[LEAPSECS] BBC radio Crowd Science

[Zefram]

GERRY ASHTON wrote:
>I guess "reduced to scalar value" means applying a procedure such as:

Yes, that's about it.  I wouldn't describe the day spans as "seconds
excluding leap seconds"; better to put it as "86400 times the number
of days".  I would use an expression such as "86400*MJDN + 3600*H +
60*M + S".

>So SV seems inadmissable in any system that is expected to accurately
>label events during the leap seconds.

As I've already said, that scalar value is not sufficient to unambiguously
represent UTC times.  We're all familiar with time_t's inadequacy in
that regard.  But it's legitimate to use that kind of scalar value for
other purposes.  That is how TAI-UTC is defined, not just in the leap
seconds era but also in the defining expressions for the rubber-seconds
era that can be seen in tai-utc.dat.

When converting from TAI to UTC, it is not sufficient to know only TAI
and the instantaneous value of TAI-UTC.  This is true not only of the
TAI-UTC that I have described, which changes value at the end of a leap
second, but also of Warner's variant TAI-UTC which changes value at the
start of a leap second.  In the vicinity of a month end one requires
more information, amounting to *two* TAI-UTC values, those for the two
months involved.  So in practice one tends not to tie a TAI-UTC value to
an instant or a second, but to some larger time unit: a minute, a day,
or a month.  In that context it's rather pointless to argue about what
value TAI-UTC takes during a leap second.  It only matters for the theory.

-zefram