Supercomputing Frontiers and Innovations

AlFaMove: Scalable Implementation of Surface Movement Method for Cluster Computing Systems

Nikolay A. Olkhovsky — 2024-10-25

The article presents a numerical implementation of the surface movement method for linear programming. The base of this implementation is the new AlFaMove algorithm, which builds on the surface of a feasible polytope an optimal objective path from an arbitrary boundary point to a point that is a solution to a linear programming problem. The optimal objective path is a path along the faces of the feasible polytope in the direction of maximizing the value of the objective function. To calculate the optimal movement direction, the pseudoprojection operation on a linear manifold is used. The pseudoprojection operation is a generalization of the orthogonal projection and is implemented using an iterative projection-type algorithm. The proposition is proved that, for a linear manifold that is the intersection of hyperplanes, the pseudoprojection coincides with the orthogonal projection. It is also proved that, in the case of a linear manifold, pseudoprojection makes it possible to calculate the movement vector in the direction of maximum increase of the objective function. A parallel implementation of the AlFaMove algorithm is described. The results of computational experiments on a cluster computing system are presented to demonstrate the high scalability of the proposed numerical implementation.

Study of the Effectiveness of Parallel Algorithms for Modeling the Dynamics of Collisionless Galactic Systems on GPUs

Sergei S. Khrapov — 2024-10-25

N-body model is a common research tool in galaxy physics and cosmology. The transition to the use of computing systems with GPUs can significantly improve the performance and quality of simulation results for gravitational systems. N-body – Particle-Particle algorithm is presented on a hybrid computing platform CPU + multi-GPUs. Using a direct method of calculating gravitational forces by summing the interactions of each particle with each other is resource-intensive, but provides the best accuracy in modeling dynamics at all scales. The main result is an analysis of the efficiency of parallel code depending on the number of GPUs and the choice of single and double precision floating-point arithmetics. The laws of conservation of energy, momentum and angular momentum are tested for a series of models, including major mergers of galaxies and the evolution of galactic stellar disc subject to the most severe gravitational instability. The general conclusion is that conservation laws are poorly implemented when using 4-byte numbers due to the accumulation of arithmetic errors. Calculations with 8-byte numbers ensure that the laws of conservation of momentum and angular momentum are satisfied to the limit of arithmetic accuracy without accumulating errors. The law of conservation of energy is determined primarily by the order of the numerical scheme for integrating the equations of motion. The additional reduction in the error of the conservation law of total energy due to the transition from 4-byte to 8-byte numbers is 1–2 orders of magnitude. Increasing the number of GPUs used helps improve the implementation of conservation laws due to a decrease in the number of particles per graphics processing unit.

Investigation of the Capability of Restoring Information on the Primary Particle from Cherenkov Light Generated by Extensive Air Showers Using the Lomonosov-2 Supercomputer

Elena A. Bonvech — 2024-10-25

The new SPHERE-3 detector is under development. Its main objectives are the primary cosmic ray spectrum and chemical composition studies in the 1–1000 PeV energy range. The detector will register both reflected and direct Cherenkov light from extensive air showers. The goal of the new approach is high precision of event-by-event estimation of the primary particle parameters, especially its mass. The reflected Cherenkov light registration technique used in earlier experiments has good energy sensitivity and some mass estimation capability. Addition of direct Cherenkov light registration will allow to further advance the detector mass sensitivity. Several approaches to direct Cherenkov light registration are considered: by the main detector camera and by a dedicated direct light detector. First tests of the proposed methods are presented both for reflected and direct Cherenkov light. The detector design is tested on a large database of simulated showers. The simulation pipeline and related challenges to it are described. Also, progress in parallelization of the CORSIKA code for Cherenkov light simulations is presented.

Quantum-Chemical Study of Some Trispyrazolobenzenes and Trispyrazolo-1,3,5-triazines

Vadim M. Volokhov — 2024-10-25

Development of new high-energy density materials and study of their properties is an important task, since such materials are in high demand in various application areas. This paper continues the study of polynitrogen fused tetracyclic systems which include three azole rings annelated with a benzene of azine ring. Such polycyclic structures attract special attention of scientists. This paper is dedicated to the study of properties of a number of promising high-energy tetracyclic compounds annelated with pyrazole nitro derivatives. For this study, we used quantum-chemical methods (the hybrid density functional B3LYP and the composite G4MP2 and G4 methods) within the Gaussian 09 and NWChem software packages at Lomonosov Moscow State University Supercomputer Complex. We used the atomization method and method of reactions to calculate the enthalpy of formation. We analyzed the dependence of the enthalpy of formation on the structural parameters of the compounds and calculated the optimized structures and IR absorption spectra. We also compare the Gaussian 09 and NWChem quantum chemical programs in terms of efficiency, parallelization and computational requirements. In the cases where the G4-level accuracy of the results is not required, the use of NWChem can significantly save computation time.

