Instability of An Inhomogeneous Bacterial Suspension Subjected To A Chemoattractant Gradient PDF

J. Fluid Mech. (2014), vol. 741, pp 619–657.
c Cambridge University Press 2014 619

doi:10.1017/jfm.2013.628
Instability of an inhomogeneous bacterial

Downloaded from https://www.cambridge.org/core. University of Washington Libraries, on 26 Jul 2017 at 05:22:35, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
suspension subjected to a chemo-attractant

gradient
T. V. Kasyap and Donald L. Koch†
School of Chemical and Biomolecular Engineering, Cornell University, Ithaca, NY 14853, USA
(Received 3 February 2013; revised 6 November 2013; accepted 26 November 2013;

first published online 17 February 2014)
The stability of a suspension of chemotactic bacteria confined in an infinitely long

channel and subjected to a stationary, linear chemo-attractant gradient is investigated.
While swimming, individual bacteria exert force dipoles on the fluid, which at the
continuum level lead to a stress depending upon the bacterial orientation and number
density fields. The presence of the attractant gradient causes bacteria to tumble less
frequently when swimming along the gradient, leading to a mean orientation and
a non-zero chemotactic drift velocity U0 in that direction. At long length and time
scales compared to those associated with the persistence of bacterial swimming,
fluxes due to chemotaxis and the random run–tumble motion of bacteria balance
to yield an exponentially varying number density profile across the channel in the
base state. The associated bacterial stress field is also exponentially varying and is
normal. This spatially non-uniform base state is unstable to fluctuations in the bacterial
concentration field when the scaled bacterial concentration β = (3C/8)hn0 iL2 H exceeds
a critical value determined by a Péclet number defined as Pe = U0 H/κ. Here, C is
a non-dimensional dipole strength, which depends on the geometry of the bacterium,
hn0 i is the bacterial concentration averaged across the channel of depth H, L is the
total length of the bacterium, κ is the bacterial diffusivity, and βcrit is a monotonically
decreasing function of Pe, with βcrit ∼ 720/Pe3 for Pe 1 and βcrit ∼ 2 for Pe 1.
The instability is the result of the coupling between the active stress-driven fluid
flow and the bacterial concentration, and manifests as rectangular convection patterns.
When β first exceeds βcrit , the unstable wavelengths are large with λ H and the
mode of instability is stationary. Although oscillatory modes appear when λ 6 O(H)
and β > 247, the most dangerous mode of instability is found to be always stationary
with a wavelength λm /H ∼ Pe−1 . To study the coupling between the previously
analysed orientation shear instability mechanism of bacterial suspensions and the
new chemotaxis-driven instability, a new set of continuum equations that consistently
account for weak chemotaxis, rotation of bacteria by weak fluid shear and weak
non-continuum effects along with their coupled effects has been derived. The stability
analysis of those equations showed that the orientation shear mechanism has only a
negligible influence on the critical concentration for the present chemotaxis-induced
instability when the suspension depth is large, and it is the latter that has the lowest
critical concentration.
https://doi.org/10.1017/jfm.2013.628
Key words: instability, micro-organism dynamics, suspensions
† Email address for correspondence: dlk15@cornell.edu

620 T. V. Kasyap and D. L. Koch
1. Introduction
Recent experimental, numerical and theoretical investigations indicate that the
stresses that swimming bacteria exert on the fluid give rise to hydrodynamic
instabilities and collective motion in sufficiently dense bacterial suspensions (Koch
& Subramanian 2011). In a chemically neutral environment, many bacteria such as
Escherichia coli perform an unbiased random walk comprising intervals of persistent
swimming lasting about a second punctuated by short intervals of ‘tumbling’, after
which they pick a new swimming direction. In the presence of a chemical gradient,
bacteria bias their random walk by reducing the tumbling frequency when swimming
up the chemical gradient, and leaving it unaltered when swimming down, resulting
in a net bacterial migration along the chemical gradient (Berg 2003). Since the
aforementioned process, ‘chemotaxis’, is often critical to the survival of bacteria,
its effects on the stability of the suspension is a pertinent question. In a previous
letter (Kasyap & Koch 2012), we have shown through a long-wavelength lubrication
analysis that a bacterial suspension confined in a microchannel or a thin liquid film
and subjected to a transverse chemo-attractant gradient is unstable to perturbations of
bacterial concentration parallel to the channel or film if the bacterial concentration
exceeds a critical value, due to the coupling between the bacterial number density
field and the active stress-driven fluid flow. In this paper, we perform a more detailed
stability analysis in which we consider finite wavelengths and cover a broad range
of parameter space. The primary objective of this paper is to show the existence
of a most dangerous mode for which the growth rate is maximum and a cut-off
wavenumber above which disturbances do not grow. Along with the scaling for
the most dangerous and cut-off wavenumbers, we demonstrate the existence of an
oscillatory instability for certain parameter values. In addition, we examine the
effect of the rotation of bacteria due to disturbance fluid shear fields, which was
neglected in Kasyap & Koch (2012). Shear rotation leads to orientational instabilities
in suspensions of swimming bacteria and self-propelled rods (Simha & Ramaswamy
2002; Saintillan & Shelley 2008a,b; Subramanian & Koch 2009) and its inclusion
in the present analysis allows us to explore both the number density chemotaxis
instability and the orientational instability.
Early studies of the instability of chemotactic bacteria were either reaction–diffusion
studies of bacteria that excreted or consumed the chemo-attractant in which
hydrodynamic effects were neglected (Budrene & Berg 1991, 1995; Brenner, Levitov
& Budrene 1998) or studies of gravitational instabilities leading to bioconvection due
to suspension mass density variations induced by chemotaxis (Pedley & Kessler
1992; Hillesdon & Pedley 1996; Jánosi, Kessler & Horváth 1998). While the
typical suspension depth required for bioconvection is of the order of millimetres
or centimetres (Jánosi et al. 1998), macroscopic fluid motion has been observed
even in the case of chemotactic bacterial suspensions confined in liquid films or
microchannels for which gravitational effects are negligible (Kim & Breuer 2007;
Sokolov et al. 2009). For example, Kim & Breuer (2007) observed a significant
enhancement in the effective diffusivity of a fluorescent dye in a suspension of
https://doi.org/10.1017/jfm.2013.628
wild-type E. coli in the presence of a horizontal chemo-attractant or chemo-repellent

gradient. An increase in the scalar mixing indicates the existence of macroscopic
stirring of the suspension. Sokolov et al. (2009) studied the dynamics of thin films of
aqueous suspensions of aerobic Bacillus subtilis cells and observed a transition from
a spatially homogeneous, quasi-two-dimensional, coherent motion in the plane of
the film to a strongly convective, spatially inhomogeneous, three-dimensional motion
Instability of an inhomogeneous bacterial suspension 621
across the film when the film thickness was raised above a critical value. The onset
of this convective motion coincided with the development of cross-film gradients
of oxygen, and the bacterial concentration was strongly inhomogeneous across the
film thickness due to chemotaxis towards oxygen. At the critical film thickness, the
Rayleigh number was much lower than the value derived by Hillesdon & Pedley
(1996) for the onset of bioconvection. This leads to the hypothesis that the bulk
convective motions observed in the experiment of Sokolov et al. (2009) was driven
by the active stresses that bacteria exert on the fluid as they swim. The analysis
presented in Kasyap & Koch (2012) shows that this is indeed the case.
The long-range fluid velocity disturbance due to a swimming bacterium results
from a force dipole or stresslet due to the absence of any net force or torque acting
on the bacterium. The fluid disturbance created by this stresslet is extensile for a
bacterium like E. coli, since its flagella bundle pushes the cell body from behind
(Hernandez-Ortiz, Stoltz & Graham 2005; Saintillan & Shelley 2007; Underhill,
Hernandez-Ortiz & Graham 2008; Lauga & Powers 2009). Observations of large-scale
fluid motion in experimental studies of bacterial suspensions (Mendelson et al. 1999;
Wu & Libchaber 2000; Dombrowski et al. 2004; Kim & Breuer 2004; Sokolov et al.
2007) and in many-body simulations of hydrodynamically interacting self-propelled
particles (Hernandez-Ortiz et al. 2005; Saintillan & Shelley 2007; Underhill et al.
2008) strongly suggest that the swimming-induced hydrodynamic disturbances lead
to collective motion in bacterial suspensions. At the continuum level, the effect of
force dipoles is to provide a stress field as a function of the local orientation and
concentration fields of bacteria forcing the Stokes equations governing the fluid
motion. The orientation and concentration fields of bacteria are in turn coupled to the
fluid motion, owing to the translation of bacteria by the fluid flow and rotation by the
fluid velocity gradients. Other factors that affect the microstructure of the suspension
are bacterial swimming and effects such as tumbling as in the case of wild-type
bacteria or rotational diffusion as in the case of smoothly swimming bacteria (Simha
& Ramaswamy 2002; Saintillan & Shelley 2008a,b; Subramanian & Koch 2009;
Hohenegger & Shelley 2010).
The active stress field has been shown to give rise to instabilities in isotropic
suspensions of swimming bacteria (Subramanian & Koch 2009) and other self-
propelled particles (Simha & Ramaswamy 2002; Saintillan & Shelley 2008a,b;
Hohenegger & Shelley 2010) owing to the coupling between the bacterial orientation
field and the fluid flow. The physical mechanism of the instability is that the shear
field associated with a disturbance flow in the suspension rotates bacteria into the
extensional quadrant of the disturbance flow, and the intrinsic force dipoles on bacteria
then reinforce the original fluid disturbance, resulting in instability. This orientation
shear instability mechanism in the context of bacteria performing chemotaxis has
been analysed by Subramanian, Koch & Fitzgibbon (2011), in which it has been
found that the biased tumbling of bacteria makes the base-state bacterial orientation
field anisotropic, with a mean orientation parallel to the chemical gradient. Previous
researchers (Simha & Ramaswamy 2002; Saintillan & Shelley 2008a,b) had also
considered the stability of a state of perfect alignment of suspensions of self-propelled
https://doi.org/10.1017/jfm.2013.628
rods, but without considering a mechanism (such as chemotaxis) for the maintenance
of this orientational order. It may also be noted that chemotaxis leads to a bias in the
orientation distribution of the cells rather than the perfect initial alignment considered
by Simha & Ramaswamy (2002) and Saintillan & Shelley (2008a,b). Recently, Lushi,
Goldstein & Shelley (2012) have analysed the stability and the nonlinear evolution of
a uniform isotropic suspension of chemotactic bacteria that are capable of producing
the chemo-attractant themselves. Their primary observation is the existence of two
independent mechanisms of linear instability, one being the usual orientation shear
instability arising from the fluid-shear-induced rotation of bacteria, and the other an
aggregation instability originating from the reaction–diffusion coupling between the

