pith. sign in

arxiv: 2606.02319 · v1 · pith:QBJZBMTJnew · submitted 2026-06-01 · 💻 cs.DC

Strategies for Molecular Dynamics using Hybrid Systems: LAMMPS Use Case

Paulo Henrique Leme Ramalho , Dennis Alves Pedersen , F\'abio Andrijauskas This is my paper

Pith reviewed 2026-06-28 12:37 UTC · model grok-4.3

classification 💻 cs.DC
keywords LAMMPSmolecular dynamicshybrid parallelizationMPI OpenMPscalability analysisbiomolecular simulationcoarse-grained modelingHPC performance
0
0 comments X

The pith

Hybrid MPI+OpenMP configurations in LAMMPS scale better than pure MPI for large-scale biomolecular simulations by reducing communication overhead.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper evaluates LAMMPS performance for coarse-grained simulations of the Tritrpticin peptide on up to 1024 cores. Pure MPI runs perform well on single nodes but lose efficiency in multi-node setups due to rising communication and synchronization costs. Hybrid MPI+OpenMP runs maintain better efficiency at large scales by cutting those costs and making fuller use of the NUMA memory layout. The breakdown shows communication and electrostatic calculations take the most time in large pure-MPI cases. This highlights how parallel granularity and communication balance determine performance in biomolecular HPC work.

Core claim

For the LAMMPS simulator applied to coarse-grained biomolecular dynamics of the Tritrpticin peptide, hybrid MPI+OpenMP parallelization delivers superior scalability on multi-node HPC systems compared to pure MPI because it lowers inter-process communication volume and better matches the NUMA memory hierarchy, leading to lower execution times at scales up to 1024 cores.

What carries the argument

The hybrid MPI+OpenMP execution model within LAMMPS, which combines message passing across nodes with shared-memory threading within nodes for the spatial domain decomposition.

If this is right

  • Pure MPI executions show scalability degradation beyond single-node due to communication overhead.
  • Hybrid configurations reduce the fraction of time spent in communication and electrostatic routines at large scales.
  • Performance in these simulations depends on balancing parallelization granularity, spatial decomposition, and distributed communication costs.
  • Hybrid MPI+OpenMP offers a more sustainable path for running coarse-grained biomolecular simulations on many-core HPC architectures.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar hybrid strategies could be tested in other molecular dynamics packages to see if the scalability benefit generalizes.
  • The results on one peptide suggest checking whether the advantage holds for larger or more complex biomolecular systems.
  • Future work might explore optimal thread-to-process ratios for different hardware NUMA configurations.

Load-bearing premise

The measured performance edge of hybrid over pure MPI on the Tritrpticin workload and the specific testbed hardware will apply to other biomolecular systems and different HPC machines.

What would settle it

A follow-up run of the same LAMMPS tests on a different peptide or protein system using the same hybrid versus pure MPI comparison at 1024 cores, checking whether the hybrid still shows lower execution time.

Figures

Figures reproduced from arXiv: 2606.02319 by Dennis Alves Pedersen, F\'abio Andrijauskas, Paulo Henrique Leme Ramalho.

Figure 1
Figure 1. Figure 1: Coarse-grained biomolecular system based on the Tritrpticin peptide (1D6X) inserted into a hydrated DOPC lipid bilayer using the SPICA force field. 3.3. Simulation parameters Simulations were conducted in the isothermal–isobaric (NPT) ensemble, with a Nose–Hoover thermostat and an anisotropic barostat compatible with coarse-grained lipid membrane systems. The lj/spica/coul/long potential used a 15 Å cutoff… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of mean execution time (loop time) between pure MPI, hybrid MPI+OpenMP, and pure OpenMP configurations across different multi-node scales [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: complements this analysis by presenting a heatmap of the execution time as a function of the number of nodes and the number of OpenMP threads per MPI rank. The existence of optimal regions of hybrid parallelization is clearly observed, particularly in the multi-node configurations with 4 and 8 OpenMP threads per MPI process. 1 2 4 8 16 128 OpenMP threads per MPI rank 1 2 4 8 Compute nodes 4.4 4.6 5.8 8.0 1… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of speedup between pure MPI and hybrid MPI+OpenMP executions at different scales of distributed parallelization. The results show that the hybrid MPI+OpenMP model presented consistently superior performance compared to pure MPI in multi-node environments. While the pure MPI executions suffered progressive degradation of speedup with the increase in the number of nodes, the hybrid configurations … view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of parallel efficiency between pure MPI and hybrid MPI+OpenMP strategies on multi-node architectures. The efficiency results evidence a strong departure from ideal linear scaling in both parallelization models, an expected behavior for small- and medium-sized coarse-grained biomolecular workloads executed across a large number of distributed cores. Even so, the hybrid configurations maintained e… view at source ↗
Figure 6
Figure 6. Figure 6: Percentage distribution of execution time among the main internal routines of LAMMPS for different parallelization strategies. The results show substantial differences between the hybrid and pure MPI executions. In the hybrid configurations considered most efficient (32 × 8, 64 × 8, and 128 × 8), a relatively balanced distribution is observed between force calculation (Pair and Kspace) and communication (C… view at source ↗
Figure 7
Figure 7. Figure 7: Growth of the relative cost of distributed communication (Kspace + Comm) with the increase in parallel scale. It is observed that the pure MPI executions reached approximately 90% of the total time concentrated in communication-related routines at the largest scales analyzed. In contrast, the hybrid configurations maintained significantly lower values in several scenarios, particularly in the 32 × 8 and 12… view at source ↗
Figure 8
Figure 8. Figure 8: Trade-off between distributed communication overhead and time available for force computation. Each marker is one experimental configuration averaged across three seeds; marker size is proportional to the total number of compute cores. The dashed line indicates the geometric constraint that Kspace + Comm + useful-compute fractions cannot exceed 100% of loop time. This behavior reinforces the central hypoth… view at source ↗
Figure 9
Figure 9. Figure 9: Percentage variability of the execution time across three independent seeds for each experimental configuration. Most configurations presented variability below 10% across seeds, reinforcing the stability of the measurements obtained and the thermodynamic equivalence between different parallelization strategies. The main exception was the 1024 × 1 configuration, whose variability exceeded 50% — consistent … view at source ↗
read the original abstract

The complexity of biomolecular simulations has substantially increased the demand for High-Performance Computing (HPC) infrastructures, particularly in molecular dynamics and coarse-grained modeling. This work presents a systematic performance and scalability analysis of the LAMMPS simulator for coarse-grained biomolecular simulations, using the antimicrobial peptide Tritrpticin (PDB ID: 1D6X) as the experimental workload. Pure MPI and hybrid MPI+OpenMP executions were evaluated in HPC environments comprising up to 8 compute nodes and 1024 simultaneous cores. Metrics of execution time, speedup, parallel efficiency, statistical variability, and internal time decomposition were investigated. Results showed that pure MPI executions deliver excellent performance in single-node environments but suffer scalability degradation in multi-node executions due to communication overhead and inter-process synchronization. Hybrid MPI+OpenMP configurations proved more efficient at large scale, reducing communication costs and better exploiting the NUMA memory hierarchy. The computational breakdown revealed that communication and electrostatic interaction routines accounted for the largest fraction of execution time at the largest pure-MPI scales. These results reinforce that performance of biomolecular HPC applications depends directly on the balance among parallelization granularity, spatial decomposition, and distributed communication costs. Hybrid MPI+OpenMP strategies represent a more sustainable alternative for coarse-grained biomolecular simulations on modern many-core architectures.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript reports a performance and scalability study of LAMMPS for coarse-grained biomolecular MD simulations using the Tritrpticin peptide (PDB 1D6X) as the sole workload. Pure MPI and hybrid MPI+OpenMP runs are compared on an HPC testbed with up to 8 nodes and 1024 cores, using metrics of execution time, speedup, parallel efficiency, statistical variability, and internal time decomposition. The central claim is that pure MPI scales well within a node but degrades at multi-node scales due to communication overhead, while hybrid MPI+OpenMP is superior at large scale by lowering communication costs and better exploiting the NUMA hierarchy, making hybrid strategies a more sustainable choice for such simulations.

Significance. If the trends hold beyond the reported case, the work supplies concrete empirical guidance on parallelization granularity for biomolecular HPC codes, with the internal time breakdown (highlighting communication and electrostatics as dominant at scale) being a useful diagnostic contribution. The systematic comparison of pure vs. hybrid on a real application is a strength for practitioners, though the single-workload design constrains broader applicability to other coarse-grained systems or hardware.

major comments (2)
  1. [Abstract] Abstract: the headline claim that hybrid MPI+OpenMP 'proved more efficient at large scale' and represents 'a more sustainable alternative for coarse-grained biomolecular simulations' is derived exclusively from the 1D6X peptide; no results are shown for other system sizes, force fields, or interaction profiles that would be needed to establish that the observed crossover in communication-to-compute ratio is representative rather than workload-specific.
  2. [Abstract] Abstract: although 'statistical variability' is listed among the investigated metrics, the text supplies no error bars, standard deviations, or hypothesis tests on the timing data, leaving the reported trends in execution time and efficiency without quantified uncertainty and making it impossible to judge whether observed differences exceed measurement noise.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claim that hybrid MPI+OpenMP 'proved more efficient at large scale' and represents 'a more sustainable alternative for coarse-grained biomolecular simulations' is derived exclusively from the 1D6X peptide; no results are shown for other system sizes, force fields, or interaction profiles that would be needed to establish that the observed crossover in communication-to-compute ratio is representative rather than workload-specific.

    Authors: We agree that the empirical results are derived from a single workload (the 1D6X Tritrpticin peptide). This system was chosen as a representative coarse-grained biomolecular case with typical interaction profiles, allowing detailed internal time decomposition. The observed communication overhead trends are expected to generalize to similar systems, but we accept that broader validation across system sizes and force fields would strengthen the claim. In revision we will qualify the abstract to state that hybrid MPI+OpenMP proved more efficient at large scale for this workload and add a note in the conclusions on the value of future multi-workload studies. revision: yes

  2. Referee: [Abstract] Abstract: although 'statistical variability' is listed among the investigated metrics, the text supplies no error bars, standard deviations, or hypothesis tests on the timing data, leaving the reported trends in execution time and efficiency without quantified uncertainty and making it impossible to judge whether observed differences exceed measurement noise.

    Authors: The manuscript does list statistical variability as a metric and performed repeated runs to assess it, yet the reported figures and text omit explicit error bars or standard deviations. This is a reporting omission that prevents readers from evaluating whether differences exceed noise. We will revise the manuscript to include error bars (standard deviation across runs) on all timing and efficiency plots and add a brief discussion of the observed variability. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical timing measurements on one workload with no derivations or self-referential equations

full rationale

The paper reports benchmark timings, speedups, efficiency, and time breakdowns for LAMMPS runs of the 1D6X peptide under pure-MPI vs. hybrid MPI+OpenMP on a fixed testbed. No equations, fitted parameters, predictions derived from inputs, or self-citation chains appear in the abstract or described methodology. All claims rest on direct measurement rather than any reduction to prior results or definitions within the paper. This matches the reader's assessment of score 0.0 and contains none of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Ledger populated from abstract only; the central claim rests on the representativeness of one workload and test environment.

axioms (1)
  • domain assumption The Tritrpticin coarse-grained simulation and the tested HPC nodes are representative of typical biomolecular workloads and modern many-core systems.
    All scaling conclusions are drawn from this single case; if the workload is atypical the hybrid advantage may not hold.

pith-pipeline@v0.9.1-grok · 5768 in / 1181 out tokens · 32197 ms · 2026-06-28T12:37:23.540895+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. How far does a random forest generalize from a 54-run LAMMPS+SPICA benchmark?

    cs.DC 2026-06 unverdicted novelty 3.0

    A random forest surrogate trained on 54 LAMMPS runs achieves 4% relative error on loop time and ranks configurations correctly within hardware regimes but degrades across architectural boundaries.

Reference graph

Works this paper leans on

34 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    T.Liu,D.Lu,H.Zhang,M.Zheng,H.Yang,Y.Xu,C.Luo,W.Zhu,K.Yu,H.Jiang,Applyinghigh-performancecomputingindrugdiscovery and molecular simulation, National Science Review 3 (1) (2016) 49–63.doi:10.1093/nsr/nww003

    work page doi:10.1093/nsr/nww003 2016

  2. [2]

    S. A. Hollingsworth, R. O. Dror, Molecular Dynamics Simulation for All, Neuron 99 (6) (2018) 1129–1143.doi:10.1016/j.neuron. 2018.08.011

    work page doi:10.1016/j.neuron 2018

  3. [3]

