I suspect you are right about the number of the domain divisions. When a 3D domain has a large hole in the centre, such as yours, unused nodes are generated in the empty space inside the hole, so you need a high node density so that enough nodes are assigned to the domain in order to find a feasible solution. You can sidestep this using user-defined nodes, as shown in LPSolver_UserNodes.gh in the example files. For auto-generated nodes, increasing the node density is generally the way we’d suggest to try to fix this – but I’m having difficulty getting this to work for your problem.
I will look into this and update you.
As I mentioned above, we would generally recommend solving problems for a domain such as yours using user-defined nodes. However, after looking into your problem, we’ve identified a potential bug in this feature, which we are currently investigating. In the mean time, increasing the nodal density may address the problem, however I appreciate this means long solving times.