TB2J keeps crashing

[adityaputatunda]

Dear Devs,

I am trying to run tb2j with siesta for a big unit-cell of about 500 atoms on a 250 GB RAM, 64 core node in colinear mode. For a smaller cell of about 100 atoms, the calculation ran out of RAM in Green's function calculation stage which I could fix by using the --use_cache option. However, in the current cell, the calculation crashes earlier (in the exchange calculation stage) with only about 10% memory use reported by top with the error message:

Traceback (most recent call last):
File "/usr/local/bin/siesta2J.py", line 139, in
run_siesta2J()
File "/usr/local/bin/siesta2J.py", line 134, in run_siesta2J
orb_decomposition=args.orb_decomposition
File "/usr/local/lib64/python3.6/site-packages/TB2J/manager.py", line 295, in gen_exchange_siesta
description=description)
File "/usr/local/lib64/python3.6/site-packages/TB2J/exchange.py", line 72, in init
self.set_tbmodels(tbmodels)
File "/usr/local/lib64/python3.6/site-packages/TB2J/exchangeCL2.py", line 35, in set_tbmodels
nproc=self.np)
File "/usr/local/lib64/python3.6/site-packages/TB2J/green.py", line 92, in init
self._prepare_eigen()
File "/usr/local/lib64/python3.6/site-packages/TB2J/green.py", line 159, in _prepare_eigen
emax=self.efermi + 10.1)
File "/usr/local/lib64/python3.6/site-packages/TB2J/green.py", line 98, in _reduce_eigens
istart, iend = ts[0], ts[-1] + 1
IndexError: index 0 is out of bounds for axis 0 with size 0

I wish to know if the crash is happening due to the system size (IndexError) or lack of memory and if there is a fix to it. Any help would be appreciated.

Thank you,
Aditya

[mailhexu]

Hi Aditya,
Thanks for the feedback. I am aware of this issue: it is because the diagonalization of the Hamiltonian fails when the size is too large . So far I haven't found a good way for the diagonalization. One solution is to read the wave functions directly from the SIESTA output, which is indeed in my plan.
Best regards,
HeXu