lnternat. J. Math. & Math. Sci. VOL. 16, NO. 4 (1993) 737-748 737 AN ORTHONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET RAAFAT RIAD RIZKALLA Mathematics Department, Faculty of Education Atn Shams University, Heltopolts, Cairo, Egypt (Received January 16, 1992 and in revised form June 22, 1992) ABSTRACT. This paper presents a new complete orthonormal system of functions [0,1] and whose supports shrink to nothlng. Thls system related to the construction of the Cantor ternary set. We defined the canonical map and proved the equivalence between this system and the Walsh defined on the interval ystem. The generalized Cantor set with any dissection ratio is established and the constructed system Is defined in the general case. KEY WORD AND PIStASES. Dyadic expansion, Walsh functlons, Cantor set. 1992 Al SU CLASSIFICATION CODE. 42C10 I. INTRODUCTION. We shall denote the set of positlve integers by P, the set of non-negatlve integers by N, the set of real numbers by R, and the set of dyadlc rationals in the unit interval [0,1} by q. In partlcular, each element of Q has the form n p/Z for some p,n e N, 0 Each n e N has a unique dyadic expansion p < Z n. nk 2k nke {0,1} k=O are called the dyadic coefficients of n. Likewise n E where n k a {unique} dyadic expansion x {1.1} ech x [0,1} has z-(k+1) Z (1.2) xk Xke {0,1} k=O the finite expansion being chosen in case x belongs to the dyadic rationals q. In terms of the dyadic expansions, the dyadic sum of two numbers x,y [0,1} is defined by x $ y (R) Yk Integers The dyadic sum of a pair of n E k=O m 2 -(k+l] [1.3) n,m e N is defined by nk mk 2-k (1.4) k=O where (n ,k e N) and (m ,k e N) are the dyadic coefficients of n and m. k k Let r be the function deflned on [0, I) by x e r(x) -1 x e [0,) [,1} extended to R by periodicity of period 1. The Rademacher system (r ,n n N) is R.R. RIZKALLA 738 rn(X) defined by, r(Z n N. The first three Rademacher functions R,n x), x are shown In Flgure I. r o 1/4 112 -1 z 314 Fig. 1. The Rademcher functions form an incomplete set of orthonormal functlons on [0,I). The Walsh system w was introduced by Paley [11] in 193Z N n as products of Rademacher functlons In the followlng way. n coefficients k N Wn Walsh functlons If n N has dyadic then to belong (1.5) rk -kTO the class of plecewlse constant basis functions that have been developed In the twentleth century and have played an Important role in sclentlflc appllcatlons. The foundations of the Walsh functlons field were made by Rademacher [IZ], Walsh [16], Flne [10], and Paley [11]. Owlng to their sallent propertles, Walsh functions proved to be very powerful In solvlng varlous problems such as the analysis of dynamic systems [4-8], the deslsn of optlmal controllers [3], and the Identlflcatlon problem systems [Z],[14]. The flrst elght Walsh functlons are shown in Flgure Z. of dynamlc w0 w w 4 5 w 6 w? +1 or +1 ill 1/2 or 3/4 -1 114 112 314 -1 Fig. 2. Notlce by deflnltlon that Wzn Wo= r n for n e N and each Walsh dlscontlnultles on function Is piecewlse constant wlth flnltely many Jump on [0,1) and takes only the values +I or -I. The Walsh system Is orthonormal [0,1) and possess the properties (see [13]), w w w (1.6) Wn(X y) Wn(X Wn(Y (1.7) n m AN ORTHONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET 739 By a dyadic interval in [0,1) p we shall mean an interval of the form p+l [--n I(p,n) n p < 2 0 n 2 [0,1) and n N n p 2 For each x length 2 -n N we shall denote the dyadlc Interval of n I (x) I(p,n) where 0 -< p < 2 Is uniquely whlch contalns x by n x determined by the relatlonshlp I(p,n). The Walsh functions, belng a complete orthonormal system (see [I],[16]) allows a representatlon of every absolutely Integrable function f(x) on [0, I) In Walsh serles In the form where a w (x) n n--O n being the Walsh-Fourler coefflclents of f. f(x) J n f(x) Wn(X) dx sum of Walsh The n-th partial DIrlchlet kernel are denoted respectively by [15] n-I n-1 and f series and the Sn D n (x) 2 Dn k=O ak wk Flne[10], has shown that satlsfles a 2 n Snf(X) Dn(X f(t) kO wk $ y) dt The Dlrlchlet kernel [0,2 -n) and zero elsewhere. for x e I (0) n THE CANTOR SET AND THE CANTOR f3JNCTION The Cantor ternary set C formed by removing middle thirds from the interval [0,1] can be defined as follows for each n e N, construct the 2. closed intervals 2n/k where k 2k +I 2kn j 3 T. k I=0 i (2 I) is defined by k k n k < 2 0 n 3 I 31 (2.2) belng the dyadic coefficients of k In {I. I}. of ratio I/3 Then the Cantor ternary set or more briefly the Cantor set C Is (R) C= 0 n=O 2n-1 U J 2n/k k=O The middle open Intervals removed In the above construction are defined by E U 2n-1 U n= 0 k=O E where 2n+k 6k’+ 3n+1 6k +I E 3n+l 2n+k Arlthmetlcally, the Cantor set consists exactly of those points which can be expanded in the ternary system without using the digit x x k=O k 3 -(k+1) Xke We deflne a class of closed intervals n :0 k < 2 J ,i.e. for every x e C {0,2) (2.3) n e N -n. n The Cantor N, n contains 2 closed Interval each of length 3 For each n set can be associated with a monotone non-decreaslng continuous function Lebesgue function and defined by the followlng process. called the Cantor For each closed interval I, let IRI(X and I Xl(t) dt where xi(t) {I 0 tI I the length of the and function characteristic represents the O 740 R.R. RIZKALLA interval I, respectively. N, let For each n Fn(X) 2 -n 2n-I {2.4) Rj ,i(x) k=O k Since X (t) Rj .(x) /k J2n*k Jz".k 0 2 x 3n[ OJ dt X (t) dt Jzn/k x < 2k .3 0 3nx xJn 2k 2 /k x > {2k +I}. 3-n Hence, from (2.4), { Fn(X) F0 Fo(x) x 2k’+ 2-nl3nx x e x e k) (2k+I).2 -m x [0,1] J1 Osk<2 Jzn+k n m-I 0 s k < 2 E2m-l/& P and for n s m s n {x),,J F (x) / F2 (x) _.__/- f / F3 (x) t’-/----/0 0 Flg.3. We observe that, the function F {x} is continuous, nonn decreasing on [0,1],and satisfies Fn{O} O, Fn{1} 1 {see figure 3}. Indei, 2k+1 Fn(X lira N, for each n 2 It is easy to see that m F(x) x E2m-l/k 0 m-I k < 2 1 < = [0, I] exists, is continuous, llm Fn (x), x nm I. O, F{1} F{O} monotone non-decreaslng and satisfies This function is called the Cantor-Lebesgue function associated with the Cantor set C. (x) =r. Xk.2 k=O where Xk xk [0,1) C Now, we introduce the mapping -(k+l) {o 1 x if c (2.s) 5= o Xk= 2, being the coefficients in the ternary expansion of x in {2.3}. The mapping satisfies the following properties: 1- is one-to-one mapping from contalns all x C/C onto [0, I) where is a countable set C whlch has ternary expansion termlnates in 2’s. In fact, C ={ 2k+1 n 3 Oak<2 n neN }. - AN ORTHONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET n N, 0 -< k < 2 2- For each n x f] k Define the Lebesgue-stleltjes measure 2 n n n which contains y. F(a) F() N n where F Is (k__ n 2 AN ORTHONORMAL SYSTEM OF FUNCTIONS 3. We construct a function system for each x In the followlng way n k I -n [a,) F 2-n F Jn+k 2 0 -< k < 2 the Cantor function. Indeed, for each k In (x) Iff J2n/k and 2n Is the dyadic interval of length In(Y) where C/C 2k" ( 741 21 k I=0 I @ |0,1) nand nne NN, on the Interval [0,I) n+1 e 0 s k < 2 define k (-1) { n(X o x e Ja 0 /k (3.1) otherwise has the constant value +1 on half of the closed nwhlle has constant value -1. on the other half n n+l (see figure 4). Consequently, is odd function about t It Is obvlous that l,ntervals Indeed, in n 1/2 [ n dF O N n O o nu 13 FI U u +1 o| -1L F! u l,, ’/, 2/9 7/9 6/9 3/9 e/9 FIg. 4. THEOREM 1. measure F The system n: n e N Is orthonormal with. respect to the l.e. n m dF nm (0 (3.21 n m. PR(X)F. Let n > m z 0 and J e Slnce Pm has constant value +1 or -I on J, and J contains an even number {namely 2n-m of closed intervals in on half of these intervals Pn has the constant value +1 whlle on the other half has constant value -I, consequently, m+l" m+l n Summing over all J &+l we get J n dF nmd Jean+ Since {J} interval, then 2 -(n+l) V J e +1 1 n d@=l z 0 O. O. If n m, then Je+ n e N Mn+1 d n contains 2 n+l c 1 osed R.R. RIZKALLA 742 For each x COROLLARY 1. (-11 n n(X) where r N Let n PROOF. k (-1) 0 then J2n+l X (-I) n rn, r g 2- n x )(x) n k=O k n C/C On Now, we define the product system N n Since ko" (-1) r ((x)) n )(x) From the N J2"/l/k n k In+l( and /k (C/C) x and (x) we have x n(X) n+l k < 2 0 properties of the mapping &0n(X) (3.3) )(x) rnO be the Rademacher system. N n n N n C/C xn N. by N n N) are the dyadic coefficients of n. where (n ,k k lf3 l-I M U I-i U i-I U U +1 o| |,l -1 [ 6/9 3/9 2/9 /9 ?/9 8/9 Fig. 5. The function system I satisfies the following properties: For each n,m e N, 2. For each n,m to and @n @m @n 2n+k, N, m @2 n n closed intervals in 0 < k < 2 n, @m an+1 2n+I’ @m is even function with respect {see figure S). respect to is odd function with 3. For each n,m N, 2n s m < is the dyadic sum. where m has constant value +! on half of the while on the other half @m has constant value -1, consequently, @m dF n m < 2 0 2n+l n N. In general, j 4. @n dF P n 0 Let n e N with dyadic coefficients n k k (3.5) N equations (3.3),{3.4), we have Cn(X) {-!) nkXk {-1) <n,Cx)> k=O where {n,(x}) Z nk k=O x k (rood 2} and x C/C From AN ORTHONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET For each x COROLLARY 2. @n(X) w where (3.6) )(x) (WnO be the Walsh system. N n n N n C/C From equations (1.5),(3.3) and (3.4), we have PROOF. @n(X) (k(X)) k=O ((rko k=O -- )(x)) Wn((x) (rk({(x))) k=O )(x). w n According to corollary 2 and the properties of the mapping obvious that 2k" n TIREM 2. measure fF J let First, let O, n,m e g 2 n aEOREM 3. it {3.7} n e N 0 ( (3.8) Since it suffices to show that i’J i(R)J Second, in case i which is clear from {3.5}. n+1 is is orthonormal with respect to the N n J, then dg Jean+ F{J) F 2 n+1. 2 -(n+l) Jean+ The system Is complete, with respect to n e N @n on the Cantor set C. F PROOF. iJ. m < 2 Om dF the measure n’ , ijd N, m n 0 s k < 2 k Wn( )= The system i.e. PROOF. J Cm 743 Let n,m e N n m < 2 0 and suppose that S is Integrable functlon for which Slnce each function ,In 2 /k Let O r @m dF 0 s m < 2 @ n 0 s m < 2 0 n (3.9) has only one flxed value on each Interval -(2k +1).3 -n r(k) fnnJH J Then {3.9} glves, Y J dF k .3 /k 2n-1 #m dPF k=O r(k) n -n dF’ m 0 -< k < 2n" n 0 s m < 2 0 From {3.7), we get @m( then 2k k w)m 0 -< k < 2 2n 2n_l r. k=O Now, the determlnant (k)n win( r m( w k k n, 0 )l n 0 -< k < 2 and does not vanlsh. Thus the numbers w k m m < 2 0 n, n 0 s m < 2 0 n m < 2 n e N, {3.10} is of order 2 n are llnearly independent and R.R. RIZKALLA 744 it follows from (3.10) that rn(k} 0 -< k < Z 0 FSd continuous function G(x) G( Uslng (3.11), we get F 0 _2k +I_ 2k--= G( n (3.11) N and let G(x) denote the m Suppose that {3.9) is satisfied for a11 n 0 -< k < 2 n n 3 3 is equivalent to zero on the and so G(x} is a constant on C. It follows that Cantor set C. Thus the proof is completed. [0,1]. The Let f be an absolutely lntegrable function on the interval is defined N n Fourier series of f with respect to the system #n by S a f n=O _n n where a [ n f(x) 0 The n-th partial sums of @n(X) fF n-1 a,. #k _’. _.Qnf this series will be denoted by k=O P Notice for n N. n [0,1] and x n-1 {Snf)(x) (j f(t),k(t)dF(t)),k k=O whe r e (x) =J f(t)Dn(x,t)dF(t) n- (x) k=O denote to the n-th Dirichlet kernel. k Dn(x,t)= COROLLARY 3. N Let n (3.12} k (t) C/C" x,t n-1 (I) Dzn(X,t) I] Dzn(X, t) ..n + i=0 II (t) J x Jn /k x,t N and (I) CC 2n-1 I=0 . Fix ,2n-1 0,1 I n (k.2-n). Dlrlchlet kernel }, + I=0 Wl((x) l(X) l(t) the the properties of 2k +1 n 3 and 2kn J 2n+k Using ). 3 mapping the in the Walsh case, we get D2n D2n(X,t) x and let 2n-1 THEORE 4. I=0 n-1 k (x) then I=0 Wl((x) wi((t) ) rl((x)) rl((t)) + i=0 (II) 2n-1 2n_1 @i(x) Oi(t) n-I [l k < Using (1.7), (3.6), and (3.12), we have Dzn(X,t) 0 for some k 2 /k 2 Let n PROOF. llx)i(t) i=O 2n (t)) wi((x) 2n ((t)) I n (k.2-") Let f be a continuous function on CC (t) J z then 2 /k n f converges uniformly to f PROOF. n n where k (x) For every CC x and n N we define 2k +1 <- x < n n 3 3 is defined in (2.2). By the canonical map and n n(X) by maps onto the dyadic interval 2k n In(k.2-n) and F([=n n for some k the set n )) (CC)N 2 -n n k < -2 0 [an, n Then the AN ORTHONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET function f 745 can be written as f(x) 2 n [ f{x) dF{t) (3.13) n Using corollary 3 (S2n f)(x) j n 2nj From 2no O2n(X,t} dF(t) f(t) f(t} < e whenever f(x} (S2n f)(x} (t} [e n ,gn dF(t) dF n {3.13} and since f is continuous, i.e f{t} f(t) t x f(x} V < 8 > 0 8 > 0 3 ,n }. [an x,t e n 2n [ (f(t} n n If(t) 2nj f(x)) f(x} Therefore, dF{t) < e dF{tl n and the proof is completed. M CANTOR SET ITH CONSTANT DISSECTION RATIO 4. in (3.1} can be descrlbe on any Cantor set The constructed system with constant dlssectlon ratlo. The generatlng intervals of the Cantor set of general ratio 0 < p < q p/q 2 /k are deflned by 2n.k ] I n O<k<2 n 2q 2q are determlned by the recurslve formula 2n/k where t (q-p)n [kn P/q p,q e P 1-k A t and {q_p) n-1 + )t 2n-l/k 2n+k n (q+p)(2q) n-1 k being the dyadic coefficients of k in {I.I). It is easy to see that k p/q e N. Then the Cantor set C for each C C/q P/q m intervals m (l) J.+ 2 (ili) p/q fl O n=O k=O of each J 2n+k closed a a 2 +k 2 n+l The closed 2n+k P/q P/q .=o, J.+ (li) 2n+k +2k+s p/q interval satisfies the following four properties: J c n Is of ratlo p/q P/q P/q =2 2 -1 end-polnts contains the R2n 2 2 2 +2k 0 s k < 2 for n+l 2r+k /2k J.+ fl 2k*! n = N, and n /2k+1 P/q (iv) llm J2n+k ma Ok<2 -) 0 . The open Intervals removed In this construction are defined by 2 p/q E U p/q p/q U k= 0 E where 2n+k 2 +k P/q For each N, the class J I 2 n+ 2q 0 2 n+l n+ P/q { J2n/k k .2k k < 2 n (q_p)n+l {2q)n+1 .2k.! contains 2 n closed 2 -1 interval each of length q-P -q and Fn (x) 2 -n R k=O p/q(x) ’J2n/ k where R.R. RIZKALLA 746 (t) dt ;[ J2n+ 0 P/q J2n, k k t x < 0 "{q-pa /k n n (2q) p/q 2 +k + (q_p)n / > X Hence, I Fn(X) " x 2n+k (2q) n p/q 2 +k (q_p)n 2 +k p/q xEEm_ 2 2k +___I m 2 0 -< k < 2 non-decreaslng on [0,1]. In fact assoclated with the Cantor set each F n /q n - is linear on the lhervals n U x e J 2n+k 0 k < E 2 and 2 .k k =0 is Jn p/q P/q 2k+l m 2 F (x) n m +k constant on each component of the set [0,1]/( llm m-1 (II), and (III) above that F is continuous and n p/q Observe by properties (I), F(x) O-<k<2n xe,J +l+k ). The Cantor function 2m-I m < +k P < then there is a continuous strictly increasing TIOIIE S. If C. function