Abstract
We study joint numerical and spectral radii defined for d-tuples of bounded operators on a Hilbert space and related to noncommutative notions of independence. The definitions are in analogy with the ones of Popescu, where his formulations turned out to be related with free creation operators, and in this way related to the free independence of Voiculescu. In our study the definitions are related with either weakly monotone creation operators, and thus associated with the monotone independence of Muraki, or with boolean creation operators, and hence related with the boolean independence.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The notion of numerical radius as well as the related notion of numerical range is an object of intensive studies since the work by Toeplitz [10] in 1918 until today. Numerical radius provides a norm on the space of bounded operators, which is equivalent to the operator norm. Its special features include unitarity invariance, the power inequality and the relation with the spectral radius, see e.g. [6]. The numerical range and radius link the properties of operators with geometry of a complex plane, allowing many interesting applications, from the approximate localization of spectrum via the stability results for differential equations (e.g. [5]) to the von Neumann type inequalities (e.g. [1]).
In 2009, Gelu Popescu [9], in his search for a free analogue of the Sz.-Nagy-Foias theory for row contractions, defined free analogues of numerical and spectral radii, namely the joint numerical radius and joint spectral radius for d-tuples of operators \((T_1, \ldots , T_d)\) acting on a Hilbert space \({\mathcal {H}}\). The definitions are as follows.
Definition 1
The (free) joint numerical radius is
where \({\mathbb {F}}_d^+\) is the free semigroup with free generators \(g_1, \ldots , g_d\) and each \(\alpha \in {\mathbb {F}}_d^+\) is a word in these generators.
Definition 2
The (free) joint spectral radius is
where for \(\alpha \in {\mathbb {F}}_d^+\) one puts \(|\alpha |=k\) if it is a word in k generators \(\alpha = g_{i_1}\dots g_{i_k}\).
These two notions are related not only to the free semigroup \({\mathbb {F}}_d^+\), but also to the model of freeness of Voiculescu [11], and more precisely to the creation operators on the full Fock space by the following result.
Theorem 3
([9], Corollary 1.2) The joint free numerical and spectral radii can be computed as the ordinary numerical radiuswand the spectral radiusrof single operators:
where\(S_1:=S(e_1), \ldots , S_d:=S(e_d)\)are the free creation operators on the full Fock space\({\mathcal {F}}(H_d)\)ond-dimensional Hilbert space\(H_d\)with an orthonormal basis\(\{e_1, \ldots , e_d\}\)andS(h) is the creation operator by the vector\(h\in H_d\): \(S(h)v=h\otimes v\)for\(v\in {\mathcal {F}}(H_d)\).
The main idea of this paper is to study analogues of these definitions in the case where we replace the (free) full Fock space by a Fock space associated to other noncommutative independences, and the free creation operators by the creation operators on the appropriate Fock space. We show that the joint (noncommutative) numerical radii, defined in analogy to Theorem 1, satisfy many basic properties similar to the free case. In particular, we show the unitarity invariance and the relation with the appropriately defined spectral radius. We also compute some examples. In the paper we treat the monotone and boolean case, but the framework is more general. Our idea establishes yet another bridge between classical operator theory and the noncommutative probability and we believe this is a starting point for further investigations. However, in this paper there is no need to define the monotone and boolean independences, it is sufficient to consider the models of both of them, built on either weakly monotone Fock space or on the Boolean Fock space, respectively.
2 General scheme
Recall that the (classical) numerical radius of a linear operator T, bounded on a Hilbert space \({\mathcal {H}}\), is defined by
and the (classical) spectral radius of T is defined by
Let us consider one of the noncommutative independence, e.g. boolean, free, monotone and let us consider a noncommutative Fock space \(F_{\star }(H)\) associated to this independence. By this we mean the full Fock space ( [11]) for the free independence of Voiculescu, the weakly monotone Fock space ( [12]) for the monotone independence of Muraki and the boolean Fock space ( [3]) for the boolean independence (c.f. [2]). On each of these Fock spaces we are given creation and annihilation operators, which we shall use to define relative joint (numerical and spectral) radii, following the work by Popescu [9] for the free case.
Let \(d\in {\mathbb {N}}\) and let H be the d-dimensional Hilbert space with the orthonormal basis \(\{e_j; 1\le j\le d\}\). Denote by \(A_1^{\star },\ldots , A_d^{\star }\) the creation operators associated to the basic vectors \(A_j^{\star }=A^{\star }(e_j)\) on the related Fock space \(F_\star (H)\) and define the \(\star\)-joint numerical radius of the d-tuple \((T_1,\ldots ,T_d)\) of operators in \(B({\mathcal {H}})\) by
Similarly, we define the \({\star }\)-joint spectral radius of \((T_1,\ldots ,T_d)\) by
The following properties of the \(\star\)-joint numerical and spectral radii are immediate consequences of the construction and of the properties of the classical numerical radius (compare with Theorem 1.1 in [9]).
Proposition 4
The\(\star\)-joint numerical radius and joint spectral radius satisfy:
- (i):
-
\({ w}_\star (\lambda T_1,\ldots ,\lambda T_d) = |\lambda | { w}_\star (T_1,\ldots ,T_d)\)for any\(\lambda \in {\mathbb {C}}\);
- (ii):
-
\({ w}_\star (T_1+T_1',\ldots ,T_d+T_d')\le { w}_\star (T_1,\ldots ,T_d)+ { w}_\star (T_1',\ldots ,T_d')\);
- (iii):
-
\({ w}_\star (U^*T_1U,\ldots ,U^*T_dU)={ w}_\star (T_1,\ldots ,T_d)\)for any unitary\(U:{\mathcal {K}} \rightarrow {\mathcal {H}}\);
- (iv):
-
\({ w}_\star (I_{\mathcal {K}}\otimes T_1,\ldots ,I_{\mathcal {K}}\otimes T_d)={ w}_\star (T_1,\ldots ,T_d)\)for any separable Hilbert space\({\mathcal {K}}\);
- (v):
-
\(r_\star (T_1,\ldots ,T_d) \le { w}_\star (T_1,\ldots ,T_d)\).
Proof
The properties (i) and (ii) follows from \({ w}_{\mathrm{cl}}(\lambda T) = |\lambda | { w}_{\mathrm{cl}}(T)\) and \({ w}_{\mathrm{cl}}(T+T') \le { w}_{\mathrm{cl}}(T)+{ w}_{\mathrm{cl}}(T')\). As for (iii), for a unitary \(U:{\mathcal {K}} \rightarrow {\mathcal {H}}\) we set \(V=I\otimes U\) and \(A:=A_1^{\star }\otimes T_1^*+\ldots A_d^{\star }\otimes T_d^*\) and observe that
The property (iv) goes exactly as in the proof of [9, Theorem 1.1], using \({ w}_{\mathrm{cl}} (I_{\mathcal {K}}\otimes T)={ w}_{\mathrm{cl}} (T)\):
Finally, to show (v) we just use the classical result \(r_{\mathrm{cl}}(T)\le { w}_{\mathrm{cl}} (T)\). \(\square\)
3 Joint boolean numerical and spectral radii
Recall after [3] that the boolean Fock space (over the d-dimensional space H) is defined as the direct sum
where \(\Omega\) is a unit vector, called the vacuum. The boolean creation and annihilation operators are given by
For a fixed orthonormal basis \(\{e_j : 1\le j\le d\}\) in H, we shall use the notation \(B_j^* := B^{*}(e_j)\), \(B_j:=B(e_j)\), \(j=1,\ldots ,d\), for the creation and annihilation operators (respectively) by the basic vectors, and \(e_0:=\Omega\). It is easy to see that \(B_j B^*_k = \delta _{jk} P_\Omega\), where \(P_\Omega\) is the projection onto the vacuum vector \(e_0=\Omega\).
Let now \((T_1,\ldots ,T_d)\) be the d-tuple of bounded operators on a Hilbert space \({\mathcal {H}}\). We define the joint boolean numerical radius of \((T_1,\ldots ,T_d)\) as
and the spectral joint boolean spectral radius of \((T_1,\ldots ,T_d)\) as
We first provide the explicit formula for computing the joint boolean numerical radius, which is the analogue of Popescu’s definition in the free case (see Definition 1).
Proposition 5
The joint boolean numerical radius can be expressed as
Proof
By the definition of the classical numerical radius, we have
Expressing \(h\in {\mathcal {F}}_B(H) \otimes {\mathcal {H}}\) as \(h=\sum _{k=0}^d e_k \otimes g_k\) with \(g_0,g_1,\ldots ,g_d\in {\mathcal {H}}\) (recall that \(e_0:=\Omega\)), we observe that
whereas, using the relation \(\displaystyle B_je_m=\delta _{jm}e_0\), for \(1 \le j, m\le d\), and \(B_je_0=0\) for \(1\le j\le d\), we get
\(\square\)
It turns out that all of the properties that were shown to hold for the (free) joint numerical radius (see [9, Theorem 1.1]), remains true in the boolean case. Some of them were already observed in Proposition 4; here we prove the remaining ones.
