I perform a geometry optimization with some DFT in a basis set A with Gaussian. After several steps of a geometry optimization procedure the next step fails because the self-consistent procedure for DFT/A in the current geometry does not converge. I tried various extra tricks from Gaussian manual, which usually improve the convergence rate, but in this case they all fail. For a single point calculation there is a possibility to perform a preliminary calculation in another basis (B), save orbitals of the DFT/B model to a checkpoint and use this checkpoint to start the DFT/A model with DFT/B orbitals. However, this procedure works for a single point only.

My question is whether it is possible to tell Gaussian to perform a preliminary calculation with a basis B each time it starts a new SCF, so that this method of dealing with a difficult convergence case can be used for geometry optimizations, too?