This iteration of the SPICE model relies on a global fit to timing profiles of arriving photons to constrain the correlation coefficient between the effective scattering and absorption. In the previous iterations this coefficient was determined mainly by the value obtained from the AHA ice model (although some freedom was allowed around that value).
Additionally, a problem that led to ~3% rate reduction was corrected (in version v23). The problem affected fits where more than one flasher event was simulated (for the same DOM). The previously fitted value of py=2.1 is unchanged (since only the faster one-event-per-flasher simulation was used to calculate py). The following calculation was redone with the fixed version; the plots in the Appendix1 demonstrate the steps described in much detail here.
As before, the ice table obtained from the correlation to dust profiles was refined with 10 more iterations of the fits to flashers (with 10 flasher events simulated):
The table so refined was scaled in scattering and absorption in the vicinity of the found solution, as shown above. The fits not using the timing information have little or no ability to resolve the value of the correlation coefficient between scattering and absorption. The minimum has an extremely oblong shape, and the direction of its longest extension is determined. The point along the line drawn in this direction is chosen to minimize the likelihood that does use the timing information (which appears to have an adequate ability to resolve the correlation coefficient).
Thus, the bulk of the fits is still performed with only the total charge information, however the relationship between the scattering and absorption coefficients is now determined by fits that do use the timing information. The best point shown on the plots above determines the scaling coefficients that are applied to the ice table before the refining procedure:
As expected, the new location of the best point is now virtually indistinguishable from (1.0, 1.0).
The method described here relies on the total charges collected by each DOM to determine the variations of the ice parameters with depth. Charge recorded by a DOM has been shown to be linear in a wide range of values (up to ~1000 p.e.) with respect to the input charge, saturation effects at the higher end being more than well covered by the systematic error belt of 20% (as used in the method described here). The more fragile (due to various observed problems) timing information is only used in a global fit to determine the correlation coefficient between scattering and absorption.
The fits could not be improved much further by allowing the scattering and absorption coefficients to vary freely (for either likelihood description: with timing or charge-only), which implies the hypothesis that the two coefficients are correlated is not only physically motivated (since both scale with concentration of dust particles in ice) but also well supported by the data.
The distance between the minimum of the likelihood using timing fits and the best value along the minimum of the likelihood that only uses charge information can be used to judge the unresolved residual uncertainty in the description of the flasher data. This uncertainty (5%, mostly in absorption coefficient, as seen from the plot above) is likely due to the approximation steps taken in describing the IceCube flashers and due to lack of some information. Here is a tentative list of such potential issues: