Minimal floating point exponent for double precision; smaller exponents are principally possible (up to -16383), but not supported in all math functions.