lp:~nickpapior/siesta/4.0-ipi
- Get this branch:
- bzr branch lp:~nickpapior/siesta/4.0-ipi
Branch merges
- Alberto Garcia: Approve
-
Diff: 32 lines (+11/-3)2 files modifiedSrc/iosockets.F90 (+10/-2)
version.info (+1/-1)
Branch information
Recent revisions
- 563. By Alberto Garcia
-
Fix mulliken populations for non-collinear spin in parallel
For non-collinear spins, the mulliken routine was not parallelized
correctly. - 562. By Alberto Garcia
-
Spin monitoring in scf cycle. Clarification of sign convention for DM
* Spin monitoring during scf cycle
If the fdf variable 'Spin.In.Scf' is set to 'true', the size and
components of the (total) spin polarization will be printed at every
scf step. This is analogous to the 'Mulliken.In.Scf' feature.The default for the Spin.In.SCF fdf flag is 'true' for calculations
involving spin.A single line of spin information is printed, prefixed by ' spin-moment:'.
Note that now, for each scf step, the spin and/or mulliken analyses are
printed *after* the line showing the energies.The initial spin polarization for every geometry iteration is also
printed in 'state_init', replacing the similar calls in 'new_dm'.* Use explicitly the same DM_12 sign-convention as in SOC versions
In the non-collinear case, the code is actually using internally the
same sign convention for the "up-down" components of H and the DM as
the spin-orbit-capable (SOC) versions (>= 4.1): The building of the
dense Hamiltonian has the same form and the off-diagonal components
of Vxc are identical. Routines diag2g and diag2k are actually using
the same convention, but with DM* as an intermediate step.By changing the interface to the outside world, namely the
correspondence between spin angles and the DM in new_dm, and the
calculation of the spin (actually moment) components in spnvec, the
code can now interoperate (i.e., reuse DMs) with the SOC versions.In order to reuse DM information produced by previous versions of
the program in the non-collinear case, the sign of D(4) of the old
DM has to be changed. This can be achieved with the
dm_noncol_sign_flip4 program in Util/DensityMatrix. - 561. By Alberto Garcia
-
Make generation of timer report more robust. More information in CML
* Generate timer report in non-convergence and out-of-time conditions
For proper accounting, the timer information should be generated even
if the program stops due to non-convergence (scf, geometry) or
out-of-time conditions.This is achieved by inserting appropriate calls to 'timer' before the
call to 'die'. To avoid a race condition in which a non-root node
might get to the 'die' call before the IO node has had time to
properly write the CML, timing, and even output information, a call to
'barrier' has been inserted before the 'die' call.It might be argued that this kind of bookeeping should be left to the
'die' call itself. But it is only possible if all nodes call 'die'.* Add no_u and nnz info to CML file to ease the extracting of
benchmarking information. - 560. By Alberto Garcia
-
Enable compatibility switch for NRTAB value
Internally the two-center integrals involved in some matrix element
calculations are tabulated with a preset number of elements (NRTAB). In
versions 4.0.1 and prior this was 128. Since 4.0.2 the number of table
elements has been increased to 1024, which translates to more accurate
matrix element calculations.If preservation of the (lower accuracy) numerical results of 4.0.1 or
prior versions is required for reproducibility purposes, the user can
includeCompat.
Matel.NRTAB T in the fdf file.
- 558. By Alberto Garcia
-
Write JSON file for tree-timer information
A more flexible output for timing information.
The 'time.json' file is produced if the optionuse-tree-timer T
is in effect.
- 557. By Nick Papior
-
Increased NRTAB to 1024.
Initial investigations shows that NRTAB in matel is important
for forces and possibly other quantities.I have added a notice in the compatibility.tex text which describes
what to do to revert the behaviour. - 555. By Nick Papior
-
Added fermi-level to PDOS output
By adding the fermi-level to the PDOS output we do not need
to read the fermi-level from elsewhere.
This makes analysis much easier.Also fixed a PDOS output bug for no_u > 9999.
Branch metadata
- Branch format:
- Branch format 7
- Repository format:
- Bazaar repository format 2a (needs bzr 1.16 or later)
- Stacked on:
- lp:siesta