Superactivation of Quantum Capacity

• Superactivation of quantum capacity is the phenomenon where two individually zero-capacity channels combine to yield a positive joint capacity, demonstrating extreme nonadditivity.
• The mechanism involves pairing a PPT (private-state) channel with an antidegradable channel, utilizing coherent information and structural properties to recover quantum information.
• Both theoretical analysis and experimental approaches, including Gaussian bosonic and finite-blocklength studies, validate superactivation as a key concept in quantum communication.

Superactivation of quantum capacity is the phenomenon in which two quantum channels with individual quantum capacities $Q(\Phi)=Q(\Psi)=0$ nevertheless satisfy $Q(\Phi\otimes\Psi)>0$. In the broader activation effect, only one factor has zero capacity and the joint channel increases the capacity of the other, $Q(\Phi\otimes\Psi)>Q(\Phi)$ when $Q(\Psi)=0$ (Lim et al., 2019). Because quantum capacity is defined by the regularized coherent information,

$$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$

with one-shot coherent information $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$, superactivation is an extreme manifestation of nonadditivity. It has been studied in discrete-variable, Gaussian bosonic, zero-error, finite-blocklength, and experimentally motivated settings, and several works explicitly describe it as having no classical analogue (Lim et al., 2019, Parentin et al., 29 Apr 2026).

1. Capacity formalism and the meaning of superactivation

The standard capacity formula used throughout the literature is the Lloyd–Shor–Devetak expression

$$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$

or equivalently the regularized maximum of coherent information over blocklengths (Bhargava et al., 9 Jun 2026, Gyongyosi et al., 2012). In practical analyses, one often uses the single-letter quantity $\max_\rho I_c(\rho,\mathcal N)$ as a lower bound, while upper bounds come from data-processing, teleportation-simulation, or semidefinite-programming arguments (Lim et al., 2019, Wang et al., 2016).

Within this framework, activation and superactivation are distinct. Activation occurs when a zero-capacity channel $\Psi$ increases the capacity of another channel $\Phi$, so that $Q(\Phi\otimes\Psi)>0$0. Superactivation is the more restrictive case

$Q(\Phi\otimes\Psi)>0$1

so that two individually useless channels jointly transmit quantum information (Lim et al., 2019). Several works emphasize that this behavior is a strong form of nonadditivity: the joint channel can have positive coherent information even when each factor has vanishing capacity and vanishing one-shot coherent information (Oppenheim, 2010, Brandão et al., 2011).

Two standard zero-capacity mechanisms recur throughout the subject. A channel can be PPT or entanglement-binding, meaning its Choi state is positive under partial transpose and it cannot carry distillable entanglement; or it can be antidegradable, meaning its complementary output can simulate the receiver’s output, which also forces $Q(\Phi\otimes\Psi)>0$2 (Smith et al., 2011, Brandão et al., 2011). The first explicit superactivation constructions paired channels from these two classes.

2. The original discrete-variable mechanism

The first explicit example described in the supplied literature uses a PPT “private-dit” channel $Q(\Phi\otimes\Psi)>0$3 and the $Q(\Phi\otimes\Psi)>0$4 erasure channel $Q(\Phi\otimes\Psi)>0$5, each of zero quantum capacity, yet with positive joint capacity (Brandão et al., 2011). The private channel produces a twisted private state

$Q(\Phi\otimes\Psi)>0$6

where $Q(\Phi\otimes\Psi)>0$7 is maximally entangled and $Q(\Phi\otimes\Psi)>0$8 is a shield system. The erasure channel is antidegradable for $Q(\Phi\otimes\Psi)>0$9, so $Q(\Phi\otimes\Psi)>Q(\Phi)$0 in precisely the regime used for superactivation (Brandão et al., 2011, Brandão et al., 2010).

The operational mechanism is that the private channel emits a state whose entanglement is hidden by the shield, while the erasure channel sometimes transmits the missing shield subsystem to the receiver. When the shield arrives, the receiver can apply the inverse twisting unitary and recover a maximally entangled state; when erasure occurs, the receiver knows it from the erasure flag (Oppenheim, 2010, Brandão et al., 2010). For the erasure family this yields the lower bound

$Q(\Phi\otimes\Psi)>Q(\Phi)$1

for every $Q(\Phi\otimes\Psi)>Q(\Phi)$2, so the joint capacity is strictly positive throughout the antidegradable regime of the erasure channel (Brandão et al., 2011).

