From "Fractional Calculus and its Applications", Springer Lecture Notes in Mathematics, volume 57, 1975, pp.1-36. A BRIEF HISTORY AND EXPOSITION OF THE FUNDAMENTAL THEORY OF FRACTIONAL CALCULUS BERTRAM ROSS Abstract: This opening lecture is intended to serve as a propaedeutic for the papers to be presented at this conference whose nonhomogeneous audience includes scientists, mathematicians, This expository and developmental lecture, engineers and educators. a case study of mathemati- cal growth, surveys the origin and development of a mathematical from its birth in intellectual curiosity to applications. mental structure of fractional calculus is outlined. for the use of fractional cated. calculus idea The funda- The possibilities in applicab]e mathematics is indi- The lecture closes with a statement of the purpose of the con- ference. Fractional calculus has its origin in the question of the ex- tension of meaning. of real numbers A well known example is the extension of meaning to complex numbers, meaning of factorials of integers and another is the extension of to factorials of complex numbers. In generalized integration and differentiation the question extension of meaning is: order dny/dx n irrational, be extended to have meaning where fractional Leibnitz invented the above notation. that prompted L'Hospital possibility that n be a fraction. Leibnitz Perhaps, n infinite product for it was naive be ½?", asked "It will lead to a paradox." "From this apparent paradox, ful consequences will be drawn." that differential is any number--- to ask Leibnitz about the "What if [i] in 1695 replied, But he added prophetically, Wallis's n or complex? play with symbols L'Hospital. of the Can the meaning of derivatives of integral one day use- In 1697, Leibnitz, referring to ~/2, used the notation d2y and stated calculus might have been used to achieve the same result. In 1819 the first mention of a derivative of arbitrary order appears in a text. The French mathematician, S. F. Lacroix [2], published a 700 page text on differential which he devoted less than two pages Starting with n a positive symbol and by replacing m he found the F by in the manner typical Lacroix obtained mth and n x a. formalists factorial, real number a, of this period, the derivative of arbitrary for (x) = dx ½ = ½F(½) yielded by the present xa-½ r (a+½) He gives the example F(3/2) the generalized by any positive of the classical to be the formula d½ because derivative n -m which denotes 1/2 dx ½ tion in x d2y = F(a+l) which expresses calculus y = xn~ integer, dmy _ n! dx m (n-m) ! Using Legendre's and integral to this topic. order y = x 1/2 of the func- and gets 2~ /-~ = ½/-# and F(2) = i. day Riemann-Liouville 279 years This result definition tional derivative. It has taken raised the question for a text to appear solely is the same of a frac- since L'Hospital devoted first to this topic~ [3]. Euler and Fourier made mention order but making they gave no applications the first application Abel applied the fractional equation which This problem, arises belongs calculus wire of the tautochrone is placed. of the wire in the s a m e The brachistochrone time problem regardless problem. is that of find- lying in a vertical plane that the time of slide of a bead placed on the wire slides lowest point of [4] in 1823. in the solution of an integral called the isochrone problem, ing the s~hape of a frictionless of arbitrary So the honor to Niels Henrik Abel in the formulation sometimes of derivatives or examples. such to the of where the bead deals with the shortest time of slide. Abel's attracted attempt solution was so elegant the attention of Liouville to give a logical definition that it is my guess it [S] who made the first major of a fractional derivative. He published three long memoirs Liouville's tives of integral in 1832 and several more through starting point is the known result 1855. for deriva- order Dme ax = ame ax which he extended in a natural way to derivatives of arbitrary order DYe ax = aVe ax He expanded the function f(x) in the series 9o (1) f(x) : I cn e anX, n=O and assumed the derivative of arbitrary order f(x) to be co (2) DVf(x) Cn a v e anx = n=O This formula obvious is known disadvantage the series as Liouville's that v [6] first definition must be r e s t r i c t e d and has the to values such that converges. Liouville's of the form second m e t h o d was applied x "a, a > O. (3) I The transformation He considered = f xu = t (4) x-a _ to explicit functions the integral ua-le-XUdu. gives the result 1 I. r(a) Then, with the use of (I) he obtained, of (4) with after operating on both sides D v, the result (5) DVx -a = (-l)Vr(a+v) x -a-v [7] r(a) Liouville problems was successful in potential theory. in applying these definitions "These concepts last," said Emil Post [8]. certain values and the second method class of of functions. v were too narrow The first definition is restricted is not suitable to to to to a wide Between 1835 and 1850 there was a c o n t r o v e r s y w h i c h c e n t e r e d on two definitions favored Lacroix's mathematicians [I0] of a fractional derivative. George Peacock g e n e r a l i z a t i o n of a case of integral order. favored Liouville's definition. Augustus [9] Other De M o r g a n ' s judgement p r o v e d to be accurate when he stated that the two versions may very p o s s i b l y be parts of a more general W i l l i a m Center versions of a fractional tive of a constant. fractional while In 1850 derivative focused on the fractional deriva- A c c o r d i n g to the P e a c o c k - L a c r o i x v e r s i o n the derivative of a constant yields according to Liouville's of a constant equals formula zero because The state of affairs cleared up. system. [ii] observed