A bit more details I found. I added debug FTriangleMesh pointer and some debug facilities to
FVelocityAndPressureField to make sure FTriangleMesh is still alive, because PointToTriangleMap is an array view to FTriangleMesh::MPointToTriangleMap. When a new crash happened, it was alive which is a good news.
The view contained 103 elements, and only [88-103) looked valid. At same time the mesh contained an array with 15 elements which are same as in view range [88-103). What actually wrong is either a view or the
Index. Because that index came from ParallelFor which parallelized valid ranges, first of which was [0-88). It came from TPBDActiveView<FSolverParticles>::RangeFor (variable MParticlesActiveView), and there are two valid ranges: [0-88) and [88-103).