Recently, Professor Shang Honghui, a specially-appointed professor at the National Key Laboratory of Precision and Intelligent Chemistry at the University of Science and Technology of China (USTC), collaborated with the research team led by Academician Yang Jinlong. Together, they have ingeniously fused the Transformer architecture from artificial intelligence with the foundational equations of quantum physics to create QiankunNet—a cutting-edge network tailored to accurately solve the multi-electron Schrödinger equation. This groundbreaking work has been featured in Nature Communications.

The multi-electron Schrödinger equation stands as a cornerstone in quantum mechanics. However, traditional computational methods face significant limitations due to the exponential surge in computational complexity as the number of electrons increases. To overcome this, the research team ingeniously incorporated the attention mechanism from the Transformer architecture to construct quantum wave functions. By integrating an end-to-end differentiable architecture with an efficient autoregressive sampling algorithm grounded in Monte Carlo tree search, they successfully breached the 'exponential wall' bottleneck.

Tests have demonstrated that QiankunNet achieves an impressive 99.9% accuracy in correlation energy within 30 spin orbitals when compared to the full configuration interaction method. Moreover, it boasts a computational speed that is 10 times faster than traditional coupled-cluster methods. Its reliability in handling systems containing transition metals was further validated through simulations of the Fenton reaction, thereby offering a potent new tool for complex molecular simulations.