Partial-differential-equation (PDE)-constrained optimization is a well-worn technique for acquiring optimal parameters of systems governed by PDEs. However, this approach is limited to providing a single set of optimal parameters per optimization. Given a differentiable PDE solver, if the free parameters are reparameterized as the output of a neural network, that neural network can be trained to learn a map from a probability distribution to the distribution of optimal parameters. This proves useful in the case where there are many well performing local minima for the PDE. We apply this technique to train a neural network that generates optimal parameters that minimize laser-plasma instabilities relevant to laser fusion and show that the neural network generates many well performing and diverse minima.
Evgeny Podryabinkin, Kamil Garifullin, Alexander Shapeev
et al.
Nowadays, academic research relies not only on sharing with the academic community the scientific results obtained by research groups while studying certain phenomena, but also on sharing computer codes developed within the community. In the field of atomistic modeling these were software packages for classical atomistic modeling, later -- quantum-mechanical modeling, and now with the fast growth of the field of machine-learning potentials, the packages implementing such potentials. In this paper we present the MLIP-3 package for constructing moment tensor potentials and performing their active training. This package builds on the MLIP-2 package (Novikov et al. (2020), The MLIP package: moment tensor potentials with MPI and active learning. Machine Learning: Science and Technology, 2(2), 025002.), however with a number of improvements, including active learning on atomic neighborhoods of a possibly large atomistic simulation.
Yi-Kai Kan, Franz X. Kärtner, Sabine Le Borne
et al.
The fast multipole method (FMM) has received growing attention in the beam physics simulation. In this study, we formulate an interpolation-based FMM for the computation of the relativistic space-charge field. Different to the quasi-electrostatic model, our FMM is formulated in the lab-frame and can be applied without the assistance of the Lorentz transformation. In particular, we derive a modified admissibility condition which can effectively control the interpolation error of the proposed FMM. The algorithms and their GPU parallelization are discussed in detail. A package containing serial and GPU-parallelized solvers is implemented in the Julia programming language. The GPU-parallelized solver can reach a speedup of more than a hundred compared to the execution on a single CPU core.
Many physical systems can be studied as collections of particles embedded in space, evolving through deterministic evolution equations. Natural questions arise concerning how to characterize these arrangements - are they ordered or disordered? If they are ordered, how are they ordered and what kinds of defects do they possess? Originally introduced to study problems in pure mathematics, Voronoi tessellations have become a powerful and versatile tool for analyzing countless problems in pure and applied physics. In this paper we explain the basics of Voronoi tessellations and the shapes they produce, and describe how they can be used to study many physical systems.
This paper reviews magnetic flux signal calculations through pick-up loops using vector spherical harmonic expansion under the quasi-static approximation, and presents a near-analytical method of evaluating the flux through arbitrary parametrizable pick-up loops for each expansion degree. This is done by simplifying the surface flux integral (2D) into a line integral (1D). For the special case of tangential circular sensors, we present a fully analytical recursion calculation. We then compare commonly-used cubature approximations to our (near-)analytical forms, and show that cubature approximations suffer from increasing errors for higher spatial frequency components. This suggests the need for more accurate evaluations for increasingly sensitive sensors that are being developed, and our (near-)analytical forms themselves are a solution to this problem.
The integration of the equations of motion of N interacting particles, represents a classical problem in many branches of physics and chemistry. The direct N-body problem is at the heart of simulations studying Coulomb Crystals. We present an hand-optimized code for the latest AVX-512 set of instructions that achieve a single core speed up of $\approx 340\%$ respect the version optimized by the compiler. The increase performance is due a optimization on the organization of the memory access on the inner loop on the Coulomb and, specially, on the usage of an intrinsic function to faster compute the $1/\sqrt{x}$. Our parallelization, which is implemented in OpenMP, achieves an excellent scalability with the number of cores. In total, we achieve $\approx 500GFLOPS$ using a just a standard WorkStation with one Intel Skylake CPU (10 cores). It represents $\approx 75\%$ of the theoretical maximum number of double precision FLOPS corresponding to Fused Multiplication Addition (FMA) operations.
