A third-party map-making software for the WMAP raw time-order data (1003.1073)

[Hao Liu]

Here is the software we used in the works 1003.1073 and
0907.2731. In these two works, we have obtained a significantly
different result to the WMAP release. According to the importance of a
reliable CMB anisotropy map, we decide to publish our software and invite
the readers to check our result.

Although it's a complex job to make CMB maps from TOD, this software is
easy to use. All you have to do is to download it as well as necessary
libraries and TOD (as listed below), compile a fortran plug-in
(recommended but optional, the program can also work without it), set a
few lines of parameters (mainly the directories), and type the command
and let it run. All will be finished in 1-7 hours depending on the band and
whether the fortran plug-in works.

We have also designed a convenient switch to help the readers to repeat
our fig2. More details can be found in "readme.txt". There are also many
notes in the source code to help readers understand the pipeline and to
do some tests they want. Crucial places in the code are marked by a line
of semicolons.



You can download the software right here: release_v1.zip or from the Tsinghua Center for Astrophysics:
http://dpc.aire.org.cn/data/wmap/090727 ... e_code/v1/

Here are the required libraries:
HEALPix: http://healpix.jpl.nasa.gov/ or http://sourceforge.net/projects/healpix/ for direct download
Astrolib: required by HEALPix, http://idlastro.gsfc.nasa.gov/ftp/astron.dir.tar.gz
WMAP_IDL_Pro: http://lambda.gsfc.nasa.gov/data/map/dr ... v40.tar.gz
For all libs, only the IDL pro are needed.

The required data files are the WMAP3 calibrated, filtered TOD (about 94GB):
http://lambda.gsfc.nasa.gov/product/map ... al_get.cfm

Responses and questions are wellcome, please write to me:
[Log in to view email]

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

Figure 2 of 1003.1073, the left is produced purely from dipole
signals, whereas the right is the WMAP quadrupole. The close similarity
between them indicates that the WMAP result is artificial.


Here are the CMB power spectrum derived by our code with the WMAP
offset setting. They definitely ensure that our code is reliable
The low-l:

The high-l:

[Boud Roukema]

Astrolib: required by HEALPix, http://idlastro.gsfc.nasa.gov/ftp/astron.dir.tar.gz
WMAP_IDL_Pro: http://lambda.gsfc.nasa.gov/data/map/dr ... v40.tar.gz
For all libs, only the IDL pro are needed.

I haven't tried this yet, but here's a link to a recent point of view on the status of GDL by astronomers:
http://aramis.obspm.fr/~coulais/IDL_et_ ... _Draft.pdf

GDL is free (libre) software; IDL is non-free. GDL should be more useful to use in the long term, for the obvious reasons.

[Hao Liu]

Astrolib: required by HEALPix, http://idlastro.gsfc.nasa.gov/ftp/astron.dir.tar.gz
WMAP_IDL_Pro: http://lambda.gsfc.nasa.gov/data/map/dr ... v40.tar.gz
For all libs, only the IDL pro are needed.

I haven't tried this yet, but here's a link to a recent point of view on the status of GDL by astronomers:
http://aramis.obspm.fr/~coulais/IDL_et_ ... _Draft.pdf

GDL is free (libre) software; IDL is non-free. GDL should be more useful to use in the long term, for the obvious reasons.

Thank you for your suggestion. I haven't used GDL yet, but by reading
the document, I noticed this: "Large parts of the Astron Library are
working well in GDL, including the FITS part". So I think my code may
also work in GDL. I'll try it as soon as possible.

Update 2010-03-13:
Now it is verified that our code can also work under GDL, the free IDL- like
GNU software (http://gnudatalanguage.sourceforge.net/) with
only a few modifications. Here are known necessary modifications:

1, You can not use the fortran plug-in under GDL, so you must set so_file=0
2, The time cost is about 9 times as IDL
3, You may have to change the program names to some simple names,
such as "mmm.pro", "nnn.pro" to make it work (a strange problem,
maybe it's just because of a GDL installation issue)
4, The "write_fits_map" subroutine doesn't work properly, you need to
manually replace all occurrence of "write_fits_map" by you custom
codes, e.g., something like "writeu", and then use other tools to view
your custom output file.

[Boud Roukema]

Presumably mapmaking.pro should still work correctly for lower resolution,
e.g. res_param = 8, and this should drop the memory requirements by
a factor of res_param^2. Is this correct?

[Hao Liu]

Presumably mapmaking.pro should still work correctly for lower resolution,
e.g. res_param = 8, and this should drop the memory requirements by
a factor of res_param^2. Is this correct?

No, because the total number of observations will not change. Although
the produced map will be smaller, the number of observations in each
pixel will be much higher.

[Boud Roukema]

Presumably mapmaking.pro should still work correctly for lower resolution,
e.g. res_param = 8, and this should drop the memory requirements by
a factor of res_param^2. Is this correct?

No, because the total number of observations will not change. Although
the produced map will be smaller, the number of observations in each
pixel will be much higher.

In fact, literally setting res_param=8 in your code not only fails to drop the memory requirements, but it also gives a wrong pixelisation if nothing else is changed. Setting both res_param=8 and nside=256 simultaneously would presumably give a correct map.

[Boud Roukema]

Now it is verified that our code can also work under GDL, the free IDL- like GNU software (http://gnudatalanguage.sourceforge.net/) with only a few modifications.

I agree. IDL/GDL seems easy enough to learn and GDL is by definition easy to hack. I found an eight-year-old bug in the IDL Astro library (presumably a violation of FITS standards: the END statement in the header of a binary table is removed!), and a few missing features in GDL-0.9~rc3 that are easy to work around. I didn't find any bugs in GDL, nor in your mapmaking package.

Here's a cylindrical equal-area/interrupted Collignon projection, a.k.a. the HEALPix H=4, K=3 spherical projection of iteration 80, 'm' mode, of

Code: Select all

run_mapmaking,'v1',3,1, /dipole

, i.e. of 1 year of V1 quaternion data run with about 3.5Gb RAM available, using the patch/script set LLmapmaking_GDLpatches-0.0.2 described below and GDL-0.9~rc3. The map orientation is conventional, i.e. GC at centre, longitude increasing towards the left. The scale is roughly from -10 (black) to +10 (white) [tex]\mu K[/tex]. As you state in Liu & Li 1003.1073, this map does not contain CMB observations and it is not an l=2 spherical harmonic [tex]a_{lm} Y_{lm}[/tex] of an observational CMB map. It only contains the effect of the attitude/pointing error ("evil attitude") on the Doppler dipole of the spacecraft's velocity history over one year: Equations (1) and (2) of 1003.1073, which contain no cosmology information. I have not made a formal comparison with the "official" WMAP quadrupole, i.e. Fig 2 right of 1003.1073, but by inspection, my quaternion dipole map looks extremely similar in position and amplitude to Fig 2 left (as it should, assuming that my customisation for GDL didn't make any significant modifications of your code), and also extremely similar to Fig 2 right - the CMB quadrupole estimated from the WMAP team official maps.



The 2D FITS file can be downloaded and displayed in any FITS image viewer e.g. ds9 (packet in debian: saods9). To convert a 1D HEALPix FITS binary array to a 2D FITS image, e.g. WMAP ILC 3yr 2D FITS map, you can use HPXcvt in wcslib 4.3 or later. HPXcvt.c requires the cfitsio and C math libraries, i.e. it should be compilable with e.g.

Code: Select all

gcc -O -o HPXcvt HPXcvt.c -lcfitsio -lm

LLmapmaking_GDLpatches_0.0.2.tbz is a package of patches and scripts for using Liu & Li's mapmaking IDL(TM) routines in GDL (GNU Data Language)

• the "fortran" loop map_gen is provided by a C++ equivalent, which is linked using GDL's LINKIMAGE procedure [solves problem in points 1 and 2 of Liu (2010-03-13)]
• an optional parameter "low_memory_Gb" is introduced to allow repeated disk reads as a substitute for large amounts of RAM [avoids requirement to have e.g. minimum 6GB RAM for W band, point 4 of LL (2010-03-12), at the cost of being slow]
• writing output FITS files to disk works with a few minor corrections to IDL Astro routines in order to correct a FITS-standard bug in fxbcreate.pro and to work around missing features in GDL-0.9~rc3 [solves problem in point 4 of Liu (2010-03-13)]
• user's directory names are set in an environment variable startup script instead of in mapmaking.pro
• read README and INSTALL!
• a minor hack for avoiding divisions by zero when using less-than-full-year data sets is introduced to mapmaking.pro (search "(nobs or ones)" in mapmaking.pro after patching)
• point 3 of Liu (2010-03-13) on routine names: I did not have this problem with GDL-0.9~rc3 (which is now a moderately old version)
• tested on debian/lenny and ubuntu/karmic x86_64 GNU/Linux systems

References:

• Liu, 2010-03-13: http://cosmocoffee.info/viewtopic.php?p=4525#4525
• LL, 2010-03-12: http://dpc.aire.org.cn/data/wmap/090727 ... Readme.txt