Thread Subject:

exp function

i s there a limit of exp(x) for which it will only give zero as answer.

In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.

Is this a limitation of Matlab or there is any alternative to get the small values also.

Thanks
Nilesh Goel

"Nilesh Goel" wrote in message <ito85l$fgm$1@newscl01ah.mathworks.com>...
> i s there a limit of exp(x) for which it will only give zero as answer.
>
> In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.
>
> Is this a limitation of Matlab or there is any alternative to get the small values also.


Instead of implementing as N*exp(-921), perhaps implement as

exp( log(N)-921 )

Hello Matt,

The problem is still there as now value of (log(N) - 921) is very less again so exponent of this gives zero o/p. :(

"Matt J" wrote in message <ito8eu$ge4$1@newscl01ah.mathworks.com>...
> "Nilesh Goel" wrote in message <ito85l$fgm$1@newscl01ah.mathworks.com>...
> > i s there a limit of exp(x) for which it will only give zero as answer.
> >
> > In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.
> >
> > Is this a limitation of Matlab or there is any alternative to get the small values also.
>
>
> Instead of implementing as N*exp(-921), perhaps implement as
>
> exp( log(N)-921 )

"Nilesh Goel" wrote in message <itobel$pbd$1@newscl01ah.mathworks.com>...
> Hello Matt,
>
> The problem is still there as now value of (log(N) - 921) is very less again so exponent of this gives zero o/p. :(
==================

Then you need to reformulate your calculations so that they can be accomplished within the precision capabilities of normal computers. Show a bigger chunk of what you are trying to do and maybe the community will have suggestions.

"Nilesh Goel" wrote in message <itobel$pbd$1@newscl01ah.mathworks.com>...
> The problem is still there as now value of (log(N) - 921) is very less again so exponent of this gives zero o/p. :(
>
> > > In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.
- - - - - - - -
  You should remember that the very smallest positive number possible in 'double' format is 2^(-1075). Your number exp(-921) is far below that and is forced to round to zero. There is no avoiding that. You should adjust your proceedings accordingly.

Roger Stafford

On Jun 20, 5:02 pm, "Roger Stafford"
<ellieandrogerxy...@mindspring.com.invalid> wrote:
> "Nilesh Goel" wrote in message <itobel$pb...@newscl01ah.mathworks.com>...
> > The problem is still there  as now value of (log(N) - 921)  is very less again so exponent of this gives zero o/p. :(
>
> > > > In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.
>
> - - - - - - - -
>   You should remember that the very smallest positive number possible in 'double' format is 2^(-1075).  Your number exp(-921) is far below that and is forced to round to zero.  There is no avoiding that.  You should adjust your proceedings accordingly.
>
> Roger Stafford

For integer arguments I'm getting

2^(-1074) <= 2^(n) <= 2^(1023)

and

exp(-745) <= exp(m) <= exp(709)

Hope this helps.

Greg

Greg Heath <heath@alumni.brown.edu> wrote in message <43d85ca6-5b2e-4af4-ad1b-f0b1411c220d@y30g2000yqb.googlegroups.com>...
> For integer arguments I'm getting
> 2^(-1074) <= 2^(n) <= 2^(1023)
- - - - - - - - -
  Yes you are right, Greg. The smallest positive double is the denormalized number 2^(-1074), making a nonzero exp(-921) even more impossible.

Roger Stafford

On Jun 20, 9:48 pm, "Nilesh Goel" <nilesh.g...@gmail.com> wrote:
> i s there a limit of exp(x) for which it will only give zero as answer.
>
> In one of my equation exp(-921) is multiplied by a large constant but the answer is zero. I observe exp(-921) returns zero so even multiply it with large no. will return zero.
>
> Is this a limitation of Matlab

No. It's a limitation of finite wordlength aritmetics.

> or there is any alternative to get the small values also.

There might be. But if you routinely mess with numbers of
that kind of magnitudes, you might want to review the
scope of your work.

Rune