Correlations, Spectra and Entanglement Transitions in Ensembles of Matrix Product States

Hugo Lóio, Guillaume Cecile, Sarang Gopalakrishnan, Guglielmo Lami, Jacopo De Nardis · 2024

Read Correlations, Spectra and Entanglement Transitions in Ensembles of Matrix Product States by Hugo Lóio, Guillaume Cecile, Sarang Gopalakrishnan, Guglielmo Lami, Jacopo De Nardis (2024) 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-2412-14261.txt.

Introduction

Hugo Lóio and four colleagues examine crowds of quantum states built for easy storage, and find the crowds remember how they were made. Matrix product states compress quantum states from circuits, measurements, and mixtures of both. The team studied the spectra of the transfer matrices behind these states.

Different circuit styles left different fingerprints in those spectra, tied to how information spreads inside the state. When measurements entered the circuits, the ensembles showed a sharp entanglement transition. A simple count, the null space size of the transfer matrix, tracked the change. The work links data structures to deep quantum phase behavior.

Catalog record

CREATORHugo Lóio, Guillaume Cecile, Sarang Gopalakrishnan, Guglielmo Lami, Jacopo De Nardis
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-14261

Browse the full library · About the Tower of Babel

Full text

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

Correlations, Spectra and Entanglement Transitions in Ensembles of Matrix Product States Hugo Lóio, Guillaume Cecile, Sarang Gopalakrishnan, Guglielmo Lami, Jacopo De Nardis Abstract We investigate ensembles of Matrix Product States (MPSs) generated by quantum circuit evolution followed by projection onto MPSs with a fixed bond dimension χ \chi . Specifically, we consider ensembles produced by: (i) random sequential unitary circuits, (ii) random brickwork unitary circuits, and (iii) circuits involving both unitaries and projective measurements. In all cases, we analyze the spectra of the MPS transfer matrices and relate them to the spreading of mutual information in the MPS state. We demonstrate how different features of the spectral density correspond to distinct types of circuits, revealing that these MPS ensembles retain crucial physical information about the underlying microscopic dynamics. Notably, in the presence of quantum monitoring, we show the existence of a measurement-induced entanglement transition (MIPT) in MPS ensembles, with the averaged dimension of the transfer matrix’s null space serving as the effective order parameter. Introduction. — Simulating highly entangled quantum states is one of the central challenges in condensed matter physics, quantum information science, and statistical physics. In one-dimensional systems with bounded entanglement Perez-Garcia et al. 2007 ; Verstraete et al. 2008 ; Verstraete and Cirac 2006 ; Schollwöck 2011 , tensor networks provide a powerful and efficient framework for representing quantum many-body states classically, and also for preparing them efficiently on digital quantum platforms Piroli et al. 2021 ; Malz et al. 2024 ; Smith et al. 2024 ; Stephen and Hart 2024 ; Zhang et al. 2024 . In such cases, the bond dimension \chi , which defines the size of each tensor in the network, can be kept relatively small and independent of the system size. Ground states of local Hamiltonians are typically weakly entangled, but in more general settings (e.g. when describing nonequilibrium dynamics Nahum et al. 2017 ; Rath et al. 2023 ; Calabrese and Cardy 2016 ; Fagotti and Calabrese 2008 ) the bond dimension \chi must scale exponentially with the system size. This exponential growth makes exact classical tensor network representations infeasible. Approximate representations, in which a state is projected onto a manifold of tensor network states with a fixed \chi , have been explored in the literature Leviatan et al. 2017 ; Cecile et al. 2024 ; Begušić et al. 2024 ; Kloss et al. 2018 ; Ayral et al. 2023 , but the extent to which such approaches can fully capture the underlying quantum dynamical process remains unclear. Figure 1: Illustrative overview of this work. We study ensembles of Matrix Product States (MPSs) characterized by a fixed bond dimension \chi , including: (left) random MPSs generated by a staircase quantum circuit; and (right) MPSs produced by quantum circuits composed of random two-qubit gates and, optionally, single-site measurements. In the latter case, the wave function is projected onto the MPS manifold with bond dimension \chi and renormalized after each layer (orange shapes). While exhibiting distinct features, such as specific spectral and correlation structures at finite \chi , we show that the two ensembles share several common properties in the large- \chi limit (with \chi/2^{N}\to 0 ). In this work, we introduce ensembles of matrix product states (MPSs) that, remarkably, despite having a finite bond dimension, can mirror most of the features of quantum dynamics and, in particular, of entanglement spreading and entanglement transitions. We consider ensembles generated by a paradigmatic family of chaotic many-body systems: namely, random quantum circuits Fisher et al. 2023 ; Hoke 2023 , where each two-site gate is randomly selected from the Haar unitary ensemble, see Fig. 1 and where the MPS is projected back on the space of finite bond dimension \chi on each layer. We consider both unitary circuits and hybrid circuits in which projective measurements are introduced at a fixed rate p Li et al. 2019 ; Chan et al. 2018 ; Skinner et al. 2019a ; Vijay et al. 2012 ; Katz et al. 2006 ; Campagne-Ibarcq et al. 2016 ; Bauer et al. 2015 ; Carollo and Alba 2022 . Our main result is that, even if one truncates the dynamics to a space of fixed \chi , the MPS ensembles retain information about the structure of the underlying dynamics. For unitary dynamics, we show that random sequential circuits and random brickwork circuits yield transfer matrices featuring a spectral gap that approaches a universal, \chi -independent value at large \chi , but with different spectral densities (defined as the eigenvalue probability density in the complex plane). It is generally believed that spatial correlations in MPSs are encoded in the transfer matrix spectrum, where the largest eigenvalue is associated with the correlation length \xi Rams et al. 2018 ; Haag et al. 2023 ; Rams et al. 2015 ; Zauner et al. 2015 ; Huang et al. 2024 ; Vanhecke et al. 2019 . Naively, this would suggest that sequential and brickwork circuits, once projected on MPS manifolds, generate the same correlations. We demonstrate that this interpretation is incomplete, as the full spectral density of eigenvalues participates in the spatial correlations. We find that correlations are finite over a region of size \xi_{\mathrm{eff}}\sim\log\chi^{\alpha} where \alpha is distinct for different circuits, reflecting the fact that these circuits have distinct entanglement dynamics. Finally, we show that in the presence of projective measurements, the spectrum of the transfer matrix is itself changed, with the appearance of midgap states for any value of measurement rate, and, mostly importantly, with a finite null space for any value of \chi , whenever the measurement rate exceeds the critical measurement-induced phase transition (MIPT) value. The latter is responsible for the emergence of the area-law phase, where \xi_{\mathrm{eff}}\sim\log\chi^{0} . Our findings elucidate how long-range correlations can also arise from transfer matrices with finite spectral gaps and demonstrate how key properties of truly quantum many-body systems are mirrored in the structure of MPS ensembles with modest values of bond dimensions. Preliminaries: MPSs, transfer matrices and correlation lengths — MPSs are characterized by d matrices of size \chi\times\chi for each qudit, with local dimension d . A generic MPS with bond dimension \chi can therefore be written as \begin{split}|\psi\rangle=\begin{gathered}\includegraphics[width]{mps_paper.pdf}\end{gathered}\end{split}\,. (1) MPSs can be efficiently contracted by choosing an appropriate gauge, characterized by a center of orthogonality. In this gauge, all tensors are either left- or right-isometries, except for the tensor at the orthogonal center Schollwöck 2011 . Expectation values of correlation functions can be expressed in terms of the transfer matrix: \mathcal{T}_{i}=\sum_{\sigma=1}^{d}(A_{i}^{\sigma})^{*}\otimes A_{i}^{\sigma}=\begin{gathered}\includegraphics[width]{transfer_matrix_paper}\end{gathered}\,. (2) In a translationally invariant MPS, A_{i}=A and \mathcal{T}_{i}=\mathcal{T} for all i . For instance, the connected correlation function of a local operator O_{i} between positions i and i+r+1 can be expressed in terms of the left and right eigenvectors, \left\langle\left\langle l_{m}\right|\right. and \left.\left|r_{m}\right\rangle\right\rangle , and the corresponding eigenvalues \lambda_{m} of the transfer matrix \mathcal{T} as Schollwöck 2011 : \langle O_{i}O_{i+r+1}\rangle^{c}=\sum_{m>1}\lambda_{m}^{r}\left\langle\left\langle l_{1}\right|\right.\mathcal{T}^{O}\left.\left|r_{m}\right\rangle\right\rangle\left\langle\left\langle l_{m}\right|\right.\mathcal{T}^{O}\left.\left|r_{1}\right\rangle\right\rangle, (3) where, similarly to Eq. ( 2 ), we define \mathcal{T}^{O}=\sum_{\sigma}O^{\sigma}(A^{\sigma})^{*}\otimes A^{\sigma} . Normalization imposes \lambda_{1}=1 , while all other eigenvalues lie within the unit disk on the complex plane: |\lambda_{m}|<1 , for m>1 . This typically motivates the use of a single leading eigenvalue \lambda_{\rm max}=\underset{\lambda\in\{\lambda_{m},\,m>1\}}{\mathrm{arg\,max}}\,|\lambda| to define the correlation length \xi , thereby reducing the sum

Read the complete text