Do I need to worry about NaNs and INF in IndirectReduction?

I’ve been running some quiet counts (no beam) on IRIS, and if I try to reduce the data on more recent runs such as 94133, the output has lots of NaN s and Inf s. Is this due to the normalisation against the incident beam for a low number of ueV?

Older runs with similar conditions seem to be more sensible (to me at least) and return 0. This only seems to be a problem for a low number of counts, but I’m a bit worried that it indicates deeper problems.


Hi Ian,

Thanks for the post. We’d like to try and recreate this but need some additional information.

Can you give us an example of an older run we can compare with 94133?

Which version of Mantid are you using?

Which operating system are you using?

Feel free to e-mail with this information if you’d prefer.



Hi Sarah,
an old run that “works” is 41814. I’ve tried a few versions, but the last stable 6.4 version on windows 10 is where I first discovered it.

Dear Ian,

Could you please email me the nxs files for the two runs? For some reason I cannot view the data for IRIS on the archive.

Kind regards,

Dear Ian,

I tried 94133 with Mantid version 5.1 and I still got the NaN’s and infs. Has anything changed with the way the data is collected or stored?

Kind regards,

Hi Anthony,
We re-installed the mica detectors in July 2017 (det 54-104). I don’t think this should have made a difference, though it is possible that tables were changed around this time. Let me know if you need help understanding these, though you might better off talking to Freddie.

You refer to low microeV. Detectors are divided by the monitor so if the latter has zeros it could lead to,problems.

Yes, I did think of that, but since the early data seemed to report zeros gracefully, rather than undefined, I was wondering if there had been any changes made, they might pop up in a more inconvenient place.

Just adding a link to the Github issue: Indirect Reduction: Nans and Infs · Issue #34784 · mantidproject/mantid · GitHub