You are on page 1of 14

# A Note: The Natural Curve of Continued Fraction Approximants

John Gill

December 2014

Abstract: Forward recursion generates consecutive normal approximants of continued fractions and other
expansions. Backward recursion does the same for reverse approximants. These approximants lie on natural
smooth curves. The comments here extend the authors paper 

f ( ) =

## A linear fractional transformation (LFT) ,

a + b
, can be written in terms of its (one
c + d

f ( )
z
=K
f ( )
z

f ( n)( )
z
= Kn
( n)
f ( )
z

( ) =
(1 K ) + ( K )
n

deduced

(n )

( K ) + ( K 1)
( ) =
(1 K ) + ( K )
t

(t )

, t 0.

## Example 1 : = 4, = 4, K = .5 + .8i with z = .5 + 4i , t :0 30

Normal iterates (approximants) occur where the color of the curve changes from green to
black. The first five are shown in purple.

~1~

## ( 4 (.8i ) 4i ) + 16i ((8i ) 1)

( ) =
(1 (.8i ) ) + ( 4(.8i ) 4i )
t

Example 1a :

(t )

, = 1 + 4i , t :0 30

This linear fractional transformation is loxodromic, but close to elliptic. Hence there is both
oscillation about fixed points and convergence to the attractor.

~2~

1 1
2 2

1 + 1 2 + 2

## From above: Set

n (t , ) =

(
(t , ) =
n

n n ( K nt 1) n

(K
t
n

n n

K nt n + n n K nt 1

(1 K ) + ( K
t
n

t
n

),

Kn =

n
n

. An adaptation of this:

an (t )
n
, with n =
a normalizing factor,
bn (t ) +
n n

## An = bn An 1 + an An2 , Bn = bn Bn 1 + an Bn 2 , where the nth approximant is

0

n (t , ) =
n n
( + )
n
n

Note that

Example 2:

Cn ( ) =

An
.
Bn

if t = 0
if t = 1

(1 + i n ) z
, CF= lim C1  C2  Cn (0) , z = 2 + 3i . n=40
n
1+ i n + z

The initial point (green) on the curve is the value of the first approximant, not z.
The algorithms for continuization described in this note work best for expansions that are limitperiodic and do not have alternating signs.

~3~

Example 3:

Example 4:

Tan( z ) =

z 13 z 2 151 z 2
1 1 1 

1
4 k 2 1

z2

1 

, Tan(5 + 8i ) 0 + i

1 1
2 2
, where n = ( x + 1n ) + i ( y n1 ) , n = ( nn+1 ) + i ( n2 ) , z = 2 i
1 + 1 2 + 2

~4~

## A general composition of LFTs:

fn ( ) =

an + bn
, Fn ( ) = f1  f2  fn ( ) F ( ) .
cn + dn

Fn ( ) =

We have

An + Bn
Cn + Dn

## An = an An1 + cn Bn1 , Bn = bn An1 + dn Bn1 , Cn = anCn1 + cn Dn1 , Dn = bnCn1 + dn Dn1 .

Defining

(t )
n

(
( ) =

K nt n + n n K nt 1

(1 K ) + ( K
t
n

t
n

)=a

(t )
n
(t )
n

+ bn(t )
c + dn(t )

## The initial point (green) is the initial value of z.

~5~

(Example 2) :(again) The continued fraction using the LFT algorithm of this section:

Example 6 : CFn ( ) =

(3 + 11 i )( 4 + 11 2i ) (3 + 12 i )( 4 + 12 2i )
(7 + 11 ) + ( 11 2) i (7 + 12 ) + ( 12 2) i

= 5 + 4i , n = 10

~6~

(3 + 1n i )( 4 + 1n 2i )
(7 + 1n ) + ( 1n 2) i

Example 7 : n = 3, n = 3, K n =

1
1
+ 1 +
i , z = 3 + i, n = 10 . These LFTs are close to
10 5 + n

elliptic, whose iterations oscillate and do not converge; loose spiraling occurs:

Example 8 :

1
n

1
n

4
5

n = 1 + , n = n, K n = + i, z = 2 + i, n = 20

~7~

z 2 154 z 2
Example 9 : C( z ) =

1+ 1+
1
3

k2
4 k 2 1

z2

1+

, F ( z ) = ArcTan( z ) =

z
1 + C( z )

The value of the arc tangent is shown as a purple bullseye. Initial value of z in green.

## II. Backward Recursion: Fk ,n ( ) = fn k  fnk +1  fnk +2  fn k + k ( ) Fn ( ) as k n 1

The idea is that backward recursion a very efficient evaluation protocol for CFs starts at
some low level, say fn ( ) , and progresses upward to fn1  fn ( ) , fn2  fn1  fn ( ) , etc. At
each step a point in the plane is determined, a sort of reverse approximant. The continuization
process simply links these discrete points together in a natural continuum.
For LFTs let t :0 1 in the following
(t )
k +1,n

(
( ) =

That is to say

K mt m Fk ,n ( ) + m m K mt 1

(1 K ) F
t
m

k ,n

t
m

( ) + K m m

, m = n k 1 , k : 0,1,2,3,..., n 1

Fk ,n( ) if t = 0
Fk(+t 1,) n ( ) = fn(t k) 1(Fk ,n ( )) =
Fk +1,n ( ) if t = 1

~8~

Example 10:

CF =

z 13 z 2 151 z 2
1 1 1 

z 13 z 2 154 z 2
Example 11 : CF =
1 + 1 + 1 +

1
4 k 2 1

z2

1 

k2
4 k 2 1

z2

1 +

## , ArcTan(.2 + 2i ) 1.51 + .53i

~9~

Additional Examples:

The procedure described for CFs can be used elsewhere. For example, suppose
(6)

## fn( z ) = K n( z ) ( z n ) + n in an appropriate region in the complex plane. Then

a fairly simple theory evolves if K n( z ) < n < 1 . This condition is not essential, however, and

## K nt ( z ) might mimic the multiplier for CFs, with t : 0 1 . If so,

z if t = 0
fn(t )( z ) = K nt ( z n ) + n =
, and allowing t to increase more or less
fn ( z ) if t = 1
continuously from 0 to 1 may fill in the spaces between either regular or reverse approximants
with smooth curves.

(7)

1 zn
z z2
Example 12: e = 1 + + + ... , f n(t ) ( ) = 1 + , Gn ( ) = f n  f n 1  f1 ( )
1! 2!
n!
z

z = 2 + 3i

~ 10 ~

Example 13 : ( s) =

1 1 1
1 1
+ s + s + , f n(t ) ( ) = 1 + s , s = 1.5 + .5i , n = 50
s
1 2 3
n

## ( s ) has an attracting fixed point ( ) = 1.82 .

Example 14 :

10 ( s) =

1
1
1
+ s + s +  , s = 1.2 + 4i , n = 40
s
10 11 12

~ 11 ~

z2
z2
Sin( z ) = z 1 2 2 , f n(t ) ( ) = 1 2 2 , z = 2 3i , n = 80 ,
n
n =1
n

Example 15:

Gn ( ) = f n  f n 1  f1 ( )

Example 16 : f n( z ) = e

z
n2 +10

( z 5) + 5 , K

t
n

(z) = e

ntz
n2 +10

## Fn( z ) = f1  f2  fn( z ) .

~ 12 ~

, z = 6 + 3i , n = 50 .

Example 17 :

ez = 1 +

z z2
1
z
+ + ... , f n(t ) ( ) = ( n ) + n , n =
,
1! 2!
1 nz
n

Fn ( z ) = f1  f 2  f n (1) e z , z = 2 + 5i , n = 20

1 s
1 1 1
Example 18 : ( s ) = s + s + s + , f n(t ) ( z ) = 1 + 1 z . s = 1.5+.5i n=50
n
1 2 3

~ 13 ~

z
1
(t )
(
)
Example 19 : Tan( z ) = R
,
f
z
=
z , Tan(1.5 + .1i ) 4.7 + 6.7i

n
2
1

k =1 1 1 z 2
1

n z
4k
4

 J. Gill, A Natural Continuous Interpolating Structure for Continued Fractions, J. Comp & Appl Mathematics, 105
(1999).

~ 14 ~