Out of memory issue while running diagonalization

[Priyanka]

Dear Experts

I'm experiencing difficulties with diagonalization due to an out of memory issue but not with haydock. I would like to examine excitons in more detail that's why I'm using diagonalization method here. Kindly find the attached files herein and help me to understand the problem and how to solve the same.

Thanks in advance
With warm regards
Priyanka
University of Modena and Reggio Emilia
Italy.

[Daniele Varsano]

Dear Priyanka,

it seems you have a very large BSE matrix and from the file you posted (report) the calculation stop when building the kernel and not when diagonalizing it. Here I can be wrong, as it is possible that the kernel building step is not reported in the report because of buffering and the log file should be inspected.

Some suggestions:
if the problem is the kernel building:
1. Try to reduce the number of bands included in the kernel hoping to not lose too much accuracy
2. If you have not yet done, you can recompile Yambo using the option --enable-memory-profile when configuring, this will provide in the log file an approximation of the memory needed and you can then tune the number of nodes needed for completing the calculation

if the problem is in the diagonalization, I think you need to resort to the slepc method. Also note that in the current version, I suspect that parallel algebra for diagonalization is not supported when using GPUs.

Best,
Daniele