Cubic and linear baseline

Konstantinos Ch...
Posts: 59
Joined: 2014-11-14
Location: United Kingdom

Hi all,

Just a question.....I have a single peak with some background and i tried 2 sets of fitting. First case, i used ''linear'' baseline and i did the fitting and the result was ok. I then tried to do it using ''cubic'' and it worked much better which i expected since the background does not look very linear. Just out f curiosity, i then tried to set to zero the K2 and K3 terms of the cubic function, essentially leaving myself with a linear function. I just wanted to see if the software will give me similar coefficients as in my initial try when i used ''linear'' baseline. However, the coefficients were very different but the goodness of fit (chi square) was identical to the one i obtained when i initially used the option of ''linear'' baseline. Any ideas for that? I hope my message is not very confusing. Thanks a lot


HJDrescher
Posts: 311
Joined: 2015-01-20
Location: Germany

This might arise from different initial guesses and too large epsilon values.
Hard to tell without displayed data and fits...

From the manual (Overview of Curve Fitting):

Unless you know a great deal about the fitting function and the data, it is unwise to assume that a solution is a good one. In almost all cases you will want to see a graph of the solution to compare the solution with the data. You may also want to look at a graph of the residuals, the differences between the fitted model and the data. Igor makes it easy to do both in most cases.

HJ


Posts: 1796
Joined: 2007-06-29
Location: United States

In order to avoid problems with a cubic polynomial overflowing on large X values, the cubic baseline fits to a normalized X range. So, yes, the fitted linear term should be quite different from the linear baseline. But the result should be pretty much identical.

John Weeks
WaveMetrics, Inc.
support@wavemetrics.com


Back to top