bacterial concentration field and the chemo-attractant concentration field.
The above-mentioned studies are limited to the case of unbounded and homogeneous
suspensions in which the long-wavelength hydrodynamic instability is purely due to
the coupling between the bacterial orientation field and the fluid flow, without
a perturbation to the number density field. Even in the case of suspensions of
attractant-producing bacteria analysed by Lushi et al. (2012), the hydrodynamic
instability at linear order involves only orientation fluctuations, and the bacterial
number density fluctuations at linear order are due to chemotactic aggregation, with
no influence of hydrodynamics. However, since the active stress is a function of both
number density and orientation fields, one can conceive a situation where number
density perturbations couple with fluid flow, resulting in instability, without needing
to have orientational fluctuations. For this to happen at the linear order, the base-state
number density needs to be inhomogeneous, a state that naturally emerges in the
presence of a chemo-attractant gradient. The mean bacterial orientation parallel to
the attractant gradient resulting from the biased tumbling of the bacteria gives rise
to a net ‘chemotactic’ velocity in the same direction driving a convective flux. There
is also a diffusive bacterial flux due to the unbiased part of bacterial tumbling
(Keller & Segel 1971). If impermeable walls or interfaces are present normal to the
chemical gradient, as in the case of bacterial suspensions studied by Sokolov et al.
(2009), the convective and diffusive fluxes balance and, consequently, the equilibrium
bacterial concentration and active stress profiles across the channel will be exponential
(Kasyap & Koch 2012). Furthermore, the anisotropic bacterial orientation field makes
the bacterial stress normal with a compressive normal stress in the cross-channel
direction and a tensile normal stress in the longitudinal direction. Above a critical
value of bacterial concentration, this normal stress difference sets up cellular flow
patterns when a perturbation to the number density field occurs, and the resulting
flow transports bacteria in such a way as to reinforce the original perturbation. The
critical concentration for instability depends upon the strength of chemotaxis, bacterial
diffusivity, the channel depth and the bacterial geometry (Kasyap & Koch 2012).
In our previous long-wavelength analysis (Kasyap & Koch 2012), the growth rate
of the instability was found to increase quadratically with the wavenumber. However,
for large enough wavenumbers, one expects diffusion to stabilize the suspension and
a mode of maximum growth rate should exist. If one neglects nonlinear effects, the
preferred length scale of the instability is likely to be set by the most dangerous mode
and hence its wavelength is a quantity of practical relevance (Chandrasekhar 1961). In
addition, the information on the cut-off wavenumber above which disturbances do not
grow is useful in numerical simulations. In the present paper, we address these through
numerical solutions based on spectral methods, since the governing equations with
exponentially varying base-state number density defy any simple analytical solution
https://doi.org/10.1017/jfm.2013.628
for finite wavenumbers. For the special case of deep channels, the governing equations
will be solved by using matched asymptotic expansions. Motivated by the extensive
literature on shear orientation instabilities, we perform a detailed analysis of the effect
of shear rotation of bacteria on the present instability mechanism. We find that shear
rotation in general has a destabilizing effect, and, below a critical suspension depth,
the critical concentration is determined by the shear rotation instability mechanism.
Above this critical depth, the present mechanism of instability starts to influence the
critical concentration, and for large suspension depths the effect of shear rotation is
negligible.
This paper is organized as follows. In § 2 we pose the problem, derive the governing
equations, and introduce the physical mechanism of the instability by revisiting
the long-wavelength analysis made in Kasyap & Koch (2012). Next, in § 3, we
present an asymptotic analysis of a thin layer of bacteria confined in a deep channel
and then proceed to a semi-analytical calculation for the case of infinitely deep
channels. A detailed numerical calculation using spectral methods for the case of
finite wavenumbers is presented in § 4, followed by the analysis of shear rotation
in § 5. We begin § 5 by presenting a consistent derivation of governing equations
that includes chemotaxis and shear rotation, and we then perform a linear stability
analysis of those equations in the long-wavelength limit to investigate the effect of
shear rotation. Finally, in § 6 we discuss our results from the perspective of the
present literature on the subject.
2. Formulation
We consider a quiescent bacterial suspension between two infinitely long,
impermeable parallel walls at z = 0 and z = H subjected to a steady linear
chemo-attractant gradient in the −z direction. Let x be the coordinate along the
channel. Even though alternative ways of locomotion and chemotaxis have been
discovered among bacteria (see Stocker 2011), we focus here on the case of
run–tumble exhibited by the common species E. coli and B. subtilis, whose motility
and chemotaxis are well characterized. In order to provide a minimal physical
mechanism for the instability, we neglect the consumption and production of the
attractant by bacteria. In fact, it is not necessary for an attractant to be consumed in
order to elicit a chemotactic response. For instance, Me-Asp (α-methyl-dl-aspartate) is
a well-known attractant that is non-metabolizable (Mesibov & Adler 1972; Budrene
& Berg 1991). Production of the attractant is not a generic phenomenon, as it often
requires special situations such as the presence of environmental stresses (Budrene
& Berg 1991, 1995; Brenner et al. 1998). We also assume that the gradient remains
steady even after the onset of instability, since typical attractant diffusivities are much
larger than the bacterial diffusivity. A quantitative justification for this assumption is
provided in appendix A. We further consider bacterial suspensions that are sufficiently
dilute to be described by the singlet probability density Ω(x, p, t), which is defined
such that Ω(x, p, t) dx dp is the probability to find a bacterium at a position near
x and an orientation near p. The probability density Ω(x, p, t) satisfies the kinetic
equation (Subramanian & Koch 2009)
1
Z
∂Ω Ω Ω
+ ∇ · [(Us p + u)Ω] + ∇ p · (ṗΩ) + − dp = 0 (2.1)
∂t τ 4π τ
and the normalization constraint Ω(x, p, t) dp = n(x, t), where n(x, t) is the number
R
https://doi.org/10.1017/jfm.2013.628
density of bacteria. Here u(x, t) is the fluid velocity field, so that the second term in
(2.1) represents the advection of the probability distribution due to bacteria swimming
with a speed Us and the fluid flow. The third term is the divergence of the orientational
flux associated with the rotation of the bacteria at a rate ṗ due to fluid velocity
gradients. The rotation rate ṗ is given by ṗ = ω · p + F (r)(I − pp) · (e · p) (Kim &
Karrila 1991), where ω = 12 (∇u† − ∇u), e = 12 (∇u† + ∇u) and F (r) = (r2 − 1)/(r2 + 1),
with r = L/d the aspect ratio of the bacterium defined as the ratio of the overall length
of the bacterium (L ≈ 10 µm) including the flagella bundle to the bacterial diameter
(d ≈ 1 µm) (Subramanian & Koch 2009). The final term on the right-hand side of
(2.1) represents the rate of change in Ω(x, p, t) due to the tumbling of bacteria
with a frequency τ −1 . This is equal to the difference between the rate at which
bacteria tumble away from orientations near p and the rate at which bacteria tumble
into orientations near p. Since the statistics of the tumbling process is Poissonian,
the former rate is simply equal to Ω/τ (Subramanian & Koch 2009). Assuming
Rthat the time evolution of Ω(x, p, t) is Markovian, the latter rate is given by
[K(p | p0 )Ω(x, p0 , t)/τ ] dp0 , where K(p | p0 ) is the transition probability between
the pre- and post-tumble orientations, which is equal to 1/(4π) if the pre- and
post-tumble orientations are uncorrelated (Subramanian & Koch 2009).
In the absence of any fluid flow, and at time and length scales larger than the
persistence time and length scales, τ0 and Us τ0 , of a bacterium’s swimming, the
dominant balance in (2.1) is between the rates of bacteriaR coming into the phase space
and leaving the phase space, so that (Ω/τ ) − (1/(4π)) (Ω/τ ) dp = 0 (Subramanian
et al. 2011; Kasyap & Koch 2012). This allows one to separate Ω(x, p, t) into a
time-dependent number density field n(x, t) and a quasi-steady orientation field f (p)
as (Kasyap & Koch 2012)
Ω(x, p, t) = n(x, t) f (p), (2.2)
where f (p) is governed by the integral equation
f (p) 1 f (p)
Z
= dp. (2.3)
τ 4π τ
It is natural to seek a treatment at long length and time scales, since chemotaxis
manifests only after a bacterium has experienced many tumbles. In this limit the
separation between the orientation and number density fields is possible since the
swimming term Us p · ∇Ω does not appear in the dominant balance in (2.1), and the
only way in which the number density can be correlated with the orientation field in
the absence of fluid flow is through bacteria swimming.
In the presence of a stationary chemo-attractant gradient, the biphasically biased
bacterial tumbling frequency τ −1 is given by (Rivero et al. 1989; Chen, Ford &
Cummings 2003; Kasyap & Koch 2012)
τ0−1 (1 − ζ p · g), p · g > 0,

−1
τ = (2.4)
τ0−1 , p · g 6 0,
where ζ = χUs G and g = −ez is the unit vector parallel to the chemical gradient. Here
G is the strength of the chemical gradient, and χ characterizes the sensitivity of the
bacterium to the attractant gradient, and is a function of the number of receptors on
https://doi.org/10.1017/jfm.2013.628
the bacterial cell body to which the attractant molecules bind, the binding kinetics
and the gain of the bacterium’s signalling system (Rivero et al. 1989; Subramanian
et al. 2011). The above equation is the linearized version for ζ 1 of the exponential
chemotactic response (see Subramanian et al. 2011) originally proposed by Rivero
et al. (1989). Inserting (2.4) into (2.3) and solving for f (p) yields (Kasyap & Koch
2012)
1

[1 + (p · g − 1/4)ζ ], p · g > 0,


4π

f (p) = (2.5)
1
[1 − ζ /4], p · g 6 0.


4π

For 0 < ζ 1 the orientation field described in (2.5) is weakly anisotropic, peaking at
g and resulting in a non-trivial mean orientation hpif parallel to the chemical gradient
and a chemotactic drift velocity U0 in the same direction (Kasyap & Koch 2012)
ζ
hpif = g, U0 = Us hpif = −U0 ez , (2.6)
6
where U0 = Us ζ /6. Here the symbol h if denotes an average over orientation space
weighted by the bacteria’s orientation distribution. Experimental observations indicate
that a typical chemotactic velocity is ∼10 % of the swimming speed (Kalinin et al.
2009; Vuppula, Tirumkudulu & Venkatesh 2010) and the chemotactic velocity can
be adjusted to still smaller values using microfluidic devices that generate controlled
chemical gradients, such as the one described by Cheng et al. (2007). Thus the
quantitative predictions made in this paper with the small-ζ approximation can be
expected to be reasonably accurate. The chemotactic velocity leads to a convective
bacterial flux U0 n parallel to the chemical gradient. In addition to this, for length
scales much larger than the bacterial persistence length, the unbiased part of the
bacterial random walk contributes a diffusive flux −κ · ∇n, so that the conservation
equation for the bacterial number density takes the form of a convection–diffusion
equation (Keller & Segel 1971; Bearon & Pedley 2000; Chen et al. 2003; Kasyap &
Koch 2012)
∂n
+ ∇ · [(U0 + u)n − κ · ∇n] = 0. (2.7)
∂t
Here the diffusivity tensor κ = κk gg + κ⊥ [I − gg], where κk and κ⊥ are in general
functions of ζ . However, for ζ 1, it can be shown that the leading-order diffusion
coefficient reduces to a scalar given by κ = Us2 τ0 /3 (Bearon & Pedley 2000; Bearon
2003; Chen et al. 2003).
In the continuum theory of bacterial suspensions (Simha & Ramaswamy 2002;
Saintillan & Shelley 2008a,b; Subramanian & Koch 2009), the fluid velocity u(x, t)
and pressure p(x, t) satisfy the Stokes equations with an additional bacterial stress
field σ B :
∇ · u = 0, (2.8a)
− ∇p + µ∇ u + ∇ · σ B = 0.
2
(2.8b)
Note that one can define a macroscopic Reynolds number based on the channel depth
and chemotactic velocity Re = U0 H/ν in addition to the microscopic Reynolds number
based on the bacterial length and swimming speed. Since the former describes the
unstable fluid flows generated by the chemotaxis of bacteria, U0 is the appropriate
https://doi.org/10.1017/jfm.2013.628
velocity scale. For typical values U0 = 2 µm s−1 and ν = 0.01 cm2 s−1 , then Re 1,
justifying the use of the Stokes equations when H is smaller than 1 cm. The deviatoric
bacterial stress σ B in (2.8) is given by (Subramanian & Koch 2009)
I
Z
B 2
ff (x, p, t) = −CµUs L Ω pp − dp, (2.9)
3
where C ≈ 0.57 is the non-dimensional dipole moment of the bacteria. Using (2.2)
and (2.5), one gets
σ B (x, t) = n(x, t)S , (2.10)
where the dipole moment S is given by (Kasyap & Koch 2012)
C I

2
S = − µUs L ζ gg − . (2.11)
16 3
Being a pusher, a bacterium exerts a pressure in the z direction (negative normal

stress) and tension (positive normal stress) in the x and y directions corresponding
to Sxx = Syy = − 12 Szz (Kasyap & Koch 2012).
The boundary condition for the number density equation (2.7) is the no-flux
condition
U0 n − κ∇n = 0 at z = 0, H. (2.12)
While it has been observed recently that the hydrodynamic and steric interactions
between swimming bacteria and solid walls give rise to interesting phenomena, such
as swimming in circular trajectories (Lauga et al. 2006), preferential swimming
towards the right-hand side (DiLuzio et al. 2005), accumulation near walls (Berke
et al. 2008; Li & Tang 2009) and motility reversal (Cisneros et al. 2006), we expect
these effects to be confined within a length scale of Us τ0 from the wall so that the
no-flux boundary condition (2.12) is appropriate for the long-length-scale treatment
presented here. For the fluid velocity field, we use the no-slip boundary condition
u=0 at z = 0, H (2.13)
for most of our calculations. Bacteria often produce surfactants (see Angelini et al.
2009), and in those situations the no-slip boundary condition (2.13) is applicable
even for suspensions confined in liquid films, since the Marangoni stress due to
surfactants prevents dilatation of the interfaces, making them effectively slip-free
for two-dimensional fluid flows such as those arising in our stability analysis. We
nevertheless briefly illustrate the effect of having a slip boundary on this instability
in § 3.
The governing equations (2.7) and (2.8) together with the boundary conditions
(2.12) and (2.13) admit the stationary base-state solution (Kasyap & Koch 2012)
U0 H −U0 z

hn0 i exp
κ κ
n0 (z) = (2.14)
−U0 H

1 − exp
κ
and
p0 (z) = Szz n0 (z), (2.15)
https://doi.org/10.1017/jfm.2013.628
where the subscript ‘0’ indicates the base state and the angle brackets now indicate
the average over the channel cross-section. Here hn0 i serves as a normalization
constant. Thus, in the base state, the diffusive flux due to bacterial random walk
and the chemotactic flux balance, yielding an exponentially varying bacterial number
density profile across the channel, and the resulting bacterial stress is balanced by
the pressure.
We now study the stability of this base state against small perturbations of the
normal mode form
(n0 , u0 , w0 , p0 ) = [N(z), U(z), W(z), P(z)] exp(ikx + σ t), (2.16)

in which u0 and w0 are the perturbations in the fluid velocity in the x and z directions,
respectively, and the wavenumber k = 2π/λ, with λ being the wavelength, is assumed
to be real. The growth rate σ is allowed to be complex and the base state is unstable
if Re(σ ) is positive. Substituting (2.16) followed by the elimination of U(z) and P(z)
in favour of W(z) yields the following non-dimensional equations for the z-dependent
perturbation amplitudes:
[D2 + PeD − (σ + k2 )]N = PeWDn0 , (2.17a)
(D2 − k2 )2 W = −βk2 DN, (2.17b)
(Pe + D)N = W = DW = 0 at z = 0, z = 1. (2.17c)
Here D indicates differentiation with respect to z and n0 (z) = Pe exp(−Pez)/[1 −
exp(−Pe)]. For non-dimensionalization, we have used the characteristic scales H for
length, 1/H for wavenumber, U0 for velocity, hn0 i for bacterial number density, H 2 /κ
for time and − 32 Szz hn0 i for pressure.
Equation (2.17) shows that the physics of the instability is governed by two non-
dimensional parameters, the Péclet number defined as
U0 H H U0
Pe = =3 (2.18)
κ Us τ0 Us
and
3Szz hn0 iH 3
β =− = Chn0 iL2 H. (2.19)
2µU0 8
Thus Pe measures the strength of the chemotactic migration relative to the random
bacterial motion and β is the ratio of the bacterial stress to the viscous stress.
Equation (2.19) indicates that the parameter β can also be viewed as a scaled
bacterial concentration in the channel, a relation obtained by using (2.6) and (2.11).
Parameters β and Pe are related by
C Us τ0 Us

