Next Article in Journal
Quaternion and Biquaternion Representations of Proper and Improper Transformations in Non-Cartesian Reference Systems
Previous Article in Journal
Reliability Analysis for Degradation-Shock Processes with State-Varying Degradation Patterns Using Approximate Bayesian Computation (ABC) for Parameter Estimation
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Some Limiting Measures Related to Free and Boolean Convolutions

by
Ayed. R. A. Alanzi
1,
Shokrya S. Alshqaq
2 and
Raouf Fakhfakh
1,*
1
Department of Mathematics, College of Science, Jouf University, Sakaka P.O. Box 2014, Saudi Arabia
2
College of Science, Jazan University, P.O. Box 114, Jazan 45142, Saudi Arabia
*
Author to whom correspondence should be addressed.
Symmetry 2024, 16(10), 1365; https://doi.org/10.3390/sym16101365
Submission received: 13 September 2024 / Revised: 8 October 2024 / Accepted: 12 October 2024 / Published: 14 October 2024
(This article belongs to the Section Mathematics)

Abstract

:
Several authors are interested in the study of limiting distributions for large symmetrical random matrices. Our approach of studying limiting distributions is different and is related to the new concept of variance functions of Cauchy–Stieltjes Kernel (CSK) families of probabilities. By means of the variance functions machinery, some novel limiting probability measures are provided involving the free multiplicative law of large numbers, using both free and Boolean (additive and multiplicative) convolutions. Several examples of these limiting probability measures are given in the context of free probability.

1. Introduction

The topic of limiting spectral distribution for symmetric matrices has received much attention from researchers, see [1,2,3,4,5]. Connected to this topic, several works have explored limiting measures for Boolean and free (additive and multiplicative) convolutions during the past few decades; see [6,7,8,9,10]. Many fundamental characteristics for sums and products of classical independent random variables apply symmetrically to free and Boolean independent random variables. In [11], the author presented research on the distribution of sums and products of Boolean independent random variables in an infinitesimal triangular array. The limit measures of Boolean convolutions are dictated by those of free convolutions, and vice versa. These findings demonstrate symmetries and similarities between the limiting distributional behavior of Boolean and classical convolutions. The authors of [12] presented the distributional model behavior of sums of identically distributed free random variables. The limiting measures of classical and free additive convolutions are explicitly related. Tucci [13] established the free multiplicative law of large numbers (FMLLN) for probabilities with bounded support. The authors of [14] presented an expanded result of [13] to cover measures with unbounded support. Further results on the topic of limiting laws in free probability are presented in [15,16,17]. Pursuing the study of this topic, and based on a new concept called variance functions of Cauchy-Stieltjes Kernel (CSK) families, in this paper, we provide some new limiting measures in connection with the FMLLN and involving Boolean and free (additive and multiplicative) convolutions. To illustrate our findings clearly, we need to review key ideas in free probability. P (respectively, P + ) represents the set of probabilities on R (respectively, R + ). The Cauchy–Stieltjes transform G σ ( · ) of σ P is introduced as
G σ ( w ) = 1 w ξ σ ( d ξ ) , for w C supp ( σ ) .
The free additive convolution σ ρ of σ and ρ P , is introduced by
C σ ρ ( ξ ) = C σ ( ξ ) + C ρ ( ξ ) ,
where C σ ( · ) is the free cumulant transform of σ provided by (see [18])
C σ ( G σ ( ξ ) ) = ξ 1 G σ ( ξ ) , ξ in   an   appropriate   domain .
σ P is ⊞-infinitely divisible if for each p N , σ p P exists, so that
σ = σ p σ p p times .
The s-fold additive free convolution of σ with itself is denoted by σ s . It is well defined ∀ s 1 , (see [19]) and
C σ s ( ξ ) = s C σ ( ξ ) .
σ P is ⊞-infinitely divisible if σ s is well defined ∀ s > 0 .
For λ P + , ( λ δ 0 ) , the S -transform is provided by
C λ ( ζ S λ ( ζ ) ) = 1 S λ ( ζ ) ζ close   enough   to 0 .
Multiplying S -transforms results in another S -transform. The multiplicative free convolution λ τ of λ and τ P + is introduced by S λ τ ( ζ ) = S λ ( ζ ) S τ ( ζ ) . Powers of multiplicative free convolution λ q are well defined at least ∀ q 1 (see [20], Theorem 2.17), by S λ q ( ζ ) = S λ ( ζ ) q .
The Boolean additive convolution σ ρ of σ and ρ P is the measure provided by
E σ ρ ( ξ ) = E σ ( ξ ) + E ρ ( ξ ) , ξ C + ,
where E σ ( ξ ) = ξ 1 G σ ( ξ ) , is the Boolean cumulant transform of σ .
σ P is ⊎-infinitely divisible if for every p N , σ p P exists, so that
σ = σ p σ p p times .
All measures σ P are ⊎-infinitely divisible; see ([21], Theorem 3.6).
A multiplicative version of the additive Boolean convolution was defined in [22]. For λ P + , the function
Ψ λ ( w ) = 0 + w y 1 w y λ ( d y ) , w C R +
is univalent in i C + . We know that Ψ λ ( i C + ) is contained in the circle with diameter ( λ ( { 0 } ) 1 , 0 ) . Furthermore, R Ψ λ ( i C + ) = ( λ ( { 0 } ) 1 , 0 ) . We have that Ψ λ ( w 1 ) = w G λ ( w ) 1 . According to [23], the η -transform of λ P + is introduced as:
η λ : C R + C R + ; w η λ ( w ) = Ψ λ ( w ) 1 + Ψ λ ( w ) .
We known that η λ ( ( , 0 ) ) ( , 0 ) , lim w 0 , w < 0 η λ ( w ) = η λ ( 0 ) = 0 and η λ ( w ¯ ) = η λ ( w ) ¯ , for w C R + . Furthermore, arg ( w ) arg ( η λ ( w ) ) < π , for w C + .
The transform
B λ ( w ) = w η λ ( w )
is well defined for w C R + . The Boolean multiplicative convolution of λ and τ P + is the unique measure in P + , denoted by λ × τ , which satisfies
B λ × τ ( w ) = B λ ( w ) B τ ( w ) , for w C R + .
Note that for λ , τ P + , which satisfies
(i)
arg ( η λ ( w ) ) + arg ( η τ ( w ) ) arg ( w ) < π for w C + ( , 0 )
(ii)
at least one of the first moments of one of measures λ or τ is finite,
So, λ × τ P + is well defined. It was proven in [22] that P + is not preserved by the multiplicative Boolean convolution. However, the multiplicative Boolean convolution power λ × t P + still exists for 0 t 1 .
Now, we present the purpose of this article. In free probability, FMLLN is investigated and formulated in [14] for positive free random variables. By means of probabilities, it may be seen as the convergence of
( λ q ) 1 q , q N
where λ P + and μ α are the push-forward of measure μ by y y α for α R . For the sequence (1), the limit law exists; it is denoted by Φ [ λ ] . If λ δ 0 , the limit law is determined by
Φ [ λ ] 0 , 1 S λ ( y 1 ) = y , y ( λ ( { 0 } ) , 1 ) and Φ [ λ ] ( { 0 } ) = λ ( { 0 } ) .
In ([24], Theorem 3.1), a description is presented for the FMLLN  Φ [ λ ] based on the pseudo-variance function V λ ( · ) of the CSK family induced by λ ; see the next paragraph for these concepts. Some specific examples are provided for Φ [ λ ] ; see [24].
In this article, we aim to expressly provide, by means of V λ ( · ) , the limiting measures Φ [ λ × α ] , Φ D 1 / s ( λ s ) × α , Φ D 1 / r ( λ r ) × α , Φ B r ( λ ) × α , Φ M t ( λ ) , Φ M t D 1 / s ( λ s ) , Φ M t D 1 / r ( λ r ) and Φ M t ( B r ( λ ) ) for t > 0 , r > 0 , s > 1 and α [ 0 , 1 ] . Here, D c ( λ ) denotes the dilation of λ by c 0 , B r ( λ ) = ( λ 1 + r ) 1 1 + r and M t ( λ ) = λ ( t + 1 ) × 1 t + 1 . Several examples are presented for important measures λ in free probability. Section 2 covers fundamental principles of CSK families and the associated pseudo-variance functions. Section 3 presents the article’s main results and some illustrations are given in Section 4.

