THE LANGEVIN EQUATION FOR RW
The Langevin equation is a stochastic equa-
tion of motion. In the present context we
shall use it as an equation of motion for ran-
dom walk.
Let us start with Newton's equation
dx
a
ae
Here m is the mass of the particle and F the
total force acting on the particle
The force F consists of three parts: 1) an
external of drift force Faq, 2) a friction force
F;, and 3) a molecular force Fim.
F=FgtFe+Fm
The drift force Fy is due to external force
fields
Gravitation
Fy =-mg , 9 = gravitation constantCharged particle, Lorentz force
Fg =e(E+v xB), v velocity , e charge
The friction force Fr is due to interaction
with the medium. For small velocities the
force is given by Stokes law and is linear in
the velocity
Fe=—av
Here a is the friction coefficient
The molecular force Fm characterizes the
molecular bombardment of the particle. The
force is random and corresponds to noise.
The molecular force produces random walk.
Fim = n(t)
We thus obtain the equation of motion
ar dx
—~+4q—=F t
mas tag = Fa tn)
Overdamped case - no drift forceIn the overdamped case the damping term
adz/dt dominates the dynamics and is much
larger that the inertial term md?a/dt?,
ae > oe
dt dt?
Setting Fy = 0 we obtain the equation of
motion
dx
~=nlt
og = 18)
and absorbing a@ by a scaling of t, t/a +t
dx
—=n(t
Eo)
This is the simplest form of a Langevin equa-
tion. It is first order in time and thus de-
scribes an irreversible process. The equation
is driven by the random force n(t) which fluc-
tuates in time. Below we sketch a time series
for (t)
Wey
a AIIARAR DR NAM ah AMMAIn discrete time we have
a(t + At) = x(t) + n@) AC)
and we note that the random force 'kicks’
the particle at time t. In order to produce
ordinary random walk we assume that the
distribution for n(t) is the same as the distri-
bution P(§) for the step in random walk
Plo We
over,
Kt)
Noise correlation function
In order to invoke the fundamental time scale
separation which is the foundation for ran-
dom walk, i.e., the fact that the time scale
of the molecular collisions happen on a muchfaster time scale than the random walk itself,
we assume that 7*(t) is correlated according
to
(nn) = 5 = (nn?)
We have here reinstated the index a for the
cartesian component. The noise is uncorre-
lated at different times. Assuming statistical
independence we moreover obtain
a8
(nn?) = 8°?) = (9?)
and
608
(n° (t)n*(t)) = SP et t+)
We also assume that the noise fluctuates
about zero, i.e.,
(n®(t)) = 0
Time average - ensemble average
We usually assume ergodicity and thus the
equivalence of a time average and an ensem-
ble average.The time average is defined as
1 _fL/2
y= in [a
In the time average we simply average over
the time evolution of A(t) for a very long
time.
The ensemble average can be performed in
the case we know the ensemble and is defined
as
(A(t) = 0 ALPi(t)
l
Here P,(t) is the probability of the configura-
tion 1.
In thermodynamic equilibrium life is simple,
we know the ensemble:
P, = exp(—Ej/kT)/ S~ exp(—E,/kT)
i
In general for nonequilibrium situation we do
not know the ensemble at the outset.White Noise - Power Spectrum
In the present case we have assumed that
the noise is delta function correlated in time.
This is called ‘white noise’ because the power
spectrum is flat.
The power spectrum of the noise is defined
according to
P,(w) = / dte™* (n™(t)n(0)) = const.
‘The power spectrum is just the Fourier trans-
form and yields the harmonic content at a
Particular frequency. In the present case all
harmonics enter with the same weight - there-
fore white noise.
Uncorrelated noise events yields a flat power
spectrum.
Gaussian white noise
So far we have only assumed that the noise is
white, i.e., correlated at short times, yieldinga flat power spectrum. We shall now fur-
thermore assume that the distribution of the
noise at a given time instant is Gaussian.
P(n?) x en PIA
with a line width of order VA
The probability distribution for the white Gaus-
sian noise signal over a period T is thus
T
P(n®(t),0