This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.
Introduction
Let Mn(ℝ) be the set of all square matrices over ℝ of order n ≥ 1 and by GL(n, ℝ) the group of invertible matrices of Mnℝ. Let G be a finitely generated abelian sub-semigroup of Mn(ℝ). By a sub-semigroup of Mn(ℝ), we mean a subset which is stable under multiplication and contains the identity matrix. For a vector v ∈ ℂn, we consider the orbit of G through v: G(v) = {Av: A ∈ G}⊂ ℝn. A subset E ⊂ ℝn is called G-invariant if A(E) ⊂ E for any A ∈ G. The orbit G(v)⊂ ℝn is dense in ℝn if $\overline{G(v)}= {\mathbb{R}}^{n}$ , where Ē denotes the closure of a subset E⊂ ℝn. The semigroup G is called hypercyclic if there exists a vector v ∈ℝn such that G(v) is dense in ℝn. We refer the reader to the recent books [4] and [9] for a thorough account on hypercyclicity. In [7], [8], Costakis and Manoussos localized the notion of hypercyclicity through the use of the J-sets, firstly for a single bounded linear operator T acting on a complex Banach space X, and secondly extended to that of semigroup. Recall that T is called a J-class operator if for every non-zero vector x in X and for every open neighborhood U ⊂ X of x and every non-empty open set V ⊂ X, there exists a positive integer n such that Tn(U) ∩ V = ∅. Recently, there are much research that study or use J-class operators, We mention for instance, the series of papers by Azimi and Müller [3], Nasseri [10], Chan and Seceleanu [5]. Among properties, there are locally hypercyclic, non hypercyclic operators and that finite dimensional Banach spaces do not admit locally hypercyclic operators (see [7]).
In [8], Costakis and Manoussos extend the notion of J-class operator to that of a J-class of semigroup G as follows: Suppose that G is generated by p matrices A1,…, Ap(p ≥ 1) then we define the extended limit set JG(x) of x under G to be the set of y ∈ ℝn for which there exists a sequence (xm)m ⊂ ℝn and sequences of non-negative integers $\{k^{(j)}_{m}:\ m\in \mathbb{N}\}$ for j = 1, 2,…, p with $k^{(1)}_{m}+k^{(2)}_{m}+\dots + k^{(p)}_{m}\to+\infty$ such that xm → x and $A_{1}^{k^{(1)}_{m}}A_{2}^{k^{(2)}_{m}}\dots A_{p}^{k^{(p)}_{m}}x_{m}\to y$ .
This describes the asymptotic behavior of the orbits of nearby points to x. Note that the condition $\ k^{(1)}_{m} + k^{(2)}_{m}+\dots+ k^{(p)}_{m}\to+\infty$ is equivalent to having at least one of the sequences $\{k^{(j)}_{m}:\ m\in\mathbb{N}\}$ for j = 1, 2,…, p containing an increasing subsequence tending to +∞. We say that G is locally hypercyclic (or J-class) if there exists a vector v ∈ ℝn\{0} such that JG(v) = ℝn. This notion is a “localization” of the concept of hypercyclicity, this can be justified by the following: JG(x) = ℝn if and only if for every open neighborhood Ux ⊂ ℝn of x and every nonempty open set V ⊂ ℝn there exists A ∈ G such that A(Ux)∩ V≠∅.
As we have mentioned above, in ℂn or ℝn, no matrix can be locally hypercyclic. However, what is rather remarkable is that in ℂn or ℝn, a pair of commuting matrices exists which forms a locally hypercyclic, non-hypercyclic semigroup (see [8]).
In the present work, we show that G is hypercyclic if and only if there exists a vector v in an open set V, defined according to the structure of G, such that JG(v) = ℝn (Theorem 1). This answer the question 1 raised by the author in [2]. Furthermore, we construct for every n ≥ 2, a locally hypercyclic abelian semigroup G generated by matrices A1,…, An+1 which is non-hypercyclic whenever JG(uk) = ℝn, k = 1,…, n, for a basis (u1,…, un) of ℝn (Theorem 4), this answers negatively the question raised by Costakis and Manoussos in [8]. However, we prove that the question is true (see Proposition 5) for any abelian semigroup G consisting of lower triangular matrices on ℝn with all diagonal elements equal.
Before stating our main results, let introduce the following notations and definitions.
Set ℕ be the set of non negative integers and write ℕ0 = ℕ\{0}. Let n ∈ ℕ0 be fixed. Denote by:
ℬ; = (e1,..., e1) the canonical basis of ℝn.
In the identity matrix on ℝn.
For each m = 1, 2,…, n, denote by:
$\mathbb{T}_{m}(\mathbb{R})$ the set of matrices over ℝ of the form:
$\mathscr{K}^{*}_{\eta}(\mathbb{R}): = \mathscr{K}_{\eta}(\mathbb{R})\cap \textrm{GL}(n, \ \mathbb{R})$ , it is a sub-semigroup of GL(n, ℝ). In particular:
If r = 1, s = 0 then $\mathscr{K}_{\eta}(\mathbb{R}) = \mathbb{T}_{n}(\mathbb{R})$ and η = (n).
If r = 0, s = 1 then $\mathscr{K}_{\eta}(\mathbb{R}) = \mathbb{B}_{m}(\mathbb{R})$ and η = (m), n = 2m.
If r = 0, s = 1 then $\mathscr{K}_{\eta}(\mathbb{R}) = \mathbb{B}_{m_{1}}(\mathbb{R})\oplus\dots\oplus\mathbb{B}_{m_{s}}(\mathbb{R})$ and η = (m1,…, ms).
For a row vector v ∈ ℝn, we will be denoting by vT the transpose of v.
uη = [eη,1,…,eη,r; fη,1,…,fη,s]T ∈ ℝn, where for every k = 1,…, r; l = 1,…, s, eη,k = [1,0,…, 0]T ∈ ℝnk, $f_{\eta,l} = [1,0,\dots, 0]^{T}\in \mathbb{R}^{2m_{l}}$ .
Let G be an abelian sub-semigroup of Mn(ℝ), we have the following “normal form of G”:
There exists a P ∈ GL(n, ℝ) such that $P^{-1}GP\subset \mathscr{K}_{\eta}(\mathbb{R})$ for some partition η of n (see Proposition 6).
Given two integers r, s ∈ ℕ, we shall say that the semigroup G has “a normal form of length r + 2s” if G has a normal form in 𝒦η(ℝ) for some partition η with length r + 2s. For such a choice of matrix P, we let:
vη = Puη.
V = P(U), it is a dense open set in ℝn.
Our principal results are the following:
Theorem 1
LetG be a finitely generated abelian semigroup of matrices on ℝn. If JG(v) = ℝnfor somev ∈ Vthen $\overline{G(v_{\eta})} = \mathbb{R}^{n}$ .
Corollary 2
Under the hypothesis of Theorem 1, the following are equivalent:
G is hypercyclic.
JG(vη) = ℝn.
$\overline{G(v_{\eta})} = \mathbb{R}^{n}$ .
Corollary 3
Under the hypothesis of Theorem 1, assume that G is not hypercyclic. Then
, where Hk (resp. Fl) are G-invariant vector subspaces of ℝnof dimension n − 1 (resp. n − 2).
Remark
If n = 1, then vη = 1 and by Corollary 2, a sub-semigroup G of ℝ is hypercylic if and only if it is locally hypercyclic.
Theorem 4
Let n ≥ 2 be an integer. Then there exists an abelian semigroup G generated by diagonal matricesA1, …,An+1 ∈ GL(n,ℝ) which is not hypercyclic such that JG(ek) = for every k = 1,…, n.
Proposition 5
Let G be a finitely generated abelian sub-semigroup of $\mathbb{T}_{n}(\mathbb{R})$ . If there exists a basis (e′1,…,e′nof ℝnsuch that JG(e′k) = ℝnfor everyk = 1,…, n, thenGis hypercyclic.
Notation and some results
Recall first the following proposition.
Proposition 6
([1], Proposition 2.2) Let G be an abelian sub-semigroup of Mn(ℝ). Then there exists a P ∈ GL(n, ℝ) such thatP−1GPis an abelian sub-semigroup of𝒦η(ℝ), for some partitionηofn.
Notation.
