From: Matthew Austern (austern_at_[hidden])
Date: 2001-09-04 12:32:13
Kevlin Henney wrote:
> > From: Matt Austern <austern_at_[hidden]>
> >I'm more familiar with the very small units than with the very large,
> >but in my experience it's quite common to convert between (say) electron
> >volts and joules, and between meters and bohr radii.
> This is true, but the number of significant figures remains the same.
> So, although the magnitudes vary, the floating point floats
Not quite true, actually, since the conversion factors in these two
examples I gave are measured quantities as opposed to defined
quantites. They have finite precision. For example,
1 eV =1.602176462(63) x 10^-19 J, where the digits in parens are
the 68% confidence limits for the last two digits.
If anyone was doing a math library that attempted to keep track of
the accuracy of physical measurements, they'd have to keep this
into account. Sounds like a very hard task, though, since
uncertainties are introduced in lots of different ways and since
they aren't always gaussian.
Incidentally, one good source for numbers like this is
numbers come from a roundup of all of the relevant experimental
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk