Nonlinear Electrophoresis of Highly Charged Nonpolarizable Particles

Soichiro Tottori, Karolis Misiunas, and Ulrich F. Keyser∗ Cavendish Laboratory, Department of Physics, University of Cambridge, JJ Thomson Avenue, Cambridge, CB3 0HE, United Kingdom

Douwe Jan Bonthuis Institute of Theoretical and Computational Physics, Graz University of Technology, 8010 Graz, Austria (Dated: July 10, 2019) Nonlinear field dependence of electrophoresis in high fields has been investigated theoretically, yet experimental studies have failed to reach consensus on the effect. In this work, we present a systematic study on the nonlinear electrophoresis of highly charged submicron particles in applied electric fields of up to several kV/cm. First, the particles are characterized in the low-field regime at different salt concentrations and the surface charge density is estimated. Subsequently, we use microfluidic channels and video tracking to systematically characterize the nonlinear response over a range of field strengths. Using velocity measurements on the single particle level, we prove that nonlinear effects are present at electric fields and surface charge densities that are accessible in practical conditions. Finally, we show that nonlinear behavior leads to unexpected particle trapping in channels.

Electrophoresis has been widely used for sensing, filtra- trokinetics found for polarizable particles [20], is not well tion, manipulation, and the separation of molecules and understood for intermediate to large κa in contrast to the particles, particularly recently using micro- and nanoflu- case of small κa in nonpolar electrolytes [21–26]. The- idic devices [1–4]. On these length scales, the applied oretical analyses have predicted that the electrophoretic electric field may reach kV/cm because the confinement velocity becomes nonlinear to the applied field at a mod- ∼ focuses the electric field, even with a moderate applied erate to high field β = aE/φth & 1 [14, 27–29]. However, voltage to the system. The electrophoretic velocity vep a well-controlled experiment in this regime has been lack- of a rigid nonpolarizable sphere in the limit of low zeta ing. Indeed, there are experimental reports showing that potential ζ and low applied electric field E is particle electrophoretic velocity is linear [30] and nonlin- ear [27] under similar conditions. Even for the nonlinear 2 v = f(κa)ζE, (1) case, the properties of particles were not clearly char- ep 3η acterized. Moreover, the zeta potential was reported to where  and η are the permittivity and viscosity of the be severalfold lower than the value used for fitting one fluid, respectively [5]. f(κa) is Henry’s function, where of the models, and the difference was then attributed to 1 κ− and a are the Debye length and particle radius, the divergence of the slip plane from the actual particle respectively [6, 7]. Within a canonical model (homo- surface [31, 32]. Here, we resolve the conflicting results geneous  and η, no specific ion-surface interactions, found in the literature by systematic experiments in com- no hydrodynamic slip) and using a low zeta potential bination with detailed simulations. ( ζ < φth = kBT/e 26 mV), ζ calculated using Eq. (1) We used high-speed video tracking to study the non- corresponds| | to the≈ homogeneous electrostatic potential linear electrophoresis of highly charged particles in mi- at the particle surface. In real systems, however, the crofluidic channels. We characterized particle mobility zeta potentials of particles may exceed φth, in which case with different salt concentration, and subsequently mea- the direct action of the electric field on the double layer sured the nonlinear electrophoretic velocity under high as well as the advection of ions by the flow result in an field up to β 3. The experimental data were compared ≈ asymmetric shape of both coion and counterion clouds, with the values obtained with coupled Stokes-Poisson- arXiv:1907.04278v1 [cond-mat.soft] 9 Jul 2019 as schematically shown in Fig. 1(a) [8, 9]. For interme- Nernst-Planck (SPNP) simulations, showing that high diate values of κa, the double-layer distortion, known as field particle electrophoresis experiments can be quan- relaxation effect, leads to a nonlinear dependence of vep titatively explained with our SPNP model. We also on ζ [7, 8, 10]. Therefore, in case ζ & φth, the model in- demonstrate that particles can be trapped in the channel cluding relaxation effect [11, 12] needs| | to be used instead at high fields due to this nonlinear effect in combination of Eq. (1) to extract the surface potential that reproduces with the electro-osmotic flow resulting from the channel vep within the canonical model. wall. In addition, the relaxation effect is field dependent, Microfluidic channels were made out of polydimethyl- complicating the interpretation of experimental results siloxane (PDMS) cast onto microfabricated molds [33]. even further (see Supplemental Material [13]). This non- The mold for the “wide” channel (w = 12.7 µm, h = linear effect, which is different from the nonlinear elec- 5.2 µm, l = 100 µm) was fabricated with SU-8 photore- 2

(a) (b) Laser Doppler Microelectrophoresis (Zetasizer Nano ZSP, E Malvern). All particles were diluted in KCl solution Ag/AgCl electrode to the final concentration (volume fraction) of approx- ζ < ϕth + + + 5 5 4 + imately 2.5 10− , 1 10− , and 2.5–10 10− for + tube KCl solution × × × + + narrow and wide microchannel experiments and Laser + + PDMS Doppler Microelectrophoresis measurement, respectively. ζ ϕ E ≳ th + + In Fig. 1(c), the electrophoretic mobility of PS particles + PS particle - + - in 10 mM KCl is plotted as a function of pH. The pH was + + access channel PDMS-coated + glass + - objective adjusted by adding small quantities of KOH or HCl. The lens + + 8 2 1 1 observed negative mobility of 7 10− m V− s− is ≈ − × (c) -4 due the negative surface charge associated with the sul- )

-1 PS 620 nm, 10 mM KCl fate end groups on PS, originating from the decomposi- s

-1 tion of the initiator used during particle synthesis (ini- V

2 + tiator: K2S2O8, end group: -OSO3−K ). The reported m -6

-8 values vary but typically are in the range of several tens 2

(10 of mC/m . As the dissociation constant pKa of sulfate ep

μ end groups is below 2 [35–37], no significant change in -8 electrophoretic mobility was observed for pH 4. We 2 4 6 8 10 decided that no buffer was necessary in the following≥ ex- pH periments, and pH values were all kept in the range of 6–9. FIG. 1. (a) Schematic illustrations of the counterion cloud around particles with low and high zeta potentials under an Coupled SPNP simulations were performed using a fi- electric field E. (b) Schematic illustration of our experimental nite element method (COMSOL Multiphysics ver. 4.4). system. Microfluidic chip is connected to two open reservoirs. The particle was fixed in a large enough space (= 100a) Particle motion in channels is imaged using a camera via the with open boundary conditions. The surface charge den- objective. (c) Bulk electrophoretic mobility of PS particle sity was set at a constant value on the particle surface. (620 nm in diameter) as a function of pH. Dotted line is the The electro-osmotic flow generated far away from the averaged value of µep for pH 4. ≥ particle was used to calculate the electrophoretic mobil- ity of the particle [5]. More details on the simulations can be found in Supplemental Material [13]. sist. The “narrow” channels (w = 750 nm, h = 750 nm, In Fig. 2, the electrophoretic mobilities µep of var- l = 10 µm) were made by depositing platinum onto silicon ious particles measured with Laser Doppler Micro- substrates with focused ion beam. The access channels electrophoresis are plotted as a function of KCl concen- for wide and narrow channels were made out of SU-8 tration c. The field used in the measurement was be- (h = 100 µm) and AZ 9260 (h = 12 µm), respectively. low 0.025 kV/cm (β 0.05); thus nonlinear effects were The bottom glass slides were coated with a thin layer of negligible. The electrophoretic≈ mobility curves of PS PDMS, then plasma bonded to the PDMS channels [34]. particles display negative peaks at about 10 mM, in a The overview of our experimental system is shown in good agreement to previous literature values for other Fig. 1(b). The channels were connected to the two reser- highly charged particles [35, 38–40]. The negative peak voirs through access channels and tubes. The reservoir of PMMA particles, which had lower charge density, was height was adjusted to eliminate the pressure difference found at around 1 mM. between both sides before each experiment. Voltage ( ≤ In order to estimate the surface charge density from 35 V) was applied through Ag/AgCl electrodes connected these data, we fitted the approximate model including the to a source measure unit (Keithley 2450). The particle relaxation effect by Chen and Keh [12] (valid for κa & 20) motion was recorded at maximum 2000 fps and tracked to the experimental data with a surface charge density as by custom-written software after experiments. Conver- the only fitting parameter (see Supplemental Material for sion coefficients k from voltage to field inside a channel, more details [13]). The extracted surface charge densities E = kV , were calibrated for each microfluidic chip de- of PS 620 nm, PS 370 nm, PMMA 520 nm, and PS-COOH sign by measuring the current I through the system, as 380 nm are 51.2, 41.7, 13.1, and 11.2 mC/m2, re- k = I/(ρAV ), where ρ and A are the conductivity of the spectively.− Note that− these− surface charge− densities are electrolyte and total channel cross sectional area, respec- not necessarily the bare surface charge densities because tively. we do not take into account the details of the surface, The particles used in our experiments were polystyrene such as local viscosity and permittivity. However, since (PS, from Polysciences and Bangs Labs) and poly(methyl these values are employed in the numerical SPNP simu- methacrylate) (PMMA, from Microparticles GmbH). lations to predict the magnitude of the nonlinearity using Bulk electrophoretic mobilities were measured with the canonical model of electrophoresis, this treatment is 3

(a) 10 0 PS 620 nm PS 620 nm PMMA 520 nm PS 372 nm linear fit 5 PMMA 520 nm (mm/s) linear fit

) -2 PS-COOH 380 nm m -1 s − v -1

V 0 2

m -4 5

-8 (b) PS 620 nm

(10 PMMA 520 nm

ep 4 2 μ -6 simulation (620 nm, 51.2 mC/m ) simulation (520 nm, 13.1 mC/m2 ) 3 ref. [14] ref. [14]

-8 (mm/s)

nl 2 10-1 100 101 102 103 − v c (mM) 1

FIG. 2. Bulk electrophoretic mobility as a function of salt con- 0 centration. Dashed and dash-dotted lines are the fitted curves to the Chen and Keh model [12] for PS 620 nm and PMMA 0 0.5 1 1.5 2 2.5 520 nm, respectively. Estimated surface charge densities of PS E (kV/cm) 620 nm, PS 370 nm, PMMA 520 nm, and PS-COOH 380 nm are 51.2, 41.7, 13.1, and 11.2 mC/m2, respectively. FIG. 3. (a) Measured particle velocity v (= v + v ) as a − − − − m ep eof function of the applied field in the wide channel (c = 1 mM). Solid and dashed lines are linear fits of the first two points. (b) Nonlinear velocity components vnl as a function of the fully consistent. applied field. Solid lines are the SPNP simulation results. Having established an estimate for the surface charge Error bars in (a) and (b) are both standard errors of the density, the particle velocity in high fields was char- mean and within the symbol size. Dashed lines are the cubic acterized using a microfluidic channel. Hydrodynamic term from the model by Schnitzer and Yariv in [14]. and electric wall effects were negligible in this wide channel since the channel height and width were large enough [15]. The two types of particles (PS 620 nm and surface charge densities of 51.2 and 13.1 mC/m2, re- PMMA 520 nm) were suspended in 1 mM KCl. Previ- spectively. The simulations− showed a− good agreement ously, two methods have been proposed to extract parti- with the experiments. This implies that we can predict cle mobility in a high field: using asymmetric oscillating the magnitude of nonlinearity well with the estimated fields [27, 31, 32, 41] and measuring bulk flow rate from surface charge density. the hydraulic height in a reservoir [30, 42]. Here we used Our results contradict the earlier report by Kumar et microfluidic channels and dc voltage to directly moni- al. [30], where they concluded that the electrophoretic tor particle motions, which allowed us to acquire non- velocity was always linear. Our results are qualitatively linear electrophoretic velocity on the single particle level in line with the measurements performed by Shilov et precisely, without transient behavior, bubble generation, al. [27] and Barany [41] using oscillating fields. How- and pressure flow. ever, a direct comparison is difficult because neither the In Fig. 3(a), the measured particle velocity vm (= zeta potential nor surface charge density of their parti- vep + veof, the sum of the electrophoretic and electro- cles were reported. Instead, the model in [27] was fitted osmotic velocities) is plotted as a function of an ap- to the experimental data with zeta potential as a fitting plied field with the maximum field strength 2.5 kV/cm, parameter. Mishchuk et al. used the same approach which is equivalent to β 3 (for 2a = 620≈ nm). The and reported that their measured particle zeta potential solid and dashed lines are≈ the linear fits to the first two ( 28 mV) was significantly lower than the zeta potential points of each data set. Clearly, the particle velocities in- estimated− using Shilov’s model ( 175 mV) [27], and at- creased superlinearly with the applied field for both types tributed this difference to the divergence− of the slip plane of particles. Figure 3(b) shows the nonlinear components and particle surface [31, 32]. However, the zeta potential of the particle velocity vnl, which was extracted by sub- was estimated using Eq. (1) even at low salt concentra- tracting the linear components obtained from linear fits tions (c = 0.1 mM). Thus, it is likely that this value was shown in Fig. 3a. This process is important because it underestimated compared to the actual zeta potential. is difficult to measure veof precisely in situ. The solid In contrast, our results suggest that nonlinear effects can lines are the nonlinear component of the SPNP simula- be accurately predicted by taking into account the relax- tion results with the actual size of the particle and the ation effect with extracted surface charge densities. 4

0.9 kV/cm E For the analytical approach in the thin double layer (a) (c) 80 8 limit and small P´ecletnumber, the first two allowed terms 0 x 2.6 kV/cm 60 6 by symmetry are described as v = µ E+µ(3)E3, where ep ep ep 40 (3) 4 3.5 kV/cm both coefficients µ and µ are independent of E [14, F (pN) ep ep 20 (3) 2 E (kV/cm) 28]. The coefficients µep obtained from our results (PS 4.4 kV/cm 0 620 nm and PMMA 520 nm) in Fig. 3(b) were approxi- 0 mately 21.4( 2.2) and 9.9( 1.0) 10 20m4s 1V 3, -1 -0.5 0 0.5 1 − − − 8.8 kV/cm x (μm) respectively.− Based± on the− model± proposed× by Schnitzer 40 et al. [13, 14] for small Dukhin and P´ecletnumbers, the (d) 3.78 kV/cm (b) 2.6 kV/cm 4.72 kV/cm nonlinear coefficients for these parameters are 37.6 and 20 5.67 kV/cm 20 4 1 3 − 6.61 kV/cm 7.9 10− m s− V− , as also shown in Fig. 3(b). Inter- − × 0 7.55 kV/cm estingly, although the model is based on the assumption 8.50 kV/cm F (pN) of small Dukhin and P´ecletnumbers, it still shows a rea- -20 0 x (3) sonable agreement to our experimental data µep . These -40 results imply that the two particles with similar µep can -1.5 -1 -0.5 0 0.5 1 (3) x (μm) be still separated by the difference of µep . Last, we discovered that this nonlinear electric field FIG. 4. Electric field induced trapping and transport. (a) Im- dependence allows for voltage-controlled particle trap- ages show 0, 3, 5, and 7 particles (PS-COOH 380 nm) trapped ping in confinement. Here, slightly lower charged par- inside the channel for E = 0.9, 2.6, 3.5, and 4.4 kV/cm, respec- ticles (PS-COOH 380 nm) in 5 mM KCl were introduced tively. No particle entered the channel at 8.8 kV/cm. Scale into PDMS-glass-bonded narrow channels. The electro- bar equals 5 µm. (b) Example of transport mode for a channel osmotic flow was stronger than the electrophoresis of with three particles at 2.6 kV/cm. Spatiotemporal diagram displaying the cascading motion of trapped particles. During the particles; thus the particles were always transported the shown time frame of 72 ms three particles entered from the by the electro-osmotic flow at low fields. The mobil- left (arrows) releasing the particles at the right end (arrows). ity of particle inside the channel µc (= µep + µeof ) was (c) Simulated force landscape near the entrance (left) and lo- 8 2 1 1 0.8 10− m V− s− at E 0.088 kV/cm (β 0.065). cal external electric field (right). Shadowed area corresponds ≈ × ≈ ≈ Interestingly, at a sufficiently high field (E & 2.6 kV/cm to the inside of the channel. (d) Simulated force landscape or β & 1.9), the particles were trapped at the inside or near the constriction. Shadowed area corresponds to the con- entrance of the channel, as shown in Fig. 4(a). The num- striction. Field strengths are the values at the center of an empty channel. ber of trapped particles increased with the applied field from 3, 5, 7 for 2.6, 3.5, and 4.4 kV/cm (β 1.9, 2.6, and 3.3), respectively. No particle entered the≈ channel at ' 8.8 kV/cm. When trapping occurred inside the chan- ing fields, resulting in larger trapping forces. The larger nel, trap-and-release cascade motion was observed: when trapping forces allow for a larger number of trapped par- one additional particle was entering, the initially trapped ticles, before the outermost particle is pushed out of particle was released. The spatiotemporal diagram of this the trapping site by steric or hydrodynamic interaction cascade motion is shown in Fig. 4b. with an incoming particle. At a sufficiently high field In order to understand this trapping mechanism, we (8.50 kV/cm), no particle can be trapped inside the chan- numerically simulated the force landscape near the en- nel since the force landscape is negative at the entrance, trance of the channel, as shown in Fig. 4(c). Force preventing particles from entering the channel. Clearly, was calculated as the sum of electrostatic and hydrody- the trapping strength is dependent on the characteris- tics of the constrictions, such as height and length, yet namic forces on the fixed particle (2a = 380 nm, σp = 11.2 mC/m2). The surface charge density of the chan- our numerical results capture the trapping and cascade − 2 motions observed in experiments. This trapping effect nel wall was estimated to be σw = 12.7 mC/m from the particle velocity inside the channel− at low fields. The due to nonlinear field dependence of electrophoresis is confinement ratio was set a/R = 0.5. The force land- intriguing, as it is clearly different in origin from well scape turns from positive to negative near the entrance characterized dielectrophoretic trapping [43]. because of the sharp increase of the field. We further We experimentally proved that highly charged noncon- investigated the effect of small inhomogeneity on force ducting particles exhibit nonlinear electrophoretic veloc- landscapes by adding a small constriction in the channel ity at high fields. Our results include measurements rang- (20 nm 0.05R in height, R in length) in Fig. 4(d). At ing from the low field to the high field limit spanning intermediate≈ fields (5.67, 6.61, and 7.55 kV/cm), force several orders of magnitude in salt concentration. Our reversal was observed near the constriction inside the high-speed video tracking results allowed for extracting channel, indicating that particles can be trapped inside the particle mobility in channels and comparing the re- the channel. The force landscapes decrease with increas- sults with the numerical models. We found a near quanti- 5 tative agreement with numerical simulations using finite of multiple colloidal spheres,” J. Fluid. Mech. 238, 251– element methods. Finally we showed that the nonlin- 276 (1992). ear mobility can give rise to unexpected particle trapping [13] See Supplemental Material at [url] for [the explanation that is controlled by the applied field. Our results are rel- on nonlinearity in electrophoresis, the detail of numerical simulation, the analytical models of the electrophoretic evant for various fluidic systems and processes, including mobility and the nonlinear term in high fields], which filtration, separation, and manipulation. includes Refs. [7, 8, 12, 14–19]. [14] O. Schnitzer and E. Yariv, “Nonlinear electrophoresis at We thank Dr. Ory Schnitzer for his suggestion and Dr. arbitrary field strengths: small-dukhin-number analysis,” Alice Thorneywork for her comments on the manuscript. Phys. Fluids 26, 122002 (2014). We also thank Bangs Lab and microParticles GmbH for [15] H. C. Chiu and H. J. Keh, “Electrophoresis of a colloidal providing us the information on the formation of mi- sphere with double-layer polarization in a microtube,” croparticles. S.T. is supported by Nakajima Founda- Microfluid. Nanofluid. 20, 64 (2016). tion Scholarship and John Lawrence Cambridge Trust [16] P. H. Wiersema, A. L. Loeb, and J. Th. G. Overbeek, International Scholarship. K.M. and U.F.K. acknowledge “Calculation of the electrophoretic mobility of a spheri- cal colloid particle,” J. Colloid Interface Sci. 22, 78–99 funding from an ERC consolidator grant (DesignerPores (1966). 647144). [17] D. J. Bonthuis and R. R. Netz, “Unraveling the combined effects of dielectric and viscosity profiles on surface ca- pacitance, electro-osmotic mobility, and electric surface conductivity,” Langmuir 28, 16049–59 (2012). [18] S. S. Dukhin, “Electrophoresis at large peclet numbers,” ∗ [email protected] Adv. Colloid Interface Sci. 36, 219–248 (1991). [1] N. Yukimoto, M. Tsutsui, Y. H. He, H. Shintaku, [19] O. Schnitzer and E. Yariv, “Macroscale description of S. Tanaka, S. Kawano, T. Kawai, and M. Taniguchi, electrokinetic flows at large zeta potentials: nonlinear “Tracking single-particle dynamics via combined optical surface conduction,” Phys. Rev. E 86, 021503 (2012). and electrical sensing,” Sci. Rep. 3, 1855 (2013). [20] M. Z. Bazant and T. M. Squires, “Induced-charge elec- [2] S. Liu, Y. Zhao, J. W. Parks, D. W. Deamer, A. R. trokinetic phenomena: theory and microfluidic applica- Hawkins, and H. Schmidt, “Correlated electrical and tions,” Phys. Rev. Lett. 92, 066101 (2004). optical analysis of single nanoparticles and biomolecules [21] S. Stotz, “Field dependence of the electrophoretic mobil- on a nanopore-gated optofluidic chip,” Nano Lett. 14, ity of particles suspended in low-conductivity liquids,” J. 4816–4820 (2014). Colloid Interface Sci. 65, 118–130 (1978). [3] H. Wu, H. Liu, S. Tan, J. Yu, W. Zhao, L. Wang, and [22] J. C. Thomas, K. L. Hanton, and B. J. Crosby, “Mea- Q. Liu, “The estimation of field-dependent conductance surement of the field dependent electrophoretic mobility change of nanopore by field-induced charge in the translo- of surface modified silica/aot suspensions,” Langmuir 24, cations of aunps-dna conjugates,” J. Phys. Chem. C 118, 10698–701 (2008). 26825–26835 (2014). [23] M. M. Gacek and J. C. Berg, “Investigation of surfac- [4] E. Angeli, A. Volpe, P. Fanzio, L. Repetto, G. Firpo, tant mediated acid-base charging of mineral oxide parti- P. Guida, R. Lo Savio, M. Wanunu, and U. Valbusa, cles dispersed in apolar systems,” Langmuir 28, 17841–5 “Simultaneous electro-optical tracking for nanoparticle (2012). recognition and counting,” Nano Lett. 15, 5696–5701 [24] Q. Guo, J. Lee, V. Singh, and S. H. Behrens, “Surfac- (2015). tant mediated charging of polymer particles in a nonpolar [5] J. H. Masliyah and S. Bhattacharjee, Electrokinetic liquid,” J. Colloid Interface Sci. 392, 83–89 (2013). and Colloid Transport Phenomena (Wiley-Interscience, [25] F. Strubbe, F. Beunis, T. Brans, M. Karvar, W. Woesten- Hoboken, 2006). borghs, and K. Neyts, “Electrophoretic retardation of [6] D. C. Henry, “The cataphoresis of suspended particles colloidal particles in nonpolar liquids,” Phys. Rev. X 3, part i - the equation of cataphoresis,” Proc. R. Soc. Lond. 021001 (2013). A 133, 106–129 (1931). [26] A. S. Khair, “Strong deformation of the thick electric [7] S. S. Dukhin, “Non-equilibrium electric surface phenom- double layer around a charged particle during sedimen- ena,” Adv. Colloid Interface Sci. 44, 1–134 (1993). tation or electrophoresis,” Langmuir 34, 876–885 (2018). [8] R. W. O’Brien and L. R. White, “Electrophoretic mo- [27] V. Shilov, S. Barany, C. Grosse, and O. Shramko, “Field- bility of a spherical colloidal particle,” J. Chem. Soc., induced disturbance of the double layer electro-neutrality Faraday Trans. 2 74, 1607–1626 (1978). and non-linear electrophoresis,” Adv. Colloid Interface [9] C. Grosse, “Out of equilibrium ion distribution around a Sci. 104, 159–173 (2003). suspended particle undergoing electrophoretic motion,” [28] O. Schnitzer, R. Zeyde, I. Yavneh, and E. Yariv, “Weakly Colloids Surf. A 467, 207–210 (2015). nonlinear electrophoresis of a highly charged colloidal [10] T. M. Squires, “Particles in electric fields,” in Fluids, Col- particle,” Phys. Fluids 25, 052004 (2013). loids and Soft Materials: An Introduction to Soft Matter [29] J. D. Sherwood and S. Ghosal, “Nonlinear electrophoresis Physics, edited by A. FernandezNieves and A. M. Puertas of a tightly fitting sphere in a cylindrical tube,” J. Fluid. (John Wiley & Sons, Inc., New York, 2016) pp. 59–79. Mech. 843, 847–871 (2018). [11] R. W. O’Brien, “The solution of the electrokinetic equa- [30] A. Kumar, E. Elele, M. Yeksel, B. Khusid, Z. Y. Qiu, tions for colloidal particles with thin double layers,” J. and A. Acrivos, “Measurements of the fluid and parti- Colloid Interface Sci. 92, 204–216 (1983). cle mobilities in strong electric fields,” Phys. Fluids 18, [12] S. B. Chen and H. J. Keh, “Axisymmetric electrophoresis 123301 (2006). 6

[31] N. A. Mishchuk, “Concentration polarization of interface silica surfaces,” J. Chem. Phys. 115, 6716–6721 (2001). and non-linear electrokinetic phenomena,” Adv. Colloid [38] M. Antonietti and L. Vorwerg, “Examination of the atyp- Interface Sci. 160, 16–39 (2010). ical electrophoretic mobility behavior of charged colloids [32] N. A. Mishchuk and N. O. Barinova, “Theoretical and in the low salt region using the o’brian-white theory,” experimental study of nonlinear electrophoresis,” Colloid Colloid. Polym. Sci. 275, 883–887 (1997). J. 73, 88–96 (2011). [39] M. Borkovec, S. H. Behrens, and M. Semmler, “Obser- [33] S. Pagliara, C. Chimerel, R. Langford, D. G. Aarts, and vation of the mobility maximum predicted by the stan- U. F. Keyser, “Parallel sub-micrometre channels with dard electrokinetic model for highly charged amidine la- different dimensions for laser scattering detection,” Lab tex particles,” Langmuir 16, 5209–5212 (2000). Chip 11, 3365–8 (2011). [40] M. Kobayashi and A. Sasaki, “Electrophoretic mobility [34] S. Deshpande, Y. Caspi, A. E. C. Meijering, and of latex spheres in mixture solutions containing mono C. Dekker, “Octanol-assisted liposome assembly on and divalent counter ions,” Colloids Surf. A 440, 74–78 chip,” Nat. Commun. 7, 10447 (2016). (2014). [35] M. Elimelech and C. R. Omelia, “Effect of electrolyte [41] S. Barany, “Electrophoresis in strong electric fields,” type on the electrophoretic mobility of polystyrene latex Adv. Colloid. Interfac. 147–148, 36–43 (2009). colloids,” Colloid Surface 44, 165–178 (1990). [42] M. R. Youssefi and F. J. Diez, “Ultrafast electrokinetics,” [36] B. R. Midmore, G. V. Pratt, and T. M. Herrington, Electrophoresis 37, 692–698 (2016). “Evidence for the validity of electrokinetic theory in the [43] K. H. Kang, X. C. Xuan, Y. J. Kang, and D. Q. Li, “Ef- thin double layer region,” J. Colloid Interface Sci. 184, fects of dc-dielectrophoretic force on particle trajectories 170–4 (1996). in microchannels,” J. Appl. Phys. 99, 064702 (2006). [37] S. H. Behrens and D. G. Grier, “The charge of glass and Supplemental Material: Nonlinear Electrophoresis of Highly Charged Nonpolarizable Particles

Soichiro Tottori, Karolis Misiunas, and Ulrich F. Keyser∗ Cavendish Laboratory, Department of Physics, University of Cambridge, JJ Thomson Avenue, Cambridge, CB3 0HE, United Kingdom

Douwe Jan Bonthuis Institute of Theoretical and Computational Physics, Graz University of Technology, 8010 Graz, Austria (Dated: July 10, 2019)

I. NONLINEAR TERMS IN PARTICLE ELECTROPHORESIS

The canonical model of electrophoresis consists of a spherical particle with constant surface potential ζ, in a medium with a constant dielectric constant , constant viscosity η, only electrostatic (no specific) ion-surface interactions and a no-slip hydrodynamic boundary condition on the surface of the particle. Using this model in the limit of low ζ and low applied electric field E, the electrophoretic mobility of a spherical particle is a linear function of ζ and E. The dependence on the particle size a and the inverse Debye length κ is given by the Henry function f(κa), 2ζ v = f(κa)E. (1) ep 3η For small values of κa the Henry function approaches unity, whereas for κa it approaches 3/2. At large values of ζ and intermediate values of κa, the mobility becomes a nonlinear function→ of∞ζ [1, 2]. In contrast, at both small and large values of κa, the electrophoretic velocity remains a linear function of ζ for all values of ζ. Studying the interfacial layer in molecular detail shows that the electrophoretic mobility of a spherical particle depends on its surface charge density σ in a complex way, depending on the inhomogeneous viscosity and the dielectric properties of the interfacial layer, as well as on specific ion-surface interactions and hydrodynamic slip [3]. Therefore, Eq. (1) effectively provides only a definition of the ζ potential as an effective potential that reproduces the particle’s electrophoretic mobility within the canonical model described above. Instead of defining a ζ potential and using Eq. (1), we can thus define the linear electrophoretic mobility µep(, η, κa, σ) and write

vep = µepE, (2) where the only necessary approximation is that the electric field is small: aEe k T [4]. Clearly, when extracting  B ζ from µep at intermediate values of κa and large values of the surface charge density with the aim of using ζ in a numerical solution of the canonical model, it should be kept in mind that a more accurate relation should be used instead of Eq. (1) [1], such as the one employed in the main text [5]. This situation changes at high electric field strengths, when the nonlinear terms in the Stokes-Nernst-Planck equa- tions become non-negligible. The magnitudes of the nonlinear terms depend on the surface charge density, as well as on the mobility of the particle and the ions. Here we give a brief explanation of the source of the nonlinearity. We start with the coupled Stokes and Nernst-Planck equations for a single ionic species,

J = D [ c + zc ψ∗] + uc, (3) − ∇ ∇ arXiv:1907.04278v1 [cond-mat.soft] 9 Jul 2019 with ψ∗ being the electrostatic potential normalized with respect to the thermal voltage ψ∗ = ψ/φth, c being the concentration of one type of ion species, J and u being the ion flux and fluid velocity, respectively, and D and z being the diffusion coefficient and the valence of the ionic species. In the steady state,

0 = [D c + zDc ψ∗ uc] . (4) ∇ · ∇ ∇ − When an electric field is applied, the system departs from equilibrium. The electric field strength E = ψ defines −|∇ |∞ a length scale at which the change of the potential is equal to the thermal voltage φth. To normalize the equations,

∗ [email protected] 2 this length scale is compared to the radius a of the particle, giving rise to the definition of the non-dimensional electric field β = aE/φth. Now we split the ion concentration and the potential into their equilibrium values, c0 and ψ0∗, and the perturbations due to the applied electric field, δc and δψ∗. In equilibrium, u = 0, and therefore [D c + zDc ψ∗] = 0, leading to ∇ · ∇ 0 0∇ 0 0 = [D δc + zDc δψ∗ + zDδc ψ∗ uc + zDδc δψ∗ uδc] . (5) ∇ · ∇ 0∇ ∇ 0 − 0 ∇ − To leading order, δc, δψ∗ and u are proportional to β. O’Brien and White employed a first order approximation with respect to the field β, neglecting the second-order terms (the final two terms) of Eq. (5) [4]. This procedure leads to an electrophoretic mobility which is a linear function of the applied electric field for all values of the surface charge density. Note that the extra flux due to the action of the equilibrium field on the perturbation of the double layer and the gradient of the double layer perturbation (both parts of the surface conductance) are taken into account in the linearized equations, and thus do not produce a nonlinear electrophoretic velocity per se, regardless of the magnitude of the surface potential. Therefore, any nonlinear effect must be produced by the final two terms of Eq. (5), which have in common that they are proportional to δc, i.e. to the distortion of the electric double layer upon application of the electric field. Rearranging Eq. (5) gives

auc0 au 0 = a δc + azc δψ∗ + azDδc ψ∗ + az δψ∗ δc . (6) ∇ · ∇ 0∇ ∇ 0 − D ∇ − D h   i For an electric field which is large on the scale of the particle, i.e. β 1, the final two terms of Eq. (6) cannot be ignored. The concentration perturbation δc depends linearly on the applied≥ electric field and on the ion concentration in the electric double layer [4]. Therefore δc will be larger for large surface charge densities, but the relation between δc, the bare surface charge density and the ζ potential estimated from mobility measurements is intricate and depends strongly on the interfacial dielectric environment, the interfacial viscosity and the specific ion-surface interactions. In the thin double-layer limit, the criterion for negligible concentration polarization is often expressed in terms of the Bikerman or Dukhin number at high values of ζ, Du = exp ζ /2/(κa) [6]. In the limit of vanishing Du, vep becomes a linear function of the electric field for any field strength,| | but as the present experiments show, a sizable nonlinear component is observed already for Dukhin numbers significantly smaller than 1 (Du 0.4 and Du 0.1 for the PS 620 nm and PMMA 520 nm particles, respectively). For the first term inside the curved≈ brackets, we≈ find a δψ∗/β 1, and therefore this term cannot be neglected regardless of the properties of the particle. The second term∇ in brackets∼ can be discarded if a u /(Dβ) 1, where u denotes the velocity in the direction of the electric field. In our experiments, both terms| inside| the curved backets| | are non-negligible. A detailed theoretical analysis using the canonical model can be found in the paper by Schnitzer et al. [7].

II. COUPLED STOKES POISSON NERNST-PLANCK SIMULATION

We used the coupled Stokes Poisson Nernst-Planck (SPNP) simulation as follows. The Poisson’s equation relates electric potential distribution ψ and volumetric free charge density,

 2ψ = F z c , (7) ∇ − i i X where , F , zi, and ci are permittivity of water, Faraday constant, valence, and ion concentration, respectively. The steady-state Nernst-Planck equation for each ion is (D c + z µ c F ψ) = u c , (8) ∇ · i∇ i i i i ∇ · ∇ i where Di, µi, and u are the diffusion coefficient, mobility of ion, and fluid velocity vector, respectively. Di and µi are related by Nernst-Eisntein relation as µi = Di/RT . The steady state Stokes and continuity equations are

η 2u p F z c ψ = 0, u = 0. (9) ∇ − ∇ − i i∇ ∇ · These equations are solved simultaneously with theX finite element method (COMSOL Multiphysics ver. 4.4) in the 2D axisymmetric domain shown in Fig. S1. A sphere is fixed at the center of the domain. The surface of the particle has a constant surface charge density σ, which is related to the surrounding potential as n ψ = σ, where n is the unit outward normal vector on the surface of the sphere. The mobility is calculated− as· the ∇ negative of the electroosmotic flow velocity far away from the particle surface. The physical parameters used for simulations are, temperature: 298.15 K, relative permittivity of water: 80, viscosity of water: 0.889 mPa s, diffusion coefficients of + 9 2 9 2 · ions K and Cl−: 1.957 10− m /s and 2.032 10− m /s, respectively. × × 3

E Rotational symmetry

Open boundary

a No slip boundary

FIG. S1. The boundary conditions for SPNP simulations (not drawn to scale).

III. ASYMPTOTIC MODEL OF ELECTROPHORETIC MOBILITY IN A LOW FIELD

Here we briefly describe how we estimated surface charge densities with the asymptotic model of electrophoretic mobility with relaxation effect by Chen and Keh [5]. The surface potential ψ0 and surface charge density σ under zero external field are related with the Grahame equation,

ψ σ = 2κφ sinh 0 . (10) th 2φ  th  The electrophoretic mobility with concentration polarization is calculated using [5],

ζ 1 µ = 2 + C + C + (C C ) ln cosh ζ¯ , (11) ep 3η 1 2 1 − 2 ζ¯  

1 ζ /2φth ¯ ¯ zeζ where C1,2 are the coefficients of κa and ζ, and C1,2 = 1/2 in the limit (κa)− e| | 0. ζ is described as ζ = 4kT . Eq. (11) is fitted to the experimental data points in Fig. 2 with least square curve fitting.→ The surface charge densities were calculated by assuming ζ = ψ0. The terms C1 and C2 are given as, 1 1 3 C = + β β β + β β β β /∆, (12a) 1 2 2 22 − 2 12 − 11 12 21 − 11 22  

1 1 3 C = + β β β + β β β β /∆, (12b) 2 2 2 11 − 2 21 − 22 12 21 − 11 22   where ∆ and βij are

∆ = 1 + β + β + β β β β , (13) 11 22 11 22 − 12 21

1 3f 12f β = 4 1 + 1 exp ζ¯ sinh ζ¯ 1 ζ¯ + ln cosh ζ¯ , (14a) 11 κa z2 − z2    

1 12f β = 1 ln cosh ζ,¯ (14b) 12 κa z2  

1 12f β = 2 ln cosh ζ,¯ (14c) 21 κa z2  

1 3f 12f β = 4 1 + 2 exp ζ¯ sinh ζ¯ + 2 ζ¯ ln cosh ζ¯ . (14d) 22 κa − z2 − z2 −    

The parameter f1, f2 and z were set f1 = f2 = 0.2 and z = 1, as suggested for KCl in [5, 8]. 4

IV. NONLINEAR TERM IN SCHNITZER AND YARIV’S MODEL

Here we describe the Schnitzer and Yariv’s model of nonlinear electrophoresis for small Dukhin numbers and arbitrary field strengths [9]. For small P´eclet numbers, the electrophoretic velocity vep is calculated as,

φ 2 v = th ζβ˜ + Duu + ... , (15) ep ηa 1     where ζ˜ is a dimensionless zeta potential ζ˜ = ζ/φth. Du is defined as, Du = (1 + 2α−)σ/ ˜ (κa), where α− is the 2 dimensionless ionic drag coefficient of counterion (not negative ion), α− = φth /(ηD−), andσ ˜ is the dimensionless surface charge density,σ ˜ = σ/ (κφth) . The dimensionless term u1 is given as,

˜ ζ 2 3 u1 4 ln cosh + ζ˜ β + β . (16) ∼ − " 4! # 21

Thus, the nonlinear velocity term vnl is,

φ 2 2 v th Du β3. (17) nl ∼ ηa 21   (3) 3 The cubic electrophoretic mobility µep (= vnl/E ) is,

(3)  (1 + 2α−) aσ µep , (18) ∼ 21ηφthze cNA

3 where Na is the Avogadro constant, and c is the concentration (mol/m ).

[1] P. H. Wiersema, A. L. Loeb, and J. Th. G. Overbeek, “Calculation of the electrophoretic mobility of a spherical colloid particle,” J. Colloid Interface Sci. 22, 78–99 (1966). [2] S. S. Dukhin, “Non-equilibrium electric surface phenomena,” Adv. Colloid Interface Sci. 44, 1–134 (1993). [3] D. J. Bonthuis and R. R. Netz, “Unraveling the combined effects of dielectric and viscosity profiles on surface capacitance, electro-osmotic mobility, and electric surface conductivity,” Langmuir 28, 16049–59 (2012). [4] R. W. O’Brien and L. R. White, “Electrophoretic mobility of a spherical colloidal particle,” J. Chem. Soc., Faraday Trans. 2 74, 1607–1626 (1978). [5] S. B. Chen and H. J. Keh, “Axisymmetric electrophoresis of multiple colloidal spheres,” J. Fluid. Mech. 238, 251–276 (1992). [6] S. S. Dukhin, “Electrophoresis at large peclet numbers,” Adv. Colloid Interface Sci. 36, 219–248 (1991). [7] O. Schnitzer and E. Yariv, “Macroscale description of electrokinetic flows at large zeta potentials: nonlinear surface con- duction,” Phys. Rev. E 86, 021503 (2012). [8] H. C. Chiu and H. J. Keh, “Electrophoresis of a colloidal sphere with double-layer polarization in a microtube,” Microfluid. Nanofluid. 20, 64 (2016). [9] O. Schnitzer and E. Yariv, “Nonlinear electrophoresis at arbitrary field strengths: small-Dukhin-number analysis,” Phys. Fluids 26, 122002 (2014).