Re: Bessel Functions and Thumbscrews

[Robert T. Short]

On 03/02/2012 11:17 AM, Jordi Gutiérrez Hermoso wrote:
> On 2 March 2012 12:58, Robert T. Short<address@hidden>  wrote:
> > Following up on a post and bug report several weeks ago.
> >
> > I have attached a patch in which I added a bunch of tests to
> > bessel{j,y,i,k}.
> Thanks, this is helpful. I've pushed your patch and made a few style fixes:
>
>      http://hg.savannah.gnu.org/hgweb/octave/rev/e28a1723d5a2
>
> For future reference, you can use the three-argument calling form of
> assert for testing relative instead of absolute tolerance (pass a
> negative argument there). This is helpful because it produces a better
> diagnostic if the assertion fails.
>
> I also changed to xtest the two tests that failed on my machine. One
> of them seems to be getting the signs wrong for some entries. The
> other one seems to be getting the precision wrong.
>
> - Jordi G. H.
OK.  I will figure out how to do the relative tests and fix that.  
Thanks for the tip - I didn't pick up on that.

If these tests are failing for anything but a precision error, I think 
there is cause for real concern.  After all, I took the data out of 
well-established tables as my benchmark, and I get the same answers in 
the tables using multiple Bessel function routines (i.e. not just octave).

All these things pass on my machine.  Is there some way you can send me 
the details for the failed tests?  I would like for this to be really 
useful and correct (although I have already spent WAY more time than I 
had available). I really need all the gory details, not just the output 
from the tests.  I did this stuff using arrays, so I need to see which 
array elements fail so I can figure out what the problem is.  If it is 
helpful, I can fix the relative error thing first.

I am running debian testing, 32 bit, on an intel core i7 860.  I 
normally run the tip of the stable branch, but in this case I built and 
tested against the tip of the default.

Bob