[ntp:questions] A Suggestion For Abolishing the Leap Second

Quadibloc jsavard at ecn.ab.ca
Sat Jun 2 01:41:37 UTC 2007

jlevine wrote:
>    The proposal to incorporate leap seconds by changing the length of
> the
> second was tried before in the early days of atomic time. It was not a
> very
> workable idea then and it would be even more difficult to implement
> now.

Well, that is entirely possible. I realize that there were
difficulties in changing the length of the second in civil time
gradually to correspond to the second in mean solar time, so I tried
to come up with a proposal for a civil time scale that, although it
still avoids leap seconds by changing the length of the second, it
does so in a more structured manner.

>   In the first place, the unit of frequency plays a central role in
> precision
> measurements and fundamental constants -- much more fundamental than
> the unit of time. If the time services distributed a non-SI frequency
> then
> all frequency calibrations would become much more difficult and
> ambiguous.

I can understand that, and I do not advocate that any different
frequency be distributed.

Here is what I do propose, after further discussions in another
newsgroup (sci.astro.amateur) where this proposal was commented on to
a greater extent.

TAI plus 10 seconds would still be GPS time, of course.

UTC would still be around, in a slightly modified form. A single leap
second would always be added or subtracted at midnight, June 30th; two
leap seconds would always be added or subtracted at midnight March
31st and September 30th, and so on. The number of leap seconds for a
given year would always be announced prior to January 1st of the
preceding year. This means that the discrepancy between the modified
UTC and mean solar time would be allowed to be somewhat larger than
0.9 seconds.

People with very high-accuracy applications could always use TAI or
the replacement for UTC.

For those who have other kinds of timekeeping applications, where
coping with leap seconds would be difficult and awkward, I propose
that the length of the second be modified *for the first 360 days of
the year* so that this modification will add the *integer number of
leap seconds for that year* evenly to all the seconds of which the
first 360 days of the year are composed.

On my web page at


I outline how, with a limited amount of additional circuitry, either a
cesium oscillator or a rubidium oscillator could be made to drive a
clock that kept within 1/90th of a second of this nominal civil time
on years with one leap second. (On years with no leap second, of
course, all the seconds are SI seconds.)

So this scheme is very much tied down to atomic time; there is a
strict and simple mathematical relationship between time in my
proposed system and atomic time for any given year, once the number of
leap-seconds for that year is known.

>     To a great extent, where you stand on this question depends on
> where you
> sit, and the characterisitc of any compromise solution is that
> everybody will be
> at least a little bit unhappy.

I understand this. What I have tried to do - and I realize that my
proposed scheme may be a bit too baroque and complicated to be
seriously considered - is to provide a scheme that minimizes the

If you need atomic time, you can continue to stick to it. If you can't
abide leap seconds, a time scale is provided that is strictly tied to
atomic time, from which the atomic time can be simply and directly
derived, *despite* the fact that it has a second of variable length.
Because the variations in the length of the second aren't arbitrary,
but are kept to a confined domain of integer steps.

While my compromise solution may be unworkable for some reasons, such
as its admitted complexity, in *some* ways it seems that it just might
be the best possible compromise solution, coming close to offering
people the *best* of both worlds instead of the worst of both worlds,
like many compromises do.

John Savard

More information about the questions mailing list