Abstract:The Burgers equation is a well-established test case in the computational modeling of several phenomena such as fluid dynamics, gas dynamics, shock theory, cosmology, and others. In this work, we present the application of Physics-Informed Neural Networks (PINNs) with an implicit Euler transfer learning approach to solve the Burgers equation. The proposed approach consists in seeking a time-discrete solution by a sequence of Artificial Neural Networks (ANNs). At each time step, the previous ANN transfers its knowledge to the next network model, which learns the current time solution by minimizing a loss function based on the implicit Euler approximation of the Burgers equation. The approach is tested for two benchmark problems: the first with an exact solution and the other with an alternative analytical solution. In comparison to the usual PINN models, the proposed approach has the advantage of requiring smaller neural network architectures with similar accurate results and potentially decreasing computational costs.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use Physics - Informed Neural Networks (PINNs) combined with the implicit Euler transfer learning method to solve the viscous Burgers equation with homogeneous Dirichlet boundary conditions. Specifically, the paper proposes a new method to iteratively learn at each time step through a series of Multilayer Perceptrons (MLPs) to obtain an approximate solution of the Burgers equation. The main advantage of this method is that it can use a smaller neural network architecture while maintaining a similar accuracy to the traditional PINN method and has the potential to reduce the computational cost. ### Background of the paper The Burgers equation is a classical partial differential equation, which is widely used in many fields such as fluid dynamics, gas dynamics, shock wave theory, and cosmology. The form of this equation is as follows: \[ u_t + u u_x = \nu u_{xx}, \quad (t, x) \in (0, t_f] \times (0, 1) \] The initial condition and the boundary condition are respectively: \[ u(0, x) = u_0(x), \quad x \in [0, 1] \] \[ u(t, 0) = u(t, 1) = 0, \quad t \in [0, t_f] \] ### Overview of the method The method proposed in the paper mainly includes the following steps: 1. **Initial condition learning**: First, train an MLP \( N^{(0)} \) to fit the initial condition \( u_0(x) \), and the loss function is: \[ L_0 := \frac{1}{n_s} \sum_{s = 1}^{n_s} \left\| \tilde{u}^{(0)}(x_s) - u_0(x_s) \right\|^2 \] 2. **Implicit Euler transfer learning**: At each time step \( t^{(k)} \), the MLP \( N^{(k - 1)} \) of the previous time step transfers its knowledge to the next MLP \( N^{(k)} \), and the latter learns the solution of the current time step by minimizing the following loss function: \[ L := \frac{1}{n_s - 2} \sum_{s = 1}^{n_s - 2} \left\| R(x; \tilde{u}^{(k)}, \tilde{u}^{(k - 1)}) \right\|^2 + \frac{1}{2} \left( \left\| \tilde{u}^{(k)}(0) \right\|^2 + \left\| \tilde{u}^{(k)}(1) \right\|^2 \right) \] where the residual \( R \) is defined as: \[ R(x; \tilde{u}^{(k)}, \tilde{u}^{(k - 1)}) := \tilde{u}^{(k)} - \tilde{u}^{(k - 1)} - h_t \left( \nu \tilde{u}^{(k)}_{xx} - \tilde{u}^{(k)} \tilde{u}^{(k)}_x \right) \] 3. **Iterative process**: Start from \( k = 1 \) and repeat the above steps until the final time step \( t_f \) is reached. ### Experimental results The paper verifies the effectiveness of the proposed method through two benchmark problems: 1. **Problem 1**: In the case of having an exact solution, the initial condition is: \[ u_0(x) = \frac{2 \nu \pi \sin(\pi x)}{2 + \cos(\pi x)} \] The exact solution is: \[ u(t, x) = \frac{2 \nu \pi e^{-\nu \pi^2 t} \sin(\pi x)}{2 + e^{-\nu \pi^2

Burgers' pinns with implicit euler transfer learning

A Combination of Deep Neural Networks and Physics to Solve the Inverse Problem of Burger's Equation

Approximate solution of KdV-Burgers equation using improved PINNs algorithm

Equation identification for fluid flows via physics-informed neural networks

DiffGrad for Physics-Informed Neural Networks

Solving Viscous Burgers' Equation: Hybrid Approach Combining Boundary Layer Theory and Physics-Informed Neural Networks

Distributed physics informed neural network for data-efficient solution to partial differential equations

A Comparative Study of the Explicit Finite Difference Method and Physics-Informed Neural Networks for Solving the Burgers' Equation

Binary structured physics-informed neural networks for solving equations with rapidly changing solutions

Multi-soliton solutions and data-driven discovery of higher-order Burgers' hierarchy equations with physics informed neural networks

iPINNs: incremental learning for Physics-informed neural networks

Recipes for when physics fails: recovering robust learning of physics informed neural networks

Physics-Informed Neural Networks with Adaptive Localized Artificial Viscosity

Efficient Error Certification for Physics-Informed Neural Networks

Investigating the Ability of PINNs To Solve Burgers' PDE Near Finite-Time BlowUp

Physics-informed neural networks for solving forward and inverse flow problems via the Boltzmann-BGK formulation

Operator Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations Characterized by Sharp Solutions

Solving stiff ordinary differential equations using physics informed neural networks (PINNs): simple recipes to improve training of vanilla-PINNs

Error estimates and physics informed augmentation of neural networks for thermally coupled incompressible Navier Stokes equations

Combining physics-informed graph neural network and finite difference for solving forward and inverse spatiotemporal PDEs