This is the mail archive of the gsl-discuss@sourceware.cygnus.com 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]

Re: tru64 failures

To: Steve ROBBINS <stever at bic dot mni dot mcgill dot ca>
Subject: Re: tru64 failures
From: Brian Gough <bjg at network-theory dot co dot uk>
Date: Mon, 15 May 2000 15:24:02 +0100 (BST)
Cc: gsl-discuss at sourceware dot cygnus dot com
References: <Pine.SGI.3.96.1000513134003.11406B-100000@bottom><Pine.SGI.3.96.1000514181859.24730B-100000@bottom>

Steve ROBBINS writes:
 > make[2]: Entering directory `/export/cgm/u3/stever/src/gsl/monte'
 > ...
 > FAIL: vegas(f2), dim=9, err=0.0003, chisq=0.6418 (0.49968100184661118 observed vs 1 expected)
 > ...

I've just had a separate report of the same error, on sparc/solaris 2.7.

 > FAIL: vegas(f2), dim=9, err=0.0003, chisq=0.6418 (0.499681001846611517 observed vs 1 expected)

Looks like there is a problem with the algorithm.  The result is
consistently half the correct value -- I'd guess the routine is only
integrating half of the double-gaussion.

I can reproduce the problem myself now using
GSL_IEEE_MODE=double-precision, so it is something to do with
rounding.

References:
- Re: prerelease checks
  - From: Steve ROBBINS
- tru64 failures
  - From: Steve ROBBINS

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]