2D Elastic FWI on Marmousi with Torch

Source file:

• examples/FWI/2d/elastic/torch/fwi_marmousi.py

What This Example Does

This example runs 2D elastic full-waveform inversion on the Marmousi model with one script that supports two propagator backends:

• eager: pure PyTorch propagation through PropTorch(..., backend="eager")
• cuda: compiled CUDA propagation through PropTorch(..., backend="cuda")

The script:

1. loads the Marmousi true and smooth vp models
2. derives vs from vp / 1.732 and uses constant rho
3. generates observed elastic shot gathers from the true model
4. inverts vp and vs from the smooth starting model

Main Components

The solver is built from:

• equation: Elastic(...)
• propagator: PropTorch(...)
• wave: a Ricker wavelet injected into sxx and szz
• sources: regularly sampled surface sources
• receivers: regularly sampled surface receivers
• models: vp, vs, and rho

Prepare the Marmousi Model Files

This example reads the prepared Marmousi vp files and derives vs and rho inside the script:

• examples/models/marmousi/true.npy
• examples/models/marmousi/smooth.npy

Generate the files before running the example:

python3 examples/models/marmousi/download_marmousi.py --extract
python3 examples/models/marmousi/extract_model_segy.py
python3 examples/models/marmousi/convert_segy_to_npy.py
python3 examples/models/marmousi/prepare_fwi_models.py \
  --input examples/models/marmousi/npy/vp_1p25m.npy \
  --source-dh 1.25 \
  --target-dh 25.0 \
  --radii 8,8 \
  --passes 3

Backend Selection

Run the example with:

PyTorchCUDA

python3 examples/FWI/2d/elastic/torch/fwi_marmousi.py --backend eager

python3 examples/FWI/2d/elastic/torch/fwi_marmousi.py --backend cuda

Key Configuration

Shared configuration comes from examples/_shared/configure_marmousi.py and includes:

• nt=2500, dt=0.002
• dh=25.0
• spatial_order=8
• abcn=20
• src_step=4, rec_step=2
• source_encoding=True
• batchsize=8

The script uses:

• source_type=["sxx", "szz"]
• receiver_type=["vx", "vz"]
• separate learning rates for vp, vs, and rho

Geometry

The final array shapes are:

• sources: (nshots, 2)
• receivers: (nshots, nreceivers, 2)

Outputs

The script saves:

• ricker.png
• observed_data.png
• loss.png
• epoch_XXXX.png

Each backend writes into its own output directory:

PyTorchCUDA

elastic_fwi_marmousi_torch

elastic_fwi_marmousi_cuda

Example Figures

The following figures come from a completed CUDA run of the 2D elastic Marmousi example.

observed_data.png: observed vx and vz shot gathers generated from the true model.

loss.png: the inversion loss curve across optimization steps.

epoch_0100.png: the saved progress panel at the final shown epoch, including the true models, current inverted models, and current gradients.

Notes

• the Marmousi elastic example defaults to source encoding
• cuda mode requires a CUDA-capable PyTorch environment and compiled bindings