Inverse Design of Novel Superconductors via Guided Diffusion

Pawan Prakash, Jason B. Gibson, Zhongwei Li, Gabriele Di Gianluca, Juan Esquivel, Eric Fuemmeler, Benjamin Geisler, Adrian Roitberg, Ellad B. Tadmor, Mingjie Liu, Stefano Martiniani, Gregory R. Stewart, James Hamlin, Peter Hirschfeld, Richard Hennig

Keywords: Guided Diffusion, Superconductivity, Materials Discovery, Generative AI, Inverse Design, Density Functional Theory (DFT), Machine Learning

TL;DR: We engineered an end-to-end workflow by developing a guided diffusion model to generate new materials and utilizing machine learning models to characterize them, ultimately discovering 773 new superconducting candidates.

Abstract: The inverse design of materials with specific desired properties, such as high-temperature superconductivity, represents a formidable challenge in materials science due to the vastness of chemical and structural space. This work introduces a guided diffusion framework to accelerate the discovery of novel superconductors. Our approach begins with a foundation model trained on crystal structures from the Alexandria Database, which is then fine-tuned using a labeled dataset of 7,217 conventional superconductors to generate new structures conditioned on critical temperature, $T_\mathrm{c}$. Employing classifier-free guidance, we generated 200,000 potential crystal structures. These candidates were subsequently subjected to a rigorous multi-stage computational screening workflow, utilizing machine learning models and density functional theory calculations to assess stability and electronic properties. Notably, our generative model demonstrated effective property-driven design by shifting the distribution of generated materials toward targeted $T_\mathrm{c}$ values within the training regime. This process successfully identified 773 promising superconducting candidates with predicted $T_\mathrm{c}>5K$. This end-to-end workflow, from generation to new candidate superconductors, illuminates a powerful pathway for materials discovery, demonstrating the significant potential of the AI-driven framework to accelerate discovery.

Submission Track: Paper Track (Full Paper)

Submission Category: AI-Guided Design

Institution Location: {Gainesville (FL), USA}, {New York City, USA}, {Minneapolis, USA}

Submission Number: 106