On Friday, 21 February 2014 at 05:21:53 UTC, Frustrated wrote:
I think though adding a "repeating" bit would make it even more accurate so that repeating decimals within the bounds of maximum bits used could be represented perfectly. e.g., 1/3 = 0.3333... could be represented perfectly with such a bit and sliding fp type. With proper cpu support one could have 0.3333... * 3 = 1 exactly.
I believe that the repeating decimals, or better, repeating binary fractions, will hardly be more useful than a rational representation like p/q. The reason is, if we take a reciprocal 1/q and represent it as a binary, decimal, or other fixed-base fraction, the representation is surely periodic, but the upper bound on the length of its period is as high as q-1, and it is not unlikely to be the exact bound.
For example, at http://en.wikipedia.org/wiki/Binary_number#Fractions , note that the binary fraction for 1/11 has period 10, and for 1/13 the period is 12.
Thus repeating decimal for a fraction p/q will take up to q-1 bits when we store it as a repeating decimal, but log(p)+log(q) bits when stored as a rational number (numerator and denominator).
Ivan Kazmenko.