Diffusion reconciliation
Three methods, one diffusion coefficient.
CV, GITT, and EIS each produce a diffusion coefficient, through different physics, assumptions, and length scales. Putting the three estimates side by side, with their assumptions explicit, is what turns ‘a D value’ into a number you can defend.
What it measures
The solid-state diffusion coefficient D controls how fast lithium moves through the active material, and three standard methods estimate it from different experiments:
- CV, Randles-Ševčík. Peak current scales with the square root of scan rate: iₚ = 2.69×10⁵ · n³ᐟ² · A · D¹ᐟ² · C · v¹ᐟ². The slope of iₚ versus √v yields D, hanging entirely on the electroactive area A and concentration C you assume.
- GITT, the relaxation transient. A current pulse, then open-circuit relaxation; the voltage against √t during the transient carries D at that state of charge, through the electrode’s molar volume and active surface area.
- EIS, the Warburg element. The 45° diffusion tail of the impedance spectrum, fit with a bounded Warburg, gives a time constant τ = L²/D, converting a diffusion length L into D.
The methods probe different length scales and average differently over the porous electrode, so disagreement of one to two orders of magnitude between them is normal, and informative, because the pattern of disagreement is itself diagnostic.
How to read the output
Plot the three estimates on a log axis at the same state of charge and temperature. Clustered within an order of magnitude: the value is robust and the assumptions are mutually consistent. CV far above GITT/EIS: the electroactive-area assumption in Randles-Ševčík is suspect, geometric area overstates D badly in porous electrodes. All three trending together against SOC, with a dip near phase transitions, is the expected physics; one method diverging alone is an assumption problem, not a materials discovery.
A real use case
A silicon-graphite anode misses its fast-charge spec and the team needs to know whether the limitation is particle-level transport or electrode-level design. GITT and EIS agree on a low D near full lithiation; the CV estimate, computed with geometric area, sits two orders higher until the area assumption is corrected, at which point all three line up. Conclusion: transport inside the particles genuinely limits at high SOC, so the fix is particle size and gradation, not calendering density. The reconciliation is what made that call defensible against the supplier’s counter-claim that the electrode design was at fault.
Common mistakes
- Using geometric electrode area in Randles-Ševčík. The true electroactive area of a porous electrode is far larger, and D scales with 1/A², this single assumption moves the answer by orders of magnitude.
- Quoting one D for the material. D varies strongly with state of charge; a single number without its SOC is incomplete.
- Comparing estimates taken at different SOC or temperatures and calling the spread “method disagreement.”
- Fitting the GITT transient through the IR drop instead of starting after it.
- Extracting a Warburg D from a spectrum that fails Kramers-Kronig validation, bad data fits circuits happily and lies.
Three estimates, every assumption on the table
Niobia computes D by each route the data supports: Randles-Ševčík slopes from CV scan-rate studies, relaxation-transient analysis from GITT pulse data, and bounded Warburg time constants from EIS, refusing to fit spectra whose Kramers-Kronig residuals exceed 2%. It will not produce a diffusion coefficient without the required metadata, electrode area, sample thickness, molar volume, temperature, electron count, it asks for them rather than assuming silently. The output puts the estimates side by side with each method’s assumptions stated, and tracks them against state of charge, temperature, or cycle number across multi-spectrum datasets, so the team chooses a number knowing exactly what it rests on.
Frequently asked
Why do CV, GITT, and EIS give different diffusion coefficients?
They measure different things under different assumptions: CV depends on an assumed electroactive area, GITT on molar volume and surface area through the relaxation transient, EIS on an assumed diffusion length. Spreads of one to two orders of magnitude are normal; the pattern of disagreement usually identifies which assumption is off.
Which method should I trust most?
None universally. GITT is closest to equilibrium and resolves SOC dependence well; EIS is fast and nondestructive but assumption-heavy; CV is quick but most sensitive to the area assumption. Agreement after honest assumptions is the trustworthy outcome.
What metadata is required before D can be computed at all?
At minimum: electrode geometric dimensions, active material loading or molar volume, temperature, and the number of electrons transferred. Without them the methods return arbitrary numbers, which is why Niobia asks instead of assuming.