Physics 127b: Statistical Mechanics
Brownian Motion
Brownian motion is the motion of a particle due to the buffeting by the molecules in a gas or liquid.
The particle must be small enough that the effects of the discrete nature of matter are apparent, but
large compared to the molecular scale (pollen in the early experiments, various plastic beads these
days). It is a convenient example to display the residual effects of molecular noise on macroscopic
degrees of freedom. I will use this example to investigate the type of physics encountered, and the
tools used to treat the fluctuations.
Random Walk
The first observation of Brownian motion is that the particle under the microscope appears to
perform a “random walk”, and it is first useful to study this aspect in its simplest form.
Lets consider first a one dimensional random walk, consisting of n jumps of �l along the x axis.
We take n to be even (the odd case is essentially the same, but differs in minor details). For the
particle after n jumps to be at x D ml there must have been 12.nCm/ forward jumps, and 12.n−m/
backwards jumps (in any order), and m must be even. The probability of arriving at x D ml is
therefore
pn.m/ D n!�1
2.n−m/
�
!
�1
2.nCm/
�
!
: (1)
For large m; n Stirling’s approximation n! ’ .2�n/1=2.n=e/n gives
pn.m/ D 2p
2�n
e−m2=2n: (2)
This is a Gaussian probability centered around m D 0 (the most probable and mean position is the
origin) and the mean square displacement 〈m2� D n, or〈
x2
� D nl2: (3)
For large n the discreteness of the displacements is unimportant compared to the root mean square
distance of the walk. Transforming to a continuous variable x and a probability density p.x; t/
usingpn.m/ D p.x/�2l (since the interval between the discrete results is dx D 2l) and introducing
time supposing there are n jumps in time t
p.x; t/ D 1p
4�Dt
exp
�
− x
2
4Dt
�
(4)
where we have written
nl2=2t D D: (5)
We recognize that this is the expression for diffusion, with p satisfying
@p
@t
D D@
2p
@x2
; p.x; t D 0/ D �.x/ (6)
1
Administrator
高亮
with the diffusion constant D. In terms of D〈
x2
� D 2Dt: (7)
These results are readily extended to 3 dimensions, since we can consider a walk with steps
.�l;�l;�l/ for example, so that the walk is the product of walks in each dimension. The mean
square distance gone after n walks is again
〈
r2
� D nL2 with L D p3l the length of each step. The
probability distribution p.Ex; t/ satisfies the 3d diffusion equation
@p
@t
D 1
.4�Dt/3=2
exp
�
− r
2
4Dt
�
(8)
with r2 D x2 C y2 C z2. This equation is simply the product of three 1d diffusion equations with
D D nl2=2t as before. The means square distance is〈
r2
� D 6Dt: (9)
(The results in 2d can similarly be constructed.)
The fact that the mean displacement is zero, and the mean square displacement grows linearly in
time can be derived by very simple arguments. Lets consider the two dimensional case of a random
walk consisting of n vectors of length s but with arbitrary angles �i taken from a uniform probability
distribution. The total displacement in the x direction is
X D
X
i
s cos �i: (10)
Clearly hXi D 0 since cos �i is equally likely to be positive or negative. On the other hand
〈
X2
� D * X
i
s cos �i
!2+
D s2
*X
i
cos �i
X
j
cos �j
+
(11)
D s2
*X
i
.cos �i/
2
+
D ns2=2 (12)
where we have used the fact that
DP
i;j 6Di cos �i cos �j
E
D 0 since again each cos �i is equally likely
to be positive or negative. Thus the mean square distance is〈
R2
� D 〈X2 C Y 2� D ns2: (13)
This specific result is useful in adding complex numbers with random phases: the average amplitude
is zero, and the mean square magnitude (the “intensity”) scales linearly with the number of vectors.
Some general nomenclature
The position x.t/ in a one dimensional random walk forms a one dimensional random process—in
general a scalar function y.t/ for which the future data is not determined uniquely by the known
initial data.
2
Administrator
高亮
Administrator
高亮
Administrator
高亮
The random process is in general characterized by probability distributions p1; p2 : : : such that
pn.y1; t1I y2; t2 : : : I yn; tn/dy1dy2 : : : dyn (14)
is the probability that a single process drawn from the ensemble of processes will take on a value
between y1 and y1 C dy1 at t1 etc. The different pn are related byZ 1
−1
pndyj ! pn−1: (15)
Ensemble averages are determined by integrating over the appropriate distribution, e.g. for the mean
and the two point correlation function
hy.t1/i D
Z 1
−1
y1p1.y1; t1/dy1; (16)
hy.t1/y.t2/i D
Z 1
−1
Z 1
−1
y1y2 p2.y1; t1I y2; t2/dy1dy2: (17)
Higher order correlation functions require the knowledge of higher order distribution functions. In
the random walk we have just looked at p1.
A stationary random process is one for which the pn depend only on time differences, or
pn.y1; t1 C � I y2; t2 C � I : : : I yn; tn C �/ D pn.y1; t1I y2; t2 : : : I yn; tn/: (18)
I have chosen to formulate the random walk as starting a particle from a particular position at time
t D 0, so that x.t/ is not stationary. Alternatively we could have considered a stationary process
(e.g. the field of vision of a microscope with many Brownian particles) and then calculated the
conditional probability P2.x1; t1jx2; t2/which is the probability of the particle being at x2 at time t2
given that it was at x1 at time t1. Then P2.0; 0jx; t/ takes the diffusive form that we have calculated
and the pn all just depend on the time difference (p1.x; t/ is just constant, for example).
Means and correlation functions are defined with respect to the ensemble average. For a stationary
random process we usually assume ergodicity, and replace the ensemble average by a time average,
e.g.
hyi D y.t/ D lim
T!1
1
T
Z T=2
−T=2
y.t/dt: (19)
The probability distribution for the random walk is a Gaussian function. A Gaussian process in
general is one in which all the probability distributions are Gaussian
pn.y1; t1I y2; t2 : : : I yn; tn/ D A exp
24− nX
jD1
nX
kD1
�jk.yj − hyi/.yk − hyi/
35 (20)
where hyi is the mean of y, �jk is a positive definite matrix and A is a normalization constant. For
a stationary process hyi is time independent and � and A depend only on time differences.
Gaussian processes are important in physics because of the central limit theorem: if
Y D 1
N
X
yi (21)
3
with yi a random process or variable with arbitrary distribution but with finite mean hyi and variance
� 2y then for N large Y is a Gaussian process or variable
p.Y / D 1q
2�� 2Y
exp
"
−.Y − hY i/
2
2� 2Y
#
(22)
with hY i D hyi and �Y D �y=
p
N . The central limit theorem is why the Gaussian distribution
of the random walk is independent of the details of the step (e.g. fixed length, or varying length)
providing the mean is zero and the variance is finite.
Equation (6) for the time evolution of the probability distribution is actually the Fokker-Planck
equation for this random process. We will return to this topic in more detail later.
Spectral description of a random process
For a conventional function y.t/ a convenient definition of the Fourier transform is
Qy.f / D
Z 1
−1
y.t/ei2�f tdt; (23a)
y.t/ D
Z 1
−1
Qy.f /e−i2�f tdf: (23b)
The correctness of the inverse is shown from the resultZ 1
−1
ei2�xydx D lim
x!1
sin 2�xy
�y
D �.y/: (24)
For a real function y.t/ we have Qy�.f / D Qy.−f /.
For a stationary random process the integral defining Qy.f / diverges, so we instead define the
auxiliary process
yT .t/ D
�
y.t/ −T=2 < t < T=2
0 otherwise (25)
and then use the finite QyT .f /.
Parseval’s theorem tells us
lim
T!1
1
T
Z T=2
−T=2
[y.t/]2 D lim
T!1
2
T
Z 1
0
j QyT .f /j2 df: (26)
Here and elsewhere we use Qy�.f / D Qy.−f / to restrict the frequency domain to positive values.
With these preliminaries in mind, we define the spectral density of the random process y.t/ as
Gy.f / D lim
T!1
2
T
����Z T=2−T=2[y.t/− Ny]ei2�f tdt
����2 : (27)
where Ny is the time average over T . Why do we use this expression? Lets suppose that the mean
has been subtracted off of y, so Ny D 0: The quantity inside the j j is the Fourier transform of the
4
process yT .t/ . How does this grow with T ? We can estimate this by supposing the interval T to
be formed ofN subintervals of length � . The Fourier transform QyT is then the sum ofN transforms
Qy� of processes y� .t/defined over the interval � . For a random process we would expect each of
the Qy� to be of similar magnitude, but with arbitrary phase, since the latter depends sensitively on
the phasing of the ei2�f t with respect to the time start of the signal. Adding N complex numbers
with random phase gives a number of magnitude / pN and random phase. Thus the transform
of y.t/ − Ny grows as pT and the phase varies over all values as T changes. The spectral density
Gy.f / is constructed to be independent of T and to contain all the useful information. Parseval’s
theorem now gives us Z 1
0
Gy.f /df D lim
T!1
1
T
Z T=2
−T=2
[y.t/− Ny]2dt D � 2y ; (28)
so that the frequency integral of the spectral density is the variance of the signal.
The spectral density is directly related to the Fourier transform of the correlation function Cy.�/.
Let’s set the mean Ny to zero for simplicity. Then, using assumption of ergodicity to replace the
ensemble average by a time average, the correlation function is
Cy.�/ D lim
T!1
1
T
Z T=2
−T=2
dt y.t/y.t C �/ (29)
D lim
T!1
1
T
Z 1
−1
dt yT .t/yT .t C �/ (30)
where the small error in replacing y.t C �/ by yT .t C �/ is unimportant in the limit. Now inserting
the Fourier transforms and using Qy�.f / D Qy.−f /
Cy.�/ D lim
T!1
1
T
Z 1
−1
dt
Z 1
−1
df
Z 1
−1
df 0 QyT .f / QyT .f 0/e−i2�f 0� ei2�.fCf 0/t : (31)
The t integrations is �.f C f 0/, and using Qy�.f / D Qy.−f / gives
Cy.�/ D lim
T!1
1
T
Z 1
−1
df j QyT .f /j2 e−i2�f � (32)
D lim
T!1
2
T
Z 1
0
df j QyT .f /j2 cos 2�f � (33)
D
Z 1
0
Gy.f / cos.2�f �/df: (34)
Thus we have the inverse pair
Cy.�/ D
Z 1
0
Gy.f / cos.2�f �/df (35a)
Gy.f / D 4
Z 1
0
Cy.�/ cos.2�f �/d� (35b)
(since Cy and Gf are both even functions, we have written the results as cosine transforms only
involving the positive domain). These equations are known as the Wiener-Khintchine theorem.
5
A particularly simple spectral density is a flat one, independent of frequency. We describe such
a random process as being white. The corresponding correlation function is a delta function, i.e.
no correlations except for time differences tending to zerp. One strength parameter g is needed to
specify the force
GF.f / D g; (36a)
CF .�/ D g2 �.� /: (36b)
The Einstein Relation
Einstein showed how to relate the diffusion constant, describing the random fluctuations of the
Brownian particle, to its mobility �, the systematic response to an externally applied force.
Under an applied force −dV=dx the drift velocity of the particle is (the definition of the mobility)
ud D −�dV
dx
: (37)
For a sphere of radius a in a liquid the viscosity � the mobility is given by the Stokes expression
� D .6��aud/−1, and so � is related to the dissipation in the fluid.
Consider now the thermodynamic equilibrium of a density n.x/ of independent Brownian particles
in the potential V .x/. We can dynamically understand the equilibrium in terms of the cancelling
of the particle currents due to diffusion and mobility
−Ddn
dx
C n
�
−�dV
dx
�
D 0: (38)
Equilibrium thermodynamics on the other hand tells us n.x/ / exp[−V .x/=kT ]. Substituting into
Eq. (38) gives the Einstein identity
D D kT �: (39)
Note the use of equilibrium constraints to relate fluctuation quantities (the diffusion constant which
gives us
〈
x2.t/
�) and dissipation coefficients (� or �). This is an example of a general approach
known as fluctuation dissipation theory, that we will take up again later. The fact that the fluctuations
and dissipation of a Brownian particle are related should not be unexpected: both are a reflection
of the molecular buffeting, the dissipation given by the net force due to the systematic component
of the collisions coming from the drift of the particle relative to the equilibrium molecular velocity
distribution, and the fluctuations coming from the random component.
Fluctuation-Dissipation Theory
The relationship between the dissipation coefficient and the fluctuations is made more explicit by
directly evaluating D in terms of the fluctuations producing the random walk
D D lim
t!1
1
2t
〈[x.t/− x.0/]2� : (40)
6
Expressing the displacement as the integral of the stochastic velocity u
x.t/− x.0/ D
Z t
0
u.t1/dt1 (41)
leads to
D D lim
t!1
1
2t
Z t
0
dt1
Z t
0
dt2 hu.t1/u.t2/i ; (42)
which depends on the velocity correlation function. The integrand is symmetric in t1; t2 and we can
replace the integral over the square by twice the integral over the triangle 0 < t1 < t; t1 < t2 < t ,
and then introducing the time difference � D t2 − t1
D D lim
t!1
1
t
Z t
0
dt1
Z t−t1
0
d� hu.t1/u.t1 C �/i : (43)
Since the correlation function hu.t1/u.t1 C �/i decays to zero in some finite relaxation time �r , as
t ! 1 the limit of the second integral can be replaced by infinity for almost all values of t1 in
the first integration. Further, hu.t1/u.t1 C �/i D Cu.�/ is independent of t1 (u.t/ is a stationary
random process if external conditions are fixed). Hence
D D
Z 1
0
d� hu.0/u.� /i (44)
and
� D 1
kT
Z 1
0
d� hu.0/u.� /i (45)
directly relating a dissipation coefficient to a correlation function.
7
Physics 127b: Statistical Mechanics
Brownian Motion
Random Walk
Some general nomenclature
Spectral description of a random process
The Einstein Relation
Fluctuation-Dissipation Theory