Quantum avalanches in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain Authors: Lv Zhang, Kai Xu, Heng Fan Published: 2025 ABSTRACT Recent numerical works have revealed the instability of many-body localized (MBL) phase in disordered quantum many-body systems with finite system sizes and over finite timescales. This instability originates from the rare Griffiths regions in the thermodynamic limit, which rapidly thermalize and induce an avalanche mechanism that destabilizes surrounding MBL regions. Here, we consider the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain, whose phase diagram is enriched by the presence of an intermediate ergodic phase separating two distinct MBL phases characterized by different long-range orders. We investigate the dynamic characteristics of the model when coupled to an infinite bath under perturbation, and through the scaling behavior of the slowest thermalization rate, we find how the critical disorder strength in finite-size systems is affected by the avalanche mechanism. We also implement the embedded inclusion model and use the time evolution of mutual information between each spin and the artificial Griffiths region to probe the diffusion of the thermal bubble. We observe that in finite-sized systems, the critical disorder strength gradually drifts away from the center. Our results show that both the MBL paramagnetic phase and MBL spin-glass phase are unstable at finite sizes. Introduction The many-body localized (MBL) phase in disordered, interacting quantum many-body systems1–6 has attracted widespread attention in recent decades, as it violates the eigenstate thermalization hypothesis (ETH)7–9. Anderson formulated the question of the possible absence of thermalization in interacting systems, while explicitly solving the non-interacting limit10. Subsequent works have suggested that systems in the MBL phase can be phenomenologically described by emergent quasi-integrability, where the information of the initial state is retained within local integrals of motion (LIOM’s)5,11–16, also known as “l-bits”. Consequently, the system does not thermalize under its own dynamics. Recent studies have revealed that the violation of the ETH can also occur in disorder-free systems, such as Stark MBL17–22. However, later work has pointed out that the localization phenomenon in Stark MBL stems from the Hilbert space fragmentation23–26, which differs from the mechanism underlying the disordered MBL systems27. Furthermore, a series of works have obtained the MBL phase on quantum simulation platforms28–35. It has now been recognized that for finite-size disordered systems and over finite timescales, there may not exist a stable MBL phase, rather a MBL regime36,37. The results obtained from previous works, derived through exact diagonalization, do not accurately reflect the critical points at which the ergodic-MBL phase transition occurs. Although the disorder strength of a system exceeds the threshold determined by earlier researches, for sufficiently large system sizes and long time, the system may ultimately reaches its thermal equilibrium37,38. Ref39. has suggested that the instability of the MBL phase originates from local, weakly disordered regions, termed Griffiths regions40–43, which emerge in the thermodynamic limit due to statistic theory. These Griffiths regions are in the thermal phase, after rapidly thermalizing within a short period of time, gradually thermalize the surrounding areas, acting as the bath, expanding the range of the thermalization regions and introducing the avalanche mechanism37,44–49. At the same time, another work has discussed an alternative scenario in which localization is destabilized by the resonance proliferation near the transition50. A subsequent work revealed a connection between many-body resonances and the avalanche instability, showing that the avalanche primarily propagates through strong, rare near-resonant eigenstates of the closed system51. Recent numerical works have focused on discussing the propagation of avalanches in the typical MBL phase. One approach to probing the avalanche instability is to treat the remainder of the chain, excluding the Griffiths regions, as an open system, revealing that the critical point of the disordered Heisenberg chain drifts from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^* \approx 8$$\end{document} to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*>20$$\end{document} with increasing system sizes37,46. Another approach involves modeling the bath using the same structure as the rest of the chain but with different disorder strength, placing it in the ergodic phase47,49. The one-dimensional non-interacting Ising Majorana chain is an important model in condensed matter physics, often used to study the Majorana edge states, as an effective model of p-wave superconductors, introduced by Kitaev52. The transition between ordered and disordered localized phases in the noninteracting Ising Majorana model was first proposed by Huse et al53. Over the following two decades, various forms of interaction were introduced into this model54–58. Recently, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain revealed a rich MBL phase diagram58–63. It has been highlighted that this model has two completely different MBL phases, including trivial MBL paramagnetic (PM) phase and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document} symmetry-breaking MBL spin-glass (SG) phase62. A few numerical works have shown that there is an emergent ergodic phase between two MBL phases with different long-range order properties, by calculating conventional landmarks like mean-level spacing ratio and half chain von-Neuman entropy. As the interaction strength increases, the extent of the ergodic phase continues to expand61–63. Previous works have shown that the thermal phase in various MBL systems expands as the system size increases. For the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain, would the Markov method (applying dissipation at one end of the chain) and the non-Markov method (connecting an artificial Griffith region at one end of the chain) yield the same results? Do the drifts of the critical disorder strengths between the thermal phase and the two topologically distinct MBL phases in the model coincide, as expected from the duality? Moreover, in the non-Markov method, can the dynamics of the mutual information between the artificial Griffiths region and an individual spin reveal the destabilization of the MBL phase? In this paper, to address these questions, we calculated the slowest thermalization rate of the system, coupling with an infinite bath, in the dissipationless limit, and obtained the ergodic-MBL phase transition points (for both PM MBL phase and SG MBL phase) with different system sizes37,46. Meanwhile, we artificially set one end of the interacting Ising Majorana chain to be in ergodic phase47, equivalent to a Griffiths region in the thermodynamic limit, and calculated the dynamic evolution of mutual information between each spin in the rest of the chain and the effective bath. Model In this paper, we focus on the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain model with open boundary conditions, which is described by1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_\mathrm{{IM}} = H_\mathrm{{TFI}} + H_\mathrm{{int}}, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_\mathrm{{TFI}}$$\end{document} describes a non-interacting transverse-field Ising (TFI) chain and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_\mathrm{{int}}$$\end{document} introduces the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document} interaction in the context of Majorana fermions chain, 2a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_\mathrm{{TFI}}= & \sum _i h_i \sigma _i^z + \sum _i J_i \sigma _i^x\sigma _{i+1}^x, \end{aligned}$$\end{document}2b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_\mathrm{{int}}= & \sum _i V_i(\sigma _i^x\sigma _{i+2}^x + \sigma _i^z\sigma _{i+1}^z), \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\sigma }_i = \{ \sigma _i^x,\sigma _i^y,\sigma _i^z \}$$\end{document} are Pauli spin-1/2 operators at site \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i=1,2,\dots ,L$$\end{document}. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_i,J_i$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_i$$\end{document} are disordered coupling strengths, they are independent random variables uniformly distributed in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-W_h,+W_h]$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-W_J,+W_J]$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-W_V,+W_V]$$\end{document}, respectively. To simplify, we focus on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_JW_h=1$$\end{document} and fix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_V = 0.7$$\end{document} in this paper. Unlike previously studied MBL models, such as the Floquet random circuit and the random-field Heisenberg chain, which exhibit only a single transition between the thermal and MBL phases37, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain hosts two distinct MBL phases (see Fig. 1). The PM MBL phase is similar to the MBL phases observed in earlier models, while the SG MBL phase features eigenstates with a long range correlation. Before studying the fate of the thermal bubbles in this model, we first review some basic properties of this interacting Ising Majorana chain. This model exhibits a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document} symmetry with the parity operator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {P}=\prod _i \sigma _i^z$$\end{document}, i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[\mathbb {P},H_\mathrm{{IM}}] = 0$$\end{document}. Recently, several works have revealed that this model has two distinct many body localized regions, which are separated by an emergent thermal phase, by analyzing mean-level spacing ratio64, half chain von-Neuman entanglement entropy65 and Edwards-Anderson (EA) order parameter61,66. For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J \gg W_h, W_V$$\end{document}, the Ising interaction term dominates the Hamiltonian and the system is in the SG MBL phase, indicating that the full spectrum is pairwise degenerate63 as the eigenstates are so-called “Cat state”, e.g., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\Psi \rangle = |\leftarrow \rightarrow \rightarrow \dots \rangle \pm |\rightarrow \leftarrow \leftarrow \dots \rangle$$\end{document}. However, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_h \gg W_J, W_V$$\end{document}, the transverse field term plays an important role, the system is in the PM MBL phase, the eigenstates are product states of the basis along the z-direction on each site, e.g., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\Psi \rangle =|\uparrow \uparrow \downarrow \dots \rangle$$\end{document}. Mean-level spacing ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle r \rangle$$\end{document} illustrates two transitions around \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J=0.32$$\end{document} (from PM MBL phase to thermal phase) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J=3.2$$\end{document} (from thermal phase to SG MBL phase) via the finite size scaling61. In two MBL phases, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle r\rangle \approx 0.39$$\end{document} corresponds to the Poisson statistics64, while in thermal phase, the energy spectrum is related to the GOE statistics with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle r\rangle \approx 0.53$$\end{document}. Unlike the mean-level spacing ratio, half chain von-Neuman entropy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{L/2}$$\end{document} and EA order parameter can easily distinguish two MBL phases. Although \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{L/2}$$\end{document} exhibits area law both in two different MBL phases, in PM MBL phase, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{L/2}$$\end{document} is close to 0 and in SG MBL phase, it leads to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ln 2$$\end{document} because the special “Cat state” structure. In both localized phases \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{L/2}$$\end{document} is independent of system size L but in the middle thermal phase (for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J/W_h\approx 1$$\end{document}) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{L/2}$$\end{document} increases with the system size gradually approaching the page value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{page}(L,L/2) = (L/2)\ln 2 -1/2$$\end{document}61,67. EA order parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi$$\end{document} is used to distinguish the distinct MBL phase by calculating long range Ising correlation. In SG MBL phase, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi$$\end{document} shows the volume law as system size increase while deeply in PM MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\ll 0.32$$\end{document}) it approaches a constant61.Fig. 1Schematic phase diagram of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain with fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_V=0.7$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_JW_h=1$$\end{document}. A thermal phase (green region) emerges between the PM MBL phase (blue region) and the SG MBL phase (red region). The two yellow diamonds represent the two phase transition points, obtained from the mean level spacing ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle r\rangle$$\end{document}, located at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 0.32$$\end{document} (from the PM MBL phase to the thermal phase) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 3.2$$\end{document} (from the thermal phase to the SG MBL phase). Two sets of circles, ranging from light gray to black, show the phase transition points obtained from the Markovian method, gradually drifting away from the center (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 1$$\end{document}) as the system size increases. Schematic phase diagram of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain with fixed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_V=0.7$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_JW_h=1$$\end{document}. A thermal phase (green region) emerges between the PM MBL phase (blue region) and the SG MBL phase (red region). The two yellow diamonds represent the two phase transition points, obtained from the mean level spacing ratio \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle r\rangle$$\end{document}, located at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 0.32$$\end{document} (from the PM MBL phase to the thermal phase) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 3.2$$\end{document} (from the thermal phase to the SG MBL phase). Two sets of circles, ranging from light gray to black, show the phase transition points obtained from the Markovian method, gradually drifting away from the center (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J\approx 1$$\end{document}) as the system size increases. Methods Avalanche instability theory In this subsection, we briefly review the avalanche instability theory37,39 in disordered quantum MBL systems. The theory postulates that the MBL systems can be destabilized by a few emergent thermal inclusions, so called Griffiths regions, in the thermodynamic limit. These Griffiths regions thermalize rapidly, regarded as thermal bubbles, and destabilize the surrounding typical MBL chains by serving as heat reservoirs. The fate of a thermal bubble is corresponded to its thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma$$\end{document}39,68. Consider a Griffiths region of length N within a sufficiently long one-dimensional spin chain, when this thermal bubble extends over L spins in each direction, the typical level spacing of the enlarged thermal inclusion is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{-(N+2L)}$$\end{document} (as it has thermalized \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N+2L$$\end{document} spins in total). If the typical level spacing is smaller than the thermalization rate, the spin would not recognize the discreteness of the spectrum, i.e., the thermal bubble serves as a bath effectively. So the avalanche will not stop until the thermalization rate and the typical level spacing have similar energies, indicating that the scaling behavior of the thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma \sim g^{-L}$$\end{document} is important. According to the typical level spacing we have discussed before, the point where the transition occurs is at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g=4$$\end{document}. The probability of obtaining a Griffiths region decreases exponetially with its length N:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} P(N)\sim \mathscr {O}(e^{-N}), \end{aligned}$$\end{document}leading us to a reasonable assumption that there is no direct interaction between any two Griffiths regions. Despite the fact that the entire long chain constitutes an isolated system, after the rapid thermalization of the Griffiths region, we can focus solely on the typical MBL region and regard the Griffiths region as an infinite bath which connects at one end of the MBL chain. In this case, the dynamic features of the MBL chain are governed by the Master equation. The non-steady eigenmodes of the Master equation can reflect the effect of the Griffiths region on the MBL chain, and the thermalization rate can be estimated from the slowest relaxation mode (the eigenvalue with the second-largest real part). In addition, we directly simulate the time evolution of the full chain using the embedded thermal inclusion model47, in which a short spin chain in the thermal phase is artificially attached to one end of a finite MBL chain. The diffusion of the thermal bubble is then probed through the mutual information between each spin in the MBL chain and the effective bath (the artificial Griffiths region). Markov process To study avalanche instability, we assume that the system couples with an infinite bath. The Master equationcan then describe the dynamics of the system,4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \partial _t \rho = \mathscr {L}\rho , \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {L}$$\end{document} is the Liouvillian superoperator,5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {L}\rho = - i[H_\mathrm{{IM}},\rho ] + \gamma \sum _{\mu } \big ( L_{\mu }\rho L_{\mu }^{\dagger } -\frac{1}{2}\{ L_{\mu }^{\dagger }L_{\mu },\rho \} \big ), \end{aligned}$$\end{document}here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{L_{\mu }\}=\{\sigma ^x_1,\sigma ^y_1,\sigma ^z_1\}$$\end{document} are Lindblad operators, which describe the couplings between the Ising Majorana chain and the infinite bath. While \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document} denotes the coupling strength. We can obtain all the modes by solving for eigenoperators from, 6a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {L}\rho ^R_i= & \lambda _i \rho ^R_i, \end{aligned}$$\end{document}6b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {L}^{\dagger }\rho ^L_i= & \lambda _i^* \rho ^L_i, \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\rho ^L_i\}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\rho ^R_i\}$$\end{document} are left and right eigenmodes, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda _i$$\end{document} is the ith eigenvalue. Thus the time dependent solution of equation (4) can be written as the sum of each right eigenmode,7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \rho (t) = \rho ^R_0+\sum _{i>0} e^{\lambda _i t}p_i\rho _i^R, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_i$$\end{document} is the overlap between the initial state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho (0)=|\phi _0\rangle \langle \phi _0|$$\end{document} and the i\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{th}$$\end{document} left eigenmode i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_i= \textrm{Tr}[\rho (0)\rho ^L_i]$$\end{document}. In addition to the steady mode \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _0=\mathbb {I}/D$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda _0=0$$\end{document}, thermalization rate of other modes is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma \equiv -\textrm{Re}[\lambda ]>0$$\end{document}. So it’s natural to get slowest thermalization rate by diagonalizing the Liouvillian superoperator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {L}$$\end{document}, but unfortunately, it requires a huge amount of computing resources to obtain the entire exact spectrum of a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4^L\times 4^L$$\end{document} matrix, which we cannot afford even for a slightly larger system size. However, if the couplings between the infinite bath and the chain are limited (in the dissipationless limit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma \ll 1$$\end{document}), we can reduce the problem to the subspace \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {P}$$\end{document} spanned by the eigen density matrix of the original Hamiltonian equation (1) using perturbation theory46. The subspace reads,8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {P} = \bigg \{|n\rangle \langle n|\ \bigg |\ H_\mathrm{{IM}}|n\rangle = E_n|n\rangle \bigg \}, \end{aligned}$$\end{document}now we can estimate the slowest thermalization rate by exactly diagonalizing the reduced matrix (a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^L\times 2^L$$\end{document} non-sparse matrix) \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {S}$$\end{document}. The elements of the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {S}$$\end{document} are9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {S}_{nm} = \langle n |\mathscr {L}[|m\rangle \langle m|]|n\rangle = -3\gamma \delta _{nm} + \gamma \sum _\mu |\langle n|L_{\mu }|m\rangle |^2, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|n\rangle$$\end{document} is the nth eigenstate of the original interacting Ising Majorana chain Hamiltonian equation (1) and the slowest thermalization rate we need is the smallest non-zero eigenvalue of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {S}$$\end{document}. Remember that the scaling behavior of the slowest thermalization rate is the vital ingredient in avalanche instability, so it’s reasonable to add a scaling factor to the slowest thermalization rate. Although the bath is attached at the edge and the avalanche spreads directionally, in our assumption the thermal bubble originates as a statistical phenomenon in the thermodynamic limit and thus acts on both sides of the spin chain, which justifies the use of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4^L$$\end{document}. Therefore, what we really need is the rescaled thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\Gamma } = \Gamma 4^L$$\end{document}. Non-Markov process We also use the embedded inclusion model47,49, as shown in Fig. 2 to probe quantum avalanches, the system is divided into two parts: the left part (region A) is the interacting Ising Majorana chain in thermal phase with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{AJ}\equiv 1$$\end{document}, which behaves like a bath (Griffiths region) in dynamic evolution. The right part (region B) is the Ising Majorana chain in the phase of interest, determined by varying the disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J$$\end{document}. To obtain how the thermal bubble spreads in the region B, we calculate the mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {I}$$\end{document} at time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau$$\end{document} between the region A (effective bath) and each spin in the region B, described as follows10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {I}_i(\tau ) = S_{A}(\tau ) + S_i(\tau ) - S_{A[NONSPACE] \& i}(\tau ), \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{\Omega }= - \textrm{Tr}[\rho _{\Omega }\ln \rho _{\Omega }]$$\end{document} is the von-Neuman entropy of the subsystem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega$$\end{document}. Here we introduce the initial state11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} |\Psi \rangle = |\Psi _\mathrm{{Harr}}\rangle _A \otimes |\textrm{Neel}\rangle _B, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\Psi _\mathrm{{Harr}}\rangle _A$$\end{document} is a random state, sampled from the Harr ensemble69, on region A. While \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\textrm{Neel}\rangle _B$$\end{document} denotes the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\uparrow \downarrow \uparrow \downarrow \dots \downarrow \rangle _B$$\end{document} when we focus on PM MBL phase and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\rightarrow \leftarrow \rightarrow \leftarrow \dots \leftarrow \rangle _B$$\end{document} for SG MBL phase on region B.Fig. 2Diagram depicting the setting of the embedded inclusion model. We artificially connect a Griffiths region (the left part), which is in the thermal phase i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JA}\equiv 1$$\end{document}, to the one end of a typical MBL Ising Majorana chain (the right part). The initial states are prepared apart: the initial states in region A are sampled from the Harr ensemble, while in region B, the initial states are selected from two Neel states according to which MBL phase is focused. Then we time evolve the whole system by implementing the Krylov subspace algorithm. We calculate the mutual information between each spin in region B and the total region A. Diagram depicting the setting of the embedded inclusion model. We artificially connect a Griffiths region (the left part), which is in the thermal phase i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JA}\equiv 1$$\end{document}, to the one end of a typical MBL Ising Majorana chain (the right part). The initial states are prepared apart: the initial states in region A are sampled from the Harr ensemble, while in region B, the initial states are selected from two Neel states according to which MBL phase is focused. Then we time evolve the whole system by implementing the Krylov subspace algorithm. We calculate the mutual information between each spin in region B and the total region A. Results Avalanche with Markovian dynamics Fig. 3Rescaled thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma 4^L$$\end{document} as a function of the disorder parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h$$\end{document}. Points are calculated by rescaling the 80th percentile of the distribution of the smallest non-zero value of the matrix defined by equation (9) with a factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4^L$$\end{document}. Curves are fitted by quadratic fitting. Different curves show different system sizes from 6 to 12. Left panel: PM MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h < -2.4$$\end{document}), the rescaled thermalization rate increases with the disorder parameter. Right panel: SG MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h> 2.0$$\end{document}), the rescaled thermalization rate decreases with the disorder parameter. Crossing points of adjacent curves indicate the avalanche instability of different system sizes. Crossing point drifts away from the middle with the system size increasing in both MBL phases. Rescaled thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma 4^L$$\end{document} as a function of the disorder parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h$$\end{document}. Points are calculated by rescaling the 80th percentile of the distribution of the smallest non-zero value of the matrix defined by equation (9) with a factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4^L$$\end{document}. Curves are fitted by quadratic fitting. Different curves show different system sizes from 6 to 12. Left panel: PM MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h < -2.4$$\end{document}), the rescaled thermalization rate increases with the disorder parameter. Right panel: SG MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h> 2.0$$\end{document}), the rescaled thermalization rate decreases with the disorder parameter. Crossing points of adjacent curves indicate the avalanche instability of different system sizes. Crossing point drifts away from the middle with the system size increasing in both MBL phases. We calculate the rescaled thermalization rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Gamma 4^L$$\end{document} with multiple system sizes from 6 to 12 with at least 1200 disorder realizations, and we choose the 80th percentile of the distribution of results for each disorder parameter and system size. The results are shown in Fig. 3.Fig. 4The crossing points of the rescaled thermalization rate for adjacent system sizes in Fig. 3. The results of the PM MBL phase and SG MBL phase are displayed in one figure by taking the absolute value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h$$\end{document}. The behavior of the results changs with the system size increasing, fitted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\zeta \lg W_{JC} = \alpha \lg L +\beta$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta =1$$\end{document} for SG MBL phase and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta = - 1$$\end{document} for PM MBL phase, showing the instability of the localization both in two MBL phases. The crossing points of the rescaled thermalization rate for adjacent system sizes in Fig. 3. The results of the PM MBL phase and SG MBL phase are displayed in one figure by taking the absolute value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h$$\end{document}. The behavior of the results changs with the system size increasing, fitted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\zeta \lg W_{JC} = \alpha \lg L +\beta$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta =1$$\end{document} for SG MBL phase and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta = - 1$$\end{document} for PM MBL phase, showing the instability of the localization both in two MBL phases. In Fig. 3, it illustrates that the rescaled rate increases (decreases) with the disorder parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J/W_h$$\end{document} in PM MBL phase (SG MBL phase). The solid lines in Fig. 3 are calculated by quadratic fitting for every system size with different disorder strength. Generally, we need to perform a scaling collapse of these rescaled slowest thermalization rate with different system sizes and fix disorder strength to determine the location of the phase transition. However, we don’t know the exact form of the scaling function, so we use the crossing point of the two curves in Fig. 3 that the sizes of the system are adjacent to estimate the position, where the avalanche-driven transition occurs, in finite size system37. Although the features of the SG MBL phase, such as the existence of long range correlation in its eigenstates, differ from those of the PM MBL phase or the MBL phases in previously studied models (such as Floquet random circuit and random-field Heisenberg chain), the drift behavior of the critical disorder strength is similar. The crossing point drifts towards different directions with system size increasing, moving away from the middle, i.e., the thermalization phase diagnosed by conventional landmarks. Because of the limited acceptable system size to numerics, we can only obtain the crossing point of rescaled thermalization rate for small size system. However, we can provide an upper bound (lower bound) in SG MBL phase (PM MBL phase) in the thermodynamic limit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L\rightarrow \infty$$\end{document}, if we assume that the crossing point monotonically drifts with the system size L. In the PM MBL phase roughly at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h < -3.05$$\end{document}, while in the SG MBL phase \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lg W_J/W_h> 3.07$$\end{document}. To show more details how the crossing point drifts with the system size, we display the crossing point \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2|\lg W_{JC}|$$\end{document} as a function of the system size L for different MBL phases, as shown in Fig. 4. We use \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2|\lg W_J|=\alpha \lg L + \beta$$\end{document} to fit all results. Although the drift of the critical disorder strength does not strictly follow this scaling behavior at large sizes, within the computationally accessible range of system sizes from 7 to 12, the fitted parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha _\mathrm{{PM}}$$\end{document}=2.0706 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha _\mathrm{{SG}}=3.5313$$\end{document}) indicates that the critical disorder strength between the SG MBL phase and the thermal phase is less stable than that between the PM MBL phase and the thermal phase. Using other fitting functions would not alter our conclusion. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain exhibits the Wannier-Kramers70 duality transformation(\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _i^x\sigma _{i+1}^x,\sigma _i^z\Leftrightarrow \tau _i^z,\tau _i^x\tau _{i+1}^x$$\end{document}) that allows the SG MBL phase and the PM MBL phase to correspond to each other12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tilde{H}=\sum _i h_i \tau _i^x\tau _{i+1}^x + J_i\tau _i^z +V_i(\tau _i^x\tau _{i+2}^x+\tau _i^z\tau _{i+1}^z), \end{aligned}$$\end{document}suggesting the emergence of a symmetric critical disorder \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^*_{\alpha } \Leftrightarrow (W^*_{\alpha })^{-1}(\alpha =J, h)$$\end{document}, in the perspective of the spectrum. However, the completely different long-range order properties of its eigenstates in different MBL phases will result in distinct dynamic behaviors when the chain is connected to an infinite bath. When the chain is deeply in PM MBL phase, the eigenstates are product states of the z-direction basis at each spin, which do not exhibit long-range correlations. Therefore, the thermalization of neighboring spins by the bath in the early stage does not affect the spins at the far end. However, when the chain is in deep SG MBL phase, the eigenstates are “Cat-state” along x-direction, which possess a strong long-range correlation. In this case, the thermalization of nearby spins by the bath in the early stage also affects the spins at the far end. This suggests that as we increase the size of the chain, the SG MBL phase requires a larger amount of disorder, compared to the PM MBL phase, to protect this long-range structure from being disrupted, indicating that its critical disorder is more sensitive to the size of system. Avalanche with non-Markovian dynamics Fig. 5Mutual information evolution of embedded inclusion model, with region B size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=14$$\end{document}, for several disorder strength. Top panel: results are calculated in PM MBL phase for disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1,\ 0.32,\ 0.2,\ 0.128,\ 0.1$$\end{document} from left to right with the initial state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\uparrow \downarrow \uparrow \downarrow \dots \downarrow \rangle$$\end{document} in region B. Bottom panel: results are calculated in SG MBL phase for disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1,\ 3.2,\ 5,\ 7.2,\ 10$$\end{document} from left to right with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\rightarrow \leftarrow \rightarrow \leftarrow \dots \leftarrow \rangle$$\end{document} in region B. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1$$\end{document}, the whole system is in the ergodic phase, the results displayed in two figures on the leftmost side illustrate the rapid thermalization in region B as the mutual information descent swiftly after it reach its maximal value in the beginning. However, as the region B gradually dives deeply into the MBL phase, the rate of decline in mutual information slows down, until the region B reaches a sufficiently high disorder strength, the mutual information would remain unchanged after a period of time evolution, indicating that the propagation of the thermal bubble fails to spread through the whole region B. The distinctly different dynamical features of mutual information show the existence of an ergodic-MBL phase transition for the system. Mutual information evolution of embedded inclusion model, with region B size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=14$$\end{document}, for several disorder strength. Top panel: results are calculated in PM MBL phase for disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1,\ 0.32,\ 0.2,\ 0.128,\ 0.1$$\end{document} from left to right with the initial state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\uparrow \downarrow \uparrow \downarrow \dots \downarrow \rangle$$\end{document} in region B. Bottom panel: results are calculated in SG MBL phase for disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1,\ 3.2,\ 5,\ 7.2,\ 10$$\end{document} from left to right with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|\rightarrow \leftarrow \rightarrow \leftarrow \dots \leftarrow \rangle$$\end{document} in region B. When \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=1$$\end{document}, the whole system is in the ergodic phase, the results displayed in two figures on the leftmost side illustrate the rapid thermalization in region B as the mutual information descent swiftly after it reach its maximal value in the beginning. However, as the region B gradually dives deeply into the MBL phase, the rate of decline in mutual information slows down, until the region B reaches a sufficiently high disorder strength, the mutual information would remain unchanged after a period of time evolution, indicating that the propagation of the thermal bubble fails to spread through the whole region B. The distinctly different dynamical features of mutual information show the existence of an ergodic-MBL phase transition for the system. In the study of the embedded inclusion model, we calculate various system sizes from 8 to 14 and 5 different disorder strengths for each MBL phase. The region B, with fixed length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A=4$$\end{document}, which is chosen to behave like a Griffiths region in dynamic evolution as we set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{AJ}\equiv 1$$\end{document}. For each system size, we performed multiple disorder realizations: 2000, 1500, 1000 and 500 samples for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L=$$\end{document}8, 10, 12 and 14, respectively. We implement the Krylov subspace algorithm71,72 to time evolve the whole model with a time step \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau =0.5$$\end{document}. The results of mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {I}$$\end{document} with system size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=14$$\end{document} are displayed in Fig. 5. Results of both PM MBL phase and SG MBL phase manifest the similar phenomenon with disorder strength increasing. The mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {I}(A:B)$$\end{document} between subsystems A and B is13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {I}(A:B) = S(A) + S(B) - S(A\cup B), \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S(\Omega )=\textrm{Tr}(\rho _\Omega \textrm{ln}\rho _\Omega )$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho _\Omega = \textrm{Tr}_{\Omega ^c}(\rho )$$\end{document} are von-Neuman entanglement entropy and reduced density matrix of the subsystem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega$$\end{document}. In pure state, it equals to conditional mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathscr {I}(A:B|C)=S(A\cup C) + S(B\cup C)-S(A\cup B\cup C)-S(C)$$\end{document}, which reflects the direct entanglement between subsystems A and B. We now consider a tripartite system consisting of three subsystems \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A\otimes \Omega \otimes R$$\end{document} with sizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A,\ 1$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_R$$\end{document}, respectively. Suppose the system is in a Haar-random state. Fixing the subsystem size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A$$\end{document}, we focus on the mutual information between A and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega$$\end{document}. The mutual information exhibits a scaling behavior as a function of the size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_R$$\end{document}:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {I}(A:\Omega ) \sim {\left\{ \begin{array}{ll} \mathscr {O}(1), & L_R\lesssim L_A+1,\\ 2^{-L_R}, & L_R\gtrsim L_A+1. \end{array}\right. } \end{aligned}$$\end{document}Here, we focus on the mutual information between the bath (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A=4$$\end{document}) and a single spin (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{\Omega }=1$$\end{document}). Since the system Hamiltonian contains at most next-nearest-neighbor interactions, the influence of the bath propagates from the coupled edge toward the other end of the chain. What we are interested in is whether the effect of the bath can eventually spread throughout the entire chain. As time evolves, the length of the thermalized region, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{\Omega }+L_{R}$$\end{document}, gradually increases, and the thermalized part together with the bath can be approximately regarded as a Haar-random state. Therefore, as thermalization proceeds, the dynamics of the mutual information between the bath and the first spin exhibits the characteristic behavior described in equation (14), namely an initial growth followed by a subsequent decay to zero. However, if the bath is only capable of thermalizing a finite portion of the chain, the mutual information will eventually saturate at a finite plateau. The peak of the mutual information (if existing) is occured at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_R\sim L_A+L_{\Omega }$$\end{document}, which captures the length of the thermalized spins of the system \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{\Omega }+L_R$$\end{document}. For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{BJ}=1$$\end{document}, the whole chain is in the thermal phase, the mutual information of first several sites in the system (region B) tend to grow in the beginning, and shrink as the evolution time gets longer. However, mutual information of each site eventually converges to the same value, which is almost zero, after evolving for a sufficiently long time. Eventually converged mutual information reveals that the system and the bath are fully coupled and reach the thermalization. While for region B deep in MBL phase (the rightmost column in Fig. 5, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{BJ}=10$$\end{document} for SG MBL phase and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{BJ}=0.1$$\end{document} for PM MBL phase), after a brief increase of the mutual information at the first few spins, the mutual information between each spin and the bath (region A) remains unchanged, indicating that only a limited number of spins are thermalized, i.e. the thermal bubble cannot spread to the whole system. According to the description above, we believe that this result can provide an upper bound (lower bound) for the critical point of disorder strength in SG MBL phase (PM MBL phase), specifically for the system size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=14$$\end{document}, which influences the stability of the avalanche, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^\mathrm{{crit},SG}_{BJ}\le 10$$\end{document} (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W^\mathrm{{crit},PM}_{BJ}\ge 0.1$$\end{document}).Fig. 6Mutual information evolution of embedded inclusion model, with the same disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=0.2$$\end{document} (deeply in the PM MBL phase according to convention landmarks), for different region B sizes. The region B sizes are chosen to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=8,10,12,14$$\end{document} from left to right. For the smallest size of region B, the mutual information between the first spin and the Griffiths region stay the same after a period of time, while as the size of the region B increases, the mutual information tends to descent more and more rapidly, revealing the avalanche instability in the PM MBL phase. Mutual information evolution of embedded inclusion model, with the same disorder strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=0.2$$\end{document} (deeply in the PM MBL phase according to convention landmarks), for different region B sizes. The region B sizes are chosen to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=8,10,12,14$$\end{document} from left to right. For the smallest size of region B, the mutual information between the first spin and the Griffiths region stay the same after a period of time, while as the size of the region B increases, the mutual information tends to descent more and more rapidly, revealing the avalanche instability in the PM MBL phase. Another question of interest is how the size of system influences the avalanche instability. To probe this, we display system sizes from 8 to 14, as shown in Fig. 6, with the system is deep in PM MBL phase (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_{JB}=0.2$$\end{document}) according to previous works. It illustrates that for the small size of the system (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=8$$\end{document}), after \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t>125$$\end{document}, the mutual information of each spin and the bath remains a constant, that is, the effect of the bath to the system is limited and the thermal bubble cannot spread to the entire system. However, for the lager system (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_B=14$$\end{document}), the mutual information between the first spin and the bath reaches its maximum value at t=10 and then decreases rapidly, implying that the range of the avalanche of the system exceeds that of the shorter size of the system. It obviously reveals that the stability of the avalanche is not robust to the size of the system. The critical point of the disorder, which destabilizes the localization of the system, tends to drift away from the center (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J=1$$\end{document}) as system size increases. Discussion In this paper, we have investigated the impact of the avalanche instability, originating from Griffiths regions in the thermodynamic limit, on the one-dimensional \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain using two complementary approaches. We first treat the Griffiths region as an effective infinite bath after a short time evolution, thus the rest of the chain is an open system and its dynamics is governed by the Master equation. In the dissipationless limit, we have extracted the slowest thermalization rate, which is related to the thermalization rate that stems from the Griffiths regions. Our results are shown in Fig. 3, where for each system size, the rescaled thermalization rate decreases as the system penetrates deeper into the MBL phase. The crossing points between the curves of adjacent system sizes serve as estimators of the critical points. The data were fitted with a quadratic form, and the crossing points of different MBL phases and different system sizes are shown in Fig. 4. We performed a linear fit between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{lg}L$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2|\textrm{lg}W_{JC}|$$\end{document}. What we can determine is that, in the range of system sizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7\le L\le 12$$\end{document}, the critical disorder strength between the thermal phase and the two MBL phases drifts away from the center (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J=W_h=1$$\end{document}) and the localization of the SG MBL phase is more unstable compared to that of the PM MBL phase, as its slope \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document} is larger. However, it remains unclear whether this drift continues to follow the same scaling behavior at larger system sizes, and thus we cannot determine whether a stable MBL phase persists in the thermodynamic limit. Although our data are more consistent with the scenario in which the critical disorder strength ultimately does not converge, we cannot exclude the possibility that stable MBL phases may still exist at larger sizes. We have also directly studied the dynamical properties of the original model, which is additionally connected to an artificial Griffiths region. The mutual information equation (10) between each spin outside the Griffiths region and the effective bath is used to probe the entanglement and the spread of avalanches. The results shown in Fig. 5 reveal that in a low-disorder system, thermalization gradually propagates from the end connected to the artificial Griffiths region to the opposite end. However, in a highly disordered system (i.e., away from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_J=1$$\end{document}), the spread of thermalization halts at a specific location. These phenomena clearly indicate the existence of an ergodic-to-MBL phase transition. We can roughly estimate the time at which the mutual information between the spin closest to the Griffiths region and the effective bath begins to decline as a signature when thermalization propagates through at least \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A+c$$\end{document} spins. Fig. 6, showing the time evolution of mutual information for different system sizes at a fixed disorder strength, illustrates that the MBL PM phase is destabilized by avalanches, and that larger disorder is required to preserve localization as the system size increases. These conclusions are in agreement with our previous analysis of the scaling behavior of the slowest thermlization rate when the system is coupled to an infinite bath. Our results indicate that the critical points of ergodic-MBL (both PM MBL phase and SG MBL phase) phase transition are not stable and these boundaries would drift away from the center for small system sizes, where the system is in ergodic phase, showing the similar behavior to different models from previous works37,46. In previous work63, the critical disorder was estimated by computing the typical localization length \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi _b$$\end{document} from the end-to-end correlations of the system’s eigenstates and comparing it with the avalanche criterion39 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\xi ^* = (\ln 2)^{-1}$$\end{document}. In addition, other ED-based indicators, such as the area to volume-law entanglement transition and the Poisson–GOE change in the parity-resolved spectral statistics, yield a consistent critical value. While the typical localization length analysis incorporates avalanche-based reasoning, all these diagnostics are nevertheless based on static spectral properties of a closed system. By contrast, our approach, whether through a boundary Lindblad bath or via coupling to an artificial Griffiths region, is intrinsically dynamical. An apparent discrepancy is observed between our results and the previously reported ED results63 for closed systems. We also note that a similar discrepancy between static ED diagnostics (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W\simeq 3.1$$\end{document}) and Lindblad-based dynamical signatures (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W>7$$\end{document}) has been reported previously in standard random-field Heisenberg model37. These observations lead us to reasonably conjecture that the discrepancy originates from the intrinsic difference between static ED probes and open-system dynamical approaches. However, recent work73 has shown that in non-interacting Anderson and Aubry-André-Harper (AAH) models, the critical disorder strength obtained via coupling to a thermal bath exhibits non-monotonic behavior with respect to system size. This suggests that for the interacting many-body system studied here, although the critical disorder exhibits monotonic behavior within computationally accessible system sizes, the employed Lindblad bath method may lead to an overestimation of the critical disorder strength in the thermodynamic limit. Consequently, our work can deepen the understanding of the effects of avalanche in one-dimensional \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain. Additionally, the study of the dynamic features can facilitate the design of the experiments to directly obtain the avalanches and thermal bubbles that spread in disordered quantum many-body systems. Jordan-Wigner transformation and Ising-Majorana chain In this appendix, we show the details of the Majorana fermions representation. We firstly invite Jordan-Wigner transformation to convert the original spins problem into the Dirac fermions problem, the Jordan-Wigner transformation reads, 15a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} c_i^{\dagger }&= \bigg (\prod _{k=1}^{i-1}\sigma _k^z \bigg )\sigma _i^-, \end{aligned}$$\end{document}15b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} c_i&= \bigg (\prod _{k=1}^{i-1}\sigma _k^z \bigg )\sigma _i^+, \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _i^+=\frac{1}{2}(\sigma _i^x + i\sigma _i^y)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _i^-=\frac{1}{2}(\sigma _i^x - i\sigma _i^y)$$\end{document}, one can quickly derive \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _i^z = 1 - 2c^{\dagger }_i c_i=1-2n_i$$\end{document}. Thus the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbb {Z}_2$$\end{document}-preserving interacting Ising Majorana chain Hamiltonian equation (1) can be written as,16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} H_\mathrm{{IM}} =\ &\sum _i h_i (1 - 2n_i) \\ &+ \sum _i J_i (c_i^{\dagger } - c_i) (c_{i+1}^{\dagger }+c_{i+1}) \\ &+ \sum _i V_i(1-2n_i)(1-2n_{i+1}) \\ &+ \sum _i V_i (c_i^{\dagger } - c_i)(1-2n_{i+1})(c_{i+2}^{\dagger }+c_{i+2}) \end{aligned} \end{aligned}$$\end{document}If we rewrite each fermions as two Majorana fermions, 17a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \gamma _{2i-1}= & c^{\dagger }_i + c_i, \end{aligned}$$\end{document}17b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \gamma _{2i}= & i(c^{\dagger }_i - c_i), \end{aligned}$$\end{document} then the Hamiltonian turns out to be, 18a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_\mathrm{{TFI}}= & -i\sum _l t_l \gamma _l \gamma _{l+1} \end{aligned}$$\end{document}18b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_\mathrm{{int}}= & - \sum _l g_l \gamma _l \gamma _{l+1} \gamma _{l+2} \gamma _{l+3} \end{aligned}$$\end{document} here \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{2i-1}=h_i,t_{2i}=J_i$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$g_{2i-1}=g_{2i}=V_i$$\end{document}. The first term equation (18a) is a standard quadratic form, thus the Hamiltonian of TFI part corresponds to a free Majorana fermions system. The second term equation (18b) introduces interactions between different free modes. Mutual information in Harr ensemble Fig. 7The mutual information between two subsystems of lengths 1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A$$\end{document}, selected from the Harr ensemble, with the horizontal axis representing the size of the removed system (system B). After the size of system B exceeds \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A+1$$\end{document}, the mutual information exhibits an exponential decay. The mutual information between two subsystems of lengths 1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A$$\end{document}, selected from the Harr ensemble, with the horizontal axis representing the size of the removed system (system B). After the size of system B exceeds \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_A+1$$\end{document}, the mutual information exhibits an exponential decay. Consider a Harr ensemble containing N spins of 1/2. If the size of a subsystem is l, then its Von-Neuman entanglement entropy is,19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S(l|N) = n\ln 2 - \frac{1}{2^{N-2n+1}}, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=\min \{l,N-l\}$$\end{document}. Therefore, the mutual information between the two subsystems of sizes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_1$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_2$$\end{document} in a three-part system \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Omega =A_1\otimes A_2\otimes B$$\end{document} can be written as20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {I} = S(L_1|N) + S(L_2|N) - S(L_1+L_2|N). \end{aligned}$$\end{document}We numerically simulate the three-part system with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_1=L_A$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_2=1$$\end{document}, sampled from the Harr ensemble, and the results of the mutual information are displayed in Fig. 7. When the size of the removed subsystem B is smaller than the total size of the two subsystems \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_1$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_2$$\end{document}, the mutual information decays at a polynomial rate; however, when the size of the subsystem B exceeds half of the total system size, the mutual information exhibits an exponential decay,21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathscr {I} \propto 2^{-L_B} \ \ \ \textrm{when} \ \ \ L_B> L_A + 1. \end{aligned}$$\end{document}Thus, when we use mutual information to probe the diffusion of the thermal bubble in the embedded inclusion model, the initial rapid increase in mutual information between the spins close to the Griffiths region and the bath indicates thermal diffusion. After a sufficiently long time, the convergence of mutual information to a value close to zero shows that the system has been fully thermalized by the Griffiths region, even if it is in a typical MBL phase. If the mutual information remains constant, it indicates that the system has not been fully thermalized, and the thermal bubble can only propgate a limited length.