Journal of Chromatography A, 1126 (2006) 70–85

Review

Transition from creeping via viscous-inertial to turbulent

flow in fixed beds
Dzmitry Hlushkou, Ulrich Tallarek ∗
Institut für Verfahrenstechnik, Otto-von-Guericke-Universität Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany
Available online 27 June 2006

This work is dedicated to Prof. Dr. Georges Guiochon on the occasion of his 75th birthday. In particular, U.T. wants to thank him for his prominent role as
scientific advisor and mentor, as well as for his friendship over the past decade.

Abstract
This review is concerned with the analysis of flow regimes in porous media, in particular, in fixed beds of spherical particles used as reactors
in engineering applications, or as separation units in liquid chromatography. A transition from creeping via viscous-inertial to turbulent flow is
discussed based on macro-scale transport behaviour with respect to the pressure drop–flow rate dependence, in particular, the deviation from
Darcy’s law, as well as direct microscopic data which reflect concomitant changes in the pore-level hydrodynamics. In contrast to the flow
behaviour in straight pipes, the transition from laminar to turbulent flow in fixed particulate beds is not sharp, but proceeds gradually through
a viscous-inertial flow regime. The onset of this steady, nonlinear regime and increasing role of inertial forces is macroscopically manifested
in the failure of Darcy’s law to describe flow through fixed beds at higher Reynolds numbers. While the physical reasons for this failure still
are not completely understood, it is not caused by turbulence which occurs at Reynolds numbers about two orders of magnitude above those
for which a deviation from Darcy’s law is observed. Microscopic analysis shows that this steady, nonlinear flow regime is characterized by the
development of an inertial core in the pore-level profile, i.e., at increasing Reynolds number velocity profiles in individual pores become flatter
towards the center of the pores, while the velocity gradient increases close to the solid–liquid interface. Further, regions with local backflow and
stationary eddies are demonstrated for the laminar flow regime in fixed beds. The onset of local fluctuations (end of laminar regime) is observed
at superficial Reynolds numbers on the order of 100. Complementary analysis of hydrodynamic dispersion suggests that this unsteady flow
accelerates lateral equilibration between different velocities in fixed beds which, in turn, reduces spreading in the longitudial (macroscopic flow)
direction.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Porous media; Fixed beds; Sphere packings; Darcy’s law; Flow regimes; Inertial flow; Turbulent flow; Velocity profiles; Critical Reynolds number;
Hydrodynamic dispersion; Zone spreading; Lateral equilibration; Liquid chromatography; Turbulent flow chromatography; Review

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2. Fluid dynamics at the macroscopic scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3. Fluid dynamics at the microscopic scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4. Flow regimes and longitudinal dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5. Flow regimes and liquid chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

∗ Corresponding author. Tel.: +49 391 61 10465; fax: +49 391 67 12028.
E-mail address: ulrich.tallarek@vst.uni-magdeburg.de (U. Tallarek).

0021-9673/$ – see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.chroma.2006.06.011
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
1. Introduction
Turbulence is one of the most important and, at the same time,
yet least understood problem in fluid dynamics. It is known
that all flows of liquids can be divided into two strictly dif-
ferent types: those referred to as laminar flows in which the
fluid moves in layers or laminas gliding smoothly over adja-
cent ones with only molecular interchange of momentum, and
their opposite, turbulent flows in which all fluid mechanical
properties (velocity, pressure, temperature, etc.) fluctuate with
extremely irregular spatio-temporal pattern. This complicated
structure of turbulent flows affects many properties of liquid
transport which differ substantially in the laminar and turbulent
cases. The difference between laminar and turbulent regimes is
revealed in a number of phenomena which are of great signifi-
cance for many engineering problems. For instance, the presence
of irregular fluctuations of the fluid velocity in turbulent flows
leads to a sharp increase in mixing of the fluid, often con-
sidered as the most characteristic feature of turbulent motion.
In turn, due to a far more efficient radial mixing, the veloc-
ity profile in turbulent flow is considerably more uniform than
in laminar flow. Based on this fact Giddings in 1965 [1] sug-
gested that “velocity equalization” in turbulent flow through an
open capillary is advantageous for chromatographic separation
efficiency which he verified experimentally in 1966 [2]. Subse-
quently, turbulent flow has been studied in gas chromatography
[3–6], but did not demonstrate great potential for retained solutes
[7].
Though the existence of two sharply different flow regimes
was pointed out in the first half of the 19th century [8], the
first theoretical approach to studying turbulence came only with
the pioneering works of Osborne Reynolds published in 1883
and 1895 [9,10]. In these studies Reynolds focused his atten-
tion, in particular, on the conditions under which laminar flow
of fluid in pipes is transformed into a turbulent one and pro-
posed a general criterion for dynamic similarity of flows of a
viscous incompressible fluid in geometrically similar systems.
Dynamic similarity of flows was proposed to be identified with
the coincidence of the Reynolds number, Re = ud/ν, calculated
for pipes, where u is the characteristic scale of velocity, d the
diameter of the pipe, and ν is the kinematic viscosity of the
fluid. The Reynolds number can be also viewed as a parameter
expressing the ratio between inertial forces (related to accel-
eration or deceleration of fluid) and viscous (frictional) forces
acting within the fluid. Viscous forces assist in the equalization
of velocities at neighbouring points, i.e., in smoothing out small-
scale heterogeneities in the flow. By contrast, inertial forces
producing mixing of different fluid volumes which move with
different velocities result in a transfer of energy from large- to
small-scale components and, as a consequence, assist in the for-
mation in the flow of heterogeneities characterizing turbulent
flow.
One of the fundamental results formulated in Reynolds’ work
is that flow will be laminar as long as Re does not exceed a crit-
ical value, while for Re > Recrit it will be turbulent. The first
experiments for verifying this criterion and actually measuring
Recrit were conducted by Reynolds himself. In these experi-
71
ments, coloured water was introduced along the axis of a tube
(glass pipe) at a smooth inlet connected to a reservoir of pure
water. Variation of Re was realized by changing the pipe diam-
eter, flow velocity, and viscosity of the water (via temperature).
For small values of Re, the coloured water took the form of a
thin jet indicating laminar flow. As Re increased, at the instant
of passing through its critical value, the form of the coloured
jet suddenly changed. At a rather small distance from the inlet
towards the pipe, the jet spread out and waves appeared in it.
Further on, separate eddies were formed, and towards the end
of the pipe the whole liquid became coloured. It was also found
that the value of Recrit corresponding to the transition from lam-
inar to turbulent flow depends on the degree of disturbance in
the fluid while entering the pipe. Recrit is the smaller the greater
the intensity of the disturbance. Subsequent experimental stud-
ies on flow regimes in pipes and tubes have shown that Re in
itself is not a unique demarcation criterion for transition to tur-
bulence in open channels [11–13]. For instance, in the case of a
tube with a sharp entrance, pushed through the plane wall of the
reservoir, the end of the tube creates a significant disturbance
and Recrit is about 2800. By contrast, onset of turbulence in flow
through a straight pipe can be delayed up to Recrit of 105 [13]
if special means are employed for damping flow disturbances at
the entrance of the tube.
Though the characterization of liquid transport occurring
within porous media is of fundamental importance for under-
standing numerous processes in natural and engineering sci-
ences, the existence of turbulent flow in porous media has been
addressed experimentally only in the second half of the 20th
century [14–18]. As porous media used in engineering applica-
tions and modern liquid chromatography like particulate fixed
beds [19] often have small interstitial pores and low hydraulic
permeabilities, and because fluid velocity is relatively small,
the predominant regime is the laminar flow regime. Mean-
while, high-speed flow may lead to turbulent flow, that is, highly
unsteady chaotic flow within the interstitial pore space. In packed
beds, viscous and inertial loss terms are additive, an effect com-
pletely at odds with the observed flow behaviour in straight pipes.
Evidently, something other than simple laminar and turbulent
flow takes place in fixed beds. Compared to pipe flow the repre-
sentative dimension of the largest flow eddy in a porous medium
is limited by the pore dimension, usually much smaller than the
macroscopic dimension of the system (pipe or column diame-
ter). Investigations of flow through porous media [20–27] allow
to conclude that (i) the transition from laminar to turbulent flow
is gradual and (ii) Recrit based on average interstitial pore dimen-
sions is several times lower than for flow through a straight pipe.
Another fundamental characteristic of turbulent flow in
porous media is the distinction between microscopic and macro-
scopic turbulence. The only experimental evidence of turbulence
in a three-dimensional porous medium is that of microscopic
turbulence where turbulent flow was detected by point-wise
probes within the pores of the medium, as pointed out by Antohe
and Lage [28]; there has been no attempt to volume-average
local signals in a representative volume to obtain a signature
of macroscopic turbulence. It is possible that by averaging a
large number of local (random) signals in a representative ele-
72 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
mentary volume the microscopic turbulence is smoothed out
and, thus, turbulent flow might not endure on a macroscopic
level.
In view of these characteristics, this review of the transition
from creeping via viscous-inertial to turbulent mobile phase
flow in fixed beds is organized as follows. We begin by ana-
lyzing the fluid dynamics in packed, i.e., particulate beds on a
macroscopic scale in view of pressure drop–flow rate behaviour.
The transition from laminar to turbulent flow is discussed in
view of deviations from Darcy’s law at increasing Reynolds
number. Macroscopic analysis is complemented by microscopic
data which, based on various measurement approaches, provide
direct insight into the pore-scale behaviour which contributes
to macro-scale transport. Then, we analyze longitudinal hydro-
dynamic dispersion in fixed beds depending on Péclet number
in order to retrieve effects of the different flow regimes in the
dispersion characteristics. We close by addressing the issue of
turbulent flow in liquid chromatography.
2. Fluid dynamics at the macroscopic scale
Up to the present days one of the commonly used macro-
scopic approaches in describing fluid flow through a homoge-
neous porous medium is to characterize the system in terms of
its hydraulic resistance to flow. In the middle of the 19th cen-
tury, Henry Darcy in his investigations of water flow through
sand filters [29] found that under certain conditions the velocity
was proportional to the pressure drop across the system in the
flow direction. This empirical relationship, known as Darcy’s
law, can be represented in the following form
p µu
= = αu, (1)
L k
where p is the pressure drop over a porous medium of length
L in the macroscopic flow direction, µ the dynamic viscosity
of the fluid, u the superficial velocity (i.e., the mean interstitial
fluid velocity times the interstitial porosity of the medium), and
k is assumed to be a constant depending on porous medium
properties and is called hydraulic conductivity or permeability.
Commonly, one refers to liquid motion satisfying Darcy’s law
as creeping flow or Darcy flow. For particulate packed beds or
fibres another way to express the resistance of the medium to
flow is by the introduction of the friction factor fp
dp p
fp = , (2)
ρu2 L
where dp is the average particle or fibre diameter and ρ is the
density of the fluid.
It has been shown by numerous experimental investigations
that if the flow rate through a porous medium is raised pres-
sure drop becomes no longer proportional to fluid velocity.
Though it is generally admitted that this deviation from Darcy’s
law with increasing fluid velocity is due to a more prominent
role of inertial forces, physical origins of these forces are still
unknown. It should be noted that similar to flow through straight
pipes a dimensionless group characterizing the ratio of iner-
tial and viscous forces, the Reynolds number, can be assigned
also for flow through porous media, Re = ud/ν, but here d refers
to some length dimension of the material. For packed par-
ticulate beds, it is common to employ for d either the mean
particle diameter (dp ) or d = (k/φ)1/2 , where φ is the interstitial
porosity.
Referring to the recent work of Panfilov and Fourar [30] we
address only some of the possible physical origins responsible
for the deviations from Darcy’s law which are discussed in a
number of papers. In particular, they are related to interstitial
pore space curvature [31] and macro-roughness of the pores
[32], the formation of a viscous boundary layer [33], kinetic
energy losses in restrictions and constrictions [34], microscopic
inertial forces manifested in the interstitial drag force [35], the
singularity of streamline patterns and/or microscale flow nonpe-
riodicity [36], and the variation of integral viscous dissipation
due to a deformation of streamline patterns [30]. It should be
mentioned that some authors explain the onset of the deviation
from Darcy’s law with the inhomogeneity of the porous medium
and by a resulting gradual start of turbulence in larger pores
with higher local fluid velocity [37,38]. However, a number of
experimental observations at the pore level, discussed in detail
in the next section of the paper where microscopic aspects of
the flow regimes are considered indicate, following Bear [39],
that “. . . turbulence occurs at Re values at least one order of
magnitude higher than the Re at which deviation from Darcy’s
law is observed.” It must be emphasized that, though inertial
forces dominate in turbulent flow, caution is advised not to iden-
tify a deviation from Darcy’s law with the onset of turbulence;
see also remarks by Nield [40] and Niven [41]. This issue was
carefully pointed out in several textbooks on fluid transport in
porous media [39,42–44].
For instance, as mentioned by Dullien [44], it is a serious
misinterpretation of the phenomenon to attribute the failure of
Darcy’s law to the onset of turbulence. Furthermore, in fol-
lowing Happel and Brenner [43], “. . . failure of Darcy’s law
results when the distortion that occurs in the streamlines owing
to changes in the direction of motion is great enough that iner-
tial forces become significant compared with viscous forces.
The incidence of turbulence will occur at much higher Reynolds
numbers.” The above assertion has been confirmed with exper-
imental studies of fluid flow through porous media, collating
flow velocities at which a deviation from Darcy’s law appears
(as revealed by macroscopic analysis) and those at which turbu-
lence begins (as revealed by microscopic analysis) [17,45–49].
For example, as mentioned by Bear [39], Chauveteau and
Thirriot [47] made experiments in several two-dimensional mod-
els which allowed the visualization of streamlines. For Re < 2
flow is in agreement with Darcy’s law; streamlines remain fixed.
As Re increases, streamlines start to shift and fixed eddies begin
to appear in diverging areas of the model. At Re = 75 turbulence
starts and begins to spread out as Re increases. Turbulence cov-
ers about 50% of the flow domain at Re = 115 and 100% of it
at Re = 180. Deviation from Darcy’s law, on the other hand, is
observed already at Re = 2–3.
Forchheimer in 1901 [50] suggested that flow through a
porous medium can be described, in contrast to linear Darcy’s
law, by a higher order empirical relationship between pressure
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 1. Characterization of different hydrodynamic regimes in fixed beds by
means of pressure drop–flow rate behaviour.
drop and flow rate which can be presented in the form
p µu
= + ρkF u2 = αu + βu2 ,
L k
where kF is the inertial (or non-Darcian) flow coefficient. One
can see that when the flow velocity is very small Eq. (3) sim-
ply reduces to Darcy’s law, while as the fluid velocity increases,
Eq. (3) predicts a deviation from the linear relationship between
pressure drop and flow velocity, and at very high velocities (tur-
bulent regime) the quadratic term on the right-hand side of Eq.
(3) is dominating. In this regime, the dependence of pressure
drop on fluid velocity can be described in good agreement with
experimental data by the Burke–Plummer equation [51]
p
= ρkF u2 = βu2 .
L popular relationship for describing the dependence of pressure
On this basis three regimes of liquid flow through porous media
can be identified via the relative contribution of linear and
nonlinear velocity terms to the overall pressure drop. This is
illustrated by Fig. 1 which differentiates between (i) Darcian
flow (viscous forces dominate and pressure drop is proportional
to the fluid velocity), (ii) turbulent flow (inertial forces dominate;
pressure drop is proportional to the squared fluid velocity) and
(iii) transient, Forchheimer, or nonlinear laminar flow (both vis-
cous and inertial forces affect the fluid dynamics; pressure drop
depends nonlinearly on fluid velocity, but flow is still laminar).
If coefficients α and β in Eq. (3) are known, one can easily
introduce a demarcation parameter which allows to distinguish
between the different flow regimes in a system via the rela-
tive contributions of the linear and quadratic terms in Eq. (3).
Formally, values of the coefficients α and β could be obtained
for a given physical system by solution of the Navier–Stokes
(partial differential) equations relating pressure and velocity
fields, subject to corresponding boundary conditions. However,
the complex geometry of porous media and nonlinearity of the
Navier–Stokes equations make the analytical/numerical solution
of the above problem impossible/very difficult. Therefore, the
main effort of the numerous investigations has been to obtain an
empirical relationship between p/L and u in porous media.
73
In particular, Ergun in 1952 [52] examined the nonlinear rela-
tionship between pressure gradient and fluid velocity. He studied
this phenomenon for gas flow through crushed porous solids
based on its dependence on flow rate and properties of the fluid,
porosity of the medium, as well as on the orientation, size, and
shape of the particles. By generalizing Forchheimer’s equation
Ergun obtained the following empirical equation
p (1 − φ)2 µu (1 − φ) ρu2
=A + B , (5)
L φ3 dp2 φ3 dp
where dp is the average size of the particles used to obtain a
packed bed. After analysis of a large quantity of experimen-
tal data, Ergun concluded that their best representation could
be obtained with A = 150 and B = 1.75. However, in subse-
quent studies these values have been found to vary considerably
[21,53–55]. In particular, after testing the Ergun equation using
many more data than ever before, Macdonald et al. [21] found
that the following values of A and B give the best fits to all of
the involved data: A = 180, and B = 1.8 (smooth particles) or 4.0
(3) (rough particles).
It should be noted that a large number of other empirical equa-
tions describing fluid flow through porous media and its depen-
dence on medium properties have been reported [21,42,56,57].
Specifically, according to Dullien [44], “Much effort has been
expended in order to find the best dp (effective particle or fibre
diameter) and porosity function f(φ) so as to be able to repre-
sent all data points by the same constant. Had these attempts
been successful, there would exist now a general formula for
the permeability.” The work of Rumpf and Gupte [56] suggests
that such general formula does not exist because the value of
k depends on the distribution functions of particle shape and
size, as well as on packing structure. Nevertheless, the most
(4)
drop/friction factor on fluid velocity and packing properties is the
Ergun equation. This can be explained by the fact that it contains
intrinsically the two additive terms corresponding individually
to the Blake–Kozeny equation [58] valid for low fluid veloci-
ties, and the Burke–Plummer equation [51] valid for high fluid
velocities. Fig. 2 illustrates the capability of the Ergun equa-
tion to cover, in good agreement with the experimental data, a
wide range of velocities in fixed particulate beds from creeping
to turbulent flow [52,59]. In addition, Fig. 2 allows to provide
values of Reinter calculated using the particle diameter and inter-
stitial velocity u/φ which can be used to demarcate creeping,
viscous-inertial, and turbulent flow regimes in packed beds hav-
ing different porosities:
Reinter ≤ 1 (creeping or Darcian regime)
1 < Reinter < 500 (nonlinear-laminar
or viscous-inertial regime)
Reinter ≥ 500 (turbulent regime)
More correctly, these values represent a range of Reinter
because a sharp demarcation of the different flow regimes in
74 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 2. Normalized plot of the Ergun equation for flow through packed beds, and
the two related asymptotes, the Blake–Kozeny equation and the Burke–Plummer
equation (after Ergun [52]). From Bird et al. [59], Copyright© 2002, reprinted
with permission of John Wiley & Sons, Inc.
porous media is impossible due to the gradual transition between
them.
To summarize, it has long been known that the linear law of
Darcy fails to describe flow through porous media at higher flow
rates. Although it is generally admitted that this failure is caused
by the increased role of inertial forces even when flow continues
to be laminar, and that turbulence develops at much higher flow
rates, some authors still use the terms “non-Darcian” and “turbu-
lent” interchangeably in a misconceptual manner [60]. Analysis
of the macroscopic properties of the porous medium, such as
the hydraulic permeability or friction factor, can be employed to
demarcate the onset of the viscous-inertial (non-Darcian) regime
for flow through a packed bed at increasing velocity. Although
the physical nature of the deviation from Darcy’s law is still
unclear and may have several reasons, empirical relationships
allow to correlate the pressure drop and average fluid velocity
in porous media, as well as to assign velocity ranges to differ-
ent flow regimes. The detailed macroscopic analysis of mass
and momentum transport inherent to the different flow regimes
is, however, of limited significance without knowing the micro-
scopic structure of the porous medium under consideration.
3. Fluid dynamics at the microscopic scale
During the last 40 years various techniques have been used to
analyse the microscopic pattern of flow through porous media.
Among these techniques are laser-Doppler anemometry [61],
particle image velocimetry [62–65], confocal laser scanning
microscopy [66,67], photoluminescent volumetric imaging [68],
magnetic resonance imaging and dynamic NMR microscopy
[69–76], direct visualization [16,20,22], as well as micro-
electrode and hot wire techniques [15,23,24,26].
Owing to a visualization technique Jolls and Hanratty [16]
studied the motion of a coloured plume in liquid flow through
packed beds of 2.54-cm diameter glass spheres, injected from
the surface of a sphere located 2–3 diameters from the wall of
the column. The column diameter was 30 cm and the interpar-
ticle void volume was 0.41. They observed that the patterns of
streamers remained steady until Resf (Reynolds number defined
with respect to the sphere diameter and superficial flow veloc-
ity) reached a value of 110. As flow velocity increased coloured
filaments started to wave, and the flow became turbulent for
Resf > 300. Flow patterns showed a vigorous eddying motion
which could be seen 4–5 sphere diameters away from the ink
inlet location. The visual studies were complemented by ana-
lyzing the fluctuations of the diffusional current on 0.5-mm
diameter circular nickel probes mounted at different positions
on the surface of one of the particles. The onset of unsteady
flow patterns (at Resf = 110) was accompanied by gentle wav-
ing of the probe signals, with excursions of the order of 2–4%
from the average signal, with a further increase up to 15% for
Resf = 150. Using the same visualization technique Wegner et
al. [20] studied flow through a regular dense array of 7.5-cm
diameter spheres with a void volume of 0.26 (face-centered
cubic structure). They observed a very complex flow scheme
demonstrating regions with locally reversed flow direction. The
transition from steady to highly unsteady flow occurred in the
superficial Reynolds number range of 90–120.
Latifi et al. [23] used a micro-electrode technique for study-
ing the hydrodynamics in columns packed with 5-mm diameter
glass spheres (column diameter and bed void volume, 50 mm
and 0.39, respectively). The onset of fluctuations in the limiting
current was observed for Resf = 110. The maximum of the rate of
fluctuation was found at Resf = 170. An analysis of the slope of
the current fluctuation spectra enabled the authors to conclude
that the flow of liquid had a laminar nature up to Resf = 370.
The same technique was used by Rode et al. [24] for studying
transitions between flow regimes in fixed beds. The onset of fluc-
tuations (unsteady regime) was found to occur at Resf between
110 and 150 (Fig. 3). It is characterized by the onset of a time-
dependent chaotic flow and change in the slope of the local shear
rate versus Reynolds number plots. In the chaotic flow regime,
the liquid seems to form aggregates having dimensions on the
order of the average pore size. The authors pointed out that,
while this is a manifestation of turbulence, it does not mean
that the flow is fully developed turbulent flow [24]. Seguin et al.
[26,77] and Comiti et al. [78] performed similar investigations
on flow regimes for a wide range of flow velocities using two
packed beds (with void volume of 0.36) of 5- and 8-mm diameter
spheres. The authors proposed a global scheme for transitions
between flow regimes in packed beds based on an analysis of
the proportion of the inertial contribution to the total pressure
drop (Fig. 4) determined as a sum of two terms representing
the viscous and inertial contributions (Eq. (17) in Ref. [77]). In
particular, liquid flow characterized by Resf < 2.7 (inertial con-
tribution ≈ 5%) can be considered as Darcian; flow with Resf up
to 120 (inertial contribution ≈ 70%) is laminar; and flow char-
acterized by Resf > 600 (inertial contribution ≈ 90%) represents
the turbulent regime. (We have recalculated Resf from the pore
Reynolds number used by Seguin et al. [26,77] originating in the
capillary model of packed beds proposed by Comiti and Renaud
[79]; cf. Fig. 4.)
Table 1 summarizes the comparatively limited knowledge of
the hydrodynamics beyond the laminar regime in view of values
for Resf corresponding to the onset of local velocity fluctuations
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85 75

