57 Internat. J. Math. & Math. Sci. (1985) 57-68 Vol. 8 No. THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS CYRIL NASIM Department of Mathematics and Statistics The University of Calgary Calgary, Alberta T2N IN4 Canada (Received December 11, 1984) ABSTRACT. A theory of a generalized form of the chain transforms of order n is developed, and various properties of these are established including the Parseval relation. Most known cases of the standard theory are derived as special cases. Also a theory of self-reciprocal functions is given, based on these general chain transforms; and relations among various classes of self-reciprocal functions are established. KEY WORDS AND PHRASES. Fourier transform, ,hain trasforms, f’,eIZin transforr.Ts, Parseval theorem, kernel functions, L2-space, self-reciprocal functions, gamma ]’unction, Riemann-Zeta function, Bessel functions, Whittaker functions. 1980 ,Ar{.L’,’TTCS SUBJECT CLASSIFICATION CODE. 44A05 i. INTRODUCTION. The standard and the simplest definition of a function g to be the integral f transforms of with respect to k is that f(t)k(xt)dt g(x) where k is called the kernal, f(x) I] [1,VIII]. If further g(t)k(xt)dt, and g are pair of k-transforms. f then we say that (1.2) Under less stringent conditions the equations (I.i) and (1.2) can be replaced by p(t)dt I where f(t)dt k l(x) I I t-l f(t)kl (Xt)t (I .3) t-lg(t)kl(Xt)d/ (1.4) x k(t)dt The above reciprocity formulae define the pair transforms. f and g, genera|ly known as Watson lany generalizations of this reciprocal relation have been given, and C. NASIM 58 one of the directions was pointed out by Fox, [2], who introduced the idea of chain He showed that for chains of the form transforms. I g2(x) gl(U)kl(UX)du gn (x) gn-1 (U)kn-1 (ux)du gn(U)kn (ux)du gl (x) 0 2, i.e., the Watson there exists a theory similar to the standard theory for n Also, Duggal, [3], considered the pair of equations. transformations. ! v()f(x)dx z (.5) k()ff(z) u 1 0 using convolution of the functions f If V and 9. v are taken 5 f and characterizing the relatonship between and to be the Heavsde step functions, then (1.5) and (i.6) reduce to the equations sm[lar to the equations (1.1) and (1.2) above, thus extend- ng the Watson’s definition. In ths paper e shall combine th- above mentioned to extensions of the usuai e atson integral transformations. shall gve a generalized form of the chain trans- forms of order n and develop properties praliel to hose of the standard theory. Nost of the ell-kno results are deduced as special cases. s self-reciprocal functions again many standard results are deduced as special cases. re gven Further a theory of developed based on these generai chain transforms; and In the end various exapies to llustrate the general nature of the main theory developed and its special cases. 