Autoregressive Neural Quantum States with Quantum Number Symmetries

Aleksei Malyshev,Juan Miguel Arrazola,A. I. Lvovsky
2023-10-06
Abstract:Neural quantum states have established themselves as a powerful and versatile family of ansatzes for variational Monte Carlo simulations of quantum many-body systems. Of particular prominence are autoregressive neural quantum states (ANQS), which enjoy the expressibility of deep neural networks, and are equipped with a procedure for fast and unbiased sampling. Yet, the non-selective nature of autoregressive sampling makes incorporating quantum number symmetries challenging. In this work, we develop a general framework to make the autoregressive sampling compliant with an arbitrary number of quantum number symmetries. We showcase its advantages by running electronic structure calculations for a range of molecules with multiple symmetries of this kind. We reach the level of accuracy reported in previous works with more than an order of magnitude speedup and achieve chemical accuracy for all studied molecules, which is a milestone unreported so far. Combined with the existing effort to incorporate space symmetries, our approach expands the symmetry toolbox essential for any variational ansatz and brings the ANQS closer to being a competitive choice for studying challenging quantum many-body systems.
Quantum Physics,Disordered Systems and Neural Networks,Chemical Physics,Computational Physics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to effectively combine quantum number symmetries in autoregressive neural quantum states (ANQS) to improve the efficiency and accuracy of variational Monte Carlo (VMC) simulations. Specifically, the authors propose a general framework that enables autoregressive sampling to conform to an arbitrary number of quantum number symmetries. This solves the problem in existing methods that these symmetries cannot be effectively handled, thereby improving the precision and speed of electronic structure calculations. ### Background of the Paper 1. **Variational Monte Carlo (VMC)** - VMC is a method for solving the ground state of quantum many - body systems. It selects a parameterized quantum state (ansatz) and optimizes these parameters to minimize the expected value of energy. - Traditional NQS methods usually use Metropolis - Hastings sampling, but this method has problems such as long autocorrelation time and difficulty in handling multimodal distributions. 2. **Autoregressive Neural Quantum States (ANQS)** - ANQS is a quantum state representation method based on autoregressive neural networks and can sample quickly and unbiasedly. - However, existing ANQS methods have difficulties in dealing with quantum number symmetries because autoregressive sampling is non - selective and cannot directly exclude samples that are not in the correct symmetry sector. ### Contributions of the Paper 1. **General Framework** - A general algorithm is proposed that can automatically identify and exclude partial basis vectors that do not satisfy quantum number symmetries during the autoregressive sampling process. - This algorithm is applicable to an arbitrary number of quantum number symmetries and has polynomial complexity and will not lead to exponential computational overhead. 2. **Electronic Structure Calculations** - This framework is applied to molecular electronic structure calculations, taking into account the number of particles, spin projection, and spatial symmetry (Z2 symmetry). - The experimental results show that this method not only reaches chemical accuracy but is also more than an order of magnitude faster than existing ANQS methods. ### Main Results 1. **Improvement in Precision** - For all the molecules studied, the median error of this method is lower than the chemical accuracy (1.6 mHa), which is a milestone that has not been reported before. - In particular, for molecules with higher spatial symmetry (such as LiCl, N2, LiF, Li2O, and C2), the improvement in precision is especially significant, up to an order of magnitude. 2. **Computational Efficiency** - Compared with existing ANQS methods, the computational time of this method is significantly reduced when the same accuracy is achieved. - The experimental results show that this method can reach CCSD accuracy within 3 × 10^4 iterations, and the time required is much less than that of traditional methods. ### Conclusion This paper successfully combines quantum number symmetries in ANQS by proposing a general framework, significantly improving the precision and efficiency of electronic structure calculations. This result provides a powerful tool for studying complex quantum many - body systems.