odd results with double precision in Java

Question

I'm simply trying to calculate percentage_imp, but instead of 0.22 (exactly 0.22, no rounding error), I get 0.22000000000000003!! I used to get similar odd results, and I've been told to move from float to double, but this one is still odd! All the variables below are double!

double percentage_imp= budget -  (sum_minlessi)/ (sum_i + sum_lessi);

Sounds like it's time to learn how floating point actually works ;) — Peter Lawrey, Oct 27 '13 at 07:18

score 3 · Accepted Answer · answered Oct 27 '13 at 05:28

3

Thats because of the floating point precision values.

You must read:- What Every Computer Scientist Should Know About Floating-Point Arithmetic

You must also read how floating point arithmetic and its internal representation works.

answered Oct 27 '13 at 05:28

Rahul Tripathi

152,732
28
233
299

1

I know this is heresy, but don't find that page very useful for beginners. It's fantastic (necessary, even!) for someone who kinda-sorta gets floats already and wants to _really_ understand them, but it's not the place to start. I posted a link above that I think is much better suited as an introduction. – yshavit Oct 27 '13 at 05:31
2

@yshavit:- I have just updated the second link for the refernce as to how floating point arithmetic works. Hope that makes sense for beginers! – Rahul Tripathi Oct 27 '13 at 05:32

score 1 · Answer 2 · answered Oct 27 '13 at 05:28

1

0.22 is not representable as a double.

As an example, 1/3 cannot be represented in base-10, so we approximate with 0.3333333333333333.

answered Oct 27 '13 at 05:28

Paul Draper

64,883
37
172
246

score 0 · Answer 3 · answered Oct 27 '13 at 05:28

Its a rounding error inherit in binary calculations. Not all rational decimal numbers can be represented as a single decimal number in binary. As such, when you perform operations such as multiply and divide, you'll get some nasty error on the last few digits.

Moving from double to float doesn't change this as double is simply a double precision floating point number.

I suggest you look at this link

As a extra bonus in java, simple operations such as binary multiplications are optimized as much as possible utilizing any sort of hardware optimizations when possible yielding different answers on different machines. If you require consistent behavior across all machines use the strictfp keyword in your class declaration.

odd results with double precision in Java

3 Answers3