Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1628 0013-4651/2003/150共12兲/A1628/9/$7.00 © The Electrochemical Society, Inc. Decomposition of LiPF6 and Stability of PF5 in Li-Ion Battery Electrolytes Density Functional Theory and Molecular Dynamics Studies Ken Tasaki,a,*,z Katsuya Kanda,b Shinichiro Nakamura,b and Makoto Uec,* a Mitsubishi Chemical Research and Innovation Center, Goleta, California 93117, USA Mitsubishi Chemical Corporation, Science & Technology Research Center, Computational Modeling of Materials Laboratory, Yokohama, Kanagawa 227-8502, Japan c Mitsubishi Chemical Corporation, Science & Technology Research Center, Electrochemistry Laboratory & Battery System Design Laboratory, Inashiki, Ibaraki 300-0332, Japan b The decomposition of LiPF6 and the stability of PF5 in organic solvents, diethyl carbonate 共DEC兲, dimethyl carbonate 共DMC兲, ␥-butyrolactone 共GBL兲, and ethylene carbonate 共EC兲, have been investigated through density functional theory 共DFT兲 calculations, in which solvent was modeled as a dielectric continuum, and also by molecular dynamics 共MD兲 simulations which treated solvents explicitly. Both calculations showed a similar trend in which the decomposition was further promoted in more polar solvents, yet the DFT calculations predicted an endothermic decomposition, while the MD simulations indicated exothermic. This sharp contrast in the results suggests strong solute-solvent interactions, especially for PF5 , which were not accounted for in the DFT calculations. The specific interaction between PF5 and solvent was further investigated by DFT calculations for adduct models and also by the MD simulations for solutions. Both calculations suggest a stable formation of a PF5 -solvent adduct in solution and its stability depends on the solvent. It was found that PF5 is more stabilized in polar and sterically compact solvents such as EC and GBL than in less polar and bulky, linear carbonates such as DMC and DEC. The reactivity of PF5 with organic solvents and the difference in the stability of LiPF6 between organic and aqueous solution are also discussed. © 2003 The Electrochemical Society. 关DOI: 10.1149/1.1622406兴 All rights reserved. Manuscript submitted December 23, 2002; revised manuscript received May 31, 2003. Available electronically October 17, 2003. Lithium hexafluorophosphate (LiPF6 ) is by far the most widely used electrolyte salt in lithium-ion batteries 共LIBs兲. Electrolyte solutions of LiPF6 dissolved in binary or ternary solvents which include cyclic carbonates such as ethylene carbonate 共EC兲 or propylene carbonate 共PC兲 and linear carbonates such as dimethyl carbonate 共DMC兲, diethyl carbonate 共DEC兲, ethyl methyl carbonate 共EMC兲, or their mixtures show high electrolytic conductivity, electrochemical stability, and enough thermal stability, although dry, nonsolvated LiPF6 exhibits poor chemical stability.1 However, there are still some safety concerns associated with the improvement in safety of the LIBs, and understanding the thermal stability of the electrolytes is essential in the design of safe LIBs.2 The thermal decomposition produces pentafluorophosphate (PF5 ) by LiPF6 → LiF ⫹ PF5 , which in turn reacts with solvents to initiate polymerization of solvents3 or to give other products such as phosphine oxides, alkyl fluorides, and ethers.4 It is well known that the reaction between LiPF6 in electrolyte solutions and trace amounts of water 共or alcohols兲 produces HF,5,6 which causes a detrimental effect in battery performance. The initial reaction between water and LiPF6 is described by the overall reaction: LiPF6 ⫹ H2 O → LiF ⫹ POF3 ⫹ 2HF. However, little is known about the reaction mechanism of this hydrolysis. It was proposed that PF5 , a product of the equilibrium of LiPF6 LiF ⫹ PF5 , undergoes hydrolysis with residual water to produce HF as PF5 ⫹ H2 O → POF3 ⫹ 2HF. 7,8 This is based on the experimental facts that LiPF6 aqueous solutions are very stable and their HF content can be determined by aqueous NaOH titration without further HF formation by the LiPF6 hydrolysis.6,8 Therefore, understanding the salt decomposition and its process is foremost in controlling the stability of LiPF6 in electrolyte solutions. Yet the difficulty of probing PF5 in solution hinders the chemical analysis of the reactions involved in the LiPF6 decomposition, although several experimental studies on the reactivity of these species have been reported.3,4,8 Computer simulation may be a useful tool for examining chemi- * Electrochemical Society Active Member. z E-mail: [email protected] cally very unstable compounds.9,10 In this study we attempt to analyze the reactivity and stability of chemical components in the salt decomposition, specifically, LiPF6 and PF5 , in organic solvents by density functional theory 共DFT兲 calculations and molecular dynamics 共MD兲 simulations in order to shed light on the salt decomposition and reactions that follow. The salt decomposition may depend on a number of environmental variables: temperature, impurity, salt concentration, solvent polarity, and others. Our interest here is in the solvent dependence of the salt decomposition in an attempt to design electrolytes to control the decomposition.11,12 Computational Protocols DFT calculations.—The solvents examined here included DEC, DMC, ␥-butyrolactone 共GBL兲, and EC to cover a wide range of solvent polarity with the dielectric constant 共⑀兲 spanning from 2.8 共DEC兲 to 90 共EC兲. DFT calculations were performed for the following reactions in each solvent by self-consistent reaction field 共SCRF兲 methods13 at the level of B3LYP/6-31G共d,p兲14,15 in conjunction with the Onsager model16 to include the solvent effect LiPF6 → LiF ⫹ PF5 关1兴 PF5 ⫹ S → PF5 ⫺ S 关2兴 The product in Eq. 2 represents an adduct formation of PF5 with a solvent 共S兲 which may lead to a cascade of reactions that follow the decomposition. Here it was intended to examine the PF5 stability in each solvent. Gaussian 9817 was used for all ab initio and DFT calculations. Table I lists the dipole moment of each solute, calculated at the SCRF-HF/6-31G共d兲 level, and the solute cavity, obtained from a spherical volume of each solute calculated at HF/6-31G共d兲 level, defined as the volume inside a contour of 0.001 electrons/bohr3 density.18 MD simulations.