Convexity properties of sections of 1-symmetric bodies and Rademacher sums

The paper establishes a monotonicity property for the volumes of central hyperplane sections of 1-symmetric convex bodies and a corresponding convexity property for Rademacher sums regarding projections, with applications to chessboard cutting.

The Gist

Imagine you have a giant, multi-dimensional block of cheese (a "convex body") and you want to slice it with a giant, invisible knife (a "hyperplane"). The question mathematicians love to ask is: What is the biggest slice you can get?

This paper, written by four researchers from Carnegie Mellon University, tackles two big, interconnected puzzles about slicing shapes and adding up random numbers. Here is the story of their discovery, explained without the heavy math jargon.

1. The Chessboard Cutting Problem

First, let's look at a 2D chessboard. If you draw a straight line across an $N \times N$ grid of squares, what is the maximum number of squares the line can touch?

• The Answer: It turns out to be $2N - 1$. You get this maximum by drawing a diagonal line that just barely "grazes" the corners of the squares.

Now, imagine a 3D chessboard (a cube made of smaller cubes) or even a 100-dimensional hyper-cube. If you slice through it with a flat plane, how many little cells can you hit?

• The Old Guess: For a long time, mathematicians knew the answer for a perfect cube, but they didn't know the rule for other weirdly shaped, symmetrical blocks.
• The New Discovery: The authors found a universal rule for a specific family of shapes called 1-symmetric bodies. These are shapes that look the same if you flip them over any axis or swap their dimensions (like a perfect sphere or a cube).

The "Aha!" Moment:
They proved that the "best" slice (the one with the most volume) always happens when you cut the shape exactly down the middle, along its most "diagonal" direction.

• The Analogy: Imagine a loaf of bread that is perfectly symmetrical. If you want the biggest slice of bread, you don't cut it at a weird angle near the crust. You cut it straight down the middle. Furthermore, if the bread is shaped like a perfect cube, the "middle" cut is the one that goes through all corners equally (the diagonal).
• The "Schur" Connection: They used a concept called "Schur-concavity." Think of this as a measure of chaos or evenness. The more "evenly" you distribute your cutting force across the dimensions, the bigger the slice you get. If you concentrate your cut on just one side, the slice gets smaller.

2. The Random Number Game (Rademacher Sums)

The second half of the paper is about a different but related game involving random signs.
Imagine you have a list of numbers: $x_1, x_2, \dots, x_n$. Now, imagine flipping a coin for each number. If it's heads, you keep the number; if it's tails, you make it negative. You add them all up.

• The Question: If you change the size of your numbers ($x$), how does the "average size" of the final sum behave?
• The Discovery: The authors proved that if you look at the logarithm of the average size of this random sum, the graph of that function is always convex (shaped like a bowl).

The Analogy:
Think of a tightrope walker.

• If the function were "concave" (shaped like a hill), the tightrope walker would be unstable; a tiny wobble could send them falling off.
• Because the function is convex (shaped like a valley), the tightrope walker is stable. If they step slightly off-center, the "slope" pushes them back toward the center.
• This stability is a powerful property. It means that mixing different random scenarios in a specific way always leads to a predictable, smooth outcome.

3. Why Does This Matter? (The "Dual" Connection)

The paper connects these two ideas through a concept called duality.

• Slicing a Cube (Geometry) is mathematically the "mirror image" of Adding Random Signs (Probability).
• The authors showed that the rule for getting the biggest slice of a symmetrical shape is the same mathematical rule that governs how random numbers behave when you add them up with random signs.

The "Toy Case" Victory:
There is a famous, unsolved problem in math called the "Logarithmic Brunn-Minkowski Inequality." It's like a "Holy Grail" that mathematicians have been trying to prove for decades. It's incredibly hard for general shapes.

• The authors couldn't solve the whole "Holy Grail" yet.
• However, they solved a specific, simpler version of it (the "toy case") involving random signs. They proved that for these specific random scenarios, the "stability" (convexity) holds true. This gives hope that the bigger, harder problem might be solvable too.

Summary in One Sentence

The authors discovered that for perfectly symmetrical shapes, the biggest slice is always the most "even" diagonal cut, and this geometric fact is mathematically identical to a rule about how random numbers behave when you add them up, proving a new type of stability in both geometry and probability.

Why should you care?
This isn't just about cheese or chessboards. These rules help engineers design better algorithms for data compression, help physicists understand high-dimensional spaces, and give mathematicians a new tool to solve problems that have been stuck for years. They turned a complex, abstract problem into a clear, stable rule.

Technical Summary

Here is a detailed technical summary of the paper "Convexity properties of sections of 1-symmetric bodies and Rademacher sums" by Kalarickal, Rotunno, Singh, and Tkocz.

1. Problem Statement and Context

The paper addresses two interconnected problems in geometric functional analysis and probability theory:

1. Chessboard Cutting Problem: Determining the maximum number of lattice cells (cubes) a hyperplane can intersect within a convex body $K$. This relates to the asymptotic behavior of the "cutting constant" $\beta_K$, which depends on the maximal volume of central hyperplane sections of $K$.
2. Convexity of Rademacher Sums: Investigating the convexity properties of functions involving the moments of sums of independent Rademacher random variables (random signs $\pm 1$). This is motivated by the Logarithmic Brunn-Minkowski inequality, a major open conjecture in convex geometry.

The authors aim to refine existing results regarding the unit cube and 1-symmetric bodies and establish new convexity properties for Rademacher sums that serve as a probabilistic analogue to geometric section properties.

2. Key Definitions and Background

• 1-Symmetric Convex Bodies: A convex body $K \subset \mathbb{R}^n$ is 1-symmetric if it is symmetric with respect to every coordinate hyperplane ($x_j=0$) and invariant under permutations of coordinates.
• The Function $V_K(a)$: Defined as $V_K(a) = \frac{\|a\|_1}{|a|} \text{vol}_{n-1}(K \cap a^\perp)$. For origin-symmetric bodies, the maximum volume section is central ($t=0$). The constant $\beta_K$ in the chessboard problem is $\max_a V_K(a)$.
• Majorization ($\prec$): A vector $x$ is majorized by $y$ ($x \prec y$) if the sum of the $k$ largest components of $x$ is less than or equal to that of $y$ for all $k$, with equality for the total sum.
• Schur-Convexity/Concavity: A function is Schur-concave if $x \prec y \implies f(x) \geq f(y)$.
• Busemann's Theorem: States that for an origin-symmetric convex body $K$, the function $N_K(x) = \frac{|x|}{\text{vol}_{n-1}(K \cap x^\perp)}$ defines a norm on $\mathbb{R}^n$.

3. Methodology

The authors employ a combination of geometric theorems, symmetry arguments, and probabilistic techniques:

• Geometric Approach (Theorem 1):

  • They utilize Busemann's Theorem to establish that the function $N_K(x)$ is a norm (and thus convex).
  • They leverage the symmetry of 1-symmetric bodies to show that $N_K$ is symmetric under coordinate permutations.
  • They apply the definition of majorization: if $x \prec y$, then $x$ can be written as a convex combination of permutations of $y$. By the convexity and symmetry of $N_K$, they prove $N_K(x) \leq N_K(y)$, which implies the Schur-concavity of $V_K$.

• Probabilistic Approach (Theorem 3):

  • They analyze the function $\Phi(t) = \log \mathbb{E} | \sum e^{t_j} \varepsilon_j |^p$.
  • Using Hölder's inequality, they express the $L_p$ norm as a maximum over the dual unit ball.
  • Crucial Innovation: They restrict the dual optimization to a subset $A_q$ of random variables that have nonnegative correlations with the Rademacher variables $\varepsilon_j$.
  • They prove that the specific maximizer $Y^*(t)$ in Hölder's inequality lies within this subset $A_q$.
  • This allows them to express the function as a pointwise maximum of log-convex functions (sums of exponentials), thereby proving convexity.

4. Key Contributions and Results

Result 1: Schur-Concavity of Section Volumes (Theorem 1)

• Statement: For any 1-symmetric convex body $K$, the function $a \mapsto V_K(a)$ is Schur-concave on $\mathbb{R}^n_+$.
• Implication: The maximum of $V_K(a)$ is attained at the most "chaotic" vector (the diagonal), i.e., $a = (1, 1, \dots, 1)$.
• Refinement: This generalizes a previous result by Aliev (who proved this specifically for the unit cube) to all 1-symmetric bodies.
• Formula: $\beta_K = \sqrt{n} \, \text{vol}_{n-1}(K \cap (1, \dots, 1)^\perp)$.

Result 2: Convexity of Rademacher Sums (Theorem 3)

• Statement: Let $\varepsilon_1, \dots, \varepsilon_n$ be independent Rademacher variables. For any $p \geq 1$, the function
  $$\Phi(t_1, \dots, t_n) = \log \mathbb{E} \left| \sum_{j=1}^n e^{t_j} \varepsilon_j \right|^p$$
  is convex on $\mathbb{R}^n$.
• Significance: This is a "toy case" or probabilistic analogue of the conjectured Logarithmic Brunn-Minkowski inequality. While the geometric conjecture for sections of cubes remains open, this probabilistic version is proven true.

Result 3: Extensions and Corollaries

• Corollary 5: The convexity result extends to sums of independent symmetric random variables (not just Rademacher).
• Corollary 7: The result extends to Hilbert space-valued coefficients.
• Theorem 6 (Monotonicity): For 1-symmetric bodies, the Busemann norm $N_K(x)$ is monotone with respect to each coordinate on $\mathbb{R}^n_+$.
• Lemma 8 & Corollary 9: The authors provide an elementary representation of the $L_1$ norm of Rademacher sums as a maximum of linear forms with nonnegative, ordered coefficients. This offers an alternative proof for the convexity of the function in Theorem 3.

5. Significance and Impact

• Resolution of a Specific Case: The paper resolves the Schur-concavity question for 1-symmetric bodies, confirming that the "diagonal" direction always yields the maximal section volume for this class of bodies.
• Bridge between Geometry and Probability: The work establishes a deep connection between the geometry of convex sections (via Busemann's theorem) and the convexity of probabilistic sums (Rademacher sums). It suggests that the difficult geometric Logarithmic Brunn-Minkowski inequality might have a tractable probabilistic counterpart.
• Methodological Advance: The technique of restricting the dual space in Hölder's inequality to variables with nonnegative correlations is a novel tool that could be applicable to other problems involving convexity of moments of random sums.
• Applications: The results refine the understanding of the "chessboard cutting" constant, providing exact asymptotics for a broad class of convex bodies, not just cubes.

In summary, the paper successfully generalizes known results about the unit cube to all 1-symmetric bodies using symmetry and Busemann's theorem, and simultaneously proves a strong convexity property for Rademacher sums that serves as a probabilistic model for the elusive Logarithmic Brunn-Minkowski inequality.