Quantum chemistry with neural-network wave functions

© 2023 Springer Nature

© 2023 Springer Nature

A recent review article in Nature Reviews Chemistry presents a comprehensive analysis of the integration of neural-network wave functions within quantum chemistry. Titled "Ab initio quantum chemistry with neural-network wave functions," this review offers insights into the potential and challenges of using machine learning techniques to solve the fundamental equations of quantum mechanics.

A recent review article published in Nature Reviews Chemistry delves into the emerging realm of quantum chemistry, where neural-network wave functions are being integrated to address fundamental challenges. Titled "Ab initio quantum chemistry with neural-network wave functions," this review paper offers an insightful analysis of the potential and limitations of employing neural networks in quantum chemistry applications.

This review article aims to provide a comprehensive overview of the innovative approach of integrating machine learning within quantum Monte Carlo methods. The collaborative effort between leading researchers from diverse institutions including Microsoft Research AI4Science, FU Berlin, DeepMind, University of Zurich, IBM Quantum, Imperial College London, and EPFL, showcases the interdisciplinary nature of this endeavor.

The review paper underscores that despite being in its nascent stages, the integration of neural-network wave functions with quantum Monte Carlo methods showcases promise in achieving remarkably accurate electronic-structure solutions for small systems. The method competes favorably with established approaches for such systems, and it holds potential for efficiently addressing electronic structures of systems with up to 100–200 electrons.

The review also outlines challenges that must be overcome, including issues related to size consistency, the accuracy limitations of neural-network ansatzes, and practical challenges in scaling the method to larger systems. These issues, while significant, are not insurmountable and can benefit from the simplicity of the framework and the ongoing advancements in deep learning architectures.

Furthermore, the review highlights that the Computational Quantum Science Lab (CQSL) at EPFL is actively engaged in the development of Neural Quantum States (neural wave functions), underscoring the lab's commitment to advancing the intersection of quantum chemistry and deep learning.

For the full insights and comprehensive analysis presented in the review, please refer to the complete paper


Hermann, J., Spencer, J., Choo, K. et al. Ab initio quantum chemistry with neural-network wavefunctions. Nat Rev Chem (2023)