Proposition 6
We have
- \((i)_B\) :
-
\({ w}_B(T_1,\ldots ,T_d)=0\)if and only if\(T_1=\ldots =T_d =0\),
- \((ii)_B\) :
-
\(\frac{1}{2} \Vert \sum _{j=1}^d T_j T_j^*\Vert ^{\frac{1}{2}} \le { w}_B(T_1,\ldots ,T_d)\le \Vert \sum _{j=1}^d T_j T_j^*\Vert ^{\frac{1}{2}}\),
- \((iii)_B\) :
-
\({ w}_B(X^*T_1X,\ldots ,X^*T_dX)\le \Vert X\Vert ^2 { w}_B(T_1,\ldots ,T_d)\)for any bounded operator\(X:{\mathcal {H}}\rightarrow {\mathcal {K}}\).
Proof
Ad \((i)_B\). By the properties of the classical numerical radius, \({ w}_B(T_1,\ldots ,T_d)=0\) implies \(\sum _{j=1}^d B_j^* \otimes T_j^*=0\). Hence for any \(h=\sum _{k=0}^d e_k \otimes g_k \in {\mathcal {F}}_B(H) \otimes {\mathcal {H}}\), one gets
Thus, for any \(j=1,\ldots ,d\) and \(g_0\in {\mathcal {H}}\), one gets \(T_j^*g_0=0\). Consequently, \(T_j^*=0\) and, equivalently, \(T_j=0\) for any \(j=1,\ldots ,d\).
Ad \((ii)_B\). Thanks to \(\frac{1}{2}\Vert T\Vert \le { w}_{cl}(T)\le \Vert T\Vert\), we have
Similarly, we prove that \({ w}_B(T_1,\ldots ,T_d) \ge \frac{1}{2} \left\| \sum _{j=1}^d T_j^*T_j\right\| ^\frac{1}{2}\).
Ad \((iii)_B\). Let \(X:{\mathcal {H}}\rightarrow {\mathcal {K}}\) be a bounded operator and let \(g_0,\ldots ,g_d\in {\mathcal {H}}\) satisfy \(\sum _{k=0}^d \Vert g_k\Vert ^2=1\). Define \(C:=\sqrt{\sum _{k=0}^d \Vert Xg_k\Vert ^2}\) and \(h_k:=\frac{1}{C}Xg_k\), \(k=0,1,\ldots ,d\). Then \(\sum _{k=0}^d \Vert h_k\Vert ^2=1\) and \(C\le \Vert X\Vert\). Consequently,
\(\square\)
Remark 7
The properties (i), (ii) and \((i)_B\) show that the joint boolean numerical radius is a norm on \(B({\mathcal {H}})^{d}\), which, by \((ii)_B\), is actually equivalent to the operator norm \(\displaystyle \left\| \sum _{j=1}^d T_j^*T_j\right\| ^\frac{1}{2}\) of the operator row matrix \([T_1,\ldots ,T_d]\), hence \({ w}_B\) is a continuous map in the norm topology.
We now compute some examples and, in particular, show that \({ w}_B\ne { w}_F\).
Example 8
For \(T_j=I_{\mathcal {H}}\), \(j=1,\ldots ,d\) we have
Using the relation between the \(\ell _1\)- and \(\ell _2\)-norm on \({\mathcal {H}}^d\):
and Proposition 5, we observe that
Hence for any \(g_0,\ldots ,g_d\in {\mathcal {H}}\) such that \(\sum _{k=0}^d \Vert g_k\Vert ^2=1\), denoting \(t:=\Vert g_0\Vert \in [0,1]\), we get
and the supremum is achieved when \(t=\frac{1}{\sqrt{2}}\). This show that \({ w}_B(I_1,\ldots ,I_d) \le \frac{\sqrt{d}}{2}\).
To see that the equality holds, take \(g_0\in {\mathcal {H}}\) with \(\Vert g_0\Vert ^2=\frac{1}{2}\) and \(g_k=\frac{g_0}{\sqrt{d}}\), \(k=1,\ldots ,d\). Then \(\sum _{k=1}^d \Vert g_k\Vert ^2=1\) whereas \(|\langle g_0, \sum _{k=1}^d g_k\rangle |= |\langle g_0,\frac{d}{\sqrt{d}} g_0\rangle |=\frac{\sqrt{d}}{2}\).
Example 9
For \(T_j=B_j\), \(j=1,\ldots ,d\), the boolean annihilators, we have
Since \(B_jh=\langle h,e_j\rangle \Omega\) for \(h\in {\mathcal {H}}\), we compute
as shown in the previous Example.
Taking \(g_0=\frac{1}{\sqrt{2}}\Omega\) and \(g_k=\frac{1}{\sqrt{2d}}e_j\) for \(j=1,\ldots ,d\), we find that \(\sum _{k=1}^d \Vert g_k\Vert ^2=1\) while
Example 10
For \(d=1\) we find out that
and \({ w}_B(T)=\frac{1}{2} { w}_F(T)\) when \({\mathcal {H}}={\mathbb {C}}\), \(Tz:=az\), \(a\in {\mathbb {C}}^*\). This shows that \({ w}_F(T_1,\ldots ,T_d) \ne { w}_B(T_1,\ldots ,T_d)\) in general.
It was shown in [9, Sect. 1] that \({ w}_F(T) = { w}_{\mathrm{cl}}(T)\). Now, observe that
On the other hand, given \(g_0,g_1\in {\mathcal {H}}\), with \(\Vert g_0\Vert ^2+\Vert g_1\Vert ^2 =1\), we can repeat the idea of Popescu, setting \(f_\theta :=g_0+e^{\mathrm{i}\theta }g_1\) for \(\theta \in [0,2\pi ]\). Then
and
So, using the fact that \(|\langle h, Th \rangle | \le { w}_{cl}(T) \Vert h\Vert ^2\), we get
For the special choice \({\mathcal {H}}={\mathbb {C}}\) and \(T_a z:=az\) for some fixed \(a\in {\mathbb {C}}^*\), we get \({ w}_B(T_a)=\frac{1}{2} { w}_{\mathrm{cl}}(T_a)\). Indeed,
It is an open problem to check if the equality \({ w}_B(T)={ w}_F(T)\) can hold.
We end this Section with an observation that the joint boolean spectral radius degenerates.
Proposition 11
The joint boolean spectral radius is always 0.
Proof
Since the boolean creation operators satisfy \(B_j^*B_k^* =0\) for any j, k, we have
\(\square\)
4 Joint monotone numerical and spectral radii
We consider the model of the monotone independence on the discrete weakly monotone Fock space \({\mathcal {F}}_{WM}(H)\), defined in [13]. This space is built upon a d-dimensional Hilbert space H with a given orthonormal basis \(\{e_j:\ 1\le j\le d\}\), as a closed subspace of the full Fock space \({\mathcal {F}}(H)\), spanned by the vacuum vector \(\Omega :=e_0\) and the simple tensors \(\displaystyle e_{j_k}\otimes \dots \otimes e_{j_1}\), where the indices are in weakly monotonic order: \(1\le j_1\le \dots \le j_k\). By the standard convention we identify \(e_0\otimes h = h\otimes e_0 = h\) for any \(h\in {\mathcal {F}}_{WM}(H)\).
The creation operator \(M_j^{*}\) by the vector \(e_j\) is defined as follows:
The annihilation operator is defined as
and they are mutually adjoint: \(\displaystyle (M_k)^{\star }=M_k^*\).
It is useful to introduce the following orthogonal projections. By definition \(P_0\) is the orthogonal projection onto \(e_0=\Omega\) and for \(m\ge 1\):
Then \(Q_m\) is the orthogonal projection onto the span of \(e_0=\Omega\) and vectors of the form \(e_{j_k}\otimes \dots \otimes e_{j_1}\) with \(j_k\le m\), and \(P_m\) is the orthogonal projection onto the span of vectors of the form \(e_{j_k}\otimes \dots \otimes e_{j_1}\) with \(j_k = m\), \(k\ge 1\).
The weakly monotone creation and annihilation operators satisfy the following commutation relations
In particular, for \(j<k\) we have
Remark 12
The monotone creation operators are bounded and generate \(*\)-subalgebras which are monotone independent in the sense of Muraki [7, 8] (for the proof see [13] and [4]).
In this setting the joint monotone numerical radius is defined as
and the joint monotone spectral radius of \((T_1,\ldots ,T_d)\) is defined as
To provide explicit formulas for the joint monotone numerical and spectral radii we introduce some operations on weakly monotone sequences. For each \(k\in {\mathbb {N}}_0:={\mathbb {N}}\cup \{0\}\) we define
Note that each \({\mathcal {M}}_k\) with \(k\ge 1\) is finite with cardinality \(\left( {\begin{array}{c}k+d-1\\ k\end{array}}\right)\). We also set \({\mathcal {M}}:=\bigcup _{k\ge 0} {\mathcal {M}}_{k}\) to be the union of all \({\mathcal {M}}_k\) for \(k\ge 0\) and \({\widetilde{{\mathcal {M}}}}:= {\mathcal {M}}\cup {\mathcal {M}}_{-1}\).
There is a natural comparison relation R of the sequences from \({\mathcal {M}}\), namely for \(\alpha =(i_k, \ldots , i_1)\in {\mathcal {M}}_k\) and \(\beta :=(j_m, \ldots , j_1)\in {\mathcal {M}}_m\) with \(k, m\ge 1\) we put
Observe that this relation is neither symmetric nor antisymmetric.
For convenience we shall write \(\alpha \ge \beta\) iff \((\alpha , \beta )\in R\).
Definition 13
The weakly monotone concatenation of such sequences is defined on \({\widetilde{{\mathcal {M}}}}\) as follows: Let \(\alpha =(i_k, \ldots , i_1)\in {\mathcal {M}}_k\) and \(\beta :=(j_m, \ldots , j_1)\in {\mathcal {M}}_m\) with \(k, m\ge 1\), then
If \(\alpha \in {\mathcal {M}}_k\), \(\beta \in {\mathcal {M}}_m\) with \(k,m\ge 0\) and \(\alpha \ge \beta\), then \(\alpha \beta \in {\mathcal {M}}_{k+m}\).
In particular, for \(j\in {\mathcal {M}}_1\) and \(\alpha =(i_k, \ldots , i_1)\in {\mathcal {M}}_k\) we shall have \(\displaystyle j \alpha = (j, i_k, \ldots , i_1)\) if \(j\ge \alpha\). As one can see for this concatenation 0 plays the role of neutral element and the empty set \(\emptyset\) behaves like 0 in multiplication of numbers.
Then, with the notation
the set \(\{e_\alpha :\alpha \in {\mathcal {M}}\}\) is an orthonormal basis of \({\mathcal {F}}_{WM}(H)\). We shall extend this notation to the operators: if \(T_1,\ldots , T_d\in B({\mathcal {H}})\) are given and \(\alpha \in {\mathcal {M}}\), then
Now we are ready to describe the explicit formula for the joint monotone numerical radius of d-tuple of operators.
Proposition 14
Let\((T_1, \ldots , T_d)\)be ad-tuple of bounded operators on a Hilbert space\({\mathcal {H}}\). Their joint monotone numerical radius can be computed as
Proof
We express \(h\in {\mathcal {F}}_M(H) \otimes {\mathcal {H}}\) as \(h=\sum _{\alpha \in {\mathcal {M}}} e_\alpha \otimes g_\alpha\) with \(g_\alpha \in {\mathcal {H}}\). Then
Since
and
the formula (11) follows. \(\square\)
Note that in the summation over \(\alpha \in {\mathcal {M}}\) the nonzero terms might be only for those \(\alpha\) for which \(j\alpha \ne \emptyset\), i.e. when \(j \ge i_k\) if \(\alpha =(i_k, \ldots , i_1)\). For example, if \(j=1\) then it must be \(\alpha = 0\) or \(\alpha =(1,\ldots ,1)\), so in this case the summation over \(\alpha \in {\mathcal {M}}\) reduces to the summation over \(k=0,1,\dots\), the length of the sequence of 1’s.
To study the properties of the joint monotone numerical radius we need the following lemma.
Lemma 1
For any \(T_1,\ldots , T_d\in B({\mathcal {H}})\) we have
Proof
Using (7) and (8) and the definition of \(Q_j\), we observe that
Now, we use the mutual orthogonality of the orthogonal projections \(P_0, P_1, \ldots , P_d\) to get
\(\square\)
In analogy to Proposition 6 we have the following properties of the joint monotone numerical radius.
Proposition 15
We have
- \((i)_M\) :
-
\({ w}_M(T_1,\ldots ,T_d)=0\)if and only if\(T_1=\ldots =T_d =0\),
- \((ii)_M\) :
-
\(\displaystyle \frac{1}{2} \max _{1\le m\le d} \Vert \sum _{j=m}^d T_j T_j^*\Vert ^{\frac{1}{2}} \le { w}_M(T_1,\ldots ,T_d)\le \max _{1\le m\le d}\Vert \sum _{j=m}^d T_j T_j^*\Vert ^{\frac{1}{2}}\),
- \((iii)_M\) :
-
\({ w}_M(X^*T_1X,\ldots ,X^*T_dX)\le \Vert X\Vert ^2 { w}_M(T_1,\ldots ,T_d)\)for any bounded operator\(X:{\mathcal {H}}\rightarrow {\mathcal {K}}\).
Proof
Ad \((i)_M\). As in the proof of Proposition 6, \({ w}_M(T_1,\ldots ,T_d)=0\) implies \(\sum _{j=1}^d M_j^* \otimes T_j^*=0\). Hence for any \(h=\sum _{k=0}^d e_k \otimes g_k \in {\mathcal {F}}_B(H) \otimes {\mathcal {H}}\), one gets
Hence \(T_j=0\) for any \(j=1,\ldots ,d\).
Ad \((ii)_M\). Due to \({ w}_{cl}(T)\le \Vert T\Vert\) and Lemma 1 we have
The other part follows from \(\frac{1}{2}\Vert T\Vert \le { w}_{cl}(T)\).
Ad \((iii)_M\). Let \(X:{\mathcal {H}}\rightarrow {\mathcal {K}}\) be a bounded operator and let \(g_0,\ldots ,g_d\in {\mathcal {H}}\) satisfy \(\sum _{k=0}^d \Vert g_k\Vert ^2=1\). Define \(C:=\sqrt{\sum _{k=0}^d \Vert Xg_k\Vert ^2}\) and \(h_k:=\frac{1}{C}Xg_k\), \(k=0,1,\ldots ,d\). Then \(\sum _{k=0}^d \Vert h_k\Vert ^2=1\) and \(C\le \Vert X\Vert\). Consequently,
\(\square\)
Remark 16
The properties \((i)_M\), \((ii)_M\) and \((i)_M\) show that the joint monotone numerical radius is a norm on \(B({\mathcal {H}})^{d}\), which, by \((ii)_M\), is actually equivalent to the maximum, over \(1\le m \le d\), of the operator norms of the operator row matrices \([T_m,\ldots ,T_d]\).
Example 17
For \(d=1\) and \(T\in B({\mathcal {H}})\) we get \({ w}_M(T) = { w}_F(T)={ w}_{\mathrm{cl}}(T)\).
This follows immediately from the fact that for \(d=1\) the set \({\mathcal {M}}_k\) consists of the unique term \((1,\ldots ,1)\) (k-times). Hence,
which is exactly the same as the joint free numerical radius for a single operator, and which in turn was shown to be equal to the classical numerical radius (see [9, Sect. 1]).
The joint monotone spectral radius can be computed similarly to the (free) joint spectral radius of Popescu.
Proposition 18
The joint monotone spectral radius satisfies
where
is a set of monotone sequences of lengthk, starting fromj.
Proof
First observe that
due to (6). This also implies
Since the projections \(Q_1, P_2, \ldots , P_d\) are mutually orthogonal, we obtain
Therefore we can write
\(\square\)
4.1 Example: joint monotone numerical radius for weakly monotone annihilation operators
We shall compute the joint monotone numerical radius for d-tuple of weakly monotone annihilators \(M_1,\ldots ,M_d\). An upper bound for \({ w}_M(M_1,\ldots ,M_d)\) is easily obtained.
Proposition 19
For any \(d\ge 2\) we have
Proof
Using the fact that the norm of any creation operator \(M_j\) is 1 and the Cauchy-Schwarz inequality, we have the following
for any family \((g_\alpha )_{\alpha \in {\mathcal {M}}}\) satisfying \(\sum _{\alpha \in {\mathcal {M}}} \Vert g_{\alpha }\Vert ^2=1\). \(\square\)
This is a rough estimate and the following calculations show that perhaps the optimal one would be related to \(\sqrt{d}\). To support this claim, we provide lower estimates of that sort.
It will be instructive to treat first the case \(d=2\).
Proposition 20
For two weakly monotone annihilation operators\(T_1=M_1\)and\(T_2=M_2\), we have\(\displaystyle { w}_M(M_1, M_2) \ge \frac{4}{3\sqrt{3}} \sqrt{2}\approx 0,77 \cdot \sqrt{2}\).
Proof
For the proof we shall analyze in details the expression
For this purpose we shall use the notation \(\alpha =[k,m]\) if \(\alpha =(2,\ldots , 2, 1, \ldots 1)\) is a sequence of m times the digit 2 followed by k times the digit 1. Then we have explicit formulas for the composition \(j\alpha\) with \(j=1,2\), namely \(2[k,m]=[k,m+1]\) if \(k,m\ge 0\), \(1[k,0]=[k+1,0]\) if \(k\ge 1\) and \(1[k,m]=0\) if \(m\ge 1\).
We shall also denote the orthonormal basis \(\displaystyle e_{\alpha }=e_{m,k}\) if \(\alpha =[k,m]\); in particular \(\displaystyle \langle e_{s,r}| e_{u,t}\rangle = 1\) if and only if \(s=u,\ r=t\). Each vector \(g_{[k,m]}\) can be written in the orthonormal basis as \(\displaystyle g_{[k,m]}=\sum _{r,s\ge 0} g_{[k,m]}^{r,s}e_{s,r}\) and then we get
and
We split this summation into two parts.
-
1.
For \(j=1\) we have necessarily \(m=0\) and \(\displaystyle \langle M_1^*e_{s,r}| e_{u,t} \rangle = \langle e_{0,r+1}| e_{u,t} \rangle\) if \(s=0\) and \(\displaystyle \langle M_j^*e_{s,r}| e_{u,t} \rangle = 0\) if \(s\ge 1\). Hence for \(j=1\) we get only the following nonzero term
$$\begin{aligned} \sum _{k \ge 0} \sum _{r \ge 0} \sum _{t,u \ge 0} g_{[k,0]}^{r,s} g_{[k+1,0]}^{t,u} \langle e_{0,r+1}| e_{u,t} \rangle&= \sum _{k, r \ge 0} g_{[k,0]}^{r,0} g_{[k+1,0]}^{r+1,0}. \end{aligned}$$ -
2.
For \(j=2\) and \(m\ge 0\) in a similar manner we get
$$\begin{aligned} \displaystyle \langle M_2^*e_{s,r}| e_{u,t} \rangle = \langle e_{s+1,r}| e_{u,t} \rangle = 1 \text{ if } \text{ and } \text{ only } \text{ if } u=s+1, t=r. \end{aligned}$$Hence the nonzero term in (15) is
$$\begin{aligned} \sum _{k\ge 0} \sum _{m\ge 1} \sum _{r \ge 0} \sum _{t,u \ge 0} g_{[k,m]}^{r,s} g_{[k,m+1]}^{t,u} \langle e_{s+1,r}| e_{u,t} \rangle&= \sum _{k\ge 0} \sum _{m\ge 1} \sum _{r,s \ge 0} g_{[k,m]}^{r,s} g_{[k,m+1]}^{r,s+1}. \end{aligned}$$
So we get
Now we consider special vectors, defined for a constant \(x\in (0, 1)\) and \(k,m,r,s \ge 0\):
We have that in this case the quantity in (14) equals
On the other hand, we have
Hence maximizing the quotient of (18) by (17) we get the function \(\displaystyle f(x):=2x-x^3\), which in (0, 1) has the local maximum \(\frac{4}{3} \sqrt{\frac{2}{3}}\) at \(x=\sqrt{\frac{2}{3}}\).
To sum up, we have
\(\square\)
We are now ready to treat the general case.
Proposition 21
For\(d\ge 3\)and the monotone annihilators\(T_j=M_j\), \(j=1,\ldots ,d\), we have
Proof
We generalize the idea from the proof of the case \(d=2\). For a fixed \(d\ge 3\), any \(\alpha \in {\mathcal {M}}\) is of the form
where \(k_d,\ldots ,k_1\ge 0\). In the sequel, such an element will be simply denoted by \(\alpha =[k_d,\ldots ,k_1]\). For arbitrary (but fixed) \(x\in (0,1)\) we shall consider the vectors
for which we have
Now, we would like to compute
For this purpose observe that, for a fixed \(1\le j \le d\), if \(\alpha =[k_d,\ldots ,k_1]\), then \(j\alpha \ne \emptyset\) if and only if there are no indices bigger than j in \(\alpha\), that is if \(k_{j+1}=0,\ldots ,k_d=0\). In such case \(j\alpha =[0,\ldots ,0,k_j+1,k_{j-1}, \ldots , k_1]\) and
Otherwise \(j\alpha =\emptyset\) and \(\langle g_{\alpha } | M_jg_{j \alpha } \rangle =0\).
This simplifies the expression (20) considerably:
Taking into account the normalization (19) of the family \((g_\alpha )_\alpha\), we arrive to the problem of maximizing the function:
over (0, 1). Setting \(y=1-x^2\) this reduces to the maximization of the function
over (0, 1).
Trying to resolve the optimization problem explicitly, one arrives to the equation
which describes the critical points of \(g_d\). We shall not go this way. We just observe that the value of \(g_d\) at the points \(y_d=1-\frac{1}{d}\) provides the required estimates. Namely,
This shows that the numerical radius satisfies
\(\square\)
Let us remark that numerical calculations reveal that the point \(y_d=1-\frac{1}{d}\) seems to be a good approximation of \(y_{d,max}\), at which the maximum of \(g_d\) is achieved. Moreover, for \(3\le d\le 100\), the relative error appearing when \(g_d(y_{d,max})\) is replaced by \(g_d(y_d)\) is smaller than 2%.
All these calculations and observations have led us to the formulation of the following statement.
Conjecture 1
There exists an absolute constant\(c>0\)such that the joint monotone numerical radius of thed-tuple of weakly monotone annihilators equals
4.2 Example: joint monotone spectral radius for weakly monotone creation operators
Using Proposition 18 we compute the joint monotone spectral radius of d weakly monotone annihilation operators \(M_1, \ldots , M_d\). In this case we have
Indeed, let us denote by \(c_k^j\) the cardinality of the set \({\mathcal {M}}_k^j\), see (13). Then \(c_{jk}=\displaystyle |{\mathcal {M}}_k^j|=\left( {\begin{array}{c}k+d-j\\ k-1\end{array}}\right)\). Since for \(\alpha \in {\mathcal {M}}_k^j\), we have \(M_{\alpha }M_{\alpha }^*=Q_j\), the right-hand side above can be written as
We see that the maximum is achieved for \(m=1\) and the last expression equals
5 Concluding remarks and open problems
This paper is a beginning of the studies of joint numerical and spectral radii related to creation operators independent in noncommutative sense: the monotone and boolean ones. We have showed that some properties are analogous as in the work of Popescu for free creation operators. However, especially for the monotone case, which we base on the weakly monotone creation operators, our results differ from the free case.
There is still a lot about the noncommutative joint numerical and spectral radii to be understood. Some open problems appeared in our study, in particular to find the exact formulas for the joint monotone numerical radius of \(d\ge 2\) weakly monotone annihilation operators (at first glance it seems to be related to \(\sqrt{d}\) rather then to d). But to see the real power of these objects one should search for analogues of the classical power inequality or von Neumann type inequalities. Finally, it would be interesting to know what are the relations between the three types (free, boolean and monotone) of numerical radii.
References
Berger, C.A., Stampfli, J.G.: Mapping theorems for the numerical range. Amer. J. Math. 89, 1047–1055 (1967)
Bożejko, M.: Positive definite functions on the free group and the noncommutative Riesz product. Boll. Un. Mat. Ital. 5(1), 13–21 (1986)
Bożejko, M.: Deformed Fock spaces, Hecke operators and monotone Fock space of Muraki. Demonstr. Math. 45(2), 399–413 (2012)
Crismale, V., Griseta, M.E., Wysoczański, J.: Weakly monotone Fock space and monotone convolution of the Wigner Law. J. Theor. Probab. 33, 268–294 (2020)
Goldberg, M., Tadmor, E.: On the numerical radius and its applications. Linear Algebra Appl. 42, 263–284 (1982)
Gustafson, Karl E., Rao, D.K.M.: Numerical range. The field of values of linear operators and matrices. Springer, New York (1997)
Muraki, N.: A new example of non-commutative “de Moivre-Laplace theorem”. In: S. Watanabe et al (eds.), Prob. Th. Math. Sat., 7th Japan-Russia Symp., World Scientific, Tokyo, 1995, pp. 353–362 (1996)
Muraki, N.: Monotonic independence, monotonic central limit theorem and monotonic law of small numbers. Infin. Dimens. Anal. Quantum Probab. Relat. Top. 4(1), 39–58 (2001)
Popescu, G.: Unitary invariants in multivariable operator theory. Mem. Am. Math. Soc. 200, 941 (2009)
Toeplitz, O.: Das algebraische Analogon zu einem Satze von Fejar. Math. Z. 2(1–2), 187–197 (1918)
Voiculescu, D.: Symmetries of some reduced free product \(C^{*}\)-algebras, In: Operator algebras and their connections with topology and ergodic theory (Busteni, 1983), Lecture Note in Math., Vol. 1132 , pp.115–144. Springer, Heidelberg (1985)
Wysoczański, J.: Monotonic independence on the weakly monotone Fock space and related Poisson type theorem. Infin. Dimens. Anal. Quantum Probab. Relat. Top. 8(2), 259–275 (2005)
Wysoczański, J.: Monotonic independence associated with partially ordered sets. Inf. Dim. Anal. Quantum Probab. Relat. Top. 10(1), 17–41 (2007)
Acknowledgements
AK was supported by the Polish National Science Center Grant SONATA 2016/21/D/ST1/03010. JW was supported by the Polish National Science Center Grant OPUS 2016/21/B/ST1/00628. AK and JW were also supported by the Polish National Agency for Academic Exchange (NAWA) within the POLONIUM program PPN/BIL/2018/1/00197/U/00021.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Communicated by Lyudmila Turowska.
Dedicated to Franciszek Hugon Szafraniec on the occasion of his 80th birthday, with appreciation of his influence on our multidimensional collaboration.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Kula, A., Wysoczański, J. Joint monotone and boolean numerical and spectral radii of d-tuples of operators. Adv. Oper. Theory 5, 1039–1060 (2020). https://doi.org/10.1007/s43036-020-00074-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s43036-020-00074-6