Topology and Spectrum in Measurement-Induced Phase Transitions

Topology is undoubtedly one of the most essential concepts in understanding stable phases of matter [1]. For ground states of local Hamiltonians with a finite excitation gap, states belonging to distinct topological phases cannot be smoothly deformed to each other without closing the gap, owing to discrete characters of their topological invariants [2]. In particular, for symmetry-protected topological phases [3, 4, 5, 6, 7, 8], nontrivial topology of the bulk leads to gapless boundary states robust against any symmetry-preserving perturbations. This phenomenon, known as the bulk-edge correspondence, has remarkable consequences in physical properties of the topological phases and has been intensively studied in the past few decades [9, 10, 11, 12, 13].

Despite notable successes in equilibrium systems, topology in out-of-equilibrium systems caused by external environment has yet to be fully understood. In particular, it is still elusive how the topology plays a role in monitored quantum systems, which gather a recent extensive interest as they host novel non-equilibrium phases concerning, e.g., entanglement [14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]. Indeed, repetitive measurements stabilizing distinct topological phases can lead to phase transitions between different entanglement phases, which can be probed by topological entanglement entropy or purification dynamics [26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42] (see also [43, 44, 45]). However, intrinsic spatiotemporal randomness caused by the probabilistic nature of measurement outcomes makes it difficult to generalize more standard notions of topological phases, such as bulk topological invariants and gapless edge modes, to monitored systems.

In this Letter, we discover gapless edge modes and a bulk topological invariant that characterize measurement-induced phase transitions of monitored Majorana circuits and discuss their bulk-edge correspondence. Introducing effective Hamiltonians obtained through the Lyapunov analysis, we reveal the presence (absence) of edge-localized zero modes inside the bulk gap in the topological (trivial) area-law phase. We also construct a bulk topological invariant based on the fermion parity, extending the method in equilibrium systems to the monitored setting; this is accomplished by introducing a unique boundary condition with twisted measurement outcomes. The invariant sharply distinguishes these gapped area-law phases, featuring the bulk-edge correspondence. Moreover, we show that the critical entanglement phase corresponds to a gapless phase concerning the Lyapunov spectrum. While the bulk-edge correspondence is obscured in this phase, we demonstrate that the topological invariant can still dynamically characterize the two gapped and gapless phases. Our result is summarized in Fig. 1.

We consider a $(1+1)$-dimensional quantum circuit acting on $2L$ Majorana fermions [see Fig. 1(a)], consisting of repeated applications of three elementary steps: (i) unitary time evolution $\hat{\mathcal{U}}$, (ii) measurements of all Majorana pairs on odd bonds, and (iii) measurements of all Majorana pairs on even bonds. These operations update $\ket{\psi_{t}}$ to $\ket{\psi_{t+1}}$ in a single time step. To be more precise, the unitary evolution is given by the Kitaev chain Hamiltonian, $\hat{\mathcal{U}}=e^{-i\hat{\mathcal{H}}_{\mathrm{Kitaev}}}$, with

where $\gamma_{\ell}$ are Majorana fermions obeying $\{\gamma_{\ell},\gamma_{\ell^{\prime}}\}=2\delta_{\ell\ell^{\prime}}$ and $J,J^{\prime}$ are real constants. The measurements of Majorana pairs on odd/even bonds with an outcome $s=\pm 1$ are described by the Kraus operators,

which are weak measurements of the strength $\theta_{e/o}=\tanh^{-1}(\mu_{e/o})$. We set $\mu_{e}=1-\mu_{o}$ hereafter. Given a (unnormalized) state $\ket{\psi}$, each outcome $s$ is obtained with the Born probability, $p^{e/o}_{j}(s)=\bra{\psi}\hat{\mathcal{K}}_{j}^{e/o}(s)^{\dagger}\hat{\mathcal{K}}_{j}^{e/o}(s)\ket{\psi}/{\langle\psi|\psi\rangle}.$ In the following, we use three boundary conditions: open (OBC), periodic (PBC), and antiperiodic (APBC). For the unitary evolution $\hat{\mathcal{U}}$, the boundary condition is simply specified by $J^{\prime}$: $J^{\prime}=0,J,$ and $-J$ for the OBC, PBC, and APBC, respectively. For the measurements, the OBC means that we discard $\hat{\mathcal{K}}^{e}_{L}(s)$ from the circuit, while we need a careful definition of the APBC, as explained later.

At the time $t=T$, an initial state $\ket{\psi_{0}}$ is evolved by the Kraus operator labeled by a sequence of outcomes $\bm{s}=\{\bm{s}_{1},\cdots,\bm{s}_{T}\}$ with $\bm{s}_{t}=\{s_{1,t},\cdots,s_{2L,t}\}$,

Since both unitary evolution and measurements are bilinear in Majorana fermions, this circuit maps a fermionic Gaussian state to another Gaussian state [46, 47, 48, 49, 50, 51, 52]. A fermionic Gaussian state is completely characterized by the Majorana covariance matrix $\Gamma_{t}$ whose element reads $(\Gamma_{t})_{\ell\ell^{\prime}}=(i/2)\bra{\psi_{t}}[\gamma_{\ell},\gamma_{\ell^{\prime}}]\ket{\psi_{t}}/\langle\psi_{t}|\psi_{t}\rangle$. Time evolution of the (unnormalized) Gaussian state is encoded in the transformation of the Majorana fermions, $\vec{\gamma}\to\hat{\mathcal{K}}_{T}^{\dagger}(\bm{s})\vec{\gamma}(\hat{\mathcal{K}}_{T}^{\dagger}(\bm{s}))^{-1}=K_{T}(\bm{s})\vec{\gamma}$, where

Here, $H_{\mathrm{Kitaev}}$ and $\Theta^{e/o}(\bm{s}_{t})$ are real antisymmetric matrices defined through $\hat{\mathcal{H}}_{\mathrm{Kitaev}}=i\vec{\gamma}^{\mathsf{T}}H_{\mathrm{Kitaev}}\vec{\gamma}/4$ and $\prod_{j}\hat{\mathcal{K}}^{e/o}_{j}(s_{2j/2j-1,t})\propto\exp({i\vec{\gamma}^{\mathsf{T}}\Theta^{e/o}(\bm{s}_{t})}\vec{\gamma}/4)$ with $\vec{\gamma}=(\gamma_{1},\cdots,\gamma_{2L})^{\mathsf{T}}$ (see Supplemental Material [53]).

Since this circuit $K_{T}(\bm{s})$ is a product of random matrices, we can perform the Lyapunov analysis [54, 55, 56, 57, 58, 59, 60]. Provided that the Oseledec theorem [61, 62, 63] holds, the Lyapunov spectrum $z_{\ell}$ does not depend on the sequence of outcomes $\bm{s}$. It is also related to the energy spectrum of the effective Hamiltonian,

and the corresponding many-body Hamiltonian $\hat{\mathcal{H}}_{\textrm{eff},T}(\bm{s})=-i\vec{\gamma}^{\mathsf{T}}H_{\textrm{eff},T}(\bm{s})\vec{\gamma}/4$. Since $H_{\textrm{eff},T}(\bm{s})$ is real antisymmetric, its eigenvalues $iz_{\ell,T}(\bm{s})\in i\mathbb{R}$ come in pairs, $z_{2j-1,T}(\bm{s})=-z_{2j,T}(\bm{s})$. This defines the single-particle energy spectrum $z_{\ell,T}(\bm{s})$, which in the asymptotic limit coincides with the Lyapunov spectrum: $z_{\ell}=\lim_{T\to\infty}z_{\ell,T}(\bm{s})$. We below arrange them as $0\leq z_{1}=-z_{2}\leq\cdots\leq z_{2L-1}=-z_{2L}$. We can also compute the corresponding Lyapunov vectors $\vec{w}_{\ell,T}(\bm{s})$. Written in the matrix form $\widetilde{O}_{T}(\bm{s})=(\vec{w}_{1,T}(\bm{s}),\cdots,\vec{w}_{2L,T}(\bm{s}))$, they give in the asymptotic limit an orthogonal matrix that brings $H_{\textrm{eff},T}(\bm{s})$ into the standard form,

In practice, we construct $\widetilde{O}_{T}(\bm{s})$ through computing Lyapunov vectors corresponding to the non-negative Lyapunov spectrum based on the complex fermion representation, which is equivalent to the Majorana representation [53]. In the following, the initial state of the trajectories is fixed to be a vacuum state $\ket{\psi_{0}}=\ket{0}$, whereas the Lyapunov analysis itself is performed with random initial vectors $\vec{w}_{\ell,0}$ for a given trajectory. We only look at a single trajectory specified by a typical sequence of $\bm{s}$ and consider a temporal average of quantities in the long-time limit unless otherwise mentioned [53].

