I will present a new open-source code "caustics" (https://github.com/fbartolic/caustics) which enables the computation of photometric light curves for binary and triple lens events. The code is implemented in the JAX Python framework which enables the computation of *exact* gradients of the model likelihood with respect to all input parameters of the model. This is made possible by the use of automatic differentiation.
"caustics" uses the contour integration method to compute the magnification of an extended, limb-darkened source at a computational cost that is comparable to VBBinaryLensing for binary lenses. The cost of evaluating the magnification for triple lens systems is only 2-3x greater than in the binary lens case.
The availability of exact gradients of the likelihood for the first time enables the use of efficient optimization and MCMC sampling algorithms (for instance, Hamiltonian Monte Carlo) for fitting binary and triple lens events. The code is highly modular and extensively tested. In addition to binary and triple lenses, "caustics" also enables efficient computation of the finite source magnification for single lenses.