Thank you so much for the clarifications Prof. Tian Lu.

1 You can try to calculate a ground-state single point energy using both of them, and see if their results are the same. If not the same, use B3LYP/G keyword in ORCA and check.

2 libxc(B3LYP) is required because functional derivative of native B3LYP is not implemented. It can be used for both ground and excited state calculations, but its speed may be slightly slower than the native code, you can perform test to confirm.

3 Both of them are acceptable.

When I optimize the ground state (S₀) geometry in ORCA using the native B3LYP functional, the calculation runs without any problem.

However, when I attempt to optimize the first excited state (S₁), ORCA stops with an error and requests to use the LibXC implementation of B3LYP instead of the native one.

I would like to clarify the following points:

Are native B3LYP and LibXC(B3LYP) formally equivalent in ORCA, or are there meaningful differences in practice?

If LibXC(B3LYP) is required for excited-state optimizations (and for ESD calculations), is it recommended to also use LibXC(B3LYP) for the ground-state (S₀) optimization and frequency calculations to maintain methodological consistency?

Is it acceptable (from ORCA’s perspective and for publication-quality work) to optimize S₀ with native B3LYP and S₁ with LibXC(B3LYP), or should the same functional backend be used consistently for both states?

Any guidance on best practices for handling this situation would be greatly appreciated.