Search papers, labs, and topics across Lattice.
This paper addresses optimization problems where the feasible region is the intersection of two manifolds, a common but geometrically complex scenario. The authors prove the equivalence of clean intersection and intrinsic transversality, enabling a tractable projection onto the tangent space of the intersection. They then introduce a geometric optimization method that retracts onto one manifold while updating iterates along two orthogonal directions to approach the other manifold and decrease the objective function, achieving convergence under intrinsic transversality.
Forget about coupled geometry: this new method makes optimization over manifold intersections tractable by cleverly navigating tangent spaces.
Optimization over the intersection of two manifolds arises in a broad range of applications, but is hindered by the coupled geometry of the feasible region. In this paper, we prove that the regularities -- clean intersection and intrinsic transversality -- are equivalent, which yields a tractable projection onto the tangent space of the intersection. Therefore, we propose a geometric method that employs a retraction on only one manifold and updates the iterate along two orthogonal directions. Specifically, the iterates stay on one manifold, and the two directions are responsible for asymptotically approaching the other manifold and decreasing the objective function, respectively. Under intrinsic transversality, we derive the convergence rate for both the feasibility and optimality measures, and show that every accumulation point is first-order stationary. Numerical experiments on problems stemming from sparse and low-rank optimization, including fitting spherical data, approximating hyperbolic embeddings on real data, and computing compressed modes, demonstrate the effectiveness of the proposed method.
