Large Simulation System Research Articles

The dipole moment of supercritical water – local vs. mean-field polarisation contributions

Supercritical water has attracted much attention from both fundamental and technological perspectives, based largely on its ability to solvate other molecules. Predicting and controlling this requires a deeper understanding of water's polarisation behaviour. Using the computationally efficient Self-Consistent Electrostatic Embedding method, we were able to calculate the water dipole moment over an unprecedented range of thermodynamic conditions, covering gas, liquid and supercritical states, with large simulation systems and a high-level quantum mechanical method. We find a discontinuous change in the dipole moment along subcritical isotherms, corresponding to the sharp transition between the vapour and liquid states, with the latter exhibiting induced dipole moments between 0.5 and 0.9 D, depending on the temperature. In contrast, the dipole moment changes continuously from gas-like to liquid-like behaviour in the supercritical regime, allowing the degree of polarisation to be controlled through manipulating temperature and pressure. The dipole moment was found to be linearly related to the average number of hydrogen-bonded neighbours of water, emphasising the key role of local interactions to the polarisation process. Mean-field approaches based on a dielectric continuum representation of the solvent are unable to predict this behaviour due to the neglect of local interactions.

GENESIS CGDYN: large-scale coarse-grained MD simulation with dynamic load balancing for heterogeneous biomolecular systems.

Residue-level coarse-grained (CG) molecular dynamics (MD) simulation is widely used to investigate slow biological processes that involve multiple proteins, nucleic acids, and their complexes. Biomolecules in a large simulation system are distributed non-uniformly, limiting computational efficiency with conventional methods. Here, we develop a hierarchical domain decomposition scheme with dynamic load balancing for heterogeneous biomolecular systems to keep computational efficiency even after drastic changes in particle distribution. These schemes are applied to the dynamics of intrinsically disordered protein (IDP) droplets. During the fusion of two droplets, we find that the changes in droplet shape correlate with the mixing of IDP chains. Additionally, we simulate large systems with multiple IDP droplets, achieving simulation sizes comparable to those observed in microscopy. In our MD simulations, we directly observe Ostwald ripening, a phenomenon where small droplets dissolve and their molecules redeposit into larger droplets. These methods have been implemented in CGDYN of the GENESIS software, offering a tool for investigating mesoscopic biological processes using the residue-level CG models.

New findings related to carbothermal reduction of polysiloxane-derived ceramics

Carbothermal reduction is the primary mechanism for SiC formation in polymer derived ceramics. However, the specific process involved in the SiOC conversion to SiC and the role of oxygen and carbon are unknown. Using ReaxFF with LAMMPS in a large simulation system containing more than 93 K atoms, this study for the first time shows that O must be removed from SiO2 clusters for SiC formation, providing insights into the carbothermal reduction mechanism. SiO2 regions need to be of a large enough size to induce O reduction. C cluster growth is inversely related to removal of O. This work shows not only the importance of the simulation system size in understanding atomic interactions but also absence of phase separation of SiOC to SiO2. Instead, SiO2 converts to SiC through C diffusion and O removal.

A Cuda Fast Multipole Method with Highly Efficient M2L Far Field Evaluation

Solving an N-body problem is a computationally quite demanding task in many scientific fields ranging from astrophysics to biomolecular simulations. However, the direct solution scales with O(N2); hence, even on modern hardware, the direct calculation becomes impractical even for moderate numbers of particles, and thus, efficient yet accurate approximations are key. For biomolecular simulations, a widely used such method is Particle Mesh Ewald (PME), which scales with O(N log N). Although extremely fast on a single node, PME runs into communication bottlenecks when parallelized for large simulation systems on many nodes. The fast multipole method (FMM) offers an attractive alternative: It requires less communication and reduces the complexity to optimal O(N). The method approximates long-range interactions by grouping the particles into clusters represented as multipoles. The cluster size grows with the interaction distance according to the underlying octree structure. Hence, further separated particles require fewer interaction computations, hence less communication. Here, we present our full NVIDIA CUDA FMM implementation, which has been optimized for the electrostatic interactions described by Coulomb's law relevant to molecular dynamics simulations. We compare different parallelization approaches to the computationally limiting part of the algorithm, the Multipole-to-Local (M2L) operator, and discuss their performance bottlenecks. The first approach can be implemented with only minimal modifications to the sequential CPU implementation. It features the Unified Memory concept, which allows for a simple utilization of the existing CPU data structures. The second approach enhances the achieved performance by exploiting CUDA Dynamic Parallelism. It introduces a significant speedup, especially for a high accuracy requirement. The third parallelization approach abstracts the underlying octree with precomputed interaction lists and it exploits operator symmetries to achieve nearly optimal performance in the whole tested accuracy range.

