On Sun, 27 Jan 2008 17:19:12 GMT, rjf <fateman@[EMAIL PROTECTED]
> wrote:
>Vladimir Bondarenko wrote:
>> ----------------------------------------------------------------
>>
>> Our little demo continues..... Hello again from the VM machine
>> which hopefully soon will not be ignored by CAS manufacturers.
>>
>> ----------------------------------------------------------------
>>
>> N[Integrate[Sin[z] BesselJ[1, z]/Sqrt[z + z^2], {z,0,Infinity}]]
>>
>> 0.800192
>
>This little demonstration illustrates how NOT to do re****ting of a bug.
>
>Either there is an error in Integrate
>or there is an error in N
>or there are errors in both
>or in neither.
>
>It is possible for N[] to produce what appears to be a 'wrong' answer
>because of limited precision resulting in excessive cancellation.
>Whether this is wrong or not depends on what you expect N[] to do in
>such cases.
>
>It is possible for Integrate[] to produce an answer that is formally
>correct but is in such a form as to cause massive cancellation, and thus
>give the appearance that N is incorrect.
>
>Someone posting a "bug" should specify whether this is the problem or
not.
>
>Perhaps the answer from Integrate is just clearly wrong, and N[] is the
>simplest way to show that it is off the mark. In which case the person
>presenting the bug should say so.
> For example, does N[Simplify[Integrate[...]],1000] also produce the
>same number, approximately?
I disagree.
A bug is a reproducible erroneous output.
No need to debug it.
No need to find an explanation.
The users who find the bug are the _customers_, not the employees.
Simply re****t it and let the CAS vendor's sup****t people take it from
there.
quasi
|
