None of those will work. You cannot use a ground state calculated with combination #1 of pseudopotentials+basis-sets (e.g. FHI-DZP or SG15-M) as initial state for a calculation with combination #2 of pesudopotential+basis-sets (e.g. SG15-H). The numerical representation of the electronic structure problem is different between those combinations, so they are not compatible with each other as initial guess for the ground state. Technically, you can do it, but ATK will complaint that the initial state is not usable and will therefore just not use it. The calculation will therefore start from scratch.
However, for calculations such as geometry optimization, what you can use from one combination to another is the final relaxed configuration: If you want to run geometry optimization with SG15-H, which is a high-quality calculation and therefore also slower than many others, it is sometimes useful to first run the optimization with a smaller basis set, e.g. FHI-DZP, and then start the SG15-H relaxation from that. Might save some BFGS steps in the SG15-H calculations, especially if the starting configuration is far from equilibrium.
