Chemical Engineering Science 149 (2016) 156168

Contents lists available at ScienceDirect

Chemical Engineering Science

journal homepage: www.elsevier.com/locate/ces

Ultra-fast microuidic mixing by soft-wall turbulence

V. Kumaran n, P. Bandaru
Department of Chemical Engineering, Indian Institute of Science, Bangalore 560 012, India

H I G H L I G H T S

Flow-soft wall interaction causes transition to turbulence in a microchannel.

Multi-fold reduction in transition Reynolds number when compared to rigid channel.
Post-transition, mixing time 5 orders of magnitude lower than that for laminar ow.
Pressure drop signicantly lower than rigid channel due to soft-wall deformation.
Simple, low-cost, energy efcient strategy for ultra-fast mixing in microuidics.

art ic l e i nf o

a b s t r a c t

Article history:
Received 3 December 2015
Received in revised form
4 March 2016
Accepted 2 April 2016
Available online 11 April 2016

Slow cross-stream mixing in micro-uidic devices poses a signicant challenge in realising efcient labon-a-chip technologies. Due to the small dimension and ow velocity, the ow is in the laminar regime,
and this results in slow molecular cross-stream diffusion (in contrast to the fast turbulent mixing by
cross-stream eddies in industrial applications). Here, we demonstrate a simple and powerful strategy for
ultra-fast mixing in a microchannel with one soft wall with height as low as 35 m at a Reynolds number
as low as 226. There is a spontaneous transition from a laminar ow to a turbulent ow state when the
ow rate increases beyond a threshold value, resulting in complete cross-stream mixing. After transition,
the mixing time across a channel of width 0.5 mm is smaller, by a factor of 105, than that for a laminar
ow, and complete mixing is achieved within a channel length of 2 cm. The increased mixing rate comes
at very little energy cost, because the pressure drop is comparable to that required in current microuidic
devices, and it increases continuously and modestly at transition. This is because the channel length
required to achieve complete mixing, 2 cm, is much smaller than that used in microuidic devices that
employ diffusive mixing; in addition, the deformation of the soft wall decreases the resistance to ow.
& 2016 Elsevier Ltd. All rights reserved.

Keywords:
Mixing
Microuidics
Laminarturbulent transition

1. Introduction
Advances in microfabrication have made possible the miniaturisation of complex reaction networks onto lab-on-a-chip devices, in which the reactors and channels have smallest dimension
in the range 10100 m or less. Miniaturisation has several potential benets, such as fast reactions in controlled environments
with small volumes of reagents in devices of small size. However,
the absence of quick and efcient sample preparation methods
presents a technological barrier to realising the true potential of
microuidic technologies. Though applications that do not require
sample preparation, such as blood glucose monitoring systems,
have already been successfully commercialised, there are no successful commercial microuidic systems which involve reactions
and transformations of multiple uid streams. This is because
n

Corresponding author.

http://dx.doi.org/10.1016/j.ces.2016.04.001
0009-2509/& 2016 Elsevier Ltd. All rights reserved.

reactions take place only when the reactant streams mix at the
molecular level, and it has long been recognised that slow mixing
could be a bottleneck in the use of microreactors in commercial
applications (Whitesides, 2006; de Mello, 2006).
The mixing rates depend on the uid ow characteristics,
which in turn depend on the dimension and the velocity of ow.
For small channel widths and velocities in microuidic applications, the ow is in the laminar regime, and consists of smooth
parallel streamlines with no cross-stream ow. Mixing across the
channel in a laminar ow takes place by molecular diffusion,
which is a very slow process with mixing times of the order of
seconds or more. The time required for two streams to mix across
a channel of width W can be estimated as (W2/D), where D is the
molecular diffusion coefcient. The diffusion coefcient for small
molecules in liquids such as water is of the order of 10  9 m2/s,
while that for complex and polymeric molecules could be up to
four orders of magnitude lower. Based on this, the time required
for diffusion across a channel of width 0.11 mm is approximately

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

101000 s.
Due to the relatively long time required, mixing in microuidic
devices is usually achieved using long channel lengths, of the order of a many tens of centimeters. A specic example of blood cell
counting devices is discussed to provide a better context for the
technological difculties in microuidic mixing. Proposed blood
cell counting devices (van Berkel et al., 2011) employ microuidic
path lengths of 60 cm1 m in order to achieve lysing of the red
blood cells, and subsequent quenching in order to count white
blood cells. Such long path lengths require high pressures for
driving the ow of the order of a few atmospheres or more, and
this increases the complexity of the auxiliary equipment required
for generating such large pressures, as well as the chip fabrication
procedures due to the strength required for withstanding high
pressures.
Several strategies have been proposed for increasing mixing in
microreactors. These can be broadly classied into passive strategies where the ow is steady but the streamlines are curved, and
active strategies where time-dependence is introduced into the
ows by external actuation. Passive mixers rely on generating
complex ow pathways with curved streamlines, where uid
elements are stretched and rotated along different directions in
order to increase mixing. These include channels with repeated
bends to curve the streamlines (Kane et al., 2008; Jiang et al.,
2004; Liu et al., 2000), wall grooves to introduce secondary ows
(Stroock et al., 2002), hydrodynamic focusing where uid streams
with substantially different ow rates come into contact (Knight
et al., 1998), split-and-recombine strategies (splitting the inlet into
a large number of small streams using channel bifurcations and
then recombining them by an inverse bifurcation) either in parallel
(Bessoth et al., 1999) or in series (Lee et al., 2006), and chaotic
advection inside microdroplets. Active strategies include pressure
pulsing (Glasgow and Aubry, 2003), electrokinetic disturbances
induced due to uctuating electric elds (Bazant and Squires,
2004; Posner and Santiago, 2006), actuation by acoustic waves
(Ahmed et al., 2009), and micron sized stirring devices (Mensing
et al., 2004). In active strategies, there is additional energy input
either by micron sized moving parts, or by external elds such as
electric and ultrasound. In passive strategies, there is no external
energy input. However, there is an energy cost because the curved
streamlines, tortuous paths in split-and-recombine strategies, or
secondary ows due to wall groves, dissipate additional energy
due to uid friction, and the pressure difference required to drive
the ow is higher than that for a straight microchannel. These
strategies are not amenable to economical scaling-up (or numbering-up) because they involve complicated micromachining of
micron sized moving parts or actuators.
In nature, fast uid mixing at large scales is facilitated by a
transition to a turbulent ow, where there are large velocity
uctuations and signicant cross-stream mixing due to turbulent
eddies. This results in mixing rates that are many orders of
magnitude higher than those for a laminar ow at the same ow
rate and conduit dimensions. The ow becomes turbulent at a
Reynolds (Re (Vh/)) number higher than about 1200 in a
channel and about 2100 in a tube (Reynolds, 1883; Patel and Head,
1969). Here and are the uid density and viscosity, V is the
average velocity and h is the height of the channel or the tube
diameter. Such high Reynolds numbers have been achieved in
Polydimethylsiloxane (PDMS) based microuidic devices of dimension about 150 m (You et al., 2015), but these require very
large velocities of the order of tens of meters per second, and very
large ow rates of the order of 40 ml/min. The large pressure
gradient required to drive the ow, of the order of 13 atmospheres, also requires very strong bonding between the surfaces.
Strong bonding required for withstanding such high pressures is a
formidable challenge, which has been overcome using advanced

