betaFeedback

WorksMathematics

Decay of Correlations for Multiplicative Functionals of Determinantal Point Processes

Abstract

The main result of this paper is an upper bound on the correlations of multiplicative functionals of determinantal point processes. In particular, for the sine process, the correlations of a multiplicative functional associated with a rapidly decaying function decay quadratically. It follows that the spectral density of such functionals belongs to the Sobolev space H1H_1.

Introduction and statement of the main result

Multiplicative functionals of determinantal point processes

Let (E,d)(E,d) be a complete separable metric space, let μ\mu be a σ\sigma-finite measure on EE, and let Π:E×E→C\Pi:E\times E\to\mathbb{C} be a Borel function of two variables that is bounded on compact sets, and such that the operator Πf(x)=∫Π(x,y)f(y)dμ(y)\Pi f(x)=\int\Pi(x,y)f(y)d\mu(y), initially defined on bounded Borel functions with compact support, extends to an orthogonal projection acting on L2(E,μ)L_2(E,\mu) (all Hilbert spaces considered in this paper are complex). By the Macchi–Soshnikov–Shirai–Takahashi theorem [10, 15, 14], the projection Π\Pi determines a determinantal measure on the space Conf⁡(E)\operatorname{Conf}(E) of locally finite point configurations on EE.

If ff is a bounded Borel function with compact support on EE, and Ψ1+f\Psi_{1+f} is the multiplicative functional associated with ff on the space of configurations Conf⁡(E)\operatorname{Conf}(E) defined by the formula

Ψ1+f(X)=∏x∈X(1+f(x)),\Psi_{1+f}(X)=\prod_{x\in X}(1+f(x)),

then (cf. [5, 4]) the determinantal measure PΠP_\Pi is uniquely determined by the condition

EPΠΨ1+f=det⁡(1+ΠfΠ)=det⁡(1+fΠf).\mathbb{E}_{P_\Pi}\Psi_{1+f}=\det(1+\Pi f\Pi)=\det(1+\sqrt{f}\Pi\sqrt{f}).

Examples of measures that would subsequently be called determinantal already appear in Andréiéf’s work [1].

The notation fΠf\sqrt{f}\Pi\sqrt{f} is well-defined for both real- and complex-valued functions ff. Either square root of ff may be chosen; the only requirement is that the same square root be used on the left and on the right hand side. Now set

Ψ~1+f=Ψ1+f⋅exp⁡(−tr⁡(ΠfΠ))=Ψ1+fexp⁡(−tr⁡(fΠf)).\widetilde{\Psi}_{1+f}=\Psi_{1+f}\cdot\exp\left(-\operatorname{tr}(\Pi f\Pi)\right)=\Psi_{1+f}\exp\left(-\operatorname{tr}(\sqrt{f}\Pi\sqrt{f})\right).

We next introduce the normalized multiplicative functionals. Let HSHS to be the space of Hilbert–Schmidt operators acting on L2(E,μ)L_2(E,\mu). Consider the space HS(Π)HS(\Pi) of Borel functions ff for which the operator fΠf\Pi is Hilbert–Schmidt, and equip this space with the Euclidean structure defined by ∥f∥HS(Π)=∥fΠ∥HS\lVert f\rVert_{HS(\Pi)}=\lVert f\Pi\rVert_{HS}; by definition, we have

⟨f,h⟩HS(Π)=tr⁡(fΠh‾)=⟨fΠ,hΠ⟩HS.\langle f,h\rangle_{HS(\Pi)}=\operatorname{tr}(f\Pi\overline{h})=\langle f\Pi,h\Pi\rangle_{HS}.

For the Hilbert space obtained by completing the space of bounded Borel functions with compact support with respect to the Euclidean structure <,>HS(Π)<,>_{HS(\Pi)}, we retain, with a slight abuse of notation, the symbol HS(Π)HS(\Pi).

Recall that the regularized Hilbert–Carleman determinant [8] is defined for a finite-rank bounded operator AA by the formula

det⁡2(1+A)=det⁡(1+A)exp⁡(−tr⁡A)\det_2(1+A)=\det(1+A)\exp(-\operatorname{tr}A)

and is then extended by continuity to the entire space of Hilbert–Schmidt operators.

Proposition 1. The map f→Ψ~1+ff\to\widetilde{\Psi}_{1+f} extends to a continuous map HS(Π)→L1(Conf⁡(E),PΠ)HS(\Pi)\to L_1(\operatorname{Conf}(E),\mathbb{P}_\Pi), and for every f∈HS(Π)f\in HS(\Pi), we have:

EPΠΨ~1+f=det⁡2(1+ΠfΠ)=det⁡2(1+fΠ).\mathbb{E}_{\mathbb{P}_\Pi}\widetilde{\Psi}_{1+f}=\det_2(1+\Pi f\Pi)=\det_2(1+f\Pi).

A general theorem on decay of correlation

The main result of this paper is an upper bound on the covariance of two normalized multiplicative functionals.

We use L2(Π)L_2(\Pi) to denote the space of square-integrable functions with respect to the measure Π(x,x)dμ(x)\Pi(x,x)d\mu(x).

We will also need another embedding of the space of bounded Borel functions with compact support into the space of Hilbert–Schmidt operators, namely f→ΠfΠf\to\Pi f\Pi. Define the Hilbert space H(2)(Π)H^{(2)}(\Pi) as the completion of the space of bounded Borel functions with compact support in the norm

∥f∥H(2)(Π)2=∥ΠfΠ∥HS2;\lVert f\rVert_{H^{(2)}(\Pi)}^2=\lVert\Pi f\Pi\rVert_{HS}^2;

defined by the inner product

⟨f,h⟩H(2)(Π)=tr⁡Πfh‾Π=⟨ΠfΠ,ΠhΠ⟩HS.\langle f,h\rangle_{H^{(2)}(\Pi)}=\operatorname{tr}\Pi f\overline{h}\Pi=\langle\Pi f\Pi,\Pi h\Pi\rangle_{HS}.

By definition, we have

⟨f,h⟩H(2)(Π)=∬E×Ef(x)∣Π(x,y)∣2h(y)‾ dμ(x)dμ(y)=∥fΠh‾∥2.\langle f,h\rangle_{H^{(2)}(\Pi)}=\iint_{E\times E}f(x)|\Pi(x,y)|^2\overline{h(y)}\,d\mu(x)d\mu(y)=\|\sqrt{f}\Pi\sqrt{\overline{h}}\|^2.

Remark 1. The same determinantal measure can be represented by different kernels, but the space H(2)(Π)H^{(2)}(\Pi) is defined in terms of the determinantal measure itself, since the measure ∣Π(x,y)∣2dμ(x)dμ(y)|\Pi(x,y)|^2d\mu(x)d\mu(y) is uniquely determined by the measure PΠP_\Pi.

For r1>0,r2>0,r3>0r_1>0,r_2>0,r_3>0, let EΠ(r1,r2,r3)\mathcal{E}_\Pi(r_1,r_2,r_3) be the family of functions f∈L∞(E,μ)f\in L_\infty(E,\mu) defined by the formula

EΠ(r1,r2,r3)=={f∈L∞(E,μ):∥f∥L∞≤r1,∥fΠ∥HS≤r2,∣det⁡2(1+ΠfΠ)∣≥r3}.\begin{aligned} \mathcal{E}_\Pi(r_1,r_2,r_3)= \\ &=\{f\in L_\infty(E,\mu):\|f\|_{L_\infty}\le r_1,\|f\Pi\|_{HS}\le r_2,|\det_2(1+\Pi f\Pi)|\ge r_3\}. \end{aligned}

For δ>0\delta>0, let NΠ(δ)\mathcal{N}_\Pi(\delta) be the family of pairs of functions f∈L∞(E,μ)f\in L_\infty(E,\mu) and h∈L∞(E,μ)h\in L_\infty(E,\mu), defined by the formula

NΠ(δ)={(f,h):f∈L∞(E,μ),h∈L∞(E,μ),∥ΠfhΠ∥HS≤δ,∥ΠfΠhΠ∥HS≤δ}.\mathcal{N}_\Pi(\delta)=\{(f,h):f\in L_\infty(E,\mu),h\in L_\infty(E,\mu), \|\Pi fh\Pi\|_{HS}\le\delta,\|\Pi f\Pi h\Pi\|_{HS}\le\delta\}.

Theorem 1. For every r1>0,r2>0,r3>0r_1>0,r_2>0,r_3>0, there exist δ>0\delta>0 and A>0A>0 such that the following holds. For any locally trace-class projection Π\Pi and the corresponding determinantal point process PΠP_\Pi, any f,h∈EΠ(r1,r2,r3)f,h\in\mathcal{E}_\Pi(r_1,r_2,r_3) and (f,h)∈NΠ(δ)(f,h)\in\mathcal{N}_\Pi(\delta), we have

∣EPΠΨ~1+f⋅Ψ~1+h−EPΠΨ~1+f⋅EPΠΨ~1+h∣≤A(⟨∣f∣,∣h∣⟩L2(Π)+⟨∣f∣,∣h∣⟩H(2)(Π)).\left|\mathbb{E}_{P_\Pi}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}-\mathbb{E}_{P_\Pi}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{P_\Pi}\widetilde{\Psi}_{1+h}\right| \le A\left(\langle|f|,|h|\rangle_{L_2(\Pi)}+\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}\right).

1.3. Correlations of multiplicative functionals of stationary determinantal point processes. Theorem 1 yields estimates on the rate of mixing for stationary determinantal point processes. Our convention for the Fourier transform is as follows:

f^(λ)=∫Re−iλtf(t) dt;\widehat{f}(\lambda)=\int_{\mathbb{R}}e^{-i\lambda t}f(t)\,dt;
f(t)=12π∫Rf^(λ)eiλt dλ.f(t)=\frac{1}{2\pi}\int_{\mathbb{R}}\widehat{f}(\lambda)e^{i\lambda t}\,d\lambda.

As usual, we denote the inverse Fourier transform of ff by fˇ\check{f}.

We now consider stationary determinantal point processes on R\mathbb{R} corresponding to Fourier projections onto bounded Borel sets. Let B⊂RB \subset\mathbb{R} be a bounded Borel set of positive measure, and let the projection ΠB\Pi^{B} be defined by the formula

ΠBf=(1Bf)∨.\Pi^{B}f=(1_{B}f)^{\vee}.

By definition, the kernel of the operator ΠB\Pi^{B} is given by the formula

ΠB(x,y)=1ˇB(x−y).\Pi^{B}(x,y)=\check{1}_{B}(x-y).

In particular, for B=[−π,π]B=[-\pi,\pi], we obtain the sine process, a determinantal point process with kernel

S(x,y)=sin⁡π(x−y)π(x−y).S(x,y)=\frac{\sin\pi(x-y)}{\pi(x-y)}.

the kernel of the projection onto the Paley–Wiener space. Let gsg_s denote the action of the translation group on the real line, gs(t)=t+sg_s(t)=t+s; by a slight abuse of notation, we use the same symbol gsg_s for the induced action of the translation group on Conf⁡(R)\operatorname{Conf}(\mathbb{R}). Recall that the tail σ\sigma-algebra of a determinantal process associated with an orthogonal projection is trivial (see [11] in the discrete case and [7] and [12] in the general case). In particular, the determinantal process PΠB\mathbb{P}_{\Pi^{B}} also has a trivial tail σ\sigma-algebra. Consequently, the translation group gsg_s induces a Kolmogorov flow with respect to the determinantal measure PΠB\mathbb{P}_{\Pi^{B}} and, consequently, has Lebesgue spectrum of countable multiplicity. Our aim is to give explicit regularity estimates for the spectral measures of multiplicative functionals with respect to the determinantal measure PΠB\mathbb{P}_{\Pi^{B}}.

To a bounded Borel set BB assign a positive increasing function ΦB\Phi_{B} by setting

ΦB(t)=(sup⁡r:∣r∣≥∣t∣∣1ˇB(r)∣2)−1.\Phi_{B}(t)=\left(\sup_{r:\lvert r\rvert\geq\lvert t\rvert}\left\lvert\check{1}_{B}(r)\right\rvert^{2}\right)^{-1}.

Note that positivity of ΦB\Phi_{B} follows from the Heisenberg uncertainty principle (the Fourier transform of the indicator function 1B(⋅)1_{B}(\cdot) is not compactly supported, and hence ΦB(t)\Phi_{B}(t) never takes the value +∞+\infty). Our key assumption is

∫0+∞dtΦB(t)<+∞.(1)\int_{0}^{+\infty}\frac{dt}{\Phi_{B}(t)}<+\infty. \tag*{(1)}

Introduce the class KB\mathcal{K}_{B} of bounded Borel functions ff on R\mathbb{R} such that

sup⁡t∈RΦB(t)∣f(t)∣<+∞.\sup_{t\in\mathbb{R}}\Phi_{B}(t)\lvert f(t)\rvert<+\infty.

and set ∥f∥KB=sup⁡t∈RΦB(t)∣f(t)∣\lVert f\rVert_{\mathcal{K}_B}=\sup_{t\in\mathbb{R}}\Phi_B(t)|f(t)|.

Corollary 1. Let B⊂RB\subset\mathbb{R} be a bounded Borel set satisfying condition (1). For every C>0C>0 and c>0c>0, there exists A>0A>0 such that for any Borel functions fi∈KBf_i\in\mathcal{K}_B, i=1,2i=1,2, such that

∣det⁡2(1+ΠBfiΠB)∣≥c,∥fi∥KB≤C,(2)\left|\det_2\left(1+\Pi_B f_i\Pi_B\right)\right|\ge c,\qquad\lVert f_i\rVert_{\mathcal{K}_B}\le C, \tag*{(2)}

the following inequality holds:

sup⁡t∈RΦB(t/4)∣EΠB(Ψ1+f1∘gt⋅Ψ1+f2)−EΠBΨ1+f1EΠBΨ1+f2∣≤A.\sup_{t\in\mathbb{R}}\Phi_B(t/4)\left|\mathbb{E}_{\Pi_B}\left(\Psi_{1+f_1}\circ g_t\cdot\Psi_{1+f_2}\right)-\mathbb{E}_{\Pi_B}\Psi_{1+f_1}\mathbb{E}_{\Pi_B}\Psi_{1+f_2}\right|\le A.

In the particular case of the sine process, the expression

(1+t2)Φ[−π,π](t)(1+t^2)\Phi_{[-\pi,\pi]}(t)

is bounded away from zero and bounded above by definition, and we obtain the following.

Corollary 2. For every C>0C>0 and c>0c>0, there exists A>0A>0 such that if the Borel functions fi:R→Cf_i:\mathbb{R}\to\mathbb{C}, i=1,2i=1,2, satisfy the inequalities

∣det⁡2(1+SfiS)∣≥c,sup⁡t∈R(1+t2)∣fi(t)∣≤C,i=1,2,(3)\left|\det_2(1+Sf_iS)\right|\ge c,\qquad\sup_{t\in\mathbb{R}}(1+t^2)|f_i(t)|\le C,\qquad i=1,2, \tag*{(3)}

then the following inequality also holds:

sup⁡t∈R(1+t2)∣EPS(Ψ1+f1∘gt⋅Ψ1+f2)−EPSΨ1+f1EPSΨ1+f2∣≤A.\sup_{t\in\mathbb{R}}(1+t^2)\left|\mathbb{E}_{P_S}\left(\Psi_{1+f_1}\circ g_t\cdot\Psi_{1+f_2}\right)-\mathbb{E}_{P_S}\Psi_{1+f_1}\mathbb{E}_{P_S}\Psi_{1+f_2}\right|\le A.

Thus, the multiplicative functional corresponding to a function ff satisfying conditions (3) has, under the sine process, continuous spectral density with square-integrable derivative; in other words, the spectral density belongs to the Sobolev space H1H^1.

The proofs of Corollaries 1 and 2 are given in § 4.

Outline of the proof of Theorem 1

It will be more convenient for us to work with operator norms rather than determinants and to estimate the multiplicative rather than the additive correlation. We therefore reformulate Theorem 1 as follows. For r>0r>0, let CΠ(r)\mathcal{C}_{\Pi}(r) be the family of functions f∈L∞(E,μ)f\in L^\infty(E,\mu), defined by

CΠ(r)={f∈L∞(E,μ):∥f∥L∞+∥fΠ∥HS+∥(1+ΠfΠ)−1∥<r}.\mathcal{C}_{\Pi}(r)=\left\{f\in L^\infty(E,\mu):\lVert f\rVert_{L^\infty}+\lVert f\Pi\rVert_{HS}+\left\lVert(1+\Pi f\Pi)^{-1}\right\rVert<r\right\}.

