[fitswcs] FITS WCS Time Paper

[Lucio Chiappetti]

This is the second promised message (and last !)

I will group my comments to V0.93 of the paper in two groups. First the 
issues which might affect inclusion in the standard and IAUFWG 
supervision, and next any other point. I will list both following the 
order of the paper.

---------------------------------------------------------------------
A) STANDARD-RELATED COMMENTS

- section 2 end

   I see no problem to have Appendix A as normative. In fact it
   REPLACES Appendix B of the 3.0 standard (which was NOT part of
   it).

   This is the first of a set of comments leading to the possible
   issue of a version 3.1 of the standard, once the Time Paper is
   approved and incorporated by reference.

- section 3 end
   section 3.5
   section 6

   (on 128-bit floating and its implication on the standard) see
   previous message in reply to Tom McGlynn.

- section 3.1 datetime "type"

   the datetime "type" defined here is an ASCII string representation,
   so it applies MOSTLY to header keywords, which are not "strongly
   typed" but "heuristically typed".

   One might possibly have cases in which a datetime string is stored
   in an ASCII table, or binary table column, but the latter is definitely
   less efficient than one of the floating point representations.

   So datetime is not properly a type (TFORM) but a format (TDISP).
   One could, if not satisfied with the heuristic interpretation,
   define a specific format descriptor (I remember I had a custom H
   format in a plot labelling program to plot times as hh:mm hh:mm:ss or
   hh:mm:ss.ff etc.) specifying things like digits for year and
   fractional part of seconds.

- section 3.1 datetime 5-digit year format

   *) the change from 4 CCYY to 5 signed digits requires a CHANGE to
      the existing FITS standard (4.4.2.1)

   *) question: is the extension to 5 digits part of "true" ISO 8601 ?
      I have no access to the expensive ISO standard, and the info on
      the net is contradictory.
      If it weren't, we should not call it ISO 8601 but ISO 8601 like.

   *) I presume the usage of 5-digit years will be limited to specific
      sub-communities. What are their needs ? Should we limit to 5 digits
      or can have an arbitrary number of digits between sign and first
      dash?

- section 5 convention timescale;reference position

   The boundary between something which is not mandatory and wishful
   thinking is thin.

   Either the "strong recommendation" is somewhat enforced in terms of
   must/shall/should (2.1 of FITS standard), or it is moved to a plain
   (optional) convention.

- section 5.2.1

   The first sentence shall be emphasized. CTYPEi='TIME' or Tab.2 is
   what discriminates a "time image" to which the paper applies.

   I presume there is one case in which alternate WCSs on the time axis
   will be provided i.e. some of those in 5.5 (I found sometime useful to
   use a (pseudo-image to stack light curves, or periodograms, or Fourier
   transforms in several energy bands ... and these light curves can
   have a "proper" (!) time axis, and an alternate WCS which is phase
   according to some ephemeris

- section 5.3

   6th para "In the context of tables ..." shall be emphasized. It is what
   discriminates a generic table to a one to which the paper applies.

- section 5.4

   Although random groups are deprecated outside of interferometry,
   shouldn't we add a column concerning their kwds to table 5 (as
   they are present in table C.1 of the FITS standard) ?

- table 5

   this table is candidate for incorporation in a table of reserved
   keywords in the FITS standard, like Tab. 22.

   I note that DATE-OBS, DATE-AVG, MJD-OBS and MJD-AVG already
   appear in such Tab. 22 also in MJDOBn etc. form (this was due
   to spectra radial velocities, 8.4.1). Should they appear in the
   appropriate columns in tab. 5 too ?

   Note d says "optional ; only for use with coordinate type PHASE"
   I would drop the word "optional". Clearly these are APPLICABLE
   only with PHASE, but not all the other keywords are compulsory
   in all cases !

---------------------------------------------------------------------
B) OTHER COMMENTS

  - section 2, 3rd item list

    if one mentions "folded or stacked light curves" (ref. 5.5) one
    should add also "Fourier transforms" or "periodograms"

  - section 2 "proper" time

    (already discussed) I feel that "proper" time is sufficiently
    different from "proper time", but if one wants one might specify
    in parenthesis  (not "proper time" !)

  - section 3.2

    "and can be used when their precision suffices" ->
    "and \textit{should} be used when their precision suffices" ->

  - section 3.4

    see previous messages on clarifications on usage

  - section 4

    An URL http://fits.gsfc.nasa.gov/registry/greenbank.html for
    the Green Bank convention won't hurt (FORMALLY it is under
    review, although Bill Pence who wrote it seems to consider it not
    a carved-in-stone reference

  - also section 4 last para

    see previous mail for "high precision floating valued"
    Irrespective of those comments, one should clarify here that one
    is talking of header keywords, therefore "high precision floating
    valued" are "floating valued" EXPRESSED AS ASCII STRINGs (i.e. 4.2.4
    of FITS standard)

  - section 4.2.3 typos

    MJDREFI MJDREFF JDREFI JDREFF
    integer-valued-valued ? integerfloating-valued ?

    (incidentally what is the reason *in this case only* to split
    integer and fractional part in two kwds ?)

  - section 4.6

    see previous mail. It is a convention as said in section 4.5 !

  - section 5

    here you quote the WCS distortion paper as "Paper IV" while in
    the introduction you have renounced to such number (it looks
    likely the Time Paper will be paper IV !)

  - section 5.5

    I am of course pleased of seeing folded light curves etc.
    mentioned. I wonder if one should not add periodograms (e.g.
    chi-square vs period "plots" resulting from folding analysis)
    added, and therefore a CTYPEi etc. of PERIOD in addition to
    FREQUENCY (one could even think of using alternate WCSs for
    them since f=1/P, which allows e.g. to compare Fourier transforms
    with periodograms on the same x-scale)

    I presume that for "unitless phase expressed in turns" you mean
    that instead of using a phase limited to the range 0.0-1.0 (as
    usual for folded light curves, sometimes duplicated 0.0-2.0 for
    plotting purposes), you consider a "normal" light curve in which
    an alternate axis for t(i) will be phi(i) in range 0-n (or n-m)

    Of course any folded data need an ephemeris (T0 and period).
    Really we do not want to consider cases with a P-dot ?

- example FITS headers

   *) table 6, how does one infer EXPLICITLY (see also comment to 5.2.1)
      that this is a "time image" ? By the presence of TIMESYS ?  By the
      fact CTYPE3 is one of Tab.2 ?

   *) table 7

      There are multiple WCSs A-F for TTYPE1, and C and G for TTYPE20.

      I am perplexed by the fact WCS C has just TCTY1C and no CRVAL,
      CRPIX etc.  What does it mean ? it is just an alternate "label"
      for TT ?

      And why instead the second WCS of TTYPE20 is labelled C ?
      Which relations it has with WCS C of TTYPE1 ?

-- 
------------------------------------------------------------------------
Lucio Chiappetti - INAF/IASF - via Bassini 15 - I-20133 Milano (Italy)
For more info : http://www.iasf-milano.inaf.it/~lucio/personal.html