betaFeedback

WorksMathematics

Finite power image families and automorphism actions in linear groups

Abstract

Let k/Qk/\mathbb{Q} be a finitely generated field extension. We characterize the subgroups G≤GLn(k)G\leq GL_n(k) for which the power images Pm(G)={gm:g∈G}P_m(G)=\{g^m:g\in G\} form a finite family: this holds precisely when G=N⋊FG=N\rtimes F, where FF is finite and NN is a uniquely divisible subgroup consisting of unipotent matrices. The subgroup NN is canonical and fully invariant.For e=exp⁡(G/N)e=\exp(G/N), we obtain Pm(G)=Pgcd⁡(m,e)(G)P_m(G)=P_{\gcd(m,e)}(G), exactly τ(e)τ(e) distinct power images, and least positive period ee, where τ(e)τ(e) is the number of positive divisors of ee.Over Q\mathbb{Q}, the group coincides with the full group of rational points of its Zariski closure. Since finite automorphism-orbit count implies finiteness of the power-image family, an affirmative answer to Problem~21.40 of the Kourovka Notebook\cite{Kourovka} follows as a corollary. We further prove that a rational linear group GG admits a decomposition G≅(Qd,+)⋊FG\cong(\mathbb{Q}^d,+)\rtimes F, with d≥0d\geq0 finite and FF finite, whenever a virtually nilpotent subgroup of Aut⁡(G)\operatorname{Aut}(G) has finitely many orbits on GG.The latter argument uses nonconstant invariant rational functions arising from nonzero Lie brackets to obstruct dense orbits. A metabelian action with five orbits on the rational Heisenberg group shows that solubility of the acting group is insufficient.

Introduction And Main Results

The structure of groups with finitely many automorphism orbits has been studied under a variety of algebraic hypotheses. For a group GG and a subgroup A≤Aut⁡(G)A \le\operatorname{Aut}(G), consider the natural left action α⋅g=α(g)\alpha\cdot g = \alpha(g) and write

ωA(G)=∣A\G∣,ω(G)=ωAut⁡(G)(G).\omega_A(G) = |A\backslash G|,\qquad\omega(G) = \omega_{\operatorname{Aut}(G)}(G).

Thus A\GA\backslash G is the set of AA-orbits on GG. All automorphisms and endomorphisms in this paper are abstract group maps unless explicitly described as algebraic or linear.

A motivating question is Problem 21.40 of the twenty-first edition of the Kourovka Notebook, proposed by A. Dantas and E. de Melo [13]: if G≤GLn(Q)G \le\mathrm{GL}_n(\mathbb{Q}) and ω(G)<∞\omega(G) < \infty, must GG be virtually soluble?

Two recent preprints present proposed affirmative answers to Problem 21.40. Jayadevan [1, Theorem 1.1] proved if G≤GLn(Q)G \le\mathrm{GL}_n(\mathbb{Q}) and ω(G)<∞\omega(G) < \infty, then GG has a normal torsion-free nilpotent subgroup UU satisfying

class⁡(U)≤n−1,[G:U]≤(2n+1)d≤(2n+1)n2,d=dim⁡Qspan⁡QG.\operatorname{class}(U) \le n - 1,\qquad[G:U] \le(2n+1)^d \le(2n+1)^{n^2},\qquad d = \dim_{\mathbb{Q}}\operatorname{span}_{\mathbb{Q}}G.

A closely related AI-generated argument for the same bound is presented in [2 §3.39]. Both arguments use bounded-degree roots of algebraic numbers to constrain the spectrum and a trace-radical construction to obtain the finite-index nilpotent subgroup; neither requires GG to be finitely generated. Jayadevan also treats finite trace sets and obtains a number-field version by restriction of scalars [1 Theorem 1.2 and Corollary 4.1]. In this paper, we give a structural characterization of subgroups G≤GLn(k)G \le\mathrm{GL}_n(k), where k/Qk/\mathbb{Q} is finitely generated, under the strictly weaker assumption that their power-image family is finite. For k=Qk = \mathbb{Q}, an affirmative answer to Problem 21.40 follows as a corollary. We further prove virtual abelianness for rational linear groups admitting a virtually nilpotent subgroup of automorphisms with finitely many orbits. This refinement is not automatic, as the rational Heisenberg group shows, and its proof connects nonzero Lie brackets with invariant rational functions.

Earlier structural results begin with solubility or virtual nilpotence. Bastos–Dantas–de Melo [4] obtain a radicable nilpotent kernel with finite complement for soluble groups of finite rank with finitely many automorphism orbits. Their later work [5] starts instead with virtual nilpotence and obtains a torsion-free radicable nilpotent characteristic kernel and a torsion complement. Dantas–de Melo–Kato [7] study mixed-order soluble groups of finite rank with four or five automorphism orbits. Dantas–de Sousa [8] prove finiteness for finitely generated locally graded groups with finitely many automorphism orbits and also study finite-orbit phenomena in Mal’cev completions of free nilpotent groups. Accordingly, the finitely generated automorphism-orbit consequence of our structural theorem is not claimed as a separate new result.

For prescribed acting groups, Jabara [11] proves that GG is finite when A≤Aut⁡(G)A \leq\operatorname{Aut}(G) is finitely generated abelian and ωA(G)<∞\omega_A(G) < \infty. Without finite generation of AA, Jabara [12] proves virtual abelianness for an arbitrary abstract group GG when AA is abelian and has finitely many orbits. These results provide the context for our second main theorem, which restricts GG to rational linear groups but permits virtually nilpotent acting subgroups.

The power-image questions of Hrushovski–Kropholler–Lubotzky–Shalev [9] and Lubotzky–Meiri [14] concern powers inside finitely generated groups. Our setting does not assume finite generation of the group, and the finiteness condition studied below need not pass to finitely generated subgroups.

Notations and Main theorems

Notations and Main theorems. For a group GG and a positive integer mm, let

Pm(G)={gm:g∈G},P(G)={Pm(G):m≥1}.P_m(G) = \{g^m : g \in G\}, \qquad\mathcal{P}(G) = \{P_m(G) : m \geq1\}.

The notation Pm(G)P_m(G) denotes the image set of the power map. For every endomorphism α\alpha of GG, the identity α(gm)=α(g)m\alpha(g^m) = \alpha(g)^m gives α(Pm(G))⊆Pm(G)\alpha(P_m(G)) \subseteq P_m(G). Thus Pm(G)P_m(G) is a union of automorphism orbits and contains the identity. Consequently,

ω(G)<∞⟹∣P(G)∣≤2ω(G)−1<∞.(1)\omega(G) < \infty\quad\Longrightarrow\quad|\mathcal{P}(G)| \leq2^{\omega(G)-1} < \infty. \tag*{(1)}

Indeed, the identity forms one orbit, and each of the other orbits can be included or omitted. Different exponents may yield the same set.

Definition 1.1. A group is divisible (or radicable) if every equation xm=gx^m = g, m≥1m \geq1, has a solution. It is uniquely divisible if that solution is always unique. A subgroup is fully invariant if every endomorphism of the ambient group maps it into itself.

The finite residual of GG is

Rf(G)=⋂{L⊴G:[G:L]<∞}.R_f(G) = \bigcap\{L \trianglelefteq G : [G : L] < \infty\}.

For a finite group BB, write exp⁡(B)\exp(B) for its exponent, with exp⁡(1)=1\exp(1) = 1, and let τ(e)\tau(e) be the number of positive divisors of ee.

Theorem A. Let k/Qk/\mathbb{Q} be a finitely generated field extension and let G≤GL⁡n(k)G \leq\operatorname{GL}_n(k). The following conditions are equivalent:

(i) P(G)\mathcal{P}(G) is finite;

(ii) G=N×FG = N \times F, where FF is finite and NN is a uniquely divisible subgroup consisting of unipotent matrices.

When these conditions hold, let HH be the Zariski closure of GG in GL⁡n(k‾)\operatorname{GL}_n(\overline{k}). Then H∘H^\circ is unipotent and

N=G∩H∘={g∈G:g is unipotent}=Rf(G).(2)N = G \cap H^\circ= \{g \in G : g \text{ is unipotent}\} = R_f(G). \tag*{(2)}

The subgroup NN is fully invariant, torsion-free and nilpotent, with nilpotency class class⁡(N)≤n−1\operatorname{class}(N) \leq n - 1. It is also the largest divisible subgroup of GG. If e=exp⁡(G/N)e = \exp(G/N), then

Pe(G)=N,Pm(G)=Pgcd⁡(m,e)(G),∣P(G)∣=τ(e).(3)P_e(G) = N, \qquad P_m(G) = P_{\gcd(m,e)}(G), \qquad|\mathcal{P}(G)| = \tau(e). \tag*{(3)}

The least positive period of the sequence (Pm(G))m≥1(P_m(G))_{m \geq1} is ee. The trivial group has nilpotency class zero.

The assumption on the field is relevant to the forward implication. The reverse implication and the power-image formula follow from a more general abstract observation about finite extensions of uniquely divisible groups, proved in Proposition 3.3.

Corollary 1.2. Let G≤GLn(Q)G \le\mathrm{GL}_n(\mathbb{Q}) and suppose that P(G)\mathcal{P}(G) is finite. With HH and NN as in Theorem A, regarded as a Q\mathbb{Q}-defined algebraic group and its canonical subgroup, respectively,

N=H∘(Q),G=H(Q).N = H^{\circ}(\mathbb{Q}), \qquad G = H(\mathbb{Q}).

Corollary 1.3. Let G≤GLn(k)G \le\mathrm{GL}_n(k), where k/Qk/\mathbb{Q} is finitely generated. If ω(G)<∞\omega(G) < \infty, all conclusions of Theorem A hold, and ω(N)≤ω(G)\omega(N) \le\omega(G). In particular, GG is virtually torsion-free nilpotent, hence virtually soluble and amenable as a discrete group. For k=Qk = \mathbb{Q}, Theorem 1.2 also applies.

Virtual nilpotence cannot in general be strengthened to virtual abelianness: UT3(Q)\mathrm{UT}_3(\mathbb{Q}) is a counterexample, as recalled in § 5. A restriction on the acting automorphism group gives the following conclusion.

Theorem B. Let G≤GLn(Q)G \le\mathrm{GL}_n(\mathbb{Q}). Suppose that there is a virtually nilpotent subgroup A≤Aut⁡(G)A \le\operatorname{Aut}(G) such that ωA(G)<∞\omega_A(G) < \infty. Then the canonical subgroup NN in Theorem A is abelian. In particular,

G≅(Qd,+)⋊FG \cong(\mathbb{Q}^{d}, +) \rtimes F

for some finite integer d≥0d \ge0 and some finite group FF. The same abstract conclusion holds when GG is linear over a number field. Neither GG nor AA is assumed to be finitely generated.

The hypothesis on AA cannot be weakened to solubility, even to metabelianness; see Proposition 5.3.

Theorems A and Corollary 1.2 are proved in § 2–§ 3. § 4 proves Theorem B, and § 5 gives examples and limits of the hypotheses.

Prime roots And the spectrum

We first give the field-theoretic facts needed in the proof.

Lemma 2.1. Let LL be a field of characteristic zero, let pp be prime, and let a∈L×a \in L^{\times}. If a∉(L×)pa \notin(L^{\times})^{p}, then Xp−aX^{p}-a is irreducible over LL.

Proof. If the polynomial were reducible, it would have a root bb with 1≤d=[L(b):L]<p1 \le d = [L(b):L] < p. Taking the field norm in bp=ab^{p}=a gives NL(b)/L(b)p=adN_{L(b)/L}(b)^{p}=a^{d}. Choose integers u,vu,v with ud+vp=1ud+vp=1. Then

a=(NL(b)/L(b)uav)p,a = \left(N_{L(b)/L}(b)^{u}a^{v}\right)^{p},

a contradiction. □

Lemma 2.2. Let L/QL/\mathbb{Q} be finitely generated and let a∈L×a \in L^{\times} be an element that is not a root of unity. Then aa has pp-th roots for only finitely many primes pp in LL, i.e., a∈(L×)pa \in(L^{\times})^{p}

Proof. Suppose first that aa is algebraic over Q\mathbb{Q}. The relative algebraic closure CC of Q\mathbb{Q} in LL is a number field. Indeed, choose a transcendence basis TT for L/QL/\mathbb{Q} and let D=[L:Q(T)]D = [L:\mathbb{Q}(T)]. Every finite subextension C0/QC_0/\mathbb{Q} of C/QC/\mathbb{Q} satisfies

[C0:Q]=[C0(T):Q(T)]≤D.[C_0:\mathbb{Q}] = [C_0(T):\mathbb{Q}(T)] \le D.

The composition of two such subextensions is also a subextension. Choosing one of maximal degree can shows that C/QC/\mathbb{Q} itself is finite. If bp=ab^p=a with b∈Lb \in L, then bb is algebraic over Q\mathbb{Q}, hence lies in CC.

If some finite-prime valuation vp(a)v_p(a) in CC is nonzero, then bp=ab^p=a forces p∣vp(a)p \mid v_p(a) and allows only finitely many primes. Otherwise aa is a unit and so is every such bb. Dirichlet’s unit theorem [17 Chapter I] identifies OC×/μ(C)\mathcal{O}_C^{\times}/\mu(C) with a finitely generated free abelian group. The nonzero class of aa is divisible by only finitely many primes.

Now suppose that aa is transcendental over Q\mathbb{Q}. Extend aa to a transcendence basis a,t2,…,tra,t_2,\ldots,t_r and let E=Q(a,t2,…,tr)E=\mathbb{Q}(a,t_2,\ldots,t_r). In the polynomial ring Q(t2,…,tr)[a]\mathbb{Q}(t_2,\ldots,t_r)[a], by using Eisenstein’s

criterion at the prime aa, we can shows that Xp−aX^p-a is irreducible over EE. If b∈Lb\in L satisfies bp=ab^p=a, then

p=[E(b):E]≤[L:E].p=[E(b):E]\le[L:E].

Again only finitely many primes satisfy the inequality.

Proposition 2.3. Let G≤GL⁡n(k)G\le\operatorname{GL}_n(k), where k/Qk/\mathbb{Q} is finitely generated. If P(G)\mathcal{P}(G) is finite, every eigenvalue of every element of GG is a root of unity. Moreover, only finitely many different eigenvalues occur in GG.

Proof. Assume not, there is a g∈Gg\in G has an eigenvalue λ\lambda which is not a root of unity, and let L=k(λ)L=k(\lambda). This is again a finitely generated field. By Lemma 2.2, there is an infinite set SS of primes p>np>n for which λ∉(L×)p\lambda\notin(L^\times)^p. If p∈Sp\in S and p∤rp\nmid r, we have λr∉(L×)p\lambda^r\notin(L^\times)^p as well. If not, assume λr=yp\lambda^r=y^p for some y∈L×y\in L^\times, since p∤r,gcd⁡(r,p)=1p\nmid r,\gcd(r,p)=1. by Bezout’s theorem, there exist u,vu,v such that ur+vp=1ur+vp=1. Then we have λ=λur+vp=(λr)u(λp)v\lambda=\lambda^{ur+vp}=(\lambda^r)^u(\lambda^p)^v. Then λ=(yp)u(λv)p=(yuλv)p\lambda=(y^p)^u(\lambda^v)^p=(y^u\lambda^v)^p. So λ∈(L×)p\lambda\in(L^\times)^p, contradiction.

If gr∈Pp(G)g^r\in\mathcal{P}_p(G), gr=xpg^r=x^p for x∈Gx\in G, then some eigenvalue μ\mu of xx satisfies μp=λr\mu^p=\lambda^r. Lemma 2.1 and the characteristic polynomial of xx give the contradictory inequalities

p=[L(μ):L]≤[k(μ):k]≤n.p=[L(\mu):L]\le[k(\mu):k]\le n.

Thus gr∉Pp(G)g^r\notin\mathcal{P}_p(G) whenever p∈Sp\in S and p∤rp\nmid r.

Next, we want to construct infinite Pp(G)\mathcal{P}_p(G). For distinct p,q∈Sp,q\in S we have gp∈Pp(G)∖Pq(G)g^p\in\mathcal{P}_p(G)\setminus\mathcal{P}_q(G) since p∤qp\nmid q. The sets Pp(G)\mathcal{P}_p(G) are therefore pairwise distinct, contrary to the hypothesis P(G)\mathcal{P}(G) is finite.

To obtain uniform finiteness of the spectrum, choose a transcendence basis TT of k/Qk/\mathbb{Q} and let D=[k:Q(T)]D=[k:\mathbb{Q}(T)]. If a root of unity ζ\zeta is as an eigenvalue, then

[Q(ζ):Q]=[Q(ζ)(T):Q(T)]≤[k(ζ):Q(T)]≤nD.[\mathbb{Q}(\zeta):\mathbb{Q}]=[\mathbb{Q}(\zeta)(T):\mathbb{Q}(T)]\le[k(\zeta):\mathbb{Q}(T)]\le nD.

A primitive rr-th root of unity has degree φ(r)\varphi(r) over Q\mathbb{Q}. Only finitely many rr satisfy φ(r)≤nD\varphi(r)\le nD, so only finitely many roots of unity can satisfy the condition.

Proposition 2.4. Under the hypotheses of Proposition 2.3, let H=GZar⁡H=G^{\operatorname{Zar}} and N0=G∩H∘N_0=G\cap H^\circ. Then H∘H^\circ is unipotent, N0⊴GN_0\trianglelefteq G has finite index, and N0N_0 is torsion-free nilpotent of class at most n−1n-1. Furthermore, N0N_0 is exactly the set of unipotent elements of GG.

Proof. By Proposition 2.3, only finitely many characteristic polynomials occur on GG. Set this finite set as Λ={λ1,…,λs}\Lambda=\{\lambda_1,\ldots,\lambda_s\}. For any g∈Gg\in G, the characteristic polynomial of gg is χg(X)=∏j=1n(X−μj)\chi_g(X)=\prod_{j=1}^n(X-\mu_j), where μj∈Λ\mu_j\in\Lambda. Hence {χg:g∈G}\{\chi_g:g\in G\} the possibilities of characteristic polynomials also finite. For any χh(X)=det⁡(XIn−h)=Xn+c1(h)Xn−1+⋯+cn(h)\chi_h(X)=\det(XI_n-h)=X^n+c_1(h)X^{n-1}+\cdots+c_n(h). ci(h)c_i(h) is a polynomial of hjkh_{jk}. Thus, ci:H∘→A1c_i:H^\circ\to\mathbb{A}^1 is regular function on H∘H^\circ. They take finitely many values on the Zariski-dense subgroup G∩H∘G\cap H^\circ, hence finitely many values on H∘H^\circ itself. Since a connected algebraic group is irreducible, each coefficient function is constant. (if not, then reducible) Its value is determined at the identity. Thus

det⁡(XIn−h)=χIn(X)=det⁡(XIn−In)=(X−1)n(∀h∈H∘).\det(XI_n-h)=\chi_{I_n}(X)=\det(XI_n-I_n)=(X-1)^n \qquad(\forall h\in H^\circ).

So H∘H^\circ is unipotent. The triangularization theorem for unipotent algebraic groups [10] places it inside a conjugate of

UT⁡n(k‾)={(1∗⋯∗01⋯∗⋮⋱⋱⋮0⋯01)}.\operatorname{UT}_n(\overline{k})= \left\{ \begin{pmatrix} 1&*&\cdots&*\\ 0&1&\cdots&*\\ \vdots&\ddots&\ddots&\vdots\\ 0&\cdots&0&1 \end{pmatrix} \right\}.

UT⁡n\operatorname{UT}_n is nilpotent and nilpotency class is n−1n-1. Consequently N0N_0 is nilpotent of class at most n−1n-1. Nonidentity unipotent matrices in characteristic zero have infinite order. The group of components H/H∘H/H^\circ is finite since HH has finite components. Considering the natural map G→H/H∘G\to H/H^\circ, so the kernel is N0=G∩H∘N_0=G\cap H^\circ. Identity component is normal subgroup of algebraic group H∘⊴HH^\circ\trianglelefteq H, so does N0N_0. We have G/N0↪H/H∘G/N_0\hookrightarrow H/H^\circ, so N0N_0 is finite index normal subgroup in GG. Next we want to show every unipotent element uu of an algebraic group in characteristic zero belongs to its identity component. If u≠Inu \ne I_n, the polynomial map

γ:A1→GL⁡nt↦exp⁡(tlog⁡u)\gamma: \mathbb{A}^{1} \to\operatorname{GL}_n \qquad t \mapsto\exp(t\log u)

is an algebraic one-parameter subgroup, with image isomorphic to the additive group. For any m∈Zm \in\mathbb{Z}, γ(m)=exp⁡(mlog⁡u)=um\gamma(m) = \exp(m\log u) = u^m. γ(Z)⊂H\gamma(\mathbb{Z}) \subset H since u∈Hu \in H. Z\mathbb{Z} is Zariski dense in A1\mathbb{A}^{1}, so does γ(Z)\gamma(\mathbb{Z}) in γ(A1)\gamma(\mathbb{A}^{1}). Hence the image lies in the algebraic group and is connected, I=γ(0)I = \gamma(0) and u=γ(1)u = \gamma(1). Applying this to HH gives u∈H∘u \in H^\circ. Every unipotent element of HH belongs to H∘H^\circ. Thus, if u0∈Gu_0 \in G be unipotent, then u∈G∩H∘=N0u \in G \cap H^\circ= N_0, N0={g∈G:g is unipotent}N_0 = \{g \in G : g \text{ is unipotent}\}.

Stable power images and the canonical divisible subgroup

For a unipotent matrix uu, the finite matrix logarithm is

log⁡u=∑j=1n−1(−1)j+1j(u−In)j.\log u = \sum_{j=1}^{n-1} \frac{(-1)^{j+1}}{j}(u-I_n)^j.

It is inverse to the finite exponential on nilpotent matrices. In particular, log⁡(um)=mlog⁡u\log(u^m) = m\log u. Thus power maps are injective on any subgroup of unipotent matrices in characteristic zero.

Lemma 3.1. Let GG be a group with finite P(G)\mathcal{P}(G). Suppose that N0⊴GN_0 \trianglelefteq G has finite index dd and that every positive-integer power map on N0N_0 is injective. Then N0N_0 is uniquely divisible and N0=Pd(G)N_0 = P_d(G). In particular, N0N_0 is fully invariant.

Proof. Fix q≥1q \ge1 and let a=dqa = dq. The descending chain

Pa(G)⊇Pa2(G)⊇Pa3(G)⊇⋯P_a(G) \supseteq P_{a^2}(G) \supseteq P_{a^3}(G) \supseteq\cdots

stabilizes, since all its terms belong to the finite family P(G)\mathcal{P}(G). Suppose j≥1j \ge1 with Paj(G)=Paj+1(G)P_{a^j}(G) = P_{a^{j+1}}(G). For u∈N0u \in N_0, we can find x∈Gx \in G satisfying

uaj=xaj+1=(xa)aj.u^{a^j} = x^{a^{j+1}} = (x^a)^{a^j}.

Since xd∈N0x^d \in N_0, also xa∈N0x^a \in N_0. Injectivity on N0N_0 gives

u=xa=(xd)q.u = x^a = (x^d)^q.

Thus every element of N0N_0 has a qq-th root in N0N_0. Such a root is unique by hypothesis. Since qq was arbitrary, N0N_0 is uniquely divisible. The inclusion Pd(G)⊆N0P_d(G) \subseteq N_0 follows from the finite quotient, and the opposite inclusion follows from divisibility of N0N_0. Every endomorphism of GG preserves Pd(G)P_d(G) as a subset.

We include the splitting argument to separate the stabilization step from the standard extension theory it subsequently uses.

Lemma 3.2. If N⊴GN \trianglelefteq G is uniquely divisible and nilpotent, and G/NG/N is finite, then G=N⋊FG = N \rtimes F for a finite subgroup F≤GF \le G.

Proof. In fact, we need to prove the short exact sequence

1⟶N→ιG→πB⟶1.1 \longrightarrow N \xrightarrow{\iota} G \xrightarrow{\pi} B \longrightarrow1.

must split under the conditions of the lemma, where B=G/NB = G/N. i.e., there is a group homomorphism s:B⟶Gs : B \longrightarrow G satisfies π∘s=id⁡B\pi\circ s = \operatorname{id}_B. Let F=s(B)F = s(B), since BB is finite and ss is injective, so does FF. For any g∈Gg \in G, we have g=nfg = nf, where n∈Nn \in N, f∈Ff \in F, so we have G=N⋊FG = N \rtimes F.

Since NN is nilpotent, so we have

1=Z0(N)≤Z1(N)≤⋯≤Zc(N)=N,1 = Z_0(N) \le Z_1(N) \le\cdots\le Z_c(N) = N,

where cc is the nilpotency class of NN. The case N=1N = 1 is immediate. We induct on the nilpotency class of NN. In the abelian case, NN is uniquely divisible, so for any v∈Nv \in N, integer qq, equation qw=vqw = v has unique solution. Then we can define pqv=p(1qv)p_qv = p\left(\frac{1}{q}v\right). It’s easy check NN is a Q\mathbb{Q}-vector space. We will show any extension of a finite group BB by a Q[B]\mathbb{Q}[B]-module VV splits. An extension is represented by an additive 22-cocycle c(a,b)c(a,b). Set

t(a)=1∣B∣∑x∈Bc(a,x).t(a)=\frac{1}{|B|}\sum_{x\in B}c(a,x).

Summing the identity c(a,b)+c(ab,x)=a⋅c(b,x)+c(a,bx)c(a,b)+c(ab,x)=a\cdot c(b,x)+c(a,bx) over x∈Bx\in B gives ∑x∈Bc(a,b)+∑x∈Bc(ab,x)=∣B∣c(a,b)+∣B∣t(ab)\sum_{x\in B}c(a,b)+\sum_{x\in B}c(ab,x)=|B|c(a,b)+|B|t(ab) on the left side, ∑x∈Ba⋅c(b,x)+∑x∈Bc(a,bx)=∣B∣a⋅t(b)+∣B∣t(a)\sum_{x\in B}a\cdot c(b,x)+\sum_{x\in B}c(a,bx)=|B|a\cdot t(b)+|B|t(a) on the right side. These give

c(a,b)=t(a)+a⋅t(b)−t(ab).c(a,b)=t(a)+a\cdot t(b)-t(ab).

In fact, this cocycle is a coboundary, so adjusting the section gives a homomorphic section. Thus, split.

Now, we prove the non-abelian part. For the induction step we let Z=Z(N)Z=Z(N). The unique mm-th root of a central element is central, because conjugation preserves that root equation. Thus ZZ is uniquely divisible. Also N/ZN/Z is uniquely divisible. Since NN is nonabelian and nilpotent, the nilpotency class of N/ZN/Z is strictly smaller than that of NN. We therefore apply the induction hypothesis to the extension

1⟶N/Z⟶G/Z⟶G/N⟶1.1\longrightarrow N/Z\longrightarrow G/Z\longrightarrow G/N\longrightarrow1.

It follows that there exists a finite subgroup F‾≤G/Z\overline{F}\leq G/Z such that G/Z=(N/Z)×F‾G/Z=(N/Z)\times\overline{F}. Let q:G⟶G/Zq:G\longrightarrow G/Z be the quotient map, and define E=q−1(F‾)E=q^{-1}(\overline{F}). Then E/Z≅F‾E/Z\cong\overline{F}, E∩N=ZE\cap N=Z. Indeed, if x∈E∩Nx\in E\cap N, then xZ∈F‾∩(N/Z)=1xZ\in\overline{F}\cap(N/Z)=1, so x∈Zx\in Z. Thus we obtain another short exact sequence

1⟶Z⟶E⟶F‾⟶1.1\longrightarrow Z\longrightarrow E\longrightarrow\overline{F}\longrightarrow1.

where ZZ is a uniquely divisible abelian group, hence a Q[F‾]\mathbb{Q}[\overline{F}]-module, and F‾\overline{F} is finite. Because the abelian case proved above, this extension splits.

Consequently, there exists a finite subgroup F≤EF\leq E such that

E=Z×F,F≅F‾.E=Z\times F,\qquad F\cong\overline{F}.

Since G/Z=(N/Z)F‾G/Z=(N/Z)\overline{F}, taking inverse images under qq gives G=NE=NFG=NE=NF. Furthermore, N∩F⊆N∩E=ZN\cap F\subseteq N\cap E=Z, while F∩Z=1F\cap Z=1 because FF is a complement to ZZ in EE. Hence N∩F=1N\cap F=1. Since N⊴GN\trianglelefteq G, we conclude that G=N⋊FG=N\rtimes F, where FF is finite.

We can use the exponent of the finite quotient to control the Pm(G)P_m(G) for GG has uniquely divisible normal subgroup NN and finite quotient, Nilpotence of NN and splitting of the extension are not required.

Proposition 3.3. Let N⊴GN\trianglelefteq G be uniquely divisible and let B=G/NB=G/N be finite of exponent ee. Let T(G)T(G) denote the set of finite-order elements of GG. Then

Pm(G)=⋃s∈T(G)smCN(s)(m≥1).(4)P_m(G)=\bigcup_{s\in T(G)}s^mC_N(s)\qquad(m\geq1). \tag*{(4)}

Where CN(s)C_N(s) is the centralizer of ss in NN. Moreover, Pe(G)=NP_e(G)=N, Pm(G)=Pgcd⁡(m,e)(G)P_m(G)=P_{\gcd(m,e)}(G), ∣P(G)∣=τ(e)|\mathcal{P}(G)|=\tau(e), and ee is the least positive period of the sequence of power images.

Proof. A uniquely divisible group is torsion-free. For g∈Gg\in G, let u∈Nu\in N be the unique ee-th root of geg^e. B=G/NB=G/N is finite and with index ee. For any g∈Gg\in G, (gN)e=N(gN)^e=N, ge∈Ng^e\in N, there exist unique u∈Nu\in N such that ue=geu^e=g^e, so (ugu−1)e=gueg−1=ggeg−1=ge(ugu^{-1})^e=gu^eg^{-1}=gg^eg^{-1}=g^e. ugu−1ugu^{-1} and uu are ee-th roots of geg^e in NN. ugu−1∈Nugu^{-1}\in N, hence ugu−1=uugu^{-1}=u. Let s=gu−1s=gu^{-1}. Then

g=su=us,se=1.g=su=us,\qquad s^e=1.

Every finite-order element of GG has order dividing ee: its ee-th power lies in the torsion-free group NN and has finite order. For s∈T(G)s \in T(G) the subgroup CN(s)C_N(s) is uniquely divisible, because for v∈CN(s)v \in C_N(s), there exist unique w∈Nw \in N, such that wm=vw^m=v, (sws−1)m=swms−1=svs−1=v(sws^{-1})^m=sw^ms^{-1}=svs^{-1}=v, sws−1sws^{-1} is also a mm-th root of vv in NN, by uniqueness, sws−1=wsws^{-1}=w, w∈CN(s)w \in C_N(s).

Next, We want to show Pm(G)=⋃s∈T(G)smCN(s)P_m(G)=\bigcup_{s\in T(G)}s^mC_N(s). For any x∈Pm(G)x \in P_m(G). there exist g∈Gg \in G, such that x=gmx=g^m. we’ve proved for g∈Gg \in G can be uniquely write as g=su=usg=su=us, where s∈T(G)s \in T(G), u∈Nu \in N. Because ss and uu commute, u∈CN(s)u \in C_N(s). gm=smumg^m=s^mu^m, um∈CN(s)u^m \in C_N(s). Thus x=smum∈smCN(s)x=s^mu^m \in s^mC_N(s). We have Pm(G)⊆⋃s∈T(G)smCN(s)P_m(G)\subseteq\bigcup_{s\in T(G)}s^mC_N(s).

For any x∈smCN(s)x \in s^mC_N(s), then x=smvx=s^mv, for any v∈CN(s)v \in C_N(s). CN(s)C_N(s) is uniquely divisible, there exist w∈CN(s)w \in C_N(s) such that wm=vw^m=v. Since sw=wssw=ws, (sw)m=smwm=smv=x(sw)^m=s^mw^m=s^mv=x. Then x∈Pm(G)x \in P_m(G). Thus, ⋃s∈T(G)smCN(s)⊆Pm(G)\bigcup_{s\in T(G)}s^mC_N(s)\subseteq P_m(G). We have Pm(G)=⋃s∈T(G)smCN(s)(m≥1)P_m(G)=\bigcup_{s\in T(G)}s^mC_N(s)(m\geq1).

It follows immediately that Pm+e(G)=Pm(G)P_{m+e}(G)=P_m(G) and that Pe(G)=NP_e(G)=N.

Let d=gcd⁡(m,e)d=\gcd(m,e). There is an integer aa prime to ee such that da≡m(mode)da\equiv m\pmod e. Indeed, the class m/dm/d is a unit modulo e/de/d, and it can be lifted to a unit modulo ee by the Chinese remainder theorem. The map s↦sas\mapsto s^a is a bijection of T(G)T(G), and CN(sa)=CN(s)C_N(s^a)=C_N(s). Substitution in (4) yields Pm(G)=Pd(G)P_m(G)=P_d(G).

Finally, we need to prove the sets Pd(G)P_d(G) for d∣ed\mid e are distinct. To see this, pass to the finite quotient BB. For each prime p∣ep\mid e, let b=vp(e)b=v_p(e). The largest pp-part of the order of an element of Pd(B)P_d(B) is pb−vp(d)p^{b-v_p(d)}. The upper bound follows by taking dd-th powers; equality follows by choosing an element whose order has pp-part pbp^b, and then taking a power to obtain an element of order exactly pb−vp(d)p^{b-v_p(d)}. Such an element exists by the definition of the exponent; an element of order ee is not required. Thus different divisors dd give different power images in BB, hence in GG. This proves ∣P(G)∣=τ(e)|\mathcal{P}(G)|=\tau(e). Finally, if t>0t>0 is a period, then Pe+t(B)=Pe(B)={1}P_{e+t}(B)=P_e(B)=\{1\}. Consequently e∣e+te\mid e+t, so e∣te\mid t. Hence ee is the least period. □

Proof of Theorem A. Assume (i). Proposition 2.4 provides N0=G∩H∘N_0=G\cap H^\circ, a finite-index normal unipotent subgroup. The matrix logarithm makes its power maps injective. Lemma 3.1 therefore makes it uniquely divisible and fully invariant. Set N=N0N=N_0 and apply Lemma 3.2 to obtain G=N⋊FG=N\rtimes F with FF finite. This proves (ii). Conversely, (ii) implies (i) by Proposition 3.3.

