Hi,
I'm doing GW/BSE with Turbomole 7.5 on some rather large structures. The G0W0 (rigw) calculations on these structures prove perfectly feasible, even if slow, but the corresponding Bethe-Salpeter calculations appear to be impossible because they require more than 400 GB of memory even with a smallish basis-set. Are there (algorithmic) options for reducing the memory cost of Bethe-Salpeter calculations other than (further) reducing the basis-set size?
Thanks,
Martijn