Yambo Community Forum

Dea developers,
I ran the BSE calculations and stopped at the process of diagonalization,
...
[09.03] Diago Solver @q1
========================
[ERROR] STOP signal received while in[09.03] Diago Solver @q1

[ERROR] LINEAR ALGEBRA driver [SERIAL_HERMITIAN_diagonalization]performing CHEEV
...
I tried the version of v5.2.1 and v5.2, but both of them failed.
Thanks!

Shudong

Dear Shudong,

can you provide more information on your calculation? (e.g. including the report file).
Best,
Daniele

Dear Daniele,
Thanks for your reply. Attached is the report file.

Best,

Shudong

Dear Shudong,

here two suggestion to investigate the problem:

1) The excitonic matrix, you are dealing with, is large but not impossible to diagonalize using serial linear algebra. Can you try to repeat your calculation without the help of the scalapack support. Just set 1 to the assignment in BS_nCPU_LinAlg_DIAGO variable.

Anyway, I do not think that scalapack is the problem, as I can see a suspicious in your report file.

2) Can you read the ndb.BS* files by using ncdump to have it in human read format and check if you do not have a matrix with all zero entries? If this is the case, we will need to inspect what happened during the build-up of the kernel.

Best,

Daniele

Dear Daniele，
I have checked the ndb.BS_PAR_Q1, it seems the problem is here as:

netcdf ndb {
dimensions:
complex = 2 ;
BS_K_dim = 16384 ;
variables:
float BSE_RESONANT(BS_K_dim, BS_K_dim, complex) ;
data:

BSE_RESONANT =
NaNf, NaNf,
0, 0,
0, 0,
0, 0,
_, _,
_, _,
_, _,
_, _,
_, _,
...

Thanks!

Best,

Shudong

Dear Shudong,

can you try to understand if the problem is in direct term (explore if the NaN already appear in the screening database ndb.em1s/ndb.pp), or
in x term (perform a minimal test with x only). Also inspect dipoles.

Best,
Daniele