Linsolve broken recently? was Re: Oscillatory integrals



On 9/28/2013 12:43 PM, John Lapeyre wrote:
>
>
> But I haven't compared the more complicated things like cos(sinh(x))....
> Maybe there are some that quadpack cannot do well, but LevinInt or
> something like it can.
>
> Then there are the other routines.... I think this is a big
> project.
Yes.  There are lots of approaches and only limited guidance to tell you
which one is best.  The Mathematica code for (numerical + symbolic assist)
integration is huge.
>
> I wonder if changevar has always been broken in that it gives
> the wrong sign sometimes. Or is that sort of recent? There
> are a couple of bug reports on it.
>
Changevar has to solve y=f(x)  for x, and if there are multiple solutions,
pick one. I assume that what is reported as a bug is related to
choosing an inappropriate solution.  I don't know if this can be
fixed by looking at assumptions or something else.
RJF

> >