2. Cauchy–Stieltjes Kernel Families

In free probability, the CSK families (of probability measures) are newly presented using the CSK ( 1 θ y ) 1 . Some characteristics of CSK families are presented in [25] using probabilities with compact support. Extended work on CSK families is presented in [26], covering probabilities with support bounded from one side, say from above. Denote P b a as the set of non-degenerate probabilities with support bounded from above. Let σ P b a . With 1 / θ + ( σ ) = max { sup   supp ( σ ) , 0 } , the function
M σ ( θ ) = 1 1 θ y σ ( d y )
is defined ∀ θ [ 0 , θ + ( σ ) ) . The CSK family induced by σ is the set of probabilities
K + ( σ ) = P ( θ , σ ) ( d y ) = 1 M σ ( θ ) ( 1 θ y ) σ ( d y ) : θ ( 0 , θ + ( σ ) ) .
The map θ K σ ( θ ) = y P ( θ , σ ) ( d y ) realizes a bijection between ( 0 , θ + ( σ ) ) and its image ( m 1 ( σ ) , m + ( σ ) ) , which is said to be the mean domain of K + ( σ ) . The inverse function of K σ ( · ) is denoted by χ σ ( · ) . For m ( m 1 ( σ ) , m + ( σ ) ) , write Q ( m , σ ) ( d y ) = P ( χ σ ( m ) , σ ) ( d y ) , then the mean-parametrization of K + ( σ ) is
K + ( σ ) = { Q ( m , σ ) ( d y ) : m ( m 1 ( σ ) , m + ( σ ) ) } .
From [26], we know that
m 1 ( σ ) = lim θ 0 + K σ ( θ ) and m + ( σ ) = B lim w B + 1 G σ ( w ) ,
with
B = B ( σ ) = max { sup   supp ( σ ) , 0 } = 1 / θ + ( σ ) .
If σ has a one-sided support boundary from below, the CSK family is denoted K ( σ ) and θ ( θ ( σ ) , 0 ) with θ ( σ ) is either 1 / A ( σ ) or and A = A ( σ ) = min { inf   s u p p ( σ ) , 0 } . The mean domain for K ( σ ) is ( m ( σ ) , m 1 ( σ ) ) with m ( σ ) = A 1 / G σ ( A ) . If σ has compact support, then K ( σ ) = K ( σ ) { σ } K + ( σ ) is the two-sided CSK family.
The variance of Q ( m , σ ) is denoted as V σ ( m ) and the function defined on ( m 1 ( σ ) , m + ( σ ) ) by
m V σ ( m ) = ( y m ) 2 Q ( m , σ ) ( d y )
is the variance function (VF) of K + ( σ ) . Unfortunately, if the moment of order 1 of σ P b a does not exists, then all probabilities in K + ( σ ) have infinite variance. This issue has led the authors in [26] to define a notion of pseudo-variance function (PVF) as
V σ ( m ) = m 1 χ σ ( m ) m .
If < m 1 ( σ ) = y σ ( d y ) < + , then the VF exists and (see [26])
V σ ( m ) = m m m 1 ( σ ) V σ ( m ) .
Next, we review some facts useful in the demonstration of the article’s main results.
Remark 1.
(i) 
Let σ P b a . Consider g : y κ y + β , where κ 0 and β R and let g ( σ ) be the image of σ by g. Then,m is sufficiently close to m 1 ( g ( σ ) ) = g ( m 1 ( σ ) ) = κ m 1 ( σ ) + β ,
V g ( σ ) ( m ) = κ 2 m m β V σ m β κ .
(ii) 
V σ ( · ) characterizes σ: If we consider Υ = Υ ( m ) = m + V σ ( m ) m , then
G σ ( Υ ) = m V σ ( m ) .
(iii) 
Let σ P b a . For s > 0 , so that σ s is defined andm close sufficiently to m 1 ( σ s ) = s m 1 ( σ ) (see [26])
V σ s ( m ) = s V σ ( m / s ) .
In addition, r > 0 andm are sufficiently close to m 1 ( σ r ) = r m 1 ( σ ) (see [27])
V σ r ( m ) = r V σ ( m / r ) + m 2 ( 1 / r 1 ) ,
Furthermore,m is sufficiently close to m 1 ( B r ( σ ) ) = m 1 ( σ ) ,
V B r ( σ ) ( m ) = V σ ( m ) + r m 2 .
(iv) 
Let λ P + . For q > 0 , so that λ q is defined and m ( m ( λ q ) , m 1 ( λ q ) ) = ( ( m ( λ ) ) q , ( m 1 ( λ ) ) q ) , (see [28])
V λ q ( m ) = m 2 2 / q V λ m 1 / q .
Also, α [ 0 , 1 ] and m ( m ( λ × α ) , m 1 ( λ × α ) ) = ( ( m ( λ ) ) α , ( m 1 ( λ ) ) α ) , (see [29])
V λ × α ( m ) = m 1 1 / α V λ m 1 / α + m 1 + 1 / α m 2 .
In addition, t > 0 and m ( m ( M t ( λ ) ) , m 1 ( M t ( λ ) ) ) = ( m ( λ ) , m 1 ( λ ) ) ,
V M t ( λ ) ( m ) = m t V λ ( m ) + m 2 ( m t 1 ) .
Remark 2.
Let ν P . For s > 0 , so that ν s is defined, an injective analytic function exists (said subordination function) W s : C + C + , so that G ν s ( ξ ) = G ν ( W s ( ξ ) ) , for ξ C + . In addition,
W s ( ξ ) = ξ / s + ( 1 1 / s ) G ν s ( ξ )
and H s ( W s ( ξ ) ) = ξ , where
H s ( ξ ) = s ξ + ( 1 s ) G ν ( ξ ) .
See ([30], Theorem 2.5).

3. Main Results

We present some new limiting measures connected to the FMLLN and involving Boolean and free (additive and multiplicative) convolutions. More precisely, we have:
Theorem 1.
Let λ be an absolutely continuous measure in P + . With the notations introduced above, α [ 0 , 1 ] , ∀ s > 1 , ∀ t > 0 and r > 0 ,
(i) 
Φ λ × α ( d m ) = 1 m 1 / α 1 1 + m 1 / α 1 V λ ( m 1 / α ) 1 ( m ( λ ) ) α , ( m 1 ( λ ) ) α ( m ) d m .
(ii) 
Φ ( D 1 / s ( λ s ) ) × α ( d m ) = 1 m 1 / α 1 1 + m 1 / α 1 s V λ ( m 1 / α ) 1 W s ( 0 ) s 1 α , m 1 ( λ ) α ( m ) d m .
(iii) 
Φ ( D 1 / r ( λ r ) ) × α ( d m ) = r m 1 / α 1 r + m 1 1 / α V λ ( m 1 / α ) 1 m ( λ ) α , m 1 ( λ ) α ( m ) d m .
(iv) 
Φ ( B r ( λ ) ) × α ( d m ) = 1 ( r + 1 ) m 1 / α 1 1 + m 1 / α 1 V λ ( m 1 / α ) 1 W 1 + r ( 0 ) r α , m 1 ( λ ) α ( m ) d m .
(v) 
Φ M t ( λ ) ( d m ) = 1 m t 2 V λ ( m ) + m t 1 1 ( m ( λ ) , m 1 ( λ ) ) ( m ) d m .
(vi) 
Φ M t D 1 / s ( λ s ) ( d m ) = s m t 2 V λ ( m ) + s ( m t 1 ) 1 W s ( 0 ) s 1 , m 1 ( λ ) ( m ) d m .
(vii) 
Φ M t D 1 / r ( λ r ) ( d m ) = r m t 2 V λ ( m ) + m t r 1 m ( λ ) , m 1 ( λ ) ( m ) d m .
(viii) 
Φ M t B r ( λ ) ( d m ) = 1 m t 2 V λ ( m ) + ( r + 1 ) m t 1 1 W 1 + r ( 0 ) r , m 1 ( λ ) ( m ) d m .
Proof. 
(i) According to ([24], Theorem 3.1), for an absolutely continuous measure λ in P + , the FMLLN Φ [ λ ] is provided by
Φ [ λ ] ( d m ) = m 2 V λ ( m ) 1 ( m ( λ ) , m 1 ( λ ) ) ( m ) d m .
As λ is an absolutely continuous measure on [ 0 , ) , it is the same for λ × α . Combining (8) and (20), we obtain ∀ α [ 0 , 1 ]
Φ λ × α ( d m ) = m 2 V λ × α ( m ) 1 m λ × α , m 1 λ × α ( m ) d m = 1 m 1 / α 1 1 + m 1 / α 1 V λ ( m 1 / α ) 1 ( m ( λ ) ) α , ( m 1 ( λ ) ) α ( m ) d m .
(ii) As λ is an absolutely continuous measure on [ 0 , ) , it is the same for λ s . We have A = A ( λ s ) = min { inf   s u p p ( λ s ) , 0 } = 0 . So,
m ( λ s ) = A 1 / G λ s ( A ) = 1 / G λ s ( 0 ) .
Relations (21) and (10) imply that ∀ s > 1
m ( λ s ) = s W s ( 0 ) s 1 .
Then,
m ( ( D 1 / s ( λ s ) ) × α ) = ( m ( D 1 / s ( λ s ) ) ) α = W s ( 0 ) s 1 α .
Furthermore,
m 1 ( ( D 1 / s ( λ s ) ) × α ) = ( m 1 ( D 1 / s ( λ s ) ) ) α = m 1 ( λ ) α .
Using (8), (4) and (2), ∀ s > 1 and ∀ α [ 0 , 1 ] we obtain
V ( D 1 / s ( λ s ) ) × α ( m ) = m 1 1 / α s V λ ( m 1 / α ) + m 1 + 1 / α m 2 .
Combining (20), (23)–(25), we obtain ∀ s > 1 and ∀ α [ 0 , 1 ]
Φ ( D 1 / s ( λ s ) ) × α ( d m ) = 1 m 1 / α 1 1 + m 1 / α 1 s V λ ( m 1 / α ) 1 W s ( 0 ) s 1 α , m 1 ( λ ) α ( m ) d m .