    B. Acun, D. J. Hardy, L. V. Kale, K. Li, J. C. Phillips, J. E. Stone, Scalable molecular dynamics with NAMD on the Summit system, IBM Journal of Research and Development 62 (6) (2018) 4:1–4:9.doi:10.1147/jrd.2018.2888986

    work page doi:10.1147/jrd.2018.2888986 2018

  4. [4]

    Kutzner, V

    C. Kutzner, V. Miletić, K. Palacio Rodríguez, M. Rampp, G. Hummer, B. L. de Groot, H. Grubmüller, Scaling of the GROMACS Molecular Dynamics Code to 65k CPU Cores on an HPC Cluster, Journal of Computational Chemistry 46 (5) (Feb. 2025).doi:10.1002/jcc.70059

    work page doi:10.1002/jcc.70059 2025

  5. [5]

    Petropoulos, S

    M.Koch,C.Arlandini,G.Antonopoulos,A.Baretta,P.Beaujean,G.J.Bex,M.E.Biancolini,S.Celi,E.Costa,L.Drescher,V.Eleftheriadis, N.A.Fadel,A.Fink,F.Galbiati,I.Hatzakis,G.Hompis,N.Lewandowski,A.Memmolo,C.Mensch,D.Obrist,V.Paneta,P.Papadimitroulas, K. Petropoulos, S. Porziani, G. Savvidis, K. Sethia, P. Strakos, P. Svobodova, E. Vignali, HPC+ in the medical field: ...

    work page doi:10.3233/thc-229015 2023

  6. [6]

    Z. D. Stephens, S. Y. Lee, F. Faghri, R. H. Campbell, C. Zhai, M. J. Efron, R. Iyer, M. C. Schatz, S. Sinha, G. E. Robinson, Big Data: Astronomical or Genomical?, PLOS Biology 13 (7) (2015) e1002195.doi:10.1371/journal.pbio.1002195

    work page doi:10.1371/journal.pbio.1002195 2015

  7. [7]

    G.Lightbody,V.Haberland,F.Browne,L.Taggart,H.Zheng,E.Parkes,J.K.Blayney,Reviewofapplicationsofhigh-throughputsequencing in personalized medicine: Barriers and facilitators of future progress in research and clinical application, Briefings in Bioinformatics 20 (5) (2019) 1795–1811.doi:10.1093/bib/bby051

    work page doi:10.1093/bib/bby051 2019

  8. [8]

    J. Dean, S. Ghemawat, MapReduce: Simplified data processing on large clusters, Communications of the ACM 51 (1) (2008) 107–113. doi:10.1145/1327452.1327492

    work page doi:10.1145/1327452.1327492 2008

  9. [9]

    Goth, Eric Brewer: Change agent, Communications of the ACM 53 (7) (2010) 24–24.doi:10.1145/1785414.1785425

    G. Goth, Eric Brewer: Change agent, Communications of the ACM 53 (7) (2010) 24–24.doi:10.1145/1785414.1785425

    work page doi:10.1145/1785414.1785425 2010

  10. [10]

    Kovatch, L

    P. Kovatch, L. Gai, H. M. Cho, E. Fluder, D. Jiang, Optimizing High-Performance Computing Systems for Biomedical Workloads, in: 2020 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), IEEE, 2020, pp. 183–192.doi:10.1109/ ipdpsw50202.2020.00040

    arXiv 2020

  11. [11]

    J. Li, S. Wang, S. Rudinac, A. Osseyran, High-performance computing in healthcare: An automatic literature analysis perspective, Journal of Big Data 11 (1) (May 2024).doi:10.1186/s40537-024-00929-2

    work page doi:10.1186/s40537-024-00929-2 2024

  12. [12]

    S.Y.Mashayak,M.N.Jochum,K.Koschke,N.R.Aluru,V.Rühle,C.Junghans,RelativeEntropyandOptimization-DrivenCoarse-Graining Methods in VOTCA, PLOS ONE 10 (7) (2015) e0131754.doi:10.1371/journal.pone.0131754

    work page doi:10.1371/journal.pone.0131754 2015

  13. [13]

    P. R. Alliata, D. Rubaga, D. Kumlin, A. Puliga, Parallelizing Drug Discovery: HPC Pipelines for Alzheimer’s Molecular Docking and Simulation (2025).doi:10.48550/ARXIV.2509.00937

    work page doi:10.48550/arxiv.2509.00937 2025

  14. [14]

    S.J.Marrink,D.P.Tieleman,PerspectiveontheMartinimodel,ChemicalSocietyReviews42(16)(2013)6801.doi:10.1039/c3cs60093a

    work page doi:10.1039/c3cs60093a 2013

  15. [15]

    Braun, J

    E. Braun, J. Gilmer, H. B. Mayes, D. L. Mobley, J. I. Monroe, S. Prasad, D. M. Zuckerman, Best Practices for Foundations in Molecular Simulations [Article v1.0], Living Journal of Computational Molecular Science 1 (1) (2018) 5957.doi:10.33011/livecoms.1.1.5957

    work page doi:10.33011/livecoms.1.1.5957 2018

  16. [16]

    E. R. Georgieva, Protein Conformational Dynamics upon Association with the Surfaces of Lipid Membranes and Engineered Nanoparticles: Insights from Electron Paramagnetic Resonance Spectroscopy, Molecules 25 (22) (2020) 5393.doi:10.3390/molecules25225393

    work page doi:10.3390/molecules25225393 2020

  17. [17]

    Ramirez, V

    D. Ramirez, V. Kohar, M. Lu, Toward Modeling Context-Specific EMT Regulatory Networks Using Temporal Single Cell RNA-Seq Data, Frontiers in Molecular Biosciences 7 (Apr. 2020).doi:10.3389/fmolb.2020.00054

    work page doi:10.3389/fmolb.2020.00054 2020

  18. [18]

    J.Herber,J.Njavro,R.Feederle,U.Schepers,U.C.Müller,S.Bräse,S.A.Müller,S.F.Lichtenthaler,ClickChemistry-mediatedBiotinylation RevealsaFunctionfortheProteaseBACE1inModulatingtheNeuronalSurfaceGlycoproteome,Molecular&CellularProteomics17(8) (2018) 1487–1501.doi:10.1074/mcp.ra118.000608

    work page doi:10.1074/mcp.ra118.000608 2018

  19. [19]

    Adrian, E

    G. Adrian, E. Konradsson, M. Lempart, S. Bäck, C. Ceberg, K. Petersson, The FLASH effect depends on oxygen concentration, The British Journal of Radiology 93 (1106) (Dec. 2019).doi:10.1259/bjr.20190702. P. H. L. Ramalho et al.:Preprint submitted to ElsevierPage 17 of 18 Strategies for Molecular Dynamics using Hybrid Systems

    work page doi:10.1259/bjr.20190702 2019

  20. [20]

    S. Seo, W. Shinoda, SPICA Force Field for Lipid Membranes: Domain Formation Induced by Cholesterol, Journal of Chemical Theory and Computation 15 (1) (2018) 762–774.doi:10.1021/acs.jctc.8b00987

    work page doi:10.1021/acs.jctc.8b00987 2018

  21. [21]

    Monticelli, D

    L. Monticelli, D. P. Tieleman, Force Fields for Classical Molecular Dynamics, in: Biomolecular Simulations, Humana Press, 2012, pp. 197– 213.doi:10.1007/978-1-62703-017-5_8

    work page doi:10.1007/978-1-62703-017-5_8 2012

  22. [22]

    R. O. Dror, R. M. Dirks, J. Grossman, H. Xu, D. E. Shaw, Biomolecular Simulation: A Computational Microscope for Molecular Biology, Annual Review of Biophysics 41 (1) (2012) 429–452.doi:10.1146/annurev-biophys-042910-155245

    work page doi:10.1146/annurev-biophys-042910-155245 2012

  23. [23]

    Lindorff-Larsen, P

    K. Lindorff-Larsen, P. Maragakis, S. Piana, M. P. Eastwood, R. O. Dror, D. E. Shaw, Systematic Validation of Protein Force Fields against Experimental Data, PLoS ONE 7 (2) (2012) e32131.doi:10.1371/journal.pone.0032131

    work page doi:10.1371/journal.pone.0032131 2012

  24. [24]

