InvDesFlow: An AI search engine to explore possible high-temperature superconductors (2409.08065v2)

Abstract: The discovery of new superconducting materials, particularly those exhibiting high critical temperature ($T_c$), has been a vibrant area of study within the field of condensed matter physics. Conventional approaches primarily rely on physical intuition to search for potential superconductors within the existing databases. However, the known materials only scratch the surface of the extensive array of possibilities within the realm of materials. Here, we develop InvDesFlow, an AI search engine that integrates deep model pre-training and fine-tuning techniques, diffusion models, and physics-based approaches (e.g., first-principles electronic structure calculation) for the discovery of high-$T_c$ superconductors. Utilizing InvDesFlow, we have obtained 74 dynamically stable materials with critical temperatures predicted by the AI model to be $T_c \geq$ 15 K based on a very small set of samples. Notably, these materials are not contained in any existing dataset. Furthermore, we analyze trends in our dataset and individual materials including B$_4$CN$_3$ (at 5 GPa) and B$_5$CN$_2$ (at ambient pressure) whose $T_c$s are 24.08 K and 15.93 K, respectively. We demonstrate that AI technique can discover a set of new high-$T_c$ superconductors, outline its potential for accelerating discovery of the materials with targeted properties.

• The paper introduces InvDesFlow, which integrates generative models, machine learning classifiers, and stability assessments to generate novel high-Tc superconductors.
• It employs an active learning strategy that refines predictions by iteratively incorporating discovered superconducting materials into the training set.
• Validated through DFT with 74 new candidates, including B4CN3 with a computed Tc of 24.08 K, the framework demonstrates significant potential for accelerating superconductor discovery.

InvDesFlow: Exploration of High-Temperature Superconductors

The field of condensed matter physics continuously seeks advancements in superconducting materials, especially those with high critical temperatures ($T_c$). The conventional methodologies often rely on existing databases and physical intuition, limiting exploration potential. This paper introduces InvDesFlow, a comprehensive AI-powered framework aimed at discovering new high-$T_c$ superconductors by leveraging generative models, machine learning, and physics-based approaches.

Methodology

InvDesFlow adopts a multifaceted approach, integrating several AI and computational techniques to enhance the discovery process of superconductors:

1. Generative Models: Utilizing symmetry-constrained diffusion models, InvDesFlow generates novel crystal structures not present in existing databases. This is bolstered by equivalent graph neural networks (EGNN) for atomic structure denoising.
2. Superconducting Classification: A pre-trained and fine-tuned graph neural network classifies the generated materials as superconducting or non-superconducting, achieving a high discrimination success rate. This step ensures only superconducting candidates proceed further.
3. Stability Assessment: An enhanced formation energy prediction model, based on MEGNET architecture, evaluates the stability of the generated materials. The model incorporates additional atomic features, improving its prediction accuracy.
4. $T_c$ Prediction: The Atomistic Line Graph Neural Network (ALIGNN) forecasts the critical temperatures of these materials. A 15 K threshold is applied as a benchmark to identify high-$T_c$ candidates.
5. Validation: High-performing candidates undergo density functional theory (DFT) calculations to verify their superconducting properties. Special attention is given to dynamically stable materials such as B$_4$CN$_3$ and B$_5$CN$_2$, with confirmed \textit{ab initio} $T_c$ of 24.08 K and 15.93 K, respectively.
6. Active Learning: The iterative incorporation of discovered superconductors into the training set enhances the AI model's predictive capability over time, extending its utility across a broader chemical space.

Results

InvDesFlow demonstrates its prowess by successfully predicting and validating 74 novel high-$T_c$ superconductors, with critical temperatures exceeding 15 K. The focus on materials like B$_4$CN$_3$ and B$_5$CN$_2$ exemplifies the successful integration of AI-driven approaches in identifying potentially transformative materials absent from existing datasets.

Implications and Future Work

The potential impact of InvDesFlow is significant both in theoretical and practical domains of materials science. The methodology not only enhances material discovery efficiency but also expands the exploration of novel superconductors beyond the confines of current knowledge. The AI engine's flexibility allows its application to a wide array of functional materials, fostering broad advancements in material science.

Looking ahead, InvDesFlow can be instrumental in further refining superconducting material predictions, driving forward our understanding of high-$T_c$ superconductivity. This technology promises to not only accelerate discoveries but also guide experimental efforts in synthesizing these computationally discovered materials. Future developments may focus on incorporating other physical properties and scalable computing solutions, amplifying the design and discovery process for targeted applications in superconductivity.