In trying to understand planet formation, one of the most powerful tools we have is to compare the distributions of the physical and orbital properties of the planets discovered with those predicted…

This is a $565,916.11 grant awarded by NASA to Las Cumbres Observatory Global Telescope Network, Inc., a private non-profit organization that operates a global telescope network. The grant supports the development of a two-stage detection and classification system to analyze data from the upcoming WFIRST mission's microlensing survey. This will help identify exoplanets in the gap between previous detection methods. The project also involves analyzing data from Las Cumbres' existing microlensing survey called ROME/REA to train and validate the detection algorithms. All software developed will be released publicly for community use. This grant supports NASA's Exoplanets Research Program by advancing preparations for the WFIRST mission and providing tools to discover and characterize exoplanetary systems.

In trying to understand planet formation, one of the most powerful tools we have is to compare the distributions of the physical and orbital properties of the planets discovered with those predicted by formation and evolutionary models. For this to be effective, we need a complete census of planetary systems at all masses and orbital separations. But no one survey technique is sensitive to the whole parameter space. Kepler provided a landmark catalog of planets at small orbital radii around their host STARS, complementing the results of other mostly ground-based projects, but there is a significant GAP in the detected population of low-mass planets at separations of few-10 AU, where most detection techniques suffer a dramatic drop in efficiency and/or sensitivity. Microlensing offers a way to bridge this GAP, and so is one of the goals of the forthcoming wfirst mission. Preparatory work for the wfirst microlensing survey is already underway. A major element of IT will be to develop tools capable of analyzing the large resulting dataset and detecting microlensing events within, while carefully distinguishing between false positives and systematics. This presents a number of challenges, since the analysis of just one binary lensing event can be computationally intensive, and there is a marked shortage of people (~10) in the US who specialize in this subject. Existing software tools for this task are mostly proprietary, and have limitations which mean they will not scale to process the wfirst dataset in a timely manner. We will address this challenge by developing a two-stage detection and classification system for the wfirst bulge survey data using a combination of machine learning techniques. The first stage builds on prototype random-forest microlensing/variable star detection software already developed by our team to distinguish astrophysical signals from systematics or false positive detections. We propose to combine this with a second stage which will leverage recent progress in deep learning software to analyze the lightcurves of binary lensing events and identify all possible models which are consistent with the data. This will address a long-standing problem in the modeling of binary lenses, for which there are a number of known degeneracies. The effectiveness of machine learning systems depends critically on the quality and quantity of the data used to train the system. We will to leverage a number of public photometric catalogs to help to train the system, and apply IT to analyze data from the ongoing microlensing key project at LCO, rome/REA, a high cadence, TRI-band optical survey for microlensing events in the galactic bulge. Our team was awarded 2650 HRS/YR for 3YRS starting in 2017 for this survey, which is currently in its second observing season. We will release a complete catalog of the imaging and photometry from rome/REA for community use through a NASA archive. We will also conduct simulated-event injection tests to evaluate the algorithm's performance and detection statistics. Once the algorithm is applied to wfirst data, this analysis will enable US to calculate the true planet frequency. All software developed through this proposal will be thoroughly tested and documented, ahead of its public release for community use. This proposal is highly responsive to the aims of the exoplanets research program, since IT will discover and characterize planetary systems and advance preparations for the wfirst mission. IT will also provide a valuable resource for a wealth of stellar astrophysics.