Update ppconvert

[ye-luo]

1. allows changing density mixing parameter via command line.
2. update ppconvert manual section

[Paul-St-Young]

@ye-luo thanks for adding more details on ppconvert to the manual!

Could you add a few sentences on how a "reference state" should be chosen?
A few examples would be greatly appreciated.

[ye-luo]

@Paul-St-Young I added my limited knowledge of how to set the reference states.

[Paul-St-Young]

@ye-luo thanks a lot! the reference state should include only the valence electrons this is what I missed. I can now get ppconvert to run through. Checking results.

[Paul-St-Young]

The following script seems to produce a reasonable upf for silicon
`wget http://www.tcm.phy.cam.ac.uk/~mdt26/pseudo_lib/si/1/summary.txt
wget http://www.tcm.phy.cam.ac.uk/~mdt26/pseudo_lib/si/1/pp.data
wget http://www.tcm.phy.cam.ac.uk/~mdt26/pseudo_lib/si/1/awfn.data_s2p2_3P

BIN=ppconvert

pot=Si.TN
ref="1s(2)2p(2)"
loc=2

cmd="$BIN --local_channel $loc --casino_pot pp.data --s_ref $ref --p_ref $ref --d_ref $ref"

$cmd --upf $pot.upf >& convert_upf.out
$cmd --xml $pot.xml >& convert_xml.out`

[Paul-St-Young]

The local part of potential agree very well with the tabulated data.
However, there seem to be some small differences in the pseudized orbitals.

[Paul-St-Young]

I should mention that my first instinct was to use 3s(2)3p(2), because Wikipedia says Si [Ne] 3s(2)3p(2). However, this naive first attempted failed miserably because ppconvert puts nodes into the pseudized orbitals. i.e. RadialWF::Solve uses int TotalNodes = n-l-1; Perhaps this is something that should be explicitly warned against in the manual.

[ye-luo]

The difference in orbitals may be caused by using a different XC functional or HF or DF.

[Paul-St-Young]

The converted pseudopotential can reproduce the Si band structure compared to the standard quantum espresso example (PP/examples/example01). I also tried manually replacing the <PP_PSWFC> and <PP_RHOATOM> entries in the UPF with data from awfn.data (<PP_R> and <PP_LOCAL> already exactly matches pp.data). I saw no change to the band structure. I think the small difference we saw is insignificant.

[prckent]

@Paul-St-Young The incorrect number of nodes clearly indicates the wrong states are being solved for by ppconvert. You need to change the specification of the reference state. From an archived page (google search):

Note that the reference state should include only the valence electrons. For example, for a sodium atom with a neon core, the reference state would be "1s(1)". Note that the "s_ref" needs to include a 1s state (potentially with 0 occupancy), the "p_ref" needs to include a 2p state, and the "d_ref" needs to include a 3d state. One reference state should be included for each channel in the pseudopotential.

[prckent]

@Paul-St-Young Do you have enough information to convert the potentials that you need? Clearly (i) a set of converted potentials would avoid this problem and (ii) a better write up is needed for ppconvert to be useful or an alternative found.

[ye-luo]

@prckent I looked at the archived page when I wrote my guidance for reference states. The following requirement is no more needed.

Note that the "s_ref" needs to include a 1s state (potentially with 0 occupancy), the "p_ref" needs to include a 2p state, and the "d_ref" needs to include a 3d state.

[Paul-St-Young]

@prckent yes, I was able to obtain the pseudopotential I needed using the bash script I posted above. i.e. ref="1s(2)2p(2)". I understand that orbitals with nodes are wrong. I was just trying to point out a gatcha for new users.

[Paul-St-Young]

I also noticed that the starting charge in QE is off by 4pi
starting charge 20.36910, renormalised to 256.00000.

If I scale the <PP_RHOATOM> entry in the generated UPF by 4pi, then the starting charge is almost correct starting charge 255.96560, renormalised to 256.00000.

The QE description of the UPF file says <PP_RHOATOM> should hold 4pi r^2* true charge. Is this a new update since ppconvert was made?

[ye-luo]

If all the pseudopotential files are created by ppconvert, this is not a big issue since QE will rescale them all. If mixed pseudopotential files (from OPIUM and BFD) are used, the starting density may be pretty bad. So fixing the density normalization is needed.

[prckent]

I updated the ghost instructions.