Documentation › Glossary

Glossary

Model: A function f(x; p) proposed to describe the data, with parameters p to be determined.
Parameter: A quantity the fit adjusts, written [0], [1], … in a ROOT expression or as a symbol in the equation editor.
Starting value (initial guess): The value of a parameter at which the minimiser begins. Not an uncertainty, and not a constraint.
Standard uncertainty: One standard deviation of a measured or derived quantity: the half-width of its 68% interval. The only kind of uncertainty the tool uses.
Residual: Data minus model at a point, y_i − f(x_i).
Pull: A residual divided by the point's effective uncertainty. For a correct model and correct uncertainties the pulls follow a standard normal distribution.
χ² (chi-square): The sum over fitted points of pull². The quantity the fit minimises.
NDF: Degrees of freedom: fitted points minus free parameters.
χ²/NDF (reduced chi-square): χ² per degree of freedom; about 1 when the model and the uncertainties are both right. Its spread is roughly √(2/NDF).
p-value: The probability that a correct model would produce a χ² at least as large as the one observed.
Poisson deviance: The likelihood-ratio statistic reported for histogram fits by Poisson likelihood; plays the role of χ² and approaches it at large counts.
Minuit: ROOT's minimiser. Its status code is the convergence status in the report.
Hesse matrix: The matrix of second derivatives of χ² at the minimum; its inverse is the covariance matrix.
Covariance matrix: The matrix whose diagonal holds the squared parameter uncertainties and whose off-diagonal elements express how pairs of parameters vary together.
Correlation: Covariance normalised to the range −1 to 1. Near ±1 means the data determine only a combination of the two parameters.
Effective variance: σ_y² + (f′·σ_x)²: the method by which an x uncertainty is included in the fit.
Confidence band: The uncertainty of the fitted curve itself at each x, from the parameter covariance. Not a prediction interval for new data.
Fit range: The interval of x inside which points are fitted; points outside are plotted but ignored.
Point exclusion: A single measurement left out of the fit, with a recorded reason, while remaining in the data and on the plot.
Stale: A fit result whose data or model have changed since it was computed. Calculations that depend on it are blocked until it is refitted.
Propagation of uncertainty: Carrying the uncertainties and covariances of inputs through a calculation to the uncertainty of its result; done here to first order with numerical derivatives.
Session: Everything in both pages: datasets, models, results and calculations. Autosaved in the browser; saved to a .json file with Save.

← All documentation