Hybrid functionals

[maurosgroi]

Dear all,
I've searched the forum with no results. I hope this argument was not already covered.
I'd like to use Abinit with the B3LYP and PBE0 hybrid functionals. My question is: are they implemented only for PAW or also for NC pseudopotentials?
Moreover, should I simply put ixc -403 for B3LYP? Nothing else to check?
From a general point of view I should generate a PSP or a PAW using the B3LYP functional. Is it acceptable to use LDA or GGA pseudopotentials or PAW?
Thanks a lot and best regards,
Mauro Sgroi.

[jzwanzig]

The current release notes for the 7.0 family address this point, the use of hybrids appears to be still under development but at quite an advanced state.

Generally speaking it would be a bad idea to use pseudopotentials made with one xc choice together with a calculation using a different xc choice, you should be consistent with both.

[maurosgroi]

Dear jzwanzig,
thanks a lot for the explanation. My main problem is that when I try to use ixc=-406 for PBE0 or ixc=-402 for B3LYP the code crashes and exit.
I'm using NC pseudopotentials. Is this behaviour related to the fact that the code cannot perform ground state calculations with LIBXC hybrid functionals? Or I've a problem in the compilation?
Thanks a lot in advance.
Best regards,
Mauro.

The crash is reported in the following:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
abinit 00000000018113D4 Unknown Unknown Unknown
abinit 0000000000F81CAA Unknown Unknown Unknown
abinit 00000000005E0986 Unknown Unknown Unknown
abinit 00000000005B4A44 Unknown Unknown Unknown
abinit 00000000005B6363 Unknown Unknown Unknown
abinit 000000000055D7F3 Unknown Unknown Unknown
abinit 000000000048718C Unknown Unknown Unknown
abinit 00000000004769E5 Unknown Unknown Unknown
abinit 000000000046D0C7 Unknown Unknown Unknown
abinit 000000000046B30C Unknown Unknown Unknown
libc.so.6 0000003074A1D9C4 Unknown Unknown Unknown
abinit 000000000046B219 Unknown Unknown Unknown
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
abinit 00000000018113D4 Unknown Unknown Unknown
abinit 0000000000F81CAA Unknown Unknown Unknown
abinit 00000000005E0986 Unknown Unknown Unknown
abinit 00000000005B4A44 Unknown Unknown Unknown
abinit 00000000005B6363 Unknown Unknown Unknown
abinit 000000000055D7F3 Unknown Unknown Unknown
abinit 000000000048718C Unknown Unknown Unknown
abinit 00000000004769E5 Unknown Unknown Unknown
abinit 000000000046D0C7 Unknown Unknown Unknown
abinit 000000000046B30C Unknown Unknown Unknown
libc.so.6 0000003074A1D9C4 Unknown Unknown Unknown
abinit 000000000046B219 Unknown Unknown Unknown
.....continues...

[jzwanzig]

I strongly suspect that the GS calculation simply isn't coded for hybrid functionals. My understanding from the release notes is that hybrid functionals are being coded into the GW part of the code at this point. Next week the abinit developers are having a workshop and I'll get an update there, I'll post more if I learn more about this issue.

Joe

[JEJohns]

Hi Joe and Mauro,
I'm also very interested in this topic. In particular, it'd be great to be able to optimize structures using the hybrid functionals. I don't know if this helps, Mauro, but the program also crashes if you try HSE06 (ixc=-428). Oddly, this doesn't happen for all hybrids (MN12-SX and N12-SX) but the energy for those calculations oscillates wildly and just doesn't converge, no idea what's going on under the hood. I also have no idea how to get an appropriate psuedopoential for the hybrids. Neither Opium nor APE nor fhi98pp calculate any HF or exact exchange, so that wouldn't that information be missing if you just used a standard GGA psuedopotential in a ground state calculation? I know that the PAW datasets from VASP are supposed to be useable with hybrids and contain kinetic information for use with MetaGGAs within that program, but I'd love hearing about an open source code that offered that functionality.
James

[maurosgroi]

Dear jzwanzig and JEJohns,
thanks a lot for the explanations.
I also suspect that hybrid functionals are not code for ground states.
Moreover I tried to use Atompaw to generate PAW based on hybrid functionals, but also on that case the hybrids are not code and the libxc support is not complete.
I'm not expert and I would like to figure out why is it difficult to include hybrid functionals in planewaves codes.
Best regards,
Mauro.

[jzwanzig]

I don't think it's conceptually difficult to include hybrid functionals, it's just very inefficient to compute the exact exchange in reciprocal space using a planewave basis. The issues are discussed for example in Bylaska et al., J. Comp. Chem. 32, 54 (2010).

[maurosgroi]

Dear jzwanzig,
many thanks for the suggestion.
I will read that article.
Best regards,
Mauro.

[JEJohns]

Hi Joseph,
I Feel really stupid for this post now about the hybrids. The features web page is pretty clear about the hybrids
"1. Available physical properties
1.A. Computation of the total energy of an assembly of nuclei and electrons placed in a repeated cell.
1.A.1. The computation can be done using (1) plane waves and norm-conserving pseudopotentials, or, (2) so-called "Projector-Augmented Waves" (PAW method), with appropriate pseudoatomic data, or (3) wavelets, see 1.A.14.

1.A.2. The total energy computation is done according to the Density Functional Theory. Many exchange-correlation functionals are available, through two different implementations : one is the native ABINIT implementation, the other is the ETSF library of XC functionals (LibXC is a plug-in to ABINIT). In the native ABINIT set, most of the important local approximations (LDA) are available, including the Perdew-Zunger one. Two different local spin density (LSD) are available, including the Perdew Wang 92, and one due to M. Teter. The Perdew-Burke-Ernzerhof, the revPBE, the RPBE and the HCTH GGAs (spin unpolarized as well as polarized) are also available. In the LibXC 2.0 library, as interfaced with ABINIT, there are 24 functional forms of the 3D LDA type, and over 80 functional forms of the GGA type. They can be used with norm-conserving pseudopotentials as well as PAW atomic data. Also, some metaGGA can be used with ABINIT (norm-conserving case only). For response-function type calculations, the native ABINIT LDA and GGA kernels can be used as well as the LibXC ones. The exchange can also be computed on the basis of the Fock expression (exact exchange), and the correlation can be computed on the basis of the RPA approximation (see the GW section). Hybrid functionals calculations (HSE06, PBE0, B3LYP) can be performed. The implementation of the exact exchange, correlation and hybrid does not deliver the forces and stresses at present. Several Van Der Waals functionals are available : Grimme, Silvestrelli."

That's not generally the place I go to look for information, and unfortunately, this isn't listed in either the release notes or under the input variables for ixc. It might be helpful to list that info in a couple of other spots. Either way, I should have found it earlier. Sorry to have bugged you,
James

[hatuey]

I am also interested in using hybrid functionals.

Looking in the documentation for ixc, I did not found any information about setting ixc=-406 for PBE0, ixc=-402 for B3LYP or ixc=-428 for HSE06.

In the online documentations there is a gap: jump from 200 to 524. From where did you get those values?

Regards,

Hatuey

[Boris]

Hi everyone

The implementation of hybrid functionals are still under development, but you can still use a simplified version for f electrons, as has been done here: http://prb.aps.org/abstract/PRB/v80/i23/e235109.

It applies the exact exchange on selected electrons in the PAW spheres.

[tongclub61]

you should be consistent with both.
ทางเข้า sbobet