The ABINIT Users Mailing List ( CLOSED )

[fer@tiziano.phys.washington.edu]

Hi..

On Thu, 31 Aug 2006, Xavier Gonze wrote:

> Dear Fer,
>
> Some technical details ...
>
> (1) TPSS needs the kinetic energy density (space-resolved), that is not 
> available in ABINIT at present
> (its implementation would not be a lot of work, though)
> So, even if the Nanoquanta XC library has it, TPSS cannot be used with 
> ABINIT 
> (at present).

The lack of the kinetic energy density was what stopped me from going 
straight ahead and programming TPSS. I realized that implementing it 
couldn't be too difficult, but it wouldn't be easier either for someone 
like me that doesn't know the code that well.

I gather from your message that the use of nanoquantaxc is not 
"transparent", i.e. you need to modify the code setting up and interface 
to provide nanoquantaxc with the stuff it needs.

> (2) More preoccupying, and independent of ABINIT : one cannot compute 
> straightforwardly the Vxc (and Kxc) corresponding to a meta-GGA like 
> TPSS (one needs an OEP method - quite unpleasant). So, I think that in 
> the available implementations (Gaussian 03, WIEN), the TPSS is computed 
> in a one-shot calculation, on top of a LDA (or GGA) density, not 
> self-consistently. So, all optimizations have to be done without the 
> gradients (so, without the forces, and stresses). It is even worse for 
> the computation of the interatomic force constants, that one will not be 
> able to do using the linear-response theory (or with an enormous 
> implementation effort)

As far as I can tell, Gaussian's implementation of TPSS is 
self-consistent (I was just digging through the source), using the TPSS 
density and not a precomputed one.

I'm probably missing something, but as far as I know, OEP's are needed for 
functionals 
that depend explicitly on the orbitals. TPSS depends on the orbitals 
through the kinetic energy density. Doesn't this allow you to calculate 
vxc and kxc without having to resort to an OEP?

> (3) Hybrid functionals (mixing with Fock contribution - like B3LYP) have 
> a much better behaviour if one does not to do SCF loops as DFT's 
> (meaning again a OEP), but with the self-consistent Fock operator. As 
> the kinetic energy density, the Fock exchange contribution is not 
> implemented in ABINIT. But, unlike the kinetic energy density, the Fock 
> exchange contribution is heavy to compute using plane waves... Still, I 
> have the impression that there is more hope along these lines to be able 
> to compute efficiently accurate (with respect to experiment) coherent 
> lattice parameters AND interatomic force constants.

I agree, we have found that we can do quite well with hybrid functionals 
when we apply our techniques to molecules. In some cases we can convert 
our solid/condensed matter problem into a cluster approach and the results 
are quite good.

> Not very encouraging in any case ...
> Xavier

Yes, that's why I'm looking into other possibilities. I recently came 
across another functional (PRB 73, 235116 (2006)), simpler than TPSS, 
which gives some encouraging results. I'm in contact with the authors who 
have this functional already implemented in ABINIT. I'll probably post 
something in this regard after some test.

Best regards and thanks for the interest.

Fer.

> On 21 Aug 2006, at 22:49, fer@tiziano.phys.washington.edu wrote:
>
>> Hi..
>>
>>  As I have probably mentioned here before, we have been working on 
>> calculating EXAFS Debye-Waller factors using the dynamical matrices 
>> generated by abinit. The whole approach works quite well, but we are not 
>> totally satisfied by the results we get from either LDA (PW92) or GGA 
>> (PBE). For us it is quite crucial getting the interatomic force constant 
>> accurately. We have found that they depend a lot on the crystal lattice 
>> that we use, and therefore we need to be able to find a way to realiably 
>> predict both the crystal lattice and the force constants simultaneously. 
>> Its seems that neither PW92 nor PBE can do this.
>> 
>> We got in contact with John Perdew and he suggested we try the meta-GGA 
>> TPSS and we have tried to do so with the implementation already available 
>> in Gaussian 03. for reasons I don't what to discuss here, this 
>> implementation has several drawbacks when it comes to solid states.
>> 
>> Therefore, I was wondering if anyone has implemented TPSS into abinit 
>> already. I did a search on the source and found that nanoquantaxc has 
>> TPSS, 
>> but I really don't know how to integrate it into my binaries nor how to 
>> call it. Any experience, instructions, etc, etc, in this regards would be 
>> greatly welcome.
>> 
>> Best regards and thanks in advance, Fer.
>> 
>> Ubi dubium ibi libertas.
>> *******************************************************************************
>> Fernando D. Vila                Voice    (206)543-9697
>> Department of Physics           Fax      (206)685-0635
>> University of Washington        E-mail   fdv@u.washington.edu
>> Seattle, WA 98195, USA          WWW      http://faculty.washington.edu/fdv
>> *******************************************************************************
>


Ubi dubium ibi libertas.
*******************************************************************************
Fernando D. Vila                Voice    (206)543-9697
Department of Physics           Fax      (206)685-0635
University of Washington        E-mail   fdv@u.washington.edu
Seattle, WA 98195, USA          WWW      http://faculty.washington.edu/fdv
*******************************************************************************