catopt-category-theoretic-compositional-/README.md

catopt-category-theoretic-compositional-

Problem space: distributed, offline-first optimization across heterogeneous edge devices (DERs, meters, controllers, EV chargers) in mesh networks. Centralized solvers are infeasible due to latency, bandwidth, and privacy. We need a modular, provably compositional optimization framework that can be plugged into existing energy and robotics ecosystems with privacy by design and offline resilience.

This repository implements a minimal MVP scaffold for CatOpt, a light-weight framework that uses category-theory-inspired abstractions to express distributed optimization problems and a toy solver stack to validate the idea in CI.

MVP Runtime (new): lightweight primitives to experiment with CatOpt concepts locally.

• LocalProblem: representation of an agent's optimization task.
• DataContract, SharedVariables, PlanDelta: lightweight data-exchange primitives for a privacy-conscious mesh.
• ADMMTwoAgentSolver: tiny, in-process ADMM-like solver to demonstrate joint optimization across two agents.
• demo_two_agent_admm(): quick in-repo demonstration function returning final variables.

Usage notes:

• Import from the package and optionally run the demo function to observe convergence on a toy problem.
• Example:

  from catopt_category_theoretic_compositional.runtime import ADMMTwoAgentSolver, demo_two_agent_admm
  result = demo_two_agent_admm()
  print(result)