bug list summary

bug list summary

[Brian Gough]

I'm also posting a summary of the current BUGS list. 

These bugs are all open, so if you think you might be able to fix any
of them see the top-level BUGS file in CVS for more details about each
one.

BUG#1 -- gsl_sf_hyperg_2F1_e fails for some arguments 
BUG#3 -- gsl_sf_bessel_Jn fails for large argument 
BUG#4 -- gsl_linalg_solve_symm_cyc_tridiag() crashes 
BUG#5 -- broken error terms for implicit odes 
BUG#6 -- gsl_isnan, isinf on Compaq Unix 
BUG#7 -- gsl_sf_coupling_3j loses accuracy for large arguments 
BUG#8 -- inexact coefficients in rk8pd.c 
BUG#9 -- problem with multimin example 
BUG#10 -- gsl_sf_fermi_dirac_int error estimates 

Can we close BUG#6 ?

[Slaven Peles]

I posted this before, but I guess there's no harm repeating it. If I compile 
GSL on Compaq Alpha with the additional flag "-ieee" everything seem to work 
fine and make check reports that all checks are passed. So it seems that bug 
is actually in documentation only. In INSTALL file hints for Compaq/DEC Alpha 
should say:

> Hints for Compaq/DEC Alpha
> ==========================
>
> The library should compile successfully with Compaq's C compiler using
> the -std option for ANSI conformance.  Use
>
>    CC=cc ./configure
>    make CFLAGS="-std -ieee"
>
> to build the library this way.
>
> Use GNU tar to unpack the tar file, as Tru64 tar gives an error
> halfway through.

Is it safe to close this bug then?
Cheers,
Slaven


-------------------------------------------------------------------
BUG#6 -- gsl_isnan, isinf on Compaq Unix STATUS: open

From: Slaven Peles <peles@cns.physics.gatech.edu>
To: gsl-discuss@sources.redhat.com
Subject: Make check fails on Alpha
Date: Fri, 20 Sep 2002 08:47:54 -0400

I compiled GSL 1.2 on Compaq Alpha with:
    CC=cc ./configure
    make CFLAGS="-std"
Compiler I used identifies itself as:
    Compaq C V6.3-025 on Compaq Tru64 UNIX V5.1 (Rev. 732)
    Compiler Driver V6.3-026 (sys) cc Driver
Make check fails for gsl_isnan, gsl_isinf and gsl_finite functions.

make  check-TESTS
FAIL: gsl_isinf(inf) (0 observed vs 1 expected)
FAIL: gsl_isinf(-inf) (0 observed vs -1 expected)
FAIL: gsl_isnan(nan) (0 observed vs 1 expected)
FAIL: gsl_finite(inf) (1 observed vs 0 expected)
FAIL: gsl_finite(nan) (1 observed vs 0 expected)



On Tuesday 08 October 2002 03:31 pm, Brian Gough wrote:
> I'm also posting a summary of the current BUGS list.
>
> These bugs are all open, so if you think you might be able to fix any
> of them see the top-level BUGS file in CVS for more details about each
> one.
>

Re: Can we close BUG#6 ?

[Brian Gough]

Slaven Peles writes:
 > I posted this before, but I guess there's no harm repeating it. If
 > I compile GSL on Compaq Alpha with the additional flag "-ieee"
 > everything seem to work fine and make check reports that all checks
 > are passed. So it seems that bug is actually in documentation
 > only. 

Thanks, BUG#6 now closed.  I'd lost track of the status that one.

Re: bug list summary

[Luc Maisonobe]

[-- Attachment #1: Type: text/plain, Size: 1188 bytes --]

Brian Gough wrote:

> BUG#8 -- inexact coefficients in rk8pd.c

I finally got the reference article from Dormand and Prince (many thanks
again Joonas). In fact, the article itself gives these false
approximations and explicitly says "they were computed using a precision
of about 24 significant digits and the rationals presented in table 2
are continued fractions accurate to 18 significant figures".

I still think there is an error in the first and fourth elements of the
b7 array, but this error was in the original article. I also think the
real coefficients given in rksuite are more accurate, and since they
thank Dormand and Prince for their help, I think they got the original
coefficients from the authors. I don't know if all 30 significant
figures given in rksuite are good or if only the first 24 ones are good.

I will check this as time permit. For the moment, I propose the
following patch, which only add some comments (corected order of the
method, article reference) and correct two coefficients. I have just
written it now to answer this mail, not tried it (and even no compiled
it, so beware of typos ...)

                                                         Luc

[-- Attachment #2: rk8pd.patch --]
[-- Type: text/plain, Size: 1367 bytes --]

--- gsl-1.2/ode-initval/rk8pd.c~	Mon Nov 19 22:39:32 2001
+++ gsl-1.2/ode-initval/rk8pd.c	Wed Oct  9 09:49:26 2002
@@ -17,7 +17,12 @@
  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  */
 
-/* Runge-Kutta 8(9), Prince-Dormand */
+/* Runge-Kutta 8(7), Prince-Dormand
+ * The original article is :
+ *  High order embedded Runge-Kutta formulae
+ *  P.J. Prince and J.R. Dormand
+ *  Journal of Computational and Applied Mathematics, volume 7, no. 1, 1981
+ */
 
 /* Author:  G. Jungman
  */
@@ -80,14 +85,21 @@
 static const double b4[] = { 1.0 / 32.0, 0.0, 3.0 / 32.0 };
 static const double b5[] = { 5.0 / 16.0, 0.0, -75.0 / 64.0, 75.0 / 64.0 };
 static const double b6[] = { 3.0 / 80.0, 0.0, 0.0, 3.0 / 16.0, 3.0 / 20.0 };
+
+/* the b7[0] and b7[3] coefficients were corrected by Luc Maisonobe
+ * since the approximation given in the original article seemed to be
+ * wrong, because the sum b7[0]+b7[1]+ ... +b7[5] was not equal to
+ * ah[5] as it should be.
+ */
 static const double b7[] = {
-  29443841.0 / 614563906.0,
+  215595617.0 / 4500000000.0, /* was 29443841/614563906 in the article */
   0.0,
   0.0,
-  77736538.0 / 692538347.0,
+  202047683.0 / 1800000000.0, /* was 77736538/692538347 in the article */
   -28693883.0 / 1125000000.0,
   23124283.0 / 1800000000.0
 };
+
 static const double b8[] = {
   16016141.0 / 946692911.0,
   0.0,

Re: bug list summary

[Filip Floegel]

hello gsl guys,

maybe one should have a look into the matpack lib which is
a c++ library in order to fix the bugs below. as i could see
this lib is under GPL. (  http://www.matpack.de )
in the source code are references to the numerical algorithms
which are used.
The author may be contacted via:  gammel@matpack.de

---------------------------------------------------------------
> BUG#1 -- gsl_sf_hyperg_2F1_e fails for some arguments

here one could have a look at the Laguerre Polynomials of matpack,
i guess that there is a connection to hyperg_2F1 function.

i haven't checked it

> BUG#3 -- gsl_sf_bessel_Jn fails for large argument 

there is a Bessel function , first kind, order n, in matpack also

but this I haven't ckecked

> BUG#7 -- gsl_sf_coupling_3j loses accuracy for large arguments 

3J-symbol in the matpack lib works perfectly is my impression even
for very large angular momenta

i have ckecked it out

------------------------------------------------------------------

sifac,

       filip floegel
    
-- 
+---------------------------------------------------------------+
| dipl.phys. Filip Floegel   e-mail:floegel@itp.uni-hannover.de |               
| ITP/Universitaet Hannover  phone: 0511/762-3381               |               
| Appelstr.2                 fax: 0511/762-4945                 |               
| 30167 Hannover                                                |               
|                                                               |               
+---------------------------------------------------------------+

Re: bug list summary

[Brian Gough]

Luc Maisonobe writes:
 >  I will check this as time permit. For the moment, I propose the
 > following patch, which only add some comments (corected order of
 > the method, article reference) and correct two coefficients. I have
 > just written it now to answer this mail, not tried it (and even no
 > compiled it, so beware of typos ...)

Thanks.  I have applied the patch to the odeext-1-2 branch of gsl on
savannah.gnu.org.