that the d i s c r e p a n c y b e t w e e n the two a result other than zero (5) the fractional derivative r(o) = ~. in the m i d - n i n e t e e n t h Harold Thayer Davis [12] states, century is now "The m a t h e m a t i c i a n s at that time were aiming for a p l a u s i b l e definition of g e n e r a l i z e d differentiation but, in fairness to them, one should note they lacked the tools to examine the consequences of their definition in the com- plex plane." Riemann posthumously [13] in 1847 while a student wrote a paper p u b l i s h e d in which he gives a definition of a fractional operation. It is my guess that Riemann was i n f l u e n c e d by one of Liouville's memoirs in which Liouville wrote, "The ordinary differential equation dny = O dx n has the complementary solution Yc = Co + ClX + c 2 x 2 + "'" + Cn-I x n - 1 Thus du , f(x) = o dx u should have clined to a corresponding believe Riemann complementary saw fit his definition of a fractional (6) Cayley D-v f(x) = 1 r (v) to add solution." So, a complementary I am i n function integration: (x-t)v-lf(t)dt + ,(x). ;c [13] remarked in 1880 that Riemann's is of indeterminate nature. complementary function to The development Peacock made several he m i s a p p l i e d which of mathematical errors in the topic of fractional the Principle of the Permanence is stated for algebra and which theory of operators. in his discussion Liouville made of one of the parameters versions sults when Heaviside century, Riemann became hope- Thus, I suggest silence Two different re- that when Oliver of the nineteenth and disdain not only because jibes he made at mathematicians the distrust mathematicians function. yielded in the last decade he was met with haughty of the hilarious apply to the complementary derivative applied to a constant. when function that the specialization led to an absurdity. p u b l i s h e d his work calculus of Equivalent Forms did not always an indeterminate of a fractional error. an error when he failed to note of a complementary lessly entangled with different ideas is not without had in the general but also because of concept of fractional operators. The subject ness of notation papers that notation of notation of fractional follow calculus in this text, I prefer was cannot be minimized. The succinct- adds to its elegance. various notations invented by Harold T. Davis. In the are used. The All the informa- tion can be conveyed by the symbols cD~ v f(x), denoting integration scripts c and a definite x of arbitrary denote integral to avoid ambiguities becomes integration itly formulate. fractional a vital part the mathematical fractional or complex, wide formalizing f(z) z, the derivative frac- but were try- class, for every function and every number D v f(z) = g(z) c z If symbol problem of defining a function ' i. The adjoin- of the operator Briefly what is w a n t e d is this: D v f(x) = g(x) when z is purely cx to the following criteria: able integration. of they well understood but did not explic- z = x + {y, of a sufficiently irrational The sub- of integration It is clear that the mathe- so far were not merely ing to solve a p r o b l e m which f(z), along the x-axis. (terminals) and differentiation. mentioned O, in applications. We now consider tional order the limits which defines ing of these subscripts maticians v real is an analytic cDzv f(z) v, or ' should be assigned subject function of the complex vari- is an analytic function of v and z. 2. as o r d i n a r y is The operation a negative same result along with D v f(x) integer, as its 3. n-1 when say ordinary The operation at C D- n X the same result integer. f(x) and C D- n X If v must produce f(x) must the vanish x = c. zero leaves the function un- = f(x) X The fractional operators cDx v [af(x) + bg(x)] must be linear: = a cD-Vx f(x) The law of exponents 5. a positive then of order DO f ( x ) 4. is integration derivatives C v v = -n, n-fold changed: must produce C X differentiation + b cD-Vx g(x) for integration of arbitrary order holds: cDxu- cD-Vx f(x) A definition of Riemann (7) This definition we have Riemann's to establish definition is discussed The definition in honor (x-t)v-lf(t)dt" of arbitrary and when (see the above named fX order stated is the same as function. c = -= , (7) [6], pp. 176-178). a set of criteria This question criteria but has no complementary definitions to fulfill these = irtvjlr ~ jc for integration definition Liouville's criteria, that will When c = O is equivalent Although it might be of interest characterize later in this text p. (7) can be obtained (7) uniquely. 379. in at least ~0 x (x-t)bt d dt = F(b+l) F(d+1)_ xb+d+l F(b+d+2) (s) b = 3 four differ- and d = 4, (8) gives r b and d > -I the result (4) x s 8.7-6 -5 If one were constant to (7) can be Euler had shown that ent ways. For fulfills is cD-Vx f(x) Riemann's shown which and Liouville = cD-U-Vx f(x) to integrate of integration the function x4 four times each time to be zero, and take the the result will be x8 1 8-7"6"5 Inquisitive experimentation of this type might lead one to guess that the above two results may be connected by the expression: x4 = - oDx 4 fox i F (4) (x-t) 3 t 4 dt, or in general (9) oD; n f(x) ~OX 1 = ?(n) The above is generalized by letting The same r e s u l t integral iterated ( x - t ) n-1 f ( t ) n = v. can be o b t a i n e d by c o n s i d e r i n g - 2 F(x) = ~cx dx 1 ~cxl d x 2 - ' This iterated integral ~c xn region as a single f(Xn) dXn ' integral by the that is, by integrating [14]. the n - f o l d Xn - 1 f dXn-1 Jc can be written method devised by Dirichlet, priate triangular dt. The result over an appro- is rX F(x) = If we denote gration 1 ~ r(n) Jc (x x - n) the operators as We may write F(x) ing by replacing and = cD; n f(x). n with v A third approach linear differential Then letting xn = t we again arrive at (7). dn - 1 dxn-I + ... + Pn(X) + Pl(X) operator whose are continuous on some closed finite Po(X) I. H and generaliz- to (7) may be deduced using the theory of dn Let and of inte- dx, Let be a linear differential on ~c x . . . . equations. L = Po(X) dx n.. > O f(Xn) dxn " of differentiation D; 1 Dx n-I coefficients interval be the one-sided Pk' 0 ~ k ~ n I = [a,b] Green's and function for L. Then if in f is any function I, then for all continuous on I, and xo is any point x C I, g(x) = f H(x,~) f(~)d~ o is the fies solution of the the boundary nonhomogeneous equation Ly = f ( x ) which g ( k ) ( x o) = 0 [For further details see, , 0 ~ k ~ n-1 for example, . K. S. Miller, Linear Differ- ential Equations in the ReaZ Domain, W. W. Norton and Co., York (19631; satis- conditions Chapter Inc., New 3.] The Green's function H is given explicitly by el(X) qb2 (x) ... ~bn(X) qbl (~) ¢2(~) ..- (~n(~) (-i) n-I H(x,{) = where and Po(~)W(~) {¢k[l ~ k # n} W is a fundamental set of solutions of Ly = O, is their Wronskian. Now s u p p o s e L = Dn Then {l Dny = O , x, x 2 , ..., x n-l) dn dx n is a fundamental set of solutions and ¢i(~) w(~) ¢2(¢) --. : i (~) "'" Cn(~) of = 1 ¢ ¢2 0 1 2¢ O O 2 o o o (n-l) . . cn -i (n-l) ¢n-2 (n-l) (n-Z) ¢ n-3 ( n - I) ! !! where n-I (n-l) !! = II k! k=O Thus in this special case 2 1 X X • 1 ¢ ¢2 . . 0 1 2¢ o o o • • X n-i cn-i (-i) n-I H(X,¢) is a p o l y n o m i a l = ( n - l ) !! of degree n-i in (-1) n-1 _ _ [(_l)n+l(n_2) ( n - l ) !! x leading 1 (n-l) ! But H(x,¢) 3xk Thus ¢ = O , 0 < k < n-2 • x=¢ is a zero of m u l t i p l i c i t y H(x,¢) (I0) Hence if xo n-I 1 -I - (n-1) l(x-¢)n = a, (n-l) ¢n-2 (n-l) ! ¢ with ! !] = • and coefficient 10 rx (11) 1 g(x) is the unique Ja (x-~)n-I ~ (n-l')! solution of the differential f(~)d~ equation dny - f(x) dx n which assumes write (II) as the initial values We may ~aX aDxnf(x) Now, 0 < k < n-l. g~k~(a)r ~ = O, = of course, we replace 1 r (n) n (x_~)n-lf(~)d~ by v (with . Re ~ > O) in the above formula~ [15]. The fourth method of arriving tour integration these generalized tegral operators in standard works in securing in 1888 used a contour in 1890. Laurent Cauchy's integral formula n But here To keep three methods values i/(t-z) v+l the function semi-infinite is n! U no longer contains starts because c, cut be the to negative Let C infinity on be the open contour c < x, on the lower edge of the real axis to A, around the circle in the positive the upper edge of the cut. of v! = F(v+l). a pole but a branch point. we let the branch t = x > O a the point the cut, then goes along (7) the generalization no difficulties single valued, at f(t) dt. (t-z) n+l = 2--~-~-~J ic the real t axis as in the figure below. It-x I < ~ at the origin as did A. Krug of obtaining creates line starting (or loop) which P.A. loop. f(n) (z) = D z f(z) to fractional starting theory. in- only passing in 1884 used a contour that started and ended at -~, now called a Laurent In the previous fact that for themselves in complex variable Nekrassov n (7) is by con- It is a curious and their connection with the Cauchy formula have succeeded references at definition in the complex plane. sense to B, and then back to c along ii Im(t) t-plsne cut of n Generalizing gives B/~ex ~.C ,Re(t) the Cauchy integral formula to arbitrary values F(v+I) D v f(z) = " 2 ~ (t-z) -v-lf( t) dt where we define (t_x) -v-I = e(-V-l)in(t-x), and where In(t-x) is real when t-x is a positive By standard methods of contour to (7) (see integration we are a~ain led [6] pp. 198-202). The general validity of definition positive real number. (7) for integer,can be established by mathematical we are concerned with criterion 2 which stipulates must produce the same result as ordinary v = n, n induction. a Here that the definition integration. There is no loss of generality by taking the lower limit of integration to be zero. We have 1 ODxnf ~ fx) = F(n) The above is obviously oD if(x)~ (x_t)n-lf(t) dt. true for = fo n = i, for f(t) dt. Now assume the formula true for n = k: 12 1 oOx- k f<x) o r(k) Replace k with ~0X( x _ t ) k - l f ( t ) k+l : oD; (k+l) f(x) = SO X ( x _ t ] k f l t ] 1 ........... kr (k) Operate on both sides of the above with (12) ODxk- Applying Leibnitz's f(x) d = 77 1 (x = O, and g(x,x) = 0 of an integral ~ (x_t)kf(t) is the integrand The last two terms d(O)/dx (x-t)kf(t)dt. fox 1 - kr(k) d(O) g and we have ~}(k)flo g(O,x) - 7 T the function dt. oDx = d/dx rule for the derivative 0Dx- k f ( x ) where dt. + g(x,x) (x-t)kf(t) gives dt dx --~ , in (12). on the right above vanish because because of the factor (x-x). T h e n we rx have oDx-k f(x) = k = n and since k Io (x-t)k-lf(t)dt' kr('~'i we have by mathematical induction fX - 1 oDxn f ( ~ ) = r(~"i This result / o (x_t)n-lf(t)dt. is the same as (9) obtained heuristically. The definition for differentiation of arbitrary be shown later to be an integration followed by ordinary tion. criterion It follows that (7) fulfills order will differentia- 2 for differentiation and integration. Criterion 3 states the function unchanged, whether D v f ÷ f as C the x The investigation that is, cDx0 f(x) = O. v ÷ 0 shows a concern with the continuity x D -v operator C that the operation of order zero leaves at v = O. We have of of 13 -X (13) c DO x f(x) as a consequence be taken general, O-~ . - F(O) 1 of letting ~£ ( x - t ) O-1 f ( t ) d t , v = O in (7). The f a c t o r 1/r(O) can equal to zero because F(O) = ~ . The i n t e g r a l would, in be divergent a n d we h a v e t o d e a l w i t h t h e i n d e t e r m i n a t e form There are several We assume ways of handling f(t) this is e x p a n s i b l