MAE 545: Lecture 17 (4/10)

Random walks
 Random walks
Brownian motion Swimming of E. coli
J. Phys. A: Math. Theor. 42 (2009) 434013

(A) (B)

..ATTATGCATGACGAT..

Figure 1. (A) Schematic representation of the protein–DNA search problem. The pr

must find its target site (red) on a long DNA molecule confined within the cell nucleo
Polymer random coils Protein search for a
or cell nucleus (in eukaryotes). Compare with figure 9(A) which shows confined
target site must be recognized with 1 base-pair (0.34 nm) precision, as displacement
in a different sequence and consequently a different site.
binding site on DNA
(A) 1d 3d (B)

Figure 2. (A) The mechanism of facilitated diffusion. The search process consists
rounds of 3D and 1D diffusion, each with average duration τ3D and τ1D , respecti
2 antenna effect [9]. During 1D diffusion (sliding) along DNA, a protein visits on a
 Brownian motion of small particles
1827 Robert Brown: observed irregular motion of
small pollen grains suspended in water

⇡ 10µm

https://www.youtube.com/watch?v=R5t-oA796to

1905-06 Albert Einstein, Marian Smoluchowski:

microscopic description of Brownian motion and
relation to diffusion equation

3
 Random walk on a 1D lattice
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
At each step particle jumps to the right with

probability q and to the left with probability 1-q.
sample trajectories for q=1/2 sample trajectories for q=2/3
(unbiased random walk) (biased random walk)
100 100

80 80

60 60
N

40 N 40

20 20

0 0
-20 -10 0 10 20 -50 -25 0 25 50
x/ℓ 4 x/ℓ
 Random walk on a 1D lattice
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
At each step particle jumps to the right with

probability q and to the left with probability 1-q.

What is the probability p(x,N) that we find

particle at position x after N jumps?

Probability that particle makes k ✓ ◆

N k
jumps to the right and N-k jumps to p(k, N ) = q (1 q)N k
k
the left obeys the binomial distribution

x = k` (N k)` = (2k N )`
Relation between k and
 ⇣ ⌘
particle position x: 1 x
k= N+
2 `
5
 Random walk on a 1D lattice
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
unbiased random walk biased random walk
✓ ◆
N k
p(k, N ) = q (1 q)N k
k
1⇣ x⌘
k= N+
2 `
Note: exact discrete
distribution has been made
continuous by replacing
discrete peaks with boxes
whose area corresponds to
the same probability.

after several steps the

probability distribution
spreads out and becomes
approximately Gaussian

6
 Gaussian approximation for p(x,N)
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
N
X
Position x after N jumps can be expressed
as the sum of individual jumps xi 2 { `, `}. x= xi
N i=1
Mean value averaged over X
hxi = hxi i = N hx1 i = N (q` (1 q)`)
all possible random walks
i=1

hxi = N ` (2q 1)
⌦ ↵ ⇣⌦ ↵ ⌘
2 2 2 2 2 2
Variance averaged over all = x hxi = N = N x1 1 hx1 i
possible random walks ⇣ ⌘
2 2 2 2
= N q` + (1 q)` hx1 i
2
= 4N `2 q(1 q)
According to the central limit 1 (x hxi)2 /(2 2
)
theorem p(x,N) approaches p(x, N ) ⇡ p e
2⇡ 2
Gaussian distribution for large N: 7
 Number of distinct sites visited
by unbiased random walks
Total number of sites inside
explored region after N steps
p
1D Ntot / N In 1D and 2D every
site gets visited after
a long time
2D Ntot / N

p In 3D some sites are

p 3D Ntot / N N never visited even
2r / N after a very long time!

Shizuo Kakutani: “A drunk man will find his way

home, but a drunk bird may get lost forever.”
p
1D Nvis ⇡ 8N/⇡
Number of distinct visited
2D Nvis ⇡ ⇡N/ ln(8N )
sites after N steps
8
3D Nvis ⇡ 0.66N
 Master equation
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
Master equation provides recursive relation for the
evolution of probability distribution, where ⇧(x, y)
describes probability for a jump from y to x.
X
p(x, N + 1) = ⇧(x, y)p(y, N )
y
For our example the master equation reads:
p(x, N + 1) = q p(x `, N ) + (1 q) p(x + `, N )
Initial condition: p(x, 0) = (x)

Probability distribution p(x,N) can be easily obtained

numerically by iteratively advancing the master equation.
9
 Master equation and Fokker-Planck equation
1 q q

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
Assume that jumps occur in regular small time intervals: t
Master equation:
p(x, t + t) = q p(x `, t) + (1 q) p(x + `, t)
In the limit of small jumps and small time intervals, we can Taylor expand
the master equation to derive an approximate drift-diffusion equation:
✓ 2
◆ ✓ 2
◆
@p @p 1 2 @ p @p 1 2 @ p
p+ t =q p ` + ` + (1 q) p + ` + `
@t @x 2 @x2 @x 2 @x2

