Linear Fitting (vertcal fit)

[GSI_Seb]

Hi there,

I have trouble with fitting vertical lines with ROOT using Minuit.
I think it comes from the fact that, on a vertical track, the slope of the fit line (defined by y = m*x + b) is infinite.
So, I end up with no fit line drawn and a slope which is crazy big (>10000!).
Maybe, I need to mention that I’m trying to fit TGraphs or TGraphErrors.

I tried to look at the Normal Equation method to maybe solve my problem, but I really don’t understand how it works (and, more importantly, if it’ll work in my situation).
In order to try it out, I slightly modified the file $ROOTSYS/tutorials/matrix/solveLinear.C to build a vertical line (ie all Xs = 1.0 in this case) and I ran it.
As expected, the Minuit method didn’t work and the Normal Equation method seems to give me something, but I don’t know what it is.

When I print the TVectorD (using .Print()) returned by the NormalEqn function, I get :

0 0.858824
1 1

The problem is that I’m not sure what those numbers are.

Could someone help me please ?

Thank you very much.

[oshadura]

perhaps @moneta or @vcroft can help here?

Thank you,
Oksana.

[moneta]

Hi,

Are your points exactly at the same x values ? In that case you will not be able to fit the data.
If not, you can try fitting them, but maybe try to parametrize the line as y = 1/A * x
However, if you have large error in y you will get a very large parameter on the slope, because it will be very little constraints by the data

Lorenzo