Diffusion and velocity relaxation of a Brownian particle immersed in a viscous compressible fluid
confined between two parallel plane walls
Ubbo Felderhof, RWTH Aachen
Many processes in physical chemistry and biology are dominated by the process of diffusion.
In geometry on a small scale, i. e. membranes, thin liquid fims, pores, one has to worry about the influence of geometry
on the diffusion coefficient.
Near a wall diffusion becomes anisotropic and one has to deal with a diffusion tensor dependent on the distance h to the wall.
In bulk Einstein 1905
with friction coefficient Stokes 1850
shear viscosity particle radius
Near a wall with mobility tensor
parallel xy-plane
( )h Dt
D kT
= ζ
6 a ζ = πη
η a
( )
h=
kTµ ( )
hD
t t µ t ( ) h = t ς ( ) h
−1( )h xx( )(h x x y y) zz( )h z z
µ
t =µ
e e +e e +µ
e e0
( ) [1 9 ]
xx 16 h a
µ = µ − h
0
( ) [1 9 ]
zz 8 h a
µ
=µ
− h Lorentz 19070
1 6 a
µ = πη Higher order correction terms first worked out by Faxén 1925 At present the mobilities and are known very precisely.
Similar results for a particle between two plane walls and
zz
( ) h µ
xx
( ) h µ
( , )
xx h L
µ µ
zz( , )h L To first order ina
h
Faxén 1925
( , ) 0[1 1.004 ]
xx 2
L a
L h
µ =µ −
( , ) 0[1 1.452 ]
zz 2
L a
L h
µ =µ − BUF 2005
So far we considered static diffusion tensor
For fast processes it may be necessary to generalize to a frequency-dependent tensor
Again there is an Einstein-type relation
where is the admittance tensor for the geometry h, L at frequency
For applied force the particle velocity is with
( , )h L Dt
( , , )h L
ω
Dt( , , ) h L ω = kT ( , , ) h L ω
D t y t
( , , ) h L ω yt ω
( )t = Re ωe−i tω
E E
( )t Re ωe i t
ω
= −
U U
( , , ) h L
ω
= ω ⋅
ωU y t E
The diffusion process is related to velocity relaxation by
with velocity correlation function For
( ) (0)t v v
0
( ) ω =
∞∫
ei tω( ) (0)
t dtD
t
v v( ) (0) t = (0) (0) e
−ζt m/ pv v v v
with (0) (0)p
kT
= m
v v 1
t
this gives the Einstein relation More generally
Corresponding to admittance In confined geometry
(0) kT D = ζ
( ) p ( )
D kT ω i m
ω ζ ω
= − +
( ) ( )
t
p
kT ω i m
ω ζ ω
= − +
y
( ) ω = kT ( ) ω D t y t
0
( ) 1 e
i t( ) (0) t dt kT
ω =
∞∫
ωy t v v
fluctuation-dissipation theoremBy inverse Fourier transform
( ) (0) ( )
2 kT
i tt ω e
ωd ω
π
∞
−
−∞
= ∫
v v yt
0 t >
This may be used to calculate v( ) (0)t v in confined geometry.
2
3 At high frequency
The behavior at low frequency is of particular interest,
since it is related to the long-time behavior. This is affected by the geometry.
Alder and Wainwright found 1970 in computer simulation
( ) 1
i mp
ω ≈ −ω
yt 1 as
ω → ∞
( ) (0) t t
−3/ 2v v
ast → ∞
This was first understood from kinetic theory, later from hydrodynamics.
The admittance of a sphere in an incompressible fluid behaves at low frequency as
1
2( ) 1 ( )
t
6 a O
ω a α α
πη ⎡ ⎤
= ⎣ + + ⎦
y
α = −iηωρThis yields by Tauberian theorem
3/ 2 3/ 2
( ) (0) 1
12 ( )
t kT t
πρ πν
≈ −
v v 1 as
t → ∞
ν η
= ρ kinematic viscosity Quite generally for
Tauberian theorem small ω behavior large t behavior of of
0
ˆ ( )
i t( )
fω =
∞∫
eω f t dtˆ ( )
f
ω
f t( )conversely large ω behavior small t behavior of offˆ ( )
ω
f t( ) The converse theorem has also played a role in physics.Following earlier remarks by Lorentz, and work by H. Weyl, there is a famous paper by Mark Kac
„Can one hear the shape of a drum?“
In his paper Kac actually reduces the acoustic problem to a diffusion problem:
Consider the conditional probability of finding a particle at r at time t when it starts out at 0 at time 0
( , | , 0) P r t 0
P
behaves like particle density and therefore satisfies P 2D P
t
∂ = ∇
∂
The fundamental solution is 2
3/ 2
( , | , 0) 1 exp[ ]
(4 ) 4
P t r
Dt Dt
π
= − r 0
Mean square displacement
i.e. size of probability cloud grows as Kac considered
Write this as integral of decaying exponentials
2 2
( , | , 0) 6 r =
∫
r P r t 0 dr = Dtt
3/ 2
( , | , 0) 1
(4 ) P t
π Dt
0 0 = t > 0
3/ 2 0
( ) exp[ ] 1
(4 )
g t d
λ λ λ Dt
π
∞
− =
∫
then
3/ 2
( ) 1
(4 ) (3 / 2)
g D
λ λ
= π
Γ
(3 / 2) 1
2
π
Γ =
in agreement with Tauberian theorem
(3 / 2) λ
Γ
3/ 21 t
large
λ
smallt
small
λ
larget
In this case both types of behavior are realized at the same time.
4
Similarly in a viscous incompressible fluid satisfies the linearized Navier-Stokes equation
2
p
ρ ∂ t = ∇ − ∇ η
∂
v v ∇ ⋅ = v 0
For d- impulse at t=0 fundamental solution
At the origin
0 t >
( , ) 1 ( , )
t
4
t= πη ⋅
v r T r P
1 1 3/ 2
( , )
4 3
t t
πη πν
= −
v 0 P all t
3/ 2 3/ 2
1
12 ( ) t ρ πν
= − P
corresponds precisely to the long-time behavior of Brownian particle found from term in
This shows that the velocity correlation function of a Brownian particle is closely related to the Green function of the hydrodynamic
equations of motion.
But the Green function depends on geometry.
One can expect that in particular the long-time behavior is strongly dependent on geometry.
Gotoh and Kaneda (1982) found that in the presence of a single plane wall the long-time behavior is
ω
yt( )ω
5 / 2
( ) (0)
x x
v t v t
−v t v
z( ) (0)
zt
−7 / 2I found (2005) that the latter result is incorrect. Both correlation functions behave as
5 / 2
xx
( )
xxC t ≈ A t
−C
zz( ) t ≈ A t
zz −5 / 2with coefficients
A
xxA
zz The second coefficient may be<0, depending on particle mass.
5
( , ) t v r
2 p ( ) ( )t
ρ
∂t − ∇ + ∇ =η δ δ
∂
v v P r
Velocity correlation function for a fluid with a single wall
was studied in computer simulation by Pagonabarraga, Hagen, Lowe, Frenkel 1998 It turned out that fluid compressibility has a significant effect.
In bulk compressible fluid one can calculate the velocity correlation function again from the admittance
Result:
t( )
ω
y3/ 2 5 / 2
( ) (0) 3/ 2
12 ( )
x x
v t v kT t At
πρ πν
− −
≈ +
with a coefficient A that is negative if the fluid is sufficiently compressible, i.e. the decay is not monotonic, but can change sign
BUF, JChemPhys 2005 ( ) 1
t ( )
i mp
ω = −ω +ζ ω
y with a complicated expression for the
friction coefficient
Zwanzig, Bixon 1970 Bedeaux, Mazur 1974 Metiu et al. 1977
ζ ω ( )
depends on shear viscosity, bulk viscosity, density, compressibility Again a wall causes modification of the behavior, but I found thatthe coefficients Axx and Azz of the t-5/2 long-time behavior are independent of compressibility, BUF 2005
(limit to bulk behavior not simple)
For a fluid confined between two walls Pagonabarraga et al.
found a dramatic change of behavior (1997,1998)
( ) ( ) (0)
2xx x x
C t = v t v ≈ At
− with A<0 no details were shownThey made more elaborate analysis in 2D: fluid between two lines.
In that case
( ) ( ) (0) 3/ 2
xx x x
C t = v t v ≈ At− with A<0 They gave expression for A.
6
Recently I have calculated the coefficient A of the t-3/2 long-time tail in 3D.
Result:
2 2
5 2 2
0
9 ( )
( ) ( ) (0)
xx x x
2
h L h kT
C t v t v
L c t
π ρ
= ≈ − −
c
0 is the adiabatic (long-wave) sound velocity Note the result is independent of viscosity.Again the behavior follows from a Tauberian theorem.
The admittance tensor in any geometry can be expressed as
0 0
( , ) ω =
t( ) 1 ω ⎡ ⎣ + A ( ) ( ) ω C ω
a( , ) ω ⎤ ⎦
y r t y F r t
bulk Faxén type coefficents calculated by Bedeaux, Mazur 1974
( , )0
a
ω
F rt
is the reaction field tensor, depends on geometry.
In point approximation
0
0 0 0 0
( , )
ω
lim[ ( , ) ( )]= → − −
r r
F rt G r r G r r
Bulk Green function is known.
I have calculated for compressible viscous fluid between two planes.
At this gives results mentioned earlier:
0
( −
0, ) ω G r r
( , , )
0ω G r r
ω = 0
(0)
0[1 6 (0)]
xx
aF
xxµ = µ + πη µ
zz(0) = µ
0[1 6 + πη aF
zz(0)]
( , ) 1 [1 1.004 ]
xx 6 h L a
a h
µ
=πη
− ath = L / 2
Faxén 1925( , ) 1 [1 1.452 ]
zz 6 h L a
a h
µ = πη − at
h = L / 2
BUF 20057
Tauberian theorem is applied to the low frequency behavior
2 2
2 2 2
0 5
1 2 ( )
( , , ) 36 ln ( )
4 3
xx
h L h
F h L X h h O
h L
ω α ξ α α α
πη
⎡ − ⎤
= ⎢ + − + ⎥
⎣ ⎦
iωρ
α η
= −
Here is given by a complicated integral over wavenumber q, coming from Fourier expansion in the xy-plane.
This gives the steady-state results.
The next term leads to cancellation of the bulk t-3/2 tail.
The mathematical origin of this term is already quite subtle.
Usually the term linear in comes from an integral over wavenumber of the form
X0
2 3
α
hω
2 2
2 2
0
( ) gq q
f e dq
α q
α
∞
= −
∫
+cutoff for large q diffusion pole
Expansion in powers of yields
α
1 2
( ) ( )
2 2
f O
g
α
=π π
−α
+α
independent of the cutoff, such a term gives rise to the bulk t-3/2 tail.
Instead the term comes from a branch cut in the complex q-plane, rather than a simple pole.
In the last term
1
2( ) exp[ ] [ ]
2 2
f g erfc g
g
α = π π − α α α
2 3
α
h0 0
c c
η ν
ξ = ρ = is the acoustic damping length.
8 The singularity comes from an acoustic diffusion pole
(overdamped sound wave), but with weight q rather than q2
2 ln
α α
The corresponding diffusion coefficient is
The xx element of the reaction field tensor can be expressed as
2 2 0
12 D c L
= ν
0
( , , ) 1 ( , )
xx 4 x
F h L
ω
f qω
q dqπη
∞
=
∫
The function f qx
( , ) ω
behaves for smallq
and asω
2 2 2 2 2 2 2 2
3 2 2 2 2
2 2 2
2 ( ) 36
( , )
2 12
x
q q q h L h
f q q L q L
α α α ξ
ω α α α ξ
+ − − −
≈ +
+ +
branch cut same as for single wall diffusion pole In the pole term we use the integral
2 2 2
2 2 1
0
exp[ ] 1 ( )
2 q g
gq dq e E g q
α α
α
∞
− =
∫
+Expansion yields the term, and this gives the t-2 tail.
Define relaxation functions and from
2ln
α α
9
xx
( ) t
γ γ
zz( ) t
( ) ( )
xx xx
p
C t kT t
m
γ
=
zz( ) zz( )p
C t kT t
m γ
=
(0) 1
γ
xx=
γzz(0)=10 5 10 15 20 25 30
0 0.01 0.02 0.03
xx( )t
γ
t
10
-3 -2 -1 0 1 2
-12 -10 -8 -6 -4 -2 0
ln | γ
xx( ) | t
log t
10-3 -2 -1 0 1 2
-12 -10 -8 -6 -4 -2 0