I've been doing some light afternoon googling on this and it would seem I'm in need of better search terms. I want to assess the cost and feasibility of simulating fracturing and approximate destructibility of pixel terrains, including handling actual physics, like cave-ins, different terrain consistencies at different points and such. But I'm having some trouble pinning down how to go about it.
The best idea I can come up with is creating a fairly low resolution representation of the terrain in the form of inter-connected breakable joints laid out either in a regular grid or attached to key locations depending on material (kind of like a Voronoi diagram). Simulating a collapse then becomes a matter of breaking a number of joints and as part of the terrain is broken loose, dynamically generating a mesh and simulating it as usual using my favorite physics engine. This seems limited, however, as the number and shape of the broken off terrain pieces can quickly become overwhelming and in terms of performance it would be far from robust to stitch them back together once they come to a rest again. As such I'd much prefer a more fine-grained GPU-based solution, which I cannot presently envisage.
Ideas?
