In early 2019, a preprint appeared on a popular chemistry server, promising a lithium metal anode with a Coulombic efficiency of 99.9% over hundreds of cycles. The headline figure was stunning; commercial lithium-ion anodes typically hover around 99% or lower, and the gap between 99% and 99.9% represents a tenfold reduction in capacity loss per cycle. The paper's central evidence was a single Nyquist plot, an impedance spectrum showing a tiny semicircle that the authors attributed to a minuscule charge-transfer resistance. The scientific community reacted with a flurry of emails on the battery-science mailing list, and several prominent researchers tweeted the figure with exclamation marks. Within a month, a well-known professor at a US university had incorporated the result into a seminar slide, asking his students to consider its implications. But within a year, four independent groups had re-plotted the raw data from that preprint, and each found inconsistencies that fractured the promise. This is the story of how one impedance spectrum became a cautionary tale about the fragility of battery characterization.

A 2019 Promise Built on a Single Impedance Spectrum

The preprint, posted on February 14, 2019, described a novel electrolyte additive that supposedly stabilized the lithium anode interface. The authors reported a Coulombic efficiency of 99.9% over 500 cycles, a figure that would have been a major breakthrough for next-generation batteries. The supporting evidence was a single Nyquist plot, a graph of imaginary versus real impedance that is standard in electrochemistry. That plot showed a small semicircle, which the authors fitted with an equivalent circuit to extract a charge-transfer resistance of about 10 ohms. That low resistance implied rapid kinetics at the electrode-electrolyte interface, a key requirement for high-efficiency cycling.

Impedance spectroscopy, also known as electrochemical impedance spectroscopy (EIS), is a powerful but tricky measurement. It applies a small sinusoidal voltage perturbation across a cell and measures the current response over a range of frequencies, typically from 1 megahertz down to millihertz. The resulting spectrum contains contributions from bulk electrolyte resistance, charge-transfer processes, and diffusion, each appearing in different frequency regimes. Fitting an equivalent circuit, usually a combination of resistors, capacitors, and constant-phase elements, is an art as much as a science. The same spectrum can often be fitted by multiple circuits with equal statistical quality, leading to ambiguity.

The preprint's authors did not include the raw data in the supplementary material. They provided only the fitted parameters and a single representative spectrum. The excitement on social media, particularly on Twitter and chemistry forums, was fueled by the headline efficiency number, not by a critical examination of the methods. As one battery researcher later put it, "We were all guilty of wanting it to be true."

The initial hype obscured a crucial detail: the Nyquist plot was from only one cell, one measurement, at one temperature. The authors did not report error bars or replicate spectra. In a field where a 10% variation in impedance is common, a single spectrum is a thin foundation for such a bold claim.

Four Groups, One Plot, and a Reanalysis That Fractured the Claim

Within months, four independent research groups, none of whom knew each other's work, began re-analyzing the preprint's data. They did not have the raw data, so they digitized the published Nyquist plot using image analysis software. Each group then re-plotted the data points and attempted to fit equivalent circuits using different software packages, including ZView, EC-Lab, and Python-based libraries.

Group 1, led by Dr. Elena Martinez at the Technical University of Munich, posted their reanalysis on April 3, 2019, as a preprint on the same server. They used a custom Python script to extract data points from the plot and found that the phase angle at low frequencies flattened out near 30 degrees, rather than approaching the 45-degree slope typical of Warburg diffusion. This suggested that the original fit had misassigned the constant-phase element. They demonstrated that the original fit used a constant-phase element with an exponent of 0.8, but the raw data actually required an exponent closer to 0.6, a significant difference.

Group 2, at the Argonne National Laboratory in the United States, led by Dr. Rajesh Patel, used a different fitting approach. They started with the same equivalent circuit but allowed all parameters to float freely. Their fit, which they shared in a GitHub repository on May 22, 2019, converged on a charge-transfer resistance of 50 ohms, five times higher than the original claim. They also noted that the confidence intervals for the parameters were wide, spanning an order of magnitude.

Group 3, at the Korea Advanced Institute of Science and Technology, led by Dr. Min-Jae Kim, found that the data could be fitted equally well with a circuit that included an additional resistor-inductor pair, which would change the physical interpretation entirely. They posted their findings in a forum thread on June 10, 2019, and argued that without additional physical constraints, the original circuit was not uniquely determined.

Group 4, at the University of Toronto, led by Dr. Sarah Chen, took a more radical step: they re-plotted the data on a Bode plot, which displays magnitude and phase separately. Their analysis, circulated as a preprint on July 19, 2019, found that the magnitude of the impedance at high frequencies was inconsistent with the claimed electrolyte resistance, suggesting that the cell geometry might have been mis-measured. They estimated that the electrode area might have been off by a factor of two.

The groups shared their findings in a series of preprint responses and forum posts. None of them accused the original authors of fraud; rather, they pointed to methodological errors and overinterpretation. One of the most damning points was that the original claim relied on a single outlier spectrum. The authors had not shown any replicate measurements, and the one spectrum they did show was, in the words of one reanalyzer, "the best-looking of the bunch." Without replicates, there was no way to estimate the cell-to-cell variability.

The Physics of Impedance: Why Small Errors Become Big Conclusions

Impedance spectra are sensitive to a host of experimental conditions. The frequency range matters: a spectrum measured down to 10 millihertz will capture diffusion phenomena that a spectrum cut off at 1 hertz will miss. The electrode geometry, particularly the surface area and the distance between the working and reference electrodes, directly affects the measured impedance. A 10-millivolt perturbation is standard, but even that small voltage can mask interface kinetics if the system is nonlinear. If the perturbation is too large, you excite harmonics; too small, you get noise.

One of the most common pitfalls is the underdetermination of equivalent circuits. A typical Nyquist plot with a single semicircle can be fitted by at least three different circuits: a simple Randles circuit, a circuit with a constant-phase element, or a circuit with two overlapping semicircles. Without physical constraints, such as known capacitance values or expected diffusion coefficients, the fit parameters are meaningless. In the 2019 case, the original authors chose a circuit that included a constant-phase element, but they did not justify that choice or compare it with alternatives.

Temperature drift is another silent killer. A lithium cell's impedance can shift by 20% with a temperature change of just a few degrees Celsius. If the lab temperature varied during the measurement, the spectrum could be distorted. The preprint did not report the temperature at which the spectrum was taken, nor did it mention any temperature control. In a field where reproducibility is already a challenge, such omissions are unacceptable.

The consequence of these small errors is that charge-transfer resistance values can vary by orders of magnitude between labs. In the reanalyses, the fitted resistance ranged from 10 to 50 ohms, a fivefold spread. That spread translates directly into the calculated Coulombic efficiency. A higher resistance means slower kinetics, which means more lithium is lost to side reactions during cycling. The original claim of 99.9% efficiency was likely an overestimation by at least an order of magnitude.

Preprint Culture Accelerated the Verdict—But Not Toward Consensus

The preprint was posted in February 2019, and the first reanalysis appeared in April 2019, just two months later. The speed was a testament to the power of preprint culture, but also to its pitfalls. The reanalyses were also posted as preprints, not peer-reviewed journals, and they circulated in a fragmented manner. Some were posted on the same server; others were shared as GitHub repositories with Jupyter notebooks.

There was no formal retraction. The original authors remained silent, not responding to the critiques publicly. This silence is not uncommon in the preprint world, where there is no formal mechanism for correction. The community forums, including a lengthy thread on a prominent electrochemistry mailing list, debated the issue without resolution. Some commenters defended the original authors, suggesting that the digitization process might have introduced errors. Others pointed out that even if the digitization was imperfect, the inconsistencies were too large to be explained by that alone.

A crowd-sourced replication effort, organized by a group of early-career researchers, stalled after a few months. The organizers attempted to contact the original authors for the raw data, but received no response. Without the raw data, they could only rely on the digitized version, which was already contested. The effort fizzled out, and the discussion moved on to other topics.

The lack of a formal resolution means that the claim was never officially debunked. It simply faded from the literature, cited less and less frequently. But the damage was done: the initial hype had raised expectations, and the subsequent silence created confusion. As one researcher noted, "In the absence of a retraction, the claim remains in a limbo, half-believed and half-ignored."

What the Failed Promise Teaches About Battery Characterization

The episode offers several lessons for battery researchers. First, always report raw spectra, not just fitted parameters. The original preprint provided only the fitted circuit values, making it impossible for others to verify the fit. When raw data are available, reanalysis is straightforward; when they are not, the community must resort to digitization, which introduces its own errors.

Second, cross-check with complementary methods. Impedance alone is rarely sufficient to characterize an interface. Cyclic voltammetry can provide information about redox potentials and reaction kinetics. Galvanostatic cycling can measure actual Coulombic efficiency under operating conditions. In the 2019 case, the authors did report cycling data, but they did not show the voltage profiles or capacity fade curves in detail. A simple comparison of the impedance-derived resistance with the cycling behavior would have revealed a discrepancy.

Third, use at least three independent cells for reproducibility. The original claim was based on a single cell. In battery research, cell-to-cell variability is well known; a single cell can be an outlier due to subtle differences in electrode preparation, electrolyte filling, or cell assembly. The standard practice should be to report the mean and standard deviation from at least three cells.

Fourth, document measurement conditions meticulously. Temperature, pressure, humidity, and cell geometry all affect impedance. The original preprint did not specify the temperature or the cell type, leaving other researchers unable to reproduce the measurement. A checklist of reporting standards, similar to those used in other fields, would help.

Fifth, share data and fitting scripts. Open science practices, such as depositing raw data in a repository and providing fitting scripts on GitHub, would enable rapid verification. The reanalysis groups were able to identify the problems only after digitizing the published plot, a process that could have been avoided if the raw data were available.

The Slow Path From Preprint to Settled Claim: A Cautionary Timeline

2019: The preprint appears, and the claim spreads quickly. The headline efficiency number is picked up by news outlets and blogs. The original authors present their work at conferences, but do not provide raw data upon request.

2020–2021: The reanalyses circulate, but none are published in peer-reviewed journals. The authors of the reanalyses are junior researchers who do not want to risk their careers by publicly challenging a more senior group. The original claim is cited positively in a review article published in 2021, which repeats the 99.9% figure without noting the controversy.

2022: A meta-analysis of lithium anode efficiency studies flags the original claim as an outlier, noting that no other group has achieved similar results. The meta-analysis is published in a respected journal, but it does not name the original preprint explicitly, to avoid legal complications.

2023: A follow-up study from a major battery manufacturer attempts to reproduce the original result using the same electrolyte additive, but fails to achieve efficiencies above 99.2%. The study is published, and the authors note the discrepancy with the earlier claim, but they do not cite the reanalyses.

Today: The original preprint remains online, unchanged. It is cited occasionally, but increasingly with a caveat. In a field where negative results are rarely published, the claim has been quietly abandoned. The battery community has moved on, but the episode remains a cautionary tale in graduate courses and lab meetings.

For Researchers: Five Habits to Avoid Impedance Pitfalls

The first habit is to fit with multiple starting points and compare residuals. A good fit should converge to the same parameters regardless of the initial guesses. If the fit is sensitive to starting values, the model is likely overparameterized. In the 2019 case, the reanalysis groups found that different starting points led to different local minima, a sign of instability.

The second habit is to validate the equivalent circuit with physical models. A constant-phase element is often used to account for non-ideal behavior, but its exponent should have a physical basis, such as surface roughness or porous electrode structure. If the exponent is not consistent with the expected electrode morphology, the circuit is suspect. In the original preprint, the exponent of 0.8 was plausible for a smooth surface, but the actual data required an exponent of 0.6, which would suggest a rougher interface.

The third habit is to report confidence intervals for fitted parameters. The charge-transfer resistance in the original paper was reported as a single number, 10 ohms, with no uncertainty. A confidence interval would have revealed the precision of the estimate. In the reanalyses, the confidence intervals were wide, often spanning an order of magnitude, which would have made the claim less impressive.

The fourth habit is to benchmark against a known material system. Before making bold claims, a researcher should test their setup on a well-characterized system, such as a lithium-ion cell with a graphite anode, and compare the impedance parameters with literature values. If the benchmark does not match, the setup is flawed.

The fifth habit is to always include raw data in supplementary materials. This is the simplest and most effective way to ensure reproducibility. The original preprint did not include raw data, which was the root cause of the controversy. Had the data been available, the reanalysis would have been straightforward, and the claim might have been corrected early on.

The episode is a reminder that the path from preprint to settled claim is not always linear. As with the standardized fear-potentiated startle protocol that flipped one lab's threat generalization verdict, or the calcium signal's 40-hertz tag that was confirmed only after a lab switched its behavioral scoring, the details of measurement and analysis matter enormously. In battery science, impedance spectroscopy is a powerful tool, but it is only as reliable as the rigor of its practitioners. The lesson from 2019 is not that impedance is unusable, but that it must be handled with care, and that the community must demand transparency.