Abstract
Let X be an underlying space with a reference measure \(\sigma \). Let K be an integral operator in \(L^2(X,\sigma )\) with integral kernel K(x, y). A point process \(\mu \) on X is called determinantal with the correlation operator K if the correlation functions of \(\mu \) are given by \(k^{(n)}(x_1,\dots ,x_n)={\text {det}}[K(x_i,x_j)]_{i,j=1,\dots ,n}\). It is known that each determinantal point process with a self-adjoint correlation operator K is the joint spectral measure of the particle density \(\rho (x)=\mathcal A^+(x)\mathcal A^-(x)\) (\(x\in X\)), where the operator-valued distributions \(\mathcal A^+(x)\), \(\mathcal A^-(x)\) come from a gauge-invariant quasi-free representation of the canonical anticommutation relations (CAR). If the space X is discrete and divided into two disjoint parts, \(X_1\) and \(X_2\), by exchanging particles and holes on the \(X_2\) part of the space, one obtains from a determinantal point process with a self-adjoint correlation operator K the determinantal point process with the J-self-adjoint correlation operator \(\widehat{K}=KP_1+(1-K)P_2\). Here \(P_i\) is the orthogonal projection of \(L^2(X,\sigma )\) onto \(L^2(X_i,\sigma )\). In the case where the space X is continuous, the exchange of particles and holes makes no sense. Instead, we apply a Bogoliubov transformation to a gauge-invariant quasi-free representation of the CAR. This transformation acts identically on the \(X_1\) part of the space and exchanges the creation operators \(\mathcal A^+(x)\) and the annihilation operators \(\mathcal A^-(x)\) for \(x\in X_2\). This leads to a quasi-free representation of the CAR, which is not anymore gauge-invariant. We prove that the joint spectral measure of the corresponding particle density is the determinantal point process with the correlation operator \(\widehat{K}\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let X be an underlying space with a reference measure \(\sigma \). (Typically \(X=\mathbb {R}^d\) and \(\sigma \) is the Lebesgue measure.) Let \(\Gamma _X\) denote the space of configurations in X, i.e., locally finite subsets of X. A point process in X is a probability measure on \(\Gamma _X\), see e.g. [13]. A point process \(\mu \) is called determinantal if the correlation functions of \(\mu \) are given by
see e.g. [6, 14, 23, 26]. The function K(x, y) is called the correlation kernel of \(\mu \). To study \(\mu \), one usually considers a (bounded) integral operator K in the (complex) space \(\mathcal H:=L^2(X,\sigma )\) with integral kernel K(x, y). One calls K the correlation operator of \(\mu \).
Assume that the correlation kernel K(x, y) is Hermitian, i.e., \(K(x,y)=\overline{K(y,x)}\), equivalently the operator K is self-adjoint. Then, the Macchi–Soshnikov theorem [23, 26] gives a necessary and sufficient condition of existence of a determinantal point process \(\mu \) with the correlation kernel K(x, y).
It was shown in [18, 20] (see also [6, 27]) that this point process \(\mu \) is the joint spectral measure of the particle density of a gauge-invariant quasi-free representation of the canononical anticommutation relations (CAR). More precisely, assume that operators \(\mathcal A^+(\varphi )\) and \(\mathcal A^-(\varphi )\) \((\varphi \in \mathcal H)\) satisfy the CAR:
Here \(\{A,B\}=AB+BA\) is the anticommutator. Assume that the operators \( \mathcal A^+(\varphi )\) and \( \mathcal A^-(\varphi )\) act in \(\mathcal{A}\mathcal{F}(\mathcal H\oplus \mathcal H)\), the antisymmetric Fock space over \(\mathcal H\oplus \mathcal H\). Let \(\textbf{A}\) be the CAR \(*\)-algebra generated by these operators, and let \(\tau \) be the vacuum state on \(\textbf{A}\). Furthermore, assume that the operators \( \mathcal A^+(\varphi )\) and \( \mathcal A^-(\varphi )\) are such that the state \(\tau \) is gauge-invariant quasi-free, i.e.,
where \(\delta _{m,n}\) is the Kronecker symbol and K is the self-adjoint bounded linear operator in \(\mathcal H\) satisfying
See [2] or [12, Subsection 5.2.3].
Define operator-valued distributions \( \mathcal A^+(x)\) and \( \mathcal A^-(x)\) on X by
The corresponding particle density is formally defined as the operator-valued distribution \(\rho (x):= \mathcal A^+(x) \mathcal A^-(x)\), and in the smeared form,
where \(\Delta \subset X\) is measurable and pre-compact. Note that, at least formally, each operator \(\rho (\Delta )\) is Hermitian and for any sets \(\Delta _1\) and \(\Delta _2\), the operators \(\rho (\Delta _1)\) and \(\rho (\Delta _2)\) commute. The main result of [20] was that, if K is an integral operator and its integral (Hermitian) kernel K(x, y) is such that the corresponding determinantal point process \(\mu \) exists, then the operators \(\rho (\Delta )\) are well-defined, essentially self-adjoint, commuting, and furthermore
where \(\gamma (\Delta ):=|\gamma \cap \Delta |\), the number of points of the configuration \(\gamma \) that belong to \(\Delta \). Formula (7) states that the moments of the operators \(\rho (\Delta )\) under the gauge-invariant quasi-free state state \(\tau \) are equal to the moments of the determinantal point process \(\mu \). It should be stressed that the set of all monomials \(\gamma (\Delta _1)\cdots \gamma (\Delta _n)\) is total in \(L^2(\Gamma _X,\mu )\).
For each measurable, pre-compact set \(\Delta \subset X\), we denote by \(\tilde{\rho }(\Delta )\) the closure of the operator \(\rho (\Delta )\). (Note that the operators \(\tilde{\rho }(\Delta )\) are self-adjoint.) According to [4, Chapter 3], formula (7) means that the determinantal point process \(\mu \) is the joint spectral measure of the operators \(\tilde{\rho }(\Delta )\).
Let us now assume that the underlying space X is divided into two disjoint parts, \(X_1\) and \(X_2\). For \(i=1,2\), let \(P_i\) denote the orthogonal projection of \(\mathcal H\) onto \(\mathcal H_i:=L^2(X_i,\sigma )\), and let \(J:=P_1-P_2\). An (indefinite) J-scalar product in \(\mathcal H\) is defined by
see e.g. [3]. A bounded linear operator operator \(\mathbb K\in \mathcal L(\mathcal H)\) is called J-self-adjoint if \([\mathbb Kf, g] = [f, \mathbb Kg]\) for all \(f, g \in H\). If an integral operator \( \mathbb K\in \mathcal L(\mathcal H)\) is J-self-adjoint, then its integral kernel \(\mathbb K(x,y)\) is called J-Hermitian. The integral kernel \(\mathbb K(x,y)\) of an integral operator \(\mathbb K\in \mathcal L(\mathcal H)\) is J-Hermitian if and only if \( \mathbb K(x,y)=\overline{\mathbb K(y,x)}\) for x and y that belong to the same part \(X_i\) and \(\mathbb K(x,y)=-\overline{\mathbb K(y,x)}\) if \(x\in X_i\), \(y\in X_j\) with \(i\ne j\).
For an arbitrary bounded linear operator \(K\in \mathcal L(\mathcal H)\), we define
Note that the transformation \(K\mapsto \widehat{K}\) is an involution. If an operator K is self-adjoint, then \(\widehat{K}\) is J-self-adjoint, and if an operator \(\mathbb K\in \mathcal L(\mathcal H)\) is J-self-adjoint, then \(\widehat{\mathbb K}\) is self-adjoint.
A necessary and sufficient condition of existence of a determinantal point process with a J-Hermitian correlation kernel \(\mathbb K(x,y)\) was given in [19]. We note that J-Hermitian correlation kernels naturally arise in the context of asymptotic representation theory of classical groups, such as symmetric or unitary groups of growing rank, see [8,9,10,11, 22].
Assume, for a moment, that the underlying space X is discrete and \(\sigma \) is the counting measure. Then any determinantal point process with a J-Hermitian correlation kernel can be obtained from a determinantal point process with a Hermitian correlation kernel through a particle-hole transformation. More precisely, let \(\gamma \) be a configuration in X, i.e., \(\gamma \subset X\). We define a new configuration
i.e., \(I\gamma \) coincides with \(\gamma \) in \(X_1\), and with the holes of \(\gamma \) (the points unoccupied by \(\gamma \)) in \(X_2\). Note that \(I:\Gamma _X\rightarrow \Gamma _X\) is an involution. It was shown in [7] that, if \(\mu \) is a determinantal point process with a correlation operator K, then \(I_*\mu \) , the pushforward of \(\mu \) induced by I, is the determinatal point process with the correlation operator \(\widehat{K}\). Therefore, \(\mu \) is a determinantal point process with a J-Hermitian correlation kernel \(\mathbb K(x,y)\) if and only if \(\mu =I_*\nu \), where \(\nu \) is the determinantal point process with the Hermitian correlation kernel \(\widehat{\mathbb K}(x,y)\). (Note that \(\nu =I_*\mu \).)
If the underlying space X is continuous and \(\sigma \) is a non-atomic measure, then a direct generalization of the above result is impossible. Indeed, for a configuration \(\gamma \) in X, the set \(I\gamma \) defined by (8) is uncountable, hence it is not anymore a configuration in X. Furthermore, the Macchi–Soshnikov theorem and [19, Theorem 3] imply that, if K(x, y), the correlation kernel of a determinantal point process, is either Hermitian or J-Hermitian, then the operator \(\widehat{K}\) is not even an integral operator, i.e., \(\widehat{K}(x,y)\) does not exist.
The aim of this paper is to prove the following main result, which involves a Bogoliubov transformation of the CAR \(*\)-algebra \(\textbf{A}\). We refer the reader to e.g. [12, Section 5.2.2] for the definition of a Bogoliubov transformation and a discussion of its properties.
