Machine learning from hard x-ray surveys: applications to magnetic cataclysmic variable studies
University of Southampton, School of Physics and Astronomy,
Within this thesis are discussed two main topics of contemporary astrophysics. The first is that of machine learning algorithms for astronomy whilst the second is that of magnetic cataclysmic variables (mCVs). To begin, an overview is given of ISINA: INTEGRAL Scouce Identifiction Network Algorithm. This machine learning algorithm, using random forests, is applied to the IBIS/ISGRI data set in order to ease the production of unbiased future soft gamma-ray source catalogues. The feature extraction process on an initial candidate list is described together with feature merging. Three trainng and testing sets are created in order to deal with the diverse time-scales encountered when dealing with the gamma-ray sky: one dealing with faint persistent source recognition, one dealing with strong persistent sources and a final one dealing with transients. For the latter, a new transient detection technique is introduced and described: the transient matrix. Finally the performance of the network is assessed and discussed using the testing set and some illustrative source examples. ISINA is also compared to the more conventional approach of visual inspection. Next mCVs are discussed, and in particular the properties arising from a hard X-ray selected sample which has proven remarkably efficient in detecting intermediate polars and asynchronous polars, two of the rarest type of cataclysmic variables (CVs). This thesis focuses particularly on the link between hard X-ray properties and spin/orbital periods. To this end, a new sample of these objects is constructed by cross-corelating candidate sources detected in INTEGRAL/IBIS observations against catalogues of known CVs. Also included in the analysis are hard X-ray Observations from Swift/BAT and SUZAKU/HXD in order to make the study more complete. It is found that most hard X-ray detected mCVs have Pspin/Porb<0.1 above the period gap. In this respect, attention is given to the very low number of detected systems in any ban between Pspin/Porb = 0.3 and Pspin/Porb = 1 and the apparent peak of the Pspin/Porb distribution at about 0.1. The observational features of the Pspin - Porb plane are discussed in the context of mCV evolution scenarios. Also presented is evidence for correlations between hard X-ray spectral hardness and Pspin, Porb and Pspin/Porb. An attempt to explain the observed correlations is made in th context of mCV evolution and accretion footpring geometrirs on the whit dwarf surface.
Actions (login required)