You are on page 1of 15

GENERAL ⎜ ARTICLE

Continued Fractions for e


Shailesh A Shirali

I nt r oduct ion

As Paul Erd}os might have said, `even babies' know t hat a


¯nit e simple cont inued fraction (SCF for short ) always yields
a rat ional number, while an in¯nite periodic SCF always
yields a quadrat ic irrat ional. Many results of great beauty
are known about CFs. Consider for instance the following
Shailesh Shirali has two results quot ed by Ramanujan to Hardy in his very ¯rst
been at the Rishi Valley
let ter (t he general pat tern in each case is clear):
School (Krishnamurti
Foundation of India),
Z a2
Rishi Valley, Andhra a
¡ x2 1p e¡
Pradesh, for more than e dx = ¼¡ ;
0 2 1
ten years and is cur- 2a +
2
rently the Principal. He a+
has been involved in the 3
2a +
Mathematical Olympiad a + ¢¢¢
Programme since 1988.
He has a deep interest and
in talking and writing
8Ã 9
about mathematics,
< p ! 1=2 p =
particularly about its 1 5+ 5 5+ 1 2¼=5
¡ 2¼ = ¡ e :
historical aspects. He is e : 2 2 ;
also interested in 1+
e¡ 4¼
problem solving 1+
1 + ¢¢¢
(particularly in the fields
of elementary number
1
theory, geometry and It is hard to resist recalling what Watson had t o say about
combinatorics). such formulae:

Such a formula gives me a t hrill which is indist in-


guishable from t he t hrill which I feel when I ent er t he
Sagriest a Nuova of Capelle Medicee and see before me
t he aust ere beauty of `Day', `Night ', `Evening' and
`Dawn' which Michelangelo has set over t he t ombs of
1
Quoted by S Chandrasekhar
Giuliano de' Medici and Lorenzo de' Medici.
in Truth and Beauty.

14 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

In this article, we examine some in¯nit e CFs t hat converge


t o numbers relat ed to e. For a quick preview, here is one
such relat ion:

1 1
= ; (1)
e¡ 1 2
1+
3
2+
4
3+
4 + ¢¢¢

which may be writ t en equivalent ly (and perhaps more ele-


gantly) as
1 1=2
= 1+ : (2)
e¡ 2 1=3
1+
1=4
1+
1 + ¢¢¢

T his relation is easy t o prove. Our main object ive in this


article will be t o prove the following pretty but none t oo
obvious identit ies:

e2 ¡ 1 1
= ; (3)
e2 + 1 1
1+
1
3+
1
5+
7 + ¢¢¢

and
1 2
p = 1+ ; (4)
e¡ 1 4
3+
6
5+
7 + ¢¢¢

T he formal de¯nit ion of a cont inued fract ion is given in


Box 1, and some well-known theorems concerning CFs are
given in Box 2. We shall have occasion to use a result due t o
Emil Cesµaro relat ing to limit s of sequences; t his is described
in Box 3.

RESONANCE ⎜ January 2000 15


GENERAL ⎜ ARTICLE

B ox 1. D e¯nit ion of C ont i nued Fr act ions

Let a0 , a1, b1 , a2, b2 , . . . beint egers. Weshall use the not at ion [a0 ; ( a1 ; b1 ) ; ( a2 ; b2) ;
( a3 ; b3 ) ; : : :] t o denot e the number recursively de¯ned as follows: [a0; ( a1 ; b1)] =
a0 + a1 =b1 , and for n > 1,

[a0 ; ( a1; b1 ) ; ( a2 ; b2 ) ; : : : ; ( an¡ 1 ; bn ¡ 1 ) ; ( an ; bn )] =

[a0; ( a1 ; b1) ; ( a2 ; b2 ) ; : : : ; ( an ¡ 1 ; bn¡ 1 + an =bn )]:

T his will be referred to as a general continued fraction (GCF for short); it may
also be writ t en in t he following manner:

a1
a0 + :
a2
b1 +
a3
b2 +
b3 + ¢¢¢

T he not ation [a; b; c; d; : : :], wit hout t he parent heses wit hin the [ ], will refer to a
simple continued fraction (SCF for short ); it is de¯ned as the GCF [a; (1; b) ; (1; c) ;
3(1; d) ; : : :]; t he numerators here are all 1. It may also be writ ten as

1
a+ :
1
b+
1
c+
d + ¢¢¢

For typographic convenience, we shall at t imes use t he notat ion


1 1 1
a+ ¢¢¢
b+ c+ d+
t o denot e t he SCF [a; b; c; d; : : :].

P r oof of Equat ion 1

Let the sequences f un g and f vn g be de¯ned as follows:


)
un = n !;
(5)
v1 = 0; v2 = 1; vn = ( n ¡ 1)( vn¡ 1 + vn¡ 2)( n ¸ 2) :

16 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

B ox 2. Fact s A b out Cont i nued Fr act i ons

We brie° y review here t he main fact s concerning CFs. For proofs t he reader
should consult t he text by Hall and Knight , or t he one by Hardy and Wright .
([1],[2])

1. Any rational number may be writt en as a ¯nit e SCF { in two possible ways.
For inst ance, t he rat ional number 4=13 may be writ ten as [0; 3; 4] and also
as [0; 3; 3; 1].
2. By t he in¯nit e SCF [a0; a1; a2; a3 ; a4; : : :] we shall mean t he limit of the
sequence
a0 ; [a0 ; a1 ]; [a0; a1; a2]; [a0 ; a1 ; a2 ; a3 ]; :::;
provided that t he limit exist s; and similarly for an in¯nit e GCF.
3. Let a1 ; a2 ; a3 ; : : : be any in¯nit e sequence of posit ive int egers, and let a0 be
any integer. T hen the in¯nit e SCF [a0 ; a1 ; a2; a3; : : :] converges t o a real
number.
4. An in¯nit e SCF corresponds t o an irrat ional real number, and t he repre-
sent at ion of an irrational real number by an in¯nite SCF is unique.
5. A convergent in¯nit e GCF does not necessarily converge to an irrat ional
number. Also, uniqueness of represent ation does not hold.
For inst ance, we have the rather trivial relat ion
2 = [1; (2; 1) ; (2; 1) ; (2; 1) ; : : :]

via t he fact t hat 2 is one of t he solut ions of t he equat ion x = 1 + 2=x .


(Convergence is easy to show.) Likewise t he equat ion x = 2 + 3=x yields
the relat ion 3 = [2; (3; 2) ; (3; 2) ; (3; 2) ; : : :].
6. By simple calculat ion we ¯nd t hat for any SCF [a0 ; a1 ; a2 ; a3; : : :],
a0 a1 + 1 a0 a1 a2 + a0 + a2
[a0 ; a1 ] = ; [a0 ; a1 ; a2 ] = ; ::::
a1 a1 a2 + 1
Let [a0; a1; a2; a3; : : : ; an ] = Pn =Qn where Pn and Qn are coprime; t hen
Pn =Qn is the n t h convergent to t he SCF. T he following very convenient
formulae may be used for computing convergents:

continued ...

RESONANCE ⎜ January 2000 17


GENERAL ⎜ ARTICLE

)
Pn = an Pn¡ 1 + Pn ¡ 2
( n ¸ 2) :
Qn = an Qn¡ 1 + Qn¡ 2

More generally, t he convergents Pn =Qn t o the GCF [a0 ; ( a1 ; b1 ) ; ( a2 ; b2) ;


( a3 ; b3 ) : : :] follow t he recursive law displayed below.
)
Pn = bn Pn¡ 1 + an Pn¡ 2
( n ¸ 2) :
Qn = bn Qn¡ 1 + an Qn ¡ 2

7. T he following result holds for any two successive convergent s of a SCF:

Pn Qn ¡ 1 ¡ Pn¡ 1 Qn = § 1:

(T here is an easy proof via induction.)