Main result. Let \(\mathbb K\in \mathcal L(\mathcal H)\) be a J-self-adjoint integral operator, and assume that the J-Hermitian integral kernel \(\mathbb K(x,y)\) of the operator \(\mathbb K\) is the correlation kernel of a determinantal point process \(\mu \). For the self-adjoint operator \(K:=\widehat{\mathbb K}\), consider the corresponding gauge-invariant quasi-free representation of the CAR, i.e., the \(*\)-algebra \(\textbf{A}\) is generated by operators \( \mathcal A^+(\varphi )\), \( \mathcal A^-(\varphi )\) in \(\mathcal{A}\mathcal{F}(\mathcal H\oplus \mathcal H)\) that satisfy (2), (3) and, for the vacuum state \(\tau \) on \(\textbf{A}\), formulas (4), (5) hold. Define a Bogoliubov transformation of the CAR \(*\)-algebra \(\textbf{A}\) by
where for each \(\varphi \in \mathcal H\),
Here \((\mathcal C\varphi )(x):=\overline{\varphi (x)}\) is the complex conjugation. (Note that the vacuum state \(\tau \) on the CAR \(*\)-algebra \(\mathbb A\) generated by the operators \(A^+(\varphi )\), \(A^-(\varphi )\) (\(\varphi \in \mathcal H\)) is still quasi-free but not anymore gauge-invariant.) Let operator-valued distributions \(A^+(x)\) and \(A^-(x)\) be determined by \(A^+(\varphi )\), \(A^-(\varphi )\) similarly to (6). Then the corresponding particle density
is a family of well-defined, essentially self-adjoint, commuting operators in \(\mathcal{A}\mathcal{F}(\mathcal H\oplus \mathcal H)\). Furthermore, for these operators \(\rho (\Delta )\) and the determinantal point process \(\mu \) with the correlation kernel \(\mathbb K(x,y)\), formula (7) holds. In other words, the determinantal point process \(\mu \) is the joint spectral measure of the family of self-adjoint operators \(\tilde{\rho }(\Delta )\).
Note that formula (10) implies that
In words, on the \(X_1\) part of X we use the creation and annihilation operators of the original representation of the CAR, while on the \(X_2\) part of X we exchange the creation and annihilation operators of the original representation. Hence, the Bogoliubov transformation (9), (10) can be thought of as a counterpart of the involution I defined by (8) in the case where the space X is discrete.
In fact, in our derivation of the main result, the existence of the determinantal point process \(\mu \) follows from a general theorem regarding the joint spectral measure of a family of self-adjoint commuting operators, see [20, Theorem 1]. Hence, as a by-product of our considerations, we obtain a new proof of existence of a determinantal point process with a J-Hermitian correlation kernel.
In the case of a discrete space X, Koshida [16] proved that each pfaffian point process appears (in the terminology of the present paper) as the joint spectral measure of the particle density of a quasi-free representation of the CAR. Koshida notes: ‘it seems highly nontrivial if our construction can be extended to the case of continuous systems.’ While the present paper does not provide a solution to this problem, it still solves it for a particular class of (non-gauge-invariant) quasi-free states.
The paper is organized as follows. In Sect. 2, we discuss necessary preliminaries regarding determinantal point processes, J-Hermitian correlation kernels, the correlation measures and the joint spectral measure of a family of commuting self-adjoint operators, and quasi-free states on the CAR algebra.
In Sect. 3, we employ heuristic considerations, involving formula (11), in order to give a rigorous definition of Hermitian operators \(\rho (\Delta )\). We also prove that these operators (algebraically) commute.
In Sect. 4, we derive rigorous formulas for the Wick (normal) product
In Sect. 5, we formulate the main theorem of the paper (Theorem 5.2), which states that the the determinantal point process \(\mu \) with the J-Hermitian correlation kernel \( \mathbb K(x,y)\) is the joint spectral measure of the family of the commuting self-adjoint operators \(\tilde{\rho }(\Delta )\). We start proving this result in Sect. 5.
Finally, in Sect. 6, we prove that the operators \(\rho (\Delta )\) possess correlation functions and these are given by the right-hand side of formula (1) in which K(x, y) is replaced by \(\mathbb K(x,y)\). This concludes the proof of our main result.
2 Preliminaries
2.1 Determinantal point processes
Let X be a locally compact Polish space, let \(\mathcal B(X)\) be the Borel \(\sigma \)-algebra on X, and let \(\mathcal B_0(X)\) denote the collection of all sets from \(\mathcal B(X)\) which are pre-compact. The configuration space over X is defined as the set of all locally finite subsets of X:
Here, for a set \(\Lambda \), \(|\Lambda |\) denotes its capacity. Elements \(\gamma \in \Gamma _X\) are called configurations. One identifies each configuration \(\gamma =\{x_i\}_{i\ge 1}\) with the measure \(\gamma =\sum _i\delta _{x_i}\) on X. Here, for \(x\in X\), \(\delta _{x}\) denotes the Dirac measure with mass at x. Through this identification, one gets the embedding of \(\Gamma _X\) into the space of all Radon (i.e., locally finite) measures on X.
The space \(\Gamma _X\) is endowed with the vague topology, i.e., the weakest topology on \(\Gamma _X\) with respect to which all maps \(\Gamma _X\ni \gamma \mapsto \langle \gamma ,f\rangle =\sum _{x\in \gamma } f(x)\), \(f\in C_0(X)\), are continuous. Here \(C_0(X)\) is the space of all continuous real-valued functions on X with compact support. We will denote by \(\mathcal B(\Gamma _X)\) the Borel \(\sigma \)-algebra on \(\Gamma _X\). A probability measure \(\mu \) on \((\Gamma _X,\mathcal B(\Gamma _X))\) is called a point process on X. For more detail, see e.g. [13, 15].
A point process \(\mu \) can be described with the help of its correlation measures. Denote \(X^{(n)}:=\{(x_1,\dots ,x_n)\in X^n\mid x_i\ne x_j\text { if }i\ne j\}\). The n-th correlation measure of \(\mu \) is the symmetric measure \(\theta ^{(n)}\) on \(X^{(n)}\) that satisfies
for all measurable symmetric functions \(f^{(n)}:X^{(n)}\rightarrow [0,\infty )\). Let \(\sigma \) be a reference Radon measure on \((X,\mathcal B(X))\). If the correlation measure \(\theta ^{(n)}\) has density \(k^{(n)}:X^{(n)}\rightarrow [0,\infty )\) with respect to \(\frac{1}{n!}\,\sigma ^{\otimes n}\), then \(k^{(n)}\) is called the n-th correlation function of \(\mu \). Under a mild condition on the growth of correlation measures as \(n\rightarrow \infty \), they determine a point process uniquely [17].
Recall that a point process \(\mu \) is called determinantal if there exists a complex-valued function K(x, y) on \(X^2\), called the correlation kernel, such that (1) holds, see e.g. [6, 26]. The integral operator K in the complex \(L^2\)-space \(\mathcal H=L^2(X,\sigma )\) which has integral kernel K(x, y) is called the correlation operator of \(\mu \).
Note that, for a given integral operator K in \(\mathcal H\), the integral kernel K(x, y) is defined up to a set of zero measure \(\sigma ^{\otimes 2}\). When calculating the value of \(\det \big [K(x_i,x_j)\big ]_{i,j=1}^n\), one has to use the values of \(K(\cdot ,\cdot )\) on the diagonal \(\{(x,x)\in X^2\mid x\in X\}\). However, the latter set is of zero measure \(\sigma ^{\otimes 2}\) if the measure \(\sigma \) is non-atomic, i.e., \(\sigma (\{x\})=0\) for all \(x\in X\). Hence, when speaking about the correlation operator K of a determinantal point process, one has to properly choose the values of the integral kernel of K on the diagonal in \(X^2\).
2.2 J-Hermitian correlation kernels
Assume that the underlying space X is split into two disjoint measurable parts, \(X_1\) and \(X_2\), of positive measure \(\sigma \). Just as in Introduction, we denote by \(P_i\) the orthogonal projection of \(\mathcal H\) onto \(\mathcal H_i=L^2(X_i,\sigma )\), and we let \(J=P_1-P_2\).
According to the orthogonal sum \(\mathcal H=\mathcal H_1\oplus \mathcal H_2\), each operator \(A\in \mathcal L(\mathcal H)\) can be represented in the block form,
where \(A^{ij}:\mathcal H_j\rightarrow \mathcal H_i\), \(i,j=1,2\). Here \(A^{ij}:=P_iAP_j\). Then the operator A being J-self-adjoint means that \((A^{ii})^*=A^{ii}\) (\(i=1,2\)) and \((A^{21})^*=-A^{12}\).
We denote by \(\mathcal S_1(\mathcal H)\) the set of all trace-class operators in \(\mathcal H\), and by \(\mathcal S_2(\mathcal H)\) the set of all Hilbert–Schmidt operators in \(\mathcal H\). For \(\Delta \in \mathcal B_0(X)\), we denote by \(P_\Delta \) the orthogonal projection of \(\mathcal H\) onto \(L^2(\Delta ,\sigma )\). For \(i=1,2\), we denote \(\mathcal B_0(X_i):=\{\Delta \in \mathcal B_0(X)\mid \Delta \subset X_i\}\).
We say that an operator \(K\in \mathcal L(\mathcal H)\) is locally trace-class on \(X_i\) (\(i=1,2\)) if, for each \(\Delta _i\in \mathcal B_0(X_i)\), we have \(K_{\Delta _i}:=P_{\Delta _i}KP_{\Delta _i}\in \mathcal S_1(\mathcal H)\).
The following lemma can be easily checked by using basic properties of trace-class and Hilbert–Schmidt operators, see e.g. [25].
Lemma 2.1
Let \( K\in \mathcal L(\mathcal H)\) satisfy \(\textbf{0}\le K\le \textbf{1}\). Define \(K_1:=\sqrt{K}\), \(K_2:=\sqrt{\textbf{1}- K}\). Then the following statements are equivalent.
-
(i)
The operator K is locally trace-class on \(X_1\) and the operator \(\textbf{1}- K\) is locally trace-class on \(X_2\).
-
(ii)
For any \(\Delta _i\in \mathcal B_0(X_i)\) (\(i=1,2\)), we have \( K_iP_{\Delta _i}\in \mathcal S_2(\mathcal H)\), or equivalently \(P_{\Delta _i}K_i\in \mathcal S_2(\mathcal H)\).
Note that, in Lemma 2.1, \(K_iP_i\) and \(P_i K_i\) (\(i=1,2\)) are integral operators and their respective integral kernels \( K_i(x, y)\) with \((x,y)\in (X\times X_i)\cup (X_i\times X)\) satisfy
for any \(\Delta _i\in \mathcal B_0(X_i)\). Without loss of generality, we may assume that \(K_i(x,y)=\overline{K_i(y,x)}\) for all \((x,y)\in (X\times X_i)\cup (X_i\times X)\), and \(\int _X | K(x,y)|^2\,\sigma (dy)<\infty \) for all \(x\in X_i\).
Now consider a J-self-adjoint operator \(\mathbb K\in \mathcal L(\mathcal H)\) and denote \(K:=\widehat{\mathbb K}=\mathbb KP_1+(\textbf{1}-\mathbb K)P_2\). Note that \(\mathbb K^{11}= K^{11}\), \(\mathbb K^{22}= (\textbf{ 1}-K)^{22}\), \(\mathbb K^{21}=K^{21}\), and \(\mathbb K^{12}=- K^{12}=-(K^{21})^*\). Let us assume that the operator K satisfies the assumptions of Lemma 2.1, equivalently the operator \(\mathbb K\) is locally trace-class on both \(X_1\) and \(X_2\) and \(\textbf{0}\le \widehat{\mathbb K}\le \textbf{1}\).
Let us show that \(\mathbb K\) is an integral operator, and let us present an integral kernel of \(\mathbb K\). For \(i=1,2\), we set
which is an integral kernel of \(\mathbb K^{ii}\). Note that, for all \((x,y)\in X_i^2\), we have \(\mathbb K(y,x)=\overline{\mathbb K(x,y)}\). Next, for any \(\Delta _i\in \mathcal B_0(X_i)\) (\(i=1,2\)), we have \(P_{\Delta _2}\mathbb KP_{\Delta _1}= P_{\Delta _2}\mathcal \mathbb KP_{\Delta _1}\in \mathcal S_2(\mathcal H)\). Hence, \(\mathbb K^{21}\) is an integral operator. We choose an arbitrary integral kernel of \(\mathbb K^{21}\), denoted by \(\mathbb K(x,y)\) with \(x\in X_2\) and \(y\in X_1\). Finally, we set \(\mathbb K(x,y)=-\overline{\mathbb K(y,x)}\) for \(x\in X_1\) and \(y\in X_2\). Thus, we have constructed an integral kernel of the operator \(\mathbb K\).
The following theorem is shown in [19, Theorem 2].
Theorem 2.2
Let a J-self-adjoint operator \(\mathbb K\in \mathcal L(\mathcal H)\) be locally trace-class on both \(X_1\) and \(X_2\) and such that \(\textbf{0}\le \widehat{\mathbb K}\le \textbf{1}\). Let the integral kernel \(\mathbb K(x,y)\) of the integral operator \(\mathbb K\) be chosen as above. Then there exists a unique determinantal point process with the correlation kernel \(\mathbb K(x,y)\).
Remark 2.3
In fact, the conditions of Theorem 2.2 are necessary for the existence of a determinantal point process with a J-Hermitian correlation kernel, see [19, Theorem 3].
Remark 2.4
While Theorem 2.2 will serve as a motivation for our studies, we will not actually use it and will derive the existence of a determinantal point process as in Theorem 2.2 by methods different to [19]. Nevertheless, we will use the choice of the integral kernel \(\mathbb K(x,y)\) as described above.
2.3 Joint spectral measure of a family of commuting self-adjoint operators
Let us now present a result from [20] on the joint spectral measure of a family of commuting self-adjoint operators. Our brief presentation essentially follows [1, Section 4].
Let \(\mathcal F\) be a separable Hilbert space and let \(\mathcal D\) be a dense subspace of \(\mathcal F\). For each \(\Delta \in \mathcal B_0(X)\), let \(\rho (\Delta ):\mathcal D\rightarrow \mathcal D\) be a linear Hermitian operator in \(\mathcal F\). We further assume:
-
for any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\) with \(\Delta _1\cap \Delta _2=\varnothing \), we have \(\rho (\Delta _1\cup \Delta _2)=\rho (\Delta _1)+\rho (\Delta _2)\);
-
the operators \(\rho (\Delta )\) commute, i.e., \([\rho (\Delta _1),\rho (\Delta _2)]=0\) for any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\).
Let \(\mathcal A\) denote the (commutative) \(*\)-algebra generated by \((\rho (\Delta ))_{\Delta \in \mathcal B_0(X)}\). Let \(\Omega \) be a fixed vector in \(\mathcal D\) with \(\Vert \Omega \Vert _{\mathcal F}=1\), and let a state \(\tau :\mathcal A\rightarrow \mathbb C\) be defined by \(\tau (a):=(a\Omega ,\Omega )_{\mathcal F}\) for \(a\in \mathcal A\).
We define Wick polynomials in \(\mathcal A\) by the following recurrence formula:
where \(\Delta ,\Delta _1,\dots ,\Delta _{n+1}\in \mathcal B_0(X)\) and \(n\in \mathbb N\). It is easy to see that, for each permutation \(\pi \in S_n\),
We assume that, for each \(n\in \mathbb N\), there exists a symmetric measure \(\theta ^{(n)}\) on \(X^n\) that is concentrated on \(X^{(n)}\) (i.e., \(\theta ^{(n)}(X^n\setminus X^{(n)})=0\)) and such that
Note that, if the measure \(\theta ^{(n)}\) exists, then it is unique. The \(\theta ^{(n)}\) is called the n-th correlation measure of the operators \(\rho (\Delta )\). If \(\theta ^{(n)}\) has a density \(k^{(n)}\) with respect to \(\frac{1}{n!}\sigma ^{\otimes n}\), then \(k^{(n)}\) is called the nth correlation function of the operators \(\rho (\Delta )\).
Theorem 2.5
[20] Let \((\rho (\Delta ))_{\Delta \in \mathcal B_0(X)}\) be a family of Hermitian operators in \(\mathcal F\) as above. In particular, these operators have correlation measures \((\theta ^{(n)})_{n=1}^\infty \) respective the state \(\tau \). Furthermore, we assume that the following two conditions are satisfied.
(LB1) For each \(\Delta \in \mathcal B_0(X)\), there exists a constant \(C_\Delta >0\) such that
(LB2) For any sequence \(\{\Delta _{l}\}_{l\in \mathbb {N}}\subset \mathcal {B}_{0}(X)\) such that \(\Delta _{l}\downarrow \varnothing \) (i.e., \(\Delta _1\supset \Delta _2\supset \Delta _3\supset \cdots \) and \(\bigcap _{l=1}^\infty \Delta _l=\varnothing \)), we have \(C_{\Delta _{l}}\rightarrow 0\) as \(l\rightarrow \infty \).
Then the following statements hold.
-
(i)
Let \(\mathfrak D:=\{a\Omega \mid a\in \mathcal A\}\) and let \(\mathfrak F\) denote the closure of \(\mathfrak D\) in \(\mathcal F\). Each operator \((\rho (\Delta ),\mathfrak D)\) is essentially self-adjoint in \(\mathfrak F\), i.e., the closure of \(\rho (\Delta )\), denoted by \(\widetilde{\rho }(\Delta )\), is a self-adjoint operator in \(\mathfrak F\).
-
(ii)
For any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\), the projection-valued measures (resolutions of the identity) of the operators \(\widetilde{\rho }(\Delta _1)\) and \(\widetilde{\rho }(\Delta _2)\) commute.
-
(iii)
There exist a unique point process \(\mu \) on X and a unique unitary operator\(U:\mathfrak F\rightarrow L^2(\Gamma _X,\mu )\) satisfying \(U\Omega =1\) and
$$\begin{aligned} U(\rho (\Delta _1)\cdots \rho (\Delta _{n})\Omega )=\gamma (\Delta _1)\cdots \gamma (\Delta _n)\end{aligned}$$(20)for any \(\Delta _1,\dots ,\Delta _{n}\in \mathcal B_0(X)\) (\(n\in \mathbb N\)). In particular,
$$\begin{aligned} \tau \big (\rho (\Delta _1)\cdots \rho (\Delta _{n})\big )= \int _{\Gamma _X}\gamma (\Delta _1)\cdots \gamma (\Delta _n)\,\mu (d\gamma ).\end{aligned}$$(21) -
(iv)
The correlations measures of the point process \(\mu \) are \((\theta ^{(n)})_{n=1}^\infty \).
According to [4, Chapter 3], the point process \(\mu \) from Theorem 2.5 is the joint spectral measure of the family of commuting self-adjoint operators \((\widetilde{\rho }(\Delta ))_{\Delta \in \mathcal B_0(X)}\).
2.4 Quasi-free states on the CAR algebra
Let \(\mathcal F\) be a separable Hilbert space, and let \(a^+(\varphi )\) and \(a^-(\varphi )\) (\(\varphi \in \mathcal H\)) be bounded linear operators in \(\mathcal F\) such that \(a^+(\varphi )\) linearly depends on \(\varphi \) and \(a^-(\varphi )=\big (a^+(\varphi )\big )^*\). Let \(a^+(\varphi )\) and \(a^-(\varphi )\) satisfy the CAR, i.e., formula (3) holds in which the operators \(\mathcal A^+(\varphi )\), \(\mathcal A^-(\varphi )\) are replaced by \(a^+(\varphi )\), \(a^-(\varphi )\). Let \(\mathbb A\) be the \(*\)-algebra generated by these operators. We define field operators \(b(\varphi ):=a^+(\varphi )+a^-(\varphi )\) (\(\varphi \in \mathcal H\)). As easily seen, these operators also generate \(\mathbb A\).
Let \(\tau :\mathbb A\rightarrow \mathbb C\) be a state on \(\mathbb A\). The state \(\tau \) is completely determined by the functionals \(T^{(n)}:\mathcal H^n\rightarrow \mathbb C\) \((n\in \mathbb N)\) defined by
The state \(\tau \) is called quasi-free if
where the summation is over all partitions \(\nu =\big \{\{i_1,j_1\},\dots ,\{i_n,j_n\}\big \}\) of \(\{1,\dots ,2n\}\) with \(i_k<j_k\) (\(k=1,\dots ,n\)) and \({\text {Cross}}(\nu )\) denotes the number of all crossings in \(\nu \), i.e., the number of all choices of \(\{i_k,j_k\},\{i_l,j_l\}\in \nu \) such that \(i_k<i_l<j_k<j_l\), see e.g. [12, Section 5.2.3].
The state \(\tau \) is called gauge-invariant if, for each \(q\in \mathbb C\) with \(|q|=1\), we have \(T^{(n)}(q\varphi _1,\dots ,q\varphi _n)=T^{(n)}(\varphi _1,\dots ,\varphi _n)\) for all \(\varphi _1,\dots ,\varphi _n\in \mathcal H\), \(n\in \mathbb N\). The state \(\tau \) can also be uniquely characterized by the n-point functions \(S^{(m,n)}:\mathcal H^{m+n}\rightarrow \mathbb C\) (\(m+n\ge 1\)) defined by
The state \(\tau \) is gauge-invariant quasi-free if and only if
Let us briefly recall the Araki–Wyss [2] construction of the gauge-invariant quasi-free states. Let \(\mathcal G\) denote a separable complex Hilbert space. Let \(\mathcal{A}\mathcal{F}(\mathcal G):=\bigoplus _{n=0}^\infty \mathcal G^{\wedge n}n!\) denote the antisymmetric Fock space of \(\mathcal G\). Here \(\wedge \) denotes the antysymmetric tensor product and elements of the Hilbert space \(\mathcal{A}\mathcal{F}(\mathcal G)\) are sequences \(g=(g^{(n)})_{n=0}^\infty \) with \(g^{(n)}\in \mathcal G^{\wedge n}\) (\( \mathcal G^{\wedge 0}:=\mathbb C\)) and \(\Vert g\Vert ^2_{\mathcal{A}\mathcal{F}(\mathcal G)}=\sum _{n=0}^\infty \Vert g^{(n)}\Vert ^2_{\mathcal G^{\wedge n}}\,n!<\infty \). The vector \(\Omega =(1,0,0,\dots )\) is called the vacuum.
For \(\varphi \in \mathcal G\), we define a creation operator \(a^+(\varphi )\in \mathcal L\big (\mathcal{A}\mathcal{F}(\mathcal G)\big )\) by \(a^+(\varphi )g^{(n)}:=\varphi \wedge g^{(n)}\) for \(g^{(n)}\in \mathcal G^{\wedge n}\). For each \(\varphi \in \mathcal G\), we define an annihilation operator \(a^-(\varphi ):=a^+(\varphi )^*\). Then,
for all \(g_1,\dots ,g_n\in \mathcal G\). Note that
The operators \(a^+(\varphi )\), \(a^-(\varphi )\) satisfy the CAR (over \(\mathcal G\)).
Let now \(\mathcal G=\mathcal H\oplus \mathcal H\), and for \(\varphi \in \mathcal H\) and \(\Diamond \in \{+,-\}\), we denote \(a^\Diamond _1(\varphi ):=a^\Diamond (\varphi ,0)\), \(a^\Diamond _2(\varphi ):=a^\Diamond (0,\varphi )\).
We fix any \(K\in \mathcal L(\mathcal H)\) such that \(\textbf{0}\le K\le \textbf{1}\), and define the operators \(K_1\) and \(K_2\) as in Lemma 2.1. We define operators
where \(\mathcal C\) is the complex conjugation in \(\mathcal H\). The operators \(\mathcal A^+(\varphi )\), \(\mathcal A^-(\varphi )\) satisfy the CAR (2), (3). Let \(\textbf{A}\) denote the corresponding CAR \(*\)-algebra. The vacuum state on \(\textbf{A}\) is defined by \(\tau (a):=(a\Omega ,\Omega )_{\mathcal{A}\mathcal{F}(\mathcal G)}\) (\(a\in \textbf{A}\)). This state is gauge-invariant quasi-free. More exactly, setting \(\mathcal F=\mathcal{A}\mathcal{F}(\mathcal G)\) and \(a^\pm (\varphi )=\mathcal A^\pm (\varphi )\), one shows that formulas (25), (26) hold, with \(S^{(1,1)}(\varphi ,\psi )=(K\varphi ,\psi )_{\mathcal H}\) (\(\varphi ,\psi \in \mathcal H\)). In fact, each gauge-invariant quasi-free state on the CAR algebra over \(\mathcal H\) can be constructed in such a way [2].
Next, just as in Sect. 2.2, we assume that the space X is divided into two disjoint parts, \(X_1\) and \(X_2\). We define operators \(A^+(\varphi )\), \(A^-(\varphi )\) (\(\varphi \in \mathcal H\)) by formula (10). Hence, by (28),
These operators also satisfy the CAR and denote by \(\mathbb A\) the corresponding CAR \(*\)-algebra. (Note that we have the equality of \(\textbf{A}\) and \(\mathbb A\) as sets.) The vacuum state \(\tau \) on \(\mathbb A\) is not anymore gauge-invariant but it is still quasi-free. More exactly, setting \(\mathcal F=\mathcal{A}\mathcal{F}(\mathcal G)\) and \(a^\pm (\varphi )=A^\pm (\varphi )\), one easily shows that formulas (22)–(24) hold, with
Here \(\mathbb J\varphi :=P_1\varphi +P_2\mathcal C\varphi \).
Remark 2.6
The n-point functions \(S^{(m,n)}\) for the state \(\tau \) on \(\mathbb A\) can be calculated as follows. First, we note that
Next, if \(m+n\) is odd, then \(S^{(m,n)}=0\), and if \(m+n=2k\) (\(k\ge 2\)), then, by using Lemma 6.1 below, we get
where the summation is over all partitions \(\nu =\big \{\{i_1,j_1\},\dots ,\{i_k,j_k\}\big \}\) of \(\{1,\dots ,2k\}\) with \(i_l<j_l\) and \(S^{(2)}_{i_l,\,j_l}:=S^{(2,0)}\) if \(j_l\le m\), \(S^{(2)}_{i_l,\,j_l}:=S^{(1,1)}\) if \(i_l\le m<j_l\), \(S^{(2)}_{i_l,\,j_l}:=S^{(0,2)}\) if \(i_l\ge m+1\) (\(l=1,\dots ,k\)).
3 Rigorous Construction of the Particle Density
Let \(\mathbb K\in \mathcal L(\mathcal H)\) be a J-self-adjoint operator satisfying the assumptions of Theorem 2.2, and let \(K:=\widehat{\mathbb K}\). Let \(\mathcal G=\mathcal H\oplus \mathcal H\), and let the operators \(A^+(\varphi ),A^-(\varphi )\in \mathcal L\big (\mathcal{A}\mathcal{F}(\mathcal G)\big )\) (\(\varphi \in \mathcal H\)) be defined by (29). Let the operator-valued distributions \(A^+(x)\), \(A^-(x)\) (\(x\in X\)) be determined by \(A^+(\varphi )\), \(A^-(\varphi )\) similarly to (6).
Recall that \(\mathcal{A}\mathcal{F}(\mathcal G)\) consists of all sequences \(g=(g^{(n)})_{n=0}^\infty \) with \(g^{(n)}\in \mathcal G^{\wedge n}\) satisfying \(\sum _{n=0}^\infty \Vert g^{(n)}\Vert ^2_{\mathcal G^{\wedge n}}\,n!<\infty \). We denote by \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\) the dense subspace of \(\mathcal{A}\mathcal{F}(\mathcal G)\) that consists of all finite sequences \(g=(g^{(n)})_{n=0}^\infty \) from \(\mathcal{A}\mathcal{F}(\mathcal G)\), i.e., for some \(N\in \mathbb N\) (depending on g), we have \(g^{(n)}=0\) for all \(n\ge N\). We endow \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\) with the topology of the locally convex direct sum of the Hilbert spaces \(\mathcal G^{\wedge n}\).
We denote by \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\) the space of continuous linear operators in \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\). Note that a linear operator A acting in \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\) is continuous if and only if, for each k there exists N such that \(A\mathcal G^{\wedge k}\subset \bigoplus _{n=0}^N\mathcal G^{\wedge n}\) and A acts continuously from \(\mathcal G^{\wedge k}\) into \(\bigoplus _{n=0}^N\mathcal G^{\wedge n}\), see e.g. [24, Chapter II, Section 6].
Recall the heuristic definition (11) of \(\rho (\Delta )\). Our aim in this section is to rigorously define \(\rho (\Delta )\) for each \(\Delta \in \mathcal B_0(X)\) as an operator from \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\) which is Hermitian in \(\mathcal{A}\mathcal{F}(\mathcal G)\). To this end, we start with heuristic considerations.
Think of \(K_1\) and \(K_2\) as self-adjoint integral operators with integral Hermitian kernels \( K_1(x,y)\) and \( K_2(x,y)\), respectively. Let \(\Delta \in \mathcal B_0(X_1)\), and denote by \(\chi _\Delta \) the indicator function of the set \(\Delta \). We have
From here, and using similar calculations when \(\Delta \in \mathcal B_0(X_i)\) (\(i=1,2\)), we formally conclude that
Denoting \(\rho (x):=A^+(x)A^-(x)\), we get, for \(x\in X_1\),
and for \(x\in X_2\),
Remark 3.1
Comparing formulas (30) and (31), we see that the right-hand side of (31) can be obtained from the right-hand side of (30) by swapping places of \(a_1^\Diamond (K_1(x,\cdot ))\) and \(a_2^\Diamond (K_2(\cdot ,x))\) (\(\Diamond \in \{+,-\}\)).
For \(\Delta \in \mathcal B_0(X_1)\), let us now rigorously define \(\int _X \rho (x)\sigma (dx)\).
The creation and annihilation operators. We will now define
First, we note that, for \(\varphi ,\psi \in \mathcal G\) and \(g^{(n)}\in \mathcal G^{\wedge n}\),
Here, for each \(n\in \mathbb N\), we denote by \(\mathcal A_n\) the antisymmetrization operator in \(\mathcal G^{\otimes n}\), i.e., the orthogonal projection of \(\mathcal G^{\otimes n}\) onto \(\mathcal G^{\wedge n}\).
For \(\varphi ^{(2)}\in \mathcal G^{\otimes 2}\), we define a creation operator \(a^+(\varphi ^{(2)})\) in \(\mathcal A\mathcal F_{\textrm{fin}}(\mathcal G)\) by
Obviously \(a^+(\varphi ^{(2)})\in \mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\). We also denote
which also belongs to \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\). One can easily derive explicit formulas for the action of this operator.
Thus, we formally have
To give the above operator a rigorous meaning, we will first identify \(\int _\Delta K_2(\cdot ,x)\otimes K_1(x,\cdot ) \,\sigma (dx) \) as an element of \(\mathcal H^{\otimes 2}\).
For a linear operator \(B\in \mathcal L(\mathcal H)\), denote \(\overline{B}:=\mathcal CB\mathcal C\), i.e., \(\overline{B}\) is the complex conjugate of B. If B is an integral operator with integral kernel B(x, y), then \(\overline{B}\) is the integral operator with integral kernel \(\overline{B(x,y)}\). In particular, if B is self-adjoint, then the integral kernel of \(\overline{B}\) is B(y, x).
Now, for any \(\varphi ,\psi \in \mathcal H\), we formally calculate
Since \( P_\Delta K_1\in \mathcal S_2(H)\) (see Lemma 2.1 (ii)), we get \(K_2P_\Delta K_1\in \mathcal S_2(H)\). Therefore, \(K_2P_\Delta K_1\) is an integral operator and we denote its integral kernel by \(\big (K_2P_\Delta K_1\big )(x,y)\). Note that \(K_2P_\Delta K_1(\cdot ,\cdot )\in \mathcal H^{\otimes 2}\). Thus, we continue (33) as follows:
Hence, we rigorously define
We define an isometry
We denote \(\big (K_2P_\Delta K_1\big )_{2,1}:=\mathcal I_{21}(K_2P_\Delta K_1)(\cdot ,\cdot )\). Then, in view of (32) and (34), we define
and so
The neutral operator. Our next aim is to rigorously define operators
For a linear operator \(B\in \mathcal G\), the differential second quantization of B is defined as a linear operator \(d\Gamma (B)\in \mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\) satisfying \(d\Gamma (B)\Omega :=0\) (recall that \(\Omega \) is the vacuum), and for any \(g_1,\dots ,g_n\in \mathcal G\),
As easily seen, for any \(\varphi ,\psi \in \mathcal G\), we have \(a^+(\varphi )a^-(\psi )=d\Gamma \big ((\cdot ,\psi )_{\mathcal G}\,\varphi \big )\), i.e., the differential second quantization of the operator \(\mathcal G\ni g\mapsto (g,\psi )_{\mathcal G}\,\varphi \). Hence, we may formally write
where the isometry \(\mathcal I_2:\mathcal H\rightarrow \mathcal G=\mathcal H\oplus \mathcal H\) is defined by \(\mathcal I_2\varphi :=(0,\varphi )\).
Let us define \(\int _\Delta \big (\cdot ,K_2(\cdot ,x)\big )_\mathcal H,K_2(\cdot ,x)\,d\sigma (x)\) as a bounded linear operator in \(\mathcal H\). For \(\varphi ,\psi \in \mathcal H\), we formally calculate
Thus, we rigorously define
In view of (37) and (38), we define
Next, using the CAR, we formally have
Since \(P_\Delta K_1\) is a Hilbert–Schmidt operator (see Lemma 2.1), we have
where \(\Vert \cdot \Vert _{2}\) denotes the Hilbert–Schmidt norm in \(\mathcal S_2(\mathcal H)\). Noting that the operator \(K_\Delta =P_\Delta K P_\Delta \) is self-adjoint, we easily see that
Hence, by (40)–(42) and similarly to (39), we rigorously define
Thus, formulas (35), (36), (39), and (43) imply a rigorous definition of \(\rho (\Delta )\) for \(\Delta \in \mathcal B_0(X_1)\):
Next, we note, if \(\varphi ^{(2)}\in \mathcal G^{\otimes 2}\) and \(\psi ^{(2)}\in \mathcal G^{\otimes 2}\) is defined by \(\psi ^{(2)}(x,y):=\varphi ^{(2)}(y,x)\), then \(a^+(\psi ^{(2)})=-a^+(\varphi ^{(2)})\). Using this observation and Remark 3.1, we similarly define \(\rho (\Delta )\) for \(\Delta \in \mathcal B_0(X_2)\):
Finally, for each \(\Delta \in \mathcal B_0(X)\), we define \(\rho (\Delta ):=\rho (\Delta \cap X_1)+\rho (\Delta \cap X_2)\).
We sum up our considerations in the following definition.
Definition 3.2
For each \(\Delta \in \mathcal B_0(X)\), we define
where \(J_\Delta :=P_{\Delta \cap X_1}-P_{\Delta \cap X_2}\). Each operator \(\rho (\Delta )\) belongs to \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\).
We note that, for each \(\Delta \in \mathcal B_0(X)\), \(\rho (\Delta )\) is a densely defined, Hermitian operator in \(\mathcal{A}\mathcal{F}(\mathcal G)\).
Let \((e_i)_{i=1}^\infty \) be an orthonormal basis for \(\mathcal H\) consisting of real-valued functions. The following proposition can be easily proved by using Definition 3.2.
Proposition 3.3
For \(\Delta \in \mathcal {B}_0(X_1)\), we have
and for \(\Delta \in \mathcal {B}_0(X_2)\), we have
In formulas (44) and (45), the series converge strongly in \(\mathcal L(\mathcal {F}_{\textrm{fin}}(\mathcal G))\), i.e., for each \(f\in \mathcal {F}_{\textrm{fin}}(\mathcal G)\), the series applied to the vector f converges in \(\mathcal {F}_{\textrm{fin}}(\mathcal G)\) (hence also in \(\mathcal {F}(\mathcal G)\)).
Remark 3.4
Formulas (44) and (45) could serve as an alternative definition of \(\rho (\Delta )\) for \(\Delta \) from \(\mathcal B_0(X_1)\) or \(\mathcal B_0(X_2)\), respectively. However, if we initially accepted (44) and (45) as the definition of \(\rho (\Delta )\), it would not be a priori clear if such a definition does not depend on the choice of a real orthonormal basis in \(\mathcal H\).
Proposition 3.5
For any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\), we have \(\rho (\Delta _1)\rho (\Delta _2)=\rho (\Delta _2)\rho (\Delta _1)\).
To prove Proposition 3.5, let us recall a result on strong convergence of bounded linear operators. Let \(E_1\) and \(E_2\) be Hilbert (or even Banach) spaces. Let \((B_n)_{n=1}^\infty \) be a sequence from \(\mathcal L(E_1,E_2)\) and assume that \((B_n)_{n=1}^\infty \) converges strongly to \(B\in \mathcal L(E_1,E_2)\). Then, the uniform boundedness principle states that \(\sup _{n\in \mathbb N}\Vert B_n\Vert <\infty \), see e.g. [5, Chapter 8, Section 2]. This immediately implies
Lemma 3.6
Let \(E_1,E_2,E_3\) be Hilbert spaces, let \(\{B_n\}_{n=1}^\infty \in \mathcal L(E_1,E_2)\) and \(\{C_n\}_{n=1}^\infty \in \mathcal L(E_2,E_3)\). Assume that the series \(\sum _{n=1}^\infty B_n\) and \(\sum _{n=1}^\infty \) strongly converge in \(\mathcal L(E_1,E_2)\) and \(\mathcal L(E_2,E_3)\), respectively. Then the series \(\sum _{m=1}^\infty \sum _{n=1}^\infty B_mC_n\) converges strongly in \(\mathcal L(E_1,E_3)\).
Proof of Proposition 3.5
The result follows from Proposition 3.3 and Lemma 3.6 by a tedious calculation that uses the CAR.
4 Wick Polynomials of the Particle Density
Recall that the Hermitian operators \(\rho (\Delta )\) were heuristically defined by (11) and the corresponding Wick polynomials \({:}\rho (\Delta _1)\cdots \rho (\Delta _n){:}\) were defined by (17). Then the heuristic formula (13) holds, see e.g. [21, Section C]. For the reader’s convenience, we will now present a heuristic proof of (13).
Define the operator-valued distribution \({:}\rho (x_1)\cdots \rho (x_n){:}\) so that the following formula holds:
for all \(\Delta _1,\dots ,\Delta _n\in \mathcal B_0(X)\). Then, by (17), we have \({:}\rho (x){:}=\rho (x)\) and
Here the generalized function \(\delta (x_1,x_2)\) is defined so that
It follows from the CAR that the operator-valued distributions \(A^+(x)\), \(A^-(x)\) satisfy the commutation relations:
Now (46) and (47) imply, by induction,
Note that formula (48) does not depend on the representation of the CAR, and just states that \({:}\rho (x_1)\cdots \rho (x_n){:}\) is the Wick (normal) ordering of the product \(\rho (x_1)\cdots \rho (x_n)\).
Formula (13) can be recurrently written as
The aim of this section is to derive a rigorous form of formula (49).
We start with presenting a rigorous form of the heuristic operator
where \(\Delta \in \mathcal B_0(X)\) and \(R\in \mathcal {L}(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\). The following result is inspired by Proposition 3.3 and formula (50).
Proposition 4.1
Let \(R\in \mathcal {L}(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\). For \(\Delta \in \mathcal {B}_0(X_1)\), we define
and for \(\Delta \in \mathcal {B}_0(X_2)\), we define
For \(\Delta \in \mathcal {B}_0(X)\), we define \(W(\Delta ,R):=W(\Delta \cap X_1,R)+W(\Delta \cap X_2,R)\). Then \(W(\Delta ,R)\in \mathcal {L}(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\) In formulas (51) and (52), the series converge strongly in \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\).
Before proving Proposition 4.1, let us present the main result of this section, which gives a rigorous form of formula (49).
Proposition 4.2
For any \(\Delta _1,\dots ,\Delta _{n}\in \mathcal B_0(X)\), \(n\ge 2\), we have
Let us now prove Propositions 4.1 and 4.2.
Proof of Proposition 4.1
We will present the proof only in the case \(\Delta \in \mathcal B_0(X_1)\). Below, we will denote by \(\textbf{1}_k\) the identity operator in \(\mathcal G^{\otimes k}\).
Step 1. Let \(R_{m,n+1}\in \mathcal {L}(\mathcal G^{\wedge (n+1)},\mathcal G^{\wedge m})\). For any \(\varphi ,\psi \in \mathcal G\) and \(g^{(n)}\in \mathcal G^{\wedge n}\), we have
We set \(e_n^{(1)}:=(e_n,0)\), \(e_n^{(2)}:=(0,e_n)\) (\(n\in \mathbb N\)), which is an orthonormal basis for \(\mathcal G\). Then, for any \(M,N\in \mathbb N\),
in \(\mathcal G^{\wedge (m+1)}\) as \(M, N\rightarrow \infty \). This implies that the series
converges strongly in \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\).
Step 2. Similarly to \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\), we define the topological vector space \(\mathcal {F}_{\textrm{fin}}(\mathcal G)\) that consists of all finite sequences \(g=(g^{(n)})_{n=0}^\infty \) with \(g^{(n)}\in \mathcal G^{\otimes n}\) (here \( \mathcal G^{\otimes 0}:=\mathbb C\)).
For \(\varphi \in \mathcal G\), we denote by \(l^+(\varphi )\) the left creation operator and by \(l^-(\varphi )\) the left annihilation operator in \(\mathcal {F}_{\textrm{fin}}(\mathcal G)\):
for \(g_1,\dots ,g_n\in \mathcal G\). Obviously, \(l^+(\varphi ), l^-(\varphi )\in \mathcal {L}( \mathcal {F}_{\textrm{fin}}(\mathcal G))\). Then, for each \(g^{(n)}\in \mathcal G^{\wedge n}\),
Let \(\varphi ^{(2)}\in \mathcal G^{\otimes 2}\). We also define an operator \(l^-(\varphi ^{(2)})\in \mathcal {L}(\mathcal {F}_{\textrm{fin}}(\mathcal G))\) by
In particular, if \(\varphi ^{(2)}=\varphi _1\otimes \varphi _2\) with \(\varphi _1, \varphi _2\in \mathcal G\), then \(l^-(\varphi _1\otimes \varphi _2)=l^-(\varphi _2)l^-(\varphi _1)\).
Consider \(R_{m,n-1}\in \mathcal {L}(\mathcal G^{\wedge (n-1)},\mathcal G^{\wedge m})\). Then, for \(\varphi ,\psi \in \mathcal G\), \(g^{(n)}\in \mathcal G^{\wedge n}\), we have
Hence, for \(M,N\in \mathbb N\),
in \(\mathcal G^{\wedge n}\) as \(M, N\rightarrow \infty \). This implies that the series
converges strongly in \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\).
Step 3. Let \(R_{m,n-1}\in \mathcal {L}(\mathcal G^{\wedge (n-1)}, \mathcal G^{\wedge m}).\) For \(\varphi ,\psi \in \mathcal G\) and \(g ^{(n)}\in \mathcal G^{\wedge n}\), we have
Hence, for any \(M, N\in \mathbb {N}\),
in \(\mathcal G^{\wedge (m+1)}\) as \(M, N\rightarrow \infty \). This implies that the series
converges strongly in \(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\).
Step 4. Recall that, for all \(x\in X_1\), we have \(K_1(x,\cdot )\in \mathcal H\). Therefore, for all \(x\in X_1\), the operators \(a_1^+\big (K_1(x,\cdot )\big )\) and \(a_1^-\big (K_1(x,\cdot )\big )\) are bounded in \(\mathcal{A}\mathcal{F}(\mathcal G)\) and have norm \(\Vert K_1(x,\cdot )\Vert _\mathcal H\).
Let \(R_{m,n+1}\in \mathcal {L}(\mathcal G^{\wedge (n+1)},\mathcal G^{\wedge m})\). We will now show that
exists as a Bochner integral with values in \( \mathcal {L}(\mathcal G^{\wedge n},\mathcal G^{\wedge (m-1)})\). For the definition and properties of a Bochner integral, see e.g. [5, Chapter 10, Section 3].
Lemma 4.3
The mappings
are strongly measurable.
Proof
Consider the product space \(\Delta \times X\) equipped with the corresponding \(\sigma \)-algebra. We will say that a function \(g:\Delta \times X\rightarrow \mathbb {C}\) is rectangle-simple if it is of the form
where \(c_i\in \mathbb C\), \(A_i\in \mathcal {B}(X_1)\), \(A_i\subset \Delta \), \(B_i\in \mathcal {B}(X)\), \(\sigma (B_i)<\infty \) (\(i=1,\dots ,n\)), and the sets \(A_i\times B_i\) are mutually disjoint. Since \(K_1(\cdot ,\cdot )\in L^2(\Delta \times X_1,\sigma ^{\otimes 2})\), there exists a sequence \((g_n)_{n=1}^\infty \) of rectangle-simple functions such that \(g_n(x,y)\rightarrow K_1(x,y)\) and \(|g_n(x,y)|\leqslant |K_1(x,y)|\) for \(\sigma ^{\otimes 2}\)-a.a. \((x,y)\in \Delta \times X\). In particular, \(g_n(\cdot ,\cdot )\rightarrow K_1(\cdot ,\cdot )\) in \(L^2(\Delta \times X,\sigma ^{\otimes 2})\). By (27), this implies that \(\big (a^+_1(g_n(x,\cdot ))\big )_{n=1}^\infty \) and \(\big (a^-_1(g_n(x,\cdot ))\big )_{n=1}^\infty \) are sequences of simple functions on \(\Delta \) with values in \(\mathcal {L}(\mathcal{A}\mathcal{F}(\mathcal G))\) such that, for \(\sigma \)-a.a. \(x\in \Delta \),
where convergence is in \(\mathcal {L}(\mathcal{A}\mathcal{F}(\mathcal G))\). This implies that the statement of the lemma holds. \(\square \)
Lemma 4.3 easily implies that the mapping
is strongly measurable. Furthermore, by (27),
Hence, by [5, Chapter 10, Theorem 3.1], the Bochner integral (54) exists.
We have,
where the series converges in \(\mathcal {L}(\mathcal{A}\mathcal{F}(\mathcal G))\). Note also that, for each \(N\in \mathbb {N}\),
Using the dominated convergence theorem for a Bochner integral (see e.g. [5, Chapter 10, Exercise 3.6]), we get
where the series converges in \(\mathcal {L}(\mathcal G^{\wedge n}, \mathcal G^{\wedge (m+1)})\).
To prove Proposition 4.2, we first need the following
Lemma 4.4
Let \(\Delta _1,\Delta _2\in \mathcal {B}_0(X)\) and \(R\in \mathcal {L}(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\). Then
Proof
By linearity, we can assume that \(\Delta _1\in \mathcal {B}_0(X_{i_{1}})\), \(\Delta _2\in \mathcal {B}_0(X_{i_{2}})\), where \(i_1,i_2\in \{1,2\}\). Consider, for example, the case where \(\Delta _1,\Delta _2\in \mathcal B_0(X_1)\). By (44), (51) and Lemma 3.6, we then write down the left-hand side of (55) as
and similarly we write down the right-hand side of (55). (The appearing series converge strongly in \(\mathcal L(\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G))\).) Through lengthy but rather straightforward calculations, one shows that both expression are equal. To give the reader a feeling how these calculations are carried out, we consider the following term appearing in (56):
where we used the CAR. The second sum on the right-hand side of (57) is equal to
On the other hand, another term appearing in (56) is:
Similarly, the second sum on the right-hand side of (59) is equal to
Thus, we see that, in formula (57), the ‘wrong’ term given by (60) cancels out, the first sum on the right-hand side of (57) and the first sum on the right-hand side of (59) come from \(W(\Delta _2,\rho (\Delta _1) R)\), and the first sum on the right-hand side of (58) comes from \(W(\Delta _1\cap \Delta _2, R)\).
We leave the rest of calculations to the interested reader. \(\square \)
Proof of Proposition 4.2
For \(\Delta _1,\Delta _2\in \mathcal B_0(X)\), we have by Lemma 4.4,
Hence, by (17),
i.e., formula (53) holds for \(n=2\). Assume formula (53) holds for n and let us prove it for \(n+1\). By Lemma 4.4 and (17),
which, by (17), implies that formula (53) holds for \(n+1\).
5 The Joint Spectral Measure of the Particle Density
Our aim now is to apply the results of Sect. 2.3 by setting \(\mathcal F=\mathcal{A}\mathcal{F}(\mathcal G)\), \(\mathcal D=\mathcal{A}\mathcal{F}_{\textrm{fin}}(\mathcal G)\), \(\mathcal A\) to be the commutative \(*\)-algebra generated by the particle density\((\rho (\Delta ))_{\Delta \in \mathcal B_0(X)}\), and \(\tau \) to be the vacuum state on \(\mathcal A\), i.e., \(\tau (a)=(a\Omega ,\Omega )_{\mathcal{A}\mathcal{F}(\mathcal G)}\) (\(a\in \mathcal A\)), where \(\Omega \) is the vacuum in \(\mathcal{A}\mathcal{F}(\mathcal G)\).
Recall the construction of the integral kernel \(\mathbb K(x,y)\) of the operator \(\mathbb K\) in Sect. 2.2.
Theorem 5.1
The operators \((\rho (\Delta ))_{\Delta \in \mathcal B_0(X)}\) have correlation measures \((\theta ^{(n)})_{n=1}^\infty \) respective the vacuum state state \(\tau \). Furthermore, the corresponding correlation functions are given by
We will prove Theorem 5.1 below, in Sect. 6, but let us now formulate and prove the main theorem of the paper.
