Path Integrals and Fokker-Planck Equations
Justin Bois
California Institute of Technology
Chemical Engineering Option
26 September, 2003
Abstract
This report briey outlines how Fokker-Planck equations may be derived using the path
integral formalism. The path integral in quantum mechanics is introduced and a connection
to statistical physics is drawn. In particular, the analogy between the axiomic properties of
the Greens function
1
to the Chapman-Kolmogorov equation is discussed. Finally, we derive
a Fokker-Planck equation governing the conformation of a semiexible polymer with inherent
bending as a nontrivial expose of the utility of the path integral formalism in deriving Fokker-
Planck equations.
1 Introduction
Fokker-Planck equations are important stochastic PDEs for describing a large number of physical
processes. They describe how the probability distribution P (x; t) describing the state of a system
evolves through time and a set of variables in phase space, x. A Fokker-Planck equation has the
form
t
P (x; t) =
x
A(x) P (x; t) +
1
2
x
B (x) P (x; t) . (1)
There are countless ways of deriving Fokker-Planck equations for various processes. One com-
mon approach is to derive a Fokker-Planck equation corresponding to a given Langevin equation
by averaging over the noise (assumed to be white and Gaussian distributed) and constructing a
Taylor expansion in powers of the noise [4]. Another approach is to expand a master equation
in a Kramers-Moyal expansion and truncate it to two terms
2
[5]. One may also directly use the
Markovian nature of a process to rst derive a Chapman-Kolmogorov equation and then perform
a Taylor expansion about small time dierences
3
[6].
While the aforementioned ways of arriving at the Fokker-Planck equation are useful, the path
integral formalism is particularly pedagogical. In the words of Richard Feynman, It is usually true
that the path integral method does not really help to get the solution to problems that cannot be
solved exactly in some other manner. Nevertheless, someone . . . familiar with path integrals will
1
We use the terminology Greens function as in [1]. This function is also called a kernel, as in [2], or a
propagator, as in [3].
2
This method also assumes that the transition rates are such that only small transitions occur and that the
probability distribution varies slowly in space.
3
This derivation is a bit more complicated than described here. See [6] for a more complete description.
1
nd its mode of expression and logic very simple and direct when applied to probability problems.
[2] Furthermore, this report reveals interesting parallels between quantum and Brownian systems.
The rst section of this report gives a brief introduction to the path integral formulation of
quantum mechanics. From the Feynman postulates, the time-dependent Schr odinger equation
is derived. The next section discusses the parallel between the development of the Schr odinger
equation from path integrals and the derivation of a Fokker-Planck equation from the Chapman-
Kolmogorov equation. Finally, we present the derivation of the Fokker-Planck equation governing
the Greens function for a wormlike chain (probability distribution function describing the position
and orientations of the end of the polymer).
2 Introduction to path integrals in quantum mechanics
Historically the quantum mechanics have been developed from dierent viewpoints, including the
Schr odinger and Heisenberg treatments, but rely on the same fundamental postulates. Of course,
the postulates have dierent forms for dierent treatments, but they are all identical. In this section
of the report, we will rst briey state the postulates for a single particle from the path integral
viewpoint, often called Feynmans postulates. Given the postulates, we will then derive the
Schr odinger equation from the path integrals. This derivation is particularly important because it
follows a very similar prescription as the derivation of Fokker-Planck equations from path integrals.
As a bibliographical note, the developments that follow appear in a number of texts, and are
therefore uncited. In our presentation, we follow the development of Feynman and Hibbs [2], though
the notes of Schulten [7] are a particularly useful companion.
2.1 Quantum mechanical postulates in the language of path integrals
Postulate 1 A particle is described by a wave function (r, t) where the probability of nding
the particle at position r at time t is given by | (r, t)|
2
. The wave function is normalized
such that
_
dr | (r, t)|
2
= 1. (2)
This means it must be somewhere in space at any given time.
