🔢 mathematics

The Equality Cases of the Weak Simplex Conjecture

This paper proves that the regular simplex is the unique maximizer of the correct-decoding probability for equiprobable equal-energy signals in additive white Gaussian noise, establishing that any non-simplex signal set strictly underperforms the bound at every positive signal-to-noise ratio.

Original authors: Mengwei Su, Kaiwen Yang, Hao Xu, Chih-Lin I

Published 2026-08-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mengwei Su, Kaiwen Yang, Hao Xu, Chih-Lin I

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

In the invisible highways of modern communication, information travels as waves of electricity or light, constantly battered by a background hum known as noise. To send a message reliably, engineers must choose a set of distinct shapes, or signals, to represent the data. Imagine trying to place a handful of marbles on a table so that they are as far apart from one another as possible; this distance acts as a buffer against the noise that might blur them together. For decades, a fundamental question has lingered in the minds of information theorists: if you have a specific number of signals to send, all carrying the same amount of energy, what is the absolute best way to arrange them? The answer would guarantee the highest possible chance that the receiver understands the message correctly, no matter how strong or weak the noise is. This is not just a theoretical puzzle; it defines the ultimate limits of how efficiently we can communicate.

For seventy years, the leading hypothesis was that the best arrangement is a regular simplex. In simple terms, if you have four signals, they should form a shape like a tetrahedron, where every point is equidistant from every other point. This idea, known as the Weak Simplex Conjecture, was widely believed to be true, but proving it was like trying to show that a specific arrangement of marbles is the only one that works, rather than just one of many good options. Previous work had shown that this arrangement was indeed optimal, but it left a critical door open: could there be some other, stranger arrangement that performed just as well? Or was the regular simplex the single, unique champion?

A new study has finally closed that door, proving that the regular simplex is not merely a good design, but the only design that works. The researchers demonstrated that any deviation from this perfect, symmetrical shape results in a strictly worse performance. It does not matter how slightly you distort the arrangement or how you shift the signals; the probability of making a mistake will always be higher than if you had used the perfect shape. This finding is absolute. It means that in the world of signal design, there is no such thing as a tie. If a computer search finds two arrangements that seem to perform equally well, the researchers explain that this is an illusion caused by the limitations of the calculation, not a reality of the physics. The perfect shape is the only one that reaches the theoretical ceiling.

The team arrived at this conclusion by translating the problem of sending signals into the language of probability and geometry. They treated the signals as points in a multi-dimensional space and analyzed how likely it was for a noisy receiver to confuse one point for another. By using advanced mathematical tools to examine the behavior of these points under every possible condition, they showed that the regular simplex creates a unique "gap" in performance. Any other arrangement, no matter how close it looks to the perfect shape, falls strictly below this gap. The proof is so rigorous that it has been verified by a computer program designed to check mathematical logic, ensuring that every step of the argument holds up without human error.

This result has profound implications for how we think about efficiency in communication. It tells engineers that there is no need to search for alternative, perhaps simpler, configurations that might offer the same performance. The regular simplex is the only path to the best possible outcome. Furthermore, the study revealed that for a signal set to be optimal, every single signal must use its entire allowed energy budget. There is no room for saving power; to achieve the best performance, every signal must be pushed to its limit. This rigidity extends beyond just the signals themselves to the very shape of the space they occupy, confirming that the geometry of the solution is as fixed and unchangeable as the laws of physics that govern the noise.

The researchers also connected their findings to a broader question in geometry regarding the "width" of shapes. They proved that among all shapes of a certain type, the regular simplex is the only one that maximizes a specific measure of size. This confirms that the optimality of the simplex is a fundamental property that appears in different mathematical disguises, whether viewed through the lens of communication, probability, or pure geometry. By settling this long-standing question, the paper provides a definitive map for the landscape of signal design, showing that while there are many ways to arrange signals, there is only one way to do it perfectly. The search for a better arrangement is over, not because a better one was found, but because the perfect one has been proven to be alone at the top.

Technical Summary: The Equality Cases of the Weak Simplex Conjecture

Problem Statement
This paper addresses the "Weak Simplex Conjecture," a foundational problem in signal design for the Additive White Gaussian Noise (AWGN) channel. The problem asks: among $n+1$ equiprobable, equal-energy signals in $\mathbb{R}^n$ , which arrangement maximizes the probability of correct decoding under maximum-likelihood (ML) decoding? While Mulgund (2026) established that the regular simplex provides an upper bound on the correct-decoding probability for every signal-to-noise ratio (SNR), the question of whether this bound is attained only by the regular simplex (uniqueness) remained open. Specifically, Mulgund's Open Problem 8.1 sought to determine the equality cases: whether a non-simplex signal set could match the regular simplex's performance at a single operating point or if strict superiority is guaranteed for any deviation.

