Abstract:Neural network quantum states (NQS) have been widely applied to spin-1/2 systems where they have proven to be highly effective. The application to systems with larger on-site dimension, such as spin-1 or bosonic systems, has been explored less and predominantly using spin-1/2 Restricted Boltzmann Machines (RBMs) with a one-hot/unary encoding. Here we propose a more direct generalisation of RBMs for spin-1 that retains the key properties of the standard spin-1/2 RBM, specifically trivial product states representations, labelling freedom for the visible variables and gauge equivalence to the tensor network formulation. To test this new approach we present variational Monte Carlo (VMC) calculations for the spin-1 antiferromagnetic Heisenberg (AFH) model and benchmark it against the one-hot/unary encoded RBM demonstrating that it achieves the same accuracy with substantially fewer variational parameters. Further to this we investigate how the hidden unit complexity of NQS depend on the local single-spin basis used. Exploiting the tensor network version of our RBM we construct an analytic NQS representation of the Affleck-Kennedy-Lieb-Tasaki (AKLT) state in the $xyz$ spin-1 basis using only $M = 2N$ hidden units, compared to $M \sim O(N^2)$ required in the $S^z$ basis. Additional VMC calculations provide strong evidence that the AKLT state in fact possesses an exact compact NQS representation in the $xyz$ basis with only $M=N$ hidden units. These insights help to further unravel how to most effectively adapt the NQS framework for more complex quantum systems.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: How to construct Neural - network Quantum States (NQS) that are more efficient and have a smaller number of parameters for spin - 1 systems, in order to improve their compactness and accuracy in representing complex quantum states. Specifically, the paper mainly focuses on the following points: 1. **Limitations of existing methods**: - For spin - 1/2 systems, NQS has achieved very good results. However, for systems with larger local dimensions (such as spin - 1 or boson systems), there are relatively few studies on the application of NQS. - Traditional methods usually use "one - hot" or "unary" encoding to handle multi - valued variables, which will significantly increase the number of visible units, resulting in a substantial increase in the number of parameters and optimization complexity, thus limiting the scale of the system. 2. **Proposed new method**: - The paper proposes a method of directly generalizing Restricted Boltzmann Machines (RBMs) to spin - 1 systems, retaining the key characteristics of spin - 1/2 RBMs, such as the ability to describe any direct - product state, the freedom of assigning values to visible variables (labelling freedom), and the equivalence with tensor network representations. - New quadratic biases and interaction weights are introduced to adapt to the characteristics of spin - 1 systems. 3. **Verify the effectiveness of the new method**: - Through Variational Monte Carlo (VMC) calculations, the spin - 1 antiferromagnetic Heisenberg model (AFH model) is studied and compared with the traditional "one - hot" encoding method. The results show that the new method can significantly reduce the number of variational parameters while maintaining the same accuracy. - The influence of different local single - spin bases on the complexity of hidden units is studied, especially the performance in the xyz basis and the Sz basis. 4. **Specific case analysis**: - Through the analysis of the Affleck - Kennedy - Lieb - Tasaki (AKLT) model, it is shown that the new method can accurately represent the AKLT state with only $ M = 2N $ or even $ M = N $ hidden units in the xyz basis, while in the Sz basis, it requires $ M \sim O(N^2) $ hidden units. In summary, this paper aims to improve the structure of NQS to make it more suitable for spin - 1 systems and improve its efficiency and accuracy in representing complex quantum states.

Neural-network Quantum States for Spin-1 systems: spin-basis and parameterization effects on compactness of representations

Neural-Network Quantum States for Spin-1 Systems: Spin-Basis and Parameterization Effects on Compactness of Representations

Compact Neural-network Quantum State representations of Jastrow and Stabilizer states

Specialising Neural-network Quantum States for the Bose Hubbard Model

Efficiency of neural-network state representations of one-dimensional quantum spin systems

Unitary-Coupled Restricted Boltzmann Machine Ansatz for Quantum Simulations

Efficiency of neural quantum states in light of the quantum geometric tensor

Neural Network Quantum States for the Interacting Hofstadter Model with Higher Local Occupations and Long-Range Interactions

Helping restricted Boltzmann machines with quantum-state representation by restoring symmetry

Optimizing Design Choices for Neural Quantum States

Entanglement Features of Random Neural Network Quantum States

Neural Network Quantum States analysis of the Shastry-Sutherland model

Universal Performance Gap of Neural Quantum States Applied to the Hofstadter-Bose-Hubbard Model

Accuracy of restricted Boltzmann machines for the one-dimensional $J_1-J_2$ Heisenberg model

Autoregressive neural quantum states of Fermi Hubbard models

Neural Quantum State Study of Fracton Models

Quantum tensor network algorithms for evaluation of spectral functions on quantum computers

Highly resolved spectral functions of two-dimensional systems with neural quantum states

Adaptive variational ground state preparation for spin-1 models on qubit-based architectures

Neural network approach to quasiparticle dispersions in doped antiferromagnets

Benchmarking discrete truncated Wigner approximation and neural network quantum states with the exact dynamics in a Rydberg atomic chain