e situation. in a T a y l o r ' s series w i t h remainder: f(t) and t a k i n g = f(x) limits lim v÷O + (t-x)f'(x) of b o t h sides oDxVf(x) = + (t-x)2f"(x)/2i of (7) as [Jo lim v÷O ~Ox ( x - t ~ , , v f ' ( x ) d t + - - " + (_l)n+ 1 f(n+l)(@) (n+l) I except ~OX l i r a r ( v ) = ~ , a l l t h e t e r m s on t h e v+O the first because the first integral , we have (x-t)v-1 F (v) f(x) dt ;ox ( x n_ t!)r (vv÷)n - 1 + (-1) n r(v) Because v÷O + --- f(n) (x)dt (x_t)v+n F (v) ....... d t right has above vanish the value xVf(x) r (v+l) due to the g a m m a no function loss of g e n e r a l i t y of relation u s i n g the vF(v) = r(v+l). We n o t e l o w e r limit of i n t e g r a t i o n there O instead c. Thus lira v÷O oDx v f(x) D O f(x) Ox We can also arrive = lira xVf(x) v÷O F (v+l) = f(x) at the same result in the f o l l o w i n g manner. Let the f u n c t i o n The integral f be c o n t i n u o u s (7) can then be w r i t t e n on the interval is (c,x). as the sum of two i n t e g r a l s : 14 c D-v x f(x) - F1(v) ~cX If (t) (14) f(x)] (x-t) v- ldt -X i f(x) r (v) The first tend to zero as intervals integral v F(v) f(t) 1 r(v) 6 tegrals on the right tends to zero. f(x)] is any small positive number. in integral B, where is continuous. these in- s depends 1 6+0 After evaluating is written If(t) on 6, and f(x) I by s. Thus, lim s(6) 6+0 = 0 because ~xX (x_t)V-ld t 6 ~v ~ (~)~v s(~)''T-: r(v+l) with IBI : o. the second integral as cDVf(x)-~_ IBI ÷ 0 of v ~ O, we have lim (14) B. We then have --< ~ where Let us designate f(x) l _-< ~(s) 1 IBI =< r--'~) ~(6) for all + the maximum If(t) Eq. (x-t) V-ldt as In B denote Then, into two sub- ~xX6 If(t)- f(x)] (x-t)V-ldt A f in (14) will be shown to It can be divided of integration: (15) where (x_t) V-ldt. : 6. = A ÷ B + (x - c ) v r (v+l) {(x) on the right in (14), 15 We can now consider maximum of If(t) f(x) I the integral by M. A in (15). Denote the Then M ~c x-6 (x-t)V-ldt IAI ~ ~7-~) r(v+l) (~v _ (x_c)V). Let e be any arbitrary positive number. IBI < e for all v ~ O. For this fixed both sides of (16) add -f(x). Ken cDxVf(x) Because f(x) -i . IBI < e, we have - f(x) I ~ 0 + ~ + 0 can be chosen as small as we wish, lim I cDx vf(x) v÷O or so that v + O. To $ [A I + IBI + If(x) l LF(v+l) lim sup I cDx-v f(x) v÷O Since Now choose 6 ~, IAI + 0 as lim v÷O D-Vf(x) it follows that f(x) I = O, = f(x). C x Another approach to the above result using the theory of Laplace transforms might be of interest. If we define in [O,L] then f can be taken as zero in x > L. Let f(x) f(x) only be such that L ~ e_aXf (x) dx exists for some real ~(s) is an analytic a. Then it follows f that e-Stf(t)dt function of s in Re(s) > a, that, with x e -sx 1 r (v) J (x_t)v-lf(t) d t v > -i, 16 exists in Re(s) > a, and in fact, that oo 7(s,v) where g(x,v) = -'o e-SXg(x,v) dx ~ s - v ~ ( s ) , denotes the right side of (7). It is also true that 1 ~ f l (x) ="~T~ where G eS x ~(s) ds is any vertical path lying in differs from f(x) Re (s) > 0 and where fl(x) on, at most, a countable number of points. Furthermore, g(x,v) - 2 1~ But, for such a path ~G e sx g(s,v) ds. G, everything is uniformly bounded and lim g(x,v) - 2 1~ v÷O ; e st f(s)ds z /c fl(x) which is the result wantedj[16]. D-u CX D-v f(x) = cDxu-v f ( x ) . CX We now consider criterion 5: By definition (17) (7) we have = 1 ~c X (x-s) U-ld s r(u) 1 F(vi i o s (S _t)v- 1 f(t)dt. The repeated integral above corresponds to a double integral to which Dirichlet's formula, mentioned earlier, may be applied. (18) I cDx u c D x-v f(x) - r(u)r(v) /cx We have f(t) dt • ~-x I (x-s)U-I (s-t) v-I ds. t 17 When either to (18) u or v is on the interval (0,I), the passage can be justified by a minor m o d i f i c a t i o n from (17) of the Dirichlet proof over a smaller triangle. Make the transformation gral on the right in (18) is a beta integral The second inte- is then (x-t) u+v-I which y = (s-t)/(x-t). fo1 (I -y)U-lyV-ldy that has the value r (u) r (v) (x_t)u+v-1 r (u+v) When this is substituted into (18), we obtain ~D~ u cDx ~ f(x) = r(u+v) ~ 1 The integral on the right the role of arbitrary above is definition order. We then have A subtle mathematical the law of indices order. (7) with u+v playing the required result. arises when one seeks to extend of arbitrary order to deriv- If we follow the preceding method, we will integral Du Vf = 1 c x cDx (x) F(-u-v) To establish (19) f (x-t) -(u+v) -if(t)dt. the relation f(x) c Du x cD~f(x ) = c _u+v Ux it will be required which vanishes This proof to impose the restriction is omitted here but details derivatives, interchange that at the lower limit of integration, The restriction ing (19). problem stated for integration atives of arbitrary get the divergent X(x_t)u+v_if (t)dt. that f as stated in criterion at x = c 2, is necessary f(c) = O. and at its n-i to justify the used in the proof of establish- the relation DD -I f(x) = DOf(x) be a function can be found in [6]. vanishes of the order of operations For example, f namely = f(x) 18 always holds. But the relation (20) is not D-1D f ( x ) always valid. = DOf(x) = f ( x ) For, by definition C D X-I cDx f(x) = cDx- 1 f'(x) = r(1) = and (20) holds only when However, sion. by means Let greater than - (7) is for integration of a simple where v, and f(c), = O. of arbitrary v = m-p (x-t)Of'(t)dt f(x) f(c) The definition For differentiation (7) of arbitrary order. order it cannot be used directly. trick, we can find a convergent for convenience O < p __< i. m is the least Then for differentiation expres- integer of arbi- trary order we have D v fix) C = X Dm C (21) X D -p f(x] C X _ dm 1 dx m r(p) ~c x (x-t)P-lf(t) dr' where we take a~Ivantage of the knowledge mth derivative definition operator that dm/dx m. from the fact that fractional operator quired analyticity, by hindsight. Dm X is an ordinary We have assumed for purposes referred Dm-p D -v. to above, namely is the analytic It is obvious of this and also the other four criteria, The question of extending answered by letting ¢(v,x) D v = Dm-p continuation that criterion v be real and greater = oDxVf(x) = F(v) 1 re- of the 1 which re- were established the definition gration of arbitrary order to differentiation (22) C Dm-p = DmD -p. The simple trick sults that (7) for inte- of arbitrary than zero. order is We have ~x (x-t)v-lf(t)dt ~O which is in general convergent for v > O. For any v we can write 19 ¢(v,x) = oD-Vx f(x) = oDmx oDxp- f(x) dm 1 dx m F(p) where -v = m-p, When m = O, I, 2, v > O choose ~0 X if (x-t) p- (t) dt, .... m = O. Thus v = p and [~(v)~oX(x-t)v-lf(t)dtl dx. ~ -- ,. Now~ (22) can be written ~(v,x) By Dirichlet's = d~oX formula, we have d 1 ~0 x (x-t)vf(t)dt = ~-~ r(v+l) ~(v,x) which is convergent ~(v,x) This process point of v > -i. = ~(V,X) for can be repeated is analytic v > -n. for Since in R1 v ~ -n, n v > O and ~ on a set of points in the right half plane, ~. m = I. for where ~ = , We then have This justifies then q in Some explicit examples For the fractional letting v = m-p, m RI~R 2 Dm-p of fractional derivative for in R2 Now for continuation Dv. derivatives of a constant the least integer integer. with a limit is the analytic the trick of writing useful. a positive is analytic will be k, we have by > v, and the use of (21), the formula k (23) oD~, k - r ( I - v ) Another example (24) -v is the integration arbitrary order of the natural By definition x and differentiation logarithm. (7) we have O xD -v In x = 1/oX F(v) (x-t)V-lln t dt, V > O. of 20 Let t : x+ t - x, : x(1 x>O + t~x). in t = In x + in(l + t;x) Then with the restriction -I < t-x < i . X Using the Taylor's = series expansion for In(l+@), we get co In t = in x + ~ (-l)n-l(t-x)n L_. nx n n=l where the interval of convergence right side of the above in x is O < t ~ 2x. into the right Substituting the side of (24) gives ~0 X (x_ t) v- Idt r (v) co 1 r(v) Term by term integration, gives f permissible xVln x In terms because xv r(v) XV oDx v in x = is Euler's In x ~ of the psi function, written order of dt • of uniform convergence, co in x = r(v+l) C (x-t) n -- nx n n=l the result oD;V where 2 v-1 (x-t) r (v+l) constant. [ In x - 1 ~<~+k~ the above result can be C - ~(v+l) ] For differentiation of arbitrary we have O xD v I n x = O xD m-p in x am dx m (p+l) xpZ r(p) k( + k=l where the usual criteria for termwise differentiation is to be applied. 21 Although we now know how to interpolate orders where of the derivative such procedures of functions such as might be applicable. between integral in x, little is known In this connection this writer s u b m i t t e d a p r o b l e m t o t h e American Mathematical Monthly t o r(x). This will permit appear in winter 1974-75, concerning 0 Dxv i n interpolation b e t w e e n i n t e g r a l o r d e r s o f t h e p s i f u n c t i o n and m i g h t h a v e u s e in t h e s u m m a t i o n o f s e r i e s o f t h e f o r m ~ 1 / ( 1 + x ) u. Eric Russell Love [17] has defined integration of pure imaginary order in such a way as to extend the properties tion and differentiation the case where Re(n) the Riemann-Liouville cosine of arbitrary = O. n where of integra- Re(n) Francis H. Northover makes definition (7) can be connected and Fourier sine transforms imaginary order > O to the claim that to the Fourier by means of derivatives of pure order as follows. l~ X 1 cDxv- F(x) I (x-t)V-IF(t)dt, ;cl - r(v) Re(v) > O. Make the transformation t = x - (x-c)e -@. The limits (terminals) of integration then become O and ~, and we have cDxv- F(x) v F(x Now let = (X-c),V" r(v) -c)V .... ( X r(v$---S{ v = -in, and assume e d~), -V~ F(#) F cD1xn F(x) exists. Then oo (26) cDlxn F(x) - (x-c)-{n r(-in) (x-c) -in r(-in) where and C(n) = (2)½ S(n) = (2)½ 50 L e ~n~ F(~) d~ ½ (2) {c(n) F(~) cos n~ d~ V(~) s i n + { s(n)} n~ d~. , 22 Love has shown that suitably derivatives of all orders of any order v function which for Re(v) where > O. restricted Re(v) functions have but have no derivative but does not possess For this reason, paragraph where = O He has also cited an example is locally integrable of any imaginary order. preceding v it was stated of a a derivative caution was exercised that cD~ ~ F(x) in the is assumed to exist. Consider now .X aDx- v I = Assume f(t) f(x) = F~v) is expansible /a (x-t)v-lf(t)dt" in a Taylor's series co f(t) = 1 (-1)n f(n)(X)n! (x-t)n n=O The substitution of the series for f(t) in the integrand above gives co 1 (26a) I = ~ (-1) n f(n)(x ) (x-a) v+n (v+n) n! V (V) n=O Now if f(x) = (x-a) p, p > -i, then = F(p+v+l) where we have noted without proof r (v) ~ = 'r ( p + v + l ) If = _ the identity 1 ( v + l ) r (p) 1 vr(p+l) f(x) = (x-b) p , (x-a)V+P, r(p+l) _ aDx v(x-a)p p > -i, then from (x_a)V(x_b)p aDxV(x-b)P 1 + (v+2) 2!r (26a) y(_l)ncx_a~ F(v) (p-l) ~x-b j r(p+l) ...... (v+n) n! r(p-n+l) n=O forO<b~8. We recall D -u-v f(x) the laws of exponents is written or indices a D x-v a D x-u f(x) = for the case when both terminals of integra- a x tion are the s ~ e . a measure With of deviation the results just given one can investigate of the index rule, say for example, 23 f(x) = x: (See Open Questions, aDx v bDx u x. Some special arbitrary functions oD;(P+½) For Re(p) function. 