Path-connectedness and topological closure of some

sets related to the non-compact Stiefel manifold

Nizar El Idrissi, Samir Kabbaj, and Brahim Moalige

June 19, 2023

Abstract
If H is a Hilbert space, the non-compact Stiefel manifold St(n, H) consists
of independent n-tuples in H. In this article, we contribute to the topological
study of non-compact Stiefel manifolds, mainly by proving two results on
the path-connectedness and topological closure of some sets related to the
non-compact Stiefel manifold. In the first part, after introducing and proving
T
an essential lemma, we prove that j∈J (U (j) + St(n, H)) is path-connected
by polygonal paths under a condition on the codimension of the span of the
components of the translating J-family. Then, in the second part, we show
that the topological closure of St(n, H) ∩ S contains all polynomial paths
contained in S and passing through a point in St(n, H). As a consequence, we
prove that St(n, H) is relatively dense in a certain class of subsets which we
illustrate with many examples from frame theory coming from the study of the
solutions of some linear and quadratic equations which are finite-dimensional
continuous frames. Since St(n, L2 (X, µ; F)) is isometric to F(X,Σ,µ),n
F , this
article is also a contribution to the theory of finite-dimensional continuous
Hilbert space frames.

Contents
1 Introduction 2
2020 Mathematics Subject Classification. 57N20; 42C15; 54D05; 54D99.
Key words and phrases. Stiefel manifold, continuous frame, path-connected space, topological
closure, dense subset.

1
2 Preliminaries 3
2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Continuous frames in Fn . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Basic topological properties of St(n, H) and Sto (n, H) . . . . . . . . 6

3 Path-connectedness of (U (j) + St(n, H)) 9

T
j∈J

4 Topological closure of St(n, H) ∩ S 11

5 Existence of solutions of some linear and quadratic equations

which are finite-dimensional continuous frames 12
5.1 Linear equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 A quadratic equation . . . . . . . . . . . . . . . . . . . . . . . . . . 17

6 Examples of subspsaces in which the space of continuous frames

is relatively dense 21

1 Introduction
Duffin and Shaeffer introduced in 1952 [15] the notion of a Hilbert space frame
to study some deep problems in nonharmonic Fourier series. However, the general
idea of signal decomposition in terms of elementary signals was known to Gabor
[18] in 1946. The landmark paper of Daubechies, Grossmann, and Meyer [14]
(1986) accelerated the development of the theory of frames which then became
more widely known to the mathematical community. Nowadays, frames have a
wide range of applications in both engineering science and mathematics: they have
found applications in signal processing, image processing, data compression, and
sampling theory. They are also used in Banach space theory. Intuitively, a frame in
a Hilbert space K is an overcomplete basis allowing non-unique linear expansions,
though technically, it must satisfy a double inequality called the frame inequality.
There are many generalizations of frames in the literature, for instance frames in
Banach space [10] or Hilbert C*-modules [17]. A general introduction to frame
theory can be found in [9, 12].
The space F(X,Σ,µ),n
F
of continuous frames indexed by (X, Σ, µ) and with values
in F is isometric to the Stiefel manifold St(n, L2 (X, µ; F)). If H is a Hilbert
n

space, the non-compact Stiefel manifold St(n, H) is the set of independent n-

frames in H, where an independent n-frame simply denotes an independent n-tuple.
Stiefel manifolds are studied in differential topology and are one of the fundamental
examples in this area. Even though the theory of finite dimensional Stiefel manifolds
is generally well-known [26, 22, 25], there are still some aspects under study [24, 27].

2
The theory of infinite dimensional Stiefel manifolds is less studied and some recent
results can be found in [6, 21].
There have been also many studies directly devoted to the geometry of frames
and their subsets. Connectivity properties of some important subsets of the frame
space Fk,n
F
were studied in [8, 29]. Differential and algebro-geometric properties of
these subsets were studied in [16, 32, 31, 33] and chapter 4 of [11] respectively.
A fiber bundle structure with respect to the L1 and L∞ norms was established
for continuous frames in [1, 2]. A notion of density for general frames analogous
to Beurling density was introduced and studied in [4]. Finally, connectivity and
density properties were studied for Gabor [5, 13, 23, 28] and wavelet [7, 19, 20, 30]
frames.

Plan of the article. This article is organized as follows. In section 2,

we set some notations, introduce the definition of continuous Bessel and frame
families and their basic properties in Fn , and present Stiefel manifolds with an
emphasis on their topological aspects. In section 3, after introducing and proving
an essential lemma, we prove that j∈J (U (j) + St(n, H)) is path-connected
T

by polygonal paths under a condition on the codimension of the span of the

components of the translating J-family. Then, in section 4, we show that the
topological closure of St(n, H) ∩ S contains all polynomial paths contained in S
and passing through a point in St(n, H). As a consequence, we prove that St(n, H)
is relatively dense in a certain class of subsets which we illustrate, in section 6,
with many examples from frame theory coming from the study of the solutions
of some linear and quadratic equations which are finite-dimensional continuous
frames (section 5).

2 Preliminaries
2.1 Notations
The following notations are used throughout this article.
N denotes the set of natural numbers including 0 and N∗ = N \ {0}.
We denote by n an element of N∗ and by F one of the fields R or C.
If K is a Hilbert space, we denote by L(K) and B(K) respectively the set of linear
and bounded operators in K. IdK is the identity operator of K.
If K is a Hilbert space, m ∈ N∗ , and θ1 , · · · , θm ∈ H, the Gram ma-
trix of (θ1 , · · · , θm ) is the matrix Gram(θ1 , · · · , θm ) whose k, l-coefficient is
Gram(θ1 , · · · , θm )k,l = ⟨θk , θl ⟩.
If σ, τ ∈ N∗ , we denote by Mσ,τ (F) the algebra of matrices of size σ × τ over the
field F. When σ = τ , we denote this algebra Mσ (F).

3
An element x ∈ Fn is a n-tuple (x1 , · · · , xn ) with xk ∈ F for all k ∈ [[1, n]].
If S ∈ L(Fn ), we denote by [S] ∈ Mn (F) the matrix of S in the standard basis of
Fn , and we write In as a shorthand for [IdFn ].
If U = (ux )x∈X is a family in Fn indexed by X, then for each k ∈ [[1, n]], we denote
by U k the family (ukx )x∈X .

2.2 Continuous frames in Fn

Let K be a Hilbert space over F and (X, Σ, µ) a measure space.
Definition 2.1. [12] We say that a family Φ = (φx )x∈X with φx ∈ K for all x ∈ X
is a continuous frame in K if
Z
∃ 0 < A ≤ B : ∀v ∈ K : A∥v∥2 ≤ |⟨v, φx ⟩|2 dµ(x) ≤ B∥v∥2
X

A frame is tight if we can choose A = B as frame bounds. A tight frame with

bound A = B = 1 is called a Parseval frame. A Bessel family is a family satisfying
only the upper inequality. A frame is discrete if Σ is the discrete σ-algebra and µ
is the counting measure. If (n, k) ∈ (N∗ )2 , we denote by F(X,Σ,µ),K , F(X,Σ,µ),n
F
and
Fk,n respectively the set of (X, Σ, µ)-continuous frames with values in K, the set
F

of (X, Σ, µ)-continuous frames with values in Fn , and the set of discrete frames
indexed by [[1, k]] and with values in Fn .
If U = (ux )x∈X with ux ∈ K for all x ∈ K is a continuous Bessel family in K,
we define its analysis operator TU : K → L2 (X, µ; F) by
∀v ∈ K : TU (v) := (⟨v, ux ⟩)x∈X .
The adjoint of TU is an operator TU∗ : L2 (X, µ; F) → K given by
Z
∀c ∈ L2 (X, µ; F) : TU∗ (c) = c(x)ux dµ(x).
X

The composition SU = TU∗ TU : K → K is given by

Z
∀v ∈ K : SU (v) = ⟨v, ux ⟩ux dµ(x)
X

and called the frame operator of U . Since U is a Bessel family, TU , TU∗ , and SU
are all well defined and continuous. If U is a frame in K, then SU is a positive
self-adjoint operator satisfying 0 < A ≤ SU ≤ B and thus, it is invertible.
We now recall a proposition preventing that a frame belongs to L2 (X, µ; K)
when dim(K) = ∞. Here the set L2 (X, µ; K) refers to Bochner square integrable
(classes) of functions in M(X; K), where the latter refers to the set of measurable
functions from X to K. It explains why we only study the L2 topology of frame
subspaces in the finite dimensional case.

4
Proposition 2.1. Let K be a Hilbert space with dim K = ∞. Then F(X,Σ,µ),K ∩
L2 (X, µ; K) = ∅
Proof. Let Φ = (φx )x∈X ∈ F(X,Σ,µ),K ∩ L2 (X, µ; K). Let {em }m∈M be an orthonor-
mal basis of K. We have
Tr(SΦ ) = Tr(TΦ∗ TΦ ) = ∥T (em )∥2
X

m∈M
X Z
= |⟨em , φx ⟩|2 dµ(x)
m∈M X
Z !
= 2
X
|⟨em , φx ⟩| dµ(x)
X m∈M
Z
= ∥φx ∥2 dµ(x).
X

Since Φ ∈ F(X,Σ,µ),K , there exists a constant A > 0 such that

SΦ ≥ A · Id,
so Z
∥φx ∥2 dµ(x) = T r(SΦ ) = +∞
X
since dim(K) = ∞. Hence Φ ∈
/ L2 (X, µ; K).
From now on, we consider K = Fn . In what follows, we will recall some
elementary facts about Bessel sequences and frames in this setting.
Proposition 2.2. A family U = (ux )x∈X with ux ∈ Fn for all x ∈ X is a continuous
Bessel family if and only if it belongs to L2 (X, µ, Fn ).
Proof. (⇒) Suppose that U = (ux )x∈X is a continuous Bessel family. For each
k ∈ [[1, n]], denote by ek the k-th vector of the standard basis of Fn .
Applying the definition to the vector ek , we have for each k ∈ {1, · · · , n} :
∥U k ∥2L2 (X,µ,F) < ∞, and so
n
∥U ∥2L2 (X,µ,Fn ) = ∥U k ∥2L2 (X,µ,F) < ∞,
X

k=1

which implies U ∈ L2 (X, µ; Fn ).

(⇐) Suppose that U = (ux )x∈I ∈ L2 (X, µ; Fn ). We have
Z
∀v ∈ Fn : |⟨v, ux ⟩|2 dµ(x) ≤ ∥U ∥2L2 (X,µ;Fn ) ∥v∥2 < ∞
x∈X

by the Cauchy-Schwarz inequality, which implies that U = (ux )x∈X is a continuous

Bessel family.

5
Lemma 2.1. If U ∈ L2 (X, µ; Fn ), then [SU ] = Gram(U 1 , · · · , U n ).
Proof. Let (ek )k∈[[1,n]] the standard basis of Fn . Let i, j ∈ [[1, n]]. Then
Z Z
[SU ]i,j = ⟨Sej , ei ⟩ = ⟨ej , ux ⟩⟨ux , ei ⟩dµ(x) = ujx uix dµ(x) = ⟨U i , U j ⟩.
X X

Proposition 2.3. [12] Suppose Φ = (φx )x∈X is a family in Fn . Then

Φ is a continuous frame ⇔ Φ ∈ L2 (X, µ; Fn ) and SΦ is invertible

⇔ Φ ∈ L2 (X, µ; Fn ) and det(Gram(Φ1 , · · · , Φn )) > 0
⇔ Φ ∈ L2 (X, µ; Fn ) and {Φ1 , · · · , Φn } is free.

Proposition 2.4. [12] Suppose Φ = {φx }x∈X is a family in Fn and let a > 0.
Then

Φ is a measurable a-tight frame ⇔ Φ ∈ L2 (X, µ; Fn ) and SΦ = aIn

⇔ Φ ∈ L2 (X, µ; Fn ) and Gram(Φ1 , · · · , Φn ) = aIn
⇔ Φ ∈ L2 (X, µ; Fn ) and (Φ1 , · · · , Φn )
is an orthogonal family of L2 (X, µ; F)
√
and (∀i ∈ [[1, n]] : ∥Φi ∥ = a).

Example 2.1. Define φ1m = m1 e2πiam and φ2m = m1 e2πibm with a, b two real numbers
such that a − b is not an integer. Then Φ1 = (φ1m )m∈N and Φ2 = (φ2m )m∈N are
2
square summable with sum π6 . Since the sequences Φ1 and Φ2 are not proportional
due to the constraint on a and b, it follows by 2.3 that Φ is a discrete frame in C2 .
It is not however a tight frame since Φ1 and Φ2 are not orthogonal.

2.3 Basic topological properties of St(n, H) and Sto (n, H)

In this subsection, we introduce St(n, H) and Sto (n, H) as well as some of their
basic topological properties. We recall that n is a fixed element of N∗ . If H is
a Hilbert space, then St(n, H) is non-empty exactly when dim(H) ≥ n. In the
following, we will always suppose this condition.
Definition 2.2. The non-compact Stiefel manifold of independent n-frames in
H is defined by St(n, H) := {h = (h1 , · · · , hn ) ∈ H n : {h1 , · · · , hn } is free}.
The Stiefel manifold of orthonormal n-frames in H is defined by
Sto (n, H) := {h = (h1 , · · · , hn ) ∈ H n : {h1 , · · · , hn } is an orthonormal system}.
Proposition 2.5. We have

6
 1. St(n, L2 (X, µ; F)) is isometric to F(X,Σ,µ),n
F
.

2. Sto (n, L2 (X, µ; F)) is isometric to the set of continuous (X, Σ, µ)-Parseval
frames with values in Fn .

Proof. Define

L2 (X, µ; Fn ) → L2 (X, µ; F)n
Transpose : .
F = (fx )x∈X 7→ ((fx1 )x∈X , · · · , (fxn )x∈X )

Then Transpose is clearly an isometry, and it sends F(X,Σ,µ),n

F
to St(n, L2 (X, µ; F))
and Sto (n, L2 (X, µ; F)) to the set of continuous (X, Σ, µ)-Parseval frames with
values in Fn by propositions 2.3 and 2.4 respectively.
Remark 2.1. Because of proposition 2.5, the reader should keep in mind that
the following topological properties and the new results of this article are also
shared, for any measure space (X, Σ, µ), by F(X,Σ,µ),n
F
or the set of continuous
(X, Σ, µ)-Parseval frames with values in F , depending on the context.
n

Proposition 2.6. We have

1. St(n, H) is open in H n .

2. Sto (n, H) is closed in H n .

Proof. 1. St(n, H) is open because St(n, H) = (det ◦ Gram)−1 ((0, ∞)).

