---
title: Entanglement-Breaking Quantum Channels
url: https://www.emergentmind.com/topics/entanglement-breaking-quantum-channels
type: topic
---

# Entanglement-Breaking Quantum Channels

Entanglement-breaking quantum channels are completely positive trace-preserving maps that destroy all output entanglement with any ancillary system. In finite dimensions, a channel $\Phi$ is entanglement breaking precisely when $(\operatorname{id}\otimes \Phi)(\rho)$ is separable for every bipartite input state $\rho$; equivalently, $\Phi$ has a measure-and-prepare form, its Choi matrix is separable, and it admits a Kraus decomposition with rank-one Kraus operators [2109.01340]. In operator-algebraic formulations, the same idea extends to channels between $C^\ast$-algebras and von Neumann algebras, while in infinite dimensions a stronger notion, strongly entanglement breaking, requires countably separable outputs for every ancillary extension [1806.05854] [2410.19270]. The subject now connects structural channel theory, Perron–Frobenius analysis, complementary channels, broadcasting, temporal memory, and dynamical thresholds for the onset of entanglement destruction.

## 1. Fundamental definitions and equivalent formulations

For a $d$-dimensional input system, a quantum channel is a linear map $\Phi:M_d(\mathbb{C})\to M_{d'}(\mathbb{C})$ that is completely positive and trace-preserving. It admits a Kraus representation
$$
\Phi(\rho)=\sum_k V_k \rho V_k^\ast,\qquad \sum_k V_k^\ast V_k=I_d.
$$
The channel is entanglement breaking if $(\Phi\otimes \operatorname{id}_d)(X)$ is separable for every bipartite state $X\in M_d\otimes M_d$. Three equivalent finite-dimensional characterizations are standard: a Holevo or measure-and-prepare form
$$
\Phi(\rho)=\sum_{i=1}^m \operatorname{Tr}(F_i\rho)\,\sigma_i,
$$
with $\{F_i\}$ a POVM and $\{\sigma_i\}$ density operators; separability of the Choi matrix
$$
J(\Phi)=(\Phi\otimes \operatorname{id})(|\Omega\rangle\langle\Omega|);
$$
and existence of a Kraus decomposition with rank-one Kraus operators [2109.01340] [2204.01685] [2211.11909].

In Holevo form, the Choi matrix has the explicit separable decomposition
$$
J(\Phi)=\sum_{i=1}^m \tau(F_i)\otimes \sigma_i,
$$
where $\tau$ denotes matrix transpose in the computational basis. Conversely, any separable $J(\Phi)$ with the appropriate marginal constraints induces a measure-and-prepare representation [2109.01340]. This equivalence is also the basis of the standard statement that entanglement-breaking channels are antidegradable: the environment can simulate the output through a completely positive map because the output is determined by classical measurement data recorded in the environment [2204.01685].

The operator-algebraic formulation replaces matrix algebras by unital $C^\ast$-algebras. A channel $\Lambda:A\to A_{\mathrm{in}}$ is entanglement breaking if, for every $C^\ast$-algebra $B$ and every state $\omega\in S(A_{\mathrm{in}}\otimes B)$, the composed state $\omega\circ(\Lambda\otimes \operatorname{id}_B)$ is separable in the injective tensor product. In this setting, entanglement breaking is equivalent to factorization through a commutative algebra, to a POVM-based Holevo form with possibly continuous outcomes, to $n$-copy compatibility for all finite $n$, and to countable compatibility [1806.05854].

In infinite-dimensional systems, the distinction between separability and countable separability becomes nontrivial. A channel $\Phi:J(H)\to J(K)$ is strongly entanglement breaking if $(\operatorname{id}_R\otimes \Phi)(\rho)$ is countably separable for every ancillary Hilbert space $R$ and every state $\rho$. This is equivalent to a countable Holevo form
$$
\Phi(X)=\sum_{k=1}^{\infty}\operatorname{Tr}(F_k X)\,R_k,
$$
and also to a rank-one Kraus representation
$$
\Phi(X)=\sum_{k=1}^{\infty}E_k X E_k^\dagger,\qquad E_k=|u_k\rangle\langle v_k|,
$$
with $\sum_k |v_k\rangle\langle v_k|=I_H$ [2410.19270].

