Answers are grounded in this paper and cite their sources.
Formulating optimization models from natural-language specifications is a major bottleneck in engineering decision-making, because accurate translation requires both domain expertise and strict modeling discipline. Existing large language model (LLM) approaches show strong code-generation ability but often produce formulations that are executable yet semantically incorrect, or structurally invalid under solver constraints. Evolutionary Chain-of-Thought Optimization (EvoCoT) is introduced as a co-evolutionary autoformulation framework that treats model synthesis as guided discrete search over reasoning--code pairs rather than one-shot generation. EvoCoT combines three tightly coupled mechanisms: chain-of-thought evolutionary refinement for semantic exploration, memory-guided exemplar retrieval for reusing successful modeling patterns, and a co-evolving verifier that continuously acquires formulation knowledge during search and delivers dense structural-faithfulness feedback unavailable to static verifiers. Verifier feedback and retrieved successes directly regulate selection and regeneration, forming a feedback-regulated closed loop that progressively improves formulation faithfulness and solver robustness. Experiments on eight benchmark suites, including NL4OPT, MAMO, BWOR, IndustryOR, OptMATH, and ComplexOR, show consistent gains over strong zero-shot, agentic, and fine-tuned baselines, with the largest improvements on difficult industrial-scale tasks. The results demonstrate that co-evolving generation and verification is a practical pathway toward reliable, scalable, and solver-executable optimization autoformulation.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: