Diagonalizing large-scale quantum many-body Hamiltonians using variational quantum circuit and tensor network

Peng-Fei Zhou, Shuang Qiao, An-Chun Ji, Shi-Ju Ran · 2025

Read Diagonalizing large-scale quantum many-body Hamiltonians using variational quantum circuit and tensor network by Peng-Fei Zhou, Shuang Qiao, An-Chun Ji, Shi-Ju Ran (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-2508-06159.txt.

Introduction

Peng-Fei Zhou and co-authors diagonalize huge quantum systems by splitting the job between two compact tools. Their 2025 paper notes that exact diagonalization gives every energy level and state but grows in cost explosively, trapping it at small sizes. Their method stores the full energy spectrum inside a matrix product state.

A shallow quantum circuit then builds the matching states from simple starting states. The pairing cuts the cost from exponential to polynomial in system size. On an Ising chain, the method matches exact energies up to 16 sites and samples all levels at 100 sites, far beyond exact reach. Comparing two other chains at special limits exposes how their states tangle differently, which governs how hard they are to simulate.

Catalog record

CREATORPeng-Fei Zhou, Shuang Qiao, An-Chun Ji, Shi-Ju Ran
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-06159

Browse the full library · About the Tower of Babel

Full text

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

Diagonalizing large-scale quantum many-body Hamiltonians using variational quantum circuit and tensor network Peng-Fei Zhou, Shuang Qiao, An-Chun Ji, Shi-Ju Ran Abstract Exact diagonalization (ED) is an essential tool for exploring quantum many-body physics but is fundamentally limited by the exponentially-scaled computational complexity. Here, we propose tensor network variational diagonalization (TNVD), which encodes the full eigenenergy spectrum of a quantum many-body Hamiltonian into a matrix product state, and encodes the eigenstates as the evolutions of product states using variational quantum circuit (VQC). Thereby, TNVD reduces the computational complexity of diagonalization from exponential to polynomial in system size N N . Numerical benchmarks up to N = 100 N=100 spins are provided, which far surpass the computational limit of ED. We further consider quantum Ising model in a random field to reveal the underlying reliance between the efficiency of TNVD and entanglement properties of eigenstates. Typical signs, including the distribution of entanglement entropy (EE) versus eigenenergy and the density of state versus EE, are suggested to indicate area law of entanglement entropy or its violation, which are essential to the TNVD efficiency. Our work establishes TNVD as a powerful and scalable diagonalization approach for large-scale quantum many-body Hamiltonians. The incorporation of VQC lays a promising pathway to applying quantum computation to address the volume-law-EE Hamiltonians that lack efficient classical approaches. Introduction.— The eigenenergy spectra of quantum many-body Hamiltonians encode not only equilibrium characteristics but also the full structure of quantum dynamics. They underpin our understanding of criticality, phase transitions. Analytical methods, such as Bethe ansatz Heisenberg 1928 ; Bethe 1931 , provide invaluable exact solutions but are inapplicable to generic non-integrable systems. Exact diagonalization (ED) computes the complete spectrum for small-size systems, making it a fundamental tool with wide applications in, e.g., thermalization, quantum chaos, and eigenstate entanglement Neumann 1929 ; Deutsch 1991a ; Srednicki 1994a . However, the computational cost of ED in general scales exponentially with the system size, which is known as the “exponential wall”, severely limiting its applicability. Taking interacting spin systems as examples, the sizes that ED can access are generally less than 30 spins, even after exploiting symmetries of the Hamiltonians Lin et al. 1993 ; Dagotto 1994 ; Luitz et al. 2015 . Tremendous efforts have been made to access the eigenenergies and eigenstates of large-scale quantum many-body Hamiltonians. Among others, density matrix renormalization group (DMRG) White 1992 ; White and Huse 1993 and its variants Xiang 1996 ; Khemani et al. 2016 ; Yu et al. 2017 ; Zhang et al. 2023 can efficiently simulate the eigenenergies and eigenstates of large quantum many-body systems, but are limited to a small part of the eigenstates, say the low-lying eigenstates or the scar states. More recently, inspiring progress has been made in developing variational diagonalization methods, such as variational unitary matrix product operator ansatz Pollmann et al. 2016 ; Wahl et al. 2017 . Still, efficient diagonalization approaches with polynomial complexity in system size are strongly desired, In this work, we aim to tackle the “exponential wall” in diagonalizing quantum many-body Hamiltonians by proposing an efficient diagoalization approach named tensor network variational diagonalization (TNVD). Our key idea is encoding the full eigenenergy spectrum of the Hamiltonian into a matrix product state (MPS), and meanwhile encodes the eigenstates as the evolutions of product states using variational quantum circuit (VQC) Cerezo et al. 2021 ; Bharti et al. 2022 , thereby reducing the computational complexity of diagonalization from exponential to polynomial in system size N . Our benchmarks demonstrate high efficiency in diagonalizing quantum spin systems of sizes up to N=100 , which far surpass the computational limit of ED and the existing variational approaches Chandran et al. 2015 ; Pollmann et al. 2016 ; Wahl et al. 2017 ; Choo et al. 2018 ; Glasser et al. 2018 . The computational error can be well controlled by a logarithmic Schmidt distance Vedral and Plenio 1998 that can be simulated classically with high efficiency. We further reveal the underlying relation between TNVD efficiency and the entanglement properties of eigenstates by considering quantum Ising model in a random longitudinal field as an example. Our results suggest high accuracy of TNVD in the standard and deep many-body localized phases Anderson 1958 ; Gornyi et al. 2005 ; Basko et al. 2006 ; Oganesyan and Huse 2007 ; Serbyn et al. 2013 ; Devakul and Singh 2015 , where the entanglement entropy (EE) of eigenstates are deemed to obey the area law Bauer and Nayak 2013 ; Kjäll et al. 2014a ; Friesdorf et al. 2015 ; Herviou et al. 2019 . Within the thermalized phase Deutsch 1991b ; Srednicki 1994b , our results imply two sub-regions: a transition sub-region from area-law to volume-law EE, and a sub-region exhibiting volume-law EE. Numerical evidences and typical signs, including the errors of TNVD, distribution of EE versus eigenenergy, and density of states versus EE, are suggested, which allow to assess the strength of engenstate entanglement and TNVD efficiency. Figure 1: (Color online) Illustration of TNVD ansatz encoding of the eigenvalue decomposition of a given Hamiltonian [Eq. ( 1 )]. The exponentially-many eigenenergies \{E_{\alpha}\} are encoded into an MPS. The eigenstates are encoded into the evolutions of VQC on product states as |\alpha\rangle=\hat{U}|r_{\alpha}\rangle . The ansatz is formed by the MPS and VQC (and its conjugate) connected by super-diagonal tensors. Tensor network variational diagonalization approach.— Considering a quantum system of N spin- 1/2 ’s, the eigenvalue decomposition of its Hamiltonian \hat{H} is given by \hat{H}=\sum_{\alpha=0}^{2^{N}-1}E_{\alpha}|\alpha\rangle\langle\alpha|, (1) where \{E_{\alpha}\} form the eigenenergy spectrum and \{|\alpha\rangle\} represent the eigenstates satisfying \hat{H}|\alpha\rangle=E_{\alpha}|\alpha\rangle . The index \alpha can be expressed in the binary form as \alpha\equiv(r_{1},r_{2},\cdots,r_{N}) with r_{m}=0 or 1 . Then we write |\alpha\rangle as the evolution by the unitary transformation \hat{U} on the product state \{|r_{\alpha}\rangle\equiv|r_{1}r_{2}\cdots r_{N}\rangle\} as |\alpha\rangle=\hat{U}|r_{\alpha}\rangle . By treating the eigenenergy spectrum as an N th-order tensor E_{r_{1}r_{2}\cdots r_{N}} , we encode it into an MPS as E_{r_{1}\cdots r_{N}}=\sum_{a_{1}\cdots a_{N}}\left(\prod_{n=1}^{N}A^{[n]}_{r_{n}a_{n}a_{n+1}}\right), (2) with \{\boldsymbol{A}^{[n]}\} the local tensors (see the yellow squares in Fig. 1 ) and \{a_{n}\} the virtual indexes. The complexity of MPS just scales linearly with N as O(2N\chi_{a}^{2}) , with the virtual dimension \chi_{a}=\dim(a_{n}) a hyper-parameter. The unitary transformation \hat{U} is represented as a VQC Cerezo et al. 2021 ; Bharti et al. 2022 . Without losing generality, we choose the “brick-wall” VQC formed by two-body gates (see the blue squares in Fig. 1 ). Its complexity scales as O(NN_{L}) with N_{L} the VQC depth. We introduce third-order super-diagonal tensors, satisfying \delta_{abc}=1 if a=b=c or \delta_{abc}=0 otherwise, to connect the MPS and the VQC (as well as its conjugate) Zhou et al. 2023 . The eigenvalue decomposition of \hat{H} is finally represented as a variational ansatz formed by an MPS, VQC and its conjugate, connected by N third-order super-diagonal tensors. The local tensors of MPS and the parameters of the gates in the VQC form the variational parameters of our TNVD ansatz. We adopt latent tensors as a universal parameterization of the gates Zhou et al. 2021 . The ansatz is updated by minimizing the logarithmic Schmidt distance Vedral and Plenio 1998 F=\log_{2}\left|\hat{H}-\tilde{H}\right|^{2}-N, (3) with \tilde{H} the Hamiltonian given by the ansatz and |\ast| the L2 norm. The variational parameters are

Read the complete text