UNIVERSITY OF CALIFORNIA Santa Barbara

Nonlinear Phenomena in Induced Charge Electroosmosis

A Dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Mechanical Engineering

by

Gaurav Soni

Committee in charge: Professor Carl D. Meinhart, Chair Professor George M. Homsy Professor Todd M. Squires Professor Hyongsok T. Soh

December 2008

The dissertation of Gaurav Soni is approved.

____________________________________________ George M. Homsy

____________________________________________ Todd M. Squires

____________________________________________ Hyongsok T. Soh

____________________________________________ Carl D. Meinhart, Committee Chair

October 2008

ii

Nonlinear Phenomena in Induced Charge Electroosmosis

Copyright 2008 by Gaurav Soni

iii

To my loving parents, Ramji and Sushila

iv

ACKNOWLEDGEMENTS

The work contained in this dissertation would have been impossible without the support of my Ph.D. advisor, Professor Carl Meinhart. The dictionary does not have enough words to express my gratitude towards Carl. He always showed confidence in my abilities. He showed me the path and prevented me from falling in the pitfalls. He gave me the freedom, however, to digress and pick up knowledge from foreign territories. I am grateful to my co-advisor and committee member, Professor Todd Squires for pulling me out of scientific deadlocks. He provided me with a unique insight and theoretical framework for solving complex problems. His advice often brought a turning point in my research. I am thankful to my committee member Professor George Bud Homsy for evaluating the merit of my work and offering useful insights through his enlightening comments during my exams. It was a wonderful experience taking lessons in fluid dynamics from him and also TAing for him. Thanks are also due to my committee member, Professor Tom Soh, not only for evaluating my work but also for allowing me to borrow equipment and books from his lab. This work would have been very difficult to finish without the support, comraderie and friendship of the past and present members of the Meinhart Lab. It is a pleasure for me to mention their names here and say thanks to them: Hope Feldman, Marin Sigurdson, Matthew Pommer, Frederic Bottausci, Caroline Cardonne, Stephen Bradford, Brian Piorek and Lisan Viel. Special thanks to Dr. Changsong Ding for teaching me Titanium microfabrication and to Dr. Adam v

Monkowski for making the nanoscale structures used in the chapter six of this dissertation. I want to take this opportunity to thank my friends Mr. Amitabh Virmani, Mr. Ankur Saxena and Mr. Amarendra Singh for their friendship and moral support. My words are not enough to express my thankfulness to my father Ramji, and mother Sushila. I simply couldnt have achieved this milestone without their love, support and the sacrifices that they made for my education. I am grateful to my brothers Tej and Vikas for always being there for me and for helping me in all phases of my life. I am thankful to my sister-in-law Ambika for her kind love. Finally, I want to say thanks to my wife Ritoo Varma for believing in me and loving me unconditionally.

vi

VITA OF GAURAV SONI October 2008 gaurav.soni@gmail.com EDUCATION Ph.D. (Mechanical Engineering) University of California Santa Barbara, CA, USA, Nov 2008 Advisor: Carl Meinhart Bachelor of Technology (Mechanical Engineering) Indian Institute of Technology Delhi, India, May 2002 INDUSTRIAL EXPERIENCE Support Engineer Comsol Inc., Los Angeles, CA, July Sept 2004 Thermal Analysis Engineer Applied Thermal Technologies, Pune, India, Sept 2002 July 2003 AWARDS Guru Gobind Singh Fellowship, 2008-09 Best paper award at the annual congress of ASME, Seattle, Nov 2007 Merit Fellowship of Department of Mechanical Engineering, UCSB, Sept 2003 Merit Certificate for outstanding GPA, IIT Delhi, 2001 Silver medal for 3rd rank in the Rajasthan Senior Secondary examination, 1997 Silver medal for 8th rank in the Rajasthan Secondary examination, 1995 JOURNAL PUBLICATIONS Impact of Surface Conduction on Induced Charge Electroosmosis (in preparation) Nonlinear Effects in Induced Charge Electroosmosis (in preparation) A Titanium Micro and Nano Structure Based Flat Heat Pipe (in preparation) A new wetting material based on titanium micro and nano structures (in preparation) REFEREED CONFERENCE PROCEEDINGS C. Ding, G. Soni, P. Bozorgi, B. Piorek, C. D. Meinhart, N. C. MacDonald, A Titanium Based Flat Heat Pipe, Proceedings of IMECE2008, Paper number: IMECE2008-68967, ASME International Mechanical Engineering Congress and Exposition, October 31-November 6, 2008, Boston, MA, USA G. Soni, T. M. Squires, C. D. Meinhart, Simulation of highly nonlinear electrokinetics using a weak formulation, Proceedings of Comsol User Conference, October 9-11, 2008, Boston, MA, USA

vii

G. Soni, T. M. Squires and C. D. Meinhart, Nonlinear phenomena in induced charge electroosmosis, Proceedings of IMECE 2007, American Society of Mechanical Engineers, November 11-15, 2007, Seattle, WA (best paper award). G. Soni, T. M. Squires and C. D. Meinhart, Nonlinear phenomena in induced charge electroosmosis, A numerical and experimental investigation, Proceedings of MicroTAS 2007, The 11th International Conference on Miniaturized Systems for Chemistry and Life Sciences, October 7-11, 2007, Paris, France. G. Soni, T. M. Squires and C. D. Meinhart, Study of nonlinear effects in electrokinetics, Proceedings of Comsol Users Conference, October 4-6, 2007, Boston, MA M. S. Pommer, A. R. Kiehl, G. Soni, N. S. Dakessian and C. D. Meinhart, A 3D3C micro-PIV method, Proceedings of NEMS 2007, 2nd IEEE International Conference on Nano/Micro Engineered and Molecular Systems, January 16-19, 2007, Bangkok, Thailand. J. Wu, G. Soni, D. Wang and C. D. Meinhart, AC electrokinetic pumps for micro/nanofluidics, Proceedings of IMECE 2004, American Society of Mechanical Engineers, November 13-19, 2004, Anaheim, CA.

viii

ABSTRACT

Nonlinear Phenomena in Induced Charge Electroosmosis

by

Gaurav Soni

Induced charge electroosmosis (ICEO) refers to production of electroosmotic slip by way of induced charges. Unlike fixed-charge-zeta potentials, the induced zeta potentials are proportional to the applied electric field strength which can be very strong in microfluidic devices. As a result, the induced zeta potentials are generally much higher than the thermal voltage ( > kT ze ). The linear theory of electrokinetics which is derived under the Debye-Huckel limit (
kT ze ) breaks

down for such large induced zeta potentials and predicts unrealistically high magnitudes of ICEO slip velocities. Moreover, many flow characteristics observed in experiments can not be explained by the linear theory. A vast discrepancy between the linear theory and the experiments creates the need for an investigation of the effects which take place at high zeta potentials (called nonlinear effects). This dissertation investigates some of these nonlinear effects by the means of experiments and numerical simulations. An effort has been made to reduce the

ix

discrepancy between the theory and the experiments and to explain the previously unexplained experimental flow characteristics. Induced charge electroosmotic flow was produced on a planar microelectrode with an AC electric field. The experimental velocity was found to be 2 orders of magnitude lower than the predictions of the linear theory. It was also found that the slip velocity saturates at high applied voltages, a feature not predicted by the linear theory. A nonlinear electrokinetic model was formulated with the intent of explaining the experiments. The nonlinear model is more advanced than the linear model. It solves for the surface conduction of ions through the diffuse layer and also models the double layer as a nonlinear capacitor which requires exponentially large amounts of charge to get charged. Surface conduction refers to excess ionic currents through the diffuse layer. We show that surface conduction through a nanoscale diffuse layer can cause micron scale gradients in the bulk electric field and cause severe reduction in the tangential electric field. We show that these nonlinear effects deteriorate the slip velocity. We are able to reduce the discrepancy between the theory and the experiments by one order of magnitude. Finally ICEO flow is produced on a rough surface with nanoscale roughness. We demonstrate that the roughness of a surface can have dramatic effects on the flow velocity. These effects are explained with the help of fundamental aspects of surface conduction.

TABLE OF CONTENTS

Chapter 1. Introduction ........................................................................................... 1 1.1 1.2 1.3 Applications of Electrokinetics ................................................... 1 Motivation ................................................................................... 3 Outline of Dissertation ................................................................ 7

Chapter 2. Theory of Electrokinetics .................................................................... 10 2.1 2.2 2.3 Introduction ............................................................................... 10 Electrokinetic Equations ........................................................... 15 Theory of a Thin Double Layer: Poisson-Boltzmann Equation 16 2.3.1 2.3.2 2.4 The Linear or Debye-Huckel Theory ................................ 19 The Nonlinear or Gouy-Chapman Theory ........................ 20

Capacitance of the Double Layer .............................................. 22 2.4.1 2.4.2 Linear Capacitance............................................................ 23 Nonlinear Capacitance ...................................................... 23

2.5 2.6

Helmholtz Smoluchowski Formula........................................... 24 Surface conduction and Dukhin Number .................................. 25 2.6.1 Stern Layer Dukhin Number ............................................. 30

Chapter 3. Induced Charge Electroosmosis: Experiments .................................... 33 3.1 3.2 3.3 3.4 Induced Charge Electroosmosis ................................................ 33 Experimental Evidence for ICEO Flow .................................... 37 Device Design and Fabrication Process .................................... 38 Device Packaging...................................................................... 40 xi

3.5 3.6 3.7 3.8 3.9 3.10 3.11

Experimental Setup ................................................................... 41 Measurement of ICEO with PIV............................................. 43 Dependence on Driving Voltage and Ionic Concentration ....... 47 Dependence on Driving Frequency........................................... 49 Effects of Finite Frame Rate ..................................................... 51 Momentum Diffusion Length................................................ 54 Numerical Simulations based on Linear Theory................... 55 Boundary Conditions......................................................... 56 Boundary Conditions on the Gate Electrode..................... 57

3.9.1 3.9.2 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 3.20

Numerical Results ................................................................. 60 Breaking the Symmetry: Field Effect Flow Control (FEFC) 63 FEFC in DC Electric Fields .................................................. 65 FEFC in AC Electric Fields .................................................. 66 Experimental Evidence of FEFC in AC Electric Fields........ 67 Linear Simulation of FEFC ................................................... 72 Uncertainty Analysis ............................................................. 74 Conclusions ........................................................................... 77 Appendix to Chapter 3 .......................................................... 79

Microelectrode Fabrication with Image Reversal Lithography .... 79 Chapter 4. Surface Conduction in Induced Charge Electroosmosis ..................... 81 4.1 4.2 4.3 Effects at High Zeta Potentials.................................................. 83 A Fundamental Picture.............................................................. 84 Bulk Equations .......................................................................... 90 xii

4.3.1 4.4

Bulk Boundary Conditions................................................ 90

Charge Conservation in the Double Layer................................ 91 4.4.1 4.4.2 Double Layer Edge Conditions ......................................... 93 Convergence Issues ........................................................... 93

4.5 4.6

Dimensionless Equations .......................................................... 95 ICEO on a Flat Metal Electrode................................................ 98 4.6.1 4.6.2 4.6.3 4.6.4 4.6.5 Parameters of Study ........................................................ 101 Results and Discussion.................................................... 101 Results of Parametric Study ............................................ 103 Normalized Quantities..................................................... 105 Streamlines ...................................................................... 112

4.7

ICEO on a Metal Cylinder ...................................................... 113 4.7.1 Surface Conduction Model for a 2D Cylinder in Cartesian Coordinates.................................................................................. 116 4.7.2 4.7.3 4.7.4 4.7.5 4.7.6 Geometry and Boundary Conditions............................... 117 Electric Field Lines around Cylinder .............................. 120 Streamlines around Cylinder........................................... 122 Normalized Quantities..................................................... 124 Parametric Study ............................................................. 128

4.8 4.9 4.10

Concentration Polarization...................................................... 130 Conclusions ............................................................................. 133 Appendix to Chapter 4 ........................................................ 133 Comments on COMSOL Multiphysics Simulations ..... 133 xiii

4.10.1

4.10.2 4.10.3

Weak Form for Finite Element Solution ....................... 136 Weak Form of Double Layer PDE ................................ 137

Chapter 5. Simulations vs. Experiments ............................................................. 140 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 Numerical Model..................................................................... 140 Depth Averaging ..................................................................... 143 Uncertainty in Diffusivity Values ........................................... 146 Results ..................................................................................... 148 Simulations vs. Experiments ................................................... 153 Contribution of Dielectrophoresis........................................... 154 Contribution of Electrothermal Flow ...................................... 156 Stern Layer and High Ionic Concentrations............................ 160 Conclusions ............................................................................. 162

Chapter 6. Induced Charge Electroosmosis on a Rough Surface........................ 165 6.1 6.2 Background ............................................................................. 165 Fundamental Picture................................................................ 168 6.2.1 Thick Double Layers.......................................................... 171 6.3 Experiments............................................................................. 171 6.3.1 Details of the Device .......................................................... 171 6.3.2 Fabrication Process ......................................................... 173

6.3.3 Experimental Results.......................................................... 174 6.4 6.5 Asymmetry in Flow................................................................. 178 Conclusions ............................................................................. 179

Chapter 7. Conclusions and Future Directions ................................................... 181 xiv

7.1 7.2 7.3 7.4 7.5

Experimental ........................................................................... 181 Numerical ................................................................................ 182 Simulations vs. Experiments ................................................... 182 ICEO Flow over Rough Surfaces............................................ 183 Future Directions..................................................................... 184 7.5.1 7.5.2 7.5.3 Bulk Equations ................................................................ 184 Surface Transport Equations ........................................... 185 Boundary Conditions on the Gate Electrode................... 185

Bibliography........................................................................................................ 187

xv

xvi

Chapter 1. Introduction
This dissertation deals with nonlinear effects in electrokinetics. Electrokinetics is an important subject because it has applications in numerous areas such as micro-nano fluidics, lab-on-a-chip, cell separation, bio-detection, molecule/colloid transport, micro pumping and power generation. Electrokinetics is a very interesting subject because it falls at the intersection of seemingly unrelated subjects such as electromagnetism, biology, colloids and hydrodynamics [1-3]. The motivation behind our work comes from the potential applications of electrokinetics. Therefore, we will begin this chapter by elaborating a little more on the applications of electrokinetics. Then we will describe the motivation behind our work and present the outline of this dissertation. 1.1 Applications of Electrokinetics Electrokinetics deals with transport of ionic liquids, charged species and polarizable particles in presence of electric fields. It has roots in 19th century with the pioneering electrophoresis experiments of Reuss in 1809 [4]. Since then, electrophoresis has been of great importance to the biochemists who frequently use gel electrophoresis for DNA fractionation and sequencing [5]. Electrokinetics has been studied in the context of colloids for a long time [1, 614]. In the last two decades, however, it has found applications in micro and nanofluidic devices popularly known as lab-on-a-chip. Recent emergence of micro and nano technology has given birth to a renewed interest in the study of

electrokinetic physics at nano liter scale [15]. A lot of practical applications have been developed with electrokinetics. For example, dielectrophoresis has been established as an effective method of manipulating micron and submicron size particles [16]. Dielectrophoretic cell sorters have been widely adapted by biochemists for separating diseased cells from healthy ones [17-19]. The AC electrothermal stirring method has been used for enhancing the speeds of biochemical reactions in diffusion-limited micro channels [20, 21]. Another electrokinetic phenomenon, namely electroosmosis has captured a lot of attention in recent times. Electroosmosis refers to the flow of an ionic liquid on a charged (fixed or induced) surface in presence of electric fields. Since the flow is generated on the surface, the flow speeds do not decrease as the size of the channel is decreased. This offers a great alternative to pressure driven pumping which suffers from decreasing flow rates and increasing pressure losses as the channel size is decreased. Various types of micropumps have been developed based on DC electroosmosis [22, 23]. DC electroosmotic pumps have been employed in applications such as chromatography [24] and cooling of VLSI chips [25]. Micropumps have been developed for AC electric fields too, for example micropumping can be achieved by using asymmetric electrode designs in AC electroosmosis [26-38]. In addition to pumping, AC electroosmosis has also been used as a method of concentrating DNA in a micro-chamber [39, 40]. More recently, a general theory of Induced Charge Electroosmosis (ICEO) has been presented [41, 42]. It shows that charges induced by an electric field can lead to electroosmotic flows on neutral but polarizable surfaces. ICEO flow has been 2

observed experimentally on surfaces of various shapes [43, 44]. Various applications such as pumps and mixers can be developed by breaking the symmetry of ICEO flows [45, 46]. Another electrokinetic phenomenon called field effect has been used to develop various fluidic applications such as field effect flow control (FEFC) [4753] and nanofluidic field effect transistors [54-57]. FEFC modifies electroosmotic slip velocity on a surface by modifying its zeta potential. FEFC has been used for enhancing the velocity of dc electroosmotic flow and for controlling band dispersion in capillary electrophoretic devices [51-53]. Nanofluidic transistors have been used for controlling transport of ions and proteins in nanochannels [56, 57]. Recent effort has been put into making nanofluidic batteries which can generate electricity from pressure driven flow of ions in nanochannels [58-60]. 1.2 Motivation We are motivated to make microfluidic pumps and mixers which can transport fluids at substantial flow rates. Electroosmosis is a surface driven method of fluid transport in micro and nano channels and appears to be an ideal candidate for microscale fluid manipulation. According to Helmholtz-Smoluchowski formula, the electroosmotic slip velocity, us , is proportional to the zeta potential of the surface, , and to the applied tangential electric field, E , us = E ,

(1.1)

where is the absolute permittivity of the fluid and is the viscosity. The natural zeta potential of a glass surface is -0.1 Volt. A simple micropump can be realized by applying a tangential DC electric field to the walls of a glass capillary. However, there are a few problems with this approach: firstly, when the capillary is long, a large voltage difference (100-1000 Volt) is required to create a substantial electric field; secondly, the zeta potential of glass is fixed at a low value and therefore a large electric field is required for producing substantial flow. Thirdly, high DC voltages create electrolysis of water and pose an engineering challenge to the pump designer. AC electric fields can solve the problem of electrolysis. The problem of low zeta potentials can be solved by inducing zeta potentials which are proportional to the applied electric fields (see chapter 3). The problem of large voltages can be solved by moving the electrodes close to each other within the channel and by applying an AC field to them to avoid electrolysis. Linear theory of electrokinetics (explained in the next chapter) predicts that such devices can generate flow velocities of the order of 1-10 mm/s for very low applied potentials such as 10 Volts and 100-1000 Hz frequencies. Encouraged by these calculations, we designed a device which had microelectrodes inside a channel. We applied an AC electric field to the electrodes and generated induced charge electroosmotic flow on a metal surface. To our dismay, we found that the measured flow velocities were 2-3 orders of magnitude lower than what the linear theory predicts. In other words, we could not produce as high flow velocities as we expected them to be.

While this creates a roadblock in the development of high flow rate pumps, our understanding of the theory of electrokinetics is also challenged. Linear theory has been derived for very low zeta potentials (
kT ze = 25 mVolt , where is the

zeta potential, k is the Boltzmann constant, T is the absolute temperature, z is the valence of the ions and e is the elementary charge) and works fine in that regime. For large zeta potentials, however, it fails and predicts unrealistic results. In induced charge electroosmosis, the zeta potentials are proportional to the applied field and therefore the zeta potentials can be 10-100 times higher than the thermal voltage, kT ze . We do expect the linear theory to break down at such large zeta potentials. The discrepancy between the linear theory of electrokinetics and experimental measurements is not new. Many experimentalists have indicated that the linear theory predicts much higher flow velocities than can be observed in the experiments [31, 44]. Apart from the magnitude discrepancy, various experimental features can not be explained by the linear theory either. For example, linear theory predicts that the slip velocity scales with the square of the applied voltage, 02 . In the experiments, however, velocity tends to saturate at high voltages [36, 44]. As another example, experimentalists have observed that the direction of asymmetric AC electroosmotic pumping mysteriously reverses direction at high frequencies [35] but linear theory does not predict so. Yet another example pertains to the observed dependence of AC electroosmotic velocity on the salt concentration. Experiments show that the velocity decreases as the salt concentration increases

[28] but the linear theory predicts no such effect. Linear theory fails to predict all these mysterious features observed in the experiments. All these problems with the linear theory of electrokinetics motivate us to develop an understanding of the phenomena which may take place at large zeta potentials and which are not incorporated in the linear theory. There are several effects which may take place at large zeta potentials. Following are a few of them: nonlinear capacitance of the double layer, surface conduction [1], faradaic reactions [2, 38], chemi-osmosis [61, 62], steric interactions between ions in the crowded environment of the double layer [63], increase in the viscosity of the double layer due to ion condensation [64], etc. In fact, study of these effects has yielded some success in predicting the experimental features. For example, the study of steric effects has been shown to explain the flow reversal in asymmetric pumping [64]. Similarly, the increase in the viscosity of the double layer has been shown to explain the concentration dependence of the velocity [64]. In this dissertation, we measure the velocity of induced charge electroosmotic flow at large zeta potentials. We show some surprising features in the experiments which can not be explained by the linear theory. We then try to explain those surprising features with the help of nonlinear simulations. We have devoted our attention to two nonlinear effects, namely surface conduction and nonlinear capacitance of the double layer. We have developed a fundamental picture (both qualitative and quantitative) of surface conduction in ICEO flows. We show some really surprising features of surface conduction which occur only in nonuniform ICEO flows. We use this fundamental picture to explain some of our experimental 6

features. We also show that these nonlinear effects indeed help in reducing the magnitude discrepancy between the theory and the experiments. 1.3 Outline of Dissertation Chapter 2 presents the basic theory of electrokinetics. It introduces the equations which govern the transport of ions and fluids in presence of electric fields. These equations are then simplified to derive the Boltzmann distribution of ionic species and distributions of potential and velocity inside a thin double layer. Important concepts such as nonlinear capacitance of the double layer and surface conductance are also introduced. Chapter 3 presents the results of our experiments on induced charge electroosmosis (ICEO). We demonstrate the dependence of velocity on various experimental parameters such as salt concentration, driving voltage and frequency. On a practical note, we use field effect to break the symmetry of ICEO flows and produce micro pumping. We also perform linear simulations and show that the numerical velocities are 2-3 orders of magnitude higher than the experimental velocities. Chapter 4 deals with nonlinear electrokinetic simulations. We develop a model to simulate surface conduction in steady and time dependents cases. We also incorporate nonlinear capacitance in our model. Via our simulations, we develop a fundamental picture of surface conduction in ICEO flows. We show some surprising consequences of surface conduction in ICEO flows.

In chapter 5, we compare the results of our simulations with the experiments. We show that the nonlinear effects reduce the discrepancy between the theory and the experiments. In chapter 6, we present the results of our experiments with rough surfaces. We show that the roughness of a surface has dramatic impact on the electroosmotic flow around the surface. We present experimental data to support our claim and give arguments to explain the results of the experiments. Chapter 7 concludes the dissertation. Finally, the Bibliography contains the references cited in the dissertation.

Chapter 2. Theory of Electrokinetics

In this chapter, we present the theory of electrokinetics. Most attention has been paid to electroosmosis and the theory of double layer. The electric double layer is an essential part of any electrokinetic phenomenon, be it electroosmosis or dielectrophoresis. We, therefore, present the equations (also called Nernst-Planck Equations) which govern the transport of ionic species in the presence of electric fields. These equations have been simplified to derive the Boltzmann distribution of ionic species. We also apply the Debye-Huckel approximation to derive the potential distribution and the flow profile within the double layer. An expression for the Helmholtz Smoluchowski slip velocity has been derived. Some other useful concepts such as the capacitance of the double layer and the surface conductance have also been introduced. These concepts are of great importance while studying the charging dynamics of the double layer and will be used intensively in all the succeeding chapters. 2.1 Introduction Electrokinetics deals with the motion of suspended particles and suspending fluids in presence of electric fields. An example of electrokinetic transport is electrophoresis [3]. Electrophoresis refers to the motion of a charged particle suspended in an ionic liquid and subjected to an external electric field. The electric field causes the particle to move with an electrophoretic velocity u proportional to its electrophoretic mobility, e , and to the electric field strength, E,

10

u = e E .

(2.1)

The electrophoretic mobility e is related to the zeta potential of the particle surface, , and can be expressed as the following in a thin electric double layer limit,

e = ,

(2.2)

where is the absolute permittivity of the suspending fluid and is the viscosity. The definition of will be clear to the reader when the concept of an electric double layer is introduced in a following section. Electrophoresis is of great importance to the biochemist who frequently uses gel electrophoresis for DNA fractionation and sequencing [5]. An example of DNA fractionation by gel electrophoresis is shown in Fig. 2.1.

Fig. 2.1: Gel electrophoretic separation of DNA fragments based on their electrophoretic mobility which is inversely proportional to the length of the fragment (in base pair, shown on the vertical axis). This image is reprinted from http://www.mun.ca/biology/scarr/Lab_5_gel_1999.gif for demonstration purposes and is not part of the original work contained in this dissertation.

11

Another example of electrokinetics deals with the creation of the electric double layer over a charged surface. When a charged surface (such as the surface of the particle discussed above) comes in contact with an ionic liquid, counter ions of the ionic liquid are attracted towards the surface and a diffuse charge cloud is formed on the surface. This charge cloud is called electric double layer [1, 2]. When an external tangential field is applied, the charge in the double layer feels a net Coulomb force. This force acts only on the double layer and not on the bulk fluid because the bulk is electroneutral. As a result of the Coulomb force, the double layer moves relative to the charged surface and creates a fluid motion in the bulk fluid due to viscous diffusion of momentum. This phenomenon is referred to as electroosmosis. The famous Helmholtz Smoluchowski equation for

electroosmotic slip velocity, us , is given as us = E .

