New and improved features¶
Multiple molecule types with identical SETTLE parameters are now supported¶
This can be useful when one, for instance, wants to apply position restraints to a subset of water molecules.
Periodic SCC-DFTB, GFN1-xTB and GFN2-xTB are available for CP2K QM/MM¶
The CP2K QM/MM interface can now generate periodic SCC-DFTB, GFN1-xTB and
GFN2-xTB inputs using short-range regularized point-charge coupling with SPME
electrostatics. For GFN1-xTB and GFN2-xTB CP2K should be compiled with support
for the tblite interface.
The user can select Gaussian Expansion of Electrostatic Potential (GEEP)
for MM charges in tight-binding methods through the
qmmm-cp2k-dftb-electrostatic-coupling = gauss MDP option,
which requires a CP2K 2027.1 or higher, whereas the default is
short-range regularized point-charge coupling
qmmm-cp2k-dftb-electrostatic-coupling = point-charge.
Regular DFT methods are still using GEEP (Gauss) for electrostatics.
CMAP interactions now support free energy perturbation¶
Energy correction map (CMAP) torsion interactions can now be
perturbed between an A state and a B state using the
bonded-lambdas free-energy lambda component. The A and B
state CMAP grids are linearly interpolated at each lambda
value, and the corresponding \(\partial H/\partial\lambda\)
contribution is accumulated for BAR/TI analysis.
To use CMAP FEP, specify a B-state CMAP type in the [ cmap ]
section of the topology. The type can be given as a 1-based index
or as an explicit name token (e.g. GLY, PZQ) that matches
the optional name field on the corresponding [ cmaptypes ]
header line. When the A and B state CMAP types differ, the
interaction is treated as perturbed; when they are identical (or
no B-state is specified) the standard non-perturbed code path is
used.
Neural network potentials can run on Intel GPUs¶
The NNPot interface now runs model inference on Intel GPUs when a LibTorch
with XPU support (PyTorch 2.5 or later) is used. The model runs on the first
Intel GPU LibTorch enumerates, by default and with GMX_NN_DEVICE=gpu.
Configuring warns when a SYCL build for Intel GPUs is paired with a LibTorch
without XPU support.
Reading and writing of mmCIF/PDBx structure and trajectory files¶
GROMACS tools can now read and write structures in the mmCIF/PDBx format
(.cif), the primary distribution format of the wwPDB. This covers entries
that the legacy PDB format cannot express, including extended entry identifiers
(the pdb_######## form, up to 12 characters) and chemical-component
identifiers of up to 5 characters. Multi-model .cif files can be read and
written as trajectories, for example with gmx trjconv.
With the -conect option of gmx editconf and gmx trjconv, atom
connectivity is written as struct_conn records, the mmCIF counterpart of the
PDB LINK and SSBOND records, and read back again.
mmCIF allows chain identifiers of more than one character, while GROMACS stores a single character. Such identifiers are truncated to their first character, and a warning naming them is printed.