Abstract:Singularly perturbed boundary value problems pose a significant challenge for their numerical approximations because of the presence of sharp boundary layers. These sharp boundary layers are responsible for the stiffness of solutions, which leads to large computational errors, if not properly handled. It is well-known that the classical numerical methods as well as the Physics-Informed Neural Networks (PINNs) require some special treatments near the boundary, e.g., using extensive mesh refinements or finer collocation points, in order to obtain an accurate approximate solution especially inside of the stiff boundary layer. In this article, we modify the PINNs and construct our new semi-analytic SL-PINNs suitable for singularly perturbed boundary value problems. Performing the boundary layer analysis, we first find the corrector functions describing the singular behavior of the stiff solutions inside boundary layers. Then we obtain the SL-PINN approximations of the singularly perturbed problems by embedding the explicit correctors in the structure of PINNs or by training the correctors together with the PINN approximations. Our numerical experiments confirm that our new SL-PINN methods produce stable and accurate approximations for stiff solutions.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is **the challenges in the numerical approximation of Singularly Perturbed Boundary Value Problems (SPBVPs)**, especially the stiffness problem of numerical solutions due to the existence of boundary layers. Specifically, this type of problem has very steep transition regions (i.e., boundary layers) near the boundaries, which makes traditional numerical methods (such as the finite element method or the finite volume method) require a large amount of mesh refinement when dealing with these problems, resulting in high computational costs and low efficiency.
To overcome these challenges, the author proposes a new Semi - Analytic Physics - Informed Neural Networks method (SL - PINNs). This method improves the stability and accuracy of the numerical approximation of stiff solutions by embedding corrector functions that describe the singular behavior of the boundary layer in the neural network structure or by training these corrector functions and the PINN approximate solution simultaneously during the training process.
### Main contributions of the paper:
1. **Boundary layer analysis**: Through detailed boundary layer analysis of singularly perturbed problems, the author derives corrector functions that describe the singular behavior of the boundary layer.
2. **Construction of the SL - PINN method**: The author proposes two methods for constructing SL - PINN: one is to directly embed the corrector functions in the neural network structure, and the other is to train the corrector functions and the PINN approximate solution simultaneously during the training process.
3. **Verification by numerical experiments**: Through a series of numerical experiments, the author verifies the effectiveness and accuracy of the SL - PINN method in dealing with stiff solutions, especially in non - characteristic boundary and characteristic boundary cases.
### Mathematical formulation of specific problems:
The paper considers the following two - dimensional singularly perturbed elliptic boundary value problem in the form:
\[
\begin{cases}
-\varepsilon \Delta u - \mathbf{b} \cdot \nabla u = f, & \text{in } \Omega, \\
u = 0, & \text{on } \partial \Omega,
\end{cases}
\]
where \(\Omega=(0,1)^2\), \(0 < \varepsilon \ll 1\) is a small perturbation parameter, \(\mathbf{b}=(b_1(x,y), b_2(x,y))\) and \(f = f(x,y)\) are assumed to be smooth enough to meet the requirements of analysis and computation.
### Main technical details:
- **Non - characteristic boundary case**: When \(b_1>0\) and \(b_2>0\), the problem has Ordinary Boundary Layers (OBLs) at the boundaries \(x = 0\) and \(y = 0\).
- **Characteristic boundary case**: When \(b_1>0\) and \(b_2 = 0\), the problem has Parabolic Boundary Layers (PBLs) at the boundaries \(y = 0\) and \(y = 1\), and has an Ordinary Boundary Layer at the left outflow boundary \(x = 0\).
By introducing corrector functions and embedding them into the PINN structure, the SL - PINN method can effectively capture the singular behavior of the boundary layer, thereby numerically approximating the stiff solution more accurately.