Impedance Data Analysis: Equivalent Circuits, DRT, and Physical Interpretation

Experimental Methods

Electrochemistry

EIS

Data Analysis

An advanced guide to analysing electrochemical impedance data using equivalent circuits, relaxation times, regularization, and physically defensible interpretation.

Author

Mostafa Elkady

Published

November 3, 2026

DATA ANALYSIS · ADVANCED GUIDE 18

Electrochemical impedance spectroscopy (EIS) produces a frequency-dependent response that can contain overlapping contributions from charge transfer, ionic transport, interfaces, contacts, diffusion and electrode reactions. The central challenge is not obtaining a visually acceptable fit; it is determining which features are resolved by the data and which physical conclusions are justified.

Two complementary analysis approaches are widely used:

Equivalent-circuit analysis, in which the spectrum is represented by a network of ideal or non-ideal electrical elements.
Distribution of relaxation times (DRT), in which the response is represented as a continuous or discretized distribution of characteristic relaxation times.

Neither method automatically identifies a unique microscopic mechanism. Reliable interpretation requires data validation, model selection, uncertainty assessment and consistency with independent physical evidence.

1. Start with data quality, not fitting

Before analysis, preserve the original frequency, real impedance, imaginary impedance, amplitude, bias and measurement-condition data. Do not overwrite raw measurements with smoothed, truncated or corrected values.

Check the following:

Frequency ordering and units: verify that frequency is in Hz and that angular frequency is calculated as \(\omega=2\pi f\).
Impedance convention: determine whether the instrument exports \(Z''\) or \(-Z''\) and use one convention consistently.
Outliers: inspect isolated points, especially at the highest and lowest frequencies.
Inductive artefacts: inspect high-frequency loops that may originate from cables, leads or cell geometry.
Low-frequency drift: confirm that the system remained stationary during long measurements.
Linearity: where needed, repeat spectra at different AC amplitudes and compare the response.
Reproducibility: repeat selected spectra and verify that the specimen and operating conditions are stable.

A spectrum measured while temperature, gas composition, state of charge or electrode chemistry is changing is not a stationary linear-response spectrum. A sophisticated fitting method cannot correct an invalid experiment.

2. Equivalent-circuit analysis

An equivalent circuit describes the measured impedance using a network of electrical elements. A common example is a series resistance followed by a parallel resistor–capacitor element:

\[ Z(\omega)=R_s+\frac{R}{1+j\omega RC} \]

Here, \(R_s\) is a series resistance, \(R\) is the resistance of the parallel process, and \(C\) is its capacitance. The characteristic relaxation time is

\[ \tau=RC \]

and the frequency at the maximum of the ideal semicircle is

\[ f_{\max}=\frac{1}{2\pi RC} \]

This relationship applies to an ideal Debye-type \(R\parallel C\) response. It should not be used uncritically for a depressed arc, a CPE-based model or overlapping processes.

Constant-phase elements

A constant-phase element (CPE) is often used to represent non-ideal capacitive behaviour:

\[ Z_{\mathrm{CPE}}= \frac{1}{Q(j\omega)^n} \]

where \(0<n\leq1\) for the common capacitive case. At \(n=1\), the CPE is an ideal capacitor. When \(n<1\), the response is distributed or non-ideal.

The CPE parameter \(Q\) is not generally a capacitance. Its units depend on the exponent \(n\). Several approximate conversions from CPE parameters to effective capacitance exist, but they depend on the circuit topology and the assumptions used. State the conversion method rather than reporting \(Q\) as \(C\).

Choosing a circuit

A circuit should be selected using several forms of evidence:

residuals should be small and lack systematic frequency-dependent structure;
fitted parameters should be stable against reasonable changes in initial guesses;
confidence intervals and parameter correlations should be acceptable;
each element should have a plausible interpretation for the specimen and measurement geometry;
the model should remain useful across related temperatures, atmospheres or bias conditions.

A lower residual alone does not justify adding another resistor, capacitor or CPE. Additional elements can fit noise and create highly correlated parameters.

3. What does a relaxation time mean?

A relaxation time characterizes the timescale over which a process responds to a perturbation. For an ideal parallel \(R-C\) element,

\[ \tau=RC \]

and the corresponding characteristic frequency is \(f_c=(2\pi\tau)^{-1}\).

In real electrochemical materials, a process may have a distribution of relaxation times rather than one sharply defined value. This can arise from microstructural heterogeneity, distributed reaction rates, nonuniform current paths, diffusion, or a range of local environments.

A peak in a DRT spectrum therefore indicates a contribution concentrated around a range of timescales. It does not, by itself, prove that the process is a grain boundary, charge-transfer reaction, surface exchange step or diffusion mechanism.

