If you do not specify errors, the system cannot (in principle) compute errors on parameters. I have implemented a simple algorithm in the new version of TGraph::Fit in CVS that compute some default errors if no errors are available on the points (case of TGraph, or TGraphErrors with null errors). The algorithm evaluates the fit function at x+epsilon and x-epsilon. If somebody has a better idea and algorithm, let me know. Rene Brun On Thu, 11 Jul 2002 kesisogl@fnal.gov wrote: > Hi, > > I have a question about the usage of TGraphErrors(n,x,y). > Can we set the errors on both x and y equal to zero > and then fit the graph with a given function? > I've tried to do so, and from what I can see the quality of the fit > is not very good (I try to avoid the usage ot TGraph since sometimes I > have errors and sometimes I don't). > Someone gets a much better quality if he specifies a small non-zero > error. > If that's the case (small non-zero errors) is there any special > value that gives the best fit or the smaller errors the better? > > Thanks! > > Stelios. >
This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:50:59 MET