simpeg.electromagnetics.time_domain.Simulation3DMagneticFluxDensity.getAdiagDeriv#

Simulation3DMagneticFluxDensity.getAdiagDeriv(tInd, u, v, adjoint=False)[source]#

Derivative operation for the diagonal system matrix times a vector.

The diagonal system matrix for time-step index k is given by:

\[\mathbf{A}_k = \mathbf{C M_{e\sigma}^{-1} C^T M_{f\frac{1}{\mu}}} + \frac{1}{\Delta t_k} \mathbf{I}\]

where

  • \(\Delta t_k\) is the step length

  • \(\mathbf{I}\) is the identity matrix

  • \(\mathbf{C}\) is the discrete curl operator

  • \(\mathbf{M_{e \sigma}}\) is the conductivity inner-product matrix on edges

  • \(\mathbf{M_{f\frac{1}{\mu}}}\) is the inverse permeability inner-product matrix on faces

See the Notes section of the doc strings for Simulation3DMagneticFluxDensity for a full description of the formulation.

Where \(\mathbf{m}\) are the set of model parameters, \(\mathbf{v}\) is a vector and \(\mathbf{b_k}\) is the discrete solution for time-step k, this method assumes the discrete solution is fixed and returns

\[\frac{\partial (\mathbf{A_k \, b_k})}{\partial \mathbf{m}} \, \mathbf{v}\]

Or the adjoint operation

\[\frac{\partial (\mathbf{A_k \, b_k})}{\partial \mathbf{m}}^T \, \mathbf{v}\]
Parameters:
tIndint

The time-step index; between [0, n_steps-1].

u(n_faces,) numpy.ndarray

The solution for the fields for the current model; i.e. \(\mathbf{b_k}\).

vnumpy.ndarray

The vector. (n_param,) for the standard operation. (n_faces,) for the adjoint operation.

adjointbool

Whether to perform the adjoint operation.

Returns:
numpy.ndarray

Derivative of system matrix times a vector. (n_faces,) for the standard operation. (n_param,) for the adjoint operation.