ADVANCES IN MATHEMATICS 41, 173-185 (1981)

Ramanujan’s “Lost” Notebook.

II. @-Function Expansions

GEORGE E. ANDREWS*

Pennsylvania State University,

University Park, Pennsylvania 16802

1. INTRODUC~~N

The idea of expanding an arbitrary function in an infinite series whose

terms belong to a specified family is an ancient technique of analysis. First
came power series expansions. Subsequently Fourier series were developed,
and numerous other expansions have played important roles in mathematics
generally.
The idea in this paper is a very natural one. We consider four families of
trigonometric polynomials:

@,;,(z, q) = 2q114sin z fi (1 - fj2n) fi (1 - 2q2m cos 2Z + q4Y9 (l-1)

n=l m=l

@,;,(z, q) = 2qv4 cos z fi (1 - q2”) fi (1 + 2qzm cos 22 + q49 U-2)

n=1 m=l

(introduced by Watson [lo; p. 671)

s,;,(z, q) = fi (1 - q2R) fi (1 + a2rn-l cos 2z + q4m-2), (1.3)

It=1 In=1
and

s4;,(z, q) = nfll (1 - 42”) mfil (1 - 24J2m-1cos 22 + q4m-2)*

(1.4)

These trigonometric polynomials are partial products for the four classical 6-
functions [ 11; Chap. XXI] :

* Partially supported by National Science Foundation Grant MCS-75-19162.

173
OOOl-8708/81/080173-13$05.00/O
Copyright 0 1981 by Academic Press, Inc.
All rights of reproduction in any form reserved.
174 [Link]

6,(z, q) = 2 f (-I)” q(n+1’2)2sin(2n + 1)z

n=o

= -i nx, (-1)” q(fl+ V*12e(*n+ I)iz

= ;+z4;N(Z,
91, (l-5)
6,(z, q) = 2 F q(n+1’*)2cos(2n + 1)z
n=O

=fq ‘?I+l’*)*e(*n+l)iZ = tz 62:N(z, q), (1.6)

n=--00

6,cz, q) = 1 + 2 2 qn*cos 2~2~= F q”2e2niz

n=1 n=--00
= lim @3;N(z,4), (1.7)
N-m

6,(z, q) = 1 + 2 5 (-1)” qn2cos 2nz

n=1

= 5 (-1)” qn2e2niz
= 1% 6,:,(z, 9) W3)
n=--00

We now ask for the coefficients T,(i, j; q, ql) in

A priori it is not clear that such expansions exist in general. Indeed in some
of the cases described by Ramanujan, we have divergent expansions for
which we must specify summability conventions.
Ramanujan’s wonderful observation is that in a number of instances these
coefficients have very elegant closed forms. We summarize his assertionsas:

THEOREM 1.
26

Lww13)= (1 -q2)(1 -4q,) .,. (1 -q4”) ; (l.W,

qn2 n:=,< 1 + q2m+ ‘)
TJ4,4;q,q2)= (lvq)(lmq3) . . . (1-q*“-~)(1-q4)(1-q*) .. . (l-qy ;

(l-11),
 RAMANUJAN’S “LOST” NOTEBOOK. II 175

2(-l)” c;& q++”

Tn(394;qyq)=* (1-q4)(1-q*)...(1-q4”)

= 2(-l)” n;=i(l - q4m+4”)

(1*W,
HZ=,(l -q4m-2) *

(The * in (1.12), is explained in the last paragraph of Section 2.)

We shall deduce Theorem 1 from a new general basic hypergeometric
series transformation given in Section 2. Ramanujan derived a large number
of q-series summations from Theorem 1. For example (stated again as (3.3)):
a, q*n*(l + q + q2)(1 + q3 + q6) *** (1 + q2”-l + q4n--2)
‘+CPI=1 (1 -q2)(1 -q4) **a (1 -q4”)

= 2 (-l)nq3+-2n mfl, (1,“;;” .

I II=--oO I
Although results like (1.13) look cumbersome at first, they have rather
startling implications for the theory of partitions. In this case, one may
deduce from (1.13):
THEOREM 2. Let p,(n) denote the number of partitions of n subject to the
following three conditions: (i) no part appears more than twice; (ii) no odd
part exceeds the number of even parts; (iii) among the even parts (arranged
in nonincreasing size) only the second, fourth, sixth, etc., may be subse-
quently repeated. Let p,(n) denote the number of partitions of n into parts
= f2, k3, k4, *:8 (mod 18). Then pi(n) =p,(n) for all n.
As an example of Theorem 2 we note that ~~(10) = 7 since it enumerates
the partitions 10, 8 + 2, 8 + 1 + 1, 6 + 4, 6 + 2 + 2, 6 + 2 + 1 + 1,4 + 2 +
2 + 1 + 1; while ~~(10) = 7 since it enumerates 10, 8 + 2, 4 + 4 + 2,
4 + 3 + 3, 4 + 2 + 2 + 2, 3 + 3 + 2 + 2, 2 + 2 + 2 + 2 + 2. We shall prove
Ramanujan’s corollaries of Theorem 1 in Section 3, and in Section 4 we shall
prove Theorem 2.
Ramanujan also found an analog of Theorem 1 for L. J. Rogers’s false 6-
series. We shall present these results and sketch their proofs in Section 5. We
conclude in Section 6 with a look at some of the open questions.

2. THE &EXPANSIONS

To facilitate our work we shall use the following standard notation [S;
Sect. 3.2.11:

(2.1)
176 GEORGE E. ANDREWS

note that for positive integral n

(a),=(l-a)(l-aq)*~* (1--q”-‘), (2.2)

(a), = fi (1 - aq”). (2.3)
m=O

LEMMA 1.

(--as47”(- $ ;92)n(b;q2Mc;
q2)n
(2) n
?L
n=o ((12; q2)2”

Proof

(-aq;q2),(- $ ;
F
?I=0

n2+2”a-“b-“c-“(-aq-2”+ l; q2)2”(b; q2)“(c; q2)n

Sq
“YO (q2; q2)zn

--n+mqn*+2n-2nm+d(b; q2),(c; q2), b-"y"

f? q a
=
n=O rzo (q2; q2Mq2; q2)2n-m

(by [l; p. 36, Eq. (3.3.6)])

=
N=-cc

(q2; q2M2; 42L2N

(by [ 1; p. 20, Cor. 2.41)

 RAMANUJAN’S “LOST” NOTEBOOK. II 177

(q2; q2)w (jg ; 91)

co
2 N

(a” + a-“) fl*(b; q’)&; q’)N 2

( 1
+z
N=l (4’; q2)2N

(bqZN; 42)n@q2N; 4% ($”

6
n=o (q2; q2)“(dN+ 2; q2)n

($t)*(~~ll2)~

= (q2;q2)* (g x2)
co

+ Ng, (a” + a-“) dJ’(b; q2),& q’)N (;)”

ZN+2),-1; q2)w(qZN+2C-1; q2)m

x (q

(q2; q2L (g ; q2)*

(by [l, p. 20, Cor. 2.41)

(~;42.)*($2?2)*
= (q2;q2)w($ ;41)*

607/d l/2-5
178 GEORGE [Link]

Theorem 1 now follows directly from Lemma 1. Equation (1.10) is

derived from (2.4) by letting a = -ezir, b -+ co and c + co. Equation (1.11)
follows from (2.4) if we set a = -ezir, b = -q and let c + co.,The reason for
the asterisk on the equals sign in (1.12) is that this expression for 7’d3,4;
q, q) makes (1.9) a divergent series. We therefore interpret (1.12) thusly: set
a = ezir, b = -c in Lemma 1; this provides an apparent singularity of the left
side at c = q. We define the value of the left side to be the limit as c + q-.
Hence we may with the above convention set -b = c = q in Lemma 1. This
proves (1.12) once we make the observation that the right side is

(q2;q4L2 F (- 1 )N qN2e2irN

m*;q2)(-q2;
q2jmNc-lcc
= Cg= -,(-l)N qNze2izN
(by [L P. 231).
2 c,“=oqn’+n

3. COROLLARIES OF THE ~-EXPANSIONS

As well as giving equivalent results to those in Theorem 1, Ramanujan

lists the following six identities that follow from Theorem 1. Identity (3.3) is
a restatement of (1.13).

q qn2b& s’>, = CLcx q3n2-n

-p’O(-q)(n’+n)~* ;
n=O @?I*”
2;~ _ m q(3n2-n)/*
p'. qwtnM2 ;

(3*4),

= c:=-,(-l)” q4”2+2” fi (1 + q12n+6). (3 5)

C,“=-,(-l)“q2”2+n n=O ’ * R
JJ:= -a, q4n2+*.
fi (1 -q12”f6), (3.6),
~,“=-,(-l)“q2”2+” n=o

To obtain identity (3.1), we take the case of (1.9) with i =j= 4, q1 = q3,
z = z/4, then utilizing (1.10) replace q by q”* and observe
 RAMANUJAN’S “LOST” NOTEBOOK. II 179

1 + 2 c;Cp(-l)” q6”* (d? 46)*

rIE1<1 - 4”) = (4; 46M&J
= et; 46)co(--Qz; 4w--4; 4*hm
w; ~6)m(--4)m(q)m
= (q6; 46Lo(--42~ 46)mt-44; q6)a,
G?*; 4*LA--c 4*LJ
c;=+ q”on-l)
= ~~~o(-q)a(n+lV2 *

To obtain (3.2), we again apply (1.10) to (1.9) with i=j=4, q, =q3,

z = 7112,and then observe

1 + 2 2:=,,(-1)R q3n2 (s3; q3Lo

~nri,<l - q2”) = (-13; q3),(q2; 4*Lc
(43; q3M-daJ
= (-4;; 43M--4LJ(42; 4*L3
= (q3; q3M--4; 43L3(--42; 43)cc
@I*; 4*M(4; 4*LJ
= CF=-, qn(3n-‘y2
y. qn(n+ lb/2 *

To obtain (3.3), we also apply (1.10) to (1.9) with i=j=4, q1 =q3,

z = 7r/3, and we note

Next we apply (1.11) to (1.9) with i =j = 4, q1 = q*, z = n/3, noting in this

instance
180 [Link]

Cn”-m(-l)n F @I*;4*jco (4;q2jm

cjy& q(n2+n)‘* = (-q*; q*)m * (42; q2),

(4)rx 1 + cn”=i(-l)” q”(3”--1)‘2(1+ q”)

= (q”; q4), = (q4; q4)m

and we thus obtain (3.4).

To obtain (3.5) we apply (1.11) to (1.9) with i =j = 4, q1 = q*, z = 7r/3,
and we note

= ~n”-m(-1)“q4n2+2” (-q6. q,2)

c,“-,(-1)” q*n2+” ’ ma

Finally for (3.6) we take i =j= 4, q1 =q* in (1.9) and apply (1.11) with
z = 5n/6. The right side of (3.6) is derived in the same manner as the right
side of (3.5).

4. CONSEQUENCES FOR PARTITIONS

In this section we content ourselves with a proof of Theorem 2 as stated in

Section 1. We begin with the observation that

is the generating function for partitions b, + b, + . . . + b, with either 2n or

2n - 1 parts wherein b, > b, > b, > b, > ..a (by [5], see also [3; p. 81).
Hence

(4.1)
 RAMANUJAN’S “LOST” NOTEBOOK. II 181

is the generating function for partitions b; + b; + .. . + bj with either 2n or

2n - 1 parts all even wherein b, > b, > b, > b, > . . . . Now it is clear tht

fi (1 + q”-’ + q”-2) (4.2)

j=l

is the generating function for partitions into odd parts each <2n - 1 and
each appearing at most twice. Consequently if we multiply together the
expressions in (4.1) and (4.2) we obtain the generating function for partitions
with 2n or 2n - 1 even parts such that (i) no part appears more than twice;
(ii) no odd part exceeds the number of even parts; (iii) among even parts
(arranged in nonincreasing size) only the second, fourth, sixth, etc., may be
subsequently repeated. Hence summing over all n we find

l&(1 +
c
O”

n=o
q2”2

(q2;
q*j-’

42)2n
+ q”-2)
= 5
II=1
h(n)q” (4.3)

