Skip to content

Inverse Diffusivity

anypinn create my-project --template inverse-diffusivity

Recovers a space-dependent diffusivity D(x) represented as a neural network Field, rather than a single scalar parameter.

Problem

\[ \frac{\partial u}{\partial t} = \nabla \cdot \bigl(D(x)\,\nabla u\bigr) \]

where D(x) is unknown and learned as a neural network.

Features Demonstrated

  • Function-valued parameter recovery (D(x) as a Field)
  • Composable differential operators from anypinn.lib.diff
  • PDE inverse problem with spatially varying coefficients

Results

Inverse Diffusivity results