Boundary transfer matrix spectrum of measurement-induced transitions

We consider monitored quantum circuits with open spatial boundary conditions, using various boundary conditions depicted in Fig 1. To begin, we consider a fixed set of space-time coordinates for the action of measurement operators and unitary gates. The dynamics is then described by the quantum channel

where $\rho$ is system’s initial density matrix and $K_{\bf{m}}=K_{t}^{m_{t}}K_{t-1}^{m_{t-1}}\dots K_{1}^{m_{1}}$ where $K_{s}^{m_{s}}=P_{m_{s}}^{s}U_{s}$ consists of the random unitary operations $U_{s}$ (possibly trivial for measurement-only dynamics) and random projectors $P_{m_{s}}^{s}$ onto the measurement outcomes $m_{s}$. Each term ($K_{\bf{m}}\rho K_{\bf{m}}^{\dagger}$) in the sum of Eq. 2 represents a quantum trajectory which occurs with the Born probability $p_{\mathbf{m}}$ ($=\rm{Tr}(K_{\bf{m}}\rho K_{\bf{m}}^{\dagger})$) for the measurement outcomes $\bf{m}$. The trajectories in the channel form an ensemble of statistical mechanical models with inherent space-time randomness coming from the measurement outcomes. Following Ref. Zabalo et al. (2022), we introduce the transfer matrix $\mathcal{T}_{j}=K_{2j}^{m_{2j}}K_{2j-1}^{m_{2j-1}}$ (shown in Fig 1) for the unitary-measurement dynamics which describes evolution for a single time period (maps $\rho\to\mathcal{T}_{j}\rho\mathcal{T}_{j}^{\dagger}$). Now at late times the singular values $\sigma^{\bf{m}}_{i}$ (where $(\sigma^{\bf{m}}_{i})^{2}$ are eigenvalues of $K_{\bf{m}}K_{\bf{m}}^{\dagger}$) of $K_{\bf{m}}$ decay exponentially $\sigma^{\bf{m}}_{i}=e^{\lambda_{i}^{\bf{m}_{t}}}$ (where $\lambda_{i}^{\bf{m}_{t}}<0$) as the state purifies under evolution of maximally mixed density matrix (where $(\sigma^{\bf{m}}_{i})^{2}$ are its eigenvalues). The trajectory averages of $\lambda_{i}^{\bf{m}_{t}}$ give the values of the Lyapunov exponents $\lambda_{0},\lambda_{1},\dots$ (in descending order) of the transfer matrix. The leading Lyapunov exponent is related to the average free energy of the statistical mechanical model up to a factor of time, i.e., $F=-\lambda_{0}t$. Interestingly, this is equivalent Zabalo et al. (2022) to the Shannon entropy of the measurement record, where

At criticality, the scaling of this averaged free energy with system size $L$ is dictated by conformal invariance. This can be seen by introducing the replicated partition functions $\bar{Z}_{k}=\sum_{\bf{m}}(p_{\bf{m}}p_{\bf{m}}^{k})$ from which the averaged free energy can be obtained as $F=\lim_{k\to 0}\frac{dF_{k}}{dk}$ in the replica limit $k\to 0$ where $F_{k}$( $=-\ln{\bar{Z}_{k}}$) is the replicated free energy. For any finite number of replicas $k$, $F_{k}$ is the free energy of a statistical mechanics model which exhibits (for $k$ small enough) a 2nd order phase transition with emerging conformal invariance Bao et al. (2020); Jian et al. (2020). In the replica limit $k\to 0$, this transition coincides with the MIPT. To find the free energy density we have to take into account the space-time area ($A$) where $A=\alpha Lt$ and $\alpha$ is a non-universal anisotropy factor that characterizes the asymmetry between the intrinsic space and time directions of the statistical mechanical model with periodic boundary conditions. Using standard CFT results Affleck (1986); Blöte et al. (1986); Affleck (1988); Francesco et al. (2012), this implies that the bulk free energy density $f(L)=F/(\alpha tL)$ – scales asZabalo et al. (2022)

for a cylindrical geometry of circumference $L$ and circuit depth $t$ (when $t\gg L$), where $f(L=\infty)$ is the extensive bulk term, and the effective central charge

is a universal number that characterizes the log-CFT. Note that the actual central charge of the MIPT CFT is $c=\lim_{k\to 0}c(k)=0$, since the partition function $\bar{Z}_{k}$ becomes trivial $\bar{Z}_{k\to 0}=1$ in the replica limit. Thus the free energy is trivial$(=0)$ with no system size dependence and hence $c=0$. In this bulk geometry, typical values of scaling dimensions are extracted from differences of higher (subleading) Lyapunov exponents, as shown in Ref. Zabalo et al., 2022.

In order to study boundary scaling properties, we now turn to a different geometry, consisting of an infinite strip of width $L$ as shown in Fig 2. This coordinate system $(x^{\prime},t^{\prime})$ with $z^{\prime}=x^{\prime}+it^{\prime}$ can be related to the upper half of the complex plane via the conformal transformation $x^{\prime}+it^{\prime}=z^{\prime}=(iL/\pi)\log(z)$ with $z=x+it$. We introduce distinct boundary conditions on the bottom of the upper half plane, where the left boundary condition is depicted as red and labelled $\alpha$ and the right boundary conditions is blue and labelled as $\beta$ as shown in Fig. 2, corresponding to the insertion of BCC operator $\Phi_{\alpha\beta}$ at the origin. Upon applying the conformal transformation, this maps to distinct boundary conditions on the left and right edges of the strip (quantum circuit). In the strip geometry, the BCC operator $\Phi_{\alpha\beta}$ is inserted at imaginary time $\tau=-\infty$, and changes the boundary conditions from $\alpha$ to $\beta$ from the left to the right side of the strip. As a result of the different (conformally invariant¹¹1Because the bulk is critical, any boundary condition implemented at the microscopic lattice scale of the circuit will at long scales result in a scale-invariant and conformally invariant boundary condition.) boundary conditions $\alpha$ and $\beta$, the scaling of the averaged free energy density is given by Blöte et al. (1986); Cardy (1984b)

Compared to (4), there is an extra non-universal $1/L$ dependence due to the presence of the surface free energy $f_{{s}}^{(\alpha|\beta)}$, and a boundary specific universal contribution from the scaling dimension $h_{\alpha|\beta}$ ($=0$, when $\alpha=\beta$). As illustrated in Fig 2, this scaling dimension corresponds to the insertion of a BCC operator $\Phi_{\alpha\beta}$ at imaginary time $\tau=-\infty$ which changes boundary conditions from $\alpha$ to $\beta$ from the left to the right side of the strip. Just like the effective central charge, the scaling dimension $h_{\alpha|\beta}$ is obtained as a derivative $h_{\alpha|\beta}=\lim_{k\to 0}\frac{dh_{\alpha|\beta}(k)}{dk}$ from the replicated theory (and represents a typical scaling dimension). The surface free energy $(f_{s}^{\alpha|\beta})$ contribution occurs in any statistical mechanics model with specified boundary ends, including for an example an Ising model on a strip.

Using eq. (6), we see that we can extract numerically the BCC scaling dimensions $h_{\alpha|\beta}$ (belonging to the boundary spectrum of the CFT) from finite size scaling using various sets of boundary conditions $(\alpha,\beta)$. The boundary conditions that we will consider are guided by the underlying replica statistical mechanics model Jian et al. (2020). We will consider both “free” boundary conditions (corresponding to open boundary conditions and denoted by $\alpha=f$), and different “fixed” boundary conditions, corresponding to fixing the boundary spins of the statistical mechanics model at the boundary. The degrees of freedom (“spins”) of the replicated statistical mechanics model are permutations of the replicas $g\in S_{k}$, and we will only consider two particular permutations corresponding to dissipation (identity permutation, label $\alpha=a$), and entanglement (cyclic/swap permutation, label $\beta=b$). This terminology is consistent with Ref. Li et al., 2021b.

We first introduce dissipators at a boundary end to implement the boundary condition shown in Fig 1(b). A single qubit maximally mixed depolarizing channel or dephasing along the $z$-axis are used to model dissipation. For random Haar circuits, dephasing and depolarizing channels are expected to flow – in the renormalization group (RG) sense – to the same conformally invariant boundary condition. These are added after each time period as shown in Fig 1(b). The resulting dynamics is described by the quantum channel

where $\mathcal{D}$ is the single qubit depolarizing/dephasing mapNielsen and Chuang (2010) that acts on the boundary qubit $x=L$. The maximally mixed depolarizing channel maps $\rho\to\mathcal{D}(\rho)=I/2$, which models absolute decoherence and the dephasing channel maps $\rho\to\mathcal{D}(\rho)=P_{\uparrow}\rho P_{\uparrow}+P_{\downarrow}\rho P_{\downarrow}$, which amounts to adding measurements on this qubit and discarding the measurement outcomes Weinstein et al. (2022); Li and Claassen (2023). This random circuit geometry will be denoted by the boundary conditions $(\alpha,\beta)=(f,a)$, and in the statistical mechanics model language it corresponds to fixing boundary spins at the right boundary to the identity permutation.

Mapping this boundary condition back to the half-plane, we see that the $(f,a)$ boundary condition corresponds to a setup in which one measures the spins on the left half-line at the final time, and computes the resulting Born probability, $\mathrm{Tr}(|\psi\rangle_{\mathbf{\tilde{m}}}\langle\psi|_{\mathbf{\tilde{m}}})$, where $\mathbf{\tilde{m}}$ is a bit-string of measurement outcomes for both the mid-circuit measurements in the spacetime bulk of the circuit and the measurements on the left half-system on the final time-step. This quantity receives a nonuniversal surface contribution (which would exist even if all the sites were independent) and a subleading part capturing the long-range correlations between the outcomes of distant measurements.

Another observable that contains direct information about the (typical) scaling dimension $h_{f|a}$ turns out to be the Shannon entropy of the measurement record of the circuit at early time, i.e. that of a shallow circuit. To see this consider a “sideways variant” of Fig. 2, where the roles of space and time are exchanged: Specifically, it is convenient to first consider a circuit with periodic boundary conditions in space of “large” circumference $L$, and “small” depth $t$, so that $t\ll L$. Then choose as initial condition the state labeled by the boundary condition $f$ (which can be represented, e.g., by a simple product state); the state of the circuit appearing at depth $t$, the ‘final’ time, which contains the physical qubits, is represented by boundary condition $a$ ²²2For both of these interpretations of the boundary conditions $f$ and $a$, as (simple product) initial state and final state containing the physical qubits, respectively, see Fig. 2 (a), and the corresponding text of Ref. Li et al., 2021b.. Now consider the partition function $Z_{\mathbf{m}}$ of the so-defined shallow circuit which equals³³3paragraph below Eq. 2 the Born probability for the measurement record, $p_{\mathbf{m}}=$ $Z_{\mathbf{m}}$. For this shallow circuit we are considering now, where $t\ll L$, the corresponding Born-probability-averaged free energy density, which by definition, Eq. 3, is the corresponding shallow-circuit Shannon entropy density, has the same form⁴⁴4 see, e.g., Ref. Li et al., 2021b as Eq. 6, except that the roles of space and time are exchanged,

where $\overline{O}_{\bf m}=\sum_{\bf m}p_{\bf m}O_{\bf m}$ refers to average over quantum trajectories, and $f(t=\infty)$ and $f_{s}^{(f|a)}$ are non-universal. Eq. 8 shows that the ${1\over t^{2}}$-decay of the Shannon entropy density at early times is directly affected by the exponent $h_{f|a}$, a universal signature of the sensitivity of this entropy to the initial state $f$. At long times, the circuit “forgets” about the initial state $f$, and at finite spatial circumference $L$ the behavior in Eq. 4 obtains in the opposite limit of a deep circuit where $t\gg L$, in the steady state, for the case of periodic spatial boundary conditions we are currently considering. (Open spatial boundary conditions do not modify the early-time $t\ll L$ behaviorLi et al. (2021b), Eq. 8, but the late-time form Eq. 6 obtains in the opposite limit of a deep circuit, $t\gg L$, in the steady state.)

In order to implement different boundary conditions, we introduce ancilla qubits at left and right edges Li et al. (2021b) as shown in Fig. 1(c). Every time step, we introduce two fresh system qubits that individually form a Bell pair with an ancilla qubit and inject them into the system at the first ($x=1$) and last ($x=L$) qubit location. Then after the evolution of system qubits for one time period we take out the first and last qubit to store them as ancillas with no further action on them. After that we repeat by introducing two fresh pre-entangled qubits at the boundary ends again, and so on. Thus for a circuit of depth $t$, we introduce $2t$ ancilla qubits. Tracing over the ancillas at a boundary effectively implements dissipation, corresponding to $\alpha=a$.

We now compute the entanglement between the right and left ancillas, while measuring all physical qubits – corresponding to a free $f$ boundary condition on the top layer of the circuit⁵⁵5see Ref. Li et al., 2021b, Appendix C, and Ref. Popp et al., 2005.. This measurement-induced entanglement (MIE) Popp et al. (2005); Lin et al. (2023) between right and left boundaries effectively implements a change in boundary conditions: one of the boundaries is traced over (boundary condition $\alpha=a$), while the other is subject to a partial trace (corresponding to a cyclic permutation of replicas, boundary condition $\beta=b$). In the statistical mechanics language, this forces the insertion of a domain wall propagating vertically between the two ends of the strip. The MIE is then directly given by the free energy cost of changing this boundary condition:

A “sideway” version of this geometry, where the space and the time directions of the circuit were exchanged, was considered in Ref. Li et al., 2021b. We thus see that this geometry allows us to extract the “entanglement” scaling dimension $h_{a|b}$ which controls the scaling of the entanglement entropy at criticality. In particular, in previous works $h_{a|b}$ was typically extracted directly from the logarithmic scaling of entanglement entropy (of the physical qubits) at criticality $S_{A}\sim 2h_{a|b}\ln L_{A}$, which exhibits the coefficient $\gamma$ of the logarithm in Eq. (1), $\gamma=2h_{a|b}$. As already mentioned, we can unfortunately not use this boundary condition with Haar random circuits as the Hilbert space dimension grows with time as one adds ancilla qubits as depicted in Fig. 1.

In the rest of this paper, we apply this boundary transfer matrix approach numerically. We average free energies over $10^{5}$–$10^{7}$ sampled quantum trajectories where each trajectory is evolved up to time $t=10L$, after an initial equilibration time ($\sim 4L$). The sampling complexity of quantum trajectories limits us to inspect small system sizes, even with state-of-the-art computing platforms. However we still manage to do larger systems, both for Clifford and Haar, as compared to the previous work with periodic boundaryZabalo et al. (2022). All results are shown at criticality $p=p_{c}$, with $p_{c}^{\rm{MOIM}}=0.5$, $p_{c}^{\rm{DC}}=0.205$, $p_{c}^{\rm{DH}}=0.14$, and $p_{c}^{C}=0.1596$, for MOIM, dual Clifford, dual Haar, and Clifford circuits, respectively Gullans and Huse (2020b); Zabalo et al. (2020). Note that adding dissipators at boundary does not influence saturation time of free energy density. However it results in additional Monte Carlo sampling of trajectories for Haar circuit as described in Appendix D ( Fig 12). This further limits the system sizes and thus results in larger error bars for the universal and non-universal quantities as compared to the Clifford circuit. This also leads to larger error bars for the dissipator boundary as compared to the open boundary condition, even within the Haar circuit. The space-time asymmetries for different monitored circuits are characterized by the anisotropy parameter ($\alpha$) and we extract this from the ratio of space and time correlators Zabalo et al. (2022) which gives $\alpha_{\rm{MOIM}}=\alpha_{\rm{DC}}=\alpha_{\rm{DH}}=1$ for MOIM, dual Clifford, dual Haar circuits (which are all isotropic), and $\alpha_{C}=0.61$ for the Clifford random circuit. All error bars are estimated using a bootstrap analysis where the data is bootstrapped over 1000 samples. To improve the estimate of universal quantities we use standard double fitting procedure where we successively remove small system sizes ($L<L_{\rm{min}}$) from the fit which in turn accounts for the leading order correction to the averaged free energy density.

Measurement-only circuits are comprised of different non-commuting measurement operators and are free from random unitary gates Nahum and Skinner (2020); Lang and Büchler (2020); Ippoliti et al. (2021); Lavasani et al. (2021b); Sang and Hsieh (2021); Sang et al. (2021); Sriram et al. (2022). We consider the measurement-only Ising model (MOIM) which describes measurement-only dynamics using non-commuting competitive measurement operators, $\sigma^{x}_{i}$ and $\sigma^{z}_{i}\sigma^{z}_{i+1}$, acting on sites $i\in\{1,2,\dots,L\}$ of a one-dimensional spin-1/2 chain. The dynamics involve the projective measurement operator $M[O]$ that measures the observable $O$, which then maps the state $|\psi\rangle$ onto the eigenspace of $O$, i.e., $M[O]|\psi\rangle=\frac{P_{\lambda}|\psi\rangle}{\sqrt{\langle\psi|P_{\lambda}|\psi\rangle}}$, with probability $p_{\lambda}=\langle\psi|P_{\lambda}|\psi\rangle$ where $\lambda$ is the eigenvalue of the corresponding eigenspace of $O$ and $P_{\lambda}$ is the projection operator onto the eigenspace of $O$ with eigenvalue $\lambda$. For MOIM, the observable $O$ is either $\sigma^{x}_{i}$ or $\sigma^{z}_{i}\sigma^{z}_{i+1}$ and the corresponding projection operator $P_{\lambda}$ is either $\frac{I+\lambda\sigma^{x}_{i}}{2}$ or $\frac{I+\lambda\sigma^{z}_{i}\sigma^{z}_{i+1}}{2}$ where $\lambda$ takes value $\pm 1$. In each discrete time step, the measurement operators $M[\sigma^{z}_{i}\sigma^{z}_{i+1}]$ ($\equiv M_{e}^{zz}$) and $M[\sigma^{x}_{i}](\equiv M_{i}^{x})$ acts randomly on each edge $e=(i,i+1)$ and site $i$ with a probability $1-p$ and $p$, respectively. The layer of $M_{e}^{zz}$ measurements precedes all probable $M_{i}^{x}$ operators. The initial state is set to $|\psi_{0}\rangle=|++\dots+\rangle$ where $|+\rangle=\frac{|0\rangle+|1\rangle}{\sqrt{2}}$ is the eigenstate of $\sigma^{x}$ operator and thus the resulting dynamics leaves the state invariant under the operator $\mathcal{C}=\prod_{i=1}^{L}\sigma_{i}^{x}$ which describes the symmetry of this model. The transfer matrix evolving the system for a single time step then acts as

where $\bf{e}$ and $\bf{i}$ are the sets of all edges and sites, respectively, on which the measurement operations ($M_{e}^{zz}$ and $M_{i}^{x}$) act. It has been shown that the model exhibits an entanglement phase transition between two area law phases and the physics at the critical point is described by bond percolation (which is a CFT with central charge $c=0$, as it should to describe a MIPT).

We now review the mapping of this model onto bond percolation Nahum and Skinner (2020); Lang and Büchler (2020), and generalize it to fully characterize the associated replicated statistical mechanics model $Z_{k}=\sum_{{\bf m}}p_{\bf m}^{k+1}$. First, note that to each realization of the circuit, one can associate a bond percolation configuration as shown in Fig. 3(a), where a measurement along an edge $M_{e}^{zz}$ corresponds to a horizontal bond along that edge, whereas the absence of a local measurement $M_{i}^{x}$ on a site maps to a vertical bond. This provides a one-to-one correspondence between percolation configurations and measurement locations. As we now show, the free energy is entirely given by the properties of this percolation configuration.

First, note that the MOIM is a stabilizer (Clifford) circuit, so that each measurement outcome is either fully deterministic (determined outcomes), or fully random (with equally probable outcomes). For example let us suppose we have a stabilizer state $|+,+\rangle$. Then measuring $\sigma^{z}_{1}\sigma^{z}_{2}$ will result in either $(|00\rangle+|11\rangle)/\sqrt{2}$ or $(|01\rangle+|10\rangle)/\sqrt{2}$, each with probability $1/2$. However if we measure $\sigma^{x}_{1}$ for the same state, we get back $|+,+\rangle$ with probability 1. We call the former as random and the latter to be a deterministic measurement outcome. Note that for MOIM whether an outcome is random or deterministic will depend on the order in which measurements are performed in a given layer of time but the total number of random measurements is independent of the choice of order. We choose the order of measurements from left to right end in each time step. As a result, the Born probability of a given trajectory is given by $p_{\bf m}=\left(\frac{1}{2}\right)^{N_{\rm rand}}$, where $N_{\rm rand}$ is the total number of random (non-deterministic) measurements. This means that the free energy is given by $F=\overline{N}_{\rm rand}\ln(2)$, where $\overline{N}_{\rm rand}$ is the average number of random measurements in those circuits.

Next, we notice that the number of random measurements is purely determined in terms of the geometrical properties of the percolation clusters. Namely, we find by inspection that

where $N_{s}$, $N_{\rm{vb}}$, and $N_{c}$ are the total number of sites, vertical bonds, and disconnected clusters in the percolation configuration. An example illustrating the validity of this expression is provided in Fig. 3 where we fix the order of measurements from left to right end in each time step. For this particular choice, the random and deterministic measurement locations are shown in Fig. 3(b). So for a fixed measurement trajectory with all measurement outcomes set to $1$, we write the evolution of stabilizer generators for one time step from $t=2$ to $t=3$ in Fig. 3(b), following the convention $X\equiv\sigma^{x}$ and $Z\equiv\sigma^{z}$ to denote these generators. The stabilizer generators at $t=2$ are given by $\{X_{1},X_{2},X_{3},X_{4},X_{5}\}$; this set evolves to $\{X_{1}X_{2},Z_{1}Z_{2},X_{3}X_{4},Z_{3}Z_{4},X_{5}\}$ upon measuring $\sigma^{z}_{1}\sigma^{z}_{2}$ and $\sigma^{z}_{3}\sigma^{z}_{4}$. Clearly both measurements are random in accordance with the two site example discussed earlier. Next we measure $\sigma^{x}_{2}$ and $\sigma^{x}_{4}$, which results in the stabilizer generators $\{X_{1},X_{2},X_{3}X_{4},Z_{3}Z_{4},X_{5}\}$, at the end of time $t=3$. The measurement on site $2$ is random while the measurement on last qubit is deterministic since the qubit already is in the $|+\rangle$ state. Doing this for all time steps in turn leads to the total random measurements that obey the formula in Eq. 11. Although we do not have a formal proof of this formula, we have checked its validity numerically for very large circuits. This equation can further be broken down into purely extensive contributions, and terms including non-extensive universal corrections. The first two terms scale extensively ($O(Lt)$) and will therefore contribute to the bulk free energy only. Dropping these non-universal extensive contributions, we thus find that

where $\overline{N}_{c}$ is the average number of percolation clusters at criticality. This quantity can be computed using the standard Fortuin-Kasteleyn (FK) Newman (1994) mapping between percolation and the $Q\to 1$ limit of the $Q$-state Potts model. Using these standard percolation results, we find that $\overline{N}_{c}=-\left.\frac{d}{dQ}F^{\rm{Potts}}\right|_{Q=1}$ where $F^{\rm{Potts}}$ is the free energy of the classical $Q$-state Potts model. This provides a direct relation between the free energy of the circuit (Shannon entropy of measurement record) and that of the classical Potts model. In fact, using the same reasoning as above, one can show that the replicated partition function $Z_{k}=\sum_{{\bf m}}p_{\bf m}^{k+1}$ maps onto a Potts model with $Q=4^{k}$ (up to non-universal contributions), with $Q\to 1$ in the replica limit $k\to 0$. We can then use standard CFT results for the Potts model to infer universal quantities for the MOIM transition. For example, the central charge of the Potts model Francesco et al. (2012) is given by $c^{\rm{Potts}}=1-\frac{6}{m(m+1)}$ where $m=\frac{\pi}{\arccos(\frac{\sqrt{Q}}{2})}-1$. We thus find that the effective central charge is given by

Scaling dimensions can be identified through this mapping as well, and fit into the “Kac table” Francesco et al. (2012) $h_{r,s}=((r(m+1)-sm)^{2}-1)/(4m(m+1))$.

We now turn to a numerical analysis of the MOIM in the various geometries summarized in Fig. 1. We will first implement the open (free) boundary conditions, Fig 1(a)), to extract $c_{\rm{eff}}$. Then we introduce dissipation at one end of the boundary to evaluate the dissipator scaling dimension ($h_{f|a}$). Finally to extract the entanglement scaling dimension $h_{a|b}$, we implement the entangled system-ancilla setup (as described in fig 1(c)). We make use of stabilizer formalism Aaronson and Gottesman (2004); Gottesman (1998); Nahum et al. (2017); Li et al. (2019); Krastanov et al. (2022), which allows for efficient classical simulation of large quantum circuits.

We first compute the numerical value of the effective central charge in the cylindrical geometry setup with periodic boundary conditions. The plot of free energy density $(f)$, shown in Appendix A (Fig. 8(a)), follows a straight line when plotted against $1/L^{2}$ as expected from Eq 4 and the slope gives $c_{\rm{eff}}=0.96(1)$. This is in good agreement with the analytic prediction in Eq (13). We also extracted a precise value of $f(L=\infty)=0.55718(2)$, using the linear double fitting procedure. We now implement MOIM with free boundary ends as shown in Fig 1(a), corresponding to $(\alpha,\beta)=(f,f)$. The free energy density scales in accordance with Eq. 6 where $h_{f|f}=0$ (as expectedLi et al. (2021b)) and the resulting fit gives $c_{\rm{eff}}=1.0(1)$ as shown in Fig 4(a). Note that this estimate is less accurate than the periodic boundary conditions case, largely because of the non-universal surface term $f_{(f|f)}^{s}$ which adds a fitting parameter. In order to reduce the number of fitting parameters, we use the extensive free energy term $f(L=\infty)$ obtained from periodic boundary conditions in Eq. 6.

Next, we implement boundary depolarizing/dephasing channels as shown in fig 1(b), to extract the dissipator scaling dimension $h_{f|a}$. In order to get rid of the extensive bulk contribution, we plot the free energy difference

where $f_{\delta s}=f^{s}_{(f|a)}-f^{s}_{(f|f)}$. The plot of $(f_{(f|a)}-f_{(f|f}))L$ vs $\frac{1}{L}$ displays a linear trend with $1/L$ as shown in 4(b). Now using the double fitting procedure we find the boundary scaling dimension $h_{f|a}(L)=0.048-\frac{0.432}{L^{2}}$ and the non-universal free energy surface term $f_{\delta s}=0.27+\frac{0.06}{L^{2}}$. We conclude that $h_{f|a}=0.048(6)$ and $f_{\delta s}=0.275(1)$. Note that the addition of dissipators still preserves the global symmetry $\mathcal{C}$. We also observe that the scaling dimension remains unchanged with the type of dissipators since dephasing along $z-$axis is equivalent to depolarizing in this model.

The mapping onto bond percolation can be generalized in the presence of boundary dissipators. We find that boundary dissipation induces additional random (non-deterministic) measurements that are associated with the hulls of percolation clusters touching the boundary, see Appendix B. The BCC operator associated with changing the fugacity of such boundary hulls was identified in Ref. Jacobsen and Saleur (2008). Combining these results, we find that $h_{f|a}=\frac{\sqrt{3}}{8\pi}\ln 2\simeq 0.048$ (see Appendix B), in very good agreement with our numerical results.

Finally, the boundary scaling dimension $h_{a|b}$ has been determined both numerically and analytically for the MOIM with cylindrical geometryLang and Büchler (2020), with $h_{a|b}=\left.\ln 4\frac{d}{dQ}h^{\rm{Potts}}_{1,2}\right|_{Q=1}\simeq 0.096$. This critical exponent corresponds to counting clusters crossing the strip. Here, we extract it using the boundary ancilla setup of Fig. 1(c). We compute the measurement-induced entanglement ($S_{\rm{MIE}}$) between ancilla qubits present on the left and right ends of the boundary where all system qubits after the last time step are measured. The result in Fig. 4(c) agrees with the scaling form Eq. 9 and we extract the entanglement scaling dimension $h_{a|b}(L)=0.097-\frac{0.06}{L^{2}}$. We conclude that $h_{a|b}=0.097(2)$, which is in good agreement with the theory prediction. Hence these results completely support our approach to obtain the boundary conformal spectrum. We thus can now generalize it to the more general unitary-measurement dynamics.

We will first start by discussing the calculation of $c_{\rm{eff}}$ in cylindrical geometry (periodic boundary conditions). In Appendix A (Fig 8(b)), we find that the free energy density $f$ shows a clear linear dependence when plotted against $1/L^{2}$ in accordance with the CFT result stated in Eq. 4. This yields $c_{\rm{eff}}=0.349(3)$ and $f(L=\infty)=0.131574(6)$. We now implement dual Clifford circuit with free boundary ends where we find $c_{\rm{eff}}=0.32(3)$ from the plot between $f_{(f|f)}$ and $1/L$ as shown in fig 5(a). More precisely, using a double fitting procedure, we find $c_{\rm{eff}}(L)=0.32-\frac{8.74}{L^{2}}$ and $f_{(f|f)}^{s}(L)=-0.0615-\frac{0.0569}{L^{2}}$. The plot respects the scaling form given in Eq. 6 and the value of $c_{\rm{eff}}$ is in agreement with the one extracted from periodic boundary conditions. As in the MOIM case, $c_{\rm{eff}}$ has relatively large error bars as a consequence of the free energy surface term. Last, we find that $f_{(f|f)}^{s}=-0.0615(2)$.