Distributed Optimization over Simplicial Complexes via Topological Dirac Synchronization

Abstract

We introduce a distributed optimization framework for systems with higher-order interactions represented by simplicial and cell complexes. Specializing global topological Dirac synchronization to gradient flows of strongly convex local objectives, we show that the synchronized equilibrium coincides with the minimizer of the aggregate cost, without Lagrange multipliers, dual variables, or saddle-point dynamics. Orientation-equivariant convex terms and cell-dependent dimensional targets allow the optimal state to be nonzero, as determined by joint per-dimension optimality conditions. Our algorithm couples signals across dimensions through the Dirac operator: nodes interact with edges, edges with nodes and faces, and faces with edges, enabling information exchange beyond graph-Laplacian methods. In the isotropic case, a Master Stability Function analysis yields a necessary and sufficient condition for local exponential stability: the product of the directional coupling strengths must be below the squared ratio of the strong convexity parameter to the spectral radius of the complex. A weighted Lyapunov argument accounting for non-normal Dirac coupling establishes global exponential convergence at the same rate. Finally, we identify a topological admissibility condition ensuring invariance of the synchronization manifold and coherent convergence of cells to the optimum, with no counterpart in classical graph-based distributed optimization.