1: \begin{abstract}

2: 

3: Estimating free energy is a fundamental problem in statistical mechanics.

4: Recently, machine-learning-based methods, particularly the variational autoregressive networks (VANs), have been proposed to minimize variational free energy and to approximate the Boltzmann distribution. 

5: VAN enjoys notable advantages, including exact computation of the normalized joint distribution and fast unbiased sampling, which are critical features often missing in Markov chain Monte Carlo algorithms.

6: However, VAN also faces significant computational challenges.

7: These include difficulties in the optimization of variational free energy in a complicated parameter space and slow convergence of learning.

8: In this work, we introduce an optimization technique based on natural gradients to the VAN framework, namely \textit{ng-VAN}, to enhance the learning efficiency and accuracy of the conventional VAN.

9: The method has computational complexity cubic in the batch size rather than in the number of model parameters, hence it can be efficiently implemented for a large VAN model.

10: We carried out extensive numerical experiments on the Sherrington-Kirkpatrick model and spin glasses on random graphs and illustrated that compared with the conventional VAN, ng-VAN significantly improves the accuracy in estimating free energy and converges much faster with shorter learning time. 

11: This allows extending the VAN framework's applicability to challenging statistical mechanics problems that were previously not accessible.

12: \end{abstract}

13: