QuantumShellNet

Efficient and precise characterization of material properties is essential in Quantum Mechanical Modeling. Density Functional Theory (DFT), while a foundational method in material property analysis, suffers from scalability issues and precision limitations, particularly with complex materials. In response, we introduce QuantumShellNet, a novel, vision-based approach with an orbital encoder and a physics-informed deep neural network, meticulously engineered for the rapid and accurate prediction of ground-state eigenvalues of materials using electronic shell structures and their fermionic properties. Our experiments across a diverse set of elements and molecules reveal that QuantumShellNet outperforms both traditional DFT and modern machine learning counterparts, including PsiFormer and FermiNet.