Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Appl. Math.

-JCU
14B(1999) ,90-98

A CLASS OF F U N C T I O N A L EQUATION AND FRACTAL I N T E R P O L A T I O N FUNCTIONS


Sha Zhen
Abstract. A new class of functional equation in Co (I) is investigated. It is proved that some class of FIF satisfies the functional equation. Another functional equation is constructed. Their solutions can approximate FIF arbitrarily. And a new approximate estimate between FIF and interpolated function is given.
e

w1

Introduction
~xN=l be a partition of the

First we introduce some notations. Let 0 = x 0 ~ x ~ . . .

interval I = [ 0 , 1 ] , I j = [-x~-i , x j ] , ) = 1,2 . . . . . N , and K = I X [a , b ] , where - - o o < a < b < -.~-c<~. Let Lj'I--,-Ij be contractive homeomorphisms satisfying :

Given a set { ( x j , y j ) E I X [ a , b ] l j = O , 1 function such that

Lj(xo) = x j 1,Lj(xN) = x j , j = 1,2 . . . . . N , (ii) [Lj(x')--Lj(x")[ ~pIx' - - x " [ , Vx',x"E


(i)

1,0~p~

1.

(1.1)

. . . . . N } , let Fj:K--~[a,b] be a continuous

When the mapping Wj.K-,-K is defined by

Fj(xo,yo) = y j - l , F j ( x N , y N ) = y j , j = (ii) ]Fj(x,y') -- f j ( x , y " ) l ~ q l y ' - - y " l ,


(i)

1,2 . . . . . N ,

V y',y"E [-a,b],O~q~l.
j= 1,2 . . . . . N ,

(1.2)

Wj(x,y) = (Li(x),Fj(x,y)),
( I F S ) corresponding with the set { (x~,yj) }~.

(1.3)

the system {K ; W j ' j = 1,2 . . . . N} is called a fractal interpolation iterated function s y s t e m

Barnsley tll proved that such IFS has a unique attractor G, which is the graph of some continuous function f ( x ) . I - - ~ [ a , b ] , with f ( x j ) : - y ~ , j = O , 1 . . . . . N. T h a t is G = { ( x , a fractal interpolation func-

f(x) IxEI}.

In the following we will call the function f ( x )

tion (abbreviated F I F ) associated with Received : 1997-09-15. 1991 MR Subject Classification..28A80,41A05. Keywords :Functional equation, fractal interpolation, approximation. The project supported by Zhejiang Provincial Natural Science Foundation of China.

No. 1

Sha Zhen

FRACTAL INTERPOLATION

FUNCTION

91

!
tional equation : L(r

Wj(x,y),j

= 1,2 . . . . . N , 0,1 . . . . . N. }.

{ ( x , , y j ) IJ

Here is the outline of this paper. In w 2 and w 3, we introduce and s t u d y a new func-

~-]~aj(o~)L(r
i=1

-4- f ( r

( i . 4)

In w 4, we will note that a kind of F I F satisfy the functional equation (~. 4 ) , and f r o m (1. 4) we construct a n o t h e r functional equations called truncated equations. mate between F I F and interpolated function. T h e i r solutions are also to a p p r o x i m a t e F I F arbitrarily. In Section 5, we will give a new a p p r o x i m a t e esti-

w2

A Class of Functional Equation

First we introduce some notations. Define Code space ~ :


g] = {~ = ( i t , i z . . . . . ik . . . . ) [i, E ( 0 , 1 . . . . . N -

1)}.

for o~= (il ,iz . . . . ) and ~o----(;~,~2 . . . . ) E g2, we introduce the distance function
I

Io, -

?ol =

~,,
h=l

N*
ik . . . .

T h e n (2 i s a compact metric space. For w = ( i l , i z . . . . .

)Eg2,if ik=O,k~n+l, by (it . . . . .


i,,,N--1).

we Let

will denote it by (i~ . . . . i . , 0 ) , similarly for i ~ = N - - l , k ~ n + l , a'2" denote the subset of I"2: O" = {(it . . . . . i . , 0 ) or (i~ . . . . .
i.,N--

1)ln~

1,i, E ( 0 , 1 . . . . . N - -

1)}.

T h e Shift m a p a:O--~g'2 is defined by


aw = (i2 ,i~ . . . . ) , if ~o ---- (il ,i2 ,is . . . . ).

