4 views

Skip to first unread message

Aug 13, 2010, 1:46:23 AM8/13/10

to sage-devel

Maxima is printing what I consider a strange result on two Solaris machines,

when computing the asinh of 1.0. The leading zero is not printed, so a doctest

fails.

when computing the asinh of 1.0. The leading zero is not printed, so a doctest

fails.

sage: maxima('asinh(1.0)')

.8813735870195429

which fails a doctest

Expected:

0.881373587019543

Got:

.8813735870195429

http://trac.sagemath.org/sage_trac/ticket/9693

The Maxima developers can reproduce similar examples on Linux machines and seem

to see this more as a feature than a bug, as it allows printing more digits in

the 17 places allocated to printing numbers.

http://www.math.utexas.edu/pipermail/maxima/2010/022230.html

How should this doctest be changed so it passes?

What seems easiest to me, is just to change the test. So instead of finding the

asinh of 1.0, we do it of 2.0

sage: maxima('asinh(2.0)')

1.44363547517881

That gives a number > 1, so the problem with the missing leading zero goes. For

the point of view of a test, I don't see it really matters what the value is

going to be chosen.

A higher precision result of this, which I computed with Mathematica is:

In[4]:= N[ArcSinh[2],30]

Out[4]= 1.44363547517881034249327674027

It so happens I get *exactly* the same result then on all these machines.

* sage.math (Linux)

* bsd.math (OS X)

* t2.math (Solaris 10 on SPARC)

* fulvia @ skynet (Solaris 10 on x86)

* My Ultra 27 (OpenSolaris on x86)

In all cases, all digits are in fact correct - at least they agree with a higher

precision result I computed with Mathematica.

So I propose we change symbolic/expression.pyx to compute the asinh(2.0) instead

of asinh(1.0), and set the expected result to 1.44363547517881, as that's that

result is correct in every digit, and is given on all the machines where I have

managed to try Sage.

Does that seem a reasonable change?

Perhaps setting the expected result to 1.4436354751788... might be better, as

that will allow a bit of room for numerical noise, though all machines I've

tested on give exactly the same result.

Dave

Aug 13, 2010, 2:10:29 AM8/13/10

to sage-...@googlegroups.com

On Thu, Aug 12, 2010 at 10:46 PM, Dr. David Kirkby

<david....@onetel.net> wrote:

> Does that seem a reasonable change?

<david....@onetel.net> wrote:

> Does that seem a reasonable change?

+1

--Mike

Aug 13, 2010, 3:08:53 AM8/13/10

to sage-...@googlegroups.com

Thank you Mike. I'll make that change if there are no objections. But I'll wait

until I get a few responses from the answer to

sage: maxima('asinh(2.0)')

as I'd like to know if other systems give other results.

How can I add a comment to the doc test, which will not be printed in the

documentation? I'd like to add the result from a high-precision computation, so

anyone looking at the doctest in future could see what's a reasonable answer and

what is not. I get very frustrated when I see an expected result of

"23.32233298" or something like that, with no explanation why that result is

expected.

Dave

Aug 13, 2010, 11:29:46 AM8/13/10

to sage-...@googlegroups.com

I don't know of a way to add a comment to the doctest that won't be

shown as part of the documentation. I'd recommend either letting it

show in the documentation or adding it as a Python comment just after

the doctest (at the first line of the function body).

Carl

Reply all

Reply to author

Forward

0 new messages

Search

Clear search

Close search

Google apps

Main menu