We first perform the Lyapunov analysis for the measurement-only circuit ($J=J^{\prime}=0$). In this case, a direct transition at $\mu_{e}\simeq 0.5$ occurs from a topological to a trivial area-law phase, as numerically checked by the topological entanglement entropy and bipartite mutual information [53], as done in [41, 34, 42].

Remarkably, we reveal that the topological area-law phase is characterized by Majorana zero modes for the Lyapunov spectrum. Figure 2(a) shows the non-negative single-particle Lyapunov spectrum $z_{k}$ against $\mu_{e}$ for the measurement-only circuit under the OBC for $L=64$. Importantly, $z_{1}$ takes a finite value for $\mu_{e}>0.5$, while it vanishes for $\mu_{e}\leq 0.5$. Vanishing of $z_{1}$ under the OBC in the topological area-law phase implies the many-body Lyapunov gap closing for $\hat{\mathcal{H}}_{\textrm{eff},T}(\bm{s})$, which is reminiscent of Majorana zero modes for the static Kitaev chain [64].

We next show that the above zero mode is a spatially localized edge mode by analyzing the spinor wavefunction $\vec{\psi}_{T}(\bm{s})$ [65] with its element $(\vec{\psi}_{T}(\bm{s}))_{j}=\sqrt{\sum_{a=1,2}\sum_{b=2j-1,2j}((\vec{w}_{a,T}(\bm{s}))_{b}^{2}/2}$ corresponding to the lowest energy $z_{1,T}(\bm{s})\simeq z_{1}$. In Fig. 2(b), We plot the long-time average of the sum of the squared amplitudes for $\vec{\psi}_{T}(\bm{s})$ at both edges, $|(\vec{\psi}_{T}(\bm{s}))_{1}|^{2}+|(\vec{\psi}_{T}(\bm{s}))_{L}|^{2}$, against $\mu_{e}$. Deep inside the topological phase ($\mu_{e}\ll 0.5$), it takes a value close to $1$ regardless of the system size, signaling localization of the zero mode near the edges. In contrast, it decreases toward a value $2/L$ in the trivial phase, indicating that the Majorana mode delocalizes over the whole lattice on average. The localization of the zero modes in the topological phase is much more visible if we directly look at the spatial profile of the $|(\vec{\psi}_{T}(\bm{s}))_{j}|^{2}$. The inset of Fig. 2(b) shows (trajectory-averaged) values of $|(\vec{\psi}_{T}(\bm{s}))_{j}|^{2}$ for $L=16$ and $\mu_{e}=0.1$. These results indicate that the Lyapunov spectrum clearly reveals edge-localized Majorana zero modes in the topological area-law phase.

We now introduce a bulk topological invariant and relate it with the presence (absence) of the Lyapunov edge modes in the topological (trivial) area-law phase. Since the circuit $\hat{\mathcal{K}}_{T}(\bm{s})$ conserves the fermion parity $\hat{P}=\prod_{j=1}^{L}i\gamma_{2j-1}\gamma_{2j}$, the eigenstates of the effective Hamiltonian $\hat{\mathcal{H}}_{\textrm{eff},T}(\bm{s})$ have definite parities $\pm 1$ unless they are degenerate. We then define the topological invariant as a difference between the fermion parities computed for the ground state under the PBC and that under the APBC. While this is analogous to the procedure for static Kitaev chains [64, 66, 67], a special care is required to determine the APBC because the trajectories depend on measurement outcomes. Namely, the APBC is here defined with respect to the PBC; we generate a circuit under the PBC with a sequence of measurement outcomes $\bm{s}$. Then the circuit under the APBC is defined by flipping the outcomes $s_{2L,t}$ only for the boundary Majorana pairs, $\hat{\mathcal{K}}^{e}_{L}(s_{2L,t})\to\hat{\mathcal{K}}^{e}_{L}(-s_{2L,t})$, with $\hat{\mathcal{U}}$ under the APBC. Using $\mathcal{\hat{H}}_{\mathrm{eff},T}^{\mathrm{PBC/APBC}}(\bm{s})$ defined through the above prescription, we introduce the topological invariant as the product of their ground states’ parities. Since the parity can be computed as the sign of the Pfaffian of the single-particle Hamiltonian for noninteracting systems, which is known as Kitaev’s formula [64], the topological invariant for monitored Majorana circuits is given by

However, the explicit form of $H_{\textrm{eff},T}(\bm{s})$ is numerically intractable for large $T$. Thus, we instead compute $\chi_{T}(\bm{s})$ defined by

Since $\chi_{T}(\bm{s})$ approximates and converges to $Q_{T}(\bm{s})$ for $T\to\infty$, we will also call $\chi_{T}(\bm{s})$ as the topological invariant. We note that $\widetilde{O}^{\textrm{PBC/APBC}}_{T}(\bm{s})$ is not an orthogonal matrix for general $T$.

