A rechargeable lithium–oxygen battery with dual mediators stabilizing the carbon cathode

At the cathode of a Li–O2 battery, O2 is reduced to Li2O2 on discharge, the process being reversed on charge. Li2O2 is an insulating and insoluble solid, leading ultimately to low rates, low capacities and early cell death if formed on the cathode surface. Here we show that when using dual mediators, 2,5-Di-tert-butyl-1,4-benzoquinone [DBBQ] on discharge and 2,2,6,6-tetramethyl-1-piperidinyloxy [TEMPO] on charge, the electrochemistry at the cathode surface is decoupled from Li2O2 formation/decomposition in solution. Capacities of 2 mAh cmareal−2 at 1 mA cmareal−2 with low polarization on charge/discharge are demonstrated, and up to 40 mAh cmareal−2 at rates ≫1 mA cmareal−2 are anticipated if suitable gas diffusion electrodes can be devised. One of the major barriers to the progress of Li–O2 cells is decomposition of the carbon cathode. By forming/decomposing Li2O2 in solution and avoiding high charge potentials, the carbon instability is significantly mitigated (<0.008% decomposition per cycle compared with 0.12% without mediators). The Li–O2 cell performance is largely limited by the insulating and insoluble nature of Li2O2. Here the authors report that dual mediators decouple the electrochemical reactions at the cathode from the formation and decomposition of Li2O2 from solutions, helping stabilize the carbon cathode.

T he Li-O 2 battery possesses the highest theoretical specific energy of any battery, 3,500 Wh kg −1 . If it could be realized in practice it would transform energy storage [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] . A typical Li-O 2 battery is composed of a lithium anode separated by a non-aqueous electrolyte from a porous carbon cathode, at which O 2 is reduced to Li 2 O 2 on discharge, the process being reversed on charge. Reduction of O 2 to Li 2 O 2 at the cathode of a Li-O 2 cell on discharge normally proceeds via the intermediate LiO 2 : A recent report described arresting discharge at step 1, LiO 2 , thus improving cyclability, although at the expense of specific energy (1 e − /O 2 for LiO 2 instead of 2 e − /O 2 for Li 2 O 2 ) 17 . Here we focus on Li 2 O 2 .
On discharging a Li-O 2 cell, O 2 reduction to Li 2 O 2 that grows on the carbon cathode surface leads to passivating films, resulting in early cell death (low capacity) and low rates 2,4 . This is exacerbated by the formation of Li 2 CO 3 at the interface between the Li 2 O 2 and the carbon cathode, associated with decomposition of the latter; a process that results in high charging potentials for the Li-O 2 cell. Decomposition of the carbon cathode is one of the major problems facing the Li-O 2 cell, prompting significant effort to find ways of stabilizing the carbon surface or to discover alternative cathode materials [18][19][20][21] . If, on the other hand, Li 2 O 2 grows from solution, high capacities ( Supplementary Fig. 1) and rates are possible 8,10 . Solution growth is possible using electrolyte solutions or additives that dissolve the intermediate LiO 2 or by using reduction mediators such as iron phthalocyanine, and 2,5-di-tert-butyl-1,4benzoquinone (DBBQ) 2,[22][23][24][25] . The last of these promotes solution growth without passing through LiO 2 as an intermediate, permitting the use of low-polarity electrolyte solvents, such as ethers, which are more stable but are unable to dissolve LiO 2 , and therefore unable to promote directly the solution reaction 1,23 . Solution growth introduces the need to mediate the oxidation of Li 2 O 2 on charge [25][26][27][28][29][30][31][32][33][34][35][36] . It has been shown that charge mediators provide the additional benefit of improved electrolyte stability, and this was ascribed to avoidance of reactive Li 2−x O 2 on charge 36 .
Here we describe the cycling of a Li-O 2 cell with mediators on discharge and charge, in a low-donor-number ether solvent, 1,2-dimethoxyethane (DME), that does not dissolve LiO 2 . We note that this differs from a recently published Li-O 2 battery, which operates in a flow cell configuration using mediators to form and decompose Li 2 O 2 in tanks outside the cell 37 . On discharge, DBBQ is reduced at the cathode surface and transfers electrons to O 2 in solution, reducing it to form Li 2 O 2 . On charge, 2,2,6,6tetramethyl-1-piperidinyloxy (TEMPO), is oxidized at the cathode surface and transfers electron-holes to Li 2 O 2 , oxidizing it in solution to O 2 . By decoupling completely, the growth/decomposition of Li 2 O 2 (the energy storage reaction) from the electrochemistry at the cathode surface, we obtain cycling with capacities of 2 mAh cm −2 areal at a rate of 1 mA cm −2 areal and discharge/charge potentials of 2.7 and 3.6 V, respectively. The capacity and rate are limited by poor O 2 mass transport. By combining dual mediators with a true gas diffusion electrode (which could deliver O 2 throughout the electrode akin to a fuel cell cathode) capacities of 40 mAh cm −2 areal , and at rates 1 mA cm −2 areal may be possible. This is the value modelling studies have shown is necessary for a Li-O 2 cell to achieve 500-600 Wh kg −1 (refs 6,15). The use of dual mediators avoids intimate contact, and hence reactivity between Li 2 O 2 and the carbon surface, charging voltages are well below 4 V at which carbon is known to decompose significantly to Li 2 CO 3 , and the surface electrochemistry involves only electron transfer between mediating molecules, which is less affected by Li 2 CO 3 . As a result, the use of dual mediators avoids the instability of carbon electrodes, previously dismissed as being too unstable in Li-O 2 cells. Carbon is by far the most attractive material from which to form porous cathodes; hence, demonstrating that it is significantly more stable with mediators addresses one of the major barriers to progress of the Li-O 2 battery. It should be noted that the stability of Li 2 O 2 in contact with the electrolyte solution remains one of the key challenges still to be addressed in Li-air.

The dual-mediator cell
Plots of the voltage versus capacity on discharge and charge for cells with and without dual mediators at a porous, gas diffusion layer (GDL), carbon cathode, are shown in Fig. 1. Experimental details, including construction of the cells, preparation of the electrode and electrolyte solutions, and the methods used for characterization are all described in Methods. As in previous studies of Li-O 2 cells, LiFePO 4 was used as the anode instead of Li to avoid unwanted reactions involving the latter 24,38 . We note that the potential of the LiFePO 4 anode, 3.45 V versus Li + /Li, would not result in a practical cell voltage. Without the dual mediators, the cell in Fig. 1 dies quickly, with a capacity of only 0.1 mAh cm −2 areal on the first discharge. Charging requires 4.5 V. Within five cycles the capacity is reduced to only 0.02 mAh cm −2 areal (Fig. 1b). These results are consistent with previous studies in ethers 24,39 . In contrast, with the reduction and oxidation mediators, at a current density of 1 mA cm −2 areal , a capacity of 2 mAh cm −2 areal , can be sustained, 20× greater on cycle 1 and 100× greater on cycle 5 than without mediators. The charging voltage is also much lower at ∼3.6 V. Note that the observed potentials of redox mediators are affected by the electrolyte solution composition and by the ratio of the reduced and oxidized forms present. Mediators must operate in the correct voltage range for O 2 reduction to Li 2 O 2 and vice versa 26 .
The charging voltage is set by the TEMPO-mediated oxidation and is slightly lower than that in previous studies using TEMPO 26,40,41 . The higher voltage in previous studies may arise from Li 2 O 2 film growth on the electrode surface inhibiting mediator access, and hence oxidation, whereas in the present work Li 2 O 2 grows from solution. There is a slight increase in the charging voltage towards the end of charge, which is discussed later. Scanning electron The capacity in Fig. 1a was deliberately limited to 2 mAh cm −2 areal , at 1 mA cm −2 areal because of pore clogging at the gas/electrode interface beyond this capacity, which results in polarization, then Li 2 O 2 film formation and further polarization, as discussed below. The socalled GDL electrode is not a true gas diffusion electrode, as used in a fuel cell, in which O 2 is delivered to the electrolyte/electrode interface along gas channels. It is simply a porous carbon cathode flooded with the electrolyte solution and exposed to the O 2 atmosphere only on the side opposite to the electrolyte separator, as shown in Supplementary Fig. 2. As a result, Li 2 O 2 particles form preferentially at the electrode/gas interface (where O 2 is readily available). SEM images collected from a cell in which discharge was extended beyond 2 mAh cm −2 areal (Supplementary Fig. 3a) indicate that the pores become clogged near the electrode/gas interface ( Supplementary Fig. 3c,d), impeding mass transport, and this is associated with the gentle decrease in discharge voltage and increase in charge voltage above 2 mAh cm −2 areal . Extending the depth of discharge further sees the gentle voltage decrease continue. As the voltage continues to decrease, it reaches a potential where Li 2 O 2 film formation commences and the voltage now drops rapidly to the end  of discharge. This rapid downturn is consistent with previous studies, in the absence of a mediator, that assigns the end of discharge to Li 2 O 2 film formation 42,43 . In practice, as the voltage decreases only gently when pore blocking commences, it is difficult to use a voltage cut-off for cycling. Hence discharge was arrested at 2 mAh cm −2 areal . To explore the rate capability, cells were operated at three different current densities up to 2 mA cm −2 areal (Supplementary Fig. 4a). At the highest current density the discharge voltage is lower and the onset of voltage downturn is significantly earlier. This is consistent with the rate performance being limited by O 2 mass transport, resulting in even more of the Li 2 O 2 deposition occurring at the gas/electrode interface, and earlier pore clogging. We note that the high charge potential at 2 mA cm −2 areal is due to the formation of a Li 2 O 2 film during the downturn on discharge, blocking the electrode. We also examined the effect of reducing the mediator concentration to 12.5 mM and 6.25 mM, and this had no significant effect on the performance between 25 mM and 12.5 mM ( Supplementary Fig. 4b), supporting the conclusion that the rate capability is limited by O 2 rather than mediator mass transport at those concentrations. When the mediator concentration was reduced to 6.25 mM, the discharge capacity dropped dramatically, suggesting that mediator mass transport became the limiting factor rather than the O 2 mass transport. The diffusion coefficients of O 2 , DBBQ and TEMPO were measured by cyclic voltammetry (CV), as described in the Methods under Characterization, and are 4 × 10 −5 cm 2 s −1 , 5.5 × 10 −6 cm 2 s −1 and 8.6 × 10 −6 cm 2 s −1 , respectively. Although the mediators are slower, their concentration is approximately three times higher than O 2 . Furthermore, the mediators are regenerated so rate limitations cannot be assumed based on a simple comparison of diffusion coefficients.
The results in Fig. 1 represent utilization of only about 4% of the total electrode pore volume due to these mass transport limitations. A true gas diffusion electrode would alleviate this problem, delivering O 2 evenly throughout the electrode, accessing a much higher proportion of the electrode volume, and increasing significantly the capacity to store charge and at higher rates. Assuming 80% of the porosity could be filled with Li 2 O 2 (leaving 20% of the porosity for the electrolyte) 9 , an encouragingly large capacity of 40 mAh cm −2 areal could be achieved, and at significantly higher rates. This exceeds the range of 5 to 30 mAh cm −2 estimated to be necessary for a practical Li-O 2 cell to achieve up to 500-600 Wh kg −1 (refs. 6,15). In short, the use of dual mediators mitigates limitations imposed by the formation/decomposition of Li 2 O 2 on the electrode surface, as shown in the comparison between Fig. 1a,b. The hurdle to achieving yet higher capacities at higher rates becomes one of designing a suitable gas diffusion electrode to alleviate O 2 mass transport limitations. Of course, stable electrolyte solutions are also required for reversible, extended cycling.

Product analysis
As shown previously, FTIR is very sensitive to the products of any side reactions 39,44 . There is evidence of some lithium acetate and carbonate at the end of the first discharge (Fig. 3), consistent with previous reports for Li-O 2 cells with ether-based electrolytes at the end of discharge and the Li 2 O 2 yield at the end of discharge of 88%, discussed later 14,39,45 . The yield on discharge is close to that reported in literature 46 , as the major side reaction is still due to the decomposition of electrolyte and a more stable solvent is required to overcome this problem. CH 3 CO 2 Li is oxidized on charge, confirmed by the FTIR spectra shown in Fig. 3 (ref. 44). Li 2 CO 3 cannot be effectively oxidized at 3.6 V, thus it accumulates during cycling 47,48 . Its persistence on extended cycling is seen in FTIR spectra (Fig. 3) and SEM images for cycle 10 and 50 (Fig. 2e,g), where Li 2 CO 3 is apparent as fine particles at the end of charge. The slight increase in voltage seen for all cycles towards the end of charge is discussed in the following section. The increase in voltage seen on cycle 50 across the whole of the charge is in accord with the accumulating Li 2 CO 3 (Fig. 3). Therefore, it is the build-up of Li 2 CO 3 on cycling and its accumulation on the electrode surfaces after 50 cycles that resulted in stopping cycling at cycle 50. As we discuss in the next section, the use of mediators suppresses carbon decomposition to Li 2 CO 3 , so the majority of the Li 2 CO 3 that accumulated and limited cycling is due to electrolyte solution decomposition.
The consumption and evolution of O 2 was investigated operando by monitoring the gas pressure change in the head space above the cell during discharge and charge 48,49 (the results are shown in Supplementary Fig. 5). The experiment is described in the Methods. The pressure drop on discharge, due to O 2 consumption, corresponds to 2.03 e − /O 2 , consistent with formation of Li 2 O 2 . Differential electrochemical mass spectrometry (DEMS) was also carried out during discharge and charge, and the ratio of O 2 evolved on charge to O 2 consumed on discharge is 86% ( Supplementary  Fig. 6), consistent with the yield of Li 2 O 2 , which was 88% on the first discharge. The latter was calculated by titrating the amount of Li 2 O 2 formed using ultraviolet-visible (UV-Vis) spectrometry with addition of TiOSO 4 solution 50 . Given the 2.03 e − /O 2 ratio on discharge, this suggests the CH 3 CO 2 Li and Li 2 CO 3 by-products, identified by FTIR, form by reaction with Li 2 O 2 rather than any intermediate in the O 2 reduction reaction. The curve on charge tracks that on discharge over most of its length, indicating that the reduction reaction is reversed on charge (Supplementary Fig. 5b). At the later stage of charge, all Li 2 O 2 is consumed, TEMPO + has nothing to oxidize and hence the pressure ceases to increase. This is consistent with the chemical yield of 88% and indicates the point at which all the Li 2 O 2 has been oxidized and the last ∼10% of charge capacity is due to TEMPO oxidation alone, resulting in an increasing ratio of [TEMPO + ]/[TEMPO] and hence increasing voltage 41 . This is the origin of the slight increase in voltage towards the end of charge on each cycle. The complete oxidation of Li 2 O 2 was confirmed by titration at the end of charge, in agreement with its absence from the FTIR data (Fig. 3). Yields after the 10th and 50th discharge of 84% and 75% indicate that the side reactions increase with cycling, probably due to increasing decomposition products in the electrolyte solution reacting with the Li 2 O 2 .

Carbon cathode and electrolyte solution stability
Carbon cathodes have been shown to be unsuitable for Li-O 2 cells, decomposing to form Li 2 CO 3 , with deleterious consequences for the performance, such as a large charging voltage, low rate and severe capacity fading, as illustrated in Fig. 1b (refs 3,12,38). However, this is not the case for the cells with dual mediators, comparing Fig. 1a,b. To investigate the effect of dual mediators on carbon stability, Li-O 2 cells were constructed using 13 C-carbon to form a porous positive electrode, and then subjected to cycling with and without dual mediators under the same conditions (as described in Methods under Materials preparation). The 13 C-carbon permits identification of decomposition products arising from the carbon electrode. The electrodes were extracted from the cell and treated with acid to decompose any Li 2 13 CO 3 to 13 CO 2 , which was detected by mass spectrometry. The details are given in Methods under Characterization and the results are shown in Fig. 4. They demonstrate that with mediators, the carbon decomposition is much suppressed. Less than 0.008% of the carbon electrode decomposed each cycle, compared with 0.12% for the same electrode in the absence of mediators; the latter result is very similar to previous studies 3 . Previous studies in fuel cells have shown that carbon can be decomposed above 4 V versus Li + /Li. However, studies in Li-O 2 cells have shown that carbon decomposition begins at 3.5 V and becomes significant on charging at about 3.6 V, when the carbon is also in contact with Li 2 O 2 (ref. 38). The much lower carbon decomposition and the much reduced effect on the performance of the Li-O 2 cell in the presence of dual mediators, as seen in Fig. 1, may be attributed to Li 2 O 2 oxidation not taking place at the carbon electrode surface and the charging potential, and hence oxidation, occurring all at ∼3.6 V (that is, no rise in charging potential, and therefore no increase, in decomposition) 3,38 . Also, the electron transfer reactions of the molecular mediators at the electrode are less affected by the presence of Li 2 CO 3 . This last factor is evident from the fact that the cell with dual mediators continues to cycle with amounts of Li 2 CO 3 that have negative effects on non-mediated Li-O 2 cells; compare cycle 19 with, and cycle 3 without mediators in Fig. 4.
Considering the implications for the GDL electrodes used here, McCloskey et al. have demonstrated that carbon decomposition is expected to be insensitive to the type of carbon, also both GDL and 13 C carbons were heated in 5% H 2 in Ar prior to use to promote similar surface chemistries 3 . As the carbon corrosion occurs on the  surface, and if it scales with surface area, then the GDL-based porous carbon electrode would exhibit a 0.1% decomposition after 1,500 cycles, assuming a constant rate of decomposition, as its surface area is 150-fold lower than the 13 C-carbon electrode for the same geometrical area of cathode. Carbon is the most attractive choice for the cathode in Li-O 2 cells, due to its low mass, low cost and high conductivity. Alternatives such as TiO x are less attractive. If carbon can truly be used as the cathode in Li-O 2 cells, it would remove one of the major barriers to progress of Li-O 2 batteries.
Given that the amount of carbon decomposition is very small, the major contribution to the side reactions, and hence amount of Li 2 CO 3 seen in the FTIR in Fig. 3, must come from reactions between the reduced oxygen species and the electrolyte solution. This is confirmed by analysing the amounts of Li 2 12 CO 3 formed, as this arises from electrolyte decomposition whereas Li 2 13 CO 3 was from carbon decomposition. As shown in Supplementary Fig. 7, the Li 2 12 CO 3 is significantly greater than the Li 2 13 CO 3 . The growth of both appears approximately linear with cycling. It should be noted that as a result of suppressing the carbon decomposition, the main factor now limiting reversibility is the electrolyte solution stability, emphasizing the importance of addressing this problem in future work. Note that the quantity of Li 2 12 CO 3 from the electrolyte solution decomposition with the mediators is less than 1/3 of the cell without, consistent with our previous studies of discharge using DBBQ (ref. 24). On the other hand, the long-term stability of the mediators themselves must also be addressed, posing a new challenge.
The Li-O 2 cell described here decouples the process of energy storage (Li 2 O 2 formation/decomposition from solution) from the surface electrochemistry. This is represented schematically in Fig. 5. The mechanisms of DBBQ-mediated O 2 reduction and Li 2 O 2 oxidation by TEMPO have been described 24,40 . On discharge, DBBQ is first reduced to LiDBBQ at the surface of the positive electrode. LiDBBQ then reacts with O 2 in solution to form the intermediate LiDBBQO 2 , which can either disproportionate or undergo a second reduction to form Li 2 O 2 and regenerate DBBQ (Fig. 5). On charging, TEMPO is first oxidized at the positive electrode surface to TEMPO + , which in turn oxidizes Li 2 O 2 particles in solution and in doing so regenerates TEMPO.

Cell with protected lithium anode
Although LiFePO 4 has been used here as the negative electrode, as in many previous studies 24,29,40,41 , in practical cells a lithium anode is required for a viable cell potential, and such an electrode will

Conclusions
A Li-O 2 cell using dual mediators has been cycled with capacities of 2 mAh cm −2 areal at a rate of 1 mA cm −2 areal and discharge/charge potentials of 2.7 and 3.6 V, respectively. By decoupling the electrochemical reactions at the cathode surface from the energy storage (growth/decomposition of Li 2 O 2 from solution) using dual mediators, Li-O 2 cell performance is no longer limited by the insulating and insoluble nature of Li 2 O 2 , but rather by O 2 mass transport and associated pore clogging at the electrode/gas interface. The results imply that combining dual mediators with a true gas diffusion electrode could deliver an encouraging capacity of 40 mAh cm −2 areal and meet the goal for a practical cell with 500-600 Wh kg −1 . Future work should focus, in part, on the design of gas diffusion electrodes that deliver facile mass transport, as for fuel cell electrodes. The carbon cathode is significantly more stable in the presence of dual mediators, exhibiting less than 0.008% decomposition per cycle compared with 0.12% without mediators, and it does not have a deleterious effect on the cell performance, unlike in the absence of mediators. This may be attributed to Li 2 O 2 growth/decomposition no longer taking place at the carbon surface, charging occurring at 3.6 V rather than 4 V, and the electron transfer reactions associated with the mediators at the carbon surface being less sensitive to the presence of Li 2 CO 3 . Carbon is by far the most attractive material for Li-O 2 cell cathodes due to its low cost, high conductivity and low mass, but has been dismissed due to instability. Coupled with dual mediators it may be possible to use carbon cathodes, and if so remove one of the major barriers to progress of Li-O 2 . The results also show that now, the major source of side reactions and hence limitation of cycling and reversibility arises from reactions with the electrolyte solution. This emphasizes that future research should focus on the search for more stable solvents and mediators, with the latter including investigation of oxidation mediators able to operate at lower potentials to further reduce the charging voltage. Using mediators on both discharge and charge can reduce one of the important sources of side reactions (that is, carbon instability) and can mitigate another barrier to cell performance, namely the limitations on rate and capacity imposed if Li 2 O 2 is formed and decomposed as a film on the cathode surface. If stable electrolyte solutions and appropriate mediators can be found, then our results suggest a Li-O 2 cell with mediation at the cathode could sustain cycling.
Due to the volatility of DME, an air-tight Swagelok cell with O 2 container was constructed. This was similar to Swagelok Li-O 2 cells described previously 24 , but with a closed-end tube for the cathode. O 2 was stored in a 20-cm-long stainless steel tube with two valves (Supplementary Fig. 8) and the cells were held with the O 2 tube vertical during discharge to ensure DME remained within the electrode structure. Binder-free gas diffusion layers (GDL, H2315, Quintech), made of binder-free carbon fibres, served as the O 2 electrode (4 mm in diameter) as described previously 24 . GDLs were heated under an Ar:H 2 (95: 5, v:v) atmosphere at 900 • C for three hours. The porosity of the GDLs is ∼80% and the Brunauer-Emmett-Teller surface area is below 1 m 2 g −1 (ref. 51). Glass fibre filters (Waterman) were used as the separator. A partially oxidized LiFePO 4 electrode was used as the anode 38,52 . The two-phase Li x FePO 4 has a fixed potential of 3.45 V versus Li + /Li. 25 mM DBBQ-25 mM TEMPO-300 mM LiClO 4 in DME served as the electrolyte solution. In the cell with Li metal as the anode a disk of the LiSICON (Ohara glass) was used to separate the catholyte and anolyte. 25 mM DBBQ-25 mM TEMPO-300 mM LiClO 4 in DME served as the catholyte and 300 mM LiClO 4 dissolved in tetraglyme served as anolyte. All cell components except Li were dried at 90 • C under vacuum prior to use. Cells were cycled inside an Ar-filled glove box. Galvanostatic cycling was carried out at a VMP3 potentiostat (Biologic). At the end of each charge, cells were held at 3.1 V (versus Li + /Li) for 10 min to reduce any remaining TEMPO + , as reported previously 41 . In a real battery this would simply add to the discharge capacity, but here we ensure all discharge capacity is associated with Li 2 O 2 formation and not adventitious TEMPO + reduction. Due to the significant resistance of LiSICON disk, the resistance of LiSICON was measured by impedance spectroscopy, then applying V = iR, to obtain the ohmic overpotential. This ohmic overpotential was subtracted in the load curves.
For 13 C-carbon electrodes, 13 C-carbon (Aldrich) was treated under an Ar:H 2 (95:5, v:v) atmosphere at 900 • C, as for the GDL electrodes. Its Brunauer-Emmett-Teller surface area is 150 m 2 g −1 . 13 C-carbon-PTFE (9:1, m:m) composite electrodes were constructed as described previously 38 . A stainless steel mesh was used as the current collector. The carbon loading was 2-3 mg and the geometric area 4 × 4 mm. The electrodes were dried under vacuum at 120 • C for 24 h and then transferred to an Ar-filled glove box without exposure to air.
The diffusion coefficients of O 2 , DBBQ and TEMPO were measured by CV, in which peak current is proportional to D 1/2 0 . For a given concentration, D 0 can be calculated 53 .
Characterization. For post-cycling characterization, the cells were dissembled in a glovebox and the cathode was rinsed with DME and dried under vacuum prior to further characterization. The morphologies of discharge electrodes were observed by FE-SEM (Zeiss-Merlin). FTIR spectra were collected within minutes of the end of discharge and charge with a Thermo IR spectrometer (Nicolet 6700) in a N 2 -filled glove box.
For the pressure cell and DEMS cell, GDLs served as working electrode and a partially oxidized LiFePO 4 composite electrode served as anode. The electrolyte was 25 mM DBBQ-25 mM TEMPO in 0.3 M LiClO 4 in DME. For the pressure cell, a pressure sensor (Omega Engineering) was connected to the head space of the cell to monitor the change of pressure during cell cycling. The head space of the cell was 4.95 (±0.03) ml. The cell was assembled inside an Ar-filled glovebox and filled with dried O 2 before cycling. The cell was cycled in a temperature-controlled oil bath (Julabo) at 20 • C to avoid any pressure change caused by temperature fluctuation 49 . The cells were cycled under the same conditions as for the Swagelok cells above. The DEMS cell was constructed as described previously 52 and galvanostatic discharge and charge was applied with a current of 0.5 mA. A continuous O 2 :Ar (20:80, V:V) mixture gas flow was purged through the cell as a carrier gas at a flow rate of 0.5 ml min −1 . As DME is volatile, only the first cycle of cell was recorded and an in-line trap filled with dry LiTFSI was used to absorb DME, preventing it going into the instrument and causing significant background noises.
The quantity of Li 2 O 2 formed was determined by UV-Vis spectrometry (Thermo Evolution 200) using a UV-Vis titration method reported previously 24,50,54 . The unwashed discharged electrode and separators were added to a vial containing a known amount of water; Li 2 O 2 reacts with water to produce H 2 O 2 in solution. 1 ml of this solution was mixed with 2 ml of 2% TiOSO 4 dissolved in 1 M H 2 SO 4 solution and a yellowish complex [Ti(O 2 )] 2+ (λ max = 405 nm) was formed. The UV-Vis absorption spectrum of the solution was measured and compared to a calibration curve, which was obtained using known standards of Li 2 O 2 . The typical error of the method is ±2%.
The amount of Li 2 13 CO 3 in the discharged electrode was quantified as described previously 38 . The 13 C-carbon electrodes were cycled at a current density of 1 mA cm −2 areal with a limited capacity of 2 mAh cm −2 areal , as for GDLs. Each cell was stopped at the end of discharge and the 13 C-carbon electrode and separators were extracted, rinsed with dry DME, dried under vacuum and transferred into an empty vial connected to a mass spectrometer with a carrier gas of pure Ar (BOC, Grade N5.5). A flow rate of 1 ml min −1 was applied. 0.3 ml of 2 M H 3 PO 4 was injected into the glass vial and CO 2 was evolved from the reaction between Li 2 CO 3 and acid. The amount of Li 2 13 CO 3 from decomposition of the carbon electrode was calculated from the amount of 13 CO 2 detected and the amount of Li 2 12 CO 3 from decomposition of electrolyte was calculated from the amount of 12 CO 2 detected. Data availability. Data supporting the findings of this study are available from the Oxford Research Archive (ORA) with the identifier doi: http://dx.doi.org/10.5287/bodleian:JNGX5EGnM (ref. 55). All other data are available from the corresponding author upon reasonable request.