From Brownian Motion To Euclidean Fields: 1 The Gaussian Kernel of The Diffusion Equation

From Brownian motion to Euclidean elds
April 17, 2014

1 The Gaussian kernel of the diusion equation
Each site has q neighbors. q =coordination number= 2d. At regular time
intervals, separated by t = 1, the walker jumps with uniform probability
of 1/q to a neighbour. Successive jumps are independent.
P(x
1
, t
1
; x
0
, t
0
) is the conditional probability that the walker is at site x
1
at time t
1
given that the walker is at x
0
at t
0
.
P(x
1
, t
0
; x
0
, t
0
) =
x
1
,x
0
.
Normalization at time t
1
implies
x
1
P(x
1
, t
1
; x
0
, t
0
) = 1.
Recurrence relation:
P(x, t + 1; x
0
, t
0
) =
1
2d
is a nbr of x
P(x
, t; x
0
, t
0
). (1)
Recast this equation using the discrete Laplacian
r
where
r
f(x) =
d
=1
[f(x + e
) + f(x e
) 2f(x)]
Then,
P(x, t + 1; x
0
, t
0
) P(x, t; x
0
, t
0
) =
r
P(x, t; x
0
, t
0
) (2)
Fourier transform!, in space. Denition: P(x, t; x
0
, t
0
) =
d
d
k
(2)
d
e
ik.x

P(k, t).
This yields:
P(k, t + 1) =
1
d
d
=1
cos(k
)

P(k, t). (3)
and
P(k, t
0
) = e
ik.x
0
1
Clearly, we have
P(x
1
, t
1
; x
0
, t
0
) =
d
d
k
(2)
d
e
ik.(x
1
x
0
)
(
1
d
cos(k
))
t
1
t
0
(4)
We are interested in its asymptotic behavior. We now rescale time and
space intervals from 1 to and a respectively. This means, t t/,
k ak and x x/a.
P(x
1
x
0
, t
1
t
0
) = a
d
/a
/a
d
d
k
(2)
d
e
ik.(x
1
x
0
)
(
1
d
cos(ak
))
(t
1
t
0
)/
Dividing by the volume element a
d
gives us a probability density.
p(x
1
x
0
, t
1
t
0
) = lim
a,0
/a
/a
d
d
k
(2)
d
e
ik.(x
1
x
0
)
(
1
d
cos(ak
))
(t
1
t
0
)/
Claim: This limit is non-trivial only when a and vanish such that

a
2
is
kept xed. Hence, WNLOG, =
a
2
2d
Expanding the cosine, we get
p(x
1
x
0
, t
1
t
0
) = lim
a,0
/a
/a
e
ik.(x
1
x
0
)
(1
a
2
k
2
2d
+ . . .)
(t
1
t
0
)/
Hence we have,
p(x
1
x
0
, t
1
t
0
) =
e
(t
1
t
0
)k
2
+ik.(x
1
x
0
)
=
1
[4(t
1
t
0
)]
d/2
e
(x
1
x
0
)
2
4(t
1
t
0
)
This nal answer is the well-known kernel of the diusion equation in
continuous space. It satises:
d
d
xp(x, t; x
0
, t
0
) = 1
lim
t
1
t
0
p(x
1
, t
1
; x
0
, t
0
) =
d
(x
1
x
0
)
For t
2
> t
1
> t
0
:
d
d
x
1
p(x
2
, t
2
; x
1
, t
1
)p(x
1
, t
1
; x
0
, t
0
) = p(x
2
, t
2
; x
0
, t
0
)
2
2 Greens function
Consider the mean time spent on a given site:
G(x
1
x
0
) =
t
1
=t
0
P(x
1
, t
1
; x
0
, t
0
) =
d
d
k
(2)
d
e
ik.(x
1
x
0
)
1
1
1
d
cos(k
)
The equation satised by P(x
1
, t
1
; x
0
, t
0
) imples the following for G(x
1
x
0
):
G(x
1
x
0
) =
x
1
,x
0
+
1
2d
G(x
1
+ e
x
0
) + G(x
1
e
x
0
)
In the notation of the discrete Laplacian, this is:
r
G(x
1
x
0
) =
x
1
,x
0
and he nce it is the Greens function of the discrete Laplacian
r
.
When k is small, the integrand for G(x
1
x
0
) behaves has
1
k
2
and hence,
there is an infrared divergence for d = 1 and d = 2. This means that the
mean time spent is . We will see that this implies that the return
probability is 1, for d 2.
We can instead consider the mean time spent at x
1
in excess of the time
spent at x
0
. This however is not a positive quantity unlike G.
G
s
(x
1
x
0
) = G(x
1
x
0
) G(0)
=
d
d
k
(2)
d
cos(k.(x
1
x
0
)) 1
1
1
d
cos(k
)
For d = 1, G
s
(x) = |x|.
We can dene the following generating function:
G(x
1
x
0
, ) =
t
1
t
1
t
0
P(x
1
, t
1
; x
0
, t
0
)
which is convergent always for || < 1.
G(x
1
x
0
, ) =
d
d
k
(2)
d
e
ik.(x
1
x
0
)
1
d
cos(k
)
One can derive the following equation:
G(x, ) =
t0
t
P(x, t; 0, 0)
=
t>0
[
t
1
2d
P(x+e
, t 1; 0, 0) +P(xe
, t 1; 0, 0)] +P(x, 0; 0, 0)
3
=
2
t0
t
1
2d
P(x + e
, t; 0, 0) + P(x e
, t; 0, 0) +
x,0
Subtracting G(x, ) from the above, we get,
[1 ]G(x, ) =
r
G(x, ) +
x,0
Simplifying to:
[
r
+ (
1
1)]G(x, ) =
x,0
This is the analog of the massive propagator in particle physics.
Continuum limit of the massive propagator: Take x x/a, k ka and
1/ = 1 + m
2
a
2
/2d, in the limit a 0.
g(x, m
2
) = lim
a0
1
2da
d2
G(
x
a
, ) =
d
d
k
(2)
d
e
ik.x
k
2
+ m
2
3 Aside: Polyas random walk constants
Let r(d) be the return probability of a random walk on a simple cubic
lattice in d dimensions. It is easy to show that:
r(d) = 1 G
1
d
(0)
where G
d
is the greens function dened above, in d dimensions.
r(1) = r(2) = 1 since G
1
(0) = G
2
(0) = .
See wolfram mathworld article on Polyas random walk constants and its
references.
4 Path integral formula for p(x, t) and g(x, m)
We rst derive a formula for g(x, m) in terms of p(x, t) analogous to the
formula for G(x, ) in terms of P(x
1
, t
1
; x
0
, t
0
).
g(x, m
2
) = lim
a0
1
2da
d2
G(
x
a
, = (1 +
m
2
a
2
2d
)
1
)
= lim
a0
a
2
2d
t/=0
(1 +
m
2
a
2
2d
)
t/
1
a
d
P(x/a, t/; 0, 0)
=

0
e
m
2
t
p(x, t)
(where we have used
a
2
2d
= 1.)
4
We now use the the convolution property of p(x
1
, t
1
; x
0
, t
0
) to write a path
integral formula for it. And then use the above formula connecting g and
p to write a pathe integral formula for g(x, m
2
).
p(x
f
, t
f
; x
i
, t
i
) =
n1
j=1
d
d
x
j
[4(t
j+1
t
j
)]
d/2
1
[4(t
1
t
0
)]
d/2
e
1
4
n1
j=0
(x
j+1
x
j
)
2
(t
j+1
t
j
)
where x
0
= x
i
and x
n
= x
f
.
We can now write this with the following symbolic formula:
=
x(t
i
)=x
i
andx(t
f
)=x
f
Dx(t) exp[
1
4
dt x
2
]
5

From Brownian Motion To Euclidean Fields: 1 The Gaussian Kernel of The Diffusion Equation

Uploaded by

From Brownian Motion To Euclidean Fields: 1 The Gaussian Kernel of The Diffusion Equation

Uploaded by

From Brownian motion to Euclidean elds

April 17, 2014

You might also like