>>>>> "Doug" == Doug Stewart writes:
Doug> I am new to the list so this might be known already.
Doug> I calculated 1000000 digits of PI and found the last 4 are wrong. I then
Doug> calculated 3142 digits and found the last 2 were wrong. The method that
Doug> is used seems to leave the last few digits wrong so you should calculate
Doug> a little farther so that you can display the correct digits to the depth
Doug> asked for.
Yes. fpprec is the number of digits you used in bigfloat numbers, so
there will be rounding errors. I see from the code that pi is
computed with 3 extra bits of precision and then rounded. I guess
that's not enough to produce an accurate value of pi. Other functions
also use a few extra bits to reduce roundoff.
It would be nice, however, if maxima would produce values for pi, e,
gamma, etc., to full accuracy, even if other bigfloats aren't.
Ray