Proof of Tomaszewski's Conjecture on Randomly Signed Sums

Abstract: We prove the following conjecture, due to Tomaszewski (1986): Let $X= \sum_{i=1}^{n} a_{i} x_{i}$, where $\sum_i a_i^2=1$ and each $x_i$ is a uniformly random sign. Then $\Pr[|X|\leq 1] \geq 1/2$. Our main novel tools are local concentration inequalities and an improved Berry-Esseen inequality for Rademacher sums.

• The paper proves Tomaszewski's 1986 conjecture using novel local concentration inequalities, refined Berry-Esseen bounds, and a semi-inductive approach.
• Key methods include new local concentration inequalities and an improved Berry-Esseen bound tailored to Rademacher sums for tighter probabilistic control.
• The techniques resolve a long-standing problem and have potential applications in statistical physics, financial mathematics, and data science.

Insightful Overview of the Proof of Tomaszewski's Conjecture

The paper "Proof of Tomaszewski's Conjecture on Randomly Signed Sums" by Nathan Keller and Ohad Klein provides a detailed and rigorous proof of a well-known problem in probability theory and combinatorics. Specifically, the authors resolve a conjecture posed by Bogusław Tomaszewski in 1986, which concerns the distribution of sums of real numbers with random signs. The central claim of Tomaszewski's problem is that for any set of real numbers whose squares sum to one, the probability that the absolute value of a signed sum is less than or equal to one is at least one-half.

Core Results and Methods

The authors' approach builds on several innovative techniques:

1. Concentration Inequalities: The paper introduces new local concentration inequalities for Rademacher sums, which are sums of independent random variables taking values in {-1, 1}. These inequalities allow the authors to compare probabilities over different intervals and are crucial for establishing bounds on these probabilities. Specifically, the segment comparison tool is pivotal in deriving the main result.
2. Berry-Esseen Type Bounds: An improved Berry-Esseen inequality tailored to Rademacher sums is developed. This result provides tighter control on how closely the distribution of these sums aligns with a Gaussian distribution. This refinement is pivotal for validating the conjecture when individual terms in the sum are relatively small.
3. Semi-Inductive Approach: A novel semi-inductive strategy is employed to handle cases where the two largest weights of the sum are relatively large. By considering sums with a reduced number of terms and using a well-crafted stopping time argument, the authors are able to incrementally extend known results to larger sums.

Structural Breakdown of the Proof

The proof is divided into several cases based on the values and relationships of the weights in the sum. This granular approach allows for the application of different mathematical tools across various scenarios:

• Small Maximum Coefficient: Demonstrated using the improved Berry-Esseen bounds, providing a foothold into tackling the conjecture for sums with coefficients up to 0.31.
• Dominant Coefficients: Managed by the semi-inductive method, allowing the authors to extend results from simpler cases to scenarios where two dominant coefficients aggregate to one.
• Intermediate Coefficients: Utilizing a sophisticated combination of the newly developed concentration inequalities and traditional Chebyshev-type inequalities to ensure the conjecture holds across more intricate distributions.

Implications and Future Directions

The proof not only resolves a long-standing mathematical conjecture but also opens avenues for further exploration in sum distributions and stochastic processes involving discrete random variables. The techniques developed herein, particularly the refined probabilistic inequalities and the tailored Berry-Esseen bound, have potential applications in statistical physics, financial mathematics, and beyond, where understanding the behavior of randomly weighted sums is of interest.

Future research might focus on generalizing these results to more complex scenarios, such as sums involving dependent random variables or expanding the methodology to encompass other types of probability distributions. Additionally, exploring the computational aspects of the proposed techniques could lead to novel algorithms in data science and machine learning.

In summation, the paper by Keller and Klein significantly advances our comprehension of probabilistic inequalities and the behavior of randomly signed sums, shedding light on intricate mathematical phenomena and stimulating further research in probability theory and its applications.