Diffusion through a plane sheet

• Diffusion through a plane sheet models flow through a

membrane
l

c1 = const c0 c2 = const

• At the start: concentration in the membrane c0

• Concentration c0 changes until steady-state is established
• Common experimental setting: c0 = c2 = 0

64

Diffusion through a plane sheet

• Boundary conditions: c0 = c2 = 0
• Solution:
q (t ) Dt 1 2 ∞
(− 1)n exp(− Dn 2π 2t / l 2 )
lKc1 l
= 2 − − 2
6 π
∑1 n2

• With t → ∞:
DKc1 ⎛ l2 ⎞
q (t ) = ⎜t − ⎟
l ⎜⎝ 6 D ⎟⎠

• Intercept (lag time):

l2
L=
6D

65
Diffusion through a plane sheet

Concentration-depth profile Cumulated drug

Donor Barrier Akzeptor

66

Permeability and lag-time

Concentration-depth profile Cumulated drug

Donor Barrier Akzeptor

permeability
coefficient kp

lag-time tlag

67
Time dependent solutions for Fick‘s second law:
Diffusion from a single spot

• Point source of particles (e.g., injection, small crystal)

• t = 0, all n0 molecules at r = 0
• Concentration is finite at all points
• Number of particles is constant

∂c
• Flux is radial J r = −D
∂r

∂c ∂J
• Remember: =− r
∂t ∂r

∂c 1 ∂ ⎛ 2 ∂c ⎞
=D 2 ⎜r ⎟
∂t r ∂r ⎝ ∂r ⎠
68

Time dependent solutions for Fick‘s second law:

Diffusion from a single spot

• Point source of particles (e.g., injection, small crystal)

• Diffusion is spherically symmetric
• t = 0, all n0 molecules at r = 0
• Concentration is finite at all points
• Number of particles is constant

• The solution is:

r2
−
c(r , t ) =
n0
e 4 Dt
(4πDt )3/ 2

69
Steady-state solutions

• Sources
– create particles
• Adsorbers
– destroy particles

• Non-uniform distribution of particles

• Steady-state:
∂c
= ∇ 2c = 0
∂t

• Solution for spherically symmetric diffusion processes:

1 ∂ ⎛ 2 ∂c ⎞
D ⎜r ⎟=0
r ∂r ⎝ ∂r ⎠
2
70

Spherical adsorber

• Spherical adsorber of radius a

• Boundary conditions:
– concentration at surface is 0, i.e., c(a,t) = 0
– concentration at infinite distance is c0 = const

c=0

71
Spherical adsorber

• Spherical adsorber of radius a

• Boundary conditions:
– concentration at surface is 0, i.e., c(a,t) = 0
– concentration at infinite distance is c0 = const

• Solution:
⎛ a⎞
c(r ) = c0 ⎜1 − ⎟
⎝ r⎠

∂c
J r (r ) = − D
a
= − Dc0 2
∂r r

72

Diffusion current

• Remember:
– Flux Jr(a) = number of particles entering the sphere per unit area in
time τ

• Particles are adsorbed by the sphere at the rate flux * surface

area:
I = − J r (a ) ∗ 4πa 2

= 4πDac0
• I is called diffusion current

73
Disk-like adsorber

• Cylindrically symmetric problem

• Boundary conditions:
– concentration at surface is 0
– concentration at infinite distance is c0 = const

74
I = 4Dsc0

Disk-like aperture

• Cylindrically symmetric problem

• Boundary conditions:
– concentration at surface is 0
– concentration at distance x = -∞ is c1 = const
– concentration at distance x = ∞ is c2 = const

75
I = 2 Ds(c2 − c1 )
Disk-like adsorbers on a sphere

• Nonadsorbing sphere of radius a

• N disk-like adsorbers of radius s << a
• Concentration at r = ∞ is c0
• For small N: I ∝ N 4Dsc0
• For large N: I = 4πDac0

76

Disk-like adsorbers on a sphere

• In analogy to electricity: V c0

I=V/R I = c0 / R
• Diffusion resistance for sphere: Ra = 1/4πDa
• Diffusion resistance for each disk: Rs = 1/4Ds
• Total resistance: R = Ra + Rs/N = 1/4πDa + 1/4DNs
R = Ra + Rs / N
1 1
= +
4πDa 4 DNs
1 ⎛ πa ⎞
= ⎜1 + ⎟
4πDa ⎝ Ns ⎠
⎛ πa ⎞
= Ra ⎜1 + ⎟
⎝ Ns ⎠ 77
Disk-like adsorbers on a sphere

• Solution:
4πDac0
I=
1 + (πa / Ns )

I 1
=
I 0 1 + (πa / Ns )

• The half maximum of I/I0 is at:

πa
N=
s
78

Disk-like adsorbers on a sphere

1,0

0,8

0,6
I/I0

0,4

0,2

0,0
0 1 2 3 4 5
N

79
Disk-like adsorbers on a sphere

• Example:
a = 5 µm
s = 10 Å

N1/2 = πa/s = 15,700

Surface fraction covered by adsorbers:

N1/2πs2/4πa2 = 1.6 x 10-4

Distance between neighboring adsorbers:

(4πa2/N1/2)1/2 = 0.14 µm = 1400 Å

80

Diffusion through apertures in a planar barrier

x=0 x=b 81
c1 c2
Diffusion through apertures in a planar barrier

• I2,1 = DA(c2-c1)/b
• Rs = ½ Ds
R
• Rtotal = R2,1 + s
N
b 1
= +
DA 2 DNs
b ⎛ A ⎞
= ⎜1 + ⎟
DA ⎝ 2 Nsb ⎠
b ⎛ 1 ⎞
= ⎜1 + ⎟
DA ⎝ 2nsb ⎠

with n = number of apertures per unit area

I 1
= 82
I 2,1 1 + 1 / 2nsb

Diffusion to capture: two adsorbing boundaries

• Particle released at x = a
• What is the probability of being adsorbed at x = 0 ?

Ileft Iright

x=0 x=a x=b 83

c=0 c = const = cm c=0
Diffusion to capture: two adsorbing boundaries

• Ileft = DAcm/a
• Iright = DAcm/(b-a)

• Probability of being adsorbed at x = 0:

I left b−a
=
I left + I right b

84

Mean time to capture: two adsorbing boundaries

• Particle released at x = a
• What is the mean time W(a) for particles to be captured at x = 0
and x = b ?

Ileft Iright

x=0 x=a x=b 85

c=0 c = const = cm c=0
Mean time to capture: two adsorbing boundaries

• Particle may step left or right, so

1
W (x ) = τ + [W (x + δ ) + W (x − δ )]
2

2τ + W (x + δ ) − W ( x ) + W (x − δ ) − W ( x ) = 0
2τ 1 1
+ [W (x + δ ) − W (x )] − [W (x ) − W ( x − δ )] = 0
δ δ δ
dW (x ) dW ( x ) 2τ
− + =0
dx x dx x −δ δ
d 2W 2τ
+ =0
dx 2 δ 2
d 2W 1
+ =0
dx 2 D 86

Mean time to capture: two adsorbing boundaries

• Boundary conditions:
– adsorbing boundary
W=0
– adsorbing boundary
dW/dx = 0

• In the above example:

– adsorbing boundaries:
W(0) = W(b) = 0

– solution: W (x ) =
1
2D
(
bx − x 2 )
– mean time to capture particle released at random position x:

b2
b
1
∫ W ( x )dx = 87
b0 12 D
Mean time to capture:
adsorbing and reflecting boundary

• Boundary conditions:
– adsorbing boundary at x = 0
– reflecting boundary at x = b

• With

W(0) = 0
dW/dx = 0 at x = b

the solution is
W (x ) =
1
2D
(
2bx − x 2 )

88

Mean time to capture:

adsorbing and reflecting boundary

Mean time to capture particle released at random position x:

b2
b
1
∫ W ( x )dx =
b0 3D

89
Mean time to capture
2
b /2D

adsorbing boundary at x = 0,
reflecting boundary at x = b

2
b /4D
W

adsorbing boundaries at x = 0, x = b
0
0 b/2 b
x

90

Mean time to capture

• In one dimension:
d 2W 1
+ =0
dx 2 D

• In two or three dimensions:

1
∇ 2W + =0
D

91
 Diffusion

1. Microscopic modelling: Molecular dynamics

2. Mesoscopic modelling: Random walk
3. Macroscopic Theory
a. Fick‘s laws of diffusion
b. exact solutions
c. numerical solutions
H.C. Berg: Random walks in biology - Princeton University Press 92
J. Crank: The mathematics of diffusion – Oxford University Press

Physical derivation

h
q1 q2

c c0

c1 c2

x 93
Physical derivation

• Amount diffusant entering through unit area in time τ:

Dτ
q1 = − (c1 − c0 )
h
• Amount diffusant leaving through unit area in time τ:
Dτ
q2 = − (c2 − c1 )
h
• Net gain of diffusant:
Dτ
q1 − q2 = (c0 − 2c1 + c2 )
h
= h(c1′ − c1 )

Dτ
∆c = c1′ − c1 = (c0 − 2c1 + c2 )
h2
94

Numerical solutions (one-dimensional)

• Select two grid constants h (space) and k (time)

• Grid points are (xi, tj) with xi = ih for i = 0, 1, …, m and tj = jk
for j = 0, 1, …, n.

wi,j+1

wi-1,j wi,j wi+1,j

t

h k

95
x
Numerical solutions (one-dimensional):
Forward scheme

• Explicit finite-difference formula:

wi , j +1 = ⎜1 − 2 ⎟ wi , j + D 2 (wi +1, j + wi −1, j )

⎛ 2 Dk ⎞ k
⎝ h ⎠ h

• Error: O(k+h2)
k 1
• Conditionally stable: D 2 ≤
h 2
• Number of multiplications: Z ≅ 2n

96

Forward scheme

With
λ = D(k / h 2 )
and

⎡(1 − 2λ ) λ 0L 0 ⎤
⎢ λ (1 − 2λ ) λ O M ⎥
⎢ ⎥
A=⎢ 0 λ O O 0 ⎥
⎢ ⎥
⎢ M O O O λ ⎥
⎢⎣ 0 L 0 λ (1 − 2λ )⎥⎦
this can be written as

w ( j ) = Aw ( j −1)
97
Numerical solutions (one-dimensional):
Backward scheme

• Implicit method

wi , j − wi , j −1 wi +1, j − 2 wi , j + wi−1, j
−D =0
k h2

• Error: O(k+h2)
• Unconditionally stable
• Needs boundary conditions
• Values may be negative

98

Backward scheme

With
λ = D(k / h 2 )
and

⎡(1 + 2λ ) −λ 0 L 0 ⎤
⎢ −λ (1 + 2λ ) − λ O M ⎥
⎢ ⎥
A=⎢ 0 −λ O O 0 ⎥
⎢ ⎥
⎢ M O O O −λ ⎥
⎢⎣ 0 L 0 − λ (1 + 2λ )⎥⎦

this can be written as

Aw ( j ) = w ( j −1)
99
 Numerical solutions (one-dimensional)

• Implicit: Crank-Nicholson method

wi , j +1 − wi. j D ⎡ wi +1, j − 2 wi, j + wi −1, j wi +1, j +1 − 2 wi , j +1 + wi −1, j +1 ⎤

− ⎢ + ⎥=0
k 2⎣ h2 h2 ⎦

• Error: O(k2+h2)
• Unconditionally stable
• Number of operations (inluding solving the tridigonal system):
Z ≅ 5n
• Needs boundary conditions
• Values may be negative

100
J. Crank, P. Nicolson. A practical method for numerical evaluation of solutions of partial
differential equations of the heat-conduction type. Proc. Camb. Phil. Soc. 43:50-67 (1947)

Crank-Nicholson method

With
λ = D(k / h 2 )

and
⎡ λ ⎤
⎢(1 + λ ) −
2
0 L 0 ⎥
⎢ λ λ ⎥
⎢ − (1 + λ ) − O M ⎥
⎢ 2 2 ⎥
λ
A=⎢ 0 − O O 0 ⎥
⎢ 2 ⎥
⎢ λ ⎥
⎢ M O O O − ⎥
⎢ 2 ⎥
λ
⎢ 0
⎢⎣
L 0 − (1 + λ )⎥⎥
2 ⎦

101
Crank-Nicholson method

and
⎡ λ ⎤
⎢(1 − λ ) 2
0 0 ⎥
L
⎢ λ λ ⎥
⎢ (1 − λ ) O M ⎥
⎢ 2 2 ⎥
λ
B=⎢ 0 O O 0 ⎥
⎢ 2 ⎥
⎢ λ ⎥
⎢ M O O O ⎥
⎢ 2 ⎥
λ
⎢ 0
⎢⎣
L 0 (1 − λ )⎥⎥
2 ⎦

this can be written as

Aw ( j +1) = Bw ( j )
102
unknown concentrations known concentrations

The tridiagonal linear system

• The equation
Aw ( j +1) = Bw ( j )
can be written as
Ax = b
which is basically a tridiagonal linear system:
b1 x1 + c1 x2 = d1 ; i = 1
ai xi −1 + bi xi + ci xi+1 = d i ; i = 2, K, n − 1
an xn −1 + bn xn = d n ; i = n

103
Solving the tridiagonal linear system

• Direct methods
– Gauss‘s elimination method without pivoting
– LU decomposition / Thomas algorithm
– Crout reduction

• Iterative methods
– Jacobi method
– Gauss-Seidel method
– Successive over-relaxation method

104

LU decomposition

b1′ = b1
d1′ = d1
for k = 2K n do
ak
ak′ = , bk′ = bk − ak′ ck −1 , d k′ = d k − ak′ d k′ −1
bk′ −1
end
d n′
xn =
bn′
for k = n − 1K1 do
d′ − c x
xk = k k k +1
bk′
end

H.R. Schwarz: Numerische Mathematik – Teubner Verlag Stuttgart 105

G.D. Smith: Numerical solution of partial differential equations – Oxford University Press
Crank-Nicolson method w/ Crout reduction

Let nx = the number of data points + 1

Let c[x] be the concentration for x = 0,...,nx with
c[0] = c[nx] = 0.

Step 1
INPUT c[x] = input data for x = 1,...,nx-1
Set c[0] = c[nx] = 0
Set lambda = Dk/h2
Set u[0] = 0

Step 2 (Initialization of C-N tridiagonal matrix)

For x = 1 to nx-1
Set l[x] = 1 + lambda + lambda/2 * u[x-1]
Set u[x] = -lambda/(2*l[x])

106

Crank-Nicolson method w/ Crout reduction

Step 3 (Iteration of C-N algorithm for nt time steps)

For t from 1 to nt
Set z[0] = 0
(Solution of tridiagonal system by Crout
reduction)
For x from 1 to nx-1
Set z[x] = ( ( 1 – lambda ) c[x] +
lambda/2 * ( c[x+1] + c[x-1] +
z[x-1] ) ) / l[x]
(Back substitution)
For x from nx-1 to 1
Set psi[x] = z[x] - u[x]*psi[x+1]

Step 4
OUTPUT c[1] through c[nx-1]

107
Boundary conditions

• Dirichlet conditions:
value at the boundary is function of time, e.g., w(0,t) = b(t)

• Neumann conditions:
specify flux at the boundary

• Mixed boundary conditions

• Periodic or wraparound boundary conditions

108

Numerical solutions:
Two dimensional forward scheme

• Basic equation:
∂c ⎛ ∂ 2 c ∂ 2c ⎞
= D⎜⎜ 2 + 2 ⎟⎟
∂t ⎝ ∂x ∂y ⎠
• Grid:
– x, y space divided in elements of length h, k with indices a, b
– timestep of length τ, index n
• Concentration at all points is known for t = 0
• Solution:
wa ,b,n +1 − wa ,b,n
=
D
(wa−1,b,n − 2wa ,b,n + wa+1,b,n )
τ h 2

+ 2 (wa ,b−1,n − 2 wa ,b,n + wa ,b +1,n )

D
k
• Stable for:
⎛ 1 1⎞ 1
D⎜ 2 + 2 ⎟τ ≤ 109
⎝h k ⎠ 2
Mathematical derivation of solutions

• Approximate ⎛ ∂c ⎞ ⎛ ∂ 2c ⎞
⎜ ⎟ = D⎜⎜ 2 ⎟⎟
⎝ ∂t ⎠ x ⎝ ∂x ⎠

by Taylor series
∂u (x ) 1 2 ∂ 2u ( x ) 1 2 ∂ 3u ( x )
u (x + h ) = u (x ) + h + h + h K
∂x 2 ∂x 2
6 ∂x 3

∂u (x ) 1 2 ∂ 2u (x ) 1 2 ∂ 3u (x )
u (x − h ) = u ( x ) − h + h − h K
∂x 2 ∂x 2 6 ∂x 3

e.g.: addition
∂ 2u ( x )
u (x + h ) + u (x − h ) = 2u ( x ) + h
∂x
2
2
+ O h4 ( )
∂ 2 c ( x ) c ( x + h ) − 2c ( x ) + c ( x − h )
⇒ ≅
∂x 2 h2 110

Miscellaneous equations

• Douglas equation
– implicit (1 − 6λ )wi −1, j +1 + (10 + 12λ )wi , j +1 + (1 − 6λ )wi +1, j +1
= (1 + 6λ )wi−1, j + (10 − 12λ )wi , j + (1 + 6λ )wi +1. j

error: O(k2+h4)
unconditionally stable

– explicit
1
wi , j +1 =
2
( )
2 − 5λ + 6λ 2 wi , j + λ (2 - 3λ )(wi +1, j + wi −1, j )
2
3
− λ (1 - 6λ )(wi + 2, j + wi − 2, j )
1
12

error: O(k2+h4)
conditionally stable: λ ≤ 2/3
111
Miscellaneous equations

• Du Fort and Frankel

– explicit
(1 + 2λ )wi , j +1 = 2λ(w i-1, j + w i+1, j ) + (1 − 2λ )wi, j −1
– unconditionally stable
– error: O(k2+h2)

• Implicit three-time level difference equation

3 wi , j +1 − wi, j 1 wi, j − wi, j −1 wi +1, j +1 − 2 wi , j +1 + wi −1, j +1
− =
2 Dk 2 Dk h2

– unconditionally stable
– error: O(k2+h2)

112