Randles Circuit - Battery Design

If you’ve ever fitted an equivalent circuit to an impedance spectrum, there’s a good chance the first thing you reached for was some version of the Randles circuit. It’s the starting point almost everyone uses to turn an EIS Nyquist plot into numbers that mean something physically: a bulk resistance, a reaction rate, a capacitance, sometimes a diffusion term. It isn’t the most accurate model of what’s actually happening inside a real porous electrode, and it was never built with lithium-ion cells in mind at all. But it’s simple enough to fit with a handful of parameters, and those parameters map cleanly enough onto real physical processes that it has stuck around for nearly 80 years as the circuit people reach for first.

Quick facts

Origin: J.E.B. Randles, 1947, developed while studying the kinetics of fast electrode reactions at a mercury drop electrode
Core topology: solution resistance (R_s) in series with a parallel combination of charge-transfer resistance (R_ct) and double-layer capacitance (C_dl)
Common extension: a Warburg element (Z_w) added in series with R_ct to capture diffusion-limited transport
Nyquist signature: a semicircle (kinetics-controlled, high-to-mid frequency) running into a straight line at roughly 45° (diffusion-controlled, low frequency)
Where it shows up: corrosion studies, biosensors, fuel cells, supercapacitors, and as the starting template for lithium-ion cell EIS fitting

Standard Randles equivalent circuit: R_s in series with the parallel combination of C_dl and (R_ct in series with Z_w). Drop the Z_w branch and you have the basic three-element Randles circuit; add it back in and you have the diffusion-extended version.

What the Circuit Actually Represents

An electrochemical interface (an electrode sitting in an electrolyte) doesn’t behave like any single textbook component. Current has to get through the bulk electrolyte to reach the electrode surface, then it has to actually cross the interface via a chemical reaction, and depending on the system there’s also a charging effect where ions pile up at the surface without reacting at all. The Randles circuit is an attempt to represent each of those distinct physical processes with a separate, familiar electrical element, wired together in a topology that mimics how they actually combine.

The three original elements are:

Solution resistance, R_s (sometimes written R_u or R_Ohm). This is the resistance of the ionic path between the working and reference electrode: the electrolyte itself, plus any separator, plus the various ohmic contacts in the measurement path. It’s a genuine resistor in the circuit-theory sense: current through it doesn’t depend on frequency, so it sits alone, in series, ahead of everything else.
Charge-transfer resistance, R_ct (sometimes R_p for polarization resistance). This represents the kinetic barrier to the actual electrochemical reaction: how easily electrons cross the interface and turn reactant into product (or vice versa). It’s governed by Butler-Volmer kinetics, and at small signal amplitude it reduces to a resistance that’s inversely proportional to the exchange current density of the reaction. A sluggish reaction gives you a large R_ct, a fast one gives you a small one.
Double-layer capacitance, C_dl: the electrode surface and the adjacent electrolyte form a charge-separated layer (the classic Helmholtz/Gouy-Chapman-Stern picture), and that layer stores charge the same way any capacitor does, without any chemical reaction taking place. Because charging current and faradaic (reaction) current both flow across the same physical interface at the same time, C_dl sits in parallel with R_ct, not in series with it. The two paths share current in proportion to how easy each one is at a given frequency.

Put those three together (R_s in series with R_ct and C_dl in parallel) and you have the basic Randles circuit. Randles introduced it in 1947 in a paper on the kinetics of fast electrode reactions, working with a dropping mercury electrode, well before anyone was thinking about lithium-ion cells or even before EIS as a routine lab technique existed in its current form [1][6]. The fact that it’s still the first circuit reached for in a modern EIS fitting package says something about how well the underlying physical reasoning holds up, even outside the system it was designed for.

Reading the Impedance Spectrum

Writing the circuit out as a complex impedance is straightforward once you treat R_ct and C_dl as a parallel combination and add R_s in series: $Z(\omega) = R_s + \frac{R_{ct}}{1 + j\omega R_{ct} C_{dl}}$

Splitting that into real and imaginary parts gives:

$\text{Re}(Z) = R_s + \frac{R_{ct}}{1 + \omega^2 R_{ct}^2 C_{dl}^2}$ $-\text{Im}(Z) = \frac{\omega R_{ct}^2 C_{dl}}{1 + \omega^2 R_{ct}^2 C_{dl}^2}$