L on [0,1] such that L(/q) < we have k 2 lOOF. For each n e N, 0 n, P/q q-P -q where n 2q p < 1,then 0<1. Let L be the monotone-lncreaslng plecewlse llnear n Since 3 p/q function, L :[0, I]--[0 I] 0 s k < 2 n. mapping the end-polnts of Then for n>m, L n and L m differ only on I So, for each x m" 3 defines a function x) on o,}. As Ln Lm II/I > [P/q} 2n+k - L(/q) J2n+k k < 2 0 m, m Xlm {nX)-mX){ So for each n and k, 0 - In fact, the sequence ( L (x)> converges L Is continuous [0, I] and Lm converges uniformly to L [9]. monotone increasing. We have L([O,I]) L j J onto those of J 2m+k {tmx)-x){= < J and Is clearly n k < 2 We need only to show that L is one-to-one. Suppose c either [0,1]. If x or y lles in removed Interval it is easy to see (x,y) then as /q is So suppose that L(y) > L(x). x and y are in /q SO C. p/q nowhere dense there is an Interval L(y) L(x) J so n .k {2q} n for some n,k. But then is strictly increasing and the proof Is completed. L According to theorem 5, for each n L (x,y) 2n+k (q_p)n } k < 2 N, 0 , 2k 3 n and L n, 1 and ] s P < 1, {2q} n } 2k +I n 3 AN ORTfIONORMAL SYSTEM ON THE CONSTRUCTION OF THE GENERALIZED CANTOR SET set 0 747 Now, the constructed system @ in (3.1) can be generallzed for any Cantor [0,1), p,q P and n N, for each x Cp/q In the following way k < 2 n n+l, k l.e. k 2 define i=O (-1) n(X) k0 P/q x 2 otherwise 0 Is defined as in (3.4), it is orthonormal and complete on the general Cantor set Cp/. Ust system with respect to the measure for the theorem 5 and the mappt L, the results obtained in sect ton 3 The product system @ F ternary Cantor set C c be easily generalized for the case C REFERENCES 1. Convergence Problems of ALEXITS, G. New York Oxford Press, Pergamon Orthoonal Serf,s, Paris, 1961. 2. Time CHEN, C.F., and HSIAO, C.H. Pro@. IEE 122 {1975}, 565-570. 3. CHEN, C.F. ,and HSIAO, C.H. Control Via Walsh Functions, 596-602. Design Domain Synthesis Vla Walsh Functions, Of Plecewlse Constant Gains for Optimal IEEE Trans. Autom. Controll.. (1975), AC-.O 4. CHEN, C.F.,and HSIAO, C.H. A State Space Approach to Walsh Serles Solution of Linear Systems, Int___..J_.Syst. Scl. 6 {1975}, 833-858. 5. Analysls and Optimal Control of Time Varying CHEN, W.L., and SHIH, Y.P. Linear Systems Vla Walsh Functlons, Int. J. Control, 27. No.6 {1978}, 917-932. 6. CHEN,W.L., Equatlons, and SHIH,Y.P. Shlft Walsh Matrlx and Delay dlfferentlal IEEE Trans. Autom. ControlL AC-23 (1978), 1023-1028. 7. Walsh Series Analysis of Multi Cl, W.L. Inst. 313 (1982}, 207-217. 8. CORRINGTON, M.S. Walsh Functions, 9. Delay Systems, J. Fr’artk.l.l.n Solution of Dlfferentlal and Integral Euatlons with Trans. Circuit Theory CT-20 (1973}, 470-476. .IEEE DEBARRA, G. Introductlon t_9o Measure Theory, Van Nostrand Relnhold press, New York London, 1975. 10. FINE, N.J. 372-414. On The Walsh Functions, Trans. er. Math..Soc. 65 (1949), 11. PALEY, R.E.A.C. A Remarkable Series of Orthogonal Functions, Proc. Math. Soc. 34 {1932}, 241-279. Lond. Elnlge Stze Ober Relhen yon Allgemelnen Orthogonalen 12. Rademacher, H. Funktlonen, Math. Ann. 87 {1922}, 122-138. 13. SCHIPP, F., WADE, W., and PETER, S. Walsh Serles- An Introduction to the DYadic Harmonlc Analvsls, Adam Hllger England, 1990. 748 R.R. RIZKALLA 14. TZAFESTAS, S.G. Identification. Walsh Series Approach to Lumped and Distributed System J. Franklin Inst. 305 (1978), 199-220. Recent Developsents In the Theory of Walsh Series, Internat. J. Hath. &_ 4ath. 5. No.4 (1982), 62S-673. 15. WADE, W. R. 16. WALSH, J.L. A Closed Set of Normal Orthogonal Functions, Amer. J__. Hat.h, 45 (1923), 5-24.