Proposition 2. For every r>0r>0, there exist δ>0\delta>0 and A>0A>0 such that the following holds. For any locally trace-class projection Π\Pi and

the corresponding determinantal point process PΠ\mathbb{P}_{\Pi}, any f,h∈GΠ(r)f,h \in\mathcal{G}_{\Pi}(r) and (f,h)∈NΠ(δ)(f,h) \in\mathcal{N}_{\Pi}(\delta), we have

exp⁡(−A(⟨∣f∣,∣h∣⟩H(2)(Π)+⟨∣f∣,∣h∣⟩L2(Π)))≤∣EPΠΨ~1+f⋅Ψ~1+hEPΠΨ~1+f⋅EPΠΨ~1+h∣≤exp⁡(A(⟨∣f∣,∣h∣⟩H(2)(Π)+⟨∣f∣,∣h∣⟩L2(Π))).(4)\exp\left(-A\left(\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}+\langle|f|,|h|\rangle_{L_2(\Pi)}\right)\right) \leq \left|\frac{\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}}{\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+h}}\right| \leq \exp\left(A\left(\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}+\langle|f|,|h|\rangle_{L_2(\Pi)}\right)\right). \tag*{(4)}

Theorem 1 immediately follows from Proposition 2 and the following simple observation.

Proposition 3. For any r1>0,r2>0,r3>0r_1>0,r_2>0,r_3>0, there exists r>0r>0 such that

GΠ(r)⊂EΠ(r1,r2,r3).\mathcal{G}_{\Pi}(r)\subset\mathcal{E}_{\Pi}(r_1,r_2,r_3).

For any r>0r>0, there exist r1>0,r2>0,r3>0r_1>0,r_2>0,r_3>0 such that

EΠ(r1,r2,r3)⊂GΠ(r).\mathcal{E}_{\Pi}(r_1,r_2,r_3)\subset\mathcal{G}_{\Pi}(r).

Thus, under the assumptions of Theorem 1, there exists c1c_1, depending only on rr, such that

∣EPΠΨ~1+f∣−1≥c1,∣EPΠΨ~1+h∣−1≥c1.\left|\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\right|^{-1}\geq c_1,\qquad \left|\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+h}\right|^{-1}\geq c_1.

Estimate (4) now implies the existence of a positive constant AA such that

∣EPΠΨ~1+fΨ~1+h−EPΠΨ~1+fEPΠΨ~1+h∣≤A(⟨∣f∣,∣h∣⟩H(2)(Π)+⟨∣f∣,∣h∣⟩L2(E,Πdμ)).\left|\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\widetilde{\Psi}_{1+h} -\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+h}\right| \leq A\left(\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}+\langle|f|,|h|\rangle_{L_2(E,\Pi d\mu)}\right).

Theorem 1 follows now from Proposition 2. We turn to the proof of Proposition 2.

Our first step is to represent the correlation ratio as a Fredholm determinant.

Lemma 1. Let f,h∈HS(Π)f,h\in HS(\Pi) such that the operators 1+ΠfΠ1+\Pi f\Pi and 1+ΠhΠ1+\Pi h\Pi are invertible. Then

EPΠΨ~1+f⋅Ψ~1+hEPΠΨ~1+f⋅EPΠΨ~1+h=det⁡(1+(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1).(5)\frac{\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}}{\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{\mathbb{P}_{\Pi}}\widetilde{\Psi}_{1+h}} = \det\left(1+(\Pi f h\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right). \tag*{(5)}

The proof of Lemma 1 is given in Section 2.

Set

KΠ(f,h)=(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1.(6)K_{\Pi}(f,h)=(\Pi f h\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}. \tag*{(6)}

Rewrite equation (5) in the formula

EΠΨ~1+f⋅Ψ~1+hEΠΨ~1+f⋅EΠΨ~1+h=det⁡(1+KΠ(f,h)).(7)\frac{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}}{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+h}}=\det(1+K_{\Pi}(f,h)). \tag*{(7)}

To estimate the Fredholm determinant det⁡(1+KΠ(f,h))\det(1+K_{\Pi}(f,h)), we first estimate the regularized determinant

det⁡2(1+KΠ(f,h))=det⁡2(1+(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1),\det_{2}(1+K_{\Pi}(f,h))=\det_{2}\left(1+(\Pi fh\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right),

and then estimate the trace tr⁡(KΠ(f,h))\operatorname{tr}(K_{\Pi}(f,h)) separately.

The following lemma gives upper and lower bounds for the regularized Hilbert–Carleman determinant det⁡2(1+KΠ(f,h))\det_{2}(1+K_{\Pi}(f,h)).

Lemma 2. (1) For every r>0r>0, there exists C2(r)>0C_{2}(r)>0 such that, for all f,h∈CΠ(r)f,h\in\mathcal{C}_{\Pi}(r), the following inequality holds:

∣det⁡2(1+KΠ(f,h))∣≤\left|\det_{2}(1+K_{\Pi}(f,h))\right|\le
≤exp⁡(C2(r)(⟨∣f∣2,∣h∣2⟩L2(Π)+⟨∣f∣2,∣h∣2⟩H(2)(Π))).(8)\le\exp\left(C_{2}(r)\left(\langle|f|^{2},|h|^{2}\rangle_{L_{2}(\Pi)}+\langle|f|^{2},|h|^{2}\rangle_{H^{(2)}(\Pi)}\right)\right). \tag*{(8)}

(2) For every r>0r>0 and δ0>0\delta_{0}>0, there exists C3(r,δ0)>0C_{3}(r,\delta_{0})>0 such that for all f,h∈CΠ(r)f,h\in\mathcal{C}_{\Pi}(r) with (f,h)∈NΠ(δ0)(f,h)\in\mathcal{N}_{\Pi}(\delta_{0}), the following inequality holds:

∣det⁡2(1+(ΠfhΠ−ΠfΠhΠ)RhRf)∣≥\left|\det_{2}(1+(\Pi fh\Pi-\Pi f\Pi h\Pi)R_{h}R_{f})\right|\ge
≥exp⁡(−C3(r)(⟨∣f∣2,∣h∣2⟩L2(Π)+⟨∣f∣2,∣h∣2⟩H(2)(Π))).(9)\ge\exp\left(-C_{3}(r)\left(\langle|f|^{2},|h|^{2}\rangle_{L_{2}(\Pi)}+\langle|f|^{2},|h|^{2}\rangle_{H^{(2)}(\Pi)}\right)\right). \tag*{(9)}

Note that the upper bound requires only that f,hf,h belong to CΠ(r)\mathcal{C}_{\Pi}(r), whereas the lower bound relies on the condition (f,h)∈NΠ(δ0)(f,h)\in\mathcal{N}_{\Pi}(\delta_{0}). The proof of Lemma 2 is given in Section 2.

After estimating the regularized Hilbert–Carleman determinant

det⁡2(1+KΠ(f,h)),\det_{2}(1+K_{\Pi}(f,h)),

in order to prove Proposition 2 and the main Theorem 1, it remains to estimate the trace tr⁡(KΠ(f,h))\operatorname{tr}(K_{\Pi}(f,h)).

Lemma 3. For any r>0r>0, there exists C4(r)>0C_{4}(r)>0, depending only on rr, such that for all f,h∈CΠ(r)f,h\in\mathcal{C}_{\Pi}(r), we have

∣tr⁡(KΠ(f,h))−⟨f,h‾⟩L2(Π)∣≤C4(r)⟨∣f∣,∣h∣⟩H(2)(Π).(10)\left|\operatorname{tr}(K_{\Pi}(f,h))-\langle f,\overline{h}\rangle_{L_{2}(\Pi)}\right|\le C_{4}(r)\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}. \tag*{(10)}

The proof of Lemma 3 is given in Section 3. Lemmas 2 and 3 directly imply Proposition 2, which in turn implies Theorem 1.

Proofs of Lemmas 1 and 2

Proof of Lemma 1

Proof of Lemma 1. Lemma 1 can be proved as follows: a direct computation gives

EΠΨ~1+f⋅Ψ~1+hEΠΨ~1+f⋅EΠΨ~1+h=EΠΨ1+f⋅Ψ1+hEΠΨ1+f⋅EΠΨ1+h=det⁡(1+(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1)(11)\frac{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}}{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+h}} = \frac{\mathbb{E}_{\Pi}\Psi_{1+f}\cdot\Psi_{1+h}}{\mathbb{E}_{\Pi}\Psi_{1+f}\cdot\mathbb{E}_{\Pi}\Psi_{1+h}} = \det\left(1+(\Pi fh\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right) \tag*{(11)}

for f,hf,h such that the operators ΠfΠ\Pi f\Pi and ΠhΠ\Pi h\Pi are trace class. In the general case, when f,hf,h satisfy fΠ∈HSf\Pi\in HS and hΠ∈HSh\Pi\in HS, the formula follows by continuity.

