In this article, an analog of a graded version of Quillen-Suslin's local-global principle was established for Steinberg groups St(Φ,−) and Chevalley groups G(Φ,−) over a reduced irreducible root system Φ.
Introduction
In the context of classical K-theory, the local-global principle was introduced by Quillen [10] and Suslin [14] as part of their solution to Serre’s problem on projective modules over polynomial extensions of a field. This principle was initially applied to prove Serre’s conjecture. The solution to Serre’s problem ultimately relied on two key analogs: the K2-analog of Quillen’s local-global principle, which deals with projective modules, and the K1-analog of Suslin’s local-global principle, which deals with elementary matrices.
Recall that for a commutative ring R, denote G(Φ,R) as a simply connected Chevalley group (see [13]) and St(Φ,R) as a Steinberg group (see Definition 2.1) with a reduced irreducible root system Φ. The subgroup of G(Φ,R) generated by the elementary root unipotents eα(r) for all α∈Φ and r∈R is called the elementary subgroup of G(Φ,R) and is denoted by E(Φ,R). There is a well-defined map ϕ:St(Φ,R)→G(Φ,R) sending xα(r)↦eα(r), and the image of ϕ is E(Φ,R), which is a normal subgroup of G(Φ,R) when rank(Φ)≥2. The groups K1(Φ,R) and K2(Φ,R) are defined as the cokernel and kernel of ϕ, respectively
1⟶K2(Φ,R)⟶St(Φ,R)ϕG(Φ,R)⟶K1(Φ,R)⟶1.
The K1-analog of the local-global principle was established for Chevalley groups of type An by Suslin [14]. Subsequent work extended it to type Dn by Suslin–Kopeiko [15], to type Cn by Grunewald–Mennicke–Vaserstein [4], and to all Chevalley groups by Abe [1] and Wendt [18]. Turning to the K2-analog of the local-global principle, it was proven for Chevalley groups of type An by Tulenbaev [16]. Recently, it was extended to type En by Sinchuk [11], to type Dn by Lavrenov-Sinchuk [9], and to type Cn by Lavrenov [8]. The present article aims to revisit these results in the setting of commutative N-graded rings.
The graded version of the K1-analog of the local-global principle has been studied for classical groups in [3, 5]. Meanwhile, for Chevalley groups and Steinberg groups, the graded versions of both the K1-analog and K2-analog are expected to hold. We therefore establish the graded versions of the local-global principle for Steinberg groups and Chevalley groups. These results constitute the main theorems of the present article and are stated below.
Theorem 1.1 (Theorem 3.4). Let R=⨁i≥0Ri be a graded ring and Φn be a either root system of type simply-laced or Cn of rank ≥3. Suppose θ∈St(Φn,R,R+). Then
Key words and phrases. Steinberg group, Chevalley group, Local-global principle, Graded ring.
θ=1 if and only if λm∗(θ)=1∈St(Φn,Rm) for all maximal ideals m of R0.(1)
θ∈Im(lR) if and only if λm∗(θ)∈Im(lRm) for all maximal ideals m of R0.(2)
Theorem 1.2 (Theorem 4.4). Let R=⨁i≥0Ri be a graded ring and Φ be a root system of rank ≥2. Assume that R has no residue fields of 2 elements if Φ=C2 or G2. Suppose θ∈G(Φ,R,R+). Then θ∈E(Φ,R) if and only if λm∗(θ)∈E(Φ,Rm) for all maximal ideals m of R0.
The centrality of K2 (see [9]) follows from the K2-analog of Quillen’s local-global principle, just as the normality theorem for the elementary subgroup is derived from the K1-analog of Suslin’s local-global principle. In other words, the local-global principle plays a foundational role in establishing both the normality of E(Φ,R) and the centrality of St(Φ,R). To extend these results from the polynomial ring to the graded ring, one must employ a combination of three key techniques: the methodology of Quillen-Suslin’s argument, the dilation principle, and Swan–Weibel’s homotopy technique, which allows one to connect polynomial rings to graded rings via a homotopy map.
Preliminaries and Notations
Throughout the article, all rings are assumed to be commutative with unity, and all root systems are reduced and irreducible of rank ≥2 unless otherwise stated.
Definition 2.1. Let A be a ring and Φ be a root system. The Steinberg groupSt(Φ,A) is the group generated by symbols xα(u), where α∈Φ and u∈A, subject to the following relations:
where [p,q]=pqp−1q−1, the product is taken over all positive integers i,j such that iα+jβ∈Φ, and Ci,j,α,β∈Z are the Chevalley structure constants (depending only on Φ). More information on Chevalley groups, the structure constants, and various types of root systems Φ can be found in [13, 17].
Definition 2.2. A root system Φ is called simply-laced if the Dynkin diagram of Φ does not contain double bonds (i.e., Φ has type An, Dn and E6,7,8).
For an ideal I of A, the subgroup St(Φ,I) is spanned by symbols xα(s), where α∈Φ and s∈I. An ideal I is called a splitting ideal if the canonical projection ring homomorphism π:A→A/I splits, i.e., there exists σ:A/I→A such that πσ=1. It is noted that TA[T] is a splitting ideal of A[T].
Definition 2.3. For a splitting ideal I, the relative Steinberg groupSt(Φ,A,I) is defined as the kernel Ker(π∗:St(Φ,A)→St(Φ,A/I)). Obviously, St(Φ,A,I) coincides with the normal closure of St(Φ,I) in St(Φ,A).
In the case of an arbitrary ideal I, the relative Steinberg group St(Φ,A,I) can be defined as a central extension of Ker(π∗). For the purposes of the present text, we concentrate on the relative Steinberg groups corresponding to splitting ideals. For more information regarding the general case, we refer the reader to [11].
Definition 2.4. (Universal Property) Let F(X) denote the free group on the generating set X. If G is any group with a generating set X and certain defining relations ui=vi, there is an isomorphism F(X)/D≅G taking xˉ↦x for each x∈X, where D is a normal subgroup of F(X) generated by the conjugates of the relators uivi−1. Then F(X) has the following universal property: If H is any group and each function f:X→H extends uniquely to a group homomorphism φ:F(X)→H that makes the diagram commute.
X↓⏐F(X)↓⏐F(X)/DfφH⏐↑f∗H
Now, to define a group homomorphism f∗:G→H, it suffices to choose a function f:X→H for which, when x is replaced by f(x) and x−1 by f(x)−1, the defining relations ui=vi for G become true equations in H. Indeed, the homomorphism φ extending f takes D to 1H; thus, it induces a homomorphism f∗ taking xˉ to f(x) for x∈X. Thus, the composite map f∗:F(X)/D→H is a homomorphism taking x∈X to f(x)∈H. For more details on universal property and classical K-theory, we refer the reader to [6].
Lemma 2.5. If f:A→B is a ring homomorphism, then f induces a group homomorphism f∗:St(Φ,A)→St(Φ,B).
Proof. Since any element of St(Φ,A) can be written as a finite product of elements of the form xα(u), α∈Φ, u∈A, it is enough to define a map on generators. We define a map f∗(xα(u))=xα(f(u)) where α∈Φ. Due to the universal property 2.4, we need to verify that this map preserves the Steinberg relations, i.e., replacing u by f(u) and v by f(v) in the defining relations (St1) and (St2) for St(Φ,A) yields equations in St(Φ,B). However, this is indeed true because f is a ring homomorphism. ■
One of the key components for handling the graded situation is the well-known “Swan–Weibel’s homotopy trick”. Let R=R0⊕R1⊕R2⊕⋯ be a graded ring and R+=R1⊕R2⊕⋯ denote the irrelevant ideal of R. An element r∈R will be denoted by r=r0+r1+r2+⋯ where ri∈Ri for each i, and all but finitely many ri are zero.
Definition 2.6. For a graded ring R=⨁i≥0Ri, the Swan–Weibel homotopy maph:R→R[T] is defined as
h(r0+r1+⋯+rm)=r0+r1T+⋯+rmTm,(ri∈Ri).
For any R0-algebra A and any a∈A, consider an evaluation map eva:R[T]→A, which is the R0-algebra homomorphism sending T↦a.
We denote the image r in A by rh(a):=(eva∘h)(r). If A=R, then it is easy to observe that ev0∘h:R→R is precisely the projection onto the degree-zero part and ev1∘h:R→R is precisely the identity map. For a fixed element c∈R0, similarly denote rh(c):=h(r)(c)=(evc∘h)(r), and note that rh(0)=r0 and rh(1)=r. In order to emphasize the choice of A, we denote by ev[T↦aR[T]→A] for R0-algebra homomorphism eva.
Remark 2.7. In view of Lemma 2.5, the map h:R→R[T] (resp. eva:R[T]→A) induces a group homomorphism h∗:St(Φ,R)→St(Φ,R[T]) (resp. eva∗:St(Φ,R[T])→St(Φ,A)) that sends the generators xα(r) to xα(h(r)) (resp. xα(p(T)) to xα(p(a))) where α∈Φ. For θ∈St(Φ,R) and c∈R0, analogously, θh∗(c):=(evc∗∘h∗)(θ) is defined. For more details on Swan–Weibel’s homotopy trick, we refer the reader to [7 Chapter V].
Notations: For a ring A, we denote by λa:A→Aa and λm:A→Am the homomorphisms of localization at an element a∈A and at a maximal ideal m of A respectively. If f:A→B is a ring homomorphism, its corresponding Steinberg group homomorphism is denoted by f∗:St(Φ,A)→St(Φ,B). If Φn is an irreducible root system of rank n, we denote by ιA:St(Φn−1,A)→St(Φn,A) the homomorphism of Steinberg groups. For visual understanding, it is noted that we have the following commutative diagram
In this section, we assume that Φn is a root system either of type simply-laced (i.e., An≥3,Dn≥3 or En=6,7,8) or Cn≥3. We begin with the following observation.
Lemma 3.1. Let R=⨁i≥0Ri be a graded ring and ‘overbar’ denote the reduction modulo the ideal R+. For θ∈St(Φ,R) one has θh∗(0)=θˉ and θh∗(1)=θ.
Proof. An element θ∈St(Φ,R) has the form ∏i=1kxαi(ri), where αi∈Φ and ri=⨁j=0mirij with rij∈Rj. Then
Now, it is easy to observe that ξ(X,0)=1. Hence, ξ(X,Y) belongs to the kernel of St(Φn,A[Y])→St(Φn,A). Thus, we can assume that ξ(X,Y)∈St(Φn,A[Y],YA[Y]). The image of ξ(X,Y) in the ring St(Φn,Aa[Y]) is
(1) Since λa∗(ξ)=1∈St(Φn,Aa[Y]) and analogously λb∗(ξ)=1∈St(Φn,Ab[Y]), by Lemma 3.2(1), for some natural number k, we have ξ(X,akY)=1 and ξ(X,bkY)=1. We choose u,v∈R0 such that uak+vbk=1. This implies ξ(X,−uakY)=1 and ξ(X,−vbkY)=1. From Lemma 3.1, we have θh∗(0)=1. Then
(2) The hypothesis implies that λa∗(ξ)∈Im(lAa[Y]) and λb∗(ξ)∈Im(lAb[Y]). By Lemma 3.2(2), both ξ(X,−uakY) and ξ(X,−vbkY) lie in the image of lA[Y] as in the case (1). Then the identities θ=ξ(1,−vbk)ξ(uak,−uak) implies that θ∈Im(lR). ■
We now prove Theorem 1.1, which we restate here for the reader’s convenience.
Theorem 3.4. Let R=⨁i≥0Ri be a graded ring. Suppose θ∈St(Φn,R,R+). Then
(1) θ=1 if and only if λm∗(θ)=1∈St(Φn,Rm) for all maximal ideals m of R0.
(2) θ∈Im(lR) if and only if λm∗(θ)∈Im(lRm) for all maximal ideals m of R0.
Proof. We first prove that the following set is an ideal of R0 (called the Quillen ideal).
Clearly 0∈Q(R0). If a,b∈Q(R0) and c∈aR0+bR0, then clearly ac,bc∈Q(R0). Note that a and b are comaximal elements of (R0)c. Since λc∗(θ)∈St(Φn,Rc,Rc+), apply Lemma 3.3 to the ring Rc, we get λc∗(θ)=1 (resp. λc∗(θ)∈Im(lRc)), and hence c∈Q(R0). Therefore, Q(R0) is an ideal of R0. It suffices to show if part of the statement. To get this we need to prove that Q(R0)=R0, i.e., 1∈Q(R0). If otherwise, there exists a maximal ideal m of R0 such that Q(R0)⊆m. Since λm∗(θ)=1 (resp. λm∗(θ)∈Im(lRm)), there exists a∈R0∖m such that λa∗(θ)=1 (resp. λm∗(θ)∈Im(lRa)). Then a∈Q(R0)⊆m which is a contradiction. ■
Local-Global Principle for Chevalley Groups
In this section, we deduce an analog of the local-global principle for the Chevalley group over graded rings, and we assume that R has no residue fields of 2 elements when Φ=C2 or G2. We first recall the definition of a relative elementary subgroup.
Definition 4.1. For an ideal I of a ring A, the congruence subgroup G(Φ,A,I) is the kernel of the canonical map G(Φ,A)→G(Φ,A/I). The subgroup E(Φ,I) of G(Φ,A,I) is spanned by elementary root unipotents eα(s) where s∈I and the relative elementary subgroup E(Φ,A,I) is a normal closure of E(Φ,I) in E(Φ,A).
Let R=⨁i≥0Ri be a graded ring. The Swan-Weibel’s homotopy map h:R→R[T] induced a group homomorphism h∗:G(Φ,R)→G(Φ,R[T]). As in Remark 2.7, if θ∈G(Φ,R), for c∈R0, we denote by θh∗(c):=(evc∗∘h∗)(θ).
The following lemma is similar to Lemma 2.5. Hence, we skip the proof.
Lemma 4.2. If f:A→B is a ring homomorphism, then f induces a group homomorphism f∗:E(Φ,A)→E(Φ,B) sending the generators eα(a) to eα(f(a)).
The following lemma generalizes the result of Suslin’s [14 Lemma 3.4], where he proved it over a polynomial ring for an elementary linear group.
Lemma 4.3. Let R=⨁i≥0Ri be a graded ring. Suppose θ∈G(Φ,R), and s∈R0. If λs∗(θ)∈E(Φ,Rs), then there exists an integer k such that, for two indeterminates X,Y, we have
θh∗(X+skY)⋅θh∗(X)−1∈E(Φ,R[X,Y],(Y)).
In particular, for any c,d∈R0 with c≡d(modskR0), we have θh∗(c)⋅θh∗(d)−1∈E(Φ,R).
Proof. Denote Γ:=Γ(Z)=θh∗(Z)=(evT↦ZR[T]→R[Z]∗∘h∗)(θ)∈G(Φ,R[Z]). Since λs∗(θ)∈E(Φ,Rs), by Lemma 4.2, we have λs∗(Γ)=(evT↦ZRs[T]→Rs[Z]∗∘h∗)(λs∗(θ))∈E(Φ,Rs[T]). Consider the following element in G(Φ,R[X,Y,Z],(Y)),
Since λs∗(Γ)∈E(Φ,Rs[Z]), we have λs∗(Γ′)∈E(Φ,Rs[X,Y,Z]). Applying [2 Lemma 1.4] to the ring Rs[X,Y,Z] and the splitting ideal YRs[X,Y,Z], we have
Therefore, λs∗(Γ′)∈E(Φ,Rs[X,Y,Z],(Y)). Then, by [2 Theorem 6.3], there exists an integer k such that Γ′(X,skY,Z)∈E(Φ,R[X,Y,Z],(Y)). Hence, substituting Z=1, we have that
Γ′(X,skY):=θh∗(X+skY)⋅θh∗(X)−1∈E(Φ,R[X,Y],(Y)).
::: For any c,d∈R0 with c=d+skb for some b∈R0 and specializing X=d and Y=b, we get
Γ∗(d,skb):=θh∗(c)⋅θh∗(d)−1∈E(Φ,R).
This gives the desired conclusion.
Theorem 4.4. Let R=⨁i≥0Ri be a graded ring. Suppose θ∈G(Φ,R) with θh∗(0)=1. Then θ∈E(Φ,R) if and only if λm∗(θ)∈E(Φ,Rm) for all maximal ideals m of R0.
Proof. It suffices to show if part of the statement. We shall first prove that if λa∗(θ)∈E(Φ,Ra) and λb∗(θ)∈E(Φ,Rb) with a,b are two co-maximal elements of R0, then θ∈E(Φ,R).
By Lemma 4.3, we can choose k such that θh∗(c)θh∗(d)−1∈E(Φ,R) whenever c≡d(modakR0) or c≡d(modbkR0). Note that aku+bkv=1 for some u,v∈R0. Since θh∗(0)=1, we write
θ=θh∗(1)(θh∗(uak))−1θh∗(uak)(θh∗(0))−1.
Then θ∈E(Φ,R) since aku≡1(modbkR0) and aku≡0(modakR0).
As in Theorem 3.4, it is easy to verify that the set Q(θ)={a∈R0∣λa∗(θ)∈E(Φ,Ra)} is an ideal of R0. To show θ∈E(Φ,R) we need to prove that 1∈Q(R0). If otherwise, there exists a maximal ideal m such that Q(R0)⊆m. Since λm∗(θ)∈E(Φ,Rm), there exists a∈R0∖m such that λa∗(θ)∈E(Φ,Ra). Then a∈Q(R0)⊆m which is a contradiction.
Corollary 4.5. Let R=⨁i≥0Ri be a graded ring and θ∈G(Φ,R). If λm∗(θ)∈G(Φ,(R0)m)⋅E(Φ,Rm) for all maximal ideals m of R0, then θ∈G(Φ,R0)⋅E(Φ,R).
Proof. Let ‘overbar’ denote the reduction modulo the ideal R+. After left-multiplying θ by θ−1, we can assume that θh∗(0)=1. For all maximal ideals m of R0, we have the decomposition λm∗(θ)=η⋅γ where η∈G(Φ,(R0)m) and γ∈E(Φ,Rm). Now, reduction modulo R+, we get I=η⋅γ, and hence η=γ−1. Thus
λm∗(θ)=γ−1⋅γ∈E(Φ,Rm).
Then, by Theorem 4.4, we get θ∈E(Φ,R). Therefore θ∈G(Φ,R0)⋅E(Φ,R) since the original θ has already been modified by a left factor from G(Φ,R0).
Acknowledgments: The author thanks Prof. Manoj Kumar Keshari for suggesting the problem. This material is based on work supported by IIT Bombay Postdoctoral Research Fellowship.
References
Department of Mathematics, Indian Institute of Technology Bombay, Powai, Mumbai 400076, India