18

Physics Informed Neural Networks

Physics Informed Neural Network, particularly to investigate the effectiveness of neural networks in solving the Burgers' equation, a fundamental model in fluid dynamics.

1.Theory and Background:

Bateman–Burgers equation is a fundamental partial differential equation and convection–diffusion equation occurring in various areas of applied mathematics, such as fluid mechanics, nonlinear acoustics, gas dynamics, and traffic flow.

dudt+ududx=d2udx2\frac{du}{dt} + u\frac{du}{dx} = \frac{d^2u}{dx^2}

ut+uux(0.01/π)uxx=0u_t +uu_x - (0.01/\pi)u_{xx}= 0

u(0,x)=sin(πx)u(0,x)=-sin(\pi x) u(t,1)=u(t,1)=0u(t,-1)=u(t,1)=0

2. Traditional Methods vs. Neural Networks

Traditional MethodsNeural Networks
Finite Difference: Approximates derivatives using small finite differences.Flexible Approximators: Neural networks can approximate complex nonlinear functions, making them suitable for solving a wide range of partial differential equations.
Finite Element: Divides the domain into smaller elements, approximating the solution within each element and assembling them into a global system of equations.Computational Efficiency: Neural networks can offer computational efficiency by bypassing the need for explicit mesh generation and discretization, making them attractive for real-time or large-scale simulations.
Spectral Methods: Represent the solution as a sum of basis functions, such as trigonometric or polynomial functions, providing high accuracy but limited to specific geometries.

3. Architecture:

An image of the neural net architecture

4. Results and Discussion:

List of Loss on Epochs

Graph of the function

5. Applications and Future Work

Neural Networks can be used as an alternative to numerical methods

  • Hyperparameter tuning for compute Efficiency
  • Robustness in architecture by varying training data
  • Solving Discrete Time Models
  • Full Vortex Simulation with hyperparameters