GROMACS 2019.4 release notes#

This version was released on October 2nd, 2019. These release notes document the changes that have taken place in GROMACS since the previous 2019.3 version, to fix known issues. It also incorporates all fixes made in version 2018.7 and earlier, which you can find described in the Release notes.

Fixes where mdrun could behave incorrectly#

Fix incorrect pressure when atoms in CMAP cross a box boundary#

The virial calculation and thus the pressure would be incorrect when the second and third atom involved in a CHARMM CMAP correction term would reside in different periodic images. This can happen when a protein is positioned over a box boundary. Note that the energy and forces were correct, but sampling was affected when pressure coupling was applied when a protein crossed a box boundary.

Issue 2845 Issue 2867

Fix incorrect LJ cut-off on GPU when rvdw < rcoulomb#

When rvdw was chosen by the user to be smaller than rcoulomb in the mdp file, the LJ cut-off would initially be set to the Coulomb cut-off for computing non-bonded interactions on the GPU. This only affected energy minimization, mdrun -rerun and the first 2*nstlist steps of a normal MD run, since the correct LJ cut-off is set when PME tuning (on by default) starts after 2*nstlist steps (unless PME tuning was disabled with -notunepme).

Issue 3056

Fix (unlikely) missing bonded forces with CUDA GPUs and domain decomposition#

Forces could be missing for bonded interactions computed on CUDA GPUs with domain decomposition when there are non-local bonded interactions, but no non-local non-bonded interactions between two domains. Note that this is extremely unlikely to happen, since the distance between the bonded atoms needs to be larger than the pair-list cut-off distance and there should be no other non-local atoms within the pair-list cut-off distance.

Issue 3063

Fix incorrect reporting of final kinetic energy and temperature#

With the leap-frog integrator, the kinetic energy and temperature reported for the last step were incorrect when the last step was not divisible by nstcalcenergy, nsttcouple or nstpcouple.

Issue 2950

Fix segmentation fault in grompp and mdrun with cosine COM pulling#

Issue 3023

Fixes for gmx tools#

Fix grompp not adding angle constraints between constraints#

When using the mdp option constraints=all-angles, angles involving bonds supplied as constraints in the topology would be removed, but not replaced by angle constraints.

Issue 3067

Fix gmx wham with angle and dihedral geometries#

gmx wham would apply an incorrect radian to degree unit conversion, leading to no overlap or not-a-number output.

Issue 2609 Issue 3094

Fix bug in gmx xpm2ps#

The tool would fail when not being provided with a library file to read in.

Issue 3012

Fix bug in gmx anaeig#

An issue was noted when reading a second set set of eigenvectors that could lead to problems when the number of eigenvectors was less than the three times the number of atoms.

Issue 2972

Fix issue with demux.pl script#

The trajectories could become discontinuous with simulations longer than 100ns and exchange strides that are not a multiple of 1 ps. This only affected the post-processing of trajectories generated from replica exchange simulations.

Made gmx disre work with non-consecutively labeled restraints#

Issue 2953

Fixed writing of gro files with index groups#

An output .gro file from from e.g. gmx editconf -f -n would take the atom names for the output file in order from the atoms in the input file, rather than in order from the atoms indicated by the indices in the index file.

Issue 3107

Made gmx make_ndx keep chain IDs#

Old style structure file reading caused the chain IDs to be overwritten with default values.

Issue 3070

Fixes that affect portability#

Disable PME OpenCL on Apple#

The Apple OpenCL compilers fail to produce a functional clFFT build. The OpenCL PME support is therefore disabled on Apple platforms.

Issue 2941

Miscellaneous#

Added AMD Zen 2 detection#

The AMD Zen 2 architecture is now detected as different from Zen 1 and uses 256-bit wide AVX2 SIMD instructions (GMX_SIMD=AVX2_256) by default. Also the non-bonded kernel parameters have been tuned for Zen 2. This has a significant impact on performance.