Quantifying Changes to the Electrolyte and Negative Electrode in Aged NMC532/Graphite Lithium-Ion Cells

1Department of Chemistry, Dalhousie University, Halifax NS, Canada 2Department of Physics and Atmospheric Science, Dalhousie University, Halifax NS, Canada 3Department of Chemistry, University of New Brunswick, Fredericton NB, Canada 4Department of Community Health and Epidemiology, Dalhousie University, Halifax NS, Canada 5Department of Earth Science, University of New Brunswick, Fredericton NB, Canada 6Tesla Canada, Dartmouth, N.S. Canada

Lithium-ion batteries are currently used in a wide range of applications: cell phones, power tools, vehicles and even grid energy storage. 1 While changes to the negative electrode, 2 positive electrode 3 and engineering components 4 can improve the lifetime, safety and energy density of Li-ion cells it has also been shown that modifying electrolyte composition is a cost effective way to increase cycle and calendar lifetime and improve safety by hindering or preventing unwanted, parasitic reactions within the cell. 5 Aged lithium-ion cells with many cycles or after testing in extreme conditions show dramatic electrolyte changes. [6][7][8] Recorded electrolyte changes include oxidation at the positive electrode, 7,9 reduction at the negative electrode, 7,9 dimerization of ethylene carbonate (EC) and linear carbonates, 8 transesterification of the linear carbonates, 10 LiPF 6 break down and subsequent reactions, 7,[11][12][13] and consumption of active Li and electrolyte in solid electrolyte interphase (SEI) formation. 7,14 These reactions may be paired with electrode transformations later in cell life including changes to the positive electrode surface structure from layered to spinel 1,15 and transition metal dissolution from the positive electrode followed by deposition on the negative electrode. 16 Here, analysis methods are presented that can track some of these changes to Li-ion cells to ultimately measure the impact of these reactions throughout cell lifetime. First, a simple, modified extraction method compared to that presented by Petibon et al. 17 is introduced to obtain pure electrolyte from pouch cells. This allows the electrolyte to undergo multiple analyses to determine its organic and inorganic components. In addition, the positive and negative electrodes can be recovered from the same cells with or without exposure to air and reserved for further analyses. The methods are applied to a sample set of cells that had been operating for more than 750 cycles at various upper cutoff voltages (UCVs) and at 55 • C.
Previous studies have used μ-X-ray fluorescence (μ-XRF) to map the three-dimensional atomic composition of positive electrodes. 18 In the experiments reported here, negative electrodes were analyzed using scanning μ-XRF to study transition metal deposition at the negative electrode. 19 The XRF method for Mn was calibrated by sputtering a thickness "wedge" of Mn (from 0 to 40 μg/cm 2 ) on the surface of a fresh negative electrode and preparing a calibration curve of μ-XRF counts to Mn loading. Calibrations were not made for Ni and Co, but the sensitivity of the instrument to Ni and Co is like that for Mn.
The major outcomes of this work are that increased transesterification occurred as the upper cutoff potential increased and that the amount of transition metal dissolution from the positive electrodes of these cells was very minimal and unlikely to have impacted cell lifetime.
1.55 g/cm 3 after compression. Prior to filling, cells were cut open below the heat seal and placed in a (100 ± 1) • C antechamber under vacuum for 14 hours to remove residual moisture before transferring to an argon-filled glove box.
Cell filling.-Once in the Ar-filled glove box, cells were filled with ∼1.0 g of prepared electrolyte and vacuum sealed at 170 • C under −90 kPa gauge pressure with a vacuum heat sealer (Model MSK-115A from MTI Corporation). Cells were weighed before and after sealing to ensure no significant solvent loss.
Cell formation.-Sealed cells were first held at 1.5 V for 24 hours to help ensure full wetting of the electrodes and then transferred to a (40.0 ± 0.1) • C temperature-controlled box to undergo formation on a Maccor 4000 series charger at C/20 (11 mAh) to 3.5 V, held at 3.5 V for 1 hour and removed to degas SEI formation products in the same atmosphere and sealing conditions as above. Degassed cells then continued formation on the Maccor to their designated upper cutoff potential.
Cell cycling.-Cells were placed in a (55.0 ± 0.1) • C temperaturecontrolled box for the duration of testing and cycled using a Newware (Shenzhen, China) charging system at C/3 (80 mAh) rate between 2.8 V and the designated upper cutoff voltage. Constant currentconstant voltage (CCCV) mode was used to charge cells until the current was less than C/20 (11 mAh Gas measurements.-The volume of gas produced during formation and during cycling was measured using Archimedes principle. 21 Cells were hung below an analytical balance (Shimadzu, AUW00D) and suspended in room temperature de-ionized water (18.2 M -cm, Thermo Scientific Barnstead Nanopure Water Purification System). Because the mass of the sealed cells remains constant, the difference in weight of the cells before and after testing is measured and proportional to gas produced. Figure 1a. The method includes removing a discharged (2.0 V) jelly roll from the pouch bag and adding to a polytetrafluoroethylene (PTFE) vial containing dichloromethane (CH 2 Cl 2 ) solvent for electrolyte extraction. The PTFE vial is then machine shaken in two directions for a total of 30 minutes. Some of the extract is removed and filtered with a syringe filter containing a PTFE membrane with a 0.45 μm pore size into a second PTFE vial containing additional CH 2 Cl 2 and deionized-water for LiPF 6 and HF extraction. 17 The new solution is then shaken for another 5 minutes, centrifuged for 20 minutes at 2200 RPM (800 g) to separate the organic and inorganic layers. The bottom, organic layer is then removed for gas chromatography-mass spectrometry (GC-MS) analysis. This method extracts approximately 95% of electrolyte in wound cell but renders the remainder of the cell unavailable for additional study. Figure 1b shows a schematic of the centrifuge method. For this method, the pouch bag of discharged cell (2.0 V) was cleaned of any markings and scored on either end of the jelly roll. Previous centrifuge techniques have removed the separator to extract electrolyte, 13 however here, the entire cell was placed in a 15 mL polypropylene (PP) vial. The cap was fitted with a 9 mm PTFE/silicon seal to prevent solvent evaporation. The tube was then centrifuged at 2200 RPM (800 g) at 30 • C for 20 minutes. The sealed vials were weighed before and after centrifuging to ensure no significant solvent loss. Of the original 0.9 g of electrolyte added during cell filling, approximately 0.1g to 0.2 g of electrolyte was recovered from this process. The pure electrolyte was then removed from the PP vial with a 1 mL syringe and diluted for GC-MS and inductively coupled plasma-mass spectrometry (ICP-MS) analyses. Electrodes were separated from the jelly roll and allowed to dry in a fume hood. A representative segment of the anode was then used for μ-XRF analysis.

Modified extraction method using a centrifuge.-
The liquid-liquid extraction and centrifuge methods are compared in Figure 2. Here, cells with three different electrolyte systems were processed through both methods. Figure 2a shows 2% VC and 98% EMC, Figure 2b shows 20% EC, 50% EMC and 30% DEC, and Figure 2c shows 30% EC and 70% EMC. Results, reported in panels a, b and c, show no significant difference between samples prepared through liquid-liquid extraction or centrifuge extraction at 95% confidence intervals.
Is the electrolyte centrifuged from a cell representative of the electrolyte within the core of the jelly roll?.-A possible concern is that the electrolyte centrifuged from a cell may be dominated by electrolyte from the periphery of the jelly roll and not from the core of the jelly roll. This is because there is normally a small excess of electrolyte added to cells. Experiments were designed to probe the time needed for the electrolyte to be homogeneous between the periphery and core of a jelly roll within a pouch cell. As will be shown, the charge-discharge cycle test times of the cells tested here (about 8 months) were much longer than the time needed for the electrolyte to homogenized between core and periphery. Thus, the centrifuge method extracts electrolyte representative of that in the core of the jelly roll.
The time needed for electrolyte to be homogeneous in the cell was explored to ensure the electrolyte removed was an accurate representation of the entire electrolyte, not just the electrolyte at the periphery of the jelly roll. A fresh, dry cell was filled with 0.8 mL of 1.1 m LiPF 6 in EC:DEC (3:7), sealed and wet for 24 hours at 1.5 V to allow for electrolyte permeation into electrodes (Figure 3a). The cells were cut just below the heat seal in an Ar-filled glove box and filled with an additional 0.7 mL of DMC ( Figure 3b). The cells were then sealed again and later opened for GC-MS and ICP-MS analysis at various times after the second filling (1 hour, 1 day, 8 days and 21 days). The organic mass ratios and Li concentration were compared to the expected homogeneous compositions to determine the time required for the jelly roll electrolyte system to become homogenous (Figures 3c and 3d). Results ( Figure 4) showed that the expected homogeneous EC content was (90 ± 10) % recovered after 1 day ( Figure  4a), the expected homogeneous DEC content was (90 ± 10) % recovered after one day ( Figure 4b) and DMC was (107 ± 7) % recovered after 1 day (Figure 4c). The errors are determined by the deviation between electrolyte analysis from duplicate cells (n = 2) at a 95% confidence interval. The data shows no statistically significant difference between mass ratios of organic species measured after one day and the expected mass ratios. After 8 days, the mass ratios matched the expected ratios for a homogeneous electrolyte suggesting full mixing after 8 days.
ICP-MS analysis for Li concentration showed that after 8 days (90 ± 10) % of the expected lithium was recovered ( Figure 4d). The data shows no statistically significant difference between measured and expected lithium concentration at 95% confidence intervals after 8 days. After 21 days the expected and measured lithium contents agreed exactly. Thus, whatever electrolyte extracted using a centrifuge from a cell tested for 21 days or more will be representative of the liquid electrolyte in the core of the jelly roll.
GC-MS sample preparation.-One drop of a pure electrolyte sample was diluted into clean PTFE vials containing ∼20 mL of a CH 2 Cl 2 organic phase and ∼0.1 mL aqueous phase to ensure complete extraction of salts. The samples were then shaken for a total of 30 minutes in two directions and centrifuged for 20 minutes at 15 • C and 2200 RPM (800g) to ensure adequate extraction and separation of phases, respectively. The bottom, organic layer was then sampled using a transfer pipette for analysis. The aqueous layer is small enough that it does not cover the entire surface and the pipette is inserted to avoid the aqueous layer. To be sure that only the organic layer is used, only the bottom portion of pipetted liquid is delivered to the sample vial.
Care was taken to determine detection limits for various electrolyte degradation components expected in the electrolyte. One drop of a stock solution of known amounts EC, DMC, DEC, EMC, DE-OHC (diethyl-2,5-dioxahexane carboxylate), DMOHC (dimethyl-2,5dioxahexane carboxylate) and four other compounds was added to a known amount of CH 2 Cl 2 . This was injected to the GC-MS in exactly the same way as the electrolyte samples described in the paragraph above. Detection limits for EC, DMC, DEC, EMC, DEOHC and DMOHC were all between 30 ppm and 50 ppm by weight. That is, if there was more than 50 ppm of any of these solvents on the electrolyte, the instrument would detect it.

GC-MS sample measurement.-An
Agilent 7890 gas chromatograph coupled to an Agilent 5977B single-quadrupole mass spectrometer was used for organic analysis. The inlet was equipped with a split/spitless injector and a 30 m BR-5MS column with an inner diameter of 0.25 mm and coating thickness of 1 um. The carrier gas  was helium (99.999%) at a constant flow of 1.3 mL/min. Samples were injected into the inlet at 250 • C and carried onto the column in an oven held at 35 • C for 8 minutes followed by a ramp at 40 • C/min until the oven temperature reached 290 • C. The final oven temperature was held for 5 minutes to ensure all compounds were evolved from column. The transfer line to the MS detector was held at 250 • C. The mass spectrometer, containing an electron impact ionization module, had a 200 • C ion source and electrons had 70 eV energy. 17 A minimum 5-point calibration curve with an r 2 value of > 0.998 was generated from standards prepared on analysis day with known concentrations. Each sample and standard were injected twice to ensure reproducibility. A full scan was performed on each injection. Data analysis used the primary mass/charge ratios to identify and quantify each peak at the appropriate retention time. Figure 5 shows a schematic of the ICP sample preparation process. About 0.10 g of pure electrolyte was weighed into PTFE vials containing ∼20.0 g of aqueous 2% HNO 3 and ∼0.2 g of CH 2 Cl 2 for organic separation. Vials were shaken for 20 minutes in one direction, 20 minutes in the perpendicular direction and centrifuged at 15 • C and 2200 RPM (800 g acceleration) for 20 minutes to adequately extract the Li + into aqueous phase and separate the aqueous and organic layers. For the second dilution, approximately 0.11 g of the aqueous top layer was added to ∼ 50.0 g of 2% HNO 3 and shaken for 5 minutes. Target concentrations for the ICP-MS instrument were between 0 and 100 μg/L Li. Samples were then sealed in PP vials and stored in a refrigerator at 10 ± 2 • C until analysis.

ICP-MS sample measurement.-
On the day of sample analysis, lithium standards were diluted from 1000 mg/L stock solution with 2% HNO 3 , to produce a six-point calibration curve of 0 μg/L, 20 μg/L, 40 μg/L, 60 μg/L, 80 μg/L and 100 μg/L of Li with an r 2 value greater than 0.9999. A freshly prepared 1.1 m LiPF 6 in EC:EMC (3:7) electrolyte with known concentration also underwent sample processing and ICP-MS analysis to ensure measured concentrations were accurate.
An iCAP Q ICP-MS (Thermo Fisher Scientific, MA, USA) paired with an ESI SC-4DXS auto sampler (Elemental Scienctific, NE, USA) was used for sample analysis following the protocol developed by Smith et al. 22 All samples were run in Kinetic Energy Discrimination (KED) mode, using high purity helium (99.999%) as the collision gas. Online internal standard addition was performed to correct for any instrumental drift if necessary using 50 μg/L scandium and an SC FAST Valve (Elemental Scientific, NE, USA). A quality control check standard was analyzed every 20 samples. Samples were measured with 0.01 s dwell time and 25 sweeps. A minimum of 3 main runs with a maximum relative standard deviation of 5.0% were taken for each sample.

Robustness of ICP measurements.-The
ICP-MS method was tested periodically with electrolytes of known concentrations to ensure the robustness, reproducibility and stability of the method over time. This test included four ICP-MS samples prepared on day 1 from an electrolyte of known salt and solvent concentration. These samples were then stored in a 10 ± 2 • C refrigerator in PP vials which were wrapped with Parafilm. On day 6, an additional four samples were prepared from a new electrolyte of known salt and solvent concentration prepared the same day. Both sets of samples were measured by ICP-MS on day 6 and on day 7. The average fraction of Li recovery for both electrolyte samples on both days are shown in Figure 6. Within a 95% confidence interval, there is no significant difference between the expected Li concentration and the measured concentration on either day of measurement and after seven days of sample storage. Each of the four samples also have good agreement (within 5%). During the same sputtering run, fresh graphite negative electrodes were mounted on the substrate table next to the aluminum foil weigh discs. The same "wedge" of Mn was deposited onto the graphite negative electrodes. Several of these Mn-coated electrodes were then used to record the Mn count per unit area versus position on the electrode (Figure 7c) to be compared with the mass per unit area versus position of the weigh discs (Figure 7b), thus allowing a calibration curve to be constructed.  a Rh X-ray tube 19 using 0-50 keV range and a tube current of 200 μA. Sample scanning was done at a rate of 4.00 mm/sec. using a 40 μm spot size. Complete elemental analysis was performed at each position. The Mn signal count for the samples was then fitted using calibration data from Figures 7b and 7c to determine the Mn mass per area on the electrodes. The average Mn loading over the sampled anode area was then calculated and is reported here. Error was calculated using a regression analysis of calibration curve at 95% confidence Figure 6. Fraction of Li concentration measured compared to expected concentration, measured on day 6 and day 7 of the electrolyte storage experiment.

Mn calibration for X-ray fluorescence analysis.-A
intervals. Ni and Co μ-XRF signals were also measured from the aged electrodes and were compared to the Mn μ-XRF signals.
A negative electrode from a dry pouch cell and from a freshly formed pouch cell were measured via μ-XRF to quantitatively show the magnitude of Mn deposition before electrolyte filling and after formation, before cycling. Figure 8a shows the Mn loading on the negative electrode of a dry cell (0.0 ± 0.0 μg/cm 2 ) and Figure 8b shows the Mn loading on the negative electrode of a freshly formed cell (0.52 ± 0.09 μg/cm 2 ). Figure 8 shows images of the Mn signal maps over the approximately 3 cm × 2 cm electrode studied. The negative electrode from a dry fresh pouch cell showed no Mn signal. Similar measurements on the copper foil current collector also showed no Mn signal. Figure 9a shows the capacity vs. cycle count for the NMC532/graphite cells studied here. The cells were tested using CCCV cycling using currents corresponding to C/3 at 55 • C. The upper cutoff potentials for the tested cells are given in the legend of Figure 9. As expected, the initial capacity increases with increased upper cutoff potential. Cells cycled at higher upper cutoff potentials    Figure 10a shows gas production after cycling plotted for the five different upper cutoff potentials. There is a general trend of increased gas production as the upper cutoff potential increases, however, one cell at 4.0 V exhibited gas production comparable to the 4.4 V cells, therefore no definitive trend can be reported. The scale of gas production for these pouch cells remains small (<0.4 mL) compared to the original volume of the cells (2.5 mL). Figure 10b shows DMC and DEC solvent fractions compared to EMC plotted as the fraction of transesterification, defined here to be:  for EMC. 10,17 In freshly formed cells the presence of these products indicates the existence of Li-alkoxides, likely formed from the reduction of EMC at an un-passivated anode. 8,10,17 Li-alkoxide formation and subsequent reaction with linear carbonate are shown in Equation  1 and Equation 2, respectively. The presence of transesterification products later in life was hypothesized to originate from the passivation of dissolved transition metals upon deposition at the negative electrode. 25 [1] [2] However, limited transition metals found at the negative electrode, as discussed below, suggests another mechanism is involved. Li-alkoxides can also facilitate reactions between linear carbonates and ethylene carbonate (EC). DEOHC (diethyl-2,5-dioxahexane carboxylate) and DMOHC (dimethyl-2,5-dioxahexane carboxylate) are dimerization products also expected in the presence of Li-alkoxides as shown by Equation 3. 8,10,17 [3] DMOHC and DEOHC were not observed in any of the electrolyte samples to the detection limit of 50 ppm by weight. This suggests that Li-alkoxides were not present and another mechanism is responsible for the transesterification of linear carbonates and that the dimerization reaction of EC and EMC can be facilitated by another species. Some studies, such as that conducted by Zhang et al. have suggested that the presence of PF 5 , a decomposition product of PF 6 (see Equation 4), can act as a Lewis acid to open the EC ring. 26,27 The opened ring can then react with EMC/DEC to form DEOHC. This reaction, as shown in Equation 4, would be accompanied by Li + consumption, which is also not observed in this matrix.

Results and Discussion
LiPF 6(sol) ↔ LiF (s) + PF 5(sol) [4] Figure 10c shows the EC to linear carbonate (EMC, DMC and DEC) ratio. The EC:linear carbonate ratio was also monitored to observe any possible changes to the initial 3:7 weight ratio (shown by solid line in Figure 10c). The plotted cells show no significant difference between original EC:EMC ratio. Even cells with high fractions of transesterification have the same linear carbonate:EC ratio as the original electrolyte. This suggests that any EC or linear carbonate loss (gas formation, SEI formation) occurs at the approximately the same rate or is very small compared to the original amount of solvents in the cell. Given that only 0.4 mL of gas was found in the cell with the most gas, that 1 mL of electrolyte was added to the cells and a typical 700:1 gas:liquid volume ratio (at 1 atm pressure) for a typical liquid:gas transition one might conclude that only a small amount of solvent could have reacted. However, Ellis et al. 28 have recently shown that many gases generated in cells are also consumed so that the amount of gas remaining in a cell may not represent the total amount generated during the life of the cell. In any event, whatever the mechanism, the EC: linear carbonate ratio in these cells does not change after about 750 cycles at 55 • C. Figure 10d shows the Li + concentration remaining in the electrolyte measured by ICP-MS. The original Li + concentration is shown as the solid horizontal line. The results show no significant difference between measured Li + concentration and the original Li + concentration. This suggests that Li + has not been consumed or has been consumed at the same rate as the solvent in these cells. Readers should take care to notice that these are very good cells showing excellent capacity retention at 55 • C (see Figure 9). Figure 10e shows the Mn loadings found at the negative electrode for each cell. The results show an increase in Mn deposited at the negative electrode in cells tested at higher upper cutoff potentials. However, one pair cell at 4.0 V also shows higher amounts of Mn, making it hard to determine a definitive trend. The Mn loading found on freshly formed cells is included as the solid line to show that the amount of Mn on the negative electrode did not increase substantially for cells tested to upper cutoffs of 4.1 and 4.3 V compared to freshly formed cells. The original NMC532 positive electrode loading of 21.3 mg/cm 2 yields a Mn loading in the positive electrode of 3.6 mg/cm 2 . If 0.1% of the Mn from the positive electrode migrated to the negative electrode, a loading on the negative electrode of 3.6 μg/cm 2 would be expected. This loading level is shown in Figure 10e as the dashed line.
For spherical NMC532 particles having a radius of 3 μm (an approximation for the single crystal NMC532 used in these cells) complete removal of 0.1% of the transition metals would represent a layer that is 0.5 nm thick or one to two atomic layers thick. 20 This is certainly an underestimate of the depth from which transition metals might dissolve in the NMC532 particles. If 10% of the transition metals in near-surface layers were to dissolve, then this might impact a depth of 5 nm or so which is consistent with many TEM images of reconstructed surfaces on NMC particles after testing. Figure 11 compares the raw μ-XRF count rates for Ni (black data points) and Co (blue data points) compared to Mn as found on the negative electrodes in the tested cells. The solid black and solid blue lines in Figure 11 represent the expected ratios based on equal probabilities for dissolution of Ni, Mn and Co from NMC532. The relative sensitivity of the μ-XRF instrument was estimated by measuring the count rates per cm 2 from pure Ni, Co and Mn metal samples which were found to be in the ratio 1:0.96:0.81. These correction factors have not been applied to the data in Figure 11. The purpose of Figure 11 is to show that Ni and Co dissolve from the NMC532 positive electrode to approximately the expected extent and that Mn does not dissolve in far greater quantities. In a recent study published by Gilbert et al. 29 it was also observed that transition metals were deposited onto the anode at the expected mass ratios. Figure 12 compares results for Mn deposition on the negative electrode from this study to those recently published by Gilbert et al. 29 Gilbert et al. reported the Mn content as weight fraction of the negative electrode so it was necessary to convert the data in Figure  10 into the same units. This required some thought, because the positive and negative electrode loadings used in ref. 29 were different than those used here. Gilbert et al. used a positive loading of 9.2 mg/cm 2 while a loading of 21.3 mg/cm 2 was used here. However, because the negative/positive electrode mass ratios in the cells ref. 29 and the cells studied here were about the same, the negative electrode loading scales with the positive electrode loading. Therefore, in the end, it was only necessary to divide the Mn mass loadings in Figure 10 by the negative electrode mass per unit area (12.8 mg/cm 2 ) to get a meaningful comparison.
Gilbert et al. found that amount of Mn deposition at the negative electrode increased as the fraction of capacity loss increased. The NMC532/graphite coin cells in their study contained 1.1 M LiPF 6 in EC:EMC (3:7) and no additives. Their cells experienced 37% capacity loss after cycling between 3.0 and 4.4 V at 30 • C for only about 900 hours (200 cycles at C-rate). By contrast, cells in this study experienced a maximum of 12.2% capacity loss after cycling for over 5000 hours at 55 • C (750 cycles at C/3 rate). Figure 12 suggests that the Mn loading found on cycled anodes in this study appears to fall within the trend identified by Gilbert et al. However, it is very important to realize that the Mn loading on the negative electrodes in the cells of this study measured directly after formation (with no capacity loss) does not agree with the trend proposed by Gilbert et al., but rather suggests a relationship schematically shown by the curved solid line in Figure 12.
The data in Figure 12 for the cells in this study suggests that when transition metal dissolution is controlled through the selection of Figure 13. a) Fraction of transesterification compared to V measured at the end of the charge-discharge testing shown in Figure 9. b) Mn loading on the anode compared to V. suitable additives and/or other strategies, transition metal dissolution and subsequent deposition on the negative electrode does not play a dominant role in the failure mechanism of NMC/graphite Li-ion cells. This is a very important point that readers need to appreciate.
Gilbert et al. 29 state in their conclusions: ". . . it is our belief that controlling TM dissolution from the oxide surfaces through various possible preventive measures can lead to the development of high voltage LIBs with extended operation time. Any method of reducing the stress and fracture of oxide particles that leads to enhanced TM dissolution should reduce both capacity fade and impedance rise. These methods could include changes in oxide synthesis conditions to strengthen primary particle boundaries, oxide coatings and electrolyte additives to minimize corrosion reactions at the oxide-electrolyte interfaces, etc." The electrolyte mixture in this study likely provides more stable SEI layers on both the positive and negative electrode which prevents severe transition metal dissolution. Since transition metal dissolution has been shown to be driven by HF formation from PF 6 − , 30 it could also be that the difference in the results from this study and the Gilbert study originates from relative amounts of formed PF 5 and HF. This is also consistent with the lack of dimerization products measured in the electrolyte in this study. 31 Additionally, Li et al found that NMC532/graphire pouch cells with single crystal positive electrode materials, as used in this study, had longer cycling lifetime and better performance than cells with polycrystalline materials. 32 Figure 12 suggests that further reductions in transition metal dissolution and deposition at the negative electrode may have little impact on lifetime, however, more data points are needed, particularly at larger capacity losses, to make an adequate comparison and prediction. Figure 9b showed that the difference in average charge and discharge voltage ( V) increased with cycle number more rapidly as the upper cutoff potential increased. Since V is a measure of impedance growth in the cells it is interesting to see how it correlates to the amounts of degradation products found. The value of V at the end of the testing in Figure 9b was used for comparison. Figure 13a shows that the fraction of transesterification products (DMC and DEC) increased as V increased, suggesting that the amount of transesterification may be related the impedance growth. It would not be prudent to suggest causality without knowing the reaction mechanisms responsible for the transesterification. Similarly, Figures 13b suggests the Mn loading at the anode increases as V increases but the data shows significant scatter. Again, it is not possible to suggest a causal relationship between Mn on the negative electrode and impedance growth, when the amount of Mn on the negative electrode is very small.

Conclusions
Quantitative methods for determining the solvents present, the Li + ion concentration and the loading of transition metals transferred to the negative electrode in aged Li-ion cells were carefully described. These methods were applied to a set of single crystal NMC532/graphite cells with effective electrolyte additives that had been tested at 55 • C for about 8 months. The cells were charged and discharged between 2.8 V and various upper cutoff potentials ranging from 4.0 to 4.4 V using currents corresponding to C/3. The cells all attained about 750 cycles over a 9-month period with approximately 10% capacity loss.
The results showed that increased upper cutoff potential increased the amount of the transesterification products DMC and DEC. The increased fraction of transesterification products in cells tested to higher potential correlated well to the increase in impedance growth rate (Figures 9b and 13a) in cells tested to higher potential. This suggests that the mechanism of impedance growth is associated with electrolyte oxidation at the positive electrode side and that transesterification is an indicator or a product of these reactions.
The amount of transition metals transferred from the positive to the negative electrode was less than 0.1% of all the transition metals in the positive electrode for all the cells tested. There may be an increase in the amount of transition metals transferred with upper cutoff potential, but scatter in the data makes it hard to discern if this trend is real. Cells tested at 4.1 V and at 4.3 V for 8 months and 55 • C showed only about twice as much Mn on the negative electrode as freshly formed cells (Figure 10e). It is therefore difficult for the authors to believe that transition metal dissolution in these cells is a significant contributor to cell capacity loss.
There was no significant change in salt content nor EC:linear carbonate ratio in these cells compared to the original electrolyte even for cells tested to 4.4 V. These results are not too surprising given that all cells had less than 10% capacity loss and were all decreasing in capacity at similar, steady rates with no indication of cell failure. Future studies will consider older cells with more extreme capacity loss.
The results from this study were compared to those of a similar study by Gilbert et al. 29 The amounts of Mn and other transition metals found on the negative electrode were an order of magnitude smaller than the Gilbert et al. study, even though cells studied here were tested for 5 times longer and at 55 • C (this work) instead of 30 • C (Gilbert et al.). These results point to the importance of electrolyte additives and other strategies in controlling transition metal dissolution.