1. Introduction
Stochastic processes invariant under distributional symmetries have been intensively studied in classical probability theory, and their natural applications to statistical mechanics and other applied fields deeply encouraged this investigation. The reader is referred to [
1] for an exhaustive account on the matter. It was then natural to address the systematic investigation of the theory of stochastic processes to the quantum setting, which indeed started with the seminal paper [
2].
Families of random variables which are not necessarily commutative provide a general framework to realise de Finetti-type theorems, and therefore to classify stochastic processes whose finite joint distributions are independent of the action of some algebraic structures. Potential applications to quantum information theory and quantum statistical mechanics promoted a huge amount of results in this subject in recent years. For an account which is far to be complete, we refer the reader to [
3,
4,
5,
6,
7], and the references cited therein.
Among the most common distributional symmetries, we mention spreadability, exchangeability and stationarity. By definition, exchangeable stochastic processes are automatically spreadable. For commutative random variables, the converse also holds. The equivalence between spreadability and exchangeability is indeed the Ryll–Nardzewki theorem (cf. [
8]), nowadays celebrated as a part of the so-called extended de Finetti theorem. This statement is not generally true in the non commutative setting (e.g., [
9]), but there are prominent examples of quantum stochastic processes for which it still holds. One of them is the boolean case, as we are going to show in these notes.
Recently, in quantum probability it has been established a one-to-one correspondence between unitarily equivalent classes of noncommutative stochastic processes on the index-set
J, for the sample
-algebra
, and states on the free product
-algebra
. This entails that exchangeable or stationary (in the case
) stochastic processes are uniquely determined by symmetric or shift invariant states on
, respectively. The same holds on concrete
-algebras, seen as the quotient of the free product
-algebra, by means of the universal property of
. The reader is referred to [
4,
10] for more details about this conceptual point.
Using these results, in [
5] spreadability was investigated for stochastic processes arising from the so-called monotone commutation relations. There, it was shown that the monoid generated by the right and left hand-side partial shifts acts on the monotone
-algebra by unital *-endomorphisms, and spreadable stochastic processes, or equivalently spreading invariant states, were also classified.
After studying the algebraic structures involved in spreadability, here we develop the analysis of such a distributional symmetry for processes belonging to the concrete
-algebra arising from boolean commutation relations. One of the main interest in this research field is motivated by the physical application in quantum optics of such boolean stochastic processes, as pointed out in [
11].
We mention that the structure of the (concrete) boolean
-algebra was investigated in [
10,
12], whereas the ergodic properties of invariant states (see e.g., [
13]) for the monotone vs. boolean
-algebras, with their similarities and differences, are described in [
14],
Section 5.
The paper is organised as follows. After recalling in
Section 2 some features on
-dynamical systems and quantum stochastic processes, in
Section 3 we present the distributional symmetries managed in the notes, and some of the basic relations among them as well. Although the definition of a spreadable stochastic process is provided in terms of the invariance of the finite joint distributions under the natural action of the monoid of strictly increasing maps on
, here we show that spreadability can be directly stated in terms of invariance with respect to the action of the monoid, denoted by
, generated by left and right hand-side partial shifts on the integers.
In
Section 4, we introduce a further monoid involved in our investigation, that is the one generated by the strictly increasing maps on
whose range has finite complement. It is denoted by
, and we see that it is properly included in the monoid of strictly increasing maps, and strictly contains
. In addition, it provides another structure to study spreadability, as we show that spreading invariant states are exactly those invariant under the action of
. This statement appears interesting in our successive investigation about boolean stochastic processes, since
offers a more flexible analysis in that case. The intimate relation between
and
is pointed out in Proposition 4, which is the main result of the section. There, we prove that
is the semi-direct product of
by the action of
generated by the conjugation by the one-step shift. Such a result of combinatory nature, has a natural self-containing interest deserving of possible applications in other fields of mathematics where the monoids of increasing maps can play a role.
