Reduced order elasticity in MCMC estimation

[brendanjmeade]

@aseyboldt

Follow-up to our Monday discussion:

In the current MCMC estimation, the elastic contribution to station deformation for each mesh is done directly with a tde -> station vel operator. There is an acceleration of this via a truncated SVD, but that approach only reduces the dimensionality a bit and the SVD mesh modes display expected patchiness.

I think it's worth considering replacing this with the modes -> station vel operator that we already have (and use for the SQP solution). This has the advantage of being small, implicitly smooth, and already built! This would reduce the size of the matrix-vector multiplies and be implicitly smooth even with a low number of modes.

Thoughts?