[Paper Review] Global Uniqueness for an Inverse Stochastic Hyperbolic Equation with Three Unknowns

Qi Lü, Xu Zhang|arXiv (Cornell University)|Jul 17, 2011

Numerical methods in inverse problems7 references3 citations

TL;DR

This paper establishes global uniqueness for an inverse stochastic hyperbolic problem with three unknowns—random force intensity $ g $, initial displacement $ z_0 $, and initial velocity $ z_1 $—by deriving a new global Carleman estimate for stochastic hyperbolic equations. The key result shows that zero boundary flux and zero terminal displacement almost surely imply $ g = 0 $, $ z_0 = 0 $, and $ z_1 = 0 $, resolving the inverse problem uniquely under suitable geometric and regularity conditions.

ABSTRACT

This paper is addressed to an inverse stochastic hyperbolic equation with three unknowns, i.e., a source term, an initial displacement and an initial velocity. The global uniqueness is proved by a new global Carleman estimate for the stochastic hyperbolic equation. It is found that both the formulation of stochastic inverse problems and the tools to solve them differ considerably from their deterministic counterpart.

Motivation & Objective

• To address the global uniqueness of an inverse stochastic hyperbolic problem involving three unknowns: random force intensity $ g $, initial displacement $ z_0 $, and initial velocity $ z_1 $.
• To establish conditions under which the unknowns can be uniquely determined from partial boundary observations $ \partial z/\partial\nu|_{(0,T)\times\Gamma_0} $ and terminal displacement $ z(T) $.
• To bridge the gap between deterministic and stochastic inverse problems by developing tools tailored to the stochastic setting, particularly for hyperbolic equations with multiplicative noise.
• To prove that zero boundary flux and zero terminal state imply all unknowns vanish almost surely, ensuring global uniqueness.

Proposed method

• Derives a new global Carleman estimate for stochastic hyperbolic equations with multiplicative noise, tailored to handle the stochastic nature of the problem.
• Introduces a weighted energy identity involving the stochastic process $ z $, its time derivative $ z_t $, and spatial gradients, using a weight function $ \theta $ and parameter $ \lambda $.
• Applies the Itô formula to the weighted energy identity to derive a stochastic integral identity involving $ dz_t $, $ (b^{ij}z_{x_i})_{x_j} $, and the noise term $ (b_4 z + g)dB(t) $.
• Uses the boundary condition $ z = 0 $ on $ \Sigma $ to simplify the boundary integral terms and isolate the normal derivative $ \partial z/\partial\nu $.
• Establishes a lower bound on the weighted energy involving $ \lambda^3 z^2 $, $ \lambda |\nabla z|^2 $, $ \lambda z_t^2 $, and $ g^2 $, using the structure of the stochastic differential equation and the Carleman estimate.
• Combines the Carleman estimate with the assumption of zero boundary flux and zero terminal state to deduce that all unknowns must vanish almost surely.

Experimental results

Research questions

• RQ1 Can the random force intensity $ g $, initial displacement $ z_0 $, and initial velocity $ z_1 $ be uniquely determined from partial boundary observations and terminal state data in a stochastic hyperbolic system?
• RQ2 Does the vanishing of the normal derivative $ \partial z/\partial\nu $ on a subset $ \Gamma_0 \subset \Gamma $ and the terminal state $ z(T) = 0 $ imply that all three unknowns vanish almost surely?
• RQ3 How does the stochastic nature of the hyperbolic equation—specifically the multiplicative noise term $ g dB(t) $—affect the uniqueness of the inverse problem compared to the deterministic case?
• RQ4 What novel analytical tools are required to handle the stochastic inverse problem, and how do they differ from classical deterministic methods?
• RQ5 Can a global Carleman estimate be constructed for stochastic hyperbolic equations to control the solution and its derivatives in terms of boundary and terminal data?

Key findings

• The global uniqueness of the inverse stochastic hyperbolic problem is established: if $ \partial z/\partial\nu = 0 $ on $ (0,T) \times \Gamma_0 $ and $ z(T) = 0 $ in $ G $ almost surely, then $ g = 0 $, $ z_0 = 0 $, and $ z_1 = 0 $ in $ G $ almost surely.
• A new global Carleman estimate for stochastic hyperbolic equations is derived, which controls the weighted $ L^2 $-norms of $ z_t $, $ \nabla z $, and $ z $, as well as the noise intensity $ g $, in terms of boundary and terminal data.
• The estimate is constructed using a carefully chosen weight function $ \theta $, a large parameter $ \lambda $, and the stochastic Itô formula, leading to a lower bound involving $ \lambda^3 z^2 $, $ \lambda |\nabla z|^2 $, $ \lambda z_t^2 $, and $ g^2 $.
• The boundary integral term simplifies due to the Dirichlet condition $ z = 0 $ on $ \Sigma $, allowing the normal derivative $ \partial z/\partial\nu $ to be isolated and used in the energy estimate.
• From the vanishing of the boundary flux and terminal state, the estimate implies $ \mathbb{E} \int_G \theta^2 (\lambda |z_1|^2 + \lambda |\nabla z_0|^2 + \lambda^3 |z_0|^2) dx = 0 $, leading to $ z_0 = z_1 = 0 $ a.s.
• The term $ \mathbb{E} \int_Q (T-t) \theta^2 g^2 dx dt = 0 $ implies $ g = 0 $ almost surely in $ Q $, completing the proof of global uniqueness.