Free-Boolean

# Free-Boolean independence with amalgamation

Weihua Liu Weihua Liu: Department of Mathematics
Indiana University
Boomington, IN 47401, USA.
and  Ping Zhong Ping Zhong: Department of Pure Mathematics
University of Waterloo
200 University Avenue West
###### Abstract.

In this paper, we develop the notion of free-Boolean independence in an amalgamation setting. We construct free-Boolean cumulants and show that the vanishing of mixed free-Boolean cumulants is equivalent to our free-Boolean independence with amalgamation. We also provide a characterization of free-Boolean independence by conditions in terms of mixed moments. In addition, we study free-Boolean independence over a -algebra and prove a positivity property.

## 1. Introduction

Free probability theory is a probability theory that studies noncommutative random variables with highest noncommutativity. This theory, due to Voiculescu, is based on the notion of free independence which is an analogue of the classical independence. In , Voiculescu generalized his notion of free independence to free independence with amalgamation over an arbitrary algebra in details. To be specific, moments of random variables are no longer scalar numbers but elements from a given algebra. On the other aspect, Voiculescu started to study pairs of random variables simultaneously thereby generalized the notion of free independence to a notion of bi-free independence . Further more, the notion of bi-free independence with amalgamation, defined by Voiculescu , was fully developed in . there are exactly two unital universal independence relations, namely Voiculescu’s free independence relation, the classical independence relation . It was mentioned that we would obtain more independence relations by decreasing the number of axioms for universal products . For instance, people introduced Boolean independence , monotone independence , conditionally independence  in various contexts. Their operator-valued generalization were studied as well [9, 12, 13]. Recently, their corresponding independence relations for pairs of random variables, analog of Voiculescu’s bi-free theory, were introduced and studied [4, 3, 6]. Furthermore, the conditionally bi-free independence with amalgamation is studied in .

In , the first-named author introduced a notion of mixed independence relations for pairs of random variables, where random variables in different faces exhibit different kinds of noncommutative independence. In particular, the combinatorics of free-Boolean independence relation was fully developed. In this paper, we generalize the notion of free-Boolean independence to an amalgamation setting. Relevant combinatorial tools are extended to study this new independence. Beyond the corresponding combinatorial results, we address the positivity of free-Boolean independence with amalgamation. Therefore, it is possible to study the relation in topological probability spaces but not only algebraic probability spaces. For instance, we can study our free-Boolean independence with amalgamation over a -algebra, which is a suitable framework to address some probabilistic questions. We plan to study probabilistic results such as operator-valued infinitely divisible laws in a forthcoming paper.

The paper is organized as follows: Besides this introduction, in Section 2, we give the definition of free-Boolean independence with amalgamation over an algebra. In Section 3, we review some relevant combinatorial tools. In Section 4, we demonstrate that free-Boolean independence can be characterized by the property of the vanishing of mixed free-Boolean cumulants. In Section 5, we prove an operator-valued version of free-Boolean central limit law. In Section 6, we provide an equivalent characterization of free-Boolean independence by certain moments-conditions. In Section 7, we study the positivity property for free-Boolean independence relation.

## 2. Preliminaries and Notations

In this section, we give the motivation and the definition for free-Boolean independence relation with amalgamation over an algebra.

###### Definition 2.1.

A --bimodule with a specified projection is a triple , where is a direct sum of --bimodules , and is the projection

 p(b⊕η)=b.

Denote by the algebra of linear operators with respect to the --bimodule structure. The expectation from onto is the linear map defined by

 EL(X)(a)=p(a(1B⊕0)).

We now recall the definition of the reduced free product of --bimodules with specified projections [16, 18]. Let be a family of --bimodules with specified projections. The reduce free product of with amalgamation over is defined to be the --bimodule with a specified projection , where and is the --bimodule defined by

 ˚X=⨁n≥1⨁i1≠i2≠⋯≠in˚Xi1⊗B⋯⊗B˚Xin.

For each , we denote by

 X(i)=B⊕⨁n≥1⨁i1≠i2≠⋯≠ini1≠i˚Xi1⊗B⋯⊗B˚Xin,

and let be the natural isomorphism of bimodules

 Vi:X→Xi⊗BX(i).

For each , is a unital homomorphism defined by

 λi(a)=V−1i(a⊗I)Vi

and is a linear map defined by

 βi(a)=Piλi(a)Pi,

where is the natural projection onto and vanishes on the other direct summands.

