Expanding Functions into Infinite Compositions

John Gill 5 January 2013

(Preliminary Discussion)

Infinite compositions of analytic functions may occur in two forms:

n n

I Inner or right compositions:

R tk (z ) = t1 t2
k =1 n

tn (z ) , T (z ) = lim R t k (z ) .
n k =1 n

II Outer or left compositions:

Ltk (z ) = tn tn 1
k =1

t1 (z ) , T (z ) = lim Lt k (z ) .
n k =1

Convergence theory of each of these may be found in [1] and [2]. Here, the emphasis will be on finding algorithms for converting closed form expressions into infinite expansions. Simple functional equations that relate a function of nz to an expression of the same function of z sometimes lead directly to such expansions. Consider the example: Example Ia

Tan (2z ) =

2Tan (z ) . 1 Tan 2 (z )

We follow the procedure (Kojima [2]):

T (z ) = Tan (z ) = 2 =

z z z z Tan ( z )= 2z Tan ( z )= 4 Tan ( z ) 2 2 4 2 1 2 1 2 1z 1 4z 1 4z 1 z2 z z

2 1 4

1 z
n

1 42

z2

4z Tan ( z )= 4

Which leads to T (z ) =

Rt
k =1

(z ) = t 1 t 2

tn (rn (z )) with

tk (z ) =

z 1
1 4k

z2

, rn (z ) = 2nTan (z / 2n ) z
n k =1

Computer experiments lead to the conjecture

Tan (z ) = lim R t k (z ) = R t k (z ) .
n k =1

~1~

Image (I)

Tan(z) (n=5)

and

Tan(z) z (n=20)

[-8<x,y<8]

Convergence in a neighborhood of z = 0 can be seen by applying the following ______________________________________________________________________ Theorem 2.6 [1] Suppose f n ( z ) = z (1 + n ( z ) ) , with n analytic for z R1 and

n ( z ) < n ,

< . Choose 0 < r < R1 , and define R = R (r ) =

R1 r

(1 + )
k k =1

Then

Fn (z) = f1 f 2

f n (z) F(z) uniformly for z R

and F ( z )

(1 + ) < where
k k =1

k =

R1 k . r

_______________________________________________________________________

Here

z 1 n (z ) = n . The actual region of convergence is larger. 2 1 4 1 4 n z

Although there are many examples showing the convergence of an inner or right composition to the function of which it is an expansion (see, e.g., analytic theory of continued fractions [3]), there are perhaps no previous non-trivial examples showing the same for outer or left compositions. Here is one:

~2~

Example IIa and

1 Set gk (z ) = tk (z ) , n (z ) = rn1 (z ) (Example Ia). Then gk (z ) z , n (z ) z

Arc tan(z ) = n gn gn 1
with

g1 (z ) gn gn 1
Set

g1 (z ) g1 (z ) .

gk (z ) =

2 4k z

( 1+

1 4k

z2 1 .

Gn (z ) = gn gn 1

Hence, the conjecture

Gn (z ) Arc tan(z ) , or

2 4k Arc tan(z ) = L z k =1

( 1+

1 4k

z 2 1 , P.V. of course, which is supported by computer

examples. This may be written in a simpler form:

Arc tan(z ) =

2z L k =0 1 + 1 + k z

z , k = k 4

Convergence in a neighborhood of z = 0 can be verified by employing the following ___________________________________________________________________________ Theorem 2.8 [1] Let {g n } be a sequence of complex functions defined on S=(|z|<M) . Suppose there exists a sequence {n } such that

n < and g n (z) z < Cn if z < M . Set

k=1

= C k
1

and R 0 = M > 0 . Then, for every z S0 = z < R 0 , G n (z) = g n g n 1 uniformly on compact subsets of S0 .

g1 (z) G(z) ,

___________________________________________________________________________

Here

z 1 gn (z ) z = n 4 1+ 1+

.
1 4n

The actual convergence region is larger.

z2

Example Ib

F (z ) = e z 1 . F (2z ) = F (z )(F (z ) + 2) gives

z2 z2 F (z ) = z (z + 2) F (z / 2) = + z 2z F (z / 2) = + z 2F (z / 2) 4 4 z2 = + z 4

(2z

+ 4z

z2 z2 F (z / 4) = + z + z 4z F (z / 4) = 4 8

~3~

With rn (z ) = 2n F

z z . We have the following n 2

z2 e z = 1 + R k +1 + z k =1 2

Theorem 2.6[1] can be used to show convergence in a neighborhood of z = 0 with n (z ) = Of course, the radius of convergence of the composition is actually infinite.

z . 2
n +1

z2 Example IIb In the previous example tk (z ) = k +1 + z , so that 2 2z 1 gk (z ) = tk (z ) = . Thus Ln (z + 1) Gn (z ) = gn gn 1 1 + 1 + 4 2k1+1 z

