Proposal NeXus Errors

The current definition for the NXdata group (which, incidentally, is misleading on the wiki since it implies that the data SDS should be called “data”) defines an error SDS with the name “errors” (in this case, the name is fixed). This is a workable solution when there is only one signal. However, the original plan for NXdata groups (again, not particularly clear on the wiki) was that they could contain more than one data SDS. The primary one has the attribute “signal=1”, but secondary data sets with “signal=2”, etc., were to be allowed. This is of particular importance to the synchrotron community, who frequently measure, for example, fluorescence counts in addition to the main detector counts. In principle, each one of these could have their own errors, but there is no way in the current scheme to associate different SDSs with those errors.

Proposal

An extra attribute be defined for each data (or signal) SDS called “errors”, which would be a text string defining the name of the SDS within the same NXdata group to contain the errors SDS. The rank and dimensions of the errors SDS must match the signal SDS. The original method, i.e., associating an SDS with the name “errors” with the primary signal SDS will be allowed in cases where there is no ambiguity, and for reasons of backward compatibility.

Discussion

This is analogous to the use of the “axes” attribute to define the independent axes associated with each signal SDS. However, this attribute will only ever contain one name, since there is no reason to associate more than one error array with a single signal array.

Here is an example NXdata group using the proposed new attribute:

<NXdata name="data">
    <primary_data signal="1" axes="x1:y1" errors="e1">...</primary_data>
    <x>...</x>
    <y>...</y>
    <e1>...</e1>
    <secondary_data signal="2" axes="x:y" errors="e2">...</secondary_data>
    <e2>...</e2>
</NXdata>

I propose to present this for discussion and a possible vote at the upcoming NIAC meeting at the SNS on October 7-8.

[Ray] (Ray_Osborn.html “wikilink”) 16:29, 29 September 2010 (UTC)

Conclusion: 01/2015

At NIAC 2014, the NIAC ratified a scheme for describing uncertainties. See 2014_axes_and_uncertainties and, in a short while, the NeXus manual.