Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userAccelerating Atomistic Simulations with Piecewise Machine Learned Ab Initio Potentials at Classical Force Field-like Cost
98
0
0.0
(
0
)
Ask ChatGPT about the research
No Arabic abstract
Machine learning methods have nowadays become easy-to-use tools for constructing high-dimensional interatomic potentials with ab initio accuracy. Although machine learned interatomic potentials are generally orders of magnitude faster than first-principles calculations, they remain much slower than classical force fields, at the price of using more complex structural descriptors. To bridge this efficiency gap, we propose an embedded atom neural network approach with simple piecewise switching function based descriptors, resulting in a favorable linear scaling with the number of neighbor atoms. Numerical examples validate that this piecewise machine learning model can be over an order of magnitude faster than various popular machine learned potentials with comparable accuracy for both metallic and covalent materials, approaching the speed of the fastest embedded atom method (i.e. several {mu}s/atom per CPU core). The extreme efficiency of this approach promises its potential in first-principles atomistic simulations of very large systems and/or in long timescale.
rate research
Read More
We develop a neuroevolution-potential (NEP) framework for generating neural network based machine-learning potentials. They are trained using an evolutionary strategy for performing large-scale molecular dynamics (MD) simulations. A descriptor of the atomic environment is constructed based on Chebyshev and Legendre polynomials. The method is implemented in graphic processing units within the open-source GPUMD package, which can attain a computational speed over $10^7$ atom-step per second using one Nvidia Tesla V100. Furthermore, per-atom heat current is available in NEP, which paves the way for efficient and accurate MD simulations of heat transport in materials with strong phonon anharmonicity or spatial disorder, which usually cannot be accurately treated either with traditional empirical potentials or with perturbative methods.
We present an implicit solvent model for ab initio electronic structure calculations which is fully self-consistent and is based on direct solution of the nonhomogeneous Poisson equation. The solute cavity is naturally defined in terms of an isosurface of the electronic density according to the formula of Fattebert and Gygi (J. Comp. Chem. 23, 6 (2002)). While this model depends on only two parameters, we demonstrate that by using appropriate boundary conditions and dispersion-repulsion contributions, solvation energies obtained for an extensive test set including neutral and charged molecules show dramatic improvement compared to existing models. Our approach is implemented in, but not restricted to, a linear-scaling density functional theory (DFT) framework, opening the path for self-consistent implicit solvent DFT calculations on systems of unprecedented size, which we demonstrate with calculations on a 2615-atom protein-ligand complex.
The electronic structure in matter under extreme conditions is a challenging complex system prevalent in astrophysical objects and highly relevant for technological applications. We show how machine-learning surrogates in terms of neural networks have a profound impact on the efficient modeling of matter under extreme conditions. We demonstrate the utility of a surrogate model that is trained on emph{ab initio} quantum Monte Carlo data for various applications in the emerging field of warm dense matter research.
The accurate prediction of singlet and triplet excitation energies is of significant fundamental interest and is critical for many applications. An area of intense research, most calculations of singlet and triplet energies use time-dependent density functional theory (TDDFT) in conjunction with an approximate exchange-correlation functional. In this work, we examine and critically assess an alternative method for predicting low-lying neutral excitations with similar computational cost, the ab initio Bethe-Salpeter equation (BSE) approach, and compare results against high-accuracy wavefunction-based methods. We consider singlet and triplet excitations of 27 prototypical organic molecules, including members of Thiels set, the acene series, and several aromatic hydrocarbons exhibiting charge-transfer-like excitations. Analogous to its impact in TDDFT, we find that the Tamm-Dancoff approximation (TDA) overcomes triplet instabilities in the BSE approach, improving both triplet and singlet energetics relatively to higher level theories. Finally, we find that BSE-TDA calculations built on good DFT starting points, such as those utilizing optimally-tuned range-separated hybrid functionals, can yield accurate singlet and triplet excitation energies for gas-phase organic molecules.
The intercombination $a^3Pi - X^1Sigma^+$ Cameron system of carbon monoxide has been computationally studied in the framework of multi-reference Fock space coupled cluster method with the use of generalized relativistic pseudopotential model for the effective introducing the relativity in all-electron correlation treatment. The extremely weak $a^3Pi_{Omega=0^+,1} - X^1Sigma^+$ transition probabilities and radiative lifetimes of the metastable $a^3Pi$ state were calculated and compared with their previous theoretical and experimental counterparts. The impact of a presumable variation of the fine structure constant $alpha=e^2/hbar c$ on transition strength of the Cameron system has been numerically evaluated as well.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
