«

Patricia Shanahan wrote:
> manzur wrote:
> > I have a bigdecimal and a double value,say
> >
> > Bigdecimal bd = 599999.45;
> > double d= 5.9;
> >
> > i want to do 599999.45^5.9
> >
> > Note:I dont want to convert bigdecimal to double
> >
> > thanks in advance.
> >
>
> As already said, non-integer powers in BigDecimal would be a significant
> project.
>
> Maybe there is a less drastic solution to your problem. Could you
> explain why you don't want to do the power calculation in double? That
> may give someone ideas for alternatives.
>
> Patricia


iam writing code for some banking applications which demand accuracy
in money
values.For money values iam using Bigdecimal. If i convert my
bigdecimal to double i fear of inaccuracy(In cases where i deal with
huge values greater than double)

manzur wrote:
> Patricia Shanahan wrote:
>
>>manzur wrote:
>>
>>>I have a bigdecimal and a double value,say
>>>
>>>Bigdecimal bd = 599999.45;
>>>double d= 5.9;
>>>
>>>i want to do 599999.45^5.9
>>>
>>>Note:I dont want to convert bigdecimal to double
>>>
>>>thanks in advance.
>>>
>>
>>As already said, non-integer powers in BigDecimal would be a significant
>>project.
>>
>>Maybe there is a less drastic solution to your problem. Could you
>>explain why you don't want to do the power calculation in double? That
>>may give someone ideas for alternatives.
>>
>>Patricia
>
>
>
> iam writing code for some banking applications which demand accuracy
> in money
> values.For money values iam using Bigdecimal. If i convert my
> bigdecimal to double i fear of inaccuracy(In cases where i deal with
> huge values greater than double)
>

You won't be dealing with values greater than double, because the
maximum double is bigger than 10^308. That is more than the probable
number of atoms in the observable universe.

However, it is entirely possible that you could get a wrong answer for
the least significant digit on large sums of money. Double is just about
precise enough to get the cents digit right on the US national debt in
dollars. If that is good enough, one solution might be to use double for
the exponentiation, but immediately convert back to BigDecimal. That
would avoid accumulating rounding errors during routine addition and
subtraction.

I'm a bit surprised by the use of exponentiation in this precise an
environment. For example, I would have expected compound interest to be
done a compounding period at a time, with the interest for each period
rounded according to fixed rules, then added to the balance.

Patricia

On 19 Apr 2006 21:58:45 -0700, "manzur" <> wrote,
quoted or indirectly quoted someone who said :

> iam writing code for some banking applications which demand accuracy
>in money
>values.For money values iam using Bigdecimal. If i convert my
>bigdecimal to double i fear of inaccuracy(In cases where i deal with
>huge values greater than double)

Hmm. What you might do is get a good initial approximation with
Math.pow on double. Then use a Newton-Raphson to home in on a very
accurate value.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Java custom programming, consulting and coaching.

Patricia Shanahan wrote:

> I'm a bit surprised by the use of exponentiation in this precise an
> environment. For example, I would have expected compound interest to be
> done a compounding period at a time, with the interest for each period
> rounded according to fixed rules, then added to the balance.

Another possibility is that "perfect" precision is not required for /this
particular/ calculation. For instance in the UK any offer of a loan must be
accompanied by (besides the exact statement of interest, etc) an indicator
figure which is supposed to make it easier for people to compare offers with
different conditions. I forget the thing's name, and the rules for computing
it are complicated, but the point is that floating-point accuracy is entirely
adequate, even though it's a financial calculation.

-- chris

[andyt]

Hi,

I've just released a package with a method in for this, but for some reason the forum won't let me post a link until I have posted 50 messages, so I've mangled the URL. Hopefully you can decode it and find what you want:

http_www_geog_leeds_ac_uk/people/a.turner/src/andyt/java/generic/

I have released under LGPL, but could release with a different license if anyone has a problem with that.

Best wishes,

Andy