Fig. 2.

The figure of merit, Eq. (13), for the Hessian representation of the NNPDF3.0 NLO set, computed for points which satisfy the gaussianity criterion Eq. (12), plotted versus the threshold $\mathit{ϵ}$ Eq. (12) and the number of eigenvectors $N_{\mathrm{eig}}$, after the choice of basis replicas has been optimized through a run of the GA with the settings of Table 1. The value of the estimator along the valley of minima, i.e. the curve determined by finding the value of $N_{\mathrm{eig}}$ at which the estimator has its absolute minimum for each $\mathit{ϵ}$ is shown (blue curve). The red curve marks the value $\mathit{ϵ}=0.25$ which is finally adopted. The projection of the valley of minima in the $(\mathit{ϵ},N_{\mathrm{eig}})$ plane is shown in Fig. 3