QANS v4: Adaptive Multi-Branch Optimization Outperforming Simulated Annealing Under Fixed-Time Constraints

This repository contains the full experimental evidence and validation results for QANS v4, an adaptive multi-branch optimization framework designed to outperform classical simulated annealing under fixed computational time constraints. We benchmark QANS v4 against simulated annealing on weighted Max-Cut instances (n = 200, p = 0.03), using equal time budgets across all experiments. Key findings: - QANS v4 consistently outperforms simulated annealing across all tested time regimes.- Statistical validation shows highly significant improvements (p < 0.001).- Effect sizes are large (Cohen’s d = 1.6–2.2).- Win rates range from 95% to 100% across randomized graph instances.- Performance gains are strongest in low-time regimes, indicating superior early-stage search efficiency. The algorithm combines:- multi-branch exploration- guided block updates- deterministic local refinement- adaptive mutation under stagnation Together, these produce a structured search dynamic that achieves faster convergence and higher-quality solutions than classical stochastic annealing. This dataset includes:- raw benchmark outputs- robustness evaluations across multiple seeds- statistical validation results- publication-quality figures This work provides reproducible evidence of a consistent algorithmic improvement over a widely used optimization baseline.