The ABINIT Users Mailing List ( CLOSED )

[Paul Fons <paul-fons@aist.go.jp>]

First of all, thank you very much for your offer of help.  I felt a little embarrassed in asking for more guidance, but I am sure there are many more like me out there.  

I wanted to start off by saying that the Cd file with rc=2.0 Angstroms was one I had placed in the atompawabinit folder by myself from the abinit PAW potentials download page (http://www.abinit.org/PAW/ATOMICDATA/048-Cd/LDA-HOLZWARTH/).  It was my mistake to attribute it to Natalie Holzwarth's group (sorry about that!).  Lacking a general recipe of what to try in the generation of potentials, I was attempting to go through the examples on the Lake Forest page (http://www.wfu.edu/%7Enatalie/papers/pwpaw/periodictable/periodictable.html).  On the Lake Forest page, the downloadable potentials seem to be of two flavours as reinforced by the somewhat cryptic label "Modified Elements" and "Unmodified Elements".  I didn't find a reference anywhere as to what these terms meant (pardon me if it was just my oversight).  I did find some excellent notes on the details of the implementation of PAW under the link (Notes on new atompaw implementation.) I found the document very informative but a little short of details on the nitty-gritty of how to go about twisting knobs and double checking a potential (I expected that it would be a lot of work).  Along these lines, I thought it might be useful to clarify some of the basic procedures so we can go about making a set of potentials that can be used by the abinit community.  

I was curious to follow up on the mail by Natalie Holzwarth. First, the general guidelines in her email appear to be useful and provided some new information to me that I hadn't considered (e.g. the inclusion of positive energy states).  I wanted to take this opportunity to ask a couple of (what I hope aren't too) simple questions.

1. On the Lake Forest periodic table of PAW potentials, most of the potentials (the "unmodified" ones?) seem to use PS of the Trouiller-Martin norm conserving form using the Blochl scheme.  This would appear to have the advantage that there are fewer parameters to tweak (I am assume this is also its disadvantage!).  The new entries seem to use T-M PS, but with the Vanderbilt scheme.  

I was hoping that Natalie (or another person from her research group) would be so kind as to comment on the advantages/disadvantages of the two techniques.

2.  On the Lake Forest page, there is the comment  "In addition to the manual prepared by Marc Torrent, notes on the method are available. (" The link pointing to the manual is dead while the page I can find on the abinit site would seem to be (http://www.abinit.org/PAW/AtomPAW2Abinit-Manual-html/HolzwarthPAW.htm) which appears to be a one page summary of how to compile the program and a couple of input files are included with a one line of explanation of what the input lines mean in a particular file, but there is not comment that I could find as to the logic of how to choose parameters or even what the naming scheme is for the different output files are. These two examples also both use the VNCT keyword (Blochl scheme Trouiller Martin PS).   

  Thanks for some advice in advance!

Dr. Paul Fons

Nano-Optics Reseach Team

Team Leader

National Institute for Advanced Industrial Science & Technology

METI

Center for Applied Near-Field Optics Research (CANFOR)

AIST Central 4, Higashi 1-1-1

Tsukuba, Ibaraki JAPAN 305-8568

tel. +81-298-61-5636

fax. +81-298-61-2939

email: paul-fons@aist.go.jp

The following lines are in a Japanese font

〒305-8562 茨城県つくば市つくば中央東 1-1-1

産業技術総合研究所

近接場光応用工学研究センター

近接場光基礎研究チーム　チーム長

ポール・フォンス

On Apr 16, 2007, at 9:40 AM, Natalie Holzwarth wrote:

Dear abinit users,

      I am very glad to help with the atompaw program and to try to help

generate basis and projector functions for any particular project when I

have time.  (In the past I have some times promised to help and then could

not find the time.)  Nick Dellaripa is working with me for a few months

(before he goes to graduate school) to up date the web pages

http://www.wfu.edu/~natalie/papers/pwpaw/man.html and we will

be happy to test and include basis functions generated by other users.

Unfortunately, there is no full-proof recipe that works every time, and it

can be very time consuming to find and test good functions.  For what it

is worth, here are some guidelines that I use:

1.  Find the maximum rc for each atom for the particular set of materials

you will study so as not to have overlapping augmentation spheres.  (Some

overlap is OK, but not too much.)   Your rc values should be that value or

less.

2.  Decide which electrons should be valence and which should be core.

If you are studying an elemental material, the outer valence electrons

should be fine.   If you are studying a material which has positive ions,

it may be necessary to use the upper core states as well as valence states

in your basis set.  If you want to work with excited states , it would be

good to include some additional positive energy states for each l value as

well.

3.  Check the logderivative plots.  If there is a ghost state in the

middle of the range of interest, you will need to choose a different rc or

choose different basis functions.  (The worst example we found was P in

which case we had to use a 2s3s2p3p basis set in order to get rid of the

ghost resonance.)  I think that Ga and As may also have this problem.

4.  Find a simple example of the materials you are interested in and try to

test the converence and accuracy of the function you generated on that.

Ideally, I like to compare our results with an independent program such as

wien2k.

Marc Torrent has written a very nice manual about all of this which he

will put on the abinit web page if it is not there already.  I should

mention that we recently discovered that while most basis sets give nearly

identical results for pwpaw and abinit, some are slightly off in the

+-0.001 Ry range.  It would be nice to understand this descrepancy ...

Thanks for your interest in the atompaw program.  Your feedback is always

welcome.

                            Thanks, Natalie Holzwarth

PS.  Did you mean Cr?   A small rc would be needed for CrN perhaps.

On Fri, 13 Apr 2007, paul fons wrote:

I would also second this request.  The documentation for atompaw has

not kept pace with it development.  While some parameters are obvious

(loggrid, scalarrelativistic), others are not.  In particular

guidelines/advice on how to pick projector energies is lacking.

There is also a dichotomy in the examples on the example potentials

on Nancy Holzwarth's periodic table (http://www.wfu.edu/%7Enatalie/

papers/pwpaw/periodictable/periodictable.html).  Here you can see

example input files for atompaw as well as equation of state data for

calculations in which the PAW potentials have been used.  The table

has two classes of entries the older (labled unmodified in the table)

entries all use the Vanderbilt scheme but do not enter any additional

projectors/energies while the (newer, recent) entries labled modified

entries do introduce such levels.   How to decide on these levels?

It would be great to have a tutorial on this subject (it would answer

a lot of people's question).  As of the moment it is a hard resource

to use and it shouldn't be as it looks to be very useful.  In my

limited tests, it seems to work fairly well, but I feel like I am

working a little in the dark.  For example, in the examples folder

for the latest version of atompaw there is a Cd example, however the

core radius for this PS is set for 2.0 Angstroms.  Why such a small

cutoff?  What was the logic behind it?  The answers to how to choose

energy levels (I know it is a try and error process) as well as some

educated comments about appropriate core radii would go a long way

toward the generation and verification of PAW PS.

Another idea would be to try and generate more elements of the period

table as a group and then verifying them by calculating known

materials EOS.  Doing this process as a group might lead to a useful

table of elements that all can use.

Just my thoughts,

Paul Fons

Paul Fons

Team Leader

Nano-Optics Research Team

paul-fons@aist.go.jp

Center for Applied Near-Field Optics

National Institute of Advanced Industrial Technology

Tsukuba Central 4, Higashi 1-1-1

Tsukuba, Ibaraki Japan 305-8562

tel. +81-298-61-5635

fax. +81-298-61-2939

The following lines are in a Japanese font

���305-8562 ��������������������������������������� 1-1-1

���������������������������

������������������������������������������

������������������������������������������������

������������������������

On Apr 13, 2007, at 7:53 AM, snow9@llnl.gov wrote:

Guillaume, and others:

I am interested in developing a Ga PAW set with atompaw, but have

only just begun using atompaw.  Would you mind sharing what you

have for Ga, that I can use as a guideline for my set?  Thanks a

million.

-- Ryan

Quoting Guillaume Dumont <dumont.guillaume@gmail.com>:

Dear Paul,

We actually had some trouble generating PAW datasets for Ga with

the USPP

package. As you suggest we figured out that the partial core

corrections had

a great impact on the physical properties one obtains with datasets

generated with this package. On the other hand, AtomPAW has

improved a lot

since the first time I tried it. Personnaly, I find it easier to

use than

USPP. There is also a great manual written by M. Torrent that

should be

available on the web (is it?).

At the last developper workshop, everyone agreed that a complete

table of

PAW should be generated by some of the developpers. So the fact

that the

list of elements is more complete with USPP should not be a

problem anymore

in the near future.

Guillaume

On 4/10/07, Paul Fons <paul-fons@aist.go.jp> wrote:

 I have tried both methods of generating ultrasoft

pseudopotentials and

was under the impression that Nancy Holzwarth's system generated

more

reasonable pseudopotentials, but perhaps it was the non-linear core

corrections that were the problem.  Has anyone systematically

compared the

two methods?   The USPP pseudopotentials have the advantage that

there are a

lot of different input files for the elements already available I

have

noted.  On the other hand Nancy Holzwarth's pages have a periodic

table

linked with several tens of elements included as well.  On thing

that has

bothered me about this table is that none of the input files

create new

generalized basis functions (whereas the input file on the abinit

PAW page

for O does).    I have generated a PAW potential for Sb and it

seemed to

give reasonable results, but I am curious about others experiences.

Dr. Paul Fons

Nano-Optics Reseach Team

Team Leader

National Institute for Advanced Industrial Science & Technology

METI

Center for Applied Near-Field Optics Research (CANFOR)

AIST Central 4, Higashi 1-1-1

Tsukuba, Ibaraki JAPAN 305-8568

tel. +81-298-61-5636

fax. +81-298-61-2939

email: *paul-fons@aist.go.jp*

The following lines are in a Japanese font

��305-8562 �������������������������� 1-1-1

������������������

����������������������������

���������������������� ��������

����������������

--

Guillaume Dumont

=========================

guillaume.dumont.1@umontreal.ca

dumont.guillaume@gmail.com

(514) 341 5298

(514) 343 6111 ext. 13279

N. A. W. Holzwarth email: natalie@wfu.edu

Department of Physics www: http://www.wfu.edu/~natalie

Wake Forest University         voice: 336-758-5510

Winston-Salem, NC 27109-7507            fax: 336-758-6142

U. S. A.