Data going negative after subtracting background, different from collette

[Nick Draper]

Reported by Ann Terry:

Just seeing some differences in the output from Mantid compared to Colette. The data in Mantid appears to go negative upon subtracting a background whereas in Colette it is still positive. Sarah and myself have had a look and can’t spot any differences in bin size, q range etc. One guess from Sarah was if it could be due to the wavelength problem Richard highlighted since it is the data at high q which is going negative.

This may need some more input from Richard Heenan

Is this right and is there a solution? For now, users are working with Colette.

[Steve Williams]

This from Richard Heenan

So to recap, I am after finding out why 61400 gives lower I(Q) in Colette than in Mantid, as per the NEW2 plot attached (same one as this morning)

S             T         DB
61404        61392      61432
61400        61388  

In the ISIS SANS GUI the instrument should be set to LOQ and the emails above typed into its boxes on the first page. When you click Load and then Reduce1D the data shouldn't be more negative than Colette's (i.e. > -0.12)

[Steve Williams]

(In [10035]) Stop FlatBackground from removing too much background re #2599

[Steve Williams]

(In [10044]) Stop FlatBackground from removing too much background re #2599

[Steve Williams]

(In [10047]) Convert to distribution before calling FlatBackground re #2599

[Steve Williams]

(In [10306]) Flatbackground now works for non-distribution data and SANS Display Mask is now clearer re #2612 fixes #2599

[Steve Williams]

(In [10528]) Account for the sample moving on loading re #2599

[Steve Williams]

It appears to be fixed From Richard

The background run on the attached, run 61400 has extra large “flat background” in its M2 detector. Having fixed a bug results are now close enough to COLETTE for this run – I’m not expecting them to be identical at small Q as the methods are slightly different and the small Q data is very sensitive to the flat background subtraction under M2.

[Anders Markvardsen]

Looks like in the right direction and Richard is happy. There is a litte bit of negative values but judged to be within errorbars.

Note that hile testing this ticket finds that the following sequence of steps crashes MantidPlot:

1. Enter the numbers above in the example Steve provide
2. Mask from SVN
3. Load data
4. Tick 'Plot Result'
5. Click the '2D reduce' button

The error message which comes up says 'vector<T> too long'.

Investigating this further we found this appears to be related to log plotting set in preferences along the y-axis

[Stuart Campbell]

This ticket has been transferred to github issue ​3446