QuantumShellNet: Ground-state eigenvalue prediction of materials using electronic shell structures and fermionic properties via convolutions

Abstract

Efficient and precise characterization of material properties is critical in quantum mechanical modeling. While Density Functional Theory (DFT) remains a foundational method for analyzing material properties, it faces scalability challenges and precision limitations, especially with complex materials. This study introduces QuantumShellNet, a novel vision-based approach that combines an orbital encoder and a physics-informed deep neural network. QuantumShellNet is specifically designed to rapidly and accurately predict ground-state eigenvalues in materials by leveraging electronic shell structures and their fermionic properties. Experiments conducted across a diverse range of elements and molecules show that QuantumShellNet outperforms traditional DFT as well as modern machine learning methods, including PsiFormer and FermiNet.