So we have ~ w = ( i t + t , i ; , + z . . . . ). We define map r

r
n--1

h=l

~. N

By the ergodic t h e o r y , for almost every point x E [-0,1) (corresponding to oJ) we have

i=0

where X~ denotes the characteristic function of E and r e ( E ) denotes the Lebesgue m e a s u r e of E. T h u s a l m o s t all set {tb.,~}7~ is dense e v e r y w h e r e in l-0,1]. Let C 0 [ - 0 , 1 ] = { f E C [ - 0 , 1 ] I f ( 0 ) = f ( 1 ) = 0 } ,

Ilfll= xE[O,l] max If(x)I.

It is not difficult to

prDve the following lemma.


Lemma 2. 1.
Let
V(r = ~,ajL(r
]=1

-4- f(tb~.),

where L , f E C o ~ O , 1 ]

and ay is real constant. If ~ ]


j=l

lail<oo,

then V is a single valued

92

Appl. Math. - J C U

Vol. 14,Set. B

function on [0,1-], and VECo[-O,1-]. Now we consider the following functional equations in C0[0,1-]: L(ib~) = ~-]ajL(r
i=1

+ fQb.),

f E Col0,1-].

(2.1)

Set 'AI "= a l , Az = alAl + az,


9 ~ .

(2. 2)

.Am-i = alAm-2 + a2A=-~ - t - . . . + a=-l,


Theorem 2. 1.

If

j=l

~ lajl<ooand ~ IAil<oo,then
j~l

equation (2. 1) has a unique solu-

tion L . LQb,) = f(r


Proof.

+ ~-]Akf(~b2,~).
h~l

(2. 3)

and f ( r

For simplicity, when ~o= (i~ . . . . . i~,()) El'/" , we will denote ~,. by [i~ . . . . . i , ] , by f [ i l . . . . . i~].
1--1

Using induction, if L is a solution of equation ( 2 . 1 ) , then we will have: L[i~ . . . . . i~-] = f [ i l . . . . . ik] + Z A j f [ i i + ,
2/=1

. . . . . ik-].

(2.4)

In fact, obviously we have L [ i l ] = f[il-], L[il,iz-] = fT_il ,i2] + a~f[i2]. then Now suppose that the formula ( 2 . 4 ) is true for k ~ m , L[i~ . . . . . i . ,i=+1] =
ra

f[i~ . . . . . i..+1] + ~ a i L [ i j + l . . . . . i,,+,-] ---j=l


m

m--j

f[ia . . . . . i=+x-] + ~ a i { f [ i j + l
J =1

. . . . . i.+~-] + ~-]A,f[ii~_,. 1. . . . . i...~-]}=


i=1 j~ -1

j* =1

j=l

f [ i l . . . . . i.+~] :?
1=1

A j f [ i ,+ ~. . . . . i=+1-].

This leads to ( 2 . 4 ) . Set


m--I

~,Ajf[ii_,
j=l

. . . . . i . ] -j--1

A j f (~bd~) =

r,,,/z]

~ ] A, {f[ij+~ . . . . . i . ] -- f Q b d . ) } +
j--1

Ajf[ij+1 . . . . . i . ] -j=[m/2]+ l j=Em/2]+ l

A J Q b d . ) ~ I~ + I , -

13.

No. 1 Sha Zhen

F R A C T A L INTERPOLATION FUNCTION

93

By assumption, it is easy to see that I ~ - ~ 0 , j = 1 , 2 , 3 , proved.

when m--~c~. Thus (2. 3) is

On the other hand, it is easy to verify that the function L, defined by ( 2 . 3 ) , is really a solution of equation (2. 1). The theorem has been completely shown. Remark. Note that, in the Theorem 2. 1, two conditions ~
j=l

[a~ I ' ~ ~ 1 7 and 6 ~


j=l

IAj [ < o o

are independent with respect to each other in general. Corollary 1. Corollary 2. Let a j = ( - - 1 ) J - ~ d ~, I d l < l , Let a i = - - d j, I d l < l , then equation (2. 1) has a unique solution: (2.5) (2.6) L(~b.) = f(~b.) + df(~b,.). then equation (2.1) has a unique solution: L(~b,) = f(~b,) -- df(~b,.). Given {d,}o u-a,

Now we discuss the other case in which the coefficients {a~} in (2. 1) are relative to co.

