🔬 materials science

Temperature-Dependent Ion Dynamics in Argyrodite Superionic Conductor Li $_{6}$ PS $_{5}$ Cl with Fine-Tuned MACE Foundation Model

By combining molecular dynamics simulations with a fine-tuned MACE foundation model and essential dynamics analysis, this study reveals that Li $^+$ ion dynamics in the argyrodite superionic conductor Li $_{6}$ PS $_{5}$ Cl transition from localized, low-dimensional phonon-like behavior below 500 K to a complex regime of strong competition with framework ions between 500–700 K, before exhibiting distinct high-temperature behavior.

Original author: Jack Yang, Lei Ao, Sean Li

Published 2026-11-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original author: Jack Yang, Lei Ao, Sean Li

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer

Inside the solid world of batteries and advanced materials, there exists a strange state of matter that behaves like both a rigid crystal and a flowing liquid. In most solids, atoms are locked in place, vibrating gently around fixed spots like people sitting in a crowded theater. In liquids, atoms roam freely, sliding past one another like a bustling crowd. Superionic conductors occupy the rare middle ground between these two states. In these special materials, a specific type of atom, usually a charged particle called an ion, breaks free from its seat and begins to zip through the solid structure with the ease of a liquid, while the rest of the material remains solid and stable. This unique behavior is the key to creating better, safer batteries that can charge faster and hold more energy, yet scientists have struggled to fully understand how these ions move so freely without the solid framework collapsing.

To solve this puzzle, researchers Jack Yang, Lei Ao, and Sean Li turned to a powerful new tool: a type of artificial intelligence designed to predict how atoms interact. Instead of relying on traditional, slower methods of calculation, they used a sophisticated machine-learning model called MACE, which acts like a highly accurate map of atomic forces. They focused on a specific material known as Li6PS5Cl, a type of argyrodite crystal that is famous for its ability to conduct lithium ions. By training this AI model on data from high-speed computer simulations, the team was able to watch the movement of thousands of atoms over time, observing how the lithium ions behaved as the temperature changed from a cool 200 degrees up to a scorching 800 degrees.

The researchers discovered that the journey of these lithium ions is not a simple, straight path that gets faster as it gets hotter. Instead, the ions follow a complex, shifting route that depends entirely on how the surrounding crystal structure vibrates. At lower temperatures, below 500 degrees, the lithium ions are trapped in small, localized pockets. They vibrate intensely but stay in one place, moving along a very specific, curved path that is difficult to predict. In this state, the ions are essentially dancing in place, confined by the rigid walls of their atomic cage.

However, as the temperature rises between 500 and 700 degrees, the story changes dramatically. The researchers found a fierce competition between the lithium ions and the framework atoms that hold the crystal together. The framework begins to shake violently, and the lithium ions struggle to find a way through. This is a chaotic transition zone where the ions are neither fully trapped nor fully free. The team observed that the movement of the lithium ions becomes highly dependent on the motion of the surrounding atoms, creating a complex geometric pattern that does not follow a simple rule. It is as if the ions are trying to squeeze through a door that is constantly being pushed open and shut by the vibrating walls.

Once the temperature climbs past 700 degrees, the system settles into a new rhythm. The framework atoms begin to move in sync with the lithium ions, effectively clearing the path for them. The lithium ions are no longer fighting against the structure; instead, they flow through it in large, sweeping motions. At this stage, the ions behave much more like a liquid, diffusing rapidly across long distances. The researchers confirmed this by running special simulations where they artificially slowed down the movement of the framework atoms. When they did this at lower temperatures, the lithium ions moved faster, suggesting that the shaking walls were actually blocking them. But at the highest temperatures, slowing the framework actually stopped the lithium ions from moving, proving that the ions now rely on the framework's motion to help them travel.

The study also revealed that the way these ions move is deeply connected to the geometry of their path. At low temperatures, the path is a tight, twisted loop. At high temperatures, the path opens up into a wide, smooth highway. The researchers used advanced mathematical techniques to visualize these paths, showing that the ions do not just move randomly; they follow specific, collective patterns that change as the material heats up. This work provides a clearer picture of how solid materials can suddenly become excellent conductors, offering a new way to design better battery materials by understanding exactly how the atomic structure helps or hinders the flow of electricity.

Technical Summary: Temperature-Dependent Ion Dynamics in Argyrodite Superionic Conductor Li6PS5Cl with Fine-Tuned MACE Foundation Model

Problem Statement
Superionic conductors, such as the argyrodite Li6PS5Cl, exist in a unique state between ordered crystals and random liquids. While the harmonic phonon model adequately describes localized ionic vibrations below transition temperatures, it fails to capture the large-amplitude, liquid-like ionic diffusion that occurs above these temperatures, even with higher-order anharmonic corrections. Understanding the complex, temperature-dependent interplay between mobile ions (Li+) and the crystal framework is critical for advancing solid-state battery technologies. However, traditional theoretical frameworks struggle to describe these dynamics over the long timescales required to observe diffusion. Furthermore, while Machine Learning Interatomic Potentials (MLIPs) offer a pathway to long-time molecular dynamics (MD), pre-trained foundation models often suffer from systematic softening and may not accurately reproduce the specific vibrational characteristics of superionic conductors without adaptation.

Methodology
The authors employed a multi-stage computational approach combining fine-tuned MLIPs with advanced statistical analysis:

Foundation Model Fine-Tuning: The study utilized the MACE (Multi-Atomic Cluster Expansion) foundation model (specifically the omat-0-medium variant). To address the limitations of pre-trained models in reproducing the full frequency range of Li6PS5Cl, the model was fine-tuned using a dataset derived from ab initio molecular dynamics (AIMD) simulations. These AIMD simulations covered a broad temperature range (10–700 K) to ensure the model could handle the diverse local chemical environments, particularly the highly mobile Li+ ions and rattling PS4 tetrahedra.
Molecular Dynamics Simulations: Using the fine-tuned MACE model, the authors performed NVE and Langevin MD simulations on a (3×3×3) supercell (1,404 atoms) across temperatures ranging from 200 K to 800 K. Langevin dynamics were specifically employed at 500, 600, and 700 K to systematically dampen the framework ion motions, isolating their influence on Li+ dynamics.
Essential Dynamics Analysis (EDA): Adapted from biomolecular simulations, EDA was applied to the MD trajectories. This involved constructing mass-weighted covariance matrices of atomic displacements to identify dominant collective modes. The analysis focused on the eigenvalue spectra and participation ratios of the full system versus Li-only and framework-only sub-blocks.
Geometric and Topological Analysis: The authors projected high-dimensional displacement vectors onto low-dimensional manifolds using principal components. They quantified the local nonlinearity of these manifolds using a tangent space rotation metric ( $\kappa_{tan}$ ) to characterize the geometric structure of ion trajectories.
Temporal Correlation Analysis: Lag-time-dependent covariance matrices were computed to analyze the correlation between Li+ and framework ions over time, distinguishing between concerted (cooperative) and anti-cooperative motions.

Key Results

Model Performance: The fine-tuned MACE model successfully reproduced key features of the phonon density of states (PDOS) for Li6PS5Cl, including the linear variation of Li PDOS at low frequencies (a signature of liquid-like conduction) and the distinct contributions of Cl, P, and S vibrations. However, the model retained a minor artifact of finite PDOS at zero frequency at 200 K, attributed to the nature of the baseline training data.
Temperature-Dependent Dynamics:
- Low Temperature (<500 K): Li+ ions exhibit localized, phonon-like vibrations. The trajectories lie on a highly nonlinear, low-dimensional manifold. The eigenvalue spectrum shows that many modes are required to describe the motion, and the system exhibits strong correlations between Li+ and framework vibrations.
- Transition Regime (500–700 K): This region is characterized by a strong competition between Li+ and framework dynamics. At 600 K, the system undergoes a transition where Li+ ions are liberated to undergo large-amplitude movements, yet the framework still plays a role. At 700 K, the leading eigenvectors exhibit the strongest contributions from the movements of framework ions, indicating that the framework moves concomitantly with the Li+ sublattice. Consequently, the diffusion pathway of the Li+ ions lies in the most complex manifold in terms of dimensionality at this temperature window, and the projected configurations show significant local nonlinearity.
- High Temperature (>700 K): As the temperature rises to 800 K, the lighter Li+ ions gain sufficient thermal energy, and the ionic dynamics become dominated by them again. This regime exhibits large-amplitude diffusive motions over a smoother, more linear manifold, indicative of the superionic state.
Framework-Li+ Interplay: Langevin dynamics revealed that damping the framework motions has temperature-dependent effects. At 500 K, damping the framework paradoxically promotes Li+ conduction by removing low-energy barriers, whereas at 700 K, damping the framework liberates the Li+ ions further, suggesting that in the superionic regime, the thermally activated framework motions act as a frictional force slowing down Li+ migration.
Geometric Signatures: The analysis of local nonlinearity ( $\kappa_{tan}$ ) confirmed that low-temperature trajectories are highly curved (nonlinear). The 700 K regime also displays significant local nonlinearity, reflecting the complex geometric structures underpinning the collective movements of Li+ ions at this temperature. The 600 K regime showed distinct, isolated clusters in the projected space, representing unique collective displacement patterns not seen at other temperatures.
Temporal Correlations: Correlations between Li+ ions are strong at low temperatures (phonon-mediated) but diminish rapidly as temperature increases. At 600 K and above, correlations become negligible, consistent with liquid-like behavior. However, short-lived, transient correlations may still exist locally.

Significance and Claims
The paper claims to contribute to the theoretical and data-driven framework for studying lattice dynamics in solids by demonstrating a unified approach that bridges localized phonon-like vibrations and large-amplitude ionic diffusion.

Methodological Contribution: The work establishes that fine-tuning pre-trained foundation models (like MACE) with targeted, temperature-diverse datasets is essential for accurately capturing the dynamics of superionic conductors, where mobile ions traverse configurational spaces not present in standard training sets.
Physical Insight: The study provides a detailed geometric and dynamic characterization of the Li6PS5Cl superionic transition. It challenges the view of a simple monotonic transition, instead revealing a complex regime (500–700 K) where framework and Li+ dynamics compete.
Unified Framework: By combining MLIP-MD, essential dynamics, and differential geometric analysis, the authors present a single mathematical framework capable of describing the full spectrum of ion dynamics, from localized vibrations to free diffusion.

The authors remain modest regarding the specific anharmonicity of the framework, noting that their analysis showed Gaussian distributions for framework displacements, suggesting that the observed effects are not driven by notable anharmonicity in the framework itself within the simulated range. They conclude that extending these methods to higher-order correlation tensors and exploring differential geometry for diffusion pathways are promising future directions.

More like this