WorksPhysics and Astronomy
Oscillating electroosmotic flow in channels and capillaries with modulated wall charge distribution
Abstract
Electrolyte-filled channels with modulated wall charge distribution subjected to an applied DC electric field, form time-independent vortices whose sense of circulation is determined by the field direction [Physical Review Letters ]. In this paper we show that an electrolyte in a channel or cylindrical capillary subjected to an external alternating (AC) electric field gives rise to various laminar flow structures, including vortices whose sense of circulation changes with the period of oscillation of the applied AC field. The introduction of a period of oscillation lifts certain degeneracies associated with its time-independent counterpart. Although, in general, the mass flux vanishes, the charge flux is nonzero. The flow is accompanied by a longitudinal (oscillating) advective current that displays hysteresis accompanied by a diverging and negative self-similar conductance that depends on the applied voltage [Nano Letters ]. We show that this behavior can be interpreted with respect to a ``memory retention time'', that depends on frequency, viscosity and the Debye length and could thus form the impetus for investigating control protocols of signal carriers.
Introduction
Microfluidic devices employ static channel networks to continuously activate liquids via time-independent drives. Such flows can be mediated by applied pressure gradients or electroosmosis in channels filled with ionic solutions. These static devices have been constructed using various methods such as lithography, micromilling, and injection moulding [1, 2, 3]. However, such predefined fabrications limit their functionalities in terms of applications that require an adjustable control of the fluid flow state. On the other hand, configurable microfluidic devices that are physically actuated by means of device assembly can be modified from a base state to a specified functional state. This has been achieved for example, through modular assembly, patterning hydrophobic surfaces, and paraffin structuring [4, 5]. Attaining regulated fluid flows in channels and capillaries that go beyond a fixed number of states is crucial for many microfluidic applications such as targeted solute delivery, mixing-induced chemical reactions, and volume-specified fluid transport.
Most studies of electrolyte transport assume that the channel walls are uniformly charged. However, acquired through fabrication or other routes, charge adsorption may change the surface charge distribution, for instance, during charged particle filtration, cf. [6]. Anderson [7] and Ajdari [8] determined the vortex flow structure of an electrolyte in channels with nonuniform charged walls. The vortices are time-independent and have a fixed sense of circulation determined by the direction of the (steady) applied DC electric field. These studies were followed up by experiments on patterning flows in soft-lithography polydimethylsiloxane (PDMS) and glass wall channels [5, 9] establishing consistency of the experimental results with theoretical models. More recently Kamsma et al. [10] showed numerically, that an inhomogeneous wall charge distribution on a conical channel exhibited stronger current rectification and more pronounced memristive properties compared to uniformly charged conical channels [10], if the charge distribution changes sign when traversing the channel along its long axis (what they call “bipolar” surface charges). A theoretical framework for understanding the combined effect of active and charged boundaries relevant in biological or soft matter systems was recently developed in Ref. [11].
The steady character of the aforementioned flows leads a microfluidic device to be classified as static or configurable such that the state of flow is either fixed at the manufacturing level and/or formatted to attain only limited predefined states [4]. Current scientific and technological needs, however, require the state of devices to be reconfigurable so to attain many different desired states ad voluntatem and in real time, cf. [12]. To address this requirement we propose here a way to generate time-dependent vortex and other laminar flows by applying an alternating (AC) electric field on a rectangular channel or cylindrical capillary with modulated charged walls. As a result, this mode of reconfiguration generates a multitude of flow states that can be actuated as desired. This work suggests a reconfigurable way to tune and control electroosmotic flow in microfluidic devices that provides an efficient means for mixing, targeted solute transport, pumping and energy conversion.
Electrolyte Flow with Modulated Wall Charge Distribution
It is well known [21 §2.10] that there are low Reynolds number flows which are unsteady. This requires the “vibrational” Reynolds number to not be small in comparison to the translational Reynolds number . Here and are characteristic length, translational velocity and frequency scales, respectively, while is the liquid kinematic viscosity. With this understanding, the Navier-Stokes equations reduce to the form
which can be employed to obtain estimates for the flow structure of time-dependent effects. Here, is the hydrodynamic pressure, is the mass density of the liquid, is its dynamic viscosity and a body force. In the context of this paper the body force is induced by an electric field acting on the free charges of an electrolyte.
We consider an external alternating electric field with (real) frequency acting on a electrolyte in a rectangular channel and in a cylindrical capillary. Because the walls are charged, an intrinsic electric field forms (unrelated to the external field in this discussion), providing the necessary equilibrium to the electrolyte charge species. When the potential is small (the Debye-Hückel approximation), it satisfies the modified Helmholtz equation
where is the Debye length. As in Ajdari [8], it is implicit in (2) that the charge distribution in the electrolyte bulk is given by where the dielectric constant (a consequence of the Debye-Hückel approximation, cf. Probstein [22]).
The total electric field exerts a body force on the liquid electrolyte (fluid + ions), whose velocity satisfies the Navier-Stokes equations (1)
where is an effective pressure. Since the expression is the gradient of a scalar function, this contribution to the body force has been absorbed into the pressure and it does not affect the form of the streamlines and the commensurate velocity field.
Motivation for studying this configuration stems from the experiments of Stroock et al. [9 figure 2], who obtained qualitatively similar flow patterns to the ones obtained here (albeit theirs being steady and expressed with respect to different boundary conditions). For clarity, our notation follows that of Ajdari [8, 23] who showed that steady vortices, in a channel with wall charge variation, can form in a time-independent electric field. In the adiabatic limit () our formulation recovers these time-independent results (this discussion is developed in Appendix A). In addition, we adopt the notation of Landau & Lifshitz [16, p.89] since our configuration features similarities with the corresponding problems of oscillating pressure-driven flows in a channel or capillary. For instance, the inertial term in (1) leads to the introduction of a penetration depth and complex wavenumber , see Eq. (13).
Rectangular channel flow with applied field parallel to surface charge variation
We consider a channel of width (see Figure 1) whose upper and lower walls are insulating and inhomogeneously charged with corresponding distributions that only vary in the -direction, satisfying the boundary conditions

Figure 1. AC electroosmosis with charge-modulated walls in a rectangular channel: A uniform alternating electric field of frequency exerts a body force on a electrolyte in a rectangular channel of width with surface charge on the upper and lower insulating walls, respectively. This gives rise to a system of vortices whose sense of rotation changes periodically in time according to the external electric field, cf. Figure 2.
There is an externally applied field parallel to the wall charge wave-vector , and in the bulk there is the potential generated by the surface charges and ions. The Navier-Stokes equations (3) become
Following [8] we consider two possible inhomogeneous surface charge distributions varying harmonically with wavevector , giving rise to the corresponding potentials as detailed below
and
where the symbol denotes surface charge distribution at the upper and lower channel walls, respectively, (see Figure 1), and
To determine the velocity field we take the curl of the Navier-Stokes equations (5) and introduce the streamfunction
| Quantity (units) | Definition |
| () | frequency of external field |
| () | electrolyte kinematic viscosity |
| (m) | liquid penetration depth (cf. Eq. (13)) |
| () | wall charge variation wavenumber |
| () | inverse Debye length |
| () | |
| () | liquid oscillating complex wave number (cf. Eq. (13)) |
| () | liquid complex wave number with charge variation (cf. Eq. (13)) |
| () | electrolyte dynamic viscosity |
| (m) | channel width, capillary diameter |
| wall charge distribution | |
| dielectric constant | |
| electric potential | |
| streamfunction, cf. (9) | |
| hydrodynamic and modified pressure | |
| (m/s) | channel horizontal and vertical electrolyte velocity components |
| (m/s) | capillary radial and axial electrolyte velocity components |
Table 1. Definitions of generic parameters and wavenumbers. , and as in [8]. and as in [16 §24].
The vorticity equation, written in terms of the streamfunction , then takes the form
When the -variation of the velocity field is proportional to and the time variation is harmonic with frequency (instantaneously following the field), the streamfunction
leads to the single equation for (with an abuse in notation)
where, with a slight abuse of notation, we’ve set in (12), and it is understood that is given by one of (6) or (7). Here,
is the familiar notation of the penetration depth and complex wavenumber of the oscillating pressure-driven flow in a channel, cf. [16 p.84-89] and is the kinematic viscosity of the electrolyte. Thus, the variation of has wavenumbers of the form
leading to a solution of the form , where is the particular solution associated with the last term in (12) and , , and are integration constants to be determined by the no-slip boundary condition at the channel walls
For their calculation see Appendix A. Two features distinguish the eigenvalue set in (14) from previous studies. First, the oscillatory character of our system has lifted the degeneracy present in the time-independent counterpart (where the four eigenvalues in (14) are instead two pairs of imaginary wavenumbers , cf. [8]). Second, the frequency of oscillation has introduced an additional length-scale to the problem (the penetration depth , cf. Eq. (13)) which competes with the existing scales , and .
For instance, in the symmetric charge distribution case leading to the electric potential (6), the streamfunction (11) satisfying the vorticity equation (10), prompts the function (by solving Eq. (12)) to acquire the form
with
This corresponds to the velocity components
and
In Figure 2 we display the dimensionless instantaneous streamlines from (16) (left) and (C1) (right) associated, respectively, with the symmetric surface charges and antisymmetric surface charges , along the channel walls. It is easy to see that the transition from the left pattern to the right occurs as the relative phase of the charge distribution on each wall is continuously varied from (in the notation of [8]) to . See the third column in [8 figure 2].

Figure 2. Oscillating vortex structure in a channel of width , induced by a uniform alternating electric field of frequency that is parallel to the wall charge variation wavevector , cf. Figure 1 for the coordinate system. Left: dimensionless instantaneous streamlines from (16) determined for symmetric surface charges along the channel walls. Right: dimensionless instantaneous streamlines from (C1), determined for antisymmetric surface charges . Compare with the upper and lower row of [8 figure 2]. It is thus clear that the pattern displayed on the left smoothly deforms and becomes the one to the right by continuously varying the relative phase of the charge distribution between the upper and lower channel walls (see discussion in the main text).
We introduce the velocity scaling factor
The unsteady streamfunction (18) reduces to its static counterpart (A3) in the limit . A distinguishing feature of the solution (16) in comparison with previous studies cf. Appendix A is that it has lifted another degeneracy; the one that exists in the time-independent problem when (although this limit is not expected to practically ever be reached), see the front factors of Eq. (A3)–(A4) in comparison to the front factors of (16) with (17).
Cylindrical capillary flow with axially applied field
The above analysis can be repeated for the case of a cylindrical capillary with circular cross-section of radius whose insulating wall carries a charge
cf. Figure 3. Again we consider the Debye-Hückel approximation where . Assuming , with a slight abuse of notation, Gauss law reduces to

Figure 3. AC electroosmosis in a cylindrical capillary with charge-modulated walls: A uniform alternating electric field of frequency exerts a body force on a 1 : 1 electrolyte in a cylindrical capillary of radius with inhomogeneous surface charges, on its insulating wall. This gives rise to a system of oscillating tori (whose cross-section are the displayed vortices) whose sense of rotation changes according to the external field’s, cf. Figure 4 for a calculated cross-section. Compare with figures 1 and 2 of the channel case.
Thus, the potential acquires the form
where and are modified Bessel functions of the first kind and with , defined in (8). If a uniform alternating electric field is applied along the axis of the cylinder (cf. Figure 3), it exerts a torque on the liquid of the form
where is the azimuthal unit vector in a system of circular cylindrical coordinates. The vorticity equation in terms of a streamfunction , expressed in cylindrical coordinates is well known cf. [21 §4.7] and we briefly revisit the salient points of its derivation. The vorticity equation can be written in the form
The flow is axisymmetric and only the radial and axial velocity components and , respectively, are non-vanishing. With a Stokes streamfunction satisfying
the incompressibility condition is automatically satisfied and the component of the vorticity can be written in the form
and this defines the operator (denoted by the symbol in [21], which however here is reserved to denote the electric field amplitude). It is then easy to show that the vorticity equation (25) becomes, cf. [21 Eq. (4-7.12)]
Assuming a streamfunction of the form (with a slight abuse of notation), its radially dependent factor satisfies
which is the counterpart of (12), and defined in (8) and (13), respectively. The general solution of (29), that is finite at , is of the form
where is the Bessel function of the first kind and and are arbitrary constants to be determined by the boundary conditions. These are the no-slip conditions at the capillary walls
The two constants are
where is the Bessel function of the first kind, of zeroth order.
In Figure 4 we display a cross-section of the flow in the plane. The flow consists of identical vortices, which, when extended along the azimuthal direction , form tori which alternate in their sense of circulation as one moves parallel to the axis of the cylinder (in the -direction). The circulation direction changes periodically depending on the orientation of the electric field . Employing (26) and (32) shows that the radial velocity is proportional to and thus vanishes when the wall charges are spatially uniform.

Figure 4. Oscillating toroidal vortex structure in a capillary of radius , induced by a uniform alternating electric field of frequency that is parallel to the wall charge variation wavevector (parallel to the cylinder axis), cf. Figure 3 for the coordinate system. Here we display an slice of the dimensionless instantaneous streamlines with from (29) and (32) determined for charges on the capillary wall located at .
Rectangular channel flow with applied field perpendicular to surface charge variation
Next consider an external alternating electric field that is applied perpendicular to wall charge variation with symmetric or antisymmetric wall distributions given corresponding potentials Eq. (6) and (7), satisfying Eq. (2) in the Debye-Hückel approximation as in the previous section. This configuration represents the oscillatory counterpart of Ajdari [8], Eq. (9) & Eq. (10).
The momentum equation becomes in this case
where in Cartesian coordinates. The vorticity equation shows that
and that . As in [8 Eq. (9) & (10)], a solution that satisfies no-slip boundary conditions at is . The velocity component satisfying (34) obtains the form , thereby
where we employed the electric potential (6) determined when both walls carry the same charge (). We recover the case where the electric potential employed is (7) (determined when the walls carry opposite charge), by performing the exchange in (35).
The resulting velocity is
where was defined in (13). The velocity profile that corresponds to the electric potential in Eq. (7), is obtained from (36) by performing the exchange and , leaving intact. It is clear that (36) recovers previous static results in the adiabatic limit . Thus, setting in (36) and employing the identity , we recover the (static) velocity profile of Ajdari [23 Eq. (26)].
A snapshot of the velocity profile (36), scaled by the factor (20) which here is denoted by , that is,
is displayed in the bottom left panel of Figure 5 over half period in the -direction. This is reminiscent of the experimental results of Stroock et al. [9, figure 2]. Moreover, it is seen that the profile is oscillating but of single sign (in the chosen interval of the -axis) and can thus be employed for unidirectional transport of electrolytes, as discussed in the following section. The bottom right panel of Figure 5 displays the scaled velocity profile by , when the upper and lower channel walls carry charges of opposite sign. Here the flow, averaged over the width of the channel, vanishes. Such a pattern can also be developed for a symmetric wall charge distribution (cf. Figure 12 in Appendix § 17).

Figure 5. AC electroosmosis in a rectangular channel with applied field perpendicular to wall charge variation. Top: An alternating electric field of frequency exerts a body force on a 1 : 1 electrolyte in a rectangular channel of width with symmetric inhomogeneous wall charges, on the upper and lower insulating walls. Over each tile, the charge distribution gives rise to a non-vanishing velocity averaged over the channel width (36). Bottom: Oscillating velocity profile in a channel of width , induced by a uniform alternating electric field of frequency that is perpendicular to the wall charge variation for the coordinate system. Snapshot of the velocity profile from (36) at time when the surface charges at the channel walls are symmetric (left). Snapshot of the velocity profile (by exchanging the cosines with sines in (36), and vice-versa) at time when the surface charges at the channel walls are antisymmetric (right).

Figure 12. Electrolyte (oscillating) velocity profile scaled by the factor , in a channel of width , induced by a uniform alternating electric field that is perpendicular to the wall charge variation, cf. Figure 1 for the coordinate system. The left column displays the velocity profiles (G9) with symmetric charge distribution giving rise to corresponding potentials (G4). The right column displays the velocity profiles (G10) with symmetric charge distribution giving rise to corresponding potentials (G5). The upper left profile when the surface charges at the channel walls are symmetric , can be employed for unidirectional transport of an electrolyte. This flow pattern is qualitatively similar to the ones obtained experimentally in [9 figure 2] (albeit theirs being steady and expressed with respect to different boundary conditions).
AC-INDUCED UNIDIRECTIONAL ELECTROLYTE MOTION IN CHANNELS AND CAPILLARIES
In addition to the vortex flow patterns of the previous sections, it is of practical interest to examine the presence of unidirectional electrolyte transport. In this section we calculate the velocity averaged over the channel width for the perpendicular field to the charge variation, and examine their limiting behaviors. The averaged velocity is
Consider the case of uniform wall charges along both walls. The motivation for the choice of symmetric wall charge comes from the observation that an optimal unidirectional velocity is expected to arise when the wall excitations are identical and of the same phase [24]. The configuration where the velocity does not vanish when averaged over the channel width, is the one with symmetric charges along the channel walls.
Average velocity in rectangular channel with field perpendicular to charge variation
In the case of the perpendicular applied field, the velocity profile (36) associated with the symmetric charge on the walls , gives rise to unidirectional charge transport. Averaging (36) over the channel width we obtain
In the left two panels of Figure 6 we display the absolute value of the amplitude contained in the square brackets of (39), for various values of the wall charge variation wavenumber .

Figure 6. Unidirectional channel averaged velocity with perpendicular applied AC field. Left panel displays the velocity amplitude versus the scaled frequency for various wavenumbers of wall charge variation, where . The plateau at the low frequency values denotes the plug-like adiabatic limit. Middle panel: Amplitude of the scaled unidirectional velocity (39) for the perpendicular case as a function of for different values of frequency , recovering the static case of Ajdari [23] in the adiabatic limit. The right panel displays the velocity (39) in the limit.
Of interest is also the adiabatic limit (and thus ). Defining the Langevin function
the limit of (39) becomes
(over times which satisfy ). Since the argument in this paper is considered to be rather large, the Langevin function appearing in (41) is approximately equal to 1. The resulting velocity is of a type obtained in standard (static) electroosmosis formulations, cf. [25 p. 10.24]. We note that this is a plug-like induced by the large values of the inverse Debye wavelength and it differs from the plug-like flow that forms when , cf. Appendix B.
In the right panel of Figure 6 we display the absolute value of the amplitude contained in (39) for versus the scaled frequency where the velocity scale was defined in (37). For low scaled frequency values the velocity amplitude reaches a plateau (to the left of each curve) denoting the plug-like adiabatic limit (41), expressed with respect to a Langevin function (40) and resembles the standard (static) electroosmosis expression met in the literature. For large frequencies, in principle, the rapid oscillation of the complex exponential should also be taken into account.
Estimates
From figure 6 and the long wavelength expression (41) it can be seen that for low frequencies the electrolyte velocity is of the order of the scaled parameter defined in (37). For higher frequencies the velocity magnitude is reduced. Thus, we can employ
as an upper bound that the electrolyte velocity can attain. Employing standard parameter values for water , , and (as in [9]) and a surface charge with the nominal value we obtain the estimate
This value is of the same order of magnitude as standard electroosmosis velocity estimates (cf. Probstein [22 below (6.5.5)]) and one order of magnitude higher than those of the experiments of Stroock et al. [9].
Unidirectional velocity in rectangular channel with field parallel to charge modulation
In the limit the velocity in the rectangular and cylindrical cases becomes unidirectional and oscillatory in the direction (channel case) and in the direction (cylinder case). When averaged over the width of the channel/radius of cylinder, they both vanish. The convective current, however, is non-zero for all values of . This observation implies that the oscillatory electric field in a capillary with wall charge variation gives rise to hysteretic flow (evident from the presence of the renormalized Stokes wavenumber in (13)) and is thus endowed with traces of memory, a subject we develop in the ensuing sections.
Memory Effects
Recent work has focused on imitating physiological processes taking place in a synapse to create highly efficient iontronic computing devices, carrying information via ion transport in an aqueous environment rather than via electrons and holes in the crystal lattice, and will thus be manipulated by both chemical and electrical signals.
Figure 7. Left panel: Hysteresis loops of advective current from Eq. (46) vs.-voltage for various values of the wall charge wavevector . Right panel: Area between advective current (46) and voltage. Currents in the left panel were evaluated at frequencies that maximize the respective areas. Thus, although the average velocity over the width of the channel vanishes, the current remains nonzero and gives rise to traces of memory, see the ensuing discussion. Both panels refer to the electrolyte oscillatory channel flow (18) developed in § 2 A when both walls carry the same charge distribution giving rise to the potential (6).
The plot of vs. in an iontronic device shows a hysteresis effect cf. [26 Fig. S13-S15], and this is interpreted as sign of “memory”. The largest the enclosed area, the largest the memory is. The area of the hysteresis loops can be calculated as a function of the frequency of the applied voltage. The frequency that maximizes the area of the loop determines the characteristic ‘memory’ time , cf. [27 Eq.(3) & Fig. 3C] and [10 Fig. 3(c)].
For the static case, Osterle and coworkers [28] pointed out that although the current in the transverse direction to the capillary walls vanishes, the current in the longitudinal direction (parallel to the walls) is nonzero, cf. [28 Eq. (23)]. This remains true in the present oscillatory configuration. We take advantage of this concept to show that rectangular channels or cylindrical capillaries in the presence of wall charge variation give rise to a current, which is hysteretic with respect to the applied voltage.
Hysteresis loops in the literature appear to be pinched (current vanishes when the voltage does). It was however argued by Pershin, Di Ventra and coworkers [17, 18, 29] that pinching is not a necessary defining characteristic displayed by a system with memory. In the context of ionic systems, it is evident from the experimental figures of Kamsma et al. [26 Fig. S13-S15] that hysteresis loops can be open. Open hysteresis loops, displaying capacitive behavior, can also be seen in recent theoretical studies, for instance, in Figures 6Aa/d and 11B of [30]. Below, we show that the natural behavior of an hysteresis loop emanating from the advective longitudinal current, within our approximate formulation, is not pinched.
Negative resistance or differential resistance has been documented in a number of recent works [19, 20, 30, 31, 32, 33]. Likewise, we will here show that the conductance displayed by our ionic system can become negative and divergent as the external voltage approaches zero. This behavior is reminiscent of capacitive systems, for instance see Figure 3 of [17] and Figure 3B of [19], both showing negative capacitance in experiments whose behavior resembles the right-most panel of Figure 8.
Figure 8. Channel geometry. Left panel: Advective current from Eq. (47) and with the scaling from (44), with arrows denoting the direction of voltage sweep in time, vs.-voltage hysteresis loop evaluated at Hz which maximizes the loop area (i.e. maximum energy dissipation, cf. the area maximum in the center panel). Center panel: Area between advective current (47) and voltage. The frequency that maximizes the area of the loop determines the characteristic ‘memory’ time (48). Right panel: Dimensionless memconductance vs. dimensionless voltage Eq. (49). Numbers indicate the order at which the voltage is swept (denoted by arrows). The memconductance diverges as the voltage approaches its zero value. This figure should be compared to Krems et al. [17 Figs. 4 & 5]. Inset: same figure but for small memconductance values. All panels refer to the electrolyte oscillatory channel flow developed in § 2 A when both walls carry the same charge distribution giving rise to the potential (6).
For discussion in this memory section, below we will need the following current and voltage scalings
where is the length of the channel or cylindrical tube over which the voltage is applied.
Channel memory
The longitudinal current , averaged over the width of the channel becomes
where is given by (18) and in the Debye-Hückel approximation is the bulk conductivity. With the scaling (44) we nondimensionalize the current (45) and obtain . The coefficient can be understood as an Osterle or Levinson number [34 Eq. (1.25)], and considering standard material parameters it can be shown that it is of . Since this term represents an oscillation with the same phase as the driving field, it is of limited interest; we concentrate instead on the first term of the right hand side of (45) which is the advective current in the channel with symmetric wall charge distribution, that is the electric potential is (6), $\rho=-\epsilon\kappa^2\phi$ and the velocity field is Eq. (18)`. Simple dimensional analysis arguments show that the advective current is self-similar
with respect to the stated dimensionless parameters, where is the Stokes penetration depth defined in (13). In (46) we also averaged over one period of wall charge variation in the longitudinal direction . In the left panel of figure 7 we display the advective current (46) vs. the voltage for various values of the wall charge wavenumber . In the right panel of the same figure we display the area of the - loops. This figure establishes what we have already stressed, that, although the average velocities over the channel width vanish, the advective current is non-zero.
The advective current (46) is a rather long expression, which simplifies somewhat in the limit of
which, can likewise become of the same order as the second term in (45) by a prudent choice of the parameters. On the left-most panel of figure 8 we display the hysteresis loop formed by the advective current (47) and the applied voltage . On the center panel of the same figure we display the area between the advective current (47) and voltage vs. the dimensionless group . The frequency that maximizes the area of the loop, determines the characteristic ‘memory’ time , cf. [27 Eq.(3) & Fig. 3C] and [10 Fig. 3(c)]. For standard material parameters, and we obtain the estimate
We note that the above analysis can be repeated also for the antisymmetric wall charge distribution leading to the electric potential (7) and the corresponding velocity fields.
Channel memconductance
Following the definition of memconductance introduced by [13, 35], cf. Appendix E, we consider expression (47) (and thus for simplicity, we consider only the case from now on) for the advective current and form the conductance , which can take the somewhat simpler self-similar form
where , is the channel cross-sectional area over which the current is flowing, and we remind the reader that , cf. Eq. (13). in (49) is history-dependent. In the left-most panel of figure 8 we display the current-voltage hysteresis loop as the voltage is swept and in the right-most panel we display the memconductance (49). It is clear that the conductance diverges as the voltage approaches its zero value and can become negative. This behavior should be contrasted to Figures 4 & 5 of Krems et al. [17] whereby the models employed and the molecular dynamics simulations displayed in the insets therein, have exactly the behavior displayed in Figure 8.
In Appendix F we argue that this behavior is generic for systems whose - hysteresis loops are ellipses (as the present loops are).
Cylinder memory
We repeat the above for the case of the cylinder, whose axial velocity is (26) with (30). The advective current is again self-similar
with respect to the stated dimensionless parameters, where is the Stokes penetration depth defined in (13) and in (50) we also averaged over one period of wall charge variation in the longitudinal direction . The advective current (50) again simplifies somewhat in the limit of
In the left panel of figure 9 we display the hysteresis loop between the advective current and the applied voltage . In the center panel of the same figure we display the area between the advective current (51) and voltage vs. the dimensionless group . The frequency that maximizes the area of the loop, determines the characteristic ‘memory’ time which is of the same order of magnitude as estimate (48) in the channel case.
The memconductance in the cylinder case is given again by Eq. (49) by effecting the change and considering as cross-sectional area . We display in the right-most panel of Figure 9 the memconductance which diverges as the voltage approaches its zero value and can become negative. This behavior can again be contrasted to Figure 4 & Figure 5 of Krems et al. [17].
Figure 9. Cylinder geometry. Left panel: Advective current from Eq. (51) and with the scaling from (44), with arrows denoting the direction of voltage sweep in time, vs.-voltage hysteresis loop evaluated at Hz which maximizes the loop area (i.e. maximum energy dissipation, cf. the area maximum in the center panel). Right panel: Area between advective current (51) and voltage. The frequency that maximizes the area of the loop determines the characteristic ‘memory’ time . Right panel: Dimensionless memconductance vs. dimensionless voltage Eq. (49) (with ). Numbers indicate the order at which the voltage is swept (denoted by arrows). The memconductance diverges as the voltage approaches its zero value. This figure should be compared to Krems et al. [17 Figs. 4 & 5]. Inset: same figure but for small memconductance values. All panels refer to the electrolyte oscillatory flow in a cylinder developed in section II B when the wall carries the same charge distribution at , giving rise to the potential (23).
Discussion
Electroosmosis is an effective means of charge transport, whereby the electrolyte adjacent to a uniformly charged wall is set into motion by an externally applied (DC) electric field. Here we showed instead that in an AC field a capillary with nonuniformly charged walls, it will give rise to vortex and other laminar flow behavior. These types of flows can be employed for mixing and solute transport. The vortex flow patterns obtained resemble those observed in experiments with soft-lithography polydimethylsiloxane (PDMS) and glass wall channels [5, 9]. In addition the non-vortex patterns shown in Appendix § 17 where the wall charge distribution can change sign (once) could lead to stronger current rectification as shown in a recent numerical comparative study [10]. We established the presence of unidirectional velocity as developed in Figure 6 and discussed in section III A.
All oscillatory systems we examined in sections II A and II B show traces of memory. Although the velocity field vanishes after averaging over the width of the channel or cylinder cross-section, the advective currents are non-zero, giving rise to - hysteresis loops, which are thereby employed to develop a concept of memory time. The concept of memory developed in section IV, and in particular the negative and diverging conductance in the context of ionic solutions in a channel or cylindrical capillary, as displayed in the right-most panel of Figure 8, resembles the presence of negative and diverging capacitance in related theoretical [17] and experimental works [19]. Analogous concepts were developed in the past in the context of solid-state configurations, see for instance [18, 36].
Multiple alternative configurations can likewise be studied. For instance, one can envision flow in the annular region between two pipes of radii and with the same or opposite wall charge distribution. Then, in addition to the Bessel functions and employed in section II B, one has to consider their (diverging at the origin) counterparts and , respectively. Hence, the vortex pattern will resemble the left or right panel of Figure 2, depending on the relative phase of charge modulation between the inner and outer tubes.
Acknowledgments
We acknowledge the support of the NSF grant number DMR-2452280. We are grateful to an anonymous referee for suggesting the differential operator decomposition in Eq. (29) and for providing numerous comments that significantly improved the manuscript.
The steady streamfunction of Ajdari [23]
When the electric field is static [8, 23], the vorticity equation in terms of the streamfunction (9), acquires the form
When the -variation of the velocity field is proportional to , the streamfunction leads to the single equation for
which is the counterpart of Eq. (12) by setting in (13). This form of solution carries a certain degeneracy associated with the twice-repeated wavenumbers . As shown by Ajdari [23 Eq. (9) & (15)-(19)], for the antisymmetric wall charge distribution giving rise to the electric potential (7), the stationary streamfunction becomes
where the subscript denotes the integration constants of the static problem, , and
Flow due to time-harmonic pressure gradient for channel and capillary
We follow the solution of [16 p.89]. A harmonic in time pressure gradient inside a rectangular channel gives rise to a flow in the -direction (in our notation) satisfying the -momentum equation
where the units of the scalar are implied. The solution, satisfying no-slip boundary conditions at is
where , as considered in this article. Its average value is
The adiabatic, small limit leads to which is of Poiseuille type. The opposite limit of fast frequencies leads to a plug-type velocity profile with a thin boundary layer adjacent to the channel walls.
For the case of an oscillating pressure gradient in a cylindrical capillary of radius we consider the solution derived by Drazin & Riley [37 §5.1.2]. The velocity field in the axial direction satisfies the Navier-Stokes for a liquid in a cylinder of radius under an oscillating pressure gradient has the form
Its solution is
where is given by the usual expression , cf. (13).
Antisymmetric charge distribution in a channel
For the asymmetric charge distribution leading to the electric potential (7), the streamfunction satisfying the vorticity equation (10), leads the function (solving Eq. (12)) to acquire the form
with , and
Limiting behaviors
Below we consider a limiting behavior that serves as reality check of our formulation. In the presence of DC electric field only, the above expressions for the streamfunction should reduce to the corresponding time-independent expressions of [8, 23]. We will perform this only for the unsteady streamfunction (C1) which was derived by consideration of antisymmetric channel wall charges leading to the electric potential (7). Let the complex wavenumber in (13) be given by ( are real numbers), where
and the sign corresponds to the real () and imaginary () parts of , respectively. First, it is clear that in the limit, the particular solutions in (A3) are identical. We thus consider the effect of the limit on the combination in the unsteady streamfunction. In this limit, each one of these terms gives rise to identical leading order expressions proportional to , but with the opposite sign, hence they cancel out. Retaining then the terms only, shows that they recover the steady streamfunction of Ajdari [23 Eq. (9) & (15)–(19)], denoted here by the terms in (A3), as a special case of our unsteady streamfunction.
Systems with memory
We repeat here the definition of systems with memory [35 Eq. (28) & (29)], driven by an applied voltage and resulting in a current
where is a memory conductance and is a vector of state variables.
Generic form of diverging conductance
A diverging conductance is expected to be present every time the current is nonzero when the voltage is zero. Here and in [17] it is clear that the current-voltage and charge-voltage loops are ellipses. This can be seen if we set the complex factor of in (47) to equal , where and are real. Thus, (47) becomes
From elementary calculus it can now be shown that (F1) along with constitute the parametric equation of an ellipse.
For small voltages and in normal coordinates, one has , where is the ordinate in normal coordinates and the sign denotes the branch of the ellipse. A diverging conductance at ensues when one expresses this current-voltage relation as a linear function of voltage: , whereby
We display Eq. (66) in Figure 10. The resemblance to the conductance that appears in the right-most panel of Figure 8 and Figure 9 is clear.
Figure 10. Conductance (F2) of an ellipse - system expressed in normal coordinates. It resembles the conductance in the right-most panel of Figure 8 and Figure 9, as well as the memcapacitance of [17] and [19]. This is a generic feature of conductance (capacitance) when the current (charge) is finite at vanishing voltage. The positive (negative) branch of the ellipse corresponds to the plus (minus) sign expression in (F2).
Channel AC electroosmosis with hyperbolic wall charge distributions
Wall charge distributions alternative to the sinusoidal ones of section II may also be employed. Apart from leading to closed-form analytical predictions, another motivation behind such choice is that they are reminiscent of those recently employed [10], to numerically establish the preeminence of the commensurate current rectification in contrast to uniformly charged channel walls [10].
Consider an external uniform alternating electric field with (real) frequency acting on a 1:1 electrolyte in a rectangular channel of width (Figure 1) whose upper and lower walls are insulating and inhomogeneously charged with one of the following symmetric or antisymmetric distributions: , , and . Thus, there will be four distinct potentials associated with one of these distributions satisfying Eq. (2), in the Debye-Hückel approximation. These are displayed below and equations (G4) and (G5) for symmetric and antisymmetric charge distributions on the upper and lower walls, respectively.
The vorticity equation, written in terms of , again takes the form (10) Assuming a streamfunction of the form
leads to the single equation for
where is the complex wavenumber introduced in (13), is defined through (or ) and is one of (G4) or (G5). The variation of has wavenumbers of the form
leading to a solution of the form , where is the particular solution associated with the last term in (G2) and thus determined accordingly to the form of one of the four potentials (G4) and (G5). , , , and are integration constants to be determined by the no-slip boundary condition at the channel walls (not shown). In Figure 11 we display streamlines (left and right columns), corresponding to symmetric and antisymmetric wall charge distributions, respectively.

Figure 11. Electrolyte (oscillating) laminar flow structure in a channel of width , induced by a uniform alternating electric field , determined by solving (G2), cf. Figure 1 for the coordinate system. Here we display the dimensionless instantaneous streamline pattern determined for symmetric (left column) surface charges and antisymmetric (right column) surface charges along the channel walls , corresponding to the potentials (G4) and (G5), respectively.
Below we only consider the symmetric wall charges of the form which, in the limits and could be employed for unidirectional electrolyte transport. The remaining three velocity profiles all vanish when averaged over the width of the channel. The motivation behind the study of these configurations lies in recent numerical experiments [10], which established that current rectification is preeminent when the charge wall changes sign, in comparison to uniformly charged channel walls.
Consider an external uniform alternating electric field with (real) frequency acting on a electrolyte in a rectangular channel of width (see Figure 1) whose upper and lower walls are insulating and inhomogeneously charged with one of the following symmetric or antisymmetric distributions: , , and . Thus, there will be four distinct potentials associated with one of these distributions satisfying Eq. (2), in the Debye-Hückel approximation, having the form
and
The hyperbolic function charge distributions require a redefinition of the parameters and which are
where was defined in (13).
Again, the velocity component satisfies (34) or
Assuming a velocity field of the form (or ), the function satisfies
where is defined as (or ) and is one of the four potentials in (G4) and (G5). Thus, the velocity field acquires one of the following one forms
for symmetric wall charges, and
for antisymmetric ones. In Figure 12 we display these four velocity profiles scaled by the factor (37). The left column displays the velocity profiles (G9) with symmetric charge distribution giving rise to corresponding potentials (G4). The right column displays the velocity profiles (G10) with symmetric charge distribution giving rise to corresponding potentials (G5). The upper left flow pattern is qualitatively similar to the ones obtained experimentally in [9 figure 2] (albeit theirs being steady and expressed with respect to different boundary conditions). We’ll show below that it can be employed for unidirectional electrolyte transport.
When , the velocity profile in (G9) averaged over the channel width leads to
which is identical to (39) if we replace with and with . Below we only consider the symmetric wall charges of the form which, in the limits and could be employed for unidirectional electrolyte transport. The remaining three velocity profiles all vanish when averaged over the width of the channel. In the limit it becomes
Thus, in the long wavelength limit the behavior of the horizontal component of the velocity can also be described by the curves displayed in Figure 6 for the amplitude of the normalized average unidirectional velocity (G12), versus the scaled frequency , and is defined in (37). The adiabatic limit leads to same behavior expressed through the Langevin function defined in (40).
References
- [1]H. Stone, A. Stroock, and A. Ajdari, Engineering flows in small devices: microfluidics toward a lab-on-a-chip, Annual Review of Fluid Mechanics 36, 381 (2004).DOI
- [2]T. Squires and S. Quake, Microfluidics: Fluid physics at the nanoliter scale, Reviews of Modern Physics 77, 977 (2005).DOI
- [3]D. Erickson and D. Li, Integrated microfluidic devices, Analytica Chimica Acta 507, 11 (2004).DOI
- [4]F. Paratore, V. Bacheva, M. Bercovici, and G. Kaigala, Reconfigurable microfluidics, Nature Reviews Chemistry 6, 70 (2022).DOI
- [5]A. Stroock and G. Whitesides, Controlling flows in microchannels with patterned surface charge and topography, Accounts of Chemical Research 36, 597 (2003).pubmed.ncbi.nlm.nih.gov/12924956
- [6]P. Chowdiah, D. Wasan, and D. Gidaspow, Electrokinetic phenomena in the filtration of colloidal particles suspended in nonaqueous media, AIChE Journal 27, 975 (1981).DOI
- [7]J. Anderson and W. Idol, Electroosmosis through pores with nonuniformly charged walls, Chemical Engineering Communications 38, 93 (1985).DOI
- [8]A. Ajdari, Electro-osmosis on inhomogeneously charged surfaces, Physical Review Letters 75, 755 (1995).pubmed.ncbi.nlm.nih.gov/10060106
- [9]A. Stroock, M. Weck, D. Chiu, W. Huck, P. Kenis, R. Ismagilov, and G. Whitesides, Patterning electro-osmotic flow with patterned surface charge, Physical Review Letters 84, 3314 (2000).pubmed.ncbi.nlm.nih.gov/11019078
- [10]T. Kamsma, W. Boon, T. ter Rele, C. Spitoni, and R. van Rooij, Iontronic neuromorphic signaling with conical microfluidic memristors, Physical Review Letters 130, 268401 (2023).DOI
- [11]A. Shrestha, E. Kirkinis, and M. Olvera de la Cruz, Self-generated electrokinetic flows from active-charged boundary patterns, Physical Review Research 7, 023223 (2025).arxiv.org/abs/2412.15397
- [12]A. Shrestha, E. Kirkinis, and M. Olvera de la Cruz, Universal behaviour in boundary-driven electrokinetic flows, Journal of Fluid Mechanics 1010, A50 (2025).DOI
- [13]L. Chua and S. Kang, Memristive devices and systems, Proceedings of the IEEE 64, 209 (1976).DOI
- [16]L. D. Landau and E. M. Lifshitz, Fluid Mechanics. Course of Theoretical Physics, Vol. 6 (Pergamon Press Ltd., London-Paris, 1987) pp. xii+515.
- [17]M. Krems, Y. Pershin, and M. Di Ventra, Ionic memcapacitive effects in nanopores, Nano Letters 10, 2674 (2010).DOI
- [18]J. Martinez-Rincon, M. Di Ventra, and Y. V. Pershin, Solid-state memcapacitive system with negative and diverging capacitance, Physical Review B—Condensed Matter and Materials Physics 81, 195430 (2010).
- [19]D. Wang, M. Kvetny, J. Liu, W. Brown, Y. Li, and G. Wang, Transmembrane potential across single conical nanopores and resulting memristive and memcapacitive ion transport, Journal of the American Chemical Society 134, 3651 (2012).DOI
- [20]L. Luo, D. Holden, W.-J. Lan, and H. White, Tunable negative differential electrolyte resistance in a conical nanopore in glass, ACS Nano 6, 6507 (2012).DOI
- [21]J. Happel and H. Brenner, Low Reynolds number hydrodynamics with special applications to particulate media (Prentice-Hall Inc., Englewood Cliffs, N.J., 1965) pp. xiii+553.
- [22]R. F. Probstein, Physical Hydrodynamics: An Introduction (John Wiley & Sons, 1994).
- [23]A. Ajdari, Generation of transverse fluid currents and forces by an electric field: Electro-osmosis on charge-modulated and undulated surfaces, Physical Review E 53, 4996 (1996).pubmed.ncbi.nlm.nih.gov/9964829
- [24]H.-C. Yeh, R.-J. Yang, and W.-J. Luo, Analysis of traveling-wave electro-osmotic pumping with double-sided electrode arrays, Physical Review E 83, 056326 (2011).pubmed.ncbi.nlm.nih.gov/21728666
- [25]J. Melcher, Continuum electromechanics (MIT press, Cambridge, MA, 1981).
- [26]T. Kamsma, J. Kim, K. Kim, W. Q. Boon, C. Spitoni, J. Park, and R. van Rooij, Brain-inspired computing with fluidic iontronic nanochannels, Proceedings of the National Academy of Sciences 121, e2320242121 (2024).DOI
- [27]P. Robin, T. Emmerich, A. Ismail, A. Niguès, Y. You, G.-H. Nam, A. Keerthi, A. Siria, A. Geim, B. Radha, et al., Long-term memory and synapse-like dynamics in two-dimensional nanofluidic channels, Science 379, 161 (2023).DOI
- [28]F. Morrison and J. Osterle, Electrokinetic energy conversion in ultrafine capillaries, The Journal of Chemical Physics 43, 2111 (1965).DOI
- [29]Y. Pershin and M. Di Ventra, Comment on ‘if it’s pinched it’s a memristor’, Semicond. Sci. Technol. 34, 098001 (2019).DOI
- [30]J. Bisquert, Inductive and capacitive hysteresis of current-voltage curves: Unified structural dynamics in solar energy devices, memristors, ionic transistors, and bioelectronics, PRX Energy 3, 011001 (2024).DOI
- [31]J. Perez-Grau, P. Ramirez, V. Garcia-Morales, J. Cervera, S. Nasir, M. Ali, W. Ensinger, and S. Mafe, Fluoride-induced negative differential resistance in nanopores: Experimental and theoretical characterization, ACS Applied Materials & Interfaces 13, 54447 (2021).
- [32]J. Bisquert, Iontronic nanopore model for artificial neurons: The requisites of spiking, The Journal of Physical Chemistry Letters 14, 9027 (2023).DOI
- [33]R. Yang, Y. Balogun, S. Ake, D. Baram, W. Brown, and G. Wang, Negative differential resistance in conical nanopore iontronic memristors, Journal of the American Chemical Society 146, 13183 (2024).DOI
- [34]S. Levine, J. Marriott, G. Neale, and N. Epstein, Theory of electrokinetic flow in fine cylindrical capillaries at high zeta-potentials, Journal of Colloid and Interface Science 52, 136 (1975).DOI
- [35]Y. Pershin and M. Di Ventra, Memory effects in complex materials and nanoscale systems, Advances in Physics 60, 145 (2011).DOI
- [36]E. Kirkinis, L. Bonds, A. Levchenko, and A. Andreev, Photogalvanic effect in hydrodynamic flows of nonreciprocal electron liquids, Physical Review B 112, 155413 (2025).DOI
- [37]P. Drazin and N. Riley, The Navier-Stokes equations: a classification of flows and exact solutions (Cambridge University Press, Cambridge, 2006).DOI



