GROMACS 2023.4 release notes#

This version was released on January 24th, 2024. These release notes document the changes that have taken place in GROMACS since the previous 2023.3 version, to fix known issues. It also incorporates all fixes made in version 2022.6 and earlier, which you can find described in the Release notes.

Fixes where mdrun could behave incorrectly#

Freezing AWH simulations with multiple MPI ranks per walker#

AWH simulations with biases shared by multiple walkers and using multiple MPI ranks per walker would hang when outputting energy, i.e., already at step 0.

There would not be any incorrect results.

Issue 4925

With AWH for FE, mdrun could exit with an assertion failure#

The would happen e.g. when using soft-core only for LJ, whereas it was also needed for Coulomb or when decoupling large molecules. Now proper error messages are generated.

Issue 4938

Missing force contribution on neighbor search steps with GPU update#

On a neighbor search step, when virial is _not_ computed (i.e., when nstlist is not a multiple of nstpcouple) and GPU update is used, there could have been a race between GPU buffer clearing and copying of forces computed on the CPU, leading to missing some of the force contribution on this step.

This should not have ever happened unless the GPU was heavily oversubscribed or hipSYCL/AdaptiveCpp was used, and even in those cases, the problem was very unlikely to manifest.

Issue 4937

Fixes for gmx tools#

Fixed a segmentation fault in grompp with AWH and FEP without FEP states#

Issue 4932

gmx msd now raises a proper error when given incompatible trajectory#

gmx msd internally converts frame times to picoseconds. As a consequence, if the time cannot be expressed as an integral number of picoseconds (usually due to too frequent trajectory output), the tool would crash or produce an incorrect result.

Now, an error is printed in such cases.

Issue 4694

Fixes that affect portability#

Fixed compilation with Boost 1.83#

Issue 4909


A grompp warning when using Parrinello-Rahman pressure coupling combining with position restraints suggested to use Berendsen instead. This suggestion has now been changed to C-rescale.