Problems using dsplp (from SLATEC-library)
Olaf Poppe
for my purposes.
But I can't even get the tiny example running that comes with it (called
lpex).
I cut and pasted the example, so there should be no mistasake in the
code.
Everything compiles just fine (I am using g77). But the program returns
the following error:
***MESSAGE FROM ROUTINE DSPLP IN LIBRARY SLATEC.
***POTENTIALLY RECOVERABLE ERROR, PROG ABORTED, TRACEBACK REQUESTED
* THE USER OPTION ARRAY HAS UNDEFINED DATA.
* ERROR NUMBER = 14
*
***END OF MESSAGE
***JOB ABORT DUE TO UNRECOVERED ERROR.
0 ERROR MESSAGE SUMMARY
LIBRARY SUBROUTINE MESSAGE START NERR LEVEL
COUNT
SLATEC DSPLP THE USER OPTION ARRA 14 1
1
This error number "14" is somehow linked to the definition of the user
option vector PRGOPT.
But, since I am using the sample code and the option used in there is
conform with the description:
C PRGOPT(*)
C ---------
C This array is used to redefine various parameters within DSPLP( ).
C Frequently, perhaps most of the time, a user will be satisfied
C and obtain the solutions with no changes to any of these
C parameters. To try this, simply set PRGOPT(1)=1.D0.
I don't see, what I did wrong. PRGOPT has dimension 1. And PRGOPT(1)
gets a value assigned to it.
So, I don't understand why this array should have "UNDEFINED DATA".
I am certain that this library routine is correct but I simply cannot
find the mistake I must have made..
So I would be glad if someone of you could tell me, what I did wrong...
Thanks!
Olaf Poppe
------------
olaf....@iw.uni-halle.de
Hans D. Mittelmann
several years ago I made progress on getting this to run, but there was
then a numerical instability stemming apparently from a Harwell
subroutine. I would suggest to use a newer LP solver. See, for example,
http://plato.la.asu.edu/guide.html#LP
http://plato.la.asu.edu/bench.html
Hans Mittelmann
--
Hans D. Mittelmann http://plato.la.asu.edu/
Arizona State University Phone: (602) 965-6595
Department of Mathematics Fax: (602) 965-0461
Tempe, AZ 85287-1804 email: mitte...@asu.edu
James R. Van Zandt
Olaf Poppe <olaf....@iw.uni-halle.de> wrote:
>I am trying to use the routine dsplp (LP-solver) from the SLATEC-library
>for my purposes.
For the architectures I use (i386 and MIPS), I found it necessary to
substitute a newer version of the customization subroutine, r1mach.f.
With that, I got SPLP (the single precision version) working.
However, DSPLP would not work. I have not managed to track down the
problem. If you get it working, please let me know.
If an interior point method can serve, I suggest you check out PCx at
http://www-c.mcs.anl.gov/home/otc/Guide/SoftwareGuide/Blurbs/PCx.html
If you have a Debian Linux system, you can try the Debian package
(which I maintain):
/debian/dists/stable/non-free/binary-i386/math/pcx_1.1-1.deb
- Jim Van Zandt
Peter Spellucci
j...@mitre.org (James R. Van Zandt) writes:
|> In article <36F2B94B...@iw.uni-halle.de>,
|> Olaf Poppe <olaf....@iw.uni-halle.de> wrote:
|> >I am trying to use the routine dsplp (LP-solver) from the SLATEC-library
|> >for my purposes.
|>
|> For the architectures I use (i386 and MIPS), I found it necessary to
|> However, DSPLP would not work. I have not managed to track down the
|> problem. If you get it working, please let me know.
|>
do this is obvious from the comments there. simply set the four values
to those you know for your architecture and throw away these seemingly
machine independent octal notations
there is a further trouble with the example file and the driver.
you must include the common areas there and declare anything as double
use implicit double precision (a-h,o-z). make sure that anything is
compatible (check with lintfor e.g.) in type. i got it running this way.
may send it by email if required.
hope this helps
peter