Machine learning for vibrational spectroscopy via divide-and-conquer semiclassical initial value representation molecular dynamics with application to N-methylacetamide

Abstract

Paper published as part of the special topic on Quantum Dynamics with ab Initio PotentialsQDAB2020 ARTICLES YOU MAY BE INTERESTED IN

e-Graphene: A Computational Platform for the Prediction of Graphene-Based Drug Delivery System by Quantum Genetic Algorithm and Cascade Protocol

“…For the rapid optimization, we have implemented Tripos FF in e-Graphene to automatically minimize the models by Powell minimization method that is natively coded in e-Graphene. At the same time, we also have developed an interface to call the external QM/SQM programs, such as Gaussian03/09/16 (Frisch et al, 2016), ORCA4.2 (Neese, 2012), MOPAC2009/2012/2016 (Stewart, 2016), and XTB6.3 (Bannwarth et al, 2019) to further minimize the graphene/GO model in a higher level of accuracy. In order to minimize all the graphene/GO models in batch mode, automatic calculation of the total formal charge for each model is developed in e-Graphene, whereas the multiplicity for each model is set to 1 by default due to our current focus on the ground state of the graphene/GO model.…”

“…According to the MTD method, the Grimme group developed a very useful conformation search program CREST (Bannwarth et al, 2019), where the root-mean-square deviation (RMSD) of a small molecule is adopted as a reaction coordinate (RC) in the MTD simulation, but this RC cannot capture the relative position between the small molecule and graphene/GO. Alternatively, semi-classical MD simulation method was proposed by the Ceotto group to explore various minima on a potential energy surface, which also is not a computationally cheap method and mainly developed for the prediction of molecular spectroscopy (Conte and Ceotto, 2020;Gandolfi et al, 2020). As a result, these methods may be not very suitable for the rapid prediction/screen of GDDS.…”

Caldeira–Leggett model vs ab initio potential: A vibrational spectroscopy test of water solvation

“…This machinery allows one to retrieve clear spectroscopic signals even for high-dimensional and complex systems. Over the past years, MC-DC SCIVR has been applied to the study of systems like small peptides and organic molecules, [54,55] glycine supramolecular systems, [56], nucleobases and nucleotides, [57,58] and molecules adsorbed on titania surfaces. [59] Furthermore, MC-DC SCIVR is being adopted to answer the question concerning the minimum number of water molecules necessary to fully solvate one.…”

Infrared Spectroscopy of Chemically Diverse Carbon Clusters: A Data-Driven Approach

“…15,16 Within the general context of relating properties to structure, several groups have recently shown the benefits of employing ML for vibrational spectroscopy [17][18][19] through a variety of approaches aiming to represent potential energy and electric dipole moment surfaces within perturbative frameworks, 20 to condense molecular information into topological descriptors, 21 or to numerically solve the quantum nuclear dynamics problem by better partitioning the various degrees of freedom. 22 In the present contribution, we explore several ML ideas to reconstruct the infrared spectrum of carbon clusters in a statistical sense, from a limited sample and using interpolation techniques in a multidimensional feature space, supervision being introduced through metric learning. We notably show how such an interpolation scheme can predict the spectral trends for clusters that are not members of the initially chosen sample.…”

“…While the two tasks of sampling the energy landscape and determining the individual spectra can be addressed successively along the lines of a multiscale description, the computational effort associated with the spectral determination can be particularly heavy for large samples, even with efficient methods such as density functional-based tight binding (DFTB). However, extracting the spectral features from a statistical sample seems also naturally suited to be tackled by machine-learning (ML) techniques, a broad range of computational approaches that have become spectacularly popular in chemical physics and physical chemistry in recent years. , Within the general context of relating the properties to the structure, several groups have recently shown the benefits of employing ML for vibrational spectroscopy − through a variety of approaches aiming to represent potential energy and electric dipole moment surfaces within perturbative frameworks, to condense molecular information into topological descriptors, or to numerically solve the quantum nuclear dynamics problem by better partitioning the various degrees of freedom …”