Research Paradigm of Materials Science Tetrahedra with Artificial Intelligence

Abstract

The classical material tetrahedron that represents the Structure-Property-Processing-Performance-Characterization relationship is the most important research paradigm in materials science so far. It has served as a protocol to guide experiments, modeling, and theory to uncover hidden relationships between various aspects of a certain material. This substantially facilitates knowledge accumulation and material discovery with desired functionalities to realize versatile applications. In recent years, with the advent of artificial intelligence (AI) techniques, the attention of AI towards scientific research is soaring. The trials of implementing AI in various disciplines are endless, with great potential to revolutionize the research diagram. Despite the success in natural language processing and computer vision, how to effectively integrate AI with natural science is still a grand challenge, bearing in mind their fundamental differences. Inspired by these observations and limitations, we delve into the current research paradigm dictated by the classical material tetrahedron and propose two new paradigms to stimulate data-driven and AI-augmented research. One tetrahedron focuses on AI for materials science by considering the Matter-Data-Model-Potential-Agent diagram. The other demonstrates AI research by discussing Data-Architecture-Encoding-Optimization-Inference relationships. The crucial ingredients of these frameworks and their connections are discussed, which will likely motivate both scientific thinking refinement and technology advancement. Despite the widespread enthusiasm for chasing AI for science, we must analyze issues rationally to come up with well-defined, resolvable scientific problems in order to better master the power of AI.

Figures (6)

• Figure 1: Overview of materials science research and AI integration. a, The periodic table lays the foundation of various materials. The lower panel shows examples of materials of different types for illustration. b, The number of publications, $\mathcal{N}_{\rm pub}$, in a variety of materials science research fields. c, Yearly $\mathcal{N}_{\rm pub}$ for different subjects with additional search keywords "machine learning or artificial intelligence", demonstrating the research activities of integrating AI with these subjects.
• Figure 2: The classical material tetrahedron. The historical overview of the research diagram of the structure-property-processing-performance-characterization relationship. There are roughly four stages with the demonstrated key components in summary.
• Figure 3: Schematic learning landscape of AI for materials science. The known materials and the associated knowledge (World Knowledge) are conceptually shown within a single local minimum for simplicity. It is only an iceberg in the potential material space. Statistical learning usually figures out the hidden pattern locally, bringing limited power to expand the knowledge boundary. Target knowledge is separated into another local minimum to show the desire of grand breakthrough. AI is promising to guide navigation on the landscape with two research targets: domain exploration and pathway optimization.
• Figure 4: AI-augmented material research tetrahedron. With Matter centered, the tetrahedral network has four components, i.e., Data, Model, Potential, and Agent. These vertices and their combinations are all significant for AI-augmented materials science. Some extended components are included for illustration.
• Figure 5: Tetrahedron for AI research. a, There are four vertices centered at Data to make up the tetrahedron, i.e., Architecture, Encoding, Inference, and Optimization. b, Schematic relationship between model complexity and the sufficient size of the training dataset. Three regimes are shown: statistical learning, deep learning, and LLMs. c, An illustration depicting the capability of a model in processing input information. With increasing model complexity, it can process information in a wider range of complexity, with the upper bound extended. Advanced LLMs like GPT-5/Gemini can process both simple and difficult requests, while statistical learning like SVM has limited capability with specific requirements.