Help needed with SCF convergence

 Dear netters,
     I am searching the energy surface of a triatomic molecule
 (doublet) with ROHF method and got convergence problem whenever I try
 to bend the molecule.  The scf procedure goes okay until the total
 energy difference between two consecutive iteration steps is at the
 7th decimal place. Then the difference remains almost constant and
 the iteration goes on forever.
     I tried vshift option with g92 but it didn't work.  G92 manual
 says scf=qc does not apply to ROHF. The program I am using now is
 meldf.  With meldf, sometimes it works, sometimes it doesn't, even if
 the guess is from a previous run with a smaller angle.
     Can anyone give me advices?  Thanks!