[fitsbits] Re: leap second alert
Rob Seaman
seaman at noao.edu
Tue Dec 21 14:15:23 EST 1999
I thought I'd request a bit more background from our contact on the
inside of the time and frequency community, Judah Levine. Note that
the date on his original message was 21 November 1999. It also seemed
prudent to gauge if events had changed over the last month. The reply
seems to indicate that the T&F folks are proceeding at a measured pace.
My typically purple prose has been rendered down a bit, but Levine's
is untouched beyond some line-wrapping and such.
Rob
----
> From seaman Fri Dec 17 16:32:22 1999
To: jlevine at india.colorado.edu
Subject: Re: leap seconds
Hello Judah,
We corresponded a few weeks ago after your inquiries to Katy Garmany
were forwarded to NOAO. I'm trying to get a better feel for this
apparent initiative to abolish the leap second.
[...]
I have some further questions and comments. I think I'll concentrate
on one specific recent time-related project, rather than tackle the
broader issues of which I'm by no means an expert.
You said:
> Thank you for your comments. The Time and Frequency community regards
> leap seconds as a nuisance and a bother, and there is an increasingly
> strong push to abolish them. This would cause UT1-UTC to grow without
> bound.
>
> Time and frequency folks generally don't care about UT1, but I thought
> that this might bother astronomers. If it does not, then so much the
> better.
Like most of us I wear several hats. At one point, for instance, I
worked on an offshoot of the GONG project to perform an "astroseismology"
experiment on Polaris. Like many stars, Polaris is variable on a small
level and the idea was to use the techniques developed by GONG to attempt
to detect similar vibration modes as they are mapping in the Sun but in
the spectra of a star.
GONG relies on a stable clock to both detect the vibratory modes and to
synchronize the multiple stations girdling the world. [...]
The "SONG" project targeted Polaris with tens of thousands of high
resolution spectra over a two month period. A requirement for this
was maintaining an even observing cadence over this long period. We
performed various initial trials to determine the best way to achieve
this. It turns out that our data acquisition software for the CCD
camera attached to that spectrograph keeps a fairly regular cadence
naturally. I collaborated on a little paper (ASP Conf Ser 125, p190)
discussing this.
Over 600 free-running exposures, the cadence remained between about
68.4 and 68.7 seconds over a complete observing session (night). There
was that very slight 0.3s (0.5%) trend. The result of this trend is to
make the observations useless for the purpose, however. If you dig up
the reference you can see a plot of the residuals showing that the phase
of the observations sweeps out 120 degrees (20 seconds) over the course
of the night relative to an evenly gridded cadence.
No big deal. The solution was just to add a trigger to the data
acquisition software to open the shutter on a precise cadence. Most of
our remaining trials before the actual observing run were determining
the best exposure parameters to balance the signal to noise with the
rapid cadence necessary to detect the vibratory modes the stellar models
suggested should exist.
No big deal since the level of precision required was on the order of
10s of milliseconds per exposure. So I have another plot showing the
"after" residuals as being constrained to +/- 0.01s. A not too shabby
improvement of 3 orders of magnitude (with the added benefit of
maintaining the cadence from night to night).
No big deal - but how to implement this quickly, easily and reliably?
This was a spare time sort of experiment with no real budget beyond
data tapes to hold the spectra and the staff labor.
Well, NTP was the answer, of course. We do have a local time signal
available at the telescopes, but this is really only easily accessible
on the real-time telescope control computers. Our data acquisition
systems are run separately these days (for all the obvious reasons),
and virtually always rely on very general purpose computers (for all the
obvious reasons). NTP was already running on the mountain and could
easily supply the accuracy required for this particular experiment.
NTP also had one other characteristic - it provided an intuitive "UT"
in a very human accessible form. The point of this is that the Earth
is not an inertial observing platform and for many purposes astronomy
requires a heliocentric (or more correctly, barycentric) clock,
corrected for the varying light travel time across the Earth's orbit.
It turns out that even though the observations were scheduled when the
Earth's motion was more-or-less transverse to Polaris (normal procedures),
and even though the projected motion is relatively small at that high
ecliptic latitude, that over the duration of the experiment the cadence
would be shifted many complete cycles. In fact, the effect is quite
noticeable during a single day for most observing lines of sight and
times of year. (The total maximum heliocentric correction is about
1000s over 6 months.)
So the data acquisition software had to include the heliocentric
correction. We also had thoughts of repeating the experiment later
in the year or in succeeding years and wanted to maintain the same
cadence.
The issue isn't whether all this could be achieved using various other
time systems. (And note that just like GONG, this experiment benefits
from ignoring leap seconds - during a single run.) The issue is how
much effort has to be applied and whether the generally available civil
and community and internet resources are of any use. It was a great
convenience to have a simple resource easily available that could
reliably be used as just "UT". I had no need to use a heliocentric
correction that varies from year to year. I had no need for a time
varying correction to the conversion between UT and sidereal time.
A simple experiment (well, this aspect was fairly simple) that
benefited from simple tools.
Folks understand the current UT. They may not understand leap seconds,
but they understand that they are necessary to keep the clocks synchronized
to the Earth. I'm sure you've already considered the various civil
constraints relating to maintaining a knowledge of various "Earth centered"
time measures. (Might religious observances be one you missed?)
>From my point of view, most fundamentally we need to preserve civil time
as a scientifically viable resource. I previously stated that for most
applications the particular segment of the community that I'm associated
with (ground-based optical/IR observational astronomy) doesn't really
need access to a high-precision time signal. There are, of course, other
parts of astronomy that need such very much indeed.
But I guess what I've really come around to thinking (and the description
above is just one example) is that it is precisely that most applications
don't require "high tech" time that mandates that the generally available
"low tech" time remain attached to our most obvious physical standard -
the spinning Earth.
[...]
The diurnal rotation of the Earth is an issue in many subtle and
not-so-subtle ways. My feeling is that estimates of many hundreds or
thousands of years until the clock diverges from the Sun don't really
capture the major impact of the ~1s/year trend on both scientific and
civil users. This is actually a very large effect (or you wouldn't
have to issue leap seconds often enough to care...)
My analysis of all this is obviously rather naive. I have some experience
with these issues, but little deep expertise. But that is exactly why a
very serious requirement on civil time must remain to tie it to intuitively
obvious standards. It is not just astronomers and navigators who expect
midnight UT to correspond to midnight in Greenwich.
> On the other hand, if there is some application that would be seriously
> affected by this change then it would be useful to know about it now
> before the move to abolish leap seconds gains something approaching
> ustoppable momentum.
[...]
A whole other cluster of concerns relates to archival data. (One of
my other hats.) Like many other fields, Astronomy is experiencing a
vast interest in improving archiving and data mining. We are a field
that continues to reference data from tens and hundreds and thousands
of years ago in the refereed literature. [...]
Time references abound in our data. Just now I'm wondering what
changes would be required to the Julian Day system that establishes
an unbroken chronology reaching indefinitely far into the past.
(Historians also use this, so I suspect you already have an answer?)
Perhaps you could also discuss the precise nature and implications
of this leap second eradication? So far I'm not entirely clear what
is being suggested...to just stop issuing them?
Maintaining a secure chronology is absolutely essential. If the
current civil time standards are secure, the most obvious change to
preserve this security is to do nothing at all. Any change will have
associated risks. Any change will also have an indeterminant, but
very large, cost associated with it. This cost will be spread over
the surface of the Earth.
[...]
I do take it from the timing of this discussion that the time and
frequency community is not feeling panicked about Y2K? ;-)
Rob Seaman
----
> From jlevine at india.colorado.edu Sat Dec 18 17:51:09 1999
To: seaman at noao.edu
Subject: Leap seconds
Hello,
> I'm trying to get a better feel for this
> apparent initiative to abolish the leap second.
There has been grumbling for some time about the hassles associated
with leap seconds, and that grumbling is slowly increasing in volume
because of the projected increase in the frequency with which leap
seconds will be needed in the next century. To my knowledge, the
chief advocates for considering changing the current system are the
folks in the Time service of the Naval Observatory (Dennis McCarthy,
for example).
In order to address the quesion, a commission has been formed by
URSI. This commission wrote to me because of my connection with time
stuff, and I forwarded the question to all of my astronomical colleagues
in JILA, since I thought they would probably be the ones who were likely
to be most strongly affected if leap seconds were abolished. (There
will probably be some kind of formal solicitation in a few months, but
I thought it would be useful for people who felt strongly to respond
early and avoid the rush.)
I don't use UT1 myself, and so I have no strong personal bias one way
or the other. Let me try and summarize the arguments.
A. Keep leap seconds more or less as they are
1. Keeps civil time, astronomical time and atomic time more or
less in sync.
2. Allows the approximation UT1 ~ UTC, since the current system
is designed so that |UT1-UTC| <= 0.9s. This is very convenient
for civil time keeping, for astronomy, for religious observances,
etc.
3. Folks who need a long-term uniform RATE and who can't tolerate
the jumps caused by leap seconds already use some form of atomic
time such as TAI, GPS, etc.
4. The formats of many time services depend on the fact that the UT1
correction is less than 1 second, and some of these formats are
realized in hardware and are difficult to change.
B. Abolish leap seconds completely
1. UT1 would diverge from UTC by about 1 minute per century. Even
if the rate increased somewhat, that is not a big time problem
for the hassles that it causes.
2. Although the TIME of UTC would diverge from UT1, the RATE of
UTC would become more uniform since the leap second jumps would
be gone. Folks who needed a long-term uniform RATE would be
pleased.
3. Computer-based clocks that keep time as the number of seconds
since some epoch can't represent the leap second and there is
no good way of defining a time-tag for an event that happens
during a leap second. This is true both within the computer
itself and in various transmission protocols such as NTP.
4. As a matter of practice, most systems observe a leap second
by effectively stopping the clock for 1 second. NTP-type time
services usually transmit the last second of the day twice.
This is awkward and ambiguous.
C. Keep leap seconds but apply a variable number of them less frequently
1. Keeps the divergence between UT1 and UTC from becoming too large.
2. Makes the rate jumps associated with a leap second less frequent.
3. If leap seconds were always inserted at the same time (on 29 Feb.,
for example), the table listing them could be simplified.
(My opinion is that option C is the worst of all possibilities. It has
almost all of the drawbacks of both option A AND option B.)
Although there are a number of time scales that have uniform rates and
therefore could be used in place of UTC, none of them is as widely available
as UTC, because most timing laboratories transmit UTC exclusively. This
could certainly be re-visited, but that is how it works now.
My understanding is that your comments provide detailed support for A1
and A2. To the extent that your experiments depend on a long-term uniform
RATE or cadence you might favor B2 if your observations crossed a leap-second
boundary.
To a great extent, where you stand on the question depends on how you
use time signals and whether you think of them as supplying time or rate.
Although the current system is awkard to implement, my guess is that it
will not be changed because nobody can think of another solution that is
obviously better.
I hope this has made the issue a bit clearer, but write again if you
still have questions or comments.
Judah Levine
Time and Frequency Division
NIST Boulder
----
> From seaman Sat Dec 18 20:58:12 1999
To: jlevine at india.colorado.edu
Subject: Re: Leap seconds
Hi,
Thanks for your detailed reply. Very interesting. As you inferred,
I prefer option A - keep things the way they are. Note that I focused
on one particular observational experiment as an example, but that it's
possible to come up with many others.
Obviously the international time standards will continue to support many
useful alternatives that provide a different balance between the rate
and the time, as you put it. Professionals can arrange to use the system
that best fits their needs. (At least - well funded professionals.)
The question is what the default system should be that the professionals
provides to the public. I assume that there is no proposal on the table
to provide more than one choice to the public - the possibilities for
confusion would be manifold and the likely risks terrifying.
Assuming that only one civil time should be provided, I can't but think
that it would be a colossal mistake to divorce our clocks from our
terrestrial roots. I'm a pretty pragmatic grounded guy - but this
almost borders on theology (new age, old age, any age) to have the
hubris to decide for the species that our clocks aren't going to track
the motions of the Earth and the sky anymore - for the rest of eternity.
Tweaking the UTC policies is one thing. Eliminating the fundamental
design requirement that ties UTC to our planet is another.
I'm sure you understand that the astronomical community continues to
regard itself (with some justification) as the original "time and
frequency" community. Allowing all the clocks on the planet (except
for some very small fraction used by professionals) to drift by a
minute over the next century is really a huge change to our philosophical
and scientific points of view. A minute is a huge amount for many
purposes. (Surely, I'm preaching to the choir?) It's a quarter of
a degree on the sky - that's half the width of the moon and the sun.
