We develop a differentiable binary microlensing model with adaptive contour integration. The code extends VBBinaryLensing and uses JAX to compute accurate light curves and derivatives for optimization and inference.
In this paper, we develop a differentiable binary microlensing model using adaptive contour integration. The code is a modified version of VBBinaryLensing (Bozza 2010, Bozza et al. 2018) that can compute the binary microlensing light curve and its derivatives both efficiently and accurately. The implementation is based on JAX and is available on GitHub. With accurate derivatives, we can use gradient-based methods for both optimization and statistical inference. We also demonstrate the application of this code to real microlensing event modeling.
Caption: An example binary-lens light curve (top panel) and its derivative with respect to the planet-to-star mass ratio q (middle panel). Only microlux with \( E^d \), the new gradient error estimator proposed in this work, yields a smooth and accurate derivative curve.
An early conference presentation of this work is available at 26th-microlensing-conference.