gpt 6 astra pro attempt on Math-230

Serre's Conjecture II on Galois Cohomology of Simply Connected Semisimple Groups

Definitions and mathematical statement

Let $K$ be perfect, so every algebraic extension is separable, and write $\Gamma_K=\operatorname{Gal}(K^{\mathrm{sep}}/K)$. Cohomological dimension at most two means vanishing of continuous Galois cohomology in degrees greater than two for all discrete torsion coefficient modules. For a semisimple simply connected algebraic $K$-group $G$, the pointed set $H^1(K,G)$ classifies $G$-torsors. Such a torsor becomes isomorphic to $G$ after a separable extension and is trivial exactly when it has a $K$-point. The assertion is \[ H^1(K,G)=\{1\} \] for every pair $K,G$ in this scope. Simply connected is the algebraic-group notion of having no nontrivial central isogeny cover, not topological simple-connectedness of a set of real points.

Formulation and concept references: [S1].

Short English statement

Over a perfect field of cohomological dimension at most two, must every torsor under a semisimple simply connected algebraic group have a rational point?

Research attempt

Current frontier and bibliographic correction.

The URL in [S1] points to a different work from the Gille survey named there: it is Izquierdo--Lucchini Arteche's Transfer principles for Galois cohomology and Serre's conjecture II, whose third version is dated 9 September 2025. Among its consequences, the characteristic-zero conjecture implies the positive-characteristic conjecture; it does not prove the characteristic-zero premise. [E1], Introduction and Section 4.3. Gille's actual survey is separately identified in [E2]. Its classical-group and quasi-split exceptional results are distinguished from the unrestricted assertion. The March 2026 manuscript of Nguyen extends the formulation to certain pseudo-reductive groups and proves an equivalence, together with local and global-field cases, not the general premise. [E4], Abstract. These observations are a source audit, not changes to the original perfect-field target.

Attack route.

A torsor can be attacked through an explicit norm equation, through cohomological invariants, or by transfer from a smaller class of fields. The first route gives a complete calculation for quaternion norm-one groups. It reveals exactly why a degree-three invariant is sufficient there. We then try to transplant that mechanism to a general simply connected group, keeping the kernel problem visible. Transfer principles cannot repair a missing invariant-detection theorem merely by changing characteristic. The lower-rank argument below is a rederivation of a known subcase, not an assertion of a newly solved family.

Attempt.

1. What the split cases actually use. For every field $K$, the exact sequence \[ 1\longrightarrow\mathrm{SL}_r\longrightarrow\mathrm{GL}_r \xrightarrow{\det}\mathbf G_m\longrightarrow1 \] and Hilbert's theorem 90 for $\mathrm{GL}_r$ identify $H^1(K,\mathrm{SL}_r)$ with the cokernel of $\det:\mathrm{GL}_r(K)\to K^*$. The diagonal matrix $\operatorname{diag}(a,1,\ldots,1)$ shows that this map is surjective. Thus this family has trivial torsors without any bound on cohomological dimension. Likewise a split symplectic torsor is a nondegenerate alternating form on a vector space; choosing a vector, a partner with pairing one, and their orthogonal complement inductively constructs a symplectic basis. These are explicit split-group arguments. They do not trivialize a nonsplit form of the group. The cohomological framework and its norm-one interpretation are recalled in [E2], Sections 1 and 3.

For a central simple algebra $D$, the corresponding exact sequence is \[ 1\longrightarrow\mathrm{SL}_1(D)\longrightarrow\mathrm{GL}_1(D) \xrightarrow{\operatorname{Nrd}}\mathbf G_m\longrightarrow1. \] Noncommutative Hilbert 90 gives \begin{equation}\label{wsref-0-33-30-30-30-36-37-39-39-3a-6e-6f-72-6d-74-6f-72-73-6f-72-73} H^1(K,\mathrm{SL}_1(D))=K^*/\operatorname{Nrd}(D^*). \end{equation} This is a genuine reduction to surjectivity of a norm, not to the vanishing of the Brauer class of $D$. A division algebra can remain nonsplit even when every torsor under its norm-one group is trivial.

2. Quaternion norm surjectivity from a dimension gap. Assume now $\operatorname{char}K\ne2$ and $\operatorname{cd}(K)\le2$. Let $D=(u,v)_K$ be a quaternion algebra. Its reduced norm is the four-dimensional Pfister form \[ q=\langle1,-u,-v,uv\rangle. \] If $D$ is split, the norm is the determinant on $M_2(K)$ and is already surjective on $K^*$. Suppose $D$ is a division algebra; then $q$ is anisotropic. For any $a\in K^*$ consider the eight-dimensional form \[ \pi=q\otimes\langle1,-a\rangle=q\perp(-a q). \] Let $I$ be the fundamental ideal of the Witt ring. The established Milnor isomorphisms identify $I^3/I^4$ with $H^3(K,{\mathbb{Z}}/2)$, and the class of $\pi$ maps to the symbol $(u)\cup(v)\cup(a)$. This cohomology group is zero, so $[\pi]\in I^4$. The Arason--Pfister dimension theorem says that an anisotropic form with nonzero class in $I^4$ has dimension at least $16$. The anisotropic part of $\pi$ has dimension at most $8$, so it must vanish. Both deep inputs, including the characteristic-not-two convention, are stated in [E3], Section 4, Theorem 4.1 and the Milnor-isomorphism discussion. They are used here as theorems, not proved by this notebook.

In particular $\pi$ is isotropic. There are $x,y\in D$, not both zero, with $q(x)=a q(y)$. Anisotropy of $q$ forces $x,y\ne0$; in a division algebra $y$ is invertible. Multiplicativity of reduced norm gives \begin{equation}\label{wsref-0-33-30-30-30-36-37-39-39-3a-72-61-74-69-6f} \operatorname{Nrd}(xy^{-1})=q(x)/q(y)=a. \end{equation} Since $a$ was arbitrary, (30006799:normtorsors) proves this quaternion subcase. The decisive mechanism is not just that an invariant vanishes: it is that vanishing moves a specific eight-dimensional form into a filtration level whose nonzero anisotropic elements require dimension at least sixteen.

3. Attempt to propagate the dimension-gap mechanism. For a simply connected absolutely almost simple group in characteristic zero, the Rost invariant is a pointed map \[ r_G:H^1(K,G)\longrightarrow H^3(K,{\mathbb{Q}}/{\mathbb{Z}}(2)). \] The right side is zero when $\operatorname{cd}(K)\le2$. Consequently a proof of \begin{equation}\label{wsref-0-33-30-30-30-36-37-39-39-3a-6b-65-72-6e-65-6c} r_G^{-1}(0)=\{1\} \end{equation} for the relevant groups and fields would prove their instances of the conjecture. For quasi-split exceptional groups of the types covered by Gille's Theorem 5.2 this is an established route. It is not a theorem that (30006799:kernel) holds for arbitrary $G,K$: the same source describes nontrivial Rost kernels for $E_8$ over fields outside the present dimension-two scope. [E2], Sections 4.2 and 5.2. They refute an unrestricted invariant-injectivity assertion, not Serre's dimension-two conjecture.

The quaternion proof suggests looking for a finite filtration on a geometric object representing the torsor, with two properties: vanishing of degree-three invariants raises the filtration, and a size bound forces a terminal filtration class to be trivial. Neither such an object nor a size bound independent of the torsor has been supplied for the remaining exceptional groups. Merely introducing further invariants of degrees $4,5,\ldots$ is not enough: all their ordinary Galois-cohomology targets also vanish under the given dimension bound. One still needs a theorem that their joint zero fiber consists only of the trivial torsor. Defining an invariant with trivial kernel by using the torsor itself would be circular.

