Sie sind auf Seite 1von 33

Essentials of Fra

tional Cal ulus


Rudolf Goren o a and Fran es o Mainardi b 1
;

Department of Mathemati s and Computer S ien e, Free University of Berlin,


Arnimallee 2-6, D-14195 Berlin, Germany.

gorenflomath.fu-berlin.de

Dipartimento di Fisi a, Universita di Bologna and INFN Sezione di Bologna,


via Irnerio 46, I{40126 Bologna, Italy.
mainardibo.infn.it

Abstra t

The aim of these introdu tory le tures is to provide the reader with the essentials
of the fra tional al ulus a ording to di erent approa hes that an be useful for our
appli ations in the theory of probability and sto hasti pro esses. We dis uss the
linear operators of fra tional integration and fra tional di erentiation, whi h were
introdu ed in pioneering works by Abel, Liouville, Riemann, Weyl, Mar haud, M.
Riesz, Feller and Caputo. Parti ular attention is devoted to the te hniques of Fourier
and Lapla e transforms for treating these operators in a way a essible to applied
s ientists, avoiding unprodu tive generalities and ex essive mathemati al rigor. Furthermore, we dis uss the approa h based on limit of di eren e quotients, formerly
introdu ed by Grunwald and Letnikov, whi h provides a dis rete view-point to the
fra tional al ulus. Su h approa h is very useful for a tual numeri al omputation
and is omplementary to the previous integral approa hes, whi h provide the ontinuous view-point. Finally, we give some information on the trans endental fun tions
of the Mittag-Leer and Wright type whi h, together with the most ommon Eulerian fun tions, turn out to play a fundamental role in the theory and appli ations
of the fra tional al ulus.

Keywords: 1991 Mathemati s Subje t Classi ation: 26A33, 33E20, 44A20,


45E10, 45J05.

A. Histori al Notes on Fra tional Cal ulus


Fra tional al ulus is the eld of mathemati al analysis whi h deals with the
investigation and appli ations of integrals and derivatives of arbitrary order.
The term fra tional is a misnomer, but it is retained following the prevailing
use.
1

Corresponding author.

Preprint submitted to MaPhySto Center, preliminary version,

28 January 2000

The fra tional al ulus may be onsidered an old and yet novel topi . It is an
old topi sin e, starting from some spe ulations of G.W. Leibniz (1695, 1697)
and L. Euler (1730), it has been developed up to nowadays. In fa t the idea
of generalizing the notion of derivative to non integer order, in parti ular to
the order 1/2, is ontained in the orresponden e of Leibniz with Bernoulli,
L'H^opital and Wallis. Euler took the rst step by observing that the result
of the evaluation of the derivative of the power fun tion has a a meaning for
non-integer order thanks to his Gamma fun tion.
A list of mathemati ians, who have provided important ontributions up to
the middle of the 20-th entury, in ludes P.S. Lapla e (1812), J.B.J. Fourier
(1822), N.H. Abel (1823-1826), J. Liouville (1832-1837), B. Riemann (1847),
H. Holmgren (1865-67), A.K. Grunwald (1867-1872), A.V. Letnikov (18681872), H. Laurent (1884), P.A. Nekrassov (1888), A. Krug (1890), J. Hadamard
(1892), O. Heaviside (1892-1912), S. Pin herle (1902), G.H. Hardy and J.E.
Littlewood (1917-1928), H. Weyl (1917), P. Levy (1923), A. Mar haud (1927),
H.T. Davis (1924-1936), A. Zygmund (1935-1945), E.R. Love (1938-1996), A.
Erdelyi (1939-1965), H. Kober (1940), D.V. Widder (1941), M. Riesz (1949),
W. Feller (1952).
However, it may be onsidered a novel topi as well, sin e only from less than
thirty years it has been obje t of spe ialized onferen es and treatises. For the
rst onferen e the merit is as ribed to B. Ross who organized the First Conferen e on Fra tional Cal ulus and its Appli ations at the University of New
Haven in June 1974, and edited the pro eedings [79. For the rst monograph
the merit is as ribed to K.B. Oldham and J. Spanier [71, who, after a joint
ollaboration started in 1968, published a book devoted to fra tional al ulus
in 1974.
Nowadays, to our knowledge, the list of texts in book form devoted to fra tional al ulus in ludes less than ten titles, namely Oldham & Spanier (1974)
[71 M Bride (1979) [63, Samko, Kilbas & Mari hev (1987-1993) [83, Nishimoto (1991) [70, Miller & Ross (1993) [64, Kiryakova (1994) [46, Rubin
(1996) [80, Podlubny (1999) [74, among whi h the en y lopaedi treatise
by Samko, Kilbas & Mari hev is the most prominent. Furthermore, we re all the attention to the treatises by Davis (1936) [17, Erdelyi (1953-1954)
[22, Gel'fand & Shilov (1959-1964) [26, Dzherbashian [19,20, Caputo [13,
Babenko [4, Goren o & Vessella [37, whi h ontain a detailed analysis of
some mathemati al aspe ts and/or physi al appli ations of fra tional al ulus, although without expli it mention in their titles.
For details on the histori al development of the fra tional al ulus we refer
the interested reader to Ross' bibliography in [71 and to the histori al notes
generally available in the above quoted texts.
2

In re ent years onsiderable interest in fra tional al ulus has been stimulated
by the appli ations that it nds in di erent elds of s ien e, in luding numeri al analysis, physi s, engineering, biology, e onomi s and nan e. In this
respe t we quote the olle tion of arti les on the topi of Fra tional Di eren ing and Long Memory Pro esses, edited by Baillie & King (1996), appeared as
a spe ial issue in the Journal of E onometri s [5, the book edited by Carpinteri and Mainardi (1997), ontaining le ture notes of a CISM Course devoted
Fra tals and Fra tional Cal ulus in Continuum Me hani s [15, and nally the
forth oming book edited by R. Hilfer (2000), ontaining invited ontributions
in some areas of physi s [41.

B. Approa hes to Fra tional Cal ulus


There are di erent approa hes to the fra tional al ulus whi h, not being all
equivalent, have lead to a ertain degree of onfusion and several misunderstandings in the literature. Probably for this the fra tional al ulus is in some
way the "bla k sheep" of the analysis. In spite of the numerous eminent mathemati ians who have worked on it, still now the fra tional al ulus is obje t
of so many prejudi es.
In these review le tures we essentially onsider and develop two di erent
view-points to the fra tional al ulus: the ontinuous view-point based on the
Riemann-Liouville fra tional integral and the dis rete view-point based on the
Grunwald-Letnikov fra tional derivative. Both approa hes turn out to be useful in treating our generalized di usion pro esses in the theory of probability
and sto hasti pro esses.
We use the standard notation N, Z, R, C to denote the sets of natural, integer,
real and omplex numbers, respe tively; furthermore, R+ and R+0 denote the
sets of positive real numbers and of non-negative real numbers, respe tively.
Let us remark that, wanting our le tures to be a essible to various kinds of
people working in appli ations (e.g. physi ists, hemists, theoreti al biologists,
e onomists, engineers) we have deliberately and ons iously as far as possible
avoided the language of fun tional analysis. We have used vague phrases like
"for a su iently well behaved fun tion" instead of onstru ting a stage of
pre isely de ned spa es of admissible fun tions. We have devoted parti ular
attention to the te hniques of Fourier and Lapla e transforms: orrespondingly
our fun tions are required to belong to the spa e L1 (R) (summable fun tions in
all of R) or Llo (R+0 ) (summable fun tion in any nite inerval of R+0 ) : We have
extended the Fourier and Lapla e transforms to the Dira "delta fun tion" in
the typi al way suitable for appli ations in physi s and engineering, without
adopting the language of distributions. We kindly ask spe ialists of these elds
of pure mathemati s to forgive us. Our notes are written in a way that makes
it easy to ll in details of pre ision whi h in their opinion might be la king.
3

The ontinuous view-point to fra tional al ulus

The starting point of the so alled Riemann-Liouville fra tional al ulus is the
integral formula (attributed to Cau hy) that redu es the al ulation of the
n fold primitive of a (su iently well behaved) fun tion (x) (x 2 [a; b  R ;
1  a < b  +1) to a single integral of onvolution type. Indeed, for any
n 2 N ; the repeated integral

Ian+ (x)

:=

x xZn 1

:::

x1

(x0 ) dx0 : : : dxn 1 ; a  x < b ;

whi h provides the n fold primitive n (x) ; vanishing at x = a with its derivatives of order 1; 2; : : : ; n 1 ; an be written be ause of the Cau hy formula
as

Ian+ (x)

= n (x) =

(n 1)!

(x  )n 1 ( ) d ; a  x < b :

(B:1)

Then, in a natural way, one is led to extend the above formula from positive
integer values of the index n to any positive real values by using the Gamma
fun tion. Indeed, noting that (n 1)! = (n) ; and introdu ing the arbitrary
positive real number ; one de nes the fra tional integral of order > 0 as

Ia + (x)

1
:=
( )

(x  ) 1 ( ) d ; a < x < b > 0 :

(B:2)

A dual form of the above integral is

Ib

1
(x) :=
( )

(

x) 1 ( ) d ; a < x < b > 0 :

(B:3)

We refer to the fra tional integrals Ia + and Ib as progressive and regressive,


respe tively. For omplementation we de ne Ia0+ = Ib0 := I (Identity operator), i.e. we mean Ia0+ (x) = Ib0 (x) = (x) : Furthermore, by Ia + (a+ ) ;
Ib (b ) ; we mean the limits (if they exists) of Ia + (x) for x ! a+ and
Ib (x) for x ! b ; respe tively; these limits may be in nite.
The most ommon hoi es on erning the interval [a; b are the whole set of
real numbers R (i.e. a = 1 ; b = +1), onsidered by Liouville [50{52, and
the set of non negative real numbers R+0 (i.e. a = 0 ; b = +1), onsidered by
Riemann [77. We note that the fra tional integrals over in nite intervals, espe ially the regressive one, were named in many later papers as Weyl integrals.
Weyl [89 arrived at these indire tly by de ning fra tional integrals suitable
for periodi fun tions. We thus agree to refer to these integrals as LiouvilleWeyl fra tional integrals. In these le tures we shall use the simpli ed notation
I+ and I when a = 1 and b = +1 ; respe tively.
4

We also note that, before Riemann, fra tional integrals with a = 0 have been
onsidered by Abel [1,2 when he introdu ed his integral equation, named after
him, to treat the problem of the tauto hrone. It was Abel who gave the rst
appli ation of fra tional al ulus to me hani s in solving his problem by inverting the fra tional integral, see e.g. Goren o & Vessella (1991) [37. In these
as to Abel-Riemann
le tures we agree to refer to the fra tional integral I0+
fra tional integral; for them we use the spe ial and simpli ed notation J in
agreement with the notation introdu ed by Goren o & Vessella (1991) [37
and then followed in any paper of ours.
Before introdu ing the fra tional derivative let us point out the fundamental
property of the fra tional integrals, namely the semi-group property a ording
to whi h
Ia + Ia + = Ia ++ ; Ib Ib = Ib + ; ;  0 :
(B:4)
The fra tional derivative of order an be introdu ed as the left inverse of
the orresponding fra tional integral, so extending the similar property of the
ommon derivative of integer order. In fa t it is straightforward to re ognize
that the derivative of any integer order n = 0; 1; 2; : : :

dn
(x) = (n) (x) ; a < x < b
dxn
satis es the following omposition rules with respe t to the repeated integrals
of the same order n ; Ian+ (x) and Ibn (x)
Dn (x) =

8
>
>
>
<

Dn Ian+ (x) = (x) ;

>
>
>
:

Ian+ Dn (x) = (x)

8
>
>
>
<

Dn Ibn (x) = ( 1)n (x) ;

>
>
>
:

Ibn

a < x < b;
(k) (a+ )
(x a)k ;
k!
k=0

nX1

Dn (x)

=(

1)n

(x)

nX1 (k) (b

k!

k=0

(b

x)k

(B:5a)

a < x < b : (B:5b)

As a onsequen e of (B.3-5) the left-inverse of the fra tional integrals Ia + ; Ib


may be de ned by introdu ing the positive integer m su h that m 1 <  m :
Then one de nes the fra tional derivative of order > 0 as
8
>
<

Da + (x) := Dm Iam+ (x) ;

>
:

Db

(x) := (

a < x < b ; m 1 <  m : (B:6)

1)m Dm Ibm (x) ;

In fa t, taking for example the progressive operators, we get

Da + Ia + = Dm Iam+ Ia + = Dm Iam+ = I :
5

The fra tional derivatives, like the fra tional integrals, turn out to be ontinuous with respe t to the order, redu ing to the to the standard repeated
derivatives when the order is an integer. However, when the order is not integer, the fra tional derivatives (namely the "proper" fra tional derivatives) do
not follow ne essarily the "semi-group" property of the fra tional integrals: in
this respe t the starting point a 6= 1 (or the ending point b 6= +1) plays
a "disturbing" role. Furthermore we stress the fa t that "proper" fra tional
derivatives are non-lo al operators being expressed by ordinary derivatives of
onvolution integrals with a weakly singular kernel. We note that the fra tional integrals ontain a weakly singular kernel only when the order is less
than one.
The dis rete view-point to fra tional al ulus

In all the above approa hes in whi h the fra tional derivatives are de ned as
the left inverse of the orresponding fra tional integrals, the fra tional differentiation is seen as a sort of integration of order : A totally di erent
approa h arises from the desire to properly generalize the fa t that ordinary
derivatives are limits of di eren e quotients. Let T h denote the translation by
a step of length h > 0
T h (x) = (x h) :
(B:7)
The ba kward nite di eren e of order is de ned as
1


(x kh) ;
k

(B:8)

( 1) : : : ( k + 1)

( + 1)
=
=
:
k
k!
(k + 1) ( k + 1)

(B:9)

 h (x)

:= (I

T h ) (x) =

where I = T 0 ; the identity, and

k=0

1)k

For 2 N it redu es to the familiar ba kward nite di eren e of integer


order. For 62 N we note the asymptoti behabviour




j sin( )j j (k )j  ( +1) j sin( )j k



=
( +1)

k

(k + 1)

!

( +1)

as k ! 1 :

Grunwald (1867) [39 and Letnikov (1868) [47 developed an approa h to fra tional di erentiation based on the de nition

D+ (x) = lim+
h!0

 h (x)
:
h

(B:10)

While the arguments of the rst author were rather formal, the latter gave a
rigorous onstru tion of the theory of fra tional integro-di erentiation on the
basis of su h a de nition. Letnikov had in parti ular shown that thus de ned
6

D+ (x) oin ides with our Liouville-Weyl fra tional integral I+ and with our
Abel-Riemann fra tional integral J under the appropriate interpretation of
the fra tional di eren e  h (x) : He proved the semigroup property within
the framework of de nition (B.10).
Hilfer (1997) [40 has pointed out that there are several possibilities to de ne
the limit of the fra tional nite di eren e quotient, e.g. point-wise, almost
everywhere, or in the norm of a Bana h spa e. The hoi e depends upon the
question at hand. Furthermore he notes that eq. (B.8) suggests it is also possible to de ne fra tional derivatives as fra tional powers of the di erentiation
operator following the approa h started by Balakrishnan (1958, 1959, 1960)
[6{8 and Westphal (1974) [88. More generally one may onsider fra tional
powers of the in nitesimal generators of strongly ontinuous semi-groups.
Finally, a mixed approa h to the fra tional al ulus, namely ontaining integrals and nite di eren es, is to de ne the fra tional derivatives trying to
repla e with dire tly in the above Riemann-Liouville fra tional integrals. However the resulting integrals are divergent (with hyper-singular kernels) and need to be regularized by using the te hniques of Hadamard's nite
part. This approa h was su essfully pursued by Mar haud (1927) [62. It is
interesting to note that both Liouville and Riemann dealt with the so- alled
" omplementary" fun tions whi h arise when one attempts to treat fra tional
di erentiation of order as fra tional integration of order ; see Samko,
Kilbas & Mari hev (1993) ([83, p. xxix and Histori al Notes in x4.1, x9.1).
The plan for the following se tions

Let us now explain the ontents of the following se tions. Se tions C and D
will be devoted to the ontinuous view-point to fra tional al uls. In se tion C
we start from the Liouville-Weyl fra tional integrals to arrive at the fra tional
derivatives in the sense of Riesz and Feller. These derivatives are suitable to
generalize the standard di usion equation by repla ing the se ond-order spa e
derivative. In view of this we shall onsider fun tions of the spa e variable x ;
denoted by (x) ; and apply the Fourier transform. In se tion D we start from
the Abel-Riemann fra tional integrals to arrive at the fra tional derivatives
in the sense of Caputo. These derivatives are suitable to generalize the standard di usion equation by repla ing the rst-order time derivative. In view of
this we shall onsider fun tions of the time variable t ; denoted by (t) ; and
apply the Lapla e transform. In se tion E we shall provide some details on
the dis rete view-point to fra tional al ulus based on the Grunwald-Letnikov
di eren e s heme and its variants. These s hemes turn out to be useful in the
interpretation of the spa e or time fra tional di usion pro esses by randomwalk models. Finally se tions F and G are devoted respe tively to the spe ial
fun tions of the Mittag-leer and Wright type, whi h play a fundamental role
in our appli ations of the fra tional al ulus.
7

C. Fra tional Cal ulus a ording to Riesz and Feller


In this Se tion the fun tions under onsideration are assumed su iently well
behaved in L1 (R) to ensure the existen e of the Fourier transform or its inverse, as required. In our notation the Fourier transform and its inverse read
8
>
>
>
b
>
>
>
>
<

() = F [(x) =

>
>
>
>
>
>
>
:

(x) = F

+
Z

e +ix (x) dx ;
1

1 Z
b
() =
e ix b() d ;
2 1
i

(C:1)

where  2 R ; denotes the Fourier parameter. In this framework we also onsider the lass of pseudo-di erential operators of whi h the ordinary repeated
integrals and derivatives are spe ial ases. A pseudo-di erential operator A,
a ting with respe t to the variable x 2 R ; is de ned through its Fourier
representation, namely
1

+
Z

e ix A (x) dx = Ab() b() ;

(C:2)

where Ab() is referred to as symbol of A : An often appli able pra ti al rule


is


(C:3)
Ab() = A e ix e +ix ;  2 R :
If B is another pseudo-di erential operator, then we have AdB () =
Ab() Bb () :
For the sake of onvenien e we shall adopt the notation  to denote the
juxtaposition of a fun tion with its Fourier transform and that of a pseudodi erential operator with its symbol, namely

(x)

 () ; A  A :
b

The Liouville-Weyl fra tional integrals and derivatives

We now onsider the pseudo-di erential operators represented by the LiouvilleWeyl fra tional integrals and derivatives. The Liouville-Weyl fra tional integrals (of order > 0 for a well-behaved fun tion (x) with x 2 R ) are
de ned
8
x
>
1 Z
>

>
>
(x  ) 1 ( ) d ; > 0 ;
I+ (x) :=
>
>
( ) 1
<
(C:4)
+
Z1
>
>
1
>


1
>
>
I (x) :=
( x) ( ) d ; > 0 :
>
:
( ) x
8

For omplementation we put I0 := I (Identity operator), as it an be justi ed


by passing to the limit ! 0 :
The Liouville-Weyl integrals possess the semigroup property, i.e.

I+ I+ = I+ + ; I I = I + ; for all ;  0 :

(C:5)

The Liouville-Weyl fra tional derivatives (of order > 0) are de ned as the
left-inverse operators of the orresponding Liouville-Weyl fra tional integrals
(of order > 0), i.e.
D+ I+ = D I = I :
(C:6)
Therefore, introdu ing the positive integer m su h that m 1 <  m and
re alling that Dm I+m = I and Dm I m = ( 1)m I ; we have

(Dm I m ) (x) ; if
D (x) = (Dm I m  ) (x) ; if

(

m 1 <  m ; m odd;
m 1 <  m ; m even.

(C:7)

For omplementation we put D0 := I (Identity operator).


We note that a su ient ondition that the integrals entering I in (C.4)
onverge is that


 (x) = O jxj

;  > 0 ; x ! 1 :

Integrable fun tions satisfying these properties are sometimes referred to as


fun tions of Liouville and Weyl lass, respe tively, see Miller & Ross (1993)
[64. For example power fun tions jxj with > > 0 and x < 0 and e x
with > 0 are of Liouville lass. For these fun tions we obtain
8
>
>
>
>
<

)
jxj + ;
( )
> > 0; x < 0;
>
(

+

)
>


>
>
=
jxj ;
: D+ jxj
( )
I+ jxj =

and

8
>
<

I+ e x =

>
:

D e x
+

e x ;

e x ;

> 0; x 2 R:

(C:8)

(C:9)

The symbols of the fra tional Liouville-Weyl integrals and derivatives an be


easily derived a ording to
8
>
<

I

>
:

 (i)
 (i)

= jj e i (sign ) =2 ;

= jj+ e i (sign ) =2 :


9

(C:10)

The Riesz fra tional integrals and derivatives

The Liouville-Weyl fra tional integrals an be ombined to give rise to the


Riesz fra tional integral (usually alled Riesz potential) of order , de ned as

I (x)
0

Z
I+ (x) + I (x)
1
=
jx  j 1 ( ) d ; (C:11)
=
2 os( =2)
2 ( ) os( =2) 1
+

for any positive with the ex lusion of odd integer numbers for whi h
os( =2) vanishes. Using (C.10-11) the symbol of the Riesz potential turns
out to be
I0  jj ;
> 0 ; 6= 1 ; 3 ; 5 : : : :
(C:12)
In fa t

I+ + I

"

1
(+i) + ( i) 2 os( =2)
1
=
+
=
( i) (+i)
jj
jj :

We note that, at varian e with the Liouville-Weyl fra tional integral, the Riesz
potential has the semigroup property only in restri ted ranges, e.g.

I0 I0 = I0 + for 0 < < 1 ; 0 < < 1 ; + < 1 :

(C:13)

From the Riesz potential we an de ne by analyti al ontinuation the Riesz


fra tional derivative D0 , in luding also the singular ase = 1 ; by formally
setting D0 := I0  jj ; where the minus sign has been put in order to
re over for = 2 the standard se ond derivative. Restri ting our attention to
the range 0 <  2 the expli it orre t de nition turns out to be
8
>
>
<

D0 (x) := I0 (x) := >


>
:

D+ (x) + D (x) if 6= 1 ;
2 os( =2)
D H (x) ;
if = 1 ;

(C:14)

where H denotes the Hilbert transform operator de ned by


1

1 Z ( )
1 Z (x  )
H (x) :=
d =
d ;
 1x 
 1

+

(C:15)

the integral understood in the Cau hy prin ipal value sense. In identally, we
note that H 1 = H : By using the rule (C.3) we an derive the symbol of
H ; namely

HH
= i sign  :
(C:16)
The expressions in (C.14) an be easily veri ed by manipulating with symbols
10

of "good" operators as below

D
0

:= I0

 jj

8
>
>
<

=>

( i) + (+i) =
2 os( =2)

jj ;

+i  isign  =  sign  =

>
:

if 6= 1 ;

jj ;

if = 1 :

In parti ular, from (C.14) we re ognize that


!


1 d2
1
d2
d2
d
D02 = D+2 + D2 =
+
=
; but D01 6= :
2
2
2
2
2 dx dx
dx
dx

We also re ognize that the symbol of D0 is just the logarithm of the hara teristi fun tion of a symmetri Levy stable pdf .
In general, the Riesz fra tional derivative D0 turns out 2 to be related to the
=2-power of the positive de nite operator D2 = dxd 2 sin e, as noted by
Feller (1952) [23,

jj

=2

D
0

d2
dx2

=2

(C:17)

whereas the two Liouville-Weyl fra tional derivatives are related to the power of the rst order di erential operator D = dxd : We note that it was
Bo hner (1949) [11 who rst introdu ed the fra tional powers of the Lapla ian
to generalize the di usion equation.
We would like to mention the "illuminating" notation introdu ed by Zaslavsky,
see e.g. Sai hev & Zaslavsky (1997) [81, to denote our Liouville-Weyl and
Riesz fra tional derivatives

D =


d
= d ; 0 <  2:
;
D
0
d(x)
djxj

(C:18)

We now point out that other expressions for D and hen eforth D0 are obtained by "regularizing" the hyper-singular integrals I for 0 <  2 :
This "regularization", based on a former idea by Mar haud, see e.g. Mar haud (1927) [62, Samko, Kilbas & Mari hev (1993) [83, Hilfer (1997) [40,
is noteworthy sin e it leads to a dis retization of the operators of fra tional
derivative, alternative to that based on the Grunwal-Letnikov method.
Let us rst onsider from (C.7) the operator

D+ := I+ :=

d 1
I ; 0 < < 1:
dx +

We have, see Hilfer (1997) [40,


11

(C:19)

D+ (x) := I+ (x) =

d 1
I (x)
dx +
x

d Z
(x  ) ( ) d
=
(1 ) dx 1
1
1
d Z
 (x  ) d
=
(1 ) dx 0
1

(1 )

0 (x

)

1+

d d ;

so that, inter hanging the order of integration,

(x) =

(1 )

(x) (x  )
d ; 0 < < 1 :
 1+

(C:20)

Here 0 denotes the rst derivative of  with respe t to its argument. The
oe ient in front to the integral in (C.20) an be re-written, using known
formulas for the Gamma fun tion, as

(1 )

( )

= (1 + )

sin 
:


(C:21)

Similarly we get for

D := I

d 1
I ; 0 < < 1;
dx

(x +  ) (x)
d ; 0 < < 1 :
(1 ) 0
 1+
Similar results an be given for 1 < < 2 :
D

(x) =

(C:22)
(C:23)

Re alling from (C.14) the fra tional derivative in Riesz's sense

D0 (x) :=

D+ (x) + D (x)
; 0 < < 1; 1 < < 2;
2 os( =2)

and using (C.20), (C.23), we get for it the following regularized representation,
valid also in = 1 ;

D (x)
0

sin ( =2)
= (1+ )


(x +  ) 2(x) + (x  )


d ; 0 < < 2 :
 1+
(C:24)

12

The Feller fra tional integrals and derivatives

A generalization of the Riesz fra tional integral and derivative has been proposed by Feller (1952) [23 in a pioneering paper, re alled by Samko, Kilbas
& Mari hev (1993) [83, but only re ently revised and used by Goren o &
Mainardi (1998) [34. The purpose of Feller was indeed to generalize the se ond order spa e derivative entering the standard di usion equation with a
pseudo-di erential operator whose symbol is the logarithm of the hara teristi fun tion of a general Levy stable pdf a ording to his his parameterization,
whi h is loser to ours.
Let us now show how obtain the Feller derivative by inversion of a properly
generalized Riesz potential, later alled Feller potential by Samko, Kilbas &
Mari hev (1993) [83. Using our notation we de ne the Feller potential I by
its symbol obtained by the Riesz potential by a suitable "rotation" of an angle
 =2 where  is a real number properly restri ted. We have

I

 jj

i (sign ) =2 ;

jj 
jj  2

if 0 < < 1 ;
if 1 <  2 :

(C:25)

As in the Riesz potential the ase = 1 is omitted.


The integral representation of the pseudo-di erential operator I turns out to
be
I (x) = ( ; ) I+ (x) + + ( ; ) I (x) ;
(C:26)
where, if 0 < < 2 ; 6= 1 ;

+ ( ; ) =

sin [( ) =2
;
sin (  )

( ; ) =

sin [( + ) =2
;
sin(  )

(C:27)

and, by passing to the limit (with  = 0)

+ (2; 0) = (2; 0) = 1=2 :

(C:28)

In the parti ular ase  = 0 we get

+ ( ; 0) = ( ; 0) =

1
;
2 os ( =2)

(C:29)

and thus, from (C.26) and (C.29) we re over the Riesz potential (C.11). Like
the Riesz potential also the Feller potential has the (range-restri ted) semigroup property, e.g.

I I = I + for 0 < < 1 ; 0 < < 1 ; + < 1 :

(C:30)

From the Feller potential we an de ne by analyti al ontinuation the Feller


fra tional derivative D , in luding also the singular ase = 1 ; by setting

D := I

 jj e +i (sign ) =2 ;


13

0 <  2;

(C:31)

with  restri ted as in (C.25). We have


8
>
<

D (x) := >
:

+ ( ; ) D+ + ( ; ) D (x) ; if 6= 1 ;

[ os(=2) D + sin(=2) D (x) ;


1
0

if = 1 :

(C:32)

For 6= 1 it is su ient to note that  ( ; ) =  ( ; ) : For = 1 we need


to re all the symbols of the operators D and D01 = DH , namely D^ = ( i)
d
1
and D
jj ; and note that
0 =
d
+i (sign ) =2 = jj os(=2) (i) sin(=2)
1
D
 = jj e
d
1

= os(=2) D
0 + sin(=2) D :

For later use we nd it onvenient to return to the "weight" oe ients  ( ; )


in order to outline some properties along with some parti ular expressions,
whi h an be easily obtained from (C.27) with the restri tions on  given in
(C.25). We obtain

0 ; if 0 < < 1 ;
 
(C:33)
 0 ; if 1 <  2 ;
and
os (=2)  > 0 ; if 0 < < 1 ;
+ + =
(A:34)
os ( =2) < 0 ; if 1 <  2 :
In the extremal ases we nd

= 1 ; = 0 ; if  = ;
0 < < 1 ; + = 0 ; = 1 ; if  = + ;
(C:35)
+
0 ; = 1 ; if  = (2 ) ;
1 < < 2 ; + =
1 ; = 0 ; if  = +(2 ) :
+ =
We also note that in the extremal ases of = 1 we get
d
D1 1 = D =  :
dx


(C:36)
(C:37)

In view of the relation of the Feller operators in the framework of stable probability density fun tions, we agree to refer to  as to the skewness parameter.
We must note that in his original paper Feller (1952) used a skewness parameter di erent from our  ; the potential introdu ed by Feller is


2
; =
: (C:38)
2

Then Samko, Kilbas & Mari hev (1993) named I as the Feller potential
operator. We note that Feller onsidered the inversion of his potential for
= 1 but limiting himself to the symmetri ase ( =  = 0) for whi h
he provided the representation as the rst derivative of the Hilbert transform.
Samko, Kilbas & Mari hev (1993) apparently ignored the singular ase = 1 :
I

jj e i (sign )

; so =

14

D. Fra tional Cal ulus a ording to Abel-Riemann and Caputo


In this Se tion the fun tions under onsideration are assumed su iently well
behaved in Llo (R+0 ) to ensure the existen e of the Lapla e transform or its
inverse, as required. In our notation the Lapla e transform and its inverse read
1

8
>
>
e
>
>
>
<

(s) = L [ (t) =

>
>
>
>
>
:

(t) = L

e st (t) dt ;

1 Z st e
(s) =
e (s) ds ;
2i Br

(D:1)

where Br denotes a Bromwi h path and s 2 C is the Lapla e parameter.


For the sake of onvenien e we shall adopt the notation  to denote the
juxtaposition of a fun tion with its Lapla e transform, namely (t)  e(s) :
The Abel-Riemann fra tional integral and derivatives

We rst de ne the Abel-Riemann (A-R) fra tional integral and derivative of


any order > 0 for a generi (well-behaved) fun tion (t) with t 2 R+ :
For the A-R fra tional integral (of order ) we have

1
(t) :=
( )

(t  )

( ) d ; t > 0 > 0 :

(D:2)

For omplementation we put J 0 := I (Identity operator), as it an be justi ed


by passing to the limit ! 0 :
The A-R integrals possess the semigroup property, i.e.

J J = J + ; for all ;  0 :

(D:3)

The A-R fra tional derivative (of order > 0) is de ned as the left-inverse
operator of the orresponding A-R fra tional integral (of order > 0), i.e.

D J = I :

(D:4)

Therefore, introdu ing the positive integer m su h that m 1 <  m


and noting that (Dm J m ) J = Dm (J m J ) = Dm J m = I ; we de ne
D := Dm J m namely
8
>
>
>
>
<

dm
1
m
D (t) := > (m ) dt
>
>
dm
>
:
(t) ;
dtm

( )
(t  ) +1

d ; m 1 < < m ;
= m:

15

(D:5)

For omplementation we put D0 := I : For ! m we thus re over the


standard derivative of order m but the integral formula loses its meaning for
= m:
By using the properties of the Eulerian beta and gamma fun tions it is easy
to show the e e t of our operators J and D on the power fun tions: we have
8
>
>
>
>
<

( + 1) +
t ;
( + 1 + )
>
( + 1)
>

>
>
t ;
:D t =
( + 1 )

J t =

t > 0;  0; > 1:

(D:6)

These properties are of ourse a natural generalization of those known when


the order is a positive integer.
Note the remarkable fa t that the fra tional derivative D (t) is not zero for
the onstant fun tion (t)  1 if 62 N : In fa t, the se ond formula in (D.6)
with = 0 tea hes us that

D 1 =

t
;  0; t > 0:
(1 )

(D:7)

This, of ourse, is  0 for 2 N, due to the poles of the gamma fun tion in
the points 0; 1; 2; : : :.
The Caputo fra tional derivative

We now observe that an alternative de nition of fra tional derivative, originally introdu ed by Caputo (1967) (1969) in the late sixties and adopted by
Caputo & Mainardi (1971) in the framework of the theory of linear vis oelasti ity, is D = J m Dm with m 1 <  m ; namely
8
>
>
>
>
<

Z
(m)
1
( )

+1
D (t) := > (m ) 0 (t  )
>
>
dm
>
:
(t) ;
dtm

d ;

m 1 < < m;

(D:8)

= m:

The de nition (D.8) is of ourse more restri tive than (D.5), in that it requires
the absolute integrability of the derivative of order m. Whenever we use the
operator D we (ta itly) assume that this ondition is met. We easily re ognize
that in general

D (t) := Dm J m

(t) 6= J m Dm (t) := D (t) ;

(D:9)

unless the fun tion (t) along with its rst m 1 derivatives vanishes at t = 0+ .
In fa t, assuming that the passage of the m-derivative under the integral is
16

legitimate, one re ognizes that, for m 1 < < m and t > 0 ;

D (t) = D (t) +

m
X1
k=0

(k

tk

( )

+ 1)

(0+ ) ;

(D:10)

and therefore, re alling the fra tional derivative of the power fun tions, see
(D.6),
!
m
X1 tk
(k )
+

(0 ) = D (t) :
(D:11)
(t)
D
k
!
k=0
The subtra tion of the Taylor polynomial of degree m 1 at t = 0+ from
(t) means a sort of regularization of the fra tional derivative. In parti ular,
a ording to this de nition, the relevant property for whi h the fra tional
derivative of a onstant is still zero an be easily re ognized,

D 1  0 ; > 0 :

(D:12)

We now explore the most relevant di eren es between the two fra tional
derivatives (D.5) and (D.8). We agree to denote (D.8) as the Caputo fra tional derivative to distinguish it from the standard A-R fra tional derivative
(D.5). We observe, again by looking at (D.6), that D t 1  0 ; > 0 ; t > 0 :
We thus re ognize the following statements about fun tions whi h for t > 0
admit the same fra tional derivative of order ; with m 1 <  m ; m 2 N ;
= D (t)

() (t) = (t) +

D (t) = D (t)

() (t) = (t) +

(t)

where the oe ients j are arbitrary onstants.

j =1
m

j =1

j t j ;

(D:13)

j tm j ;

(D:14)

For the two de nitions we also note a di eren e with respe t to the formal
limit as ! (m 1)+ ; from (D.5) and (D.8) we obtain respe tively,

D (t) ! Dm J (t) = Dm
D (t) ! J Dm (t) = Dm

(t)

(t) ;
m

1)

(D:15)
(0+ ) :

(D:16)

We now onsider the Lapla e transform of the two fra tional derivatives. For
the A-R fra tional derivative D the Lapla e transform, assumed to exist,
requires the knowledge of the (bounded) initial values of the fra tional integral J m and of its integer derivatives of order k = 1; 2; : : : ; m 1 : The
orresponding rule reads, in our notation,

D (t)  s fe(s)

mX1
k=0

Dk J (m

(0+ ) sm
17

k;

m 1 <  m : (D:17)

For the Caputo fra tional derivative the Lapla e transform te hnique requires
the knowledge of the (bounded) initial values of the fun tion and of its integer
derivatives of order k = 1; 2; : : : ; m 1 ; in analogy with the ase when = m :
In fa t, noting that J D = J J m Dm = J m Dm ; we have

J D (t) = (t)

m
X1
k=0

( )

(0+ )

tk
;
k!

(D:18)

so we easily prove the following rule for the Lapla e transform,

D (t)  s e(s)

m
X1
k=0

( )

(0+ ) s

k;

m 1 <  m:

(D:19)

Indeed the result (D.19), rst stated by Caputo (1969) [13, appears as the
"natural" generalization of the orresponding well known result for = m :
Goren o & Mainardi (1997) [33 have pointed out the major utility of the
Caputo fra tional derivative in the treatment of di erential equations of fra tional order for physi al appli ations. In fa t, in physi al problems, the initial
onditions are usually expressed in terms of a given number of bounded values
assumed by the eld variable and its derivatives of integer order, despite the
fa t that the governing evolution equation may be a generi integro-di erential
equation and therefore, in parti ular, a fra tional di erential equation.

E. Fra tional Cal ulus a ording to Grunwald-Letnikov


A ording to the familiar de nition based on the "ba kward di eren e limit",
the rst-order derivative of a fun tion (x) is obtained as

d
(x) (x h)
(x) = (1) (x) = hlim
:
!0
dx
h
Applying this rule twi e gives the se ond-order derivative
(1) (x) (1) (x h)
d2
(2)

(
x
)
=

(
x
)
=
lim
h!0
dx2
h
(
)
1 (x) (x h) (x h) (x 2h)
= hlim
!0 h
h
h
(x) 2(x h) + (x 2h)
= hlim
:
!0
h2
We then obtain
d3
(x) 3(x h) + 3(x 2h) (x 3h)
(x) = (3) (x) = hlim
:
3
!
0
dx
h3
and, by indu tion,
18

Dn (x) =

1
dn
(x) = (n) (x) = hlim
n
!0 hn
dx

k=0

1)k

n
(x kh) : (E:1)
k

We note that the n-th derivative is seen to be a linear ombination of the


fun tion's values at the (n + 1) values x; x h; x 2h; : : : x nh of the independent variable. The oe ients are binomial oe ients and alternate in
sign. There is a divisor of hn :
Starting from the so- alled Riemann sum de nition of an integral and iterating
we get, see e.g. Oldham & Spanier (1974) [71, Podlubny (1999) [74

Ian+ (x)

= hlim
!0

hn

xX
a)=h

k=0

1)k

n
(x kh) ;
k

(E:2)

for k = 1; 2; : : :

(E:3)

with
(

1)k

n(n + 1) : : : (n + k
n
=
k
k!

1)

We have impli itly assumed that (x a)=h is an integer. We then note that the
n-th repeated integral is again a weighted sum of the fun tions's values, but
that the length of the sum tends to in nity as h ! 0 : The weights are again
de ned by binomial oe ients, but they are all positive. In this framework
we also re ognize the formal identities Ian+ = D n ; or Dn = Ia+n :
The de nition of the Grunwald-Letnikov fra tional derivative is an extension
of (E.1) to any real positive order , namely
1

GL Da+ (x) = hlim
!0 h

xX
a)=h

[(

k=0

( 1)k


(x kh) ; a < x < b : (E:4a)
k

Similarly, by taking the "forward di eren e limit", we de ne the dual


Grunwald-Letnikov fra tional derivative of order as

GL Db

1
(x) = hlim
!0 h

bX
x)=h

[(

k=0

1)k


(x + kh) ; a < x < b : (E:4b)
k

In eqs (E.4) the [  denotes the integer part. In analogy with the RiemannLiouville fra tional al ulus we may refer to (E.4a) and (E.4b) as the progressive and regressive Grunwald-Letnikov fra tional derivative, respe tively.
For our future purposes it is onvenient to introdu e a general notation to denote the ( nite or in nite) series of the ba kward/forward di eren e quotients
entering the limit in the above de nitions, namely
19

1 X

(x  kh) ;
( 1)k
h  :=
h k=0
k
where the upper limit of the sum is the appropriate one.

(E:5)

Under ertain onditions one an prove the following onne tions with the
Riemann-Liouville fra tional integrals and derivatives, (B.2-3) and (B.6), for
a < x < b;
Z

1
=
( )

GL Da+

GL Db

(x)

and

1
( )

(x) =


GL Da+ (x) =

(x  ) 1 ( ) d = Ia + (x) ; > 0 ;

(E:6a)

x) 1 ( ) d = Ib (x) ; > 0 ;

(E:6b)

Dm Iam+ (x) = Da + (x) ; m 1 < m  m ;

(E:7a)

(

(x) = ( 1)m Dm Ibm (x) = Db (x) ; m 1 <  m : (E:7b)


We point out the relation with the Caputo fra tional derivative (D.8), for t > 0
and m 1 < < m ;

GL Db

GL

0+

(t) =

mX1
k=0

(k

tk

Z
(m)
( )
1
+

+ 1)
(m ) 0 (t  ) +1

d :

(E:8)

The Grunwald-Letnikov s heme may be useful in numeri al omputations of


fra tional integrals and derivatives. However, as re ently reviewed by Goren o
(1997), see [28 and referen es therein, it has mostly a on eptual value sin e
it yields only a rst-order approximation in the step-length h : Higher-order
methods have been introdu ed for getting better approximations; so, in [87
se ond-order a ura y has been obtained by shifting the s heme by a noninteger multiple of h :
Let us on lude this se tion by noting that the formal identity between the
fra tional derivative and the fra tional integral (with the obvious hange of
sign in the order) stated in in (E.6a) has indu ed several authors to unify the
notions of fra tional derivatives and integrals with the joint name of di erintegration. Giving priority to the derivative they use the notation D for the
fra tional integral, attributing to it a negative order of di erentiation. The
present authors oppose to this notation sin e it is misleading and generally
not orre t, even if it is used in distinguished treatises as Davis (1936) [17,
Oldham & Spanier (1974) [71, Miller & Ross (1993) [64 and Podlubny (1999)
[74. It is well known that derivation and integration operators are not inverse
to ea h other, even if their order is integer, and therefore su h uni ation of
symbols appears not justi ed.
20

F. The Mittag-Leer type fun tions


The Mittag-Leer fun tion is so named from the great Swedish mathemati ian
who introdu ed it at the beginning of the XX -th entury in a sequen e of ve
notes [65{69. In this Se tion we shall onsider the Mittag-Leer fun tion
and some of the related fun tions whi h are relevant for their onne tion with
fra tional evolution pro esses. It is our purpose to provide a short referen ehistori al ba kground and a review of the main properties of these fun tions,
based on our papers, see [32, [33, [57, [58, [59.
Referen e-histori al ba kground

We note that the Mittag-Leer type fun tions, being ignored in the ommon
books on spe ial fun tions, are unknown to the majority of s ientists. Even in
the 1991 Mathemati s Subje t Classi ation these fun tions annot be found.
However, in the new AMS lassi ation foreseen for the year 2000, a pla e for
them has been reserved: 33E12 ("Mittag-Leer fun tions").
A des ription of the most important properties of these fun tions with relevant
referen es an be found in the third volume of the Bateman Proje t [21, in
the hapter XV III of the mis ellaneous fun tions. The treatises where great
attention is devoted to them are those by Dzherbashian [19, [20. We also
re ommend the lassi al treatise on omplex fun tions by Sansone & Gerretsen
[84 and the re ent book on fra tional al ulus by Podlubny [74.
Sin e the times of Mittag-Leer several s ientists have re ognized the importan e of the Mittag-Leer type fun tions, providing interesting mathemati al
and physi al appli ations, whi h unfortunately are not mu h known.
As pioneering works of mathemati al nature in the eld of fra tional integral
and di erential equations, we like to quote those by Hille & Tamarkin [42
and Barret [9. In 1930 Hille & Tamarkin have provided the solution of the
Abel integral equation of the se ond kind in terms of a Mittag-Leer fun tion,
whereas in 1954 Barret has expressed the general solution of the linear fra tional di erential equation with onstant oe ients in terms of Mittag-Leer
fun tions.
As former appli ations in physi s we like to quote the ontributions by Cole
in 1933 [16 in onne tion with nerve ondu tion, see also [17, and by Gross
[38 in 1947 in onne tion with me hani al relaxation. Subsequently, in 1971,
Caputo & Mainardi [14 have shown that Mittag-Leer fun tions are present
whenever derivatives of fra tional order are introdu ed in the onstitutive
equations of a linear vis oelasti body. Sin e then, several other authors have
pointed out the relevan e of these fun tions for fra tional vis oelasti models,
see e.g. Mainardi [57.
21

The Mittag-Leer fun tions E (z ) ; E ; (z )

The Mittag-Leer fun tion E (z ) with > 0 is de ned by the following series
representation, valid in the whole omplex plane,

E (z ) :=

n=0

zn
; > 0; z 2 C:
( n + 1)

(F:1)

It turns out that E (z ) is an entire fun tion of order  = 1= and type 1 :


This property is still valid but with  = 1=Ref g ; if 2 C with positive real
part, as formerly noted by Mittag-Leer himself in [68.
The Mittag-Leer fun tion provides a simple generalization of the exponential
fun tion be ause of the substitution of n! = (n + 1) with ( n)! = ( n + 1) :
Parti ular ases of (F.1), from whi h elementary fun tions are re overed, are


E2 +z 2 = osh z ;

E2

z 2 = os z ;

z 2 C;

(F:2)

and

E1=2 (z 1=2 ) = ez 1 + erf (z 1=2 ) = ez erf (z 1=2 ) ; z 2 C ;


h

(F:3)

where erf (erf ) denotes the ( omplementary) error fun tion de ned as
2
erf (z ) := p


2
e u du ; erf (z ) := 1 erf (z ) ; z 2 C :

(F:4)

In (F.4) by z 1=2 we mean the prin ipal value of the square root of z in the
omplex plane ut along the negative real semi-axis. With this hoi e z 1=2
turns out to be positive/negative for z 2 R+ . A straightforward generalization
of the Mittag-Leer fun tion, originally due to Agarwal in 1953 based on a
note by Humbert, see [3, [44, [45, is obtained by repla ing the additive
onstant 1 in the argument of the Gamma fun tion in (F.1) by an arbitrary
omplex parameter : For the generalized Mittag-Leer fun tion we agree to
use the notation

E ; (z ) :=

n=0

zn
; > 0; 2 C; z 2 C:
( n + )

(F:5)

Parti ular simple ases are


ez

sinh (z 1=2 )
:
(F:6)
z
z 1=2
We note that E ; (z ) is still an entire fun tion of order  = 1= and type 1 :
For la k of spa e we prefer to ontinue with the Mittag-Leer fun tions in one
parameter: sin e here we shall limit ourselves to onsider evolution pro esses

E1;2 (z ) =

; E2;2 (z ) =

22

hara terized by a single fra tional order, the more general fun tions with
two parameters turn out to be redundant. However, we nd it onvenient to
introdu e other fun tions depending on a single parameter whi h turn out to
be related by simple relations to the original Mittag-Leer fun tions, and to
onsider them as belonging to the lass of Mittag-Leer type fun tions.
The Mittag-Leer integral representation and asymptoti expansions

Many of the most important properties of E (z ) follow from Mittag-Leer's


integral representation

E (z ) =

1 Z  1 e 
d ; > 0 ; z 2 C ;
2i Ha  z

(F:7)

where the path of integration Ha (the Hankel path) is a loop whi h starts and
ends at 1 and en ir les the ir ular disk j j  jz j1= in the positive sense:
  arg   on Ha : To prove (F.7), expand the integrand in powers of
 , integrate term-by-term, and use Hankel's integral for the re ipro al of the
Gamma fun tion.
The integrand in (F.7) has a bran h-point at  = 0. The omplex  -plane is ut
along the negative real semi-axis, and in the ut plane the integrand is singlevalued: the prin ipal bran h of  is taken in the ut plane. The integrand has
poles at the points m = z 1= e2 i m= ; m integer, but only those of the poles
lie in the ut plane for whi h  < argz + 2 m <  : Thus, the number
of the poles inside Ha is either [ or [ + 1, a ording to the value of arg z .
The most interesting properties of the Mittag-Leer fun tion are asso iated
with its asymptoti developments as z ! 1 in various se tors of the omplex
plane. These properties an be summarized as follows. For the ase 0 < < 2
we have

E (z ) 

1
exp(z 1= )

k=1

z k
;
k=1 (1 k )
For the ase  2 we have
E (z ) 

E (z ) 

z k
;
(1 k)

jzj ! 1 ; jarg zj < =2 ; (F:8)

jzj ! 1 ; =2 < arg z < 2 =2 : (F:9)

exp z 1= e2 i m=

k=1

z k
;
(1 k)

jzj ! 1 ;

(F:10)

where m takes all integer values su h that =2 < arg z +2 m < =2 ; and
arg z an assume any value between  and + in lusive. From the asymptoti
properties (F.8-10) and the de nition of the order of an entire fun tion, we infer
that the Mittag-Leer fun tion is an entire fun tion of order 1= for > 0;
23

in a ertain sense ea h E (z ) is the simplest entire fun tion of its order. The
Mittag-Leer fun tion also furnishes examples and ounter-examples for the
growth and other properties of entire fun tions of nite order.
The Lapla e transform pairs related to the Mittag-Leer fun tions

The Mittag-Leer fun tions are onne ted to the Lapla e integral through
the equation
Z1
1
; > 0:
(F:11)
e u E (u z ) du =
1 z
0
The integral at the L.H.S. was evaluated by Mittag-Leer who showed that
the region of its onvergen e ontains the unit ir le and is bounded by the
line Re z 1= = 1. Putting in (F.11) u = st and u z =  t with t  0 and
 2 C ; and using the sign  for the juxtaposition of a fun tion depending
on t with its Lapla e transform depending on s, we get the following Lapla e
transform pairs

s 1
; Re s > jj1= :
(F:12)
s + 
In our CISM Le ture notes [33 we have shown the key role of the Mittag-Leer
type fun tions E (  t ) in treating Abel integral equations of the se ond
kind and fra tional di erential equations, so improving the former results by
Hille & Tamarkin (1930) [42 and Barret (1954) [9, respe tively. In parti ular,
assuming  > 0 ; we have dis ussed the pe uliar hara ters of these fun tions
(power-law de ay) for 0 < < 1 and for 1 < < 2 related to fra tional
relaxation and fra tional os illation pro esses, respe tively, see also [56, [32.
E (  t ) 

Other formulas: summation and integration

For ompleteness hereafter we exhibit some formulas related to summation and


integration of ordinary Mittag-Leer fun tions (in one parameter ), referring
the interested reader to [19, [74 for their proof and for their generalizations
to two parameters. Con erning summation we outline

E (z ) =



1 pX1
E =p z 1=p ei2h=p ; p 2 N ;
p h=0

(F:13)

from whi h we derive the dupli ation formula


i
1h
(F:14)
E (z ) = E =2 (+z 1=2 ) + E =2 ( z 1=2 ) :
2
Con erning integration we outline another interesting dupli ation formula
1

x2 =(4t) E (x ) dx =

24

 t E =2 (t =2 ) ; t > 0 :

(F:15)

G. The M-Wright type fun tions

Let us rst re all the more general Wright fun tion W; (z ) ; z 2 C ; with
 > 1 and  > 0 : This fun tion, so named from the British mathemati ian
who introdu ed it between 1933 and 1940 [92{95 is de ned by the following
series and integral representation, valid in the whole omplex plane,

W; (z ) :=

zn
1 Z  + z
:=
e
2i Ha
n=0 n! (n + )
X

d
:


(G:1)

Here Ha denotes an arbitrary Hankel path, namely a ontour onsisting of


pie es of the two rays arg  =  extending to in nity, and of the ir ular
ar  =  ei ; jj   ; with  2 (=2;  ) ; and  > 0 ; arbitrary.
It is possible to prove that the Wright fun tion is entire of order 1=(1 + ) ;
hen e of exponential type if   0 : The ase  = 0 is trivial sin e W0; (z ) =
e z = () : The ase  =  ;  = 1  with 0 <  < 1 provides the fun tion M
fun tion of the Wright type, M (z ) ; that is of spe ial interest for us, see e.g.
Mainardi (1995, 1996a, 1996b, 1997) [54{57 and Goren o, Lu hko & Mainardi
(1999, 2000) [30,31. Spe i ally, we have

M (z ) := W

;1  (

z) =

1
W
z

;0 (

z) ; 0 <  < 1 ;

(G:2)

and therefore from (G.1-2)

M (z ) :=

n=0

1 Z 
( z )n
:=
e
n! [ n + (1  ) 2i Ha

z  d :
1 

(G:3)

It turns out that M (z ) is an entire fun tion of order  = 1=(1  ) ; whi h


provides a generalization of the Gaussian and of the Airy fun tion. In fa t we
obtain


1
(G:4)
M1=2 (z ) = p exp z 2 =4 ;



(G:5)
M1=3 (z ) = 32=3 Ai z=31=3 ;
where Ai denotes the Airy fun tion. Furthermore it an be proved that M1=q (z )
satis es the di erential equation of order q 1

dq 1
( 1)q
M
(
z
)
+
z M1=q (z ) = 0 ;
1
=q
dz q 1
q

(G:6)

subje ted to the q 1 initial onditions at z = 0, derived from the series


expansion in (G.1),
h)
M1(=q
(0) =

( 1)h
[(h + 1)=q sin[ (h + 1)=q ; h = 0; 1; : : : q

25

2 : (G:7)

We note that, for q  4 ; Eq. (G.6) is akin to the hyper-Airy di erential


equation of order q 1 ; see e.g. Bender & Orszag (1987)[10. Consequently, the
fun tion M (z ) is a generalization of the hyper-Airyfun tion. In the limiting
ase  = 1 we get M1 (z ) = (z 1) :
From now on let us onsider only the M fun tion for positive (real) argument,
M (r) ; whi h will be the relevant fun tion for our purposes. The asymptoti
representation of M (r) ; as r ! 1 an be obtained by using the ordinary
saddle-point method. Choosing as a variable r= rather than r the omputation is easier and yields, see Mainardi & Tomirotti (1995) [61,

M (r= )  a( ) r
where

1=2


exp

b( ) r 1

; r ! +1 ;

(G:8)

1 
1
> 0 ; b( ) =
> 0:
(G:9)

2 (1  )
Be ause of the above exponential de ay, any moment of order > 1 for
M (r) is nite and results

a( ) = q

r M (r) dr =

( + 1)
; > 1:
( + 1)

(G:9)

The following Lapla e transform pairs an be proved, see Mainardi (1997) [57,

M (r)
and

t +1

 E ( s) ;

M (1=t )

0 <  < 1:

 exp ( s ) ;

(G:10)

0 <  < 1;

(G:11)

1
M (1=t )  s1  exp ( s ) ; 0 <  < 1 ;
(G:12)
t 
As a parti ular ase of (G.10) we re over the well-known Lapla e transform
pair, see e.g. Doets h (1974),

M1=2 (r) :=

p1 exp

r2 =4

 E = ( s) := exp(s ) erf (s) :


1 2

(G:13)

We also note that, transforming term-by-term the Taylor series of M (r) (not
being of exponential order) yields a series of negative powers of s ; whi h
represents the asymptoti expansion of E ( s) as s ! 1 in a se tor around
the positive real axis. As parti ular ases of (G.11-12) we re over the wellknown Lapla e transform pairs, see e.g. Doets h (1974) [18 ,




1
1
1=2
3=2
2
1=2
p
; (G:14)
M
(1
=t
))
:=
t
exp
1
=
(4
t
)

exp
s
1
=
2
2t3=2
2 




1 1=2
1
1=2
2
1=2
1=2
p
M
(1
=t
))
:=
: (G:15)
t
exp
1
=
(4
t
)

s
exp
s
t1=2 1=2


26

Referen es
[1 N.H. Abel, Solution de quelques problemes a l'aide d'integrales de nie
[Norwegian, Magazin for Naturvidenskaberne , Aargang 1, Bind 2, Christiana
1823. Fren h translation in Oeuvres Completes , Vol I, pp. 11-18. Nouvelle edition
par L. Sylow et S. Lie, 1881.
[2 N.H. Abel, Au oesung einer me hanis hen Aufgabe, Journal fur die reine und
angewandte Mathematik (Crelle), Vol. I (1826), pp. 153-157.
[3 R.P. Agarwal, A propos d'une note de M. Pierre Humbert, C.R. A ad. S i. Paris
236 (1953) 2031{2032.
[4 Yu.I. Babenko, Heat and Mass Transfer (Chimia, Leningrad, 1986) [in Russian
[5 R.T. Baillie and M.L. King (Editors), Fra tional Di eren ing and Long Memory
Pro esses, Journal of E onometri s 73 No 1 (1996) 1-324.
[6 A.V. Balakrishnan, Representation of abstra t Riesz potentials of the ellipti
type, Bull. Amer. Math. So . 64 No 5 (1958) 266{289.
[7 A.V. Balakrishnan, Operational al ulus for in nitesimal generators of semigroups, Trans. Amer. Math. So . 91 No 2 (1959) 330-353.
[8 A.V. Balakrishnan, Fra tional powers of losed operators and the semi-groups
generated by them, Pa i J. Math. 10 No 2 (1960) 419-437.
[9 J.H. Barret, Di erential equations of non-integer order, Canad. J. Math. 6 (1954)
529{541.
[10 C.M. Bender and S.A. Orszag, Advan ed Mathemati al Methods for S ientists
and Engineers (M Graw-Hill, Singapore, 1987), Ch 3.
[11 S. Bo hner, Di usion equation and sto hasti pro esses, Pro . Nat. A ad.
S ien es, USA 35 (1949) 368{370.
[12 M. Caputo, Linear models of dissipation whose Q is almost frequen y
independent, Part II Geophys. J. R. Astr. So . 13 (1967) 529{539.
[13 M. Caputo, Elasti ita e Dissipazione (Zani helli, Bologna, 1969). [in Italian
[14 M. Caputo and F. Mainardi, Linear models of dissipation in anelasti solids,
Riv. Nuovo Cimento (Ser. II) 1 (1971) 161{198.
[15 A. Carpinteri and F. Mainardi, eds, Fra tals and Fra tional Cal ulus in
Continuum Me hani s (Springer Verlag, Wien, 1997).
[16 K.S. Cole, Ele tri al ondu tan e of biologi al systems in Pro . Symp. Quant.
Biol. (Cold Spring Harbor, New York, 1933) 107-116.
[17 H.T. Davis, The Theory of Linear Operators (The Prin ipia Press, Bloomington,
Indiana, 1936).

27

[18 G. Doets h, Introdu tion to the Theory and Appli ations of the Lapla e
Transformation (Springer Verlag, Berlin, 1974).
[19 M.M. Dzherbashian, Integral Transforms and Representations of Fun tions in
the Complex Plane (Nauka, Mos ow, 1966). [in Russian
[20 M.M. Dzherbashian, Harmoni Analysis and Boundary Value Problems in the
Complex Domain (Birkhauser Verlag, Basel, 1993.
[21 A. Erdelyi (Editor) Higher Trans endental Fun tions, Bateman Proje t, Vols.
1-3 (M Graw-Hill, New York, 1953-1955).
[22 A. Erdelyi (Editor) Tables of Integral Transforms, Bateman Proje t, Vols. 1-2
(M Graw-Hill, New York, 1953-195).
[23 W. Feller, On a generalization of Mar el Riesz' potentials and the semi-groups
generated by them, Meddelanden Lunds Universitets Matematiska Seminarium
(Comm. Sem. Mathem. Universite de Lund), Tome suppl. dedie a M. Riesz, Lund
(1952), 73-81.
[24 W. Feller : An Introdu tion to Probability Theory and its Appli ations, Vol. 2,
2nd edn (Wiley, New York, 1971). [1st edn. 1966
[25 C. Fox, The G and H fun tions as symmetri al Fourier kernels, Trans. Amer.
Math. So . 98 (1961) 395-429.
[26 I.M. Gel'fand and G.E. Shilov, Generalized Fun tions, Vol. 1 (A ademi Press,
New York, 1964). [English translation from the Russian (Nauka, Mos ow, 1959)
[27 R. Goren o, Abel Integral Equations with spe ial Emphasis on Appli ations ,
Le tures in Mathemati al S ien es Vol. 13 (The University of Tokyo, Graduate
S hool of Mathemati al S ien es, 1996). [ISSN 0919-8180
[28 R. Goren o, Fra tional al ulus: some numeri al methods, in: A. Carpinteri
and F. Mainardi, eds, Fra tals and Fra tional Cal ulus in Continuum Me hani s
(Springer Verlag, Wien, 1997) 277-290.
[29 R. Goren o, G. De Fabritiis and F. Mainardi, Dis rete random walk models for
symmetri Levy-Feller di usion pro esses, Physi a A 269 (1999) 79{89.
[ < http://xxx.lan.gov/abs/ ond-mat/9903264
[ < http://www.e onophysi s.org >: sele t "papers" and then "mainardi"
[30 R. Goren o, Yu. Lu hko and F. Mainardi, Analyti al properties and
appli ations of the Wright fun tion, Fra tional Cal ulus and Applied Analysis
2 No 4 (1999) 383-414.
[31 R. Goren o, Yu. Lu hko and F. Mainardi, Wright fun tions as s ale-invariant
solutions of the di usion-wave equation, J. Computational and Appl. Mathemati s
(2000) to appear.

28

[32 R. Goren o and F. Mainardi, Fra tional os illations and Mittag-Leer


fun tions, Pre-print A-14/96, Fa hberei h Mathematik und Informatik, Freie
Universitat, Berlin, 1996.
[ < http://www.math.fu-berlin.de/publ/index.html >
[33 R. Goren o and F. Mainardi, Fra tional al ulus: integral and di erential
equations of fra tional order, in: A. Carpinteri and F. Mainardi, eds, Fra tals
and Fra tional Cal ulus in Continuum Me hani s (Springer Verlag, Wien, 1997)
223{276.
[34 R. Goren o and F. Mainardi, Random walk models for spa e-fra tional di usion
pro esses, Fra tional Cal ulus and Applied Analysis 1 No 2 (1998) 167{191.
[ < http://www.e onophysi s.org >: sele t "papers" and then "mainardi"
[35 R. Goren o and F. Mainardi, Approximation of Levy-Feller di usion by random
walk, J. Analysis and its Appli ations (ZAA) 18 No 2 (1999) 231{246.
[ < http://www.e onophysi s.org >: sele t "papers" and then "mainardi"
[36 R. Goren o, F. Mainardi and H.M. Srivastava, Spe ial fun tions in fra tional
relaxation-os illation and fra tional di usion-wave phenomena, in: D. Bainov,
ed, Pro eedings VIII International Colloquium on Di erential Equations, Plovdiv
1997 (VSP, Utre ht, 1998) 195{202.
[37 R. Goren o and S. Vessella, Abel Integral Equations: Analysis and Appli ations ,
Le ture Notes in Mathemati s, Vol. 1461 (Springer-Verlag, Berlin, 1991).
[38 B. Gross, On reep and relaxation, J. Appl. Phys. 18 (1947) 212{221.

[39 A.K. Grunwald, Uber
"begrenzte" Derivation und deren Anwendung, Z. angew.
Math. und Phys. 12 (1867) 441-480.
[40 R. Hilfer, Fra tional Derivatives in Stati and Dynami S aling, in B. Dubrulle,
F. Graner and D. Sornette (Editors), S ale Invarian e and Beyond (Springer
Verlag, Berlin and EDP S ien e, Fran e, 1977), pp. 53-62. Le ture 3, Les Hou hes
Workshop, Mar h 10-14, 1997.
[41 R. Hilfer (Editor), Appli ations of Fra tional Cal ulus in Physi s (World
S ienti , Singapore, 2000) to appear.
[42 E. Hille and J.D. Tamarkin, On the theory of linear integral equations Ann.
Math. 31 (1930) 479{528.
[43 Hj. Holmgren, Om di erentialkalkylen med indi es af hvad natur som???, Kongl.
Svenska Vetenskaps-Akad. Hanl. Sto kholm 5 No.11 (1965-1866) 1{83.
[44 P. Humbert, Quelques resultats relatifs a la fon tion de Mittag-Leer, C.R.
A ad. S i. Paris 236 (1953) 1467{1468.
[45 P. Humbert and R.P. Agarwal, Sur la fon tion de Mittag-Leer et quelquesunes de ses generalisations, Bull. S i. Math (Ser. II) 77 (1953) 180{185.

29

[46 V. Kiryakova, Generalized Fra tional Cal ulus and Appli ations, Pitman
Resear h Notes in Mathemati s # 301, Longman, Harlow 1994.
[47 A.V. Letnikov, Theory of di erentiation with an arbitrary index, Mat. Sb. 3
(1868) 1{66. In Russian.
[48 A.V. Letnikov, On histori al development of di erentiation theory with an
arbitrary index, Mat. Sb. 3 (1868) 85{112. In Russian
[49 A.V. Letnikov, Investigations on the theory of integrals of the form ax (x
p 1 f (u) du ; Mat. Sb. 7 No 1 (1874) 5{205. In Russian.
u)
R

[50 J. Liouville, Memoire sur quelques questions de geometrie et de me anique,



et sur un nouveau genre de al ul pour resoudre es questions, J. l'E ole
Roy.
Polyte hn. 13 Se t. 21 (1832) 1{69.
[51 J. Liouville, Memoire sur le al ul des di erentielles a indi es quel onques, J.

l'E ole
Roy. Polyte hn. 13 Se t. 21 (1832) 71{162.
[52 J. Liouville, Memoire sur l'integration des equations di erentielles a indi es

fra tionnaires, J. l'E ole
Roy. Polyte hn. 15 No 55 (1837) 55{84.
[53 Ch. Lubi h, Dis retized fra tional al ulus, SIAM J. Math. Anal. 17 (1986)
704-719.
[54 F. Mainardi, The time fra tional di usion-wave equation, Radio sika 38 No
1-2 (1995) 20{36. [English Transl.:Radiophysi s & Quantum Ele troni s
[55 F. Mainardi, The fundamental solutions for the fra tional di usion-wave
equation, Applied Mathemati s Letters 9 No 6 (1996) 23{28.
[56 F. Mainardi, Fra tional relaxation-os illation and fra tional di usion-wave
phenomena, Chaos, Solitons & Fra tals 7 (1996) 1461{1477.
[57 F. Mainardi, Fra tional al ulus: some basi problems in ontinuum and
statisti al me hani s, in: A. Carpinteri and F. Mainardi, eds, Fra tals and
Fra tional Cal ulus in Continuum Me hani s (Springer Verlag, Wien, 1997) 291{
248.
[58 F. Mainardi and R. Goren o, The Mittag-Leer fun tion in the RiemannLiouville fra tional al ulus, in: A.A. Kilbas, Ed., Boundary Value Problems,
Spe ial Fun tions and Fra tional Cal ulus (Belarusian State University, Minsk,
1996) 215{225.
[59 F. Mainardi and R. Goren o, On Mittag-Leer type fun tions in fra tional
evolution pro esses, J. Computational and Appl. Mathemati s (2000) to appear.
[60 F. Mainardi, P. Paradisi and R. Goren o, Probability distributions generated by
fra tional di usion equations, in J. Kertesz and I. Kondor (Editors), E onophysi s:
an Emerging S ien e (Kluwer, Dordre ht, 1999) pp. 39, to appear.
[ < http://www.e onophysi s.org >: sele t "papers" and then "mainardi"

30

[61 F. Mainardi and M. Tomirotti, On a spe ial fun tion arising in the time
fra tional di usion-wave equation, in: P. Rusev, I. Dimovski and V. Kiryakova
(Eds.), Transform Methods and Spe ial Fun tions, So a 1994 (S ien e Culture
Te hnology, Singapore, 1995), pp. 171-183.
[62 A. Mar haud, Sur les derivees et sur les di eren es des fon tions de vriables
reeelles, I. Math. Pures et Appl. 6 No 4 (1927) 337-425.
[63 A.C. M Bride, Fra tional Cal ulus and Integral Transforms of Generalized
Fun tions, Pitman Resear h Notes in Mathemati s # 31, Pitman, London 1979.
[64 K.S. Miller and B. Ross: An Introdu tion to the Fra tional Cal ulus and
Fra tional Di erential Equations, Wiley, New York 1993.
[65 G.M. Mittag-Leer, Sur l'integrale de Lapla e-Abel, C.R. A ad. S i. Paris (ser.
II) 136 (1902) 937{939.
[66 G.M. Mittag-Leer, Une generalisation de l'integrale de Lapla e-Abel, C.R.
A ad. S i. Paris (ser. II) 137 (1903) 537{539.
[67 G.M. Mittag-Leer, Sur la nouvelle fon tion E (x), C.R. A ad. S i. Paris (ser.
II) 137 (1903) 554{558.
[68 G.M. Mittag-Leer, Sopra la funzione E (x), R. A ad. Lin ei, Rend. (ser. V)
(1904) 13 3{5.
[69 G.M. Mittag-Leer, Sur la representation analytique d'une bran he uniforme
d'une fon tion monogene, A ta Math. 29 (1905) 101{181.
[70 K. Nishimoto, An Essen e of Nishimoto's Fra tional Cal ulus, Des artes Press,
Koriyama 1991.
[71 K.B. Oldham and J. Spanier, The Fra tional Cal ulus, A ademi Press, New
York 1974.
[72 S. Pin herle, Sulle funzioni ipergeometri he generalizzate, Atti R. A ademia
Lin ei, Rend. Cl. S i. Fis. Mat. Nat., (Ser. 4) 4 (1888) 694-700, 792-799. Reprinted
in Salvatore Pin herle: Opere S elte, edited by UMI (Unione Matemati a Italiana)
Vol. 1, pp. 223-230, 231-239 (Cremonese, Roma, 1954).
[73 S. Pin herle, Sulle derivate ad indi e qualunque, Rend. R. A . S ienze Bologna
(Ser. V) 9 (1902) 745-758.
[74 I. Podlubny, Fra tional Di erential Equations, A ademi Press, San Diego 1999.
[75 H. Pollard, The ompletely monotoni hara ter of the Mittag-Leer fun tion
E ( x) ; Bull. Amer. Math. So . 54 (1948) 1115{1116.
[76 J. Pruss, Evolutionary Integral Equations and Appli ations (Birkhauser, Basel,
1993).
[77 B. Riemann, Versu h einer allgemeinen Au assung der Integration und
Di erentiation, in Bernhard Riemann's gesammelte mathematis he Werke und

31

wissens haftlikerNa hlass , pp 331{344 (Teubner, Leipzig, 1876). New edition

edited by H. Weber (Dover, New York, 1953).


[This paper, written by Riemann in January 1847 when he was still a student,
was published only in 1876, ten years after his death
[78 M. Riesz, L'integrales de Riemann-Liouville et le probleme de Cau hy, A ta
Math. 81 No 1-2 (1949) 1-223.
[79 B. Ross, Fra tional Cal ulus and its Appli ations, Le ture Notes in Mathemati s
# 457, Springer Verlag, Berlin 1975. [Pro . Int. Conf. held at Univ. of New Haven,
USA, 1974

[80 B. Rubin, Fra tional Integrals and Potentials, Pitman Monographs and Surveys
in Pure and Applied Mathemati s #82, Addison Wesley Longman, Harlow 1996.
[81 A. Sai hev and G. Zaslavsky, Fra tional kineti equations: solutions and
appli ations, Chaos 7 (1997) 753-764.
[82 W.R. S hneider, Stable distributions: Fox fun tion representation and
generalization, in S. Albeverio, G. Casati and D. Merlini (Eds), Sto hasti
Pro esses in Classi al and Quantum Systems, Le ture Notes in Physi s # 262
(Springer Verlag, Berlin, 1986) 497-511.
[83 S.G. Samko, A.A. Kilbas and O.I. Mari hev, Fra tional Integrals and
Derivatives, Theory and Appli ations (Gordon and Brea h, Amsterdam, 1993).
[English translation from the Russian, Integrals and Derivatives of Fra tional
Order and Some of Their Appli ations (Nauka i Tekhnika, Minsk, 1987)
[84 G. Sansone and J. Gerretsen, Le tures on the Theory of Fun tions of a Complex
Variable, Vol. I. Holomorphi Fun tions (Nordho , Groningen, 1960) 345{349.
[85 B. Stankovi , On the fun tion of E.M. Wright, Publ. de l'Institut Mathematique,
Beograd, Nouvelle Ser. 10 (1970) 113-124.
[86 H.M. Srivastava, K.C. Gupta, and S.P. Goyal, The H-Fun tions of One and
Two Variables with Appli ations (South Asian Publishers, New Delhi and Madras,
1982).
[87 Vu Kim Tuan and R. Goren o, Extrapolation to the limit for numeri al
fra tional di erentiation, ZAMM 75 (1995) 646-648.
[88 U. Westphal, An approa h to fra tional powers of operators via fra tional
di eren es, Pro . London Math. So . 29 No 3 (1974) 557-576.
[89 H. Weyl, Bemerkungen zum Begri des Di erentialquotientten gebro hener
Ordnung, Vierteljahress hrift der Naturfors henden Geselis haft in Zuri h 62 No
1-2 (1917) 296-302.

[90 A. Wiman, Uber
die Nullstellen der Funktionen E (x) ; A ta Math. 29 (1905),
217{234.
[91 G. Witte, Die analytis he und die numeris he Behandlung einer Klasse von
Volterras hen Integralglei hungen im Hilbertraum, PhD Thesis, Free University
of Berlin (Logos Verlag, Berlin, 1997). ISBN 3-89722-005-9.

32

[92 E.M. Wright, On the oe ients of power series having exponential singularities,
Journal London Math. So . 8 (1933)71-79.
[93 E.M. Wright, The asymptoti expansion of the generalized Bessel fun tion,
Pro . London Math. So . (Ser. II) 38 (1935) 257-270.
[94 E.M. Wright, The asymptoti expansion of the generalized hypergeometri
fun tion, Journal London Math. So . 10 (1935) 287-293.
[95 E.M. Wright, The generalized Bessel fun tion of order greater than one, Quart.
J. Math., Oxford ser. 11 (1940) 36-48.

33