Documentation of EM-Bright¶
Introduction:¶
Binary system of neutron stars and black holes are some of the strongest and best understood emitters of gravitational waves. Additionally, when a system has a neutron star then there is also a finite probability of disrupted baryonic matter present after the coalescence. This disrupted matter could be either from physical collision of the two compact objects as in the case of a binary neutron star (BNS), or from tidal interactions between the two compact objects with at least one of them being a neutron star (NS). This baryonic matter is generally extremely neutron rich, and will undergo r-process neucleosynthesis producing heavy elements that subsequently goes through nuclear fission, producing large quantity of energy. This may result in an electromagnetic signal, commonly known as a Kilonova. Simultaneous observation of Kilonova and gravitational wave resulting from the coalescence of the gravitational wave is one of the most sought after astrophysical transient phenomenon. However, these events are much rapidly evolving than supernova, and the localization region of gravitational wave can be several tens of square degree in size. Thus, early knowledge about the possibility of an electromagnetic counterpart of a gravitational wave event can be helpful for observers.
Similarly, another interesting astrophysical event would be the binary coalesence of compact objects in the “mass gap” region. Stellar evolution models predict that black holes with masses in two ranges cannot be directly formed by the gravitational collapse of a star. These mass ranges are distinguished as the “lower” and “upper” mass gaps. The lower mass gap is traditionally considered to be in mass range fom 3 to 5 M_{odot}. Astronomers may be interested in following up gravitational-wave sources whose component masses lie in this “lower mass-gap” region between neutron stars and black holes.
The EM-Bright package provides tools for computing a machine learning based score for the potential presence of an electromagnetic counterpart and the presence of a mass gap object in a merger of two compact binary objects. For a given compact binary coalescence event the EM-Bright code provides three scores, HasNS, HasRemnant and HasMassGap. The first quantity gives a score for the presence of a neutron star in the binary. The second quantity is the score of non-zero tidally disrupted matter to be present outside the final object after coalescence. Both these quantities are EoS maginalized. In the case of a neutron star -black hole we use a fitting formula of numerical relativity results as provided in Foucart. The third quantity gives the score to quantify whether the binary system has at least one compact object which lies in the lower mass gap region.
EM-Bright Calculation:¶
The knowledge of the masses and spins of the binary will allow us to compute the HasNS , HasRemnant, and HasMassGap scores. However, the source parameter information are poorly known in the low-latency, it might be hours before we get the first results from rapid parameter estimation to directly compute the EM-Bright scores. To address this issue, we implement a supervised learning technique to compute HasNS and HasRemnant EMBright-paper. In its current implementation we apply a nearest neighbor supervised learning technique to train the classifier based on a large set of simulations. Similary, we compute HasMassGap using a supervised learning technique, Random Forest, trained on a similarly large set of simulations.
In this study we inject compact binary coalescence signals in noise stream of LIGO and Virgo detectors. We recover these injections using the detection pipelines used by the LIGO/Virgo collaboration. The recovered parameters exhibit deviation from the injected parameters due to pipeline systematics. We train the classifier to identify the “true” nature of the event based on the injected parameters where the feature set is the recovered parameters.