Run all tests, solved merge problems and updated convergence criteria
- Re-run all tests.
This showed a couple of interesting things.
1. There was a mistake in the compute_dm code after the PEXSI merge when spin-orbit coupling was introduced 2. I have removed Hprev as it essentially is the same as Hold. Either way it shouldn't produce a huge difference in the tracking of the dEbs, etc. (after all they are not physically used other than for convergence criteria) 3. The change to SCF.Mix Hamiltonian resulted in a huge number of changes in the output. This is because the first step prints out the energies at INIT. However, the Hamiltonian is different because it is initialized after the compute_dm step.
- Changed the logic in convergence criteria. Now the convergence criterias are additive and may be fully controlled. However, at least one convergence criteria must be used.
Now the default convergence criteria is both the Hamiltonian and the density matrix.
This is updated in the manual and the compatibility note.
- Initially I thought the above differences in the energies was due to inconsistencies after r538. Hence I have created some simple routines in the m_energies.f90 code which updates a selected few of the energies. I think this should be adopted in the future to ensure that all calls to update energies are consistent. This will make changes to energy calculations less error-prone.
- Implemented the spin-type in the following routines: compute_dm final_H_f_stress state_init
- Changed the m_compute_max_diff to an interface code with appropriate size calculations. There was no reason for explicitly using the sparse pattern.
- Fixed a bug in the molecularmechanics code (introduced by Nick r542)
- Added a cyclediffs.sh script which loops on OUT.diffs files and it lets one easily cycle the diffs, simply do:
cd Tests make check ./cyclediffs.sh
and answer all the questions. Basically it makes deletes OUT.diffs which you have agreed isn't really a change.
- The tests may now be runned via:
make MPI="mpirun -np 4"
which then uses the default SIESTA location.
Currently the script checks whether mpirun/mpiexec is in SIESTA variable, and if so, does not use MPI variable. This makes it easier to decide on the number of cores without writing the full path.