floating point numbers don't add right

[janice]

Hi all,

I read a lot of info on this issue, but still don't get it. Why does
this happen?

document.write( ".05" + " \+ " + ".05" + " \= " + (.05 + .05) +
"<br>");
document.write( "1.05" + " \+ " + ".05" + " \= " + (1.05 + .05) +
"<br>");
document.write( "2.05" + " \+ " + ".05" + " \= " + (2.05 + .05) +
"<br>");

Outputs:

..05 + .05 = 0.1
1.05 + .05 = 1.1
2.05 + .05 = 2.0999999999999 996
I also tried the code provided in this group's FAQ, and it worked
except for when it reaches an "even" number which should display 6.00
for example, it displays 5.100.

thx!

J.

[Philip Ronan]

On 23/9/04 9:14 pm, janice wrote:

Hi all,

I read a lot of info on this issue, but still don't get it. Why does
this happen?
(snip) 2.05 + .05 = 2.0999999999999 996

In binary, these numbers go on for ever:

2.05 = 10.000011001100 110011001100110 0...
0.05 = 0.0000110011001 100110011001100 ...
2.10 = 10.000110011001 100110011001100 1...

With a fixed number of bits it is impossible to represent these numbers
exactly, so they contain errors right from the start. Addition makes the
errors even larger. For example, if I added together the first two binary
numbers shown above, the last digit of the result would be 0, not 1.

It's just like adding 1/3 to 2/3 in base 10:

2/3 = 0.6666666666666 66
+ 1/3 = 0.3333333333333 33
= 0.9999999999999 99 (not 1.000000)

It is just as impossible to write 1/3 in decimal as it is to write 0.05 in
binary.

I hope that makes sense.

--
Philip Ronan
ph***********@v irgin.net
(Please remove the "z"s if replying by email)

[Douglas Crockford]

> I read a lot of info on this issue, but still don't get it. Why does

this happen? .05 + .05 = 0.1
1.05 + .05 = 1.1
2.05 + .05 = 2.0999999999999 996

That's really annoying, isn't it? JavaScript uses binary floating point
notation internally to represent numbers. Floating point is very useful
for some scientific applications, but for other applications it has some
severe shortcomings.

In case, you are seeing that it cannot exactly represent decimal
fractions. That would be ok on a planet where people do not commonly use
decimal fractions. It might seem that floating point was a poor choice.

http://www.crockford.com/#javascript

[Lee]

Douglas Crockford said:

I read a lot of info on this issue, but still don't get it. Why does
this happen?

.05 + .05 = 0.1
1.05 + .05 = 1.1
2.05 + .05 = 2.0999999999999 996

That's really annoying, isn't it? JavaScript uses binary floating point
notation internally to represent numbers. Floating point is very useful
for some scientific applications, but for other applications it has some
severe shortcomings.

In case, you are seeing that it cannot exactly represent decimal
fractions. That would be ok on a planet where people do not commonly use
decimal fractions. It might seem that floating point was a poor choice.

The advantages of using standard IEEE floating point representation
far outweigh any disadvantage. The infinitesimal error encountered
in typical web-page arithmetic is easily corrected by rounding.

When representing numbers in binary, each bit represents either
0 or 1 times 2 raised to some positive or negative integer power,
so a binary representation of a value between 0 and 1 is the sum
of the series (b1)/2 + (b2)/4 + (b3)/16 + (b4)/32 + ...
where each (bn) value is either 0 or 1. It's mathematically
impossible to represent some decimal fractions as a finite
series of such terms.

[Douglas Crockford]

Lee wrote:

Douglas Crockford said:

I read a lot of info on this issue, but still don't get it. Why does
this happen?

.05 + .05 = 0.1
1.05 + .05 = 1.1
2.05 + .05 = 2.0999999999999 996

That's really annoying, isn't it? JavaScript uses binary floating point
notation internally to represent numbers. Floating point is very useful
for some scientific applications, but for other applications it has some
severe shortcomings.

In case, you are seeing that it cannot exactly represent decimal
fractions. That would be ok on a planet where people do not commonly use
decimal fractions. It might seem that floating point was a poor choice.

The advantages of using standard IEEE floating point representation
far outweigh any disadvantage. The infinitesimal error encountered
in typical web-page arithmetic is easily corrected by rounding.

When representing numbers in binary, each bit represents either
0 or 1 times 2 raised to some positive or negative integer power,
so a binary representation of a value between 0 and 1 is the sum
of the series (b1)/2 + (b2)/4 + (b3)/16 + (b4)/32 + ...
where each (bn) value is either 0 or 1. It's mathematically
impossible to represent some decimal fractions as a finite
series of such terms.

That is precisely why floating point is poorly suited to most
applications. For applications that require decimal precision (such as
those involving money), there are other representations that are much
better. Scaled integers and BCD, for example.

2.0999999999999 996 shows 3.0 to 16 decimal places, nearly all of them
wrong. This result, to most people, is surprising.

[Lee]

Douglas Crockford said:

That is precisely why floating point is poorly suited to most
applications . For applications that require decimal precision (such as
those involving money), there are other representations that are much
better. Scaled integers and BCD, for example.

2.099999999999 9996 shows 3.0 to 16 decimal places, nearly all of them
wrong. This result, to most people, is surprising.

When handling money, all you need to do is round to thousandths
and display to hundredths. It's pretty trivial.

The fact that it's surprising to neophytes isn't much of
a reason to change it.

[Lasse Reichstein Nielsen]

Lee <RE************ **@cox.net> writes:

When handling money, all you need to do is round to thousandths
and display to hundredths. It's pretty trivial.
No code is so trivial that it can't contain an error (except perhaps
the empty statement :).
The fact that it's surprising to neophytes isn't much of
a reason to change it.

It's perhaps too late to change it, although I don't think a change to
BCD-numbers like C#'s "decimal" type would be noticed by most
applications.

However, as a purely theoretical observation, I'd say that for a
script language that is used by so many (until recently)
non-programmers, it would probably be better to use a decimal number
format to avoid exactly these questions. It happens often enough
to be in the FAQ.

/L
--
Lasse Reichstein Nielsen - lr*@hotpop.com
DHTML Death Colors: <URL:http://www.infimum.dk/HTML/rasterTriangleD OM.html>
'Faith without judgement merely degrades the spirit divine.'

[Dr John Stockton]

JRS: In article <5b************ **************@ posting.google. com>,
dated Thu, 23 Sep 2004 13:14:59, seen in news:comp.lang. javascript,
janice <go****@janic e-eirich.com> posted :

I also tried the code provided in this group's FAQ,

The FAQ says : Operations on integers are exact if the true result and
all intermediates are integers within that range.

If you want results exact to the cent, then work in cents, not in euros.

Integers are exact up to as little over 9,000,000,000,0 00,000; enough
for most budgets.

--
© John Stockton, Surrey, UK. ?@merlyn.demon. co.uk Turnpike v4.00 IE 4 ©
<URL:http://www.jibbering.c om/faq/> JL/RC: FAQ of news:comp.lang. javascript
<URL:http://www.merlyn.demo n.co.uk/js-index.htm> jscr maths, dates, sources.
<URL:http://www.merlyn.demo n.co.uk/> TP/BP/Delphi/jscr/&c, FAQ items, links.

[Dr John Stockton]

JRS: In article <ek**********@h otpop.com>, dated Fri, 24 Sep 2004
19:20:58, seen in news:comp.lang. javascript, Lasse Reichstein Nielsen
<lr*@hotpop.com > posted :

However, as a purely theoretical observation, I'd say that for a
script language that is used by so many (until recently)
non-programmers, it would probably be better to use a decimal number
format to avoid exactly these questions. It happens often enough
to be in the FAQ.

The same applies in Delphi groups of news:borland.*, except that they
have no newsgroup FAQs.

--
© John Stockton, Surrey, UK. ?@merlyn.demon. co.uk Turnpike v4.00 MIME. ©
Web <URL:http://www.merlyn.demo n.co.uk/> - FAQish topics, acronyms, & links.
For news:borland.*, use their server newsgroups.borl and.com ; but first read
Guidelines <URL:http://www.borland.com/newsgroups/guide.html> ff. with care.