arXiv:2609.26032
Hyperbolic Restricted Boltzmann Machine Neural Quantum State
H. L. Dao
Abstract
We construct the first type of non-Euclidean non-autoregressive neural quantum state (NQS) in the form of the hyperbolic Restricted Boltzmann Machine (HRBM), which is studied in the variational Monte-Carlo (VMC) setting of the Quantum Sherrington-Kirkpatrick (QSK) model whose ground state exhibits volume-law entanglement. Across a 512-fold increase in the Hilbert space dimension corresponding to a system size increase from $N=14$ to $N=24$, HRBM NQS robustly outperforms its Euclidean version, the RBM NQS, in terms of better ground state energy optimization as well as lower Renyi-2 $S_2$ and von Neumann $S_{vN}$ absolute entanglement entropy reconstruction errors. More importantly, for all tested QSK system sizes, HRBM NQS demonstrates a superior expressivity in faithfully reproducing the entire entanglement spectrum of the QSK model from the top eigenvalues down to the tail end across 15 orders of magnitude, while RBM NQS consistently overestimates the sub-dominant modes. This work furnishes a proof-of-concept demonstrating that hyperbolic non-autoregressive NQS ansatze, thanks to the exponential volume of the hyperbolic geometry underlying their constructions, might be more natural at representing volume-law quantum systems than conventional Euclidean NQS. Furthermore, an interesting byproduct of this work is the polynomial scaling result of RBM-type NQS ansatze in the QSK volume-law system as the Hilbert space dimension increases exponentially.
AI Summary
The paper asks whether hyperbolic latent geometry improves neural quantum-state representations of volume-law-entangled ground states. It constructs a hyperbolic restricted Boltzmann machine (HRBM) using Lorentz-hyperboloid exponential/logarithmic maps, matrix multiplication and bias addition, with separate embeddings for real and imaginary components. HRBM retains the same trainable parameter count as a conventional complex RBM but introduces a spatial-cutoff hyperparameter, $L_{\max}$. The comparison uses variational Monte Carlo for the all-to-all disordered quantum Sherrington–Kirkpatrick model at fixed transverse field $\Gamma=-1$, with $N=14,16,\ldots,24$. Each size includes 12 disorder realizations and five network initializations per architecture, benchmarked against exact diagonalization. Both networks use $N^2+2N$ parameters, ranging from 224 to 624. HRBM has lower disorder-averaged relative energy error and energy variance at every tested size. At $N=24$, relative energy errors are $2.82\times10^{-4}$ for HRBM and $3.26\times10^{-4}$ for RBM; absolute Rényi-2 entropy errors are $2.41\times10^{-3}$ and $4.67\times10^{-3}$, respectively, while von Neumann entropy errors are similar. Reported entanglement-spectrum comparisons show HRBM tracking the exact spectral decay down to approximately $10^{-15}$ for most realizations, whereas RBM generally overestimates small eigenvalues. The study is explicitly a proof of concept limited to one model and sizes accessible to exact diagonalization. Some individual realizations favor RBM. Performance depends on cutoff selection: $L_{\max}$ is increased from 10 to 20 at $N=24$. Larger-size behavior and extensions to other architectures or fermionic systems remain untested.
Central claim may not hold: The Hilbert-space dimension increases by 1024-fold between N=14 and N=24, not the repeatedly stated 512-fold. The finite-training benchmarks do not establish superior representational expressivity or the claimed geometric restriction on Euclidean RBMs. The paper's assertion that quantum entanglement spectra decay exponentially in eigenvalue index is not generally valid and is not established for its benchmark family. The reported N=24 von Neumann entropy errors do not establish a robust HRBM advantage without paired uncertainty analysis, and individual disorder realizations sometimes favor RBM. The N=14–24 experiments support finite-size performance with quadratic parameter counts, not the claimed numerical proof of asymptotic polynomial scaling.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- v2: more QSK disorder realizations added in expanded VMC experiments - more robust results obtained, conclusions unchanged + more references & descriptions added