Abstract:In kinetic equations, external fields play a significant role, particularly when their strength is sufficient to balance collision effects, leading to the so-called high-field regime. Two typical examples are the Vlasov-Poisson-Fokker-Planck (VPFP) system in plasma physics and the Boltzmann equation in semiconductor physics. In this paper, we propose a generic asymptotic-preserving multiple-input DeepONet (AP-MIONet) method for solving these two kinetic equations with variable parameters in the high-field regime. Our method aims to tackle two major challenges in this regime: the additional variable parameters introduced by electric fields, and the absence of an explicit local equilibrium, which is a key component of asymptotic-preserving (AP) schemes. We leverage the multiple-input DeepONet (MIONet) architecture to accommodate additional parameters, and formulate the AP loss function by incorporating both the mass conservation law and the original kinetic system. This strategy can avoid reliance on the explicit local equilibrium, preserve the mass and adapt to non-equilibrium states. We demonstrate the effectiveness and efficiency of the proposed method through extensive numerical examples.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are the two major challenges encountered when solving collision kinetic equations (such as the Vlasov - Poisson - Fokker - Planck (VPFP) system and the semiconductor Boltzmann equation) in the high - field limit: 1. **Multiple input parameters**: In the high - field limit, the existence of an external electric field introduces additional variable parameters. These parameters include not only initial conditions and boundary conditions but also external electric field parameters controlled by coupled equations. Traditional neural operator methods are usually only able to handle operator learning within a single Banach space and have difficulty dealing with multiple input functions. 2. **Lack of explicit local equilibrium states**: In the high - field limit, kinetic equations often do not have explicit local equilibrium states, which is a key component in asymptotic - preserving (AP) schemes. This makes methods designed based on modern AP losses difficult to implement because these designs usually rely on micro - macro decompositions or parity decompositions. To solve these problems, the author proposes a new framework - Asymptotic - Preserving Multi - Input DeepONet (AP - MIONet). Specifically, this method addresses the above challenges in the following ways: - **Utilizing the MIONet architecture**: MIONet can accept multiple input functions, thus being able to handle the multi - input problems brought about by the external electric field and initial - boundary conditions. MIONet maps these inputs to the corresponding solutions, thereby expanding the capabilities of the traditional DeepONet. - **Introducing the AP loss function**: To deal with the problem of the lack of explicit local equilibrium states in the high - field limit, the author incorporates both the law of mass conservation and the original kinetic equation into the loss function. This strategy avoids reliance on explicit local equilibrium states and ensures adaptability to mass and non - equilibrium states. Through these improvements, AP - MIONet can effectively solve the VPFP system and the semiconductor Boltzmann equation in the high - field limit and has demonstrated its high efficiency and accuracy in numerical experiments. ### Summary of Mathematical Formulas - **Kinetic Equation**: \[ \frac{\partial f}{\partial t}+v\cdot\nabla_x f-\frac{1}{\varepsilon}\nabla_x\phi\cdot\nabla_v f = \frac{1}{\partial\varepsilon}Q(f) \] \[ -\Delta_x\phi(t, x)=\rho(t, x)-h(x) \] - **Law of Mass Conservation**: \[ \partial_t\rho+\nabla_x\cdot j = 0 \] where \(j = \int_{\mathbb{R}^N}v f(t, x, v)dv\) - **VPFP System in the High - Field Limit**: \[ \begin{cases} \partial_t\rho-\nabla_x\cdot(\rho\nabla_x\phi)=0\\ -\Delta_x\phi=\rho - h(x) \end{cases} \] - **Semiconductor Boltzmann Equation in the High - Field Limit**: \[ \partial_t\rho+\nabla_x\cdot(\rho(t, x)\sigma(E(t, x)))=0 \] where \(\sigma(E)=\int_{\mathbb{R}^N}v F_E(v)dv\) Through these formulas and the improved neural network architecture, AP - MIONet can effectively solve complex kinetic equations in the high - field limit.

AP-MIONet: Asymptotic-preserving multiple-input neural operators for capturing the high-field limits of collisional kinetic equations

Asymptotic-preserving neural networks for multiscale Vlasov-Poisson-Fokker-Planck system in the high-field regime

Asymptotic-Preserving Neural Networks for Multiscale Vlasov–Poisson–Fokker–Planck System in the High-Field Regime

Asymptotic-preserving neural networks for the semiconductor Boltzmann equation and its application on inverse problems

State-space Modeling for Electrochemical Performance of Li-ion Batteries with Physics-Informed Deep Operator Networks

DeepM&Mnet: Inferring the electroconvection multiphysics fields based on operator approximation by neural networks

A conservative hybrid physics-informed neural network method for Maxwell-Ampère-Nernst-Planck equations

Asymptotic-Preserving Neural Networks for Multiscale Time-Dependent Linear Transport Equations

Structure-Preserving Operator Learning: Modeling the Collision Operator of Kinetic Equations

Macroscopic auxiliary asymptotic preserving neural networks for the linear radiative transfer equations

Complex Valued Deep Operator Network (DeepONet) $[\mathcal{G}]$ for Three Dimensional Maxwell's Equations: $\mathcal{G} \in \mathbb{C}^{m \times n}$

Data Generation-based Operator Learning for Solving Partial Differential Equations on Unbounded Domains

A conservative hybrid deep learning method for Maxwell–Ampère–Nernst–Planck equations

Enriched Physics-informed Neural Networks for Dynamic Poisson-Nernst-Planck Systems

Fourier-MIONet: Fourier-enhanced multiple-input neural operators for multiphase modeling of geological carbon sequestration

B-LSTM-MIONet: Bayesian LSTM-based Neural Operators for Learning the Response of Complex Dynamical Systems to Length-Variant Multiple Input Functions

Interfacing finite elements with deep neural operators for fast multiscale modeling of mechanics problems

Physics-informed neural networks with parameter asymptotic strategy for learning singularly perturbed convection-dominated problem

Solving forward and inverse PDE problems on unknown manifolds via physics-informed neural operators

MIONet: Learning multiple-input operators via tensor product

SPI-MIONet for surrogate modeling in phase-field hydraulic fracturing