[R] Problem fitting a non-linear regression model with nls

[Bert Gunter]

> My question is how could I estimate those initial values so that the nls
> fitting works.
>
You can't. Your parameters are almost certainly nonidentifiable (which is
what Gabor told you more gracefully).

Just because you believe in a complex (often mechanistic) nonlinear model
and have some data does not assure that the model parameters can be
estimated. If you do not understand why this is so, consider fitting even a
simple 4 parameter logistic when the data do not level off at the top and/or
bottom end. There are then infinitely many solutions in which the parameters
"trade off" with one another to give essentially identical fits. That is
what the "singular gradient" message is trying to tell you.

Bert Gunter
Genentech Nonclinical Statistics