If I want to fracture a building and make it collapse using an RBD dynamics sim in to a FLIP container so the fractured pieces splash the water, can FLIP recognise packed geometry from an RBD bullet simulation, or does it have to be converted to unpacked (real) geometry for collisions with the fluid sim?

Any advice on the basic node setup or links to tutorials would be appreciated.

In one tutorial I see a Collision Source node was used to convert an animated fractured object in to a VDB volume so it can collide with a FLIP sim. Must the animated object be converted to real geometry before feeding in to theCollision Source node rather than using a packed RBD Object, i.e. unpack it first?

In a different tutorial an RBD Packed Object with a Rigid Body Solver is fed in to the FLIP sim.

It seems there are multiple ways.Are there any advantages/disadvantages to these two methods?

The simple method is merging rigid body solver and flip solver, and implement operation through a sop solver.

That confirms one method. Thanks! I like your gif too.

In one tutorial I see a Collision Source node was used to convert an animated fractured object in to a VDB volume so it can collide with a FLIP sim. Is there some reason to not just se the normal geometry? Why convert to VDB?

litote
Is there some reason to not just se the normal geometry? Why convert to VDB?

FLIP doesn't collide with surfaces (apart from particles that can be corrected using particle collision), for computing proper velocity pressure projection it needs volume representation of collider object
and it's generally faster to compute it using VDB (or even being able to cache it), and just sample in Static Object or sample directly into Collision field rather than letting Static Object compute it

Thank you again Thomas! I appreciate your expertise. Would you always convert geometry to VDB to use as a collider in a FLIP sim? I see many basic tutorials just feed the polygon geo straight in to the sim without converting it to VDB. Presumably they are mot aware this is less efficient.

always, especially for deforming geo

there are ways to optimize though
if you are using just transforming object from object level it may not matter as much as the volume is built just once and not per timestep
I'd still use VDB volume though as it's more precise and also faster to compute, even if the speed benefit may not be as crucial if its built just once

there is also a possibility to use packed sets for colliders, which allows to reuse tansformed packed primitives as colliders efficiently, in a similar way of building the vdb just once per piece and just instance on points over time with v and w inherited from points

this is not built into FLIP however Pyro Solver SOP uses those so it's possible to rebuild for DOP level FLIP as they share the ay collision fields are built

for SOP level FLIPs the workflow is a bit different as there you manage your vdbs in SOPs and source into the global collision field before it even gets to solver

Thanks for the detailed explanation Thomas!