[QE-users] Regarding the optimization of SCF calculations in supercells

[Paolo Giannozzi]

On 22/05/2025 22:38, Mateus Corradini Lopes wrote:

> export OMP_NUM_THREADS=48.

too many!

Paolo

> mpirun -np 10 pw.x -in input/“$INPUT” -nb 1 -nt 2 > output/"$OUTPUT”
> 
> 
> I'm trying to run MPI+openMP in a hybrid way, since using pure MPI has 
> prohibitively increased the use of RAM.  Regarding the second point, I 
> used K_POINTS gamma for the supercell calculations, I don't know if 
> that's exactly your question. As for ecutwfc and ecutrho, I used 50 Ry 
> and 400 Ry respectively. Regarding the fourth point, as these are steps 
> in a finite displacement process, at least one coordinate will be subtly 
> disturbed, referring to the displacement of the atom of the method 
> itself, but in fact as you mentioned, there were other coordinates that 
> were a little disturbed due to the relaxation I did with vc-relax before 
> generating the supercell, which changed the atomic positions a little 
> for these strange coordinates. I'm already testing the calculation 
> without this relaxation to see if it improves. I'm still looking for 
> more suggestions to improve the calculation.
> 
> Sincerely,
> 
> M.Sc. Mateus Corradini Lopes
> PhD. student
> Gleb Wataghin Institute of Physics - Unicamp
> 
> 
> Em 22/05/2025 16:34, Giovanni Cantele escreveu:
>> Hi,
>>
>> first of all it would be nice also to have the command you use to run 
>> pw.x (because if you for example use pools, this replicates the memory 
>> requirements).
>>
>> Second: which k-point grid do you use to sample the Brillouin zone? In 
>> the case of a 1x1x1 grid, do you use gamma-only algorithms?
>>
>> Third: which ecutwfc and ecutrho do you use?
>>
>> Fourth: I understand you need a "perfect" supercell as derived from 
>> replicating a bulk cubic cell. Coordinates like 0.0013807042891237 
>> or 0.1874999906250000 are harmful (unless you need to break symmetry 
>> or they are exactly what you want), in that they prevent pw.x from 
>> finding symmetries. The latter might be quite helpful in (sometimes 
>> even significantly) reducing the calculation load (in terms of k- 
>> points but not only!). So, I would round all coordinates so as to 
>> obtain exactly the number you expect.
>>
>> Hope this helps.
>> Giovanni
>>
>> -- 
>>
>> Giovanni Cantele, PhD
>> CNR-SPIN
>> c/o Dipartimento di Fisica
>> Universita' di Napoli "Federico II"
>> Complesso Universitario M. S. Angelo - Ed. 6
>> Via Cintia, I-80126, Napoli, Italy
>> e-mail: giovanni.cantele at spin.cnr.it <mailto:giovanni.cantele at spin.cnr.it>
>> Phone: +39 081 676910
>> Skype contact: giocan74
>>
>> ResearcherID: http://www.researcherid.com/rid/A-1951-2009
>> Web page: https://sites.google.com/view/giovanni-cantele/home
>>
>>
>> Il giorno gio 22 mag 2025 alle ore 20:03 Mateus Corradini Lopes 
>> <mclopes at ifi.unicamp.br> ha scritto:
>>
>>     Dear all,
>>
>>     I have recently been working on obtaining transport properties for
>>     certain types of materials. One of the parts of the work is to obtain
>>     the lattice thermal conductivity (\kappa_latt), so I'm trying to use
>>     phono3py+Quantum Espresso to calculate the supercell and obtain the
>>     force constants.
>>
>>     At the moment I'm trying to calculate the \kappa_latt of GaAs as a
>>     test,
>>     I've already managed to generate the supercells (4x4x4 and 3x3x3
>>     based
>>     on the conventional GaAs cell, which gives 512 and 216 atoms
>>     respectively) and I've used phono3py's --cutoff-pair to reduce the
>>     number of inputs, but even though I've tried various ways of
>>     running the
>>     calculation I haven't found a way to make it feasible in terms of
>>     computational time and RAM (the lowest I could get was “Estimated
>>     total
>>     dynamical RAM > 115.43 GB”). I have access to a cluster with 480
>>     processors (10 nodes with 48 processors each) and I would like to
>>     know
>>     what is the best way to run this system, in terms of optimizing the
>>     input and the parallelism used, balancing precision and
>>     computational time.
>>
>>     Ps. I'm attaching an example of the input I'm currently using.
>>
>>     Thank you in advance,
>>
>>     M.Sc. Mateus Corradini Lopes
>>     PhD. student
>>     Gleb Wataghin Institute of Physics - Unicamp
>>     _______________________________________________________________________________
>>     The Quantum ESPRESSO Foundation stands in solidarity with all
>>     civilians worldwide who are victims of terrorism, military
>>     aggression, and indiscriminate warfare.
>>     --------------------------------------------------------------------------------
>>     Quantum ESPRESSO is supported by MaX (www.max-centre.eu <http://
>>     www.max-centre.eu>)
>>     users mailing list users at lists.quantum-espresso.org
>>     https://lists.quantum-espresso.org/mailman/listinfo/users
>>
>>
>> _______________________________________________________________________________
>> The Quantum ESPRESSO Foundation stands in solidarity with all civilians worldwide who are victims of terrorism, military aggression, and indiscriminate warfare.
>> --------------------------------------------------------------------------------
>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
>> users mailing listusers at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
> 
> _______________________________________________________________________________
> The Quantum ESPRESSO Foundation stands in solidarity with all civilians worldwide who are victims of terrorism, military aggression, and indiscriminate warfare.
> --------------------------------------------------------------------------------
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
> 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 206, 33100 Udine Italy, +39-0432-558216