Many-Body Neural Network Wavefunction for a Non-Hermitian Ising Chain

Lavoisier Wah, Remmy Zen, Flore K. Kunst · 2025

Read Many-Body Neural Network Wavefunction for a Non-Hermitian Ising Chain by Lavoisier Wah, Remmy Zen, Flore K. Kunst (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-11222.txt.

Introduction

Lavoisier Wah and co-authors test neural network wavefunctions on a non-Hermitian quantum chain in a 2025 paper. Open quantum systems are often described by non-Hermitian models, but their ground states are hard to compute because the state space grows exponentially and exact diagonalization quickly hits its limits.

The team builds many-body wavefunctions from three network types: a recurrent network, a restricted Boltzmann machine, and a multilayer perceptron. On small systems the networks match exact results well. On larger ones the recurrent network wins, though transfer learning lifts the other two a long way. The model studied is a parity-time-symmetric Ising chain with a complex spectrum.

Catalog record

CREATORLavoisier Wah, Remmy Zen, Flore K. Kunst
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-11222

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Many-Body Neural Network Wavefunction for a Non-Hermitian Ising Chain Lavoisier Wah, Remmy Zen, Flore K. Kunst Abstract Non-Hermitian (NH) quantum systems have emerged as a powerful framework for describing open quantum systems, non-equilibrium dynamics, and engineered quantum optical materials. However, solving the ground-state properties of NH systems is challenging due to the exponential scaling of the Hilbert space, and exotic phenomena such as the emergence of exceptional points. Another challenge arises from the limitations of traditional methods like exact diagonalization (ED). For the past decade, neural networks (NN) have shown promise in approximating many-body wavefunctions, yet their application to NH systems remains largely unexplored. In this paper, we explore different NN architectures to investigate the ground-state properties of a parity-time-symmetric, one-dimensional NH, transverse field Ising model with a complex spectrum by employing a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), and a multilayer perceptron (MLP). We construct the NN-based many-body wavefunctions and validate our approach by recovering the ground-state properties of the model for small system sizes, finding excellent agreement with ED. Furthermore, for larger system sizes, we demonstrate that the RNN outperforms both the RBM and MLP. However, we show that the accuracy of the RBM and MLP can be significantly improved through transfer learning, allowing them to perform comparably to the RNN for larger system sizes. These results highlight the potential of neural network-based approaches—particularly for accurately capturing the low-energy physics of NH quantum systems. I Introduction Traditional quantum mechanics relies on Hermitian Hamiltonians to ensure real eigenvalues and unitary time evolution. However, non-Hermitian (NH) quantum mechanics has emerged as a powerful framework to describe open quantum systems, where interactions with the environment may lead to energy dissipation and gain Ashida et al. 2020 ; Bergholtz et al. 2021 . A particularly intriguing subclass of non-Hermitian systems consists of those obeying parity-time ( PT ) symmetry, where the Hamiltonian remains invariant under simultaneous spatial reflection and time-reversal operations Bender and Boettcher 1998 . Such systems exhibit real eigenvalues in the unbroken PT -symmetric phase, and undergo a phase transition through an exceptional point (EP), where two or more eigenvalues and their corresponding eigenvectors coalesce. Beyond this point, the system enters the broken PT -symmetric regime, where eigenvalues become complex El-Ganainy et al. 2018 . This rich phenomenology has inspired studies across condensed matter physics, quantum optics, and cold-atom systems Ateuafack et al. 2025 ; Zhang and Wei 2024 ; Bergholtz et al. 2021 ; Özdemir et al. 2019 ; El-Ganainy et al. 2018 ; Li et al. 2019 ; Ren et al. 2022 ; Ding et al. 2021 . Recent advances have extended NH physics from single-particle systems to many-body quantum systems, unveiling novel exotic phenomena and a richer understanding of EPs and the NH skin effect leading to novel phases Gohsrich et al. 2025 , unconventional localization and distinct quantum criticality Luitz and Piazza 2019 ; Zhang et al. 2020 ; Lu et al. 2024 ; Sun et al. 2021 . These developments have been explored in various contexts, including NH extensions of the transverse field Ising model (TFIM), where complex interactions lead to unconventional symmetry breaking and dynamical properties Lu et al. 2024 ; Lenke et al. 2021 ; Li et al. 2014 ; Guerra et al. 2025 . Despite significant progress in theoretical formulations, an efficient numerical method to determine the ground-state properties of non-Hermitian many-body systems remains an open challenge due to the exponential scaling of the Hilbert space as well as the instabilities induced by the emergence of EPs. The application of artificial intelligence and machine learning techniques in quantum many-body physics has provided innovative approaches to studying strongly correlated systems Krenn et al. 2023 ; Carleo et al. 2019 ; Carrasquilla 2020 . Neural-network-based variational wavefunctions or neural quantum states (NQSs), such as restricted Boltzmann machines (RBMs), multilayer perceptrons (MLPs), and recurrent neural networks (RNNs) have demonstrated remarkable success in approximating ground states of complex quantum systems Carleo and Troyer 2017 ; Hibat-Allah et al. 2020 ; Pilati et al. 2019 ; Inack et al. 2018 ; Taud and Mas 2017 . By parameterizing the wavefunction with a neural-network model, it can capture entanglement and correlation structures efficiently, making them well-suited for studying quantum phases and transitions Carleo and Troyer 2017 ; Hibat-Allah et al. 2020 ; Sharir et al. 2020 . However, NQSs has so far been restricted to Hermitian many-body systems and their application to NH many-body physics offers a promising alternative to traditional numerical techniques, such as exact diagonalization (ED) and tensor network methods, which become computationally prohibitive for large system sizes. Very recently, these techniques have only been applied to study NH fermionic systems with real spectrum using the RNN Ibarra-García-Padilla et al. 2025 . Here, we propose to further the study to the realm of many-body bosonic NH systems with a complex spectrum , which is the “raison d’être” of this paper. Our goal is to develop fast and efficient algorithms capable of capturing the many-body wavefunction of bosonic NH systems and propose a method to investigate their ground state. Extending many-body neural-network wavefunctions to NH systems enables the exploration of exotic ground states, particularly in PT -symmetric models, where standard Hermitian variational techniques may fail. This work aims to develop and apply a neural-network-based variational Monte Carlo (VMC) method to study the ground-state properties of a PT -symmetric non-Hermitian TFIM with complex spectrum. The TFIM is the perfect model to test VMC in NH systems because the model can be solved analytically Lenke et al. 2021 ; Li et al. 2014 . Using our numerical techniques, we not only find that our results are in perfect agreement with ED but also provide new insights into metastable states that are not fully accessible with ED as discussed later. In what follows, we present our model in Sec. II and discuss the concept of the ground state in NH systems. In Sec. III , we demonstrate how to recover the ground state for our model. We then show in Sec. IV how transfer learning (TL) can enhance the performance of the neural networks. In Sec. V , we study the unbroken-to-broken phase transition and show the emergence of EPs. Finally, our results are summarized in Sec. VI . II Model and non-Hermitian symmetry II.1 Hamiltonian Figure 1: 1D Ising chain in a staggered magnetic field . A (red) and B (blue) represent the sublattices in which spin- 1/2 particles occupy the odd and even positions, respectively. The complex field g ( g^{*} ) is applied to the A ( B ) sublattices. Neighboring spins are coupled via the coupling constant J . The model of interest consists of a one-dimensional (1D) quantum Ising chain of N spin- 1/2 particles on two sublattices A and B coupled via a spin coupling J and subject to a complex, staggered magnetic field that introduces non-Hermiticity into the system (see Fig. 1 ). The NH field is chosen so that g=\eta+i\xi is applied perpendicularly to the spins on sublattice A (red) and g^{*}=\eta-i\xi to the spins on sublattice B (blue) with \eta,\xi\in\mathbb{R} . The Hamiltonian describing such a system is given by \displaystyle H=-J\sum_{j=1}^{N}\sigma_{j}^{z}\sigma_{j+1}^{z}-g\sum_{j\in A}\sigma_{j}^{x}-g^{*}\sum_{j\in B}\sigma_{j}^{x}, (1) where \sigma_{j}^{z} and

Read the complete text