Table 1
Superficial Reynolds numbers characterizing the onset of fluctuations and tur-
bulence in fixed beds
Authors Mean Onset of Onset of
porosity of fluctuations (end turbulence
fixed bed of laminar regime)

Jolls and Hanratty 0.41 110–150 300

[16]
Wegner et al. [20] 0.26 90–120 –
Latifi et al. [23] 0.39 110 400
Rode et al. [24] 0.4 110–150 300
Seguin et al. 0.36 120 600
[26,77]
McFarland and 0.48 123–153 –
Dranchuk [37]

and turbulence in liquid flow through fixed beds of spheres.

Fig. 3. Recordings of the diffusional current (reflecting the local velocity insta-
bility) measured by electrochemical probes at different superficial Reynolds
numbers in a fixed bed of glass spheres; x-axis: time (0–1 s), y-axis: voltage
(0–120 mV). Reprinted from Rode et al. [24], Copyright© 1994, with permis-
sion from Elsevier.
Fig. 4. Proportion of inertial stress to the global pressure drop inside packed
beds as a function of the pore Reynolds number. The pore Reynolds number
(Rep ) is based on the mean pore velocity and pore diameter. Reprinted from
Seguin et al. [26], Copyright© 1998, with permission from Elsevier.
As different techniques, as well as particle diameters and bed
porosities have been employed in these studies, these results can
be considered as quite homogeneous for fixed beds of spherical
particles.
Using interstitial velocity measurements by laser anemom-
etry and flow visualization experiments Dybbs and Edwards
[22] also identified four different flow regimes in two porous
media consisting, respectively, of glass spheres and glass rods;
as demarcation parameter they employed a pore Reynolds num-
ber, Repore , based on the average pore size and average pore
velocity: (i) creeping flow for Repore < 1, (ii) steady nonlinear
flow with 1–10 < Repore < 150, (iii) unsteady laminar flow with
150 < Repore < 300 and (iv) chaotic flow at Repore > 300–350. The
steady nonlinear flow regime was characterized by the formation
of a hydrodynamic boundary layer near the solid–liquid interface
and subsequent development of an inertial core in the center of
the pores. However, streaklines were very well defined and indi-
cated a laminar nature of the flow. Dybbs and Edwards concluded
that the development of this inertial core-flow was the reason for
a deviation from the linear relationship between pressure drop
and flow rate [22]. The unsteady laminar flow was characterized
by the onset of laminar wake oscillations which took the form of
travelling waves of different period, amplitude, and growth rate.
Finally, the chaotic flow regime was characterized by break-up
of the dye streaklines and immediate dye dispersion. This regime
resembled turbulent flow. Similar results have been obtained by
Masuoka et al. [80] who studied the transition between laminar
and turbulent regimes in liquid flow through a staggered tube
bank (Fig. 5).
The development of an inertial core in liquid flow through
packed beds in the viscous-inertial regime was confirmed by
investigations of Johns et al. [81]. They used magnetic reso-
nance imaging to measure fluid velocity distributions in the
interstitial pore space of a bed of 5-mm diameter glass ballo-
tini in a 4.6 cm × 70 cm glass column. Flows of pure water and
a glucose solution were studied over a range of Resf from 3.63
to 14.52 and from 0.84 to 6.75, respectively. In particular, the
influence of flow rate on the pore-scale flow pattern was studied.
It was found that with increasing Resf velocity profiles in indi-
vidual pores develop a flatter profile towards the centre of the
76 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 5. Visualization of water flow through a staggered bank of 24-mm diameter tubes. Values of the Reynolds number are calculated on the basis of hydraulic
diameter and superficial velocity. From Masuoka et al. [80], Copyright® 2002, reproduced by permission from Taylor & Francis Group, LLC.
pores and an increasing velocity gradient in the fluid close to the
solid–liquid interface. Fig. 6 shows selected profiles of absolute
and normalized velocity in an individual pore [81].
In order to quantify the observed trend in pore-scale velocity
profiles for increasing Resf Johns et al. [81] defined the variance
of the shape of normalized profiles as the sum of differences
squared between each point in the profile and the maximum
in the profile, divided by the number of points in that profile.
As the inertial core grows in the profiles and curves become
flatter around the maximum, the variance defined by this way
should decrease. In Fig. 7, the change in variance for velocity
profiles extracted from 18 randomly selected pores is shown
[81]. Each line in Fig. 7a shows how the variance evolves as the
Reynolds number is increased for a particular pore. The variance
for each pore has been normalized relative to that which existed
at the lowest Reynolds number considered (Resf = 0.84) to aid
the comparison between different pores. Although the variance
for the majority of profiles remains approximately constant, few
pores experience a marked drop in variance which is a strong
evidence for the developing inertial core. This confirms that there
is no sudden transition from creeping to viscous-inertial flow
in packed beds. Next, for each velocity profile represented in
Fig. 7a, the authors calculated a local Reynolds number Reloc
with respect to the length of the velocity profile and average
velocity calculated from that profile and found a sharp transition
in the slope of the variance as a function of Reloc occurring at
Reloc ≈ 30 (Fig. 7b). Consequently, Johns et al. [81] proposed
this value as being critical for the pore-level transition between
creeping and viscous-inertial flow regimes.
Although Johns et al. [81] did not observe an indication
for flow instability or turbulence even at the highest flow rate
(Resf = 14.52), the obtained distribution of local fluid velocity
components in the superficial flow direction (Fig. 8) indicates the
presence of regions with countercurrent flow even at Resf = 0.84,
i.e., in the creeping flow regime. The fraction of countercurrent
fluid increased with flow velocity (Fig. 8c). This observation of
time-independent local backflow in packed beds (even at small
values of Resf ) is in agreement with results of other experimen-
tal investigations [82–85]. In particular, Kutsovsky et al. [82]
employed NMR methods for studying spatially resolved veloc-
ity profiles and spatially nonresolved velocity distributions in
a packing of 0.25-mm diameter beads in a 10 mm i.d. cylin-
der (column-to-particle-diameter ratio, 40:1) and in a packing
of 6-mm diameter beads in a 40 mm i.d. cylinder (column-to-
particle-diameter ratio, 6.7:1). Flow rates corresponded to a
range of Resf from 2 to 6 and from 14.9 to 44.8 for the 0.25-
and 6-mm diameter beads, respectively. The authors found that
flow remained locally steady even at Resf = 44.8, while regions
of interstitial pore space with negative axial velocity compo-
nent were seen even at the smallest flow rate (Resf = 2). The
authors also observed that the shape of pore-space velocity pro-
files changed with the flow rate.
Ren et al. [85] studied steady-flow velocity the distribution
and its pattern in columns (19 mm i.d. × 300 mm) packed with
glass beads of different diameter (13.8, 7.0, 4.0, 2.0, and 0.6 mm)
by magnetic resonance imaging and pulsed field gradient nuclear
magnetic resonance. In particular, Fig. 9 shows the obtained
axial velocity distribution for an individual cross-section of the
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 6. Velocity profiles through a single pore in a sphere packing. (a) Distribu-
tion of the absolute velocity for water at a flow rate of 1 cm3 /s (Resf = 3.63; solid
line), 2 cm3 /s (Resf = 7.26; dashed line), 4 cm3 /s (Resf = 14.52; dotted line); and
glucose solution at a flow rate of 1 cm3 /s (Resf = 0.84; solid line, bold), 4 cm3 /s
(Resf = 3.38; dashed line, bold), 8 cm3 /s (Resf = 6.75; dotted line, bold). The pro-
files shown in (a) are shown again in (b) for water and in (c) for glucose solution,
normalized to the maximum velocity within each profile. From Johns et al. [81],
reproduced with permission, Copyright© 2000 AIChE.
bed with 2-mm diameter beads at Resf = 20.7 [85]. In this fig-
ure, negative velocity components are highlighted by a special
colour scheme and can be identified in green; they are localized
mostly in the cusp regions between two neighboured beads. The
authors point out two possible explanations for the occurrence of
backflow in packed beds: (i) circulating flow patterns in imme-
diate vicinity of the surfaces and (ii) vortex-like structures at the
meeting points of streamlines.
It is worth mentioning that the occurence of stationary eddy
patterns in laminar flows is a known, but obviously astound-
77
Fig. 7. (a) Change in variance of 18 randomly selected velocity profiles extracted
from pores in a sphere packing. (b) Same data plotted as a function of the local
Reynolds number. A transition occurs for local Reynolds numbers of about 30,
indicating the development of an inertial core in the velocity profiles. From Johns
et al. [81], reproduced with permission, Copyright© 2000 AIChE.
ing phenomenon in fluid mechanics [86–92]. In particular, after
publication of Moffatt’s paper [87] dedicated to the study of
slow flow of a viscous fluid in a sharp corner or crevice and
analysis of the formation of stationary eddies in such flow it was
shown that this type of liquid motion has a universal charac-
ter near sharp corners appearing in many situations [93–98].
Though only one parameter, namely the Reynolds number,
enters the Navier–Stokes equations governing the velocity field
of an incompressible Newtonian fluid, other parameters origi-
nating in the boundary geometry and related motion can and do
significantly influence the velocity field [99]. One of the most
essential factors determining the flow pattern thus is the flow
geometry, not only the value of the Reynolds number.
For instance, Fig. 10, after Hasimoto and Sano [100] and
Taneda [91], is a photograph of streamlines which demonstrate
the existence of stationary eddies around two equal cylinders
(separated by a distance of one cylinder diameter) in uniform
fluid flow already at Resf = 0.02. Furthermore, it has recently
been generally appreciated in fluid dynamics that turbulent flow
is not necessary for the occurrence of complex particle trajec-
tories, and that laminar flow, once thought to have a simple
dynamics, can result in chaotic behaviour of Lagrangian par-
ticle trajectories, even though the Eulerian velocity at any given
point in space is fixed in time [99,101–104]. Thus, in addition
to the conventionally quoted Reynolds number the actual flow
pattern in packed beds has to be considered as deterministically
78 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85

Fig. 9. Velocity-encoded image for water flow through a packing of glass beads,
highlighting negative velocity components in green colour. The resolution of
the image is 195 ␮m × 390 ␮m. From Ren et al. [85]. Copyright© 2005 AIChE,
reprinted with permission of John Wiley & Sons, Inc.

computational fluid dynamics (CFD) methods, also triggered by

Fig. 8. Distributions of velocity in the superficial flow direction. (a and b) Data
obtained for water (dashed lines) and glucose solution (solid lines) at flow rates of
1 and 4 cm3 /s, respectively. (c) Evolution in the velocity distribution for water
as the flow rate is increased; flow rates: 1 cm3 /s (solid line), 2 cm3 /s (dashed
line), 4 cm3 /s (dotted line). From Johns et al. [81], reproduced with permission,
Copyright© 2000 AIChE.
the development of high-performance (parallel) computer sys-
tems and new numerical approaches, in particular, the lattice
Boltzmann equation (LBE) method [105–107]. This method
is a gas kinetics-based approach invented for solving mainly
the hydrodynamics in systems described by the Navier–Stokes
equation. It has already found wide application in different areas
of CFD including the simulation of flow in porous media.
For example, Maier et al. [108] used the LBE approach to
simulate fluid flow through columns packed with spherical-
nonporous, i.e., impermeable particles. The comparison of
simulated velocity distributions with data measured by NMR
microscopy resulted in qualitative agreement. Further analysis of
simulated velocity fields in the sphere packings allowed to detect
regions with negative velocities or recirculation cells. Cells were
found to be ring vortices which form around the contact points

influencing band spreading, e.g., in liquid chromatography. In

other words, even though the macroscopic Reynolds numbers
for different fixed beds coincide, the different flow patterns on
microscopic scale may result in different hydrodynamic disper-
sion.
The existence of regions with local backflow and station-
ary eddies has also been confirmed by the results of numerical
simulations of laminar flow through fixed beds. However, it
first should be recalled that the three-dimensional large-scale
numerical simulation of flow through pore space with a complex
morphology (like in random-close sphere packings) is a difficult Fig. 10. Streamline pattern around two equal circular cylinders placed stream-
computational task. In this respect, there has been rapid progress wise in uniform fluid flow. Reproduced with permission from Hasimoto and
over the past 20 years in resolving hydrodynamic problems by Sano [100], Copyright© 1980 by Annual Reviews, Inc.
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 11. Visualization of (a) creeping flow with Resf → 0 and (b) moderate-
Reynolds number flow (Resf = 46.6) in a random array of spheres with an
interparticle void fraction of 0.412. Reproduced with permission from Hill et al.
[111], Copyright© 2001 Cambridge University Press.
of the beads. The ring diameter was approximately 20–30% of
the bead diameter, while the ring cross-section was about 5–10%
of the bead diameter. Similar regions were found by Schure et
al. [109] in the simulation of longitudinal dispersion in viscous
fluid flow through fixed beds of spherical particles. Hill et al.
[110,111] employed lattice-Boltzmann simulations to investi-
gate the effect of fluid inertia at small and moderate values of
Resf on flow through random packings of impermeable spheres.
They compared velocity fields obtained in the same fixed bed for
various values of Resf . Fig. 11 shows visualizations of the flow
patterns corresponding to Resf → 0 and Resf = 46.6. It demon-
strates the formation of backflow regions and stationary eddies
already in the laminar flow regime (Fig. 11b).
To summarize, microscopic investigations of flow through
porous media, in particular, fixed beds of particles have demon-
79
strated the existence of unsteady flow and turbulence at the
interparticle pore level. Results allow to conclude that the tran-
sition from laminar to turbulent flow takes place gradually via a
viscous-inertial flow regime accompanied by the development
of an inertial core in the pore-level velocity profile. Critical
Reynolds numbers which characterize this transition are sev-
eral times lower than for the sharp laminar–turbulent transition
of liquid flow through a straight pipe. For porous media regions
with local backflow and stationary eddy patterns exist in the lam-
inar flow regime and should not be misinterpreted as turbulence.
As pointed out by Antohe and Lage [28] the analysis of a sim-
plified turbulence model for steady unidirectional fully devel-
oped flow indicates that macroscopic turbulence cannot persist in
a porous medium once this flow configuration is achieved. Even
with turbulence at the pore level, if one measures the fluctua-
tions of a flow parameter at several points within a representative
volume, the volumetric averaging of all these quantities might
lead to a smooth macroscopic result [28]. This hypothesis con-
firms the comment by Nield [112] that “. . . true turbulence, in
which there is a cascade of energy from large eddies to smaller
eddies, does not occur on a macroscopic scale in a dense porous
medium” and also the remark by Nield and Bejan [113] that
“. . . it does not make sense to talk about turbulence on a macro-
scopic scale in a natural porous medium because one cannot
have unimpeded eddies of arbitrary size.”
4. Flow regimes and longitudinal dispersion
When a slug of tagged fluid is injected with an initially sharp
front into a porous medium (e.g., into a packed bed) saturated
with the same, but untagged fluid a diffuse, mixed zone develops
after some time across the front. In a macroscopically homo-
geneous and isotropic medium, under stagnant conditions, the
tracer fluid will diffuse at the same rate in all directions. Net
transport of the tracer fluid across any arbitrary plane can be
represented by Fick’s second law of diffusion
∂C
= Dm ∇ 2 C, (6)
∂t
where C denotes tracer concentration, t the time, and Dm is the
molecular diffusion coefficient.
Such a mixing process is independent of whether or not there
is convective transport through the medium. However, if the fluid
is flowing, then the tracers will spread faster in the macroscopic
flow direction than in the directions perpendicular to it, and some
additional mixing of a different kind appears: convective mix-
ing. The complex system of interconnected channels comprising
the microstructure of the packed bed results in continuous divi-
sion and rejoining of flow streamtubes. Consequent tangling and
divergence of streamlines is accentuated by widely varying ori-
entations of flow passages. Variations in local velocity, both in
magnitude and direction, along tortuous flow paths and between
adjacent flow paths as a result of the velocity distribution in each
pore (which, in turn, depends on the flow regime in an individual
pore) cause any initial tracer distribution to spread and occupy an
ever-increasing volume of the material [114]. In addition to het-
erogeneity on a microscopic scale, variations in the permeability
80 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85