2. PRELININARY SLTS. Definition 2.1 [f();] s 1 Let n < Then define f()z- y() + g integral exists g2(0,). f < . F() s the mean-square sense. (.) f(); + g), said to be the Mell[n transform of If further P() I g2(_ i, the then (2.2) F(s)-Sds sense; we say that the integral existing in the mean-square 1 < [l,I]. < i, ellin transform of F(s), s The Parseval theorem, If f and g is the inverse =+ [l,III]. L2(O, ) (zt)(t)d f(x) , respectively, then and have Mellin transforms F(s) and G(s) -g ’ a’-s’-d (" ) THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS A function Definition 2.2. k if f is its own 59 is said to be self-reciprocal with respect to the kernel k-transform, satisfying the identity f(t)k(xt)dt, f(x) 0 We then say that or its equivalent. f Rk (2.4) Although most of the results hold for functions involved under less stringent conditions, but we shall mainly work in the Hilbert space L 2, for convenience and elegance of the theory of integral transforms in this setting. CHAIN TRANSFOrmS. 3. We consider the system of n integral equations ki(ux)/i+l (x)dx where i vi(ux)fi(x)dx n and f(x) 1,2 f(). For simplicity we shall assume that no two are equal, although it is not essential. We shall refer to the functions k. and v. Now, we give one of our major results. as kernels. Theorem 3.1. If (i) (ii) (iii) 7,’.i,vi,fi and ki(ux)fi+l (x)dx }(x 71 f() and n n V/(s) E S(s) i= uo then and fn+ (x)d P (x) uhere Vg() and fn+l 2- Kg() 1 + g, 0 1/2-i, Ki(s) L2(0,=), i all (ux)fi(x)dx, i n. 1,2 n where (3.1) L 1 (- , + i(R)), 1 (3.2) x-lP (u) (x)dx 1 P() x (. 3) d denote the Neiln transforms of < 1,2 . vg(z) PROOF. and kg(z) respectively on Due to the hypothesis () and the fact that g(0,) henever the ntegrals in (3.1) exist and are n fact absoluteIy convergent. the 2-functons n () have MellOn transforms h[ch belong to g, + g) due to definition (2.1) above. fi fi (0,), Also,1 2(1 No due to the Parsevai theorem for MellOn transforms of boh sides of (3.1), we have Ki(s)Fi+ 1(s) on s 1 we obtain + i, < (s)Fi(s =, i 1,2 a.e. n. 2-functtons, applied to (3.5) On successive elimination and using (3.2), C. NASIM 60 y (s) P()Yl(S) n+l Next we have M-l ,f-+l(x) F (8) x], n+l therefore [%+{" P(S)fl (s)x-Sds’ 1 fn+l (x) i. (3.6) ioo the integral converging in the mean-square sense, since I i F I(8) + fro). 0r from (3.6), we have i, 1 P( + iT) is bounded and L2( u F =I (x) P(s)F (s) u -s the integration inside the integral sign is justified because of uniform convergence due to hypotheses (i) and (iii). 1 1 + i ) and hence z, L2(- X - exists and Now since P(s) is bounded, therefore P(s) 1 8 Pl (x) w-[LlL2(0,). Also (s) fl(x) I[F l(l_s); U xl, therefore due to the Parseval theorem applied to the right-hand side of (3.7), we obtain u x- Ip (ux) fn+l (x)dx (x)dx as desired. DEFINITION 3.1. The sequence of functions {fi is said to be the general chain transform of order n, with respect to the kernels k the system (3.1) along with (3.3) is satisfied and vi, i 1,2,...,n, whenever i It is now an easy matter, to prove a stronger result than that of Theorem i, which we give below without proof. Let the conditions of (i) and (ii) of THEOREM i hold. THEOREM 3.2. Then necessary and sufficient condition that (3.3) holds is that v.(s) n P(s) 7 I-i =1 .gi () Next we shall give some special cases of the result of THEOREM i. I. First let n COROLLARY i. If i ()72 (X)C and V (2)i (X)C V (s) P(s) then f2 (x)dx where Pl (X) 1 x- (ux) [+i i (x)dx P(s)_ s xl-s ds. (3.8) THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS Thus if k 1,v and function If we further set 2" fl 0 v 61 are known, we have an inversion formula to retrlve the unknown if 0<x< i, (x) i if x> i we obtain a familiar unsymmetric transformation of COROLLARY 2. If u I fl I(R)u kl )f2 u (X)C Watson’s type [1,VIII]; that is (x)o and 1 P(s )K (s) 1 then Pl (U)fl (X)C f2 (X)OL Next let n 2. Then THEOREM 2, reduces to a known result [3], that is: If (i) and E L2(0,), i 1,2, COROLLARY 3. ki,fi f3 and then x-lpl (ux)fl (x)dx f3 (x)dx where Suppose, now, 1%+i P(s) xl-S ds 1 P (x) J1/2_iil- Z Pl(X) p (x) is differentiable, then from (3.4), we have p(s)x-Sds p (X) 1/2-i i.e. p(x) -i M [P(s); x]. Further let k i(x) i x), (i 1,2,...,n and 6 being the Dirac delta function. Then the sequence (3.1), reduces to fi+1 (u) v i (um)fi (x)cLz and consequently we have, a non-integrated version of Theorem 1: THEOREM 3.3. If (i) (iii) vi,fi P(s) and H i=I fn+l Vi(8), L2(0’)’ i 1,2 n C. NASIM 62 then n+l (u) p(x) where Letting p(x) o --v p()71 P(s)x ds | (l-x), and hence, its Mellin transform, P(s) 1 the above result, then, defines the chain transforms introduced by Fox. Note that all the above mentioned results are specializations of our Theorem i, showing the general nature of that result. Next we shall deduce a Perseval type relation for the general chain transforms as defined by Definition 3. Let sequences forms of order n with respect to the kernels k. and v.. {fi and {gi} be chain trans- Now consider where + + () () By virtue of THEOREM i and the consequence (3.3), we have x-lf:+1(ux)1(x)dx x-11(x)dx t-l t-lp (uxt)?1(t)dt (t)dt x 0 t -lpl (t)l (t)gn+l (ut)dt Hence x-l+* t-l ()gn+l (ut)dt (uz)g (x)d establishing, formally, the Parseval relation. (3.9) Further on differentiating both sides, formally, we have 71 (t)L+1 (ut)dt fn+1 (ux)1 (x) 1, reduces the above to the usual arseval Theorem, [1, It is now an east matter to develop an inversion theor for the general chain transforms. For instance Sf we assume that ettng n P(s)P(I ) i, then from (3.3), due to the standard inversion formula, we have 1 (X), x-lpl (zeX)fn+l (x) Thus the unknown function 1 EO, P(e E0, 0 I is retrieved. -x On the other hand if we assume that d being the Lagunerre-Polya class [4, VII], then due to the known inversion tech- niques, [5] 1 p(o) [fn+l (x)] giving us the fl (x) function I" THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS 4. 63 SELF-RECIPROCAL FUNCTIONS. is a sequence of general chain transforms of order n, with respe-t {i Suppose to the kernels v..z kz. and If we then set L2(O,), ki,vi,i If (i) THEOREM 4.1. 1,2 i and n-l, then in Theorem I, we have 1,2 (x) f( LI( i, n V.(s) z H K.(8) i=i P() (iii) n+l(X) l(X)’ 1 0 d. 1 From the conclusion (4.3), t follows that kernel Pl(z)’ s The particular case when n t,. Sbol[cally we say that in the sense of definition 2.2. fl self-reciprocal with respect to R P In this case, many useful 2 is of special interest. properties of self-reciprocal functions are established, including a procedure for generating self-reciprocal functions with respect to a given kernel. Now let n 2. Then, COROLLARY i. If (i) (ii) and ki,v i f L2(0’)’ fi f (um)2(x)o!r k k 2 (ux) fl (x)dm i (um)l (x)ci V2 ()2 (x) P() then i PROOF. , (4.5) + g) R Pl () where