InvDesFlow: An AI-driven materials inverse design workflow to explore possible high-temperature superconductors

TL;DR

InvDesFlow, an artificial intelligence (AI)-driven materials inverse design workflow that integrates deep model pre-training and fine-tuning techniques, diffusion models, and physics-based approaches for the discovery of high-Tc superconductors is developed.

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.

Figures (3)

• Figure 1: (a) The Proposed InvDesFlow framework. AI-accelerated discovery of high-$T_c$ superconductors includes generative model for predicting crystal structures, pre-trained model for superconductivity classification, formation energy prediction model, screening model for superconducting transition temperature prediction, and validation using DFT calculation. (b) Symmetry-constrained crystal generation model. The generation of superconducting crystals defines two Markov processes: the black arrows represent the gradual addition of noise to a BCS superconducting crystal, resulting in a random unit cell, while the red arrows indicate the gradual denoising from a prior atomic distribution to generate the original superconducting crystal structure. The structures predicted by the generative AI have not yet converged in terms of energy and forces, requiring further post-processing. Here, we use DPA2 zhang2023dpa to predict the interatomic potentials (at DFT accuracy) and employ the atomic simulation environment ase-jcp to simulate the structural relaxation. (c) Formation Energy Prediction Model. A lower formation energy of a material indicates that its constituent elements adopt the lowest-energy configuration arrangement, which generally implies thermodynamic stability. In this step, we predict the formation energies of structurally optimized materials for subsequent screening of promising candidates. This model covers the sources of training data for formation energy prediction, crystal data representation using atomic graphs with an 8 Å cutoff radius, and the interactions between node, edge, and global state representations within the model’s architecture. (d) Superconducting Classification Model. This model includes a high-throughput screening process for pre-training and fine-tuning data, along with a graph auto-encoder architecture based on a graph neural network for superconductivity classification.
• Figure 2: (a)-(b) Electronic structure and DOS of B$_5$CN$_2$ at ambient pressure. (c)-(d) Electronic structure and DOS of B$_4$CN$_3$ under 5 GPa. The blue soild lines and red circles represent the bands obtained by DFT and Wannier projection, respectively. The Fermi level is set to be zero.
• Figure 3: (a)-(b) Phonon spectrum with a color representation of $\lambda_{q\nu}$, Eliashberg spectral function $\alpha^2F(\omega)$, and accumulated EPC constant $\lambda(\omega)$ for B$_5$CN$_2$ at ambient pressure. (c)-(d) Phonon spectrum with a color representation of $\lambda_{q\nu}$, Eliashberg spectral function $\alpha^2F(\omega)$, and accumulated EPC constant $\lambda(\omega)$ for B$_4$CN$_3$ under 5 GPa. The scale of $\alpha^2F(\omega)$ is omitted.