An elementary counterexample to a Nullstellensatz for finite-type link invariants
Alec Kriebel · ORCID 0009-0001-9320-500X
12 October 2026 · Zenodo version 1.0 · Unrefereed preprint
The manuscript is dated 12 October 2026 and prints no version number; 1.0 is the Zenodo version.
Original abstract
Let V be the ordinary, uncompleted complex algebra of all finite-order finite-type invariants of links. We give a principal ideal with a nonempty actual-link zero set whose vanishing ideal strictly contains its radical. The generator is g(4g-1), where g is the sum of squared pairwise linking numbers. The obstruction is a character on the entire algebra, obtained by evaluating the standard twist-restriction polynomial at a nonintegral parameter. This gives a negative answer to the arbitrary-ideal formulation of Bar-Natan's Problem 2.27, as recorded by Ohtsuki; the separate linking-number-generated ideal question is not decided.
What the result means
A link invariant assigns a number to a collection of intertwined circles. Finite-type invariants form an algebra: they can be added and multiplied. One can ask whether knowing which actual links make an ideal vanish determines that ideal up to taking powers, as in a Nullstellensatz.
This note gives an explicit negative answer to the arbitrary-ideal, actual-link formulation. Its ideal has plenty of actual zeros, including the unlink, yet an invariant vanishing on all those zeros has no positive power in the ideal. The distinction comes from an algebra character that no actual link realizes; it does not contradict the classical Hilbert Nullstellensatz.
The full result
For every integer m≥2, let X_m be the oriented, unordered m-component links in S³, including split links. Let V_m be their ordinary, uncompleted unital complex algebra of finite-type invariants, with pointwise operations. Order at most d means that the positive-minus-negative crossing extension vanishes on every link with d+1 double points. Each invariant has some finite order; there is no common order cutoff on the whole algebra and no completion by infinite-order series.
For an ideal A, Z_act(A) means its common zero set among actual links, and J_act(S) means all invariants vanishing on S. The radical rad(A) consists of invariants that have a positive power in A. The always-valid containment is rad(A)⊆J_act(Z_act(A)).
Theorem 1. Set g(L)=Σ_{i<j}lk(L_i,L_j)², f=g(4g−1), and A=fV_m. Then Z_act(A) is exactly the nonempty set of algebraically split links: every pairwise linking number is zero. Moreover g∈J_act(Z_act(A)) but g∉rad(A), so the containment is strict. The unlink belongs to the zero set.
The same construction works in the algebra V on all finite-component links, where each invariant has one finite order valid across that whole domain. The finite-type orders are at most two for g and four for f, uniformly in the number of components. The global argument restricts each such invariant to one fixed-component family; it does not assume that every fixed-component invariant extends to all links.
The construction also works for ordered or labelled links, after forgetting orientations by pullback, and over every characteristic-zero coefficient field. The linking-square expression is unchanged by component permutations or orientation reversals. Its actual integer values are considered before embedding into the coefficient field; no ordering of that field is assumed.
The proof companion also records the integer-valued ring version: restriction gives rational Newton polynomials and a ring homomorphism V_Z→Q, yielding the same obstruction. This is not an arbitrary coefficient-ring or positive-characteristic assertion.
Why the obstruction works
On a crossing-resolution cube, an order-d finite-type invariant has a multilinear representative of degree at most d. Products of orders d and e therefore have order at most d+e. Each pairwise linking number is affine on such a cube, so g has order at most two and f has order at most four.
For an actual link, g is a nonnegative integer, being a sum of squares of integer linking numbers. Thus g(4g−1)=0 exactly when g=0, which is exactly when every pairwise linking number vanishes. These are actual integer linking numbers, not arbitrary complex coordinates. Because the algebra contains the unit and A is generated by f, zeros of f are precisely the common zeros of A.
Let L_n, for every integer n≥0, be the coherently oriented closure of the two-strand braid σ^(2n), with m−2 split unknots added. It has m components and g(L_n)=n². Changing a designated crossing in each added two-crossing block either retains that full twist or cancels the block. The resulting signed resolution cube is the forward difference Δ^r[v(L_n)]. This is a crossing-change argument, not smoothing; every resolution keeps the same component count.
If v has order at most d, then Δ^(d+1)[v(L_n)]=0 for every n≥0. Newton's difference formula gives a unique polynomial P_v(t) of degree at most d agreeing with v(L_n) at all nonnegative integers. Agreement on infinitely many integers shows P_(uv)=P_uP_v and preserves sums and the unit. This constructs a homomorphism on the entire finite-order algebra, rather than assuming a character on C[g] extends to a larger ring.
Evaluate that polynomial formally at one half: χ(v)=P_v(1/2). This is a character on the whole algebra, with χ(g)=1/4 and χ(f)=0. For every N≥1 and every finite-order multiplier h, an equation g^N=fh would imply 4^(−N)=0, which is impossible in characteristic zero. Thus no positive power of g lies in A. The proof has no upper bound on N or on the order of h.
The same contradiction can be read as polynomial divisibility: t^(2N) cannot equal t²(4t²−1)P_h(t), because evaluation at t=1/2 makes the right side zero and the left side nonzero. The finite-order polynomial argument proves the universal statement; the accompanying exact-rational checks support its formulas and conventions.
Formal evaluation at one half is not evaluation on an actual link, including the one-component closure of σ. Actual-link values of g are integers, whereas this character gives 1/4. The counterexample concerns using actual links as the available evaluation points; it is not a failure of the classical Hilbert Nullstellensatz on its usual algebraic spectrum.
Scope and limitations
The principal ideal A=fV_m is different from the ideal generated by all indexed pairwise linking numbers. That indexed ideal belongs to the ordered or labelled theory, where individual component pairs are distinguished. In that theory the two ideals have the same actual zero set, but g already belongs to the indexed linking-number ideal, so this witness does not decide its radical/vanishing-ideal question.
Algebraically split means that all pairwise linking numbers vanish. It does not mean that the link is geometrically split. The theorem includes split links and does not impose a non-split hypothesis.
No conclusion about completed invariant algebras, homogeneous-ideal variants, positive characteristic, or a knot-only counterexample is asserted. With only one component, g is identically zero, so the displayed witness does not give a knot-only result. Replacing actual links by all algebra characters would change the question.
Polynomiality along twist and braid families is established prior work. The contribution recorded here is its elementary application to the explicit principal ideal and nonempty actual-link zero set. Historical firstness, independent discovery, an exhaustive absence of predecessors and present worldwide openness are not established.
Background and credit
Bar-Natan's Problem 2.27, recorded by Ohtsuki on printed page 422, asks the arbitrary-ideal question using actual links as zeros. The collection has volume year 2002 and was published in 2004. Its separate indexed linking-number example remains distinct from the principal-ideal counterexample here.
The polynomiality input is classical. Stanford–Trapp's Theorem E treats full-twist differences; Eisermann develops finite-type polynomiality and the Newton interpretation. Stoimenow's 2010 Theorem 2.1 and Definitions 2.2–2.3 provide an earlier braid-sequence interface and explicitly credit earlier braiding-sequence work. No new polynomiality method is claimed.
Stoimenow's theorem is read with its local odd-exponent and default non-split conventions. One fixed crossing together with local exponent 2n−1 gives even total exponent 2n. The manuscript's direct singular-link argument independently handles the stated full-twist family and all split components. Jeong–Park's connected-sum polynomiality is a related older route, not silently identified with a varying-linking family.
The source comparison supports these specific classical interfaces, including the exact 2010 theorem and its conventions. It does not certify every earlier or final-version proof, exhaustive literature coverage, or the absence of an earlier equivalent ideal counterexample. These limits do not replace the self-contained all-order proof.
Contribution and collaboration
Independent verification, a well-supported primary-source comparison or substantive work toward a journal submission would be welcome. Authorship would reflect actual contributions.
AI and review disclosure
AI tools were used extensively in mathematical development, proof checking, source research, computations and writing. This is an unrefereed preprint. Internal AI checks are not conventional human peer review, and no formal proof-assistant certification is claimed.
Read and download
Read the complete unchanged four-page paper below, or open or download the original PDF. The complete original ZIP contains manuscript source and authored proof and verification materials. The original standalone TeX is also provided separately. The README documents optional exact finite checks; the universal claims rest on the written proof.
Open original PDF · Download PDF · BibTeX · Zenodo record and DOI
Citation
Alec Kriebel. An elementary counterexample to a Nullstellensatz for finite-type link invariants. Zenodo version 1.0, 12 October 2026. doi:10.5281/zenodo.23331295.
Primary background
- Ohtsuki's problem collection — Bar-Natan's Problem 2.27, printed page 422
- Stanford–Trapp, actual 1999 preprint — Theorem E
- Eisermann, author-revised 2003 preprint — finite-type polynomiality
- Stoimenow, 2010 — Theorem 2.1 and Definitions 2.2–2.3
- Jeong–Park, 2002 — the related connected-sum polynomiality route
Rights
The original paper, TeX and authored verification materials, including code, are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0), as stated in the Zenodo metadata and deposited README. Cited third-party works retain their own rights and their full texts are not included as authored source bodies.