4. Distribution of relaxation times: the basic formulation

For a common representation of a spectrum containing resistive relaxation processes, the impedance can be written as

\[ Z(\omega)=R_{\infty}+ \int_{-\infty}^{+\infty} \frac{\gamma(\ln\tau)} {1+j\omega\tau}\,d\ln\tau \]

where:

\(R_{\infty}\) is the high-frequency series or ohmic contribution within the adopted model;
\(\tau\) is the relaxation time;
\(\gamma(\ln\tau)\) is the distribution of resistive contributions per logarithmic interval of relaxation time;
\(j\) is the imaginary unit.

The exact form depends on the DRT convention and kernel used by the implementation. Some formulations include additional terms for inductance, capacitive contributions or boundary behaviour. Always state the convention and software method.

The DRT transforms a spectrum from frequency space into relaxation-time space. Instead of requiring the analyst to specify a fixed number of arcs, it estimates how much resistive response is associated with each relaxation-time interval.

Why use logarithmic relaxation time?

Electrochemical processes may span many orders of magnitude in timescale. Defining

\[ u=\ln\tau \]

allows the distribution to be evaluated over a logarithmic grid. A peak width then describes a range of timescales rather than a uniform interval in seconds.

The area associated with a peak is often used to estimate its resistance contribution, provided the normalization and integration convention are clearly stated. For the formulation above, the contribution over a selected interval is

\[ R_{\mathrm{peak}}= \int_{u_1}^{u_2}\gamma(u)\,du \]

Peak area is generally more robust to moderate changes in peak width than peak height, but it still depends on regularization, grid limits, overlap and the selected baseline.

5. DRT as an inverse problem

Measured EIS data provide a finite set of complex impedance values. Recovering a continuous distribution from these measurements is an ill-posed inverse problem: many distributions can reproduce the data within experimental uncertainty.

After discretization, a simplified problem can be expressed as

\[ \mathbf{Z}\approx\mathbf{A}\boldsymbol{\gamma} \]

where \(\mathbf{Z}\) contains measured impedance data, \(\mathbf{A}\) is the discretized kernel, and \(\boldsymbol{\gamma}\) contains the unknown DRT values.

A common regularized estimate minimizes an objective of the form

\[ \min_{\boldsymbol{\gamma}} \left[ \left\|\mathbf{W} (\mathbf{A}\boldsymbol{\gamma}-\mathbf{Z})\right\|_2^2 +\lambda^2 \left\|\mathbf{L}\boldsymbol{\gamma}\right\|_2^2 \right] \]

where:

\(\mathbf{W}\) weights data according to the chosen error model;
\(\mathbf{L}\) is a regularization operator, often related to smoothness;
\(\lambda\) controls the balance between fitting the data and stabilizing the solution.

A non-negativity constraint may also be imposed in implementations that assume a non-negative distribution of resistive processes. This constraint is a modelling assumption, not a universal law for every possible electrochemical response.

The regularization trade-off

If regularization is too weak, the DRT may contain narrow oscillations and artificial peaks that fit noise. If it is too strong, nearby processes may merge and genuine features may be suppressed.

The regularization parameter should not be selected solely because the resulting DRT looks smooth. Consider:

the data residual and its structure;
measurement uncertainty and noise level;
stability of peak positions and areas;
sensitivity to the relaxation-time grid;
sensitivity to the regularization parameter;
agreement between reconstructed and measured impedance.

Methods for choosing regularization include cross-validation, evidence-based approaches, residual analysis and L-curve inspection. No single selection method is universally optimal for every dataset.

6. Practical DRT workflow

A defensible workflow can be organized as follows.

Step 1 — Validate the spectrum

Inspect Nyquist and Bode representations, identify outliers and possible inductive contributions, and check whether the measurement is stationary and sufficiently linear.

Step 2 — Define the analysis window

Select a relaxation-time interval that is supported by the measured frequency range. Do not interpret peaks near the limits as fully resolved without additional evidence. A process slower than the lowest measured frequency or faster than the highest measured frequency may be truncated.

Step 3 — Choose the kernel and error model

State whether the analysis uses real and imaginary components jointly, how data are weighted, and whether the method includes inductive or other additional terms. Weighting by relative error can behave poorly when a component approaches zero; explain the selected weighting scheme.

Step 4 — Select regularization

Evaluate a reasonable range of regularization strengths. Compare residuals, reconstructed spectra and peak stability rather than choosing a single setting by visual appearance.

Step 5 — Reconstruct the impedance

Use the estimated DRT to calculate the impedance predicted by the forward model. Overlay reconstructed and measured real and imaginary components, and inspect residuals versus frequency.

Step 6 — Test robustness

Vary the grid density, relaxation-time bounds, regularization strength and reasonable preprocessing choices. A feature that appears only under one narrow parameter choice should be treated cautiously.

Step 7 — Interpret with independent evidence

Compare peak evolution with temperature, atmosphere, gas partial pressure, electrode composition, specimen thickness, microstructure and independently measured transport properties.

7. Linking DRT peaks to physical processes

In ceramic electrochemical systems, candidate contributions may include:

bulk ionic transport;
grain-boundary response;
electrode charge transfer;
surface exchange and gas conversion;
contact and current-collection effects;
diffusion or gas-transport limitations.

These are hypotheses to test, not automatic labels for particular relaxation-time ranges. Characteristic times overlap across materials and change with temperature, geometry, electrode morphology and measurement configuration.

Use capacitance as supporting evidence

For a resolved ideal \(R\parallel C\) process, an effective capacitance can be estimated from

\[ C=\frac{\tau}{R} \]

where \(R\) is the resistance associated with that same process. This estimate is not directly transferable to a broad DRT peak without defining how its resistance and characteristic time are extracted.

Capacitance magnitude can help distinguish candidate processes because bulk, grain-boundary and electrode responses often have different effective capacitance scales. However, geometry, dielectric response, porosity and distributed interfaces influence the result. Capacitance should be treated as supporting evidence rather than a unique fingerprint.

Use temperature dependence carefully

If a characteristic relaxation frequency follows thermally activated behaviour, an Arrhenius-type relationship may be examined:

\[ f_{\max}=f_0 \exp\left(-\frac{E_a}{k_BT}\right) \]

Taking logarithms,

\[ \ln f_{\max}=\ln f_0- \frac{E_a}{k_B}\frac{1}{T} \]

where \(E_a\) is an apparent activation energy and \(k_B\) is Boltzmann’s constant.

The slope of \(\ln f_{\max}\) versus \(1/T\) is \(-E_a/k_B\). The activation energy is meaningful only if the same process is tracked across temperature and the peak position is reliably resolved. Overlapping peaks, changing electrode conditions or a change in rate-limiting mechanism can invalidate a simple fit.

8. Equivalent circuits and DRT are complementary

Question	Equivalent-circuit analysis	DRT analysis
How many model elements are needed?	Must be specified and tested	Peaks emerge from regularized inversion
Can overlapping responses be separated?	Sometimes, if the model and data support it	May reveal overlapping timescales, subject to resolution
Are parameters directly physical?	Only with justified assignments	Peak positions and areas still require interpretation
Main risk	Overfitting and non-unique circuits	Regularization artefacts and non-unique inversion
Useful output	Element parameters and model comparison	Distribution of resistive response over timescale

A strong analysis may use DRT to identify the number and approximate timescales of contributions, then use physically justified equivalent circuits to quantify selected processes. The two analyses should not be forced to agree if their assumptions differ.

9. Common DRT artefacts and failure modes

Artificial peak splitting: weak regularization or noise can create multiple peaks from one broad contribution.
Peak merging: excessive regularization can combine nearby processes.
Boundary peaks: a peak near the edge of the chosen relaxation-time interval may be truncated or poorly constrained.
Grid dependence: a coarse grid can miss narrow features; an excessively fine grid does not create additional information.
Inductive contamination: unmodelled high-frequency inductance can distort the recovered distribution.
Incorrect weighting: a few high-magnitude points may dominate the objective, or low-magnitude points may be over-weighted.
False physical labels: assigning a peak to a microstructural feature from its position alone is not justified.
Ignoring uncertainty: a smooth curve is not evidence that the inverse problem has a unique solution.

10. Reporting a reproducible DRT analysis

A publication-quality report should state:

Instrument, frequency range, AC amplitude, DC bias and measurement conditions.
Data convention, preprocessing and treatment of inductive contributions.
DRT formulation and kernel, including any additional terms.
Relaxation-time grid and bounds.
Regularization operator and selection method for its strength.
Weighting or noise model and any non-negativity constraints.
Software name, version and relevant algorithm settings.
Reconstructed-spectrum comparison and residual plots.
Sensitivity tests for regularization and grid choices.
How peak positions and areas were extracted, including uncertainty.
Independent evidence used to support physical assignments.

Provide the measured spectrum or sufficient numerical data whenever possible. A DRT plot without analysis settings and reconstruction quality is difficult to reproduce.

11. Recommended interpretation strategy

For a polycrystalline oxide or electrochemical cell, a practical sequence is:

Establish the high-frequency series contribution and check for wiring artefacts.
Fit a simple physically plausible circuit as a baseline.
Calculate a DRT using a documented regularization procedure.
Compare the number, positions and areas of stable DRT features with the circuit response.
Repeat measurements at different temperatures or atmospheres while holding other conditions constant.
Use geometry, capacitance estimates, electrode controls and microstructural characterization to test candidate assignments.
Report ambiguous or unresolved contributions explicitly.

For example, a peak that changes strongly with oxygen partial pressure may be consistent with an oxygen-electrode process, but that observation alone does not identify a specific elementary reaction. A peak that scales with specimen thickness may support a bulk contribution, but contact and geometry effects must also be considered.