week ending

PRL 110, 184503 (2013) PHYSICAL REVIEW LETTERS 3 MAY 2013

Nonlinear Amplification in Electrokinetic Pumping in Nanochannels in the Presence of

Hydrophobic Interactions
Suman Chakraborty,1,* Dipankar Chatterjee,2 and Chirodeep Bakli1
1
Department of Mechanical Engineering, Indian Institute of Technology, Kharagpur 721302, India
2
Simulation and Modeling Laboratory, CSIR-Central Mechanical Engineering Research Institute, Durgapur 713209, India
(Received 7 February 2012; revised manuscript received 27 February 2013; published 3 May 2013)
We discover a nonlinear coupling between the hydrophobicity of a charged substrate and electrokinetic
pumping in narrow fluidic confinements. Our analyses demonstrate that the effective electrokinetic
transport in nanochannels may get massively amplified over a regime of bare surface potentials and
may subsequently get attenuated beyond a threshold surface charging condition because of a complex
interplay between reduced hydrodynamic resistance on account of the spontaneous inception of a less
dense interfacial phase and ionic transport within the electrical double layer. We also show that the
essential physics delineated by our mesoscopic model, when expressed in terms of a simple mathematical
formula, agrees remarkably with that portrayed by molecular dynamics simulations. The nontrivial
characteristics of the initial increment followed by a decrement of the effective zeta potential with a
bare surface potential may open up the realm of hitherto-unexplored operating regimes of electro-
hydrodynamically actuated nanofluidic devices.

DOI: 10.1103/PhysRevLett.110.184503 PACS numbers: 47.61.
k, 47.65.
d

The physics of electrically modulated fluidic transport in
narrow confinements is central to the understanding of
several processes of practical relevance, with applications
ranging from the transport and separation of charged col-
loids to the realization of miniaturized energy conversion
devices and systems. Many of these challenging applica-
tions involve the formation of a charged interfacial layer in
the vicinity of the solid substrates, also known as the
electrical double layer (EDL) [1]. Under the action of an
external driving force, excess ionic charges present within
this layer may be preferentially transported across the
channel. In certain cases, this pumping effect may get
further amplified by an effective reduction of the resistive
forces in the interfacial layer, as attributed to the inception
of a less dense phase [2–4]. The reduced interfacial friction,
in turn, may amplify the rate of ionic pumping, leading to
the possibilities of realizing an apparently hyperfluidic
electrohydrodynamic transport over small scales. While
related consequences for microscale devices [5] sound
more trivial, several aspects of the corresponding implica-
tions over nanometer length scales remain unclear.
Considering the hydrodynamics within the EDL over
slipping surfaces, researchers have demonstrated that an
effective interfacial slipping phenomenon may amplify
the extent of electrokinetic pumping [2] in nanochannels
to an extent, as manifested through augmentations in the
effective interfacial (zeta) potential,
eff (which is linearly
related to the bare surface potential). While this kind of
linear response has been verified by extensive molecular
dynamics (MD) simulations, its validity over a broader
regime has by far remained questionable.
Can complex electrochemical-hydrodynamic interac-
tions over the relevant physical scales give rise to nonlinear
and nonmonotonic amplifications in electrokinetic pump-
ing in nanofluidic channels, beyond the above regimes
hitherto addressed? Unlike the concerned linear regime
addressed in the previous studies, here we demonstrate
that the same is indeed a possibility, although not a trivi-
ality. Our studies do reveal that the concerned relative
enhancement in the effective zeta potential, in general,
may be characterized through a complex nonlinear depen-
dence on the bare surface potential, in sharp contrast to the
classical linear regime. We arrive at this intriguing con-
jecture by interpreting the concerned complex physical
interactions over reduced length scales in a combined
phase field-lattice Boltzmann (PFLB) framework (which
is different from standard LB models for electrokinetics
[6]), instead of following computationally more intensive
multiscale simulation routes [7]. Irrespective of the under-
lying complexities in the physical issues prevalent within
the newly discovered nonlinear regime, we finally establish
an apparently universal and generalized functional form
for the amplification in the effective zeta potential by using
phase-field thermodynamics [8]. Our prediction agrees
remarkably with the effective zeta potential predicted
directly from MD simulations, albeit in a computationally
efficient paradigm [9,10].
To begin with our PFLB model description, we first
consider generic distribution functions (f , fu , f c ) for
the phase fraction () evolution, velocity (u) field, and
the electrical potential ( c ) distribution, respectively. Here,
we capture the phase fraction evolution on account of
hydrophobic interactions through an order parameter,
defined as  ¼ ðn1
n2 Þ=ðn1 þ n2 Þ, where n1 and n2 rep-
resent the number densities of the depleted and liquid
phases, respectively. In the absence of any external forcing

0031-9007=13=110(18)=184503(5) 184503-1 Ó 2013 American Physical Society

 week ending
PRL 110, 184503 (2013) PHYSICAL REVIEW LETTERS 3 MAY 2013

parameters, these distribution functions may evolve accord-
ing to the discretized LB equation as [11] fi ðxþci
t;tþ

tÞ
fi ðx;tÞ¼
ð1= Þ½fi ðx;tÞ
fi;eq ðx;tÞ, where  2
ð; u; c Þ, fi;eq ðx; tÞ are the local equilibrium distribution
functions, and ci are the discrete lattice velocity vectors.
The distribution functions are related to the pertinent trans-
P P P tive, we choose the following parameter values [11]: B ¼
port parameters as  ¼ i fiu , u ¼ i fiu ci ,  ¼ i fi .
The corresponding equilibrium distribution functions may
be derived by expanding those in terms of power series in
the local velocity [12]. We take the parameters ci following
We set up a model problem which we first investigate
through MD simulations. The problem essentially addresses
electrokinetic transport of a sodium chloride-water solution
of a given concentration through a nanofluidic channel of a
given dimension as well as substrate wettability [14].
To implement the above problem from a PFLB perspec-

A ¼ 0:003, k ¼ 0:01, ug ¼ 1, ul ¼ 0:8. This yields the
following
qffiffiffiffiffiffifor the liquid-vapor surface tension coefficient:
lv ¼ 8kB 9 . Further, as measured in lattice units, lv 
lu

a standard D2Q9 model [11] with weighting functions for 0:105  0:002 [3]. One may thus obtain the lattice spacing
the nine lattice directions w0 ¼ 4=9, wi ¼ 1=9 for i 2 in physical units by relating lv with lu lv in the following
ð1; 4Þ and wi ¼ 1=36 for i 2 ð5; 8Þ, whereas we relate the manner [3]: lv ¼ kB T=
x2 lu lv , so that
x  0:3 nm for
pffiffifficonstant (
x) and time step (
t) as c ¼
x=
t;
lattice T ¼ 300 K. This dimension is indeed comparable with the
c ¼ 3cs ; cs being the sonic speed. For implementing atomic ranges of potentials used in typical MD simula-
position-dependent body forces dictating the momentum tions. Further, we consider  c ¼ 1 [6].
transport [13], we add a forcing term to the right-hand side We first depict a representative variation of slip length
of the evolution equation of the velocity distribution (ls ) with a nondimensional bare surface potential (
 ¼
u ¼ wi ð1
ð1=2 u 
equation as fsource P ÞÞ½3ðci
u Þ þ 9ðci  ze
o =kB T). The complex variations in slip length, as por-
u Þci   FðxÞ, where u ¼ ð1=Þ½ i fiu ci þ ðF=2Þ. trayed in Fig. 1, essentially stem from the fact that an
Physically consistent prescription of FðxÞ, mentioned uncharged hydrophobic surface would not have favored
as above, turns out to be a critically important considera-
tion for implementing the present model, in consistency
with the electrochemical-thermodynamic interactions in
narrow confinements subjected to hydrophobic interac-
tions. Fundamentally, momentum transport due to electri-
cal effects is explicitly linked with an electrokinetic body
force, as
 
1 1 @"el
FEK ¼ e E
E  Er"el þ r E  E  ;
2 2 @
where E is the local electric field (¼
r c ), "el is the local
permittivity of the medium, and e is the electrical charge
density. This forcing term, thus, is directly linked with the
Poisson equation for electrical potential distribution. In
order to ensure that an unsteady term does not feature in
the governing equation for electrical potential distribution
[7], we append an additional source term in the right-hand
side of the discretized LB form of the evolution equation
for fic , in the form:
tw i RD, where R is the right-hand
side of the Poisson equation (which cooperates the wall- FIG. 1 (color online). Variation of slip length, ls , with non-
adjacent density variations [2] in the description of the dimensional bare surface potential shown for a representative
concentration profiles, over and above the traditional case relating salt concentration to the surface charge by the ratio
Poisson-Boltzmann picture), D ¼ c2 ð c
ð1=2ÞÞð
t=3Þ, Nþ =Nw ¼ 3:2 (Case 1) and Nþ =Nw ¼ 2:4 (Case 2), where Nþ is
the number of Naþ ions in an unit cell and Nw e is the amount of
and w i ¼ wi for i 2 ð1; 8Þ; w i ¼ ðw0
1Þ for i ¼ 0. The charge on a channel wall confining the unit cell. The channel
corresponding equilibrium distribution function is given by wettability is defined by a static contact angle of 110. The slip
f0c ;eq ¼ ðw=0
1Þf0c ; fic ;eq ¼ w=i fic (i
0), where w=0 ¼ lengths tend to gradually decrease with an increase of bare surface
1=9, w=i ¼ 1=8 (i
0). The resultant electrical potential potential, beyond a critical limit. Extremely large bare surface
P
distribution: c ¼ i fic =ð1
w=0 Þ, satisfies the Poisson
potential causes an abrupt drop in slip length, making the slip
independent of salt concentration, wall charge, or further rise in
equation, which may be recovered by executing a zeta potential. Markers represent MD simulation data, and solid
Chapman-Enskog expansion. We augment the interaction lines represent corresponding PFLB model predictions. Several
forces in the PFLB model further to address some addi- other values of Nþ =Nw are studied, resulting no qualitative
tional facets of electromechanical interactions over changes in the trends observed (not reported in the figure, for
reduced scales [14,15]. the sake of clarity).

184503-2
 week ending
PRL 110, 184503 (2013) PHYSICAL REVIEW LETTERS 3 MAY 2013

accumulation of water in the near-surface region and the surface potential. Interestingly, beyond a threshold limiting
same would get depleted having a lower than bulk density value of the surface potential, this additional driving force
[16]. However, introduction of a surface potential favors tends to saturate. Since an increase (or decrease) in this
increments in the local number density of water at the interfacial driving force may be interpreted to result in an
solvent-substrate interface, beyond a threshold limit of increase (or decrease) in the extent of the effective slip,
the surface potential. It is important to mention in this this clearly explains the observed trends in slip length,
context that despite an inherently mesoscopic character purely from mesoscopic simulation considerations. These
of the PFLB model employed here, it implicitly takes trends may further be justified from the interfacial density
into account some subgrid scale physical considerations variations [14].
depicting certain facets of substrate-solvent-solute interac- How do the above variations in slip length translate in
tions, through the description of an augmented free energy the form of an effective zeta potential? Towards answering
functional. Importantly, these considerations include the this question, we first obtain the effective zeta potential
solvation (hydration) interactions between the ionic con- directly from the MD simulations by calculating the ionic
stituents [17], resulting in a modification in the description current as [2] Ie ¼
ðð"d
eff Þ=ÞAf0 , where  is the shear
of the free energy. In our PFLB model, the corresponding viscosity, A is the fluid slab cross area, and f0 is the force
effective chemical potential (obtained from the variation per unit volume applied in streaming current experiment.
in the additional component of the free energy with respect The consequent variation is essentially plotted in Fig. 2.
to the order parameter), eff , eventually gives rise to an From Fig. 2, it is apparent that there occurs an initial
amplification in
eff with increments in
,  followed by a
additional effective forcing function: Feff ¼ eff r in
the momentum equation (where  is the order parameter dip in the same. This kind of trend appears to be a con-
field), so as to address the corresponding physical impli- sequence of complicated nonlinear interactions between
cations in a mesoscopic sense. the potential profile and the slip length, both being sensi-
From Fig. 1 it is evident that extremely low surface tively dependent on the ionic concentration, as well as the
potentials do not seem to induce any variation in the slip bare surface potential, for a given substrate wettability.
We next formulate a simple yet insightful continuum-
length. As the potential is increased, slip lengths gradually
 slip lengths based semianalytical platform. Towards that, we first reit-
decrease. Beyond a threshold increase in
,
erate that intricate interfacial interactions over small scales
start decreasing abruptly. These avalanches in the slip
may alter the near-wall arrangement of water molecules.
length values are observed for all values of salt concen-
This, in turn, modifies the Boltzmann distribution of
trations and occur almost at the same bare surface potential
ions by appending an extra potential term as n ¼
value for different values of ionic concentration [18].
no expððze c
c ext Þ=kB TÞ, where c is electrostatic
Finally, at very high surface charge densities, the slip
lengths are observed not only to become insensitive to
salt concentration but also to a further rise of wall charge
density. A further increase of surface potential, in effect,
leads to overcrowding of the surface with counterions and
slip length variation tends to plateau. In effect, with a large
counterion population guided by the high bare surface
potential, the near-wall fluid-ion layering attains saturation
and a further increase in surface potential does not induce
any further appreciable change in the interfacial density
distribution. The high electrostatic force for a charged
surface, under such circumstances, overrules the solvation
forces (which could otherwise have introduced a concen-
tration dependency), leading to slip lengths being indepen-
dent of the solution concentration. As is evident from Fig. 1,
very good agreements may indeed be observed between the
PFLB simulation predictions and MD predictions.
In order to delve deeper into the underlying mesoscopic
physics addressed through our PFLB simulations, we have
clearly revealed that below a critical magnitude of the
FIG. 2 (color online). Comparison between the effective zeta
surface potential, the value of Feff due to additional subgrid potential predictions as obtained from MD simulations and the
scale interactions implicitly accounted for by the PFLB same obtained from semianalytical (MPB) theory in conjunction
model does not vary appreciably, leading to nearly invari- with PFLB simulation database for slip length as well as water
ant slip lengths. However, beyond this critical limit, the density fluctuations. Simulation data are the same as those
value of Feff decreases with an increase in the value of the considered for plotting Fig. 1.

