New global Carleman estimates and null controllability for forward/backward semi-linear parabolic SPDEs

Basic Information

• Paper ID: 2401.13455
• Title: New global Carleman estimates and null controllability for forward/backward semi-linear parabolic SPDEs
• Authors: Lei Zhang, Fan Xu, Bin Liu (School of Mathematics and Statistics, Huazhong University of Science and Technology)
• Classification: math.OC (Mathematics - Optimal Control), math.PR (Mathematics - Probability Theory)
• Submission Date: January 2024
• Paper Link: https://arxiv.org/abs/2401.13455

Abstract

This paper investigates the null controllability problem for parabolic stochastic partial differential equations (SPDEs) involving the state and its gradient. First, improved global Carleman estimates are established for linear forward (backward) parabolic SPDEs with general random coefficients and square-integrable source terms. Based on this foundation, new global Carleman estimates are developed for linear forward (backward) parabolic SPDEs with source terms in negative-order Sobolev spaces, enabling the treatment of global null controllability for linear backward (forward) parabolic SPDEs with gradient terms. As a byproduct, explicit weighted energy-type estimates depending on parameters λ, μ, and weight function θ are obtained for the controlled system, making it possible to extend linear null controllability to semi-linear backward (forward) parabolic SPDEs through fixed-point arguments in appropriate Banach spaces.

Research Background and Motivation

Importance of the Problem

1. Theoretical Challenge: Carleman estimates, originally introduced by T. Carleman to study unique continuation for elliptic PDEs, have become powerful tools for studying PDEs, inverse problems, and control problems. For stochastic partial differential equations, establishing such estimates presents significant challenges.
2. Technical Difficulties: Compared to the deterministic case, the main difficulties in the stochastic setting are:
   • Loss of temporal regularity of solutions
   • Absence of compact embedding properties in the state space
   • Making the fixed-point arguments commonly used in the deterministic case inapplicable
3. Practical Demands: Such SPDEs have important applications in stochastic optimal control and filtering theory, attracting widespread attention from the SPDE theory perspective.

Limitations of Existing Methods

1. Challenges in Literature 40, 37: As pointed out in the pioneering work of Tang and Zhang, the main challenge in extending deterministic results to the stochastic setting is the loss of temporal regularity of solutions and the absence of compact embeddings in the state space.
2. Difficulties with Gradient-Dependent Terms: Existing L²-Carleman estimates 20, 40 cannot be directly applied to handle gradient-dependent terms, and newly established H⁻¹-Carleman estimates 4, 30 remain insufficient for solving nonlinear controllability problems.
3. Coefficient Regularity: The appearance of general random coefficients introduces additional interaction terms requiring detailed and precise analytical work.

Core Contributions

1. Improved L²-Carleman Estimates: By employing appropriate weight functions, new global Carleman estimates are established for forward (backward) linear parabolic SPDEs with general random coefficients and L² source terms.
2. H⁻¹-Carleman Estimates: Using duality arguments and Lions' penalty HUM method, new global Carleman estimates are derived for forward (backward) parabolic SPDEs with source terms in L²_F(0,T;H⁻¹(O)).
3. Linear Null Controllability: Based on H⁻¹-Carleman estimates, global null controllability is established for linear backward (forward) parabolic SPDEs involving the state and gradient, while obtaining interesting energy-type estimates related to parameters λ, μ, and weight function θ.
4. Semi-linear Null Controllability: Through fixed-point arguments (without using compact embedding results from deterministic counterparts), global null controllability is proved for semi-linear backward (forward) parabolic SPDEs.

Methodology Details

Problem Formulation

The null controllability of the following semi-linear backward parabolic SPDE is studied:

dy + ∇·(A∇y)dt = (F(ω,t,x,y,∇y,Y) + 1_{O'}u)dt + YdW_t  in O_T
y = 0  on Σ_T, y(T) = y_T  in O

where (y,Y) is the unique solution associated with the control variable u and terminal state y_T.

Weight Function Construction

Inspired by literature 20, for parameters m ≥ 1 and μ ≥ 1, the weight functions are:

• φ(x,t) = γ(t)(e^{μ(β(x)+6m)} - μe^{6μ(m+1)})
• ξ(x,t) = γ(t)e^{μ(β(x)+6m)}
• θ(x,t) = e^{ℓ(x,t)}, ℓ(x,t) = λφ(x,t)

where γ(t) is a piecewise C² function with specific monotonicity and singularity properties in different time intervals.

Core Technical Innovations

1. Improved L²-Carleman Estimates (Theorem 1.2)

For the linear forward system, the following estimate is established:

E∫_O e^{2λφ(T)}(λ²μ³e^{2μ(6m+1)}z²(T) + |∇z(T)|²)dx 
+ E∫_{O_T} λμ²ξθ²(|∇z|² + λ²μ²ξ²z²)dxdt
≤ C[E∫_{O_T} θ²(λ²μ²ξ³φ₂² + |∇φ₂|² + φ₁²)dxdt + E∫_{O'_T} λ³μ⁴ξ³θ²z²dxdt]

2. H⁻¹-Carleman Estimates (Theorem 1.4)

By combining L²-Carleman estimates, Lions' HUM method, and duality arguments:

E∫_O λμ²(ξθ²)(T)z²(T)dx + E∫_{O_T} λμ²ξθ²(λ²μ²ξ²z² + |∇z|²)dxdt
≤ C[E∫_{O'_T} λ³μ⁴ξ³θ²z²dxdt + E∫_{O_T} θ²φ₁² + λ²μ²ξ²θ²(φ₂² + |b|²)dxdt]

3. Fixed-Point Argument Framework

Define a mapping in the weighted Banach space B_{λ,μ}:

K: φ ∈ B_{λ,μ} ↦ F(ω,t,x,y,∇y,Y) ∈ B_{λ,μ}

By choosing sufficiently large λ, μ such that Cλ⁻¹μ⁻² < 1, K becomes a contraction mapping, and the Banach fixed-point theorem guarantees the existence of a unique fixed point.

