GW calculation stopped

[ezekiel]

Hi,
My GW calculations stopped. Pls, help me out.
My log file is attached.
Thank you

[Daniele Varsano]

Dear Ezekiel,
it can be a memory issue. In order to improve the memory distribution, try to put all the CPUs on the Gbands by setting in input:

Code: Select all

SE_CPU= " 1 1 64"       # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b" 

Best,

Daniele

[ezekiel]

Thank you Daniele. I tried your suggestion but the problem persisted.

[Daniele Varsano]

Dear Ezekiel,

do you have any error message in the report file or the output of your job submission that can help to spot the problem?
Can you also post your input and report file?

Best,
Daniele

[ezekiel]

Dear Daniele,
I have attached my input and report files.
Thank you

[Daniele Varsano]

Dear Ezekiel,
are you sure that input and report comes from the same run? There are some discrepancies!
You have in input
SE_CPU= "1 1 64"
but then in the report is reported:
* CPU-Threads :SE(environment)-2 32 1(CPUs)-q qp b(ROLEs)
the latter parallelization is highly memory consuming.

Here some suggestion:


*note that %QPkrange is not closed by the symbol %, this is because the file is cut? Please check as this can give problem.
* You define GTermKind, but the variable for the fictitious pole is missing: you should define something like:
GTermEn=40 eV
Probably it comes by defualt, but I'm not totally sure.
* I discourage the use of XTermKind= "BG" , it is memory consuming and does not help much, also it is not alwauys stable.

Anyway, yambo stops when trying to calculate the exchange self-energy so the problem is not related with the issues above.
If it is a memory problem, please be sure to set the right parallelization strategy: SE_CPU= "1 1 64"

if the problem persists, you can split your calculation in smaller pieces, e.g.
first run:
%QPkrange # [GW] QP generalized Kpoint/Band indices
1|100|45|48|
#

Second run considering k points 101-200, third runs 201-288.
When doing this use the -J option to save the database in different directory, or rename the ndb.QP file after each run ndb.QP_1, ndb.QP_2 to avoid yambo overwrite them.

Finally, you can merge the databases using ypp or yambo-py.

Best,
Daniele

[ezekiel]

Hi,
I am running a GW calculation. The run stopped after sometimes. Could anyone assist in resolving the issue? I have attached the report and LOG file.

Thank you
Ezekiel

[Daniele Varsano]

Dear Ezekiel,

I'm afraid this is a memory issue. The code stops before starting the calculation of the IP response function Xo which is a memory intensive step.

To have an idea of how much memory is needed, the code should be compiled using the flag --enable-memory-profile in the configure procedure.

Anyway, you can try to reduce the number of CPUs per node to have more memory per task and see if this solves the memory problem. Of course, the calculation will be slower.

Alternatively, you can check if the code runs reducing the number of bands in the response function calculation.

Best,

Daniele