grompp no longer modifies nstcomm¶
grompp will no longer set nstcomm, the interval for center of mass motion removal, equal to nstcalcenergy when nstcomm < nstcalcenergy. A note is still printed in that case.
Bonded atom types names can now start with a digit¶
Bonded atom types names in topologies were not allowed to start with a number. Now all names are supported that contain at least one non-digit character.
grompp now warns when exclusion forces might be missing¶
When using PME, exclusions between non-perturbed,atom pairs should be within the cut-off distance, otherwise mdrun might not compute grid correction forces and energies. grompp now computes these distance for the starting structure and warns when they are beyond 90% of the cut-off distance and generates an error when they are beyond the cut-off distance.
The AWH cover diameter for angles now has units degrees¶
Using old tpr files that apply AWH to angles or dihedrals and have a non-zero cover diameter results in an error with the suggestion to regenerate the tpr file.
Core spin-up code is removed¶
Formerly, on non-x86 and non-PowerPC platforms, mdrun ran some multi-threaded code to try to wake up any cores that the OS might have powered down. This caused problems on some Arm platforms, and does not seem to suit a significant number of platforms for use of GROMACS. So now it is removed.
If required, please manually spin-up the cores with, e.g.,
stress --cpu $(nproc --all).
Add documentation for linear angle potential¶
Added documentation and reference for the linear angle potential. Also added please_cite entry, but there is no call to reference it yet.
gmxapi.mdrun guarantees trajectory output¶
gmxapi simulations now always run with full-precision trajectory output (
in order to guarantee the availability of a usable output trajectory through the
gmxapi.mdrun accepts arbitrary runtime arguments¶
Arbitrary mdrun arguments can be passed through gmxapi with the new runtime_args key word argument, accepting a dictionary of flags and values.
Improved MPI awareness and task uniqueness for gmxapi Python runner¶
Previously, only the Python components in
gmxapi.simulation reacted to the presence
of an MPI context. This could result in duplicate work or even invalid file access.
gmxapi.commandline_operation() now executes tasks in unique working directories.
For all gmxapi operations, tasks are only launched from one process (per ensemble member).
If mpi4py is available,
the MPI environment is inspected.
If multiple ranks are discovered, the
ResourceManager instances on the various ranks coordinate
to make sure that
update is only called for each member of each task once. Results are
broadcast to all ranks from the
ResourceManager where the work occurred.
These changes merely constitute a bug-fix. Additional development is needed for more optimal use of resources and to reduce unnecessary data transfers.
Further discouraged use of Berendsen coupling algorithms¶
Those algorithms have been proven to cause incorrect sampling of their respective distributions and are mainly provided as a means to provide backwards compatibility for older simulations. This is why their use has been further discouraged by changing the current notes about their use to actual warnings at grompp time.