Accelerating superconductor discovery through tempered deep learning of the electron-phonon spectral function

Abstract

Integrating deep learning with the search for new electron-phonon superconductors represents a burgeoning field of research, where the primary challenge lies in the computational intensity of calculating the electron-phonon spectral function, ${\alpha }^{2}F(\omega )$, the essential ingredient of Midgal-Eliashberg theory of superconductivity. To overcome this challenge, we adopt a two-step approach. First, we compute ${\alpha }^{2}F(\omega )$ for 818 dynamically stable materials. We then train a deep-learning model to predict ${\alpha }^{2}F(\omega )$, using an unconventional training strategy to temper the model's overfitting, enhancing predictions. Specifically, we train a Bootstrapped Ensemble of Tempered Equivariant graph neural NETworks (BETE-NET), obtaining an MAE of 0.21, 45 K, and 43 K for the Eliashberg moments derived from ${\alpha }^{2}F(\omega )$: $\lambda$, ${\omega }_{\log }$, and ${\omega }_2$, respectively, yielding an MAE of 2.5 K for the critical temperature, $T_c$. Further, we incorporate domain knowledge of the site-projected phonon density of states to impose inductive bias into the model's node attributes and enhance predictions. This methodological innovation decreases the MAE to 0.18, 29 K, and 28 K, respectively, yielding an MAE of 2.1 K for $T_c$. We illustrate the practical application of our model in high-throughput screening for high-$T_c$ materials. The model demonstrates an average precision nearly five times higher than random screening, highlighting the potential of ML in accelerating superconductor discovery. BETE-NET accelerates the search for high-$T_c$ superconductors while setting a precedent for applying ML in materials discovery, particularly when data is limited.

12 pages, 5 figures, 1 table

References in corpus (33)

• Quantum ESPRESSO: a modular and open-source software project for quantum simulations of materials
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Restoring the density-gradient expansion for exchange in solids and surfaces
• Generalized gradient approximation for solids and their surfaces
• Optimized norm-conserving Vanderbilt pseudopotentials
• Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties
• Conventional superconductivity at 203 K at high pressures
• The PseudoDojo: Training and grading a 85 element optimized norm-conserving pseudopotential table
• Quantum ESPRESSO toward the exascale
• A General-Purpose Machine Learning Framework for Predicting Properties of Inorganic Materials
• Superconductivity at 250 K in lanthanum hydride under high pressures
• Reconciling modern machine learning practice and the bias-variance trade-off
• Evidence for superconductivity above 260 K in lanthanum superhydride at megabar pressures
• Optimization Algorithm for the Generation of ONCV Pseudopotentials
• A Universal Graph Deep Learning Interatomic Potential for the Periodic Table
• Atomistic Line Graph Neural Network for Improved Materials Property Predictions
• Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
• The Joint Automated Repository for Various Integrated Simulations (JARVIS) for data-driven materials design
• Machine learning modeling of superconducting critical temperature
• Exploration of new superconductors and functional materials and fabrication of superconducting tapes and wires of iron pnictides
• Superconducting materials: Challenges and opportunities for large-scale applications
• Adaptive Genetic Algorithm for Crystal Structure Prediction
• Deep Learning Model for Finding New Superconductors
• Room Temperature Superconductivity: the Roles of Theory and Materials Design
• Ultra-fast interpretable machine-learning potentials
• Designing High-Tc Superconductors with BCS-inspired Screening, Density Functional Theory and Deep-learning
• Predicting new superconductors and their critical temperatures using unsupervised machine learning
• Direct prediction of phonon density of states with Euclidean neural networks
• Electron-phonon coupling, superconductivity and nontrivial band topology in NbN polytypes
• Automatic extraction of materials and properties from superconductors scientific literature
• High critical field superconductivity at ambient pressure in MoB${}_2$ stabilized in the P6/mmm structure via Nb substitution
• Diboride compounds doped with transition metals$\unicode{x2013}$a route to superconductivity through structure stabilization as well as defects
• Benign, Tempered, or Catastrophic: A Taxonomy of Overfitting

Cited by in corpus (5)

• AI-driven inverse design of materials: Past, present and future
• Machine-learning Guided Search for Phonon-mediated Superconductivity in Boron and Carbon Compounds
• Developing a Complete AI-Accelerated Workflow for Superconductor Discovery
• HTSC-2025: A Benchmark Dataset of Ambient-Pressure High-Temperature Superconductors for AI-Driven Critical Temperature Prediction
• Real-space orbital tiling approach for the design of novel superconductors