Page 1 of 2
Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 2:10 am
by jasonhan0710
Dear Developers,
I want to get the optical properties by yambo_ph, however, an error happens in BSE calculations, which reports,
Code: Select all
[03.04] BSE solver(s) @q1
=========================
[03.04.01] Diago Solver @q1
===========================
[WARNING]Parallel non-herm diag not yet implemented. Switching to LK
(END)
and the output file reports
Code: Select all
** On entry to CGEBAL parameter number 3 had an illegal value
the log file reports
Code: Select all
<03s> P1-c39.hpc: [MEMORY] Alloc BS_V_left( 136.1110 [Mb]) TOTAL: 275.5410 [Mb] (traced) 63.61200 [Mb] (memstat)
<03s> P1-c39.hpc: [MEMORY] Alloc BS_V_right( 136.1110 [Mb]) TOTAL: 411.6520 [Mb] (traced) 63.61200 [Mb] (memstat)
<03s> P1-c39.hpc: [MEMORY] Alloc BS_overlap( 136.1110 [Mb]) TOTAL: 547.7630 [Mb] (traced) 63.61200 [Mb] (memstat)
<03s> P1-c39.hpc: BSK diagonalize | | [000%] --(E) --(X)
The input file was uploaded as an attachment.
Would you please help me solve the problem? Thank you!
Best,
Jason
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 10:08 am
by Daniele Varsano
Dear Jason,
can you please also attach your report file ?
Best,
Daniele
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 3:19 pm
by jasonhan0710
Dear Daniele,
Thanks for your reply. The report file and log file are as attached.
Best,
Jason
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 3:29 pm
by Daniele Varsano
Dear Jason,
it seems to me there is something wrong in your setup.
You have 2Si atom in a FCC unit cell but yambo sees the system having 12 metallic bands:
Something wrong in the QE calculations or in the yambo setup?
Best,
Daniele
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 4:16 pm
by jasonhan0710
Dear Daniele,
Thank you for your help. I set a wrong Bosonic temperature in the GW and BSE calculation. The Bosonic temperature should be 300 K rather than 300 eV, which leads to a metallic system. However, when I change the input parameter and recalculate the quasiparticle energy and optical properties, the error still happened.
Code: Select all
Intel MKL ERROR: Parameter 3 was incorrect on entry to CGEBAL
The new report file was attached.
Best,
Jason
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 4:31 pm
by Daniele Varsano
Quite strange,
can you try to not include the coupling terms in the BSE (which should not be relevant for silicon) and see if it works?
Best,
Daniele
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 5:02 pm
by jasonhan0710
Dear Daniele,
I have changed the
Code: Select all
BSEmod= "resonant" # [BSE] resonant/retarded/coupling
, it didn't report any error but it still stopped at the same place
Code: Select all
[04] BSE solver(s) @q1
======================
[04.01] Diago Solver @q1
========================
Best,
Jason
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Mon Oct 11, 2021 5:29 pm
by Daniele Varsano
Dear Jason,
1. any message in the log file?
2. Have you deleted previous ndb.BS* databases before running the BS in resonant mode?
3. Are you using yambo or yambo_ph executable?
Best,
Daniele
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Tue Oct 12, 2021 1:31 pm
by Davide Sangalli
Dear all,
it could also be a memory issue.
In case you do not need the BSE eigenvectors you can try the Haydock solver.
Otherwise, you can consider the iterative solvers based on the SLEPC library.
They both work in parallel, are more efficient and much better at distributing memory over MPI tasks.
Best,
D.
Re: Errors in BSE calculation at Diago Solver @q1
Posted: Thu Oct 14, 2021 8:57 am
by jasonhan0710
Dear Daniele,
Thanks for your reply.
I have tried to use Haydock solver, however, it reports
Code: Select all
[03.04] BSE solver(s) @q1
=========================
[WARNING]K with QP widths not supported in the Haydock solver
Is there and special requirement in the configuration process for yambo-5.0? Sometimes, the same case works well with yambo-4.5 but cannot get the correct results in yambo-5.0. For example, in 2D-hBN cases, in the GW process, yambo 4.5 gives the correct Eigen energy
Code: Select all
#
#
# GPL Version 4.5.3 Revision 9122. (Based on r.17117 h.03e635bbc)
#
# MPI+SLK+OpenMP Build
# http://www.yambo-code.org
#
# GW [Newton Solver]
#
# == PPA-GN GW ==
#
# PPA imaginary pt [ev]: 27.21138
# RIM G`s : 111
# RIM random pts : 1000000
# GW SC iterations :0
# dS/dw steps : 2
# dS/dw step [ev]: 0.10000
# X G`s [used]: 111
# X G`s [disk]: 111
# X bands : 1 20
# X poles [o/o]: 100.0000
# X e/h E range [ev]:-1.000000 -1.000000
# X xc-Kernel : none
# X BZ energy Double Grid: no
# X BZ Double Grid points:0
# Sc/G bands : 1 20
# Sc/G damping [ev]: 0.10000
# Sc bands terminator : no
# Sx RL components : 4971
#
# QP @ K 1 - 70 : b 1 - 20
#
# K-point Band Eo E-Eo Sc|Eo
#
1.00000 1.00000 -17.66856 -1.39721 8.58841
1.000000 2.000000 -5.272483 -1.490242 3.677860
1.000000 3.000000 -1.339043 -0.622278 3.701504
1.000000 4.000000 -1.339034 -0.622376 3.701499
1.000000 5.000000 4.724442 0.864567 -1.377525
1.000000 6.000000 5.651101 0.512426 -0.812870
1.000000 7.000000 6.069533 -0.196305 -0.524733
1.000000 8.000000 7.327153 0.281225 -0.666494
1.000000 9.000000 7.845431 0.172939 -0.823885
...
However, in yambo-5.0, it gives 'NaN' in the output file
Code: Select all
#
# Version 5.0.3 Revision 19584 Hash 1cce2ba63
# Branch is
# MPI+OpenMP+SLK+HDF5_IO Build
# http://www.yambo-code.org
#
#
# GW solver : Newton
# GW approximation : PPA
# PPA imaginary Energy : 27.2113800 [ev]
# RIM G`s : 111
# RIM random pts : 1000000
# GW SC iterations : 0
# dS/dw steps : 2
# dS/dw step : 0.100000001 [ev]
# X G`s : 111 [used]
# X G`s : 111 [disk]
# X bands : 1 20
# X poles : 100.000000 [o/o]
# X e/h E range : -1.00000000 -1.00000000 [ev]
# X xc-Kernel : none
# X BZ energy Double Grid : no
# Sc/G bands : 1 20
# Sc/G damping : 0.100000001 [ev]
# Sc bands terminator : no
# Sx RL components : 4971
#
# QP @ state[ 1 ] K range : 1 70
# QP @ state[ 1 ] b range : 1 20
# GF energies kind : Perdew, Burke & Ernzerhof(X)+Perdew, Burke & Ernzerhof(C)
# GF WF`s kind : Perdew, Burke & Ernzerhof(X)+Perdew, Burke & Ernzerhof(C)
# Xs energies kind : Perdew, Burke & Ernzerhof(X)+Perdew, Burke & Ernzerhof(C)
# Xs WF`s kind : Perdew, Burke & Ernzerhof(X)+Perdew, Burke & Ernzerhof(C)
#
# Vxc =Perdew, Burke & Ernzerhof(X)+Perdew, Burke & Ernzerhof(C)
# Vnlxc=Hartree-Fock
#
# K-point Band Eo [eV] E-Eo [eV] Sc|Eo [eV]
#
1 1 -17.66856 NaN NaN
1 2 -5.272483 NaN NaN
1 3 -1.339043 NaN NaN
1 4 -1.339034 NaN NaN
1 5 4.724442 NaN NaN
1 6 5.651101 NaN NaN
1 7 6.069533 NaN NaN
1 8 7.327153 NaN NaN
1 9 7.845431 NaN NaN
1 10 9.999383 NaN NaN
1 11 10.64105 NaN NaN
1 12 11.70873 NaN NaN
1 13 11.70874 NaN NaN
1 14 12.78220 NaN NaN
1 15 13.55611 NaN NaN
1 16 15.08026 NaN NaN
1 17 16.13510 NaN NaN
I don't know what's the difference between v4.5 and v5.0 rely on HPC environment? Could it be a reason why yambo_ph in v5.0 doesn't work well?
Best,
Jason