The linear positive torsor presentation and arithmetic purity of strong approximation

Let x\in U(\mathbb A_K)^{\operatorname{Br}(U)}. When G is of multiplicative type, we choose x\in U(\mathbb{A}_{K})^{\mathrm{Br}_{1}(U)}. Let \mathcal N be an open neighbourhood of x in U(\mathbb A_K).

By ([CDX19], Theorem 3.5) , there exist \xi\in H^1(K,G) and \eta\in Y_U^\xi(\mathbb A_K), such that p_U^\xi(\eta)=x. Let f=p^{\xi}_{U}.

The central cocharacter descends to V^\xi with the same positive weights. We continue to denote it by \lambda; it preserves Y^{\xi}_{U} and satisfies

There exists nonempty open subsets

such that

By Proposition 2.1, there exist y\in Y^{\xi}_{U}(K) and r\in\mathbb Q_{>0} satisfying

Consequently,

Since f is invariant under \lambda, the element u:=f(y)=f(\eta')\in\mathcal N. And we conclude the theorem as x and \mathcal N were arbitrary.

Let q:Y\to X be the torsor defined in propostion 4.1, and \lambda:\mathbb{G}_{m}\to G be the cocharacter defined in lemma 4.2. Then the datum (G,V,Y,q,\lambda) is a postive linear torsor presentation for X. Since G is of multiplicative type, we conclude the result by theorem 1.5.

The equivalence of (i) and (ii) follows from the fact that:

(ii) implies (iii): This is a direct consequence of (lemma 5.1).

(iii) implies (iv): Trivial.

(iv) implies (i): If x,y is connected by one chain, let f:\mathbb P^1_K\to X with f(0)=x and f(\infty)=y. Denote P=Y_x. The Scheme

is a torsor under the smooth K-group \operatorname{Aut}_G(P) with trivial fibre at 0. By [bouthier2025generically, Theorem 7.2.1], its fibre at \infty is also trivial. Thus Y_y\simeq P=Y_x, and hence

In the general case, apply the argument repeatedly.

Finally, let \xi\in H^1(K,G) be an arbitrary element. By Lemma 3.1, V^\xi is isomorphic to \mathbb A_K^{\dim_K V}. Since K is infinite, every nonempty open subset of affine space has a K-point. Therefore we may choose y\in Y^\xi(K) and let x=q^\xi(y)\in X(K). Then we have \delta_q(x)=\xi, thus \delta_q is surjective. Together with the equivalence of (i) and (iv), we conclude that X(K)/\sim_{\mathbb{P}^1,\mathrm{ch}} \cong H^1(K,G).

First assume K=\mathbb{Q}. Let n=\mathrm{dim}(V) and

Changing the signs of the weight-basis vectors if necessary, we may assume W_\infty\cap\mathbb R_{>0}^n\neq \varnothing and fix an element w\in W_\infty\cap\mathbb R_{>0}^n. By ([MR3873133], proposition 4.6), we have

Hence there exists b\in Y(\mathbb{Q}) such that b_{f}\in W_{f} and b_{\infty}>0. Now consider the following morphism:

Note that \Phi_{b}(ct)=\lambda(c)\Phi_{b}(t).

Let Y_{b}:=\Phi_{b}^{-1}(Y), since \Phi_{b} is faithfully flat, the preimage Y_{b} is again an open subset of affine space with co-dimensional two complement. Moreover, \Phi_{b}^{-1}(W_{f}) is non-empty since it contains 1, and \Phi_{b}^{-1}(W_{\infty}) is non-empty since (\frac{w_{i}}{b_{i}})^{1/a_{i}} maps to w\in W_{\infty}.

Let B\subset \Phi^{-1}_{b,\infty}(W_{\infty})\cap \mathbb{R}^{n}_{+} be a convex open ball and let v_{1},...,v_{n}\in B\cap \mathbb{Q}^{n} be linearly independent vectors. Then we define an automorphism of affine space:

For z\in \mathbb{R}^n_{+}, define r(z):=\sum^{n}_{j=1}z_{j}. Then the convexity of B gives

Now let U_{f}=(\Phi_{b}\circ \theta)^{-1}(W_{f}). Applying ([MR3873133], proposition 4.6) to U_{f} yields an element z_{0}\in \theta^{-1}(Y_{b})(\mathbb{Q}), such that

Finally, let y=\Phi_{b}(\theta(z_{0}))\in Y(\mathbb{Q}) and r=r(z_{0}). Then

In the general case, let d=[K:\mathbb{Q}] and take

Let Z=V\setminus Y. Then base change to \overline{\mathbb{Q}} gives

Each term has dimension at most (n-2)+(d-1)n=dn-2. There are natural identifications

Restricting the \lambda-action to the diagonal \mathbb G_m\subseteq\operatorname{Res}_{K/\mathbb{Q}}\mathbb G_m repeats every original weight d times. Hence the general case follows from the K=\mathbb{Q} case.

As the action \rho:G\to \mathrm{GL}(V) is linear, the twist V^{\xi} is a vector bundle over \mathrm{Spec}(K) and is isomorphic to the affine space.

