I) PCA of spectral profiles of ~500 S0 MaNGA galaxies: Radial binning and stacking of spaxels to generate the spectral profile of each galaxy (mean spectrum as a function of the galactocentric distance). Projection of the profiles on their first 2 principal components (~90% of the sample variance [4]). The closer to the PS in the PC1 - PC2 space, the weaker the activity (SF/AGN)[2].
II) Vectorization and quenching classification: Profiles are converted into vectors. Then, galaxies are classified according to the orientation of these vectors wrt the PS ridge. This orientation establishes how quenching proceeds in galaxies.
More details on the method
The set of principal components obtained for the MaNGA spectral profiles is equivalent to that derived in [4] for a sample of SDSS single-fiber spectra of ~70,000 S0 galaxies.
Figure 1. Black lines: Mean spectrum and first three principal components (ES) of the mean flux of the radial bins of the ~500 S0 galaxies. Red lines: residuals from subtracting to each component represented here the corresponding component derived from the SDSS single fiber spectra of S0 galaxies from [4].
In [4], S0 galaxies were classified in the PC1 - PC2 diagram into Passive (PS), Transition (TR) and Active (AC). Each class correspds to 70%, 5% and 25% of the population. The properties of galaxies of each class differ from each other, hence it is convenient to study them appart.
Figure 2. Projections of the SDSS single fiber spectra of ~70,000 S0 galaxies on their first 2 principal components derived in [4].
Figure 3. Distribution of probabilities of belonging to the PS class obtained from a logistic regression employed for classifying galaxies in the PC1 - PC2 diagram. Classes are deffined at the 95% of the interclass variance.
From [2] we know that the level of activity (SF/AGN) of a spectrum strongly correlates with its distance to the PS.