Avoiding the Leap Second

In the beginning, there was Ephemeris Time.

Then, it got replaced by TAI.

Both were made up of seconds the length of which was based on the
length of the second in 1900. But the Earth's rotation has been
slowing down, due to tidal forces.

Because TAI started off from civil time at the time of its adoption,
while Ephemeris time presumably started on noon, December 31st, 1899,
GMT, a clock showing Ephemeris Time would be 32.184 seconds ahead of
one showing TAI.

And a clock showing TAI would be 19 seconds ahead of one showing the
time used in the GPS system.

Civil time switched over to atomic time with inserted leap seconds
when TAI was already 10 seconds ahead of civil time.

Anyways: a while back, there was a message in these newsgroups about
how a group, shrouding its activities in mystery, came forwards with a
proposal to just forget about leap seconds. We could always adjust our
clocks an hour at a time, if we really felt strongly about wanting
local 12 noon to happen around lunchtime.

I don't think the mass of humanity really considers it so important to
be "modern" and "scientific" that they would willingly allow the clock
to be independent of the real time of day.

But it is true that leap seconds are awkward and confusing. I would
like to suggest an alternative for those whose concerns are precise
ones.

There are 86,400 seconds in a day, and about 365 days in a year. If we
add one second to a year, then, that lengthens the year by one part in
365 times 86,400.

If we instead increased the length of every *second* in that year by
the same proportion, we would be making the civil second equal to 1 SI
second and 31.70979198... SI nanoseconds.

This would be an approximation to UT1, or mean solar time, the way
civil time was kept *before* we had leap seconds.

Adding 30, 32, or 33 1/3 nanoseconds... or going from 9,192,631,770
cesium atom oscillations to 9,192,632,061.5 cesium atom oscillations
(adding 291.5)... or whichever approximation might be most convenient,
for the forthcoming year to the length of the second; or subtracting,
or adding twice that, as necessary, would allow us to have a civil
time without leap seconds.

The length of the civil second would vary, but there would only be a
limited number of possible variations, separated by uniform steps.

I presume this would be good enough for those applications where leap
seconds are disruptive; those where a second absolutely fixed in
length is required would have to cope with the difference between TAI
and civil time and its changes.

Since I'm proposing changing the length of a second, though, by an
approximation, rather than the *exact* proportion that adding an extra
second would make, this would not lead to TAI minus civil time being
an integer number of seconds at least at the start of each new year.
There are two possible cures: use the exact proportion instead
(adjusted in leap years!), or switch from a longer second to a regular
one before the end of the year (for example, splitting the leap second
up among all the seconds of the first 360 days of the year would lead
to an "even" lengthening of the second in some senses).

John Savard