hsiloP : A note on accuracy

What a chimera then is man! What a novelty! What a monster, what a chaos, what a contradiction, what a prodigy! Judge of all things, feeble earthworm, depository of truth, a sink of uncertainty and error, the glory and shame of the universe. (Pascal, 1657)

The numbers produced by hsiloP are accurate enough for everyday needs, but if you need more precision or want to work with very large numbers, you may want to use a different program, such as the free and cross-platform math program PARI/GP.

HsiloP stores numbers as double-precision floating point values. This means that numbers are accurate to about 16 significant figures (base 10). For small, integral values, exact calculations can be made. But with fractional values, there is a slight loss of accuracy with every operation. This error may become greater and greater with each successive calculation.

Also, operations that produce large, integral values may "overflow," sometimes without warning you that anything unusual has happened. For this reason, the factorial operator is limited to 20! (the largest factorial that will fit in a double without loss of precision). The combinatorial operators, if given large enough values, may "overflow" the bounds of the 64-bit integers used for their calculations, without warning you.

xkcd's take on floating point errors.