(2.3)

This expression is very similar to the expression for the electrophoretic velocity (2.1) of the suspended particle discussed above. When the particle is freely suspended, the electric field causes the particle to move with the electrophoretic velocity. If the particle is fixed in position, an electroosmotic flow is developed around it with the same magnitude of slip velocity as electrophoresis but in the opposite direction. Electroosmosis can be produced not only on surfaces with fixed charge but also on surfaces which do not have any charge. The surprising phenomenon of electroosmosis over uncharged surfaces has been termed as Induced Charge

12

Electroosmosis (ICEO) [41-44]. In ICEO, the charge on the surface is induced due to polarization. Since the charge is induced and moved by the same electric field, it is a nonlinear phenomenon unlike the linear phenomenon of fixed charge electroosmosis. The nonlinearity of this phenomenon produces steady flows in AC fields unlike fixed charge electroosmosis which produces zero time averaged flow in AC fields. Infact, a steady ICEO flow in AC fields was first discovered on a pair of closely spaced coplanar electrodes and was labeled as AC Electroosmosis. It was studied in detail by [26, 28, 29, 31]. AC electroosmosis has been used as a method of concentrating DNA in a micro-concentrator [39, 40]. AC electroosmosis has also been shown as a mechanism for micro pumping by introducing asymmetry in the electrode design [32-38]. Later on a general theory of ICEO on any polarizable surface was given by [41]. There are two other phenomena which take place only in nonuniform electric fields, namely 1) Dielectrophoresis, and 2) Electrothermal flow. Dielectrophoresis (DEP) is a force which acts on the suspended particles in presence of non-uniform electric fields [26]. DEP arises due to the difference in the polarizability of the particle and the fluid medium. DEP can be used to move very small particles in a controlled manner. Recent work in this field has established DEP as an effective method of manipulating micron and submicron size particles [16].

Dielectrophoresis has been used widely for microfluidic cell separation, concentration and bio-analysis [18, 19]. Electrothermal flow defines the motion of the suspending medium in the presence of non-uniform electric fields [26]. Electrothermal flow arises due to 13

gradients in the fluids electric properties (namely, the permittivity and the conductivity). Non-uniform electric fields lead to non-uniform Joule heating of the fluid and generate temperature gradients. Electric properties like permittivity and conductivity depend on the temperature and therefore temperature gradients cause gradients in these properties too. These gradients combined with the electric field lead to an electrothermal force on the fluid and cause the electrothermal flow. One important observation of the electrothermal flow is that it creates viscous drag on the suspended particles. At the micro scale, this viscous drag can be of equal or even more importance than the DEP force. Therefore, the study of electrothermal flow is quite important. Recently, electrothermal flow has been used to enhance the binding reaction between an immobilized ligand and an antigen in a microfluidic immuno-assay [20, 21]. This opens a lot of scope for the use of electrokinetics in biological sciences where high yield and throughput are highly desired. In this thesis we will not pursue the subjects of dielectrophoresis or electrothermal phenomena. Instead, electroosmosis will be the main subject. Study of electroosmosis starts with certain electrokinetic equations which govern the transport of ionic species in a liquid medium in presence of electric fields. In a very thin double layer limit, these equations can be used to derive distributions of concentration, electrostatic potential and flow velocity within the double layer. The canonical case of fixed charge electroosmosis (also called DC electroosmosis) will be discussed. Important concepts (such as double layer capacitance and surface conduction) will be established so that they can be used for the detailed study of induced-charge electroosmosis in the succeeding chapters. 14

2.2

Electrokinetic Equations We describe the relevant equations for transport of various ionic species in a

dilute electrolyte solution in the presence of electric fields. These equations are also called the Nernst-Planck equations. For simplicity, we present the equations for a symmetric electrolyte (z:z), made of ions of valence z (e.g. KCl). The electrostatic potential obeys Poissons equation:

2 =

(n+ n ) ze

(2.4)

where n+ and n are the number densities of positive and negative ions respectively, e is the elementary charge and is the absolute permittivity of water which is approximately equal to 80.1 0 at 200 C, where 0 is the permittivity of vacuum. For some symmetric electrolytes, the positive and negative ions have equal diffusivities and ionic mobilities. For example, the difference in the diffusivities of K+ and Cl- ions is less than 4% with Dk + = 1.96 109 and DCl = 2.03 109 m2/s [65]. With these simplifications, the number densities of ions obey the following conservation equations n + .J = 0 , t where the ion current densities, J , are given by J = n Dn + n u , (2.6) (2.5)

15

where u is the local fluid velocity vector, and and D are the mobility and diffusivity of the ions, respectively. These two quantities are related to each other as = zeD / kT [1]. The three terms on the right hand side of (2.6) represent the flux of ionic species due to a Coulomb force (same as Ohms law,
ohm J = / ze , with the conductivity = n ( ze) 2 D / kT ), diffusion due to

concentration gradient and advection by the fluid flow respectively. The fluid flow is governed by Navier-Stokes equations:

u u = p + 2u ze(n+ n ) ,
u = 0 ,

(2.7) (2.8)

where the third term on right hand side of (2.7) represents the Coulomb body force on the fluid.
2.3 Theory of a Thin Double Layer: Poisson-Boltzmann Equation

Here we present a theoretical framework for a thin double layer ( D

a,

where a is the radius of curvature of the surface). Consider an ionic solution which has equal densities of positive and negative charges. The bulk of this solution will be electrically neutral due to equal ion densities. When a charged surface comes in contact with such ionic solution, it attracts counter ions of the solution and repels co-ions. This creates a narrow region of non-zero charge density close to the surface. This narrow charged region is referred to as the electric double layer or Debye layer (see Fig. 2.2). The thickness of the double layer is a function of the ionic strength of the solution and is generally very small (1-100 nm). The electric

16

potential caused by the surface charge is the highest at the surface and zero in the bulk. The drop in the potential is caused by the double layer charge. Therefore, the double layer can be said to act as a capacitor which screens the surface charge from the bulk of the solution.

us
+

Plane shear Plane of of Slip Stern Layer

+ + + + Diffuse Layer + + D + +++ + + + + + ++ + + + - + + ++ + ++ + + + + + + + + + + + + + + + + + + + + + + + + + + + ++ ++ + + + + - - - - - - - - - - - -q - - - - - - - - - - +

+ +

y x

Fig. 2.2: Electrical double layer (EDL) formation on a charged surface. EDL is much thinner than the bulk. The potential drops exponentially through the charge cloud. The potential difference across the diffuse part of the EDL is called the zeta potential of the surface. A tangential electric field can drive the charge cloud with a velocity that grows exponentially in the double layer. Far field velocity is called the electroosmotic slip velocity.

The double layer is viewed as consisting of two layers, an inner or Stern layer and an outer or diffuse layer. The Stern layer consists of those ions which are closest to the charged surface and the fluid in this region experiences a no-slip condition. As a result, the Stern layer does not contribute to the electroosmotic flow. The diffuse layer resides on top of the Stern layer and ions in this layer are mobile. The interface between the two layers is called the plane of shear. The net

17

potential drop across the diffuse layer is defined as the zeta potential of the surface. Lets consider an infinite flat surface with a fixed charge density - q (per unit area) (Fig. 2.2). Lets assume that the double layer is much thinner than the dimensions of the surface ( D
a ). If we assume that the fluid is static (i.e. u=0),

then combining (2.5) and (2.6), we obtain a balance between the electromigration and diffusive fluxes of the ions close to the surface,
D dn d = n . dy dy

(2.9)

The boundary conditions away from the surface, i.e. y , can be written as,

( y ) = 0 ,
and
n ( y ) = n0 ,

(2.10)

(2.11)

where n0 is the number density of ions in the bulk. With the help of boundary conditions (2.10) and (2.11), (2.9) can be integrated to give the distribution of ionic concentration in the double layer, also known as the Boltzmann distribution,
n = n0 exp( ze ). kT (2.12)

Now combining (2.4) and (2.12) yields the famous Poisson-Boltzmann equation,

18

d 2 2n0 ze ze = sinh( ). 2 dy kT

(2.13)

which can be integrated to determine the potential distribution in the double layer. The results of the integration are shown in the next two subsections.
2.3.1 The Linear or Debye-Huckel Theory

In the electrokinetic literature, a Debye-Huckel approximation is used for linearizing the Poisson-Boltzmann equation. This approximation states that when the zeta potential is much smaller than the thermal voltage ( nonlinear terms in the electrokinetic equations can be linearized, i.e., sinh( ze ze ) for kT kT kT ze . (2.14)
kT ze ), the

Under the Debye-Huckel assumption, (2.13) can be linearlized to, d 2 = 2, dy 2 D where (2.15)

D =

kT
2n0 ( ze) 2

(2.16)

is the approximate double layer thickness. Note that D decreases with the salt concentration, n0 . The double layer charge counterbalances the charge on the surface. At higher ionic concentrations, a thinner region of ionic charge is sufficient to counterbalance the charge on the surface. Following are the boundary conditions for the integration of (2.15):

19

 ( y = 0) = s ,
and

(2.17)

( y ) = 0 ,

(2.18)

where s is the potential at the surface. A simple integration of (2.15) then yields the potential distribution in the double layer

= s exp(

).

(2.19)

If we assume that the potential at the surface, i.e. s , is not much different from the potential at the plane of slip, i.e. , then

exp(

).

(2.20)

The preceding analysis shows that, in a thin double layer limit and under the Debye Huckel assumption, the potential drops exponentially across the double layer (also shown in Fig. 2.2).
2.3.2 The Nonlinear or Gouy-Chapman Theory

Under the Debye-Huckel approximation, we obtained an exponential decay of potential in the double layer. This is a canonical result and gives us a great insight into the double layer potential distribution. This approximation works well for many surfaces of interests. For example, the zeta potential of a glass surface (in contact with water of pH<4.0) is about -100 mVolt which is only four times higher than the thermal voltage at room temperature (~25 mVolt).

20

In many induced charge electroosmotic situations, however, the zeta potential is caused by the electric field and not by a chemical change. In such cases, the induced zeta potentials can be much higher than the thermal voltage and thus the Debye-Huckel approximation does not hold true. Infact, linear theory predicts much higher slip velocities than can be observed in ICEO experiments [44]. Gouy-Chapman offered the integration of Poisson-Boltzmann equation without linearization. The nonlinear integration was done by noting that, d 2 1 d d = , dy 2 2 d dy and hence (2.13) becomes, d 4n0 ze ze d sinh( )d . = kT dy Integrating the preceding equation yields, d 4n0 kT ze cosh( ) + constant . = kT dy
2 2 2

(2.21)

(2.22)

(2.23)

The constant of integration can be evaluated by noting that at distances far from the surface, = 0 and ( d dy = 0 ) . As a result, d 4n0 kT = dy
2

ze cosh( kT ) 1 .

(2.24)

The preceding equation can be further simplified to

d ze 8n kT = 0 sinh( ), dy 2kT
12

(2.25)

21

which can then be integrated in the following manner to find the potential distribution in the double layer,
8n kT d sinh( ze 2kT ) = 0w

12 y

dy ,
0

(2.26)

where is taken to be the potential at y = 0 (approximately). The result of integration is the following: 2kT tanh( ze 4kT ) 8kTn0 ln = ze tanh( ze 4kT ) or
12

y,

(2.27)

4kT ze tanh 1 tanh ze 4kT

y D e .

(2.28)

2.4

Capacitance of the Double Layer

According to Helmholtz model of double layer [1], the electrode surface and the double layer can be seen together as two plates of a parallel plate capacitor. The charge residing in the double layer counterbalances the charge on the electrode surface. We can define a capacitance for the double layer. This capacitance gives us a way to relate the amount of charge in the double layer to the voltage drop across it. In the following section we present expressions for the capacitance of the diffuse layer. The capacitance of the Stern layer is unknown because the ionic properties such as diffusivity and mobility in the Stern layer may not be the same as the bulk.

22

2.4.1 Linear Capacitance

Under the Debye Huckel approximation, a capacitive relation for the diffuse layer can be expressed as follows q = Cd , (2.29)

where q is the double layer charge per unit area and Cd is the diffuse layer capacitance per unit area. According to Gauss's law, the charge density per unit area of an infinite sheet of charge is related to the normal electric field in the following manner,
q = En = d dy

.
y =0

(2.30)

Under the linear assumption, d dy y =0 = D (see (2.20)). As a result,

q=

. D

(2.31)

Comparing the two expressions for q , i.e. (2.29) and (2.31), we get an expression for the linear capacitance of the diffuse layer

Cd =

. D

(2.32)

This result is analogous to the parallel plate capacitance with a plate separation of D .
2.4.2 Nonlinear Capacitance

When the zeta potential is higher than the thermal voltage, the preceding linear capacitance expression becomes inadequate. In such circumstances, we need to use

23

the results of the Gouy-Chapman theory. We can get the following nonlinear expression easily by combining (2.25) and (2.30),

q = (8 kTn0 )1 2 sinh(
which can be simplified to

ze ). 2kT

(2.33)

q=

ze 2kT sinh D ze 2kT

(2.34)

A nonlinear differential capacitance of the diffuse layer can then be defined in the following manner [2]
Cd = dq ze = cosh( ). d D 2kT

(2.35)

Note that, for

2kT ze , the preceding equation yields Cd = D which is

equivalent to the linear capacitance formula given by (2.32).

2.5 Helmholtz Smoluchowski Formula

Lets consider a surface with a fixed charge density i.e. a fixed zeta potential. Lets subject the surface to a constant tangential electric field E . If there is no pressure gradient in the system, the fluid flow equation for a parallel flow can be simplified to,

d 2u d 2 = 2 E . dy 2 dy

(2.36)

Then using u (0) = 0, (0) = , u y = 0 and y = 0 for a thin double layer, the preceding equation can be integrated to yield

24

u=

E 1

(2.37)

Under Debye-Huckel approximation, can be substituted from (2.20) to yield

u=

y E [1 exp( )] . D

(2.38)

Then, the electroosmotic slip velocity, us , at the edge of the double layer (i.e.,

y ) can be written as,

us =

E .

(2.39)

This is the well known Helmholtz-Smoluchowski equation for electroosmotic slip velocity. This equation was derived for a thin double layer under DebyeHuckel approximation ( ze / kT 1 ), however, it works very well up to

ze / kT 2 according to [66]. According to Squires & Bazant 2004 [41], this

equation remains valid in nonlinear regime as long as

D
a

exp(

ze ) 2kT

1,

(2.40)

where a is the radius of curvature of the surface.

2.6 Surface conduction and Dukhin Number

In the previous sections, we looked at the mechanism by which a tangential flow of ions (and the fluid) is generated (i.e. electroosmosis). A solution was also obtained which satisfied all the equations and boundary conditions. However, the flow of ions tangential to the electrode surface creates a current which has not been

25

accounted for yet in the mathematical model. The ions can move along the electrode surface not only by convection but also by conduction due to the tangential component of the electric field (see Fig. 2.3).

Bulk

----+ + + + +

- + + - + +
s

j=E js=sE

Fig. 2.3: Surface conduction

Usually, the net ionic current in the body of the fluid can be approximated by

j = E . However, when the zeta potentials are large, a lot of charge can reside in
the double layer and the conductivity in the double layer might be much higher than the bulk. Therefore the approximation ( j = E ) yields a lower value for the current. The excess current caused by the excess conductivity of the double layer is referred to as surface conduction (see Fig. 2.3). Surface conduction has been studied in the colloidal science for a long time [6-8]. Surface current is very important to estimate because excessive amount of surface current can affect the electric field in the bulk and deteriorate the

26

electroosmotic slip velocity (refer to chapter 4). Surface conduction becomes important in cases where zeta potentials are large (such as induced charge electroosmosis in which the zeta potential is induced by an applied electric field). The excess current (or surface current) can be characterized by a excess surface conductivity, s ,

js = s E ,

(2.41)

where js is the excess surface current density per unit width in C s-1 m-1 = A m-1,

s is the excess surface conductivity in A V-1 = S and E is the electric field

causing the current. If y is the direction normal to the electrode surface, then it has been shown that [1]

js = [ j ( y ) j ()]dy ,
0

(2.42)

where j ( y ) is the bulk current density in C s-1 m-2 = A m-2 . Bikerman, 1940 [6] realized that besides conduction, the charge also gets advected with the electroosmotic flow. Including the electroosmotic advection in (2.42) and expressing j ( y ) and j () in terms of ionic concentrations and mobilities, we get j ( y ) j () = i [ni ( y ) ni ()] zi ei E i zi eni ( y ) [ ( y )]E , (2.43) where, the first term is the current density due to conduction of charge relative to the fluid and the second term is the current density due to the electroosmotic flow. Here ni is the number density of the ith ionic species, zi is the valence, e is the

27

charge of an electron, i is the ionic mobility, is the zeta potential and ( y ) is the potential distribution in the double layer. The ionic mobility, i , can also be expressed in terms of diffusivity using the Nernst-Einstein Equation,

i =

zi eDi . kT

(2.44)

Substituting for i in (2.43) from (2.44) yields,

j ( y ) j ( ) = e2 kT

[n ( y) n ()] z
i i i

2 i

Di E

e [ ( y )] i zi ni ( y )E .(2.45)

Substituting this expression in (2.42) and taking into account that all concentrations follow from Gouy-Chapman theory, Bikerman derived for a symmetric electrolyte,

s =
where,

2 z 2 e 2 nD D+ ( e ze kT

2 kT

1) (1 + m+ ) + D ( e ze

2 kT

1) (1 + m ) , (2.46)

kT 2 m = ze D

(2.47)

is a dimensionless parameter indicating the relative contribution of electroosmosis to surface conduction. When the cations and the anions have similar diffusivities (i.e. D+ = D ), (2.46) can be simplified to

s =

4 z 2 e 2 nDD ze (1 + m ) cosh kT 2kT

1 .

(2.48)

28

For aqueous solutions of a binary electrolyte (such as KCl), m 0.45 at room temperature and therefore the contribution of electroosmosis is lower than conduction. Usually, the surface conductivity is expressed in terms of a dimensionless parameter called Dukhin number, Du , defined as. Du =

s , a

(2.49)

where is the bulk conductivity and a is the characteristic length scale. The Dukhin number is the dimensionless ratio of the surface and the bulk conductivities. This dimensionless ratio has been used in the electrokinetic literature for a long time and has been used for discriminating between various regimes of electrokinetics [1]. The bulk conductivity can also be expressed in terms of ionic mobility and concentration as

= 2zen ,
which, using = zeD kT , can be further simplified to 2z 2 e 2 nD = . kT Now combining (2.48), (2.49) and (2.51) yields Du = which can also be written as 2D ze (1 + m ) cosh a 2kT 1 ,

(2.50)

(2.51)

(2.52)

29

Du =

4D ze (1 + m ) sinh 2 . a 4kT

(2.53)

2.6.1 Stern Layer Dukhin Number

The previous discussions of surface conduction and the Dukhin number were solely about excess surface currents in the diffuse part of the double layer. It was easy to estimate the current density and Dukhin number in the diffuse part of the double layer because the ionic mobility and diffusivity in the diffuse layer can be considered to be the same as the bulk. However, the current density in the Stern layer is hard to estimate because the diffusivity and mobility in the Stern layer are unknown. The Stern layer surface conductivity, si is generally expressed as

si = i ii ii ,

(2.54)

where i in the superscript represents inner or Stern layer whereas the i in the subscript represents the ith ionic species. Here ii is the surface charge density in the Stern layer. Often there is only one ionic species (denoted by i) in the Stern layer and therefore,

si = ii ii =

zi eDii ii . kT

(2.55)

Then using the definition of the bulk conductivity i = 2 zi 2 e 2 ni Di kT , we can express the Stern layer Dukhin number as follows, Du =
i

ii Dii
2azi eni Di

(2.56)

30

The methods of estimation of Stern layer Dukhin number have been described in [1]. Assuming an adsorption model, ii can be related to ni and the specific Gibbs binding energy ads Gmi . Making Gmi more negative increases ii but does not necessarily enhance si because tighter bound ions may have a lower lateral mobility. In some cases, the Stern layer surface current can be stronger than the diffuse current. This is expected for porous surfaces, containing thick hydrodynamically stagnant layers with mobile ions in them, as is the case with porous glass and bacterial cells.

31

32

Chapter 3. Induced Charge Electroosmosis: Experiments

We present an experimental evidence for the existence of induced charge electroosmosis on a flat microelectrode. The experimental data contains basic features of symmetric ICEO flow over a metallic surface. We also present an experimental method called field effect flow control (FEFC) to break the symmetry of the symmetric ICEO flows. FEFC can be used for making micropumps whose pumping velocity as well as the direction of pumping can be modified by applying a radial voltage to the flow surface. The existence of both ICEO and FEFC has been demonstrated through experiments. Finite element simulations based on linear theory (i.e. Debye Huckel approximation) have been performed to verify the results of the experiments. Not surprisingly, the results of the simulations do not match with the experiments; the simulations predict almost 2-3 orders of magnitude higher slip velocities. The applied voltages in the experiments are high and they induce high zeta potentials ( 10 40 kT ze ). We expect the linear theory (and the Debye Huckel approximation) to fail at such high zeta potentials. In chapter 4 and 5, we will introduce a nonlinear model which reduces the numerical velocity and reduces the discrepancy between the numerical simulations and the experiments.
3.1 Induced Charge Electroosmosis

ICEO refers to a phenomenon in which a DC or AC electric field induces charge on a polarizable surface (metal or dielectric), and produces an

33

electroosmotic slip by applying a body force on the electric double layer [41, 42]. Since the double layer is created and moved by the same electric field, this phenomenon gives rise to steady flows in both DC and AC electric fields. Flat electrodes are of much interest in microfluidic devices. Consider a finite flat conductor surface in contact with an electrolytic solution. When it is subjected
to an external electric field, E = E0 x , at t = 0 , the electric field lines intersect the

surface at right angles and a charge density is induced on the surface because of charge separation (Fig. 3.1a). However, the field lines start changing their configuration as a current J = E drives positive ions towards one half of the surface ( x < 0 ) and negative ions to the other half ( x > 0 ). This process develops a double layer on the surface which grows as long as the normal electric field drives ions into it. In the steady state, assuming that there is no surface conduction or Faradaic injection, the double layer insulates the surface completely and no electric field lines can penetrate into it. In this state, all the electric field lines are tangential to the surface (Fig. 3.1b) and cause an electroosmotic slip directed from the edges toward the center giving rise to two symmetric rolls above the surface (Fig. 3.1c). An AC field will drive an identical flow as the change in the direction of the field changes the polarity of the induced charge as well.

34

E0

Conductor

- - - - - - - - - - - +++++++++++ x
(a) Electric field at t=0,

E0

++++ ++ + +++ + - - - - - - - - - - Double layer

- - - - - - - - - - - +++++++++++ x
(b) Steady State electric field

Conductor

(a) Steady ICEO Flow

Fig. 3.1: Induced charge electroosmosis on a conducting surface. (a) At time t=0, there is no double layer, hence all the field lines are perpendicular to the surface (shown by electric field lines), (b) In the steady state, the double layer gets completely charged and all the field lines become tangential to the surface, (c) The tangential field causes the double layer to move and produces symmetric ICEO flow (shown by streamlines).

35

The slip velocity us is given by the Helmholtz Smoluchowski equation, us = E ,

(3.1)

where and are respectively the absolute permittivity and viscosity of the ionic solution, is zeta potential of the surface and E is the electric field component tangential to the surface. is defined as the potential drop across the diffuse part of the double layer. In an ideal case when there is no nonlinear effect present (e.g. Stern layer, surface conduction or faradaic reactions), E = E0 and = E0 x . By substituting these in (3.1), we get us =
2 E0 x,

x .

a 2

(3.2)

where a is the length of the conducting plate. This shows that the flow is symmetric about x = 0 (i.e. the center of the surface) and the maximum slip velocity occurs at the left and right edges. The velocity depends on the square of the electric field which implies that an AC field also drives a similar flow as long as the frequency, , is low enough that the double layer has time to form,

< c1 ,
where c is the characteristic double layer charging time defined as

(3.3)

c =

a , D

(3.4)

where D is the double layer thickness, and is the electrical conductivity of the solution. Existence of c can be readily seen by considering an RC circuit with a

36

bulk resistor, R = a / , and a double layer capacitor, C = / D , [27]. The expression for c can also be obtained by a scaling analysis of (3.25). For a symmetric electrolyte for which the cations and anions have the same magnitude of valence and diffusivity,

=
and

2z 2e2 n0 D , kT

(3.5)

kT = 2 2 2z e n 0

12

(3.6)

where D is the diffusivity of ions, n0 is the numeric concentration of ions in the bulk, k is the Boltzmann constant, e is the elementary charge, and z is the valence of the ions. Combining the preceding definitions of c , and D , we get

c =
3.2

D a
D

(3.7)

Experimental Evidence for ICEO Flow

Squires and Bazant, 2004 [41] theoretically predicted ICEO flows around polarizable surfaces when such surfaces are subject to an external electric field in the presence of an electrolytic solution. Levitan et al. 2005 [43] verified the existence of ICEO around a cylindrical wire in an ionic liquid. We present experimental evidence of ICEO flow on a flat metal electrode in the following sections.

37

3.3

Device Design and Fabrication Process

ICEO experiments were performed in a microfluidic device which simply consisted of three parallel, equally spaced, coplanar electrodes laid on a glass substrate (see Fig. 3.2). Image reversal photolithography was used to transfer the electrode pattern in a photoresist film (AZ5214). Electrodes were made by depositing 10 nm of titanium and 200 nm of platinum using electron beam evaporation. Residual photoresist was removed by lift-off in acetone. The fabrication process is depicted in Fig. 3.3. A step-wise fabrication procedure is given in an appendix to this chapter.

10 mm

Driving Electrodes

Gate Electrode 20 mm

Fig. 3.2: The ICEO device. Three 200 m wide electrodes are deposited on a glass substrate. The middle electrode is called gate electrode; the outer two electrodes are called driving electrodes.

38

1) 2) 3) 4) 5) 6)

Glass Photoresist Metal PDMS

Fig. 3.3: Fabrication process for the ICEO device: (1) A glass wafer was cleaned respectively in Acetone, Iso propanol, de-ionized water (DI) and finally in O2 plasma. (2) AZ 5214 photoresist was spin coated on the wafer. (3) The device design was transferred from the mask into the photoresist by image reversal photolithography. (4) Thin layers of metals (Ti/Pt, respectively 10/200 nm) were evaporated on the wafer and (5) then lifted off in acetone. (6) A PDMS chamber was placed on the glass substrate. PDMS readily adheres to the clean glass surface. The chamber was closed by placing another glass piece on top.

The electrodes were placed in contact with a microchamber filled with an ionic solution. An AC signal was applied between the two outer electrodes (also called driving electrodes); the middle electrode was kept floating i.e. it was not connected to any power source. The floating (i.e. the middle) electrode is analogous to the polarizable surface shown in Fig. 3.1. The AC electric field produced by the two driving electrodes causes the floating electrode to get polarized and leads to formation of double layer on the floating electrode. The external field also causes the double layer to move, resulting in a symmetric ICEO flow.

39

The floating electrode is also denoted as the gate electrode. This name for the electrode is chosen because of its analogy with a solid state field effect transistor (FET). In a solid state FET, a gate voltage controls the electronic current between a source and a drain terminal. In our device, the driving electrodes are used for producing an electric field and set up a symmetric ICEO flow on the gate electrode. The symmetry of the flow can be broken by applying an external potential on the gate electrode (equivalent to field effect). A simple micropump based on AC electric fields with directional control of the flow can thus be made. We have discussed field effect in ICEO in a later section of this chapter.
3.4 Device Packaging

The

overall

dimensions

of

the

bottom

glass

substrate

were

20 mm10 mm500m (Fig. 3.4). The gap between the two driving electrodes
was 800 m (Fig. 3.5). The gate electrode was located symmetrically between the two driving electrodes. A 2 mm4 mm125m flow chamber was cut into a 125

m thick PDMS sheet and placed on the electrodes. The net volume of the flow
chamber was 1 Liter. A 1 Liter volume of a KCl solution was dropped into the chamber using a micro pipette. The ionic solution was seeded with 700 nm florescent polystyrene beads for flow tracing. A 500m thick blank glass chip was placed atop the chamber to close it. A cross section of the ICEO chamber is shown in Fig. 3.5.

40

Gate electrode

Driving electrodes ICEO flow chamber Top glass cover PDMS sheet

glass substrate 20 mm Connection pads

mm 10

Fig. 3.4: Picture of a fully packaged device. A cut PDMS sheet is sandwiched between the bottom glass substrate and a top glass cover to form a closed flow chamber.

Driving electrode 1
0.125 mm

Flow chamber

Gate electrode
y x

Driving electrode 2

0.2 mm

0.2 mm 0.8 mm 2 mm

0.2 mm

Fig. 3.5: Cross section of the flow chamber. The gap between the driving electrodes is 800 m. The flow chamber is 125 m deep.

3.5

Experimental Setup

An AC function generator produces a signal which is applied to the driving electrodes (Fig. 3.6). In order to visualize the ICEO flow, 700 nm red-fluorescent polystyrene particles (Duke scientific, Fremont, CA) are suspended in the ionic fluid. The final concentration of fluorescent particles in the working fluid is 0.02%

41

by volume. A 100 W mercury arc lamp (Optiquip, Highland Mills, NY), an epifluorescence microscope (Nikon Eclipse E600FN), an optical filter cube (excitation 532 nm, emission 612nm, Chroma, Rockingham, VT), a 10x objective lens (NA 0.25) and a CCD camera (Hamamatsu, 1280 1024 12 -bit) are the main components of the imaging system. The fluorescent particles are excited using green light (532 nm wavelength) and upon excitation, they emit red light (612 nm) which is recorded with the CCD camera.

PIV CCD Excitation filter White light Hg lamp Excitation light Objective lens ICEO device Floating electrode Focusing lens Filter cube Emitted light Fluorescent particles in KCl
glass
PDMS

glass

AC function generator

Fig. 3.6: The experimental setup for flow measurement.

42

3.6

Measurement of ICEO with PIV

The velocity of the fluid was determined by measuring the velocity of the suspended fluorescent particles which were 0.7 m in diameter. The particles close to the electrodes have high velocity (due to ICEO slip) whereas the particles close to the top-cover are slow (due to no-slip). Since the particle diameter is much larger than the double layer thickness (~1030 nm), its impossible to measure the velocity within the double layer with such large particles. Apart from this, the excitation light illuminates the entire flow field and the lens collects light from all the particles in the illuminated flow field. Therefore, what we measure is not the velocity in the focal plane; instead it is a weighted depth average of the entire flow field. However, the highest relative contribution to the depth average comes from the focal plane and the relative contribution decays sharply as the distance from the focal plane increases. For these reasons, the measured velocity (i.e. the weighted depth average) is a good approximation for the velocity in the focal plane and we will call it the slip velocity. The reader must keep in mind that its not the true value of the slip velocity but a very good approximation. For a discussion of the weighted depth averaging, the reader is referred to section 5.2. For a typical experiment, approximately 50 consecutive image frames were recorded using a PIV system [67]. A PIV program calculated the cross correlations from the sequential image pairs and averaged the correlations before producing a final velocity vector field [68]. The uncertainty in the experimental

43

data is about 3.6% (with respect to full scale) and has been discussed in section 3.18.

Fig. 3.7: A typical image of a PIV experiment. The gate electrode is visible in the center. The inner edges of the two driving electrodes are also shown. Fluorescent particles are also clearly visible.

For the symmetric ICEO flow, the velocity has a symmetric vector field. A typical vector field is shown in Fig. 3.8. The fluid moves from the edges towards the center on all three electrodes. Flow is observed to be symmetric on the gate electrode. The vector field on the gate electrode alone is also shown in Fig. 3.9 for clarity. The fluid flows symmetrically from the two edges ( x = 100 m ) towards the center of the gate ( x = 0 ). The velocity is highest close to the edges and zero at the center. At the center, the fluid moves out of the plane (not shown in this vector field). The highest velocity is observed to be 45 m/s (Fig. 3.10) for a driving voltage of 0 = 20 Vpp , and a frequency of

f = 100 Hz in purified water

( = 17 S/cm). The frequency of 100 Hz was chosen such that there is enough

44

time

for

the

double

layer

to

charge.

Taking

= 17 S/cm

and

= 80.1 8.854 1012 F/m for the purified water and D = 1.995 109 m2/s for K+
and Cl- ions, we find D = D = 28.85 nm (refer to (3.5) and (3.6)). Then taking a = 200 m, we find c = D a D = 0.0029 s (refer to (3.4)). We can now calculate a dimensionless frequency, f c = 0.29 < 1 . In a linear regime (i.e.

< kT ze ), a dimensionless frequency less than 1 will ensure that the double layer
has enough time to charge. In a nonlinear regime, however, the double layer requires an exponentially large amount of charge and the double layer might not charge completely even at very low frequencies (refer to (2.34)). The electrical conductivity of the purified water was measured to be 17 S/cm (Oakton Ectest microprocessor based conductivity meter). The purified water was actually a 50:1 mixture of deionized (DI) water and Duke Scientific fluorescent microsphere polymer suspension (Fremont, CA). DI water has very low conductivity (0.055 S/cm) and a tiny fraction (1/50th) of microsphere suspension raises its conductivity to 17 S/cm. Microspheres were added essentially for visualizing the flow. The content of the microsphere suspension is not disclosed by Duke Scientific due to proprietary issues. Therefore, we dont know what ions are present in the particle solution.

45

Fig. 3.8: A typical vector field for symmetric ICEO flow on a planar electrode.

Fig. 3.9: Velocity vectors on the gate electrode, for 0 = 20 Vpp and f = 100 Hz in purified water with = 17 S/cm.

46

5 4 3 Slip Velocity, us (m/s) 2 1 0 -1 -2 -3 -4

x 10

-5

-5 -1

-0.5 0 0.5 Distance from the center x (m)

1 x 10
-4

Fig. 3.10: Variation of velocity along the gate electrode for 0 = 20 Vpp and f = 100 Hz in purified water ( = 17 S/cm).

3.7

Dependence on Driving Voltage and Ionic Concentration

Fig. 3.11 shows the maximum slip velocity, umax , as a function of the driving voltage and the ionic concentration. The experiments were conducted for a range of driving voltage at a constant frequency of 100 Hz and in two different solutions: purified water ( = 17 S/cm) and 1 mM KCl (165 S/cm). The velocity in the purified water is significantly higher than that in 1 mM KCl solution. Many workers have also observed that the electroosmotic slip velocity decreases as the ionic concentration increases [28]. Such a reduction in slip velocity has been attributed to Stern layer. At high ionic concentrations, the double layer becomes very thin and its capacitance becomes comparable to that of the Stern layer. As a

47

result, a large part of the surface potential is dropped across the Stern layer, leaving a small drop across the diffuse layer and reducing the slip velocity [30]. The velocity scales with 02 for low values of 0 confirming the trend predicted by the linear theory. However, it becomes almost constant at high values of 0 . The saturation of velocity at high voltages can be attributed to the nonlinear effects (such as surface conduction and nonlinear surface capacitance) which become strong at high voltages and do not let the velocity grow. Its important to note that the saturation takes place at a voltage of around 15 Vpp in the purified water and at about 20 Vpp in 1 mM KCl. In other words, the flow in the purified water saturates sooner than in 1 mM KCl. This behavior can be explained by comparing the double layer thicknesses in the two solutions. The double layer is much thicker in the purified water than in 1 mM KCl. The surface currents are stronger for thicker double layers because there is more diffuse charge (see chapter 4). In other words, surface conduction is much stronger in the purified water than in 1 mM KCl. Therefore, the saturation in the purified water takes place at a lower voltage than it does in 1 mM KCl.

48

3.5 Maximum Slip Velocity, u (m/s) max 3 2.5 2 1.5 1 0.5

x 10

-5

Purified Water

1 mM KCl u =4E-72
s 0

0 0

10 15 20 25 Driving Voltage, 0 (Vpp)

30

35

Fig. 3.11: Variation of maximum slip velocity with the driving voltage. Initially the velocity scales quadratically with the voltage but eventually saturates. This saturation behavior is attributed to surface conduction. Surface conduction is stronger in the purified water than in 1 mM KCl; therefore the velocity saturates at a lower voltage in purified water than in 1 mM KCl. The measurements were performed at 100 Hz.

3.8

Dependence on Driving Frequency

Fig. 3.12 exhibits the effect of frequency on the maximum slip velocity. The velocity stays almost a constant for f<300 Hz but decreases significantly for higher frequencies. This indicates that the characteristic frequency of the double layer charging is less than 300 Hz. The experiments were performed in 1 mM KCl at 18 Vpp.

49

10 Maximum Slip Velocity, u (m/s) max

-4

10

-5

10

-6

10

-7

10

10 10 10 Driving Signal Frequency, f (Hz)

10

Fig. 3.12: Variation of maximum slip velocity with driving frequency. As the frequency increases, the velocity decreases. This is because at high frequencies, the double layer does not get enough time to get charged. The experiments were performed in 1 mM KCl at 18 Vpp.

We chose to perform our ICEO experiments in presence of AC electric fields because AC fields yield the advantage of reduced electrolysis. In presence of DC electric fields, a lot of gas bubbles are generated due to electrolysis. Apart from this, DC electric field will cause the driving electrodes to get shielded by double layer completely. In such a case, there will be no electric field in the bulk because entire potential will be dropped across the double layers on the driving electrodes. These problems motivate us to run our experiments in presence of AC electric fields. However, the frequency of the driving AC field affects the velocity. As the frequency increases, the available time for the double layer formation decreases.

50

As a result, the double layer does not charge completely and the zeta potential is reduced.
3.9 Effects of Finite Frame Rate

The frequency of frame acquisition is generally much smaller than the driving AC frequencies. For example, in our experiments, the PIV frames (i.e. images) were captured at a frame rate of 13.2 Hz, whereas the driving frequency was 100 Hz. As a result, PIV images yield a time averaged velocity field. However, since the frame acquisition is started at a random time, the frame acquisition process might have a phase lag with respect to the driving ac cycle. If the phase lag is nonzero, the time averaged slip velocity might have a contribution from the time dependent component. Lets quantify how much effect a finite frame rate has on PIV velocity calculations. In a linear regime, both and E can be assumed to vary in synch with the driving voltage. In other words, sin(2 ft ) and E sin(2 ft ) . As a result, the ICEO flow velocity can be represented as
u = 2u0 sin 2 (2 ft )

(3.8)

where u0 is the time averaged flow velocity. Note that, when integrated from t = 0 to 1/f, (3.8) yields a time averaged velocity equal to u0 . The preceding expression can now be separated into a steady and a time dependent component as follows,
u = u0 u0 cos(4 ft )

(3.9)

51

The time variation of the velocity is shown in Fig. 3.13. Lets say a frame is captured at an arbitrary time, t1, and the second frame is captured at a time t1+t, where t the time gap between the two frames. Since, the frame capturing was started at an arbitrary time, a phase lag, say 1 , occurs between the velocity variation and the frame acquisition. Since the frequency of frame acquisition is different from the ac frequency, a different phase lag, , say 2 , when the second frame is captured. The phase lag keeps changing as more and more frames are acquired. When the first two frames are cross-correlated, a time averaged velocity is obtained. The time average can be represented as follows,
u=

1 t ( u0 u0 cos(4 ft + 1 ) )dt , t 0

(3.10)

which can be simplified to

u = u0 u0 ( cos(4 f t ) sin (1 ) + sin(4 f t ) cos (1 ) ) . 4 f t

(3.11)

Note that the time dependent component of the velocity no longer produces a zero average and leaves a non-zero residual which depends upon the phase 1 . The phase dependent residual can be viewed as a source of error because the exact value of 1 is not known. The phase dependent residual can be made negligible by averaging over several pairs of images. Lets say n pairs of images are captured and the velocity is obtained by averaging n cross correlations. As a result, the average velocity can be roughly estimated as,

52

u = u0

n n u0 cos(4 f t ) sin (i ) + sin(4 f t ) cos (i ) . (3.12) 4 nf t i =1 i =1

If n is large, the phase dependent residual in (3.12) will become very small because the phase i keeps changing its value from one frame to another. Although the values of i have a particular order because the images were captured continuously but this order depends on f and t. If we assume that that the ac cycle and the image acquisition start in synch and then digress from each other, the phase i can be expressed as

i = 2 2if t intL ( 2if t ) ,

where intL (

n=0,1,2,...,n ,

(3.13)

) is the highest lower integer. In general, f and t might not have an

integer product (i.e. f t 1 ) and therefore i may attain various non-integer values. Averaging over a large number of phases will then yield a negligible phase dependent residual in (3.12). In our experiments, we have

f = 100 Hz,

t = 1/13.2 = 0.0757 and n = 50 . We find that for these parameters, the phase

dependent residual is approximately 4 orders smaller than the steady component, u0 . Hence, we can say that the measurement error due to a finite frame rate and due to the phase lag is negligible.

53

Frame time, t Frame 1 t1 Frame 2 t1+ t

2u0

Velocity, u

u0

1
Phase lag for first PIV pair

2
Phase lag for second PIV pair

1/f

2/f 3/f Time, t

4/f

5/f

Fig. 3.13: Effect of finite frame rate. The finite frame rate yields a time averaged velocity. But the phase lag between ac cycle and frame acquisition causes a phase dependent component in the average velocity. When averaged over a large number of frames, the phase dependent component becomes negligible.

3.10

Momentum Diffusion Length

Since the experiments are performed under AC electric fields, we must check if the time dependent component of the velocity has enough time to diffuse to a significant distance from the electrode. The momentum diffusion length can be derived from a scaling analysis of the following equation,
u 2u = 2 . t y

(3.14)

54

A quick scaling analysis yield a momentum diffusion length, y f , where is the kinematic viscosity of water. Taking = 106 m2/s and f=100 Hz, we find y = 100 m. The chamber height is approximately 125 m. This shows that the time dependent component of the velocity can diffuse upto a significant portion of the chamber, under the experimental ac frequencies. The PIV measurements yield only a time averaged velocity (as was discussed in the previous section) and therefore a momentum diffusion length smaller than the chamber height does not have much impact on our PIV measurements.
3.11 Numerical Simulations based on Linear Theory

ICEO flow can be simulated numerically by solving the electrokinetic equations described in chapter 2. The electrokinetic equations govern the electrostatic potential and the concentrations of various ionic species. However, the double layer, which is only 1-100 nm thick, is orders of magnitude smaller than the depth of the chamber (125 m). Therefore, one needs a really high resolution mesh in order to discretize the electrokinetic equations in the double layer. Since enormous computer resources will be needed for simulating the double layers in a microfluidic environment, workers tend to model the double layer as a set of effective boundary conditions. One of the effective boundary conditions is derived by writing a charge conservation equation for the double layer. The other effective boundary condition pertains to the mean concentration of the salt [61, 62]. The double layer charging process not only affects the electric field in the bulk but also causes gradients in the bulk salt concentration. The salt gradients, however, can be 55

ignored when the zeta potentials are low and the salt concentration can be assumed to be uniformly equal to the far field concentration. This leaves us with the problem of solving only for the electrostatic potential in the bulk and the double layer charge density. Once the electrostatic potential and the double layer charge density are obtained, we can find the slip velocity from the Helmholtz Smoluchowski equation and use that as a boundary condition in the Navier Stokes flow simulation. The electrostatic potential in the bulk obeys Laplaces equation, 2 = 0 . (3.15)

The time averaged flow velocity in the bulk is obtained by solving steady state Navier-Stokes equations,

u u = p + 2u ,
u = 0 .

(3.16) (3.17)

Fluid transport in microfluidic devices is mostly dominated by diffusion and therefore the nonlinear inertial terms can be ignored in (3.16).

3.9.1 Boundary Conditions

The walls of the chamber are insulated, i.e. n = 0 ,

u = 0.

(3.18) (3.19)

56

On the left driving electrode we can apply a sinusoidal potential to simulate AC fields,

0
2

sin t ,

(3.20)

and similarly, on the right driving electrode,

0
2

sin t ,

(3.21)

where 0 is the amplitude of the driving voltage. The voltage in peak-to-peak units will be reported as 20 . In our simulations, we assume that there is no slip on any boundary except the gate. In that case, the side walls, the top wall and the entire bottom wall except the gate electrode will have the following b.c.
u = 0.

(3.22)

3.9.2 Boundary Conditions on the Gate Electrode

A double layer is considered on the gate electrode. If we ignore nonlinear effects such as Faradaic reactions, surface conduction and concentration polarization [61, 62], the charge conservation equation for the double layer charge is given as q = n E . t (3.23)

where q is the double layer charge per unit area. The left hand side (lhs) of the preceding equation represents the rate of accumulation of double layer charge per

57

unit area. The right hand side (rhs) represents the normal flux of the charge into
the double layer. Note that n is the outward normal unit vector.

Under linear assumption, q can be related to the zeta potential, , by treating the double layer as a linear capacitor, i.e. q=

. D

(3.24)

Combining (3.23) and (3.24) and using E = yields

= n . D t
The zeta potential of the gate electrode is given as

(3.25)

= el +

q , Cs

(3.26)

where el is potential of the equipotential gate electrode and Cs is the capacitance of the Stern layer. The third term on rhs represents the drop across the Stern layer. Note that the Stern layer has been modeled as a linear capacitor which is a good assumption according to [2]. Combining (3.24) and (3.26) yields

el , 1 + D C s

(3.27)

which shows that the zeta potential is reduced by a factor (1 + ) where

= D Cs is a parameter representing the ratio of diffuse to Stern layer

capacitance. This parameter accounts for the Stern layer effect. The higher the value of , the lower the zeta potential .

58

Combining (3.25) and (3.27) yields the following form of the double layer charge conservation equation, (el ) = n . D (1 + ) t

(3.28)

Since the preceding equations are linear, all field quantities can be represented as phasors when using AC fields, e.g. = Im eit

where is the complex

amplitude of . Note that phasors can only be used with linear equations. Using phasors and rearranging the terms of (3.28), we get a simple boundary condition for the bulk potential,
n = i . D (1 + ) el

(3.29)

The time averaged Helmholtz Smoluchowski slip velocity can also be represented in terms of phasors us =

2 Im eit Im Es eit dt . 2 0

) (

(3.30)

After some algebra, the preceding equation can be simplified to us = 1 conj( Es ) + conj( ) Es , 4

(3.31)

where conj represents the complex conjugate and Es is the phasor amplitude of the

tangential component of the electric field. This equation can be used as the slip boundary conditions in the fluid flow solution.

59

3.12

Numerical Results

The linear model was simulated using a commercial finite element package COMSOL (Comsol Inc., Stockholm, Se). The geometry used in the simulations is the same as that shown in Fig. 3.5. Its not clear what ions are present in the purified water because purified water is a mixture of deionized (DI) water and fluorescent particle suspension. For numerical purpose, we have assumed that K+ and Cl- are the main ions present. An average diffusivity value of D = 1.995 109 m2/s at room T=200C was used in the numerical simulations. The gate potential el was taken to be zero. Following are the constants used in simulations.

Table 3.1: Constants for Numerical Simulations Cons tant a D k T Value 200e-6 m 1.995e-9 m2/s 1.381e-23 J/K 293.15 K 1.002e-3 Pa-s 80.1 8.854e-12 F/m 1.602e-19 Coulomb 1 Description Width of the gate electrode Average diffusivity of K+ and ClBoltzmann constant Room temperature Viscosity of water at 200 C Relative permittivity of water at 200 C Absolute permittivity of vacuum Elementary charge Valence of K+ and Cl- ions

r
e Z

Fig. 3.14 shows the numerical value of the slip velocity for purified water ( = 17 S/cm) when 20 Vpp (i.e. 0 = 10 Volt) at 100 Hz is applied to the driving electrodes. The magnitude of the velocity is also shown in Fig. 3.15 along with the

60

experimental data. The numerical value of the slip velocity is almost two orders of magnitude higher than the experiments. This indicates that the linear theory is very inadequate in predicting the slip velocities for high zeta potentials. Please note that in this particular experiment, the zeta potential is of the order 1 Volt which is 40 times higher than the thermal voltage ( kT ze = 25.9 mVolt) at room temperature. The zeta potential can be estimated as follows. The applied voltage of 20 Vpp means that the maximum instantaneous potential difference between the driving electrodes is 0 = 10 Volt. The driving electrodes have a gap of 1 mm approximately. This yields a bulk electric field of Es 104 V/m. The maximum zeta potential is then Es a / 2 = 1 Volt. Here a (the width of the gate electrode) is taken to be 200 m.

61

5 4 3 Slip Velocity, us (m/s) 2 1 0 -1 -2 -3 -4

x 10

-3

Linear simulation

-5 -1

0 Distance from the center, x (m)

1 x 10
-4

Fig. 3.14: The slip velocity on the gate electrode as predicted by a linear simulation. The simulations were performed for purified DI water ( = 17 S/cm) and driving conditions of 20 Vpp at 100 Hz.
10
-2

Linear simulation

Slip Velocity, us (m/s)

10

-3

10

-4

10

-5

Experimental data

10

-6

-1

0 Distance from the center, x (m)

1 x 10
-4

Fig. 3.15: Comparison of numerical and experimental slip velocity magnitudes for purified DI water ( = 17 S/cm) and driving conditions of 20 Vpp at 100 Hz. The numerical values are almost 2 orders of magnitude higher than the

62

experiments. It shows that the linear theory fails to predict the correct magnitudes of the slip velocity.

3.13

Breaking the Symmetry: Field Effect Flow Control (FEFC)

The symmetry of the flow can be useful for several purposes such as mixing and enhanced transport of suspended species; but it becomes a roadblock when it comes to pumping. The symmetric flows do not pump the fluid in any one direction; instead the fluid moves in fixed vortices. In order to pump, the symmetry has to be broken. The symmetry of ICEO flow can be broken by breaking the symmetry of the induced charge. ICEO flows are symmetric when the charge residing on the metal surface is symmetric. This condition is automatically broken when the surface has a native charge (such as a pre-charged surface). In such cases, the native charge gives rise to an electroosmotic flow which moves in one direction in presence of DC electric field. In presence of AC fields, however, such a surface again yields zero time averaged flow. Apart from this, we dont have native charge on many surfaces of interest, such as the electrodes in a microfluidic system. In such cases, there is another way of breaking the symmetry of ICEO flow, that is, by applying a radial voltage to the metal surface. The radial voltage essentially produces a uniform charge density in addition to the symmetric induced charge. This method yields an asymmetric ICEO flow leading to a scheme for micro pumping. This method can also be understood in terms of zeta potentials. In case of symmetric ICEO, the zeta potential of the surface has a symmetric nature.

