THE JOURNAL OF CHEMICAL PHYSICS 126, 134706 共2007兲

Diffusion in a tube of varying cross section: Numerical study of reduction

to effective one-dimensional description
A. M. Berezhkovskii
Mathematical and Statistical Computing Laboratory, Division of Computational Bioscience,
Center for Information Technology, National Institutes of Health, Bethesda, Maryland 20892
M. A. Pustovoit
St. Petersburg Nuclear Physics Institute, Gatchina 188300, Russia
S. M. Bezrukov
Laboratory of Physical and Structural Biology, National Institute of Child Health and Human Development,
National Institutes of Health, Bethesda, Maryland 20892
共Received 9 January 2007; accepted 1 March 2007; published online 4 April 2007兲

Brownian dynamics simulations of the particle diffusing in a long conical tube 共the length of the
tube is much greater than its smallest radius兲 are used to study reduction of the three-dimensional
diffusion in tubes of varying cross section to an effective one-dimensional description. The authors
find that the one-dimensional description in the form of the Fick-Jacobs equation with a
position-dependent diffusion coefficient, D共x兲, suggested by Zwanzig 关J. Phys. Chem. 96, 3926
共1992兲兴, with D共x兲 given by the Reguera-Rubí formula 关Phys. Rev. E 64, 061106 共2001兲兴, D共x兲
= D / 冑1 + R⬘共x兲2, where D is the particle diffusion coefficient in the absence of constraints, and R共x兲
is the tube radius at x, is valid when 兩R⬘共x兲兩 艋 1. When 兩R⬘共x兲兩 ⬎ 1, higher spatial derivatives of the
one-dimensional concentration in the effective diffusion equation cannot be neglected anymore as
was indicated by Kalinay and Percus 关J. Chem. Phys. 122, 204701 共2005兲兴. Thus the reduction to
the effective one-dimensional description is a useful tool only when 兩R⬘共x兲兩 艋 1 since in this case one
can apply the powerful standard methods to analyze the resulting diffusion equation. © 2007
American Institute of Physics. 关DOI: 10.1063/1.2719193兴

I. INTRODUCTION sumes that the distribution of the solute in any cross section
of the tube is uniform as it is at equilibrium. The point is that
The problem of diffusion in a tube of varying cross sec- this assumption allows one to reduce the three-dimensional
tion arises in different contexts. Examples include diffusion problem with a complex boundary to the one-dimensional
of ions and macromolecular solutes through the channels in problem of diffusion along the tube axis in the presence of an
biological membranes,1 transport in zeolites2 and nanostruc- entropy potential.8
tures of complex geometry,3 controlled drug release,4 and Directing the x axis along the centerline of the tube and
diffusion in man-made periodic porous materials.5 It is intu- denoting the cross-section area of the tube at x by A共x兲 one
itively appealing to formulate the problem as one dimen- can introduce the effective one-dimensional concentration of
sional, i.e., in terms of the effective one-dimensional concen- the solute, c共x , t兲,
tration of diffusing molecules that satisfies a one-
dimensional diffusion equation. However, reduction of the
three-dimensional diffusion equation with reflecting bound-
c共x,t兲 = 冕
A共x兲
C共x,y,z,t兲dydz. 共1.2兲
ary condition on the wall of the tube to the effective one-
dimensional one is a tricky problem, which has been dis- When distributions in the cross sections are uniform, this
cussed in the literature for a long time.6,7 Real progress in concentration satisfies the Fick-Jacobs 共FJ兲 equation6
understanding this reduction has been made in recent papers
by Zwanzig,8 Reguera and Rubí,9 and Kalinay and
Percus.10–13
⳵c共x,t兲
⳵t
=D
⳵
⳵x
再 冋 册冎
A共x兲
⳵ c共x,t兲
⳵x A共x兲
, 共1.3兲

