Sie sind auf Seite 1von 6

Aequationes Math.

69 (2005) 257–262 c Birkhäuser Verlag, Basel, 2005


°
0001-9054/05/030257-6
DOI 10.1007/s00010-004-2738-6 Aequationes Mathematicae

A continued fraction from Ramanujan’s lost notebook


Bruce C. Berndt1 and Geumlan Choi

Summary. In his lost notebook, Ramanujan recorded without a proof the value of an elegant
periodic continued fraction. The authors provide an elementary proof of a corrected version of
Ramanujan’s claim.

Mathematics Subject Classification (2000). 11A55.

Keywords. Periodic continued fraction, Ramanujan’s lost notebook, recurrence relations.

In the spring of 1976, George Andrews discovered a sheaf of 138 pages of Ramanu-
jan’s mathematics in the library at Trinity College, Cambridge. This manuscript
was in the possession of the English mathematician G. N. Watson at the time of
his death in 1965, and was sent to Trinity College Library on December 26, 1968
by R. A. Rankin of the University of Glasgow. Although technically neither lost
nor a notebook, in view of the fame of Ramanujan’s notebooks, it was natural
for Andrews to call his discovery Ramanujan’s lost notebook. When Ramanujan’s
lost notebook was photocopied and published in 1988, other manuscripts and frag-
ments by Ramanujan from the libraries at Cambridge and Oxford were included
in the volume [2]. The continued fraction in the title of this paper does not appear
in Ramanujan’s lost notebook, but in one of these fragments published with the
lost notebook [2, p. 341]. We state next Ramanujan’s claim about this continued
fraction.

Theorem 1 (p. 341). If



a2 + 4
µn := à √ !n à √ !n , (1)
a + a2 + 4 a − a2 + 4

2 2

1 Research partially supported by grant MDA904-00-1-0015 from the National Security

Agency.
258 B. C. Berndt and G. Choi AEM

then
 sµ 
¶2
1 µn+1 µn+1
−c + b + c+b + (−1)n µ2n+1 
2 µn µn
1 1 1 b 1 1
= , (2)
a + a + ··· + a + c + a + a + ···
1
where in each grouping, there are n fractions a.

We first remark that this entry is difficult to read. In the demoninator of µn


the “4” at the left is hardly legible, and the other “4” in the denominator is more
illegible. Second, we can easily see that (2) is false, in general. For example,
suppose that a = b = c = n = 1. Then µ1 = µ2 = 1, and (2) yields
1³ p ´ √3 1 1 1
2
−1 + 1 + (1 + 1) − 1 = = .
2 2 1 + 1 + 1 + ···
But it is well-known and √ easy to prove that the continued fraction on the right
side above has the value ( 5 − 1)/2. It is surprising that Ramanujan would have
made such a mistake.
The entry is an isolated one on page 341 of [2], and, in fact, it may be that
this entry is on a scrap of paper attached to a larger page for photocopying. The
remainder of the page is devoted to generating a family of solutions to Euler’s
Diophantine equation a3 + b3 = c3 + d3 , and nothing on adjoining pages is related
to continued fractions. Furthermore, immediately to the right of Theorem 1 are
two vertical lines drawn with a straightedge. It is possible that the entry has
been cropped, and so the entry may be incomplete, providing an explanation for
Ramanujan’s “mistake.”
We are therefore faced with the problem of finding the “correct” theorem that
Ramanujan likely possessed. We have two choices: we could try to find a continued
fraction for the left side of (2), or we could find an algebraic representation for the
continued fraction on the right side of (2). Because the continued fraction is an
extremely elegant continued fraction, the latter tack is desirable. In fact, we at-
tempted both strategies. However, we were not able to find any kind of a continued
fraction representation for the left side resembling anything similar to the contin-
ued fraction on the right side. On the other hand, we were indeed successful in
finding an algebraic representation for Ramanujan’s beautiful continued fraction.
Of course, it is then tempting to convert our representation into a form resembling
what Ramanujan claimed on the left side of (2). Our attempts, partially with
computer algebra, to “correct” Ramanujan in this way were fruitless.
Our goal in this short note is to determine an evaluation for the continued
fraction on the right side of (2). Most likely, Ramanujan intended a, b, and c
to be positive real numbers, and so we make this assumption in the statement
of our theorem. After the conclusion of our proof, we discuss the values of the
continued fraction for other real values of a, b, and c. Although we could easily
Vol. 69 (2005) A continued fraction 259

examine the convergence and values for complex a, b, and c, even for real values of
the parameters, it is very difficult to relate all the possibilities for the convergence
and values of the continued fraction in an efficient manner. The sizes, signs, and
possible zero values for each parameter, a, b, and c, and the parity of n present
a large variety of cases that must be individually examined, yielding a variety of
results.

