GAMESS: SCF convergence

[FyD <fyd { *at * } u-picardie.fr>]

Dear All,
 I want to optimize a model to the B3LYP/6-31++G** theory level (TL) using
 GAMESS. If I optimize it directly to this TL the first SCF computation does not
 converge (with GUESS=HUCKEL). So, to be able to get a B3LYP/6-31++G** optimized
 structure, I proceed in two times. I first optimize my model to the 6-31++G** TL
 (with GUESS=HUCKEL) and then from the new 6-31++G** optimized Cartesian
 coordinates I reoptimized it to the B3LYP/6-31++G** TL (with GUESS=MOREAD)
 providing the 6-31++G** $VEC group. I have the same probleme with a single point
 energy determination using the same B3LYP/6-31++G** basis set if I do not
 provide a previous $VEC group.
 - However, I am interested in getting B3LYP/6-31++G** energy value using a
 SINGLE input and not TWO as I am doing. Is there a way to do this ?
 - I do not have this SCF convergence problem if I use MP2 computations for
 instance. Is it a known probleme for B3LYP/DFT calculations ?
 Thanks, regards, Francois