Molecular dynamics is expected to produce accurate results over a wide range of conditions and timescales. However, this is not always the case since the field has been too reluctant to abandon historically popular techniques known to introduce artefacts. Two recent papers have suggested there are reliability issues in the GROMACS code since it no longer uses a legacy twin-range algorithm. Here, we show there are order-of-magnitude differences in accuracy favoring the modern Trotter decomposition, and that a force field relying on the old algorithm will have errors parametrized into the force field. Similarly, the suggestions about incorrect virial calculations turn out to be explained by insufficient accuracy in the default SHAKE settings used for GROMOS, while the GROMACS default choices are accurate. This highlights the importance of being more critical to error cancellation in simulations in order for algorithms and parameters to both gradually converge to more perfect ones.
Erik Lindahl KTH Royal Institute of Technology
download asset hess2019.pdf 0.23 MB [opens in a new tab] cloud_download
pdf : 0.23 MB