Rogers replied:
"
Just beware the namespace clash - there is already a
MAUS::CovarianceMatrix defined in Optics. This is potentially an object
we will want in Globals, so you may want to define it as a
ReconstructionCovarianceMatrix (a bit verbose) or similar.

I would use a TSymMatrix or whatever the ROOT object is rather than a 2D
vector like this. Ian should also comment.

For the processor, best would be nested arrays. I think that a
ValueArrayProcessor of ValueArrayProcessors will work, but obviously let
me know if there is a problem.
"

The ROOT question in object is a TMatrixDSym (which is a typedef of a TMatrixTSym<double>). This has helpful GetMatrixArray and SetMatrixArray functions, which provides the data in a double* array. It would be sensible to document the order of the elements, so that some-one accessing the JSON can still use the data.

I'm assuming that ValueArrayProcessor will deal with a double*, and that the array size is only that of the unique elements. To quote Under Siege 2, "Assumption is the mother of all f**k-ups", but I'm making Ed's problem as of right... now!

Well, the JsonCppProcessors are mine so if you need a feature or whatever then complain to me and I will see what I can do.

Yes, use of MAUS::CovarianceMatrix in code (i.e. aside from DataStructure/ classes) is encouraged. It already has double array initialization, so adding a constructor that accepts a TMatrixDSym is trivial. I'll add that to my dev branch, but if you need it before I merge just let me know.<br>
<br>
Should the DataStructure/Global classes also use covariance matrices for uncertainties? Right now in my global recon code I take the uncertainty arrays, square them, and make a diagonal matrix out of them anyway.

Yes, you can close it.

My matrices are of the type TMatrixD, so I used GetMatrixArray() like Ian suggested. I then converted this array into a std::vector so that I could use ValueArrayProcessor<double>.

Thanks!