Re: dblquad issues for 3.2.4 and 3.6.3

[twinclouds]

After my first posting, I received a number of replies.  Thanks for all of
your help.  Here I would like to give an update on my findings on this
issue.
The function I want to integrate is not very well behaved.  I made some
changes to make it better behaving and the dblquad in both 3.2.4 and 3.6.x
worked.  The one in 3.2.4 is much faster still.  
However, after I made some more changes of the function to have more "close
to singular" points in the integral, I had the same problem again.  What I
observed is that the default base integrators in dblquad are different for
3.2.4 and 3.6.x.  For 3.2.4 the default integrator is quadgk.  I am not sure
which one is in 3.6.x but it is not quadgk for sure.  If I force the base
integrator to be quadgk in 3.6.x, the result is the same as 3.2.4, i.e., it
will claim the integrand overflowing in bad cases but will be as fast as
3.2.4 if it works.  If don't specify the baseline integrator, it will work
but will take very long time (> 10 hours!).  When the function is relative
well behaved, it only take a few seconds.
I asked a friend to see how it behaves in matlab (2011b?).  For the bad
cases, matlab also has the same overflow problem if not specify the base
integrator or use quadgk.  However, what he found is that if he specify the
base integrator to be quadl, it works without any problem and the operation
only take less than 30 seconds.  I am trying the quadl also on Octave, it
has been over one hour but still haven't get the result yet.  I don't know
what is difference between the implementations of quadl in Matlab and Octave
and don't know if it is possible to make Octave behave more like Matlab when
using quadl.  Right now, the only working version in Octave (3.6.x with
default base integrator) took more than 10 hours and it is too long!  Hope
some more experienced people will find out why and find a solution.  Thanks
again.



--
View this message in context: 
http://octave.1599824.n4.nabble.com/dblquad-issues-for-3-2-4-and-3-6-3-tp4648989p4649072.html
Sent from the Octave - General mailing list archive at Nabble.com.