Question

I’m writing a C library with GMP and its arbitrary precision floats. I need to chop off “small numbers”, but I don’t know how to determine what is small.

Say I set the precision of GMP floats (mpf_t) to n bit. Then what is to be considered small in the calculation?

For any of you who might be familiar with GSL (GNU Scientific Library), I need the equivalent of their GSL_DBL_EPSILON, which for double, on my 32-bit computer, happpens to be 2.2204460492503131e-16.

Thanks in advance,j.

Answer 1

I think I found the magical number: it is 2 ^{-(bits of significand precision)}
as detailed on this Wikipedia article.