Huseyin Sener
Sen
* and
Bora
Karasulu
*
Department of Chemistry, University of Warwick, Coventry CV4 7AL, UK. E-mail: huseyin-sener.sen@warwick.ac.uk; bora.karasulu@warwick.ac.uk
First published on 17th May 2025
Despite the wide range of emerging solid electrolytes with promising characteristics, such as high ionic conductivity, the inherent thermodynamic instability against lithium metal remains a significant challenge. We have previously introduced a new family of solid electrolytes based on thiophosphates with inherent stability against the metallic Li anode and high conductivity [C. Szczuka, B. Karasulu, M. F. Groh, F. N. Sayed, T. J. Sherman, J. D. Bocarsly, S. Vema, S. Menkin, S. P. Emge, A. J. Morris and C. P. Grey, J. Am. Chem. Soc., 2022, 144, 16350–16365]. In this study, we employ density functional theory (DFT) together with ab initio molecular dynamics (AIMD) simulations to investigate the diffusion mechanisms and underlying factors contributing to the high ionic conductivity of these novel Li–P–S ternary electrolytes, including Li7PS2, Li5PS, Li8P2S, and Li11P3S. Our findings reveal that these materials exhibit ionic conductivities comparable to the well-known superionic conductor, Li7P3S11, positioning them as promising candidates for solid-state battery applications. Additionally, we show that, unlike Li7P3S11 which forms a solid-electrolyte interphase, the novel Li–P–S ternaries exhibit remarkable stability against Li metal anode due to their unique Li2S-like structural framework. The absence of a solid-electrolyte interphase layer is particularly significant, as it eliminates additional resistance at the electrolyte–anode interface, a common challenge in many solid-state battery systems. Our study not only highlights the suitability of these novel ternaries, particularly Li5PS, as high-performance solid electrolytes but also underscores the importance of structural design in developing next-generation battery materials. The ability of these materials to maintain high ionic conductivity and stability over extended periods makes them ideal candidates for future solid-state lithium batteries, offering a pathway to safer, more efficient, and longer-lasting energy storage solutions.
ASSBs exhibit superiority over conventional LIBs, particularly in high-temperature applications, offering reduced flammability risks and a diminished threat of thermal runaway.16,17 Furthermore, the absence of bulk polarization effects, as Li+-ions exclusively conduct ionic charge, suggests the potential for faster (dis)charging in ASSBs compared to traditional LIBs.18 Despite these advantages, the commercialization of ASSBs faces challenges, primarily the absence of a highly conductive SE that is stable toward lithium metal and cathode materials, while accommodating mechanical stresses induced during battery cycling. Dendrite-free plating of lithium metal also remains an unresolved issue.
The exploration of promising SEs has witnessed a proliferation of interest in diverse materials. One class, represented by cation-substituted Li7La3Zr2O12 (LLZO)-derived compounds,19 offers high ionic conductivity and kinetic stability toward lithium metal but faces challenges related to mechanical stiffness and susceptibility to cracking as well as lithium dendrite formation particularly at high current densities.18 A second promising class involves materials containing sulphide ions, exhibiting extremely high ionic conductivity and stress accommodation.20 In these materials, oxide anions are substituted by larger and more polarizable sulphide anions. This results in reduced Li-ion jump barriers which enhances their conductivity enormously.21 In fact, materials incorporating thiophosphate polyhedra as fundamental units exhibit remarkably high ionic conductivities comparable to those observed in liquid electrolytes. Studying glasses and crystalline phases within the pseudo-binary (Li2S)x–(P2S5)1−x were one of the first attempts to utilize the presence of thiophosphate polyhedra.22,23 Recent developments have highlighted that doped crystalline derivatives such as Li10GeP2S12 (LGPS, σ = 12 mS cm−1),24 Li6−xPS5−xClBrx (σ = 24 mS cm−1),25 and Li9.54Si1.74P1.44S11.7Cl0.3 (σ = 25 mS cm−1)26 are some of the most conductive SEs to date. More recently, a third class, lithium-rich ternary phosphides emerged as potential SE candidates, featuring phosphide anions (P3−) with even greater polarizabilities than sulphide anions. These phases are structured around anionic TtP4 tetrahedra (Tt = Al, Si, Ge, Ga, or Sn). Notably, Li9AlP4 demonstrates conductivities of up to 3 mS cm−1 at room temperature.27–33 Despite their appeal as SEs due to a high number of charge carriers and low density, these materials exhibit sensitivity to oxygen and moisture.
