Sie sind auf Seite 1von 25

Nonlinear electrochemical relaxation around conductors

Kevin T. Chu1,2 and Martin Z. Bazant1


1
Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139
2
Department of Mechanical and Aerospace Engineering, Princeton University, Princeton, NJ 08544
(Dated: March 5, 2006)
We analyze the simplest problem of electrochemical relaxation in more than one dimension –
the response of an uncharged, ideally polarizable metallic sphere (or cylinder) in symmetric, binary
electrolyte to a uniform electric field. In order to go beyond the circuit approximation for thin double
layers, our analysis is based on the Poisson-Nernst-Planck (PNP) equations of dilute solution theory.
Unlike most previous studies, however, we focus on the nonlinear regime, where the applied voltage
across the conductor is larger than the thermal voltage. In such strong electric fields, the classical
model predicts that the double layer adsorbs enough ions to produce bulk concentration gradients
and surface conduction. Our analysis begins with a general derivation of surface conservation laws
in the thin double-layer limit, which provide effective boundary conditions on the quasi-neutral
bulk. We solve the resulting nonlinear partial differential equations numerically for strong fields
and also perform a time-dependent asymptotic analysis for weaker fields, where bulk diffusion and
surface conduction arise as first-order corrections. We also derive various dimensionless parameters
comparing surface to bulk transport processes, which generalize the Bikerman-Dukhin number. Our
results have basic relevance for double-layer charging dynamics and nonlinear electrokinetics in the
ubiquitous PNP approximation.

PACS numbers: 82.45.Gj, 82.45.Jn, 66.10.-x

E
I. INTRODUCTION (a) (b)

Diffuse-charge dynamics plays an important role in the


response of electrochemical and biological systems sub-
ject to time-dependent voltages or electric fields [1]. The
classical example is impedance spectroscopy in electro-
chemistry [2–5], but electrochemical relaxation is also be-
ing increasingly exploited in colloids and microfluidics [6].
For example, alternating electric fields have been used to (c) (d)
pump or mix liquid electrolytes [7–18], to separate or self-
assemble colloids near electrodes [19–25], and to manip-
ulate polarizable particles [16, 26–30] or biological cells
and vesicles [31–33].
In this paper, we analyze some simple problems exem-
plifying the nonlinear response of an electrolyte around FIG. 1: (a) Schematic diagram of metallic colloidal
an ideally polarizable object due to diffusion and electro- sphere in a binary electrolyte subjected to an ap-
plied electric field, which has the same relaxation
migration. As shown in Figure 1, we consider the ionic
as a metallic hemisphere on a flat insulting surface,
relaxation around a metallic sphere (a) or cylinder (b) shown in (c). We also consider the analogous two-
subject to a suddenly applied uniform background elec- dimensional problems of a metallic cylinder (b), or
tric field, as in metallic colloids. Equivalently, we con- a half cylinder on an insulating plane (d).
sider a metallic hemi-sphere (c) or half-cylinder (d) on an
insulating plane, to understand relaxation around metal-
lic structures on channel walls in micro-electrochemical
devices. Although we do not consider fluid flow, our for over a century [1], the standard model of electrochem-
analysis of nonlinear electrochemical relaxation is an nec- ical relaxation has been an equivalent circuit, where the
essary first step toward understanding associated prob- neutral bulk is represented by an Ohmic resistor and the
lems of induced-charge electro-osmosis in the same ge- double layer by a surface impedance [2–5], which reduces
ometries [16–18, 29, 30], and thus it also has relevance to a linear capacitor at an ideally polarizable surface. For
for the case of AC electro-osmosis at planar electrode our model problems, this “RC circuit” model was first
arrays [7–14]. applied to electrochemical relaxation around a sphere
In electrochemistry [34–37] and colloid science [38, 39], by Simonov and Shilov [40] and around a cylinder by
it is common to assume that the charged double-layer at Bazant and Squires [16, 17]. Similar RC-circuit analysis
a metal surface is very thin and thus remains in quasi- has been applied extensively to planar electrode arrays in
equilibrium, even during charging dynamics. As a result, microfluidic devices, following with Ramos et al. [7] and
2

Ajdari [8]. problems, they find that the relaxation of the cell to the
While convenient for mathematical analysis and often steady state requires bulk diffusion processes that appear
sufficiently accurate, circuit models neglect the possi- as a small correction at O(), where the small parameter
bility of bulk concentration gradients, which can arise  is the ratio of the Debye length to a typical scale of the
at large applied voltages [1] and/or when the surface geometry. (In contrast, in Refs. [41–44], it appears that
is highly charged [41–44], as well as nonuniform sur- primarily the the strength of the applied electric field is
face transport of ions through the double layer [45, 46]. assumed to be small, although the mathematical limit
Dukhin and Shilov [41, 44] and later Hinch, Sherwood, is not explicitly defined.) For applied voltages in the
Chew, and Sen [42, 43] made significant progress beyond strongly nonlinear regime, they show that bulk concen-
the simple circuit model by including bulk diffusion in tration gradients can no longer be considered small, since
their studies of double layer polarization around highly O(1) concentration variations appear. In both regimes,
charged spherical particles (of fixed charged density) in the absorption of neutral salt by the double layer (and
weak applied fields. One of the main results of their anal- therefore build up of surface ion density) is the key driv-
ysis is that for weak applied electric fields, bulk concen- ing force for bulk diffusion. This positive salt adsorption
tration gradients appear as a small correction (on the or- in response to an applied voltage, first noted in Ref. ??,
der of the applied electric field) to a uniform background is opposite to the classical “Donnan effect” of salt expul-
concentration. In this work, we will lift the weak-field sion [39], which occurs if the surface chemically injects
restriction and consider the nonlinear response of the or removes ions during the initial creation of the double
system to strong applied fields, using the same mathe- layer [49].
matical model as in all prior work – the Poisson-Nernst- Bazant, et al. also emphasize the importance of both
Planck (PNP) equations of dilute solution theory [37]. As the charging and the diffusion time scales in the evolution
we shall see, nonlinear response generally involves non- of the electrochemical systems [1]. Because circuit mod-
negligible bulk diffusion. els inherently neglect diffusion processes, the only char-
There are also difficulties with the traditional macro- acteristic dynamic time scale that appears is the so-called
scopic picture of the double layer as an infinitely thin sur- RC charging time, τc = λD L/D, where λD is the Debye
face impedance, or possibly some more general nonlinear length, L is the system size, and D is the characteristic
circuit element. At the microscopic level, the double layer diffusivity of the ions [68]. However, when concentration
has a more complicated structure with at least two dif- gradients are present, diffusion processes and dynamics
ferent regimes: a diffuse layer where the ions move freely at the diffusion time scale, τL = L2 /D may be important.
in solution and a compact surface layer, where ions may Most theoretical analyses of electrochemical systems only
be condensed in a Stern mono-layer with its own physical consider the dynamics at one of these two dominant time
features (such as surface capacitance, surface diffusivity, scales – effectively decoupling the dynamics at the two
and surface roughness) [34–36]. The surface capacitance time scales. This decoupling of the dynamics is natural
may also include the effect of a dielectric coating, where when one considers the wide separation in the time scales
ions do not penetrate [8]. Mathematically, in one dimen- that govern the evolution of the system: τL  τc . An
sion the capacitor model can be derived as an effective interesting discussion of how the two time scales are cou-
boundary condition for the neutral bulk (Nernst-Planck) pled using ideas related to time-dependent asymptotic
equations by asymptotic analysis of the PNP equations matching is presented in Ref.[1].
in the thin double-layer limit [1]. Extending this analysis The paper is organized as follows. We begin in sec-
to higher dimensions, however, requires allowing for tan- tion II by carefully considering the thin double-layer limit
gential “surface conduction” through the double layer on of our model problems, which leads to effective boundary
the conductor for large applied electric fields. Here, we conditions for the neutral-bulk equations in section ??.
derive effective boundary conditions for the neutral bulk The most interesting new boundary conditions are sur-
in the PNP model by following a general mathematical face conservation laws, whose physical content we discuss
method for surface conservation laws at microscopically in detail in section IV, where we also define dimensionless
diffuse interfaces [47]. parameters governing the importance of various surface
The nonlinear response of electrochemical systems to transport processes. In section V, we explore the steady
strong applied fields seems to be relatively unexplored. response to large applied electric fields in our model prob-
To our knowledge, the only prior mathematical study lems; a notable prediction is the formation of recirculat-
of nonlinear relaxation with the PNP model is the re- ing bulk diffusion currents coupled to surface transport
cent work of Bazant, Thornton, and Ajdari on the one- processes. We then turn to relaxation dynamics in the
dimensional problem of parallel-plate blocking electrodes three regimes identified by Bazant et al. [1], using simi-
applying a sudden pulsed voltage [1]. (The same analy- lar methods, albeit with nontrivial modifications for two
sis has been recently extended to “modified-PNP” mod- or three dimensions. We begin with the linear response
els accounting for steric effects in concentrated solu- to a weak field in section VI, where we obtain exact solu-
tions [48], which would also be an important extension of tions using transform methods for arbitrary double-layer
our work.) For applied voltages in the weakly nonlinear thickness and also consider AC electric fields. Next, in
regime, which is analogous to weak applied fields in our section VII we use boundary-layer methods in space and
3

time to analyze “weakly nonlinear” relaxation in some- dynamics, we also assume an ideally polarizable surface
what larger fields, in the asymptotic limit of thin dou- with no-flux boundary conditions:
ble layers. Finally, in section VIII we comment on the
challenges of “strongly nonlinear” relaxation, where bulk ∂C+ zeD ∂Φ
D + C+ = 0 (6)
diffusion and surface conduction dominate the dynamics. ∂n kT ∂n
We conclude in section IX by discusing limitations of the ∂C− zeD ∂Φ
D − C− = 0 (7)
PNP model and directions for future research. ∂n kT ∂n
where the direction of the unit normal is taken to point
inwards towards the center of the sphere (i.e., outwards
II. MATHEMATICAL MODEL
from the region occupied by the electrolyte solution).
In the far field, we assume that both the concentration
A. PNP Initial-Boundary-Value Problem and potential profiles tend toward their initial conditions,
given everwhere by
As a model problem, we consider the response of an iso-
lated, ideally polarizable sphere (or cylinder) subjected C± (R, θ, φ, t = 0) = Co (8)
to a uniform, applied electric field, as shown in Figure 1. a3
 
For simplicity, we focus only on a symmetric, binary elec- Φ(R, θ, φ, t = 0) = −Eo R 1 − 3 cos θ. (9)
R
trolyte where both ionic species have the same diffusivity
and charge number. In order to study nonlinear effects where a is the radius of the sphere, Co is the bulk con-
and avoid imposing a time scale, we assume that the uni- centration far away from the conductor, and Eo is the
form electric field is suddenly applied at t = 0. applied electric field.
As in most (if not all) prior work on electrochemical dy-
namics, we assume the Poisson-Nernst-Planck equations
of dilute solution theory [37], B. Different Contributions to Ion Transport
 
∂C+ zeD The transport equations (1) and (2) represent conser-
= ∇ · D∇C+ + C+ ∇Φ (1)
∂t kT vation laws for the ionic species where the flux of each
species is a combination of diffusion and electromigration.
 
∂C− zeD
= ∇ · D∇C− − C− ∇Φ (2) Because this paper examines ionic fluxes in detail, let us
∂t kT
take a moment to fix the notation that we shall use to
−εs ∇2 Φ = ze (C+ − C− ) (3) discuss different contributions to charge and mass trans-
port. All fluxes will be denoted with the by the variable F
where D is the diffusivity, z is the charge number of the (or F for scalar components of flux). Superscripts will be
positively charged species, e is the charge of a proton, k used to distinguish between the diffusion, (d), and elec-
is Boltzmann’s constant, T is the absolute temperature, tromigration, (e), contributions to the flux. Subscripts
and εs is the electric permittivity of the solution. As will be used to denote the species or quantity with which
usual, we have used the Nernst-Einstein relation to write the flux is associated. Finally, normal and tangential
the mobility in terms of the the diffusivity, b = D/kT . components of a flux will denoted through the use of an
It is also useful to define the chemical potentials of the extra subscript: n for the normal component and t for the
ions, (e)
tangential component. As examples, F+ = − zeD kT C+ ∇Φ
µ± = kT log C± ± zeΦ (4) represents the flux of the positively charged species due
(d)
to electromigration and Fn,c = −D ∂C ∂n represents the
from which their fluxes are defined as F± = −bC± ∇µ± . flux of the neutral salt concentration, C = (C+ + C− )/2,
At the conductor’s surface, we assume the same bound- normal to a surface arising from diffusion. Table I pro-
ary conditions as in Ref. [1]. We adopt a linear surface- vides a summary of the various bulk fluxes that shall arise
capacitance condition on the electrostatic potential [50], in our discussion. (We also abuse notation and use the
same symbol µ for chemical potential with dimensions
∂Φ and scaled to kT .)
Φ + λS =V (5)
∂n

where λ−1
S is a length characterizing the compact-layer
C. Dimensionless Formulation
surface capacitance (e.g. due to a Stern monolayer or
a thin dielectric coating), where V is the potential of To facilitate the analysis of the model problem, it is
the conductor, which is set either externally or by the convenient to nondimensionalize the governing equations
condition of fixed total charge [16? , 17]. (We will focus and boundary conditions. Scaling length by the radius of
on the case of zero total charge, where symmetry implies the sphere, a, the time to the diffusion time τD = a2 /D,
V = 0 in our simple geometries.) To focus on charging and the electric potential by the thermal voltage divided
4

Because the charge density and neutral salt concen-


TABLE I: Summary of notations for for the various ion fluxes
tration are important for understanding the behavior of
and chemical potentials, before and after scaling.
electrochemical transport at high applied fields, it is of-
a
Dimensional Formula Dimensionless Formula ten useful to formulate the governing equations in terms
of the average concentration, c = (c+ + c− ) /2, and half
µ± kT log C± ± zeΦ log c± ± φ the charge density, ρ = (c+ − c− ) /2 [1, 50, 51]. Using
these definitions (10) – (12) can be rewritten as
F± −bC± ∇µ± −c± ∇µ±
∂c
− D∇C± ± zeD = ∇ · (∇c + ρ∇φ) (18)
` ´
C± ∇Φ − (∇c± ± c± ∇φ)
(d)
kT ∂t
F± −D∇C± −∇c± ∂ρ
(e) = ∇ · (∇ρ + c∇φ) (19)
F± ∓ zeD
kT
C± ∇Φ ∓c± ∇φ ∂t
−2 ∇2 φ = ρ (20)
zeD
` ´
Fc − D∇C + ρ∇Φ − (∇c + ρ∇φ)
(d)
kT Throughout our discussion, we shall alternate between
Fc −D∇C −∇c this formulation and equations (10) – (12) depending on
Fc
(e)
− zeD
kT
ρ∇Φ −ρ∇φ the context. The initial and boundary conditions for this
set of equations are easily derived from (13) – (17). Here

`
− D∇ρ + zeD
C∇Φ
´
− (∇ρ + c∇φ) we summarize those boundary conditions that change:
kT
(d)
Fρ −D∇ρ −∇ρ ∂c ∂φ
+ρ = 0 (21)
(e)
Fρ − zeD C∇Φ −c∇φ ∂n ∂n
kT
∂ρ ∂φ
+c = 0 (22)
a In these formulae, ρ is half the charge density (not the total ∂n ∂n
charge density). Also, we have abused notation and used the same c(r, θ, φ, t = 0) = 1 (23)
variable ρ for both the dimensional and dimensionless formulae. ρ
in the dimensional formulae is equal to Co ρ in the dimensionless
ρ(r, θ, φ, t = 0) = 0 (24)
formulae. For the remainder of this paper, we shall work primar-
ily with dimensionless equations. On occasion, we will
by the cation charge number, kT /ze, the governing equa- mention the dimensional form of various expressions and
tions (1) – (3) become equations to help make their physical interpretation more
apparent.
∂c+
= ∇ · (∇c+ + c+ ∇φ) (10)
∂t
∂c− D. Electroneutral Bulk Equations
= ∇ · (∇c− − c− ∇φ) (11)
∂t
2 2
− ∇ φ = (c+ − c− ) /2 (12) In the context of electrokinetics, it is desirable to
reduce the complexity of the electrochemical transport
where c± , φ, and t are the dimensionless concentra- problem by replacing the PNP equations with a sim-
tions, electric potential and time, respectively, the spatial pler set of equations that treats the bulk electrolyte and
derivatives are with respect to the dimensionless posi- the double layer as separate entities. Circuit models
tion, and
q  is the ratio of the Debye screening length, [1, 12, 16, 17] have been used extensively to achieve
λD = 2zε2sekT this goal by reducing the transport problem to an elec-
2 C , to the radius of the sphere. The bound-
o
trostatics problem. However, circuit models make the
ary conditions at the surface of the sphere and the initial rather stringent assumption that bulk concentrations re-
conditions become main uniform at all times. Unfortunately, at high applied
∂φ electric fields, this assumption is no longer valid because
φ + δ = v (13) concentration gradients become important [1].
∂n
∂c+ ∂φ In the present analysis, we consider an alternative sim-
+ c+ = 0 (14) plification of the PNP equations that allows for bulk con-
∂n ∂n
∂c− ∂φ centration variations. Since we are interested in colloidal
− c− = 0 (15) systems where particle diameters are on order of microns,
∂n ∂n
 is very small which suggests that we consider the thin
c± (r, θ, φ, t = 0) = 1 (16)
  double layer limit ( → 0). In this limit, the bulk re-
1 mains locally electroneutral, so it is acceptable to replace
φ(r, θ, φ, t = 0) = −Eo r 1 − 3 cos θ. (17)
r Poisson’s equation with the local electroneutrality condi-
tion [37, 52]:
In the far field, the dimensionless concentrations
approach 1 and the electric potential approaches
X
zi ci = 0. (25)
−Eo r cos θ. i
5

For the case of symmetric, binary electrolytes, (25) leads Fn,i


to the electroneutral Nernst-Planck equations: Jt,i

∂c Γi
= ∇2 c (26)
∂t a
0 = ∇ · (c∇φ) (27)
ρ = 0. (28)

Notice that the common practice of modeling the bulk


electrolyte as a linear resistor obeying Ohm’s law, ∇2 φ =
0, arises from these equations if the concentration profile
is uniform.
We emphasize that these equations only describe the
electrochemical system at the macroscopic level (i.e., in FIG. 2: Schematic diagram of normal and tangential
the bulk region of the solution). The microscopic struc- fluxes involved in the surface conservation law (31).
ture within the double layer, where local electroneutral- The shaded circle represents the metal particle; the
dotted circle represents the outer “edge” of the double
ity breaks down, is completely neglected. Therefore, any
layer.
physical effects of double layer structure can only be
incorporated into the mathematical model via effective
boundary conditions.
edge of the diffuse charge layer) [1, 50]:

III. EFFECTIVE BOUNDARY CONDITIONS ζ + 2δ c sinh (ζ/2) = v − φ. (29)
OUTSIDE THE DOUBLE LAYER
Here ζ is the potential drop across the diffuse part of the
double layer (i.e., zeta-potential), v is the potential of
When local electroneutrality is used in place of Pois- the metal sphere in the thermal voltage scale, and φ and
son’s equation, the physical boundary conditions im- c are the values of the electric potential and average con-
posed at at electrode surfaces must be modified to ac- centration at the outer edge of the diffuse charge layer.
count for the microscopic structure within the double With dimensions, (29) is given by
layer. Fortunately, in the  → 0 limit, the double layer re-
mains in quasi-equilibrium, so the Gouy-Chapman-Stern
r  
2kT C z+ eζ̄
model [36] can be used to derive effective boundary condi- ζ̃ + 2λs sinh =V −Φ (30)
εs 2kT
tions for the system. We emphasize that the GCS model
is not an assumption; rather, it emerges as the leading- where ζ̃ is the dimensional zeta-potential. Note that the
order approximation in an asymptotic analysis of the thin GCS model for the double layer leads to the conclusion
double-layer limit. (See Ref. [1] for the history of this that the only dependence of the normal derivative of
well known result.) Because the general form for the ef- the electric potential in on the double layer structure is
fective boundary conditions of locally electroneutral elec- through the zeta-potential. A more detailed derivation of
trochemical systems has not been extensively discussed this form of the Stern boundary condition can be found
in the literature, we provide a detailed derivation of these in [50].
boundary conditions and discuss some associated dimen-
sionless parameters.
B. Diffuse-layer Surface Conservation Laws

A. Compact-layer Surface Capacitance The effective boundary conditions for ionic fluxes are
more complicated. Because the physical domain for the
We begin our discussion of effective boundary con- macroscopic equations (26) – (28) excludes the diffuse
ditions by considering the “Stern boundary condition” part of the double layer, the no-flux boundary conditions
(13), which describes the (linear) capacitance of a pos- do not apply – it is possible for there to exist ion flux
sible compact layer on the surface. Because Eq. (13) between the bulk region and the double layer. More-
involves the electric potential and electric field at the in- over, there is also the possibility of ion transport within
ner edge of the diffuse charge layer, it cannot be directly the double layer itself (often neglected) which must be
used as a boundary condition for the locally electroneu- accounted for.
tral equations (26) – (28). However, by rearranging (13) The derivation of the effective flux boundary condi-
and using the GCS model, it is is possible to rewrite the tions (38) is based on a general theory of surface conser-
Stern boundary condition so that it only explicitly in- vation laws at microscopically diffuse interfaces, which
volves the electric potential and average concentration at we develop in Ref. [47]. The basic physical idea is to in-
the macroscopic surface of the electrode (i.e., the outer tegrate out the spatial variation within the double layer
6

in the direction normal to the electrode-electrolyte inter- number should appear for alternative choices of the elec-
face. While intuitively obvious, carrying out the integra- tric potential scale; in the following discussion, the zi
tion involves careful use of asymptotic analysis to address are essentially the sign of the “dimensional” ionic charge
the mathematical subtleties of integration over boundary numbers.
layers. The theory tells us that effective flux boundary There are a few important features of (38) worth men-
conditions have the physically apparent form tioning. First, the surface transport term (first term on
the right hand side) does not always contribute to the
∂Γi leading order effective flux boundary condition. Whether
= −∇s · Jt,i + Fn,i (31)
∂t the surface conduction term must be retained at leading
where (∇s ·) denotes surface divergence, Jt,i is the tan- order depends on the magnitudes of Γi (which in turn de-
gential flux within and Fn,i is the normal flux into the pends on the zeta-potential) and the tangential compo-
boundary layer (see Figure 2). The effective fluxes Jt,i nent of the bulk electric field. Interestingly, when the the
and Fn,i are directly related to the flux Fi for the trans- surface transport term is significant, the flux boundary
port process via condition depends explicitly on the small parameter .
Z ∞ Also, (38) allow for two important physical effects that
 only arise for 2- and 3-D systems: (i) non-uniform dou-
Jt,i =  F̃t,i − F̂t,i dz (32) ble layer charging and (ii) surface transport within the
0
Fn,i = Fi · n̂ (33) double layer itself. The presence of these effects lead to
richer behavior for 2- and 3-D systems compared to the
where F̃ and F̂ denote the flux within the boundary layer 1D system studied in [1].
and the flux in the bulk just outside of the boundary layer To put (38) into a more useful form, we use the GCS
and the integration is over the entire boundary layer (i.e., model of the double layer to express the surface flux den-
z is the inner variable of a boundary layer analysis). For sities in terms of the zeta-potential and the bulk concen-
electrochemical transport, the flux is given by the Nernst- tration. From the GCS model [36, 37], we know that the
Planck equation excess concentration of each ionic species is given by
 
Fi = − (∇ci + zi ci ∇φ) . (34) γi = c̃i − ĉi = ĉ e−zi ψ̃ − 1 (39)
Substituting this expression into (32) – (33), rearranging
a bit, and using the definition of a surface excess concen- and that
tration [38, 39, 47] √
!
∂ ψ̃ z+ ψ̃
Z ∞ Z ∞ = −2 ĉ sinh . (40)
∂z 2
Γi ≡  γi dz =  (c̃i − ĉi ) dz, (35)
0 0
Using these expressions, it is straightforward to show that
we obtain the surface excess concentration is
 Z ∞ 

Jt,i = − ∇s Γi + zi Γi ∇s φ̂ + zi c̃i ∇s ψ̃dz (36) 2 ĉ  −zi ζ/2 
0 Γi = e −1 . (41)
|zi |
∂ci ∂φ
Fn,i = − − z i ci . (37)
∂n ∂n Therefore, the first two surface flux density terms in (38)
can be written as
where ψ̃ ≡ φ̃− φ̂ is the excess electric potential within the
boundary layer and ∇s denotes a surface gradient. As be- Γi
fore, the tilde (˜) and hat (ˆ) accents denote the quantities ∇s Γi + zi Γi ∇s φ̂ = ∇s log ĉ + zi Γi ∇s φ̂
2 √
within boundary layer and the in the bulk immediately
−  sgn(zi ) ĉ e−zi ζ/2 ∇s ζ. (42)
outside of the boundary layer. Finally, the effective flux
boundary conditions for electrochemical transport follow
To evaluate the last term in the surface flux density, we
by using these results in (31):
observe that
 Z ∞ 
∂Γi
= ∇s · ∇s Γi + zi Γi ∇s φ + zi c̃i ∇s ψ̃ dz
 
∂ ψ̃ ∇s ζ z
∂t 0 ∇s ψ̃ = − √ + ∇s log ĉ , (43)
  ∂z 2 ĉ sinh (ζ/2) 2
∂ci ∂φ
− + z i ci (38)
∂n ∂n which follows directly by comparing the normal and sur-
face derivatives of the leading order expression for the
Note that even though our choice of electric potential
electric potential within the double layer [1, 50, 51]
scale eliminates the need to explicitly refer to the ionic
charge numbers, zi , we opt to continue using them in  √ 
the present discussion so that it is clear where the charge ψ̃(z) = 4 tanh−1 tanh(ζ/4)e− ĉz . (44)
7

Using this result, the integral in (38) greatly simplifies GCS model for the double layer; equations (36) – (38)
and yields are valid even for more general models of the boundary
layer.
Z ∞ √
zi c̃i ∇s ψ̃ dz =  sgn(zi ) ĉ e−zi ζ/2 ∇s ζ Before moving on, we write the effective boundary con-
0 ditions in dimensional form to emphasize their physical
Γi interpretation:
+ ∇s log ĉ. (45)
2    
∂ Γ̄i C zi eD
Finally, combining (42) and (45), the effective flux = ∇s · DΓ̄i ∇s log + Γ̄i ∇s Φ
boundary condition (38) becomes ∂t Co kT
 
∂Ci zi eD ∂Φ
∂Γi h i − Di + Ci , (49)
= ∇s · Γi ∇s log ĉ + zi Γi ∇s φ̂ ∂n kT ∂n
∂t  
∂ci ∂φ where Γ̄i is the dimensional surface excess concentration
− + z i ci (46)
∂n ∂n of species i and is defined as
where (dropping hats) c and φ understood to be in the Z  
bulk, just outside the double layer. Notice that the tan- Γ̄i ≡ C̃i − Ci dZ (50)
dl
gential gradients in the zeta-potential have vanished in
this equation so that the surface flux density of the indi- where the integration is only over the double layer. With
vidual species is independent of ∇s ζ. the units replaced, it becomes clear that the effective
In terms of the (dimensionless) chemical potentials of boundary conditions is a two-dimensional conservation
the ions, law for the excess surface concentration Γi with a driv-
ing force for the flux and a source term that depend only
µi = log ci + zi φ̂, (47)
on the dynamics away from the surface. Thus, the effec-
the surface conservation law (46) reduces to a very simple tive boundary conditions natually generalize the simple
form, capacitor picture of the double layer to allow for flow of
ionic species tangentially along the electrode surface.
∂Γi
= ∇s · (Γi ∇µi ) − n̂ · ci ∇µi (48)
∂t
where it is clear that the tangential gradient in the bulk C. Surface Charge and Excess Salt Concentration
chemical potential just outside the double layer drives the
transport of the surface excess concentration of each ion, Since the governing equations (26) – (28) are formu-
as well as the bulk transport. This form of the surface lated in terms of the salt concentration and charge den-
conservation law should hold more generally, even when sity, it is convenient to derive boundary conditions that
the chemical potential does not come from dilute solution are directly related to these quantities. Toward this end,
theory (PNP equations). we define q and w to be the surface charge density and
Effective boundary conditions similar to (46) describ- surface excess salt concentration, respectively:
ing the dynamics of the double layer have been known
for some time. In the late 1960s, Dukhin, Deryagin, and Z ∞
1
Z ∞
Shilov essentially used (46) in their studies of surface q = ρ̃dz = (γ+ − γ− ) dz (51)
2
conductance and the polarization of the diffuse charge Z0 ∞ 0
1 ∞
Z
layer around spherical particles with thin double layers w = (c̃ − ĉ) dz = (γ+ + γ− ) dz. (52)
at weak applied fields [41, 44, 46]. Later, Hinch, Sher- 0 2 0
wood, Chew, and Sen used similar boundary conditions
in their extension of the work of Dukhin et al. to explic- By integrating the expressions for the diffuse layer salt
itly calculate the tangential flux terms for a range of large concentration and charge density [1, 36–38],
surface potentials and asymmetric electrolytes [42, 43].
A key feature of both of these studies is the focus on c̃ = ĉ cosh ψ̃ (53)
small deviations from bulk equilibrium. As a result, the ρ̃ = −ĉ sinh ψ̃ (54)
effective boundary conditions used are basically applica-
tions of (46) for weak perturbations to the background and using (40), both q and w can be expressed as simple
concentration and electric potential. Our work differs functions of the zeta-potential and the bulk concentration
from these previous analyses because we do not require just outside of the double layer:
that the bulk concentration only weakly deviates from a
uniform profile, and we more rigorously justify the ap- √
proxiation through the use of matched asymptotics. In q = −2 ĉ sinh(ζ/2) (55)

addition, our analysis is not restricted to the use of the w = 4 ĉ sinh2 (ζ/4). (56)
8

Thus, we can combine the effective flux boundary condi- charged particles (with large, but nearly uniform sur-
tions for individual ions (38) to obtain face charge) [41, 44, 55]. As in other situations, surface-
 Z ∞  conduction effects in electrophoresis are controlled by Bi,
∂q which thus came to be known in the Russian literature
 = ∇s · ∇s q + w∇s φ̂ + c̃∇s ψ̃ dz
∂t 0 as the “Dukhin number”, Du. (Dukhin himself denoted
∂φ it by Rel).
−c (57) Recently, Bazant, Thornton and Ajdari pointed out
∂n
∂w
Z ∞  that the (steady) Bikerman-Dukhin number is equal to
 = ∇s · ∇s w + q∇s φ̂ + ρ∇s ψ̃ dz the ratio of the excess surface salt concentration to its
∂t 0 bulk counterpart at the geometrical scale a,
∂c
− . (58) Γ+ + Γi
∂n Bi = (62)
2aC0
Notice that, as in the PNP equations written in terms of
c and ρ, there is a symmetry between q and w in these and they showed its importance in a one-dimensional
equations. As in the previous section, we can use the problem of electrochemical relaxation between parallal-
GCS model to rewrite (57) and (58) solely in terms of plate electrodes [1]. This surprising equilvalence (in light
bulk field variables and the zeta-potential: of the definition of Bi) demonstrates that surface con-
duction becomes important relative to bulk conduction
∂q   ∂φ simply because a signficant number of ions are adsorbed
 = ∇s · q∇s log ĉ + w∇s φ̂ − c (59)
∂t ∂n in the double layer compared to the nearby bulk solution.
∂w   ∂c
This means that salt adsorption leading to bulk diffusion
 = ∇s · w∇s log ĉ + q∇s φ̂ − , (60)
∂t ∂n should generally occur at the same time as surface con-
duction, if the double layer becomes highly charged dur-
which is the form of the effective flux boundary condi-
ing the response to an applied field or voltage, as in our
tions we shall use in our analysis below.
model problems below.
Unlike prior work, however, our multi-dimensional
IV. SURFACE TRANSPORT PROCESSES nonlinear analysis allows for nonuniform, time-dependent
charging of the double layer, so the Bikerman-Dukhin
number can only be defined locally and in principle could
Before proceeding to the analysis, we discuss the rel- vary wildly across the surface. Moreover, since we sepa-
ative importance of the various surface transport pro- rate the contributions to surface transport from diffusion
cesses, compared to their neutral bulk counterparts. For and electromigration, we can define some new dimension-
clarity, we summarize our results for tangential surface less numbers. From Eqs. ()-(60), we find that the follow-
fluxes in Table II, with and without dimensions. As with ing surface-to-bulk flux ratios are related to the excess
the associated bulk fluxes summarized in Table ??, we surface-to-bulk ratio of neutral salt concentration
introduce different notations for contributions by diffu-
sion and electromigration (superscripts (d) and (e), re- (e)
|Jt,q | |Jt,w |
(d)
spectively) to the tangential (subscript t) surface fluxes α = = (63)
aJ0 aF0
of cations, anions, charge, and neutral salt (subscripts r  
+, −, q, and w, respectively). This allows us to define λD C zeζ
= w = 4 sinh2 (64)
dimensionless parameters comparing the various contri- a C0 4kT
butions.
J. J. Bikerman pioneered the experimental and the- Note that when surface diffusion is neglected as in most
oretical study of double-layer surface transport [53, 54] prior work, then α = Bi, since surface currents arise
and first defined the dimensionless ratio of surface cur- from electromigration alone. The other surface-to-bulk
rent to bulk current, across a geometrical length scale, a, flux ratios are given by the surface-to-bulk ratio of the
for a uniformly charged double layer and a uniform bulk charge density,
solution [45]. Using our notation, the Bikerman number (d) (e)
is |Jt,q | |Jt,w |
β = = (65)
aJ0 aF0
Jt,q
Bi = (61)
r  
aJ0 λD C zeζ
= |q| = 2 sinh (66)
a C0 2kT
where J0 = zeF0 is a reference bulk current in terms of
the typical diffusive flux, F0 = DC0 /a. B. V. Deryagin For thin double layers (λD  a), we see that the surface-
and S. S. Dukhin later added contributions from electro- to-bulk flux ratios, α and β, are only significant when ζ
osmotic flow, using the GCS model of the double layer significantly exceeds the thermal voltage kT /e.
(as did Bikerman) [46], and Dukhin and collaborators To better understand how the charge and neutral-salt
then used this model to study electrophoresis of highly fluxes are carried, it is instructive to form the ratio of
9

TABLE II: Summary of surface flux formulae for electrochemical transport.


Dimensional Formulaa,b,c Dimensionless Formula

Jt,± −bΓ± ∇s µ± −Γ± ∇s µ±


zeD
ˆ ˜
− DΓ± ∇s log (C/Co ) ± kT
Γ± ∇ s Φ − (Γ± ∇s log c ± Γ± ∇s φ)
(d)
Jt,± −DΓ± ∇s log (C/Co ) −Γ± ∇s log c
(e)
Jt,± ∓ zeD
kT
Γ± ∇ s Φ ∓Γ± ∇s φ

zeD
ˆ ˜
Jt,q − DQ∇s log (C/Co ) + kT
W ∇s Φ − (q∇s log c + w∇s φ)
(d)
Jt,q −DQ∇s log (C/Co ) −q∇s log c
(e)
Jt,q − zeD
kT
W ∇s Φ −w∇s φ

zeD
ˆ ˜
Jt,w − DW ∇s log (C/Co ) + kT
Q∇s Φ − (w∇s log c + q∇s φ)
(d)
Jt,w −DW ∇s log (C/Co ) −w∇s log c
(e)
Jt,w − zeD
kT
Q∇s Φ −q∇s φ

a We have abused notation and used the same variable Γ for both
i
the dimensional and dimensionless formulae. Γi in the dimensional
formulae is equal to Co aΓi in the dimensionless formulae.
b The dimensional surface charge density, Q, is defined by Q ≡

Co λD q.
c The dimensional excess neutral salt concentration, W , is defined

by W ≡ Co λD w.

these numbers, Similarly, the flux boundary conditions (IV) – (60) be-
(d) (e)
come
α |Jt,w | |Jt,q |
= (e)
= (d) (67)
β |Jt,w | |Jt,q | ∂φ
0 = ∇s · (q∇s log c + w∇s φ) − c (71)
∂n
 
zeζ
= tanh (68) ∂c
4kT 0 = ∇s · (w∇s log c + q∇s φ) − . (72)
∂n
For weakly charged double layers, zeζ  4kT , where
α  β, the (small) surface flux of salt is dominated by
electromigration, while the surface flux of charge (sur- Notice that we have retained the surface transport terms
face current) is dominated by diffusion (if bulk concen- in these boundary conditions even though they appear at
tration gradients exist). For highly charged double layers O(). At large applied fields, it is no longer valid to order
zeζ > 4kT where α ∼ β, the contributions to each flux the terms in an asymptotic expansions by  alone. We
by diffusion and electromigration become comparable, as also need to consider factors of the form eζ or  sinh(ζ)
counterions are completely expelled (q ∼ w). since eζ may be as large as O(1/E) for large applied
fields. Since both q and w contain factors which grow
exponentially with the zeta-potential, we cannot discard
V. NONLINEAR STEADY RESPONSE FOR the surface conduction terms in (71) and (72). Finally,
THIN DOUBLE LAYERS the Stern boundary condition (29) remains unchanged
because it does not involve any time derivatives.
A. Effective Equations As mentioned earlier, the steady problem exhibits in-
teresting features that have not been extensively ex-
Using the mathematical model developed in the pre- plored. Unfortunately, the nonlinearities present in the
vious section, we now examine the steady response of governing equations and boundary conditions make it dif-
a metal sphere or cylinder with thin double layers sub- ficult to proceed analytically, so we use numerical meth-
jected to a large, uniform applied electric field. At ods to gain insight into the behavior of the system. In
steady-state, the unsteady term is eliminated from the this section, we first briefly describe the numerical model
governing equations (26) – (28), so we have we use to study the system. We then use the numerical
model to study the development of O(1) bulk concen-
0 = ∇2 c (69) tration variations and their impact on transport around
0 = ∇ · (c∇φ) . (70) metal colloid spheres.
10

B. Numerical Model
1.2
2
1
ψ 0
c 0.8

To solve (69) – (70) numerically in a computationally 0.6


0.4
−2

efficient manner, we use a pseudospectral method [56– 2 2

58]. For problems in electrochemical transport, pseu- 0


−2
2
4 0
−2
2
4

0 0
dospectral methods are particularly powerful because z x z x

they naturally resolve boundary layers by placing more


FIG. 3: Numerical solutions for the concentration c
grid points near boundaries of the physical domain [56– (left) and excess electric potential ψ (right) for E = 10,
58]. We further reduces the computational complexity  = 0.01, δ = 1. Notice the large gradients in the
and cost of the numerical model by taking advantage of concentration profile near the surface of the sphere.
the axisymmetry of the problem to reduce the numerical
model to two dimensions (as opposed to using a fully 3-D
description). 2 20
E=15
E=15 E=5 15

For the computational grid, we use a tensor product 1.5


E=10 E=1
10 E=10

5
grid of a uniformly spaced grid for the polar angle direc- cs φs
1 0
tion and a shifted semi-infinite rational Chebyshev grid −5 E=5
for the radial direction [57]. To obtain the discretized 0.5 −10 E=1

form of the differential operators on this grid, we use −15

Kronecker products of the differentiation matrices for 0


0 1.57 3.14
−20
0 1.57 3.14
θ θ
the individual one-dimensional grids [56]. The numeri-
cal model is then easily constructed using collocation by
FIG. 4: Bulk concentration ĉ and electric potential φ̂
replacing field variables and continuous operators in the
at the surface of the sphere for varying values of the
mathematical model by grid functions and discrete oper- applied electric field. In these figures,  = 0.01 and
ators. The resulting nonlinear, algebraic system of equa- δ = 1. Notice that for E = 15, the poles (θ = 0 and
tions for the values of the unknowns at the grid points θ = π) are about to be depleted of ions (i.e., ĉ ≈ 0).
is solved using a standard Newton iteration with contin-
uation in the strength of the applied electric field. The
Jacobian for the Newton iteration is computed exactly
by using a set of simple matrix -based differentiation rules C. Enhanced Surface Excess Concentration and
derived in the appendix of [52]. By using the exact Ja- Surface Conduction
cobian, the convergence rate of the Newton iteration is
kept low; typically less than five iterations are required Perhaps the most important aspects of the steady re-
for each value of the continuation parameter before the sponse at high applied electric fields are the enhanced
residual of the solution to the discrete system of equa- surface excess ion concentration and surface transport
tions is reduced to an absolute tolerance of 10−8 [69]. within the double layer. As shown in Figure 5, at
Directly computing the Jacobian in matrix form had the high applied fields, the excess surface concentrations is
additional advantage of making it easy to implement the O (1/E), so surface transport within the double layer
numerical model in MATLAB, a high-level programming
language with a large library of built-in functions. It is
also worth mentioning that we avoid the problem of deal- 0.1
0.1
ing with infinite values of the electric potential by formu- E=15 E=15

lating the numerical model in terms of φ + Er cos θ, the 0.05


0.08
E=10
E=10
deviation of the electric potential from that of the uni- 0.06
εw
form applied electric field, rather than φ itself. εq 0
E=5
0.04
E=5
−0.05 E=1
E=1 0.02
In our numerical investigations, we used the numerical
−0.1
method described above with 90 radial and 75 angular 0 1.57 3.14
0
0 1.57 3.14
θ θ
grid points. This grid resolution balanced the combined
goals of high accuracy and good computational perfor- FIG. 5: Surface charge density q (left) and excess sur-
mance. Figure 3 shows typical solutions for the concen- face concentration of neutral salt w (right) and for
tration and electric potential (relative to the background varying values of the applied electric field. In these
applied potential) for large applied electric fields. A com- figures,  = 0.01 and δ = 1. Notice that for large ap-
parison of the concentration and electric potential at the plied fields, w = O(1/E) and q = O(1/E) so that the
surface of the sphere for varying values of E are shown surface conduction terms in (IV) and (60) are O(1).
in Figure 4.
11

0 1.2 20
E=15
E=15
0.8
E=1 E=10
−0.2 E=10 15
0.4
E=5
F
−0.4
w
0 |E|||
F 10 E=5
q
E=10 −0.4
E=5
−0.6
−0.8 E=1 5
E=15 E=1

−0.8 −1.2
0 1.57 3.14 0 1.57 3.14
θ
0
θ 0 1.57 3.14
θ

FIG. 6: Tangential surface fluxes for the surface charge


density |Jt,q | (left) and excess surface concentration FIG. 8: Tangential component of bulk electric field,
of neutral salt |Jt,w | (right) for varying values of the |Et | at surface of sphere for varying values of the ap-
applied electric field. In these figures,  = 0.01 and plied electric field. In these figures,  = 0.01 and δ = 1.
δ = 1. Notice that for large applied fields, the surface
fluxes are O(1) quantities and make a non-negligible
leading-order contribution in (IV) and (60). 3 3
E=15 E=15
2 2
E=10 E=10
1 1
E=5 E=5
−c∂φ/∂n −∂c/∂n
1.25 1.25
0 0
surface
conduction
1 surface 1 −1 −1
conduction E=1 E=1

0.75 0.75 −2 −2

surface surface −3 −3
0.5 0.5 diffusion 0 1.57 3.14 0 1.57 3.14
diffusion θ θ

0.25 0.25
∂φ
FIG. 9: Normal flux of current c ∂n (left) and neutral
0 0 ∂c
0 1.57 3.14 0 1.57 3.14 salt ∂n (right) into the double layer for varying values
θ θ
of the applied electric field. In these figures,  = 0.01
and δ = 1.
FIG. 7: KTC Comparison of the magnitudes of |J(d) |
(solid lines) and |J(e) | (dashed lines) for the excess
surface fluxes of q (left) and w (right) for an ap-
plied electric field value of E = 15. In these figures, the bulk into the double layer (see Figure 9). Notice
 = 0.01 and δ = 1. Notice that in both cases, the sur-
that the normal flux of neutral salt into the double layer,
face diffusion is on the order of 1/ĉE times the surface
conduction. which is given by −∂c/∂n, shows an injection of neutral
salt at the poles (−∂c/∂n > 0), where the charging is
strongest, and an ejection of neutral salt at the equator
(−∂c/∂n < 0), where there is essentially no excess neu-
(shown in Figure 6) becomes non-negligible in the lead- tral salt build up. This configuration of fluxes leads to
ing order effective flux boundary conditions (71) – (72). the neutral salt depletion at the poles and accumulation
(e) (e)
Interestingly, surface conduction, Jt,q and Jt,w , are the at the equator shown in Figures 3 and 4. Similarly, the
dominant contributions to surface transport (see Fig- normal current density −ĉ∂φ/∂n, shows an influx of neg-
ure 7). While there are clearly surface gradients in con- ative (positive) current density at the north (south) pole
centration, surface diffusion is smaller than surface mi- and a positive (negative) current density closer to the
gration by a factor on the order of 1/ĉE. Also, we reit- equator. At the equator, there is no normal current den-
erate that the driving force for surface transport is solely sity because the normal diffusion currents of cations and
from surface gradients of the bulk concentration and bulk anions exactly balance and there is no normal electric
electric potential; gradients in the zeta-potential do not field to drive a migration current.
play a role because they are completely cancelled out.
An important feature of the surface fluxes, Jt,q and
Jt,w , shown in Figure 6 is that they are non-uniform. D. Bulk Diffusion and Concentration Gradients
This non-uniformity is strongly influenced by the non-
uniformity in the tangential electric field (see Figure 8). One major consequence of surface conduction is gen-
The non-uniformity of the surface excess salt concentra- eration of large fluxes of neutral salt into the double
tion and charge surface density also play a role but to layer. These cause strong concentration gradients to ap-
a lesser extent. For the steady problem, the surface ex- pear near the surface of the sphere (Figures 3 and 4),
cess concentration of ions remains constant in time, so indicating that the usual assumption of a uniform con-
the non-uniformity in the surface fluxes leads to non- centration profile is invalid at high electric fields. The
uniform normal fluxes of current and neutral salt from presence of these large concentration gradients at rela-
12

25 2.5
θ=π/2
1 E = 15 20 2
θ=π/4 θ = π/2

15 1.5
0.5
10 1
θ=0 θ = 0,π/4
z
0
5 0.5

−0.5 0 0
1 2 3 4 5 1 2 3 4 5
r r

−1
FIG. 11: KTC - which fluxes?? Comparison of the
0 0.5 1 1.5 2 2.5 magnitudes of bulk electromigration |F(e) | (solid lines)
x
and diffusion |F(d) | (dashed lines) fluxes as a function
FIG. 10: Diffusion currents drive transport of neu- of distance from the surface of the sphere at θ = 0, π/4,
tral salt near the surface of the sphere. In this figure, and π/2 (left). The figure on the right zooms in on
E = 10,  = 0.01 and δ = 1. Notice that streamlines the diffusion flux. In these figure, E = 10,  = 0.01
of neutral salt are closed; current lines start on the and δ = 1. Notice that magnitude of the electromi-
surface of the sphere near the equator and end closer gration dominates diffusion and that the electromigra-
to the poles. tion term itself becomes dominated by the contribu-
tion from the applied electric field a short distance
from the surface of the sphere.

tively low electric fields (E ≈ 5) should not be surpris-


ing since it is well-known that large voltage effects often a short distance from the sphere, the electromigration
begin with voltages as low as a few times the thermal term itself becomes dominated by the contribution from
voltage [1]. The dramatic influence of the voltage arises the applied electric field. Thus, the concentration gradi-
from the exponential dependence of double layer concen- ent only serves to slightly bias the flux densities so that
trations on the zeta-potential. cation (anion) motion is slightly retarded near the north
Since the transport of neutral salt is driven by concen- (south) pole.
tration gradients, the presence of these strong variations It is more interesting to consider the surface transport
leads to diffusion currents (see Figure 10). An important of the individual ions within the double layer. In the
feature of these diffusion currents is that they are closed; northern hemisphere, the double layer is dominated by
current lines start on the surface of the sphere near the anions; similarly, the southern hemisphere is dominated
equator where neutral salt is ejected into the bulk (as a by cations. As a result, transport in each hemisphere is
result of neutral salt transport within the double layer) primarily due to only one species (see Figure 12). This
and end close to the poles where neutral salt is absorbed
by the double layer. These recirculation currents are im- 1 1
portant because they allow the system to conserve the to- 0.75 0.75
E=15

tal number of cation and anions locally. Without them, 0.5 E=1 0.5 E=10

the local depletion and accumulation of ions would re- 0.25 0.25 E=5
Fc Fc
quire global changes to the bulk concentration (i.e., the +
0 −
0

−0.25 −0.25
concentration at infinity would be affected). E=5
E=1
−0.5 E=10 −0.5
While the presence of diffusion currents is interesting, −0.75 −0.75
E=15
we must be careful in how they are interpreted in terms −1 −1
0 1.57 3.14 0 1.57 3.14
of the motion of individual ion molecules. In actuality, no θ θ

ions are moving purely under the influence of diffusion.


Rather, the cation and anion migration flux densities are FIG. 12: Tangential surface fluxes for the cation |Jt,+ |
(left) and anion |Jt,− | (right) for varying values of the
slightly imbalanced due to the presence of a concentra-
applied electric field. In these figures,  = 0.01 and
tion gradient which results in a net flux of neutral salt δ = 1. Notice that for large applied fields, the surface
concentration. fluxes are O(1) quantities (in the appropriate hemi-
sphere) which leads to a non-negligible leading-order
contribution in (46).
E. Individual Ion Currents
observation provides a direct explanation for the deple-
Since cations and anions are the physical entities that tion and accumulation regions in the concentration pro-
are transported through the electrolyte, it is useful to file in terms of the the motion of individual ions. As
consider the cation and anion flux densities individually. mentioned earlier, the transport is from the poles to the
As shown in Figure 11, the contribution of electromigra- equator because surface conduction is driven by the tan-
tion to transport dominates diffusion. Moreover, within gential component of the bulk electric field (see Figure 8).
13

Since the double layer in the northern hemisphere is pre- When the potential drop across the particle is much
dominantly anions, the surface ion transport is from the smaller than the thermal voltage (Eo  1), it is possible
poles towards the equator. A similar argument in the to analyze the response of the system without assuming
southern hemisphere shows that surface transport of the that the double layers are thin; that is, we need not as-
majority ion is again from the poles towards the equator. sume that  is small and may describe the system using
Thus, influx of ions into the double layer occurs in the the full unsteady PNP equations, (18) – (22). Instead,
regions near the poles and outflux occurs by the equator. we assume that the response of the system is only a small
The dominance of a single species within the double deviation from the equilibrium solution:
layer for each hemisphere leads to a small conundrum:
how does the bulk electrolyte near the surface of the c ≡ 1 , ρ ≡ 0 , φ = Er cos θ. (73)
sphere remain locally electroneutral? In the northern In this limit, we can linearize the ionic concentra-
hemisphere, it would seem that more anion should be tions around a uniform concentration profile so that
absorbed at the pole and ejected at the equator leading c = 1 + δc. Using this expression in (19) and mak-
to bulk charge imbalance. Analogous reasoning involving ing use of Poisson’s equation (12) to eliminate the elec-
cation leads to the same conclusion in the southern hemi- tric potential, we find that the charge density, ρ =
sphere. The resolution to the conundrum comes from re- (c+ − c− ) /2 = (δc+ − δc− ) /2, obeys the (dimensionless)
membering that both diffusion and electromigration con- Debye-Falkenhagen equation [59]:
tribute to transport. The imbalance in the normal flux
required to sustain an imbalanced concentration profile ∂ρ 1
in the double layer is achieved by carefully balancing dif- ≈ ∇2 ρ − 2 ρ. (74)
∂t 
fusion (which drives both species in the same direction)
and electromigration (which drives the two species in op- Similarly, the flux boundary condition corresponding to
posite directions) so that the normal flux of the appropri- this equation reduces to
ate species dominates. For example, at the north pole, ∂ρ ∂φ
an electric field pointing away from the surface of the + = 0. (75)
∂n ∂n
sphere will suppress the cation flux towards the surface
while enhancing the anion flux. Note that (74) and Poisson’s equation (together with the
In this situation, the electric field plays a similar role no-flux and Stern boundary conditions) are a linear sys-
as the diffusion potential for electrochemical transport in tem of partial differential equations. Thus, we can take
an electroneutral solution. Recall that when the cation advantage of integral transform techniques.
and anion have different diffusivities, the electric field
acts to slow down the species with the higher diffusiv-
ity and speed up the species with the lower diffusivity in B. Transform Solutions for Arbitrary  and δ
such a way that both species have equal flux densities.
In the current situation, the electric field serves to cre- Since for weak applied fields, the model problem is a
ate the necessary imbalance in the cation and anion flux linear, initial value problem, it is natural to carry out the
densities so that the surface excess concentrations within analysis using Laplace transforms in time. Transforming
the double layer can be maintained while preserving local the Debye-Falkenhagen and Poisson equations, we obtain
electroneutrality in the bulk.
∇2 ρ̌ = β 2 ρ̌ (76)
− ∇2 φ̌ = ρ̌
2
(77)
VI. LINEAR RELAXATION FOR ARBITRARY
where
DOUBLE LAYER THICKNESS
1
β(s)2 = s + (78)
A. Debye-Falkenhagen Equation 2
and the check accents (ˇ) are used to denote Laplace
Although the focus of this paper is on nonlinear relax- transformed variables. Similarly, the boundary condi-
ation, leading to the steady state described in the pre- tions become
vious section, it is instructive to first consider linear re-
sponse to a weak applied field, where exact solutions are ∂ ρ̌ ∂ φ̌
+ = 0 (79)
possible. Moreover, the linear analysis also has relevance ∂n ∂n
for the early stages of nonlinear relaxation before sig- ∂ φ̌
φ̌ + δ = vs−1 (80)
nificant double layer charge has accumulated, such that ∂n
max{ζ(θ)}  kT /e, as long as the metal surface is un- −∇φ̌ → Eo s−1 as r → ∞. (81)
charged when the field is applied. The linear reponse also
allows a more general analysis, including AC periodic re- To solve the resulting boundary value problem, we take
sponse, in addition to our model problem with sudden advantage of the spherical geometry to write the solu-
DC forcing. tion in terms of spherical harmonics. Since ρ̌ satisfies
14

the modified Helmholtz equation, we can expand it in a By independently equating the coefficients of the differ-
series with terms that are products of spherical harmon- ent spherical harmonics in (86) and (87), we obtain (after
ics, Ylm (θ, φ), and modified spherical Bessel functions, a little algebra)
kl (βr). Moreover, we can reduce the series to a single
term A = vs−1 (88)
−1 2
ρ̌(r, θ, φ) = R k1 (βr) Y10 (θ, φ) = R k1 (βr) cos θ (82) −3Eo s (β)
R = h i (89)
2
2k1 (β) + 1 − (β) (1 + 2δ) βk10 (β)
by taking into account the symmetries of the charge den- !
sity: (i) axisymmetry, (ii) antisymmetry with respect to Eo s−1 R 1
θ = π/2, and (iii) the dipolar nature of the externally B = − − − 1 βk10 (β). (90)
2 2 (β)2
applied field. Note that we have only retained the term
involving the modified spherical Bessel functions that de-
cays as r → 0 because ρ̌ vanishes at infinity. Similarly, Finally, by writing k1 (x) in terms of elementary func-
the general solution for φ̌ may be expressed as tions [60],

cos θ e−x (x + 1)
φ̌(r, θ, φ) = −Eo s−1 r cos θ + A + B + C k1 (βr) cos θ k1 (x) = , (91)
r2 x2
(83)
where the first term accounts for the boundary condition we can express R as
on the electric field at infinity, the next two terms are
the general solution to Laplace’s equation possessing the −3Eo s−1
R= i. (92)
required symmetries, and the last term is the particular
h 
2
e−β 1 + β + β22 (1 + 2δ) − 1
(β)2
solution to Poisson’s equation. Note that we have left
out the monopolar term in the potential because it is
only necessary for charged spheres. Our analysis is not Following Bazant, Thornton, and Ajdari [1], we focus
made any less general by neglecting this term; the case on times that are long relative to the Debye time (t =
of a charged sphere in an weak applied field is handled O(2 ) in dimensionless units). In this limit, s  1/2 so
by treating it as the superposition of a charged sphere in that the charge density on the surface of the sphere is
the absence of an applied field with an uncharged sphere given by
subjected to an applied field.
−Kρ s−1
 
The coefficients in (82) and (83) are determined by en- ρ̌(r = 1, θ, s) ∼ cos θ (93)
forcing Poisson’s equation and the boundary conditions 1 + τρ s
(79) – (80). Plugging (82) and (83) into Poisson’s equa-
tion (77), we obtain with

R 3Eo (1 + )
C=− Kρ = (94)
2. (84) 2γ
(β)  
(1 + 2δ)(1 + ) − δ
To apply the boundary conditions, note that on the τρ =  (95)
∂ ∂
2γ(1 + )
sphere ∂n = − ∂r r=1
so that
γ = (1 + 2δ) (1 + ) + δ. (96)

∂ φ̌ −1 1 ∂ ρ̌
= Eo s cos θ + 2B cos θ + 2 (85) Inverting the Laplace transform, we see that at long
∂n (β) ∂r r=1 times, the charge at the surface of the sphere has an
where the last term was obtained by using the relation exponential relaxation with a characteristic time on the
between C and R. Thus, the no-flux boundary condition order of :
(79) becomes  
! ρ(r = 1, θ, t) ∼ −Kρ 1 − e−t/τρ cos θ. (97)

−1 1 ∂ ρ̌
0 = Eo s cos θ + 2B cos θ + 2 −1
(β) ∂r r=1 Note that τρ is on the order of the dimensionless RC time,
, which is much larger than 2 , the dimensionless Debye
Eo s−1
( )
 + 2B  time.
= 1 cos θ (86)
+ R (β) 2 − 1 βk10 (β) To obtain the linear response of cylinders, we follow the
same procedure as above. The main differences are that
Similarly, the Stern boundary condition (80) becomes the series solution is written in terms of a cosine serices
with the radial dependence given by modified Bessel func-
vs−1 = A + tions. Without going through the algebra, the results for
(δ − 1) Eo s−1 + (1 + 2δ) B
 
cylinders are given in Table III along side the analogous
R 0 cos θ (87)
+ (β) 2 [−k1 (β) + δβk1 (β)] results for spheres.
15

TABLE III: Table of formulae for the linear response of metallic cylinders and spheres to weak applied electric fields.
Cylindera Sphere

ρ̌ RK1 (βr) cos θ Rk1 (βr) cos θ


h i h i
RK1 (βr) Rk1 (βr)
φ̌ vs−1 − Eo s−1 r cos θ + B
r
− (β)2
cos θ vs−1 − Eo s−1 r cos θ + B
r2
− (β)2
cos θ

−2Eo s−1 (β)2 −3E«os


−1
R K1 (β)+[1−(β)2 (1+δ)]βK1
0 (β)
»„ –
e−β 2 2
1+ β + 2 (1+2δ)− 1
β (β)2
“ ” “ ”
B −Eo s−1 − R 1
(β)2
− 1 βK10 (β) − 21 Eo s−1 − 12 R 1
(β)2
− 1 βk10 (β)

2Eo K1 (1/) 3Eo (1+)


Kρ K1 (1/)−δK1 0 (1/) 2[(1+2δ)(1+)+δ]

8 9
< 2K1 (1/)+ 2+δ−δ K10 (1/) K10 (1/)+δK100 (1/) =
» –
K1 (1/)
n o
(1+2δ)(1+)−δ
τρ − 
: [
2 K1 (1/)−K1 0 (1/)
] ; 2(1+)[(1+2δ)(1+)+δ]

2Eo K0 (1/) 3Eo 


Kq K1 (1/)−δK10 (1/) (1+2δ)(1+)+δ

8„ 0 « » 0 – 9
< + K0 (1/) K (1/)+ 1+2δ−δ K0 (1/) K 0 (1/)+δK 00 (1/) =
K (1/) 1 K (1/) 1 1
n o
0 0 (1+2δ)(1+)
τq − 2[K1 (1/)−K1

: 0 (1/)
] ; 2[(1+2δ)(1+)+δ]

aθ measured from vertical axis.

C. Response to a Weak, Oscillatory Field observations, the surface charge density is given by

R k0 (β)
Due to the close relationship between Fourier and q̌(θ, s) = cos θ
β
Laplace transforms, the algebra involved in analyzing the
response of the sphere to a weak, oscillatory field is al- R e−β
= cos θ. (99)
most identical to the response to a suddenly applied field. β2
Thus, for sufficiently low frequencies (ω  1/2 ), we can
immediately write down the responseto a weak, oscilla- In the long time limit s  1/2, we find that the sur-
tory field of the form E = Eo Re eiωt : face charge density
 an exponential relaxation q(θ, t) =
−Kq 1 − e−t/τq cos θ with

eiωt 3Eo 
 
ρ(r = 1, θ, t) = −Kρ Re cos θ. (98) Kq = (100)
1 + iωτρ γ
 
(1 + 2δ)(1 + )
τq =  . (101)

As with ρ, we see that the characteristic relaxation time


D. Accumulated Surface Charge Density
for q is on the order of the dimensionless RC time.

Because of its importance in many physical processes,


the accumulated surface charge density, q̌, is an interest- E. Time Scales for Linear Response
ing quantity to consider. It is easily computed from the
(volume) charge density ρ by integrating it in the radial We have seen that at long times both ρ and q relax ex-
direction from r = 1 to r = ∞. While the identifica- ponentially with characteristic time scales on the order
tion of this integral with a surface charge density really of the RC time, . However, as Figure 13 shows, the re-
only makes sense in the thin-double layer limit, the ac- laxation times for the two quantities are not exactly the
cumulated surface charge density is still worth examin- same and have a nontrivial dependence on the diffuse
ing. Fortunately, the integral is straightforward because layer thickness, , and Stern capacitance, δ. Notice that
k1 (z) = −k00 (z) [60] and the radial dependence of the for infinitely thin double layers ( = 0) the relaxation
charge density is independent of the angle. Using these times for the surface charge density and the accumulated
16

0.6 0.6

Steady Diffuse
δ = 0.0
0.5 δ = 0.1 0.5
Bulk Diffusion

Layer
0.4 δ = 1.0 0.4
δ = 10

Electrode
τq / ε
τ /ε

0.3 0.3
ρ

δ = 0.0 t
0.2 0.2 δ = 0.1
0.1 0.1 δ = 1.0
δ = 10 Uniform Bulk Ο(ε)
0 0
0 1 2 3 4 5 0 1 2 3 4 5
ε ε

z
FIG. 13: Exponential relaxation time constants for the Ο(ε)
charge density ρ and the accumulated surface charge Charging RC−time
Diffusion Layer
density q at weak applied fields as a function of  and Diffuse
Layer Ο( ε )
δ. The left panel shows the relaxation time constant
for the charge density at the surface of the sphere,
τρ (r = 1). The right panel shows the relaxation time FIG. 14: Five asymptotically distinct regions of space-
constant for the accumulated surface charge density, time that govern the dynamic response of a metal col-
τq . loid sphere to an applied electric field. Note the nested
spatial boundary layers at the RC time (t = O()).

charge density are identical. This behavior is expected RC time λD a/D and (2) the bulk diffusion time a2 /D.
since for thin double layers, almost all of the charge den- We proceed by seeking the leading order term (and in
sity in the diffuse layer is located very close to the surface some cases, the first-order correction) to the governing
of the particle. In this limit, the relaxation time has a equations (18) – (20) with both the spatial and the time
strong dependence on the Stern capacitance. For thick coordinate scaled to focus on the space-time region of
double layers (  1), this dependence on the Stern ca- interest.
pacitance disappears and the relaxation time curves for For the analysis, it is important to realize that the
all δ values converge. space-time domain is divided into five asymptotically dis-
Physically, the difference in the relaxation times for the tinct regions (see Figure 14). At the RC time, there ex-
surface charge and the accumulated charge densities for ist three spatially significant regions: (i) an O() quasi-
nonzero  values is an indication of the complex spatio- √
steady double layer, (ii) an O( ) dynamically active
temporal structure of the double layer charging. For thin diffusion layer, and (iii) a quasi-equilibrium, uniform
double layers, the difference τρ and τq is relatively small bulk electrolyte layer with time-varying harmonic elec-
because the charge in the double layer is restricted to a tric potential. At this time scale,√the charging dynamics
thin region. For thick double layers, however, the dif- are completely driven by the O( ) diffusion layer. At
ference in relaxation times is accentuated because the the diffusion time, there are only two important spatial
charge in the double layer is spread out over a larger spa- regimes: (i) a quasi-steady double layer and (ii) a dy-
tial region, which does not necessarily charge uniformly. namic bulk that evolves through locally electroneutral,
In fact, that τρ (r = 1) is smaller than τq for larger values diffusion processes.
of  (Figure 13) suggests that regions closer to the surface
of the sphere charge faster than regions that are further
away. B. Dynamics at the RC Time

1. Uniform Bulk and Equilibrium Double Layers


VII. WEAKLY NONLINEAR RELAXATION
FOR THIN DOUBLE LAYERS To examine the dynamics at the RC time, we rewrite
(18) – (20) by rescaling time to the RC time using t̃ = t/:
A. Dynamical Regimes in Space-Time
∂c
= ∇ · (∇c + ρ∇φ) (102)
For weak applied fields in linear response, the compli- ∂ t̃
cated dependence of the Laplace transform solution for ∂ρ
= ∇ · (∇ρ + c∇φ) (103)
large s (short times) above hints at the presence of multi- ∂ t̃
ple time scales in the charging dynamics. In this section, −2 ∇2 φ = ρ. (104)
using boundary-layer methods, we derive asymptotic so-
lutions in the thin double-layer limit (  1) at some- Since the spatial coordinate is scaled to the bulk length,
what larger electric fields (defined below) for the con- the leading order solution of these equations describes the
centrations and electric potential by solving the leading dynamics of the bulk during the double layer charging
order equations at the two dominant time scales: (1) the phase. Substituting a regular asymptotic expansions of
17

the form have been lost since both the bulk and the boundary
layers are in quasi-equilibrium at leading order; neither
c(x, t) ∼ c0 + c1 + 2 c2 + . . . (105) layer drives its own dynamics. As discussed in [1], the
resolution to this paradox lies in the time-dependent flux
into equations (102)-(104) yields a hierarchy of partial
matching between the bulk and the boundary layer. Un-
differential equations. By sequentially solving the equa-
fortunately, it is inconsistent to directly match the bulk √
tions using the initial conditions, it is easy to show that
to the boundary layer; there must exist a nested O ( )
the “outer” solutions at the RC charging time scale are
diffusion layer in order to account for the build up of both
c0 (x, t) ≡ 1 (106) surface charge and surface excess neutral salt concentra-
tion.
cj (x, t) ≡ 0 , j = 1, 2, 3, . . . (107)
Mathematically, the presence of the diffusion layer at
ρj (x, t) ≡ 0 , j = 0, 1, 2, . . . (108) the RC time scale appears as a dominant balance in the
transport equations by rescaling the spatial √ coordinate
with φj is harmonic at all orders. In other words, the bulk
in the normal direction to the surface by  to obtain:
solution can be completely expressed (without a series ex-
pansion) as a uniform concentration profile, c (x, t) ≡ 1, ∂c̄
= 2 ∇s · ∇s c̄ + ρ̄∇s φ̄

and a time-varying harmonic electric potential, φ. Note ∂ t̃
that by taking advantage of spherical geometry and ax-
 
∂ ∂c̄ ∂ φ̄
isymmetry in our problem, we can write the potential as + + ρ̄ (111)
∂z 0 ∂z 0 ∂z 0
a series in spherical harmonics with zero zonal wavenum-
∂ ρ̄
= 2 ∇s · ∇s ρ̄ + c̄∇s φ̄

ber (i.e., Legendre polynomials in cos θ):
∂ t̃
∞  
X Al (t) ∂ ∂ ρ̄ ∂ φ̄
φ(r, θ, t) = −Eo r cos θ + Pl (cos θ) (109) + + c̄ 0 (112)
rl+1 ∂z 0 ∂z 0 ∂z
l=0
∂ 2 φ̄
where the radial dependence of each term has been se- −2 ∇2s φ̄ −  02 = ρ̄. (113)
∂z
lected so that φ automatically satisfies Laplace’s equation
Here the bar accent denotes√the “diffusion layer” solution
at all times.
at the RC time and z 0 = Z/  is the spatial coordinate in
At the RC charging time, the O() double layer is in
the direction normal to the surface. Notice that at this
quasi-equilibrium, which is easily verified by rescaling the
length scale, the system is not in quasi-equilibrium as
spatial coordinate normal to the particle surface by the
it is at the bulk and Debye length scales. It is, however,
dimensionless Debye length . As mentioned earlier, a
locally electroneutral at leading order as a result of (113).
quasi-equilibrium double layer possesses a structure de-
As the double layer charges, it absorbs
√ an O() amount
scribed by the Gouy-Chapman-Stern model. For conve-
of charge and neutral salt from the O( ) diffusion layer.
nience, we repeat the GCS solution here:
Therefore, we expect that concentration√ changes within
c̃± = ĉe∓ψ̃ ,c̃ = ĉ cosh ψ̃ , ρ̃ = −ĉ sinh ψ̃ the diffusion to be on the order of , which motivates
 √  the use of an asymptotic expansion of the form
ψ̃(z) = 4 tanh−1 tanh(ζ/4)e− ĉz . (110)
c̄(x, t) ∼ c̄0 + 1/2 c̄1/2 +  c̄1 + . . . . (114)
Recall that ψ̃ is the excess voltage relative to the bulk, Using this expansion in (111) – (113), we find that the
ψ̃ = φ̃ − φ̂ and that the tilde and hat accents are used to leading order equations are:
indicate quantities within and just outside of the double ∂c̄0 ∂ 2 c̄0
layer, respectively. Also, we reiterate that unlike the 1D = (115)
∂ t̃ ∂z 02
case, ζ is not a constant but is a function of spatial along ∂

∂ φ̄0

the electrode surface. c̄0 0 = 0 (116)
Since the bulk concentration at the RC time is uni- ∂z 0 ∂z
form, the double layer structure is given by (110) with ĉ ρ̄0 = 0. (117)
set equal to 1. Notice that at the RC time, bulk concen- The initial conditions and boundary condition as z 0 → ∞
tration gradients have not yet had time to form, so the for these equations are c̄0 (t = 0) ≡ 1 and c̄0 (z 0 → ∞) =
variation of the diffuse layer concentration and charge 1, respectively. The boundary condition at z 0 = 0 is given
density along the surface of the sphere is solely due to a by flux matching with the double layer. Rescaling space
non-uniform zeta-potential. and time in (IV) and (60), we find that the appropriate
flux boundary conditions for the diffusion layer are

2. O ( ) Diffusion Layer ∂q  1 ∂ φ̄
= ∇s · q∇s log c̄ + w∇s φ̄ + √ c̄ 0 (118)
∂ t̃  ∂z
The analysis in the previous section leads us to an ap- ∂w  1 ∂c̄
= ∇s · w∇s log c̄ + q∇s φ̄ + √ . (119)
parent paradox. The dynamics of the system seem to ∂ t̃  ∂z 0
18

Thus, the leading


√ order flux boundary conditions, which field
√ in the diffusion layer is a constant and appears at
appear at O(1/ ), are O( ) so that

∂ φ̄1/2
 
∂ φ̄0 ∂φ 1 ∂ φ̄ ∂q
c̄0 = 0 (120) = √ ∼ = .
∂z 0 ∂Z Z=0
 ∂z 0
z 0 →∞
0
∂z z0 →∞ ∂ t̃
∂c̄0 (128)
= 0. (121)
∂z 0 Similarly, the definition of the zeta-potential and the
Stern boundary condition (29) across the entire diffusion
The leading order solutions in the diffusion layer, c̄0 ≡ 1 layer require that φ̄ ∼ φ (Z → 0) at leading order.
and φ̄0 = φ (Z → 0), are easily determined by applying
the initial and boundary conditions to (115) – (117).
To obtain dynamics, we must examine the first-order 4. Leading-order Dynamics
correction to the solution. At the next higher order, the
governing equations are
Using the results of our discussion, we now derive the
∂c̄1/2 2
∂ c̄1/2 leading order equations that govern the charging dynam-
= (122) ics on the surface of the sphere. Since charging is non-
∂ t̃ ∂z 02 uniform over the electrode surface, the equations take
2
∂ φ̄1/2 the form of partial differential equations on the surface
= 0 (123)
∂z 02 of the sphere. Defining Ψ(θ, φ) ≡ v − φ(r = 1, θ, φ),
ρ̄1/2 = 0 (124) we can write the Stern boundary condition (29) and the
double layer charging equation (128) as
where we have made use of the leading order solution to
simplify (123). Again, the initial conditions and bound- ζ + 2δ sinh (ζ/2) = Ψ (129)
ary condition at infinity are simple: c̄1/2 (t = 0) ≡ 0 ∂Ψ ∂φ
and c̄1/2 (z 0 → ∞) = 0. The flux boundary conditions C(Ψ) = (130)
∂ t̃ ∂n
at z 0 = 0, however, are more interesting because they
involve the charging of the double layer: where we have introduced the leading order differential
double layer capacitance C(Ψ) = −∂q/∂Ψ and used the
∂ φ̄1/2 fact that ĉ = 1 at the RC time. Together with (55), (129)

∂q
= (125) and (130) form a complete set of boundary conditions for
∂ t̃ ∂z 0 z=0
the leading order electric potential in the bulk region. To
∂c̄1/2

∂w
= . (126) compute the double layer capacitance, we can combine
∂ t̃ ∂z 0 z=0 (55) with (129) to obtain

Thus, simple diffusion of neutral salt at x = O( ) driven 1
by absorption into the O() double layer dominates the C= . (131)
sech (ζ/2) + δ
dynamics of the diffusion layer. From (123) and (125),
we see that the electric potential possesses a linear profile Since C depends on Ψ̃ (via ζ), the charging equation
with slope given by rate of surface charge build up in the (130) is nonlinear which makes the problem analytically
double layer: intractable.
  For small applied fields (where ζ ≈ Eo is a reasonable
∂q approximation), analytical progress can be made by lin-
φ̄ ∼ φ (Z → 0) + 1/2 z0. (127)
∂ t̃ earizing the the double-layer capacitance around ζ = 0
√ to obtain C ∼ 1/(1 + δ). This approximation leads to
Note that constant term at O( ) is forced to be zero a linear charging equation making it possible to solve
by matching with the bulk electric potential since all the equations as an expansion in spherical harmonics.
higher order corrections to the bulk potential are identi- Substituting the expansion (109) into (130) and the def-
cally zero. inition of Ψ, we obtain a decoupled system of ordinary
differential equations for the expansion coefficients:
3. Effective Boundary Conditions Across Entire Diffusion dA0 dv
Layer + (1 + δ)A0 =
dt̃ dt̃
dA1 dEo
It is precisely the fact that the electric potential has + 2(1 + δ)A1 = −(1 + δ)Eo +
dt̃ dt̃
the form (127) that justifies the common approach of dAl
asymptotic matching directly between the bulk and the + (1 + l)(1 + δ)Al = 0 , l > 1. (132)
dt̃
double layer [1, 16, 17]. For instance, the time-dependent
matching used by Bazant, Thornton, and Ajdari [1] rests To retain generality, we have allowed for the possibility
on the implicit assumption that the leading order electric that the applied electric field and surface potential are
19

time-varying. For the case of a steady surface potential 5. Numerical Model


and uniform applied field, we find that the bulk electric
potential is
For larger applied fields, the nonlinear double layer
v −(1+δ)t̃

1   capacitance forces us to use numerical solutions to gain
−2(1+δ)t̃
φ= e − Eo r cos θ 1 + 3 1 − 3e physical insight. Since the bulk electric potential is a
r 2r
time-varying harmonic function, it is most natural to nu-
(133)
merically model the evolution equation for the potential
where we have assumed that both the surface potential
using a multipole expansion with harmonic terms. Trun-
and the electric field are suddenly switched on at t̃ =
cating (109) after a finite number of terms yields a dis-
0. The initial condition in this situation is that of a
crete solution of the form:
conducting sphere at potential v in a uniform applied
electric field Eo : φ = v − Eo r cos θ 1 − 1/r3 .

N
Using (133), the double layer potential and total diffuse X Al (t̃)
φ̂(r, θ, t̃) = −Eo r cos θ + Pl (cos θ) (136)
charge are easily determined to be rl+1
l=0
 
Ψ = v 1 − e−(1+δ)t̃ where the unknowns are the time-dependent coefficients
3   in the expansion. By using (136) and enforcing that (130)
+ Eo cos θ 1 − e−2(1+δ)t̃ (134) is satisfied at the collocation points θi = iπ/N , we de-
2
v   rive a system of ordinary differential equations for the
q ∼ − 1 − e−(1+δ)t̃ coefficients Al (t̃):
1+δ
3  
− Eo cos θ 1 − e−2(1+δ)t̃ . (135) ~
2 (1 + δ) dA ~
CP = Eo P(:, 2) + QA (137)
dt̃
These results are consistent with the calculations by where A ~ is the vector of expansion coefficients, P and Q
Bazant and Squires [16, 17], which is expected since the are collocation matrices that relate the expansion coeffi-
low applied field limit corresponds to the regime where cients to φ and ∂φ/∂n, respectively, and C is a diagonal
the total diffuse charge is linearly related to the zeta- matrix that represents the double layer capacitance at
potential (and therefore the total double layer poten- the collocation points. Note that P(:, 2) (which repre-
tial). It is worth mentioning that in the common situa- sents the second column of P in MATLAB notation) is
tion where the surface potential is set well before t = 0, the discrete representation of P1 (cos θ) = cos θ, so the
then the first term in each of the above three expressions term Eo P(:, 2) accounts for the applied background po-
is not present. Analogous double layer charging formulae tential. The explicit forms for the collocation matrices
for the cylinders are shown in Table IV. and the discrete double layer capacitance are given by

   
P0 (cos θ0 ) P1 (cos θ0 ) . . . PN (cos θ0 ) P0 (cos θ0 ) 2P1 (cos θ0 ) . . . (N + 1)PN (cos θ0 )
 P0 (cos θ1 ) P1 (cos θ1 ) . . . PN (cos θ1 )   P0 (cos θ1 ) 2P1 (cos θ1 ) . . . (N + 1)PN (cos θ1 ) 
P= .. .. .. , Q= .. .. ..
   
..  .. 
 . . . .   . . . . 
P0 (cos θN ) P1 (cos θN ) . . . PN (cos θN ) P0 (cos θN ) 2P1 (cos θN ) . . . (N + 1)PN (cos θN )

C = diag (C(Ψ(θ0 ), C(Ψ(θ1 ), . . . , C(Ψ(θN )) . (138)

The system of ODEs for the expansion coefficients is eas- grounded, charging is dominated by the dipolar contribu-
ily solved using MATLAB’s built-in ODE solvers by mul- tion to the response (see Figure 15). While the nonlinear
tiplying (137) through by (CP)−1 and writing a simple capacitance does in fact allow higher harmonics to con-
function to evaluate the resulting right-hand side func- tribute to the response, the higher harmonics only play
tion. a small role even at larger applied fields. As expected,
when the sphere is kept at zero voltage, the antisymme-
try between the upper and lower hemisphere is not bro-
6. Dipole-Dominated Charging ken and only odd terms contribute to the series solution
(109). However, as shown in Figure 16, if a nonzero po-
tential is applied to the sphere, all harmonics contribute
From the numerical solution of the charging equa- to the solution. In this case, the dominant contributions
tion (130), we see that when the sphere is electrically
20

TABLE IV: Table of formulae for double layer charging of metallic cylinders and spheres at weak applied electric fields at the
RC-time. In these formulae, the potential, v, of the particle is set to zero.
Cylindera Sphere
h “ ”i h “ ”i
φ −Eo r cos θ 1 + r12 1 − 2e−(1+δ)t −Eo r cos θ 1 + 1
1 − 3e−2(1+δ)t
2r 3
“ ” “ ”
Ψ 2Eo cos θ 1 − e−(1+δ)t 3
E cos θ 1 − e−2(1+δ)t
2 o
“ ” “ ”
q − 1+δ2
Eo cos θ 1 − e−(1+δ)t 3
− 2(1+δ) Eo cos θ 1 − e−2(1+δ)t

aθ measured from vertical axis.

6 6
v=0 v=3
A
4 A
1
E=5 4 0 E=5
δ = 0.1 A
2
A
δ = 0.1
2 A3 2
3

A (t) A (t)
l l
0 0

−2 −2
A A
4 1
−4 −4
0 2 4 6 8 10 0 2 4 6 8 10
t t

FIG. 15: Time evolution of the dominant coefficients FIG. 16: Time evolution of the dominant coefficients
in the Legendre polynomial expansion of the bulk elec- in the Legendre polynomial expansion of the bulk elec-
tric potential in the weakly nonlinear regime at the RC tric potential in the weakly nonlinear regime at the RC
time. In this figure v = 0, E = 5 and δ = 0.1. Notice time when the sphere has a nonzero applied voltage.
that the dipolar term dominates the solution. In this figure v = 3, E = 5 and δ = 0.1. Note that
both symmetric and antisymmetric terms make non-
negligible contributions.

come from the monopolar and dipolar terms.


The dipolar nature of double layer charging forms the
foundation of much of the work on the charging of col- For sufficiently small δ, charging is slowed at higher ap-
loid particles over the past half century. For instance, plied fields because the sech(ζ/2) term in the denomina-
the non-equilibrium double layer is often characterized tor of the double layer capacitance (131) becomes negli-
in terms of the induced dipole moment [55]. As shown gible when ζ  −2 ln(δ/2).
in (133) – (135), the monopole and dipole contributions
are the only contributions in a linearized theory. Our
numerical investigations demonstrate that, even for the
nonlinear theory, the monopole and dipole response dom-
inates the total response. Our results provides additional C. Dynamics at the Diffusion Timescale
theoretical support for the focus on the dipole response
when studying colloid particles in applied fields.
In this section, we examine the response of the system
at the diffusion time scale. We find that the only dy-
namic process is diffusion of neutral salt within the bulk
7. Extended Double Layer Charging in response to surface adsorption that occurred at the
RC time scale. Since the total amount of neutral salt ab-
The slowing of double layer charging is one important sorbed by the diffuse charge layers during the charging
consequence of nonlinearity in the double layer capaci- phase is an O() quantity, the bulk concentration only
tance (see Figure 17). However, as shown in Figure 17 needs to decrease by O() to compensate. Thus, we find
extended charging only occurs for δ  1. Mathemati- that dynamics are not present at leading order; rather,
cally, we only see slowed charging at small values of δ be- they appear only in higher-order corrections. Also, at
cause the double layer capacitance (131) can only become the diffusion time, surface transport, which is negligible
as small as 1/δ, which occurs at large zeta-potentials. at the RC time scale, becomes important.
21

6 6 2. Higher-Order Bulk Diffusion


v=0 v=0
4
A E=5 4 E=5
A1
1
δ=1 δ = 0.01 In order to see dynamics, we need to consider the first-
2 2 A3
Al(t) Al(t) order correction to (139) and (140):
0 0
∂c1
−2 −2 = ∇2 c1 (144)
∂t
−4
0 1 2 3 4 5
−4
0 5 10 15 20 25 ∇2 φ1 = −∇c1 · ∇φ0 . (145)
t t

FIG. 17: Time evolution of the dominant coefficients


The boundary conditions for these equations are a bit
in the Legendre polynomial expansion of the bulk elec- more complicated. Using (IV) – (60) and taking into ac-
tric potential in the weakly nonlinear regime at the count the leading order solutions, we find that the bound-
RC time for low (left) and high (right) Stern capaci- ary conditions for the O() equations are
tance values. In these figures v = 0 and E = 5. Note
that double layer charging is retarded when δ is small ∂φ1
q0 δ + (t) = ∇s · (w0 ∇s φ0 ) − (146)
but that the slowdown in double layer charging is sup- ∂n
pressed for larger δ values. + ∂c1
w0 δ (t) = ∇s · (q0 ∇s φ0 ) − (147)
∂n
where q0 and w0 are the leading order equilibrium sur-
1. Leading Order Bulk and Double Layer Solutions face charge density and surface excess neutral salt con-
centration. As mentioned earlier, at the diffusion time
scale, these quantities are static in time. Also, note the
Substituting an asymptotic series into (18) – (20), we presence of the delta-functions in time, which account
obtain the leading order bulk equations: for the “instantaneous” adsorption of charge and neutral
salt from the bulk during the at the RC-time [1].
∂c0 Mathematically, the appearance of the delta-functions
= ∇2 c0 (139)
∂t is a consequence of the connection between the time
∇2 φ0 = 0. (140) derivative of double layer quantities at the two time scales
in the asymptotic limit  → 0. To illustrate this connec-
Applying the initial conditions (obtained by matching to tion, consider the time derivative of the surface charge
the solution at the RC time) and the leading order bound- density, q. Let t and t̃ be scaled to the diffusion and RC
ary conditions (derived from the effective flux boundary times, respectively, so that t = t̃. At these two time
conditions (IV) and (60)) yields the simple leading order scales, the
 surface charge density can be written as q(t)
solutions and q̃ t̃ which are simply related by q(t) = q̃ t̃ . The
time derivatives, however, are related by
c0 (~x, t) ≡ 1 (141) ∂q 1 ∂ q̃

1
 (t) = (t/) . (148)
φ0 (~x, t) = −Eo r cos θ 1 + 3 . (142) ∂t  ∂ t̃
2r
Therefore, for nonzero t, ∂q ∂t = 0 in the asymptotic limit
∂ q̃
Notice that there is no time dependence for either the because ∂ t̃ (t/) approaches zero faster than linearly as
concentration or the electric potential. It is worth men-  → 0 (as an example, see (135) ). In contrast, for t = 0,
tioning that for a general geometry, the initial condition ∂q ∂ q̃
∂t is infinite because ∂ t̃ (0) has a fixed nonzero value.
is a uniform concentration profile with the electric poten- Next, consider the following integral with t2 > 0:
tial of an insulator in an applied field and the boundary
Z t2 Z t2 /
conditions (which are consistent with the initial condi- ∂q ∂ q̃
tions) are dt = dt̃ = q̃ (t2 /) − q̃ (t1 /) . (149)
t1 ∂t t1 / ∂ t̃

∂φ0 ∂c0 In the asymptotic limit, the integral approaches zero for
c0 = 0 and = 0. (143)
∂n ∂n nonzero t1 but approaches q̃(∞) when t1 equals zero.
Putting the above properties together, we see that ∂q ∂t (t)
At the diffusion time scale, the double layer contin- is indeed a one-sided delta-function of strength q̃(∞) =
ues to remain in quasi-equilibrium, so its leading order q0 .
structure is given by (110) and (??) with the bulk concen- In contrast to one-dimensional systems, the boundary
tration set equal to 1. However, unlike the double layer layers play a more active role in the evolution of the bulk
at the RC time scale, the leading order zeta-potential is concentrations because surface conduction continues to
not evolving, so the leading order double layer structure play a role well beyond the initial injection of ions at
is static in time. t = 0. Note, however, that the surface conduction terms
22

in (146) and (147) are static, so they essentially impose As discussed in section IV, this condition also implies
fixed normal flux boundary conditions on the O() bulk that surface adsorption of ions from the bulk is larger
equations. enough to trigger significant bulk diffusion and surface
transport through the double layer, in steady steady
state. However, as noted by Bazant et al. [1], weakly
3. Comparison with the Analysis of Dukhin and Shilov nonlinear asymptotics breaks down during relaxation
√ dy-
namics at somewhat smaller voltages, α/ π < 1, or
(with units restored)
It is interesting to compare and contrast our weakly
nonlinear analysis with the work of Dukhin and Shilov
 
kT λS  πa 
[41, 44] on the polarization of the double layer for highly E0 > 1+ log , (151)
zea λD λ
charged, spherical particles in weak applied fields. In
both cases, bulk concentration variations and diffusion since large surface adsorption can occur only temporarily
processes appear as a higher-order correction to a uni- at certain positions. (The factor π in this formula comes
form background concentration and are driven by surface from a one-dimensinal analysis of bulk diffusive relax-
conduction. However, the significance of the surface con- ation, which does not strictly apply here, but the rough
duction term arises for different reasons. As mentioned scale should be correct.
earlier, the small parameters that controls the size of the Beyond the weakly nonlinear regime, there are two
correction is different in the two analyses. In Dukhin main effects that occur: (1) transient, local depletion of
and Shilov’s analysis, the small parameters are  and Eo . the leading order bulk concentration and (2) surface con-
Because they essentially use asymptotic series in Eo as duction at the leading order. As in the case of a steady
the basis for their analysis, they require that the double applied field, perhaps the greatest impact of α = O(1) is
layer be highly charged in order for surface conduction to that we must pay attention to factors of the form eζ or
be of the same order of magnitude as the O(Eo ) normal  sinh(ζ), in addition to factors of , when ordering terms
flux of ions from the bulk. In other words, because the in asymptotic expansions.
size of the surface conduction is proportional to Eo , in In the thin double-layer limit, the boundary layers are
order for surface conduction to be of the same order of still in quasi-equilibrium, as long as the bulk concen-
magnitude as the normal flux, the surface charge density tration does not approach zero, which would typically
must be an O(1) quantity: q = 2 sinh(ζ0 /2) = O(1). require Faradaic reactions consuming ions at the con-
In contrast, we use asymptotic series in  in our analysis ducting surface [50, 61]. Since we only consider ideally
and do not restrict Eo to be small, so the surface conduc- polarizable conductors, bulk depletion is driven solely by
tion and normal flux of ions from the bulk have the same adsorption of ions in the diffuse layer, which is unlikely to
order of magnitude for small O() surface charge densi- exceed diffusion limitation and produce non-equilibrium
ties regardless of the strength of the applied electric field space charge, although this possibility has been noted [1].
(as long as it is not so large that the asymptotic anal- Therefore, we would like to proceed as in the previous
ysis breaks down). Thus, our weakly nonlinear analysis sections and treat the bulk as locally electroneutral with
complements the work of Dukhin and Shilov by extend- effective boundary conditions.
ing their analysis to stronger applied electric fields and
the case where the surface charge density is induced by
the applied field rather than fixed by surface chemistry B. Leading-Order Equations
of the colloid particle.
Unfortunately, the analysis of the leading order equa-
tions derived in this manner does not appear to be as
VIII. STRONGLY NONLINEAR RELAXATION straightforward as the analysis in the weakly nonlinear
limit. The main problem is that it seems difficult to de-
A. Definition of “Strongly Nonlinear” rive a uniformly valid leading order effective boundary
conditions along the entire surface of the sphere. For
this reason, we merely present the apparent leading or-
The weakly nonlinear analysis of the thin double-layer der equations for the strongly nonlinear regime and leave
limit in the previous section assumes that the dimension- a thorough analysis for future work.
less surface charge density α and surface salt density β At the leading order in the bulk, we find the usual
remain uniformly small. For our PNP model, we can equations for a neutral binary electrolyte:
estimate when this assumption becomes signficantly vio-
lated using Eq. (64) with ζ = E0 a/(1 + δ) and C = C0 , ∂c0
which occurs at fields above a critical value at least a few = ∇2 c0 (152)
∂t
times the thermal voltage, 0 = ∇ · (c0 ∇φ0 ) (153)

with ρ = O(2 ). The structure of the boundary layers


 
2kT λS a
E0 > 1+ log . (150)
zea λD λ is described by GCS theory with the concentration and
23

charge density profiles given by (110). Effective bound- electric fields, using the standard mathematical model
ary conditions for (152) – (153) are derived in the same of the Poisson-Nernst-Planck (PNP) equations. Unlike
manner as for a steady applied field except that unsteady most (if not all) prior theoretical studies, we have focused
terms are retained. Recalling that q and w grow expo- on the nonlinear response to “large” electric fields, which
nentially with the zeta-potential, we find that both the transfer more than a thermal voltage to the double layer
surface conduction terms and the time derivatives of the after charging. We have effectively extended the recent
total diffuse charge and excess concentration appear in nonlinear analysis of Bazant, Thornton and Ajdari [1] for
the leading order boundary conditions: the one-dimensional charging of parallel-plate blocking
electrodes, to some new situations in higher dimensions,
∂ q̃0 ∂φ0 where the potential of the conductor is not controlled.
 = ∇s · (w̃0 ∇s φ0 ) − c0 (154)
∂t ∂n Instead, electrochemical relaxation in our model prob-
∂ w̃0 ∂c0 lems is driven by a time-dependent background electric
 = ∇s · (q̃0 ∇s φ0 ) − . (155)
∂t ∂n field applied around the conductor, whose charges are
completely free to relax. We are not aware of any prior
analysis of nonlinear response in such problems (whether
C. Challenges with Strongly Nonlinear Analysis or not using the PNP equations), and yet it is important
in many applications, in microfluidics, colloids, and elec-
It is important to realize that these equations are trochemical systems, where applied fields or voltages are
mathematically much more complicated than the anal- often well beyond the linear regime.
ogous equations in the weakly nonlinear regime. First, We have focused on the structure and dynamics of the
the nonlinearity due to the electromigration term explic- double layer and the development of bulk concentration
itly appears in the bulk equations at leading order; the gradients. A major goal has been to move beyond the tra-
nonlinearity is not removed by the asymptotic analysis. ditional circuit models commonly used to study the linear
Furthermore, the diffuse layer concentrations depend on response of electrochemical systems. Through a combi-
time explicitly through the bulk concentration at the sur- nation of analytical and numerical results, we have shown
face in addition to the zeta potential: that significantly enhanced ion concentration within the
double layer is a generic feature of nonlinear electrochem-
c̃± (t) = c± (t)e∓ψ̃(t) . (156) ical relaxation around a conductor. By interacting with
the bulk electric field, the enhanced concentration within
Already, these features of the equations greatly increases the double layer leads to large surface current densities.
the challenge in analyzing the response of the system. In addition, because the double layer does not charge
However, the greatest complication to the mathemat- uniformly over the surface, tangential concentration gra-
ical model in the strongly nonlinear regime is that effec- dients within the double layer itself lead to surface diffu-
tive boundary conditions (154) and (155) are not uni- sion. Due to their coupling to bulk transport processes
formly valid over the surface of the sphere (or cylinder). via normal fluxes of ions into the double layer, these sur-
Near the poles, the double layer charges quickly, so the face transport process drive the formation of bulk con-
time-dependent and surface transport terms in the ef- centration gradients.
fective boundary conditions become O(1) at very short We have also found that bulk concentration gradients
times. In contrast, the amount of surface charge in the are always present to some degree and play an important
double layer near the equator is always a small quantity. role in allowing the system to relax to the steady-state.
Thus, it would seem that near the equator, the only sig- For weak applied fields, they are often neglected because
nificant terms in the effective boundary conditions are they only appear as a first-order correction to a uniform
the normal flux terms. Together, these observations sug- background concentration profile. However, for strong
gest that the appropriate set of boundary conditions to applied fields, the variations in the bulk concentration
impose on the governing equations (152) and (153) de- becomes as large as the background concentration, so
pends on the position on the surface of the sphere. More- they cannot be ignored. These bulk concentration gra-
over, the position where the boundary conditions switch dients lead to bulk diffusion currents which result in net
from one set to the other depends on time as double-layer circulation of neutral salt in the region near the metal
charging progresses from the pole towards the equator. colloid particle.
Another key contribution of this work is a careful
mathematical derivation of general effective boundary
IX. CONCLUSION
conditions for the bulk transport equations in the thin
double layer limit by applying matched asymptotic ex-
A. Predictions of the PNP Model pansions to the PNP equations. We derive a set of bound-
ary equations that relate the time evolution of excess sur-
In this paper, we have analyzed electrochemical relax- face (double-layer) concentrations to surface transport
ation around ideally polarizable conducting spheres and processes and normal flux from the bulk. An interesting
cylinders in response to suddenly applied background feature of the effective boundary conditions is that they
24

explicitly involve the small parameter  and may have electrochemical systems.
different leading order forms depending on the charac- Of course, electrochemists are well aware of this prob-
teristic time scale and the magnitude of double layer ion lem; in fact, Stern originally proposed the compact layer
concentrations. of adsorbed ions, outside the continuum region, as a way
It is beyond the scope of this article, but worth pur- to cutoff the unbounded double-layer capacitance [62].
suing, to study “strongly nonlinear” dynamics in detail, Since then, various empirical models of the compact layer
where (according to the PNP equations) the double layer have appeared [34, 35], but steric effects (or other non-
adsorbs enough ions to significantly perturb the bulk con- linearities) in concentrated solutions have received much
centration and cause surface transport to bulk transport less attention. Borukhov, Andelman and Orland recently
of ions. The challenging issues discussed at the end of postulated a continuum free energy for ions, taking into
Section VIII need to be addressed in order to gain a account steric repulsion with the usual mean-field elec-
deeper understanding of the rich behavior of metal col- trostatics and minimized it to derive a modified Poisson-
loid systems in this practially relevant regime. Also, it Boltzmann equation for potential in the double layer [63].
would be beneficial to validate solutions of the effective Their model predicts a steady, equilibrium profile of ions
bulk and boundary conditions in the thin double layer with a condensed layer at the steric limit for large zeta
limit against solutions for the full PNP equations. The potentials. By extending this approach to obtain the
utility of the thin double layer approximation cannot be chemical potential, Kilic, Bazant and Ajdari [48] have de-
fully appreciated until this comparison is completed. rived modified-PNP equations and studied steric effects
on double layer charging, but further model development
is still needed, not only for steric effects, but also for field-
B. The Need for Better Continuum Models dependent permittivity and/or diffusivity. The validity
of a continuum model at the scale of several atoms in
Unfortunately (or fortunately, depending on one’s per- the most condensed part of the double layer must also be
spective), the theoretical challenge of strongly nonlinear viewed with some skepticism.
electrochemical relaxation is much deeper than just solv- Nevetheless, in spite of these concerns, it is a natural
ing the PNP equations in a difficult regime: It is clear first step to study nonlinear electrochemical relaxation
that the dynamical equations themselves must be modi- using the standard PNP equations as we have done. The
fied to better describe the condensation of ions in highly details of our results will surely change with modified
charged double layers. As emphasized by Newman [37], transport equations and/or surface boundary conditions,
the PNP equations are only justified for dilute solu- but we expect that some key features to be robust. For
tions, since each ion moves in response to an independent example, steric effects could decrease the capacitance of
stochastic force with constant diffusivity and mobility, the double layer at large zeta potentials, but surface con-
interacing with other ions only through the mean long- duction and adsorption, coupled to bulk diffusion, should
range Coulomb force. Important effects in concentrated still occur, albeit perhaps with smaller magnitude for a
solutions, such as short-range forces, many-body correla- given applied field. Also the mathematical aspects of our
tions, steric constraints, solvent chemistry, and nonlinear boundary-layer analysis, such as the derivation of the
mobility, diffusivity, or permittivity, are all neglected. surface conservation law (48), could be applied to any
It is tempting to trust the PNP equations in the case of continuum transport equations.
a dilute bulk electrolyte around a initially uncharged sur-
face. However, as we have shown, the model predicts its
own demise when a large electric field is applied, due to Acknowledgments
enormous increases in surface concentration in the diffuse
part of the double layer, even if the bulk concentration This work was supported in part by the Department
is small. Note that the condition for strongly nonlinear of Energy Computational Science Graduate Fellowship
relaxation (150) or (151) is similar to the condition for Program of the Office of Science and National Nuclear
the breakdown of the PNP equations – both require ap- Security Administration in the Department of Energy
plied voltages across the double layer only several times under contract DE-FG02-97ER25308 (KTC) and by the
the thermal voltage. In particular, the Gouy-Chapman MRSEC Program of the National Science Foundation un-
solution to the PNP equations predicts an absurd con- der award number DMR 02-13282 (MZB and KTC). The
centration of counterions of one per Å3 at the surface for authors thank A. Ajdari, J. Choi and L. H. Oleson for
a surprisingly small zeta potential around 5kT /ze ≈ 0.2 many helpful discussions. We are particularly grateful to
V even in a fairly dilute 0.01 M electrolyte. This crit- L. H. Olesen for pointing out an important term that we
ical voltage is routintely applied to the double layer in had missed in deriving Eq. (46).

[1] M. Z. Bazant, K. Thornton, and A. Ajdari, Phys. Rev. [2] M. Sluyters-Rehbach and J. H. Sluyters, Electroanalytical
E 70, 021506 (2004).
25

Chemistry (Marcel Dekker, New York, 1970), vol. 4, pp. University Press, Oxford, 2001).
1–128. [39] J. Lyklema, Fundamentals of Interface and Colloid Sci-
[3] J. R. Macdonald, Electrochim. Acta 35, 1483 (1990). ence. Volume II: Solid-Liquid Interfaces (Academic Press
[4] R. Parsons, Chem. Rev. 90, 813 (1990). Limited, San Diego, CA, 1995).
[5] L. A. Geddes, Ann. Biomedical Eng. 25, 1 (1997). [40] I. N. Simonov and V. N. Shilov, Colloid J. USSR 39, 775
[6] T. M. Squires and S. R. Quake, Rev. Mod. Phys. 77 (1977).
(2005). [41] S. S. Dukhin and V. N. Shilov, Colloid. J. USSR. 31, 564
[7] A. Ramos, H. Morgan, N. G. Green, and A. Castellanos, (1969).
J. Colloid Interface Sci. 217, 420 (1999). [42] E. J. Hinch and J. D. Sherwood, J. Fluid. Mech. 132,
[8] A. Ajdari, Phys. Rev. E 61, R45 (2000). 337 (1983).
[9] A. González, A. Ramos, N. G. Green, A. Castellanos, [43] E. J. Hinch, J. D. Sherwood, W. C. Chew, and P. N. Sen,
and H. Morgan, Phys. Rev. E 61, 4019 (2000). J. Chem. Soc. Faraday. Trans. II 80, 535 (1984).
[10] N. G. Green, A. Ramos, A. González, A. Castellanos, [44] V. N. Shilov and S. S. Dukhin, Colloid. J. USSR. 32, 90
and H. Morgan, Phys. Rev. E 66, 026305 (2002). (1970).
[11] N. G. Green, A. Ramos, A. González, H. Morgan, and [45] J. J. Bikerman, Trans. Faraday Soc. 36, 154 (1940).
A. Castellanos, Phys. Rev. E 61, 4011 (2000). [46] B. V. Deryagin and S. S. Dukhin, Colloid. J. USSR. 31,
[12] A. Ramos, A. González, A. Castellanos, N. G. Green, 277 (1969).
and H. Morgan, Phys. Rev. E 67, 056302 (2003). [47] K. T. Chu and M. Z. Bazant (2006), in preparation.
[13] V. Studer, A. Pépin, Y. Chen, and A. Ajdari, Microelec. [48] M. S. Kilic, M. Z. Bazant, and A. Ajdari (2006).
Eng. 61, 915 (2002). [49] Lyklema, Phys. Rev. E 71, 032501 (2005).
[14] V. Studer, A. Pépin, Y. Chen, and A. Ajdari, Analyst [50] M. Z. Bazant, K. T. Chu, and B. J. Bayly, SIAM J. Appl.
129, 944 (2004). Math. 65, 1463 (2005).
[15] F. Nadal, F. Argoul, P. Kestener, B. Pouligny, C. Ybert, [51] A. Bonnefont, F. Argoul, and M. Bazant, J. Electroanal.
and A. Ajdari, Eur. Phys. J. E 9, 387 (2002). Chem. 500, 52 (2001).
[16] M. Z. Bazant and T. M. Squires, Phys. Rev. Lett. 92, [52] K. T. Chu, Ph.D. thesis, Massachusetts Institute of Tech-
066101 (2004). nology, Department of Mathematics (2005).
[17] T. M. Squires and M. Z. Bazant, J. Fluid Mech. 509, 217 [53] J. J. Bikerman, Z. Phys. Chem. Abt. A 163, 378 (1933).
(2004). [54] J. J. Bikerman, Kolloid-Z. 72, 100 (1935).
[18] J. A. Levitan, S. Devasenathipathy, V. Studer, Y. Ben, [55] S. S. Dukhin, Adv. Colloid Interface Sci. 44, 1 (1993).
T. Thorsen, T. M. Squires, and M. Z. Bazant, Colloids [56] L. N. Trefethen, Spectral Methods in MATLAB (SIAM,
and Surfaces A 267, 122 (2005). Philadelphia, PA, 2000).
[19] M. Trau, D. A. Saville, and I. A. Aksay, Langmuir 13, [57] J. P. Boyd, Chebyshev and Fourier Spectral Methods
6375 (1997). (Dover Publications, Inc., Mineola, NY, 2001), 2nd ed.
[20] S. Yeh, M. Seul, and B. Shraiman, Nature (London) 386, [58] B. Fornberg, A Practical Guide to Pseudospectral Meth-
57 (1997). ods (Cambridge University Press, New York, NY, 1998).
[21] C. Faure, N. Decoster, and F. Argoul, Eur. Phys. J. B 5, [59] P. Debye and H. Falkenhagen, Physik. Z. 29, 121 (1928).
87 (1998). [60] E. W. Weisstein, Modified spherical bessel function of
[22] N. G. Green, A. Ramos, and H. Morgan, J. Phys. D 33, the second kind, From MathWorld–A Wolfram Web Re-
632 (2000). source, http://mathworld.wolfram.com/
[23] C. Marquet, A. Buguin, L. Talini, and P. Silberzan, Phys. ModifiedSphericalBesselFunctionoftheSecondKind.html.
Rev. Lett. 88 (2002). [61] K. T. Chu and M. Z. Bazant, SIAM J. Appl. Math. 65,
[24] F. Nadal, F. Argoul, P. Hanusse, B. Pouligny, and A. Aj- 1485 (2005).
dari, Phys. Rev. E 65 (2002). [62] O. Stern, Z. Elektrochem. 30, 508 (1924).
[25] W. D. Ristenpart, I. A. Aksay, and D. A. Saville, Phys. [63] I. Borukhov, D. Andelman, and H. Orland, Phys. Rev.
Rev. Lett. 90 (2003). Lett. pp. 235–258 (1997).
[26] V. A. Murtsovkin, Colloid Journal 58, 341 (1996). [64] J. D. Ferry, J. Chem. Phys. 16, 737 (1948).
[27] E. Yariv, Phys. Fluids 17, 051702 (2005). [65] S. S. Dukhin and V. N. Shilov, Adv. Colloid Interface
[28] T. M. Squires and M. Z. Bazant, J. Fluid Mech. in press Sci. 13, 153 (1980).
(2006). [66] A. A. Kornyshev and M. A. Vorotyntsev, Electrochimica
[29] K. Rose and J. Santiago (????), preprint. Acta 26, 303 (1981).
[30] D. Saintillon, E. Darve, and E. Shaqfeh (????), preprint. [67] J. R. Macdonald, Trans. Faraday Soc. 16, 934 (1970).
[31] W. Helfrich, Z. Naturforsch. C 29, 182 (1974). [68] Note that when charging is driven by an externally ap-
[32] M. D. Mitov, P. Méleard, M. Winterhalter, M. I. An- plied voltage or field, the relevant relaxation time for
gelova, and P. Bothorel, Phys. Rev. E 48, 628 (1993). double layer charging is not the often quoted Debye
[33] R. Pethig, Crit. Rev. Biotechnol. 16, 331 (1996). time, τD = λ2D /D [38? ]. The Debye time is the cor-
[34] P. Delahay, Double Layer and Electrode Kinetics (New rect characteristic response time for double layer charging
York, 1965). only in the unphysical scenario where charge is instan-
[35] J. O. Bockris and A. K. N. Reddy, Modern Electrochem- taneously placed on the particle (as opposed to trans-
istry (New York, 1970). ported through the electrolyte) [64]. This result has been
[36] A. J. Bard and L. R. Faulkner, Electrochemical Methods discovered many times by different scientific communi-
(John Wiley & Sons, Inc., New York, NY, 2001). ties [1, 55, 65–67] but only recently seems to be gaining
[37] J. Newman, Electrochemical Systems (Prentice-Hall, widespread understanding.
Inc., Englewood Cliffs, NJ, 1991), 2nd ed. [69] The residual is computed using the L∞ norm.
[38] R. J. Hunter, Foundations of Colloid Science (Oxford