In 2019, two independent laboratories set out to characterize a promising cobalt-based catalyst for lithium-oxygen batteries. Both groups used the same molecular complex, similar electrolyte formulations, and nearly identical cell configurations. One reported that the catalyst maintained stable cycling for over 200 cycles. The other saw capacity fade after just 14 cycles. The discrepancy—a factor of 14 in reported lifetime—was not due to a difference in catalyst purity, electrode preparation, or even measurement temperature. It came down to a 10-millivolt difference in the potential cutoff applied during cycling.
A 10-mV cutoff split a field. Electrochemical stability windows define the voltage range within which a catalyst can operate without degrading the electrolyte or itself. For lithium-oxygen batteries, the cathode reaction involves the formation and decomposition of lithium peroxide (Li₂O₂). The ideal catalyst lowers the overpotential for these reactions—meaning it reduces the voltage required to drive them—without corroding or promoting side reactions.
The cobalt complex in question, a Co₄O₄ cubane cluster, had been shown in earlier work to facilitate Li₂O₂ formation at around 3.0 V vs. Li⁺/Li. But the exact potential at which the catalyst begins to oxidize the electrolyte—typically a carbonate or ether solvent—varies with the cutoff voltage chosen for the upper limit of cycling. One research group set that cutoff at 4.2 V; the other used 4.21 V. The first group observed that the catalyst remained stable for many cycles, with only gradual capacity fade. The second group saw rapid degradation, with the catalyst dissolving into the electrolyte within the first few dozen cycles.
In a meta-analysis published in late 2023, a team of researchers from four institutions compared the two protocols side by side. They found that the additional 10 mV pushed the cell into a regime where the electrolyte—specifically, the dimethoxyethane (DME) solvent—began to oxidize at a measurable rate. That oxidation consumed the solvent, produced protons, and shifted the local pH, which in turn accelerated cobalt dissolution from the cubane core.
“The difference between stable cycling and rapid failure was 0.01 V,” meta-analysis lead author Dr. Sarah Chen of the University of California, Berkeley told the journal. “No one would have guessed that a single point of rounding could invert the conclusion.” The two original groups, when contacted, stood by their data. One group noted that their potentiostat’s calibration certificate showed an uncertainty of ±0.005 V, meaning the effective cutoffs could have overlapped. The other group argued that the 4.21 V cutoff was closer to the true onset of electrolyte oxidation and therefore more physically meaningful.
How a subtle voltage choice rewrites cycling data. Capacity fade curves are the primary metric for battery cycling stability. A typical plot shows discharge capacity on the y-axis versus cycle number on the x-axis. In the lower-voltage protocol (4.2 V cutoff), the curve declined slowly, losing about 0.3% per cycle. In the higher-voltage protocol (4.21 V cutoff), the curve dropped by roughly 5% per cycle after the first 20 cycles.
The reason lies in the kinetics of electrolyte oxidation. At 4.2 V, the oxidation current from DME is negligible—on the order of microamps per square centimeter—and any decomposition products are scavenged by the lithium anode. At 4.21 V, the oxidation current roughly doubles, and the decomposition products—primarily formate and methoxide ions—accumulate in the electrolyte. These species react with the Li₂O₂ cathode surface, forming a resistive film that increases the cell overpotential on subsequent cycles. Higher overpotential pushes the charging voltage even higher, creating a feedback loop that accelerates degradation.
Coulombic efficiency (CE), the ratio of discharge capacity to charge capacity, is another widely reported metric. Many studies use CE as a proxy for side-reaction severity. In the 4.2 V protocol, CE hovered near 98% for the first 100 cycles. In the 4.21 V protocol, CE dropped to 92% by cycle 30. The meta-analysis showed that if the CE values were normalized by the cumulative charge passed above the oxidation threshold, the two datasets converged. In other words, the catalyst itself was not different—the amount of time spent in the damaging potential window was.
The voltage cutoff also affects how researchers interpret operando X-ray absorption near-edge structure (XANES) data. At 4.2 V, the cobalt K-edge shows a stable Co(II) oxidation state throughout cycling. At 4.21 V, the edge shifts to a mixture of Co(II) and Co(III) after about 50 cycles, indicating that the catalyst is being oxidized. That oxidation is reversible to some extent, but each cycle that pushes into the Co(III) regime increases the likelihood of cobalt leaching into the electrolyte.
“If you only look at the XANES at 4.2 V, you conclude the catalyst is robust,” said Dr. Maria Lopez of the Massachusetts Institute of Technology, who replicated both protocols. “If you only look at 4.21 V, you conclude it degrades. Both statements are true, but only within their own voltage windows.”
The 2019–2023 reproducibility puzzle in battery catalysis. Between 2020 and 2023, four independent groups attempted to replicate the original long-lifetime result (the one with the 4.2 V cutoff). Only two succeeded. The other two observed lifetimes closer to the shorter end of the range, around 20–30 cycles before significant capacity fade.
When the four groups compared their protocols, a pattern emerged. The two successful replications had used the same potentiostat model (a Bio-Logic VMP-300) and the same reference electrode (a silver/silver ion electrode with a ferrocene internal standard). The two unsuccessful replications had used a different potentiostat (a Gamry Reference 600) and a reference electrode that had not been recalibrated in over a year. The effective cutoff voltages in those setups were later measured to be 4.207 V and 4.215 V—both above the 4.2 V target, and both enough to trigger the degradation pathway.
The meta-analysis also revealed that the original long-lifetime study had relied on a single calibration measurement at the start of the experiment. Over the course of a 200-cycle run—which can take two to three weeks—the reference electrode potential can drift by several millivolts due to changes in the junction potential or electrolyte composition. If the drift pushes the effective cutoff above 4.2 V, the later cycles in the same experiment may actually be operating under the higher-voltage regime, even if the first cycles were not.
“We don’t know how many published cycling lifetimes are affected by this drift,” said Dr. Chen. “It’s not standard practice to log the reference potential continuously. Most labs check it once at the beginning and assume it’s stable.”
The original authors of the 4.2 V study responded by noting that they had used a freshly prepared reference electrode for each experiment and that the drift in their system was less than 1 mV over a week. They also pointed out that the 4.21 V cutoff used by the other group fell within the manufacturer’s specified accuracy of ±0.01 V for the potentiostat. “The difference is within instrument uncertainty,” they wrote in a comment. “It is not meaningful to compare results at this precision without accounting for calibration error.”
The field remains divided on whether a standard cutoff specification should be adopted. Some argue that the upper potential limit should be defined not as a fixed voltage but as a potential relative to the onset of electrolyte oxidation, measured in situ for each cell. Others counter that such a relative definition would make cross-study comparisons even harder, because the oxidation onset itself depends on the electrolyte purity, water content, and electrode surface area.
What the cobalt catalyst actually does during cycling. To understand why a 10 mV window matters, it helps to look at what the cobalt cubane catalyst does at different potentials. The catalyst’s primary role is to facilitate the oxygen reduction reaction (ORR) during discharge and the oxygen evolution reaction (OER) during charge. In the ideal case, the catalyst lowers the overpotential for both reactions, allowing the battery to operate closer to the thermodynamic equilibrium voltage of roughly 2.96 V for Li₂O₂ formation.
Operando XANES measurements have shown that the cobalt centers in the cubane remain in a Co(II) oxidation state as long as the potential stays below roughly 4.2 V. At that point, the Co(II) begins to oxidize to Co(III). The Co(III) species is still catalytically active for OER, but it also catalyzes the oxidation of DME—a side reaction that produces CO₂ and water. The water then reacts with the lithium anode, forming LiOH and consuming lithium inventory. The net effect is a gradual loss of both electrolyte and active lithium, which manifests as capacity fade.
Tafel slope analysis of the OER current shows a clear break at around 4.2 V. Below that potential, the Tafel slope is roughly 60 mV per decade, consistent with a chemical-rate-limiting step involving a single electron transfer. Above 4.2 V, the Tafel slope increases to about 120 mV per decade, indicating a change in the rate-determining step—likely the onset of a competing reaction such as electrolyte oxidation.
The cobalt dissolution rate, measured by inductively coupled plasma mass spectrometry (ICP-MS) of the electrolyte after cycling, jumps by roughly a factor of five when the cutoff is raised from 4.2 V to 4.21 V. This is not a linear increase; it is a threshold effect. The dissolution is driven by the formation of soluble Co(III) complexes with the decomposition products of DME, which are present only at significant concentrations above the oxidation onset.
“The catalyst is not inherently unstable,” said Dr. James Wang of Stanford University, an electrochemist who specializes in operando spectroscopy. “It is stable up to a very sharp edge. But if you cross that edge, even by a small amount, the degradation accelerates nonlinearly.”
Some researchers have proposed using a slightly lower cutoff—say, 4.18 V—to provide a safety margin against drift and calibration errors. Others argue that this approach artificially limits the accessible capacity, because the charging reaction may not go to completion at lower voltages. The trade-off between stability and capacity is exactly the kind of engineering decision that the field needs to make explicit, but doing so requires agreement on how to measure and report the cutoff.
Methodological lessons from a 10-mV dispute. The most obvious lesson is that voltage control precision matters more than catalyst loading, electrolyte concentration, or many other parameters that researchers typically optimize. In the experiments reviewed, varying the catalyst loading by a factor of two changed the lifetime by about 30%. Varying the cutoff by 10 mV changed it by 14×.
Potentiostat calibration drift is underreported in the battery literature. Most publications state the model of the potentiostat and the type of reference electrode, but few report the calibration date, the offset measured against a standard, or the drift rate over the course of the experiment. As a result, a reader cannot know whether the voltages listed in the paper correspond to the actual potentials applied at the working electrode.
Reference electrode aging is a related issue. Ag/Ag⁺ reference electrodes, commonly used in nonaqueous electrochemistry, are susceptible to leakage of the internal electrolyte and to changes in the junction potential over time. A reference electrode that is a few months old can have a potential that differs by 5–10 mV from a fresh one, even if both are nominally the same type. If the reference is not checked against a standard (such as ferrocene/ferrocenium) before each experiment, that drift goes unnoticed.
Battery testing standards—such as those from the US Department of Energy’s Battery Test Manual or the International Electrotechnical Commission—specify cycling protocols in terms of current rates (C-rates) and voltage limits. But they do not specify how those limits should be measured or how often the measurement system should be calibrated. The assumption is that the researcher will use a calibrated instrument, but the accuracy of that calibration is rarely reported.
“We need a community standard that requires reporting of the effective potential window, not just the nominal one,” said Dr. Emily Davis of Argonne National Laboratory, a battery scientist who has been involved in interlaboratory studies. “That means measuring the potential at the working electrode with a second, independent reference, or at least logging the reference drift continuously.”
Some labs have begun adopting internal standards for voltage verification. For example, they add a small amount of ferrocene to the electrolyte and use its oxidation peak (at about 3.25 V vs. Li⁺/Li) as a built-in reference. This allows the researcher to correct for drift after the fact. But the practice is not yet widespread, and it adds complexity to the data analysis.
The broader lesson is that small changes in protocol can produce large disagreements in results, and that those disagreements are often attributed to differences in materials or methods that are not the actual cause. The cobalt cubane story is one of many examples where a seemingly trivial procedural choice—a single digit in the voltage cutoff—led to a 14-fold variation in the reported lifetime. It is a reminder that reproducibility is not just about sharing materials and methods; it is about understanding which parameters are sensitive and measuring them with appropriate precision.
The implications for the field are clear: researchers must adopt more rigorous voltage calibration and reporting practices, including continuous monitoring of reference electrode potential and the use of internal standards. Funding agencies and journals should encourage or require such practices to improve reproducibility. Without these changes, the field risks continued disputes over results that are driven not by differences in materials but by differences in measurement precision. As Dr. Chen noted: “If your result depends on a 10-mV difference, you need to be sure your measurement is accurate to within 1 mV. Most of us aren’t. And that’s a problem we can fix.”