Although the newly introduced SEs offer several advantages, they share a significant drawback: they are inherently thermodynamically unstable when in contact with lithium metal, as demonstrated for metal/metalloid-containing oxides, sulphides, and thiophosphates.34–37 For instance, thiophosphates degrade into a mixture of lithium sulphides and phosphides, driven by the reduction of phosphorus ions from a P5+ oxidation state to P3− in Li3P. The resulting lithium binary compounds (Li2O, Li2S, Li3P, and LiX (X = a halide)) are suboptimal as electrolytes, creating a conductivity bottleneck at the SE//Li metal interface. Additionally, the reduction of other cations in the material may lead to short circuits via the formation of lithium intermetallics. For example, Ge4+ in LGPS38 and phosphidogermanates can form electronically conductive germanides,36,37 posing a risk of further decomposition and short circuits through the creation of mixed ionic-electronic conducting interphases.38
Quantum chemical calculations employing density functional theory (DFT) have been widely used to investigate the structural, electronic, transport, and spectral properties of various Li-ion battery materials, particularly SEs.39–41 Moreover, DFT is often combined with crystal structure prediction approaches, such as Ab Initio Random Structure Search (AIRSS), genetic algorithm (GA), USPEX, and CALYPSO, in the in silico discovery of novel functional materials.42,43 While these tools have been applied to predict chemically-doped thiophosphate electrolytes, such as Li3Y(PS4)2 (ref. 44) and Lix(PS4)yXz (X = Cl, Br, I),45 the pure Li–P–S phase diagram has largely remained unexplored, with the exception of individual studies focusing exclusively on a few Li2S–P2S5 phases.22
Although novel SEs show promise, their instability with lithium metal underscores the need for a comprehensive approach focusing on both structural and thermodynamic properties, as well as synthesis methods to developing highly Li+-ion conductive SEs that are thermodynamically stable against Li metal. In our previous work,46 we successfully explored the Li–P–S ternary phase space, employing a combination of high-throughput crystal structure predictions and solid-state synthesis (via ball milling) to isolate the most promising compositions, particularly those within the Li3P–Li2S tie line. We conducted a systematic characterization of the structural properties and Li-ion mobility of these materials using techniques such as X-ray and neutron diffraction, solid-state nuclear magnetic resonance spectroscopy (relaxometry), and electrochemical impedance spectroscopy. This led to the identification of several Li3P–Li2S metastable solid solutions, where the phases adopted a fluorite (Li2S) structure with phosphorus substituting for sulphur and extra Li+-ions occupying octahedral voids. The combined analysis of experimental data and quantum-chemical calculations46 reveals that four of the new ternary compounds—Li7PS2, Li5PS, Li8P2S, and Li11P3S—not only exhibit high structural stability and ionic conductivity (hull distances of 76.3 meV, 38.5 meV, 12.4 meV, 45.4 meV, respectively) but also possess low activation barriers for Li+-ion transport (193 meV, 190 meV, 232 meV, 250 meV at room temperature, respectively). Moreover, each of these structures are shown to be insulators with a direct band gap of at least 1.4 eV making them perfect candidates as SEs if they can maintain thermodynamic stability at the interfaces with electrodes.46 As shown in Fig. 1, these compounds adopt a Li2S-like orthorhombic (slightly distorted cubic) structural framework, which suggests potential thermodynamic stability in the presence of lithium metal anodes similar to Li2S.47–49
![]() | ||
Fig. 1 Supercells of highly conductive Li3P–Li2S solid solutions studied in this work; (a) Li7P2S, (b) Li5PS, (c) Li8P2S and (d) Li11P3S. |
In this work, we delve into the thermodynamic stability of these four Li–P–S ternaries (see Fig. 1) in contact with lithium metal anodes by means of DFT simulations. The results are compared with established Li–P–S electrolytes (Li7P3S11 and γ-Li3PS4) and end members (Li2S and Li3P) of the tie-line (see Fig. S1† for their unit cells). The article is structured as follows: Section 2 details the methodology and computational approach employed, Section 3 presents the results, beginning with an examination of the bulk structures, followed by an analysis of the surface structures, and concluding with an evaluation of the interfaces and their properties, and finally, Section 4 offers a summary of our findings and outlines the implications for future research.
Our approach represents a significant advancement in the development of SEs by addressing critical challenges in the field, particularly the instability of SE materials against lithium metal anodes. The novel Li–P–S ternaries we study not only demonstrate exceptional stability against lithium anodes—comparable to the highly stable Li2S—but also exhibit high ionic conductivity without the formation of a solid-electrolyte interphase (SEI), even at elevated temperatures. This absence of SEI formation eliminates internal resistance at the interface, maintaining the high conductivity observed in their bulk structures. Furthermore, our investigation into the underlying mechanisms reveals that phosphorus substitution for sulphur atoms stabilizes extra interstitial sites and coordinates additional lithium atoms, creating new conduction channels that further enhance ionic transport. This dual achievement of stability and conductivity presents a pivotal step forward, offering a robust solution to the longstanding issues in SE design and paving the way for more efficient and safer energy storage systems.
In our previous study,46 we presented both computational and experimental data on the ionic conductivities and activation energies of newly synthesised and other commonly studied Li–P–S ternary systems. Our findings revealed that the new ternaries, particularly Li7PS2, Li5PS, Li8P2S, and Li11P3S (see Fig. 1) exhibit exceptionally high conductivities and low activation energies, for which we are yet to clarify the underlying reasons, comparable to those of the superionic conductor Li7P3S11. While traditional experimental techniques offer valuable insights into macroscopic ionic conductivity, they often lack the resolution necessary to elucidate the atomic-scale mechanisms that govern ion movement, particularly in complex solid-state systems. AIMD enables the detailed investigation of microscopic processes underlying ionic diffusivity, offering an atomistic perspective of ion transport that accounts for the intricate interactions within the material's crystal lattice and electronic structure in a fully quantum mechanical framework. These simulations are therefore indispensable for the design and optimisation of new SE materials with enhanced ionic conductivity, directly influencing the development of next-generation energy storage technologies. In this section, we provide an understanding of the fundamental mechanisms behind the superior diffusion properties of the four newly-developed ternaries through AIMD simulations.
The self-diffusivity for particle i, also referred to as the tracer diffusivity (D*), can be computed from the mean square displacement (MSD) of the particle during an AIMD simulation as:
![]() | (1) |
![]() | (2) |
The bulk ionic conductivity of an electrolyte, such as a Li–P–S ternary, can be estimated using the Nernst–Einstein relation,67 given by:
![]() | (3) |
Fig. 2a illustrates the MSD of lithium atoms for bulk-Li–P–S systems at 525 K. Li2S and Li3P are not included in this figure due to their almost negligible diffusivities at 525 K. The data reveals that while the lithium mobility in the superionic conductor Li7P3S11 is higher than in all other systems, the mobilities in Li5PS and Li8P2S are quite comparable. Notably, while Li7P3S11 has the highest tracer diffusivity (D*), it is only about four times that of Li5PS and Li8P2S, indicating that the difference in mobility is moderate. Additionally, all the novel ternary compounds demonstrate significantly higher lithium diffusivity compared to γ-Li3PS4 meanwhile Li5PS and Li8P2S ternaries present higher lithium diffusivities than that of highly conductive β-Li3PS4 (see Table S1† for the list of diffusivities and the calculated error values68,69 for the systems shown in Fig. 2a). These observations underscore the enhanced ionic transport properties of the newly developed ternary systems.