In
Section 5, it is seen that
acts by unital *-endomorphisms on the concrete boolean
-algebra. In particular, this result is preliminary achieved on the unital *-algebra of finite rank operators on the boolean Fock space. Then, it is extended to its closure in the uniform topology (i.e., the
-algebra of the compact linear operators), which indeed coincides with the (concrete) boolean
-algebra, and the action is completely described also in this case. This result is also aimed to yield the structure of boolean spreading invariant states, given in the last part of the notes. Indeed, we get the suitable version of the Ryll–Nardzewki theorem in our setting, finding that the states invariant under the aforementioned action are exactly the symmetric, or equivalently shift invariant positive normalised functionals, whose common structure was obtained in [
15] and [
10], respectively.
In the final section, we briefly summarise some open problems for further investigations.
2. Preliminaries
The present section is devoted to collect some features and properties useful in the forthcoming part of the notes. To shorten the notations, we indifferently denote by
the set of all natural numbers, with or without 0, if this causes no confusion. In addition, we put
In our setting, the triplet is said a -dynamical system if is a -algebra with unit , M is a monoid, and finally is a representation of M by completely positive identity preserving (i.e., unital) maps of .
In some cases,
M is replaced by a group
G, and in the
-dynamical system
is indeed a representation of
G into the group of the *-automorphisms
of
. In the latter case, one speaks of reversible dynamics, whereas dissipative dynamics appears in absence of bijections, see, e.g., [
16].
By we denote the convex of the states on , that is the positive normalised linear functionals on . is weakly *-compact as is unital.
Let
be invariant under the action of each element of
M, i.e.,
,
, and consider the Gelfand–Naimark–Segal (GNS for short) representation
. Then there exists a unique contraction
such that
and
see, e.g., [
17], Lemma 2.1. The quadruple
is called the covariant GNS representation associated to the invariant state
. If the
are multiplicative, the
are isometries. If in addition the
are invertible, then the
are unitaries.
The convex, compact in the *-weak topolgy, subset of all invariant states is
The set of the extremal invariant states (i.e., the extreme boundary) is denoted by
. Those are, by definition, nothing else than the ergodic states under the action
of
M.
Among the groups we deal with, we mention that consisting of all permutations
of an arbitrary index-set
J, leaving fixed all elements but finitely many. It is given by
where
is the symmetric group associated to the finite set
. If
J is the linearly ordered
, we also mention the group generated by the one-step shift
of the integers
, which is canonically identified with
itself.
Recall that, for a given arbitrary set
J and unital
-algebras
, their unital free product
-algebra
(cf. [
18]) is the unique unital
-algebra, together with unital monomorphisms
such that for any unital
-algebra
and unital morphisms
, there exists a unique unital homomorphism
making commutative the following diagram
Here, we consider unital free product
-algebras based on a single unital
-algebra
, the algebra of the samples, called the free product
-algebra, and denoted simply as
. We refer the reader to [
4,
10,
19] for further details.
The central aspect in the theory of stochastic processes is to construct a process based on the sample algebra and the index-set J, starting from the knowledge of the collection of its finite dimensional distributions. In the abelian case the suitable conditions are summarised in the Kolmogorov Reconstruction Theorem, whereas the quantum generalisation is provided by the GNS construction.
To be more precise, fix
,
with contiguous different indices, and
. The finite joint distributions are the values
which arise from multilinear functionals
on
. They satisfy some natural positivity and consistency conditions given by
The classical case is characterised by
,
X being a (locally) compact space, and
for
,
, and
.
The above properties indeed reduce to the Kolmogorov requests, and Equation (
1) gives that a classical stochastic process is uniquely determined, up to equivalence, by the finite joint distributions
such that the sets of indices
are all different, and independent of their order.
Thus, the Kolmogorov theorem allows to construct a probability measure on the Tikhonoff product of J copies of X. In the quantum setting, the aforementioned properties permit to perform the GNS representation (defined up to unitary equivalence), and so give rise to general stochastic processes as defined in the forthcoming lines.
By taking into account the previous considerations, we can assume as starting point (i.e., by definition) that the process under consideration is directly realised on a Hilbert space.
A (realisation of a) possibly quantum, stochastic process, labelled by the index-set J and determined up to unitary equivalence is, in our language, a quadruple , where is a -algebra, is an Hilbert space, the maps are *-homomorphisms of in , and is a unit vector, cyclic for the von Neumann algebra naturally acting on .
In [
10], Theorem 3.4, it was proved that states on
uniquely correspond to quantum stochastic processes. More in detail, one sees that the quadruple
determines a unique state
, and a representation
of
on the Hilbert space
, such that
is the GNS representation of the state
.
Conversely, each state defines a unique stochastic process, just by looking at its GNS representation uniquely determined up to unitary equivalence.
For more details and proofs, the interested reader is referred to the above mentioned paper, and [
4] as well.
3. Stochastic Processes and Their Symmetries
In the present section, we investigate some natural invariance properties for the stochastic processes. Among those, we will deal with the so-called stationarity, spreadability and exchangeability. To simplify the matter, we suppose that in order to compare the above three mentioned symmetries.
We consider the set of all maps . It provides a monoid , where the product is the composition “∘” between maps and the unit is the identity-map of .
The following two sub-monoids of
are of interest for our analysis. The first one
is given by all the strictly increasing maps of
, or equivalently maps which determine all subsequences of
. Obviously, if
then their composition
, and therefore
is endowed with a structure making it a monoid
.
The second one is the monoid generated by all partial shifts on
. Namely, the
h-right hand-side partial shift,
, is the one-to-one map
such that
Analogously, the
h-left hand-side partial shift,
, is the one-to-one map
such that
We note that
.
Let us denote by the sub-monoid of generated by all forward and backward partial shifts , and . We will see later that .
From now on, we drop the composition symbol simply by writing that as a product: . Thus, we indicate with the n-fold composition of f with itself. If is invertible, we put and therefore is defined for all when it is meaningful.
We often also write the relative monoids without pointing out the composition and the unit. For example, and will be denoted simply as and , respectively.
Remark 1. The powers of the one-step shift τ and its inverse act in a natural way on by conjugacy: for ,Therefore, for any . We report, without the proof, the following useful results (cf. [
10], Lemma 2.2, and [
5], Proposition 2.1) in order to manage the symmetries of stochastic processes which will be introduced below.
Proposition 1. The following holds true for a finite interval .
- (i)
There exists a cycle such that .
- (ii)
For each there exists such that .
By universality, the groups
and
act in a natural way as *-automorphisms on the free product
-algebra
by shifting and permuting the indices of the generators, respectively. Moreover, it is possible to see (cf. [
5], Section 4) that there is an action by *-endomorphisms of both the monoids
and
on
.
Denoting by , , , and the corresponding dynamical systems, we have the following immediate consequences of Proposition 1.
Corollary 1. The following assertions hold true:
- (i)
.
- (ii)
.
Proof. Taking into account the proof of Proposition 2.1 in [
4], (i) and the relation
follow directly from (i) and (ii) of Proposition 1, respectively. Since
, and therefore
, (ii) holds true as well. □
Definition 1. For the -algebra , , , , the stochastic process is said to be
- -
- -
exchangeable if for each , - -
spreadable if for each ,
By the above mentioned equivalence between stochastic processes for the sample algebra on the index-set , and states on , Corollary 1 leads to:
- -
If a process is exchangeable, then it is stationary;
- -
A process is spreadable if and only if it is invariant under the natural action of the monoid .
Therefore, we can indifferently define spreadability as the invariance under the action of , or equivalently the invariance under its sub-monoid . We will see later that, for particular models, spreadability can be conveniently investigated by a suitable monoid included between and .
4. Monoids of Increasing Maps
In order to manage spreadability, it appears useful to define and study the structure of further sub-monoids of .
Let us denote by
and
the sub-monoids of
generated by all forward and backward partial shifts
and
, respectively. In addition, let us take
is also a sub-monoid of
. As usual, we denote such monoids simply by
,
and
, respectively. We note that all such monoids are sub-monoids of
.
Remark 2. We have .
Proof. For , there exists such that . On the other hand, if then there exists such that for each . This concludes the proof. □
For
, we provide the following notation
We have:
- -
,
- -
for some .
Remark 3. Let us take . Then one has Proof. Fix
. Since
f is one-to-one, it follows
This gives
□
Here, we investigate the relations among the above introduced structures. To our goal, we start with the following technical
Lemma 1. Let . Namely , where , or for some , , and . Then .
Proof. We start by noticing that, if , one finds . Now we proceed by induction.
Indeed, suppose that for
, where
,
and
, one has
. After taking
,
, we prove
. We reduce the matter to
, the other case being similar. Here, since
for some
, one finds for
and
,
whereas
gives
and finally
when
. □
Remark 4. Corollary 1 and Lemma 1 lead to
- (iii)
.
Therefore, spreadability for a stochastic process can be equivalently defined in terms of invariance w.r.t. the action of the monoid . This turns out to be more convenient in the case of (concrete) boolean processes, as we will see later.
Lemma 1 allows to prove that some of the monoids introduced above are strictly ordered with respect to set inclusion.
Proposition 2. Under the notations above, one finds Proof. Lemma 1 gives
and, by definition,
. Moreover, the strictly increasing map
does not belong to
, as
. In addition, the one-step shift
clearly belongs to
. However, since
,
by Lemma 1. □
The structure of is strictly related to that of . For this purpose, we recall the notion of semi-direct product of two monoids M and N, generalising the analogous notion for groups.
Indeed, fix a monoid
M acting by morphisms
on a second monoid
N. The semi-direct product
is defined as follows. As a set,
, whereas a binary operation is given by
It is easy to check that
, equipped with the multiplicative law in Equation (
4) defines a monoid whose unit is
.
Proposition 3. The group acts on the monoids , , , and , through the powers of one-step shift τ and its inverse .
Proof. Fix
. It is immediate to see that the map
defined in Equation (
2) realises an automorphism when restricted to
, that is
acts on
. Since
,
, one has that
acts by restriction on
. By [
5], Section 2.3,
acts by restriction, separately on
and
, and therefore also on
which is generated by the latter monoids. □
Here, there is the main result of the present section.
Proposition 4. Under the above notations, one has Proof. We first note that, for
,
If
, we show that
f is uniquely decomposed as
, for
and
.
In order to prove the claim, we preliminary observe that for any
there exist uniquely determined
(depending on
f) such that
If
, it results
. Hence,
for some integers
, and
. Note that
, as for the strictly increasing
g one finds
.
For each , one can have either or . Thus, one defines , and denotes where, as usual, if .
In the case
, the aforementioned decomposition is achieved, since one finds
the last equality following from Equation (
5).
If instead
A is non-void, we write
for some
, and obtain
where
As a concrete example, take
uniquely determined by
and
. Then
as
, and consequently
,
,
,
, and finally
.
After defining
, one achieves the desired decomposition in this case too, by getting
where the last equality comes again from Equation (
5).
Suppose now that
Since there exist uniquely determined
such that
,
, one firstly gets
. Uniqueness of the above decomposition then follows if
. To this goal, we reduce the matter to
, the other cases being trivial. Here, since
one finds the following “normal order” for
,
:
where
,
,
, and finally
,
, for
,
. This immediately yields
.
Summarising, the map
realises an isomorphism between monoids. Therefore, for
, the unique expression
with
and
, provides the description of
as inner semi-direct product between elements of
and powers of
and
.
Concerning the sub-monoid
, we get
and thus we have the equality of monoids
The equality
follows analogously. □
Once having established (cf. Remark 4) that spreadability of stochastic processes on the index-set can be indifferently investigated by the monoids , and , in the next section we will see that the intermediate monoid provides a more flexible analysis in the case of the (concrete) boolean -algebra.
5. Spreading Invariant States on the Boolean Algebra
As an application of the previous results, the present section is devoted to the investigation of the spreadability for stochastic processes arising from the so-called boolean commutation relations (
7). For such a purpose, the main step will be to show that the monoid
acts on the concrete boolean unital
-algebra by unital *-endomorphisms.
Let be a complex Hilbert space with inner product , linear in the first argument. The boolean Fock space over is the direct sum , and the vacuum vector is . The vacuum vector state is denoted by .
For
and
, the creation and annihilation operators are defined as follows:
They are mutually adjoint and bounded.
The concrete boolean
-algebra
is that generated by all creators and the identity
. Since the *-algebra generated by the
consists of all finite rank operators, we easily get
where
denotes the
-algebra of compact linear operators on
. Here, we deal with the case
, where the canonical basis is
. Therefore
With the notations
,
, we can see that the following boolean commutation relation (in the spirit of [
20], pag. 109)
holds true with
,
being the usual Kronecker symbol. The above infinite sum is meant (i.e., converges) in the strong operator topology of
, as it was seen in [
15], Proposition 3.2.
For the convenience of the reader, we report the following result proved in [
10], Proposition 7.1: the unital
-algebra generated by the position operators
coincides with
and therefore, differently from the analogous one generated by the position operators of the free commutation relations, it acts irreducibly on
.
From now on, we use the shorthand notation . For each subset , will denote the self-adjoint projection onto the closed subspace of generated by the , .
Thus,
is the
-algebra of compact linear operators acting on
, and for the canonical system of matrix–units
in
, one has
It is well known that the following groups naturally act on
by *-automorphisms (e.g., [
4,
10,
15]):
- -
the integers by all powers of the one-step shift and its inverse;
- -
the group of all permutations moving only finitely many elements of .
Such actions are directly implemented by the (2
nd quantised action of the) corresponding actions on the canonical basis of
, that is by Bogolyubov automorphisms (cf. [
21]).
In what follows, we show that also
acts by unital *-endomorphisms on
. Such an action determines the structure of positive normalised functionals which are invariant, that is the spreadable stochastic processes arising from the boolean commutation relations. The reader is referred to [
5] for the similar situation involving monotone (and anti-monotone) commutation relations.
Let
be the *-algebra of finite rank operators on the boolean Fock space. On the unital *-algebra
, dense in the norm topology in
, as for the above mentioned actions of
and
, we can define
such that
where
The
are well defined because
is a Hamel basis. In accordance to the action of
on the monotone
-algebra (cf. [
5]), we can prove that
extends to an action of
by unital *-endomorphisms of the boolean
-algebra
by providing an explicit formula (see Theorem 1) for such an action.
On the canonical basis
of
, for any
we define
which extends to an isometry on
, denoted again by
. Indeed,
Lemma 2. For any , one has Proof. Since the
are injective, the
are isometries. In addition, as for
and
we get
and therefore the second identity in Equation (
9) easily follows.
Since for
and
, as
, one finds
the last identity in Equation (
9) is achieved. □
By Equation (
6), any
is decomposed as
, where
and
. Thus, the state at infinity
is well defined as
Lemma 3. For and , we get:
- (i)
,
- (ii)
.
Proof. We start by noticing that is a two-sided ideal, and implies is finite rank.
Fix
. By Equation (
9), we get:
- (i)
.
- (ii)
. □
Our next goal consists in showing it is possible to extend
in Equation (
8) as an action of *-endomorphisms on the whole
-algebra
.
For
, define the linear maps
Theorem 1. The mapgiven in Equation (11) provides a representation of the monoid in extending the linear maps given in Equation (8). Proof. We start by noticing that the maps in Equation (
11) preserve the *-operation, and are unital and bounded.
In addition, for
and
, Equations (
10) and (
9) give
For
,
, we now check
and therefore the
are unital *-endomorphisms of
.
Now we check that
provides an action of
on
. To this aim, fix
. Exploiting Lemma 3 (ii), Equations (
9) and (
3), one obtains
Finally, consider
, and a generic matrix unit
, for
. For any
and
Therefore, by linearity
. □
The following result establishes an equivalence between stationary, spreadable and exchangeable processes on the concrete boolean
-algebra, thus realising a version of Ryll–Nardzewski Theorem [
8] in our setting.
Proposition 5. The states on which are spreading invariant coincide with the stationary and symmetric states: Proof. By Remark 7.4 in [
15], one gets
Moreover, Remark 4 and Proposition 4 give
The thesis then follows after showing that
Indeed, recall that
is invariant for
,
. Hence, for any
with
and
, Equation (
11), Equations (
10) and (
9) entail
Since for
,
is finite rank, Lemma 3 yields
□
6. Conclusions
The investigation of the so-called quantum probability started with the seminal paper [
22]. After that, several applications to many fields of mathematics and physics have been established. We mention natural applications to models of quantum statistical mechanics and quantum information, and refer the reader to [
3,
23] and the literature cited therein for more details, even though the list is very far to be complete, compared with hundreds of interesting papers on the topics. On the other hand, a self-containing treatment of quantum probability, similar to the classical one, is nowhere close to being satisfactory.
An attempt towards a unified version of the probability scheme, including as particular cases both the various models arising from noncommutative realm, and the classical one too, was carried out in [
4,
10]. There, the concept of quantum stochastic process on a discrete index-set was a main topic of the investigation, and some natural distributional symmetries like stationarity and exchangeability were analysed. We also point out the relevance of quantum stochastic processes on continuous index-sets which was firstly outlined in [
24]. Therefore, its development towards a systematic theory could be a very interesting direction for future research.
In order to present some open questions closely related with the present notes, we first recall that, in commutative probability, the extended de Finetti theorem states that sequences of random variables which are either spreadable, or exchangeable, or finally conditionally independent and identically distributed w.r.t. the tail algebra, coincide. In [
12], Theorem 1, it was proved that exchangeable boolean stochastic processes are indeed those conditionally independent and identically distributed w.r.t. the tail algebra (known for physical applications as the algebra at infinity, see, e.g., [
16]). Consequently, Proposition 5 allows us to achieve the boolean version of the aforementioned theorem. Since similar results were obtained for monotone and
q-deformed stochastic processes in [
5,
10], open problems for future investigation could be:
- -
studying all natural symmetries like stationarity, exchangeability and spreadability, in the case of stochastic processes associated to anomalous commutation relations arising from quantum physics, such as the Fermi case [
25], or the more general case of Yang–Baxter–Hecke quantisation [
26];
- -
investigating another prominent example of distributional symmetry, that is the rotatability, for general families of noncommutative random variables, see, e.g., [
9,
27] for commutative and free cases, respectively.
Coming back to more general problems in quantum probability, one finds direct connections with physics, e.g., in Bose, Fermi, or Boltzamnn particle models. On the other hand, up to our knowledge, there are not yet direct physical applications for exotic commutation relations, such as general
q-deformations and the other ones satisfying Equation (
7) and described in [
20]. Among those, we however mention the commutation rules corresponding to the anyonic statistics investigated from a mathematical viewpoint in [
28], which could provide promising applications to the physical models described in [
29].
Concerning the boolean case treated in the present paper, up to the best knowledge of the authors, the only physical motivations is described in [
11] as already pointed out above. There, the first direct relation between the boolean commutation relations and the boolean independence (cf. [
30]) has been also established.
Boolean independence might also be related to quantum measurement processes as formulated in [
31], and successively in [
32] (we acknowledge an anonymous referee for bringing to our attention this possible connection and the interesting reference [
32]). Therefore, further open problems for possible future investigation could be:
- -
providing physical applications of random variables exhibiting exotic commutation rules such as the
q-relations,
(e.g., [
33,
34]), and the monotone and anti-monotone ones (e.g., [
5,
15,
30]);
- -
investigating potential connections between boolean independence and quantum measurement processes, in particular for the models fitted to quantum physics described in [
32].