## 2. Classicalization, stochastic matrices, and operator-algebraic structure

A measure-and-prepare representation carries an intrinsic classical dynamics. If
$$
\Phi(\rho)=\sum_{k=1}^r \operatorname{Tr}(F_k\rho)\,R_k,
$$
then on the convex hull of the prepared states $\{R_k\}$ the coefficients evolve by a column-stochastic matrix
$$
S_{ij}=\operatorname{Tr}(F_i R_j),\qquad \sum_i S_{ij}=1.
$$
If $\rho=\sum_m c_m R_m$, then one application of $\Phi$ maps $c\mapsto Sc$ and
$$
\Phi\!\left(\sum_m c_m R_m\right)=\sum_k (Sc)_k R_k.
$$
Thus an entanglement-breaking channel induces a classical Markov evolution on the preparation simplex [2109.01340].

The induced matrix does more than record probabilities. Writing $A\in \mathbb{C}^{d^2\times r}$ with $k$-th column $\operatorname{vec}(R_k)$ and $B\in \mathbb{C}^{r\times d^2}$ with $k$-th row $\operatorname{vec}(F_k^T)^T$, one has
$$
[\Phi]=AB,\qquad S=BA.
$$
By Flanders’ theorem, $AB$ and $BA$ have the same Jordan form on the non-zero spectrum. Consequently, the non-zero spectrum of $\Phi$ coincides with that of $S$, including Jordan multiplicities [2109.01340]. This identifies a precise classical reduction for all non-zero spectral data of an entanglement-breaking channel.

The same representation controls primitivity. A channel is primitive if there exists $k$ such that $\Phi^k(X)\succ 0$ for every $X\ge 0$, $X\neq 0$; equivalently, it has a unique full-rank fixed point and trivial peripheral spectrum. For an entanglement-breaking channel in Holevo form, primitivity is equivalent to two conditions:
$$
S\ \text{primitive}\qquad\text{and}\qquad \sum_k R_k\succ 0.
$$
Under these hypotheses, if $p(S)$ is the primitivity index of the stochastic matrix and $q(\Phi)$ the primitivity index of the channel, then
$$
|q(\Phi)-p(S)|\le 1.
$$
If $r$ is the Holevo rank, Wielandt’s bound yields
$$
q(\Phi)\le r^2-2r+3.
$$
The examples in the paper show that the $\pm1$ slack is tight [2109.01340].

The fixed-point structure is equally transparent. If $Sv=v$, then
$$
\rho_\ast=\sum_{k=1}^r v_k R_k
$$
is a fixed point of $\Phi$. For primitive channels, the subleading spectral radius
$$
r_\ast=\max\{|\lambda|:\lambda\in \operatorname{spec}(S),\ \lambda\neq 1\}<1
$$
governs convergence for both the matrix and the channel [2109.01340]. A plausible implication is that many mixing-time and perturbative results for stochastic matrices should transfer to entanglement-breaking channels whenever the associated Holevo form is explicit.

The operator-algebraic generalization reaches a complementary conclusion from a different direction. An entanglement-breaking channel is precisely one that factors through a commutative algebra, and this factorization is equivalent to the existence of joint channels for arbitrarily many copies. In that sense, “classicalization” appears either as a stochastic-matrix reduction on the prepared-state simplex or as factorization through a commutative outcome algebra [1806.05854].

## 3. Complementary channels, PPT criteria, and entanglement-breaking rank

Complementary-channel methods supply further criteria for deciding when a channel is entanglement breaking. Given a Stinespring isometry $U:A\to B\otimes E$,
$$
\Lambda(\rho)=\operatorname{Tr}_E[U\rho U^\dagger],\qquad
\Lambda^c(\rho)=\operatorname{Tr}_B[U\rho U^\dagger].
$$
A channel is called PPT in the sense used in [2204.01685] when $\Lambda\circ T$ is completely positive, equivalently when $J_\Lambda^{T_{\mathrm{in}}}\ge 0$. The central structural result is that if both $\Lambda$ and one complementary channel $\Lambda^c$ are PPT, then both are entanglement breaking. More precisely, for complementary completely positive maps $(\Phi,\Psi)$ with $\Phi$ PPT, the following are equivalent for $\Psi$: being PPT, being entanglement breaking, and having undistillable Choi state. As a corollary, every degradable channel that stays completely positive under composition with transposition is entanglement breaking [2204.01685].

The proof uses the shared purification of complementary Choi operators and low-rank PPT criteria. If $J(\Phi)$ and $J(\Psi)$ arise from a pure tripartite state $|L\rangle\in A\otimes B\otimes C$, then PPT of one Choi operator induces rank inequalities on the other. In the regime
$$
\operatorname{rank} X\le \max\{\operatorname{rank}\operatorname{Tr}_A X,\operatorname{rank}\operatorname{Tr}_B X\},
$$
PPT, separability, and undistillability coincide [2204.01685]. This converts complementary-channel structure into an entanglement-breaking criterion.

A different complementary-channel method addresses the entanglement-breaking rank. For a channel $\Phi:M_d\to M_{d'}$, the entanglement-breaking rank $r_{EB}(\Phi)$ is the minimal number of terms in any rank-one Kraus decomposition; equivalently, it is the optimal ensemble length of the separable Choi state. If $\{K_i\}_{i=1}^d$ is a minimal Kraus set and $\Phi^C$ its complementary channel, then
$$
\Phi^{C\dagger}(E_{ij})=K_i^\ast K_j.
$$
The channel is entanglement breaking if and only if there exist vectors $\{w_i\}$ and $\{v_i\}$ such that
$$
\sum_i w_i w_i^\ast=I,\qquad \Phi^{C\dagger}(w_i w_i^\ast)=v_i v_i^\ast
$$
for all $i$. This yields a complementary-channel formula for $r_{EB}(\Phi)$ [2211.11909].

The multiplicative domain becomes decisive when the Choi matrix is a projection. In that case, $\Phi^{C\dagger}$ is unital and trace-preserving, and $\Phi$ is entanglement breaking if and only if
$$
\mathrm{MD}(\Phi^{C\dagger})\simeq \bigoplus_{k=1}^M M_{j_k}(\mathbb{C}),
$$
equivalently, up to unitary equivalence the multiplicative domain contains rank-one projections summing to the identity. In this projection-Choi class,
$$
r_{EB}(\Phi)=r_{Choi}(\Phi)=\operatorname{rank}J(\Phi).
$$
The same paper shows that an entanglement-breaking channel has projection Choi matrix exactly when it is internally or externally unitarily equivalent to the complement of a Schur product channel $X\mapsto X\circ C$, where $C$ is a correlation matrix [2211.11909].

## 4. Entanglement breaking and neighboring channel classes

Entanglement breaking is distinct from entanglement annihilation. If $\Phi_A$ acts on a multipartite subsystem $A$, then $\Phi_A$ is entanglement annihilating when $\Phi_A[S(H_A)]\subset \operatorname{Sep}(H_A)$, meaning it destroys all internal entanglement inside the acted-upon subsystem. By contrast, entanglement breaking requires that $(\Phi_A\otimes I_B)(\rho_{AB})$ be separable across $A|B$ for every ancilla $B$. The two notions do not contain one another. An entanglement-breaking channel need not be entanglement annihilating: the constant map to a fixed entangled state on $A$ is entanglement breaking but not entanglement annihilating. Conversely, entanglement annihilation does not imply entanglement breaking: for qubit depolarizing noise $\Lambda_\lambda$, the map $\Lambda_{1/\sqrt{3}}\otimes \Lambda_{1/\sqrt{3}}$ is entanglement annihilating on two qubits but not entanglement breaking as a channel on the two-qubit system with an ancilla [1006.2502].

Local entanglement annihilation also exhibits a strict hierarchy. If a single-party channel $\Lambda$ is $k$-locally entanglement annihilating, then it is $(k-1)$-locally entanglement annihilating, so the admissible set shrinks with the number of parties. By contrast, local entanglement breaking collapses:
$$
EB^1=2\text{-LEB}=3\text{-LEB}=\cdots.
$$
For the qubit depolarizing channel $\Lambda_\lambda(\rho)=\lambda \rho +(1-\lambda)I/2$, the exact thresholds are: entanglement breaking iff $\lambda\le 1/3$, $2$-locally entanglement annihilating iff $\lambda\le 1/\sqrt{3}$, and not $3$-locally entanglement annihilating for $\lambda>0.5567\ldots$ [1006.2502].

A higher-rank generalization replaces separability by Schmidt-number constraints. A channel is $k$-partially entanglement breaking if it cannot transmit Schmidt number greater than $k$. This is equivalent to $SN(J_\Phi)\le k$, and also to existence of a Kraus decomposition with $\operatorname{rank}(K_\alpha)\le k$ for all $\alpha$. Operationally, such channels are exactly those simulable by one-way LOCC from an entangled resource with Schmidt number $k$ [1307.2727]. Recent work reframes this as $r$-Schmidt-number-breaking channels, which satisfy
$$
SN[(\operatorname{id}\otimes S)(\rho)]\le r
$$
for all bipartite $\rho$; equivalently, $SN(C_S)\le r$ and the Kraus operators can be chosen with rank at most $r$. Entanglement breaking is the case $r=1$ [2411.19315].

The same 2024 analysis introduces Schmidt-number-annihilating channels, which reduce Schmidt number within a composite subsystem rather than across a system–reference cut. It proves, among other relations, that every $r$-Schmidt-number-breaking channel is $2$-local $r$-Schmidt-number-annihilating, while tensor products need not preserve the Schmidt-number-breaking property because Kraus ranks multiply [2411.19315].

A different generalization places the problem in the category of proper convex cones. There, a map is entanglement breaking when its Choi tensor lies in a minimal tensor product cone, and entanglement annihilating when all tensor powers map the maximal tensor product of one cone into the minimal tensor product of another. The main resilience theorem states that if either the input or output cone is a Lorentz cone, then every entanglement-annihilating map is already entanglement breaking [2110.11825]. In the PSD specialization, this recovers the qubit robustness associated with the isomorphism $PSD(\mathbb{C}^2)\simeq L_3$.

Quantum-correlation–breaking channels refine the picture further. Channels that break correlations down to the QC type are exactly measurement maps
$$
\Lambda(\rho)=\sum_i \operatorname{Tr}(E_i\rho)\,|e_i\rangle\langle e_i|,
$$
so they form a strict subclass of entanglement-breaking channels. In the CC case, a CC Choi state forces a commuting POVM structure but does not imply that all outputs are CC. This leads to an unexpected relation between local broadcasting and finite Markov chains, with Perron–Frobenius stationary distributions generating spectrum-broadcastable states [1208.2162].

## 5. Dynamical thresholds, breaking times, and amendment

For qubit channels, the Bloch-affine normal form gives explicit entanglement-breaking criteria. After unitary equivalence, any qubit channel has
$$
r'\,=\,\operatorname{diag}(\lambda_1,\lambda_2,\lambda_3)\,r+n.
$$
In the unital case $n=0$, the channel is entanglement breaking if and only if
$$
|\lambda_1|+|\lambda_2|+|\lambda_3|\le 1.
$$
This describes the octahedral entanglement-breaking region inside the cube of Pauli-diagonal contractions. In both the unital and non-unital cases, if at least one singular value $\lambda_i$ vanishes, the channel is entanglement breaking. When exactly two components of $n$ vanish, the paper gives a necessary-and-sufficient inequality involving the remaining $n_{i_3}$ and the corresponding $\lambda$’s [1906.01999].

These criteria recover standard thresholds. For the qubit depolarizing channel, entanglement breaking occurs at $p\ge 2/3$. For phase damping, only complete dephasing is entanglement breaking. For amplitude damping, only the full-relaxation limit is entanglement breaking [1906.01999]. The same work studies time-dependent channels and obtains explicit entanglement-breaking transitions for depolarization and quantum homogenization.

A complementary dynamical viewpoint treats the entanglement-breaking time of a Lindblad semigroup $\Phi_t=e^{t\mathcal{L}}$ through entanglement witnesses and quantum-speed-limit bounds. For a witness $W$ and input $\rho$, let
$$
w_\rho(t)=\operatorname{Tr}[W(\operatorname{id}\otimes \Phi_t)(\rho)].
$$
The configuration-breaking time is the first $t$ for which $w_\rho(t)\ge 0$, and the channel’s entanglement-breaking time is the maximum over inputs and witnesses. A general Mandelstam–Tamm-type lower bound is
$$
t_{CB}\ge \frac{|w_\rho(0)|}{\sum_\alpha r_\alpha |\lambda_\alpha|},
$$
where the $\lambda_\alpha$ are eigenvalues of $\mathcal{L}^\ast$ and the $r_\alpha$ depend on the chosen configuration. For a canonical symmetric configuration based on the maximally entangled state and the witness
$$
W_{\Psi_+}=I^{\otimes 2}-d|\Psi_+\rangle\langle \Psi_+|,
$$
the resulting lower bounds depend only on the spectrum of the generator [2209.08689].

