using lsqcurvefit with non continuous function

Hi all,

I'm using lsqcurvefit to fit a model made in COMSOL Multiphysics to some experimental data. Sometimes the model can't converge for specific parameters (haven't found out yet why, it seems kind of random as well). Is there a way to return something to lsqcurvefit (returning all 0's now) without screwing up the fitting process?

[John D'Errico]

1. lsqcurvefit assumes well-defined continuous functions. As such, it may randomly fail to succeed on some data.
2. You have found some instances where lsqcurvefit fails, AND your function is known to be discontinuous.

Why should we be surprised?

Arbitrarily returning all zeros to lsacurvefit some of the time? A great way to create exactly the circumstance where it will fail.