Fig. 7

a The easiest way to localize the border between the anisotropic and isotropic regions in Fig. 5 is to track the sequential couples of scatter plots along the blue-dashed arrow while analyzing their anisotropy as shown in (a). The point where the scatter plot turns from anisotropic to isotropic denotes a reasonable cut-off for the meaningful principal components. To make the procedure automatic the quantitative criterion for anisotropy is needed; b–e compare the different anisotropy criteria: covariance (b), multivariate skewness (c), matrix pureness (d) and projected histograms (e). The component index plotted in the horizontal axis corresponds to the lowest index of the evaluated couple of components. Among all the considered methods, the method of projected histograms (e) performs best in separating the anisotropic and isotropic regions. The anisotropy criterion oscillates very closely to zero in the isotropic region as apparent from the inset in (e)