The same line of work also shows that the $Q(\Phi\otimes\Psi)>Q(\Phi)$3 erasure channel is not unique. Superactivation can be obtained with a wide class of inequivalent erasure channels, and the paper explicitly states that none of these channels can simulate each other. It further gives examples with anti-degradable depolarizing channels, again using the private-state/shield mechanism (Brandão et al., 2011).

A complementary interpretation was developed through symmetric-side and erasure assistance. In that formulation, all known single-copy superactivation protocols with the erasure channel can be understood as converting mutual independence or weak mutual independence into distillable entanglement. The erasure channel acts as the assisting zero-capacity resource that completes the conversion from private correlation to EPR pairs (Brandão et al., 2010). A later private-state perspective investigates channels whose Choi–Jamiolkowski operators are private states, and states that information encoded in the shield system that would otherwise leak to the environment can be recycled when paired with an assisting channel; that work also gives an alternative proof of superactivation for approximate private channels and extends its validity to a broader parameter regime (Wu et al., 6 Oct 2025).

3. Gaussian and bosonic superactivation

Superactivation was subsequently extended to Gaussian bosonic channels. A Gaussian channel acts on covariance matrices as

$Q(\Phi\otimes\Psi)>Q(\Phi)$4

with complete positivity requiring $Q(\Phi\otimes\Psi)>Q(\Phi)$5 (Smith et al., 2011, Lercher et al., 2012). For single-mode phase-insensitive Gaussian channels, $Q(\Phi\otimes\Psi)>Q(\Phi)$6 and $Q(\Phi\otimes\Psi)>Q(\Phi)$7 are proportional to the identity, and the main physical families are thermal attenuators $Q(\Phi\otimes\Psi)>Q(\Phi)$8, thermal amplifiers $Q(\Phi\otimes\Psi)>Q(\Phi)$9, and additive-noise channels (Lim et al., 2019).

The first Gaussian bosonic construction paired a two-mode PPT Gaussian channel with a single-mode $Q(\Psi)=0$0 attenuator. The attenuator is antidegradable for loss at least $Q(\Psi)=0$1, hence has zero quantum capacity, while the PPT Gaussian channel is entanglement-binding and also has zero capacity. Using an entangled three-mode Gaussian input and evaluating coherent information from the Williamson symplectic spectrum, the combined channel $Q(\Psi)=0$2 was shown to have strictly positive coherent information, establishing superactivation in an optical continuous-variable setting (Smith et al., 2011).

Lim, Takagi, Adesso, and Lee broadened this analysis to all single-mode phase-insensitive Gaussian channels assisted by a two-mode PPT channel (Lim et al., 2019). For the combined channel

$Q(\Psi)=0$3

they optimize coherent information over a three-mode Gaussian input and compare it with the best known upper bound $Q(\Psi)=0$4 for the thermal attenuator. Superactivation is witnessed whenever

$Q(\Psi)=0$5

(Lim et al., 2019).

The main conclusion is that superactivation is not confined to the beam-splitter point $Q(\Psi)=0$6. The positive activation gap extends over a broad region of the $Q(\Psi)=0$7 plane, including thermal attenuators with quite low transmissivity. The numerical thresholds extracted in that study state that for $Q(\Psi)=0$8 in the range $Q(\Psi)=0$9, superactivation already appears at transmissivity as small as $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$0, and over a wide strip the minimal transmissivity stays below $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$1 (Lim et al., 2019). The same work also reports that for $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$2, adding thermal noise can induce activation where the quantum-limited channel alone fails to activate.

This Gaussian literature also clarifies the resource requirements. Lercher, Giedke, and Wolf prove that for gauge-covariant Gaussian channels, PPT implies entanglement-breaking; hence such channels cannot transmit quantum information and cannot participate in standard superactivation. They also construct a counterexample outside the gauge-covariant class, using passive interactions with a squeezed environment, and explicitly conclude that standard super-activation for Gaussian channels requires squeezing (Lercher et al., 2012). A plausible implication is that squeezing is not merely a convenient implementation detail but a structural separator between PPT channels that are already entanglement-breaking and PPT channels that can still contribute to superactivation.

4. Structural constraints, no-go theorems, and capacity bounds

A recurrent misconception is that any zero-capacity channel might become useful under tensoring. The supplied papers draw a much sharper boundary. In finite dimensions, if $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$3 is entanglement-breaking and $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$4 is arbitrary, then $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$5, so $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$6; therefore entanglement-breaking channels cannot be superactivated. Lim et al. extend this statement to infinite-dimensional bosonic entanglement-breaking channels under finite-energy constraints, proving

$$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$7

for all finite energy bounds $$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$8 (Lim et al., 2019). In the Gaussian setting, this gives a no-go theorem for physically relevant bounded-energy inputs.