Postulate 2 The Greens function, G(r, t |r
, t
) is dened to describe the time evolution of the
particle from position r
(t
) at time t
to postion r (t) at time t by
(r, t) =
_
dr
G
_
r, t
, t
_
r
, t
_
. (3)
Postulate 3 The Greens function is given by
G
_
r, t
, t
_
=
r
_
r
Dr exp
_
i
h
S
_
r (t)
_
t
_
_
, (4)
where the functional S [r (t) |r
(t
)] is the classical action integral, given by
S
_
r (t)
_
t
_
=
t
_
t
dt
0
L( r, r, t
0
) , (5)
2
with L( r, r, t) being the classical Lagrangian and t
0
being a dummy integration variable. The
symbol
_
r
r
Dr denotes a path integral, meaning that we integrate over all paths that begin at
r
(t
) and end at r (t). The appropriate normalization constants are absorbed in the symbol.
A couple of important properties about the Greens function may be derived from the postulates,
namely that the Greens function obeys the superposition principle and is unitary. We proceed by
deriving these two properties, starting with the superposition principle.
2.1.1 The superposition principle
Consider three time points, t
< t
< t. Then, according to equation 5 the action along and path
between r
(t
) and r (t) can be written as
S
_
r (t)
_
t
_
=
t
_
t
dt
0
L( r, r, t
0
) =
t
_
t
dt
0
L( r, r, t
0
) +
t
_
t
dt
0
L( r, r, t
0
)
= S
_
r
_
t
_
t
_
+ S
_
r (t)
_
t
_
. (6)
So, by equation 4,
G
_
r, t
, t
_
=
r
_
r
Dr exp
_
i
h
S
_
r
_
t
_
t
_
_
exp
_
i
h
S
_
r (t)
_
t
_
_
, (7)
We can perform the integral in equation 7 by splitting the path into two parts. We rst integrate
from r
(t
) to r
(t
). The result is simply G(r
, t
|r
, t
) by equation 4. However, we must also
integrate over all intermediate positions r
to consider all paths. Thus, we have
G
_
r, t
, t
_
=
_
dr
G
_
r
, t
, t
_
_
Dr exp
_
i
h
S
_
r (t)
_
t
_
_
(8)
We see immediately that the path integral in equation 8 is G(r, t |r
, t
) by equation 4. Thus, we
have arrived at the superposition principle (also called the completeness relationship):
G
_
r, t
, t
_
=
_
dr
G
_
r, t
, t
_
G
_
r
, t
, t
_
(9)
This means that to fully determine the state of a system, (r, t), at time t given that we knew the
state of the system to be (r
, t
) at time t
, we must know the state of the system at all points r
for some time t
with t
< t
< t. One should notice that the superposition principle is analogous
to the Chapman-Kolmogorov equation.
The superposition principle is easily generalized for further divisions of the path from r
to r by
repeated separations of path integrals as we have just done. Consider breaking the path up into
N subpaths that terminate at positions r
1
, r
2
, . . . , r
N
, with corresponding time intervals. We then
obtain
G
_
r, t
, t
_
=
_
dr
1
_
dr
2
_
dr
N1
G(r, t |r
N1
, t
N1
) G(r
N1
, t
N1
|r
N2
, t
N2
)
G(r
i+1
, t
i+1
|r
i
, t
i
) G
_
r
1
, t
1
, t
_
. (10)
3
As N gets large in equation 10, the time step between points on the segregated path gets small.
We dene a small time step as t
i+1
t
i
. For small , the action, according to equation 5 is
given by
S [r
i+1
, t
i+1
|r
i
, t
i
] =
t
i+1
_
t
i
dt
0
L( r, r, t
0
) L
_
r
i+1
r
i
,
r
i+1
+r
i
2
,
t
i+1
+ t
i
2
_
. (11)
Thus, for two points separated by a very small time dierence,
G(r
i+1
, t
i+1
|r
i
, t
i
) =
1
A
exp
_
i
h
S [r
i+1
, t
i+1
|r
i
, t
i
]
_
=
1
A
exp
_
i
h
L
_
r
i+1
r
i
,
r
i+1
+r
i
2
,
t
i+1
+ t
i
2
__
, (12)
where A is the normalization constant. In substituting equation 12 into equation 10 and taking
the appropriate limits, we recover the postulate given by equation 4.
G
_
r, t
, t
_
= lim
N
0
_
dr
1
A
_
dr
2
A
_
dr
N1
A
exp
_
i
h
N1
i=1
S [r
i+1
, t
i+1
|r
i
, t
i
]
_
= lim
N
0
_
dr
1
A
_
dr
2
A
_
dr
N1
A
exp
_
i
h
S
_
r, t
, t
_
=
r
_
r
Dr exp
_
i
h
S
_
r (t)
_
t
_
_
. (13)
2.1.2 Unitarity of the Greens function
We now prove that the Greens function is unitary. Using equations 2 and 3, we see that
_
dr | (r, t)|
2
=
_
dr
_
dr
G
_
r, t
, t
_
r
, t
_
_
dr
_
r, t
, t
_
r
, t
_
=
_
dr
_
r, t
2
= 1 (14)
Equation 14 must be true for any arbitrary , so
_
dr G
_
r, t
, t
_
G
_
r, t
, t
_
=
_
r
_
. (15)
Equation 15 means that the Greens function is unitary.
2.2 Derivation of the time-dependent Schr odinger equation
Although not immediately obvious from the stated postulates in the previous section, we may
derive dierential equations describing the wave function and the Greens function from the inte-
gral equations given in the postulates. The result, as we will see, is the famous time-dependent
4
Schr odinger equation for the wave function. We will also see that the Greens function is indeed a
Greens function for the Schr odinger equation, which is why is has its name.
We begin be dening the Lagrangian for a single particle quantum system.
L =
T
V =
m
2
r
2
V (r, t) , (16)
where
T and
V are the kinetic energy and potential energy operators, respectively, and m is the
mass of the particle. We will consider a very short path going from r (t
) r
to r (t) r with
an intermediate point in the path, r (t
) r
. We dene t
t as a small time step. Using
equations 9 and 12 and the Lagrangian given by equation 16, we obtain, for small ,
G
_
r, t
, t
_
=
1
A
_
dr
exp
_
i
h
_
m(r r
)
2
2
2
V
_
r +r
2
, t
_
__
G
_
r
, t
, t
_
(17)
We see that when r
is varies signicantly from r the quantity (r r
)
2
/ becomes very large.
As a consequence, the exponential oscillates rapidly as r
varies. This means that the integral is
very small in this regime. Therefore, the only appreciable contribution to the integral occurs when
(r r
)
2
is small. It is then logical to substitute b = r
r and consider small b. Making this
substitution and splitting the exponential for clarity, we get
G
_
r, t
, t
_
=
1
A
_
db exp
_
im
2 h
b
2
_
exp
_
i
h
V
_
r +
b
2
, t
__
G
_
r +b, t
, t
_
(18)
We notice that the argument of the rst exponential is of order unity (and therefore making the
major contribution to the integral before the rapid oscillations begin) when b
2
is of order O().
Thus, to derive an expression that is accurate to rst order in , we expand G in the right hand side
to second order in b. Performing this expansion and also on the potential energy operator yields
G
_
r, t
, t
_
=
1
A
_
db exp
_
im
2 h
b
2
_
_
1
i
h
V
_
r, t
_
__
1 +b +
1
2
bb :
_
G
_
r, t
, t
_
(19)
To determine the normalization constant, we note that equation 19 must hold to zero order in
, which means
1
A
_
db exp
_
im
2 h
b
2
_
=
1
A
db
x
exp
_
im
2 h
b
2
x
_
_
db
y
exp
_
im
2 h
b
2
y
_
_
db
z
exp
_
im
2 h
b
2
z
_
=
1
A
_
_
du exp
_
im
2 h
u
2
_
_
_
3
=
1
A
_
i2 h
m
_3
2
= 1
A =
_
i2 h
m
_3
2
. (20)
In the calculation of A, we used the fact that
I
0
(a)
dx exp
_
iax
2
_
=
_
i
a
, (21)
5
which can be found from contour integration in the complex plane. In fact, this integral belongs
to a larger family of integrals which will be useful in subsequent calculations (and many quantum
mechanical calculations for that matter), so we present them here. We dene the integral I
n
as
I
n
(a) =
dx x
n
exp
_
iax
2
_
, (22)
where a is real and positive. We see that for n = 1, the integrand is odd, so the integral I
1
vanishes.
In fact, all I
n
with odd n are zero. We also see that we can relate I
n+2
to I
n
by dierentiating with
respect to a. Thus, we have
I
0
(a) =
_
i
a
, I
1
(a) = 0;
I
n+2
(a) =
1
i
a
I
n
(a) . (23)
Given that we now have the normalization constant, A, and have evaluated the integral associ-
ated with it, we need only to evaluate integrals of the type in equation 23 with a = m/ (2 h). In
doing so, we get
1
A
_
db exp
_
im
2 h
b
2
_
= 1
1
A
_
db bexp
_
im
2 h
b
2
_
= 0, (24)
and
1
A
_
db bbexp
_
im
2 h
b
2
_
=
i h
m
I, (25)
where I is the identity matrix. Plugging the results of equations 20 and 24 through 25 into equation
19 yields
G
_
r, t
, t
_
=
_
1 +
i h
2m
i
h
V (r, t) +
2
2m
V (r, t)
2
_
G
_
r, t
, t
_
(26)
Multiplying both sides of equation 26 by i h/, noting that t = t
+ , and rearranging we get, to
order O(),
i h
G(r, t
+ |r
, t
) G(r, t
|r
, t
=
_
h
2
2m
2
+ V
_
r, t
_
_
G
_
r, t
, t
_
. (27)
We notice that for small , the left-hand side of equation 27 is i h/t
. Thus, upon renaming the
variable t
to t and rearranging, we have
_
i h
t
+
h
2
2m
2
V (r, t)
_
G
_
r, t
, t
_
= 0 (28)
We dene the Greens function to be zero for all time t < 0, which trivially satises the PDE of
equation 28. To establish the boundary condition, we note that as t
t, the Greens function
6
goes to (r r
). This is dictated by equation 3 in the second postulate. Thus, we arrive at the
complete PDE for the Greens function
_
i h
t
+
h
2
2m
2
V (r, t)
_
G
_
r, t
, t
_
=
_
r r
_
t t
_
(29)
Given that equations 3 and 29 hold, (r, t) must satisfy the homogeneous version of the PDE
governing the Greens function. Thus, we arrive at
i h
t
(r, t) =
H (r, t) , (30)
with
H =
T +
V =
h
2
2m
2
+ V (r, t) . (31)
This result is the famous time-dependent Schr odinger equation.
3 From quantum to statistical mechanics
The ultimate goal of this report is to show how the methods of quantum mechanics can be applied
to derive Fokker-Planck equations for statistical systems. Thus far, we have been very specically
describing the behavior of quantum particles. Feynman and Hibbs [2] provide an excellent discussion
on application of path integrals to systems described by quantum statistical mechanics and the
reader is encouraged to read this (it is chapter 10 in the book), but our aim is to describe systems
of larger scale. We will do this by simply dening a path for a system and then noting the
analogies to the quantum system
4
.
Consider a system going from state x
(t
) at time t
to state x(t)
5
at time t. The system can go
from state x
to x along any of a number of paths, so each path has associated with it a probability,
or statistical weight [x, t
0
]. The statistical weight of a given path is proportional to the Boltzmann
weighted energy associated with that path, or
[x, t
0
] exp
_
1
kT
_
t
t
dt
0
H(x, t
0
)
_
, (32)
where H is the Hamiltonian per dierential time. Then, the transition probability of the system
going from state x
to state x is the Boltzmann-weighted sum over all paths going from x
to x.
