Formalize, Don't Optimize: The Heuristic Trap in LLM-Generated Combinatorial Solvers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Haoyu, Song, Yuliang, Li, Tao, Deng, Zhiwei, Wang, Yaqing, Ramachandran, Deepak, Cohen, Eldan, Roth, Dan
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916005528207360
author Wang, Haoyu
Song, Yuliang
Li, Tao
Deng, Zhiwei
Wang, Yaqing
Ramachandran, Deepak
Cohen, Eldan
Roth, Dan
author_facet Wang, Haoyu
Song, Yuliang
Li, Tao
Deng, Zhiwei
Wang, Yaqing
Ramachandran, Deepak
Cohen, Eldan
Roth, Dan
contents Large Language Models (LLMs) struggle to solve complex combinatorial problems through direct reasoning, so recent neuro-symbolic systems increasingly use them to synthesize executable solvers. A central design question is how the LLM should represent the solver, and whether it should also attempt to optimize search. We introduce CP-SynC-XL, a benchmark of 100 combinatorial problems (4,577 instances), and evaluate three solver-construction paradigms: native algorithmic search (Python), constraint modeling through a Python solver API (Python + OR-Tools), and declarative constraint modeling (MiniZinc + OR-Tools). We find a consistent representational divergence: Python + OR-Tools attains the highest correctness across LLMs, while MiniZinc + OR-Tools has lower absolute coverage despite using the same OR-Tools back-end. Native Python is the most likely to return a schema-valid solution that fails verification, whereas solver-backed paths preserve higher conditional fidelity. On the heuristic axis, prompting for search optimization yields only small median speed-ups (1.03-1.12x) and a strongly bimodal effect: many instances slow down, and correctness drops sharply on a long tail of problems. A paired code-level audit traces these regressions to a recurring heuristic trap. Under an efficiency-oriented prompt, the LLM may replace complete search with local approximations (Python), inject unverified bounds (Python + OR-Tools), or add redundant declarative machinery that overwhelms or over-constrains the model (MiniZinc + OR-Tools). These findings support a conservative design principle for LLM-generated combinatorial solvers: use the LLM primarily to formalize variables, constraints, and objectives for verified solvers, and separately check any LLM-authored search optimization before use.
format Preprint
id arxiv_https___arxiv_org_abs_2605_12421
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Formalize, Don't Optimize: The Heuristic Trap in LLM-Generated Combinatorial Solvers
Wang, Haoyu
Song, Yuliang
Li, Tao
Deng, Zhiwei
Wang, Yaqing
Ramachandran, Deepak
Cohen, Eldan
Roth, Dan
Artificial Intelligence
Large Language Models (LLMs) struggle to solve complex combinatorial problems through direct reasoning, so recent neuro-symbolic systems increasingly use them to synthesize executable solvers. A central design question is how the LLM should represent the solver, and whether it should also attempt to optimize search. We introduce CP-SynC-XL, a benchmark of 100 combinatorial problems (4,577 instances), and evaluate three solver-construction paradigms: native algorithmic search (Python), constraint modeling through a Python solver API (Python + OR-Tools), and declarative constraint modeling (MiniZinc + OR-Tools). We find a consistent representational divergence: Python + OR-Tools attains the highest correctness across LLMs, while MiniZinc + OR-Tools has lower absolute coverage despite using the same OR-Tools back-end. Native Python is the most likely to return a schema-valid solution that fails verification, whereas solver-backed paths preserve higher conditional fidelity. On the heuristic axis, prompting for search optimization yields only small median speed-ups (1.03-1.12x) and a strongly bimodal effect: many instances slow down, and correctness drops sharply on a long tail of problems. A paired code-level audit traces these regressions to a recurring heuristic trap. Under an efficiency-oriented prompt, the LLM may replace complete search with local approximations (Python), inject unverified bounds (Python + OR-Tools), or add redundant declarative machinery that overwhelms or over-constrains the model (MiniZinc + OR-Tools). These findings support a conservative design principle for LLM-generated combinatorial solvers: use the LLM primarily to formalize variables, constraints, and objectives for verified solvers, and separately check any LLM-authored search optimization before use.
title Formalize, Don't Optimize: The Heuristic Trap in LLM-Generated Combinatorial Solvers
topic Artificial Intelligence
url https://arxiv.org/abs/2605.12421