VeriSimpl: Robust Optimization Modeling from Natural Language using Simplification-based Verification
06:00 · July 24, 2026 · arXiv cs.AI RSS
Natural language interfaces can greatly benefit the accessibility and usability of optimization modeling, and recent advances in large language models (LLMs) show promise in automatically translating textual problem descriptions into executable solver formulations. However, a key challenge for existing approaches is to ensure that the inferred formulation correctly implements the intended task, even if it may execute without errors. We introduce VeriSimpl, a solver LLM framework for robust natural-language-to-optimization formalization. Our approach is based on the idea of simplification-based verification, where the optimization solver is leveraged to generate simplified diagnostic queries about a candidate formulation to allow the LLM to tractably reason about the correctness of the formulation with respect to the task description. We present such simplification strategies along different dimensions with respect to problem constraints and decision variables, which allow the LLM to reason locally under fixed global contexts. Evaluations on a range of optimization benchmarks show how our approach provides consistent improvements in accuracy over existing methods, while also providing a novel high-precision self-verification signal.
Summary
Natural language interfaces promise to lower the barrier to optimization modeling, yet large language models still struggle to produce formulations that are not only executable but semantically faithful to a given task description. VeriSimpl addresses this gap with a solver-LLM framework that interleaves code generation and verification. Rather than relying on the LLM to invent test cases, the approach uses the solver itself to create simplified diagnostic queries that probe the candidate formulation along selected dimensions of constraints and decision variables.
These simplifications preserve the global structure of the original problem while reducing its complexity, so that the LLM can reason locally about feasibility or optimality properties under fixed context. For each query the solver supplies a ground-truth outcome; the LLM then checks whether that outcome aligns with the original natural-language specification. Consistent agreement across multiple queries yields a high-precision self-verification signal that flags formulations the system can treat with elevated confidence.
Evaluations on four optimization benchmarks spanning different domains show that the method delivers consistent gains in end-to-end formulation accuracy relative to prior prompting, agentic, and fine-tuning baselines. At the same time, the verification mechanism identifies a substantial subset of outputs for which manual inspection can be safely reduced, offering a practical route toward more reliable natural-language interfaces for operations-research tasks.
Why it matters
This research is highly relevant for Dutch AI researchers and practitioners, particularly those working in operations research, logistics, and supply chain optimization. By improving the reliability of LLM-generated optimization models, it lowers the barrier to entry for SMEs and enterprises seeking to deploy complex decision-making algorithms.
