I haven't followed this closely, partly because the code was never
very clean and tlimit was generally much better. But maybe there
is another issue worth pointing out, namely that the limit program
may be oriented to solving real limits, and that it makes mistakes
in the complex plane not because of bugs, but because it wasn't
thought about in the original design.
RJF
Raymond Toy wrote:
>>>>>>"Raymond" == Raymond Toy writes:
>>>>>>
>>>>>>
>>>>>>">" == SourceForge net writes:
>>>>>>
>>>>>>
> >>> Bugs item #1281737, was opened at 2005-09-04 15:30
>
> Raymond> [snip]
> >>> Summary: limit(atan(x)/(1/exp(1)-exp(-(1+x)^2)),x,inf,plus) - wrong
>
>[snip]
>
>While looking for other stuff, I notice that this exact limit problem
>was discussed at the beginning of this year. See the thread
>http://www.math.utexas.edu/pipermail/maxima/2005/009063.html
>
>There's even the same analysis of the issue with log(-1-%i*x) from
>Barton. Stavros also mentions some bugs with the limit of atan2,
>reported as bug 1103515. From the code, I see that atan2 gets the
>limits correctly only by chance because it computes limit(atan2) =
>atan(limit), so my fix for the log issue is correct only by chance
>too.
>
> Raymond> With this fix we get the expected limits for the original bug report,
> Raymond> and the two simplified cases. All of the tests pass, but that's
> Raymond> hardly conclusive.
>
>I'm going to apply this patch because it does fix the issue with log
>(with a pointer to the MIT TR where many (most) of the ideas for limit
>come from).
>
>Now I just have to figure out how to make atan2 work. This seems
>significantly harder than log because atan2 takes 2 args, so extra
>care is needed.
>
>Ray
>