[LEAPSECS] What's the point?

[Rob Seaman]

Tony Finch wrote:


> Warner Losh wrote:

> 

>> Rob Seaman wrote:

>> 

>>> C) As pointed out on numerous occasions in the past, these kaleidoscopic timezones would accelerate quadratically just like leap seconds.

>> 

>> This problem isn't solved by this method either.  True.

> 

> Except that timezone adjustments continue to work much further into the future than leap seconds.


No - the 2nd derivative is the same whether the leap-second-equivalents (LSEs) are batched one-by-one or 3600 at a time.  (Putting aside the question of whether timezone adjustments would meet the project requirements in the first place.)

The current leap second policies are constrained to twice per year - this would correspond to a timezone do-se-do of 1800 years.  The actual standard, though, is 12 per year - that brings it down to 300 years, which seems similar in level of intrusiveness.  Larger interruptions must occur less frequently to be tolerated.

However, a leap second per day (or even multiples) is not logistically out of the question.  This is Mark Calabretta's epsilon.  One-per-day would mean a timezone reorganization every ten years, which would be absurdly unacceptable compared to taking our daily epsilon vitamin.

Leap seconds would be much more robustly tolerated into the far distant future than rubber timezones.

...which is all just to say that in sampling theory a finer grid is generally preferred.  For instance, the usual sqrt(1/12) quantization error applies as the LSE binning grows.

The usual disclaimer:  none of this is in the actual ITU draft.  There are complications with any position.  There should be a coherent plan developed in advance of any change to the UTC standard.  Due diligence has not been served.

Rob