P2y error: when writting down ns.db1

[pandachang97]

Hello Yambo experts:
Good morning. I hope you are all doing well. Recently, I compiled yambo 5.2.4 with intel compiler and MKL library. The compilation is done without any errors. I also include compilation log files in the attachment in case you need to check. When I run the p2y to read the QE wavefunction(in the save folder), it reads the lattice vector, RL vectors and atom information. Then, when it is writting the ns.db1, it reports one error and stoped:
[ERROR] Writing File ./SAVE//ns.db1; Variable NOT DEFINED; NetCDF: HDF error
I also include my configure option below:

Code: Select all

 ./configure --prefix=/work/09846/pandachang97/Software/yambo_5.2.4_ifort_local_lib  MPIF90=mpiifort MPICC=mpiicc FC=ifort CC=icc F77=ifort MPIF77=ifort  --enable-parallel --with-blas-libs="-L$MKLROOT/lib/intel64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core"  \
--enable-mpi --enable-uspp --enable-options-check --enable-memory-profile  --enable-iotk --enable-hdf5-p2y-support  --with-blacs-libs="-L$MKLROOT/lib/intel64 -lmkl_blacs_intelmpi_lp64"   --with-scalapack-libs="-L$MKLROOT/lib/intel64 -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64"  \
 --with-scalapack-libs="-L$MKLROOT/lib/intel64 -lmkl_lapack95_lp64" --with-libxc=yes --with-libxc-prefix="/work/09846/pandachang97/Software/libxc-7.0.0"  --with-hdf5-libs=$TACC_PHDF5_LIB --with-hdf5-include=$TACC_PHDF5_INC

Please help me check them and let me know what proble I have in compiling Yambo or QE and in calculating the PWSCF. Thank you so much and best regards,
Xin

[pandachang97]

Hello Yambo experts,

An update. When I turn off the option, --enable-hdf5-p2y-support, this error disappeared (The other configuration options are the same). It seem that, in the src of p2y, there may be some bugs on reading the hdf5 files. Please check the source code of p2y. Thank you and best regards,
Xin

[andrea.ferretti]

Dear Xin,

thanks a lot for reporting. In order to investigate this issue further we need a few extra info:
* which version of QE have you been using ? was it compiled with hdf5 support (if you look into $prefix.save, .dat files indicate no hdf5 support, while .hdf5 or alike indicate that hdf5 support is active in QE.)
* can you send a reproducer of the problem ? (scf/nscf input files of QE)

thanks
Andrea

[pandachang97]

Hello Andrea,

Thank you for your quick response. For the QE calculation, I am now using 7.3.1 version (the newest) WITHOUT hdf5. Thus, in the save folder, they are all wfc(n).dat files. The reason I turn on the enalbe-p2y-hdf5-support is that I tried to compile QE with hdf5 support, I found with hdf5 the QE calculation is faster in geometry relax and phonon calculation. But, the problem is that the p2y can not read the wavefunction. So, yesterday, I turned off the hdf5 support in both QE and Yambo. The problem disappeared. I am thinking there may be bugs in the p2y source code to read the hdf5 files. I just attached my input file of QE scf calculation in the attachment. I am using QE-7.3.1 (compiled by myself with ifort and mkl library). The wavefunction is saved as wfc(n).dat. When I use p2y (without hdf5 support), it works well. Nothing error is reported. I hope you will find the bug soon. Best regards,
Xin

[andrea.ferretti]

Dear Xin,

thanks for the input.
While I'm trying to reproduce the problem, I've noticed that you use
ecutwfc=55, and ecutrho=440, which seems to point to the use of ultrasoft pseudos.
Is it the case or is it a typo ?
Anyway, only norm-conserving PPs are permitted when using yambo.

Andrea

[pandachang97]

Hello Andrea,

Thank you for your quick response. For the input, I follow the settings from my previous SCF calculation. I did not change them. I am not sure if these settings really change the SCF result or affect the Yambo calculation? Please let me know. I am a relative new user on PWSCF and Yambo package.
Also, as to the pseudopotential, I download them from pseudodojo, NC SC (ONCVPSP v0.5) for PBE. I double checked the files in the folder. It is not super soft type. Best regards,
Xin

[andrea.ferretti]

quick update:
using gfortran, and having changed the input to ecutwfc=55, ecutrho=220, it works with me.
I'll investigate further.

Concerning the point about norm-conserving PPs (NCPP):
* since rho(r) = sum |psi(r)|^2, the kinetic energy cutoff for the
charge-density is mathematically 4 times that of the wavefunctions
* so, for NPCC, the is no point, neither mathematical nor numerical, in using a
larger ratio (however hard the pseudos are)
* I'm actually wonder whether this is what generates issues in p2y... (it should not, but surely it is a little tested situation)

[pandachang97]

Hello Andrea,

Thank you so much for your quick response and instrunctions on QE. I will also have a try on our super cluster with rhocutoff=220. I am now using ifort to compile all packages with the MKL library. I am not sure if that is the internal confilicts between Yambo code and Intel compiler. Thank you again and best regards,
Xin