Strong and weak conjectures on essential specializations and the discriminant ideal

Let II be the ideal under consideration, and let GG be its generic Gröbner basis. A specialization is essential when its reduced Gröbner basis is not the generic basis GG, and singular when it is a singular specialization. Let JJ and NN be the ideals defined in the paper, with NN the relevant intersection of the radical ideals NiN_i.

Strong and weak conjectures. The strong conjecture asserts that all essential specializations are singular. The weak conjecture asserts

JN.J\supseteq N.

The paper proves the reverse inclusion JNJ\subseteq N, so the weak conjecture would imply J=NJ=N. The supplied text does not give evidence that either conjecture has been resolved.

Sources & referencesView supporting material

Primary source

Montserrat Manubens and Antonio Montes, “Improving DISPGB Algorithm Using the Discriminant Ideal”, arXiv:math/0601763 (2006).

Progress summary

Refreshed
Open

No public discussion or published progress was found for this conjecture.

No public discussion or published progress was found for either the strong or weak conjecture.

Current status (as of August 2026): the conjectures appear open, with no recorded public activity.

Solutions 1

Counterexample

Both conjectures are false, even when the singular-specialization locus is nonempty and both discriminant ideals are proper.

The construction amplifies the one-parameter example in M. Wibmer, Gröbner bases for families of affine or projective schemes, Journal of Symbolic Computation 42 (2007), Example 3, https://arxiv.org/abs/math/0608019. A related example was also recorded by A. Montes, On the canonical discussion of polynomial systems with parameters, https://arxiv.org/abs/math/0601674.

Let

R=Q[u,v]R=\mathbb Q[u,v]

and consider

I=vu(ux1),vx(ux1)=v(ux1)u,xR[x].I=\langle vu(ux-1),\,vx(ux-1)\rangle =v(ux-1)\langle u,x\rangle \subseteq R[x].

Over Q(u,v)\mathbb Q(u,v), the generic reduced Gröbner basis is

G={g},g=x1u,G=\{g\},\qquad g=x-\frac1u,

with leading-power-product set {x}\{x\}.

For (u,v)=(c,b)(u,v)=(c,b), the specialized ideal is

Ic,b={x1/c,b0, c0,x,b0, c=0,(0),b=0.I_{c,b}= \begin{cases} \langle x-1/c\rangle,&b\neq0,\ c\neq0,\\ \langle x\rangle,&b\neq0,\ c=0,\\ (0),&b=0. \end{cases}

Under the original definition, a specialization is singular precisely when its leading-power-product set differs from the generic set. Therefore the singular locus is exactly V(v)V(v), and hence

N=I(V(v))=(v).(1)N=I(V(v))=(v). \tag{1}

On the other hand,

Jg={aR:agI}.J_g=\{a\in R:ag\in I\}.

If agR[x]ag\in R[x], its constant coefficient a/u-a/u forces a=uha=uh for some hRh\in R. Cancelling ux1ux-1 in the integral domain R[x]R[x] gives

ag=h(ux1)v(ux1)(u,x)hv(u,x)R=(uv).ag=h(ux-1)\in v(ux-1)(u,x) \quad\Longleftrightarrow\quad h\in v(u,x)\cap R=(uv).

Thus

Jg=(u2v),J=Jg=(uv).(2)J_g=(u^2v), \qquad J=\sqrt{J_g}=(uv). \tag{2}

At the point (u,v)=(0,1)(u,v)=(0,1),

Jg(u,v1),J_g\subseteq(u,v-1),

so the specialization is essential. Nevertheless its specialized ideal is x\langle x\rangle, whose leading-power-product set is still {x}\{x\}; therefore it is not singular. This disproves the strong conjecture.

Finally, (1) and (2) give

J=(uv)(v)=N,J=(uv)\subsetneq(v)=N,

so vNJv\in N\setminus J. Hence the conjectured inclusion JNJ\supseteq N also fails, disproving the weak conjecture.

0 endorsements
Shivam Patel ·