Brian

Re: bug list summary

[Brian Gough]

Filip Floegel writes:
 > hello gsl guys,
 > 
 > maybe one should have a look into the matpack lib which is
 > a c++ library in order to fix the bugs below. as i could see
 > this lib is under GPL. (  http://www.matpack.de )
 > in the source code are references to the numerical algorithms
 > which are used.
 > The author may be contacted via: gammel@matpack.de

I contacted the author to check the license, some of the source files
have a different copyright notice (not GPL). I haven't heard back yet.

Brian

gsl compilation with gcc

[Gaurav Bansal]

Hello there,

I am new to this discussion group. I want to use GSL eigen value lib in my
C++ program. so i need to include the header file for eigen value.

Problem is that the header file is located in c:/programFiles/GSL
directory and my program is in other directory. Compiler cant find the
required files.

How can i make sure that the compiler will look at GSL library?

Thank you,
Gaurav

Re: gsl compilation with gcc

[Brian Gough]

Gaurav Bansal writes:
 > Hello there,
 > 
 > I am new to this discussion group. I want to use GSL eigen value lib in my
 > C++ program. so i need to include the header file for eigen value.
 > 
 > Problem is that the header file is located in c:/programFiles/GSL
 > directory and my program is in other directory. Compiler cant find the
 > required files.
 > 
 > How can i make sure that the compiler will look at GSL library?

The manual for your compiler should describe how to add directories to
its include path.

Re: gsl compilation with gcc

[Gaurav Bansal]

Hello Brian,

Thank you. I am able to work it with visual studio. I want to use the same
program with g++ on unix. how can i include header files there?

thank you once again.

regards,
Gaurav


On Fri, 18 Oct 2002, Brian Gough wrote:

> Gaurav Bansal writes:
>  > Hello there,
>  >
>  > I am new to this discussion group. I want to use GSL eigen value lib in my
>  > C++ program. so i need to include the header file for eigen value.
>  >
>  > Problem is that the header file is located in c:/programFiles/GSL
>  > directory and my program is in other directory. Compiler cant find the
>  > required files.
>  >
>  > How can i make sure that the compiler will look at GSL library?
>
> The manual for your compiler should describe how to add directories to
> its include path.
>
>

Re: gsl compilation with gcc

[Atakan Gurkan]

On Sun, Oct 20, 2002 at 02:11:36PM -0500, Gaurav Bansal wrote:
> Hello Brian,
> 
> Thank you. I am able to work it with visual studio. I want to use the same
> program with g++ on unix. how can i include header files there?
> 
> thank you once again.
> 
> regards,
> Gaurav
> 
> 
GCC comes with a manual. `man gcc' shows the command line options. `info gcc'
shows more information. also you can find the manual in fsf's pages:
http://gcc.gnu.org/onlinedocs/
have fun,

ato

Make File - VPATH usage problem using GSL eigen value program

[Gaurav Bansal]

Hello There,

I am using g++ compiler in UNIX.

I have a couple of .h and .c files that i am using from GNU-GSL library.
These files are stored in my program subdirectory called "gsl".

I am using VPATH variable in makefile to make sure that subdirectory is
checked for files.
I am also using -Igsl in the CFLAGS for compiler.

The program compiles but there is a problem with the linker. Do i have to
specify somewhere in make file for the linker to know that some files are
in "gsl" subdirectory.

Thank you ,
Gaurav

Re: Make File - VPATH usage problem using GSL eigen value program

[Joakim Hove]

Gaurav Bansal <gaurav@godot.urol.uic.edu> writes:


> The program compiles but there is a problem with the linker. Do i have to
> specify somewhere in make file for the linker to know that some files are
> in "gsl" subdirectory.

1. Post (relevant portions) of your makefile, so the people who might
   be able to help don't have to guess.

2. make -n prints to stdout what the make program intends to do,
   without actually doing it. This output might also be useful for
   debugging purposes.

Regards - Joakim

-- 
==== Joakim Hove      www.phys.ntnu.no/~hove/   =======================
|| Institutt for fysikk  (735) 93637 / E3-141  |  Skøyensgate 10D    ||
|| N - 7491 Trondheim    hove@phys.ntnu.no     |  N - 7030 Trondheim ||
================================================= 73 93 31 68 =========