New and improved features ^^^^^^^^^^^^^^^^^^^^^^^^^ .. Note to developers! Please use """"""" to underline the individual entries for fixed issues in the subfolders, otherwise the formatting on the webpage is messed up. Also, please use the syntax :issue:`number` to reference issues on GitLab, without the a space between the colon and number! Hybrid Quantum-Classical simulations (QM/MM) with CP2K interface """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" Simulations of chemical reactions pathways can provide an atomistic insight into many biological and chemical processes. To perform such kind of modelling in complex systems, that includes solvent and/or proteins Multi-scale Quantum Mechanics / Molecular Mechanics (QM/MM) approaches are often used. Here we introduce a whole new interface to perform QM/MM simulations in fully periodic systems using MDModule that couples |Gromacs| with CP2K quantum chemistry package. This enables hybrid simulations of systems in systems where chemical reactions occurs. The interface supports most of the simulations techniques available in |Gromacs| including energy minimization, classical MD and enhanced sampling methods such as umbrella sampling and accelerated weight histogram method. Transformation pull coordinate for mathematical transformations of pull coordinates """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" A new pull coordinate type named transformation has been added. This enables mathematical transformation of previously defined pull coordinates using a user supplied formula in a string. This allows for example non-linear transformation of a distance, e.g. a contact coordinate or (non-)linear combinations of multiple pull coordinates. This is a powerful tool for defining complex reaction coordinates and it can be combined with the Accelerated Weight Histogram Method to enhance sampling. Replica-exchange molecular dynamics simulations with GPU update """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" Replica-exchange molecular dynamics now works with GPU update. A new formulation of soft-core interactions for free energy calculations """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" With this addition Gromacs allows to choose from two schemes to soften non-bonded interactions during alchemical perturbations: Beutler *et al.*\ :ref:`100 <refBeutler94>` and Gapsys *et al.*\ :ref:`185 <refGapsys2012>` soft-core functions. More flexible sharing of biases in AWH """""""""""""""""""""""""""""""""""""" With the accelerated weight histogram method, biases can now be shared between subsets of all simulations, without restrictions. The allows for more flexible ensemble simulation setups, as well as simpler launches of sets of simulations. More features implemented in modular simulator """""""""""""""""""""""""""""""""""""""""""""" Several features were added to the modular simulator, including all temperature and pressure coupling algorithms available in the legacy simulator, expanded ensemble and pull. Free energy calculations now support all non-perturbed bonded interactions """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" Previously |Gromacs| did not permit any usage of a few more special bonded interactions (restricted angles/dihedrals or combined bending-torsion potentials) in free energy calculations. These are now allowed, as long as the interaction itself is not perturbed. :issue:`3691` Adapt number of threads to actually permitted hardware """""""""""""""""""""""""""""""""""""""""""""""""""""" Previously, |Gromacs| would attempt to start as many threads as there are processors in the system, and try to pin threads on processing units. This would fail whenever we are not allowed to use all those processors, e.g. when Slurm only provides part of a node to a job, or on A64fx where some processors are reserved for the system. We would also start far too many threads in container environments. As part of improved hardware detection, we now only detect processors on which we are allowed to run, and adapt the number of threads whenever there is a cpu limit set, which will improve performance both for containers and make |Gromacs| do the right thing when Slurm or other queue systems allocate part of a node. Enable use of more OpenMP threads """"""""""""""""""""""""""""""""" The thread-force-reduction code in |Gromacs| will now allow up to 128 OpenMP threads by default, and we have changed the internal logic so we just limit the number of threads rather than refuse to run. This only applies within each rank; you can use an unlimited number of threads by combining OpenMP threading with multiple ranks. For large machines with many cores this is usually faster since the domain decomposition used with multiple ranks is better adapted to non-uniform memory access hardware. :issue:`4370` Centering and symmetrization supported in gmx potential """"""""""""""""""""""""""""""""""""""""""""""""""""""" gmx potential now supports the same centering and symmetrization options as gmx density, which is particularly useful for membranes. :issue:`3579`