EPW interpolation suddenly stops for high nkf/nqf

[fabiocosta4]

Dear EPW fellows

I'm facing a strange issue when calculating the electron-phonon interaction and the critical temperature for a simple test system. At first, I tried these calculations with a very low-settings coarse grid ( q and k = 2^3 ), and did the interpolation process for several small meshes, reaching the maximum of 20^3 K and 10^3 Q. This first test was successful, even though not very precise (as expected, since the coarse grid was very coarse indeed).

As a second step, I tried the same calculation, now with a coarse grid of 8 8 8 K and Q. My first interpolation was for a grid of 16^3 K and 8^3 Q, followed by interpolations for higher orders (32^3 K 16^3 Q and 48^3 K 24^3 Q). Only the first one finished successfully, and the other ones finish with no clear error message.
After a line about the size of the allocated memory, there is just a message informing that the mpiexec process is killed, but it is unclear to me what went wrong.

The output files and the input model are in the following link: https://github.com/fabiocosta4/epw-error.git .

To run the code, I'm using the command $MPI_DIR/mpiexec -np $np $QE_DIR/epw.x -npool $np -i epw.in > epw.out , with $np set to 6.

Also, I'm working with QE 6.4 and EPW 5.1.

In advance, many thanks for any assistance

Fbio Costa
MSc student in physics
Federal University of Bahia, Brazil

[liu xinbao]

Dear Fabio,

As you can see, the program stop after it estimate the memory demand.

Code: Select all

Size of allocated memory per pool : ~=    5.5034 Gb

so it is very likely that the memory demand exceed the RAM of your server.
I would suggest you to set 'etf_mem = 2 ' in your epw.in file which will reduce the memory usage.

Bests,
Xinbao

[fabiocosta4]

Dear Xinbao

Thank you for your reply

My main suspicion was that the memory was actually the main issue, just as you said. These past days I've done many attempts, with this variable etf_mem set to 1 and 2, along with several values for max_memlt, but no success. For instance, if I set max_memlt to 4gb (since the system has 16G for each 4 cores), the program still complains that the required memory exceeds max_memlt. If I set it to higher values, the program exceeds the maximum memory (naturally), and crashes. At this point it seems to me that the server is not powerful enough to run these higher order interpolations.

Any further suggestions will be appreciated

Thank you for your attention

Fbio Costa