—MD simulations were performed for a system with the reactant or the product 共one molecule兲 as a solute in Eq. 1 in each solvent 共50 solvent molecules兲 under normal temperature and pressure conditions at T ⫽ 300 K. Solvent molecules which had van der Waals overlaps with the solute were removed from the system upon immersion of the solute molecule in the box of solvents. Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1629 Table I. Calculated dipole moments „… and solute cavities „a… of solutes in LiPF6 decomposition. , D a, Åb a a b LiPF6 LiF PF5 7.08 3.55 6.17 2.61 0.00 3.38 Obtained from SCRF-HF/6-31G共d兲 calculations. See the text. The effective concentration of each solution was around 1 ⫻ 10⫺4 M and the density ranged from 1.09 to 1.18 g cm⫺3. The concentration was set to be very small to eliminate interactions between the solutes. This was necessitated by calculations of thermodynamic properties. COMPASS™19 was used for the force field. The simulations were carried out by using a Cerius2 package.20 The Ewald summation was employed for evaluating long-range electrostatic interactions. Each system was first equilibrated long enough for the ensemble properties 共pressure, temperature, potential energy, etc.兲 to settle, typically for 50 ps. Production runs were then performed for 500 ps more. The potential energy used for the thermodynamic analysis was time-averaged from the production runs. Results and Discussion Previously we detected a small amount of POF3 and PO2 F⫺ 2 as the decomposition products in a LiPF6 -based electrolyte solution by 19 F-NMR as shown in Fig. 1; however, no PF5 was observed.8 Even immediately after the PF5 gas was introduced into a pure PC solvent in a nuclear magnetic resonance 共NMR兲 sample tube, no product other than POF3 was detected, which shows the extremely high reactivity of PF5 with a carbonate solvent. Thus, we decided to use computer simulation as a tool for examining the stability and the reactivity of PF5 in various solvents. DFT calculations.—We start with the gas-phase decomposition which serves as a reference point in this work. Figure 2 illustrates the structures of solutes, LiPF6 and PF5 . The optimized geometries of all solutes and the atomic charges are listed in the first column of Table II, obtained from HF/6-31G共d兲 calculations. In the optimized structure of LiPF6 , the Li atom coordinates with three F atoms in equidistance, with Li and three F’s forming a tripod in the most stable, undisturbed state. The distance between Li and F is 1.882 Å. In order for the decomposition to occur, Li has to move to the extension of one of the P-F bonds with the P-F¯Li angle being Figure 1. The 19F-NMR spectrum for an electrolyte solution, 1 M LiPF6 /EC-EMC 共3:7 in volume ratio兲, after the salt decomposition. The estimated concentration for PO2 F⫺ 2 and POF3 were 93 and 9 ppm, respectively. Figure 2. The optimized geometries of LiPF6 and PF5 at the HF/6-31G共d兲 level in the gas phase. The structure of PF5 is D 3h . See Table II for the detailed geometries. 180°, the same direction as the P-F anti-bonding orbital. This coordination allows the maximum overlap of atomic orbitals between the Li and the F atoms, thus for a reaction of Li and F to take place, producing LiF and PF5 . When enough thermal energy is given to the salt, this movement can take place relatively easily. Conversely, in a highly polar solvent, the salt tends to dissociate itself, thus removing the Li and the F distance further away from each other, reducing the possibility of decomposition. The optimized geometry for PF5 has the D 3h symmetry, and the transitional state of the pseudorotation with the C 4v symmetry 共not shown兲 was also found at a higher energy by 5.6 kcal mol⫺1. The free energy differences for the decomposition in the gas phase calculated at two temperatures, 300 and 400 K, are listed in the first two columns of Table III. As is shown, the decomposition is endothermic in the gas phase for which the level of calculations used here is highly accurate. This result demonstrates the inherent stability of LiPF6 and suggests that the decomposition in solid or solution is caused by the thermal and the condensed phase effect. Next we introduce solvents to the system. It is not known how critically the solvent plays a role in the salt decomposition. There are basically two models treating solvents: a model treating solvent implicitly and one treating explicitly. The former is based on the Onsager-type reaction field theory,21 treating solvent as a dielectric continuum, collectively known as the SCRF methods, representing simple, less computationally expensive methods, yet sometimes resulting in inaccurate description of solute-solvent interactions. In the SCRF treatment, in general, a solute is placed in a cavity in a continuum of solvent with the cavity representing the volume of the solute. In light of interpreting the results, further description of the model may be helpful. In the simplest model, a solute occupies the fixed spherical cavity within the solvent field. A dipole of the solute induces the polarization of the surrounding solvent, which in turn interacts with the solute dipole, stabilizing the solute. The stabilization energy, ⌬E, depends on the solute dipole moment, , the cavity, a, and the dielectric constant of the solvent, Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1630 Table II. Optimized geometries and atomic charges for solutes in various solvents.a Table III. Thermodynamic properties for decomposition of LiPF6 in various solvents „kcal molÀ1…. LiPF6 r LiP r PF1 b r PF4 c r LiF1 d F1 PF2 e F1 PF4 f F4 PF5 g LiF1 Ph q Li qP qF Gas DEC DMC GBL EC 2.432 1.662 1.561 1.882 83.9 90.6 94.4 86.4 0.68 2.07 ⫺0.46 2.471 1.650 1.569 1.922 84.5 90.6 93.6 87.1 0.70 2.07 ⫺0.46 2.474 1.650 1.569 1.926 84.6 90.7 92.6 87.1 0.70 2.06 ⫺0.46 2.522 1.639 1.576 1.978 85.4 90.8 92.7 88.1 0.72 2.04 ⫺0.46 2.525 1.638 1.576 1.979 85.4 90.9 92.6 88.1 0.72 2.04 ⫺0.46 Gas DEC DMC GBL EC 1.555 0.66 ⫺0.66 1.583 0.68 ⫺0.68 1.584 0.68 ⫺0.68 1.611 0.70 ⫺0.7 1.611 0.70 ⫺0.7 Gas DEC DMC GBL EC 1.534 1.568 1.535 90.0 120.0 1.95 ⫺0.39 1.535 1.567 1.534 90.0 120.0 1.95 ⫺0.39 1.535 1.567 1.534 90.0 120.0 1.95 ⫺0.39 1.535 1.567 1.534 90.0 120.0 1.95 ⫺0.39 1.535 1.567 1.534 90.0 120.0 1.95 ⫺0.39 ⌬E 1 ⌬H 1 ⌬S 1 ⌬G 1 Gasa Gasb DECa DMCa GBLa ECa 1 52.72 52.01 0.041 39.78c 1 52.72 44.43 0.039 28.92c 2.3 49.43 48.71 0.040 36.66d 3.1 49.21 48.48 0.040 36.45d 42 47.16 46.43 0.039 34.71d 90 47.02 46.29 0.039 34.59d a At 300 K. At 400 K. ⌬G 1 ⫽ ⌬H 1 兵 ⌬E 1 关 B3LYP/6-31G共d, p) 兴 ⫹ thermal correction 关HF/6-31G共d)] 其 ⫺ T⌬S 1 关 HF/6-31G共d)]. d ⌬G 1 ⫽ ⌬H 1 兵 ⌬E 1 关 SCRF-B3LYP/6-31G共d, p) 兴 ⫹ thermal correction 关SCRF-HF/6-31G共d)] 其 ⫺ T⌬S 1 关 SCRF-HF/6-31G共d)] with Onsager’s model. b c LiF r LiF q Li qF ⌬G 1 ⫽ G 1 共 LiF兲 ⫹ G 1 共 PF5 兲 ⫺ G 1 共 LiPF6 兲 PF5 r PF1 r PF2 i r PF4 j F1 PF2 k F1 PF4 l qP qF r, , and q refer to the distance in angstroms, the angle in degrees, and the atomic charge, respectively. All results were obtained from SCRF-HF/6-31G共d兲 calculations with the Onsager model, except for the gas phase for which HF/6-31G共d兲 calculations were used. b r PF1 ⫽ r PF2 ⫽ r PF3 . c r PF4 ⫽ r PF5 ⫽ r PF6 . d r LiF1 ⫽ r LiF2 ⫽ r LiF3 . e F1 PF2 ⫽ F2 PF3 ⫽ F1 PF3 . f F1 PF4 ⫽ F1 PF5 ⫽ F2 PF4 ⫽ F2 PF6 ⫽ F3 PF5 ⫽ F3 PF6 . g F4 PF5 ⫽ F4 PF6 ⫽ F5 PF6 . h LiF1 P ⫽ LiF2 P ⫽ LiF3 P . i r PF2 ⫽ r PF3 . j r PF4 ⫽ r PF5 . k F1 PF2 ⫽ F1 PF3 ⫽ F2 PF4 ⫽ F2 PF5 ⫽ F3 PF4 ⫽ F3 PF5 . l F1 PF4 ⫽ F1 PF5 ⫽ F4 PF5 . a ⌬E ⫽ 冉 冊 ⫺ 1 2 2 ⫹ 1 a 3 vent’s role in the reactions in question, provided that the explicit method adequately models solute-solvent interactions. We start with the implicit model to calculate the following free energy difference for the decomposition 关3兴 Thus, a greater dipole moment is further stabilized by more polar solvent. Such a treatment of solvent significantly reduces the cost of computational time, compared to calculating solute-solvent interactions individually, but it may not be recommended in general for a system with strong solute-solvent interactions. Alternatively, the second model, treating solvent molecules explicitly, thus specific solute-solvent interactions being accounted for, is yet more computationally demanding. MD simulations generally fall in this category. Thus, which protocol is chosen is a trade-off between cost and accuracy. Here, we have chosen both models, implicit and explicit. A benefit is that a comparison between two such protocols inevitably discloses the consequence of the difference in treating solvents, thus revealing a degree of significance of the sol- 关4兴 The optimized geometries and the atomic charges are summarized in Table II, obtained from SCRF-HF/6-31G共d兲 calculations and the CHelpG method22 using the SCRF-HF/6-31G共d兲 density, respectively, for each solute in each solvent. The changes in the geometries and the atomic charges are what is generally expected from the treatment used, SCRF. The deviation from the gas phase, the geometry or the atomic charge, becomes greater as the polarity of the solvent increases. Note that the distance between Li and P of LiPF6 increases in going from DEC to EC by 2%, predicting qualitatively the experimental trend of a larger dissociation of LiPF6 in a more polar solvent.23,24 The exception is PF5 whose geometry and atomic charge show little change over a wide range of solvents. This is due to PF5 ’s nonpolarity, having no dipole moment 共see Table I兲 due to its molecular symmetry. This invariability of both the geometry and the atomic charge of PF5 reflects on the following thermodynamic properties. Table III summarizes the thermodynamic properties (⌬E 1 , ⌬H 1 , ⌬S 1 , and ⌬G 1 ) obtained from SCRF-DFT calculations at the B3LYP/6-31G共d,p兲 level in various solvents represented by their dielectric constants 共⑀兲. The frequency calculations were performed at the level of SCRF-HF/6-31G共d兲 from which the thermal corrections to the enthalpy and the entropy were obtained. Though the entropic contributions, T⌬S, are not small, ⬃12 kcal mol⫺1, the values stay almost the same across the solvents and are thus inconsequential to the comparison among the solvents. As is known, LiF is insoluble in the solvents used in this study, so the numbers in Table III cannot be directly compared to experimental data, if any. Still, they offer helpful insight for discussion regarding the relative stability of each solute, which is the main objective of this work. According to the SCRF-DFT calculations, the decomposition is largely endothermic in these solvents and shifts more toward the products with increasing polarity of the solvent, which is expected from the model used. The relative stability of each solute expressed by the solvation free energy, ⌬G solv关 ⫽ ⌬G 1 (solvent)⌬G 1 (gas) 兴 , is illustrated in Fig. 3 as a function of the dielectric constant of the solvent. The stability of PF5 stays virtually unchanged, whereas the other solutes become more stable as the polarity of the solvent increases with the stability increase of LiF being the largest. This is in general explained by Eq. 3. Though the dipole moment of LiPF6 is higher than that of LiF, its larger cavity offsets the solvation energy by a power of 3. This model does not treat solvent molecules individually. A nonpolar solute such as PF5 does not change its stability over a wide range of solvents due to its very small dipole moment, Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1631 Table IV. Free energies for PF5 -solvent adduct „kcal molÀ1….a ⌬G 2 a Figure 3. The solvation free energy of each solute as a function of the dielectric constant of the solvent obtained by SCRT-DFT calculations. but experimental data seem to show otherwise. The amount of PF5 in different solvents has been reported to vary from solvent to solvent.4 The experimental findings seem to contradict with the results from the implicit model. In order to take into account a thermal effect, the frequency calculations were also performed at 400 K, at which the salt decomposition was clearly observed.4 Yet, this only reduces the free energy by around 10 kcal mol⫺1, leaving the reaction still endothermic. Next, we improve the model by using a more explicit treatment of the solvent. Since the previous decomposition reactions were estimated to be all endothermic, our attention was directed toward the stability of the products which may be underestimated by the continuum model. PF5 is a strong Lewis acid and may form a solvent adduct in solution. In this calculation, a solvent adduct of PF5 was immersed in each solvent continuum in order to treat somewhat explicitly the PF5 -solvent interaction. Similar calculations were reported for the PF5 -H2 O adduct in the gas phase.9 Here, the structures of the adducts were constructed from the optimized gas-phase structure of PF5 with the P atom coordinating with the carbonyl oxygen of the solvent, then optimized in each solvent. The adducts with the P atom pointing at the ether oxygen of the solvent were also examined. It was found that the former adduct formation is more stable than the latter for all the solvent. Thus, the results for the former are only shown here. The optimized geometry obtained from SCRF-HF/6-31G共d兲 calculations in solution are displayed in Fig. 4. The structure of PF5 in each adduct has no longer the D 3h symmetry due to the interactions with the adjacent solvent. Table IV lists the free energies (⌬G 2 ) of each PF5 -solvent adduct calculated in each solvent, obtained similarly to the calculations in Table III, for the process defined by Eq. 2. As the negative values of the free energies indicate, the process is voluntary for the all the solvents, moreso in a polar solvent, suggesting a stable formation of DEC DMC GBL EC ⫺1.47 ⫺1.85 ⫺5.2 ⫺5.3 Obtained from SCRF-B3LYP/6-31G共d,p兲 calculations with Onsager’s model. PF5 -solvent adduct in solution. The results are also in line with the experimental observation that the reactivity of PF5 depends on the solvent.4 To examine the differences in the relative stability of PF5 adducts, the atomic charges were calculated for each adduct based on the CHelpG method using the SCRF-HF/6-31G共d兲 density, shown for the carbonate group or the carboxyl group of each adduct by italic in Fig. 4. The charges of the carbonate group are larger than those of the lactone group due to the greater polarization around the O-C(⫽O兲-O moiety than the C-C(⫽O兲-O moiety. The higher charges of the carbonyl and the ether oxygens in general result in unfavorable interactions with the electron-rich F atoms of PF5 , which make the adduct less stable. These interactions are reflected on the geometries of each adduct, especially on the distances between the solvent and PF5 , with the longer P-共C⫽)O distances for the linear carbonate solvents, as is seen in Fig. 4. The smaller oxygen charges give rise to a more stable adduct in GBL. In fact, PF5 -GBL adduct has the shortest P-共C⫽)O distance. The large charges of the carbonate group are screened by the polar solvent, EC. Unfavorable interactions with the F atoms are thus shielded, having a comparable geometry to that of the GBL adduct. These results have an important implication in the reactivity of PF5 in different solvents. The stability of these adducts affects the reaction of PF5 with solvent that follows the LiPF6 decomposition, and these adducts could well be some of the reaction intermediates. For example, a similar adduct PF5 -H2 O with the P atom coordinating with the water oxygen is an intermediate for PF5 hydrolysis.25 The adduct stabilization (⌬G 2 ) energy was added to Eq. 4 to give the following equation ⌬G ⬘ ⫽ ⌬G 1 ⫹ ⌬G 2 关5兴 Table V gives the values of resultant ⌬G ⬘ . The decomposition processes become less endothermic than they are with ⌬G 1 alone, and yet, are still prohibited in solution. In the next model we treat solvent explicitly. Figure 4. The structures of PF5 -solvent adducts used for SCRFDFT calculations with the adduct P-O bond drawn in the figure. The geometry was optimized by HF/6-31G共d兲 geometry optimization starting with the P atom pointing at the carbonyl oxygen. The optimized distances and the atomic charges are also partially included. The P-O-C bond angles are 132° for all the adducts. See the text for the atomic charge calculations. Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1632 Table V. Free energies for LiPF6 decomposition recalculated from Table IV „kcal molÀ1…. ⌬G ⬘ a a DEC DMC GBL EC 35.19 34.59 29.51 29.29 Table VI. Enthalpies of decomposition of LiPF6 in various solvents „kcal molÀ1…. Enthalpies and their components DEC LiPF6 ⌬G ⬘ ⫽ ⌬G 1 ⫹ ⌬G 2 . E gas solvent E sol E sol ⌬E sol ⌬H sol MD simulations.—To treat the solvent effect more rigorously, we performed MD simulations in which all interactions between the solute and the solvent and among the solvent molecules are explicitly included. Using the thermodynamic cycle shown in Fig. 5, one decomp may estimate the decomposition enthalpy in solution (⌬H sol PF5 LiPF6 LiF ⫽ H sol ⫹ H sol ⫺ H sol ) according to Eq. 6 PF LiPF6 decomp decomp LiF ⫽ ⌬H gas ⫹ ⌬H sol ⫹ ⌬H sol5 ⫺ ⌬H sol ⌬H sol DMC E gasa solvent E sol E sola ⌬E sol ⌬H sol 关6兴 where ⌬H sol represents the heat of solution, e.g. PF PF ⌬H sol5 ⫽ ⌬E sol5 ⫹ P⌬V ⫺ RT PF PF PF solvent ⌬E sol5 ⫽ E sol5 ⫺ E sol ⫺ E gas5 关7兴 关8兴 where P, V, R, and T are the pressure, volume, gas constant, and PF PF absolute temperature, respectively, and E sol5 and E gas5 the internal energies of PF5 in solvent and in the gas phase, respectively, and solvent E sol the internal energy of solvent. The simulations were performed under zero pressure; thus, P⌬V disappears in Eq. 7. Each quantity in Eq. 6-8 was obtained from an ensemble average over 500 ps of simulation for a box of solvent and one solute, LiPF6 , PF5 , or LiF (Esol), and over 100 ps for those in the gas phase (E gas). For such simulations, the thermodynamics properties can depend on the initial configuration of the system. The final values were obtained by averaging over five independent simulations to increase the statistical significance of each property. Table VI summarizes the results from the MD simulations. For decomp ⌬H gas the values from the DFT calculations in the gas phase were used. Again, these values are primarily used for comparison of the relative stability of each solute. The number of solvent molecules in each system was 50 for the LiF solution and 49 for the LiPF6 and the PF5 solution, respectively. The enthalpy of decomposition now is negative in all solvents, the process being more exothermic with the increasing polarity of the solvent. Figure 6 illustrates the heat of solution for each solute as a function of the dielectric constant of the solvent. The dependence of PF5 stability on the solvent is well demonstrated and the contrast from Fig. 3 is apparent. The sharp contrast in the results between the DFT calculations and the MD simulations unambiguously suggests the strong interactions of the solute with the solvent, especially for PF5 , because the difference in the results between the two protocols, the implicit and the explicit model, is most profound. In order to further examine the PF5 -solvent interactions, the trajectories from the MD simulations were inspected. Figure 7 depicts the first shell solvation of each solute shown by snapshots taken Figure 5. The thermodynamics cycle for the LiPF6 decomposition. GBL E gasa solvent E sol E sola ⌬E sol ⌬H sol EC E gasa solvent E sol E sola ⌬E sol ⌬H sol a a a LiF PF5 10.51 ⫺2636.69 ⫺2723.14 ⫺96.96 ⫺97.56 ⫺171.88 ⫺2689.42 ⫺2947.71 ⫺86.41 ⫺87.01 64.98 ⫺2636.69 ⫺2638.68 ⫺66.97 ⫺67.57 10.51 ⫺1640.18 ⫺1747.71 ⫺118.04 ⫺118.64 ⫺171.88 ⫺1672.98 ⫺1953.59 ⫺108.72 ⫺109.32 64.98 ⫺1672.98 ⫺1675.63 ⫺67.63 ⫺68.22 10.51 ⫺1343.60 ⫺1470.92 ⫺137.83 ⫺138.43 ⫺171.88 ⫺1343.60 ⫺1651.52 ⫺136.04 ⫺136.64 64.98 ⫺1343.60 ⫺1373.36 ⫺94.74 ⫺95.33 10.51 ⫺3316.15 ⫺3436.08 ⫺130.44 ⫺131.04 ⫺171.88 ⫺3382.47 ⫺3641.50 ⫺129.67 ⫺130.27 64.98 ⫺3316.15 ⫺3346.45 ⫺95.28 ⫺95.88 Enthalpies of decomposition DEC decomp ⌬H gas LiPF6 ⌬H sol LiF ⌬H sol PF ⌬H sol5 decomp ⌬H sol a DMC GBL EC 52.01 52.01 52.01 52.01 ⫺97.56 ⫺118.64 ⫺138.43 ⫺131.04 ⫺87.01 ⫺109.32 ⫺136.64 ⫺130.27 ⫺67.57 ⫺68.22 ⫺95.33 ⫺95.88 ⫺1.33 ⫺3.21 ⫺41.53 ⫺43.10 The errors for the condensed phase simulations are within 10 and 0.2 kcal mol⫺1 for the gas-phase simulations. from the trajectories. It was found that PF5 is well solvated by the cyclic solvents, EC and GBL, closely surrounded by the solvent molecules, whereas the linear carbonates are somewhat distant from the solute, creating an extra cavity between the solute and the solvent. According to Eq. 3, a larger cavity gives rise to a poor solvation energy in general. Pair distribution functions were calculated from the trajectories to evaluate the number of solvent molecules near PF5 . Table VII summarizes the solvation of PF5 by each solvent. The numbers of the carbonyl oxygen, the ether oxygen, and the methyl carbon 共DMC and DEC兲 or the methylene carbon 共EC and GBL兲 are calculated by integrating the relevant pair distribution function up to 5 Å. In the case of EC, the number of the ether oxygen can include those in the same molecule or those in different molecules due to the symmetry of the molecules. The results in the tables demonstrate the stronger affinity to PF5 by EC and GBL, compared to the linear carbonates. The number of the carbonyl oxygen within 5 Å from the P atom of PF5 in EC or GBL is almost eightfold higher than that in DEC. Further, the total number of oxygens tends to be higher for the polar solvents than the nonpolar solvents. These observations can be rationalized in that in addition to the polarization effect by the solvents, the sterically compact solvents such as EC or GBL have a Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 A1633 Table VII. The solvation structure around PF5 in various solvents. a r P¯共C-兲O⫽ , Å r P¯共C-兲O- , b Å n (C-兲O⫽c n (C-兲O- d n CH3 /CH2 e DEC DMC GBL EC 5.21 4.71 0.37 1.98 1.07 3.63 4.21 1.68 1.63 1.08 3.63 3.77 3.09 3.01 2.07 2.95 4.85 2.92 1.54 1.68 a Figure 6. The heat of solvation for each solute in various solvents as a function of the dielectric constant of the solvent obtained from MD simulations. better access to PF5 , while the bulky, flexible molecules such as DEC create a large cavity between PF5 and the solvent. As to the solvation site, the polar solvents tend to solvate the P atom of PF5 with the carbonyl oxygen, while the less polar solvents are likely to solvate with the ether oxygen. The solvation by GBL and DMC falls somewhere in between EC and DEC. In particular, in the case of GBL, a larger negative atomic charge was assigned for the ether oxygen than for the carbonyl charge in the force field used for simulations. What plays in the preference in the solvation site is not clear at this point. Table VIII lists the values for the Li-P distance of LiPF6 averaged over 500 ps of simulations in various solvents, compared to those obtained from the SCRF-HF optimizations listed in Table II. It is clear that the SCRF-HF calculations largely underestimate the dissociation of the salt pair, the Li-P distance, moreso in the polar solvents. Our previous study showed that the trend in the dissociation of LiPF6 in various organic solutions can be reasonably reproduced by MD simulations.12 Comparison of two protocols.—Though the results from MD simulations are consistent with the SCRF-DFT calculations in that the decomposition is further promoted in more polar solvent, it is The averaged shortest distance between the P atom and the carbonyl oxygen of solvent. b The averaged shortest distance between the P atom and the ether oxygen of solvent. c The number of the carbonyl oxygen of solvent within 5 Å from the P atom of PF5 . d The number of the ether oxygen of solvent within 5 Å from the P atom of PF5 . e The number of the methyl 共DEC or DMC兲 or the methylene 共GBL or EC兲 carbon of solvent within 5 Å from the P atom of PF5 . unambiguously found from the MD simulations that the stability of PF5 strongly depends on the solvent and apart from the polarization effect by the solvent, the steric effect of the solvent on solvation also seems to be at work, which is absent in the SCRF-DFT calculations. The difference in the results between the SCRF-DFT calculations, including the PF5 adduct model, and the MD simulations primarily stems from the insufficient treatment of solvation. In particular, the PF5 adduct model included only one solvent molecule in the SCRFDFT calculations. MD simulations revealed more than one solvent molecule are involved in solvation of PF5 . Remember that the two protocols, SCRF-DFT calculations and MD simulations, are based on different theoretical foundations: the DFT calculations on quantum mechanics and the MD simulations on classical mechanics, i.e., the Newtonian mechanics, using a semiempirically derived force field for energy evaluation. The higher level of accuracy for the former over the latter in general is undisputable in principle. Yet it is the way solvent molecules are treated that sharply differentiated the results between the two protocols. The MD simulations predicted the strong solvent dependence of PF5 stability, as experimentally suggested, while the SCRF calculations basically do not distinguish its stability over a wide variety of solvents. It is shown in this study that the explicit solute-solvent interactions described by the classical MD simulations still seem to provide more accurate results, compared to otherwise more accurate DFT calculations using an implicit solvent model. The difference also demonstrates that long-range interactions are important in solvation of species involved in the LiPF6 decomposition. It is of interest to compare the two protocols in the energetics. For example, the energy for the formation for the PF5 -EC adduct in Eq. 2 estimated from the force field calculations was ⫺4.02 kcal mol⫺1, very close to ⫺3.90 kcal mol⫺1 obtained from the gas-phase B3LYP/631G共d,p兲 calculations, for example. Hence, it seems that the solutesolvent interactions are described by the MD simulations reasonably well. Table VIII. The averaged distance between the Li and the P atom in LiPF6 . r LiP , a Å r LiP , b Å DEC DMC GBL EC 2.71 2.47 2.78 2.47 2.93 2.52 3.14 2.53 a Figure 7. Snapshots for the PF5 solvation in various solvents. Only the first solvation shell is shown, taken from dynamics trajectory. The average distance between Li and the P atom of PF6 obtained from MD simulations. b The optimized distance between PF6 obtained from SCRT-HF calculations 共see Table II兲. Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). A1634 Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 Figure 8. HOMO/LUMO energy level diagram for PF5 , BF3 , and various solvents, obtained from HF/6-31G共d兲 calculations. The entropic effect in the decomposition is not included in the previous discussion based on the MD simulations. In the previous section, the SCRF-DFT calculations showed the entropic contribution to the decomposition thermodynamics to be almost identical for all the solvents, and thus it made little difference among the solutes in the entropic effect on the relative stability in the different solvents. For example, the entropic change of LiPF6 in going from EC to DEC is comparable to that of PF5 in the same solvent change. This entropic contribution arises primarily from the change in the molecular vibrations and translations of the solute in different solvents. Other entropic contribution stems from the solvent molecular rearrangement upon solvation of the solute, which is not accounted for in the SCRF calculations. This contribution should be roughly cancelled out in the discussion of the relative stability of the solutes in structurally similar solvents such as DMC and DEC or EC and GBL. As to the solvation entropy difference between unlike solvents such as EC and DEC, it is not expected to overturn the overall outcome of the calculations. Considering the similarity in molecular shape, the solvation of PF5 by EC, both roughly sphere, may be entropically more favorable than the PF5 solvation by DEC, a linear solvent. In this study, LiF was treated as a solute, though it precipitates as it is produced. Still, the difference in the results from the two protocols is caused largely by the increased stability of PF5 in solvent by MD simulations which treat solute-solvent interactions more explicitly. Thus, a treatment of LiF as a solute in this study made little difference on the overall conclusion. Reactivity of PF5.—For a thorough investigation of the reactivity of PF5 with solvents, the reaction pathways involving PF5 and solvents need to be clearly defined. Though some of the products such as POF3 and PO2 F⫺ 2 have been found by our experiments, the reaction mechanism is yet to be clearly understood. Thus, here we examine the reactivity of PF5 in terms of the highest occupied and lowest unoccupied molecular orbital 共HOMO-LUMO兲 interactions between PF5 and the solvents. Figure 8 illustrates a diagram for the HOMO and LUMO energy levels for PF5 and the solvents used, including those for CH3 F and BF3 for comparison. BF3 is a product of the decomposition of another well-known Li salt, LiBF4 . The strong acidity of PF5 as a Lewis acid is well demonstrated by the very low LUMO energy level, 0.15 eV. The HOMO orbitals of the solvents reside mainly on their carbonyl oxygens. According to the molecular orbital theory, the smaller the energy gap between orbital levels belonging to different molecules, the stronger the interaction between the orbitals becomes and thus, the higher the reactivity between the molecules in Figure 9. The hydrolysis of LiPF6 and Et3 MeNPF6 in water and in PC with 0.5 wt % of water monitored by ion chromatography. The vertical axis is the decomposition percentage based on F atoms, while the horizontal line is time in days. general. It follows that with such a low LUMO energy level PF5 easily attacks the Lewis base’s carbonyl oxygen for electrophilic reaction. As a comparison, a weaker Lewis basis CH3 F has a lower HOMO energy than the carbonate solvents, and thus is less likely for the reaction with PF5 . In fact, PF5 has been identified in a CH3 F solution by NMR spectroscopy.26 The LUMO energy level for BF3 is comparable to those for the carbonate solvents and thus, the reactions between BF3 and the solvents are less likely, which is supported by experimental observations, a reason for the stability of LiBF4 in carbonate solvents. Hydrolysis of LiPF6.—Previously, we have investigated the hydrolysis of LiPF6 by ion chromatography.8 Figure 9 shows the progression of hydrolysis over time for LiPF6 and Et3 MeNPF6 in water and also in PC including 0.5 wt % water, monitored by ion chromatography. The figure demonstrates an intriguing behavior of LiPF6 being stable in aqueous solutions, while extremely prone to hydrolysis in organic solutions. We have also found that the high ion dissociation by using a quaternary ammonium cation such as Et3 MeN⫹ suppresses the hydrolysis of PF⫺ 6 , because larger cations tend to prevent the formation of the contact ion pairs.27-29 In order to understand the sharp contrast in the behavior of LiPF6 hydrolysis between aqueous and organic solutions, we also performed MD simulations of LiPF6 in a box of 50 water molecules. The simulation protocol was similar to that used for LiPF6 decompositions. Figure 10 illustrates a snapshot from the MD trajectory, depicting the hydration of LiPF6 with a water molecule intersecting between the Li⫹ and the PF⫺ 6 ions. Table IX lists the hydration structure of LiPF6 in water obtained from integrating the pair distribution functions up to 5 Å. The most profound difference from the solvation in organic solvents is the large Li-P distance, a 40% increase from that in EC 共see Table VIII兲, demonstrating a much stronger dissociating power of water for the LiPF6 ion pair. This long distance, thus more space, between the Li and the P atom allows water molecules to intervene the two ions, as shown in Fig. 10. At this loose association, the shortest Li-F distance is 3.75 Å, averaged over 500 ps, far enough for a reaction between the Li and the F atom Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 Figure 10. A snapshot taken from the dynamics trajectory for LiPF6 in water. Only water molecules in the vicinity of LiPF6 are shown. not to occur. In EC the averaged Li-F distance is as short as 1.87 Å. Though water has a smaller dielectric constant than EC, the size of water molecule, smaller than EC, allows easier access to the space between the cation and the anion, promoting the salt dissociation. Conclusion SCRF-DFT calculations, using an implicit solvent model, showed the LiPF6 decomposition to be further promoted in more polar solvent, but predicted a virtually identical stability for the decomposition product PF5 over a wide range of solvents, predicting an endothermic reaction in solution. The overall thermodynamics did not change, even at a higher temperature at which the salt decomposition tends to be promoted. Inclusion of a PF5 -solvent adduct, in order to account for the stability of PF5 by solvent more explicitly, did not significantly alter the overall outcome. MD simulations, treating solvent molecules explicitly, on the contrary, demonstrated not only a significant stabilization of PF5 , but also its considerable stability variation over various solvents, predicting the decomposition exothermic in all the solvents. The sharp contrast of the results between the two calculations using distinctively different solvent models suggests the strong solute-solvent interactions in the salt decomposition. In general, the continuum model fails when the solvation cannot be described by the dielectric constant of the solvent and the size and shape of the solute cavity. Those systems with specific interactions, such as weak bond formations between the solvent and the solute including hydrogen bonds, may fall into this category. Though the continuum model has been successfully applied to a series of systems for efficiently estimating solvation energies in general,30 it has been pointed out that the model may not be suffi- Table IX. The hydration structure around LiPF6 in water.a r LiP , b Å 4.41 a r LiO , c Å r LiF , d Å r PO , e Å r FH , f Å n OLi , g Å n OP , h Å 2.53 3.75 3.17 1.73 8.87 12.63 Averaged over 200 ps. The averaged shortest distance between Li and the P atom of the PF⫺ 6 anion. c The averaged shortest distance between Li and the water oxygen. d The averaged shortest distance between Li and the F atom of the PF⫺ 6 anion. e ⫺ The averaged shortest distance between the P atom of the PF6 anion and the water oxygen. f The averaged shortest distance between the F atom of the PF⫺ 6 anion and the water hydrogen. g The number of water oxygens within 5 Å from Li. h The number of water oxygens within 5 Å from P. b A1635 cient for systems with a polar solute in a polar solvent.31 Yet, for a system with a non-polar solute, PF5 , in organic solvent, it is not clear how strong the specific solvation of PF5 may be and there are few experimental studies reporting detailed interactions between PF5 and the organic solvents. Our modeling study may shed light into the solvation of PF5 . Based on the inspection of the trajectories from MD simulations, along with the results form SCRF-DFT calculations, it is concluded that PF5 is more stabilized in polar and compact solvents such as EC and GBL than in bulky, linear carbonates such as DMC and DEC. Thus, it is expected that the decomposition of LiPF6 is promoted in a solvent which has a large dielectric constant and is small in volumetric size. Yet it should not be forgotten that these polar solvents are also effective in dissociating the salt which prevents the decomposition in the first place. Still, at a relatively high concentration used in LIB electrolytes, ⬃1 M, a large degree of ion pairs is believed to remain intact, even in a polar solvent like EC. Then the stability of PF5 by the solvent becomes an important issue. A dilemma is that a polar solvent not only helps dissociate a LiPF6 ion pair but also stabilizes the decomposition product, PF5 . A guideline for a better electrolyte is a design that promotes an ion pair dissociation while effectively destabilizing PF5 . Two factors which are found to play a role in stabilizing PF5 from this study, the dielectric constant and the volumetric size, may be used to optimize the electrolyte characteristics. Our MD simulations of LiPF6 in water provide a good example as to how these two effects contribute to the decomposition of the salt. The calculations on the LiPF6 hydrolysis support the model that the hydrolysis occurs via a pathway involving PF5 produced by the decomposition of LiPF6 and also demonstrate that the LiPF6 salt is more dissociated in water than in carbonate solvents, which may explain the difference in the hydrolysis behavior between the two solutions. Acknowledgment The authors are grateful to Accelrys’s assistance in providing software tools. We also thank M. Takehara, N. Sato, and Y. Sakata for their experimental work, and N. Mine and T. Aoshima for valuable discussions at Mitsubushi Chemical Corporation. Mitsubishi Chemical Corporation assisted in meeting the publication costs of this article. References 1. D. Salmon, D. Barnette, and D. McDonald, Abstract 33, p. 55, The Electrochemical Society Meeting Abstracts, New Orleans, LA, Oct 10-15, 1993. 2. G. G. Botte, R. E. White, and Z. Zhang, J. Power Sources, 97, 570 共2001兲. 3. S. E. Sloop, J. K. Pugh, S. Wang, J. B. Kerr, and K. Kinoshita, Electrochem. Solid-State Lett., 4, A42 共2001兲. 4. C. L. Campion, B. L. Lucht, R. Boris, J. DiCario, R. Gitzendanner, and K. M. Abraham, Abstract 58, The 224th American Chemical Society Meeting, Boston, MA, Aug 18-22, 2002. 5. C. G. Barlow, Electrochem. Solid-State Lett., 2, 362 共1999兲. 6. U. Heider, R. Oesten, and M. Jungnitz, J. Power Sources, 81, 119 共1999兲. 7. D. Aurbach, J. Power Sources, 89, 206 共2000兲. 8. M. Takehara, M. Ue, N. Sato, and Y. Sakata, in The 41st Battery Symposium in Japan, p. 278, Nagoya, Nov 20-22, 2000. 9. A. F. Janzen, X. Ou, and M. G. Sowa, J. Fluorine Chem., 83, 27 共1997兲. 10. T. Sonoda, H. Kamizori, S. Ikeda, H. Nagashima, K. Momota, T. Hashimoto, A. Kimiura, J. Yamaki, R. Hagiwara, and F. Kita, in The 40th Battery Symposium in Japan, p. 447, Kyoto, Nov 14-16, 1999. 11. K. Tasaki and S. Nakamura, J. Electrochem. Soc., 148, A984 共2001兲. 12. K. Tasaki, J. Electrochem. Soc., 149, A418 共2002兲. 13. J. B. Foresman and Æ. Frisch, in Exploring Chemistry by Electronic Structure Theory, 2nd ed., p. 194, Gaussian, Inc., Pittsburgh, PA 共1999兲. 14. A. D. Becke, J. Chem. Phys., 98, 5648 共1993兲. 15. M. J. Frisch, J. A. Pople, and J. S. Binkley, J. Chem. Phys., 80, 3265 共1984兲. 16. M. W. Wong, M. J. Frisch, and K. B. Wiberg, J. Am. Chem. Soc., 113, 4776 共1991兲. 17. M. J. Frisch, G. W. Trucks, H. B. Schlegel, P. M. W. Gill, B. G. Johnson, M. A. Robb, J. R. Cheeseman, T. Keith, G. A. Peterson, J. A. Montgomery, K. Raghavachari, M. A. Al-Lahoresman, V. G. Zakrzewski, J. V. Ortiz, J. B. Foresman, C. Y. Peng, P. Y. Ayala, W. Cheng, M. W. Wong, J. L. Andres, E. S. Replogle, R. Gomperts, R. L. Martin, D. J. Fox, J. S. Binkley, D. J. Defrees, J. Baker, J. P. Stewart, M. Head-Gordon, C. Gonzalez, and J. A. Pople, GAUSSIAN 98, Rev. A.1, Gaussian, Inc., Pittsburgh, PA 共1998兲. Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract). A1636 Journal of The Electrochemical Society, 150 共12兲 A1628-A1636 共2003兲 18. Æ. Frisch and M. J. Frisch, in Gaussian 98 User’s Reference, 2nd ed., Gaussian, Inc., Pittsburgh, PA 共1999兲. 19. H. Sun, J. Phys. Chem. B, 102, 7338 共1998兲. 20. Cerius2 version 4.5 Modeling Environment, Accelrys, San Diego, CA 共2002兲. 21. L. Onsager, J. Am. Chem. Soc., 58, 1486 共1936兲. 22. C. M. Breneman and K. B. Wiberg, J. Comp. Chem., 11, 361 共1999兲. 23. M. Ue and S. Mori, J. Electrochem. Soc., 142, 2577 共1995兲. 24. M. Ue, Prog. Batteries Battery Mater., 16, 332 共1997兲. 25. J. A. Gibson, D. G. Ibbott, and A. F. Janzen, Can. J. Chem., 51, 3203 共1973兲. A. F. Janzen, Coord. Chem. Rev., 130, 355 共1994兲. M. Ue, J. Electrochem. Soc., 141, 3336 共1994兲. M. Ue, Curr. Top. Electrochem., 7, 49 共2000兲. M. Ue, A. Murakami, and S. Nakamura, J. Electrochem. Soc., 149, A1385 共2002兲. D. M. Dolney, G. D. Hawkins, P. Winget, D. A. Liotard, C. J. Cramer, and D. G. Truhlar, J. Comp. Chem., 21„5…, 340 共2000兲. 31. T. Brinck, A. G. Larsen, K. M. Madsen, and K. Daasbjerg, J. Phys. Chem. B, 104, 9887 共2000兲. 26. 27. 28. 29. 30. Downloaded on 2016-02-20 to IP 130.203.136.75 address. Redistribution subject to ECS terms of use (see ecsdl.org/site/terms_use) unless CC License in place (see abstract).
© Copyright 2025 Paperzz