X-coordinate equal spacing requirement in Levenberg-Marquardt fit VI

[mavernon]

Good day everyone,

I've heard that most fit algorithms require relatively equal spaced independent coordinate (X-coord) values in order to do a proper fit.  I will have to convert my current array of equally spaced x-coordinates into an array of values which are no longer equally spaced.

Will this be a problem for the LabView Non-Linear Fit (LM) VI?

Is there any literature on this subject that anyone has come across?  Please let me know if any further info is needed.

Many thanks,

Mark

[Bob_Schor · Accepted solution]

@mavernon wrote:

I've heard that most fit algorithms require relatively equal spaced independent coordinate (X-coord) values in order to do a proper fit.  

Will this be a problem for the LabView Non-Linear Fit (LM) VI?

Is there any literature on this subject that anyone has come across? 

---

I don't know where you heard this, but it is not true.  Common sense (and a little "statistical imagination") should convince you that having most of your points clumped together with one or two outliers will result in the outliers perhaps having "more of an effect" than if the points were more evenly distributed, but you certainly can do fits with outliers.  Uneven spacing will not be a problem for LabVIEW.  For literature, simply read the Help file for Nonlinear Curve Fit VI (which says nothing about even spacing).  For more information, you'll need to dive into some texts on statistics and parameter estimation ...

Bob Schor

[mavernon]

Thank you for the response Bob.  Your explanation makes sense.  I'm fitting some spectroscopy data to a model function.  However, previously I've done this with the X-axis in terms of wavelength which was evenly spaced due to the experimental equipment.  But now, the model is calculated in terms of energy and I'd like to fit the data in terms of energy. 

When converting the measured wavelengths from nm to eV, the values go from being evenly spaced to no longer evenly spaced.  And based on your explanation I'm sure this should not be a problem, as the the LM fit can handle outliers.  

I think as long as I calculate the simulated model with the same X-values as the experimental data, then I should be good.

Thanks again!