Interactive-quantum-chemical-descriptors enabling accurate prediction of an activation energy through machine learning. (26th August 2020)

• Record Type: Journal Article
• Authors: Mikami, Koichiro
• Is Part Of: Polymer. Volume 203(2020)
• Publication Date: 2020-08-26
• Subjects: Machine-learning -- DFT calculation -- Metallocene -- Mechanism -- Descriptor; Polymers -- Periodicals; Polymerization -- Periodicals; Polymères -- Périodiques; Polymérisation -- Périodiques; 547.7
• DOI: 10.1016/j.polymer.2020.122738
• Languages: English
• ISSNs: 0032-3861

Abstract

Abstract: Artificial intelligence- and machine learning (ML)-assisted reaction/material development are an emerging research area in organic, organometallic, polymer chemistry and materials science. Quantum chemical descriptors (QCDs) that are classically constructed with steric/electrostatic parameters make the process of the prediction through ML easily understood and allow us to find new chemical pictures for reaction, materials and functionality. Herein, I present the development of novel QCDs—interactive-quantum-chemical-descriptors (IQCDs)—well-expressing an intermolecular interaction among target molecules. The use of IQCDs drastically improved the prediction-accuracy rather than the use of only the classical QCD. One of the IQCDs consists of natural energy decomposition analysis (NEDA), well-expressing a chemical interaction among the molecules/materials, which would be applicable for dynamic processes including formation of chemical bonding, organometallic complex, and supramolecular complex. Graphical abstract: Image 1 Highlights: The IQCDs were serendipitously discovered; their advantage was demonstrated through the prediction of an activation energy using MLs. The use of IQCDs drastically improved the prediction-accuracy rather than the use of only the classical QCD. One of the IQCDs consists of natural energy decomposition analysis (NEDA), which would be applicable for dynamic processes.