*Sandro Wieser*

Metal-organic frameworks (MOFs) are a class of highly porous hybrid materials with many possible applications, for example, in catalysis, in gas storage and separation, or as electronic components. Many of these application involve processes generating heat. For optimal large-scale implementation this heat needs to be dissipated effectively, which is why it is our objective to understand structurally dependent heat transfer processes in MOFs.

In this work, non-equilibrium molecular dynamics (NEMD) simulations were employed for an accurate prediction of the thermal conductivity and to analyze the flow of thermal energy in real space. Due to the associated high computational cost, it was necessary to develop computationally efficient force field potentials (FFPs) based on ab-initio reference data. The focus was on an accurate description of phonon properties in the respective materials, as those play a crucial role for thermal transport. As an initial approach, MOF-FF [1] type force fields were parametrized to model the atomistic interactions in a group of isoreticular metal-organic frameworks (IRMOFs). This allowed to identify the interface between node and linker as the primary heat transfer bottleneck in MOFs [2]. To develop structure-to-property relations for such systems, the structure of the base MOF was systematically varied, changing the nature of the metal ions, and changing the length and nature of the organic linkers [3].

A drawback of this approach is that developing MOF-FF type potentials is extremely costly regarding the use of human resources and for more complex MOFs also turned out to be not sufficiently accurate. To overcome these problems, recently developed machine-learning methodologies were explored to provide an easier approach for generating force-field potentials. This led to the creation of astonishingly accurate moment tensor potentials (MTPs) [4], which can be generated with minimal human input and only slightly increased computational costs. The MTPs were extensively tested based on the elastic and phonon properties of a set of the most commonly studied MOFs, yielding highly promising results. This will greatly increase the multitude of MOFs for which structurally-dependent heat transport properties can be estimated at close to ab-initio accuracy.

[1] S. Bureekaew, S. Amirjalayer, M. Tafipolsky, C. Spickermann, T. K. Roy, R. Schmid, Phys. Status Solidi Basic Res. 2013, 250, 1128.

[2] S. Wieser, T. Kamencek, J. P. Dürholt, R. Schmid, N. Bedoya-Martínez, E. Zojer, Adv. Theory Simulations 2021, 4, 2000211.

[3] S. Wieser, T. Kamencek, R. Schmid, N. Bedoya-Martínez, E. Zojer, Nanomaterials 2022, 12, 2142.

[4] I. S. Novikov, K. Gubaev, E. V Podryabinkin, A. V Shapeev, Mach. Learn. Sci. Technol. 2021, 2, 025002.