GROMACS 2022.2 release notes¶
This version was released on June 16th, 2022. These release notes document the changes that have taken place in GROMACS since the previous 2022.1 version, to fix known issues. It also incorporates all fixes made in version 2021.5 and earlier, which you can find described in the Release notes.
Fixes where mdrun could behave incorrectly¶
Fix incorrect constraining in highly parallel runs¶
With domain decomposition, OpenMP and connected constraints (so not only bonds with hydrogens constrained), constraint correction could be applied when a domain would not have any constraints while it had constraints before. This is unlikely to have gone unnoticed in longer runs, as the chance is high that atoms would clash and the system would become unstable. But short runs might not crash and could therefore produce incorrect results. The correctness can be judged from the drift in the conserved energy quantity, which is reported at the end of the log file, and which will be one to two orders of magnitude larger than for a correct run.
Fix missing CPU-GPU synchronization when doing free-energy calculations¶
When GPU halo exchange with direct communication is enabled, CPU based free-energy kernels were being run without waiting on non-local co-ordinates to be available on host. This resulted in use of stale data on CPU side. This issue can cause incorect output when GPU direct communication is enabled using GMX_ENABLE_DIRECT_GPU_COMM environment variable and simulation contains free-energy calculations.
Fix missing PME mesh dV/dlambda with PME on GPU on a seperate PME rank¶
When doing free-energy calculations with PME running on GPU on a separate PME rank, the dV/dlambda contribution for the PME mesh part was missing. The same contribution was missing from the foreign lambda energy differences. Note that the energies and forces were correct.
Removed the (incorrect) output file size note/warning with mdrun -rerun¶
Wait for PME coordinate padding clearing to complete after reinitialization¶
As part of the irregular reinitialization of PME coordinate buffers, the padding area of the buffer on the GPU must be set to zero. Previously, a dependency was missing such that, with GPU direct communications enabled via the GMX_ENABLE_DIRECT_GPU_COMM, it was possible that the PME kernel could preceed before this initialization had completed, and this has now been fixed. The incorrect ordering is only expected to have occured in extreme benchmarking cases, causing an obvious crash.
Note known issues with Verlet buffer estimation¶
There are known issues with the Verlet buffer estimate for imhonogeneous systems as well as for potentials with only the repulsive part of the Lennard Jones potential. These issues and workarounds are listed in the known issues section.
Clarified pdb2gmx error message for undefined atom type in terminal database¶
Lower severity of grompp exclusion distance issues with energy minimization¶
With energy minimization, the error/warning about pair distances beyond/close to the cut-off has been changed to a warning/note, as energy minimization might fix such issues.
Fixed missing cmap torsion correction for periodic pepdites¶
When using pdb2gmx with periodic peptides and the CHARMM27 force field, the CMAP correction across the periodic boundary was missing (but not the torsion itself). Reprocess your PDB file with pdb2gmx from version 2022.2 or later to get a correct topology, or add it manually to the topology.
Avoid crash in gmx bar with invalid input¶
gmx bar could try to read an invalid input data file and would fail with a hard crash instead of a helpful error message.
Fix printing of incorrect data from analysis tools¶
A change on how we handle selections caused the analysis tools in the trajectoryanalysis framework to print data for wrong selections.
Remove broken charge zeroing function in convert-tpr¶
This had been broken for a long time and no discernable use.
Fixes that affect portability¶
Warn when using gcc version 7 with CUDA builds¶
Different versions of gcc 7 behave differently in a way that makes it
hard for GROMACS to check whether CUDA’s
nvcc compiler will
accept compiler flags. GROMACS 2022 and 2022.1 sometimes would
spuriously detect flags as invalid, avoid using them, and so produce
slow CUDA kernels. Now GROMACS assumes all nvcc flags are valid in
this case and the build system warns when this is occuring. If you
then experience build failures, please use a newer version of gcc.
Fix external tinyXML version to below 7¶
Newer versions are not compatible with GROMACS.
Fixed detection of external TinyXML-2¶
Updated the code to properly detect the presence and the version of external
TinyXML-2 (only relevant if
-DGMX_EXTERNAL_TINYXML2=ON is used).
Fixed warnings when using module-specific OpenMP thread-count environment variables¶
One of the arrays used to construct message strings wasn’t updated properly, so sometimes an information message was wrong or may have printed garbage.