A system with many degrees of freedom can be described as a point in a space with many dimensions.

One approach to the analysis of these high dimensional data is to look for a linear projection onto a lower dimensional space. Principal components analysis focuses on the eigenvalues of the covariance matrix, hoping to find a lower dimensional description. But when the spectrum of this matrix is nearly continuous, any distinction between eigenvalues that we keep and those that we ignore becomes arbitrary. I will present a framework, developed in physics in the early fifties, to tackles this problem and I will suggest how to generalize this approach to analyse real data in complex systems. At the end of the talk I will touch upon the role of journals and the offer of Physical Review journals for publishing interdisciplinary research.