3
β = Pehn0 iL . (2.20)
8 L U0
For a typical wild-type bacterium like E. coli, Us τ0 /L = O(1), and using C = 0.57 and
the nominal value of U0 /Us as 0.1, one obtains β = O(Pehn0 iL3 ).
Equation (2.17) has a close resemblance to the governing equations for bioconvection
in Childress, Levandowsky & Spiegel (1975) and Hill, Pedley & Kessler (1989)
without gyrotaxis terms, except that the momentum equation (2.17b) here does not
contain the inertial term and it is the bacterial number density gradient on the
right-hand side of the equation that drives the fluid flow rather than the number
density itself as in bioconvection. The latter is a consequence of the fact that fluid
https://doi.org/10.1017/jfm.2013.628
flow here is driven by the divergence of the active stress rather than buoyancy. A
caveat in deriving the amplitude equations (2.17) is the neglect of the effects of
the bacterial rotation imparted by fluid velocity gradients. Shear rotation of bacteria
changes the orientation field in (2.5) and hence changes the bacterial stress tensor
σ B and the chemotactic drift velocity U0 . We address these later in § 5 by doing a
rigorous long-wavelength analysis including the shear rotation of bacteria. It is found
103
102
101
100
0 20 40 60 80 100
Pe
F IGURE 1. Neutral curve obtained from the long-wavelength analysis. The suspension is
unstable in the region that lies above the solid line. The asymptotes of βcrit are βcrit ∼
720/Pe3 for Pe 1 and βcrit ∼ 2 for Pe 1 (Kasyap & Koch 2012).
that the effect of shear rotation is negligible in determining the critical bacterial
concentration required for the instability if the suspension is deep enough.
The instability mechanism and the critical condition for the instability were obtained
in our previous publication (Kasyap & Koch 2012) through a classical lubrication
analysis of linearized governing equations by assuming that the wavelength of the
disturbance is much larger than the channel depth, λ H. It can be shown that the
small-wavenumber solution of (2.17) is equivalent to the lubrication solution given in
Kasyap & Koch (2012), since the small parameter for lubrication analysis, = H/λ,
used in Kasyap & Koch (2012) is related to the wavenumber through k = 2π. The
primary result of the lubrication analysis in Kasyap & Koch (2012) is an analytical
expression for the critical concentration βcrit required for the instability as a function
of the Péclet number as
1
βcrit = , (2.21)
Peα
where the quantity α is given by
1 exp(−Pe)(1 + Pe) − 1 Pe[1 + exp(−Pe)] + 2[exp(−Pe) − 1] 2

α= + −3 .
2Pe Pe2 [1 − exp(−Pe)] Pe2 [1 − exp(−Pe)]
(2.22)
The suspension becomes unstable if β exceeds βcrit , and figure 1 illustrates

the variation of βcrit as a function of Pe. It is clear from figure 1 that bacterial
chemotaxis is the sole reason behind the instability, since βcrit diverges in the limit of
vanishing chemotaxis, Pe → 0, and since increasing chemotaxis by increasing Pe for a
https://doi.org/10.1017/jfm.2013.628
given suspension depth decreases the critical bacterial concentration required for the
instability. The source term in the number density equation (2.17a) and the equation
for the z component of the fluid velocity (2.17b) indicate that the present instability
mechanism involves a coupling between the bacteria stress-driven fluid flow and the
bacterial concentration field. In particular, (2.17b) shows that perturbations in bacterial
concentration field drive fluid flows through the chemotactic bacterial stress, and the
(a) 1.00 (b) 1.00

0.75 0.75
z 0.50 0.50
0.25 0.25
0
0 50 100 150 200 –1 0 1 2 3 4 5 6
x
F IGURE 2. (a) Physical mechanism of the chemotaxis-driven instability with the fluid
disturbance created by a single bacterium aligned with the chemical gradient direction
(−z direction) shown at the bottom. This extensile fluid disturbance sets up convective
motions in response to sinusoidal fluctuations in the bacterial concentration field. The
latter is shown by the grey scale (colour online) shading, with brighter regions indicating
higher bacterial concentrations. The fluid motion acting on the base-state bacteria profile
brings more bacteria towards the region of high perturbed bacterial concentration, leading
to growth of the perturbation (Kasyap & Koch 2012). (b) Profiles of the base-state number
density field n0 (solid line), fluid velocity in the x direction u0 /k (dotted line) and the flux
u0 n0 at x = 75 (dashed line) in (a) for k = 0.02π, Pe = 5 and β = 25. The exponentially
decaying n0 (z) results in a net flux hu0 n0 i towards the brighter region.
aforementioned fluid flow when acting on the inhomogeneous base-state bacterial

concentration field reinforces the original bacterial concentration perturbations, thus
making the suspension unstable. This can be further elucidated by averaging the
number density equation (2.17a) in the z direction, which yields the expression
(σ + k2 )hNi = −PehWDn0 i = −ikhUn0 i, (2.23)
which makes use of the boundary conditions on the number density and fluid velocity
amplitudes and the continuity equation. Equation (2.23) reveals that the growth of the
z-averaged bacterial concentration perturbations is due to a net bacterial flux in the
x direction, hUn0 i, arising from the bacteria stress-driven fluid flow. To understand
how this net flux arises, consider figure 2, in which the grey scale (colour online)
shading represents the perturbed number density field, with brighter regions indicating
a positive perturbation and darker regions indicating a negative perturbation from the
mean number density. The associated spatially varying bacterial stress field drives a
https://doi.org/10.1017/jfm.2013.628
flow as shown by the streamlines. Since the hydrodynamic disturbance created by a

bacterium draws fluid inwards towards its sides and pushes fluid out at the front and
the back, the disturbance fluid flow near the bottom wall is into the regions of higher
number density. Fluid returns back to regions of lower number density near the top
wall. Since the base-state number density is strongly peaked at the bottom wall, more
bacteria are convected into brighter regions at small z than are convected back into
darker regions at large z, and there is a net bacterial flux towards brighter regions.
This is made clear in figure 2(b), in which we show the profiles of the base-state
number density field, the fluid velocity in the x direction and the convective bacterial
flux along the x direction. When this convective flux exceeds the diffusive flux of
bacteria from brighter regions to darker regions, the original concentration perturbation
is enhanced, resulting in instability. It is obvious that a suspension of pullers that
generate contractile fluid disturbances with β < 0 will be unconditionally stable against
this mechanism of instability since the hydrodynamic disturbance field of a puller is
exactly opposite to that of a pusher.
The growth rate of the instability obtained from the long-wavelength lubrication
analysis in Kasyap & Koch (2012) when expressed in terms of the characteristic time
scale H 2 /κ used in deriving (2.17) is
σ = k2 (Peαβ − 1) + O(k3 ), (2.24)
which is quadratic in the wavenumber. Thus the ‘most dangerous’ wavenumber,

defined as the wavenumber at which the growth rate is maximum, is not predicted
by the long-wavelength theory. This is due to the fact that, for long wavelengths, the
diffusion of bacteria in the x direction (the −k2 N term in (2.17a)) does not participate
in determining the leading-order number density field. Thus we need to relax the
restriction of long-wavelength disturbances to obtain a most dangerous wavenumber,
and we do this in the following two sections through an analytical treatment in the
limit Pe 1 and a full numerical solution for Pe = O(1). Although an analytical
solution is also possible for Pe 1, the asymptotic behaviour of βcrit for Pe 1 in
figure 2 indicates the impossibility of instability with realistic bacterial concentrations
in that regime, and so we do not pursue the analysis for a shallow channel. Instead,
we proceed to the analysis for deep channels for which Pe 1.
3. Theory for deep channels

When the depth of the channel is large, Pe 1, the exponential form of the
base-state number density n0 (z) implies that bacteria occupy only an O(1/Pe) layer
near the bottom boundary z = 0 of the channel. Thus the characteristic length
scale within this boundary layer is its thickness δ = κ/U0 and hence the validity
of continuum equations requires that δ Us τ0 . This condition is satisfied when
ζ 1 since the boundary layer thickness δ = κ/U0 = Us2 τ0 /(3U0 ) = 2Us τ0 /ζ Us τ0
for ζ 1. However, for a typical bacterial swimming speed of 20 µm s−1 and
tumbling at a frequency of 1 s−1 , κ/U0 = 66.67 µm is just 3.33 times larger
than the bacterial persistence length Us τ0 = 20 µm for the typical chemotactic
velocity U0 /Us = 0.1 observed in experiments (Kalinin et al. 2009; Vuppula et al.
2010), rendering the quantitative predictions made by the continuum theory only
approximate when Pe 1. Nevertheless, as mentioned earlier in § 2, the chemotactic
velocity can be made smaller in experiments by controlling the magnitude of the
chemical gradient using microfluidic techniques (Cheng et al. 2007), thus making
the boundary layer thickness κ/U0 large enough to make the present continuum
https://doi.org/10.1017/jfm.2013.628
theory quantitatively accurate. We solve the deep-channel problem by means of

matched asymptotic expansions in which the dependent variables are expanded
as a power series in 1/Pe for both the boundary layer (inner) and outer regions.
Subsequent matching of solutions in these two regions determines the growth
rate and eigenmodes (Childress et al. 1975; Hill et al. 1989). We first perform
the analysis for the case with the no-slip condition on both boundaries, and we
then proceed to the case with the z = 0 boundary being a shear-free interface and
z = 1 being a no-slip boundary. We choose to analyse the latter case since the z = 0
boundary condition has the strongest effect in this Pe 1 regime due to the fluid
flow being driven by the bacterial crowd in the boundary layer. We also provide a
discussion of the results at the end of this section. In the present section ‘case I’
refers to the problem with no-slip boundaries and ‘case II’ to the problem with a slip
boundary at z = 0 and a no-slip boundary at z = 1. In both cases, we focus on the
typical case with k and β being O(1) quantities.
3.1. Analysis of case I

We start with the outer region, for which the amplitude equations are
[Pe−1 (D2 − σ − k2 ) + D] N
b = 0, (3.1a)
(D − k ) W = −βk2 DN,
2 2 2b b (3.1b)
(1 + Pe−1 D)N
b =W b = DW b = 0 at z = 1, (3.1c)
which admit perturbation solutions N(z)
b = 0 for all powers of 1/Pe and
W(z)
b = [A + B(z − 1)] sinh[k(z − 1)] − kA(z − 1) cosh[k(z − 1)], (3.2)
with A = j=0 Pe−j Aj and B = ∞
P∞ −j
j=0 Pe Bj (Hill et al. 1989). The values of coefficients
P
Aj and Bj are to be obtained by matching the outer solution with the inner solution.
Since the characteristic vertical length scale in the inner region is the boundary layer
thickness δ = O(1/Pe), the inner coordinate variable z̄ is obtained by stretching the
outer coordinate variable z by a factor of Pe as z̄ = Pez and the amplitude equations
in the inner region take the form
2
[D + D − Pe−2 (σ + k2 )]N = −PeW exp(−z̄), (3.3a)
2
[D − k2 Pe−2 ]2 W = −Pe−3 βk2 DN, (3.3b)
(1 + D)N = W = DW = 0 at z̄ = 0, (3.3c)
N(z̄ → ∞) = 0, W(z̄ → ∞) = W(z b → 0), (3.3d)
where D is the differentiation with respect to z̄ and the last line in (3.3) gives the
conditions for matching with the outer solution. In order to solve these equations, one
needs to know the orders of magnitude of different variables. It is clear that N = O(1),
suggesting the expansion
∞
Pe−j N j (z̄),
X
N(z̄) = (3.4)
j=0
R∞
which should satisfy the arbitrary normalization condition 0
N(z̄) d z̄ = 1. Equation
(3.3b) then suggests the expansion for W(z̄) as
∞
Pe−(j+3) W j+3 (z̄),
X
W(z̄) = (3.5)
https://doi.org/10.1017/jfm.2013.628
j=0
and balancing the growth and coupling terms in (3.3a) gives the expansion for σ as
∞
Pe−j σj .
X
σ= (3.6)
j=0
Solving (3.3) with the above asymptotic expansions is straightforward and yields the
following:
N(z̄) = exp(−z̄) + O(Pe−2 ), (3.7a)

W(z̄) = Pe−3 βk2 [z̄ + exp(−z̄) − 1]
k − sinh k cosh k 2

−4 3
− Pe βk z̄ + O(Pe−5 ). (3.7b)
k2 − sinh2 k
One can now determine the growth rate using the solvability condition
Z ∞
σ= [Pe3 W exp(−z̄) − k2 N] dz̄ (3.8)
0
obtained by integrating (3.3a) across the boundary layer as

2βk3 k − sinh k cosh k

β
σ = k2 −1 − + O(Pe−2 ). (3.9)
2 Pe k2 − sinh2 k
Since σ is purely real, the instability is stationary in the Pe 1 limit. Solving (3.9)
for σ = 0 gives a critical β as
8k k − sinh k cosh k

βc = 2 + + O(Pe−2 ), (3.10)
Pe k2 − sinh2 k
which is different from the βcrit defined earlier in § 2, since the neutral curve here is
a function of the wavenumber also. One may now define βcrit as βcrit = min(βc ) and
from (3.10) βcrit = 2 since βc has the minimum value at k = 0. Thus we have recovered
the previous asymptotic result for Pe 1 from the long-wavelength analysis in Kasyap
& Koch (2012). We shall also see in § 4 that, for any Pe, the minimum of βc occurs
at k = 0.
3.2. Analysis of case II

Since the outer boundary condition is the same here as in case I, the outer solutions
for the number density and fluid velocity fields remain the same as N(z) b = 0 and
(3.2). The inner number density and velocity fields are still governed by (3.3) but
the no-slip boundary condition U = 0 changes to the no-shear-stress condition DU = 0
2
at z̄ = 0, which yields the boundary conditions W = D W = 0 at z̄ = 0 through the
incompressibility constraint. The expansion for the number density field is the same
as in (3.4), and to proceed further we need to know the scaling for the velocity field
in this case. Since the governing equations in the inner region are the same as in
case I, except for the boundary condition at z = 0, one might think that the same
scaling W ∼ Pe−3 would hold here too (see (3.5)). This scaling gives an O(Pe−2 ) fluid
velocity in the x direction, which satisfies the no-slip boundary condition at z = 0 and
varies across the boundary layer according to the scaled incompressibility constraint
ikU + PeDW = 0 within the boundary layer. However, unlike case I, the shear-free
https://doi.org/10.1017/jfm.2013.628
boundary condition here permits a fluid velocity in the x direction that is constant
across the boundary layer, which requires W = O(Pe−2 ) instead of the Pe−3 scaling.
This suggests the expansion
∞
Pe−(j+2) W j+2 (z̄).
X
W(z̄) = (3.11)
j=0
Equation (3.3a) then suggests the expansion for growth rate as
∞
Pe−j+1 σj−1 .
X
σ= (3.12)
j=0
Boundary layer solutions with the above-mentioned asymptotic expansions are

