Comparing the Expense and Accuracy of Methods to Simulate Atomic Vibrations in Rubrene

Abstract

Atomic vibrations can inform about materials properties from hole transport in organic semiconductors to correlated disorder in metal−organic frameworks. Currently, there are several methods for predicting these vibrations using simulations, but the accuracy−efficiency tradeoffs have not been examined in depth. In this study, rubrene is used as a model system to predict atomic vibrational properties using six different simulation methods: density functional theory, density functional tight binding, density functional tight binding with a Chebyshev polynomial-based correction, a trained machine learning model, a pretrained machine learning model called ANI-1, and a classical forcefield model. The accuracy of each method is evaluated by comparison to the experimental inelastic neutron scattering spectrum. All methods discussed here show some accuracy across a wide energy region, though the Chebyshev-corrected tight-binding method showed the optimal combination of high accuracy with low expense. We then offer broad simulation guidelines to yield efficient, accurate results for inelastic neutron scattering spectrum prediction.

Elucidating correlated defects in metal organic frameworks using theory-guided inelastic neutron scattering spectroscopy

“…The high costs of the DFT calculations have, for the most part, limited this approach to simple crystalline materials with high structural symmetry (including a few nearly defect-free MOFs). 53,54 For example, Deacon et al 42 combined DFT calculations with INS spectra to show the pressure-induced phase transition of ZIF-L to ZIF-8. Because defects in MOFs reduce the overall symmetry of the structure, INS simulations of defective MOFs require larger simulation volumes and are orders of magnitude more computationally expensive than those of their pristine analogs.…”

Multireference Generalization of the Weighted Thermodynamic Perturbation Method

“…The “reweighting entropy (RE)” is an index developed specifically for this purpose, and we shall discuss it in more detail in subsequent sections. We were motivated to develop the gwTP method by the growing interest in supplementing semiempirical QM/MM Hamiltonians with machine learning potentials (MLP) that are trained to reproduce ab initio QM/MM energies and forces. ,− If one had a priori confidence in the trained MLP, then one obviously can use it to estimate FESs without additional correction; however, if one is applying an MLP to a system that was not explicitly included in the training, then an MLP-corrected model naturally serves as an excellent reference potential to estimate the ab initio FES from reweighting. Furthermore, the active learning procedure used to train MLPs produces several neural network parameter sets (several potentials), , and the gwTP method provides a means to estimate the ab initio FES from the aggregate sampling performed with each potential.…”

Davis Computational Spectroscopy Workflow—From Structure to Spectra

“…Molecular dynamics (MD) force fields can generally overcome these system size limitations, but are frequently difficult to parameterize with sufficient accuracy to model the low-frequency modes of interest and largely exhibit poor transferability to new materials. 31,32 Hence, the intrinsic complexity of OEMs with disordered, mixed, and amorphous structures requires electronic modeling that combines a much lower computational cost with a greater accuracy than the options that are widely available.…”

“…DCS-Flow thus provides an efficient workflow to create databases of inelastic neutron scattering simulations. 50 In its core, we have linked powerful packages such as the atomic simulation environment (ASE), 51 phonopy, 52 and OCLIMAX. 53 It provides an almost identical flow of calculations despite using a diverse group of calculators such as VASP, 54 CASTEP, 55 DFTB+, 56 and DFTB/ChIMES.…”

“…DFT has proven to yield accurate INS spectra for a number of crystalline materials including small-molecule and polymer OEMs. , However, DFT calculations are extremely computationally intensive and are thereby limited to studies of highly ordered systems with repeating boundary conditions, thus excluding inclusion of realistic concentrations of point defects or long-range thermally induced disorder. Molecular dynamics (MD) force fields can generally overcome these system size limitations, but are frequently difficult to parameterize with sufficient accuracy to model the low-frequency modes of interest and largely exhibit poor transferability to new materials. , Hence, the intrinsic complexity of OEMs with disordered, mixed, and amorphous structures requires electronic modeling that combines a much lower computational cost with a greater accuracy than the options that are widely available.…”