of the porous medium from one portion of the flow domain to

the next (e.g., due to macroscopic heterogeneity) also contribute
to the overall spreading of the tracers initial distribution. Mix-
ing caused by heterogeneity on microscopic and macroscopic
scales of the velocity field is called miscible displacement or
hydrodynamic dispersion.
Theoretical approaches to hydrodynamic dispersion in
porous media have been developed including regular
models [115–120], volume-averaging models [121–125],
statistical–kinetic models [126–128], hydrodynamic models
[129–133], and network models [134–140]. As noted by Bear
[114] all the commonly employed approaches to hydrodynamic
dispersion modelling can be divided into two groups. In the
first one, a porous medium is replaced by a fictitious, greatly
simplified model in which the mixing that occurs can be ana-
lyzed using exact mathematical methods. A single tube, a bundle
of tubes, an array of cells, etc., are examples of such models
which, being amenable to exact mathematical analysis, allow to Fig. 12. Dependence of longitudinal dispersion on the Péclet number in fixed
take into account other factors affecting dispersion. The second particulate beds, with an indication of the five dispersion regimes (after Fried
and Combarnous [143]). Reproduced from Sahimi [144], Copyright© 1995, with
group of approaches assumes the construction of a conceptual permission from Wiley-VCH.
model of the microscopic motion of tracer fluid particles and
averaging this motion to obtain macroscopic description.
The dependence of the flow pattern in a porous medium on the where αi and βi , and vαi and vβi are the displacement projec-
morphology (geometry and topology of the pore space) and flow tions and the velocity projections, respectively, of the ith particle
regime in an individual pore makes the development of a gen- onto the corresponding Cartesian axes; the overbar denotes aver-
eral and universal model of hydrodynamic dispersion in porous aging over the ensemble. Commonly, one is interested in only
media an exceedingly hard or even impossible task. According the long-time asymptotic values of the diagonal elements of the
to Bear [114] “An experiment is also the only way to determine dispersion matrix, Dxx (t), Dyy (t), Dzz (t), used to characterize
the various coefficients that appear in the macroscopic equa- hydrodynamic dispersion in the direction of flow (longitudinal
tions derived from these models. There is no way to obtain them dispersion coefficient, DL or D|| ) and perpendicular to it (trans-
from the mathematical analysis itself, although some models verse dispersion coefficient, DT or D⊥ ).
may be considered more refined as they relate gross coefficients Numerous experimental studies were carried out to analyse
to the more elementary medium properties, which still must be the correlation between hydrodynamic dispersion in packed beds
determined experimentally.” Perhaps the most important param- and parameters like the length of the packed column, fluid prop-
eter characterizing hydrodynamic dispersion in porous media is erties, ratio of column diameter to particle diameter, particle size
the time-dependent dispersion tensor Dαβ (α, β = x, y, z). Its distribution, particle shape, effect of fluid velocity, etc. (For a
elements can be obtained either by the tracer particles displace- recent comprehensive survey of experimental work devoted to
ments or their velocity autocorrelation function [141] the study of hydrodynamic dispersion in packed beds, see Del-
gado [142].) In particular, Fried and Combarnous [143] have
2
t compiled experimental data concerning the dependence of the
1 dσαβ (t)
Dαβ (t) = = Covαβ (t  ) dt  , (7) longitudinal dispersion coefficient DL in sandpacks on the flow
2 dt 0
velocity. Fig. 12 represents these data in form of a variation of
where σαβ 2 (t) denotes the elements of the covariance matrix of DL /Dm with the Péclet number, Pe = uav dp /Dm , where uav is the
tracer particles displacements and Covαβ (t) is the autocovariance average fluid velocity in the interstitial pore space and dp denotes
of the tracer particles velocities. Hence, if we have an ensemble the mean size of a grain or particle. According to Sahimi [144],
of N tracer particles, an element Dαβ of the dispersion matrix the physical significance of the five different dispersion regimes
can be calculated either as in Fig. 12 is as follows:

1 d
N
2 (I) Pe < 0.3. Diffusion regime: Convection is so slow that dif-
Dαβ (t) = (αi − βi ) , α, β = x, y, z (8) fusion controls dispersion almost completely.
2N dt
i=1 (II) 0.3 < Pe < 5. Transition regime: Convection contributes to
or as dispersion, but the effect of diffusion is still quite strong.
(III) 5 < Pe < 300. Power-law regime: Convection dominates dis-