From AIMD trajectories, one can also extract the Li-ion transport pathways to elucidate the underlying factors contributing to the conductivity. In the Li2S structure, lithium atoms occupy the 8c tetrahedral sites, while sulphur atoms reside in the 4a octahedral sites, as illustrated in the upper panel of Fig. 2b. The 4b octahedral sites are energetically unfavourable for both sulphur and lithium atoms, rendering them unstable and vacant. Three primary ion migration paths can be anticipated in Li2S and Li–P–S ternaries: (1) the 8c–8c (T–T) path as shown by red arrows, involving Li hopping between thermodynamically stable tetrahedral sites (i.e., vacancy hopping);40,70 (2) the 8c–4b–8c (T–O–T) path as shown by black arrows, where Li jumps between regular and interstitial Li sites;40,70 and (3) the 4b–4b (O–O) path as shown by blue arrows, which is only possible for Li–P–S ternaries as Li2S lacks interstitials. In the lower panel of Fig. 2b, we depict the lithium ion trajectory during the whole simulation of 100 ps at 525 K with green dots, while the initial positions of sulphur atoms are given as reference. For pristine Li2S, Li conduction does not occur at 525 K. Li mobility is rather limited and confined to the regular Li sites, primarily due to the full occupancy of the 8c sites in the model used. Although Li migration through the 8c–8c (T–T) path appears to be the most probable diffusion mechanism, introducing Li lattice vacancies at the 8c sites is not expected to significantly enhance Li mobility at low temperatures.40,70 Moreover, T–O–T type migration is also absent, consistent with the reported high activation barrier (Ea = 0.39 eV).40 However, at 1075 K, slightly above the previously reported superionic phase transition temperature (T ≈ 900 K),70–72 Li ions begin to diffuse via the T–O–T path, which then becomes the predominant mode of conduction.
Next, we investigate the ionic conduction in the novel Li2S–Li3P solid solutions. These novel ternary compounds exhibit significantly higher ionic conductivities compared to Li2S, primarily due to the incorporation of phosphorus atoms. In these ternary compounds, each phosphorus atom substitutes a sulphur atom, occupying a 4a site. Phosphorus atoms, with their 3− oxidation state, exert stronger coulombic interactions with lithium ions compared to sulphur atoms with a 2− oxidation state. This interaction not only stabilises the adjacent 4b sites, enabling the coordination of additional lithium atoms, as depicted in the upper panel of Fig. 2c, but also reduces the energy barrier for T–O migration. Consequently, diffusion channels primarily form around the phosphorus atoms and most lithium diffusion in these ternary compounds occurs in the vicinity of phosphorus atoms due to the increased availability of stable interstitial sites, as demonstrated for bulk-Li7PS2 structure shown in the lower panel of Fig. 2c. Substituting S atoms with P does not appear to alter the 3D Li migration pathways significantly, as both T–T (more visible in other ternaries) and T–O–T paths (but not O–O) are utilised in all ternaries. A clear distinction between the parent Li2S and the Li2S–Li3P solutions is the significantly higher prevalence of T–O–T (8c–4b–8c) conduction channels even at low temperatures, making the T–O–T path the primary mode of conduction. This is likely facilitated by the additional interstitial Li ions. The diffusion process typically initiates via a knock-on mechanism, wherein a lithium ion occupying an interstitial 4b site facilitates the displacement of a lithium ion from a lattice 8c site into a neighbouring, higher-energy vacant 4b interstitial site, constituting the T–O segment of the T–O–T conduction pathway. Subsequently, the resulting vacancy at the 8c site is filled by another lithium ion migrating from a nearby 4b interstitial site, completing the O–T segment of the pathway. Due to this mechanism and the limited number of 4b sites the phosphorus-to-sulphur ratio becomes a critical factor in optimising ionic conduction. Fig. 2d shows the calculated lithium ion conductivity at 525 K as a function of phosphorus content. Li2S and Li3P, representing the extremes, exhibit negligible conductivity. The data indicate that conductivity peaks around a phosphorus content of 50% (Li5PS) where the phosphorus atoms are anticipated to distribute throughout the electrolyte more homogenously. Insufficient phosphorus results in a low number of stabilised 4b sites, reducing the number of available hopping sites for lithium ions. Conversely, an excess of phosphorus leads to an over-coordination of lithium ions, decreasing the number of vacant, stable 4b sites and thus limiting ionic conduction.
Based on these findings, we propose that incorporating phosphorus sites into the Li2S–Li3P solutions enhances Li migration through two mechanisms: (1) by lowering the migration energy barrier between nearby 8c–4b (T–O) sites due to strong Coulomb interactions between Li+–P3− ions, and (2) by coordinating additional Li ions at the higher-energy interstitial octahedral (4b) sites. The former mechanism aligns with previous reports on other superionic conductors.73 Both factors facilitate migration along the 8c–4b–8c and 8c–8c paths, with the former being more prominent, while the 4b–4b path remains less populated.
Due to the fundamentally distinctive structural configurations of the superionic conductors Li7P3S11 and Li3PS4 compared to Li2S and novel ternary compounds, their lithium-ion conduction pathways are not directly comparable. In the case of Li7P3S11, studies have shown that the dynamic flexibility of P2S7 ditetrahedral units facilitates Li+ diffusion by effectively reducing energy barriers and generating an open framework for ion transport. This mechanism enables lithium-ion mobility without relying on well-defined interstitial sites or vacancy-mediated diffusion.64–66,74,75 In contrast, the new Li–P–S solid solutions have a knock-on mechanism, which is facilitated by the additional Li atoms located at the interstitials that are coordinated by the phosphorous atoms.
![]() | (4) |
![]() | (5) |
In our study, we assess the chemical potentials associated with each element (Li, S and P) using the decomposition approach, analysing energies inherent to bulk lithium, lithium sulphide (Li2S), and lithium phosphide (Li3P), or alternatively, employing the starting material approach, analysing lithium sulphide, lithium phosphide, and phosphorus pentasulfide (P2S5), as well as the elemental approach, analysing bulk lithium (bcc-Li), bulk sulphur, and bulk phosphorus (black phosphorus). Each distinct configuration of these subsystems yields disparate chemical potentials, necessitating comparison with experimental data to validate their accuracy. Due to the lack of empirical methods to check the validity of these diverse set of chemical potentials which leads to a substantial ambiguity and the significant disparities observed in the formation energies of the structures studied in this work when employing chemical potentials derived from the aforementioned set of subsystems (see ESI, Tables S2 and S3†), we have opted to forgo the incorporation of chemical potential considerations and instead maintain consistent stoichiometry across all scenarios (even if it results in symmetry-breaking in some cases) similar to the approach of Haruyama et al.84
As previously stated, the determination of surface formation energy necessitates the utilization of symmetric surfaces, a requirement that may not be generally achievable across all Miller indices. Indeed, the presence of symmetric surfaces within the systems under investigation is notably scarce. Consequently, we have devised a methodological approach guided by a predefined set of principles governing the generation of slabs, with particular emphasis on prioritizing specific surface characteristics.
The set of principles can be outlined as follows:
I. Ensure the stoichiometric integrity: This principle arises from the inherent non-uniqueness of chemical potentials and the lack of testing mechanisms, where outcomes are intricately tied to the chosen parameters.
II. Where feasible, create symmetric surfaces: Surface symmetry is imperative for the application of eqn (4), facilitating the precise determination of surface formation energies. The slabs featuring symmetric surfaces are denoted by the symbol “s” within Table 1.
γ s (J m−2) | Li | Li3P | γ-Li3PS4 | Li7P3S11 | Li2S | Li7PS2 | Li5PS | Li8P2S | Li11P3S |
---|---|---|---|---|---|---|---|---|---|
(001) | 0.46 s | 0.48 s | 0.45q | 0.25s | 0.88s | 0.65q | 0.65s | 0.65q | 0.71q |
(010) | 0.21s | 0.18s | 0.67q | 0.67s | 0.66q | 0.68q | |||
(100) | 0.51s | 0.34s | 0.09 s | 0.66q | 0.61s | 0.65q | 0.74q | ||
(011) | 0.50s | 0.67s | 0.41q | 0.12s | 0.50s | 0.51q | 0.56s | 0.59q | 0.65q |
(01![]() |
0.40n | 0.17s | 0.50q | 0.56s | 0.62q | 0.64q | |||
(101) | 0.27q | 0.12s | 0.51q | 0.59s | 0.57q | 0.69q | |||
(10![]() |
0.46n | 0.15s | 0.55q | 0.59s | 0.58q | 0.59q | |||
(110) | 0.73s | 0.18 s | 0.20s | 0.53q | 0.62s | 0.60q | 0.66q | ||
(1![]() |
0.18s | 0.20s | 0.55q | 0.62s | 0.57q | 0.65q | |||
(111) | 0.54s | 0.85s | 0.33n | 0.18s | 0.33 s | 0.34 q | 0.43 s | 0.46q | 0.56q |
(11![]() |
0.29n | 0.16s | 0.36q | 0.43s | 0.42 q | 0.54 q | |||
(1![]() |
0.36n | 0.17s | 0.34q | 0.43s | 0.43q | 0.58q | |||
(![]() |
0.50n | 0.13s | 0.37q | 0.43s | 0.44q | 0.56q | |||
(11![]() |
0.22s | ||||||||
(10![]() |
0.21s | ||||||||
(210) | 0.51s | 0.54s | |||||||
(211) | 0.54s | 0.29n |
III. In case symmetric surfaces are unattainable, form quasi symmetric surfaces: We focus on aligning the surface atoms while disregarding those situated beneath. This approach aims to maximize symmetry within the outermost layers (while minimizing the net dipole moment), thereby enabling the utilization of eqn (4) under the assumption of comparable formation energies between the two surfaces. The slabs featuring quasi-symmetric surfaces are denoted by the symbol “q” within Table 1.
IV. Maintain the integrity of the S–P bonds: While not a strict rule, preserving the integrity of the S–P bonds in PS4 tetrahedra or P2S7 (if any) during surface formation prevents substantial energy penalties.
V. In cases where neither symmetric nor quasi-symmetric surfaces can be achieved, the surfaces are constructed while adhering solely to stoichiometric considerations. For the structures in this category, the results will only be important if the calculated formation energy for the fully non-symmetric slab is found to be the lowest among all Miller indices indicating the need for subsequent investigation into the corresponding structure. Fully non-symmetric slabs are denoted as “n” within Table 1.
We systematically studied the free surface formation for bcc-Li, Li3P, γ-Li3PS4, Li7P3S11, Li2S and the new ternaries (Li7PS2, Li5PS, Li8P2S, Li11P3S) for all Miller indices up to 1 (i.e., −1, 0, and 1). For some cases, we further analysed specific Miller indices to compare with the literature. The results are presented in Table 1 where we show the lowest formation energies in bold font and the corresponding surfaces are depicted in Fig. S2.† (001) surface in bcc-Li (symmetrically equivalent to (100) surface) and (001) surface in Li3P present the lowest formation energies of 0.46 J m−2 and 0.48 J m−2, respectively, which are in perfect agreement with previous studies.85–87 Our results indicate that for γ-Li3PS4, (110) surface has the lowest surface energy, at 0.18 J m−2. The surfaces of γ-Li3PS4 have not been extensively explored in the literature. Lepley et al.88 reported a surface formation energy of 0.32 J m−2 for the (010) surface though our calculations yield a lower formation energy of 0.21 J m−2 for the same Miller index. Discrepancies in energy assessments may stem from the myriad of conceivable surfaces for Li3PS4, rendering comprehensive analysis impractical. For instance, the (010) surface presents at least three distinct cuts, the fewest among all Miller indices, further compounded by the lack of symmetry in many slabs, introducing additional ambiguity to model selection and ensuing energy calculations. For Li7P3S11, (100) surface yields the lowest surface formation energy of 0.09 J m−2, consistent with prior literature.89 Due to the cubic symmetry of Li2S, we focus on the formation of surfaces corresponding to three Miller indices: (100), (110), and (111). Although the cubic symmetry is disrupted in the new ternary compounds, their structural framework remains analogous to Li2S. Consequently, their surfaces can be categorized into three distinct families of Miller indices (100), (110), and (111), as illustrated in Fig. S3.† These families can easily be identified by their structural motifs when viewed from above the surface. For example, in (100) family, the anions form a square shape (see the black dashed lines in Fig. S3† top view), whereas, in (110) and (111) families, a parallelogram and a triangular shape emerges, respectively. Specifically, the (100) family includes (010) and (001) surfaces; the (110) family includes (10), (101), (10
), (011), and (01
) surfaces; and the (111) family includes (11
), (1
1) and (
11) surfaces. Although the surfaces for different Miller indices within each family are not equivalent, they still exhibit similar surface formation energies (Table 1). The (111) surface of Li2S demonstrates the lowest formation energy at 0.33 J m−2, which aligns well with values reported in the literature.78,85,87 The red circles and lines on the far left of Fig. S3† represent the Li atoms and Li–S bonds lost during the cleavage to create the surface. Notably, for the (111) surface, sulphur atoms lose only a single Li bond, whereas for the (100) and (110) surfaces, two Li bonds are broken. We note that the breaking of fewer bonds during surface formation leads to a smaller increase in total energy, which in turn results in lower formation energy. Given the similar framework to Li2S, the new ternary compounds also exhibit the lowest surface formation energies for (111) type surfaces. Specifically, for Li7PS2 and Li5PS, (111) surfaces have the lowest formation energies of 0.34 J m−2 and 0.43 J m−2, respectively while, the lowest formation energy for Li8P2S and Li11P3S is observed for (11
) surfaces, with values of 0.42 J m−2 and 0.54 J m−2, respectively. Notably, the surface formation energy increases with the phosphorus content. In these systems, each phosphorus ion can coordinate additional lithium ions. Consequently, when the bulk structure is cleaved to form a surface, the phosphorus ions at the surface lose more lithium coordination than sulphur ions. This implies that with higher phosphorus content, more Li–P bonds are broken upon surface formation, thereby increasing the surface formation energy.
The Wulff shape, is a geometric method utilized to depict the equilibrium morphology of a crystal.90,91 In Fig. 3, we present the Wulff shapes derived from the surface formation energies listed in Table 1. The shape for bcc-Li aligns perfectly with the literature,86 showing (001) and (110) surfaces contributing 39% and 34% of the total surface area, respectively. Contributions from (111), (210), and (211) are approximately 9% each. The shape for Li3P shows some differences from what Canepa et al.85 obtained as they also investigate Miller indices higher than 1. The Wulff shape for Li3P is composed of (100) and (001) type surfaces, with 65% and 35% surface areas, respectively. (110) and (101) surfaces dominate the Wulff shape of γ-Li3PS4 by contributing 58% and 33% to the surface area, respectively. For Li7P3S11, the surface area contributions are distributed among many surfaces, with (100) surface contributing 31%, (011) surface 19%, (101) surface 16%, and (11) surface 13%. For Li2S, the Wulff shape exclusively contains the (111) surface. In new ternaries, similar to Li2S, the Wulff shape is dominated by the (111) family surfaces. However, as the phosphorus content increases, other surface families begin to appear, resulting in a less anisotropic shape. This is attributed to the rapid increase in surface formation energy for the (111) family with increasing phosphorus content. For Li11P3S, the Wulff shape comprises (111), (100), and (110) surfaces, contributing 78%, 14%, and 8% to the total area, respectively.
![]() | (6) |
Due to the imposed boundary conditions in the simulation system, a lattice strain is artificially introduced while aligning the lattices of materials a and b. However, in real materials, such boundary conditions are absent, potentially resulting in negligible strain. Consequently, the calculated interface formation energy is usually overestimated, encapsulating an energy component attributed to in-plane strain. The actual interface energy (γabact) therefore exists within upper and lower bounds represented as γabmin < γabact < γabmax, where γabmax = and γabmin denotes the minimal interface energy devoid of strain energy (Estr), calculated as;
γabmin = γab − Estr | (7) |
The strain energy can be determined through various methods47,48,92,93 but we introduce only two of them here. The first method involves subtracting the energy of relaxed slabs from that of the strained slabs, which we refer to as “Comparison” method. Adopted from Ferrari et al.,47 we allow interface structure to relax to its minimum energy configuration, inducing strain in both slabs. Thus, the strain energy for each slab can be calculated as follows;
![]() | (8) |
γab = γablim + nbc | (9) |
γab = γablim + naca + nbcb | (10) |
A significant limitation of the extrapolation method is the rapid escalation in the number of atoms as n is increased for the Li–P–S systems. For the majority of the studied interfaces, we employ the minimal n values for Li–P–S slabs, which are already considerably thick. Consequently, this renders the extrapolation approach impractical for our purposes. Nevertheless, we conducted a series of tests using both approaches to evaluate the differential impacts between the comparison and the extrapolation (see Fig. S4b†) methods for two exemplary systems: Li2S(111)//Li(100) and Li5PS(111)//Li(100). The results are summarized in Table 2. The strain values obtained from both methodologies are of the same order, and the effect of the interface (γint) is less than 10%. Hence, we applied the comparison method to the remainder of the interfaces studied.
(J m−2) | Li2S(111)//Li(100) | Li5PS(111)//Li(100) | ||
---|---|---|---|---|
Comparison | Extrapolation | Comparison | Extrapolation | |
E Li str | 2.9 × 10−2 | 3.2 × 10−2 | 6.1 × 10−2 | 5.6 × 10−2 |
E electrolyte str | 5.4 × 10−2 | 4.6 × 10−2 | 8.8 × 10−2 | 8.0 × 10−2 |
E str | 8.3 × 10−2 | 7.9 × 10−2 | 0.15 | 0.14 |
γ min | 0.31 | — | 0.14 | — |
γ lim | — | 0.32 | — | 0.15 |
γ lnt | 0.01 (1.5%) | 0.01 (7.5%) | ||
c Li | — | 1.3 × 10−4 | — | 3.8 × 10−4 |
c electrolyte | — | 6.8 × 10−4 | — | 1.7 × 10−3 |
The work of adhesion (Wadh) represents the energy required to separate two adherent surfaces to an infinite distance and quantifies the mechanical stability of an interface. In practical terms, Wadh is defined as the energy necessary to separate the two slabs and can be formulated as follows,
![]() | (11) |
We define it as the minimum work of adhesion since it inherently contains the strain energy which lowers the value. To get the maximum work of adhesion, the strain energy should be added to this value (Wadhmax = Wadhmin + Estr). Similar to the interface formation energy, the actual work of adhesion lies between Wadhmin and Wadhmax. Typically, low (positive value) interfacial energy and high (positive value) work of adhesion are indicative of strong interfacial stability.
The calculated values for the interface formation energy, work of adhesion and strain energy are summarized in Table 3. The calculated maximum and minimum formation energies for Li3P(001)//Li(100) and Li2S(111)//Li(100) interfaces are γLi3P//Limax = 0.37 J m−2, γLi3P//Limin = 0.35 J m−2, γLi2S//Limax = 0.40 J m−2 and γLi2S//Limin = 0.31 J m−2, respectively, which are in agreement with the literature.85 The interface formation energies for new ternaries are even lower than Li2S and Li3P, indicating higher stability. Interface formation energies are typically positive because the formation of interfaces involves creating surfaces from the bulk forms of two systems, which are inherently lower in energy. However, the interfaces of Li3PS4 and Li7P3S11 with Li metal exhibit negative interface formation energies. This anomaly indicates the presence of chemical activities at the interface that lowers the total energy of the interface system. We present the Li5PS//Li and Li7P3S11//Li interfaces as examples of inactive and active interfaces, respectively, in Fig. 4 (See Fig. S5† for all other structures). The dashed lines indicate the position of the interface before relaxation. Notably, the PS4 tetrahedra near the interface breaks up and sulphur atoms diffuse across the Li7P3S11//Li interface to coordinate with lithium atoms and form Li2S-like structures as indicated by the oval (Fig. 4b, bottom panel), whereas no diffusion of sulphur or phosphorus atoms is observed across the Li5PS//Li interface. The S–P bond breaking and sulphur atom diffusion results in the undercoordinated phosphorus atoms bond with lithium atoms in the vicinity. These changes in the bond formation from S–P to Li–P and Li–S significantly reduce the total energy of the system, leading to negative formation energy. Other works that study the apparent reactivity of Li3PS4//Li and Li7P3S11//Li interfaces agree on the formation of a similar Li2S-like buffer layer.47,48,88,94,95
Interface | γ ab max | γ ab min | W adh min | W adh max | E str |
---|---|---|---|---|---|
Li3P(001)//Li(100) | 0.37 | 0.35 | 0.56 | 0.58 | 2.0 × 10−2 |
γ-Li3PS4(110)//Li(100) | −0.90 | −1.11 | 1.41 | 1.63 | 0.22 |
Li7P3S11(100)//Li(100) | −1.87 | −2.04 | 2.31 | 2.48 | 0.17 |
Li2S(111)//Li(100) | 0.40 | 0.31 | 0.42 | 0.50 | 8.3 × 10−2 |
Li7PS2(111)//Li(100) | 0.27 | 0.25 | 0.50 | 0.52 | 1.9 × 10−2 |
Li5PS(111)//Li(100) | 0.28 | 0.14 | 0.49 | 0.64 | 0.15 |
Li8P2S(111)//Li(100) | 0.25 | 0.22 | 0.56 | 0.60 | 3.2 × 10−2 |
Li11P3S(111)//Li(100) | 0.28 | 0.15 | 0.58 | 0.74 | 0.13 |
The calculated maximum and minimum work of adhesion for Li3P(001)//Li(100) and Li2S(111)//Li(100) interfaces are WLi3P//Limax = 0.58 J m−2, WLi3P//Limin = 0.56 J m−2, WLi2S//Limax = 0.50 J m−2 and WLi2S//Limin = 0.42 J m−2, respectively, which are consistent with the values reported in the literature.85,87 The work of adhesion for the new ternary compounds with lithium is higher than that for Li2S and Li3P, indicating greater stability. Furthermore, the phosphorus content has an increasing effect on the work of adhesion, similar to the trend observed for surface formation energy. Having more phosphorus atoms, results in the formation of more bonds at the interface, making it more difficult to separate the surfaces. In the cases of Li3PS4//Li and Li7P3S11//Li interfaces, the chemical reactions at the interface lead to a significant reduction in energy, yielding much higher work of adhesion compared to the other interfaces.
The positional histograms (perpendicular to the interface – along the c-axis) for phosphorus and sulphur atoms, presented in Fig. 5c and d, elucidate the instability of the Li7P3S11//Li interfaces, while simultaneously demonstrating the structural integrity of the Li2S framework. In this stable framework, the S atoms wobble around their initial positions, retaining their original peak position, shape, and height in the histogram. In contrast, the P and S atoms in the highly-conductive Li7P3S11 electrolyte become widely dispersed.
A similar conclusion can be drawn from the pair distribution functions (PDFs) for various atom pairs in these systems, as shown in Fig. 5e and f. In Li2S//Li system, the separation distances between Li–S and S–S pairs remain relatively unchanged over time. However, in Li7P3S11//Li system, the initial peak diminishes due to the dissolution of PS4 tetrahedra at the interface, which decreases the number of P–S bonds.
Fig. 6a–c illustrate the distribution of atomic coordination numbers for various atoms at different times, using a coordination radius of 3 Å. While there is no significant change in Li2S//Li system, in Li7P3S11//Li system, the coordination number with S atoms remains relatively stable, whereas it increases for P atoms. This is due to the aforementioned dissolution of PS4 units to form new Li–P bonds which increases the coordination number of P atoms from around 7 to 9.
To gain a deeper understanding of the microstructural changes, we further analysed the Li7P3S11//Li system. Given the interface instability, already at t = 20 ps, the interface region could be divided into two subregions, as shown in Fig. 6d: the decomposing region, also referred to as the SEI layer, and the crystalline region, which resembles the Li7P3S11 crystal structure. Detailed examination of randomly-selected S and P atoms from these regions reveals significant differences in their coordination. Notably, P atoms, which are initially surrounded by S atoms in the crystalline region, become solely coordinated by Li atoms in the decomposing region. The coordination numbers of Li for P and S atoms in the decomposing region correspond well with those in the bulk Li2S and bulk Li3P. Therefore, a Li2S-like layer where P atoms occasionally substitute S atoms forms, as suggested by other studies.47,48,88,94,95 With a much longer simulation on a much larger system, it might even be possible to observe the formation of separate Li2S and Li3P regions. This is, however, limited by the high cost of AIMD simulations, calling for other approaches, like machine-learned interatomic potentials. The decomposition reaction for the γ-Li3PS4 and Li7P3S11 molecules are therefore as follows:
Li3PS4 + 8Li → 4Li2S + Li3P + 11.4 eV |
Li7P3S11 + 24Li → 11Li2S + 3Li3P + 35 eV |
By analysing the microstructure at each time step, we derived an estimation of the SEI layer thickness as a function of time, as depicted in Fig. 6e. Our model assumes that phosphorus and sulphur atoms coordinated solely with lithium atoms form Li3P- and Li2S-like structures, respectively. The SEI layer thickness was then calculated based on the relaxed bulk volumes of Li3P and Li2S as follows:
![]() | (12) |
Fig. 7a illustrates the structural evolution of the Li5PS//Li interface at 525 K. The dashed black lines represent the initial interface positions at t = 0. Throughout the simulation, even at an elevated temperature of 1075 K, no migration of phosphorus or sulphur atoms is observed, indicating that the interfaces maintain their original structural integrity over time (unlike the γ-Li3PS4 and Li7P3S11 cases). Further evidence of the high stability is provided by the positional histogram shown in Fig. 7b which reveals that after 100 ps, P and S atoms remain at their initial positions without diffusing into the lithium metal, mirroring the behaviour of Li2S//Li system. A similar conclusion is drawn from the PDFs, as the separation distances of each pair remain almost unchanged, indicating stable atomic configurations over time (Fig. 7c).
To quantitatively assess the interface stability, the MSD of sulphur atoms across different systems is plotted in Fig. 7d. As shown in the previous section, the instability of the γ-Li3PS4//Li and Li7P3S11//Li interfaces results in the diffusion of sulphur and phosphorus atoms into the lithium metal, leading to higher MSD values and tracer diffusion coefficients. In both systems, the sulphur migration tends to decelerate after most of the sulphur atoms are consumed for SEI layer formation in a similar fashion to SEI layer growth shown in Fig. 6e. In contrast, the sulphur atoms in the Li5PS//Li system exhibit mobility and a tracer diffusion coefficient comparable to that of the stable Li2S//Li system indicating high stability.
The coordination numbers for phosphorus and sulphur atoms within the Li5PS//Li system exhibit minimal variation over time (Fig. 7e and f), further supporting the conclusion that the interface remains stable. In Fig. 7g, the microstructural environment surrounding the phosphorus and sulphur atoms in the Li5PS layer is analysed. A closer examination of individual phosphorus and sulphur atoms reveals that they are situated in Li3P-like and Li2S-like environments, respectively. This suggests that, unlike in the γ-Li3PS4 and Li7P3S11 cases (see Fig. 6d), the phosphorus atoms in Li5PS (and other Li2S–Li3P solid-solutions) are already in a 3− oxidation state, which diminishes the driving force for decomposition reactions. This observation is corroborated by the energetics of the corresponding decomposition reaction of Li5PS:
Li5PS → Li2S + Li3P +0.35 eV. |
The energy gain from this reaction is negligible, approximately two orders of magnitude lower than that for Li7P3S11, further underscoring the thermodynamic stability of the Li5PS//Li interface.
Next, we assessed the electrochemical stability of novel ternary electrolytes in contact with Li metal by computing both the decomposition electrochemical stability window (DESW) and the intrinsic electrochemical stability window (IESW), following the methodology outlined by Wagemaker et al.97 For DESW calculations, grand potential phase diagrams were employed to evaluate the thermodynamic stability of the ternary compounds against Li metal and to determine their decomposition products as a function of applied voltage. The results, summarized in Table S4,† indicate that the DESW spans from 0 to 0.87 V, suggesting that these ternaries remain thermodynamically stable against a Li-metal anode only at low voltages. Although the stability window is relatively narrow and the decomposition into Li2S and LiP occurs beyond 0.87 V due to delithiation, these ternaries are still promising candidates for use as protective coatings to prevent SEI formation in highly conductive electrolytes that are otherwise unstable in contact with Li metal.
The IESW, in contrast, considers the indirect decomposition of the ternaries through delithiation. To compute the IESW, we fully delithiated the ternary solid electrolytes and applied the following equation:39,98
![]() | (13) |
We note that high stability in an AIMD simulation for 100 ps does not guarantee the interface stability at a much later time or in real life batteries. However, these findings indicate that the newly discovered ternaries present remarkable improvement over the well-established counterparts. For real life applications, experimental verification is required.
Fig. 8 compares the AIMD lithium ion diffusion trajectories for bulk Li2S, Li7P3S11, and Li5PS systems, as well as their respective interfaces with lithium metal (Li2S//Li, Li7P3S11//Li, and Li5PS//Li). Li2S systems are depicted at an elevated temperature of 1075 K, as they exhibit negligible diffusion at 525 K. In bulk Li2S, although very limited, T–O–T type diffusion pathways are visible at 1075 K. In the Li2S//Li interface system, on the other hand, there is an increase in diffusion, likely due to the possibility of Li atom exchange at the interface. However, the increase in diffusion is predominantly in the direction parallel to the interface, which limits its contribution to overall ionic conduction. In bulk Li7P3S11, well-defined (yet somewhat irregular) conduction pathways are observed in agreement with previous studies,64–66,74,75 indicating a stable structural framework at 525 K. Conversely, in the interface structure, these pathways are disrupted due to rapid interface degradation and SEI layer formation. Notably, the formation of SEI layer impedes the movement of lithium ions which is visible as a grid-like pattern on the left side of the image for Li diffusion in Li7P3S11//Li system. Li5PS exhibits both high structural stability and high ionic conductivity, leading to a well-defined and similar ionic diffusion network in both bulk Li5PS and Li5PS//Li interface systems. The diffusion is three-dimensional and continues seamlessly (percolated) across the interface.
![]() | ||
Fig. 8 Li+-ion trajectory (green) for bulk Li2S, Li7P3S11, and Li5PS systems, and their respective interfaces with lithium metal (Li2S//Li, Li7P3S11//Li, and Li5PS//Li). Li7P3S11 and Li5PS systems are presented at 525 K but Li2S systems are depicted at 1075 K, due to negligible diffusion at 525 K. The positions of the phosphorus and sulphur atoms at t = 0 are shown in the background. Similar Li-ion trajectory plots for other systems can be found in ESI, Fig. S9.† |
Consider a system of N particles, specifically Li atoms, with time-dependent position coordinates ri(t), where i = 1, …, N represents the particle index, and t denotes time. The Van Hove correlation function is defined as the probability of finding a particle at position r at time t, given that a particle was located at the origin at t = 0:
![]() | (14) |
![]() | (15) |
For a given distance r and time t, the self-part Gs(r, t) quantifies the likelihood that a particle has moved from its initial position by a distance r after a time interval t. Conversely, the distinct part Gd(r, t) describes the radial distribution of the remaining N − 1 particles at time t relative to the initial reference particle. Notably, Gd(r, t) reduces to the static PDF when t = 0.
In this study, we calculated the distinct part of the Van Hove correlation function for the charge carrier Li+ ions. If the system possesses well-defined lattice sites for Li atoms and some of these sites remain occupied throughout the AIMD simulation, pronounced peaks emerge in the function. However, if the initial lattice structure is disrupted and the well-defined sites are lost, the reference positions of Li atoms become less meaningful, leading to a broadening and eventual disappearance of the peaks.
Fig. 9 presents the distinct part of the Van Hove correlation function for Li2S, Li7P3S11, and Li5PS systems. For bulk Li2S at 525 K, the lack of lithium diffusion results in a plot that closely resembles the PDF over time, exhibiting the first and second coordination shells at approximately 3 Å and 4 Å, respectively. However, at the elevated temperature of 1075 K, the lithium atoms in Li2S acquire sufficient energy to diffuse, leading to a pronounced peak in Gd(r, t) near r = 0, indicating highly correlated Li+-ion motions and a high probability that a vacated Li site is quickly occupied by another Li atom. A similar behaviour is observed in bulk Li7P3S11, bulk Li5PS, and Li5PS//Li systems at 525 K. In Li7P3S11//Li system, however, the peak near r = 0 is weak and diffuse. This can be attributed to degradation at the interface and the formation of the SEI layer, which disrupts the well-defined initial Li occupation sites. Additionally, the coordination shells in both bulk Li7P3S11 and Li7P3S11//Li systems appear less distinct. In contrast, both bulk Li5PS and Li5PS//Li systems exhibit pronounced peaks near r = 0 and well-defined coordination shells due to high stability and diffusivity as discussed in previous sections.
![]() | ||
Fig. 9 The distinct part of the Van Hove correlation function for Li atoms in (a) bulk Li2S at 525 K, (b) bulk Li2S at 1075 K, (c) bulk Li7P3S11 at 525 K, (d) Li7P3S11//Li at 525 K, (e) Li5PS at 525 K and, (f) Li5PS//Li at 525 K. Similar Van Hove correlation plots for other bulk and interface systems can be found in ESI, Fig. S10.† |
In Fig. 10b, we compare the bidirectional flux of Li+-ions—the total count of Li atoms passing through the zone in either direction—for bulk Li5PS and bulk Li7P3S11, as well as their interfaces with the lithium metal anode at 525 K. For Li5PS, the flux at both centre and interface zones is very similar to that observed at bulk zone, with a constant increase over time. This indicates the absence of significant internal resistance to Li+-ion flux at the interface. In contrast, for Li7P3S11, the flux at the “interface” is lower than that at the “centre”, suggesting the presence of resistance, likely due to the formation of the SEI layer that impedes Li+-ion flux. However, the fluxes at both interface and centre zones are significantly higher than at bulk zone, indicating the presence of a different diffusion mechanism in the Li7P3S11//Li system than its bulk counterpart.
Fig. 10c presents the directional flux of Li+-ions—the total count of Li atoms passing through the zone in one direction—into and out of the electrolytes in Li5PS//Li and Li7P3S11//Li systems at 525 K. Initially, the slope of the Li+-ion flux into the electrolyte is much steeper than the flux out, indicating a rapid influx of Li atoms into the electrolyte in Li7P3S11//Li system. After 40 ps, the slopes of the fluxes in each direction become similar, suggesting a steady state. The initial surge of Li ions into the electrolyte is attributed to the porous nature of the material. In our simulations, in bulk Li7P3S11, ionic conduction occurs via Li ion jumps from one site to another. However, when Li7P3S11 is interfaced with the Li metal, the excess Li atoms rapidly infiltrate the pores of Li7P3S11, not only leading to the formation of an SEI layer but also providing an alternative conduction pathway through vacant interstitial sites, leading to a change in the principle conduction mechanism. In contrast, the structural framework of Li5PS is robust and without such pores, which dictates that the conduction mainly occurs through stable vacant interstitial sites, as discussed in Section 3.1. Consequently, the flux of Li ions in each direction through the interface is balanced and the steady-state is observed throughout the simulation.
Our investigation also extended to the thermodynamic stability of these ternaries in contact with lithium metal anodes, a crucial factor for practical Li-metal battery applications. Compared to Li7P3S11, which rapidly forms an SEI layer due to the reduction of phosphorus from an oxidation state of 5+ to 3−, the novel Li–P–S ternaries demonstrate remarkable stability. The SEI formation rate in Li7P3S11 is a time-dependent process, where the thickness of the interphase layer grows proportionally to the square root of time, progressively increasing the internal resistance and reducing the battery's overall efficiency. The stability of novel ternaries is rooted in their unique Li2S-like structural framework, which inherently positions phosphorus atoms in the 3− oxidation state, thereby eliminating the main (chemical) driving force for SEI formation. The absence of SEI layer is particularly significant, as it means that no additional resistance is introduced at the electrolyte–anode interface, which is a common issue in many solid-state battery systems. These finding not only highlights the suitability of novel ternaries as high-performance SEs but also underscore the importance of structural design in developing next-generation battery materials. By avoiding the issues associated with SEI formation, the novel Li–P–S ternaries can maintain high ionic conductivity and stability over extended periods, making them ideal candidates as electrolytes and coating materials for use in solid-state lithium batteries.
In summary, our study provides a detailed understanding of the physical mechanisms driving the superior performance of these new Li–P–S ternary electrolytes. The insights gained from our work pave the way for the development of advanced solid-state batteries that combine high ionic conductivity with robust chemical and electrochemical stability. These findings suggest that the novel ternaries, particularly Li5PS, could play a crucial role in the future of energy storage technologies, offering a pathway to safer, more efficient, and longer-lasting batteries.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5ta00585j |
This journal is © The Royal Society of Chemistry 2025 |