Fokker-Planck equation: `
2 drift velocity v = (2q 1)
@p @p @ p t
= v +D 2 diffusion `2
@t @x @x D=
coefficient 2 t
10
 Fokker-Planck equation

x
5` 4` 3` 2` ` 0 ` 2` 3` 4` 5`
In general the probability distribution ⇧ of jump
lengths s can depend on the particle position x
⇧(s|x)
Generalized master equation:
X
p(x, t + t) = ⇧(s|x s)p(x s, t)
s
Again Taylor expand the master equation above
to derive the Fokker-Planck equation:
 2

@p(x, t) @ @
= v(x)p(x, t) + 2
D(x)p(x, t)
@t @x @x
drift velocity diffusion coefficient
(external fluid flow, external potential) (e.g. position dependent temperature)
X s ⌦ ↵
hs(x)i X s 2 s (x) 2
v(x) = ⇧(s|x) = D(x) = ⇧(s|x) =
s
t t 11 s
2 t 2 t
 Lévy flights Lévy flight
⇢ trajectory
Probability of
 ↵
C|~s | , |~s | > s0
jump lengths in
 ⇧(~s ) = ↵ = 3.5, D = 2
0, |~s | < s0
D dimensions
Z
Normalization
condition
dD ~s ⇧(~s ) = 1 ↵>D
⇢
Moments of ⌦ 2↵ AD s20 , ↵>D+2
h~s i = 0 ~s =
distribution 1, ↵<D+2
Lévy flights are better strategy than 2D random walk
random walk for finding prey that is scarce trajectory

D. W. Sims et al.
Nature 451, 1098-1102 (2008) 12
 Probability current
Fokker-Planck equation
 2

@p(x, t) @ @
= v(x)p(x, t) + 2
D(x)p(x, t)
@t @x @x
Conservation law of probability
(no particles created/removed)
@p(x, t) @J(x, t)
=
@t @x
Probability current:

@
J(x, t) = v(x)p(x, t) D(x)p(x, t)
@x
Note that for the steady state distribution, where @p⇤ (x, t)/@t ⌘ 0
the steady state current is constant and independent of x

⇤ ⇤ @
J ⌘ v(x)p (x) D(x)p⇤ (x) = const
@x
Equilibrium probability distribution:
If we don’t create/remove
Z x
⇤ 1 v(y)
particles at boundaries then J*=0 p (x) / exp dy
13 D(x) 1 D(y)
 Spherical particle suspended
U (x) in fluid in external potential
Newton’s law:
@2x @U (x)
m 2 = v(x) + Fr
@t @x
fluid external random
drag potential Brownian
force force
@2x
For simplicity assume overdamped regime: 2
⇡0
@t
x Drift velocity
 1 @U (x)
hv(x)i =
averaged over time @x
R particle radius
Equilibrium probability distribution
⌘ fluid viscosity
p⇤ (x) = Ce U (x)/ D
= Ce U (x)/kB T
= 6⇡⌘R Stokes drag coefficient (see previous slide) (equilibrium physics)
kB Boltzmann constant Einstein - Stokes equation
T temperature
kB T kB T
D diffusion constant D= =
14 6⇡⌘R
 Translational and rotational diffusion
for particles suspended in liquid

Translational
 Rotational
 ✓
diffusion x diffusion

2
⌦ 2↵
hx i = 2DT t ✓ = 2DR t
3
Stokes viscous drag: T = 6⇡⌘R Stokes viscous drag: R = 8⇡⌘R

Einstein - Stokes kB T Einstein - Stokes kB T

DT = DR =
relation 6⇡⌘R relation 8⇡⌘R3
Time to move one body length Time to rotate by 900
in water at room temperature in water at room temperature
⌦ 2
↵ 3⇡⌘R3 ⌦ 2↵ 4⇡⌘R3
x ⇠ R2 t⇠ ✓ ⇠1 t⇠
kB T kB T
R ⇠ 1µm t ⇠ 1s Boltzmann constant kB = 1.38 ⇥ 10 23
J/K

t ⇠ 100 years water viscosity ⌘ ⇡ 10 3 kg m 1

s 1
R ⇠ 1mm
15 room temperature T = 300K
 Adolf Fick 1855
Fick’s laws
N noninteracting Local concentration
Brownian particles
of particles
c(x, t) = N p(x, t)
Fick’s laws are equivalent to Fokker-Plank equation
First Fick’s law
@c
Flux of particles J = vc D
@x
Second Fick’s law
 
Diffusion of @c @J @ @ @c
= = vc + D
particles @t @x @x @x @x

Generalization to higher dimensions

@c ~ = ~ · (c~v ) + r
~ · (Drc)
~
J~ = c~v ~
Drc = rJ r
@t
16
Further reading

17