Re: Besseli Function

[John W. Eaton]

On  9-Sep-2003, David Bateman <address@hidden> wrote:

| My question is then, why not just used the scaled calculation always? A quick
| test with the unpatched code shows that there seems to be little speed 
| difference. 

| octave:1> x = 20*rand(100,100)-10; tic; y0 = airy(0,x); toc
| ans = 0.14151
| octave:2> x = 20*rand(100,100)-10; tic; y1 = exp( - 2. / 3. .* x .* 
sqrt(x)).*airy(0,x,true); toc
| ans = 0.13462
| 
| Is there possibly a problem with precision?

Perhaps.  It would help if someone could do some checking.  I did a
spot check of besseli (0, x) and found that for some numbers in the
range x = 79.5:0.0001:80, one method was off from the other by 1e+17,
but the result value is on the order of 1e+33, so the relative error
is less than eps, which should be OK.

| If not I propose to just used
| the scaled calculations always internally, and return the scaled or unscaled 
| values depending on the the optional argument..

I think this is a good solution unless some checks reveal big
differences in the results.  Then you can still get the
scaled result back if that's what you want.  And we are probably doing
a reasonable job of computing the unscaled results too.

Thanks,

jwe



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------