Independence, infinite dimension, and operators

Nizar El Idrissi and Samir Kabbaj

June 19, 2023

Abstract
In [Appl. Comput. Harmon. Anal., 46(3):664–673, 2019] O. Christensen and M.
Hasannasab observed that assuming the existence of an operator T sending en to
en+1 for all n ∈ N (where (en )n∈N is a sequence of vectors) guarantees that (en )n∈N
is linearly independent if and only if dim(span{en }n∈N ) = ∞. In this article, we
recover this result as a particular case of a general order-theory-based model-theoretic
result. We then return to the context of vector spaces to show that, if we want to
use a condition like T (ei ) = eϕ(i) for all i ∈ I where I is countable as a replacement
of the previous one, the conclusion will only stay true if ϕ : I → I is conjugate to
the successor function succ : n 7→ n + 1 defined on N. We finally prove a tentative
generalization of the result, where we replace the condition T (ei ) = eϕ(i) for all i ∈ I
where ϕ is conjugate to the successor function with a more sophisticated one, and to
which we have not managed to find a new application yet.

Contents
1 Introduction 1

2 Notations 2

3 Lemmas in order theory 3

4 Application to model theory 4

5 The only possible countable extensions 5

6 A tentative generalization 7
2020 Mathematics Subject Classification. 15A03; 15A04; 06A12; 03C07.
Key words and phrases. vector space, operator, linear independence, dimension, ordered structures,
σ-structure.

1
1 Introduction
Linear algebra is an entrenched subject of mathematics that started with the intro-
duction of coordinates in geometry by René Descartes. Its modern theory emerged in
the late nineteenth century after Peano gave the definition of a vector space. This theory
makes heavy use ot the concepts of linear independence and dimension, which often allow
to state important theorems and conjectures.
Generally, linear independence of an infinite sequence implies that it spans an infinite-
dimensional space, but not the opposite. As a result, it is interesting to consider the
conditions of a reverse statement. Such reverse statements may allow to solve standing
problems on linear independence.
In [1], O. Christensen and M. Hasannasab observed that assuming the existence of
an operator T sending en to en+1 for all n ∈ N (where (en )n∈N is a sequence of vectors)
guarantees that (en )n∈N is linearly independent if and only if dim{en }n∈N = ∞. To wit:

Proposition 1.1. (O. Christensen and M. Hasannasab)

Let E be a vector space and (en )n∈N a family in E indexed by N. Then

(∃T ∈ L(span{en }n∈N , E) : ∀n ∈ N : T (en ) = en+1 ) and dim span{en }n∈N = +∞] ⇒

(en )n∈N is free.

In this article, we prove some additional results related to proposition 1.1. First,
we recover proposition 1.1 as a particular case of a general order-theory-based
model-theoretic result. We then return to the context of vector spaces to show
that, if we want to use a condition like T (ei ) = eϕ(i) for all i ∈ I where I is
countable as a replacement of the previous one, the conclusion will only stay true
if ϕ : I → I is conjugate to the successor function succ : n 7→ n + 1 defined on
N. We finally prove a tentative generalization of the result, where we replace the
condition T (ei ) = eϕ(i) for all i ∈ I where ϕ is conjugate to the successor function with
a more sophisticated one, and to which we have not managed to find a new application yet.

Plan of the article. We dedicate section 2 to the notations. We then prove

in section 3 some order-theoretic lemmas. These lemmas will allow us to prove in the next
section 4 a model-theoretic result and recover proposition 1.1 as a particular case. We
then return in section 5 to the context of vector spaces and show that proposition 1.1 can
at most be generalized in the countable case to families (ei )i∈I indexed by a countable
set I and maps ϕ : I → I that are conjugate to the successor function succ : n 7→ n + 1
defined on N, at least if we want to preserve a condition like T (ei ) = eϕ(i) for all i ∈ I.
We finally prove in section 6 a tentative generalization of the result, where we replace
the condition T (ei ) = eϕ(i) for all i ∈ I where ϕ is conjugate to the successor function
with a more sophisticated one.

2
2 Notations
In the sequel, N denotes the set {0, 1, 2, · · · } of natural numbers including 0. N∗
denotes N \ {0}.
If A is a set, we denote by |A| the cardinality of A, P(A) the powerset of A, Pω (A) the
set {B ⊆ A : |B| < ∞}, and Pω,∗ (A) the set {B ⊆ A : 0 < |B| < ∞}.
If A is a set, ϕ : A → A a self map and n ∈ N∗ , we denote by ϕn the composition of ϕ
with itself n times : ϕ ◦ · · · ◦ ϕ : A → A. In addition, we define ϕ0 to be the identity
function on A. Moreover, if a ∈ A, we denote by Orbϕ (a) the forward orbit of a under
the iterates of ϕ : {ϕn (a) : n ∈ N}.
If E and F are two vector spaces, L(E, F ) denotes the set of linear operators from E to
F . When E = F , we simply write L(E). If E is a vector space, we denote by L(∗, E)
the class {T ∈ L(E ∗ , E) : E ∗ is a vector space}.

3 Lemmas in order theory

Definition 3.1. Consider some set P and a binary relation ≤ on P . Then ≤ is a preorder
if it is reflexive and transitive; i.e., for all a, b and c in P , we have that:
• a ≤ a (reflexivity)

• if a ≤ b and b ≤ c then a ≤ c (transitivity)

A set that is equipped with a preorder is called a preordered set.
Definition 3.2. Consider a preordered set (P, ≤) and a map p : P → P . Then p is
called a projection if for all a and b in P , we have that:
• a ≤ b implies p(a) ≤ p(b) (p is monotone/increasing/order-preserving/isotone)

• p(p(a)) = p(a) (idempotence)

Lemma 3.1. Let (P, ≤) be a preordered set, (an )n∈N a sequence in P , b ∈ P and p a
projection on P .
Suppose there exists an increasing map f : P → P such that f (p(b)) ≤ p(f (b)), ∀n ∈ N :
an+1 ≤ f (an ), and f (b) ≤ a0 .
Then: (a0 ≤ p(b)) ⇒ (∀n ∈ N : an ≤ p(b)).
Proof. Indeed, suppose that a0 ≤ p(b). Let’s show by induction that ∀n ∈ N : an ≤ p(b).
The base case is the hypothesis a0 ≤ p(b). Suppose we have that an ≤ p(b) for some
n ∈ N. Then an+1 ≤ f (an ) ≤ f (p(b)) ≤ p(f (b)) ≤ p(a0 ) ≤ p(p(b)) = p(b). Hence
an ≤ p(b) holds for all n ∈ N and the lemma is proved.

Definition 3.3. Consider some set P and a binary relation ≤ on P . Then ≤ is a partial
order if it is a preorder and for all a and b in P , we have a ≤ b ≤ a implies a = b
(antisymmetry).
A set that is equipped with a partial order is called a partially ordered set or poset.

3
Definition 3.4. A set S partially ordered by the binary relation ≤ is a join-semilattice
if for all elements x and y of S, the smallest upper bound of the set {x, y} exists in S.
The smallest upper bound of the set {x, y} is called the join of x and y, denoted x ∨ y.

Lemma 3.2. Let (S, ≤, ∨) be a join-semilattice, p a projection on S, (am,n )(m,n)∈N2 a

double sequence in S, (bm )m∈N an increasing sequence in S such that ∀m ∈ N : am,0 ≤
p(bm+1 ).
Suppose there exists m• ∈ N and an increasing map f : S → S such that f (p(bm• )) ≤
p(f (bm• )), ∀(m, n) ∈ N2 : am,n+1 ≤ f (am,n ), and f (bm• ) ≤ m
W •
i=0 ai,0 .
Then: (am• ,0 ≤ p(bm• ) ⇒ (∀n ∈ N : am• ,n ≤ p(bm• )).

Proof. For all (m, n) ∈ N2 , let ag m,n = i=0 ai,n . Suppose that am• ,0 ≤ p(bm• ). Then
m
W

m• ,0 = i=0 ai,0 ≤ p(bm• ) since ∀i ∈ [[0, m• − 1]] : ai,0 ≤ p(bi+1 ), am• ,0 ≤ p(bm• ), and
Wm•
a]
(p(bi ))i∈N is increasing. Moreover, f being increasing implies ∀a, b ∈ S : f (a) ∨ f (b) ≤
f (a ∨ b), and so ∀n ∈ N : a^m• ,n+1 = i=0 ai,n+1 ≤ i=0 f (ai,n ) ≤ f ( i=0 ai,n ) = f (^ am• ,n ).
Wm• Wm • Wm •

Therefore, applying lemma 3.1 to (^ am• ,n )n∈N , bm• , and f , we have that ∀n ∈ N : a
^ m• ,n ≤
p(bm• ) which implies ∀n ∈ N : am• ,n ≤ p(bm• ).

Lemma 3.3. Let (S, ≤, ∨) be a join-semilattice, p a projection on S, (en )n∈N and (bn )n∈N
two sequences in S such that (bn )n∈N is increasing and ∀n ∈ N : en ≤ p(bn+1 ).
Suppose there exists m• ∈ N and an increasing map f : S → S such that f (p(bm• )) ≤
p(f (bm• )), ∀n ∈ N : en+1 ≤ f (en ), and f (bm• ) ≤ m
W •
i=0 ei .
Then: (em• ≤ p(bm• )) ⇒ (∀n ∈ N : em• +n ≤ p(bm• )).

Proof. Define am,n as em+n for all (m, n) ∈ N2 . Then ((am,n )(m,n)∈N2 , (bn )n∈N , f ) is a
triple satisfying the conditions of lemma 3.2 and so we deduce the result.

4 Application to model theory

In this section, we will use lemma 3.3 to prove a model-theoretic result. For a quick
reference on model theory, see the book [2]. The model-theoretic result allows to recover
proposition 1.1 when we choose the σ-structure to be a vector space over a field. The
idea of the following model-theoretic proposition is to consider an unsorted algebraic
σ-structure A = (A, σ) and regard the map sending a set X ⊆ A to the set TermsA [X]
of interpreted terms with variables taken from X as a special projection map on the
boolean algebra P(A).

Proposition 4.1. Let σ be an algebraic signature and A = (A, σ) a σ-structure.

Let (en )n∈N be a sequence of elements in A.
Suppose there exists a map f ∈ Endσ (A) such that ∀n ∈ N : f (en ) = en+1 .
Then: (∃m ∈ N : em ∈ TermsA [e0 , · · · , em−1 ]) ⇒ (∃m ∈ N : ∀n ∈ N : em+n ∈
TermsA [e0 , · · · , em−1 ]).

Proof. Notice that (P(A), ⊆, ∪, ∩, ∅, A) is a boolean algebra and so a join-semilattice

(P(A), ⊆, ∪). Define p as the map sending a set X ⊆ A to the set TermsA [X] of

4
interpreted terms with variables taken from X, and bn = {e0 , · · · , en−1 }. The sequence
(en )n∈N in A induces a sequence of singletons (
({en })n∈N in P(A). Also, the endomorphism
P(A) → P(A)
f induces the direct image map f• = which is increasing, satisfies
X 7→ f (X)
∀X ∈ P(A) : f• (p(X)) = p(f• (X)), ∀i ∈ N : {ei+1 } = f• ({ei }), and f• ({e0 , · · · , em−1 }) =
{f (e0 ), · · · , f (em−1 )} = {e1 , · · · , em } ⊆ mi=0 {ei }. Moreover (bn )n∈N is increasing and
S

∀n ∈ N : {en } ⊆ TermsA [e0 , · · · , en ]. Suppose ∃m• ∈ N : em• ∈ TermsA [e0 , · · · , em• −1 ].

Then the implication follows from lemma 3.3 applied to (S, ≤, ∨) := (P(A), ⊆, ∪), p,
({en })n∈N , (bn )n∈N , f• , and m• .

The previous proposition can be better appreciated after considering its contrapositive:
Corollary 4.1. Let σ be an algebraic signature and A = (A, σ) a σ-structure.
Let (en )n∈N be a sequence of elements in A.
Suppose there exists a map f ∈ Endσ (A) such that ∀n ∈ N : f (en ) = en+1 .
Then: (∀m ∈ N : ∃φ(m) ≥ m : eφ(m) ∈ / TermsA [e0 , · · · , em−1 ]) ⇒ (∀m ∈ N : em ∈
/
Terms [e0 , · · · , em−1 ]).
A

Example 4.1. Consider the case of a vector space over a field, with its classical signature
described for instance in [2] pp. 3-4. Then the corollary means that if we have an infinite
sequence (en )n∈N of vectors and a linear map with the property f (en ) = en+1 for all
n ∈ N, then in order to show the linear independence of the sequence (en )n∈N , it is
sufficient to prove that it spans an infinite-dimensional space. This is the result appearing
in the paper [1] by O. Christensen and M. Hasannasab (proposition 1.1 of the present
article).
Corollary 4.2. Let E be an infinite-dimensional vector space, e ∈ E, and S a linear
operator in E. Define the infinite sequence (en )n∈N in E as en = S n (e) for all n ∈ N.
Suppose that dim span(en )n∈N = ∞. Then (en )n∈N is free.
Proof. We have ∀n ∈ N : S(en ) = en+1 and we conclude by proposition 1.1.

We do not know if it is possible to formulate a definition of infinite dimension or

independence in the context of general algebraic σ-structures as in corollary 4.1. If
possible, corollary 4.1 may be advantageously applied to other algebraic σ-structures like
groups, rings, algebras, etc., and have the same intuitive meaning of establishing a link
between infinite dimension and independence.

