tracking normal to find out where/why it is hanging

leonada · Mon Dec 08, 2014 9:36 pm

LIB "normal.lib";
ring r=2,(y,x),wp(37,31);
ideal i=y31+y12x17+x37;
list nor=normal(i);nor;

has been running for maybe 200 hours (3.7 GB),
on this problem that can almost be done by hand.
It takes my QthPower algorithm maybe 9 seconds in M2.
How can I rerun this and trace what is taking so much time?
[I assume it is a kernel computation, though it could be a GB computation.]

Re: tracking normal to find out where/why it is hanging

hannes · Tue Dec 09, 2014 11:45 am

To find out where it is hanging: press ctrl-c and you will be asked

Code:

// ** Interrupt at cmd:`$INVALID$` in line:'        iwork=phi(iwork);'
abort after this command(a), abort immediately(r), print backtrace(b), continue(c) or quit Singular(q) ?

Answering with b presents the backtrace.
The problem here is that the routine simplifyIdeal from primdec.lib
does not work in char 2.
To fix it, substitute simplifyIdeal by

Code:

static proc simplifyIdeal(ideal i)
{
  ASSUME(1, hasFieldCoefficient(basering) );
  ASSUME(1, not isQuotientRing(basering) ) ;
  ASSUME(1, hasGlobalOrdering(basering) ) ;

  def r=basering;

  int j,k;
  map phi;
  poly p;

  ideal iwork=i;
  ideal imap1=maxideal(1);
  ideal imap2=maxideal(1);

  if (char(r)!=2)
  {
  for(j=1;j<=nvars(basering);j++)
  {
    for(k=1;k<=ncols(i);k++)
    {
      if(deg(iwork[k]/var(j))==0)
      {
        p=-1/leadcoef(iwork[k]/var(j))*iwork[k];
        imap1[j]=p+2*var(j);
        phi=r,imap1;
        iwork=phi(iwork);
        iwork=subst(iwork,var(j),0);
        iwork[k]=var(j);
        imap1=maxideal(1);
        imap2[j]=-p;
        break;
      }
    }
  }
  }
  return(iwork,imap2);
}

Re: tracking normal to find out where/why it is hanging

leonada · Tue Dec 09, 2014 5:00 pm

Well I tried ctrl-c,
then ctrl-c ctrl-c gives me only the choices
(a)bort current command, (q)uit, (r)estart Singular or (c)ontinue?

Then I tried this in Singular instead of ESingular to finally get the print backtrace(b) option.

But this seems to be characteristic independent.

ring r=0,(y,x),dp;
ideal i=y31+y12x17-2y6x27+x37;
list nor=normal(i);nor;

seems to hang at the same command

iwork=phi(iwork)

Re: tracking normal to find out where/why it is hanging

hannes · Thu Dec 11, 2014 5:59 pm

True, the problem was that the simplification was using larger and larger maps
which led to huge computation times.
A re-ordering of the variables to consider for simplifications helps
(independend of characateristic).
Get a new primdec.lib from
[url]
https://github.com/Singular/Sources/blo ... rimdec.lib
[/url]

Re: tracking normal to find out where/why it is hanging

leonada · Tue Dec 30, 2014 9:08 pm

Is there now a bug in primdec.lib with something such as changeordTo?
Try the example for given integralBasis to see if you get the same error I did.

Re: tracking normal to find out where/why it is hanging

hannes · Wed Dec 31, 2014 1:23 pm

The example given above does not lead to an error.
But the development continues: now the new primdec.lib requires also
the new version of ring.lib for changeordTo.