Abstract:Inspired by the gradient flow viewpoint of the Landau equation and corresponding dynamic formulation of the Landau metric in [<a class="link-https" data-arxiv-id="2007.08591" href="https://arxiv.org/abs/2007.08591">arXiv:2007.08591</a>], we develop a novel implicit particle method for the Landau equation in the framework of the JKO scheme. We first reformulate the Landau metric in a computationally friendly form, and then translate it into the Lagrangian viewpoint using the flow map. A key observation is that, while the flow map evolves according to a rather complicated integral equation, the unknown component is merely a score function of the corresponding density plus an additional term in the null space of the collision kernel. This insight guides us in approximating the flow map with a neural network and simplifies the training. Additionally, the objective function is in a double summation form, making it highly suitable for stochastic methods. Consequently, we design a tailored version of stochastic gradient descent that maintains particle interactions and reduces the computational complexity. Compared to other deterministic particle methods, the proposed method enjoys exact entropy dissipation and unconditional stability, therefore making it suitable for large-scale plasma simulations over extended time periods.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop a new implicit particle method for solving the Landau equation. Specifically, the main objectives of the paper include:
1. **Reformulating the Landau metric**: The author first reformulates the Landau metric into a computationally - friendly form for numerical calculations.
2. **Switching to the Lagrangian perspective**: By using the flow map, the Landau metric is transformed from the Eulerian perspective to the Lagrangian perspective.
3. **Simplifying the approximation of the flow map**: The key observation is that although the flow map follows a complex integral equation, the unknown part is mainly the score function corresponding to the density plus an additional term in the null space of the collision kernel. This observation guides the author to use neural networks to approximate the flow map and simplify the training process.
4. **Designing an efficient optimization algorithm**: Since the objective function is in the form of a double sum, it is very suitable for using stochastic methods. Therefore, the author designs a customized stochastic gradient descent (SGD) algorithm to maintain the interaction between particles and reduce the computational complexity.
5. **Achieving unconditional stability and accurate entropy dissipation**: Compared with traditional deterministic particle methods, the proposed method has the characteristics of unconditional stability and accurate entropy dissipation, and is suitable for long - time large - scale plasma simulations.
### Specific problem description
The Landau equation is a fundamental kinetic equation used to model the behavior of charged particles under the Coulomb force, especially suitable for cases where the collision effect is significant in plasma. When the spatial dependence is ignored, the Landau equation can be expressed as:
\[
\frac{\partial f}{\partial t}=\nabla_{v} \cdot\left(\int_{\mathbb{R}^{d}} A(v - v_{*})\left(f(t, v_{*})\nabla_{v} f(t, v)-f(t, v)\nabla_{v_{*}} f(t, v_{*})\right) d v_{*}\right)
\]
where \( f(t, v) \) is the mass distribution function of charged particles at time \( t \) and velocity \( v \), and \( A(z) \) is the collision kernel, defined as:
\[
A(z)=C_{\gamma}|z|^{\gamma + 2}\left(I_{d}-\frac{z\otimes z}{|z|^{2}}\right)
\]
where \( C_{\gamma}>0 \) is the collision intensity, \( I_{d} \) is the identity matrix, and \( \Pi(z) \) represents the projection onto \( \{z\}^{\perp} \). The range of the parameter \( \gamma \) is from \(-d - 1\) to \(1\). In specific cases, \( d = 3 \) and \( \gamma=-3 \) correspond to the Coulomb interaction in plasma.
### Numerical calculation challenges
The main challenges in numerically solving the Landau equation include:
- The complexity of the operator \( Q \)
- High - dimensional velocity space
- Maintaining the physical properties of the solution
### Paper contributions
The method proposed in the paper aims to solve all of the above challenges simultaneously. The main contributions include:
- Based on the gradient - flow perspective of the Landau equation, ensuring that the method structure remains unchanged.
- Using the approximation capabilities of particle methods and neural networks to efficiently handle high - dimensional problems.
- Designing an implicit scheme to ensure that the method is unconditionally stable and has accurate entropy dissipation, suitable for large - scale long - time plasma simulations.
### Method overview
1. **Gradient - flow perspective**: Consider the Landau equation as the gradient flow of entropy \( H \) with respect to the Landau metric \( d_{L} \).
2. **Dynamic JKO scheme**: Based on De Giorgi's minimizing movement scheme, construct a time - discrete approximate solution.
3. **Lagrangian formulation**: Transform the variational problem into a Lagrangian formulation to facilitate the implementation of the particle method.
4. **Particle method**: Use the particle representation method to represent the objective function and the constrained PDE as a double sum of particles.