If G is a sub-semigroup of $\mathbb{T}_{n}(\mathbb{K})$ $(\mathbb{K} = \mathbb{R}$ or ℂ, denote by
$$F_{G} = \textrm{vect}\big(\{(B - \mu_{B}I_{n})e_{i}\in \mathbb{K}^{n}: \ 1\leq i\leq n-1, \ B\in G\}\big)$$
the vector subspace generated by the family of vectors {(B − μBIn)ei ∈ K;n : 1 ≤ i ≤ n − 1 , B ∈ G}.
- rank(FG) the rank of FG. We have rank(FG) ≤ n − 1.
If G is an abelian sub-semigroup of $\mathbb{B}_{m}(\mathbb{R})$ , (n = 2m), denote by:
If G is an abelian sub-semigroup of 𝒦n(ℝ), then for every B ∈ 𝒦η(ℝ), we have $B = \textrm{diag}(B_{1},\dots, B_{r}; \ \widetilde{B_{1}},\dots,\widetilde{B_{s}})$ with $B_{k}\in\mathbb{T}_{n_{k}}(\mathbb{R})$ , $\widetilde{B_{l}}\in\mathbb{B}_{m_{l}}(\mathbb{R})$ , k = 1,…, r, l = 1,…, s. Denote by:
Gk = {Bk : B ∈ G}, k = 1,…, r, it is an abelian sub-semigroup of $\mathbb{T}_{n_{k}}(\mathbb{R})$ .
$\widetilde{G}_{l} = \{\widetilde{B_{l}}\ : \ B\in G\}$ , l = 1,…, s, it is an abelian sub-semigroup of $\mathbb{B}_{m_{l}}(\mathbb{R})$ .
For every $x = [x_{1},\dots,x_{r}; \widetilde{x_{1}},\dots,\widetilde{x_{s}}]^{T}\in \mathbb{R}^{n}$ , where $x_{k} = [a_{k,1},\dots, a_{k,n_{k}}]^{T}\in\mathbb{R}^{*}\times\mathbb{R}^{n_{k}-1}$ , $\widetilde{x_{l}} = [b_{l,1},\dots, b_{l,m_{l}}]^{T}\in(\mathbb{R}^{2}\backslash\{(0,0)\})\times\mathbb{R}^{2m_{l}-2}$ , we let:
$H_{x_{k}} = \mathbb{R}x_{k}+F_{G_{k}}$ , k = 1,…, r.
$\widetilde{H}_{\widetilde{x_{l}}} = \mathcal{C}\widetilde{x_{l}} + F_{\widetilde{G}_{l}}$ , l = 1,…, s, where $\mathcal{C}= \{\textrm{diag}(C,\dots, C): \ C\in \mathbb{S}\}$ .
LetGbe an abelian sub-semigroup of 𝒦η(ℝ). Under the notations above, for everyx ∈ ℝn, Hx is G-invariant.
Proof
It suffices to prove that $H_{x_{k}}$ is Gk-invariant (resp. $\widetilde{H}_{\widetilde{x_{l}}}$ is $\widetilde{G}_{l}$ -invariant): write for short G = Gk and x = xk (resp. $G = \widetilde{G}_{l}$ and $x = \widetilde{x_{l}})$ . One has Hx = ℝx (resp. $\widetilde{H}_{x} = \mathcal{C}x + F_{G})$ .
- If w = [w1,…, wn]T ∈ Hx and B ∈ G with eigenvalue μ ∈ ℝ, then $Bw = \mu w+(B-\mu I_{n})w = \mu w+\underset{i=1}{\overset{n-1}{\sum}}w_{k}(B-\mu I_{n})e_{i}\in H_{x}$ (since w, (B−μ In)ei ∈ Hx and Hx is a vector space).
- If $\widetilde{w} = [\widetilde{w_{1}},\dots,\widetilde{w_{s}}]^{T}\in \widetilde{H}_{x}$ and B ∈ G, then we have $B\widetilde{w} = C\widetilde{w}+ (B- C I_{n})\widetilde{w} = C \widetilde{w} + \underset{i=1}{\overset{n-1}{\sum}}\widetilde{w_{k}}(B- C I_{n})e_{i}\in \widetilde{H}_{x}$ (since $C\widetilde{w}, \ (B-C I_{n})e_{i}\in \widetilde{H}_{x}$ and $\widetilde{H}_{x}$ is a vector space).
Proposition 8
Let G be an abelian sub-semigroup of 𝒦η(ℝ). If JG(u) = ℝnfor someu ∈ U, then rank(FGk) = nk − 1 and $\mathrm{rank}(F_{\widetilde{G}_{l}}) = 2m_{l}-2$ , for everyk = 1,…, r; l = 1,…, s.
Proof
Let $u = [u_{1},\dots,u_{r}; \ \widetilde{u_{1}},\dots, \widetilde{u_{s}}]^{T}\in \mathbb{R}^{n}$ , where $u_{k}\in \mathbb{R}^{n_{k}}$ , $\widetilde{u_{l}}\in \mathbb{R}^{2m_{l}}$ , for every k = 1,…, r; l = 1,…, s.
i) First, we will show that $J_{G_{k}}(u_{k}) = \mathbb{R}^{n_{k}}$ and $J_{\widetilde{G}_{l}}(\widetilde{u_{l}}) = \mathbb{R}^{2m_{l}}$ . For this, let $x_{k}\in \mathbb{R}^{n_{k}}$ , $\widetilde{x_{l}}\in \mathbb{R}^{2m_{l}}$ and set $y = [y_{1},\dots,y_{r};\widetilde{y_{1}},\dots,\widetilde{y_{s}}]^{T}\in \mathbb{R}^{n}$ such that
Therefore xk ∈ JGk(uk) and $\widetilde{x_{l}}\in J_{\widetilde{G}_{l}}(\widetilde{u_{l}})$ . It follows that JGk(uk) = ℝnk and $J_{\widetilde{G}_{l}}(\widetilde{u_{l}})= \mathbb{R}^{2m_{l}}$.
ii) Second, one can then suppose that G ⊂ 𝕋n(ℝ) and u ∈ ℝ* × ℝn – 1 (resp. G ⊂ 𝔹m(ℝ) and u ∈ (ℝ2\{(0, 0)}) × ℝ2m – 2. It is clear that u ∉ FG.
Assume that G ⊂ 𝕋n(ℝ). By Lemma 7, He1 = ℝe1 + FG is G-invariant.
Suppose that ℝn\He1 ≠ Ø, so there exist y ∈ ℝn\He1 and two sequences (xm)m ⊂ ℝn and (Bm)m ⊂ D such that $\underset{m\to +\infty}\lim x_{m}=e_{1}$ and $\underset{m\to +\infty}\lim B_{m}x_{m}=y$ . Let Hxm = ℝxm + FG for every m ∈ ℕ. By Lemma 7, Hxm is G-invariant, so Bmxm, for every m ∈ ℕ. Write Bmxm = αmxm + zmαm ∈ ℝ and zm ∈ FG. We distinguish two cases:
If (αm)m is bounded, one can suppose by passing to a subsequence, that (αm)αm ≥ 1 is convergent, say $\underset{m\to +\infty}\lim \alpha_{m} = a\in\mathbb{R}$ . It follows that $\underset{m\to +\infty}\lim z_{m} = y-au\in F_{G}$ and so y ∈ He1, a contradiction.
If (αm)m is not bounded, one can suppose by passing to a subsequence, that $\underset{m\to +\infty}\lim|\alpha_{m}| = +\infty$ , then $\underset{m\to +\infty}\lim \frac{1}{\alpha_{m}}z_{m} = -u\in F_{G}$ , a contradiction. We conclude that He1 = ℝn and so dim (FG) = n – 1.
Assume that G ⊂ 𝔹m(ℝ) (n = 2m). By Lemma 7, He1 = 𝒞e1 + FG is G-invariant. Suppose that ℝn\He1 ≠ Ø and let y ∈ ℝn\He1. Then there exist two sequences (xk)k ⊂ ℝn and (Bk)k ⊂ G such that $\underset{k\to +\infty}\lim x_{k} = e_{1}$ and $\underset{k\to +\infty}\lim B_{k}x_{k}=y$ . Let Hxk = 𝒞xk + FG for every k ∈ ℕ. By Lemma 7, Hxk is G-invariant, so Bkxk ∈ Hxk, for every k ∈ ℕ. Write Bkxk = Ckxk + zk, with Ck = diag (Rk,..., Rk), $R_{k}=\begin{bmatrix} \alpha_{k} & \beta_{k} \\ -\beta_{k} & \alpha_{k} \\ \end{bmatrix},$ , and zk ∈ FG. Take ||Ck|| = ||Rk|| = sup(|αk|, |βk|). We distinguish two cases:
If (||Ck||)k is bounded, one can suppose by passing to a subsequence, that (αk)k and (βk)k converge, say $\underset{k\to +\infty}\lim \alpha_{k} = \alpha\in\mathbb{R}$ and $\underset{k\to +\infty}\lim \beta_{k} = \beta\in\mathbb{R}$ . As $\underset{k\to +\infty}\lim C_{k}e_{1}= \alpha e_{1}-\beta e_{2}=Ce_{1}$ , where C = diag(R,..., R) with
$R=\begin{bmatrix} \alpha& \beta \\ -\beta & \alpha \\ \end{bmatrix},$ and as Bkxk = Ckxk = Ck(xk – e1) + Cke1 + zk, then $\underset{k\to +\infty}\lim z_{k} = y-Ce_{1}\in F_{G}$ and therefore y ∈ Ce1 + FG ⊂ He1, a contradiction.
If (||Ck||)k is not bounded, one can suppose by passing to a subsequence, that $\underset{k\to +\infty}\lim\|C_{k}\| = +\infty$ , then Ck is invertible for k large, so from Bkxk = Ckxk = Ck(xk – e1) + Cke1 + zk, we have $ \frac{1}{\|C_{k}\|}C_{k}e_{1}+ \underset{k\to +\infty}\lim \frac{1}{\|C_{k}\|}z_{k} = 0$ . In particular, we get $\underset{k\to +\infty}\lim \frac{\alpha_{k}}{\|C_{k}\|} = \underset{k\to +\infty}\lim \frac{\beta_{k}}{\|C_{k}\|}=0$ , this is a contradiction with ||Ck|| (|αk|, |βk|).
We conclude that He1 = ℝn and so dim (FG) = n – 1.
LetGbe an abelian sub-semigroup of 𝕋n(𝕂) (𝕂= ℝ or 𝕔) such that rank (FG) = n – 1. Let u, ∈ 𝕂* × 𝕂n – 1. If two sequences (um}m∈ℕin 𝕂* × 𝕂n – 1 and (Bm)m∈ℕinGsuch that $\underset{m\to+\infty}\lim u_{m} = u$ and $\underset{m\to+\infty}\lim B_{m}u_{m} = v$ then (Bm)m∈ℕis bounded.
Proof
The proof is down in ([2], Lemma 3.4) if $\mathbb{K}= \mathbb{C}$ and the same proof works if $\mathbb{K}= \mathbb{R}$ .
Now consider the following basis change:
Assume that n = 2m, m ∈ ℕ0. For every k = 1, …, m, we let:
$e^{\prime}_{k} = \frac{e_{2k-1}-ie_{2k}}{2}$ and $\mathscr{C}_{0} = (e^{\prime}_{1},\dots,e^{\prime}_{m}, \ \overline{e^{\prime}_{1}},\dots,\overline{e^{\prime}_{m}})$ , where $\overline{u}=[\overline{z_{1}},\dots,\overline{z_{m}}]^{T}$ is the conjugate of u = [z1,…, zm]T. Then 𝒞0 is a basis of ℂ2m. Denote by Q ∈ GL(2m, ℂ) the matrix of basis change from ℬ0 to 𝒞0. Then a simple calculation yields that:
Lemma 10
Under the notation above, for every $B\in\mathbb{B}_{m}(\mathbb{R})$ , $Q^{-1}BQ = \mathrm{diag}(B^{\prime}_{1},\ \overline{B^{\prime}_{1}})$ , where $B^{\prime}_{1}\in \mathbb{T}_{m}(\mathbb{C})$ .
Lemma 11
Let G be an abelian sub-semigroup of $\mathbb{B}_{m}(\mathbb{R})$ such that rank(FG) = 2m − 2. Letv ∈ (ℝ2\{(0,0)}) × ℝ2m−2. If two sequences (um)m∈ℕ ⊂ (ℝ2\{(0,0)}) × ℝ2m−2and (Bm)m∈ℕ ⊂ Gsuch that $\underset{m\to+\infty}\lim u_{m} = u$ and $\underset{m\to+\infty}\lim B_{m}u_{m} = v$ then (Bm)m∈ℕis bounded.
Proof
For B ∈ G, we have $Q^{-1}BQ = \textrm{diag}(B^{\prime}_{1},\ \overline{B^{\prime}_{1}})$ , where $B^{\prime}_{1}\in \mathbb{T}_{m}(\mathbb{C})$ . Write G′1 = {B′1 : B ∈ G} and $\overline{G^{\prime}_{1}} = \{\overline{B^{\prime}_{1}}: B\in G\}$ . Then G′1 (resp. $\overline{G^{\prime}_{1}}$ ) is an abelian sub-semigroup of $\mathbb{T}_{m}(\mathbb{C})$ .
First we prove that $\textrm{rank}(F_{G^{\prime}_{1}}) = m-1$ . Write C = diag(R,…,R) with $R = \begin{bmatrix} \alpha& \beta \\ -\beta & \alpha \\ \end{bmatrix}$ and μ = α + iβ. One has
As rank(FG) = 2m−2 then rank (Q−1(FG))= 2m−2 = 2rank(FG′1). So rank(FG′1) = m − 1.
Second, we let u′m = Q-1um = [u′m,1, u′m,2]T], u′ = Q-1u = [u′1, u′2]T and v′ = Q-1v = [v′1,v′2]T, where u′m,i, u′i, v′i ∈ ℂm, i = 1, 2. As u, v, um ∈ (ℝ2\{(0,0)}) × ℝ2m−2 then u′1, v′1, u′m,1 ∈ ℂ* × ℂm−1. We have $\underset{m\to+\infty}\lim u_{m}^{\prime} = u^{\prime}$ and so $\underset{m\to+\infty}\lim u_{m,1}^{\prime} = u_{1}^{\prime}$ . Take B′m = Q-1BmQ. By Lemma 10, $B_{m}^{\prime} = \textrm{diag}(B_{m,1}^{\prime},\ \overline{B_{m,1}^{\prime}})$ , where $B_{m,1}^{\prime}\in \mathbb{T}_{m}(\mathbb{C})$ . Then B′m,1 ∈ G′1 and $\underset{m\to+\infty}\lim B_{m}^{\prime}u_{m}^{\prime} = v^{\prime}$ . So $\underset{m\to+\infty}\lim B_{m,1}^{\prime}u_{m,1}^{\prime} = v^{\prime}_{1}$ . By Lemma 9, (Bm,1)m∈ℕ is bounded, so is (B′m)m∈ℕ. We conclude that (Bm)m∈ℕ bounded.
It follows from Lemmas 9 and 11, the following:
Corollary 12
Let G be a finitely generated abelian sub-semigroup of 𝒦η(ℝ). Suppose that rank(FGk) = nk−1, $\mathrm{rank}(F_{\widetilde{G}_{l}})= 2m_{l}-2$ , k = 1,…, r; l = 1,…, s. Ifx, y ∈ Uand two sequences (Bm)m∈ℕ ⊂ Gand (xm)m∈ℕ ⊂ ℝnsuch that $\underset{m\to+\infty}\lim x_{m}=x$ and $\underset{m\to+\infty}\lim B_{m}x_{m} = y$ then (Bm)m∈ℕis bounded.
Proposition
Let G be a finitely generated abelian sub-semigroup ofMn(ℝ). Then G is hypercyclic if and only ifJG(x) = ℝnfor everyx∈ℝn.
Proof
[Proof of Theorem 1] One can assume, by Proposition 6, that G is a sub-semigroup of 𝒦η(ℝ). Suppose that JG(u) = ℝn. Then by Proposition 8, rank(FGk) = nk−1 and $\mathrm{rank}(F_{\widetilde{G}_{l}})= 2m_{l}-2$ , for every k = 1,…, r; l = 1,…, s. Let y ∈ U, then there exist two sequences (Bm)m∈ℕ ⊂ G and (xm)m∈ℕ ⊂ ℝn satisfying:
Thus $\underset{m\to+\infty}\lim B_{m}u=y$ and so $y\in \overline{G(u)}$ . It follows that $U\subset\overline{G(u)}$ . Since $\overline{U} = \mathbb{R}^{n}$ , we get $\overline{G(u)} = \mathbb{R}^{n}$ . □
Proof of Corollary 2. (i)⟹(ii) follows from Proposition 13.
(ii)⟹(iii) results from Theorem 1. (iii)⟹(i) is clear.
JG(v) ≠ ℝn for any v ∈ V. Thus V ∩ E = ∅ and therefore $E\subset\mathbb{R}^{n}\backslash V = P(\underset{k=1}{\overset{r}{\bigcup}}H_{k} \cup \underset{l=1}{\overset{s}{\bigcup}}F_{l})$ , where
[6], Lemma 2.6]. Leta, b ∈ ℝ with −1 < a < 0, b > 1 and $\dfrac{\log|a|}{\log b}$ is irrational. Then the set {akbl : k, l ∈ ℕ} is dense in ℝ.
Proof
[Proof of Theorem 4) Consider the abelian sub-semigroup G of GL(n, ℝ) generated by B, A1,…, An, where B = bIn and $A_{k} = \textrm{diag}(\underset{(k-1)-terms}{\underbrace{1,\dots\dots,1}},a,1\dots,1),\ k=1,\dots, n$ . Then G is a sub-semigroup of $\mathscr{K}^{*}_{\eta}(\mathbb{R})$ with r = n and η = (1,…, 1). One has uη = [1,…, 1]T.
First, we will show that G is not hypercyclic: for this, it is equivalent to prove, by Corollary 2, that $\overline{G(u_{\eta})}\neq \mathbb{R}^{n}$ : We have
Observe that for every x = [x1,…, xn]T ∈ G(uη), we have $\dfrac{x_{1}}{x_{2}}= a^{k_{1}-k_{2}}$ . Since the set $\left\{a^{p} :\ \ p\in \mathbb{Z}\right\}$ is not dense in ℝ, the orbit G(uη) cannot be dense in ℝn.
Second, we will show that JG(e1) = ℝn (the other ek work in the same way). Fix a vector y = [y1, …, yn]T ∈ ℝn such that y1 ≠ 0. By Lemma 14, choose two sequences of positive integers (im)m ∈ℕ and (jm)m∈ℕ with im, jm → +∞ such that $\underset{m\to +\infty}\lim a^{i_{m}}b^{j_{m}}= y_{1}$ . Let $x^{(m)} = (1, x^{(m)}_{2},\dots, x^{(m)}_{n})$ with $x^{(m)}_{k} = \frac{y_{k}}{a^{i_{m}}b_{j_{m}}a^{m}}$ , k = 2,…, n. Since −1 < a < 0 and y1 ≠ 0, we see that $\underset{m\to+\infty}{\lim}x^{(m)}=e_{1}$ . On the other hand, consider $B_{m}: = B^{j_{m}}A_{1}^{i_{m}}A_{2}^{i_{m}+m}A_{3}^{i_{m}+m}\dots A_{n}^{i_{m}+m}$ . Then $B_{m}x^{(m)} = \left[a^{i_{m}}b^{j_{m}},y_{2},\dots,y_{n}\right]^{T}$ and so, $\underset{m\to+\infty}{\lim}B_{m}x^{(m)}=y$ . We conclude that y ∈ JG(e1) and therefore JG(e1) = ℝn.
Proof of Proposition 5. Since (e′1,…,e′n) is a basis of ℝn, there exists i0 ∈ {1,…, n} such that
ℝ* × ℝn−1. As V = U = ℝ* × ℝn−1 and ${\rm J}_{G}(e^{\prime}_{i_{0}}) = \mathbb{R}^{n}$ then by Theorem 1, $\overline{G(e^{\prime}_{i_{0}})} = \mathbb{R}^{n}$ and hence G is hypercyclic.