In Fig. 2(c,d), we show the lowest single-particle Lyapunov spectrum $z_{1}$ for the measurement-only circuit under the (c) PBC and (d) APBC. It takes a finite value for both cases, implying a finite bulk Lyapunov gap, within each area-law phase. As shown in Fig. 2(e), the topological invariant $\chi_{T}(\bm{s})$ clearly separates the topological and trivial area-law phases. Note that these results are qualitatively independent of the measurement outcomes $\bm{s}$. In fact, $P^{\mathrm{PBC}}_{T}(\bm{s})=1$ is satisfied in the whole parameter region, since the gap does not close at the transition $\mu_{e}\simeq 0.5$ for PBC due to finite-size splitting $z_{1}\sim 1/L$. On the other hand, $P^{\mathrm{APBC}}_{T}(\bm{s})$ abruptly changes from $-1$ to $+1$ across the transition point for APBC, implying an exact gap closing near $\mu_{e}=0.5$. This results in the observed transition in $\chi_{T}(\bm{s})$. The behaviors of the bulk Lyapunov gap resemble those of the bulk energy gap $\Delta E$ of the static translation-invariant Kitaev chain [64]; in that case, $\Delta E=0$ for PBC and $\Delta E\sim 1/L$ for APBC at the transition, across which the Pfaffian invariant changes its sign.

The edge-localized zero mode under the OBC in Fig. 2(a,b) and the topological invariant in Fig. 2(e) indicate that the bulk-edge correspondence applies even to nonequilibrium phases under temporally random measurements, if the bulk gap is open. Such an exploration based on topology in random dynamics is accomplished by the Lyapunov analysis and the appropriate definition of the topological invariant via effective Hamiltonians. Note that a naive quantity $P_{T}^{\textrm{PBC}}(\bm{s})P^{\textrm{APBC}}_{T}(\bm{s}^{\prime})$ with $\bm{s}\neq\bm{s}^{\prime}$ does not provide a good topological invariant in general. Rather, focusing on the trajectory under the APBC dependent on the trajectory under PBC makes their fermion parities meaningful to define the invariant under random measurements.

We now study the monitored circuit with unitary dynamics by the Kitaev Hamiltonian $\mathcal{\hat{H}}_{\mathrm{Kitaev}}$ and discuss how the Lyapunov spectrum and the topological invariant behave. Reference [51] considered a continuous-time version of our model to find three different entanglement phases as $\mu_{e}$ increases. Specifically, the unitary part leads to a critical phase whose entanglement scales as $S\sim(\ln L)^{2}$ between the topological and trivial area-law phases (see also [37, 38, 68, 69, 39, 70, 40, 41, 71, 42]). In our circuit model, we numerically find three entanglement phases separated by two transition points $\mu_{e}\simeq 0.4$ and $\simeq 0.6$ for $J=0.5$, using the topological entanglement entropy and bipartite mutual information [53].

To discuss the topology of the three phases, we first investigate the Lyapunov spectrum in the circuit with OBC, finding that the critical phase corresponds to a gapless phase without Majorana zero modes. Figure 3(a) shows the $\mu_{e}$ depedence of the non-negative single-particle Lyapunov spectrum. As observed in the measurement-only case, the lowest mode $z_{1}$ become gapless (gapped) in the topological (trivial) phase. In the critical phase for $0.4\lesssim\mu_{e}\lesssim 0.6$, not only the lowest mode $z_{1}$ but also higher modes $z_{k(\neq 1)}$ appear to become gapless, implying the closing of the bulk Lyapunov gap. Interestingly, the $L$-dependence of $z_{1}$ seems different between the topological area-law phase and the critical phase; $z_{1}$ decays faster than power law in the topological area-law phase ($\mu_{e}\lesssim 0.4$), while it appears to decay in $1/L$ for our available system sizes in the critical phase ($0.4\lesssim\mu_{e}\lesssim 0.6$) [53].

The above observation of the bulk spectrum is also confirmed from the lowest spectrum $z_{1}$ under the PBC and APBC. Indeed, as shown in Fig. 3(b,c), $z_{1}$ goes to zero for $0.4\lesssim\mu_{e}\lesssim 0.6$, whereas the bulk gap is open for other $\mu_{e}$. While the scaling form of $z_{1}$ right at the transitions $\mu_{e}\simeq 0.4$ and $0.6$ is close to $1/L$, it appears to decay faster than $1/L$ inside the critical phase (except the APBC circuit at $\mu_{e}=0.5$, where exact gap closing is anticipated nearby) [53].

The above results for the spectral gaps indicate that the bulk-edge correspondence is obscured for the gapless phase if we consider $\chi_{\infty}\equiv\lim_{T\rightarrow\infty}\chi_{T}(\bm{s})$ for finite $L$, which is confirmed to be convergent irrespective of $\bm{s}$. That is, since $z_{1}$ can only be exactly zero near $\mu_{e}=0.5$ for APBC, the asymptotic topological invariant can switch the sign only at this point for finite system sizes. Indeed, we have numerically observed that $\chi_{T}$ flows towards $-1$ for $\mu_{e}<0.5$ and $+1$ for $\mu_{e}>0.5$ as $T$ increases [53]. In contrast, as discussed above, the Majorana zero mode seems absent even for $0.4\lesssim\mu_{e}<0.5$. Therefore, while the critical phase is unstable in static Majorana chains and is thus unique to monitored Majorana chains, the bulk-edge correspondence is obscured since it corresponds to the gapless phase. This is reminiscent of the fact that topological phase is usually well defined in the gapped phase in static systems.

Remarkably, however, we find that the topological invariant dynamically characterizes the different phases, including the critical gapless phase. Specifically, we consider the sample-averaged invariant, $\chi_{T}$, for the timescale $T=\mathcal{O}(L)$. Figure 3(d) shows $\chi_{T=L}$ for different system sizes, from which we find three crossings with $\mu_{e}\simeq 0.4,0.5,$ and $0.6$. The locations of the first and last crossings coincide well to the transitions between the area-law gapped phases and critical gapless phase. Notably, this characterization is attributed to the divergent relaxation time of the invariant in the gapless phase. If we denote the timescale as $\tau_{\mathrm{relax}}$, it is at least longer than $\sim 1/z_{1}$, which increases faster than $L$ inside the gapless phase. Therefore, $\chi_{T=L}$ will not converge and almost stay zero for the gapless phase. This is contrasted to the gapped phases with $z_{1}=\mathcal{O}(L^{0})$, where $\chi_{T}$ rapidly converge to $\pm 1$ for $T=\mathcal{O}(L)$. Note that the crossing point at $\mu_{e}\simeq 0.5$ is different from the boundaries of the three phases. The above argument indicates that this crossing is a finite-size artifact, and $\chi_{T=L}$ will flatly become zero in the entire critical phase for $L\rightarrow\infty$.

We have shown that monitored Majorana circuits in the gapped phases exhibit the bulk-edge correspondence through investigating the Lyapunov spectrum, gapless edge modes, and topological invariants based on the fermion parity. To define the topological invariant, we have used a unique methodology where the trajectory under APBC is defined with respect to that under PBC, according to twisted measurement outcomes at the boundary. We have also shown that the critical phase corresponds to the gapless phase in terms of the Lyapunov spectrum and is dynamically characterized by the topological invariant in a timescale $t=\mathcal{O}(L)$.

Our study opens the way to explore topological features in monitored systems through the Lyapunov analysis. An important future direction is to examine many-body systems, where our topological invariant based on the twisted measurement outcomes readily applies. For example, it is intriguing to analyze monitored circuits with $Z_{2}\times Z_{2}$ symmetry, for which measurements can stabilize an area-law phase with a symmetry-protected topological order of the cluster state [26, 38, 32]. Such a topological area-law phase will be charactarized by (nearly) four-fold degenerate Lyapunov spectrum under the OBC from gapless edge modes and a topological invariant measuring the change of one $Z_{2}$ charge by twisting boundary measurement outcomes for another $Z_{2}$ symmetry.

Note added.— While this work was in its final stage, we learned about a recent paper which investigates a similar issue [72].

Data availability.— Data and simulation codes are available upon reasonable request.