Abstract
This article is a survey of conjectures and results on reductive algebraic groups having good reduction at a suitable set of discrete valuations of the base field. Until recently, this subject has received relatively little attention, but now it appears to be developing into one of the central topics in the emerging arithmetic theory of (linear) algebraic groups over higherdimensional fields. The focus of this article is on the Main Conjecture (Conjecture 5.7) asserting the finiteness of the number of isomorphism classes of forms of a given reductive group over a finitely generated field that have good reduction at a divisorial set of places of the field. Various connections between this conjecture and other problems in the theory of algebraic groups (such as the analysis of the globaltolocal map in Galois cohomology and the genus problem) are discussed in detail. The article also includes a brief review of the required facts about discrete valuations, forms of algebraic groups, and Galois cohomology.
Introduction
Over the past six decades, the analysis of various properties of linear algebraic groups over local and global fields, the origins of which can be traced to the works of Lagrange and Gauss, has developed into a wellestablished theory known as the arithmetic theory of algebraic groups (cf. [94]). While this subject remains an area of active research, there is growing interest in the arithmetic properties of linear algebraic groups over fields of an arithmetic nature that are not global (such as function fields of curves over various classes of fields, including padic fields and number fields). These recent developments rely on a symbiosis of methods from the theory of algebraic groups on the one hand, and arithmetic geometry on the other. At this stage, it is too early to give a comprehensive account of these new trends, so the goal of the present article is to discuss one important, and somewhat surprising, instance of the propagation of the ideas of arithmetic geometry into the theory of algebraic groups. Curiously, reduction techniques that have been used in the analysis of diophantine equations since antiquity, and the notion of good reduction, which is central to modern arithmetic geometry, were utilized in the classical arithmetic theory of algebraic groups in a rather limited way (see the discussion in Sect. 5).
A key novelty in the current work is that the consideration of algebraic groups having good reduction at an appropriate set of discrete valuations of the base field has moved to the forefront. In fact, one of the important conjectures in the area states that under suitable assumptions, the number of isomorphism classes of such absolutely almost simple groups having a given type should be finite (see Conjecture 5.7 for the precise statement). Philosophically, this conjecture can be viewed as an analogue of Shafarevich’s conjecture, proved by Faltings [38], on the finiteness of the number of isomorphism classes of abelian varieties defined over a given number field and having good reduction outside a fixed finite set of places of the field. More importantly, just like the work of Faltings is now the centerpiece of finiteness results in arithmetic geometry, this conjecture and related ones are likely to become the cornerstone for various finiteness properties involving linear algebraic groups over higherdimensional fields. Here we just mention that these conjectures deal with such classical aspects of the theory as the localglobal principle (formulated in terms of properties of the globaltolocal map in Galois cohomology—cf. [118] and the recent survey [87]) as well as ways of extending the theorem on the finiteness of class numbers for groups over number fields to more general situations. It should also be pointed out that, if proven, the finiteness conjecture for groups with good reduction would have numerous applications: we will discuss the genus problem for absolutely almost simple algebraic groups and weakly commensurable Zariskidense subgroups of these that play a crucial role in the analysis of lengthcommensurable locally symmetric spaces, particularly those that are not arithmetically defined (such as, for example, nonarithmetic Riemann surfaces). We hope that this new direction of research in the theory of algebraic groups, which brings together various topics from a number of areas and has rather unexpected applications, will be of interest to a broad audience.
The structure of the article is as follows. In Sect. 2, we give a brief overview of the use of reduction techniques and the notion of good reduction in arithmetic geometry, focusing, in particular, on the (weak) Mordell theorem and Shafarevich’s theorem on elliptic curves with good reduction that eventually culminated in the work of Faltings. Next, in Sect. 3, we consider some motivating examples that naturally lead to the formal definition of good reduction for reductive groups. Since the formulation of our Main Conjecture relies on the notion of forms of a given (reductive) algebraic group, we recall this in Sect. 4, along with the required facts about Galois cohomology. In Sect. 5, we first review previous work on the analysis of groups with good reduction, which focused exclusively on the case where the base field is the fraction field of a Dedekind domain and the relevant set of places consists of the discrete valuations of the field associated with maximal ideals of the ring. We then formulate the Main Conjecture (Conjecture 5.7) for groups having good reduction at a divisorial set of places of a finitely generated field. In Sect. 6, we discuss several other finiteness conjectures in this setting, which deal with the properness of the globaltolocal map in Galois cohomology and finiteness conditions on the class set of an algebraic group. Available results on these conjectures are presented in Sect. 7. We should point out that recently, all the finiteness properties discussed in this article were established for algebraic tori [106]. On the other hand, results for absolutely almost simple groups are more modest (see, in particular, [20,21,22]), and a great deal of work here still lies ahead. In Sect. 8, we apply these results to the genus problem, which focuses on understanding absolutely almost simple algebraic groups that have the same isomorphism classes of maximal tori over the field of definition (cf. [21, 22]). In Sect. 9, we discuss some applications to Zariskidense subgroups of absolutely almost simple groups related to the idea of eigenvalue rigidity; the latter is based on the notion of weak commensurability and grew out of the investigation of isospectral and length commensurable locally symmetric spaces (cf. [96, 104]). Finally, for completeness, in Sect. 10, we discuss two aspects of the arithmetic theory of algebraic groups over higherdimensional fields that are not directly related to the Main Conjecture, but which play an important role: strong approximation and rigidity.
Good reduction in arithmetic geometry
In order to provide some historical and philosophical context for the recent developments in the study of algebraic groups with good reduction, in this section, we give a brief overview of the use of reduction techniques and the notion of good reduction in arithmetic geometry.
To begin with, reduction techniques are among the most basic and oldest tools in number theory. Indeed, it has been known since antiquity that they can be used to show that certain algebraic equations do not have integral or rational solutions (in other words, the corresponding algebraic varieties have no integral or rational points). For example, consider the equation
and suppose that \((x_0, y_0)\) is an integral solution. Reducing this equation modulo 7, we obtain
However, there is no class modulo 7 that satisfies this condition as otherwise the multiplicative group \(\left( {{\mathbb {Z}}}/7{{\mathbb {Z}}}\right) ^{\times }\), which has order 6, would contain an element of order 4. This means that the original equation has no integral solutions.
At the beginning of the twentieth century, it was realized that beyond simply detecting the absence (and sometimes also the presence) of rational points, reduction techniques can be used to analyze the structure of the solution set, i.e., the set of rational points of the corresponding variety over a given field. One of the earliest, and perhaps most telling, examples of this arose in the study of elliptic curves. For simplicity, suppose that K is a field of characteristic \(\ne 2,3.\) We recall that an elliptic curve E over K is given by an equation of the form
where \(f(x) = x^3 + ax + b\) is a cubic polynomial over K without multiple roots. More precisely, E is the corresponding projective curve that is obtained by adding one point at infinity. It is well known (see, for example, [125, Ch. 1]) that the set of Krational points E(K) has the structure of an abelian group (with the group operation defined geometrically by the chordtangent law). One of the cornerstone results in the arithmetic of elliptic curves is that when K is a number field (or, more generally, a finitely generated field), then the group E(K) is finitely generated. This statement, known as the MordellWeil Theorem, is actually true for any abelian variety (cf. [71, Ch. VI]). The proof for elliptic curves over the field \(K = {{\mathbb {Q}}}\) of rational numbers was given by Louis Mordell [80] in 1922. One of the key steps, which is now usually referred to as the Weak Mordell Theorem, is to show that the quotient E(K)/2E(K) is finite (in fact, E(K) /nE(K) is finite for any \(n \ge 1\)); the argument here relies heavily on reduction.
For the purposes of this discussion, let us assume that the coefficients a and b of f(x) in (1) are integers, and let \(p > 3\) be a prime. Reducing modulo p, we obtain the equation
over the finite field \({\mathbb {F}}_p\) with p elements, where \({\bar{a}}\) and \({\bar{b}}\) denote the images of a and b in \({\mathbb {F}}_p\), respectively. If \({\bar{f}}\) does not have multiple roots, the reduced equation still defines an elliptic curve, in which case we say that Eq. (1) has good reduction at p. On the other hand, if \({\bar{f}}\) has multiple roots, we say that (1) has bad reduction at p. In this case, after reduction modulo p, the original elliptic curve degenerates into a singular rational curve. We note that the primes of bad reduction are precisely those that divide the discriminant \(\Delta (f) =4a^327b^2\) of f, and therefore form a finite set. Furthermore, we say that an elliptic curve E has a good reduction at a prime \(p > 3\) if (after a possible \({{\mathbb {Q}}}\)defined change of coordinates) it can be given by an Eq. (1) that has good reduction at p.^{Footnote 1} For example, the equation \(y^2 = x^3  625x\) has bad reduction at \(p = 5\), but the elliptic curve it defines is isomorphic to the elliptic curve given by \(y^2 = x^3  x\), which has good reduction at \(p = 5\). Otherwise, we say that E has bad reduction at p (we refer the reader to [124, Ch. VII] for a systematic account of these issues).
We will now sketch a proof of the Weak Mordell Theorem. Fix an elliptic curve E over \({{\mathbb {Q}}}\) and let
(as noted above, S is a finite set of primes). Furthermore, for a point \(R \in E({\bar{{{\mathbb {Q}}}}})\), where \({\bar{{{\mathbb {Q}}}}}\) is a fixed algebraic closure of \({{\mathbb {Q}}}\), we let \({{\mathbb {Q}}}(R)\) denote the residue field of R, which is \({{\mathbb {Q}}}\) if R is the point at infinity, and the field generated by the coordinates of R for all other points. We now consider the isogeny
Since \(\pi \) has degree 4, it follows that for any \(P \in E({{\mathbb {Q}}})\) and any \(R \in \pi ^{1}(P)\), the field extension \({{\mathbb {Q}}}(R)/{{\mathbb {Q}}}\) is of degree \(\le 4.\) Moreover, using the fact that E has good reduction at any prime \(p \notin S\), one shows that the extension \({{\mathbb {Q}}}(R)/{{\mathbb {Q}}}\) is unramified at p (this part of the argument can be carried out either through the analysis of formal groups or Hensel’s lemma). We now recall that according to the classical Hermite–Minkowski theorem, \({{\mathbb {Q}}}\) has only finitely many extensions of a bounded degree that are unramified at all primes outside of a fixed finite set of primes (cf. [83, Ch. III, Theorem 2.13]). Applying this to the set S, we see that among the fields \({{\mathbb {Q}}}(R)\), where \(R \in \pi ^{1}(P)\) and P runs through \(E({{\mathbb {Q}}})\), there are only finitely many distinct ones. Consequently, the field \({{\mathbb {Q}}}(\pi ^{1}(E({{\mathbb {Q}}})))\), which is the compositum of all such \({{\mathbb {Q}}}(R)\), is a finite extension of \({{\mathbb {Q}}}.\) One then derives the required finiteness of the quotient \(E({{\mathbb {Q}}})/2E({{\mathbb {Q}}})\) by combining the Kummer sequence with the inflationrestriction sequence in Galois cohomology. The reader who wishes to fill in the details of this sketch can consult [124, Ch. VII and VIII] for a comprehensive account of the theory of elliptic curves over local and global fields.
Now, while this argument certainly demonstrates the utility of considering the places of good reduction of an elliptic curve, it still does not fully reveal their effect on the curve itself. In his 1962 ICM talk [121], Shafarevich pointed out that if S is a finite set of rational primes, then there are only finitely many isomorphism classes of elliptic curves over \({{\mathbb {Q}}}\) having good reduction at all primes \(p \notin S\) (in fact, he stated his theorem for an arbitrary number field K and a finite set of places S). For the sake of completeness, we sketch an elegant proof of this theorem that appears in Serre’s book [116, IV7] and which Serre attributes to Tate. The argument is based on Siegel’s theorem from diophantine geometry and goes as follows. Let A be the localization of \({{\mathbb {Z}}}\) with respect to the multiplicative set generated by S, which we can assume contains 2 and 3. First, using unique factorization in A, one shows that a given elliptic curve E having good reduction at all \(p \notin S\) is isomorphic to an elliptic curve given by (1), where the coefficients a and b of the polynomial f belong to A and the discriminant \(\Delta (f)\) belongs to the unit group \(A^{\times }\). Next, if we have two elliptic curves with discriminants \(\Delta _1 , \Delta _2 \in A^{\times }\), and \(\Delta _2 = \Delta _1 u^{12}\) for some \(u \in A^{\times }\), then one curve can be replaced by an isomorphic one, whose equation is still of the form (1), so that the discriminants actually become equal. Since the quotient \(A^{\times }/{A^{\times }}^{12}\) is finite, it is enough to show that the polynomials f with \(a , b \in A\) and a fixed value \(\Delta _0\in A^{\times }\) of the discriminant form a finite set. But the equation \(\Delta (f) = \Delta _0\) can be written in the form \(27b^2 =4a^3 + \Delta _0\), which itself defines an elliptic curve, hence has only finitely many solutions in A by Siegel’s Theorem (cf. [71, Ch. VII, §§12]).
The preceding argument is obviously very specific to elliptic curves, but Shafarevich felt that his theorem was an instance of a far more general phenomenon, which prompted him to formulate the following finiteness conjecture for abelian varieties (which are higherdimensional analogues of elliptic curves):
Let K be a number field and S be a finite set of primes of K. Then for every \(g \ge 1\), there exist only finitely many Kisomorphism classes of abelian varieties of dimension g having good reduction at all primes \({\mathfrak {p}} \notin S\).
This conjecture was proved by Faltings [38] in 1982 as a culmination of research in diophantine geometry on finiteness properties over the course of several decades. Its numerous implications include the Mordell conjecture (a smooth projective curve of genus \(\ge 2\) over a number field K has only finitely many Krational points) as well as Shafarevich’s conjecture for curves (for any \(g \ge 2\), there are only finitely many isomorphism classes of smooth projective curves over K of genus g having good reduction at all \({\mathfrak {p}} \notin S\)). We refer the reader to the survey [34] for an account of these developments as well as a discussion of other instances of the analysis of good reduction for various classes of “objects” over number fields. It will come as no surprise to the reader that, to this day, this subject remains one of the major themes in arithmetic geometry.
Reductive algebraic groups with good reduction
Having discussed the notion of good reduction in arithmetic geometry in the previous section, we will now transition to looking at good reduction in the context of linear algebraic groups. We begin this section with a series of examples that highlight several important points that arise in the consideration of reductions of algebraic groups. With these motivations in place, we will then formally define what it means for a reductive group to have good reduction with respect to a discrete valuation of the base field.
We refer the reader to Borel’s book [10] for a detailed account of the theory of linear algebraic groups. For our purposes, we recall that a linear algebraic group is a subgroup \(G \subset \mathrm {GL}_n (\Omega )\) that is defined by polynomial equations over \(\Omega \) in terms of the matrix entries (where \(\Omega \) is some algebraically closed field). Moreover, if K is a subfield of \(\Omega \) and the ideal of all polynomial functions over \(\Omega \) that vanish on G is generated by polynomials with coefficients in K, we say that G is Kdefined or is an algebraic Kgroup. We note that if \(\mathrm {char}\,\Omega = 0\), then G is Kdefined if and only if it can be defined by polynomial equations with coefficients in K (cf. [10, AG, 12.2]). For simplicity, we will consider examples over \(K = {{\mathbb {Q}}}\), but these can be easily generalized.
Example 3.1
Intuitively, the reduction of the general linear group \(G = \mathrm {GL}_n\) modulo a prime p should be the group \(\mathrm {GL}_n\) over the field \({\mathbb {F}}_p = {{\mathbb {Z}}}/ p {{\mathbb {Z}}}.\) To justify this fact formally, one uses the following considerations. We observe that G can be viewed as a \({{\mathbb {Z}}}\)group scheme \(\mathrm {Spec}\, A\), where
in the sense that for any commutative ring R, one can identify the group \(\mathrm {GL}_n(R)\) with \(\mathrm {Hom}_{{{\mathbb {Z}}}\text {}\mathrm {alg}}(A , R)\), and this identification is natural in R. For any prime p, we can reduce A modulo p:
Then it is easy to see that \(A_p\) represents \(\mathrm {GL}_n\) over the category of \({{\mathbb {F}}}_p\)algebras, i.e., for any such algebra R, there is an identification of \(\mathrm {GL}_n(R)\) with \(\mathrm {Hom}_{{\mathbb {F}}_p\text {}\mathrm {alg}}(A_p , R)\) that is natural in R. Thus, we can say that the \({{\mathbb {Q}}}\)group \(\mathrm {GL}_n\) has a \({{\mathbb {Z}}}\)structure given by the algebra A, and that the reduction modulo p of the latter represents the group \(\mathrm {GL}_n\) over \({\mathbb {F}}_p\).
In particular, the 1dimensional split torus \({\mathbb {G}}_m = \mathrm {GL}_1\) is represented by \({{\mathbb {Z}}}[x , x^{1}]\); the reduction of this \({{\mathbb {Z}}}\)algebra modulo p is \({\mathbb {F}}_p[x , x^{1}]\), which represents the 1dimensional split torus over \({\mathbb {F}}_p\). More generally, the reduction modulo p of the ddimensional split torus \({\mathbb {G}}_m^d\), which is represented by the Laurent polynomial ring
is the ddimensional split torus over \({\mathbb {F}}_p.\)
Example 3.2
The \({{\mathbb {Q}}}\)group \(G= \mathrm {SL}_n\) is defined inside \(\mathrm {GL}_n\) by the single equation \(\det (x_{ij}) 1 = 0.\) Its reduction modulo p gives an equation of a similar shape over \({\mathbb {F}}_p\), suggesting that the reduction modulo p should be \(\mathrm {SL}_n\) over \({\mathbb {F}}_p\). Again, to justify this formally, we view G as an affine group scheme over \({{\mathbb {Z}}}\) represented by the Hopf \({{\mathbb {Z}}}\)algebra
The reduction of this algebra modulo p is \({\mathbb {F}}_p[x_{11}, \ldots , x_{nn}]/(\det (x_{ij})  1)\), which represents \(\mathrm {SL}_n\) as a group scheme over \({\mathbb {F}}_p\).
Example 3.3
Let G be the special orthogonal group \(\mathrm {SO}_n(q)\), where \(q = x_1^2 + \cdots + x_n^2 \in {{\mathbb {Z}}}[x_1, \dots , x_n]\) and \(n \ge 3\). Then the reduction of G modulo any prime \(p > 2\) is again the special orthogonal group \(\mathrm {SO}_n({\bar{q}})\) of the quadratic form \({\bar{q}} = x_1^2 + \cdots + x_n^2\) over \({\mathbb {F}}_p\).
The common feature in these three examples is, in essence, that reduction modulo p yields an algebraic group of the same type. To be more precise, the groups in Example 3.1, i.e., \(\mathrm {GL}_n\) and the split tori, are (connected and) reductive, and so are their reductions modulo all primes p (in this context, connectedness is always understood in terms of the Zariski topology). The groups in Examples 3.2 and 3.3 are (connected and) semisimple, and all of their reductions are again semisimple.^{Footnote 2} On the other hand, the next two examples exhibit a different type of behavior.
Example 3.4
Let \(p > 2\) be a prime and consider the quadratic extension \(L = {{\mathbb {Q}}}(\sqrt{p})\). Recall that the norm of an element \(z = a + b \sqrt{p} \in L\) is given by \(\mathrm {N}_{L/{{\mathbb {Q}}}}(z) = a^2  pb^2\). There exists an algebraic \({{\mathbb {Q}}}\)group G whose group of \({{\mathbb {Q}}}\)points \(G({{\mathbb {Q}}})\) consists precisely of the elements \(z \in L^{\times }\) with \(\mathrm {N}_{L/{{\mathbb {Q}}}}(z) = 1\). Explicitly, for any \({{\mathbb {Q}}}\)algebra R, the group G(R) consists of matrices of the form \(\left( \begin{array}{cc} a &{}\quad pb \\ b &{}\quad a \end{array} \right) \), with \(a, b \in R\), having determinant 1. In other words, G is defined by the following equations on a \((2\times 2)\)matrix \(X = (x_{ij})\):
We note that the matrix \(\left( \begin{array}{cc} \sqrt{p} &{}\quad \sqrt{p} \\ 1 &{}\quad 1 \end{array} \right) \) conjugates G (over \({\overline{{{\mathbb {Q}}}}}\)) into the group of diagonal matrices \(\left( \begin{array}{cc} u &{}\quad 0 \\ 0 &{}\quad v \end{array} \right) \) with determinant 1, which means that structurally, G is a 1dimensional (\({{\mathbb {Q}}}\)anisotropic) torus, usually denoted \(\mathrm {R}^{(1)}_{L/{{\mathbb {Q}}}}({\mathbb {G}}_m)\) and called the norm torus associated with the extension \(L/{{\mathbb {Q}}}\).
Now, considering the closed subscheme of the \({\mathbb {Z}}\)scheme \(\mathrm {GL}_2\) given by (2) and reducing the defining equations modulo p, we obtain the equations
The solutions to these equations are matrices of the form \(\pm \left( \begin{array}{cc} 1 &{}\quad 0 \\ u &{}\quad 1 \end{array} \right) \). Thus, the reduced equations define an algebraic group over \({\mathbb {F}}_p\) which is disconnected and whose connected component is a 1dimensional unipotent group! At the same time, reducing the equations (2) modulo any prime \(q > 2\) different from p, we still get a 1dimensional torus.
Example 3.5
We now consider a noncommutative version of Example 3.4. Let \(p > 2\) be a prime and D be the quaternion algebra over \({{\mathbb {Q}}}\) corresponding to the pair \((1 , p)\). Explicitly, D is a 4dimensional vector space over \({{\mathbb {Q}}}\) with basis 1, i, j, k and the following multiplication table:
We recall that the reduced norm of a quaternion \(z = a + bi + cj + dk \in D\) is given by \(\mathrm {Nrd}_{D/{{\mathbb {Q}}}}(z)= a^2 + b^2 pc^2  pd^2\). Again, there is a \({{\mathbb {Q}}}\)defined algebraic group G, which is usually denoted \(\mathrm {SL}_{1 , D}\), whose group of \({{\mathbb {Q}}}\)points is
Using the regular representation of D, we can represent G explicitly by matrices of the form
One can easily find a matrix over \({{\mathbb {Q}}}(\sqrt{1})\) or \({{\mathbb {Q}}}(\sqrt{p})\) that conjugates G into matrices of the form \(\left( \begin{array}{cc} A &{}\quad O \\ O &{}\quad A \end{array} \right) \), with \(A \in \mathrm {SL}_2\), which means that over these extensions (and hence over \({\bar{{{\mathbb {Q}}}}}\)), this group is isomorphic to \(\mathrm {SL}_2\). In other words, \(\mathrm {SL}_{1 , D}\) is a \({{\mathbb {Q}}}\)form of \(\mathrm {SL}_2\), hence, in particular, a simple algebraic group (see Sect. 4 for a more detailed discussion of forms).
Taking the obvious linear equations defining the matrices (3) and reducing them modulo p, we obtain a system that defines the following group
The matrices of the form \(\left( \begin{array}{cc} I_2 &{}\quad O \\ * &{}\quad I_2 \end{array} \right) \) in this group form a connected unipotent normal subgroup, which shows that the reduction in this case is not a reductive group.
We see that the reductive \({{\mathbb {Q}}}\)groups in Examples 3.1–3.3 admit a structure of a group scheme over \({\mathbb {Z}}\) (or, more generally, over \({\mathbb {Z}}_{(p)}\)—the localization of \({\mathbb {Z}}\) at the prime ideal (p)), which is described by a system of polynomial equations with coefficients in the respective ring such that the reductions of these equations modulo p still define a reductive group. On the other hand, in Examples 3.4 and 3.5, we have situations where a reductive group is given by a system of polynomials with coefficients in \({{\mathbb {Z}}}\) (or \({{\mathbb {Z}}}_{(p)}\)) such that the reduced system no longer defines a reductive group.
So, in analogy with the case of elliptic curves, we are naturally led to the following definition for \({{\mathbb {Q}}}\)groups: we say that a reductive \({{\mathbb {Q}}}\)group G has good reduction at a prime p if it can be defined by a system of polynomial equations with coefficients in \({{\mathbb {Z}}}_{(p)}\)^{Footnote 3} such that the reduction of the system modulo p still defines a reductive group; otherwise we say that G has bad reduction at p. Thus, the groups in Examples 3.1–3.3do have good reduction at the specified primes. On the other hand, with some additional work, one can show that for any prime \(p > 2\), the group in Example 3.4 has bad reduction at p and good reduction at every odd prime \(q \ne p\), and the group in Example 3.5 has bad reduction at p whenever \(p \equiv 3 \pmod {4}\). (We note that for \(p \equiv 1 \pmod {4}\), the group \(\mathrm {SL}_{1 , D}\) in Example 3.5 is in fact isomorphic to \(\mathrm {SL}_2\), hence has good reduction at p. So, a group defined by a system whose reduction is not reductive may still have good reduction at p since it may be possible to describe the group by another system whose reduction does yield a reductive group.)
Now, for (reductive) linear algebraic groups defined over general fields, one considers the notion of good reduction with respect to discrete valuations. We recall that a (normalized) discrete valuation on a field K is a surjective map \(v :K^{\times } \rightarrow {\mathbb {Z}}\) such that

(a)
\(v(ab) = v(a) + v(b)\);

(b)
\(v(a + b) \ge \min (v(a) , v(b))\) whenever \(a + b \ne 0\).
Then
is a subring of K called the valuation ring of v. It is a local ring with the maximal ideal
which is called the valuation ideal of v. An element \(\pi \in K^{\times }\) with \(v(\pi ) = 1\) is called a uniformizer. It is easy to see that \(\pi \) generates \({\mathfrak {p}}(v)\), and in fact every nonzero ideal of \({\mathcal {O}}(v)\) is of the form \((\pi ^k)\) for some \(k \ge 0.\) In other words, \({\mathcal {O}}(v)\) is a discrete valuation ring (DVR)—we refer the reader to [4, Ch. 9] or [15, Ch. VI, §3, n\(^{\circ }\) 6] for a discussion of various equivalent definitions of a DVR. Next, the quotient \({\mathcal {O}}(v)/{\mathfrak {p}}(v)\) is called the residue field of v, and will be denoted by \(K^{(v)}\). Furthermore, the function \(\vert \cdot \vert _v :K \rightarrow {{\mathbb {R}}}\) defined by
where \(\rho \) is a fixed real number with \(0< \rho < 1\), is a (nonarchimedean or ultrametric) absolute value on K. We will write \(K_v\) for the completion of K with respect to the metric associated with \(\vert \cdot \vert _v\). One shows that v naturally extends to a discrete valuation on \(K_v\); for ease of notation, we will denote this extension simply by v. The corresponding valuation ring and valuation ideal in \(K_v\) will be denoted \({\mathcal {O}}_v\) and \({\mathfrak {p}}_v\), respectively; we note that the quotient \({\mathcal {O}}_v/{\mathfrak {p}}_v\) coincides with the residue field \(K^{(v)}\) defined above. We also set \(U_v = {\mathcal {O}}_v^{\times }\) to be the group of units in \({\mathcal {O}}_v\).
Example 3.6
Here are several useful examples of discrete valuations.
(a) To every rational prime p, there corresponds the (normalized) padic valuation \(v_p\) on \({\mathbb {Q}}\) defined as follows: if \(a \in {\mathbb {Q}}^{\times }\) is of the form \(\displaystyle a = p^{\alpha } \cdot \frac{m}{n}\) with m and n relatively prime to p, then \(v(a) = \alpha \). The corresponding valuation ring is the localization \({\mathbb {Z}}_{(p)}\) considered earlier, and the residue field is \({\mathbb {F}}_p\). Furthermore, the completion is the field of padic numbers \({\mathbb {Q}}_p\), and the valuation ring of \({{\mathbb {Q}}}_p\) is the ring of padic integers \({\mathbb {Z}}_p\).
(b) Let \(K = k(x)\) be the field of rational functions in one variable over a field k, and let \(p(x) \in k[x]\) be a (monic) irreducible polynomial. Then the same construction as in part (a) enables us to associate to p(x) a discrete valuation \(v_{p(x)}\) on K. There is one additional discrete valuation on K given by
Note that all of these valuations are trivial on the field of constants k (i.e., \(v(a) = 0\) for all \(a \in k\)), and cumulatively they constitute all valuations of K with this property. These valuations are often called “geometric” since they naturally correspond to the closed points of the projective line \({\mathbb {P}}^1_k\).
(c) Let again \(K = k(x)\), but now assume that we are given a discrete valuation \(v_0\) of k. Then \(v_0\) can be extended to a discrete valuation v on K by first extending it to the polynomial ring k[x] using the formula
and then extending v to K by multiplicativity
This extension is often called “Gaussian” (see [15, Ch. VI, §10, n\(^{\circ }\) 1] for further details). In particular, we can start with the padic valuation \(v_p\) on \({{\mathbb {Q}}}\) and then, using this procedure, extend it to a valuation v on \({{\mathbb {Q}}}(x).\)
We are now ready to give a formal definition of good reduction for reductive groups. It requires language from the theory of reductive group schemes, for which we refer the reader to [31] or [36] (see also [35] or [144] for general introductions to group schemes). However, it turns out that in a number of particular cases, some of which will be discussed below, good reduction can be characterized in very concrete terms. We note that, in a broad sense, the theory of affine groups schemes shifts attention from concrete matrix groups to the Hopf algebras that represent these groups and on which one can consider various “integral” structures. For our purposes, this change in perspective provides an indispensable framework for discussing the reduction of algebraic groups (as we have already seen in Examples 3.1 and 3.2).
In our discussion, we will use the following standard notation: given an affine scheme \(X = \mathrm {Spec}\, A\), where A is a commutative algebra over a commutative ring R, and a ring extension \(R \subset R'\), we will denote by \(X \times _R R'\) the affine scheme \(\mathrm {Spec}\, (A \otimes _R R')\) over \(R'\) (usually referred to as “base change”).
Definition 3.7
Let K be a field equipped with a discrete valuation v (with corresponding completion \(K_v\) and valuation ring \({\mathcal {O}}_v \subset K_v\)), and let G be a connected reductive Kgroup. We say that G has good reduction at v if there exists a reductive group scheme \({\mathscr {G}}\) over \({\mathcal {O}}_v\) with generic fiber \({\mathscr {G}} \times _{{\mathcal {O}}_v} K_v\) isomorphic to \(G \times _K K_v\).
We recall that a smooth \({\mathcal {O}}_v\)group scheme \({\mathscr {G}}\) is called reductive if for every \(x \in \mathrm{Spec}({\mathcal {O}}_v)\), the geometric fiber \({\mathscr {G}} \times _{{\mathcal {O}}_v} \overline{\kappa (x)}\) (where \(\overline{\kappa (x)}\) is an algebraic closure of the residue field \(\kappa (x)\)) is a (connected) reductive algebraic group over \(\overline{\kappa (x)}.\) We also note that the \({\mathcal {O}}_v\)scheme \({\mathscr {G}}\) with generic fiber \(G \times _K K_v\) is unique (cf. [84, Théorème 5.1] for semisimple groups as well as [50, §6] for the general case); in particular, this means that the reduction (or special fiber)
is well defined. Furthermore, when G is an absolutely almost simple Kgroup, the reduction \({\underline{G}}^{(v)}\) has the same CartanKilling type as G (see [36, Exp. XXII, Proposition 2.8]). It should be observed that the definition of good reduction is sometimes given in terms of reductive \({\mathcal {O}}(v)\)schemes instead of \({\mathcal {O}}_v\)schemes. This makes essentially no difference, but the above definition is more convenient for our purposes, particularly for discussing connections of good reduction with localglobal principles (see Sect. 6).
We conclude this section with several examples of semisimple groups with good reduction that will be sufficient for understanding the rest of the article.
Example 3.8
(a) An absolutely almost simple simply connected Ksplit group G has good reduction at any v. This follows from the Chevalley construction, which provides a \({\mathbb {Z}}\)structure for G, described in detail in [9]. (This generalizes the above discussion of \(\mathrm {SL}_n\) in Example 3.2.)
(b) The algebraic Kgroup \(G = \mathrm {SL}_{1 , A}\) associated with the group of elements of reduced norm 1 in a central simple Kalgebra A (which generalizes the group \(\mathrm {SL}_{1 , D}\) considered in Example 3.5) has good reduction at v if and only if \(A \otimes _K K_v\) comes from an Azumaya algebra \({\mathscr {A}}\) over \({\mathcal {O}}_v\) (which means, in particular, that the reduction \({\mathscr {A}} \otimes _{{\mathcal {O}}_v} K^{(v)}\) is a central simple algebra over the residue field \(K^{(v)}\)). Another way of putting this is to say that A is unramified at v—see the discussion at the beginning of Sect. 7.2.
(c) Assuming that \(\mathrm {char}\, K^{(v)} \ne 2\), the spinor group \(G = \mathrm {Spin}_n(q)\) of a nondegenerate quadratic form q in \(n > 2\) variables over K has good reduction at v if and only if, over \(K_v\), the form q is equivalent to a quadratic form
Forms with good reduction and Galois cohomology
The purpose of this section is to recall some key points concerning forms of algebraic groups and (nonabelian) Galois cohomology. These are needed, on the one hand, in order to formulate an appropriate analogue of Shafarevich’s conjecture for reductive algebraic groups (see Question 4.3 and Conjecture 5.7), and, on the other hand, are indispensable for the discussion of localglobal principles in Sect. 6.
F/Kforms
As we saw in Sect. 2, Shafarevich’s conjecture asserts that the number of Kisomorphism classes of abelian varieties of a given dimension over a number field K that have good reduction outside a given finite set of places of K is finite. So, as a straightforward analogue of this conjecture for reductive linear algebraic groups, one can consider the following:
Let K be a field equipped with a set V of discrete valuations. What are the reductive algebraic groups of a given dimension that have good reduction at all \(v \in V\)? More specifically, what are the situations in which the number of Kisomorphism classes of such groups is finite?
Of course, to make these questions meaningful, one needs to specialize K and V, which we will do in Sect. 5. However, we would first like to point out that in the case of reductive algebraic groups, considering reductive algebraic groups of a given dimension is far less natural than considering abelian varieties of the same dimension. Indeed, in a very coarse sense, all complex abelian varieties of dimension d “look the same”: they are all analytically isomorphic to complex tori \({\mathbb {C}}^d / \Lambda ,\) where \(\Lambda \subset {\mathbb {C}}^d\) is a lattice of rank 2d (see, for example, [82, Chapter 1]). At the same time, the “fine” structure and classification of these varieties are highly involved as these varieties have nontrivial moduli spaces, which leads to infinite continuous families of nonisomorphic varieties. As a simple example, we recall that the isomorphism class of a complex elliptic curve is uniquely determined by its jinvariant, which can be any complex number (in fact, isomorphism classes of elliptic curves are classified by the jinvariant over arbitrary algebraically closed fields—see [124, Ch. III, Proposition 1.4]).
On the contrary, reductive algebraic groups of the same dimension may look completely different. For example, the absolutely almost simple group \(\mathrm {SL}_2\) is 3dimensional, as is the torus \(({\mathbb {G}}_m)^3\), which is a reductive group that is not semisimple. Furthermore, the absolutely almost simple simply connected group of type \(\textsf {B}_3\), which can be realized as the spinor group \(\mathrm {Spin}_7\), has the same dimension as the product of 7 copies of \(\mathrm {SL}_2\), which is a semisimple but not absolutely almost simple group. A fundamental result in the structure theory of reductive groups is that over an algebraically (or separably) closed field, for each integer \(d \ge 1\), there are only finitely many isomorphism classes of (connected) reductive groups of dimension d (thus, there are no nontrivial moduli spaces for such groups—see, for example, [31, Theorem 1.3.15] for the precise statement). So, in the analysis of finiteness phenomena for reductive groups, it makes sense to focus on those classes of groups that become isomorphic over a separable closure of the base field; the main issue then becomes the passage from an isomorphism over the separable closure to an isomorphism over the base field (socalled Galois descent). This brings us to the following.
Definition 4.1
Let G be a linear algebraic group over a field K, and let F/K be a field extension. An algebraic Kgroup \(G'\) is called an F/Kform of G if there exists an Fisomorphism
(where \(G \times _K F\) denote the algebraic Fgroup obtained by base change from K to F).
Although we will restrict ourselves primarily to forms of algebraic groups in the present article, we should point out that the consideration of forms of various other “algebraic objects” comes up naturally in many different contexts—we refer the reader to [118, Ch. III, §1] for a general discussion of forms as well as a number of concrete examples. For our purposes, we will be interested mostly in \(K^{\mathrm{sep}}/K\)forms of a linear algebraic group G, where \(K^{\mathrm{sep}}\) is a fixed separable closure of K; these will often be referred to simply as Kforms of G.^{Footnote 4}
Let us now look at several illustrative examples of forms of algebraic groups, which will suffice for understanding most of the article.
Example 4.2
We let K be a field and fix a separable closure \(K^{\mathrm{sep}}\) of K.
(a) Let \(T = ({\mathbb {G}}_m)^d\) be a ddimensional Ksplit torus. Any other ddimensional Ktorus \(T'\) splits over \(K^{\mathrm{sep}}\), i.e., we have a \(K^{\mathrm{sep}}\)isomorphism
This means that all ddimensional Ktori are Kforms of T.
(b) Similarly, let G be an absolutely almost simple simply connected split algebraic group over K, and let \(G'\) be any absolutely almost simple simply connected Kgroup of the same (KillingCartan) type as G. Again, \(G'\) splits over \(K^{\mathrm{sep}}\), hence there is a \(K^{\mathrm{sep}}\)isomorphism
This means that \(G'\) is a Kform of G. Thus, every absolutely almost simple simply connected Kgroup is a Kform of an absolutely almost simple simply connected split Kgroup. For later use, let us also mention the following variant of this statement. First, we recall that a Kgroup G is said to be Kquasisplit (or simply quasisplit if there is no risk of confusion) if it contains a Borel subgroup defined over K. Then one shows that every absolutely almost simple simply connected Kgroup is an inner form of an absolutely almost simple simply connected Kquasisplit group (see Example 4.5 below for a brief discussion of these matters).
(c) Let A be a central simple Kalgebra of degree n, and let \(G = \mathrm {SL}_{1 , A}\) be the algebraic Kgroup associated with the group of elements of reduced norm 1 in A. Since \(A \otimes _K K^{\mathrm{sep}}\) is isomorphic to the matrix algebra \(\mathrm {M}_n(K^{\mathrm{sep}})\), the group \(G \times _K K^{\mathrm{sep}}\) is \(K^{\mathrm{sep}}\)isomorphic to \(\mathrm {SL}_n\). In other words, G is a Kform of \(\mathrm {SL}_n\). In fact, it is an inner form, and all inner forms are obtained this way—see [94, Ch. II, §2.3.4] for the details.
(d) Assume that \(\mathrm{char}~K \ne 2\). Let q be a nondegenerate quadratic form in n variables over K, and let \(G = \mathrm {Spin}_n(q)\) be the corresponding spinor group (which is an absolutely almost simple simply connected Kgroup for \(n \ge 3\), \(n \ne 4\)). Then any other nondegenerate quadratic form \(q'\) over K in n variable is equivalent to q over \(K^{\mathrm{sep}}\), implying that for \(G' = \mathrm {Spin}_n(q')\), there is a \(K^{\mathrm{sep}}\)isomorphism similar to (4). Thus, \(G'\) is a Kform of G. If n is odd, then the groups \(G' = \mathrm {Spin}_n(q')\) account for all Kforms of G. For n even however, there may be other Kforms of G defined in terms of the (universal covers of) the unitary groups of (skew)hermitian forms over noncommutative central division Kalgebras with an involution of the first kind (such as, for example, quaternion algebras). For the details on this, as well as a discussion of the Kforms of other groups of classical types, we refer the reader to [94, Ch. II, §2.3.4].
We will close this subsection with the following adjusted version of our original question concerning reductive groups with good reduction.
Question 4.3
Let K be a field equipped with a set V of discrete valuations. What are the Kforms (or even just inner Kforms) of a given reductive algebraic Kgroup G that have good reduction at all \(v \in V\)? More specifically, in what situations is the number of Kisomorphism classes of such Kforms finite?
(In fact, it follows from Example 4.1(b) that when G is an absolutely almost simple simply connected algebraic Kgroup, we can assume that it is quasisplit over K).
We will return to this question in Sect. 5. First, however, we would like to review some of the main points of (nonabelian) Galois cohomology and its connection to forms of algebraic groups.
A brief review of Galois cohomology
Since Galois cohomology will play a significant role in subsequent sections, we will now quickly review some of the basic notions, focusing particularly on nonabelian aspects (for further details, the reader can consult various sources, including [94, Ch. 2] and [118, Ch. I, §5]).
First, suppose a group \(\Gamma \) acts by automorphisms on another group A (which may be noncommutative). We then define the 0th (noncommutative) cohomology by \(H^0(\Gamma , A) = A^{\Gamma }\) (the subgroup of \(\Gamma \)fixed elements). Furthermore, a (noncommutative) 1cocycle is a map \(f :\Gamma \rightarrow A\) satisfying
It should be noted that while the set \(Z^1(\Gamma , A)\) of 1cocycles is of course an abelian group when A is abelian, it may fail to be a group in the general case; instead, it will be treated simply as a pointed set, whose distinguished element is the trivial cocyle. Next, two cocycles \(f , g \in Z^1(\Gamma , A)\) are said to be equivalent if there exists an element \(a \in A\) such that
One easily checks that this gives an equivalence relation on \(Z^1(\Gamma , A)\), and we define \(H^1(\Gamma , A)\) to be the quotient of \(Z^1(\Gamma , A)\) by this relation. Again, in the general case, \(H^1(\Gamma , A)\) will be viewed as a pointed set whose distinguished element is the equivalence class of the trivial cocycle.
As in the classical (abelian) situation, to every short exact sequence
of \(\Gamma \)groups and \(\Gamma \)homomorphisms (i.e., group homomorphisms commuting with the \(\Gamma \)action), one can associate a long exact sequence
of cohomology sets—the exactness of the second half of the sequence is understood in terms of pointed sets, where the kernel is defined as the preimage of the distinguished element. (Unfortunately, in general, this sequence cannot be extended beyond degree 1 since \(H^i(\Gamma , A)\) is undefined for \(i \ge 2\) if A is noncommutative; however, as described in [118, Ch. I, §5, Proposition 43], if A is a central subgroup of B, then there is a natural map \(H^1(G,C) \rightarrow H^2(G,A)\) that extends (6)).
We also note that the standard functoriality properties of cohomology remain valid in the noncommutative situation. More precisely, let A be a \(\Gamma \)group and B be a \(\Delta \)group, and suppose we are given group homomorphisms
Then there is a map of pointed sets \(H^1(\Delta , B) \rightarrow H^1(\Gamma , A)\), which, on the level of cocycles, is defined as follows: the image of \(g \in Z^1(\Delta , B)\) is \(g^* \in Z^1(\Gamma , A)\) given by
In particular, given a \(\Gamma \)group A, for any subgroup \(\Delta \subset \Gamma \), we can apply this construction by taking f to be the identity map \(A \rightarrow A\) and \(\varphi \) to be the natural embedding (in which case (7) obviously holds) to define the restriction map \(H^1(\Gamma , A) \rightarrow H^1(\Delta , A)\). Next, given a normal subgroup \(\Phi \unlhd \Gamma \), we can consider \(A^{\Phi }\) as a group with the natural action of \(\Delta = \Gamma /\Phi \). We can then apply the above construction by taking f to be the natural embedding \(A^{\Phi } \hookrightarrow A\) and \(\varphi \) to be the canonical map \(\Delta \rightarrow \Delta /\Phi \), to obtain the inflation map \(H^1(\Gamma /\Phi , A^{\Phi }) \rightarrow H^1(\Gamma , A)\).
In order to transition to the Galois cohomology of algebraic groups, we need to specialize the preceding definitions to one situation involving topological groups. More precisely, we will now assume that \(\Gamma \) is a profinite group (i.e., a compact and totally disconnected topological group) that acts continuously on a discrete group A (the continuity assumption amounts to the requirement that for each \(a \in A\), its stabilizer \(\Gamma (a)\) is an open subgroup of \(\Gamma \)). We then consider continuous 1cocycles \(f :\Gamma \rightarrow A\). Since we will deal exclusively with continuous cocycles in this context, we will keep the notation \(Z^1(\Gamma , A)\) for the set of all continuous 1cocycles. It is easy to see that an abstract cocycle that is equivalent to a continuous cocycle is itself continuous, and we again define \(H^1(\Gamma , A)\) to be the pointed set of equivalence classes of continuous 1cocycles. The basic properties of abstract noncommutative cohomology remain valid in this setting. In particular, to every exact sequence (5) of discrete groups with continuous \(\Gamma \)action and \(\Gamma \)homomorphisms, there corresponds the long exact sequence (6) of (continuous) cohomology. Furthermore, for every closed subgroup \(\Delta \subset \Gamma \) and any discrete group A with a continuous \(\Gamma \)action, we have the restriction map \(H^1(\Gamma , A) \rightarrow H^1(\Delta , A)\), and for every closed normal subgroup \(\Phi \unlhd \Gamma \), we have the inflation map \(H^1(\Gamma /\Phi , A^{\Phi }) \rightarrow H^1(\Gamma , A)\).
Let us now briefly indicate how the continuous cohomology of profinite groups is related to the cohomology of finite groups (see, for example, [122, Ch. II] for the details). For a profinite group \(\Gamma \), we let \({\mathcal {N}}\) denote the family of all open normal subgroups \(N \subset \Gamma \). Then we have the identification
where the inverse limit is taken over all \(N \in {\mathcal {N}}\) with respect to the canonical homomorphisms \(\Gamma / N_1 \rightarrow \Gamma / N_2\) for \(N_1 , N_2 \in {\mathcal {N}}\) with \(N_1 \subset N_2\). Then for any discrete group A with a continuous \(\Gamma \)action, \(H^1(\Gamma , A)\) defined as above in terms of continuous cocycles is naturally identified with the direct limit
taken over all \(N \in {\mathcal {N}}\) with respect to the inflation maps \(H^1(\Gamma /N_2 , A^{N_2}) \rightarrow H^1(\Gamma /N_1 , A^{N_1})\) for \(N_1 \subset N_2\).
We will apply this general setup only in situations where the profinite group \(\Gamma \) is the Galois group \(\mathrm {Gal}(L/K)\) of a (possibly infinite) Galois extension L/K; in fact, the most important case for us will be where \(\Gamma = \mathrm {Gal}(K^{\mathrm{sep}}/K)\) is the Galois group of a separable closure of K, i.e., the absolute Galois group of K. Then a discrete group A with a continuous \(\mathrm {Gal}(L/K)\)action will be called a Galois L/Kmodule (note that, as above, we allow A to be noncommutative). The cohomology set \(H^1(\mathrm {Gal}(L/K) , A)\) will be denoted simply by \(H^1(L/K , A)\), which will be shortened to \(H^1(K , A)\) if \(L = K^{\mathrm{sep}}\). For any subextension \(K \subset M \subset L\), the Galois group \(\mathrm {Gal}(L/M)\) is a closed subgroup of \(\mathrm {Gal}(L/K)\), so for any Galois L/Kmodule A, we have the restriction map \(H^1(L/K , A) \rightarrow H^1(L/M , A)\). In particular, for any subextension M of \(K^{\mathrm{sep}}\), we have the restriction map \(H^1(K , A) \rightarrow H^1(M , A)\) (where A is a Galois Kmodule, i.e., a \(K^{\mathrm{sep}}/K\)module).
Here is another important example of a restriction map. Let K be a field equipped with a (discrete) valuation v, and let \(K_v\) be the corresponding completion. It is well known (cf. [2, Ch. VII, Proposition 1.2]) that the absolute Galois group \(\mathrm {Gal}(K^{\mathrm{sep}}_v/K_v)\) can be identified with the decomposition group \(D({\bar{v}})\) of a fixed extension \({\bar{v}}\) of v to \(K^{\mathrm{sep}}\), which is a closed subgroup of \(\mathrm {Gal}(K^{\mathrm{sep}}/K)\). This gives rise to the restriction map \(H^1(K , A) \rightarrow H^1(K_v , A)\). Moreover, in a certain obvious sense, this map is independent of the choice of the extension \({\bar{v}}\) (see [77, Ch. VII, Remark 2.4] for the details); in particular, its kernel is well defined.
Now, given an algebraic Kgroup G, for any Galois extension L/K, the group of Lpoints G(L) is naturally a Galois L/Kmodule. We then have the corresponding cohomology set \(H^1(L/K, G(L))\), which is usually denoted \(H^1(L/K, G)\); when \(L = K^{\mathrm{sep}}\), we will simply write \(H^1(K, G).\) If K is equipped with a (discrete) valuation v, then composing the restriction map discussed above with the map induced by the natural embedding \(G(K) \hookrightarrow G(K_v)\), we obtain a map \(H^1(K , G) \rightarrow H^1(K_v , G)\). By abuse of terminology, we will refer to this map, which will play an important role in Sect. 6, also as a restriction map.
Forms and Galois cohomology
To wrap up this section, we will now briefly review the description of F/Kforms in terms of Galois cohomology. As before, we will focus mostly on forms of algebraic groups, but in fact the description remains valid in many other situations, and we will mention some relevant examples. For further details, the reader can consult [67, Ch. VII], [94, Ch. 2], or [118, Ch. III, §1], among other sources.
Let G be an algebraic Kgroup and F/K be a Galois extension. Note that although the group \(A_F\) of Fdefined automorphisms of G (or rather of \(G \times _K F\)) may not naturally be the group of Fpoints of an algebraic Kgroup, it is always a Galois F/Kmodule. Next, let \(G'\) be an F/Kform of G. By definition, there exists an Fdefined isomorphism \(t :G \rightarrow G'\). Since the groups G and \(G'\) are defined over K, the Galois group \(\mathrm {Gal}(F/K)\) acts on the set of Fisomorphisms between G and \(G'\). So, for every \(\sigma \in \mathrm {Gal}(F/K)\), we can consider the isomorphism \(\sigma (t) :G \rightarrow G'\). Then
belongs to \(A_F\), and the mapping \((\sigma \mapsto f(\sigma ))\) defines a 1cocycle in \(Z^1(F/K , A_F)\), which we will denote simply by f. Moreover, one shows that the equivalence class of f is independent of the choice of the isomorphism t. Then it turns out that sending \(G'\) to the equivalence class of f in \(H^1(F/K, A_F)\) defines a bijection between the set of Kisomorphism classes of F/Kforms of G and \(H^1(F/K , A_F)\). In particular, there is a natural bijection between the set of Kisomorphism classes of Kforms of G and \(H^1(K , A_{K^{\mathrm{sep}}})\).
Here are several explicit examples of this classification.
Example 4.4
Let \(T = ({\mathbb {G}}_m)^n\) be an ndimensional Ksplit torus. Then the corresponding automorphism group \(A_{K^{\mathrm{sep}}}\) is \(\mathrm {GL}_n({\mathbb {Z}})\) equipped with the trivial action of \(\mathrm {Gal}(K^{\mathrm{sep}}/K)\). Since, according to Example 4.2(a), any ndimensional Ktorus is a Kform of T, we see that the Kisomorphism classes of ndimensional Ktori are in bijection with the equivalence classes (in this case, conjugacy classes) of continuous homomorphisms \(f :\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow \mathrm {GL}_n({\mathbb {Z}})\). Let us now take an ndimensional Ktorus \(T'\) and consider the corresponding homomorphism f. Then \(N := \ker f\) is an open normal subgroup of \(\mathrm {Gal}(K^{\mathrm{sep}}/K)\), so the corresponding fixed subfield \(L = (K^{\mathrm{sep}})^N\) is a finite Galois extension of K with Galois group \({\mathscr {G}} = \mathrm {Gal}(K^{\mathrm{sep}}/K)/N\). In fact, L is the minimal splitting field of \(T'\), i.e., the minimal Galois extension of K over which \(T'\) becomes isomorphic to the split torus T and it is uniquely determined by this property. We conclude that the Kisomorphism classes of ndimensional tori with the minimal splitting field L correspond bijectively to equivalence classes of ndimensional faithful representations \({\mathscr {G}} \rightarrow \mathrm {GL}_n({\mathbb {Z}})\).
Example 4.5
Let G be an absolutely almost simple simply connected Kgroup. Then the automorphism group \(A_{K^{\mathrm{sep}}}\) fits into an exact sequence
of Galois Kmodules, where \(I_{K^{\mathrm{sep}}}\) is the group of inner automorphisms, which can be identified with the group of \(K^{\mathrm{sep}}\)points \({\overline{G}}(K^{\mathrm{sep}})\) of the corresponding adjoint group \({\overline{G}}\), and \(S_{K^{\mathrm{sep}}}\) is the group of symmetries of the Dynkin diagram of G (see [94, Ch. 2, §2.1.13] for the details). Next, let \(G'\) be another absolutely almost simple simply connected algebraic group of the same type as G. Then both G and \(G'\) become split, hence isomorphic, over \(K^{\mathrm{sep}}\). Thus, \(G'\) is a Kform of G, implying that \(H^1(K , A_{K^{\mathrm{sep}}})\) in this case classifies the Kisomorphism classes of all absolutely almost simple simply connected Kgroups of the same type as G. Associated to (8), we have the following exact sequence of Galois cohomology
Then the Kforms of G that correspond to the cohomology classes in the image of \(\alpha \) are called inner, while all other forms are called outer (note that \(\alpha \) may not be injective!).
For concreteness, let us take \(G = \mathrm {SL}_n\). Then the corresponding adjoint group is \({\overline{G}} = \mathrm {PGL}_n\), and it follows from the Skolem–Noether theorem that the elements of \(H^1(K , {\overline{G}})\) are in onetoone correspondence with the isomorphism classes of central simple Kalgebras of degree n. Then if a cohomology class \(\xi \in H^1(K , {\overline{G}})\) corresponds to an algebra A, the image \(\alpha (\xi )\) corresponds to the norm 1 group \(\mathrm {SL}_{1 , A}\) introduced in Example 4.1(c). We note that the norm 1 groups corresponding to an algebra A and its opposite algebra \(A^{\small \mathrm {opp}}\) are Kisomorphic, which means that the corresponding cohomology classes in \(H^1(K , {\overline{G}})\), which are generally distinct, have the same image under \(\alpha \)!
Next, if G is Ksplit, then it is known that the Galois action on \(S_{K^{\mathrm{sep}}}\) is trivial (so, the elements of \(H^1(K , S_{K^{\mathrm{sep}}})\) correspond to conjugacy classes of continuous homomorphisms \(\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow S_{K^{\mathrm{sep}}}\)), and the sequence (8) has a \(\mathrm {Gal}(K^{\mathrm{sep}}/K)\)equivariant splitting. Let \(\iota :H^1(K , S_{K^{\mathrm{sep}}}) \rightarrow H^1(K , A_{K^{\mathrm{sep}}})\) be the corresponding splitting for \(\beta \). Then for \(\xi \in H^1(K , S_{K^{\mathrm{sep}}})\), the image \(\iota (\xi )\) corresponds to a quasisplit group \(G(\xi )\) (see, for example, [126, Lemma 16.4.8]), and the inner forms of \(G(\xi )\) correspond precisely to the elements of the fiber \(\beta ^{1}(\xi )\). Thus, every form is an inner form of some quasisplit group (note that the quasisplit group is unique up to isomorphism).
Continuing with the assumption that G is Ksplit, let us consider a Kform \(G'\) of G. Let \(c(G') \in H^1(K, A_{K^{\mathrm{sep}}})\) be the corresponding cohomology class and let \(d(G') \in H^1(K, S_{K^{\mathrm{sep}}})\) be the image of \(c(G')\) under \(\beta \). Then \(d(G')\) is represented by a continuous homomorphism \(\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow S_{K^{\mathrm{sep}}}\) called the \(*\)action associated with \(G'\) (cf. [136]). Thus, \(G'\) is an inner form of the split group G if and only if the \(*\)action is trivial. Furthermore, if \({\mathcal {H}}\) denotes the kernel of the \(*\)action, then the fixed field \(L = (K^{\mathrm{sep}})^{{\mathcal {H}}}\) is the (uniquely defined) minimal Galois extension of K over which \(G'\) becomes an inner form of the split group G. We also note that if \(G'\) is Kquasisplit, then L is the minimal Galois extension of K over which \(G'\) splits. We refer the reader to [96, Lemma 4.1] for a detailed discussion of these matters.
In our final example, we will briefly demonstrate how the above method for classifying the Kforms of algebraic groups can be used to classify (the isomorphism classes of) some other objects.
Example 4.6
Fix a nondegenerate quadratic form in n variables over a field K of characteristic \(\ne 2\), and let \(G = \mathrm {O}_n(q)\) be the corresponding orthogonal group. It is well known that any other nondegenerate quadratic form \(q'\) in n variables over K becomes equivalent to q over \(K^{\mathrm{sep}}\). Thus, if Q and \(Q'\) are the matrices of q and \(q'\), then there exists \(X \in \mathrm {GL}_n(K^{\mathrm{sep}})\) such that \(Q = X^t Q' X\). For each \(\sigma \in \mathrm {Gal}(K^{\mathrm{sep}}/K)\), the element
belongs to \(G(K^{\mathrm{sep}})\), and the mapping \((\sigma \mapsto f(\sigma ))\) defines a 1cocycle \(f :\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow G(K^{\mathrm{sep}})\). One then shows that associating to \(q'\) the cohomology class of f sets up a bijection between the set of equivalence classes of nondegenerate quadratic forms in n variables over K and \(H^1(K , G)\). In fact, this easily follows from the fact that \(H^1(K , \mathrm {GL}_n) = 1\) (the noncommutative version of Hilbert’s Theorem 90). Furthermore, for \(H = \mathrm {SO}_n(q)\), the elements of \(H^1(K , H)\) correspond to the equivalence classes of nondegenerate ndimensional quadratic forms that have the same discriminant as q (see [94, Ch. 2, §2.2.2] for the details).
The finiteness conjecture for reductive groups with good reduction
Having reviewed the necessary material on Galois cohomology, we now return to Question 4.3, our central question concerning forms with good reduction. In our discussion, we will deal with two natural choices for the field K and the set of discrete valuations V of K: we will first consider the case where K is the field of fractions of a Dedekind ring R and V is the set of discrete valuations associated with the maximal ideals of R, after which we will turn to finitely generated fields K equipped with a divisorial set of places V.
The Dedekind case
Early interest in this case can be traced back to the work of G. Harder [54] and J.L. ColliotThélène and J.J. Sansuc [28]. For example, the cohomology set introduced in [54, Lemma 4.1.3] (basically) yields, as a particular case, the set of Kisomorphism classes of Kforms of a given semisimple group that have good reduction at all the relevant valuations. In [28], the authors define and analyze similar unramified cohomology sets in connection with the study of torsors under reductive group schemes over lowdimensional base schemes. The basic case where \(K = {\mathbb {Q}}\) and \(R = {\mathbb {Z}}\), and hence V is the set of all padic valuations of \({{\mathbb {Q}}}\), was considered by B.H. Gross [49] and B. Conrad [30]. One of the key observations in [49] is the following.
Theorem 5.1
([49, Proposition 1.1]) Let G be an absolutely almost simple simply connected algebraic group over \({\mathbb {Q}}\). Then G has a good reduction at all primes p if and only if G is split over all \({\mathbb {Q}}_p\).
Note that the fact that an absolutely almost simple simply connected split group G over \({\mathbb {Q}}\) has good reduction at all primes is a particular case of Example 3.8(a). (For comparison, we recall that there are no abelian varieties over \({\mathbb {Q}}\) that have good reduction at all primes—see [1] and [40].) As we will see below, the proof of Theorem 5.1 yields the fact that the \({\mathbb {Q}}\)forms \(G'\) of the \({\mathbb {Q}}\)split group G that have good reduction at all primes are necessarily inner. Furthermore, using deeper results on Galois cohomology, such as the Hasse principle for adjoint semisimple groups over number fields (see [94, §6.5], and particularly Theorem 6.22 therein), one concludes that such forms are uniquely determined by their isomorphism class over \({\mathbb {R}}\), opening thereby a way to their classification up to \({\mathbb {Q}}\)isomorphism (and of course implying the finiteness of the number of such classes). In fact, this analysis is taken much further in [49] and [30] by explicitly constructing such nonsplit forms for certain types in terms of \({\mathbb {Z}}\)lattices and even considering the problem of the classification of such lattices. We note that the choice of a \({\mathbb {Z}}\)lattice determines the structure of a group scheme over \({\mathbb {Z}}\) on the corresponding \({\mathbb {Q}}\)algebraic group, so the classification of such lattices really amounts to the classification of forms with good reduction at all primes up to \({\mathbb {Z}}\)isomorphism rather than just \({\mathbb {Q}}\)isomorphism, which of course is a much harder problem (to get a sense of the difficulty, the reader may want to consider the problem of classifying unimodular integral quadratic forms up to equivalence versus rational equivalence—cf. [117, Ch. V]). It turns out that in some cases, the required lattices can in fact be classified, but in other cases one shows using the mass formula that such lattices are so numerous that no reasonable classification appears possible.
We will now briefly sketch a proof of Theorem 5.1, mainly to indicate the ingredients that go into the argument. One of the key facts that is used is the following very explicit local description of groups with good reduction in the situation at hand (we refer the reader to Example 4.5 for the relevant terminology). Namely, suppose \({\mathscr {G}}\) is an absolutely almost simple simply connected algebraic group over a field \({\mathscr {K}}\) that is a finite extension of \({{\mathbb {Q}}}_p\). Then \({\mathscr {G}}\) has good reduction if and only if it is quasisplit over \({\mathscr {K}}\) and the minimal Galois extension \({\mathscr {L}}\) of \({\mathscr {K}}\) over which it becomes an inner form of the split group is unramified over \({\mathscr {K}}\)—cf. [31, Corollary 5.2.14]. Now, let G be an absolutely almost simple simply connected \({\mathbb {Q}}\)group that has good reduction at all primes, and let L be the minimal Galois extension of \({\mathbb {Q}}\) over which G becomes an inner form of the split group. Applying the above local description of groups with good reduction, we conclude that \(L/{{\mathbb {Q}}}\) is unramified at all primes. On the other hand, a wellknown consequence of Minkowski’s estimate for the discriminant of a number field is that \({\mathbb {Q}}\) does not have nontrivial extensions with this property. Thus, \(L = {{\mathbb {Q}}}.\) It follows that for every prime p, the group G is quasisplit over \({\mathbb {Q}}_p\) and at the same time is an inner form over \({{\mathbb {Q}}}_p\), implying that it actually splits over \({\mathbb {Q}}_p\). Note that since \(L = {{\mathbb {Q}}}\), our argument also shows that G is an inner form of the split group over \({{\mathbb {Q}}}\).
We should point out that Theorem 5.1 easily extends to any number field K and any set V of discrete valuations of K if we limit ourselves to absolutely almost simple simply connected algebraic groups of those types for which the Dynkin diagram does not have nontrivial symmetries (recall that these types are \(\textsf {A}_1\), \(\textsf {B}_{\ell }\), \(\textsf {C}_{\ell }\), \(\textsf {E}_7\), \(\textsf {E}_8\), \(\textsf {F}_4\), and \(\textsf {G}_2\)). However, if we consider groups of other types over a number field that has an everywhere unramified quadratic extension, the above argument and the result itself become invalid even if we take V to be the set of all nonarchimedean valuations of K. Nevertheless, we have the following finiteness result (cf. [60]—we refer the reader to this paper for analogues of Shafarevich’s conjecture for several classes of varieties). In the statement below, we denote by \(V_f^K\) the set of all nonarchimedean places of a number field K.
Proposition 5.2
Let G be an absolutely almost simple simply connected algebraic group over a number field K, and let \(V \subset V_f^K\) be a set of discrete valuations of K such that \(V_f^K \setminus V\) is finite. Then the number of Kisomorphism classes of Kforms of G that have good reduction at all \(v \in V\) is finite.
Proof
First, we recall that according to the Hermite–Minkowski theorem, for every integer \(d \ge 1\), the number field K has only finitely many extensions of degree \(\le d\) that are unramified at all \(v \in V\). Now let S be the group of symmetries of the Dynkin diagram of G. Then the Hermite–Minkowski theorem implies that there are only finitely many continuous homomorphisms
such that the fixed field \(L(\omega ) = (K^{\mathrm{sep}})^{\ker \omega }\) of \(\ker \omega \) is unramified at all \(v \in V\). Let \(\omega _1, \ldots , \omega _r\) be representatives of the conjugacy classes (in S) of such homomorphisms, which we view as elements of \(H^1(K,S).\) As we discussed in Example 4.5, for each \(i = 1, \ldots , r\), there is a quasisplit Kform \(G_i\) of G for which \(L(\omega _i)\) is the minimal Galois extension of K over which \(G_i\) becomes split. Suppose now that \(G'\) is a Kform of G that has good reduction at all \(v \in V\), and let L be the minimal Galois extension of K over which \(G'\) becomes an inner form of the split group. Then, as discussed in Example 4.5, L is the fixed subfield of the kernel of the natural homomorphism \(\omega :\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow S\) given by the \(*\)action. Besides, it follows from [31, Corollary 5.2.14] that L/K is unramified at all \(v \in V\). By our construction, this means that \(\omega \) coincides with one of the \(\omega _i\), in which case \(G'\) is an inner form of \(G_i\). Thus, it is enough to prove, for each \(i = 1, \dots , r\), the finiteness of the number of isomorphism classes of those Kforms of G that have good reduction at all \(v \in V\) and are inner forms of \(G_i\). By definition (see Example 4.5), any such \(G'\) corresponds to an element \(\xi \in H^1(K , \overline{G_i})\). Furthermore, as we already saw in our discussion of Theorem 5.1, for each \(v \in V\), the group \(G'\) becomes quasisplit over \(K_v\), hence \(K_v\)isomorphic to \(G_i\). In the language of Galois cohomology, this means that \(\xi \) lies in the kernel of the restriction map \(H^1(K , \overline{G_i}) \rightarrow H^1(K_v , \overline{G_i})\). Since this is true for all \(v \in V\), we conclude that \(\xi \) is contained in the kernel of the product map
However, it is well known that over number fields, the map \(\theta _{G_i , V}\) is proper (i.e., the preimage of a finite set is finite), and hence, in particular, its kernel is finite—see [11, §5] and [12, §7] (in the next section we will discuss a conjectural extension of this property to a more general situation). So, the number of possible \(\xi \) arising in this setup is finite, and the finiteness of the number of Kisomorphism classes of such Kforms \(G'\) follows. \(\square \)
Another basic example of a Dedekind ring is the polynomial ring \(R = k[x]\) over a field k. We let V denote the set of discrete valuations of the field of rational functions \(K = k(x)\) associated with monic irreducible polynomials \(p(x) \in k[x]\) (see Example 3.6(b)). We then have the following.
Theorem 5.3
(cf. [102, Theorem 1.1] and [45]) Let \(G_0\) be a (connected) semisimple simply connected algebraic group over a field k. If \(G'\) is a Kform of the group \(G = G_0 \times _k K\) that has good reduction at all \(v \in V\) and splits over \(k^{\mathrm{sep}}(x)\), then \(G' = G'_0 \times _k K\) for some kform \(G'_0\) of \(G_0\).
The original result in [102] is more general and is formulated in terms of torsors (principal homogeneous spaces): Let \(G_0\)be a connected reductive algebraic group over a field k, and let \(\pi :B \rightarrow {\mathbb {A}}^1_k\)be a \(G_0\)torsor over the affine line \({\mathbb {A}}^1_k = \mathrm {Spec}\, k[x]\). If \(\pi \)is trivialized by the base change from k to \(k^{\mathrm{sep}}\), then \(\pi \)is obtained by the base change \({\mathbb {A}}^1_k \rightarrow \mathrm {Spec}\, k\)from a \(G_0\)torsor \(\pi _0 :B_0 \rightarrow \mathrm {Spec}\, k\). To derive Theorem 5.3 from this statement, one argues as follows. Since, by our assumption, \(G'\) splits over \(k^{\mathrm{sep}}(x)\), the homomorphism \(\mathrm {Gal}(K^{\mathrm{sep}}/K) \rightarrow S\) to the group of symmetries of the Dynkin diagram that corresponds to the \(*\)action associated with \(G'\) (see Example 4.5) factors through a homomorphism \(\mathrm {Gal}(k^{\mathrm{sep}}/k) \rightarrow S\). Let \(G_0^{\sharp }\) be the quasisplit kform of \(G_0\) corresponding to this homomorphism, and set \(G^{\sharp } = G_0^{\sharp } \times _k K\). Then \(G'\) is an inner Kform of \(G^{\sharp }\), so it corresponds to a cohomology class \(\xi \in H^1(K , {\overline{G}}^{\sharp })\) with values in the corresponding adjoint group. Since \(G'\) has good reduction at all \(v \in V\), it follows that this cocycle gives rise to a torsor of \({\overline{G}}_0^{\sharp }\) over \({\mathbb {A}}^1_k\); in other words, \(\xi \) is the image of some \(\zeta \in H^1_{_{\mathrm {\acute{e}t}}}(k[x] , {\overline{G}}_0^{\sharp })\) under the natural map
(where \(H^1_{_{\mathrm {\acute{e}t}}}(k[x] , {\overline{G}}_0^{\sharp })\) denotes the étale cohomology set over \(\mathrm{Spec}(k[x])\))—cf. [46, Corollary A.8]. But since \(G'\) splits over \(k^{\mathrm{sep}}\), the cocycle \(\xi \) lies in the kernel of the restriction map \(H^1(K , {\overline{G}}^{\sharp }) \rightarrow H^1(k^{\mathrm{sep}} K , {\overline{G}}^{\sharp })\). Then \(\zeta \) lies in the kernel of the map \(H^1_{et}(k[x] , {\overline{G}}_0^{\sharp }) \rightarrow H^1_{et}(k^{\mathrm{sep}}[x] , {\overline{G}}_0^{\sharp })\), hence by the result from [102] is the image of some \(\omega \in H^1(k , {\overline{G}}_0^{\sharp })\) under the map \(H^1(k , {\overline{G}}_0^{\sharp }) \rightarrow H^1_{_{\mathrm {\acute{e}t}}}(k[x] , {\overline{G}}_0^{\sharp })\). Then the kform \(G'_0\) of \(G_0\) corresponding to \(\omega \) is as required.
We note that the assumption that \(G'\) splits over \(k^{\mathrm{sep}}(x)\) holds automatically when k has characteristic zero. Indeed, in this case, the separable closure \(k^{\mathrm{sep}}\) coincides with the algebraic closure \({\overline{k}}\), so the field \(k^{\mathrm{sep}}(x) = {\overline{k}}(x)\) has cohomological dimension \(\le 1\) by Tsen’s theorem (cf. [118, Ch. II, §3]). Hence, applying Steinberg’s theorem, we conclude that \(G'\) is quasisplit over \({\overline{k}}(x)\) (see [118, Ch. III, §2.3]). On the other hand, since \(G'\) has good reduction at all \(v \in V\), the minimal Galois extension L/K over which \(G'\) becomes an inner form of the split group is unramified at all \(v \in V\). Using the Riemann–Hurwitz formula, one concludes that \({\overline{k}}L = {\overline{k}}(x)\), implying that \(G'\) is actually split over \({\overline{k}}(x)\).
Yet another case where forms with good reduction have been considered in full involves the ring of Laurent polynomials \(R = k[x , x^{1}]\). Here, the set V consists of the discrete valuations of \(K = k(x)\) corresponding to monic irreducible polynomials \(p(x) \in k[x]\), with \(p(x) \ne x\).
Theorem 5.4
(cf. [16, Theorem 2.5]) Let \(G_0\) be a (connected) semisimple simply connected algebraic group over k. Assume that \(\mathrm {char}\, k\) is prime to the order of the Weyl group of \(G_0\). Then there is a natural bijection between the isomorphism classes of inner Kforms \(G'\) of the group \(G = G_0 \times _k K\) that have good reduction at all \(v \in V\) and the elements of the Galois cohomology set \(H^1(k(\!(x)\!) , {\overline{G}})\) of the corresponding adjoint group over the field of Laurent series.
Again, this easily follows from the following more general result proved in [16]: Let \(G_0\)be a connected reductive group over a field k. Assume that the characteristic of k is good.^{Footnote 5}Then there is a natural bijection between the isomorphism classes of \(G_0\)torsors over the punctured affine line \({\mathbb {A}}^{\times }_k = {\mathbb {A}}^1_k \setminus \{ 0 \}\)and the elements of \(H^1(k(\!(x)\!) , G)\). We refer the reader to [16, §5] for the details concerning this result, and only mention here that it plays a crucial role in the proof of the conjugacy of the analogues of Cartan subalgebras in certain infinitedimensional Lie algebras [17].
The finiteness conjecture for function fields of curves
An important class of examples of Dedekind rings consists of the rings of regular functions \(R = k[C]\), where C is a smooth geometrically integral affine curve over a field k. In view of the bijection between the maximal ideals of R and the closed points of C, the corresponding V is then the set of discrete valuations of the function field \(K = k(C)\) associated with the closed points of C. Unfortunately, in most cases, no explicit description of the Kforms of a given reductive Kgroup G that have good reduction at all \(v \in V\), similar to Theorems 5.3 and 5.4, seems to be available. A more tractable problem in this setting appears to be the qualitative question about the conditions that ensure the finiteness of the number of isomorphism classes of such forms. We observe that if \(G_0\) is a reductive algebraic kgroup and \(G = G_0 \times _k K\) is the base change of \(G_0\) to K, then for any kform \(G'_0\) of \(G_0\) the group \(G' = G'_0 \times _k K\) is a Kform of G that has good reduction at all \(v \in V\). So, even though nonisomorphic kforms may become isomorphic after base change to K, in order to have the affirmative answer to the above question in a sufficiently general situation, one needs to assume that over k, the groups at hand have only finitely many nonisomorphic forms. This basically amounts to a hypothesis on the finiteness of Galois cohomology.
In [118, Ch. III, §§4.14.3], Serre described a class of fields over which one does have the required finiteness by introducing the following condition on a profinite group \({\mathscr {G}}\):
(F) For every integer \(m \ge 1\), the group \({\mathscr {G}}\)has only finitely many open subgroups of index m
(such profinite groups are sometimes called “small”). He then defined a field K to be of type (F) if it is perfect and its absolute Galois group \(\mathrm {Gal}(K^{\mathrm{sep}}/K)\) satisfies (F). The key result is that if K is a field of type (F), then the set \(H^1(K,G)\) is finite for any linear algebraic Kgroup G (see [118, Ch. III, §4.3, Theorem 4]). Moreover, if K is of characteristic 0 and of type (F), then any linear algebraic group has finitely many Kforms (see [118, Ch. III, §4.3, Remark 1] and [12, §6]).
Recently, in [112], the secondnamed author proposed a generalization of condition (F) that holds in some situations where (F) fails and is still sufficient to establish certain finiteness results. For this, let K be a field, and \(m \ge 1\) be an integer prime to \(\mathrm {char}\, K\). We say that K is of type \((\mathrm {F}'_m)\) if
\((\mathrm {F}'_m)\)For every finite separable extension L/K, the quotient \(L^{\times }/{L^{\times }}^m\)of the multiplicative group \(L^{\times }\), is finite.
If K is of type (F), then it satisfies \((\mathrm {F}'_m)\) for all m prime to \(\mathrm {char}\, K\)—we refer the reader to [112] for a proof of this fact as well as a discussion of various examples of fields of type \((\mathrm {F}'_m)\) and the associated finiteness results. It is reasonable to expect that given an absolutely almost simple algebraic Kgroup G whose Weyl group has order w, the fact that \(\mathrm {char}\, K\) is prime to w and K satisfies condition \((\mathrm {F}'_p)\) for all prime divisors of w should imply the finiteness of \(H^1(K , G)\); however, this has not yet been established. Along these lines, we would like to propose the following finiteness conjecture for forms with good reduction.
Conjecture 5.5
Let \(K = k(C)\) be the function field of a smooth geometrically integral affine curve C over a field k, and let V be the set of discrete valuations associated with the closed points of C. Consider an absolutely almost simple simply connected algebraic Kgroup G, and let w be the order of the Weyl group of G. Assume that \(\mathrm {char}\, k\) is prime to w and that the field k satisfies \((\mathrm {F}'_p)\) for all prime divisors p of w. Then the number of isomorphism classes of Kforms \(G'\) of G that have good reduction at all \(v \in V\) is finite.
As we will see in Sect. 7, for certain types, already condition \((\mathrm {F}'_2)\) for k implies the finiteness of the number of isomorphism classes of forms with good reduction, which suggest that the assumptions on k in the above conjecture could probably be weakened.
The finiteness conjecture for arbitrary finitely generated fields
We now turn to the main finiteness conjecture for forms with good reduction over arbitrary finitely generated fields (see Conjecture 5.7 below). Let us point out that, unlike the Dedekind situation discussed in the previous subsections, forms with good reduction have never been previously considered in the higherdimensional setting.
We begin with a description of our general setup. Let K be an arbitrary finitely generated field, i.e., a field that can be generated over its prime subfield by finitely many elements. Then K possesses natural sets of discrete valuations called divisorial. More precisely, let \({\mathfrak {X}}\) be a model of K, i.e., a normal separated irreducible scheme of finite type over \({{\mathbb {Z}}}\) (if \(\mathrm {char}\, K = 0\)) or over a finite field (if \(\mathrm {char}\, K > 0\)) such that K is the field of rational functions on \({\mathfrak {X}}\). It is well known that to every prime divisor \({\mathfrak {Z}}\) of \({\mathfrak {X}}\), there corresponds a discrete valuation \(v_{{\mathfrak {Z}}}\) on K (cf. [33, 12.3], [56, Ch. II, §6]). Then
is called the divisorial set of places of K corresponding to the model \({\mathfrak {X}}\). Any set of places V of K of this form (for some model \({\mathfrak {X}}\)) will be simply called divisorial. In terms of commutative algebra, this construction amounts to finding a subring R of K whose fraction field is K and such that R is integrally closed (in K) and a finitely generated \({\mathbb {Z}}\)algebra (or \({\mathbb {F}}_p\)algebra). For any height one prime ideal \({\mathfrak {p}} \subset R\) (i.e., a minimal nonzero prime ideal), the localization \(R_{{\mathfrak {p}}}\) is a discrete valuation ring, hence gives rise to a discrete valuation \(v_{{\mathfrak {p}}}\) of K. Then V consists of these valuations \(v_{{\mathfrak {p}}}\) corresponding to all height one prime ideals of R. We note that given a divisorial set of places V associated with a model \({\mathfrak {X}}\), one can consider an affine open subscheme \({\mathfrak {U}} = \mathrm {Spec}\, R\) of \({\mathfrak {X}}\). Then for the divisorial set \(V'\) associated with \({\mathfrak {U}}\), we have \(V' \subset V\) and \(V \setminus V'\) is finite. This means that, in most situations, it is enough to consider divisorial set associated with affine models, where, as we have just seen, they can be described in terms of height one prime ideals.
Furthermore, we observe that any two divisorial sets \(V_1\) and \(V_2\) associated with two different models of K are commensurable, i.e., \(V_i \setminus (V_1 \cap V_2)\) is finite for \(i = 1, 2\) (this makes a divisorial set of places almost canonical), and that for any finite subset S of a divisorial set V, the set \(V \setminus S\) contains a divisorial set. For the sake of completeness, let us briefly justify these properties. We first note that any divisorial set of places V of a finitely generated field K satisfies the following condition
(A) For any \(a \in K^{\times }\), the set \(V(a) := \{ v \in V \, \vert \, v(a) \ne 0 \}\) is finite.
Now, in order to show that two divisorial sets \(V_1\) and \(V_2\) are commensurable, we can assume, without loss of generality, that \(V_i\) is associated with a model \({\mathfrak {X}}_i = \mathrm {Spec}\,R_i\), where \(R_i\) is a finitely generated \({\mathbb {Z}}\)algebra with fraction field K, for \(i = 1, 2\). The finite generation of \(R_i\) implies that there exist nonzero elements \(a_1 \in R_1\) and \(a_2 \in R_2\) such that \(R_2 \subset R_1[1/a_1]\) and \(R_1 \subset R_2[1/a_2]\). Then \(V_1\setminus V_1(a_1) \subset V_2\) and \(V_2 \setminus V_2(a_2) \setminus V_1\). Consequently,
and the commensurability of \(V_1\) and \(V_2\) follows. Furthermore, given a divisorial set V associated with \({\mathfrak {X}} = \mathrm {Spec}\, R\) and a finite subset \(S \subset V\), using weak approximation for K we can find a nonzero \(a \in R\) such that \(v(a)> 0\) for all \(v \in S\). Then the divisorial set \(V_a\) associated with \({\mathfrak {X}}_a := \mathrm {Spec}\, R[1/a]\) satisfies \(V_a \subset V\setminus S\), as required.
Example 5.6
The field of rational functions \(K = {\mathbb {Q}}(x)\) is the fraction field of \(R = {\mathbb {Z}}[x]\) (which, of course, is an integrally closed finitely generated \({\mathbb {Z}}\)algebra). It is well known that any height one prime ideal \({\mathfrak {p}}\) of R is principal with a generator p of one of the following two types: (a) \(p \in {\mathbb {Z}}[x]\) is an irreducible polynomial with content 1; or (b) \(p \in {\mathbb {Z}}\) is a rational prime. In the first case, the associated discrete valuation of K coincides with the one corresponding to the monic irreducible polynomial in \({\mathbb {Q}}[x]\) obtained by dividing p by its leading coefficient (cf. Example 3.6). We will call such discrete valuations geometric and denote the set of all these valuations by \(V_0\). In the second case, the associated discrete valuation coincides with the Gaussian extension of the padic valuation of \({\mathbb {Q}}\). We will refer to such valuations as arithmetic and denote the set of all these valuations by \(V_1\). Thus, the set of divisorial valuations of K associated with the model \({\mathfrak {X}} = \mathrm {Spec}\, R\) is \(V = V_0 \cup V_1\).
We are now in a position to formulate our central conjecture for forms with good reduction over arbitrary finitely generated fields.
Conjecture 5.7
(Main Conjecture for forms with good reduction) Let G be a connected reductive algebraic group over a finitely generated field K, and V be a divisorial set of places of K. Then the set of Kisomorphism classes of (inner) Kforms \(G'\) of G that have good reduction at all \(v \in V\) is finite (at least when the characteristic of K is “good”).
(When G is an absolutely almost simple algebraic group, we say that \(\mathrm {char}\, K = p\) is “good” for G if either \(p = 0\) or \(p > 0\) and does not divide the order of the Weyl group of G. For nonsemisimple reductive groups, only characteristic 0 will be considered good.)
To conclude this section, we would now like to make a few brief comments on the use of the word “main” in the above designation. As we will see in Sect. 6, this conjecture has links to several other finiteness conjectures for reductive algebraic groups over finitely generated fields. In fact, the Main Conjecture, on the one hand, automatically implies the truth of some of these conjectures in certain cases (cf. the discussion following Conjecture 6.1), and on the other hand, some of the other conjectures are likely to provide tools for attacking the Main Conjecture. Second, the Main Conjecture has important applications to the genus problem (cf. Sect. 8) and to the analysis of weakly commensurable Zariskidense subgroups of absolutely almost simple algebraic groups and the related concept of eigenvalue rigidity (cf. Sect. 9). We observe that these developments go back to geometric problems involving lengthcommensurable locally symmetric spaces [96] and can be viewed as a rather inspiring example of the application of techniques from number theory and arithmetic geometry to differential geometry. It should also be noted that we were led to the Main Conjecture by our earlier work on these applications, which indicated the necessity of considering the above statement in a higherdimensional setting. To the best of our knowledge, this point of view has never come up before in the context of algebraic groups.
As a final remark, we note that one can also consider the original conjecture of Shafarevich for abelian varieties (as well as some other classes of varieties) over higherdimensional fields. Let us first observe that, in the case of elliptic curves, the argument of SerreTate that we sketched at the end of Sect. 2 yields, with minimal changes, the following statement:
If K is a finitely generated field of characteristic zero and V is a divisorial set of places of K, then the set of Kisomorphism classes of elliptic curves over K that have good reduction at all \(v \in V\)is finite.
Important results in this direction, particularly over function fields of characteristic \(p > 0\), are due to Parshin [88], Zarhin [147], Szpiro ([132] and [133]), and MoretBailly [81]. We also refer to the reader to [142] for a survey of results in diophantine geometry in positive characteristic.
Some other finiteness conjectures
While Conjecture 5.7 on forms of algebraic groups with good reduction appears to be most important (and in line with results on abelian varieties), it should really be viewed as part of a “package” of several conjectures on finiteness properties of linear algebraic groups over higherdimensional fields. One of these finiteness properties is related to the localglobal principle in this setting. We will begin with the general formulation of the localglobal principle in terms of Galois cohomology, and will then indicate how it translates into statements about norms in finite separable extensions, finitedimensional simple algebras, and quadratic forms.
Let K be a field equipped with a set V of valuations (not necessarily discrete), and let G be a linear algebraic Kgroup. We say that the (cohomological) localglobal principle holds for G with respect to V if the globaltolocal map in Galois cohomology
given by the product of restriction maps, is injective. As we noted in Sect. 4.2, these Galois cohomology sets do not, in general, have a natural group structure, and should instead be viewed as pointed sets whose distinguished element is the cohomology class of the trivial cocycle. For such a map \(\theta _{G,V}\), one defines the corresponding Tate–Shafarevich set to be the kernel \(\ker \theta _{G,V}\), i.e., the preimage of the distinguished element. We should emphasize that, as a reflection of the absence of a natural group structure, different fibers of \(\theta _{G , V}\) may have different sizes. In particular, while is certainly trivial (i.e., reduces to a single element) when \(\theta _{G,V}\) is injective, in general, the injectivity of \(\theta _{G,V}\) is not a formal consequence of the triviality of and typically requires additional considerations involving twisting (cf. [94, Ch. 1, §1.3.2], [118, Ch. I, §5]).
We now briefly recall how the cohomological localglobal principle is interpreted in several concrete situations. First, let L/K be a finite separable field extension, and let \(T = \mathrm {R}^{(1)}_{L/K}({\mathbb {G}}_m)\) be the corresponding norm torus (see Example 3.4 for a special case of such a torus and, e.g., [94, Ch. 2, §2.1.7] for a discussion of the general case). As a consequence of Hilbert’s Theorem 90 and Shapiro’s Lemma, we have a natural isomorphism
Therefore, the cohomological localglobal principle for T is equivalent to the statement (known as the localglobal norm principle) that an element \(a \in K^{\times }\) is a norm in the extension L/K (i.e., a belongs to the norm subgroup \(N_{L/K}(L^{\times }) \subset K^{\times }\)) if and only if it is a norm locally at all \(v \in V\) (i.e., \(a \in N_{L \otimes _K K_v/K_v}((L \otimes _K K_v)^{\times })\) for all \(v \in V\)).
In slightly different terms, if L/K has degree n, then the norm \(N_{L/K}(x)\) of an element \(x \in L\) is given by a homogeneous polynomial \(\nu (x_1, \ldots , x_n)\) over K in terms of the coordinates \(x_1, \ldots , x_n\) of x with respect to a fixed basis of L/K. Then the norm principle asserts that the equation
has a solution \((x_1, \ldots , x_n) \in K^n\) if and only it has a solution \((x_1^v, \ldots , x_n^v) \in K_v^n\) for every \(v \in V\). Among the first results on the localglobal norm principle was the famous Hasse Norm Theorem asserting that this principle indeed holds for all cyclic Galois extensions of number fields when V is the set of all valuations of K, including the archimedean ones. So, to recognize the pioneering role of this result in the subject, the localglobal principle is often referred to as the Hasse principle.
Next, it follows from Example 4.5 that the cohomological localglobal principle for the group \(G = \mathrm {PGL}_n\) is equivalent to the statement that two central simple algebras \(A_1\) and \(A_2\) of degree n over K are isomorphic if and only the algebras \(A_1 \otimes _K K_v\) and \(A_2 \otimes _K K_v\) are isomorphic over \(K_v\) for all \(v \in V\). Thus, the truth of the localglobal principle for \(G = \mathrm {PGL}_n\) for all \(n \ge 2\) amounts to the fact that the natural map of Brauer groups
defined by sending the Brauer class [A] of a finitedimensional central simple Kalgebra A to \(([A \otimes _K K_v])_{v \in V}\), is injective. Furthermore, the localglobal principle for \(G = \mathrm {O}_n(q)\) (the orthogonal group of a nondegenerate ndimensional quadratic form q over K) means that two nondegenerate ndimensional quadratic forms \(q_1\) and \(q_2\) over K are Kequivalent if and only if they are \(K_v\)equivalent for all \(v \in V\). Similar interpretations can be given in the context of simple algebras with involution, hermitian forms, and so on.
Initially, the study of localglobal principles focused almost exclusively on the case where K is a number field and V is the set of all valuations of K, and dealt primarily with some concrete situations rather than with the general cohomological setup that we just described. In particular, the norm principle was thoroughly investigated for arbitrary finite field extensions of number fields using techniques from class field theory (cf. [2, Ch. VII, §11.4]). Another consequence of class field theory is the theorem of Albert–Brauer–Hasse–Noether stating that the map (9) is injective in this case (see [90, Ch. 18, §18.4] for number fields and [48, Ch. 6, §6.5] for function fields). As we mentioned above, this implies the cohomological Hasse principle for \(G = \mathrm {PGL}_n\) for all \(n \ge 2\), and, more generally, for \(G = \mathrm {PGL}_{1 , A}\) for any finitedimensional central simple algebra A over K. Furthermore, the HasseMinkowski Theorem in the theory of quadratic forms implies the localglobal principle for equivalence of quadratic forms, hence the cohomological localglobal principle for the orthogonal groups (see [117, Ch. IV, §3] for a discussion of the HasseMinkowski theorem in the special case \(K = {{\mathbb {Q}}}\) and [85, Ch. VI, §66] for the general case). Eventually, these results and their variations led to the cohomological Hasse principle for all semisimple simply connected groups with components of classical types. This result was subsequently extended, using structural information provided by the theory of algebraic groups, to include exceptional types, ultimately culminating in the proof of the cohomological Hasse principle for all semisimple simply connected groups over number fields (cf. [94, Ch. 6, §§6.76.8]). In fact, this result implies that the principle also holds for all absolutely almost simple groups, and consequently for all adjoint groups—see [94, Ch. 6, §6.5].
On the other hand, it was discovered rather early that the Hasse norm principle may fail for noncyclic finite extensions of number fields, which, in turn, entails the failure of the cohomological principle for the corresponding norm torus. Later, examples of the failure of the cohomological Hasse principle were also found for semisimple groups—cf. [118, Ch. III, §4.7, Theorem 8] (note that such groups are, of course, neither simply connected nor adjoint). Nevertheless, it was proved in [11] using reduction theory for adelic groups (see also [12]) that when K is a number field, the map \(\theta _{G,V}\) is proper (i.e., the preimage of a finite set is finite) for any linear algebraic group G whenever V contains almost all valuations of K. The analogous statement for reductive groups over global fields of positive characteristic follows from results of Harder [55]. Informally, these results mean that while the localglobal principle for groups over global fields may fail, the deviation from it is always of finite size.
In the past 1015 years, the Hasse principle has been analyzed over certain other classes of fields, including function fields of padic curves, cf. [27, 52, 53]. This work has provided numerous examples where the localglobal principle holds for fields other than global. In this article, however, we would like to focus on the arithmetic situation, where the classical results over global fields and our own work over higherdimensional global fields (see Sect. 7.3) strongly suggest that the following statement should be true.
Conjecture 6.1
Let G be a (connected) reductive algebraic group defined over a finitely generated field K, and let V be a divisorial set of places of K. Then the globaltolocal map \(\theta _{G , V}\) is proper. In particular, the Tate–Shafarevich set is finite.
Thus, Conjecture 6.1 expresses a broad expectation that the deviation from the localglobal principle should be finite in all situations involving reductive linear algebraic groups over a finitely generated field K with respect to any divisorial set of places V of K. We will discuss available results on Conjecture 6.1 in Sect. 7.3 At this point, we would like to indicate how the truth of Conjecture 5.7 automatically implies that of Conjecture 6.1 for adjoint groups (cf. [19, §6], [104, §7]).
To fix notations, let G be an absolutely almost simple simply connected algebraic group over a finitely generated K of “good” characteristic, and let V be a divisorial set of discrete valuations of K. Note that since V satisfies condition (A) (see Sect. 5.3), it follows that we can pick a finite subset \(V_0 \subset V\) so that G has good reduction at all \(v \in V \setminus V_0\). As we observed in Sect. 5.3, \(V \setminus V_0\) contains a divisorial set \(V'\). Thus, replacing V by \(V'\), we may assume that G has good reduction at all \(v \in V\). Now set \({\overline{G}}\) to be the corresponding adjoint group. Suppose and let \(G' = G(\xi )\) be the corresponding (inner) Kform of G. By our assumption, \(G' \times _K K_v \simeq G \times _K K_v\) for all \(v \in V\), and consequently \(G'\) has good reduction at all \(v \in V\). Therefore, assuming Conjecture 5.7, we conclude that the groups \(G(\xi )\) for form finitely many Kisomorphism classes. In cohomological terms, this means that that the image of under the canonical map \(H^1(K , {\overline{G}}) {\mathop {\longrightarrow }\limits ^{\alpha }} H^1(K , A_{K^{\mathrm{sep}}})\) is finite (as in Example 4.5, we denote by \(A_{K^{\mathrm{sep}}}\) the group of automorphisms of \(G \times _K K^{\mathrm{sep}}\)). But since \({\overline{G}} \simeq I_{K^{\mathrm{sep}}}\) has finite index in \(A_{K^{\mathrm{sep}}}\), the map \(\alpha \) has finite fibers, which yields the finiteness of (see [118, Ch. I, §§5.35.5]) for a discussion of the fibers of such maps in nonabelian cohomology).
Another fundamental finiteness property in number theory is the finiteness of the ideal class group of a number field. In the higherdimensional situation, one replaces the class group by the Picard group. While this group can be infinite, it is known that if X is a scheme that is normal and of finite type over \(\mathrm{Spec}({{\mathbb {Z}}})\), then the Picard group \(\mathrm{Pic}~X\) is finitely generated (cf. [62]). In order to transport this notion into the context of linear algebraic groups, one uses adeles.
So, let K be a field equipped with a set V of (pairwise inequivalent) discrete valuations, and let G be a linear algebraic Kgroup with a fixed matrix realization \(G \subset \mathrm {GL}_n\). For each \(v \in V\), we set
where \({\mathcal {O}}_v\) is the valuation ring in the completion \(K_v\). We then define the corresponding adelic group as
In other words, \(G({\mathbb {A}}(K , V))\) is the restricted (topological) product of the groups \(G(K_v)\) for \(v \in V\) with respect to the (open) subgroups \(G({\mathcal {O}}_v)\) (cf. [94, Ch. 5, §5.1] for the details). The product
is called the subgroup of integral adeles. Henceforth, we will assume that V satisfies condition (A) introduced in Sect. 5.3 (we recall that this condition holds automatically for a divisorial set of places of a finitely generated field). Then one has a diagonal embedding \(G(K) \hookrightarrow G({\mathbb {A}}(K , V))\), whose image is called the subgroup of principal adeles and which we will still denote by G(K). The set of double cosets
is called the class set of G (we should point out that the class set is sometimes defined using rational adeles rather than the full adeles we introduced above). The following examples link this definition with classical notions.
Example 6.2
Let \(G = {\mathbb {G}}_m\) be the 1dimensional Ksplit torus. Then \(G({\mathbb {A}}(K , V))\) is the group of ideles \({\mathbb {I}}(K , V)\) and \(G({\mathbb {A}}^{\infty }(K , V))\) is the subgroup of integral ideles
So, there is a natural bijection between the class set \(\mathrm {cl}(G, K, V)\) and the quotient \({\mathbb {I}}(K , V)/{\mathbb {I}}^{\infty }(K , V) K^{\times }\). On the other hand, it is easy to see that the latter quotient is isomorphic to the Picard group \(\mathrm {Pic}(K , V)\), which is defined as follows. Let \(\mathrm {Div}(V , K)\) be the free abelian on the set V, which we call the group of divisors. By virtue of condition (A), we can define a group homomorphism
the image of which is the called the subgroup of principal divisors \(\mathrm {P}(K , V)\). We set
The isomorphism \({\mathbb {I}}(K , V)/{\mathbb {I}}^{\infty }(K , V)K^{\times } \simeq \mathrm{Pic}(K,V)\) is then induced by the map
If K is the fraction field of a Dedekind domain A and V is the set of discrete valuations of K corresponding to the maximal ideals of A, then \(\mathrm {Pic}(K , V)\) is precisely the ideal class group of A. More generally, if K is the function field of an integral regular scheme X and V is the set of discrete valuations associated with prime divisors of X, then \(\mathrm {Pic}(K , V)\) coincides with the usual Picard group \(\mathrm {Pic}(X)\). So, it follows from classical results that \(\mathrm {Pic}(K , V)\) is finite in the first situation when A is the ring of Sintegers in a number field, and is finitely generated in the second situation when X is integral, normal, and of finite type over \(\mathrm{Spec}({{\mathbb {Z}}})\).
Example 6.3
Let K be a number field with the ring of integers \({\mathcal {O}}\), and let V be the set of all nonarchimedean valuations of K. Furthermore, let q be a nondegenerate quadratic form in n variables with coefficients in \({\mathcal {O}}\), and \(G = \mathrm {O}_n(q)\) be the corresponding orthogonal group. It is well known that in this case, there is a natural bijection between the class set \(\mathrm {cl}(G, K, V)\) and the set of classes in the genus of q (see [94, Ch. 8, Proposition 8.4] for the details and relevant definitions). We recall that when \(n \ge 3\) and q is indefinite (i.e., there exists an archimedean place \(v \in V^K_{\infty }\) that is either complex (\(K_v = {\mathbb {C}}\)) or is real and q is indefinite in the usual sense over \(K_v = {\mathbb {R}}\)), then \(\mathrm {cl}(G, K, V)\) has a natural structure of an abelian group, which is finite of order a power of 2 (see [94, Ch. 8, §8.2, Theorem 8.6]). On the contrary, if q is definite, then \(\mathrm {cl}(G, K, V)\) is a finite set which in general does not have a natural group structure and whose size can be made divisible by any given integer if one changes q to a rationally equivalent form (see [94, Ch. 8. §8.3, Theorem 8.9]).
More generally, it was shown by Borel [11] using reduction theory that if K is a number field and V is the set of all nonarchimedean valuations of K, then the class set \(\mathrm {cl}(G, K, V)\) is finite for any linear algebraic Kgroup G. This finiteness result was extended to global fields of positive characteristic by Conrad [29], who employed the theory of pseudoreductive groups developed by Conrad–Gabber–Prasad [32]. On the other hand, for \(G = {\mathbb {G}}_m\) over a finitely generated field K with V a divisorial set of places, \(\mathrm {cl}(G, K, V) \simeq \mathrm {Pic}(K , V)\) may be an infinite group, which is nevertheless finitely generated. For an arbitrary linear algebraic group, however, \(\mathrm {cl}(G, K, V)\) may not have a natural group structure, so no general finiteness condition on the class set can conceivably be stated on the basis of finite generation. We have proposed the following path towards a possible generalization, which consolidates the two cases discussed above and appears to be quite useful. First, we observe that if \(\mathrm {cl}(G, K, V)\) is either finite or a finitely generated group (in the presence of a natural group structure), then one easily shows that there exists a finite subset \(S \subset V\) such that the class set \(\mathrm {cl}(G, K, V\setminus S)\) reduces to a single element (see the argument following [23, Definition 3.4] for the details). This suggests the following condition on the triple (G, K, V):
Condition (T). There exists a finite subset \(S \subset V\) such that \(\vert \mathrm {cl}(G, K, V \setminus S) \vert = 1.\)
While one does not expect Condition (T) to hold for an arbitrary reductive algebraic group G over a general finitely generated field K and a divisorial set V, it is likely to be true for all G in certain important situations, including when

K is a 2dimensional global field (i.e., the function field of a smooth geometrically integral curve over a number field or the function field of a smooth geometrically integral surface over a finite field—see [22] and [64]) and V is a divisorial set of places; and

\(K = k(C)\), the function field of a smooth geometrically integral curve C over a finitely generated field k and V is the set of places of K associated with the closed points of C.
In fact, one can expect the following weaker property to be true even more generally: for a reductive group \(G \subset \mathrm {GL}_n\) over a finitely generated field K and a divisorial set of places V, there exists a finite subset \(S \subset V\) such that
Finally, we would like to observe that although at this point no direct connections between Conjectures 5.7 and 6.1 and Condition (T) have been established in the general case, ideas involving Condition (T) were used in a very essential way in the proof of Conjecture 6.1 for tori—see Sect. 7.1 for a brief discussion and [106] for complete details. Moreover, as we pointed out in [23], Condition (T) can also be used in the analysis of some finiteness questions for unramified cohomology in degree 3 and hence the genus problem for the groups of type \(\textsf {G}_2\) (we will touch upon some aspects of these issues in Sects. 7.2 and 8.4).
Results
In this section, we will give an overview of the currently available results on Conjectures 5.7 and 6.1 as well as Condition (T). We begin with the case of algebraic tori, where all conjectures were recently resolved in [106]. We then move on to absolutely almost simple algebraic groups, where there has been notable progress (see, in particular, [21] and [22]), but much work still remains to be done.
Algebraic tori
First, we have the following finiteness result for tori with good reduction, which completely settles Conjecture 5.7 in this case.
Theorem 7.1
[106, Theorem 1.1] Let K be a finitely generated field of characteristic zero and V be a divisorial set of places of K. Then for any integer \(d \ge 1\), the set of Kisomorphism classes of ddimensional Ktori that have good reduction at all \(v \in V\) is finite.
As we discussed in Examples 4.2(a) and 4.4, all ddimensional Ktori are Kforms of the ddimensional Ksplit torus \(({\mathbb {G}}_m)^d\), and their Kisomorphism classes are classified by the minimal splitting field L and the equivalence class of a faithful representation \(\mathrm {Gal}(L/K) \rightarrow \mathrm {GL}_d({\mathbb {Z}})\). Furthermore, it is a general fact (see [94, Ch. 4, §4.4, Theorem 4.3]) that a given finite group has only finitely many equivalence classes of integral representations in each dimension. So, it is enough to prove that there are only finitely many possibilities for the extension L/K. The key observations here are, first, that the degree [L : K] is bounded by a constant depending only on d (as are the orders of finite subgroups of \(\mathrm {GL}_d({\mathbb {Z}})\)) and, second, that this extension is unramified at all \(v \in V\). Let X be a model of K used to define V. Then the fact that there are only finitely many possible extensions L/K is derived from the result that the fundamental group of X is “small,” i.e., satisfies Serre’s condition (F) (see [51]), which can be viewed as a higherdimensional analogue of the Hermite–Minkowski theorem that we have already mentioned several times in our previous discussion.
It should be mentioned that Theorem 7.1 is no longer true in positive characteristic. Indeed, an infinite family of pairwise nonisomorphic Ktori with good reduction can be constructed using Artin–Schreier extensions over the global field \(K = {\mathbb {F}}_p(t)\) (where \({\mathbb {F}}_p\) is the field with p elements), with V being the set of discrete valuations corresponding to all monic irreducible polynomials \(f(t) \in {\mathbb {F}}_p[t]\) (as in Example 3.6(b)).
Now, let G be an absolutely almost simple simply connected Kgroup, and let L/K be the minimal Galois extension over which G becomes an inner form of the split group (see the discussion in Example 4.5). If G has good reduction at a discrete valuation v of K, then L/K is unramified at v. Replacing the use of the Hermite–Minkowski theorem in the proof of Proposition 5.2 with its higherdimensional analogue, we see that in the case where K is a finitely generate field of characteristic zero and V is a divisorial set of places, there are only finitely many continuous homomorphisms
to the symmetry group S of the Dynkin diagram of G such that \(L(\omega ) = (K^{\mathrm {sep}})^{\ker \omega }\) is unramified at all \(v \in V\). Let \(\omega _1, \ldots , \omega _r\) be representatives of the conjugacy classes of such homomorphisms, and let \(G_1, \ldots , G_r\) be the corresponding Kquasisplit forms of G. Then any Kform of G that has good reduction at all \(v \in V\) is an inner form of one of the \(G_i\). This implies that it is enough to prove Conjecture 5.7 for inner forms of all quasisplit Kgroups. In fact, this conclusion remains valid over finitely generated fields of positive characteristic \(p > 3\), but some care needs to be exercised in characteristic 2 and 3, even as far as the formulation of Conjecture 5.7 is concerned—cf. [106, Remark 2.6] for further details.
Finally, if G is a nonsemisimple reductive group over a finitely generated field K, then applying Theorem 7.1 to the maximal central torus, we see that it is enough to prove Conjecture 5.7 for the derived group of G (cf. [10, Ch. IV, §14.2, Proposition] for the relevant structural information), which reduces the conjecture to the semisimple case. In turn, the semisimple case can essentially be reduced to the case of absolutely almost simple simply connected groups, which we will consider in the next subsection. Now, however, we turn our attention to the finiteness of the Tate–Shafarevich group of an algebraic torus.
Theorem 7.2
[106, Theorem 1.2] Let K be a finitely generated field and V be a divisorial set of places of K. Then for any algebraic Ktorus T, the Tate–Shafarevich group
is finite.
As we already indicated in Sect. 6, this result can be interpreted in a variety of concrete situations. For example, let L be a finite separable extension of a finitely generated field K with a divisorial set of places V. Let
be the group of “local norms.” Then Theorem 7.2 applied to the norm torus \(T = \mathrm {R}^{(1)}_{L/K}({\mathbb {G}}_m)\) implies that the quotient \({\mathcal {N}}/ N_{L/K}(L^{\times })\) of \({\mathcal {N}}\) modulo the subgroup of “global norms” is always finite.
It should be noted that the classical proof of the finiteness of when K is a number field and V is set of all places of K (including the archimedean ones) relies on Tate–Nakayama duality (see [143, Ch. 4, §11.3, Theorem 6] and the subsequent discussion), which is not available in the general situation. So, in [106], we gave two different proofs of Theorem 7.2. The first one requires an additional assumption on the characteristic of K, but it develops an approach that is applicable in some situations involving fields that are not finitely generated. The second one systematically uses adelic techniques in the context of arbitrary finitely generated fields and their divisorial sets of valuations, which, to the best of our knowledge, have not been previously employed in this generality. More specifically, the argument relies on the validity of Condition (T) in the present case—see Theorem 7.3. Our second proof demonstrates, in particular, that in the classical situation where K is a number field, the finiteness of the Tate–Shafarevich group can be established without Tate–Nakayama duality, and is actually a direct consequence of two basic facts: the finite generation of the group of Sunits and the finiteness of the class number. Moreover, this argument applies to cohomology groups in all degrees and yields the following result: Let K be a finitely generated field, V be a divisorial set of places, and F/K be a finite separable extension. Then for any \(i \ge 1\), the group
is finite (here for each \(v \in V\), we pick one extension w to F).
The following result verifies Condition (T) for groups more general than tori.
Theorem 7.3
[106, Theorem 3.4] Let K be a finitely generated field and V be a divisorial set of places of K. Then any linear algebraic Kgroup G whose connected component is a torus satisfies Condition (T).
We would now like to mention analogues of Theorems 7.1 and 7.2 for function fields of curves over base fields of type (F) (see Sect. 5.2 for the definition). To fix notations, suppose k is a field of type (F) and having characteristic 0. Let \(K = k(C)\) be the function field of a smooth geometrically integral curve C over k, and let V be the set of discrete valuations of K associated with the closed points of C. We then have the following statement concerning tori with good reduction.
Theorem 7.4
With notations as above, for each \(d \ge 1\), there are finitely many Kisomorphism classes of ddimensional Ktori that have good reduction at all \(v \in V\).
The proof of this (unpublished) result proceeds essentially along the same lines as that of Theorem 7.1 sketched above, with the key input being the fact that the étale fundamental group of C is small. For this, we fix an algebraic closure \({\bar{k}}\) of k, set \({\overline{C}} = C \times _{\mathrm{Spec}(k)} \mathrm{Spec}({\bar{k}})\), and let \({\bar{x}}\) be the corresponding geometric point of \({\overline{C}}.\) We then have the following standard exact sequence of profinite groups
induced by the natural maps \({\overline{C}} \rightarrow C\) and \(C \rightarrow \mathrm{Spec}(k).\) By our assumption, \(\mathrm {Gal}({\bar{k}}/k)\) is small. Furthermore, it is well known that \(\pi _1({\overline{C}}, {\bar{x}})\) is topologically finitely generated (see, for example, [131, Theorem 4.6.7]), and hence is small by [118, Ch. III, §4, Proposition 9]. Applying [51, Lemma 2.7], we conclude that \(\pi _1(C, {\bar{x}})\) is small, as required.
Next, the first proof of Theorem 7.2 alluded to above, together with the finiteness results for the unramified cohomology of tori obtained by the second author in [112, §5], yield the following.
Theorem 7.5
With notations as above, for any Ktorus T, the Tate–Shafarevich group is finite.
Absolutely almost simple groups: Conjectures 5.5 and 5.7.
We begin with the results that have been obtained so far on the Main Conjecture 5.7 for absolutely almost simple groups.
First, for inner forms of type \({\mathsf {A}}_n\), Conjecture 5.7 has been proved completely.
Theorem 7.6
Let K be a finitely generated field and V be a divisorial set of places of K. Then for any \(n \ge 2\) that is prime to \(\mathrm {char}\, K\) and any simply connected inner form G of type \(\textsf {A}_{n1}\), the number of Kisomorphism classes of inner Kforms of G that have good reduction at all \(v \in V\) is finite.
We recall that the group G in the statement of the theorem is of the form \(\mathrm {SL}_{1,A}\)—the algebraic group associated with the group of elements of reduced norm 1 in a central simple Kalgebra A of degree n over K (cf. [94, Ch. 2, §2.3, Proposition 2.17]). As we remarked in Example 3.8(b), such a group has good reduction at a discrete valuation v of K if and only if the algebra A is unramified at v, i.e., there exists an Azumaya algebra \({\mathcal {A}}\) over the valuation ring \({\mathcal {O}}_v \subset K_v\) such that there is an isomorphism of \(K_v\)algebras
A detailed discussion of Azumaya algebras can be found, for example, in [68, 78, Ch. IV], and [114, Ch. 2]. For our purposes, we merely recall that an \({\mathcal {O}}_v\)algebra \({\mathcal {A}}\) is called an Azumaya algebra if it is a free \({\mathcal {O}}_v\)module of finite rank and if the canonical homomorphism of \({\mathcal {O}}_v\)algebras
(where \({\mathcal {A}}^{\mathrm {op}}\) denotes the opposite algebra) that sends a simple tensor \(a \otimes a'\) to the endomorphism \((x \mapsto axa')\), is an isomorphism. In this case, the quotient \({\mathcal {A}}/{\mathfrak {p}}_v{\mathcal {A}}\) (where \({\mathfrak {p}}_v \subset {\mathcal {O}}_v\) is the valuation ideal) is a central simple algebra over the residue field \(K^{(v)} = {\mathcal {O}}_v/{\mathfrak {p}}_v\).
The key input in the proof of Theorem 7.6 is a finiteness result for the unramified Brauer group, which we now describe. First, we recall that the Brauer group \(\mathrm {Br}(K)\) of a field K consists of the Brauer equivalence classes of finitedimensional central simple Kalgebras (for the details of this construction, the reader can consult [39, 90], or [114]). Given a central simple algebra A over K, we denote by [A] the corresponding class in \(\mathrm {Br}(K).\) It is well known that if A is a central simple Kalgebra of degree n, then [A] is annihilated by multiplication by n in \(\mathrm {Br}(K)\), i.e., belongs to the ntorsion subgroup \({}_n\mathrm {Br}(K)\). Furthermore, given a discrete valuation v of K, we say that \(a \in \mathrm {Br}(K)\) is unramified at v if it can be represented by a central simple Kalgebra A that is unramified at v, as defined above. Now, if V is a set of discrete valuations of K, we let \(\mathrm {Br}(K)_V\) denote the subgroup of \(\mathrm {Br}(K)\) consisting of elements that are unramified at all \(v \in V\) (this group is usually referred to as the unramified Brauer group of K with respect to V).
With these preliminaries, we now come to the following finiteness statement from which Theorem 7.6 is derived (cf. [20]):
Let K be a finitely generated field and V be a divisorial set of places of K. Then for any \(n \ge 1\)that is prime to \(\mathrm {char}\, K\), the group \({}_n\mathrm {Br}(K)_V = {}_n\mathrm {Br}(K) \cap \mathrm {Br}(K)_V\)is finite.
Now, with notations as in Theorem 7.6, any inner Kform \(G'\) of G that has good reduction at all \(v \in V\) is of the form \(\mathrm {SL}_{1 , A'}\), where \(A'\) is a central simple Kalgebra of degree n with \([A'] \in {}_n\mathrm {Br}(K)_V\). Since the latter group is finite, the required finiteness in the theorem follows from the fact that if \(A_1\) and \(A_2\) are two central simple Kalgebras of degree n and \([A_1] = [A_2]\) in \(\mathrm {Br}(K)\), then the algebras \(A_1\) and \(A_2\) are isomorphic, implying that the corresponding algebraic groups \(G_1 = \mathrm {SL}_{1 , A_1}\) and \(G_2 = \mathrm {SL}_{1 , A_2}\) are also isomorphic.
The proof of the finiteness of \({}_n\mathrm {Br}(K)_V\) relies on cohomological techniques. We will not go into the details here and will only briefly indicate the basic setup. To begin with, it is well known that \(\mathrm {Br}(K)\) can be identified with \(H^2(K , (K^{\mathrm{sep}})^{\times })\), and under this identification, for any \(n \ge 1\) that is prime to \(\mathrm {char}\, K\), the ntorsion subgroup \({}_n\mathrm {Br}(K)\) corresponds to \(H^2(K , \mu _n)\), where \(\mu _n\) is the group of nth roots of unity in \(K^{\mathrm{sep}}\) (cf. [90, Ch. 14]). Furthermore, for any discrete valuation v of K such that \(\mathrm {char}\, K^{(v)}\) is prime to n, there exists a residue map
where \({\mathbb {Z}}/n{\mathbb {Z}}\) denotes \(\mu _n\) with the trivial Galois action—see [48, §6.8] for the details. It turns out that an element \(a \in H^2(K , \mu _n)\) is unramified at v as defined above if and only if \(\partial _v^2(a) = 0\). So, assuming that n is prime to \(\mathrm {char}\, K^{(v)}\) (which we can always do in terms of proving Theorem 7.6 as n is prime to \(\mathrm {char}\, K\)), one can define the unramified cohomology group
and then identify it with \({}_n\mathrm {Br}(K)_V\). The proof then proceeds via a careful analysis of the group \(H^2(K, \mu _n)_V.\) Let us point out that in [20], we not only established the finiteness of the latter group, but also provided explicit estimations of its order in certain situations.
Next, we will discuss results on Conjecture 5.7 for groups of several other types over a special class of finitely generated fields—the socalled twodimensional global fields. Following Kato [64], by a twodimensional global field, we mean either the function field \(K = k(C)\) of a smooth geometrically integral curve C over a number field k, or the function field \(K = k(S)\) of a smooth geometrically integral surface S over a finite field k.
Theorem 7.7
[22, Theorem 1.1] Let K be a twodimensional global field of characteristic \(\ne 2\) and let V be a divisorial set of places of K. Fix an integer \(n \ge 5\). Then the set of Kisomorphism classes of spinor groups \(G = \mathrm {Spin}_n(q)\) of nondegenerate quadratic forms in n variables over K that have good reduction at all \(v \in V\) is finite.
Whereas the proof of Theorem 7.6 was based on the study of the Brauer group, the proof of Theorem 7.7 requires an analysis of the Witt ring (we refer the reader to [37] and [70] for the construction and basic properties of the Witt ring). We will assume henceforth that \(\mathrm {char}\, K \ne 2\), and denote by W(K) the Witt ring of K and by I(K) its fundamental ideal. For a nondegenerate quadratic form q over K, we set [q] to be the corresponding class in W(K). Now, a consequence of Voevodsky’s proof of Milnor’s conjecture is that for \(d \ge 1\), there are natural isomorphisms of abelian groups
(cf. [86]). On the other hand, as above, for any discrete valuation v such that \(\mathrm {char}\, K^{(v)} \ne 2\) and any \(d \ge 1\), there exists a residue map
which actually factors through the restriction map \(H^d(K , \mu _2) \rightarrow H^d(K_v , \mu _2)\) (where, as before, \(K_v\) denotes the completion of K at v). Then one says that \(a \in H^d(K , v)\) is unramified at v if \(\partial _v^d(a) = 0\). Moreover, if K is equipped with a set V of discrete valuations such that \(\mathrm {char}\, K^{(v)} \ne 2\) for all \(v \in V\), one defines the corresponding unramified cohomology group by
To factor in good reduction, one proves the following technical statement. Let v be a discrete valuation of K such that \(\mathrm {char}\, K^{(v)} \ne 2\). We let \(W_0(K_v)\) denote the subring of \(W(K_v)\) generated by the classes of 1dimensional forms \(ux^2\) with \(u \in U_v = {\mathcal {O}}_v^{\times }\). Then we show the following:
If q is a nondegenerate form over \(K_v\)such that \([\lambda q] \in I(K_v)^d \cap W_0(K_v)\)(where \(d \ge 1\)) for some \(\lambda \in K_v^{\times }\), then \([q] \in I(K_v)^d\)and
is unramified at v.
(See [22, Lemma 3.3]). On the other hand, as we remarked in Example 3.8(c), the spinor group \(G = \mathrm {Spin}_n(q)\) has good reduction at v if and only if q is \(K_v\)equivalent to a quadratic form \(\lambda q_0\) with \(\lambda \in K_v^{\times }\) and \(q_0 = u_1 x_1^2 + \ldots + u_n x_i^2\), where \(u_i \in U_v\) for all \(i = 1, \ldots , n\).
Suppose now that K is a finitely generated field of characteristic \(\ne 2\) equipped with a divisorial set of places V. Without loss of generality we may assume that \(\mathrm {char}\, K^{(v)} \ne 2\) for all \(v \in V\). Using the above discussion, in conjunction with some facts from the theory of quadratic forms (most notably, the Hauptsatz—see [70, Ch. X, §5]), one shows that to prove Theorem 7.7, it is enough to establish the finiteness of the unramified cohomology groups \(H^d(K, \mu _2)_V\) for all \(d \ge 1\) (see [22, Theorem 2.1] for a more precise statement). So, to complete the argument, we prove the required finiteness over twodimensional global fields of characteristic \(\ne 2\). Here, we will comment on this fact assuming that \(\mathrm{char}~K =0\) (for the positive characteristic case, the reader can consult [22, §7]). Then the finiteness of \(H^1(K , \mu _2)_V\) is a standard result and, as mentioned above, the finiteness of \(H^2(K , \mu _2)_V\) was established in the course of the proof of Theorem 7.6. On the other hand, the finiteness of \(H^d(K , \mu _2)_V\) for \(d \ge 4\) can be derived from results of PoitouTate (cf. [22, Proposition 4.2]). The most challenging case is the finiteness of \(H^3(K, \mu _2)_V\), for which we gave two proofs in [22]. One proof makes use of several powerful results, first and foremost, those of Kato [64] on cohomological Hasse principles. The second proof requires considerably less input; in particular, it does not rely on Kato’s localglobal principle, but instead is based on a modification of Jannsen’s proof of the latter [58]. This argument appears to be more amenable to generalizations in the spirit of Jannsen’s proof [59] of Kato’s conjecture on the localglobal principle for higherdimensional varieties, which extended his original argument in [58].
The above sketches of the proofs of Theorems 7.6 and 7.7 indicate an intimate connection between Conjecture 5.7 and finiteness properties of unramified cohomology. The analysis of unramified cohomology in general and of unramified \(H^3\) in particular, in both arithmetic and geometric situations, is a major direction of independent interest, which, however, lies beyond the scope of the current paper. We refer the reader to [25] for an informative survey of this subject.
Next, we should mention that Theorem 7.7 yields similar results for some other types of groups. Here is the statement for simply connected outer forms of type \(\textsf {A}_{n1}\) that split over a quadratic extension L/K. We recall that such forms are obtained as follows. Let L/K be a quadratic extension with a nontrivial automorphism \(\tau \), and let h be a nondegenerate \(\tau \)hermitian form of dimension \(n \ge 2\). Then the groups in question are the special unitary groups \(G = \mathrm {SU}_n(L/K , h)\) (cf. [94, Ch. 2, §2.3.4]).
Theorem 7.8
[22, Theorem 8.1] Let K be a twodimensional global field of characteristic \(\ne 2\) and V be a divisorial set of places of K. Fix a quadratic extension L/K, and let \(n \ge 2\). Then the number of Kisomorphism classes of special unitary groups \(G = \mathrm {SU}_n(L/K , h)\) of nondegenerate hermitian L/Kforms in n variables that have good reduction at all \(v \in V\) is finite.
Since the number of quadratic extensions L/K that are unramified at all \(v \in V\) is finite, Theorem 7.8 in effect yields the finiteness of the number of Kisomorphism classes of special unitary groups with good reduction at all \(v \in V\) of ndimensional nondegenerate hermitian forms associated with all quadratic extensions L/K.
A result similar to Theorem 7.8 is also valid for the special unitary groups \(G = \mathrm {SU}_n(D , h)\) of nondegenerate hermitian forms of dimension \(n \ge 2\) over a quaternion division algebra D with center K with the canonical involution, which are precisely the absolutely almost simple simply connected groups of type \(\textsf {C}_n\) that split over a quadratic extension of K.
Over twodimensional global fields, we also have the following finiteness result for groups of type \(\textsf {G}_2\).
Theorem 7.9
[22, Theorem 9.1] Let K be a twodimensional global field of characteristic \(\ne 2\) and V be a divisorial set of places of K. The number of Kisomorphism classes of Kgroups of type \(\textsf {G}_2\) that have good reduction at all \(v \in V\) is finite.
These preceding results suggest that the proof of the following general fact should be within reach in the near future: if K is a twodimensional global field, then for each type, there are only finitely many Kisomorphism classes of Kforms that split over a quadratic extension of K and have good reduction at all discrete valuations in some divisorial set of places of K.
The very recent results on finiteness of unramified cohomology obtained in [106, §5] make it possible to extend the above results beyond the class of twodimensional global fields. We will just mention that Theorems 7.7 and 7.8 remain valid for a purely transcendental extension \(K = k(x_1 , x_2)\) of transcendence degree two of a number field k, while Theorem 7.9 remains valid for K a purely transcendental extension of a number field k of any (finite) transcendence degree as well as the function field of a Severi–Brauer variety corresponding to a central simple algebra of either odd degree or degree 2 (in all cases V can be any divisorial set of places of the field at hand).
To close this subsection, we will briefly survey the results on Conjecture 5.5 in the case where \(K = k(C)\) is the function field of a smooth geometrically integral curve C over a base field k that satisfies conditions \((\mathrm {F})\) or \((\mathrm {F}'_m)\) (see Sect. 5.2 for the definitions), and V is the set of discrete valuations of K associated with the closed points of C.
Theorem 7.10
With notations as above, assume that k is of characteristic \(\ne 2\) and satisfies condition \((\mathrm {F}'_2)\). Then the number of Kisomorphism classes of

spinor groups \(G = \mathrm {Spin}_n(q)\) of nondegenerate quadratic forms q over K of dimension \(n \ge 5\),

special unitary groups \(G = \mathrm {SU}_n(L/K , h)\) of nondegenerate hermitian forms of dimension \(n \ge 2\) over a quadratic extension L/K with respect to its nontrivial automorphism,

special unitary groups \(G = \mathrm {SU}_n(D , h)\) of nondegenerate hermitian forms of dimension \(n \ge 1\) over a central quaternion division algebra D over K with respect to the canonical involution,

groups of type \(\textsf {G}_2\)
that have good reduction at all \(v \in V\) is finite.
The proof proceeds along the same lines as the proofs of the Theorems 7.7–7.9 and relies on the finiteness results for unramified cohomology established in [112]. We should point out that the list of groups in Theorem 7.10 was recently augmented by S. Srinivasan [127]. She proved the corresponding finiteness statement in the same situation as considered in the theorem for the universal covers \(G = \widetilde{\mathrm {SU}}_n(D , h)\) of the special unitary groups \(\mathrm {SU}_n(D , h)\), where D is a central quaternion Kalgebra and h is a nondegenerate skewhermitian form of dimension \(n \ge 4\) with respect to the canonical involution. Recall that these groups are of type \(\textsf {D}_n\), and in fact all simply connected Kgroups of this type that split over a quadratic extension of K are of the form \(\widetilde{\mathrm {SU}}_n(D , h)\) (cf. [94, Ch. 2, §2.3.4]). So, combining Theorem 7.10 with the main result of [127], we see that for each \(r \ge 1\), there exist only finitely many Kisomorphism classes of absolutely almost simple simply connected algebraic Kgroups G of rank r that belong to one of the types \(\textsf {A}_n, \textsf {B}_n, \textsf {C}_n, \textsf {D}_n\) or \(\textsf {G}_2\), split over a quadratic extension of K, and have good reduction at all \(v \in V\). The types \(\textsf {E}_6, \textsf {E}_7, \textsf {E}_8\), and \(\textsf {F}_4\) have not been considered yet.
Absolutely almost simple groups: Conjecture 6.1
As we already remarked in Sect. 6, the truth of Conjecture 5.7 for the inner forms of an absolutely almost simple simply connected group G automatically implies the properness of the globaltolocal map
for the corresponding adjoint group \({\overline{G}}\). Thus, it follows from Theorem 7.6 that if K is a finitely generated field equipped with a divisorial set of places V, then for a central simple Kalgebra A of degree n which is prime to \(\mathrm {char}\, K\), the map
is proper; in particular, the map \(\theta _{\mathrm {PSL}_n , V} :H^1(K , \mathrm {PSL}_n) \longrightarrow \prod _{v \in V} H^1(K_v , \mathrm {PSL}_n)\) is proper.^{Footnote 6} It should be pointed out, however, that the properness of the globaltolocal map for the adjoint group does not automatically imply its properness for the corresponding simply connected group (or another isogenous group). In fact, the properness of the map globaltolocal \(\theta \) remains an open problem for the group \(G = \mathrm {SL}_{1 , A}\) in the general case. We have the following partial result over twodimensional global fields.
Theorem 7.11
[22, Theorem 5.7] Let K be a twodimensional global field, V a divisorial set of places of K, and n a squarefree integer that is prime to \(\mathrm {char}\, K\). Then for a central simple Kalgebra A of degree n and \(G = \mathrm {SL}_{1 , A}\), the map
is proper.
Next, for odd integers \(n \ge 5\), the adjoint group for \(G = \mathrm {Spin}_n(q)\) is \({\overline{G}} = \mathrm {SO}_n(q)\). So, in this case, Theorem 7.7 automatically implies the properness of the map
when K is a twodimensional global field of characteristic \(\ne 2\) and V is a divisorial set of places of K. In fact, one proves that this map is proper for all \(n \ge 5\)—see [22, Theorem 1.3]. Moreover, still assuming that K is a twodimensional global field and V is a divisorial set of places of K, we establish in [22] that this map is proper for special unitary groups \(\mathrm {SU}_n(L/K , h)\) and \(\mathrm {SU}_n(D , h)\) over quadratic extensions and quaternion algebras, and for groups of type \(\textsf {G}_2\). In addition, in [106], we show that the map \(\theta _{G , V}\) is proper for the groups \(\mathrm {SO}_n(q)\), \(\mathrm {SU}_n(L/K , h)\), and \(\mathrm {SU}_n(D , h)\) in the previous notations when K is a purely transcendental extension \(k(x_1 , x_2)\) of transcendence degree two of a number field k and V is any divisorial set of places of K. The next result, also obtained in [106], deals with purely transcendental extensions of number fields of any (finite) transcendence degree.
Theorem 7.12
Let k be a number field and suppose \(K = k(x_1, \ldots , x_r)\) is a purely transcendental extension of k or \(K = k(X)\) is the function field of a Severi–Brauer variety X over k associated with a central division algebra D over k of degree \(\ell \), and let V be a divisorial set of places of K. Then in each of the following situations

\(G = \mathrm {SL}_{1 , A}\) where A is a central simple Kalgebra of a squarefree degree m such that k contains a primitive mth root of unity, and either m is relatively prime to \(\ell \) or \(\ell \) is a prime number if K is the function field of a Severi–Brauer variety;

G is a simple algebraic group of type \(\textsf {G}_2\) and either \(\ell \) is odd or \(\ell = 2\) if K is the function field of a Severi–Brauer variety,
the globaltolocal map \(\theta _{G , V} :H^1(K , G) \rightarrow \prod _{v \in V} H^1(K_v , G)\) is proper.
We conclude with the following theorem, which collects available results on the properness of the globaltolocal map over function fields of curves. The proofs are based on a combination of the arguments used to establish properness in [22], together with the finiteness results for unramified cohomology obtained in [112].
Theorem 7.13
Let \(K = k(C)\) be the function field of a smooth geometrically integral curve C over a field k, and let V be the set of discrete valuations of K associated with the closed points of C. In each of the following situations:

\(G = \mathrm {SL}_{1 , A}\), where A is a central simple Kalgebra of a squarefree degree n prime to \(\mathrm {char}\, k\) and such that k satisfies condition \((\mathrm {F}'_n)\);
and assuming that k is of characteristic \(\ne 2\) and satisfies \((\mathrm {F}'_2)\)

\(G = \mathrm {SO}_n(q)\), where q is a nondegenerate quadratic form over K of dimension \(n \ge 5\);

\(G = \mathrm {SU}_n(L/K , h)\), where h is nondegenerate hermitian form of dimension \(n \ge 2\) over a quadratic extension L/K with respect to its nontrivial automorphism;

\(G = \mathrm {SU}_n(D , h)\), where h is a nondegenerate hermitian form of dimension \(n \ge 1\) over a central quaternion division Kalgebra with respect to the canonical involution’

G is of type \(\textsf {G}_2\)
the globaltolocal map \(\theta _{G , V} :H^1(K , G) \rightarrow \prod _{v \in V} H^1(K_v , G)\) is proper.
Condition (T)
As mentioned previously, we have been able to establish Condition (T) for algebraic tori in all situations—see Theorem 7.3 for the precise statement. However, for general reductive groups, the analysis of Condition (T) is only unfolding. We begin with the following statement over function fields of curves.
Theorem 7.14
[23, Theorem 4.1] Let \(K = k(C)\) be the function field of a smooth geometrically integral affine C curve over a finitely generated field k, and let V be the set of discrete valuations of K associated with closed points of C. Then Condition (T) with respect to V holds for any connected reductive split Kgroup G.
Using considerations involving strong approximation, the proof of this result essentially reduces to verifying Condition (T) for a maximal split torus of G, where it follows from the finite generation of the Picard group \(\mathrm {Pic}(K , V)\) (note that V can be included in a divisorial set of places of K). We should point out that questions about strong approximation over fields other than global are interesting in their own right, and we will briefly comment on the initial steps in their study in Sect. 10.1.
In the higherdimensional situation, the investigation of Condition (T) is expected to be a very challenging problem. Indeed, as we will see shortly, already in the case \(G = \mathrm {GL}_n\), it is related to a famous conjecture of H. Bass [5], which has not seen much progress since it was posed in 1972. Now, it makes sense to consider this problem first in the more general context of commutative algebra without any finite generation assumptions. So, let R be a noetherian integral domain that is integrally closed in its field of fractions K. We denote by \(\mathrm {P}\) the set of height one primes of R and let V be the associated set of discrete valuations of K. As in Sect. 6, we set \(G({\mathbb {A}}(K , V))\) to be the corresponding adele group, and let \(G({\mathbb {A}}^{\infty }(K , V))\) and G(K) denote the subgroups of integral and principal adeles, respectively. We first observe that for \(G = \mathrm {GL}_n\) the class set
has the following interpretation in terms of reflexive Rmodules. Given an Rmodule M, we let \(M^* = \mathrm {Hom}_R(M , R)\) denote the dual module. Then there is a natural homomorphism of Rmodules \(M \rightarrow M^{**} = (M^*)^*\), and M is called reflexive if this homomorphism is an isomorphism. An Rsubmodule M of \(W = K^n\) is called a lattice if it is finitely generated and contains a Kbasis of W. We let \(\textsf {Refl}_n(R)\) (resp., \(\textsf {Proj}_n(R)\)) denote the set of isomorphism classes of lattices in \(W = K^n\) that are reflexive (resp., projective) Rmodules. One proves that there is a natural bijection between the class set \(\mathrm {cl}(\mathrm {GL}_n, K, V)\)and \(\textsf {Refl}_n(R)\)—see [23, Proposition B.2]. (To be more precise, one actually constructs a bijection between the class set defined using rational adeles and \(\textsf {Refl}_n(R)\), but since \(\mathrm {GL}_n\) has weak approximation with respect to any finite set of places, the full adeles and the rational adeles result in the same class set.) We note that in general, \(\textsf {Proj}_n(R)\) is a proper subset of \(\textsf {Refl}_n(R)\), but it is known that if R is a regular integral domain of Krull dimension \(\le 2\), then \(\textsf {Refl}_n(R) = \textsf {Proj}_n(R)\) for all \(n \ge 1\) (cf. [115, Proposition 2] and [119, Corollary 6, p. 78]).
Next, let \(K_0(R)\) be the Grothendieck group of R—cf., for example, [79, §1] for the definition and basic properties. The following conjecture was proposed by Bass ([6, §9.1]).
Conjecture 7.15
(Bass) Let R be a finitely generated \({\mathbb {Z}}\)algebra which is a regular ring. Then the group \(K_0(R)\) is finitely generated.
(We note that Bass actually conjectured the finite generation of all groups \(K_n(R)\)\((n \ge 0)\) for such R.)
The following statement reveals a rather surprising connection between Conjecture 7.15 and Condition (T).
Proposition 7.16
([23, Corollary 6.16]) Let R be an integral domain which is a finitely generated \({\mathbb {Z}}\)algebra and a regular ring of Krull dimension \(\le 2\), and let V be the set of discrete valuations of the fraction field K associated with the height one prime ideals of R. If Conjecture 7.15 is true then \(G = \mathrm {GL}_n\) for \(n \ge 3\) satisfies Condition (T) with respect to V.
The proof relies on the equality \(\textsf {Refl}_n(R) = \textsf {Proj}_n(R)\) in the situation at hand, in conjunction with the following Cancellation Theorem due to Bass [5]: Let R be a noetherian commutative ring of Krull dimension \(d < \infty \), and let P and Q be finitely generated projective Rmodules of constant rank \(r > d\). If \(P \oplus F \simeq Q \oplus F\), with F free and finitely generated, then \(P \simeq Q\).
As we observed in [23, §3], Condition (T) can be used to show the finiteness of certain subgroups of unramified cohomology in degree 3 with \(\mu _2\)coefficients (which is needed, in particular, for the analysis of groups of type \({\mathsf {G}}_2\) with good reduction). However, to implement this approach, we need Condition (T) to hold not for \(\mathrm {GL}_n\), but rather for its Kforms \(\mathrm {GL}_{1 , A}\), where A is a central simple Kalgebra of degree n. For this, we developed in [23, Appendix C] a descent procedure that, under some additional assumptions, enables one to derive Condition (T) for \(\mathrm {GL}_{1 , A}\) from the fact that Condition (T) holds for \(\mathrm {GL}_n\) over a suitable finite Galois extension L/K that splits A.
While the investigation of Condition (T) is still in its initial stages, the range of its potential applications as well as its connections to various other problems, make this a very natural avenue for future work.
Applications to the genus problem
After surveying the available results on Conjectures 5.5, 5.7, and 6.1, we now turn to applications. In this section, we will consider applications of the Main Conjecture 5.7 to the genus problem for absolutely almost simple algebraic groups, while in the next one, we will relate it to the analysis of weakly commensurable Zariskidense subgroups of such groups, which in turn is linked to questions in differential geometry about isospectral and lengthcommensurable locally symmetric spaces. Historically, many of these developments can be traced back to [96]: although this work focused primarily on geometric problems, it became apparent that some of the ideas introduced therein should be considered in a much more general context. Eventually, research in this direction has led to the Main Conjecture. So, the reader should regard Sects. 8 and 9 of this article as an overview of the problems that motivated the Main Conjecture, and for which it provides a uniform approach (as it does with regard to some other issues, such as the localglobal principle).
The genus problem for algebraic groups
Given two reductive algebraic groups \(G_1\) and \(G_2\) over a field K, we say that \(G_1\) and \(G_2\)have the same isomorphism classes of maximal Ktori if every maximal Ktorus \(T_1\) of \(G_1\) is Kisomorphic to some maximal Ktorus \(T_2\) of \(G_2\), and vice versa.
Definition 8.1
Let G be an absolutely simple simply connected algebraic group over a field K. The genus \(\mathbf{gen }_K(G)\) of G is the set of Kisomorphism classes of (inner) Kforms \(G'\) of G that have the same isomorphism classes of maximal Ktori as G.
(We note that if K is a finitely generated field, then any Kform \(G'\) of G that has the same isomorphism classes of maximal Ktori as G is necessarily inner—cf. [100, Lemma 5.2]. So, restricting ourselves to just inner forms in this context does not result in a loss of generality.)
In the most general terms, the goal of the genus problem is to characterize the genus of a given group, which is of course crucial for understanding how two (absolutely almost simple simply connected) algebraic groups are related given the fact that they have the same isomorphism classes of maximal Ktori. From a variety of more precise questions that one can ask in connection with the genus problem, we will focus on the following two.
Question 8.2
When does \(\mathbf{gen }_K(G)\) reduce to a single element?
Question 8.3
When is \(\mathbf{gen }_K(G)\) finite?
The basic case where K is a number field was considered in [96, Theorem 7.5], where the following result was established (although the term “genus,” which appeared later, was not used).
Theorem 8.4
Let G be an absolutely almost simple simply connected algebraic group over a number field K. Then

(1)
\(\mathbf{gen }_K(G)\) is finite;

(2)
if G is not of type \(\textsf {A}_n\), \(\textsf {D}_{2n+1}\)\((n > 1)\), or \(\textsf {E}_6\), we have \(\vert \mathbf{gen }_K(G) \vert = 1\).
A noteworthy feature here is the completely different behavior of the groups of type \(\textsf {D}_n\) for n even and odd. This difference was worked out in [97] in the context of algebras with involution and in [42] in the context of algebraic groups. Another observation is that the types excluded in part (2) are precisely the types for which the automorphism of multiplication by \((1)\) of the corresponding root system is not in the Weyl group of the root system. In fact, these types are honest exceptions: indeed, it follows from [96, Theorem 9.12] that the genus for each of those types can be arbitrarily large.
Having addressed number fields, the next question is what can one expect regarding \(\mathbf{gen }_K(G)\) over more general fields? In order to provide some context for the conjecture that we will formulate in Sect. 8.4 and also to motivate the genus problem in general, we will next briefly review the genus problem for division algebras (see [19] for a more detailed account).
The genus problem for division algebras
Let \(D_1\) and \(D_2\) be two central division Kalgebras of degree n. We say that \(D_1\) and \(D_2\)have the same maximal subfields if a degree n field extension P/K admits a Kembedding \(P \hookrightarrow D_1\) if and only if it admits a Kembedding \(P \hookrightarrow D_2\). Then one can ask the following natural question:
\((*)\) Let \(D_1\) and \(D_2\) be central division algebras of the same degree. How are they related given the fact that they have the same maximal subfields?
This question can be seen as an extension of the following famous theorem of Amitsur [3]:
Theorem 8.5
(Amitsur) Let \(D_1\) and \(D_2\) be finitedimensional central division algebras over a field K that have the same splitting fields, i.e., for a field extension F, the algebra \(D_1 \otimes _K F\) is Fisomorphic to a matrix algebra \(\mathrm {M}_{n_1}(F)\) if and only if the algebra \(D_2 \otimes _K F\) is isomorphic to a matrix algebra \(\mathrm {M}_{n_2}(F)\). Then \(n_1 = n_2\) and the classes \([D_1]\) and \([D_2]\) in the Brauer group \(\mathrm {Br}(K)\) generated the same subgroup, \(\langle [D_1] \rangle = \langle [D_2] \rangle \).
The important point is that the proof of this result relies in a very essential way on infinite (nonalgebraic) extensions of K—namely, socalled generic splitting fields (concrete examples of which are function fields of Severi–Brauer varieties). So, one may wonder if it is possible to prove Amitsur’s Theorem, or perhaps another statement along the same lines, using only finite extensions of K. In other words, is it enough to assume only that \(D_1\) and \(D_2\) have the same finitedimensional splitting fields or just the same maximal subfields? It turns out that the conclusion of Amitsur’s Theorem is false in this setting. In fact, using the Albert–Brauer–Hasse–Noether theorem (see [90, Ch. 18, §18.4]), one can easily construct arbitrary large collections of pairwise nonisomorphic cubic division algebras having the same maximal subfields over number fields (the same construction actually works for division algebras of any degree \(d > 2\)—cf. [19, §1]). On the other hand, two quaternion division algebras over a number field that have the same quadratic subfields are necessarily isomorphic (as we will see in Sect. 8.3, this fact turns out to have important consequences for Riemann surfaces). Thus, even over number fields, the question \((*)\) appears to be interesting. Moreover, until about 10 years ago, no information at all was available on \((*)\) over any fields other than global. The following question along these lines was first asked in [96, Remark 5.4]:
Are quaternion division algebras over \({\mathbb {Q}}(x)\) determined uniquely up to isomorphism by their maximal subfields?
Shortly after it was formulated, this question was answered in the affirmative by D. Saltman. In subsequent work, he and S. Garibaldi [43] showed that the answer is still affirmative over the field of rational functions k(x), where k is any number field, and also in some other situations. This was the starting point of the investigation of question \((*)\) over fields more general than global, and we will now present the results that have been obtained since then. We note that a similar question, formulated in terms of finitedimensional splitting fields, was considered in [69].
For our discussion, it will be convenient to quantify the problem by introducing the notion of the genus of a division algebra (this terminology was suggested by L.H. Rowen).
Definition 8.6
Let D be a finitedimensional central division algebra over a field K. Then the genus \(\mathbf{gen }(D)\) of D is defined to be the set of classes \([D'] \in \mathrm {Br}(K)\) represented by central division Kalgebras \(D'\) having the same maximal subfields as D.
Let us remark that Definition 8.1 in the previous subsection is a straightforward generalization of this definition (which historically was given earlier) to algebraic groups, in which maximal subfields are replaced with maximal tori. Moreover, just as in the case of algebraic groups, the general question \((*)\) essentially reduces to the analysis of the genus \(\mathbf{gen }(D)\). In the present situation, we would like to focus on the analogues of Questions 8.2 and 8.3 in this situation, i.e.,

When does \(\mathbf{gen }(D)\)reduce to a single element? (Note that this is the case if and only if D is determined uniquely up to isomorphism by its maximal subfields.)

When is \(\mathbf{gen }(D)\)finite?
Over a number field K, the description of the Brauer group \(\mathrm {Br}(K)\) provided by the Albert–Brauer–Hasse–Noether Theorem enables one to resolve both questions. Namely, it turns out that the genus of every quaternion division algebra is trivial (i.e., reduces to a single element), while the genus of any division algebra of higher degree is nontrivial but always finite (see [19, Proposition 3.1] for the details).
Next, the following theorem for the field of rational functions was established in [105].
Theorem 8.7
(Stability Theorem) Assume that \(\mathrm {char}\, k \ne 2\). If \(\vert \mathbf{gen }(\Delta ) \vert = 1\) for any central division quaternion algebra \(\Delta \) over k, then \(\vert \mathbf{gen }(D) \vert = 1\) for any quaternion algebra D over k(x).
We note that the same statement remains valid for all division algebras having exponent two in the Brauer group (cf. [18]). On the other hand, \(\vert \mathbf{gen }(D) \vert > 1\) whenever D does not have exponent two since in that case, the opposite algebra \(D^{\small \mathrm {op}}\) is not isomorphic to D, but clearly has the same maximal subfields as D. Now, a consequence of Theorem 8.7 is that the genus of a quaternion algebra over the purely transcendental extension \(k(x_1, \ldots , x_r)\) of a number field k of any (finite) transcendence degree reduces to a single element. At the same time, the following question remains open.
Question 8.8
Does there exist a central quaternion division algebra D over a finitely generated field K of characteristic \(\ne 2\) having nontrivial genus?
Turning now to the question of the finiteness of the genus, we first should point out that over general fields, the genus \(\mathbf{gen }(D)\) can be infinite. Indeed, adapting a construction that has been suggested by a number of people, including M. Schacher, A. Wadsworth, M. Rost, S. Garibaldi and D. Saltman, J. Meyer [76] produced examples of quaternion algebras over “large” fields with infinite genus.^{Footnote 7} (By construction, these fields have infinite transcendence degree over the prime subfield.) In particular, there exist quaternion algebras over such fields with nontrivial genus (this was actually already observed by Garibaldi and Saltman [43]), indicating that the finite generation assumption in Question 8.8 cannot be omitted. Subsequently, S. Tikhonov [135] extended this approach to construct examples of division algebras of any prime degree having infinite genus. On the other hand, in [20] and [23], we proved the following.
Theorem 8.9
Let K be a finitely generated field. Then for any finitedimensional central division Kalgebra D, the genus \(\mathbf{gen }(D)\) is finite.
There are two versions of the proof of Theorem 8.9. Both of them rely on the analysis of ramification, but they differ in the amount of information about the unramified Brauer group that is needed as an input. The proof given in [20] requires the additional assumption that the degree n of the division algebra D is relatively prime to \(\mathrm {char}\, K\), and proceeds along the following lines. Fix a divisorial set of places V of K. First, since \(\mathrm {char}\, K\) is prime to n, we can assume without loss of generality that for each \(v \in V\), the characteristic of the residue field \(K^{(v)}\) is prime to n. Consequently, the residue map
that we encountered in our discussion of the proof of Theorem 7.6 is defined. One then shows that if \(D'\) is a central division algebra with \([D'] \in \mathbf{gen }(D)\) and
are the corresponding characters of the absolute Galois group of \(K^{(v)}\), then \(\ker \chi _v = \ker \chi '_v\) (see [18, Lemma 2.5]). In particular, we see that for D and \(D'\) as above, the algebras are either simultaneously ramified or simultaneously unramified at a place v. This fact leads to the following estimate:
where r is the number of \(v \in V\) where D ramifies (which is necessarily finite for a divisorial set). Thus, we obtain an upper bound on the size of the genus that is uniform over all central division Kalgebras of a given degree with a fixed number of ramification places; in particular, this estimate is uniform over all quaternion division Kalgebras.
Our second proof of Theorem 8.9, which we gave in [23], also uses the analysis of ramification, but avoids imposing restrictions on the characteristic of the field K. The reason for this is that the argument does not require the finiteness of the (ntorsion of the) unramified Brauer group, but only the finiteness of certain subgroups of the latter. On the other hand, since these subgroups depend on the division algebra at hand, we do not obtain a nice estimate on the size of the genus as provided by our first proof.
To finish up this discussion, let us mention that the proof of Theorem 8.7 applies similar considerations to the set V of geometric places of the field \(K = k(x)\) (i.e., the set of those v that correspond to the closed points of \({\mathbb {P}}^1_k\)). Namely, let D be a quaternion division algebra over K and let \(D'\) be another quaternion division Kalgebra such \([D'] \in \mathbf{gen }(D)\). As above, for each \(v \in V\), letting \(\chi _v\) and \(\chi '_v\) denote the characters defined by (10), we have \(\ker \chi _v = \ker \chi '_v\). But since \(n = 2\), this means that actually \(\chi _v = \chi '_v\). It follows that the class \([D] \cdot [D']^{1}\) in \({}_2\mathrm {Br}(K)\) is unramified at all \(v \in V\). But according to a result of Faddeev (cf. [48, Theorem 6.9.1]), we have
implying that \([D'] = [D] \cdot [\Delta \otimes _k K]\) for some quaternion algebra \(\Delta \) over k. Finally, using the assumption that the genus of every quaternion algebra over k is trivial and applying a specialization argument, one concludes that the class \([\Delta ] \in \mathrm {Br}(k)\) is trivial, hence \(D = D'\).
Quaternion algebras and Riemann surfaces
In this subsection, we will briefly describe how quaternion algebras sharing “many” (although a priori not all) quadratic subfields arise in the investigation of Riemann surfaces. These considerations, which provide a geometric context for our general discussion, will be extended in Sect. 9 to locally symmetric spaces of arbitrary simple real algebraic groups.
Let \({\mathbb {H}} = \{ x +iy \in {\mathbb {C}} \, \vert \, y > 0 \}\) be the complex upper halfplane equipped with the standard hyperbolic metric \(ds^2 = y^{2}(dx^2 + dy^2)\). The action of \(\mathrm {SL}_2({\mathbb {R}})\) by fractional linear transformations is transitive and isometric, allowing us to identify \({\mathbb {H}}\) with the homogeneous (symmetric) space \(\mathrm {SL}_2({\mathbb {R}}) / \mathrm {SO}_2({\mathbb {R}})\). Let \(\pi :\mathrm {SL}_2({\mathbb {R}}) \rightarrow \mathrm {PSL}_2({\mathbb {R}})\) be the canonical projection. It is well known (cf., for example, [41, Theorem 27.12]) that any compact Riemann surface of genus \(> 1\) can be presented as a quotient \(\Gamma \backslash {\mathbb {H}}\) by some discrete subgroup \(\Gamma \subset \mathrm {SL}_2({\mathbb {R}})\) containing \(\{ \pm I \}\) and having torsionfree image \(\pi (\Gamma )\). It was demonstrated in [72] that some properties of M can be understood in terms of the associated quaternion algebra \(A_{\Gamma }\), which is constructed as follows.
Let \(\Gamma ^{(2)}\) denote the subgroup of \(\Gamma \) generated by the squares of all elements, and set \(A_{\Gamma }\) to be the \({\mathbb {Q}}\)subalgebra of \(\mathrm {M}_2({\mathbb {R}})\) generated by \(\Gamma ^{(2)}\). One shows that \(A_{\Gamma }\) is a quaternion algebra (not necessarily a division algebra) with center
(the socalled trace field)—cf. [72, Ch. 3]. Furthermore, it turns out that if \(\Gamma _1\) and \(\Gamma _2\) are commensurable (i.e., their intersection has finite index in both of them), then \(A_{\Gamma _1} = A_{\Gamma _2}\); in other words, \(A_{\Gamma }\) is an invariant of the commensurability class of \(\Gamma \). Moreover, if \(\Gamma \) is an arithmetic Fuchsian group, then \(K_{\Gamma }\) is a number field and \(A_{\Gamma }\) is the quaternion algebra involved in the description of \(\Gamma \) (cf. [72, §8.2]). It follows that if \(\Gamma _1\) and \(\Gamma _2\) are arithmetic subgroups and the algebras \(A_{\Gamma _1}\) and \(A_{\Gamma _2}\) are isomorphic, then \(\Gamma _1\) is commensurable with a conjugate of \(\Gamma _2\). Thus, in the arithmetic case, \(A_{\Gamma }\) completely determines the commensurability class of \(\Gamma \) (up to conjugation). This is no longer true for nonarithmetic subgroups, but nevertheless \(A_{\Gamma }\) remains an important invariant of the commensurability class.
Next, in differential geometry, one attaches various spectra to a Riemannian manifold M: particularly when M is compact, one considers the Laplace spectrum E(M), which consists of the eigenvalues of the BeltramiLaplace operator with multiplicities; in the general case, one can also look at the (weak) length spectrum L(M), which is defined as the set of lengths of all closed geodesics in M. Then two Riemannian manifolds \(M_1\) and \(M_2\) are said to be

(1)
isospectral if \(E(M_1) = E(M_2)\) (assuming that \(M_1\) and \(M_2\) are compact);

(2)
isolengthspectral if \(L(M_1) = L(M_2)\);

(3)
lengthcommensurable if \({\mathbb {Q}} \cdot L(M_1) = {\mathbb {Q}} \cdot L(M_2)\).
On the other hand, \(M_1\) and \(M_2\) are called commensurable if they have a common finitesheeted cover, with the covering maps being local isometries. In general, one would like to understand how \(M_1\) and \(M_2\) are related if they satisfy one of the above conditions (1)(3) or similar ones. Probably the most famous version of this general question is due to M. Kac [61], who asked “Can one hear the shape of a drum?” In other words, are two compact isospectral Riemannian manifolds necessarily isometric? For our purposes, however, we will focus on lengthcommensurable Riemann surfaces and their commensurability. We refer the reader to [100] for a more detailed discussion of these conditions, and only mention here that for compact locally symmetric spaces of simple real algebraic groups (in particular, for compact Riemann surfaces), condition (1) implies condition (2), which in turn trivially implies condition (3). What is interesting is that as far as questions of commensurability are concerned, condition (3), despite being the weakest, has essentially the same consequences as the strongest condition (1).
We will now examine how the lengthcommensurability of two compact Riemann surfaces of genus \(> 1\) impacts the associated quaternion algebras. First, let \(M = \Gamma \backslash {\mathbb {H}}\) be a compact Riemann surface as above. It is well known that every closed geodesic in M corresponds to a semisimple element \(\gamma \in \Gamma \), \(\gamma \ne \pm I\), and we will denote it by \(c_{\gamma }\). For our discussion, we will only need the following formula for the length \(\ell (c_{\gamma })\) of \(c_{\gamma }.\) Since \(\Gamma \subset \mathrm {SL}_2({\mathbb {R}})\) is discrete and \(\pi (\Gamma ) \subset \mathrm {PSL}_2({\mathbb {R}})\) is torsionfree, any semisimple element \(\gamma \in \Gamma \) is automatically hyperbolic, hence is conjugate to a matrix of the form \(\left( \begin{array}{cc} t_{\gamma } &{}\quad 0 \\ 0 &{}\quad t_{\gamma }^{1} \end{array} \right) \) with \(t_{\gamma } \in {\mathbb {R}}\). Then
where \(n_{\gamma }\) is an integer (in fact, it is the winding number—we refer the reader to [96, §8] for the details). It follows that
Now, let \(M_1 = \Gamma _1 \backslash {\mathbb {H}}\) and \(M_2 = \Gamma _2 \backslash {\mathbb {H}}\) be two compact Riemann surfaces as above, and assume that they are lengthcommensurable. One then shows that
(this is a consequence of the more general Theorem 8.15 in [96]—see also Theorem 9.4). Furthermore, it follows from (12) that for any semisimple \(\gamma _1 \in \Gamma _1^{(2)}\) different from \(\pm I\), there exists a semisimple \(\gamma _2 \in \Gamma _2^{(2)}\) such that
for some nonzero integers m, n, and consequently \(\gamma _1^m\) and \(\gamma _2^n \in \mathrm {M}_2({\mathbb {R}})\) are conjugate. Then for \(i = 1, 2\), the algebra \(K[\gamma _i]\) is a maximal étale subalgebra of \(A_{\Gamma _i}\), and we have an isomorphism of Kalgebras
Thus, the geometric condition of lengthcommensurability translates into the algebraic condition that \(A_{\Gamma _1}\)and \(A_{\Gamma _2}\)have a common center and the same isomorphism classes of maximal étale subalgebras that have a nontrivial intersection with \(\Gamma _1^{(2)}\)and \(\Gamma _2^{(2)}\), respectively.
We recall from our discussion in Sect. 8.2 that the genus of a quaternion algebra over a number field reduces to a single element. Note, however, that the preceding condition implied by lengthcommensurability is technically weaker than the condition that \(A_{\Gamma _1}\) and \(A_{\Gamma _2}\) belong to the same genus. Nevertheless, it was observed by A. Reid [113] (prior to the systematic investigation of the genus problem) that if \(M_1\) and \(M_2\) are isospectral (hence also isolength spectral) Riemann surfaces with arithmetic fundamental groups \(\Gamma _1\) and \(\Gamma _2\), respectively, then the latter is still sufficient to conclude that \(A_{\Gamma _1} \simeq A_{\Gamma _2}\). As we remarked above, this implies that \(\Gamma _1\) is commensurable to a conjugate of \(\Gamma _2\), and hence the Riemann surfaces \(M_1\) and \(M_2\) are commensurable. In Sect. 9, we will give a brief overview of similar results for arithmetically defined locally symmetric spaces of arbitrary real simple algebraic groups (cf. [96, 100]). We will also see that even for not necessarily arithmetic Riemann surfaces, the above condition leads to the same finiteness results as we have for the genus.
Conjectures and results on the genus problem
Juxtaposing the treatment of the genus of absolutely almost simple simply connected algebraic groups over number fields in Theorem 8.4 with the results in Sect. 8.2 on the genus of division algebras over general fields, one is led to the following.
Conjecture 8.10

(1)
Let \(K = k(x)\) be the field of rational functions in one variable over a number field k. If G is an absolutely almost simple simply connected algebraic Kgroup with center Z(G) of size \(\le 2\), then the genus \(\mathbf{gen }_K(G)\) reduces to a single element.

(2)
Let G be an absolutely almost simple simply connected algebraic group over a finitely generated field of “good” characteristic. Then the genus \(\mathbf{gen }_K(G)\) is finite.
(Here, “good” characteristic is used in the same sense as in Conjecture 5.7.)
The general hope is that this conjecture will be proved through a farreaching extension to algebraic groups of the techniques developed for the study of the genus of division algebras; in this extension, groups with good reduction are expected to play a role similar to that of unramified division algebras. More precisely, as we pointed out in our sketch of the first proof of Theorem 8.9, conceptually, one of the critical observations needed to establish the finiteness of the genus is that if D is a finitedimensional central division Kalgebra that is unramified at a discrete valuation v of K, then every central division Kalgebra \(D'\) such that \([D'] \in \mathbf{gen }(D)\) is also unramified at v. While this fact is certainly nontrivial, at the same time, it is not particularly surprising, and can be proved by exploiting the equivalence of several different characterizations of unramified algebras. Informally speaking, it means that the maximal subfields of a division algebra detect whether or not the algebra is unramified. On the other hand, there were no indications in the literature as to why the maximal tori of a reductive group should be able to detect whether or not the group has good reduction in a sufficiently general situation. So, the following result from [21] and [24] is quite surprising.
Theorem 8.11
Let G be an absolutely almost simple simply connected algebraic group over a field K, and let v be a discrete valuation of K. Assume that the residue field \(K^{(v)}\) is finitely generated and that G has good reduction at v. Then every \(G' \in \mathbf{gen }_K(G)\) also has good reduction at v.
Since the proof relies on techniques that we will not address in this article (specifically, considerations involving socalled generic tori—cf. [99, §9] for an overview of these), we refer the reader to [24] for the details; here, we will only discuss the consequences of this statement for the genus problem. We would also like to draw the reader’s attention to Theorem 9.9 that elaborates on Theorem 8.11.
Suppose now that G is an absolutely almost simple simply connected algebraic group defined over a finitely generated field K, and let V be a divisorial set of places of K. Using the fact that V satisfies condition (A) (see Sect. 5.3), it is easy to show that there exists a finite subset \(S \subset V\) such that G has good reduction at all \(v \in V \setminus S\). Then it follows from Theorem 8.11 that every Kform \(G'\) of G whose Kisomorphism class lies in the genus \(\mathbf{gen }_K(G)\) also has good reduction at all \(v \in V \setminus S\). Since \(V \setminus S\) contains a divisorial set of places of K, we conclude that the truth of Conjecture 5.7 would automatically imply the finiteness of \(\mathbf{gen }_K(G)\) (at least if \(\mathrm {char}\, K\) is “good” for the type of G). We will now list some results on the genus in the spirit of Conjecture 8.10 that have already been established, beginning with inner forms of type \(\textsf {A}_n\), where the conjecture has been proved in full.
Theorem 8.12

(1)
Let D be a central division algebra of exponent two over the field of rational functions \(K = k(x_1, \ldots , x_r)\), where k is either a number field or a finite field of characteristic \(\ne \) 2. Then for \(G = \mathrm {SL}_{m , D}\)\((m \ge 1)\), the genus \(\mathbf{gen }_K(G)\) reduces to a single element.

(2)
Let \(G = \mathrm {SL}_{m , D}\), where D is a central division algebra over a finitely generated field K of degree prime to \(\mathrm {char}\, K\). Then \(\mathbf{gen }_K(G)\) is finite.
The discussion preceding the statement of Theorem 8.12 shows that part (2) follows from Theorem 7.6. On the other hand, we should point out that part (2) is not a direct consequence of Theorem 8.9 on the finiteness of \(\mathbf{gen }(D)\), even for \(m = 1\). The problem is that while every maximal Ktorus of the group \(G = \mathrm {SL}_{1 , D}\), with D a central division Kalgebra, is the norm torus \(\mathrm {R}^{(1)}_{F/K}({\mathbb {G}}_m)\) for some maximal separable subfield F of D, the fact that two such tori \(\mathrm {R}^{(1)}_{F_1/K}({\mathbb {G}}_m)\) and \(\mathrm {R}^{(1)}_{F_2/K}({\mathbb {G}}_m)\) are Kisomorphic does not in general imply that the field extensions \(F_1\) and \(F_2\) are isomorphic over K. Thus, the fact that the algebraic Kgroups \(\mathrm {SL}_{1 , D_1}\) and \(\mathrm {SL}_{1 , D_2}\), for some central division Kalgebras \(D_1\) and \(D_2\), are in the same genus may not imply that the algebras \(D_1\) and \(D_2\) are themselves in the same genus. So, some additional considerations (involving generic tori) are needed to derive the finiteness of \(\mathbf{gen }_K(G)\) from that of \(\mathbf{gen }(D)\) (cf. [18, Theorem 5.3]). Analogous considerations implemented in the context of the proof of the Stability Theorem (Theorem 8.7) yield part (1) of Theorem 8.12. Since there are no restrictions on the characteristic in Theorem 8.9, it would be interesting to determine if these are necessary in Theorem 8.12(2).
Next, we will consider the genus of spinor groups.
Theorem 8.13
Suppose K is either the field of rational functions k(x, y) in two variables or the function field k(C) of a smooth geometrically integral curve C over k, where, in both cases, k is a number field. Let \(G = \mathrm {Spin}_n(q)\) be the spinor group of a nondegenerate quadratic form q over K of odd dimension \(n \ge 5\). Then \(\mathbf{gen }_K(G)\) is finite.
We note that the argument sketched prior to the statement of Theorem 8.11, combined with Theorem 7.7 and [106, Theorem 5.5], shows that for \(G = \mathrm {Spin}_n(q)\), where q is a nondegenerate quadratic form over K of any dimension \(n \ge 5\), the number of Kisomorphism classes of spinor groups \(G' = \mathrm {Spin}_n(q')\) that have the same maximal Ktori as G is finite. When n is odd, all Kforms of G are again spinor groups, and we obtain the above theorem. On the other hand, when n is even, G has Kforms coming from skewhermitian forms over noncommutative central division algebras over K, and so far, we have not been able to eliminate these as potential members of the genus \(\mathbf{gen }_K(G)\) (see, however, [22, Theorem 1.2] for a partial result in this direction). We also know that \(\mathbf{gen }_K(G)\) is finite if K is the same as in Theorem 8.13 and G is either \(\mathrm {SU}_n(L/K , h)\), where h is a nondegenerate hermitian form of dimension \(n \ge 2\) over a quadratic extension L/K, or \(\mathrm {SU}_n(D , h)\), where h is a nondegenerate hermitian form of dimension \(n \ge 1\) over a central quaternion division algebra D over K with the canonical involution (the case where K is a 2dimensional global field is handled in [22, Theorem 8.3 and Remark 8.6]; for the field of rational functions in two variables, one proceeds analogously, making use of [106, Theorem 5.1]).
We conclude this section with the following result for groups of type \(\textsf {G}_2\).
Theorem 8.14
Let G be a simple algebraic Kgroup of type \(\textsf {G}_2\).

(1)
If K is the field of rational functions k(x), where k is a number field, then \(\vert \mathbf{gen }_K(G) \vert = 1\).

(2)
If k is a number field and K is one of the following:

\(K = k(x_1, \ldots , x_r)\) is the field of rational functions in any (finite) number of variables;

\(K = k(C)\) is the function field of a smooth geometrically integral curve C over k;

\(K = k(X)\) is the function field of a Severi–Brauer variety X over k associated with a central division algebra D over k of degree \(\ell \), where \(\ell \) is either odd or \(\ell = 2\),
then \(\mathbf{gen }_K(G)\) is finite.

The proofs of these statements ultimately rely on an analysis of unramified cohomology, and are treated in detail in [22, Theorems 9.1 and 9.3] and [106, Proposition 5.3].
The genus and base change
To conclude our overview of the genus problem, in this subsection, we will briefly discuss how the genus \(\mathbf{gen }_K(G)\) varies under base change. As a first example, we observe that if \(G = \mathrm {SL}_{1,D}\), where D is a cubic division algebra over a number field K, then, using the description of the Brauer group of a number field provided by Albert–Brauer–Hasse–Noether theorem, one can construct a sequence of finite extensions \(F_i/K\) so that the sizes of \(\mathbf{gen }_{F_i}(G \times _K F_i)\) grow unboundedly (of course, this cannot happen if D is a quaternion algebra). On the other hand, the following theorem from [24] shows that the genus cannot grow under purely transcendental extensions.
Theorem 8.15
Let G be an absolutely almost simple simply connected algebraic group over a finitely generated field k of characteristic 0, and let \(K = k(x)\) be the field of rational functions in one variable. Then any \(G' \in \mathbf{gen }_K(G \times _k K)\) is of the form \(G' = G'_0 \times _k K\) for some \(G'_0 \in \mathbf{gen }_k(G)\).
To sketch the idea of the proof, we let V denote the set of geometric places of K (i.e., those valuations that correspond to the closed points of \({\mathbb {P}}_k^1\)). Then the group \(G \times _k K\) has good reduction at all \(v \in V\). So, it follows from Theorem 8.11 that \(G'\) also has good reduction at all \(v \in V\). The desired conclusion is then derived from Theorem 5.3.
In conjunction with Theorem 8.4, this yields the following.
Corollary 8.16
Let G be an absolutely almost simple simply connected algebraic group over a number field k, and let \(K = k(x_1, \ldots , x_r)\) be the field of rational functions in \(r \ge 1\) variables. Then the genus \(\mathbf{gen }_K(G \times _k K)\) is finite, and in fact reduces to a single element if the type of G is different from \(\textsf {A}_n\), \(\textsf {D}_{2n+1}\)\((n > 1)\), and \(\textsf {E}_6\).
We note that for the exceptions \(\textsf {A}_n\), \(\textsf {D}_{2n+1}\)\((n > 1)\), and \(\textsf {E}_6\), we have \(\vert Z(G) \vert > 2\), so this corollary proves Conjecture 8.10 in full for groups of the form \(G \times _k K\) in the above notations.
It is important to point out that for \(G'_0 \in \mathbf{gen }_k(G)\), the group \(G' = G'_0 \times _k K\) may not lie in \(\mathbf{gen }_K(G \times _k K)\). In fact, we have the following two results from [24], which point to a new phenomenon that we refer to as “killing the genus” by a purely transcendental extension of the base field.
Theorem 8.17
Let A be a central simple algebra of degree n over a finitely generated field k, and let \(G = \mathrm {SL}_{1 , A}\). Assume that \(\mathrm {char}\, k\) is prime to n, and let \(K = k(x_1, \ldots , x_{n1})\) be the field of rational functions in \((n1)\) variables. Then \(\mathbf{gen }_K(G \times _k K)\) consists of (the isomorphism classes of) groups of the form \(H \times _k K\), where \(H = \mathrm {SL}_{1 , B}\) and B is a central simple kalgebra of degree n such that its class [B] in the Brauer group \(\mathrm {Br}(k)\) generates the same subgroup as the class [A].
Theorem 8.18
Let G be a group of type \(\textsf {G}_2\) over a finitely generated field k of characteristic \(\ne 2\), and let \(K = k(x_1, \ldots , x_6)\) be the field of rational functions in 6 variables. Then \(\mathbf{gen }_K(G \times _k K)\) reduces to a single element.
We note that the genus of a group of type \(\textsf {G}_2\) can be nontrivial (cf. [8]). The proof of Theorem 8.17 uses Amitsur’s Theorem (Theorem 8.5) on generic splitting fields, while the proof of Theorem 8.18 relies on properties of quadratic Pfister forms. These two results, and also similar ones for the genus of division algebras ([24]), point to the following conjecture.
Conjecture 8.19
Let G be an absolutely almost simple simply connected algebraic group over a finitely generated field k, and assume that \(\mathrm {char}\, k\) is “good” for the type of G. Then there is a purely transcendental extension \(K = k(x_1, \ldots , x_r)\) of transcendence degree r depending only on the type of G such that every \(G' \in \mathbf{gen }_K(G \times _k K)\) is of the form \(G'_0 \times _k K\), where \(G'_0\) has the property that
for any field extension F/k.
This can be reformulated using a different, more functorial, notion of the genus (proposed by A.S. Merkurjev), which is also based on the consideration of maximal tori, but at the same time incorporates infinite extensions like those involved in Amitsur’s theorem. Namely, one defines the motivic genus \(\mathbf{gen }_m(G)\) of an absolutely almost simple simply connected algebraic kgroup G as the set of kisomorphism classes of kforms \(G'\) of G such that
for all field extensions F/k. Then according to Theorem 8.17, the motivic genus of \(G = \mathrm {SL}_{1 , A}\) is always finite and reduces to one element if A has exponent two. Furthermore, by Theorem 8.18, for a kgroup of type \(\textsf {G}_2\), the motivic genus always reduces to one element (at least when \(\mathrm {char}\, k \ne 2\)). Next, it was shown by Izhboldin [57] that for nondegenerate quadratic forms q and \(q'\) of odd dimension over a field k of characteristic \(\ne 2\), the condition
\((\bullet )\)q and \(q'\) have the same Witt index over any field extension F/k
is equivalent to the fact that q and \(q'\) are scalar multiples of each other (this conclusion being false for evendimensional forms). It follows that \(\vert \mathbf{gen }_m(G) \vert = 1\) for the spinor group \(G = \mathrm {Spin}_n(q)\) with n odd. We note that the condition \((\bullet )\) amounts to the fact that the motives of q and \(q'\) in the category of Chow motives are isomorphic (cf. [140, 141, Theorem 4.18], and [63]), which prompted the choice of terminology for this version of the notion of the genus. One can expect the motivic genus to be finite for all absolutely almost simple simply connected groups (at least over fields of “good” characteristic). On the other hand, Conjecture 8.19 asserts that the genus gets reduced to the motivic genus (i.e., becomes as small as possible) after a suitable purely transcendental extension of the base field.
Applications to Zariskidense subgroups
The analysis of Zariskidense (thin) subgroups of semisimple algebraic groups is a very broad and active area (see, for example, [134]). Our goal in this section is to give some indications of how reduction techniques, and particularly Conjecture 5.7, can be applied in this context. More specifically, we will focus on the geometry of locally symmetric spaces, and at the end will demonstrate how this approach leads to a finiteness result for lengthcommensurable Riemann surfaces without any assumptions of arithmeticity—see Theorem 9.11.
We begin by quickly recalling the standard geometric setup. Let G be a simple algebraic group over \({\mathbb {R}}\). We view the group of \({{\mathbb {R}}}\)points \({\mathcal {G}} = G({\mathbb {R}})\) as a Lie group, pick a maximal compact subgroup \({\mathcal {K}}\) of \({\mathcal {G}}\), and consider the associated symmetric space \({\mathfrak {X}} = {\mathcal {G}} / {\mathcal {K}}\). Furthermore, given a discrete torsionfree subgroup \(\Gamma \subset {\mathcal {G}}\), we let \({\mathfrak {X}}_{\Gamma } = \Gamma \backslash {\mathfrak {X}}\) denote the corresponding locally symmetric space. We say that \({\mathfrak {X}}_{\Gamma }\) is arithmetically defined if the subgroup \(\Gamma \) is arithmetic (see [96, §1] for the details). As in the case of Riemann surfaces that we saw in Sect. 8.3, closed geodesics in \({\mathfrak {X}}_{\Gamma }\) correspond to nontrivial semisimple elements of \(\Gamma \). However, the formula relating the length of the closed geodesic \(c_{\gamma }\) attached to a semisimple element \(\gamma \in \Gamma \) to the eigenvalues of \(\gamma \) is substantially more complicated than Eq. (11) for Riemann surfaces, particularly when the rank of \({\mathfrak {X}}\), i.e., the real rank of G, is \(> 1\) (see [96, Proposition 8.5] for the precise statement). Consequently, the fact that two locally symmetric spaces are lengthcommensurable does not translate into a simple condition like (13) on the eigenvalues of semisimple elements. Instead, the characterization of lengthcommensurable locally symmetric spaces requires the following relationship that was introduced in [96].
Definition 9.1
Let F be an infinite field.

(1)
Let \(\gamma _1 \in \mathrm {GL}_{n_1}(F)\) and \(\gamma _2 \in \mathrm {GL}_{n_2}(F)\) be semisimple (diagonalizable) matrices, and let
$$\begin{aligned} \lambda _1, \ldots , \lambda _{n_1} \ \ \text {and} \ \ \mu _1, \ldots , \mu _{n_2} \end{aligned}$$be their eigenvalues (in a fixed algebraic closure \({\overline{F}}\)). We say that \(\gamma _1\) and \(\gamma _2\) are weakly commensurable if there exist integers \(a_1, \ldots , a_{n_1}\) , \(b_1, \ldots , b_{n_2}\) such that
$$\begin{aligned} \lambda _1^{a_1} \cdots \lambda _{n_1}^{a_{n_1}} = \mu _1^{b_1} \cdots \mu _{n_2}^{b_{n_2}}. \end{aligned}$$(WC) 
(2)
Let \(G_1 \subset \mathrm {GL}_{n_1}\) and \(G_2 \subset \mathrm {GL}_{n_2}\) be reductive algebraic Fgroups. We say that Zariskidense subgroups \(\Gamma _1 \subset G_1(F)\) and \(\Gamma _2 \subset G_2(F)\) are weakly commensurable if every semisimple element \(\gamma _1 \in \Gamma _1\) of infinite order is weakly commensurable to some semisimple element \(\gamma _2 \in \Gamma _2\) of infinite order, and vice versa.
One can view this relationship simply as a higherdimensional version of (13) in the case where \(G_1 = G_2 = \mathrm {SL}_2\). But in fact, even irrespective of locally symmetric spaces, it provides a way of matching the eigenvalues of semisimple elements of \(\Gamma _1\) and \(\Gamma _2\) that does not depend on the choice of the matrix realizations of the ambient algebraic groups \(G_1\) and \(G_2\). In the context of locally symmetric spaces though, one proves the following (cf. [96, Corollary 8.14]):
Let \(G_1\)and \(G_2\)be simple real algebraic groups, and let \({\mathfrak {X}}_i\)be the symmetric space of \({\mathcal {G}}_i = G_i({\mathbb {R}})\)for \(i = 1, 2\). If \(\Gamma _i \subset {\mathcal {G}}_i\)is a torsionfree lattice and the locally symmetric spaces \({\mathfrak {X}}_{\Gamma _i} := \Gamma _i \backslash {\mathfrak {X}}_i\), for \(i = 1, 2\)are lengthcommensurable (in particular, compact isospectral or isolengthspectral), then their fundamental groups \(\Gamma _1\) and \(\Gamma _2\)are weakly commensurable.
On the other hand, the locally symmetric spaces \({\mathfrak {X}}_{\Gamma _1}\) and \({\mathfrak {X}}_{\Gamma _2}\) are commensurable if there exists an \({\mathbb {R}}\)isomorphism \(\varphi :G_1 \rightarrow G_2\) such that \(\varphi (\Gamma _1)\) and \(\Gamma _2\) are commensurable in the usual sense (in this case, we say that \(\Gamma _1\) and \(\Gamma _2\) are commensurable up to an isomorphism between \(G_1\) and \(G_2\)). This suggests that to attack the geometric problem of when lengthcommensurable locally symmetric spaces are commensurable, one should try to prove that in the cases of interest, the weak commensurability of Zariskidense subgroups implies their commensurability (in a suitable sense). At first glance, however, the chances of obtaining a sufficiently general result along these lines appear to be rather slim. Indeed, the matrices
are weakly commensurable because
On the other hand, no powers \(A^m\) and \(B^n\) for \(m , n \ne 0\) are conjugate, so the subgroup \(\langle A \rangle \) is not commensurable to any conjugate of \(\langle B \rangle \). It turns out, though, that the situation changes dramatically if, instead of “small subgroups” (such as cyclic ones), one considers “big subgroups” (e.g., Zariskidense subgroups) of simple algebraic groups. In fact, the case of arithmetic subgroups was worked out almost completely [96].^{Footnote 8} Since the results that we will discuss deal with fields of characteristic zero, we will assume for the remainder of this section that the base field has characteristic zero.
Theorem 9.2
(cf. [96, Theorems 4 and 5]) Let \(G_1\) and \(G_2\) be absolutely almost simple algebraic groups over a field F of characteristic zero.

(1)
Assume that \(G_1\) and \(G_2\) are of the same type, which is different from \(\textsf {A}_n\), \(\textsf {D}_{2n+1}\)\((n > 1)\), or \(\textsf {E}_6\). If Zariskidense arithmetic subgroups \(\Gamma _1 \subset G_1(F)\) and \(\Gamma _2 \subset G_2(F)\) are weakly commensurable, then they are commensurable.^{Footnote 9}

(2)
In all cases, the Zariskidense arithmetic subgroups \(\Gamma _2 \subset G_2(F)\) that are weakly commensurable to a given Zariskidense arithmetic subgroup \(\Gamma _1 \subset G_1(F)\) form finitely many commensurability classes.
Let us note that this theorem was established in [96] not only for arithmetic, but also for Sarithmetic subgroups. Furthermore, just as in Theorem 8.4, the types excluded in part (1) of the theorem are honest exceptions: for each of them, one can construct arbitrarily large, but finite, families of weakly commensurable and pairwise noncommensurable arithmetic subgroups (cf. [96, §9]). As we will see in Theorem 9.3, the only situation in which \(G_1\) and \(G_2\) of different types can contain finitely generated Zariskidense weakly commensurable is when one of the groups is of type \(\textsf {B}_n\) and the other of type \(\textsf {C}_n\)\((n \ge 3)\). Weakly commensurable arithmetic subgroups in this case were completely classified in [44].
As we already indicated above, this line of work was initially motivated by questions in differential geometry, and Theorem 9.2 has served as a basis for various geometric applications to isospectral and lengthcommensurable locally symmetric spaces. For example, it implies the following statement:
Let \(M_1\)and \(M_2\)be arithmetically defined hyperbolic manifolds of the same dimension \(d \not \equiv 1(\mathrm {mod}\, 4)\). If they are lengthcommensurable (in particular, if they are compact and isospectral) then they are commensurable.
(To be more precise, it is actually enough to assume that only one manifold is arithmetically defined and the other is of finite volume.) We refer the reader to [96, 98], and [104] for a number of other geometric applications.
The proof of Theorem 9.2 does not deal with arithmetic subgroups directly, but rather uses the following criterion for their conjugacy. Let G be an absolutely almost simple algebraic group over a field F of characteristic zero. Then the commensurability classes of Zariskidense arithmetic subgroups \(\Gamma \) of G(F) can be parametrized by pairs \((K , {\mathscr {G}})\), where K is a number field contained in F and \({\mathscr {G}}\) is an F/Kform of the adjoint group \({\overline{G}}\). More precisely, every arithmetic subgroup is \((K , {\mathscr {G}})\)arithmetic for some pair \((K , {\mathscr {G}})\) as above, which means that there is an Fisomorphism \(\theta :{\mathscr {G}} \times _K F \rightarrow {\overline{G}}\) such that the image \({\overline{\Gamma }}\) of \(\Gamma \) in \({\overline{G}}(F)\) is commensurable with \(\theta ({\mathscr {G}}({\mathcal {O}}_K))\), where \({\mathcal {O}}_K\) is the ring of algebraic integers in K (this description is very similar to the standard description of arithmetic Fuchsian groups). Then, given absolutely almost simple Fgroups \(G_1\) and \(G_2\), Zariskidense arithmetic subgroups \(\Gamma _1 \subset G_1(F)\)and \(\Gamma _2 \subset G_2(F)\)corresponding to the pairs \((K_1 , {\mathscr {G}}_1)\)and \((K_2 , {\mathscr {G}}_2)\)are commensurable (up to an Fisomorphism between \({\overline{G}}_1\) and \({\overline{G}}_2\)) if and only if
(cf. [96, Proposition 2.5]). So, in the proof of Theorem 9.2, one first shows that in both parts, there is an equality \(K_1 = K_2 =: K\); then, using various localglobal considerations, it is shown that \({\mathscr {G}}_1 \simeq {\mathscr {G}}_2\) in part (a) and that there are finitely many Kisomorphism classes of possible groups \({\mathscr {G}}_2\) for a given \({\mathscr {G}}_1\) in part (b).
The important point here is that such characteristics as K and \({\mathscr {G}}\) can be defined and analyzed not only for arithmetic, but in fact for arbitrary Zariskidense subgroups (although then K does not have to be a number field). Although, in general, they do not necessarily determine the commensurability class of a subgroup, they still carry important information. We begin with two results for arbitrary finitely generated Zariskidense subgroups that were established in [96]. To fix notations, let \(G_1\) and \(G_2\) be absolutely almost simple algebraic groups defined over a field F of characteristic zero, and for \(i = 1, 2\), let \(\Gamma _i \subset G_i(F)\) be a finitely generated Zariskidense subgroup.
Theorem 9.3
([96, Theorem 1]) If \(\Gamma _1\) and \(\Gamma _2\) are weakly commensurable, then either \(G_1\) and \(G_2\) have the same type or one of them is of type \(\textsf {B}_n\) and the other of type \(\textsf {C}_n\) for some \(n \ge 3\).
(We note that arithmetic Zariskidense subgroups in the groups of type \(\textsf {B}_n\) and \(\textsf {C}_n\) can be weakly commensurable—cf. [44, 96, Example 6.7].)
Next, given a semisimple Fgroup G and a Zariskidense subgroup \(\Gamma \subset G(F)\), we let \(K_{\Gamma }\) denote the trace field, i.e., the subfield of F generated by the traces \(\mathrm {tr}(\mathrm {Ad}\, \gamma )\) of all elements \(\gamma \in \Gamma \) in the adjoint representation on the corresponding Lie algebra \({\mathfrak {g}} = L(G)\). According to a result of E.B. Vinberg [139], the field \(K = K_{\Gamma }\) is the minimal field of definition of \(\Gamma \). This means that K is the minimal subfield of F with the property that all transformations in \(\mathrm {Ad}\, \Gamma \) can be simultaneously represented by matrices having all entries in K in a suitable basis of \({\mathfrak {g}}\). If such a basis is chosen, then the Zariskiclosure of \(\mathrm {Ad}\, \Gamma \) in \(\mathrm {GL}({\mathfrak {g}})\) is a semisimple algebraic Kgroup \({\mathscr {G}}\). It is an F/Kform of the adjoint group \({\overline{G}}\), and we will call it the algebraic hull of \(\mathrm {Ad}\, \Gamma \). One proves that if \(\Gamma \) is a Zariskidense \((K , {\mathscr {G}})\)arithmetic subgroup of an absolutely almost simple Fgroup G, then the trace field of \(\Gamma \) coincides with K and the algebraic hull with \({\mathscr {G}}\).
Theorem 9.4
(cf. [96, Theorem 2]) If \(\Gamma _1\) and \(\Gamma _2\) are weakly commensurable, then \(K_{\Gamma _1} = K_{\Gamma _2}\).
For the sake of completeness, we mention one further result. Assume that \(\Gamma _1\) and \(\Gamma _2\) as above are weakly commensurable, and let K be their common trace field and \({\mathscr {G}}_i\) be the algebraic hull of \(\mathrm {Ad}\, \Gamma _i\) for \(i = 1, 2\). We set \(L_i\) to be the minimal Galois extension of K over which \({\mathscr {G}}_i\) becomes an inner form of the split group (see Example 4.5 for the relevant definitions).
Proposition 9.5
(cf. [100, Lemma 5.2]) In the above notations, we have \(L_1 = L_2\).
What makes these results interesting is that \(\Gamma _1\) and/or \(\Gamma _2\) may very well be free groups (in fact, by a theorem of Tits [137], the group \(G_i(F)\) always contains a Zariskidense subgroup which is a free group of rank two). In this case, structurally these groups carry no imprint of the ambient simple algebraic group, but nevertheless, according to Theorem 9.3, information about the eigenvalues of their elements is sufficient to recover the type of this group. Furthermore, classical rigidity results due to Mostow and Margulis imply that an isomorphism between higher rank arithmetic group/lattices yields an isomorphism between their fields of definition^{Footnote 10}. However, again, the structural approach does not extend to arbitrary finitely generated Zariskidense subgroups, while information about eigenvalues can be used to identify the fields of definition.^{Footnote 11} So, Theorems 9.3 and 9.4 point to a new form of rigidity, which is potentially applicable to arbitrary finitely generated Zariskidense subgroups and is based on information about the eigenvalues of elements in the subgroups—for these reasons, it was named eigenvalue rigidity in [104]. The critical issue in this analysis that is not addressed by the above results is the precise relationship between the algebraic groups containing \(\Gamma _1\) and \(\Gamma _2\). Let us fix an absolutely almost simple algebraic Fgroup \(G_1\) and a finitely generated Zariskidense subgroup \(\Gamma _1 \subset G_1(F)\), and denote by \(K = K_{\Gamma _1}\) and \({\mathscr {G}}\) its trace field and algebraic hull, respectively. Next, let \(G_2\) be another absolutely almost simple Fgroup with a finitely generated Zariskidense subgroup \(\Gamma _2 \subset G_2(F)\) that is weakly commensurable to \(\Gamma _1\). What can we say about the algebraic hull \({\mathscr {G}}_2\) of \(\Gamma _2\)? By Theorem 9.4, the trace field \(K_{\Gamma _2}\) must coincide with K, so assuming that \(G_2\) is adjoint (which we always may), we conclude that \({\mathscr {G}}_2\) is an F/Kform of \(G_2\). In general, as \(\Gamma _2\) varies, while remaining weakly commensurable to \(\Gamma _1\), the Kisomorphism class of \({\mathscr {G}}_2\) can also vary. However, one expects that this class always remains within a finite set of possibilities.
Conjecture 9.6
(Finiteness Conjecture for Algebraic Hulls of Weakly Commensurable Subgroups) Let \(G_1\) be an absolutely almost simple algebraic Fgroup, and \(\Gamma _1 \subset G_1(F)\) be a finitely generated Zariskidense subgroup with trace field \(K = K_{\Gamma _1}\). Given another absolutely simple adjoint Fgroup \(G_2\), there exists a finite collection \({\mathscr {G}}_2^{(1)}, \ldots , {\mathscr {G}}_2^{(r)}\) of F/Kforms of \(G_2\) such that if \(\Gamma _2 \subset G_2(F)\) is a finitely generated Zariskidense subgroup that is weakly commensurable to \(\Gamma _1\), then it is conjugate to a subgroup of one of the \({\mathscr {G}}_2^{(i)}(K)\)\((\subset G_2(F))\).
Now, instead of fixing an absolutely simple Fgroup \(G_2\), one can consider all possible absolutely simple adjoint Fgroups \(G_2\) such that there exists a finitely generated Zariskidense subgroup \(\Gamma _2 \subset G_2(F)\) that is weakly commensurable to \(\Gamma _1\). First, according to Theorem 9.3, apart from the ambiguity between types \(\textsf {B}_n\) and \(\textsf {C}_n\), \(G_2\) must have the same type as \(G_1\). We may therefore assume that the type of \(G_2\) is fixed. Second, replacing the field F with its algebraic closure \({\overline{F}}\), we can assume that \(G_2\) (or, more precisely, \(G_2 \times _F {\overline{F}}\)) itself is fixed as an \({\overline{F}}\)group. In other words, taking into account the ambiguity between types \({\mathsf {B}}_n\) and \({\mathsf {C}}_n\), there are either one or two possibilities for \(G_2\) as an \({\overline{F}}\)group. Applying Conjecture 9.6, we then conclude that, even without initially fixing \(G_2\) as an Fgroup, there are only finitely many Kisomorphism classes of algebraic hulls \({\mathscr {G}}_2\).
Thus, if proven, Conjecture 9.6, in conjunction with the preceding results, would tell us that if we fix a finitely generated Zariskidense subgroup \(\Gamma _1 \subset G_1(F)\) with trace field \(K = K_{\Gamma _1}\), then the finitely generated Zariskidense subgroups of absolutely simple algebraic groups Fgroups that are weakly commensurable to \(\Gamma _1\) will all have the same trace field K and there will be only finitely many possibilities for their algebraic hulls. Such a statement would completely resolve one of the keys issues in the study of eigenvalue rigidity (we refer the reader to [99, §10] for several other open problems in this area). In more informal terms, Conjecture 9.6 would basically imply that an absolutely simple algebraic group is almost determined (i.e., guaranteed to be in a finite set of possibilities) by the eigenvalues of the elements of a finitely generated Zariskisubgroup, regardless of how small/thin this subgroup may be (e.g., it can be a free group on two generators). For a slightly different perspective, recall that in Conjecture 8.10, we considered the problem of whether an absolutely almost simple algebraic group G over a finitely generated field K of good characteristic is almost determined by the set of Kisomorphism classes of its maximal Ktori. From this point of view, Conjecture 9.6 would imply that G is still almost determined (at least in characteristic zero) by the Kisomorphism classes of those maximal Ktori that intersect nontrivially a given Zariskidense subgroup—cf. [100, §5]. (Note that this is consistent with our discussion at the end of Sect. 8.3). Here is what Conjecture 9.6 yields in some concrete situations.
Example 9.7
Let A be a central simple algebra of degree n over a finitely generated field K of characteristic zero, set \(G = \mathrm {SL}_{1 , A}\) to be the corresponding algebraic Kgroup associated with group of elements of reduced norm 1, and suppose \(\Gamma \subset G(K)\) is a finitely generated Zariskidense subgroup with trace field \(K = K_{\Gamma }\). Now, let \(G'\) be another absolutely almost simple simply connected Kgroup such that there exists a finitely generated Zariskidense subgroup \(\Gamma ' \subset G'(K)\) that is weakly commensurable to \(\Gamma '\). Then we deduce from Theorem 9.3 and Proposition 9.5 that \(G'\) is necessarily of the form \(G' = \mathrm {SL}_{1 , A'}\), where \(A'\) is a central simple Kalgebra of the same degree n (note that Proposition 9.5 implies that \(G'\) is an inner form of G). Furthermore, Conjecture 9.6 in this situation would tell us that there are only finitely many possibilities for \(A'\) up to Kisomorphism.
Example 9.8
Let q be a nondegenerate quadratic form in \(n \ge 5\) variables over a finitely generated field K of characteristic zero, let \(G = \mathrm {SO}_n(q)\) be the corresponding special orthogonal group, and take \(\Gamma \subset G(K)\) to be a finitely generated Zariskidense subgroup with trace field \(K = K_{\Gamma }\). Then Conjecture 9.6 would imply that there exist finitely many similarity classes of nondegenerate ndimensional quadratic forms \(q'\) over K such that for \(G' = \mathrm {SO}_n(q')\), the group \(G'(K)\) contains a finitely generated Zariskidense subgroup \(\Gamma '\) that is weakly commensurable to \(\Gamma \).
The important point in the context of our discussion in this article is that the assertion of Conjecture 9.6 can be derived from Conjecture 5.7. The argument hinges on the following result from [24] that links weak commensurability with good reduction.
Theorem 9.9
Let G be an absolutely almost simple simply connected algebraic group over a finitely generated field K of characteristic zero, and let V be a divisorial set of places of K. Given a finitely generated Zariskidense subgroup \(\Gamma \subset G(K)\) with trace field K, there exists a finite subset \(S(\Gamma ) \subset V\) such that every absolutely almost simple simply connected algebraic Kgroup \(G'\) with the property that there exists a finitely generated Zariskidense subgroup \(\Gamma ' \subset G'(K)\) that is weakly commensurable to \(\Gamma \) has good reduction at all \(v \in V \setminus S(\Gamma )\).
The proof of this theorem is based on a significant elaboration of the ideas involved in the proof of Theorem 8.11. The key point is to show that in order to characterize good reduction, one does not need to consider all maximal tori over the base field, but only those that intersect nontrivially a given finitely generated Zariskidense subgroup and also are generic (cf. [100, §5]).
In the case where the trace field of \(\Gamma \) is a number field (although \(\Gamma \) does not need to be arithmetic), Conjecture 9.6 was proved in [100, §5]. Furthermore, arguing as in Example 9.7 and combining Theorems 7.6 and 9.9, one proves Conjecture 9.6 in the case where the algebraic hull \({\mathscr {G}}_1\) of \(\Gamma _1\) is an inner form of type \(\textsf {A}_n\)\((n \ge 1)\). We leave it to the reader to consider various other cases of Conjecture 9.6 that can be obtained by combining Theorem 9.9 with the results from Sect. 7.2 on Conjecture 5.7. In the case of lattices, the preceding observations lead to the following consequence. Let G be an absolutely almost simply algebraic \({\mathbb {R}}\)group, and let \(\Gamma \) be a lattice in \({\mathcal {G}} = G({\mathbb {R}})\). It follows from [101, 7.67, 7.68] that if G is not isogenous to \(\mathrm {SL}_2\), then the trace field \(K_{\Gamma }\) is necessarily a number field. Thus, we obtain the following.
Theorem 9.10
Conjecture 9.6 is true when \(F = {\mathbb {R}}\) and \(\Gamma _1\) is a lattice (not necessarily arithmetic) in \(G_1({\mathbb {R}})\).
We conclude this section with a finiteness statement for Riemann surfaces that does not require any assumptions of arithmeticity or even the finiteness of the volume. Motivated by the wellknown result that a family of isospectral compact Riemann surfaces consists of finitely many isometry classes [75], one may wonder if a family of lengthcommensurable Riemann surfaces necessarily consists of finitely many commensurability classes. While this question remains open, the following result for the associated quaternion algebras is established in [24] (here we use the notations introduced in Sect. 8.3).
Theorem 9.11
Let \(M_i = {\mathbb {H}}/\Gamma _i\)\((i \in I)\) be a family of lengthcommensurable Riemann surfaces, where the subgroups \(\Gamma _i \subset \mathrm {SL}_2({\mathbb {R}})\) are finitely generated, discrete, and Zariskidense, and have torsionfree images in \(\mathrm {PSL}_2({\mathbb {R}})\). Then all of the \(\Gamma _i\) have the same trace field K, and the corresponding quaternion algebras \(A_{\Gamma _i}\)\((i \in I)\) split into finitely many Kisomorphism classes.
Indeed, it follows from the discussion in Sect. 8.3 that the subgroups \(\Gamma _i\) are pairwise weakly commensurable. Hence, by Theorem 9.4, they all have the same trace field K, which is finitely generated. Let V be a divisorial set of places of K. Then it follows from Theorem 9.9 that there exists a finite subset \(S \subset V\) such that all algebras \(A_{\Gamma _i}\) are unramified at all \(v \in V \setminus S\). Consequently, the finiteness assertion follows from the finiteness of the unramified Brauer group, as discussed in the proof of Theorem 7.6. (In the present case, one can also give a rather elementary argument by observing that all of the algebras \(A_{\Gamma _i}\) share a quadratic extension of K and arguing as in [23, §2]). Curiously, this theorem is probably one of the first applications of reduction techniques and arithmetic geometry to differential geometry.
Afterword
Our primary goal in this article was to introduce Conjecture 5.7 (our Main Conjecture) and discuss its links with several other problems in the investigation of algebraic groups over higherdimensional fields. For the sake of completeness, we would like to conclude with a brief discussion of two other issues that play a critical role in the classical arithmetic theory of algebraic groups, but whose extension to the more general setting is not directly related to the Main Conjecture: strong approximation and rigidity.
Strong approximation
Let K be a field equipped with a set V of discrete valuations satisfying condition (A) (see Sect. 5.3 and also Sect. 6 for notations and terminology pertaining to adelic groups). Given an algebraic Kgroup G, the corresponding adelic group \(G({\mathbb {A}}(K , V))\) is endowed with a natural topology that has sets of the form
where \(T \subset V\) is a finite subset and \(U_v \subset G(K_v)\) is an arbitrary open subset for \(v \in T\), as a basis of open neighborhoods. We say that G has strong approximation with respect to V if the diagonal embedding \(G(K) \hookrightarrow G({\mathbb {A}}(K , V))\) has dense image.^{Footnote 12} This property plays a fundamental role in the study of various questions about algebraic groups over global fields. We refer the reader to [94, Ch. 7] for a detailed account (based on Platonov’s proof [91]) of strong approximation for algebraic groups over number fields, and to the original papers of Margulis [73] and Prasad [95] for groups over arbitrary global fields. The survey [103] also contains a discussion of some variations, generalizations, and applications of strong approximation. In this section, we will briefly consider strong approximation over fields other than global.
Example 10.1
Let R be a Noetherian integral domain that is integrally closed in its field of fractions K, and let V be the set of discrete valuations of K associated with height one prime ideals of R. Take G to be the additive group \({\mathbb {G}}_a\), so that the corresponding adelic group is simply the additive group of the adele ring \({\mathbb {A}}(K , V)\). If G has strong approximation with respect to V, then it easily follows from the definition of the adelic topology that for any height one prime ideals \({\mathfrak {p}}_1 , {\mathfrak {p}}_2\) of R, \({\mathfrak {p}}_1 \ne {\mathfrak {p}}_2\), the diagonal map
must be surjective. However, if the Krull dimension of R is \(> 1\), then typically R contains height one prime ideals \({\mathfrak {p}}_1 \ne {\mathfrak {p}}_2\) such that \({\mathfrak {p}}_1 + {\mathfrak {p}}_2 \ne R\), in which case the above map is not surjective, and hence G fails to have strong approximation with respect to V. On the other hand, if the Krull dimension of R is 1, i.e., R is a Dedekind domain, then any two distinct prime ideals \({\mathfrak {p}}_1 , {\mathfrak {p}}_2\) of R satisfy \({\mathfrak {p}}_1 + {\mathfrak {p}}_2 = R\). It follows from the Chinese Remainder Theorem that R is dense in
easily implying that G does have strong approximation with respect to V.
There are two takeaways from this example. First, we should confine the consideration of strong approximation to the situation where K is the fraction field of a Dedekind domain R and V is the set of discrete valuations of K associated with the nonzero prime ideals of R. The most interesting case is where \(R = k[C]\) is the coordinate ring of a smooth geometrically integral affine curve C over a field k and V is the corresponding set of geometric places of the function field \(K = k(C)\). (Of course, if k is a finite field, then the function field K is global, in which case strong approximation is completely understood. However, the problem for other fields remains wide open—see below.)
Second, the Ksubgroups of G isomorphic to \({\mathbb {G}}_a\) may be helpful for proving strong approximation in G. More precisely, let us assume that G is an absolutely almost simple simply connected Kgroup. We recall that G is called Kisotropic if it contains a nontrivial Ksplit torus (in which case it also contains an abundance of 1parameter unipotent subgroups), and Kanisotropic otherwise. Assuming that G is Kisotropic, one can consider the (normal) subgroup \(G(K)^+ \subset G(K)\) generated by the Krational elements of the unipotent radicals of Kdefined parabolic subgroups. In [92], V.P. Platonov constructed the first examples where \(G(K)^+ \ne G(K)\), thereby disproving the KneserTits conjecture (see [47] for a relatively recent survey of the KneserTits problem). At the same time, the quotient
known as the Whitehead group of G over K, is conjectured to be abelian (in which case it will automatically be of finite exponent depending only on the type of G), which has been established in many cases. (In fact, W(G, K) is expected to be finite abelian group whenever K is finitely generated, but only partial results are available so far—cf. [26]). Motivated by these properties, we will say that a normal subgroup N of an abstract group H is “big” if the quotient H/N is an abelian group of finite exponent. Now, if K is the fraction field of a Dedekind domain R and V is the set of discrete valuations of K associated with the nonzero prime ideals of R, then using strong approximation in \({\mathbb {G}}_a\), one easily proves that for any absolutely almost simple simply connected Kisotropic group G, the closure \(\overline{G(K)}\) of G(K) in \(G({\mathbb {A}}(K , V))\) always contains
Thus, for those types where the Whitehead group W(G, F) is known to be abelian for any field extension F/K, the closure \(\overline{G(K)^+}\) is a “big” subgroup of \(G({\mathbb {A}}(K , V))\). We observe that Platonov [93] has found examples of isotropic groups where G(K) is not dense in \(G(K_v)\), so G may not have strong approximation in the general case. It would be interesting to determine if \(\overline{G(K)}\) may be of infinite index in \(G({\mathbb {A}}(K , V))\) when \(K = k(C)\) is the function field of a smooth geometrically integral affine curve C over a finitely generated field k and V is the set of geometric places of K.
Example 10.2
Let \(K = {\mathbb {C}}(t)\), take V to be the set of discrete valuations of K corresponding to all linear polynomials \(t  a\), \(a \in {\mathbb {C}}\), and let G be an absolutely almost simple simply connected Kgroup. It follows from Tsen’s theorem that K has cohomological dimension \(\le 1\), which implies that G is quasisplit (cf. [118, Ch. III, §2]). Then the Whitehead group W(G, F) is trivial for any field extension F/K (cf. [138]). So, the above discussion shows that G has strong approximation with respect to V.
Of course, techniques using 1parameter unipotent subgroups are inapplicable if G is Kanisotropic. So, we would like to propose the following.
Problem 10.3
Let K be the field of fractions of a Dedekind domain R, and let V be the set of discrete valuations of K corresponding to the nonzero prime ideals of R. Investigate the problem of strong approximation for an absolutely almost simple simply connected Kanisotropic group G. When can one guarantee that the closure \(\overline{G(K)}\) of G(K) in \(G({\mathbb {A}}(K , V))\) is a “big” subgroup of the latter? More specifically, for the coordinate ring \(R = k[C]\) of a smooth geometrically integral affine curve C over a finitely generated field k and V the corresponding set of geometric places of \(K = k(C)\), in what cases is \(\overline{G(K)}\) of finite index in \(G({\mathbb {A}}(K , V))\)? Can one always ensure that it is of finite index by deleting from V a finite set of places?
To the best of our knowledge, in the anisotropic case over fields other than global, strong approximation has been established only for the groups \(\mathrm {SL}_{1 , D}\), where D is a quaternion division algebra over \({\mathbb {R}}(t)\) (see [146]). It is likely that this result can be extended to function fields \({\mathbb {R}}(C)\) of arbitrary smooth geometrically integral affine real curves C. Furthermore, since every semisimple algebraic group over \({\mathbb {R}}(C)\) becomes quasisplit over its quadratic extension \({\mathbb {C}}(C)\) (cf. Example 10.2), using strong approximation for the quaternionic group, in conjunction with the techniques of dealing with semisimple groups that split over a quadratic extension of the base field (cf. [145]), one may be able to establish strong approximation for many (and maybe even all) absolutely almost simple simply connected groups over \({\mathbb {R}}(C)\). Because of the paucity of research done in this area so far, it is difficult to predict what methods may be useful, but one should probably reexamine Kneser’s approach to strong approximation ([65, 66]), which relies primarily on considerations from Galois cohomology.
Rigidity
The analysis of representations, and more generally actions, of arithmetic groups and lattices has been one of the central subjects in the theory of arithmetic groups and discrete subgroups of Lie groups in the past 60 years. So, to complete our account of the trends in the arithmetic theory of algebraic groups over higherdimensional fields, we would like to discuss briefly one result on representations of higherdimensional analogues of arithmetic groups. The reader may want to consult [110] for more information. This subject goes back to the classical paper of Bass, Milnor, and Serre [7], where, as a consequence of the solution of the Congruence Subgroup Problem, it was shown that for \(n \ge 3\), every finitedimensional complex representation
is almost algebraic, i.e., there exists a morphism of algebraic groups \(\sigma :\mathrm {SL}_n({\mathbb {C}}) \rightarrow \mathrm {GL}_k({{\mathbb {C}}})\) such that for a suitable finiteindex subgroup \(\Delta \subset \mathrm {SL}_n({\mathbb {Z}})\), the restrictions \(\rho \vert \Delta \) and \(\sigma \vert \Delta \) coincide (cf. [7, Theorem 16.2]). Serre [120] proved a similar result for the group \(\mathrm{SL}_2({{\mathbb {Z}}}[1/p])\). Subsequently, very general results about representations of higherrank arithmetic groups were obtained by Margulis in his Superrigidity Theorem (cf. [74, Ch. VII]). At the same time, Steinberg [129] showed that the above results for representations of \(\mathrm{SL}_n ({{\mathbb {Z}}})\) (\(n \ge 3\)) can be derived directly from the commutator relations for elementary matrices.
Taking inspiration from Steinberg’s generatorsrelations approach, the second author introduced in [107] a novel method for analyzing abstract representations of elementary subgroups of higherrank Chevalley groups over arbitrary commutative rings. To fix notations, let \(\Phi \) be a reduced irreducible root system of rank \(\ge 2\), and let G be the simplyconnected Chevalley group scheme over \({{\mathbb {Z}}}\) of type \(\Phi \). Given a commutative ring R, we denote by E(R) the subgroup of G(R) generated by the Rpoints of the canonical 1parameter root subgroups \(e_{\alpha } :{\mathbb {G}}_a \rightarrow G\) for all \(\alpha \in \Phi \) (this group is usually called the elementary subgroup of G(R)). In [107], the second author studied in detail the finitedimensional representations
where K is an algebraically closed field of characteristic 0. He showed that if \((\Phi , R)\) is a nice pair,^{Footnote 13} then such a representation often has a standard description, i.e., there exists a commutative finitedimensional Kalgebra B, a ring homomorphism \(f :R \rightarrow B\) with Zariskidense image, and a morphism \(\sigma :G(B) \rightarrow \mathrm{GL}_n(K)\) of algebraic Kgroups such that for a suitable finiteindex subgroup \(\Gamma \subset E(R)\) of, we have
where \(F :E(R) \rightarrow E(B)\) is the group homomorphism induced by f (see [107, Main Theorem] for the precise statement). A key step in the proof of this theorem is the construction of an algebraic ring that is naturally associated to the representation \(\rho \) and which captures information about the images of all root subgroups of E(R) (see [107, Theorem 3.1]). This result, on the one hand, confirmed in the case of split groups a longstanding conjecture of Borel and Tits [13] on the structure of abstract homomorphisms of algebraic groups, and, on the other hand, subsumed most previous rigidity statements for Chevalley groups over commutative rings. Subsequently, an analogous statement was also obtained for representations of the groups G(k), where \(G = \mathrm{SL}_{n, D}\) with \(n \ge 3\) and D is a finitedimensional central division algebra over a field k of characteristic 0 (cf. [108]). Moreover, in [108] and [109], these results were applied to the analysis of character varieties of certain finitely generated groups. Additionally, by studying the structure of algebraic rings in positive characteristic, D. Boyarchenko and the second author [14] established, in many situations, the existence of standard descriptions for representations of elementary subgroups of Chevalley groups over fields of characteristic p.
Now, while the rigidity properties of arithmetic groups and lattices are well understood, over the last 20 years, there has been a great deal of interest in the representations and related properties (such as Kazhdan’s property (T)) of groups over higherdimensional rings, particularly the groups \(\mathrm{SL}_n({{\mathbb {Z}}}[x_1, \dots , x_k])\) with \(n \ge 3\) (these are sometimes called universal lattices). In the context of the analysis abstract homomorphisms, we should mention that, using a variation of the method of Bass, Milnor, and Serre, Shenfeld [123] showed that any completely reducible finitedimensional complex representation of \(\mathrm{SL}_n({{\mathbb {Z}}}[x_1, \dots , x_k])\) (\(n \ge 3)\) has a standard description, thereby answering a question of Kazhdan. However, until recently, there were no rigidity statements available for arbitrary finitedimensional representations of universal lattices. In [111], using the general framework developed in [107], in conjunction with a careful analysis of certain central extensions, the second author obtained the following rigidity result.
Theorem 10.4
(cf. [111, Corollary 1.3]) Let \(\Phi \) be a reduced irreducible root system of rank \(\ge 2\), G be the corresponding simplyconnected Chevalley group scheme over \({{\mathbb {Z}}}\), and be K an algebraically closed field of characteristic 0. If \({\mathcal {O}}\) is the ring of Sintegers in a number field such that \((\Phi , {\mathcal {O}})\) is a nice pair, then any representation
has a standard description.
In fact, the results of [111] deal, more generally, with representations of the groups E(R), where R is a ring with “few” derivations, which, in particular, explains the classical rigidity results for Chevalley groups of rank \(\ge 2\) over number rings. Now, by a wellknown result of Suslin [130], if \(\Phi \) is of type \({\mathsf {A}}\), then \(E({\mathcal {O}}[x]) = \mathrm{SL}_n({\mathcal {O}}[x])\), so Theorem 10.4 shows, in particular, that any finitedimensional representation of the universal lattice \(\mathrm{SL}_n({{\mathbb {Z}}}[x])\) has a standard description. We should note that Suslin’s result has recently been extended by Stavrova [128] to all simplyconnected Chevalley groups of rank \(\ge 2\), thereby yielding the existence of standard descriptions for arbitrary representations of \(G({\mathcal {O}}[x]).\)
Notes
 1.
More formally, let \(v_p\) denote the (normalized) padic valuation on \({{\mathbb {Q}}}\) and let \({{\mathbb {Z}}}_{(p)}\) be the corresponding valuation ring. The elliptic curve E is said to have good reduction at p if there exists an abelian scheme \(E_{(p)}\) over the valuation ring \({{\mathbb {Z}}}_{(p)}\) with generic fiber E (the scheme \(E_{(p)}\) is then unique, which leads to a welldefined notion of reduction modulo p).
 2.
We recall that one defines the unipotent radical of a connected algebraic group G to be the largest connected unipotent normal subgroup, and one says that G is reductive if its unipotent radical is trivial. For example, all tori (i.e., connected diagonalizable algebraic groups) are reductive. An algebraic group is (absolutely almost) simple if it does not contain any proper connected normal subgroups, and semisimple if it admits a surjective morphism from a direct product of simple groups. We refer the reader to [10] and [32] for the details.
 3.
In more technical terms, this system defines a scheme over \({{\mathbb {Z}}}_{(p)}\) with generic fiber G.
 4.
For comparison, we would like to point out the following finiteness theorem for the forms of abelian varieties (cf. [89, §3, Finiteness theorem for forms]): Let X be an abelian variety over a field K, and let F/K be a finite separable extension. Then the set of Kisomorphism classes of abelian Kvarieties \(X'\)such that there exists an Fisomorphism \(X \times _K F \simeq X' \times _K F\)is finite. On the contrary, for a semisimple linear algebraic Kgroup G, and a finite separable extension F/K, the set of Kisomorphism classes of F/Kforms \(G'\) of G is infinite in many cases, even when K is a number field (see, however, the discussion of fields of type (F) in Sect. 5.2). So, the problem of classifying forms of (semisimple) linear algebraic groups with special properties, which is central to the current article, comprises some challenges that do not arise in the context of abelian varieties.
 5.
As defined in [32, §5].
 6.
Using twisting, one shows that the properness of \(\theta _{\mathrm {PSL}_n , V}\) in fact implies the properness of \(\theta _{\mathrm {PSL}_{1 , A} , V}\) for any central simple Kalgebra A of degree n.
 7.
We observe that if the genus \(\mathbf{gen }(D)\) is infinite for a central division Kalgebra D, then the genus \(\mathbf{gen }_K(G)\) is also infinite for the corresponding algebraic group \(G = \mathrm {SL}_{1 , D}\).
 8.
We will not go into the details of this analysis here and would only like to point out that one of the important factors is the existence of socalled generic elements in every Zariskidense subgroup—see [99] for a detailed discussion. The reader interested in the technical ingredients can also review the Isogeny Theorem (Theorem 4.2) in [96], which provides a farreaching generalization of the following fact used in section Sect. 8.3: if \(\gamma _1 , \gamma _2 \in \mathrm {SL}_2(F)\)are semisimple elements of infinite order that are weakly commensurable, then for any subfield K that contains the traces of \(\gamma _1\)and \(\gamma _2\), the subalgebras \(K[\gamma _1]\)and \(K[\gamma _2]\)are Kisomorphic.
 9.
This means that that there exists an Fisomorphism \(\varphi :{\overline{G}}_1 \rightarrow {\overline{G}}_2\) between the corresponding adjoint groups such that \(\varphi ({\overline{\Gamma }}_1)\) is commensurable with \({\overline{\Gamma }}_2\), where \({\overline{\Gamma }}_i\) denotes the image of \(\Gamma _i\) in \({\overline{G}}_i(F)\).
 10.
See Sect. 10.2 for some rigidity results over rings more general than rings of algebraic Sintegers
 11.
Of course, the traces of elements in the adjoint representation that generate the field of definition can be easily expressed in terms of the eigenvalues, but in our setup, all we can work with are relations like (WC) in Definition 9.1 for \(\gamma _1 \in \Gamma _1\) and \(\gamma _2 \in \Gamma _2\), which do not immediately yield any relation between \(\mathrm {tr}(\mathrm {Ad}\, \gamma _1)\) and \(\mathrm {tr}(\mathrm {Ad}\, \gamma _2)\).
 12.
Let us point out that here we deviate from the standard terminology. Namely, recall that in the classical setting where K is a number field, \(V^K\) is the set of all places of K, and \(S \subset V^K\) is a finite subset with complement \(V = V^K \setminus S\), we say that G has strong approximation with respect to S if the diagonal embedding \(G(K) \hookrightarrow G({\mathbb {A}}(K , V))\) has dense image (cf. [94, Ch. 7]).
 13.
We will say that \((\Phi , R)\) is a nice pair if 2 is a unit in R whenever \(\Phi \) contains a subsystem of type \({\mathsf {B}}_2\), and 2 and 3 are units in R whenever \(\Phi \) is of type \({\mathsf {G}}_2\).
References
 1.
Abrashkin, V.A.: \(p\)divisible groups over \({{{\mathbb{Z}}}}\). Izv. Akad. Nauk SSSR Ser. Mat. 41(5), 987–1007, 1199 (1977)
 2.
Algebraic Number Theory, ed. by J.W.S. Cassels and A. Fröhlich, 2nd edn. London Mathematical Society (2010)
 3.
Amitsur, S.A.: Generic splitting fields of central simple algebras. Ann. Math. 62, 8–43 (1955)
 4.
Atiyah, M.F., MacDonald, I.G.: Introduction to Commutative Algebra. Westview Press, Boulder (1969)
 5.
Bass, H.: \(K\)theory and stable algebra. Publ. Math. IHES 22, 5–60 (1964)
 6.
Bass, H.: Some problems. In: “classical” algebraic \(K\)theory, Algebraic \(K\)theory, II: “Classical” algebraic \(K\)theory and connections with arithmetic (Proceedings of the Conference in Battelle Memorial Institute, Seattle, Washinton, 1972). Lecture Notes in Mathematics, vol. 342, pp. 3–73. Springer, Berlin (1973)
 7.
Bass, H., Milnor, J., Serre, J.P.: Solution of the congruence subgroup problem for \(SL_ n(n \ge 3)\) and \(Sp_{2n} (n \ge 2)\). Publ. Math. Inst. Hautes Etud. Sci. 33, 59–137 (1967)
 8.
Beli, C., Gille, P., Lee, T.Y.: Examples of algebraic groups of type \({{\sf G}}_2\) having the same maximal tori. Proc. Steklov Inst. Math. 292(1), 10–19 (2016)
 9.
Borel, A.: Properties and linear representations of Chevalley groups. In: Seminar on Algebraic Groups and Related Finite Groups. Lecture Notes in Mathematics, vol. 131, pp. 1–55. Springer (1970)
 10.
Borel, A.: Linear Algebraic Groups, vol. 126, 2nd edn. Springer, New York (1997)
 11.
Borel, A.: Some finiteness properties of adele groups over number fields. Inst. Hautes Études Sci. Publ. Math. 16, 5–30 (1963)
 12.
Borel, A., Serre, J.P.: Théorèmes de finitude en cohomologie galoisienne. Comment. Math. Helv. 39, 111–164 (1964)
 13.
Borel, A., Tits, J.: Homomorphismes ”abstraits” de groupes algébriques simples. Ann. Math. 97(3), 499–571 (1973)
 14.
Boyarchenko, M., Rapinchuk, I.A.: On abstract representations of the groups of rational points of algebraic groups in positive characteristic. Arch. Math. (Basel) 107(6), 569–580 (2016)
 15.
Bourbaki, N.: Commutative Algebra, vol. 17. Springer, New York (1989)
 16.
Chernousov, V.I., Gille, P., Pianzola, A.: Torsors over the punctured affine line. Am. J. Math. 134(6), 1541–1583 (2012)
 17.
Chernousov, V.I., Neher, E., Pianzola, A., Yahorau, U.: On conjugacy of Cartan subalgebras in extended affine Lie algebras. Adv. Math. 290, 260–292 (2016)
 18.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: The genus of a division algebra and the unramified Brauer group. Bull. Math. Sci. 3, 211–240 (2013)
 19.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: Division algebras with the same maximal subfields. Russ. Math. Surv. 70(1), 91–122 (2015)
 20.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: On the size of the genus of a division algebra. Proc. Steklov Inst. Math. 292(1), 63–93 (2016)
 21.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: On some finiteness properties of algebraic groups over finitely generated fields. C. R. Acad. Sci. Paris Ser. I 354, 869–873 (2016)
 22.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: Spinor groups with good reduction. Compos. Math. 155, 484–527 (2019)
 23.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: The finiteness of the genus of a finitedimensional division algebra, and generalizations. Israel J. Math. (to appear)
 24.
Chernousov, V.I., Rapinchuk, A.S., Rapinchuk, I.A.: Simple algebraic groups with the same maximal tori, weakly commensurable Zariskidense subgroups, and good reduction (in preparation)
 25.
ColliotThélène, J.L.: Birational invariants, purity and the Gersten conjecture, in \(K\)theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992). In: Proceedings of Symposia in Pure Mathematics, vol. 58, Part 1, pp. 1–64. American Mathematical Society, Providence (1995)
 26.
ColliotThélène, J.L.: Quelques résultats de finitude pour le groupe \(SK_1\) d’une algèbre de biquaternions. KTheory 10(1), 31–48 (1996)
 27.
ColliotThélène, J.L., Parimala, R., Suresh, V.: Patching and localglobal principles for homogeneous spaces over function fields of padic curves. Comment. Math. Helv. 87(4), 1011–1033 (2012)
 28.
ColliotThélène, J.L., Sansuc, J.J.: Fibrés quadratiques et composantes connexes réelles. Math. Ann. 244(2), 105–134 (1979)
 29.
Conrad, B.: Finiteness theorems for algebraic groups over function fields. Compos. Math. 148(2), 555–639 (2012)
 30.
Conrad, B.: Nonsplit reductive groups over \({{{\mathbb{Z}}}}\), Autour des schémas en groupes II. In: Panor. Synthèses, vol. 46, pp. 193–253. Societe Mathematique de France, Paris (2015)
 31.
Conrad, B.: Reductive group schemes, Autour des schémas en groupes, École d’été “Schémas en groupes,” vol. I (Luminy 2011). Societe Mathematique de France, Paris (2014)
 32.
Conrad, B., Gabber, O., Prasad, G.: PseudoReductive Groups. New Mathematical Monographs, 2nd edn. Cambridge University Press, Cambridge (2015)
 33.
Cutkosky, S.D.: Introduction to Algebraic Geometry, GSM, vol. 188. AMS, New York (2018)
 34.
Darmon, H.: Rational points on curves. Arithmet. Geomet. Clay Math. Proc. 8, 7–53 (2009)
 35.
Demazure, M., Michel, G.P.: Algébriques, Groupes: Tome I: Géométrie algébrique, généralités, groupes commutatifs. NorthHolland Publishing Co., Amsterdam (1970)
 36.
Demazure, M., Grothendieck, A.: Schémas en groupes, Séminaire de géométrie algébrique du Bois Marie 1962/64 (SGA 3). Lecture Notes in Mathematics, vol. 151153. Springer, New York (1970)
 37.
Elman, R., Karpenko, N., Merkurjev, A.: The Algebraic and Geometric Theory of Quadratic Forms, vol. 56. Amer. Math. Soc. Colloq. Publ., Providence (2008)
 38.
Faltings, G.: Endlichkeitssätze für abelsche Varietäten über Zahlkörpern. Invent. Math. 73(3), 349–366 (1983)
 39.
Farb, B., Dennis, R.K.: Noncommutative Algebra, GTM, vol. 144. Springer, New York (1993)
 40.
Fontaine, J.M.: Il n’y a pas de variété abélienne sur \({{{\mathbb{Z}}}}\). Invent. Math. 81(3), 515–538 (1985)
 41.
Forster, O.: Lectures on Riemann surfaces, GTM, vol. 81. Springer, New York (1981)
 42.
Garibaldi, S.: Outer automorphisms of algebraic groups and determining groups by their maximal tori. Michigan Math. J. 61(2), 227–237 (2012)
 43.
Garibaldi, S., Saltman, D.: Quaternion Algebras with the Same Subfields, Quadratic Forms, Linear Algebraic Groups, and Cohomology. Developmental Mathematics, vol. 225238, p. 18. Springer, New York (2010)
 44.
Garibaldi, S., Rapinchuk, A.S.: Weakly commensurable \(S\)arithmetic subgroups in almost simple algebraic groups of types \({{\sf B}}\) and \({{\sf C}}\). Algebra Number Theory 7(5), 1147–1178 (2013)
 45.
Gille, P.: Torseurs sur la droite affine. Transform. Groups 7(3), 231–245 (2002)
 46.
Gille, P., Pianzola, A.: Isotriviality and étale cohomology of Laurent polynomial rings. J. Pure Appl. Algebra 212(4), 780–800 (2008)
 47.
Gille, P.: Le problème de KneserTits, Séminaire Bourbaki. Vol. 2007/2008, Astérisque No. 326 (2009), Exp. No. 983, vii, pp. 39–81 (2010)
 48.
Gille, P., Szamuely, T.: Central Simple Algebras and Galois Cohomology. Cambridge University Press, Cambridge (2006)
 49.
Gross, B.H.: Groups over \({{{\mathbb{Z}}}}\). Invent. math. 124(1–3), 263–279 (1996)
 50.
Guo, N.: The GrothendieckSerre Conjecture over Semilocal Dedekind Rings. arXiv:1902.02315
 51.
Harada, S., Hiranouchi, T.: Smallness of fundamental groups for arithmetic schemes. J. Number Theory 129(11), 2702–2712 (2009)
 52.
Harbater, D., Hartmann, J., Krashen, D.: Applications of patching to quadratic forms and central simple algebras. Invent. Math. 178(2), 231–263 (2009)
 53.
Harbater, D., Hartmann, J., Krashen, D.: Localglobal principles for torsors over arithmetic curves. Am. J. Math. 137(6), 1559–1612 (2015)
 54.
Harder, G.: Halbeinfache Gruppenschemata über Dedekindringen. Invent. math. 4, 165–191 (1967)
 55.
Harder, G.: Über die Galoiskohomologie halbeinfacher algebraischer Gruppen. III. J. Reine Angew. Math. 274(5), 125–138 (1975)
 56.
Hartshorne, R.: Algebraic Geometry, GTM, vol. 52. Springer, New York (1977)
 57.
Izhboldin, O.T.: Motivic equivalence of quadratic forms. Doc. Math. 3, 341–351 (1998)
 58.
Jannsen, U.: Principe de Hasse cohomologique. Séminaire de Théorie des Nombres, Paris 1989–90, pp. 121–140
 59.
Jannsen, U.: Hasse principles for higherdimensional fields. Ann. Math. 183(1), 1–71 (2016)
 60.
Javanpeykar, A., Loughran, D.: Good reduction of algebraic groups and flag varieties. Arch. Math. 104(2), 133–143 (2015)
 61.
Kac, M.: Can One Hear the Shape of a Drum?, vol. 4, 73rd edn, pp. 1–23. American Mathematical Monthly, New York (1966)
 62.
Kahn, B.: Sur le groupe des clsses d’un schéma arithmétiques. Bull. Soc. Math. France 134(3), 395–415 (2006)
 63.
Karpenko, N.A.: Criteria of motivic equivalence for quadratic forms and central simple algebras. Math. Ann. 317(3), 585–611 (2000)
 64.
Kato, K.: A Hasse principle for two dimensional global fields. J. Reine Angew. Math. 366, 142–183 (1986)
 65.
Kneser, M.: Starke approximation in algebraischen Gruppen. I. J. Reine Angew. Math. 218, 190–203 (1965)
 66.
Kneser, M.: Strong approximation. In: Algebraic Groups and Discontinuous Subgroups (Proceedings of Symposia in Pure Mathematics, Boulder, CO, 1965), pp. 187–196, American Mathematical Society, Providence
 67.
Knus, M.A., Merkurjev, A., Rost, M., Tignol, J.P.: The Book of Involutions, vol. 44. American Mathematical Society Colloquium Publications, New York (1998)
 68.
Knus, M.A., Ojanguren, M.: Théorie de la descente et algèbres d’Azumaya. Lecture Notes in Mathematics, vol. 389. Springer, New York (1974)
 69.
Krashen, D., McKinnie, K.: Distinguishing division algebras by finite splitting fields. Manuscripta Math. 134(1–2), 171–182 (2011)
 70.
Lam, T.Y.: Introduction to Quadratic Forms over Fields, GSM 67. AMS, New York (2005)
 71.
Lang, S.: Fundamentals of Diophantine Geometry. Springer, New York (1983)
 72.
Maclachlan, C., Reid, A.: The Arithmetic of Hyperbolic 3Manifolds, GTM, vol. 219. Springer, New York (2003)
 73.
Margulis, G.A.: Cobounded subgroups in algebraic groups over local fields. Funkcional. Anal. i Prilozen. 11(2), 45–57, 95 (1977)
 74.
Margulis, G.A.: Discrete Subgroups of Semisimple Lie Groups. Springer, New York (1991)
 75.
McKean, H.P.: Selberg’s trace formula as applied to a compact Riemann surface. Commun. Pure Appl. Math. 25, 225–246 (1972)
 76.
Meyer, J.S.: A division algebra with infinite genus. Bull. Lond. Math. Soc. 46, 463–468 (2014)
 77.
Milne, J.S.: Class Field Theory. https://www.jmilne.org/math/CourseNotes/cft.html
 78.
Milne, J.S.: Etale Cohomology. Princeton University Press, Princeton (1980)
 79.
Milnor, J.: Introduction to Algebraic \(K\)Theory. Annals of Mathematics Studies, vol. 72. Princeton University Press, Princeton (1971)
 80.
Mordell, L.J.: On the rational solutions of the indeterminate equations of the third and fourth degrees. Proc. Camb. Philos. Soc. 21, 179–192 (1922)
 81.
MoretBailly, L.: Pinceaux de variétés abéliennes. Astérisque, vol. 129 (1985)
 82.
Mumford, D.: Abelian Varieties, vol. 5. Tata Institute of Fundamental Research Studies in Mathematics, Mumbai (1970)
 83.
Neukirch, J.: Algebraic Number Theory. Grundlehren der Mathematischen Wissenschaften, vol. 322. Springer, New York (1999)
 84.
Nisnevich, E.A.: Espaces homogènes principaux rationnellement triviaux et arithmétique des schémas en groupes réductifs sur les anneaux de Dedekind. C. R. Acad. Sci. Paris Ser. I Math. 299(1), 5–8 (1984)
 85.
O’Meara, O.T.: Introduction to Quadratic Forms. Die Grundlehren der mathematischen Wissenschaften, vol. 117. Springer, New York (1963)
 86.
Orlov, D., Vishik, A., Voevodsky, V.: An exact sequence for \(K_*^M/2\) with applications to quadratic forms. Ann. Math. 165(1), 1–13 (2007)
 87.
Parimala, R.: A Hasse principle for quadratic forms over function fields. Bull. Am. Math. Soc. (N.S.) 51(3), 447–461 (2014)
 88.
Parshin, A.N.: Minimal models of curves of genus 2, and homomorphisms of abelian varieties defined over a field of finite characteristic. Izv. Akad. Nauk SSSR Ser. Mat. 36, 67–109 (1972)
 89.
Parshin, A.N., Zarhin, Y.G.: Finiteness problems in Diophantine geometry, English version of the Appendix to the Russian translation of S. Lang’s Fundamentals of Diophantine geometry. arXiv:0912.4325
 90.
Pierce, R.S.: Associative Algebras, GTM, vol. 88. Springer, New York (1982)
 91.
Platonov, V.P.: The problem of strong approximation and the KneserTits hypothesis for algebraic groups. Izv. Akad. Nauk SSSR Ser. Mat. 33, 1211–1219 (1969)
 92.
Platonov, V.P.: On the TannakaArtin problem. Dokl. Akad. Nauk SSSR 221(5), 1038–1041 (1975)
 93.
Platonov, V.P.: Reduced \(K\)theory and approximation in algebraic groups. Trudy Mat. Inst. Steklov. 142, 198–207 (1976)
 94.
Platonov, V.P., Rapinchuk, A.S.: Algebraic Groups and Number Theory. Academic Press, Cambridge (1994)
 95.
Prasad, G.: Strong approximation for semisimple groups over function fields. Ann. Math. 105(3), 553–572 (1977)
 96.
Prasad, G., Rapinchuk, A.S.: Weakly commensurable arithmetic groups and isospectral locally symmetric spaces. Publ. Math. IHES 109, 113–184 (2009)
 97.
Prasad, G., Rapinchuk, A.S.: Localglobal principles for embedding of fields with involution into simple algebras with involution. Comment. Math. Helv. 85(3), 583–645 (2010)
 98.
Prasad, G., Rapinchuk, A.S.: On the fields generated by the lengths of closed geodesics in locally symmetric spaces. Geom. Dedicata. 172, 79–120 (2014)
 99.
Prasad, G., Rapinchuk, A.S.: Generic elements in Zariskidense subgroups and isospectral locally symmetric spaces. In: Thin Groups and Superstrong Approximation, vol. 61, pp. 211–252. Cambridge University Press, Cambridge (2014)
 100.
Prasad, G., Rapinchuk, A.S.: Weakly Commensurable Groups, with Applications to Differential Geometry in Handbook of Group Actions I. Advanced Lectures in Mathematics, vol. 31. International Press, Somerville (2015)
 101.
Raghunathan, M.S.: Discrete Subgroups of Lie Groups Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 68. Springer, New York (1972)
 102.
Raghunathan, M.S., Ramanathan, A.: Principal bundles on the affine line. Proc. Indian Acad. Sci. (Math. Sci.) 93(2–3), 137–145 (1984)
 103.
Rapinchuk, A.S.: Strong Approximation for Algebraic Groups in Thin Groups and Superstrong Approximation, vol. 61, pp. 269–298. Cambridge University Press, Cambridge (2014)
 104.
Rapinchuk, A.S.: Towards the eigenvalue rigidity of Zariskidense subgroups. In: Proceedings of ICM2014 (Seoul), vol. II, pp. 247*269
 105.
Rapinchuk, A.S., Rapinchuk, I.A.: On division algebras having the same maximal subfields. Manuscr. Math. 132, 273–293 (2010)
 106.
Rapinchuk, A.S., Rapinchuk, I.A.: Some finiteness results for algebraic groups and unramified cohomology over higherdimensional fields. arXiv:2002.01520
 107.
Rapinchuk, I.A.: On linear representations of Chevalley groups over commutative rings. Proc. Lond. Math. Soc. 102(5), 951–983 (2011)
 108.
Rapinchuk, I.A.: On abstract representations of the groups of rational points of algebraic groups and their deformations. Algebra Number Theory 7(7), 1685–1723 (2013)
 109.
Rapinchuk, I.A.: On the character varieties of finitely generated groups. Math. Res. Lett. 22(2), 579–604 (2015)
 110.
Rapinchuk, I.A.: Abstract homomorphisms of algebraic groups and applications. In: Handbook of Group Actions II. Advanced Lectures in Mathematics (ALM), vol. 32, pp. 397–447. International Press, Somerville (2015)
 111.
Rapinchuk, I.A.: On abstract homomorphisms of Chevalley groups over the coordinate rings of affine curves. Transform. Gr. 24(4), 1241–1259 (2019)
 112.
Rapinchuk, I.A.: A generalization of Serre’s condition \((\rm F)\) with applications to the finiteness of unramified cohomology. Math. Z. 291(1–2), 199–213 (2019)
 113.
Reid, A.: Isospectrality and commensurability of arithmetic hyperbolic 2 and 3manifolds. Duke Math. J. 65(2), 215–228 (1992)
 114.
Saltman, D.: Lectures on division algebras. In: CBMS Regional Conference Series, vol. 94, AMS (1999)
 115.
Samuel, P.: Anneaux gradués factoriels et modules réflexifs. Bull. Soc. Math. France 92, 237–249 (1964)
 116.
Serre, J.P.: Abelian \(\ell \)adic representations and elliptic curves. McGill University Lecture Notes Written with the Collaboration of Willem Kuyk and John Labute, W. A. Benjamin, Inc. (1968)
 117.
Serre, J.P.: A Course in Arithmetic, GTM, vol. 7. Springer, New York (1973)
 118.
Serre, J.P.: Galois Cohomology. Springer, New York (1997)
 119.
Serre, J.P.: Local Algebra. Springer, New York (2000)
 120.
Serre, J.P.: Le problème des groupes de congruence pour \(SL_2\). Ann. Math. 92, 489–527 (1970)
 121.
Shafarevich, I.R.: Algebraic number fields. In: Proceedings of ICM1962 (Stockholm), pp. 163–176
 122.
Shatz, S.: Profinite Groups, Arithmetic, and Geometry. Annals of Mathematics Studies, vol. 67. Princeton University Press, Princeton (1972)
 123.
Shenfeld, D.: On semisimple representations of universal lattices. Groups Geom. Dyn. 4(1), 179–193 (2010)
 124.
Silverman, J.: The arithmetic of elliptic curves, GTM, vol. 106, 2nd edn. Springer, New York (2009)
 125.
Silverman, J., Tate, J.: Rational Points on Elliptic Curves. Undergraduate Texts in Mathematics, 2nd edn. Springer, New York (2015)
 126.
Springer, T.A.: Linear Algebraic Groups. Progress in Mathematics, vol. 9, 2nd edn. Birkhäuser Boston Inc., Boston (1998)
 127.
Srinivasan, S.: Good Reduction of Unitary Groups of Quaternionic SkewHermitian Forms. arXiv:1906.01414
 128.
Stavrova, A.: Chevalley groups of polynomial rings over Dedekind domains. J. Group Theory 23(1), 121–132 (2020)
 129.
Steinberg, R.: Some consequence of elementary relations in \(SL_n\), finite groups—coming of age. In: Proceedings of the Canadian Mathematical Society Conference held on June 15–28, 1982. Contemporary Mathematics, vol. 45. AMS (1985)
 130.
Suslin, A.A.: On the structure of the special linear group over rings of polynomials. Izv. Akad. Nauk SSSR Ser. Mat. 41, 235–252 (1977) [English transl. Math.USSR Izv. 11, 221–238 (1977)]
 131.
Szamuely, T.: Galois Groups and Fundamental Groups. Cambridge Studies in Advanced Mathematics, vol. 117. Cambridge University Press, Cambridge (2009)
 132.
Szpiro, L.: Sur le théorème de rigidité de Parshin et Arakelov. Journées de Géométrie Algébrique de Rennes (Rennes, 1978), Vol. II, pp. 169–202, Astérisque 64, Soc. Math. France, Paris (1979)
 133.
Szpiro, L.: Propriétés numériques du faisceau dualisant relatif, seminar on pencils of curves of genus at least two. Astérisque 86, 44–78 (1981)
 134.
Thin groups and superstrong approximation, Selected expanded papers from the workshop held in Berkeley, CA, February 6–10, 2012, edited by E. Breuillard and H. Oh, Mathematical Sciences Research Institute Publications, vol. 61. Cambridge University Press, Cambridge (2014)
 135.
Tikhonov, S.V.: Division algebras of prime degree with infinite genus. Proc. Steklov Inst. 292, 256–259 (2016)
 136.
Tits, J.: Classification of algebraic semisimple groups. In: Algebraic Groups and Discontinuous Subgroups, Proceedings of Symposia in Pure Mathematics, Boulder, CO, 1965, pp. 33–62, American Mathematical Society, Providence (1966)
 137.
Tits, J.: Free subgroups in linear groups. J. Algebra 20, 250–270 (1972)
 138.
Tits, J.: Groupes de Whitehead de groupes algébriques simples sur un corps (d’après V. P. Platonov et al.), Séminaire Bourbaki (1976/77), Exp. No. 505, pp. 218–236, Lecture Notes in Mathematics, vol. 677. Springer, New York (1978)
 139.
Vinberg, E.B.: Rings of definition of dense subgroups of semisimple linear groups. Izv. Akad. Nauk SSSR Ser. Mat. 35, 45–55 (1971)
 140.
Vishik, A.: Integral Motives of Quadrics, Preprint MPI199813. Max Planck Institute für Mathematik, Bonn 1998, p. 82. http://www.mpimbonn.mpg.de/node/263
 141.
Vishik, A.: Motives of quadrics with applications to the theory of quadratic forms. Geometric methods in the algebraic theory of quadratic forms. Lecture Notes in Mathematics, vol. 1835, pp. 25–101. Springer, Berlin (2004)
 142.
Voloch, J.F.: Diophantine geometry in characteristic \(p\): a survey. In: Arithmetic Geometry (Cortona, 1994), Symposium in Mathematics, vol. XXXVII, pp. 260–278. Cambridge University Press, Cambridge (1997)
 143.
Voskresenskii, V.E.: Algebraic Groups and Their Birational Invariants. Translations of Mathematical Monographs, vol. 179. American Mathematical Society, Providence (1998)
 144.
Waterhouse, W.: Introduction to Affine Group Schemes, GTM, vol. 66. Springer, New York (1979)
 145.
Weisfeiler, B.: Semisimple algebraic groups that are decomposable over a quadratic extension. Izv. Akad. Nauk SSSR Ser. Mat. 35, 56–71 (1971)
 146.
Yamasaki, A.: Strong approximation theorem for division algebras over \({{{\mathbb{R}}}}(X)\). J. Math. Soc. Jpn. 49(3), 455–467 (1997)
 147.
Zarhin, Y.G.: Torsion of abelian varieties in finite characteristic. Mat. Zametki 22(1), 3–11 (1977)
Acknowledgements
Special thanks are due to Brian Conrad, who carefully read the article and made a number of suggestions that helped to improve the exposition. We are also grateful to Uriya First, Ariyan Javanpeykar, Boris Kunyavskiī, Daniel Loughran, Alexander Merkurjev, Dipendra Prasad, C. Rajan, Zinovy Reichstein, Charlotte Ure, Uzi Vishne, and the anonymous referee for helpful comments.
Author information
Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Rapinchuk, A.S., Rapinchuk, I.A. Linear algebraic groups with good reduction. Res Math Sci 7, 28 (2020). https://doi.org/10.1007/s40687020002263
Received:
Accepted:
Published: