Sources and Receivers

This guide summarizes source and receiver tensor conventions for both 2D and 3D APIs.

Source Amplitude

• Expected shape: [n_shots, n_sources, nt].
• Typical dtype: float32 on CPU or CUDA.

Time axis convention: - Time is the last axis in source_amplitude. - In returned receiver data, time is the first axis: [nt, n_shots, n_receivers].

Source Locations

• Expected shape: [n_shots, n_sources, ndim].
• Coordinate order:
  • 2D: [y, x]
  • 3D: [z, y, x]
• dtype should be torch.long/int64.

Receiver Locations

• Expected shape: [n_shots, n_receivers, ndim].
• Coordinate order follows source conventions.

Batching

• Shot axis is the first axis in source_amplitude, source_location, and receiver_location.
• Multiple shots run in one call when these tensors share the same n_shots.
• Output receiver_amplitudes preserves shot order.

Example shape mapping: - source_amplitude: [8, 2, 1500] - source_location: [8, 2, 2] - receiver_location: [8, 32, 2] - receiver output: [1500, 8, 32]

Validation Rules

• Each coordinate must satisfy 0 <= idx < model_size along that dimension.
• Mismatched n_shots across source and receiver tensors will fail at runtime.
• For source-free propagation, pass source_amplitude=None and provide nt explicitly.