Elastic3D

class Elastic3D(
    spatial_order=4,
    device="cpu",
    backend="torch",
)

Implementation:

• src/sweep/equations/elastic3d.py

First-order 3D elastic velocity-stress equation.

Parameters

• spatial_order (int, optional): Finite-difference order. It should be an even number.
• device (device or str, optional): Target device used to place derivative operators and backend-specific tensors.
• backend (str, optional): Numerical backend, typically "torch". If you plan to run with PropCUDA, this should still normally be "torch" rather than "cuda".

Models

• models (list[str]): ["vp", "vs", "rho"]

Required models:

• vp: P-wave velocity
• vs: S-wave velocity
• rho: density

Wavefields

• wavefields (list[str]): ["vx", "vy", "vz", "sxx", "syy", "szz", "sxy", "sxz", "syz", "m_vxx", "m_vxy", "m_vxz", "m_vyx", "m_vyy", "m_vyz", "m_vzx", "m_vzy", "m_vzz", "m_sxxx", "m_szzz", "m_sxyx", "m_sxyy", "m_sxzx", "m_sxzz", "m_syyy", "m_syzy", "m_syzz"]

This extends the 2D elastic formulation with y-direction physical and CPML memory variables.

Backend Behavior

• implemented as a first-order 3D equation using PartialDerivative
• supports compiled CUDA binding through _C()

Torch Binding

• supports_torch_binding() : True