Second-Order KKT Guarantees for Bregman ADMM in Nonconvex and Non-Lipschitz Optimization
It provides second-order optimality guarantees for a broad class of nonconvex optimization problems where standard Lipschitz gradients fail, benefiting practitioners in matrix/tensor factorization and distributed optimization.
The paper proves that Bregman ADMM for nonconvex linearly constrained problems with two-sided relative smoothness converges to strict-saddle KKT points with probability zero from random initialization, ensuring almost-sure second-order stationarity. Numerical experiments on distributed matrix factorization and symmetric tensor factorization validate the theory.
We analyze Bregman ADMM for nonconvex linearly constrained problems under two-sided relative smoothness, a condition that replaces the standard Lipschitz gradient assumption with a Hessian comparison relative to a Bregman kernel. This setting covers polynomial objectives arising in matrix and tensor models for which a global Lipschitz-gradient constant need not exist. We show that on an invariant open state-space domain, one iteration of Bregman ADMM defines a smooth primal--dual fixed-point map whose strict-saddle KKT points are unstable fixed points; consequently, from random initialization the iterates converge to a strict saddle with probability zero. Combined with existing first-order convergence results, this yields almost-sure second-order stationarity of limiting KKT points. We extend the analysis to a multi-block star consensus formulation for distributed optimization. The technical novelty lies in a determinant reduction with a Bregman-specific symmetrization and scaling step in the two block spectral argument, together with a null space cancellation exploiting the star graph structure in the consensus case. Numerical experiments on distributed matrix factorization illustrate the theory, and a symmetric tensor factorization example demonstrates the broader Bregman proximal splitting idea beyond the separable consensus setting.