This is given by
G
_
x, t
, t
_
=
x
_
x
Dx exp
_
_
_
1
kT
t
_
t
dt
0
H(x, t
0
)
_
_
_
, (33)
where we have again included normalization constants into the symbol
_
x(t)
x
(t
)
Dx.
4
We will only be dealing with conserved Hamiltonians in this simplistic look at path integrals in statistical
mechanics. For dissipative systems, path integrals still work, but properly dening the action requires a bit more
rigorous treatment. Chapter 10 in Feynman and Hibbs [2] gives a good introduction to such methods.
5
The variable x describes the state of the system in phase space.
7
u s
R
u
u
s
R
R
s
Figure 1: A representation of the Kratky-Porod wormlike chain. The vector R(s) represents the position of the
chain in space as a function of the path length, s. The unit vector u(s) is the unit tangent to the chain at as a
function of s, thereby representing its orientation.
We immediately see analogies to the third quantum mechanical postulate given on page 2. We
see that the transition probability, or Greens function, has precisely the same form at the quantum
mechanical Greens function given by equation 4 except i/ h is replaced by 1/kT. Similarly, the
action integral in equation 5 becomes the energy associated with the path. In summary, for the
statistical treatment, we have
G
_
x, t
, t
_
=
x
_
x
Dr exp
_
1
kT
H
_
x, t
, t
_
, (34)
where H[x, t |x
, t
] is the energy associated with the path.
Given these denitions, the superposition principle also applies to the statistical Greens func-
tion as it did to the quantum Greens function. Furthermore, if the Greens function is a function
of many variables, say x
1
, x
2
, . . ., x
N
, and t, the superposition principle is generalized as
G
_
x
1
, . . . , x
N
, t
1
, . . . , x
N
, t
_
=
_
dx
1
. . .
_
dx
N
G
_
x
1
, . . . , x
N
, t
1
, . . . , x
N
, t
_
G
_
x
1
, . . . , x
N
, t
1
, . . . , x
N
, t
_
.(35)
We see that the superposition principle is entirely analogous to the Champman-Kolmogorov equa-
tion. Indeed, the path integrals tell us explicitly how the probability distribution function is
propagated through time.
4 An example calculation: the Greens function for a wormlike
chain
We will now consider the Kratky-Porod wormlike chain (WLC), the simplest model for a polymer
that captures its inherent bending rigidity. Despite its simplicity, it is a very powerful model.
8
Figure 1 shows a drawing of a WLC. Notice that the vectors R(s) and u(s) represent the
position and orientation, respectively, of the chain as a function of path length, s. In our analysis,
the path length is analogous to time. Note also that u is given by
u
R
s
(36)
with
|u| = 1. (37)
The stipulation given by equation 37 that u be a unit vector comes from the denition of curvature
from elementary dierential geometry and is a very important constraint in our analysis.
Treating the WLC as a dierentiable space curve and following the development in [8] which
draws from elasticity theory, the Hamiltonian for a dierential section of the chain is given by
H =
2
_
u
s
_
2
+ U
e
, (38)
where is the bending modulus
6
and U
e
is an external eld. For a WLC, the bending modulus is
related the the persistent length,
p
, by
=
p
kT =
kT
2
where
p
1
2
. (39)
In such a way, is a measure of how the thermal energy compares to the bending energy.
The Greens function of a WLC, G(R, u, s |R
, u
, s
), is the probability that the chain has
position R and orientation u at path length s, given that it has at position R
with orientation
u
at path length s
. The Greens function contains all the thermodynamic information about the
chain [8], since it represents a statistical weight, or partition function, of a chain with the prescribed
starting and ending points.
Corresponding with equations 38 and 39, the potential energy for a section of rod going from
path length s
to s with bending modulus (which may be a function of the path length) and
subject to external eld U
e
is given by
U =
kT
4
s
2
_
s
1
ds
_
u
s
_
2
+
s
2
_
s
1
ds U
e
. (40)
The conformational distribution function is then simply the Boltzmann weighting of this potential.
