Abstract:We present a structure-preserving Eulerian algorithm for solving $L^2$-gradient flows and a structure-preserving Lagrangian algorithm for solving generalized diffusions. Both algorithms employ neural networks as tools for spatial discretization. Unlike most existing methods that construct numerical discretizations based on the strong or weak form of the underlying PDE, the proposed schemes are constructed based on the energy-dissipation law directly. This guarantees the monotonic decay of the system's free energy, which avoids unphysical states of solutions and is crucial for the long-term stability of numerical computations. To address challenges arising from nonlinear neural network discretization, we perform temporal discretizations on these variational systems before spatial discretizations. This approach is computationally memory-efficient when implementing neural network-based algorithms. The proposed neural network-based schemes are mesh-free, allowing us to solve gradient flows in high dimensions. Various numerical experiments are presented to demonstrate the accuracy and energy stability of the proposed numerical schemes.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop a new numerical algorithm to solve the gradient flows problem efficiently and stably. Specifically: 1. **Structure - preserving discretization method**: The paper proposes a method of directly constructing numerical discretization based on the law of energy dissipation, rather than on the strong or weak form of PDE. This method can ensure that the free energy of the system decreases monotonically, thus avoiding unphysical states and ensuring the long - term stability of numerical calculations. 2. **Challenges of high - dimensional problems**: Traditional numerical discretization methods (such as finite difference, finite element and finite volume methods) face the "curse of dimensionality" when dealing with high - dimensional problems, that is, the computational complexity increases exponentially with the problem dimension. Since neural networks perform well in high - dimensional function approximation, the paper uses neural networks as a tool for spatial discretization to solve high - dimensional gradient flow problems. 3. **Order of time and space discretization**: In order to overcome the difficulties brought by neural network discretization, the paper proposes a method of performing time discretization first and then space discretization. This approach has the advantage of high computational memory efficiency when implementing neural - network - based algorithms. 4. **Specific types of gradient flows**: - **L2 - gradient flow**: It satisfies the energy dissipation law $\frac{d}{dt}F[\varphi]=-\int_{\Omega}\eta|\varphi_t|^2\,dx$. - **Generalized diffusion**: It satisfies the energy dissipation law $\frac{d}{dt}F[\rho]=-\int_{\Omega}\eta(\rho)|u|^2\,dx$, where $\rho$ satisfies the mass conservation equation $\rho_t+\nabla\cdot(\rho u) = 0$. 5. **Algorithm framework**: The paper proposes two structure - preserving algorithms based on neural networks: - **Eulerian - format algorithm**: Used to solve the L2 - gradient flow. - **Lagrangian - format algorithm**: Used to solve the generalized diffusion problem. Through these methods, the paper aims to provide a numerical solution scheme that is efficient, robust and applicable to high - dimensional problems, while ensuring energy stability and the preservation of physical constraints.

Energetic Variational Neural Network Discretizations of Gradient Flows

High-order energy stable discrete variational derivative schemes for gradient flows

Numerical Analysis on Neural Network Projected Schemes for Approximating One Dimensional Wasserstein Gradient Flows

A new flow dynamic approach for Wasserstein gradient flows

Gauss Newton Method for Solving Variational Problems of PDEs with Neural Network Discretizaitons

Large-Scale Wasserstein Gradient Flows

Lagrangian Flow Networks for Conservation Laws

Eulerian algorithms for computing some Lagrangian flow network quantities

Efficient and stable SAV-based methods for gradient flows arising from deep learning

Evolving Neural Network (ENN) Method for One-Dimensional Scalar Hyperbolic Conservation Laws: I Linear and Quadratic Fluxes

Variational Extrapolation of Implicit Schemes for General Gradient Flows

Discrete approximation of the minimizing movement scheme for evolution equations of Wasserstein gradient flow type with nonlinear mobility

Energy stable neural network for gradient flow equations

Gradient Flow Based Phase-Field Modeling Using Separable Neural Networks

High order spatial discretization for variational time implicit schemes: Wasserstein gradient flows and reaction-diffusion systems

Energy-dissipative spectral renormalization exponential integrator method for gradient flow problems

Neural Sinkhorn Gradient Flow

Two novel numerical methods for gradient flows: generalizations of the Invariant Energy Quadratization method

Conformal variational discretisation of infinite dimensional Hamiltonian systems with gradient flow dissipation

Neural Vortex Method: from Finite Lagrangian Particles to Infinite Dimensional Eulerian Dynamics