Hello,
It seems that energy values larger than 1 000 000 a.u. won't be printed in the "gradient" file, resulting in following output:
cycle = 1 SCF energy =****************** |dE/dxyz| = 0.060736
Because the energies cannot be found from "gradient", relax and statpt can get a bit confused. Both consider all energy changes between geometries to be 0.0000 -> energy changes will be neglected when optimization convergence is checked -> optimization can end prematurely. So far I have circumvented this by tightening geometry change criterion to with "jobex -gcart" and checking the last energy change manually from "energy file" after convergence. So this is not such a huge problem, but it can cause some headache. Can the problem be fixed easily or would it cause compatibility issues in some modules or scripts?
This problem mostly affects structures consisting of late 3rd row elements (Ga->Br). For example, a structure with about 450 arsenic atoms already has an energy of over one million hartrees at B3LYP/def-SVP level. And the system size is still reasonable for optimization, one geometry step takes less than one day on one CPU without even using RI (B3LYP/def-SVP, ~15000 basis functions, O-symmetry).