    Gelpi, A

    J. Gelpi, A. Hospital, R. Goñi, M. Orozco, Molecular dynamics simulations: Advances and applications, Advances and Applications in Bioinformatics and Chemistry (2015) 37doi:10.2147/aabc.s70333

    work page doi:10.2147/aabc.s70333 2015

  25. [25]

    Vanommeslaeghe, E

    K. Vanommeslaeghe, E. Hatcher, C. Acharya, S. Kundu, S. Zhong, J. Shim, E. Darian, O. Guvench, P. Lopes, I. Vorobyov, A. D. Mackerell, CHARMM general force field: A force field for drug-like molecules compatible with the CHARMM all-atom additive biological force fields, Journal of Computational Chemistry 31 (4) (2009) 671–690.doi:10.1002/jcc.21367

    work page doi:10.1002/jcc.21367 2009

  26. [26]

    Y.Nochez,S.Arsene,N.Gueguen,A.Chevrollier,M.Ferré,V.Guillet,etal.,Acuteandlate-onsetopticatrophyduetoanovelOPA1mutation leading to a mitochondrial coupling defect, Molecular Vision 15 (2009) 598–608

    2009

  27. [27]

    M. C. Righetti, L. Aliotta, N. Mallegni, M. Gazzano, E. Passaglia, P. Cinelli, A. Lazzeri, Constrained Amorphous Interphase and Mechanical Properties of Poly(3-Hydroxybutyrate-co-3-Hydroxyvalerate), Frontiers in Chemistry 7 (Nov. 2019).doi:10.3389/fchem.2019.00790

    work page doi:10.3389/fchem.2019.00790 2019

  28. [28]

    Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J

    S. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, Journal of Computational Physics 117 (1) (1995) 1–19. doi:10.1006/jcph.1995.1039

    work page doi:10.1006/jcph.1995.1039 1995

  29. [29]

    U.Essmann,L.Perera,M.L.Berkowitz,T.Darden,H.Lee,L.G.Pedersen,AsmoothparticlemeshEwaldmethod,TheJournalofChemical Physics 103 (19) (1995) 8577–8593.doi:10.1063/1.470117

    work page doi:10.1063/1.470117 1995

  30. [30]

    A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T.D.Nguyen,R.Shan,M.J.Stevens,J.Tranchida,C.Trott,S.J.Plimpton,LAMMPS-aflexiblesimulationtoolforparticle-basedmaterials modelingattheatomic,meso,andcontinuumscales,ComputerPhysicsCommunications271(2022)108171.doi:10.1016/j...

    work page doi:10.1016/j.cpc.2021 2022

  31. [31]

    Rohskopf, C

    A. Rohskopf, C. Sievers, N. Lubbers, M. A. Cusentino, J. Goff, J. Janssen, M. McCarthy, D. M. de Oca Zapiain, S. Nikolov, K. Sargsyan, D. Sema, E. Sikorski, L. Williams, A. P. Thompson, M. A. Wood, FitSNAP: Atomistic machine learning with LAMMPS, Journal of Open Source Software 8 (84) (2023) 5118.doi:10.21105/joss.05118

    work page doi:10.21105/joss.05118 2023

  32. [32]

    G. M. Amdahl, Validity of the single processor approach to achieving large scale computing capabilities, in: Proceedings of the April 18-20, 1967,SpringJointComputerConferenceon-AFIPS’67(Spring),AFIPS’67(Spring),ACMPress,1967,p.483.doi:10.1145/1465482. 1465560

    work page doi:10.1145/1465482 1967

  33. [33]

    Ghedira, O

    K. Ghedira, O. Khamessi, C. Hkimi, S. Kamoun, N. Dhamer, K. Daassi, W. Ben Salah, H. Othman, W. Belhadj, Y. Ghorbal, Design and implementation of a scalable high-performance computing (HPC) cluster for omics data analysis: Achievements, challenges and recommendations in LMICs, GigaScience 13 (2024).doi:10.1093/gigascience/giae060

    work page doi:10.1093/gigascience/giae060 2024

  34. [34]

    Miyazaki, Y

    Y. Miyazaki, Y. Teppei, W. Shinoda, onefive13, SPICA-group/spica-tools: V1.0.0, Zenodo (Feb. 2024).doi:10.5281/ZENODO.10611578. P. H. L. Ramalho et al.:Preprint submitted to ElsevierPage 18 of 18

    work page doi:10.5281/zenodo.10611578 2024