(iii) As λ is an absolutely continuous measure on [ 0 , ) , it is the same for λ r . We have A ( λ r ) = min { inf   s u p p ( λ r ) , 0 } = 0 . So,
m ( λ r ) = 1 / G λ r ( 0 ) = r / G λ ( 0 ) = r m ( λ ) .
Then, ∀ r > 0 and ∀ α [ 0 , 1 ]
m ( ( D 1 / r ( λ r ) ) × α ) = ( m ( D 1 / r ( λ r ) ) ) α = m ( λ ) α .
Furthermore,
m 1 ( ( D 1 / r ( λ r ) ) × α ) = ( m 1 ( D 1 / r ( λ r ) ) ) α = m 1 ( λ ) α .
Using (8), (5) and (2), ∀ r > 0 and ∀ α [ 0 , 1 ] we get
V ( D 1 / r ( λ r ) ) × α ( m ) = m 1 1 / α r V λ ( m 1 / α ) + m 1 + 1 / α r m 2 .
Combining (20), (27)–(29), we obtain ∀ r > 0 and ∀ α [ 0 , 1 ]
Φ ( D 1 / r ( λ r ) ) × α ( d m ) = r m 1 / α 1 r + m 1 1 / α V λ ( m 1 / α ) 1 m ( λ ) α , m 1 ( λ ) α ( m ) d m .
(iv) Relations (22) and (26) produce
m ( B r ( λ ) ) = m λ 1 + r 1 1 + r = 1 1 + r m ( λ 1 + r ) = W 1 + r ( 0 ) r .
Thus, ∀ r > 0 and ∀ α [ 0 , 1 ] we have
m ( ( B r ( λ ) ) × α ) = ( m ( B r ( λ ) ) ) α = W 1 + r ( 0 ) r α .
We also have,
m 1 ( ( B r ( λ ) ) × α ) = ( m 1 ( B r ( λ ) ) ) α = m 1 ( λ ) α .
Using (8) and (6), ∀ r > 0 , ∀ α [ 0 , 1 ] , and ∀ m < m 1 ( λ ) α close enough to m 1 ( λ ) α , we have
V ( ( B r ( λ ) ) × α ) ( m ) = ( r + 1 ) m 1 / α + 1 m 2 + m 1 / α + 1 V λ ( m 1 / α )
Combining (20), (30)–(32), we obtain ∀ r > 0 and ∀ α [ 0 , 1 ]
Φ ( B r ( λ ) ) × α ( d m ) = 1 ( r + 1 ) m 1 / α 1 1 + m 1 / α 1 V λ ( m 1 / α ) 1 W 1 + r ( 0 ) r α , m 1 ( λ ) α ( m ) d m .
(v) Relation (16) follows by combining (20) and (9).
(vi) ∀ t > 0 and ∀ s > 1 , we have
m M t D 1 / s ( λ s ) = m D 1 / s ( λ s ) = W s ( 0 ) s 1 ,
and
m 1 M t D 1 / s ( λ s ) = m 1 D 1 / s ( λ s ) = m 1 ( λ ) .
Furthermore, by combining (9), (4) and (2), we obtain
V M t D 1 / s ( λ s ) ( m ) = m t s V λ ( m ) + m 2 ( m t 1 ) .
Relation (17) follows by combining (20), (33)–(35).
(vii) ∀ t > 0 and ∀ r > 0 , we have
m M t D 1 / r ( λ r ) = m D 1 / r ( λ r ) = m ( λ ) ,
and
m 1 M t D 1 / r ( λ r ) = m 1 D 1 / r ( λ r ) = m 1 ( λ ) .
Furthermore, by combining (9), (5) and (2), we obtain
V M t D 1 / r ( λ r ) ( m ) = m t r V λ ( m ) + m 2 ( m t / r 1 ) .
Relation (18) follows by combining (20), (36)–(38).
(viii) ∀ t > 0 and ∀ r > 0 , we have
m M t B r ( λ ) = m B r ( λ ) = W 1 + r ( 0 ) r ,
and
m 1 M t B r ( λ ) = m 1 B r ( λ ) = m 1 ( λ ) .
Furthermore, by combining (9) and (6), we obtain
V M t B r ( λ ) ( m ) = m t V λ ( m ) + m 2 ( ( r + 1 ) m t 1 ) .
Relation (19) follows by combining (20), (39)–(41). □

4. Examples

We provide some examples of the limiting laws presented in Theorem 1, using important probability measures λ in free probability.
Example 1.
Consider the Semicircle law
γ ( d y ) = 4 y 2 2 π 1 ( 2 , 2 ) ( y ) d y ,
with m 1 ( γ ) = 0 . We have
V γ ( m ) = 1 = V γ ( m ) , m ( m ( γ ) , m 1 ( γ ) ) = ( 1 , 0 ) a n d G γ ( w ) = 1 2 w w 2 4 .
Consider g : y y + 2 . Then,
λ ( d y ) = g ( γ ) ( d y ) = y ( 4 y ) 2 π 1 ( 0 , 4 ) ( y ) d y ,
with m 1 ( λ ) = 2 . We have
V λ ( m ) = m m 2 , m ( m ( λ ) , m 1 ( λ ) ) = ( 1 , 2 ) a n d G λ ( w ) = 1 2 w 2 w ( w 4 ) .
From (11), we have that
H s ( w ) = s w 2 ( s 1 ) w 2 w ( w 4 )
and, consequently,
W s ( u ) = 2 s 2 s 2 + u + s u ( s 1 ) 4 s ( s 1 ) 4 s u + u 2 2 s
u C ( 2 s 2 s , 2 s + 2 s ) . Then,
W s ( 0 ) = ( s 1 ) 1 s 1 s .
Then, we have the following:
(i) α [ 0 , 1 ] ,
Φ λ × α ( d m ) = m 1 / α 2 ( m 1 / α 2 ) ( m 1 / α 1 1 ) + 1 / m 1 1 , 2 α ( m ) d m .
(ii) s > 1 and α [ 0 , 1 ] ,
Φ ( D 1 / s ( λ s ) ) × α ( d m ) = m 1 / α 2 ( m 1 / α 2 ) ( m 1 / α 1 1 ) + 1 s m 1 1 + s 1 s α , 2 α ( m ) d m .
(iii) r > 0 and α [ 0 , 1 ] ,
Φ ( D 1 / r ( λ r ) ) × α ( d m ) = r ( m 1 / α 2 ) ( m 1 / α 2 ) ( m 1 / α 1 r ) + 1 / m 1 1 , 2 α ( m ) d m .
(iv) r > 0 and α [ 0 , 1 ] ,
Φ ( B r ( λ ) ) × α ( d m ) = ( m 1 / α 2 ) ( m 1 / α 2 ) ( ( r + 1 ) m 1 / α 1 1 ) + 1 / m 1 1 + r r + 1 α , 2 α ( m ) d m .
(v) t > 0 ,
Φ M t ( λ ) ( d m ) = m 2 m t 1 + ( m t 1 ) ( m 2 ) 1 1 , 2 ( m ) d m .
(vi) t > 0 and s > 1 ,
Φ M t D 1 / s ( λ s ) ( d m ) = s ( m 2 ) m t 1 + s ( m t 1 ) ( m 2 ) 1 1 + s 1 s , 2 ( m ) d m .
(vii) t > 0 and r > 0 ,
Φ M t D 1 / r ( λ r ) ( d m ) = r ( m 2 ) m t 1 + ( m t r ) ( m 2 ) 1 1 , 2 ( m ) d m .
(viii) t > 0 and r > 0 ,
Φ M t B r ( λ ) ( d m ) = ( m 2 ) m t 1 + ( ( r + 1 ) m t 1 ) ( m 2 ) 1 1 + r 1 + r , 2 ( m ) d m .
Example 2.
For 0 < a 2 < 1 , the Marchenko-Pastur measure is provided by
γ ( d y ) = 4 ( y a ) 2 2 π ( 1 + a y ) 1 ( a 2 , a + 2 ) ( y ) d y .
We have
V γ ( m ) = 1 + a m = V γ ( m ) , m ( m ( γ ) , m 1 ( γ ) ) = ( 1 , 0 ) a n d G γ ( w ) = a + w ( a w ) 2 4 2 ( 1 + a w ) .
Consider f : y a y + 1 . We have,
λ ( d y ) = f ( γ ) ( d y ) = ( a + 1 ) 2 y y ( a 1 ) 2 2 π a 2 y 1 ( ( a 1 ) 2 , ( a + 1 ) 2 ) ( y ) d y ,
with m 1 ( λ ) = 1 , and
V λ ( m ) = a 2 m 2 m 1 .
The Cauchy transform of λ is
G λ ( w ) = 1 a G γ w 1 a = 1 2 w a a + ( w 1 ) a a ( w 1 ) a 2 4 .
For m > 0 , solving m + V λ ( m ) / m = 0 , one obtains m = 1 a 2 . Relation (3) implies G λ ( 0 ) = 1 1 a 2 . Thus, m ( λ ) = A ( λ ) 1 / G λ ( A ( λ ) ) = 1 / G λ ( 0 ) = 1 a 2 . So, ( m ( λ ) , m 1 ( λ ) ) = ( 1 a 2 , 1 ) .
We have that
H s ( w ) = s w 2 w a ( s 1 ) a + ( w 1 ) a a ( w 1 ) a 2 4
and
W s ( u ) = a 2 + s + a 2 s s 2 + u + s u ( s 1 ) ( s a 2 ) 2 2 u ( a 2 + s ) + u 2 2 s ,
u C ( a 2 + s 2 s a 2 , a 2 + s + 2 s a 2 ) . Then
W s ( 0 ) = ( s 1 ) ( s a 2 ) s .
We have:
(i) α [ 0 , 1 ] ,
Φ λ × α ( d m ) = m 1 / α 1 ( m 1 / α 1 ) ( m 1 / α 1 1 ) + a 2 m 1 / α 1 1 ( 1 a 2 ) α , 1 ( m ) d m .
(ii) s > 1 and α [ 0 , 1 ] ,
Φ ( D 1 / s ( λ s ) ) × α ( d m ) = m 1 / α 1 ( m 1 / α 1 ) ( m 1 / α 1 1 ) + a 2 s m 1 / α 1 1 s a 2 s α , 1 ( m ) d m .
(iii) r > 0 and α [ 0 , 1 ] ,
Φ ( D 1 / r ( λ r ) ) × α ( d m ) = r ( m 1 / α 1 ) ( m 1 / α 1 ) ( m 1 / α 1 r ) + a 2 m 1 / α 1 1 ( 1 a 2 ) α , 1 ( m ) d m .
(iv) r > 0 and α [ 0 , 1 ] ,
Φ ( B r ( λ ) ) × α ( d m ) = ( m 1 / α 1 ) ( m 1 / α 1 ) ( ( r + 1 ) m 1 / α 1 1 ) + a 2 m 1 / α 1 1 r + 1 a 2 r + 1 α , 1 ( m ) d m .
(v) t > 0 ,
Φ M t ( λ ) ( d m ) = m 1 a 2 m t + ( m t 1 ) ( m 1 ) 1 1 a 2 , 1 ( m ) d m .
(vi) t > 0 and s > 1 ,
Φ M t D 1 / s ( λ s ) ( d m ) = s ( m 1 ) a 2 m t + s ( m t 1 ) ( m 1 ) 1 s a 2 s , 1 ( m ) d m .
(vii) t > 0 and r > 0 ,
Φ M t D 1 / r ( λ r ) ( d m ) = r ( m 1 ) a 2 m t + ( m t r ) ( m 1 ) 1 1 a 2 , 1 ( m ) d m .
(viii) t > 0 and r > 0 ,
Φ M t B r ( λ ) ( d m ) = ( m 1 ) a 2 m t + ( ( r + 1 ) m t 1 ) ( m 1 ) 1 r + 1 a 2 r + 1 , 1 ( m ) d m .
Example 3.
The free Gamma measure is provided, for a 0 , by
γ ( d y ) = 4 ( 1 + a 2 ) ( y 2 a ) 2 2 π ( a 2 y 2 + 2 a y + 1 ) 1 ( 2 1 + a 2 + 2 a , 2 1 + a 2 + 2 a ) ( y ) ( d y ) ,
with m 1 ( γ ) = 0 . We have
V γ ( m ) = V γ ( m ) = ( 1 + a m ) 2 a n d G γ ( w ) = 2 a + w + 2 a 2 w ( 2 a w ) 2 4 ( 1 + a 2 ) 2 ( 1 + a w ) 2 .
Consider h : y a y + 1 . Then,
λ ( d y ) = h ( γ ) ( d y ) = ( ( a + a 2 + 1 ) 2 y ) ( y ( a + a 2 + 1 ) 2 ) 2 π a 2 y 2 1 ( ( a 2 + 1 | a | ) 2 , ( a 2 + 1 + | a | ) 2 ) ( y ) d y ,
with m 1 ( λ ) = 1 . We have
V λ ( m ) = a 2 m 3 m 1 , m ( m ( λ ) , m 1 ( λ ) ) = 1 1 + a 2 , 1
and
G λ ( w ) = w 1 a + 2 a w ( 2 a w 1 a ) 2 4 ( 1 + a 2 ) 2 a w 2 .
We also have
H s ( w ) = s w 2 ( s 1 ) a w 2 w 1 a + 2 a w ( 2 a w 1 a ) 2 4 ( 1 + a 2 )
and
W s ( u ) = s s 2 + u ( 1 + 2 a 2 + s ) ( s 1 ) s 2 ( 4 a 2 + 2 s ) u + u 2 2 ( a 2 + s )
u C ( 2 a 2 + s ( 2 a 2 + s ) 2 1 , 2 a 2 + s + ( 2 a 2 + s ) 2 1 ) . Then,
W s ( 0 ) = s ( s 1 ) a 2 + s .
We have:
(i) α [ 0 , 1 ] ,
Φ λ × α ( d m ) = m 1 / α 1 ( m 1 / α 1 ) ( m 1 / α 1 1 ) + a 2 m 2 / α 1 1 1 1 + a 2 α , 1 ( m ) d m .
(ii) s > 1 and α [ 0 , 1 ] ,
Φ ( D 1 / s ( λ s ) ) × α ( d m ) = m 1 / α 1 ( m 1 / α 1 ) ( m 1 / α 1 1 ) + a 2 s m 2 / α 1 1 s s + a 2 α , 1 ( m ) d m .
(iii) r > 0 and α [ 0 , 1 ] ,
Φ ( D 1 / r ( λ r ) ) × α ( d m ) = r ( m 1 / α 1 ) ( m 1 / α 1 ) ( m 1 / α 1 r ) + a 2 m 2 / α 1 1 1 1 + a 2 α , 1 ( m ) d m .
(iv) r > 0 and α [ 0 , 1 ] ,
Φ ( B r ( λ ) ) × α ( d m ) = ( m 1 / α 1 ) ( m 1 / α 1 ) ( ( r + 1 ) m 1 / α 1 1 ) + a 2 m 2 / α 1 1 r + 1 r + 1 + a 2 α , 1 ( m ) d m .
(v) t > 0 ,
Φ M t ( λ ) ( d m ) = m 1 a 2 m t + 1 + ( m t 1 ) ( m 1 ) 1 1 1 + a 2 , 1 ( m ) d m .
(vi) t > 0 and s > 1 ,
Φ M t D 1 / s ( λ s ) ( d m ) = s ( m 1 ) a 2 m t + 1 + s ( m t 1 ) ( m 1 ) 1 s s + a 2 , 1 ( m ) d m .
(vii) t > 0 and r > 0 ,
Φ M t D 1 / r ( λ r ) ( d m ) = r ( m 1 ) a 2 m t + 1 + ( m t r ) ( m 1 ) 1 1 1 + a 2 , 1 ( m ) d m .
(viii) t > 0 and r > 0 ,
Φ M t B r ( λ ) ( d m ) = ( m 1 ) a 2 m t + 1 + ( ( r + 1 ) m t 1 ) ( m 1 ) 1 r + 1 r + 1 + a 2 , 1 ( m ) d m .

5. Conclusions

The law of large numbers (LLN) for the classical additive convolution in classical probability can be used to derive the LLN for the classical multiplicative convolution. In the situation of non-commutative probability, this is not true. For probability measures with bounded support, the free additive law was proven in [31], and it was extended to all measures with finite first moment in [15]. For measures with restricted support, the FMLLN was proven in [13], and it was extended to measures with unbounded support in [14]. Except in the situation where the original measure is a Dirac measure, the limiting measure for the FMLLN is not a Dirac measure, in contrast with the situation of classical multiplicative convolution. Using the variance functions, an intriguing explanation of the FMLLN is provided in [24]. The FMLLN for different kinds of non-commutative probability measures including the Boolean and free (additive and multiplicative) convolutions has been clearly shown in this study. Researchers studying non-commutative probability may find the results helpful as they are explained using a variety of probability metrics. The approach used in this paper provides an explicit expression of the limiting distributions basing on the variance function, which plays a crucial role in the setting of CSK families.

Author Contributions

Conceptualization, A.R.A.A. and S.S.A.; methodology, R.F.; software, A.R.A.A.; validation, A.R.A.A., S.S.A. and R.F.; formal analysis, S.S.A.; investigation, S.S.A.; resources, S.S.A.; data curation, R.F.; writing—original draft preparation, R.F.; writing—review and editing, R.F.; visualization, A.R.A.A.; supervision, A.R.A.A.; project administration, A.R.A.A.; funding acquisition, A.R.A.A. All authors have read and agreed to the published version of the manuscript.

Funding

This research is funded by the Deanship of Graduate Studies and Scientific Research at Jouf University through the Fast-Track Research Funding Program.

