The attached input is an (apparently) streaightforward instance of feeding a converged $VEC group at the starting geometry. However, the SCF convergence at the initial point (corresponding to the $VEC group read with guess=moread) is exceedingly slow: more than 100 SCF steps are needed, even with soscf=.t. switched on. Since I do not often find these problems, and the most salient difference between my "usual" cases and this instance is the use of ecp=read, I wonder whether there is some problem arising from the interplay of these groups.
This message contains the 242 kb attachment
[ input.tar.gz ]