[QE-users] "vc-relax" and "relax" with hybrids

Giuseppe Mattioli giuseppe.mattioli at ism.cnr.it
Wed Apr 10 14:08:18 CEST 2019


> However, as I understand they are a bit outdated?

Well, not at all IMHO. In the "roaring twenties" of GGA, US PPs  
guaranteed a significant speed up of calculations, but the density  
cutoff was almost the same of NC. VASP steered towards PAW some years  
ago, and all the EXX machinery has been built to be PAW-compliant. But  
there is nothing "outdated" in norm-conserving, and the density cutoff  
may be the actual bottleneck of EXX calculations. The ecutfock  
variable is a very clever tunable parameter in this regard. You can  
verify that if ecutwfc is well converged, then ecutfock=ecutwfc is a  
sensible choice for *very fast* geometry optimizations, and  
ecutfock=1.5~2*ecutwfc is a sensible choice for properties  
calculation. ONCV pseudopotentials are not that "hard" (80 Ry should  
be a sensible ecutwfc for carbon), give them a try.

HTH
Giuseppe

> Few additional questions if You don't mind:
> - Should I expect constrained orbital calculations (delta SCF) work well
> with hybrids?

If you are calculating ionization energies and/or electron affinities  
no. Total energies for charged and neutral supercells are not additive  
because of PBC.

> - Any suggestions for hybrid parallelisation? We have shared memory 36 core
> mashine with hyperthreading (72 threads).
>   System size is 200-500 carbon atoms + defect.

I'm not an expert of k-points parallelization with hybrids. If your  
system is large (and your system seems to be quite large) and you use  
gamma-only sampling, then you can perform a preliminary test (e.g. a  
sinle point calculation of a relevant system) with band  
parallelization. You may also test the new implementation of SCDM  
localization

Experimental version of SCDM localization with k-points, activated like for
k=0 by specifying in &system namelist a value > 0 for "localization_thr".

HTH
Giuseppe

Lukas Razinkovas <lukas.razinkovas at ftmc.lt> ha scritto:

> Thank You, Paolo and Giusepe
>
> It is really awesome to get response from You :). I will try norm
> conserving pseudopotentials.
> However, as I understand they are a bit outdated?
>
> Few additional questions if You don't mind:
> - Should I expect constrained orbital calculations (delta SCF) work well
> with hybrids?
> - Any suggestions for hybrid parallelisation? We have shared memory 36 core
> mashine with hyperthreading (72 threads).
>   System size is 200-500 carbon atoms + defect.
>
> Thanks for Your time,
> Lukas
>
>
>
> On Wed, Apr 10, 2019 at 1:26 PM Giuseppe Mattioli <
> giuseppe.mattioli at ism.cnr.it> wrote:
>
>>
>> That is: you should use NC pseudopotentials! EXX works very well if
>> you learn a couple of further tricks (ecutfock, adaptive_thr).
>> HTH
>> Giuseppe
>>
>> Paolo Giannozzi <p.giannozzi at gmail.com> ha scritto:
>>
>> > The crash is due to a stupid error (a missing argument) in a call to the
>> > "errore" routine, but the routine is called because forces with USPP and
>> > PAW are not currently implemented with hybrid functionals
>> >
>> > Paolo
>> >
>> >
>> > On Wed, Apr 10, 2019 at 11:53 AM Lukas Razinkovas <
>> lukas.razinkovas at ftmc.lt>
>> > wrote:
>> >
>> >> Dear QE users and developers,
>> >>
>> >> I have read about Adaptively Compressed Exchange Operator (ACE) approach
>> >> to hybrid
>> >> functionals and that it is implemented in PWSCF code. This motivates
>> >> transition from VASP
>> >> to QE code for our research calculations. We are working with crystal
>> >> defects and are interested in modeling excited states.
>> >>
>> >> However it seems that only "scf" calculations are possible. Is it right?
>> >> I have tried "vc-relax" and "relax" calculations and it failed.
>> >>
>> >> *Input file:*
>> >>
>> >>  &CONTROL
>> >>                  calculation = 'vc-relax' ,
>> >>                 restart_mode = 'from_scratch' ,
>> >>                       outdir = 'tmp' ,
>> >>                   pseudo_dir = '../../psps/' ,
>> >>                       prefix = 'diamond' ,
>> >>                    verbosity = 'high'
>> >>  /
>> >>  &SYSTEM
>> >>                        ibrav = 1,
>> >>                    celldm(1) = 6.755024314,
>> >>                          nat = 8,
>> >>                         ntyp = 1,
>> >>                      ecutwfc = 45 ,
>> >>                      ecutrho = 350 ,
>> >>                    input_dft = HSE06,
>> >>    nqx1 = 4, nqx2 = 4, nqx3 = 4
>> >>  /
>> >>  &ELECTRONS
>> >>              diagonalization = 'david' ,
>> >>  /
>> >>  &IONS
>> >>  /
>> >>  &CELL
>> >>  /
>> >> ATOMIC_SPECIES
>> >>     C   12.01070  C.pbe-n-kjpaw_psl.1.0.0.UPF
>> >> ATOMIC_POSITIONS alat
>> >>     C      0.000000000    0.000000000    0.000000000
>> >>     C      0.500000000    0.500000000    0.000000000
>> >>     C      0.000000000    0.500000000    0.500000000
>> >>     C      0.500000000    0.000000000    0.500000000
>> >>     C      0.250000000    0.250000000    0.250000000
>> >>     C      0.750000000    0.750000000    0.250000000
>> >>     C      0.750000000    0.250000000    0.750000000
>> >>     C      0.250000000    0.750000000    0.750000000
>> >> K_POINTS automatic
>> >>   4 4 4   0 0 0
>> >>
>> >>
>> >> *gives error:*
>> >>
>> >> forrtl: severe (174): SIGSEGV, segmentation fault occurred
>> >> Image              PC                Routine            Line
>> >> Source
>> >> pw.x               0000000000D0C1D3  Unknown               Unknown
>> Unknown
>> >> libpthread-2.17.s  00007F9B508AF5D0  Unknown               Unknown
>> Unknown
>> >> pw.x               0000000000B9ED47  errore_                    10
>> >> error_handler.f90
>> >> pw.x               00000000005A8E2F  setup_                    137
>> >> setup.f90
>> >> pw.x               0000000000579E23  run_pwscf_                 99
>> >> run_pwscf.f90
>> >> pw.x               000000000040AA85  MAIN__                     98
>> >> pwscf.f90
>> >> pw.x               000000000040A8E2  Unknown               Unknown
>> Unknown
>> >> libc-2.17.so       00007F9B501F33D5  __libc_start_main     Unknown
>> >> Unknown
>> >> pw.x               000000000040A7E9  Unknown               Unknown
>> Unknown
>> >> forrtl: severe (174): SIGSEGV, segmentation fault occurred
>> >> Image              PC                Routine            Line
>> >> Source
>> >> pw.x               0000000000D0C1D3  Unknown               Unknown
>> Unknown
>> >> libpthread-2.17.s  00007F08180CB5D0  Unknown               Unknown
>> Unknown
>> >> pw.x               0000000000B9ED47  errore_                    10
>> >> error_handler.f90
>> >> pw.x               00000000005A8E2F  setup_                    137
>> >> setup.f90
>> >> pw.x               0000000000579E23  run_pwscf_                 99
>> >> run_pwscf.f90
>> >> pw.x               000000000040AA85  MAIN__                     98
>> >> pwscf.f90
>> >> pw.x               000000000040A8E2  Unknown               Unknown
>> Unknown
>> >> libc-2.17.so       00007F0817A0F3D5  __libc_start_main     Unknown
>> >> Unknown
>> >> pw.x               000000000040A7E9  Unknown               Unknown
>> Unknown
>> >>
>> >> What are the main reasons for this error/behavior? Is it possible to
>> >> enable this functionality?
>> >>
>> >> Another side question is: is is possible to use constrained orbital
>> >> calculations (delta SCF) with hybrids?
>> >>
>> >>
>> >> Lukas
>> >> _______________________________________________
>> >> users mailing list
>> >> users at lists.quantum-espresso.org
>> >> https://lists.quantum-espresso.org/mailman/listinfo/users
>> >
>> >
>> >
>> > --
>> > Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
>> > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
>> > Phone +39-0432-558216, fax +39-0432-558222
>>
>>
>>
>> GIUSEPPE MATTIOLI
>> CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
>> Via Salaria Km 29,300 - C.P. 10
>> I-00015 - Monterotondo Scalo (RM)
>> Mob (*preferred*) +39 373 7305625
>> Tel + 39 06 90672342 - Fax +39 06 90672316
>> E-mail: <giuseppe.mattioli at ism.cnr.it>
>>
>> _______________________________________________
>> users mailing list
>> users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>



GIUSEPPE MATTIOLI
CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
Via Salaria Km 29,300 - C.P. 10
I-00015 - Monterotondo Scalo (RM)
Mob (*preferred*) +39 373 7305625
Tel + 39 06 90672342 - Fax +39 06 90672316
E-mail: <giuseppe.mattioli at ism.cnr.it>



More information about the users mailing list