If the number of bands is not prohibitively large you can try to run with the option -nd 1, which disables the usage of scalapack. I also noticed that compiling with elpa parallel diagonalization is more stable.
I hope this helps
Greetings Pietro
Il 26 apr 2019 5:10 PM, "IORI, Federico" <[email protected]> ha scritto:
Hi all.I found such error randomly in almost my calculations at the moment (no matter the size, the number of nodes or cpus used)Davidson diagonalization with overlap
forrtl: severe (71): integer divide by zero
Image PC Routine Line Source
pw.x 0000000000CFE5CE Unknown Unknown Unknown
libpthread-2.22.s 00002AAAB2C08C10 Unknown Unknown Unknown
libmkl_scalapack_ 00002AAAAB0F713F pdsbrdb_ Unknown Unknown
libmkl_scalapack_ 00002AAAAAF303B0 pdsyrdb_ Unknown Unknown
libmkl_scalapack_ 00002AAAAAF1AC6D mkl_pdsyevd0_ Unknown Unknown
libmkl_scalapack_ 00002AAAAAF194CC mkl_pdsyevdm_ Unknown Unknown
libmkl_scalapack_ 00002AAAAAF184E7 pdsyevd_ Unknown Unknown
pw.x 0000000000B63358 dspev_module_mp_p 741 dspev_drv.f90
pw.x 0000000000B5C1EC prdiaghg_ 317 rdiaghg.f90
pw.x 0000000000A9D02E pregterg_ 872 regterg.f90
pw.x 00000000006580CF diag_bands_ 346 c_bands.f90
pw.x 000000000065633B c_bands_ 101 c_bands.f90
pw.x 000000000040CAF1 electrons_scf_ 566 electrons.f90
pw.x 0000000000409C04 electrons_ 152 electrons.f90
pw.x 00000000005764EA run_pwscf_ 133 run_pwscf.f90
pw.x 00000000004077B5 MAIN__ 98 pwscf.f90
pw.x 000000000040761E Unknown Unknown Unknown
libc-2.22.so 00002AAAB3132725 __libc_start_main Unknown Unknown
pw.x 0000000000407529 Unknown Unknown UnknownDoes someone have an idea where it could come for ? is a matter of scalapack or from the diagonalizatino algoritm of pw ?Or maybe suggest a test to reproduce it ?Thanks and sorry to bother again with such issue.--Federico IORIComputational material scientist
Paris Innovation Campus
1 chemin de la Porte des Loges
Les Loges en Josas – 78354 Jouy en Josas cedex
Mail: [email protected]Phone: +33 7 621 605 15
_______________________________________________ Quantum Espresso is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list [email protected] https://lists.quantum-espresso.org/mailman/listinfo/users
