This table present a readable representation of the traffic on the gmx-revision list. Since changes for both the stable and the unstable version are recorded on this list, you'll have to deduce from the date whether or not a change made it into a certain version.
| Person | Date | Summary | Description |
|---|---|---|---|
| Berk Hess | 17 Aug 2004 | eneconv crash with starting time > 0 | When the first file would start at a non-zero time eneconv would crash with a segmentation violation. This has been fixed. |
| Berk Hess | 17 Aug 2004 | Dispersion correction and virial | Fixed the virial correction which was incorrect in 3.2.0 and 3.2.1 (the pressure correction was correct). |
| Berk Hess | 17 Aug 2004 | Dispersion correction and shifted potentials | Implemented dispersion correction for shifted potentials. Improved the efficiency of the dispersion correction calculation. |
| Berk Hess | 17 Aug 2004 | Free-energy with rcoulomb > rvdw | Fixed a bug which would cause free-energy simulations to crash with a segmentation fault when rcoulomb > rvdw. |
| Berk Hess | 17 Aug 2004 | Speed up with free-energy | Free-energy runs now use much more solvent optimization. This gives a significant speed-up, especially with water. Also the neighbor-search with free-energy has been made more efficient. There is now almost no overhead for free-energy runs. |
| Berk Hess | 17 Aug 2004 | Settle stability problem in single precision | (AMBCONV) A precision problem has been fixed which could occur depending on the masses of the water atoms. With some combination of masses, for instance with the Amber forcefield produced with the AMBCONV script, the center of mass of each water molecule would shift every step by half a bit of a float. This would cause molecules to fly through the box. All forcefields supplied with Gromacs use the same water masses and have not been affected by this problem. |
| Berk Hess | 17 Aug 2004 | Xmgrace displays g_anaeig output incorrectly | Fixed incorrect scaling in xmgrace of the x-axis in files with multiple graphs produced by g_anaeig. |
| Berk Hess | 17 Aug 2004 | eneconv bug when init_step > 0 | When in any of the input files for eneconv the step number did not start at zero, which will happen when tpbconv is used in version 3.2, the averages in the output would be slightly off. This has been fixed. |
| Berk Hess | 17 Aug 2004 | PME energy accuracy in single precision | The PME and Ewald exclusion correction energy is now accumulated in a double precision variable. Previously the PME energy could have a relatively large error, 100 kJ/mol for a protein/water system, due to rounding problems. The forces and thus the sampling were correct. |
| Berk Hess | 17 Aug 2004 | Reaction-field exclusion correction | The reaction-field correction is now also applied to excluded atom pairs. The 1-4 energies are now reported without reaction-field correction. This mainly results in a shift in the Coulomb energy. There are minor changes in forces between excluded pairs that do not have 1-4 interactions. For a water box RF with a cut-off of 0.9 or 1.4 nm now gives a Coulomb energy within a few promille of PME. With old tpr files the old implementation will still be used, it can also be selected with reaction-field-old. |
| Berk Hess | 17 Aug 2004 | Free-energy bug with soft-core | Fixed a bug which was present in 3.2.0 and 3.2.1 which caused the normal LJ parameters to change after the first MD-step when soft-core potentials are used and 1-4 interactions are present. This will probably not have caused unnoticed errors as the LJ energies and dgdl changed significantly after this first step. |
| Berk Hess | 17 Aug 2004 | Time in pull output | Fixed the time in the pull output when init_t is not zero. |
| Berk Hess | 17 Aug 2004 | Pull code and freeze groups | The pull code is now aware of frozen atoms. Frozen atoms will no longer move during pulling. |
| David van der Spoel | 25 Aug 2004 | rerun + dummies bug resolved | Fixed a bug when doing a rerun with dummies. The graph would not be in sync with the coordinates since it only is created at neighborsearching time. Added an extra (conditional) mk_mshift call before constructing dummies. Added a further feature to the rerun option, that update (and shake) is only performed when velocities are present in the trajectory (i.e. never for xtc files) |
| Berk Hess | 27 Aug 2004 | Removed mdp option bd-temp | The mdp option bd-temp has been removed, its function is replaced by ref-t. Old run input files are converted correctly. |
| Berk Hess | 27 Aug 2004 | New mdp option deform | The non-equilibrium MD option deform has been introduced. It sets velocities for deformation of the (triclinic) box. It can be used to strain or shear a solid or a liquid. |
| Erik Lindahl | 29 Aug 2004 | Added the Encad and Gromos45a3 force fields | I've had Michel's Encad force fields working for several months, but finally decided to add it to the repository. This also involved added the Encad-style shift functions and water model (f3c) to be able to reproduce results exactly. While I was at it, I also added the version of Gromos45a3 that Alessandra contributed. |
| Erik Lindahl | 30 Aug 2004 | Fixed bug with vectorized 1/sqrt with IBM MASS | I've fixed two minor bugs that caused complete crashes: 1. When aligning memory it didn't work on 64-bit machines. I don't think we ever did it for any 64-bit machine, thouh. 2. There was a bug in the innerloop that caused it to crash hard when using the IBM MASS library for vectorized invsqrt. |
| David van der Spoel | 30 Aug 2004 | eneconv without time 0 | Patch for the case that the energy files did not start at time 0 Included in 3.2 patch branch and normal CVS |
| Berk Hess | 30 Aug 2004 | Replica exchange | Replica exchange has been implemented. It uses the -multi functionality of mdrun to run replicas in parallel. |
| Berk Hess | 31 Aug 2004 | Fixed bug with ewald_geometry = 3dc | Fixed bug where the ewald_geometry=3dc correction was only used when surface_epsilon was not zero, which is the default value, meaning ewald_geometry=3dc had no effect in most cases. |
| David van der Spoel | 6 Sep 2004 | fixed bug in mpich-parallel runs | Implemented a workaround for a bug in some parallel MPICH runs that never finished. The background for this was that the MPI_Finalize call did not return, and hence user jobs would not stop. |
| David van der Spoel | 6 Sep 2004 | error messages | The fatal_error function was replaced by a more flexible set of routines that give more informative error messages. |
| David van der Spoel | 7 Sep 2004 | new feature in g_analyze | Added option in g_analyze to compute an integral of e.g. a correlation function with error bars. |
| David van der Spoel | 7 Sep 2004 | tip4p.itp updated | Fixed an error which would compute intramolecular interaction in the case a flexible model was used. |
| David van der Spoel | 24 Sep 2004 | MNO solvent bug | A bug in the inner loop for special solvents where both the Coulomb and the LJ are shifted was fixed. The bug would lead to the LJ interaction being computed incorrectly, and would most often result in a crash with weird energies. |
| Erik Lindahl | 27 Sep 2004 | MNO solvent bug in 3dnow assembly loops | Fixed the previously mentioned bug in the 3DNow assembly loops too. |
| Erik Lindahl | 27 Sep 2004 | Added GROMOS96 53a5 and 53a6 force fields | The new versions of Gromos96 (contributed by Allesandra) have been added to the distribution, and the static choices in pdb2gmx. |
| David van der Spoel | 28 Sep 2004 | special solvents bug | A bug in the inner loop for special solvents where both the Coulomb and the LJ are shifted or switched was fixed. The bug would lead to the LJ interaction being computed incorrectly, and would most often result in a crash with weird energies. Coulomb with PME plus shifted/switched potential is also affected (but only for special solvents, like TIP4P water, and not for SPC(/E) or TIP3P). |
| Berk Hess | 29 Sep 2004 | neighborlist optimization | Optimal neighborlists are now generated. In previous versions pairs of 'i-atoms' without LJ and 'j-atoms' with LJ would still end up in innerloops which also calculated LJ interations. This optimization can provide a significant speedup for systems which contain many atoms with and without LJ, such as solvents (note that 3-site water models were already fully optimized). |
| Berk Hess | 4 Oct 2004 | g_cluster bugfixes | Fixed bug which caused incorrect avg rmsd for clusters and incorrect (rmsd for) middle structures. Changed name full linkage to single linkage (fixed several months ago). Fix error which almost always occured during the writing of the xpm output. Removed the -keepfree option. |
| Berk Hess | 15 Oct 2004 | trjorder pdb B-factor output | Fixed a bug which could cause trjorder to write nonsense distances in the B-factor field of pdb files. The ordering itself was correct. |
| Berk Hess | 18 Oct 2004 | g_sas, area of a subgroup | g_sas can now plot the solvent accessible of a subgroup of a molecule, for instance of one residue of a protein. |
| Berk Hess | 19 Oct 2004 | g_energy RMSD and Fluct. | Fixed g_energy RMSD and Fluct. output for energy files which do not start at step 0. Such energy files are produced when continuing runs with tpbconv. |
| Berk Hess | 21 Oct 2004 | grompp, pairtypes with gen-pairs=yes | Fixed problems which caused grompp to produce warnings and a fatal error when pairtypes where used in combination with gen-pairs=yes. |
| David | 24 Oct 2004 | pmetest | Developed a new program src/contrib/pmetest that reads a tpr file and a trajectory file and computes the longe range Coulomb interactions using the PME algorithm. Works in parallel as well. |
| Berk Hess | 27 Oct 2004 | trjconv with -pbc nojump and fitting | Fixed the problem in trjconv which caused the combination of -pbc nojump and fitting to produce 'jumped' results. |
| Berk Hess | 9 Nov 2004 | pull code umbrella sampling | Fixed problems in the pull code with umbrella sampling with multiple groups or an absolute reference. |
| Berk Hess | 15 Nov 2004 | 20 pull groups | The pull parameter file now has an extra entry "ngroups". The maximum number of pull groups has been increased from 4 to 20. |
| David | 15 Nov 2004 | segv with mdrun -h | Fixed a SEGV in mdrun when called with the -h flag |
| Berk Hess | 8 Dec 2004 | changed grompp warnings to errors | All warnings issued by grompp about incorrect topology input have been converted to errors. |
| Berk Hess | 14 Dec 2004 | Renamed dummies to virtual sites | To avoid confusion with dummy masses, the notion dummy atom in Gromacs has been renamed to virtual (interaction-)site. The topology directives "dummies?" have been renamed to "virtual_sites?". Particle type "D" has been renamed to "V". Old topology files can still be read. |
| Berk Hess | 17 Dec 2004 | FENE and quatic angle potentials | Implemented the FENE (finite extensible nonlinear elastic) bond potential. Implemented a fourth order polynomial angle potential. |
| Erik Lindahl | 27 Dec 2004 | Fixed block creation bug | When creating blocks with stupid_fill, the last group was left out. This has lead to the genbox neighborsearching neglecting the last group in the system. |
| Erik Lindahl | 27 Dec 2004 | Added TIP4P loop optimization | The nonbonded kernels and neighborsearching has been updated to allow TIP4p-specific optimization as well as SPC. All references to the MNO general solvent loops have been removed. |
| Erik Lindahl | 27 Dec 2004 | TIP4p assembly loops | Assembly loops optimized for TIP4p have been added both for SSE & SSE2. There are also new assembly loops for standard coulomb or reaction-field combined with tabulated LJ, so now _all_ interactions except for buckingham are accelerated. |
| Erik Lindahl | 27 Dec 2004 | X86-64 assembly loops | SSE (single) and SSE2 (double) assembly loops have been added for the x86-64 architecture. These loops _should_ work on both Opteron/Athlon64 and Intel EMT64, but this far they have only been tested on AMD systems. Erik |
| Erik Lindahl | 27 Dec 2004 | ia64 assembly loops | Nonbonded kernels in both single and double precision ia64 assembly have been added. The kernels use software pipelining scheduled for the instruction latencies on Itanium2. The performance will likely be _very_ bad on Itanium1, but since that generation is pretty uncommon now they are enabled by default on ia64. If instruction latencies change in future generations, the innermost loops might have to be rescheduled. |
| Erik Lindahl | 27 Dec 2004 | Multithreading for nonbonded kernels | All assembly kernels now use compare-exchange instructions for multithreading. We still don't use multiple threads by default (need to fix neighborsearching), but this means they might require a slightly newer version of the gas assembler - one that supports "lock cmpxchg" to lock the bus in SMP systems. |
| Erik Lindahl | 27 Dec 2004 | New solvent detection | The solvent detection code has been rewritten from scratch: It now finds all charge groups that fulfill the conditions for either the SPC/TIP3p nonbonded kernels, or the TIP4P kernels. Parameters are compared between groups, and only the most abundant type is set to the optimized solvent. (The previous code never checked if the parameters matched between groups). |
| Erik Lindahl | 27 Dec 2004 | Test systems for nonbonded kernels | A new test_kernels directory has been added to the "test" module in cvs. This is not run by default, but inside the directory there are two perl scripts which will run all the systems and compare the results with those of Gromacs 3.2/3.3. Note that not all kernels will be available on all architectures; on e.g. ia64 the water loops are disabled when using assembly. |
| Erik Lindahl | 27 Dec 2004 | Reverted atom order in TIP4p | The TIP4p coordinates and itp files now have the atoms in standard order: OW HW1 HW2 MW Note that this is the only order which will enable the new TIP4p-specific assembly loops. |
| Berk Hess | 5 Jan 2005 | pull code constraint distance | Added an option to the pull code parameter file to set the constraint distance, instead of taking the distance from the input coordinates (which can still be done). |
| Berk Hess | 12 Jan 2005 | g_analyze block averaging | Improved the fit of the block averaging by fitting the sqaure, and improving the weights and the initial guess. Invalid fits should not occur anymore. |
| Berk Hess | 12 Jan 2005 | Ewald and PME with epsilon_r not 1 | Ewald and PME now also work correctly with dielectric constant epsilon_r not equal to 1. |
| David van der Spoel | 23 Jan 2005 | support for dihedral principal component analysis | Added support for dihedral PCA to g_angle. |
| Erik Lindahl | 2 Feb 2005 | Fixed nonbonded parameter bug | Hi, The code committed by Berk 2-3 weeks ago to enable sigma/epsilon values for individual nonbonded parameters caused a bug for force fields that uses scaling for 1,4 parameters. Previously, the parameters were first converted to c6/c12 and then everything was scaled. In the new version the are kept longer as sigma/epsilon, but this resulted in both the sigma and epsilon values being scaled. I've fixed it by checking for the combination rule, and only scale epsilon values when sigma/epsilon parameters are read. |
| Erik Lindahl | 2 Feb 2005 | LAPACK, BLAS, and ARPACK stuff | I've added a subset of the LAPACK and BLAS libraries to gmxlib, written in portable C. The user can optionally override it and use vendor-provided BLAS/LAPACK libraries instead. The ql77 diagonalizer code from EISPACK has been replaced with much faster LAPACK versions throughout the code. I've also added a small number of routines from ARPACK to support Arnoldi iterations for general eigenvalue problems. |
| Erik Lindahl | 2 Feb 2005 | Sparse matrix representation for normal modes | I've implemented sparse matrix representation both for the Hessian calculation and the actual diagonalization. The sparse representation can be used whenever cut-offs are used (instead of e.g. PME), and makes it possible to determine an arbitrary number of low eigenvectors for very large systems - the memory requirements are linear in system size! To support this I have changed the mtx format to be either full or sparse. The selection of full/sparse format is completely automatic and does not require any user intervention. |
| Erik Lindahl | 3 Feb 2005 | Eigenvector format & eigenfrequencies | The eigenvalues are now stored in the eigenvector files, so you don't have to provide the eigenval.xvg file separately. g_nmeig now also writes a separate file with eigenfrequencies correctly scaled to cm-1. |
| David van der Spoel | 8 Feb 2005 | PME-User support | Added support for a combination of PME and User determined coulomb potentials. The user has to take care that the combination of Coulomb and standard PME makes sense. |
| Berk Hess | 8 Feb 2005 | tip5p atom order | The tip5p atom order has been changed to the 'normal' order. Because of more optimization in the neighbor search the normal Gromacs innerloops now run tip5p much faster than the MNO solvent optimized loops. |
| Berk Hess | 21 Feb 2005 | g_densmap | Added the tool g_densmap which calculates 2D number-density maps. The map can be planar or axial-radial. Also xpm2ps has been adapted to correctly plot ticks for spatial instead of time or matrix element axes. |
| Berk Hess | 1 Mar 2005 | test particle insertion | Test particle insertion has been implemented. It is very efficient as it only calculates the interaction energy (no forces) of the test particle with the rest of the system. |
| Berk Hess | 1 Mar 2005 | reaction-field NEC | Renamed reaction-field-old (the reaction-field implementation of Gromacs version 3.2 and older) to reaction-field-nec: no exclusion correction, which gives a more useful description. |
| Berk Hess | 2 Mar 2005 | Pull code AFM and umbrella fixes | The contribution of the AFM and umbrella forces to the virial are now taken into account. The reaction forces on the reference group for AFM and umbrella pulling have now been added. |
| Berk Hess | 2 Mar 2005 | Pull code AFM and umbrella fixes | The contribution of the AFM and umbrella forces to the virial are now taken into account. The reaction forces on the reference group for AFM and umbrella pulling have now been added. |
| Berk Hess | 7 Mar 2005 | pbc=full now works with constraints | Simulations with pbc=full (for periodic molecules) can now be performed properly with constraints. Only LINCS is supported, not SHAKE. |
| Berk Hess | 11 Mar 2005 | Pull code constraint virial | With constraint pulling the constraint virial (and thus the total virial) is now also correct when the constraints between two pull groups passes over a box boundary. |
| Berk Hess | 11 Mar 2005 | Virtual sites and virial | Fixed (relatively) small jumps in the virial which could occur when the virtual site and/or atoms involved in the construction of the virtual site lie over a box boundary. |
| David van der Spoel | 11 Mar 2005 | XTC - random access | Added support for random access to xtc files, allowing for e.g. fast forward. Implemented by Filipe Maia. |
| David | 1 Apr 2005 | changed format of hdb and tdb files | The format of hydrogen database and termini database files has been changed in order to allow for user-specified names on generated hydrogens. |
| David | 2 Apr 2005 | fixed bug in g_msd | A bug was fixed that gave a SEGV when the -mol option was used in combination with the -trestart option when trestart was < trajectory length, i.e. when there were actual restarts. |
| David | 3 Apr 2005 | new features in g_hbond | Added functionality to compute hydrogen bond lifetimes and fixed SEGV bug in the hbmap option. |
| Berk Hess | 7 Apr 2005 | Temperature coupling | The Berendsen and Nose-Hoover temperature coupling now use the temperature at the half step. Nose-Hoover now uses a reversible integrator: Holian et al. Phys Rev E 52(3) : 2338, 1995 |
| Berk Hess | 7 Apr 2005 | Kinetic energy | The kinetic energy in Gromacs is now determined as the average of the kinetic energy of the velocities at the half steps around the full step. This is four times more accurate than the old method which used the full step particle velocities which were obtained as the averages of the half steps. |
| Berk Hess | 12 Apr 2005 | trjconv -center | Changed the -center option of trjconv to an enum which has the values no, tric, rect and zero so the user can define where the box should be centered. |
| Berk Hess | 12 Apr 2005 | distances and pbc in tools | The pbc_dx distance function which is used in many analysis tools now corrects for mutiple box-vector shifts and thus always produces correct distances. For normal mdrun output there was never a problem. But problems could occur with manipulated trajectories or output from other software packages. |
| Berk Hess | 20 Apr 2005 | skipping frames in double precision | When using a double precision compilation frame skipping with a single precision trajectory is now performed with a single precision margin instead of double precision. |
| Berk Hess | 21 Apr 2005 | dispersion correction in large systems | Fixed bug in the dispersion correction for systems with more than 46340 atoms. Reduced time for the average C6 calculation for dispersion correction in large systems from minutes to seconds. |
| Berk Hess | 22 Apr 2005 | g_hbond reduced memory usage | g_hbond used to always allocate an amount of memory proportional to the total number of donors times the total number of acceptors. This made analysis of more than a few thousand water molecules impossible. For half of the output options this memory allocation has been removed. |
| David van der Spoel | 7 May 2005 | OPLS parameters | Modified incorrect SIGMA values in ffoplsaanb.itp for united atom types. These types are not generally used and hence old simulations are correct. |
| Berk Hess | 19 May 2005 | implemented mulitple user tables | Seperate user tables can now be used for selected pairs of energy groups. This is set in the mdp file with the option energygrp_table. |
| Berk Hess | 19 May 2005 | PME-User | With coulombtype=PME-User mdrun now subtracts the PME mesh contribution from the user tables. |
| Berk Hess | 24 May 2005 | corrected averages and fluctuations in md.log | Since version 3.2 incorrect averages and fluctuations would be reported at the end of the md.log for simulations which were continued with tpbconv. This has been corrected. |
| Berk Hess | 1 Jun 2005 | minimization with constraints | During constraining in energy minimization atoms were moved inversly proportional to their mass. This has been fixed. Steepest descents with constraints now converges properly. |
| David van der Spoel | 2 Jun 2005 | Added new filetype .pqr | Input files for the MEAD electrostatics program can now be produced directly by editconf from a tpr file. |
| Berk Hess | 6 Jun 2005 | generalized reaction-field | Fixed an error in the generalized reaction-field. There was an incorrect factor 2 in the (kappa r_c)^2 terms in both the code and the manual. (The normal reaction-field was not affected) |
| Berk Hess | 6 Jun 2005 | epsilon_rf | Introduced a seperate mdp parameter for the relative dielectric constant of the reaction field: epsilon_rf. This avoids confusion and allows the use of a different epsilon_r and epsilon_rf for reaction field. Old mdp and run input files are still read correctly. |
| David van der Spoel | 9 Jun 2005 | New option -[no]xvgr to analysis tools | Added a new option to analysis tools to prevent the output of xvgr specific info like legends, title, axis title etc. |
| David | 11 Jun 2005 | ndx file support in gmxcheck | Added feature to analyze the content of an index file in the gmxcheck program. |
| Berk Hess | 17 Jun 2005 | special pair interaction functions | Up to now when shift, switch or user functions were used for the nonbonded interactions, these functions were also used for the pair interactions. Now with shift and switch the normal cut-off interactions are used for the pairs. For user functions a second table file is now read by mdrun when pair interactions are present. Berk |
| David | 5 Aug 2005 | Constant pressure REMD | Added support for constant pressure replica exchange MD. Can be turned off using a command line argument for backwards compatibility. |
| Berk Hess | 29 Aug 2005 | finalization with mdrun -multi | mdrun -multi (and replica exchange runs) would not call MPI_Finalize before finishing. In some cases this would lead to truncated or missing output files. This has been fixed. |
| Berk Hess | 30 Aug 2005 | soft-core lambda power | Implemented lambda power of 1 (now default) for the soft-core interactions in addition to the old lambda power of 2. |
| Erik Lindahl | 30 Aug 2005 | Fixed table bug for switch/shift interactions | The bCoulomb[] field was initialized incorrectly in the CVS version, which lead to coulomb cutoffs being assigned to LJ and vice versa for switch/shift interactions. This does not affect 3.2.1. |
| Erik Lindahl | 30 Aug 2005 | Fixed logfile runaway | mdrun now stops with an error message after 10,000 LINCS warnings. |
| Berk Hess | 30 Aug 2005 | PBC problem with special topologies | A PBC problem that has been in the code for many years has been fixed. This would only occur for special topologies, where not all atoms in a molecule are be connected by bonds, but they would still interact via other interactions, for instance distance restraints. The problem could also occur when running on multiple cpu's without constraints, which could cause molecules to be split over cpu's. |
| Erik Lindahl | 31 Aug 2005 | Fixed LJ bug in x86-64 loops | A rare bug in the LJ parameter assignment caused small potential errors in the updated x86-64 loops introduced last week. |
| Erik Lindahl | 31 Aug 2005 | EM64t assembly loops | The x86-64 loops have been ported to compile and work on em64t (Intel 64-bit extensions) architectures. |
| Erik Lindahl | 31 Aug 2005 | Fixed segfault on ia32 SSE2 | Fixed a stack error that caused segmentation faults on some double precision ia32 nonbonded loops. |
| David | 2 Sep 2005 | Fixed cool quotes | Cool quotes can now be turned off by setting an environment variable GMX_NO_QUOTES or by removing the gurgle.dat and/or bromacs.dat files. |
| David | 5 Sep 2005 | Header in xvg files | The header in xvg files now prints the command line used to generate the output file. |
| David | 24 Sep 2005 | Moved system splitting code to mdrun | The code that splits a simulation system into bits for a parallel calculation has been moved from grompp to mdrun. One does not need special multiprocessor topologies anymore. Shuffling has not yet been added. [4.0 only] |
| David van der Spoel | 4 Oct 2005 | g_hbond bug fixes and new options | Fixed horrible bug in g_hbond which would completely mess up the hbmap option. Added new feature to compute number of HB per hydrogen for comparison with Raman spectroscopy. In 3.3rc and 4.0. |
| Erik Lindahl | 5 Oct 2005 | Builting parallel FFTs | Added code to perform our own parallel FFTs using single-CPU 1d and 2d transforms. |
| Erik Lindahl | 5 Oct 2005 | More FFT libraries | Since we now have our own parallel FFTs we are no longer dependent on FFTW-2.1.x. The release-3-3-patches branch now contains wrappers for fftw2, fftw3, Intel MKL, and a builtin (slower) version of FFTPACK. Use the configuration switch --with-fft=XXX to select which implementation to use. /Erik |
| Berk Hess | 5 Oct 2005 | fixed error in dihedral angle free-energy | contribution In free energy simulations with perturbed proper and improper dihedrals a degrees to radians conversion factor had been missing in the contribution to dV/dlambda of the change in equilibrium angle. Note that the force and potential were correct, as well as the contribution of the change in force constant. |
| Erik Lindahl | 9 Oct 2005 | Modified default cpp path | The default preprocessor is now /usr/bin/cpp , which seems to be more common on modern systems. |
| Erik Lindahl | 11 Oct 2005 | Using gromacs subdirectories in share & include | We now use the standard pkgdatadir and pkgincludedirs by default, which means actual files are installed in share/gromacs/... include/gromacs/... even if the top-level prefix contains gromacs (e.g. /usr/local/gromacs). Previously the configure script contained an ugly hack where the subdirectory was added if the prefix did not contain gromacs or gmx, but this doesn't work with modern versions of rpmbuild. (It is also better to standardize on the same solution as the rest of the world rather than risk overwriting stuff in /usr/include...) |
| Erik Lindahl | 12 Oct 2005 | Fixed incorrect dvdl term from R-B dihedrals | Fixed reading of bonded types with explicitly specified parameters, so that B-state parameters are automatically set from the A-state if the are not given. This applies e.g. to the OPLS torsions that were specified with #defines, and gave rise to incorrect contributions to dvdl. |
| Berk Hess | 21 Oct 2005 | frame counters | The frame counters for reading trajectory and energy files now progessively print each, each 10th, each 100th and every 1000th frame. |
| Berk Hess | 25 Oct 2005 | trjcat dt problems | Fixed a problem in trjcat which could cause missing frames in the output when the timestep for the different input trajectories were not identical. |
| Berk Hess | 26 Oct 2005 | gmxcheck -ab | Added -ab option to gmxcheck which compare the A and B state topology from one run input file for free energy simulations. |
| Berk Hess | 3 Nov 2005 | g_rdf normalization | g_rdf did not take self exclusions into account for the normaliztion of rdf's when the reference and analysis group overlapped. For an rdf of a group of N particles with itself the rdf needs to be scaled up by N/(N-1). This correction has been added now. |
| Berk Hess | 10 Nov 2005 | grompp default bonded parameters for perturbed atoms | For free-energy sytems where atomtypes are different in the A and B state and default parameters for bonded interactions were searched for both the A and B state and could be found for the A state, but could not be found for the B-state, the parameters for both the A and B state would silently be set to zero. Now the A-state parameters are copied to be B-state and a warning is issued. |
| Berk Hess | 6 Dec 2005 | SD and pull code constraints | Fixed the combination of SD and pull code constraints. The velocities and pull forces would be far too large (could not have caused silent errors). |
| Berk Hess | 15 Dec 2005 | trjconv -split file names | Fixed a bug which caused trjconv -split to produce filename like: (null)_0_(null) |
| Berk Hess | 15 Dec 2005 | Frame selection problems with -tu | Fixed rounding problems with the time unit option for analysis tools -tu was set to something different than the default ps. This problem could cause frame to be skipped or selected incorrectly. |
| David van der Spoel | 29 Dec 2005 | g_energy selection | The selection mechanism in g_energy was changed to accept strings rather than numbers. |
| David van der Spoel | 2 Feb 2006 | removed disco and friends | The disco and cdist programs, an implementation in GROMACS of the CONCOORD program suite by Bert de Groot have been removed. Users are advised to use CONCOORD instead. [3.3 and 4.0] |
| Berk Hess | 3 Feb 2006 | pair types for free-energy decoupling | Added new pair type 2, which is identical to the standard type 1, except that it only the A-state charges and LJ parameters. Added new pair type 3, which use the non-scaled charges and the normal non-bonded parameters, also only from the A-state. |
| Berk Hess | 10 Feb 2006 | fixed bug with angle restraints and free energy | The angle restraint parameters were not stored correctly in the tpx file. This would cause the angle restraint parameters to be zero in the B-state. This only had effect on the angle restraints in free-energy simulations. This bug has also been fixed for the release-3-3-patches branch, but there the B-state parameters will always be copied from the A-state. |
| Berk Hess | 16 Feb 2006 | grompp check on T-coupling groups | grompp will now generate a fatal error instead of a rest group when atoms are not part of any of the T-coupling groups. |
| David van der Spoel | 22 Feb 2006 | Fixed bug in genbox | Using the -shell in genbox works again. |
| Erik Lindahl | 6 Mar 2006 | Tarball filenames > 99 chars | Added support for ustar format in the tarballs created by "make dist", in order to support total filename paths above 99 characters. Unfortunately this means developers now need automake version 1.9 or later. Standard on linux, but on OS X you'll need to install a newer automake than the default 1.6. |
| Erik Lindahl | 6 Mar 2006 | OS X/intel SSE assembly loops | Added support for both Intel and AT/T syntax in the assembly loops. The latter have auto-translated from the first ones using intel2gas. This was required for SSE assembly loops on newer Intel Macs, which now compiles correctly out-of-the-box. |
| David van der Spoel | 8 Mar 2006 | New genbox option | Added option -maxsol to specify max number of solvent molecules to be added, provided the box is large enough. |
| David van der Spoel | 9 Mar 2006 | New genion option | Added options to generate ions at a certain concentration and to neutralize the system automatically (or both). |
| Erik Lindahl | 9 Mar 2006 | Fixed internal XDR on big-endian 64-bit archs with | long>int Normally the system libraries are used everywhere except on windows (which is a 32-bit platform), so this bug is unlikely to have caused any problems. |
| Erik Lindahl | 13 Mar 2006 | Fixed NPT REMD pressure test initialization bug | Fixed a bug where pressure tests for initiating NPT replica-exchange was done outside the loop over nodes, which causes the init to fail. This is a wonderful example why you should always use braces even for single-statement loops... /Erik |
| Erik Lindahl | 15 Mar 2006 | QM/MM manual and mdp option documentation | Added to the "manual" CVS module. |
| hessb@mpip-mainz.mpg.de | 21 Mar 2006 | soft-core lambda power check | Added a check on the soft-core power being 1 or 2 when free-energy and soft-coring is used. Set the default value of sc_power to 0, so users can not accidentally use the incorrec lambda power. This change is also in the 3.3.1 release. |
| David van der Spoel | 29 Mar 2006 | New option -pbc for g_angle and g_chi | Added new (default) flag -pbc to g_angle and g_chi which computes dihedrals modulo 2 Pi. In the old g_angle and average dihedral would be plotted with -ov which was meaningless since it incorporated a random (integer) times 180 before averaging. |
| Berk Hess | 4 Apr 2006 | check for rcoulomb=rlist with PME, Ewald and PPPM | grompp now check that rcoulomb is equal to rlist with PME, Ewald and PPPM. |
| Erik Lindahl | 4 Apr 2006 | Changed PME spacing for triclinic cells | PME grid dimensions are now calculated from spacing along (non- orthogonal) triclinic box vectors rather than orthogonal cartesian distances. This improves the accuracy of free energy calculations with pme_order=4. Note that the grid calculation occurs in grompp, so you will need to regenerate the tpr files for it to have effect. |
| David van der Spoel | 16 Apr 2006 | removal of center of mass motion | Fixed bug where the combination of comm_mode = Angular and nstcomm = -1 (the old way of specifying removal of angular momentum) would lead to turning off Angular, and only doing Linear removal of center of mass motion. |
| David van der Spoel | 17 Apr 2006 | charge group bug in pdb2gmx | Fixed bug in tdb files where pdb2gmx would put the Ca in the N-terminal residue in another charge group than the NH3 group leading to two partially charged groups. |
| David van der Spoel | 17 Apr 2006 | special bonds in pdb2gmx | Fixed bug in pdb2gmx where special bonds (e.g. Cys-Cys or HEME-FE) were not generated anymore (they were in 3.2.1). |
| David van der Spoel | 5 May 2006 | x2top additions | Added feature in x2top to write an rtp entry |
| David van der Spoel | 19 Jul 2006 | New options to configure | Configure now checks for the presence of the GNU scientific library and the XML library (libxml2). Both of these are optional, and can be turned off on the command line. |
| Berk Hess | 20 Jul 2006 | PME-Switch | PME-Switch electrostatics has been implemented. This is a combination of PME and Switch and is mainly useful for testing energy conservation. Now both PME-Switch and PME-User allow for rcoulomb<=rlist. |
| David van der Spoel | 21 Jul 2006 | New volume option to g_sas | Added a new option to g_sas in order to compute the volume and density of a sample. The volume was already computed by the double-cube lattice method but we never used the result. Option is -tv. |
| Berk Hess | 14 Aug 2006 | GMXRC update problem | Fixed GMXRC files not being updated when running configure multiple times with different --prefix options. |
| Berk Hess | 14 Aug 2006 | G53a5/6 rtp HISB impoper dihedral problem | Fixed an incorrect atom list for an improper for HISB in ffG53a5.rtp and ffG53a6.rtp: CE1 ND1 NE2 HD2 gi_1 should be: CE1 ND1 NE2 HE1 gi_1 This would cause puckering in the ring. |
| David van der Spoel | 17 Aug 2006 | g_energy fixes | Restored option to use numbers to select energy terms. One can now use either numbers or strings or a mix and terminate the list with a zero or a newline or EOF. |
| David van der Spoel | 17 Aug 2006 | Bug in minimization with long range | Fixed bug where the long range forces where discarded during energy minization, normal mode analysis and TPI. |
| Berk Hess | 24 Aug 2006 | distance restraints with multiple molecules | Fixed a bug where distance restraints involving multiple atoms and which were not ensemble averaged were split into separate restraints for all but the first copy of a molecule. Thus this bug only occured with multiple copies of the same molecule. |
| David van der Spoel | 24 Aug 2006 | Fixed bug in dihedral analysis tools | A bug in dihedral analysis tools (g_chi and g_angle) was fixed that yielded incorrect answers for intermolecular dihedrals and angles. Although there is no perfect solution to this problem angles dihedrals are now computed by taking the nearest images for computing the distance vectors. |
| David van der Spoel | 29 Aug 2006 | Regression analysis in g_analyze | Added an option to do linear regression analysis in g_analyze with optional error bars (sigma) on the y value. |
| David van der Spoel | 30 Aug 2006 | demuxing in trjcat | Fixed the demultiplexing algorithm in trjcat. It works in conjunction with the script demux.pl (in src/contrib/scripts) that derives the exchange data from a standard unaltered md0.log file and generates an xvg file (replica_index.xvg) that should be passed as input to trjcat. |
| Berk Hess | 14 Sep 2006 | trjconv -pbc | Several trjconv -pbc options have been added and renamed. Now mol, res and atom put the center of mass of molecules residues and atoms respectively in the box. The -pbc whole options now only makes molecules whole and does not put things in the box. Especially the -mol option is an improvement, as it uses on the center of mass, instead of the first atom of a molecule as with the old whole option. |
| Berk Hess | 5 Oct 2006 | interaction table spline change | The splines interpolation for tabulated interactions (both standard and user tables) has been changed from V and V'' to V and F=-V' matching. This means that the forces are now continuous instead of V''. This increases the accuracy of tabulated potentials and forces by a factor of 5. The user tables should now contain V and F. |
| Berk Hess | 16 Oct 2006 | velocity constraining | Up to now Gromacs constrained the velocities by doing: v_(n+1/2) = (x_(n+1) - x_(n))/dt This introduces rounding noise on v that increases with decreasing timestep. Now the velocities are corrected using the constraint lagrange multipliers. The rounding noise still increases with decreasing timestep, however the energy drift has decreased by roughly an order of magnitude. |
| Berk Hess | 8 Dec 2006 | tabulated bonded interactions | Implemented tabulated bond, angle and dihedral potentials. These potentials use the same type of cubic spline interpolated tables as the user non-bonded interactions. |
| David van der Spoel | 10 Jan 2007 | New programs g_sdf & g_spatial | Two new programs for computing distribution functions contributed by users have been added to the repository for 3.3 and 4.0. |
| Berk Hess | 17 Jan 2007 | g_dipoles with charged molecules | g_dipoles can now handle charged molecules. For molecules with net charge, the net charge is subtracted at the center of mass of the molecule. |
| Berk Hess | 17 Jan 2007 | fixed triclinic correction vectors bug | For some triclinic boxes Gromacs programs would issue a warning about too many trilinic correction vectors. This could lead to incorrect distances for particles that are far apart. This bug has been fixed. |
| Berk Hess | 13 Feb 2007 | g_rdf normalization | Removed the density correction factor for self exclusions introduced in 3.3.1, as the proper definition of the pair correlation function does not contain this factor. This correction would scale the rdf for a set of N particles with itself by N/(N-1). |
| Berk Hess | 13 Feb 2007 | pbc=xy | Implemented 2D boundary conditions: pbc=xy, i.e. no periodic boundary in the z direction. |
| Berk Hess | 13 Feb 2007 | wall potentials | Implemented several wall potentials at z=0 and optionally at z=z_box, to be used with pbc=xy. |
| Berk Hess | 13 Feb 2007 | renamed unconstrained_start to continuation | The mdp option unconstrained_start has been renamed to continuation. This should be less confusing and also reflects that it now not only turns of constraints for the starting configuration, but also for instance the shell optimization. |
| David van der Spoel | 21 Feb 2007 | Insert molecules in genbox | Fixed bug in inserting molecules (-ci flag) into a triclinic box. The distance check was done on a rectangular box even for a triclinic one leading to obvious problems. Also removed the bug notion that only one molecule at a time can be inserted. |
| David van der Spoel | 20 Mar 2007 | Bugfix in g_mdmat | Fixed a bug in g_mdmat where the average mdmatrix was not written, but rather the last instantaneous one. This means that average matrices generated by the program up until now were incorrect. |
| David van der Spoel | 5 Apr 2007 | New option to g_disre | An option to dump a matrix corresponding to time averaged violations per residue has been added to the g_disre program. |
| David van der Spoel | 7 Apr 2007 | moved genpr to genrestr | Added code to generate a distance restraint matrix to the genpr program and renamed it to genrestr to make clear that the functionality has been extended. A distance restraint matrix allows to have a mobile and semi-rigid molecule, and should not cause problems with pressure coupling. |
| Berk Hess | 11 Apr 2007 | tpbconv and missing energy frames | tpbconv used to read until the last trr frame (unless -time was set). If -e was set and the energy file did not contain the corresponding frame, for instance because the energy file missed the last part due to buffering, tpbconv would not write a tpr file. This has been fixed by simutaneously checking for trr and energy frames. |
| Berk Hess | 18 Apr 2007 | energy conservation with Nose-Hoover | With an NVT ensemble produced by the Nose-Hoover thermostat, mdrun now writes out the conserved energy quantity. Note that the state of this quantity is not stored, and therefore there will be jumps in this quantity between continuated runs. |
| Berk Hess | 24 Apr 2007 | mdrun TERM signal handling with init_step | When mdrun received a term signal, it would not stop at the next step, but init_step steps later. This has been fixed. Also the environment variables GMX_NO_TERM and GMX_NO_USR1 have been added to turn off the signal handling of mdrun. |
| Berk Hess | 2 May 2007 | trjconv xy only fit | Added option for xy only translational and rotational fit for trjconv. |
| Berk Hess | 7 May 2007 | reaction-field potential shift with epsilon_rf=0 | With epsilon_rf=0, meaning an infinite dielectric constant, the constant term in the reaction field correction c_rf was set to zero instead of to 1.5/rc. This has been corrected. This bug only affected the potential (and with charge groups with net charge zero it would even cancel). The forces and sampling were unaffected. |
| Berk Hess | 8 May 2007 | Reaction-Field-zero | Added coulombtype Reaction-Field-zero, which is zero beyond rcoulomb and rcoulomb can be smaller than rlist. Can only be used with epsilon_rf=0 (inifinity). This provides proper energy conservation with reaction field, but is significantly slower than normal reaction field, due to rlist > rcoulomb and table lookup. |
| hessb@mpip-mainz.mpg.de | 16 May 2007 | g_analyze integral error estimate | Fixed bugs in the error estimate of the integral for g_analyze -xydy -integrate The error estimate was the average variance divided by n-1. This had no meaning. Now it is the proper standard error estimate, assuming that all the errors are independent. It now also works correctly for unevenly spaced x values. |
| David van der Spoel | 6 Jun 2007 | Bug fixed when splitting molecules over processors | Fixed bug 109 which gave spurious forces due to splitting molecules over processors and not taking exclusion forces into account duly. The same bug was present in the reaction-field code and the shift code. In practice this was not a problem if you use constraint = all-bonds because that prevents splitting molecules over processors. |
| David van der Spoel | 19 Jun 2007 | entropy flag to g_anaeig | Added a flag -entropy to g_anaeig that computes the entropy according to the Quasiharmonic approximation or Schlitters approximation. |
| Berk Hess | 28 Jun 2007 | FENE bug | Bugs have been fixed in the FENE potential and force. The force was so wrong that this would not have caused silent errors. |
| hessb@mpip-mainz.mpg.de | 3 Jul 2007 | complete rewrite of the pull code | The COM pull code has been completely rewritten. All parameters have been moved to the mdp parameters (see mdp_opt.html). There is a new option for constant force pulling. There are some new geometry options. For pulling with an umbrella or linear potential, the potential is now written to the energy file and it is added to the total energy. |
| Berk Hess | 13 Jul 2007 | single log file with mdrun | When running mdrun is parallel now only a single log file (md.log) is written. Only when the -seppot option is used multiple log files are written. |
| David van der Spoel | 21 Jul 2007 | splitting topologies & maxwarning | Made splitting topologies work with constraints=h-bonds or constraints=none. Set maxwarning in grompp to 0 by default. |
| Berk Hess | 10 Aug 2007 | LINCS improvement with angle constraints | LINCS now adds an extra lincs_order terms to the matrix expansion, only for the couplings between constraints that occur in triangles of constraints. This improves the accuracy of isolated angle constraints, especially COH angle constraints that occur when using v-sites. It would also help when one would like to use a rigid tri-atomic solvent in combination with for instance a protein. |
| Berk Hess | 31 Aug 2007 | fixed hang of pdb2gmx on certain rtp files | Removed an infinite loop in pdb2gmx that occurred when rtp files ended on a bonded directive. This problem would cause pdb2gmx to eat all the memory. |
| David van der Spoel | 31 Aug 2007 | cpp replacement for grompp implemented | This code replaces the call to an external preprocessor for topologies. It does not write any temporary files anymore, unless explicitly requested using the -pp option. The topology produced with the -pp option yields identical tpr files as the original. |
| Berk Hess | 5 Sep 2007 | reversed parameter order for tabulated bondeds | The order of the parameters for the tabulated bonded potentials is now: table index, k. |
| Berk Hess | 6 Sep 2007 | free energy bonded parameters | Allowed B-state parameters in bonded type definitions (only for 4.0). Added an explanatory table for this to the manual. Changed the note on copying A parameters on the bonded interaction definition line to the B-state into a warning, which is now only issued when the B-state atom types are not identical to the A-state atom types (for 4.0 and 3.3.2). |
| Berk Hess | 13 Sep 2007 | automated neighbor list updates | Which switched and/or shifted potentials the neighbor list update frequency can now be determined dynamically by setting nstlist=-1. The neighbor list is then only updated when particles moved more than half the ns buffer size from the charge group center of mass as determined at the previous neighbor search. |
| David van der Spoel | 15 Sep 2007 | New test set | A new test has been implemented in CVS. It works with 3.3.2 only at the moment. We should probably make versions like with the code itself. More info on the wiki: http://wiki.gromacs.org/index.php/Test-set |
| Berk Hess | 17 Sep 2007 | g_rdf COM/COG | Implemented options for molecule and residue center of mass and geometry RDF's. Fixed the -com option to really use the COM of the first group instead of the center of geometry. |
| Berk Hess | 4 Oct 2007 | fixed double precision / xtc writing issue | Fixed the box and time variables being converted from double precision to single precision when writing an xtc frame with double precision code. This could cause crashes in parallel double precision runs with range checking error messages during neighbor search. |
| Berk Hess | 17 Oct 2007 | mdrun deform bug | Fixed a bug in the deform option when the z/y component was non-zero and the y and z box sizes were different. |
| hessb@mpip-mainz.mpg.de | 14 Nov 2007 | PME free energy bug fixed | With PME and free energy, for particles that had zero charge in the A-state and non-zero charge in the B-state the splines would not be constructed. This would lead to incorrect PME mesh contributions for such particles. In most cases the A-state particles would have charge and the B-state not, in which case the mesh contribution was correct. Even for the incorrect case the direct space contribution would be usually be larger than the incorrect mesh part. This has been fixed for Gromacs 3.3.3 and 4.0. |
| Berk Hess | 18 Dec 2007 | invalid mdp values | A check has been added for mdp integer and real values. When the supplied value is not an integer or a real an error is now generated. This check has been added for Gromacs 3.3 and 4.0. |
| Berk Hess | 3 Jan 2008 | COM pull code integration | The COM pull code has now been fully integrated into the force evaluation and constraint calls. This mean that pulling now also works with all minimizers and with normal mode analysis. |
| Berk Hess | 3 Jan 2008 | Additional stochastic dynamics algorithm | In addition to the 2-step third order leap-frog algorithm for stochastic dynamics integration, there is now a much simpler and computationally cheaper 1-step algorithm. This algorithm is chosen automatically when the relative difference with the 2-step scheme is smaller than 10^-5. A note on this is printed in the log file. |
| Berk Hess | 10 Jan 2008 | constraints in energy minimization | Constraints can now not only be used with steepest descent minimization, but also with CG and L-BFGS. The force is now constrained in a more accurate way, but this change means that SHAKE can no longer be used with energy minimization (use LINCS). A force constraining equivalent for SETTLE has been implemented. |
| Berk Hess | 11 Jan 2008 | two stochastic dynamics integrators | The new single-step stochastic dynamics integrators is now a separate integrator with name sd. The old, more accurate two-step stochastic dynamics integrator is now called sd2. Old tpr files used with a new mdrun binary will use integrator sd2. |
| Berk Hess | 21 Jan 2008 | grompp and large charge group | grompp now (for 3.3.3 and 4.0) prints a warning when there are charge groups of more than 10 atoms. Hints are given about appropriate charge group compostion. |
| Berk Hess | 28 Jan 2008 | analysis tools -tu option | With Gromacs 3.3, 3.3.1 and 3.3.2, the -tu option for analysis tools did not set the time conversion for the -b, -e and -dt options. This has been fixed. |
| hessb@mpip-mainz.mpg.de | 2 Feb 2008 | constraint dV/dlambda print | With free-energy simulations and the mdrun -sepdvdl or -seppot option one contribution to dV/dlambda was missing in the log file prints, that of the constraints. This has been added for releases 3.3.3 and 4.0. |
| David van der Spoel | 3 Feb 2008 | Removed -check14 flag in grompp | The check14 flag could be used to remove 1-4 interactions between atom pairs in case there was no Lennard Jones interaction. The algorithm did not check for Coulomb interactions, and when applied to a protein gave wrong energies. Therefore it was removed. If this kind of functionality is needed in the future it will be implemented in pdb2gmx (or its successor). |
| Berk Hess | 13 Feb 2008 | box distortion with pressure coupling | In all Gromacs versions before 4 the box shape would "diffuse" over time when using (semi)isotropic pressure coupling. This issue (only noticable for very long simulations) has been fixed for version 4.0. |
| David van der Spoel | 28 Feb 2008 | Editconf fix | Fixed bug which prevented editconf from writing Mead/Grasp files. |
| David van der Spoel | 28 Feb 2008 | RIP Motif | Finally killed the Motif interface to all programs (CVS head only). |
| hessb@mpip-mainz.mpg.de | 28 Feb 2008 | extreme triclinic box problems | A bug has been fixed in Gromacs 3 for extreme triclinic boxes. When the box vector component combination a_x - |b_x| - |c_x| < cutoff, a very small number of interactions between atoms in two opposite box corners could be missing. This has been fixed. For common triclinic boxes, such as rhombic dodecahedrons this situation would never occur. |