###### Proposition 2.2.

For any , we have .

###### Proof.

Notice that the reduced free product of --bimodules with specified projections can be decomposed as

 X=(B⊕˚Xi)⊕X′i,whereX′i=⨁j≠i˚Xj⊕⨁n≥2⨁ci1≠i2≠⋯≠in˚Xi1⊗B⋯⊗B˚Xin.

The space is invariant under for any . We can check directly that the space is also invariant under for any by the definition of . Hence the result follows. ∎

The preceding result implies the next corollary.

###### Corollary 2.3.

The map is a homomorphism.

###### Definition 2.4.

A -valued probability space is a pair consisting of an algebra over and an --bimodule map , i.e. a linear map such that

 E(b1ab2)=b1E(a)b2

for all and .

###### Definition 2.5.

Let be a -valued probability space. A family of -faces of is a family of (not necessarily unital) subalgebras of such that are --bimodules for each . The family of -faces is said to be free-Boolean with amalgamation over if

• are unital algebras

• there are --bimodules with specified projections such that there are unital homomorphisms , (not necessarily unital) homomorphisms ,

• Let be the reduce free product of , so that the joint distribution of the family in is equal to the joint distribution of the family of operators in the probability space . That is,

 EL(X)(λi1(γi1(c1))βi1(δi1(d1))⋯λin(γin(cn))βin(δin(dn)))=E(c1d1⋯cndn),

where

## 3. Interval-noncrossing partitions

In this section, we review some combinatorial tools which will be used

to define operator-valued free-Boolean cumulants. We give a characterization of free-Boolean independence with amalgamation thereby generalizes results in  to the operator-valued framework. In noncommutative probability theory, non-crossing partitions are used in the combinatorics of free probability and the interval partitions are used in the combinatorics of Boolean independence. It turns out the partitions used in the combinatorics of free-Boolean independence are so-called interval-noncrossing partitions introduced in . All results without proof in this section are taken from .

### 3.1. Interval-noncrossing partitions

Throughout this section, we let , and , for some fixed index set . We will denote by the set for .

###### Definition 3.1.

Let be a linearly ordered set. A partition of the set consists of a collection disjoint, nonempty sets whose union is . The sets are called the blocks of . Given , we write if the two elements are in the same block.

• A partition is called noncrossing if there is no quadruple such that , , and , where are two disjoint blocks of . The set of all noncrossing partitions of will be denoted by .

• A block of is called interval if for any and , we have . A partition is called an interval partition if all blocks are interval blocks.

• A block of a partition is said to be inner if there is another block and such that for all . A block is outer if it is not inner.

• Let . We denote by the partition whose blocks are the sets .

• Given two partitions and , we say if each block of is contained in a block of . This relation is called the reversed refinement order.

• We denote by the partition of consists of blocks and by the partition of consists of exactly one block.

###### Definition 3.2.

Let . A partition of is said to be interval-noncrossing with respect to if is noncrossing, and are in the same block whenever , and . We denote by the set of all interval-noncrossing partitions of the set with respect to .

###### Remark 3.3.

The set does not depend on the value of at and . In particular, when , we have .

For example, let such that . Given two noncrossing and of the set , then and . In pictures below, we use ""to denote elements in and ""to denote elements in .

Assume now and set . We denote by the interval . For each , we denote by the restriction of to the interval . We also denote by the restriction of to the interval and the restriction of to the interval . Note that each can be any noncrossing partition of the set , since there is no such that .

###### Proposition 3.4.

Let be defined by

 α′1(π)=(α1(π),α′(π))

and be defined by

 α(π)=(α1(π),⋯,αm(π)).

Then and are isomorphisms of partial ordered sets. The set is a lattice with respect to the reverse refinement order on partitions.

We provide pictures below to illustrate the preceding proposition. Let , and which is an interval-noncrossing of the set with respect to as shown in the following diagram.

In the above diagram, , , . Therefore, , , , and are illustrated in the following diagrams:

###### Proposition 3.5.

Let and let such that , i.e. each block of is contained in a block of . Then if and only if for all .

###### Proposition 3.6.

Let . Denote by the set of all such that . Then

 [0n,π]≅INC(χ|V1)×⋯×INC(χ|Vp).