Wing Noise Simulation of Supersonic Business Jet in Landing Configuration

Alexey P. Duben — 2024-10-25

The paper presents the results of wing noise simulations for the prototype of supersonic business jet in landing mode. The near-field airflow is modeled according to Delayed Detached Eddy Simulation approach. The finite-volume vertex-centered scheme with the low weight of upwind component is used for convective flux approximation. The noise at the far-field points is calculated by the Ffowcs Williams–Hawkings method. The noise spectra at the near-field points are presented, and the impact of local mesh resolution and numerical instability on the near-field acoustics is discussed. For the Ffowcs Williams–Hawkings method due to the features of the wing geometry and the resulting flow configuration, we used non-standard integration surfaces to reduce computational costs of the scale-resolving simulations. Additionally, we employed optimized mesh resolution on the integration surfaces to significantly reduce the dick space required for storing the data for far-field noise calculations. The tests performed for the near-field and far-field points demonstrated applicability of the proposed optimizations.

Tool and Algorithm for the Determination of Aptamers in Nanopore Sequencing Data: AptaLong

Maria A. Grigoryeva — 2024-10-25

Nanopore sequencing is a third generation sequencing technology that allows direct, real-time sequencing of individual DNA or RNA molecules. It utilizes a nanopore – an extremely small pore – in a membrane to pass a single strand DNA or RNA. As the sequence passes through the nanopore, changes in electrical current are detected and used to determine the nucleotide sequence. Nanopore sequencing has several advantages. It offers long read lengths, allowing for the sequencing of difficult regions of the genome, such as repetitive regions. It also enables real-time sequencing, providing immediate data generation without the need for extensive library preparation. Many bioinformatics pipelines and tools have been developed specifically for nanopore sequencing data analysis, addressing the unique characteristics and challenges of this technology, while dealing with non-standard long reads, derived from the ligation process of shorter oligonucleotides, might be challenging. In this research we present a new algorithm that extracts an aptamer sequence from the results of nanopore sequencing of several SELEX experiments with single-stranded DNA. The algorithm is based on statistical methods, based on known primer sequences and length of searching aptamer. We used step-by-step displacement of the reference sequence with positional alignment and calculated the positional frequencies of each nucleotide. As a result, the nucleotide frequencies obtained at each step are averaged, and thus, we find the sequence that is more likely to represent the aptamer.

Modeling Microtubule Dynamics on Lomonosov-2 Supercomputer of Moscow State University: from Atomistic to Cellular Scale Simulations

Nikita B. Gudimchuk — 2024-10-25

Cytoskeletal polymers of tubulin, the microtubules, are critically important for cellular physiology. Their remarkable non-equilibrium dynamics and unusual mechanical properties have nurtured interest in exploring microtubules with diverse experimental methods and modeling their properties at different scales. In this work, we overview the studies of microtubules from the atomistic level of detail to the cellular dimension, focusing on the computational modeling work that has been carried out by our group on Lomonosov-2 supercomputer of Moscow State University since 2015. Our computational efforts have been aimed at understanding of microtubules through a set of models at multiple spatial and temporal scales, starting from examining the properties of tubulin dimers, as the building blocks, and further elucidating how those properties enable more complex assembly/disassembly and force-generation behaviors of microtubules, emerging at larger scales. Our methodology includes different approaches, from atomistic molecular dynamics to more coarse-grained techniques, such as Brownian dynamics and Monte Carlo simulations. We describe the motivation and the context for each model, overview the major conclusions from the simulations, which we believe were instrumental in building an integrative understanding of these polymers. We also discuss some technical aspects of the modeling, such as the computational performance of different types of simulations, current limitations and potential future directions for description of the microtubule dynamics, using the multi-scale approach.

Numerical Analysis of OECD/NEA HYMERES Project Benchmark Tests Using CABARET-SC1 CFD Code

Anton A. Kanaev — 2024-10-25

The benchmark tests carried out within the OECD/NEA HYMERES (HYdrogen Mitigation Experiments for Reactor Safety) international project allowed to assess the capability of computational tools and to develop methodology for improving the modelling of complex safety issues relevant for the analysis and mitigation of a severe accident leading to hydrogen release into a nuclear plant containment. The paper presents the results of numerical simulation of two OECD/NEA HYMERES benchmark tests using CABARET-SC1 code. The code is based on the eddy resolving CABARET technique, which allows implicit modeling of the subgrid turbulence scales without using tuning parameters (ILES approximation). The absence of tuning parameters in the numerical approach allowed evaluating the influence of a separate physical phenomenon of radiative heat transfer. The influence of the mesh resolution in flow regions with complex geometries and the use of a porous medium model was also investigated.