Equality in the Bosonic Quantum Entropy Power Inequality

The bosonic quantum entropy power inequality establishes conditions under which an inequality relating input and output entropies becomes an exact equality, providing a complete characterization for Gaussian states in this context.

Main Results

(i) Linear Equality:

1. The linear equality, defined as the condition where the retained output state's entropy equals the weighted sum of input entropies, is equivalent to three conditions: (i) the entropy equality holds; (ii) the joint output characteristic function factorizes into a product form; and (iii) both inputs are Gaussian states with identical covariance matrices.

2. The equivalence between product outputs and Gaussian inputs is characterized by the quantum Darmois–Skitovich characterization.

3. The new implication established is that linear equality, S(C) = ηS(A) + (1 − η)S(B), forces the two outputs to be independent.

Exponential Equality

1. Exponential equality, defined by e S(C)/m = ηe S(A)/m + (1 − η)e S(B)/m, holds if and only if the inputs are Gaussian with the same covariance matrix, allowing arbitrary displacements.

2. The paper shows that the two qEPIs have the same equality class.

3. For unequal input entropies, the qEPI bound is strict at every finite-energy input pair; identical thermal inputs attain it.

Proof Strategy

1. The proof uses a doubling construction, which involves four independent finite-energy m-mode systems and applying a sequence of beam splitters and unitary transformations to generate a joint state, denoted as the final four-output state of the doubled circuit (65).

2. The core identity used is Lemma 3, which relates the deficits of different input pairs: Fp(A1, B1) + Fp(A2, B2) = Fp(A+, B+) + Fp(A−, B− M) + I(C−; E+ C+)Ξ (27).

3. By showing that the three terms on the right side are nonnegative, the identity implies that for fixed p = η, I(C−; E+ C+)Ξ = 0 (64).

Gaussian Characterization

1. The finite-energy quantum Darmois–Skitovich characterization (Proposition 14) is used to prove Gaussianity and the equality of covariance matrices when independent outputs are obtained under a beam splitter with transmissivity 0 < η < 1.

2. This proof relies on Lemma 13, the Scalar finite-variance Darmois–Skitovich lemma, which shows that if linear combinations of independent random variables satisfy specific independence conditions, the variables must be Gaussian with equal variance.

3. The equality condition for the covariance matrices is established by showing that the joint spectral law of the output state forces VA = VB through polarization of real quadratic forms.

Connection to Classical and Quantum Limits

1. The paper distinguishes between linear equality (Fη = 0) and exponential equality, noting that they are equivalent under specific affine weight variations (76).

2. The proof for exponential equality proceeds by propagating the zero affine deficit to a Gaussian limit using the n-fold quantum convolution and applying the trace-norm quantum CLT.

3. Strictness within the Gaussian family is used to recover linear equality for original inputs, demonstrating that VA = VB and p = η are necessary conditions for exact equality in both forms.

Auxiliary Tools

1. A two-outcome instrument is used to show that zero conditional mutual information implies the original input must be a product state, establishing the link between output independence and input independence.

2. The thermal auxiliary states (e.g., τ2 with mean photon number 2) are introduced at a fixed finite energy to obtain strict information loss, which is crucial for proving the equality condition in Theorem 1.

3. The use of the Poincaré representation and characteristic functions allows for the verification of key identities, such as Lemma 4, which relates the channel formula to the input state via a measure-and-prepare representation.

Conclusion and Extensions

1. The equality condition is expressed entirely in terms of output systems: an independent finite-energy pair attains either entropy inequality precisely when its beam-splitter outputs are independent.

2. The method separates two roles of Gaussian states: auxiliary thermal states enforce structural conditions on the outputs, while the Gaussian limit in the exponential proof determines affine weights and input covariances.

3. A natural extension is a quantitative version of the qEPI, which would control information lost in vacuum branches and thermal noise, though exact injectivity does not immediately yield stability estimates.

Based on the provided scientific paper, here are specific improvements that could be made to AI systems, categorized by the capability they would gain:

)Improvement 1: Gaussian State Recognition and Characterization (GSR-AI)

By integrating the characterization derived in Theorem 1 and Proposition 14, an AI system could reliably classify input data into specific quantum state classes. The system would utilize the covariance matrix comparison criterion.

• Improved Capability: The AI could take arbitrary mixed bosonic state inputs (from sensors or simulation outputs) and determine, with high precision, whether the inputs are Gaussian. Furthermore, it could identify if two independent input streams share the same covariance matrix, allowing for immediate application of optimization results (Corollary 9).

)Improvement 2: Optimal Input State Generation (OIS-AI)

Leveraging Theorem 1 and Corollary 9, an AI system could be designed to generate optimal inputs for specific beam-splitter operations.

• Improved Capability: Given a target output entropy bound or a desired linear equality condition, the AI would determine the exact Gaussian input states (including optimal mean displacements) that achieve this bound. This is crucial for designing quantum communication protocols or signal processing circuits where energy constraints and information bounds are critical.

)Improvement 3: Quantum Channel Characterization and Entanglement Breaking Detection (QCCD-AI)

By utilizing the results from Lemma 4, Proposition 6, and Section 3.2, an AI system could rigorously analyze quantum channels for their structural properties concerning entanglement breaking.

• Improved Capability: The AI could receive a quantum channel as input and determine if it is an entanglement-breaking channel (e.g., by checking if the vacuum branch has positive probability). It could also precisely characterize the effect of such channels, distinguishing between different types of noise or preparation processes (measure-and-prepare) by analyzing their characteristic functions.

)Improvement 4: Quantum Information Processing with Noise Modeling (QIPN-AI)

The proof structure in Section 3.4 and Theorem 2 allows for a sophisticated modeling of information loss through noisy channels, specifically using auxiliary thermal states.

• Improved Capability: The AI could model complex quantum systems subject to realistic noise by treating the noise as an auxiliary thermal input. It could then use the derived doubling construction (Section 3) to predict how information loss propagates across multiple stages of a beam-splitter network, allowing for the prediction of whether a channel preserves mutual information under specific noisy conditions.

)Improvement 5: Quantitative Error Estimation for Quantum Inequalities (QE-AI)

The paper introduces rigorous tools like the affine deficit and the variational weight to establish strict bounds on entropy inequalities.

• Improved Capability: The AI could move beyond qualitative inequality holds statements to provide quantitative error bounds for quantum information measures. If an input state is slightly non-Gaussian, the AI could estimate how much its entropy power inequality is violated, providing a measure of the distance from Gaussianity or equality in any given circuit configuration.

The bosonic quantum entropy power inequality bounds the entropy of a beam-splitter output in terms of the input entropies. We determine its complete equality class among independent inputs of finite mean photon number: for any number of modes and any transmissivity strictly between zero and one, equality in either the linear or the exponential form holds if and only if the inputs are Gaussian with the same covariance matrix, allowing arbitrary displacements. The main step shows that equality for one output forces the two outputs to be independent: a thermal auxiliary converts equality into preservation of mutual information by a fixed noisy channel, and a two-outcome instrument shows that preservation of mutual information requires a product state with the reference. The quantum Darmois--Skitovich theorem then gives Gaussianity, while a central-limit argument reduces exponential equality to linear equality.