Provides tools for working with nonlinear least squares problems. It is intended to eventually supersede the nls() function in the R distribution. For example, nls() specifically does NOT deal with small or zero residual problems. Its Gauss-Newton method frequently stops with 'singular gradient' messages.