157

vapour deposition techniques (You et al., 2015), and this represents a signicant advance in fabricating high pressure microuidic devices.
Recent experiments (Verma and Kumaran, 2013; Srinivas and
Kumaran, 2015) have shown that there could be a transition to
turbulence in a microchannel of smallest dimension as low as
100 m if one of the walls is made sufciently soft. These experimental studies followed the theoretical prediction of different
types of instability in a channel/tube with soft walls. When the
walls are soft, the transition Reynolds number depends on the
parameter = (GR2/2), the ratio of thicknesses of the wall material and uid and the ratio of dissipation (viscosity) of the wall
and uid materials. Here, and are the uid density and viscosity, G is the shear elasticity of the wall material1 and R is the
characteristic dimension (tube diameter or channel height). The
transition Reynolds number decreases as the ratio of the wall
thickness and uid characteristic dimension increases, but it tends
to a constant in the limit where the wall thickness is much larger
than the uid thickness, since the penetration depth for the wall
displacement uctuations is comparable to the uid thickness. The
transition Reynolds number does depend on the ratio of wall and
uid viscosities, as well as the details of the wall model, for the
low Reynolds number instability (Chokshi and Kumaran, 2008),
but there is little dependence on the wall dissipation at high
Reynolds number. The transition Reynolds number shows systematic variations with the parameter , which is the ratio of
elastic stresses in the wall material and viscous stresses in the
uid. Different scalings of the transition Reynolds numbers with
have been derived in different asymptotic limits. It has been
shown that there could be an instability even in the limit of zero
Reynolds number (Kumaran et al., 1994; Kumaran, 1995; Chokshi
and Kumaran, 2008) when the parameter (V/GR ) exceeds a critical
value, and the transition Reynolds number increases proportional
to in this limit. At high Reynolds number, there are primarily
two modes of destabilisation. The high Reynolds number inviscid
modes (Kumaran, 1998; Shankar and Kumaran, 1999, 2000) are a
modication of the mode of instability in a rigid channel, and the
transition Reynolds number increases proportional to 1/2 for
these modes. There is another class of instabilities called the high
Reynolds number wall modes (Kumaran, 1996; Shankar and Kumaran, 2001), where the viscous effects are restricted to a thin
region near the wall of the channel/tube; the transition Reynolds
number increases proportional to 3/4 in this case. The ow is
destabilised by the transfer of energy from the mean ow to the
uctuations due to the shear work done at the soliduid interface. Weakly non-linear analysis has shown that the low Reynolds
number instability is sub-critical (Shankar and Kumaran, 2001;
Chokshi and Kumaran, 2008), whereas the high Reynolds number
instability (which triggers the transition to turbulence in the
present application) is super-critical (Chokshi and Kumaran,
2009). A summary of the linear stability studies is provided in
Kumaran (2003) and Shankar (2015).
The low Reynolds number instability has been veried in experiments (Kumaran and Muralikrishnan, 2000; Muralikrishnan
and Kumaran, 2002). Recently experiments in exible tubes of
diameter 1 mm (Verma and Kumaran, 2012, 2015) have demonstrated that there is a ow instability at a Reynolds number as
low as 500, which is signicantly lower than the transition Reynolds number of 2100 for the ow through rigid tubes. The transition is induced by a dynamical coupling between the uid and
the soft wall material which results in wall motion and uid
1
The polymer gels that are used for the soft walls have a compression modulus
that is an order of magnitude larger than the shear modulus, so they can be considered incompressible to a good approximation (Verma and Kumaran, 2012).

158

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

velocity uctuations. In a microchannel of height 100 m, a transition Reynolds number as low as 200 was achieved in the experiments (Verma and Kumaran, 2013); this is in contrast to the
transition Reynolds number of 1200 in a rigid channel. A summary
of the experimental studies on transition in soft-walled conduits
can be found in Kumaran (2015).
It has further been shown that the microchannel ow at Reynolds numbers in the range 200400 exhibits all the salient features of uid turbulence in channels at much higher Reynolds
numbers (Srinivas and Kumaran, 2015), including the large magnitude of the velocity uctuations, the near-wall maximum in the
streamwise root mean square velocity, and the characteristic
Reynolds stress proles; in fact, the magnitude of the velocity
uctuations, when scaled by the mean velocity, is larger in a softwalled channel at a Reynolds number in the range 200400 than
that in a rigid channel at a Reynolds number in the range 5000
20,000. However, there are also signicant differences with the
turbulent ow in a rigid channel, the most important being that
the turbulence production mechanism in a soft-walled channel
appears to involve wall motion. So the ow after transition in a
microchannel can be characterised as turbulent, though it is a
different type of turbulence in comparison to the ow in a rigid
channel. Dye-stream experiments have shown that the soft-wall
turbulence can induce mixing at a Reynolds number as low as 300
in experiments (Verma and Kumaran, 2013) with channels of
height 160 m. Here, we proceed to investigate lower channel
heights as low as 35 m to examine whether the soft-wall turbulence can induce mixing in systems of interest in microuidic
applications. The pressure drop has also been examined as a
function of the ow rate, and it has been shown that the pressure
drop after transition is higher than that expected for a laminar
ow in Verma and Kumaran (2013) and Srinivas and Kumaran
(2015). Here, the detailed shape of the deformed wall is measured
in the experiments at each value of the Reynolds number. The
geometry is reconstructed using ANSYS FLUENT simulation software, and the velocity proles and pressure drop are calculated as
a function of the ow rate assuming the ow is laminar. Below the
transition Reynolds number, the velocity proles and pressure
drop from experiments are found to be in good agreement with
simulation results. When the transition Reynolds number is exceeded, the pressure drop calculated in experiments is higher than
that in the simulations (where the ow is considered laminar) due
to turbulent uctuations.
Here, we show that the transition to turbulence induced by a
soft wall can be used to mix uid streams in microchannels with
height as low as 35 m. The Reynolds number required for complete mixing varies from about 226 for the microchannel of height
35 m, to about 300 for the microchannel with height about
100 m. This is smaller than the transition Reynolds number of
about 1200 in a rigid channel, though it is larger than the Reynolds
numbers of up to 10 encountered in microuidic applications. The
ow velocities are rather large, of the order of 36 m/s. However,
the mixing time for complete cross-stream mixing across a width
of about 0.5 mm is of the order of a few milliseconds, which is ve
to six orders of magnitude smaller than the time required for
molecular diffusion across a channel of width 500 m used here,
and complete mixing is achieved over a path length of 2 cm. Despite the large velocities, the pressure drop required to drive the
ow is much smaller than expected for two reasons. The rst is
that the length required to achieve complete mixing, less than
2 cm, is smaller by 12 orders of magnitude than the path lengths
of tens of centimeters typically required for diffusive mixing in
laminar ows. Secondly, due to the deformation of the soft wall,