Theorem 2. Set
1 1 1 b 1 1
α := (3)
a + a + ··· + a + c + a + a + ···
and recall that µn is defined by (1).
Then, for any positive numbers a, b, and c,
sµ 
µ ¶2
1 µn+1 µn+1
α= −c + (1 − b) + c + (1 + b) + 4b(−1)n µ2n+1  . (4)
2 µn µn

Proof. It will be convenient to define


√ √
a + a2 + 4 a − a2 + 4 1
σ := and τ := =− .
2 2 σ
Furthermore, define, for each nonnegative integer n,
1
νn := . (5)
µn
Then, by (5),
σn − τ n
νn = √ . (6)
a2 + 4
It will be more convenient to work with νn . Using (6), it is easy to verify that νn
satisfies the recurrence relation
νn = aνn−1 + νn−2 , n ≥ 2, ν0 = 0, ν1 = 1. (7)
Then, from the elementary recurrence formulas for the numerator and denominator
of a continued fraction [1, p. 9, eq. (1.2.9)],
νn 1 1 1
= , (8)
νn+1 a + a + ··· + a
where there are n fractions a1 .
Now, by (3) and (8), write α in the form
1 1 1 b 1 1
α :=
a + a + ··· + a + c + a + a + ···
1 1 1 b
=
a + a + ··· + a + c + α
(c + α)νn + bνn−1
= , (9)
(c + α)νn+1 + bνn
260 B. C. Berndt and G. Choi AEM

where we have employed (8) and again used the elementary recurrence relations
for a continued fraction’s numerator and denominator [1, p. 9, eq. (1.2.9)]. Solving
(9) for α, we find that
α2 νn+1 − (νn − bνn − cνn+1 )α − bνn−1 − cνn = 0. (10)
Solving (10) and taking the requisite positive root, we find that
p
(1 − b)νn − cνn+1 + ((1 − b)νn − cνn+1 )2 + 4νn+1 (bνn−1 + cνn )
α= . (11)
2νn+1
We now utilize another elementary relation for the numerators and denomi-
nators of continued fractions [1, p. 9, eq. (1.2.10)] and apply it to (8) to deduce
that
νn2 − νn+1 νn−1 = (−1)n−1 . (12)
Using the elementary relation (A + B)2 = (A − B)2 + 4AB and employing (12)
under the radical sign, we conclude that
 sµ 
¶2
1 νn νn 1
α = −c + (1 − b) + c + (1 + b) + 4b(−1)n 2  . (13)
2 νn+1 νn+1 νn+1

Since, by (6), νn = 1/µn , we see that (13) is the same as (4). The convergence
of (4) and its convergence to the given value follow from a general theorem on
periodic continued fractions found in Lorentzen and Waadeland’s book [1, p. 104,
Thm. 6]. This completes the proof.

We conclude our paper with a more thorough, but by no means complete


discussion of the conditions under which Ramanujan’s continued fraction converges
to either the right side of (4) or to its conjugate. For brevity, set
√ √
(1 − b)νn − cνn+1 + D (1 − b)νn − cνn+1 − D
α1 := and α2 := , (14)
2νn+1 2νn+1
where
2
D := (cνn+1 + (1 + b)νn ) + 4b(−1)n . (15)
Set
√ √
|(1 + b)νn + cνn+1 + D| |(1 + b)νn + cνn+1 − D|
kα1 k := and kα2 k := .
2 2
(16)
From [1, p. 104, Thm. 6], α converges to αi if kαi k > kαj k for i, j = 1, 2, i 6= j.
Observe that
kα1 k > kα2 k, if D > 0 and (1 + b)νn + cνn+1 > 0,
kα2 k > kα1 k, if D > 0 and (1 + b)νn + cνn+1 < 0,
α1 = α2 , if D = 0.
Vol. 69 (2005) A continued fraction 261

Denote
δn := (1 + b)νn + cνn+1 .
Thus, by (15),
D = δn2 + 4b(−1)n .
Suppose first that abc 6= 0. Then, using the aforementioned theorem in [1], we
conclude that α converges to α1 in the following cases:
b>0 b <√
0
n even δn > √
0 δn > 2 −b
n odd δn > 2 b δn > 0
Moreover, α converges to α2 in the following cases:
b>0 b < 0√
n even δn < 0√ δn < −2 −b
n odd δn < −2 b δn < 0
We do not give any details but provide some examples as an illustration. If n
is odd, ac > 0, and −1 < b < 0, then α converges to α1 . Using (7), we can bound
νn from above and below in terms of Fibonacci numbers in various cases and then
give alternative criteria for convergence. If n is even, b, c > 0, and a < −1, then α
converges to α1 if
(1 + b)|a|n−1 Fn+1
< ,
c Fn
where Fj , j ≥ 0, denotes the jth Fibonacci number.
If abc = 0, then, as above, we must consider separately several cases. We state
one such result. Suppose that n is even, a = 0, c 6= 0, and c2 + 4b ≥ 0. Then the
continued fraction α converges to

−c + (sgn c) c2 + 4b
,
2
where (
+1, if c > 0,
sgn c =
−1, if c < 0.
Suppose that n is odd, a = 0, |b| > 1, and c 6= 0. Then the continued fraction α
converges to
c
.
b−1
Note that if b = 0, α trivially converges, since it terminates.
Lastly, note that there are cases when α does not converge, e.g., when a = c = 0
and b > 0, and when a = 0 and c2 + 4b < 0.
We are grateful to both referees for valuable suggestions, especially the referee
who advised us that we need to pay more attention to the convergence of (4).
262 B. C. Berndt and G. Choi AEM

References
[1] L. Lorentzen and H. Waadeland, Continued Fractions with Applications, North Hol-
land, Amsterdam, 1992.
[2] S. Ramanujan, The Lost Notebook and Other Unpublished Papers, Narosa, New Delhi,
1988.

B. C. Berndt and G. Choi


Department of Mathematics
University of Illinois
1409 West Green Street
Urbana, IL 61801
USA
e-mail: berndt@math.uiuc.edu
e-mail: g-choi1@math.uiuc.edu

Manuscript received: September 8, 2003 and, in final form, February 22, 2004.

Das könnte Ihnen auch gefallen