---
title: Approximate Private Channels
url: https://www.emergentmind.com/topics/approximate-private-channels
type: topic
---

# Approximate Private Channels

Searching arXiv for relevant papers on approximate private quantum channels, fermionic Gaussian systems, and related private-capacity frameworks.
Approximate private channels are quantum channels that replace exact output randomization by a controlled approximation to the maximally mixed state. In the finite-dimensional formulation, a private quantum channel (PQC) maps every input state to $I_d/d$, while an $\varepsilon$-private quantum channel requires the output to be within a prescribed Schatten $p$-norm distance of that state. In fermionic Gaussian systems, this idea becomes the fermionic $\varepsilon$-private quantum channel, or $\varepsilon$-FPQC: a randomizing channel on $M$-mode fermionic Gaussian states that is close to maximally mixed rather than exactly maximally mixed. The central construction in the fermionic setting uses Majorana-based random unitaries and shows that approximate privacy can be achieved with substantially fewer unitaries than exact privacy, reducing the required cardinality from about $4M^2$ in the exact case to $O(M\log M)$ in the approximate regime [2003.11907].

## 1. Exact and approximate privacy criteria

A standard PQC is defined by the exact relation
\[
\Lambda(\varrho)=\frac{I_d}{d}.
\]
This is the quantum analogue of perfect one-time-pad secrecy: all inputs are rendered indistinguishable at the output. The approximate version replaces equality by norm control. For an operator $A$, the Schatten $p$-norm is
\[
\|A\|_p=\left[\operatorname{Tr}\big((A^\dagger A)^{p/2}\big)\right]^{1/p}, \qquad 1\le p\le \infty,
\]
and a channel $\Lambda:\mathcal{B}(\mathbb{C}^d)\to\mathcal{B}(\mathbb{C}^d)$ is an $\varepsilon$-PQC if, for every state $\varrho$,
\[
\left\|\Lambda(\varrho)-\frac{I_d}{d}\right\|_p \le \frac{\varepsilon}{d^{(p-1)/p}}.
\]
In the trace-norm case $p=1$, this reduces to
\[
\left\|\Lambda(\varrho)-\frac{I_d}{d}\right\|_1\le \varepsilon.
\]
The fermionic paper adopts the same pattern for $M$-mode fermionic Gaussian states and calls the resulting object an $\varepsilon$-FPQC [2003.11907].

This definition places approximate private channels in the family of norm-controlled privacy primitives: privacy is not represented by zero leakage in an absolute sense, but by bounded deviation from an ideal randomizing map. A recurrent misconception is that there is a unique approximation notion for privacy channels. The literature instead uses several inequivalent metrics depending on the task: trace or Schatten norms for output randomization, diamond norm for approximate degradability, fidelity for approximate private states, and QLDP inequalities for local privacy constraints. This suggests that “approximate private channel” is not a single universal definition but a task-dependent family of relaxations [1412.0980].

## 2. Fermionic Gaussian formulation

The fermionic construction is formulated on an $M$-mode system with annihilation and creation operators $\{\hat f_j,\hat f_j^\dagger\}_{j=1}^M$, and more conveniently with Majorana operators
\[
\{\hat c_1,\ldots,\hat c_{2M}\},
\]
satisfying the canonical anti-commutation relations
\[
\{\hat c_k,\hat c_\ell\}=2\delta_{k\ell}, \qquad \hat c_k^\dagger=\hat c_k.
\]
A fermionic Gaussian state is generated by a quadratic Hamiltonian,
\[
\rho_F=\frac{1}{Z}e^{\hat H}, \qquad 
\hat H=\frac{i}{2}\hat{\mathbf c}^T\Gamma \hat{\mathbf c}+\hat{\mathbf c}^T\mathbf x,
\]
with $\Gamma=-\Gamma^T\in \mathcal M_{2M}(\mathbb R)$. The paper then restricts attention to the even Gaussian case,
\[
\hat H'=\frac{i}{2}\hat{\mathbf c}^T\Gamma \hat{\mathbf c},
\]
which is the natural sector for the explicit privacy construction [2003.11907].

The corresponding normal form is
\[
\rho_F=\bigotimes_{k=1}^M \rho_{F,k}, \qquad 
\rho_{F,k}=\frac{1+\lambda_k}{2}\,|0\rangle\langle0|+\frac{1-\lambda_k}{2}\,|1\rangle\langle1|,
\]
with $\lambda_k\in[0,1]$. Pure fermionic Gaussian states are obtained when $\lambda_k=1$ for all $k$. This parametrization makes clear that fermionic Gaussian randomization is structurally analogous to discrete and bosonic Gaussian randomization, but it is built from CAR algebra and antisymmetric covariance data rather than Pauli operators or CCR-based phase-space variables [2003.11907].

## 3. Explicit fermionic private-channel construction

The explicit fermionic private channel is a random unitary channel
\[
\Lambda_F(\rho_F)=\frac{1}{|\mathcal U|}\sum_{\ell=1}^{|\mathcal U|}U_\ell \rho_F U_\ell^\dagger,
\]
with fermionic unitaries chosen as
\[
U_\ell=i\pi \hat c_\ell,
\]
and parity operator
\[
\pi=(-1)^M \hat c_1\cdots \hat c_{2M}.
\]
This is the concrete Majorana-based construction introduced for fermionic Gaussian privacy [2003.11907].

Its role is directly comparable to Pauli twirling in qubit systems: conjugation by these unitaries randomizes fermionic Gaussian states so that the output is driven toward the maximally mixed state. The paper stresses that this similarity should not obscure the algebraic distinction. Fermionic channels are organized by Majorana operators and CAR structure, whereas bosonic Gaussian channels use displacement and squeezing transformations over CCR phase space. The resemblance is therefore operational rather than algebraically literal [2003.11907].

The same broad random-unitary paradigm appears in earlier approximate randomization results outside the fermionic setting. In approximate quantum state sharing, for example, random unitary channels of the form
\[
\mathcal{N}(\varphi)=\sum_{i=1}^n p_i\,U_i\varphi U_i^\dagger
\]
are used as approximate private quantum channels, with the trace-distance condition
\[
\left\|\mathcal{N}(\varphi)-\frac{\mathbb{I}}{d}\right\|_1\le\epsilon
\]
playing the same operational role [1011.3297]. The fermionic construction can be viewed as the Gaussian-CAR specialization of this general randomization paradigm.

## 4. Resource scaling and proof architecture

A principal result of the fermionic paper is the gap between exact and approximate privacy cost. The exact fermionic PQC requires about
\[
|\mathcal U|=(2M)^2=4M^2
\]
unitaries in the optimal exact case, whereas approximate fermionic privacy can be achieved with only
\[
|\mathcal U|=O(M\log M)
\]
unitaries [2003.11907]. In the trace-norm case, the highlighted proposition states that
\[
\Lambda_F(\rho_F)=\frac{1}{|\mathcal U|}\sum_{\ell=1}^{|\mathcal U|}\pi\hat c_\ell\,\rho_F\,(\pi\hat c_\ell)^\dagger
\]
is an $\varepsilon$-FPQC if
\[
|\mathcal U|\ge 2\kappa M,
\]
with the proof carrying an additional $\log(10/\varepsilon)$ factor. The emphasized message is therefore linear scaling in $M$ up to constants and logarithms, rather than quadratic scaling [2003.11907].

The proof strategy is a standard randomization argument adapted to pure fermionic Gaussian states. First, an $\varepsilon$-net $N$ of pure fermionic states is used, with size bound
\[
|N|\le \left(\frac{5}{\varepsilon}\right)^{4M}.
\]
Second, for
\[
Y_{\varphi_F}:=\left\|\Lambda_F(\varphi_F)-\frac{I_{2M}}{2M}\right\|_1,
\]
the bounded-difference estimate
\[
\left|Y_{\varphi_F}-Y'_{\varphi_F}\right|\le \frac{2}{|\mathcal U|}
\]
allows McDiarmid concentration,
\[
\Pr\left[Y_{\varphi_F}\ge t+\left(\frac{2M}{|\mathcal U|}+\frac{1}{2M}\right)\right] \le e^{-|\mathcal U|t^2/2}.
\]
Third, unitary invariance and the triangle inequality extend the estimate from net points to all pure states. Finally, a union bound over the net yields exponentially small failure probability once the ensemble size scales as above [2003.11907].

This proof architecture mirrors earlier discrete and bosonic approximate randomization arguments, but the fermionic specialization is technically nontrivial because the geometry of the state space and the randomizing ensemble are Majorana-based rather than Pauli- or displacement-based. A plausible implication is that the $O(M\log M)$ phenomenon is less about any one operator basis than about the generic effectiveness of random-unitary coverings in Gaussian state families.

## 5. Relation to private capacity and approximate channel structure

The fermionic paper motivates FPQCs by the same two themes that drive the broader PQC literature: quantum cryptographic tasks and quantum channel capacity problems [2003.11907]. In that wider literature, approximate structure often appears not through output randomization but through approximate degradability. A channel $\Phi$ is $\varepsilon$-degradable if there exists a degrading channel $\Xi$ such that
\[
\bigl\|\Phi^{c}-\Xi\circ\Phi\bigr\|_{\diamond}\le \varepsilon,
\]
and this condition yields explicit upper bounds on both quantum capacity and private classical capacity [1412.0980]. This is a different relaxation from $\varepsilon$-PQC randomization, but both frameworks replace exact structural equalities by norm-bounded defects.

For bosonic Gaussian channels, approximate privacy is usually discussed through private-capacity converses rather than randomization. One work derives upper bounds on the energy-constrained private capacity of single-mode bosonic Gaussian channels using the conditional quantum entropy power inequality, including the attenuator bound
\[
\mathcal{P}(\Lambda_{\tau,\rho_E},N)\le 2g(\tau N+(1-\tau)N_E^*)-2(1-\tau)g(N_E^*),
\]
and analogous amplifier bounds for thermal and general Gaussian environmental noise [2001.01179]. Another develops energy-constrained $\varepsilon$-degradable and $\varepsilon$-close-degradable bounds for phase-insensitive bosonic Gaussian channels, placing explicit upper limits on private communication rates in thermal, amplifier, and additive-noise models [1708.07257]. These results do not define approximate private channels via randomization, but they show how approximate structure governs private communication in continuous-variable settings.

A related exact-capacity perspective is supplied by the notions of more capable and less noisy channels. If the complementary channel has zero private capacity, then the main channel is less noisy and satisfies
\[
C_p({\cal N}_B)=Q({\cal N}_B)=Q^{(1)}({\cal N}_B),
\]
while the more capable condition gives
\[
C_p({\cal N}_B)=Q({\cal N}_B)
\]
[1110.5746]. This provides a distinct benchmark: instead of approximating an ideal private channel, one identifies channel classes for which the environment is fundamentally too weak to sustain private or quantum communication.

## 6. Structural extensions and adjacent notions of approximate privacy

Approximate privacy also appears at the state level. An $\varepsilon$-approximate private state $\omega$ is one that is close in fidelity to an exact private state $\gamma$, and its key size is bounded by squashed entanglement through
\[
\log_2 K \le E_{\mathrm{sq}(AA';BB')_{\omega} + f_1(\sqrt{\varepsilon},K),
\]
with an error term depending only on $\varepsilon$ and $\log_2 K$, not on shield dimension [1606.08028]. This establishes a converse principle parallel to approximate private channels: approximate secrecy remains quantitatively constrained by a robust entanglement measure.

At the opposite end of the spectrum, exact PQCs admit an operator-algebraic characterization. For a conditional expectation channel $E:M_n\to A$, a triple $[S,E,\rho_0]$ is a private quantum channel if and only if $S$ is a set of trace vectors of $A$ with respect to $\rho_0$ [1210.6625]. The paper is explicitly exact rather than approximate, but it suggests a structural route to approximation: replace exact trace-vector identities by norm- or expectation-controlled perturbations.

More recently, privacy-constrained quantum channels have also been analyzed through QLDP and contraction coefficients. Under $\varepsilon$-QLDP, the exact trace-distance contraction coefficient is
\[
\eta_T^\varepsilon=\frac{e^\varepsilon-1}{e^\varepsilon+1},
\]
and under $(\varepsilon,\delta)$-QLDP it becomes
\[
\eta_T^{\varepsilon,\delta}=\frac{e^\varepsilon-1+2\delta}{e^\varepsilon+1}
\]
[2406.18651]. These results concern local privacy rather than output randomization to the maximally mixed state, but they reinforce the same general lesson: approximate privacy is naturally expressed through quantitative contraction or closeness bounds, and those bounds propagate to operational tasks such as hypothesis testing, fairness, and Holevo-information stability.

Taken together, these developments indicate that approximate private channels form a broad research area rather than a single formalism. In the fermionic Gaussian setting, the defining achievement is the explicit $\varepsilon$-FPQC construction from Majorana-derived unitaries together with the sharp reduction in randomization cost from $4M^2$ to $O(M\log M)$ [2003.11907]. In adjacent settings, approximation is captured instead by degradability defects, finite-energy simulation errors, state fidelity to private states, or privacy-constrained contraction coefficients. This suggests that approximate privacy is best understood as a family of controlled relaxations of exact secrecy, tailored to the algebraic structure of the system and the operational quantity being optimized.

Source: https://www.emergentmind.com/topics/approximate-private-channels