N(z̄) = exp(−z̄) + O(Pe−1 ), (3.13a)
k2 − sinh2 k z̄2

βk
W(z̄) = Pe−2 z̄ + Pe−3 βk2 exp(−z̄) − − 1 , (3.13b)
2 k − sinh k cosh k 2
and finally the solvability condition (3.8) gives the growth rate
βk(k2 − sinh2 k) 3β

2
σ = Pe −k 1+ + O(Pe−1 ). (3.14)
2(k − sinh k cosh k) 2
The critical β at finite wavenumber then obtained from (3.14) is
1 2k(k − sinh k cosh k)
βc = + O(Pe−2 ), (3.15)
Pe (k2 − sinh2 k)
and βc has the minimum value of βcrit = 4/Pe at k = 0.
3.3. Discussion
We now reflect upon the role of the boundary condition at z = 0 on the fluid
velocity field set up by the instability. The primary observation here is that the fluid
velocity in the z direction in case I is O(Pe−3 ) (see (3.7b)) while the same in case
II is O(Pe−2 ). The corresponding fluid velocities in the x direction are O(Pe−2 ) and
O(Pe−1 ), respectively, for case I and II, due to the scaled incompressibility constraint
ikU + PeDW = 0 within the boundary layer. Thus the zero tangential stress boundary
condition at z = 0 results in O(Pe) larger fluid velocity than in the no-slip case, and
the origin of these scalings can be understood from a dominant balance analysis of
the momentum equation. When k = O(1), the characteristic length scales in the x
and z directions are the channel thickness H and the boundary layer thickness δ,
respectively. Since δ H the fluid velocity in case I is determined by the balance
between the shear stress and the bacteria stress in the x momentum equation as
µu0 /δ 2 ∼ Sxx n0 /H. This yields the amplitude of the non-dimensional fluid velocity
in the x direction as U ∼ βPe−2 and the incompressibility constraint gives the fluid
velocity amplitude in the z direction as W ∼ (δ/H)U ∼ βPe−3 for case I. While most
of the bacterial stress in the boundary layer is utilized for overcoming the wall shear
stress induced by the no-slip boundary condition in case I, the lack of any wall shear
stress in case II leaves the entire bacterial stress available for driving the outer flow.
As a result, the magnitude of the fluid velocity field in case II, which is larger than
case I by a factor of Pe, can be obtained by balancing the shear stress µu0 /δ in the x
https://doi.org/10.1017/jfm.2013.628
momentum equation with the bacteria stress Sxx n0 , which gives the scaling U ∼ βPe−1
and consequently W = βPe−2 for case II. The associated βcrit from (3.10) and (3.15)
also differ by a factor of Pe, with βcrit = O(1) for case I and βcrit = O(Pe−1 ) for case II.
The difference between the magnitudes of the fluid velocity in cases I and II can
also be reasoned in terms of the images that the individual force dipoles of bacteria
create when they are near a no-slip or shear-free boundary. It is well known that
1.0
z 0.5
0
–2 0 2 4
F IGURE 3. Profiles of the leading-order fluid velocity in the x direction for case I (solid
line) and case II (dashed line) along with the leading-order number density perturbation
(dash-dotted line) at Pe = 50, β = 5 and k = 1. These profiles are computed by applying the
incompressibility constraint on the fluid velocity in the z direction in the outer and inner
layers; and in order to plot both cases I and II on the same graph, we rescale velocities
as Pe2 U(z) for case I and as PeU(z) for case II.
a stokeslet with the point force normal to a no-slip boundary creates a velocity
field with no net force dipole in the far field but rather acts as a force quadrupole
(Blake 1971). On the other hand, the image of a stokeslet normal to a shear-free
boundary is just another stokeslet with an opposite force, and hence the velocity field
is that of a net force dipole in the far field (Aderogba & Blake 1978). Since a force
dipole is composed of two forces equal in magnitude but opposite in direction, the
far-field velocity produced by a dipole near a no-slip boundary has no net dipolar or
quadrupolar component, while the far-field velocity due to a dipole near a shear-free
boundary is a force-dipole velocity field with twice the magnitude of the original
dipolar field. Thus the fluid disturbance created by a force dipole near a shear-free
boundary is much stronger than near a no-slip boundary.
Now the magnitude of the x fluid velocity in the outer region for both cases I
and II is set by the respective inner x velocities, since there is no driving force for
fluid flow in the outer region. Thus U b ∼ βPe−2 for case I and Ub ∼ βPe−1 for case II,
respectively. One important question now is whether the fluid flow resulting from the
instability is confined to the boundary layer or not. To answer this, we plot profiles
of the leading-order fluid velocity in the x direction in figure 3, and the observation
in figure 3 is that the flow indeed covers the entire channel for both cases I and II.
Thus the vertical length scale of fluid motion is of the same order as the horizontal
length scale, which in this case with k = O(1) is the suspension depth. This is a simple
consequence of the low-Reynolds-number nature of the fluid motion and the fact that
the fluid velocity profile in the outer region is determined primarily by the constraint
https://doi.org/10.1017/jfm.2013.628
of incompressibility.
3.4. Analysis for a semi-infinite channel

Equation (3.9) does not exhibit a most dangerous mode since the diffusion of bacteria
in the x direction and the growth of the perturbation do not arise at leading order in
the boundary layer bacteria conservation equation (3.3a) when k = O(1). In order to
have them at the leading order in (3.3a), one needs k = O(Pe) and σ = O(Pe2 ), so that
the most dangerous wavelength is of the order of the boundary layer thickness κ/U0 .
Since the fluid motion would be confined to the boundary layer rather than the whole
channel in this limit, we ignore the z = 1 boundary and instead solve the problem for
the semi-infinite domain covering the positive half-space. The result from this analysis
will serve as a tool for the validation of the numerical solution presented in the next
section. We would like to point out that the Stokes equations remain valid in this case
since the Reynolds number based on the boundary layer thickness is small. Thus we
solve the modified inner problem
[D2 + D − (σ̃ + k̃2 )]N = −W exp(−z), (3.16a)

2 2 2 2
[D − k̃ ] W = −β k̃ DN, (3.16b)
(1 + D)N = W = DW = 0 at z = 0, (3.16c)
N = DN = W = DW → 0 as z → ∞, (3.16d)
obtained after dropping overbars in (3.3) and rescaling σ and k to yield σ̃ = σ /Pe2 and
k̃ = k/Pe. Since the gap-averaged number density hn0 i vanishes in the limit of infinite
depth, one needs to rescale the number density by na U0 /κ, where na is the number
of bacteria per unit area. The quantity β can then be redefined as β = 83 Cna L2 .
To solve (3.16) we use a series method similar to the one used by Childress et al.
(1975) and expand the dependent variables as truncated series,
M
X
N(z) = {N1,j exp[−(m + j)z] + (N2,j + N3,j z) exp[−(k̃ + j)z]}, (3.17a)
j=0
M
X
W(z) = {W1,j exp[−(m + j)z] + (W2,j + W3,j z) exp[−(k̃ + j)z]}, (3.17b)
j=0
q
1
with m = 2
[1 + 1 + 4(σ̃ + k̃2 )], so that the solutions in (3.17) and their derivatives
decay as z → ∞ as indicated in the boundary conditions (3.16c) for σ̃ > 0. It is
also assumed that σ is purely real here. Here N2,0 = N3,0 = 0 and N1,0 exp(−mz)
is the decaying homogeneous solution of the number density equation (3.16a) with
the arbitrary amplitude N1,0 , which, without loss of any generality, is taken as
unity. Recurrence relationships between successive coefficients of functions of z in
series expansions in (3.17) can be obtained by equating the coefficients of respective
functions on both sides of (3.16a) and (3.16b), and the only unknowns in the problem
are the growth rate σ , and W2,0 and W3,0 , the coefficients of the homogeneous
solutions of (3.16b). The boundary condition at z = 0 in (3.16c) provides three
algebraic equations for these three unknowns, which are solved by Newton–Raphson
iteration. In figure 4(a) we show the comparison between the growth rates obtained
from the series solution and the long-wavelength lubrication theory given in (2.24)
at Pe 1, and the excellent agreement between the two in the figure establishes the
https://doi.org/10.1017/jfm.2013.628
validity of the series solution. Growth rates for finite values of wavenumber for the
unbounded suspension obtained from the series solution are given in figure 4(b). We
ensure the convergence of the series solution by noting the good agreement between
solutions with M = 10 and M = 20 in figure 4. As expected, for large enough k̃,
diffusion starts stabilizing the suspension with the associated decrease in the growth
rate, and eventually the suspension becomes stable after the cut-off wavenumber k̃cut ,
(a) (b)
4 0.8
3 0.6
40
2 0.4
30
1 0.2 20
10
0 0.002 0.004 0.006 0.008 0.010 0 0.2 0.4 0.6 0.8 1.0
F IGURE 4. Growth rates for semi-infinite suspensions obtained from the series solution.
(a) Comparison with the long-wavelength theory prediction given in (2.24) for M = 10
and β = 10. (b) Growth rates at finite k̃ for various β as indicated. Solid lines are with
M = 10 and asterisks are for M = 20.
defined as the non-zero horizontal intercepts of curves in figure 4(b). Within the
range 0 < k̃ < k̃cut , most dangerous modes exist for all cases with a wavenumber k̃m
defined as the point at which the growth rate is the maximum. Figure 4(b) shows
that k̃m = O(1) (though less than unity), which confirms that the wavelength of the
most dangerous mode is of the order of the boundary layer thickness κ/U0 .
4. Numerical solution for finite Pe and k

Determining the nature of solutions for Pe = O(1) and finite k requires full
numerical solution of the governing equations (2.17). As noted in § 3, such a solution
is required to find the fastest growing mode for Pe = O(1). We adopt a spectral
collocation method in which the dependent variable N(z) is expanded as a truncated
series of basis function φj (z) values that satisfy the no-flux boundary conditions,
M
X
N(z) = aj φj (z), (4.1)
j=0
where
φ (z) = exp(−Pez), (4.2a)

0
φj (z) = exp − Pe2 z sin jπz − 2jπ cos jπz , j = 1, 2, 3, . . . , (4.2b)

Pe
in which φ0 corresponds to the eigenfunction in the long-wavelength case (see

https://doi.org/10.1017/jfm.2013.628
equation (9) in Kasyap & Koch (2012)). These basis functions are the solutions
of the Sturm–Liouville problem closely related to the left-hand side of bacteria
conservation equation (2.17a),
(D2 + PeD + λ2j )φj = 0, (4.3a)

(Pe + D)φj = 0 at z = 0, 1. (4.3b)
The series expansion for W(z),
M
X
W(z) = aj ψj (z), (4.4)
j=0
is obtained by substituting (4.1) into (2.17b) and solving the resultant inhomogeneous
fourth-order ordinary differential equation analytically with no-slip boundary conditions.
The form of ψj (z) is too lengthy to be reproduced here. Substituting (4.4) in (2.17a)
and rewriting it yields
X M M
X
aj Hj (z) = σ aj φj (z), (4.5)
j=0 j=0
where
Hj (z) = (D2 + PeD − k2 )φj (z) − Peψj (z)Dn0 (z). (4.6)
The differential eigenvalue problem (4.5) is then converted to a generalized linear
algebraic eigenvalue problem by collocating at M discrete points in z ∈ (0, 1),
M
X M
X
Hij aj = σ φij aj , (4.7)
j=0 j=0
where Hij = Hj (zi ) and φij = φj (zi ). We choose equidistant collocation points such that
zi = (1 + i)/(M + 1) with i = 0, 1, 2, . . . , M − 1 and solve the generalized linear
eigenvalue problem (4.7) using the QZ algorithm. The neutral curves (σ = 0) are then
obtained by using the bisection method.
4.1. Validation of the numerical solution

The numerical solution has been validated by comparing it with the growth rates and
neutral curves obtained from the long-wavelength theory and the deep-channel theory.
The comparison presented in figure 5 shows that the spectral solution converges to
analytical solutions after ∼80 collocation points. As anticipated earlier, it is clear
from figure 5(c,d) that the asymptotic solution for Pe 1 is accurate when k = O(1).
We also found that the spectral solution is accurate only when Pe < 80 or so, and
the accuracy could not be improved by increasing the number of collocation points.
Nonetheless, a comparison with the series solution for semi-infinite suspensions shows
that the spectral solution correctly reproduces the scaling σ ∼ Pe2 as seen in figure 6.
In figure 6, when the growth rate and wavenumber are appropriately scaled, all data
points collapse onto a single curve, which is in excellent agreement with the series
solution even for a moderate Pe = 30. Thus the Pe < 80 range is good enough to
explore all the dynamics of the problem.
4.2. Results
https://doi.org/10.1017/jfm.2013.628
We begin by presenting the neutral curves, βc versus k, for various values of Pe in

figure 7. Since the critical wavenumber at which the neutral curve has a minimum is
zero for all cases, the marginal stability limit obtained from the long-wavelength
analysis in Kasyap & Koch (2012) is indeed the true marginal stability limit.
Nonetheless, βc (k) would be the critical wavenumber in a numerical simulation
conducted with periodic boundary conditions in the x direction with a box length
(a) (b)
1.5 103
1.0 102
0.5 101
0 20 40 60 100 101 102

Pe Pe
(c) 250 (d) 3.0
200
2.5
150
100
2.0
50
1.5
0 1 2 3 4 5 0 1 2 3 4 5
k k
F IGURE 5. Validation of the spectral solution. (a) Comparison of growth rates and
(b) neutral curves obtained from the spectral solution with different numbers of grid points
(shown by symbols) and the long-wavelength theory at k = 0.01. (c) Comparison of the
growth rates from the spectral and the asymptotic solutions for deep channels at Pe = 60.
(d) Neutral curves from the spectral and the deep-channel asymptotic solutions at Pe = 60.
Growth rates in (a) and (c) are at β = 20. Numbers of grid points in the spectral solution
corresponding to various symbols are: 4, M = 40; 5, M = 80; , M = 120; , M = 160;
and , M = 200.
of 2π/k. When β > βc , all modes with wavenumber lying between kcrit = 0 and the
cut-off wavenumber kcut will be unstable, with the wavenumber of the most dangerous
mode km lying somewhere between the two limits. Here kcut for a given value of
β > βc and Pe is given by the intersection of the horizontal line corresponding
to the chosen value of β and the neutral curve corresponding to the Pe specified.
Figure 8(a,b) shows the variation of the growth rate with wavenumber for different
values of β and Pe, respectively. It is clear that the suspension is destabilized when
either the bacterial chemotaxis, the suspension depth or the bacterial concentration is
increased. For all cases, a mode of maximum growth rate exists with a wavenumber
km along with a cut-off mode with wavenumber kcut . The variations of km and kcut
https://doi.org/10.1017/jfm.2013.628
with Pe are given in figure 9(a,b), respectively, in which it is seen that both of them
scale with Pe. Thus, unless β is arbitrarily close to βcrit , the wavelength of the most
dangerous mode and the cut-off wavelength are comparable to the channel gap width
for Pe = O(1) or to the boundary layer thickness κ/U0 for Pe 1. We also provide
the profiles of the fluid velocity in the x direction for the most dangerous modes at
Pe = 50 and β = 5 (km = 8.03) and β = 50 (km = 24.37) in figure 10 to show the
0.08
0.06
0.04
0.02
0 0.1 0.2 0.3 0.4 0.5
F IGURE 6. Comparison of the spectral solution at different Pe (symbols) and the series
solution for semi-infinite channels in § 3 (solid line) for different values of Pe: , Pe = 30;
and , Pe = 50; β = 10 for all cases.
100
75
4
50 8
16
25
32
0 5 10 15 20
k
F IGURE 7. Neutral curves βc versus k for different Pe. The critical wavenumber for
instability kcrit at which the neutral curve has a minimum is zero for all cases, implying
that βcrit obtained from the long-wavelength analysis given in (2.21) is the true critical
concentration for instability.
vertical extent of the fluid flow. It is interesting to note that, while the flow covers
the entire channel for the β = 5 case, there is practically no flow after a quarter
of the channel depth in the β = 50 case. The difference in the vertical length scale
emerges from the fact that the most dangerous mode wavelength is of the order of
the channel depth for the first case since km = O(1) and of the order of the boundary
layer thickness κ/U0 for the second case since km = O(Pe). As pointed out in § 3.3,
outside the boundary layer the vertical length scale and horizontal length scales of
https://doi.org/10.1017/jfm.2013.628
the fluid flow are of the same order since they are determined primarily by the
incompressibility constraint.
For all cases discussed so far, the growth rate σ is purely real and the instability
is stationary. However, for larger values of β and Pe, oscillatory solutions do appear,
with the imaginary part of σ being non-zero. Figure 11 depicts such a case for Pe = 15
and β = 300. In figure 11, for sufficiently large values of k, a second mode becomes
(a) 100 (b) 240

75 80
160
30
50
60
80
25 40 20
20 10
0 4 8 12 0 8 16 24
k k
F IGURE 8. Variation of the growth rate σ with wavenumber k for (a) different values of
β at Pe = 8 and (b) different values of Pe at β = 20.
(a) 40 (b) 80
30 60
km 20 kcut 40
10 20
0 20 40 60 0 20 40 60
Pe Pe
F IGURE 9. Variation of the wavelength of (a) the most dangerous mode km and (b) the
cut-off wavelength kcut with Pe for β = 20 (solid line), 40 (dotted line) and 60 (dashed
line).
(a) 1.0 (b) 1.0
z 0.5 0.5
0 0
– 0.5 0 0.5 1.0 – 0.5 0 0.5 1.0
https://doi.org/10.1017/jfm.2013.628
F IGURE 10. Eigenfunctions corresponding to the most dangerous mode for Pe = 50 and
(a) β = 5 with km = 8.03 and (b) β = 50 for km = 24.37. Solid lines show the profiles
of the number density disturbance N(z) and dashed lines show the profiles of the x fluid
velocity disturbance U(z). In order to be represented in the same plot as number density,
velocities are rescaled by their maximum magnitudes, which are 0.007 for the first case
and 0.068 for the second case.
1200
800
400
0 5 10 15 20
k
F IGURE 11. Growth rate Re(σ ) versus wavenumber k for Pe = 15 and β = 300. The
solid lines indicate stationary solutions with Im(σ ) = 0 and the broken line indicates the
complex conjugate oscillatory solution.
unstable, and both the first and second modes are stationary until k reaches some k∗ ,
after which both modes share the same Re(σ ) and have equal and opposite Im(σ ).
These complex conjugate growth rates signify travelling waves propagating along both
x and −x directions. For k > kcut the waves are damped. Figure 12 shows the number
density disturbances and streamline patterns for the stationary and oscillatory branches
of the solution, respectively. The mode corresponding to the upper stationary branch
of figure 11 shown in figure 12(a) contains only two cells per wavelength, while
the mode corresponding to the lower stationary branch of figure 11 contains four
cells per wavelength as seen in figure 12(b). The additional cells that appear near the
bottom wall in the latter case convect bacteria back from regions with excess number
density to regions of lesser number density, offsetting part of the destabilizing flux
and reducing the growth rate. Thus the fastest growing mode in the stationary regime
will always be the two-cell mode shown in figure 12(a). The oscillatory modes shown
in figure 12(c,d) have mirror symmetry as expected and are remarkably different from
the stationary modes in the sense that the cells bend left or right at the bottom wall
depending upon their direction of propagation.
From an experimental point of view, it is important to anticipate the mode of
instability that is going to be realized. To aid this, we provide the boundary separating
the stationary and oscillatory instabilities in β–Pe space in figure 13(a) and the
transition wavenumber k∗ above which the modes become oscillatory versus Pe
in figure 13(b). In figure 13(a) the dashed line shows the critical β required for
the instability calculated using (2.21) and the solid line shows those values of β
above which oscillatory solutions exist. The latter shows that oscillatory solutions
exist only if β > 247, and to see whether this regime is accessible we plot the
https://doi.org/10.1017/jfm.2013.628
accessible value of β as a function of Pe as a dash-dotted line in figure 13(a).

To calculate the latter, we make use of (2.20) with a bacterial concentration of
the order of the maximum bacterial concentration reached in experiments (see
Sokolov et al. 2009), hn0 i = 1010 cm−3 , along with typical values for bacterial
parameters C = 0.57, Us = 20 µm s−1 , L = 10 µm, τ0−1 = 1 s−1 and U0 /Us = 0.1.
It is now clear from figure 13(a) that the oscillatory modes are accessible when
(a) (b)
(c) (d)
F IGURE 12. Number density disturbances (shown by the grey scale (colour online)
shading) and streamline patterns for stationary (at k = 10) and oscillatory (at k =
15) instabilities at Pe = 15 and β = 300. Brighter and darker regions indicate
a higher and lower bacterial number density than the base state. (a) Leading
stationary mode; (b) lagging stationary mode; (c) left-propagating oscillatory mode; and
(d) right-propagating oscillatory mode.
Pe > 18, which corresponds to a channel depth greater than 1 mm for the values of
bacterial parameters given above. Furthermore, the likelihood for oscillatory modes
to be realized in an experiment depends upon whether the most dangerous mode
is oscillatory or not. Since km < k∗ always in figure 13(b), we conclude that the
most dangerous mode is always stationary, at least in the parameter space that we
tested here. Nevertheless, one could isolate the oscillatory modes in a simulation
by performing the simulations with periodic boundary conditions in the x direction
that admit only the wavenumbers corresponding to the oscillatory modes. Thus in
https://doi.org/10.1017/jfm.2013.628
experiments one may expect oscillatory modes to be realized only if the channel is
longitudinally confined.
5. Effect of shear rotation

In general, the orientations of bacteria are influenced by both the chemical attractant
gradient, which alters the tumbling frequency of the bacteria, and the fluid velocity
(a) 400 (b) 60

300
40
200
20
100
0
0 20 40 60 10 20 30 40 50 60
Pe Pe
F IGURE 13. (a) Boundary separating the stationary and oscillatory solutions in β–Pe
space. The dash-dotted line shows the maximum β attainable for a given Pe for typical
bacteria parameters and assuming the maximum attainable bacterial concentration to be
1010 ml−1 . For comparison, the marginal stability curve is shown by the dashed line.
(b) Variation of the wavenumber of the most dangerous mode km (open symbols) and the
critical wavenumber k∗ above which oscillatory modes exist (filled symbols) with Pe for
β = 260 (circles) and 500 (squares).
gradient, which rotates the cells. Previous studies (Simha & Ramaswamy 2002;
Saintillan & Shelley 2008a,b; Subramanian & Koch 2009; Subramanian et al. 2011)
have focused on instabilities induced by rotation due to the fluid shear. In the
preceding sections of the present paper we have discussed an instability resulting
from coupling of the bacterial stress due to varying bacterial number density and the
convective transport of the bacteria. For simplicity, we neglected the rotation of the
cells by the fluid velocity gradient and assumed that their orientation distribution was
governed solely by their biased tumbling in the chemical gradient field. In the present
section we will first derive the governing equations for the large-length-scale behaviour
of a bacterial suspension whose orientation distribution responds to both chemical
and fluid velocity gradients and then perform a stability analysis of these equations
to determine the effect of shear-induced rotation on the number density instability.
5.1. Derivation of conservation equations of a chemotactic bacterial suspension

Previous investigations of shear-rotation-induced instabilities of micro-swimmer
suspensions (Simha & Ramaswamy 2002; Saintillan & Shelley 2008a,b; Subramanian
& Koch 2009; Subramanian et al. 2011) have performed stability analyses directly
on the kinetic equation, the equation for the one-swimmer orientation and spatial
probability Ω(p, x, t). However, over large length and time scales, we may expect
the suspension behaviour to be described by continuum conservation equations for
the bacterial number density n(x, t) and the fluid velocity u(x, t). Our analysis
of the chemotaxis-driven active-stress instability in the preceding sections, like the
previous analyses of gravity-driven bioconvection (Hill et al. 1989), was based
on such continuum equations. In order to perform a continuum analysis of a
https://doi.org/10.1017/jfm.2013.628
suspension including both fluid shear-induced rotation and chemo-gradient-induced

biased tumbling, we require a systematic derivation of the continuum equations
coarse-grained over length scales much larger than the persistence length of bacteria,
Us τ0 , with a weak imposed chemical gradient. The first, partial effort in this
direction was made by Bearon & Pedley (2000) and Bearon (2003), who derived
macro-transport equations for suspensions of chemotactic bacteria in a specified fluid
flow. We will point out the differences between their continuum equations and the
equations derived here later in this section.
We begin the formulation by writing down the non-dimensional form of the kinetic
equation (2.1) as
1
Z
−1 ∂Ω Ω Ω
H
b + ∇ · (pΩ) + ∇ · (uΩ) + ∇ p · (ṗΩ) + ∗ − dp = 0, (5.1)
∂t τ 4π τ∗
where
τ ∗ −1 = 1 − ζ p · gH(p · g) (5.2)
is the tumbling frequency non-dimensionalized by τ0−1
(see (2.4)) with weak
chemotaxis and H = H/(Us τ0 ) is the ratio of the channel thickness to the bacterial
b
persistence length. The symbol H in (5.2) stands for the Heaviside step function.
We assume H b −1 1 for a coarse-grained description of the suspension, and we also
assume that bacteria are slender so that their rotation rate due to fluid shear is given
by ṗ = ω · p + (I − pp) · (e · p) with e = 12 (∇u + ∇u† ) and ω = 12 (∇u† − ∇u). To derive
(5.1) we have used the suspension depth H, bacteria swimming velocity Us and the
ratio H/Us as the characteristic scales for length, velocity and time, respectively. Thus
when H b −1 1, the fluid shear acting on bacteria is O(Us /H), which is much weaker
than the tumbling frequency τ0−1 .
The corresponding evolution equation for the bacterial number density obtained after
integrating (5.1) in the orientation space is
∂n
+ ∇ · (q + un) = 0, (5.3)
∂t
in which the swimming flux q = Ωp dp is given by the first moment of the
R
probability density. As mentioned in § 2, the fluid velocity field u(x, t) in (5.3) is
governed by the Stokes equations (2.8), which when written in non-dimensional form
are
∇ · u = 0, (5.4a)
8β
− ∇p + ∇ 2 u + ∇ · σ B = 0. (5.4b)
3
Here, the pressure has been scaled by the viscous stress µUs /H and the definition
of β remains the same as in (2.19). The non-dimensional bacterial stress given by
theR second moment of the probability density in the orientation space is ff B (x, p, t) =
− Ω(pp − I /3) dp.
Equations (5.3) and (5.4) govern the continuum behaviour of the bacterial
suspension, and our goal now is to obtain continuum constitutive equations for the flux
q and the bacteria stress ff B . (this needed) In the following, we derive coarse-grained
constitutive equations for the aforementioned quantities valid for length scales much
larger than the bacterial persistence length Us τ0 , which corresponds to the limit of
b −1 1. Since a non-zero swimming flux and a non-zero bacterial stress can arise
H
only from the deviation of the bacterial orientation field from isotropy, we have
https://doi.org/10.1017/jfm.2013.628
Z
q = Ω 0 p dp, (5.5a)
I
Z
B 0
ff = − Ω pp − dp, (5.5b)
3
where Ω 0 = Ω − n/(4π) is the anisotropy R 0 in the bacterial orientation field, which0
satisfies the normalization condition Ω dp = 0. The governing equation for Ω
obtained by subtracting H b −1 /(4π) times the number density equation (5.3) from (5.1)
and by using (5.2) is

0
1
Z
−1 ∂Ω 0 0 0 0
H
b +∇· Ω p− Ω p dp + ∇ · (uΩ ) + ∇ p · (ṗΩ )
∂t 4π
Ω0 1 Ω0 n 1
Z
+ ∗ − dp = ζ p · gH(p · g) −
τ 4π τ∗ 4π 4
h n n i
−1
−H b ∇p · ṗ + p · ∇ . (5.6)
4π 4π
The terms on the right-hand side of (5.6) are the sources of orientational anisotropy
through the biased tumbling of bacteria, rotation of bacteria by fluid shear, and the
b −1
spatial variations in the number density, respectively. In the limit of ζ 1 and H
1, the deviation from isotropy is weak and hence (5.6) admits a perturbative solution
of the form
n 1 b −1 3n ∇u : pp − H b −1 p · ∇ n

Ω0 = ζ p · gH(p · g) − +H
4π 4 4π 4π
+ higher-order terms, (5.7)
which satisfies the required normalization condition. The second term on the right-
hand side of the above expression makes use of the relationship −∇ p · (nṗ) = 3n∇u :
pp. Using (5.7) in (5.5) gives the following constitutive relations for the bacterial
swimming flux and the stress:
ζ b −1
H
q = gn − ∇n + higher-order terms, (5.8a)
6 3
ζ

I
b −1
2H
σB = − gg − n− ne + higher-order terms. (5.8b)
16 3 5
The first term in (5.7) is the same as the product of the number density and the O(ζ )
anisotropic orientation field in (2.5) due to chemotaxis, and as a result the order ζ
flux and stress terms in (5.8) when written in the dimensional form are identical to
the chemotaxis flux and stress terms, the U0 n term and the ff B term, respectively,
in (2.7) and (2.10). The second term in (5.7) is the deviation from the isotropic
state arising from the fluid shear acting on the isotropic orientation field, and since
the latter is even in p, no net swimming flux arises from this term in (5.8a). This
is a consequence of the fact that only the straining component of the fluid shear
alters the orientation field, since the vorticity component merely rotates an isotropic
orientation field, leaving it unaltered. Owing to the symmetrically located extensional
and compressional regions of straining motion, no mean orientation results from its
https://doi.org/10.1017/jfm.2013.628
action on the isotropic orientation field. Instead, one gets an enhanced probability
of bacterial orientations near the extensional axis of the strain (Subramanian &
Koch 2009). The associated bacterial stress obtained from (5.8b), −(5H) b −1 n(2e ), is
Newtonian since it is linear in the strain-rate tensor but with a negative viscosity, since
the stress constituted by it acts in the opposite direction to the usual viscous stress.
This is due to the higher fraction of bacteria being oriented along the extensional
axis of the fluid flow reinforcing the flow with their force dipoles. Finally, the third
term in (5.7) is the result of the spatial variations in bacterial number density and
it gives rise to the diffusive flux (Hb −1 /3)∇n in (5.8a). It is clear from (5.7) that
the deviation from isotropy caused by a non-uniform bacterial concentration field

peaks along the opposite direction to the number density gradient. Since the O(1)
bacterial orientation field is isotropic, bacteria are equally likely to swim in any
direction, and consequently, if there exists a spatial variation in the bacterial number
density, then there will be a net bacterial swimming from high to low bacterial
concentration regions. This flux is diffusive, as it is proportional to the gradient
of bacterial concentration. In dimensional form, this flux is driven by the scalar
diffusivity κ = Us2 τ0 /3 (Bearon & Pedley 2000; Bearon 2003; Chen et al. 2003).
Since the orientational anisotropy arising from bacterial number density gradients is
odd in p, no bacteria stress arises from this term. Thus the leading-order bacteria
stress in (5.8b) is constituted by chemotaxis and shear rotation of bacteria, while the
leading-order bacterial swimming flux is constituted by chemotaxis and diffusion of
bacteria. The latter makes the bacteria conservation equation (5.3) identical to (2.7)
when written in dimensional form.
The higher-order flux and stress terms in (5.8), which would involve coupling
between chemotaxis, shear rotation and non-uniformity of the suspension, are ignored
in this paper, and we instead focus on the first effects of shear rotation of bacteria
on the chemotaxis-driven instability. The analysis presented above has elucidated
that the primary role of shear rotation of bacteria is to reduce the effective viscosity
of the suspension due to the increased probability of bacterial orientations near
the extensional axis of the fluid flow. As mentioned in § 1, the orientation shear
instability of bacterial suspensions arises from the aforementioned mechanism and
in fact Subramanian & Koch (2009) consider the long-wavelength orientation shear
instability of homogeneous suspensions without any chemical gradient as a result
of the effective viscosity being negative. It is clear from (5.4b) and (5.8b) that the
non-dimensional effective suspension viscosity has the form
8β
b eff (x, t) = 1 −
µ n(x, t). (5.9)
15Hb
The critical concentration for the orientation shear instability is then given by the
condition of zero effective viscosity, and for an unbounded, homogeneous suspension,
the latter condition corresponds to βcrit = 15/8H b (Subramanian & Koch 2009). Using
the definitions of β and H, one finds that the critical condition is given by the non-
b
dimensional concentration γ = (C/5)hn0 iL2 Us τ0 = 1 (Subramanian & Koch 2009).
Finally, we wish to point out how the present derivation differs from the works
of Bearon & Pedley (2000) and Bearon (2003) in which the coupling between shear
rotation and chemotaxis has been analysed. The primary objective of Bearon & Pedley
(2000) and Bearon (2003) is to obtain a macroscopic transport equation for bacteria
in the presence of an imposed chemical gradient and a simple shear flow. As a result,
https://doi.org/10.1017/jfm.2013.628
Bearon & Pedley (2000) and Bearon (2003) assume only one-way coupling between
bacterial and fluid motion, which involves advection and the rotation of bacteria by
the fluid velocity and its gradients, respectively, but without the bacterial force dipoles
affecting the fluid motion. The latter has been incorporated in our derivation, and that
leads to coupled equations for bacterial number density and fluid motion. Equation
(5.4b) shows that bacterial force dipoles will have a strong influence particularly if
the suspension is dense enough to have β/H b = O(1). For typical bacteria, Uτ0 /L =
3
O(1), so that β/H = O(hn0 iL Us τ0 /L) will be of order unity even for moderately dense
b
suspensions with nL3 = O(1).
5.2. Stability analysis with shear rotation

The analysis in the previous section shows that the primary effect of reorientation of
bacteria under weak shear is to reduce the effective viscosity of the suspension.
We now examine how this reduced viscosity of the suspension influences the
critical concentration for the chemotaxis-driven instability analysed in this paper
by performing a stability analysis of (5.3) and (5.4) together with (5.8). By using the
fact that Pe = Hζ
b /2, we rewrite the governing equations and boundary conditions as
∂n
3H
+ ∇ · (Peng + 3Hun
b b − ∇n) = 0, (5.10a)
∂t
∇·u = 0, (5.10b)
8β I

βζ
− ∇p + ∇ · 1 − n (∇u + ∇u† ) − gg − · ∇n = 0, (5.10c)
15Hb 6 3
Peng − ∇n = u = 0 at z = 0, 1,(5.10d)
with g = −ez . Since the dimensional form of (5.10a) is the same as the number density
equation without any shear rotation (2.7) used in previous sections, the base-state
number density field n0 (z) = Pe exp(−Pez)/[1 − exp(−Pe)] remains the same as before.
Linearizing (5.10) for small perturbations of the form in (2.16) about the base state
yields the amplitude equations
[D2 + PeD − (σ + k2 )]N = 3HWDn

b 0, (5.11a)
ikU + DW = 0, (5.11b)
8β 8β

βζ
D 1− n0 (DU + ikW) = 2k2 1 − n0 U + ik P − N ,
15Hb 15Hb 18
(5.11c)
8β 8β

2D 1 − n DW = −ik 1 − n (DU + ikW)
15Hb 0 15Hb 0

βζ
+D P+ N , (5.11d)
9
with the boundary conditions (Pe + D)N = U = W = 0 at z = 0 and z = 1. When
deriving (5.11a) from (5.10a), the factor of 3H
b has been absorbed into σ so that the
growth rate σ is non-dimensionalized by H 2 /κ as before. The factor of 3H b in the
unsteady term in (5.10a) arose because time had been non-dimensionalized by H/Us
in this section rather than by the diffusive time scale H 2 /κ as in § 2. In addition,
the fluid convective terms in (2.17a) and (5.11a) differ by a factor of ζ /6 since the
https://doi.org/10.1017/jfm.2013.628
fluid velocity has been scaled by the swimming velocity in the former and by the
chemotactic velocity in the latter.
We now solve (5.11) in the long-wavelength limit, k 1, by expanding the
dependent variables and the growth rate as N(z) = N0 (z) + O(k), P(z) = P0 (z) + O(k),
U(z) = kU1 (z) + O(k2 ), W(z) = k2 W2 (z) + O(k3 ) and σ = k2 σ2 + O(k3 ), substituting
these in (5.11), and solving the resulting ordinary differential equations to yield the
leading-order solutions for the number density and the pressure:
Pe exp(−Pez)
N0 (z) = , (5.12a)
1 − exp(−Pe)
3

βζ
P0 (z) = −N0 + C2 , (5.12b)
9 2
in which C2 is the constant of integration. The physical balances that lead to the
solutions given in (5.12) are identical to those in the case without shear rotation
(Kasyap & Koch 2012). The exponential number density field in (5.12) is the result
of the balance between chemotaxis and diffusion, and the pressure field results from
the balance of the divergence of the bacteria stress and the pressure gradient in the
z momentum equation. The leading-order x momentum equation

βζ
D{[1 − β exp(−Pez)]DU1 } = i P0 −
b N0 , (5.13a)
18
U1 (0) = U1 (1) = hU1 i = 0, (5.13b)
where
8βPe γ Pe
βb = = , (5.14)
15H[1
b − exp(−Pe)] 1 − exp(−Pe)
is just a balance between the viscous shear stress with a vertically stratified effective
viscosity 1 − βb exp(−Pez), pressure and the bacterial stress. The condition of no net
fluid flux in the x direction, hU1 i = 0, in (5.13) is a result of the incompressibility
constraint and the no-slip boundary condition, W(0) = W(1) = 0. The stratification of
the effective viscosity is due to the non-uniform base-state bacterial concentration field.
The solution of (5.13) is
Z z
dU1 0 iβζ I1 (z)

U1 (z) = dz = + C2 I2 (z) + C3 I3 (z) , (5.15)
0 dz 6 1 − exp(−Pe)
0
where C3 is the second integration constant, and the forms of the integrals I1 (z), I2 (z)
and I3 (z) are given in appendix B. The constants C2 and C3 are then determined from
the boundary condition on the top wall, U1 (1) = 0, and the no net fluid flux constraint,
hU1 i = 0.
Now averaging the number density equation (5.11a) in the z direction yields the
O(k2 ) growth rate as
σ2 = −(1 + 3iHhU
b 1 n0 i). (5.16)
The marginal stability curve given by the condition σ2 = 0 is no longer simply
described by β and Pe as in figure 1, owing to the presence of the non-dimensional
number βb in the expression for the velocity in (5.15). The definition of βb in (5.14),
i.e. βb = γ Pe/[1 − exp(−Pe)], shows that, for a given Pe, the latter is fixed by a
https://doi.org/10.1017/jfm.2013.628
scaled bacterial concentration γ (recall that γ = (C/5)hn0 iL2 Us τ0 ) that does not
involve the suspension depth. This is unlike the case without shear rotation described
in previous sections in which the bacterial concentration influences the problem only
through the depth-dependent scaled concentration β. Since both concentrations β and
βb influence the velocity field in (5.15), it is useful to separate the effects of bacterial
concentration from those of the suspension depth. Moreover, the effect of chemotaxis
can be described by ζ , and thus all three effects can be studied independently of each
other if we describe the problem in terms of ζ , H b (defined as H
b = H/(Us τ0 )) and γ .
Physically, this is a consequence of the fact that the effective viscosity is dependent
only on the local bacterial concentration. A direct comparison with the case without
shear rotation shown in figure 1 is still possible by making use of the relationship
15 b
β= 8
γ H. (5.17)
since β has been defined as β = 83 Chn0 iL2 H.

Since our primary objective here is to understand the effect of the orientation shear
mechanism on the critical concentration for the chemotaxis-driven instability, we
restrict the present analysis to the case where the effective viscosity 1 − βb exp(−Pez)
is positive everywhere in the domain z ∈ [0, 1]. This, along with (5.14), leads to an
upper bound for γ of
1 − exp(−Pe)
γmax = . (5.18)
Pe
Thus, our analysis yields a prediction of a chemotaxis-driven instability only when
there exists a critical concentration γ < γmax for which the growth rate is zero.
If such a critical concentration does not exist, then the suspension is destabilized
by the orientation shear instability mechanism when γ = γmax . Otherwise, the
chemotaxis-driven instability mechanism helps to reduce the critical concentration
below that for the pure orientation shear instability. The advantage of using γ instead
of β to describe the critical condition is apparent now, since, in the absence of
any chemotaxis (ζ = 0), we recover the critical condition for the orientation shear
instability γcrit = γmax = 1 obtained previously by Subramanian & Koch (2009).
The results of our calculations depicted in figure 14 show that a critical bacterial
concentration γcrit that is smaller than the value for the shear rotation instability, γmax ,
is indeed predicted. In figure 14(a) we present the variation of γcrit obtained from the
treatments with and without shear rotation with the non-dimensional suspension depth
H
b at ζ = 0.1. For small H, b the critical concentration is determined by the orientation
shear mechanism, as indicated by the merging of the γcrit and γmax curves. When H b is
above the value H b min shown by the circle in figure 14(a), the number density instability
mechanism significantly reduces the critical concentration from the orientation shear
value γmax . We choose to put the circle where γcrit is smaller than γmax by around 10 %.
The influence of the number density instability mechanism continues to increase with
increasing H.b The transition between the shear-driven instability and the chemotaxis
instability is more or less complete at a value H b =H b max . Beyond this value, shear
rotation plays little role in the instability, as noted by the fact that, at H b max , γcrit is
90 % of the value of γcrit without shear rotation. For a given bacterial concentration
or γ , the curves in figure 14(a) represent the critical suspension depth required for
the instability, and it is clear from the figure that, for concentrations at which the
https://doi.org/10.1017/jfm.2013.628
shear rotation and the chemotaxis instability mechanisms are comparable in strength,
the critical depth is only modestly affected by the shear rotation mechanism.
The effect of chemotaxis on the critical concentration is shown in figure 14(b). In
the absence of any chemotaxis, for which ζ = 0, the suspension is destabilized purely
by the orientation shear mechanism if γ > γcrit = 1 (Subramanian & Koch 2009). For
a given H,b increasing the strength of chemotaxis by increasing ζ decreases not just
(a) 1.2 (b) 1.2