Proposition 2.4 proves the first two equalities in (1.2) and the nilpotency-class bound. Every homomorphism from a divisible group to a finite group is trivial: if the target has exponent rr, write an element of the source as an rr-th power. Hence NN lies in every finite-index normal subgroup of GG. Since G/NG/N is finite, N=Rf(G)N=R_f(G). The same argument applied to an arbitrary divisible subgroup of GG shows that it is contained in NN. Finally, Proposition 3.3 proves (1.3) and the least-period assertion. The subgroup in (ii) is canonical, although the chosen finite complement is not claimed to be canonical. □

Proof of Corollary 1.3. Apply (1.1) and Theorem A. Since NN is characteristic, the restricted action of Aut⁡(G)\operatorname{Aut}(G) on NN has at most ω(G)\omega(G) orbits. The full group Aut⁡(N)\operatorname{Aut}(N) has at most that many orbits. Nilpotent groups and their finite extensions are amenable. □

Before we prove the Corollary 1.2. We use the following standard form of Mal’cev completion from Baues–Grunewald [6 §4.2].

Lemma 3.4 (Mal’cev completion). Let UU be a unipotent algebraic group defined over Q\mathbb{Q}, and let Γ≤U(Q)\Gamma\leq U(\mathbb{Q}) be finitely generated and Zariski dense. Then for every x∈U(Q)x\in U(\mathbb{Q}) there is m≥1m\geq1 with xm∈Γx^m\in\Gamma. Equivalently, U(Q)U(\mathbb{Q}) is the radicable hull of Γ\Gamma.

Lemma 3.5. Every Zariski-dense subgroup DD of a unipotent algebraic group UU in characteristic zero contains a finitely generated Zariski-dense subgroup.

Proof. Every closed algebraic subgroup of UU is connected: each of its elements lies in a connected one-parameter subgroup, as in Proposition 2.4. Considering all the finitely generated subgroups Λ=⟨d1,…,dr⟩≤D\Lambda=\langle d_1,\ldots,d_r\rangle\leq D. Their Zariski closure satisfies 0≤dim⁡Λ‾Zar≤dim⁡U0\leq\dim\overline{\Lambda}^{\mathrm{Zar}}\leq\dim U. Choose Γ≤D\Gamma\leq D be the finitely generated group whose Zariski closure has maximal dimension. For any x∈Dx\in D, the closure of ⟨Γ,x⟩\langle\Gamma,x\rangle contains Γ‾Zar\overline{\Gamma}^{\mathrm{Zar}} and has the same dimension by maximality. These are connected algebraic groups, so they are equal. Thus D⊆Γ‾ZarD \subseteq\overline{\Gamma}^{\mathrm{Zar}}, and density gives Γ‾Zar=U\overline{\Gamma}^{\mathrm{Zar}}=U. □

Proof of Corollary 1.2. The Zariski closure HH of the subgroup of rational matrices GG is Q\mathbb{Q}-defined; so is its identity component U=H∘U=H^{\circ}. The canonical subgroup N=G∩UN=G\cap U is Zariski dense in UU. Choose a finitely generated Zariski-dense subgroup Γ≤N\Gamma\leq N by Lemma 3.5. If x∈U(Q)x\in U(\mathbb{Q}), then xm∈Γ⊆Nx^{m}\in\Gamma\subseteq N for some mm by Lemma 3.4. Unique divisibility of NN supplies y∈Ny\in N with ym=xmy^{m}=x^{m}. Injectivity of powers on unipotent matrices implies y=xy=x. Therefore N=U(Q)N=U(\mathbb{Q}).

Write G=N⋊FG=N\rtimes F as in Theorem A. G=⋃f∈FNfG=\bigcup_{f\in F}Nf. Since FF is finite, we have

H=G‾Zar=⋃f∈FNf‾Zar=⋃f∈FN‾Zarf.H=\overline{G}^{\mathrm{Zar}}=\bigcup_{f\in F}\overline{Nf}^{\mathrm{Zar}}=\bigcup_{f\in F}\overline{N}^{\mathrm{Zar}}f.

right multi ff is algebraic isomorphism and Zariski homeomorphism. Thus

Nf‾Zar=N‾Zarf=Uf.\overline{Nf}^{\mathrm{Zar}}=\overline{N}^{\mathrm{Zar}}f=Uf.

So

H=⋃f∈FUf.H=\bigcup_{f\in F}Uf.

Each representative ff is a rational matrix. Taking rational points therefore gives H(Q)=U(Q)F=NF=GH(\mathbb{Q})=U(\mathbb{Q})F=NF=G. □

Corollary 3.6. For n≥1n\geq1, define

Mn=lcm⁡{r≥1:φ(r)≤n}.M_n=\operatorname{lcm}\{r\geq1:\varphi(r)\leq n\}.

If G≤GLn(Q)G\leq\mathrm{GL}_n(\mathbb{Q}) and P(G)\mathcal{P}(G) is finite, then

exp⁡(G/N)∣Mn,PMn(G)=N,Pm+Mn(G)=Pm(G)(m≥1).\exp(G/N)\mid M_n,\qquad P_{M_n}(G)=N,\qquad P_{m+M_n}(G)=P_m(G)\quad(m\geq1).

Proof. A finite-order rational n×nn\times n matrix is diagonalizable over C\mathbb{C}. The order of each eigenvalue ζ\zeta satisfies φ(ord⁡(ζ))≤n\varphi(\operatorname{ord}(\zeta))\leq n, so the matrix order divides MnM_n. Apply this to the finite complement FF, which is isomorphic to G/NG/N, and use Proposition 3.3. □

Corollary 3.7. Under the hypotheses of Theorem A, GG is residually finite if and only if it is finite. If GG is finitely generated, it is finite.

Proof. The first conclusion follows from Rf(G)=NR_f(G)=N and the finiteness of G/NG/N. For the second, Schreier’s theorem makes NN finitely generated. Its abelianization is a finitely generated divisible abelian group, hence trivial. A perfect nilpotent group is trivial, so N=1N=1. □

Virtual abelianness and nilpotent automorphism actions

We begin with an abstract observation that identifies the obstruction to virtual abelianness in Theorem A.

Lemma 4.1. A divisible group has no proper finite-index subgroup. If N⊴GN\trianglelefteq G is divisible and of finite index, then

G is virtually abelian⟺N is abelian.G\text{ is virtually abelian}\Longleftrightarrow N\text{ is abelian}.

Proof. Every homomorphism from a divisible group DD to a finite group QQ is trivial: Considering the group homomorphism φ:D→Q\varphi:D\to Q, let r=exp⁡(Q)r=\exp(Q) for any x∈Dx\in D. By divisible, there exists a y∈Dy\in D, such that yr=xy^{r}=x, then φ(x)=φ(yr)=φ(y)r=1\varphi(x)=\varphi(y^{r})=\varphi(y)^{r}=1. So φ(D)={1}\varphi(D)=\{1\}. If [D:L]=s<∞[D:L]=s<\infty, considering the set of cosets D/L={xL:x∈D}D/L=\{xL:x\in D\}, which has ss elements. DD acts on it by d⋅(xL)=(dx)Ld\cdot(xL)=(dx)L. We get a group homomorphism ρ:D→Ss\rho:D\to S_s. DD is divisible and SsS_s is finite, so ρ\rho is trivial. For each d∈Dd\in D, dL=LdL=L. Hence d∈Ld\in L holds for each d∈Dd\in D, D=LD=L, divisible group has no proper finite-index subgroup.

If N⊴GN\trianglelefteq G, [G:N]<∞[G:N]<\infty, and NN is divisible. If NN is abelian, it’s finite abelian subgroup of GG, so GG is virtually abelian. Conversely, if GG is virtually abelian, then there exist an

abelian subgroup D≤GD \leq G, [G:D]<∞[G : D] < \infty. Then [N:N∩D]≤[G:D]<∞[N : N \cap D] \leq[G : D] < \infty by the injective N/(N∩D)⟶G/DN/(N \cap D) \longrightarrow G/D, n(N∩D)⟼nDn(N \cap D) \longmapsto nD. NN is divisible, it has no proper finite index subgroup, so N∩D=NN \cap D = N, N≤DN \leq D, NN is also abelian.

The next lemma is the algebraic ingredient. The Lie algebra in its statement is not assumed to be nilpotent.

Lemma 4.2 (A dense-orbit obstruction). Let KK be an algebraically closed field of characteristic zero, and let l\mathfrak{l} be a finite-dimensional Lie algebra over KK. Suppose that a connected nilpotent linear algebraic group BB acts rationally on l\mathfrak{l} by Lie algebra automorphisms. If l\mathfrak{l} is nonabelian, then its rational function field contains a nonconstant BB-invariant function. Consequently, if BB has a Zariski-dense orbit in l\mathfrak{l}, then l\mathfrak{l} is abelian.

