The code is here.

For each pixel position on the screen already containing information about the simulation, the gravity force vector is subtracted from it (see later), its components are rounded to the nearest integer, and the pixel at this new position is picked. Then this new pixel's colour is "added" (literal addition from grade 1) to the pixel's colour at the original position, component by component.

This in itself would merely cause duplication of image data, so the code actually subtracts the gravity force from the pixel position in a tricky way: instead of converting each gravity force unit to a single pixel position unit, it converts a gravity force unit to 0.750, 0.875, 1.000 pixel position units for the R, G, B colour channels, respectively. This causes the image data to be duplicated at different distances from its origin *per colour channel*, which causes a chromatic aberration effect.