Page 1 of 1

NetCDF: Numeric conversion not representable when BSE

Posted: Sun Jul 14, 2019 9:01 am
by chwolf
Dear all,

I have MgO with 72 k-points and 200 KS bands. I can run IPA, RPA and GW but when attempting to do BSE I get the following error:

Code: Select all

 [03] Transferred momenta grid
 =============================

 [RD./SAVE//ndb.kindx]---------------------------------------
  Fragmentation                    :no
  Polarization last K   :  72
  QP states             :   1   72
  X grid is uniform     :yes
  BS scattering         :yes
  COLL scattering       :no
 - S/N 000495 -------------------------- v.04.02.01 r.00110 -

 [ERROR] STOP signal received while in :[03] Transferred momenta grid

 [ERROR] File ./SAVE//ndb.kindx; Variable Bindx; NetCDF: Numeric conversion not representable
Does anyone know what the problem might be? Apparently this is not a common problem and I could not find anything related to it in the forum.

I have compiled QE 6.4.1 and Yambo 4.3.2 on Intel 19, if you need any specifics regarding the compilation please let me know.

Thanks in advance and have a nice Sunday everyone!
Chris

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Mon Jul 15, 2019 11:55 am
by Daniele Varsano
Dear Christoph,
indeed it is the first time I see this kind of error. In order to spot the problem, we need to reproduce your error.
Can you please provide us you qe input/pseudo and yambo input file?

Many thanks,
Daniele

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Tue Jul 16, 2019 8:53 am
by chwolf
Dear Daniele,

I attached the input in the zip file - I now think this might be related to the memory issue in my other thread which I should investigate in more details first I guess...

the pseudopotentials used are from the ONCV library
inputs.zip
PP.zip

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Tue Jul 16, 2019 9:56 am
by Daniele Varsano
Dear Christoph,
I now think this might be related to the memory issue
did you spot some messages in the LOG files? If so, can you post it?

Thanks,
Daniele

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Tue Jul 16, 2019 11:57 am
by chwolf
Dear Daniele,

there is no warning in the logs but a single process of yambo keeps eating up to 190 GB of Ram (max RAM per node on the cluster) and then crashes. Sorry that I made another post as I initially thought those two are unrelated

viewtopic.php?f=13&t=1604

Best,
Chris

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Tue Jul 16, 2019 12:16 pm
by Daniele Varsano
Dear Christoph,
thanks for reporting, I'm not sure the probems are related.
Looking at the point where yambo stops:

Code: Select all

STOP signal received while in :[03] Transferred momenta grid
this part should be not memory intensive.
Actually I never tested before Intel 19 and I do not have it available in any machine.

Best,
Daniele

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Tue Jul 16, 2019 4:28 pm
by chwolf
Dear Daniele,

thanks for your swift reply - is there any compiler set you can recommend other than intel19?

Does Yambo usually play nice with gfortran?

Thanks for your continuous support!

Chris

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Wed Jul 17, 2019 4:21 am
by chwolf
Sorry, I guess that my mistake was to use Ry instead of RL for BSENGBlk :shock: :shock:

I must have copied the input over from the previous attempt without noticing. It seems to be working (for the very small values of BSENGBlk I tried) - I will try to run a more realistic calculation and let you know.

Best,
Chris

Re: NetCDF: Numeric conversion not representable when BSE

Posted: Wed Jul 17, 2019 9:18 am
by Daniele Varsano
Dear Christoph,
ok let us know if you encounter problems.
I think that different problems are mixing. Inputs in Ry for BSENGBlk should also work.
The netcdf problem you encountered at the beginning contrary to what I said you in my first reply was already spotted in the past and fixed in the devel version of the code, but it appeared for very large k point samplings and this is not your case.

About gfortran, this is also an option for Yambo.

best,
Daniele