Abstract:Partial Differential Equations are foundational in modeling science and natural systems such as fluid dynamics and weather forecasting. The Latent Evolution of PDEs method is designed to address the computational intensity of classical and deep learning-based PDE solvers by proposing a scalable and efficient alternative. To enhance the efficiency and accuracy of LE-PDE, we incorporate the Mamba model, an advanced machine learning model known for its predictive efficiency and robustness in handling complex dynamic systems with a progressive learning strategy. The LE-PDE was tested on several benchmark problems. The method demonstrated a marked reduction in computational time compared to traditional solvers and standalone deep learning models while maintaining high accuracy in predicting system behavior over time. Our method doubles the inference speed compared to the LE-PDE while retaining the same level of parameter efficiency, making it well-suited for scenarios requiring long-term predictions.

What problem does this paper attempt to address?

This paper attempts to solve the following problems: 1. **Computational complexity and efficiency issues of traditional PDE solvers**: - Traditional partial differential equation (PDE) solvers have high computational complexity when dealing with large - scale and complex dynamic systems, and it is difficult to provide solutions quickly. Especially when dealing with time steps involving millions of units, traditional methods are particularly inefficient. - Inverse optimization tasks (such as inferring system parameters) also face similar challenges, further increasing the difficulty of modeling forward evolution. 2. **Limitations of existing deep - learning models**: - Existing deep - learning - based surrogate models can accelerate PDE solving, but they rely on an end - to - end mapping structure and use convolutional neural networks (CNN) as the base model, resulting in significant convolutional computation time and thus increasing the time complexity. - The training mechanisms of these models are not fully defined yet, lacking effective learning methods, making it difficult for the models to learn how to learn efficiently. 3. **Continuity conflicts in multi - step prediction**: - In the training objective function, the multi - step bundling strategy and the continuity of the loss may conflict in highly nonlinear problems. The former ensures the multi - step prediction ability of the model, while the latter requires that the continuity in the latent space should not be too large. To solve these problems, the paper introduces the **LE - PDE++ framework**, which combines the Mamba model and improves the existing LE - PDE method in the following ways: - **Using the Mamba model to accelerate inference in the latent space**: The Mamba model can significantly improve the inference speed while maintaining parameter efficiency. - **Introducing a progressive learning mechanism**: Through the progressive sampling strategy, the model can gradually adapt and improve, so as to better handle long - term prediction tasks. - **Optimizing the training process**: By adjusting the training objective function, the continuity conflict problem in multi - step prediction is solved, and the accuracy and stability of the model are improved. These improvements make LE - PDE++ show higher inference speed and better prediction accuracy in multiple benchmark tests, especially when dealing with complex dynamic systems.

LE-PDE++: Mamba for accelerating PDEs Simulations

Learning to Accelerate Partial Differential Equations via Latent Global Evolution

Meta-PDE: Learning to Solve PDEs Quickly Without a Mesh

Solving Partial Differential Equations Using Deep Learning and Physical Constraints

Latent Neural PDE Solver: a reduced-order modelling framework for partial differential equations

Deep-learning based discovery of partial differential equations in integral form from sparse and noisy data

Learning a Neural Solver for Parametric PDE to Enhance Physics-Informed Methods

Solving High-Dimensional PDEs with Latent Spectral Models

Physics Informed Extreme Learning Machine (PIELM) -- A rapid method for the numerical solution of partial differential equations

DL-PDE: Deep-learning based data-driven discovery of partial differential equations from discrete and noisy data

Local Extreme Learning Machines and Domain Decomposition for Solving Linear and Nonlinear Partial Differential Equations

Multi-resolution partial differential equations preserved learning framework for spatiotemporal dynamics

PDE-LEARN: Using deep learning to discover partial differential equations from noisy, limited data

Uncertainty Quantification for Forward and Inverse Problems of PDEs via Latent Global Evolution

Monte Carlo Neural PDE Solver for Learning PDEs via Probabilistic Representation

End-to-End Mesh Optimization of a Hybrid Deep Learning Black-Box PDE Solver

Meta-Auto-Decoder for Solving Parametric Partial Differential Equations

M$^{2}$M: Learning controllable Multi of experts and multi-scale operators are the Partial Differential Equations need

State-space models are accurate and efficient neural operators for dynamical systems

A machine learning framework for data driven acceleration of computations of differential equations