The gradient of likelihood might be sped up by computing jacobson with respect to initial positions, and then just carrying out a single time the multiplication by the Jacobian of the initial conditions. This saves having to do this for each and every transit!
And this might be a better test of the numerical accuracy of the initial condition Jacobian.
The gradient of likelihood might be sped up by computing jacobson with respect to initial positions, and then just carrying out a single time the multiplication by the Jacobian of the initial conditions. This saves having to do this for each and every transit!
And this might be a better test of the numerical accuracy of the initial condition Jacobian.