2014MNRAS.437..241H

The mid-infrared (MIR) spectra observed with the Spitzer Infrared Spectrograph (IRS) provide a valuable data set for untangling the physical processes and conditions within galaxies. This paper presents the first attempt to blindly learn fundamental spectral components of MIR galaxy spectra, using non-negative matrix factorization (NMF). NMF is a recently developed multivariate technique shown to be successful in blind source separation problems. Unlike the more popular multivariate analysis technique, principal component analysis, NMF imposes the condition that weights and spectral components are non-negative. This more closely resembles the physical process of emission in the MIR, resulting in physically intuitive components. By applying NMF to galaxy spectra in the Cornell Atlas of Spitzer/IRS sources, we find similar components amongst different NMF sets. These similar components include two for active galactic nucleus (AGN) emission and one for star formation. The first AGN component is dominated by fine structure emission lines and hot dust, the second by broad silicate emission at 10 and 18µm. The star formation component contains all the polycyclic aromatic hydrocarbon features and molecular hydrogen lines. Other components include rising continuums at longer wavelengths, indicative of colder grey-body dust emission. We show an NMF set with seven components can reconstruct the general spectral shape of a wide variety of objects, though struggle to fit the varying strength of emission lines. We also show that the seven components can be used to separate out different types of objects. We model this separation with Gaussian mixtures modelling and use the result to provide a classification tool. We also show that the NMF components can be used to separate out the emission from AGN and star formation regions and define a new star formation/AGN diagnostic which is consistent with all MIR diagnostics already in use but has the advantage that it can be applied to MIR spectra with low signal-to-noise ratio or with limited spectral range. The seven NMF components and code for classification are available at https://github.com/pdh21/NMF_software/.