Atomic structures and orbital energies of 61,489 crystal-forming organic molecules

Abstract

Data science and machine learning in materials science require large datasets of technologically relevant molecules or materials. Currently, publicly available molecular datasets with realistic molecular geometries and spectral properties are rare. We here supply a diverse benchmark spectroscopy dataset of 61,489 molecules extracted from organic crystals in the Cambridge Structural Database (CSD), denoted OE62. Molecular equilibrium geometries are reported at the Perdew-Burke-Ernzerhof (PBE) level of density functional theory (DFT) including van der Waals corrections for all 62k molecules. For these geometries, OE62 supplies total energies and orbital eigenvalues at the PBE and the PBE hybrid (PBE0) functional level of DFT for all 62k molecules in vacuum as well as at the PBE0 level for a subset of 30,876 molecules in (implicit) water. For 5,239 molecules in vacuum, the dataset provides quasiparticle energies computed with many-body perturbation theory in the G0W0 approximation with a PBE0 starting point (denoted GW5000 in analogy to the GW100 benchmark set (M. van Setten et al. J. Chem. Theory Comput. 12, 5076 (2016))).

Low-Order Scaling G0W0 by Pair Atomic Density Fitting

“… 149 , 150 , 211 We performed PADF- G 0 W 0 @PBE and PADF- G 0 W 0 @PBE0 calculations for all molecules in the GW100 database 82 as well as PADF- G 0 W 0 @PBE0 212 , 213 calculations for selected molecules from the GW5000 database. 214 For GW100, we used the structures as published in the original work, 82 except for vinyl bromide and phenol for which we used the updated structures. 215 To preclude potential confusion, we emphasize that all results from the other codes we refer to herein have been taken from the literature and have not been calculated by us.…”

“… 82 Using localized basis functions, even on the QZ level, BSEs for HOMO and especially LUMO QP energies can exceed several hundred meV, necessitating a CBS limit extrapolation to obtain very accurate reference values. 46 , 82 , 214 Using localized AOs, one needs to rely on heuristics since the expansion of MOs in this basis does not converge uniformly, unlike expansions in terms of PW 224 or finite elements in RS. 225 HOMO QP energies obtained with these basis set types are generally in good agreement with the original ones by van Setten et al., 82 while differences for unbound LUMO energies often exceed 1 eV.…”

“…We now turn our attention to systems large enough for local approximations to take effect and discuss the HOMO and LUMO energies of 20 organic molecules with in between 85 and 99 atoms from the GW5000 database. 214 These tests are crucial for our purpose. First, they allow us to assess the effect of the values of the thresholds controlling distance effects.…”

Active discovery of organic semiconductors

“… 46 , 47 As molecular properties, both ϵ HOMO and λ h can be determined by efficient first-principles calculations as detailed in Supplementary Note 2 , where the density-functional theory (DFT) B3LYP 48 – 50 level of theory constitutes a well established accuracy standard 27 , 31 , 39 , 40 , 51 , matching experimental data 44 , 52 . We emphasize though that using the lowest-energy gas-phase conformer for the descriptor calculation disregards packing-effects in the molecular crystal 53 – 55 and we further discuss the influence of conformers on descriptor values in Supplementary Note 3 .…”

Machine learning in chemical reaction space

“…State-of-the-art methods actually surpass chemical accuracy (ca. 0.05 eV) when applied to standard benchmarks like the QM9 database 34 – 38 . Similarly, ML models can be applied to conformational space (e.g., when trained on ab initio molecular dynamics trajectories) or even interpolate across chemical and conformational space at the same time 39 – 41 .…”