Capturing intensive and extensive DFT/TDDFT molecular properties with machine learning

Abstract

Abstract. Machine learning has been successfully applied to the prediction of chemical properties of small organic molecules such as energies or polarizabilities. Compared to these properties, the electronic excitation energies pose a much more challenging learning problem. Here, we examine the applicability of two existing machine learning methodologies to the prediction of excitation energies from time-dependent density functional theory. To this end, we systematically study the performance of various 2-and 3-body descriptors as well as the deep neural network SchNet to predict extensive as well as intensive properties such as the transition energies from the ground state to the first and second excited state. As perhaps expected current state-of-the-art machine learning techniques are more suited to predict extensive as opposed to intensive quantities. We speculate on the need to develop global descriptors that can describe both extensive and intensive properties on equal footing.

From DFT to machine learning: recent approaches to materials science–a review

“…Popular descriptors with these properties can be categorized into few cases: structural data such as Coulomb matrices [248,457,470], molecular strings or graphs [264,455,457], and polymer fingerprinting [457][458][459]; simple atomic properties of the constituent species [460,462,466], and DFT-derived data, such as PBE/LDAlevel bandgaps and hybrid-level electronic density [206,272,452,456,461,471,472]. Frequently a combination of two or more classes of descriptors [453,457,460,473] as well as experimental data as features [272,465] is found in the literature. The overall picture is that the community is aware of the importance of careful selection of the descriptor set.…”

Chemical diversity in molecular orbital energy predictions with kernel ridge regression

“…[8,9] In the context of ab initio molecular science, ML has been applied to learn a variety of molecular properties such as atomization energies, [10][11][12][13][14][15][16][17][18][19] polarizabilities, [12,14,16,19,20] electron ionization energies and affinities, [12,16,19,60] dipole moments, [12][13][14][20][21][22][23] enthalpies, [12,13,18] band gaps, [12][13][14]20], binding energies on surfaces [59] as well as heat capacities. [12][13][14] A few studies addressed the prediction of spectroscopically relevant observables, such as electronic excitations, [12,16,19,24] ionization potentials, [12,16,19,25] nuclear chemical shifts,…”

Deep Learning for Deep Chemistry: Optimizing the Prediction of Chemical Patterns

“…However, the computational cost inherent to these classical approximations have limited the size, flexibility, and extensibility of the studies. Larger searches on relevant chemical patterns, have been successfully conducted since several research groups have developed ML models and algorithms to predict chemical properties using training data generated by DFT, which have also contributed to the increase of public collections of molecules coupled with vibrational, thermodynamic and DFT computed electronic properties (e.g., Behler and Parrinello, 2007;Rupp et al, 2012;Behler, 2016;Hegde and Bowen, 2017;Pronobis et al, 2018;Chandrasekaran et al, 2019;Iype and Urolagin, 2019;Marques et al, 2019;Schleder et al, 2019).…”

“…These are extracted from recent contributions, that can be regarded as complementary and providing an overall perspective of the applications. These include different approaches for (i) understanding and controlling chemical systems and related behavior (Chakravarti, 2018;Fuchs et al, 2018;Janet et al, 2018;Elton et al, 2019;Mezei and Von Lilienfeld, 2019;Sanchez-Lengeling et al, 2019;Venkatasubramanian, 2019;Xu et al, 2019;Zhang et al, 2019), (ii) calculating, optimizing, or predicting structure-property relationships (Varnek and Baskin, 2012;Ramakrishnan et al, 2014;Goh et al, 2017;Simões et al, 2018;Chandrasekaran et al, 2019), density functional theory (DFT) functionals, and interatomic potentials (Snyder et al, 2012;Ramakrishnan et al, 2015;Faber et al, 2017;Hegde and Bowen, 2017;Smith et al, 2017;Pronobis et al, 2018;Mezei and Von Lilienfeld, 2019;Schleder et al, 2019), (iii) driving generative models for inverse design (i.e., produce stable molecules from a set of desired properties) (White and Wilson, 2010;Benjamin et al, 2017;Kadurin et al, 2017;Harel and Radinsky, 2018;Jørgensen et al, 2018b;Kang and Cho, 2018;Li et al, 2018b;Sanchez-Lengeling and Aspuru-Guzik, 2018;Schneider, 2018;Arús-Pous et al, 2019;Freeze et al, 2019;Jensen, 2019), (iv) screening, synthesizing, and characterizing new compounds and materials…”