Time-Temperature-Transformation Diagrams to Navigate the Nucleation and Quenchability of Metastable α-Li₃PS4 Authors: Akira Miura, Woohyeon Baek, Yuta Fujii, Kiyoharu Tadanaga, Rana Hossain, Aichi Yamashita, Yoshikazu Mizuguchi, Chikako Moriyoshi, Shintaro Kobayashi, Shogo Kawaguchi, Ding Jiong, Shigeo Mori, Atsushi Sakuda, Akitoshi Hayashi, Wenhao Sun Published: 2026 ABSTRACT α-Li₃PS₄ is a promising solid-state electrolyte with the highest ionic conductivity among its polymorphs. However, its formation presents a thermodynamic paradox: the α-phase is the equilibrium phase at high temperature and transforms to the stable γ-Li3PS4 polymorph when cooled to room temperature; however, α-Li3PS4 can be synthesized and quenched in a metastable state via rapid heating at relatively low temperatures. The origin of this synthesizability and anomalous stability has remained elusive. Here, we address this paradox by presenting an experimental and computational time-temperature-transformation (TTT) diagram, constructed from a temperature-size phase diagram and experimental high-time-resolution isothermal measurements. Our density functional theory calculations reveal that at the nanoscale, the α-phase is stabilized by its low surface energy, which drastically lowers its nucleation barrier across a wide temperature range. This size-dependent stabilization is directly visualized using in-situ sub-second synchrotron X-ray diffraction and electron microscopy, capturing the rapid nucleation of nano-sized α-Li3PS4 and its subsequent slow transformation. This work presents a generalizable framework that integrates thermodynamic and kinetic factors for understanding nucleation and phase transformation mechanisms, providing a rational strategy for the targeted synthesis of functional metastable materials. Introduction Metastable materials, ranging from traditional steel and glass to advanced battery materials, are crucial for a wide range of functional applications1–3. Numerous synthesis techniques have been developed to obtain metastable materials, including topotactic intercalation1, ball milling2, and rapid temperature control3. Among these, controlling the heating profile remains a key strategy for producing metastable phases with desired properties. For instance, the classic composition-temperature phase diagram of the Fe-C system shows thermodynamically stable phases, while the temperature-time-transformation (TTT) diagram provides kinetic guidance for processing both stable and metastable phases in steel4. Multidimensional phase diagrams, which extend beyond conventional two-dimensional temperature-pressure or temperature-composition diagrams, can be powerful tools for capturing the thermodynamics and processing conditions for producing metastable materials5. By combining first-principles theories and computations, we can achieve a predictive understanding to guide the synthesis and processing pathways of specific metastable phases6,7. For example, DFT-computed size-dependent phase diagrams can explain the nanoscale stability of metastable manganese oxides8,9, or competitive nucleation of Sc-Zn quasicrystals by accounting for time- and temperature-dependent phase selectivity10. Chemical potential diagrams of multicomponent materials have also been proposed to understand the chemical reaction pathway to target materials11,12. The theory of TTT diagrams is well-established for metallurgical and glass systems4,13, and recently TTT diagrams have been expanded to electronic systems14. However, in functional ceramics, such as battery materials, TTT diagrams are typically obtained experimentally15,16, and there is currently not a strong theoretical foundation to predict or interpret TTT diagrams for ceramics. Moreover, experimentally measured TTT diagrams are usually constructed from thermal analysis (e.g., TG-DTA) or ex-situ XRD characterization of samples subjected to different heating protocols, which may not resolve distinct thermal events when multiple signals are convolved during fast nucleation and transformation processes. Furthermore, ex-situ XRD lacks the necessary time resolution to capture the rapid reactions that occur within seconds, making it difficult to understand the early stages of structure selection12. Recent advances in in-situ synchrotron X-ray techniques have enabled data acquisition with high time resolution, offering a powerful perspective for understanding metastable phase formation during reactions. By combining a preheated atmosphere with a rapidly inserted sample holder, along with rapid heating and subsequent isothermal heating experiments, we and others17 have now developed the capability to capture the fast, short timescale formation of metastable phases. These experiments provide essential kinetic insights into their metastable nature and the origin of phase selectivity depending on synthesis protocol17,18. One recent mystery in the temperature-time processing of metastable phases is the quenchability of α-Li3PS4. Lithium thiophosphate (Li3PS4) is a well-established solid sulfide electrolyte for all solid-state batteries and has four polymorphs: amorphous, low-temperature γ phase, medium-temperature β phase, and high-temperature α phase (crystal structures in Fig. S1)19,20. The α phase is predicted to have the highest Li-ionic conductivity for solid-state electrolytes, but is only thermodynamically stable at temperatures above 480 °C, and typically converts to the less ionically-conductive γ phase (σγ = 10⁻⁷ S cm⁻¹ at room temperature) upon cooling19. Recently, it was observed that the α phase with the high conductivity (σα > 10⁻3 S cm⁻¹ at room temperature) can be synthesized by rapid thermal annealing (400 °C/min) of the amorphous glass precursor at 243–374 °C followed by quenching3; whereas a slow heating produces the β phase (σβ = 2.0 × 10–4 S cm⁻¹ at room temperature)3. Why can the high-temperature polymorph α-Li3PS4 form at low temperatures, and more importantly, why can it be retained when quenched quickly? If we could build a thermodynamic and kinetic understanding of this process, it could broadly enable the construction of TTT diagrams to synthesize and retain high-performance metastable materials. In this study, we conduct a comprehensive theoretical and experimental investigation of the stability and transformation kinetics in the Li₃PS₄ system. Ab initio thermodynamics shows that α-Li3PS4 is bulk metastable, but is stabilized at the nanoscale over a broad temperature range by its relatively low surface energy. Next, using in-situ synchrotron XRD with millisecond time resolution, we find that this low surface energy drives rapid nucleation of α-Li3PS4 even at low temperatures where it is a non-equilibrium phase. Moreover, inhibited crystal growth results in nanocrystalline α-Li3PS4, which enables the quenchability and persistence of α-Li3PS4 as a metastable phase, in contrast to bulk α-Li3PS4 which transforms to γ-Li3PS4 over time. Our combined approach provides a holistic view of the thermodynamic and kinetic factors governing phase selection in metastable functional ceramics. Results First, we performed ab initio thermodynamic calculations to map the phase stability of Li3PS4 polymorphs as a function of both temperature and particle size (Fig. 1). For all polymorphs, we computed bulk energies, as well as their surface energies (Fig. S3) using the surface slab generation scheme21; using density functional theory (DFT) calculations with the r2SCAN-rVV10 meta-GGA functionals22,23. From the equilibrium Wulff shapes, we calculated the morphology-averaged surface energies of the crystalline polymorphs, shown in Fig. 1a (Methods)24,25. The morphology-averaged surface energy is lowest for α-Li3PS4 (0.43 J/m2), intermediate for the β phase (0.50 J/m2), and highest for the γ phase (0.58 J/m2). This ordering is the inverse of the calculated bulk formation free energy at ambient temperatures (Ebulk,γ < Ebulk,β < Ebulk,α), which can lead to nanoscale crossovers in phase stability26,27. In other words, the bulk metastable α-phase can be nano-stabilized at high surface area to volume ratios (1/R), where R is the radius of a particle (Table S4).Fig. 1Temperature- and size-dependent phase stability of Li₃PS₄ polymorphs from first-principles calculations.a Calculated Wulff shapes of α-, β-, and γ-Li₃PS₄, with facets colored according to their surface energy. b The three-dimensional and projected two-dimensional phase diagram showing the Gibbs free energy, ΔG(T, R), as a function of temperature (T) and particle radius (R). The colored surfaces represent the most thermodynamically stable phase in each region. c–e Two-dimensional plots of Gibbs free energy versus temperature, showing cross-sections of the 3D diagram for particle radii of c, R = 4 nm, d R = 8.5 nm, and e the bulk limit (R = ∞). All energies are referenced to bulk γ-Li3PS4 (R = ∞) at each temperature. Markers in c–e indicate the equilibrium phase transition temperatures where the free energy of phases is the same. Arrows indicate the driving force to transform from amorphous to nanosized α-Li₃PS₄ at 350 °C (dashed arrow), and to β-Li₃PS₄ at 350 °C and from β-Li₃PS₄ to bulk α-Li₃PS₄ at 547 °C (dotted arrow), respectively (Figs. 2–4). f Calculated nucleation energy barrier at the critical nucleus (ΔGc) of bulk Li₃PS₄ polymorphs from the bulk amorphous matrix at different temperatures. Temperature- and size-dependent phase stability of Li₃PS₄ polymorphs from first-principles calculations. a Calculated Wulff shapes of α-, β-, and γ-Li₃PS₄, with facets colored according to their surface energy. b The three-dimensional and projected two-dimensional phase diagram showing the Gibbs free energy, ΔG(T, R), as a function of temperature (T) and particle radius (R). The colored surfaces represent the most thermodynamically stable phase in each region. c–e Two-dimensional plots of Gibbs free energy versus temperature, showing cross-sections of the 3D diagram for particle radii of c, R = 4 nm, d R = 8.5 nm, and e the bulk limit (R = ∞). All energies are referenced to bulk γ-Li3PS4 (R = ∞) at each temperature. Markers in c–e indicate the equilibrium phase transition temperatures where the free energy of phases is the same. Arrows indicate the driving force to transform from amorphous to nanosized α-Li₃PS₄ at 350 °C (dashed arrow), and to β-Li₃PS₄ at 350 °C and from β-Li₃PS₄ to bulk α-Li₃PS₄ at 547 °C (dotted arrow), respectively (Figs. 2–4). f Calculated nucleation energy barrier at the critical nucleus (ΔGc) of bulk Li₃PS₄ polymorphs from the bulk amorphous matrix at different temperatures. The combination of temperature and size variations is captured in the free energy surfaces presented in Fig. 1b. The vibrational and configurational entropy contributions of the α, β and γ polymorphs were calculated using quasi-harmonic vibrational phonons28–31 and Wang-Landau sampling of the configurational density of states for Li partial occupation sites32 (Methods, Supplementary Information, and Fig. S4). For the bulk polymorph ordering (R = ∞), the calculated phase transition temperatures are 296 °C for γ → β; and 385 °C for β → α (Fig. 1e), which are in good agreement with experimental measurements (250–300 °C and 450–500 °C, respectively)3,19,20. This behavior changes dramatically for nanoparticles, as shown in the three-dimensional phase diagram (Fig. 1b) and its two-dimensional cross-sections (Fig. 1c, d, e). Here, the nanoscale free energy of crystalline phases are calculated from size-dependent energy relations G(T, R) = H – TS + γη/ρ(1/R), where γ is the surface energy, ρ is the atomic density and η is the shape factor (Supplementary Equation 2)10. At a particle radius of 8.5 nm, the γ → β and β → α transition temperatures are suppressed to 127 °C and 180 °C, respectively (Fig. 1d). The crystalline polymorphs nucleate from an amorphous matrix, and so we also need the free energy of the amorphous phase. Unfortunately, the enthalpy and entropy of amorphous phases are difficult to obtain from DFT, as phonons are difficult to converge for the structurally disordered amorphous phase. Here we introduce a scheme that references the free energy of the amorphous phase from experimental differential scanning calorimetry (DSC) data of the amorphous → β transition at 230 °C3. The DSC-measured enthalpy evolution is ΔHamorphous→β = -4.746 kJ/mol (Methods and Supplementary Equation 6), and the ΔS between β → amorphous can be determined by integrating the measured heat capacity for each phase in the temperature window before the phase transition (Supplementary Equation 7). The calculated free energy of the bulk amorphous phase, derived from experimental data, is then included into the phase diagram (Fig. 1b–e). The free energy of the amorphous matrix is fixed to be the bulk free energy (i.e., we do not evaluate its size-dependent free energy) because the amorphous phase is the precursor phase and is only in a bulk state in this scenario. At the nanoscale, the bulk amorphous phase crosses over in free energy first to crystalline α-Li₃PS₄ at a radius of 8.5 nm and 380 °C (Fig. 1d) and is metastable with respect to all crystalline phases at bulk sizes (Fig. 1e). Lastly, to address the nucleation kinetics of polymorph selectivity, the critical nucleation barrier, ΔGc, was calculated for each crystalline polymorph from a parent bulk amorphous matrix (Fig. 1f and Fig. S5). The steady-state nucleation rate is obtained from J = A exp(–ΔGc/kBT), where ΔGc is the nucleation barrier (Supplementary Equation 9) and A is the kinetic prefactor to steady-state nucleation that physically corresponds to the diffusivity of precursor monomers in the parent phase. Here, we assume that A is the same for all three polymorphs, and competitive nucleation from the amorphous matrix is evaluated solely based on the ratios of their nucleation barriers. We plot ΔGc/kBT as a function of temperature in Fig. 1f. Because the α-phase is the polymorph with the lowest surface energy, this leads it to have the lowest nucleation barrier at all temperatures. A characteristic ‘nose’ shape appears in this diagram, where α has the smallest barrier at 217 °C, due to a trade-off between increasing ΔGamorphous-solid driving force at low temperatures, but being divided by kBT at low temperatures. Overall, Fig. 1f indicates that at low temperatures, the α-phase can nucleate first and significantly faster than the β- and γ-phases. To experimentally map the crystallization kinetics, we employed high-resolution in- situ synchrotron XRD at the SPring-8 beamline, where we achieved 0.5-sec time resolution. The time-resolved diffraction patterns and the resulting Time-Temperature-Transformation (TTT) diagram reveal temperature-dependent transformation pathways when annealing from the amorphous precursor (Fig. 2). The observed crystallization behavior can be categorized into four distinct regimes. First, at low temperatures around 220 °C (Fig. 2a), the thermodynamically favored β-phase is the dominant product, though its crystal growth at these low temperatures is slow, resulting in low crystallinity. The α-phase also appears as a very minor co-existing phase.Fig. 2Experimental mapping of the time- and temperature-dependent phase evolution of Li₃PS₄ from an amorphous precursor.Time-resolved XRD patterns with the wavelength of 0.354 Å collected during isothermal annealing at a 220 °C, b 340 °C, c 380 °C, d 440 °C, and e 547 °C. The amorphous sample was rapidly heated to the target temperature (before t = 0), and the subsequent phase evolution was monitored over time. Red and blue arrows indicate characteristic peaks for the α- and β-phases, respectively. f, g Time-Temperature-Transformation (TTT) diagrams constructed from the in-situ XRD data. The diagram is divided into two halves: the left side (red colormap) displays the phase fraction of α-Li₃PS₄ (αf), while the right side (blue colormap) displays the phase fraction of β-Li₃PS₄ (βf) as a function of time and temperature. Experimental mapping of the time- and temperature-dependent phase evolution of Li₃PS₄ from an amorphous precursor. Time-resolved XRD patterns with the wavelength of 0.354 Å collected during isothermal annealing at a 220 °C, b 340 °C, c 380 °C, d 440 °C, and e 547 °C. The amorphous sample was rapidly heated to the target temperature (before t = 0), and the subsequent phase evolution was monitored over time. Red and blue arrows indicate characteristic peaks for the α- and β-phases, respectively. f, g Time-Temperature-Transformation (TTT) diagrams constructed from the in-situ XRD data. The diagram is divided into two halves: the left side (red colormap) displays the phase fraction of α-Li₃PS₄ (αf), while the right side (blue colormap) displays the phase fraction of β-Li₃PS₄ (βf) as a function of time and temperature. In the intermediate temperature range of 340 °C to 440 °C (Fig. 2b, d), a classic kinetic-to-thermodynamic transition is observed. The α-phase consistently appears first, confirming that its nucleation is kinetically favored, and persists as a metastable intermediate. With prolonged isothermal holding, the α-phase then transforms into the β-phase, which is more thermodynamically stable for larger particle sizes. Because the rate of particle growth is highly temperature-dependent, the α → β transformation proceeds significantly more rapidly and results in higher final crystallinity at 440 °C than at 340 °C. A peculiar kinetic inhibition is observed at 380 °C (Fig. 2c). While trace amounts of the α-phase appear fleetingly within the first second, crystallization is largely suppressed for over 100 seconds, after which only weak β-phase peaks emerge. Interestingly, at 380°C, the amorphous phase has minimal thermodynamic driving force to transform to nanophase α‑ and β‑phases, as illustrated in Fig. 1d, which supports the phenomenological lack of nucleation and growth at this specific temperature window. Lastly, at high temperatures, such as 547 °C (Fig. 2e), the behavior aligns with the bulk thermodynamic predictions from our calculations. The α-phase forms rapidly and remains the sole, stable crystalline phase throughout the measurement, achieving high crystallinity. This result experimentally confirms the thermodynamic stability of α-Li₃PS₄ at elevated temperatures, as predicted by the phase diagram in Fig. 1. Crystallite size at 100 sec is analyzed from the in-situ XRD diffraction patterns using the Scherrer equation, which revealed that the crystallite size of metastable α-phase is smaller than that of β-phase at intermediate temperatures: at 340 °C, the crystallite size of the metastable α-phase is solved to be ~35 nm, smaller than that of the metastable β-phase at ~52 nm. At 440 °C, both phases exhibited significant growth, reaching approximately 72 nm for the α-phase and 78 nm for the β-phase. The observed trends suggest that the phase transformation from α to β is facilitated by grain growth of the α polymorph at intermediate temperatures, by a mechanism that will be rationalized later. Furthermore, at 547 °C, the size of the α-phase reached 92 nm, indicating further grain growth at elevated temperatures. The effect of heating rate and quenchability of the α-phase was examined by ex-situ SXRD and TEM after naturally cooling from 350 °C (Fig. 3). The thermal treatments for the TEM samples are shown in Fig. 3a, where glassy samples were heated to 350 °C using either a fast or a slow heating profile. The approximate corresponding phase fractions, as inferred from the in-situ XRD measurements in Fig. 2f and 2g, are overlaid onto Fig. 3a. The fast-heated sample ( ~ 10 s) exhibits a mixture of α- and β-phases, demonstrating the concurrent nature of the α-phase (Fig. 3b and Fig. S6). In contrast, the sample subjected to slow heating ( ~ 5 min) shows a doublet peak, which is a clear fingerprint of the well-crystallized ββ-phase, the equilibrium phase at this temperature. This phase formation is consistent with the computed TTT diagram (Fig. 2), where at 350 °C and short holding times, the system is in a regime where the kinetically favored αα-phase has nucleated, but the transformation to the ββ-phase proceeded during the cooling. The formed αα-phase can be partially retained and quenched even after cooling to room temperature. This quenched sample exhibited a total ionic conductivity of 5.8 × 10⁻⁴ S cm⁻¹ (Fig. S8), which reflects the intrinsically high ionic conductivity of the α-phase and also perhaps amorphous and grain boundary diffusivity contributions.Fig. 3Effect of heating rate on Li₃PS₄ polymorphs and microstructure after cooling to room temperature.a Schematic illustration of two thermal profiles applied to the amorphous precursor: a fast heating process with a short hold ( ~ 10 s) and a slow heating process with an extended hold ( ~ 5 min), both to a target temperature of 350 °C and naturally cooled down to room temperature. In the thermal profiles, the dotted and solid lines represent the amorphous and crystalline states of the material, respectively. b Ex-situ XRD patterns (λλ = 0.248 Å) of the final products at room temperature after natural cooling. The fast-heated sample exhibits a mixture of quenchable α-Li₃PS₄ and β-Li₃PS₄ phases. In contrast, the slowly heated sample shows a distinct doublet, a clear signature of the well-crystallized β-Li₃PS₄ phase. c Hollow cone dark-field TEM images, where the bright areas correspond to crystalline regions. The insets show the corresponding selected area electron diffraction (SAED) patterns. Fast heating results in dispersed, yellow-highlighted small nanoparticles ( ~ 15 nm) and larger aggregates above 40 nm. d Dark field TEM image showing that slow heating leads to the formation of larger crystallites of 20–200 nm in size. The inset shows the SAED pattern. Effect of heating rate on Li₃PS₄ polymorphs and microstructure after cooling to room temperature. a Schematic illustration of two thermal profiles applied to the amorphous precursor: a fast heating process with a short hold ( ~ 10 s) and a slow heating process with an extended hold ( ~ 5 min), both to a target temperature of 350 °C and naturally cooled down to room temperature. In the thermal profiles, the dotted and solid lines represent the amorphous and crystalline states of the material, respectively. b Ex-situ XRD patterns (λλ = 0.248 Å) of the final products at room temperature after natural cooling. The fast-heated sample exhibits a mixture of quenchable α-Li₃PS₄ and β-Li₃PS₄ phases. In contrast, the slowly heated sample shows a distinct doublet, a clear signature of the well-crystallized β-Li₃PS₄ phase. c Hollow cone dark-field TEM images, where the bright areas correspond to crystalline regions. The insets show the corresponding selected area electron diffraction (SAED) patterns. Fast heating results in dispersed, yellow-highlighted small nanoparticles ( ~ 15 nm) and larger aggregates above 40 nm. d Dark field TEM image showing that slow heating leads to the formation of larger crystallites of 20–200 nm in size. The inset shows the SAED pattern. Analysis of crystalline size from XRD gives statistically-representative, bulk-averaged information, but to better characterize the local particle size distribution and morphology, we further perform ex-situ transmission electron microscopy (TEM) analysis on naturally cooled samples following the temperature-time treatments (Fig. 3c and d). We observe that the fast-heated sample ( ~ 10 sec) consists of small, dispersed crystalline nanoparticles ( ~ 15 nm) and aggregated ones (40–60 nm), confirming the presence of a nanocrystalline phase. The slower heating profile ( ~ 5 min), however, allowed for significant crystal growth into larger crystallites of 20–200 nm in size (Fig. 3d). These TEM observations show crystallite growth is arrested in the fast-heat-and-quenched process, yielding some particles smaller than 20 nm, whereas the slow-heating sample enabled ripening and grain growth to particles much larger than 20 nm. The culmination of our combined computational and experimental investigation is summarized in the TTT diagram in Fig. 4, which provides a unified picture of the competing thermodynamic and kinetic processes surrounding the long-standing synthesis puzzle of α-Li₃PS₄. Below 220 °C and above 500 °C, the synthesized products are consistent with the thermodynamically stable phases. However, in the intermediate temperatures, the small free energy differences between phases can lead to nucleation kinetics dominating structure selection, resulting in formation of the α polymorph in a temperature window where it is a non-equilibrium phase. Figure 4 captures the key fact that synthesis of Li₃PS₄ is not governed by a simple temperature scale, but rather by a multidimensional landscape where thermodynamics (dependent on temperature and size) and kinetics (dependent on temperature and time) are intimately coupled.Fig. 4A conceptual Time-Temperature-Transformation (TTT) diagram summarizing the synthesis pathways for Li₃PS₄ polymorphs.The colored background represents the thermodynamically stable phase at a given temperature for different particle sizes (time is a proxy for particle size). The white characters and arrows illustrate two thermal pathways: (1) Rapid heating and quenching of the precursor (dashed arrow) leads to the formation of nanocrystalline α-Li₃PS₄, which can be retained at room temperature (“Quenchable α“) due to nanoscale stabilization. (2) Slow isothermal heating (dotted arrow) allows for sufficient time for crystal growth, leading the system to form the more stable β-phase and eventually the high-temperature α-phase (“Unquenchable α“). Upon cooling, this bulk α-phase transforms into the stable γ-phase. This diagram provides a unified framework for understanding why the thermal history is critical for selectively synthesizing the quenchable, metastable α-phase. A conceptual Time-Temperature-Transformation (TTT) diagram summarizing the synthesis pathways for Li₃PS₄ polymorphs. The colored background represents the thermodynamically stable phase at a given temperature for different particle sizes (time is a proxy for particle size). The white characters and arrows illustrate two thermal pathways: (1) Rapid heating and quenching of the precursor (dashed arrow) leads to the formation of nanocrystalline α-Li₃PS₄, which can be retained at room temperature (“Quenchable α“) due to nanoscale stabilization. (2) Slow isothermal heating (dotted arrow) allows for sufficient time for crystal growth, leading the system to form the more stable β-phase and eventually the high-temperature α-phase (“Unquenchable α“). Upon cooling, this bulk α-phase transforms into the stable γ-phase. This diagram provides a unified framework for understanding why the thermal history is critical for selectively synthesizing the quenchable, metastable α-phase. The rapid heating pathway leverages the low nucleation barrier of the α-phase (Fig. 1f) to kinetically select it over other polymorphs at low temperatures. Simultaneously, the short heating protocol prevents significant particle growth, confining the system to the nanoscale regime where the low surface energy of the α-phase (Fig. 1a) minimizes its thermodynamic propensity to transform to the γ-phase, which is the equilibrium phase at room temperature and bulk particle sizes (Fig. 1b). This nanoscale stabilization is the definitive reason why the low-temperature α-phase is quenchable, a conclusion directly supported by our experimental characterization of particle sizes by TEM (Fig. 3). Conversely, the slow heating pathway provides sufficient time for growth of α-Li3PS4 out of the nanoscale regime and into the stability fields of β- and eventually γ-Li₃PS₄ upon cooling, explaining why the bulk, high-temperature α-phase cannot be quenched20. As illustrated in Fig. 1e, these larger particles are in a size regime with much stronger thermodynamic driving force to transition from α- to the β- and γ-phases. The TTT diagram rationalizes that retention of the α‑phase is governed by nanoscale stabilization and fast nucleation, and that quenching from the high‑temperature stability region would be difficult, which indeed has not been experimentally demonstrated yet. The present framework thereby establishes generalizable principles for accessing metastable phases outside of their bulk thermodynamic stability window, as exemplified by α‑Li3PS4. The peculiar kinetic inhibition at 380 °C, where only very small amounts of crystalline phases appear at a short time (Fig. 2c), can also be anticipated from the driving forces and nucleation kinetics. At this temperature, the free energy difference between the parent amorphous phase and nanoscale Li3PS4 polymorphs is nearly zero (ΔGamorphous-solid ~ 0) (Fig. 1d). This effectively leads to a larger nucleation barrier for all phases; for example, the critical nucleation radius of α-Li3PS4 is relatively large at 5.6 nm at 380 °C (Fig. S5c). Regarding the stability of the obtained metastable phases, our computational predictions are in overall agreement with experimental observations, capturing the key trends of phase selection and stabilization across temperature and size scales. However, it should be noted that the calculated critical radius of a phase transformation does not necessarily represent a strict experimental limit beyond which the most stable phase forms spontaneously. When the free energy difference between competing metastable phases is small, the kinetic barrier suppresses rapid transition to the equilibrium phase33,34. For example, recent X-ray total scattering and Reverse Monte Carlo analyses have shown that the evolution toward the γ-phase proceeds gradually via continuous lattice-parameter shifts and progressive orientational ordering of PS₄ tetrahedra35, indicating that the metastable state remains kinetically robust beyond the theoretical size limit. These observations support the view that the metastable α-phase, as well as the β-phase, can be retained at room temperature, as the kinetic barrier effectively prevents spontaneous transformation into the equilibrium γ-phase. Discussion Here we rationalized the nanoscale stabilization of quenchable metastable α-Li₃PS₄ by clarifying the underlying mechanisms that enable its synthesizability. By integrating computational thermodynamics with high time-resolution experimental kinetic measurements, we constructed a TTT diagram that quantitatively captures how rapid thermal processing promotes the remnant metastability of α-Li3PS434,36. First-principles calculations showed that the low surface energy of the α phase thermodynamically stabilizes it at the nanoscale, a state achieved and maintained by rapid heating at relatively low temperatures that kinetically favors its nucleation while suppressing crystal growth. Validated by in-situ synchrotron XRD, this quantitative thermodynamic and kinetic picture provides a unified understanding of its nucleation and transformation pathways. This integrated methodology not only resolves the puzzle of α-Li₃PS₄, but also establishes a versatile platform applicable to other functional battery materials and glass-ceramics, paving the way for targeted synthesis of novel metastable phases. The success of our TTT framework hinged on the synergistic combination of experimental and computational data to deliver essential quantities that are inaccessible from either theory or experiment alone. While the bulk and surface energies of crystalline phases are readily computable, determining the temperature-dependent energy of an amorphous phase from first principles is difficult due to its non-periodic nature. To bridge this gap, here we used experimental differential scanning calorimetry (DSC) data to reference the bulk energy and heat capacity of the amorphous phase against the DFT-computed free energy of a crystalline phase. Similarly, when quantifying the kinetics, DFT-calculated bulk and surface energies can be used to estimate nucleation barriers, but the crystal growth rate is not easily accessible through computation alone. Therefore, integration of theory predictions with experimental high-time resolution measurements, such as diffraction or electron microscopy7,37–39, will be essential for a holistic understanding of these kinetic factors. Methods Density Functional Theory (DFT) calculations were performed for atomistic structure calculations using Quantum Espresso (QE)40,41. The revised regularized strongly constrained and appropriately normed (r2SCAN) meta-Generalized Gradient Approximation (meta-GGA)22 functional with revised Van der Waals 10 (rVV10) nonlocal correlation functional23, and norm-conserving pseudopotentials by Y. Yao and Y. Kanai42 were applied to all DFT calculations of crystalline phases. The energy cutoff of wavefunction is set to 170 Ry and kinetic energy cutoff for charge density is set to 680 Ry from convergence test (Fig. S2). The convergence threshold of the total energy, force, and stress is set to 10-4 eV/atom, 3 × 10-3 eV/Å, and 0.5 kbar for structure relaxations, respectively. The k-points were optimally adapted from Pymatgen43. Surface energies of the crystalline Li3PS4 were calculated from slab structures (Supplementary Equation 1), generated by cleaving bulk phases using the Pymatgen package21. The maximum Miller indices were set to 3, and the minimal slab size and vacuum length normal to the surface were set to 10–20 and 20 Å, respectively. The Wulff shape was also calculated using Pymatgen, and the weighted surface energy and shape factor from calculated surface energies was calculated by the γ-plot construction (Tables S1–S3, Fig. 1a)25. Table S4 summarizes all calculated parameters. The vibrational free energy of three crystalline polymorphs was calculated from quasi-harmonic approximation. The ratio of unit cell three lattice parameters of length (volume) was varied from 0.97 to 1.03 keeping the relaxed fractional atomic coordinates from structure relaxation using DFT. The atomic position displacement per each unit cell with varied volumes was generated with symmetry-adapted displacements for α-Li3PS4 and β-Li3PS4, and with 100 random displacements of γ-Li3PS4 on 2 × 2 × 2 supercell. The total atomic energy of all generated structures was calculated using QE. The calculated energy with respect to the volume was fitted in the Rose-Vinet equation of states44 and then obtained the lowest energy. The displaced structure generation and free energy calculations were performed using Phonopy28,29 and with Alamode for random displacement30,31. The k-mesh parameters for phonon calculations were set to 8-12 times larger than the k-points for DFT calculations. The configurational entropy contribution by Li disorder of α-Li3PS4 and β-Li3PS4 was also considered to the total free energy. We randomly sampled 100 α-Li3PS4 and all 70 possible configurations of β-Li3PS4 by placing the Li to one of partial sites but not overlapping the same configuration of sampled ones, keeping the original stoichiometry of Li3PS4 from the experimentally defined crystal structures and atomic positions with partial occupation sites19 and calculated atomic energies of each configuration using QE (Supplementary Data 1). The configurational entropy was calculated from Boltzmann entropy formula (Supplementary Equations 3 and 4). To define the contribution of the partition function of configurations, the optimization of the energy contribution to the statistical distribution was performed using the Wang-Landau algorithm32. The energy was sampled from Monte Carlo up to 100,000 steps in maximum and repeated the calculation with various temperatures. The free energy of the amorphous phase was calculated from the experimental differential scanning calorimetry (DSC) reference data3. The 3 mg Li3PS4 glass powder was placed in an Al pan under a dry Ar atmosphere and performed DSC measurements, heating to 500 °C and then cooled to 25 °C at a scanning rate of 10 °C/min. The DSC during the heating cycle observed an exothermic peak at 230 °C, showing phase transition from amorphous to β-Li3PS4. The enthalpy difference between amorphous and β-Li3PS4 was calculated by the integration of analyzed thermal peaks from the corrected baseline (Supplementary Equation 6). The entropy of amorphous Li3PS4 was calculated from the integrated heat capacity into the entropy change equation at constant pressure (Supplementary Equations 5 and 7). The total free energy of amorphous Li3PS4 is calculated from 25 to 230 °C before phase transition. Calculated free energy data points of amorphous and the melting temperature of crystalline Li3PS4 at 700 °C were fit to a power function for the entire temperature range (Fig. S4d). All the starting materials and products were handled without exposure to air. The Li2S-P2S5 samples of ~2 g were first prepared by hand milling with a mortar and pestle. Subsequent ball-milling was performed with a zirconia pot (45 ml) and a zirconia ball (6 mm diameter). In-situ synchrotron X-ray diffraction was performed at the BL13XU beamline in the SPring-8 with the approvals of 2023B1669, 2023B1942, and 2025B1753. The samples were sealed in a quartz capillary with a diameter of 0.3 mm. The samples were heated by a nitrogen flow, and diffraction patterns were recorded by LAMBDA 750 K detectors45. The wavelength was 0.354 Å or 0.248 Å. In-situ SXRD was performed in two ways. One was at the constant heating rate of 60 °C/min, and the other was an isothermal experiment. For the isothermal experiment, a nitrogen gas flow for heating the capillary samples was preheated at 220, 237, 260, 300, 340, 360, 367, 380, 440, 500, and 547 °C without the samples. The diffraction measurements were collected every 0.5 second, and the samples were inserted in a preheated nitrogen flow. The samples were kept during the isothermal measurements. We defined the time of 0.5 seconds when the diffraction data was detected. We measured the change in lattice parameters of silver particles in the vacuumed capillary in the same setting of isothermal reactions. This measurement showed that the lattice parameters are almost constant after 2.0 seconds. Thus, this measurement is the rapid heating within 2.0 seconds and subsequent isothermal heating after 2.0 seconds (Fig. S7). Rietveld analysis was performed to quantify the phase fraction using TOPAS software46. We normalized the mass fraction to 100% using the maximum sum of the alpha and beta phase scale factors in each isothermal measurement, and constructed a time-temperature-transformation diagram. The XRD data shown in Fig. 3 were collected after the samples were naturally cooled to room temperature. Guided by the TTT diagram, glass samples were crystallized in a short time. A few tens of milligrams of glass sample produced by ball milling were put on the hot plate and pressed with a plastering spatula for 10 seconds and collected. The product was characterized using SXRD and laboratory XRD with CuKa radiation by Miniflex600 (Rigaku). Impedance spectra were measured by SP-200 (BioLogic) to estimate its ion conductivity (Fig. S8). In-situ TEM observation was conducted by JEM-2100Plus (JEOL Co. Ltd) with an accelerated voltage of 200 kV. In this in-situ experiment, a 4-electrode Vacuum Transfer Holder (Mel Build Co. Ltd.) was used to prevent air exposure during sample transfer from the glovebox to the TEM. In addition, a high-performance heating micro-electromechanical systems (MEMS) chip was used to control the rate of temperature change. Supplementary information Supplementary Information Transparent Peer Review file Description of Additional Supplementary Files Supplementary Data 1 Supplementary Information Transparent Peer Review file Description of Additional Supplementary Files Supplementary Data 1 Source data Source Data Source Data