| | 3 | The initial guess for L is in general too large for point-like objects. First, there's an error in the code that chooses between the elliptical major axis and the first radial moment, resulting in a large initial guess. Fixing this doesn't completely work, so the idea was to introduce an offset, such that an object with a size similar to the width is probably point like, so we should normalize by subtracting that width off. This shrinks the length again, but insufficiently. Looking at large numbers of objects (click on full skycell tables in attachments below), it's clear that dL/dIteration is almost universally positive, suggesting an issue with the calculation (typo in the code?) or algorithm. |
