[LEAPSECS] real-time UT1 realisation

[Zefram]

Tom Van Baak wrote:

>     Handling day boundaries smoothly would depend on how

>close to UT1 you want to be, philosophically. The IERS publishes

>past, present, and future DUT1 to sub-millisecond resolution, but

>I don't know if you want to interpolate from day to day, or if the

>IERS quantized time/rate jumps every midnight are acceptable.


I thought about the same thing, with a view to displaying UT1 on a
conventional UTC-synched system.  The biggest issue, it seems to me,
is not interpolating within days (for which linear interpolation is a
fine first implementation), but switching from one set of predictions
to another.  The DUT1 that is available for current time is necessarily
always an extrapolation from older data, and extrapolations made at
different times will disagree, so the sequence of predicted-current-UT1
has serious discontinuities.  Suppose you always use the predictions
from the latest Bulletin A; every Thursday you'll do a little jump as
you switch from using last week's Bulletin A to the new one.

Whether to jump now or slew, and over what time period to slew, is
of course the same dilemma faced by NTP and by each UTC(k).  In the
programming realm, I'd like to see explicit distinction, in time
acquisition APIs, between "instantaneous best guess of X" and "local
time scale steered to X".

-zefram