Regularization

Use the Settings > Regularization option to define the fitting method for the inversion.

Regularization Settings dialog options

Method	Select one of the fitting method options available: Auto-Fit, Fixed Coefficient, or Fit with starting coefficient. These are further described below in the Application Notes section. The Fit with starting coefficient method is available only for DCIP and EM Modelling.
Data fit	The normalized difference between the observed and predicted response at which to terminate the inversion.
Attempts	The maximum number of tries to attempt. When this number is reached, even if the desired fit level has not been achieved, the inversion terminates and returns the results.
Coefficient	The regularization coefficient; it depends on the method chosen. See the Application Notes section for more details.

Application Notes

In the Tikhonov regularization approach, we seek to minimize the total objective function, φ_T=φ_D+λφ_M, where φ_D is the data misfit, λ is the regularization coefficient, and φ_M is the model norm.

By selecting the Auto-Fit method, the regularization coefficient will be chosen such that the total objective function is minimized and φ_D approaches the defined data fit (default 1). You can also adjust the number of attempts by choosing an appropriate λ (default 20).

The Fixed Coefficient method gives you control over the regularization parameter choice. Therefore, it is up to you to vary the regularization parameter manually to provide a meaningful result (i.e., fit the data). Equivalently, this method provides one point on the L-curve. By selecting the Fixed Coefficient method, the regularization coefficient is specified by you (default 2500), and the total objective function will be minimized to that effect.

By selecting the Fit with starting coefficient method, you may provide an initial regularization parameter starting value (default is 2500). Subsequent regularization coefficients will be chosen such that the total objective function is minimized and φ_D approaches the defined data fit (default 1). You can also adjust the number of attempts by choosing an appropriate λ (default 20).
The ability to set the initial regularization parameter makes it possible for you to restart and continue a previous inversion from its termination state. Also, for non-linear inversions such as those for conductivity (DCIP and EM), the starting coefficient can affect the path the inversion takes and the final model the inversion returns.