Exact Lagrangian Realization and Robust Strain Sensing in Incompressible Flow

What the math gives to ML

The paper provides a concrete algebraic design principle for sensing symmetric strain tensors through finitely many directional quadratic measurements that remain identifiable after arbitrary volume-preserving deformation. Its key transferable asset is the congruence action P_i to F P_i F^T: because F is invertible, a spanning family of rank-one projectors remains spanning. In three dimensions, six unoriented axes associated with a regular icosahedron provide a structured sensing frame. This can become a geometric strain-sensing layer for fluid neural operators, deformation-aware graph networks, or learned simulators, with explicit reconstruction and conditioning diagnostics.