The present paper deals with diffusion in a cylindrical which is the Smoluchowski equation for diffusion in the en-
tube of varying cross section. Local concentration of diffus- tropy potential U共x兲 defined as8
ing solute molecules, C共x , y , z , t兲, satisfies the diffusion equa-
A共x兲
tion U共x兲 = − kBT ln , 共1.4兲
A共x0兲
⳵C共x,y,z,t兲
⳵t
=D
⳵
⳵x
冉2 +
2
⳵
⳵y 2 +
2
⳵
⳵z2
2
冊
C共x,y,z,t兲, 共1.1兲 where kB and T are the Boltzmann constant and the absolute
temperature, and U共x兲 at x = x0 is taken to be zero.
where D is the solute diffusion coefficient in space with no Reduction to the one-dimensional description based on
constraints, with reflecting boundary condition on the wall of the local equilibrium assumption is obviously oversimplified.
the tube. The description dramatically simplifies if one as- Zwanzig derived a corrected form of the FJ equation in

0021-9606/2007/126共13兲/134706/5/$23.00 126, 134706-1 © 2007 American Institute of Physics

Downloaded 05 Apr 2007 to 85.142.10.65. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
134706-2 Berezhkovskii, Pustovoit, and Bezrukov J. Chem. Phys. 126, 134706 共2007兲

which small deviations from local equilibrium are taken into
account.8 Assuming that the tube radius, R共x兲, does not
change too fast, i.e., 兩R⬘共x兲兩 Ⰶ 1, he showed that c共x , t兲 satis-
fies the conservation probability equation
⳵c共x,t兲 ⳵ j共x,t兲
=− , 共1.5兲
⳵t ⳵x
in which the flux, j共x , t兲, is given by
justified when 兩R⬘共x兲兩 Ⰶ 1 共in spite of the fact that ␧ = 1兲 since
Zwanzig’s result gives the leading correction due to the de-
viation from local equilibrium in this limiting case.
In Ref. 13 Kalinay and Percus consider the stationary
flux through a long tube of varying cross section at fixed
concentrations of the molecules at the tube ends. They show
that the general relation between the stationary flux jst and
the stationary concentration cst共x兲,

j共x,t兲 = − A共x兲D共x兲 冋 册
⳵ c共x,t兲
⳵x A共x兲
. 共1.6兲 jst = − A共x兲D̂ x,冉 冊 冋 册
⳵ ⳵ cst共x兲
⳵x ⳵x A共x兲
, 共1.12兲

The expression for the position-dependent effective diffusion dramatically simplifies. The infinite sum of the derivatives of
coefficient, D共x兲, derived by Zwanzig, has the form cst共x兲 / A共x兲 can be summed up and the stationary flux can be

冋 册
written in the conventional form

冋 册
1 D
DZw共x兲 = D 1 − R⬘共x兲2 ⯝ . 共1.7兲 ⳵ cst共x兲
2 1 + 共1/2兲R⬘共x兲2 jst = − A共x兲Dst共x兲 . 共1.13兲
⳵x A共x兲
Later Reguera and Rubí generalized Zwanzig’s result. Based
on heuristic arguments they suggested that D共x兲 entering into The effective position-dependent diffusion coefficient Dst共x兲
Eq. 共1.6兲 is given by9 can be found solving the equation

DR−R共x兲 =
D
冑1 + R⬘共x兲 2
. 共1.8兲 冉 冊冋
A共x兲D̂ x,
⳵
⳵x
1
A共x兲Dst共x兲
册
= 1. 共1.14兲

The approximate expression for the flux in Eq. 共1.6兲 con- Introducing the inverse diffusivity operator, D̂共x , ⳵ / ⳵x兲−1,
tains only the first spatial derivative of c共x , t兲. However, one can write the solution to Eq. 共1.14兲 as
based on general ideas one might expect that the exact ex-
pression for the flux obtained by reduction to the one-
dimensional description must contain all derivatives,
1
Dst共x兲
= A共x兲D̂ x,冉 冊冋 册
⳵
⳵x
−1
1
A共x兲
. 共1.15兲

⳵kc共x , t兲 / ⳵xk, k = 1 , 2 , . . .. Such an expression has been ob- Kalinay and Percus show that for a long conical tube
tained by Kalinay and Percus.10–13 Based on their analysis with R共x兲 = R共xL兲 + ␭共x − xL兲, where R共xL兲 is the tube radius at
we can write the flux as x = xL and ␭ = R⬘共x兲 is a constant, Dst共x兲 = const, given by the

冉 冊 冋 册
formula suggested by Reguera and Rubí, Eq. 共1.8兲,
⳵ ⳵ c共x,t兲
j共x,t兲 = − A共x兲D̂ x, , 共1.9兲 D
⳵x ⳵x A共x兲 Dst共x兲 = 共1.16兲
冑1 + ␭2 .
where we have introduced the operator D̂共x , ⳵ / ⳵x兲 which we
will call the diffusivity operator. Note that a conical tube may be considered as long when its
To write an expression for this operator we first assume length is greater than variation of its radius, i.e., when ␭
that diffusion in the tube is highly anisotropic: it occurs ⬍ 1.
much slower along the tube axis than in the normal direction Reduction to the effective one-dimensional diffusion
so that corresponding diffusion coefficients, Dx and D⬜, sat- equation is a useful tool to analyze diffusion in a tube of
isfy ␧ = Dx / D⬜ Ⰶ 1. Then the diffusivity operator can be writ- varying cross section only if this equation is not too compli-
ten as cated. In this respect the conventional form of the one-

冋 册
dimensional diffusion equation, Eqs. 共1.5兲 and 共1.6兲, has an
冉冏 冏冊
⬁
⳵ ⳵k important advantage over the generalized form, Eqs. 共1.5兲,
D̂ x, ␧ = Dx 1 − 兺 ␧k+1␨k共x,␧兲 k , 共1.10兲
⳵x k=0 ⳵x 共1.9兲, and 共1.10兲. The point is that powerful techniques have
been developed to analyze the conventional diffusion equa-
where functions ␨k共x , ␧兲 are also expressed as Taylor series in tion, while there are no standard methods of analysis of the
␧. The first three terms of the ␧ expansion of D̂共x , ⳵ / ⳵x 兩 ␧兲 generalized version. The purpose of this study is to establish
are13 the range of applicability of the reduction to the conventional

冉冏 冏冊
D̂ x,
⳵
⳵x
再␧
␧ = D x 1 − R ⬘2 −
2
␧ 2R ⬘ 2
24
冋
R R⵮ + RR⬘R⬙
form of the diffusion equation and to indicate geometrical
constraints under which such a reduction is justified. It seems
natural to formulate the constraints in terms of R⬘共x兲. We will

− 7R⬘3 + R共RR⬙ + R⬘2兲

⳵
⳵x
册冎 . 共1.11兲
see that although initially Zwanzig8 showed that the reduc-
tion is justified only when 兩R⬘共x兲兩 Ⰶ 1, in fact, the range of its
applicability is much broader, 兩R⬘共x兲兩 艋 1. Understanding
The diffusivity operator entering into Eq. 共1.9兲 is the opera- these constraints seems important for potential applications
tor in Eq. 共1.10兲 with Dx = D and ␧ = 1. The diffusivity opera- of the effective one-dimensional diffusion equation, for ex-
tor reduces to DZw共x兲 in Eq. 共1.7兲 if one approximates the ample, in studies of diffusion in quasi-one-dimensional peri-
infinite sum in Eq. 共1.10兲 by the first two terms. This is odic porous structures.14

Downloaded 05 Apr 2007 to 85.142.10.65. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
134706-3 Diffusion in a tube
FIG. 1. Conical tube used in our Brownian dynamics simulations. The local
radius of the tube, R共x兲, is given by Eq. 共2.1兲.
In the present paper we report on our numerical study of
the reduction to the effective one-dimensional description us-
ing Brownian dynamics simulations. Our goal is to under-
stand 共i兲 under which conditions one can neglect higher spa-
tial derivatives of c共x , t兲 and use conventional expression for
the flux given in Eq. 共1.6兲 and 共ii兲 the relation between the
effective diffusion coefficient found numerically and those
given by the Zwanzig and Reguera-Rubí formulas, Eqs. 共1.7兲
and 共1.8兲. Our main results are as follows. One can use the
conventional expression for the flux in Eq. 共1.6兲 with D共x兲
given in Eq. 共1.8兲 when 兩R⬘共x兲兩 艋 1. For larger values of
兩R⬘共x兲兩 higher spatial derivatives of c共x , t兲 cannot be ne-
glected. Note that even when 兩R⬘共x兲兩 = 1 Eqs. 共1.7兲 and 共1.8兲
lead to close values of D共x兲: DZw共x兲 ⬵ 0.66D, DR−R共x兲
⬵ 0.71D, which are not much less than D.
II. RESULTS OF NUMERICAL STUDY
To study the reduction to the effective one-dimensional
description we run Brownian dynamics simulations in the
long conical tube of length L shown in Fig. 1. The tube
radius, R共x兲, is given by
R共x兲 = 1 + ␭x, 0 ⬍ x ⬍ L,
where we have chosen the radius of the narrow end of the
tube as a unit length and ␭ = R⬘共x兲 is a positive constant, ␭
艌 0. A cylindrical tube of unit radius corresponds to ␭ = 0.
Particle trajectories start from one end of the tube, which, as
well as the wall of the tube, is a perfectly reflecting bound-
ary, and are terminated at their first contact with the opposite
end, which is a perfectly absorbing boundary. In simulations
we find the mean first passage times from one end of the tube
to the other, ␶␭共n → w兲 and ␶␭共w → n兲, where n and w denote
the narrow and wide ends of the tube, as functions of ␭ for
L = 20. When running simulations we take D = 1 and the time
step ⌬t = 2 ⫻ 10−4, so that 冑2D⌬t = 2 ⫻ 10−2 Ⰶ 1. Each mean
first passage time is obtained by averaging the first passage
times of 104 trajectories whose starting points are uniformly
distributed over the reflecting end of the tube. To estimate the
accuracy of our numerical results we run 105 trajectories in
the cylindrical tube 共␭ = 0兲 and in the conical tube with ␭
= 2. We divide the entire set of trajectories into the subsets of
104 trajectories and determine the mean first passage time for
each subset. For the cylindrical tube these times are com-
pared with the exact value, L2 / 共2D兲, while for the conical
tube the times are compared with the mean first passage time
Downloaded 05 Apr 2007 to 85.142.10.65. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
J. Chem. Phys. 126, 134706 共2007兲
found by averaging the times of all 105 trajectories. The
comparison shows that the relative error of the first passage
times found in our simulations is less than 2% in both cases.
The mean first passage times found in simulations are
used to determine the effective diffusion coefficients, D␭共n
→ w兲 and D␭共w → n兲, assuming that the flux entering into Eq.
共1.5兲 is given by the conventional expression in Eq. 共1.6兲. We
chose the conical geometry of the tube because for this ge-
ometry R⬘共x兲 = ␭ = const and, as a consequence, the effective
diffusion coefficient, D共x兲, in Eq. 共1.6兲 is constant. The fact
that the diffusion coefficient is independent of x allows us to
use standard simple expressions for the first passage times,
L2 3 + ␭L
␶␭共n → w兲 = 共2.2兲
6D␭共n → w兲 1 + ␭L
and
L2
␶␭共w → n兲 = 共3 + 2␭L兲. 共2.3兲
6D␭共w → n兲
Respectively, the diffusion coefficients for transitions in both
directions are
L2 3 + ␭L
D␭共n → w兲 = 共2.4兲
6␶␭共n → w兲 1 + ␭L
and
L2
D␭共w → n兲 = 共3 + 2␭L兲. 共2.5兲
6␶␭共w → n兲
The ratio D␭共w → n兲 / D␭共n → w兲, determined from our simu-
lations of the first passage times according to Eqs. 共2.4兲 and
共2.5兲, may be considered as an indicator whether the conven-
tional expression for the flux is applicable or not. When the
expression is applicable the two diffusion coefficients are
equal and their ratio must be unity. Deviation of the ratio
from unity indicates that the conventional expression for the
共2.1兲
flux is inapplicable, and higher spatial derivatives of c共x , t兲
cannot be neglected.
The results of our simulations are presented in Fig. 2
which shows the ratio of ␶␭共n → w兲 and ␶␭共w → n兲 to the
corresponding mean first passage time in the cylindrical tube
of uniform cross section given by L2 / 共2D兲. The squares rep-
resent our numerical results while solid curves show the de-
pendences obtained on the basis of the Fick-Jacobs equation,
Eq. 共1.3兲, and its modified version, Eqs. 共1.5兲 and 共1.6兲, with
D共x兲 given in Eqs. 共1.7兲 and 共1.8兲, as indicated by letters near
the curves. As might be expected ␶␭共w → n兲 monotonously
increases with ␭ 关see Fig. 2共a兲兴. This happens because both
the entropic repulsion and slowdown of diffusion lead to the
increase of ␶␭共w → n兲 with ␭. Figure 2共a兲 shows that apply-
ing the Reguera-Rubí formula in Eq. 共1.8兲 one can predict
variation of ␶␭共w → n兲 over a broad range of times.
In contrast to the monotonic growth of ␶␭共w → n兲 with ␭,
the ␭ dependence of ␶␭共n → w兲 is nonmonotonic. This can be
understood if one considers the effect of the entropy poten-
tial which pulls the particle towards the wider end that leads
to the decrease of ␶␭共n → w兲 at small ␭. In the opposite lim-
iting case, ␭ → ⬁, the problem reduces to that of one-
dimensional diffusion. For this reason, ␶␭共n → w兲 returns to
134706-4 Berezhkovskii, Pustovoit, and Bezrukov J. Chem. Phys. 126, 134706 共2007兲

FIG. 3. The ratio of the diffusion coefficients, D␭共n → w兲 / D␭共w → n兲, as a

function of ␭.

expression for the flux, in Fig. 3 we show the ratio of the

diffusion coefficients obtained by means of Eqs. 共2.4兲 and
共2.5兲 using the mean first passage times found in simulation.
One can see that the ratio monotonically increases with ␭.
For ␭ = 1 the ratio is approximately 1.09. We consider ␭ = 1
as the upper boundary for the range of applicability of the
conventional expression for the flux with the Reguera-Rubí
formula for the diffusion coefficient. In Fig. 4 we show the
ratios of the two diffusion coefficients to DR−R. This figure
shows that D␭共n → w兲 deviates from DR−R much stronger
than D␭共w → n兲. Nevertheless, the relative deviation does not
exceed 10% for ␭ 艋 1.
FIG. 2. The ratios of the mean first passage times between the two ends of In summary, our numerical study of diffusion in a long
the conical tube, ␶␭共w → n兲 and ␶␭共n → w兲, to the mean first passage time in conical tube 共the length of the tube is much greater than its
a tube of uniform cross section, L2 / 共2D兲, as functions of ␭ 关panels 共a兲 and
共b兲, respectively; see the details in the text兴. smallest radius兲 of varying cross section has shown that the
reduction to the effective one-dimensional description is jus-
tified when 兩R⬘共x兲兩 艋 1, where R共x兲 is the tube radius at x.
its value for the cylindrical tube of uniform cross section that
When this condition is fulfilled, one can use the conventional
corresponds to ␭ = 0. Equation 共2.2兲 shows that the estima-
expression for the flux, Eq. 共1.6兲, with D共x兲 given in Eq.
tion based on the Fick-Jacobs equation, which neglects slow-
共1.8兲. Such a reduction provides significant simplification of
down of diffusion with ␭, leads to the monotonic decrease of
the analysis of diffusion in periodic porous structures dis-
the ratio 2D␶␭共n → w兲 / L2 from unity at ␭ = 0 to 1 / 3 as ␭
→ ⬁ 关Fig. 2共b兲兴. The slowdown of diffusion leads to the in-
crease of ␶␭共n → w兲. Competition between the decrease of
this time with ␭ due to the entropy potential and its increase
with ␭ due to the slowdown of diffusion determines the non-
monotonic behavior of the ratio 2D␶␭共n → w兲 / L2 shown in
Fig. 2共b兲.
From this figure one can see that the prediction based on
the conventional expression for the flux with D共x兲 given by
the Reguera-Rubí formula is in better agreement with the
numerical results than the two other predictions shown in the
figure. One can also see that even the best of the three pre-
dictions fails at ␭ ⬎ 1. The point is that Eq. 共2.2兲 with D共x兲
given in Eq. 共1.8兲 leads to incorrect asymptotic behavior of
␶␭共n → w兲 as ␭ → ⬁. As we discussed earlier, in this limiting
case ␶␭共n → w兲 tends to its value in the tube of uniform cross
section, L2 / 共2D兲, that corresponds to ␭ = 0, while Eq. 共2.2兲
predicts the linear growth of ␶␭共n → w兲 with ␭ at large ␭, FIG. 4. The ratios of the diffusion coefficients, D␭共n → w兲 and D␭共w → n兲 to
␶␭共n → w兲 ⯝ ␭L2 / 共6D兲 as ␭ → ⬁. DR−R = D / 冑1 + ␭2, as functions of ␭. Squares and diamonds represent D␭共n
To evaluate the range of applicability of the conventional → w兲 / DR−R and D␭共w → n兲 / DR−R, respectively.

Downloaded 05 Apr 2007 to 85.142.10.65. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
134706-5 Diffusion in a tube
cussed recently14 on the basis of the modified Fick-Jacobs
equation. When this approach fails, diffusion in periodic po-
rous structures can be analyzed using an alternative
approach,15 which in a sense is complementary to the first
one.
ACKNOWLEDGMENTS
The authors are grateful to Miguel Rubí, Attila Szabo,
and Bob Zwanzig for helpful discussions of the problem and
related issues. This study was supported by the Intramural
Research Program of the NIH, Center for Information Tech-
nology, and National Institute of Child Health and Human
Development. One of the authors 共M.A.P.兲 also thanks the
Russian Foundation for Basic Research 共Project No. 05-02-
17626兲, the Scientific Council “Superconductivity,” and the
State programs “Quantum Macrophysics,” and “Strong Cor-
related Electrons in Semiconductors, Metals, Superconduct-
ors and Magnetic Materials” for partial support.
Downloaded 05 Apr 2007 to 85.142.10.65. Redistribution subject to AIP license or copyright, see http://jcp.aip.org/jcp/copyright.jsp
J. Chem. Phys. 126, 134706 共2007兲
1
B. Hille, Ion Channels in Excitable Membranes 共Sinauer Associates, Sun-
derland, MA, 2001兲.
2
J. Kärger and D. M. Ruthven, Diffusion in Zeolites and Other Mi-
croporous Solids 共Wiley, New York, 1992兲.
3
N. F. Sheppard, D. J. Mears, and S. W. Straks, J. Controlled Release 42,
15 共1996兲.
4
J. T. Santini, Jr., M. J. Cima, and R. Langer, Nature 共London兲 397, 335
共1999兲; R. A. Siegel, J. Controlled Release 69, 109 共2000兲.
5
S. Matthlas and F. Müller, Nature 共London兲 424, 53 共2003兲.
6
M. H. Jacobs, Diffusion Processes 共Springer, New York, 1967兲.
7
E. E. Petersen, AIChE J. 4, 343 共1958兲; A. S. Michaels, ibid. 5, 270
共1959兲; C. S. Patlak, Bull. Math. Biol. 35, 81 共1973兲; B. Ballal and K.
Zygourakis, Chem. Eng. Sci. 40, 1477 共1985兲.
8
R. Zwanzig, J. Phys. Chem. 96, 3926 共1992兲.
9
D. Reguera and J. M. Rubí, Phys. Rev. E 64, 061106 共2001兲.
10
P. Kalinay and J. K. Percus, J. Chem. Phys. 122, 204701 共2005兲.
11
P. Kalinay and J. K. Percus, Phys. Rev. E 72, 061203 共2005兲.
12
P. Kalinay and J. K. Percus, J. Stat. Phys. 123, 1059 共2006兲.
13
P. Kalinay and J. K. Percus, Phys. Rev. E 74, 041203 共2006兲.
14
D. Reguera, G. Schmid, P. S. Burada, J. M. Rubí, P. Reimann, and P.
Hänggi, Phys. Rev. Lett. 96, 130603 共2006兲; B. Q. Ai and L. G. Liu,
Phys. Rev. E 74, 051114 共2006兲.
15
A. M. Berezhkovskii, V. Yu. Zitserman, and S. Y. Shvartsman, J. Chem.
Phys. 118, 7146 共2003兲; 119, 6991 共2003兲; O. K. Dudko, A. M. Berezh-
kovskii, and G. H. Weiss, J. Phys. Chem. B 109, 21296 共2005兲.