# Is is possible to specify use of a non-Euclidean norm for NonlinearModelFit

When using NonlinearModelFit with Method->NMinimize I have received the warning message NonlinearModelFit::lmnl, which includes the text ""The model -cVar+Log[1+0.00821596 t] is linear in the parameters {cVar}, but a nonlinear method or non-Euclidean norm was specified, so nonlinear methods will be used."

The warning is not a problem; the model fit is good. Also, I understand that I could use LinearModelFit for this, but am not for unimportant reasons.

This implies that I could specify a non-Euclidean norm, presumably for use in summing the residuals when determining the objective function. I would love to be able to do this, but can not find any mention of this in the documentation.

My question is: Is it possible to specify a non-Euclidean norm for NonlinearModelFit? How?

• Did you already look at the documentation for NormFunction? May 27 at 15:11
• Switch to FindFit and you can specify a custom NormFunction. NonlinearModelFit does not have that option because some of the properties it returns along with the fit are only defined for the 2-norm, or not implemented for other norms. See also this answer to "What is the difference between FindFit and NonlinearModelFit" and the following comment. May 27 at 15:16