Setting Relative Tolerance of VBBL Will Lead to Significant Speedup of MulensModel in High Magnification Region

[zhangjiyuan22]

Currently, explicitly setting relative error goal of VBBL (RelTol) hasn't been incorporated into MulensModel (see source/VBBL/VBBinaryLensingLibrary_wrapper.cpp, only setting absolute error goal, Tol, is implemented). If we set absolute error goal to 1e-2 and relative error goal to 1e-3, then when magnification is larger than 10, the relative error goal will be met first and the computation can stop earlier.

The following figure shows the speedup from RelTol=0 (the default value in VBBL if you don't set it explicitly) to setting RelTol=1e-3 (sufficient for ground base observation). The speedup is 5(10) at magnification 100(200).