Sparse control for VCHE with abstract J

Abstract: We investigate a distributed optimal control problem for the viscous Camassa--Holm equations with sparse controls and a general cost functional. Considering three different forms of sparsity-promoting terms, we prove the existence of optimal solutions, derive the corresponding optimality conditions and analyze the stability of optimal solutions with respect to the sparsity parameter.

• The paper introduces a novel framework for sparse optimal control of the 3D VCHE, demonstrating existence and precise first-order optimality conditions using L¹-based penalties.
• It establishes well-posedness, regularity, and second-order stability, proving convergence rates of O(κ) for some formulations and O(κ^(2/3)) for others.
• By leveraging advanced subdifferential calculus and adjoint techniques, the study derives practical projection formulas for efficient numerical implementation.

Sparse Optimal Control for the 3D Viscous Camassa-Holm Equations: A Technical Overview

Problem Setting and Motivation

The paper addresses distributed optimal control of the three-dimensional viscous Camassa-Holm equations (VCHE, also known as LANS-α or Navier-Stokes-α) on bounded domains with smooth boundary, focusing on the case of sparsity-promoting controls. These equations are regularized modifications of Navier-Stokes that preserve essential vorticity and circulation properties, with applications ranging from turbulence modeling to geophysical flows. The core challenge is to design controls $u$ that are spatially or temporally sparse—active only on small subsets of the domain or time interval—while ensuring effective manipulation of the system’s evolution.

The cost functional considered generalizes standard quadratic forms, incorporating three distinct sparsity-inducing terms:

• $j_1(u) = \|u\|_{L^1(Q)^3}$: standard $L^1$ penalty promoting joint space-time sparsity.
• $j_2(u) = \|u\|_{L^2(0,T;L^1(\Omega)^3)}$: time-sparse (group-Lasso-like) penalty.
• $j_3(u) = \|u\|_{L^1(\Omega;L^2(0,T)^3)}$: space-sparse penalty.

This framework leads to three families of optimal control problems—$(P_1)$, $(P_2)$, $(P_3)$—subject to a standard box constraint on the control amplitude, and dynamic constraints given by the VCHE. Notably, the cost functional is allowed to be strictly convex and twice continuously Fréchet differentiable in the state variable $y$, broadening the admissible class of objective functions beyond recent literature.

Analytical Results

Well-posedness, Regularity, and Control-to-state Map

The paper establishes the well-posedness of VCHE with controls in $L^2$-spaces and derives crucial regularity results for weak solutions in appropriate Sobolev-Bochner spaces. The control-to-state operator $j_1(u) = \|u\|_{L^1(Q)^3}$0 is shown to be twice continuously Fréchet differentiable from $j_1(u) = \|u\|_{L^1(Q)^3}$1-controls to the regularity class of states, permitting the use of standard sensitivity analysis and the derivation of first- and second-order optimality conditions via the adjoint approach.

The paper rigorously analyzes the nonlinear terms, specifically the trilinear operator structure, using detailed properties of the Stokes operator and fractional powers for the embedding and compactness results.

Existence of Optimal Solutions

For each of the three sparsity-promoting formulations, the paper proves existence of optimal controls using direct methods of calculus of variations. Weak lower semicontinuity of the objectives—owing to convexity and continuity of both the cost $j_1(u) = \|u\|_{L^1(Q)^3}$2 and the $j_1(u) = \|u\|_{L^1(Q)^3}$3-type penalty—is essential for the passage to the limit in minimizing sequences.

First-Order Necessary Optimality Conditions

The main technical contribution is the derivation of precise first-order optimality conditions for all three sparsity functional cases. The result is a system involving:

1. The state equation (VCHE);
2. The adjoint equation, a backward-in-time linearized VCHE with nontrivial final conditions;
3. A variational inequality or projection formula for the optimal control, incorporating the subdifferential of the sparsity functional (expressed explicitly for each case).

For $j_1(u) = \|u\|_{L^1(Q)^3}$4, the first-order condition leads to, almost everywhere,

$j_1(u) = \|u\|_{L^1(Q)^3}$5

with $j_1(u) = \|u\|_{L^1(Q)^3}$6 encoding the subdifferential of $j_1(u) = \|u\|_{L^1(Q)^3}$7 and enforcing sparsity. Analogous expressions are produced for $j_1(u) = \|u\|_{L^1(Q)^3}$8 and $j_1(u) = \|u\|_{L^1(Q)^3}$9, with explicit subdifferential characterizations given.

Structural Implications:

• For sufficiently large $L^1$0, the optimal control is identically zero, and for $L^1$1 below a threshold determined by the adjoint, the sparsity structure can exhibit "bang-bang-bang" (take values at the bounds or zero).
• Temporal and spatial supports of the optimal control are precisely connected to level sets of the adjoint variable, depending on the choice of sparsity norm.

Second-order Optimality Conditions

The authors derive both necessary and sufficient second-order conditions exploiting the differentiability of the reduced objective functional and the structure of the so-called critical cone. The results characterize when strict local quadratic growth holds in the presence of nonsmooth $L^1$2 penalties (via directional derivatives and variational inequalities), and connect this to strong stability of the local minimum.

Stability with Respect to Sparsity Parameter

A novel and significant part of the analysis concerns the effect of the sparsity parameter $L^1$3. Theorems demonstrate:

• Strong convergence (in $L^1$4 and even $L^1$5 norms) of sparse optimal controls to optimizers of the non-sparse problem as $L^1$6.
• Quantitative rates of convergence: Lipschitz stability ($L^1$7) for $L^1$8 and $L^1$9, and Hölder stability ($j_2(u) = \|u\|_{L^2(0,T;L^1(\Omega)^3)}$0) for $j_2(u) = \|u\|_{L^2(0,T;L^1(\Omega)^3)}$1.
• Techniques include detailed estimates of terms arising from the sparsity penalties and the use of compactness and monotonicity arguments.

Numerical and Theoretical Implications

The explicit characterization of sparsity in space and/or time for controls of regularized fluid systems provides both a theoretical foundation for practical actuator placement (minimal intervention) and an efficient route to numerical solution algorithms: the subdifferential calculations and projection formulas can be directly used in nonsmooth optimization schemes (e.g., semismooth Newton methods, active set strategies).

The detailed stability analysis implies robust performance and support structure of controls as penalization parameters are tuned, answering an open question in the field about the persistence of sparse structures and their sensitivity.

On a theoretical front, the extension to general convex objectives and the inclusion of abstract regularization terms paves the way for further generalizations to hybrid functionals, measure-valued controls, or more complex parabolic/hyperbolic PDEs.

Outlook and Open Directions

Given the generality and rigor of the analytic approach, several avenues for future research are natural:

• Extension to state constraints: Many practical scenarios require state constraints; integrating sparsity with pointwise or integral state constraints remains challenging.
• Numerical strategies: The explicit structure of subdifferentials can be leveraged for efficient large-scale solvers, including inexact or adaptive methods.
• Nonlinear and stochastic cost functionals: Broadening the functional class to stochastic settings or incorporating uncertainty quantification would enhance relevance to real-world applications (e.g., weather prediction, turbulence control).
• Measure-valued and actuator design problems: For truly minimal actuator design, measure-valued controls linked directly to Dirac measures and their approximations should be considered, requiring further functional analytic development.

Conclusion

This paper provides a comprehensive functional analytic treatment of sparse optimal control for the three-dimensional viscous Camassa-Holm equations with various $j_2(u) = \|u\|_{L^2(0,T;L^1(\Omega)^3)}$2-type regularization schemes. The authors derive sharp conditions for existence, first- and second-order optimality, and strong quantitative stability of optimal controls with respect to the sparsity parameter. This advances both the theoretical understanding and practical applicability of sparsity-promoting control methods in the context of nonlinear PDEs modeling fluid dynamics.