smk78
commented
4 years ago After further testing, it appears that both 4.2.2 and 5.0.2 mostly give ~same extracted parameters as the CCP13 CORFUNC (as noted above absolutely identical parameters are not expected). What has, however, been highlighted, is the much greater impact that the choice of high-Q extrapolation fitting range and background level can have on the shape of the 1D correlation function.
Parameter__________________________CCP13 CF__________4.2.2__________5.0.2
Average Long Period 76.00 74.07 74.1
Average Hard Block Thickness 13.95 12.56 12.6
Average Interface Thickness 2.00 0.00 0.0
Average Core Thickness 4.00 2.64 2.65
Local Hard Block Fraction 0.183 0.169 0.17
Polydispersity 0.223 0.330 0.331Whilst I still regard the Interface & Core Thicknesses from the CCP13 CF as suspicious, the Interface Thickness from SasView needs investigation, as does the fairly significant discrepancy in the Polydispersity.
butlerpd
lucas-wilkins
Elliot Gilbert has been playing with Correlation Function analysis in SasView (in 4.x) in an attempt to understand what impact systematically varying aspects of the scattering data have on the derived correlation function and, more importantly, the extracted parameters that quantify the underlying lamellar morphology.
When @rprospero's corfunc-py code was assimilated into SasView I did test it against the old CCP13 code. I have done so again using the BSL \convertible_files\Z83000 & Z98000 and their regular \1d_data\ISIS_83404 & ISIS_98929 equivalents and, whilst the extrapolations and transforms compare pretty favourably, they are not identical. But then I would not expect them to actually be identical since the two implementations are not using identical code/libraries.
However, there is indeed some variation in the extracted parameters (in the following SV=SasView & CF=CCP13 CORFUNC). Testing the Z98000/ISIS_98929 data we have:
Background values of 0.3085 (SV) vs 0.3095 (CF) - not too bad at all. Hard block thickness is 17.1 (SV) vs 13.9 (CF) – close enough?
But then:
Long period is 31.6 (SV) vs 76 (CF) – I think here the SV extract algorithm has picked on a little hump in the base of the big minima in Gamma1(x), not present in the CF transform, rather than the big peak/hump around 70-80. In essence the code needs to look for a change in slope and in the SV case it's been misled. Interface thickness is 4.6 (SV) vs 2.0 (CF) Core thickness is 9.7 (SV) vs 4.0 (CF) *But I regard the fact that these are 2.000000000000 and 4.00000000000 as deeply suspicious!!! Local crystallinity is 0.54 (SV) vs 0.18 (CF) – but since this is calculated as hardblock thickness/long period it is bound to be different.
A lot of this will be down to the subtleties of the differences in the actual shapes of the 1D CF because of the way the way the lamellar morphology parameters are extracted. But we should satisfy ourselves that the code is doing what we expect.