I have finished the GW calculation. The log and report are attached in this email.
But I meet the NaN problem again on my HPC.
I find where NaN from.
It is in the ns.kb_pp_pwscf_fragment_1 like this.
I have renamed ns.kb_pp_pwscf_fragment* (do not use them).
From TianxiangQian: