Errors computation: general question

2 very general questions on error computation in root.

When I am fitting some data in a TGraphErrors and I have such a small number of points that the NDF=0 (eg a line on two data points), how does root computes the error on the fit parameters?

When I am fitting a function on a TGraph without errors, how does the parameters errors are computed in this case?

Thank you,

The error is computed from the covariance matrix which is obtained from the inverse of the Hessian.
When fitting a graph without errors, the obtained parameter errors are re-scaled using the obtained chi2 from the fit.
In case you have only two data points and you are fitting a function with two parameters, the chi square will be zero so the rescaling will not work anymore, and you cannot interpret the obtained values as parameter errors.
Anyway if you want reliable error by fitting a graph, you need to assign some uncertainties on the data points.



I am sorry to ask, what do you mean by rescaling, exactly?

See note 5 in ROOT: TGraph Class Reference