Astronomical Image Processing System
Mac OSX 31DEC19 version changed
The next Mac OS version (10.15) will no longer run 32-bit
applications. Since the Intel compiler we have been using
creates 32-bit load modules, we have been forced to change to
gfortran version 6.3. A MNJ will download all of the recompiled
load modules and libraries. You will need to invoke the bash
script fix_aips_elcap.sh to link up the 3 new gfortran
run-time libraries provided with AIPS. Please contact the
helpdesk or firstname.lastname@example.org if there are problems. Note that the
31DEC18 and earlier versions of AIPS will not run on Mac OS
EOP web site has moved
A new web location for the Goddard EOP data files has appeared.
The procedure VLBAEOPS in VLBAUTIL has been updated in both the
31DEC19 and 31DEC18 versions. The old web site at
gemini.gsfc.nasa.gov is gone. We thank the folks at Goddard
for providing these files through a back channel to support
NRAO's correlators during the outage and for creating this new
Solaris binary support has ended
The support for the Solaris operating system in Socorro has
been terminated due to the failure of shire in a power outage.
The binary installations of Solaris (and of Mac PPC) are not
provided in the 31DEC18 and later releases. They are available
in the 31DEC17 and earlier releases. If you have this sort of
computer, you may still take AIPS in text form and compile it
locally. The code was corrected a while back to survive modern
Fortran compilers and the Solaris scripts for compiling AIPS
have been used for a great many years.
Warning about modern Linux operating systems
Good news: I tried compiling the
LNX64 binary installation with the latest Intel compiler
(18.0.0, 20170811). Everything ran well on NRAO computers,
but the installation failed in Leiden without even getting to
the NaN issue. The binary installation has been rebuilt,
therefore, with gcc and gfortran (version 6.4). That
installation works normally in Leiden. Note
that gfortran versions 7.x do not work with dynamic memory
pseudo-AP and put serious limits on fixed memory pseudo-AP
versions. The following is left, for now, to document
the Intel compiler issue.
Computers that ran AIPS perfectly well with the binary
installation have suddenly run into mysterious issues after an
operating system upgrade. These issues are discussed below.
Both Leiden and Jive report that the issues go away if they
compile from source using gcc and gfortran. I (Eric) use gcc
version 6.3 on my machine and some of the gcc versions 5.x also
worked fine (latest here was 5.4). The problem is being
One of our users had his computer "upgraded" to the Fedora OS
version 25. Instantly, multiple AIPS tasks started producing
NaNs scattered amongst the correct values. This happened in
multiple versions of AIPS using binaries downloaded from NRAO
over a long span of time. The results do depend on the inputs,
but if the inputs are not changed, the results even from
multiple versions of AIPS remain the same. We will attempt to
investigate this further, but, for the moment, we recommend
strongly that you do not install this version of Fedora. Note
that AIPS worked fine under previous versions, 24 and less.
And the contagion is spredding: A second site has now
reported the same problems on both Fedora 25 and 26. The
RedHat 7 system in Leiden has suddenly developed the same
issues after having worked well for quite some time. And now
Jive has reported NaN issues with Arch Linux 4.12.12-1-ARCH
kkernel 4.12.12) and Debian 9. Yet another site has reported
this failure with Scientific Linux 7.4, while SL6 still works
Further study in Leiden has found that computers with
varied but rather modern hardware fail under both Fedora and
RedHat 7, while somewhat older computers run AIPS perfectly
well under both. A local compilation with gfortran 6.? was
done on one of the failing Fedora machines and the resulting
AIPS worked properly. I am concerned that the Intel compiler
we use for 64-bit Linux is too old; it dates from January
2009. We will be looking into something more modern.
AIPS has now been tested with a Fedora 25 virtual machine
and the tasks that failed in Leiden worked fine on the same
data when run in Socorro. Perhaps the only issue was
something wrong with some installations of Fedora 25
including the one in Leiden.
Another user has reported problems with IMAGR under Ubuntu
17.04 that went away when she reverted to Ubuntu 16.04 LTS.
NRAO has instituted a "helpdesk" system
similar to those at other astronomical institutions. AIPS will
participate in this effort which is intended to avoid having
problems fall between the cracks. The helpdesk is accessed via
that the help site will forward you to my.nrao to create a login
if you do not already have one and the my.nrao site will then
offer a tab to reach the helpdesk. The system for the AIPS
"department" will make use of the email@example.com list manager
system that we have used for years. Thus, unlike the rest of the
NRAO helpdesk, messages to the AIPS portion will be made public in
the daip archive. This archive is accessible via the web and is
well known to Google among other things. Please switch to the new
helpdesk to allow more systematic handling of your AIPS queries
and bug reports.
The development version 31DEC19 is
available, the previous development version 31DEC18 is "frozen"
(except for patches), and the old frozen version 31DEC17 may not
be available for installation much longer. (A MNJ on 31DEC17, if
it has not been run since December 12, 2018, might still be useful.)
The new cvs form of the Midnight Job has been very easy to use and
effective in keeping numerous sites up to date. In 2019, the MNJ
will work sometimes on 31DEC18 if there has been a released patch.
The 31DEC19 and 31DEC18 releases are also available as binary
installations (and MNJs). The 31DEC16 and later MNJs no longer
require cvs for binary installations, just rsync.
TeK and message server problems
Users have encountered problems getting
the TeK and message servers to work in binary installations.
These servers are programs which run in an xterm and that xterm
must come up with some knowledge of AIPS libraries. This requires
users to modify their login scripts (e.g. .cshrc file for c shell,
.bashrc file for bash). See
Linux compiler troubles
There appear to be problems with many of the versions of the
gcc/g77/gfortran compiler suites. GNU compiler versions 2.* and
3.* are now obsolete and the performance under g77 left much to be
desired. The 31DEC15 version of AIPS received attention so that
it may be compiled under gfortran/gcc. Versions 4.4.7, 4.8.3,
4.9.2, and 5.1.0 and 6.3 have been tested and appear to work.
Versions 4.x, 7.x, and 8.x require changes to make the pseudo-AP
fixed, rather than dynamic, memory and to limit all COMMONs to
less than 2 Gbytes. See the 31DEC18 release instructions for the
specifics of these changes. Versions
5.1.0, 5.3.0, and 6.3 also appear to work and no longer have the
issues with the pseudo-AP. The LNX64 binary version now uses
gfortran 6.3. gfortran versions 7.x and 8.3 all fail to
implement the dynamic memory pseudo-AP code as a COMMON.
We have therefore re-wriiten the pseudo-AP to be used as a variable
passed down from above. This does work with gfortran 7.x and 8.x
which otherwise compile AIPS correctly.
Also, see the Users'
FAQ: answers to Frequently Asked Questions, the
AIPS Managers' FAQ:
answers to AIPS Managers (and installers) Frequent Questions; and our
page on who is in the AIPS
group. The AIPS Manager FAQ page is
especially helpful if you are having trouble getting the AIPS TV to
run properly or are encountering other general difficulties or wish to
configure new machines and disks in your AIPS environment.
is the new primary development version of AIPS. It is ready
for installation. The 6-month progress report on 31DEC19 is
available as 31DEC19
AIPSLetter dated 30 June 2018 (PS, 869314 bytes).
It is also available gzipped,
(402257 bytes) and PDF,
is the previous development version of AIPS. It is ready for
installation. The final report on 31DEC18 is available as 31DEC18
AIPSLetter dated 31 December 2018 (PS, 8022909 bytes).
It is also available gzipped,
(1008277 bytes) and PDF,
(244403 bytes). The 6-month progress report on 31DEC18 is available as 31DEC18
AIPSLetter dated 30 June 2018 (PS, 867221 bytes).
It is also available gzipped,
(401469 bytes) and PDF,
is the now obsolete version of AIPS. It is available for
installation for a while. The final report on 31DEC17 is available as 31DEC17
AIPSLetter dated 31 December 2017 (PS, 904027 bytes).
It is also available gzipped,
(420144 bytes) and PDF,
(159226 bytes). The 6-month progress report on 31DEC17 is available as 31DEC17
AIPSLetter dated 30 June 2017 (PS, 902569 bytes).
It is also available gzipped,
(419491 bytes) and PDF,
was once the development version of AIPS. It was updated
occasionally in 2017 for patches. It is no longer available for
installation. The final report on 31DEC16 is available as 31DEC16
AIPSLetter dated 31 December 2016 (PS, 957655 bytes).
It is also available gzipped,
(431102 bytes) and PDF,
(185628 bytes). The 6-month progress report on 31DEC16 is available as 31DEC16
AIPSLetter dated 30 June 2016 (PS, 907865 bytes).
It is also available gzipped,
(414726 bytes) and PDF,
- 31DEC06 and later releases contain a port to
the MacIntosh OS/X system running on Intel cpus. This includes
a full binary installation produced with Intel's compiler. An
AIPSMark of 150 was obtained on our iMac which was used to
support this "new" architecture. The iMac has been replaced
with a more modern Mac mini.
- Since the end of January 2009, Linux systems
running on 64-bit hardware with 64-bit operating systems have
the choice of using LNX64 rather than LINUX as their AIPS
architecture. The binary version for LNX64 is compiled with a
a 64-bit gfortran compiler and runs somewhat faster than the
LINUX version run on the same machines.
- From September 1, 2005 through October, 2017,
the binary versions for Linux are compiled with the Intel
compiler version 9.0. Those load modules developed oddities on
newer computers, especially in Europe. We have switched to
gfortran 6.3 which has performance about equal to the Intel load
modules (on those computers that ran them properly).
- If you own a MacIntosh OS X computer, you
may now run AIPS. All available AIPS versions contain the
MACINT port. Note that the somewhat expensive Intel compiler
produces code which runs at about the same speed as modern
gfortran compilers. However, the Intel compiler allows the load
modules to be built to support older Mac OS versions back to
10.7. Read also the somewhat obsolete Mac
installation instructions which are supplements to the A Guide to the Install Wizard
- When you get
the new install wizard; we have a good on-line guide that will help you
through it. Read it!
- The 15OCT99, 31DEC00, 31DEC01, 31DEC02, 31DEC03 31DEC04 31DEC05 31DEC06 31DEC07 31DEC08 31DEC09 31DEC10 31DEC11 31DEC12 31DEC13 31DEC14 releases of AIPS
are no longer available; the number of bug fixes and differences
between them and the
31DEC17 version have become too
great. The continued significant demand for a frozen, even if
obsolete, version was the main reason we decided to resume
TWO VLBI TUTORIALS
Two extensive tutorials on VLBI data reduction in AIPS have been
prepared, complete with data sets, introductory material about AIPS,
and detailed instructions. They are simple VLBA project including
spectral-line VLBA project plus astrometry. Users new to, or
rusty in, VLBI data reduction are encouraged to try these tutorials.
Appendix C and Chapter 9 of the AIPS CookBook
are also recommended.
The VLBA pipeline procedure has been improved and may be useful to you.
In addition, a pipeline for reducing VLBA data from secondary
phase-stopping positions has been written. See VLBARUN information for details.
There are sometimes considerations which users of older versions of
AIPS must take into account due to changes in the latest version. In
addition, there are sometimes bugs introduced into the latest version
which then get repaired in that version. Such matters may affect
those users who keep their AIPS up to date. These bugs often do not
get reported in the AIPSLetter since they were temporary but might be
of concern for diligent users. They are reported in
but those reports are sometimes fragmentary and confusing. We have
begun a new web page for problems of this
sort. It is hoped that we will keep it up to date.
You really should be getting 31DEC18 or 31DEC19 instead of patching an older version!
With previous releases, we did not change the
frozen release tar ball; user sites were required to download the text
files and do any compilations themselves. But, beginning with
31DEC04, we provide binary releases for which user sites are not
required to have compilers. Therefore, we have reversed our policy.
These patches have been applied to the official code of 31DEC17 and
may be downloaded using a 31DEC17 "Midnight Job". Any tarball taken
after the change date listed below will already have the patches
applied. Of course, you may still download and compile the files as
There are three large and fourteen normal patches
for 31DEC18 as yet.
