[Schrödinger]

I notice that it is accepted practice to perform geometry optimizations and frequency calculations at one level of theory in DFT (using a particular functional/basis set combo), and to then use this optimized structure to perform single point energy calcs and TD-DFT calculations at a different level of theory.

Is this practice technically correct, or is it a big approximation that has just become the norm? Because the way I see it, every level of theory produces its own Potential Energy Surface (PES). Maybe I just don't understand enough, but once you change anything, like functional, or basis set, or solvent model, you will be generating a totally new PES, right?

[Corribus]

I think it's a necessary compromise for the sake of computing power.

[pm133]

You don't use a different level of theory as such. The functional stays the same as that used for the geometry optimisation but you improve the basis set and potentially add dispersion corrections. If you ran geometry optimisation on the larger basis set you would use significantly more computational power but there would be no noticeable change in the structure of the molecule.
What matters more than minute changes in geometry is getting the best energy result before you start calculating properties and this CAN be significant.