Peter Melchior

The first line you linked needs to stay in place because otherwise the return value isn't the spectrum of a point source. The second line is indeed problematic for very...

Sure, no question about that, but read the docstring. The return value of this method is the SED of a point source! This needs to be stated (and obeyed) because...

I have an idea for this problem. In weak-lensing land we use moment-based estimates like the galaxy resolution R = 1 - T_psf / T_gal as an indicator how close...

Are you running `set_spectra_to_match`?

Then it isn't the mechanism we identified. Remy's report indicated that the mismatch in the spectrum amplitude is driving the ringing. This mismatch is now absent. We also concluded that...

What's the difference between the top and bottom row in each case?

And how is the second set not also multi-resolution? There are clearly different numbers of pixels involved?

This is a different version of the earlier problem: massive mismatch of the initial model. In this case it happened since the morphology of the model in multi-resolution is determined...

autograd's `hessian` only works for a single-parameter model. Other autodiff frameworks may not even have a Hessian implementation. However, a cheaper and viable way is the approximation of the Hessian...

This would be a good issue to get started. However, as I said in #51, we're shuffling the code, so I'd hold off for a few days, but then it...