Turning to the right side of (1.13), we find

= (q6;q6L&&Q6M?5;46LJ(4g;4%
(da3
= (q6;q18)m(q12;
418)m(418;
d8)&?; 418)oo(47;
PMd3; P)m
x (q5;4’8)co(411;
41*)oo(417;
418)03(4g;
d8LJ
(da2
1
= (q2; q18)m(q3; q1*)oo(q4; q18)m(q*; q18)03(q10; 4%

x (q14; q18)03(q15; 41*)m(416; dS)oo

( )

=z.p2(n)
4”.
In light of (1.13), we may identify (4.3) and (4.4). Therefore p,(n) = p2(n)
for all n as asserted in Theorem 2.

5. THE FALSE &EXPANSIONS

Rogers introduced false &functions [7]. He found them arising in some

very elegant q-series identities, and these results have subsequently been
exploited to obtain numerous partition theorems [2; Chap. 3B]. Their
182 [Link]

analytic behavior while more complex than that of ordinary &functions has
been analyzed masterfully by Mordell [6; Sect. 41. The results Ramanujan
stated in the “Lost” Notebook for false &functions are (with one exception)
special cases of the following result.

LEMMA 2.

(-a),+ ,(-qlaMMc)n g,”

;
n=O (4)Zntl
(b)N(C)N (g)”
cc 7 +N+l + a-N) q(N2+NU2
=
q2 N’=O
(9) (-)
lx2 bc o.

Proof

(-a),, 1 (- t) (b)“(C), ($
?
n=o on 2n+l -
p
(~2+~)‘2+2nb-nC-“a-“(-uq-n)2”+ ,(b),(c),
=
n=O (4) 2n+1
-n+m (n2tn)/z+zn nm+(“I-“)/2(b)n(C)n
b-“c-“q-
=\“‘-?” q
,fzo zo Mk?) Zntl--m

(by [l, p. 36, Eq. (3.3.6)1)

= $ aNq(N2-N)/Z $
(b),(c),
(;) ’
N=Eco ,:O (d, tNh)n+ I-N

(b)n+N--1(C)ntN-,
(;)n+N-’
= ‘$

NY1
caN + .l -N) qW-N)/2 5

n=O (q)n(dn t ZN- 1

@IN--I(C),+-1 (;)“-’
= F @N + a1 -N) qW-N)/2

N=l (q)ZN- 1

b“+lL(d’-l)n ($)”

mk72N)”
 RAMANUJAN’S “LOST” NOTEBOOK. II 183

= $, (a” + d-N) q(@-N)‘2

k)2N- 1

(by [ 1, p. 20, Cor. 2.41)

tb)NtC)N (g) N
= q2 c (#+I + u-N)q(N2+N~2 ,
N=O
(4) (-)bc
N

which is the desired result. 1

The first three of the following identities constitute counterparts of
Theorem 1. The remainder is a false &series counterpart of the results in
Section 3.

-f .(-1)” @,,,(z; 4)

n=O (q2; q4)n+1

= * 1 fi (1 - 92”) 1 5 (-1)” q(n+U2)* cos(2n + l)z, (5.2,,

n=l n=O

q z. ‘-y12”:;%.::‘:; 4)

= 1 ZI (1 - 42”) 1 go (-1)” q2(n+1’2)* cos(2n + l)z, (5.3),

If0;.;$yf = set z n=O

5 (-1)” qn*+n cos(2n + l)z, (54)R

5 (-1)” q”(“+yq3; 43).

= go q!Jn’+Jn - ZI q9+3n. (5.5),
n=O (4) 2n+l

We note that the left side of (5.2) is a divergent series. The mitigation of
this problem is precisely that used to treat (1.12) at the end of Section 2. To
obtain (5.2) appropriately interpreted we first replace q by q2 in Lemma 2
then set u = e-“’ and let b = -c + q2; the result is multiplied by qu4eiz l-IF=,
(1 - 42”).
To obtain (5.3) we replace q by q2 in Lemma 2 and then let a = e”‘,
b = q2, c + +co. The result is multiplied by 1/2qV2e” l-I:= r(l - q2”).
184 [Link]

Identity (5.4) appears not to be directly deducible from Lemma 2.