Howard Yanxon, David Zagaceta, Brandon C. Wood
et al.
In this article, we present a systematic study in developing machine learning force fields (MLFF) for crystalline silicon. While the main-stream approach of fitting a MLFF is to use a small and localized training sets from molecular dynamics simulation, it is unlikely to cover the global feature of the potential energy surface. To remedy this issue, we used randomly generated symmetrical crystal structures to train a more general Si-MLFF. Further, we performed substantial benchmarks among different choices of materials descriptors and regression techniques on two different sets of silicon data. Our results show that neural network potential fitting with bispectrum coefficients as the descriptor is a feasible method for obtaining accurate and transferable MLFF.
Oleksandr Koshkarov, Gianmarco Manzini, Gian Luca Delzanno
et al.
We discuss the development, analysis, implementation, and numerical assessment of a spectral method for the numerical simulation of the three-dimensional Vlasov-Maxwell equations. The method is based on a spectral expansion of the velocity space with the asymmetrically weighted Hermite functions. The resulting system of time-dependent nonlinear equations is discretized by the discontinuous Galerkin (DG) method in space and by the method of lines for the time integration using explicit Runge-Kutta integrators. The resulting code, called Spectral Plasma Solver (SPS-DG), is successfully applied to standard plasma physics benchmarks to demonstrate its accuracy, robustness, and parallel scalability.
Continuum kinetic simulations of plasmas, where the distribution function of the species is directly discretized in phase-space, permits fully kinetic simulations without the statistical noise of particle-in-cell methods. Recent advances in numerical algorithms have made continuum kinetic simulations computationally competitive. This work presents the first continuum kinetic description of high-fidelity wall boundary conditions that utilize the readily available particle distribution function. The boundary condition is realized through a reflection function that can capture a wide range of cases from simple specular reflection to more involved first principles models. Examples with detailed discontinuous Galerkin implementation are provided for secondary electron emission using phenomenological and first-principles quantum-mechanical models. Results presented in this work demonstrate the effect of secondary electron emission on a classical plasma sheath.
D. Naranchimeg, L. Khenmedekh, G. Munkhsaikhan
et al.
We have developed the Coulomb wave function discrete variable representation (CWDVR) method to solve the imaginary time dependent Kohn - Sham equation on the many - electronic second row atoms. The imaginary time dependent Kohn - Sham equation is numerically solved using the CWDVR method. We have presented that the results of calculation for second row Li, Be, B, C, N, O and F atoms are in good agreement with other best available values using the Mathematica 7.0 programm.
Stefan Klus, Andreas Bittracher, Ingmar Schuster
et al.
We present a novel machine learning approach to understanding conformation dynamics of biomolecules. The approach combines kernel-based techniques that are popular in the machine learning community with transfer operator theory for analyzing dynamical systems in order to identify conformation dynamics based on molecular dynamics simulation data. We show that many of the prominent methods like Markov State Models, EDMD, and TICA can be regarded as special cases of this approach and that new efficient algorithms can be constructed based on this derivation. The results of these new powerful methods will be illustrated with several examples, in particular the alanine dipeptide and the protein NTL9.
We present Fiend - a simulation package for three-dimensional single-particle time-dependent Schrödinger equation for cylindrically symmetric systems. Fiend has been designed for the simulation of electron dynamics under inhomogeneus vector potentials such as in nanostructures, but it can also be used to study, e.g., nonlinear light-matter interaction in atoms and linear molecules. The light-matter interaction can be included via the minimal coupling principle in its full rigour, beyond the conventional dipole approximation. The underlying spatial discretization is based on the finite element method (FEM), and time-stepping is provided either via the generalized-α or Crank-Nicolson methods. The software is written in Python 3.6, and it utilizes state-of-the-art linear algebra and FEM backends for performance-critical tasks. Fiend comes along with an extensive API documentation, a user guide, simulation examples, and allows for easy installation via Docker or the Python Package Index.
GiovanniMaria Piccini, Dan Mendels, Michele Parrinello
We introduce an extension of a recently published method\cite{Mendels2018} to obtain low-dimensional collective variables for studying multiple states free energy processes in chemical reactions. The only information needed is a collection of simple statistics of the equilibrium properties of the reactants and product states. No information on the reaction mechanism has to be given. The method allows studying a large variety of chemical reactivity problems including multiple reaction pathways, isomerization, stereo- and regiospecificity. We applied the method to two fundamental organic chemical reactions. First we study the \ce{S_N2} nucleophilic substitution reaction of a \ce{Cl} in \ce{CH_2 Cl_2} leading to an understanding of the kinetic origin of the chirality inversion in such processes. Subsequently, we tackle the problem of regioselectivity in the hydrobromination of propene revealing that the nature of empirical observations such as the Markovinikov's rules lies in the chemical kinetics rather than the thermodynamic stability of the products.
We consider the problem of computing first-passage time distributions for reaction processes modelled by master equations. We show that this generally intractable class of problems is equivalent to a sequential Bayesian inference problem for an auxiliary observation process. The solution can be approximated efficiently by solving a closed set of coupled ordinary differential equations (for the low-order moments of the process) whose size scales with the number of species. We apply it to an epidemic model and a trimerisation process, and show good agreement with stochastic simulations.
Homogenization of a thin micro-structure yields effective jump conditions that incorporate the geometrical features of the scatterers. These jump conditions apply across a thin but nonzero thickness interface whose interior is disregarded. This paper aims (i) to propose a numerical method able to handle the jump conditions in order to simulate the homogenized problem in the time domain, (ii) to inspect the validity of the homogenized problem when compared to the real one. For this purpose, we adapt an immersed interface method originally developed for standard jump conditions across a zero-thickness interface. Doing so allows us to handle arbitrary-shaped interfaces on a Cartesian grid with the same efficiency and accuracy of the numerical scheme than those obtained in an homogeneous medium. Numerical experiments are performed to test the properties of the numerical method and to inspect the validity of the homogenization problem.
We revisit the image charge method for the Green's function problem of the Poisson-Boltzmann equation for a dielectric sphere immersed in ionic solutions. Using finite Mellin transformation, we represent the reaction potential due to a source charge inside the sphere in terms of one dimensional distribution of image charges. The image charges are generically composed of a point image at the Kelvin point and a line image extending from the Kelvin point to infinity with an oscillatory line charge strength. We further develop an efficient and accurate algorithm for discretization of the line image using Padé approximation and finite fraction expansion. Finally we illustrate the power of our method by applying it in a multiscale reaction-field Monte Carlo simulation of monovalent electrolytes.
The recent results of applying the parallel numerical code SELFAS-3 to modelling of electrodynamic aggregation of magnetized nanodust are presented. The modelling describes evolution of a many-body system of basic blocks which are taken as strongly magnetized thin rods (i.e., one-dimensional static magnetic dipoles), with electric conductivity and static electric charge, screened with its own plasma sheath. The code provides continuous modelling of the following stages of evolution: (i) alignments of randomly situated solitary basic blocks in an external magnetic field and formation of stable filaments, (ii) percolation of electric conductivity in a random filamentary system, and electric short-circuiting in the presence of an external electric field, (iii) evolution of electric current profile in a filamentary network with a trend towards a fractal skeletal structuring.
The "weighted ensemble" method, introduced by Huber and Kim, [G. A. Huber and S. Kim, Biophys. J. 70, 97 (1996)], is one of a handful of rigorous approaches to path sampling of rare events. Expanding earlier discussions, we show that the technique is statistically exact for a wide class of Markovian and non-Markovian dynamics. The derivation is based on standard path-integral (path probability) ideas, but recasts the weighted-ensemble approach as simple "resampling" in path space. Similar reasoning indicates that arbitrary nonstatic binning procedures, which merely guide the resampling process, are also valid. Numerical examples confirm the claims, including the use of bins which can adaptively find the target state in a simple model.