0.8 0.8
0.4 0.4
0.01
0.05
0 50 100 150 0 200 400 600
(c) 500 (d ) 10
250 5
0 0.05 0.10 0 0.04 0.08 0.12
F IGURE 14. (a) Variation of the critical concentration γcrit (solid line) with suspension
depth Hb at ζ = 0.1. For comparison, variation of γcrit for the case without shear rotation
obtained from (2.21) and (5.17) (dashed line) and the critical concentration for the purely
orientational instability γmax (dotted line) are also shown. The circle indicates the point
at which γcrit is 90 % of γmax , and the square when γcrit is 90 % of γcrit obtained for the
case without shear rotation. We denote the values of H b at these points as H b min and H b max ,
respectively. (b) Plot of γcrit versus H
b for various values of ζ . (c) Variation of H
b min (circles)
and Hb max (squares) with ζ . (d) Normalized growth rate σ /k2 versus concentration γ for
the case with shear rotation (solid line) and without shear rotation (dashed line) at ζ = 0.1
and Hb = 150.
γcrit but also the range of H

b in which shear rotation has an influence. The latter point
is seen in figure 14(c).
Although shear rotation does not play a significant role in determining the critical
https://doi.org/10.1017/jfm.2013.628
concentration for deep channels, as seen in figure 14(a), it has a strong destabilizing
effect when γ is sufficiently larger than γcrit . To elucidate this, we study the variation
of the growth rate with the concentration in figure 14(d) at a suspension depth
H
b = 150 and ζ = 0.1, for which the effect of shear rotation on the critical condition
is negligible, as seen in figure 14(a). In figure 14(d), when γ is close to γcrit (the
horizontal intercept), the growth rates predicted by theories with and without shear
rotation do not differ appreciably from each other. However, the presence of shear
rotation causes a sharper increase in the growth rate thereafter. The reduction in
the effective viscosity due to shear rotation enables the chemical gradient-induced
active stresses to drive a larger fluid velocity disturbance, leading to a faster growth
of the number density variation. The destabilizing role of shear rotation is further
evidenced by the fact that the value of γcrit obtained from the calculation that includes
shear rotation is always smaller than the γcrit predicted by the theory without shear
rotation. However, as seen in figure 14(a), this difference is only infinitesimal for
large enough suspension depth, since the reduction in viscosity happens only within
a thin boundary layer near z = 0.
6. Concluding discussion
In the preceding sections we have presented a detailed analysis of the chemotaxis-
driven instability mechanism presented in Kasyap & Koch (2012). The balance of
chemotaxis and diffusion leads to a non-uniform base-state bacterial concentration
field and active stress field. The coupling between fluid flow and bacterial concentration
field through the swimming-induced active stress and convection of bacteria leads to
an instability of the aforementioned base state when the scaled bacterial concentration
β exceeds a critical value dictated by the Péclet number Pe measuring the strength
of chemotaxis relative to diffusion. The critical concentration for the instability,
βcrit , is set by the modes with the largest wavelength and βcrit ∼ Pe−3 for Pe 1,
while βcrit ∼ 2 for Pe 1 in the case with both boundaries being no-slip walls. If
the boundary closer to the region of larger concentration of the chemical attractant
is a shear-stress-free interface, the large-Pe asymptote changes to βcrit ∼ 4/Pe. An
analysis of semi-infinite suspensions has revealed the existence of a most dangerous
wavenumber of the instability for which the growth rate is maximum and a cut-off
wavenumber above which the disturbances cease to grow. From the full numerical
solution of suspensions of finite depth, these wavenumbers have been found to scale
with the Péclet number, indicating the fact that the most dangerous wavenumber
and cut-off wavelength are of the order of the boundary layer thickness κ/U0 when
Pe is large. While the instability is stationary in the limit of small wavenumbers,
the numerical solution shows the presence of oscillatory solutions for wavenumbers
of O(Pe) and for β > 247. Further analysis, however, has shown that the most
dangerous mode is always stationary. A long-wavelength analysis that includes the
rotation of bacteria by fluid velocity gradients shows that the critical concentration
of shallow suspensions is set by the orientation shear instability mechanism and the
latter has only negligible influence in determining the critical concentration of deeper
suspensions. Nevertheless, the growth rate of the instability in the super-critical
regime is strongly enhanced by shear rotation.
While bioconvection and the present chemotaxis instability involve the coupling of
fluid flow and bacterial number density fluctuations, which destabilize a quiescent,
inhomogeneous base state, the fluid velocity in the present context arises from the
swimming-induced force dipoles of the cells, whereas bioconvection results from their
gravitational forces. As a result of this distinction, bioconvection is manifested in
https://doi.org/10.1017/jfm.2013.628
thicker suspensions where a critical Rayleigh number is exceeded, while the present
mechanism can occur on lengths of 100 µm to 1 mm. In addition, the present
mechanism arises only for pushers, while bioconvection can occur for pushers or
pullers. The active stress in a suspension can be modulated by bacterial concentration
and/or orientation. Previously investigated instabilities due to active stresses have
involved the coupling of cell orientation with the disturbance shear flow and have
100
75
50
25
Hcrit
0 150 300 450 600
F IGURE 15. Variation of the value of β obtained in Sokolov et al.’s (2009) experiment,
βexp (solid line), and the βcrit (dashed line) required for instability with the thickness of
the film. The instability persists when H > Hcrit ≈ 310 µm.
exhibited no bacterial concentration fluctuations in the large-wavelength limit, whereas

the present mechanism is critically dependent on concentration fluctuations that persist
at large wavelengths. Unlike the shear rotation instability in an unbounded suspension,
the present instability mechanism exhibits a most dangerous wavelength. The critical
concentration for the chemotaxis-driven instability can be much smaller than that for
non-chemotactic collective motion when the suspension thickness is sufficiently large
for significant inhomogeneous base-state cell concentration to develop.
While controlled experiments for validating our predictions are yet to be done,
we believe that the present mechanism of instability operated in the experiments of
Sokolov et al. (2009). The observation of cross-film convective patterns in liquid
films containing oxytactic B. subtilis cells when the film thickness is raised above
200 µm suggests this. We note that the fundamental requirement for the present
active stress instability, namely the inhomogeneous bacterial concentration field, is
fulfilled in Sokolov et al.’s (2009) experiments at sufficiently large film thicknesses
owing to the chemotactic accumulation of bacteria near the interfaces. The spatially
homogeneous, in-plane collective motion that the authors observe for thinner films is
presumably due to the orientation shear instability. When the film thickness becomes
sufficiently large, the spatially inhomogeneous active stress instability takes over and
one starts to see cross-film convective patterns. Furthermore, our estimate of the
critical film thickness for the onset of the present instability compares well with the
experimental observation of Sokolov et al. (2009). To show this, we plot βexp , the
value of β reached in their experiment, and the theoretical βcrit required for instability
(2.21) versus the film thickness in figure 15. The mean bacterial concentration hn0 i
used for calculating βexp taken from Sokolov et al. (2009) is 2 × 1010 ml−1 . The
swimming velocity and the mean tumbling frequency used in the calculations are the
https://doi.org/10.1017/jfm.2013.628
typical values 20 µm s−1 and 1 s−1 , respectively, along with the typical ratio 0.1 of
the chemotactic velocity to the swimming speed. Here we use the film half-thickness
H/2 for all calculations, since the gradient in Sokolov et al.’s (2009) experiment is
from the mid-plane of the film to the interfaces. The chemotaxis-driven instability
described here persists when βexp > βcrit , which happens when H > Hcrit ≈ 310 µm.
The determination of a critical film thickness in an experiment is not entirely
straightforward. Sokolov et al. (2009) reported the size of the largest region in
which they observed number density inhomogeneities as a function of the film
thickness and reported a critical thickness of 200 µm as the minimum film thickness
at which any inhomogeneity could be observed. However, if we take the critical

film thickness by extrapolating the plot of inhomogeneity size versus film thickness
to zero inhomogeneity size, assuming a simple linear or quadratic dependence, we
obtain a critical film thickness of ∼290 µm. In either case, the observed critical
thickness is near the theoretical prediction, but the latter method, which requires a
significant inhomogeneity to constitute an unstable film, yields a critical thickness that
is quite close to the theoretical prediction of 310 µm in figure 15. This agreement is
excellent considering the fact that the actual chemotactic velocity in the experiment
is not known. Moreover, the Pe based on the film half-thickness for those film
thicknesses above which the cross-film convective motions persist in Sokolov et al.’s
(2009) experiment is of O(1), which means that the most dangerous mode scales
with the film thickness, as seen in § 4. This seems to agree with the map of bacterial
number density given in figure 4(a) of Sokolov et al. (2009). These observations
strongly suggest the existence of the new instability mechanism discussed here in
their experiments.
The existence of the chemotaxis-induced instability presented in this paper and
in Kasyap & Koch (2012) has been confirmed by Ezhilan, Pahlavan & Saintillan
(2012) through full nonlinear simulations of the kinetic equation for Ω(x, p, t). The
primary objective of this paper is the investigation of the dynamics of suspensions of
aerotactic bacteria under the experimental setting of Sokolov et al. (2009). Thus the
suspension is confined in a cubical domain with periodic boundary conditions in two
directions and zero-shear-stress boundary condition in the third direction, and the latter
boundaries are kept at constant oxygen concentration. The oxygen gradient is formed
due to the consumption by bacteria, and, in addition to the bacterial consumption, the
oxygen concentration evolves in time due to fluid convection and diffusive transport.
Ezhilan et al. (2012) varied the thickness of the film in the third direction to mimic
the experiment of Sokolov et al. (2009) and observed three principal regimes of
suspension behaviour. For film thicknesses less than 200 µm (classified as ‘regime I’
in Ezhilan et al. (2012)), the bacterial and the oxygen concentration profiles reached
a steady state. Gross fluctuations in the bacterial concentration field and the fluid
velocity field started to appear in ‘regime II’, which spans the film thicknesses
between 200 and 400 µm. Further increase in the film thickness beyond 400 µm
(region III) resulted in an oxygen depletion layer along with the intensification in the
magnitude of fluctuations closer to the film edges and attenuation of the magnitude
near the centre.
Ezhilan et al. (2012) point out that, in regime I, in which no unsteady motion
was observed, the oxygen concentration was uniform across the film, so that the
chemotaxis of bacteria was too weak for any fluctuations to grow. The latter started
growing and eventually reached a level of saturation in regime II, which had sufficient
thickness to create a strong oxygen gradient. A typical observation in this regime
was strong chemotactic concentration of bacteria near the boundaries and plumes
https://doi.org/10.1017/jfm.2013.628
dense in bacteria originating from those boundaries to the bulk. This is consistent
with the pattern of motion predicted by our theory, and, in addition, Ezhilan et al.
(2012) report a general agreement on the critical conditions for the onset of the
unsteady motion with our predictions in Kasyap & Koch (2012). An important
insight based on this agreement is that the two approaches for analysing the stability
of bacterial suspensions (viz. solving an equation for the number density, as we do,
and solving the full kinetic equation for the position–orientation probability density
in Ezhilan et al. (2012)) are equivalent. However, the former will be considerably
simpler especially for nonlinear simulations, since one does not have to solve for the
orientational degrees of freedom, and in this regard the continuum equations derived
in § 5.1 are valuable. The applicability of continuum equations is, however, restricted
to the case where the effect of chemotaxis, fluid shear and the non-continuum nature
of the suspension are weak.
Acknowledgement
This work was supported by NSF grant CBET-1066193.
Appendix A. Justification for neglecting the attractant transport by fluid flow

In § 2 we neglected the attractant transport by the fluid flow generated by instability
since attractant diffusivity is typically much larger than bacterial diffusivity. We now
provide a justification for this assumption by first writing down the non-dimensional
attractant transport equation,
∂Ca κa
+ Pe∇ · (uCa ) − ∇ 2 Ca = 0, (A 1)
∂t κ
in which length, time and velocity have been scaled by the channel depth H, the
bacterial diffusion time H 2 /κ and the chemotactic velocity U0 , respectively, as
followed for deriving the non-dimensional bacterial transport equation (2.17a). For
typical attractants like Me-Asp or oxygen, the diffusivity κa is of the order of
103 µm2 s−1 (see Ferrell & Himmelblau 1967; Sartori & Tu 2011), while a typical
bacterium swimming at a speed of 20 µm s−1 and tumbling at a rate of 1 s−1
has κ ≈ 133 µm2 s−1 , which is an order of magnitude smaller than the attractant
diffusivity, so that κa /κ 1. Thus when Pe = O(1), the transport of the attractant by
fluid flow along with the unsteady term in (A 1) can be neglected. Even when Pe 1,
the diffusion term dominates, since the fluid velocity is only O(Pe−1 ) at most (see
§ 3.3). A comparison of oxygen concentration profiles from nonlinear simulations of
Ezhilan et al. (2012) also indicates that the instability-generated fluid flow causes
little perturbation to the oxygen concentration field.
Appendix B. Integrals in (5.15)

The full form of the integrals appearing in (5.15) are
Z z
exp(−Pez0 ) dz0
I1 (z) =
0 1−β b exp(−Pez0 )
" #
1 1 − βb exp(−Pez)
= ln , (B 1a)
Peβb 1 − βb
Z z
z0 dz0
I2 (z) =
https://doi.org/10.1017/jfm.2013.628
( " #
1 Pe2 z2 1 − b exp(−Pez)
β
= + ln(β)
b ln
Pe2 2 1 − βb
)
+ dilog[βb exp(−Pez)] − dilog(β) b , (B 1b)
Z z 0
dz
I3 (z) =
" #
1 1 − βb exp(−Pez)
= z+ ln , (B 1c)
Pe 1 − βb
Rx
where the dilogarithmic function is defined as dilog(x) = − 1 ((ln t)/(t − 1)) dt
(Abramowitz & Stegun 1972).
REFERENCES
A BRAMOWITZ , M. & S TEGUN , I. A. 1972 Handbook of Mathematical Functions with Formulas,

Graphs, and Mathematical Tables. (Applied Mathematics Series), vol. vol. 55. National Bureau
of Standards, (10th printing).
A DEROGBA , K. & B LAKE , J. R. 1978 Action of a force near the planar surface between two
semi-infinite immiscible liquids at very low Reynolds numbers. Bull. Austral. Math. Soc. 18
(3), 345–356.
A NGELINI , T. E., ROPER , M., K OLTER , R., W EITZ , D. A. & B RENNER , M. P. 2009 Bacillus subtilis
spreads by surfing on waves of surfactant. Proc. Natl Acad. Sci. USA 106, 18109–18113.
B EARON , R. N. 2003 An extension of generalized Taylor dispersion in unbounded homogeneous
shear flows to run-and-tumble chemotactic bacteria. Phys. Fluids 15, 1552–1563.
B EARON , R. & P EDLEY, T. 2000 Modelling run-and-tumble chemotaxis in a shear flow. Bull. Math.
Biol. 62, 775–791.
B ERG , H. C. 2003 E. coli in Motion. Springer.
B ERKE , A. P., T URNER , L, B ERG , H. C. & L AUGA , E. 2008 Hydrodynamic attraction of swimming
microorganisms by surfaces. Phys. Rev. Lett. 101, 038102.
B LAKE , J. R. 1971 A note on the image system for a stokeslet in a no-slip boundary. Math. Proc.
Camb. Philos. Soc. 70 (2), 303–310.
B RENNER , M. P., L EVITOV, L. S. & B UDRENE , E. O. 1998 Physical mechanisms for chemotactic
pattern formation by bacteria. Biophys. J. 74, 1677–1693.
B UDRENE , E. O. & B ERG , H. C. 1991 Complex patterns formed by motile cells of Escherichia
coli. Nature 349, 630–633.
B UDRENE , E. O. & B ERG , H. C. 1995 Dynamics of formation of symmetrical patterns by chemotactic
bacteria. Nature 376, 49–53.
C HANDRASEKHAR , S. 1961 Hydrodynamic and Hydromagnetic Stability. Dover.
C HEN , K. C., F ORD , R. M. & C UMMINGS , P. T. 2003 Cell balance equation for chemotactic bacteria
with a biphasic tumbling frequency. J. Math. Biol. 47, 518–546.
C HENG , S. Y., H EILMAN , S., WASSERMAN , M., A RCHER , S., S HULER , M. L. & W U , M. 2007 A
hydrogel-based microfluidic device for the studies of directed cell migration. Lab on a Chip
7, 763–769.
C HILDRESS , S., L EVANDOWSKY, M. & S PIEGEL , E. A. 1975 Pattern formation in a suspension of
swimming microorganisms: equations and stability theory. J. Fluid Mech. 69, 591–613.
C ISNEROS , L., D OMBROWSKI , C., G OLDSTEIN , R. E. & K ESSLER , J. O. 2006 Reversal of bacterial
locomotion at an obstacle. Phys. Rev. E 73, 030901.
D I L UZIO , W. R., T URNER , L., M AYER , M., G ARSTECKI , P., W EIBEL , D. B., B ERG , H. C. &
W HITESIDES , G. M. 2005 Escherichia coli swim on the right-hand side. Nature 435 (7046),
https://doi.org/10.1017/jfm.2013.628
1271–1274.
D OMBROWSKI , C., C ISNEROS , L., C HATKAEW, S., G OLDSTEIN , R. E. & K ESSLER , J. O. 2004
Self-concentration and large-scale coherence in bacterial dynamics. Phys. Rev. Lett. 93, 098103.
E ZHILAN , B., PAHLAVAN , A. A. & S AINTILLAN , D. 2012 Chaotic dynamics and oxygen transport
in thin films of aerotactic bacteria. Phys. Fluids 24 (9), 091701.
F ERRELL , R. T. & H IMMELBLAU , D. M. 1967 Diffusion coefficients of nitrogen and oxygen in
water. J. Chem. Engng Data 12 (1), 111–115.
H ERNANDEZ -O RTIZ , J. P., S TOLTZ , C. G. & G RAHAM , M. D. 2005 Transport and collective
dynamics in suspensions of confined swimming particles. Phys. Rev. Lett. 95, 204501.
H ILL , N. A., P EDLEY, T. J. & K ESSLER , J. O. 1989 Growth of bioconvection patterns in a suspension
of gyrotactic micro-organisms in a layer of finite depth. J. Fluid Mech. 208, 509–543.

H ILLESDON , A. J. & P EDLEY, T. J. 1996 Bioconvection in suspensions of oxytactic bacteria: linear
theory. J. Fluid Mech. 324, 223–259.
H OHENEGGER , C. & S HELLEY, M. J. 2010 Stability of active suspensions. Phys. Rev. E 81, 046311.
JÁNOSI , I. M., K ESSLER , J. O. & H ORVÁTH , V. K. 1998 Onset of bioconvection in suspensions of
Bacillus subtilis. Phys. Rev. E 58, 4793–4800.
K ALININ , Y. V., J IANG , L., T U , Y. & W U , W. 2009 Logarithmic sensing in Escherichia coli bacterial
chemotaxis. Biophys. J. 96, 2439–2448.
K ASYAP, T. V. & K OCH , D. L. 2012 Chemotaxis driven instability of a confined bacterial suspension.
Phys. Rev. Lett. 108, 038101.
K ELLER , E. F. & S EGEL , L. A. 1971 Model for chemotaxis. J. Theor. Biol. 30, 225–234.
K IM , M. J. & B REUER , K. S. 2004 Enhanced diffusion due to motile bacteria. Phys. Fluids 16,
L78–L81.
K IM , M. J. & B REUER , K. S. 2007 Controlled mixing in microfluidic systems using bacterial
chemotaxis. Analyt. Chem 79, 955–959.
K IM , S. & K ARRILA , S. J. 1991 Microhydrodynamics. Butterworth-Heinemann.
K OCH , D. L. & S UBRAMANIAN , G. 2011 Collective hydrodynamics of swimming microorganisms:
living fluids. Annu. Rev. Fluid Mech. 43, 637–659.
L AUGA , E., D I L UZIO , W. R., W HITESIDES , G. M. & S TONE , H. A. 2006 Swimming in circles:
motion of bacteria near solid boundaries. Biophys. J. 90 (2), 400–412.
L AUGA , E. & P OWERS , T. R. 2009 The hydrodynamics of swimming microorganisms. Rep. Prog.
Phys. 72, 096601.
L I , G. & TANG , J. X. 2009 Accumulation of microswimmers near a surface mediated by collision
and rotational Brownian motion. Phys. Rev. Lett. 103, 078101.
L USHI , E., G OLDSTEIN , R. E. & S HELLEY, M. J. 2012 Collective chemotactic dynamics in the
presence of self-generated fluid flows. Phys. Rev. E 86, 040902.
M ENDELSON , N. H., B OURQUE , A., W ILKENING , K., A NDERSON , K. R. & WATKINS , J. C. 1999
Organized cell swimming motions in Bacillus subtilis colonies: patterns of short-lived whirls
and jets. J. Bacteriol. 181, 600–609.
M ESIBOV, R. & A DLER , J. 1972 Chemotaxis toward amino acids in Escherichia coli. J. Bacteriol.
112, 315–326.
P EDLEY, T. J. & K ESSLER , J. O. 1992 Hydrodynamic phenomena in suspensions of swimming
microorganisms. Annu. Rev. Fluid Mech. 24, 313–358.
R IVERO , M. A., T RANQUILLO , R. T., B UETTNER , H. M. & L AUFFENBURGER , D. A. 1989 Transport
models for chemotactic cell populations based on individual cell behaviour. Chem. Engng Sci.
44, 2881–2897.
S AINTILLAN , D. & S HELLEY, M. J. 2007 Orientational order and instabilities in suspensions of
self-locomoting rods. Phys. Rev. Lett. 99, 058102.
S AINTILLAN , D. & S HELLEY, M. J. 2008a Instabilities and pattern formation in active particle
suspensions: kinetic theory and continuum simulations. Phys. Rev. Lett. 100, 178103.
S AINTILLAN , D. & S HELLEY, M. J. 2008b Instabilities, pattern formation, and mixing in active
suspensions. Phys. Fluids 20, 123304.
S ARTORI , P. & T U , Y. 2011 Noise filtering strategies in adaptive biochemical signaling networks. J.
Stat. Phys. 142 (6), 1206–1217.
S IMHA , R. A. & R AMASWAMY, S. 2002 Hydrodynamic fluctuations and instabilities in ordered
https://doi.org/10.1017/jfm.2013.628
suspensions of self-propelled particles. Phys. Rev. Lett. 89, 058101.

S OKOLOV, A., A RANSON , I. S., K ESSLER , J. O. & G OLDSTEIN , R. E. 2007 Concentration
dependence of the collective dynamics of swimming bacteria. Phys. Rev. Lett. 98, 158102.
S OKOLOV, A., G OLDSTEIN , R. E., F ELDCHTEIN , F. I. & A RANSON , I. S. 2009 Enhanced mixing
and spatial instability in concentrated bacterial suspensions. Phys. Rev. E 80, 031903.
S TOCKER , R. 2011 Reverse and flick: hybrid locomotion in bacteria. Proc. Natl Acad. Sci. USA
108 (7), 2635–2636.
S UBRAMANIAN , G. & K OCH , D. L. 2009 Critical bacterial concentration for the onset of collective
swimming. J. Fluid Mech. 632, 359–400.

S UBRAMANIAN , G., K OCH , D. L. & F ITZGIBBON , S. R. 2011 The stability of a homogeneous
suspension of chemotactic bacteria. Phys. Fluids 23, 041901.
U NDERHILL , P. T., H ERNANDEZ -O RTIZ , J. P. & G RAHAM , M. D. 2008 Diffusion and spatial
correlations in suspensions of swimming particles. Phys. Rev. Lett. 100, 248101.
V UPPULA , R. R., T IRUMKUDULU , M. S. & V ENKATESH , K. V. 2010 Chemotaxis of Escherichia
coli to L-serine. Phys. Biol. 7, 026007.
W U , X. L. & L IBCHABER , A. 2000 Particle diffusion in a quasi-two-dimensional bacterial bath.
Phys. Rev. Lett. 84, 3017–3020.
https://doi.org/10.1017/jfm.2013.628

Instability of An Inhomogeneous Bacterial Suspension Subjected To A Chemoattractant Gradient PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Instability of An Inhomogeneous Bacterial Suspension Subjected To A Chemoattractant Gradient PDF

Uploaded by

Copyright:

Available Formats

J. Fluid Mech. (2014), vol. 741, pp 619–657.

c Cambridge University Press 2014 619

Instability of an inhomogeneous bacterial

suspension subjected to a chemo-attractant

(Received 3 February 2013; revised 6 November 2013; accepted 26 November 2013;

The stability of a suspension of chemotactic bacteria confined in an infinitely long

Key words: instability, micro-organism dynamics, suspensions

† Email address for correspondence: dlk15@cornell.edu

wild-type E. coli in the presence of a horizontal chemo-attractant or chemo-repellent

aggregation instability originating from the reaction–diffusion coupling between the

Ω(x, p, t) = n(x, t) f (p), (2.2)

where f (p) is governed by the integral equation

τ0−1 (1 − ζ p · g), p · g > 0,

where the dipole moment S is given by (Kasyap & Koch 2012)

Being a pusher, a bacterium exerts a pressure in the z direction (negative normal

(n0 , u0 , w0 , p0 ) = [N(z), U(z), W(z), P(z)] exp(ikx + σ t), (2.16)

1 exp(−Pe)(1 + Pe) − 1 Pe[1 + exp(−Pe)] + 2[exp(−Pe) − 1] 2

The suspension becomes unstable if β exceeds βcrit , and figure 1 illustrates

(a) 1.00 (b) 1.00

aforementioned fluid flow when acting on the inhomogeneous base-state bacterial

(σ + k2 )hNi = −PehWDn0 i = −ikhUn0 i, (2.23)

flow as shown by the streamlines. Since the hydrodynamic disturbance created by a

σ = k2 (Peαβ − 1) + O(k3 ), (2.24)

which is quadratic in the wavenumber. Thus the ‘most dangerous’ wavenumber,

3. Theory for deep channels

theory quantitatively accurate. We solve the deep-channel problem by means of

3.1. Analysis of case I

N(z̄) = exp(−z̄) + O(Pe−2 ), (3.7a)

obtained by integrating (3.3a) across the boundary layer as

3.2. Analysis of case II

Boundary layer solutions with the above-mentioned asymptotic expansions are

3.4. Analysis for a semi-infinite channel

[D2 + D − (σ̃ + k̃2 )]N = −W exp(−z), (3.16a)

4. Numerical solution for finite Pe and k

φ (z) = exp(−Pez), (4.2a)

in which φ0 corresponds to the eigenfunction in the long-wavelength case (see

(D2 + PeD + λ2j )φj = 0, (4.3a)

4.1. Validation of the numerical solution

We begin by presenting the neutral curves, βc versus k, for various values of Pe in

0 20 40 60 100 101 102

0 0.1 0.2 0.3 0.4 0.5

(a) 100 (b) 240

(a) 1.0 (b) 1.0

accessible value of β as a function of Pe as a dash-dotted line in figure 13(a).

5. Effect of shear rotation

(a) 400 (b) 60

5.1. Derivation of conservation equations of a chemotactic bacterial suspension

suspension including both fluid shear-induced rotation and chemo-gradient-induced

and by using (5.2) is

the deviation from isotropy caused by a non-uniform bacterial concentration field

5.2. Stability analysis with shear rotation

[D2 + PeD − (σ + k2 )]N = 3HWDn

since β has been defined as β = 83 Chn0 iL2 H.

(a) 1.2 (b) 1.2

0 50 100 150 0 200 400 600

0 0.05 0.10 0 0.04 0.08 0.12

γcrit but also the range of H

exhibited no bacterial concentration fluctuations in the large-wavelength limit, whereas

at which any inhomogeneity could be observed. However, if we take the critical

Appendix A. Justification for neglecting the attractant transport by fluid flow

Appendix B. Integrals in (5.15)

A BRAMOWITZ , M. & S TEGUN , I. A. 1972 Handbook of Mathematical Functions with Formulas,

of gyrotactic micro-organisms in a layer of finite depth. J. Fluid Mech. 208, 509–543.