5 The only possible countable extensions

Before establishing the main proposition 5.1, we need to recall some well-known
lemmas.
Lemma 5.1. Let I be an infinite set, a ∈ I and ϕ : I → I.
Suppose that Orbϕ (a) is infinite.
Then a, ϕ(a), ϕ(ϕ(a)), · · · are distinct.

5
Proof. We use euclidean division. Suppose that ϕn (a) = ϕm (a) for n < m. By induction,
we have ϕn+j (a) = ϕm+j (a) for all j ∈ N. Let e ≥ n. Let e − n = q(m − n) + r
be the division with remainder of e − n ∈ N by m − n ∈ N∗ . If q ≥ 1, we have
ϕe (a) = ϕn+q(m−n)+r (a) = ϕm+(q−1)(m−n)+r (a) = ϕn+(q−1)(m−n)+r (a). By immediate
induction, we have that ϕe (a) = ϕn+r (a), where 0 ≤ r < m − n. So Orbϕ (a) =
{a, ϕ(a), · · · , ϕm−1 (a)} is finite, contradiction.

Lemma 5.2. Let I be an infinite set, (a, b) ∈ I 2 and ϕ : I → I.

Suppose that Orbϕ (a) is infinite and Orbϕ (b) is cofinite.
Then (∃(m, n) ∈ N2 ) : ϕm (a) = ϕn (b).

Proof. Assume by way of contradiction that (∀(m, n) ∈ N2 : ϕm (a) ̸= ϕn (b)). Then

Orbϕ (a) ⊆ I \ Orbϕ (b). But this is impossible since Orbϕ (a) is infinite and I \ Orbϕ (b) is
finite. Hence the result.

Lemma 5.3. Let I be a countably infinite set and ϕ : I → I. Then

(
N →N
(∃a ∈ I : Orbϕ (a) = I) ⇔ (ϕ is conjugate to succ : in the sense that
n 7→ n + 1
∃α : N → I such that α is bijective and α ◦ succ = ϕ ◦ α).
(
N →I
Proof. (⇒) Suppose that ∃a ∈ I : Orbϕ (a) = I. Define α : . By lemma
n 7→ ϕn (a)
5.1, α is bijective. Moreover, we have clearly
( α ◦ succ = ϕ ◦ α.
N →N
(⇐) Suppose ϕ is conjugated to succ : by α. Then we have Orbϕ (α(0)) =
n 7→ n + 1
{ϕn (α(0))}n∈N = {α(succn (0))}n∈N = {α(n)}n∈N = I.

The following proposition is the main result of this section.

Proposition 5.1. Let E be an infinite dimensional vector space and I a countably

infinite set.
For all (ei )i∈I ∈ E I and ϕ : I → I, let P ((ei )i∈I , ϕ) be the proposition

(∃T ∈ L(span(ei )i∈I , E) :∀i ∈ I : T (ei ) = eϕ(i) and dim span(ei )i∈I = +∞)
⇒ (ei )i∈I is free.

Then we have
h  i
∀ϕ : I → I : ∀(ei )i∈I ∈ E I : P ((ei )i∈I , ϕ) ⇔ ∃a ∈ I : Orbϕ (a) = I

Proof. Let ϕ : I → I.
(⇒) : Suppose that ∀(ei )i∈I : P ((ei )i∈I , ϕ).
Let’s show first that all the orbits of ϕ are cofinite. Suppose by way of contradic-
tion that ∃a ∈ I : |I \ Orbϕ (a)| = +∞. Let (an )n∈N be a free family in E. We let

6
 0 if i ∈ Orbϕ (a)
(
u : I \ Orbϕ (a) → N be a bijection, and set for all i ∈ I, ei = . We de-
au(i) otherwise
fine an operator on span(ei )i∈I = span(an )n∈N by setting T (an ) = eϕ(u−1 (n)) for all n ∈ N
and extending linearly. Then we have ∀i ∈ I : T (ei ) = eϕ(i) . Indeed, if i ∈ Orbϕ (a), then
ϕ(i) ∈ Orbϕ (a) and so ei = eϕ(i) = 0 by definition which makes the relation true. Other-
wise, we have ei = au(i) and so T (ei ) = T (au(i) ) = eϕ(u−1 (u(i))) = eϕ(i) . Besides, we have
dim span(ei )i∈I = +∞ since span(ei )i∈I = span(an )n∈N and dim span(an )n∈N = +∞.
Since P ((ei )i∈I , ϕ) is true, it follows that (ei )i∈I is free which is impossible since ea = 0.
Now take a ∈ I such that |I \ Orbϕ (a)| is minimal. From lemma 5.2, ∀b ∈ I :
∃(m(b), n(b)) ∈ N2 : ϕm(b) (b) = ϕn(b) (a). If we choose m(b) and n(b) such that
m(b) + n(b) is minimal, then n(b) ≥ m(b) by minimality of I \ Orbϕ (a). Let (an )n∈N
be a free family in E. For all i ∈ I, we set ei = an(i)−m(i) . We define an operator on
span(ei )i∈I = span(an )n∈N by setting T (an ) = an+1 for all n ∈ N and extending linearly.
Then we have ∀i ∈ I : T (ei ) = eϕ(i) . Indeed, if i ∈ Orbϕ (a), then m(i) = m(ϕ(i)) = 0 and
n(ϕ(i)) = n(i) + 1 which implies T (ei ) = T (an(i) ) = an(i)+1 = an(ϕ(i)) = eϕ(i) . Otherwise,
m(i) ≥ 1, m(ϕ(i)) = m(i) − 1 and n(ϕ(i)) = n(i) since ϕm(i) (i) = ϕn(i) (a) with m(i) + n(i)
minimal and m(i) ≥ 1 implies ϕm(i)−1 (ϕ(i)) = ϕn(i) (a) with (m(i) − 1) + n(i) minimal.
So T (ei ) = an(i)−m(i)+1 = an(i)−(m(i)−1) = an(ϕ(i))−m(ϕ(i)) = eϕ(i) . Besides, we have
dim span(ei )i∈I = +∞ since span(ei )i∈I = span(an )n∈N and dim span(an )n∈N = +∞.
Since P ((ei )i∈I , ϕ) is true, it follows that (ei )i∈I is free which implies Orbϕ (a) = I (other-
wise, ∃b ∈ I \ Orbϕ (a), and so eb = an(b)−m(b) = eϕn(b) (a) with b ̸= ϕn(b) (a), contradicting
the independence of (ei )i∈I ).
(⇐) : Suppose that ∃a ∈ I : Orbϕ (a) = I. By lemma 5.3, ϕ is conjugated to succ :
(
N →N
by α. Let (ei )i∈I ∈ E I and suppose there exists T ∈ L(span(ei )i∈I , E) such
n 7→ n + 1
that for all i ∈ I : T (ei ) = eϕ(i) and dim span(ei )i∈I = +∞. Define (fn )n∈N by fn = eα(n)
for all n ∈ N. Then (fn )n∈N satisfies ∀n ∈ N : T (fn ) = fn+1 and dim span(fn )n∈N = +∞
which by proposition 1.1 implies that (fn )n∈N is free. Since α is a bijection, it follows
that (ei )i∈I is free.

6 A tentative generalization
Definition 6.1. Consider two sets X and Y . A binary relation R on X and Y is a subset of
the cartesian product X×Y . The direct image of a subset S ⊆ X under a binary relation R
on X and Y is written R[S] and refers to {y ∈ Y : ∃x ∈ S : (x, y) ∈ R}. If X = Y , a binary
relation on X and Y is simply called a relation on X. The relation ∆X := {(x, x) : x ∈ X}
is called the identity relation on X. The composition of two binary relations R1 and R2
over X and Y , and Y and Z (respectively), is the binary relation over X and Z, denoted
R1 ◦ R2 , and given by the subset {(x, z) ∈ X × Z : ∃y ∈ Y : (x, y) ∈ R1 ∧ (y, z) ∈ R2 }. If
R is a binary relation on X, we define R0 as ∆X and for all n ≥ 1, Rn as R ◦ Rn−1 .
The following proposition which uses the language of functions with set-valued inputs
or/and outputs is the main result of this section.

7
Proposition 6.1. Let E be an infinite dimensional vector space, V a finite dimensional
subspace of E, (ei )i∈I a family of vectors in E, and J an infinite set. Suppose that
1. There exist two functions u : Pω,∗ (I) → I and G : Pω,∗ (I) → Pω,∗ (I) such that for
all I ∗ ∈ Pω,∗ (I), u(I ∗ ) ∈ I ∗ and I ∗ ⊆ G(I ∗ ),

2. There exists a function T : Pω,∗ (I) × J → L(∗, E) such that for all (I ∗ , j) ∈
Pω,∗ (I) × J, T (I ∗ , j) ∈ L(span(ei )i∈G(I ∗ ) , E),

3. There exists a finite subset J0 of J and a relation R on J such that R[j] ⊆ J0 for
all j ∈ J0 and (∀j ∈ J)(∃nj ∈ N)Rnj [j] ⊆ J0 ,

4. ∀i ∈ I : dim span{T (I ∗ , j)ei }j∈J = ∞,

5. ∀(I ∗ , j) ∈ Pω,∗ (I) × J : ∀i ∈ G(I ∗ ) \ {u(I ∗ )} :

T (I ∗ , j)ei ∈ span{T (I ∗ , j ′ )ei′ }(j ′ ,i′ )∈R[j]×G(I ∗ ) + V.

