Professional Documents
Culture Documents
summation convention. The jt are Gaussian random tion is performed. If we introduce new variables in Eq. 1
variables with zero mean value and correlation functions such that
it = 0, it jt = 2ijt t. 2 = ,t, 13
The Fokker-Planck or forward Kolmogorov equation for the then the Langevin equation in the new variables is
transition probability density Px , t x , t associated with
the Langevin equation 1, together with the property 2, = hi,t + gij,t jt, 14
reads as t t
where
P
t
=
xi
Dix,t +
2
xi x j
Dijx,t P, 3
hi =
i i
+
t k
hk ; gij =
i
k
gkj . 15
where Dix , t is the drift coefficient and Dijx , t the
diffusion coefficient defined, respectively, as Hence, the transformed drift and diffusion coefficients
writing xi instead of in the argument are given by
Dix,t = hix,t + gkjx,t gijx,t, 4 xi xi 2xi
xk Di = + Dk + Drk , 16
t x xk xr xk
Dijx,t = gikx,tg jkx,t, 5
xi xj
with initial condition Dij = Drk , 17
xr xk
Px,tx,t = x x. 6
where Di and Dij are given, respectively, by Eqs. 4 and 5.
Here kt = xk for k = 1 , 2 , . . . , N, represents the sharp value The Fokker-Planck equation 7 is also transformed in such a
of the variable t at time t. If we multiply Eq. 3 by way that, for the new variables, it can be written as
Wx , t and integrate over x we obtain the Fokker-Planck
equation for the probability density Wx , t, i.e., W 2
= Di + Dij W , 18
t x xi xi xj
2
W= Dix,t + Dijx,t W. 7
t xi xi x j where the transformed drift Di and diffusion Dij coefficients
agree with Eqs. 16 and 17. The probability density W is
The complete information of a Markov process is con- transformed as W = JW = W / J, where J is the Jacobian of
tained in the joint probability distribution W2x , t ; x , t the transformation defined by
which can be expressed as
J dNx/dNx = Det xi/ xj = 1/J = 1/Det xi/ x j,
W2x,t;x,t = Px,tx,tWx,t. 8
19
If the drift and diffusion coefficients do not depend on time,
a stationary solution may exist. In this case, P can depend and dNx and dNx are the volume elements.
only on the time difference t t, and we may write for
t t the joint probability distribution in the stationary state B. The Fokker-Planck equation for free Brownian motion in
W2x,t;x,t = Px,t tx,0Wstx. 9 the velocity space
A particular case of Eq. 1 is the Ornstein-Uhlenbeck The three-dimensional Brownian motion for the velocity
10,11 process, which is described by ut in the absence of external force is the simplest form of
Eq. 10 and it is given by
i = ij j + jt, 10
u = u + At, 20
with -correlated Gaussian distributed forces
where u is the systematic force representing the dynami-
jt = 0, it jt = qijt t, 11 cal friction. The components of vector At satisfy the same
where the coefficients qij = q ji describing the strength of the properties as that of Eq. 11, but
noise do not depend on the variables k and ij is a constant AitA jt = 2qijt t, 21
matrix. For this type of process, the drift coefficient is linear
and the diffusion coefficient constant, that is, q being the noise intensity given by q = kBT / m. If this pa-
rameter is absorbed in the function gij, then it can be shown
Di = ijx j, Dij = D ji , 12 that drift and diffusion coefficient read as
where the matrix Dij is also constant. D i = u i, Dij = qij . 22
A. Transformation of variables The Fokker-Planck equation for the transition probability
The Langevin equations are very convenient to calculate density of the velocity conditioned by initial data u0 at time
the drift and diffusion coefficients if the variable transforma- t0 = 0, i.e., Pu , t u0, as required by Eq. 3 is
041117-2
FOKKER-PLANCK-KRAMERS EQUATION FOR A PHYSICAL REVIEW E 74, 041117 2006
P 2 0 0 0 1 0 0
= ui P + q 2 P = divu uP + qu2 P, 23
t ui ui 0 0 0 0 1 0
0 0 0 0 0 1
where the operator divu and the Laplace operator u2 act with = , 31
respect to the velocity coordinates. The transition probability 0 0 0 0 0
satisfies the initial condition 0 0 0 0 0
0 0 0 0 0
lim Pu,tu0 3u u0
t0
0 0 0 0 0 0
= u1 u01u2 u02u3 u03. 24 0 0 0 0 0 0
Following Chandrasekhars proposal 3, the general so- 0 0 0 0 0 0
lution of Eq. 23 is connected with its associated first-order D=
0 0 0 q 0 0
.
equation, i.e.,
0 0 0 0 q 0
P 0 0 0 0 0 q
divu uP = 0. 25
t In this case we have the vector x = x1 , x2 , x3 , x4 , x5 , x6
= x , y , z , ux , uy , uz. From Eq. 3 we construct the Fokker-
The general solution of this first-order equation involves the Planck-Kramers equation in the absence of a external force
first three integrals of the Lagrangian subsidiary system for the transition probability Pr , u , t u0 , r0 governing
the probability of the simultaneous occurrence of the veloc-
u = u. 26 ity u and the position r at time t given that u = u0 and
r = r0 at t = 0, which can be written as
The required integrals are therefore
P 2
= ui P + ui P + q 2 P, 32
uet = u0 = I1 = const. 27 t xi ui ui
28
lim Pr,u,tu0,r0 3u u03r r0. 34
t0
As the time t this probability density becomes the
Maxwellian distribution, i.e., Again, following Chandrasekhars proposal, the
solution of Eq. 33 is connected with the solution of the
Pu = m
2 k BT
3/2
exp
mu2
2kBT
, 29
first-order equation
P
+ u r P divuuP = 0. 35
t
which corresponds to the stationary distribution.
The general solution of this equation can expressed in terms
of six independents integrals of the Lagrangian subsidiary
C. The Fokker-Planck-Kramers equation for free Brownian system
motion
u = u, r = u, 36
The stochastic differential equation for free Brownian
motion taking into account the position r = x , y , z and the namely
velocity u = ux , uy , uz is uet = I1, r + 1u = I2 , 37
where the constants I1 and I2 are defined as I1 = u0 and
r = u,
I2 = r0 + 1u0. If we define PR , S Pr , u , t u0 , r0, where
u0
u = u + At, 30 R = r r0 1 et,
where At satisfies the same properties given in Sec. II A.
S = u u 0e t , 38
This set of six stochastic differential equations can be written
as the Ornstein-Uhlenbeck process 10, in such a way that then the general solution of Eq. 33 can be expressed in the
the drift matrix and the diffusion matrix D are given by form
041117-3
J. I. JIMNEZ-AQUINO AND M. ROMERO-BASTIDA PHYSICAL REVIEW E 74, 041117 2006
u = u + Wu + At, 46
where
where W is a real antisymmetric matrix given by
q
F = 3 2t 3 + 4et e2t,
0 0
W= 0 0 , 47
q
G= 1 e2t, 0 0 0
= eB / mc being the Larmor frequency. To establish the
q Fokker-Planck equation associated to Eq. 46, this last one
H = 2 1 et2 , must be written as
u = u + At, 48
FG H2 = ab h2e2t , 40
together with the parameters where now reads as
q q 2t 2q 0
a=2 t, b= e 1, h= 1 et. 41
2 2 = 0 . 49
The transition probability PR Pr , t r0 , u0 governing 0 0
the probability of the position r at time t, given that the
particle is at r0 with a velocity u0 at t = 0, can be calculated Equation 48 is a coupled system of equations in the plane
through the integral ux , uy and independent from the coordinate uz, for which
the corresponding evolution equation is the Langevin equa-
PR Pr,tr0,u0 = PR,SdS. 42
tion of the ordinary Brownian motion. The drift and diffusion
coefficients are
Pr,tr0,u0 = 3
2q2t 3 + 4et e2t
3/2
Dij = qij , 50
P
43 = divuuP + qu2 P. 51
t
For large times such that t 1, it reduces to
To solve this equation by following Chandrasekhars idea we
1 r r02 require to solve the first-order equation of Eq. 51
Pr,tr0,u0 exp , 44
4Dt3/2 4Dt without the Laplacian term, which involves the Lagrangian
subsidiary system
D being the Einstein diffusion constant given by
D = q / 2 = k BT / m . u = u, 52
III. THE FOKKER-PLANCK EQUATION FOR A together with the required first integrals
CHARGED PARTICLE IN A MAGNETIC FIELD
The Langevin equation 20 can be adapted to the case uet = u0 = I1 = const. 53
of diffusion of charged particles Brownian gas in the
However, such a solution is not easy to calculate because the
presence of a constant magnetic field that acts upon particles
resulting equations are also a coupled system. To solve this
via the Lorentz force. In this case the corresponding
problem we propose the following strategy. If we make the
Langevin equation is 2,4
change of variables u = eWtu, then the Langevin equation
e 46, in the new velocity space, takes the very simple form
u = u + u B + At, 45
mc
u = u + R1tAt, 54
where e denotes the charge of the particle of mass m and
At satisfies the same properties as that given in Eq. 21. where Rt = eWt is an orthogonal rotation matrix given by
041117-4
FOKKER-PLANCK-KRAMERS EQUATION FOR A PHYSICAL REVIEW E 74, 041117 2006
cos t sin t 0 1
Pu,tu0 =
Rt = sin t cos t 0 . 55 2q1 e2t/3/2
0 0 1
inverse and therefore R1t = eWt. As we can see, the Lange- Returning to the variables u, we first notice that
vin equation 54 is the same as that of the free Brownian
motion except by the factor R1t multiplying the noise u u0et2 u u0et u u0et = u etu02 ,
At, which amounts to a rotation of the noise At. We can 62
observe immediately from Eq. 54 that the drift and the
where is the same as in Eq. 49. Therefore, in the u
diffusion coefficients are
velocity space the transition probability density is given by
Di = ui , 1
Pu,tu0 =
2q1 e2t/3/2
Dij = qR1tikR1t jk = qij .
result as that given in Eq. 50 because (R1t)ik(R1t) jk which is exactly the same as that calculated by Czopnik and
= ij. The interesting point that we can remark from this re- Garbaczewski 4 by other method. Also, in the long-time
sult is the following: the term R1tAt, which represents a limit, it reduces to the Maxwellian distribution
rotation of the noise At, has the same statistical properties
m 3/2
mu2
as At if this last one satisfies the property 21, being the Pu = exp . 64
reason why Dij is the same as Dij. This fact physically means 2 k BT 2kBT
that the rotation matrix with elements sin t and cos t has Since the process is Markovian, then Eq. 63 can be easily
been absorbed by the noise At and therefore such oscillat- extended to the case of arbitrary t0 = t 0. For this case
ing functions will not appear any more in the diffusion con- Pu , t u , t can be derived by substituting anywhere t t
stant of the associated Fokker-Planck equation. On the other instead of t and u instead of u0, where u = ut here we
hand, the drift and diffusion coefficients given in Eq. 50 must not confuse this u with the transformation of variables
can also be derived from the transformations 16 and 17. defined earlier. In the stationary state the joint probability
The Fokker-Planck equation for the transition probability density W2u , t ; u , t may be expressed by the product of
density of the velocity conditioned by initial data u0 at time
the aforementioned transition probability and Eq. 64. So,
t0 = 0, i.e., Pu , t u0, is then for both t t and t t, it can be written as
P
t
2
= divuu P + qu P . 57 W2u,t;u,t =
2q
3
1
1 e2tt3/2
65
lim Pu,tu0 3u u0
t0
This joint probability density uniquely determines a station-
= u1 u01
u2 u02
u3 u03
, ary Markovian stochastic process for which we can calculate
58 various mean values. For instance, the mean values of the
velocity components uit for i = 1 , 2 , 3 are equal to zero, that
the same as in the ordinary Brownian motion. The solution is,
of Eq. 57 is connected with the solution of the associated
first-order equation, which involves the first three integrals of
the Lagrangian subsidiary system
uit =
ui Pudu = 0. 66
t
uitu jt =
uiuj W2u,t;u,tdudu , 67
ue = u0 = I1 = const. 60
with i , j = 1 , 2 , 3. For instance, the autocorrelation function of
Finally, the solution of Eq. 57 reads as the x component of the velocity yields
041117-5
J. I. JIMNEZ-AQUINO AND M. ROMERO-BASTIDA PHYSICAL REVIEW E 74, 041117 2006
q tt P
u1tu1t = e cos t t, 68 + u r P = divuuP + qu2 P. 75
t
with the same expression for the y component. On the other Proceeding as in the previous section, the first-order solution
hand, the z component satisfies the usual expression for free may be expressed in terms of six integrals of the Lagrangian
Brownian motion, i.e., u3tu3t = q / ett. subsidiary system
From these autocorrelation functions, the mean square u = u, r = u. 76
displacement MSD for the position components
r r1 , r2 , r3 x , y , z can easily be calculated if the par- These integrals are
ticle starts at time t = 0 at r = r0 with the velocity u = u0. The uet = I1, r + 1u = I2 , 77
x component is
where now I1 = u0, I2 = r0 + u0, and
1 1
is the inverse of
2 q q matrix given by
x2 = 2D 2 2 t 2 1 e
1t
1 e2t,
+ 1 22
0
69 2 + 2 2 + 2
where D = q / 2 = kBT / m, which is consistent with the
1 = 2 0 . 78
value of the Einsteins diffusion constant, 1 = i, and + + 2
2 2
similar to those of ordinary Brownian motion. This can be
2 achieved if we define r = u in such a way that
x2 = y2 = 2D t, 71
2 + 2
u = eWtu, r = eWtr . 79
and along the magnetic field
In this case Eq. 73 is transformed into
z = 2Dt,
2
72 r = u ,
where D = q / = kBT / m is the Einsteins diffusion constant.
2
r = u, uet = I1, r + 1u = I2 , 83
where I1 = u0 and I2 = r0 + 1u0.
u = u + At. 74 Before proceeding, we must establish the transformation
which makes the connection between the representations
The Fokker-Planck-Kramers equation associated with the r , u and r , u and which is also consistent with the trans-
system 74 is then formation 80. The desired transformation can be obtained
041117-6
FOKKER-PLANCK-KRAMERS EQUATION FOR A PHYSICAL REVIEW E 74, 041117 2006
with the aid of Eqs. 77 and 83, together with the trans- 2 + 2
formation u = eWtu, such that JS = 1, JR = , 95
2
r = 1eWtr I2 + I2 . 84 and therefore P = JR P. On the other hand, if we define the
If we use this result together with the transformation for u vectors on the xy plane as S = S1 , S2 and R = R , R , then
1 2
in Eqs. 16 and 17 we obtain the same drift and diffu-
S = S , S3 and R = R , R3, where S3 and R3 are the z compo-
sion D matrices as those given in Eq. 31. Under these nents of the vectors S and R, such that S3 = uz u0zet and
circumstances, if we define PR , S Pr , u , t u0 , r0, R3 = z z0 1u0z1 et, with u0z = uz0 and z0 = z0.
then the solution of Eq. 81, with the initial condition
From these definitions, the vectors S and R can be written as
lim Pr,u,tu0,r0 3u u03r r0, 85
t0 S = u e t u , = r r
R u , 96
0 0 0
can be written in the same form as Eq. 39, i.e.,
11 e t
where
, with the matrices
1
PR,S =
8 FG H23/2
3
= , 2 + 2 2 + 2
exp
FS2 2HR S + GR2
2FG H2
,
1 =
2
+ + 2
2 2
. 97
86
Under these circumstances, it can be shown that
where now
2 + S2 ,
S2 = S2 = S 98
u 3
R = r r0 0 1 et,
R2 = R
2 + R2 ,
3 99
t
S = u u0e , 87
+ S
S R = S R R
+S R , 100
and the parameters F, G, and H are the same as in Eqs. 40 z 3 3
and 41.
Returning to the transition probability density P we use where S R = S R S R is the z component
z 1 2 2 1
the fact that PdSdR = PdSdR, where the volume element of the cross product and = 2 + 2 / 2. Substituting
transforms as Eqs. 98100 into Eq. 86 we can verify that the
dSdR = JdSdR , 88 TPD for P can be written as the product of two
independent transition probability densities, that is
J being the Jacobian of the transformation and therefore PR , S PR
, SP z , u , t u , z , where
z z 0z 0
JP = P. It can be shown that J = JSJR, where
JR
JS Det Si/ Sj = 1/JS = 1/Det Si/ S j. 89 PR
,S
=
4 FG H2
2
Similarly
JR Det Ri/ Rj = 1/JR = 1/Det Ri/ R j.
Then, J = JSJR = 1 / JSJR = 1 / J. The explicit form of these
transformations can be calculated if we first observe from
90
exp
FS S + 2 HS
2 2HR
2FG H2
R
z
+ GR
2
Eq. 87 that 101
t
S = u u0e =e Wt
S, 91 corresponds to the planar TPD which describes the diffusion
process on the xy plane and
where we define
1
S = u etu0 . 92 Pzz,uz,tu0z,z0 =
42FG H21/2
For R we use Eq. 84 to show that
R = r r0
u0
1 et = eWtR, 93 exp FS23 2HR3S3 + GR23
2FG H2
, 102
with is the TPD which describes the diffusion process along the z
axis and is the same as that of the ordinary Brownian motion,
R = r r0 u0 , 94 as expected. From Eq. 101, we can calculate the spatial
and 1 e
1 t
. It can be shown, from Eqs. 91 and transition probability density PR Pr
, t r0 , u0 through
93 that the integral
041117-7
J. I. JIMNEZ-AQUINO AND M. ROMERO-BASTIDA PHYSICAL REVIEW E 74, 041117 2006
PR
= PS
,R
dS
. 103
the planar JPD PR , S, where now the vectors R
are defined on the xy plane, that is R
and S
= R , R and S
1 2
After a long but straightforward algebra it reduces to = S1 , S2, and have the same expressions as that given by Eq.
87, i.e.,
Pr 1
= r r u0 1 et,
,tr 0 =
0,u
2De/2t 3 + 4et e2t R
0
u 2
r0
r 0
exp ,
2De/2t 3 + 4et e2t S = u uet .
0 110
104 We can easily see from Eq. 80 that
where De = D / + , which corresponds to a normaliza-
2 2 2
t
tion of the diffusion constant D. For large times such that S = et e sA
sds, 111
t 1, it immediately reduces to 0
Pr
,tr 0
0,u
1
4Det
exp
r02
r
4Det
. 105 = 1et
R t
0
e sA
sds + 1
0
t
sds, 112
A
x2 = y2 = 2D 2
2 + 2
t. 106
out the third row and the third column of matrix given in Eq.
55. Both vectors S and R correspond to Gaussian distri-
butions, each one with zero mean value. Therefore the planar
Evidently, PS
Pu
, t u0 can also be obtained from the JPD is given by
integral
PR
,S
= 1 1
1/2 exp cij xix j ,
1
4 det C
2
2 i,j
PS
= PS
,R
dR
, 107
113
with the result where C = cij = xix j is the matrix of variances and covari-
ances, such that the variable x = x1 , x2 , x3 , x4
1
Pu 0 =
,tu R1 , R2 , S1 , S2 presents a vector in the four-dimensional
2q1 e2t/ phase space and c1
ij represents the components of the inverse
etu02
u matrix C1. From Eq. 111 it can be shown that
exp , 108
2q1 e2t/ q
G = S1S1 = S2S2 = 1 e2t, 114
which corresponds to the planar TPD for the velocity space.
Clearly, Eq. 63 can also be written as the product of two while S1S2 = S2S1 = 0. From Eq. 112 we have
independent TPD; one is the TPD 108 and the other is
q
1 F = R1R1 = R2R2 = 2t 3 + 4et e2t
Pzuz,tu0z = 3
2q1 e2t/1/2
115
exp
uz u0zet2
2q1 e2t/
, 109 and R1R2 = R2R1 = 0. We also show that
q
corresponding to the ordinary Brownian motion. H = R1S1 = R2S2 = 1 et2 116
In the following section we will show that the general 2
TPD 101 is equivalent to the joint probability distribution
and R1S2 = R2S1 = 0. Thus, the covariance matrix C = cij
JPD PR
, S of the vectors R
and S defined above.
reads as
B. General planar JPD through the correlation matrix G 0 H 0
formalism 0 G 0 H
C= , 117
The interesting point of the general expression 101 is H 0 F 0
that, although very similar, it is not the same to the calculated 0 H 0 F
by Czopnik and Garbaczewski. To show this argument, we
will use the same strategy used by these authors to calculate and its inverse matrix will be
041117-8
FOKKER-PLANCK-KRAMERS EQUATION FOR A PHYSICAL REVIEW E 74, 041117 2006
F 0 H 0 such periodic functions, but it does not appear in our pro-
2 0 0 H posal because of the absorption of such periodic functions by
FG H F
C1 = , 118 the noise. Thus, k is an independent parameter in our pro-
det C H 0 G 0 R appearing
posal which multiplies the cross product S z
0 H 0 G in the JPD of Ref. 4. From the above arguments we can
where det C = FG H . Therefore, the planar JPD of Eq.
2 2 point out the following: when we make = 0, it can be
113 can be written as checked that both TPDs reduce to the planar TPD of ordi-
nary Brownian motion, as expected. In this case the param-
PR
,S
= 1 eters f, g, and h will be the same as those given in Eq. 40
42FG H2 together with k = 0.
exp 2 2HR
FS S + GR
FG H
2
, V. CONCLUSIONS
041117-9
J. I. JIMNEZ-AQUINO AND M. ROMERO-BASTIDA PHYSICAL REVIEW E 74, 041117 2006
7 I. Holod, A. Zagorodny, and J. Weiland, Phys. Rev. E 71, and Applications Springer-Verlag, Berlin, 1984.
046401 2005. 10 E. Nelson, Dynamical Theories of Brownian Motion Princeton
8 A. Zagorodny and I. Holod, Condens. Matter Phys. 3, 295 University Press, Princeton, 1967.
2000. 11 J. L. Doob, Ann. Math. 43, 351 1942.
9 H. Risken, The Fokker-Planck Equation: Methods of Solution
041117-10