t
N persion, but the effect of diffusion cannot be neglected.
1
Dαβ (t) = (vαi (t  ) − vαi )(vβi (0) − vβi ) dt  , (IV) 300 < Pe < 105 . Pure convection or mechanical dispersion
N 0 i=1
regime: Dispersion is the result of a stochastic velocity field
α, β = x, y, z (9) induced by the randomly distributed pore boundaries.
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 13. Summary of experimental data on the longitudinal dispersion coefficient
in natural porous media [114,143]. Reprinted with permission from Brenner
[146], Copyright© 1980, The Royal Society.
(V) Pe > 105 . Turbulent dispersion regime: The Péclet number is
no longer the only correlating parameter, and the Reynolds
number should be used in addition.
Pfannkuch [145] also carried out a graphical compilation
of experimentally measured values of the longitudinal disper-
sion coefficient in packed beds in dependence of the flow rate
[44,114]. By analyzing these data, in particular, the slope of
curve fitted to these experimental data, Bear [114] distinguished
as well five different dispersion regimes. Brenner [146] replot-
ted the data of Fried and Combarnous in form of the dependence
of the longitudinal dispersivity coefficient, CL = DL /dp uav , on
the Péclet number (Fig. 13) and discriminated, by analysis of
this curve, four different dispersion regimes (joining together
dispersion regimes II and III introduced by Sahimi):
(I) Pe < 0.2. Molecular diffusion dominant.
(II) 0.2 < Pe < 500. Superposition of molecular diffusion and
mechanical dispersion effects.
(III) 500 < Pe < 2 × 105 . Mechanical dispersion dominant.
(IV) Pe > 2 × 105 . Dispersion outside the Darcy-flow domain.
It should be mentioned that, while Sahimi [144] used the
term “turbulent” in order to define the last, high-flow-velocity
dispersion regime, Brenner [146] associated this dispersion
regime with the deviation from the Darcy-flow regime, but
not solely with turbulence. Assuming Dm = 2.15 × 10−9 m2 s−1
and ν = 10−6 kg m−1 s−1 (water), as well as an average poros-
ity φ = 0.4, one can estimate values of Resf corresponding to
the onset of the last dispersion regime proposed by Sahimi
(Resf ≈ 86) and by Brenner (Resf ≈ 172). These values of Resf
are in good agreement with those corresponding to the onset of
velocity fluctuations obtained by observation of the local pattern
of the flow velocity in packed beds (see Table 1). This finding
allows to suggest that the decline in the longitudinal dispersion
coefficient at high velocities (regime V in Fig. 12; regime IV in
Fig. 13) can be a result of instabilities in the flow pattern which
81
accelerate “lateral equilibration” between velocity extremes in
the fluid. However, a further increase in flow velocity and the
occurence of turbulence do not result in any qualitative change
in the behaviour of the dispersion coefficient.
5. Flow regimes and liquid chromatography
The hydrodynamic longitudinal dispersivity coefficient
(CL = DL /dp uav ) introduced in the previous section and used
by Brenner to replot data of Fried and Combarnous (Fig. 13) has
the same physical meaning as the reduced axial plate height, one
of the most important parameters characterizing the efficiency
of chromatographic separations. It allows to translate easily the
results expressed in terms of the hydrodynamic dispersion char-
acteristics, more common in the fluid mechanics and chemical
engineering literature, into plate height which is conventional in
separation science.
In contrast to fixed beds of nonporous, i.e., impermeable par-
ticles analyzed so far in this work, most of the columns used
in modern liquid chromatography are packed with completely
porous particles in order to increase the surface-to-volume ratio
of the porous medium. Under typical conditions, the transport
of analytes through a packed chromatographic column occurs
mainly by interparticle convection and intraparticle diffusion,
and chromatographic separation is generally achieved by a dif-
ferential adsorption of analytes onto the large inner surface
area of the packing material. The adsorption process requires
that the analyte first moves through a boundary layer of fluid
at the external surface of the adsorbent particles, where diffu-
sion normal to the surface is at least the dominating transport
mechanism, and then diffuses through the pools of stagnant
fluid entrained in the intraparticle pore network. In many prac-
tical cases, the adsorption–desorption process is fast and the
equilibrium kinetics are controlled by the mass transport resis-
tances. In this respect, the stagnant mobile phase mass transfer,
i.e., the diffusion of solute molecules into and out of the nar-
row pores of the particles, with 10 nm being a typical intra-
particle pore size, has been identified as a major source of
band dispersion in liquid chromatography [1,147–149]. Espe-
cially with a smaller partition coefficient of the analyte and
a larger particle diameter this diffusion-limited mass transfer
within the particles may dramatically limit the overall kinet-
ics [150]. In particular, Fig. 14 is compiling the experimental
data obtained by Tallarek et al. [151] using dynamic NMR
microscopy for studying stagnant mobile phase mass transfer
in chromatographic columns packed with 50 and 30 ␮m parti-
cles. This NMR approach which relies on motion-encoding by
pulsed magnetic-field gradients allows to study directly the equi-
libration between stagnant intraparticle and flowing interparticle
fluid in the packed columns. Data in Fig. 14 were normalized by
the respective intraparticle diffusion coefficient. The resulting
collapse of the data onto a unique curve demonstrates that the
diffusion behaviour of the (small, unretained) fluid molecules
is governed by the morphology of the intraparticle pore space
reflected by the effective intraparticle diffusion coefficient. For
example, with 50-␮m diameter porous particles the exchange
time needed for complete equilibration between the intraparti-
82 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
Fig. 14. Intraparticle stagnant mobile phase mass transfer kinetics for spherical,
porous particles. Stagnant fluid volume fraction inside the porous particles as
a function of observation time ∆, normalized by ∆e = τ intra rp2 /2Dm (τ intra is the
intraparticle tortuosity factor and rp the particle radius). The respective tortuosity
factor accounts for the pore space morphology of the employed particles and its
influence on effective intraparticle diffusion. Reprinted with permission from
Tallarek et al. [151], Copyright© 1999, American Chemical Society.
cle and the interparticle fluid is a few hundred milliseconds
[151].
The importance of intraparticle stagnant zones originates in
their influence on dispersion. Fluid molecules entrained in the
deep, purely diffusive pools cause a substantial holdup contribu-
tion and thereby affect the time scale of transient dispersion, as
well as the magnitude of the asymptotic longitudinal dispersion
coefficient [152–155]. Consequently, the associated kinetics of
mass transfer between fluid percolating through the bed and
stagnant fluid becomes rate limiting in a number of dynamic
processes including the efficiency in chromatographic separa-
tions. For fixed beds of nonporous (impermeable) particles, the
long-time longitudinal dispersion coefficient is dominated by the
boundary-layer contribution (resulting from the no-slip condi-
tion at the solid–liquid interface) or by medium and large-scale
velocity fluctuations in the flow field, depending on the materials
disorder and Péclet number (Pe = tc /td ). This behaviour contrasts
with beds of porous (permeable) particles. In that case, liquid
holdup associated with intraparticle stagnant zones dominates
dispersion as the convective times (tc = uav t/dp ) significantly
exceed the dimensionless time for diffusion (td = Dm t/dp 2 ) [155].
Koch and Brady [124] developed a theoretical model for disper-
sion in fixed beds. They recognized four different contributions
in the long-time limit of the longitudinal dispersion coefficient
in dependence of Pe
DL
= τ + Θm Pe + Θb Pe ln(Pe) + Θh Pe2 , (10)
Dm
where Θm , Θb , and Θh denote the coefficients for mechanical,
boundary layer mass transfer, and holdup contributions, respec-
tively, while the first term represented by the tortuosity factor
(τ) stands for effective longitudinal molecular diffusion in the
packed bed. Eq. (10), which reveals a quadratic dependence of
the holdup contribution on Pe, can be easily transformed into
Fig. 15. Dependence of the (asymptotic) longitudinal dispersion on the Péclet
number in fixed beds of porous and nonporous spherical particles. Reprinted
with permission from Kandhai et al. [155], Copyright© 2002 by the American
Physical Society.
a form of reduced axial plate height equation by dividing both
sides by (Pe/2).
Thus, complete equilibration between stagnant and flowing
mobile phase in fixed beds of porous particles adds a significant
holdup contribution to longitudinal dispersion which is aggra-
vated as the interparticle velocity increases. This is illustrated
in Fig. 15 for liquid flow through fixed beds of nonporous and
fully porous particles [155] which combines experimental data
measured directly in the packed beds and results of numerical
simulations. One can see that already at Pe ≈ 100 the existence
of stagnant zones in the porous particles leads to an increase in
the longitudinal dispersion coefficient (and plate height) by one
order of magnitude with respect to the packing of nonporous,
spherical particles. At increasing velocity the flattening local
(pore-level) velocity profile and more efficient lateral mixing
related to the viscous-inertial and turbulent flow regimes can
contribute to a reduction of longitudinal dispersion, but at the rel-
evant velocities (with Pe > 105 , referring to regime V in Fig. 12
and regime IV in Fig. 13) these contributions are completely
obscured by the then tremendous holdup contribution. In other
words, the utilization of turbulent flow in liquid chromatography
with fixed beds of relatively large (dp ≈ 50 ␮m), porous parti-
cles at high velocities in itself is an ill-defined concept, from
a purely hydrodynamic point of view. With (strongly) retained
analytes in chromatography, this situation becomes a hopeless
case as they spend (much) more time in the large porous particles
than an unretained hydrodynamic tracer. By contrast, the adverse
effects of intraparticle stagnant mobile phase mass transfer on
dispersion may be simply removed using nonporous particles.
However, this realization will suffer from the very small sur-
face area available for chromatographic separation in columns
packed with particles that must be relatively large (to realize suf-
ficiently high Reynolds numbers) and impermeable (to eliminate
the massive intraparticle liquid holdup) at the same time.
Despite of conclusive data collected in this review on the
local development of unsteady flow and subsequent onset of
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
a microscopic turbulence in fixed beds of nonporous particles
(Table 1), as well as the hydrodynamic longitudinal disper-
sion characteristics in beds of nonporous and porous particles
(Figs. 12–15), it appears that successful attempts have been
reported on the implementation of turbulent flow in liquid chro-
matography using packed beds of porous particles [60,156–167].
Yet, after closer inspection of that work, particularly in the light
of this review, it seems that most of it originates in a severe
misinterpretation of the observed phenomena and misconcep-
tual use of the term “turbulent flow”. In particular, the claimed
onset of turbulent flow in chromatographic columns is charac-
terized by superficial Reynolds numbers on the order of 1–10. In
this review, we demonstrate that such values for Resf are indica-
tive for a just beginning nonlinear (non-Darcian), laminar flow
regime which has nothing to do with turbulence! Local fluc-
tuations in the flow (end of the laminar regime) begin for Resf
about an order of magnitude higher, and the onset of microscopic
turbulence is observed for Resf about two orders of magnitude
higher (see Table 1) than values for which the deviation from
Darcy’s law is noticed. Thus, the statement by Edge [60] that
“. . . in packed beds of uniform spheres it is generally agreed
that the critical Reynolds value above which turbulence occurs
is between 1 and 10 [168]” is simply false. In the same line, a