I remember pointing (well, zeroing the pointing model of) a telescope
with a submillimeter instrument that could only detect the moon for
the required "peaking up". (The sun was too bright, of course.)
We had to detect the two opposing limbs of the moon and take the
difference to zero the coordinates. This task was hard enough as it
was - and vitally important, since all observing was done through
blind pointings (through some small aperture).
If we hadn't had a "UT" clock - or rather several clocks, one in the
dome, one in the control room, one on my wrist - this would have been
an awful chore with a very high likelihood of mistakes.
Think of all the movies in which the characters synchronize their
watches before jumping into action. What is being suggested is that
none of these watches will ever be synchronized with the original
definition of time ever again.
I see that the Smithsonian Magazine has a picture of Harrison's
Number Four clock on the cover. This pleased me no end, since my
Augustinian history of astronomy professor at Villanova had doted
on Harrison's clock. Harrison's clock was one of the triumphs of
navigation (and the Enlightenment, for that matter). To remove the
connection between the Earth and humanity's clocks is to say that
no tinkerer can ever build (and calibrate from first principles)
a good clock again.
If the issue is just the headaches associated with maintaining the
bookkeeping and infrastructure associated with leap seconds, might I
suggest that improving that infrastructure is where the effort be put?
And if you guys think you have headaches now, just wait...
[...]
Rob Seaman
----
> From jlevine at india.colorado.edu Tue Dec 21 08:30:03 1999
To: seaman at noao.edu
Subject: Re: Leap seconds
Hello,
> The question is what the default system should be that the
> professionals provides to the public.
I agree with this general idea, and I am not really in favor of
changing things. One of my initial concerns was that this argument
would prove to be not strong enough, and my idea in raising the
issue among the astronomical community was the hope that a stronger
(or at least an additional) argument would emerge.
> I assume that there is no proposal on the table
> to provide more than one choice to the public - the possibilities for
> confusion would be manifold and the likely risks terrifying.
No, this is not an explicit choice to my knowledge, but the GPS system
actually works this way now. The GPS time scale itself does not incorporate
leap seconds, but the leap second value is transmitted as part of the
navigation message. Thus you can use the same system to either have leap
seconds or not by either using the transmitted time directly or including
the correction value from the navigation message.
> Allowing all the clocks on the planet (except
> for some very small fraction used by professionals) to drift by a
> minute over the next century is really a huge change ...
I agree with this idea too. However, keep in mind that the other
side will respond that there is already an annual variation of
about 30 minutes p-p in the difference between what you might call
"sundial time" and a conventional clock. It is true that a secular
change of about 1 minute per century is different from a much
larger periodic effect, but the other side will argue that dealing
with the secular effect is not any harder in principle than dealing
with the much larger periodic one.
In my opinion, the strongest argument that I have heard so far is
that the astronomical community depends in an implicit way on the fact
that UT1 ~ UTC, and that removing this approximate equality will cause
enormous trouble in lots of obscure places. (Sort of like the Y2K
situation). I think I see this point in your examples, although you
may not have said it explicitly in exactly this way. You make the
additional point that this approximation is deeply rooted in the
civilian time system as well -- people expect the sun to be more or
less at the zenith at noon and would be very disturbed if this were not
the case. I have used the length and position of my shadow as a rough
navigation aid since I was a child. In fact, I didn't realize how much
I had come to depend on this system until it failed me on my first visit
to Sydney, Australia. Until then, I had never been in a place where the
Sun was North of me, and I had trouble re-calibrating my semi-automatic
navigation system.
> Unless you have an objection, I'd like to copy your reply to the FITS
> newsgroup.
Certainly. You can add this note too if you like.
> and I suspect you may get other more thoughtful (or at least, more
> pointed) replies.
That is fine. However, it would be nice if folks remembered that I
am just the messenger and that I am really trying to help. The real
fight will be next Summer within the URSI committee.
Best regards,
Judah Levine
--
seaman at noao.edu, http://iraf.noao.edu/~seaman
NOAO, 950 N Cherry Ave, Tucson AZ 85719, 520-318-8248
PGP: 98 8D 8B 49 74 9A 41 88 3A 43 87 54 51 BF 30 4B
More information about the fitsbits
mailing list