Understanding and suppressing gas evolution in lithium metal batteries with ether-based electrolytes

Abstract

Understanding and suppressing gas evolution is critical to enabling high-energy-density lithium metal batteries (LMBs). Yet, comprehensive investigations in ether-based systems remain limited. Here we quantify gas generation in ether-based LMBs and elucidate the underlying mechanisms. We link CO and CO₂ production to the cathode, and CH₄ evolution to the anode. Notably, CO and CO₂ are consumed at the Li anode to form Li-containing species such as Li₂CO₃. Although CH₄ ultimately dominates the gaseous products, its evolution during cycling is delayed until a distinct onset point. We show that in a high-concentration ether electrolyte, anode activation improves Li deposition morphology and suppresses interfacial reactions, extending the number of cycles to gas onset and cell failure by an order of magnitude. Achieving these gains without altering the electrolyte enables the reconsideration of seemingly impractical electrolytes, highlighting a practical strategy to enhance the safety and performance of commercial LMBs.

Gas evolution compromises the safety and commercial viability of high-energy lithium metal batteries. Now it has been shown that CH₄ generated by ether-based electrolyte decomposition at the Li anode dominates gas evolution. A facile anode activation strategy suppresses CH₄-forming interfacial reactions, dramatically reducing gas evolution and extending cycle life.

Main

Lithium metal anodes, with a theoretical capacity of 3,860 mAh g⁻¹ (refs. ^1–3), are attractive for next-generation batteries targeting electric aviation applications such as electric vertical take-off and landing (eVTOL) aircraft and long-range electric vehicles^4,5. Despite their promise, lithium metal battery (LMB) technology, which preceded but was largely displaced by lithium-ion battery (LIB) technology following safety-related recalls (for example, Moli Energy in 1989), has yet to regain market share. Widespread commercialization remains contingent on simultaneously optimizing electrochemical performance^6–17, cost and safety^2,18. Among these, safety is critical because uncontrolled gas evolution can rupture pouch cells or trigger premature venting in prismatic and cylindrical formats, leading to electrolyte leakage and the release of toxic, flammable gases^19–23.

Replacing graphite with metallic lithium fundamentally alters the interfacial chemistry and gas evolution pathways. Consequently, gassing mechanisms established for LIBs are not directly transferable to LMBs^20,22. Existing studies have largely focused on carbonate electrolytes, drawing on LIB-derived frameworks^19,23–25. In contrast, operando investigations of ether-based electrolytes remain limited^20–23,26, despite their widespread use in LMBs due to superior stability against Li metal^6,27–29. Identifying the evolved gases and their formation pathways is therefore essential for mitigating gas generation in ether-based LMB systems.

Elevated temperatures accelerate electrolyte decomposition and gas generation^21,30,31, making mechanistic understanding particularly important for aviation applications. Under constrained pack-level thermal management, high-rate operation can drive cell temperatures above 50 °C (refs. ^32–35). Simulations of high-energy prismatic cells under air-cooled conditions predict temperatures approaching 62 °C during high-rate discharge (8 C; Supplementary Fig. 1), highlighting the importance of studying thermally accelerated degradation pathways.

We aim to address the existing gap in our mechanistic understanding of gas evolution in pouch-format LMBs employing high-concentration ether-based electrolytes (HCEs) and lithium–nickel–manganese–cobalt oxide (NMC811) cathodes^23,26 using operando gas analysis, high-fidelity pouch cell testing, and first-principles simulations. We will identify dominant gas species, elucidate their formation pathways, and quantify how electrolyte composition and operating temperature influence gas generation. Earlier studies have reported reductive C–O bond cleavage for carbonate solvents^24,26, raising the possibility that analogous pathways occur for ether solvents such as 1,2-dimethoxyethane (DME). By directly probing gas evolution in pouch-format cells, this work extends previous carbonate-focused frameworks to ether-based LMBs, identifying dominant pathways of gas generation and providing guidance for interfacial and electrolyte engineering to suppress gas evolution while maintaining high-rate performance.

Overview of gas evolution in LMBs

Gas evolution in Li‖NMC single-layer-stack (SLS) pouch cells (Fig. 1a) containing 4 M lithium bis(fluorosulfonyl)imide (LiFSI) in DME was investigated using Archimedes methods³⁶ to quantify cell volume changes. Ex situ gas chromatography (GC) was used to determine gas composition. Figure 1b shows the voltage and in situ volume profiles during cycling at 60 °C over 2.8–4.3 V, with charging at 0.2 C (28 mA) and discharging at 0.5 C (70 mA). The volume profile exhibits a slight initial rise, a gradual decline (from ~500 to 3,000 min), and a sharp increase after five cycles (at ~3,000 min), marking the onset of gas evolution. Approximately 9 ml of gas was produced after 28 cycles, leading to pronounced cell swelling (Supplementary Fig. 2). Low-amplitude volume oscillations reflect reversible expansion during cycling, consistent with the periodic voltage trace. Replicate in situ volume profiles are presented in Supplementary Information.

Fig. 1. Investigating and quantifying gas evolution in LMBs.

a, Schematic of Li‖NMC full cell, LFP counter electrodes, Li‖LFP anode half-cell and d-LFP‖NMC cathode half-cell systems used to study gas evolution. b, Voltage (black) and cell volume changes (orange) over time for a Li‖NMC full cell with 4 M LiFSI/DME electrolyte cycled at 60 °C. c, Normalized volumes of gas components generated in Li‖NMC, Li‖LFP and d-LFP‖NMC cells. Data are presented as mean ± s.d. (n = 2, each from a separate cell). d, Schematic showing the presence or absence of crosstalk occurring in cells where CO and CO₂ are evolved in the presence of NMC. e, Schematic and CO₂ signal over time for the titration–mass spectrometry experiment used to quantify the amount of a crosstalk product, Li₂CO₃.

To isolate electrode-specific gas evolution, anode and cathode half-cells (Fig. 1a) were prepared by pairing each electrode with an ‘inert’ lithium iron phosphate (LFP, LiFePO₄) counter electrode, as delithiated LFP‖LFP (d-LFP‖LFP) cells produced no measurable gas during 30 cycles between 0 and 100% state-of-charge (SOC) at 60 °C (Supplementary Fig. 3). Justification for using the anode (Li‖LFP) and cathode (d-LFP‖NMC) half-cells instead of symmetric cells is provided in Supplementary Note 1.

Figure 1c shows the normalized gas volumes and compositions for the full cell, anode half-cell and cathode half-cell. Cell volume changes measured ex situ were normalized by design capacity and cycle time (Methods), and gas compositions were measured using GC. The anode half-cell behaved similarly to the full cell, producing ~0.3 ml Ah⁻¹ h⁻¹ of gas dominated by methane (CH₄, ~97%; Supplementary Tables 2 and 5), together with trace quantities of ethane (C₂H₆), ethylene (C₂H₄), acetylene (C₂H₂), propylene (C₃H₆), propane (C₃H₈), hydrogen (H₂) and oxygen (O₂). In contrast, the cathode half-cell produced ~0.1 ml Ah⁻¹ h⁻¹ of gas consisting primarily of carbon monoxide (CO, ~36%) and carbon dioxide (CO₂, ~62%) (Supplementary Table 3), neither of which was detected in the full cell during endpoint GC measurements.

This discrepancy is explained by crosstalk between the electrodes (Fig. 1d). CO and CO₂ generated by cathode-side oxidation are subsequently consumed at the Li anode through the formation of lithium-containing species such as lithium carbonate (Li₂CO₃)^37–39. This process was confirmed by titration–mass spectrometry (T–MS)^15,40 using sulfuric acid (H₂SO₄) as the titrant (Fig. 1e). A strong T–MS signal from the Cu electrode in a Cu‖NMC cell (blue curve) indicates substantial cathode-to-anode crosstalk. Cu was used as a Li surrogate for safety, and its representativeness was validated experimentally (Supplementary Fig. 4). A small but nonzero signal was also observed in Cu‖Li cells (orange curve), which lack a cathode-side oxidative source of CO/CO₂. This signal is fully attributable to reactions between trace dissolved CO₂ and deposited Li (see Supplementary Note 2 for CO₂ solubility). In contrast, d-LFP‖NMC cathode half-cells generated CO and CO₂ (Fig. 1c), but lacked conditions necessary for Li₂CO₃ formation (Fig. 1d), resulting in no detectable T–MS signal (Fig. 1e, grey curve). Baseline titrations of fresh Cu and delithiated LFP electrodes produced no measurable signal (Supplementary Fig. 4).

The in situ volume profile for the Li‖NMC full cell (Fig. 1b, orange) captures the temporal evolution of these processes. Corresponding measurements of the cathode half-cell show that cathode-associated gas evolution occurs primarily during the first cycle (Supplementary Fig. 5, blue curve). The small initial volume increase in the full cell is therefore consistent with, at least in part, CO and CO₂ generation during early cycling. Notably, the onset of DME decomposition to CO and CO₂ occurs at ~4.13 V versus Li/Li⁺, close to the ~4.3 V reported by Jung and colleagues for chemical oxidation of carbonate solvents in the presence of NMC811⁴¹. The subsequent gradual volume decrease probably reflects crosstalk consumption of CO and CO₂, a feature absent from the cathode half-cell profile (Supplementary Fig. 5). Some cathode-side gas may also dissolve into the electrolyte, although this contribution is expected to be minor given the low CO₂ solubility (0.32 ml g⁻¹ at 25 °C; Supplementary Note 2). After five cycles (~3,000 min), gas production becomes dominated by anode-derived CH₄ production and increases sharply.

Cathode-side gas generation

Figure 2a shows that normalized gas volume from d-LFP‖NMC cathode half-cells cycled at 60 °C decreases as the LiFSI/DME electrolyte concentration increases from 2 to 5 M. At a fixed 4 M concentration (Fig. 2b), the gas volume increases with temperature from 45 to 70 °C. For all conditions, CO and CO₂ comprise ~98% of the evolved gas (Supplementary Tables 3 and 4).

Fig. 2. Gas evolution in d-LFP‖NMC cathode half-cells.

a,b, Normalized (norm.) gas volumes and gas component distributions at different concentrations of LiFSI (a, 2, 4, 5 M) and at different temperatures (b, 45, 60, 70 °C). Data are presented as mean ± s.d. (n = 2, each from a separate cell). c, Simulated pathways for the oxidation of DME into CO and CO₂.

The mechanisms for CO and CO₂ formation were identified using first-principles simulations. Both gases are primarily formed via exergonic DME chemical oxidation pathways mediated by the NMC cathode (Fig. 2c). DME first adsorbs onto the NMC surface through interactions between its O atom and surface transition metal (TM) sites. In the CO-forming pathway (dark blue), two successive dehydrogenations of a methyl end group are followed by C–O bond cleavage, producing an O=CH^• radical, which then transfers a hydrogen (H) to a reactive surface oxygen species, yielding CO. The CO₂-forming pathway (light blue) also initiates with methyl dehydrogenation. Oxygen originating from the NMC surface becomes incorporated into the DME fragment, which undergoes further H loss and C–O bond cleavage to form an O=C–O^• radical. A final dehydrogenation step yields CO₂. Consistent with these pathways, experiments with an oxidatively stable electrolyte confirmed that CO/CO₂ evolution arises predominantly from DME oxidation rather than from the cathode itself (Supplementary Note 3 and Supplementary Fig. 6).

The decrease in the normalized gas volume (ml Ah⁻¹ h⁻¹) with increasing LiFSI concentration (Fig. 2a) arises from several factors. Lower DME content slows gas generation, while a higher FSI⁻ content promotes formation of a passivating cathode electrolyte interphase (CEI) that suppresses oxidation^16,42. Interfacial side reactions drive NMC surface erosion and microcracking⁴³, primarily through electrode–electrolyte surface reactivity rather than bulk structural degradation³¹. Scanning electron microscopy (SEM) images show more pronounced microcracking at 3 M than at 5 M (Supplementary Fig. 7), consistent with enhanced DME reactivity and increased gas production at lower concentrations. This cracking does not indicate bulk cathode degradation, as X-ray diffraction (XRD) reveals no structural changes between concentrations (Supplementary Fig. 8). Cathode thickness increased at 3 M (Supplementary Fig. 7), probably reflecting reaction-induced deformation, whereas the 5 M sample retained the thickness of fresh NMC.

The increase in normalized volume with temperature (Fig. 2b) indicates thermally activated oxidation kinetics. For a given concentration, faster DME decomposition produces a larger volume of CO and CO₂ within the same experimental time window.

Anode-side gas generation

In Li‖LFP anode half-cells, CH₄ was the dominant gas species (~96–98%) across electrolyte concentrations (2–5 M LiFSI/DME; Fig. 3a and Supplementary Table 5) and temperatures (45–70 °C; Fig. 3b and Supplementary Table 6). Gassing profiles show earlier onset at lower concentrations (Fig. 3c) and higher temperatures (Fig. 3d). For the 2 M electrolyte, gas evolution began after only two cycles (Fig. 3c) and produced the largest normalized gas volume (Fig. 3a). At 5 M, gas evolution occurred near end of life (~150 cycles), beyond the practical limits of in situ measurements (~20 cycles), but was confirmed by pouch swelling. Higher temperatures produced earlier onset (Fig. 3d) and larger normalized volumes (Fig. 3b).

Fig. 3. Gas evolution in Li‖LFP anode half-cells.

a,b, Normalized gas volumes and gas component distributions at different concentrations of LiFSI (a, 2, 4, 5 M) and at different temperatures (b, 45, 60, 70 °C). Data are presented as mean ± s.d. (n = 2, each from a separate cell). c,d, Corresponding cell volume changes at different concentrations of LiFSI (c, 2, 4, 5 M) and at different temperatures (d, 45, 60, 70 °C). e, Electrolyte consumption results showing the decrease in the masses of DME (squares) and LiFSI (circles), as well as the overall decrease in the LiFSI/DME concentration (triangles) over a number of cycles. Error bars represent 90% confidence intervals of the mean (n = 2–4). Four samples were prepared for each cycle number; two samples for cycle 4 were unexpectedly damaged. f, CH₄ signal over time for the T–MS experiment used to quantify the amount of LiCH₃. g–i, Simulated pathways for the formation of CH₄ at the anode. g, Step 1 involves C–O bond cleavages that result in two key intermediates (coloured). h,i, Two possible subsequent pathways: step 2A, in which LiCH₃ reacts with another DME to produce CH₄ (h), and step 2B, in which the two intermediates from step 1 react to produce CH₄ (i).

To probe conditions around the onset point, electrolyte content was measured across cycles for the 4 M sample at 60 °C using our recently developed electrolyte consumption analysis method, extraction-gas and ion chromatography (E-G&IC)³⁶ (Methods). As shown in Fig. 3e, the masses of both LiFSI (circles) and DME (squares) decreased with cycle number, with faster LiFSI consumption leading to an overall decrease in electrolyte concentration (triangles). The linear decrease in both components over 15 cycles indicates that the anode–electrolyte reaction rate remained unchanged before and after the onset point (cycle 4). However, in situ profiles (Fig. 3c,d and Supplementary Fig. 9) showed negligible gas formation prior to the onset point, with pronounced evolution only thereafter.

First-principles simulations reconciled this observation by identifying a two-step DME decomposition pathway in which DME first forms non-gaseous intermediates that subsequently generate CH₄ gas (Fig. 3g–i). In step 1 (Fig. 3g), DME adsorbs onto the Li surface. The less exergonic pathway (grey, ΔG = −2.69 eV) cleaves a C–O bond in the ethane backbone to yield surface-bound methoxy (Li–OCH₃) and methoxyethyl (Li–C₂H₄OCH₃, blue) species. The second pathway (orange, ΔG = −3.58 eV) cleaves the methoxy C–O bond to form methyllithium (LiCH₃, orange). LiCH₃ can in principle react with another DME to produce CH₄ (step 2A, Fig. 3h), but this pathway requires additional DME consumption, inconsistent with the linear trend in Fig. 3e. More likely, LiCH₃ reacts with DME-derived fragments at the Li surface. Among these, methoxyethyl is the most thermodynamically favoured (lowest ΔG, Supplementary Table 13), yielding CH₄ (step 2B, Fig. 3i). This pathway also exhibits a 20% lower activation barrier (0.88 versus 1.14 eV), supporting step 2B as the dominant CH₄ production pathway.

Evidence for LiCH₃-like intermediates was obtained using T–MS, and the onset of CH₄ evolution was captured by in situ differential electrochemical mass spectrometry (DEMS). The T–MS experiment is analogous to that for Li₂CO₃ detection; here, LiCH₃ reacts with H₂SO₄ to generate CH₄. A strong signal was observed for anodes cycled to the onset point (Fig. 3f), but none from fresh anodes or cycled electrolyte, indicating surface LiCH₃ accumulation from DME–anode reactions. Ex situ GC analysis (Supplementary Table 2) established CH₄ as the dominant gas species (~97%), and in situ DEMS (Extended Data Fig. 1) confirmed CH₄ as the major onset species. Final CH₄ and C₂H₆ proportions from DEMS (95% and 2%, respectively; Extended Data Fig. 1c) closely match the GC values (97% and 3%). As CO and CO₂ were directly routed to the detector in DEMS, crosstalk was unlikely. Removing their contributions to emulate crosstalk consumption improves the agreement with the GC results (97% and 2%). The coincident onset of C₂H₆ (green) and CH₄ (orange) in DEMS profiles (Extended Data Fig. 1, zoomed-in panels) further supports LiCH₃ as the intermediate, consistent with parallel pathways where LiCH₃ protonation to CH₄ dominates while minor coupling yields C₂H₆.

Extended Data Fig. 1. In situ DEMS results for a Li‖NMC cell cycled at 60 °C with 4 M LiFSI/DME.

a, Gas flow rates of different gas species. b, Accumulated gas volumes of different gas species; the dashed line denotes the total volume of gas generated. We note that detection timing in DEMS may not perfectly align with in situ volume measurements due to differences in instrument sensitivity and carrier gas effects (for example, solubility, vaporization, reaction kinetics, and potential loss of volatile species). These factors can lead to either earlier or delayed detection, depending on the gas species. Nonetheless, DEMS offers critical information both before the onset point and during extended cycling, supporting conclusions drawn from GC and volume data. c, Table showing the composition of gas produced using DEMS and endpoint GC measurements.

To examine LiCH₃ reactivity in the full cell, gas evolution in Li‖LFP and Li‖NMC (containing DME oxidation species) was compared (Supplementary Fig. 10). In Li‖LFP, LiCH₃ produced in step 1 was converted directly to CH₄ (and C₂H₆ in part). In Li‖NMC, gas onset was delayed by one cycle, consistent with partial LiCH₃ consumption through reaction with oxidation species that crosstalk from the cathode^37,38,44,45, such as CO₂ (ref. ⁴⁶). Taken together, these results confirm CH₄ as the dominant product and LiCH₃ as the key intermediate, with onset timing governed by LiCH₃ accumulation and only minor influence from crosstalk or C₂H₆ formation.

Considering the anode half-cell alone, delayed CH₄ onset at higher LiFSI/DME concentrations arises from both DME availability and changes in Li⁺ solvation. Simulations show that the activation barriers for LiCH₃ and LiOCH₃ intermediate formation (step 1 pathways in Fig. 3g) increase as the electrolyte environment shifts from predominantly solvent-separated ion pairs (SSIP) to contact ion pairs (CIP) with higher LiFSI concentration⁴⁷. The barrier for LiCH₃ formation rises from 0.69 eV (SSIP) to 0.88 eV (CIP), and from 0.58 to 0.75 eV for LiOCH₃ (Supplementary Fig. 11). These increases reflect the increased presence of the FSI⁻ anion in the first Li⁺ solvation shell, weakening Li⁺–DME coordination and reducing C–O bond polarization, making C–O bond cleavage less favourable and suppressing CH₄ production.

The shift to earlier onset with increasing temperatures reflects faster reaction kinetics, affecting both the formation and consumption rate of the LiCH₃ intermediate in the two-step pathway from DME to CH₄.

Electrolyte engineering to suppress gas evolution

As cathode-associated CO and CO₂ generation occurs during early cycles and can be efficiently removed, CH₄ remains the dominant gas in the 4 M LiFSI/DME system and the primary target for mitigation.

Notable swelling of the Li‖NMC cell was observed after only 20 cycles at 60 °C (Fig. 4a). Although CH₄ production originating from the DME terminal methyl group decreases with increasing LiFSI concentration, further salt addition is impractical due to cost, high viscosity (67 cP for 8 M; Supplementary Table 15), poor rate performance (Supplementary Fig. 12) and intrinsic solubility limits (1:1.2 LiFSI:DME mole ratio, per our measurements). A localized high-concentration electrolyte (LHCE; Fig. 4b) preserves the high-salt solvation environment by incorporating a non-Li solvating diluent, 1,1,2,2-tetrafluoroethyl-2,2,3,3-tetrafluoropropyl ether (TTE)⁴². This increases local LiFSI coordination and the activation barriers for DME reduction without the viscosity (Supplementary Table 15) or performance penalties (Supplementary Fig. 12) associated with additional salt. In the LHCE, no gas generation was observed for over 150 cycles at 60 °C, and the pouch cell remained flat (Fig. 4b).

Fig. 4. Schematics and properties of different electrolyte systems.

a, For the HCE (4 M LiFSI/DME) system, abundant free DME molecules with terminal methyl groups react to form CH₄, resulting in marked gas evolution, as observed in a Li‖NMC cell after 20 cycles at 60 °C. b, For the LHCE (LiFSI–1.2DME–3TTE) system, there are fewer free DME molecules. The reduction in DME content greatly reduces CH₄ production, and hence no gas evolution was observed after 150 cycles at 60 °C. c, For the FEE (1.2 M LiFSI/F5DEE) system, because the F5DEE molecules do not contain methyl groups, no CH₄ generation occurs and no gas evolution was observed after 120 cycles at 60 °C. Inset photographs in a–c show Li‖NMC cells cycled at 60 °C with corresponding electrolytes. Cells with LHCE and FEE remain flat. d, Cycling performance at 60 °C. e, Rate performance at 25 °C. f, Ionic conductivity values at 25 °C.

Another strategy is to eliminate the CH₄ source by replacing terminal methyl groups with fluorinated moieties. The fluorinated ether electrolyte (FEE), 2-[2-(2,2-difluoroethoxy)ethoxy]-1,1,1-trifluoroethane (F5DEE)⁴⁸ (Fig. 4c), terminates in difluoromethyl (CF₂) and trifluoromethyl (CF₃) groups. Because the critical LiCH₃ intermediate cannot form, the CH₄ pathway is eliminated, and no gas evolution was observed for over 120 cycles at 60 °C (Fig. 4c).

Although LHCE and FEE formulations suppress gas evolution and extend cycle life (~200 cycles versus 14 cycles for HCE; Fig. 4d), their lower ionic conductivities (Fig. 4f) constrain rate performance (Fig. 4e). This limits their suitability for high-power applications such as eVTOL, where a high rate capability is required for both take-off/landing and emergency response^32,35,49. These trade-offs motivate shifting from electrolyte-only solutions to engineering the Li metal anode itself.

Anode activation to suppress gas evolution

The thick native passivation film that forms on fresh, untreated Li metal (Fig. 5a, brown layer) is typically oxide-based, ionically conducting but electronically insulating. Metallic Li (pale green) consequently deposits at defect sites (mauve stripes) within this passivation layer, where Li⁺ transport and localized electron access are less hindered. As shown in Extended Data Fig. 2c, this process requires overcoming a relatively high nucleation overpotential, reflected in the sharp initial voltage spike. As the defect sites are limited in number, Li repeatedly plates from the same locations, producing high-surface-area, whisker-like morphologies^50,51 (Fig. 5a). Consistent with this schematic, SEM imaging of the untreated anode showed long, whisker-like Li deposits growing from sparse defect sites (Fig. 5g, 100% SOC). These high-surface-area structures accelerate interfacial reactions that drive electrolyte consumption, gas generation and poor cycling performance. After eight cycles, the surface was covered with a thick, porous layer containing a small amount of electrically isolated ‘dead’ Li¹⁵, reaching a combined thickness of ~28 μm (Fig. 5h, 0% SOC). The bulk of this material is referred to as the residual solid–electrolyte interphase (rSEI)^17,52. Newly deposited Li remains separated from the bulk by a persistent interfacial layer consistent with the native passivation film (Extended Data Fig. 2a).

Fig. 5. Li deposition morphology and performance of untreated and activated samples.

a,b, Schematic showing Li deposition on untreated and activated Li anode surfaces. a, Untreated Li is covered by a native passivation film. Whisker-like Li deposits grow from defect sites on the passivated surface. b, Activation of the Li anode by pre-stripping the surface to expose fresh Li, forming many pits. Li deposits within these pits in a more controlled manner. c–j, SEM micrographs of untreated and activated Li anode surfaces at various SOC. c, Untreated Li anode. d, Untreated anode with Li deposited at defect sites after an initial charge to 10% SOC. e, Pre-stripped (activated) anode with pits. f, Activated anode with Li deposited in the pits after an initial charge to 10% SOC. g, Untreated anode with whisker-like Li deposits covering the surface after an initial charge to 100% SOC. h, Cross-section of the untreated anode at 0% SOC after 8 cycles. The top layer is rSEI and dead Li. i, Activated anode with Li particles covering the surface after an initial charge to 100% SOC. j, Cross-section of the activated anode at 0% SOC after 8 cycles. The thin, top layer is rSEI. k,l, Electrolyte consumption results for untreated and activated samples. Consumption of LiFSI (circles) and DME (squares) (k) and decrease in LiFSI/DME concentration in Li‖NMC cells (l) cycled at 60 °C with activated (orange) and untreated (blue) Li anodes. Error bars represent 90% confidence intervals of the mean (n = 2–4). Four samples were prepared for each cycle number; two samples for cycles 6 and 15 were unexpectedly damaged. m,n, Corresponding cycling performance (m) and changes in cell volumes (n).

Extended Data Fig. 2. Comparison of Li deposition on untreated and activated Li anodes.

Cross-sectional SEM images of Li deposition at 20% SOC on a, untreated and b, activated Li anodes. In the untreated sample (a), Li deposits above bulk Li, separated by an interface consistent with the passivation layer, whereas in the activated sample (b), Li grows directly from pits where fresh metallic Li was exposed. c, Voltage profiles showing a sharp spike during initial charging for the untreated sample due to high nucleation overpotential from the native passivation layer, while exposed metallic Li in the activated sample enables smoother plating kinetics and voltage profile.

To mitigate interfacial reactions, two anode activation methods were developed to alter the Li deposition morphology: anode constant-voltage pre-stripping (ACVP) and cathode charging-reconstitution and pre-stripping (CCRP) (Methods). Effective activation requires sufficiently high polarization to strip Li through the native passivation film and expose fresh metallic Li (refs. ^50,53; schematic in Fig. 5b). Inadequate stripping leaves more of the passivation layer intact and limits the formation of fresh metallic Li sites, reducing the effect on subsequent Li deposition morphology and surface area.

ACVP is the more aggressive protocol, over-discharging a freshly assembled cell and holding it at 1.5 V (Supplementary Fig. 13). The high initial stripping current generates a high density of pits, while lower currents subsequently remove most of the Li, totalling to roughly half the design capacity (50–60 mAh) for the 4 M LiFSI/DME system. CCRP is a more conservative analogue designed to reduce cathode perturbation, employing a milder stripping protocol and a pre-treated cathode (Supplementary Note 4). The cathode is pre-cycled to create a lithium vacancy, then reassembled with a fresh Li anode. The reconstituted cell is discharged to 2.8 V (avoiding over-discharge), followed by stepwise current reduction, stripping only ~10% of the capacity (10–20 mAh). Due to the lower stripping extent, CCRP is inherently a less thorough activation method. Accordingly, the following morphology analysis focuses on ACVP.

Defect sites are not readily apparent on the untreated anode (Fig. 5c), which shows only sparse, localized Li deposits within an ~0.2-mm² region (Fig. 5d, 10% SOC). In comparison, the activated anode exhibits a high density of pits over the same area (Fig. 5e), which serve as preferred Li nucleation sites (Fig. 5f, 10% SOC), consistent with cross-sectional imaging (Extended Data Fig. 2b, 20% SOC). Although a thin passivation layer (Fig. 5b, dark green) forms within activation-generated pits, the newly exposed surfaces remain more favourable for Li⁺ migration and reduction. As demonstrated in Extended Data Fig. 2b, Li grows directly from these pits. The abundance of nucleation sites lowers the nucleation overpotential and produces the smoother voltage profile in Extended Data Fig. 2c, enabling particle- or globule-like Li deposition (Fig. 5b,i), rather than outward whisker-like growth^50,51. The lower surface area suppresses side reactions, and the resulting rSEI is over an order of magnitude thinner, at <1 μm (Fig. 5j, 0% SOC). These morphologies were reproducible (Supplementary Fig. 14).

Electrolyte consumption analysis further quantifies the reduced interfacial reactivity after activation. Under identical conditions (4 M LiFSI/DME, 60 °C), LiFSI and DME contents decreased far more slowly in the activated sample than in the untreated sample (Fig. 5k and Supplementary Fig. 15). Correspondingly, the electrolyte concentration declined from 4 M to 3.9 M after 15 cycles, compared with a more drastic decrease to 2.9 M for the untreated sample (Fig. 5l).

Gas evolution was also substantially suppressed. The untreated cell began gassing after five cycles (Fig. 5n, blue), whereas no gassing was observed in situ for the activated cell through 80 cycles (orange). Complementary cycling tests showed the untreated cell was swollen by cycle 20 (Fig. 4a), whereas the activated cell remained flat at 20 cycles and exhibited no gassing through 130 cycles (Extended Data Fig. 3f).

Extended Data Fig. 3. Comparison of anode activation methods: ACVP (orange) and CCRP (light blue), relative to untreated samples (blue) for Li‖NMC cells with 4 M LiFSI/DME.

a, Cycling performance at 60 °C, b, in situ cell volume change at 60 °C, c, rate performance at 25 °C and d, discharge capacity at 60 °C after three 0.2-0.5 C cycles and deep discharge (additional discharge steps at 7, 3.5, and 1 mA to 2.8 V). e, Photographs of cells activated using different methods after 20 cycles. The cells remain flat, indicating no gas generation. f, Table comparing different anode activation methods in terms of capacity at various points: at cycle 1, onset of gas evolution, cycle 12, and end-of-life (EOL).

Cycle life is defined as the cycle number at which sudden capacity failure (end-of-life) occurs, enabling direct comparison across varying capacity-fade behaviours. Cycle life increased from 14 cycles for the untreated cell to 140 cycles for the activated cell (Fig. 5m), an order of magnitude improvement that renders an otherwise unusable electrolyte viable. Rate performance was only minimally impacted (Extended Data Fig. 3c) and remained superior to FEE and LHCE (Fig. 4e).

Outlook for anode activation

Although ACVP yielded dramatic improvements, the considerable anode stripping and consequent over-lithiation on the cathode side led to a ~4% loss of early-time capacity (Extended Data Fig. 3d) and more rapid capacity fade (Fig. 5m), though neither X-ray photoelectron spectroscopy (XPS) nor XRD revealed detectable structural or CEI changes (Supplementary Fig. 16). On the anode side, XPS indicated that activation promoted the formation of a more anion-derived, inorganic-rich SEI (Supplementary Fig. 17), which may stem from and reinforce the improved deposition morphology (Supplementary Note 5).

To evaluate broader applicability, three additional electrolytes—a conventional carbonate and two concentrated ethers—were investigated. Following ACVP activation, all systems exhibited extended cycle lives and reduced anode-side reactivity (Extended Data Fig. 4a–c and Supplementary Note 6), demonstrating that the benefits of activation are not limited to the 4 M LiFSI/DME electrolyte. Some improvement was even observed for the FEE (1.2 M LiFSI/F5DEE).

Extended Data Fig. 4. Comparison of untreated and activated Li‖NMC cells cycled with different electrolytes.

Cycling performance of untreated (blue) and activated (orange) Li‖NMC cells cycled with a, 1 M LiPF₆/EC/DMC, b, 2 M LiFSI/DMP, c, 2 M LiFSI/DEE, d, LiFSI–1.2DME–3TTE (LHCE) and e, 1.2 M LiFSI/F5DEE (FEE). Photographs of untreated and activated Li‖NMC cells cycled at 60 °C with f, 2 M LiFSI/DMP and g, 2 M LiFSI/DEE. No gassing occurred for the activated cells.

Across these electrolytes, the benefits of reduced anode-side reactions outweigh the accompanying cathode perturbations. The more conservative CCRP method, which reduced such perturbations, did not impact the early-time capacity (Extended Data Fig. 3d, 4 M LiFSI/DME). However, less complete anode activation led to earlier gassing (30 cycles; Extended Data Fig. 3b) and earlier cell failure despite slower initial capacity fade (80 cycles; Extended Data Fig. 3a). This reinforces that suppression of anode-side interfacial reactions is more critical to performance in these systems than minimizing cathode perturbations.

Overall, ACVP, by controlling the Li deposition morphology and reducing the surface area, suppressed interfacial reactions and improved multiple performance metrics. Reduced electrolyte consumption, extended cycling performance, and delayed gas generation were achieved without altering the electrolyte formulation or cycling conditions. Anode activation therefore provides a broadly applicable strategy for enhancing safety and durability in LMBs. Synergistic use of anode activation with rationally designed materials, protocols or complementary strategies could further unlock their potential.

Conclusions

In this Article, gas evolution in ether-based LMBs has been quantified and mitigated through a combination of experimental and computational analyses. Although gas evolution was initially minimal, anode-side CH₄ production emerged at an onset point that occurred earlier under elevated temperatures or reduced electrolyte concentrations. Meanwhile, CO and CO₂ formed through cathode-related processes were consumed via crosstalk reactions. The dominant source of gas generation was DME reductive C–O bond cleavage at the Li metal anode, which produced LiCH₃ intermediates that subsequently converted to CH₄.

Guided by these mechanistic insights, two interfacial mitigation strategies were evaluated: electrolyte engineering and anode activation. Electrolyte engineering using alternative formulations—specifically LHCE and FEE—effectively suppressed gas evolution but compromised rate performance. Conversely, the anode-activation strategy pre-stripped the Li surface to induce nucleation sites, improving the subsequent deposition morphology and reducing CH₄-forming reactions while preserving the high rate performance required for demanding applications. In the 4 M LiFSI/DME system, this method reduced Li salt and DME consumption by ~83% relative to the untreated anode, delaying the number of cycles to gas onset and cell failure by approximately one order of magnitude.

These findings underscore the importance of evaluating engineering solutions against application-relevant metrics using state-of-the-art prototypes. While recent LMB research focuses on cycle life^6–14, applications such as eVTOL demand a holistic balance of high power capability and minimal gas evolution at elevated temperatures to ensure safe, reliable operation and robust emergency response^32,33,54. The anode activation strategy demonstrated here achieves this balance across multiple electrolytes, representing a step toward safe, high-performance LMBs.

Future work should explore complementary approaches, such as optimizing the conditions for Li-deficient NMC or sulfur cathodes, and benchmarking new electrolyte designs against the performance metrics established here. Ultimately, a product-oriented research framework that bridges mechanistic insight with practical deployment requirements will be essential to drive LMB commercialization and support broader electrification efforts.

Methods

Materials and cells

LiFSI (99.99%, Kaixin), DME (99.95%, Guotai), TTE (99.5%, Aladdin) and F5DEE (99%, synthesized with the assistance of Sikang Chemical) were used for the electrolyte solutions. Specifically, 2, 4 and 5 M LiFSI/DME (denoted as HCE; see Supplementary Table 14 for compositions), LiFSI–1.2DME–3TTE (by molar ratio, denoted as LHCE) and 1.2 M LiFSI/F5DEE (denoted as FEE) solutions were prepared and injected (2.1 g Ah⁻¹) into dry cells in a dry room with a relative humidity of <1%. Cell assembly and teardown were also performed in the same dry room.

Li anodes consisted of 13-μm Cu foils coated with 50 μm of Li on one or both sides. The NMC cathode was NMC811 (LiNi_0.8Mn_0.1Co_0.1O₂, a mono- and poly-crystalline mix, Rongbai). Sealed pouch cells were clamped between two metal plates covering the full electrode area and tightened to a fixed spacing corresponding to a 0.4-MPa initial stack pressure. Clamped cells were equilibrated at 45, 60 or 70 °C for 2 h before cycling.

Additional chemicals used in supplementary experiments included 1,2-diethoxyethane (DEE, 99%, Shanghai Macklin Biochemical Technology), 1,2-dimethoxypropane (DMP, 99%, Shanghai Macklin Biochemical Technology), dimethyl carbonate (DMC, 99.99%, Suzhou DoDoChem), lithium hexafluorophosphate (LiPF₆, 99.99%, Kaixin), ethylene carbonate (EC, 99.99%, Kaixin) and 3-methylsulfolane (3-MS, 98%, Shanghai Macklin Biochemical Technology). These chemicals were used to prepare 2 M LiFSI/DEE, 2 M LiFSI/DMP, 1 M LiPF₆ in 1/1 EC/DMC (wt/wt) and 1 M LiFSI/3-MS electrolyte solutions. Their ionic conductivities and dynamic viscosities are shown in Supplementary Table 15.

Li‖NMC and Li‖LFP

Li‖NMC full cells and Li‖LFP anode half-cells were fabricated on a CATL pilot line using double-sided NMC and LFP electrodes with design capacities of 140 mAh (cathode areal capacity of 3.37 mAh cm⁻²) and 150 mAh (3.61 mAh cm⁻²), respectively. After electrolyte injection, the cells were cycled at a 0.2-C charge/0.5-C discharge rate between 0 and 100% SOC (28–70 mA, 2.8–4.3 V for Li‖NMC; 30–75 mA, 2.8–3.65 V for Li‖LFP) using a constant-current–constant-voltage (CC-CV) protocol with a cutoff current of 0.1 C (14 mA for Li‖NMC and 15 mA for Li‖LFP), without formation cycles. Cycling was performed on a Neware CT-4008Tn-5V6A-S1 system (Neware Technology).

d-LFP‖NMC

Preparation of the d-LFP‖NMC cathode half-cell involved multiple steps. A Li‖LFP cell (4 M LiFSI/DME, single-sided 75-mAh LFP electrode, 3.61 mAh cm⁻²) was first cycled once at 0.1 C–0.1 C, then charged at 7 mA for 10 h to generate a controlled ‘Li vacancy’ (specifically, 70 mAh) while maintaining the LFP plateau voltage. The resulting delithiated LFP electrode was extracted, soaked for 10 min in DME, rinsed with DME, and used as the ‘anode’ for the d-LFP‖NMC cathode half-cell. The NMC cathode was designed with a capacity of 40 mAh (1.92 mAh cm⁻²) such that its usable capacity aligned with the LFP plateau voltage. Injected d-LFP‖NMC cells were cycled at 0.2 C–0.5 C (CC-CV with a cutoff current of 0.1 C) between 0 and 100% SOC (−0.63 to 0.87 V versus E_{LFP plateau} = 3.43 V), without formation cycles.

d-LFP‖LFP

The d-LFP‖LFP cell was prepared similarly. A delithiated LFP electrode (from a Li‖LFP cell, as discussed for d-LFP‖NMC) served as the ‘anode’ and was paired with a single-sided, 60-mAh LFP ‘cathode’. After electrolyte injection, the d-LFP‖LFP cell was cycled at 0.2 C–0.5 C between 0 and 100% SOC (−0.63 to 0.22 V versus E_{LFP plateau} = 3.43 V) for 30 cycles at 60 °C.

Cu‖NMC and Cu‖Li

The Cu‖NMC and Cu‖Li cells used for T–MS experiments were prepared with 13-μm Cu foil, double-sided NMC electrodes (140 mAh) and 13-μm Cu foils coated with 50 μm of Li on each side. Cu‖NMC cells were cycled 10 times at 0.2 C–0.5 C between 2.8 and 4.3 V. A deep discharge was conducted to strip the remaining Li (7 mA to 2.8 V, 3.5 mA to 2.8 V, and 1 mA to 2.8 V, with 5-min rest between each step). For Cu‖Li cells, 200 mAh of Li was deposited on the Cu electrode at 28 mA, and 125 mAh was stripped at 100 mA for the first cycle. Subsequent cycles, for a total of 10, deposited 125 mAh at 28 mA and stripped 125 mAh at 100 mA. A final deep discharge stripping step (as described for Cu‖NMC) removed the remaining Li from the Cu electrode.

Gas volume determination

Changes in the cell volume were determined using the Archimedes principle³⁶. The pouch cell (weighted to remain fully submerged) was suspended by a fine wire in a container of fluid. As the cell volume changed, the displaced volume changed accordingly, altering the buoyant force on the cell and thus the measured weight of the fluid-filled container.

Ex situ measurements

For ex situ measurements (25 °C), a beaker containing 1,000 ml of filtered water (18.2 MΩ cm at 25 °C, Thermo Scientific Barnstead NANOpure Water Purification System) was tared on an analytical balance (Shimadzu, AUW200D). After submerging the cell, the beaker system was weighed again. The immersed cell volume was calculated from the additional measured mass divided by the density of water (that is, the displaced water volume). The gas volume generated during cycling was determined by comparing measurements before and after cycling. The cell tabs were taped for insulation. Normalized gas volumes were calculated as:

$$V_{\mathrm{N}}=\frac{V_{\mathrm{total}}}{C_{\mathrm{d}}\times t}$$

where V_N is the normalized gas volume (ml Ah⁻¹ h⁻¹), V_total is the total gas volume (ml), C_d is the design capacity (Ah), and t is the total cycle time (h).

In situ measurements

In situ gas volume measurements were performed using an Archimedes-based buoyancy system (Initial Energy Science & Technology, IEST, GVM2200). The cell was submerged in silicone oil, and volume changes were recorded in real time as the cells were cycled using a Neware BTS-5V1mA testing system (Neware Technology). Cells tested at 45, 60 and 70 °C were equilibrated for 4 h before testing.

Gas component determination

Generated gases were sampled from cycled pouch cells by puncturing the aluminium–plastic pouch with a syringe inside an Ar-filled glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm). Reported gas compositions excluded N₂ (background contamination), while O₂ levels remained uncorrected due to their low abundance. The collected gas was injected into an Agilent 8890 GC instrument with an inlet temperature of 120 °C and a 10:1 split ratio. The Ar carrier gas flow rate was set to 8 ml min⁻¹. Two columns were used in series: a 5-Å molecular sieve column (CP-Molsieve 5 Å; 25 m, 0.53 mm, 50 μm) and a PLOT column (HP-PLOT Q PT; 30 m, 0.53 mm, 40 μm). These were followed by a thermal conductivity detector (TCD, Agilent) and a flame ionization detector (FID, Agilent), enabling simultaneous quantification of permanent gases (for example, H₂, O₂, N₂, CO, CO₂) and light hydrocarbons (for example, CH₄, C₂H₆, C₂H₄) in a single injection. The TCD sensitivity was optimized at 220 °C with a makeup flow of 5 ml min⁻¹ and reference flow of 30 ml min⁻¹.

A custom gas mixture of known composition was used to determine the retention times and calibrate detector responses. The mixture (Jining Xieli Special Gas) contained, by volume percentage: 8.99% CO₂, 8.98% CO, 9.02% C₂H₆, 8.93% C₂H₄, 0.497% C₂H₂, 9.02% C₃H₈, 8.99% C₃H₆, 9.24% H₂, 0.633% O₂, 8.99% CH₄ and 26.71% N₂.

In situ gas evolution monitoring via DEMS

Complementing the in situ gas volume measurements, the gas composition was quantified in situ using DEMS. Two opposing edges of the pouch cells were cut in an Ar glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm). The cut cell was then sealed within an airtight electrochemical sample holder, with screws securing the top and bottom compartments. Stack pressure was applied through a central metal plate before final tightening of the screws. O-ring sealed metal posts provided electrical contact with the cell tabs for cycling.

Inlet and outlet ports in the sample holder directed gas flow through tubing sealed with O-rings. A carrier gas system connected to the inlet port swept evolved gases towards the outlet using high-purity Ar gas (99.999%, Fuzhou Zhongming Qiti). A 2.0-μm filter was installed downstream to prevent particulates from entering the mass flow controller or mass spectrometer. A cold trap (−78.5 °C) was added to capture volatile organic species and enhance detection sensitivity. The Ar flow rate was maintained at 0.6 ml min⁻¹ to ensure high sensitivity for trace gas analysis.

MS signals were calibrated using gas standards (Jining Xieli Special Gas), enabling conversion of ion current to volume fractions via linear regression for H₂, CH₄, C₂H₂, C₂H₄, C₂H₆, O₂, CO and CO₂. The time-dependent volume fractions (VFᵢ) obtained were smoothed using a Savitzky–Golay filter (window length = 21, polynomial order = 2) and converted into gas flow rates Qᵢ (ml min⁻¹) for each species using the Ar carrier flow rate (0.6 ml min⁻¹):

$$Q_{\mathrm{i}}=0.6\times ({\mathrm{VF}}_{\mathrm{i}}/{\mathrm{VF}}_{\mathrm{Ar}})$$

where VF_Ar = 100% − ΣVFᵢ is the calculated Ar volume fraction. The baseline drift arising from the instrument background and residual gas signals was removed using an asymmetric least-squares algorithm with species-specific smoothness parameters (λ = 10⁶–10¹²) and asymmetry weight p = 0.01. Accumulated gas volumes (ml) were obtained by trapezoidal integration of the baseline-corrected flow rates over time. All processing was performed in Python using NumPy, SciPy and pandas.

Titration–mass spectrometry

Crosstalk validation

To quantify the solid Li₂CO₃ crosstalk product on the cycled Li anodes, T–MS was performed (Shanghai LingLu Instruments, QAS 100). The extracted Li anodes were titrated with 10 M H₂SO₄ to evolve CO₂ gas from Li₂CO₃ according to the following reaction:

$${\mathrm{Li}}_2{\mathrm{CO}}_3+{\mathrm{H}}_2{\mathrm{SO}}_4\to {\mathrm{Li}}_2{\mathrm{SO}}_4+{\mathrm{H}}_2\mathrm{O}+{\mathrm{CO}}_2\uparrow$$

As a safety precaution to prevent the violent reaction between metallic Li and H₂SO₄, a Cu‖NMC configuration was utilized as the ‘full cell’, substituting the Li anode with a Cu electrode. Any Li deposited on the Cu electrode was fully stripped prior to cell disassembly. For the Li‖NMC verification experiment (Supplementary Fig. 4), the cycled Li anode was subjected to an over-discharge protocol to remove both deposited and bulk Li before titration. Fresh Cu and d-LFP electrodes were also tested to establish baseline background signals (Supplementary Fig. 4) in comparison to control systems designed to exhibit no crosstalk, including cycled d-LFP‖NMC and Cu‖Li cells.

A calibration curve for Li₂CO₃ (Supplementary Fig. 4) was established by titrating a mass series of Li₂CO₃ standard samples (99.5%, Macklin). Accordingly, the mass of Li₂CO₃ on the experimental electrodes was quantified based on the integrated CO₂ signal.

Reaction intermediate validation

To demonstrate the existence of a LiCH₃ intermediate, T–MS experiments were conducted following a procedure similar to that used for the titration of Li₂CO₃. The reaction of LiCH₃ with the H₂SO₄ titrant proceeds as follows:

$$2\mathrm{Li}{\mathrm{CH}}_3+{\mathrm{H}}_2{\mathrm{SO}}_4\to {\mathrm{Li}}_2{\mathrm{SO}}_4+{2\mathrm{CH}}_4\uparrow$$

To determine where LiCH₃ accumulates, the anode and the electrolyte extracted from cycled cells disassembled in an Ar glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm) were titrated separately. The fresh anode was also titrated as a baseline, confirming that LiCH₃ appeared at the anode only after cycling.

CO₂ solubility measurement

CO₂ solubility was determined via the Archimedes principle, adapting a methodology similar to ex situ gas volume measurements. A Mylar pouch, configured to replicate a standard pouch cell geometry, was filled with 10 g of 4 M LiFSI/DME electrolyte, heat-sealed and clamped. Subsequently, 10 ml of CO₂ gas (99.9%, Luoyuan Shunfa Industrial Gases) was injected into the electrolyte-filled pouch via a syringe, and the injection site was promptly sealed. The pouch was fully submerged into a beaker of deionized (DI) water, and the initial mass of the beaker system was immediately recorded. To ensure complete CO₂ saturation within the electrolyte, the pouch was equilibrated at 25 °C for 24 h. Following equilibration, the beaker system was re-weighed. The volume of dissolved CO₂ was calculated based on the mass difference, yielding a consumed volume of 3.22 ml. Thus, the CO₂ solubility in 4 M LiFSI/DME was determined to be 0.32 ml g⁻¹.

Electrolyte consumption analysis

Electrolyte consumption analysis was performed to quantify the loss of LiFSI and DME with cycling. LiFSI was quantified using IC and DME by GC. Both LiFSI and DME were extracted from the cell using an extraction agent. Hence, the complete method was termed E-G&IC¹⁵.

Preparation of extraction agent

An internal standard, DEE (99.9%, Aladdin), was added to correct for matrix effects and sample loss during preparation for GC analysis. The internal standard was added to the sample along with diethylene glycol dimethyl ether (diglyme, 99.99%, Aladdin) in the form of an extraction agent, which was prepared by dissolving 4 g of DEE in diglyme, and making up the solution to 200 ml in a volumetric flask. The stability of the extraction agent towards lithium and other cell components was demonstrated in previous work¹⁵.

Electrolyte extraction

A small incision was made in the experimental cell inside an Ar glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm). After injecting 5 ml of the extraction agent, the cell was immediately re-sealed and gently agitated. The cell was allowed to rest at room temperature for seven days for thorough extraction. The liquid content was then extracted from the cell and syringe-filtered for further processing.

Ion chromatography for quantifying LiFSI

Six standard LiFSI solutions with 0, 25, 50, 100, 150 and 200 mg LiFSI were prepared by dissolving the respective amounts in 1,000 ml of DI water, then 5 ml aliquots were analysed with ion chromatography (IC; Dionex Aquion RFIC, Thermo Scientific). Linear regression was performed between the measured peak areas and LiFSI concentrations, forming a calibration curve (Supplementary Fig. 18).

Samples extracted from experimental cells were diluted with DI water using a dilution factor (DF) of 200. A 5-ml aliquot of the diluted solution was measured with IC. The concentration of the diluted sample (c_LiFSI) was determined from the measured peak area (S_LiFSI):

$$S_{\mathrm{LiFSI}}=k_{\mathrm{LiFSI}}\times c_{\mathrm{LiFSI}}$$

where k_LiFSI is the slope of the calibration curve. To obtain the absolute mass of LiFSI (m_LiFSI) remaining in the cycled cell, the concentration of the diluted sample was multiplied by the dilution factor (DF = 200) and the total volume of the extracted solution:

$$m_{\mathrm{LiFSI}}=c_{\mathrm{LiFSI}}\times 200\times 5.245\mathrm{ml}$$

where 5.245 ml represents the sum of 5 ml of extraction agent and 0.245 ml for the initial 0.3 g of 4 M LiFSI/DME.

Gas chromatography for quantifying DME

A standard solution was prepared by adding 0.3 g of fresh electrolyte (4 M LiFSI/DME) to 5 ml of extraction agent. This solution was then diluted with diglyme with DF = 5. A 1.5-ml aliquot was measured by GC (Nexis GC-2030, Shimadzu). A response factor, f_DME, was calculated for DME using the known masses of DME and DEE (m_DME, m_DEE) in the standard solution and the peak areas extracted from the GC measurement (S_DME, S_DEE):

$$f_{\text{DME}}=\frac{m_{\text{DME}}/S_{\text{DME}}}{m_{\text{DEE}}/S_{\text{DEE}}}$$

Samples extracted from experimental cells were diluted with diglyme with DF = 5. A 1.5-ml aliquot of the diluted solution was measured with GC. To obtain the absolute mass of DME (m_DME,exp) remaining in the cycled cell, the ratio of the obtained peak areas (S_DME,exp, S_DEE,exp) were multiplied by the mass of the DEE added (m_DEE,exp) in the form of the extraction agent, and the response factor (f_DME):

$$m_{\text{DME},\exp }=m_{\text{DEE},\exp }\times f_{\text{DME}}\times \frac{S_{\text{DME},\exp }}{S_{\text{DEE},\exp }}$$

First-principles simulation

Density functional theory (DFT) calculations were performed with the Vienna ab initio simulation package (VASP)^55,56. The generalized gradient approximation and the Perdew–Burke–Ernzerhof functional within the DFT framework were applied to determine the electron exchange-correlation energies⁵⁷. Transition metals were treated using the DFT + U augmented approach with U values of 4, 4.4 and 5 eV for Mn, Co and Ni, respectively⁵⁸. Weak interactions were accounted for using the DFT + D3 method with dispersion correction. All calculations were spin-polarized with a plane-wave cutoff energy of 520 eV. Surface calculations were performed with a 2 × 2 × 1 Monkhorst–Pack k-point scheme. A 15-Å vacuum layer was introduced to avoid interactions between periodic slabs.

A five-layer slab of lithium (110) was used to investigate the DME reduction processes, and a charged slab of NMC811 was used to investigate the DME oxidation processes. Geometric structural optimization was performed until the force on all atoms was less than 0.02 eV Å⁻¹, with an energy convergence criterion set to less than 10⁻⁵ eV atom⁻¹. Transition states were located using the climbing image nudged elastic band (CI-NEB)⁵⁹ and dimer methods⁶⁰, with only one imaginary vibrational frequency being identified for each transition state along the reaction coordinate.

The ground-state molecular and ion geometries were optimized using the DFT method at the B3LYP/6-311G+(d,p) level. All DFT calculations were performed using the Gaussian 09 software package. The partial charges on the atoms of the solvent were computed by fitting the molecular restrained electrostatic potential (RESP) at the atomic centres using the B3LYP/aug-cc-pVDZ basis set⁶¹.

Anode activation

Anode constant-voltage pre-stripping

To activate the anode, an electrolyte-filled Li‖NMC cell was over-discharged to 1.5 V (over-discharge) at 25 °C, and held at 1.5 V with a cutoff current of 13 mA. To determine the optimal activation voltage, a series of potentiostatic discharge experiments (1.3 to 2.1 V) were performed (Supplementary Fig. 19). Optimal cycling stability and delayed gas evolution were achieved in the range of 1.5–1.7 V. We standardized the process at 1.5 V to ensure sufficient activation while maintaining the observed improvements.

Cathode charging-reconstitution and pre-stripping

An NMC cathode with 5% lithium vacancy was prepared via controlled pre-cycling and charging: a Li‖NMC cell was subjected to a single 0.2-C (28 mA to 4.3 V) charge and 0.2-C (28 mA to 2.8 V) discharge cycle at 25 °C. The cell was then charged at 0.02-C (2.8 mA) current for 2.5 h to 5% SOC (7 mAh), yielding a 5% delithiated cathode. This cathode was extracted in a dry room, soaked in DME for 3 min, and briefly air-dried. A reconstituted cell was assembled using this cathode and a fresh Li anode. To emulate the constant-voltage discharge protocol in a more controlled manner, the discharge current was stepped down from 1 C (140 mA) to 0.2 C (28 mA) in increments of 0.2 C, and finally to 0.1 C (14 mA) (Supplementary Fig. 13), with 2-min rest intervals between each step. Each discharge step, performed at 25 °C, was terminated at a less-aggressive cutoff voltage of 2.8 V.

SEM characterization

SEM images of the anode samples were acquired with an ultrahigh-resolution cold field-emission SEM system (Hitachi SU8600). Before imaging, anodes extracted from cycled cells were washed and dried three times in a dry room, with a 10-min DME soak each time. To prepare a cross-sectional sample, a narrow strip of washed anode was quenched in liquid nitrogen for 60 min, then swiftly split on a stainless-steel blade to expose a clean cross-section on each of the split pieces.

Extracted cathode samples were washed and dried in the same manner as the anode. A small piece of the sample was cut and transferred to an ion beam cross-section polisher (CP, Hitachi ArBlade 5000) under a dry atmosphere and polished at room temperature under inert conditions to expose a clean cross-section for imaging.

XRD characterization

XRD measurements were carried out using a Bruker D8 X-ray diffractometer over a 10° to 90° 2θ range at a scanning rate of 2° min⁻¹. Electrode samples were extracted from cycled cells in an Ar glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm), soaked in a Petri dish filled with DME for 5 min, and gently agitated with tweezers to remove Li salt and residual electrolyte. The electrodes were then dried in the glovebox, cut into 1 × 1-cm² pieces, and sealed in Mylar bags for transfer to the XRD instrument.

XPS characterization

XPS measurements were carried out using a Shimadzu Axis Supra+ imaging X-ray photoelectron spectrometer. An Al Kα source (1,486.7 eV) was used for excitation, and a hemispherical electron energy analyser was used to collect data over a 700 × 300-µm² area at an emission power of 120 W. Electrode samples were extracted from cycled cells in an Ar glovebox (H₂O < 0.1 ppm, O₂ < 0.1 ppm), soaked in a Petri dish filled with DME for 5 min, and gently agitated using tweezers to remove Li salt and residual electrolyte. The electrodes were then dried in the glovebox, cut into 1 × 1-cm² pieces, sealed in glass scintillation vials with parafilm and additionally heat-sealed in Mylar bags for transfer to the XPS laboratory. Samples were loaded onto the XPS sample holder in another Ar glovebox.

In the C 1s spectrum, C–C, C–O, C=O and Li₂CO₃ signals were detected at 284.8 eV, 286.2 eV, 288.4 eV and 290.8 eV, respectively. The C–C peak was used to calibrate the XPS spectrum. In the F 1s spectrum, S–F and LiF signals were detected at 685.0 eV and 688.7 eV, respectively. In the S 2p spectrum, S⁴⁺ and S⁶⁺ species signals were detected at 167.6 eV and 169.7 eV, respectively. In the O 1s spectrum, M–O, RCO₂Li/Li₂CO₃ and C–O–C signals were detected at 529.2 eV, 531.7 eV and 533.0 eV, respectively.

Ionic conductivity and viscosity

Electrolyte solutions for ionic conductivity and dynamic viscosity measurements were prepared in a dry room and sealed in glass scintillation vials with parafilm for transfer to a separate laboratory.

Ionic conductivity was measured using a benchtop conductivity meter (LeiCi, DDSJ-318) at 25 °C under ambient atmospheric conditions. A 10-ml aliquot of the solution was transferred into a 50-ml Falcon tube, sealed and equilibrated in a 25 °C water bath for 10 min. Meter calibration was verified using low- and high-conductivity standard solutions. The probe head was cleaned with ethanol and deionized water between measurements.

Dynamic viscosity was measured with a Brookfield DV2T viscometer equipped with an SC4-18 spindle at 25 °C under ambient atmospheric conditions. After levelling and auto-zeroing the instrument, an 8-ml aliquot of the solution (sufficient to fully submerge the spindle) was transferred to the sample holder and equilibrated at 25 °C for 10 min. Measurements were performed with periodic stirring.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-026-02219-1.

Author contributions

S.T.H., Y.W., Z.C., H.W., X.L., N.L. and C.O. conceived the study. H.W., Y.W. and S.T.H. designed the experiments. Z.C. and J.H. conducted the cycling tests and electrolyte consumption measurements. Y.W., F.C., H.L., S.T.H. and Z.S. developed and conducted the DEMS experiments. Z.C., Y.W. and H.W. developed the anode activation methods. C.W., Y.W. and F.Z. conducted XPS, XRD and SEM characterizations. H.L. advised on all characterization and testing. J.S., E.W., L.B. and J.L. performed the first-principles calculations, molecular dynamics simulations and temperature simulations. X.Y. conducted the titration–mass spectrometry experiments. S.T.H. organized, wrote and revised the paper with input from all authors. C.W., Y.W., Z.C., U.U. and H.W. further revised the paper. H.W. and C.O. supervised the project and secured funding. All authors discussed the results and commented on the paper.

Peer review

Peer review information

Nature Chemistry thanks Mei Cai, Tianye Zheng and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Funding

H.W. acknowledges support from the National Natural Science Foundation of China (grant no. 22209085) and the Fujian Provincial Natural Science Foundation of China (grant no. 2024J010048). C.O. acknowledges support from the National Natural Science Foundation of China (grant no. 51962010). This work was also supported by Contemporary Amperex Technology Co., Ltd (CATL).

Data availability

The data that support the findings of this study are available from the corresponding authors upon reasonable request.

Competing interests

All authors are employees of Contemporary Amperex Technology Co., Ltd (CATL). Z.C., H.W. and Y.W. are inventors on patent application PCT/CN2025/100273, filed by CATL covering the anode activation method reported in this article.

Extended data

is available for this paper at https://doi.org/10.1038/s41557-026-02219-1.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41557-026-02219-1.