Dear Matias,

That's a very good point.

To address it, I just looked at this paper:

Guillermo Román-Pérez and José M. Soler
Phys. Rev. Lett. 103, 096102 – Published 27 August 2009

and saw they benchmarked carbon nanotubes.


 I suggest to look at papers by Cooper, Soler, and other teams developing vdw 
pseudos, and see if they benchmarked water. You can also look at other quantum 
chemistry works beyond the DFT level and see what they report for water in 
terms of binding energy and bond lengths, to say something.


 I believe Fernandez-Serra has worked with water using vdw pseudos as well, so 
have a look at some of her papers too.


 The last leg of this process thus involves some literature search, but I am 
glad to read that your pseudos are ghost-free at the moment. As indicated 
before, it would be great if you share your results and benchmarks with us.


Cheers!

-Salvador


________________________________
From: [email protected] <[email protected]> on behalf 
of Matías Soto C. <[email protected]>
Sent: Tuesday, October 6, 2015 1:02 PM
To: [email protected]
Subject: Re: [SIESTA-L] Question about pseudopotential vdw for Hydrogen atom

Thanks for the advice Salvador.

It takes a little bit to change the radius and test the new pseudo for no ghost 
states and then test it for water molecule bond lengths and bond angle. But it 
compares well to experimental bond lengths and bond angle.

However, I have a question: how do you ensure eigenvalues and spin polarization 
look good?
I can look at the eigenvalues and compare to those from GGA and LDA, but other 
than that I don't know if there's a more straightforward way of comparison.

Thanks again,

-- Matias Soto



Matias Soto C.
Ph.D. Program
MSNE Department
Rice University

On Thu, Oct 1, 2015 at 10:52 AM, Salvador Barraza-Lopez 
<[email protected]<mailto:[email protected]>> wrote:

 Indeed,

 I have encountered similar phenomena: For instance, working with the BH vdw 
exchange-correlation potential, I documented ghost states for Te when taking 
the INP from the webpage. I feel that the expectation that plugging a new xc 
flag on the INP file should not be expected to hold always (and why should 
it?)... So you have to put some time to get it done.


 H should not be problematic, and that may be the reason you have not received 
a file yet...


 I may say, increase/decrease the radius for the second component (down the INP 
file) by about 10% and get a pseudo. Plug that pseudo back into your 
calculation and see if it shows no ghosts. If not, just ensure bond lengths and 
eigenvalues, and maybe spin polarization look good. Then you should be good to 
go. There is some work involved, but there are things to learn as well. As I 
said, it may not be that difficult. Could put you back one day, but it will be 
worth doing it.


 Again, it may be that someone in the list could send you the file you need. 
But while you wait, it may be worth giving it a go yourself. Or in other words, 
why wait if you can do it. You could then share your own INP, so that others 
know what rs to use.


 Cheers!

-Salvador



________________________________
From: [email protected]<mailto:[email protected]> 
<[email protected]<mailto:[email protected]>> on 
behalf of Matías Soto C. <[email protected]<mailto:[email protected]>>
Sent: Thursday, October 1, 2015 10:10 AM
To: [email protected]<mailto:[email protected]>
Subject: Re: [SIESTA-L] Question about pseudopotential vdw for Hydrogen atom

Thanks for your reply Salvador.

The pseudo was taken from the siesta database which  uses the cutoff radii 
shown. Such pseudopotential does not show any ghost states when running under 
LDA or GGA. I've seen other messages in this listserve where people had similar 
problems but the solution is not shown.

Hopefully I can follow your advice and develop my own pseudpotential for 
Hydrogen in van der waals DFT. However, I was hoping that someone had already 
found a solution to the same problem and was willing to share the 
pseudopotential.

Thank,

-- Matias

Matias Soto C.
Ph.D. Program
MSNE Department
Rice University

On Wed, Sep 30, 2015 at 1:10 PM, Salvador Barraza-Lopez 
<[email protected]<mailto:[email protected]>> wrote:
 Dear Matias,
 Let me provide the following advice.

 The channel showing ghosts is the p-one (L=1), so it is not that relevant for 
the problem, L=0 is not giving you ghosts.

 So I suggest that you vary the radii for the L=1 channel on the INP file. I 
could not tell you how to change it, since one has to try by hand.

 Then, you would want to run a calculation with these pseudos on known 
environments (water molecule, and so on), and see if you match all-electron 
results (bond distances, eigenvalues). That way you would know your pseudo is 
good, or you could improve if it is not.

 Of course, there may be somebody who could provide a working one for you; I've 
not worked with that element and that pseudo. But even in that case, you may 
want to ask the person who is giving it to you, or you may want to look at it 
yourself, as of how the pseudo is performing.

 I hope these comments are helpful.

Best regards,
-Salvador

________________________________________
From: [email protected]<mailto:[email protected]> 
<[email protected]<mailto:[email protected]>> on behalf of Matias 
Soto <[email protected]<mailto:[email protected]>>
Sent: Wednesday, September 30, 2015 12:59 PM
To: [email protected]<mailto:[email protected]>
Subject: [SIESTA-L] Question about pseudopotential vdw for Hydrogen atom

Hello SIESTA users,

I know this question has been asked before but I couldn't find the message
where the solution has been shown.

I need to generate psf file for hydrogen. I changed from ca to vw in inp file
when generating psf file in ATOM program. When I run calculation for single
hydrogen atom using siesta-trunk-462 I get messages about ghost states:

atom: Called for H                     (Z =   1)

read_vps: Pseudopotential generation method:
read_vps: ATM3.3.2  Troullier-Martins
Total valence charge:    1.00000

xc_check: Exchange-correlation functional:
xc_check: VDW Dion-Rydberg-Schroeder-Langreth-Lundqvist
V l=0 = -2*Zval/r beyond r=  1.3816
V l=1 = -2*Zval/r beyond r=  1.3816
V l=2 = -2*Zval/r beyond r=  1.3644
All V_l potentials equal beyond r=  1.2977
This should be close to max(r_c) in ps generation
All pots = -2*Zval/r beyond r=  1.3816

VLOCAL1: 99.0% of the norm of Vloc inside     25.776 Ry
VLOCAL1: 99.9% of the norm of Vloc inside     58.745 Ry
atom: Maximum radius for 4*pi*r*r*local-pseudopot. charge    1.52711
atom: Maximum radius for r*vlocal+2*Zval:    1.28155
GHOST: No ghost state for L =  0
GHOST: WARNING: Ghost state for L =  1
GHOST: No ghost state for L =  2
KBgen: WARNING: Ghost states have been detected
KBgen: WARNING: Some parameter should be changed in the
KBgen: WARNING: pseudopotential generation procedure.
Stopping Program from Node:    0

I am using:
XC.functional VDW
XC.authors    DRSLL

If someone could please provide working parameters for hydrogen
pseudopotential for van der waals exchange-correlation.

Thank you!


Matias Soto C.
Ph.D. Program
MSNE Department
Rice University



Responder a