Open JohannesBuchner opened 5 months ago
You may also want to have a look at all calls to np.exp, and see if np.expm1 may not be preferrable for numerical stability.
For example
np.exp(1.43877735e8 / (tbb*wav)) - 1.0
can be
np.expm1(1.43877735e8 / (tbb*wav))
Hi Matthew,
This part of the code
is probably not too fast because of the convolution.
On another project I spent some time looking into the BC. It turns out one can find an analytic approximating form, by considering that the BC can be approximated by a triangle (linear function with cut-off), which is convolved with a Gaussian.