[Paper Review] A Three Dimensional Signed Small Ball Inequality

Dmitriy Bilyk, Michael T. Lacey+2 more|arXiv (Cornell University)|Sep 28, 2009

Mathematical Approximation and Integration11 references3 citations

TL;DR

This paper establishes a three-dimensional signed small ball inequality for Haar functions on dyadic rectangles in the unit cube, proving that the $ L^∞ $-norm of a signed sum of Haar functions at scale $ 2^{-n} $ is bounded below by $ n^{9/8} $, improving upon the trivial $ L^2 $-based lower bound of $ n $, and providing a key step toward resolving the small ball inequality conjecture in higher dimensions, with implications for discrepancy theory and stochastic processes.

ABSTRACT

The Small Ball Inequality is a conjectural lower bound on sums the L-infinity norm of sums of Haar functions supported on dyadic rectangles of a fixed volume in the unit cube. The conjecture is fundamental to questions in discrepancy theory, approximation theory and probability theory. In this article, we concentrate on a special case of the conjecture, and give the best known lower bound in dimension 3, using a conditional expectation argument.

Motivation & Objective

• To establish a non-trivial lower bound for the $ L^\infty $-norm of signed sums of Haar functions over dyadic rectangles in three dimensions.
• To address the small ball inequality conjecture in $ d \geq 3 $, particularly for signed coefficients $ a_R = \pm 1 $, which is central to discrepancy theory.
• To improve upon the trivial $ L^2 $-based lower bound of $ n $, aiming toward the sharp conjectured $ n^{3/2} $ exponent.
• To provide a quantitative estimate that supports recent progress on the star-discrepancy function in dimensions $ d \geq 3 $.
• To contribute to the understanding of mixed derivative estimates and entropy numbers in high-dimensional function spaces.

Proposed method

• Focus on dyadic rectangles $ R \subset [0,1]^3 $ with volume $ |R| = 2^{-n} $, and consider sums of $ \pm 1 $-coefficient Haar functions $ h_R $.
• Use the $ L^\infty $-norm of the sum $ \sum a_R h_R $ over rectangles with $ |R| \geq 2^{-n} $, with coefficients $ a_R \in \{\pm 1\} $.
• Apply techniques from harmonic analysis, including Littlewood-Paley theory and properties of Haar functions in product spaces.
• Leverage the structure of dyadic rectangles and the orthogonality and cancellation properties of Haar functions to derive $ L^\infty $-norm estimates.
• Use a recursive or combinatorial decomposition of the cube to control the pointwise sup-norm of the signed sum.
• Draw connections to the Brownian sheet and mixed derivative estimates to motivate the sharpness of the bound.

Experimental results

Research questions

• RQ1 What is the best possible lower bound for the $ L^\infty $-norm of a signed sum of Haar functions over dyadic rectangles in three dimensions?
• RQ2 Can the $ L^\infty $-norm of such signed sums be bounded below by a power of $ n $ better than the trivial $ n $ from $ L^2 $-theory?
• RQ3 How does the three-dimensional signed small ball inequality relate to the star-discrepancy function in higher dimensions?
• RQ4 Is the exponent $ 9/8 $ in the lower bound $ n^{9/8} $ optimal, or can it be improved toward the conjectured $ n^{3/2} $?
• RQ5 What role do the geometric constraints $ |R_1| \geq 2^{-n/2} $ and volume $ |R| = 2^{-n} $ play in the construction of the lower bound?

Key findings

• The paper proves that for all $ n \geq 1 $, the $ L^\infty $-norm of the signed sum $ \sum a_R h_R $, where $ |R| = 2^{-n} $ and $ |R_1| \geq 2^{-n/2} $, satisfies $ \left\| \sum a_R h_R \right\|_{L^\infty} \gtrsim n^{9/8} $.
• This lower bound improves upon the trivial $ L^2 $-based estimate of $ n $, which arises from the number of rectangles at scale $ 2^{-n} $.
• The exponent $ 9/8 $ is the best known lower bound for the signed small ball inequality in three dimensions, to the authors' knowledge.
• The result supports the conjectured sharp bound of $ n^{3/2} $, which would match the expected behavior in the small ball inequality for $ d=3 $.
• The inequality is motivated by and contributes to recent advances in the theory of star-discrepancy in dimensions $ d \geq 3 $, where a logarithmic exponent improvement over Roth's bound is now known.
• The result is derived using deep tools from harmonic analysis, including Haar function theory, dyadic martingale differences, and entropy estimates in product spaces.