Problems with floating point and modulus

Thread view:

Tasperian Jigs - 18 Apr 2006 12:30 GMT

Hi,

I try to get a 'next value' (depending on 'nextNumber') for a double with
this method.

public static double nextNearest(double value, double nextNumber)
{
final double remainder = value % nextNumber;
return value - remainder + (remainder > 0 ? nextNumber : 0);
}

But the 'remainder' variable does not get the exact remainder after the
modulus. Is there another way to realise this method so the return result
is the exact double that i want? I'm aware of floating point inaccuracies,
that's why I ask this question ..

Thanks,

J

Reply to this Message

Tasperian Jigs - 18 Apr 2006 13:14 GMT

> But the 'remainder' variable does not get the exact remainder after the
> modulus.

With the example values of 0.7 and 0.01. The outcome is in that case
0.7000000000000001

Reply to this Message

Patricia Shanahan - 18 Apr 2006 15:13 GMT

> Hi,
>
[quoted text clipped - 15 lines]
>
> J

Can you explain a bit more of the context and requirements?

There are generally a couple of ways of dealing with rounding error when
you need to do tests:

1. Use a small fudge factor, traditionally called "epsilon". To apply
this, you need to find a value for epsilon that is smaller than any
non-zero number you would get if you were doing exact real arithmetic,
but larger than the rounding error.

Something around 1e-12 times the larger input value might work.

Your test would become remainder>epsilon rather than remainder>0.

2. Do the controlling stuff in some form of exact arithmetic, such as
scaled integer. If all interesting numbers are representable as decimal
fractions with a bounded number of digits, consider BigDecimal.

Patricia

Reply to this Message

Problems with floating point and modulus

Free Magazines