Journal of Fourier Analysis and Applications (2025) 31:39

https://doi.org/10.1007/s00041-025-10174-7

The Duality Principle and Biorthogonality Relation on

Quaternionic Gabor Systems

Xiao-Li Zhang1,2 · Yun-Zhang Li1

Received: 31 January 2024 / Revised: 12 February 2025 / Accepted: 18 April 2025

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025

Abstract
In recent years, quaternionic Fourier analysis has attracted increasing interest due to
its potential application in signal analysis and image processing. This paper addresses
the duality principle and biorthogonality relation on quaternionic Gabor systems. It is
well known that the Ron–Shen duality principle and the Wexler–Raz biorthogonality
relation play an important role in traditional Gabor analysis. We in this paper show
that neither the Ron–Shen duality principle nor the traditional Wexler–Raz biorthogo-
nality relation holds for quaternionic Gabor systems. And under the condition that the
products of time-frequency shift parameters are rational numbers, we characterize the
quaternionic Gabor Riesz sequences, present a sufficient condition on “G(g, α, β) is a
frame for L 2 (R2 , H)⇔G(g, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H)", and establish
a biorthogonality relation on quaternionic Gabor systems.

Keywords Frame · Riesz sequence · Quaternionic Gabor frame · Duality principle ·

Biorthogonality relation

Mathematics Subject Classification 42C15 · 42C40

Communicated by Maurice De Gosson.

Supported by National Natural Science Foundation of China (Grant No. 12371091) and Science Research
Project of Hebei Education Department (Grant No. QN2025067).

B Yun-Zhang Li
yzlee@bjut.edu.cn
Xiao-Li Zhang
zhangxiaolistudy@163.com

1 School of Mathematics, Statistics and Mechanics, Beijing University of Technology, Beijing

100124, People’s Republic of China
2 College of Statistics and Mathematics, Hebei University of Economics and Business, Shijiazhuang
050061, People’s Republic of China

0123456789().: V,-vol
 39 Page 2 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

1 Introduction

In the past several decades, Gabor analysis in L 2 (R) has seen great achievements in
theory and application. This paper addresses quaternionic Gabor frames in the setting
of L 2 (R2 , H), where H is the quaternion algebra: H = {η | η = η0 + iη1 + jη2 +
kη3 , η0 , η1 , η2 , η3 ∈ R} with multiplication rule

i j = − ji = k, i 2 = j 2 = k 2 = i jk = −1,

L 2 (R2 , H) denotes the space of all square-integrable quaternion-valued functions

f : R2 → H with finite norm

 1
2
 f  L 2 (R2 , H) = | f (x)| dx
2
< ∞, (1.1)
R2

where dx = d x1 d x2 is the usual Lebesgue measure in R2 with x = (x1 , x2 ).

Obviously, the norm (1.1) can be induced by both

( f , g) L 2 (R2 , H) = f (x)g(x)dx for f , g ∈ L 2 (R2 , H) (1.2)
R2

and 
 f , g L 2 (R2 , H) = Sc f (x)g(x)dx for f , g ∈ L 2 (R2 , H), (1.3)
R2

where
Sc[η] = η0 for η = η0 + iη1 + jη2 + kη3 ∈ H,
 
it is called the scalar part of η. It is worth noting that L 2 (R2 , H), (·, ·) is a left
 2 2 
Hilbert module but not a Hilbert space, and that L (R , H), ·, · is a real Hilbert
space. Given η = η0 + iη1 + jη2 + kη3 ∈ H, denote its vectorial part, conjugate
 by V ec[η] = iη1 + jη2 + kη3 , η = η0 − iη1 − jη2 − kη3 and |η| =
and modulus
√
η η = η02 + η12 + η22 + η32 , respectively. And write Cr (η) and Cl (η) for the right
carrier and left carrier operators given by

Cr (η)ξ = ξ η and ξ Cl (η) = ηξ (1.4)

for ξ ∈ H, respectively. Recall that

ξ η = η ξ for η, ξ ∈ H.

Formally take conjugates to (1.4), we have

ξ Cr (η) = ξ Cl (η) and Cl (η) ξ = Cr (η)ξ for ξ ∈ H.

Journal of Fourier Analysis and Applications (2025) 31:39 Page 3 of 29 39

Therefore, Cr (η) and Cl (η) are understood as Cl (η) and Cr (η), respectively. This idea
was adopted by [15]. The following proposition collects the properties of quaternions
and carriers, which was frequently used in [15, 17, 19–21, 30, 31, 37, 45].

Proposition 1.1 (i) Cr (η) = Cl (η) and Cl (η) = Cr (η) for η ∈ H;

(ii) Cr (η)1 = 1Cl (η) = η and Cr (ξ )1Cl (η) = ηξ for η, ξ ∈ H;
(iii) η1 η2 Cr (η3 ) = η3 η1 η2 for η1 , η2 , η3 ∈ H;
(iv) Sc[η γ ξ ] = Sc[γ ξ η] = Sc[ξ η γ ] for η, γ , ξ ∈ H.

By [30, Remark 1.2], given two measurable quaternion-valued functions g and h

satisfying hg ∈ L 2 (R2 , H), gCr (h) is identified with hg in the sense that

 f , gCr (h) L 2 (R2 , H) =  f , hg L 2 (R2 , H) for f ∈ L 2 (R2 , H).

It is well known that H is a noncommutative associative algebra. Noncommutativ-

ity limits the flexibility of computation. This makes the study of quaternionic Fourier
analysis nontrivial and challenging. The quaternionic Fourier transform (QFT) was
first proposed by Ell [12, 13]. In recent years, it has received increasing attention due
to its potential application in signal analysis and image processing [2–4, 22, 32, 33,
35, 36]. To better represent and compute local information of signals, the windowed
quaternionic Fourier transform (WQFT) and quaternionic Gabor systems were intro-
duced and studied [1, 6, 11, 15, 19–21, 24, 26, 31, 45]. Cerejeiras, Hartmann and
Orelma [6] studied the Wexler–Raz biorthogonality, Janssen’s and Walnut’s represen-
tations on quaternionic
 Gabor systems,
 and the duality principle on tight quaternionic
Gabor frames in L 2 (R2 , H), (·, ·) , where the window functions are real-valued.
Some results of [6] are generalized to the case of quaternion-valued window functions
by Li and He [26]. The right quaternionic Fourier transform (RQFT) was also intro-
duced in [12, 13, 22]. Recall that, given a function, the switch between the Fourier
transform of its translation (modulation) and the modulation (translation) of its Fourier
transform plays a significant role in Fourier analysis. From [19] we know that QFT
admits this switch by defining suitable modulation operator,  but RQFT does not.
And from [22], for QFT, the Plancherel theorem holds in L 2 (R2 , H), ·, · , but
 
it does not hold in L 2 (R2 , H), (·, ·) . That explains why Fourier analysis associ-
 2 2   
ated with QFT in L (R , H), ·, · are more studied than in L 2 (R2 , H), (·, ·) .
A Gabor system  is a translation-and-modulation
  system. Therefore,  studying Gabor
analysis in L 2 (R2 , H), ·, · instead of in L 2 (R2 , H), (·, ·) should be a wise
choice. For convenience,
 throughout the rest of this paper, L 2 (R2 , H) always denotes
L (R , H), ·, · .
2 2

Given g ∈ L 2 (R2 , H) and α, β > 0, we denote by G(g, α, β) the quaternionic

Gabor system:


(η ,η )
G(g, α, β) = Mβm1 2 Tαn g : m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} , (1.5)
 39 Page 4 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

(η1 ,η2 )
where the translation operator Ta and the modulation operator Mb with a, b ∈ R2
and (η1 , η2 ) ∈ {1, i} × {1, j} are defined by

(η1 ,η2 )
Ta f (x) = f (x − a) and Mb f (x) = η2 e2π jb2 x2 f (x)Cr (η1 e2πib1 x1 )

for f ∈ L 2 (R2 , H) and a.e. x ∈ R2 , respectively. A Gabor system G(g, α, β) is called

a frame for L 2 (R2 , H) if there exist constants A, B > 0 such that

(η ,η2 ) 2
A f 2 ≤  f , Mβm1 Tαn g L 2 (R2 , H) ≤ B f 2 (1.6)
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2

for f ∈ L 2 (R2 , H), where A and B are called frame bounds. It is called a Bessel
sequence in L 2 (R2 , H) if the right-hand side of inequality (1.6) holds. Define 2 (Z2 ×
Z2 × {1, i} × {1, j}, R) as the real Hilbert space:

2 (Z2 × Z2 × {1, i} × {1, j}, R)


(η1 , η2 ) (η , η )
= c = cm,n : m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} : cm,1n 2 ∈ R
⎛ ⎞1
2 
(η , η )
and c = ⎝ |cm,1n 2 |2 ⎠ <∞
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2

with inner product

(η , η2 ) (η1 , η2 )
c, d = cm,1n dm, n for c, d ∈ 2 (Z2 × Z2 × {1, i} × {1, j}, R).
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2

We denote by 0 (Z2 × Z2 × {1, i} × {1, j}, R) the set of all finitely supported
sequences in 2 (Z2 × Z2 × {1, i} × {1, j}, R). Obviously, it is dense in 2 (Z2 × Z2 ×
{1, i} × {1, j}, R). G(g, α, β) is called a Riesz sequence in L 2 (R2 , H) if there exist
constants A, B > 0 such that
 2
 
 (η1 ,η2 ) (η1 ,η2 ) 

Ac ≤ 
2
cm, n Mβm Tαn g  ≤ Bc2

(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2  2
L (R2 , H)

for c ∈ 0 (Z2 × Z2 × {1, i} × {1, j}, R).