Plot $-\text{Im}(Z)$ −Im(Z) against $\text{Re}(Z)$ Re(Z) across a frequency sweep and this traces out a semicircle. At very high frequency, the capacitor branch is effectively a short circuit, so all the current takes that path and the impedance collapses to just R_s: that’s the point where the semicircle touches the real axis on the left. At very low frequency, the capacitor branch is effectively open, so all the current is forced through R_ct, and the impedance settles at R_s + R_ct: the point where the semicircle closes back onto the real axis on the right. The diameter of the semicircle is R_ct itself, and its peak (maximum $-\text{Im}(Z)$ ) occurs at the characteristic angular frequency

$\omega_{max} = \frac{1}{R_{ct} C_{dl}}$

which is often reported directly as a time constant, $\tau = R_{ct} C_{dl}$

The Randles-with-Warburg signature: a kinetics-controlled semicircle (peak at $\omega = 1/(R_{ct}C_{dl})$ ) running straight into a diffusion-controlled 45° line, with no clean break between the two. This is the same circuit as the schematic above, plotted as impedance rather than topology.

That last relationship is worth sitting with, because it’s the reason a single Nyquist semicircle can’t tell you R_ct and C_dl independently. It gives you their product, and from the diameter, R_ct on its own; C_dl then falls out once you know Rct and the peak frequency. It’s also why people frequently report the characteristic frequency of a semicircle as a diagnostic on its own, even before doing a full circuit fit, since a shift in that frequency between two spectra is a fast way to tell that either the kinetics or the interfacial capacitance changed, without committing to which one yet.

Adding Diffusion: The Warburg Element

The plain Rs-(Rct‖Cdl) circuit assumes the reaction has all the reactant it needs, right at the interface, at every frequency. That’s a reasonable assumption when the measurement is fast enough that diffusion never becomes the bottleneck, but at low enough frequency, real systems run into the same wall every diffusion-limited process runs into: the reactant has to physically travel to the electrode, and that travel takes a finite, frequency-dependent amount of time. Randles’ full treatment already accounted for this by adding what’s now called the Warburg element in series with Rct, inside the same parallel branch as before [1].

For semi-infinite linear diffusion (a diffusion layer thick enough that its far boundary never matters within the measurement) the Warburg impedance takes the form:

$Z_w(\omega) = \sigma\, \omega^{-1/2} – j\sigma\, \omega^{-1/2} = \sigma(1-j)\,\omega^{-1/2}$

where $\sigma$ σ is the Warburg coefficient, a term that bundles together the diffusion coefficients and bulk concentrations of the reacting species. What makes this element distinctive on a Nyquist plot is that its real and imaginary parts are always equal in magnitude, which means it plots as a straight line at exactly 45° to the real axis, rather than curving into a circle. So a Randles circuit with a Warburg term produces the classic two-stage signature: a semicircle at high-to-mid frequency, where charge-transfer kinetics dominate, running into a 45° line at low frequency, where diffusion takes over.

Real cells are rarely semi-infinite, though: a graphite particle, a thin coating, or an electrolyte-filled separator all have a finite diffusion length, and the boundary at the far end matters once the AC period gets long enough to “see” it. That’s handled with a finite-length Warburg element instead of the plain one above, and the boundary condition at the far side changes the low-frequency behaviour. A reflective (blocking) boundary, the usual case for lithium diffusing into a finite particle with nowhere else to go, replaces the $\omega^{-1/2}$ term with a hyperbolic tangent function and turns the low-frequency tail into another capacitive arc rather than a straight line, while a transmissive boundary uses a hyperbolic cotangent and behaves more like an added resistance at low frequency [4]. Recognising which finite-Warburg case fits a given spectrum is one of the more useful diagnostic steps in EIS analysis, because the two boundary conditions correspond to physically different transport geometries.

Why the Capacitor Rarely Stays a Capacitor

An ideal capacitor assumes a perfectly smooth, uniform surface. Real electrodes are almost never that: they’re rough, porous, and often chemically non-uniform across the surface, whether that’s a corroding metal, a catalyst layer, or a lithium-ion electrode coating full of active particles, binder, and conductive additive. The practical effect is that the phase angle of the capacitive response doesn’t sit at the ideal 90° a pure capacitor gives you; it comes in somewhat below that, and the shortfall tends to be reasonably consistent across frequency rather than looking like some other discrete component.

The standard fix is to swap C_dl for a constant phase element (CPE), defined as:

$Z_{CPE} = \frac{1}{Q\,(j\omega)^n}$

with $0 < n \le 1$ . When $n = 1$ , this is just an ideal capacitor with $Q$ equal to capacitance; as $n$ n drops below 1, the element represents a distribution of local time constants across the electrode surface rather than one single value, and $Q$ stops having simple capacitance units. The physical causes proposed for this are the ones you’d expect: surface roughness and fractal geometry, uneven distribution of R_ct and C_dl across the interface, and localized chemical heterogeneities from adsorption or impurities [5]. It’s worth being honest about the trade-off here: the CPE is what makes a Randles-type fit match a rough or porous electrode’s data well, but it’s also a curve-fitting convenience with a somewhat fuzzy physical basis, and a low n value should be read as “this interface is heterogeneous in some way,” not as a precise statement about which mechanism is responsible.

The Randles Circuit as a Starting Point for Lithium-Ion Cells

None of this was built with lithium-ion cells in mind, but the same reasoning about series resistance, a kinetic parallel branch, and a diffusion tail carries over reasonably well, which is why the Randles circuit, usually modified, is the default first fit for a lithium-ion EIS spectrum too.

In a Li-ion cell, the mapping typically goes something like this: the high-frequency real-axis intercept is read as the bulk/ohmic resistance, that is, electrolyte, separator, current collectors, and contact resistances all lumped together, the same role Rs plays in the original circuit. Moving to lower frequency, one or more semicircles appear, and these are usually attributed to distinct interfacial processes rather than a single charge-transfer step: commonly a smaller, higher-frequency arc associated with the solid-electrolyte interphase (SEI) film that forms on the graphite anode, and a second, larger arc associated with the actual charge-transfer reaction and its double-layer capacitance [3]. At the lowest frequencies, the spectrum turns toward a diffusion-controlled tail, reflecting solid-state lithium diffusion inside the active material particles as well as electrolyte-phase diffusion through the porous electrode structure. This is the domain where a finite-length Warburg element, rather than the simple semi-infinite one, tends to fit better given how confined those diffusion paths actually are [3][4].

A schematic extended-Randles spectrum for a Li-ion cell: Rohmic at the high-frequency intercept, a small arc for contact/interfacial resistance, a larger arc for charge-transfer resistance, and a Warburg tail at low frequency. Real cell data is rarely this tidy (the two arcs often overlap partially), but the sequence and the underlying circuit logic hold.

Because a single Rct‖Cdl pair can’t represent two separate interfacial processes at once, the model most people actually use on a real cell is an extended Randles circuit: an additional R‖C (or R‖CPE) branch added in series, giving something like R_s + (R_SEI‖CPE_SEI) + (R_ct‖CPE_dl) + Z_w. Some treatments add a third branch for other interfacial layers, and there’s ongoing work on whether extra branches genuinely separate distinct physical processes or just soak up curve-fitting degrees of freedom, particularly at elevated temperature or in cells that have already aged [3].

Note: In circuit shorthand, R‖C means a resistor and capacitor connected in parallel with each other, as opposed to R+C which would mean in series.

Where the Model Breaks Down

It’s worth being upfront about the limits, since the circuit’s popularity can make it easy to over-trust a fit that happens to look clean:

Non-uniqueness. More than one circuit topology, and more than one parameter set within the same topology, can produce nearly identical fits to the same spectrum. A low residual doesn’t by itself prove the circuit’s physical interpretation is correct; it only proves the maths matches the data.
Lumped, not distributed. A real porous electrode has charge-transfer resistance and double-layer capacitance distributed continuously through its thickness and pore structure, interacting with distributed ionic and electronic resistance along the way. The Randles circuit, even the extended version, replaces that distributed reality with a handful of lumped elements. Transmission-line models handle porous electrodes more faithfully, at the cost of more parameters and a less intuitive fit [3].
Small-signal and near-equilibrium only. EIS itself assumes a small enough perturbation that the system’s response is linear, and that assumption, along with the values of Rct and Cdl you extract, only holds near whatever open-circuit condition the measurement was taken at. Fit the same cell at a different state of charge, state of health, or temperature and expect different parameter values, not a fixed set of “true” component values for the cell.
CPE overuse. It’s tempting to reach for a CPE any time a fit doesn’t quite close, since it almost always improves the residual. That’s exactly the failure mode to watch for: a CPE should represent a genuine, physically justifiable non-ideality in the interface, not a way to paper over a topology that’s fundamentally wrong for the data.

None of this makes the Randles circuit the wrong tool. It makes it what it’s always been: a good first approximation, with a clear physical story attached to each element, that needs to be checked against the data (and against a bit of scepticism) rather than trusted on the strength of a low fit residual alone.