erfc failure with large numbers

[John W. Eaton]

On 30-Oct-2008, Sergei Steshenko wrote:

| 
| 
| 
| --- On Thu, 10/30/08, Shea, Philip <pshea at dtri.net> wrote:
| 
| > From: Shea, Philip <pshea at dtri.net>
| > Subject: erfc failure with large numbers
| > To: bug at octave.org
| > Date: Thursday, October 30, 2008, 5:29 AM
| > I'm a little surprised that the library routine
| > can't handle large
| > numbers.  There are good approximations to the erfc for
| > large numbers
| > (see Proakis, John G., Digital Communications, pg. 40, eq.
| > (2-1-96)).  I
| > have attached the output of the octave bug_report function.
| >  
| > octave-3.0.1.exe:4> erfc(1000)
| >  ***MESSAGE FROM ROUTINE DERFC IN LIBRARY SLATEC.
| >  ***POTENTIALLY RECOVERABLE ERROR, PROG ABORTED, TRACEBACK
| > REQUESTED
| >  *  X SO BIG ERFC UNDERFLOWS
| >  *  ERROR NUMBER = 1
| >  *
| >  ***END OF MESSAGE
| >  
| >  ***JOB ABORT DUE TO UNRECOVERED ERROR.
| > 0          ERROR MESSAGE SUMMARY
| >  LIBRARY    SUBROUTINE MESSAGE START             NERR    
| > LEVEL
| > COUNT
| >  SLATEC     DERFC      X SO BIG ERFC UNDERF         1      
| >   1
| > 1
| >  
| >  
| > Philip A. Shea
| > DTRI
| > 1835 Alexander Bell Dr., Suite 400
| > Reston, VA 20191
| > 703-707-8555 x6270 (Office)
| > 877-383-8802 (Pager)
| >  
| 
| I can't reproduce this problem on my self-built Linux 3.0.3 'octave':
| 
| "
| octave:2> erfc(1000)
| ans = 0
| ".

Octave uses

  #if !defined (HAVE_ERFC)
  double
  erfc (double x)
  {
    double retval;
    F77_XFCN (xderfc, XDERFC, (x, retval));
    return retval;
  }
  #endif

So if configure detects that an erfc function is available in the C
library, it uses that instead of calling derfc from the slatec
library.  It's the function from slatec that needs to be fixed or
replaced with something that isn't broken.

jwe