I have already published this question in other forum. What is the best method to calculate point dynamic stiffness (mm/sec/Newton) in nastran for automotive application.
Modal frequency means that first the modes are computed and after that the total response. That means that to get accurate results you have to include "enough" modes in the calculation.
Direct means that the equations of motion are solved directly. That means that all modes are included in the calculation.
The "best" choice depends on the problem. I think you should study that a bit before making the choice but the results from both methods should be the same provided your solution parameters are correct.
As for boundary conditions, the requirement is the obvious. They should reflect the phisics/mechanics of the model. If you have double symmetri, sure use it.
If you look at an example on how to perform spektrum-to-spektrum analys it might be helpful.
Basically, say that you apply you load in dof T1, you should be able to combine that with constants in T2, T3, R1, R2 and R3. This will give you a singular solution for a static load due to one rigid body motion. But sol 111 should work as far as I know. Provided I understand you correct.
I would not "play" with the bailout card for other reasons than "error detection", the limitation are there for a reason.