Methodology
The authors resolve this problem by translating the coding question into two equivalent frameworks: the probabilistic domain (Gaussian maxima) and the convex geometric domain (mean width). The core methodology involves strengthening the variational arguments used in Mulgund's proof to establish strict inequalities.

Probabilistic Translation: The decoding probability is expressed via the moment-generating function of the maximum of a correlated Gaussian vector. The problem becomes determining if a correlated Gaussian maximum can stochastically match an independent one.
Strict Inequalities: The proof relies on two new strict inequalities:
- Strict Symmetric Rectangle Inequality (Lemma 9): A sharpening of the classical Šidák–Khatri inequality. It proves that for a centered Gaussian vector with a correlation matrix $S \neq I$ , the probability of falling within a symmetric rectangle is strictly greater than the product of the marginal probabilities. This holds for singular matrices as well, avoiding approximation limits that might lose strictness.
- Strict Product Inequality for Truncated Exponentials (Lemma 10): This lemma applies to the truncated exponential factors generated by Mulgund's adaptive tilting technique. It demonstrates that the self-convolution of these factors yields a strictly positive deficit unless the correlation structure is trivial.
Synthesis: By combining Royen's Gaussian correlation theorem with these strict inequalities, the authors show that any deviation from the identity correlation matrix (in the transformed domain) or the regular simplex Gram matrix (in the coding domain) results in a strictly lower performance metric.

Key Results
The paper provides a complete characterization of the equality cases, proving that the regular simplex is the unique optimizer up to orthogonal transformations and vertex relabeling.

Theorem 1 (Single-threshold strictness): For any correlation matrix $R \neq I$ satisfying the problem's constraints, the lower-orthant probability $P\{X \leq c\mathbf{1}\}$ is strictly greater than $\Phi(c)^{n+1}$ for every finite threshold $c$ . Equality holds if and only if $R = I$ .
Theorem 2 (MGF Rigidity): For the moment-generating function comparison, equality $E[e^{\lambda \max \xi_i}] = E[e^{\lambda \max \xi_{\Delta, i}}]$ holds at any single $\lambda > 0$ if and only if the Gram matrix $G$ corresponds to a regular simplex ( $G = G_\Delta$ ). If $G \neq G_\Delta$ , the inequality is strict for every $\lambda > 0$ .
Theorem 3 (Coding Consequences):
- Equal-Energy Case: A code achieves the regular simplex decoding probability at a single positive SNR if and only if it is a centered regular simplex inscribed in the unit sphere. No code in ambient dimension $d < n$ can attain the bound.
- Finite-Energy Case: Under an energy budget $E$ with unrestricted blocklength, the optimal codebook is uniquely the regular simplex of circumradius $\sqrt{E}$ . Crucially, every optimal codeword must exhaust its full energy allowance ( $\|c_i\|^2 = E$ ); partial energy usage is suboptimal.
Corollary 4 (Mean Width): The equality case of the Simplex Mean Width Conjecture is resolved: the mean width of a simplex in the unit ball is maximized uniquely by the regular simplex.

Significance and Claims
The paper claims to settle Open Problem 8.1 in a form stronger than uniqueness. It establishes that the regular simplex is not merely "one good design" but the only design that can achieve the theoretical upper bound.

Strictness: The results imply that any departure from the regular simplex incurs a strictly positive performance loss at every positive SNR. Consequently, numerical searches for optimal constellations cannot terminate on a "tie" with a non-simplex configuration; any reported tie is a numerical artifact.
Design Implications: For finite-energy short codes, the optimal codebook is pinned exactly to the regular simplex, and the requirement that codewords use full energy is a necessary consequence of optimality, not a design choice.
Robustness: The proofs hold for singular correlation matrices and do not rely on approximation arguments that might fail at the boundaries.
Verification: The single-parameter rigidity of the moment-generating comparison has been machine-checked in Lean 4, providing a formal verification of the core algebraic steps.

The paper concludes that while the average-energy variant (Strong Simplex Conjecture) is false (as shown by Steiner), the equal-energy formulation admits a rigid, unique solution. The work leaves open the configuration for $m \neq n+1$ signals and the biorthogonal conjecture for $m=2n$ , but fully resolves the $n+1$ equal-energy case.

Technical Summary: The Equality Cases of the Weak Simplex Conjecture

More like this