This repository was created to showcase an independent research report on the concept of Physics Informed Neural Networks (PINNs). Here I explain briefly the concepts of Backward Propagation, Forward Propagation, Loss Function formulation, optimization etc.
I then demonstrate a PiNNs approach to approximating the solutions to the 1D heat equation using PyTorch and compare the behavior of solutions based on the number of training steps.
We aim to solve the 1D heat equation using Physics-Informed Neural Networks. The mathematical model we are considering is:
The 1D heat equation is given by:
u_t = u_{xx}
Where:
u_t
denotes the partial derivative ofu
with respect tot
.u_{xx}
denotes the second partial derivative ofu
with respect tox
.
Initial and Boundary Conditions
The initial condition is:
u(x, 0) = sin(πx)
And the boundary conditions are:
- At
x = 0
:u(0, t) = 0
- At
x = 1
:u(1, t) = 0
Consider the Navier-Stokes equations, which describe the motion of a fluid:
Navier-Stokes Equations:
- Momentum equation:
ρ(∂u/∂t + u · ∇u) = -∇p + μ∇²u + f
- Continuity equation:
∇ · u = 0
Where:
u
is the velocity field.p
is the pressure.ρ
is the density.μ
is the dynamic viscosity.f
is the body force.
We implement the Physics-Informed Neural Network (PINN) solution to the 2D Navier-Stokes problem. You can find the code and plot the pressure field here.
To read further on my independent research on PINNs, please refer to this report.