Zero-Average Capacity in Communications

• Zero-average capacity is defined as the regime where channels, under specific input constraints or parameter combinations, yield vanishing information transmission despite nuanced operational behavior.
• In classical Poisson channels, stringent power constraints and dark current effects cause capacity to vanish at superlinear or doubly logarithmic rates, quantifying performance in photon-starved scenarios.
• Quantum channels exhibit superactivation, where combining individually zero-capacity channels results in a positive overall capacity, challenging traditional additivity principles.

Zero-average capacity concerns the ability (or lack thereof) of a communication channel to transmit information when averaged over certain input constraints or under specific combinations of channel parameters. In both classical and quantum settings, it captures regimes in which the capacity vanishes under prescribed circumstances (e.g., low power, composition of otherwise “useless” channels), yet exhibits surprisingly rich and sometimes counterintuitive behavior regarding information transmission, especially in quantum theory.

1. Classical Zero-Average Capacity: The Poisson Channel Paradigm

The discrete-time Poisson channel under stringent power constraints provides a foundational example of zero-average capacity in classical information theory. The model permits real, nonnegative input $X \in [0, \infty)$ and delivers output %%%%1%%%%, distributed as $Y \sim \mathrm{Poisson}(\lambda + X)$ where $\lambda \geq 0$ represents the “dark current.” The channel is subject to an average-power constraint $E[X] \leq E$ and, optionally, a peak constraint $X \leq A$.

As the average power $E \to 0$, the asymptotic channel capacity $C(\lambda, E, A)$ exhibits distinct scaling depending on the dark current regime:

• Zero or vanishing dark current ($\lambda = 0$ or $\lambda = c E$):

$C(\lambda, E, A) = E \log(1/E) + o(E \log(1/E))$

The capacity vanishes as $E \rightarrow 0$ but with a super-linear rate in $E$ (0810.3564).

• Constant nonzero dark current ($\lambda > 0$):
  • With finite peak constraint ($A < \infty$), the capacity per unit power exists:

  $$\lim_{E \to 0} \frac{C(\lambda, E, A)}{E} = (1+\lambda/A)\log(1+A/\lambda) - 1$$ - With no peak constraint ($A = \infty$), capacity decays as

  $C(\lambda, E, \infty) = \Theta(E \log \log(1/E))$

  with

  $$\frac{1}{2} \leq \liminf_{E\to 0} \frac{C(\lambda, E, \infty)}{E \log \log(1/E)} \leq \limsup_{E\to 0} \frac{C(\lambda, E, \infty)}{E \log \log(1/E)} \leq 2$$

  (0810.3564).

This establishes zero-average capacity asymptotically in power-limited Poisson channels as a robust operational characteristic, quantifying information limits in photon-starved or low-energy optical regimes.

2. Channel Combinations and Additivity: Contrasts Between Classical and Quantum

In classical Shannon theory, channel capacity is additive and convex. Specifically, composing two zero-capacity channels in parallel, or forming convex mixtures, invariably yields a combined capacity of zero. “Zero-average capacity” in this context is unambiguously tied to the statistical independence of input and output—the channel is useless for reliable communication, in isolation or aggregation (Smith et al., 2008).

In contrast, this strict additivity fails in the quantum regime. The quantum theory allows for the superactivation phenomenon: two individually zero-capacity quantum channels can, when used together, exhibit nonzero overall capacity.

3. Quantum Zero-Capacity Channels and Superactivation

A quantum channel $\mathcal{N}$ is specified as a completely positive, trace-preserving (CPTP) map on density matrices. Quantum capacity $Q(\mathcal{N})$ is defined through regularized coherent information:

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

where $$I_c(\rho, \mathcal{N}) = S(\mathcal{N}(\rho)) - S[(\mathrm{id}_R \otimes \mathcal{N})(|\psi_\rho\rangle\langle\psi_\rho|)]$$ and $S(\cdot)$ denotes von Neumann entropy (Smith et al., 2008).

Superactivation is the property that there exist channels $\mathcal{N}_1, \mathcal{N}_2$ with individually zero quantum capacity $Q(\mathcal{N}_1) = Q(\mathcal{N}_2) = 0$ but positive joint capacity:

$Q(\mathcal{N}_1 \otimes \mathcal{N}_2) > 0$

(Smith et al., 2008).

Superactivating Channel Constructions

The canonical construction demonstrating this effect employs:

• A private Horodecki channel $\mathcal{N}_H$ (such as a four-dimensional PPT channel $\mathcal{N}_H^{(4)}$), which has $Q(\mathcal{N}_H)=0$ but strictly positive private classical capacity $P(\mathcal{N}_H)>0$.
• The 50% erasure channel $A_e$ on a four-dimensional input, also with $Q(A_e)=0$.

Smith and Yard show explicitly that $Q(\mathcal{N}_H \otimes A_e) > 0$ by exhibiting input states that achieve strictly positive coherent information, thereby establishing the regularized capacity lower bound (Smith et al., 2008).

Relevant Table: Quantum Channel Properties

Channel	Quantum Capacity $Q$	Private Capacity $P$	Key Property
$\mathcal{N}_H$	0	$>0$	Zero $Q$, positive $P$ (private Horodecki)
$A_e$	0	$0$	50% erasure, symmetric channel
$\mathcal{N}_H \otimes A_e$	$>0$	$-$	Superactivation: zero $Q$ inputs, $Q>0$ joint

4. Mathematical Techniques and Proof Strategies

Classical Poisson channel analysis leverages:

• Two-mass-point (binary) input ensembles to bound mutual information from below.
• Relative entropy (Donsker–Varadhan duality) upper bounds with carefully selected output measures.
• Asymptotic expansions, Taylor series, and Poisson-tail Chernoff bounds.

In the quantum setting, superactivation proofs depend on:

• The relationship between private classical and quantum capacities, particularly $Q_A(\mathcal{N}) \geq \frac{1}{2} P(\mathcal{N})$ for the “assisted” capacity with symmetric channel assistance.
• Explicit construction of input states, with purifications ensuring nontrivial correlations between joint channel outputs (Smith et al., 2008).

5. Conceptual and Operational Implications

Zero-average capacity regimes are pivotal in understanding fundamental limitations of communication, especially in systems constrained by resources such as energy or in the presence of strong noise or decoherence sources. The classical Poisson results quantify the “photon-starved” regime, where capacity vanishes with energy but at a superlinear or doubly logarithmic rate, providing guidance for system design in ultralow-power optical communications (0810.3564).

Quantum superactivation overturns long-standing classical principles of additivity and convexity of capacity. In particular, channels previously considered entirely useless for quantum communication can, when appropriately combined, unlock the potential for error-corrected quantum information transfer. This undermines the sufficiency of capacity as a singular operational metric and motivates reconsideration of channel resources in a broader, multidimensional sense (Smith et al., 2008).

6. Broader Impact and Future Directions

The phenomena delineated by zero-average capacity, especially in quantum communication, highlight the limitations of existing channel categorizations. The discovery of superactivation exposes the necessity for richer operational classes of channels, further regularization techniques, and systematic exploration of multipartite and assisted capacities. Open questions include the generality of superactivation, its relationship to classical and private capacity trade-offs, and the structural sources of such exotic behavior in quantum information theory (Smith et al., 2008).

A plausible implication is that channel utility in quantum networks is context-dependent, dictated not solely by individual metrics but by their interplay with other communication resources and network protocols. This suggests new architectural principles for both classical and quantum information systems operating near their absolute physical limits.