A critical evaluation of the principal component analysis detection of polarized signatures using real stellar data

The general context of this study is the post-processing of multiline spectropolarimetric observations of stars, and in particular the numerical analysis techniques aiming at detecting and characterizing polarized signatures. Using real observational data, we compare and clarify several points concerning various methods of analysis. We applied and compared the results of simple line addition, least-squares deconvolution, and denoising by principal component analysis to polarized stellar spectra available from the TBLegacy database of the Narval spectropolarimeter. This comparison of various approaches of distinct sophistication levels allows us to make a safe choice for the next implementation of on-line post-processing of our unique database for the stellar physics community