Investigate IntegratePeaksMD issue reported by Pascal

[Owen Arnold]

2 Days

[Owen Arnold]

I've tracked this down to the changes made in #4998. The basic procedure for determining whether an integration region is valid is to:

• Generate 64 test points on the sphere using the integration radius or the integration background radius.
• Convert those points into fictional peaks themselves.
• Using the QLab coordinates of those peaks, determine the Kf vector for each. Then using the sample location, use ray tracing to determine what detector would be hit. Use that as the detector for each peak. If no detector is found, then the sphere must sit over the edge of the detectors, so give-up the rest of the fictional peak processing and show a warning.

However, this isn't working for WISH. It seems that the detector locations found are much further away from the origin of the peak than one would expect. I'll attach a script to demonstrate this problem.

[Owen Arnold]

Further to above, I wonder if this is a scenario where multiple detectors may contribute to the same QLab region, because the algorithms workings seem sound to me.

[Owen Arnold]

This does still need to be done, but it can wait till 3.2.

[Owen Arnold]

Accidentally tried to close this.

[Owen Arnold]

There are now 4 algorithms that can be used for SC peak integration.

IntegratePeaksMD IntegrateEllipsoids IntegratePeaksUsingClusters (new for 3.2) IntegratePeaksHybrid (new to 3.2)

The latter two work on WISH and can be used for non-spherical peaks.

The original issue still remains, (the report of the false positive), but this is now a critical issue rather than a Mantid blocker. I'm moving it to 3.3.

[Owen Arnold]

We should at least have an answer as to why this is happening, as Pascal has repeatedly asked about it. This is almost definitely an issue of false positives in the warning reports.

[Nick Draper]

Martyn,

Could you talk to Owen and take a look into this?

[Andrei Savici]

IntegratePeaksUsingClusters and IntegratePeaksHybrid are using a different method to check if detectors are on the edge. I found a strange omission in the old method. In IntegratePeaksMD2, the test peaks are constructed on detectors at 1 meter away from the sample. It is not the one causing problems in the test, but it should be fixed. Also, to comment on comment 4, There cannot be 2 detectors contributing to Qlab (but they could for the same Qsample, if appropriate goniometer/wavelengths are used)

[Martyn Gigg]

It looks like the issue for WISH is that the IDF is defined such that there are very small gaps between the tubes. This means that occasionally a point on the integration sphere is generated such that it's traced position falls down one of the gaps and doesn't register a hit and the algorithm then assumes that this means it has fallen of the edge of a detector (which technically it has). It's just that in this case we want the detector to mean the whole bank and not a tube.

[Martyn Gigg]

Fix indentation error in usage example.

Refs #8369

Changeset: b27c77cdc27790183cb8acf600560d9086b0030b

[Martyn Gigg]

Relax contraint for determing in a given Q is on a detector

It now only requires 75% of points tried to hit something, which will allow for small gaps between pixels/tubes in banks not to affect the overall test of whether a pixel found. Refs #8369

Changeset: d8842edc94a80daa2709cdc0ad87c911bb46e889

[Martyn Gigg]

Minor refactoring of method signatures to avoid copies.

Refs #8369

Changeset: c66397b960af075f7428fc930982b25c1c704f7e

[Martyn Gigg]

Reorganise headers into alphabetical order

Also removes unrequired headers. Refs #8369

Changeset: d64eec6ba4bfae6a5a0384e9454cf24b938e70ea

[Martyn Gigg]

Similar problems with gaps are also present on CORRELLI at SNS, which #10254 has been opened to deal with. This ticket was more about investigation and the fixes on #10254 amount to the same as those here so we'll simply drop those from here.

[Owen Arnold]

Agree

[Owen Arnold]

Agreed. Pascal has suggested solving this problem by repeating the ray tracing around around a circle to determine if there are neighbouring tubes, or if this is really an edge. Read - a ray with thickness.

[Owen Arnold]

NOTHING MERGED.

[Stuart Campbell]

This ticket has been transferred to github issue ​9214