[Wien] Problems with hybrid calculation
tran at theochem.tuwien.ac.at
tran at theochem.tuwien.ac.at
Tue Mar 3 10:16:31 CET 2015
I still don't understand why the script run_lapw does not find the ibz and
fbz files. Maybe not related to this, what are you specifying for the
SCRATCH directory? With hybrid functionals it is mandatory to use ./
(the directory of the calculation) for SCRATCH because of the files
case.vectorhf(_old).
It is difficult to say for the computation time, but it will certainely
be (too?) huge for a self-consistent calculation. The problem with
Hartree-Fock is that it consists of two loops over orbitals which makes
the calculations very expensive for plane-wave type basis sets.
Nevertheless, you have to consider a few points:
-Do the calculation with only one k-point. Maybe the accuracy is enough,
in particular if your system is non-metallic.
-Use the onsite-hybrid functionals (as cheap as LDA/GGA) if your system
contains strongly localized d- or f-electrons.
-Use a proper .machines file. It is useless to have more lines in
.machines than the number of k-points in the IBZ. For one k-point
calculation this should be only one line (specifying the list of 96
processors for the MPI). More details in the UG.
-If you are interested only in the electronic structure, then you
should consider the diagonal approximation (-diaghf), see the UG,
which consists of only one (and faster) iteration.
F. Tran
On Tue, 3 Mar 2015, Paul Fons wrote:
> Hi I did run init_hf_lapw and I saw no sign of errors upon running it, however, I ran it from the web interface the first time. I just
> restarted the calculation using bash and have confirmed the ibz and fbz files exist.The reason that I am interested in hybrid mode is that
> I actually have 96 processors available with infiniband interconnects. Is a 96 atom system too expensive to run with PBE0 even with 96
> cores? How much time should I expect such a calculation to take?
>
> Paul Fons
>
>
> On Mar 3, 2015, at 4:56 PM, tran at theochem.tuwien.ac.at wrote:
>
> Hi,
>
> init_hf_lapw executes run_kgenhf_lapw which creates the files
> case.klist_ibz, case.klist_fbz, case.kgen_ibz and case.kgen_fbz.
> So, I don't understand why they are not present in your directory.
> Are you sure that init_hf_lapw ran without problems?
>
> Beside this, I really think that you should forget about running hybrid
> functionals for 96 atoms, in particular with this small number of
> processors (unless you are willing to wait until retirement).
> Depending on the system/properties that you are considering,
> the use of a GGA functional maybe give results which are fairly reliable.
>
> F. Tran
>
> On Tue, 3 Mar 2015, Paul Fons wrote:
>
> Hi All,
> I am trying to calculate the density of states of a small cluster of 96 atoms. I am using the hybrid mpi/k-point
> mode. The
> calculation ran to completion without problems for PBE. My .machines file is listed below. As the cluster is amorphous
> and
> roughly cubic in shape, I used a 2x2x2 MP mesh. I then attempted to use the PBE0 hybrid functional. After using the
> init_hf_lapw script, I started the SCF loop. The calculation ran for several hours and then stopped. I have attached
> the STDOUT
> output below. I am using intel ifort (15) as well as the intel mpi environment. I do have a scratch directory set up
> on this
> same node.
> These calculations were run on a single 24 core machine. I have five of these machines connected by Infiband and would
> like to
> scale up the calculation to use several nodes if I can find and solve the hybrid problem as I am aware of the cost of
> the PBE0
> calculations.
> Any advice?
> Best wishes,
> Paul Fons
> .machines
> lapw0:localhost:24
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> 1:localhost:2
> granularity:1
> extrafine:1
> STDOUT
> LAPW0 END
> LAPW0 END
> cp: cannot stat `aCGT.kgen_fbz': No such file or directory
> cp: cannot stat `aCGT.klist_fbz': No such file or directory
> LAPW1 END
> mv: cannot stat `aCGT.vector': No such file or directory
> cp: cannot stat `aCGT.kgen_ibz': No such file or directory
> cp: cannot stat `aCGT.klist_ibz': No such file or directory
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> LAPW1 END
> cp: cannot stat `aCGT.kgen_fbz': No such file or directory
> cp: cannot stat `aCGT.klist_fbz': No such file or directory
> mv: cannot stat `aCGT.vectorhf_old': No such file or directory
> forrtl: severe (24): end-of-file during read, unit 1001, file /usr/local/share/wien2k/Fons/aCGT/031
> Image PC Routine Line Source
> lapw2c 0000000000648027 Unknown Unknown Unknown
> lapw2c 0000000000669FF3 Unknown Unknown Unknown
> lapw2c 0000000000482FEE outp_ 180 outp.f
> lapw2c 00000000004701EC l2main_ 2125 l2main_tmp_.F
> lapw2c 000000000047CEA5 MAIN__ 607 lapw2_tmp_.F
> lapw2c 000000000040407E Unknown Unknown Unknown
> libc.so.6 000000343F61ED5D Unknown Unknown Unknown
> lapw2c 0000000000403F89 Unknown Unknown Unknown
> > stop error
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at: http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>
>
> Dr. Paul FonsSenior Research Scientist
> Functional Nano-phase-change Research Team
> Nanoelectronics Research Institute
> National Institute for Advanced Industrial Science & TechnologyMETI
> AIST Central 4, Higashi 1-1-1Tsukuba, Ibaraki JAPAN 305-8568
> tel. +81-298-61-5636fax. +81-298-61-2939
> email: paul-fons at 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
> ナノエレクトロニクス研究部門
> 相変化新規機能デバイス研究チーム
> 上級主任研究員
> ポール・フォンス
>
>
>
>
>
>
>
>
More information about the Wien
mailing list