Boundedness of Functions on Product Spaces by Sums of Functions on the Factors

Stefan Born, Alexander DirmeierarXiv 2013 · DOI · arXiv

Abstract

We investigate sufficient conditions for real-valued functions on product spaces to be bounded from above by sums or products of functions which depend only on points in the respective factors.

Introduction

In mathematical analysis and multivariable calculus, it is often useful to decompose or bound a complex function $f(t, x)$ using functions of fewer variables. A common "folk theorem" suggests that any continuous real-valued function on a product of two "nice" topological spaces can be bounded from above by a sum of functions $F(t)$ and $G(x)$, where each depends only on a single factor. While this result is frequently assumed in various subfields, a rigorous treatment of the necessary topological conditions has been largely absent from the literature. This paper addresses this gap by identifying the specific requirements—local compactness and countability at infinity—needed to guarantee the existence of such bounds.

Problem Formulation

The central question involves the "coupling" of variables. Given a function $f: M \times N \to \mathbb{R}$, we seek to determine if there exist continuous functions $F: M \to \mathbb{R}$ and $G: N \to \mathbb{R}$ such that:

$$f(t, x) \leq F(t) + G(x)$$

for all $(t, x) \in M \times N$. This problem is particularly relevant in optimal transport theory, where cost functions are often analyzed based on whether they can be bounded by sums of functions on their marginal spaces. Establishing the conditions under which this decoupling is possible provides a more robust theoretical foundation for these applications.

Methodology

The construction of the bounding functions $F$ and $G$ relies on two fundamental tools from topology and analysis:

  1. Compact Exhaustions: The authors assume the spaces $M$ and $N$ are $\sigma$-compact (countable at infinity). This allows the spaces to be represented as a nested sequence of compact sets $K_1 \subset K_2 \subset \dots$ that eventually grow to cover the entire space.
  2. Partitions of Unity: To ensure the resulting functions $F$ and $G$ are continuous, the authors use partitions of unity. This technique allows them to take local maximum values of $f$ on specific compact regions and "glue" them together into globally defined continuous functions.

The proof strategy involves evaluating the maximum value of $f$ on expanding "rectangles" within the product space $M \times N$ and then distributing these maximums into the factor functions $F$ and $G$ using the partition of unity to maintain smoothness or continuity.

Key Results

The Main Theorem

The primary contribution of the paper is the formalization of the following theorem: For locally compact Hausdorff spaces $M$ and $N$ that are countable at infinity, any continuous function $f: M \times N \to \mathbb{R}$ has continuous bounds $F: M \to \mathbb{R}$ and $G: N \to \mathbb{R}$ such that $f(t, x) \leq F(t) + G(x)$.

Extensions to Smooth Manifolds

The authors further demonstrate that if $M$ and $N$ are differentiable manifolds, the construction can be refined. In such cases, the bounding functions $F$ and $G$ can be chosen to be smooth ($C^\infty$), preserving the regularity of the underlying spaces beyond simple continuity.

Multiplicative Bounds

The additive result naturally extends to a multiplicative one for positive functions. By applying an exponential transformation to the additive bound, the authors show that for a positive continuous function $f$, there exist positive continuous functions $F$ and $G$ such that $f(t, x) \leq F(t) \cdot G(x)$.

Counter-examples and Constraints

The Necessity of Local Compactness

A significant insight of the paper is that local compactness is a non-negotiable requirement. The authors provide a counter-example using an evaluation map on a specific function space. They demonstrate that if a space is paracompact but lacks local compactness, the theorem fails. This highlights that the "local" structure of the space is just as critical as its "global" countability.

Failure in $L^1$ Spaces

The paper highlights a crucial boundary for this logic: it does not extend to integrable functions. The authors construct a specific function on $\mathbb{R}^2$ that is $L^1$ (integrable) but cannot be bounded by the sum of two $L^1$ functions on the respective factors. This distinction is vital for researchers working in measure theory and functional analysis.

Conclusion

This research formalizes a widely used intuition in analysis, providing clear sufficient conditions (locally compact, Hausdorff, and $\sigma$-compact) for the existence of additive bounds on product spaces. While the paper establishes these sufficient conditions, the authors note that the necessary conditions remain an open question. Future work may explore whether these bounds hold in even more general topological settings or further integrate these findings into specific problems within optimal transport.

Summary updated

Abstract

We investigate sufficient conditions for real-valued functions on product spaces to be bounded from above by sums or products of functions which depend only on points in the respective factors.
BibTeX
@misc{born2013boundedness,
  author = {Stefan Born and Alexander Dirmeier},
  title = {{Boundedness of Functions on Product Spaces by Sums of Functions on the Factors}},
  year = {2013},
  eprint = {1401.0130},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1401.0130},
}
Figure