the area of cross section of the channel increases when a pressure

gradient is applied. This results in a signicant reduction in the
pressure gradient at xed ow rate in comparison to a rigid nondeformable channel of the same dimensions. Due to these two
reasons, the pressure difference required to pump the uid is
comparable to, or smaller than, that required for mixing by molecular diffusion in a laminar ow.
The present experiments are restricted to mass transfer across
the channel. Enhancing heat transfer in microuidic devices is also
an area of active research (Ameel et al., 1997; Chen et al., 2013;
Rosa et al., 2009) where great progress is being made in enhancing
heat transfer coefcients. Heat transfer is not considered in the
present experimental analysis. However, the Colburn J factor
analogy (Bird et al., 1994) could be used to deduce that a signicant increase in the mass transfer rates also implies a signicant increase in the heat transfer rates. The Reynolds analogy
can also be used to relate momentum and heat transfer coefcients. These analogies are especially applicable to turbulent ows
where the same mechanism, the transport due to turbulent eddies,
is responsible for signicant enhancement of both heat and mass
transfer. Thus, our results regarding mass transfer do indicate that
heat transfer coefcients will also be signicantly enhanced due to
microuidic turbulence.
The experimental set-up is discussed in the next section, and
the results are provided in Section 3. The main implications are
summarised in Section 4.

2. Experimental conguration
Microchannels of rectangular cross-section of width 0.5 mm
and four different heights, 35, 50, 75 and 100 m and length 2 cm
are fabricated. These have two inlets IA and IB connected by a Y
junction, and two symmetric outlets OA and OB also connected by
a symmetric Y junction, as shown in Fig. 1 (b). Fluids with two
different colours in the dye-stream experiments, or with two different concentrations, either of chemical species (acid and base) or
of suspended particles, are pumped into the two inlets with equal
ow rates, and the mixing between the two streams is inferred
from the collections at the two outlets. When there is no mixing,
the outlet concentrations in OA (OB) will be equal to the inlet
concentration IA (IB). When there is complete mixing, the concentration in both OA and OB will be equal to the average of the
concentrations in IA and IB.
Three walls are made of hard PDMS (polydimethylsiloxane)
gel with a shear modulus of about 0.55 MPa, fabricated using the
Sylgard 184 kit (Dow Corning), while one of the walls is made of
soft PDMS with shear modulus of about 17 kPa, as shown in Fig. 1
(c), by reducing the cross-linker concentration during the fabrication process. The details of the fabrication procedure are provided in Appendix A. It should be noted that there is a deformation
of the soft wall of the microchannel due to the applied pressure
gradient (Verma and Kumaran, 2013; Srinivas and Kumaran, 2015),
as shown in Fig. 1(d); this deformation signicantly increases the
cross-sectional area and fortuitously reduces the pressure difference required to drive the ow in comparison to that required for
a channel of constant height (Eq. (2)), as we shall see later.
The pumping assembly used here is also different from that
usually used in microuidic applications, where external syringe
pumps are used to pump uids. Here, cylindrical reservoirs are
mounted directly on the chip for both the inlets and outlets, as
shown in Fig. 1(a). These are made by cutting 2 cm cylindrical
sections of Dispo Van 5 ml disposable syringes, which are then

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

159

Fig. 1. Schematic side view of the experimental conguration with the microchannel, inlet and outlet reservoirs and piston for pumping (a), the top view of the microchannel conguration with the two inlets and outlets used for studying the mixing of the dye-stream, dissolved solutes and suspended particles (b), cross section of the
undeformed microchannel (c) and the deformed microchannel (d) due to the applied pressure difference.

tight-tted into cylindrical holes cut into the top PDMS slab above
the inlets and outlets in a leak-proof manner. The plungers in the
Dispo Van 5 ml syringes are inserted into the cylindrical barrels of
the reservoirs, and two mechanical pushers, which are metal rods
which can t into the cylindrical barrel of the reservoirs, are used
to push the plungers. The mechanical pushers are driven by programmable linear actuators, and have a maximum speed of 2 mm/
s, which results in a maximum ow rate of about 38 ml/min. The
speed of the external mechanical pusher can be adjusted to vary
the ow rates from each of the inlets. In this design, the uids are
all pre-loaded on to the chip, and the chip is uidically insulated
from the surroundings. Only external mechanical actuation is used

IA

Pressure
tapping

to pump the uids from the inlets to the outlets.

The microchannel conguration shown in Fig. 2 is used for the
pressure measurement. These microchannels are fabricated separately from those used for the mixing experiments, there is no
outlet manifold in these microchannels, and the outlet is open to
the atmosphere. There is a side pressure tapping at the inlet of the
microchannel, as shown in Fig. 2.
The total pressure difference between the inlet and outlet is
determined using a U shaped mercury manometer, in which the
inlet is connected to the side pressure tapping in Fig. 2, and the
outlet is open to the atmosphere. The pressure difference is determined as a function of the Reynolds number, which is dened
as

Re =

IB
2 cm
Fig. 2. The conguration, with no outlet manifold and a pressure tapping near the
inlet, used for measuring the pressure difference across the microchannel.

Q
W

(1)

where Q is the ow rate, W is the width of the microchannel and

is the kinematic viscosity (ratio of dynamic viscosity and density).
The Denition (1) is identical to the usual denition of the Reynolds number for a microchannel based on the height, Re = (Vh/ ),
because the average velocity is V = Q /(Wh), the ratio of the ow
rate Q and the area of cross section Wh. The Denition (1) does not
depend on the height of the channel, and so it is advantageous in
the present channel conguration with a soft wall, because the
height of the channel could change due to wall deformation
caused by an applied pressure gradient. The width of the channel,
W0.5 mm, is much larger than the height which is in the range
35100 m, and so the ow in the channel can be considered as
two dimensional. In the absence of deformation, the ow can be
approximated as a two-dimensional parabolic ow in the channel,
and the theoretical pressure gradient is related to the ow rate and
the Reynolds number by

160

pth
L

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

12Q
h3 W

12 Re
h3

(2)

where L (2 cm) is the length across which the pressure difference

is measured.
For visual observation of mixing, shown later in Figs. 3 and 4,
water mixed with black ink is pumped into inlet IA, while pure

water is pumped into inlet IB. The progress of the two streams is
lmed using a high speed camera from above at the three locations indicated by dashed rectangles in Fig. 1(b), one at the upstream junction, the second in the middle and the third at the
downstream junction.
A quantitative measure of the mixing between the two dyestreams is calculated as follows (Verma and Kumaran, 2013; Lee

Fig. 3. Images from the dye-stream experiments at the inlet (left), mid-way between inlet and outlet (centre) and outlet (right), shown in Fig. 1(b), for a microchannel with
height 50 m at Reynolds number 133 (rst row), 200 (second row), 226 (third row) and 254 (fourth row). The reference length is provided by the width of the channel,
which is 0.5 mm.

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

161

Fig. 4. Images from the dye-stream experiments at the inlet (left), mid-way between inlet and outlet (centre) and outlet (right), shown in Fig. 1(b), for a microchannel with
height 100 m at Reynolds number 133 (rst row), 200 (second row), 266 (third row) and 300 (fourth row). The reference length is provided by the width of the channel,
which is 0.5 mm.

162

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

et al., 2010). From the image, at the pixel indices i and j in the
horizontal and vertical directions, the grey scale intensity Gij is
determined. The image average of the grayscale intensity, Gav, is

Gav =

1
N

Gij

(3)

i, j

where N is the total number of pixels used for the averaging. The
variance of the distribution is given by

(G)2 =

1
N

(Gij Gav)2
i, j

(4)

(G)2
(G max G min)/2

NA
(NA + NB )/2
NB
NB* =
(NA + NB )/2

N *A =

The mixing index based on image analysis, MI, is given by

MI = 1

of particles (number per unit volume) at the outlets OA and OB is

determined using a BD FACSVerse Flow Cytometer at a low ow
rate of 15 l/min. A total of 7.5 l of the sample from each outlet is
passed through the ow cytometer over a period of 30 s, and the
maximum particle count is about 800 over this period. If NA and NB
are the number of particles counted in equal volumes of uid from
outlet OA and OB, the normalised number density of particles is
dened as

(5)

where G max G min is the difference between the pixel intensities of

the light and dark regions in the laminar ow. For a well-mixed
uid, the intensity is uniform across the entire image, (G2) is
equal to zero, and the mixing index MI is equal to 1. When the two
streams are completely segregated, one half of the channel is light
while the other half is dark. In this case, Gav = (G max + G min) /2,
G = (G max G min) /2. If the light intensity of the dark regions
Gmin = 0, the mixing index is 0 when there is no mixing. There are
some variations in the mixing index as calculated above due to
variations in the light intensity in the images, despite the care
taken to avoid reections. When a completely mixed dye sample is
passed through the channel, the mixing index is found to be less
than 1, but greater than 0.9. Therefore, the mixing index value of
0.9 is considered to indicate complete mixing in the experiments.
Two other mixing tests are carried out, one for dissolved chemicals and the other for suspended particles. In the test for the
mixing of dissolved chemicals, two solutions of different pH,
0.01 M NaOH solution with pH 12 and 0.01 M HCl solution with
pH 2, are pumped into inlets IA and IB respectively. The pH of the
solutions in the outlets OA and OB are measured using a digital pH
meter DP 505 manufactured by Digital Instruments Corporation,
with an accuracy of about 0.1 in the pH. The pH in the outlets OA
and OB are expected to be 12 and 2 respectively when there is no
mixing, and the pH in both outlets are expected to be 7 if there is
perfect mixing. The mixing is considered to be complete if the pH
at both outlets is 7 0.1, since the resolution of the pH meter is
0.1.
In the second mixing test, water with suspended spherical
particles of diameter 6 m and concentration about 100 particles/
l is pumped in inlet IA, while clear water is pumped into inlet IB.
The particles used are Polybead polystyrene black dyed particles
from Polysciences Inc., with a mean diameter of 6.02 m and
standard deviation 0.17 m as reported by the manufacturer. The
size of the particles is carefully chosen. For particles less than
about 1 m, there is signicant Brownian motion, and some apparent diffusion of particles due to Brownian motion even though
the uid streams do not mix. For particles greater than about
10 m, the particle inertia becomes a factor, and so the particles do
not follow the uid streamlines. Particles in the range 110 m are
chosen so that both particle inertia and Brownian motion are not
important, and the particles follow the uid streamlines crossstream motion of the particles is possible only if there are crossstream uid velocity uctuations. This size range is also of practical importance in the context of cells suspended in the blood, for
example, which are mostly in the 110 m range. The volume
fraction is also sufciently small, less than 10  5, so that particle
interactions and coagulation are not important. The concentration

(6)

The values of N *A and NB* are 0 and 2 respectively, when there is no

mixing, and N *A = NB* = 1 when there is perfect mixing. Due to the
small number of particles (about 800) counted in the cytometer,
there is expected to be a statistical error of about
(1/ 800 ) 0.0353 in the normalised number densities N *A and NB*
in the experiments. The two uids are considered to be perfectly
mixed if N *A and NB* are in the range 1 0.0353.
The mixing experiments for dye, dissolved chemicals and suspended particles are carried out over a range of Reynolds numbers
from 0 to about 400, by changing the ow rates at which the uid
streams are pumped onto the channel from the reservoirs
mounted on the chip. Each experiment is repeated four times, and
the mean and the standard deviation are determined. In all gures,
the error bars show one standard deviation above and below the
mean. The measurements of the pressure difference are repeated
three times for each Reynolds number, and the error bars in the
pressure difference measurements show one standard deviation
above and below the mean.

3. Results
Figs. 3 and 4 show images of the dye-stream experiments in
microchannels with height 50 m and 100 m respectively. These
images are recorded from above; the schematic of the top view is
shown in Fig. 1(b). Water mixed with dye is pumped into inlet IA,
while pure water is pumped into inlet IB. The images are recorded
at the three location shown by the dashed rectangles in Fig. 1(b)
the Y junction from the inlet is shown on the left and the Y
junction to the outlet is shown on the right, while the middle row
shows images at the centre or the channel. The regions with pure
water appear bright, while containing dye is mixed with water
appear dark.
For the microchannel with height 50 m, there is no mixing
when the Reynolds number is 133, but signicant mixing is observed at a Reynolds number of 200. The mixing further increases
when the Reynolds number is increased to 226, and complete
mixing at the centre of the microchannel is observed at a Reynolds
number of 254. In a similar manner, for the microchannel with
height 100 m, there is no mixing at a Reynolds number of 133,
and there is very little mixing at a Reynolds number of 200. There
is signicant mixing at a Reynolds number of 266, and complete
mixing is observed at a Reynolds number of 300. Thus, the Reynolds number for complete mixing increases as the channel height
is increased, as reported in previous studies (Verma and Kumaran,
2012; Srinivas and Kumaran, 2015). The mixing index MI (Eq. (5)),
is shown as a function of Reynolds number for the images at the
centre of the channel (the central column of images in Figs. 3 and
4) in Fig. 5. For the channel with height 35 m, we nd no mixing
at a Reynolds number of 66, but there is some mixing even at a

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

1
0.9
0.8
0.7
MI

0.6
0.5
0.4
0.3
0.2
0.1
0

100

200
Re

300

400

Fig. 5. The results for the mixing index MI (Eq. (5)) calculated for the images at the
centre of the channel (central images in Fig. 4) for a microchannel with height
100 m (), 75 m ( ), 50 m () and 35 m ( ). The horizontal dashed lines shows
the value MI 0.9, and the vertical lines show the Reynolds numbers where the
mixing index falls within this variation, Re226 for the channel with height 35 m,
Re 254 for the channel with height 50 m, Re 266 for the channel with height
75 m, and Re 300 for the channel with height 100 m.

Reynolds number as low as 133. The mixing index increases with

Reynolds number, and there is complete mixing at a Reynolds
number of 226. The Reynolds number for the onset of mixing increases as the height of the microchannel increases, to about 166
for the microchannel with height 50 m, about 200 for the

163

microchannel with height 75 m and about 233 for the microchannel with height 100 m. Similarly, the Reynolds number for
complete mixing also increases to about 254 for the microchannel
with height 50 m, 266 for the microchannel with height 75 m
and about 300 for the microchannel with height about 100 m.
The results of the mixing experiment for dissolved chemicals
and suspended particles at different Reynolds numbers are shown
in Fig. 6. Fig. 6(a) shows the pH of the two outlet streams when the
two inlet streams have pH of 2 and 12, and Fig. 6(b) shows the
normalised particle count at the two outlets when N *A = 2 and
NB* = 0 at the inlet. In both the mixing tests, there is very little
mixing at low Reynolds number. However, as the Reynolds number is increased, the pH and normalised particle count at the two
outlets approach each other and become equal to within experimental errors at high Reynolds number. A sharp convergence in
the pH and the normalised particle count values in the two outlets
are seen at Reynolds numbers as low as 133 for the channel with
height 35 m, and about 233 for the channel with height 100 m.
This coincides with the Reynolds numbers at which mixing between the two uids is observed in the dye-stream experiments.
The pH and normalised particle count values are within the perfect mixing ranges (7 70.1 for the pH and 1 70.0353 for the
normalised particle count) when the Reynolds number increases
beyond a critical value of about 226 for the channel with height
35 m, 254 for the channel with height 50 m, 266 for the channel
with height 75 m and about 300 for the channel with height
100 m. These are also in agreement with the results of the dyestream experiments.
The pressure difference applied across the channel is shown as
a function of the Reynolds number in Fig. 7 (a). It is observed that
the pressure difference required for pumping the uid in a channel
of length 2 cm is about 40 kPa for the channel with height 100 m
at a Reynolds number of 300, and it increases to about 80 kPa for a
channel of height 35 m at the same Reynolds number. It is clear
that the experimental pressure difference does not increase linearly with Reynolds number, as expected for a rigid channel. This is

Fig. 6. The results for the pH (a) and the normalised particle count shown in Eq. (6) (b) in the two outlets shown in Fig. 1(b), outlet A (red line) and outlet B (blue line) as a
function of the Reynolds number for a microchannel with height 100 m (), 75 m ( ), 50 m () and 35 m ( ). The horizontal dashed lines show the variation of 70.1 in
the average pH value of 7 in sub-gure (a), and a variation of 7 0.0353 about the average normalised particle count N *A = NB* = 1. The vertical lines show the Reynolds
numbers where the pH and particle counts in both outlets fall within this variation, Re 226 for the channel with height 35 m, Re254 for the channel with height 50 m,
Re 266 for the channel with height 75 m, and Re 300 for the channel with height 100 m. (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.).

164

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

Fig. 7. The pressure drop p (a), and the ratio of the pressure drop and the theoretical pressure drop in the absence of channel deformation, (p/pth ) (b), as a function of
Reynolds number for the microchannel of length 2 cm and height 100 m (), 75 m (), 50 m () and 35 m ( ). The theoretical pressure drop in the absence of channel
deformation, pth is given in Eq. (2). Lines show the Reynolds numbers where complete mixing is observed as shown in Fig. 6.

because of the deformation of the wall of the channel due to the

applied pressure difference, and the resultant increase in the
cross-sectional area which results in a decrease in the ratio of the
ow rate and the pressure difference. The ratio of the experimental pressure difference and the theoretical prediction,
(p/pth ), is shown in Fig. 7(b). Here, pth , the pressure drop in the
absence of channel deformation, is given in Eq. (2). The ratio decreases from about 0.80.5 when the Reynolds number is increased from 50 to 300 for the channel with height 100 m, but it
has a much smaller value less than 0.1 for the channel with height
35 m. This is because the fractional increase in the cross-sectional area due to expansion is much larger for the smaller channel. Thus, in addition to inducing an instability due to a dynamical
uidwall coupling at very low Reynolds numbers, the use of the
soft wall also has the advantage of signicantly reducing the
pressure difference required for a specied ow rate due to the
deformation of the wall and the consequent increase in the crosssectional area.

4. Discussion
A range of experimental techniques have been used to examine
the cross-stream mixing in a microchannel due to soft-wall

turbulence, where a transition to turbulence is induced by an instability due to the dynamical coupling between the uid and the
wall material. The experiments include image analysis for examining the mixing of dye, the mixing of two chemical species, in
this case acid and base, and the mixing of suspended particles. All
of these techniques unambiguously indicate that complete mixing
across a width of 0.5 mm is achieved at Reynolds number as low as
226 for a channel of height 35 m, and about 300 for a channel of
height 100 m. This perfect mixing is achieved at a lower Reynolds
number than that required for the transition to turbulence in a
rigid channel, which is about 1200 (Patel and Head, 1969). The
mixing time, which is of the order of 110 ms, is ve orders of
magnitude lower than that required for molecular diffusion across
a width of 0.5 mm in a laminar ow, and the path length of 2 cm is
12 orders of magnitude lower than that used in microuidic
devices which employ diffusive mixing. The pressure difference
required to drive the ow, which is less than one atmosphere, is
also comparable to that currently used in microuidic devices.
A quantitative comparison of the parameters for mixing due to
soft-wall turbulence are shown in Table 1. Also shown are the
parameters for microuidic mixing due to hard-wall turbulence
from the previous work of You et al. (2015). Due to the small dimensions, relatively large ow rates in the range of a few ml/min,
and velocities in the range of 36 m/s, are required to attain the

Table 1
For a channel with length L (cm), height h (m) and width W (m), the Reynolds number Remix at which complete mixing is observed, the ow rate Q in ml/min, the average
velocity V in m/s, the residence time tres (ratio of length (2 cm) and average velocity) in milliseconds, the average pressure gradient p in (MPa/m), and the pressure
difference p across the length of the channel in kPa.
L

Remix

tres

500
500
500
500

226
254
266
300

6.78
7.62
7.98
9.00

6.46
5.08
3.55
3.00

3.11
5.08
5.64
6.66

3.545
3.066
2.138
1.975

70.91
61.33
42.76
39.49

150

4423

40.00

29.63

0.51

37.470

280

Present experiments
2.0
2.0
2.0
2.0

35
50
75
100

Experiments of You et al. (2015)

1.5

150

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

165

Table 2
For a channel with height h and width W, the time required for mixing due to molecular diffusion in a laminar ow across a width W, tmix = (W2/D) , the average velocity V,
the length required for mixing due to molecular diffusion in a laminar ow L = tmix V , the pressure gradient required to drive the ow p in (kPa/m), the pressure difference

p = Lp in kPa, and the Reynolds number (Q /W), where the ow rate is considered to be 30 l/min, and the molecular diffusion coefcient D is assumed to be 109 m2/s .
h (m)

W (m)

tmix (s)

V (m/s)

L (m)

p (kPa/m)

p (kPa)

Re

50
50
100
100

50
500
100
500

2.5
250
10
250

0.2
0.04
0.05
0.01

0.5
10
0.5
2.5

2276
96
142
12

1138
960
71
30

10
1
5
1

transition Reynolds numbers, as shown in Table 1. Since the channel

is of length 2 cm, the residence time (ratio of channel length and
average velocity) required for complete mixing is only a few milliseconds. Despite the high ow velocity, the pressure difference
required to drive the ow varies from about 40 kPa for a channel
with height 100 m to about 70 kPa for a channel with height
35 m. This is much lower than that required for the turbulent ow
in a rigid channel, and is easily achieved using linear actuators
driving the ow through reservoirpiston assemblies mounted on
the chip. In contrast, in a hard microchannel, the ow velocities
required for complete mixing across a width of 150 m are of the
order of 40 m/s, the Reynolds number is more than 4000 and the
pressure gradient required to drive the ow is about 3 atm.
For comparison of turbulent mixing observed in our experiments with conventional mixing due to diffusion in a laminar ow,
consider typical dimensions shown in Fig. 2. The diffusion time
across the channel width W is estimated as W2/D , where the
molecular diffusion coefcient D is considered to be a typical value
of 10  9 m2/s for the diffusion of a small molecule in water. The
mixing times required range from about 2.5 s across a width of
50 m, to about 250 s across a width of 0.5 mm. Note that for
W0.5 mm, the mixing time of 250 s for diffusive mixing is about
56 orders of magnitude higher than that for turbulent mixing, as
shown in Table 1. For a typical ow rate of 30 l/min, the velocities
are in the range of 120 cm/s, and the Reynolds number is in the
range 110; both of these are much lower than those (Table 1)
used for turbulent mixing. However, due to the combination of
low velocities and long mixing time required, the channel length
required for laminar mixing is very large, in the range 0.510 m.
The pressure gradient required for driving the ow for diffusive
mixing in Table 2 is much smaller than that for turbulent mixing in
Table 1, due to the much smaller ow velocity. However, due to
the large channel lengths required, the pressure difference across
the channel (pressure gradient times the channel length) for turbulent mixing in Table 1 is either comparable (for channel of
height 100 m) or much smaller (for channel of height 50 m) in
comparison to that for laminar mixing in Table 2. It should be
noted that the pressure gradient and pressure difference for diffusive mixing in Table 2 are calculated assuming a straight channel; channels of equal length with multiple beds (for tting the
channel in a small area) could have a pressure difference 210
times higher than that stated in Table 2.
It is difcult to make a complete comparison with other passive
mixing strategies, such as serpentine channels, split-and-recombine strategies and ow focusing strategies. It is difcult to
quantify one of the primary advantages of the present strategy,
which is the extremely simple and low cost fabrication in comparison to those for the complex geometries in other passive
mixing strategies. In addition, most studies do not report the
pressure drop or the friction factor, and so it is difcult to compare

the energy requirements here with those in passive mixing.

However, we attempt to examine some of the relative advantages/
disadvantages with select studies on passive mixing with comparable dimensions to ours, with the caveat that the complexity
and cost of fabrication, not factored here, heavily favours the
present strategy.
1. In chaotic mixers (Stroock et al., 2002), grooves of length about
200 m are etched into the bottom of a microchannel of width
200 m and height about 70 m. These generate a secondary
ow that could be chaotic based on the groove design, and this
causes enhanced mixing. The length required is between 1 and
3 cm for mixing across a width of about 200 m; in contrast,
with the present strategy, we get complete mixing across a
width of 500 m over a length of 2 cm. The maximum Reynolds
number (about 10) and ow rate (about 0.1 ml/min) are also 12
orders of magnitude lower than those used here. The time
required for complete mixing, about 0.1 s, is an order of
magnitude higher than that used here. Since the channels used
are rigid, the pressure drop per unit length is about 10 times
that required here for a channel of the same dimensions, as can
be seen from the comparison in Fig. 7(b) . In addition, sophisticated micromachining is required to cut grooves in the wall of
the channel.
2. Curved and serpentine channels (Liu et al., 2000; Jiang et al.,
2004) have been used to enhance mixing in microchannels.
Two- and three-dimensional channels (Liu et al., 2000) have
been created by replicating C sections using sophisticated
microfabrication techniques. The total length of these microchannels is close to 2 cm, and the cross-section is of dimension
300 m  150 m. The Reynolds number attained is up to 70,
the mean velocities are up to 0.5 m/s, while the ow rates are
up to 1 ml/min, about an order of magnitude lower than those
considered here. In the 3D serpentine channels, mixing takes
place at Reynolds numbers above 10 within distances of the
order of 1 cm and times of the order of 0.05 s. The mixing is
much poorer in 2D and straight channels, and so it is necessary
to fabricate channels in 3D to achieve good mixing.
The authors have not reported the pressure difference required
to drive the ow, but for sharp bends of the type used in Liu
et al. (2000), the excess pressure due to the bend has been
reported (Maharudrayya et al., 2004). In the limit of low
Reynolds number (up to about 32), the excess pressure drop is
about
(49U /Dh ), where
Dh is the hydraulic diameter
(2 Wh/(W + h)), where W and h are the width and height
respectively. For a straight channel, the friction factor is given
by (14.2/Re) for a laminar ow, and so the pressure drop is
about (7.1UL/Dh2), where L is the channel length. The ratio of
the excess pressure drop in the bend and in the straight section
is, therefore, about (7Dh/L ). Thus, the pressure drop in a bend is

166

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

equal to that in a straight section of length about 7 hydraulic

diameters.
In the two-dimensional conguration of Liu et al. (2000), there
are two bends in each C section of length about 0.9 mm, and
pressure drop in the bends is equivalent to an added straight
channel length of 14 hydraulic diameters, or about 2.8 mm. In
the three-dimensional conguration, there are four bends in
each C section of length about 0.9 mm, and the pressure drop in
the bends is equivalent to an added straight channel of length
28 hydraulic diameters, or about 5.6 mm. Thus, the pressure
drop in the bends is about 3 times that in the straight section for
the two-dimensional channel, and about 6 times that in the
straight section in a three-dimensional channel. Thus, the
energy cost for mixing in a channel with three-dimensional C
sections is 7 times that in a straight channel. In contrast, in the
present soft-wall strategy, the energy cost is an order of
magnitude lower than that in a rigid straight channel due to
wall deformation, as shown in Fig. 7(b).
3. In the split-and-recombine strategy (Lee et al., 2006), the inlet
streams are split and recombined multiple times in repeat units
of length about 600 m, height 100 m and width 400 m.
About 510 repeat units are required for the variance in
concentration within about 10% between two adjacent inlet
streams. The total channel length required for the split-andrecombine mixer, which is about 36 mm, is smaller than the
2 cm length used here, but the ow rates that could be attained
are much lower, from 10  4 to 1 ml/min (in comparison to the
ow rates of 610 ml/min here). Despite the low ow rates, the
pressure drop is signicantly higher, about 35 kPa for each
repeat unit (or about 175 kPa for 5 repeat units or 350 kPa for
10 repeat units) at a ow rate of 1 ml/min, in comparison to the
pressure drop of about 40 kPa for the entire 2 cm length in the
present microchannel at a ow rate of 10 ml/min. In addition,
fabrication is much simpler and cheaper for a straight microchannel in comparison to the split-and-recombine mixer.
4. Flow focusing is another strategy used to increase mixing rates
(Knight et al., 1998). Here, a thin central focused stream of uid
is passed between two side streams of much higher width. Fast
mixing is achieved because the width of the focused stream
could be made very small, as low as 10 nm, and so the diffusion
time across the focused stream is very low. Though this method
achieves efcient infusion of the side streams into the focused
stream, due to the small width of the focused stream, the
reverse diffusion of the material from the focused stream into
the side streams is still a slow process due to the large width of
the side streams. Moreover, this technique is useful only when
the channel width is small (10 m width was used in Knight
et al. (1998)) and if the ow rates of one of the streams, which is
to be focused, is very low (ow rates of 5 nl/s were used in
Knight et al. (1998)).
The soft-wall mixing strategy is extremely well suited for
multi-step microuidic reactiondiffusion processes, where sample preparation is currently a severe barrier to commercialisation
due to slow mixing. As an example, in counting devices for blood
cells, two parallel processes need to be operated for erythrocyte
and leucocyte counts (van Berkel et al., 2011). For the leucocyte
counts, it is necessary to mix blood rst with a lysing solution, so
that the erythrocytes (which are far more numerous than leucocytes and approximately the same size) are lysed, and then a
quenching solution after a pre-determined time delay to ensure
that lysing of the more robust leucocytes does not take place. In
current microuidic chips (van Berkel et al., 2011), this is carried
out in microchannels with path lengths of the order of tens of
centimeters, and at low velocities, so that there is adequate time
for diffusion to take place. It is also necessary to drive the ow

using pumps due to the high pressure drop. Using soft-wall mixing, this entire sample preparation process can be carried out
using channels of length about 2 cm each. The reagents can be preloaded in on-chip reservoirs, and pistons mounted in these reservoirs can be used for pumping at very low pressures.

5. Conclusions
Our experiments demonstrate a simple method by which ultrafast mixing can be induced in microchannels at low Reynolds
numbers by making one of the walls soft. The physical mechanism
is a dynamical instability due to the uidwall coupling which
results in a transition from the laminar ow to turbulent ow,
resulting in the catastrophic break-up of streamlines in the ow.
The mixing time across our channel of width 0.5 mm is smaller by
ve orders of magnitude in comparison to that required for diffusive mixing, and complete cross-stream mixing is achieved
within tens of milliseconds. The transition is tunable, since the
ow rate or Reynolds number for transition can be adjusted by
varying the shear modulus of the soft wall. Signicantly enhanced
mixing is attained at relatively small cost, since the pressure difference required to drive the ow increases only modestly and
continuously at transition. The fabrication is simple, since the
standard soft-lithography techniques for straight microchannels of
small length are used. The fabrication is also scalable, since the
procedure used for one microchannel can easily be replicated to
fabricate microchannel arrays. There is no requirement of expensive micromachining of complicated features as in passive
mixers, or replicating of moving small parts as in active mixers.
The pressure difference required for driving the ow was easily
achieved by a linear actuation assembly coupled with reservoirs
mounted on the chip, which enables the chip to be uidically insulated from its surroundings. All of these features make this an
ideal mechanism for sample preparation in microuidic devices.

Acknowledgements
The authors would like to thank the Department of Science and
Technology, Government of India, for nancial support.

Appendix A. Microchannel fabrication

The microchannel is fabricated with three hard walls made of
Polydimethyl siloxane (PDMS) gel using the catalyst concentration
of 10% recommended in the Sylgard 184 kit from Dow Corning,
which results in a shear modulus of about 0.55 mPa, using PDMS
soft lithography technique. The fourth wall is made much softer
using a lower catalyst concentration using a different technique.
For the three hard walls, a negative pattern of the channel is
fabricated using SU8 photoresist on a silicon wafer, where the
spin-coating protocol of the photoresist is adjusted in order to
obtain the desired height. After curing and xing the pattern, the
pre-polymer is poured onto this negative and cross-linked by
standard procedures (Verma and Kumaran, 2013; Srinivas and
Kumaran, 2015). After cross-linking, the polymer is peeled off to
provide an open channel.
In contrast to the usual fabrication of PDMS channels, the PDMS
stamp is not bonded onto glass, but onto a soft PDMS layer of
thickness 2 mm prepared separately, as shown in Fig. A1 (a). This
lm is prepared in two steps. First, the pre-polymer mixture with
catalyst concentration 1.75% (which results in a soft gel with shear
modulus about 17 kPa) is poured onto a glass slide with conning

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

167

Fig. A1. Preparation of the soft wall of channel (a), and bonding of the soft wall with the PDMS stamp (b).

glass walls of height 2 mm. This is cured to obtain a soft gel, and
the conning walls are removed. Initially, this lm was bonded
onto the PDMS stamp. However, when a high pressure was applied
at the inlet, there was a deformation of the soft wall below the
inlet resulting in leakage. In order to prevent this, the region below
the inlets and outlets was further hardened by pouring additional
catalyst, resulting in a lm with hard gel below the inlet and
outlet, and soft gel below the microchannel in the PDMS stamp.

References
Ahmed, D., Mao, X., Shi, J., Juluri, B., Huang, T., 2009. A millisecond micromixer via
single-bubble-based acoustic streaming. Lab Chip 9, 27382741.
Ameel, T.A., Warrington, R.O., Wegeng, R.S., Drost, M.K., 1997. Miniaturisation
technologies applied to energy systems. Energy Convers. Manage. 34, 969982.
Bazant, M.Z., Squires, T.M., 2004. Induced-charge electro-kinetic phenomena theory
and microuidic applications. Phys. Rev. Lett. 92, 066101.
Bessoth, F., de Mello, A., Manz, A., 1999. Microstructure for efcient continuous ow
mixing. Anal. Commun. 36, 213215.
Bird, R.B., Stewart, W.E., Lightfoot, E.N., 1994. Transport Phenomena. John Wiley &
Sons, New York.
Chen, L., Zhang, X.-R., Okajima, J., Maruyama, S., 2013. Numerical investigation of
near-critical uid convective ow mixing in microchannels. Chem. Eng. Sci. 97,
6780.
Chokshi, P.P., Kumaran, V., 2008. Weakly nonlinear analysis of viscous instability in
ow past a neo-hookean surface. Phys. Rev. E 77, 056303.
Chokshi, P.P., Kumaran, V., 2009. Weakly nonlinear stability analysis of a ow past a
neo-Hookean solid at arbitrary Reynolds numbers. Phys. Fluids 21, 014109.
de Mello, A.J., 2006. Control and detection of chemical reactions in microuidic
systems. Nature 442, 394402.
Glasgow, I., Aubry, N., 2003. Enhancement of microuidic mixing using time pulsing. Lab Chip 3, 114.
Jiang, F., Drese, K.S., Hardt, S., Kupper, M., Schonfeld, F., 2004. Helical ows and
chaotic mixing in curved micro channels. AIChE J. 50, 22972305.
Kane, A.S., Hoffmann, A., Baumgartel, P., Seckler, R., Reichardt, G., Horsley, D.A.,
Schuler, B., Bakajin, O., 2008. Microuidic mixers for the investigation of rapid
protein folding kinetics using synchrotron radiation circular dichroism spectroscopy. Anal. Chem. 80, 95349541.
Knight, J.B., Vishwanath, A., Brody, J.P., Austin, R.H., 1998. Hydrodynamic focusing
on a silicon chip mixing nanoliters in microseconds. Phys. Rev. Lett. 80,
38633866.
Kumaran, V., Muralikrishnan, R., 2000. Spontaneous growth of uctuations in the
viscous ow of a uid past a soft interface. Phys. Rev. Lett. 84, 33103313.
Kumaran, V., Fredrickson, G.H., Pincus, P., 1994. Flow induced instability at the interface between a uid and a gel at low reynolds number. J. Phys. France II 4,
893911.
Kumaran, V., 2003. Hydrodynamic stability of ow through compliant channels and
tubes. In: Carpenter, P.W., Pedley, T.J. (Eds.), Proceedings of IUTAM Symposium

on Flow in Collapsible Tubes and Past Other Highly Compliant Boundaries.

Kluwer Academic Publishers, London.
Kumaran, V., 1995. Stability of the viscous ow of a uid through a exible tube. J.
Fluid Mech. 294, 259281.
Kumaran, V., 1996. Stability of an inviscid ow in a exible tube. J. Fluid Mech. 320,
117.
Kumaran, V., 1998. Stability of wall modes in a exible tube. J. Fluid Mech. 362,
115.
Kumaran, V., 2015. Experimental studies on the ow through soft tubes and
channels. Sadhana 40, 911923.
Lee, S., Kim, D., Lee, S., Kwon, T., 2006. A split and recombination micromixer
fabricated in a pdms three-dimensional structure. J. Micromech. Microeng. 16,
1067.
Lee, M.G., Choi, S., Park, J.-K., 2010. Rapid multivortex mixing in an alternately
formed contraction-expansion array microchannel. Biomed. Microdevices 12,
10191026.
Liu, R.H., Stremler, M.A., Sharp, K.V., Olsen, M.G., Santiago, J.G., Adrian, R.J., Aref, H.,
Beebe, D.J., 2000. Passive mixing in a three-dimensional serpentine microchannel. J. Microelectromech. Syst. 9, 190197.
Maharudrayya, S., Jayanti, S., Deshpande, A.P., 2004. Pressure losses in laminar ow
through serpentine channels in fuel cell stacks. J. Power Sources 138, 113.
Mensing, G.A., Pearce, T.M., Graham, M.D., Beebe, D.J., 2004. An externally driven
magnetic micro-stirrer. Philos. Trans. R. Soc. Lond. A 362, 10591068.
Muralikrishnan, R., Kumaran, V., 2002. Experimental study of the instability of the
viscous ow past a exible surface. Phys. Fluids 14, 775780.
Patel, V.C., Head, M.R., 1969. Some observations of skin friction and velocity proles
in fully developed channel and pipe ows. J. Fluid Mech. 38, 181201.
Posner, J., Santiago, J.J., 2006. Convective instability of electrokinetic ows in a
cross-shaped microchannel. J. Fluid Mech. 555, 142.
Reynolds, O., 1883. An experimental investigation of the circumstances which determine whether the motion of water in parallel channels shall be direct or
sinuous and of the law of resistance in parallel channels. Philos. Trans. R. Soc.
174, 935982.
Rosa, P., Karayiannis, T.G., Collins, M.W., 2009. Single-phase heat transfer in microchannels the importance of scaling effects. Appl. Therm. Eng. 29, 34473468.
Shankar, V., Kumaran, V., 1999. Stability of non-parabolic ows in a exible tube. J.
Fluid Mech. 395, 211236.
Shankar, V., Kumaran, V., 2000. Stability of non-axisymmetric modes in a exible
tube. J. Fluid Mech. 407, 291314.
Shankar, V., Kumaran, V., 2001. Weakly nonlinear stability of viscous ow past a
exible surface. J. Fluid Mech. 434, 337354.
Shankar, V., 2015. Stability of uid ow through deformable tubes and channels an
overview. Sadhana 40, 925943.
Srinivas, S.S., Kumaran, V., 2015. After transition in a soft-walled microchannel. J.
Fluid Mech. 780, 649686.
Stroock, A.D., Dertinger, S.K.W., Ajdari, A., Mezic, I., Stone, H.A., Whitesides, G.M.,
2002. Chaotic mixer for microchannels. Nature 295, 647651.
van Berkel, C., Gwyer, J.D., Deane, S., Green, N., Holloway, J., Hollisb, V., Morgan, H.,
2011. Integrated systems for rapid point of care (poc) blood cell analysis. Lab
Chip 11, 12491255.
Verma, M.K.S., Kumaran, V., 2012. A dynamical instability due to uid-wall coupling
lowers the transition reynolds number in the ow through a exible tube. J.
Fluid Mech. 705, 322347.

168

V. Kumaran, P. Bandaru / Chemical Engineering Science 149 (2016) 156168

Verma, M.K.S., Kumaran, V., 2013. A multifold reduction in the transition reynolds
number, and ultra-fast mixing, in a micro-channel due to a dynamical instability induced by a soft wall. J. Fluid Mech. 727, 407455.
Verma, M.K.S., Kumaran, V., 2015. Stability of the ow in a soft tube de- formed due
to an applied pressure gradient. Phys. Rev. E 91, 043001.

Whitesides, G., 2006. The origins and the future of microuidics. Nature 442,
368373.
You, J.B., Kang, K., Tran, T.T., Park, H., Hwang, W.R., Kim, J.M., Im, S.G., 2015. Pdmsbased turbulent microuidic mixer. Lab Chip 15, 17271735.