Top gradient: The two colours interpolated in perceptual space (inaccurate) Bottom gradient: The two colours, converted to linear space, interpolated, then converted back (accurate) Our eyes see light non-linearly: a light dimmed to produce 50% of the photons will look 70% as bright to our eyes. So, our monitors produce light output non-linearly: a monitor asked to produce 50% brightness (or an R/G/B value of 128) will only produce 25% so it looks like 50% to our eyes. But our eyes still see *colour mixing* linearly: colours blend much better if you blend them 'raw'. Since monitors expect perceptual brightness values, we have to convert them to linear values, blend them, then convert them back to perceptual values. A colour that's 50% bright should be converted to 70% photons, blended with the other colour, and -then- converted back to brightness, and that's what the bottom gradient does.