( )

g1 (z ) and

2z Ln (z + 1) = L k =1 1 + 1 + 4

( )
1 2k +1

. n = 50 gives ten decimal place accuracy for Ln (6 8i ) . z

Image (2)

Ln(1+z) - z , n=10, -8<x,y<8 and Ln(1+z)

~4~

Image (3)

z2 1 F (z ) = R k + z k =1 10

n=5

-8<x,y<8 G10 (z ) =

z2 1 k + non-convergent L z k = 0 10
10

Image (4) F (z ) =

k 1 k + 1 z + 2k z R k =1
50

50

and G (z ) =

k =0

z + 2 z L k + 1
k

-8<x,y<8

~5~

Example Ic

F (z ) = Sin (z )

Sin (z ) = 2z 1 z 2 Sin (z / 2) = z 1 =z 1

1 2 1 z 2Sin (z / 2) = z 1 z 2 4z 1 z 2 Sin (z / 4) 4 4

1 2 1 z z 1 2 z 2 4Sin (z / 4) = 4 4 n n with rn (z ) = 2 Sin (z / 2 ) z . We have

1 Sin (z ) = R z 1 k z 2 4 k =1

Where the positive sign is valid in Q1 and Q4 and the negative sign in Q2 and Q3. For Sin (1 + 4i ) the value is accurate to ten decimal places for n = 20 . Image (5) Sin(z) n=10

Continued Fractions (CFs) are a special case of inner composition (I), involving two complex variables. One type is F (z ) =

a1 (z ) a2 (z ) 1+ a (z ) 1+ 3 1+

, defined by tn (z ; ) =

an (z ) and 1+

T1 (z ; ) = t1 (z ; ) , Tn (z ; ) = Tn 1 (z; tn (z ; )) . Then

~6~

F (z ) = lim Tn (z ; ) or R (tn (z ; ) ) = 0 .
n
n =1

Although = 0 normally, other values of the variable frequently lead to the same value of F (z ) . The essential difference between the examples cited previously and CFs is that the former represent compositions on z that lead to functions F (z ) , whereas the latter evolve from compositions on an auxiliary variable , leading to F (z ) . Example Id tn (z ; ) =

an (z ) 1 where < R R and an (z ) < R (1 R ) , 0 < < 1 , 1+ 2

with an (z ) analytic for z S . Then tn (z ; ) < R and these functions contract uniformly.

Therefore F (z ) =

R 1+
n =1

an (z )

, analytic for < R and z S (see Contraction Theorems in [1]).

III

Implicit functions and Zeno contours: Consider an expression defining a function implicitly:

( , f ( ) ) = 0 or ( , z ) = 0 , z = f ( ) .
The following definition is from [4]:

Zeno contour: Let gk ,n ( z ) = z + k ,n ( z ) where z S and gk ,n ( z ) S for a convex set S in the complex plane. Require lim k ,n = 0 , where (usually) k = 1,2,..., n . Set G1,n ( z ) = g1,n ( z )
n

, Gk ,n ( z ) = gk ,n ( Gk 1,n ( z ) ) and Gn ( z ) = Gn, n ( z ) with G ( z ) = lim Gn ( z ) , when that limit exists.

n

The Zeno contour is a graph of this iteration. Normally, for a vector field, F = F , (z ) = F (z ) z , and under the right conditions G (z ) = , an attractive fixed point of F . In the context of this discussion gk ,n (z ) = z + k ,n ( F ( , z ) z ) , and if
F < < 1 S , z

a suitable domain, then Gn (z ) ( ) = f ( ) a fixed point for each value of , starting with an initial value z in some neighborhood of the fixed points [4]. Thus, from the notation II,
n n

III

Lgk ,n (z ) = gn ,n gn 1,n
k =1

g1,n (z )

and

G (z ) = lim Lgk ,n (z ) .
n k =1

~7~

z z Example IIIa: ( , z ) = Cos + z = 0 . Then F ( , z ) = Cos and the Zeno 10 10 contour terminates at z = f ( ) for in a neighborhood of the origin and initial values of z near the fixed points. For example,

f (1 2i ) 1.3289 + i (2.0128) and f (1 + i ) 1.0209 + i (1.0209) ,

starting with z0 = 4 + 2i .

References
[1] J. Gill, Convergence of infinite compositions of complex functions, Comm. Anal. Th. Cont. Frac., Vol XIX (2012) [2] S. Kojima, Convergence of infinite compositions of entire functions, arXiv:1009.2833v1 [3] L. Lorentzen, H. Waadeland, Continued Fractions with Applications, North Holland (1992) [4] J. Gill, Zeno contours and attractors, Comm. Anal. Th. Cont. Frac., Vol XIX (2012)

~8~