8. An in¯nit e SCF is periodic if the sequence a0; a1; a2 ; a3 ; : : : is ult imat ely
periodic (t hat is, if an+ p = an for some p > 0 and for all n beyond some
point). T he following is known: A periodic SCF converges p to a quadratic
irrational, that is, to a number of the form ( a + b) =c where a; b; c are
integers
p ( b; c > 0). For inst ance, the in¯nite
p SCF [1; 2; 2; 2; : : :] converges
to 2, and [1; 1; 2; 1; 2; : : :] converges to 3. A part icularlyp pretty SCF is
[1; 1; 1; 1; : : :], which converges t o t he `golden number' ( 5 + 1) =2.
Int erest ingly, cubic and higher order algebraic irrat ionals never seem t o
turn up. Indeed, the SCFs for cube roots and higher order roots of rat ional
numbers show no discernible pat tern at all.

More result s concerning CFs are known, and may be found in t he books list ed in
t he Suggested Reading.

We observe ¯rstly t hat t he two sequences follow the same


recursive law, for
h i
n ! = ( n ¡ 1) ( n ¡ 1)! + ( n ¡ 2)! :

T he ¯rst few values of un and vn are shown in t he following


t able:

18 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

B ox 3. A R esul t D ue t o Em i l C esµ
ar o

Suppose that f an gn ¸ 0 and f bn gn¸ 0 are two in¯nit e sequences of positive numbers
such that

1 the series A ( t ) = a0 + a1 t + a2t 2 + ¢¢¢and B ( t ) = b0 + b1 t + b2 t 2 + ¢¢¢have


equal radii of convergence, say R , where R > 0 ; and

2 the limit of an =bn as n ! 1 exist s, say an =bn ! s.

If A ( t ) and B ( t ) ! 1 as t ! R ¡ , i.e. t ! R from the left then

A(t)
lim¡ = s:
t! R B (t )

Exam pl e. Let an = n + 1 and bn = 2n + 1. The series

1 + 2t + 3t 2 + 4t 3 + ¢¢¢; 1 + 3t + 5t 2 + 7t 3 + ¢¢¢

have radii of convergence equal to 1, as may easily be shown. For t = 1 both


series diverge, and for j t j < 1 t he two series converge respect ively t o t he quant it ies
1 1+ t
A(t) = ; B (t) = :
(1 ¡ t ) 2 (1 ¡ t ) 2

Observe t hat bot h t hese quantit ies diverge as t ! 1. T heir rat io is equal t o
1=(1 + t ), which t ends to 1=2 as t ! 1. Not e t hat an =bn t ends t o the same limit
as n ! 1 .

n 1 2 3 4 5 6 7 :::
un 1 2 6 24 120 720 5040 : : :
vn 0 1 2 9 44 265 1854 : : :

We ¯nd t hat vn =un is the ( n ¡ 1) t h part ial sum of t he series


(1=2 ¡ 1=6 + 1=24 ¡ ¢¢¢+ (¡ 1) n =n ! + ¢¢¢). T he proof, which

RESONANCE ⎜ January 2000 19


GENERAL ⎜ ARTICLE

is easy and left for t he reader, is via induct ion. It follows


t hat vn =un ! 1=e, and t herefore that un =vn ! e.

On t he ot her hand, we ¯nd using t he recursion relations (see


Box 2), that un =vn is t he ( n ¡ 1) t h convergent of the GCF
2 3 4 5
[2; (2; 2) ; (3; 3) ; (4; 4) ; (5; 5) : : :] = 2+ ¢¢¢:
2+ 3+ 4+ 5+
So t he GCF equals e, and we obt ain a result equivalent t o
(1).

T he M ain R esult

We shall now prove t hat

e2 ¡ 1 1 1 1 1
= ¢¢¢ = [0; 1; 3; 5; 7; : : :]:
e2 + 1 1+ 3+ 5+ 7+

Let an =bn denot et he n t h convergent to t heSCF [0; 1; 3; 5; 7; : : :].


T hen
a1 0 a2 1 a3 3 a4 16
= ; = ; = ; = ; : : : ; (6)
b1 1 b2 1 b3 4 b4 21

and in general, via t he recursion law (it em 6 in Box 2), for


n > 0, )
an + 2 = (2n + 1) an+ 1 + an
: (7)
bn+ 2 = (2n + 1) bn + 1 + bn

Observe t hat the two sequences follow t he same recursion.


For reasons of convenience, we de¯ne a0 = 1 and b0 = 0.
T he ¯rst few values of an and bn are shown below.
n 0 1 2 3 4 5 6 7
an 1 0 1 3 16 115 1051 11676
bn 0 1 1 4 21 151 1380 15331

Bot h sequences grow wit h t remendous rapidity. Thus, we


have,

a20 = 9508271497633004786551 ¼ 9:51 £ 1021 ;


b20 = 12484695980499706867555 ¼ 1:25 £ 1022 :

20 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

We now introduce t he exponent ial generat ing funct ions


X1 xn X1 xn
A(x) = an ¢ ; B (x ) = bn ¢ : (8)
n= 0 n! n= 0 n!

Our ¯rst st ep will be t o show t hat bot h series have radius


of convergence equal t o 12 . For this we need t o show that
µ ¶ 1=n µ ¶ 1=n
an bn
lim
n! 1
= 2; lim
n! 1
= 2:
n! n!

P r oof. By de¯nition, we have an = (2n ¡ 3) an¡ 1 + an¡ 2


for n > 1. Since f an g is an increasing sequence of posit ive
integers, it follows t hat an lies between (2n ¡ 3) an¡ 1 and
(2n ¡ 2) an¡ 1, and therefore t hat
an
2n ¡ 3 < < 2n ¡ 2 (for n > 2; equality holds when n = 2) :
an¡ 1

Multiplying t ogether the corresponding terms of n ¡ 2 such


inequalit ies, we get

(2n ¡ 3) ¢(2n ¡ 5) ¢¢¢¢¢3 < an < (2n ¡ 2) ¢(2n ¡ 4) ¢¢¢¢¢4;

which yields, on division by n ! 2n ,


(2n ¡ 3)! an 1
2n ¡ 2 < n < :
2 ( n ¡ 2)!n ! n! 2 4n

Using St irling's approximation for t he factorial function we


obt ain, via `sandwiching',
µ ¶ 1=n
an
lim = 1; (9)
n! 1 n ! 2n

showing t hat t he radius of convergence of A ( x ) is 12 . T he


proof for B ( x ) is t he same, and t he same result is obt ained.
So both series have radius of convergence equal t o 12 .

Exact ly t he same st at ements may be made about t he deriv-


at ives A 0( x ), B 0( x ): bot h series have radius of convergence
equal t o 12 .

RESONANCE ⎜ January 2000 21


GENERAL ⎜ ARTICLE

We now proceed t o ¯nd a di®erent ial equation sat is¯ed by


bot h A ( x ) and B ( x ). Di®erent iat ion yields the following
(in t he summat ions shown below we have omit t ed t he sub-
script s, t o avoid clut ter):
X xn X xn
0 00
A (x) = an+ 1 ¢ ; A (x) = an+ 2 ¢ ; (10)
n! n!
with similar equations for B 0( x ) and B 00( x ). Now the rela-
t ion an+ 2 = (2n + 1) an + 1 + an may also be writ ten as
an+ 2 an+ 1 an+ 1 an
¡ 2¢ = + :
n! ( n ¡ 1)! n! n!

Multiplying t hroughout by x n , summing over n and using


(8) and (10), we ¯nd t hat A ( x ) sat is¯es the following second
order linear di®erent ial equat ion:

(1 ¡ 2x ) A 00( x ) = A 0( x ) + A ( x ) :

T he funct ion B ( x ) sat is¯es the same kind of equation. So


bot h A ( x ) and B ( x ) are solut ions of t he equation

(1 ¡ 2x ) y00 = y0 + y (11)

in t he unknown function y. Observe that (11) is a homo-


geneous linear equat ion wit h non-const ant coe± cient s, and
t hat x = 12 is a singular point for the equat ion. Indeed, t he
solution cannot be easily carried beyond t his point .

T he most nat ural move towards solving t he equation is t o


make t he subst itution u = 1 ¡ 2x . We get, easily enough,

dy 1 d2y 1 00
= ¡ y0; 2 = y ;
du 2 du 4
and the di®erent ial equat ion gets transformed to

d2 y dy
4u + 2 ¡ y = 0: (12)
du2 du

Fortunat ely t he general solut ion of t his equat ion can be


guessed. T he form of t he coe± cient (4u) in t he 1st term

22 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

p
u
suggest
p s t hat t he solut ion will cont ain the t erm e ; for if
y = e u , then
p µ ¶ µ ¶
dy p 1 d2 y e u 1 y 1
= e u¢ p ; = 1¡ p = 1¡ p :
du 2 u du2 4u u 4u u
p
It may be easily veri¯edp t hat y = e u does indeed sat isfy
(12), and so does y = e¡ u . Since the equat ion is linear and
of 2nd order, t he general solut ion of (12) is
p p
y( u) = ce
u
+ de¡ u

where c; d are const ants.

To ¯nd c and d we use t he following: (a) A (0) = 1, A 0(0) =


0 (the derivat ive being wit h respect t o x ); (b) B (0) = 0,
0
B (0) = 1; (c) u = 1 when x = 0. Also:
¯
dA dA dx 1 dA dA ¯
¯
= ¢ = ¡ ; so ¯ = 0;
du dx du 2 dx du u= 1
and similarly,
¯
dB 1 dB dB ¯
¯ 1
= ¡ ; ¯ = ¡ :
du 2 dx du u= 1 2

So to ¯nd A as a funct ion of u we need t o solve t he equa-


t ions y(1) = 1, y0(1) = 0 for c and d. We get , aft er some
manipulation,

e¡ 1 e p
c= ; d= ; so y( u) = cosh( u ¡ 1) :
2 2
In t he case of B we need to solve t he equat ions y(1) = 0,
1
y0(1) = ¡ 2 . We get ,
¡
e 1 e p
c= ¡ ; d= ; so y( u) = sinh(1 ¡ u) :
2 2

Remembering that u = 1 ¡ 2x we have ¯nally,


p )
A ( x ) = cosh(1 ¡ p 1 ¡ 2x ) ;
(13)
B ( x ) = sinh(1 ¡ 1 ¡ 2x ) :

RESONANCE ⎜ January 2000 23


GENERAL ⎜ ARTICLE

We are almost at t he end of our long journey. We need


t o ¯nd limn! 1 an =bn , and we shall use Cesµ aro's result (see
Box 3) wit h the value R = 12 . However we cannot immedi-
at ely use t he result , because the functions
p p
cosh(1 ¡ 1 ¡ 2x ) ; sinh(1 ¡ 1 ¡ 2x )

do not diverge as x ! 12 (divergence plays a crucial role


in t he t heorem); indeed, t he two funct ions t ake the values
cosh 1, sinh 1, respect ively. However all is not lost { we can
work inst ead wit h t he derivatives of t hese two funct ions,
t hat is, wit h the funct ions
p p
sinh(1 ¡ 1 ¡ 2x ) cosh(1 ¡ 1 ¡ 2x )
p ; p :
1 ¡ 2x 1 ¡ 2x

T he crucial fact s are: (a) t hederivatives t end to 1 as x ! 12 ;


(b) t he rat io of the coe± cient s of x n in A 0( x ) and B 0( x )
is the same as t he corresponding rat io for A ( x ) and B ( x ).
T herefore t he required limit is equal t o t he limit, as x ! 12 ,
of t he quant ity
p p
sinh(1 ¡ 1 ¡ 2x ) cosh(1 ¡ 1 ¡ 2x )
p ¥ p :
1 ¡ 2x 1 ¡ 2x

T his limitp is easy t o compute; t he expression simpli¯es t o


t anh(1 ¡ 1 ¡ 2x ), which tends t o t anh 1 when x ! 12 . So
t he limit of an =bn as n ! 1 is t anh 1; t hat is,

an e ¡ e¡ 1
e2 ¡ 1
! = ;
bn e + e¡ 1
e2 + 1
and we have t he st at ed result .

T he result may be extended to the following: for any real


number a,
a a2 a2 a2
t anh a = ¢¢¢: (14)
1+ 3+ 5+ 7+
Equivalent ly:

e2a ¡ 1 a a2 a2 a2
= ¢¢¢: (15)
e2a + 1 1+ 3+ 5+ 7+

24 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

T he other result (4) follows in the same way. We write an =bn


for the n t h convergent t o
2 4 6
¢¢¢:
3+ 5+ 7+
T hen we have, for n > 0,
)
an + 2 = (2n + 3) an+ 1 + (2n + 2) an
: (16)
bn + 2 = (2n + 3) bn+ 1 + (2n + 2) bn

We also have a0 = 1, a1 = 0, b1 = 0, b1 = 1 (as earlier).


Going t hroughPthe same mot ions, we ¯nd P
that t he two func-
t ions A ( x ) = an x n =n ! and B ( x ) = bn x n =n ! have radii
of convergence equal to 12 , and bot h obey t he homogeneous
linear di®erent ial equat ion

(1 ¡ 2x ) y00 = (3 + 2x ) y0 + 2y: (17)

T his is solved as earlier by t he substit ut ion u = 1 ¡ 2x . We


obt ain, after some e®ort,
à ! µ ¶
¡
e x 1
y= c +d
1 ¡ 2x 1 ¡ 2x

for suit able const ants c and d. Plugging in t he boundary


condit ions, we ¯nd t hat

2e¡ x ¡ 1 1 ¡ e¡ x
A(x) = ; B (x ) = : (18)
1 ¡ 2x 1 ¡ 2x

T he closed forms are valid, as earlier, only for j x j < R (wit h


1
R = 2 ), but in t his case, both expressions do diverge as
1
x ! 2 . So we do not need t o compute t he derivat ives; t he
required limit is
p
A (x ) 2e¡ x ¡ 1 2 ¡ e
lim = lim ¡ x = p ; (19)
1 ¡ B (x) 1¡ 1 ¡ e e¡ 1
x! 2 x! 2

by st raightforward substit ut ion. It follows that


2 4 6 1
1+ ¢¢¢= p :
3+ 5+ 7+ e¡ 1

RESONANCE ⎜ January 2000 25


GENERAL ⎜ ARTICLE

T he met hod we have used is quit e general; t he only qual-


i¯cat ion is that t he di®erent ial equat ion we get should be
solvable in closed from. If we had start ed with t he CF
1 1 1 1
¢¢¢; (20)
1¡ 3¡ 5¡ 7¡
where minus signs have replaced t he plus signs in (3), t hen
t he di®erent ial equation reached is (1 ¡ 2x ) y00 = y0 ¡ y. T he
same subst it ut ion works ( u = 1 ¡ 2x ; or, much bettper, u2 =
1 ¡ 2x ). T he result pobt ained is: A ( x ) = cos(1 ¡ 1 ¡ 2x )
and B ( x ) = sin(1 ¡ 1 ¡ 2x ) (wit h A ( x ) and B ( x ) de¯ned
exact ly as earlier). Once again we must go t o t he ¯rst deriv-
at ives before applying t he theorem of Cesµ aro, and we ¯nd
¯nally t hat t he CF converges t o t an 1. More generally we
have, for any real number a,

a a2 a2 a2
t an a = ¢¢¢: (21)
1¡ 3¡ 5¡ 7¡
We could also, just for fun, apply t he method t o a familiar
SCF:
1 1 1 1
1+ ¢¢¢: (22)
1+ 1+ 1+ 1+
Herept he di®erent ial equation is y00 = y0 + y. Let ® =
(1 + 5) =2 denot e t he golden rat io, and let ¯ = 1=®. This
t ime the radius of convergence is in¯nit e for both A ( x ) and
B ( x ). We ¯nd, aft er wading through lot s of manipulat ions,
t hat
®e®x + ¯e¡ ¯x
e®x ¡ e¡ ¯x
A (x ) = p ; B (x) = p ;
5 5
so the SCF converges to limx ! 1 A ( x ) =B ( x ) = ®. (This
could have been obtained much more simply!) More gen-
erally, we have t he following result which is t rue for any
posit ive real number a:
p
1 + 1 + 4a a a a a
= 1+ ¢¢¢: (23)
2 1+ 1+ 1+ 1+

