Davis Computational Spectroscopy Workflow—From Structure to Spectra

Abstract

We describe an automated workflow that connects a series of atomic simulation tools to investigate the relationship between atomic structure, lattice dynamics, materials properties, and inelastic neutron scattering (INS) spectra. Starting from the atomic simulation environment (ASE) as an interface, we demonstrate the use of a selection of calculators, including density functional theory (DFT) and density functional tight binding (DFTB), to optimize the structures and calculate interatomic force constants. We present the use of our workflow to compute the phonon frequencies and eigenvectors, which are required to accurately simulate the INS spectra in crystalline solids like diamond and graphite as well as molecular solids like rubrene. We have also implemented a machine-learning force field based on Chebyshev polynomials called the Chebyshev interaction model for efficient simulation (ChIMES) to improve the accuracy of the DFTB simulations. We then explore the transferability of our DFTB/ChIMES models by comparing simulations derived from different training sets. We show that DFTB/ChIMES demonstrates ∼100× reduction in computational expense while retaining most of the accuracy of DFT as well as yielding high accuracy for different materials outside of our training sets. The DFTB/ChIMES method within the workflow expands the possibilities to use simulations to accurately predict materials properties of increasingly complex structures that would be unfeasible with ab initio methods.

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

“…This means that DFTB with ChIMES is theoretically equivalent to the other methods. Recently, the entire DFTB/ChIMES workflow has been wrapped in a command-line tool and python package known as DCS flow …”

“…These include correcting the unit cell based on a DFT unit cell and reassigning phonon frequencies using a Γ-point DFT phonon calculation . We note that our comparison does not include recent machine learning corrections to DFTB to yield high-level quantum chemical accuracy for molecular energies, specifically. − The main difference between ChIMES and these other phonon correction methods in DFTB is that ChIMES has demonstrated the ability to transfer a model determined from a single training set to multiple molecules, , rather than requiring the creation of a unique parameter set for each system of interest. This means that when studying a series of similar molecules, the computational expense for the correction is only incurred once for ChIMES, enhancing its general appeal for predictions related to the vibrational properties of molecular crystals.…”

“…Recently, the entire DFTB/ChIMES workflow has been wrapped in a command-line tool and python package known as DCS flow. 45 Machine Learning Potentials. In the case of ML, the atomistic simulation environment (ASE) 46 along with the atomistic machine learning package (AMP) 37 was used to train a neural net to calculate forces and energies based on the atomic configuration of crystalline rubrene.…”

Catching the Killer: Dynamic Disorder Design Rules for Small‐Molecule Organic Semiconductors

“…[31][32][33] This slow process is incompatible with the type of multi-system studies necessary to search design space computationally and discover new synthetic design rules. Previously, we introduced DCS-Flow: [34] a workflow to compute the phonon modes using various calculators including DFT, density functional tight binding (DFTB), [35] or DFTB with a Chebyshev Interaction Model for Efficient Simulation (ChIMES) [36] correction from just a .cif file. Here, we introduce another workflow, ElPh, which works alongside DCS-Flow to compute the electronic structure using the method popularized by the Troisi and Fratini groups.…”

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

“…To work toward overcoming these limitations, we developed a Python-based workflow (denoted as DCS-Flow 55 ) that facilitates seamless integration between the separate components of an INS simulation. DCS-Flow is an accurate, reproducible, efficient workflow for calculating phonon modes and the INS spectrum using DFT.…”