What resolution and S/N does APOGEE/APOGEE-2 need?

[andycasey]

Some immediate tasks so that I don't lose track of this:

• [ ] Get measured LSF for all fibres
• [ ] Are there repeat spectra of the same star in two opposing fibres (i.e., with very different resolutions)?
• [ ] Write and test code to convolve from fibre-LSF to a specified R, and re-sample with flux conservation!
• [ ] Convolve DR12 individual visit spectra from the labelled set to common R, then re-stack.
• [ ] Train at certain grid points of R:
  • [ ] 20,000
  • [ ] 15,000
  • [ ] 10,000
  • [ ] 5,000
  • [ ] 2,500
  • [ ] 1,000
  • [ ] 500
  • [ ] 50
• [ ] Do test time on individual R tests for all individual visit spectra in the labelled set (i.e., we are not doing the test step for all stacked spectra at all R)

[mkness]

Once you do the tests at lower resolution it would be excellent to share this with the collaboration for R = 20,000. Jon Holtzman asked for exactly this test, as the resolution of the South may be slightly lower than that of the North so they want to assess performance impact. Presumably resolution is only dropping a few percent so results around R = 20,000 is what will be relevant for the collaboration.

[andycasey]

Absolutely!

I think to make this test realistic we will have to solve wavelength censoring first (otherwise we have no idea about how much correlated information is being passed around at low resolution), but the rest of it should be trivial to code up and burn.

[mkness]

J. Holtz said going down to R = 10K for tests would be very helpful to see.

[andycasey]

Did you tell him R = 50 was in the list? 😃

[davidwhogg]

I have a tiny issue with how we "bin down". This either has to be part of this issue or a separate, related issue. The point is about having a well-defined LSF at the end of the day, and not strongly correlating the noise...