---
title: Steady-State Entanglement Transition
url: https://www.emergentmind.com/topics/steady-state-entanglement-phase-transition
type: topic
---

# Steady-State Entanglement Transition

A steady-state entanglement phase transition is a non-equilibrium transition in which the long-time entanglement structure changes as a control parameter is varied. The relevant steady object depends on the setting: it may be an ensemble of pure-state trajectories in monitored dynamics, a Floquet diagonal ensemble or long-time periodic regime, a stationary mixed non-equilibrium steady state (NESS) of an open system, or the normalized right eigenstate selected by non-Hermitian time evolution. The literature therefore does not describe a single universality class or a single entanglement pattern. Reported transitions include volume-law to area-law changes, area-law to area-law transitions with long-range mutual information, volume-law to volume-law transitions distinguished by separability or spin-glass order, mixed-state critical points with finite but universal entanglement signatures, and first-order steady-state switches induced by level crossings [2006.09748], [1511.03668], [2107.11354], [2410.05370], [2101.04115], [2206.05384].

## 1. Conceptual scope and definitions

In the monitored-dynamics literature, a steady-state entanglement phase transition is a change in how the entanglement of long-time pure-state trajectories scales with subsystem size, even when the ensemble density matrix does not encode the transition. In boundary-driven and Lindbladian settings, the transition refers instead to the entanglement or information structure of a stationary mixed NESS. In Floquet systems, the phrase can refer either to nonanalyticities in the steady-state entanglement of the diagonal ensemble or to entanglement-spectrum transitions that become intrinsic features of the long-time periodic regime. In non-Hermitian systems, the “steady state” is typically the normalized right eigenstate with maximal imaginary part of the complex spectrum [2006.09748], [1511.03668], [1902.00025], [2510.13970], [2101.04115].

| Setting | Control parameter | Steady-state entanglement signature |
|---|---|---|
| Measurement-only PTIM | measurement probability \(p\) | area-law to area-law at \(p_c=\tfrac{1}{2}\); critical logarithmic entanglement and \(I(r)\sim r^{-2/3}\) [2006.09748] |
| Continuously monitored Heisenberg chain | measurement strength \(\lambda\) | volume-law to area-law; \(\nu=0.685(5)\) [2107.11354] |
| Periodically driven integrable systems | drive frequency \(\omega\) | steady-state nonanalyticities in \(S_\infty(\omega)\); relaxation-exponent change [1511.03668] |
| Open long-range Ising chain | spontaneous-emission rate \(\Gamma\) | finite entanglement with universal critical values; \(I(A\!:\!B)\sim \tfrac{1}{4}\log N\) [2410.05370] |
| Non-Hermitian skin-effect and Yang–Lee settings | \(\gamma/\Delta\) or imaginary field | volume/log/area-law or first-order volume-to-area transitions [2206.05384], [2101.04115] |

A recurrent source of confusion is that entanglement entropy density is not always an order parameter. In some models both phases are area-law; in others both are volume-law. Likewise, local observables may remain smooth through a sharp entanglement transition, and in monitored trajectory problems the ensemble density matrix may be blind to the transition altogether [2006.09748], [2005.03052], [2510.13970].

## 2. Measurement-only steady-state transitions: the projective transverse-field Ising model

The measurement-only benchmark is the projective transverse field Ising model (PTIM), introduced in "Entanglement Transition in the Projective Transverse Field Ising Model" [2006.09748]. It is defined on a chain of \(L\) spin-\(\tfrac{1}{2}\) degrees of freedom, initialized in \( |\Psi(0)\rangle=|+\rangle^{\otimes L}\), and evolved with no unitary dynamics. The dynamics consists solely of two noncommuting projective measurements,
\[
M_i^x \equiv M[\sigma_i^x],\qquad M_e^{zz}\equiv M[\sigma_i^z\sigma_{i+1}^z],
\]
applied stochastically with relative measurement probability \(p\). Bond measurements \(M_e^{zz}\) nucleate, enlarge, and merge Bell clusters, while site measurements \(M_i^x\) locally erode them. Their competition generates a steady-state entanglement transition at
\[
p_c=\tfrac{1}{2}.
\]

The two phases are both area-law. For \(p>p_c\), the measurement-erosion phase has short-range entanglement, rapidly saturating subsystem entropy, and exponentially decaying mutual information between distant spins. For \(p<p_c\), the cluster-condensation phase contains a macroscopic GHZ-like Bell cluster with finite density, yet bipartite entanglement remains area-law because a single macroscopic cluster contributes only \(O(1)\) Bell crossings across a cut. The mutual information between distant spins instead saturates to a finite value, providing a nonlocal signature of long-range entanglement. Exactly at criticality, entanglement across intervals acquires a logarithmic term,
\[
S_L(l)\simeq s_0+\frac{\tilde c}{3}\log_2\!\left[\frac{L}{\pi}\sin\!\left(\pi\frac{l}{L}\right)\right],
\]
with
\[
\tilde c=\frac{3\sqrt{3}\,\ln 2}{2\pi}\approx 0.573,
\]
while the mutual information decays algebraically as \(I(r)\sim r^{-\kappa}\) with \(\kappa=\tfrac{2}{3}\). The critical theory is the \(c=0\) logarithmic conformal field theory of 2D percolation, with \(\nu=\tfrac{4}{3}\). A central conceptual point is that \(\tilde c\) is a universal entanglement prefactor distinct from the CFT central charge.

The PTIM is also notable for its exact reduction to a classical nonlocal colored cluster model and then to isotropic bond percolation on the space-time square lattice. That mapping makes the transition analytically tractable and clarifies why the ensemble density matrix is uninformative: the evolution preserves the global Ising symmetry, so the steady-state ensemble density matrix is maximally mixed within the \(\mathcal U\)-invariant sector, whereas the universal structure resides in the wavefunction ensemble. The same paper gives a quantum-error-correction interpretation via the Majorana chain code, where \(p_c\) acts as a passive-protection threshold: below \(p_c\) the average lifetime of the initial logical cluster grows exponentially with \(L\), at \(p_c\) it scales as \(\tau\sim L^\beta\) with \(\beta\simeq 1\), and above \(p_c\) it scales as \(\tau\sim \log L\).

## 3. Monitored unitary dynamics: volume-to-area, volume-to-volume, and spin-glass variants

Adding unitary dynamics produces the more familiar measurement-induced transitions between volume-law and area-law steady states, but even within this class the order parameter and universality can differ sharply. In a continuously monitored disordered Heisenberg chain with local Gaussian measurements of \(\sigma_{j,z}\), the half-chain entropy \(\bar S_\infty\) interpolates from the Page value
\[
\bar{S}_\infty^{\mathrm{erg}}=\tfrac{L}{2}\ln 2-\tfrac{1}{2}
\]
at weak monitoring to an area law at strong monitoring. For \(\xi_r=1.3\) and \(\xi_s=0\), the critical measurement strength is \(\lambda_c=0.533(5)\) from the variance maximum and \(\lambda_c=0.537(4)\) from tripartite mutual-information crossings; with static disorder \(\xi_s=3\), these estimates shift to \(\lambda_c=0.441(5)\) and \(\lambda_c=0.4323(31)\). The extracted exponent \(\nu=0.685(5)\) is reported to be independent, within errors, of both static disorder and non-static noise strength, so the noise and disorder move the location of the transition without changing the continuous-measurement universality class [2107.11354].

The literature also contains monitored transitions that are not volume-law to area-law. In a class of solvable IQP-type dynamics with local projective measurements, the transition occurs within a volume-law phase. The low-measurement regime is a fully-entangled phase containing a giant non-separable cluster \(C_\infty\); the high-measurement regime is a separable-cluster phase in which the steady-state density matrix factorizes over extensively many finite clusters,
\[
\rho_{\mathrm{ss}}=\bigotimes_{i=1}^{N_c}\rho_{C_i},
\]
with \(|C_i|=O(1)\) and \(N_c\propto L\). Since the entropy density is finite in both phases, it is not an order parameter. The paper instead introduces the entangling power \(\overline{\Delta I_{AB}}\), defined from the change in mutual information between finite subsystems \(A\) and \(B\) after local measurements on the complement \(C\). The transition occurs at \(g_c=\tfrac{2}{3}\) and maps to mean-field percolation with \(\sigma=\tfrac{1}{2}\) and \(\tau=\tfrac{5}{2}\) [2005.03052].

A further variant is an exact spin-glass mapping in a random Clifford circuit ensemble with output-only measurements. There the post-measurement entanglement entropy is
\[
S(|\psi_{\mathrm{out},A}\rangle)=\operatorname{rank}(B)-\operatorname{rank}(B_A),
\]
which, for typical \(M\ge N\), reduces to the ground-state entropy of an unfrustrated 3-spin model restricted to the measured interactions. The control parameter is \(\alpha=|A|/N\), and the transition at \(\alpha_{c,\mathrm{scaling}}=0.9175\pm 0.0006\) separates a paramagnetic-like volume-law phase with \(\langle S\rangle \approx N(1-\alpha)\) from a spin-glass-like volume-law phase with a smaller slope and finite Edwards–Anderson overlap
\[
q(\alpha)=\frac{1}{N}\sum_{i=1}^N \langle \sigma_i\rangle^2.
\]
Finite-size scaling of the entanglement susceptibility gives \(\nu=1.8\pm 0.1\), distinct from the percolation-like exponents often quoted in other monitored Clifford settings [2112.06939].

These monitored examples establish that steady-state entanglement transitions are not exhausted by the entropy-scaling dichotomy “volume law versus area law.” Depending on the model, the relevant long-distance structure may be cluster percolation, separability, overlap order, or information that can be activated by measurements on the complement.

## 4. Floquet steady states and temporal entanglement transitions

Floquet systems realize a different notion of steady-state entanglement transition. In periodically driven integrable models with quadratic fermionic Hamiltonians,
\[
H(t)=\sum_{\vec k}\psi_{\vec k}^{\dagger}\big[(g(t)-b_{\vec k})\,\tau_3+\Delta_{\vec k}\,\tau_1\big]\psi_{\vec k},
\]
the entanglement entropy after \(n\) drive cycles obeys
\[
S_n(l)\sim l^{\alpha(n,\omega)},\qquad d-1\le \alpha(n,\omega)\le d.
\]
This describes a crossover from area law to volume law governed by the range \(R_t(n,\omega)\) of a parent quadratic Hamiltonian \(\mathcal H_t\): for \(l\gg R_t(n,\omega)\), the state satisfies area law, whereas for \(l\ll R_t(n,\omega)\), area law is violated and the system crosses toward volume law. The relaxation to the steady state exhibits two dynamical phases,
\[
S_n-S_\infty\sim
\begin{cases}
(\omega/n)^{(d+2)/2},& \text{fast driving},\\[3pt]
(\omega/n)^{d/2},& \text{slow driving},
\end{cases}
\]
separated by a topology-changing transition in the Floquet spectrum characterized by the appearance of interior stationary points of \(|\vec\epsilon_{\vec k}|\). In one dimension, the steady-state entropy density \(s_{\mathrm{DE}}(\omega)\) develops cusp or kink singularities at Floquet band crossings \(\omega=\omega_*\), so the steady-state entanglement itself has nonanalyticities as a function of drive frequency [1511.03668].

A more recent development is the temporal entanglement transition in a periodically driven transverse-field Ising chain,
\[
H(t)=-J\sum_{i=1}^{L-1}\sigma_i^z\sigma_{i+1}^z-h(t)\sum_{i=1}^{L}\sigma_i^x,\qquad h(t)=\frac{h_0}{2}\cos(\omega t),
\]
where the transition occurs in the spectrum of the entanglement Hamiltonian \(H_E(t)\), defined by \(\rho_A(t)=e^{-H_E(t)}\). The diagnostics are the Schmidt-gap closure
\[
\Delta(t)=\lambda_0(t)-\lambda_1(t),
\]
the entanglement echo
\[
E(t)=\langle \lambda_0(0)|\lambda_0(t)\rangle,
\]
and symmetry-quantum-number flips of the subsystem parity \(P_A=\prod_{i\in A}\sigma_i^x\). The transition is entanglement-driven: it requires initially entangled states and remains invisible to local observables such as subsystem magnetization and the Loschmidt echo. At high driving frequency, the critical times become equally spaced by an emergent intrinsic timescale \(\tau_E\) decoupled from the drive period \(T=2\pi/\omega\), so the transition becomes a genuine steady-state feature of the entanglement dynamics. Finite-size scaling yields \(\nu\simeq 1.00\), with
\[
\frac{t^*}{L_A}\propto L_A^{-1/\nu},
\qquad
\frac{\epsilon_0(t^*)}{L_A}\propto L_A^{-a},\quad a\simeq 1,
\]
consistent with Ising universality in the entanglement sector rather than in local observables [2510.13970].

The Floquet literature therefore broadens the notion of steady-state entanglement transition in two ways. First, the long-time state can acquire nonanalytic entanglement features purely from coherent spectral topology, without measurements. Second, the critical degrees of freedom can reside in the entanglement Hamiltonian rather than in conventional correlation functions.

## 5. Open-system NESS criticality: finite entanglement, mixed-order behavior, and operator-space scaling

In driven-dissipative systems, steady-state phase transitions often exhibit diverging classical correlations while entanglement remains finite. A detailed example is the long-range open quantum Ising chain with power-law interactions and local spontaneous emission,
\[
\frac{d\rho}{dt}=-i[H,\rho]+\sum_i \mathcal D[L_i]\rho,\qquad L_i=\sqrt{\Gamma}\,\sigma_i^-,
\]
whose infinite-range limit \(\alpha=0\) maps closely to the open Dicke model. The disorder–order transition occurs at
\[
\Gamma_c=4\sqrt{\Delta(2J-\Delta)}.
\]
At criticality, the logarithmic negativity has the universal finite value
\[
E_{\mathcal N}=\frac{1}{2},
\]
the optimal quantum Fisher information density saturates at \(f_{\mathrm{opt}}=2\), and the spin-squeezing parameter reaches the universal \(3\,\mathrm{dB}\) value \(\xi=\tfrac{1}{2}\). The mutual information between equal halves diverges as \(I(A\!:\!B)\sim \tfrac{1}{4}\log N\), and the paper reports “hidden criticality” into the ordered phase, where \(I\sim c(\Gamma)\log N\) persists although two-point correlators no longer reflect that scaling. The resulting picture is that classical correlations diverge, mutual information exhibits critical logarithms, but genuine entanglement is bounded and sharply nonanalytic rather than divergent [2410.05370].

Boundary-driven free-fermion systems supply a second family of NESS entanglement transitions. In the 3D Anderson model coupled to clean leads at different chemical potentials, the Gaussian NESS is characterized by the correlation matrix \(G_{ij}=\langle c_i^\dagger c_j\rangle\). The mutual coherence
\[
M(A\!:\!B)=2\,\mathrm{Tr}[G_{AB}G_{BA}]
\]
is an entanglement witness and a lower bound on mutual information for fermionic Gaussian states. In the diffusive phase and at the Anderson critical point \(W_c\approx 16.5\), the NESS exhibits volume-law entanglement and correlations across a bipartition, with \(\overline{M_d(L\!:\!R)}\propto L_0^3\times \delta\mu\). In the localized phase, the scaling crosses to an area law,
\[
M(L\!:\!R)\propto \xi\,L_0^2,
\qquad
\xi\sim |W-W_c|^{-\nu},\quad \nu\approx 1.57.
\]
This realizes a steady-state entanglement transition in a mixed, current-carrying Gaussian state, controlled by the transport transition itself [1902.00025].

A different open-chain realization appears in the boundary-driven XY model solved by third quantization. There the relevant quantity is the operator-space entanglement entropy (OSEE) of the NESS. The critical line is
\[
h_c(\gamma)=1-\gamma^2.
\]
For \(h<h_c\), the NESS displays long-range magnetic correlations and extensive operator-space entanglement,
\[
S(L)=sL+\mathrm{const},
\]
whereas for \(h\ge h_c\), \(S(L)=O(1)\). At criticality, the connected correlator decays as \(C^z(r)\propto r^{-4}\), and the Liouvillian gap closes as \(L^{-5}\) rather than the generic \(L^{-3}\) [0805.2878]. This is not bipartite entanglement of a pure state but a structural transition in the operator-state representation of a NESS.

In a boundary-driven transverse-field Ising chain coupled to magnetic reservoirs, non-equilibrium bias induces a mixed-order transition. At the threshold \(m_1=2(1-h)\), the magnetic order parameter drops discontinuously to zero while the correlation length diverges as \(\xi\propto |m-m_1|^{-1/2}\) in the generic case. In the conducting phase, the block entropy of the NESS takes the form
\[
S(\ell)=l_0 \ell + c_0 \ln \ell + c_1,
\]
with \(c_0\neq 0\) only when the chain conducts. The logarithmic term is bias dependent and differs from the equilibrium Ising result \(c/3=1/6\), so the current-carrying NESS has an entanglement structure with no direct equilibrium CFT analog [1809.07219].

Finally, in a coherently coupled dimer subject to pure dephasing, non-Markovianity can act as the control parameter for steady-state entanglement. Keeping the effective dephasing strength fixed while varying the memory time, the Markovian limit governed by a Lindblad equation yields a separable steady state, whereas non-Markovian dynamics generated by localized damped modes can stabilize an entangled steady state, approaching one e-bit near resonance. The paper explicitly notes that no non-analytic critical scaling is reported, so this is best understood as a non-Markovianity-controlled steady-state entanglement transition or crossover rather than a thermodynamic critical point [1106.2841].

## 6. Non-Hermitian steady states, exceptional points, and unifying themes

Non-Hermitian systems provide a distinct mechanism: the long-time normalized state
\[
|\phi_{\mathrm{ss}}\rangle=\lim_{t\to\infty}\frac{e^{-iHt}|\psi(0)\rangle}{\|e^{-iHt}|\psi(0)\rangle\|}
\]
is selected by the right eigenvector with maximal imaginary part. In \(\mathcal{PT}\)-symmetric models realizing the Yang–Lee edge singularity, the imaginary field drives a critical point at which the lowest two real-energy levels coalesce and then split into a complex conjugate pair. At a larger field \(h_{\mathrm{EE}}>h_{\mathrm{YL}}\), a level crossing in the imaginary spectrum makes the dominant state switch from a highly excited volume-law eigenstate to the ordered ground state of the Hermitian part, producing a first-order steady-state entanglement transition,
\[
S_A\sim \alpha L_A \longrightarrow S_A\sim \mathrm{const}.
\]
This mechanism is demonstrated in the transverse-field Ising chain with imaginary fields, the spin-1 Blume–Capel model, and the three-state Potts model, and it maps onto a forced-measurement transition in a postselected circuit with \(h_z=\gamma/4\) [2101.04115].

A second non-Hermitian route is the non-Hermitian skin effect (NHSE). In the symplectic Hatano–Nelson chain,
\[
H(k)=-J\cos k+(i\gamma \sigma_z+\Delta \sigma_x)\sin k,
\]
the competition between unitary hopping and nonreciprocal dissipation produces a continuous entanglement transition at \(|\gamma|=|\Delta|\). Under open boundary conditions, \(|\gamma|<|\Delta|\) yields a volume-law steady state, \(|\gamma|=|\Delta|\) yields logarithmic entanglement with nonunitary-CFT scaling,
\[
S(L,l)=\frac{c}{6}\ln\!\left[\sin\!\left(\pi\frac{l}{L}\right)\right]+S_0,
\]
and \(|\gamma|>|\Delta|\) yields an area law. The transition is controlled by an exceptional point, with skin-mode localization length
\[
\xi_{\mathrm{sHN}}=\frac{J}{\sqrt{\gamma^2-\Delta^2}}\propto (|\gamma|-|\Delta|)^{-1/2},
\]
and critical power-law localization of the skin modes at \(|\gamma|=|\Delta|\). A notable feature is extreme boundary-condition sensitivity: the effective central charge is strongly parameter dependent under open boundaries, while under periodic boundaries the critical scaling is consistent with \(c\simeq 2\) [2206.05384].

Chaotic non-Hermitian spin chains add a third variant. In the non-Hermitian transverse-field Ising model with complex longitudinal field and in a non-Hermitian XX model with transverse field, the interaction term commutes with the non-Hermitian term, but not with the transverse field. The long-time state is again the right eigenvector with maximal imaginary part, and the relevant spectral quantity is the complex gap
\[
\Delta(\gamma)=\operatorname{Im}(\lambda_{\max})-\operatorname{Im}(\lambda_2).
\]
As the dissipation rate increases, both models display a gapless–gapped transition in the imaginary spectrum and a corresponding steady-state entanglement change from volume law to area law. In the NHTFI with \(\Omega>J\), the transition is tied to a Yang–Lee singularity of the ground state; in the NHXX model, the dominant state in the gapped phase remains excited. In both cases, \(\Delta(\gamma)\) is oscillatory in the gapless regime because repeated level crossings change the identity of the maximal-imaginary-part state, and this produces non-monotonic steady-state entanglement even within the volume-law phase [2511.10410].

Across these disparate realizations, several general lessons recur. First, “steady-state entanglement phase transition” is an umbrella notion, not a single field-theoretic fixed point. Second, the natural order parameter may be entanglement entropy, mutual information, mutual coherence, OSEE, entangling power, Schmidt-gap closure, symmetry-sector exchange in the entanglement Hamiltonian, or an overlap such as \(q=(1/N)\sum_i\langle \sigma_i\rangle^2\). Third, the entanglement singularity need not track local observables, and it may be invisible to the steady-state density matrix when the transition is defined at the trajectory level. Fourth, the transition need not be continuous: the literature contains continuous percolation-type cases, mixed-order symmetry-breaking cases, and explicitly first-order non-Hermitian examples. Finally, the most important misconception is that steady-state entanglement transitions are synonymous with the measurement-induced volume-law to area-law transition. The broader record instead includes area-law to area-law transitions, volume-law to volume-law transitions, finite universal entanglement signatures in mixed states, and entanglement criticality governed by percolation, spin-glass, Ising, exceptional-point, or Yang–Lee structures [2006.09748], [2005.03052], [2112.06939], [2410.05370], [1809.07219], [2206.05384].

Source: https://www.emergentmind.com/topics/steady-state-entanglement-phase-transition