Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
Initialising ...
and machine learning potentials for modeling nuclear quantum effects in waterThomsen, B.; Nagai, Yuki*; Kobayashi, Keita; Hamada, Ikutaro*; Shiga, Motoyuki
Journal of Chemical Physics, 161(20), p.204109_1 - 204109_18, 2024/11
Times Cited Count：0 Percentile：0.00(Chemistry, Physical)We introduce the self-learning path integral hybrid Monte Carlo with mixed and machine learning potentials (SL-PIHMC-MIX) method which allows the application of hybrid Monte Carlo for both path integrals and for larger system sizes. The method shows savings of an order of magnitude with respect to the number of DFT calculations needed to calculate and converge the structure of room temperature water when using SL-PIHMC-MIX over ab initio path integral molecular dynamics (PIMD).
Nakata, Yuto; Sasaki, Takehiko*; Thomsen, B.; Shiga, Motoyuki
Chemical Physics Letters, 845, p.141285_1 - 141285_9, 2024/06
Times Cited Count：0 Percentile：0.00(Chemistry, Physical)Using density functional theory and metadynamics simulations, we study cellobiose hydrolysis and glucose hydrogenation with silica-supported platinum and palladium catalysts in hot water, relevant to green cellulose conversion. It is found that cellobiose hydrolysis can proceed by the attack of hydrogen atoms adsorbed on metal or protons spilled over to silica forming glucose. Glucose can then be hydrogenated by hydrogen atoms adsorbed at platinum/water interface forming sorbitol. The reaction barriers of hydrolysis and hydrogenation at platinum/water interface are both lower than that at palladium/water interface, which explains the experimental finding that the platinum performs as a better catalyst than palladium.
Shiga, Motoyuki; Thomsen, B.; Kimizuka, Hajime*
Physical Review B, 109(5), p.054303_1 - 054303_12, 2024/02
Times Cited Count：2 Percentile：72.59(Materials Science, Multidisciplinary)Inelastic neutron scattering spectra of hydrogen in palladium were calculated considering nuclear quantum effects at finite temperatures. A computational method combining semiclassical molecular dynamics based on path integrals and machine learning potentials was used. The calculated spectra agree well with the experimental spectra with respect to the positions and intensities of the peaks corresponding to the fundamental and first harmonic of the vibrational excitation of hydrogen atoms. Comparison with classical molecular dynamics shows that nuclear quantum effects play an essential role in the inelastic neutron scattering spectra.
Shiga, Motoyuki; Thomsen, B.; Nagai, Yuki
Ansanburu, 25(4), p.303 - 310, 2023/10
The parallel molecular simulation software "PIMD" will be presented. The use of PIMD will be explained through specific examples such as water structure by ab initio path integral molecular dynamics, quantum diffusion of hydrogen in metal by ring polymer molecular dynamics, machine learning potential generation and phonon properties of superconductors, and polyalcohol dehydration reaction by metadynamics.
Kimizuka, Hajime*; Thomsen, B.; Shiga, Motoyuki
Journal of Physics; Energy (Internet), 4(3), p.034004_1 - 034004_13, 2022/07
Times Cited Count：17 Percentile：76.96(Chemistry, Physical)Artificial neural network-based interatomic potential for a system of palladium and hydrogen was developed, and path integral molecular dynamics simulations were performed to study the quantum diffusion of hydrogen isotopes in palladium crystals. Diffusion coefficients of light and heavy hydrogen were calculated over a wide temperature range of 50-1500 K to clarify the difference in diffusion mechanisms at low and high temperatures.
path integral simulationsThomsen, B.; Shiga, Motoyuki
Physical Chemistry Chemical Physics, 24(18), p.10851 - 10859, 2022/05
Times Cited Count：5 Percentile：46.64(Chemistry, Physical)
study of nuclear quantum effects on sub- and supercritical waterThomsen, B.; Shiga, Motoyuki
Journal of Chemical Physics, 155(19), p.194107_1 - 194107_11, 2021/11
Times Cited Count：8 Percentile：50.20(Chemistry, Physical) path integral molecular dynamicsThomsen, B.; Shiga, Motoyuki
Journal of Chemical Physics, 154(8), p.084117_1 - 084117_10, 2021/02
Times Cited Count：13 Percentile：67.07(Chemistry, Physical)In this study we investigate the nuclear quantum effects on the acidity constant of liquid water isotopologues at the ambient condition by path integral molecular dynamics simulations. This technique not only reproduces the acidity constants of liquid D
O experimentally measured but also allows for a theoretical prediction of the acidity constants of liquid TO, aqueous HDO and HTO, which are unknown due to its scarcity. The results indicate that the nuclear quantum effects play an indispensable role in the absolute determination of acidity constants.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Thomsen, B.; Shiga, Motoyuki
no journal, ,
We report our ongoing efforts to investigate how nuclear quantum effects (NQEs) influence the structure and dynamics properties of water. To model the NQEs we employ path integral molecular dynamics (PIMD), this does however require several ab initio calculations to be performed in each timestep to model the NQEs. In order to make the simulations more computationally affordable, we are currently working to describe the system using a machine learned potential (MLP). This MLP should be transferable across the isotopologues of water, and work across the phase diagram of water. We will here give an update on our ongoing work to improve the MLP description of water, and the results of PIMD simulations using this MLP.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Water is a liquid which structure and properties are highly influenced by nuclear quantum effects (NQEs), as evidenced by the observed differences between light (H
O) and heavy (DO) water. These differences are in theory revealed by conducting so called path integral molecular dynamics (PIMD) simulations. In the past we have used ab initio DFT based potentials to conduct such studies, which come with a large computational cost. We will here use machine learned potentials (MLPs) in place of DFT in order to reduce the computational cost and allow longer and more converged simulations of light and heavy water.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
In this talk, we will discuss our recently proposed self-learning path integral hybrid Monte Carlo with mixed potentials (SL-PIHMC-MIX) and its application to the structure of water and its hydrogen bond patterns at high temperatures and pressures. Briefly, this method allows for the on-the-fly training of a machine learning potential (MLP). The structural energies from propagation using path integral molecular dynamics (PIMD) and the MLP are employed in the hybrid Monte Carlo algorithm, along with ab initio (AI) reference calculations, to ensure that the phase space explored corresponds to that of the underlying AI method. This approach enables us to evaluate the performance of various DFT functionals for studying high-temperature water using an MLP to accelerate sampling, thereby allowing an extensive exploration of the temperature, pressure, and functional space that would be infeasible with AI-PIMD due to computational cost.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Nuclear quantum effects (NQEs) play a crucial role in accurately modeling water, as shown in our studies using ab initio path integral molecular dynamics (AI-PIMD). To overcome the computational cost of AI-PIMD, we developed and applied the self-learning path integral hybrid Monte Carlo (SL-PIHMC-MIX) method, which effectively uses a combination of machine-learned and ab initio potentials to simulate water systems. Our results indicate that while RPBE-D3 performs well at room temperature, SCAN and rev-PBE0-D3 show better performance at higher temperatures.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Thomsen, B.; Shiga, Motoyuki
no journal, ,
We present our work based on self-learning path integral hybrid Monte Carlo to develop machine learned potentials to model the structure and dynamics of water including nuclear quantum effects. Using these methods we are able to model the structure of water using only one tenth count of ab initio calculations, while maintaining the accuracy of the pure ab initio result.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
We present the self-learning path integral hybrid Monte Carlo with mix potentials method (PIHMC-MIX) which is able to on the fly train a machine learned potential (MLP). The trained MLP can be used to accelerate the study of the structure and other equilibrium properties of the system using the PIHMC-MIX method. The PIHMC-MIX is a method which only requires one tenth of the computationally expensive ab initio calculations needed for ab initio path integral molecular dynamics simulations (AI-PIMD), while maintaining the same accuracy as is obtained from the highly accurate AI-PIMD method.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Nuclear quantum effects (NQEs) are crucial in modeling water and its isotopologues due to the low mass of hydrogen and significant differences in mass between hydrogen, deuterium, and tritium. We apply the self-learning hybrid Monte Carlo method with mixed ab initio and machine-learned potentials (SL-PIHMC-MIX) to efficiently simulate sub- and supercritical water, capturing NQEs and testing the accuracy of various DFT functionals. This method significantly reduces computational cost by converging the structure of water with and without NQEs using only a tenth of the calculations needed for traditional AI-PIMD or AI-MD simulations, while accurately reproducing their results.
Thomsen, B.; Shiga, Motoyuki
no journal, ,
Nuclear quantum effects (NQEs) are critical for accurately modeling water, as demonstrated in our prior studies using first-principles path integral molecular dynamics (AI-PIMD). Although effective, AI-PIMD is computationally intensive, limiting its use in long simulations and large systems. To address this, we developed the self-learning path integral hybrid Monte Carlo (SL-PIHMC-MIX) method, which combines first-principles and machine learning potentials to achieve AI-PIMD-level accuracy with significantly fewer calculations. We applied this method to analyze high-temperature water and compared various DFT functionals, finding that rev-PBE0-D3 performed best at elevated temperatures due to its superior handling of non-hydrogen bond interactions.