I & l < l , and o~=(ia,iz . . . . ) E l ] . Set


d~i) -- dq 9 dq. . . di .

We consider the following functional equation in C0[0,1]: L(~b,) = ~d,~j~L(~bg~) + f G b , ) .


j~l

(2.7)

Using the similar method, we have (the details are omitted).


Theorem 2 . 2 .

If d =

O~i~N--

max

{Id, I } ~ 1 / 2 , then the equation (2. 7) has a unique solution:


(2. 8)
k=l

Theorem 2 . 3 .

If [ d ~ ] % 1 , i = 0 , 1 . . . . . N - - l , then equation L(~b.) = - - ~d.(i)L(~b.,~) + f(~b~),


j=l

(2. 9)

has a unique solution: L(~b~,) = Theorem 2 . 4 . If d =


o<i<~N- 1

d.,~f(~b,.) + f(~b.).

(2. 10)

max { Id, I } ~ 1 / 2 , L is a solution of equation (2.7) and f E L i p a ,

then we have

.{o( llog t),

,/o

i
(2. 11)

[ 0 ( ~ ) , i f a ~ fl, 7 = min(a,/~ , where o~(L;~) denotes the modulus of continuity of L, and f l = l o g 2 d / l o g ( 1 / N ) . Proof. Given 0 % 8 % 1 , then there is a nonnegative interger m, such that 1 / N "+1 ~ ~ ~ 1 / N ' . Setting oJ= ( i l , i 2 , . . . ) , w r= (i'm,irz . . . . ) E ~ , and i ~ = i ' , , l ~ k ~ m ,
m

form ( 2 . 8 ) , we have

L(~b,.)

L ( ~ a ) =f(~b~) -- f(~ba) + ~--]2*-ld~(,)[-f(~b~D -- fGb,aa)-] +

l~m+l

k~m-t- 1

94 and IL(r --L(r

Appl. Math.-JCU

Vol. 14,Ser. B

~ K ~ ? O + K ~ 2 k _ l d ,. N ~"1 k~. + 2IISH 9


k=l k=m+l

2k-'d k.

Notice that 2 d = N -~ and using the similar method as showed in[-2] we can complete the proof of the theorem. Next we consider more general functional equation as follows

L(r

= ~aj(oa)f(r
j=l

+ f(r

(2.12)

Suppose that for any j , the aj(oa) is a single valued function on ~ , and aj(co)=a1(~o) for any oa=(ij . . . . . i . , D ) , ~ o = ( i 1. . . . . i . - - 1 , N - - 1 ) E ~ ' . "A1 (oJ) = al (oD, Setting (2.13)

~-1 A~(o~) = ~]ai(o~)A. ~(o~w) + a.,(oo),


i~l

m > 1,

We have the following theorem. Theorem 2.5. Let a i ( w ) E C ( ~ ) , V j and f 0 f f C 0 [ - 0 , 1 ] . Both series ~
/=1

la~(o~)l and

~2] ]Aj(w) [ are uniformly convergent on J~, then the equation (2. 12) has the following
)--1

solution in C0[-0,1] : L(r = f(r + ~Ai(w)f(r162


9 j=l

(2.14)

w3

Example

As an example we discuss a special equation: L(~&) = d L ( G ~ ) -+- dZL(r

~ f(r

f E C0[O,

l],d ~ O,

(3.1)

We assert that the following three statements are true: (1) If O ( d ~

"r

'

then the equation (3.1) has a unique solution:

(3.2) (2) If O ~ d ~ - 2 and f E L i p a , then we have

( 0 ( 6 ~]log31), ~o(L;3) = ~ (O(3r), where fl = log/f d . i:f

if

a = fl

a :/~ fl, )' = m i n ( a , f l ) ,

(3.3)

1 -+- 4 ~ - ] ' / / l o g ( 1 / N ) . 2

(3.4)

(3) Suppose that the function f ( x ) in (3.1) is an " h a t " function

No. 1

Sha Zhen

FRACTAI. INTERPOLATION FUNCTION Ix, ~l--x, 0 ~< x ~< 1/2, 1/2<x-..<1.

95

f(x)
Simultaneously, set N = 2 and

(3. 5)

,/-g- - !
4

<d<

J-g-2

(3.6) w h e r e / 3 is defined b y (3.

then the Box-dimension of graph of L . dimn g r a p h L = 2 - - f l ~ l , 4). In fact, for equation ( 3 . 1 ) we have A a = d , A z = 2 d

2, and Ak=a~d k, where ak is some

constant. Using ( 2 . 2 ) we can obtain a different equation about a~.


~m ~-- r 3 " am--2.

By a simple calculation, we have

_ ,/-5~" 2 ,/-g-

3(1
z )

"' 32

+ 2

,/-r)" 1

' ,m = 1,2 .....

By T h e o r e m 2 . 1 , then (1) is proved. The proof of (2) is similar to that of T h e o r e m 2 . 4 , so we omit the details. N o w we prove (3). By assumption and ( 2 ) , we have 0 < f l < l dimBgraph Note L -~ 2 -- ft. and L E L i p / 3 . Thus (3.7)

f[-i.,.
and

,i,.,13 -- f~i~.

"

"

,i,.,O~

- -

(~

- h + Z

'

L[-i,
'''" '

i=,l~ -- L~i~
'''"

,ira,03 -- ( 2~,+~ 3" E a k d '


k=l

--

1)q

m--I

(--

~--7~

1 ) i~+1

3" a~,d" 9 -2"

F r o m ( 3 . 6 ) we know 1 < ( 1 3 . - / 5 - ) d < 2 teger ~, such that 2 -- (1 3. ~ ) d

and ( ,/--if-- 1 ) d < l .

T h u s there is a positive in-

-- 2[-(1 3" "/-5-)d-] -~ > 0.

For a fixed v and m > v , set L,-~=i,~-,+I = . 9 9 = i , , ~=L,. By calculation, we can find a constant K , such that

IL[-i,, .

. , i ,., 1 ~ . -. L[-ii . .

, i , , 0 ~ l >~ K 9 [(1 + T' ~/-g-)d~ m, if m is large enough 9 which can cover L is larger than

It is clear that the number of sets of square with side ~

K 9 2--~ 9 [-(1 3. 4'-5-)d~ m. Thus dimagraph L ~ lim l o g K 9 2~-~ 9 [-(1 3" 4 ' - 5 - ) d ] " log(1 3" ~ - 5 - ) d .... log2,~+1 ----- 1 3" log2 = 2 -- ft. Combining it with ( 3 . 7 ) , (3) is proved.

96

Appl. M a t h . - J C U

Vol. 14,Ser. B

w4 Lemma 4. 1.
If two f j ( x ) ( j = l , 2 )

Approximation of FIF
are the F I F associated with (x)), n = 1,2 . . . . . N ; j = 1,2,

(x,y) = (L.(x),d.y+q.

( { ( x , , y ~ j))li = 0,1 . . . . . N ; j =
respectively, then f l

1,2}, n = 1,2 . . . . . N ,

are the F I F associated with

.~W . ( x , y )

I (x" -= " ~.. ), d . y - v - q . <l)--t-q.(2)),

[ { ( x , , y ( 1 ) y~Z))li = 0,1 . . . . . N } ,
respectively.

Lemma 4. 2.

Let l ( x )

be a segment which connects two points (x0,y0) and ( x N , y u ) .

9T h e n l ( x ) is also a F I F associated with

t
Corollary.

I~.(x,y)

= (L.(x),d.y+a.x+b.),
,N},

n = 1,2 . . . . . N ,

{ ( x , , l ( x , ) ) [i = 0 , 1 , . . .

where a.,/~, are some suitable constants being dependent on {d.}. If f ( x ) is the F I F associated with

~W2(x,y) L{(x,,yi)[i W.(x,y)

= (L.(x),d.y-~-q.(x)),
0,1, .... N

n---- 1,2 . . . . . N ,

},

l ( x ) and B~. are showed in Lemma 4.2. T h e n ( f - - l ) E C o [ - 0 , 1 - ] is the F I F associated with =- ( L ~ ( x ) , d . y + q . ( x ) -- a . x -- b . ) , n = 1,2 . . . . . N.
T h u s in the following we can suppose that Y o = y N = O (without loss of generality). N o w we consider the following I F S .

W.(x,y)
where

= (L.(x),F.(x,y)),n

= 1,2 . . . . , N ,

(4.1)

I
and 0 ~ , d . _ , [ < l , Assume f(x) ~,y~
l

L.(x)=

. n--1

LF.(x,y)

= d._~y -k- h ( L . ( x ) ),

V n,hECo[O,1-],h( N)=Y~,l~i~N--1. is the F I F associated with (4. 1 ) , it passes t h o u g h the points ( 0 , 0 ) ,

and ( 1 , 0 ) . F r o m [-2-], we have f(r ---- h ( r +

~d,~<k)h(~b2,~).

(4.2) T h e n it is a solution of the

Theorem 4. 1.

Let f ( x )

be 9

F I F associated with ( 4 . 1 ) .

following functional equation:

LQb,~) ~- = ~dwu>L(r
where t$(~b~)=h(~b~) +
j=l

+ ~(r

(4.3)

1 ~d,,,(j)h ( r

No. 1

Sha Zhen

FRACTAL INTERPOLATION FUNCTION

97

From equation ( 4 . 3 ) , we consider the following functional equation, called truncated equation : L~(r
T h e o r e m 4. 2.

j=l ~d.~j)L~(~b~) +

h(r

+ ~

i=1 ~1 --d~i~h(r

(4. 4)

Let f(2) be the F I F associated with (4. 1), L= be the solution of equation
i

( 4 . 4 ) . S e t t i n g d - - - - m a x [ d ~ [ < l , then we have [If


--

2--d g~,[[ ~ (1 - - d ) 2

d~
" " 2+ - - - ~ "

Ilhll.

(4. 5)

The proofs of Theorems 4.1 and 4.2 are not difficult, so we omit the details.

w5 L e m m a 5. 1. [z]

A N e w A p p r o x i m a t e Estimate o f FIF

Let

f(x)

be the FIF associated with ( 4 . 1 ) . If h ( x ) E L i p l , o~(f,3) = 0 ( ~ ) ,

then we have (5.1)

where log(max [d, ])


Ot

lo (1)
f(x)
be the FIF associated with ( 4 . 1 ) , where h E ~-~ti~ E (0,1 . . . . . N - 1) . (5.2)

Theorem 5.1.

Given r

Let

C0[-0,1],hELipl,

h( N) =~( N) , l ~ i ~ N - - l and O~[d,[~l,V i. Set K(N~n)


-k=

Suppose that L" is a solution of the following equation: L(r Setting = 2h(r + ~Td.(j~ LQb./.). (5.3)

e(N;n) =
then we have [r

max{ IL" (x) - r

h(x) l lx E K(N,n)}, K ) + N-~,


such that

(5.4)

-- f ( x ) [ <~ e(N;n)

w(#,

ifm~n, Ix--x"

(5.5)

where the eonstant K is independent of n. Proof. Let x E [ 0 , 1 J , then there is a x" E K ( N ; n ) I~... And let ~"

be the solution of the following equation: L(~b.) = 2hQb,) +


.=

~d~j)L(r ~-]d~)h(~.).
j=l

From Theorem 2 . 2 , we have ~" (~b.) = 2h(~.) +

98 Using (4.2), Therefore, we o b t a i n L ' = f + h .

Appl. Math. -JCU


Whenm~n, we h a v e L ' ( x ' ) = L ' ( x ' ) ,

Vol. 14,Ser. B

--f(x) r
(r

= -- r
-- r

) A- ( r
) ) A- ( r

-4- h ( x " )) -- ( f ( x " )


+ h(x') we have

+ h ( x " )) -4- f ( x " ) -- f ( x ) -- f ( x ) ) .

----

-- L" ( x " ) ) -4- ( f ( x " )

U s i n g Lemma 5.1 and ( 5 . 4 ) ,

I~(x)T h e T h e o r e m is p r o v e d .

f(x),

~to(~;~,)+e(N;n)

d- NK----~.

References
1 2 3 Barnsley, M.F. , Fractal functions and interpolation, Constructive Approximation, 1986,2:303~329. Sha Zhen, Holder property of fractal interpolation function, Approx. Theory Appl. , 1992,4:45~57. Sha Zhen ~. Chen Gang. , Haar expansions of a class of fractal interpolation function and their logical derivatives, Approx. Theory Appl. , 1993,4:73~88.

Dept. of Appl. Math. , Zhejiang Univ. , Hangzhou 310027.

You might also like