Search papers, labs, and topics across Lattice.
The paper introduces the Infinite Problem Generator (IPG), an agentic framework that synthesizes verifiable physics problems by constructing solutions as executable Python programs, ensuring mathematical consistency. IPG expands a small set of expert-defined seed problems into a diverse corpus of problems with guaranteed solvability, addressing the scarcity of high-quality reasoning data. The authors release ClassicalMechanicsV1, a dataset of 1,335 classical mechanics problems, along with the IPG pipeline, and demonstrate a strong correlation between formula count and verification code length, enabling controllable curriculum generation.
Tired of hallucinated physics problems? This agentic framework generates infinite, verifiable problems with guaranteed solvability by constructing solutions as executable Python programs.
Training large language models for complex reasoning is bottlenecked by the scarcity of verifiable, high-quality data. In domains like physics, standard text augmentation often introduces hallucinations, while static benchmarks lack the reasoning traces required for fine-tuning. We introduce the Infinite Problem Generator (IPG), an agentic framework that synthesizes physics problems with guaranteed solvability through a Formula-as-Code paradigm. Unlike probabilistic text generation, IPG constructs solutions as executable Python programs, enforcing strict mathematical consistency. As a proof-of-concept, we release ClassicalMechanicsV1, a high-fidelity corpus of 1,335 classical mechanics problems expanded from 165 expert seeds. The corpus demonstrates high structural diversity, spanning 102 unique physical formulas with an average complexity of 3.05 formulas per problem. Furthermore, we identify a Complexity Blueprint, demonstrating a strong linear correlation ($R^2 \approx 0.95$) between formula count and verification code length. This relationship establishes code complexity as a precise, proxy-free metric for problem difficulty, enabling controllable curriculum generation. We release the full IPG pipeline, the ClassicalMechanicsV1 dataset, and our evaluation report to support reproducible research in reasoning-intensive domains.
