New system for sub-pixel integration

[ConnorStoneAstro]

A key component of accurately generating light models is sub-pixel integration. If a light model changes significantly across a pixel it must be integrated over that pixel to produce a reliable flux estimate. This is already incorporated in AutoProf using a recursive box integration region near the center of each model. However a more flexible system which detects regions of high flux variability could be more robust and possibly more efficient. This would be a system to look at a model evaluated on a pixel grid and identify regions with large gradients, it would then call a sub-pixel integrator for the relevant pixels.

The challenges here include:

• efficiently identifying the high gradient pixels.
• efficiently integrate the pixels without oversampling and wasting resources.
• Only perform sub-pixel integration on the relevant pixels, potentially resulting in odd shapes with integration.

A system which could evaluate models at a flattened per pixel level could be beneficial, though it would be tricky to then add these back into the main model

This discussion is a place to keep track of ideas related to this topic

[ConnorStoneAstro]

This is now incorporated in the code! Though it uses an approximation of the second derivative, since that is actually where errors appear.