Page 1 of 1

Variable nhtoj; Numeric conversion not representable in nscf calculation with uspp

Posted: Sat Nov 27, 2021 11:49 am
by minyez
Dear developers,

In attempt to prepare a database with ultrasoft psedopotential from the non-SCF calculation for el-ph, I obtained the following error with `p2y`

Code: Select all

 <---> DBs path set to : .
 <---> detected QE data format : qexsd-hdf5
 <---> == PWscf v.6.x generated data (QEXSD-HDF5 fmt) ==
 <---> Header/K-points/Energies... done
 <---> Cell data...
 <---> [WARNING]Two or more crystal structures fit the given cell done
 <---> Atomic data... done
......
 <---> == DB1 (Gvecs and more) ...
 <---> ... Database done
 <---> == DB2 (wavefunctions)  ...
 <---> [p2y] WF I/O |                                        | [000%] --(E) --(X
 <---> [p2y] WF I/O |########################################| [100%] --(E) --(X) done ==
 <---> == DB3 (PseudoPotential) ...               file ./Cu.pbesol-dn-rrkjus_psl.1.0.0.UPF: wavefunction(s)  3D renormalized

[ERROR] Writing File ./SAVE//ns.uspp_pp_pwscf; Variable nhtoj; NetCDF: Numeric conversion not representable
This seems to make the database incomplete, and the following dvscf ypp_ph will fail.
Using `ncdump -v nhtoj SAVE/ns.uspp_pp_pwscf` gives the output

Code: Select all

dimensions:
	D_0000000003 = 3 ;
	D_0000000001 = 1 ;
	D_0000000002 = 2 ;
	D_0000000005 = 5 ;
	D_0000000004 = 4 ;
	D_0000000010 = 10 ;
	D_0000000006 = 6 ;
	D_0000000018 = 18 ;
variables:
	......
	float nhtoj(D_0000000003, D_0000000018) ;
data:

 nhtoj =
  0, 3.370452e+30, 0, 0, -Infinityf, 0, -Infinityf, 0, 0, 0, 0, 0, 0, 0, 0,
    Infinityf, 0, 0,
  Infinityf, 0, 0, -Infinityf, -5.6295e+14, 0, 0, NaNf, NaNf, NaNf, 0, 0,
    Infinityf, Infinityf, 0, 0, 0, 0,
  0, 0, 0, 0, 0, Infinityf, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 5.140654e+25, 0 ;
From this link (https://code.mpimet.mpg.de/boards/1/topics/588), it seems that the error is related to Inf and NaN.
By the way, what is the meaning of the variable `nhtoj`?

The pseudopotentials I used are the USPP from pslib v1.0.0 downloaded from the QE official site, namely
- Cu.pbesol-dn-rrkjus_psl.1.0.0.UPF http://www.quantum-espresso.org/upf_fil ... .1.0.0.UPF
- Al.pbesol-n-rrkjus_psl.1.0.0.UPF http://www.quantum-espresso.org/upf_fil ... .1.0.0.UPF
- O.pbesol-n-rrkjus_psl.1.0.0.UPF http://www.quantum-espresso.org/upf_fil ... .1.0.0.UPF

Versions are QE 6.8 and Yambo 5.0.4 compiled along with `--enable-uspp`.

With NCPP (pseudo-dojo), everything are okay. Thank you very much for your continuous help and support!
Please let me know if you need more information to reproduce the error.

Sincerely,
Min-Ye Zhang

Re: Variable nhtoj; Numeric conversion not representable in nscf calculation with uspp

Posted: Sat Nov 27, 2021 12:29 pm
by Daniele Varsano
Dear Min-Ye,
thanks for reporting, we will have a look.
Anyway, let me stress that the use of uspsp is somehow experimental and thus discouraged.
We started the porting some years ago, but then it was not developed further also because ONCVPSP pseudopotentials work very well and they are not so hard.

Best,
Daniele

Re: Variable nhtoj; Numeric conversion not representable in nscf calculation with uspp

Posted: Sat Nov 27, 2021 1:22 pm
by minyez
Dear Daniele,

Thank you very much for the reply. I started with the pslib uspp as they are recommended by the QE official and by one of my colleagues. I will then try the ONCVPSP series.

By the way, may I ask your experience about setting ecutwfc of Cu and O with ONCVPSP, for tasks like el-ph and BSE?
I have an idea that for localized Cu 3d and O 2p, the cut-off of NCPP is rather large.

Best Regards,
Min-Ye

Re: Variable nhtoj; Numeric conversion not representable in nscf calculation with uspp

Posted: Sun Nov 28, 2021 10:52 am
by Daniele Varsano
Dear Min-Ye,
you need to perform convergence tests to have a well-converged ground state and phonons, so standard convergences for DFT calculations. Next the cutoff used to converge the screening for GW/BSE e.g. the cutoff for dielectric matrix are about one order of magnitude lower.

Best,
Daniele