[LEAPSECS] the big artillery

[Zefram]

Warner Losh wrote:
>On Nov 5, 2014, at 10:54 AM, Zefram <zefram at fysh.org> wrote:
>>    TAI(k) = TAI + (UTC(k)-UTC) = UTC(k) + (TAI-UTC)
>
>Except that's not how others define it.

Michael Deckers has now pointed at ITU Rec TF.536-2 which defines "TAI(k)"
in the same way as I do.  What conflicting definitions do you see?

Anyway, this isn't about the notation, it's about the concept.

>What you should be writing is something more like TAI(UTC(x)) to denote that
>you are deriving TAI form UTC(x), not that x is realizing TAI.

I wouldn't object to using such notation.  However, I reject the claim
that this doesn't amount to x realising TAI.

In the rest of your message, I'm not clear whether your uses of "TAI(x)"
refer to the concept I was talking about or something else.  If it's
something else, I'm mystified as to what, but then we'd be talking at
cross purposes whatever it was.  To be clear, I'm talking about the thing
that I called "TAI(x)", that is the TAI realisation implied by UTC(x), not
anything else that someone might have given the same name.  My argument
would certainly not apply with any other concept substituted there.

>                  FOO(x) is the FOO timescale as realized by x. You have to
>have actual clocks or oscillators ticking the signals out. To while UTC(x) exists for
>a large number of x, TAI(x) doesn't.

How does TAI(x) not exist?  I have explained how the time signals that
deliver UTC(x) serve equally to deliver TAI(x).  Are you claiming that
the TAI(x) doesn't count merely because the time signal supplier doesn't
*intend* the signal to provide TAI(x) per se?  That would be a poor
reason to reject the validity of TAI(x), given that TAI(x) is directly
implied by UTC(x).  Or perhaps you claim it doesn't count because the
signal doesn't fully encode TAI(x) in-band?  But as I explained, MSF
doesn't fully encode UTC(NPL) in-band either.

>                                     You can find the corresponding TAI time for any
>x's UTC(x) after the fact when  BIPM publishes the data.

Red herring.  I'm not saying you get canonical TAI in real time;
that's impossible.  You equally can't get canonical UTC in real time.
What you get in real time is TAI(x), along with UTC(x).  These are pretty
good approximations of canonical TAI and canonical UTC respectively.

-zefram