Metal nanoparticles are useful as catalysts having specific reactivity owing to highly reactive site and strong size dependency. Structural information of metal nanoparticles is essential for interpretation and prediction of their reactivity. Wulff theorem predicts the equilibrium structures of crystals by using the surface energies of plane indices such as (111), (110), and (100). In this study, we evaluated the surface energies of well-defined Rh surfaces by the first principles calculations, followed by systematically constructing various sizes of Rh nanoparticles based on the Wulff theorem. For small nanoparticles with radii of 2 nm or less, only the (111) and (100) planes were present. On the other hand, high index surfaces appeared at large nanoparticles, of which the radii were more than 2.5 nm.

We previously proposed an excitation configuration analysis for the divide-and-conquer (DC)-based excited-state calculation method using dynamical polarizability to interpret the nature of excited states [J. Chem. Phys. 160, 244103 (2024)]. This article reviews the proposed DC-based excitation configuration analysis and applies the natural transition orbital analysis based on the (de-)excitation coefficients obtained from the proposed method to the lowest excited state of tris-triphenylacetylammonia. The singular values of excitation coefficients were matched to the results of the TDHF method. Also, the shapes of the natural transition orbitals coincided as well.

To verify the possibility to simulate depolymerization, simulations of the depolymerization of polystyrene were performed. Molecular Dynamics simulations (MD) using neural network potentials were found to be similar in accuracy to MD using density functional theory calculations. It was also found that long-time simulations using neural network potential-MD predicted styrene monomer yields close to those obtained experimentally, and that the monomer yields tended to decrease with increasing pressure.

Optical spectrum prediction based on first-principles calculations is important for the development of optical materials. In particular, Time Dependent Density Functional Theory (TDDFT) in real-time is one of the most widely used calculation methods. In real-time TDDFT, the dynamic dipole moment is used to obtain the polarizability by Fourier transform (FT). The optical spectrum can be obtained from this polarizability. However, if the time length is not sufficient, the spectrum becomes ambiguous. To solve this problem, we introduced singular spectrum analysis (SSA), which decomposes time series data into multiple orthogonal oscillations. By extracting only the main components related to the peak of interest, the corresponding time series data is reproduced. In this process, high-frequency oscillations recognized as noise are removed. We applied this method to TDDFT time-series data for ethylene and small molecules of benzene, naphthalene, anthracene and tetracene. We focused on band edges, which are important for understanding optical properties, to clarify the signals. Even when the time-series data is insufficient, we found that it is possible to obtain time-series data of sufficient time length by isolating the oscillation components contributing to the band edges and expanding and complementing them with time-series prediction.

Molecular dynamics simulation of y{(1−x)Na₂O-xK₂O}-(1−y)SiO₂ glasses used an improved interatomic potential was performed to investigate the mixed alkali effect. The relation of self-diffusion coefficient of potassium and of sodium was improved, but the trend with x of the self-diffusion coefficient of potassium has become worse than the previous work.

For example， dissolved carbon dioxide and carbonic acid in aqueous solution are important compounds for life activities， serving as a buffer system for pH regulation in the body. Theoretical calculations for carbon dioxide and carbonic acid， and the transition states， which are the states mediated by water molecules have been vigorously performed， however， there are limitations in terms of computational speed and resources for exploring various conformations and reaction mechanisms in the water molecule network through ab initio calculations. The machine learning potential (MLP) was created by using the DeepMD framework， which is one of the MLPs that have been reported to be applied in various systems. We analyzed the conformation of the dissociation process of carbonic acid by using the Nudged Elastic Band method in the molecular dynamics (MD) calculations. The energy difference between MLP-MD and ab-initio MD (AIMD) at the beginning and the end of the reaction was 0.109 kcal/mol/atoms. Looking at the trend of the barrier value with respect to the number of water molecules in the water molecular network， the barrier value to decrease with the increase in the number of water molecules has similar tendency of the previous quantum chemical and AIMD calculations.

It is known that it is possible to develop a chemically intelligent robot that can perceive spatial extent using the BZ reaction. If the BZ reaction can be used for learning, there is a possibility of developing a chemically intelligent robot that can make complex judgments about situations like living organisms. In the physical reservoir calculation, it is possible to construct a learning system using a material system, so in this study, we build a physical reservoir calculation using the BZ reaction and confirmed that learning is possible. The physical reservoir computation was capable of performing tasks such as time series prediction and speech recognition. BZ reaction, nonlinear chemical reaction, reservoir computing, time series prediction, speech recognition

We constructed a mathematical model to predict the 2,2-diphenyl-1-picrylhydrazyl (DPPH) free radical scavenging capacity (IC₅₀) for recently synthesized ferulic acid derivatives by machine learning with molecular orbital energy as an explanatory variable and IC₅₀ as an objective variable. We compared 96 regression models including xgbLinear and neuralnet included in R/caret package. We were able to construct IC₅₀ prediction models for these new ferulic acids by using xgbLinear, M5, ppr, and neuralnet as regression methods.