[Pw_forum] Converting pseudopotentials from CPMD
Axel Kohlmeyer
akohlmey at gmail.com
Fri Nov 25 17:24:20 CET 2011
On Fri, Nov 25, 2011 at 11:02 AM, Merlin Meheut
<merlin.meheut at get.obs-mip.fr> wrote:
>
> Hi Axel,
>
> Thanks for your help,
>
> To have some understanding on the problem, I compared the original and
> converted pseudopotential files: if they are not exactly the same,
> the spatial grid (<PP_R> in the UPF file, first column of all fields for
> CPMD format), the wavefunction values (for 2S and 2P: <PP_PSWFC>
> in the UPF file, &WAVEFUNCTION in CPMD) , and the NLCC fields
> (<PP_NLCC> in the UPF file, &NLCC in CPMD) are exactly the same in both
> files. I was not too much worried about the nlcc correction, for this
> reason...
>
> However, for the rest of the file, the CPMD file has only a &POTENTIAL
> field wit 2S and 2P potentials, whereas the UPF file have <PP_RAB> ,
> <PP_LOCAL>, and <PP_NONLOCAL> fields, and I do not see any agreement
> between the two. What should be the connection?
with CPMD potentials you have to indicate the local potential
in the CPMD input file and then the kleinman-bylander procedure
is done internally in CPMD, whereas in UPF this is already
encoded in the pseudopotential. CPMD also offers to use
gauss-hermite integration instead of kleinman-bylander
(more expensive, but may be needed to avoid ghost
states, e.g. for transition metals).
> The cpmd file seems quite simple, with two wavefunctions, two potentials
> and one nlcc field. Where is the difficulty of the conversion?
> How are the two formats differing for the potentials? Since I have the
> same values for the wavefunctions, spatial grid and nlcc,
> shouldn't I have the same also for potentials? Maybe this is related to
> a wrong choice for the local part?
the problems are related to having to anticipate correctly
the transformations, regridding, and re-normalizations that
CPMD is doing. add to that that the occasional factor 2
since CPMD uses hartree atomic units while Q-E uses
rydberg atomic units.
> By the way, I began by the most difficult case: the O, H and Si
> pseudopotentials I have do not contain nlcc corrections:
> do you think that I can trust the conversion, even if I do not give the
> correct local l?
it is always better to start with the easiest case and
make sure that it works. i am not saying that it is
impossible what you want to do, but one has to make
fairly detailed tests and know how the potentials were
created to fill in the blanks. for exactly that reason,
i found it simpler to start from a consistent atomic
code that can produce both, CPMD format and UPF
output, and then feed those potentials to both codes
to see, if they can work well.
> My problem for re-generating a pseudo is that I am not quite confident
> in which recipe to use... Especially for Li....
Li is a quite difficult case. i think the NLCC version
only works well for metallic Li and you may be better
off to do a "semi-core" style potential for ionic Li.
> Since you say cpmd comments can't be trusted, do you think rc are all
> bogus?
cannot trusted means, that you don't know. some are correct,
some are bogus. it all depends on who created the potential
and what this person assumed about how and by whom the
potential will be used. most are more-or-less accurate.
often you can tell bogus entries if the energy eigenvalue is
identical to the previous state. but i also have seen potentials,
where the comment section was that of a completely different potential...
that is what i mean by unreliable... you can't really know from the file.
axel.
> Best regards,
>
> Merlin Méheut
>
>
>
> On 25/11/2011 16:03, Axel Kohlmeyer wrote:
>> On Fri, Nov 25, 2011 at 9:27 AM, Merlin Meheut
>> <merlin.meheut at get.obs-mip.fr> wrote:
>>> Dear pwscf users,
>>>
>>> I am trying to obtain consistent pseudopotentials to use in both CPMD and
>>> pwscf, precisely for O,H, Si and Li atoms.
>>> I therefore took pseudopotentials from CPMD and tried the cpmd2upf.x
>>> converter.
>>>
>>> I am however not sure of what I should do. In particular, I am not sure of
>>> the local reference state used in these pseudos.
>>> Does this have an influence on the conversion, or is that just an
>>> information written somewhere in comment in the UPF pseudopotential?
>>> More generally, I am not sure of the recipes used to generate these
>>> pseudos, and in which
>>> extent they should be considered in the conversion. THe only information I
>>> have is inside the preamble of the pseudopotential files
>>> (Pseudopotential report), which is reproduced below:
>>>
>>> For example, for the Li atom, I attached the cpmd pseudopotential and here
>>> are the instructions I gave for conversion:
>>>
>>> $ ./cpmd2upf.x
>>> Input file> Li_MT_BLYP_NLCC.psp
>> i don't think that anybody every got the NLCC conversion
>> from CPMD to UPF working correctly. i remember i tried
>> several iterations back and forth with paolo many years
>> ago and then finally went the other route and regenerated
>> pseudopotentials with from the same settings with paulo's
>> atomic code and then converted the output to cpmd format
>> and UPF alike.
>>
>>
>>> amesh set to: 1.002086
>>> ============================================================
>>> | Pseudopotential Report Thu Aug 22 12:57:11 1996 |
>>> ------------------------------------------------------------
>>> | s channel: |
>>> | Atomic Symbol : LI |
>>> | Atomic Number : 3 |
>>> | Number of core states : 1 |
>>> | Number of valence states : 1 |
>>> | Exchange-Correlation Functional : |
>>> | Slater exchange : .6667 |
>>> | LDA correlation : Lee-Yang-Parr |
>>> | Exchange GC : Becke (1988) |
>>> | Correlation GC : Lee-Yang-Parr |
>>> | Electron Configuration : N L Occupation |
>>> | 1 S 2.0000 |
>>> | 2 S 1.0000 |
>>> | Full Potential Total Energy -7.571372 |
>>> | Trouiller-Martins normconserving PP |
>>> | n l rc energy |
>>> | 2 S 2.2000 -.11501 |
>>> | 2 P 2.2000 -.11501 |
>>> | Nonlinear core correction |
>>> | Core charge cutoff radius : 1.80000 |
>>> | Number of Mesh Points : 575 |
>>> | Pseudoatom Total Energy -.205704 |
>>> ============================================================
>>> | P-channel taken from: |
>>> | 2 S 2.2000 -.21051 |
>>> | 2 P 2.2000 -.13639 |
>>> | 3 D 1.7031 -.13639 |
>>> | Nonlinear core correction |
>>> | Core charge cutoff radius : 1.80000 |
>>> | Number of Mesh Points : 575 |
>>> | Pseudoatom Total Energy -.106791 |
>>> ============================================================
>>> lmax to use. (max. 1)> 1
>>> l local (max. 1)> 0
>>> Wavefunction # 1: label, occupancy> 2S, 1.0
>>> Wavefunction # 2: label, occupancy> 2P, 0.0
>>> Pseudopotential successfully converted
>>> Output PP file in UPF format : Li_MT_BLYP_NLCC.psp.UPF
>>> *** PLEASE TEST BEFORE USING!!! ***
>>> review the content of the PP_INFO fields
>>>
>>>
>>> As you can see in the comment of the cpmd potential, the pseudopotential
>>> contains potentials (and wavefunctions) for 2S and 2P,
>>> with the energy of 2P fixed at the same value that 2S.
>> never trust those comments in CPMD pseudopotentials.
>> the atomic code creates bogus entries in there (and
>> nobody seemed to care to fix that issue) and some
>> people "fix" them afterwards, others don't and some
>> actually "fix" those tables incorrectly. several potentials
>> also mix results from different calculations and then
>> the comments can be completely bogus.
>>
>>> I am not sure of what recipe it corresponds to: is that some sort of
>>> Hamann's approach?
>> the atomic code shipped with CPMD is based on
>> hamann's code, but heavily hacked.
>>
>>> Of course, I could test these converted pseudopotentials, but I would like
>>> to be aware of potential problems before.
>> i would be very, *very* careful. particularly
>> on pseudopotentials with NLCC. i suspect,
>> but never found proof, that there are some
>> cases where a bug in the atomic code related
>> to NLCC is compensated by a corresponding
>> bug in the CPMD code.
>>
>> axel.
>>
>>> I would be very thankfull for any help,
>>>
>>> Best regards,
>>>
>>> Merlin Méheut
>>>
>>> --
>>> Merlin Méheut
>>> adresse labo:
>>> Géosciences Environnement Toulouse
>>> OMP - Université Paul Sabatier
>>> 14 avenue Edouard Belin
>>> 31400 Toulouse
>>> FRANCE
>>> tel: (+33) 5 61 33 26 21
>>> adresse personelle:
>>> 3 rue du Puits Vert
>>> 31000 Toulouse
>>> port: (+33) 6 34 67 57 02
>>> pseudo skype: quantification
>>>
>>>
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://www.democritos.it/mailman/listinfo/pw_forum
>>>
>>>
>>
>>
>
>
> --
> Merlin Méheut
> adresse labo:
> Géosciences Environnement Toulouse
> OMP - Université Paul Sabatier
> 14 avenue Edouard Belin
> 31400 Toulouse
> FRANCE
> tel: (+33) 5 61 33 26 21
> adresse personelle:
> 3 rue du Puits Vert
> 31000 Toulouse
> port: (+33) 6 34 67 57 02
> pseudo skype: quantification
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
--
Dr. Axel Kohlmeyer
akohlmey at gmail.com http://goo.gl/1wk0
College of Science and Technology
Temple University, Philadelphia PA, USA.
More information about the users
mailing list