2. Sto (n, H) is closed because Sto (n, H) = Gram−1 (In ).

By joining continuously each element of St(n, H) to its corresponding Gram-

Schmidt orthonormalized system in Sto (n, H), we can prove

Proposition 2.7. Sto (n, H) is a deformation retract of St(n, H).

Concerning the connectedness properties of Stiefel manifolds, we have the

following

Definition 2.3. Let X be a topological space and m ∈ N. Then X is said to be

m-connected if its homotopy groups πi (X, b) are trivial for all i ∈ [[0, m]] and b ∈ X
(of course, π0 (X, b) is not a group but merely a pointed set).

Proposition 2.8. (see pp. 382-383 of [22]) We have

1. St(n, Rk ) is (k − n − 1)-connected.

7
 2. St(n, Ck ) is (2k − 2n)-connected.
3. If H is infinite dimensional, then St(n, H) is contractible.
Moreover, we have
Proposition 2.9. If H is infinite dimensional, then St(n, H) is contractible.
A proof can be found in this math.stackexchange thread.
The following proposition asserts the density of St(n, H) in H n . A corollary
of one of our results in this article (corollary 4.1) gives a generalization of this
proposition.
Proposition 2.10. St(n, H) is dense in H n .
Proof. Consider h = (h1 , · · · , hn ) ∈ H n . Pick some θ = (θ1 , · · · , θn ) ∈ St(n, H).
Let γ be the straight path connecting θ to h, i.e. for each t ∈ [0, 1] : γ(t) =
th + (1 − t)θ ∈ H n . Let Γ(t) = det(Gram((γ(t)1 , · · · , γ(t)n ))). Clearly, Γ(t) is a
polynomial function in t which satisfies Γ(0) ̸= 0 since θ ∈ St(n, H). Therefore

Γ(t) ̸= 0 except for a finite number of t′ s.

Moreover,
n n
||γ(t) − h||2H n = ||γ(t)k − hk ||2H = |1 − t|2 ||θk − hk ||2H → 0 when t → 1
X X

k=1 i=1

Hence, there exists t ∈ [0, 1] such that γ(t) is close to h and Γ(t) ̸= 0, and so
γ(t) ∈ St(n, H).
We also include the following proposition on the differential structure of the
Stiefel manifolds.
Proposition 2.11. [26] We have
1. St(n, Rk ) is a real manifold of dimension nk.

2. Sto (n, Rk ) is a real manifold of dimension nk − n(n+1)

2
.
3. St(n, Ck ) is a real manifold of dimension 2nk.
4. Sto (n, Ck ) is a real manifold of dimension 2nk − n2 .
5. If dim(H) = ∞, then St(n, H) and Sto (n, H) are Hilbert manifolds of infinite
dimension.
We ask the reader to keep in mind remark 2.1 when reading the remaining parts
of this article.

8
3 Path-connectedness of (U (j) + St(n, H))
T
j∈J
Definition 3.1. Let E be a topological vector space and γ : [0, 1] → E be a
continuous path. We say that γ is a polygonal path if there exist q ∈ N∗ , (ek )k∈[[1,q+1]]
a finite sequences with ek ∈ E for all k ∈ [[1,q + 1]], and (γk )k∈[[1,q]] a finite sequence
[0, 1] → E
of (continuous) straight paths with γk = such that
t 7→ (1 − t)ek + tek+1
γ = γ1 ∗ · · · ∗ γq , where ∗ is the path composition operation. We say that a
subset S ⊆ E is polygonally connected if every two points of S are connected by a
polygonal path.
In the following, when we say that S ⊆ E is polygonally connected, we mean
that each two points of S are connected by a polygonal path of the type γ1 ∗ γ2
where γ1 and γ2 are two straight paths.
Before we prove the main proposition of this section, let’s prove a useful lemma.
Lemma 3.1. 1. Suppose we have a family (a(j))j∈J indexed by J where each
a(j) belongs to St(n, H). Then if (a(j)k )j∈J,k∈[[1,n]] is free, we have

span({a(j)}j∈J ) \ {0} ⊆ St(n, H)

2. Suppose we have a family indexed by J where each a(j) belongs to Sto (n, H)
for all j ∈ J. Then if (a(j)k )j∈J,k∈[[1,n]] is an orthonormal system, we have
√
{x ∈ span({a(j)}j∈J ) : ∥x∥ = n} ⊆ Sto (n, H)

Proof. 1. Let h = (h1 , · · · , hn ) ∈ span({a(j)}j∈J ) \ {0}. We can write h =

u=1 λu a(ju ) with λu ∈ F for all u ∈ [[1, r]] and the λu ’s are not all zeros.
Pr

We need to show that (h1 , · · · , hn ) is an independent system. Suppose

otherwise nk=1 ck hk = 0. This means that nk=1 ck ( ru=1 λu a(ju )k ) = 0, and
P P P

so nk=1 ru=1 (λu ck )a(ju )k = 0. Since j∈J k∈[[1,n]] {a(j)k } is free, we deduce
P P S S

that λu ck = 0 for all u ∈ [[1, r]] and k ∈ [[1, n]], which implies that ck = 0 for
all k ∈ [[1, n]] since the λu ’s are not all zeros. Therefore, h ∈ St(n, H).
√
2. Let h = (h1 , · · · , hn ) ∈ span({a(j)}j∈J ) such that ∥h∥ = n. We can
write h = ru=1 λu a(ju ) with λu ∈ F for all u ∈ [[1, r]]. We need to show
P

that ⟨hk , hl ⟩ = δk,l for all k, l ∈ [[1, n]]. For k ̸= l, we have : ⟨hk , hl ⟩ =
⟨ ru=1 λu a(ju )k , ru=1 λu a(ju )l ⟩ = ru=1 rv=1 λu λv ⟨a(ju )k , a(jv )l ⟩ = 0 since
P P P P

j∈J k∈[[1,n]] {a(j)k } is an orthogonal system. Moreover, ∥hk ∥2 =

S S

λu λv ⟨a(ju )k , a(jv )k ⟩ = u=1 |λu |2 ∥a(ju )k ∥2 = ru=1 |λu |2 since

Pr Pr Pr P
u=1 v=1
j∈J k∈[[1,n]] {a(j)k } is an orthonormal system. By hypothesis, n = ∥h∥ =
S S 2

9
 ∥hk ∥2 = n( |λu |2 ), so ∥hk ∥2 = |λu |2 = 1 for all k ∈ [[1, n]] as
Pn Pr Pr
k=1 u=1 u=1
desired.

Proposition 3.1. Let H be a Hilbert space with dim(H) ≥ n, J an index set and
(U (j))j∈J a family with U (j) ∈ H n for all j ∈ J. If codimH (span({u(j)k : j ∈
J, k ∈ [[1, n]]})) ≥ 3n, then j∈J (U (j) + St(n, H)) is polygonally-connected.
T

Proof. Let X = (x1 , · · · , xn ) and Y = (y1 , · · · , yn ) in j∈J (U (j) + St(n, H)).

T

Let (z1 , · · · , zn ) be an independent family in H such that

span({zk : k ∈ [[1, n]]})

\
 
span {xk : k ∈ [[1, n]]} {yk : k ∈ [[1, n]]} {u(j)k : j ∈ J, k ∈ [[1, n]]} = {0}
[ [

This is possible since codimH (span({u(j)k : j ∈ J, k ∈ [[1, n]]}) ≥ 3n.

This ensures that we have for all j ∈ J

span({−u(j)k + zk : k ∈ [[1, n]]}) ∩ span({−u(j)k + xk : k ∈ [[1, n]]}) = {0},

span({−u(j)k + zk : k ∈ [[1, n]]}) ∩ span({−u(j)k + yk : k ∈ [[1, n]]}) = {0},

and
(−u(j)k + zk )k∈[[1,n]] is independent.
We define the straight paths

γ
1 : [0, 1] → H n
γ2 : [0, 1] → H n

by γ1 (t) = tZ + (1 − t)X and γ2 (t) = tY + (1 − t)Z respectively.

We have γ1 (0) = X, γ1 (1) = γ2 (0) = Z, and γ2 (1) = Y .
Since for all j ∈ J

span({−u(j)k + zk : k ∈ [[1, n]]}) ∩ span({−u(j)k + xk : k ∈ [[1, n]]}) = {0},

we have for all t ∈ [0, 1] and j ∈ J

−U (j) + tZ + (1 − t)X = t(−U (j) + Z) + (1 − t)(−U (j) + X) ∈ St(n, H)

by lemma 3.1, and so γ1 (t) ∈ j∈J (U (j) + St(n, H)).

T

Similarly, γ2 (t) ∈ j∈J (U (j) + St(n, H)) for all t ∈ [0, 1].
T

Composing γ1 with γ2 , we see that j∈J (U (j) + St(n, H)) is polygonally connected.
T

10
4 Topological closure of St(n, H) ∩ S
Before moving on, we need a definition and a small lemma.
Definition 4.1. Let V be a F-vector space, q ∈ N, and v, v ′ ∈ V . We say that
γ : [0, 1] → V is a polynomial path up to reparametrization joining v and v ′ if there
exist q ∈ N, a finite sequence of vectors (v k )k∈[[0,q]] with v k ∈ V for all k ∈ [[0, q]] and a
homeomorphism ϕ : [0, 1] → [a, b] ⊆ R such that ∀t ∈ [a, b] : γ(ϕ−1 (t)) = qk=0 tk v k ,
P

γ(0) = v and γ(1) = v ′ . If V is equipped with a topology, then we say that γ is a

continuous polynomial path when it is continuous as a map from [0, 1] to V .
Remark 4.1. Every expression of the form qk=0 Pk (t)v k where v k ∈ V and Pk ∈ F[X]
P
P ′
for all k ∈ [[0, q]] can be written in the form qk=0 tk wk where q ′ ∈ N and wk ∈ V
for all k ∈ [[0, q ′ ] (group by increasing powers of t). Therefore there is no difference
whether we define polynomial paths using the first expression or the second.
Lemma 4.1. Let E be a normed vector space, and v, v ′ ∈ E. Then each polynomial
path γ : [0, 1] → E up to reparametrization joining v and v ′ is continuous.
Proof. Let γ : [0, 1] → E be a polynomial path up to reparametrization joining v
and v ′ . Hence we can write ∀t ∈ [0, 1] : γ(t) = qk=0 ϕ(t)k v k . The continuity of γ
P

follows from q
∥γ(t) − γ(t′ )∥ ≤ ∥v k ∥|ϕ(t)k − ϕ(t′ )k )|
X

k=0

which goes to 0 when t goes to t′ by continuity of ϕk for all k ∈ [[0, q]].

The following is our first original proposition in this section.
Proposition 4.1. Let S ⊆ H n and

E := {γ : [0, 1] → S such that γ is a polynomial path up to reparametrization and

∃aγ ∈ [0, 1] : γ(aγ ) ∈ St(n, H) ∩ S}.

Range(γ) ⊆ St(n, H) ∩ S
S
Then γ∈E

Proof. Let γ ∈ E. We have ∀t ∈ [a, b] : γ(ϕ−1 (t)) = qk=0 tk V k .

P

Let Γ(t) := det(Gram((γ(ϕ−1 (t))1 , · · · , γ(ϕ−1 (t))n ))) for all t ∈ [a, b].
Since for all i, j ∈ [[1, n]] and t ∈ [a, b]
* q q +
⟨γ(ϕ (t))i , γ(ϕ (t))j ⟩ =
−1 −1
tk vik , tk vjk
X X

k=0 k=0
q
′ ′
D E
= vik , vjk tk+k
X

k,k′ =0

11
is a polynomial function in t ∈ [a, b], and the determinant of a matrix in Mn,n (F) is
a polynomial function in its coefficients, Γ(t) is a polynomial function in t ∈ [a, b]
which satisfies Γ(ϕ(aγ )) ̸= 0 since γ(aγ ) ∈ St(n, H). Therefore
Γ(t) ̸= 0 for t in a cofinite set L ⊆ [a, b].
Hence Range(γ)\{γ(ϕ−1 (t))}t∈[a,b]\L ⊆ St(n, H)∩S ⊆ St(n, H) ∩ S. Since [a, b]\L
is finite, the continuity of γ (see lemma 4.1) at {ϕ−1 (t))}t∈[a,b]\L implies

Range(γ) = Range(γ) = Range(γ) \ {γ(ϕ−1 (t))}t∈[a,b]\L ⊆ St(n, H) ∩ S.

This being true for all γ ∈ E, the result follows.
The following is a corollary.
Corollary 4.1. Let S ⊆ H n such that for all U ∈ S there exists a polynomial path
up to reparametrization connecting U to some Θ(U ) ∈ St(n, H) ∩ S and contained
in S. Then St(n, H) ∩ S is dense in S.
Proof. For all U ∈ S, there exists by the hypothesis γ ∈ E such that γ(0) = U .
By proposition 4.1, we have Range(γ) ⊆ St(n, H) ∩ S. It follows that U ∈
St(n, H) ∩ S since U ∈ Range(γ).
This result shows the abundance of independent n-frames not only in H n but
also in many subsets S of the form of corollary 4.1. Importantly, notice that for
S ̸= ∅, there should exist at least one Θ ∈ St(n, H) ∩ S (i.e. St(n, H) ∩ S ≠ ∅) for
the result to follow.
As an example, this is true when S is a star domain of H n with respect to some
Θ ∈ St(n, H)∩S; if it is a convex subset of H n and contains some Θ ∈ St(n, H)∩S;
and if it is in particular an affine subspace containing some Θ ∈ St(n, H) ∩ S.
In the next section, we will find sufficient conditions under which some sets
of the form f −1 ({d}) ⊆ L2 (X, µ; Fn ) where f is some linear or quadratic function
contain a continuous frame. This will allow us to apply corollary 4.1 to these
examples, which will be done in section 6.

5 Existence of solutions of some linear

and quadratic equations which are finite-
dimensional continuous frames
In this section, we show how to construct continuous finite-dimensional frames
that are mapped to a given element by a linear operator or a quadratic function.
In other words, we show the existence of frames in the inverse image of singletons
by these functions.

12
5.1 Linear equations
Proposition 5.1. Let n ∈ N∗ , V an F-vector field of dimension ≥ n and T : V n →
F a non-zero linear form. Then for all d ̸= 0, there exists (a1 , · · · , an ) ∈ V n such
that the system (a1 , · · · , an ) is free and T (a1 , · · · , an ) = d.
The proof relies on the following lemma, which may be of independent interest.
Lemma 5.1. Let n ∈ N∗ and V be a vector space of dimension ≥ n.
 (x1 , · · · , xn ) ∈ V \ {(0, · · · , 0)}. Then there exist e ∈ N with e =
n ∗
Let
1 if k = n
2 if k ∈ [[1, n − 1]]
where k = dim(span{x1 , · · · , xn }), and e independent sys-

tems (au1 , · · · , aun ) in V for u ∈ [1, e] such that (x1 , · · · , xn ) = eu=1 (au1 , · · · , aun ).
P

Proof. Without loss of generality, suppose that (x1 , · · · , xk ) is free where k =

dim(span{x1 , · · · , xn }) ≥ 1. Let (bi )i∈[[1,n−k]] be an (n − k)-tuple of vectors of V
such that the system (x1 , · · · , xk , b1 , · · · , bn−k ) is free.
• If k = n, then (x1 , · · · , xn ) is already free and so we can choose e = 1 and
a1i = xi for all i ∈ [[1, n]].

