A Generalized Variational Principle

Philip D. Loewen, Xianfu WangCanadian 2001 · DOI

Abstract
We prove a strong variant of the Borwein-Preiss variational principle, and show that on Asplund spaces, Stegall's variational principle follows from it via a generalized Smulyan test. Applications are discussed.

Introduction

In the fields of optimization and nonsmooth analysis, variational principles serve as indispensable tools for handling functions that lack a classical minimum. The core utility of these principles, such as the seminal Ekeland’s Principle, is the ability to assert that a slight perturbation of a lower semicontinuous function will result in a new function that attains a true minimum. However, the nature of this minimum—specifically its stability and the behavior of sequences approaching it—remains a critical point of study in infinite-dimensional analysis.

This paper by Loewen and Wang addresses existing gaps between the Borwein-Preiss principle, which offers smooth perturbations but lacks a guarantee of "strong" minimality, and the Deville-Godefroy-Zizler (DGZ) principle, which proves the existence of strong minimizers but does so non-constructively. The authors bridge this gap by providing an explicit construction for a perturbation that ensures a strong minimizer at a known location.

The Challenge of Strong Minimality

In infinite-dimensional spaces, a function may possess a unique minimum without being stable. A "strong minimizer" is a more robust concept: it ensures that any sequence of points where the function value approaches the minimum (a minimizing sequence) must actually converge in norm to the minimizer itself.

Existing frameworks often forced a trade-off. The Borwein-Preiss principle provided a specific recipe for perturbations using power-of-the-norm functions, but it did not always force the convergence of minimizing sequences. Conversely, the DGZ principle established that "most" perturbations (in a Baire category sense) would result in a strong minimum, but it offered no specific construction for the penalty function or the location of the resulting minimizer.

Methodology: Constructing the Penalty Function

The authors employ a constructive inductive approach to build a specific penalty function, denoted as $\rho_\infty$. This process involves three primary components:

  1. Iterative Refinement: Starting from an initial "almost" minimizer, the authors define a sequence of nested sets ($D_n$) with diameters shrinking to zero, narrowing the search space for the minimizer.
  2. Infinite Series Perturbation: The final penalty function is constructed as an infinite sum of shifted and scaled versions of a base function $\rho$. This allows the authors to "sharpen" the function's profile around the desired minimizer, ensuring the strong minimality property.
  3. Generalized Šmulyan Test: To link these results to the underlying geometry of Banach spaces, the authors develop a generalized Šmulyan test. This mathematical tool connects the Fréchet differentiability of a function to the existence of a strong minimum, providing a bridge between the smoothness of the perturbation and the stability of the solution.

A Unified Framework for Variational Principles

A significant contribution of this work is its role as a "master key" for variational theory. By selecting different base functions for $\rho$, the authors demonstrate that their Strong Variant can recover several classic theorems:

  • Ekeland’s Theorem: Recovered by using a norm-based penalty.
  • Borwein-Preiss Principle: Recovered by using a power-of-the-norm penalty.
  • DGZ Principle: Recovered through the use of smooth bump functions.

Furthermore, the paper establishes that the ability to find these strong minimizers is a defining characteristic of specific mathematical environments. Specifically, the existence of such perturbations characterizes Asplund spaces and spaces possessing the Radon-Nikodym Property (RNP).

Key Results and Applications

The theoretical core of the paper is found in Theorem 2.2, which proves that for any lower semicontinuous function $f$, a perturbation exists such that the modified function has a strong minimizer $v$ located arbitrarily close to the original starting point.

In the latter half of the paper (Theorems 4.2 and 6.2), the authors simplify the proof of Stegall’s Variational Principle. They also explore the maximality of subdifferentials, proving that for a broad class of continuous functions, the set of points where the function is "highly nonsmooth"—where the approximate subdifferential is as large as possible—is topologically significant.

Conclusion and Future Directions

While the construction of the perturbation is mathematically explicit via an infinite series, it remains a theoretical abstraction. The paper does not address the computational complexity of implementing such perturbations in numerical optimization algorithms. However, by unifying disparate variational principles and providing a constructive path to strong minimality, the work offers a deeper understanding of the geometric structure of Banach spaces. Future work may involve exploring the utility of this principle in applied fields such as control theory and economics, where nonsmooth analysis is frequently employed to model complex systems.

Summary updated

Abstract

Abstract We prove a strong variant of the Borwein-Preiss variational principle, and show that on Asplund spaces, Stegall's variational principle follows from it via a generalized Smulyan test. Applications are discussed.
BibTeX
@inproceedings{loewen2001generalized,
  author = {Philip D. Loewen and Xianfu Wang},
  title = {{A Generalized Variational Principle}},
  booktitle = {Canadian},
  year = {2001},
  doi = {10.4153/cjm-2001-044-8},
}
Figure