An explicit counterexample to three-summand Kato-kernel splitting
ORCID: 0009-0001-9320-500X · Zenodo record and DOI
Combining field extensions can kill a cohomology class even when the class cannot be assembled from the kernels of the individual extensions. This note gives an explicit characteristic-two example where that obstruction survives the addition of a third, bilinear Pfister function-field kernel.
Read the paper
Open the original PDF · Download the PDF · BibTeX
The PDF prints 17 October 2026 without a numeric version. Version 1.0.0 is the Zenodo record’s version field. The record declares that publication date and was created on 18 October 2026 UTC.
A class outside all three kernels together
Write Kn(E/F) for the kernel of the restriction map H2n+1(F) → H2n+1(E). Laghribi’s Question 2 in Oberwolfach Report 5/2016 asks whether the kernel for the composite is always the sum
Kn(LK(B)/F) = Kn(L/F) + Kn(K/F) + Kn(F(B)/F).
Here L/F is finite purely inseparable, K/F is separable quadratic, and B is an anisotropic bilinear Pfister form that remains anisotropic over L.
For the counterexample, let b, z, s and t be algebraically independent over the two-element field and set
F = 𝔽2(b,z,s,t), L = F(β), β4 = b,
K = F(θ), θ2 + θ = s−1, B = ⟨1,t⟩b.
The extension L/F has degree four and exponent two. The form B remains anisotropic over L. At n = 1, consider the class ξ in H22(F) represented by
s−4 · bz2/(1 + bz2) · dlog z.
The class ξ dies over LK, and hence over LK(B), but does not belong to K1(L/F) + K1(K/F) + K1(F(B)/F). This is a counterexample to the proposed general equality at its allowed original index n = 1. It does not assert failure at every index.
What the binary function field means
F(B) is the function field of the projective quasilinear quadric defined by v ↦ B(v,v). This convention includes binary forms: for B = ⟨1,t⟩b, the equation X2 + tY2 = 0 in the projective line defines an integral, zero-dimensional quadric. Its function field is
F(B) = F(γ), γ2 = t.
The example therefore uses the binary case allowed by the original question. The independent variable t matters: using z in that Pfister slot instead would kill dz and absorb the displayed class.
Why the third summand does not remove the obstruction
Under the identification H22(E) ≅ Br(E)[2], write [u,v) for the characteristic-two quaternion class. Over L, put f = 1 + β2z. The cohomological relations give ξL = [s−1,f), which vanishes after adjoining θ.
Suppose ξ were a sum of classes from the three individual kernels. Every class split by K/F has the form [s−1,c) for some arbitrary rational element c ∈ F×. Restrict to M = LF(B) = 𝔽2(β,z,s,γ). Both the L-kernel term and the Pfister-kernel term vanish there. The proposed decomposition would force f/c to be a norm from M(θ) to M.
Complete at s, with k′ = 𝔽2(β,z,γ) and D = k′((s)). For π = θ−1, the norm formulas are
N(π) = s, N(A + Bπ) = A2 + sAB + sB2.
The two possible leading valuations have opposite parity; the mixed term has larger valuation. Consequently every unit norm has square residue in k′. For any rational c = src0(1 + O(s)), with r any integer and c0 ∈ 𝔽2(b,z,t)×, multiplication by the norm sr forces
(1 + β2z)/c0 ∈ k′×2.
This square condition is impossible. Set R = 𝔽2(β4,z2,γ2) and S = k′2. Every such c0 has a unique expression u + vz with u,v ∈ R, including arbitrary rational denominators. Since z ∉ S, comparing coefficients in 1 + β2z = q(u + vz), q ∈ S×, gives qu = 1 and qv = β2. Thus β2 would lie in R, contradicting its odd valuation against the even valuations of nonzero elements of R.
The obstruction covers every rational parameter c, including negative s-valuations. It is a universal norm-and-residue argument, rather than a finite test of selected polynomials.
The degree-one boundary is positive
The same proposed equality holds at n = 0, in H21, for all of the stated data with B a positive-fold bilinear Pfister form. The composite kernel is exactly the K/F kernel; the purely inseparable and Pfister kernels are zero. Artin–Schreier roots cannot first appear in the purely inseparable or rational extensions used in the projective-quadric description.
The negative result is therefore a specific n = 1, exponent-two example. The positive exponent-one regime is treated in the earlier work of Aravire, Laghribi and O’Ryan.
Prior work and scope
The mixed class and its vanishing mechanism are a specialization of Aravire–Laghribi–O’Ryan’s Definition 1.3 and Lemma 3.1 in their 15 August 2019 author manuscript, published in Journal of Algebra 542 (2020). This note gives a direct proof that the class remains outside the sum after the additional independent Pfister kernel is included.
No firstness claim is made for the class, example or deduction. Bounded source comparisons do not certify historical priority, causal independence or continued worldwide openness. A final comparison with the mixed paper’s typeset body and access to adjacent nonnormal-extension and transfer papers remain limits on the historical assessment.
Original manuscript abstract
Question 2 in Laghribi’s contribution to Oberwolfach Report 5/2016 asks whether a mixed Kato kernel is the sum of the kernels of a finite purely inseparable extension, a separable quadratic extension, and a bilinear Pfister function field. We give an explicit negative instance in H₂², using an exponent-two extension and a binary bilinear Pfister form that remains anisotropic. The mixed class is exactly a specialization of the generators of Aravire–Laghribi–O’Ryan (2019/2020). We prove directly that it remains outside the sum after the additional Pfister kernel is included, by a universal norm and residue obstruction. The original degree-one cohomology case is separately positive. No firstness claim is made.
Original files and citation
The files are provided in the native record’s entries order:
- mixed-kato-counterexample.pdf (71,463 bytes) · Zenodo original PDF
- mixed-kato-counterexample-source-and-verification.zip (372,851 bytes) · Zenodo original ZIP
Alec Kriebel. An explicit counterexample to three-summand Kato-kernel splitting. Native version 1.0.0, 17 October 2026. doi:10.5281/zenodo.23444050. Download BibTeX.
The authored paper, code and analysis are licensed under CC BY 4.0. Third-party material retains its own rights.
Status and collaboration
AI tools were used extensively in developing, writing and checking this work. The preprint is unrefereed; internal AI-assisted review is not external human peer review, journal acceptance or formal proof-assistant certification. The analytic proof is in the paper and does not depend on running the supporting programs.
Independent checking, substantive mathematical extensions and development toward a journal submission would be useful contributions. Collaboration and any authorship would reflect actual scholarly contributions.
Selected references
- A. Laghribi, joint with R. Aravire and M. O’Ryan, Differential and Quadratic Forms in Characteristic Two, Oberwolfach Report 5/2016, 248–251. Question 2, p. 251. Publisher DOI.
- R. Aravire, A. Laghribi and M. O’Ryan, Graded Witt kernels of the compositum of multiquadratic extensions with the function fields of Pfister forms, Journal of Algebra 449 (2016), 635–659. Publisher DOI.
- R. Aravire, A. Laghribi and M. O’Ryan, Cohomological kernels of mixed extensions in characteristic 2, Journal of Algebra 542 (2020), 249–276. Definition and lemma numbering follows the 15 August 2019 author manuscript. Publisher DOI · Author manuscript.
- A. Chapman and A. Quéguiner-Mathieu, Minimal quadratic forms for the function field of a conic in characteristic 2, arXiv:2111.15251v3 (2022). Versioned preprint.
- J. S. Milne, Class Field Theory, version 4.03 (2020), Chapter IV. Author notes.
- J. Voight, Quaternion Algebras, Springer (2021), Chapter 6. Publisher DOI.