[LEAPSECS] Reliability

[Adi Stav]

On Sun, Jan 04, 2009 at 08:58:29PM -0700, Rob Seaman wrote:

> Adi Stav wrote:

>

>> Then why 4 seconds? Because they could be predicted a decade in  

>> advance?  Isn't that putting the cart before the horses?

>

> Yes, indeed.  You asked a question.  I provided a guess.  Personally, I 

> think the current standard is better than allowing celestial coordinates 

> to slosh around by an arcminute, but it is not the astronomers here who 

> have refused to dicker.

>

>> If I reckon correctly, people on this list specified 20 or 60 minutes 

>> as their guesses for the limit

>

> Some people said such things.  Why lend them greater weight than  

> alternate opinions?


They don't have greater weight, but they have their own direct
justifications and so can be discussed by both those who agree and
disagree.

We know that human tolerance to DUT is higher than 20 minutes because we
don't usually bother to compensate for apparent solar time. We know that 
it is probably not much higher than one or two hours because time zones 
generally have about that resolution. We guess that it is might be about 
one hour because many areas in the world choose time zones that are about
one-hour offsets from their local mean time.

These justifications are not necessarily valid, or maybe there are other
or better justifications for smaller DUT maxima. I am just trying to
find out (for myself) what these are. This is why I asked.


> Nobody has specified anything.  Specifications relate to solution space.  

> We have yet to discover the requirements describing the problem space.  

> There is nothing to specify against.


True. I should have said instead "suggested", or "said".


>> Clearly, you think DUT should be smaller. Why? For practical reasons  

>> of astronomy? For other reasons?

>

> Yes and yes.  A static geographic offset is different from introducing a 

> permanent bias in the rate.  A zero-mean periodic variation as in the 

> equation of time is different from a permanent bias.  Seasonal step 

> function jumps are different yet again.

>

> Embargoing leap seconds (or their equivalent) for periods of decades or 

> centuries is the same as not making intercalary adjustments at all.


Why is that? Even the Gregorian reform does not come into effect except
every one or two centuries. Yet it is followed exactly.


> It 

> will introduce a tilted quadratic bias in the solar rate.  The issue 

> isn't about offsets at all, it is about preserving the correct functional 

> form of civil timekeeping.


(I hope) I fully understand this requirement. So I asked about the
maximal DUT instead.


> We have heard numerous times that the Gregorian calendar is acceptable  

> because the schedule is predictable.  


I can think of properties that make the Gregorian calendar acceptable 
and that not every UTC reform will necessarily have:

 * It does not introduce a new intercalary mechanism, and instead
   modifies (slightly) the frequency of a pre-existing and accepted one.
 * It is not only infinitely predictable, but also easy; every
   schoolchild can know it.


> But why is the Gregorian calendar 

> desirable?

>

> It is desirable because it stabilizes the civil calendar against the  

> natural annual rhythms of life on Earth AND because it does so at a fine 

> enough resolution to permit smoothing across the decades and centuries.  

> When considering dates 400 years from now - or in colonial America - we 

> don't have to wonder whether April occurs before or after the Vernal 

> Equinox.


This is a property going back to the Julian calendar. But I see your
meaning -- you are saying that one second is acceptable simply because 
it uses *our smallest standard* time unit (and four seconds likewise,
with a decade of advanced warning).

By the way, it can be argued that the smoothness property is not strictly 
necessary for calendars. Consider popular and long-used artihmetic 
lunisolar calendars, such as the Hebrew, Hindu, and Chinese calendars, 
that intercalate their years to a resolution of a month.