deviation from the linear pressure drop–flow rate behaviour is
not an indication for a transition to turbulent flow, but for a tran-
sition from the creeping (linear-laminar) to the viscous-inertial
(nonlinear-laminar) flow regime. Consequently, turbulent flow
(of any intensity) also cannot explain a substantial reduction
of longitudinal dispersion in fixed beds of porous particles at
superficial Reynolds numbers on the order of unity [60].
Instead, it is possible that the observed reduction of the axial
plate height [60,159] is caused by a form of nonequilibrium
chromatography, meaning that analyte zones move through the
fixed bed so quickly that equilibration in the mobile phase char-
acterized by an effective, asymptotic longitudinal dispersion
coefficient (or plate height) cannot be achieved. The intensity
of the nonequilibrium condition at the column outlet increases
with the interstitial velocity and, in good agreement with experi-
mental findings [60], with decreasing diffusion coefficient of an
analyte. Nonequilibrium more specifically refers to an insuffi-
cient time for exchange of analyte molecules between different
velocities in the mobile phase, including exchange between
intraparticle stagnant and interparticle convective fluid (if porous
particles are used), or between velocity extremes in the inter-
particle pore space [1,152,169]. In other words, by increasing
the velocity through fixed beds and decreasing analyte diffu-
sivity the intensity of nonequilibrium effects is aggravated and
plate height can become artificially small [155]. However, under
such conditions with persistent transients a plate height is not
defined, i.e., the asymptotic dispersion regime (with the disper-
sion coefficient as a global parameter characterizing the hydro-
dynamics in fixed beds at a given Péclet number) is not reached.
To eliminate, for example, the flow rate-dependent nonequilib-
rium effects corresponding to the intraparticle stagnant mobile
phase mass transfer, the ratio of packed bed length to particle
diameter should always be much larger than the Péclet number
[152].
83
6. Summary and conclusions
In this review, we have collected and analyzed material that
directly or indirectly demonstrates the transition from creeping
via viscous-inertial to turbulent flow in fixed particulate beds.
Macroscopic analysis is based on the pressure drop–flow rate
behaviour (Fig. 1) and friction factor–Reynolds number relation
(Fig. 2). A transition from the linear-laminar (creeping) flow
regime, where Darcy’s law is valid, to the nonlinear-laminar
(viscous-inertial) flow regime, where the deviation from Darcy’s
law begins, is observed at Reynolds numbers Reinter (defined
with respect to the interparticle porosity of fixed beds) on the
order of unity. The viscous-inertial flow regime lasts up to Reinter
of about 500 above which the turbulent (inertial) flow regime
dominates. Thus, in contrast to the flow behaviour in straight
open pipes the transition from creeping to turbulent flow in
fixed beds is gradual and proceeds via a viscous-inertial flow
regime which is a steady, laminar regime in which inertial forces
to the overall pressure drop cannot be neglected any longer.
The Ergun equation is the most popular empirical relationship
describing pressure drop–flow rate behaviour in fixed beds from
the creeping to the inertial flow regime depending on mobile
phase properties, characteristics of the particles, and packing
density.
Microscopic analysis in fixed beds shows that a transition
from creeping to the viscous-inertial flow regime is characterized
by the development of an inertial core within the interparticle
pore-level velocity profile (Fig. 6) meaning that flow profiles in
individual pores develop a flatter distribution towards the centre
of the pores and an increasing velocity gradient in the fluid close
to the solid–liquid interface. A pore Reynolds number of approx-
imately 30 is found to characterize the transition from creeping
to viscous-inertial flow in fixed beds on a pore-level (Fig. 7).
The onset of unsteady flow indicating the end of the laminar
flow regime is observed at superficial Reynolds numbers Resf
(defined with respect to the column cross-sectional area) on the
order of 100, while microscopic turbulence begins for Resf ≈ 300
(Fig. 3). It should be mentioned that microscopic investigations
not only result in a consistent picture on the pore-level transition
from creeping via viscous-inertial to turbulent flow in fixed beds
(Table 1), but also confirm the flow regime transitions which have
been identified based on the macroscopic analysis (Fig. 2), after
translating interstitial to superficial Reynolds numbers.
Further macroscopic evidence for flow regime transitions in
fixed beds, in particular, for the flow regime dominated by iner-
tial forces comes from analysis of the longitudinal dispersion
coefficient in dependence of the flow velocity (Figs. 12 and 13).
At values of the Péclet number (Pe > 105 ) high enough that val-
ues of the associated superficial Reynolds numbers are on the
order of 100 (end of pore-level laminar flow regime based on
microscopic analysis) a reduction of the longitudinal dispersion
coefficient is observed for fixed beds of nonporous, i.e., imper-
meable particles. It may be attributed to a more efficient lateral
mixing due to unsteady flow patterns and subsequent onset of
local turbulence. For fixed beds of porous particles, by contrast,
these effects will be hardly distinguishable at Pe > 105 because
the associated intraparticle liquid holdup is the far dominating
84 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85
contribution to the longitudinal dispersion coefficient recorded
at long enough times (Fig. 15) which is relatively unaffected by
turbulence in the interparticle pore space.
Acknowledgements
This work was supported by the Deutsche Forschungsge-
meinschaft (Bonn, Germany) under grants TA 268/1, TA 268/2,
and HL 56/1, as well as by the Fonds der Chemischen Industrie
(Frankfurt a.M., Germany).
References
[1] J.C. Giddings, Dynamics of Chromatography. Part 1. Principles and The-
ory, Marcel Dekker, New York, 1965.
[2] J.C. Giddings, W.A. Manwaring, M.N. Myers, Science 154 (1966) 146.
[3] V. Pretorius, T.W. Smuts, Anal. Chem. 38 (1966) 274.
[4] M. Martin, G. Guiochon, C. R. Acad. Sci. Paris Ser. II 294 (1982) 899,
1259.
[5] M. Martin, G. Guiochon, Anal. Chem. 54 (1982) 1533.
[6] A. Vanes, J. Rijks, C. Cramers, J. Chromatogr. 477 (1989) 39.
[7] S.R. Sumpter, M.L. Lee, J. Microcolumn Sep. 3 (1991) 91.
[8] G. Hagen, Pogg. Ann. 46 (1839) 423.
[9] O. Reynolds, Philos. Trans. R. Soc. Lond. 174 (1883) 935.
[10] O. Reynolds, Philos. Trans. R. Soc. Lond. 186 (1895) 123.
[11] H.T. Barnes, E.G. Coker, Proc. R. Soc. Lond. 74 (1904) 341.
[12] V.W. Ekman, Ark. Mat. Astron. Fys. 6 (1911) 1.
[13] W. Pfenninger, in: G.V. Lachmann (Ed.), Boundary Layer and Flow Con-
trol, vol. 2, Pergamon Press, New York, 1961, pp. 970–980.
[14] J.C. Ward, J. Hydraul. Div. ASCE 90 (1964) 1.
[15] H.S. Mickley, K.A. Smith, E.I. Korchak, Chem. Eng. Sci. 20 (1965) 237.
[16] K.R. Jolls, T.J. Hanratty, Chem. Eng. Sci. 21 (1966) 1185.
[17] C.R. Dudgeon, Houille Blanche 21 (1966) 785.
[18] C.R. Kyle, R.L. Perrine, Can. J. Chem. Eng. 49 (1971) 19.
[19] J.R. Mazzeo, U.D. Neue, M. Kele, R.S. Plumb, Anal. Chem. 77 (2005)
460A.
[20] T.H. Wegner, A.J. Karabelas, T.J. Hanratty, Chem. Eng. Sci. 26 (1971)
59.
[21] I.F. Macdonald, M.S. El-Sayed, K. Mow, F.A.L. Dullien, Ind. Eng. Chem.
Fundam. 18 (1979) 199.
[22] A. Dybbs, R.V. Edwards, in: J. Bear, M.Y. Corapcioglu (Eds.), Funda-
mentals of Transport Phenomena in Porous Media, Martinus Nijhoff
Publishers, Boston, 1984, pp. 201–258.
[23] M.A. Latifi, N. Midoux, A. Storck, J.N. Gence, Chem. Eng. Sci. 44 (1989)
2501.
[24] S. Rode, N. Midoux, M.A. Latifi, A. Strock, E. Saatdjian, Chem. Eng.
Sci. 49 (1994) 889.
[25] I. Kececioglu, Y. Jang, J. Fluids Eng. Trans. ASME 116 (1994) 164.
[26] D. Seguin, A. Montillet, J. Comiti, F. Huet, Chem. Eng. Sci. 53 (1998)
3897.
[27] J.L. Lage, B.V. Antohe, D.A. Nield, J. Fluids Eng. Trans. ASME 119
(1997) 700.
[28] B.V. Antohe, J.L. Lage, Int. J. Heat Mass Transfer 40 (1997) 3013.
[29] H. Darcy, Les Fontaines Publiques de la Ville de Dijon, Victor Dalmont,
Paris, 1856.
[30] M. Panfilov, M. Fourar, Adv. Water Resour. 29 (2006) 30.
[31] R.E. Hayes, A. Afacan, B. Boulanger, Transport Porous Media 18 (1995)
185.
[32] E.M. Minsky, Dokl. Akad. Nauk. SSSR 58 (1951) 409.
[33] S. Whitaker, Transport Porous Media 25 (1996) 27.
[34] A. Houpeurt, Rev. Inst. Fr. Petr. 5 (1953) 129;
A. Houpeurt, Rev. Inst. Fr. Petr. 6 (1953) 248;
A. Houpeurt, Rev. Inst. Fr. Petr. 8 (1953) 129.
[35] H. Ma, D.W. Ruth, Transport Porous Media 13 (1993) 139.
[36] M. Panfilov, C. Oltean, I. Panfilova, M. Buès, C.R. Acad. Sci. Paris Ser.
Mécanique 331 (2003) 41.
[37] J.D. McFarland, P.M. Dranchuk, J. Can. Petrol. Technol. 15 (1976) 71.
[38] M. Muskat, The Flow of Homogeneous Fluids through Porous Media,
International Human Resources Development Corporation, Reprint,
1982.
[39] J. Bear, Dynamics of Fluids in Porous Media, Dover Publications, New
York, 1988, pp. 181–182.
[40] D.A. Nield, Int. J. Heat Mass Transfer 40 (1997) 2499.
[41] R.K. Niven, Chem. Eng. Sci. 57 (2002) 527.
[42] A.E. Scheidegger, The Physics of Flow through Porous Media, University
of Toronto Press, Toronto, 1960.
[43] J. Happel, H. Brenner, Low Reynolds Number Hydrodynamics, with Spe-
cial Application to Particulate Media, Prentice-Hall, Englewood Cliffs,
NJ, 1965.
[44] F.A.L. Dullien, Porous Media: Fluid Transport and Pore Structure, Aca-
demic Press, San Diego, 1992.
[45] G. Schneebeli, Houille Blanche 10 (1955) 141.
[46] G. Chauveteau, Essai sur la Loi de Darcy, Doctoral Thesis, University of
Toulouse, 1965.
[47] G. Chauveteau, C. Thirriot, Houille Blanche 22 (1967) 1.
[48] D.E. Wright, Proc. Am. Soc. Civil Eng. Hydraul. Div. 94 (1968) 851.
[49] J. Comiti, N.E. Sabiri, A. Montillet, Chem. Eng. Sci. 55 (2000) 3057.
[50] P.H. Forchheimer, Z. Ver. Deutsch. Ing. 45 (1901) 1782.
[51] S.P. Burke, W.B. Plummer, Ind. Eng. Chem. 20 (1928) 1196.
[52] S. Ergun, Chem. Eng. Prog. 48 (1952) 89.
[53] F.A.L. Dullien, Chem. Eng. J. 10 (1975) 1.
[54] J.C. Giddings, Unified Separation Science, Wiley, New York, 1991.
[55] T. Farkas, G. Zhong, G. Guiochon, J. Chromatogr. A 849 (1999) 35.
[56] H. Rumpf, A.R. Gupte, Chem. Ing. Tech. 43 (1971) 367.
[57] J. Bear, Dynamics of Fluids in Porous Media, Dover Publications, New
York, 1988, pp. 176–184.
[58] F.C. Blake, Trans. Am. Inst. Chem. Eng. 14 (1922) 415;
J. Kozeny, Sitzungsber. Akad. Wiss. Wien 136 (1927) 271.
[59] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, John
Wiley and Sons, New York, 2002.
[60] T. Edge, in: I.D. Wilson (Ed.), Bioanalytical Separations, Handbook of
Analytical Separations, vol. 4, Elsevier, Amsterdam, 2003, pp. 91–128.
[61] A.R. Yevseyev, V.E. Nakoriakov, N.N. Romanov, Int. J. Multiphase Flow
17 (1991) 103.
[62] J.L. Stephenson, W.E. Stewart, Chem. Eng. Sci. 41 (1986) 2161.
[63] S. Saleh, J.F. Thovert, P.M. Adler, AIChE J. 39 (1993) 1765.
[64] M. Rashidi, A. Tompson, T. Kulp, L. Peurrung, J. Fluids Eng. Trans.
ASME 118 (1996) 470.
[65] M. Moroni, J.H. Cushman, Phys. Fluids 13 (2001) 81.
[66] U. Tallarek, E. Rapp, H. Sann, U. Reichl, A. Seidel-Morgenstern, Lang-
muir 19 (2003) 4527.
[67] F.C. Leinweber, U. Tallarek, J. Phys. Chem. B 109 (2005) 21481.
[68] C.D. Montemagno, W.G. Gray, Geophys. Res. Lett. 22 (1995) 425.
[69] M.D. Shattuck, R.P. Behringer, G.A. Johnson, J.G. Geordiadis, Phys. Rev.
Lett. 75 (1995) 1934.
[70] J.D. Seymour, P.T. Callaghan, AIChE J. 43 (1997) 2096.
[71] A.J. Sederman, M.L. Johns, A.S. Bramley, P. Alexander, L.F. Gladden,
Chem. Eng. Sci. 52 (1997) 2239.
[72] A.J. Sederman, M.L. Johns, P. Alexander, L.F. Gladden, Chem. Eng. Sci.
53 (1998) 2117.
[73] U. Tallarek, E. Bayer, D. van Dusschoten, T. Scheenen, H. Van As, G.
Guiochon, U.D. Neue, AIChE J. 44 (1998) 1962.
[74] E. Fukushima, Annu. Rev. Fluid Mech. 31 (1999) 95.
[75] N.C. Irwin, R.A. Greenkorn, S.A. Altobelli, J.H. Cushman, AIChE J. 46
(2000) 2345.
[76] U. Tallarek, T.W.J. Scheenen, H. Van As, J. Phys. Chem. B 105 (2001)
8591.
[77] D. Seguin, A. Montillet, J. Comiti, Chem. Eng. Sci. 53 (1998) 3751.
[78] J. Comiti, N.E. Sabiri, A. Montillet, Chem. Eng. Sci. 55 (2000) 3057.
[79] J. Comiti, M. Renaud, Chem. Eng. Sci. 44 (1989) 1539.
[80] T. Masuoka, Y. Takatsu, T. Inoue, Microscale Thermophys. Eng. 6 (2002)
347.
 D. Hlushkou, U. Tallarek / J. Chromatogr. A 1126 (2006) 70–85 85

[81] M.L. Johns, A.J. Sederman, A.S. Bramley, L.F. Gladden, P. Alexander, [128] N.M. Chaudhari, A.E. Scheidegger, Can. J. Phys. 43 (1965) 1776.
AIChE J. 46 (2000) 2151. [129] A.E. Scheidegger, J. Appl. Phys. 25 (1954) 994.
[82] Y.E. Kutsovsky, L.E. Scriven, H.T. Davis, B.E. Hammer, Phys. Fluids 8 [130] G. de Josselin de Jong, Trans. Am. Geophys. Union 39 (1958) 67.
(1996) 863. [131] P.G. Saffman, J. Fluid Mech. 6 (1959) 321.
[83] M. Rashidi, L. Peurrung, A.F.B. Tompson, T.J. Kulp, Adv. Water Resour. [132] P.G. Saffman, J. Fluid Mech. 7 (1960) 194.
19 (1996) 163. [133] R.E. Haring, R.A. Greenkorn, AIChE J. 16 (1970) 477.
[84] J. Götz, K. Zick, C. Heinen, T. König, Chem. Eng. Process. 41 (2002) [134] L. Torelli, A.E. Scheidegger, J. Hydrol. 15 (1972) 23.
611. [135] M. Sahimi, A.A. Heiba, H.T. Davis, L.E. Scriven, Chem. Eng. Sci. 41
[85] X. Ren, S. Stapf, B. Blümich, AIChE J. 51 (2005) 392. (1983) 2123.
[86] W.R. Dean, Proc. Cambridge Philos. Soc. 40 (1944) 19, 214. [136] M. Sahimi, B.D. Hughes, L.E. Scriven, H.T. Davis, J. Chem. Phys. 78
[87] H.K. Moffatt, J. Fluid Mech. 18 (1964) 1. (1983) 6849.
[88] A.M. Davis, M.E. O’Neill, Quart. J. Mech. Appl. Math. 30 (1977) 355. [137] M. Sahimi, A.O. Imdakm, J. Phys. A 21 (1988) 3833.
[89] K.B. Ranger, J. Eng. Math. 11 (1977) 81. [138] L. de Arcangelis, J. Koplik, S. Redner, D. Wilkinson, Phys. Rev. Lett. 57
[90] J.M. Dorrepaal, J. Eng. Math. 12 (1978) 177. (1986) 996.
[91] S. Taneda, J. Phys. Soc. Jpn. 46 (1979) 1935. [139] J. Koplik, S. Redner, D. Wilkinson, Phys. Rev. A 37 (1988) 2619.
[92] H. Hasimoto, J. Phys. Soc. Jpn. 47 (1979) 347. [140] S. Roux, C. Mitescu, E. Charlaix, C. Baudet, J. Phys. A 19 (1986) L687.
[93] M.E. O’Neill, Angew. Math. Phys. 28 (1977) 439. [141] G.I. Taylor, Proc. Lond. Math. Soc. 20 (1921) 196.
[94] S. Wakiya, J. Phys. Soc. Jpn. 39 (1975) 1113. [142] J.M.P.Q. Delgado, Heat Mass Transfer 42 (2006) 279.
[95] S. Wakiya, J. Phys. Soc. Jpn. 45 (1978) 1756. [143] J.J. Fried, M.A. Combarnous, Adv. Hydrosci. 7 (1971) 169.
[96] S.Z. Quin, H.H. Bau, Appl. Math. Model. 29 (2005) 726. [144] M. Sahimi, Flow and Transport in Porous Media and Fractured Rock,
[97] V.S. Malyuga, J. Fluid Mech. 522 (2005) 101. VCH, Weinheim, 1995.
[98] P.N. Shankar, J. Fluid Mech. 539 (2005) 113. [145] H.O. Pfannkuch, Rev. Inst. Fr. Petrol. 18 (1963) 215.
[99] J.H.E. Cartwright, M. Feingold, O. Piro, in: H. Chaté, E. Villermaux, J.- [146] H. Brenner, Philos. Trans. R. Soc. Lond. A 297 (1980) 81.
M. Chomaz (Eds.), Mixing: Chaos and Turbulence, Kluwer, Doordrecht, [147] J.N. Done, J.H. Knox, J. Chromatogr. Sci. 10 (1972) 606.
1999, pp. 307–342. [148] J.F.K. Huber, Ber. Bunsen Ges. Phys. Chem. 77 (1973) 179.
[100] H. Hasimoto, O. Sano, Annu. Rev. Fluid Mech. 12 (1980) 335. [149] Cs. Horváth, H.-J. Lin, J. Chromatogr. 149 (1978) 43.
[101] H. Aref, S. Balachandar, Phys. Fluids 29 (1986) 3515. [150] G.E. Boyd, A.W. Adamson, L.S. Meyers Jr., J. Am. Chem. Soc. 69 (1947)
[102] K. Bajer, H.K. Moffatt, J. Fluid Mech. 212 (1990) 337. 2836.
[103] D.V. Khakhar, J.G. Franjione, J.M. Ottino, Chem. Eng. Sci. 42 (1987) [151] U. Tallarek, F.J. Vergeldt, H. Van As, J. Phys. Chem. B 103 (1999) 7654.
2909. [152] D.L. Koch, J.F. Brady, Chem. Eng. Sci. 42 (1987) 1377.
[104] V.S. Malyuga, V.V. Meleshko, M.F.M. Speetjens, H.J.H. Clercx, G.J.F. [153] J. Salles, J.-F. Thovert, R. Delannay, L. Prevors, J.-L. Auriault, P.M. Adler,
Van Heijst, Proc. R. Soc. Lond. A 458 (2002) 1867. Phys. Fluids A 5 (1993) 2348.
[105] S. Chen, G.D. Doolen, Annu. Rev. Fluid Mech. 30 (1998) 329. [154] J.P. Hulin, Adv. Colloid Interface Sci. 49 (1994) 47.
[106] D. Yu, R. Mei, L.-S. Luo, W. Shyy, Prog. Aerospace Sci. 39 (2003) 329. [155] D. Kandhai, D. Hlushkou, A.G. Hoekstra, P.M.A. Sloot, H. Van As, U.
[107] R.R. Nourgaliev, T.N. Dinh, T.G. Theofanous, D. Joseph, Int. J. Multi- Tallarek, Phys. Rev. Lett. 88 (2002) 234501.
phase Flow 29 (2003) 117. [156] J. Ayrton, G.J. Dear, W.J. Leavens, D.N. Mallett, R.D. Plumb, Rapid
[108] R.S. Maier, D.M. Kroll, Y.E. Kutsovsky, H.T. Davis, R.S. Bernard, Phys. Commun. Mass Spectrom. 11 (1997) 1953.
Fluids 10 (1998) 60. [157] D. Zimmer, V. Pickard, W. Czembor, C. Müller, J. Chromatogr. A 854
[109] M.R. Schure, R.S. Maier, D.M. Kroll, H.T. Davis, J. Chromatogr. A 1031 (1999) 23.
(2004) 79. [158] M. Jemal, Biomed. Chromatogr. 14 (2000) 422.
[110] R.J. Hill, D.L. Koch, A.J.C. Ladd, J. Fluid Mech. 448 (2001) 213. [159] C.J. Oberhauser, A.E. Niggebrugge, D. Lachance, J.J. Takarewski, M.M.
[111] R.J. Hill, D.L. Koch, A.J.C. Ladd, J. Fluid Mech. 448 (2001) 243. Pegram, H.M. Quinn, LC–GC North Am. 18 (2000) 716.
[112] D.A. Nield, Int. J. Heat Fluid Flow 12 (1991) 269. [160] L. Ramos, N. Brignol, R. Bakhtiar, T. Ray, L.M. Mc Mahon, F.L.S. Tse,
[113] D.A. Nield, A. Bejan, Convection in Porous Media, Springer, New York, Rapid Commun. Mass Spectrom. 14 (2000) 2282.
1992. [161] H.K. Lim, K.W. Chan, S. Sisenwine, J.A. Scatina, Anal. Chem. 73 (2001)
[114] J. Bear, Dynamics of Fluids in Porous Media, Dover Publications, New 2140.
York, 1988, pp. 579–663. [162] A. Asperger, J. Efer, T. Koal, W. Engewald, J. Chromatogr. A 960 (2002)
[115] G.I. Taylor, Proc. R. Soc. Lond. A 219 (1953) 186. 109.
[116] R. Aris, Proc. R. Soc. Lond. A 235 (1956) 67. [163] R.P. Grant, C. Cameron, S. Mackenzie-McMurter, Rapid Commun. Mass
[117] P.V. Dankwerts, Chem. Eng. Sci. 2 (1953) 1. Spectrom. 16 (2002) 1785.
[118] G.A. Turner, Chem. Eng. Sci. 7 (1957) 156. [164] L. Ynddal, S.H. Hansen, J. Chromatogr. A 1020 (2003) 59.
[119] J. Bear, Ph.D. Thesis, University of California, Berkeley, 1960. [165] A. Vintiloiu, W.M. Mullett, R. Papp, D. Lubda, E. Kwong, J. Chromatogr.
[120] T.K. Perkins, O.C. Johnston, J. Soc. Petrol. Eng. 19 (1963) 70. A 1082 (2005) 150.
[121] S. Whitaker, AIChE J. 13 (1967) 420. [166] L. Du, D.G. Musson, A.Q. Wang, Rapid Commun. Mass Spectrom. 19
[122] Y. Bachmat, Water Resour. Res. 5 (1969) 139. (2005) 1779.
[123] W.G. Gray, Chem. Eng. Sci. 30 (1975) 229. [167] N. Sadagopan, B. Pabst, L. Cohen, J. Chromatogr. B 820 (2005)
[124] D.L. Koch, J.F. Brady, J. Fluid Mech. 154 (1985) 399. 59.
[125] O.A. Plumb, S. Whitaker, Water Resour. Res. 24 (1988) 913. [168] C.O. Bennett, J.E. Myers, Momentum, Heat, and Mass Transfer,
[126] M.J. Beran, Statistical Continuum Theories, Interscience, New York, McGraw-Hill, New York, 1982, pp. 202–209.
1968. [169] H. Van As, W. Palstra, U. Tallarek, D. van Dusschoten, Magn. Reson.
[127] P.N. Todorovic, Water Resour. Res. 6 (1970) 211. Imag. 16 (1998) 569.