CMLGS
Simplified coupled-map-lattice gradient smoothing for non-convex periodic optimization at O(D) per step.
Mechanism
A computer-implemented zeroth-order optimization method for non-convex objective functions with regular spatial structure or periodic topology. At each step the method draws perturbation directions from a fixed-magnitude distribution, applies a one-dimensional coordinate-aligned coupled-map-lattice discrete-diffusion chain to each direction, evaluates a symmetric finite-difference at the perturbed coordinates, aggregates into a coupled gradient estimator, and applies gradient descent. The method achieves O(D) per-step computational complexity in problem dimension.
The independent claim explicitly excludes chaotic modulation, error-gating, and geometric annealing of the perturbation magnitude (these appear only in dependent claims with workload-dependent operability honestly disclosed).
Prior-art differentiator
Evolutionary covariance-update methods are computationally infeasible at problem dimension D greater than approximately one thousand. On a high-dimensional Rastrigin scaling benchmark, the evolutionary baseline hits its wall-clock cap on every seed; CMLGS converges within seconds, establishing a feasibility crossover at the upper end of the dimension sweep.
Globally-coupled coupled-map-lattice swarm constructions in recent prior art rely on chaotic amplitude controllers for ergodicity injection. CMLGS coupling is local one-dimensional coordinate-aligned on perturbation directions, with chaos explicitly disabled. A pre-registered ablation establishes that disabling the chaos modulation produces a substantial residual reduction on the canonical benchmark, directly contradicting the central teaching of the closest swarm-CML prior art and supporting a Graham v. John Deere teaching-away rebuttal.
Workload-class scope demarcation
The independent claim is scoped to objectives exhibiting at least one of regular spatial structure in one or more coordinate directions, periodic topology, or physical-lattice geometric origin. The claim explicitly excludes the tiny-NN training workload (per the Phase 11 negative result on a small CIFAR-10 task), providing 35 USC Section 112(a) enablement inoculation.
Empirical validation
On multi-dimensional Rastrigin scaling benchmarks, CMLGS shows convergence advantages of one to several orders of magnitude over CMA-ES-family and AdamW baselines across D=100 to D=1000, with a pre-registered mechanism ablation confirming the coupled-map-lattice coupling is load-bearing (chaos modulation is mildly harmful). The antenna sidelobe transfer carries the result to a real-world engineering workload. The non-operability disclosure for Griewank and Rosenbrock parabolic landscapes is preserved.
Counsel posture
Provisional self-filed. Twelve-month conversion and PCT deadline 2027-05-07. Track One Prioritized Examination at non-provisional conversion recommended to capitalize on the favorable USPTO Section 101 environment. PCT geographic scope targeted to the European Patent Office, Japanese Patent Office, and Korean Intellectual Property Office to address the global phased-array and metasurface antenna markets driven by Asian and European telecommunications giants. Joint-inventor USPTO assignment recordation in progress under 37 CFR Section 3.11, coordinated with the MRRO companion.
Parent operator-T family. CMLGS is the multi-dimensional O(D)-per-step embodiment of the MRRO unified operator family, identified in MRRO Section 4.6.
Probe-routing meta-optimizer. HAMO routes optimization budget to a component pool that includes the CMLGS-derived graduated-smoothing method.