ggnfs lattice siever misses some primes

[fivemack]

We have f(a,b) = 1361*a^6-b^6, g(a,b) = a-1361^7*b

The siever in GGNFS outputs the line

417266,227287:22061,2e64f,e9c75,6f1,dde715:13b08b,d8d,1273,4eb5,4c1,b732f,cf0671

and msieve relation-parsing gives an error -15 (wrong factors)

Converting from hex, the line says
rational-side factors 139361, 190031, 957557, 1777, 14542613
algebraic-side factors 1290379, 3469, 4723, 20149, 1217, 750383, 13567601

gp says

? factor(A-1361^7*B)
%5 = 
[-1 1]
[3 1]
[1777 1]
[139361 1]
[190031 1]
[957557 1]
[14542613 1]
? factor(1361*A^6-B^6)
%3 = 
[1217 1]
[1361 1]
[3469 1]
[4723 1]
[20149 1]
[750383 1]
[1290379 1]
[13567601 1]

That is, GGNFS is not reporting the factor 1361 in the algebraic side. I can see why 1361 is special, it's a non-tiny prime dividing the leading coefficient and so possibly not in the factor base, but I don't see why it makes sense to miss it out in the output.

This isn't a serious issue, it cost about 0.5% of the relations on a small calculation, but I tend to want to diagnose all -15 relation-reading messages, given what a mess a wrong factorisation could cause in the matrix stage.

[R.D. Silverman]

Originally Posted by fivemack

We have f(a,b) = 1361*a^6-b^6, g(a,b) = a-1361^7*b

The siever in GGNFS outputs the line

417266,227287:22061,2e64f,e9c75,6f1,dde715:13b08b,d8d,1273,4eb5,4c1,b732f,cf0671

and msieve relation-parsing gives an error -15 (wrong factors)

Converting from hex, the line says
rational-side factors 139361, 190031, 957557, 1777, 14542613
algebraic-side factors 1290379, 3469, 4723, 20149, 1217, 750383, 13567601

gp says

? factor(A-1361^7*B)
%5 = 
[-1 1]
[3 1]
[1777 1]
[139361 1]
[190031 1]
[957557 1]
[14542613 1]
? factor(1361*A^6-B^6)
%3 = 
[1217 1]
[1361 1]
[3469 1]
[4723 1]
[20149 1]
[750383 1]
[1290379 1]
[13567601 1]

That is, GGNFS is not reporting the factor 1361 in the algebraic side. I can see why 1361 is special, it's a non-tiny prime dividing the leading coefficient and so possibly not in the factor base, but I don't see why it makes sense to miss it out in the output.

This isn't a serious issue, it cost about 0.5% of the relations on a small calculation, but I tend to want to diagnose all -15 relation-reading messages, given what a mess a wrong factorisation could cause in the matrix stage.

The code is probably not handling projective primes properly. Or perhaps
not at all.