Given a Bessel sequence G(g, α, β) in L 2 (R2 , H), its synthesis operator Tg :
 (Z2 × Z2 × {1, i} × {1, j}, R) → L 2 (R2 , H) and analysis operator Tg∗ :
2

L 2 (R2 , H) → 2 (Z2 × Z2 × {1, i} × {1, j}, R) are defined by

(η ,η ) (η ,η2 )
Tg c = cm,1n 2 Mβm1 Tαn g
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2
Journal of Fourier Analysis and Applications (2025) 31:39 Page 5 of 29 39

and

  
(η ,η )
Tg∗ f = f, Mβm1 2 Tαn g 2 2 : m, n ∈ Z , (η1 , η2 ) ∈ {1, i} × {1, j}
2
L (R , H)

for f ∈ L 2 (R2 , H) and c ∈ 2 (Z2 × Z2 × {1, i} × {1, j}, R), respectively. Given a
frame G(g, α, β) for L 2 (R2 , H), another frame G(h, α, β) is called a dual frame of
G(g, α, β) for L 2 (R2 , H) if

Sg, h = Tg Th∗ = I on L 2 (R2 , H).

It is easy to check that G(g, α, β) is also a dual of G(h, α, β) if G(h, α, β) is a dual of

G(g, α, β) . So, in this case, we say (G(g, α, β), G(h, α, β)) is a pair of dual frames
or a dual frame pair.
 Fu, Kähler and Cerejeiras [15] and Hartmann [19–21] studied the system  of the form
(1,1) (1,1)
M m Tαn g : m, n ∈ Z in L (R , H). However, every system M m Tαn g :
2 2 2
α α

m, n ∈ Z2 cannot be a frame for L 2 (R2 , H) since it is incomplete in L 2 (R2 , H) by

[31, Remark 3.2].
We investigate the quaternionic Gabor systems of the form (1.5) under the following
general setup:
Generalsetup : α, β > 0, and αβ = qp with p and q being coprime positive
integers.
This general setup is a technical condition that reduces studying frame properties
of Gabor systems to designing matrix-valued functions of finite order.
Under the general setup, we in [45] presented a density theorem that the products
of time-frequency shift parameters being less than or equal to 1 is necessary and
sufficient for the existence of complete quaternionic Gabor systems (quaternionic
Gabor frames) in L 2 (R2 , H); and in [31] characterized when two quaternionic Gabor
Bessel sequences in L 2 (R2 , H) form a pair of dual Gabor frames, and showed that,
given a overcomplete quaternionic Gabor frame for L 2 (R2 , H), its canonical dual
frame need not have Gabor structure, but it admits other dual frames with Gabor
structure. Recall that the canonical dual frame of a Gabor frame for L 2 (R) has the
Gabor structure. This demonstrates that quaternionic Gabor analysis is essentially
different from Gabor analysis in L 2 (R).
Let us first recall the duality principle and the Wexler–Raz biorthogonality relation
in Gabor analysis on L 2 (Rd ). Write

G(g, a, b) = {e2πibm·x g(x − an) : m, n ∈ Zd } for g ∈ L 2 (Rd ), a, b > 0.

The duality principle which is so called the Ron–Shen duality principle states that
G(g, a, b) is a frame for L 2 (Rd ) if and only if G(g, b1 , a1 ) is a Riesz sequence in
L 2 (Rd ) [9, 16, 23, 34]. And the Wexler–Raz biorthogonality relations [42] (see also
[16, Theorem 7.3.1]) states that two Bessel sequences G(g, a, b) and G(h, a, b) form
 39 Page 6 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

a pair of dual frames for L 2 (Rd ) if and only if

 m· n 
(ab)−d h(·), e2πi a g(· − ) = δm,0 δn,0 for m, n ∈ Zd .
b

They inspired follow-up research of various duality principles (associated with “R"-
duals) in abstract space frame theory [5, 7, 8, 10, 14, 27–29, 38–40].
Motivated by the above works, this paper addresses the duality principle and
biorthogonality relation on quaternionic Gabor systems in L 2 (R2 , H) under the gen-
eral setup. In Sect. 2, using the quaternionic Zak transformation matrix techniques, we
characterize the quaternionic Gabor Riesz sequences and orthonormal sequences. In
Sect. 3, we prove that the Ron–Shen duality principle dose not hold for quaternionic
Gabor systems in L 2 (R2 , H), and present a sufficient condition on “G(g, α, β) is a
frame for L 2 (R2 , H) if and only if G(g, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H)".
In Sect. 4, we show that traditional Wexler–Raz biorthogonality relation does not
hold for quaternionic Gabor systems, and then establish a biorthogonality relation on
quaternionic Gabor systems.
Before proceeding, we list some notations and notions for later use. For n ∈ N, we
denote by In the n × n identity matrix, by Nn the set {0, 1, · · · , n − 1}, by N2n the
set Nn × Nn , by car d( ) the cardinality of a subset in Z2 , by Q μ the set [0, μ)2
for μ > 0 (write Q = Q 1 for simplicity), and by | | its measure for a Lebesgue
measurable set in Rn . Let F = C or H and be a Lebesgue measurable set in Rn
with | | > 0, we denote by L 2 ( , Fm ) the real Hilbert space:
⎧  1 ⎫
⎨ m 2 ⎬
L 2 ( , Fm ) = f : → Fm :  f  L 2 ( , Fm ) = | f s (x)|2 dx <∞
⎩ ⎭
s=1

endowed with the symmetric real scalar inner product

 m
 f , g L 2 ( , Fm ) = Sc f s (x)gs (x)dx for f , g ∈ L 2 ( , Fm ), (1.7)
s=1

where dx = d x1 d x2 · · · d xn represents the usual Lebesgue measure in Rn with

x = (x1 , x2 , · · · , xn ), f s and gs denote the s-th component of f and g, respec-
tively. We denote by L ∞ ( , Fm ) the set of all essentially bounded functions on , by
2 (Rn , Fm ) the set of locally square-integrable functions from Rn into Fm . Through-
L loc
out this paper, L 2 ( , Fm ) is always referred to the real Hilbert space with the inner
product (1.7).
Given m, n ∈ N, we denote by Mm×n (H) (Mm×n (C)) the collection of all m × n
matrices with quaternion (complex) entries. For A = (ar , s )r ∈Nm , s∈Nn ∈ Mm×n (H),
we denote by A, AT and A∗ its conjugate, transpose, and conjugate transpose,
respectively, and define

ξ A = (ξ ar , s ) and Aξ = (ar , s ξ ) for ξ ∈ H.

Journal of Fourier Analysis and Applications (2025) 31:39 Page 7 of 29 39

Let A = A1 + A2 j with A1 , A2 ∈ Mm×n (C). XA denotes the complex adjoint matrix

of A:
 
A1 A2
XA = .
−A2 A1

It is easy to check that

  2
 η1 
Aη2 =  X 
 A −η2  for η = η1 + η2 j ∈ H with η1 , η2 ∈ C
n n
(1.8)

and
XA∗ = (XA )∗ . (1.9)
)∗ X∗.
For simplicity, write (XA = A One refers to [43, 44] for the fundamentals of
quaternion matrices. By [44, Theorem 4.3], for A ∈ Mn×n (H), A is invertible if and
only if {η ∈ Hn : Aη = 0} = {0}. By [25] and [44, Theorem 4.2],

XAB = XA XB for A ∈ Mm×n (H), B ∈ Mn×t (H). (1.10)

Recall from [45, Definition 3.1] that the quaternionic Zak transform 
Zαq from
L 2 (R2 , H) into L loc
2 (R4 , H) is defined by


Zαq f (x, ω) = e2π jm 2 ω2 f (x − αqm)e2πim 1 ω1
m∈Z2

for f ∈ L 2 (R2 , H) and a.e. (x, ω) ∈ R4 . Then 

Zαq has the quasi-periodicity:


Zαq f (x, ω+n) =  Zαq f (x+αqn, ω) = e2π jn 2 ω2
Zαq f (x, ω);  Zαq f (x, ω)e2πin 1 ω1

for n ∈ Z2 , f ∈ L 2 (R2 , H) and a.e. (x, ω) ∈ R4 . Define : L 2 (R2 , H) →

2 (R4 , H p 2 ) by
L loc
 
s
f (x, ω) = 
Zαq f (x + , ω) for f ∈ L 2 (R2 , H).
β s∈N2p

By [45, Lemma 3.8], under the general setup, 

Zαq is a unitary operator from L 2 (R2 , H)
onto L (Q αq × Q, H), and is a unitary operator from L 2 (R2 , H) onto L 2 (Q 1 ×
2
β
2
Q, H p ). For every g ∈ L 2 (R2 , H), we associate it with a q 2 × p 2 quaternionic
matrix-valued function  g;α,β by
 
s
 g;α,β (x, ω) = 
Zαq g(x + − αr, ω) for a.e. (x, ω) ∈ R2 × R2 .
β r∈Nq , s∈N p
2 2

(1.11)
 39 Page 8 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

2 Quaternionic Gabor Riesz Sequences

This section is devoted to characterizing quaternionic Gabor Riesz sequences. For

this purpose, we first give some lemmas. The first two lemmas are repeated from [45,
Lemmas 3.7 and 3.14].
Lemma 2.1 For μ > 0,

(η2 ,η1 , j,i,μ)
m,
ϒ n : m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}

is an orthonormal basis for L 2 (Q μ × Q, H), where

(η2 ,η1 , j,i,μ) 1 m 1 x1 m 2 x2

m,
ϒ n (x, ω) = η1 e2πi( μ +n 1 ω1 ) η2 e2π j( μ +n 2 ω2 ) , (x, ω) ∈ Q μ × Q.
μ

Lemma 2.2 Let α and β be as in the general setup, and let g ∈ L 2 (R2 , H). Then, for
every n = qk(n) + p(qm(n) − r(n)) ∈ Z2 with (k(n), r(n), m(n)) ∈ N2p × Nq2 × Z2 ,
there exist unitary matrices E n (x, ω) and Dn (x, ω) such that
 
n
 g;α,β
x+ , ω =E n (x, ω) g;α,β (x, ω)Dn (x, ω) for a.e. (x, ω) ∈ R2 × R2 .
βq

By a standard argument, we have following Lemmas 2.3 and 2.4.

Lemma 2.3 Let A be a positive operator on Cn . Then, for λ ≥ 0, A ≤ λIn if and only
if A2 ≤ λA.
Lemma 2.4 For A ∈ Mm×n (H), we have that
∗
AI2n ≤ XA XA ≤ B I2n on C2n

if and only if

Ac2Hn ≤ Ac2Hm ≤ Bc2Hn for c ∈ Hn .

By Lemma 2.4 and [45, Theorem 4.5] we have the following lemma.
Lemma 2.5 Let α and β be as in the general setup. Then
(i) G(g, α, β) is a Bessel sequence in L 2 (R2 , H) with Bessel bound B if and only if

X∗ g;α,β (x, ω) X g;α,β (x, ω) ≤ β 2 B I2 p2 on C2 p for a.e. (x, ω) ∈ Q

2
1 × Q,
βq
(2.1)

and if and only if

 2
 g;α,β  2
 (x, ω) c q 2 ≤ β 2 Bc2 2 for c ∈ H p and a.e. (x, ω) ∈ Q 1 × Q.
H Hp βq
Journal of Fourier Analysis and Applications (2025) 31:39 Page 9 of 29 39

(ii) G(g, α, β) is a frame for L 2 (R2 , H) with frame bounds A and B if and only if

β 2 AI2 p2 ≤ X∗ g;α,β (x, ω) X g;α,β (x, ω)

2
≤ β 2 B I2 p2 on C2 p for a.e. (x, ω) ∈ Q 1 × Q, (2.2)
βq

and if and only if

 2
 
β 2 Ac2 p2 ≤  g;α,β (x, ω) c q 2
H H
p2
≤β 2
Bc2 p2 for c ∈ H and a.e. (x, ω) ∈ Q 1 × Q. (2.3)
H βq

Remark 2.1 Observe that X∗ g;α,β (x, ω) X g;α,β (x, ω) and X g;α,β (x, ω) X∗ g;α,β (x, ω) share
same nonzero eigenvalues. By Lemmas 2.3 and 2.5, (2.1) can be replaced by one of
the following conditions:
 2
(i) X∗ g;α,β (x, ω) X g;α,β (x, ω) ≤ β 2 BX∗ g;α,β (x, ω) X g;α,β (x, ω) ;
(ii) X g;α,β (x, ω) X∗ g;α,β (x, ω) ≤ β 2 B I2q 2 ;
 2
(iii) X g;α,β (x, ω) X∗ g;α,β (x, ω) ≤ β 2 BX g;α,β (x, ω) X∗ g;α,β (x, ω) .

Remark 2.2 Observe that all norms on Mm, n (C) are equivalent to each other. By
Lemma 2.5 (i), G(g, α, β) is a Bessel sequence in L 2 (R2 , H) if and only if X g;α,β (x, ω)
has L ∞ (Q 1 × Q, C)-entries, equivalently,  g;α,β (x, ω) has L ∞ (Q 1 × Q, H)-
βq βq
entries.

The following two lemmas establish the links between 2 (Z2 × Z2 × {1, i} ×
2 2
{1, j}, R) and L 2 (Q 1 × Q, Hq ), and between L 2 (Q 1 × Q, Hq ) and L 2 (Q 1 ×
β β β
2
Q, C2q ), respectively, in terms of unitary operators.

Lemma 2.6 Let α and β be as in the general setup. Define  : 2 (Z2 × Z2 × {1, i} ×
2
{1, j}, R) → L 2 (Q 1 × Q, Hq ) by
β

⎛ ⎞
1
(η , η2 )  (η2 ,η1 , j,i, β )
c(·, ·) = ⎝ cm,1qn+r ϒm, −n (·, ·)⎠ on Q 1 × Q
β
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2 r∈Nq2
(2.4)

(η1 , η2 )
for c = cm,n : m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} ∈ 2 (Z2 × Z2 × {1, i} ×
{1, j}, R). Then  is a unitary operator from 2 (Z2 × Z2 × {1, i} × {1, j}, R) onto
2
L 2 (Q 1 × Q, Hq ).
β
 39 Page 10 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

Proof Observe that, by Lemma 2.1,

(η , η )
c(·, ·)2 2 2 = |cm,qn+r
1 2
|2 = c2
L (Q 1 ×Q, Hq )
β (η1 , η2 )∈{1, i}×{1, j} m,n∈Z2 r∈Nq2

for c ∈ 2 (Z2 × Z2 × {1, i} × {1, j}, R). Thus  is an isome-

2
try. Arbitrarily fix (dr (·, ·))r∈Nq2 ∈ L 2 (Q 1 × Q, Hq ). Define c =
 β
(η1 , η2 )
cm,n : m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} ∈ 2 (Z2 ×Z2 ×{1, i}×{1, j}, R)
by

(η , η )
2 
(η2 ,η1 , j,i, β1 )
dr (x, ω) = cm,1qn+r ϒm, −n (x, ω)
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2

 (η2 ,η1 , j,i, 1 )

m, n
for r ∈ Nq2 and a.e. (x, ω) ∈ Q 1 × Q. Herein, c is well defined due to ϒ β
:
β

m, n ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} being an orthonormal basis for L 2 (Q 1 ×

β
Q, H). Then

c(x, ω) = (dr (x, ω))r∈Nq2 for a.e. (x, ω) ∈ Q 1 × Q.

β

Therefore,  is an isometric surjection from 2 (Z2 × Z2 × {1, i} × {1, j}, R) onto

2
L 2 (Q 1 × Q, Hq ), and thus it is unitary. The proof is completed. 
β

2
Lemma 2.7 Let α and β be as in the general setup. Define  : L 2 (Q 1 × Q, Hq ) →
β
2
L 2 (Q 1 × Q, C2q ) by
β

 
F1 (x, ω)
F(x, ω) = for (x, ω) ∈ Q 1 × Q (2.5)
−F2 (x, ω) β

2 2
and F = F1 + F2 j ∈ L 2 (Q 1 × Q, Hq ) with F1 , F2 ∈ L 2 (Q 1 × Q, Cq ). Then 
β β
2 2
is a unitary operator from L 2 (Q 1 × Q, Hq ) onto L 2 (Q 1 × Q, C2q ).
β β

2
Proof Arbitrarily fix F = F1 + F2 j ∈ L 2 (Q 1 × Q, Hq ) with F1 , F2 ∈ L 2 (Q 1 ×
β β
 2
Q, Cq ). Since −F2  L 2 (Q ×Q, Cq 2 ) = F2 2 2
2
q2
, we have
1 L (Q 1 ×Q, C )
β β

 2
F2 2 2 = F1 2 2 2 + −F2  L 2 (Q ×Q, Cq )
2
L (Q 1 ×Q, C2q ) L (Q 1 ×Q, Cq ) 1
β β β

= F1 2 2 2 + F2 2 2 2
L (Q 1 ×Q, Cq ) L (Q 1 ×Q, Cq )
β β
Journal of Fourier Analysis and Applications (2025) 31:39 Page 11 of 29 39

= F2 2 2 .
L (Q 1 ×Q, Hq )
β

 
G1 2
Thus  is an isometry. Arbitrarily fix G = ∈ L 2 (Q 1 × Q, C2q ) with G 1 ,
G2 β
2 2
G 2 ∈ L 2 (Q 1 × Q, Cq ). It is easy to check that (G 1 − G 2 j) ∈ L 2 (Q 1 × Q, Hq ),
β β
and
(G 1 − G 2 j) = G.
2
Therefore,  is an isometric surjection from L 2 (Q 1 × Q, Hq ) onto L 2 (Q 1 ×
β β
2
Q, C2q ), and thus it is unitary. The proof is completed. 

Using (2.4), [31, Lemma 3.5] can be rewritten as the following lemma.
Lemma 2.8 Let α and β be as in the general setup and G(g, α, β) be a Bessel sequence
in L 2 (R2 , H) with synthesis operator Tg . Then, for f ∈ L 2 (R2 , H) and c ∈ 2 (Z2 ×
Z2 × {1, i} × {1, j}, R), the following are equivalent:
(i) f = Tg c.
 ∗
(ii) f (x, ω) = β −1  g;α,β (x, ω) c(x, ω) for a.e. (x, ω) ∈ Q 1 × Q, where 
β
is defined as in (2.4).
The following theorem characterizes the quaternionic Gabor Riesz sequences and
orthonormal sequences in L 2 (R2 , H).
Theorem 2.1 Let α and β be as in the general setup. Then, for g ∈ L 2 (R2 , H), we
have
(i) G(g, α, β) is a Riesz sequence in L 2 (R2 , H) with Riesz bounds A and B if and
only if
2
β 2 AI2q 2 ≤ X g;α,β (x, ω) X ∗ g;α,β ≤ β 2 B I2q 2 on C2q for a.e. (x, ω) ∈ Q 1 × Q. (2.6)
 (x, ω) βq

(ii) G(g, α, β) is an orthonormal sequence in L 2 (R2 , H) if and only if

X g;α,β (x, ω) X∗ g;α,β (x, ω) = β 2 I2q 2 on C2q for a.e. (x, ω) ∈ Q
2
1 × Q.
βq

Proof Since an orthonormal sequence is exactly a Riesz sequence with Riesz bound
1, (i) implies (ii). Next we prove (i). By Lemma 2.2, (2.6) holds if and only if it holds
on Q 1 × Q. Also by Lemma 2.5 and Remark 2.1, G(g, α, β) is a Bessel sequence in
β
L 2 (R2 , H) with bound B if and only if the right-hand side inequality in (2.6) holds.
So we may as well assume that G(g, α, β) is a Bessel sequence in L 2 (R2 , H) with
bound B, and we only need to prove the equivalence between
 2
Tg c 2 ≥ Ac2 for c ∈ 2 (Z2 × Z2 × {1, i} × {1, j}, R) (2.7)
L (R2 , H)
 39 Page 12 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

and

X g;α,β (x, ω) X∗ g;α,β (x, ω) ≥ β 2 AI2q 2 on C2q for a.e. (x, ω) ∈ Q 1 × Q
2
(2.8)
β

under this assumption. Next we do this. By the unitarity of , Lemmas 2.6 and 2.8,
(2.7) can be written as
  2 
 g;α,β ∗ 
( (x, ω)) F(x, ω) p2 dxdω ≥ β A 2
F(x, ω)2 q 2 dxdω
Q 1 ×Q H Q 1 ×Q H
β β
(2.9)