Computationally Efficient Modeling of DC-DC Converters for PV Applications

In this work, a computationally efficient approach for the simulation of a DC-DC converter connected to a photovoltaic device is proposed. The methodology is based on a combination of a highly efficient formulation of the one-diode model for photovoltaic (PV) devices and a state-space formulation of the converter as well as an accurate steady-state detection methodology. The approach was experimentally validated to assess its accuracy. The model is accurate both in its dynamic response (tested in full linearity and with a simulated PV device as the input) and in its steady-state response (tested with an outdoor experimental measurement setup). The model detects automatically the reaching of a steady state, thus resulting in lowered computational costs. The approach is presented as a mathematical model that can be efficiently included in a large simulation system or statistical analysis.

GPU-based acceleration of Monte Carlo simulations for migration-coalescence evolution of gas bubbles in materials

Monte Carlo (MC) simulations based on the migration-coalescence model are important tools for studying the evolution of gas bubbles in materials, a subject of concern across numerous fields. The acceleration of MC simulations is always desired especially when a large simulation system is required to extract highly accurate statistics. The present work focuses on the algorithmic study. A graphics processing unit (GPU)-based implementation of an MC simulation of the migration-coalescence evolution of bubbles is described. The acceleration of simulations involves two factors. In addition to acceleration due to the parallel computing features of GPUs, algorithmic strategies are proposed to improve performance. Considering that bubbles have a finite radius and grow with time, this paper proposes generating a nearest-neighbor-list (NNL) instead of a cutoff-radius-neighbor-list. An evaluation demonstrated that an NNL strategy can improve computational efficiency and, more importantly, conserve valuable GPU memory. This paper also proposes a strategy using many small boxes in one run instead of one large box. An evaluation of this strategy shows that significant promotion of the computational efficiency for generating neighbor-lists can be achieved. Regarding the migration and coalescence process, this paper proposes a method in which a merging table is generated and CUDA atomic operations are applied to bubbles in a bi-directional procedure that handles coalescence events. This method avoids asynchronous reading and writing conflicts that may occur in the GPU implementation of the migration and coalescence process without losing numerical accuracy. The algorithms and implementations presented in this paper would be instructive to the development of simulation codes, for an example the code of object kinetic Monte Carlo, in which the defects have finite sizes.

Time sequence of energetic particle spectra in quasiparallel shocks in large simulation systems

We have performed collisionless shock simulations using a one-dimensional hybrid particle-in-cell method to investigate the energy spectra of the differential intensity around the quasiparallel shocks. The system size is sufficiently large (200 000 ion inertia length) in order to eliminate the unphysical effect caused by the upstream boundary. The obtained spectrum of the differential intensity have the shape of the power-law with exponentially falling off in higher energy as predicted in previous simulations, however, the power-law indices and e-folding energy do not depend on the shock parameters, the shock Mach number MA (7.1–11.7) and the shock angle θBN (10°–40°). The power-law index is ∼1.0. This number is close to the prediction by the standard diffusive shock acceleration theory. However, the reason for the agreements between the values is not clear because an additional acceleration process is also observed in the present runs; this process was reported in [Sugiyama and Terasawa, Adv. Space Res. 24, 73 (1999)] and [Sugiyama et al., J. Geophys. Res. 106, 21657 (2001)]. The e-folding energy linearly increases in time for all runs. One of the reasons for these independent profiles on shock parameters is that the pitch angle distribution in the upstream region shows similar profiles in each run.