Diffusion

Part 1

Diffusion equation
random walk derivation

Diffusion equation
Random walk reconsidered (PBoC 13.2.1)
Imagine you have calculated the probability p(x,t) that a random walker finds himself at x
after a time t.
p(x,t+!t) depends on p(x,t) in the immediate vicinity of x (!x, to be exact):

x-!x

t
k!

k!
t

1-2k!t

t
k!

k!

1-2k!t

k!

x+!x

t+!t
x-!x

x+!x

Our previous random walk notation had k!t=l=r and !x=h

k!t=1/2 is the classic random walk; k!t < 1/2 is a random walk with drunken pauses

p(x, t + t) = kt p(x x) + (1 2kt) p(x, t) + kt p(x + x)

Diffusion equation
To first order in !t this becomes

p(x, t + t) = p(x, t) + kt {p(x + x) + 2p(x, t) + p(x x)}

which becomes (to second order in !x)
2
p
2 p
= kx
t
x2

p
2p
= D 2 with diffusion coefficient D kx2
This is the diffusion equation
t
x
To connect this to our previous expressions for the unbiased 1D random walk, use k!t=1/2
to identify k ! (2!t)-1, x ! R, !x ! h, and t/!t ! N giving Dt=Nh2/2.
We will see below that the free-space solution to the diffusion equation is p(x, t) =
which is identical to our 1D random walk expressions with the substitutions above.

x2 /4Dt
1
e
4Dt

Not surprisingly, the random walk with drunken pauses corresponds to Dt < Nh2/2: slower
diffusion.

Diffusion equation
kinetic theory derivation

Diffusion equation
Kinetic theory of gases

area A

dx

Ficks law

J(x)
J+(x)
J(x+dx)
J+(x+dx)

The net particle flux is:

Jx = D

dc
dx

!
J! = Dc

#
area time

1D

3D

length2
D
time

Diffusion equation
Einstein showed that: D =

kB T

where ! is the drag coefficient defined by F = ! v.

For a sphere of radius a in a fluid with viscosity ", = 6a
For a typical small ion, D 1 m2 /ms
Since a

m , we expect D m1/3

globular proteins

Since the drag of a nonspherical

object at low Reynolds number
is dominated by its longest
dimension, thin proteins
diffuse less than their mass suggests.
needlelike proteins

Diffusion equation
Conservation of mass (or number)
area A

dx

J(x!dx/2)

J(x+dx/2)

n(x) = c(x) A dx

leads to the continuity equation:

Jx
c
=
t
x

1D

c
" J"
=
t

3D

Diffusion equation
Together, Ficks law + conservation law lead to the diffusion equation:

3D:

!
"
c
"
"
= Dc = D2 c
t

1D:

2c
c
=D 2
t
x

This is identical to (and often called) heat conduction

This is a second order PDE: it requires an initial condition and two boundary conditions to
be solved.
This form assumes that D is constant in space.
We could add source or sink term to right hand side if appropriate.

This is identical to the equation for p(x,t) from the random walk, except
here we talk about the c(x,t).
Since c(x,t) = N p(x,t) (where N is the total number of particles in the system), I will use p and
c interchangeably.

Free-space solutions

Free-space diffusion
Solutions to diffusion equation in free space (with c=0 at infinity) look
like spreading Gaussians
c(x, t) =
or

c(x, t) = !

2
N
ex /4Dt
4Dt

N
2 2 (t)

x2 /2 2 (t)

2 (t) 2Dt

You can derive this using Fourier transforms, but its easier to guess it and plug back into PDE
to check its validity
Mass is conserved:

c(x, t)dx = N

Diffusion profiles always smear out: they never sharpen.

Free-space diffusion
The 1D solution looks like

c
1.0

0.8

0.6

0.4

0.2

!10

!5

snapshots in time

10

animation

Free-space diffusion

The edge (position of the half-height, for instance) propagates

like (t) 2Dt

details can change, but any diffusive process ends up with

So the speed of propagation is constantly decreasing!

Diffusion is a reasonable transport process only over short times and distances.
This is not the typical speed of a diffusing molecule, which is similar to the speed of sound (100 m/s or so)
Typical values for D range from 1 "m2/s to 1 "m2/ms:
d"/dt will slow to a few "m/s within a millisecond or second

Free-space diffusion
Even if your initial concentration profile is weird, if it has a
finite width you always end up with a spreading Gaussian:

Free-space diffusion
We can solve diffusion equation in spherical coordinates:
c
1
=D
t
r r

r2

c
r

"

by noting that its a superposition of 3 individual 1D diffusions:

4r2
r 2 /2Dt
c(x)c(y)c(z) dx dy dz = N
e
dr c(r) dr
(4Dt)3/2

Similarly for cylindrical coordinates, where

2
2 /2Dt
c(x)c(y) dx dy = N
e
d c() d
(4Dt)2/2

Free-space diffusion
In summary, we have

1D

1
x2 /4Dt
c(x, t) = N
e
(4Dt)1/2

2D

2
2 /2Dt
e
c(, t) = N
(4Dt)1

3D

4r2
r 2 /4Dt
c(r, t) = N
e
(4Dt)3/2

"
x2 =

! 2"
=

! 2"
r =

x2 c(x) dx = 2Dt

2 c() d = 4Dt

r2 c(r) dr = 6Dt

Diffusion = random walk

The end-to-end distance of a random coil is the same as the
net displacement in a random walk is the same as the
diffusion of concentrated burst.
Always find <x2> ! t if correlation between links is short:
for instance, finite length m (including m=0) or exponentially decaying

Long-distance correlations (for instance, power law) lead to unusual

superdiffusive behavior:
defined as <x2>

! tn with n>1

often called Lvy flights

not to be confused with small t ballistic regime
has been demonstrated in
membranes (active transport?)
cell interiors (active transport)
ecology (foraging / search algorithms of albatrosses, bumblebees and deer.)
(Nature 449:1044 or Ecology 88:1962)

great stock market crash of 2008

(The Economist January 24th 2009)

Advection-diffusion

Advection-diffusion
What if your random walk is not unbiased?
Go back to the earlier derivation, but make rightward and leftward step rates (k+ and k-)
unequal:

k!
- t

1-(k-+k+)!t
k!
- t

x+!x

!t
k+

!t
k+

k!
- t

1-(k-+k+)!t

x-!x

t+!t
x-!x

x+!x

The expression for p(x,t+!t) is now

p(x, t+t) = k+ tp(xx)+(1(k+ +k )t)p(x, t)+k tp(x+x)

which eventually becomes the Smoluchowski equation (v1) PBoC eqn 13.54:

c
2c
c
=D 2 v
t
x
x
where we identify D and v with the microscopic rates: D = (k++k) !x2/2 and v = (k+k) !x.

Advection-diffusion
What if you have bulk flow as well as diffusion?
Moving fluid will transport its local concentration c(x). This is called advection (or
sometimes, incorrectly, convection).
This contributes an advective flux J!adv = !v c
! and put the total flux into the
We add this to our original diffusive flux J! = Dc
continuity equation to give

3D:

"
c
"
"
= Dc + "v c
t

1D:

=
t
x

!
"

D
+ vc
c

In 1D with constant D and v, this is the same as Smoluchowski v1 from the previous slide.
In 1D with constant D but not assuming v is constant, this becomes Smoluchowski v2:

c
2c
(vc)
=D 2
t
x
x
This PDE describes the evolution of a concentration profile due to diffusion and advection simultaneously.

Advection-diffusion
For constant v, the solution is simple: a diffusing Gaussian profile that drifts at speed v.

Advection-diffusion
Microscopic Smoluchowski equation
The movement v could be caused by an applied force F rather than bulk fluid flow.
Using the definition of !, (F= !v), the Einstein relation (D=kBT/!) and the definition of force
(F = - #G/#x), and writing p for c, we get Smoluchowski v3

p
p
1
=D
+
t
x2
kB T x
2

"

#$

G
p
x

Now we apply this to a generalized energy landscape G(x). It describes how the population
(or probability) p(x,t) flows in time and space.
The only unknown is the microscopic diffusion parameter D. We have to estimate it, but since everything is
proportional to D as long as its constant it will change the time scale but not the shape of p. To see this,

Nondimensionalize the Smoluchowski equation:

x*=x/l, G* = G/G0, t* = t D/l2, T*=kBT/G0, giving

! 2
" #$
p
p
1
G
=
+
p
t
x2
T x x

Advection-diffusion
We drop the *s to give Smoluchowki v4 (the last one):

p
p
1
=
+
2
t
x
T x
2

"

#$

G
p
x

Everything in this equation is of order 1. This is easier to visualize and easer to solve numerically.
2000 Naturetemperature
America Inc. k
http://structbio.nature.com
There is only one parameter that matters, the dimensionless
BT/G0.

When you seenews

a complicated
free energy surface G(x), imagine the protein diffusing around
and views
(generally toward the minimum of G) according to the Smoluchowski equation.
Free
Fig. 1. Schematic of the
folding energy landscape of
a protein molecule where
the energy of the protein is
displayed as a function of
the topological arrangements of the atoms. The
multiple states of the
unfolded protein located at
the top fall into a folding
funnel consisting of an
almost infinite number of
local minima, each of which
describes possible folding
arrangements in the protein. Most of these states
represent transient folding
intermediates in the process
of attaining the correct
native fold. Some of these
intermediates retain a more
stable structure such as the molten globule, whereas other local minima act as folding traps irreversibly capturing the protein in a misfolded state (see ref. 16 for details).

side chains of hydrophobic amino acids

into the interior of the protein. The folding process of a protein generally depends
on the strength of this hydrophobic effect
as well as on the stability of the protein
and can be described by the energy landscape theory16,19. The topology of the
folding energy landscape can best be
described as a funnel with a rough surface
(Fig. 1). The intermediate states populate
the local minima within the rough surface
and the inner funnel wall, sometimes
entrapping misfolded conformations.
The roughness of the funnel surface, and
thus the formation of the native state can
vary significantly between different types
of proteins. Pure !-helical proteins
almost simultaneously undergo an enormous reduction in conformational space

The Smoluchowski equation

The Smoluchowski equation

The steady-state solution to the Smoluchowski equation is the
Boltzmann distribution.
Smoluchowski describes the process of reaching thermal equilibrium.

In general, the equation is hard to solve.

Ive only seen analytical solutions for
constant G(x) (the diffusion equation)
linear G(x) (the advection-diffusion equation)
quadratic G(x) (no name: thermalization in a springlike potential)

The Smoluchowski equation

Diffusion over a barrier shows thermalization within the barrier (for T*<1) followed by
barrier crossing. At lower T it takes longer to cross the barrier.

T*=1/2

T*=1/8

The Smoluchowski equation

When the landscape is asymmetrical, the system ends up predominantly in the lower-energy
state. At lower T, this is more pronounced and takes longer.

T*=1/4

T*=1/16

The Smoluchowski equation

More complicated landscapes lead to more complicated behaviors, including intermediate
states. These appear as kinetic intermediates even if they arent present in the long-time,
thermalized, Boltzmann distribution limit.

T*=1/8; different biases to the landscape

The Smoluchowski equation

When T* is small (meaning either large energies G0 or small temperatures kBT), the
distribution tends to linger in local minima. There is little probability of being outside a
local minimum, so the local minima can plausibly be considered the states of this
landscape, with the deepest minimum (the global minimum) the ground state.

T*=2

T*=1/8

The Smoluchowski equation

A quench simulation shows the idea behind the conformational substate model: a single
state at high T resolves into three distinct states at low T.

The Smoluchowski equation

The rate of transition from state A to B depends on the barrier between them and on the
temperature. This can be calculated analytically for the Smoluchowski equation if T* is
small. Kramers theory ultimately gives an expression for the transition from A to B:
G /kB T

kAB = e

# depends on details of the model, but in its simplest form is =

!
kA k /2

where kA is the curvature of the free energy surface at the bottom of the well and k is the curvature at the top.
For our purposes the details dont matter: just that transition rates are exponential in !G = G-GA, referenced to
temperature.
This is simply the Arrhenius relation. Different treatments give different prefactors #, but all have the activation energy
dependence.

Exponential kinetics

Exponential kinetics
Sequential transitions
If there is a mandatory path from A to B to ... to the native state and the barriers between
states are large, then the folding process can be represented by

A k!
B k!
C ! ... ! N
A!B
B!C
If the rates are similar, you will see intermediate states accumulate. If not, the rate to N will be dominated (at long
times) by the slowest of all the individual rates.
kA

<< kI

0.9

0.9

0.8

0.8

0.8

0.7

0.7

0.7

0.6

0.6

0.6

A
intermediate
B

number

0.5

number

0.9

number

kA

equal rates

0.5
0.4

0.4

0.3

0.3

0.3

0.2

0.2

0.2

0.1

0.1

0.1

6
time

10

6
time

10

>> kI

0.5

0.4

6
time

10

Exponential kinetics
For a finite (small) number of states, the kinetics of the final state will reflect all the
individual rates.
If there are only a few states and rate constants, you may be able to identify different rates and infer how many
intermediates there are.

If there are many intermediates, and especially if there are many intermediates with closely
spaced rates (i.e. energy barriers), you may not be able to identify individual rates and the
long-time behavior may not look exponential.
This type of non-exponential kinetics is typical of an energy landscape that is of intermediate roughness:
not so rough that the protein is trapped in free minima
rough enough that many minima contribute and no single barrier definitively controls the folding process.

Structure / folding summary

2000 Nature America Inc. http://structbio.nature

news and views

primary structure
random coil

Free

Schultz, Illuminating folding

intermediates, Nature Structural
Biology 7:7-10 (2000)

side c
into t
ing pr
on the
as we
and ca
scape
foldin
descri
molten globule
(Fig. 1
the lo
and t
entrap
native conformation
The ro
thus t
vary s
of p
almos
intermediate mous
misfold
and h
Extensive research in protein folding and structure predictions may therefore suc- ments
unfolding have uncovered rate constants ceed in finding the correct fold of an some
Fig. 1. Schematic of the
folding energy landscape of
a protein molecule where
hydrophobic
the energy of the protein is
collapse
displayed as a function of
the topological arrangements of the atoms. The
multiple states of the
unfolded protein located at
the top fall into a folding
funnel consisting of an
almost infinite number of
local minima, each of which
describes possible folding
arrangements in the protein. Most of these states
represent transient folding
intermediates in the process
of attaining the correct
native fold. Some of these
intermediates retain a more
rate-limiting
stable structure such as the
molten globule, whereas other local minima act as folding traps irrekinetic
versibly capturing the proteinbarrier
in a misfolded state (see
ref. 16 for details).
kinetically
trapped

secondary structure

ture.com

tertiary structure

Structure / folding summary

Folding on a free energy landscape

from the Dill group (UCSF)