2
for F ∈ L 2 (Q 1 × Q, Hq ), equivalently,
β

  2 
 g;α,β 
( (x, ω))∗ F(x, ω) p2 dxdω ≥ β 2 A F(x, ω)2 q 2 dxdω
Q 1 ×Q H Q 1 ×Q H
β β
(2.10)

2 2
for F ∈ L 2 (Q 1 × Q, Hq ) due to F ∈ L 2 (Q 1 × Q, Hq ) if and only if F ∈
β   β
2  
L 2 (Q 1 × Q, Hq ) and  F(x, ω) q 2 = F(x, ω)Hq 2 . By (1.8), (1.9) and Lemma
β H
2.7, (2.10) can be rewritten as
  2 
 ∗ 
X g;α,β (x, ω) G(x, ω) 2 p2 dxdω ≥ β 2 A G(x, ω)2 2q 2 dxdω
Q 1 ×Q C Q 1 ×Q C
β β
(2.11)

2
for G ∈ L 2 (Q 1 × Q, C2q ). Now we prove (2.11)⇔(2.8) to finish the proof.
β
Obviously, (2.8) implies (2.11). For the converse implication, suppose (2.11) holds.
2
Arbitrarily fix d ∈ C2q and E ⊂ Q 1 × Q with |E| > 0. Taking
β

d if (x, ω) ∈ E;
G(x, ω) =
0 if (x, ω) ∈ (Q 1 × Q) \ E
β

in (2.11) gives
  2 
1  ∗ 
X g;α,β (x, ω) d  2 p2 − β A d 2q 2 dxdω ≥ 0.
2 2
|E| E C C

This leads to
 2
 ∗  2
X g;α,β (x, ω) d  2 p2 ≥ β 2 A d2 2q 2 for d ∈ C2q and a.e. (x, ω) ∈ Q 1 × Q
C C β
Journal of Fourier Analysis and Applications (2025) 31:39 Page 13 of 29 39

by [41, Theorem 1.40] and the arbitrariness of E. Equivalently, (2.8) holds. The proof
is completed. 

3 The Duality Principle

This section focuses on the duality principle on quaternionic Gabor systems. We prove
that the Ron–Shen duality principle dose not hold for quaternionic Gabor systems
in L 2 (R2 , H), and present a sufficient condition on window functions g such that
G(g, α, β) is a frame for L 2 (R2 , H) if and only if G(g, β1 , α1 ) is a Riesz sequence in
L 2 (R2 , H). For this purpose, we first give some lemmas. The first lemma characterizes
matrices A in M2×2 (H) such that one of the column vector collections of A and A is
right linearly independent, the other is right linearly dependent.
 
a a
Lemma 3.1 For A = 0,0 1,0 ∈ M2×2 (H), we have
a0,1 a1,1
(i) The columns of A are right linearly independent but the columns of A are right lin-
−1
early dependent if and only if am,n = 0 for (m, n) ∈ N22 and a1,1 = a0,1 a0,0 a1,0 =
−1
a1,0 a0,0 a0,1 .
(ii) The columns of A are right linearly independent but the columns of A are right lin-
−1
early dependent if and only if am,n = 0 for (m, n) ∈ N22 and a1,1 = a1,0 a0,0 a0,1 =
−1
a0,1 a0,0 a1,0 .

Proof Obviously, if (i) holds, replacing A by A leads to (ii). Next we prove (i). It is
easy to verify that, if am,n = 0 for some (m, n) ∈ N22 , then the columns of A are
right linearly dependent if and only if the columns of A are right linearly dependent.
So we may as well assume that am,n = 0 for all (m, n) ∈ N22 . The columns of A
−1 −1
are right linearly independent if and only if a1,0 a1,1 = a0,0 a0,1 , and if and only if
−1
a1,1 = a1,0 a0,0 a0,1 . The columns of A are right linearly dependent if and only if
−1
(a0,0 )−1 a0,1 = (a1,0 )−1 a1,1 , and if and only if a1,1 = a0,1 a0,0 a1,0 . Therefore, (i)
holds. The proof is completed. 
Remark 3.1 The conditions of Lemma 3.1 are easily satisfied due to H being
 noncom-

1 ξ
mutative. For example, if λ ∈ R\{0}, ξ , η ∈ H and ξ η = ηξ , then λ and
η ξη
 
1 ξ
λ satisfy (i) and (ii), respectively.
η ηξ
g; β1 , α1
The following lemma gives the link between  g;α,β and  .
Lemma 3.2 Let α and β be as in the general setup. Then there exist unitary Mq 2 ×q 2 (H)-
valued measurable function U (x, ω) and unitary M p2 × p2 (H)-valued measurable
function V (x, ω) such that, for every g ∈ L 2 (R2 , H),
 ∗
g; β1 , α1
 (x, ω) =U (x, ω) g;α,β (x, ω)V (x, ω) for a.e. (x, ω) ∈ R2 × R2 .
 39 Page 14 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

Proof Replacing α, β, p and q in (1.11) by β1 , α1 , q and p respectively, we have that

 
g; β1 , α1 ! r
 (x, ω) = Z p g(x + αs − , ω)
β β r∈N2p , s∈Nq2
 
s
= 
Zαq g(x + αr − , ω)
β s∈N2p , r∈Nq2

for a.e. (x, ω) ∈ R2 × R2 . For λ ∈ N, define

0 if t ∈ Nλ ; t if t ∈ Nλ ;
τλ (t) = τλ (t) =

1 if t ∈ Nλ + λ, t − λ if t ∈ Nλ + λ.

Arbitrarily fix (r, s) ∈ Nq2 × N2p with r = (r1 , r2 ) and s = (s1 , s2 ). Let us check the
g; 1 , 1  ∗
g; 1 , 1
(r, s)-entry s, rβ α (x, ω) of  β α (x, ω) . Observe that

g; 1 , α1
s, rβ (x, ω)
 
s
=Zαq g x + αr − , ω
β
 
 ( p − ( p − s1 ), p − ( p − s2 ))
= Zαq g x + α (q − (q − r1 ), q − (q − r2 )) − ,ω
β
 
( p − s1 , p − s2 )
=Zαq g x − α(q − r1 , q − r2 ) + ,ω
β

 
= e2π j (τ p ( p−s2 )−τq (q−r2 ))ω2 Zαq g x − α  τq (q − r1 ), 
τq (q − r2 )
  
τ p ( p − s1 ), 
 τ p ( p − s2 )
+ ,ω
β
e2πi (τ p ( p−s1 )−τq (q−r1 ))ω1

for a.e. (x, ω) ∈ R2 × R2 by the quasi-periodicity of 

Zαq . It follows that

g; 1 , α1
s, rβ (x, ω) = e−2πi (τ p ( p−s1 )−τq (q−r1 ))ω1 Hr, s (x, ω)e−2π j (τ p ( p−s2 )−τq (q−r2 ))ω2
(3.1)

where

   
  
τ p ( p − s1 ), 
τ p ( p − s2 )
Hr, s (x, ω) = 
Zαq g x − α τq (q − r1 ), 
τq (q − r2 ) + ,ω .
β
Journal of Fourier Analysis and Applications (2025) 31:39 Page 15 of 29 39

Obviously, Hr, s (x, ω) is exactly the (

τq (q−r1 ), 
τq (q−r2 )), ( τ p ( p−s1 ), 
τ p ( p−s2 ))-
th entry of  g;α,β (x, ω). Thus (Hr, s (x, ω))r∈Nq2 , s∈N2p can be obtained by performing
a sequence of column interchanges and row interchanges to  g;α,β (x, ω). This leads
to the lemma by (3.1). 

Observe that, by [44, Theorem 4.2], a matrix A in Mn×n (H) is unitary if and only
if XA is unitary as a matrix in M2n×2n (C). As an immediate consequence of Theorem
2.1, Lemmas 2.4 and 3.2, we have the following lemma.

Lemma 3.3 Let α and β be as in the general setup. Then G(g, β1 , α1 ) is a Riesz
sequence in L 2 (R2 , H) with Riesz bounds (αβ)2 A and (αβ)2 B if and only if

β 2 AI2 p2 ≤ X ∗ g;α,β
2
X g;α,β (x, ω) ≤ β 2 B I2 p2 on C2 p for a.e. (x, ω) ∈ Q 1 × Q,
 (x, ω) βq
(3.2)

and if and only if

 2
  2
β 2 Ac2 2 ≤  g;α,β (x, ω) c q 2 ≤ β 2 Bc2 2 for c ∈ H p and a.e. (x, ω) ∈ Q 1 × Q. (3.3)
Hp H Hp βq

The following theorem gives a class of examples of the Ron–Shen duality principle
not holding.

Theorem 3.1 Given α, β as in the general setup with 1 < p < q, let { f m }m∈Nn+1 with
2
n = [ p2 ] be a sequence of quaternion-valued functions defined on Q 1 × Q such
βq
that

0< ess inf | f m (x, ω)| ≤ ess sup | f m (x, ω)| < ∞ for every m ∈ Nn+1 ,
(x, ω)∈Q 1 ×Q (x, ω)∈Q 1 ×Q
βq βq

 "
(m) (m)
a0,0 a1,0
and let {Am }m∈Nn = (m) (m) be a sequence in M2×2 (H) such that, for
a0,1 a1,1
m∈Nn
every m ∈ Nn , the columns of Am are right linearly independent but the columns of
Am are right linearly dependent. Take block diagonal matrix (·, ·) on Q 1 × Q such
βq
that

diag(0 (·, ·), 1 (·, ·), · · · , n−1 (·, ·)) if p 2 = 2n;

(·, ·) =
diag(0 (·, ·), 1 (·, ·), · · · , n−1 (·, ·), f n (·, ·)) if p 2 = 2n + 1,

where m (·, ·) = f m (·, ·)Am . Define g and h by

 
(·, ·)
 g;α,β (·, ·) = U (·, ·)  V (·, ·) on Q 1 × Q,
(·, ·) βq
 39 Page 16 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

 
(·, ·)
 h;α,β
(·, ·) = U (·, ·) V (·, ·) on Q 1 × Q,
(·, ·)
 βq

where U (·, ·) and V (·, ·) are q 2 × q 2 and p 2 × p 2 real and unitary matrix-valued
measurable functions defined on Q 1 × Q, respectively,  (·, ·) is a (q 2 − p 2 ) × p 2
βq
quaternionic matrix-valued function with L ∞ (Q 1 × Q, H)-entries such that, on
βq
some E ⊂ Q 1 × Q with |E| > 0,
βq

(·, ·)F(·, ·) = 0
 (3.4)

with
(F0 (·, ·), F1 (·, ·), · · · , Fn (·, ·))T if p 2 = 2n;
F(·, ·) =
(F0 (·, ·), F1 (·, ·), · · · , Fn (·, ·), 0)T if p 2 = 2n + 1,
 
(m) (m) (m)
and Fm (·, ·) = f m (·, ·)a0,0 a0,1 , − f m (·, ·)|a0,0 |2 for m ∈ Nn . Then

(i) G(g, α, β) is a frame for L 2 (R2 , H) but G(g, β1 , α1 ) is not a Riesz sequence in
L 2 (R2 , H).
(ii) G(h, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H) but G(h, α, β) is not a frame for
L 2 (R2 , H).

Proof (i) Observe that

 
(x, ω)V (x, ω)c2 q 2 − p2 ≤ MV (x, ω)c2 2 = Mc2 2
H p p H H

2
for some M > 0, all c ∈ H p and a.e. (x, ω) ∈ Q 1 (·, ·) being of
× Q due to 
βq
L ∞ (Q 1 × Q, H)-entry and the unitarity of V (x, ω). It follows that
βq

 
 (x, ω)V (x, ω)c 2
(x, ω)V (x, ω)c2 p2 ≤ g;α,β
(x, ω)c2 q 2 =
 

(x, ω)V (x, ω)c  q 2
H H
H
≤ (x, ω)V (x, ω)c2 p2 + Mc2 p2
H H

2
for c ∈ H p and a.e. (x, ω) ∈ Q 1 × Q. By the unitarity of V (x, ω) and a simple
βq
computation, we have

2
Ac2 2 ≤ (x, ω)V (x, ω)c2 2 ≤ Bc2 2 for c ∈ H p and a.e. (x, ω) ∈ Q 1 × Q,
Hp Hp Hp βq
Journal of Fourier Analysis and Applications (2025) 31:39 Page 17 of 29 39

where
⎧
⎪
⎪
⎪ min A−1
m 
−2 ess inf | f m (x, ω)|2 if p2 = 2n;
⎪m∈Nn
⎪ (x, ω)∈Q 1 ×Q
⎪
⎪ βq
⎨ ⎧ ⎫
A= ⎪
⎪ ⎪
⎪
⎪
⎪ ⎨ ⎬
⎪
⎪min min A−1 −2 | f m (x, ω)|2 , | f n (x, ω)|2 if p2 = 2n + 1,
⎪
⎪ ⎪ m ess inf ess inf
⎪
⎪
⎩ ⎪
⎩ m∈N n (x, ω)∈Q 1 ×Q (x, ω)∈Q 1 ×Q ⎪
⎭
βq βq

and ⎧
⎪
⎪ max Am 2 ess sup | f m (x, ω)|2 if p2 = 2n;
⎪
⎪m∈Nn (x, ω)∈Q 1 ×Q
⎪
⎪
⎪
⎨ ⎧ βq ⎫
B= ⎪
⎪ ⎪
⎪
⎪
⎪ ⎨ ⎬
⎪max max A 2
⎪ | (x, ω)|2, | (x, ω)|2 if p2 = 2n + 1.
⎪
⎪ ⎪ m ess sup f m ess sup f n
⎪
⎪
⎩ ⎪m∈Nn
⎩ (x, ω)∈Q 1 ×Q (x, ω)∈Q 1 ×Q ⎪
⎭
βq βq

This leads to
2
Ac2 2 ≤ g;α,β (x, ω)c2 q 2 ≤ (B + M) c2 2 for c ∈ H p and a.e. (x, ω) ∈ Q 1 × Q.
Hp H Hp βq
(3.5)

Therefore, G(g, α, β) is a frame for L 2 (R2 , H) by Lemma 2.5 (ii). However, by

a direct computation, we have
 2
 g;α,β 
 (x, ω)V ∗ (x, ω)F(x, ω) 2 =(x, ω)F(x, ω)2 2
(x, ω)F(x, ω)2 2 2
+  q −p
Hp Hp H
=(x, ω)F(x, ω)2 p2
H
 ⎛ ⎞2
 (m) (m) 
 a a 
= | f m (x, ω)| 
2
Am
⎝ 0,0 0,1 ⎠
(m) 2 
m∈Nn  −|a 0,0 |  2
H

  2
 (m) (m) 
 a0,0 a0,1 
for (x, ω) ∈ E by (3.4). Also observe that Am (m)  = 0 by Lemma
 −|a0,0 |2 H2
3.1. It follows that
 2
 g;α,β 
 (x, ω)V ∗ (x, ω)F(x, ω) p2 = 0 for (x, ω) ∈ E. (3.6)
H

Obviously,
 
(m) (m) (m)
V ∗ (x, ω)F(x, ω)2 2 = F(x, ω)2 2 = | f m (x, ω)|2 |a0,0 |2 |a0,1 |2 + |a0,0 |4
Hp Hp
m∈Nn
⎛ ⎞2
⎜ ⎟
≥ ⎝ min ess inf | f m (x, ω)|⎠
m∈Nn (x, ω)∈Q 1 ×Q
βq
 
(m) (m) (m)
|a0,0 |2 |a0,1 |2 + |a0,0 |2 > 0
m∈Nn
 39 Page 18 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

for a.e. (x, ω) ∈ Q 1 × Q. Therefore, G(g, β, α)

1 1
is not a Riesz sequence in
βq
L 2 (R2 ,
H) by Lemma 3.3.
(ii) Observe that  h;α,β (·, ·) =  g;α,β (·, ·) on Q 1 × Q. By (3.5) and (3.6), we have
βq

 2
  2
Ac2 ≤  h;α,β (x, ω)c ≤ (B + M) c2 for c ∈ H p and a.e. (x, ω) ∈ Q 1 ×Q
βq

and

 h;α,β (x, ω)V ∗ (x, ω)F(x, ω)2 =0 for (x, ω) ∈ E.

It follows from Lemma 2.5 (ii) and Lemma 3.3 that G(h, β1 , α1 ) is a Riesz sequence
in L 2 (R2 , H) but G(h, α, β) is not a frame for L 2 (R2 , H).


(·, ·) = 0 makes it hold.

Remark 3.2 Condition (3.4) is easily satisfied. In particular, 

The following theorem gives a sufficient condition on “G(g, α, β) is a frame for

L 2 (R2 , H) ⇔ G(g, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H)".

Theorem 3.2 Let α and β be as in the general setup. For g ∈ L 2 (R2 , H), if one of
the following conditions is satisfied, then G(g, α, β) is a frame for L 2 (R2 , H) with
frame bounds A and B if and only if G(g, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H)
with Riesz bounds (αβ)2 A and (αβ)2 B:
(i) p = 1.
(ii) For a.e. (x, ω) ∈ Q 1 × Q,  g;α,β (x, ω) ∈ Mq 2 × p2 (C) or  g;α,β (x, ω) j ∈
βq
Mq 2 × p2 (C).


Proof If (i) holds, then  g;α,β (x, ω) ∈ Mq 2 ×1 (H). In this case, (2.3) is equivalent to
(3.3). It is easy to verify that

X ∗ g;α,β X g;α,β (x, ω) =

 (x, ω)

X∗ g;α,β (x, ω) X g;α,β (x, ω) if  g;α,β (x, ω) ∈ Mq 2 × p2 (C);

X∗ g;α,β (x, ω) X g;α,β (x, ω) if  g;α,β (x, ω) j ∈ Mq 2 × p2 (C).

So (2.2) is equivalent to (3.2) if (ii) holds. The theorem therefore follows by Lemma
2.5 (ii) and Lemma 3.3. 

Remark 3.3 It is worth noting that the converse of Theorem 3.2 is not true. That is
to say, both (i) and (ii) are only sufficient conditions on “G(g, α, β) is a frame for
L 2 (R2 , H) ⇔ G(g, β1 , α1 ) is a Riesz sequence in L 2 (R2 , H)". Following Theorem
3.3 shows this.
Journal of Fourier Analysis and Applications (2025) 31:39 Page 19 of 29 39

Theorem 3.3 Let α and β be as in the general setup with p > 1. Let A1 (x, ω),
A2 (x, ω) ∈ Mq 2 × p2 (C) with L ∞ (Q 1 × Q, C)-entries. Suppose there exists a
βq
constant λ > 0 such that

inf A1 (x, ω)d2 + inf A2 (x, ω)d2 − 2A1 (x, ω)A2 (x, ω) ≥ λ (3.7)
2 2
d∈C p :d=1 d∈C p :d=1

for a.e. (x, ω) ∈ Q 1 × Q. Define g by

βq

 g;α,β (x, ω) = A1 (x, ω) + A2 (x, ω) j for a.e. (x, ω) ∈ Q 1 × Q.

βq

Then G(g, α, β) is a frame for L 2 (R2 , H) and G(g, β, α)

1 1
is a Riesz sequence in
L 2 (R2 , H).

Proof By Lemma 2.5 (ii) and Lemma 3.3, we only need to prove that both (2.2) and
(3.2) hold. Observe that
   
A1 (x, ω) A2 (x, ω) A1 (x, ω) −A2 (x, ω)
X g;α,β (x, ω) = and X g;α,β (x, ω) = .
−A2 (x, ω) A1 (x, ω) A2 (x, ω) A1 (x, ω)
(3.8)

It follows that
 2  2
X g;α,β   
(x, ω) c C2q 2 ≤ Bc 2 p2 and X g;α,β (x, ω) c 2q 2 ≤ Bc 2 p2
2 2
 C C
C

2
for some B > 0, all c ∈ C2 p and a.e. (x, ω) ∈ Q 1 × Q due to A1 (x, ω) and
βq
A2 (x, ω) being of L ∞ (Q
1 × Q, C)-entry. This proves the right-hand inequality in
βq
(2.2) and
 (3.2).
 Next we verify the left-hand inequalities in (2.2) and (3.2). Arbitrarily
c(1)
fix c = (2) ∈ C2 p with c = 1 and c(1) , c(2) ∈ C p . Then, for (x, ω) ∈ Q 1 ×Q,
2 2

c βq
 2 
  
X g;α,β (x, ω) c 2q 2 = A1 (x, ω)c(1)
C
2  2
  
+A2 (x, ω)c(2)  q 2 + −A2 (x, ω)c(1) + A1 (x, ω)c(2)  q 2
C C
  2   2
 A1 (x, ω) 0   A2 (x, ω) 0 
=
 c
 +
 c
 2q 2
0 A1 (x, ω) C 2q 2 0 A 2 (x, ω) C
& ' & '
+ 2Sc A1 (x, ω)c(1) , A2 (x, ω)c(2)  q 2 − 2Sc A2 (x, ω)c(1) , A1 (x, ω)c(2)  2
C Cq
≥ inf A1 (x, ω)d2 + inf A2 (x, ω)d2 + (∗) (3.9)
2 2
d∈C p :d=1 d∈C p :d=1

where
& ' & '
(∗) = 2Sc A1 (x, ω)c(1) , A2 (x, ω)c(2)  q 2 − 2Sc A2 (x, ω)c(1) , A1 (x, ω)c(2)  q 2 . (3.10)
C C
 39 Page 20 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

Also observe that

|(∗)| ≤ 4A1 (x, ω)A2 (x, ω)c(1) C p2 c(2) C p2 ≤ 2A1 (x, ω)A2 (x, ω),
(3.11)

we have
 2
X  2
≥ λc2 2 p2 for c ∈ C2 p and a.e. (x, ω) ∈ Q ×Q
 g;α,β (x, ω) c C2q 2 C
1
βq

by (3.7). Replacing A1 (x, ω) and A2 (x, ω) in (3.9)-(3.11) by A1 (x, ω) and

−A2 (x, ω) respectively, we have
 2
  2
X g;α,β (x, ω) c 2q 2 ≥ λc2 2 p2 for c ∈ C2 p and a.e. (x, ω) ∈ Q 1 ×Q
C C βq

by (3.7) and (3.8). The proof is completed. 

Remark 3.4 The condition (3.7) in Theorem 3.3 is easily satisfied. By [18, Corollary
2.4], there exist unitary measurable functions U1 (·, ·), U2 (·, ·) ∈ Mq 2 ×q 2 (C) and
V1 (·, ·), V2 (·, ·) ∈ M p2 × p2 (C) defined on Q 1 × Q such that
βq

  
1 (·, ·) 2 (·, ·)
A1 (·, ·) = U1 (·, ·) V1 (·, ·) and A2 (·, ·) = U2 (·, ·) V2 (·, ·)
0 0

on Q 1 × Q, where
βq

   
(1) (2)
1 (·, ·) = λs, s (·, ·) and 2 (·, ·) = λs, s (·, ·)
s, s ∈N2p s, s ∈N2p

satisfy

(l) λ(l) ∞
s, s (·, ·) ∈ L (Q 1 × Q, C) if s = s;
λs, s (·, ·) = βq for l = 1, 2,
0 otherwise

and
2 2
min ess inf λ(1)
s, s (x, ω) + min ess inf λ(2)
s, s (x, ω)
s∈N2p (x, ω)∈Q 1 ×Q s∈N2p (x, ω)∈Q 1 ×Q
βq βq
⎛ ⎞⎛ ⎞
⎜ ⎟⎜ ⎟
− 2 ⎝max ess sup λ(1)
s, s (x, ω) ⎠ ⎝max ess sup λ(2)
s, s (x, ω) ⎠ > 0.
s∈N2p (x, ω)∈Q 1 ×Q 2 s∈N p (x, ω)∈Q 1 ×Q
βq βq

Then (3.7) holds.

Journal of Fourier Analysis and Applications (2025) 31:39 Page 21 of 29 39

4 The Biorthogonality Relation

This section focuses on the biorthogonality relation on quaternionic Gabor systems. We

show that the traditional Wexler–Raz biorthogonality relation does not hold for quater-
nionic Gabor systems, and then establish a biorthogonality relation on quaternionic
Gabor systems.
Given two Bessel sequences G(g, α, β) and G(h, α, β) in L 2 (R2 , H), if the tra-
ditional Wexler–Raz biorthogonality relation holds on quaternionic Gabor systems, it
should be
 
(η ,η )
Sg, h = I on L 2 (R2 , H) ⇔ (αβ)−2 h, M m 1 2 T k g = δm,0 δk,0 δ(η1 , η2 ),(1, 1)
α β L 2 (R2 , H)
(4.1)
for m, k ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}. But following Example 4.1 shows that, in
general, (4.1) does not hold. For this purpose, we need to give two lemmas. The first
one is repeated from [45, Lemma 4.4].
Lemma 4.1 Let α and β be as in the general setup, and let g ∈ L 2 (R2 , H). Define

g (r) (·) = g(· − αr) for r ∈ Nq2 .

Then
 
(η , η )
f , Mβm1 2 Tαqn g (r)
L 2 (R2 , H)

1 s s (η ,η , j,i, β1 )
= Sc 
Zαq f (x + , ω) m,2−n1
Zαq g(x + − αr, ω)ϒ (x, ω)dxdω
β Q 1 ×Q 2
β β
β s∈N p

for f ∈ L 2 (R2 , H), r ∈ Nq2 , m, n ∈ Z2 and (η1 , η2 ) ∈ {1, i} × {1, j}.

The following lemma characterizes C in terms of H.
Lemma 4.2

Z
C = η ∈ H : e2πiθ ηe−2πiθ = η for some θ ∈ R \ .
2

Z
Proof Obviously, C ⊂ η ∈ H : e2πiθ ηe−2πiθ = η for some θ ∈ R \ 2 . Next we
prove the converse inclusion. Suppose η ∈ H satisfies

Z
e2πiθ ηe−2πiθ = η for some θ ∈ R \ . (4.2)
2

Let η = η1 + η2 j with η1 , η2 ∈ C. Then

e2πiθ ηe−2πiθ = η1 + e4πiθ η2 j

 39 Page 22 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

by a simple computation. Thus e4πiθ η2 = η2 by (4.2). This implies that η2 = 0. Thus

η ∈ C. The proof is completed. 
The following example shows that (4.1) does not hold for quaternionic Gabor
systems.
Example 4.1 Let α and β be as in the general setup with p = 1, q > 1. Define
h ∈ L 2 (R2 , H) by

 A(x, ω) if (x, ω) ∈ E;
Zαq h(x, ω) = for (x, ω) ∈ Q αq × Q,
0, if (x, ω) ∈ (Q αq × Q) \ E
(4.3)
& ( '
where E = Q 1 Q 1 + α(1, 0)T × Q, A(x, ω) is a real-valued measurable
βq βq
function on E such that

0 < ess inf |A(x, ω)| ≤ ess sup |A(x, ω)| < ∞.
(x, ω)∈E (x, ω)∈E

Take g such that 

Zαq g(x, ω) ∈ L ∞ (Q αq × Q, H),
   
 −1 2πiω1 1
Zαq g(x, ω) = β A(x, ω) − e
2
Zαq g x + α , ω e−2πiω1 (4.4)
0

for (x, ω) ∈ Q 1 × Q, and   ⊂ (Q 1 +α(1, 0)T )× Q

Zαq g (x, ω) ∈ H\C on some E
βq βq
 > 0. Then Sg, h = I on L 2 (R2 , H), but
with | E|
 
(η ,η )
(αβ)−2 h, M m 1 2 T k g = δm,0 δk,0 δ(η1 , η2 ),(1, 1) (4.5)
α β L 2 (R2 , H)

does not hold.

Proof Since p = 1,  g;α,β (x, ω) and  h;α,β (x, ω) are both q 2 × 1 matrices:
   
 g;α,β (x, ω) = 
Zαq g(x − αr, ω) ,  h;α,β (x, ω) = 
Zαq h(x − αr, ω) .
r∈Nq2 r∈Nq2

We first calculate ( h;α,β (x, ω))∗  g;α,β (x, ω) for a.e. (x, ω) ∈ Q 1 × Q. By the
βq

quasi-periodicity of 
Zαq , we have that

( h;α,β (x, ω))∗  g;α,β (x, ω)

Zαq h(x, ω)

= Zαq g(x, ω) + Zαq h(x − αr, ω)
 Zαq g(x − αr, ω)
r∈Nq2 \{(0,0)T }
   
1 1
Zαq h(x, ω)
= Zαq g(x, ω) + e2πiω1 
Zαq h(x − αr + αq , ω)
Zαq g(x − αr + αq , ω)e−2πiω1
0 0
r∈Nq2 \{(0,0)T }
Journal of Fourier Analysis and Applications (2025) 31:39 Page 23 of 29 39

for a.e. (x, ω) ∈ Q 1 × Q. Observe that

βq

(Q 1 + α(q(1, 0)T − r) + αqk) ∩ E = ∅ for k ∈ Z2

βq

/ {(0, 0)T , (q − 1, 0)T }. It follows from (4.3) and the quasi-periodicity of 

if r ∈ Zαq
that

( h;α,β (x, ω))∗  g;α,β (x, ω)

       
1 1
Zαq h(x, ω)
= Zαq g(x, ω) + e2πiω1 
Zαq h x + α ,ω  Zαq g x + α , ω e−2πiω1
0 0
     
1
=A(x, ω)  Zαq g(x, ω) + e2πiω1 
Zαq g x + α , ω e−2πiω1 (4.6)
0

for a.e. (x, ω) ∈ Q 1 × Q. Substituting (4.4) into (4.6) leads to

βq

( h;α,β (x, ω))∗  g;α,β (x, ω) = β 2 for a.e. (x, ω) ∈ Q 1 × Q,

βq

equivalently, Sg, h = I on L 2 (R2 , H) by the proof of [31, Theorem 3.1]. Next we

prove that (4.5) does not hold to finish the proof. Suppose (4.5) holds. By Lemma 4.1,
replacing α, β, p and q by β1 , α1 , q and p respectively, we have

 
(η ,η2 )
h, M m 1 Tk g
α β L 2 (R2 , H)

= α Sc 
Zαq h(x + αr, ω)  (η2 ,η1 , j,i,α) (x, ω)dxdω
Zαq g(x + αr, ω) ϒm, −k
Q α ×Q
r∈Nq2

for m, k ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}. Also observe that α = 1

βq and


ϒ (η2 ,η1 , j,i,α) : m, k ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}
m, −k

is an orthonormal basis for L 2 (Q α × Q, H). It follows that (4.5) holds if and only if


Zαq h(x + αr, ω)
Zαq g(x + αr, ω) = β 2 for a.e. (x, ω) ∈ Q 1 × Q. (4.7)
βq
r∈Nq2

Substituting (4.3) into (4.7), we have that (4.5) holds if and only if
    
1
A (x, ω) 
Zαq g (x, ω) + 
Zαq g x + α ,ω = β 2 for a.e. (x, ω) ∈ Q 1 × Q,
0 βq
 39 Page 24 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

equivalently,
    
  1
A (x, ω) Zαq g (x, ω) + Zαq g x + α ,ω = β 2 for a.e. (x, ω) ∈ Q 1 × Q
0 βq

(4.8)

due to A(x, ω) being a real-valued measurable function.

However, (4.4) implies that

     
1
A(x, ω) Zαq g(x, ω) + e2πiω1 
Zαq g x + α , ω e−2πiω1 = β 2 for a.e. (x, ω) ∈ Q 1 × Q.
0 βq

This together with (4.8) leads to

       
1 1
e2πiω1 
Zαq g x + α , ω e−2πiω1 = 
Zαq g x + α , ω for a.e. (x, ω) ∈ Q 1 × Q.
0 0 βq

In particular,
       
1 1
e2πiω1 
Zαq g x + α , ω e−2πiω1 = 
Zαq g x + α 
, ω for a.e. (x, ω) ∈ E.
0 0

It is a contradiction by Lemma 4.2 due to   Therefore, (4.5)

Zαq g (x, ω) ∈ H \ C on E.
does not hold. The proof is completed. 

Now we will establish the biorthogonality relation on quaternionic Gabor systems.

For this purpose, we need to give two lemmas. By [45, Lemma 3.6], we have the
following lemma.

Lemma 4.3 For μ > 0,


(η ,η ,μ)
εm 1 2 : m ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}

is an orthonormal basis for L 2 (Q μ , H), where

(η ,η2 ,μ) 1 m 2 x2 m 1 x1
εm 1 (x) = η2 e2π j μ Cr (η1 e2πi μ ) for x ∈ Q μ .
μ

The following lemma expresses Sg, h in a weak sense.

Lemma 4.4 Let α and β be as in the general setup, and let G(g, α, β), G(h, α, β) be
two Bessel sequences in L 2 (R2 , H). Then

) * −2
γ , Sg, h f L 2 (R2 , H)
=β Sc γ (x) T− k f · G k (x)dx (4.9)
R2 β
k∈Z2
Journal of Fourier Analysis and Applications (2025) 31:39 Page 25 of 29 39

for all f , γ ∈ L ∞ (R2 , H) with bounded support, where

 
k
G k (x) = g x + − αn h(x − αn) for k ∈ Z2 .
2
β
n∈Z

Proof Arbitrarily fix f , γ ∈ L ∞ (R2 , H) with bounded support. Then

k k k k
γ (· + )h(· − αn + ), )g(· − αn + ) ∈ L 2 (Q 1 , H),
f (· + (4.10)
β β β β β
k∈Z2 k∈Z2
 
k
card k ∈ Z2 : x + ∈ supp(γ ) < ∞, (4.11)
β

and
card ( ) < ∞, (4.12)

where = k ∈ Z2 : x + k
β ∈ supp( f ), x ∈ supp(γ ) . By (4.10) and a simple
computation, we have

  
(η , η2 ) 1 k k (η1 ,η2 , β1 )
γ , Mβm1 Tαn h = Sc )h(x − αn + ) εm
γ (x + (x)dx
L 2 (R2 , H) β Q1 2
β β
β k∈Z
+ ,
1 k k (η1 ,η2 , β1 )
= γ (· + )h(· − αn + ), εm .
β 2
β β
k∈Z L 2 (Q 1 , H)
β

Similarly,
+ ,
  1 k k (η1 ,η2 , β1 )
(η , η )
f, Mβm1 2 Tαn g 2 2 = f (· + )g(· − αn + ), εm .
L (R , H) β β β
k∈Z2 L 2 (Q 1 , H)
β

It follows that
) *  (η , η )
  (η , η ) 
γ , Sg, h f L 2 (R2 , H) = γ , Mβm1 2 Tαn h 2 2 Mβm1 2 Tαn g, f 2 2
L (R , H) L (R , H)
(η1 , η2 )∈{1, i}×{1, j} m, n∈Z2

1
=
β2 Q1
n∈Z2 β
k k k k
γ (x + )h(x − αn + ) g(x − αn + ) f (x + )dx
β β β β
k∈Z2 k∈Z2

1 k k
= 2 γ (x)h(x − αn) g(x − αn +
β
) f (x + )dx
β
(4.13)
β R2
n∈Z2 k∈Z2

by (4.11) and Lemma 4.3. Since supp(γ ) is bounded, there exists a finite subset F of
Z2 such that supp(γ ) ⊂ ∪ (Q α + αm). Then
m∈F
 39 Page 26 of 29 Journal of Fourier Analysis and Applications (2025) 31:39


k k
γ (x)h(x − αn)g(x − αn + ) f (x + ) dx
R2 β β
n∈Z2 k∈Z2

k k
≤ γ (x)h(x − αn)g(x − αn + ) f (x + ) dx
Q α +αm β β
m∈F n∈Z2 k∈Z2

k k
= card(F) γ (x)h(x)g(x + ) f (x + ) dx
R2 β β
k∈Z2
≤ card(F)card( )γ ∞  f ∞ g L 2 (R2 , H) h L 2 (R2 , H)
<∞

by (4.12). This together with (4.13) leads to (4.9). 

The following theorem gives a biorthogonal characterization of quaternionic Gabor

dual frame pairs.

Theorem 4.1 Let α and β be as in the general setup, and let G(g, α, β), G(h, α, β)
be two Bessel sequences in L 2 (R2 , H). Then the following are equivalent.
(i) Sg, h = -I on L 2 (R2 , H). .
m 2 k2 m 1 k1
−2 2π j αβ 2πi αβ (η1 ,η2 )
(ii) (αβ) e e g, M m Tk h = δm,0 δk,0 δ(η1 , η2 ),(1, 1) for
α β
L 2 (R2 , H)
m, k ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j}.

Proof By Lemma 4.4, (i) holds if and only if


−2
γ , f  L 2 (R2 , H) = β Sc γ (x) T− k f · G k (x)dx,
R2 β
k∈Z2

equivalently,
⎡ ⎛ ⎞⎤

Sc ⎣γ (x) ⎝β −2 T− k f · G k (x) − f (x)⎠⎦ dx = 0 (4.14)
R2 β
k∈Z2

for γ , f ∈ L ∞ (R2 , H) with bounded support. Arbitrarily fix f ∈ L ∞ (R2 , H) with

bounded support, E ⊂ R2 with |E| > 0. Taking γ = η1 E with η ∈ {1, i, j, k} in
(4.14) leads to
⎡ ⎛ ⎞⎤

Sc ⎣η ⎝β −2 T− k f · G k (x) − f (x)⎠⎦ dx = 0.
E β
k∈Z2
Journal of Fourier Analysis and Applications (2025) 31:39 Page 27 of 29 39

It follows that

f = β −2 T− k f · G k for f ∈ L ∞ (R2 , H) with bounded support (4.15)

β
k∈Z2

by [41, Theorem 1.40] and the arbitrariness of E and η. Obviously, (4.15) implies
(4.14). Therefore, (i) is equivalent to (4.15). By a standard argument, (4.15) is
equivalent to

G k (x) = β 2 δk, 0 for k ∈ Z2 and a.e. x ∈ R2 . (4.16)

Next we prove the equivalence between (4.16) and (ii) to finish the proof. By
Proposition 1.1 (iv), we have
- m 2 k2 m 1 k1
.
2π j αβ 2πi αβ (η1 ,η2 )
e e g, M m Tk h
α β
L 2 (R2 , H)
 m 1 k1 m 2 k2
2πi (η1 ,η2 ) 2π j
= Sc e αβ g(x)M mdx T k h(x)e αβ
α β
R 
2
m 1 x1  m 1 k1 k m 2 x2 m 2 k2
−2π j αβ
e−2πi α η1 e g (x)h(x − )e−2π j α η2 e
2πi αβ
= Sc dx
R2 β

for m, k ∈ Z2 and (η1 , η2 ) ∈ {1, i} × {1, j}. It follows that

- m 2 k2 m 1 k1
.
2π j 2πi (η ,η2 )
e αβ e αβ g, M m 1 Tk h
α β
L 2 (R2 , H)
  
m 1 x1 k m 2 x2
= Sc e−2πi α η1 g x + h(x)e−2π j α η2 dx
R2 β

(η ,η ,α)
= α Sc G k (x)εm 1 2 (x)dx
R2

for m, k ∈ Z2 and (η1 , η2 ) ∈ {1, i} × {1, j} by a simple com-

putation.
 This implies the equivalence between (4.16) and (ii) due to
(η ,η ,α)
εm 1 2 : m ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} being an orthonormal basis for
L 2 (Q α , H). The proof is completed. 

