Igor Schein on Thu, 9 Oct 2003 21:37:29 -0400


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: polredabs failure


On Thu, Oct 09, 2003 at 09:32:35PM +0200, Karim BELABAS wrote:
> On Thu, 9 Oct 2003, Igor Schein wrote:
> > On Thu, Oct 09, 2003 at 09:04:42PM +0200, Karim BELABAS wrote:
> >> A complicated typo. The smallest element was found, then deleted due to a typo
> >> which assessed to it the norm of another (larger) element, during the
> >> "Sorting" phase (when there are too many small vectors, the list overflows,
> >> and we need to discard some of them). Fixed in CVS.
> >
> > Immediate regression:
> >
> > ? allocatemem()
> >   ***   Warning: doubling stack size; new stack = 8000000 (7.629 Mbytes).
> > ? polredabs(x^16-4*x^15+16*x^13-48*x^11+16*x^10-32*x^9+360*x^8-616*x^7+352*x^6-112*x^3+96*x^2-32*x+4)
> >   ***   bug in GP (Segmentation Fault), please report
> 
> The (delicate) old GC scheme did not work due to my slightly careless
> modification. Fixed.

Thanks.

Now, continuing the same topic:

%1 = x^16 - 4*x^15 + 8*x^14 - 8*x^13 + 6*x^12 - 16*x^11 + 40*x^10 - 32*x^9 - 50*x^8 + 160*x^7 - 176*x^6 + 40*x^5 + 140*x^4 - 192*x^3 + 112*x^2 - 32*x + 4
? polredabs(%)
%2 = x^16 - 4*x^15 + 12*x^14 - 36*x^13 + 88*x^12 - 164*x^11 + 252*x^10 - 324*x^9 + 354*x^8 - 324*x^7 + 252*x^6 - 164*x^5 + 88*x^4 - 36*x^3 + 12*x^2 - 4*x + 1
? polredabs(%)
%3 = x^16 - 8*x^15 + 36*x^14 - 108*x^13 + 244*x^12 - 436*x^11 + 636*x^10 - 780*x^9 + 831*x^8 - 780*x^7 + 636*x^6 - 436*x^5 + 244*x^4 - 108*x^3 + 36*x^2 - 8*x + 1
\\ Takes 2 iterations to stabilize; Also...
? #polredabs(%1,4)
%4 = 13
? #polredabs(%2,4)
%5 = 22
? #polredabs(%3,4)
%6 = 24

What is causing discrepancy between the last 3 results, precision
issues or probabilistic nature of algorithm?

Thanks

Igor