🔢 mathematics

Local Information-Theoretic Security via Euclidean Geometry

This paper proposes a Euclidean information theory framework that transforms the non-convex optimization of secure communication over wiretap channels into a tractable quadratic program, enabling the derivation of an analytical local secrecy capacity formula and new secret local contraction coefficients characterized by generalized eigenvalues.

Original authors: Emmanouil M. Athanasakos, Nicholas Kalouptsidis, Hariprasad Manjunath

Published 2026-05-14

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Emmanouil M. Athanasakos, Nicholas Kalouptsidis, Hariprasad Manjunath

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Secrecy in a Small Room

Imagine you are trying to whisper a secret to a friend (Bob) in a noisy room, while a spy (Eve) is standing right next to you, listening. In the world of information theory, we usually ask: "What is the maximum amount of secret data we can send if we have an infinite amount of time and a perfect code?"

This paper asks a different, more practical question: "If we are sending just a tiny, specific piece of information, how can we whisper it as clearly as possible to Bob while making sure Eve hears almost nothing?"

The authors call this "Local Information-Theoretic Security." Instead of looking at the whole ocean of data, they zoom in on a single drop of water to understand its shape and behavior.

The Problem: A Tricky Puzzle

The authors set up a game with three rules:

Help Bob: Maximize how much Bob understands.
Stop Eve: Keep the amount of information Eve hears below a strict limit.
Save Energy: Don't use too much "encoding power" (the effort to turn the secret into a signal).

Mathematically, this is a nightmare. It's like trying to find the highest point on a mountain range that is covered in fog, where the ground is bumpy and non-smooth. Standard math tools often get stuck or take forever to solve this.

The Solution: Flattening the Mountain (Euclidean Geometry)

The authors use a clever trick called Euclidean Information Theory (EIT).

Imagine the mountain of possible solutions is so complex it's hard to climb. EIT says: "Let's zoom in so close to our current spot that the mountain looks flat."

The Metaphor: If you stand on a giant beach ball, the ground looks curved. But if you look at just the patch of sand under your feet, it looks perfectly flat and square.
The Math: By treating the problem as if it were happening on a flat, square grid (Euclidean space) rather than a curved, complex one, they can turn the impossible "bumpy mountain" problem into a simple Linear Programming problem.

Think of it like this: Instead of trying to navigate a winding, foggy maze, they draw a straight line through the center. They prove that for small, local steps, the straight line is a perfect guide.

The Discovery: The "Secret Contraction Coefficient"

One of the paper's biggest findings is a new number they call the Secret Local Contraction Coefficient.

The Analogy: Imagine you have a leaky bucket (Eve) and a solid bucket (Bob). You pour water (information) into a pipe.
- Some pipes are great: they send almost all the water to Bob and very little to Eve.
- Some pipes are bad: they leak a lot to Eve.
The Coefficient: This new number measures the best possible pipe in your system. It tells you the maximum ratio of "Helpful Water" (Bob) to "Leaky Water" (Eve) you can achieve locally.

The authors discovered that this number isn't just a random guess; it is the largest "eigenvalue" of a specific matrix derived from the channel. In simple terms, it's a specific number hidden inside the math of the channel that tells you exactly how efficient your secrecy can be.

How They Solved It: The "Price Tag" System

The authors turned the complex secrecy problem into a Linear Program (LP).

The Metaphor: Imagine you are a shop owner. You have a budget for "Rate" (how much you can talk) and a budget for "Leakage" (how much you can afford to let the spy hear).
The Solution: They created a system of "price tags" (Lagrange multipliers).
- If the "Leakage" budget is tight, the price tag for leakage goes up, and the system automatically chooses a strategy that leaks less.
- If the "Rate" budget is tight, the price tag for rate goes up.
The Result: They proved that you don't need to guess these prices. You can find the perfect prices by solving a simple, standard math puzzle (a Linear Program) based on the channel's geometry.

The "Binary Symmetric" Example

To prove their idea works, they tested it on a classic, simple scenario called the Binary Symmetric Wiretap Channel (BSWC).

The Setup: Think of a light switch. You can flip it Up (1) or Down (0).
- Bob sees the switch correctly most of the time, but sometimes it flips by accident (noise).
- Eve also sees the switch, but her view is even noisier.
The Result: The authors showed that for this simple switch, their "local" math gives a very accurate answer that matches the "true" global answer when the secret is small. It also clearly showed two different modes of operation:
1. Leakage-Dominant: When the spy is very good, you must whisper so quietly that you barely speak at all.
2. Rate-Dominant: When the spy is bad, you can shout as much as your energy budget allows.

Summary of Claims

Local is Better for Small Data: For small amounts of data or specific operating points, looking at the "local" geometry is more useful than looking at the "global" asymptotic limits.
It's a Linear Problem: By using Euclidean geometry, a very hard, non-convex problem becomes a solvable Linear Program.
The "Secret Coefficient": They defined a new metric (the Secret Local Contraction Coefficient) that quantifies the channel's intrinsic ability to hide secrets. It is calculated as the largest generalized eigenvalue of the channel matrices.
Design Rules: The solution tells engineers exactly when to prioritize speed (Rate) and when to prioritize secrecy (Leakage) based on the channel's specific "eigenvalues" (its spectral properties).

In short, the paper provides a mathematical toolkit to analyze and design secure communication systems for small, specific tasks by flattening complex curves into simple lines, revealing that the best way to keep secrets is often determined by the fundamental "shape" of the communication channel itself.

Technical Summary: Local Information-Theoretic Security via Euclidean Geometry

Problem Statement
The paper addresses the challenge of characterizing secure communication over discrete memoryless wiretap channels (DM-WTC) in regimes where asymptotic assumptions (infinite blocklength) are insufficient. Specifically, it targets scenarios involving finite resources, constraints on encoding complexity, and the transmission of small data units, such as control signals in cyber-physical systems or IoT updates. The core problem is formulated as the Secret Information Coupling (SIC) problem: maximizing the legitimate user's information rate ( $I(U;Y)$ ) subject to explicit upper bounds on information leakage to an eavesdropper ( $I(U;Z) \leq \Theta$ ) and the informational cost of encoding the secret message ( $I(U;X) \leq R$ ). Unlike classical secrecy capacity formulations that optimize over asymptotic rates, the SIC problem seeks to understand performance trade-offs under small perturbations of a fixed reference input distribution.

Methodology
The authors employ Euclidean Information Theory (EIT) to transform the inherently non-convex SIC optimization problem into a tractable algebraic structure. The methodology proceeds as follows:

Local Approximation: The paper leverages the differential geometry of statistical manifolds to approximate Kullback-Leibler (KL) divergences and mutual information terms using quadratic forms. By defining perturbation vectors ( $L_u$ ) around a reference input distribution $P_X$ , the mutual information terms $I(U;X)$ , $I(U;Y)$ , and $I(U;Z)$ are approximated as weighted squared Euclidean norms of these vectors.
Quadratic Programming Formulation: The SIC problem is recast as a quadratic programming problem where the objective is to maximize a quadratic form of the perturbation vectors subject to quadratic constraints representing rate and leakage budgets.
Dual Analysis and Linear Programming (LP): Using Lagrangian duality and Karush-Kuhn-Tucker (KKT) conditions, the authors demonstrate that the optimal Lagrange multipliers governing the approximated problem can be determined by solving a Linear Program (LP). The constraints of this LP are derived from the generalized eigenvalues of channel-derived matrices (specifically, the Divergence Transfer Matrices for the legitimate and eavesdropper channels).
Spectral Characterization: The solution structure is shown to be governed by the spectral properties of the channel matrices. The optimal signaling strategy shifts between distinct operational regimes (rate-dominant, leakage-dominant, and intermediate) based on the solution to the LP.

Key Contributions
The paper presents several theoretical and analytical contributions:

EIT-Approximated SIC Framework: A systematic transformation of the non-convex SIC problem into a quadratic form, enabling analytical solutions for local secrecy capacity.
Linear Program for Capacity: A proof that the optimal Lagrange multipliers (and thus the approximate local secrecy capacity) are found by solving a general LP. The constraints of this LP are explicitly defined by the generalized eigenvalues of the channel matrices $V$ (related to the legitimate channel) and $\Lambda$ (related to the eavesdropper channel).
Analytical Formula for Local Secrecy Capacity: An explicit formula for the approximate local secrecy capacity ( $C_{SIC}$ ) expressed as a linear combination of the rate and leakage budgets ( $C_{SIC} = \rho^* R + \nu^* \Theta$ ), where the weights are the optimal dual variables.
Secret Local Contraction Coefficient: The introduction of a new information-theoretic quantity, the secret local contraction coefficient ( $\eta_{loc}^{sec}$ ), defined as the largest generalized eigenvalue of the matrix pencil $(V, \Lambda)$ . This coefficient quantifies the maximum achievable ratio of approximate utility to approximate leakage, serving as a measure of the channel's intrinsic local leakage efficiency.
Bounds on Global vs. Local Coefficients: The establishment of upper and lower bounds connecting the EIT-derived local contraction coefficient to the global secret contraction coefficient defined over exact mutual information measures.

Results
The paper validates the proposed framework through detailed numerical analysis and case studies:

Operational Regimes: The analysis identifies three distinct regimes for the local secrecy capacity:
- Rate-Dominant: The system is constrained by the encoding budget; capacity scales linearly with $R$ .
- Leakage-Dominant: The system is constrained by the secrecy requirement; capacity scales linearly with $\Theta$ and is determined by the secret local contraction coefficient.
- Intermediate: A trade-off regime where the optimal solution lies at the intersection of multiple constraints.
Binary Symmetric Wiretap Channel (BSWC): The framework is applied to the BSWC, where the commuting nature of the channel matrices simplifies the analysis. The results show that the EIT approximation accurately captures the initial slope of the true secrecy capacity curve and correctly predicts the transition between operational regimes.
Validation of LP Formulation: Numerical experiments on multi-mode channels (with non-commuting matrices) confirm that the LP solver yields results identical to exhaustive searches over feasible vertices, validating the tractability of the approach.
Convergence of Source Distribution: Numerical results support the theoretical claim that the optimal value of the objective function is independent of the specific source distribution $P_U$ , provided the cardinality of the message alphabet is sufficient to satisfy the average constraints (consistent with the Support Lemma).

Significance and Claims
The paper claims to provide a new geometric perspective on the structure of secure communication, moving beyond "black-box" numerical solutions to derive analytical expressions for performance sensitivities. By linking local secrecy performance to the spectral theory of channel matrices, the framework offers a constructive method for evaluating and designing secure systems under finite-resource constraints.

The authors emphasize that their approach is particularly relevant for emerging communication paradigms requiring high efficiency and low latency, such as IoT and cyber-physical systems. The introduction of the secret local contraction coefficient provides a new metric for quantifying the intrinsic efficiency of secrecy mechanisms. The paper modestly positions its results as approximations valid in the local regime (small perturbations), acknowledging that while they capture the fundamental trade-offs and structural properties, they differ from global asymptotic capacities at larger operating points. The work establishes a bridge between Euclidean geometry, linear algebra, and information-theoretic security, offering a tractable alternative to solving complex non-convex optimization problems directly.