Journal of Electroanalytical Chemistry 655 (2011) 23–31 Contents lists available at ScienceDirect Journal of Electroanalytical Chemistry journal homepage: www.elsevier.com/locate/jelechem How well does simple RC circuit analysis describe diffuse double layer capacitance at smooth micro- and nanoelectrodes? Edmund J.F. Dickinson, Richard G. Compton ⇑ Department of Chemistry, Physical and Theoretical Chemistry Laboratory, University of Oxford, South Parks Road, Oxford OX1 3QZ, United Kingdom a r t i c l e i n f o Article history: Received 9 November 2010 Received in revised form 7 February 2011 Accepted 16 February 2011 Available online 20 February 2011 Keywords: Diffuse double layer Nernst–Planck–Poisson equations Capacitive charging RC circuit Electrochemical impedance spectroscopy Nanoelectrode a b s t r a c t The capacitive charging of a diffuse double layer is discussed. Results from simple RC circuit analysis (an ideal resistor and capacitor in series) are compared with results from a more complete model in which the Nernst–Planck–Poisson equations are solved in a hemispherical space, both analytically and by simulation. This complementary approach allows an assessment of certain conditions which are required in order for RC circuit analysis to be suitable to describe the diffuse double layer. In particular, deviations are noted for nanoscale electrodes. Additionally, RC circuit behaviour breaks down for applied overpotentials greater than 25 mV (RT/F), such that values for solution resistance and double layer capacitance inferred from impedance spectroscopy may not apply to other experimental techniques. These conclusions apply to a smooth electrode and so are not associated with ‘‘constant phase angle’’ effects. Ó 2011 Elsevier B.V. All rights reserved. 1. Introduction circuit is at all physically meaningful for describing the diffuse double layer and its interface with a charged electrode. 1.1. Preamble It has been well understood since the early days of solutionphase electrochemistry that the excess charge on the surface of an electrode, an unmeasurable property, induces a ‘double layer’ with opposite charge. This exists in the form of a charge separation due to unequal concentrations of cations and anions in the electrolyte solution close to the electrode interface. The model due to Stern [1] and Grahame [2] considered this double layer to consist of a compact layer containing adsorbed solvent, and possibly adsorbed ionic species, and a diffuse layer in which charge separation decays smoothly into bulk solution. Charging processes, particularly when occurring in tandem with electrolysis, are commonly analysed using the RC equivalent circuit (Fig. 1), in which an ideal resistor and capacitor are considered in series with a voltage source [3]. The solution is therefore considered to have an ideal, constant and homogeneous resistance, Rs, and the double layer is considered to have an ideal, constant and homogeneous capacitance, Cd. In the following discussion we shall use the Nernst–Planck– Poisson equations (Eqs. (1.1) and (1.2)) in conjunction with the law of conservation of mass (Eq. (1.3)) to describe both equilibrium and charging problems, to determine under what conditions the RC ⇑ Corresponding author. Tel.: +44 (0) 1865 275413; fax: +44 (0) 1865 275410. E-mail address: richard.compton@chem.ox.ac.uk (R.G. Compton). 1572-6657/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jelechem.2011.02.016 zi F C i r/ Ji ¼ Di rC i þ RT X F r2 / þ zi C i ¼ 0 s 0 @C i ¼ r Ji @t ð1:1Þ ð1:2Þ i ð1:3Þ For species i, Ji is the flux vector, Ci is concentration, Di is diffusion coefficient, and zi is charge number in unit electron charges. / is potential, F is the Faraday constant, R is the gas constant, T is temperature and s0 is the permittivity of the solvent medium. t is time. The compact layer will not be discussed, but its incorporation into the arguments ought not be problematic. Here our arguments will concern ideally smooth electrodes, and so we are not concerned with deviations from ideal RC behaviour associated with ‘‘constant phase angles’’ due to electrode surface roughness [4–6]. Rather, we are concerned with defining regimes in which the charging behaviour of an ionic solution differs from the ideal circuit sometimes used to describe it. In particular we shall discuss the effects of size, such that double layers at nanoelectrodes behave differently from those at microelectrodes, due in part to the pronounced curvature of the double layer [7,8]. Additionally the dependence of the meaning of certain parameters upon the experiment used to determine them will be underlined. 24 E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 Ecap(t) V V¼ I¼ C Z¼ i (t ) ð1:11Þ E 1 þ jRs C d x 1 1 1 ¼ Rs þ ¼ Rs j jx C d jx Cd x EC d ð1:12Þ 0 E (t ) Fig. 1. Schematic of the RC equivalent circuit. 1.2. Results from the RC equivalent circuit In the RC circuit (Fig. 1), we may use Kirchhoff’s voltage law to take the sum of voltages as zero. If the applied voltage is E, the charge on the capacitor is q, and the current is i: q ¼0 Cd EC d 1 þ jRs C d x so A E iRs ð1:10Þ and R V E jx ð1:4Þ 00 Therefore Z = Rs and Z = 1/Cdx, such that the resistive and capacitive components of charging are separated into the real and imaginary components of the impedance, respectively. The problems above, and others, have been reviewed in the textbook of Bard and Faulkner [11]. In all of the above q = 0 has been taken to be zero when t = 0, i.e. the step or sweep is beginning at the potential of zero charge (PZC) of the system. This can rarely be guaranteed in practice. In this paper we shall consider the merits of this idealised RC circuit description as compared to a more complete analysis on the basis of transport equations, still assuming a smooth electrode. In particular the physical meaningfulness of the parameters Rs and Cd will be assessed. Noting that i = dq/dt, and rearranging: dq q E þ ¼ dt Rs C d Rs ð1:5Þ such that if a potential step is applied from initial conditions of q = 0: t q ¼ EC d 1 exp Rs C d ð1:6Þ and i¼ dq E t ¼ exp dt Rs Rs C d ð1:7Þ It is evident from the above results that q = ECd at t ? 1, thus confirming that the capacitor is ideal, and that the charging current decays according to a characteristic time constant RsCd, such that more resistive or more capacitive interfaces will charge more slowly. The RC circuit falls under a class of problems in circuit analysis which are described as linear time-invariant [9]. For such problems, the output (current) is a linear function of the input (voltage) and therefore the current for any voltage waveform can be determined from the current for a single example. Therefore the solution for the potential step experiment can be converted into a solution for the problem with swept potential, E = vt, by the mathematical process of convolution. The result is that: t i ¼ EC d v 1 exp Rs C d ð1:8Þ Therefore a constant charging current is attained which is only a function of the capacitance; after a certain delay, the system attains the condition of ideal capacitance. In impedance spectroscopy [10], the impedance spectrum 0 00 Z(jx) = Z (x)jZ (x) may be inferred from Ohm’s law for the linear system by Laplace transformation under a Laplace frequency jx [9]: ZðjxÞ ¼ VðjxÞ IðjxÞ ð1:9Þ where V and I are the Laplace (frequency) space transforms of the potential step transient for charging, j is the imaginary unit and x is the spectroscopic frequency. It follows that: 2. Theory 2.1. Development of a self-consistent model based on transport equations The major alternative interpretation of the double layer is in terms of the Nernst–Planck (electrodiffusion) equation for the mass transport of a species, together with the Poisson equation relating the charge density and the electric field. At steady state, the Nernst–Planck equation reduces to the thermodynamic Boltzmann equation, and solution of the combined Poisson–Boltzmann equation was first achieved for a planar geometry by Gouy [12,13] and Chapman [14] independently, yielding the Gouy–Chapman equation for the capacitance of the double layer. A great deal of work has since been done in terms of considering diffuse double layers, both at steady state and dynamically, and how their formation varies with applied voltage and solution composition. This work has been thoughtfully reviewed by Bazant et al. [15] and further study by Beunis et al. [16] introduced a variety of perceived limiting cases. Bazant et al. [15] additionally extended the range of analytical solution, providing an exact Laplace transform for the transient capacitive current for low applied voltages where the Nernst–Planck equation can be linearised, as well as applying asymptotic analysis to non-linear cases. Common to almost all of these studies, however, is the adoption of a linear geometry for the study. The argument is that since the size of the electrode greatly exceeds the typical extent of the double layer, the effect of double layer curvature only arises for very small electrodes, as reported by Dickinson et al. [7], and so the double layer is indeed physically planar, to all intents and purposes for electrodes larger than nanoscale. This argument applies exclusively at equilibrium, however. Immediately following a potential step, the electric field will extend in all directions – and so has spherical symmetry, assuming that the cell is large with respect to the electrode. Unless the solution is confined to a capillary leading directly to the reference electrode, there is very little linearity in the system, and an attempt to consider a simple model problem underlines this. Let us suppose that the cell is being potentiostatically held at its potential of zero charge Epzc, such that solution is everywhere electroneutral. Then suppose the cell potential is rapidly stepped to a E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 new potential E. If we assume that reorientation of the polar solvent is much faster than the motion of solvated ions, an electric field is established across solution instantaneously, with the solution remaining roughly electroneutral. If the cell is very large compared to the electrode, a typical situation, we must solve: r2 / ¼ 0 ð2:1Þ such that / = E Epzc at the electrode, and / = 0 at an infinite distance from the electrode. If the geometry is linear (with coordinate x), we immediately encounter a problem since there exists no solution consistent with these boundary conditions. If we introduce an arbitrary double layer length x = L where / = 0, we may write / ¼ ðE Epzc Þ 1 xL , but L must then be a strong function of the displacement of the reference and working electrodes. We then expect the initial capacitive current to depend strongly on this displacement, even when it is very large, which is a rather nonsensical situation and entirely ignores bulk solution which is equally distant from both electrodes, in which range electroneutrality is also retained. In a hemispherical space (with coordinate r), however, the Cou lombic solution / ¼ ðE Epzc Þ rre arises straightforwardly, giving a simple dependence on electrode radius, re, and no dependence at all on the exact position of the reference electrode, provided it is sufficiently distant and bulk solution is plentifully available. This conforms to expected behaviour and has the added advantage that all length scales are quantified easily and in a manner that is not arbitrary. We shall therefore consider all transport in a hemispherical space. In the limit re rD, where rD is the Debye length, the linear result should be recovered at steady state. 2.2. Choice of transformed coordinates for analysis To generalise the process of solution as much as possible, dimensionless variables and coordinates are employed. We note that the work of Bazant et al. [15] clarified the presence of mixed diffusional and Debye timescales. In a spherical space, these scales are: Dt sdiff ¼ 2 re Dt sD ¼ 2 rD ð2:2Þ ð2:3Þ The former is widely applied in voltammetric problems. In the latter, the Debye length rD is a representation of the screening length of electric fields, which may be determined by dimensional analysis of the Poisson equation (Eq. (1.2)). Potential is normalised to the thermal volt RT/F and represented h, and concentration is normalised to bulk concentration C⁄ and represented ci: r2 h þ 1 X zi c i ¼ 0 2r 2D i ð2:4Þ where rD ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi RT s 0 2F 2 C ð2:5Þ A factor of two is incorporated at this stage to simplify future equations. It is therefore clear that a space normalisation by rD will render the Poisson equation ideally dimensionless, i.e. without extraneous coefficients. However, a space normalisation by re and time normalisation by r 2e is standard for diffusional processes and renders a Coulombic field ideally dimensionless. Following Bazant et al. [15] we shall compromise by normalising the coordinates according to a geometric mean of the two scales: r R ¼ pffiffiffiffiffiffiffiffiffi re rD D t s¼ A re rD 25 ð2:6Þ ð2:7Þ where DA is a chosen normal species. With the incorporation of conservation of mass (Eq. (1.3)) we achieve a dimensionless equation set: @ci Di ¼ r ðrci þ zi ci rhÞ @ s DA 1 X r2 h þ R2e zi c i ¼ 0 2 i ð2:8Þ ð2:9Þ where Re is the value of R where r = re, and hence R2e ¼ re =r D , the ratio of the electrode size to the Debye length. For most systems, Re 1 will therefore hold. 3. Differential capacitance The capacitance of an interface is a measure of how much charge can be separated across it for a given potential difference. Differential capacitance considers this property for different potentials under an equilibrium condition. Therefore i = 0, corresponding to the t ? 1 condition of the RC circuit. Hence steady-state problems are not strictly RC circuit problems, but physical theories at steady state can provide a value for Cd and therefore it is entirely relevant to discuss steady-state problems in order to find the physical meaning of this value in the RC circuit problem, if any. In the absence of electrolysis, the fluxes of all species are zero and hence the equilibrium Boltzmann equation, ci = exp(zih), will apply, as may confirmed by integration of Eq. (1.1) when Ji = 0. This may be substituted into the Poisson equation to yield the Poisson– Boltzmann equation, which can be solved exactly in linear space (Gouy–Chapman theory) and has been considered using simulation in cylindrical and spherical spaces [7,8]. The Gouy–Chapman equation in the limit of low potential is: C 0d;GC ¼ s 0 rD ð3:1Þ where C 0d is a capacitance per unit electrode area, in units F m2, or more typically lF cm2. Here we shall find a spherical solution in the low potential (Debye–Hückel) limit. For an inert monovalent binary electrolyte AX, our problem is: 1 2 r2 h þ R2e ðexpðhÞ expðhÞÞ ¼ r2 h R2e sinh h ¼ 0 ð3:2Þ subject to h = h0 at R = Re and h ? 0 as R ? 1. The Debye–Hückel approximation takes h 1 (i.e. / RT/F) which is a common situation for impedance problems but much less so for traditional voltammetric study. In this limit, sinh(h) h, thus linearising the equation. Linear problems in spherical symmetry of this type are often solved by a substitution of the form v = R h. By this method we can solve the problem (see Supporting information) and we find, in real variables: C 0d ¼ s 0 rD rD rD ¼ C 0d;GC 1 þ 1þ re re ð3:3Þ where C 0d indicates a capacitance per unit electrode area. Therefore there is an ‘excess’ double layer capacitance due to the hemispherical nature of the diffuse double layer. At low applied potentials, the deviation from classical Gouy–Chapman theory is given by the equation above. It is evident that the Gouy–Chapman theory is obeyed very closely where re rD, i.e. it is only in error for nanoscale electrodes. These results show very close agreement with those predicted by simulation in past work for h0 1. [7] 26 E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 4. Construction of a spherical Debye–Falkenhagen equation that the Laplace coordinate s corresponds roughly to write: We may consider dynamic processes under the same condition of low applied potential. The largest deviation of any concentration from bulk will be the surface excess, at equilibrium, of the ion which has the opposite charge as the electrode. This ion has a normalised surface concentration of exp(h0) according to the Boltzmann equation. So, if h0 1, we can write ci 1 + dci where dci 1 everywhere. Hence: 1 R2 pffiffiffi < e s an zi ci rh zi rh ð4:1Þ thus linearising the Nernst–Planck equation. For the case of inert binary monovalent electrolyte, if we initially assume DA = DX (but see below, Section 6) then the Nernst– Planck equation for X may be subtracted from that for A, such that: @q ¼ r ðrq þ 2rhÞ ¼ r2 q þ 2r2 h @s ð4:2Þ where q = cA cX and is a measure of charge separation. The Poisson equation, similarly, gives: 1 2 r2 h þ R2e q ¼ 0 ð4:3Þ Therefore, combining these, we achieve the Debye–Falkenhagen equation [17]: @q ¼ r2 q R2e q @s ð4:4Þ which is a linear equation describing dynamic mass transport of charged species at low potential and hence low deviation from electroneutrality. The properties of the Debye–Falkenhagen equation following a potential step were discussed in detail for linear space by Bazant et al. [15]. In the spherical case, the complexity of r2 makes direct solution more challenging. The spherical equation, however, has the estimable advantage of being entirely independent of the reference electrode position and involving no arbitrary length scales. The technique for solution is the removal of radial dependence by introducing a variable u = R q, such that the resulting equation is amenable to Laplace transformation by analogy to the linear case. The Poisson equation may be transformed similarly, using v = R h as before. 5.1. The charging transient following a potential step It can be demonstrated that the theory of linear time-invariant systems is applicable to charging of a solution with equal diffusion coefficients (see Supporting information). Defining capacitive current [18] such that: jcap ¼ icap 2 @ 2 h ¼ Re @ s@R 2pFC Dre ð5:1Þ ð5:3Þ and hence a2n s> ! ð5:4Þ R4e Substituting back into dimensional coordinates, this gives a range of accuracy of rD t > a2n tD re 1 1 qffiffiffiffiffiffiffiffiffiffiffiffiffi s 1 þ s s þ 1 þ R2 jcap ðsÞ ¼ 2h0 L1 s!s 1 s þ Rke ! 0 and L1 s!s 1 ¼ expðsÞ 1þs B ¼ 2h0 L1 s!s @ sþ 1 C qffiffiffiffiffiffiffiffiffiffiffiffiffiA 1 þ Rs2 ð5:2Þ e qffiffiffiffiffiffiffiffiffiffiffiffiffi where s is the Laplace transform coordinate, k ¼ s þ R2e , and L1 represents an inverse Laplace transform. This inversion has no exact solution, but it may be usefully approached under certain limits. qffiffiffiffiffiffiffiffiffiffiffiffiffi Let us approximate 1 þ Rs2 by its Taylor series. To order n, this e is accurate to 1% when s R2e < a1n , where a0 50 and a1 3. Noting ð5:7Þ i.e. at long time and low voltage, and with equal diffusion coefficients, an exponential decay analogous to the RC circuit, with a decay constant exactly equal to 1 in our choice of a mixed time coordinate. This approximation is accurate to 1% for t > (2500 tD (rD/re)). This inequality is satisfied for any experimentally measurable timescale, except when the electrode is nanoscale and so re ? rD. Less stringently, the second term of the Taylor series may be included: 1 1 qffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ s 1 þ Rs2 s þ 1 þ 2R2e þ . . . ð5:8Þ e This is 1% accurate for s < ð3 R2e Þ which is equivalent to t > (9 tD (rD/re)). Further terms of this Taylor series may reasonably be neglected: we are excluding the contribution of large values of s as corresponding to (vanishingly) small values of s. So: 1 1 qffiffiffiffiffiffiffiffiffiffiffiffiffi s 1 s þ 1 þ R2 1 þ 2R2 s þ 1 e ð5:9Þ e and so rearranging, we find that to this next order of approximation jcap ðsÞ 2h0 1 ð5:6Þ e R¼Re we find that, for the potential step problem: ð5:5Þ where tD is the characteristic Debye time and is defined r2D =D. For typical electrolyte concentrations and diffusion coefficients, the Debye time is at most 1 ls and more typically is tens of nanoseconds. It is clear, then, that the requirement of ‘long time’ is in fact not too stringent; since Re is of the order 1 – 103 for most real systems, the available t range is large even for a low order of approximation. qffiffiffiffiffiffiffiffiffiffiffiffiffi Considering the zeroth order Taylor series, we take 1 þ Rs2 ! 1, e so: sþ 5. Low potential charging for equal diffusion coefficients s12 , we can 2R2e 2R2e þ 1 ! exp 2R2e 2R2e þ 1 ! ! s ð5:10Þ Again we recover an exponential decay at long time, with a slightly altered decay constant; for large Re (large electrodes) this further term rapidly tends to unity and so the first order correction is unnecessary. It should be noted here that since the additional term in the decay constant also appears as a multiplying factor to the current as a whole, it can be identified with a change in the resistance, Rs, and not of the capacitance. Converting back into real units, we note that for the charging problem at low potential, to the first approximation: E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 Rs;0 ¼ C 0d ¼ RT 1 1 r 2D ¼ F 4pFDC r e 2ps 0 r e D s 0 ð5:11Þ ð5:12Þ rD Therefore, the capacitance associated with the charging process is the same as that given for differential capacitance by the Gouy– Chapman theory. This strongly suggests that this capacitance is physically meaningful, and in the limit of a large electrode and low applied potential, charging is well described by an RC circuit. To the second approximation, the capacitance is unchanged, but: rD Rs;0 Rs;1 ¼ 1 þ 2r e ð5:13Þ i.e. a slightly elevated resistance over that predicted by Gouy–Chapman theory is seen as re ? rD. At first sight our determination here is entirely contradictory with the steady state. The inclusion of a spherical term in our analysis has introduced a correction to the capacitance at steady state, but an unaltered capacitance and a correction to the resistance for the dynamic charging process. In both cases the correction is only relevant when re is a few Debye lengths or less. The resolution of this problem lies in appreciating that although the initial electroneutral condition for the dynamic solution requires that the solution charge [19], qsoln, is zero at s = 0, it does not constrain that the electrode charge, qe = 0. Indeed, since the applied potential is well defined, the enclosed charge on the electrode immediately following the potential step, before any ionic migration, is easily calculated, since this follows simply from Coulomb’s law: qe ¼ 2ps 0 re /0 ð5:14Þ and therefore q0e ¼ s 0 re /0 Since we have developed an exact result (under the low potential limit) for the Laplace space solution for the spherical system, the impedance spectrum follows directly [20] by substituting jx for s and taking ZðjxÞ ¼ h0 ðjxÞ=iðjxÞ (see Supporting information). If re rD, i.e. the electrode is not nanoscale, and x R2e : 1 1 j 2 2x ZðjxÞ ZðjxÞGC ð5:16Þ R2e ! R2e þ 1 þ jx ð5:17Þ which is similar to the first order correction noted for the transient resistance. It must be noted, however, that in this case the resistance is slightly reduced. Therefore the nanoscale correction to the complex impedance is opposite to the correction to the transient resistance, underlining that properties due to an ‘‘electrical circuit analogy’’ cease to be meaningful at the nanoscale. 6. The case of unequal diffusion coefficients When DA – DX, the above analysis may not apply. If one ion is more mobile than the other, it will react more quickly to a step in potential and so the capacitive properties of the double layer are altered. We approach the problem analytically by choosing to normalise time more generally by the arithmetic mean of the diffusion coefficients: s¼ 5.2. Impedance for low applied potential ZðjxÞ ¼ which corresponds to the ideal (zeroth order) resistance as real impedance and the Gouy–Chapman capacitance as imaginary impedance respectively, in complete agreement with the predictions of RC circuit analysis. The condition x R2e corresponds to frequencies less than the Debye frequency D=r2D , typically 1 MHz–1 GHz, which is not an experimentally achievable range with typical electrochemical impedance spectroscopy equipment, and is not a function of electrode size. Hence even when Re is small we can take x R2e ; applying this but including the nanoscale terms achieves a more complicated result which we will not report here (but see Supporting information). The result is a correction to the impedance of the form: ð5:15Þ which is the correction term noted at steady state. Therefore the correction to the steady state capacitance is associated with the ‘‘Coulomb charge’’ on the electrode. This charge density becomes increasingly significant for nanoelectrodes, as compared to the double layer charge density which is only a function of Debye length. Since no ionic motion is required to achieve this charge, it occurs effectively immediately (at s = 0 in tandem with the potential step). It is not measured by the charging transient since this measures the change of electrode charge after s = 0 and so does not alter the apparent capacitance associated with the decay constant for this process. However, this charge must still be compensated by the development of the double layer. Therefore, for a nanoelectrode, the change of charge in solution over the charging transient is not exactly opposite to the change of charge on the electrode. This marginally lengthens the time required to form the double layer and arises in the RC expression as an excess resistance. The Cd predicted at steady state does not, however, dictate the charging capacitance for a nanoelectrode and so there is no single ‘system capacitance’ as is normally understood. The ‘solution’ resistance, Rs, is a strong function of re and so is as much a property of the electrode as it is of the solution. 27 ðDA þ DX Þt Dmean t ¼ 2r e rD re rD ð6:1Þ The solution of the NPP equations for unequal diffusion coefficients is feasible in the Laplace space under the Debye–Falkenhagen approximation (see Supporting information). It arises that the effect of unequal diffusion coefficients is parameterised by a constant d = (DA DX)/(DA + DX) which for less than infinite disparity in rates of diffusion takes values jdj < 1. For the equal diffusion coefficients case above, d = 0. Unfortunately, both the analysis and the result are extremely cumbersome for unequal diffusion coefficients. Under the approximation that s2 R4e ð1 d2 Þ2 =4d2 , long time values for Rs and Cd can be derived, with the charging transient again taking an ideal exponential form. The calculation is detailed in the Supporting information, and has the result that, for a large electrode at low overpotential: Rs ¼ C 0d ¼ 1 re ps 0 ðDA þ DX Þ r2D C 0d;GC ¼ 2DA Rs;0 DA þ DX ð6:2Þ ð6:3Þ i.e. the elevated diffusion coefficients increase the conductance of the solution without affecting the capacitance, which remains equivalent to its value as inferred from the system equilibrium. Therefore RC circuit analysis is still relevant at low potential even in the case of unequal diffusion coefficients; the resistance is altered such that the rate of charging is proportional to the arithmetic mean of the diffusion coefficients of the ions involved. In effect, the component behaves as two resistances in series, one associated with each ion. A ‘‘very long time’’ solution exists (see Supporting information) which predicts that the resistance will increase so as to behave as though the component resistances from each ion are in parallel 28 E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 rather than in series. This regime is not significant until long after the charging current has decayed to zero, however, except in the case of very small electrodes, when more terms are required for an accurate expression in any case. This appears to be a mathematical curiosity of the system without any real physical relevance since it only concerns behaviour once the double layer is indistinguishable from its equilibrium state. 7. Simulation results 7.1. The effect of size The analysis above was confirmed and extended using numerical simulation techniques. The Nernst–Planck–Poisson equations were solved dynamically in a transformed space from R = Re to R ? 1, using a fully implicit finite difference method. A normalised space is used and the simulations are formulated to solve the Nernst–Planck–Poisson equations with ln ci as the variables in place of ci, as this was found to provide better convergence for dynamic simulation, by taking advantage of natural exponential relations within the system and reducing the dominance of non-linear terms. Two representative capacitive transients are shown at Figs. 2 and 3: one for re/ rD = 10 and one for re/rD = 103, which are typical nano- and microelectrode situations respectively. A very small applied potential of h0 = 0.01 is assumed, from an initially uncharged electrode. According to the first-order theory developed above, the correction from the classical resistance is 5% in the first case and 0.05% in the second. In the simulations presented we note the strong agreement of only the first-order corrected case for the nanoelectrode transient, whereas the zeroth order theory is entirely adequate for the microelectrode where the Debye length is vanishing on the electrode scale. Fig. 2 also contains a comparison of the nanoelectrode and microelectrode cases, since the latter is given accurately by the zeroth order treatment; a distinction in the decay constant of charging is evident between the two electrode sizes. Simulation confirms the prediction that for microelectrodes an RC description using Gouy–Chapman theory to describe capacitance and a classical resistance as shown above is an excellent description of the diffuse Fig. 3. Simulated capacitance transient for a step from the PZC to h0 = 0.01, with re/ rD = 1000. The zeroth order analytical treatment is shown for comparison. double layer, noting, however, that this resistance is a property of the electrode as well as of the solution. For the nanoelectrode, the curve can still be described in terms of Rs and Cd, but as demonstrated above these parameters are dependent on the experiment in question and cannot be directly associated with those inferred from impedance spectroscopy or differential capacitance, thus undermining the physical validity of the model. 7.2. The effect of large overpotential The analysis above does not apply to overpotentials similar to or greater than RT/F. The low potential range is typical for impedance studies where small overpotentials are preferred, but is not typical in chronoamperometry or cyclic voltammetry procedures where ranges of many units RT/F are explored in order to access different thermodynamic regimes of an electrolysis reaction. The linear range of the Nernst–Planck–Poisson equation set has recently been discussed for the related problem of an ion exchange membrane by Moya [21]. The Gouy–Chapman theory, which is applicable to the limit of large electrodes, gives the form of the double layer surface charge density as: Q 0DL ¼ Fig. 2. Simulated capacitance transient for a step from the PZC to h0 = 0.01, with re/ rD = 10. Zeroth and first-order analytical treatments are shown for comparison. The zeroth order transient is an accurate approximation of the case re/ rD = 1000, normalised appropriately, so a comparison of micro- and nanoelectrode transients may also be seen. q0DL h0 ¼ 4 sinh FC r D 2 ð7:1Þ which clearly is approximately linear only in the limit of small h0. For higher overpotentials the double layer charge is predicted to vary exponentially with applied potential. Since proportionally more charge has to be assembled on the double layer, more migration of material is required, and so capacitive charging takes longer. In particular, the significance of non-linear terms in the high overpotential limit implies deviation from an ideal exponential behaviour. This is indeed observed; a comparison between charging transients for h = 0.05 and h = 5 is shown for jcap vs. s and ln jcap vs. s at Figs. 4 and 5 respectively, at a microelectrode where re/rD = 1000. The significant feature is the much longer non-exponential ‘tail’ of the high potential transient, corresponding to the assembly of excess charge at the double layer. More specifically, in this case a greater accumulation of anion is required than the depletion of cation, and so the formation of the double layer requires the establishment of a significant ionic strength gradient in addition to the separation of charge. This process clearly requires more time than the ideal ‘linear’ charging of a low potential double layer in E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 Fig. 4. Simulated capacitance transient for a step from the PZC to h0 = 0.05 and to h0 = 5, with re/rD = 1000, comparing the low and high potential limits. Fig. 5. Simulated capacitance transient for a step from the PZC to h0 = 0.05 and to h0 = 5, with re/rD = 1000, comparing the low and high potential limits. In this scale, an exponential decay is a straight line. which the variation in surface excess is roughly equivalent for both ions. The extent of deviation from exponential decay (RC behaviour) may be explored by considering the ‘‘apparent’’ exponential decay of a charging transient. Since an ideal exponential has @ ln jcap/ @ s = k where k is the decay constant, we can plot the apparent k, kapp @ ln jcap/@ s as a s-dependent function. In regions where it is constant, the RC circuit is a good description, but not where it is variable. The results are shown for a series of applied h0 values, at different times s, in Figs. 6 and 7. The predicted constant k from the spherical Debye–Falkenhagen equation is also plotted for clarity. It is clear that while the RC description is excellent at all times for h0 1, it becomes slightly inaccurate at longer times for h0 1 and is wholly inaccurate for h0 > 1. We may otherwise remark that at very short times, all transients deviate from exponential behaviour for a nanoelectrode, since the asymptotic requirement that s R2 e is no longer so well obeyed. What is more, all transients show similar behaviour, exponential or otherwise, at short times – only once the surface excesses of the two ions are no longer symmetric does the rate of 29 Fig. 6. kapp @ ln jcap/@ s, for various values s in a capacitance transient and at various applied h0 from h0 = 0.01 to h0 4, with re/rD = 10. Fig. 7. kapp @ ln jcap/@ s, for various values s in a capacitance transient and at various applied h0 from h0 = 0.01 to h0 4, with re/rD = 1000. the charging process become potential-dependent. Of course, a higher applied potential induces this situation more rapidly. In the high potential range, capacitive charging is a non-linear process. Consequently low potential analysis does not apply and since the form of the transient is not exponential, RC circuit analysis is also inappropriate as the charging ‘constant’ RsCd is a strong function of time. Therefore RC circuit analysis is not a good description of the diffuse double layer for techniques in which applied potential varies by much more than RT/F, such as, for instance, conventional chronoamperometry or cyclic voltammetry. 7.3. The effect of unequal diffusion coefficients A range of numerical simulations were conducted to consider the effect of unequal diffusion coefficients on capacitive charging transients. It was suggested by the analysis above that the characteristic decay time would be reduced according to the arithmetic mean of the diffusion coefficients. Additionally, some deviation from pure exponential behaviour may be expected for unequal diffusion coefficients, particularly the observation of a reduced decay constant at very long time. 30 E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 Three normalised transients are shown at Fig. 8 on a constant time scale (s = DAt/rerD), for modest differences in diffusion coefficients; DX and DA are in the ratios 2:3, 1:1 and 3:2, respectively. From this it is clear that while the decay constant is inversely proportional to the mean diffusion coefficient, the deviations from exponential behaviour are in practice negligible. A more complete study at Fig. 9 shows the variation of kapp, as defined above, in time and with varying diffusion coefficient ratio. It is clear that for more unequal diffusion coefficients, a greater acceleration of charging at short time is balanced by a greater deceleration at long time; this is associated with the formation of the symmetric double layer (at low potential) requiring the system to ‘‘wait’’ for the slower ion to migrate. Since the diffusion coefficients do not affect the charge on an equilibrated double layer, the integral of the charging transient must be constant and hence faster charging at some time will correspond to slower charging at some other time. In general, however, the observation of kapp very close to unity in a timescale normalised to the mean of the diffusion coefficients Fig. 8. Simulated normalised capacitance transients for a step from the PZC to h0 = 0.05, with re/rD = 104 and DX/DA = 0.666. . ., 1, and 1.5. In this scale, an exponential decay is a straight line. confirms the dominance of the long (but not ‘‘very long’’) time regime cited above, in which the two ions behave as two component resistances in series. This indicates that the effect of the charge interaction between the ions on their migration is to rapidly balance the rates of migration to a mean transport rate. The rate of double layer formation is hence the average of the contributions of each individual ion. 8. Conclusions In the arguments and results presented above, we have developed a self-consistent approach for the analysis of diffuse double layer charging that avoids the traditional use of arbitrary length scales. Simulations and mathematical analysis were used in a complementary manner and with strong mutual agreement. The exponential charging transient and impedance spectrum predicted by RC circuit analysis and those based on a complete treatment using the Nernst–Planck–Poisson equations were compared. RC circuit analysis was found to agree with the more complete model only subject to certain conditions. These conditions are: an electrode larger than nanoscale; an applied potential not larger than RT/F (25 mV); and moderately similar diffusion coefficients. Where these conditions are violated, either ideal RC behaviour breaks down altogether, as is the case for the non-exponential transients observed at high applied potential, or the meanings of the ‘constants’ Rs and Cd change depending on the experiment under consideration. In particular we highlight that the condition of low applied potential, common in electrochemical impedance spectroscopy, is strictly necessary for equivalent circuit analysis to be meaningful to describe a diffuse double layer. Common experimental techniques in Faradaic analysis, such as chronoamperometry or cyclic voltammetry, regularly probe much larger potential ranges, and so the applicability of simple models will break down for the double layer in these situations. Consequently, values of Rs and Cd inferred from impedance spectroscopy must be treated with caution, since they are not representative universal properties of the system. They are not applicable to descriptions of the double layer under potential conditions where a Faradaic reaction is strongly driven by high overpotential. Equally, they are uncertain for nanoelectrodes due to their variation depending on the experiment in question. Acknowledgement E.J.F.D. thanks St. John’s College, Oxford, for support via a graduate scholarship. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.jelechem.2011.02.016. References Fig. 9. The effect of the diffusion coefficient ratio DX/ DA on kapp @ ln jcap/@ s, at different times during a capacitance transient with h0 = 0.05 and re/rD = 104. [1] O. Stern, Z. Elektrochem. 30 (1924) 508–516. [2] D.C. Grahame, Chem. Rev. 41 (1947) 441–501. [3] Note that while the capacitance C describes the ratio of charge on the capacitor to voltage across it (Ecap(t)), the applied potential E(t) is the potential difference associated with the voltage source. These are equivalent only when the current is zero. In the discussion hereafter, capacitance will be understood to refer to the circuit (and the physical system it represents) as a whole, and so E refers to the applied voltage. As shall be demonstrated, this circuit capacitance equals C at zero current. [4] W. Scheider, J. Phys. Chem. 79 (1975) 127–136. [5] R.D. Armstrong, R.A. Burnham, J. Electroanal. Chem. 72 (1976) 257–266. [6] T.C. Halsey, M. Leibig, Ann. Phys. (San Diego) 219 (1992) 109–147. [7] E.J.F. Dickinson, R.G. Compton, J. Phys. Chem. C 113 (2009) 17585–17589. E.J.F. Dickinson, R.G. Compton / Journal of Electroanalytical Chemistry 655 (2011) 23–31 [8] M.C. Henstridge, E.J.F. Dickinson, R.G. Compton, Chem. Phys. Lett. 485 (2010) 167–170. [9] R.J. Smith, R.C. Dorf, Circuits, Devices and Systems, fifth ed., John Wiley & Sons, New York, 1992. [10] E. Barsoukov, J.R. Macdonald (Eds.), Impedance Spectroscopy: Theory, Experiment, and Applications, John Wiley & Sons, New York, 2005. [11] A.J. Bard, L.R. Faulkner, Electrochemical Methods: Fundamentals and Applications, 2nd ed., John Wiley & Sons, New York, 2001. [12] L.G. Gouy, C.R. Hebd, Séances Acad. Sci. 149 (1909) 654–657. [13] L.G. Gouy, J. Phys. Theor. Appl. 9 (1910) 457–467. [14] D.L. Chapman, Philos. Mag. 25 (1913) 475–481. [15] M.Z. Bazant, K. Thornton, A. Ajdari, Phys. Rev. E 70 (2004) 021506. [16] F. Beunis, F. Strubbe, M. Marescaux, J. Beeckman, K. Neyts, A.R.M. Verschueren, Phys. Rev. E 78 (2008) 011502. 31 [17] P. Debye, H. Falkenhagen, Phys. Z. 29 (1928) 121–132. [18] The use of j for dimensionless current is standard in the voltammetric literature and is used here to emphasise the correlation between Faradaic and capacitive currents, and their relative magnitudes. We are conscious of the possible confusion with the imaginary unit j in other parts of the manuscript and its Supporting Information and therefore consistently ‘‘cap’’ as a subscript when current is being discussed. [19] We use the notation q with appropriate subscripts here for the charges on the electrode and in the solution in the double layer system, in order to emphasise the analogy with the charge q on the capacitor in the discussion of the RC circuit in Section 1.2. [20] A. Bezegh, J. Janata, J. Electroanal. Chem. 215 (1986) 139–149. [21] A.A. Moya, Electrochim. Acta 55 (2010) 2087–2092.