Quantitative stability for eigenvalues of Schr\"{o}dinger operator, Quantitative bathtub principle \& Application to the turnpike property for a bilinear optimal control problem

Idriss Mazari, Domènec Ruiz-BaletarXiv 2020 · arXiv

TL;DR

This paper bridges the gap between spectral optimization and optimal control theory by establishing a quantitative inequality for Schrödinger operators and applying it to prove a "turnpike property" for bilinear control systems. This work provides a novel approach to understanding the behavior of complex control systems, particularly those with bilinear dynamics, which have been less explored than their linear counterparts.

What Problem Does This Paper Address?

The paper tackles two main problems:

  1. Quantitative inequalities for spectral optimization: The authors investigate how much the first eigenvalue of a Schrödinger operator changes when the potential function is perturbed, subject to constraints on the potential's $L^\infty$ and $L^1$ norms. This is crucial for understanding the stability of optimal configurations in physical systems described by such operators.
  2. Turnpike property for bilinear optimal control: The turnpike phenomenon in optimal control suggests that for long time horizons, optimal control trajectories tend to stay close to a stationary optimal control. This paper aims to prove this property for bilinear control systems, which are more complex than linear systems and have received less attention in the context of turnpike phenomena. Specifically, they want to show that the optimal control at any given time is close to a static optimal control derived from the spectral optimization problem.

Methods

The authors employ a combination of advanced mathematical techniques:

  • Quantitative Bathtub Principle: This is a novel method introduced by the authors, which is central to proving the quantitative inequalities for the spectral optimization problem. It provides a way to bound the difference between a function and its optimal counterpart based on certain geometric properties.
  • Shape Derivatives: The paper extensively uses shape derivatives (first and second order) to analyze how the spectral eigenvalue changes with respect to deformations of the domain's boundary. This is a standard tool in shape optimization.
  • Lagrangian Multipliers: Used to handle the volume constraint in the optimization problems.
  • Bootstrap Argument: Employed in the technical preliminaries to establish regularity estimates for eigenfunctions.
  • Schwarz Rearrangement: Used in the proof of the bathtub principle to transform functions into radially symmetric ones, simplifying analysis.

The approach is novel in its direct application of quantitative spectral inequalities to prove turnpike properties for bilinear control, bypassing traditional methods that rely on explicit characterization of optimal controls.

Datasets

Not discussed in the paper. The paper is theoretical and focuses on mathematical proofs and concepts rather than empirical evaluation on datasets.

Key Insights and Ideas

  • Bridging Spectral Optimization and Optimal Control: The core insight is the direct link established between quantitative inequalities for spectral optimization problems and the turnpike property in bilinear optimal control. This suggests a powerful general framework for analyzing complex control systems.
  • Quantitative Bathtub Principle as a Key Tool: The introduction of this principle is a significant methodological contribution, enabling the derivation of quantitative stability estimates where previous methods might have fallen short.
  • Turnpike Property for Bilinear Systems: Proving the turnpike property for bilinear control systems is a notable achievement, as these systems are more challenging to analyze than linear ones. This opens up new avenues for understanding and controlling such systems.
  • Bang-Bang Solutions: The paper highlights that optimal potentials in the spectral optimization problem are "bang-bang" functions (taking only two values), which simplifies their analysis and characterization.

Key Results

  • Theorem I: Establishes a quantitative inequality for the spectral optimization problem. Specifically, for a potential $V$, the difference between its spectral eigenvalue $\lambda(V)$ and the optimal eigenvalue $\lambda(V^*)$ is bounded below by a constant times the squared $L^1$ distance between $V$ and the set of optimal potentials $I^*$. This quantifies the stability of the optimal spectral configuration.
  • Theorem II: Proves an integral turnpike property for bilinear optimal control problems. It shows that for a sufficiently large time horizon $T$, the $L^1$ distance between the optimal control trajectory $V(t, \cdot)$ and the set of static optimal potentials $I^*$ is bounded uniformly in $T$. This means the control stays close to a static optimal configuration for most of the time.

Limitations and Future Work

  • Regularity Assumptions: The authors acknowledge that their results rely on certain regularity assumptions on the domain $\Omega$ and its boundary (e.g., $C^2$ boundedness, specific properties of optimal spectral sets). They suggest that exploring weaker regularity assumptions would be an interesting direction for future work.
  • Generalization to Other Systems: The authors express interest in extending their approach to other types of optimal control problems, such as parabolic and hyperbolic systems, using similar techniques from sensitivity analysis.
  • Explicit Computations: While the paper provides theoretical results, explicit computations for optimal controls are only possible in very specific geometries (e.g., a ball). The turnpike property offers a way to gain qualitative information in more general cases where explicit solutions are intractable.

Summary updated

Abstract

International audience
BibTeX
@misc{mazari2020quantitative,
  author = {Idriss Mazari and Domènec Ruiz-Balet},
  title = {{Quantitative stability for eigenvalues of Schr\textbackslash\{\}"\{o\}dinger operator, Quantitative bathtub principle \textbackslash\{\}\& Application to the turnpike property for a bilinear optimal control problem}},
  year = {2020},
  eprint = {2010.10798},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2010.10798},
}
Figure