If you ran a RKS calculation on a closed shell (CS) molecule, you'll typically easily converge to the right answer. If you run an UKS calculation on a CS molecule, it will naturally converge to the same CS solution found by RKS (although at up to 2x the cost, as the number of spatial orbtials is doubled).

If try to run a RKS calculation on an open shell (OS) system e.g. a triplet excited state of water, ORCA will print this warning:

WARNING: your system is open-shell and RHF/RKS was chosen

===> : WILL SWITCH to UHF/UKS

and then the calculation will proceed as normal, just as it would have done if you'd have used the UKS keyword, or simply omitted both RKS and UKS .

I want to add to Isaac F. Leach 's helpful answer that you want to avoid optimizing closed-shell systemts in unrestricted DFT next to the unnecessary cost also because you can introduce spin contamination (see e.g. https://en.wikipedia.org/wiki/Spin_contamination). The restricted solution is typically cheaper and better.

Thank you Isaac F. Leach and Dr Florian Mulks for helping me gain better understanding of the precautionary measures in the practice of DFT methods with regards to UKS/RKS treatment.