Cepstral Analysis to accelerate Green-Kubo thermal conductivity calculations of Metal-Organic Frameworks

Abstract

Metal-organic frameworks (MOFs) are promising porous materials for applications such as gas storage and separation, where heat transport can critically affect device performance. However, reliable computational prediction of their thermal conductivities remains challenging. In particular, equilibrium molecular-dynamics-based Green-Kubo (GK) simulations, as the most widely used approach, are severely affected by statistical noise. Moreover, they rely on multiple ambiguous, user-defined parameters, which hinder transferability and automation. Here, we demonstrate for metal-organic frameworks that cepstral analysis in combination with GK simulations provides a robust route to massively mitigate these problems, while simultaneously reducing the required sampling times. This is shown for three prototypical frameworks, MOF-5, HKUST-1, and ZIF-8, employing machine-learned moment tensor potentials trained on DFT reference data. In contrast to conventional, direct GK analysis, which shows erratic convergence and strong sensitivity to ad hoc choices of parameters, the cepstral approach yields stable results across a wide range of correlation lengths and achieves convergence within about 1-2 ns of total sampling time. This establishes cepstral analysis base Green-Kubo simulations combined with machine-learned potentials as an efficient, reproducible and automation-ready framework for near ab initio accuracy prediction of thermal transport in MOFs and other complex low-thermal-conductivity materials.