
1+
1
zq zq(1 + zq 3 )
c3 (z, q) = 1+ = 1 +
zq 2 1 + zq 2 + zq 3
1+ 3
zq
1+
1
1 + zq + zq 2 + zq 3 + z 2 q 4
= ;
1 + zq 2 + zq 3
zq zq(1 + zq 3 + zq 4 )
c4 (z, q) = 1+ = 1 +
zq 2 1 + zq 2 + zq 3 + zq 4 + z 2 q 6
1+ 3
zq
1+
zq 4
1+
1
1 + zq + zq 2 + zq 3 + zq 4 + z 2 q 4 + z 2 q 5 + z 2 q 6
= .
1 + zq 2 + zq 3 + zq 4 + z 2 q 6
The pattern is clear.
Let Hn (z, q) be the numerator of cn (z, q). Its denominator is Hn−1 (zq, q). In-
deed, for n = 2, 3, 4, we have
Hn (z, q) zqHn−2 (zq 2 , q)
cn (z, q) = =1+ .
Hn−1 (zq, q) Hn−1 (zq, q)
 HOW TO DISCOVER THE ROGERS–RAMUNUJAN IDENTITIES 7

This suggests the substitution:

H(z, q)
c(z, q) = .
H(zq, q)
Substituting for c(z, q) in the recurrence relation (6) yields:
H(z, q) = H(zq, q) + zqH(zq 2, q). (7)
So far, our calculations have been analogous to our work in §3. To solve this
recurrence relation we require an additional trick. We assume that the solution has
the power series expansion:
X
∞
H(z, q) = ak z k .
k=0

If the solution is really of this form, then it must satisfy the recurrence relation.
We will use this idea to actually compute the coefficients ak of the solution.
By substituting the power series expansion in (7), we obtain
X
∞ X
∞ X
∞
ak z k = ak q k z k + ak q 2k+1 z k+1 .
k=0 k=0 k=0
k
On comparing coefficients of z on both sides, we get, for k = 1, 2, 3, . . . :
q 2k−1
ak = ak q k + ak−1 q 2k−1 , or ak = ak−1 .
1 − qk
On iteration, this yields:
2
qk
ak = a0 .
(1 − q)(1 − q 2 ) · · · (1 − q k )
Take a0 = 1 to find that
X
∞
qk
2
X qk ∞ 2
k
H(z, q) = z = zk,
(1 − q)(1 − q 2 ) · · · (1 − q k ) (q; q)k
k=0 k=0

where to shorten our displays, we have used the notation:

(q; q)k := (1 − q)(1 − q 2 ) · · · (1 − q k ).
We are interested in (4), which is c(1, q):
H(1, q)
c(1, q) = ;
H(q, q)
or, in other words,
X
∞
qk
2

q (q; q)k
k=0
1+ = . (8)
q2 X∞ 2
q k +k
1+
q3 (q; q)k
1+ k=0
1 + ···
Thus, the Rogers–Ramanujan continued fraction is the ratio of the sums:
X
∞
qk
2
q q4 q9
=1+ + + + ···
(q; q)k (1 − q) (1 − q)(1 − q 2 ) (1 − q)(1 − q 2 )(1 − q 3 )
k=0
8 GAURAV BHATNAGAR

and
X∞ 2
q k +k q2 q6 q 12
=1+ + + + ··· .
(q; q)k (1 − q) (1 − q)(1 − q 2 ) (1 − q)(1 − q 2 )(1 − q 3 )
k=0

Here they are—the sum sides of the Rogers–Ramanujan identities.

6. The product side

The sum sides of the Rogers–Ramunujan identities arose from the analysis of
the Rogers–Ramanujan continued fraction. Next, we show how one can guess the
product sides. Let us first consider the LHS of the first Rogers–Ramunujan identity
(1):
q q4 q9
H(1, q) = 1 + + + + ··· .
(1 − q) (1 − q)(1 − q 2 ) (1 − q)(1 − q 2 )(1 − q 3 )
Note that, in view of (3), the sum H(1, q) is of the form
1 + q + higher powers of q.
There is a useful to trick to eliminate the smallest power of q (in this case q 1 ) on
the RHS: Multiply both sides by 1 − q.
q4 q9
H(1, q)(1 − q) = 1 + 2
+ + ···
1−q (1 − q )(1 − q 3 )
2

= 1 + q 4 + higher powers of q.
Now the smallest power on the RHS is q 4 . So multiply both sides by 1 − q 4 to
obtain:
q 9 (1 + q 2 ) q 16 (1 + q 2 )
H(1, q)(1 − q)(1 − q 4 ) = 1 + q 6 + 3
+ + ···
1−q (1 − q 3 )(1 − q 4 )
= 1 + q 6 + higher powers of q.
Next, multiply by 1 − q 6 to obtain:
q 16 (1 + q 2 )(1 + q 3 )
H(1, q)(1 − q)(1 − q 4 )(1 − q 6 ) = 1 + q 9 + q 11 + q 14 + + ···
(1 − q 4 )
= 1 + q 9 + higher powers of q.
We can continue in this fashion and hope that
H(1, q)(1 − q)(1 − q 4 )(1 − q 6 )(1 − q 9 )(1 − q 11 )(1 − q 14 )(1 − q 16 ) · · · = 1.
Note the powers 1, 4, 6, 9, 11, 14, 16. The guess is that the powers are 5m + 1 and
5m + 4, for m = 0, 1, 2, 3, . . . . Thus we have the conjecture:
X
∞
qk
2
Y∞
1
H(1, q) = = 5m+1 )(1 − q 5m+4 )
, (9)
(q; q)k m=0
(1 − q
k=0

where we have used the Π notation for products, which is analogous to the Σ
notation for sums.
The reader can check that a similar approach on H(q, q) yields
H(q, q)(1 − q 2 )(1 − q 3 )(1 − q 7 )(1 − q 8 )(1 − q 12 )(1 − q 13 )(1 − q 17 ) · · · = 1.
 HOW TO DISCOVER THE ROGERS–RAMUNUJAN IDENTITIES 9

So we can conjecture that:

X∞ 2
q k +k Y∞
1
H(q, q) = = 5m+2
. (10)
(q; q)k m=0
(1 − q )(1 − q 5m+3 )
k=0
Equations (9) and (10) are the Rogers–Ramanujan Identities.
The trick used here works in other contexts too. Try it on the Riemann zeta
function
1 1 1 1
ζ(s) = 1 + s + s + s + s + · · ·
2 3 4 5
and discover for yourself a famous identity of Euler!

7. Credits and closing remarks

We have seen an approach to discover the Rogers–Ramunujan identities. First
of all, we gave a q-generalization of the continued fraction for the golden mean
to obtain the Rogers–Ramunujan continued fraction. Next, an analysis of this
continued fraction led to the sum sides of the two Rogers–Ramunujan identities.
This part of our approach is based on remarks by Askey [3], who suggested that
Ramanujan could have discovered the Rogers–Ramanujan identities in this manner.
Our method to conjecture the product sides in §6 is the same as the one used by
Euler [5] to find a product representation of the Riemann zeta function.
Our computations do not amount to a proof of the Rogers–Ramunujan identities.
If you want a proof, look at Andrews [1] or Andrews and Baxter [2]. A longer
introduction to Ramanujan’s mathematics has been given by Berndt [4], a book I
heartily recommend to you. See Sury [10] for an informative book review of this
book.
Meanwhile, I hope this article has given you, dear reader, a small taste of Ra-
manujan’s own approach to mathematics—described by Hardy [6, p. xxxv] as fol-
lows:
He worked, far more than the majority of modern mathematicians,
by induction from numerical examples; . . . But with his memory,
his patience, and his power of calculation, he combined a power of
generalisation, a feeling of form, and a capacity for rapid modifica-
tion of his hypotheses, that were often really startling, and made
him, in his own peculiar field, without a rival on his day.

References
1. G. E. Andrews, q-Series: Their development and application in analysis, number theory,
combinatorics, physics and computer algebra, NSF CBMS Regional Conference Series 66
1986.
2. G. E. Andrews and R. J. Baxter, A motivated proof of the Rogers–Ramunujan Identities,
Amer. Math. Monthly, 96 (No. 5) (1989) 401–409.
3. R. Askey, Ramanujan and hypergeometric and basic hypergeometric series, in Proc. Ra-
manujan Symposium on Classical Analysis, N. K. Thakare (ed.), Macmillan India, Madras,
1989, 1–83; Russian translation in Usp. Math. Nauk. 45 (1) (1990), 33–76.
4. B. C. Berndt, Number Theory in the Spirit of Ramanujan, in Student Mathematical Library,
34, American Mathematical Society, 2006.
5. L. Euler, Variae observationes circa series infinitas (Various observations about infinite
series), Commentarii academiae scientiarum Petropolitanae 9, (1744) 160–188; Reprinted
in Opera Omnia Series 1, Volume 14, 217–244.
6. G. H. Hardy, P. V. Seshu Aiyar, B. M. Wilson, Collected Papers of Srinivasa Ramanujan,
Cambridge University Press, 1927; reprinted Chelsea, NY, 1962.
10 GAURAV BHATNAGAR

7. C. D. Olds, Continued Fractions, Random House, 1963.

8. S. A. Shirali, Continued fractions for e, Resonance, 5 (01) (2000), 14–28; available at
http://www.ias.ac.in/resonance/Volumes/05/01/0014-0028.pdf
9. B. Sury, Bessels contains continued fractions of progressions, Resonance, 10 (03) (2005),
80–87; available at http://www.ias.ac.in/resonance/Volumes/10/03/0080-0087.pdf
10. B. Sury, Review of [4], Current Science, 103 (07) (2012), 833–834; available at
http://www.currentscience.ac.in/Volumes/103/07/0833.pdf.

Educomp Solutions Ltd., Delhi, India

E-mail address: bhatnagarg@gmail.com