This is the mail archive of the
`gsl-discuss@sourceware.org`
mailing list for the GSL project.

Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
---|---|---|

Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |

Other format: | [Raw text] |

*From*: Lorenzo Moneta <Lorenzo dot Moneta at cern dot ch>*To*: Andrew W.Steiner <awsteiner at gmail dot com>*Cc*: GSL Discuss Mailing List <gsl-discuss at sourceware dot org>*Date*: Tue, 28 Apr 2009 22:05:08 +0200*Subject*: Re: [Bug-gsl] problem in cubic solver*References*: <254EE974-5E07-4EE0-87F2-B2A4FAF11472@cern.ch> <63c059b10904280931o6600dacg80f040ff32a1ac71@mail.gmail.com>

(but only finding real roots).

Thank you for the link, I will try your code and check as well if it gives better numerical stability. In that case it wold be good also to have it in GSL.

We are using presently in ROOT this code, it would be good if we could use instead GSL directly.

On Apr 28, 2009, at 6:31 PM, Andrew W. Steiner wrote:

On the topic of cubics, have you tried translating CERNLIB's rrteq3 into C for root? I have found it to give better results in general than GSL's cubic routine, and thus it might be worth adding a translation of the CERNLIB code to GSL.

My attempt at this translation is given at line 517 of http://o2scl.svn.sourceforge.net/viewvc/o2scl/trunk/src/other/poly.cpp?revision=35&view=markup

Take care, Andrew

On Tue, Apr 28, 2009 at 11:11 AM, Lorenzo Moneta <Lorenzo.Moneta@cern.ch > wrote:

Hello,

I have found a problem with solver for cubic equation (both complex and real one)

In some particular case a NaN is returned, as shown in the attached text example.

The problem is observed mainly on 64 bit architectures (for example Linux 64-bit with gcc 4.3) and not on 32 bit architectures.

This is due to a problem in a sqrt. A patch is attached fixing the problem for the complex routine (gsl_poly_complex_solve_cubic ).

A similar patch can be probably done also for the real routine

This patch fails the current test for polynomial in gsl, however in my opinion, this is acceptable, because the test condition is too strict. With a slight change of the test coefficient, you will have a similar failure also with the current version.

Best Regards

Lorenzo Moneta ROOT project (http://root.cern.ch ) CERN

_______________________________________________ Bug-gsl mailing list Bug-gsl@gnu.org http://lists.gnu.org/mailman/listinfo/bug-gsl

**Follow-Ups**:**Re: [Bug-gsl] problem in cubic solver***From:*Andrew W. Steiner

**References**:**Re: [Bug-gsl] problem in cubic solver***From:*Andrew W. Steiner

Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
---|---|---|

Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |