Escolar Documentos
Profissional Documentos
Cultura Documentos
1998
PHYSICS LETTERS A
ELSl3VIER
13-16
Chaves a,b,
March 1997; revised manuscript received 15 October 1997; accepted for publication 21 November
Communicated
by A.R. Bishop
1997
Abstract
A fractional-derivatives diffusion equation is proposed that generates the Levy statistics. The fractional derivatives are
defined by the eigenvector equation a: eax = an eoX and for one dimension the diffusion equation in an isotropic medium
reads &n = (D/2) (a, + a,) n + u&n, 1 < ru Q 2. The equation is based on a proposed generalization of Ficks law which
reads j = -(D/2) (Oy-* - '7";'
)n + vn. The diffusion equation is also written for an anisotropic medium, and in this
case it generates an asymmetric Levy statistics. @ 1998 Published by Elsevier Science B.V.
PACS:
02.50.-r;
02.50.Ey;
02.60.Nm;
05.4O.+j
Non-Gaussian
probability distributions of stochastic variables have been a subject of interest for a long
time [ l-31. More recently, these distributions have
been applied to the description of many physical processes, including turbulent flow [ 41, diffusion in complex systems [ 51, chaotic dynamics of classical conservative systems [ 61, and others [ 71.
In this Letter I propose a generalization
of Ficks
law in which the particle current is proportional to
the fractional derivatives of the particle density. With
the generalized Ficks law I obtain a fractional derivatives equation which generates the Levy distribution.
The proposed diffusion equation might be useful for
the investigation of the mechanism of superdiffusion.
With the present approach I also show that superdiffusion in a medium which is not symmetric with respect to space inversion is also asymmetric. This is
very distinct from the behavior of normal diffusion,
which always presents space inversion symmetry. An
0375-9601/98/$19.00
@ 1998 Published
PII SO375-9601(97)00947-X
important consequence of the asymmetry in superdiffusion is that linear transport in such conditions is not
symmetric with respect to the reversion of the bias.
Besides the scientific interest, this could also result in
novel devices.
Ordinary diffusion is an important process described by a Gaussian distribution. In one dimension,
the probability
density P(x, t) of a particle, initially (t = 0) at x = 0, being at x the instant t is
P(x, t) = (4TDt)-1/2exp(
-x2/4Dt),
where D is
the Einstein diffusivity constant. A main feature of
the process is the linear relation between the mean
square displacement and time, namely (x2) = 2Dt. In
anomalous diffusion one might find (x2) cx tY, y # 1,
or else (x) might be a divergent integral for t f 0.
My interest here is in this latter process, so-called
Levy flights, and the associated distribution,
which
is the Fourier transform of p(k, t) = exp( -atlkj),
with (Y < 2. Ordinary diffusion is generated as the
t> +
DajP(n,t) ,
(1)
for the initial condition P ( X, 0) = S(n) , where a nonzero mean flow speed u was allowed for the particles,
I propose a generalization of Eq. ( 1) using fractional
derivatives ~?,a,1 < a 6 2, in substitution for $, as a
generator of the Levy flights.
In contrast to the Levy flights, the ordinary diffusion equation is well understood and can be deduced
from the BoItzmann-Gibbs statistics. Another way of
looking at the problem is to examine the physical basis
of Eq. ( 1) . If one combines the continuity equation
V.j+$=O
(2)
(5)
where LY,b and a are complex numbers This form of
derivation can be applied to any function which can
be expanded as a sum of exponentials and therefore
to any distribution. The multiplication rule
(6)
follows immediately from the definition. It is opportune here to notice that by using the definition of fractional derivative given by Eq. (4) one obtains
U-1
e-
&
1 -
>
I-(--a)
J-= -DVn+vn,
x>o
(31
1 <cr<2.
(8)
c?n D
at=2
-(v,a-t-vnt)n-u.vn,
(10)
Eq. (10) can easily be solved by using the Fourier
transform
n(x,t)
OOdk
--$(
.I
k, t>eikx.
The solution is
where (D-, = xTa-i ,fr( -cY), r is the gamma function and x+ is a generalized function defined on the
positive semi-axis x >, 0.
n(x, t) =
mZ;;n(x,O)P(x
dn
- x, t) ,
(12)
A.S. Chaves/Physics
13-16
dk
; cos k( x - or)
Dfk]
(13)
a@
at=2
-[(ik)+(-ik)]@-iku@.
( 14)
Plot
of
expl k cos(av/2)
cos[kx + ksin(mr/2)]
Seshadri and West [ 91 first studied the L&y distribution as a solution to a similar equation for the characteristic function. They investigated the distribution of
particle speeds p( u, t) and defined the Fourier transform
P(L), f) =
J
*
dk
%&(
The differential
k, t) e?.
equation
for 4( k, t) is
(17)
where the left (Dr) and right (0,)
distinct. The associated kernel is
OOdk
%p
J
--ca
function
is
( 19)
corresponds
D,=Oand
(15)
(16)
P(X.f)
I.
Fig.
(k, t) ek(x-) ,
diffusivities
are
(18)
P(-k,t)
=P*(k,r),
~(0, 1) = 1,
p(k,r+
t) =~(k,t)p(k,t)
means that
(a) P(x,t)
E Iw,
(b) J:aP(X>t)dx=
1,
16
j=-D.Vn+un,
AS.
Chmes/Physics
(20)
References