Accelerating First-Principles Molecular-Dynamics Thermal Conductivity Calculations for Complex Systems

Area of Science:

• Computational Materials Science
• Condensed Matter Physics
• Nanotechnology

Background:

• Atomistic simulations of heat transport are computationally expensive and difficult to converge.
• Noise-reduction techniques for equilibrium molecular dynamics (MD) simulations have been developed to address these challenges.
• InAs nanowires, with their complex structures and phonon spectra, serve as a benchmark for evaluating these techniques in quasi-1D systems.

Purpose of the Study:

• To analyze the performance of noise-reduction strategies for atomistic heat transport simulations.
• To evaluate the effectiveness of cepstral analysis for low- and high-thermal-conductivity systems.
• To investigate alternative methods, including uncertainty propagation and covariance matrix contributions, for accurate error assessment.

Main Methods:

• Benchmarking noise-reduction techniques using InAs nanowires.
• Applying cepstral analysis to atomistic simulations of heat transport.
• Utilizing uncertainty propagation from independent simulations, including covariance matrix contributions.
• Integrating machine-learning interatomic potentials (MLIPs), specifically a transferable MACE potential, into the workflow.

Main Results:

• Cepstral analysis effectively reduces computational cost and provides accurate results for low-thermal-conductivity systems without data discarding.
• Cepstral analysis significantly underestimates thermal conductivity in high-thermal-conductivity systems.
• Including covariance matrix contributions is crucial for quantitative error assessment in thermal conductivity calculations.
• The combination of noise-reduction strategies and MLIPs offers an accelerated and robust simulation workflow.

Conclusions:

• Cepstral analysis is a valuable tool for specific material types but requires complementary methods for others.
• Accurate assessment of thermal conductivity in complex materials necessitates careful error analysis, including covariance.
• Machine-learning potentials significantly enhance the efficiency and applicability of these simulation techniques across diverse materials.