Source-linked AI summary
ms2: A molecular simulation tool for thermodynamic properties, new version release
Colin W. Glass, Steffen Reiser, Gábor Rutkai, Stephan Deublein, Andreas Köster, Gabriela Guevara Carrión, Amer Wafai, Martin Horsch, Martin F. Bernreuther, Thorsten Windmann, Hans Hasse, Jadran Vrabec
TL;DR
Version 2.0 addresses scalability constraints in molecular dynamics for system sizes where domain decomposition is excluded. It combines hybrid MPI/OpenMP parallelization with systematic Massieu-potential derivative sampling, improving multicore performance and broadening accessible thermodynamic properties.
Problem
For system sizes where the cut-off radius approaches half the simulation-volume edge length, domain decomposition is excluded, limiting standard highly scalable molecular-dynamics parallelization.
Method
Version 2.0 hybridizes existing MPI parallelization with OpenMP and adds systematic calculation of Massieu potential derivatives.
Results
Improved performance on multi-core processors and broader property sampling support development of competitive fundamental equations of state from experimental VLE data and molecular simulation results.
Takeaways & Limitations
The release expands the accessible properties of ms2 and supports combining experimental VLE data with molecular simulation results for fundamental equations of state.
Takeaways & Limitations
Massieu potential derivative sampling and hybrid MPI/OpenMP parallelization are not available together.
Abstract
from arXiv · showhide
A new version release (2.0) of the molecular simulation tool ms2 [S. Deublein et al., Comput. Phys. Commun. 182 (2011) 2350] is presented. Version 2.0 of ms2 features a hybrid parallelization based on MPI and OpenMP for molecular dynamics simulation to achieve higher scalability. Furthermore, the formalism by Lustig [R. Lustig, Mol. Phys. 110 (2012) 3041] is implemented, allowing for a systematic sampling of Massieu potential derivatives in a single simulation run. Moreover, the Green-Kubo formalism is extended for the sampling of the electric conductivity and the residence time. To remove the restriction of the preceding version to electro-neutral molecules, Ewald summation is implemented to consider ionic long range interactions. Finally, the sampling of the radial distribution function is added.
1. Introduction
Version 2.0 extends ms2 with hybrid MPI/OpenMP molecular-dynamics parallelization and broader property calculations. The release adds systematic Massieu potential derivatives, ionic-substance simulation, and reduced transport-property resource demands.
- Version 2.0 hybridizes existing MPI parallelization with OpenMP, improving performance on multi-core processors.
- The release systematically calculates Massieu potential derivatives, expanding the range of sampled properties.These derivatives support developing fundamental equations of state from experimental VLE data and molecular simulation results.
- ms2 can now simulate ionic substances, removing the preceding restriction to electro-neutral molecules.
- Transport-property calculations require significantly less time and memory in the new release.
- The implementation remains available as open-source code for academic users.
2. Hybrid MPI & OpenMP Parallelization
Because the cut-off radius can approach half the simulation-volume edge, domain decomposition is infeasible for the system sizes targeted by ms2, limiting MPI scalability. Version 2.0 addresses this with hybrid MPI/OpenMP parallelization, which improves scalability when MPI communication becomes a bottleneck.
- Motivation: For ms2 system sizes, the cut-off radius can be comparable to half the simulation-volume edge, excluding domain decomposition and limiting MPI scalability.
- Hybrid MPI & OpenMP approach: OpenMP is hybridized with MPI so a process can distribute remaining load to multiple threads when MPI communication becomes a bottleneck.
- Implementation: Three ms2 components were parallelized with OpenMP: interaction-partner search, energy calculation, and force calculation.
- Implementation: OpenMP loop parallelism is used for compute-intensive loops over molecules, while force and torque accumulation avoids atomic-update or critical-section overheads.
- Performance evaluation: The hybrid evaluation varied threads per MPI process and molecule count on a CRAY XE6 supercomputer.
- Performance evaluation: Around 20% speed-up was achieved with 2 to 4 threads per MPI process for 2’048 cores.
3. Massieu potential derivatives
Version 2.0 extends ms2 with systematic sampling of Massieu potential derivatives, enabling broad thermodynamic-property evaluation from single NVT simulations while retaining explicit limits for phase equilibria and certain derivatives.
- Systematic derivative sampling: Version 2.0 implements systematic free-energy derivative evaluation, expanding the thermodynamic property types sampleable in single simulation runs.The approach uses the Lustig formalism for simultaneous derivative sampling in a single NVT ensemble simulation at a given state point.
- Thermodynamic formulation: The Massieu potential F/T(N, V, 1/T) is preferred in molecular simulations because of practical considerations.Its independent variables are represented through particle number, volume, and inverse temperature.
- Thermodynamic formulation: The derivative formalism separates each derivative into ideal and residual parts, with molecular simulation targeting the residual contribution.The ideal part can instead be obtained from spectroscopic data, ab initio calculations, or known density dependence.
- Thermodynamic-property coverage: The first five derivatives A10, A01, A20, A11, and A02 express every measurable thermodynamic property except phase equilibria.The paper refers readers to supplementary material for the property list and describes the implementation there in detail.
- Implementation requirements: Calculating derivatives through order n = 2 requires analytical expressions for ∂U/∂V and ∂2U/∂V 2 for the applied molecular interaction potential.Long-range corrections compensate for finite system size, and their form depends on the interaction potential and cutoff method.
- Long-range corrections: For a 200-water/50-methanol liquid example, electrostatic long-range corrections were still << 1% for both ∂U/∂V and ∂2U/∂V 2.The example used T = 298 K, ρ = 45.86 mol/l, and a cutoff radius equal to 20% of the simulation-volume edge length.
4. Algorithmic Developments
ms2 extends its equilibrium-MD property calculations with Green–Kubo-based transport sampling and Ewald electrostatics. The release also reduces autocorrelation-related costs, adds ionic applicability, and retains a limitation on combining selected new features.
- Transport-property sampling: Green–Kubo sampling determines transport properties from flux autocorrelation functions and their time integrals.The formalism is used for transport-property calculations in equilibrium molecular dynamics simulations.
- Transport-property sampling: Using an extended time step evaluates autocorrelation functions every n-th MD step, reducing their memory demand and restart-file size.The extended step is n times longer than the specified MD time step, with n user-defined.
- Transport-property sampling: The revised autocorrelation treatment significantly reduces the overall computing time of MD simulations.
- Electrostatic interactions: Ewald summation extends ms2 to thermodynamic properties of ions in solutions.This removes the preceding restriction to electro-neutral molecules for ionic long-range interactions.
- Electrostatic interactions: Ewald summation divides electrostatic interactions into short-range real-space and long-range Fourier-space contributions.The approach transforms the final long-range value back into real space for efficient charge-interaction calculation.
- Feature scope: Massieu-potential derivatives and hybrid MPI/OpenMP parallelization are not currently available together.
5. Property Calculations
Version 2.0 adds or extends microscopic property calculations, including radial distribution functions, electric conductivity, mixture thermal conductivity, and residence time. These properties are evaluated from structural or time-correlation information during molecular-dynamics simulations.
- Radial distribution function: The radial distribution function g(r) measures microscopic structure and can be sampled on the fly for pure components and mixtures.Sampling occurs between all Lennard-Jones sites; arbitrary positions require user-defined dummy sites.
- Residence time: The solvation number uses component-j number density and a cutoff distance to quantify local surrounding molecules.
- Electric conductivity: Electric conductivity is evaluated through the Green–Kubo relationship between conductivity and the electric-current flux autocorrelation function.The electric current flux is defined from ion charges and velocities, and only ions in solution are included.
- Thermal conductivity: Thermal-conductivity calculation is extended from pure substances to multicomponent mixtures.For mixtures, the thermal conductivity is obtained from autocorrelation functions of microscopic heat-flow components.
- Residence time: Residence time measures how long a component-j molecule remains within a specified distance of a molecule i.Its autocorrelation formulation permits short intervals in which the distance exceeds the specified threshold, and the solvation number can be evaluated during the simulation.