### 3.2. Möbius functions on interval-noncrossing partitions

One can define the convolution for functions on the lattice following the standard procedure for partially ordered sets (see ). Once the map is fixed, the lattice structure of caputred from the lattice of the product of noncrossing partitions according to the natural isomorphism described in Proposition 3.4.

Let . Given two complex-valued functions defined on the set . The convolution of and is given by

 f∗g(σ,π)=∑ρ∈INC(χ)σ≤ρ≤πf(σ,ρ)g(ρ,π).

The delta function defined as follows:

 δINC(σ,π)={1,ifσ=π,0,otherwise.

We then define the zeta function by

 ζINC(σ,π)={1,ifσ≤π,0,otherwise.

and the Möbius function is the inverse of the zeta function in the following sense:

 μINC∗ζINC=ζINC∗μINC=δINC.

We will use the following product formula in [7, Section 6].

###### Proposition 3.7.

Let , and such that . Suppose that and , then

 μINC(χ)(σ,π)=m∏i=1μINC(χ|Vi)(σi,πi)=p∏s=1μINC(σ|Vs,1Vs)=m∏i=1p∏s=1μNC(σi|~αi(Vs),1~αi(Vs)),

where , , and is the restriction of to the set .

###### Corollary 3.8.

Let and . Denote by , and . Then,

 [0n,π]≅INC(χ|V)×INC(χ|V′).

In particular, we have

 μINC(σ,π)=μINC(σ|V,1V)μINC(σ|V′,π|V′)

for and .

In this section, we introduce the notion of operator-valued free-Boolean cumulants for pairs of random variables and give an alternative characterization of free-Boolean independence by using the free-Boolean cumulants.

### 4.1. Free-Boolean cumulants

Let be a -valued probability space . Let be the -B-linear map from to defined as

 Φ(n)(a1,⋯,an)=E(a1⋯an).

Then, for each noncrossing partition , we can write , where is an interval block of and . We define an -B-linear map recursively as follows:

 (1) Φπ(a1,⋯,an)=Φπ1(a1,⋯,al,Φ(s)(al+1,⋯,al+s)al+s+1,⋯,an).

For example, let be a noncrossing partition of . Then,

 Φπ(a1,⋯,a8)=Φ{{1,5,8},{6,7}}(a1,E(a2a3a4)a5,a6,a7,a8)=E(a1E(a2a3a4)a5E(a6a7)a8).
###### Definition 4.1.

Given any , and a tuple of elements in , the free-Boolean cumulant is an --linear map defined as follows:

 κχ,π(a1,⋯,an)=∑σ≤πσ∈INC(χ)μINC(σ,π)Φσ(a1,⋯,an).

We start to show that the operator-valued free-Boolean cumulants have the following multiplicative property.

###### Theorem 4.2.

Let and be noncommutative random variables in a -valued probability space . Suppose that is an interval block of , then

where

###### Proof.

For any , , one can decompose it into a union of two interval-noncrossing partitions , where , and , . By Proposition 3.7 and Corollary 3.8, we have

 κχ,π(a1,⋯,an)\par=∑σ≤πσ∈INC(χ)μINC(σ,π)Φσ(a1,⋯,an)\par=∑σ≤πσ∈INC(χ)μINC(σ,π)Φσ|V′(a1,⋯,al,Φσ|V(al+1,⋯,al+s)al+s+1,⋯,an)\par\par=∑σ≤πσ∈INC(χ)μINC(σ|V′,π|V′)μINC(σ|V,1V)Φσ|V′(a1,⋯,al,Φσ|V(al+1,⋯,al+s)al+s+1,⋯,an)\par\par=∑σ1≤π|V′σ1∈INC(χ|V′)σ2∈INC(χ|V)μINC(σ1,π|V′)μINC(σ2,1V)Φσ|V′(a1,⋯,al,Φσ|V(al+1,⋯,al+s)al+s+1,⋯,an)\par=∑σ1≤π|V′σ1∈INC(χ|V′)μINC(σ1,π|V′)Φσ|V′(a1,⋯,al,∑σ2∈INC(χ|V)μINC(σ2,1V)Φσ|V(al+1,⋯,al+s)al+s+1,⋯,an)\par=∑σ1≤π|V′σ1∈INC(χ|V′)μINC(σ1,π|V′)Φσ|V′(a1,⋯,al,κχ|V,1V(al+1,⋯,al+s)al+s+1,⋯,an)\par\par=κχ|V′,π|V′(a1,⋯,al,κχ|V,1V(al+1,⋯,al+s)al+s+1,⋯,an).

