Submodularity of entropy under quantum convolution

arXiv:2609.40211 · quant-ph, cs.IT, math.CO, math.IT · Submitted 2026-09-30 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: I'm Kai, and with me are Mira and Lev, guest researcher.

Mira: Today's paper: "Submodularity of entropy under quantum convolution".

Kai: The paper develops a submodular framework for von Neumann entropy under discrete quantum convolutions,

Mira: First, who's behind it and why it matters.

Title and authors: Kai: So, we're talking about "Submodularity of entropy under quantum convolution," which sounds pretty dense on the surface, but Mira, can you distill what that actually means in plain language for our listeners?

Mira: Well, essentially, this paper is building a mathematical structure to handle how entropy behaves when you perform quantum convolutions. Instead of treating these convolutions as just arbitrary operations, they are organizing them into a framework where the entropy gains follow a specific pattern called submodularity. This is like finding a consistent rule for how information accumulates when you combine different quantum inputs in this specific way.

Lev: From my side, I'm curious if this structure translates to anything practical for error correction; does this submodularity imply any sort of predictable bound on how much noise we can tolerate during sequential quantum operations?

Kai: That’s a huge question, Lev. It suggests that the way information grows isn't completely random or chaotic under these convolutions; it has an underlying geometric structure—polymatroidal geometry, according to the authors.

Mira: Exactly. The authors show that relative to any fixed set of inputs, the entropy gains have this normalized, monotone submodular extension across all subsets of the remaining inputs. It converts bounds from smaller sets into bounds for the full collection, which is a big conceptual step for entropic additive combinatorics.

Lev: If the structure is so well-defined, it might actually give us a way to design more robust quantum processes where we know exactly how much information we can expect to gain or lose at each step.

Kai: That’s what I'm hoping for, Lev. It moves us from just observing results to understanding the underlying mechanism that governs the growth of entropy in these circuits.

The paper's summary: Kai: So, we’ve touched on what submodularity means, and now let’s look at the core summary of "Submodularity of entropy under quantum convolution" to really nail down what they achieved in this work.

Mira: The authors are introducing globally weighted quantum convolutions and a compatible family indexed by admissible subsets as their starting point. They then use the characteristic-kernel method to show that quantum convolution essentially becomes entrywise multiplication of characteristic kernels.

Lev: I see the connection there; if it’s just entrywise multiplication, we might be able to analyze the spectral properties of these kernels more directly, which is important for hardware implementation analysis.

Kai: Right. The key finding is that this method allows them to relate the entropy of an auxiliary state—which is easier to handle—to the actual physical convolution entropy through a simple logarithmic difference, D.

Mira: That bridge is what makes the whole thing work; it lets them transfer ordinary entropy inequalities from an auxiliary state onto the quantum convolution itself. This is a powerful tool for proving things about quantum operations that we couldn't do before.

Lev: So, by using this kernel method, they’ve essentially found a way to translate complex quantum state evolution into a more manageable algebraic structure involving these characteristic kernels.

Kai: Precisely. They then use this foundation to prove the main results, showing that the physical entropy gains extend to that normalized, monotone, submodular function on the full subset lattice of the remaining inputs.

The paper's improvements: Kai: Now we get to the actual meat of the results—the specific inequalities and growth estimates that they derive from this framework in "Submodularity of entropy under quantum convolution."

Mira: They’ve established several key inequalities, starting with convolutional strong subadditivity, which states that if certain admissibility conditions are met, the entropy gains satisfy S(rho s tau) l,m sigma + S(sigma) at most S(rho s sigma) + S(sigma t t tau) whenever specific coefficient relations hold.

Lev: That strong subadditivity is significant because it’s a direct analogue to classical additive combinatorics, but the authors are applying it to the quantum setting, which is what we need for real hardware analysis.

Kai: They also proved the quantum Ruzsa triangle inequality under specific conditions, like 2s squared = one which simplifies to S(rho s tau) + S(sigma) at most S(rho s sigma) + S(sigma t t tau).

Mira: And then they have the fractional cover inequalities, which are the quantum entropic Plünnecke–Ruzsa inequalities, showing how entropy behaves when you consider a fractional cover F where R J is admissible whenever alpha J > zero. This allows them to use tools from classical fractional coverings to establish bounds on quantum states.

Lev: The growth estimates are also quite telling; they show that the entropy gain per added input is nonincreasing along admissible repetition scales, and the multiplicative entropy growth of an admissible m-fold convolution is at most delta qrho m-one.

