Convergence Guarantees

The theorem guarantees convergence of primal residuals, dual residuals, and objective values. The convergence theorem requires convexity, a saddle point, solvable subproblems, bounded parameters, and summable parameter changes. The convergence argument controls learned adaptation by constraining parameter sequences instead of relying on training itself. The summable-change condition is experimentally enforced for Gamma_k by stopping updates after 500 OSQP iterations. The theorem does not guarantee convergence to a unique primal-dual solution.