Speakers
Wednesday, May 27 at 11am – Anton Oliynyk (CUNY Hunter College)
Thursday, May 28 at 3pm – Chris Bartel (University of Minnesota)
Thermodynamic checks on AI-accelerated materials discovery
Abstract: Quantum chemical calculations using density functional theory (DFT) are now commonplace in materials research. There are several publicly available materials databases (e.g., Materials Project) that house DFT calculation results for millions of inorganic crystals, providing an excellent starting point for finding new materials with interesting properties. High-throughput computational screening of materials using DFT is now being augmented and accelerated using machine learning (ML). In this talk, I’ll discuss some shortcomings in the “early days” of ML-driven materials discovery (circa 2020) before transitioning to state-of-the-art approaches. Modern approaches leverage universal machine learning interatomic potentials (“foundation” potentials) along with generative artificial intelligence (AI) to rapidly search materials space. From the standpoint of materials discovery, the first challenge is efficiently identifying new inorganic crystals that are thermodynamically stable. Once these materials are identified on the computer, the next step is to make them in the lab. Unsurprisingly, ML models are now emerging with the goal of predicting which hypothetical materials are likely to be synthesizable. While these developments are exciting, a pervasive challenge for any ML problem is to anticipate the performance of models outside the curated experiments used for initial training and evaluation. This talk will detail my group’s efforts to systematically assess the quality of generative models and synthesis prediction models through the lens of first-principles thermodynamics.
Expediting Tools for Materials Informatics with LLMs
Abstract: With the advent of large language models (LLMs) such as ChatGPT and Claude, the barrier to developing custom computational tools in materials science has been significantly reduced. In the context of solid-state chemistry, with our lab focusing on the discovery of intermetallic materials, we leverage explainable machine learning to develop crystal structure design principles and predict properties with experimental validation. To accelerate this workflow for an eventual closed-loop pipeline, we have integrated domain expertise with LLM-assisted coding agents using a human-in-the-loop framework, enabling rapid development while maintaining scientific rigor. Rather than relying on unconstrained “vibe coding,” we systematically delegate well-defined programming tasks to LLMs, followed by validation against physical constraints and experimental data. Herein we present a suite of tools developed through this approach, including automated parsing and visualization frameworks for electronic structure calculations (e.g., DOS, bandstructure, COHP), user-friendly GUI-based applications for X-ray diffraction and analysis, and interactive platforms for compositional and phase-space exploration integrating experimental datasets as well as for featurizers for ML uses with novice collaborators. Additionally, we demonstrate how LLM-assisted workflows enable rapid prototyping of high-throughput pipelines and simulation interfaces for novice users. These examples will highlight how LLMs can be effectively incorporated into a closed-loop materials discovery pipeline, enhancing productivity while preserving both interpretability and scientific reliability.
Wednesday, May 27 at 3pm – Jill Wenderott (Drexel University)
Guiding solid-state material synthesis with computational-experimental feedback loop
Abstract: The challenge of discovery and design of novel advanced functional solid-state materials has shifted from identifying new chemistries and phases to that of finding efficient reaction pathways for practical synthesis. With the rapid progress in high-throughput density functional theory (DFT) calculations and the correspondingly emerged databases, new material chemistries with exceptional functional properties are being identified at an unprecedented pace. Nevertheless, these computational findings have rarely been converted into practical material applications by missing key knowledge regarding material synthesizability. Ultimately, the knitting together of computational and experimental materials discovery and synthesis in a responsive feedback loop will enable efficient realization of advanced materials. Here, the development of a framework for rapid assessment of synthesizability of inorganic compounds through a combination of the CALculation of PHAse Diagram (CALPHAD) approach and experimental validation is presented. The success of this computational-experimental feedback loop is demonstrated with the mixed anion material tantalum oxynitride (β-TaON), which displays a narrow synthetic window highly dependent on temperature and reactive gas flow partial pressures. Progress extending this feedback loop framework to other mixed anion materials, including perovskite oxynitrides, is also discussed. This knitting together of computational and experimental materials discovery and synthesis in a responsive feedback loop is promising for the on-demand synthesis of functional inorganic materials with targeted properties.
Thursday, May 28 at 11am – James Martin (North Carolina State University)
Solving Complex Problems: Measurement, Mindframe, and Models
Abstract: Most chemical/physical structures and processes exist in spatial and time domains that are difficult to observe directly. As a result, most measurements provide an observation averaged over the spatial and time domains of the probe. Thus, development of models is necessary to extract scientific significance from such measurements. While the iterative cycle of measurement and modeling forms the basis of the scientific method, advanced computational techniques, including AI and machine learning, are greatly enhancing the speed and apparent granularity with which atomic and molecular structures and processes can be modeled. Given that cognitive assumptions can distort the interpretation of an observation, it is increasingly necessary to apply the scientific-method principle of rigorous skepticism to ensure detailed pictures generated from models, whether de-novo or fit to data, are not simply confirmation bias. Frequently, multiple models can yield equivalent representations of an observation, and/or, distinct real scenarios can be represented by a single model. Consider the simplistic example for which the number pairs (0,100), (25, 75), and (50,50) are all represented by the same average model of 50; each pair, however, exhibits different realities only recognized with a mindframe distinct from considering the simple average. In this talk, examples from our research will be used to understand problems related to liquid/solution structure and the mechanism of crystallization. I will discuss cases where rigorous critique of the “scientific mindframe” and the integration of observations from distinct measurements is needed to distinguish between models and solve complex problems. This approach emphasizes the critical principles of: (a) examining the same problem using multiple tools, conditions, and measurements, while (b) being certain all model parameterization has real-world chemical and physical significance.