Siesta

"PAO.BasisType nodes" does not work properly with siesta 4.0

Bug #1625725 reported by Seungchul Kim on 2016-09-20

This bug affects 2 people

	Status	Importance	Assigned to	Milestone
Siesta	Fix Released	Medium	Nick Papior	Siesta 4.0.1
4.0	Fix Released	Medium	Alberto Garcia	Siesta 4.0.1
4.1	Fix Released	Medium	Nick Papior	Siesta 4.1-b2

Bug Description

SIESTA 4.0 does not give right total energies when nodes type of PAO is used. Total energies of Ti atom and single crystal (two atoms) calculated using siesta 3.2 and 4.0 are below.

# calculation setting is ...
pseudopotential: scalar relativistic correction is added to the pbe INP file in siesta library (FHI 98).
Ecut: 240 Ry
Basis: DZP, PAO.EnergyShift = 0.005 Ry
Ti crystal: HCP, two atoms in an unitcell

When I used "PAO.BasisType splite", I've got Etot(atom) = -109.22 eV and Etot(bulk) = -228.49.
But, when "PAO.BasisType nodes" is used, Etot(atom) = -57.83 eV and Etot(bulk) = -201 eV.

So, I've tested nodes type PAOs using siesta 3.2, then the results are
Etot(atom) = -109.242 eV and Etot(bulk) = -228.304 eV.

SIESTA 3.2 gives resonable results, although error in cohesive energy is large, but 4.0 does not.

arch.make files of siesta 3.2 and 4.0 are the same. Intel compiler + mkl (2013 version)+ openmpi were used. I also tested with gfortran compiled excutable (serial compilation), but the results are same as Intel compiler+mkl+openmpi one.

Revision history for this message

Alberto Garcia (albertog) wrote on 2016-09-27:

Thanks for your report.
Please provide your .fdf files and the .psf pseudopotential file.

Revision history for this message

Nick Papior (nickpapior) wrote on 2016-09-29:

Sadly one cannot attach files using the email client on launchpad.

Could you please go to this page:
https://bugs.launchpad.net/siesta/+bug/1625725
and manually attach the files?

Revision history for this message

Seungchul Kim (hrsckim) wrote on 2016-10-07:

files.zip Edit (149.6 KiB, application/zip)

Hi,

Here I attached psf, input & output of siesta 3.2 and 4.0, as well as INP and OUT of ATOM.

Thank you very mucch.

Revision history for this message

Alberto Garcia (albertog) wrote on 2016-10-13:

Patch for 'nodes' basis-generation option Edit (524 bytes, text/plain)

Thank you for your report.

There is indeed a bug in the "NODES" routine in 4.X versions. A patch is attached and will be applied to the code shortly.

Note however that the preferred way to generate basis sets in Siesta is the default 'split' scheme.

Regarding the slowness in convergence and the wrong final spin value for the Ti atom calculations: I have tracked it down to the use of an electronic temperature of 300K. In this case, this seems to blur out small differences in energy levels and confuse the algorithm. When you set it to something very small such as 0.1 K, the convergence is better and the final spin is 2, as it should be.

Nick Papior (nickpapior) on 2016-11-28

no longer affects:

siesta/trunk

Nick Papior (nickpapior) on 2017-02-20