deletions | additions       diff --git a/Statistical analysis.tex b/Statistical analysis.tex  index 7ce3ae0..1fc873c 100644  --- a/Statistical analysis.tex  +++ b/Statistical analysis.tex 
...
\end{equation}  The covariance matrix obtained from the numerical minimization of the negative log-likelihood is then inverted to get the uncertainties on the form factors, $\sigma(\delta_i)$. It can be shown~\cite{Davier_1993} that, in the linear form given in Eq.~\ref{eq:optimal}, this method leads to optimal statistical uncertainties on the form factors. The functions $f_i(x,\cos\theta)$ are therefore called "optimal observables". It turns out~\cite{Diehl_1994} that the covariance matrix, hence the statistical uncertainties on the form factors, can be obtained analytically in the limit of large number of events, which is the case considered in this letter. If the total event rate is included in the derivation of the likelihood, as is the case in Eq.~\ref{eq:likelihood}, the elements of the covariance matrix $V$ are given by (${\rm d}\Omega = {\rm d}x {\rm d}\cos\theta$)  \begin{equation}  \label{eq:rate}  V_{ij} = {\cal L} \int {\rm d}\Omega { f_i \times f_j \over S^0} \ ,  \end{equation}  while if the total event rate is not included in the likelihood, namely by removing the first term of the product in Eq.~\ref{eq:likelihood}, these elements take the form  \begin{equation}  \label{eq:norate}  V_{ij} = {\cal L} \left[ \int {\rm d}\Omega {f_i \times f_j \over S^0} - { \int {\rm d}\Omega f_i \int {\rm d}\Omega f_i f_j  \over \int {\rm d}\Omega S_0 } \right], \end{equation}  and the uncertainty on the form factor $\delta_i$ is simply   \begin{equation}  \sigma(\delta_i) = \sqrt{ [ V^{-1} ]_{ii} } \ .  \end{equation}  This analytical procedure is used in Ref.~\cite{Grzadkowski_2000} to determine the sensitivity to top-quark electroweak couplings in $500\,{\rm fb}^{-1}$ of ${\rm e^+ e^-}$ collisions at $\sqrt{s} = 500$\,GeV. In this article, the authors use Eq.~\ref{eq:rate}, but they let the total number of events float by adding a fictitious form factor $\delta_0$ in front of $S^0$. It was numerically checked that this work-around is equivalent to using Eq.~\ref{eq:norate}, {\it i.e.}, to not use the total event rate in the likelihood.