Release Notes for Initial Milne Eddington Inversion of HMI Magnetic Field Data (VFISV)

hmi.ME_720s_fd10 ([http://jsoc.stanford.edu/ajax/lookdata.html?ds=hmi.ME_720s_fd10 click here for data]), hmi.ME_720s_fd10_HARP ([http://jsoc.stanford.edu/ajax/lookdata.html?ds=hmi.ME_720s_fd10_HARP click here for data]), and hmi.ME_720s_fd10_HARP_nrt ([http://jsoc.stanford.edu/ajax/lookdata.html?ds=hmi.ME_720s_fd10_HARP_nrt click here for data]).

These three data series contain ME inversion data processed using VFISV ([http://hmi.stanford.edu/doc/magnetic/section4.pdf details]) using the codes implemented on April 13, 2012. There are known issues with these versions of VFISV, but this remains the best ME release version as of the date of implementation. We consider this a beta-test version of the HMI vector field inversion.

• The prime data products are field strength, inclination, and azimuth. Errors and other quantities are included.
• No disambiguation has been performed, so azimuths are in the range 0-180, not 0-360.
• fd10 in the series name is an index of the choice we think best among those we had tested, since [:ME720se15w1332: the last release].

Our efforts to optimize the VFISV for the HMI data pipeline are on going, and we will create the series to fill the data processed by the newer codes when ready. The eventual pipeline code for HMI vector inversions will be different. We will announce the release of the new data through Solar News.

Temporal Coverage for These Three Series

[http://jsoc.stanford.edu/doc/data/hmi/coverage_maps/inversion/ Monthly charts] show the 12-minute time slots for which the ME inversion data processed with the VFISV module are available. The coverage charts also contain the status flags of the HMI observation, thus the quality and availability of input [http://jsoc.stanford.edu/ajax/lookdata.html?ds=hmi.S_720s Stokes] data. See [:ME720se15w1332CoverChart: details] for an explanation of the coverage charts.

[http://sun.stanford.edu/~keiji/HMI/nprcs/invnprcs.html A temporal version] of the coverage chart is avaiable, until we will fully adjust scripts making the formal, informative monthly chart.

• hmi.ME720s_fd10 - Full Disk

  • Synoptic data, on twice per day basis at 7 and 19UT daily, are stored in this series. The latest version of VFISV is being used to extend this series using the latest definitive Stokes data, on the 2 per day basis, usually about a week after the observation. Data for scattered additional time intervals are also available. Checking the version of VFISV codes is strongly recommended.

• hmi.ME720s_fd10_HARP - HMI Active Region Patches

  • The latest version of VFISV has been used to generate the ME data stored in this series. Many of the active regions were recognized automatically by the [:HARPDataSeries:HARP module], while some were defined manually. This series includes 12-minute (720s) cadence data for several of the ARs that created X-class flares in 2011, including the one occurring on 2011 February 15 (at NOAA AR11158).

• hmi.ME720s_fd10_HARP_nrt - Near-Real-Time HMI Active Region Patches

  • The latest version of VFISV has been used to generate the ME data stored in this series. The input Stokes data and the HARP info. are of preliminary though the time lag between observation and ME process is minimized. Current setting is to start, at 7PM Pacific Time, one-per-hour run for the same day.

Versions of VFISV in the FITS files exported at JSOC database

• The VFISV version is stored as a string value in the keyword INVCODEV.

  • The ME data for the series above were processed with the latest version, labeled with a date, "Apr 13, 2012".

  • In IDL, for example, if the exported data is read with [http://hmi.stanford.edu/doc/magnetic/fitsio.pdf fitsio_read_image()] then, the version keyword can be printed with sxpar():

  •     IDL> img=fitsio_read_image('img.fits',header)
        IDL> print, sxpar(header,'INVCODEV') 

• Major differences between 2011-October-14 (or 2011-November-27) version and the current FD10 version

  • VFISV core part
    1. A regularization term, whose purpose is to penalize high values of the parameter eta_0, was added to the merit function in order to tackle a double-minima problem.
    2. The selection process that decides when the inversion of a given pixel needs to be re-started with a different (randomized or tailored) initial guess atmosphere, has been implemented and fine-tuned.
    3. We have settled on a convergence criterion based on the value of the Levenberg-Marquardt Lambda parameter. The algorithm exits the iteration loop when Lambda becomes too large and no re-start of the inversion is enforced for any other reason. The average number of iterations per pixel is now about 30.
  • Misc., other part
    1. Parallelism, by means of MPI, has been modified and optimized to allocate the tasks to each Process Element as evenly as possible.