26 RESONANCE ⎜ January 2000


GENERAL ⎜ ARTICLE

T he SC F for e

From (15) we obt ain, using a = 1=2 and repeat edly carrying
t he `2' int o the denominat or,
e¡ 1 1 1 1 1
= ¢¢¢; (24)
e + 1 2+ 6+ 10+ 14+
and therefore
2
e= 1+ 1 1 1 : (25)
1+ 6+ 10+ 14+ ¢¢¢

T his is not a SCF. However a SCF may be obt ained from


(25) using simple algebra, via the following two easily veri-
¯ed ident it ies:
2 1
= ; (26)
1 1
2k + k+
a+ x 2( a + x )
2 1
= : (27)
1 1
(2k + 1) + k+
a+ x 1
1+
2
1+
( a ¡ 1) + x

Using (26) and (27) repeat edly, we obt ain Euler's elegant
and very famous result,
1 1 1 1 1 1
e = 2+ ¢¢¢
1+ 2+ 1+ 1+ 4+ 1+
= [2; 1; 2; 1; 1; 4; 1; 1; 6; 1; 1; 8; 1; : : : ; 1; 2k; 1; : : :]:(28)

N umb er Cr unching

It is interest ing t o see how close are t he approximat ions to e


provided by (24). The GCF gives t he following convergents
t o ( e ¡ 1) =( e + 1):
6 61 860 15541 342762 8927353
; ; ; ; ; ;::::
13 132 1861 33630 741721 19318376

RESONANCE ⎜ January 2000 27


GENERAL ⎜ ARTICLE

T herefore t he following are rat ional approximat ions for e:


19 193 2721 49171 1084483 28245729
; ; ; ; ; ;:::: (29)
7 71 1001 18089 398959 10391023

Comparing t hese numbers wit h e, we ¯nd t hat t he errors are


roughly as follows:

4£ 10¡ 3 ; ¡ 3£ 10¡ 5 ; 1£ 10¡ 7 ; ¡ 3£ 10¡ 10 ¡ 13


; 5£ 10 ; ¡ 6£ 10
¡ 16
;::

An impressive convergence rate! { t he error in the 10th con-


vergent is less t han 3 £ 10¡ 25 .

Post scr ipt

T he reader may wonder why we have not at t empt ed to com-


pute the SCF
1 1 1 1
[0; 1; 2; 3; 4; : : :] = ¢¢¢;
1+ 2+ 3+ 4+
which is surely much more `nat ural looking' t han [0; 1; 3; 5; : : :].
T he answer is a prosaic one. T he recursion equat ions ob-
t ained for this SCF are an = nan¡ 1 + an¡ 2 and bn = nbn ¡ 1 +
bn¡ 2, and t he di®erent ial equat ion obtained as a result is:

(1 ¡ x ) y00 = 2y0 + y:

Unfort unat ely this equat ion does not seem t o be solvable in
closed form; at any rat e I am unable t o ¯nd such a solu-
t ion! And of course this very e®ectively put s an end t o t he
investigat ion . . . .

Suggested Reading
[1] Hall and Knight, Higher Algebra, Macmillan & Co. Ltd., London, 1960.
[2] Hardy and Wright, Introduction to the theory of numbers, Oxford Uni-
versity Press, 1960.
[3] Polya and Szego, Problems and theorems from analysis, Problem 85),
Address for Correspondence
Springer -Verlag , Vol. I, 1972.
Shailesh A Shirali
[4] S Barnard and J M Child, Higher Algebra, Macmillan and Company,
Rishi Valley School
London.
Chittoor District
[5] C D Olds, The Simple Continued Fraction Expansion of e, in The
Rishi Valley 517 352
Chauvenet Papers, Vol 2 (MAA).
Andhra Pradesh, India.
[6] Mathematics Magazine, Problem 1254,Vol. 61, No. 1 (February 1988).

28 RESONANCE ⎜ January 2000

You might also like