63

By applying an external radial voltage to this surface, we can modify its zeta potential and create asymmetries. This method is called field effect flow control (FEFC) and is portrayed in Fig. 3.16.

E Symmetric flow

E Net pumping

+ + + + + - - - - Floating gate surface

+ + + + + + + + - Vg

Fig. 3.16: Concept of field effect flow control. When no external voltage is applied to the gate electrode, it developed a symmetric bipolar charge density and produces a symmetric ICEO flow. However, when an external voltage is applied to the gate, its charge density no longer remains symmetric and breaks the symmetry of the flow. As a result, a net pumping is achieved on the gate surface. The method of modifying the flow field by applying a gate voltage is called field effect flow control (FEFC).

The name field effect is derived from semiconductor physics. Our FEFC pump has a strong analogy with a semiconductor field effect transistor (FET). A field effect transistor has three electrodes: a source, a gate and a drain. When an electric field is applied between the source and the drain, electrons flow from the source to the drain. However, the flux of the electrons depends strongly on the gate voltage. By modifying the potential of the gate, one can modify the amount of current flowing between the source and the drain. In a similar fashion, the gate potential of an FEFC pump determines the direction and the magnitude of the pumping velocity.

64

3.14

FEFC in DC Electric Fields

Field effect has been used in the past for enhancing the flow velocities of DC electroosmotic pumps. In these works, an electroosmotic flow was established on a glass surface by applying a tangential DC field. A gate electrode was embedded in the surface of the glass. When the potential of the gate is fixed by connecting it to a power supply, the zeta potential of the glass surface is modified. As a result, the slip velocity is modified. This method has been utilized for enhancing the velocity of DC electroosmotic flows [48, 53] for controlling band dispersion in capillary electrophoretic devices [51] and recently for developing nanofluidic field effect transistors [54-56]. Lets derive an expression for FEFC pumping velocity based on linear theory. If we ignore the Stern layer, then the zeta potential of the gate surface can be expressed as

= g ,

(3.32)

where g is the gate potential. If E is the external electric field, then under the linear assumption, the slip velocity can be given as us =

( ) E , g

(3.33)

If we assume that the bulk potential decays linearly, then

E x ,
and therefore,

(3.34)

65

us =

( + E x ) E . g

(3.35)

Note that the slip velocity will be symmetric about the center x = 0 if g = 0 . However, for any other applied gate potential i.e. g 0 , the slip velocity is not symmetric. The net pumping velocity can be expressed as, u pump = 1 a/2 us dx . a a / 2 (3.36)

Substituting for us from (3.35) and integrating we get u pump = g E .

(3.37)

Note that if g < 0 , the net pumping velocity is from left to right and vice versa. The pumping velocity varies linearly with the gate potential g in the linear theory limit.
3.15 FEFC in AC Electric Fields

In order to enable field effect flow control in AC electric fields, the gate potential also has to change polarity in unison with the driving voltage. Then only a non-zero time average flow can be achieved. The phase gap between the gate and the driving voltages can either be zero or 1800. These two phase gaps will produce pumping in opposite directions. Evidence of FEFC in AC electric fields was first presented by Mutlu et al. in 2004 [69]. They embedded a metal electrode in an insulating material and applied very high potentials to the embedded gate electrode. We develop the idea of AC FEFC further and present a detailed

66

experimental study. We do not insulate our gate electrode with any insulating material. Instead, our gate electrode makes direct contact with the ionic liquid and therefore we dont have to apply high potentials to the gate.
3.16 Experimental Evidence of FEFC in AC Electric Fields

AC FEFC experiments were performed with the device also used for symmetric ICEO flow described in the previous sections. The experiments were performed in three different KCl solutions: 1 mM ( = 165 S/cm), 10 mM ( = 1400 S/cm) and 50 mM ( = 2950 S/cm). A driving voltage of

0 = 60 Vpp at f = 500 Hz was applied between the driving electrodes in all the
experiments. Lets say that the left driving electrode is at 60 Vpp while the right driving electrode is grounded. Since the gate electrode is symmetrically placed between the two driving electrodes, the floating potential of the gate can be assumed to be Vg = 0 / 2 = 30 Vpp at f = 500 Hz . Therefore, if we apply a potential of Vg = 30 Vpp at 500 Hz to the gate, a symmetric flow will be produced. However, if the applied gate potential is different from 30 Vpp, the flow will no longer remain symmetric. For Vg < 30 Vpp , the net flow will be from left to right and for Vg > 30 Vpp , the net flow will be from right to left. This behavior has been demonstrated experimentally (see Fig. 3.17).

67

Fig. 3.17: Field effect flow control. Shown are the experimentally measured velocity vectors on the gate electrode for various values of gate voltage. The large arrow shows the direction of net pumping. For Vg < 30 Vpp , the fluid flows from
left to right whereas for Vg > 30 Vpp , the flow goes from right to left. The experiments were performed in 1 mM KCl under the driving conditions of 60 Vpp at 500 Hz.

Velocity profiles for several gate voltages are shown in Fig. 3.18. The velocity profile is not symmetric for any Vg 30 Vpp . Infact, the velocity is mostly positive for Vg < 30 Vpp and negative for Vg > 30 Vpp .

68

8 6 Slip Velocity, us (m/s) 4 2 0

x 10

-5

Vg=18.8 Vpp 22.2 Vpp 25.2 Vpp 28.1 Vpp

31.4 Vpp

-2 -4 -6 -8 -1 -0.5

35.5 Vpp

Vg=39.2 Vpp

0 x (m)

0.5 x 10

1
-4

Fig. 3.18: Experimentally measured slip velocity profile on gate electrode for various values of gate voltage. The experiments were performed in 1 mM KCl under the driving conditions of 60 Vpp at 500 Hz.

These velocity profiles can be integrated over the width of the gate electrode in order to find the net pumping velocity, u pump = 1 a/2 us dx . a a / 2 (3.38)

Naturally, u pump = 0 when the flow is symmetric (i.e. Vg = 30 Vpp ). The net pumping velocity u pump has been plotted for three different ionic concentrations as a function of Vg in Fig. 3.19. As the ionic concentration increases, the net pumping velocity decreases. This behavior is again attributed to an increased Stern layer effect at high ionic concentrations. The net pumping velocity is symmetric about

69

Vg = 30 Vpp . The direction of pumping can be easily reversed by modifying the potential on the gate. Note that, u pump > 0 for Vg < 30 Vpp and u pump < 0 for

Vg > 30 Vpp . FEFC thus provides us with a great tool to not only modify the pumping velocity but also the direction of pumping with the ease of modifying the gate voltage.
5 Net pumping velocity, u (m/s) pump x 10
-5

1 mM 10 mM

2.5

0 50 mM -2.5

-5 10

15

20

25 30 35 40 Gate Voltage, Vg (Vpp)

45

50

Fig. 3.19: Net pumping velocity as a function gate voltage. For Vg < 30 Vpp , the
net pumping velocity is positive whereas its negative for Vg > 30 Vpp . The experiments were performed in 1 mM KCl under the driving conditions of 60 Vpp at 500 Hz. Net pumping velocities of 50 m/s can be achieved while working with 1 mM KCl solution (see Fig. 3.19). The pumping velocity does not increase further even if the gate voltage is biased further. We define the bias as Vg 30 i.e. the difference between the applied and the floating gate potentials. The saturation

70

behavior at high bias is attributed to surface conduction. The saturation effect is more prominent for low ionic concentration (1 mM) than for high ionic concentrations (10 mM and 50 mM). This behavior can be justified by comparing the double layer thicknesses for these solutions. The double layer thickness decreases as the ionic concentration increases. When the double layer is very thin, the surface conduction currents are negligible because there is not enough diffuse charge to create the surface current. In other words, surface conduction is almost absent in the 10 mM and 50 mM experiments and therefore the velocity does not saturate at high bias in the gate voltage. The pumping velocity is also a function of the driving frequency. As the frequency increases, the net pumping velocity decreases (see Fig. 3.20).
10 Net Pumping Velocity, u (m/s) pump
-4

10

-5

10

-6

10

10 10 Driving Frequency, f (Hz)

10

Fig. 3.20: The net pumping velocity as a function of the driving frequency. The velocity decreases as the frequency increases. The experiments were performed in 1 mM KCl under the driving conditions of 60 Vpp at 500 Hz. The gate voltage was fixed at 15 Vpp.

71

3.17

Linear Simulation of FEFC

The net pumping velocity for the preceding experimental conditions was also calculated through linear numerical simulations. The gate voltage was varied to obtain the pumping velocity as a function of the gate voltage. The numerical pumping velocity as a function of the gate voltage is shown in Fig. 3.21. The numerical pumping velocity varies linearly with Vg, unlike the experiments. The linear theory does not take into account surface conduction, the reason to which the saturation behavior of experimental velocity (Fig. 3.19) was attributed. The net pumping is zero for Vg=30 Vpp also observed in the experiments. The numerical magnitudes of u pump have also been compared with the experimental data (see Fig. 3.22). The numerical values are almost three orders of magnitude higher than the experiments. This time the discrepancy (3 orders) is higher than before (see Fig. 3.15). This is because the experimental data in Fig. 3.22 was obtained in 1 mM KCl whereas the data in Fig. 3.15 was obtained in purified water. The Stern layer effect is more prominent at higher ionic concentration and reduces the experimental velocity. We have not taken the Stern layer into account in our simulations. Therefore, the discrepancy in 1 mM KCl is higher. The numerical simulations do not predict the saturation behavior of velocity at high voltages which was observed in the experiments (Fig. 3.19). This indicates that the linear theory is inadequate for simulating high zeta potential situations such as those encountered in our experiments.

72

0.03 Net Pumping Velocity, u (m/s) pump 0.02 0.01 0 -0.01 -0.02 -0.03 10 Linear Simulations

20 30 40 Gate Voltage, Vg (Vpp)

50

Fig. 3.21: Numerical values of the net pumping velocity as a function of gate voltage. The simulations were performed in 1 mM KCl under the driving conditions of 60 Vpp and 500 Hz.
10 Magnitude of Net Pumping Velocity upump (m/s)
-1

Linear Simulations 10
-2

10

-3

10

-4

Experimental Data

10

-5

10

20 30 40 Gate Voltage, Vg (Vpp)

50

Fig. 3.22: Comparison of numerical and experimental values of the net pumping velocity. Numerical values are 3 orders of magnitude higher than the experiments. The simulations and experiments were both performed in 1 mM KCl under the driving conditions of 60 Vpp and 500 Hz.

73

3.18

Uncertainty Analysis

The uncertainty of a PIV measurement originates from the uncertainty in the measurement of two quantities: the displacement of the particles, x , and the time interval between the two frames of correlation, t . The velocity can be expressed as u= x t (3.39)

If ( x ) and ( t ) are the uncertainties in x and t , then the standard error analysis yields the following expression for the uncertainty in the velocity,

u ,

( x ) ( t ) = + u x t
2

(3.40)

In our experiments, the frames were obtained at a frame rate of 13.2 Hz, which yields t = 0.075 s . The time interval, t , is controlled by an electronic camera controller and the uncertainty ( t ) is generally very small. The uncertainty in the displacement, ( x ) , on the other hand, depends on many factors such as the Brownian motion, numerical aperture of the microscope, wavelength of the light, size of the particles, pixel resolution of the CCD camera, choice of the interrogation method and the choice of the correlation-peak finding method. Therefore, the relative uncertainty in displacement, i.e. ( x ) / x , is generally much larger than the relative uncertainty in the time interval, ( t ) / t . In other

74

words, the uncertainty in the PIV measurement is mainly due to the uncertainty in determining the displacement of the particles,

( x ) u x

(3.41)

Since submicron size particles (0.7 m) were used for tracing the flow, the uncertainty due to Brownian motion must be taken into account. A first order estimate of this error relative to the displacement in the x-direction is given as [67],

B =

s2

1/ 2

1 2D u t

(3.42)

where s 2 is the random mean square particle displacement associated with Brownian motion and D is the Brownian diffusion coefficient of the particles. Note that B is equivalent to an inverse Peclet number, i.e. B = Pe 1 where Pe = u B D and B = 2 Dt . The Brownian error is small when the flow is fast or the particles are large. We can make an estimate of the Brownian error in our experiments by estimating the diffusion coefficient of the particles. According to the Einstein equation, the diffusion coefficient of a particle of radius a is given as D= kT 6 a (3.43)

For 0.7 m particles we estimate D 3 1013 m2/s. The characteristic velocity and the time interval are respectively u ~ 100 106 m/s and t = 0.075 s. Substituting these values in (3.42) yields B = 2% .

75

We also have to consider presence of a random error in our measurements. The random error is caused by the imperfections in the particle images. If a particle image diameter (when projected back into the flow field) is big enough to cover 23 pixel of the CCD array, its displacement can be determined within 1/10th of the its image diameter (projected back into the flow field). The diffraction limited diameter of a particle image is given by [70]
2 1.22 2 de = M d p + NA
1/ 2

(3.44)

where d p is the diameter of the particle (0.7 m), is the wavelength of the emitted light (612 nm), NA is the numerical aperture of the lens (0.25) and M is the magnification of the lens (10x). When projected back into the flow, the image diameter becomes d e / M which is approximately equal to 3 micron in our case and covers 4-5 pixels of our CCD array. As mentioned before, we estimate that the random errors is approximately 10% of the diameter of the particle image when projected back into the flow field [71], i.e. ( x ) = 0.1 3 = 0.3 m. In our PIV analysis, a 20 micron wide interrogation spot was used with a 50% overlap. As a result, the full scale displacement that we can measure, xFS , is approximately 10 micron (half of the interrogation region width). Referencing the error to full scale displacement, we get a relative uncertainty of 3%

( u / u FS = ( x ) / xFS = 0.3 /10 = 0.03 ). This relative uncertainty can be reduced

76

by increasing the time gap between the two frames so that the particle displaces by a longer distance between the frames. In a nutshell, the relative uncertainty due to Brownian motion is approximately 2% and the random error is approximately 3% (when referenced to the full scale displacement). Since different types of uncertainties are independent of each other, a total uncertainty can be found by taking the root of the sum of the squares, i.e. total uncertainty =
3.19
22 + 32 = 3.6% .

Conclusions

We have presented extensive experimental data which verifies the existence of ICEO flows in microfluidic devices. We observed symmetric ICEO flows on a planar microelectrode and measured the flow velocities using micro particle image velocimetry. Slip velocity of 40 m/s was observed in purified water under driving conditions of 20 Vpp at 100 Hz. From our experiments, we learned that the slip velocity decreases as the ionic concentration increases. This behavior was attributed to the Stern layer effects. We also learned that the slip velocity is a function of the driving voltage and the frequency. As the driving voltage increases, the slip velocity increases quadratically. However, at very high voltages, the slip velocity stops growing and either decreases or stays constant as the voltage is increased further. This saturation behavior has been attributed to surface conduction. Our claim about surface conduction is bolstered by the observation that the saturation behavior occurs at a lower voltage in purified water than in 1 mM KCl. Surface conduction is higher when the ionic concentration is lower and

77

thus causes a saturation of slip velocity at a lower voltage. The velocity decreases as the frequency increases. A pumping strategy based on field effect flow control (FEFC) was discussed and demonstrated experimentally. Field effect proves to be a good way of pumping at microscale. It gives the user flexibility of modifying the magnitude and the direction of pumping by simply modifying the voltage applied to the gate electrode. Pumping velocity of 50 m/s was obtained in 1 mM KCl under driving condition of 60 Vpp at 500 Hz and an applied gate voltage of 15 Vpp. The relative uncertainty in the experiments was estimated to be approximately 20%. We also performed numerical simulations based on linear Debye Huckel theory. Our simulations predict results which are 2-3 orders of magnitude higher than the experimental data. Such a high discrepancy between the linear simulations and the experiments imply that the linear theory is inadequate for simulating our experiments. In our experiments, the induced zeta potentials are much higher than the thermal voltage (10-40 times higher). Under such condition, we do expect the linear theory to fail. In the next chapter, we will introduce a nonlinear model of double layer which takes into account nonlinear effects such as nonlinear surface capacitance and surface conduction. Surface conduction in particular will be explored in detail. We will show that high amounts of surface currents can flow when the induced zeta potentials are high. These currents cause significant reduction in the tangential component of the electric field and therefore deteriorate the slip velocity. In chapter 5, a full AC nonlinear model will be simulated to make better predictions 78

about our experiments. It will be shown that the combined effects of nonlinear surface capacitance and surface conduction indeed help in bridging the discrepancy observed in the current chapter.
3.20 Appendix to Chapter 3

Microelectrode Fabrication with Image Reversal Lithography

Following are the steps for the fabrication of microelectrodes in ICEO device 1. Clean a glass wafer in acetone (3 minutes in ultrasound bath) 2. Clean in isopropanol (3 minutes in ultrasound bath) 3. Dry by blowing N2 gas 4. Stir clean in a piranha solution (H2SO4:H2O2 = 2:1) at 800 C for 10 minutes 5. Dry by blowing N2 gas 6. Dehydrate on a hotplate at 1100 C for 5 min 7. Spin coat HMDS at 4000 rpm for 45s 8. Spin coat AZ5214 photoresist at 4000 rpm for 45s 9. Soft bake on a hotplate at 950 C for 90s 10. Mask-expose with a contact aligner for 8s 11. Post exposure bake on a hotplate at 1100 C for 1 min 12. Flood expose on a contact aligner for 1 min 13. Develop in AZ400:H20 (1:4) for 25s 14. Rinse with deionized water 15. Dehydrate on a hotplate at 950 C for 5 min 16. Deposit metal with electron beam evaporator, 10 nm Ti, 200 nm Pt 17. Lift off residual metal in acetone with ultrasound bath

79

80

Chapter 4. Surface Conduction in Induced Charge Electroosmosis

We have simulated highly nonlinear electrokinetic phenomena which occur at high zeta potentials. Special attention has been paid to the understanding of surface conduction in ICEO flows. We show that surface conduction through nanoscale diffuse layer can cause micron scale gradients in the bulk electric field and reduce the slip velocity thereby. Our work reveals some dramatic and surprising consequences of surface conduction in ICEO flows. In ICEO flows, the zeta potentials are induced by the applied electric field. As a result, the zeta potentials are generally high and very nonuniform on the electrode surface. Nonuniform zeta potentials give rise to nonuniform surface conduction currents which lead to unique results not observed in phenomena with uniform zeta potentials. Khair and Squires 2008 [72] had analyzed the case of a surface with a uniform zeta potential and demonstrated that surface conduction pulls the electric field lines into the double layer close to the stagnation points (or corners) of the surface. The electric field lines heal from this pulling after a short region (called healing region) close to the corners and become parallel to the surface again. We find these corner healing regions in our analysis of nonuniform ICEO flows too; however, we find another healing region in the middle of the surface where the lines are expelled out of the double layer. This middle healing region is caused by the nonuniform surface currents inherent to ICEO flows. We will solve two canonical cases of ICEO on a flat metal electrode and on a metallic cylinder. In cases where surface

81

currents are strong, the length of the healing regions can be comparable to the surface dimensions and the tangential component of the electric field is significantly reduced in the healing zones. Surprisingly the tangential field is increased in the center of the surface. Overall, surface conduction reduces the tangential field and the zeta potential and therefore reduces the slip velocity significantly. Our analysis also shows that surface conduction grows not only with the induced zeta potential but also with the double layer thickness. In cases where double layer is infinitesimally thin, surface conduction becomes negligible and has minor effect on the slip velocity. We develop a fundamental picture (both qualitative and quantitative) of surface conduction in ICEO flows, showing its unique features. We also introduce a numerical model to simulate surface conduction both in steady and time dependent cases. In time dependent cases (such as AC), nonlinear capacitance will also play a role. In steady cases, however, surface capacitance does not play any role because enough time is provided for the double layer (capacitor) to charge. In the presented numerical model, the electric double layer is realized by solving a partial differential equation (PDE) on the surface. The PDE is derived by conserving ionic charge in the double layer. A normal ohmic current and a tangential surface conduction current are balanced with the rate of charge accumulation. This model is more advanced than the linear model based on Debye Huckel approximation because the linear model ignores the presence of surface conduction altogether and uses a linearized surface capacitance. These simplifications are not valid for situations in which the zeta potential is much higher than the thermal voltage and 82

therefore our nonlinear model serves as a tool for simulating high zeta potential situations.
4.1 Effects at High Zeta Potentials

When a surface acquires charge because of a spontaneous chemical change, we can expect its zeta potential to be of the same order as the thermal voltage ( kT ze ) because chemical changes are facilitated by thermal energy. For example, when a glass surface comes in contact with water, it acquires negative charge and develops a zeta potential of about -0.1 Volt (only four times higher than the thermal voltage at room temperature). This is, however, not the case in ICEO where the zeta potential is induced by the electric field. The zeta potential in such cases is proportional to the electric field strength. In many microfluidic experiments, electric fields are very strong and can induce several times higher zeta potentials than the thermal voltage. There are several effects which can take place at high zeta potentials, e.g., nonlinear double layer capacitance, surface conduction, chemi-osmosis [61, 62] and faradaic reactions [2, 38]. At high zeta potentials, the double layer behaves as a nonlinear capacitor requiring exponentially large amount of ionic charge, q, which according to the Gouy-Chapman model is expressed as q=
ze 2kT sinh . D ze 2kT

(4.1)

83

This means that an exponentially long time is required for charging the double layer completely. This results in very poor charging when using AC electric fields. The double layer can not acquire enough charge even at very low frequencies, yielding a low zeta potential, a low tangential electric field and therefore a low slip velocity. The surface conductivity, s , also becomes very high at high zeta potentials. According to the Gouy-Chapman model for surface conductivity [1], s is an exponentially growing function of zeta potential, ,

s = 4D (1 + m ) sinh 2

ze , 4kT

(4.2)

where m is a dimensionless parameter indicating the relative contribution of electroconvection to surface conduction. In presence of large zeta potentials, a significant amount of current flows parallel to the surface through the double layer. When a lot of charge flows through the double layer, a similarly large amount of charge has to be supplied from the bulk in form of a normal flux [72]. This causes the electric field to have a large normal component and a low tangential component. This creates a deteriorating impact on the slip velocity. Next we expand this argument by drawing a fundamental picture of surface conduction.
4.2 A Fundamental Picture

Before discussing ICEO, lets consider a steady electroosmotic flow on a finite size plate with a fixed negative charge. In a case when the zeta potential of the surface is low (
kT ze ), the electric field outside the double layer can be

84

assumed to be equal to the applied uniform field of E [73]. This is, however, not true when the surface is highly charged, i.e. > kT ze . In this case, large surface current flows through the diffuse layer for which the ions have to be supplied from the bulk. Khair and Squires, 2008 [72] showed that the electric field lines are drawn into the double layer close to the edges of plate (Figs. 4.1 and 4.2) bringing ions into the double layer, conserving charge thereby. The field lines again become parallel to the surface after a healing length LH and the electric field becomes equal to E . By a simple current counting argument, Khair and Squires, 2008 [72] showed that the healing length is equal to the ratio of surface-to-bulk electrolyte conductivities, LH

s ze = 4D (1 + m ) sinh 2 . 4kT

(4.3)

The healing length can be derived easily by applying charge conservation on a volume V extending a distance D in normal direction and a tangential distance LH from the left edge of the plate. A surface current s Ex leaves the right boundary of this volume; however no surface current enters on the left boundary because there is no surface charge upstream of the plate. To maintain the current balance, a normal current E y LH enters from the bulk. Note that the surface conductivity, s , has units of S, whereas the bulk conductivity, , has units of S/m. By equating the two currents, we reach the aforementioned expression for LH . The same argument works for the right edge of the electrode. For highly

85

charged surfaces, the healing length can be much larger than the debye length D . For example, taking = 250 mV and D = 10 nm gives LH 1.5 m . In other words, nanoscale surface conduction can cause micro scale gradients in the bulk electric field (see Fig. 4.2). Surface conduction has been studied in colloidal science for a long time and has serious consequences for the electrophoretic mobility of charged particle [7]. In fact, at sufficiently large zeta potentials, the electrophoretic mobility of a charged particle decreases with increasing because the tangential field around the particle is reduced due to surface conduction [8-11]. In context of electrophoresis, the ratio of healing length to the particle radius is also known as Dukhin number, Du, and is given as [1, 13] Du = LH s 4D ze = = (1 + m ) sinh 2 4kT a a a (4.4)

The preceding equation accounts only for diffuse layer surface conduction. We have not accounted for the Stern layer surface conduction because the ionic mobility and diffusivity in the Stern layer are unknown. Strategies to estimate the Stern layer Dukhin number are enumerated in [1].

86

jy Ey
+ + +

jy Ey

+ + +

js = 0

+ + +

js s E x

- - - - - - - - - - - LH
constant constant s

+ +

js = 0
js = 0 =0

js = 0 =0

LH

Fig. 4.1: The fundamental picture of surface conduction on a highly charged plate. Consider charge conservation in a control volume close to the left edge of the plate. A surface current, s Ex , leaves the inner boundary of the volume but no surface current enters the outer boundary because there is no surface charge outside the plate. As a result, a normal current, E y LH , enters from the bulk causing the electric field lines to be pulled into the diffuse layer and maintaining the charge conservation. This picture is adapted from Khair and Squires 2008 [72].

---- ---- ---- ---- ---- ---- ---- ---LH LH

Fig. 4.2: Electric field lines on a highly charged plate. The field lines are pulled into the double layer as a result of surface current flowing through the diffuse layer. The field heals from the discontinuity in the surface current in a distance LH and becomes parallel to the surface again.

Now lets discuss the main subject of this chapter, i.e. surface conduction effects in ICEO flows. Consider ICEO flow on a floating metal electrode subjected to a uniform applied field E . Since the electrode is not charged, it must acquire a bipolar charge distribution in order to maintain its charge neutrality. This yields a

87

nonuniform zeta potential distribution along the width of the electrode, with the maximum zeta potential, max , on the two edges and a zero zeta potential at the center. In a linear case, zeta potential will vary linearly, = E x , with x = 0 at the center of the electrode. However, when is large, surface current will flow from the two edges (composed of opposite ions, see Fig. 4.3). A current counting analysis similar to the one above will yield healing regions close to the two edges. However, at the center of the electrode, the surface conductivity is zero (because

= 0 ). As a result, the ions of the surface current are forced to escape into the
bulk before they reach the center. In turn, the electric field lines are expelled out from the double layer (focus on left to the center), yielding another region in which the field has a large normal component (see Fig. 4.4). A similar but reverse argument works on the other half of the electrode (right to the center).

jy Ey
+ +

jy Ey

+ + + +

- -

js = 0 js = 0 =0

- - - - - - + + + + ++
max

+ +

- js = 0 js = 0 =0

LH

js ,max

js = 0 =0

js ,max LH

+ max

Fig. 4.3: Fundamental picture of surface conduction on a floating electrode subjected to an applied field. The electrode acquires a bipolar charge and a nonuniform zeta potential distribution. Surface currents yield healing regions at the edges. However, in the center, where zeta potential is zero, no surface current can flow and therefore the ions are expelled out of the double layer. This yields another healing region in the middle of the electrode.

88

---- -------- ---LH

++++ ++++++++ ++++

LH

Fig. 4.4: Electric field lines on a floating electrode subject to an applied electric field. The field lines are first pulled into the double layer and then expelled out because of nonuniform surface currents. This feature is unique to ICEO flows where nonuniform zeta potentials are induced by the applied electric field.

Note that the surface current in ICEO is very nonuniform; its zero not only at the two edges but also at the center of the electrode. The surface current has a maximum somewhere between the edge and the center. The location of the maximum is also the location where the electric field lines start being expelled out (left to the center). Note that, nonuniform surface currents virtually leave no width where the electric field is completely tangential to the surface. On almost all the points, the field has a normal component, either being pulled in or expelled out. Only in the very center of the electrode, the electric field is completely tangential and can attain a very large value. A reduced tangential field renders a low zeta potential and low slip velocity.

89

4.3

Bulk Equations

Under the assumption that the bulk is electroneutral and the salt concentration in the bulk is uniform, the electrostatic potential in the bulk, , satisfies Laplaces equation, 2 = 0 . (4.5)

The time averaged flow velocity in the bulk is obtained by solving steady state Navier-Stokes equations,

u u = p + 2u ,
u = 0 .

(4.6) (4.7)

4.3.1 Bulk Boundary Conditions

At all the walls, electric insulation and no-slip are applied as boundary conditions, i.e.,

n = 0 ,
and
u = 0.

(4.8)

(4.9)

At the electrodes where electroosmotic slip occurs, following are the boundary conditions: Assuming that the potential drop across the Stern layer is negligible, the bulk potential outside the double layer can be expressed as

= el ,

(4.10)

90

where el is the potential of the equipotential surface formed by the electrode. We will show a strategy to obtain in the next section. The slip velocity on the electrode is given by Helmholtz Smoluchowski equation
u=

s s ,

(4.11)

where s is a vector tangential to the electrode surface.

4.4

Charge Conservation in the Double Layer

Assuming that the double layer is infinitely thin ( D a

1 ) and that the double

layer charging does not cause any gradients in the bulk salt concentration, we can derive a conservation-law for the double layer surface charge density q. Consider a small patch of a thin double layer (Fig. 4.5). The charge is brought into it by a normal ohmic flux. This charge can accumulate or leak tangentially from the edges.

Ohmic flux in n. ( E ) ds

ds

dl

nl . ( s Es ) dl Surface conduction flux out

Fig. 4.5: Conservation of surface charge. A tangential surface conduction flux and a normal ohmic flux constitute the conservation law for the double layer surface charge density q.

91

A conservation law can then be simply written as Rate of accumulation = Flux in Flux out, which in mathematical form can be expressed as (4.12)

s t ds = s n ( E) ds l nl ( s Es ) dl ,

(4.13)

where n is the outward normal vector and nl is a vector tangential to the surface

but perpendicular to the edge. The second term on right hand side (rhs) represents a linear integral on the boundary of the surface. We can convert it into an area integral by applying divergence theorem, s t ds = s n ( E) ds s s ( s Es ) ds . q (4.14)

Here, s is the tangential gradient operator on the outer surface of the double layer. If the electrode surface lies in the x-y plane, the tangential gradient operator will simply be s = x x + y y . Since (4.14) holds true for any arbitrarily small surface, the integrals can be removed and we get q = n ( E ) s ( s Es ) . t (4.15)

Using the electrostatic relations E = and Es = s (where is bulk electrostatic potential right outside the double layer), we get q = n ( ) + s ( s s ) , t (4.16)

92

4.4.1 Double Layer Edge Conditions

Since the surface PDE (4.16) contains spatial derivatives, we need some spatial boundary conditions (called edge conditions here, to distinguish them from the bulk boundary conditions) to solve it. We have previously argued that the surface conduction flux is zero at the edge of the electrode and therefore the following can be used as an edge condition (also see Fig. 4.6)
nl ( s s ) = 0 ,

(4.17)

nl ( s s ) = 0

nl

n
s

nl
nl ( s s ) = 0

Electrode Surface
q = n ( ) + s ( s s ) t
nl ( s s ) = 0

nl
Fig. 4.6: Conditions on the edges of the electrode. The surface conduction flux is zero on the edges. Here n is the outward surface normal vector, s is an arbitrarily oriented tangential vector, and nl is a vector tangential to the surface but perpendicular (outward) to the edge.

4.4.2 Convergence Issues

We can solve (4.16) for q but several convergence problems may arise because this PDE does not contain any spatial diffusion of q. The spatial derivatives of , which occur in this PDE, are supplied from the bulk and therefore they behave as

93

Ed ge

nl

nl ( s s ) = 0

source terms. Essentially, there is no diffusive term available for q in this equation. The absence of diffusive terms combined with the presence of source terms leads to an infinite Pclet number and can create problems in convergence. Moreover, magnitude of q is supposed to grow exponentially as predicted by (4.1); as a result the computer will solve for a variable which is very large, leading to large numerical errors. One way to counter these problems is to change the dependent variable from q to . This is explained below. By differentiating (4.1) with respect to time, we can obtain an expression for
q dt in terms of dt ,
ze q . = cosh t D 2kT t

(4.18)

Now, by using the relation = el ,

where el is the potential of the

electrode surface and by ignoring the Stern layer, we can obtain an expression for the tangential divergence of in terms of

s ( s s ) = s ( s s ) .

(4.19)

Please note that el is a constant because the metal electrode forms an equipotential surface. Combining the two preceding equations with (4.16), we get
D n ( ) s ( s s ) . = t cosh ( ze 2kT ) (4.20)

Now, using the definitions c = a D and Du = s a , the preceding equation reduces to

94

 an a 2 s ( Du s ) , = c t cosh ( ze 2kT )

(4.21)

which is a well behaved equation in . The first term in the numerator of the rhs ( an ) is supplied from the bulk and behaves as a source. The second term, on the other hand, behaves as a diffusion term in . Another advantage of changing the dependent variable from q to is that it produces a factor of 1/ cosh ( ze 2kT ) on the rhs; in other words, the rhs is divided by a large number making the rhs small; this makes the problem easier to converge. Magnitudes of are generally much smaller than q and therefore we prefer to solve for instead of q. When one wants to solve for a steady state, (4.21) is reduced to n a s ( Du s ) = 0 .
4.5 Dimensionless Equations

(4.22)

Following are the scales for making the equations dimensionless. Asterisk in the superscript represents a dimensionless quantity.

x = ax* ,

= TH * ,
q = q* TH D , t = ct* ,
u = uHS u* ,

(4.23)

95

p = p* uHS a ,

where, TH = kT ze is the thermal voltage and uHS is an appropriate scale for the

Helmholtz Smoluchowski velocity; uHS will be chosen later based on geometry and application. The dimensionless equations for the bulk can then be written as *2 * = 0 ,
Re u* *u* = * p* + *2u* ,

(4.24) (4.25) (4.26)

* u* = 0 ,

where Re = uHS a is the Reynolds numbers. In most of microfluidic situations,

Re 1 and therefore the convective terms in (4.25) can be neglected.

The boundary conditions for the walls are

n * * = 0 ,

(4.27)

and
u* = 0 .

(4.28)

At the electrodes where electroosmotic slip occurs

* * = el * ,

(4.29)

and
u* =
2 TH a

uHS

* * s ) s .

(4.30)

96

The dimensionless form for the double layer charge conservation (see (4.21)) is expressed as
n * * * ( Du* * ) s s * , = * * t cosh 2

(4.31)

which, for a steady state analysis, simplifies to

n * * * ( Du* * ) = 0 . s s

(4.32)

The edge conditions in form of dimensionless variables are given as nl ( Du* * ) = 0 . s (4.33)

Dukhin number can also be expressed in terms of dimensionless variables,

Du = 4 * (1 + m ) sinh 2 * 4 ,

(4.34)

where * is the dimensionless double layer thickness, * = D a . Similarly, the nonlinear capacitive relation reduces to
q* = 2sinh * 2 .

(4.35)

We will now use the preceding nonlinear numerical model for solving two canonical cases: (1) ICEO flow on a flat metal electrode in presence of a steady DC electric field, and (2) ICEO flow on a metal cylinder in presence of a steady DC electric field.

97

4.6

ICEO on a Flat Metal Electrode

Consider a horizontal electrode of width a placed at the bottom wall of a chamber of side L. The chamber side is much larger than the width of the electrode, L*=L/a=15 (see Fig. 4.7). The chamber is filled with a symmetric ( z+ = z = z and D+ = D = D ) electrolyte solution such as KCl+H20. We will refer to the horizontal electrode as a gate electrode. We will carry out a steady state ICEO flow analysis in presence of a DC electric field.
n * * = 0, u* = 0
L*

*2 * = 0

*2u* * p* = 0

u* = 0
* =
L* 2

* u* = 0

u* = 0

* =
u * ( x) = 2

* 2

* x* * = *

L* 2

u* = 0 n * * = 0
p*=0

y
1

u* = 0 n * * = 0
Du * =0 x*

Du

* =0 x*

* * * Du * = 0 y* x x

Fig. 4.7: Steady state simulation model for ICEO flow on a flat electrode with L/a=15. In dimensionless form, the width of the electrode is 1. Geometry is not shown to scale.

98

The two vertical sides of the chamber serve as a pair of driving electrodes. A DC electric field is produced in the chamber by applying a potential difference between the two driving electrodes. Assume that the driving electrodes do not saturate (i.e. no double layer form on them) and can maintain a constant DC electric field in the chamber. Dimensionless potentials of * = L* 2 are applied on the left and right driving electrodes respectively (see Fig. 4.8). If none of surface conduction, Stern layer and faradaic injections is present, in the steady
* state a DC electric field of E0 = is produced as a result of the applied potential

difference. This electric field produces a zeta potential of * = x* on the gate surface (see Fig. 4.8).

L*/2

Bulk potential, *

/2
0

- /2

- L /2 * -L /2

-1/2

1/2

L /2

Fig. 4.8: Steady state bulk potential in the ICEO chamber when surface conduction and faradaic injections are absent. The net potential drop across the width of the electrode is equal to . In absence of Stern layer, this yields * = x* . Note that the maximum zeta potential is equal to / 2 .

99

In dimensional form the linear estimates can be written as E0 TH a and

TH 2 . Therefore, a natural scale for normalizing the slip velocity would be

uHS =
2 E0 2TH = . 2 a

(4.36)

Using this scale, we can define the normalized slip velocity on the gate as
u* =

* 1 2 x = 2* * x. uHS x x

(4.37)

The gate electrode is kept floating, i.e., not connected to voltage source. Since the gate electrode is placed at equal distances from the two driving electrodes, its
* floating potential vanishes i.e. el = 0 . One consequence of zero floating potential

is that the bulk boundary condition becomes

* = * .

(4.38)

Assume that the depth of the chamber into the paper is infinite. Therefore, only a 2D model needs to be simulated. The charge conservation PDE for this system is simply (see (4.32)) * * * ( Du * ) = 0 , * y x x and the condition on the two edges can be expressed as Du * =0 x* (4.40) (4.39)

100

4.6.1 Parameters of Study

The parameters of study are

and the dimensionless double layer

thickness * . As described in Fig. 4.8, / 2 is the maximum linear value of * . Naturally, as increases, * increases. As a result, both Dukhin number and the surface conduction flux increase. Similarly, from (4.4), Dukhin number (and surface conduction flux) is proportional to * . Surface conduction effects grow with * . We will vary these two parameters to quantity the effect of surface conduction on ICEO flow.
4.6.2 Results and Discussion

The problem was solved with COMSOL Multiphysics (Comsol Inc., Stockholm, Se), a commercial finite element software package. The details of the weak formulation, the grid size and the solution procedure are given in an appendix to this chapter. Lets first consider the case of very thin double layer and small induced zeta potential, namely * = 104 and = 0.1 . In this case, surface conduction is negligible. In the steady state, the double layer gets completely charged and the electric field lines become completely tangential to the electrode, showing no existence of a healing region (Fig. 4.9a). In other words, the field heals instantaneously. The normal component of the electric field is uniformly zero on the electrode. This picture is similar to what the linear theory predicts.

101

However, when and * are large ( = 25 , * = 0.01 ), the steady state configuration of electric field lines is quite different (Fig. 4.9b). In this case, the electric field lines have a large normal component on the entire electrode surface, especially close to the edges. This leads to loss of tangential electric field and consequently to loss of slip velocity (Fig. 4.10). Fig. 4.10 shows the normalized slip velocity, us* , on the gate electrode. us* is linear in the former case but has a surprising profile in the latter case. In the latter case, us* is zero on the edges and peaks close to the center at a point on either side of the center. This is because the tangential field is zero on the edges and maximum in the center of the gate.

y* x*
(a) =0.1, *=10-4

y* x*
(b) =25, *=0.01

Fig. 4.9: Steady state electric field lines in the vicinity of a flat electrode subjected to a DC electric field directed from left to write. (a) For very thin double layer and a very low induced zeta potential (i.e. * = 104 and = 0.1 ), the electric field is completely tangential to the electrode. (b) For a thick double layer and a large induced zeta potential ( * = 0.01 , = 25 ), surface conduction become

102

important. Healing regions are created and the normal component of the electric field becomes large so that the nonuniform surface currents can be maintained.

=0.1, *=10-4 =25, *=10-2

Fig. 4.10: Steady-state slip velocity on the gate electrode. Two cases are shown, (1) * = 104 , = 0.1 and (2) * = 0.01 , = 25 . In the first case, the normalized slip velocity is much higher than the second case. The loss in normalized velocity in the second case is due to surface conduction which becomes prominent at high zeta potentials (i.e. large ) and in thick double layers (i.e. large * ).

4.6.3 Results of Parametric Study

* Fig. 4.11 shows the maximum normalized slip velocity, umax , on the gate

electrode as a function of and * . For * = 0 , i.e. an infinitely thin double

103

* layer, umax is independent of , indicating that surface conduction becomes * unimportant for infinitely thin double layers. For non-zero * , however, umax is a * strong function of . For * > 0 , umax decreases with increasing , implying that

the importance of surface conduction increases as induced zeta potentials increase.

* Similarly, as * increases at a fixed , umax decreases. This indicates that surface

conduction effects grow as * increases. The double layer thickness is inversely proportional to ( ) . Therefore, thin double layers take place in low conductivity
1/ 2

solutions. When the bulk conductivity is low, the ratio of excess surface conductivity to bulk conductivity is high, yielding a high Dukhin number.

* * =0

1 0.8

u* max

0.6 0.4
=0.1
*

=1E-4 * =3.16E-4
* =1E-3 * =3.16E-3 *

0.2 10
-1

=0.01 * =0.0316

10

10

10

* Fig. 4.11: Maximum normalized velocity umax as a function of the parameters * and * . umax decreased with both and * , because surface conduction increases as either of or * increases

104

4.6.4 Normalized Quantities

In the last section, we examined the normalized slip velocity (see (4.37)) as * and were varied. Now lets examine some other normalized quantities defined in table 4.1.

Table 4.1: The normalized quantities for ICEO on a flat electrode Quantity Zeta potential Tangential electric field Normal electric field Slip velocity Surface conduction flux Normalized Expression 2 * 1 * x* 1 * y* 2 * * 2 x* 1 * Du x*

105

The following results were obtained for a constant * ( = 0.01 ) while was varied. Fig. 4.12 shows the variation of Dukhin number on the gate electrode for several values of . Dukhin number increases with increasing , indicating the increasing importance of surface conduction.

=25

=10

=1
Fig. 4.12: Variation of Dukhin Number on the gate electrode for = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The Dukhin number increases with indicating the increasing importance of surface conduction. The Dukhin number is zero at the center of the gate because the zeta potential is zero at the center

106

Fig. 4.13 shows the normalized zeta potential, 2 * , on the gate electrode for several values of . For < 1 , 2 * has a linear variation, with a maximum value of 1, matching the predictions of the Debye-Huckel theory. As increases,
2 * decreases. This indicates that surface conduction creates a loss in zeta

potential.

=25 =10 =1

Fig. 4.13: Variation of normalized zeta potential, 2 * , on the gate electrode for = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The normalized zeta potential decreases as increases. The loss in zeta potential is due to increased surface conduction at high values of .

107

* Fig. 4.14 shows the normalized tangential field, Ex , on the gate electrode. * For < 1 , Ex has a uniform value of 1 over the entire gate electrode, indicating

that, in the absence of surface current, the field outside the double layer is
* completely tangential and equal to the applied field. At higher values of , Ex

goes to zero at the edges, indicating that the field lines are drawn into the double
* layer at the edges. As a result, Ex is reduced significantly in the healing regions * close to the edges. In the center, however, Ex attains a large value.

=1

=10 =25

* Fig. 4.14: Variation of normalized tangential electric field, Ex , on the gate electrode for = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The normalized tangential field decreases as increases. The loss in the tangential field is due to increased surface conduction at high values of .

108

Fig. 4.15 shows the normalized normal electric field, E * , on the gate y electrode. For < 1 , E * is uniformly zero on the entire gate electrode y indicating absence of any substantial bending of field lines into the double layer. As grows, E * grows too, indicating that a lot of lines are drawn into (or y expelled out) of the double layer.

=10 =1

=25

Fig. 4.15: Variation of normalized normal electric field, E * , on the gate y

electrode = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The normalized normal field increase as increases. For large values of , surface currents become strong and force the electric field to acquire a large normal component. Also notice that the normal field flips sign of either side of the center. The points at which the flip takes place mark the locations where the ions start escaping out of the double layer.

109

Fig. 4.16 shows the normalized slip velocity, u* , on the gate. For < 1 , the slip velocity varies linearly on the gate, with the maximum value of 1, matching with the Debye Huckel theory. As increases, u* goes to zero at the edges and is decreased at other points too (especially close to the edges). As increases, the location of the maximum velocity shifts towards the center. Interestingly, u* is large in a small region on either side of the center of the gate. This is because the tangential field is large in those regions.

=1

=10 =25

Fig. 4.16: Variation of normalized slip velocity, u* , on the gate electrode for = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The normalized slip velocity decreases as increases. The loss in the velocity is due to increased surface conduction at high values of .

110

* The normalized surface conduction flux, DuEx , is shown in Fig. 4.17. * * DuEx increases with . DuEx is zero not only on the edges but also at the

center of the gate. Its zero at the center because * is zero there. Note that Du is
* also zero at the center. On either side of the center, DuEx first increases from

zero at the edge, attains a maximum and then falls back to zero at the center. This results in the surprising pattern of electric field lines shown in Fig. 4.9b. The location of the maxima corresponds to the point where the normal electric field flips sign. This is the point where the ions start leaking out of the double layer.

Normalized Surface Conduction Flux

=25

=10

=1
* Fig. 4.17: Variation of normalized surface conduction flux, DuEx , on the gate electrode for = 25, 10, 7.5, 5, 2.5, 1 and a constant value of * ( = 0.01 ). The normalized surface conduction flux increases as increases. The flux is zero not only at the two edges but also at the center of the gate. The flux attains maxima at a point between the center and the edge. The location of the maxima also marks

111

the location at which the ions start escaping out of the double layer. The nonuniformity of the flux is the main reason behind the surprising electric field lines shown in Fig. 4.9b.

4.6.5

Streamlines

Fig. 4.18 shows fluid flow streamlines for = 0.1 and 25 at a constant value of

* ( = 0.01 ). For = 0.1 , the streamline density is highest at the two edges because
the slip velocity is highest at the edges. For = 25 , the stream line density shifts towards the center because the slip is highest at a point close to the center on either side of the center (see Fig. 4.16); at the edges the velocity is zero due to surface conduction.

(a) = 0.1

112

\ (b) = 25 Fig. 4.18: Flow stream lines for * = 0.01 and (a) = 0.1 and (b) = 25 . For = 0.1 , surface conduction is negligible and the streamlines show a linear flow profile. At = 25 , the streamline density shifts towards the center of the gate indicating that the region of slip shrinks as surface conduction grows.

4.7

ICEO on a Metal Cylinder

ICEO flow over a cylinder is of interest in many practical applications. Consider a metal cylinder of radius a with its center at the origin (Fig. 4.19). When a horizontal DC electric field of magnitude E0 is applied, double layer forms on the cylinder surface and ICEO flow is produced (Figs. 4.20a-c). In an ideal case when none of surface conduction and faradaic injections is present, the double layer gets completely charged. In the steady state and ignoring the Stern layer, the zeta

113

potential, the tangential electric field and the slip velocity on the cylinder surface are given by [41]

( ) = 2 E0 a cos ,
s E = 2 E0 sin ,

(4.41) (4.42) (4.43)

us = 2

2 E0 a sin 2 s ,

where s is a unit vector tangential to the cylinder surface.

E0 y
s n

Fig. 4.19: A metal cylinder subjected to an external electric field.

114

E0

E0

+ + + + + + -++ + ++ + + + + + + + ++

--

+ ++- - +++ -++++ + -++ +

++ - + ++ +++++ + + -++ -

(a)

(b)

(c)
Fig. 4.20: ICEO on a metal cylinder. (a) When a horizontal DC field of magnitude E0 is applied at time t=0, cylinder surface forms an equipotential surface leading to induction of charge on its surface (also called polarization). However, since the cylinder is kept in an ionic solution, the ions travel towards the surface of the cylinder. (b) In the steady state, the ionic cloud shields the cylinders surface charge completely and insulates it from the electric field. As a result, all the electric field lines become tangential (shown by field lines). (c) The double layer is moved by the tangential electric field and thus a left-right, top-bottom symmetric ICEO flow is produced around the cylinder (shown by streamlines).

115

4.7.1 Surface Conduction Model for a 2D Cylinder in Cartesian Coordinates

The double layer model described in section 4.5 can be implemented on a cylinder surface as well. We need to transform the model in Cartesian coordinates so that it can be easily implemented in commercial software packages such as COMSOL. Lets derive the dimensionless steady state double layer PDE for a 2D cylinder in Cartesian coordinates. The double layer PDE for a cylindrical surface can be simply written as * * * Du * = 0 , n* s s (4.44)

where n* and s* represent the normal and tangential derivatives on the surface respectively. Now consider a 2D cylinder of dimensionless radius 1 with its center at the origin. The normal and tangential vectors for the surface of the cylinder are defined as
n cos = s sin

sin x . y cos

(4.45)

However, for this particular geometry, cos = x* and sin = y* and therefore,
n x* = * s y

y x*

x . y

(4.46)

Using the preceding definition of the normal vector, the normal derivative of a quantity on a 2D cylindrical surface can be obtained as

116

 * * * = n * * = x* * + y* * . * n x y Similarly, the tangential derivative can be obtained as * * * = s * * = y* * + x* * . s* x y

(4.47)

(4.48)

More complicated expressions can also be calculated in similar fashion; for example the tangential divergence is obtained as,
* * * * . +x Du * = y* * + x* * Du y* s* s x y x* y*

(4.49)

Combining the preceding three equations with (4.44) and doing some algebra yield the double layer PDE in Cartesian coordinates, which is simply written as, n * * * ( DuJ s ) = 0 , where
n = x* x + y* y ,

(4.50)

(4.51)

and
J s = y*2

* * * * *2 * * * * x+x y. x y x y x* y* y* x*

(4.52)

4.7.2 Geometry and Boundary Conditions

We have solved the problem only on a half cylinder because of symmetry (See Fig. 4.21). Consider a half cylinder of radius a placed at the bottom wall of a

117

chamber of side L. The chamber side is much larger than the radius of the cylinder, L*=L/a=20.

n * * = 0, u* = 0
L*

*2 * = 0

*2u* * p* = 0

u* = 0
* =
L* 4

* u* = 0

u* = 0

* =
u * ( ) = 2

L* 4

=
*

* * *

n * * = 0
p*=0

Flow Symmetry

y
1 a

Flow Symmetry

n * * = 0
Js = 0

Js = 0

n * * * ( DuJ s ) = 0
Fig. 4.21: Surface conduction model for a 2D metal cylinder. Only half the cylinder is simulated due to top-bottom symmetry. The chamber is much larger than the cylinder, i.e. L*=L/a>>1. In nondimensional units, the radius of the cylinder is 1.

When dimensionless potentials of * = L* 4 are applied on the left and right

* walls respectively, a DC electric field E0 = 2 is produced away from the

cylinder. In the linear regime, this electric field causes a zeta potential of

* = cos on the cylinder surface (refer to (4.41)). In dimensional form, these

118

estimates can be written as E0 TH 2 a and TH . Using these estimates, a natural scale for normalizing the slip velocity can be obtained as uHS =
2 E0 2TH = , 2 a

(4.53)

using which we can define the normalized slip velocity on the gate as
u* (x) =

1 2 ( s ) s = 2 * * * s s . uHS

(4.54)

* Here we would like to introduce a new variable us for the magnitude of the

slip velocity, given as

* us =

* * * s .

(4.55)

Using the definition of s from (4.45) and combining (4.54) and (4.55), the

velocity on the cylinder surface can be written as

* * u* (x) = us s = us y* x + x* y .

(4.56)

As mentioned above equal and opposite potentials are applied to the two driving electrodes ( * = L* 4 ); the cylinder surface is kept floating; the floating
* potential on the cylinder surface vanishes i.e. el = 0 because it is placed at equal

distances from the two driving electrodes. One consequence of zero floating potential is that the bulk boundary condition becomes

* = * .
The condition on the two poles (i.e. = 0 and ) of the half cylinder is
s Du* * = 0 ,

(4.57)

(4.58)

119

which is equivalent to

Js = 0 .

(4.59)

4.7.3 Electric Field Lines around Cylinder

Fig. 4.22 shows the steady state electric field lines around the cylinder respectively for = 1, 20 and 100 at a constant * of 0.001. For low induced zeta potential (i.e. = 1 ), the electric field lines are completely tangential to the cylinder surface which is equivalent to the linear regime; For = 20 and = 100 , a lot of electric field lines terminate perpendicular to the cylinder showing large healing regions due to surface conduction. The extent of healing region grows as

increases. For = 100 , the healing region extends up to a short angle from the
equator (i.e the top or = / 2 ) of the cylinder. This shows that in presence of strong zeta potentials, the healing regions can be as big as the dimension of the cylinder itself.

120

(a) Electric field lines around cylinder for = 1

(b) Electric field lines around cylinder for = 20

121

(c) Electric field lines around cylinder for = 100 Fig. 4.22: Electric field lines around the cylinder for * = 0.001 and = 1, 20 and 100 respectively . For = 1 , the lines are perfectly tangential to the cylinder, indicating that surface currents are negligible. For = 20 , the field lines acquire a large normal component and give rise to a healing zone between the poles and the equator of the cylinder. For = 100 , the healing zone is very large and extends upto a short angle from the equator of the cylinder. At the equator, the lines are expelled out because of nonuniform surface currents.

4.7.4 Streamlines around Cylinder

Fig. 4.23 shows flow streamlines respectively for = 0.01 and = 20 at a constant * of 0.001. For = 1 , fluid slip at the entire cylinder surface is apparent. For = 20 , the stream line density shifts towards the equator (i.e. = / 2 ). This shows that the region, in which the slip is substantial, shrinks with increasing surface conduction.

122

Fig. 4.23(a): Flow streamlines around the cylinder for = 1

Fig. 4.23(b): Flow streamlines around the cylinder for = 20 Figure 4.23: Flow streamlines around the cylinder for (a) = 1 and (b) = 20 . For = 1 , the fluid slip all around the cylinder. For = 20 , the region of slips shrinks significantly and the streamline density shifts towards the equator of the cylinder..

123

4.7.5 Normalized Quantities

Similar to the flat electrode case, lets examine various normalized quantities on the surface of the cylinder as a function of . Various normalized quantities are defined in table 4.2.

Table 4.2: Normalized quantities on a metal cylinder Quantity Zeta potential Tangential electric field
Normal electric field Slip velocity Surface conduction flux

Normalized Expression

1 1

s * *

n * * 2 * s * * 2

Du

s * *

Fig. 4.24a shows Dukhin number for several values of and a constant * of 0.001. Dukhin number increases as increases. Its always zero at = 2 due to asymmetric nature of induced zeta potential. Figs. 4.24b-f show that, at very low values of (up to = 1 ), all the quantities correspond to the linear estimates (refer to (4.41), (4.42), (4.43)). With the increase of , the normalized tangential electric field decreases close to the poles but increases at the equator; the normalized normal electric field increases everywhere but flips sign a point between the pole and the equator on either side of the equator; the normalized zeta potential decreases; the normalized surface

124

conduction flux increases and finally the normalized slip velocity decreases close to the poles but increases close to the equator. The surface conduction flux is zero the two poles and also at the equator. It attains a maximum between the equator and the pole, on either side of the equator. These plots indicate that at high induced zeta potentials, surface conduction becomes important and deteriorates electroosmotic slip velocity. Similar to the flat electrode case, the surface conductivity is zero at the equator of the cylinder and therefore, the ionic current leaks out of the double layer on either side of the equator causing the surprising electric field lines shown in Fig. 4.22.

Fig. 4.24 (a): Dukhin number on the cylinder for * = 0.001 and = 1, 10, 15, 20 and 30 . The Dukhin number increases with indicating the increasing importance of surface conduction. The Dukhin number is zero at the equator (i.e. = / 2 ) of the cylinder because the zeta potential is zero at the equator.

125

Fig. 4.24 (b): Normalized zeta potential on the cylinder for * = 0.001 and = 1, 10, 15, 20 and 30 . The normalized zeta potential decreases with . The loss in zeta potential is caused by the increasing surface conduction effect.

Fig. 4.24 (c): Normalized tangential electric field on the cylinder for = 0.001 and = 1, 10, 15, 20 and 30 . The normalized tangential field decreased close to the poles ( = 0 and ) but increases around the equator ( = / 2 ) as the values of increases. The loss in the field is caused by the increasing surface conduction effect.
*

126

Fig. 4.24 (d): Normalized normal electric field on the cylinder for * = 0.001 and = 1, 10, 15, 20 and 30 . The normalized normal electric field increases as increases. Increased surface conduction causes the field to acquire a large normal component. The field flips sign between a pole and the equator due to nonuniform surface currents around the cylinder surface.

Fig. 4.24 (e): Normalized slip velocity on the cylinder for * = 0.001 and = 1, 10, 15, 20 and 30 . The slip velocity decreases as increases. The region of

127

significant velocity shrinks to a minimum as increases. The location of maximum slip velocity shifts towards the equator indicating the flow region shrinks towards the equator. The loss in velocity is caused by the increased surface conduction.

* = 0.001 and = 1, 10, 15, 20 and 30 . The flux increases with indicating the

Fig. 4.24 (f): Normalized surface conduction flux on the cylinder for

increasing importance of surface conduction. The flux is zero not only at the poles but also at the equator of the cylinder. The flux acquires a maximum between a poles and the equator. The nonuniformity of the flux causes the surprising electric field gradients shown in Fig. 4.22.

4.7.6 Parametric Study

The parameters of study are * and . Fig. 4.25 shows the normalized slip
* * velocity us at = / 4 as a function of the parameters * and . For * = 0 , us

is independent of indicating that surface conduction is unimportant for

* infinitesimally thin double layers. As * increases, us decreases. Similarly, as

128

* increases, us decreases. Increasing surface conduction causes the loss of slip * velocity. At very high induced zeta potentials i.e. > 50 , us decays almost * quadratically with i.e us 2 (focus on the straight line behavior on the log-

log plot for > 50 ). Note that the normalizing scale for slip velocity grows
2 quadratically with ( uHS = 2TH 2 a ). Multiplying the dimensionless velocity * 2 ( us 2 ) with its scale ( uHS = 2TH 2 a ) will yield the slip velocity which is

independent of . In other words, for > 50 , the velocity stops growing with and becomes saturated. Increased surface conduction arrests any further increase in the slip velocity. This behavior was also observed in the experiments of chapter 3.

* Fig. 4.25: Normalized slip velocity us at = / 4 as a function of the parameters * and . The values of * increase in the direction of the arrow, in * the following order: 0, 0.0001, 0.001, 0.01, and 0.1. us decreases with both * * and because of increasing surface conduction. For > 50 , us scales as

129

* us 2 , shown as a straight line on the log-log plot. This indicates that at high values of , the real slip velocity saturates and does not grow with increasing voltages. The further growth of the velocity is stopped by surface conduction.

4.8

Concentration Polarization

In our preceding analysis, we have ignored the concentration gradients in the bulk. We assumed that the salt concentration is uniform in the entire bulk. However, when zeta potential is high, surface transport of ions can cause gradients in the bulk salt concentration. The length scale of these gradients can be as big as the healing length of the bulk electric field. This phenomenon is generally known as concentration polarization [61, 62]. Fig. 4.26 shows a fundamental picture of how surface conduction can lead to concentration polarization. At the two edges of the electrode, the bulk electric field not only drives counter-ions into the diffuse layer but also drives away the co-ions, leading to a salt sink. In the middle, however, the counter-ions are reunited with co-ions leading to salt source. As a result, a high concentration zone is set up in the middle of the electrode whereas a salt depletion zone at the two edges. The difference in the concentration drives a chemi-osmotic flow. This flow is in a direction opposite to the electroosmotic flow and can slow the flow down.

130

Salt Sink

Salt Source

Salt Sink
+ +

- + + + + + +

- + +

+ +

- -

- - - - - - + + + + ++
Fig. 4.26: Surface conduction leads to concentration polarization. The electric field drives counter-ions into the double layer, but at the same time the coions are repelled away, creating a salt sink at the two edges of the electrode. In the middle of the electrode, however, the counter-ions are reunited with the co-ions leading a salt source. This leads to gradients in salt concentration in the bulk. For large , the gradients can extend into the bulk, giving rise to a chemi-osmotic slip velocity which opposes electroosmotic slip velocity.

- D

To account for concentration polarization, we need to solve the convectiondiffusion equations for the salt concentration.

c + u c = D 2 c . t

(4.60)

The electrostatic potential follows the following current conservation equation:

(c ) = 0 .

(4.61)

The modified Stokes equations for fluid flow and the effective boundary conditions for the excess double layer charge and excess double layer salt concentrations are given in chapter 7. The net slip velocity is the sum of electroosmosis and chemi-osmosis

uslip

4 kT ze = ln cosh x ze kT

1 c . c x

(4.62)

131

For large , chemi-osmosis can be of the same order as electroosmosis. However, we have not accounted for chemi-osmosis in our model. Directions to incorporate chemi-osmosis are given in chapter 7. An approximate dimensionless solution for the salt concentration was obtained by neglecting the convection of the salt and by neglecting the effect of salt gradients on the bulk electric field. Fig. 4.27 shows the salt concentration in the bulk for = 20 . A high concentration zone is established above center of the gate whereas salt depletion zones are established near the edges. The initial concentration in the bulk was 1.

Fig. 4.27: Concentration polarization due to surface transport. The color shows the concentration whereas the streamlines show the electric field. A high concentration zone is established above center of the gate whereas salt depletion zones are established near the edges. The initial concentration in the bulk was 1 and = 20 . The concentration solution was obtained by neglecting convection and the effect of concentration gradient on the electric field.

132

4.9

Conclusions

A qualitative and quantitative picture of surface conduction effects in ICEO has been developed via numerical simulations. It has been found that the nonuniform induced zeta potentials, which are inherent in ICEO, lead to nonuniform surface currents and cause surprising gradients in the bulk electric field. Healing regions, in which the electric field has large normal component, were found at the corners as well as in the middle of the inducing surface. We predict that surface conduction creates large normal electric fields and thus reduces the tangential field and the slip velocity. A numerical model for simulating surface conduction in steady and time dependent cases has been presented. In this model, the electric double layer is handled by solving charge conservation PDE on the electrode surface. A variable substitution has been presented for improving the convergence of the model. The new model has excellent convergence properties and can be used for solving electrokinetic problems for very high induced zeta potentials (100-1000 times higher than the thermal voltage). The model also allows for nonlinear surface capacitance.
4.10 Appendix to Chapter 4

4.10.1 Comments on COMSOL Multiphysics Simulations

A commercial finite element package COMSOL Multiphysics (Comsol Inc., Stockholm, Sweden) was used for solving the mathematical model described

133

above. Lagrange quadratic elements were used for all the dependent variables. Since the problem deals with DC electric fields, a stationery solver was used with zero initial conditions for all the variables. A Direct (UMFPACK) solver was chosen for solving the system of linear algebraic equations. First the electrostatics and the double layer PDE were solved simultaneously. Then their solution was used as a linearization point for the bulk fluid flow problem. The maximum element size in the bulk was chosen to be 0.1. The mesh element size on the ICEO electrode boundary was chosen to be 10-4. This ensures that a mesh independent solution is achieved. Mesh independence is demonstrated in Fig. 4.28. The normalized kinetic energy has been plotted as function of * and the size of the mesh element on the flat gate electrode, h. The kinetic energy, KE, is defined as the integral of the square of the slip velocity on the gate electrode.

KE(h) is divided by KE(hmin) to obtain the normalized kinetic energy. The

minimum element sized was hmin=5x10-5. Plot shows that the solution is almost mesh independent at h=10-4.

134

Normalized Kinetic Energy KE(h)/KE(hmin)

0.95

=10 =10 =10

* * * *

-4 -3 -2 -1

0.9

=10 =1

0.85

0.8 10
-1

10

-2

10

-3

10

-4

10

-5

Mesh Element Size on the Gate Electrode, h

Fig. 4.28: The normalized kinetic energy as a function of * and the element size, h. Normalized KE becomes mesh almost mesh independent for h<10-4.

Weak non-ideal constraints were used for handling the electrostatic dirichlet boundary condition ( * = * ). The weak non-ideal constraints remove bidirectional coupling between the bulk and the surface electrostatic fields and therefore give the most accurate solution for this situation. If one uses weak ideal constraints instead, a slight quantitative difference arises in the results, but the results do not change qualitatively and all the conclusions stay exactly the same. Weak ideal constraints cause this slight error in the results because of bidirectional coupling between the two fields. However, the weak non-ideal constraints make the problem harder to solve because of unidirectional coupling. To mitigate this

135

problem, one should first solve the problem with weak ideal constraints and then use this solution to as an initial condition to find the final solution with weak nonideal constraints. Its easier to achieve convergence with weak ideal constraints because of its bidirectional nature. For a description of various weak constraints in COMSOL, please refer to the COMSOL manuals available from Comsol Inc. (Stockholm, Sweden). For a mathematical introduction to weak solutions, the reader can refer to [74, 75]. For the convenience of the reader, however, we give a brief introduction to the weak forms and weak solutions in this appendix. We have also derived the weak form of the double layer PDE so that it can be directly implemented in any finite element code.
4.10.2 Weak Form for Finite Element Solution

Finite element methods generally seek a weak solution to the PDE in question. As an example consider Poisson equation,
2u (x) = f (x),

(4.63)

Suppose H is an infinite-dimensional function space that is rich enough to include in its closure all functions that may be of interest as candidates for the solution u. For all v H , we define the defect d (v) = 2 v f . Then we say that

w is a weak solution to the problem equation (4.63) provided that the defect d(w) is
orthogonal to the entire function space, that is, if

vd (w)dx = 0

v H .

(4.64)

136

v is generally referred to as a test function. Equation (4.64) is also referred as the weak form of (4.63). For more information, the reader is referred to [74, 75]. We can further simplify (4.64) as

v wdx vfdx = 0 .
2

(4.65)

The first term on left hand side can be simplified to

v wdx = ( vw) dx v wdx .

2

(4.66)

Applying divergence theorem on the first term on the right hand side yields v wdx = vn wds v wdx ,
2

(4.67)

where is the boundary of . The boundary integrals can be handled by invoking the boundary conditions. For example, if a Neumann boundary condition

is specified on the boundary, i.e., n w = b( x) , then we can simply substitute it in

the boundary integral. Dirichlet conditions are handled by implementing constraints [75].
4.10.3 Weak Form of Double Layer PDE

To implement the charge conservation equation in COMSOL, its necessary to derive its weak form. When doing a steady state analysis, weak form is simply given as (refer to (4.32)) s vn ds + s Dus v s
* * * * *

ds = 0

(4.68)

where s represents the surface and v is a test function for * . Note that the
edge condition nl Du* * = 0 was invoked in deriving (4.68). s

137

For a time dependent analysis, however, the weak form is given as (refer to (4.31)),

n * * * ( Du* * ) * s s ds = v ds * s t cosh * 2

(4.69)

The second term on the right hand side of the preceding equation can be expanded as follows
( Du )
* s * s *

cosh * 2

ds =

* v * * v tanh * 2 * * * * ( ) s s s s v 2 ds * Du* * ds Du t s s cosh ( * 2 ) cosh ( * 2 )

(4.70) The first term on the right hand side of the preceding equation can be simplified by applying the divergence theorem

* s

cosh

2
*

Du* * ds = s

vDu s nl * *dl * cosh 2

(4.71)

s However, we have s * * = 0 on the edges and therefore

* s s

cosh

( 2 )
*

Du* s

ds = 0

(4.72)

Finally combining (4.69), (4.70) and (4.72) yields the desired weak form,
vn * * v * ds = ds s t s cosh( * 2) s
*

v Du * * * v tanh( * 2)* * * * s s s s 2 ds . cosh( * 2)

(4.73)

138

139

Chapter 5. Simulations vs. Experiments

In this chapter we report the results of our nonlinear numerical simulations which were intended to simulate the conditions of our experiments. We show that the nonlinear effects indeed help us reduce the discrepancy between the simulations and the experiments. In chapter three, we had reported that the linear simulations predict two to three orders of magnitude higher velocities than the experiments. In this chapter, we reduce that discrepancy by one order by employing the nonlinear model incorporating the nonlinear surface capacitance and surface conduction.
5.1 Numerical Model

Since the experiments were performed under AC electric fields, the time dependent equations from section 4.5 were used. For the convenience of the reader, the time dependent numerical model is being repeated below and is portrayed in Fig. 5.1. The simulation geometry represents a two dimensional cross section of the ICEO chamber used in our experiments. The geometry has been nondimensionalized with the width of the gate electrode ( a = 2 104 m ). In the bulk, we solve the following equations, *2 * = 0 ,
* p* + *2u* = 0 ,

(5.1) (5.2) (5.3)

* u* = 0 .

140

*2 * = 0

*2u* * p* = 0

* u* = 0

y
n = 0

* = 0.50* sin(2 f *t * )
Driving electrode 1

n = 0

* = *
Gate

n = 0

* = 0.50* sin(2 f *t * ) n = 0

Driving electrode 2

p* = 0

* Du * = 0 x

Du

* =0 x*

u* = 0 on all boundaries except the gate

* * * * 1 = * * ( Du * * ) * * t y cosh 2 y y

* us =

* x* * where = 0 10 1
2

Fig. 5.1: The dimensionless geometry and the time-dependent numerical model incorporating surface conduction and nonlinear surface capacitance. All dimensions shown to the scale

On the chamber walls, we apply insulation and no slip boundary conditions.

n * * = 0 ,

(5.4) (5.5)

u* = 0 .

On the two driving electrodes we specify sinusoidal potentials and a no slip boundary condition,

* = + =
*

0*
2

sin ( 2 f *t * ) On left driving electrode sin ( 2 f t

* *

0*
2

) On right driving electrode

u* = 0 ,

(5.6)

(5.7)

* where 0 is the dimensionless amplitude of the driving voltage and f * = f C is

the dimensionless driving frequency.

141

On the gate electrode, the bulk electrostatic potential is related to its zeta potential and is given as
* * = el * ,

(5.8)

* In case of symmetric ICEO flow, we set el = 0 . We solve the following time

dependent PDE on the gate electrode

* * 1 = * * ( Du * * ) , t * x cosh * 2 y x

(5.9)

with the following conditions on the edges of the gate electrode Du where
Du = 4 * (1 + m ) sinh 2 * 4 ,

* =0, x*

(5.10)

(5.11)

where m is a dimensionless parameter indicating the relative contribution of electroconvection to surface conduction. For aqueous solutions of KCl at room temperature, m 0.45 . Finally, the time dependent normalized slip velocity on the gate electrode is given by
* us (t ) =

* 2

* x*

(5.12)

* where = 0 10 . This value of was chosen because in this particular

* geometry, the linear scale for maximum instantaneous * is equal to 0 10 (= ).

The rms (root mean square) * is approximately 0* 10 2 ( =

2 ) and the

142

* dimensionless rms electric field is approximately 0 5 2 ( = 2 ). Product of * these two quantities yields (0 10 ) which is equal to 2 . Using the rms zeta 2

potential and the rms electric field, we can get the following scale for the velocity uHS =
2 2TH . a

(5.13)

Scaling the slip velocity with the preceding scale yields (5.12).
* The time averaged slip velocity, us

can be obtained by averaging the slip

velocity over the duration of the solution,

* us =

1 Ts*

Ts*

* us dt * ,

(5.14)

where Ts* is the dimensionless time up to which the solution is obtained. The solution will be periodic after a short transient period. During the transient period, the solution develops and achieves periodicity. The transient period is generally shorter than one period of the AC cycle.
5.2 Depth Averaging

Before we compare our numerical results with the experiments, we would like to point out a particular feature of PIV experiments. In PIV experiments, entire volume of the fluid is illuminated uniformly. The particles in the object plane (focal plane) are in perfect focus and contribute highest to the PIV image but the particles which are away from the object plane are also visible, of course out of focus. They might not be perfectly focused but they do contribute to the PIV

143

image. Hence, PIV does not yield the velocity field at a single plane; instead it yields a certain weighted average of the velocity over the depth of the interrogation region. As a result, the measured velocity is slightly different from the real velocity in the focal plane. The relative contribution of the light from a particle away from the focal plane decreases as its distance from the focal plane increases. In other words, the highest contribution to the velocity measurement comes from the focal plane. Lets define a parameter PIV indicating the relative contribution of the particles which are a distance y away from the focal plane to the velocity measurement. The weighted depth-averaged velocity, u0 , can then be represented as,
u0 ( x , z ) =

u ( x, y, z ) ( y)dy . ( y)dy
PIV PIV

(5.15)

If we arbitrarily set PIV ( y = 0) = 1 , then according to [76],

PIV =

1 3 y 2 1 + ycorr
2

(5.16)

where ycorr is an experimental parameter (called the depth of correlation) and depends upon the properties of the imaging system (the numerical aperture, NA, and the magnification, M) and the diameter of the particles, dp [76]. Depth of correlation is defined as the axial distance from the object plane in which a particle becomes sufficiently out of focus so that it no longer contributes significantly to

144

the signal peak in the particle-image-correlation function. From (5.16), we can see that indeed PIV ( ycorr ) = 0.01 (i.e. the contribution decreases to 1/100th at y=ycorr). We want ycorr to be as small as possible so that the maximum contribution comes from the focal plane (y=0) and the contribution from other planes is minimized. ycorr can be made small by using a lens with high NA and by using small particles. In our experiments, we have used a 10x lens with NA=0.25 and particles with diameter of dp=0.7 m. For these conditions, following the analysis of [76], we determine that ycorr 18.58 m and thus we can plot PIV for our experiments (see Fig. 5.2).
1 0.8 0.6
PIV

0.4 0.2 0 -3

-2 -1 0 1 2 Distance from the focal plane (m)

3 x 10
-5

Fig. 5.2: Relative contribution of particles to the velocity measurement as a function of their distance from the focal plane for NA=0.25, M=10 and dp=0.7 m. The contribution of a particle decreases as its distance from the focal plane increases. The depth of correlation, ycorr , for these experimental conditions is approximately 18.58 m. Note that PIV ( ycorr ) = 0.01 .

145

Depth averaging in PIV has significant consequences for our experiments. First, the depth of correlation is much larger ( ycorr 18.58 m) than the double layer thickness (10-50 nm). Second, the particles used in our experiments are also much larger (0.7 m) than the double layer thickness. These two observations indicate that its impossible for us to measure the real slip velocity on the gate electrode. At best, we can measure a weighted depth-average of the velocity. Therefore, in order to draw a meaningful comparison between the experiments and the simulations, we should perform the weighted depth-averaging in our numerical simulations also.
5.3 Uncertainty in Diffusivity Values

In many of our experiments, we have used an aqueous fluid with a measured conductivity of = 17 104 S/m . We call this solution purified water because its a 50:1 mixture of DI water ( = 0.055 104 S/m ) and a proprietary particle solution obtained from Duke scientific, Fremont, CA. The resultant conductivity of this solution is clearly due to the content of the particle solution because the original conductivity of the DI water is very low ( = 0.055 104 S/m ) and the conductivity rises to the reported high value ( = 17 104 S/m ) as soon as the particle solution is added. We specifically dont know what ions are present in the particle solution. The knowledge about ions and specifically their diffusivity is critical to determining the double layer thickness, D = D and the charging time, C = D a D . These parameters are important because D is used in

146

determining Dukhin number and c determines if the double layer can charge fast enough in an AC electric field. Since, we dont have the accurate knowledge of D, we exercise the freedom to make assumptions about ions present in the solution. We have considered two different cases in our simulations: Case 1: Assume that the present ions are K+ and Cl-. We measured the pH of water before and after adding the particles and found that the two measurements are very close to each other (5.7 and 5.46 respectively). This implies that particle solution probably does not contain any acidic or basic material; instead it might contain a salt which does not alter the pH of water. Based on this observation, we assume the presence of K+ and Cl- because KCl is a common salt in chemistry. The diffusivities of K+ and Cl- ions are respectively 1.957 10-9 and 2.032 10-9 m2/s [65]. The average of the two, D = 1.995 109 m2/s, was used in our simulations. Or
-1 Case 2: Assume that the present ions are H+ and HCO3 . According to [66],

when DI water is exposed to atmosphere, it absorbs a lot of CO2 and forms carbonic acid, H 2 CO3 . This is evident from the pH values (5.46-5.7). H 2 CO3
-1 solution is generally dominated by H+ and HCO3 ions. The diffusivities of H+ and

-1 HCO3 ions are respectively 9.31110-9 and 1.185 10-9 m2/s [65]. Following the

analysis of ref. [66], we use the average of the two ionic diffusivities (i.e. D =5.248 10-9 m2/s) in our simulations.

147

Using the preceding two values of average D and the measured conductivity

= 17 104 S/m , we calculate D = 2.885 108 and D = 4.679 108 m

respectively. The characteristic time scales are

c = 2.892 103 s

and

c = 1.783 103 respectively. The second case yields a thicker double layer but its
charging time is shorter. In other words, the second case suffers from higher amount of surface conduction but at the same time has a faster charging dynamics. These two effects counter act against each other. Therefore, we expect that the results of the simulations are not so much affected by the exact value of D. Results of the simulation for these two values of D will be presented next.
5.4 Results

The preceding nonlinear numerical model was solved in COMSOL (Comsol Inc., Stockholm, Se). A time dependent solver was used for solving the electrostatic problem. The electrostatic solution was used for calculating the time
* averaged slip velocity us . The flow solution was then obtained with a stationary

* solver using us as a boundary condition. The linear algebraic system was solved

with a direct (UMFPACK) linear system solver. Mesh was created by using a maximum element size of 5e-4 on the gate boundary, 1e-5 on the two edges of the gate and 0.05 in the rest of the domain (see Fig. 5.3). Non-ideal weak constraints
* were used for handling the dirichlet conditions ( * = el * , and u* = us* ) on the

gate electrode.

148

Gate Electrode

Fig. 5.3: Mesh inside the simulation geometry. The mesh was highly refined on the gate electrode and on the edges of the gate electrode. The mesh was also sufficiently refined in the bulk for obtaining an accurate bulk flow solution.

First we present the results of simulations under the following conditions:

0 = 10 V , f = 100 Hz , = 17 104 S/m , D = 1.995 109 m2/s at T = 200 C. The

values of various other constants were reported in chapter 3. These conditions yield the following:

D = 2.885 108 m ,

c = 2.892 103 s , 0* = 395.856 ,

f * = 0.289 , = 39.586 and * = 1.442 104 . Note that the characteristic zeta potential scale, , is almost 40 times higher than the thermal voltage. The solution was obtained for durations longer than the period of the AC signal so that the solution has time to develop fully and reach periodicity. In this case, the solution was obtained upto t * 14 which is four times longer than the AC signal period ( 1 f * = 3.458 ). Fig. 5.4 shows the dimensionless zeta potential, * , at the left edge of the gate electrode (i.e. x* = 0.5, y* = 0 ) as a function of time. The

149

solution is almost periodic for t * > 2 . Note that the maximum value of * is around 10, meaning that the zeta potential is 10 times higher than the thermal voltage. We can expect the nonlinear effects to be strong at such high zeta potentials.

Fig. 5.4: Temporal evolution of dimensionless zeta potential * , at the left edge of the gate electrode (i.e. x* = 0.5, y* = 0 ). The solution first goes through a short transient period ( t * < 2 ) and then achiever periodicity with the same period as the AC cycle. The periodic solution can be called fully developed. Results are shown for 0 = 10 V , f = 100 Hz , = 17 104 S/m and D = 1.995 109 m2/s at T = 200 C. Fig. 5.5 shows the streamlines for the time averaged flow. The flow is symmetric about the center and shows the general features of symmetric ICEO flow.

150

Fig. 5.5: Time averaged streamlines on top of the gate electrode for 0 = 10 V , f = 100 Hz , = 17 104 S/m and D = 1.995 109 m2/s at T = 200 C. The ICEO flow is symmetric about the center of the gate.

* Fig. 5.6 shows the time averaged value of the normalized slip velocity, us

for D =1.995 10-9 m2/s (i.e. for KCl), while the other parameters are the same as before. The velocity is zero on the edges which is a feature of surface conduction dominated flows (as explained in chapter 4). We have also shown the weighted depth-average of the velocity on the same figure. The weighted depth-average is lower than the slip velocity and will help in reducing the discrepancy between the simulations and the experiments. In Fig. 5.7 we show the slip and the weighted depth-averaged velocity for

D =5.248 10-9 m2/s (i.e. for HCO3), while the other parameters are the same as
before. A quick comparison between Figs. 5.6 and 5.7 reveals that the velocities in the latter case are slightly lower due to increased surface conduction. However, the difference between the two is small. The reason behind this was explained in section 5.3.

151

0.15 0.1 0.05

<u*>

Slip Velocity Weighted Depth Average

0 -0.05 -0.1

-0.5

0
x
*

0.5

Fig. 5.6: The dimensionless time averaged slip velocity on the gate electrode for D = 1.995 109 m2/s and 0 = 10 V , f = 100 Hz , = 17 104 S/m at T = 200 C. The weighted depth-averaging produces a lower velocity than the slip velocity.
0.15 0.1 0.05
<u*>

Slip Velocity Weighted Depth Average

0 -0.05 -0.1

-0.5

0
x
*

0.5

Fig. 5.7: The dimensionless time averaged slip velocity on the gate electrode for D = 5.248 109 m2/s and 0 = 10 V , f = 100 Hz , = 17 104 S/m at

152

T = 200 C. The weighted depth-averaging produces a lower velocity than the slip velocity.

5.5

Simulations vs. Experiments

Fig. 5.8 shows the experimental data along with simulation results for

0 = 10 V , f = 100 Hz , = 17 104 S/m and D = 1.995 109 m2/s at T = 200 .

The linear velocity is more than two orders of magnitude higher than the experimental velocity. The nonlinear simulation reduces the velocity by one order of magnitude. Depth averaging of nonlinear velocity reduces the discrepancy even further and brings the velocity within a factor of 6 of the experiments. Depthaveraging is important because a direct comparison of the numerical slip velocity with experimental data is not meaningful. In our experiments, we can not measure the real slip velocity; instead we measure a weighted depth-average of the velocity. Therefore, its more meaningful to compare the experimental data with the weighted depth-average of the numerical results.

153

10

-2

Linear simulation

Slip Velocity, us (m/s)

10

-3

Nonlinear simulation
-4

10

10

-5

Experimental data Depth averaged nonlinear simulation

10

-6

-1

0 Distance from the center, x (m)

1 x 10
-4

Fig. 5.8: Comparison of different estimates of ICEO flow velocity for 0 = 10 V , f = 100 Hz , = 17 104 S/m . The linear theory predicts two orders of magnitude higher slip velocity than the experiments. The nonlinear theory reduces the discrepancy by more than one order of magnitude. Depth averaging of the nonlinear flow field brings the velocity within a factor of 6 of the experimental data. Simulations (both linear and nonlinear) were performed for D = 1.995 109 m2/s.

5.6

Contribution of Dielectrophoresis

In the experiments, the flow tracing particles may be subjected to dielectrophoresis (DEP). In order to determine whether DEP is responsible for discrepancy between the numerical and experimental velocities, DEP force [26] was calculated from FDEP = 2 r 3{K } Erms ,
2

(5.17)

154

where r is the particle radius, {K } is the real part of the Clausius-Mossotti factor and E is the magnitude of the electric field. The difference between the particle velocity u p and the fluid velocity u f was calculated from Stokes law, ur = u p u f = FDEP 6 r , (5.18)

where ur denotes the difference between the particle velocity and the fluid velocity. Since the electric field varies significantly within the healing length, a good estimate for Erms
2

can be found by
Erms
2

1 0 , LH 2 L

(5.19)

where a is the length of the gate electrode, L is the gap between the driving electrodes, LH is the healing length predicted by (4.3) and 0 is the amplitude of
* the driving voltage. Taking max = 10 (from Fig. 5.4) and D = 28.85 nm (from

section 5.4), we estimate the healing length to be LH 6 m. Finally taking

0 = 10 Volt, a=200 m, L=800 m, r=350 nm & {K } 1 and combining (5.17),

(5.18) and (5.19), we find ur 0.4 m/s, which is approximately 2 orders or magnitude smaller than the experimental ICEO velocity (45 m/s). The contribution of dielectrophoresis to the particle velocity measurements can therefore be ignored.

155

5.7

Contribution of Electrothermal Flow

The electrothermal flow is generated when an electric field acts on a fluid with nonuniform conductivity and permittivity. Nonuniform electric fields cause nonuniform Joule heating of the fluid. As a result, the temperature of the fluid varies within the device and leads to gradients in fluid properties such as conductivity, permittivity, density, viscosity etc. The gradients in these properties are proportional to the gradients in the temperature. A simplified expression for the electrothermal force for water under low AC frequencies can be written as [26],
f E = 0.012T E0

E0 . 1 + ( ) 2

(5.20)

We have performed time averaged simulations (using steady state solvers) to obtain an approximate scale for the electrothermal flow velocity in our experiments. The simulations were performed for our experimental conditions:

0 = 10 V , f = 100 Hz , = 17 104 S/m , D = 1.995 109 m2/s at Tambient = 200 C.

The electrostatic potential is governed by the Laplaces equation. Other governing equations are shown later in this section along with a scaling analysis. The electrostatic boundary conditions are: insulation on the all non-metallic walls, constant potentials ( 5 V ) on the two driving electrodes and = on the gate electrode. was obtained by solving (4.22) on the gate electrode. The thermal boundary conditions are: insulation of the left and right side walls and outward flux of k w *(T Tambient ) d glass on all top and bottom boundaries (including the three electrodes). Here kw is the thermal conductivity of water and d glass = 500

156

m. The outward flux boundary condition was derived by assuming that the temperature outside the device is ambient. The flow boundary conditions are: noslip at all the boundaries. A reference pressure is defined at an arbitrary point in the domain. The time averaged electric field lines inside the device are shown in Fig 5.9. The electric field causes Joule heating and gives rise to two hot zones, one on the left of the gate electrode, and the other on the right (Fig. 5.10). The maximum temperature rise is negligible (0.0027 K). The electrothermal force is highest close to the edges of the electrodes and has a magnitude of approximately 0.02 N/m3 (Fig. 5.11). The resultant electrothermal flow velocity is approximately 0.6 nm/s (Fig. 5.12), which is 4 orders of magnitude smaller than the experimentally measured ICEO slip velocity. Therefore, the electrothermal flow can be ignored in our electrokinetic simulations. Now we will show a scaling analysis which predicts magnitudes of various quantities very reliably, without performing any simulations. The temperature rise of water as a result of Joule heating is governed by the following equation,
2 ( k wT ) + Erms = 0 ,

(5.21)

where k w is the thermal conductivity of water. Since the temperature is highest in the gaps between the driving and the gate electrodes (Fig. 5.10), an estimate for the temperature rise can made as follows
k w T
2

(G 2)

0 , 2L

(5.22)

157

where G is the gap between the driving and the gate electrodes (see Fig. 5.9). Note that G is the most relevant length scale for calculating the temperature gradients. Rearranging (5.22), we find
T

02 G

. 8k w L

(5.23)

Taking k w = 0.6 W/mK (at room temperature), = 17 104 S/m , 0 = 10 Volt, G=300 m and L=800 m, we find, T 0.003 K, which is very close to the simulation results. Now using (5.20), the electrothermal force on the fluid can be
2 T 0 3 roughly estimated to be f E 0.012 2 = 0.017 N/m , which is also very G/2 L

close to the simulation results. The fluid flow in the absence of any pressure gradients is governed by

2 u + f E = 0 ,
The electrothermal flow velocity can then be

(5.24) approximated as

u f E L2 6 1010 m/s which is again very close to the simulation results. H Here LH is the healing length and has a value of 6 m for our experimental conditions. Since the electrothermal force is highest in a very small region close to the edges, the healing length, LH , is the most relevant length scale for estimating the electrothermal flow velocity. The electrothermal flow velocity is approximately 4 orders of magnitude smaller than the ICEO slip velocity measurements (~45 m/s). Therefore, electrothermal flow can be neglected while simulating ICEO.

158

The electrothermal flow is negligible in our experiments because the electric conductivity of our fluid is very low and the Joule heating is negligible.

G L

Fig. 5.9: Time averaged electric field lines for the experimental conditions.

Fig. 5.10: Temperature rise in water as a result of Joule heating for the experimental conditions.

159

Fig. 5.11: Electrothermal force magnitude and vectors for the experimental conditions.

Fig. 5.12: Electrothermal flow velocity magnitude and vectors for the experimental conditions.

5.8

Stern Layer and High Ionic Concentrations

For the purified water of our experiments, we obtained a reasonable agreement between the simulations and the experiments. However, many other experiments were performed in higher salt concentrations and the agreement at higher

160

concentrations was found to be poorer. We can explain this in the following way: At higher concentrations, the double layer gets thinner and the surface conduction effects become smaller. Therefore, in the simulations, one would get higher velocity in higher ionic concentrations. However, from our experimental experience, we know that this is not the case in real life. In real life, the velocities decrease as the ionic concentrations increase. This behavior is generally blamed to the Stern layer which becomes very important at high ionic concentrations. The model presented in the previous sections does not take the Stern layer into account and therefore predicts the unrealistic effect of higher velocities at higher concentrations. To encounter this problem, we present a way to incorporate the Stern layer in our nonlinear model. In order to incorporate the Stern layer, the double layer model (5.9) is modified in the following way.
* * 1 = * * ( Dumodified * * ) t * x cosh * 2 y x

(5.25)

where Dumodified = 1 + cosh

*
2

(5.26)

where = D CS is a parameter determining the importance of the Stern layer. The Stern layer has been modeled as a linear capacitor with a capacitance CS . The bulk boundary condition (5.8) is modified to
* * = el * + q*

(5.27)

161

Derivation of (5.26) becomes clear when one makes the substitution, q* = 2sinh ( * / 2 ) in (5.27), differentiates it with respect to x* . The parameter, = D CS , is unknown and its value depends solely on the discretion of the user. In electrokinetic literature, workers have used as a fitting parameter and chosen a value which matches the simulations with the experiments [43]. In cases, where induced zeta potential is not large, its reasonable to blame the entire discrepancy on the Stern layer. In such cases, reasonably low values of

(<1) match the results of the simulations with the experimental data [43].
However, in cases where induced zeta potentials are large, one can not ignore the possibility of other nonlinear effects such as surface conduction and nonlinear surface capacitance. In such cases, its not reasonable to blame the entire discrepancy on the Stern layer. In fact, when we try to chose a value of while ignoring other nonlinear effects, we ends up getting an unreasonably high value of

1 ) for matching the results. This again bolsters our claim that at high

induced zeta potentials, nonlinearities play very important role in electrokinetics.

5.9 Conclusions

In this chapter, we solved a time dependent nonlinear model encompassing nonlinear surface capacitance and surface conduction and compared the numerical results with our experimental data. We showed that the combined effects of nonlinear capacitance, surface conduction and depth-averaging bring the numerical velocity within a factor of 6 of the experimental measurements for the driving

162

conditions of 0 = 10 V , f = 100 Hz , = 17 104 S/m . The results of our nonlinear simulations still do not match the experimental data exactly. The cause of the remaining discrepancy is not clear to us but there are several other effects which can be responsible for this remaining discrepancy. Stern layer is the most probable reason for it. We have presented a modified model for incorporating the Stern layer. We also can not ignore the possibilities of chemi-osmosis [61, 62] and steric effects between the finite size ions [63, 64]. Directions for incorporating chemi-osmosis are given in chapter 7.

163

164

Chapter 6. Induced Charge Electroosmosis on a Rough Surface

We present the results of our electrokinetic experiments on a rough surface. We show experimentally that the roughness of a surface can have dramatic impact on the electroosmotic flow on the surface. In the previous chapters, we had analyzed the flow around smooth surfaces; but in many practical situations, surfaces are not smooth and have roughness. We have produced induced charge electroosmotic flow on a rough surface containing randomly distributed nano structures of characteristic dimension 200 nm. Experiments show that the flow is upto 50% slower on the nanoscale roughness as compared to the flow on a smooth surface. This behavior is attributed to enhanced surface conduction on the nanoscale features of the rough surface. As described previously, the surface conductivity is inversely proportional to the characteristic dimension of the surface. In case of a smooth electrode, the characteristic dimension is equal to the width of the electrode (200 micron). In case of a rough electrode, however, the characteristic dimension is on the order of the roughness (200 nm). Due to a small length scale, surface conduction is greatly amplified on the rough surface and leads to slower ICEO flows. Here we present an experimental evidence for this effect.
6.1 Background

In order to understand the electroosmotic flow around small features, it is useful to think about the electrophoretic motion of a small particle in an ionic

165

liquid. In a reference frame attached to the particle, the electrophoretic motion of the particle relative to the fluid appears as the electroosmotic flow of the fluid relative to the particle. In other words, the electroosmotic flow of the fluid around the particle causes the particle to move relative to the fluid with the same relative speed as the speed of electroosmosis but in the opposite direction. It has been established that the electrophoretic mobility of a charged particle depends not only on its zeta potential but also on the ratio of the double layer thickness to the size of the particle [8-11]. OBrien and White, 1978 [9] calculated the electrophoretic mobility of a spherical particle as a function of its dimensionless zeta potential ( * = ze kT ) and the product a where is the
reciprocal debye length (i.e. = D1 ) and a is the radius of the particle. They

showed that for a sufficiently large particle ( a > 4 ), the dimensionless mobility
* ( e = 6 e kT ) decreases as a decreases at a fixed * (see Fig. 6.1). In other

words, as the dimension of the particle, a, gets smaller for a fixed and * , its mobility also gets lower. According to Squires et al. [77], the reduction in electrophoretic mobility with decreasing a can be understood in terms of surface conduction. In a previous chapter, we reported that Dukhin number is a function of

a ,
Du =
4 ze (1 + m ) sinh 2 a 4kT (6.1)

For small values of a , we can expect surface conduction to be very prominent. Surface conduction reduces the tangential field around the particle (see

166

Fig. 6.2b) and increases the normal electric field. This reduction in tangential electric field reduces electrophoretic velocity of the particle.

* Fig. 6.1: Dimensionless electrophoretic mobility ( e = 6 ee kT ) of a

particle as a function of its dimensionless zeta potential * = ze kT and a in a KCl solution. The mobility increases with a for a > 4 . This picture has been reprinted from OBrien and White 1978 [9]. Royal Society of Chemistry 1978. For a very small particle or a very thick double layer ( a
1 ), the preceding

analysis does not hold true. In this limit, the particle becomes so small compared to

167

the double layer that it does not feel the forces acting on the double layer and the electrophoretic mobility is simply derived by equating the Coulomb force on the particle to the Stokes drag. In other words, its not required to consider the presence of the double layer in the calculation of the electrophoretic mobility when the particle is too small ( a
1 ). The reader is referred to references [3] and [9]

for an analysis of this regime of electrophoretic mobility.

6.2 Fundamental Picture

Lets consider the electric field lines around a positively charged cylinder subject to a horizontal electric field. When the zeta potential of the cylinder is low, the surface currents are absent, and the electric field lines become tangential to the surface of the cylinder (see Fig. 6.2a). However, when the zeta potential is high, the surface currents are high and the electric field line configuration appears as shown in Fig. 6.2b. Since the surface conduction current is highest at the equator of the cylinder (because the tangential electric field, Es, is highest at the equator) and zero at the poles (because Es=0 on the poles), electric field acquires a large normal component at between the poles and the equator to maintain the current at the equator. The normal component of the electric field supplies the ions for the high surface current at the equator. As a result, the tangential field around the cylinder is reduced and electroosmotic slip velocity around the cylinder is severely deteriorated.

168

Pole, js=0

E0

Equator, js0

+ + + + + + + + +

Low Enormal=0

+ + + + + + +

Fig. 6.2 (a): Electric field lines around a lowly charged cylinder
Pole, js=0 E0 Equator, high js

+ + + + + + + + +

High

+ + + + + + +

High Enormal

Fig. 6.2(b): Electric field lines around a highly charged cylinder

Fig. 6.2: Electric field lines around (a) a lowly charged cylinder and (b) a highly charged cylinder. In the former case the zeta potential and surface conduction are low and therefore the electric field is completely tangential to the cylinder surface. In the latter case, the zeta potential and surface conduction are high. The surface current is zero at the pole but highest at the equator. To maintain the high surface current at the equator, field acquires a large normal component between the pole and the equator. The tangential field around the cylinder is reduced which reduces the electroosmotic slip velocity.

169

According to Khair and Squires 2008 [72], the length of the region in which the electric field has large normal component (called healing region) is roughly proportional to the ratio of the surface conductivity to the bulk conductivity

ze LH s = 4D (1 + m ) sinh 2 4kT

(6.2)

When is large, LH can be several times longer than D . When the dimension of the cylinder is small, a large portion of the cylinder surface is covered with the healing zone, yielding a negligible electroosmotic flow rate. Now lets consider a positively charged surface with its roughness approximated by a regular shape such as sinusoid (Fig. 6.3). Suppose that the dimension of the roughness is small and the zeta potential is high. In such a case, the healing length can be comparable to the dimension of the roughness and the tangential field is severely reduced on most of the surface. Consequently, the electroosmotic flow rate is also reduced.

E0
+ + + + +

high Enormal Low js high js + + + + + + + + + + + + + + + + + + + + + +

Fig. 6.3: Electric field lines on a charged rough surface. The roughness has been approximated as sinusoids. When the zeta potential and the surface currents are high, the electric field lines acquire a large normal component and the tangential component is reduced. As a result, the electroosmotic flow rate is also reduced.

170

6.2.1 Thick Double Layers

When the roughness is much smaller than the double layer ( a

1 ), it cant

affect the charging dynamics of the double layer and the surface current is again determined by the global dimension of the electrode such as the width of the electrode.
6.3 Experiments

6.3.1 Details of the Device

We measured induced charge electroosmotic flow in a device similar to the one described in chapter 3 but the gate electrode had a specially grown nanoscale roughness. As a brief recapitulation, this device had three coplanar microelectrodes (two driving electrodes and a gate electrode) laid on a glass substrate. Each microelectrode was 200 m in width. A PDMS chamber was carefully placed on the electrodes. The chamber was filled with a KCl solution seeded with fluorescent particles. An AC signal was applied between the two driving electrodes and the flow was measured on the gate electrode using PIV. The roughness of the electrode was realized by growing a nano structured titania (NST) film on a titanium electrode. NST is automatically formed when a Ti surface is oxidized in hydrogen peroxide. NST was covered with a thin layer of metal to make its surface conductive for achieving ICEO flow. The characteristic size of NST was 200 nm (see Fig. 6.4). NST was grown only on the gate electrode because thats where the ICEO flow of interest takes place; the driving electrodes

171

were kept smooth i.e. no NST was grown on them. However, it was only a half length of the gate electrode where NST was grown. The remaining half was kept smooth. This way we obtained two different portions, rough and smooth, on the same gate electrode. The interface between the two portions was clearly visible under the microscope (see Fig. 6.5). This allowed us to measure the slip velocity on both of the portions, rough and smooth, simultaneously and enabled us to compare the velocities on the two portions under exactly the same conditions.

200 nm
Fig. 6.4: An SEM image of the Nano-Structured Titania (NST) covered with a thin layer of metal. The characteristic dimension of NST is 200 nm. NST device was fabricated by Dr. Adam Monkowski who, at the time of fabrication, was a graduate student in Prof. Noel MacDonalds group at UCSB [78].

172

NST portion

Interface

Smooth portion 200 m

Fig. 6.5: A microscopic picture of the gate electrode. NST is grown on one half (top) of the electrode whereas the other half (bottom) is kept smooth. The interface between the two portions is clearly visible.

6.3.2 Fabrication Process

A glass wafer was cleaned respectively in Acetone, Iso propanol, de-ionized water (DI) and finally in O2 plasma. AZ 5214 photoresist was spin coated on the wafer and the device design was transferred from the mask into the photoresist by image reversal photolithography. Thin layers of metals (Ti/Au/Ti, respectively 50/2000/5000 ) were evaporated on the wafer and then lifted off in acetone. AZ 4110 positive resist was spin coated and exposed through a mask yielding a pattern which hid half length of the gate electrode and revealed rest of the pattern. The top Ti layer of the revealed pattern was etched in a solution of (49%HF): H2O2: DI (5: 5: 100). The gold layer was exposed due to etching of Ti. The resist was removed with Acetone. Then the Ti layer remaining on one half of the gate electrode was

173

oxidized in 10% H2O2 preheated to 850C on a hotplate. Oxidation was carried out for 8 minutes. Nano-Structured Titania (NST) is formed on Ti surface as a result of this oxidation [78-82]. NST was annealed at 3000C for 1 hour in air. Then another AZ 5214 image reversal lithography, exactly similar as before, was carried out and layers of Ti/Pt (25/250 ) were evaporated on all the electrodes. During evaporation, the wafer was held at an angle of 200-300 and rotated at about 0.5 rpm to get uniform metal coverage on NST. The residual metal was then lifted off in acetone.
6.3.3 Experimental Results

ICEO flow velocity on the gate electrode was measured in various KCl concentrations and at various driving voltages at a constant frequency of 100 Hz. Figs. 6.6a-d show the velocity profiles for the following ionic concentrations and driving voltages: (a) purified water ( = 17 S/cm), 20 Vpp, (b) 5 mM ( = 835 S/cm), 14 Vpp, (c) 10 mM ( = 1400 S/cm), 20 Vpp, and (d) 10 mM ( = 2950 S/cm), 20 Vpp. In all of the above, the maximum velocity on the rough portion of electrode is 1.1-1.4 times lower than the maximum velocity on the smooth portion.

174

x 10 4 3 Slip Velocity (m/s) 2 1 0 -1 -2 -3 -4 -1.5

-5

Rough

-1

-0.5

Smooth 0 0.5 x (m)

1.5 x 10
-4

(a) Purified Water, 20 Vpp

1.5 1 Slip Velocity (m/s) 0.5 0

x 10

-5

Rough -0.5 -1 Smooth -1.5 -1.5 -1 -0.5 0 x (m/s) 0.5 1 1.5 x 10

-4

(b) 5 mM KCl, 14 Vpp

175

1.5 1 Slip Velocity (m/s) 0.5 0 -0.5 -1

x 10

-5

Rough

Smooth -1.5 -1.5 -1 -0.5 0 x (m) 0.5 1 1.5 x 10

-4

(c) 10 mM KCl, 20 Vpp

6 4 Slip Velocity (m/s) 2 0 -2 -4 -6 -1.5 Smooth 0 0.5 x (m) Rough x 10
-6

-1

-0.5

1.5 x 10
-4

(d) 20 mM KCl, 20 Vpp

Fig. 6.6: ICEO flow velocity on the gate electrode for various KCl concentrations and driving voltages at 100 Hz. The maximum velocity on rough

176

portion is approximately 1.1-1.4 times lower than the maximum velocity on the smooth portion.

Fig. 6.7 shows the maximum slip velocity as a function of the driving voltage at 100 Hz in 5 mM KCl. The maximum slip velocity on the rough portion is consistently lower than velocity on the smooth portion for all driving voltages. The velocity first increases with the driving voltage and then starts decreasing. The decrease in the velocity at high voltages is attributed to enhanced surface conduction due to large zeta potentials.

1.6 Maximum Slip Velocity, u (m/s) max 1.4 1.2 1 0.8 0.6 0.4 0.2

x 10

-5

Smooth

Rough

0 6

10 12 14 16 Driving Voltage, 0 (Vpp)

18

Fig. 6.7: The maximum slip velocity on the gate electrode as a function of driving voltage in 5 mM KCl at 100 Hz. The velocity is always lower on the rough portion of the gate electrode than the velocity on the smooth portion.

177

6.4

Asymmetry in Flow

The observed slip velocity is not symmetric about the center of the gate (note that the velocity is not zero at the center in Fig. 6.6). The reason behind the asymmetry is not related to the roughness. It was observed in many other experiments (with and without NST). There is a stray electroosmotic flow which superimposes on the symmetric ICEO flow and gives rise to the asymmetry in the results. The stray electroosmotic flow originates from the capacitive coupling between the gate electrode and the metallic stage of the microscope. The stage of the microscope can be considered grounded and therefore it is at a lower potential than the floating potential of the gate electrode. The glass between the gate and the stage can be seen as a dielectric material of the capacitor. The potential difference between the gate and the stage creates a stray zeta potential on the gate electrode which gives rise to a unidirectional stray electroosmotic flow. This effect can also bee seen as Field Effect (see chapter 3) caused by the grounded stage. This flow is always in one direction even if the electric field changes direction (as in AC Field Effect shown in chapter 3). We can easily identify the stray electroosmotic velocity. Its equal to the velocity at the center of the gate. We have not subtracted the stray velocity from our data so that the readers attention can be drawn to this surprising effect. To learn more about the origin of this stray flow, the reader is referred to Mansuripur, Pascall and Squires 2009 [83].

178

6.5

Conclusions

We experimentally demonstrated that the slip velocity of an ICEO flow is reduced due to the roughness of the surface. We found that a 200 nm roughness can reduce the velocity by a factor of 1.4 in high concentrations of KCl (10 mM) and under high induced zeta potentials (~ 1 Volt). We attribute this reduction of velocity to surface conduction. Surface conductivity is inversely proportional to the product a . On a smooth surface, a represents the global length scale of the surface (e.g. the width of the electrode). However, when the surface has roughness, the dimension a becomes equal to the dimension of the roughness which is much smaller than the global length scale. As a result, for small roughness features, the localized surface conduction becomes very prominent and causes reduction in the local tangential electric field which reduces the slip velocity. However, when the roughness is much smaller than the double layer, it can be ignored altogether because it cant affect the double layer charging dynamics.

179

180

Chapter 7. Conclusions and Future Directions

Following are the conclusions of our work.
7.1 Experimental

1. Symmetric ICEO flow was observed on a planar microelectrode. Slip velocity of 40 m/s was observed in purified water under driving conditions of 20 Vpp at 100 Hz applied over a distance of 1 mm. The linear theory predicts a slip velocity of ~4 mm/s which is 100 times higher than the experimental data. We estimate that the induced zeta potentials in our experiments are 10-40 times higher than the thermal voltage ( 10 40 kT ze ) and therefore we can expect the linear theory to break down. 2. The velocity is observed to depend on the salt concentration. The higher the salt concentration, the lower the velocity. We attribute this to Stern layer effects. 3. At higher applied voltages, the velocity does not scale as 02 . Instead it saturates and does not increase further. The voltage at which saturation starts depends on salt concentration. The higher the salt concentration, the higher the voltage at which saturation starts. We attribute this to surface conduction. 4. Field effect flow control has been demonstrated with AC electric fields. Net pumping velocity of 50 m/s was obtained. Field effect proves to be a good pumping method with the flexibility of modifying the flow rate and the freedom to reverse the flow direction by just varying the gate voltage.

181

7.2

Numerical

5. A novel method to simulate nonlinear electrokinetic effects has been formulated. This model can simulate surface conduction in steady and time dependent cases. In the time dependent cases (such as AC), a nonlinear surface capacitance affects the charging dynamics of the double layer. In the steady cases, the capacitance does not play any role because enough time is given for the double layer to charge. 6. The nonuniform induced zeta potentials, which are inherent in ICEO, lead to nonuniform surface currents and cause surprising gradients in the bulk electric field. Surface conduction through a nanoscale diffuse layer causes micron scale gradient in the bulk electric field. Surface current pulls the electric field lines into the double layer to maintain the surface current. This gives rise to healing regions at the corners. However, the lines are expelled out of the surface once again in the middle of the surface due to nonuniform surface current. Overall, surface conduction creates large normal electric field and reduces the tangential field causing a reduction in the slip velocity.
7.3 Simulations vs. Experiments

7. The combined effect of nonlinear capacitance, surface conduction and volumetric averaging brings the numerical velocity within a factor of 7-10 of the experimental data. The nonlinear model reduces the discrepancy between the theory and the experiments by atleast one order of magnitude.

182

8. The remaining discrepancy (factor of 7-10) could not be explained. However, there are several other nonlinear effects which have not been accounted for in our work. The two major effects which can be incorporated in the classical theory are Stern layer and chemi-osmosis. Contribution of Stern layer cant be predicted and has to be tuned by a user-defined parameter. Chemi-osmosis refers to concentration gradient driven flow of ions. Surface currents cause gradients not only in the bulk electric field but also in the bulk salt concentration. The concentration gradient driven flow counters the electroosmotic flow and further reduces the slip velocity. Directions for simulating chemi-osmosis are given in the later part of this chapter.
7.4 ICEO Flow over Rough Surfaces

9. ICEO flow was produced on a nanoscale rough surface. The characteristic dimension of the roughness was 200 nm. It was found that the flow over roughness is slower by a factor of 1.4 than the flow over a smooth electrode. 10. The reduction in velocity due to roughness was explained in terms of surface conduction. On a rough surface, the amount of surface current is determined by the roughness length scale which is much smaller than the length of the electrode. As a result surface conduction is much more prominent on a rough surface than on a smooth surface. This reduces the tangential field and the slip velocity.

183

7.5

Future Directions

We showed that surface flow of ions through nano scale diffuse layer creates microscale bulk gradients in the electric field. It can also create bulk gradients in the salt concentration. Bulk salt concentrations set up a chemi-osmotic flow of ions (and fluid) in a direction opposite to electroosmosis. The nonlinear formulation can be expanded to include the surface transport of the salt as well. Following are the simplified governing equations and effective boundary conditions for surface transport of charge and salt both.
7.5.1 Bulk Equations

Following the analysis of [62], we present simplified equations for the concentration field, c, the excess surface charge, q, and the excess surface salt concentration, w for a planar gate electrode. In the following, electroosmotic convection of q and w has been neglected. For a complete formulation, the reader is referred to [62]. 1. Conservation of conduction current in the bulk

(c ) = 0 .
2. Advection-diffusion of salt in the bulk (units of c are number per unit volume)

(7.1)

c + u c = D 2 c . t
3. Navier-Stokes equation for bulk fluid flow

(7.2)

184

 2u p = ln

c c

(7.3)

The term on the right hand side of the preceding equation represents chemiosmotic transport of fluid. The continuity equation is given as
u = 0 .

(7.4)

7.5.2 Surface Transport Equations

4. Conservation of excess charge, q, in the double layer (units of q are number per unit area)

q 1 c Dze Dze . = qD +w c + c x kT x kT y t x

(7.5)

5. Conservation of excess salt, w, in the double layer (units of w are number per unit area)

w 1 c Dze c = wD +q + D . c x kT x t x y
7.5.3 Boundary Conditions on the Gate Electrode
q = 2D c c ze sinh c 2kT c ze sinh 2 c 4kT
. .

(7.6)

(7.7)

w = 4D c

(7.8)

The slip velocity is given by,

uslip

4 kT ze 1 c . = ln cosh x ze kT c x

(7.9)

185

The second term on the right hand side represents the contribution from chemiosmosis. At large , chemi-osmotic velocity can be as high as the electroosmotic velocity.

186

Bibliography
1. Lyklema, J., Fundamentals of interface and colloid science, Volume II:

Solid-liquid interfaces. 2001: Academic press.

2. Bard, A.J. and L.R. Faulkner, Electrochemical methods, Fundamentals and

applications. second ed. 2001: John Wiley & Sons, Inc.

3. Morgan, H. and N.G. Green, AC Electrokinetics: colloids and

nanoparticles. 2003: Research Studies Press Ltd., Baldock, Hertfordshire,

England. 4. Reuss, F., Sur un nouvel effet de le electricite glavanique. Mem. Soc. Imp. Nat. Mosc., 1809. 2: p. 327. 5. Southern, E.M., Detection of specific sequences among DNA fragments

separated by gel electrophoresis. J Mol Biol., 1975. Nov 5;98(3): p. 50317. 6. Bikerman, J.J., Electrokinetic equations and surface conductance. A survey

of the diffuse double layer theory of colloidal solutions. Trans. Faraday

Soc. , 1940. 35: p. 154-160. 7. Deryaguin, B.V. and S.S. Dukhin, Theory of surface conductance. Colloid J. USSR, 1969. 31: p. 277. 8. Dukhin, S.S. and B.V. Deryaguin, Electrokinetic phenomena. Serface and Colloid Science, 1974. 7.

187

9.

O'Brien, R.W. and L.R. White, Electrophoretic mobility of a spherical

colloidal particle. J. Chem. Soc., Faraday Trans. 2, 1978. 74: p. 1607 1626. 10. O'Brien, R.W., The solution of the electrokinetic equations for colloidal

particles with this double layers. Journal of Colloid and Interface Science,
1983. 92: p. 204. 11. Russel, W.B., D.A. Saville, and W.R. Schowalter, Colloidal dispersions. 1989: Cambridge University Press. 12. Gamayunov, N.I., V.A. Murtsovkin, and A.S. Dukhin, Pair interaction of

particles in electric field. 1. Features of hydrodynamic interaction of polarized particles. Colloid J. USSR, 1986. 48: p. 197.
13. Dukhin, S.S., Non-equilibrium electric surface phenomena. Advances in Colloid and Interface Science, 1993. 44: p. 1. 14. Murtsovkin, V.A., Nonlinear flows near polarized disperse particles. Colloid J. Russ. Acad. Sci., 1996. 53: p. 947. 15. Squires, T.M. and S.R. Quake, Microfluidics: Fluid physics at the nanoliter

scale. Reviews of Modern Physics, 2005. 77(3): p. 977-1026.

16. Gascoyne, P.R.C. and J. Vykoukal,

Particle

separation

by

dielectrophoresis. ELECTROPHORESIS, 2002. 23(13): p. 1973-1983.

17. Gascoyne, P.R.C., et al., Dielectrophoretic separation of cancer cells from

blood. Industry Applications, IEEE Transactions on, 1997. 33(3): p. 670678.

188

18.

Pethig, R. and G.H. Markx, Applications of dielectrophoresis in

biotechnology. Trends in Biotechnology, 1997. 15(10): p. 426-432.

19. Hu, X.Y., et al.,

Marker-specific

sorting

of

rare

cells

using

dielectrophoresis. Proceedings of the National Academy of Sciences of the

United States of America, 2005. 102(44): p. 15757-15761. 20. Sigurdson, M., D.Z. Wang, and C.D. Meinhart, Electrothermal stirring for

heterogeneous immunoassays. Lab on a Chip, 2005. 5(12): p. 1366-1373.

21. Feldman, H.C., M. Sigurdson, and C.D. Meinhart, AC electrothermal

enhancement of heterogeneous assays in microfluidics. Lab on a Chip,

2007. 7(11): p. 1553-1559. 22. Yao, S.H., et al., Porous glass electroosmotic pumps: design and

experiments. Journal of Colloid and Interface Science, 2003. 268(1): p.

143-153. 23. Yao, S.H. and J.G. Santiago, Porous glass electroosmotic pumps: theory. Journal of Colloid and Interface Science, 2003. 268(1): p. 133-142. 24. Jacobson, S.C., et al., Open-Channel Electrochromatography on a

Microchip. Analytical Chemistry, 1994. 66(14): p. 2369-2373.

25. Jiang, L.N., et al., Closed-loop electroosmotic microchannel cooling system

for VLSI circuits. Ieee Transactions on Components and Packaging

Technologies, 2002. 25(3): p. 347-355.

189

26.

Ramos, A., et al., Ac electrokinetics: a review of forces in microelectrode

structures. Journal of Physics D-Applied Physics, 1998. 31(18): p. 23382353. 27. Ramos, A., et al., AC electric-field-induced fluid flow in microelectrodes. Journal of Colloid and Interface Science, 1999. 217(2): p. 420-422. 28. Green, N.G., et al., Fluid flow induced by nonuniform ac electric fields in

electrolytes on microelectrodes. I. Experimental measurements. Physical

Review E, 2000. 61(4): p. 4011-4018. 29. Gonzalez, A., et al., Fluid flow induced by nonuniform ac electric fields in

electrolytes on microelectrodes. II. A linear double-layer analysis. Physical

Review E, 2000. 61(4): p. 4019-4028. 30. Ramos, A., et al., Comment on "Theoretical Model of Electrode

Polarization and AC Electroosmotic Fluid Flow in Planar Electrode Arrays". Journal of Colloid and Interface Science, 2001. 243(1): p. 265266. 31. Green, N.G., et al., Fluid flow induced by nonuniform ac electric fields in

electrolytes on microelectrodes. III. Observation of streamlines and numerical simulation. Physical Review E, 2002. 66(2): p. 11.
32. Ajdari, A., Pumping liquids using asymmetric electrode arrays. Physical Review E, 2000. 61(1): p. R45-R48.

190

33.

Ramos, A., et al., Pumping of liquids with ac voltages applied to

asymmetric pairs of microelectrodes. Physical Review E, 2003. 67(5): p.

11. 34. Seibel, K., et al. Transport properties of ac electroosmotic micropumps on

labchips. 2003. Cambridge, UK.

35. Ramos, A., et al., Pumping of electrolytes using arrays of asymmetric pairs

of microelectrodes subjected to ac voltages, in Electrostatics 2003. 2004,

Iop Publishing Ltd: Bristol. p. 187-192. 36. Studer, V., et al., An integrated AC electrokinetic pump in a microfluidic

loop for fast and tunable flow control. Analyst, 2004. 129(10): p. 944-949.
37. Garcia-Sanchez, P., et al., Experiments on AC electrokinetic pumping of

liquids using arrays of microelectrodes. Ieee Transactions on Dielectrics

and Electrical Insulation, 2006. 13(3): p. 670-677. 38. Olesen, L.H., H. Bruus, and A. Ajdari, ac electrokinetic micropumps: The

effect of geometrical confinement, Faradaic current injection, and nonlinear surface capacitance. Physical Review E, 2006. 73(5).
39. Wong, P.K., et al., Electrokinetic bioprocessor for concentrating cells and

molecules. Analytical Chemistry, 2004. 76(23): p. 6908-6914.

40. Bown, M.R. and C.D. Meinhart, AC electroosmotic flow in a DNA

concentrator. Microfluidics and Nanofluidics, 2006. 2(6): p. 513-523.

41. Squires, T.M. and M.Z. Bazant, Induced-charge electro-osmosis. Journal of Fluid Mechanics, 2004. 509: p. 217-252.

191

42.

Bazant, M.Z. and T.M. Squires, Induced-charge electrokinetic phenomena:

Theory and microfluidic applications. Physical Review Letters, 2004. 92(6): p. 4.

43. Levitan, J.A., et al., Experimental observation of induced-charge electro-

osmosis around a metal wire in a microchannel. Colloids and Surfaces aPhysicochemical and Engineering Aspects, 2005. 267(1-3): p. 122-132. 44. Soni, G., Squires, T.M., Meinhart, C.D., Nonlinear Phenomena in Induced

Charge Electroosmosis. Proceedings of 2007 ASME International

Mechanical Engineering Congress and Exposition, Seattle, Washington, 2007: p. IMECE2007-41468. 45. Squires, T.M. and M.Z. Bazant, Breaking symmetries in induced-charge

electro-osmosis and electrophoresis. Journal of Fluid Mechanics, 2006. 560: p. 65-101.

46. Gangwal, S., et al., Induced-charge electrophoresis of metallodielectric

particles. Phys. Rev. Lett., 2008. 100: p. Art no. 058302.

47. Hayes, M.A., I. Kheterpal, and A.G. Ewing, Effects of buffer pH on

electroosmotic flow-control by an applied radial voltage for capillary zone electrophoresis. Analytical Chemistry, 1993. 65(1): p. 27-31.
48. Schasfoort, R.B.M., et al., Field-effect flow control for microfabricated

fluidic networks. Science, 1999. 286(5441): p. 942-945.

192

49.

Polson, N.A. and M.A. Hayes, Electroosmotic flow control of fluids on a

capillary electrophoresis microdevice using an applied external voltage.

Analytical Chemistry, 2000. 72(5): p. 1088-1092. 50. Lee, C.Y., C.H. Lin, and L.M. Fu, Band spreading control in

electrophoresis microchips by localized zeta-potential variation using field-effect. Analyst, 2004. 129(10): p. 931-937.
51. Lee, G.B., et al., Dispersion control in microfluidic chips by localized zeta

potential variation using the field effect. Electrophoresis, 2004. 25(12): p.

1879-1887. 52. Sniadecki, N.J., et al., Induced pressure pumping in polymer microchannels

via field-effect flow control. Analytical Chemistry, 2004. 76(7): p. 19421947. 53. van der Wouden, E.J., et al., Field-effect control of electro-osmotic flow in

microfluidic networks. Colloids and Surfaces a-Physicochemical and

Engineering Aspects, 2005. 267(1-3): p. 110-116. 54. Daiguji, H., P.D. Yang, and A. Majumdar, Ion transport in nanofluidic

channels. Nano Letters, 2004. 4(1): p. 137-142.

55. Fan, R., et al., Inorganic nanotube nanofluidic transistors for single

molecule detection. Abstracts of Papers of the American Chemical Society,

2005. 229: p. U141-U142. 56. Karnik, R., et al., Electrostatic control of ions and molecules in nanofluidic

transistors. Nano Letters, 2005. 5(5): p. 943-948.

193

57.

Karnik, R., K. Castelino, and A. Majumdar, Field-effect control of protein

transport in a nanofluidic transistor circuit. Applied Physics Letters, 2006. 88(12): p. 3.

58. van der Heyden, F.H.J., et al., Electrokinetic Energy Conversion Efficiency

in Nanofluidic Channels. Nano Lett., 2006. 6(10): p. 2232-2237.

59. van der Heyden, F.H.J., et al., Power Generation by Pressure-Driven

Transport of Ions in Nanofluidic Channels. Nano Lett., 2007. 7(4): p. 10221025. 60. Pennathur, S., J.C. Eijkel, and A. van den Berg, Energy conversion in

microsystems: is there a role for micro/nanofluidics? Lab Chip, 2007. 7: p.

1234-1237. 61. Chu, K.T. and M.Z. Bazant, Nonlinear electrochemical relaxation around

conductors. Physical Review E (Statistical, Nonlinear, and Soft Matter

Physics), 2006. 74(1): p. 011501. 62. Khair, A.S. and T.M. Squires, Fundamental aspects of concentration

polarization arising from nonuniform electrokinetic transport. Physics of

Fluids, 2008. 20: p. 087102. 63. Kilic, M.S., M.Z. Bazant, and A. Ajdari, Steric effects in the dynamics of

electrolytes at large applied voltages. I. Double-layer charging. Physical

Review E, 2007. 75(2). 64. Bazant, M.Z., et al., Nonlinear Electrokinetics at large applied voltages. arXiv:cond-mat/0703035v2 [cond-mat.other], 2007.

194

65. 66.

CRC Handbook of Chemisty and Physics, 88th Edition. 2007-2008.

Chen, C.H. and J.G. Santiago, A planar electroosmotic micropump. Journal of Microelectromechanical Systems, 2002. 11(6): p. 672-683.

67.

Santiago, J.G., et al., A particle image velocimetry system for microfluidics. Experiments in Fluids, 1998. 25(4): p. 316-319.

68.

Meinhart, C.D., S.T. Wereley, and J.G. Santiago, A PIV algorithm for

estimating time-averaged velocity fields. Journal of Fluids EngineeringTransactions of the Asme, 2000. 122(2): p. 285-289. 69. Mutlu, S., et al., Enhanced electro-osmotic pumping with liquid bridge and

field effect flow rectification. Micro Electro Mechanical Systems, 17th

IEEE International Conference on MEMS, 2004: p. 850- 853. 70. Meinhart, C.D., S.T. Wereley, and J.G. Santiago, PIV measurements of a

microchannel flow. Experiments in Fluids, 1999. 27(5): p. 414-419.

71. Prasad, A.K., et al., Effect of resolution on the speed and accuracy of

particle image velocimetry interrogation. Experiments in Fluids, 1992. 13:

p. 105-116. 72. Khair, A.S. and T.M. Squires, Surprising consequences of ion conservation

in electro-osmosis over a surface charge discontinuity. Under consideration

for publication in J. Fluid Mech., 2008. 73. Yariv, E., Electro-osmotic flow near a surface charge discontinuity. J. Fluid Mech., 2004. 521: p. 181.

195

74.

Johnson, C., Numerical solution of partial differential equations by the

finite element method. Studentlitteratur/Cambridge University Press, 1987.

75. Olesen, L.H., Computational Fluid Dynamics in Microfluidic Systems Technical University of Denmark, 2003(Masters Thesis). 76. Olsen, M.G. and R.J. Adrian, Out-of-focus effects on particle image

visibility and correlation in microscopic particle image velocimetry.

Experiments in Fluids, 2000. 29(7): p. S166-S174. 77. Squires, T.M., A.S. Khair, and R. Messinger. Electrokinetic Flows Over

Rough Surfaces. in 2008 AIChE Annual Meeting. November 16-21, 2008.

Philadelphia, PA. 78. Monkowski, A., PhD Thesis, Microfabricated Structures and Devices

Featuring Nanostructured Titania Thin Films, in Department of Mechanical Engineering. 2007, University of California: Santa Barbara,
CA. 79. Tengvall, P., et al., Interaction between hydrogen peroxide and titanium: a

possible role in the biocompatibility of titanium. Biomaterials, 1989. 10(2):

p. 118-120. 80. Ichinose, I., H. Senzu, and T. Kunitake, A Surface Sol-Gel Process of TiO2

and Other Metal Oxide Films with Molecular Precision. Chem. Mater.,
1997. 9(6): p. 1296-1298.

196

81.

Wu, J.-M., et al., Porous titania films prepared from interactions of

titanium with hydrogen peroxide solution. Scripta Materialia, 2002. 46(1):

p. 101-106. 82. Jane P. Bearinger, C.A.O.J.L.G., Effect of hydrogen peroxide on titanium

surfaces: In situ imaging and step-polarization impedance spectroscopy of commercially pure titanium and titanium, 6-aluminum, 4-vanadium.
Journal of Biomedical Materials Research Part A, 2003. 67A(3): p. 702712. 83. Mansuripur, T.S., A.J. Pascall, and T.M. Squires, Asymmetric induced

charge electro-osmotic flows over symmetric surfaces: the role of capacitive coupling. Physical Review E, 2009. To be submitted.

197