Then (ei )i∈I is free.

Proof. (of proposition 6.1) Assume by way of contradiction that (ei )i∈I is dependent.
Then, there exists I ∗ ∈ Pω,∗ (I) and coefficients {ci }i∈I ∗ in C∗ such that

ci ei = 0,
X

i∈I ∗

which implies
−ci
eu(I ∗ ) =
X
ei .
c ∗
i∈I ∗ \{u(I ∗ )} u(I )

By linearity of T (I ∗ , j), we then have for all j ∈ J

−ci
T (I ∗ , j)eu(I ∗ ) = T (I ∗ , j)ei . (∗)
X
c ∗
i∈I ∗ \{u(I ∗ )} u(I )

We will prove by induction on n ∈ N that

∀(n, j, i) ∈ N × J × (G(I ∗ ) \ {u(I ∗ )}) : T (I ∗ , j)ei ∈ span{T (I ∗ , j ′ )e′i }(j ′ ,i′ )∈Rn [j]×G(I ∗ ) + V.
(∗∗)
• For n = 0, the relation is true since R0 (j) = ∆J [j] = {j} and i ∈ G(I ∗ ) (take
(j ′ , i′ ) := (j, i) and v := 0).

• Suppose the induction hypothesis is true at the order n ∈ N. So for all (j, i) ∈
J × (G(I ∗ ) \ {u(I ∗ )}), there exist complex coefficients {αjj,i′ ,i′ }(j ′ ,i′ )∈R[j]×G(I ∗ ) and
{βjj,i′ ,i′ }(j ′ ,i′ )∈Rn [j]×G(I ∗ ) , and vectors v j,i and wj,i in V such that

T (I ∗ , j)ei = αjj,i′ ,i′ T (I ∗ , j ′ )ei′ + v j,i , (∗ ∗ ∗)

X

(j ′ ,i′ )∈R[j]×G(I ∗ )

8
 and
T (I ∗ , j)ei = βjj,i′ ,i′ T (I ∗ , j ′ )ei′ + wj,i , (∗ ∗ ∗∗)
X

(j ′ ,i′ )∈Rn [j]×G(I ∗ )

because of condition 5 and the induction hypothesis.

Let (j, i) ∈ J × (G(I ∗ ) \ {u(I ∗ )}). We have
T (I ∗ , j)ei = βjj,i′ ,i′ T (I ∗ , j ′ )ei′ + wj,i
X

(j ′ ,i′ )∈Rn [j]×G(I ∗ )

= βjj,i′ ,i′ T (I ∗ , j ′ )ei′

X

(j ′ ,i′ )∈Rn [j]×(G(I ∗ )\{u(I ∗ )}

+ βjj,i′ ,u(I ∗ ) T (I ∗ , j ′ )eu(I ∗ ) + wj,i

X

j ′ ∈Rn [j]

= βjj,i′ ,i′ T (I ∗ , j ′ )ei′

X

(j ′ ,i′ )∈Rn [j]×(G(I ∗ )\{u(I ∗ )}

−ci′
+ βjj,i′ ,u(I ∗ ) T (I ∗ , j ′ )ei′ + wj,i ,
X

(j ′ ,i′ )∈Rn [j]×(G(I ∗ )\{u(I ∗ )})

cu(I ∗ )

where we have used equation (∗) and extended {ci }i∈I ∗ to {ci }i∈G(I ∗ ) by setting
∀i′ ∈ G(I ∗ ) \ I ∗ : ci′ = 0. Simplifying, we have
ci′
T (I ∗ , j)ei = (βjj,i′ ,i′ − βjj,i′ ,u(I ∗ ) )T (I ∗ , j ′ )ei′ + wj,i .
X

(j ′ ,i′ )∈Rn [j]×(G(I ∗ )\{u(I ∗ )}

cu(I ∗ )

Using equality (∗ ∗ ∗) for each term T (I ∗ , j ′ )ei′ and then rearranging the resulting
sum, we have that T (I ∗ , j)ei is equal to
 !
′ ′ ci′
αjj ′′,i,i′′ βjj,i′ ,i′ βjj,i′ ,u(I ∗ )
X X
 − 
(j ′ ,i′ )∈F (j ′′ ,j,n)×G(I ∗ )\{u(I ∗ )}
cu(I ∗ )
(j ′′ ,i′′ )∈Rn+1 [j]×G(I ∗ )
 
ci′ j ′ ,i′
· T (I ∗ , j ′′ )fi′′ +  (βjj,i′ ,i′ − βjj,i′ ,u(I ∗ ) )v + wj,i 
X

(j ′ ,i′ )∈Rn [j]×(G(I ∗ )\{u(I ∗ )}

cu(I ∗ )
∗ ′
∈ span{T (I , j )ei′ }(j ′ ,i′ )∈Rn+1 [j]×G(I ∗ ) + V,
where F (j ′′ , j, n) := {j ′ ∈ Rn [j] : j ′′ ∈ R[j ′ ]}.
Hence the claim is proved. Now letting j ∈ J and using condition 3, there exists nj ∈ N
such that Rnj [j] ⊆ J0 , which implies by (∗) and the claim (∗∗) that
T (I ∗ , j)eu(I ∗ ) ∈ span{T (I ∗ , j ′ )ei′ }(j ′ ,i′ )∈J0 ×G(I ∗ ) + V.
Therefore
span{T (I ∗ , j)eu(I ∗ ) }j∈J ⊆ span{T (I ∗ , j ′ )ei′ }(j ′ ,i′ )∈J0 ×G(I ∗ ) + V.
Since span{T (I ∗ , j ′ )ei′ }(j ′ ,i′ )∈J0 ×G(I ∗ ) + V is finite-dimensional, this contradicts condition
4.

9
Example 6.1. (from O. Christensen and M. Hasannab [1])
Let E be an infinite-dimensional vector space and (en )n∈N a sequence in E.
Suppose that dim span(en )n∈N = ∞ and there exists an operator S : span(en )n∈N → E
such that S(en ) = en+1 for all n ∈ N. Then (en )n∈N is free.
Indeed, define u : Pω,∗ (N) → N and G : Pω,∗ (N) → Pω,∗ (N) by u(I ∗ ) = max(I ∗ ) and
G(I ∗ ) = [[0, max(I ∗ )]] for all I ∗ ∈ Pω,∗ (N). Moreover, define T : Pω,∗ (N) × N → L(∗, E)
by T (I ∗ , m) = (S m )| span(en )n∈[[0,max(I ∗ )]] for all (I ∗ , m) ∈ Pω,∗ (N) × N. Take J0 = {0} and
define R = (0, 0) ( m∈N∗ (m, m − 1)) ⊆ N × N. Examining the conditions of proposition
S S

6.1, we see that conditions 1-2 are valid, 3 is true because Rm [m] = {0} = J0 for
all m ∈ N, 4 is valid because {T (I ∗ , m)eu(I ∗ ) }m∈N = {eu(I ∗ )+m }m∈N spans an infinite
dimensional space since (em )m∈N does, and finally condition 5 is also valid since for all
(I ∗ , m) ∈ Pω,∗ (N) × N∗ and for all n ∈ [[0, max(I ∗ ) − 1]] we have T (I ∗ , m)en = en+m =
T (I ∗ , m − 1)en+1 (notice that m − 1 ∈ R[m] and n + 1 ∈ [[0, max(I ∗ )]]), and for all
I ∗ ∈ Pω,∗ (N) and n ∈ [[0, max(I ∗ ) − 1]] we have T (I ∗ , 0)en = en = T (I ∗ , 0)en (notice that
0 ∈ R[0] and n ∈ [[0, max(I ∗ )]]). Hence the result.

Before we move on to the next example, we need to prove two additional lemmas.

Lemma 6.1. Let I be an infinite set, (a, b) ∈ I 2 and ϕ : I → I.

Suppose that Orbϕ (a) is cofinite. Then

1. If ∃n ∈ N : b = ϕn (a), then Orbϕ (b) is cofinite,

2. If ∃n ∈ N : a = ϕn (b), then Orbϕ (b) is cofinite.

Proof. First, notice that by lemma 5.1, a, ϕ(a), ϕ2 (a), · · · are distinct.
(1) : Suppose ∃n ∈ N : b = ϕn (a). Then I \ Orbϕ (b) = I \ {ϕm (a)}m≥n = (I \ Orbϕ (a)) ∪
{ϕm (a)}m<n is finite.
(2) : Suppose ∃n ∈ N : a = ϕn (b). Then I \ Orbϕ (b) = (I \ {ϕm (b)}m≥n ) \ {ϕm (b)}m<n =
(I \ Orbϕ (a)) \ {ϕm (b)}m<n is finite.

Corollary 6.1. Let I be an infinite set, a ∈ I and ϕ : I → I.

Suppose that Orbϕ (a) = I.
Then ∀b ∈ I : Orbϕ (b) is cofinite

Lemma 6.2. Let I be an infinite set and ϕ : I → I such that ∀a ∈ I : Orbϕ (a) is infinite.
Then we have

(∃a ∈ I : Orbϕ (a) = I) ⇒

(there exist two functions u : Pω,∗ (I) → I and G : Pω,∗ (I) → Pω,∗ (I) such that
∀I ∗ ∈ Pω,∗ (I) : u(I ∗ ) ∈ I ∗ , I ∗ ⊆ G(I ∗ ), and ϕ(G(I ∗ ) \ {u(I ∗ )}) ⊆ G(I ∗ )).

Proof. Define
Pω,∗ (I) → N
(
n:
I∗ 7→ max{n ∈ N : ϕn (a) ∈ I ∗ }

10
which is well-defined because {a, ϕ(a), ϕ2 (a), · · · } are distinct (lemma 5.1) and let

Pω,∗ (I) → I
(
u: ∗
I∗ 7→ ϕn(I ) (a)

and
Pω,∗ (I) → Pω,∗ (I)
(
G: ∗ .
I∗ 7→ {a, ϕ(a), · · · , ϕn(I ) (a)}
With this choice, since Orbϕ (a) = I, u and G do satisfy the requirements.

Example 6.2. Let E be an infinite dimension vector space, I a countably infinite set,
(ei )i∈I a family in E and ϕ : I → I such that ∃a ∈ I : Orbϕ (a) = I.
Suppose that dim span{ei }i∈I = ∞ and there exists an operator S : span{ei }i∈I → E
such that S(ei ) = eϕ(i) for all i ∈ I. Then {ei }i∈I is free.
Indeed, from lemma 6.2, there exist two functions u : Pω,∗ (I) → I and G : Pω,∗ (I) →
Pω,∗ (I) such that for all I ∗ ∈ Pω,∗ (I), u(I ∗ ) ∈ I ∗ , I ∗ ⊆ G(I ∗ ) and ϕ(G(I ∗ ) \ {u(I ∗ )}) ⊆
G(I ∗ ). Define T : Pω,∗ (I) × N → L(∗, E) by T (I ∗ , m) = (S m )| span{ei }i∈G(I ∗ ) for all
(I ∗ , m) ∈ Pω,∗ (I) × N. Take J0 = {0} and define R = (0, 0) ( m∈N∗ (m, m − 1)) ⊆
S S

N × N. Examining the conditions of proposition 6.1, we see that conditions 1-2

are valid, 3 is true because Rm [m] = {0} = J0 for all m ∈ N, 4 is valid because
{T (I ∗ , m)eu(I ∗ ) }m∈N = {eϕm (u(I ∗ )) }m∈N spans an infinite-dimensional space since (ei )i∈I
does and |I \ {ϕm (u(I ∗ ))}m∈N | < ∞ (corollary 6.1), and finally condition 5 is also
valid since for all (I ∗ , m) ∈ Pω,∗ (I) × N∗ and for all i ∈ G(I ∗ ) \ {u(I ∗ )} we have
T (I ∗ , m)ei = eϕm (i) = T (I ∗ , m − 1)eϕ(i) (notice that m − 1 ∈ R[m] and ϕ(i) ∈ G(I ∗ )),
and for all I ∗ ∈ Pω,∗ (I) and i ∈ G(I ∗ ) \ {u(I ∗ )} we have T (I ∗ , 0)ei = ei = T (I ∗ , 0)ei
(notice that 0 ∈ R[0] and i ∈ G(I ∗ )). Hence the result.

Acknowledgement
The first author is financially supported by the Centre National pour la Recherche
Scientifique et Technique of Morocco.

Conflict of interest
On behalf of all authors, the corresponding author states that there is no conflict of
interest.

References
[1] O. Christensen and M. Hasannasab. Frame properties of systems arising via iterated
actions of operators. Appl. Comput. Harmon. Anal., 46(3):664–673, 2019.

[2] W. Hodges. A Shorter Model Theory. Cambridge University Press, 2017.

11
[3] J. S. Moghaddam, A. Najati, and Y. Khedmati. Fibonacci representations of sequences
in hilbert spaces. U.P.B. Sci. Bull., Series A, 83(1), 2021.

Nizar El Idrissi.
Laboratoire : Equations aux dérivées partielles, Algèbre et Géométrie spectrales.
Département de mathématiques, faculté des sciences, université Ibn Tofail, 14000 Kénitra.
E-mail address : [email protected]
Pr. Samir Kabbaj.
Laboratoire : Equations aux dérivées partielles, Algèbre et Géométrie spectrales.
Département de mathématiques, faculté des sciences, université Ibn Tofail, 14000 Kénitra.
E-mail address : [email protected]

12