Now we define a cocharacter \lambda^{\xi}:\mathbb{G}_{m}\to \mathrm{GL}(V^{\xi}) as:

It is well-defined since \lambda(t)\in Z(G).

Let V=\bigoplus_a V_a for the weight decomposition under \rho\circ\lambda. Each V_a is G-stable Since \lambda is central. Hence

The induced action of \mathbb G_m on P\times^G V_a is multiplication by t^a, and \dim_K(P\times^G V_a)=\dim_K V_a. Thus \lambda^\xi has the same weights as \rho\circ\lambda with the same multiplicities. In particular, all weights of \lambda^{\xi} are positive.

Since q is a torsor under the smooth group G, it is smooth and faithfully flat. Therefore

Let Z be the closure of q^{-1}(X\setminus U) in V. If Z contains a codimension-one irreducible component E, then E could not be contained in V\setminus Y. Hence E\cap Y would be a codimension-one subset of Y contained in q^{-1}(X\setminus U), a contradiction, and the result follows.

. Let C=\sum_{\rho\in\Sigma(1)}\mathbb R_{\geq0}v_\rho \subseteq N_{\mathbb R}, and let C^\vee\subseteq M_{\mathbb R} be its dual cone. Inside the group algebra \overline{K}[M], restriction to the torus identifies

It follows that C^\vee\cap M=\{0\}, and we conclude that C^{\vee}=0 since C^\vee is a rational polyhedral cone. Thus C=(C^\vee)^\vee=N_{\mathbb R} by the duality of closed convex cones.

Consequently, we have -v_\rho\in C for every \rho\in\Sigma(1). Choose b_{\rho\tau}\in\mathbb Q_{\geq0} such that v_\rho+\sum_{\tau\in\Sigma(1)}b_{\rho\tau}v_\tau=0. Summing these relations over \rho, we obtain

Every coefficient is a strictly positive rational number. Clearing denominators gives positive integers n_\tau such that \sum_\tau n_\tau v_\tau=0.

Let L/K be the splitting field of T and let \Gamma=\mathrm{Gal}(L/K). For each \rho, let a_\rho=\sum_{\gamma\in\Gamma}n_{\gamma^{-1}\rho}, then we have

Define \ell:D\to\mathbb Z by \ell(D_\rho)=a_\rho, then \ell is Galois-equivariant. For every m\in M, we have

Hence \ell factors through a Galois-equivariant homomorphism \overline\ell:P\to\mathbb Z. Taking the dual of \overline{\ell} yields a K-cocharacter \lambda:\mathbb G_m\to G.

The composite \mathbb G_m\xrightarrow{\lambda}G\hookrightarrow R is the dual of \ell:D\to\mathbb Z. Consequently, after base change to \overline K, the map sends t to (t^{a_\rho})_{\rho\in\Sigma(1)}, and its induced action on V_{\overline{K}} is

which proves our claim.

Let Z=W\setminus U with its reduced scheme structure, and let u=(u_i) and v=(v_i) be their coordinates. Let C=\mathbb A^n_K with coordinates c_1,\ldots,c_n and let [S:T] be the homogeneous coordinates of \mathbb P^1_K with 0=[1:0] and \infty=[0:1]. Define

Then F(c,1,0)=u, F(c,0,1)=v, and F(c,rS,rT)=\lambda(r^2)F(c,S,T).

The ideal of Z is homogeneous for the grading \deg z_i=a_i since Z is \lambda-stable. Thus we may choose weighted homogeneous generators h_1,\ldots,h_b of weighted degrees d_1,\ldots,d_b. Each h_j(F(c,S,T)) is homogeneous of degree 2d_j in S,T. Thus these equations define a closed subscheme

Note that the fibres of \mathcal{M}\to \mathbb{P}^1 over 0 and \infty are empty, since u,v\notin Z.

We identify D_+(ST)\subseteq\mathbb P^1_K with \mathbb G_m via \tau=T/S. Consider the morphism

It is an isomorphism, with inverse

Under the identification D_+(ST)\simeq\mathbb G_m, we have

In particular, we have \dim\mathcal M=\dim Z+1\leq n-1.

Let pr_{2}:C\times_K\mathbb P^1_K\to C and B=\operatorname{pr}_2(\mathcal M). Then B\neq C, and we may choose a rational point c_0\in(C\setminus B)(K). Then F(c_0,S,T)\notin Z for every geometric point (S,T)\neq(0,0), thus F(c_0,-,-) factors through U.

On the charts S\neq0 and T\neq0 of \mathbb{P}_{K}^1, we write the coordinates \tau=T/S and \sigma=S/T respectively, and define

On the overlap, we have \sigma=\tau^{-1} and F(c_0,\sigma,1) =\lambda(\sigma^2)F(c_0,1,\sigma^{-1}). The \lambda-invariance of p gives f_T(\sigma)=f_S(\sigma^{-1}). Thus these morphisms glue to a K-morphism f:\mathbb P^1_K\to X satisfying f(0)=p(u) and f(\infty)=p(v).