Theorem 5.2
The operators \((\rho (\Delta ))_{\Delta \in \mathcal B_0(X)}\), together with their correlation measures \((\theta ^{(n)})_{n=1}^\infty \), satisfy the assumptions of Theorem 2.5. Thus, the following statements hold:
-
(i)
Let \(\mathfrak D:=\{a\Omega \mid a\in \mathcal A\}\) and let \(\mathfrak F\) denote the closure of \(\mathfrak D\) in \(\mathcal{A}\mathcal{F}(\mathcal G)\). Each operator \((\rho (\Delta ),\mathfrak D)\) is essentially self-adjoint in \(\mathfrak F\), i.e., the closure of \(\rho (\Delta )\), denoted by \(\widetilde{\rho }(\Delta )\), is a self-adjoint operator in \(\mathfrak F\).
-
(ii)
For any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\), the projection-valued measures (resolutions of the identity) of the operators \(\widetilde{\rho }(\Delta _1)\) and \(\widetilde{\rho }(\Delta _2)\) commute.
-
(iii)
There exist a unique point process \(\mu \) in X and a unique unitary operator\(U:\mathfrak F\rightarrow L^2(\Gamma _X,\mu )\) satisfying \(U\Omega =1\) and (20). In particular, (21) holds.
-
(iv)
The correlations functions of the point process \(\mu \) are given by (61).
Proof
We first prove the following
Lemma 5.3
For any \(\Delta \in \mathcal B_0(X)\),
and for any \(\Delta _1,\dots ,\Delta _n\in \mathcal B_0(X)\) (\(n\ge 2\)), we have
where, for a cycle \(\psi =(l_1 l_2 \cdots l_k)\) in a permutation \(\xi \in S_{n}\), we have
Proof
Recall that that, if \(S, T\in \mathcal S_2(\mathcal H)\) with integral kernels S(x, y) and T(x, y), respectively, then
Hence, formula (62) holds by statement (ii) of Lemma 2.1 and the construction of the integral kernel \(\mathbb K(x,y)\).
Next, we note that, for any \(\Delta _1,\Delta _2\in \mathcal B_0(X)\), we have \(P_{\Delta _1}\mathbb K P_{\Delta _2}\in \mathcal S_2(\mathcal H)\), hence the operator appearing in (65) is indeed of trace class. Formula (63) obviously holds with \(\mathbb T_\psi \) given, for \(k=1\) by (64), and for \(k\ge 2\),
\(\square \)
Let us prove that condition (LB1) of Theorem 2.5 is satisfied. For \(l\ge 2\), we have
where \(\Vert \cdot \Vert _1\) denotes the norm in \(\mathcal S_1(\mathcal H)\). Theorem 5.1, Lemma 5.3 and formula (66) imply that condition (LB1) is satisfied with
In the case where \(\Delta \) is a subset of either \(X_1\) or \(X_2\), we can find a finer estimate of \(\theta ^{(n)}(\Delta ^n)\). Indeed, assume, for example that \(\Delta \subset X_1\). Then, for any \(x,y\in \Delta \), we have, by (16),
and so, for any \(x_1,\dots ,x_n\in \Delta \),
which implies
Hence, by Theorem 5.1, condition (LB1) is satisfied with
Similarly, for \(\Delta \in \mathcal B_0(X_2)\), (LB1) is satisfied with
It follows from the proof of Theorem 2.5 in [20] that, when checking condition (LB2), it is sufficient to assume that all sets in the sequence \(\{\Delta _{l}\}_{l\in \mathbb {N}}\) are subsets of either \(X_1\) or \(X_2\). But then (LB2) is an immediate consequence of formulas (67) and (68). \(\square \)
6 Proof of Theorem 5.1
We will now prove Theorem 5.1. Our strategy here is to prove the existence of correlation measures and that of correlation functions at the same time.
We first state, for any \(x_1,\dots ,x_n\in X\),
Indeed, if all points \(x_1,\dots ,x_n\) belong to the same part, \(X_1\) or \(X_2\), then the matrix \(\big [\mathbb K(x_i,x_j)\big ]_{i,j=1,\dots ,n}\) is Hermitian, hence its determinant is \(\ge 0\). Otherwise, without loss of generality, we may assume that for some m with \(1<m<n\), we have \(x_1,\dots ,x_m\in X_1\) and \(x_{m+1},\dots ,x_n\in X_2\). But then formula (69) follows from [22, Proposition 1.4].
Next, it is easy to see that, if among points \(x_1,\dots ,x_n\), at least two points coincide, then \(\det \big [\mathbb K(x_i,x_j)\big ]_{i,j=1,\dots ,n}=0\). Therefore, the measure
is concentrated on \(X^{(n)}\).
Hence, to prove Theorem 5.1, it suffices to show that, for any \(\Delta _1,\dots ,\Delta _m\in \mathcal {B}_0(X_1)\), \(\Delta _{m+1},\dots ,\Delta _{m+n}\in \mathcal {B}_0(X_2)\), \(m,n\in \mathbb N_0\), \(m+n\ge 1\), we have
We divide the proof of this formula into several steps.
Step 1. To shorten our notations, we denote, for \(i,j\in \mathbb N\) and \(\Delta \in \mathcal B_0(X_1)\),
for \(\Delta \in \mathcal {B}_{0}(X_2)\),
and
Then, by Proposition 3.3, for \(\Delta \in \mathcal {B}_0(X_1)\),
and for \(\Delta \in \mathcal {B}_0(X_2)\),
We define an ordered set
(the elements of \(\mathfrak E\) being listed in (71) in the increasing order). By Proposition 4.2,
Step 2. We remind the reader that \(\tau \) is a quasi-free state. As easily seen, the following lemma holds.
Lemma 6.1
Let \(n\in \mathbb {N}\) and let \(g_1,\dots ,g_{2n}\in \mathcal G\). Let \(\Diamond _1,\dots ,\Diamond _{2n}\in \{+,-\}\) and assume that the number of plusses among \(\Diamond _1,\dots ,\Diamond _{2n}\) is the same as the number of minuses. Then,
where the summation is over all partitions \(\nu =\big \{\{i_1,j_1\},\dots ,\{i_{n},j_{n}\}\big \}\) of \(\{1,\dots ,2n\}\) with \(i_k<j_k\) and such that \(\Diamond _{i_k}=-\), \(\Diamond _{j_k}=+\) (\(k=1,\dots ,n\)).
We define four (ordered) subsets of \(\mathfrak E\) as follows:
Denote by \(\mathfrak R\) the collection of all partitions \(\nu \) of \(\mathfrak E\) into n two-point sets such that, if \(\{i,j\}\in \nu \) with \(i<j\), then one of the following four statements holds: (i) \(i\in \mathfrak A\) and \(j\in \mathfrak B\); (ii) \(i\in \mathfrak B\) and \(j\in \mathfrak C\); (iii) \(i\in \mathfrak C\) and \(j\in \mathfrak D\); (iv) \(i\in \mathfrak A\) and \(j\in \mathfrak D\).
By Lemma 6.1 and in view of the definition of \(A^{\Diamond }_i\) and \(B^{\Diamond }_i\) \((\Diamond \in \{+, -\})\), we continue (72) as follows:
Here, for \(u\in \mathfrak E\), we denote \(\Diamond _u(\nu ):=-\) if \(\{u,v\}\in \nu \) and \( u< v\), and \(\Diamond _u(\nu ):=+\) if \(\{v,u\}\in \nu \) and \(u<v\).
Step 3. Let \(I:\mathfrak E\rightarrow \mathfrak E\) be the bijective map that is defined as follows: I acts as the identity on \(\mathfrak A\) and \(\mathfrak D\) and swaps the elements of \(\mathfrak B\) and \(\mathfrak C\), i.e., \(I(i)=i'\) for all \(i\in \mathfrak B\) and \(I(i')=i\) for all \(i'\in \mathfrak C\).
Denote by \(\mathfrak S\) the collection of all partitions \(\nu \) of \(\mathfrak E\) into n two-point sets such that, for each \(\{i,j\}\in \nu \), we have \(i\in \mathfrak A\cup \mathfrak B\) and \(j\in \mathfrak C\cup \mathfrak D\). As easily seen, the map I induces a bijection (still denoted by I) of \(\mathfrak R\) onto \(\mathfrak S\).
Lemma 6.2
(i) For \(\nu \in \mathfrak R\), denote by \(k_1(\nu )\) the number of all \(\{i,j\}\in \nu \) such that \(i\in \mathfrak A\), \(j\in \mathfrak B\), and denote \(k_2(\nu )\) the number of all \(\{i,j\}\in \nu \) such that \(i\in \mathfrak C\), \(j\in \mathfrak D\). Then \(k_1(\nu )=k_2(\nu )\) and we denote \( \mathfrak l(\nu ):=k_1(\nu )=k_2(\nu )\).
(ii) We have, for each \(\nu \in \mathfrak R\)
Proof of Proposition 4.2
Part (i) is obvious, so we only prove part (ii). For \(\nu \in \mathfrak R\), under the map I, the change in the number of crossings happens in the following three cases:
-
(a)
Let \(\{i_1,j_1\}, \{i_2,j_2\}\in \nu \), \(i_1,i_2\in \mathfrak A\), \(j_1, j_2\in \mathfrak B\). Then \(\{I(i_1),I(j_1)\}\), \(\{I(i_2),I(j_2)\}\) have a crossing if and only if \(\{i_1, j_1\}\), \(\{i_2,j_2\}\) do not have a crossing;
-
(b)
Let \(\{i_1,j_1\}, \{i_2,j_2\}\in \nu \), \(i_1,i_2\in \mathfrak C\), \(j_1, j_2\in \mathfrak D\). Then \(\{I(i_1),I(j_1)\}\), \(\{I(i_2),I(j_2)\}\) have a crossing if and only if \(\{i_1, j_1\}\), \(\{i_2,j_2\}\) do not have a crossing;
-
(c)
Let \(\{i_1,j_1\}, \{i_2,j_2\}\in \nu \), \(i_1\in \mathfrak A\), \(j_1\in \mathfrak B\), \(i_2\in \mathfrak C\), \(j_2\in \mathfrak D\). Then \(\{i_1,j_1\}\), \(\{i_2,j_2\}\) do not have a crossing while \(\{I(i_1),I(j_1)\}\) and \(\{I(i_2),I(j_2)\}\) have a crossing.