Proof. A connected nilpotent algebraic group in characteristic zero decomposes as

B=T×W,B = T \times W,

where TT is a algebraic torus and WW is connected and unipotent, the characteristic of algebraic torus TT is χ:T⟶K×\chi:T \longrightarrow K^{\times}.; see [16 Corollary 16.48]. Write the weight-space decomposition for the TT-action as

l=⨁χ∈Slχ,lχ={v:t⋅v=χ(t)v for all t∈T}.\mathfrak{l} = \bigoplus_{\chi\in S} \mathfrak{l}_{\chi}, \qquad\mathfrak{l}_{\chi} = \{v : t \cdot v = \chi(t)v \text{ for all } t \in T\}.

Here SS is the finite set of characters χ:T⟶K×\chi:T \longrightarrow K^{\times} that occur in the representation; see [16 Chapter 12]. Because TT centralizes WW, every lχ\mathfrak{l}_{\chi} is WW-stable.

A nonzero representation of a unipotent group has a nonzero fixed vector [16 Section 14b]. Applied to lχ∗=Hom⁡K(lχ,K)\mathfrak{l}_{\chi}^{*} = \operatorname{Hom}_{K}(\mathfrak{l}_{\chi},K) this yields a nonzero WW-invariant linear form fχf_{\chi} on lχ\mathfrak{l}_{\chi}. Define the action of WW on the dual space: (w⋅f)(v)=f(w−1⋅v)(w \cdot f)(v) = f(w^{-1} \cdot v). If fχf_{\chi} is invariant, then fχ(w−1⋅v)=fχ(v)f_{\chi}(w^{-1} \cdot v) = f_{\chi}(v). lχ∗\mathfrak{l}_{\chi}^{*} is invariant vector. Changing ww to w−1w^{-1}, we have fχ(w⋅v)=fχ(v)f_{\chi}(w \cdot v) = f_{\chi}(v). Now fχf_{\chi} only defined on lχ\mathfrak{l}_{\chi}. For any v∈lv \in\mathfrak{l} has unique decomposition

v=∑ψ∈Svψ,vψ∈lψ.v = \sum_{\psi\in S} v_{\psi}, \qquad v_{\psi} \in\mathfrak{l}_{\psi}.

Extend fχf_{\chi} to the whole l\mathfrak{l}, define fχ(v)=fχ(vχ)f_{\chi}(v) = f_{\chi}(v_{\chi}). For v∈lχv \in\mathfrak{l}_{\chi}, we have t⋅v=χ(t)vt \cdot v = \chi(t)v. fχ(t⋅v)=χ(t)fχ(v)f_{\chi}(t \cdot v) = \chi(t)f_{\chi}(v) and since For l\mathfrak{l}, fχ((tw)⋅v)=χ(t)fχ(v)f_{\chi}((tw) \cdot v) = \chi(t)f_{\chi}(v) (t∈T,w∈W)(t \in T, w \in W).

This is an evaluation formula; it avoids any convention about the contragredient action on functions.

Suppose that l\mathfrak{l} is nonabelian. Some χ,ψ∈S\chi,\psi\in S satisfy [lχ,lψ]≠0[\mathfrak{l}_{\chi},\mathfrak{l}_{\psi}] \neq0. Since TT preserves the bracket, [lχ,lψ]⊆lχψ[\mathfrak{l}_{\chi},\mathfrak{l}_{\psi}] \subseteq\mathfrak{l}_{\chi\psi}. Thus θ=χψ\theta= \chi\psi also belongs to SS. On the nonempty open set fθ≠0f_{\theta} \neq0, define

R=fχfψfθ.R = \frac{f_{\chi}f_{\psi}}{f_{\theta}}.

eq:4.1 and θ=χψ\theta= \chi\psi show that R((tw)⋅v)=R(v)R((tw) \cdot v) = R(v). Hence RR is BB-invariant. It is nonzero, since the coordinate ring of a vector space is a domain. Moreover,

R(cv)=cR(v)(c∈K×).R(cv) = cR(v) \qquad(c \in K^{\times}).

It therefore cannot be constant. This remains valid when some of χ,ψ,θ\chi,\psi,\theta coincide and factors cancel in the quotient.

Finally, suppose that BB has a dense orbit, O=B⋅v0\mathcal{O} = B \cdot v_{0}, O‾Zar=l\overline{\mathcal{O}}^{\mathrm{Zar}} = \mathfrak{l}. We have R=fχfψfθR = \frac{f_{\chi}f_{\psi}}{f_{\theta}}. Because Ω={fθ≠0}\Omega= \{f_{\theta} \neq0\} is nonempty open set, O\mathcal{O} is dense, O∩Ω≠∅\mathcal{O} \cap\Omega\neq\varnothing. And Ω\Omega is invariant under BB, since fθ((tw)⋅v)=θ(t)fθ(v)f_{\theta}((tw) \cdot v) = \theta(t)f_{\theta}(v), where θ(t)≠0\theta(t) \neq0. Hence, O⊆Ω\mathcal{O} \subseteq\Omega. Take v∈Ov \in\mathcal{O}, let λ=R(v)\lambda= R(v). Since RR is invariant, for all b∈Bb \in B, R(b⋅v)=R(v)=λR(b \cdot v) = R(v) = \lambda. So R=λR = \lambda on the O\mathcal{O}. On O\mathcal{O}, fχfψfθ=λ\frac{f_{\chi}f_{\psi}}{f_{\theta}}=\lambda, F=fχfψ−λfθF=f_{\chi}f_{\psi}-\lambda f_{\theta} is zero on the O\mathcal{O}, O\mathcal{O} is Zariski dense, so FF should be zero on the l:fχfψ=λfθl:f_{\chi}f_{\psi}=\lambda f_{\theta}. The rational function R=λR=\lambda is contradict with RR is non-constant.

Proposition 4.3. Let l\mathfrak{l} be a finite-dimensional Lie algebra over Q\mathbb{Q}. If a virtually nilpotent subgroup Γ≤Aut⁡Q(l)\Gamma\leq\operatorname{Aut}_{\mathbb{Q}}(\mathfrak{l}) has finitely many orbits on l\mathfrak{l}, then l\mathfrak{l} is abelian.

Proof. The zero-dimensional case is immediate. Choose a finite-index nilpotent subgroup Γ0≤Γ\Gamma_{0}\leq\Gamma. An orbit under a group splits into at most its subgroup index many orbits under a finite-index subgroup. Thus Γ0\Gamma_{0} still has finitely many orbits on l\mathfrak{l}.

Set V=l⊗QQ‾V=\mathfrak{l}\otimes_{\mathbb{Q}}\overline{\mathbb{Q}}, and let CC be the Zariski closure of Γ0\Gamma_{0} in Aut⁡Q‾(V)\operatorname{Aut}_{\overline{\mathbb{Q}}}(V). The latter is a closed algebraic subgroup of GL⁡(V)\operatorname{GL}(V): preserving the bracket is expressed by polynomial equations in matrix entries. If Γ0\Gamma_{0} has nilpotency class at most cc, its (c+1)(c+1)-fold commutator identities extend to CC by Zariski density, since word maps on algebraic groups are regular. Hence CC is nilpotent.

Let B=C∘B=C^{\circ} and Γ1=Γ0∩B\Gamma_{1}=\Gamma_{0}\cap B. The component group is finite, so [Γ0:Γ1]<∞[\Gamma_{0}:\Gamma_{1}]<\infty and Γ1\Gamma_{1} still has finitely many orbits. Choose representatives v1,…,vs∈lv_{1},\ldots,v_{s}\in\mathfrak{l}. Then

l=⋃i=1sΓ1⋅vi⊆⋃i=1sB⋅vi.\mathfrak{l}=\bigcup_{i=1}^{s}\Gamma_{1}\cdot v_{i}\subseteq\bigcup_{i=1}^{s}B\cdot v_{i}.

The rational vector space l\mathfrak{l} is Zariski dense in VV. Therefore

V=⋃i=1sB⋅vi‾Zar⁡.V=\bigcup_{i=1}^{s}\overline{B\cdot v_{i}}^{\operatorname{Zar}}.

Since VV is irreducible, at least one B⋅viB\cdot v_{i} is dense. Lemma 4.2 implies [V,V]=0[V,V]=0, hence [l,l]=0[\mathfrak{l},\mathfrak{l}]=0.

Proof of Theorem B. Since ω(G)≤ωA(G)<∞\omega(G)\leq\omega_{A}(G)<\infty, Theorems A and 1.2 give

G=N⋊F,N=U(Q),U=H∘.G=N\rtimes F,\qquad N=U(\mathbb{Q}),\qquad U=H^{\circ}.

where UU is a unipotent Q\mathbb{Q}-group and NN is characteristic in GG. Thus AA acts on NN with finitely many orbits.

Let n=Lie⁡(U)(Q)\mathfrak{n}=\operatorname{Lie}(U)(\mathbb{Q}). The exponential and logarithm identify NN with the finite-dimensional rational nilpotent Lie algebra n\mathfrak{n}, equipped with its Baker–Campbell–Hausdorff group law: X∗Y=log⁡(exp⁡Xexp⁡Y)=X+Y+12[X,Y]+112[X,[X,Y]]+112[Y,[Y,X]]+⋯X*Y=\log(\exp X\exp Y)=X+Y+\frac{1}{2}[X,Y]+\frac{1}{12}[X,[X,Y]]+\frac{1}{12}[Y,[Y,X]]+\cdots. The Mal’cev correspondence is functorial for abstract homomorphisms between uniquely divisible nilpotent groups; in particular,

α⟼(X⟼log⁡(α(exp⁡X)))\alpha\longmapsto\left(X\longmapsto\log(\alpha(\exp X))\right)

is an isomorphism from Aut⁡(N)\operatorname{Aut}(N) to Aut⁡Q(n)\operatorname{Aut}_{\mathbb{Q}}(\mathfrak{n}); see [3] for the group–Lie algebra correspondence and [16 Theorem 14.37] for its algebraic-group form. Thus no assumption of linearity of the original abstract automorphisms is being imposed.

The image of AA in Aut⁡Q(n)\operatorname{Aut}_{\mathbb{Q}}(\mathfrak{n}) is virtually nilpotent and has finitely many orbits. Proposition 4.3 therefore gives [n,n]=0[\mathfrak{n},\mathfrak{n}]=0. The Baker–Campbell–Hausdorff law reduces to addition, and hence N≅(Qd,+)N\cong(\mathbb{Q}^{d},+), where d=dim⁡Qn<∞d=\dim_{\mathbb{Q}}\mathfrak{n}<\infty. The existing splitting gives the required description of GG.

Finally, if G≤GL⁡n(K)G\leq\operatorname{GL}_{n}(K) for a number field KK, restriction of scalars embeds the same abstract group in GL⁡n[K:Q](Q)\operatorname{GL}_{n[K:\mathbb{Q}]}(\mathbb{Q}). The given abstract action of AA and its orbits are unchanged, so the rational case applies.

Remark 4.4. The essential restriction on the action is finite orbit count for a virtually nilpotent subgroup of the abstract automorphism group. Finiteness for the full automorphism group alone is insufficient. The passage from rational orbits to algebraic orbits above establishes one dense orbit, not a bijection of orbit spaces. For more general finitely generated fields, the proof does not immediately apply: a divisible unipotent subgroup can have infinite dimension

as a rational Mal’cev Lie algebra, and its abstract automorphisms need not be linear over the larger field.

Examples and limits of the hypotheses

The additional condition ω(G)<∞\omega(G)<\infty is not characterized by Theorem A; the theorem characterizes the strictly weaker condition ∣P(G)∣<∞|\mathcal{P}(G)|<\infty.

Example 5.1. The group

U(Q)={(1t01):t∈Q}U(\mathbb{Q})=\left\{\begin{pmatrix}1&t\\0&1\end{pmatrix}:t\in\mathbb{Q}\right\}

is isomorphic to (Q,+)(\mathbb{Q},+), is uniquely divisible, and has two automorphism orbits. All its power images coincide with the group. For the affine group G=Q⋊C2G=\mathbb{Q}\rtimes C_{2}, where the involution acts by inversion, the three automorphism orbits are the identity, the nonzero translations, and the involutions. Here e=2e=2 and

Pm(G)={G,m odd,Q,m even.P_m(G)= \begin{cases} G,&m\text{ odd},\\ \mathbb{Q},&m\text{ even}. \end{cases}

For example, in coordinates (ε,t)(\varepsilon,t) with ε∈{1,−1}\varepsilon\in\{1,-1\}, the automorphisms which scale translations by c∈Q×c\in\mathbb{Q}^{\times} and send a chosen involution rr to u(d)ru(d)r, d∈Qd\in\mathbb{Q}, act transitively on the last two sets separately.

In contrast, UT6(Q)\mathrm{UT}_{6}(\mathbb{Q}) is uniquely divisible, so it has just one power-image set, but has infinitely many automorphism orbits by [15 Theorem 1.2]. Thus finiteness of power images is strictly weaker than finiteness of automorphism orbits, even for rational unipotent groups.

Example 5.2 (The rational Heisenberg group). Let H=UT3(Q)\mathcal{H}=\mathrm{UT}_{3}(\mathbb{Q}), with symmetric coordinates

h(x,y,t)=(1xt+xy/201y001)(x,y,t∈Q).h(x,y,t)=\begin{pmatrix}1&x&t+xy/2\\0&1&y\\0&0&1\end{pmatrix}\qquad(x,y,t\in\mathbb{Q}).

Writing v=(x,y)Tv=(x,y)^{\mathsf{T}} and Ω(v,w)=xy′−yx′\Omega(v,w)=xy'-yx' for w=(x′,y′)Tw=(x',y')^{\mathsf{T}}, we have

h(v,t)h(w,s)=h(v+w,t+s+12Ω(v,w)).h(v,t)h(w,s)=h\left(v+w,t+s+\frac{1}{2}\Omega(v,w)\right).

The group commutator convention is [a,b]=aba−1b−1[a,b]=aba^{-1}b^{-1}. Thus

[h(v,t),h(w,s)]=h(0,Ω(v,w)),Z(H)=[H,H]={h(0,t):t∈Q}.[h(v,t),h(w,s)]=h(0,\Omega(v,w)),\qquad Z(\mathcal{H})=[\mathcal{H},\mathcal{H}]=\{h(0,t):t\in\mathbb{Q}\}.

In particular, H\mathcal{H} is nonabelian of class two. Since h(v,t)m=h(mv,mt)h(v,t)^m=h(mv,mt), it is uniquely divisible and has no proper finite-index subgroup by Lemma 4.1. It is therefore not virtually abelian.

For M∈GL2(Q)M\in\mathrm{GL}_{2}(\mathbb{Q}) and ℓ∈Hom⁡Q(Q2,Q)\ell\in\operatorname{Hom}_{\mathbb{Q}}(\mathbb{Q}^{2},\mathbb{Q}), ΦM,ℓ(h(v,t))=h(Mv,(det⁡M)t+ℓ(v))\Phi_{M,\ell}(h(v,t))=h(Mv,(\det M)t+\ell(v)) is an automorphism, since Ω(Mv,Mw)=det⁡(M)Ω(v,w)\Omega(Mv,Mw)=\det(M)\Omega(v,w). These maps are transitive on the nonidentity central elements, and also on the noncentral elements. For the latter assertion, first choose MM with Mv=wMv=w for prescribed nonzero v,wv,w, and then choose ℓ\ell with ℓ(v)=s−(det⁡M)t\ell(v)=s-(\det M)t. The identity and the center are characteristic, so ω(H)=3\omega(\mathcal{H})=3. This is the rational case of [15 Theorem 3.1]; the calculation is included to distinguish full automorphism orbits from the smaller acting group in the next proposition.

Proposition 5.3 (A metabelian action with five orbits). There is a metabelian subgroup A≤Aut⁡(H)A\leq\operatorname{Aut}(\mathcal{H}) such that ωA(H)=5\omega_A(\mathcal{H})=5. In particular, the conclusion of Theorem B fails if “virtually nilpotent” is replaced by “metabelian” or “soluble”.

Proof. For a,b∈Q×a,b\in\mathbb{Q}^{\times} and p,q∈Qp,q\in\mathbb{Q}, define

ϕa,b,p,q(h(x,y,t))=h(ax,by,abt+px+qy).\phi_{a,b,p,q}(h(x,y,t))=h(ax,by,abt+px+qy).

These are the automorphisms ΦM,ℓ\Phi_{M,\ell} above with M=diag⁡(a,b)M=\operatorname{diag}(a,b) and ℓ(x,y)=px+qy\ell(x,y)=px+qy. They form a group AA: composition is given by

ϕa,b,p,q∘ϕc,d,r,s=ϕac,bd,ar+pc,as+qd.\phi_{a,b,p,q}\circ\phi_{c,d,r,s}=\phi_{ac,bd,ar+pc,as+qd}.

The map ϕa,b,p,q↦(a,b)\phi_{a,b,p,q} \mapsto(a,b) is an epimorphism onto (Q×)2(\mathbb{Q}^{\times})^2, with abelian kernel

{ϕ1,1,p,q:p,q∈Q}≅(Q2,+).\{\phi_{1,1,p,q}:p,q\in\mathbb{Q}\}\cong(\mathbb{Q}^2,+).

Thus AA is metabelian.

The five invariant subsets and representatives are

conditionrepresentative
x=y=t=0x=y=t=0h(0,0,0)h(0,0,0)
x=y=0, t≠0x=y=0,\ t\ne0h(0,0,1)h(0,0,1)
x≠0, y=0x\ne0,\ y=0h(1,0,0)h(1,0,0)
x=0, y≠0x=0,\ y\ne0h(0,1,0)h(0,1,0)
x≠0, y≠0x\ne0,\ y\ne0h(1,1,0)h(1,1,0)

Table 5.3.

Each subset is one orbit. When (x,y)≠(0,0)(x,y)\ne(0,0), choose a,ba,b to normalize the nonzero coordinates and then choose p,qp,q to make abt+px+qy=0abt+px+qy=0. When x=y=0x=y=0 and t≠0t\ne0, choose ab=t−1ab=t^{-1}. The zero patterns cannot change, so these five orbits are distinct. Example 5.2 shows that HH is not virtually abelian, proving the asserted failure.

For completeness, AA is not virtually nilpotent. Let

D=ϕ2,1,0,0,Uq=ϕ1,1,0,q.D=\phi_{2,1,0,0},\qquad U_q=\phi_{1,1,0,q}.

Then DUqD−1=U2qDU_qD^{-1}=U_{2q}. Any finite-index subgroup A0≤AA_0\le A contains V={Uq:q∈Q}V=\{U_q:q\in\mathbb{Q}\}, because VV is divisible, and it contains DrD^r for some positive integer rr. Consequently

[Dr,Uq]=U(2r−1)q.[D^r,U_q]=U_{(2^r-1)q}.

Since multiplication by 2r−12^r-1 is surjective on Q\mathbb{Q}, every term of the lower central series of A0A_0 contains VV. Thus A0A_0 cannot be nilpotent. □

Corollary 5.4. If G≤GL⁡2(Q)G\le\operatorname{GL}_2(\mathbb{Q}) and P(G)\mathcal{P}(G) is finite, then GG is virtually abelian. For every n≥3n\ge3, there is a subgroup of GL⁡n(Q)\operatorname{GL}_n(\mathbb{Q}) with three automorphism orbits that is not virtually abelian.

Proof. Theorem A gives class⁡(N)≤1\operatorname{class}(N)\le1 in dimension two. For n≥3n\ge3, use HH and the block embedding g↦diag⁡(g,In−3)g\mapsto\operatorname{diag}(g,I_{n-3}), with the identity block omitted for n=3n=3. □

Remark 5.5 (Limits of the field statements). The equality G=H(k)G=H(k) cannot replace Theorem 1.2 over a general finitely generated field. Embed the group U(Q)U(\mathbb{Q}) from Example 5.1 in GL⁡2(Q(2))\operatorname{GL}_2(\mathbb{Q}(\sqrt{2})). Its Zariski closure is the full one-dimensional unipotent group, whose Q(2)\mathbb{Q}(\sqrt{2})-points are strictly larger than U(Q)U(\mathbb{Q}).

The finite-generation hypothesis on the field cannot be discarded from the forward implication of Theorem A: every invertible complex matrix has an mm-th root for every m≥1m\ge1, so Pm(GL⁡2(C))=GL⁡2(C)P_m(\operatorname{GL}_2(\mathbb{C}))=\operatorname{GL}_2(\mathbb{C}) for every mm. Nevertheless GL⁡2(C)\operatorname{GL}_2(\mathbb{C}) is not virtually soluble (its connected Zariski closure is not soluble). This example concerns the power-image hypothesis; it makes no claim about finite automorphism-orbit count for this group.

Acknowledgements

I am deeply grateful to my supervisor, Professor Hou Yong, for his continued support, many valuable discussions, and guidance throughout my studies. His suggestion to investigate virtual abelianness inspired Theorem B. He also encouraged me to look beyond the resolution of an individual problem and seek a deeper understanding of the underlying structures; this encouragement led me to develop the present work beyond an affirmative answer to Problem 21.40 of the Kourovka Notebook.

This paper is dedicated to the memory of my grandfather, Feng Bengui.

Declaration of AI Tool Use

The main mathematical work presented in this paper was completed in July 2026.In April, Gemini 3 Pro suggested the example presented in Example 5.1. After the manuscript was completed, ChatGPT was used to assist in verifying references and related literature. AI tools were also employed to review the details of the proofs and to check grammar.

The main mathematical results, arguments, and exposition are the author’s own work. The entire manuscript was written and typeset by the author, rather than generated by AI. All AI-assisted suggestions were independently examined, and the author takes full responsibility for the mathematical correctness, originality, and final content of this paper.

References

DEPARTMENT OF MATHEMATICS, SCHOOL OF SCIENCE AND ENGINEERING, THE CHINESE UNIVERSITY OF HONG KONG, SHENZHEN, GUANGDONG, CHINA

Email address: [email protected]

References

  1. [1]A. Jayadevan, Bounded-index nilpotent subgroups of rational linear groups with finitely many automorphism orbits, preprint (2026). https://kouroukanotebookorg.wordpress.com/wp-content/uploads/2026/09/kourovka-21-40-paper.pdf.DOI
  2. [2]Codex gpt-6-astra, M. Kinyon and J. Urban, Results on problems from the Kourouva Notebook, preprint, mathematical edition, version 4, 19 September 2026. https://kouroukanotebookorg.wordpress.com/wp-content/uploads/2026/09/kourovka-mathematics.pdf.
  3. [3]B. Assmann, Algorithmic use of the Mal’cev correspondence, in Groups St Andrews 2005, Vol. 1, London Math. Soc. Lecture Note Ser., vol. 339, Cambridge University Press, Cambridge, 2007, pp. 158–169. doi:10.1017/CBO9780511721212.007.
  4. [4]R. Bastos, A. C. Dantas and E. de Melo, Soluble groups with few orbits under automorphisms, Geom. Dedicata 209 (2020), 119–123. doi:10.1007/s10711-020-00525-7.DOI
  5. [5]R. Bastos, A. C. Dantas and E. de Melo, Virtually nilpotent groups with finitely many orbits under automorphisms, Arch. Math. (Basel) 116 (2021), 261–270. doi:10.1007/s00013-020-01566-w.DOI
  6. [6]O. Baues and F. Grunewald, Automorphism groups of polycyclic-by-finite groups and arithmetic groups, Publ. Math. Inst. Hautes Études Sci. 104 (2006), 213–268. doi:10.1007/s10240-006-0003-3.DOI
  7. [7]A. C. Dantas, E. de Melo and J. Kato, On soluble groups of finite rank with finitely many automorphism orbits, Comm. Algebra 54 (2026), no. 6, 2341–2347. doi:10.1080/00927872.2025.2578695.DOI
  8. [8]A. C. Dantas and J. F. de Sousa, Some families of locally graded groups with finitely many orbits under automorphisms, preprint (2026), arXiv:2605.02090v1.arxiv.org/abs/2605.02090
  9. [9]E. Hrushovski, P. H. Kropholler, A. Lubotzky and A. Shalev, Powers in finitely generated groups, Trans. Amer. Math. Soc. 348 (1996), no. 1, 291–304. doi:10.1090/S0002-9947-96-01456-0.DOI
  10. [10]J. E. Humphreys, Linear Algebraic Groups, Graduate Texts in Mathematics, vol. 21, Springer, New York, 1975.DOI
  11. [11]E. Jabara, Actions of abelian groups on groups, J. Group Theory 10 (2007), no. 2, 185–194. doi:10.1515/JGT.2007.014.DOI
  12. [12]E. Jabara, Abelian automorphisms groups with a finite number of orbits, J. Algebra 323 (2010), no. 10, 2798–2817. doi:10.1016/j.jalgebra.2010.03.013.DOI
  13. [13]E. I. Khukhro and V. D. Mazurov (eds.), Unsolved Problems in Group Theory: The Kourovka Notebook, 21st ed., 2026, Problem 21.40; arXiv:1401.0300v45, version dated 3 July 2026.
  14. [14]A. Lubotzky and C. Meiri, Sieve methods in group theory I: Powers in linear groups, J. Amer. Math. Soc. 25 (2012), no. 4, 1119–1148. doi:10.1090/S0894-0347-2012-00736-X.DOI
  15. [15]E. de Melo and J. Kato, Automorphism orbits of the group of unitriangular matrices, preprint (2025), arXiv:2510.09353v1.arxiv.org/abs/2510.09353
  16. [16]J. S. Milne, Algebraic Groups: The Theory of Group Schemes of Finite Type over a Field, Cambridge Studies in Advanced Mathematics, vol. 170, Cambridge University Press, Cambridge, 2017.
  17. [17]J. Neukirch, Algebraic Number Theory, translated by N. Schappacher, Grundlehren der mathematischen Wissenschaften, vol. 322, Springer, Berlin, 1999.

Paper details

vlnc:W_768yrau5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Gemini 3 Pro, ChatGPT, AI tools

View statement
Looking for related papers…
Contents