Skip to content

Project Overview

TIDE is a PyTorch-first electromagnetic modeling and inversion library built around finite-difference time-domain (FDTD) Maxwell solvers.

What You Can Do With TIDE

  • 2D TM Forward Modeling

    Run 2D TM-mode forward simulations with tide.maxwelltm, including CPML absorbing boundaries.

  • 3D Forward Modeling

    Run full 3D forward simulations with tide.maxwell3d using high-order finite-difference stencils.

  • Gradient Computation

    Autodiff with respect to epsilon and sigma, integrating directly with PyTorch optimizers.

  • Inversion Loops

    Build complete inversion workflows in raw PyTorch or with MaxwellTM / Maxwell3D modules.

Core Concepts

Concept Description
Model tensors epsilon, sigma, mu — shaped to match the grid
Source amplitude Shape [n_shots, n_sources, nt]
Receiver traces Returned as [nt, n_shots, n_receivers]
Boundary conditions CPML absorbing boundaries, finite-difference stencils, CFL-driven internal resampling

Coordinate Conventions

Axis Ordering

  • 2D TM: uses [y, x] layout
  • 3D: uses [z, y, x] layout

Typical Workflow

graph LR
    A[Build model tensors] --> B[Define source & receiver geometry]
    B --> C[Forward simulation → synthetic traces]
    C --> D[Compute misfit against observed data]
    D --> E[Backpropagate]
    E --> F[Update model]
    F -->|iterate| C

Learning Path

  1. Run a small 2D forward example from getting-started.md
  2. Read guides/api-orientation.md
  3. Read guides/modeling.md
  4. Read guides/inversion.md
  5. Before scaling up, review guides/configuration.md, guides/limitations.md, and guides/verification.md