Agenda for CfA meeting
THURSDAY 9:00 -- 12:00 (EDT)
- Review status and next actions for Paper I.
- Review of Recent Progress
- New calculations of CHIANTI emissivities with "seed" saved; completed!
- New "fake" intensities, should we account for Poisson noise?
ints -> ints + N(0,err)?; completed!
- New MHD intensities; in progress!
- New observed intensities, new lines have been added; in progress!
- Review methodology of Paper I.
- Partial list of questions:
- Do we have the best algorithm for identifying the "best" atomic data among many realizations?
- How do we handle the "save the seed" problem? Is there a way now
to work back from the "best index" that comes out of FullBayes to the deviations in the
atomic data that produced that emissivity curve?
- How to deliver a useful quantification/summary of the uncertainties. [This is feedback from the SPD Meeting]
- Is there now a better way to construct atomic data deviations? (We had some discussions on how to filter out obviously bad deviations, but not sure if anything came of it). Does the GDZ approach scale to many ions?
- Questions from scientists to understand better the exact Bayesian
techniques being applied.
- There were quite a few techniques that were used during the stats development (Prag/Full,
STAN/Laplacian) -- should figure out which one to settle on.
- (Note that there is a Bayesian tutorial scheduled for Friday afternoon, too.)
- Why are the inferred path lengths so long?
- Review status and next actions for Paper I.
THURSDAY 1:30 -- 5:00 (EDT)
- Review motivation and tasks for the Test Cases (Paper I) for the afternoon.
- Work on three tasks:
a) Test Case 1: Compute 1000 realizations of CHIANTI, pick one and
compute intensities, run algorithm and identify the best-fit atomic
data. Does it match?
b) Test Case 2: repeat using MHD simulation data. This introduces
inhomogeneity into the mix. There is no single density that describes
the plasma, only an intensity-weighted one. Does it still work?
c) Real Data: repeat the analysis that we did at the last meeting
with observed intensities but saving the collision rates (or at least
the seed!).
- Review afternoon's progress.
FRIDAY 9:00 -- 12:00 (EDT)
- Brief update on any progress since yesterday.
- Discussion on O VII/VIII "bridge" paper (Paper II)
- are the maths clear?
- for statisticians: do we know how to go from toy problem with only n_e to this bridge problem with ne|T?
- what are the next steps?
- and how to handle full DEMs (Paper III)
- what complications can we include if we had lots of lines covering lots of temperatures and densities?
FRIDAY 1:30 - END
- Bayesian tutorial by Yang Chen