In dening the Greens function, we must include a delta function forcing the denition RR
=
s
_
s
ds u. Thus, we have the Greens function given by
G
_
R, u, s
, u
, s
_
=
_
DR
_
Du
_
_
RR
s
_
s
ds
0
u
_
_
exp
_
_
_
1
4
s
_
s
ds
0
_
u
s
1
_
2
_
_
_
exp
_
_
_
1
kT
s
_
s
ds
0
U
e
_
_
_
. (41)
6
This is not the same that is a smallness parameter in our expansions; is is the bending modulus of a rod
from elasticity theory. It is an unfortunate coincidence that the same symbol is traditionally used to describe both
quantities. So as not to confuse these two symbols, we will work with the persistent length and not the bending
modulus for the rest of the discussion.
9
The superposition principle is written as
G
_
R, u, s
, u
, s
_
=
_
dR
_
du
G
_
R, u, s
, u
, s
_
G
_
R
, u
, s
, u
, s
_
. (42)
We proceed in the same was as in the derivation of the Schr odinger equation. We rst note that
for small s s
,
s
_
s
ds
0
u u,
s
_
s
ds
0
U
e
U
e
, (43)
and
s
_
s
ds
0
_
u
s
0
_
2
_
u u
_
2
=
u
+u u 2u
= 2
_
1 u
_
. (44)
We perform an expansion on the external eld term and get, for small ,
G
_
R, u, s
, u
, s
_
=
1
A
exp
_
1
2
u
__
1
U
e
kT
_
, (45)
where a factor of exp
_
(2s)
1
_
has been absorbed into the normalization constant A.
We now consider the other Greens function in the integral, G(R
, u
, s
|R
, u
, s
). We dene
b R
R and v u
u such that
G
_
R
, u
, s
, u
, s
_
= G
_
R+b, u +v, s
, u
, s
_
. (46)
Expanding this Green function as a Taylor series to second order about b and v yields
G
_
R
, u
, s
, u
, s
_
= LG
_
R, u, s
, u
, s
_
(47)
where
L = 1 +v
u
+b
R
+
1
2
vv :
u
u
+
1
2
bb :
R
R
+bv :
R
u
(48)
Combining equations 42, 45, 47, and 48 yields
G
_
R, u, s
, u
, s
_
=
1
A
_
1
U
e
kT
__
du
exp
_
1
2
u
_
LG
_
R, u, s
, u
, s
_
. (49)
As we evaluate the integrals in equation 49, it is useful to make some denitions to keep the
notation concise. We dene a (2)
1
u with a |a| = (2)
1
. Further, we dene the integral
operator A operating on some function f as A[f] =
1
A
_
du
exp(u
a)f. The value of A is found
by noting that in the limit of going to zero, the leading term in the Taylor expansion must go to
zero, i.e.
1
A
_
du
exp
_
u
a
_
= A[1] = 1. (50)
10
Upon dening a axis system where the z-axis points along u (and therefore along a) and noting
that |a u
| = a, evaluation of the integral yields
A =
_
du
exp
_
u
a
_
=
2
_
0
d
_
0
d sin exp{a cos } =
4
a
sinha. (51)
To complete our derivation, we have only to calculate all the quantities A[l] where l is one of
the terms in L as dened by equation 48. First, we calculate A[v].
A[v] = A
_
u
= A
_
u
u =
1
A
_
du
exp
_
u
a
_
u =
1
A
c
a
u
=
4
A
u
a
sinha
a
u = u
_
a
sinha
_
a cosha sinha
a
2
_
1
_
= u
_
cotha
1
a
1
_
.(52)
Given that the quantity cotha 1 goes to zero faster than 1/a as a gets large (or gets small), we
get
A[v] =
u
a
= 2u. (53)
Next, noting that b u for small , we can calculate
A[b] = A[u] = A[1] u = u. (54)
Now we calculate A[vv].
A[vv] = A
__
u u
_ _
u u
_
= uuA[1] 2uA
_
u
+A
_
u
=
2uu
a
uu +A
_
u
. (55)
We perform the calculation of the last term in equation 55 in the limit of large a.
A
_
u
=
1
A
_
du
exp
_
u
a
_
=
1
A
a
_
du
exp
_
u
a
_
=
1
A
a
c
=
4
A
__
cosha
a
2
sinha
a
3
_
I +
_
sinha
a
3
3 cosha
a
4
+
3 sinha
a
5
_
aa
_
=
_
cotha
a
1
a
2
_
I +
_
1
3 cotha
a
+
3
a
2
_
uu (56)
We now note that for large a,
cotha
a
1
a
2
1
a
(57)
and
1
3 cotha
a
+
3
a
2
1
3
a
. (58)
Substituting these values into equation 56 and then using equation 55, we get
A[vv] =
1
a
(I uu) = 2 (I uu) . (59)
11
We can conveniently use previous results to compute A[bb] and A[bv] in the limit of large a.
A[bb] =
2
A[uu] =
2
A[1] uu =
2
uu, (60)
and
A[bv] = A[uv] = A
_
u
_
u u
_
= A[uu] A
_
uu
= uu uA
_
u
= uu uucotha +
uu
a
= 2
2
uu. (61)
Here, we have noted that for large a, cotha 1.
We substitute the results from equations 53, 54, 59, 60, and 61 into equation 49 to get
G
_
R, u, s
, u
, s
_
=
_
1
U
e
kT
_
_
1 2u
u
u
R
+
2
u
+
2
2
uu :
R
R
+ 2
2
uu :
R
u
_
G
_
R, u, s
, u
, s
_
(62)
Noting that u
u
= 0 and uu :
u
u
= 0, neglecting terms of order O
_
2
_
, substituting s s
+,
and rearranging we get
G(R,u,s
+|R
,u
;s
)G(R,u,s
|R
,u
,s
=
_
u
R
+
2
u
Ue
kT
_
G(R, u, s
|R
, u
, s
) (63)
We note that in the limit of small the left hand side of equation 63 is simply the partial derivative
of G(R, u, s
|R
, u
, s
) with respect to s
. Therefore, upon renaming the variable s
to s, we get
_
s
+u
R
2
u
+
U
e
kT
_
G
_
R, u, s
, u
, s
_
= 0. (64)
We nish the derivation by adding delta function forcing to the PDE to enforce the boundary
conditions. Thus, we arrive at the governing PDE for a WLC.
_
s
+u
R
2
u
+
U
e
kT
_
G
_
R, u, s
, u
, s
_
=
_
RR
_
u u
_
s s
_
. (65)
5 Conclusions
We have seen that path integrals are a powerful tool in deriving Fokker-Planck equations in sta-
tistical mechanics. The development is analogous to that of the derivation of the time-dependent
Schr odinger equation from the Feynman postulates. The path integral picture is particularly in-
sightful because it describes explicitly how the probability distribution propagates in time. This
report serves as a basic introduction; the utility of the path integral formalism extends deeply
throughout quantum and statistical physics.
References
1. Giorgio Parisi. Statistical Field Theory. Addison-Wesley Publishing Company, Inc., 1988.
2. Richard P. Feynman and Albert R. Hibbs. Quantum Mechanics and Path Integrals. McGraw-Hill, 1965.
12
3. Claude Cohen-Tannoudji, Bernard Diu, and Franck Laloe. Quantum Mechanics, volume 1. John Wiley and Sons,
1977.
4. Robert Zwanzig. Nonequilibrium Statistical Mechanics. Oxford University Press, 2001.
5. N.G. van Kampen. Stochastic Processes in Physics and Chemistry. Elsevier Science B.V., 4th edition, 2003.
6. John F. Brady. Introduction to suspension dynamics. 1998 DRAFT, used as course notes, Caltech, 2003.
7. Klaus Schulten. Notes on quantum mechanics. [Link] PDF/QM [Link],
2000.
8. Masao Doi and Sam F. Edwards. The Theory of Polymer Dynamics. Oxford University Press, 1986.
13