Evolutionary algorithm-based analysis of gravitational microlensing lightcurves

TL;DR

This paper introduces a new evolutionary algorithm for autonomous, real-time fitting of gravitational microlensing lightcurves, demonstrating high success rates and efficiency compared to traditional methods, with potential for complex modeling tasks.

Contribution

A novel, simple, and robust evolutionary algorithm tailored for fitting microlensing lightcurves, capable of high success rates and parallelization for real-time applications.

Findings

Achieves over 90% success rate on synthetic noisy binary-lens lightcurves.

Fits each lightcurve in under 20 minutes on a desktop computer.

Outperforms conventional and neural network-based fitting approaches.

Abstract

A new algorithm developed to perform autonomous fitting of gravitational microlensing lightcurves is presented. The new algorithm is conceptually simple, versatile and robust, and parallelises trivially; it combines features of extant evolutionary algorithms with some novel ones, and fares well on the problem of fitting binary-lens microlensing lightcurves, as well as on a number of other difficult optimisation problems. Success rates in excess of 90% are achieved when fitting synthetic though noisy binary-lens lightcurves, allowing no more than 20 minutes per fit on a desktop computer; this success rate is shown to compare very favourably with that of both a conventional (iterated simplex) algorithm, and a more state-of-the-art, artificial neural network-based approach. As such, this work provides proof of concept for the use of an evolutionary algorithm as the basis for real-time, autonomous modelling of microlensing events. Further work is required to investigate how the algorithm will fare when faced with more complex and realistic microlensing modelling problems; it is, however, argued here that the use of parallel computing platforms, such as inexpensive graphics processing units, should allow fitting times to be constrained to under an hour, even when dealing with complicated microlensing models. In any event, it is hoped that this work might stimulate some interest in evolutionary algorithms, and that the algorithm described here might prove useful for solving microlensing and/or more general model-fitting problems.