184503-3

PRL 110, 184503 (2013) PHYSICAL REVIEW LETTERS
potential, c ext ¼
kB T logðf ðyÞ=fb Þ [2], fb being the
bulk density of water and f is the water density variation
along the y (transverse) direction. no , nþ , and n
are the
bulk ion density, cation density, and anion density, respec-
tively. Thus, we obtain a modified Poisson-Boltzmann
(MPB) equation, for a z:z symmetric electrolyte, as
   
d dc f ðyÞ ze c
"d ¼ 2zeno sinh :
dz dy fb kB T
Traditionally, the ratio ðf ðyÞÞ=fb is assumed to be unity
[2], which very well reproduces the simulation results for
low values of zeta potential (typically, within Debye-
Hückel limits). However, here we assess the specific
contribution from this term by appealing to the PFLB
simulation predictions that offer predictions to the near-
wall density profiles. Based on this variation, we obtain the
gradient of c , numerically, by integrating the modified PB
equation described as above. Next, we integrate the
Navier-Stokes equation with neglected acceleration terms
and in presence of an applied electric field, Ex , along the
channel axis: l ðd2 u=dy2 Þ
"l ðd2 c =dy2 ÞEx ¼0. Applying
the partial slip boundary condition as obtained from com-
prehensive numerical simulations, a solution to the above
equation (with centerline symmetry) yields:
  
c l dc   electrokinetic pumping in nanochannels. Our studies have
"
E 

u¼
l x 1

s  :
l

dy  y¼0
Noting that along the center line, uc ¼
ðð"l
eff Ex Þ=l Þ, it
follows that
  

eff c c ls d c  

 formalism. This, in turn, provides an effective tool for the
¼ 1

 ;



dy y¼0
where y ¼ 0 represents a transverse location where the
partial slip boundary condition is applied. This simple
estimate of effective amplification in electrokinetic pump-
ing, with ls data obtained through the PFLB route,
compares excellently with the MD based simulation
predictions, as evident from Fig. 2. The implications of
this elusively obvious agreement, however, are not far from
being nontrivial. This may be attributed to the fact that ls
itself is a parameter that is sensitively dependent on
,
ionic concentration, and substrate wettability. We have
already established that this dependence may be success-
fully captured through PFLB formalism (see Fig. 1). Once
a substitution of this ls dependence in Eq. (1) has been
made, the data corresponding to a wide range of combina-
tions of substrate wettability, ionic concentration, and bare
surface potential are found to remarkably converge to
identical characteristics of
eff =
, in line with the corre-
sponding MD simulation predictions. Such universal
trends are also physically reminiscent of the fact that
electrokinetic pumping in the presence of hydrophobic
interactions is a generic nonlinear function of the intrinsic
interactions between the underlying electromechanical and
184503-4
week ending
3 MAY 2013
hydrodynamic influences, as manifested through the inter-
facial physics delineated above.
How can we explain the nontrivial physics associated
with the nonmonotonic trends in the effective interfacial
potential? Towards addressing this question, it is first
imperative to note that the same is essentially decided by
an intricate interplay between molecular slip and the den-
sity fluctuations over interfacial scales. As shown in Fig. 1,
the slip lengths remain virtually unaltered for low bare
surface potentials, and subsequently tend to decrease
with increments in the same till a ‘‘sticking’’ condition is
attained. On the other hand, the interfacial density fluctua-
tions, as appearing in the MPB equation, first increase with
increasing surface charge, until a threshold charging con-
dition is reached [14]. Below this threshold condition incre-
ments in surface charge induce more fluctuations in the
water density profile, which in turn mobilizes the layering
of the ions adjacent to the interface and generates greater
current in the flow. However, beyond the threshold limit, the
fluctuations are reduced owing to the overcrowding of the
channel with salt ions. The resultant counteracting effects of
slip and density profiles essentially give rise to the effective
pumping characteristics as depicted in Fig. 2.
To summarize, we have discovered a hitherto-
unexplored nonlinear regime of massive amplification in
demonstrated that despite the underlying complexities, the
corresponding amplification factor may be expressed in the
form of a universal nonlinear functional dependence, in
which the slip lengths as well as the EDL potential gra-
dients may be obtained successfully through a mesoscopic
design and synthesis of nanofluidic devices with enhanced
capabilities, and possibly leading towards the practical
realizability of hypertransportive electrohydrodynamic
systems over nanoscopic scales.
*suman@mech.iitkgp.ernet.in
[1] R. J. Hunter and L. R. White, Foundations of Colloid
Science (Clarendon, Oxford, 1989).
[2] L. Joly, C. Ybert, E. Trizac, and L. Bocquet, Phys. Rev.
Lett. 93, 257805 (2004); J. Chakraborty, S. Pati,
S. K. Som, and S. Chakraborty, Phys. Rev. E 85, 046305
(2012); C. Bakli and S. Chakraborty, J. Chem. Phys. 138,
054504 (2013); A. Bandopadhyay and S. Chakraborty,
Langmuir 28, 17 552 (2012).
[3] M. Sbragaglia, R. Benzi, L. Biferale, S. Succi, and
F. Toschi, Phys. Rev. Lett. 97, 204503 (2006).
[4] A. Ajdari and L. Bocquet, Phys. Rev. Lett. 96, 186102
(2006); E. Lauga, M. P. Brenner, and H. A. Stone,
Handbook of Experimental Fluid Dynamics (Springer,
New York, 2007).
[5] V. Tandon and B. J. Kirby, Electrophoresis 29, 1102
(2008).
[6] Z. Chai and B. Shi, Phys. Lett. A 364, 183 (2007).

PRL 110, 184503 (2013) PHYSICAL REVIEW LETTERS
[7] R. Qiao and N. R. Aluru, Int. J. Multiscale Comput. Eng.
2, 173 (2004).
[8] A. Fakhari and M. H. Rahimian, Phys. Rev. E 81, 036707
(2010).
[9] Despite incurring fewer approximations than other meth-
ods, MD simulations on electrokinetic transport are re-
strained by the following challenges [10]: (i) if the ionic
concentration is in order of magnitude lower than the
solvent concentration then statistical accuracy is achieved
only at the expense of long simulation runs, and
(ii) typical electrokinetic speeds are much slower than
the characteristic thermal speed, necessitating additional
care to extract the signal amidst the background noise.
These considerations, in addition to favorable computa-
tional costs, impart the motivation for developing a PFLB
model that can be closely sensitive to MD simulation
data. As a quantified measure, a system containing 7176
water molecules simulated for 5 ns takes 100 physical
hours on a 3.4 GHz Core i7-2600 processor for simulating
the MD results presented here, whereas the corresponding
computational time is close to 10 physical hours for the
PFLB simulation.
[10] U. M. B. Marconi and S. Melchionna, Langmuir 28, 13727
(2012).
184503-5
week ending
3 MAY 2013
[11] J. Horbach and S. Succi, Phys. Rev. Lett. 96, 224503 (2006).
[12] A. J. Briant and J. M. Yeomans, Phys. Rev. E 69, 031603
(2004).
[13] Z. Guo, C. Zheng, and B. Shi, Phys. Rev. E 65, 046308
(2002).
[14] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.110.184503 for addi-
tional discussion on electro-mechanical interactions, and
details of MD simulations.
[15] D. M. Huang, C. Cottin-Bizonne, C. Ybert, and L.
Bocquet, Phys. Rev. Lett. 98, 177801 (2007); D. M.
Huang, P. L. Geissler, and D. Chandler, J. Phys. Chem.
B 105, 6704 (2001); L. Onsager and N. N. T. Samaras,
J. Chem. Phys. 2, 528 (1934).
[16] S. Chakraborty, Phys. Rev. Lett. 99, 094504 (2007); 100,
097801 (2008); 101, 184501 (2008); D. R. Berard, P.
Attard, and G. N. Patey, J. Chem. Phys. 98, 7236 (1993).
[17] Y. Burak and D. Andelman, Phys. Rev. E 62, 5296 (2000);
J. Chem. Phys. 114, 3271 (2001).
[18] Here we express the salt concentration in terms of the
channel height to Debye length ratio. The Debye length is
an equivalent representation
P of the ionic concentration,
given by D ¼ ð"l kB T= N n q2 Þ0:5 ; n being the con-
j¼1 j j j
centration of species j with charge qj .