Singlet Oxygen Generation as a Major Cause for Parasitic Reactions during Cycling of Aprotic Lithium-Oxygen Batteries

Non-aqueous metal-oxygen batteries depend critically on the reversible formation/decomposition of metal oxides on cycling. Irreversible parasitic reactions cause poor rechargeability, efficiency, and cycle life and have predominantly been ascribed to the reactivity of reduced oxygen species with cell components. These species, however, cannot fully explain the side reactions. Here we show that singlet oxygen forms at the cathode of a lithium-oxygen cell during discharge and from the onset of charge, and accounts for the majority of parasitic reaction products. The amount increases during discharge, early stages of charge, and charging at higher voltages, and is enhanced by the presence of trace water. Superoxide and peroxide appear to be involved in singlet oxygen generation. Singlet oxygen traps and quenchers can reduce parasitic reactions effectively. Awareness of the highly reactive singlet oxygen in non-aqueous metal-oxygen batteries gives a rationale for future research towards achieving highly reversible cell operation.

Many researchers have investigated the origin of parasitic reactions and proposed strategies to mitigate them 7,8,10,[16][17][18][19] . Superoxide has been most widely mentioned in causing side reactions on discharge since it forms as an intermediate in O 2 reduction and is a strong nucleophile and base 3,11,14,20,21,27 . Also, Li 2 O 2 was found to react with the electrolyte and carbon on discharge 3,[21][22][23][24] . These reactivities were used to explain the observation that on discharge typically close to the ideal value of 2 electrons per one O 2 molecule are consumed despite significant amounts of side products such as Li 2 CO 3 , Li formate and Li acetate being formed 17,24 . On charge typically the e -/O 2 ratio deviates significantly from 2 and more of the side products form 5,7,15,24,25 . These parasitic reactions occur at charging potentials well within the stability window (oxidative stability) of carbon and electrolyte in the absence of Li 2 O 2 21,23,25 . It was therefore suggested that some sort of reactive intermediates of Li 2 O 2 oxidation cause electrolyte and carbon decomposition on charge 11,23,25,28 .
Chemical oxidation of alkaline peroxides in non-aqueous media is known to generate singlet oxygen ( 1  g or 1 O 2 ), the first excited state of triplet ground state dioxygen ( 3  g -) [29][30][31][32] . Based on the reversible potential of Li 2 O 2 formation and the energy difference between triplet and singlet oxygen, the formation of 1 O 2 in the Li-O 2 cell has been hypothesized to be possible at charging potentials exceeding 3.5 to 3.9 V vs. Li/Li +11, 23 . Only recently 1 O 2 was reported to form in small quantities between 3.55 and 3.75 V 28 . Overall, the hitherto known processes cannot consistently explain the observed irreversibilities. Only better knowledge of parasitic reactions may allow them to be inhibited so that progress towards fully reversible cell operation can continue.
Here we show that 1 O 2 forms in the Li-O 2 cathode during discharge and from the onset of charge and that it is responsible for a major fraction of the side products in the investigated system with ether electrolyte. The lower abundance on discharge and higher one on charge can consistently explain the typically observed deviations of the e -/O 2 ratio from the ideal value of 2. The origin of the 1 O 2 on charge appears to be superoxide and peroxide. The presence of trace water enhances the formation during both discharge and charge. We also show that 1 O 2 traps and quenchers as electrolyte additives can significantly reduce the amount of side products associated with 1 O 2 .

Reactivity of the electrolyte with singlet oxygen
The discharge product formed at the Li To detect 1 O 2 at quantities, which would be responsible even for small amounts of parasitic products, we  All electrolytes had an initial DMA concentration of 1.6×10 -5 M.
To probe whether 1 O 2 is also formed during discharge, we cycled electrodes in the fluorescence setup.
Results for galvanostatic cycling of a porous carbon cathode in dry 0.1 M LiClO 4 in TEGDME are shown in Fig. 2b. Upon discharge the DMA concentration remains nearly unchanged within the measurement accuracy. However, immediately after switching to charging, starting from ~3 V, the signal drops with increasing slope as charging progresses and the voltage climbs towards 4.3 V, where full recharge is reached. The cumulatively consumed DMA corresponds to ~4% of the expected O 2 being in ethers and also higher than in the non-deuterated solvents 32 . We also added 1000 ppm D 2 O since above experiments have shown higher 1 O 2 generation when trace water was present, besides that the lifetime is longer in D 2 O than in H 2 O. Results for galvanostatic cycling are shown in Fig. 3. The signal/noise ratio does not permit a clear statement about the abundance of 1 O 2 during discharge, which is to be expected given the low generation rate detected with operando fluorescence in Fig. 2. In accord with above results there is, however, unambiguous proof of 1 O 2 generation from the start of charging and increasing rate as charging progresses to higher voltages.