376this text). as an integral of We wish to show the con- function: cos¢~¢~ = 2p /7 up-Zjo(/~). > -½ , we have Jp(X) = (x/2) p ¢7 P (P+½) Make the t r a n s f o r m a t i o n Jp(X) Let can be represented order of an elementary nection with the Bessel # 3, p. = (l-t2) p acos xt dt. [18] xt = w, the above becomes 2 (2x) P/7 r (p+½) SO x (x2-w2)p-½cos w dw. x 2 = u, w 2 = v, and the above becomes ~U 2pfg uP/2jp(¢~) = F(p+½)l *)Oi (u-v) p-½ c°scrv¢~ dv. These transformations forms COS ~/~ /g to our definition have given us an integral which (7), of arbitrary So, the above may be written 2p~-~ uP/2jp(~-~) = oD~(P+½) order p+½, and con - f(u) = in the form cos/~ ¢g which is the result we sought to verify. Here we show how a hypergeometric sented by the fractional operation function of a product can be repre- of elementary func- tions. (27) 1 + lab ~ g x + a ( a + l ).b ( .b + l.) .x 2 . + 2!g(g+l) is called a h y p e r g e o m e t r i c the geometric series series because 1 + x + x 2 -'- it is a generalization The following notations in common use: (r)n = (r+l) (r+2)'''(r+n-l), 2Fl(a,b ;g;x) • of are 24 The subscript 2 preceding F denotes two parameters in the numerator. The subscript 1 denotes one parameter in the denominator. Using this notation, (27) can conveniently be written in summation form: (28) 2Fl(a,b;g;x) (a)n(b)n n n!(g)n x . = n=O Some properties of the gamma and beta functions which will be needed later are briefly outlined. (b)n = b(b+l)'''(b+n-l) (29) (g)n g(g+l)...(g+n-1) Using the gamma-beta relation (30) Thus, (b)n ~= r(b+n) =~ F(b) r(~) "r(g+n) B(p,q) = F(p)F(q)/F(p+q), (29) becomes B(b+n. g-b] B(b, g-b) (28) becomes co ~ 1 2FI ta'b;g;xJ'' - B(b, g-b) (31) (a)/~B (b+n,g-b) .x n n! n-O where the factor I/[B(b,g-b)] is placed before the summation sign because it is independent of n. Writing symbol 2FI B(b+n, g-b) instead of as a beta integral, 2Fl(a,b;g;x), and using the we then have co (32) 1 2F1 = B ( b , g - b ) r ~ /__z n! xn /oI(l_t) g_b_itb+n_id t " n=O The interchange of the summation sign and the integral sign is permissible because of the uniform convergence of the series: oo (33) 1 2FI = B(b~g-b) ;oI(l-t) g_b_itb- 1 (a)n(Xt) n ni = Using the fact that dr. 25 n•=O (©n n! (xt)n = (1-xt)-a' we find that (33) becomes 1 2FI = ~ valid if (l-t)g Ixl < i, and All the right let the integral of the form of the definition on (7). xt = s, and we have x_g+l the relation gral above adt, that is required now is to transform ~x 2FI = B(b,g-b) Using - -l(l-xt) g,b > O. above to an integral To do this -b -Itb (x-s) g-b- isb-i (l-s) -ads . 10 B(b,g-b) in operator = r(b)r(g-b)/r(g), notation, we obtain and writing the inte- the result xg-lr (b) -(g-b)x(b-l(l_x )-a F(g) 2Fl(a'b ;g ;x) = oDx Before tional calculus, fractional There turning it will be useful integration appears our attention to some applications to mention and another another definition access to a fractional to be two representations of Hermann of frac- Weyl's of derivative. definition. One is f(x) = F(v) 1 rjJx~ x W-v~ The significant Liouville function W ~+B differences definition here being for all ~ (t-x) v-l. and L B • definition S. Miller differential integral derives operator d n-I + Pl (x) dx n- 1 + "'" + Pn (x)~ > O. and the Riemann- of integration When the Weyl Kenneth be the linear dx n this dt, Re(v) way. dn L = Po(X) between are the terminals gral in the following Let (t-x) v-I f(t) and the kernel exists, the Weyl W~W B = inte- 26 whose coefficients finite interval adjoint if f of L Pk' 0 ~ k ~ n, are of class I = [a,b] and and H (x,~) Po(X) its one-sided is any function continuous then for all > 0 on C~ on I. Green's I, and xo on some closed Let L be the function. Then is any point in I, x ~ I, (36) g(x) = H (x,~)f(~)d~ Xo is the solution satisfies of the nonhomogeneous the boundary g(k)( Now let xo = b cited on p.90 .) that function H (x,~) for Then if we let g(x) = L. =-H(~,x) where H(x,~) (See p. 37 of Miller's is text xo = b t t ( ~ , x ) f ( ~ ) d~ * is the solution of which 0 =< k =< n-l. and recall the one-sided Green's L y = f(x) conditions ) = O, X O equation L y = f(x) with initial g(k)(b) = O, We recall that conditions 0 < k < n-l. Now if g d n - ~ ~ dx n then L is formally self-adjoint for this p a r t i c u l a r L, since L = (-l)nL. (as in (i0)), H(x,~) i - (n-l)! g(x) r(n) (x_~)n-I Thus 1 - is the unique solution (-i) n ~xb (~_x)n-lf(~) d ~ of the adjoint equation dnX,' = f(x) dx n (with the initial may call conditions g(k) (b) = O, 0 =< k < n - 1 . ) So we 27 if xWbv f ( x ) - F(w) (g-x)W-lf(g)dg the adjoint fractional integral it) , Re v > 0 , (unless someone else has already named . Now for x fixed, lim sufficient a condition that xWb~f(x) b÷~ exists is f(x) , = 0 x < 0 and foo j < ix,21zfZ(x)d x 0 (Apply the Cauchy-Schwarz inequality.) Formally dx x w~'~f(x) = _x%(,~- 1 ) f(x) and, for e x a m p l e , xW~ (,J Make - 1) e x = -~ r-~) x = y the t r a n s f o r m a t i o n and we have d~-~ ~,(,;) e -x xW-(V~ -I) e -x d (¢ x)'~ le-~d~' e -x yU_le -Ydy r(,~) -X = One n o t e s e that -d m w -v f(x) dxmX~ = ( i] m w m - v ---x~ so that W-½ e-X ~ e-X, x ~ f(x) X > O. 28 and wm-½ x co for any nonnegative e-X integer = e-X m. The laws of exponents hold for The argument _ Re ~ > O and Re v > O. is similar to (17) and (18): _ , xW'~[x W~vf(x)] - F(~)F(~) (t-x)~-idt = r (~)r(~) f(<)d< fx (~-t)~-if(~)d~ (t-x)~-l(~-t)W-ldt oo B(~ ,v) = ~x f(~) (~-x) ~+V-ld~ X w-(~+v)f(x) co , which is a law of exponents. See also the paper by Kenneth S. Miller, this text, pp. 80 - 90. Another gorm of Weyl's series Mikol&s involving periodic definition in his paper this afternoon. by A. Zygmund in the treatment These definitions of certain One of the most recent methods fractional tend but their paper appears of Butzer and Westphal, function of defining ~f(t) the derivative = ~(-l)J(~) derivatives different are, unfortunately, later in this text. xY/F(y+l) of arbitrary is obtained fy (x) = I XY _ of qnotient. unable to at- Using the notation order as follows. f(x-tj) j=O ~ have been used Fourier series. order is by the limit of a fractional Paul L. Butzer and Ursula Westphal nonperiodic is that of an infinite functions which will be mentioned by Mikl6s (x > O) r(O+l) (x < O) ~ of the We define 29 [t-~&~f (x)] t y (s) = t -a 1--~ sY+l _ 1 sy-a+l (1-e-St) ~l-e-St) ~ st =~[(f~-a(u) ,1g pa(~))(x)](s) where the function pa(x) pa(x) - r(~) is defined by (-1)J (]) (x-j)a-1 X > O. O__<j<x pa(x) belongs to Thus we h a v e LI(o,~) t - a A at f y ( x ) The f o l l o w i n g the purposes is admittedly kernel consideration equation is for an opening transform a bit s-a(i-e-S) function contrived, of the form in the physical it is not a by fractional serves (x-t) v sciences. too far a n d money f l o w i n s t e a d solved The problem is to determine notch, Laplace t ÷ O+ formulated of economics,demand The i n t e g r a l for problem, equation deserves a backdrop + fy-a o f s h o w i n g how t h e a n d an i n t e g r a l frequency and has the fetched of fluid Its to use flow. operations. the shape f(y) of a weir in a dam, in which the volume flow rate of fluid, Q, through the notch is expressed the notch~ [19]. We first establish Q(h) Consider = c as a function h of (h-y)½f(y)dy. front and side views y of the height the equation of the notch below: Y JlJiiiiJlJlJiiiJiiJJ ! , Th Side view Fig. 1 Front view Fig. 2 30 Assuming the points a (37) y in Fig. equation I, [20],we can be applied between obtain Pa V2 V2 - - + gh + a = Py + gY + _y_ p 2 p 2 ' where and that Bernoulli's and Pa' Py' Va' Vy y; g are the pressures is the gravitational and velocities acceleration and p at points is the fluid density. The pressures atmospheric so that be negligible ing. Thus, at a and Pa = Py' (V a = O) since y are both and the velocity the fluid behind taken to be nearly at a is assumed the notch to is slow mov- (37) becomes o (38) gh = gy + V~/2, so that (59) gives V the velocity Y = (2g) ½(h-y) ½ y of the fluid at distance above the notch floor (x-axis). The elemental (40) Denoting 2(2g) ½ the elemental by c volume (42) Q(h) We find (43) region in Fig. 2) is given by volume flow rate through dA is to y = h dQ = V dA = 2 ( 2 g ) ½ ( h - y ) 2 f ( y ) d y . Y gives the total fractional (shaded dA = 2f(y) dy. So, by definition, (41) area f(y) integration Q(h) and integrating flow rate through = c from y = O the notch: (h-y) ½f(y)dy. by finding (7)~ Eq. = r(~) (41) (42) f(h). By the definition can be written in the form oDh 3/2 f(h). C Operating (44) on both sides of (43) with f(h) i F(3/2) oD~/2 oD~/2qc(h) gives of the result 31 312 But - oDh = O D2 oDh ½ g(h), we can write 2 where (44) oDh is d21dh 2 . Denoting Q(h)/c by as follows: -h 1 M 2 1 fO (h-Y) -½g(y) dY" f ( h ) - F(3/2) dh 2 F(%) (45) Since Q(h)/c i s known, then after evaluating second the beta derivative for and if a > -½. a = 3/2, R.I. lar symmetry fluid exits by suggests above where the weir a similar height We will now = O. its (45) yields like a parabola, are h at and H. of circu- and maximum heights H respectively. of The The time for the fluid to the ground, Determine Robert M. Hashway, A fluid reservoir The minimum above consider for the nth derivative Then is shaped problem. an orifice t(H) = ha . to this writer, the ground through g(h) and taking f(y). is a rectangle. communication a particular t(z) a = 7/2, is to be designed. the reservoir reach let above, we have Then, r (a-h) the weir In his Warwick, If Thus, are known. r(a+l__.___~) ha_3/2 7[ 2 valid g(y) on the right f(h). example f(h) - 2 and integral we obtain As a specific g(h) say height the shape a generalization z, is given of the reservoir. of Leibnitz's rule of a product: oo oDVf(x) g(x) (46) = ~, ~(n) (v) OVx f(x) oD(V-n) g (x) . n=O D (n) is ordinary entiation. differentiation Consider and D (v-n) is fractional differ- the identity x a+b = xax b Operate on both sides with tional derivative right hand side like powers, F(a+b+l) r(a+b+v+l) D v. Treat the left hand side as a frac- in accord with the definition in accord with we get an infinite = r(b+l) I I F(b+v+l) (46). series (21) By equating and treat the coefficients of of gamma functions: va F(b+v+2) v(v+l)a(a -I) + 2!F(b+v+3) I "'" " 32 The case for fractional plicity it offers the Volterra type. (483 on b o t h of explicitly for convenience F(½) D -½ f(x). O x we have = ¢~ D "½ f ( x ) . sides D2xf(x) of the above with D½ yields = ¢~ f ( x ) . (46) to get (50) xD½f(x) Substituting equations f(x) (x-t) -½ f(t)dt. subscripts xf(x) Apply formula = (7) the right hand side above is Omitting (49) integral 2x xf(x) Operating might well lie in the sim- Consider the problem of finding given the equation By definition calculus in the solution of certain + 1D-½f(x) = ¢~ f ( x ) . (48) into (50) gives (51) xD½f(x) x f ( x ) .... + 2vr~ We can get an expression for = ¢~ f ( x ) . D½f(x) by operating on both sides of (48) with D: (52) D[xf(x)] = ¢~ D½ f ( x ) , or (53) xf'(x) Our objective + f(x) = ~ D ½ f(x). has been reached when We arrive at the ordinary x2f,(x) (53) is substituted differential into (51). equation + ( -3yx _ 7 ) f(x) = 0 which has the solution f(x) = ke-~/Xx -3/2, Murray R. Spiegel, texts in the Schaum's outline to the previous problem. author of Laplace Transforms and other series, suggests the following solution 33 xF(x) = _ d_ ~0X( x - u ) - ½ F ( u ) d u = x -½ * F ( x ) . f(s) = ~ = £(½---~) f ( s ) ds - s½ f'(s)/f(s) f(s) s~ ' = / ' ~ s -½ In f(s) = 2/-~ s ½ + c I f(s) = ce -2/-~ F(x) = c X -I e - - e -4~/4x 2 ~JTTT~ = cx-3/2e-~/x No claim can be made that the fractional is better than some other approach. Parker Higgins who confided However, calculus approach to paraphrase Theodore in me tha~ he paraphrased there is a succinctness of notation the fractional that might suggest a solution functional calculus equation