Abstract:We propose to combine the Carleman estimate and the Newton method to solve an inverse source problem for nonlinear parabolic equations from lateral boundary data. The stability of this inverse source problem is conditionally logarithmic. Hence, numerical results due to the conventional least squares optimization might not be reliable. In order to enhance the stability, we approximate this problem by truncating the high frequency terms of the Fourier series that represents the solution to the governing equation. By this, we derive a system of nonlinear elliptic PDEs whose solution consists of Fourier coefficients of the solution to the parabolic governing equation. We solve this system by the Carleman-Newton method. The Carleman-Newton method is a newly developed algorithm to solve nonlinear PDEs. The strength of the Carleman-Newton method includes (1) no good initial guess is required and (2) the computational cost is not expensive. These features are rigorously proved. Having the solutions to this system in hand, we can directly compute the solution to the proposed inverse problem. Some numerical examples are displayed.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the **global inverse source problem for nonlinear parabolic equations**. Specifically, the authors aim to reconstruct the source term of the nonlinear parabolic equation from the lateral boundary data. This problem is very important in many practical applications, such as determining the spatial temperature distribution inside solids, monitoring the pollutant levels on the surfaces of rivers or lakes, and effectively monitoring the heat conduction processes in the steel industry, glass and polymer molding, and nuclear power plants.
### Problem Description
Given the lateral boundary data:
\[ g_0(x,t) = u(x,t) \quad \text{and} \quad g_1(x,t) = \frac{\partial u}{\partial \nu}(x,t) \]
for all \((x,t) \in \partial \Omega_T\), the goal is to determine the source term \( p(x) \), where \( \Omega \) is an open and bounded region, \( \partial \Omega \) is its smooth boundary, and \( \Omega_T=\Omega\times(0, T) \).
### Equation Form
Consider the initial value problem of the following nonlinear parabolic equation:
\[ \begin{cases}
u_t = \Delta u+F(x, u, \nabla u) & (x,t) \in \mathbb{R}^d\times(0, T), \\
u(x, 0) = p(x) & x \in \mathbb{R}^d.
\end{cases} \]
### Main Challenges
1. **Stability Problem**: Due to the existence of the nonlinear term \( F(x, u, \nabla u) \), traditional least - squares optimization methods may be unreliable because their stability is conditionally logarithmically stable.
2. **Initial Guess**: Traditional methods usually require a good initial guess, while the method in this paper does not require prior knowledge.
3. **Computational Cost**: Traditional methods have a high computational cost.
### Solution
To enhance stability and reduce computational cost, the authors propose a new algorithm that combines **Carleman estimates** and **Newton's method**, called the **Carleman - Newton method**. The main features of this method include:
1. **No Need for a Good Initial Guess**: The algorithm does not depend on the quality of the initial guess.
2. **Low Computational Cost**: The computational cost is relatively low.
Through this method, the authors are able to directly calculate the solution to the inverse problem and present the results of numerical experiments.