Dear Rooters:
I have about 1000 sets of experimental data. each set of data has 6~10 points. The 1st point equals to the sum of other points multiplied by the coefficients. I know there is a subfunction called Fumili written in Fortran. You can call this Fumili subfunction to get the coefficients by minimizing chi2 distribution. Is there a c++ example for this function in root?

Hi,
in ROOT is possible to use the Fumili algorithm. It has been implemented both in the TFumili class and also in the Minuit2 package.
You can easly use for fitting an histogram or graph by switching the minimizer. Otherwise, for customized chi2 functions, I have already posted some examples in an old thread in the forum: root.cern.ch/phpBB2/viewtopic.php?t=7821

Hi, Lorenzo:
Thank you for your reply. I have read your example. Maybe I didn’t understand your example well, I don’t think your example has resolved my problem.
I want to do a calibration. I have total 20 PMTs and each physical event will fire 6~8 PMTs. The energy of this event will be deposited in these fired PMTs completely. I know the total energy of each event and the deposited energy in each PMT. If I want to get the coefficient of each PMT, I can let the total energy equal to the sum of the energy in each PMT multiplied by an initial coefficient. Then use Fumili to get the real coefficient. It seems difficulty to fit them in a histogram or graph.

In your case you can write a chi2 from the difference between the total energy (ETOT) and the reconstructed one (EREC). EREC will depend on the calibration constants which are the parameters of your fit.
If you have a linear dependency on these constant you can do a simple linear fitter, which is the fastest one, or using Fumili with a user provided chi2 function as in the example exampleFumili.C provided in the old forum thread ( root.cern.ch/phpBB2/download.php?id=3525 )

Yes, it is linear dependency on these constants. I check the example of fitLinear2.C and make a simple test. Unfortunately, it seems not working.
Please see my simple example. test.C (831 Bytes)

However, with this data the linear fitter for f(x,y) = Ax+By will fail because the matrix is singular, since for every point x = y and A is 100% correlated with y. So you can reduce this problem to a simple fit for f(x) = Ax
If x is independent of y, the linear fit will work