1 GROMACS 2021.3 release notes
2 ----------------------------
4 This version was released on TODO, 2021. These release notes
5 document the changes that have taken place in GROMACS since the
6 previous 2021.2 version, to fix known issues. It also incorporates all
7 fixes made in version 2020.6 and earlier, which you can find described
8 in the :ref:`release-notes`.
10 .. Note to developers!
11 Please use """"""" to underline the individual entries for fixed issues in the subfolders,
12 otherwise the formatting on the webpage is messed up.
13 Also, please use the syntax :issue:`number` to reference issues on GitLab, without the
14 a space between the colon and number!
16 Fixes where mdrun could behave incorrectly
17 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
19 Fix mdrun -ddorder pp_pme
20 """""""""""""""""""""""""
22 When rank ordering PP-PME, mdrun would deadlock during the initialization
27 Fixed gmxapi MD plugin binding
28 """"""""""""""""""""""""""""""
30 Molecular Dynamics extension code was not properly handled when added to a
31 simulation through the gmxapi Python interface.
32 This meant that restraint potentials would silently fail to be applied with
33 gmxapi versions >= 0.1.
34 Updates have been applied internally to gmxapi.
36 The gmxapi 0.2.2 Python package supports the updated GROMACS API and will
37 issue errors if a simulation attempts to bind external plugin code with
38 a compatible-but-broken API (GROMACS 2021 through 2021.2).
40 Third party code should not need to be updated, but developers will
41 note an additional "null restraint" in
42 https://gitlab.com/gromacs/gromacs/-/tree/master/python_packaging/sample_restraint
43 (for illustration and testing purposes).
45 :issue:`4078` and :issue:`4102`
47 Fixed multi-rank restarts from checkpoints written by single-rank simulations
48 """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
50 Currently a single-rank simulation never uses update groups, however a
51 multi-rank run can do so. This fix ensures that the atoms within
52 update groups always start in the same periodic image, which was not
53 guaranteed if the checkpoint was written by a single-rank simulation.
57 Fixes for ``gmx`` tools
58 ^^^^^^^^^^^^^^^^^^^^^^^
60 Fix gmx nmr -viol option
61 """"""""""""""""""""""""
63 The tool would previously fail with a cryptic error.
64 Also enforces that this option is exclusive with other analysis modes.
68 Fixed gmx dipoles -quad option
69 """"""""""""""""""""""""""""""
71 The tool now reports correct values.
75 Make sure gmx convert-tpr -until works
76 """"""""""""""""""""""""""""""""""""""
78 This got broken during reworking the internals of the tool and didn't
79 calculate the number of remaining steps correctly.
83 Fixed dihedral transition counting in gmx chi and gmx angle
84 """""""""""""""""""""""""""""""""""""""""""""""""""""""""""
86 When a trajectory of only 1 frame is passed, transition counting is
87 avoided (formerly it was attempted and crashed).
89 When a trajectory of multiple frames is passed, transition counting is
90 correct (formerly it did not take place).
92 Fixed possible crash in gmx chi histogramming
93 """""""""""""""""""""""""""""""""""""""""""""
95 Formerly an invalid reference to a temporary string was used for a
96 residue name, which might have caused a crash.
98 Fixed gmx chi -chi_prod
99 """""""""""""""""""""""
101 Formerly it could crash or produce garbage results when the number of
102 relevant dihedrals differed from the number of residues
105 Fixes that affect portability
106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
108 Check that necessary python modules are available
109 """""""""""""""""""""""""""""""""""""""""""""""""
111 The source code validation could otherwise fail a build with cryptic errors.
115 Ensure that NB-LIB and gmxapi can be build even without tests enabled
116 """""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
118 Could otherwise lead to cryptic build errors.
123 Removed performance loss in the mdrun domain decomposition
124 """"""""""""""""""""""""""""""""""""""""""""""""""""""""""
126 With 16 or more so-called PP MPI ranks, the domain decomposition
127 repartitioning could incur large performance overheads due to a sub-optimally
128 sized hash table. This has now been fixed.