c( ( (4.4) V and () 1,2, _g By the Parseval theore for 1 Nell[n transforms, equation (4.4) and (4.5) give respectively, ()F() ()F () ()F (-) i a.e. on tons. + i, (4.) (4.7) () (), < < , invlng the Nell[n transfor of the respective func- Next define functions C() and ((_ ) ) () () () (), C (g- , g+ + i, i) . C. NASIM 64 Then from (4.6) and (4.7),weobtain, a.e. on I + iT, 8 , < T < F2(8) M(8)F2(I 8) (4.8) F I(8) L(8)F2(I 8). (4.9) and Also, L(I- 8) M(8) L() P(8) By the Parseval theorem for Mellin transform of L2-functions (4.8) and (4.9) give, re spec t ively, x-lm(um) f2 f2 (X)Clr IU fl I x- (x)dx gl (UX) f2 (x)dx where and e (X) [1/2+i 1(8) x 1 m l(x) = [1/2+i L(s) 1 1/2-i The functions m e and 1-s dx 3:1-8 ds 8 )1/2-i 1 S are defined by the integrals which exist in the mean-square sense, since both M(8) and L(8) are bounded on 1 8 + i, < M(8) < =, due to the and hypothesis (iii) of Corollary 1 above, and consequently 1 iL 1 1 + i). Thus, combining the above results, we have, i, L(8) both 8 L2( Corollary 2. fl (i) If and L2(0’)’ f2 then fl where gl,ml and Pl R are defined above. This result gives us a procedure for generating a new class of self-reciprocal functions, given a self-reciprocal function and the kernels. (ii) shows that I 2 Note that the hypothesis If all the functions involved are differentiable, we obtain a simpler form of the Corollary 2 above, that is: Corollary 3. If (i) fl and f2 L2(0,) (ii) -2 (u) m()f2 (x)o, (iii) fl (u) e()f2 (x)o (iv) P(8) and L(1 L(s) (i.e. f2 Rm)’ THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS 65 then For example, let re(x) Now -2[si+J(4x1/2) + co+{y(4x’) K(4x1/2)}], 1 (I s -,)(i (s) i where [ (z) n -z s +,) (8 + -2-) (8 R(z) _> 0 1 i ) the Riemann-Zeta function i Also n=l F(I 1 1 i (2) 1-28(8 + )cos (8 + ) l l l + -)sin (8 1 M(8) 8 by making use of the identity (8) 8-128F( I 98 8)sin 8). (i It is an easy matter to see that M(8)(l- 8) l, hence m(m) is a Fourier-Watson kernel. L(8) (8 cosec Further let 1 -), then 1 2+ and p(x) I[p(s); M 1/2) i 2 J (4x x] < Re 8 < Thus, Corollary 4. If (i) (ii) f Rm, g(x)= m defined by (4.10), [ 2 where 8(x) then (X) Rp, where p(X) 1 2 + x2 J(4). In particular, let f(x) 1/2 cos(2X). X f One can show that g(x) I 2 x m" Then (xt)f(t)d X -1/2 I cs(2$)2 i + x2t dt (I+7#) -2lx e [6;221(55)]. Now p(xt)g(t)dt 2 t e -2t J (4 x’t1/2)dt 3 (4.10) C. NASIM 66 xC-2?x [6;29(10)]. g (x), Corollary 4. as predicted, thus verifying the result of different version of Theorem I 4.1, so that the function Next we shall give a slightly fl’ rather than the function is involved in the conclusion. Let the conditions (i) and (ii) of Theorem 4.1 hold. THEOREM 4.2. n V.(l s) z U -s) i:i K/(i Q(s) then where o f (x) qi(x) f i (- x- ql (ux)f = "e Q(s) 1 _i x s i Again the special case when n i, i i) + (x)Ix i- 2, Further if (r f R ql ds is interesting, since it gives us a procedure for generating self-reciprocal functions. COROLRY 5. (i) If fi,vi () and k 1,2, (z) ()f() k ()f () e()f() () then L (0,), i i ( )( ru ) q .e. f (R q Now from the hypothesis ([[), by using the Parseval theore or Nellin transfos, the to equations imply respectively, KI(8)E2(8 VI(8)FI(8 K2(8)FI(1 a.e., where 8 V2(8)F2(8), 8) 1 + < T < T, . If we consider the functions L(8) and F2(8) M(s)F2(I s) El(S) L(s)F2(I s), M(s), defined earlier, then as before, we obtain and a.e. The last two equations, then give, respectively, and 0 f2(x)dx fl(X)dz where I (X) i IIm(ux)f2(m)dzxx1/2+i I #_ioo (ux)f2(x)dx L(s) S xi-Sds (4.11) (4.12) THE GENERAL CHAIN TRANSFORM AND SELF-RECIPROCAL FUNCTIONS and 1 m 1(x) =-i-4 + M(.8) xl-Sd8 s 1 1/2-i 67 Thus we, formally, have COROLLARY 6. (i) If fl and f2 6 L2(O,==), (4.11) and (4.12) are satisfied, L(8) (iii) Q(8) 8) L(I 8) M(I (ii) then u i x- q l(uZ)fl(z) fl(x)/ where I’ ml’ql are as defined above. It is not difficult to prove the above result rigorously, the functions and 1 and M(8), L(s) and Q(8) are bounded on 8 < T < + iT, hence the Parseval theorem can be applied to give us the desired result. A non-lntegrated fo of the above result is as follows: , L2(0, ) COROLLARY 7. then J fl(u) The kernel (i) f2(u) (ii) fl(U) If 0 I fl f2 m(ux)f2(x)dx e(um)f2 q(uz)fl(x)c5:. 0 ql (u) q(z)ce and the kernels (z) and m(m) are defined similarly. Note that the above results gives us a procedure for generating self-reciprocal functions. For example, let re(x) x J (x) then its Mellin transform is given by 1 M(s) 2 s- F(} + 1 + ) 3 +1 1 r( [7; 326(1)] Now, let (xI2)1/2(++I)K+1/2 (_) (x) (x) K being the usual Bessel function of third 2 8- L(8) r(1/2+ +Z-) 1 1 Therefore L(8) L(i Z 8) M(1 (s) 2-1/2 whence q(x) r(1/2 1 (gu 1 kind and 1 + /Z-) 8) + +) 1 3 1/2 J (x). x Thus, we have as a special case of Corollary 7. COROLLARY 8. 1 (x) 0 (xt If f H 7F(++1) and (._) (xt)f(t)dt, its Mellin transform is C. NASIM 68 then Here the class H Hankel transforms of order x v+ e -2 f(x) and H f(x) . denotes the class of self-reciprocal functions with respect to the To varify this result, let Then, [6; 132(25)], I(-"-")r(+z)r( _I g(x) _1/2(P++I) (:t) K1/2(,_) (xt)f(t)dt I e Is n ey mae o d cul ce o I i + 1)x1/2 (v+- ex2 /4W_(++),(_) (x/) b g(x) obed, be boe eul CORO 9. + H e peieed, [6;84(15)]. e A i- i.e. R f e_Xtf(t) dt g(x) then Rc, g(x) where R and R denote the classes of self-reciprocal functions with respect to Fourier sine and Fourier cosine transforms, respectively. In the operational notation, we have that if F [f] f and L[f] g, then F [g] where Fs,Fc respectively. f(x) and L denote the Fourier sine, Fourier cosine and the Laplace transforms +1/2 2 1"(41-1) x 23 such that f(x) - For example, consider e -x 2 / M3, (x2/2), Re > i 4 R8’ [7 ;115(5)]. Now, 2-X 29-I [7;215(13)] as predicted. where M "’2/4 ’ -3), and W - (x2/O) are Whittaker’s functions and then g(x) Re, [7;61(7)], REFERENCES i. 2. 3. 4. 5. E.C. Titchmarsh, Fourier Integrals, Clarendon Press, Oxford, 1948. C. Fox, Chain transforms, J. London Math. Soc., 2_3 (1948), 229-235. B.P. Duggal, On Fourier transform type operators, Period. Math. Hungary, 9 (1978), 55-61. D.V. Widder, An Introduction to Transform Theory, Academic Press, New York, 1971. C. Nasim, An inversion formula for a class of integral transforms, J. Math. Anal. Appl. 52 (1975), 525-537. 6. 7. A. Erdelyi, et al., Tables of Integral Transforms II, McGraw-Hill, New York, 1953. A. Erdelyi, et al., Tables of Integral Transforms I, McGraw-Hill, New York, 1953.