However, its proof precisely parallels that of Lemma 2. Indeed if we multiply
both sides of (5.4), by 2e@ cos z nz= i(1 - q’“)- ’ and take the coefficient
of eziNz on the left side (for N > 0) we obtain

- 42N+1(92;q2)2n \

(q2;q2)n+N+ 1(q2; q2)n-N-1 I

"+N(q)2n+2N
=(-l>"(l -q2N+')qN2 j$ (q2,qq2) (q2.q2)
Y n 7 nt2Nt1

= (-1)N qN2+N

(by the q-analog of Gauss’s theorem [ 1, p. 20, Cor. 2.41). Since cos z is an
even function we see that if we repeat the previous process first replacing z
by -z and then multiplying by eeiZ we obtain (-l)N 6““” as the coefficient
of eKcNt ‘jZir also for N > 0. This establishes (5.4).
Identity (5.5) follows from (5.3) with z = 5n/6 upon the observation that

(- 1)” cos 32, + 1))

1 if n = 0 (mod 3),
cos(5~/6) =
= 0 if n= 1 (mod3),

=- 1 if n = 2 (mod 3),

6. CONCLUSION

Our object was to present one aspect of Ramanujan’s “Lost” Notebook by

sticking fairly closely to his actual assertions. The presentation was also
designed to indicate the possible extent of further investigations. Indeed the
generality of (1.9) calls attention to the fact that Theorem 1 treats only three
possible cases of a triply infinite family of results.
We should note that (3.1) follows also from Watson’s q-analog of
Whipple’s theorem [9] and (3.3) is a confluent specialization of the q-analog
of Gauss’s theorem [ 1, p. 20, Cor. 2.41. Furthermore (5.5) has been
attributed to F. J. Dyson by Bailey [4, p. 434, Eq. (E2)]. Thus while most of
the results given have not appeared before, we nonetheless see points of
contact with classical results. Our main tools, Lemmas 1 and 2, do not
 RAMANUJAN’S “LOST” NOTEBOOK. II 185

appear to be special cases of anything in the literature. In any event,

placement of Lemmas 1 and 2 in the standard hierarchy of basic
hypergeometric series identities would surely serve to illuminate the possible
extensions of (1.9) and to suggest specializations of interest in number
theory.

REFERENCES

1. G. E. ANDI~EWS, The Theory of Partitions, Encyclopedia of Mathematics and Its

Applications, Vol. 2, Addison-Wesley, Reading, Mass., 1976.
2. G. E. ANDREWS, “Partitions: Yesterday and Today,” New Zealand Math. Sot.,
Wellington, 1979.
3. G. E. ANDREWS AND R. ASKEY, Enumeration of partitions: The role of Eulerian series
and q-orthogonal polynomials, in “Higher Combinatorics” (M. Aigner, Ed.), pp. 3-26,
Reidel, Dordrecht/Boston, 1977.
4. W. N. BAILEY, Some identities in combinatory analysis, Proc. London Math. Sot. (2) 49
(1947), 421-435.
5. B. GORDON, Some continued fractions of the Rogers-Ramanujan type, Duke Math. J. 32
(1965), 74 l-748.
6. L. J. MORDELL, The definite integral j:, eoxztbx &/(e” + d) and the analytic theory of
numbers, Acta Math. (Stockholm) 61 (1933), 323-360.
7. L. J. ROGERS, On two theorems of combinatory analysis and some allied identities, Proc.
London Math. Sot. (2) 16 (1917), 315-336.
8. L. J. SLATER, “Generalized Hypergeometric Functions,” Cambridge Univ. Press,
Cambridge, 1966.
9. G. N. WATSON, A new proof of the Rogers-Ramanujan identities, J. London Math. Sot.
4 (1929), 4-9.
10. G. N. WATSON, The final problem: An account of the mock theta functions, J. London
Math. Sot. 11 (1936), 55-80.
11. E. T. WHIITAKER AND G. N. WATSON, “A Course of Modem Analysis,” 4th ed.,
Cambridge Univ. Press, Cambridge, 1927.