Professional Documents
Culture Documents
The theory of Brownian motion is perhaps the simplest approximate way to treat the
dynamics of nonequilibrium systems. The fundamental equation is called the Langevin
equation; it contain both frictional forces and random forces. The fluctuation-dissipation
theorem relates these forces to each other.
The random motion of a small particle (about one micron in diameter) immersed in a
fluid with the same density as the particle is called Brownian motion. Early investigations
of this phenomenon were made by the biologist Robert Brown on pollen grains and also
dust particles or other object of colloidal size.
The modern era in the theory of Brownian motion began with Albert Einstein. He ob-
tained a relation between the macroscopic diffusion constant D and the atomic properties
of matter. The relation is
RT kB T
D= =
NA 6πηa 6πηa
where R is the gas constant, NA = 6.06 × 1023 /mol is Avogadros number, T is the tem-
perature, η is the viscosity of the liquid and a is the radius of the Brownan particle. Also
kB = R/NA is Boltzmanns constant.
The theory of Brownian motion has been extended to situations where the fluctuating
object is not a real particle at all, but instead some collective porperty of a macroscopic
system. This might be, for example, the instantaneous concentration of any component
of a chemically reacting system near thermal equilibrium. Here the irregular fluctuation
in time of this concentration corresponds to the irregular motion of the dust particle.
75
76 Chapter 6 Brownian Motion: Langevin Equation
Figure 6.1: A large Brownian particle with mass M immersed in a fluid of much smaller
and lighter particles.
and τr is the relaxation time for the Brownian particle, i.e. the time the particle have
diffused its own radius
a2
τr =
D
In general
τs τB τr .
In dense colloidal suspensions τr can become very long of the order of minutes or hours.
While the motion of a dust particle performing Brownian motion appears to be quite
random, it must nevertheless be describable by the same equation of motion as is any other
dynamical system. In classical mechanics these are Newton’s or Hamiltons equations. For
simplicity we will consider motion in one dimension. The results can easily be generalised
to three dimensions. Newtons equation of motion for the particle (radius a, mass m,
position x(t), velocity v(t)) in a fluid medium (viscosity η) is
dv(t)
m = F (t) (6.1)
dt
where F (t) is the total instantaneous force on the particle at time t. This force is due to
the interaction of the Brownian particle with the surrounding medium. If the pssitions
of the moelcules in the surrounding medium are known as a function of time, then in
principle this force is a known function of time. In this sense it is not a random force at
all.
It is usually not practical or even desirable to look for an exact expression for F (t).
Experience tells us that in typical cases this force is dominated by a friction force −γv(t),
6.1 Langevin equation 77
proportional to the velocity of the Brownian particle. The friction coefficient is given by
Stokes law
γ = 6πηa (6.2)
We also expect a random force ξ(t) due to random density fluctuations in the fluid. The
equations of motion of the Brownian particle are:
dx(t)
= v(t)
dt
dv(t) γ 1
= − v(t) + ξ(t) (6.3)
dt m m
This is the Langevin equations of motion for the Brownian particle.
The random force ξ(t) is a stochastic variable giving the effect of background noise
due to the fluid on the Brownian particle. If we would neglect this force (6.3) becomes
dv(t) γ
= − v(t) (6.4)
dt m
which has the familiar solution
m
v(t) = e−t/τB v(0), τB = (6.5)
γ
According to this, the velocity of the Brownian particle is predicted to decay to zero at
long times. This cannot be true since in equilibrium we must have the equipartion theorem
kb T
hv 2 (t)ieq = (6.6)
m
while (6.5) gives
hv 2 (t)ieq = e−2t/τB hv 2 (0)ieq → 0 (6.7)
The random force in (6.3) is therfore necessary to obtain the correct equilibrium. In the
conventional view of the fluctuation force it is supposed to come from occasional impacts
of the Brownian particle with molecules of the surrounding medium. The force during an
impact is supposed to vary extremely rapidly over the time of any observation. The effect
of the fluctuating force can be summarized by giving its first and second moments
The average h· · ·iξ is an average with respect to the distribution of the realizations of the
stochastic variable ξ(t).
Since we have extracted the average force −γv(t) in the Langevin equation the average
of the fluctuating force must by definition be zero. g is a measure of the strength of the
fluctuation force. The delta function in time indicates that there is no correlation between
impacts in any distinct time intervals dt1 and dt2 . This loss of correlation is a consequence
of the separation of time scales discussed above. During a short time interval dt on scale
τB = 10−3 s, say dt = 10−5 s, there are still roughly dt/τs ≈ 107 collisons with the atoms
in the liquid. Therefore any memory between forces at different times will be lost due to
these frequent collisions.
78 Chapter 6 Brownian Motion: Langevin Equation
The remaining mathematical specification of this dynamical model is that the fluctu-
ating force has a Gaussian distribution determined by the moments in (6.8).
The property (6.8) imply that ξ(t) is a wildly fluctuating function, and it is not at
all obvious that the differential equation (6.3) has a unique solution for a given initial
condition, or even that dv/dt exists. There is a standard existence theorem for differential
equations which guarantee the existence of a local solution if ξ(t) is continous. A local
solution is one which exists in some neighborhood of the point at which the initial value
is given. But even if a solution exists it may be only local, or it may not be unique, unless
some stronger conditions are imposed on ξ(t).
We can obtain an explicit formal solution of (6.3) as
1
Z t
−t/τB
v(t) = e v(0) + dse−(t−s)/τB ξ(s) (6.9)
m 0
but this only transfers the problem elsewhere. How do we know that the integral in (6.9)
exists, that it is more than just a formal symbol?
To obtain a meaning to (6.3) and (6.9) we write (6.3) as
γ 1
dv(t) = − v(t)dt + dU (t) (6.10)
m m
where
dU (t) = ξ(t)dt (6.11)
For an arbitrary nonstochastic continous function f (t) we then find
Z t γ
Z t 1
Z t
f (s)dv(s) = − f (s)v(s)ds + f (s) dU (s) (6.12)
0 m 0 m 0
γ t 1
Z
v(t) − v(0) = − v(s)ds + [U (t) − U (0)]
m 0 m
γ 1
= − [x(t) − x(0)] + [U (t) − U (0)] (6.13)
m m
We now discuss the integral noice U (t) for long times t. Dividing t into intervals we
have
n
X
U (t) − U (0) = [U (tk ) − U (tk−1 )] (6.14)
k=1
very small even though during it very many collisions occur. These numerous impacts
destroy all correlations between what happens during the time interval (ti , ti−1 ) and what
has happended before ti−1 . This implies that U (t) is a Markov process, i.e. U (tn ) depends
only on U (tn−1 ) etc.
On account of the randomness of the motion the random force ξ(t) must average to
zero, which is also implied by the separation in (6.3). If we now choose our time origin so
that U (0) = 0 we must have hU (tk )i = 0.
From the discussion above about the random collisions we also argue that the incre-
ments
U (t1 ) − U (0), U (t2 ) − U (t1 ), · · · , U (tn ) − U (tn−1 ) (6.16)
are independent. For long times we suppose that the random motion of the medium has
attained a steady state. Then the increments (6.16) are also stationary and identically
distributed with zero mean. Applying the central limit theorem to (6.14) we deduce that
U (t) is Gaussian with zero mean. Therefore, it has all the requirements for a Wiener
process, i.e.
U (t) = W (t) (6.17)
We can now write (6.10) as
γ 1
dv(t) = − v(t)dt + dW (t) (6.18)
m m
and the solution in (6.9) becomes
1
Z t
v(t) = e−t/τB v(0) + e−(t−s)/τB dW (s) (6.19)
m 0
6.2 Examples
Ornstein-Uhlenbeck process
In the Ornstein-Uhlenbeck process we study a Brownian particle where the equation of
motion is given by (6.3) or
Z t
x(t) = x0 + v(s)ds
0
1
Z t
v(t) = e−t/τB v0 + e−(t−s)/τB dW (s) (6.20)
m 0
where x0 = x(0) and v0 = v(0). Since dW is a Gaussian process this is also the case
for v(t) and x(t). We can therefore obtain the distribution functions for these variables
knowing the first and second moments. Taking the average of the second eq. in (6.20) and
using hdW (t)i = 0 we find
hv(t)i = v0 e−t/τB (6.21)
From (6.20) we can also calculate the correlation function
where we in step two changed the order of integration. The average displacement is then
h i
µx (t) = hx(t)iξ = x0 + v0 τB 1 − e−t/τB (6.31)
An important quantity is the mean squared displacement of the particle from the starting
point. This is obtained as
h i2
h[x(t) − x0 ]2 iξ = v02 τB2 1 − e−t/τB
τB2
Z tZ th ih i
+ 1 − e−(t−u)/τB 1 − e−(t−v)/τB hdW (u)dW (v)iξ
m2 0 0
gτ 2 t h
i2 i2
h Z
= v02 τB2
1−e −t/τB
+ B2 1 − e−(t−u)/τB
m 0
i2 g g h
h i
2 −t/τB 2
= τB 1 − e v0 − + 2 t − τB 1 − e−t/τB (6.32)
2mγ γ
In equilibrium the first term vanishes as before. From the remaining second term we get
for short and long times
kB T 2
(
hh(x(t) − x0 ) ii = 2 m t t→0
(6.33)
2kB T
γ t t→∞
The result for short times is the free particle term, where for short times x(t) − x0 = v0 t.
The result for long times can be compared with the diffusion result
2kB T τB2 3 1
σx2 (t) = h(x(t) − µx (t))2 i = t/τB − + 2e−t/τB − e−2t/τB . (6.36)
m 2 2
The conditional distribution function for the displacement is then
1/2 (x−µx (t))2
1
− 2 (t)
%x (xt|x0 0) = e 2σx
(6.37)
2πσx2 (t)
From the relation between the position and velocity we can also get a general Kubo
relation Z t
x(t) − x(0) = ds v(s) (6.38)
0
which for a stationary process gives
Z t Z t
2
h(x(t) − x(0)) i = ds1 ds2 hv(s2 )v(s1 )i
0 0
82 Chapter 6 Brownian Motion: Langevin Equation
Z t Z s2 Z t Z t
= ds2 ds1 hv(s2 )v(s1 )i + ds2 ds1 hv(s2 )v(s1 )i
0 0 0 s2
Z t Z s2 Z t Z s1
= ds2 ds1 hv(s2 − s1 )v(0)i + ds1 ds2 hv(s1 − s2 )v(0)i
0 0 0 0
Z t Z s Z t Z t
= 2 ds duhv(u)v(0)i = 2 duhv(u)v(0)i ds
0 0 0 u
Z t
= 2 du(t − u)hv(u)v(0)i
0
dx
= v
dt
dv γ 1
= − v − ω02 x + ξ(t)
dt m m
where ω02 = k/m. Here the random force is a Gaussian random process with moments
hξ(t1 )i = 0
hξ(t1 )ξ(t2 )i = 2γkB T δ(t1 − t2 )
−iωx(ω) = v(ω)
γ 1
−iωv(ω) = − v(ω) − ω02 x(ω) + ξ(ω)
m m
Then we can solve for x(ω) in terms of the fluctuating force to obtain
1 ξ(ω)
x(ω) = 2 γ
m ω0 − ω 2 − m iω
6.2 Examples 83
1
(a) (b)
5
4 0.5
Sx(ω)
Cx(t)
3
2 0
−0.5
0 0.5 1 1.5 2 2.5 3 0 1 2 3 4 5 6 7 8
ω t
Figure 6.2: (a) The spectrum Sx (ω)/hx20 i for a Brownian particle in a harmonic potential
plotted versus ω/ω0 for γ/m = 0.5, 1.0 and 1.5ω0 (b) The correspondning results for the
normalized correlation function Cx (t)/Cx (t = 0) plotted versus tω0 .
The integral can be calculated as a contour integral in the complex plane by taking
the residues at the poles. These are located at
s
γ γ γ2 γ
ω02 − ω 2 = ± iω, ⇒ ω = ±i ± ω02 − 2
= ±i ± ω1
m 2m 4m 2m
q
with ω1 = ω02 − γ 2 /4m2 . For t > 0 and ω = x + iy we see that the exponent −iωt =
−ixt + yt, and the contour in the complex ω-plane have to be closed by a semicircle in the
lower half plane. The poles are then at ω = ±ω1 − iγ/2m. By working out the residues
one find
2γkB T ∞ e−iωt
Z
Cx (t) = dω γ γ γ γ
2πm2 −∞ (ω − ω1 − i 2m )(ω − ω1 + i 2m )(ω + ω1 − i 2m )(ω + ω1 + i 2m )
" γ γ #
−i(ω1 −i 2m )t −i(−ω1 −i 2m )t
2γkB T e e
= −2πi 2 γ γ + γ γ
2πm (−i m )(2ω1 − i m )(2ω1 ) (−2ω1 − i m )(−2ω1 )(−i m )
kB T − γ t γ
= 2 e 2m cos ω1 t + sin ω1 t
mω0 2mω1
We notice that at t = 0 we have Cx (t = 0) = hx20 i = kB T /mω02 which is just the
equipartiton theorem
1 2 2 1
ω0 hx0 i = kB T
2 2
84 Chapter 6 Brownian Motion: Langevin Equation
Sv (ω) = ω 2 Sx (ω)
and from this we can get the velocity correlation function Cv (t).
2πω 2 β ∞
Z
α(ω) = dte−iωt hM (t) · M (0)ieq
3nc −∞
where c is the speed of light in vacuum, β = 1/kB T and n is the index of refraction. M (t)
is the total electric dipole moment of the system at time t.
Suppose the system being investigated is a single dipolar molecule. Then M is just
its permanent dipole moment. It has a constant magnitude µ and a time dependent
orientation specified by the unit vector u(t) so that
For a planar rotor with θ(t) the instantaneous angle with the x-axis
and
hu(t) · u(0)ieq = hcos θ(t) cos θ(0) + sin θ(t) sin θ(0)ieq = hcos(θ(t) − θ(0))i
= Reheiθ(t) e−iθ(0) ieq
We can calculate this quantity using the Langevin equation for rotational Brownian mo-
tion. The position x is replaced by the angle θ and the velocity v by the angular velocity
Ω, and the mass m by the moment of inertia I
dθ
= Ω
dt
dΩ
I = −γΩ + ξ(t)
dt
where the fluctuating torque has correlation function
1
Z t
−t/τR
Ω(t) = Ω(0)e + ds e−(t−s)/τR ξ(s)
I 0
6.2 Examples 85
and
Z t 1
Z t Z s
θ(t) = θ(0) + dsΩ(0)e−s/τR + ds due−(s−u)/τR ξ(u)
0 I 0 0
1
Z t
= θ(0) + τR Ω(0) 1 − e−t/τR + ds 1 − e−(t−s)/τR ξ(s)
γ 0
But ∆θ(t) = θ(t) − θ(0) is linear in the noise and in the initial angular velocity, and both
these have a Gaussian distribution. Then ∆θ(t) also has a Gaussian distribution with a
zero mean value and a second moment h(∆θ(t))2 i. For a Gaussian vaiable X with mean
value µX and variance σX2 we have the characteristic function
2 σ 2 /2
heikX i = eikµX −k X
Therefore
2kB T I h
i
−h(∆θ(t))2 i/2
−t/τR
CR (t) = e = exp − t/τR − 1 − e
γ2
For short and long times we then have
1 kB T 2
CR (t) = exp − t , t τR
2 I
kB T
CR (t) = exp − t , t τR (6.40)
γ
The rotational correlation function is directly related to the dielectric susceptibility
which is measured in dielectric relaxation measurements. The relation is
Z ∞
dCR (t)
−iωt
(iω) = ∞ + (0 − ∞ ) e − dt
0 dt
= ∞ + (0 − ∞ ) [1 − iωCR (iω)] (6.41)
Here 0 is the static dielectric constant for ω = 0 and ∞ the corresponding one for high
frequencies. An exponential relaxation for long times, as obtained above, is seen for single
molecules or low densities. For high densities the relaxation is more complex and one often
observes a so called Kohlrausch- Williams-Watts form for CR (t)
β
CR (t) = e−(t/τ ) , 0<β<1
with the relaxation time τ increasing rapidly with decreasing temperature or increasing
density. The corresponding imaginary part of the dielectric susceptibility 00 (ω) has then
a much broader peak than the Lorentzian shape obtained for an exponential function.
86 Chapter 6 Brownian Motion: Langevin Equation
d2 d
m r (`, t) = −γ r(`, t) + k [r(` + 1, t) − 2r (`, t) + r(` − 1, t)] + ξ(`, t)
dt2 dt
Here γ is the friction constant. Neglecting inertial effects, i.e. the acceleration term this
gives
d k 1
r (`, t) = [r (` + 1, t) − 2r (`, t) + r (` − 1, t)] + ξ(`, t)
dt γ γ
Let’s consider a ring polymer or a chain with periodic boundary conditions. Then a
discrete Fourier transform is
N
eiq` r (`, t)
X
R(q, t) =
`=1
and where
2πk
, k = 1, . . . , N
qk = −π +
N
This follows since periodic boundary conditions imply r (`, t) = r(`+N, t) and so exp(iqN ) =
1. We also have the relation
N 0
(
X
i(q−q 0 )` i(q−q 0 ) 1 − ei(q−q )N 0 q=6 q0
e =e =
1 − ei(q−q0 ) N q = q0
`=1
and similarly when summing over qk . For R(q, t) we find the equation
d 2k 1
R(q, t) = − [1 − cos q] R(q, t) + ξ(q, t)
dt γ γ
1
Z t
R(q, t) = ds e−(1−cos q)(t−s)/τ ξ(q, s)
γ −∞
1 X ∞
Z
r(`, t) = ds e−(1−cos q)(t−s)/τ e−iq` ξ(q, t − s)
Nγ q 0
6.2 Examples 87
which is a measure of the mean squared displacement for the different beads. Then
1 XX ∞
Z Z ∞
0 0
C(`, t) = ds ds0 e−(1−cos q)(t−s)/τ e−(1−cos q )(t−s )/τ
N 2 γ 2 q q0 0 0
Dh i h 0
iE
× e−iq` ξ(q, t − s)ξ(q, −s) · e−iq ` ξ(q 0 , t − s0 )ξ(q 0 , −s0 )
Now, the fluctuating force ξ is uncorrelated on different beads and at different times
and so
N X
N N
α β 0 0 iq` iq 0 `0 α β 0 0 0
ei(q+q )` δ(t − t0 )δα,β
X X
hξ (q, t)ξ (q , t i = e e hξ (`, t)ξ (` , t i = g
`=1 `0 `=1
= N gδq+q0 ,0 δ(t − t0 )δα,β
Inserting this into the expression for C we can perform the time integrals and one sum-
mation over q, to obtain
3g X X ∞
Z Z ∞
0
C(`, t) = 2 2
ds ds0 e−(1−cos q)(t−s)/τ e−(1−cos q)(t−s )/τ
N γ q q0 0 0
h 0
i
× δq+q0 ,0 δ(s − s0 ) − e−iq` δ(t − s + s0 ) − e−iq ` δ(t − s + s0 ) + δ(s − s0 )
3g X 1 − e−(1−cos q)t/τ cos q`
=
N 2kγ q 1 − cos q
10
C(l,t)/C(1,0)
5
0
0 10 20 30
t/τ
Figure 6.3: The function C(`, t)/C(1, 0) versus tτ for ` = 0− 4. The asymptotoc behaviour
for t/τ 1is shown as dashed curves.
To find a more explicit expression for the timedependence we can take the derivative
of C(`, t)
∂ 3g 1 π Z
3g 1
Z π
C(`, t) = 2
dq e−(1−cos q)t/τ cos q` = 2 e−t/τ dq eiq` ecos q(t/τ )
∂t γ 2π −π γ 2π −π
3g −t/τ
= e I` (t/τ )
γ2
where I` is the modified Bessel function of order `. Integrating we find
g
C(`, t) = C(`, 0) + S(`, t/τ )
2kγ
where
`−1
S(`, x) = xe−x [I0 (x) + I1 (x)] + ` e−x I0 (x) − 1 + xe−x
X
(` − n)In (x)
n=1
In figure 6.3 C(`, t) is plotted for ` = 0 − 4. For t τ we have the asymptotic expansion
ex
In (x) = , x1
(2πx)1/2
p
which gives a (t) dependence of C(`, t) for large times. In particular for ` = 0 this gives
the long time behaviour
g kt 1/2
C(0, t) =
γk πγ
The dashed curves in fig. 6.3 shows this long time behaviour. For small `-values this
asymptotic behaviour is reached for rather short times.