New and improved features#

The AWH exponential histogram growth can now be controlled#

The accelerated weight histogram growth factor during the initial phase was hard-coded to 3. Now this value can be controlled by the user. It is set to 2 by default for increased stability.

If the TPR was generated with an earlier GROMACS version, the old default value of 3 will be used.

Added support for instrumentation based on wallcycle regions using NVTX/ROCTX/ITT#

Basic support has been added for GPU tracing libraries so wallcycle main and sub-regions will show up in tracing timelines which can help with performance analysis. The tracing instrumentation support can be enabled with one of the following CMake variables: GMX_USE_NVTX, GMX_USE_ROCTX, GMX_USE_ITT.

Issue 4446

Collective variables (Colvars) module support#

The collective variables (Colvars) library for enhanced sampling simulations has a new and improved interface, which simplifies greatly its use and distribution with current and future GROMACS releases. The new interface requires no patching and supports a full integration of the Colvars input and of its restart data with the GROMACS TPR and CPT files, respectively.

For documentation and details, please see this section of the GROMACS doc along with the Colvars doc page for GROMACS. Additionally, messages in the GROMACS discussion forum can also be tagged with the colvars keyword for easier consultation.

Automatic metric scaled AWH target distribution#

The AWH target distribution can now be automatically scaled by sqrt(AWH friction metric). Regions with higher friction (slower diffusion) will get a higher target distribution. This should generally lower the statistical error of the estimated free energy landscape. The new option is called ‘awh1-target-metric-scaling’ and can be applied to further modify all AWH target distributions and/or AWH user input, but is not recommended in general in combination with Boltzmann or Local-Boltzmann target distributions, due to the risk of feedback loops between the two adaptive update mechanisms.