comparing the double result of and terminating a loop

[zskillz]

I'm running a loop where I sum very small numbers. I really think that I only need to be accurate to 35 digits or so. How do I compare only the first 35 digits of a double number to another number?

thanks
-Z

[Creosote]

What do you mean by compare? See if they are the same or something?

[strangerstill]

double epsilon = 10^-35;

if (abs(a - b) < epsilon ) {
// they are the same to within +- epsilon
} else if ( a < b ) {
// a < (b - epsilon)
} else {
// a > b + epsilon
}

[zskillz]

hmmm... i tried something along those lines stranger, but i need to know how big a double number really is before i pick an acceptable epsilon. I'm having trouble finding it.... apparently my google search technique needs some work.

-Z

[Creosote]

There has to be a better way than this--but you could convert it to a string and count t the characters.

[strangerstill]

So... use

double epsilon = max(abs(a), abs(b)) * 10^-35;

if (abs(a - b) <= epsilon) {
...

Is that what you mean? Or do you need to know the precision of the double type?

Floating point numbers are standardised in IEEE 754 - try http://www.google.com/search?hl=en&l...=Google+Search

The libraries should contain constants for the least and greatest positive values that exist in each type.