• If k ∈ [[1, n − 1]], we can write

x1 xk x1 xk
(x1 , · · · , xn ) = ( , · · · , , b1 , · · · , bn−k )+( , · · · , , xk+1 −b1 , · · · , xn −bn−k ).
2 2 2 2
( x21 , · · · , x2k , b1 , · · · , bn−k ) is free by assumption, and we can easily show that
( x21 , · · · , x2k , xk+1 − b1 , · · · , xn − bn−k ) is free by expressing xk+1 , · · · , xn in
terms of x1 , · · · , xk . Hence we can choose e = 2 and a1i = a2i = x2i for all
i ∈ [[1, k]], a1i = bi−k , and a2i = xi − bi−k for all i ∈ [[k + 1, n]].

Proof. (of proposition 5.1)

Let’s show that there exists a free system (a1 , · · · , an ) such that T (a1 , · · · , an ) ̸= 0.
Suppose to the contrary that T (a1 , · · · , an ) = 0 for all free systems (a1 , · · · , an ) in
V . Let (x1 , · · · , xn ) ∈ V n \ {(0, · · · , 0)}. By lemma 5.1, (x1 , · · · , xn ) decomposes
as a finite sum of free systems. By linearity of T , we thus have T (x1 , · · · , xn ) = 0.
Hence T is the zero form, a contradiction. Hence there exists a free system
(a1 , · · · , an ) such that T (a1 , · · · , an ) ̸= 0. Then T (a1 ,···
d
,an )
(a1 , · · · , an ) satisfies the
requirement.
Corollary 5.1. Let n ∈ N∗ and T : L2 (X, µ; Fn ) → F be a non-zero linear form.
Suppose that dim(L2 (X, µ; F)) ≥ n.
Then for all d ̸= 0, there exists a continuous frame Φ = (φx )x∈X such that T (Φ) = d.

13
Proposition 5.2. Let n ∈ N∗ , V be a vector space over F, and S : V → F be a
non-zero linear form.
Define the linear operator

V n → Fn
T : .
(x1 , · · · , xn ) 7→ (S(x1 ), · · · , S(xn ))
For all d ∈ Fn , suppose that
• if d ̸= 0, then dim(V ) ≥ n,
• if d = 0 then dim(V ) ≥ n + 1.
Then T −1 ({d}) contains an independent n-tuple (a1 , · · · , an ).
Proof. • Suppose that d ̸= 0 and dim(V ) ≥ n. Let h ∈ V such that S(h) = 1,
(h(2) , · · · , h(n) ) be an independent system in Ker(S) (this is possible since
codim(Ker(S)) = 1), and (d, d(2) , · · · , d(n) ) be an independent system in Fn .
 
h
 h(2) 
 
Consider the V -valued column matrix H =  ..  
 and the F-valued square
 . 
h(n)
 1
d1(2) · · · d1(n)
  
d a1
 . . . .  . 
matrix D =  . .. .. .. . Moreover, we set A =  ..   = DH.

 .
 
n n n
d d(2) · · · d(n) an
We have T (a1 , · · · , an ) = (S(a1 ), · · · , S(an )) = (S(di h +
k=2 d(k) h(k) ))i∈[[1,n]] = (d )i∈[[1,n]] = d since h(2) , · · · , h(n) belong to
Pn i i

Ker(S).
Let’s show that (a1 , · · · , an ) is free. Let λ = (λ1 , · · · , λn ) ∈ Fn
λ1
such that ni=1 λi ai = 0. Consider Λ =  .. 
 . . Therefore we have
P 

λn
0 = Λ⊤ A = Λ⊤ DH ∈ V . At this point, we can complete (h, h(2) , · · · , h(n) )
into a Hamel basis of V , and denote by (h∗ , h∗(2) , · · · , h∗(n) ) the first n linear
forms of its h dual basis. i Applying these linear forms to Λ⊤ DH, we see that
Λ⊤ D = 0 · · · 0 , therefore λ = 0 as D⊤ is obviously invertible since
(h, d(2) , · · · , d(n) ) is an independent system in Fn .
• Suppose that d = 0 and dim(V ) ≥ n + 1. Let h ∈ V such that S(h) = 1 and
(a1 , · · · , an ) be an independent system in Ker(S).
We have T (a1 , · · · , an ) = (S(a1 ), · · · , S(an )) = 0 ∈ Fn .
Moreover, we have by construction that (a1 , · · · , an ) is free.

14
Remark 5.1. In the previous proposition, if d = ̸ 0, the condition dim(V ) ≥ n is
necessary for the existence of an independent n-tuple because the existence of
an independent n-tuple implies that dim(V ) ≥ n, and if d = 0, the condition
dim(V ) ≥ n + 1 is also necessary for the existence of an independent n-tuple in
T −1 ({0} because the existence of an independent n-tuple (a1 , · · · , an ) in T −1 ({0}
implies that S(ai ) = 0 for all i ∈ [[1, n]], and since there exists h ∈ V such
that S(h) = 1 because S is non-zero, then (a1 , · · · , an , h) is free which implies
dim(V ) ≥ n + 1.
Corollary 5.2. Let h ∈ L2 (X, µ; F) ̸= 0. Define the linear operator

L2 (X, µ; Fn ) → Fn
T : .
= (fx )x∈X 7→ X h(x)fx dµ(x)
R
F

For all d ∈ Fn , suppose that

• if d ̸= 0, then dim(L2 (X, µ; F)) ≥ n,
• if d = 0 then dim(L2 (X, µ; F)) ≥ n + 1.
Then T −1 ({d}) contains a continuous frame Φ = (φx )x∈X .
Proposition 5.3. Let h ∈ L2 (X, µ; F) ̸= 0. Define the linear operator

L2 (X, µ; Fn ) → Fn
T : .
F = (fx )x∈X 7→ X h(x)fx dµ(x)
R

If there exists a measurable subset Y ⊂ X such that dim(L2 (Y, µ; F)) ≥ n and
µ((X \ Y ) ∩ h−1 (F∗ )) > 0, then for all d ∈ Fn , T −1 ({d}) contains a continuous
frame Φ = (φx )x∈X .
Proof. Since dim(L2 (Y, µ; F)) ≥ n, there exists a continuous frame (φy )y∈Y ∈
F(Y,µ),n
F
. We extend (φy )y∈Y by setting
h(x)
 Z 
φx := d− h(y)φy dµ(y) for all x ∈ X \ Y
∥h∥2L2 (X\Y,µ;F) Y

Let Φ = (φx )x∈X . We have

Z
T (Φ) = h(x)φx dµ(x)
X
 
Z Z
h(x) Z
= h(y)φy dµ(y) +  h(x) dµ(x) (d − h(y)φy dµ(y))
Y X\Y ∥h∥2L2 (X\Y,µ;F) Y

= d.
Moreover, Φ ∈ F(X,Σ,µ),n
F
since we have only completed (φy )y∈Y by a function in
L (X \ Y, µ; F ).
2 n

15
Proposition 5.4. Let (X, Σ, µ) be a measure space, l ∈ N∗ , (Xj )j∈[[1,l]] a partition
of X by measurable subsets, and h ∈ L2 (X, µ; F) such that there exist a family
(Yj )j∈[[1,l]] with Yj a measurable subset of Xj for all j ∈ [[1, l]], µ((Xj \Yj )∩h−1 (F∗ )) >
l
0 for all j ∈ [[1, l]], and dim(L2 (Yj , µ; F)) ≥ n.
P
j=1
Define the operator

L2 (X, µ; Fn ) → j∈[[1,l]] Fn
Q
W : .
F = (fx )x∈X 7 ( Xj h(x)fx dµ(x))j∈[[1,l]]
→
R

Then for all D = (dj )j∈[[1,l]] ∈ Fn , W −1 ({D}) contains at least one continuous
Q
j∈[[1,l]]
frame Φ ∈ F(X,Σ,µ),n
F
.