4. A restriction--corestriction test. A second attempted repair was to trivialize a torsor over finitely many extensions $K_i/K$ whose degrees have gcd one, then combine these trivializations using a Bezout identity. For an abelian cohomology group one often has \[ \operatorname{cor}_{K_i/K}\operatorname{res}_{K_i/K}(x) =[K_i:K]x, \] so such an argument can work. But $H^1(K,G)$ here is a pointed set, not an abelian group with addition, integer multiples and a universally available corestriction. The required substitute is the separate injectivity assertion \[ \ker\left(H^1(K,G)\to\prod_i H^1(K_i,G)\right)=\{1\} \quad\text{when }\gcd_i[K_i:K]=1. \] This is exactly the kind of additional descent question isolated in [E2], Section 4.3. Replacing a rational point by a zero-cycle of degree one without proving this descent step would weaken the target. The transfer paper itself distinguishes the zero-cycle and rational-point versions. [E1], Introduction and Section 4.

Stress test.

Cohomological dimension two does not make all degree-two obstructions vanish. For example, the Brauer group of a nonarchimedean local field contains nonzero quaternion classes; they give nontrivial $\mathrm{PGL}_2$ torsors. This group is not simply connected, so the example is outside the target and explains why that hypothesis matters. The distinction and the local-field results are part of [E2], Sections 1--3. Nothing in (30006799:ratio) asserted that $D$ itself became split.

The quadratic argument assumes characteristic different from two explicitly; it is not a proof for every perfect field. The characteristic-zero Rost calculation likewise avoids importing an incorrect ordinary-cohomology description of positive-characteristic primary invariants. The 2025 transfer theorem could carry a full characteristic-zero solution to the remaining characteristic, but a proof for quaternion groups alone is not its full premise. Split groups, inner forms, products and restrictions of scalars must not be identified without their separate descent arguments. None of the numerical invariants considered here is being assumed complete.

Outcome.

Outcome: Exploratory approach with unresolved gap.

The attempt supplies the explicit quaternion norm calculation, explains its filtration/dimension mechanism, and identifies precisely the invariant-kernel and descent statements needed to extend it. The calculation uses established deep theorems and belongs to a known subcase; no new family of Serre-II cases is claimed.

No complete invariant-detection or coprime-degree descent argument has been obtained in the remaining generality. In particular, the vanishing of the target of the Rost invariant has not been confused with the vanishing of its source. Both the all-group and all-perfect-field quantifiers remain unresolved by this attempt.

Sources

• S1 Philippe Gille, "Serre's conjecture II: a survey," in Quadratic forms, linear algebraic groups, and cohomology, Dev. Math. 18, Springer, New York $2010$, 41-56. https://arxiv.org/abs/2308.00903. Catalogue formulation source.
• E1 D. Izquierdo and G. Lucchini Arteche, Transfer principles for Galois cohomology and Serre's conjecture II, arXiv:2308.00903v3, 9 September 2025, Introduction and Sections 3--4, particularly 4.3. https://arxiv.org/html/2308.00903v3. This is the actual work reached by the original [S1] URL.
• E2 P. Gille, Serre's conjecture II: a survey, in Quadratic Forms, Linear Algebraic Groups, and Cohomology, Developments in Mathematics 18, Springer $2010$, 41--56. Consulted author manuscript, Sections 1--5, especially 4.2--4.3 and Theorems 5.1--5.2. https://www.math.uni-bielefeld.de/lag/man/326.pdf. Its historical status discussion is supplemented by [E1] and [E4].
• E3 A. Quéguiner-Mathieu, Pfister forms in the algebraic and geometric theory of quadratic forms, author-hosted exposition, Section 4, Theorem 4.1 and the Milnor-isomorphism discussion on pp. 6--7. https://www.math.univ-paris13.fr/~queguin/fichiers/Pfister.pdf. The cited inputs are the Arason--Pfister theorem and the established Milnor conjectures, not conjectural additional hypotheses.
• E4 M. N. T. Nguyen, Serre conjecture II for pseudo-reductive groups, arXiv:2603.08061v1, 9 March 2026, Abstract and formulation of the pseudo-reductive extension. https://arxiv.org/abs/2603.08061. All four research references were consulted 27 September 2026.