Nanoscale Transport

Diffusion through Nanopores

o Ficks law of diffusion:
o Alternative formulation (Onsager/Maxwell-Stefan):
o The two relations can be combined to give:
Three regimes of diffusion:
o Bulk or macroscopic regime (d >> ! and d >> ")
o D
T
roughly similar to bulk
o Knudsen regime (d << ! and d >> ")
o collisions with the pore wall dominate
o Configurational diffusion regime (d ~ ")
o Diffusing species undergoes cage-like diffusion
!
J = "D
T
(c)#c
!
J = "
L(c)
k
B
T
#
!
D
T
=
L(c)
k
B
T
"
"c
Transport contribution thermodynamic contribution
For fast transport, not only should the transport coefficient L be large
but the chemical potential should also be sensitive to concentration
D
T
: Transport diffusivity
"
d
!
Knudsen diffusion

o Diffusion coefficient from MSD vs. time plot
as D
T
and D
s
are similar:
o However, the MSD dictated by wall collisions
o Knudsen theory
o If all collisions are diffuse => velocities v
xn
and v
xm



are uncorrelated <v
xn
v
xm
>=0 for n ! m
o Assuming Maxwell-Boltzmann distributed x and y velocities
o Self-Diffusivity:
!
D
T
" D
s
= lim
t #$
%x
( )
2
2t
n
th
collision time
x-velocity before
n
th
collision
(for slit pore)
Knudsen diffusion, continued

Smoluchowski theory
o In reality, collisions may not be diffuse, so define tangential momentum
accommodation coefficient f that denotes fraction of collisions that are
diffuse
o Probability that the molecule randomizes its velocity after <$>, 2<$>, 3<$>
times are f, f(1-f), f(1-f)
2
, etc, respectively
Improved theory
o In reality, each collision is not purely diffuse or purely specular; f is fraction
of tangential momentum lost during each collision
The mean and
mean square
collision times are
Self-Diffusivity:
Self-Diffusivity:
Equilibrium MD for computing D
T
o The above simple Knudsen analyses fails when
o molecule-molecule cannot be neglected
o pore dimensions comparable to molecular size and molecules
interact strongly with the walls (configurational)
o MD simulations offer the best approach for computing diffusion
coefficients
o Need to compute the transport diffusivity
o L can be computed from Green-Kubo relationship:
o And then Dt can be computed as follows:
!
D
T
=
L(c)
k
B
T
"
"c
!
L =
V
3
dt' J(t' ) J(0)
0
"
#
!
D
T
=
1
3N
dt' v
i
i=1
N
"
(t' ) v
j
j=1
N
"
(0)
0
#
$
%
&
'
'
(
)
*
*
+ln
+lnc
%
&
'
(
)
*
T
!
D
T
= D
S
+
1
3N
dt' v
i
(t' ) v
j
(0)
0
"
#
j$i
N
%
i=1
N
%
&
'
(
(
)
*
+
+
,ln
,lnc
&
'
(
)
*
+
T
Self-diffusivity
Correlation term that goes to zero in dilute conditions
Transient MD for computing D
T
o Involves setting up a concentration gradient inside the simulation cell
o Relaxation of the concentration gradient using MD simulations and its
matching to the continuum solution from Ficks equation allows one to
compute D
T
o Advantages: good physical basis and efficient
o Disadvantages: difficulty in setting up concentration profile, and
assessing if one is in the linear response regime or not
Non-equilibrium MD for computing D
T
o Drive a macroscopic flow through simulation cell using an external
field
o External field: apply external force to each molecule that is equal to
the supposed chemical potential gradient

o Advantages: efficient, easy to implement
o Disadvantages: use of an external field in place of chemical potential
gradient not formally demonstrated
!
F
ext
= "#
Modified
equations of
motion:
Streaming
velocity:
Gauss
constraint to
keep T constant
Boundary driven MD for computing D
T
o Involves actual construction of high and low concentration reservoirs
within the simulation cell using Grand-Canonical Monte Carlo
simulations
o The molecular flux is then simulated using molecular dynamics and
measured and related to concentration gradient to obtain D
T
o Advantages: conceptually attractive and simple
o Disadvantages: not so easy to implement, more computationally
intensive as it involves simulation of extra molecules of the reservoirs,
streaming velocity implementation, etc
Dual control volume
grand canonical
molecular dynamics
(DCV-GCMD)
Rare event diffusion: Transition state theory
k
%G>>kT
How to deal with this?
Transition state theory (TST)
o Many systems diffuse very slowly at time scales not accessible by
molecular dynamics simulations
o flipping of lipid molecules across the two layers in a bilayer
membrane
o diffusion of large molecules within nanoporous materials
TS
R
P
The rate is dictated
by the transition
state, (bottleneck):
3N-1 dimensional
surface that must be
crossed for a
reaction or diffusion
to take place
Transition state theory, continued
k
TST
%G>>kT
o Formally, there are infinite TST rate constants k
TST
and the true rate
constant k is always <= k
TST

o Trick is to find the most severe bottleneck (energy barrier) for estimating
k
TST
closest to k
o Derivation of TST
o Probability of being located on TS is given by
where <>
R
is an ensemble average over the reactant
o But, where the Z are partition functions
o TST rate constant is equal to probability of being at TS multiplied by the
rate at which the particle crosses the TS in the forward direction
!
P
TS
= "(r # r
$
)
R
!
P
TS
= Z
"
/ Z
R
!
P
TS
=
e
"V(r)/ kT
#(r " r
$
)dr
R
%
e
"V(r) / kT
dr
R
%
!
k
TST
=
1
2
v"(r # r
$
)
R
The factor " implies that at
the TS, trajectories can go
towards forward or backward
with equal probability
Transition state theory, continued
o Average momentum velocity is given by
o The rate constant is then given by:
o Lets now examine the particular case of a particle crossing an
energy barrier from a quadratic energy well
o Extension to multiple dimensions
o All forward moving trajectories may not be reactive
!
v =
ve
"mv
2
/ 2kT
dp
v
#
e
"mv
2
/ 2kT
dp
v
#
=
2kT
$m
k
TST
can be interpreted as the attempt
frequency of oscillations in the energy
well times of the probability of reaching
the transition state
!
k
TST
=
kT
2"m
#(r $ r
%
)
R
!
k
TST
=
k
B
T
2"m
e
#kx
2
/ 2k
B
T
$(x # x
%
)dx
x
&
e
#kx
2
/ 2k
B
T
dx
x
&
k
TST
=
k
B
T
2"m
e
#'E / k
B
T
2"k
B
T /k
k
TST
=
1
2"
k
m
e
#'E / k
B
T
= fe
#'E / k
B
T
!
k
TST
=
f
i
min
i=1
3N
"
f
i
sad
i=1
3N#1
"
e
#$E / k
B
T
3N real normal modes
at the minimum
3N-1 normal modes at
the saddle point
!
k
true
= f
D
k
TST
f
D
< 1: dynamical correction factor