The other part follows from the bi-module property. This finishes the proof. ∎

The preceding theorem shows that is completely determined by cumulant functionals of the form for and .

###### Definition 4.3.

Let be a family of pairs of -faces of in a -valued probability space . We say that the family is combinatorially free-Boolean independent with amalgamation over if

 κχ,1n(a1,⋯,an)=0

whenever , , and is not a constant.

###### Proposition 4.4.

Let be a family of pairs of -faces in a -valued probability space . Then has the following additivity property:

 κχ,1n(a1,1+a2,1,⋯,a1,n+a2,n)=κχ,1n(a1,1,⋯,a1,n)+κχ,1n(a2,1,⋯,a2,n)

whenever , , , and .

###### Proof.

By a direct calculation, we have

 κχ,1n(a1,1+a2,1,⋯,a1,n+a2,n)=∑i1,...in∈{1,2}κχ,1n(ai1,1,⋯,ain,n).

Since are combinatorially free-Boolean independent, by Definition 4.3, we have

 κχ,1n(ai1,1,⋯,ain,n)=0

if for some . The result follows. ∎

###### Proposition 4.5.

Let be a combinatorially free-Boolean independent family of pairs of -faces in a -valued probability space . Assume that . Then

 κχ,π(a1,⋯,an)=0

whenever , , and is not a constant on a block of .

###### Proof.

We prove the statement by induction on the number of blocks of .

When , then the statement follows from Definition 4.3. Suppose now that , let be an interval block of . By Proposition 4.2, we have

 κχ,π(a1,⋯,an)=κχ|V′,π|V′(a1,⋯,al,κχ|V,1V(al+1,⋯,al+s)al+s+1,⋯,an),

where If is not a constant on , then . Otherwise, in not a constant on a block of The statement follows from an induction argument. ∎

### 4.2. Free-Boolean independence is equivalent to combinatorially free-Boolean independence

In this subsection, we will prove that free-Boolean independence defined in Definition 2.5 is equivalent to the combinatorially free-Boolean independence given in Definition 4.3. We will show that mixed moments are uniquely determined by lower order mixed moments in the same way for both free-Boolean independence and combinatorially free-Boolean independence.

The proof for following result is essentially the same as the proof of in [11, Proposition 10.6] in free probability context and we thus leave the details to the reader. Applying Theorem 4.2, we have the following result.

###### Lemma 4.6.

Let and be noncommutative random variables in a -valued probability space . Then

 E(a1⋯an)=∑π∈INC(χ)κχ,π(a1⋯an).

For combinatorially free-Boolean independent random variables, we have the following result.

###### Lemma 4.7.

Let be a family of combinatorially free-Boolean independent pairs of -faces in a -valued probability space . Assume that , where , . Let . Then,

 (★) E(a1⋯an)=∑σ∈INC(χ)⎛⎜ ⎜⎝∑π∈INC(χ)σ≤π≤ϵμINC(σ,π)⎞⎟ ⎟⎠Φσ(a1⋯an).
###### Proof.

By Lemma 4.6, we have

 E(a1⋯an)=∑π∈INC(χ)κχ,π(a1⋯an).

For each , write its blocks as . Since are combinatorially free-Boolean independent, by Lemma 4.5, we have

 κχ,π(a1⋯an)=0,

if is not a constant on some block of . In other words, only if is a constant on for all , which implies that is contained in a block of for all , i.e., . Therefore, we have

 E(a1⋯an)=∑π∈INC(χ),π≤ϵκχ,π(a1,⋯,an)=∑π∈INC(χ),π≤ϵ⎛⎜⎝∑σ∈INC(χ)σ≤πμINC(σ,π)Φσ(a1,⋯,an)⎞⎟⎠=∑σ∈INC(χ)⎛⎜⎝∑π∈INC(χ)σ≤π≤ϵμINC(σ,π)⎞⎟⎠Φσ(a1,⋯,an).

This finishes the proof. ∎

We now turn to consider the case that the family is free-Boolean independent in in the sense of Definition 2.5. In what follows, we assume that , where , . Let , the kernel of . Recall that . Let (or ) be the restriction of (or ) to respectively. Let (or ) be the restriction of (or ) to the interval