Improve rebinning of SofQW

[Nick Draper]

The 2D rebinning of SofQW is simplistic and needs to be improved.

The underlying data is not simple but need to be rebinned into a regular grid with error analysis. Sample code from lamp at the ILL is attached. and some potentially useful code from matlab.

[Nick Draper]

Bulk move of tickets at the end of iteration 27

[Nick Draper]

Bulk move of tickets at the end of iteration 28

[Nick Draper]

"New" tickets moved at the code freeze of iteration 29

[Martyn Gigg]

In [13369]:

Refs #2233. New SofQW looks correct but is slow so needs improvement. Test is disabled for this reason at the moment but I'd like to checkpoint my work.

[Martyn Gigg]

In [13370]:

Fix for OpenMP loop type for windows. Refs #2233

[Martyn Gigg]

In [13372]:

Fix for multiple constructor matches on Mac. Refs #2233

[Martyn Gigg]

In [13411]:

Improved the performance of SofQW2. Enabling the unit test. Still room for improvement though I think. Refs #2233

[Russell Taylor]

In [13428]:

Fix for Intel compiler getting stuck. Re #2233.

[Martyn Gigg]

In [13472]:

Added dedicated Quadrilateral class that has no dynamic vertex allocation to improve speed for rebinning. Refs #2233

[Martyn Gigg]

In [13473]:

Speed improvements for rebinned SofQW. At the moment the algorithm name is SofQW2 so that it doesn't supercede the other until I get verification that it is doing the right thing. Re #2233

[Martyn Gigg]

In [13623]:

Refs #2233. Correct the theta binning to only take in to account the detector's angular width rather than a continuum between this and the next.

[Martyn Gigg]

In [13624]:

Refs #2233 Update tests values.

[Stuart Campbell]

I tried the following script...

LoadEventNexus(Filename='/home/scu/Code/Mantid/Test/Data/BSS_16715_event.nxs',OutputWorkspace='WS')
ConvertUnits(InputWorkspace='WS',OutputWorkspace='WS',Target='DeltaE',EMode='Indirect')
Rebin(InputWorkspace='WS',OutputWorkspace='WS',Params='-0.012,0.0004,0.012')
SofQW(InputWorkspace='WS',OutputWorkspace='16715_sqw',QAxisBinning='0.2,0.2,2',EMode='Indirect')
SofQW2(InputWorkspace='WS',OutputWorkspace='16715_sqw2',QAxisBinning='0.2,0.2,2',EMode='Indirect')

The output from SofQW looks ok, but the output from SofQW2 is all 0.0's.

I also get the same for CNCS, but it takes about 10mins to run.

[Stuart Campbell]

After discussing with Nick, this can go to the next iteration.

[Martyn Gigg]

Refs #2233. Make SofQW2 a better at checking what information is required.

The algorithm will now throw if Ei has not been provided in Direct mode, whether within the workspace from GetEi or set explicity. In Indirect mode it will throw if it cannot find efixed on a component or in its tree.

Changeset: 45e67dccdc37761e6bdb1370ccc43398850177c3

[Martyn Gigg]

I think this should fix the problems with everything being zero.

[Martyn Gigg]

Refs #2233. Make SofQW2 a better at checking what information is required.

The algorithm will now throw if Ei has not been provided in Direct mode, whether within the workspace from GetEi or set explicity. In Indirect mode it will throw if it cannot find efixed on a component or in its tree.

Changeset: 45e67dccdc37761e6bdb1370ccc43398850177c3

[Nick Draper]

There remains a known issue about how this handles masked data, but that is dealt with in a separate ticket

[Stuart Campbell]

• Load CNCS_7860_event.nxs from AutoTestData directory.
• ConvertUnits into DeltaE, Direct Mode, Efixed = 3.0
• Rebin -2.9,0.01,2.9

Run SofQW:(QAxisBinning="0.0,0.2,2.0")
without specifying Efixed/Ei:

Error in execution of algorithm SofQW:
Workspace2D::getSpectrum, histogram number out of range

specifying Ei explicitly: --> Works OK - except Q range looks to go -ve.

Run SofQW2:(QAxisBinning="0.0,0.2,2.0")
without specifying Efixed/Ei:

No errors - took 577 seconds.

specifying an Ei explicitly:

No errors - took 810 seconds.

Again data looks like it is going -ve, but I think that this is a centre/edge of bin issue.

[Stuart Campbell]

Uploaded screenshot to show how the data seems to go into negative Qs'.

[Stuart Campbell]

• Load BSS_15600_event.nxs
• ConvertUnits into DeltaE, Indirect, default Efixed.
• Rebin, -0.012,0.001,0.012

Run SofQW2 (QAxisBinning="0.0,0.2,2.0"): Using default Ef:

SofQW2: Cannot find EFixed parameter for component "0". This is required in indirect mode. Please check the IDF contains these values.
Exception thrown in parallel region
Error in execution of algorithm SofQW2:
SofQW2: error (see log)

Specifying Ef=2.08:

Works ok - Q goes -ve.

SofQW shows the same behaviour.

[Stuart Campbell]

Doesn't give zeros for BASIS anymore. The Q axis still seems to go -ve if you specify q binning of 0.0,0.2,2.0. But this can be a ticket for the next iteration.

[Stuart Campbell]

About the -ve issue - looks like just a plotting issue - the underlying data has the correct limits.

[Stuart Campbell]

This ticket has been transferred to github issue ​3080