Bonding descriptors as features in machine learning of materials properties Aakash Ashok Naik 1,3 , Christina Ertural 1 , Nidal Dhamrait 1 , Philipp Benner 2 , Janine George 1,3 11 Federal Institute for Materials Research and Testing, Department Materials Chemistry, Berlin, 12205, Germany 2 Federal Institute for Materials Research and Testing, eScience Group, Berlin, 12205, Germany 3 Friedrich Schiller University Jena, Institute of Condensed Matter Theory and Solid-State Optics, Jena, 07743,Germany Understanding the interactions between constituent atoms in crystalline materials can pave the way for developing novel materials with desired properties. 1 In the case of thermoelectric materials, it is possible to obtain ultra-low lattice thermal conductivity by weakening interatomic interactions. 2 One can also construct 2D/3D maps to define new design rules for materials. 3 Albeit such known benefits, no studies involved a large set (>1000) of materials that could provide novel insights between bonding and material properties of interest. Extracting chemical bonding information from crystal structures can be done using either density-based or orbital-based methods. In this project, we use the LOBSTER 4 software package that relies on the orbital-based method. The bonding information is extracted by projecting the plane wave-based wave functions of modern density functional theory computations (DFT) onto a local, atomic orbital basis. Through this research, we aim to build a large database consisting of bonding descriptors and then use these data as input to ML algorithms for investigating the relationship between material properties (elasticity, thermal conductivity) and bonding descriptors. To build the database, we use our automated workflow 5 , which combines the VASP and LOBSTER software packages. We present here the results of this approach by building a database of about 1500 materials 6 consisting largely of semiconductors and insulators for which harmonic phonons have been computed. The quality of curated data is validated by benchmarking the projected densities of states and bonding indicators on standard density- functional theory computations and available heuristics, respectively. Lastly, we demonstrate the utility of our data by building a simple machine-learned model to predict harmonic phonon properties, which demonstrates the predictive power of the bonding features. References 1. R. Hoffmann, Angew. Chem. Int. Ed. Engl. 1987 , 26 , 846–878 2. J. He, Y. Xia, W. Lin, K. Pal, Y. Zhu, M. G. Kanatzidis, C. Wolverton, Adv. Funct. Mater. 2022 , 32 , 2108532. 3. J.-Y. Raty, M. Schumacher, P. Golub, V. L. Deringer, C. Gatti, M. Wuttig, Adv. Mater. 2019 , 31 , 1806280 4. R. Nelson, C. Ertural, J. George, V. L. Deringer, G. Hautier, R. Dronskowski, J. Comput. Chem. 2020 , 41 , 1931–1940 5. J. George, G. Petretto, A. Naik, M. Esters, A. J. Jackson, R. Nelson, R. Dronskowski, G. Rignanese, G. Hautier, ChemPlusChem 2022 , DOI 10.1002/cplu.202200123. 6. G. Petretto, S. Dwaraknath, H. P.C. Miranda, D. Winston, M. Giantomassi, M. J. van Setten, X. Gonze, K. A. Persson, G. Hautier, G.-M. Rignanese, Sci. Data 2018 , 5 , 180065.
P41DD
Made with FlippingBook Learn more on our blog