In an interesting paper by Antilogus, et al subtle effects of electron interactions inside the CCD material itself are discussed, which could be a tool for understanding and perhaps quantifying CCD nonlinearity.

We have tested our own Andor BU987 CCD camera for these effects and see them.

Plots of row- or column-neighbour means vs correlations. Top row: first panel: full set of values of means vs column-correlations
– we note the strong non-linearities that set in at about 50.000, and
some outliers.

Top row: second panel: for just the range of mean values below the onset
of non-linearity a robust regression is performed (red line).

Bottom row: same as top row but now for row-correlations.

We note the general correspondence to results in the ‘brighter-fatter’
paper: rows are more strongly correlated with their neighbours than are
columns – the ratio of slopes is about 2.0.

Can this be used quantitatively to correct for non-linearity in our CCD? The camera linearity was already tested for linearity while in Lund. See Figure 8 in the report.