There are five large and sixteen normal patches
for 31DEC17 as yet.
VBCAL failed to write the output data set
TYCOP and other tasks had old number of columns for
the SY table. 2019-02-13
- Model computation had problems when computing multiple CC
- Conversion of linear polarization to Stokes had a wrong sign.
TYSMO applied clipping and flags inconsistently.
FILIT had an infinite loop displaying the image
facets and did not handle the roam function properly
KNTR and other tasks truncated one of the
"brightness" strings. 2019-04-03
OOSUB did not do channel-dependent division
SPIXR did not handle sub-images with an
FQ axis correctly. 2019-05-03
RFLAG did not handle missing data correctly,
making too many flags. 2019-06-14
KNTR computed the number of planes to plot poorly.
BPWAY, RLCOR did not change uvw for
BIF greater than one. 2019-06-14
TAPLT tested for no data backwards and so did
not do good plots. 2019-06-21
BPASS messed up some of the standard source
spectral indices. 2019-07-12
CLIP messed up reporting its results with linear or
other non-standard polarization data. 2019-07-26
UVFLG did not use the correct antenna data when
looping over subarrays for shadowing. 2019-08-22
bpway did not close the table IO while looping over
There are fifteen 31DEC16 patches.
- Linear polarization application was done incorrectly.
- Batch processing did not work. 2018-01-16
CALIB would not do on
multi-source data sets 2018-01-16
UVSUB did not do
MODL correctly for
SETJY printed erroneous error bars with the latest
2017 flux scale2018-03-01
UVSUB did not do
weights came out zero.2018-03-14
- Large gridded models did not allocate sufficient memory.
- Antenna file reformatting in place failed on files with large
numbers of IFs. 2018-04-04
FITLD did not correct duplicate FQ IDs properly,
DBAPP had an old version of the data pointers
and failed to copy modern data properly. 2018-04-09
FQUBE wrote many
CG tables when
none were input. 2018-04-23
FITLD did not handle antenna files properly when
DOBAND 3 interpolation flagged too much data when
there were failed bandpass solutions. 2018-04-27
- Model computation made errors in DFT mode when four
polarizations were being computed 2018-06-20
UVSUB did not do
MODL correctly with
compressed data 2018-06-20
DOBAND 3 interpolation still flagged too much data
when there were failed bandpass solutions and the
LNX64 version confused the compiler.
DOPOL true did not handle magic blanks in the
polarization solutions. 2018-07-16
OOSUB did not handle multi-source data correctly.
AIPS did not do
TPUT correctly when
there was an exact alignment of the number of adverb values with
record boundaries. 2018-08-17
RM could not handle more than 5 or 10 spectral
channels due to a very small buffer. 2018-10-03
RLDIF used the wrong position angles for 3C48 and
There are three messy and nine normal 31DEC15 patches so far:
POSSM would not label correlation functions in
ATLOD would not open disk files properly.
CCRES did not honor blanked pixels
DFTPL did not use the correct frequencies when
BIF > 1. 2017-01-31
DBCON scaled uvw wrongly when combining data sets
of different frequencies. 2017-02-24
DFTPL did not address data correctly when the input
had more than one Stokes. 2017-02-28
BPASS did not normalize linear polarization
bandpasses correctly for normalization type 1.
- Automatic spectral index finding from the SU table had problems
when one or more IFs had no flux value.
UVMOD did not handle frequencies correctly with
SPIXR did not label the output images with the
frequency for which they were determined (1 GHz).
PCAL miscomputed the flux at 1 GHz for known
low-frequency sources. 2017-06-06
UVLSF copied tables twice to the output continuum
data set. 2017-06-30
SCIMG messed up the self-cal
step when the number of IFs was not the same as the number of
PCAL left a file open when an IF had no data in
spectral mode, causing failure on the next IF.
GC tables used to allow 200 values of gain curve,
so restored this limit used by some correlators.
PBEAM scaled Stokes I data incorrectly when adding
right and left data files. 2016-01-12
DBCON re-instated flagged table rows.
DTSUM did not handle new internal format correctly.
PCAL typo caused errors in antennas used with the
new ANTENNA1, ANTENNA2 format. 2016-04-15
PCAL lost Faraday Rotation calibration when doing
SPECTRAL solutions. 2016-04-29
IMAGR found automatic boxes in only part of the
IMSIZE(2) < IMSIZE(1)
OOP editing had trouble with source number zero
sometimes found in tables. 2016-05-19
UVFIX used the actual observing frequency rather
than the one in the header to scale UVW.
OOSUB and other model subtraction/division could
get the scaling between frequency channels wrong.
UVLOD, FITLD did not find the subarray value
accurately enough. 2016-08-25
SPLIT did not write the correct header frequency
when averaging IFs with BIF greater than 1.
UVLOD did not test data
flagging correctly when writing and reading UV tables in
compressed format. 2016-08-31
With the advent of the binary versions of AIPS, it has become more
difficult to compile special local tasks. If one wants to do this and
also take advantage of the improved performance of the NRAO-purchased
compilers (Intel for LINUX and MACINT; GNU for LNX64), the user must
purchase the compiler(s) for himself. This can be rather expensive.
There are two ways out of this bind. If the local task is of
sufficient interest, you could try to pursuade the AIPS group to
include it in the general AIPS distribution. This requires the code
to be written up to a least some minimal standard. A recent example
ATLOD which translates older Autralia Telescope data
into AIPS. The alternative is to send the code and help file to the
AIPS group and ask them to compile the routines, leaving the results
on the AIPS ftp site. This has now been done for 4 tasks:
ORFIT in a CfA directory and
CALIV in a WSRT directory.
Both of these site directories are located in a
directory at the main AIPS public ftp site.
The load modules are stored under the name
.EXE.arch to identify the operating
system (arch) to which the load module corresponds. To
install them in your system, copy the help file
$TST/HELP/ and copy
the appropriate load module to
that one has to rename the load module dropping the architecture
string. For LNX64 machines two load modules are provided, one
compiled with the gnu compiler and one with the intel compiler. In
general, for a binary installation, you will want the intel version.
... will be of interest to you if you want the latest
31DEC16 versions updated nightly,
weekly, or occasionally. It can do either text or binary updates
depending on which sort you did with install.pl.
The WENSS/WISH files have been corrected (2 July 2004) to omit the
sources which are the sum of other "components" also appearing in the
lists. Note that the survey files shipped with AIPS releases include
the shorter lists of the stronger sources. The full lists are only
available from this ftp site.
AIPS Memos, Reports, old AIPSLetters, the Cookbook, Going
AIPS, access to the newest versions of the
CHANGE.DOC and more. The most recent AIPS Memos
- number 124,
discussing TVIEW, a new task for exploring image cubes,
dated November 15, 2018,
- number 123,
discussing the new pulse-cal cabilities of DiFX and AIPS,
dated October 30, 2017,
- number 122,
discussing the usage of AGAUS and ZAMAN spectral absorption modeling tasks,
dated March 7, 2017,
- number 121,
discussing the usage of the UFLAG gridded data flagging task, dated
September 12, 2016,
- number 120,
discussing the usage of the TVSPC data exploration task,
revision dated July 22, 2017,
- number 119,
discussing the usage of the TVSAD Gaussian modeling task, dated
December 15, 2014,
- number 118,
discussing the usage of XGAUS, ZEMAN, and RMFIT spectral modeling tasks revision dated November 8, 2017,
- number 117,
discussing the detailed FITS format used by AIPS, revision dated October 14, 2019,
- number 116,
discussing new RFI mitigation options in AIPS, dated June 2010,
- number 115,
discussing new auto-boxing for Clean capabilites in AIPS, dated June 29, 2009,
- number 114,
re-stating the FITS-IDI convention with corrections and added
tables, revision dated February 16, 2016,
- number 113,
discussing a "new" geometric way to handle the W problem, dated May 22, 2009,
- number 112,
discussing the AIPS pipeline for VLA data reduction, dated March
- number 111,
discussing phase referencing with more than one calibrator using
ATMCA, dated January 6, 2005,
- number 110,
discussing strategy for removing troposppheric and clock errors
using DELZN, revised October 21, 2009, dated August 31, 2004,
- number 109,
discussing AIPS on DVDs and read-only data systems, dated
January 20, 2004,
- number 108,
discussing weights for VLA data, dated January 21, 2003,
- number 107,
discussing the differences between FRING and KRING, dated April
- number 106,
discussing the Brandeis scheme for making movies in AIPS dated
June 6, 2001,
- number 105,
discussing the new RUN file to simplify VLBA data reduction
dated February 28, 2001 and April 26, 2001,
- number 104,
discussing the new Y2K performance test dated September 8, 2000, and
- number 103,
discussing data weighting in AIPS dated 2000-03-21.
The AIPS group maintains three e-mail exploder lists. The first,
called daip, is restricted to immediate members of the AIPS group and
a few NRAO scientist advisors. The other two are, however, open to
any suitable person. Membership is moderated, of course, because SPAM
has become so annoying.
The group intended for discussion of AIPS issues is called
bananas and is used, among other things, to announce patches
to the "frozen" versions of AIPS. Traffic is light. You can subscribe
yourself to this list on-line.
The group intended for discussion of problems related to the
"Midnight Job" is called mnj and is used to announce AIPS
Manager things related to the current MNJ. Traffic is light, but the
announcements can have a serious impact on any site attempting to stay
current with the development version of AIPS. You can subscribe
yourself to this list on-line.
The FITS Interferometry Data Interchange
format is a set of FITS
binary table definitions and a set of conventions for using them to
transport radio interferometry data between different software
The FITS Interferometry Data Interchange Convention (formerly Format)
has been accepted as an IAU-acknowledged convention. This convention
was documented by Chris Flatters in AIPS
Memo 102 which, unfortunately, was not accompanied by an editable
text version. Therefore, Eric Greisen had to re-type the document.
While doing so, he updated some matters, corrected minor wording
issues, and added 5 new sections for 5 new tables, 3 of which are
actually in use. Interested parties should review AIPS
Memo 114 revised (ps) or AIPS
Memo 114 revised (pdf); it has been submitted to the IAU FITS
Committee. Greisen has attempted to highlight the changed parts in
red in this draft. The current draft is dated August 2011 and
includes the MODEL_COMPS table and comments from Lorant
Sjowerman, Bill Cotton, and others. Please e-mail
firstname.lastname@example.org with comments.
Help for the weary AIPS Installer. Very much experimental and
incomplete, and probably needs updating, but better than nothing.
The AIPS Benchmark was known as the "DDT" (Dirty
Dozen Tasks); we have results for a variety of systems. A more modern
test called "Y2K" has been developed in 2000 to test faster
computers and to use IMAGR rather than MX. For 31DEC04 we have new master images and data sets. In
addition, we have a new HUGE test using multi-resolution IMAGR on
Cygnus A. Changes to CALIB and VTESS in particular
forced this update, while faster computers make Y2K LARGE run in as
little as 5 minutes.
The former designated-AIP-of-the-week program; rotation has been
halted due to the lack of personnel. The e-mail address
email@example.com) was the best method of reaching all
the members of the AIPS Group and will still work. However, we ask
you to use the new NRAO-wide helpdesk instead. Go to
and login with your my.nrao.edu username and password. There you may
peruse the "knowledge base" (currently empty), enter new tickets in
any of several departments including AIPS, and review and update any
tickets you have submitted.
AIPS is copyrighted by Associated Universities, Inc. using the GNU
copright form. The full text is available with every AIPS release and
may be read here.
This is the Flexible Image Transport System, and is the
standard data interchange format used by most Astronomical Software,
including AIPS. Follow the above reference for a very extensive
set of documentataion, including a large resource list for Astronomy on
The primary AIPS address is the
referenced on our main Contact page, and general
queries, requests for distribution tapes, documents, ftp access, and so on
should be directed there. Application-oriented questions should be
directed to the Designated AIP.
Modified on $Date: 2019/11/14 00:09:14 $
Eric W. Greisen