Singlet oxygen from cation driven superoxide disproportionation and consequences for aprotic metal–O₂ batteries

Abstract

Aprotic alkali metal–oxygen batteries require reversible formation of metal superoxide or peroxide on cycling. Severe parasitic reactions cause poor rechargeability, efficiency, and cycle life and have been shown to be caused by singlet oxygen (¹O₂) that forms at all stages of cycling. However, its formation mechanism remains unclear. We show that disproportionation of superoxide, the product or intermediate on discharge and charge, to peroxide and oxygen is responsible for ¹O₂ formation. While the overall reaction is driven by the stability of peroxide and thus favored by stronger Lewis acidic cations such as Li⁺, the ¹O₂ fraction is enhanced by weak Lewis acids such as organic cations. Concurrently, the metal peroxide yield drops with increasing ¹O₂. The results explain a major parasitic pathway during cell cycling and the growing severity in K–, Na–, and Li–O₂ cells based on the growing propensity for disproportionation. High capacities and rates with peroxides are now realized to require solution processes, which form peroxide or release O₂via disproportionation. The results therefore establish the central dilemma that disproportionation is required for high capacity but also responsible for irreversible reactions. Highly reversible cell operation requires hence finding reaction routes that avoid disproportionation.

---

Broader context

Decarbonizing the energy system requires energy storage with large capacity but equally low economic and ecological footprint. Alkali metal–O₂ batteries are considered outstanding candidates in this respect. However, they suffer from poor cycle life as a result of cathode degradation. Formation of the highly reactive singlet oxygen has been proposed to cause this degradation, but formation mechanisms have remained unclear. Here, we show that the singlet oxygen source is the disproportionation of thermodynamically unstable superoxide intermediates to the peroxides. The revealed mechanism conclusively explains the strongly growing degree of degradation when going from K–O₂ to Na–O₂ and Li–O₂ cells. A major consequence is that highly reversible cell operation of Li–O₂ and Na–O₂ cells requires them to form and decompose the peroxides without disproportionation. Achieving this requires finding new reaction routes. The work lays the mechanistic foundation to fight singlet oxygen as the predominant source of degradation in metal–O₂ cells.

Introduction

Advancing electrochemical storage beyond the limits of current batteries has become the focus of much cutting-edge research and hence has caused immense interest in rechargeable non-aqueous alkali metal–O₂ batteries. They operate by reversibly forming/decomposing superoxides or peroxides of Li, Na, or K at a porous cathode according to

O₂ + xe⁻ + xM⁺ ↔ M_xO₂ (M = Li, Na, K) (1)

The products that typically form are Li₂O₂, Na₂O₂, NaO₂, or KO₂.^1–5 Practically realizing such cells faces, however, two major barriers. First, these (su)peroxides are insulating solids that passivate the electrode and lead to low capacities. Large capacities and high rates are now realized to require solution processes on both discharge and charge, which may be favored by solvating additives or mediators.^2,6–13 Second, severe parasitic reactions decompose the electrolyte and electrode and cause high charging voltage, poor reversibility and cycle life.^{1–3,7,14–25}

The parasitic reactions cause deviations from the ideal cell reaction in eqn (1). Key measures for parasitic chemistry are the ratio of O₂ consumed/evolved and peroxide or superoxide formed/decomposed per electron passed on discharge/charge. Parasitic reactions form significant amounts of side products such as alkali carbonate, carboxylates, or CO₂.^{2,3,13,15,17,20–22,26,27} The severity of parasitic chemistry increases in the order K–O₂, Na–O₂, and Li–O₂ with typical (su)peroxide yields of ∼98–100%, ∼90–95%, and 50–90%, respectively, and similar O₂ yields on recharge.^{3,5,10,13,15,17,20–22,26–29} Peroxide rather than superoxide as the product increases the severity, particularly on charge, where the voltage climbs inexorably due, in large parts, newly formed parasitic products.^{15,17,20–22,25,30,31}

The parasitic reactions have long been predominantly ascribed to the direct reactivity of electrolyte or carbon with superoxides and peroxides owing to their basicitiy, nucleophilicity, or radical nature.^{2,3,15,17,19–22,27,28} Nevertheless, these reactivities fail to conclusively explain the mentioned pattern of parasitic reactions. Specifically, the extent of side reactions would suggest the reactivity to seemingly severely grow in the order KO₂ < NaO₂ < LiO₂, and superoxides to be less reactive than peroxides, which opposes chemical intuition suggesting KO₂ to be the most reactive. KO₂ can, however, cycle highly reversible as recently shown by Lu et al.,⁵ which forcefully demonstrates that other degradation pathways than superoxide attack must prevail. Only recently, the highly reactive singlet oxygen (¹O₂ or ¹Δ_g), the first excited state of ground state triplet oxygen (³O₂ or ³Σ_g⁻), has been revealed to form upon cycling in Li–O₂ and Na–O₂ cells and to predominantly cause the side reactions.^32–341O₂ forms during discharge, rest, and from the onset of charge at rates that match the rates of parasitic chemistry occurring in cells. How ¹O₂ forms is unclear but must be deeply rooted in the way (su)peroxides form or decompose.

Discharge commences with O₂ reduction to superoxide (MO₂). Whether it further reacts to the peroxide via a second electrochemical 1 e⁻ transfer

MO₂ + e⁻ + M⁺ → M₂O₂ (2)

or disproportionation

2MO₂ → M₂O₂ + O₂ (3)

is governed by the relative thermodynamic stability of peroxide and superoxide with the respective cation as illustrated in Fig. 1a. Strong Lewis acids such as Li⁺ or Na⁺ favour the peroxide versus the superoxide, albeit only slightly for sodium.^{4,14,28,31,35,36} K⁺ and even weaker Lewis acids (e.g., quaternary amines like tetrabutylammonium (TBA⁺) and imidazoliums favour the superoxide.^10,37–39 The latter constitute often used ionic liquid electrolytes. Superoxide disproportionation (eqn (3)) is now accepted to be involved on discharge and charge of the peroxides.^{1,2,9,30,36,37,40–43} For example, Li₂O₂ oxidation during charge commences with surface delithiation to form Li_2−xO₂ species or soluble superoxide, which release O₂ by disproportionation.^1,40,42,44 Large capacities require solution processes that favour the second electron transfer to/from peroxide to pass via disproportionation.^{7–9,11,36,40,41} Pathways towards ¹O₂ in this environment are unclear. Only better knowing the formation mechanism may allow finding strategies to inhibit ¹O₂ formation, which is indispensable for progress towards fully reversible, high capacity metal–O₂ cells.

Fig. 1 Thermodynamics of alkali peroxides and superoxides and the Lewis acidity of the here used cations. (a) Standard potentials of the O₂/MO₂ and O₂/M₂O₂ redox couples on the M/M⁺ scales with M = Li, Na, K as well as for the O₂/H₂O₂ couple. The scales are brought to a common scale based on their M/M⁺ standard potentials. The dashed horizontal line indicates the O₂/KO₂ couple. The O₂/LiO₂ potential is adopted from ref. 45, but also values between 2.29 and 2.46 V vs. Li/Li⁺ have been reported.^28,43 O₂/O₂⁻ denotes the potential range reported for O₂ reduction in TBA⁺ electrolytes.^{9,37,41,45,46} (b) Schematic Lewis acidity order of the used cations.

