TL;DR
This paper establishes a crucial link between spectral optimization problems for Schrödinger operators and the "turnpike property" in bilinear optimal control. By proving new quantitative inequalities for spectral optimization, the authors demonstrate an integral turnpike property for bilinear control systems, showing that optimal controls tend to stay close to a stationary optimal state over time.
What Problem Does This Paper Address?
The paper tackles two interconnected problems in mathematical physics and control theory. First, it investigates how stable the optimal potential is for minimizing the first eigenvalue of a Schrödinger operator under specific constraints ($L^\infty$ and $L^1$ on the potential). This is a "shape optimization" problem where the "shape" is determined by the potential. Second, it addresses the "turnpike property" in bilinear optimal control. This property suggests that for long time horizons, optimal control trajectories tend to stay close to a stationary optimal control, rather than constantly deviating. The challenge lies in proving these properties for bilinear systems, which are more complex than linear ones, and under $L^\infty$ and $L^1$ constraints, which are not standard in existing turnpike literature.
Methods
The core of the paper's contribution lies in its novel approach to proving quantitative inequalities for spectral optimization. This is achieved through a "quantitative bathtub principle," a new technical tool developed by the authors. This principle, combined with assumptions on the domain's geometry and the behavior of optimal shapes (bang-bang property), allows them to derive local quantitative inequalities for deformations of optimal sets. These inequalities are then used to prove the main spectral optimization result. Subsequently, this spectral result is leveraged to establish the integral turnpike property for bilinear optimal control problems. The authors also employ standard techniques like the Rayleigh quotient, shape derivatives, and bootstrap arguments for regularity estimates.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Quantitative Bathtub Principle: This is a novel technical tool that forms the backbone of the paper's spectral optimization results. It provides a way to quantify how much the eigenvalue changes when a shape is deformed, linking it to the $L^1$ distance of the deformation.
- Bridging Spectral Optimization and Turnpike Property: The paper demonstrates a powerful connection: quantitative inequalities in spectral optimization directly imply an integral turnpike property for bilinear control. This is a significant conceptual advance, showing how insights from one area can illuminate another.
- Bilinear Control Turnpike: Proving the turnpike property for bilinear control systems is a notable achievement, as existing methods are often tailored for linear systems. The paper shows that even with the added complexity of bilinear control and $L^\infty$/$L^1$ constraints, the turnpike phenomenon still holds.
- Bang-Bang Property of Optimal Potentials: The paper relies on the fact that optimal potentials for the spectral optimization problem are "bang-bang," meaning they take only two values (0 and 1 in this case). This property simplifies the analysis and is crucial for relating optimal potentials to optimal spectral sets.
Key Results
- Quantitative Inequality for Spectral Optimization: The paper proves that for a Schrödinger operator with Dirichlet boundary conditions, the difference between the eigenvalue for any potential $V$ and the optimal potential $V^*$ is bounded below by a constant times the square of the $L^1$ distance between $V$ and the set of optimal potentials $I^*$. This is Theorem I.
- Integral Turnpike Property for Bilinear Control: The paper proves that for a bilinear optimal control problem, the integral of the squared $L^1$ distance between the optimal control trajectory and the set of stationary optimal controls is bounded uniformly in the time horizon $T$. This is Theorem II.
Limitations and Future Work
- Regularity Assumptions: The authors acknowledge that their results rely on certain regularity assumptions on the domain and optimal spectral sets (e.g., smooth boundaries, finite perimeter). They suggest that exploring whether these assumptions can be relaxed (e.g., using recent results for Lipschitz sets) would be an interesting avenue for future research, though it might present technical difficulties.
- Generalization of Bathtub Principle: While the quantitative bathtub principle is a key tool, its application here is specific to the spectral optimization problem. Exploring its broader applicability to other optimization problems could be fruitful.
- Other Control Systems: The paper focuses on bilinear control systems. Extending the findings to other types of nonlinear control systems (e.g., parabolic or hyperbolic) is mentioned as a future direction.
- Explicit Computations: The paper notes that explicit computations of optimal controls are generally hopeless except in very specific geometries. The turnpike property provides a valuable qualitative insight in these more general cases.
Summary updated