In the qubit example with one pure decay mode and one oscillatory pair, the paper gives the closed-form dynamics-only bounds
$$
T_{M\text{--}T}=\frac{2}{\gamma_\parallel+2\sqrt{\omega^2+\gamma_\perp^2}},
$$
and
$$
T_{M\text{--}L}=\frac{\pi-2}{(\pi/2)\gamma_\parallel+2\omega+(\pi-2)\gamma_\perp},
$$
together with a stronger “good-configuration” bound obtained from a transcendental fixed-point equation [2209.08689]. These are lower bounds on when entanglement breaking can first occur under the semigroup.

Entanglement-breaking behavior can also emerge through concatenation even when each segment is individually non-entanglement-breaking. This is the setting of amendment by intermediate unitary operations. The optical experiment in [1707.00161] studies repeated applications of rotated phase-damping and rotated amplitude-damping maps and shows that a suitable unitary filter inserted between two noisy segments can restore entanglement transmission. Crucially, the restoration concerns a concatenation $\Phi\circ\Phi$ that becomes entanglement breaking although each $\Phi$ is not. A genuinely entanglement-breaking channel cannot be “unbroken” by unitary pre- or post-processing, because entanglement breaking is closed under unitary composition [1707.00161].

The paper introduces strongly entanglement-breaking channels as entanglement-breaking channels that cannot be amended by any local unitary interleaving between repeated uses of the same elementary map. A sufficient condition is again rank deficiency: if a qubit channel has at least one vanishing singular value, then it is strongly entanglement breaking [1906.01999].

## 6. Operational subtleties: experiments, temporal memory, and network effects

Spatially, entanglement-breaking channels destroy entanglement. Operationally, however, that statement does not exhaust their uses. The experiment reported in “Entanglement’s Benefit Survives an Entanglement-Breaking Channel” shows that an entanglement-based protocol for quantum illumination can retain a decisive advantage even when the propagation channel is entanglement breaking in the sense of the returned–retained Gaussian state. The entanglement-breaking threshold is expressed as
$$
N_B\ge N_B^{\mathrm{thresh}}=\kappa_1\kappa_A\kappa_B G_B,
$$
and the reported operating point had measured noise $N_B=1.46\times 10^4$ per mode, threshold $N_B^{\mathrm{thresh}}=2.14\times 10^3$ per mode, and therefore $8.3$ dB beyond threshold. At a secure operating point with $N_S=7.81\times 10^{-4}$ and $G_A-1=2.48\times 10^{-5}$, the measured error rates were
$$
BER_A=1.78\times 10^{-6},\qquad BER_E\approx 0.5,
$$
and the lower bound on the information advantage peaked at approximately $0.8$ bits per bit for the stated tap settings [1303.5343]. The paper’s interpretation is not that the channel fails to be entanglement breaking, but that residual classical correlations generated by an entangled transmitter remain operationally useful.

Temporal scenarios are subtler still. In a single-system multi-time setting, an entanglement-breaking channel inserted between repeated uses of a quantum instrument induces an effective classical finite-state machine with $m$ states, where $m$ is the number of measure-and-prepare outcomes:
$$
[T_a]_{ij}=\operatorname{Tr}[I_a(\sigma_i)E_j].
$$
Because one may have $m>d$ for a $d$-dimensional quantum system, a qudit passing through an entanglement-breaking channel can outperform a classical $d$-state memory on memory-based output-generation tasks. For one-tick sequences, the paper gives explicit violations of the optimal classical $d$-state bound: for $L=4$, $d=3$, the classical optimum is $0.31640625$ while an entanglement-breaking-assisted quantum strategy achieves at least $0.359523$; for $L=5$, $d=4$, the classical optimum is $0.32768$ while the quantum strategy achieves at least $0.368445$ [2402.16789]. This directly contradicts the unrestricted identification of entanglement-breaking channels with “classical memory” in temporal settings.

Networked environments create a different kind of exception. A pair of individually entanglement-breaking channels can, when embedded into a correlated-noise environment, reactivate entanglement distribution. For qudits, correlated twirling channels of the form
$$
\Phi_{\mathrm{corr}}(\rho_{AB})=\int dU\,(U\otimes V)\rho_{AB}(U^\dagger\otimes V^\dagger),
$$
with $V=U$ or $V=U^\ast$, preserve Werner or isotropic states even though each single use is entanglement breaking. In continuous variables, anti-correlated phase rotations preserve EPR states. The environment implementing the correlation can be separable and of zero discord, so the reactivation is driven by purely classical correlations in the environment [1307.5699].

A more recent network effect is super-activation of quantum memory by entanglement-breaking channels. The 2024 construction exhibits two compatible qubit entanglement-breaking margins $E_{A\to B}$ and $E_{A\to C}$ such that every broadcasting realization $N_{A\to BC}$ necessarily outputs an entangled state on $BC$ from the maximally mixed input $I_A/2$. A fixed local filter on $B$ followed by a Bell projection on $AB$ then induces a channel $M_{A\to C}^{(N)}$ whose Choi state is entangled, hence $M_{A\to C}^{(N)}$ is not entanglement breaking [2410.13499]. This is a genuine activation phenomenon: no single margin preserves entanglement, but the compatible broadcasting architecture does.

These results clarify a common misconception. Entanglement breaking is a statement about a channel’s action on spatial entanglement across a system–reference cut. It is not, by itself, a complete statement about temporal memory, correlated-noise architectures, or broadcast networks. The distinction is also visible in bipartite channel theory: no-signaling and entanglement breaking are logically independent. There exist no-signaling bipartite channels that are neither entanglement breaking nor localizable, and the convex hull of entanglement-breaking and localizable channels is strictly smaller than the full no-signaling set [1007.1177].

## 7. Open problems and current directions

Several open problems now organize the modern theory. For stochastic-matrix representations, the outstanding questions include whether there is a meaningful topology under which nearby entanglement-breaking channels admit nearby stochastic representations, which column-stochastic matrices can arise from a given channel, and when the primitivity index satisfies equality rather than the general bound
$$
|q(\Phi)-p(S)|\le 1.
$$
Another active direction is refinement of the Holevo-rank bound
$$
q(\Phi)\le r^2-2r+3
$$
inside constrained families [2109.01340].

In the operator-algebraic setting, entanglement breaking is equivalent to compatibility of all finite and countably infinite copies, but extending these equivalences beyond the injective $C^\ast$-tensor product framework raises the problem of tensor-product nonuniqueness in general probabilistic theories [1806.05854]. In infinite dimensions, strongly entanglement-breaking channels admit countable measure-and-prepare descriptions and channels with commutative range are strongly entanglement breaking, but the distinction between entanglement breaking and strong entanglement breaking still invites finer analysis [2410.19270].

For complementary-channel methods, the projection-Choi class is now well understood, yet outside that class the equality $r_{EB}=r_{Choi}$ can fail, as illustrated by the Werner–Holevo channel. Identifying larger classes in which multiplicative-domain methods determine the entanglement-breaking rank exactly remains open [2211.11909].

The Schmidt-number program raises analogous questions. The 2024 study of Schmidt-number-breaking and Schmidt-number-annihilating channels explicitly calls for Choi- or Kraus-type structural characterizations of annihilating channels, efficient criteria beyond positive-map tests, and a systematic treatment of closure properties and capacities [2411.19315]. In the cone-theoretic framework, Lorentz cones are resilient, but the boundary of resilience outside the Euclidean setting remains a central structural problem [2110.11825].

A plausible synthesis of these directions is that entanglement breaking is no longer a terminal classification. It is a node in a larger hierarchy linking separability, Schmidt-number reduction, classical compatibility, broadcasting, primitivity, and network activation. The recent literature sharpens that hierarchy rather than collapsing it [2109.01340] [2402.16789]

Source: https://www.emergentmind.com/topics/entanglement-breaking-quantum-channels