Here, we show that superoxide disproportionation in aprotic media releases significant fractions of ¹O₂ and we derive mechanistic descriptors for ¹O₂vs.³O₂ release backed by simulations. While the strong Lewis acids Li⁺ and Na⁺ stabilize peroxide versus superoxide and drive the overall reaction, the ¹O₂ fraction is higher with Na⁺, the weaker Lewis acid. Also present even weaker Lewis acids enhance the ¹O₂ fraction massively. Larger ¹O₂ fractions go along with smaller peroxide yield. The results explain a major degradation pathway, explain the growing parasitic chemistry in K–, Na–, and Li–O₂ cells based on the growing propensity for disproportionation, and show that counteracting ¹O₂ formation requires finding reaction routes that avoid superoxide disproportionation. Given that large capacities and rates require solution processes that rely on disproportionation steps, the results establish a central dilemma for high capacity metal–O₂ cells.

Experimental

Materials

Salts contained either the bis(trifluoromethane)sulfonimide (TFSI⁻) or ClO₄⁻ anion. These anions have similar donor numbers and exert therefore minor changes on the considered solution equilibria^47,48 and an analogous effect of TBA⁺ addition as confirmed in Fig. S1 (ESI†). LiTFSI, NaClO₄, KClO₄, TBATFSI were dried under vacuum for 24 h at 80 °C. Dimethoxyethane (DME) and tetraethylene glycol dimethyl ether (TEGDME) were dried over lithium, distilled and further dried and stored over activated molecular sieves. The water content as measured by Karl–Fischer titration was below 5 ppm. 9,10-Dimethylanthracene (DMA) was recrystallized from ethanol. Lithium peroxide (Li₂O₂) was synthesized as described previously.⁴⁹ Its purity was confirmed by XRD, FTIR spectroscopy, and carbonate/carboxylate analysis.⁵⁰

Electrochemical methods

Metal–O₂ cells with integrated pressure transducer were of the type PAT-Cell-Press from EL-Cell GmbH (Hamburg, Germany) with custom modified cathode plunger as described earlier.⁵¹ Electrochemical tests were run on a potentiostat/galvanostat (SP-300 or MPG-2, Bio-Logic). Free standing carbon/PTFE electrodes were made from a slurry of Super P carbon/PTFE binder (9/1, w/w) using isopropanol. Li_1−xFePO₄/C black/PTFE (8/1/1, m/m) counter electrodes were prepared analogously. For the Li₂O₂/C/PTFE (1/8/1, m/m) electrodes, Li₂O₂ was first ground with Super P (1/9, m/m) for 1.5 h in a planetary ball mill (Pulverisette 7, Fritsch) at 200 rpm with ZrO₂ grinding balls under Ar. ATR-FTIR and XRD confirmed purity thereafter. A Super P/PTFE mixture (1/1, m/m) was made with isoproanol and dried under vacuum at 120 °C. Then, Li₂O₂/C and C/PTFE powders were mixed and pressed onto steel grids. Celgard separators and the electrodes were first washed with isoproanol and water (1/1, v/v) and subsequently with acetone. Electrodes and separators were dried under vacuum at 120 °C for 24 h. The counter electrode had three-fold the expected capacity of the working electrode. Typical working electrodes had a carbon mass loading of 1 mg and the cells were assembled with 100 μL electrolyte. Before discharge, cells were purged with high-purity O₂ (N5.0).

Spectroscopic methods

The mass spectrometry (MS) setup was built in-house and described previously.^33,50 The sample setup consisted of a glass vial with a volume of 7 mL equipped with a stirring bar. A PEEK plug with glued-in PEEK tubes and an exchangeable septum is sealed against the glass vial with a flat rubber seal. Reagents were added through a septum using a gas-tight syringe (Hamilton). All solutions were degassed with N₂ to remove dissolved CO₂ and O₂. The headspace was purged to the MS using 5 mL min⁻¹ high purity Ar 6.0. To measure the rate of O₂ evolution during the disproportionation reaction, a high-precision pressure transducer (Omega, PAA35X) was connected to the closed vessel instead of the MS. Reagents were added with a gas-tight syringe through glued-in tubing.

High-performance liquid chromatography (HPLC) was used to determine the degree of the DMA-to-DMA-O₂ conversion as described earlier.³³ From chemical experiments, the filtered electrolyte was diluted with DME to ∼1 mg_DMA mL⁻¹. From cells, the electrolyte was extracted from all cell components using 400 μL DME, sonicated for 10 min under exclusion of light and heat, centrifuged and the supernatant was transferred and DME removed under a N₂ stream at room temperature. The residue was dissolved in 500 μL DME and a volume of 2 μL was injected into the HPLC.

The amount of peroxide in a sample was measured by UV-vis spectroscopy of the Ti(IV)-peroxo complex in combination with mass spectrometry as described previously.⁵⁰ The acidic environment also evolves CO₂ from carbonates which was measured by MS. The samples in Fig. 2 and 3 from which Li₂O₂ and CO₂ yield were obtained were prepared separately to the ones for ¹O₂ and ³O₂ yield since DMA is incompatible with the Ti(IV) peroxo complex. Measurements given in bar graphs are from typically three or more repetitions. Repeatability is illustrated in Fig. S4 (ESI†).

Computational methods

Energies were calculated for solvated species with a solvent dielectric constant of 7.28 (1,1,2-trichloroethane, a value close to short chain glymes, like DME) by density functional theory (DFT) calculations by adopting a computational approach validated previously and benchmarked on post-Hartree–Fock calculations.¹⁶ The M06-2X functional and the 6-31++G(d,p) basis set (unrestricted)⁵² was used and solvation effects incorporated using a self-consistent reaction field (SCRF) in continuum solvation model C-PCM.⁵³ The final computational accuracy for the reaction energies that do not involve the ¹O₂ species is estimated to be 0.05 eV. The pure O₂ (¹Δ_g) molecule computed at the unrestricted M062X level shows unsatisfactory geometry and frequencies, similarly to the B3LYP functional, due to the well-known spin-contamination problem.⁵⁴ This unavoidable computational limitation at DFT level leads to an underestimate of the ³Σ_g⁻ → ¹Δ_g energy difference and thus to a worse computational accuracy for all ¹O₂ release reactions, estimated to 0.1–0.15 eV. All structures were relaxed to their energy ground state and vibrational stability checked for all the reported reagents, intermediates, and products. The Gibbs energy of each molecular/ionic species was calculated at 298 K by considering zero-point energies and thermal effects. All calculations were done using Gaussian16.⁵⁵ Superoxide dimers we checked for all symmetric and asymmetric cases for all four conformers suggested by Bryantsev et al.⁵⁶ and reported values are for the most stable ground state structures. The structures are shown in Fig. S2 and S3 (ESI†). The reaction energy for the precipitation to solid peroxides was calculated with thermochemical cycles starting from DFT calculations, the assessed thermodynamic properties of solid phases and for neutral atoms in the gas phase.⁵⁷ The thermodynamics of the TBA⁺O₂⁻ ion couple was calculated at the same level of theory by relaxing the solid ionic couple in the simulated solvent to a ground state minimum.⁵⁸

Results and discussion

Probing singlet oxygen from superoxide disproportionation

We studied the disproportionation reaction

2O₂⁻ → O₂²⁻ + x³O₂ + (1 − x)¹O₂ (4)

in presence of the cations shown in Fig. 1b that cover a wide range of Lewis acidity. These are the alkali cations Li⁺, Na⁺, and K⁺ as well as H⁺ from common protic electrolyte impurities and organic cations that are common constituents in ionic liquid electrolytes. Next to tetrabutylammonium (TBA⁺) we also used 3-ethyl-1-methyl-1H-imidazol-3-ium (EMIm⁺) and 3-ethyl-1,2-dimethyl-1H-imidazol-3-ium (EM₂Im⁺) as organic cations since they were used as cations in ionic liquid electrolytes and the Lewis acidity of EMIm⁺ and EM₂Im⁺ covers a range between TBA⁺ and K⁺.^19,37,39 As superoxide source we used both KO₂ as a chemical source and the electrochemical reactions during cycling of Li–O₂ cells.

As a bimolecular reaction, superoxide disproportionation passes via M(O₂)₂M dimers (with M being any of the cation in Fig. 1b).^35,41,56 We hypothesize that the energetics of pathways to ³O₂ and ¹O₂ will sensitively depend on the cations involved. Therefore, to learn about the reaction mechanism, we intentionally influence the intermediates by using, next to pure Li⁺ or Na⁺ electrolytes, also their mixtures with TBA⁺ that itself would not drive disproportionation; the overall driving force to Li₂O₂ or Na₂O₂ remains unchanged while an asymmetric M(O₂)₂TBA intermediate dimer can be expected to be destabilized due to weaker O₂⁻–TBA⁺ than O₂⁻–M⁺ interactions^9,37,45 and hence to change the energetics and relative yields of ³O₂ and ¹O₂ evolution.

To probe for ¹O₂, we used 9,10-dimethylanthracene (DMA) as a chemical trap that fulfils the requirements for the non-aqueous (electro)chemical environment: it selectively forms the endoperoxide (DMA-O₂) in contact with ¹O₂. DMA and DMA-O₂ can be quantified by HPLC as detailed in the Methods, are electrochemically stable in the required potential range, and are stable towards superoxide and peroxide.^33,34

Disproportionation of chemically produced superoxide

To probe for ³O₂ and ¹O₂ yields from superoxide disproportionation, we first brought solid KO₂ in contact with Li⁺, Na⁺, K⁺, and TBA⁺ electrolytes based on tetraethyleneglycol dimethylether (TEGDME) that also contained DMA, Fig. 2a. TEGDME was used since it is a common solvent for metal–O₂ cells.^1,2,9,13 The reaction was done in a closed vessel with the head space continuously purged to a mass spectrometer (MS) for gas analysis and the DMA-to-DMA-O₂ conversion measured at the end of the experiment. When KO₂ was brought in contact with the Li⁺ electrolyte, the O₂ concentration rose sharply and ceased within 2 h, which indicates disproportionation as reported before.^9,23,41 Quantifying the total O₂ reveals that the KO₂ has nearly quantitatively reacted and has resulted in ∼93% ³O₂ and ∼2% ¹O₂ of the total O₂ amount expected from eqn (4), i.e., 1 mol O₂ per 2 mol KO₂. KO₂ in Na⁺ electrolyte equally resulted in disproportionation as reported recently.³⁶ We found a continuous reaction, which does not come to completion within 2 h. The lower rate is in accord with the lower driving force (Fig. 1a) and the total O₂ after this time shows that ∼8% of the KO₂ have reacted of which 12% resulted in ¹O₂. KO₂ in contact with K⁺ and TBA⁺ electrolyte did not evolve an appreciable amount of O₂ as expected. These results show that superoxide disproportionation in presence of alkali cations yields significant fractions of ¹O₂ with its fraction increasing as Lewis acidity of the cations decreases.

Fig. 2 ¹O₂ from superoxide disproportionation in presence of various cations. (a) O₂ evolution versus time upon mixing KO₂ with TEGDME electrolytes containing 0.1 M of the indicated cations and 30 mM 9,10-dimethylanthracene (DMA). The inset shows the evolved ³O₂ (as measured by MS) and ¹O₂ (as measured as DMA-O₂ by HPLC) after 2 h reaction time. (b) Obtained O₂, ¹O₂, and Li₂O₂ (or Na₂O₂) upon reacting KO₂ in TEGDME that contained equimolar 18-crown-6, 30 mM DMA, 0.5 M Li⁺ (or Na⁺), and either no additive, 0.1 M TBA⁺, EMIm⁺, or EM₂Im⁺, or F₃CCOOH. The scale means mol of O₂, ¹O₂, Li₂O₂, or CO₂ per 2 mol of KO₂. I.e., ideally 1 mol O₂ and 1 mol M₂O₂ would form according to 2KO₂ + 2M⁺ → M₂O₂ + O₂ + 2K⁺. Error bars are given in Fig. S4 (ESI†).

To avoid the complexities of reactions at a solid, we further investigated the disproportionation of KO₂ solvated by the crown ether 18-crown-6 (1,4,7,10,13,16-hexaoxacyclooctadecane). Additionally to ³O₂ and ¹O₂, we also measured the Li₂O₂ or Na₂O₂ yield, respectively, after O₂ evolution ceased using established procedures with photometry of the [Ti(O₂)OH]⁺ complex after adding acidic TiOSO₄ solution, which also evolves CO₂ from formed carbonate.^4,50 The CO₂ amount serves as a proxy for the amount of side products. We added either pure Li⁺ electrolyte or combination of Li⁺ with H⁺, TBA⁺, EMIm⁺, or EM₂Im⁺. We further added pure Na⁺ or Na⁺/TBA⁺ electrolytes. The results are shown in Fig. 2b. The result with pure Li⁺ resembles the one with solid KO₂ in Fig. 2a; the ³O₂ and Li₂O₂ yields were ∼93% and the ¹O₂ ∼2%. With F₃CCOOH as H⁺ source we found ∼91% for ³O₂ and peroxide yield and 3% for ¹O₂ yield and thus vanishingly more ¹O₂ than without acid. This is in accord with reported minor ¹O₂ yields from proton assisted superoxide disproportionation in Li-free media.^59,60

Mixtures of Li⁺ and the weakly Lewis acidic organic cations, however, increase ¹O₂ very strongly; the ³O₂ and Li₂O₂ yields dropped to ∼80–85% and the ¹O₂ rose to ∼10–20%. Carbonaceous side products as indicated by CO₂ evolution also rose similarly. Adding weak Lewis acids into the disproportionation reaction not only raised the ¹O₂ yield massively, but concurrently boosted the reaction rates. We measured the ³O₂ evolution kinetics from superoxide disproportionation by means of the pressure rise in a closed reaction vessel (Fig. S5, ESI†). Values compared to the kinetics with Li⁺ alone are 5-fold with EMIm⁺ and TBA⁺ and 8-fold with EM₂IM⁺. The mechanistic implications of this finding are discussed later together with the theoretical results.

Given that organic cations provoke high ¹O₂ amounts, we assessed their stability in the system. Tetraalkylammoniums have been shown previously to be stable with superoxide.⁴¹ We probed whether imidazoliums would be reactive with superoxide or ¹O₂ and whether they would quench the latter and thus reduce the DMA-O₂ yield, which then would underestimate the ¹O₂ yield. When EMIm⁺ and EM₂Im⁺ were exposed to KO₂ in TEGDME for 1 h, ¹H-NMR spectra show a large number of new peaks (Fig. S6, ESI†), indicating decomposition in accord with previous reports.⁶¹ Exposing imidazoliums for 1 h to ¹O₂, generated photochemically as detailed in the Supplementary Methods (ESI†), left the ¹H-NMR spectra largely unchanged (Fig. S7, ESI†). We do, however, not exclude a certain reactivity. Imidazoliums in high concentrations show a noticeable ability to quench ¹O₂ to ³O₂, which suggests that measured ¹O₂ yields with imidazolium are likely underestimated (Fig. S8 and Supplementary Note 1, ESI†). Overall, enhanced ¹O₂ formation and instability with superoxide both make imidazoliums unsuitable for metal–O₂ cells.

Turning to superoxide disproportionation in Na⁺ and mixed Na⁺/TBA⁺ electrolytes, we find for pure Na⁺ an analogous result to Fig. 2a: ∼13% ¹O₂, 85% ³O₂ and 87% Na₂O₂. For the mixed Na⁺/TBA⁺ electrolyte the ³O₂ and Na₂O₂ yields further dropped to around 70% and the ¹O₂ rose to ∼16%. Together with the results for Li⁺/TBA⁺ mixtures, the higher levels of ¹O₂ with the less Lewis acidic Na⁺ suggest that weaker Lewis acidic cations favour ¹O₂ evolving pathways. Another common result for all conditions in Fig. 2b is that the amounts of ³O₂ and alkali peroxide closely match each other and that a larger fraction of missing peroxide is related to a larger amount of ¹O₂ formed.

Disproportionation during Li–O₂ cell cycling

To probe whether the above observed disproportionation phenomena that yield ¹O₂ also explain ¹O₂ formation in cells, we performed analogous electrochemical experiments in Li–O₂ cells. Li–O₂ was chosen since disproportionation is most significantly driven by thermodynamics (Fig. 1a). We focus on TBA⁺ as the weak Lewis acid since it avoids the further complications of imidazolium instability with O₂⁻. Considering first discharge, we constructed cells as detailed in the Experimental section with carbon black cathodes and TEGDME electrolytes containing 30 mM DMA and either only 0.1 M Li⁺ or a total of 1 M salt with a Li⁺ : TBA⁺ ratio of 1 : 9 or 1 : 99. The cells were discharged at constant current and the O₂ consumption followed using a pressure transducer as shown in Fig. 3a and Fig. S9 (ESI†). At the end of discharge, electrolyte and cathodes were extracted and analysed for the amount of ¹O₂, Li₂O₂, and carbonate. The results are shown in Fig. 3b with all values expressed as mol per 2 mol e⁻ passed. Hence, ideally 2 mol e⁻ would give 1 mol Li₂O₂ according to eqn (1).

Fig. 3 Superoxide disproportionation and ¹O₂ formation during Li–O₂ cell discharge. (a) O₂ consumption vs. capacity upon discharge of carbon black electrodes at a rate of 100 mA g_C⁻¹ in O₂ saturated TEGDME electrolytes that contained 30 mM DMA and either 0.1 M Li⁺ or a total of 1 M salt with a Li⁺ : TBA⁺ ratio of 1 : 9 or 1 : 99. The 1 : 99 ratio is given in Fig. S9 (ESI†) and voltage profiles in Fig. S10 (ESI†). (b) Obtained Li₂O₂, ¹O₂, and Li₂CO₃ (expressed as CO₂) per 2 e⁻ passed in the cells shown in (a).

Discharge in pure Li⁺ electrolyte resulted in a ratio of 1.98 e⁻/O₂, close to the ideal ratio of 2, and a Li₂O₂ yield of 94%, which both is in accord with previous reports for similar cells.^{11,15,20–22,33,50} The ¹O₂ yield was ∼3% and hence similar to that found in Fig. 2 for O₂⁻ disproportionation in Li⁺ electrolyte. With mixed Li⁺/TBA⁺ electrolytes with a Li⁺ : TBA⁺ ratio of 1 : 9 (1 : 99), the e⁻/O₂ ratio and Li₂O₂ yield dropped to 1.74 (1.70) e⁻/O₂ and 85% (81%), respectively. Concurrently, the amount of ¹O₂ and carbonate increased as the Li₂O₂ yield decreased. Increasing ¹O₂ yield together with decreasing Li₂O₂ yield as the electrolyte is changed from Li⁺ to Li⁺/TBA⁺ mix mirrors the results in Fig. 2 for the chemical experiments. Considering further the e⁻/O₂ ratios, the ideal value of 2 results from the sinks for the initially formed O₂⁻: a second 1 e⁻ reduction to peroxide or disproportionation to ³O₂, which both give an overall 2 e⁻/O₂ process. e⁻/O₂ ratios lower than 2 imply more efficient sinks to exist for the 1 e⁻ product O₂⁻ than a second reduction or disproportionation to ³O₂. Given the known stability of TBA⁺ with O₂⁻,^36,41 their reaction can be excluded as the sink to cause the decrease to 1.74 (1.70) e⁻/O₂. Instead, the lower ratio is in accord with TBA⁺ enhancing the ¹O₂ fraction from O₂⁻ disproportionation. Discharge with imidazoliums instead of TBA⁺ further corroborates their unsuitability as seen in even lower e⁻/O₂ ratios of 1.42 and 1.2 for EM₂Im⁺ and EMIm⁺, respectively (Supplementary Note 2 and Fig. S11, ESI†). The results on discharge are in accord with the chemical experiments shown in Fig. 2, which have shown that O₂⁻ disproportionation partly releases ¹O₂ and that the ¹O₂ fraction increases with the presence of TBA⁺. Overall, the results show that O₂⁻ disproportionation is the source of ¹O₂ on discharge, which further implies that discharge in the investigated TEGDME electrolyte passes at least significantly via disproportionation next to a possible second 1 e⁻ reduction of the LiO₂ intermediate viaeqn (2).

Turning to cell charge, we probed whether TBA⁺ analogously reveals ¹O₂ formation by O₂⁻ disproportionation. Li_2−xO₂ or soluble superoxide species were reported as intermediates on charge that disproportionate to form Li₂O₂ and O₂.^1,40,42,43 This reaction may hence equally be the source of ¹O₂ and sensitive to cations. We constructed Li₂O₂-packed working electrodes as detailed in the Experimental section. Li₂O₂ was ball milled with carbon black to ensure intimate contact between the two and the resulting powder was used to form composite electrodes using PTFE binder. We charged them in electrolytes that contained either only Li⁺ or a Li⁺/TBA⁺ mix and measured the amount of ³O₂ and ¹O₂ by means of the pressure in the cell head space and DMA conversion, Fig. 4. The charge voltage was limited to 3.95 V since this voltage was reported to be the upper limit for quasi-equilibrium decomposition in TEGDME.⁴⁰ Pressure evolution with pure Li⁺ electrolyte (Fig. 4a) shows similarly to previous reports^11,20,21,40 an elevated value of 2.40 e⁻/O₂ and thus ∼83% of the expected O₂ evolved based on charge passed. ¹O₂ formation shows that the ³O₂ loss is connected with ¹O₂ formation. When Li₂O₂ was charged in Li⁺/TBA⁺ electrolyte (Fig. 4b), the e⁻/O₂ ratio rose to 2.95 and hence only ∼68% of the expected ³O₂ evolved. Roughly doubled missing ³O₂ evolution goes along with the ¹O₂ amount being more than doubled. To exclude the suggested ¹O₂ evolution from a direct 2 e⁻ oxidation of Li₂O₂ above 3.5 V,³² we also restricted the charging voltage to 3.45 V, which shows similar results as with charge limited to 3.95 V (Supplementary Note 3, ESI†). Analogously to the experiments on discharge (Fig. 3), presence of TBA⁺ increased the fraction of ¹O₂ from O₂⁻ disproportionation with concurrently dropping ³O₂ fraction. Proportional correlation between missing ³O₂ evolution and ¹O₂ yield suggest in either case superoxide disproportionation to be a major O₂ evolution and ¹O₂ generation pathway.

Fig. 4 Superoxide disproportionation and ¹O₂ formation during Li–O₂ cell charge. (a and b) O₂ evolution vs. time upon charge of carbon black/Li₂O₂/PTFE (9/1/1, m/m) composite electrodes in TEGDME electrolyte containing 30 mM DMA and 0.1 M Li⁺ (a) or 0.1 M Li⁺ and 0.9 M TBA⁺ (b). Electrodes were charged at a rate of 10 mA g_C⁻¹ until 3.95 V and then kept at open circuit until the pressure was stable. (c) ³O₂ and ¹O₂ obtained per 2 e⁻ passed for the cells shown in (a) and (b).

Taken together, the results from the chemical and electrochemical experiments show that superoxide disproportionation, driven by the higher stability of the peroxide with strong Lewis acids, generates in part ¹O₂. Simultaneous presence of weakly Lewis acidic organic cations increases the ¹O₂ yield markedly in the chemical and electrochemical experiments. These results (a) corroborate that superoxide disproportionation is a main pathway for the second electron transfer from superoxide to peroxide during discharge and O₂ evolution during charge and (b) show that superoxide disproportionation is the ¹O₂ source during cell cycling.

A direct consequence of this finding is that the extent to which ¹O₂ can form on discharge and charge is governed by the extent to which disproportionation is responsible for the second electron transfer. The latter has been subject to many important studies recently and current understanding is that dominance of one or the other is governed by the LiO₂ solvation vs. surface adsorption.^2,62 Except for very poorly LiO₂ solvating electrolytes such as MeCN, disproportionation appears to dominate even in only slightly more solvating glymes and certainly in any solvent with higher donor number, which is further enhanced by Li salt anions that dissociate weakly.^6,63 Partition between second reduction/disproportionation has been, for example, investigated in glyme and DMSO by Shao-Horn who found at least significant disproportionation in glyme at low overpotentials.⁶⁴ Peng calculated potentials where O₂⁻* could be directly reduced to Li₂O₂* in DMSO and found disproportionation to dominate above 2 V vs. Li/Li⁺ on Au(111). Considering the latter, catalysts could potentially favour a second electron transfer already at higher voltages. We are, however, not aware of any study showing this possibility on discharge, but recent work by Lu suggests that catalysts could favour direct oxidation on charge.^40,65,66 Another potential way could be redox mediators as suggested for quinones on discharge.^7,8 However, proof that this fully suppresses superoxide disproportionation is still missing.

Energetics of singlet oxygen generation

To better understand the energetics of disproportionation and particularly why weak Lewis acids boost ¹O₂ formation, we performed density functional theory (DFT) calculations for pathways leading to ³O₂ and ¹O₂. We considered LiO₂ and NaO₂ disproportionation as well as the asymmetric pairings of LiO₂ and NaO₂ with HO₂ and TBAO₂. Energies were calculated for solvated species using the continuous C-PCM solvation model with a mean dielectric constant of ε = 7.28 (resembling glyme) and using the hybrid GGA DFT M06-2X functional and the most favourable pathways are shown in Fig. 5. All energies are relative to the free superoxide monomers (2LiO₂ or 2NaO₂) to help understanding how cations other than Li⁺ or Na⁺ change the energetics relative to pure Li⁺ or Na⁺ electrolytes due to ion association/dissociation equilibria. A reaction energy beyond 1 eV implies an at least as high activation barrier and hence slow kinetics at room temperature.¹⁸ Starting from the doublet superoxide monomers, reactions may follow singlet or triplet pathways through the formation of a superoxide dimer M(O₂)₂M′ (M being Li⁺ or Na⁺, and M′ being M, H⁺ or TBA⁺). The dimers release singlet or triplet O₂ and singlet MO₂M′ peroxide that may further ion exchange to M₂O₂ and precipitate as solid M₂O_2(s).

Fig. 5 Reaction free energy profiles for superoxide disproportionation. (a) LiO₂ disproportionation with itself, O₂⁻ or HO₂ to Li₂O₂ and molecular oxygen. (b) NaO₂ disproportionation with itself or O₂⁻ to Na₂O₂ and molecular oxygen. Pathways to release ³O₂ and ¹O₂ are indicated by full and dashed lines, respectively. All species are computed in the solvated state except for the final solid peroxide. The computational method is M06-2X6-31**G++C-PMD (ε = 7.28). Numerical values are given in Tables S2–S4 (ESI†). Further asymmetric alkali superoxide pairings are considered in Supplementary Note 5 (ESI†).

We consider first the symmetric LiO₂ and NaO₂ cases (Fig. 5a and b, red traces). For LiO₂, the triplet ³Li(O₂)₂Li dimer is slightly stabilized compared to two monomers and releases Li₂O₂ + ³O₂ weakly endergonic, followed by strongly stabilizing Li₂O₂ precipitation to solid Li₂O_2(s), which is the main overall driving force (Fig. 5a). Our results are in accord with previous works that analysed the route from LiO₂ to ³O₂ in the gas^35,56 and solution phase⁴¹ and which are summarized in Fig. S13 (ESI†) for comparison. The path that we find for ¹O₂ release appears possible but slower with a thermodynamic barrier of ∼1 eV to the singlet ¹Li(O₂)₂Li dimer followed by downhill ¹O₂ release and Li₂O_2(s) precipitation. The symmetric NaO₂ case (Fig. 5b) is in either case uphill to the dimers but with their order being reversed (relative energies of singlet/triplet M(O₂)₂M dimers are analysed in detail in Supplementary Note 4, ESI†); the singlet ¹Na(O₂)₂Na dimer forms with an energy increase of 0.83 eV less endergonic than the triplet ³Na(O₂)₂Na with ∼1.2 eV barrier. However, ongoing ¹O₂ release is further endergonic by 0.5 eV while ³O₂ release is exergonic by −0.5 eV. The single step thermodynamic barrier towards ¹O₂ release from NaO₂ is hence ∼0.1 eV higher than the barrier towards ³O₂. The following precipitation of solid Na₂O_2(s) makes both singlet and triplet path overall exergonic, but less than for LiO₂ disproportionation. Together, the relative single step barriers and overall driving forces rationalize our experimental findings: LiO₂ disproportionates fast and the strongly differing barriers between singlet and triplet path cause relatively small ¹O₂ fractions. NaO₂ disproportionates slowly and the more similar barriers cause larger ¹O₂ fractions.

Turning to proton mediated O₂⁻ disproportionation, our thermodynamic calculations for the asymmetric LiO₂ + HO₂ pairing suggest much easier ³O₂ than ¹O₂ formation (Fig. 5a, blue traces). ³O₂ and the mixed HLiO₂ peroxide form in an exergonic single step reaction without a stable intermediate dimer. In contrast, the singlet path faces a barrier of 0.46 eV to the singlet ¹Li(O₂)₂H dimer, which releases HLiO₂ and ¹O₂ in a by −0.52 eV exergonic step. Analogous results were obtained for the NaO₂ + HO₂ pairing (Fig. S14, ESI†). The singlet path is in either case much more demanding and will result in minor ¹O₂ yields. This is in accord with our experimental finding in Fig. 2, which shows insignificant additional ¹O₂ with protons compared to pure Li⁺ electrolyte. It is also in accord with reported negligible ¹O₂ yields from proton mediated O₂⁻ disproportionation in Li⁺ and Na⁺ free media.^59,60 We conclude from the calculations, in accord with the experiments, that proton sources cause minor additional ¹O₂ compared to disproportionation in Li⁺ electrolyte.

Turning to the case of the asymmetric pairing of superoxide with Li⁺ and the weakly Lewis acidic TBA⁺, our initial hypothesis was that weaker O₂⁻–TBA⁺ than O₂⁻–M⁺ interactions^9,37,45 would destabilize intermediates, reduce barriers, and hence make ¹O₂ more favourable. In support of that, the experiments have shown higher kinetics and ¹O₂ yields with TBA⁺ (Fig. 2–4 and Fig. S5, ESI†) and the calculations in Fig. 5 confirm the suggested reasons. Considering the weak association of the TBA⁺O₂⁻ ion pair even in low dielectric constant solvents like DME , TBAO₂ may be approximated by the free solvated O₂⁻ anion. Solvent dependent O₂/LiO₂ and O₂/TBAO₂ standard potentials have been measured and computed by Shao-Horn et al.⁴⁵ and found to differ by 1.24 V in DME, which agrees well with our estimate of 1.21 eV for the dissociation energy of LiO₂ to free O₂⁻ anions (Fig. 5a, black traces). Note that O₂⁻ does not have to form via dissociation of LiO₂, but may form as a transient species upon O₂⁻ generation. Ongoing triplet and singlet paths initially form ³Li(O₂)₂⁻ and ¹Li(O₂)₂⁻ dimers that are stabilized versus LiO₂ + O₂⁻ by −0.52 eV and −0.49 eV, respectively. Ongoing pathways to the charged LiO₂⁻ peroxide species plus ¹O₂ or ³O₂ would face prohibitively high barriers because of the large dissociation energy of Li₂O₂ → LiO₂⁻ + Li⁺. Instead, our calculations reveal other facile pathways: the Li(O₂)₂⁻ dimers can easily exchange TBA⁺ for Li⁺ and hence feed into the symmetric Li(O₂)₂Li pathways discussed above and shown in the red traces in Fig. 5a. Crucially, the presence of TBA⁺ decreases the barrier towards ¹O₂, the endergonicity of the most unfavourable step to the ¹Li(O₂)₂Li dimer, from ∼1 eV to a mere 0.27 eV. Analogously, the asymmetric NaO₂ + O₂⁻ pairing passes via Na(O₂)₂⁻ and Na(O₂)₂Na dimers and the barrier towards ¹O₂ decreases from 1.2 eV to 0.4 eV. Overall, the weak Lewis acid TBA⁺ opens paths to bypass the most unfavourable reaction steps und hence strongly facilitates ¹O₂ evolution.

Consequences for metal–O₂ batteries

Recognizing that ¹O₂ formation is deeply rooted in the way current metal–O₂ cells operate has serious consequences on aspects to avoid and on directions that should be gone. First, caution must be exercised with weak Lewis acids as electrolytes or additives. This is supported by the selected quaternary ammonium and imidazolium cations, which are prototypical motifs for the cations used so far in ionic liquid electrolytes for metal–O₂ cells. Imidazoliums readily decompose with superoxide. Significantly, we could show that the organic cation's weak Lewis acidity rather than its chemical nature massively boosts ¹O₂ formation. Given that ionic liquid cations suitable for electrolytes are most typically weak Lewis acids, the effect can likely be generalized. Favoured ¹O₂ formation explains why quantitative studies of metal–O₂ chemistry with a broad variety of ionic liquids have shown worse parasitic chemistry on discharge and charge than molecular electrolytes.^19,20,24

Second, protic additives drive ¹O₂ formation insignificantly but may drive parasitic chemistry in other ways. This is in accord with reports that found increased side reactions when water or other Brønsted acids were added.^11,29,67 The previous suggestion that proton sources could cause ¹O₂ in Na–O₂ cells³⁴ can now be revised to NaO₂ disproportionation being the ¹O₂ source. Protons may be a remaining source of instability in K–O₂ cells despite thermodynamic stability of KO₂ in K⁺ electrolytes.^5,10,26,38 Meticulously excluding impurities has hence allowed for impressive cyclability of K–O₂ batteries.⁵

Finally, the most prominent consequence is that situations bound for superoxide disproportionation must be avoided. Cells based on metastable LiO₂ or NaO₂ as target products likely lack the practically required tolerance to slow discharge and rest periods; the superoxides gradually convert to peroxide and side products.^{3,14,25,27,30,34,35} Peroxide products are preferred as they are much higher in energy density and the thermodynamically stable products.^28,31,68 Cycling them highly reversible requires finding routes to form and decompose them without superoxide disproportionation steps. Potential ways to do so are catalysts^40,65,66 or redox mediators.^7,8

Conclusions

In conclusion, we describe the mechanism for ¹O₂ formation and hence a main driver parasitic chemistry across alkali metal–O₂ cells. We show that superoxide disproportionation forms the ¹O₂, and we clarify the reaction mechanism and governing factors in detail. The mechanism explains the growing parasitic chemistry in K–, Na–, and Li–O₂ cells as well as between superoxide and peroxide products based on the growing propensity for disproportionation. The strong Lewis acids H⁺, Li⁺ and Na⁺ stabilize peroxide versus superoxide and hence drive disproportionation. ¹O₂ yields grow in this order with H⁺ causing insignificant ¹O₂ and strongly growing ¹O₂ fractions with Li⁺ and Na⁺. Importantly, weak Lewis acids such as TBA⁺ alone do not drive disproportionation, but, when combined with strong Lewis acids, strongly reduce the reaction barriers towards ¹O₂ and cause substantially larger fractions of ¹O₂. This calls for caution with ionic liquid electrolytes that comprise such weak Lewis acidic cations. The results explain major degradation routes of metal–O₂ cells. Given that achieving high capacities and rates requires solution routes on discharge and charge, which in turn favour disproportionation reactions, the results establish the central dilemma that disproportionation is both important for high capacity/high rate and responsible for degradation. Future work should hence focus on finding routes for peroxide discharge and charge that avoid superoxide disproportionation.

Authors contributions

E. M., Y. K. P., A. S., C. L., N. M., S. A. F. developed the methods and/or did the experiments. R. S., F. F. S., S. B. did calculations. C. P., C. S., O. F., S. B., S. A. F. discussed the results. S. A. F. and S. B. conceived the work and wrote the paper.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

S. A. F. is indebted to the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation program (grant agreement no. 636069) and the Austrian Federal Ministry of Science, Research and Economy and the Austrian Research Promotion Agency (grant no. 845364). We thank EL-Cell GmbH (Hamburg, Germany) for the test cells. Likewise, we thank M. Winkler of Acib GmbH and G. Strohmeier for HPLC measurements, S. Borisov, W. Yin, and A. Grimaud for discussions, and R. Saf for help with the MS.

Notes and references

1. Y.-C. Lu, B. M. Gallant, D. G. Kwabi, J. R. Harding, R. R. Mitchell, M. S. Whittingham and Y. Shao-Horn, Energy Environ. Sci., 2013, 6, 750–768.
2. D. Aurbach, B. D. McCloskey, L. F. Nazar and P. G. Bruce, Nat. Energy, 2016, 1, 16128.
3. T. Liu, G. Kim, M. T. L. Casford and C. P. Grey, J. Phys. Chem. Lett., 2016, 7, 4841–4846.
4. P. Hartmann, C. L. Bender, M. Vračar, A. K. Dürr, A. Garsuch, J. Janek and P. Adelhelm, Nat. Mater., 2012, 12, 228–232.
5. G. Cong, W. Wang, N.-C. Lai, Z. Liang and Y.-C. Lu, Nat. Mater., 2019, 18, 390–396.
6. C. M. Burke, V. Pande, A. Khetan, V. Viswanathan and B. D. McCloskey, Proc. Natl. Acad. Sci. U. S. A., 2015, 112, 9293–9298.
7. X. Gao, Y. Chen, L. Johnson and P. G. Bruce, Nat. Mater., 2016, 15, 882–888.
8. T. Liu, J. T. Frith, G. Kim, R. N. Kerber, N. Dubouis, Y. Shao, Z. Liu, P. C. M. M. Magusin, M. T. L. Casford, N. Garcia-Araez and C. P. Grey, J. Am. Chem. Soc., 2018, 140, 1428–1437.
9. L. Johnson, C. Li, Z. Liu, Y. Chen, S. A. Freunberger, J.-M. Tarascon, P. C. Ashok, B. B. Praveen, K. Dholakia and P. G. Bruce, Nat. Chem., 2014, 6, 1091–1099.
10. W. Wang, N.-C. Lai, Z. Liang, Y. Wang and Y.-C. Lu, Angew. Chem., Int. Ed., 2018, 57, 5042–5046.
11. N. B. Aetukuri, B. D. McCloskey, J. M. García, L. E. Krupp, V. Viswanathan and A. C. Luntz, Nat. Chem., 2014, 7, 50–56.
12. C. Xia, R. Black, R. Fernandes, B. Adams and L. F. Nazar, Nat. Chem., 2015, 7, 496–501.
13. L. Lutz, W. Yin, A. Grimaud, D. Alves Dalla Corte, M. Tang, L. Johnson, E. Azaceta, V. Sarou-Kanian, A. J. Naylor, S. Hamad, J. A. Anta, E. Salager, R. Tena-Zaera, P. G. Bruce and J.-M. Tarascon, J. Phys. Chem. C, 2016, 120, 20068–20076.
14. I. Landa-Medrano, R. Pinedo, X. Bi, I. Ruiz de Larramendi, L. Lezama, J. Janek, K. Amine, J. Lu and T. Rojo, ACS Appl. Mater. Interfaces, 2016, 8, 20120–20127.
15. B. D. Adams, R. Black, Z. Williams, R. Fernandes, M. Cuisinier, E. J. Berg, P. Novak, G. K. Murphy and L. F. Nazar, Adv. Energy Mater., 2015, 5, 1400867.
16. M. Carboni, A. G. Marrani, R. Spezia and S. Brutti, Chem. – Eur. J., 2016, 22, 17188–17203.
17. R. Black, A. Shyamsunder, P. Adeli, D. Kundu, G. K. Murphy and L. F. Nazar, ChemSusChem, 2016, 9, 1795–1803.
18. V. S. Bryantsev, V. Giordani, W. Walker, M. Blanco, S. Zecevic, K. Sasaki, J. Uddin, D. Addison and G. V. Chase, J. Phys. Chem. A, 2011, 115, 12399–12409.
19. S. Das, J. Højberg, K. B. Knudsen, R. Younesi, P. Johansson, P. Norby and T. Vegge, J. Phys. Chem. C, 2015, 119, 18084–18090.
20. B. D. McCloskey, D. S. Bethune, R. M. Shelby, T. Mori, R. Scheffler, A. Speidel, M. Sherwood and A. C. Luntz, J. Phys. Chem. Lett., 2012, 3, 3043–3047.
21. B. D. McCloskey, J. M. Garcia and A. C. Luntz, J. Phys. Chem. Lett., 2014, 5, 1230–1235.
22. B. D. McCloskey, A. Valery, A. C. Luntz, S. R. Gowda, G. M. Wallraff, J. M. Garcia, T. Mori and L. E. Krupp, J. Phys. Chem. Lett., 2013, 4, 2989–2993.
23. R. Black, S. H. Oh, J.-H. Lee, T. Yim, B. Adams and L. F. Nazar, J. Am. Chem. Soc., 2012, 134, 2902–2905.
24. M. Piana, J. Wandt, S. Meini, I. Buchberger, N. Tsiouvaras and H. A. Gasteiger, J. Electrochem. Soc., 2014, 161, A1992–A2001.
25. J. Kim, H. Park, B. Lee, W. M. Seong, H.-D. Lim, Y. Bae, H. Kim, W. K. Kim, K. H. Ryu and K. Kang, Nat. Commun., 2016, 7, 10670.
26. N. Xiao, R. T. Rooney, A. A. Gewirth and Y. Wu, Angew. Chem., Int. Ed., 2018, 57, 1227–1231.
27. S. Y. Sayed, K. P. C. Yao, D. G. Kwabi, T. P. Batcho, C. V. Amanchukwu, S. Feng, C. V. Thompson and Y. Shao-Horn, Chem. Commun., 2016, 52, 9691–9694.
28. C. L. Bender, P. Hartmann, M. Vračar, P. Adelhelm and J. Janek, Adv. Energy Mater., 2014, 4, 1–10.
29. K. U. Schwenke, M. Metzger, T. Restle, M. Piana and H. A. Gasteiger, J. Electrochem. Soc., 2015, 162, A573–A584.
30. D. Zhai, H.-H. Wang, J. Yang, K. C. Lau, K. Li, L. A. Curtiss and K. Amine, J. Am. Chem. Soc., 2013, 135, 15364–15372.
31. C. L. Bender, D. Schröder, R. Pinedo, P. Adelhelm and J. Janek, Angew. Chem., Int. Ed., 2016, 55, 4640–4649.
32. J. Wandt, P. Jakes, J. Granwehr, H. A. Gasteiger and R.-A. Eichel, Angew. Chem., Int. Ed., 2016, 55, 6892–6895.
33. N. Mahne, B. Schafzahl, C. Leypold, M. Leypold, S. Grumm, A. Leitgeb, G. A. Strohmeier, M. Wilkening, O. Fontaine, D. Kramer, C. Slugovc, S. M. Borisov and S. A. Freunberger, Nat. Energy, 2017, 2, 17036.
34. L. Schafzahl, N. Mahne, B. Schafzahl, M. Wilkening, C. Slugovc, S. M. Borisov and S. A. Freunberger, Angew. Chem., Int. Ed., 2017, 56, 15728–15732.
35. U. Das, K. C. Lau, P. C. Redfern and L. A. Curtiss, J. Phys. Chem. Lett., 2014, 5, 813–819.
36. C. Sheng, F. Yu, Y. Wu, Z. Peng and Y. Chen, Angew. Chem., Int. Ed., 2018, 57, 9906–9910.
37. C. J. Allen, J. Hwang, R. Kautz, S. Mukerjee, E. J. Plichta, M. A. Hendrickson and K. M. Abraham, J. Phys. Chem. C, 2012, 116, 20755–20764.
38. Y. Chen, Z. P. Jovanov, X. Gao, J. Liu, C. Holc, L. R. Johnson and P. G. Bruce, J. Electroanal. Chem., 2018, 819, 542–546.
39. C. V. Amanchukwu, H.-H. Chang, M. Gauthier, S. Feng, T. P. Batcho and P. T. Hammond, Chem. Mater., 2016, 28, 7167–7177.
40. Y. Wang, N.-C. Lai, Y.-R. Lu, Y. Zhou, C.-L. Dong and Y.-C. Lu, Joule, 2018, 2, 2364–2380.
41. Y. Zhang, X. Zhang, J. Wang, W. C. McKee, Y. Xu and Z. Peng, J. Phys. Chem. C, 2016, 120, 3690–3698.
42. S. Ganapathy, B. D. Adams, G. Stenou, M. S. Anastasaki, K. Goubitz, X.-F. Miao, L. F. Nazar and M. Wagemaker, J. Am. Chem. Soc., 2014, 136, 16335–16344.
43. S. Kang, Y. Mo, S. P. Ong and G. Ceder, Chem. Mater., 2013, 25, 3328–3336.
44. Y.-C. Lu and Y. Shao-Horn, J. Phys. Chem. Lett., 2012, 4, 93–99.
45. D. G. Kwabi, V. S. Bryantsev, T. P. Batcho, D. M. Itkis, C. V. Thompson and Y. Shao-Horn, Angew. Chem., Int. Ed., 2016, 55, 3129–3134.
46. Z. Peng, S. A. Freunberger, L. J. Hardwick, Y. Chen, V. Giordani, F. Bardé, P. Novák, D. Graham, J.-M. Tarascon and P. G. Bruce, Angew. Chem., Int. Ed., 2011, 50, 6351–6355.
47. M. Schmeisser, P. Illner, R. Puchta, A. Zahl and R. van Eldik, Chem. – Eur. J., 2012, 18, 10969–10982.
48. W. Linert, A. Camard, M. Armand and C. Michot, Coord. Chem. Rev., 2002, 226, 137–141.
49. M. M. Ottakam Thotiyl, S. A. Freunberger, Z. Peng and P. G. Bruce, J. Am. Chem. Soc., 2013, 135, 494–500.
50. B. Schafzahl, E. Mourad, L. Schafzahl, Y. K. Petit, A. R. Raju, M. O. Thotiyl, M. Wilkening, C. Slugovc and S. A. Freunberger, ACS Energy Lett., 2017, 3, 170–176.
51. Y. K. Petit, C. Leypold, N. Mahne, E. Mourad, L. Schafzahl, C. Slugovc, S. M. Borisov and S. A. Freunberger, Angew. Chem., Int. Ed., 2019, 58, 6535–6539.
52. Y. Zhao and D. G. Truhlar, Theor. Chem. Acc., 2008, 120, 215–241.
53. M. Cossi, N. Rega, G. Scalmani and V. Barone, J. Comput. Chem., 2003, 24, 669–681.
54. M. González, R. Valero, J. M. Anglada and R. Sayós, J. Chem. Phys., 2001, 115, 7015–7031.
55. M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone, G. A. Petersson, H. Nakatsuji, X. Li, M. Caricato, A. V. Marenich, J. Bloino, B. G. Janesko, R. Gomperts, B. Mennucci, H. P. Hratchian, J. V. Ortiz, A. F. Izmaylov, J. L. Sonnenberg, Williams, F. Ding, F. Lipparini, F. Egidi, J. Goings, B. Peng, A. Petrone, T. Henderson, D. Ranasinghe, V. G. Zakrzewski, J. Gao, N. Rega, G. Zheng, W. Liang, M. Hada, M. Ehara, K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao, H. Nakai, T. Vreven, K. Throssell, J. A. Montgomery, Jr., J. E. Peralta, F. Ogliaro, M. J. Bearpark, J. J. Heyd, E. N. Brothers, K. N. Kudin, V. N. Staroverov, T. A. Keith, R. Kobayashi, J. Normand, K. Raghavachari, A. P. Rendell, J. C. Burant, S. S. Iyengar, J. Tomasi, M. Cossi, J. M. Millam, M. Klene, C. Adamo, R. Cammi, J. W. Ochterski, R. L. Martin, K. Morokuma, O. Farkas, J. B. Foresman and D. J. Fox, Gaussian 16, Revision B.01, Gaussian, Inc., Wallingford CT, 2016.
56. V. S. Bryantsev, M. Blanco and F. Faglioni, J. Phys. Chem. A, 2010, 114, 8165–8169.
57. M. W. Chase, Jr., NIST-JANAF thermochemical tables, 4th edn, Washington, DC: American Chemical Society; New York: American Institute of Physics for the National Institute of Standards and Technology, 1998.
58. P. D. C. Dietzel, R. K. Kremer and M. Jansen, J. Am. Chem. Soc., 2004, 126, 4689–4696.
59. J. M. Aubry, J. Rigaudy, C. Ferradini and J. Pucheault, J. Am. Chem. Soc., 1981, 103, 4965–4966.
60. E. J. Nanni, R. R. Birge, L. M. Hubbard, M. M. Morrison and D. T. Sawyer, Inorg. Chem., 1981, 20, 737–741.
61. M. Hayyan, M. A. Hashim and I. M. AlNashef, Chem. Rev., 2016, 116, 3029–3085.
62. H.-D. Lim, B. Lee, Y. Bae, H. Park, Y. Ko, H. Kim, J. Kim and K. Kang, Chem. Soc. Rev., 2017, 46, 2873–2888.
63. D. Sharon, D. Hirsberg, M. Salama, M. Afri, A. A. Frimer, M. Noked, W. Kwak, Y.-K. Sun and D. Aurbach, ACS Appl. Mater. Interfaces, 2016, 8, 5300–5307.
64. D. G. Kwabi, M. Tułodziecki, N. Pour, D. M. Itkis, C. V. Thompson and Y. Shao-Horn, J. Phys. Chem. Lett., 2016, 7, 1204–1212.
65. Y. Wang and Y.-C. Lu, Angew. Chem., Int. Ed., 2019, 58, 6962–6966.
66. Y. Wang, Z. Liang, Q. Zou, G. Cong and Y.-C. Lu, J. Phys. Chem. C, 2016, 120, 6459–6466.
67. D. G. Kwabi, T. P. Batcho, S. Feng, L. Giordano, C. V. Thompson and Y. Shao-Horn, Phys. Chem. Chem. Phys., 2016, 18, 24944–24953.
68. S. A. Freunberger, Nat. Energy, 2017, 2, 17091.

---

Footnote

† Electronic supplementary information (ESI) available: Supporting methods, figures, tables and notes. See DOI: 10.1039/c9ee01453e

---