Issue 231 in mpmath: mpmath and mathematica diverge at nearly 186 digits

[mpm...@googlecode.com]

Status: New
Owner: ----

New issue 231 by vick1...@orange.mu: mpmath and mathematica diverge at
nearly 186 digits
http://code.google.com/p/mpmath/issues/detail?id=231

What steps will reproduce the problem?
1. definite integral using mpmath quadts for tanh-sinh quadrature:
2. function to integrate is
1/sqrt((.0000812*(1+x)^4)+(.2719188*(1+x)^3)+(0*(1+x)^2)+.728)
3. from 0 to 1e+13
4. in excess of 200 digits accuracy

What is the expected output? What do you see instead?
expected output is as per mathematica result in attached file
resultmath.docx and what i see is in attached file test.txt

What version of the product are you using? On what operating system?
python 2.7 mpmath-0.17, python 2.7.3 on windows 7

Please provide any additional information below.



Attachments:
test.txt 658 bytes
resultmath.docx 21.7 KB

[mpm...@googlecode.com]

Comment #1 on issue 231 by fredrik....@gmail.com: mpmath and mathematica

diverge at nearly 186 digits
http://code.google.com/p/mpmath/issues/detail?id=231

I think this is due to the large integration interval. In fact the error
estimate returned with error=True seems accurate. As a workaround you could
perhaps replace the single integration interval with something like
[0,1e1,1e2,1e3,1e4,1e5,1e6,1e13]. It would obviously be a great improvement
to have space-adaptive quadrature that deals
automatically with this kind of integral.

[mpm...@googlecode.com]

Comment #2 on issue 231 by vick1...@orange.mu: mpmath and mathematica

diverge at nearly 186 digits
http://code.google.com/p/mpmath/issues/detail?id=231

Actually it is the online mathematica version that was at fault! I checked
the result from a student version and a trial version and both of them
concur with mpmath! Thanks!