Equality (11) can also be verified by direct calculation as follows. By definition, we have

Ψ~1+h⋅Ψ~1+f=Ψ~(1+h)(1+f)×exp⁡(tr⁡ΠhfΠ).\widetilde{\Psi}_{1+h}\cdot\widetilde{\Psi}_{1+f} = \widetilde{\Psi}_{(1+h)(1+f)}\times\exp(\operatorname{tr}\Pi h f\Pi).

Recall also that for any Hilbert–Schmidt operators A,BA,B, the following identity holds:

det⁡2((1+A)(1+B))=det⁡2(1+A)⋅det⁡2(1+B)⋅exp⁡(−tr⁡(AB)).\det_{2}\left((1+A)(1+B)\right)=\det_{2}(1+A)\cdot\det_{2}(1+B)\cdot\exp\left(-\operatorname{tr}(AB)\right).

We can therefore write

EΠΨ~1+f⋅Ψ~1+hEΠΨ~1+f⋅EΠΨ~1+h=det⁡2(1+Π(f+h)Π+ΠfhΠ)det⁡2(1+Π(f+h)Π+ΠfΠhΠ)×exp⁡(tr⁡(ΠfhΠ−ΠfΠhΠ)).\frac{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h}}{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+h}} = \frac{\det_{2}\left(1+\Pi(f+h)\Pi+\Pi fh\Pi\right)}{\det_{2}\left(1+\Pi(f+h)\Pi+\Pi f\Pi h\Pi\right)} \times\exp\left(\operatorname{tr}(\Pi fh\Pi-\Pi f\Pi h\Pi)\right).

Furthermore, for any Hilbert–Schmidt operator AA such that 1+A1+A is invertible, and any Hilbert–Schmidt operator Δ\Delta, we have the identity

det⁡2(1+A+Δ)det⁡2(1+A)=det⁡2(1+Δ(1+A)−1)⋅exp⁡(−tr⁡(ΔA(1+A)−1)).\frac{\det_{2}(1+A+\Delta)}{\det_{2}(1+A)} = \det_{2}\left(1+\Delta(1+A)^{-1}\right)\cdot\exp\left(-\operatorname{tr}\left(\Delta A(1+A)^{-1}\right)\right).

Applying this identity, we obtain

EΠΨ~1+h⋅Ψ~1+fEΠΨ~1+h⋅EΠΨ~1+f=det⁡2(1+(ΠhfΠ−ΠhΠfΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1)×exp⁡(tr⁡(ΠhfΠ−ΠhΠfΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1)=det⁡(1+(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1).\begin{aligned} \frac{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+h}\cdot\widetilde{\Psi}_{1+f}}{\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+h}\cdot\mathbb{E}_{\Pi}\widetilde{\Psi}_{1+f}} &=\det_{2}\left(1+(\Pi hf\Pi-\Pi h\Pi f\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right) \\ &\quad\times\exp\left(\operatorname{tr}(\Pi hf\Pi-\Pi h\Pi f\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right) \\ &=\det\left(1+(\Pi fh\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right). \end{aligned}

Lemma 1 is proved. ■

Remark 2. Recall here that multiplying a determinantal measure by a regularized multiplicative functional again yields a determinantal point process; see [3].

Proof of Lemma 2

Set

Rh=(1+ΠhΠ)−1,Rf=(1+ΠfΠ)−1.(12)R_h = (1+\Pi h\Pi)^{-1}, \qquad R_f = (1+\Pi f\Pi)^{-1}. \tag*{(12)}

We have

Rh=1−ΠhΠRh=1−RhΠhΠ,Rf=1−ΠfΠRf=1−RfΠfΠ.R_h = 1-\Pi h\Pi R_h = 1-R_h\Pi h\Pi,\qquad R_f = 1-\Pi f\Pi R_f = 1-R_f\Pi f\Pi.

By assumption, we have

∥Rf∥≤r,∥Rh∥≤r.(13)\lVert R_f\rVert\le r,\qquad\lVert R_h\rVert\le r. \tag*{(13)}

Our aim is to estimate

det⁡2(1+KΠ(f,h))=det⁡2(1+(ΠfhΠ−ΠfΠhΠ)(1+ΠhΠ)−1(1+ΠfΠ)−1).\det\nolimits_2(1+K_\Pi(f,h))=\det\nolimits_2\left(1+(\Pi fh\Pi-\Pi f\Pi h\Pi)(1+\Pi h\Pi)^{-1}(1+\Pi f\Pi)^{-1}\right).

To obtain the upper bound, it suffices to bound the Hilbert–Schmidt norm of the operator (ΠfhΠ−ΠfΠhΠ)(\Pi fh\Pi-\Pi f\Pi h\Pi) from above and apply the classical Carleman inequality. Indeed, the upper bound follows directly from Carleman’s inequality (cf. (17)), the identity

∥ΠfhΠ∥HS2=⟨∣f∣2,∣h∣2⟩L2(Π)\lVert\Pi fh\Pi\rVert_{HS}^{2}=\langle|f|^{2},|h|^{2}\rangle_{L^{2}(\Pi)}

and the inequality

∥ΠfΠhΠ∥HS2≤⟨∣f∣2,∣h∣2⟩H(2)(Π).\lVert\Pi f\Pi h\Pi\rVert_{HS}^{2}\le\langle|f|^{2},|h|^{2}\rangle_{H^{(2)}(\Pi)}.

To obtain the lower bound, apply Lemma 5 to A=KΠ(f,h)A=K_\Pi(f,h). In view of condition (13), for every δ1>0\delta_{1}>0 there exists δ0>0\delta_{0}>0 such that (f,h)∈NΠ(δ0)(f,h)\in\mathcal{N}_{\Pi}(\delta_{0}) implies ∥KΠ(f,h)∥HS<δ1\lVert K_\Pi(f,h)\rVert_{HS}<\delta_{1} and, a fortiori, ∥KΠ(f,h)∥HS<δ1\lVert K_\Pi(f,h)\rVert_{HS}<\delta_{1}. The second part of Lemma 2 now follows directly from Lemma 5.

Estimating the trace in terms of Hilbert–Schmidt norms and proof of Lemma 3

Proposition 4. For every r>0r>0, there exists C4(r)>0C_{4}(r)>0, depending only on rr, such that for all f,h∈CΠ(r)f,h\in\mathcal{C}_{\Pi}(r), we have

∣tr⁡((ΠfhΠ−ΠfΠhΠ)RhRf−ΠfhΠ+ΠfΠhΠ+ΠfhΠ(f+h)Π)∣≤C4(r)∥∣f∣Π∣h∣∥HS2=C4(r)⟨∣f∣,∣h∣⟩H(2)(Π).\left|\operatorname{tr}\left((\Pi fh\Pi-\Pi f\Pi h\Pi)R_hR_f-\Pi fh\Pi+\Pi f\Pi h\Pi+\Pi fh\Pi(f+h)\Pi\right)\right| \le C_{4}(r)\lVert\sqrt{|f|}\Pi\sqrt{|h|}\rVert_{HS}^{2} =C_{4}(r)\langle|f|,|h|\rangle_{H^{(2)}(\Pi)}.

By definition,

tr⁡(ΠfhΠ)=⟨f,h‾⟩L2(Π),\operatorname{tr}(\Pi fh\Pi)=\langle f,\overline{h}\rangle_{L^{2}(\Pi)},
tr⁡(ΠfΠhΠ)=⟨f,h‾⟩H(2)(Π),\operatorname{tr}(\Pi f\Pi h\Pi)=\langle f,\overline{h}\rangle_{H^{(2)}(\Pi)},
tr⁡(ΠfhΠ(f+h)Π)=⟨fh,f‾+h‾⟩H(2)(Π).\operatorname{tr}(\Pi fh\Pi(f+h)\Pi)=\langle fh,\overline{f}+\overline{h}\rangle_{H^{(2)}(\Pi)}.

It is clear that

⟨fh,f+h‾⟩H(2)(Π)≤(∣f∣L∞+∣h∣L∞)⟨∣f∣,∣h∣⟩H(2)(Π).\left\langle fh,\overline{f+h}\right\rangle_{H^{(2)}(\Pi)} \leq\left(|f|_{L^\infty}+|h|_{L^\infty}\right)\left\langle|f|,|h|\right\rangle_{H^{(2)}(\Pi)}.

Thus, Proposition 4 yields

Corollary 3. For every r>0r>0, there exists C5(r)>0C_5(r)>0, depending only on rr, such that for all f,h∈CΠ(r)f,h\in\mathbb{C}_\Pi(r), we have

∣tr⁡((ΠfhΠ−ΠfΠhΠ)RhRf)−⟨f,h‾⟩L2(Π)∣≤C5(r)⟨∣f∣,∣h∣⟩H(2)(Π).\left|\operatorname{tr}\left((\Pi fh\Pi-\Pi f\Pi h\Pi)R_hR_f\right)-\left\langle f,\overline{h}\right\rangle_{L_2(\Pi)}\right| \leq C_5(r)\left\langle|f|,|h|\right\rangle_{H^{(2)}(\Pi)}.

Corollary 3 follows directly from Proposition 4 and it in turn implies Lemma 3. We proceed to the proof of Proposition 4. The proof of Proposition 4 is based on the following simple observation: if Q:L2(E,μ)→L2(E,μ)Q:L_2(E,\mu)\to L_2(E,\mu) is a bounded operator, then we have

∣tr⁡ΠfΠhQh∣≤∥Q∥⋅∥fΠh∥HS2,∣tr⁡ΠhΠfQf∣≤∥Q∥⋅∥hΠf∥HS2,∣tr⁡ΠfhΠhQh∣≤∥Q∥⋅∥h∥L∞⋅∥fΠh∥HS2,∣tr⁡ΠfhΠhQf∣≤∥Q∥⋅∥f∥L∞⋅∥hΠf∥HS2.(14)\begin{aligned} \left|\operatorname{tr}\Pi f\Pi hQh\right| &\leq\|Q\|\cdot\|\sqrt{f}\Pi h\|_{HS}^{2},\\ \left|\operatorname{tr}\Pi h\Pi fQf\right| &\leq\|Q\|\cdot\|\sqrt{h}\Pi f\|_{HS}^{2},\\ \left|\operatorname{tr}\Pi fh\Pi hQh\right| &\leq\|Q\|\cdot\|h\|_{L^\infty}\cdot\|\sqrt{f}\Pi h\|_{HS}^{2},\\ \left|\operatorname{tr}\Pi fh\Pi hQf\right| &\leq\|Q\|\cdot\|f\|_{L^\infty}\cdot\|\sqrt{h}\Pi f\|_{HS}^{2}. \tag*{(14)} \end{aligned}

It is clear that

∥fΠh∥HS2≤∥h∥L∞⋅∥fΠh∥HS2,\|\sqrt{f}\Pi h\|_{HS}^{2}\leq\|h\|_{L^\infty}\cdot\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2},

and similarly for the norm ∥hΠf∥HS2\|\sqrt{h}\Pi f\|_{HS}^{2}.

Write

(ΠfhΠ−ΠfΠhΠ)RhRf=ΠfhΠ−ΠfΠhΠ−ΠfhΠhΠRh−ΠfhΠfΠRf+ΠfΠhΠhΠRh+ΠfΠhΠfΠRf+(ΠfhΠ−ΠfΠhΠ)ΠhΠRhRfΠfΠ.\begin{aligned} (\Pi fh\Pi-\Pi f\Pi h\Pi)R_hR_f={}&\Pi fh\Pi-\Pi f\Pi h\Pi\\ &-\Pi fh\Pi h\Pi R_h-\Pi fh\Pi f\Pi R_f+\Pi f\Pi h\Pi h\Pi R_h\\ &+\Pi f\Pi h\Pi f\Pi R_f+(\Pi fh\Pi-\Pi f\Pi h\Pi)\Pi h\Pi R_hR_f\Pi f\Pi. \end{aligned}

It is clear that

∣tr⁡(ΠfhΠ−ΠfΠhΠ)ΠhΠRhRfΠfΠ∣≤C6(r)∥fΠh∥HS2.\left|\operatorname{tr}(\Pi fh\Pi-\Pi f\Pi h\Pi)\Pi h\Pi R_hR_f\Pi f\Pi\right| \leq C_6(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2}.

It follows from the above estimates that we have

∣tr⁡ΠfΠhΠhΠRh∣=∣tr⁡ΠfΠhΠRhΠhΠ∣≤C7(r)∥fΠh∥HS2,\left|\operatorname{tr}\Pi f\Pi h\Pi h\Pi R_h\right| =\left|\operatorname{tr}\Pi f\Pi h\Pi R_h\Pi h\Pi\right| \leq C_7(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2},
∣tr⁡ΠfhΠfΠRf∣=∣tr⁡ΠfhΠRfΠfΠ∣≤C8(r)∥fΠh∥HS2.\left|\operatorname{tr}\Pi fh\Pi f\Pi R_f\right| =\left|\operatorname{tr}\Pi fh\Pi R_f\Pi f\Pi\right| \leq C_8(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2}.

Next, we have

ΠfhΠRh=ΠfhΠhΠ−ΠfhΠhΠRhΠhΠ.\Pi fh\Pi R_h=\Pi fh\Pi h\Pi-\Pi fh\Pi h\Pi R_h\Pi h\Pi.

It follows from the above estimates that we have

∣tr⁡ΠfhΠhΠRhRfΠfΠ∣≤C9(r)∥fΠh∥HS2;\left|\operatorname{tr}\Pi fh\Pi h\Pi R_hR_f\Pi f\Pi\right| \leq C_9(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2};
∣tr⁡ΠfΠhΠhΠRhRfΠfΠ∣=∣tr⁡ΠfΠhΠRhΠhΠfΠRf∣≤≤C10(r)∥fΠh∥HS2≤C10(r)∥f∥L∞2⋅∥h∥L∞2⋅∥fΠh∥HS2,\begin{aligned} \left|\operatorname{tr}\Pi f\Pi h\Pi h\Pi R_hR_f\Pi f\Pi\right|=\left|\operatorname{tr}\Pi f\Pi h\Pi R_h\Pi h\Pi f\Pi R_f\right|\le\\ &\le C_{10}(r)\|f\Pi h\|_{HS}^{2}\le C_{10}(r)\|f\|_{L^\infty}^{2}\cdot\|h\|_{L^\infty}^{2}\cdot\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2}, \end{aligned}

which in turn implies the inequality

∣tr⁡ΠfΠhΠRhΠhΠ∣≤C11(r)∥fΠh∥HS2.\left|\operatorname{tr}\Pi f\Pi h\Pi R_h\Pi h\Pi\right|\le C_{11}(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2}.

Next, we have the identity

ΠfΠhΠfΠRf=ΠfΠhΠfΠ−ΠfΠhΠfΠRfΠfΠ,\Pi f\Pi h\Pi f\Pi R_f=\Pi f\Pi h\Pi f\Pi-\Pi f\Pi h\Pi f\Pi R_f\Pi f\Pi,

and we clearly have the inequality

∣tr⁡ΠfΠhΠfΠRfΠfΠ∣≤C12(r)∥fΠh∥HS2.\left|\operatorname{tr}\Pi f\Pi h\Pi f\Pi R_f\Pi f\Pi\right|\le C_{12}(r)\|\sqrt{f}\Pi\sqrt{h}\|_{HS}^{2}.

Proposition 4 is proved. ■

Spectral density of multiplicative functionals of stationary determinantal point processes

Proofs of Corollaries 1 and 2

We will need a simple lemma on the convolution of rapidly decaying functions.

Let Φ:[0,+∞)→(0,+∞)\Phi:[0,+\infty)\to(0,+\infty) be a positive increasing function satisfying

∫0+∞dtΦ(t)<+∞.\int_{0}^{+\infty}\frac{dt}{\Phi(t)}<+\infty.

Extend Φ\Phi to an even function by setting Φ(−t)=Φ(t)\Phi(-t)=\Phi(t) for t>0t>0.

Lemma 4. Let f,h:R→Rf,h:\mathbb{R}\to\mathbb{R} be Borel functions such that

sup⁡t∈RΦ(t)(∣f(t)∣+∣h(t)∣)=A<+∞.\sup_{t\in\mathbb{R}}\Phi(t)(|f(t)|+|h(t)|)=A<+\infty.

We then have the inequality

sup⁡R∈RΦ(R2)⋅∣f∗h(R)∣≤A(∥f∥L1+∥h∥L1).\sup_{R\in\mathbb{R}}\Phi\left(\frac{R}{2}\right)\cdot|f*h(R)|\le A(\|f\|_{L^1}+\|h\|_{L^1}).

Proof of Lemma 4. Let R>0R>0. We write

f∗h(R)=∫−∞+∞f(R−u)h(u) du==∫−∞R/2f(R−u)h(u) du+∫R/2+∞f(R−u)h(u) du.\begin{aligned} f*h(R)=\int_{-\infty}^{+\infty}f(R-u)h(u)\,du= \\ &=\int_{-\infty}^{R/2}f(R-u)h(u)\,du+\int_{R/2}^{+\infty}f(R-u)h(u)\,du. \end{aligned}

The absolute value of the first integral is at most A∥h∥L1/Φ(R/2)A\|h\|_{L^1}/\Phi(R/2), while the absolute value of the second is at most A∥f∥L1/Φ(R/2)A\|f\|_{L^1}/\Phi(R/2). The proof of Lemma 4 is now complete. ■

Next, for α>0\alpha>0, let KB,α\mathcal{K}_{B,\alpha} be the set of Borel functions ff on R\mathbb{R} such that sup⁡t∈RΦB(αt)∣f(t)∣<+∞\sup_{t\in\mathbb{R}}\Phi_B(\alpha t)|f(t)|<+\infty, and define

∥f∥KB,α=sup⁡t∈RΦB(αt)∣f(t)∣.\|f\|_{\mathcal{K}_{B,\alpha}}=\sup_{t\in\mathbb{R}}\Phi_B(\alpha t)|f(t)|.

Recall that we make the key condition (1), as above. For any f,h∈KBf,h\in\mathcal{K}_B, we now have f∗h∈KB,1/2f*h\in\mathcal{K}_{B,1/2}, and

∥f∗h∥KB,1/2≤(∥f∥KB+∥h∥KB)⋅(∫RdtΦB(t)).\|f*h\|_{\mathcal{K}_{B,1/2}}\leq(\|f\|_{\mathcal{K}_B}+\|h\|_{\mathcal{K}_B})\cdot\left(\int_{\mathbb{R}}\frac{dt}{\Phi_B(t)}\right).

The following simple proposition, which estimates the inner product ⟨f,h(⋅+R)⟩H(2)(ΠB)\langle f,h(\cdot+R)\rangle_{H^{(2)}(\Pi_B)} in terms of ΦB\Phi_B, plays a key rôle in deriving Corollaries 1 and 2 from Theorem 1.

Proposition 5. If a bounded Borel set BB satisfies assumption (1), then for all f,h∈KBf,h\in\mathcal{K}_B we have

ΦB(R4)⋅∣⟨f,h(⋅+R)⟩H(2)(ΠB)∣≤4(∥f∥KB+∥h∥KB)(1+∥ΦB−1∥L1).\Phi_B\left(\frac{R}{4}\right)\cdot\left|\langle f,h(\cdot+R)\rangle_{H^{(2)}(\Pi_B)}\right|\leq4(\|f\|_{\mathcal{K}_B}+\|h\|_{\mathcal{K}_B})(1+\|\Phi_B^{-1}\|_{L^1}).

Proof of Proposition 5. By definition,

⟨f,h(⋅+R)⟩H(2)(ΠB)=∫R∫Rf(u)h(u+R+k)∣ΠB(k)∣2 du dk.\langle f,h(\cdot+R)\rangle_{H^{(2)}(\Pi_B)} =\int_{\mathbb{R}}\int_{\mathbb{R}}f(u)h(u+R+k)|\Pi^B(k)|^2\,du\,dk.

Define f~\widetilde{f} by the formula f~(k)=f(−k)\widetilde{f}(k)=f(-k). Set g=f~∗hg=\widetilde{f}*h. By Lemma 4, g∈KB,1/2g\in\mathcal{K}_{B,1/2} and we have

∥g∥KB,1/2≤(∥f∥KB+∥h∥KB)⋅∥ΦB−1∥L1.\|g\|_{\mathcal{K}_{B,1/2}}\leq(\|f\|_{\mathcal{K}_B}+\|h\|_{\mathcal{K}_B})\cdot\|\Phi_B^{-1}\|_{L^1}.

Write

⟨f,h(⋅+R)⟩=∫Rg(R+k)∣ΠB(k)∣2 dk.\langle f,h(\cdot+R)\rangle=\int_{\mathbb{R}}g(R+k)|\Pi^B(k)|^2\,dk.

Applying Lemma 4 once again proves the proposition.

Finally, note that if f,h∈KBf,h\in\mathcal{K}_B, then for every δ>0\delta>0, the condition (f,h(⋅+R))∈NΠB(δ)(f,h(\cdot+R))\in\mathcal{N}_{\Pi_B}(\delta) automatically holds for all sufficiently large RR.

For the correlation coefficient of multiplicative functionals from Theorem 1, we conclude that for any bounded Borel set B⊂RB\subset\mathbb{R} satisfying condition (1), and for any r>0r>0 and δ>0\delta>0, there exists A>0A>0 such that, if

f,h∈KB∩VΠB(r)f,h\in\mathcal{K}_{B}\cap\mathcal{V}_{\Pi^{B}}(r)

and ∥f∥KB,∥h∥KB<D\lVert f\rVert_{\mathcal{K}_{B}},\lVert h\rVert_{\mathcal{K}_{B}}<D then

Φ(R4)⋅∣EPΠBΨ~1+f⋅Ψ~1+h(⋅+R)−EPΠBΨ~1+f⋅EPΠBΨ~1+h∣≤A.(15)\Phi\left(\frac{R}{4}\right)\cdot\left|\mathbb{E}_{P_{\Pi^{B}}}\widetilde{\Psi}_{1+f}\cdot\widetilde{\Psi}_{1+h(\cdot+R)}-\mathbb{E}_{P_{\Pi^{B}}}\widetilde{\Psi}_{1+f}\cdot\mathbb{E}_{P_{\Pi^{B}}}\widetilde{\Psi}_{1+h}\right|\leq A. \tag*{(15)}

This proves Corollaries 1 and 2.

The case of additive functionals

We compare our estimates for the decay of correlation of multiplicative functionals with the formulas for the spectral densities of additive functionals obtained by A. B. Soshnikov (cf. (3.11) in [15]). To a bounded Borel function ff with compact support on EE assign the additive functional SfS_f on Conf⁡(E)\operatorname{Conf}(E) by the formula

Sf(X)=∑x∈Xf(x).S_f(X)=\sum_{x\in X}f(x).

We have

ESf=∫Ef(x)Π(x,x) dμ(x).\mathbb{E}S_f=\int_E f(x)\Pi(x,x)\,d\mu(x).

Set S~f=Sf−ESf\widetilde{S}_f=S_f-\mathbb{E}S_f. A direct calculation gives

EPΠ∣S~f∣2=12∫E∫E∣f(x)−f(y)∣2∣Π(x,y)∣2dμ(x)dμ(y).\mathbb{E}_{P_{\Pi}}\lvert\widetilde{S}_f\rvert^2=\frac{1}{2}\int_E\int_E\lvert f(x)-f(y)\rvert^2\lvert\Pi(x,y)\rvert^2d\mu(x)d\mu(y).

Define the Hilbert space H˙1/2(Π)\dot{H}_{1/2}(\Pi) as the completion of the family of all bounded Borel functions with compact support on EE with respect to the Euclidean norm ∥f∥H˙1/2(Π)\lVert f\rVert_{\dot{H}_{1/2}(\Pi)}, defined by

∥f∥H˙1/2(Π)2=12∫E∫E∣f(x)−f(y)∣2∣Π(x,y)∣2dμ(x)dμ(y).\lVert f\rVert_{\dot{H}_{1/2}(\Pi)}^2=\frac{1}{2}\int_E\int_E\lvert f(x)-f(y)\rvert^2\lvert\Pi(x,y)\rvert^2d\mu(x)d\mu(y).

The map f↦S~ff\mapsto\widetilde{S}_f extends to an isometric embedding of the Hilbert space H˙1/2(Π)\dot{H}_{1/2}(\Pi) into L2(Conf⁡(E),PΠ)L_2(\operatorname{Conf}(E),P_{\Pi}).

We now return to the special case of a stationary determinantal point process on R\mathbb{R} with the kernel

ΠB(x,y)=1ˇB(x−y).\Pi^{B}(x,y)=\check{\mathbf{1}}_{B}(x-y).

A direct calculation gives

EPΠBSf~⋅Sf(⋅+R)~=12∫R∫R(f(x)−f(y))(f(x+R)−f(y+R))‾∣ΠB(x,y)∣2 dx dy=12∫R∫R(f(x+h)−f(x))(f(x+h+R)−f(x+R))‾∣1B^(h)∣2 dx dh=14π∫R∫R∣f^(λ)∣2∣eiλh−1∣2∣1B^(h)∣2e−iλR dλ dh=12π∫R∣f^(λ)∣2(∣1B^∣2(0)−∣1B^∣2(λ))e−iλR dλ.(16)\begin{aligned} \mathbb{E}_{P_{\Pi^B}}\widetilde{S_f}\cdot\widetilde{S_{f(\cdot+R)}} &= \frac{1}{2}\int_{\mathbb{R}}\int_{\mathbb{R}}(f(x)-f(y))\overline{(f(x+R)-f(y+R))}|\Pi^B(x,y)|^2\,dx\,dy \\ &= \frac{1}{2}\int_{\mathbb{R}}\int_{\mathbb{R}}(f(x+h)-f(x))\overline{(f(x+h+R)-f(x+R))}|\widehat{\mathbf{1}_B}(h)|^2\,dx\,dh \\ &= \frac{1}{4\pi}\int_{\mathbb{R}}\int_{\mathbb{R}}|\widehat{f}(\lambda)|^2|e^{i\lambda h}-1|^2|\widehat{\mathbf{1}_B}(h)|^2e^{-i\lambda R}\,d\lambda\,dh \\ &= \frac{1}{2\pi}\int_{\mathbb{R}}|\widehat{f}(\lambda)|^2\left(|\widehat{\mathbf{1}_B}|^2(0)-|\widehat{\mathbf{1}_B}|^2(\lambda)\right)e^{-i\lambda R}\,d\lambda. \tag*{(16)} \end{aligned}

In the last step, we used the fact that the function ∣1B^∣2|\widehat{\mathbf{1}_B}|^2 is even. Set

QB=∣1B^∣2.Q_B=|\widehat{\mathbf{1}_B}|^2.

Let ρB,f\rho^{B,f} stand for the spectral density of the regularized additive functional Sf~\widetilde{S_f} with respect to the determinantal measure PΠBP_{\Pi^B}. Following A. B. Soshnikov, observe that equation (16) implies the identity

ρB,f(λ)=∣f^(λ)∣2(QB(0)−QB(λ)).\rho^{B,f}(\lambda)=|\widehat{f}(\lambda)|^2(Q_B(0)-Q_B(\lambda)).

Using Lemma 4, one can directly verify that if f∈KBf\in\mathcal{K}_B, then we have

sup⁡R∈RΦB(R/4)∣EPΠBSf~⋅Sf(⋅+R)~∣<+∞.\sup_{R\in\mathbb{R}}\Phi_B(R/4)\left|\mathbb{E}_{P_{\Pi^B}}\widetilde{S_f}\cdot\widetilde{S_{f(\cdot+R)}}\right|<+\infty.

Therefore, Theorem 1 gives the same rate of correlation decay for multiplicative functionals as Soshnikov’s spectral formula does for additive functionals.

Question. What lower bounds can be obtained for the correlation ratio of multiplicative functionals of determinantal point processes?

Bounds for Fredholm and Hilbert–Carleman determinants

For completeness, this section provides proofs of the elementary estimates for Fredholm and Hilbert–Carleman determinants used in the paper. We first recall Carleman’s inequality for the regularized Hilbert–Carleman determinant; see, for example, Gohberg, Goldberg, and Krupnik [9], Theorem 7.4, formula (7.10), p. 69: for every Hilbert–Schmidt operator AA,

∣det⁡2(1+A)∣≤exp⁡(12∥A∥HS2).(17)\left|\det_{2}(1+A)\right| \le\exp\left(\frac{1}{2}\|A\|_{HS}^{2}\right). \tag*{(17)}

We now obtain a simple bound for the Hilbert–Schmidt norm of the inverse operator in terms of the Hilbert–Schmidt norm of the original operator and its Hilbert–Carleman determinant.

Proposition 6. Let AA be a Hilbert–Schmidt operator such that 1+A1+A is invertible. Then

∥A1+A∥HS≤exp⁡(∥A∥HS2+1)∣det⁡2(1+A)∣.\left\|\frac{A}{1+A}\right\|_{HS} \le\frac{\exp\left(\|A\|_{HS}^{2}+1\right)}{\left|\det_{2}(1+A)\right|}.

The proof is a straightforward adaptation of the proof of Corollary 5.2 in Simon [13].

Set

B=(A1+A)∗∥A1+A∥HS.B=\frac{\left(\frac{A}{1+A}\right)^{*}}{\left\|\frac{A}{1+A}\right\|_{HS}}.

Clearly, tr⁡BA(1+A)−1=∥A1+A∥HS\operatorname{tr}B A(1+A)^{-1}=\left\|\frac{A}{1+A}\right\|_{HS}.

Consider the function

F(λ)=det⁡2(1+A+λB)=F(\lambda)=\det_{2}(1+A+\lambda B)=
det⁡2(1+A)⋅det⁡2(1+λB(1+A)−1)⋅exp⁡(−λ∥A1+A∥HS).\det_{2}(1+A)\cdot\det_{2}\left(1+\lambda B(1+A)^{-1}\right)\cdot\exp\left(-\lambda\left\|\frac{A}{1+A}\right\|_{HS}\right).

We have

dF(λ)dλ∣λ=0=−det⁡2(1+A)∥A1+A∥HS.\left.\frac{dF(\lambda)}{d\lambda}\right|_{\lambda=0} =-\det_{2}(1+A)\left\|\frac{A}{1+A}\right\|_{HS}.

By Cauchy’s estimate for the derivative of an entire function,

∣dF(λ)dλ∣λ=0∣≤exp⁡(∥A∥HS2+1).\left|\left.\frac{dF(\lambda)}{d\lambda}\right|_{\lambda=0}\right| \le\exp\left(\|A\|_{HS}^{2}+1\right).

Proposition 6 is proved. ■

Next, let J1\mathcal{J}_{1} be the ideal of trace-class operators acting on the Hilbert space L2(E,μ)L_{2}(E,\mu):

J1={A:tr⁡AA∗<+∞}.\mathcal{J}_{1}=\{A:\operatorname{tr}\sqrt{AA^{*}}<+\infty\}.

As usual, write ∥A∥J1=tr⁡AA∗\|A\|_{\mathcal{J}_{1}}=\operatorname{tr}\sqrt{AA^{*}}.

Recall that if A∈J1A\in\mathcal{J}_{1}, and the operator 1+A1+A is invertible, then we have

∣det⁡((1+A)−1)∣=∣det⁡(1−A(1+A)−1)∣≤exp⁡tr⁡(∣A(1+A)−1∣)≤exp⁡(∥A∥J1∥(1+A)−1∥).(18)\begin{aligned} \left|\det\left((1+A)^{-1}\right)\right| =\left|\det\left(1-A(1+A)^{-1}\right)\right| \\ &\le\exp\operatorname{tr}\left(\left|A(1+A)^{-1}\right|\right) \le\exp\left(\|A\|_{\mathcal{J}_{1}}\|(1+A)^{-1}\|\right). \tag*{(18)} \end{aligned}

Similarly, if AA is a Hilbert–Schmidt operator and 1+A1 + A is invertible, then we have

∣det⁡2((1+A)−1)∣=∣det⁡2(1−A(1+A)−1)∣≤exp⁡(∥A(1+A)−1∥HS22)≤exp⁡(∥A∥HS2⋅∥(1+A)−1∥22).(19)\left|\det_{2}\left((1 + A)^{-1}\right)\right| = \left|\det_{2}\left(1 - A(1 + A)^{-1}\right)\right| \le\exp\left(\frac{\left\|A(1 + A)^{-1}\right\|_{HS}^{2}}{2}\right) \le\exp\left(\frac{\|A\|_{HS}^{2} \cdot\left\|(1 + A)^{-1}\right\|^{2}}{2}\right). \tag*{(19)}

We also have

det⁡2((1+A)−1)det⁡2((1+A))=exp⁡(−tr⁡A2(1+A)−1).(20)\det_{2}\left((1 + A)^{-1}\right)\det_{2}\left((1 + A)\right) = \exp\left(-\operatorname{tr} A^{2}(1 + A)^{-1}\right). \tag*{(20)}

Proposition 6 now yields the following corollary.

Corollary 4. For every R>0R > 0, there exist positive constants c13c_{13} and C13C_{13} depending only on RR such that, for every Hilbert–Schmidt operator AA whose Hilbert–Schmidt norm does not exceed RR and such that the operator 1+A1 + A is invertible, the following inequalities hold:

c13(∥(1+A)−1∥HS)−1≤∣det⁡2(1+A)∣≤C13(∥(1+A)−1∥HS)−1;(21)c_{13}\left(\left\|(1 + A)^{-1}\right\|_{HS}\right)^{-1} \le\left|\det_{2}(1 + A)\right| \le C_{13}\left(\left\|(1 + A)^{-1}\right\|_{HS}\right)^{-1}; \tag*{(21)}
c13∣det⁡2(1+A)∣−1≤∥(1+A)−1∥HS≤C13∣det⁡2(1+A)∣−1.(22)c_{13}\left|\det_{2}(1 + A)\right|^{-1} \le\left\|(1 + A)^{-1}\right\|_{HS} \le C_{13}\left|\det_{2}(1 + A)\right|^{-1}. \tag*{(22)}

If we additionally assume that our Hilbert–Schmidt operator is a contraction, then the regularized Hilbert–Carleman determinant admits a lower bound in terms of the Hilbert–Schmidt norm of the operator:

Lemma 5. Let AA be a Hilbert–Schmidt operator such that ∥A∥<1\|A\| < 1. Then

exp⁡(−∥A∥HS22(1−∥A∥)2−∥A∥HS22(1−∥A∥))≤∣det⁡2(1+A)∣.\exp\left(-\frac{\|A\|_{HS}^{2}}{2(1-\|A\|)^{2}}-\frac{\|A\|_{HS}^{2}}{2(1-\|A\|)}\right) \le\left|\det_{2}(1 + A)\right|.

To prove this, write the identity

det⁡2(1+A)⋅det⁡2((1+A)−1)=exp⁡(tr⁡(A2(1+A)−1)),\det_{2}(1 + A) \cdot\det_{2}\left((1 + A)^{-1}\right) = \exp\left(\operatorname{tr}\left(A^{2}(1 + A)^{-1}\right)\right),

which follows from the general identity

det⁡2(1+A)⋅det⁡2(1+B)=det⁡2((1+A)(1+B))⋅exp⁡(tr⁡(AB))\det_{2}(1 + A) \cdot\det_{2}(1 + B) = \det_{2}\left((1 + A)(1 + B)\right) \cdot\exp\left(\operatorname{tr}(AB)\right)

by substituting B=−A(1+A)−1B = -A(1 + A)^{-1} (so that 1+B=(1+A)−11 + B = (1 + A)^{-1}).

Under our assumptions, we have

∥(1+A)−1∥=∥1−A+A2−…∥≤(1−∥A∥)−1\left\|(1 + A)^{-1}\right\| = \left\|1 - A + A^{2} - \ldots\right\| \le(1 - \|A\|)^{-1}

and

∣tr⁡A2(1+A)−1∣≤∥A∥HS21−∥A∥.\left|\operatorname{tr} A^{2}(1 + A)^{-1}\right| \le\frac{\|A\|_{HS}^{2}}{1 - \|A\|}.

Furthermore, we have

∣det⁡2((1+A)−1)∣=∣det⁡2(1−A(1+A)−1)∣≤≤exp⁡(12∥A(1+A)−1∥HS2)≤exp⁡(∥A∥HS22(1−∥A∥)2).(23)\begin{aligned} \left|\operatorname{det}_{2}\left((1+A)^{-1}\right)\right|=\left|\operatorname{det}_{2}\left(1-A(1+A)^{-1}\right)\right|\le\\ &\le\exp\left(\frac{1}{2}\left\|A(1+A)^{-1}\right\|_{HS}^{2}\right)\le\exp\left(\frac{\|A\|_{HS}^{2}}{2(1-\|A\|)^{2}}\right). \tag*{(23)} \end{aligned}

It follows that we have

∣det⁡2(1+A)∣=∣exp⁡(−tr⁡12A2(1+A)−1)det⁡2((1+A)−1)∣≥≥exp⁡(−∥A∥HS22(1−∥A∥)2−∥A∥HS22(1−∥A∥)),(24)\begin{aligned} \left|\operatorname{det}_{2}(1+A)\right|=\left|\frac{\exp\left(-\operatorname{tr}\frac{1}{2}A^{2}(1+A)^{-1}\right)}{\operatorname{det}_{2}\left((1+A)^{-1}\right)}\right|\ge\\ &\ge\exp\left(-\frac{\|A\|_{HS}^{2}}{2(1-\|A\|)^{2}}-\frac{\|A\|_{HS}^{2}}{2(1-\|A\|)}\right), \tag*{(24)} \end{aligned}

as required. ■

Note added in proof. After this paper was completed, I became aware of the beautiful paper [6] in which speed of mixing is investigated for the sine-beta process.

Acknowledgements

I am deeply grateful to A.S. Arakcheev, M.A. Borovikov and S.M. Gorbunov for fruitful discussions. I am deeply grateful to the anonymous referees whose detailed comments greatly helped me to improve the presentation.

The author is a winner of the “Leader” competition of the BA-SIS Foundation for the Advancement of Mathematics and Theoretical Physics, and would like to thank the Jury and sponsors.

Part of this paper was written while I was a guest at the Sirius Mathematics Centre, and I am deeply grateful to the Centre for its warm hospitality.

Steklov Mathematical Institute of Russian Academy of Sciences, Moscow

Email address: [email protected]

References

  1. [1]C. Andréief. “Note sur une relation les intégrales définies des produits des fonctions”. Mém. de la Soc. Sci. Bordeaux, 2 (1883), 1–14.
  2. [3]A. I. Bufetov, “On multiplicative functionals of determinantal processes”. Uspekhi Mat. Nauk, 67:1(403) (2012), 177–178; Russian Math. Surveys, 67:1 (2012), 181–182.
  3. [4]Alexander I. Bufetov. “Quasi-symmetries of determinantal point processes”. The Annals of Probability, 46:2 (2018), 956–1003.arxiv.org/abs/1409.2068
  4. [5]A. I. Bufetov. “Conditional measures of determinantal point processes”. Functional Analysis and Its Applications, 54:1 (2020), 11–28.
  5. [6]L. Dumaz and M. Malvy “Long-Range Correlation of the Sineβ point Process”. arXiv, 2026, arXiv:2603.15289.arxiv.org/abs/2603.15289
  6. [7]Alexander Bufetov, Y. Qui and Alexander Shamov “Kernels of conditional determinantal measures and the Lyons-Peres completeness conjecture”. Jour. Eur. Maths Soc (JEMS), 23:5 (2021), 1477–1519.
  7. [8]Carleman, Torsten. Zur Theorie der linearen Integralgleichungen. Mathematische Zeitschrift. 9 (3–4): 196–217.
  8. [9]Israel Gohberg, Seymour Goldberg, Nahum Krupnik, Traces and Determinants of Linear Operators, Birkhäuser Verlag 2000.
  9. [10]O. Macchi. “The coincidence approach to stochastic point processes”. Adv. Appl. Prob., 7 (1975), 83–122.DOI
  10. [11]R. Lyons. “Determinantal probability measures”. Publ. Math. Inst. Hautes Etudes Sci., 98 (2003), 167–212.
  11. [12]Osada, H., Osada, S. Discrete Approximations of Determinantal Point Processes on Continuous Spaces: Tree Representations and Tail Triviality. J Stat Phys 170, 421–435 (2018). https://doi.org/10.1007/s10955-017-1892-2
  12. [13]Simon B. Trace Ideals and Their Applications. — 2nd ed. — Providence, RI : American Mathematical Society, 2005. — viii, 150 p. — (Mathematical Surveys and Monographs ; vol. 120).
  13. [14]T. Shirai and Y. Takahashi. “Fermion process and Fredholm determinant”. Proceedings of the Second ISAAC Congress, vol. 1, 15–23, Kluwer 2000.DOI
  14. [15]A. B. Soshnikov. “Determinantal random point fields”. Uspekhi Mat. Nauk, 55:5(335) (2000), 107–160.

Paper details

vlnc:W_4865u2j2
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents