A new HPC-driven method for accurate estimation of transport coefficients in complex materials.
Understanding how heat, charge, and mass move through materials is essential for a wide range of technologies. From energy conversion and thermal management to electronic devices and solid-state batteries, the ability to predict and control transport properties directly affects efficiency and performance. Transport theory, rooted in the early 20th-century work of Einstein, Onsager, Green, and Kubo, connects microscopic particle dynamics to macroscopic properties such as conductivity, viscosity, and diffusivity.
Estimating transport coefficients using equilibrium molecular dynamics (MD) has long been a challenge. The traditional Green–Kubo (GK) approach requires extremely long simulations to produce statistically reliable results. Current analysis methods struggle to quantify uncertainties, and off-diagonal transport coefficients, which describe coupled phenomena like the thermoelectric effect, are particularly difficult to evaluate accurately. Non-equilibrium MD methods exist, but they are prone to finite-size effects and complex error analysis.
The research introduces a unified, spectral-based method to estimate the full Onsager matrix of transport coefficients, including both diagonal (thermal/electrical conductivity, viscosity) and off-diagonal (thermoelectric) elements, from a single statistical model. Benchmarks on molten salts, liquid water, and the Li₃PS₄ solid-state electrolyte show that this method provides highly accurate results, overcoming limitations of previous approaches while reducing computational cost.
Advanced Spectral Methods and HPC for Transport Coefficients
The research leverages equilibrium MD simulations to generate long time-series of current data. These are analyzed using a novel spectral estimation method that generalizes existing techniques such as cepstral analysis and Bayesian regression. The method models the statistical distribution of the Onsager matrix in the frequency domain, allowing for a one-shot estimation of all transport coefficients using a maximum likelihood (NLL) approach based on Wishart processes.
High-performance computing (HPC) resources were essential for this work. Long MD simulations and subsequent spectral analysis demand massive computational power, memory, and data handling capabilities. MaX Center of Excellence provided the infrastructure necessary to run these simulations efficiently, enabling precise statistical analysis and validation against Monte Carlo methods.
The new approach successfully predicts the full Onsager matrix from single simulations, with results in excellent agreement with computationally intensive Monte Carlo benchmarks. Key achievements include:
- Accurate estimation of thermal and electrical conductivity, and shear viscosity for molten salts and liquid water.
- Reliable evaluation of off-diagonal coefficients such as the ionic Seebeck coefficient.
- Application to Li₃PS₄ solid-state electrolyte, showing its potential as a low-temperature ionic thermoelectric material.
This framework provides a significant improvement in both accuracy and efficiency over previous methods, bridging a long-standing gap in transport property prediction.
Impact, Insights, and Future Directions in Transport Theory
The study demonstrates that spectral-based analysis combined with Bayesian regression offers a unified, robust, and computationally efficient method for estimating transport coefficients in complex materials. By enabling the simultaneous evaluation of all elements of the Onsager matrix, this method overcomes major limitations of traditional approaches and provides a reliable tool for materials research.
These results have wide-ranging implications for materials design and discovery. Accurate prediction of heat, charge, and mass transport can inform the development of high-performance electrolytes, thermoelectric materials, and devices requiring precise thermal management. This research highlights the critical role of HPC in enabling simulations and analyses that were previously infeasible.
Future work could explore the application of this method to even more complex systems, including multi-component electrolytes, nanostructured materials, and high-entropy alloys. Further integration with machine learning approaches may allow for accelerated prediction and optimization of transport properties across broader material spaces. Additionally, refining the statistical models for systems with strong low-frequency spectral features could enhance the method’s accuracy in challenging cases.
Acknowledgment
This research was partially supported by the European Commission through the MAX Center of Excellence for supercomputing applications (Grant No. 101093374).
Reference article
Paolo Pegolo, Enrico Drigo, Federico Grasselli, Stefano Baroni; Transport coefficients from equilibrium molecular dynamics. J. Chem. Phys. 14 February 2025; 162 (6): 064111. https://doi.org/10.1063/5.0249677