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Banna, M.; Merlevède, F. Limiting Spectral Distribution of Large Sample Covariance Matrices Associated with a Class of Stationary Processes. J. Theor. Probab. 2015, 28, 745–783. [Google Scholar] [CrossRef]
  2. Anderson, G.; Zeitouni, O. A law of large numbers for finite-range dependent random matrices. Comm. Pure Appl. Math. 2008, 61, 1118–1154. [Google Scholar] [CrossRef]
  3. Banna, M.; Merlevède, F.; Peligrad, M. On the limiting spectral distribution for a large class of symmetric random matrices with correlated entries. Stoch. Processes Their Appl. 2015, 125, 2700–2726. [Google Scholar] [CrossRef]
  4. Merlevède, F.; Peligrad, M. On the empirical spectral distribution for matrices with long memory and independent rows. Stoch. Processes Their Appl. 2016, 126, 2734–2760. [Google Scholar] [CrossRef]
  5. Pastur, L.; Shcherbina, M. Eigenvalue distribution of large random matrices. In Mathematical Surveys and Monographs; American Mathematical Society: Providence, RI, USA, 2011; Volume 171. [Google Scholar]
  6. Liu, W. Relations between convolutions and transforms in operator-valued free probability. Adv. Math. 2021, 390, 107949. [Google Scholar] [CrossRef]
  7. Capitaine, M.; Donati-Martin, C.; Féral, D.; Février, M. Free Convolution with a Semicircular Distribution and Eigenvalues of Spiked Deformations of Wigner Matrices. Electron. J. Probab. 2011, 16, 1750–1792. [Google Scholar] [CrossRef]
  8. Bai, Z.D.; Yin, Y.Q. Necessary and sufficient conditions for almost sure convergence of the largest eigenvalue of a Wigner matrix. Ann. Probab. 1988, 16, 1729–1741. [Google Scholar] [CrossRef]
  9. Benaych-Georges, F.; Nadakuditi, R.R. The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices. Adv. Math. 2011, 227, 494–521. [Google Scholar] [CrossRef]
  10. Furedi, Z.; Komlos, J. The eigenvalues of random symmetric matrices. Combinatorica 1981, 1, 233–241. [Google Scholar] [CrossRef]
  11. Wang, J.-C. Limit laws for Boolean convolutions. Pac. J. Math. 2008, 237, 349–371. [Google Scholar] [CrossRef]
  12. Bercovici, H.; Pata, V. Stable laws and domains of attraction in free probability theory (with an appendix by Philippe Biane). Ann. Math. 1999, 149, 1023–1060. [Google Scholar] [CrossRef]
  13. Tucci, G.H. Limits laws for geometric means of free random variables. Indiana Univ. Math. J. 2010, 59, 1–13. [Google Scholar] [CrossRef]
  14. Haagerup, U.; Möller, S. The Law of Large Numbers for the Free Multiplicative Convolution. In Operator Algebra and Dynamics; Proceedings in Mathematics and Statistics; Springer: Berlin/Heidelberg, Germany, 2013; Volume 58. [Google Scholar]
  15. Lindsay, J.M.; Pata, V. Some weak laws of large numbers in noncommutative probability. Math. Z. 1997, 226, 533–543. [Google Scholar] [CrossRef]
  16. Chistyakov, G.P.; Götze, F. Limit theorems in free probability theory I. Ann. Probab. 2008, 36, 54–90. [Google Scholar] [CrossRef]
  17. Ueda, Y. Max-convolution semigroups and extreme values in limit theorems for the free multiplicative convolution. Bernoulli 2021, 27, 502–531. [Google Scholar] [CrossRef]
  18. Bercovici, H.; Voiculescu, D. Free convolution of measures with unbounded support. Indiana Univ. Math. J. 1993, 42, 733–773. [Google Scholar] [CrossRef]
  19. Nica, A.; Speicher, R. On the multiplication of free N-tuples of noncommutative random variables. Amer. J. Math. 1996, 118, 799–837. [Google Scholar] [CrossRef]
  20. Belinschi, S.T. Complex Analysis Methods in Noncommutative Probability; ProQuest LLC: Ann Arbor, MI, USA, 2005. [Google Scholar]
  21. Speicher, R.; Woroudi, R. Boolean convolution. Fields Inst. Commun. 1997, 12, 267–279. [Google Scholar]
  22. Bercovici, H. On Boolean convolutions. Oper. Theory 2006, 20, 7–13. [Google Scholar]
  23. Anshelevich, M.; Wang, J.-C.; Zhong, P. Local limit theorems for multiplicative free convolutions. J. Funct. Anal. 2014, 267, 3469–3499, ISSN 0022-1236. [Google Scholar] [CrossRef]
  24. Fakhfakh, R. Explicit free multiplicative law of large numbers. Commun.-Stat.-Theory Methods 2023, 52, 2031–2042. [Google Scholar] [CrossRef]
  25. Bryc, W. Free exponential families as kernel families. Demonstr. Math. 2009, 42, 657–672. [Google Scholar] [CrossRef]
  26. Bryc, W.; Hassairi, A. One-sided Cauchy-Stieltjes kernel families. J. Theoret. Probab. 2011, 24, 577–594. [Google Scholar] [CrossRef]
  27. Fakhfakh, R. Variance function of boolean additive convolution. Stat. Probab. Lett. 2020, 163, 108777. [Google Scholar] [CrossRef]
  28. Fakhfakh, R.; Hassairi, A. Cauchy-Stieltjes kernel families and free multiplicative convolution. Commun. Math. Stat. 2023. [Google Scholar] [CrossRef]
  29. Fakhfakh, R. Boolean multiplicative convolution and Cauchy-Stieltjes Kernel families. Bull. Korean Math. Soc. 2021, 58, 515–526. [Google Scholar] [CrossRef]
  30. Belinschi, S.T.; Bercovici, H. Atoms and regularity for measures in a partially defined free convolution semigroup. Math. Z. 2004, 248, 665–674. [Google Scholar] [CrossRef]
  31. Voiculescu, D. Voiculescu, D. Addition of certain noncommuting random variables. J. Funct. Anal. 1986, 66, 323–346. [Google Scholar] [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Alanzi, A.R.A.; Alshqaq, S.S.; Fakhfakh, R. Some Limiting Measures Related to Free and Boolean Convolutions. Symmetry 2024, 16, 1365. https://doi.org/10.3390/sym16101365

AMA Style

Alanzi ARA, Alshqaq SS, Fakhfakh R. Some Limiting Measures Related to Free and Boolean Convolutions. Symmetry. 2024; 16(10):1365. https://doi.org/10.3390/sym16101365

Chicago/Turabian Style

Alanzi, Ayed. R. A., Shokrya S. Alshqaq, and Raouf Fakhfakh. 2024. "Some Limiting Measures Related to Free and Boolean Convolutions" Symmetry 16, no. 10: 1365. https://doi.org/10.3390/sym16101365

APA Style

Alanzi, A. R. A., Alshqaq, S. S., & Fakhfakh, R. (2024). Some Limiting Measures Related to Free and Boolean Convolutions. Symmetry, 16(10), 1365. https://doi.org/10.3390/sym16101365

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop