January 2010

 -- Focusing on the over-skysubtraction issue
        e.g. 3702A in Order 97 (and 96 but not as bad)
         -- Systematically 0.5 electrons per pix too subtracted
         -- Focus on 237, 1756  (3860A in order 93) , image B0142

hires = hires_ar('hiresstrct.fits')
hires.setup =1
hires.obj_id =1
hires.arc_img ='Arcs/Arc_B0140_I.fits'
hires.obj_fil ='Extract/Obj_0142B.fits'
hires_extract, hires, 1, 1, [1], [1], ordrs=[93]
hires_extract, hires, 1, 1, [1], [1], ordrs=[93], /TRY_IVAR

  ;; Image is already over-subtracted before x_extechopt
      -- Scattered light? Must be
      -- It is set to a constant of 5 electrons (about right)

  ;; Chi2 appears minimized
      -- Offsetting by a constant does not help
      -- Appears to be a big chi2 penalty in the highest flux, single pixel
      -- No obvious wavelength dependence
      -- A scalar change to ivar does nothing

  ;; Key problem -- Negative data has poorer variance!
      -- Trying
          readno = 3.0
          new_ivar = 1.0/((img>0.) + 2*readno^2)
     hires_extract, hires, 1, 1, [1], [1], ordrs=[93], /TRY_IVAR
          :: No luck
      -- Trying
          tst_ivar[inorder[ys]] = 1.0/( (img[inorder[ys]]-yfit - 7.)>0. $
            + 7. + 2*readno^2)
          :: Success!!
          :: Key = Flat variance across the zero point
          :: Not too sensitive to the zero-point kludge

  ;; Final product mainly unchanged, however...
      -- The modeling gives a non-zero object which means
        a non-zero sky.  This is going to be very hard to avoid
      -- I suspect this is driven by the scattered light issue
          :: Trying no scattered light sub == profile will be wrong