3.146 FAQ-884 How to get a good fit with a very large or a very small parameter?

Last Update: 9/22/2017

When you do curve fitting, you might meet these problems:

  • Origin returns missing value in standard errors when you perform fitting with a large parameter.
  • Fit can not converge when you perform fitting with a very small parameter.

In both cases, we need to redefine the equations to avoid a very large or a very small parameter. For example:

  • y = A \cdot x; if A is a large parameter, we can redefine the equation as y=(A' \cdot 1E3) \cdot x. After fitting, we can get A = A'\cdot 1E3;
  • y = A \cdot x; if A is a small parameter, we can redefine the equation as y=(A' \cdot 1E-3) \cdot x. After fitting, we can get A = A' \cdot 1E-3;


Another actual example is the nonlinear implicit diode function:

f = Is \cdot {e^{(\frac{{V - I \cdot Rs}}{{k \cdot T}} - 1)}} + \frac{{V - I \cdot Rs}}{{Rsh}} - I\,\!, k is in eV{K^{ - 1}}\,\! as a unit.

To do fitting with this function, firstly, we can reset the parameter Is to I's, where I's=Is*exp(-20), then the parameter I's won't be too small.

f = I's \cdot [{e^{(\frac{{V - I \cdot Rs}}{{k \cdot T}} - 20)}} - {e^{ (- 20)}}] + \frac{{V - I \cdot Rs}}{{Rsh}} - I\,\!

In this way, we can avoid the very small parameter and finally get the fit converge.


Keywords:good fit, not converge, missing value in standard error,diode function