The same boundary appears in a different form in the Gaussian PPT literature. For gauge-covariant channels, the implication

$$Q(\mathcal N)=\lim_{n\to\infty}\frac{1}{n}\sup_{\rho_n} I_c(\mathcal N^{\otimes n},\rho_n),$$9

means that no superactivation is possible inside that class, because the would-be PPT ingredient is already EB (Lercher et al., 2012). This isolates a structural obstruction: PPT is not by itself sufficient for superactivation; what matters is a PPT zero-capacity channel that is not entanglement-breaking.

Several works formulate superactivation in terms of quantum relative entropy. Gyongyosi and Imre write the one-shot joint capacity as a relative-entropy optimization over Bob and environment outputs, and state that superactivation occurs exactly when the joint relative entropy does not factorize into single-channel contributions. In their formulation,

$$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$0

is equivalent to non-factorizability of the relevant relative-entropy expression and to entanglement of the joint optimal and average states (Gyongyosi et al., 2012). This suggests a structural criterion: product-state optimization geometry is incompatible with superactivation.

Upper-bounding the superactivated capacity is itself nontrivial. Wang and Duan introduce the additive semidefinite-programming upper bound $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$1, show

$$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$2

and explicitly state that $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$3 can be used to bound the super-activation of quantum capacity (Wang et al., 2016). Because $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$4 is single-letter and additive, it gives explicit nontrivial upper limits even when the actual capacity is known only through regularized or lower-bound arguments.

5. Geometric and algorithmic perspectives

A separate strand of work proposes an information-geometric description. In a PhD thesis, Gyongyosi defines the smallest enclosing quantum informational ball

$$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$5

with radius $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$6, and introduces a “superball” for the joint output set of $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$7 (Gyongyosi, 2012). In that construction, the radius of the smallest enclosing ball in the enlarged output space equals $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$8. The key theorem in that thesis states that if the center of the optimal joint ball does not factor as $$I_c(\mathcal N,\rho)=S[\mathcal N(\rho)]-S[\mathcal N^c(\rho)]$$9, then the radius is positive and the joint one-shot capacity is positive.

The same thesis proposes an algorithmic search procedure for superactive pairs. Candidate channel families are scanned, output core-sets are precomputed, the joint core-set is formed, and an approximate minimax ball is found via iterative farthest-point updates analogous to Bădoiu–Clarkson, with $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$0 iterations for a $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$1 approximation (Gyongyosi, 2012). Within that framework, the Horodecki channel plus the $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$2 erasure channel is treated as a case study, with

$$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$3

and hence $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$4 bit/use (Gyongyosi, 2012).

These geometric formulations are not replacements for the standard coherent-information definition. Rather, they recast the same phenomenon in optimization language and provide algorithmic heuristics for locating superactive channel pairs. This suggests that superactivation can be studied not only as an existence theorem but also as a search problem over structured channel families.

6. Non-asymptotic and experimental developments

For many years superactivation was primarily discussed as an asymptotic capacity effect. A 2026 study introduces a finite-blocklength notion using the $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$5-shot channel fidelity

$$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$6

and the corresponding $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$7-shot $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$8-error quantum capacity $$Q(\mathcal N)=\lim_{n\to\infty}\frac1n I_c(\mathcal N^{\otimes n}), \qquad I_c(\rho,\mathcal N)=S[\mathcal N(\rho)]-S[(id_R\otimes\mathcal N)(|\Psi\rangle\langle\Psi|)],$$9 (Parentin et al., 29 Apr 2026). In that framework, two channels exhibit $\max_\rho I_c(\rho,\mathcal N)$0-shot superactivation if neither channel alone can surpass fidelity $\max_\rho I_c(\rho,\mathcal N)$1 for a target dimension $\max_\rho I_c(\rho,\mathcal N)$2, regardless of how many uses are permitted, but the joint channel can exceed that fidelity after $\max_\rho I_c(\rho,\mathcal N)$3 tandem uses.

Focusing on the original pair—a private-Horodecki PPT channel $\max_\rho I_c(\rho,\mathcal N)$4 and the $\max_\rho I_c(\rho,\mathcal N)$5 erasure channel $\max_\rho I_c(\rho,\mathcal N)$6—the study combines second- and third-order blocklength bounds with a direct “symmetric-seesaw” SDP optimization that exploits permutation symmetry, Schur–Weyl reduction, flagged structure, and sparsity (Parentin et al., 29 Apr 2026). The headline result is that for $\max_\rho I_c(\rho,\mathcal N)$7 and $\max_\rho I_c(\rho,\mathcal N)$8,

$\max_\rho I_c(\rho,\mathcal N)$9

yet

$\Psi$0

Thus, as few as $\Psi$1 uses of the joint channel already produce a fidelity unattainable by any number of uses of either constituent alone (Parentin et al., 29 Apr 2026).

A distinct operational route is provided by coherent superposition of channels rather than tensoring them. A 2026 NMR experiment superposes the Stinespring dilation unitaries of two single-qubit depolarizing channels $\Psi$2 and $\Psi$3, both in the entanglement-breaking regime $\Psi$4, and shows that the effective channel is again depolarizing with strength $\Psi$5 (Bhargava et al., 9 Jun 2026). For a maximally entangled input, the one-shot coherent information becomes

$\Psi$6

and is strictly positive whenever $\Psi$7 falls below the threshold $\Psi$8 (Bhargava et al., 9 Jun 2026). The paper states that for $\Psi$9 and $\Phi$0, there is a window $\Phi$1 with $\Phi$2, and at $\Phi$3 one even gets $\Phi$4 and $\Phi$5. Experimentally, the superposed depolarizing channel was implemented on a five-qubit NMR register, with postselection probability on the order of $\Phi$6–$\Phi$7 in the superactivation window and Bloch-vector attenuation consistent with theory within $\Phi$8–$\Phi$9 in most cases (Bhargava et al., 9 Jun 2026). This is not the standard tensor-product notion of superactivation, but it shows that closely related capacity-recovery effects can now be realized in the laboratory.

7. Related notions: zero-error and multipartite superactivation

A substantial related literature concerns one-shot zero-error quantum capacity, usually denoted $Q(\Phi\otimes\Psi)>0$00 or $Q(\Phi\otimes\Psi)>0$01, rather than the regularized unassisted quantum capacity $Q(\Phi\otimes\Psi)>0$02. In this setting, superactivation means that a channel has no noiseless subspace individually but develops one under tensoring. Shirokov and Shulman construct a channel $Q(\Phi\otimes\Psi)>0$03 with $Q(\Phi\otimes\Psi)>0$04 and $Q(\Phi\otimes\Psi)>0$05 such that $Q(\Phi\otimes\Psi)>0$06 but $Q(\Phi\otimes\Psi)>0$07, and they further show the existence of channels $Q(\Phi\otimes\Psi)>0$08 with $Q(\Phi\otimes\Psi)>0$09 yet $Q(\Phi\otimes\Psi)>0$10 for arbitrary $Q(\Phi\otimes\Psi)>0$11 (Shirokov et al., 2013). The same work reformulates the phenomenon in measurement-theoretic language as the appearance of an indistinguishable subspace for tensor products of observables that individually have none.

This zero-error phenomenon also admits genuine multipartite forms. Shirokov constructs an explicit $Q(\Phi\otimes\Psi)>0$12 example for one-shot zero-error quantum capacity in which channels $Q(\Phi\otimes\Psi)>0$13 satisfy

$Q(\Phi\otimes\Psi)>0$14

while every proper subset has zero capacity, provided $Q(\Phi\otimes\Psi)>0$15 (Shirokov, 2014). The corresponding two-dimensional code is explicitly spanned by

$Q(\Phi\otimes\Psi)>0$16

(Shirokov, 2014).

These zero-error results are not statements about the regularized quantum capacity $Q(\Phi\otimes\Psi)>0$17. Their significance is conceptual: they show that superactivation is not confined to a single capacity notion, and that the tensor-product emergence of transmissible subspaces is a broader operator-algebraic feature of quantum channels and measurements (Shirokov et al., 2013, Shirokov, 2014).

Taken together, the literature presents superactivation of quantum capacity as a sharply delimited but robust nonadditivity phenomenon. The original private-channel plus erasure construction established the effect in finite dimensions; Gaussian bosonic analyses showed that it can occur in realistic optical models; no-go theorems identified entanglement-breaking channels as a boundary; relative-entropy, geometric, and SDP methods supplied structural and computational tools; and recent non-asymptotic and experimental works moved the subject beyond purely asymptotic existence results (Brandão et al., 2011, Lim et al., 2019, Wang et al., 2016, Parentin et al., 29 Apr 2026, Bhargava et al., 9 Jun 2026).