Advancing Universal Deep Learning for Electronic-Structure Hamiltonian Prediction of Materials

Researchers developed NextHAM, a deep learning method for predicting electronic-structure Hamiltonians of materials, offering significant computational efficiency advantages over traditional DFT methods. The system introduces neural E(3)-symmetry architecture and a new dataset Materials-HAM-SOC with 17,000 material structures spanning 68 elements.

• →NextHAM uses zeroth-step Hamiltonians as informative descriptors to simplify the input-output mapping for neural regression models.
• →The method employs a neural Transformer architecture with strict E(3)-symmetry for high expressiveness in Hamiltonian prediction.
• →Materials-HAM-SOC dataset comprises 17,000 material structures covering 68 elements from six periodic table rows.
• →The system prevents error amplification and 'ghost states' through novel training objectives for both real and reciprocal space accuracy.
• →Experimental results demonstrate excellent accuracy and efficiency in predicting Hamiltonians and band structures compared to traditional methods.