Integrating the Theory of Inventive Problem Solving with Large Language Models: Enhancing Reasoning for Innovation in Materials Science at the Molecular Scale

The Theory of Inventive Problem Solving (TRIZ) has long been a cornerstone for systematic innovation in engineering domains, including chemical and materials science. This paper proposes a novel framework that integrates TRIZ principles with large language models (LLMs) to emulate researcher-like reasoning in atomistic materials science. By structuring LLM prompts around TRIZ tools—such as patterns of evolution, contradiction matrices, and inventive principles—we enable models to identify problems, frame contradictions, and generate inventive solutions for challenges like data scarcity, poor interpretability, and unphysical predictions in quantum-chemical simulations and machine learning (ML) models. Drawing on recent artificial intelligence-TRIZ hybrids, like AutoTRIZ and TRIZ-GPT (generative pre-trained transformer), we demonstrate applications in molecular design, such as resolving contradictions in shape-memory polymers. This approach not only amplifies current trends in physics-informed ML and generative design but also democratizes advanced problem-solving, accelerating discoveries toward ideality.