When running a calculation on a large system (864 bands) using local exact exchange (useexexch 1), I get consistent SCF convergence failures, as the total energy diverges to ca. -1e5 (see attached). I am able to run the test (t18.abi) cleanly, as well as a calculation on the large system with ecut 2, so the problem seems to be dependent on the number of plane waves.
I have obtained the same results using intel compilers, MKL, and DFTI, as well as GNU compilers with OpenBLAS and Goedecker FFT, so the problem does not seem to be compiler or library dependent.
Is this problem known and if so is there any way around it?
Thanks,
Carl
taoi-c.abi (12.9 KB)
taoi-c_gnu.abo (778.7 KB)
taoi-c_gnu.abi (12.9 KB)
log.log (176.0 KB)
taoi-c.abo (71.9 KB)