-
Notifications
You must be signed in to change notification settings - Fork 5
Paper II
Investigation of systematic uncertainties in nebular abundance determinations
Analysis of a large number of line lists, using different combinations of
- ICF (PTP92 or KB94)
- reddening law (CCM, Fitzpatrick, Howarth galactic, Howarth LMC, Prevot SMC)
- atomic data for collisional lines (ad hoc compilation, chianti 5.2, chianti 6.0, chianti 7.0)
- atomic data for helium (Smits 96, Porter et al. 2012)
Additional effects that could be investigated
- effect of systematic uncertainties in the uncertainty estimation itself, ie artificially vary the estimated uncertainty on line fluxes and see how it affects the derived quantities.
- effect of bugs. Download and compile every revision of NEAT, analyse a line list with every one of them, see how the derived abundances vary with time. I think I could script this. Not sure how revealing it would be but I'd quite like to try it.
Basic idea is to compare the statistical uncertainty distributions resulting from different analyses, to see if the different choices made have a statistically significant effect on the result.
We might be able to say which factors introduce the largest systematic uncertainty into abundance determinations. For example, maybe the choice of reddening law can result in statistically significant differences in derived abundances, while the choice of helium atomic data does not. This would be extremely useful for nebular people to know.
Github wikis seem incredibly stupid and provide an entirely useless "add image" button that does nothing. So here's a link to a highly preliminary figure that might in some sense resemble a way to present our data meaningfully.
http://zuserver2.star.ucl.ac.uk/~rwesson/neatplot.png
I did a test run for NGC 6543, using all combinations of 5 reddening laws, 2 He atomic datasets and 2 ICFs. This only shows the results for one reddening law. Four separate He/H probability distributions are shown, with the distribution of all results also shown. I don't really know what the summed distribution of all results really means, but possibly we could say that its standard deviation (about 30% larger in this case than the individual ones) is some kind of estimate of a truer uncertainty than the statistical uncertainties alone.