Let \(N_a(\nu )\), \(N_b(\nu )\), and \(N_c(\nu )\) denote the number of \(\big \{\{i_1,j_1\}, \{i_2,j_2\}\big \}\subset \nu \) as in (a), (b), and (c), respectively. Since for any \(i,j\in \mathbb N_0\), \((-1)^{i-j}=(-1)^{i+j}\), we therefore have:
By part (i), \(N_a(\nu )=N_b(\nu )\), while \(N_c(\nu )=\mathfrak l(\nu )^2\). Hence, (76) implies (75).\(\square \)
Step 4. We will now identify the set of partitions \(\mathfrak R\) with the symmetric group \( S_{m+n}\). We define \(\mathfrak m:\mathfrak C\cup \mathfrak D\rightarrow \mathfrak A\cup \mathfrak B\) by \(\mathfrak { m}i'=i\) for \(i'\in \mathfrak C\cup \mathfrak D\).
We define \(\mathfrak I: \mathfrak R\rightarrow S_{m+n}\) as follows: for \(\nu \in \mathfrak R\) let \(\xi =\mathfrak I(\nu )\) be given by:
-
for each \(\{u,v\} \in \nu \) such that \(u\in \mathfrak {A}\) and \(v\in \mathfrak D\), \(\xi (u)=\mathfrak { m} v\),
-
for each \(\{u,v\}\in \nu \) such that \(u\in \mathfrak {A}\) and \(v\in \mathfrak B\), \(\xi (u)= v\),
-
for each \(\{u,v\}\in \nu \) such that \(u\in \mathfrak {B}\) and \(v\in \mathfrak C\), \(\xi (\mathfrak { m}v)= u\),
-
for each \(\{u,v\}\in \nu \) such that \(u\in \mathfrak {C}\) and \(v\in \mathfrak D\), \(\xi (\mathfrak { m}u)=\mathfrak {m} v\).
As easily seen, for each \(\nu \in \mathfrak R\),
Then formula (75) implies
Hence, we continue (74) as follows:
Step 5. We define mappings \(\mathfrak r_1:\mathfrak A\cup \mathfrak B\rightarrow \mathfrak A\cup \mathfrak C\) and \(\mathfrak r_2:\mathfrak A\cup \mathfrak B\rightarrow \mathfrak B\cup \mathfrak D\) as follows:
-
for \(u\in \mathfrak A\), \(\mathfrak r_1(u):=u\) and \(\mathfrak r_2(u):=u'\);
-
for \(u\in \mathfrak B\), \(\mathfrak r_1(u):=u'\) and \(\mathfrak r_2(u):=u\).
Then, for each \(\xi \in S_{m+n}\), we have \(\{u,v\}\in \mathfrak I^{-1}(\xi )\) if and only if, for some \(i\in \{1,\dots ,m+n\}\), we have \(\{u,v\}=\{\mathfrak r_1(u), \mathfrak r_2(\xi (u))\}\). Hence, by (72), (74), and (77),
Step 6. Next, we prove
Lemma 6.3
Let \(\xi \in S_{m+n}\) and let \((l_1\,l_2\cdots l_k)\) be a cycle in \(\xi \). Then
Here, for \(u,v\in \{1,2,\dots ,m+n\}\),
Proof
Let us first consider the case where \(l_1,l_2,\dots ,l_k\in \{1,\dots ,m\}\). Then the left-hand side of (79) becomes
We have
Next,
Continuing by analogy, we conclude:
which implies
We similarly treat the case where \(l_1,l_2,\dots ,l_k\in \{m+1,\dots ,m+n\}\).
Finally we consider the case where
Without loss of generality, we may assume that \(l_1\in \{1,\dots ,m\}\) and \(l_k\in \{m+1,\dots ,m+n\}.\) To simplify the notation, we will additionally assume that, for some \(\alpha \in \{1,\dots ,k-1\}\),
The interested reader can easily extend our arguments to the more general case.
We consider separately three cases:
Case 1: \(\alpha =k-1\). Then the left-hand side of (79) becomes
Analogously to the above calculations, we get
Then
and
Case 2: \(\alpha =k-2\). Then the left-hand side of (79) becomes
Case 3: \(\alpha \leqslant k-3\). Then the left-hand side of (79) becomes
Similarly to Case 2, we obtain:
Then
Finally,
Step 7. For a given cycle \(\theta = (l_1 l_2\cdots l_k)\) in a permutation \(\xi \in S_{m+n}\), we denote by \(\widetilde{\mathbb T}_\theta \) the value given by (the right hand-side of) formula (79). Denote by \(\mathfrak {t}(\theta )\) the number of \(i\in \{1,\dots ,k\}\) such that \(l_i\in \{m+1,\dots ,m+n\}\) but \(l_{i+1}\in \{1,\dots ,m\}\), where \(l_{k+1}:=l_1\). Then, by (64), (65), and (80),
By Lemma 6.2 (i),
Thus, by (78), (81), (82) and Lemma 6.3,
Data Availability
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
References
Alshehri, M.G.A., Lytvynov, E.: Hafnian point processes and quasi-free states on the CCR algebra. Infin. Dimens. Anal. Quantum Probab. Relat. Top. 25, no. 1, Paper No. 2250002, 25 pp (2022)
Araki, H., Wyss, W.: Representations of canonical anticommutation relation. Helv. Phys. Acta 37, 136–159 (1964)
Azizov, T.Ya., Iokhvidov, I.S.: Linear Operators in Spaces with an Indefinite Metric. Wiley, Chichester (1989)
Berezansky, Y.M., Kondratiev, Y.G.: Spectral Methods in Infinite-Dimensional Analysis, vol. 2. Kluwer Academic Publishers, Dordrecht (1995)
Berezansky, Y.M., Sheftel, Z.G., Us, G.F.: Functional Analysis, vol. I. Birkhäuser Verlag, Basel (1996)
Borodin, A.: Determinantal point processes. In: The Oxford Handbook of Random Matrix Theory, pp. 231–249. Oxford University Press, Oxford (2011)
Borodin, A., Okounkov, A., Olshanski, G.: Asymptotics of Plancherel measures for symmetric groups. J. Am. Math. Soc. 13, 481–515 (2000)
Borodin, A., Olshanski, G.: Point processes and the infinite symmetric group. Math. Res. Lett. 5, 799–816 (1998)
Borodin, A., Olshanski, G.: Distributions on partitions, point processes, and the hypergeometric kernel. Commun. Math. Phys. 211, 335–358 (2000)
Borodin, A., Olshanski, G.: Harmonic analysis on the infinite-dimensional unitary group and determinantal point processes. Ann. Math. 2(161), 1319–1422 (2005)
Borodin, A., Olshanski, G.: Random partitions and the gamma kernel. Adv. Math. 194, 141–202 (2005)
Bratteli, O., Robinson, D.W.: Operator Algebras and Quantum Statistical Mechanics, vol. 2. Equilibrium states. Models in Quantum Statistical Mechanics, 2nd edn. Springer, Berlin (1997)
Daley, D.J., Vere-Jones, D.: An Introduction to the Theory of Point Processes. Springer, New York (1988)
Georgii, H.-O., Yoo, H.J.: Conditional intensity and Gibbsianness of determinantal point processes. J. Stat. Phys. 118, 55–84 (2005)
Kallenberg, O.: Random Measures. Akad.-Verl, Berlin (1983)
Koshida, S.: Pfaffian point processes from free fermion algebras: perfectness and conditional measures. SIGMA Symmetry Integrability Geom. Methods Appl. 17, Paper No. 008, 35 pp (2021)
Lenard, A.: Correlation functions and the uniqueness of the state in classical statistical mechanics. Commun. Math. Phys. 30, 35–44 (1973)
Lytvynov, E.: Fermion and boson random point processes as particle distributions of infinite free Fermi and Bose gases of finite density. Rev. Math. Phys. 14, 1073–1098 (2002)
Lytvynov, E.: Determinantal point processes with \(J\)-Hermitian correlation kernels. Ann. Probab. 41, 2513–2543 (2013)
Lytvynov, E., Mei, L.: On the correlation measure of a family of commuting Hermitian operators with applications to particle densities of the quasi-free representations of the CAR and CCR. J. Funct. Anal. 245, 62–88 (2007)
Menikoff, R., Sharp, D.H.: Representations of a local current algebra: their dynamical determination. J. Math. Phys. 16, 2341–2352 (1975)
Olshanski, G.: Point processes and the infinite symmetric group. Part V: analysis of the matrix Whittaker kernel. arXiv preprint, http://arxiv.org/abs/math/9810014 (1998)
Macchi, O.: The coincidence approach to stochastic point processes. Adv. Appl. Probab. 7, 83–122 (1975)
Schaefer, H.H., Wolff, M.P.: Topological Vector Spaces, 2nd edn. Springer, New York (1999)
Simon, B.: Trace Ideals and Their Applications. Cambridge University Press, Cambridge (1979)
Soshnikov, A.: Determinantal random point fields. Russ. Math. Surv. 55, 923–975 (2000)
Tamura, H., Ito, K.R.: A canonical ensemble approach to the fermion/boson random point processes and its applications. Commun. Math. Phys. 263, 353–380 (2006)
Acknowledgements
EL is grateful to Grigori Olshanski for many useful discussions on determinantal point processes with J-Hermitian correlation kernel. The authors are grateful to the anonymous reviewers for many useful comments and suggestions.
Funding
The authors did not receive support from any organization for the submitted work.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Additional information
Communicated by Y. Kawahigashi.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
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
Alshehri, M.G.A., Lytvynov, E. Particle-Hole Transformation in the Continuum and Determinantal Point Processes. Commun. Math. Phys. 403, 627–659 (2023). https://doi.org/10.1007/s00220-023-04803-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00220-023-04803-9