NaN appeared in [06] BSE solver(s)

[emerarud]

Hi all,

I am using Yambo GPL Version 4.0.4 Revision 107 build with MPI+OpenMP.

When I was calcuating a SiO2 quartz, NaNs appeared in [06] BSE solver(s).

<31s> P0008: [06] BSE solver(s)
<34s> P0008: [06.01] Haydock solver
<37s> P0008: [Haydock] Iteration 1
<39s> P0008: [Haydock] Iteration 2
<41s> P0008: [Haydock] Iteration 3 Accuracy : NaN| -0.02000
<42s> P0008: [Haydock] Iteration 4 Accuracy : -NaN| -0.02000
<43s> P0008: [Haydock] Iteration 5 Accuracy : NaN| -0.02000
<44s> P0008: [Haydock] Iteration 6 Accuracy : -NaN| -0.02000
<45s> P0008: [Haydock] Iteration 7 Accuracy : NaN| -0.02000
<46s> P0008: [Haydock] Iteration 8 Accuracy : -NaN| -0.02000
<47s> P0008: [Haydock] Iteration 9 Accuracy : NaN| -0.02000
<48s> P0008: [Haydock] Iteration 10 Accuracy : -NaN| -0.02000

All the input files are attached.

At first I chose (d)iagonalization as a BSE solver,
but the calculation was interruppted after
[05.04] Bethe-Salpeter block due to a memory problem.

Then I restarted the calculation with (h)aydock method and
the above error appeared.

It would be helpful if you had any advice.

Best regards.

Kousuke Nakano

[Daniele Varsano]

Dear Kosuke,
please note files are not attached. Please post input/reports and standard outputs.
The forum complains when fails have not specific suffixes, but you can tar/zip as this is allowed.

Best,
Daniele

[emerarud]

Dear Daniele

I am sorry.

I failed to attach the files.

Best regards.

Kousuke.

[Daniele Varsano]

Dear Kosuke,
at first glance I can't see anything wrong.
Can you post also the qe input files, and eventually pseudos if they are not in the qe website, as we can reproduce the error?

Best,
Daniele

[emerarud]

Dear Daniele

I am sorry for my late reply.

I attached the qe_input, the pseudo potentials and the list of SAVE directory.

Best regards.

Kousuke

[Davide Sangalli]

Dear Kosuke,
I had a look to your input files.

Unfortunately the run is quite big and "computer demanding".
Whcih means for us it is not easy to reproduce.

I can propose the following:
1) To have an idea where the problem comes from, you could inspect the ndb file produced by yambo.
ndb.* are databases produced with netcdf library. Thus you can visualize their content with the command "ncump $filename"
Can you try: "ncdump $filename | grep NaN > $filename.dat "
to see whether any of the databases generated by yambo has some NaN inside ?
In case, which one ?

2) To spot the source of the problem the best thing would be to have it in a small run.
Can you try to reduce the parameters to minimal values and see if you still get NaN ?
If not can you identify at which point you would get the NaN ?

Best,
D.

[emerarud]

Dear Davide,

1) NaN appeared in [06] BSE solver(s)
-----------------------------------------------------------------------------------------------------

1) To have an idea where the problem comes from, you could inspect the ndb file produced by yambo.
ndb.* are databases produced with netcdf library. Thus you can visualize their content with the command "ncump $filename"
Can you try: "ncdump $filename | grep NaN > $filename.dat "
to see whether any of the databases generated by yambo has some NaN inside ?
In case, which one ?

2) To spot the source of the problem the best thing would be to have it in a small run.
Can you try to reduce the parameters to minimal values and see if you still get NaN ?
If not can you identify at which point you would get the NaN ?

Thank you for your advise, I solved the error!!

NaN appeared in ndb.BS_Q1_CPU_XXX.

Finally, I found the reason of the error is OpenMP failure.
NaN did not appear when I used flat-mpi.

I am cheking the compiler options again.

-----------------------------------------------------------------------------------------------------
2) BSE solver stops after 117 iterations

Although the BSK problems were solved with flatmpi, I encountered a strange error in BSE solver.

The BSE solver suddenly stopped after 117 iterations and the problem was not solved even when I restarted the calculation.

I did not find any error messages.

Do you have any suggestions?

I attached the input and output files.

Best regards

Kousuke.

[emerarud]

Dear Davide,

I tried other options such as

BSHayTrs= 0.02000 # [BSS] [o/o] Haydock treshold. Strict(>0)/Average(<0)
BSHayTer # [BSS] Terminate Haydock continuos fraction

but I have not solved the problems yet...

I used 288 nodes (2304 cores) for the BSK (and BSS) calculations.

Any information would be helpful.

Best regards.

Kousuke Nakano.