that is not readily In 1940 and 1941 Erd&lyi of a generalization tions. Professor calculus sixties, lay relatively Higgins, techniques dormant for the solution, began. AI-Bassam, Of particular More papers Osler and interest by means of fractional equations of mathematical to the of some operations, that stem from mixed boundary physics. The pair of equations f O ~ K(x,t)G(t)f(t)dt The from 1941 to the in the last decade was the development of dual and triple integral value problems properties and of the Weyl defini- resurgence Mikol~s, in the 1960's and early 1970's. applied mathematician formal and Kober investigated when a modest in to a complicated obtained by other means. of the Riemann-Liouville were published by Erd~lyi, others of formulation Sneddon will survey some of these results. topic of fractional early nineteen and simplicity A. Erd&lyi, = g(x) O<x<l, 34 O~ K ( x , t ) f ( t ) d t where the f(t) is kernel K(x,t), G(t), t o be d e t e r m i n e d to are a specific g(x), is gral equations, tial in the field of an electrified for ient to determine gral equations. problem finding known f u n c t i o n s integral to an e x p r e s s i o n of fractional fill this void. for calculus it is often conven- is to popularize to include ricula. is to exchange Another purpose in the in his lectures. Many matheThe wide will help to singular purposes, One obvious the topic in the hope it will induce tists and mathematicians inte- further trends at this conference has several hold a set of triple singular being taken in the sense of Sherlock Holmes. purpose the poten- conditions are unfamiliar with this topic. This conference inte- boundary is old but studied little. to be presented The of dual parts of the same boundary. parts of the same boundary, and scientists of papers a pair is such that different calculus and equations. disc where different Professor Mikol~s will discuss Fractional variety are dual the solution by constructing theory and applications maticians in hold over two different over three different h(x) physical example, When the problem x > 1 known as idea conditions reduce = h(x) it in their research scien- and cur- and impart information which may serve to suggest new areas of research. Fractional matics. proper objects in the last decade, in various chemistry, can be categorized fields: scattering rheology, theory, theory and elasticity. and scientists are unfamiliar However, another objective with mathematical fractional calculus electro- theory, probabil- many mathematicians with this topic possibly because calculus has developed, of this conference formal methods they Thus, while the theory its use has lagged behind. is to encourage of representing attempts physical So, to dis- phenomena models that can be treated with the elegance calculus. are and applied biology, transport have not been exposed to its applications. cover additional Scientists quantitative diffusion, mathe- operators found the fractional ity, potential of fractional as applicable and theory of these fractional of study in their own right. mathematicians, useful calculus The properties of 35 REFERENCES [I] Leibnitz, G.W., Leibnitzen's Mathematische Schriften, Germany: Georg Olm, 1962, v. 2, pp. 301-302. Hildesheim, [2] Lacroix, S.F., Trait~ du Calcul DiffJrentiel et du Calcul Integral, Paris: Mme. vecourcier, 1819, Tome Troisi&me, seconde &dition, pp. 409-410. [3] Spanier, Jerome and Oldham, Keith B., The Fractional Calculus, New York: Academic Press, 1974. [4] Abel, Niels Henrik, "Solution de quelques probl~mes a'l'aide d'int&grales d~finies," Oeuvres Completes, Christiania, 1881, tome premiere, 16-18. [5] Liouville, Joseph, "M&moire sur quelques Qu&stions de G&ometrie et de M&canique, et sur un nouveau genre de Calcul pour r&soudre ces Qu&stions," Journal de l'Ecole Polytechnique, 1832, tome XIII, XXI e cahier, pp. 1-69. [6] A more detailed discussion of Liouville's first and second definitions and also of their connection with the Riemann definition can be found in The Development of the Gamma Function and A Profile of Fractional Calculus, by Bertram Ross, New York University dissertation, 1974, Chapter V, pp. 142-210. University Microfilms, Ann Arbor, Mich., #74-17154, PO #45122. [7] Debnath, Lokenath and Speight, T.B., "On Generalized Derivatives," Pi Mu Epsilon Journal, v. 5, 1971, ND 5, pp. 217-220, East Carolina University. [8~through[ll] Details will be found in "A Chronological Bibliography of Fractional Calculus with Commentary," by Bertram Ross in The Fractional Calculus [3], pp. 3-15, and in [6]. [12] Davis, Harold Thayer, The Theory of Linear Operators, ton, Indiana: The Principia Press, 1936; p. 20. [13] Blooming- See [6], pp. 158-162. [14] The first one to apply Dirichlet's method to kernels of the form (x-t) v is Wallie Abraham Hurwitz in 1908. Cited by Whittaker and Watson, A Course in Modern Analysis, 4th edition, 1963, p. 76. [15] I am indebted to Dr. Kenneth S. Miller, Riverside Research Institute, New York City, for this contribution. [16] This approach was recommended by George F. Carrier, Harvard University. [17] Love, Eric Russell, "Fractional Derivatives of Imaginary Order," The Journal of the London Mathematical Society, Volume III (Second Series), 1971, pp. 241-259. [18] Farrell, Orin J. and Ross, Bertram, Solved Problems in Analysis, New York: Dover Publications, 1971, 279. First published in 1963, New York: The Macmillan Co. 36 [19] Brenke, W.C., "An Application of Abel's Integral Equation," American Mathematical Monthly, 1922, v. 29, 58-60. [20] Bernoulli's equation is strictly valid for steady, frictionless flow in a stream tube. It is used, however, in engineering for flows with friction by modification of solutions with a suitable friction factor.