On Wed, Mar 28, 2012 at 11:22 AM, Edwin Woollett <woollett at charter.net>wrote:
> On Mar. 27, 2012, Richard Fateman wrote:
> -----------------
>
> Another approach is to say that the quadrature program provides
>> only "absolute error" bounds if the answer is near zero, instead of
>> "relative error" (or X digits of accuracy) elsewhere.
>>
> ------------------------------**----
> This suggests that a quadpack wrapper redo a calculation
> which returns an answer "near" zero (together with
> error code = 2) by replacing the rel error criterion with
> an abs error criterion. Using an example suggested
> by Ray Toy:
> ----------------------
>
Perhaps you've seen this already, but in case you haven't, take a look at
the end of src/numerical/slatec/quadpack.lisp. There are a bunch examples
taken from the quadpack book. You might be interested in the failure
cases. (The listed failure cases and values are from the book, I think.
The actual failures appear to be somewhat different, if you run the tests
yourself.)
Ray