Trapping and quenching singlet oxygen
The above results show that 1  (DABCO) as quencher since it has been reported to be effective in non-aqueous environment 41 . DABCO also allows access to a relevant potential range between ~2.0 and 3.6 V and is stable with superoxide (Supplementary Fig. 10 and 11).
Li-O 2 cells with porous carbon black electrodes were constructed as described in the Methods section.
Three electrolytes were used: 0.1 M LiClO 4 in TEGDME that either contained no additive, 30 mM DMA, or 10 mM DABCO. Cycling was carried out at constant current in O 2 atmosphere. Cells were cycled to various discharge and charge capacities, then stopped and subjected to further analysis. A typical load curve is shown in Fig. 4a. As DABCO is oxidized at ~3.6 V, cells containing this additive were only recharged to 3.5 V and then held there until the first recharge capacity was reached.
To quantify the amount of carbonaceous side products (Li 2 CO 3 and Li carboxylates) formed at each stage of discharge and charge, the electrodes were analysed with a previously established procedure 25   Considering the cell without additive, there is continuous growth of the amount of side products with increasing discharge capacity. The amount further increases to the sampling point at one third recharge, and then vanishes nearly completely towards full recharge. This is in accord with previous investigations on the build-up and removal of the side products during cycling 15,24,25 . It was shown that on discharge side products originate predominantly from the electrolyte. At early stages of charge, the electrolyte further decomposes to solid products, accompanied by Li 2 CO 3 from the carbon electrode 24,25 .
As charge continues to higher voltages, carbon decomposition becomes more significant and carbon and electrolyte decomposition go along with CO 2 evolution from already present parasitic products. Turning to the cells with 1 O 2 trap or quencher, a significant reduction of side products during discharge is evident for both additives, Fig. 4b. Considering first the cell with DMA, the side products amount to between a half and a third of those without additive up to the second sampling point. Thereafter, the side products grow close to the level without the DMA. This can be explained considering the conversion of DMA to DMA-O 2 , Fig. 4c. At the first sampling point, 76% of the initially present DMA was consumed, and it was fully consumed at the second point. Thereafter, no effect on side product formation can be expected as is seen in the carbonate/carboxylate data, Fig. 4b. By considering the charge passed at the first sampling point and the DMA conversion, a ratio of ~1 mol DMA consumed per 10 mol of O 2 reduced can be determined.
Turning to the cells with DABCO as quencher, side products amount to consistently less than in the case of DMA additive and to 10 to 30% of the additive-free case on discharge, Fig. 4b. From these values, we can estimate the fraction of parasitic products on discharge originating from 1 O 2 to be at least 70%.
DABCO is also effective upon charging and significantly reduces the side products at the first sampling point on charge. We assume the reason for the lower efficiency on charging to be the much higher 1   With the DMA additive the recharge voltage is lower throughout than without DMA. With DMA the O 2 evolution reaches ~93% of the theoretical value at the beginning and fades to ~2/3 towards the end of charge. Without DMA the O 2 evolution is significantly lower throughout charging and reaches a maximum of 2/3 of the theoretical value. An even stronger difference is seen in the CO 2 evolution.
Significant CO 2 evolution without DMA is contrasted by a 30 fold reduced CO 2 amount with DMA (based on the integral peak area in Supplementary Fig. 14b and f). The strong reduction of the CO 2 amount in combination with the observed O 2 evolution suggests that the majority of the parasitic products that form during charge at voltages below the oxidative stability limit of electrolyte and carbon are due to the occurrence of 1 O 2 . A more in depth discussion for this assignment is given in the section Supplementary Discussion in the Supplementary Information. Taken together, the trap and quencher experiments contribute more evidence that 1 O 2 is responsible for the majority of side products upon discharge and charge, and that suitable additives can effectively reduce side reactions. The required oxidation stability of such additives can be reduced by using redox mediators that greatly reduce the charging voltage 8,42,43 .

Pathways to singlet oxygen
The results are consistent with 1 O 2 being to a large part responsible for commonly reported observations about the O 2 balance and side products. First, on discharge the e -/O 2 ratio is typically found within several percent of the ideal value of 2 despite significant amounts of side products such as Li 2 CO 3 , Li formate and Li acetate with Li 2 O 2 yields reported below 90% [24][25][26] . Second, on charge the e -/O 2 ratio typically deviates significantly by more than 10% from the ideal value of 2 from the start with the deviation increasing as charging progresses 24 . This deviation goes along with the formation of more of the mentioned solid side products until the charging voltage is sufficiently high to oxidize them to release CO 2 and other fragments 15,25,26 . So far the formation of these products could not be consistently explained by the reactivity of the known reactive species superoxide and peroxide alone 14,[23][24][25]   ).
Note that superoxide is both a proficient source and efficient quencher of 1 O 2 via Eq. (3) 50 .
We therefore believe that our observation of less 1 O 2 on discharge and more on charge in the ether electrolyte results at least in part from the differing abundance of superoxide that can reduce the 1 O 2 lifetime by quenching, which counteracts equally superoxide concentration driven formation. More precisely, net formation of 1 O 2 will depend on the relative kinetics of all superoxide sources and sinks (with 1 O 2 being involved in both) and not solely on the superoxide concentration. These sources and sinks are both electrochemical and chemical and change with discharge and charge, electrolyte, current, and potential. We thus further suggest that the current density and electrolyte properties will influence the 1 O 2 formation in much the same way it governs the occurrence of superoxide on discharge and charge below 3.5V 5,33 . Further, charge current will drive 1 O 2 production if it causes charging voltages above 3.5 V.

Conclusions
By combining complementary methods we could give evidence that 1  abundance makes traps less likely to be effective for long term cycling since they will be consumed rapidly. Physical quenchers are preferred since they are not consumed. Future work should therefore focus on finding quenchers that are entirely compatible with the cell environment, with the electrochemical potential window, compatibility, and stability against superoxide and peroxide being the most prominent requirements. Equally it needs to be compatible with anodes such as possibly protected Li metal. Alkaline superoxides in the cycling mechanism suggest that the Na-O 2 and K-O 2 systems would merit investigating whether 1 O 2 is involved.

Materials.
Ethylene glycol dimethyl ether (DME, >99.0%), 9,10-dimethylanthracene (DMA, >98.0 %) and 9,10-diphenylanthracene (DPA, >98.%) were purchased from TCI Europe. Operando NIR spectroscopy to detect the emission of singlet oxygen was performed using a germanium detector (model 261, UDT Instruments, Gamma Scientific Company, USA). It was cooled to -30 °C using a Peltier cooling unit. A longpass-filter with a cut-on wavelength of 1200 nm and a shortpass-filter with a cut-off wavelength of 1350 nm (Edmund optics) were placed directly in front of the sensor. The cell for operando NIR spectroscopy was a 1 mm absorption high precision quartz cell (Hellma Analytics) with a purpose made gas-tight PTFE-lid. The working electrode was an Au-grid electrode (ALS). The reference and counter electrodes were partly delithiated LiFePO 4 attached to an Al-grid. The cell was placed directly in front of the filters followed by an Au-mirror. The optical set-up containing the measurement cell was located in a blackbox to avoid ingress of stray light. The detector signal was amplified by a photodiode amplifier PDA-750 (Tetrahertz Technologies) and the signal recorded on the potentiostat which controlled the cell.
The operando electrochemical mass spectrometry setup was built in-house and is similar to the one described previously 52,53 . It consisted of a commercial quadrupole mass spectrometer (Balzers) with a turbomolecular pump (Pfeiffer) that is backed by a membrane pump and leak inlet which samples from the purge gas stream. The electrochemical cell was based on a three-electrode Swagelok design. The setup was calibrated for different gases (Ar, O 2 , CO 2 , H 2 , N 2 and H 2 O) using calibration mixtures in steps over the anticipated concentration ranges to capture nonlinearity and cross-sensitivity. During measurements either a gas mixture consisting of 95% O 2 and 5% Ar or pure Ar was used. All calibrations and quantifications were performed using in-house software. The purge gas system consisted of a digital mass flow controller (Bronkhorst) and stainless steel tubing. The procedure for the carbonate / carboxylate analysis was as described earlier 25 .
High-performance liquid chromatography coupled with mass spectrometry (HPLC-MS) was used for determining the degree of the DMA to DMA-O 2 conversion. The sample handling was performed inside an Ar filled glovebox. The electrolyte was extracted from the cell using DME that was then removed by evaporation at room temperature. The residue was dissolved in 50 µL DME and a volume of 1 µL was injected into the HPLC. The HPLC instrument was a 1200 Series (Agilent Technology, USA) with a multiple wavelength UV-Vis detector (Agilent Technology G1365C MWD SL) coupled to a mass spectrometer using atmospheric pressure chemical ionization (APCI) as ionisation method (Agilent Data availability. The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.