Abstract:Computational methods for thermal radiative transfer problems exhibit high computational costs and a prohibitive memory footprint when the spatial and directional domains are finely resolved. A strategy to reduce such computational costs is dynamical low-rank approximation (DLRA), which represents and evolves the solution on a low-rank manifold, thereby significantly decreasing computational and memory requirements. Efficient discretizations for the DLRA evolution equations need to be carefully constructed to guarantee stability while enabling mass conservation. In this work, we focus on the Su-Olson closure leading to a linearized internal energy model and derive a stable discretization through an implicit coupling of internal energy and particle density. Moreover, we propose a rank-adaptive strategy to preserve local mass conservation. Numerical results are presented which showcase the accuracy and efficiency of the proposed low-rank method compared to the solution of the full system.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is that in the problem of thermal radiation transfer, when the space and direction domains are finely divided, the computational method exhibits high computational cost and huge memory occupation. To solve this problem, the paper proposes a Dynamical Low - Rank Approximation (DLRA) strategy, which significantly reduces the computational and memory requirements by representing and evolving the solution on a low - rank manifold. Specifically, the paper focuses on the linearized internal energy model under the Su - Olson closure condition and proposes a method of implicitly coupling internal energy and particle density to ensure energy stability and mass conservation. In addition, a rank - adaptive strategy is also proposed to maintain local mass conservation. The paper shows the accuracy and efficiency of the proposed low - rank method compared with the full - system solution.
### Key problems and solutions
1. **High computational cost and memory occupation**:
- **Solution**: Adopt Dynamical Low - Rank Approximation (DLRA) to reduce computational and memory requirements by representing and evolving the solution on a low - rank manifold.
2. **Stability problems**:
- **Solution**: Propose a method of implicitly coupling internal energy and particle density to ensure energy stability. Through the implicit time - discretization method, the problem of total energy increase that may be caused by the explicit IMEX scheme is avoided.
3. **Mass conservation**:
- **Solution**: Introduce a rank - adaptive strategy, and ensure local mass conservation through the basis function augmentation step. This enables the numerical method to adjust the rank without violating mass conservation.
### Main contributions
- **Energy - stable numerical scheme**: Prove that the simple IMEX scheme cannot ensure energy stability, and propose a new implicit coupling scheme to ensure energy stability.
- **Mass - conservation and rank - adaptive integrator**: Through the basis function augmentation step and the conservative truncation strategy, local mass conservation and rank - adaptivity are achieved.
- **Numerical experiments**: Show that the proposed low - rank method significantly reduces computational cost and memory requirements while maintaining stability and conservation properties.
### Mathematical model
The thermal radiation transfer equations studied in the paper can be expressed as:
\[
\frac{1}{c} \frac{\partial f(t, x, \mu)}{\partial t}+\mu \frac{\partial f(t, x, \mu)}{\partial x}=\sigma(B(t, x)-f(t, x, \mu)),
\]
\[
\frac{\partial B(t, x)}{\partial t}=\sigma(\langle f(t, x, \cdot)\rangle_\mu - B(t, x)),
\]
where \( f(t, x, \mu) \) is the particle density, \( B(t, x) \) is the internal energy, \( \sigma \) is the medium absorption rate, \( c \) is the speed of light, and \( \langle\cdot\rangle_\mu \) represents the integral with respect to the direction variable \(\mu\).
### Dynamical low - rank approximation
The core idea of dynamical low - rank approximation is to approximate the solution \( f(t, x, \mu) \) as:
\[
f(t, x, \mu)\approx\sum_{i,j = 1}^r X_i(t, x)S_{ij}(t)V_j(t, \mu),
\]
where \( \{X_i\} \) and \( \{V_j\} \) are orthogonal basis functions in space and direction respectively, \( S_{ij} \) is the coefficient matrix, and \( r \) is the rank.
### Time - discretization
To construct an energy - stable time - discretization scheme, the paper adopts the following steps:
1. **K - step**: Update the spatial basis functions \( X_i \).
2. **L - step**: Update the directional basis functions \( V_j \).
3. **S - step**: Update the coefficient matrix \( S_{ij} \).
4. **Truncation**: Truncate the augmented low - rank solution to a new rank.