Remark 4.1 If one of g and h in Theorem 4.1 is a real-valued function in addition,

then (ii) can be rewritten as
 
(η ,η )
(αβ)−2 h, M m 1 2 T k g = δm,0 δk,0 δ(η1 , η2 ),(1, 1)
α β L 2 (R2 , H)

for m, k ∈ Z2 , (η1 , η2 ) ∈ {1, i} × {1, j} by a simple computation, that is, the

traditional Wexler–Raz biorthogonality relation holds.
 39 Page 28 of 29 Journal of Fourier Analysis and Applications (2025) 31:39

Acknowledgements The authors would like to extend their sincere gratitude to referees for their careful
reading this article and valuable comments, which helped to improve the readability of this article.

Funding This work was supported by National Natural Science Foundation of China (Grant No. 12371091)
Science Research Project of Hebei Education Department (Grant No. QN2025067).

References
1. Bahri, M., Hitzer, E., Ashino, R., Vaillancourt, R.: Windowed Fourier transform of two-dimensional
quaternionic signals. Appl. Math. Comput. 216(8), 2366–2379 (2010)
2. Bahri, M., Hitzer, E., Hayashi, A., Ashino, R.: An uncertainty principle for quaternion Fourier
transform. Comput. Math. Appl. 56(9), 2398–2410 (2008)
3. Bayro-Corrochano, E., Trujillo, N., Naranjo, M.: Quaternion Fourier descriptors for the preprocessing
and recognition of spoken words using images of spatiotemporal representations. J. Math. Imaging
Vis. 28(2), 179–190 (2007)
4. Bülow, T.: Hypercomplex spectral signal representations for the processing and analysis of images.
Ph.D. Thesis, Institut für Informatik und Praktische Mathematik, University of Kiel, Germany (1999)
5. Casazza, P.G., Kutyniok, G., Lammers, M.C.: Duality principles in frame theory. J. Fourier Anal. Appl.
10(4), 383–408 (2004)
6. Cerejeiras, P., Hartmann, S., Orelma, H.: Structural results for quaternionic Gabor frames. Adv. Appl.
Clifford Algebr. 28(5), 86 (2018)
7. Christensen, O., Kim, H.O., Kim, R.Y.: On the duality principle by Casazza, Kutyniok, and Lammers.
J. Fourier Anal. Appl. 17(4), 640–655 (2011)
8. Christensen, O., Xiao, X.C., Zhu, Y.C.: Characterizing R-duality in Banach spaces. Acta Math. Sin.
29(1), 75–84 (2013)
9. Daubechies, I., Landau, H.J., Landau, Z.: Gabor time-frequency latices and the Wexler-Raz identity.
J. Fourier Anal. Appl. 1(4), 437–478 (1995)
10. Dong, J., Li, Y.-Z.: Duality principles in Hilbert-Schmidt frame theory. Math. Methods Appl. Sci.
44(6), 4888–4906 (2021)
11. El Kassimi, M., Fahlaoui, S.: The two-sided Gabor quaternionic Fourier transform and uncertainty
principles. In: Recent advances in mathematics and technology, Appl. Numer. Harmon. Anal., pp.
3-19. Birkhäuser/Springer, Cham (2020)
12. Ell, T.A.: Hypercomplex Spectral Transformations. Ph.D. Dissertation, University of Minncsota (1992)
13. Ell, T.A.: Quaternion-fourier transforms for analysis of two-dimensional linear time-invariant partial
differential systems. In: Proceeding of the 32nd Conference on Decision and Control, pp. 1830-1841.
San Antonio, Texas (1993)
14. Enayati, F., Asgari, M.S.: Duality properties for generalized frames. Banach J. Math. Anal. 11(4),
880–898 (2017)
15. Fu, Y., Kähler, U., Cerejeiras, P.: The Balian-Low theorem for the windowed quaternionic Fourier
transform. Adv. Appl. Clifford Algebr. 22(4), 1025–1040 (2012)
16. Gröchenig, K.: Foundations of Time-Frequency Analysis. Birkhäuser, Boston (2001)
17. Gürlebeck, K., Habetha, K., Sprössig, W.: Funktionentheorie in der Ebene und im Raum, 1, aufl
Springer, Berlin (2006)
18. Han, B.: Homogeneous wavelets and framelets with the refinable structure. Sci. China Math. 60(11),
2173–2198 (2017)
19. Hartmann, S.: Quaternionic Gabor Expansion. Technische Universität Bergakademie Freiberg,
Freiberg (2015)
20. Hartmann, S.: Some results on the lattice parameters of quaternionic Gabor frames. Adv. Appl. Clifford
Algebr. 26(1), 137–149 (2016)
21. Hartmann, S.: Relaxed quaternionic Gabor expansions at critical density. Math. Methods Appl. Sci.
40(5), 1666–1678 (2017)
22. Hitzer, E.: Quaternion Fourier transform on quaternion fields and generalizations. Adv. Appl. Clifford
Algebr. 17(3), 497–517 (2007)
23. Janssen, A.J.E.M.: Duality and biorthogonality for Weyl-Heisenberg frames. Duality and biorthogo-
nality for Weyl-Heisenberg frames. J. Fourier Anal. Appl. 1(4), 403–436 (1995)
Journal of Fourier Analysis and Applications (2025) 31:39 Page 29 of 29 39

24. Kamel, B., Tefjeni, E.: Uncertainty principle for the two-sided quaternion windowed Fourier transform.
Integral Transforms Spec. Funct. 30(5), 362–382 (2019)
25. Lee, H.C.: Eigenvalues and canonical forms of matrices with quaternion coefficients. Proc. R. Irish
Acad. Sect. A 52, 253–260 (1949)
26. Li, J.X., He, J.X.: Some results for the two-sided quaternionic Gabor Fourier transform and quaternionic
Gabor frame operator. Adv. Appl. Clifford Algebr. 31(1), 1 (2021)
27. Li, Y.-Z., Dong, J.: On a class of weak R-duals and the duality relations. Banach J. Math. Anal. 14(2),
450–469 (2020)
28. Li, Y.-Z., Dong, J.: Duality relations associated with weak g-R-duals. Linear Multilinear Algebra
70(20), 5482–5501 (2022)
29. Li, Y.-Z., Hussain, T.: Duality principles for Fa -frame theory in L 2 (R+ ). Bull. Malays. Math. Sci.
Soc. 44(4), 2401–2423 (2021)
30. Li, Y.-Z., Zhang, X.-L.: A class of quaternionic Fourier orthonormal bases. Forum Math. 36(3), 825–834
(2024)
31. Li, Y.-Z., Zhang, X.-L.: Characterization of rationally sampled quaternionic dual Gabor frames. Math.
Methods Appl. Sci. 47(2), 1095–1112 (2024)
32. Mejjaoli, H.: Paley-Wiener theorems for the two-sided quaternion Fourier transform. Adv. Appl.
Clifford Algebr. 27(2), 1611–1631 (2017)
33. Pei, S.C., Ding, J.J., Chang, J.H.: Efficient implementation of quaternion Fourier transform, con-
volution, and correlation by 2-D complex FFT. IEEE Trans. Signal Process. 49(11), 2783–2797
(2001)
34. Ron, A., Shen, Z.: Weyl-Heisenberg systems and Riesz bases in L 2 (Rd ). Duke Math. J. 89(2), 237–282
(1997)
35. Sangwine, S.J.: Fourier transforms of color images using quaternion or hypercomplex numbers.
Electron. Lett. 32(21), 1979–1980 (1996)
36. Sangwine, S.J., Ell, T.A.: Hypercomplex Fourier transforms of color images. IEEE Trans. Image
Process. 16(1), 22–35 (2007)
37. Shapiro, M., Tovar, L.M.: On a class of integral representations related to the two-dimensional
Helmholtz operator. Contemp. Math. 212, 229–244 (1998)
38. Stoeva, D.T., Christensen, O.: On R-duals and the duality principle in Gabor analysis. J. Fourier Anal.
Appl. 21(2), 383–400 (2015)
39. Stoeva, D.T., Christensen, O.: On various R-duals and the duality principle. Integr. Equ. Oper. Theory
84(4), 577–590 (2016)
40. Takhteh, F., Khosravi, A.: R-duality in g-frames. Rocky Mt. J. Math. 47(2), 649–665 (2017)
41. Walter, R.: Real and Complex Analysis, 3rd edn. McGraw-Hill Book Co., New York (1987)
42. Wexler, J., Raz, S.: Discrete Gabor expansions. Signal Process. 21, 207–220 (1990)
43. Wolf, L.A.: Similarity of matrices in which the elements are real quaternions. Bull. Am. Math. Soc.
42(10), 737–743 (1936)
44. Zhang, F.Z.: Quaternions and matrices of quaternions. Linear Algebra Appl. 251, 21–57 (1997)
45. Zhang, X.-L., Li, Y.-Z.: Quaternionic Gabor frame characterization and the density theorem. Banach
J. Math. Anal. 17(4), 64 (2023)

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.