Remark 5.2. Proposition 5.3 results from proposition 5.4 by taking l = 1.

Remark 5.3. Proposition 5.4 can be generalized to l = +∞ or to partitions indexed
by a general index set J if we restrict to D = 0 (due to convergence issues).
Proof. For each i ∈ [[1, n]], let ei be the i-th vector of the standard basis of Fn .
l
Since dim(L2 (Yj , µ; F)) ≥ n, we can find distinct j1 , · · · , jr ∈ [[1, l]] such that for
P
j=1
each u ∈ [[1, r]], dim(L2 (Yju , µ; F)) ≥ 1 and ru=1 dim(L2 (Yju , µ; F)) ≥ n. Take a
P

partition P1 , · · · , Pr of {e1 , · · · , en } with |Pu | ≤ dim(L2 (Yju , µ; F)) for all u ∈ [[1, r]].
For all u ∈ [[1, r]], let (gup )p∈Pu be an orthonormal family in L2 (Yju , µ; F) and define
(φx )x∈Xju by
φy = gup (y)p for all y ∈ Yju
X

p∈Pu

and
h(x)
 Z 
φx := dju − h(y)φy dµ(y) for all x ∈ Xju \ Yju .
∥h∥2L2 (Xju \Yju ,µ;F) Y

For all j ∈
/ {ju : u ∈ [[1, r]]}, define (φx )x∈Xj by

φy = 0 for all y ∈ Yj

and
h(x)
φx := dj for all x ∈ Xj \ Yj .
∥h∥2L2 (Xj \Yj ,µ;F)

16
Let Φ = (φx )x∈X . We have
Z !
W (Φ) = h(x)φx dµ(x)
Xj j∈[[1,l]]

Z
= h(y)φy dµ(y)
Yj
  !
Z
h(x) Z
+ h(x) dµ(x) dj − h(y)φy dµ(y) 
Xj \Yj ∥h∥2L2 (Xj \Yj ,µ;F) Yj
j∈[[1,l]]

= (dj )j∈[[1,l]] = D.

Moreover, Φ ∈ F(X,Σ,µ),n
F
since
r Z Z
∀v ∈ Fn : ∥v∥2 = |⟨v, φx ⟩|2 dµ(x) ≤ |⟨v, φx ⟩|2 dµ(x)
X

u=1 Yju X

and
 
r
∥dju − Y h(y)φy dµ(y)∥2 
Z R
|⟨v, φx ⟩|2 dµ(x) ≤ ∥v∥2 +  ∥v∥2
X
2
X u=1 ∥h∥ L2 (Xju \Yju ,µ;F)
 
∥dj ∥2
+  ∥v∥2
X

j ∈{j
/ u :u∈[[1,r]]}
∥h∥2L2 (Xj \Yj ,µ;F)

5.2 A quadratic equation

Proposition 5.5. Let (X, Σ, µ) be a σ-finite measure space, b ∈ Cn \ {0}, ϵ > 0
and h ∈ L∞ (X, µ; C).
Consider 
L2 (X, µ; Cn ) → Cn
q: .
F = (fx )x∈X 7→
X h(x)⟨b, fx ⟩fx dµ(x)
R

Then for all d ∈ Cn such that

• if d ̸= 0, then there exists a measurable subset Y ⊆ X, two measurable subsets

B1 ⊆ {z ∈ C : Re(⟨b, d⟩z) > ϵ and Im(⟨b, d⟩z) < −ϵ} and B2 ⊆ {z ∈ C :
Re(⟨b, d⟩z) > ϵ and Im(⟨b, d⟩z) > ϵ} such that dim(L2 (Y, µ; C)) ≥ n and
µ((X \ Y ) ∩ h−1 (B1 )), µ((X \ Y ) ∩ h−1 (B2 )) > 0,

17
 • if d = 0, then there exist a measurable subset Y ⊆ X such that
dim(L2 (Y, µ; C)) ≥ n and h(x) < 0 µ-almost everywhere on Y , and [(two
measurable subsets B1 ⊆ {z ∈ C : Re(z) > 0 and Im(z) < 0} and
B2 ⊆ {z ∈ C : Re(z) > 0 and Im(z) > 0} such that µ((X \ Y ) ∩
h−1 (B1 )), µ((X \ Y ) ∩ h−1 (B2 )) > 0) or (a measurable subset B3 ⊆ {z ∈ C :
Re(z) > 0 and Im(z) = 0} such that µ((X \ Y ) ∩ h−1 (B3 )) > 0)],

there exists a continuous frame Φ = (φx )x∈X ∈ q −1 ({d}).

Proof. • Suppose d = ̸ 0. Let B f = (X \ Y ) ∩ h−1 (B ) and B

1 1
f = (X \
2
Y ) ∩ h (B2 ). There is no loss in generality in assuming that µ(B
−1 f)
1
and µ(B2 ) are finite since µ is σ-finite. Let 0 < a < ∥h∥2 ∥b∥2 . Since
f ϵ
∞
dim(L2 (Y, µ; C)) ≥ n, we can pick an a-tight frame (φy )y∈Y ∈ F(Y,µ),n)C
.
Let h(x) = (⟨b, d⟩ − Y h(y)|⟨b, φy ⟩|2 dµ(y)) h(x) for all x ∈ X. Notice that
e R

we have
Re(h(x))
e >0 µ − almost everywhere on B f,
1

Im(h(x))
e <0 µ − almost everywhere on B
f,
1

Re(h(x))
e >0 µ − almost everywhere on B
f,
2

and
Im(h(x))
e >0 µ − almost everywhere on B
f.
2

Let
1
A= ⟨− Im(h̃),Re(h̃)⟩L2 (B˜ ⟨Im(h̃),Re(h̃)⟩L2 (B˜
>0
∥Im(h̃)∥2 2
1 ,µ;C)
+ ∥Im(h̃)∥2 2
2 ,µ;C)
˜
L (B1,µ;C) ˜
L (B2,µ;C)

and
q q
√ − Im(h(x))
e √ Im(h(x))
e
g(x) = A 1B
f1 (x) + A 1B
f2 (x)
∥Im(h)∥
e
L2 (B
f1 ,µ;C) ∥Im(h)∥
e
L2 (B
f2 ,µ;C)

for all x ∈ X \ Y .
Then it is easily seen that
Z ! Z 
2
h(x)|g(x)| dµ(x) −⟨b, d⟩ + h(y)|⟨b, φy ⟩|2 dµ(y) = −1. (1)
X\Y Y

Consider (φx )x∈X\Y defined by φx = g(x) (−d + Y h(y)⟨b, φy ⟩φy dµ(y)) for
R

all x ∈ X \ Y . Then Φ = (φx )x∈X ∈ F(X,Σ,µ),n

C
since we have only completed

18
 (φy )y∈Y by a function in L2 (X \ Y, µ; Cn ) . Moreover
Z
q(Φ) = h(x)⟨b, φx ⟩φx dµ(x)
ZX
= h(y)⟨b, φy ⟩φy dµ(y)
Y
Z  Z 
+ h(x)⟨b, g(x) −d + h(y)⟨b, φy ⟩φy dµ(y) ⟩
X\Y Y
 Z 
.g(x) −d + h(y)⟨b, φy ⟩φy dµ(y) dµ(x)
Y
Z
= h(y)⟨b, φy ⟩φy dµ(y)
Y
"Z  Z  #
− h(x)|g(x)| 2
−⟨b, d⟩ + 2
h(y)|⟨b, φy ⟩| dµ(y) dµ(x) d
X\Y Y
| {z }
=−1
"Z  Z  #
+ h(x)|g(x)| 2
−⟨b, d⟩ + 2
h(y)|⟨b, φy ⟩| dµ(y) dµ(x)
X\Y Y
| {z }
=−1
Z 
. h(y)⟨b, φy ⟩φy dµ(y)
Y
=d

using equality 1.

• Suppose d = 0. Let B f = (X \ Y ) ∩ h−1 (B ) and B

1 1
f = (X \ Y ) ∩ h−1 (B ).
2 2
Suppose first that µ(B f ), µ(B
1
f ) > 0. There is no loss in generality
2
in assuming that µ(B f ) and µ(B
1
f ) are finite since µ is σ-finite. Since
2
dim(L (Y, µ; C)) ≥ n, we can pick a frame (φy )y∈Y ∈ F(Y,µ),n
2 C
. Let
h(x) = − ( Y h(y)|⟨b, φy ⟩| dµ(y)) h(x) for all x ∈ X. Notice that we have
2
e R

Re(h(x))
e >0 µ − almost everywhere on B
f,
1

Im(h(x))
e <0 µ − almost everywhere on B
f,
1

Re(h(x))
e >0 µ − almost everywhere on B
f,
2

and
Im(h(x))
e >0 µ − almost everywhere on B
f.
2

Let
1
A= ⟨− Im(h̃),Re(h̃)⟩L2 (B˜ ⟨Im(h̃),Re(h̃)⟩L2 (B˜
>0
∥Im(h̃)∥2 2
1 ,µ;C)
+ ∥Im(h̃)∥2 2
2 ,µ;C)
˜
L (B1,µ;C) ˜
L (B2,µ;C)

19
and
q q
√ − Im(h(x))
e √ Im(h(x))
e
g(x) = A 1B
f1 (x) + A 1B
f2 (x)
∥Im(h)∥
e
L2 (B
f1 ,µ;C) ∥Im(h)∥
e
L2 (B
f2 ,µ;C)

for all x ∈ X \ Y .
Then it is easily seen that
Z ! Z 
h(x)|g(x)| dµ(x) 2
h(y)|⟨b, φy ⟩| dµ(y) = −1.
2
(2)
X\Y Y

Consider (φx )x∈X\Y defined by φx = g(x) Y h(y)⟨b, φy ⟩φy dµ(y) for all x ∈
R

X \ Y . Then Φ = (φx )x∈X ∈ F(X,Σ,µ),n

C
since we have only completed (φy )y∈Y
by a function in L (X \ Y, µ; C ) . Moreover
2 n

Z
q(Φ) = h(x)⟨b, φx ⟩φx dµ(x)
ZX
= h(y)⟨b, φy ⟩φy dµ(y)
Y
Z Z 
+ h(x)⟨b, g(x) h(y)⟨b, φy ⟩φy dµ(y) ⟩
X\Y Y
Z 
.g(x) h(y)⟨b, φy ⟩φy dµ(y) dµ(x)
Y
Z
= h(y)⟨b, φy ⟩φy dµ(y)
Y
"Z Z  #
+ h(x)|g(x)| 2 2
h(y)|⟨b, φy ⟩| dµ(y) dµ(x)
X\Y Y
| {z }
=−1
Z 
. h(y)⟨b, φy ⟩φy dµ(y)
Y
=0
using equality 2.
Now let Bf = (X \ Y ) ∩ h−1 (B ) and suppose instead that µ(B
3 3
f ) > 0. There
3
is no loss in generality in assuming that µ(B3 ) is finite since µ is σ-finite.
f
Since dim(L2 (Y, µ; C)) ≥ n, we can pick a frame (φy )y∈Y ∈ F(Y,µ),nC
. Let
= − ( Y h(y)|⟨b, φy ⟩|2 dµ(y)) h(x) for all x ∈ X. Notice that we have
R
h(x)
e

h(x)
e >0 µ − almost everywhere on B
f.
3

Let q
h(x)
e
g(x) = e 1B
f3 (x) for all x ∈ X \ Y
∥h∥L2 (B
f3 ,µ;C)

20
 Then it is easily seen that
Z ! Z 
2
h(x)|g(x)| dµ(x) h(y)|⟨b, φy ⟩| dµ(y) = −1.
2
(3)
X\Y Y

Consider (φx )x∈X\Y defined by φx = g(x) Y h(y)⟨b, φy ⟩φy dµ(y) for all x ∈
R

X \ Y . Then Φ = (φx )x∈X ∈ F(X,Σ,µ),n

C
since we have only completed (φy )y∈Y
by a function in L (X \ Y, µ; C ) . Moreover we can prove that q(Φ) = 0 as
2 n

before using equality 3.

6 Examples of subspsaces in which the space of

continuous frames is relatively dense
Corollary 6.1. Let n ∈ N∗ and T : L2 (X, µ; Fn ) → F be a non-zero linear form.
Suppose that dim(L2 (X, µ; F)) ≥ n.
Then since T −1 ({d}) is affine and by corollaries 4.1 and 5.1, for all d = ̸ 0,
F(X,Σ,µ),n ∩ T ({d}) is dense in T ({d})
F −1 −1

Corollary 6.2. Let (X, Σ, µ) be a measure space, d ∈ Cn , and h ∈ L2 (X, µ; F)

such that T −1 ({d}) contains a continuous frame Φ = (φx )x∈X (see for instance
corollary 5.2 and proposition 5.3), where

L2 (X, µ; Fn ) → Fn
T : .
F = (fx )x∈X 7→ X h(x)fx dµ(x)
R

Then since T −1 ({d}) is affine, by corollary 4.1, F(X,Σ,µ),n

F
∩ T −1 ({d}) is dense in
T −1 ({d}).
Corollary 6.3. Let (X, Σ, µ) be a measure space, l ∈ N∗ , (Xj )j∈[[1,l]] a partition
of X by measurable subsets, h ∈ L2 (X, µ; F), and D ∈ Fn such that W −1 ({D})
contains a continuous frame Φ = (φx )x∈X (see for instance proposition 5.4), where

L2 (X, µ; Fn ) → j∈[[1,l]] Fn
Q
W : .
F = (fx )x∈X 7 ( Xj h(x)fx )j∈[[1,l]]
→
R

Then since since W −1 ({d}) is affine, and by corollary 4.1, F(X,Σ,µ),n

F
∩ W −1 ({D})
is dense in W ({D}).
−1

Remark 6.1. Consider the function q of proposition 5.5. If Φ = (φx )x∈X ∈ q −1 ({0}),
then we also have Φ ∈ q −1 ({0}) ∩ (q −1 ({0}) − Φ) since q(2Φ) = 0.

21
Proposition 6.1. Consider the function q of proposition 5.5. Let Φ = (φx )x∈X ∈
q −1 ({0}) and U = (ux )x∈X ∈ q −1 ({0}) ∩ (q −1 ({0}) − Φ). Then for all λ, µ ∈ R,
λΦ + µU ∈ q −1 ({0}) ∩ (q −1 ({0}) − Φ).
In particular, q −1 ({0}) ∩ (q −1 ({0}) − Φ) is a star domain relatively to Φ.

Proof.
 Let λ, µ ∈ R. Let s be the sesquilinear form
L2 (X, µ; Cn ) → Cn
(F, G)
. We have
7→ X ⟨b, gx ⟩fx dµ(x)
R

q(λΦ + µU ) = λ2 q(Φ) +λµ(s(Φ, U ) + s(U, Φ)) + µ2 q(U )

| {z } | {z }
0 0
= λµ(q(Φ + U ) − q(Φ) − q(U ))
| {z } | {z } | {z }
0 0 0
= 0.

We can show similarly that q((λ + 1)Φ + µU ) = 0.

Corollary 6.4. Let (X, Σ, µ) be a σ-finite measure space, b ∈ Cn \ {0}, and

h ∈ L∞ (X, µ; C) such that q −1 ({0}) contains a continuous frame Φ = (φx )x∈X (see
for instance proposition 5.5), where

L2 (X, µ; Cn ) → Cn
q: .
F = (fx )x∈X 7→ X h(x)⟨b, fx ⟩fx dµ(x)
R

Then by proposition 6.1 and corollary 4.1, F(X,Σ,µ),n

C
∩ q −1 ({0}) ∩ (q −1 ({0}) − Φ) is
dense in q −1 ({0}) ∩ (q −1 ({0}) − Φ).

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.

22
References
[1] D. Agrawal. The complete structure of linear and nonlinear deformations of
frames on a hilbert space. Master’s thesis, 2016.

[2] D. Agrawal and J. Knisley. Fiber bundles and parseval frames, 2015.
arxiv:1512.03989.

[3] S. T. Ali, J. P. Antoine, and J. P. Gazeau. Continuous frames in hilbert space.

Annals of physics, 222(1):1–37, 1993.

[4] R. Balan, P. G. Casazza, C. Heil, and Z. Landau. Density, overcomplete-

ness, and localization of frames. i. theory. Journal of Fourier Analysis and
Applications, 12(2):105–143, 2006.

[5] R. Balan, P. G. Casazza, C. Heil, and Z. Landau. Density, overcompleteness,

and localization of frames. ii. gabor systems. Journal of Fourier Analysis and
Applications, 12(3):307–344, 2006.

[6] E. Bardelli and A. C. G. Mennucci. Probability measures on infinite-

dimensional stiefel manifolds. Journal of Geometric Mechanics, 9(3):291–316,
2017.

[7] M. Bownik. Connectivity and density in the set of framelets. Mathematical

Research Letters, 14(2):285–293, 2017.

[8] J. Cahill, D. Mixon, and N. Strawn. Connectivity and irreducibility of algebraic

varieties of finite unit norm tight frames. SIAM Journal on Applied Algebra
and Geometry, 1(1):38–72, 2017.

[9] P. G. Casazza. The art of frame theory. Taiwanese Journal of Mathematics,

4(2):129–201, 2000.

[10] P. G. Casazza, D. Han, and D. R. Larson. Frames for banach spaces. Contem-
porary Mathematics, 247:149–182, 1999.

[11] P. G. Casazza and G. Kutyniok. Finite frames, theory and applications.

Applied and Numerical Harmonic Analysis. Birkhäuser/Springer, 2013.

[12] O. Christensen. An Introduction to Frames and Riesz Bases. Applied and

Numerical Harmonic Analysis. Birkhäuser/Springer, second edition, 2016.

[13] O. Christensen, B. Deng, and C. Heil. Density of gabor frames. Applied and
Computational Harmonic Analysis, 7(3):292–304, 1999.

23
[14] I. Daubechies, A. Grossmann, and Y. Meyer. Painless nonorthogonal expan-
sions. J. Math. Phys., 27(5):1271–1283, 1986.

[15] R. J. Duffin and A. C. Schaeffer. A class of nonharmonic fourier series.

Transactions of the American Mathematical Society, 72(2):341–366, 1952.

[16] K. Dykema and N. Strawn. Manifold structure of spaces of spherical tight

frames. Int. J. Pure Appl. Math., 28(2):217–256, 2006.

[17] M. Frank and D. R. Larson. Frames in hilbert c*-modules and c*-algebras.

Journal of Operator Theory, 48(2):273–314, 2002.

[18] D. Gabor. Theory of communications. Jour. Inst. Elec. Eng. (London),

93(3):429–457, 1946.

[19] G. Garrigos, E. Hernandez, H. Šikić, F. Soria, G. Weiss, and E. Wilson.

Connectivity in the set of tight frame wavelets (tfw). Glasnik matematički,
38(1):75–98, 2003.

[20] D. Han and D. R. Larson. On the orthogonality of frames and the density and
connectivity of wavelet frames. Acta Applicandae Mathematicae, 107(1-3):211–
222, 2009.

[21] P. Harms and A. C. G. Mennucci. Geodesics in infinite dimensional stiefel and

grassmann manifolds. Comptes Rendus Mathematique, 350(15-16):773–776,
2012.

[22] A. Hatcher. Algebraic topology. Cambridge University Press, 2005.

[23] C. Heil. History and evolution of the density theorem for gabor frames. Journal
of Fourier Analysis and Applications, 13(2):113–166, 2007.

[24] O. Henkel. Sphere-packing bounds in the grassmann and stiefel manifolds.

IEEE Transactions on Information Theory, 51(10):3445–3456, 2005.

[25] D. Husemoller. Fibre Bundles. Springer-Verlag, third edition, 1994.

[26] I. M. James. The topology of Stiefel manifolds. Cambridge University Press,

1976.

[27] V. Jurdjevic, I. Markina, and F. S. Leite. Extremal curves on stiefel and

grassmann manifolds. J. Geom. Anal., 30(4):3948–3978, 2020.

[28] D. Labate and E. Wilson. Connectivity in the set of gabor frames. Applied
and Computational Harmonic Analysis, 18(1):123–136, 2005.

24
[29] T. Needham and C. Shonkwiler. Symplectic geometry and connectivity of
spaces of frames. Advances in Computational Mathematics, 47(1 : 5), 2021.

[30] D. M. Speegle. The s-elementary wavelets are path-connected. Proc. Amer.

Math. Soc., 127:223–233, 1999.

[31] N. Strawn. Geometry and constructions of finite frames. Master’s thesis, 2007.

[32] N. Strawn. Finite frame varieties: nonsingular points, tangent spaces, and
explicit local parameterizations. J. Fourier Anal. Appl., 17(5):821–853, 2011.

[33] N. Strawn. Geometric structures and optimization on spaces of finite frames.

PhD thesis, 2011.

[34] S. F. D. Waldron. An introduction to finite tight frames. Applied and Numerical

Harmonic Analysis. Birkhäuser/Springer, 2018.

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]
Pr. Brahim Moalige.
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]

25