Speaker
Description
Metal-organic frameworks (MOFs) are highly porous, mostly crystalline materials that hold considerable promise for addressing major societal challenges, for example in gas storage (CO2 capture) and separation.[1] For many of these applications, mechanical response and thermal transport are key material properties.[2] As MOFs are typically insulating materials, both properties are governed by lattice vibrations. However, their quantitative prediction is challenging, as their structural complexity and large unit cells place them beyond the reach of routine density functional theory (DFT) simulations. This is particularly severe for observables such as thermal conductivity that require long molecular dynamics simulations of large supercells in order to account for finite size effects.
In recent years, machine-learned interatomic potentials (MLIPs) have emerged as powerful tools combining near-DFT accuracy with the efficiency needed to access the length and time scales relevant for simulating properties of complex materials.[3]
In this contribution, we show how MLIPs enable the investigation of both mechanical and thermal properties in complex materials like MOFs. For mechanical properties, we focus on long-wavelength limit properties i.e. elastic tensors.[4] We combine MLIP-based simulations with Brillouin light scattering (BLS),[5] a contactless, non-invasive spectroscopic technique that measures direction-dependent sound velocities and thereby provides access to the elastic tensor. This combined approach not only provides experimental validation of MLIP predictions for complex materials, but also demonstrates the potential of MLIPs as powerful tools for interpreting BLS measurements, particularly in structurally complex, disordered, or glassy systems.
We then extend this perspective to thermal transport in MOFs. While MLIPs enable detailed insights into heat-transport mechanisms through reciprocal-space analyses based on phonon transport formalisms,[6,7] we focus here on a complementary real-space approach: Green-Kubo (GK) theory.
This is particularly relevant also for cases in which a simple phonon picture might become less useful, for example in the case of freely moving guest molecules within the pores of a MOF. For such cases, GK simulations provide a powerful framework, but their application is often hindered by statistical noise and slow convergence. We show that this problematic can be overcome elegantly by applying Cepstral analysis to GK-based transport calculations, significantly improving both their efficiency and their reliability.[8] Quantitative comparison with single-crystal experiments for MOF-5 and HKUST-1 reveals excellent agreement with the predicted thermal conductivities.[9]
This is of considerable practical importance, because recent progress in MLIPs increasingly points toward more expressive and transferable architectures, often at the cost of computational speed. The substantially reduced simulation times enabled by cepstral-analysis-based GK simulations make the use of such more advanced MLIPs feasible also for more realistic scenarios involving for example guest molecules or defects. Taken together, these results demonstrate how MLIPs are evolving into powerful tools for the quantitative exploration of vibrationally governed properties in complex materials beyond the reach of direct DFT simulations.
[1] J.-B. Lin, T. T. T. Nguyen, R. Vaidhyanathan, J. Burner, J. M. Taylor, H. Durekova, F. Akhtar, R. K. Mah, O. Ghaffari-Nik, S. Marx, N. Fylstra, S. S. Iremonger, K. W. Dawson, P. Sarkar, P. Hovington, A. Rajendran, T. K. Woo, G. K. H. Shimizu, Science 2021, 374, 1464.
[2] N. C. Burtch, J. Heinen, T. D. Bennett, D. Dubbeldam, M. D. Allendorf, Adv. Mater. 2018, 30, 1704124.
[3] S. Wieser, E. Zojer, Npj Comput. Mater. 2024, 10, 18.
[4] F. P. Lindner, N. Strasser, M. Schultze, S. Wieser, C. Slugovc, K. Elsayad, K. J. Koski, E. Zojer, C. Czibula, J. Phys. Chem. Lett. 2025, 16, 1213.
[5] I. Kabakova, J. Zhang, Y. Xiang, S. Caponi, A. Bilenca, J. Guck, G. Scarcelli, Nat. Rev. Methods Primer 2024, 4, 8.
[6] Florian P. Lindner, Marco Moser, Lukas Legenstein, Sandro Wieser, Egbert Zojer, in preparation;
[7] L. Legenstein, L. Reicht, S. Wieser, M. Simoncelli, E. Zojer, Npj Comput. Mater. 2025, 11, 29.
[8] S. Wieser, Y.-J. Cen, G. K. H. Madsen, J. Carrete, J. Chem. Theory Comput. 2026, 22, 513.
[9] Florian P. Lindner, Egbert Zojer, and Sandro Wieser, in preparation;