Kai: That bound on the multiplicative entropy growth being controlled by the quantum doubling constant, delta qrho, which is already among states diagonal in the computational basis, feels like a very tight constraint on how fast information can explode during repeated convolutions.

Conclusion: Mira: So to wrap up, "Submodularity of entropy under quantum convolution" provides a consistent framework connecting physical entropy gains to polymatroidal geometry. This structure yields convolutional strong subadditivity, the quantum Ruzsa triangle inequality, and those fractional cover inequalities.

Lev: For error correction researchers like me, the sharp comparisons between repeated convolutions at different admissible scales are particularly useful because they give us a clear idea of how to manage the information flow across multiple sequential steps in a circuit.

Kai: I think what’s most exciting here is that this whole mechanism connects physical convolution entropy to a separable auxiliary state, giving us systematic routes to establish these important inequalities.

Mira: It sets up a direct growth structure on the direct side of entropic additive combinatorics, which means we can use powerful tools from that area to understand quantum entropy growth.

Lev: Overall, this work provides the necessary theoretical foundation for us to start translating these abstract submodular properties into concrete limits on how much information we can expect to extract from large, complex quantum computations.

Kai: It’s a solid piece of theory that gives us the language needed to talk about the scaling and stability of quantum operations under convolution, and we're ready for whatever comes next in this line of research.

Milad M. Goodarzi

Centre for Quantum Technologies, National University of Singapore

quant-ph, cs.IT, math.CO, math.IT

Submitted: 2026-09-30

Updated: 2026-09-30

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 92/100

The gist: The paper develops a submodular framework for von Neumann entropy under discrete quantum convolutions, providing a noncommutative counterpart to entropic additive combinatorics and revealing that

Key concepts

Globally Weighted Quantum Convolutions: This construction involves assigning a single global weight to every input in a convolution. Weights are then normalized independently for every admissible subset. This process creates a compatible family of multi-input convolutions, allowing the framework to handle complex convolutions that cannot be built from simple binary ones.
Characteristic-Kernel Method: This core mechanism shows that quantum convolution simplifies to entrywise multiplication of characteristic kernels. Furthermore, the normalized characteristic kernel's spectrum is directly related to the state's eigenvalues, meaning its entropy differs from the state entropy only by a constant factor (log D). This allows standard entropy inequalities to be transferred to the quantum setting.
Polymatroidal Geometry: The growth structures of these entropic gains are governed by polymatroidal geometry. This geometric structure provides the mathematical foundation for understanding how physical entropy increases when combining different input sets, leading to powerful inequalities like strong subadditivity and Ruzsa triangle inequalities.

Terminology

Trace — 6× in this paper · explained in 2 episodes
Admissibility — 4× in this paper · explained in 2 episodes
Admissibility Conditions — 2× in this paper · explained in 1 episode
Core — 2× in this paper · explained in 2 episodes
Entanglement — 1× in this paper · explained in 3 episodes
Hilbert Space — 1× in this paper · explained in 2 episodes
alpha — 1× in this paper · explained in 1 episode

Summary

The paper develops a submodular framework for von Neumann entropy under discrete quantum convolutions, providing a noncommutative counterpart to entropic additive combinatorics and revealing that these growth structures are governed by polymatroidal geometry.

How it works

The construction begins with globally weighted quantum convolutions, where a fixed global weight is assigned to each input, and weights are normalized separately on every admissible subset, resulting in a compatible family of multi-input convolutions. This construction extends the binary framework and includes multi-input convolutions that cannot be obtained by successively applying admissible binary convolutions.
The core mechanism relies on the characteristic-kernel method, which establishes two key observations: first, that quantum convolution becomes entrywise multiplication of characteristic kernels, and second, that the normalized characteristic kernel has a spectrum related to the state's eigenvalues such that its entropy differs from the state entropy only by log D. This allows ordinary entropy inequalities for an auxiliary state to be transferred to quantum convolution.
The main result (Theorem 4.1) demonstrates that the physical entropy gains extend to a normalized, monotone, submodular function on the full subset lattice of the remaining inputs relative to any fixed admissible block R. This extension is defined even for subsets J for which the physical convolution CR∪J is not.

Key Results and Inequalities

The framework yields several significant inequalities:

(1) Convolutional Strong Subadditivity:

The paper derives convolutional strong subadditivity (Corollary 5.2), stating that if certain admissibility conditions are met, the entropy gains satisfy:

S(ρ ⊠s,s τ) ⊠l,m σ + S(σ) ≤ S(ρ ⊠s,s σ) + S(σ ⊠s,t τ), whenever specific coefficient relations hold (e.g., lbt = ms).

(2) Quantum Ruzsa Triangle Inequality:

Under the condition 2s2 = 1, the paper proves the quantum Ruzsa triangle inequality (Corollary 5.4), which is equivalent to:

S(ρ ⊠s,s τ) + S(σ) ≤ S(ρ ⊠s,s σ) + S(σ ⊠s,t τ).

(3) Fractional Cover Inequalities:

The paper establishes quantum entropic Plünnecke–Ruzsa inequalities (Theorem 4.1), showing that for a fractional cover F of M where R∪J ∈ A whenever αJ > 0, the inequality holds:

S(CR∪M) − S(CR) ≤ X J∈F αJ [S(CR∪J) − S(CR)].

Growth and Scaling Estimates

The results also provide sharp comparisons for repeated inputs:

(1) Entropy Gain Per Added Copy:

Corollary 5.8 shows that for an integer j ≥ 1 with a specific condition on the weights, the entropy gain per copy is bounded:

S(Tk) − S(ρ) / k ≤ S(Tl) − S(ρ) / l.

(2) Balanced Repetition Hierarchy:

Corollary 5.9 establishes a hierarchy for balanced repetitions:

S(C[m] (ρ)) − S(ρ) / m − 1 ≤ S(C[l] (ρ)) − S(ρ) / l − 1, where l and m are nonzero squares in Zd.

(3) Optimal Exponents:

The paper shows that the multiplicative entropy growth of an admissible m-fold convolution is at most δqρ, where δq[ρ] is the quantum doubling constant, and this exponent m − 1 is optimal, already among states diagonal in the computational basis.

Conclusion

The characteristic-kernel representation connects physical convolution entropy to a separable auxiliary state whose marginal entropies reproduce the physical convolution entropies up to explicit constants. This mechanism provides a systematic route to broad families of convolutional entropy inequalities, including strong subadditivity and Ruzsa triangle inequalities, resolving long-standing conjectures in the field. The resulting theory lies on the direct-growth side of entropic additive combinatorics.

The gist: Relative to any fixed nonempty admissible input block R, the physical entropy gains extend to a normalized, monotone, submodular function on the full subset lattice of the remaining inputs.

Improvements for AI systems

Based on the provided scientific paper, here are the specific improvements that could be made to AI systems, and what those improved systems could achieve:

)

Improve representation of quantum state evolution in complex circuits (e.g., variational quantum algorithms).
Enable more rigorous analysis of entanglement growth under specific unitary operations by utilizing the framework of globally weighted quantum convolutions and their associated polymatroidal geometry.
Develop robust methods for bounding the information gained or lost during successive noisy processing steps (convolution) in quantum computing, specifically leveraging the convolutional strong subadditivity inequality (5.4).
Improve the analysis of resource management in quantum machine learning by quantifying entropy gain as a measure of complexity or information flow across different subsets of data/inputs, using the fractional-cover inequalities (4.6) to establish optimal bounds for multi-input tasks.
Create more precise theoretical guarantees for algorithms involving sequential processing where the growth rate of information is nonincreasing along admissible scales, utilizing the sharp comparisons between repeated convolutions derived from the quantum doubling constant (5.9).

The improved AI systems could:

Perform more efficient and rigorous design of quantum circuits by minimizing complexity while satisfying specific entropy constraints during state transformation, leading to smaller, more stable quantum algorithms.
Diagnose and predict the rate at which entanglement increases or decreases when a system undergoes successive, weighted quantum operations (convolutions), allowing for better control over noise and coherence in large-scale quantum processors.
Quantify the information bottleneck in complex generative models by using the submodular framework to determine how much useful information is retained when combining different subsets of input data/features, leading to better feature selection or dimensionality reduction techniques specific to quantum states.
Design more optimized algorithms for quantum sampling or inference tasks where multiple inputs are combined sequentially, ensuring that the total information gain is optimally bounded by the sum of individual gains (via fractional-cover bounds).
Develop quantum complexity metrics for iterative processes in quantum neural networks, ensuring that each successive layer's output does not yield a disproportionately larger entropy gain than previous steps, leading to more stable and efficient deep quantum circuits.