Main Theorems

Theorem 1.6 (Linear Null Controllability)

Assume condition (A1) holds. Then for each terminal state y_T ∈ L²_(Ω;L²(O)), there exists a control û ∈ L²_F(0,T;L²(O')) such that the corresponding solution (ŷ,Ŷ) satisfies ŷ(0) = 0.

Theorem 1.8 (Semi-linear Null Controllability)

Assume conditions (A1)-(A2) hold. Then for any y_T ∈ L²_(Ω;L²(O)), there exists a control variable u ∈ L²_F(0,T;L²(O')) such that the corresponding unique solution (y,Y) satisfies y(0) = 0.

Theorem 1.12 (Forward Semi-linear Null Controllability)

Assume conditions (A1) and (A3) hold. Then for each initial state y₀ ∈ L²_{F₀}(Ω;L²(O)), there exists a control pair (u,U) such that the unique solution y satisfies y(T) = 0.

Technical Difficulties and Solutions

1. Complexity Introduced by Random Coefficients

General random coefficients a^{ij}(ω,t,x) introduce additional interaction terms, including a^{ij}_, a^{ij}t, and a^{ij} terms, which do not appear in the constant coefficient Laplacian case.

Solution: These additional terms are handled through refined estimation techniques, particularly in steps 2-3 where estimates of J_i (i=1,...,5) are performed.

2. Non-Classical Choice of Weight Function

Unlike classical weight functions 4,40, a weight function that is non-degenerate at t=T is employed, resulting in the exponent of ξ in the Carleman estimate being cubic rather than quadratic.

Solution: Careful analysis of weight function behavior in different time intervals, particularly the sign of γ_t and estimates of (2.15d) in each interval.

3. Gradient-Dependent Nonlinear Terms

Handling gradient dependence in F(ω,t,x,y,∇y,Y) requires H⁻¹-Carleman estimates rather than standard L² estimates.

Solution: A new Carleman estimate framework is established by combining Lions' penalty HUM method and duality arguments.

Comparison with Related Work

Relationship with Literature 20

• Improvement: Considers general random coefficients rather than only constant coefficient cases
• Extension: Handles gradient-dependent nonlinear terms
• Technique: Employs similar weight function construction but requires handling more technical details

Relationship with Literature 4,30

• Development: Extends linear results to nonlinear cases
• Method: Combines multiple techniques (Carleman estimates, HUM method, duality arguments)
• Application: Solves more general control problems

Relationship with Literature 40

• Foundation: Built upon the pioneering work of Tang-Zhang
• Innovation: Handles gradient-dependent terms and negative-order Sobolev space source terms
• Contribution: Partially answers open questions in 40, Remark 2.5

Conclusions and Discussion

Main Conclusions

1. Successfully establishes null controllability theory for semi-linear parabolic SPDEs involving the state and gradient
2. Develops new Carleman estimate techniques capable of handling source terms in negative-order Sobolev spaces
3. Avoids the use of compact embeddings through clever fixed-point arguments

Limitations

1. Coefficient Regularity: Requires W^{2,∞} regularity of principal coefficients; minimal regularity conditions remain undetermined
2. Nonlinear Growth: Only handles Lipschitz nonlinearity; superlinear nonlinearity cases remain open
3. Control Location: Forward case requires additional control U on the diffusion term

Future Directions

1. Regularity Optimization: Study minimal regularity requirements for coefficients
2. Nonlinear Extension: Consider nonlinear terms with superlinear growth
3. Control Optimization: Investigate whether the number or scope of control variables can be reduced
4. Higher-Order Equations: Extend to fourth-order parabolic random equations

In-Depth Evaluation

Strengths

1. Theoretical Contribution: Achieves significant progress in stochastic control theory, partially resolving open problems in the field
2. Technical Innovation: Cleverly combines multiple mathematical tools, establishing a new analytical framework
3. Complete Methodology: Provides a complete theoretical chain from linear to semi-linear cases
4. Clear Presentation: Technical details are thorough and proofs are rigorous

Weaknesses

1. Technical Limitations: High regularity requirements for coefficients may limit practical applications
2. Control Cost: Certain cases require additional control variables, increasing implementation complexity
3. Open Problems: Important theoretical questions remain unresolved

Impact

1. Academic Value: Provides important theoretical tools for stochastic control theory
2. Methodological Contribution: The established technical framework may be applicable to other types of random equations
3. Application Prospects: Potential applications in stochastic optimal control and filtering theory

Applicable Scenarios

1. Theoretical Research: Control theory research for stochastic partial differential equations
2. Numerical Methods: Provides theoretical foundation for numerical algorithm design
3. Engineering Applications: Distributed parameter control systems with stochastic perturbations

References

The paper cites 50 important references covering classical and frontier works in Carleman estimates, stochastic partial differential equations, and control theory, reflecting the depth and breadth of the research.