Influence of the Treatment of Nonbonded Interactions on the Thermodynamic and Transport Properties of Pure Liquids Calculated Using the 2016H66 Force Field

Abstract

The effect of different treatments of the nonbonded interactions in simulations employing the recently introduced GROMOS-compatible 2016H66 force field is evaluated based on calculations carried out with the GROMACS software. This is done considering four thermodynamic and transport properties (pure liquid density, vaporization enthalpy, surface-tension coefficient, and self-diffusion constant) of 58 organic liquids representative of the chemical groups alcohol, ether, aldehyde, ketone, carboxylic acid, ester, amine, amide, thiol, sulfide, disulfide, and aromatic compounds, also including water (SPC model). A dipalmitoylphosphatidylcholine bilayer system is considered as well. The simulated properties are found to be very sensitive to the treatment of the long-range dispersion interactions, notably for the least polar systems. In general, the treatment of the long-range electrostatic or Lennard-Jones interactions using homogeneous correction terms or lattice-sum approaches yield similar results, with punctual discrepancies. The combination of a lattice-sum approach for the electrostatic interactions with a straight-cutoff truncation of the Lennard-Jones interactions at a distance of at least 1.2 nm is found to represent a good compromise setup within GROMACS for achieving compatibility with the reference results obtained using GROMOS as well as a comparable level of agreement with the experimental data. This study also reveals two potential issues with the GROMACS software, related to an incorrect calculation of the pressure when using LINCS in version 4.0.7 and an inadequate implementation of the twin-range scheme in version 5.1.2.

On The Importance of Accurate Algorithms for Reliable Molecular Dynamics Simulations

“…Furthermore, the authors 15 claimed to have found a potential issue with the implementation of the LINCS constraint algorithm 21 in GROMACS 4.0.7. Although it is of course within the scientific freedom of the authors to report suspected errors in a paper, we think it would have been more productive to first discuss an observation with the developers to confirm, potentially adress it, and make sure other users are informed as well.…”

“…Two recent papers from the GROMOS community have suggested there are reliability issues in the GROMACS software as a result of this change 14,15 . In short, both papers notice that the GROMOS force field that was developed in conjunction with the GROMOS software and legacy twin-range cutoffs does not produce the same results when used with Trotter decompositions as implemented in GROMACS a decade ago, and to the best of our knowledge other modern codes using multiple-time-step integration.…”

Reaction-field electrostatics in molecular dynamics simulations: development of a conservative scheme compatible with an atomic cutoff

“…20,[25][26][27][28] For the LJ interactions, the straight-cutoff scheme is less problematic. 29 Mainly a cutoff-dependent underestimation (relative to the long-cutoff limit) of the densities, vaporization/sublimation enthalpies and surface-tension coefficients 21,30,31 occurs due to the neglect of the predominantly dispersive (i.e. attractive) interactions beyond the cutoff.…”

“…at the surface of a macromolecule or for systems in interfacial/slab geometries. 21,56,57 Note that anisotropic mean-field variants have also been proposed for these situations, considering both electrostatic interactions 50,57 and LJ interactions. [58][59][60][61][62][63][64][65][66][67] See also Ref.…”

“…86 The use of a lattice-sum approach is most appropriate for truly periodic systems (crystals, neglecting static and dynamic disorder), but only approximate for inherently non-periodic systems such as liquids and solutions. In practice, for the comparatively shortranged LJ interactions, lattice-sum and mean-field approximations typically lead to comparably accurate results for bulk systems, 21,78 but generally not for interfacial systems, where latticesum is to be preferred. 30,56,83,85,86,88 There are three additional advantages to the use of mean-field over lattice-sum schemes: (i) for large systems, they enable an O(N) scaling of the computational cost, compared to an O(N log N) scaling for FFT-based lattice-sum schemes, 74,89 (ii) they enable a pairwise (or groupwise) partitioning of the interaction energy without additional computational cost, whereas such a direct partitioning is not possible in lattice-sum schemes (i.e.…”