Abstract
For integers n,q=1,2,3,… , let Pol n,q denote the \({\mathbb{C}}\)-linear space of polynomials in z and \(\bar{z}\), of degree ≤n−1 in z and of degree ≤q−1 in \(\bar{z}\). We supply Pol n,q with the inner product structure of
the resulting Hilbert space is denoted by Pol m,n,q . Here, it is assumed that m is a positive real. We let K m,n,q denote the reproducing kernel of Pol m,n,q , and study the associated determinantal process, in the limit as m,n→+∞ while n=m+O(1); the number q, the degree of polyanalyticity, is kept fixed. We call these processes polyanalytic Ginibre ensembles, because they generalize the Ginibre ensemble—the eigenvalue process of random (normal) matrices with Gaussian weight. There is a physical interpretation in terms of a system of free fermions in a uniform magnetic field so that a fixed number of the first Landau levels have been filled. We consider local blow-ups of the polyanalytic Ginibre ensembles around points in the spectral droplet, which is here the closed unit disk \(\bar{\mathbb{D}}:=\{z\in{\mathbb{C}}:|z|\le1\}\). We obtain asymptotics for the blow-up process, using a blow-up to characteristic distance m −1/2; the typical distance is the same both for interior and for boundary points of \(\bar{\mathbb{D}}\). This amounts to obtaining the asymptotical behavior of the generating kernel K m,n,q . Following (Ameur et al. in Commun. Pure Appl. Math. 63(12):1533–1584, 2010), the asymptotics of the K m,n,q are rather conveniently expressed in terms of the Berezin measure (and density)
For interior points |z|<1, we obtain that \({\mathrm{d}}B^{\langle z\rangle}_{m,n,q}(w)\to{\mathrm{d}}\delta_{z} \) in the weak-star sense, where δ z denotes the unit point mass at z. Moreover, if we blow up to the scale of m −1/2 around z, we get convergence to a measure which is Gaussian for q=1, but exhibits more complicated Fresnel zone behavior for q>1. In contrast, for exterior points |z|>1, we have instead that \({\mathrm{d}}B^{\langle z\rangle}_{m,n,q}(w) \to{\mathrm{d}}\omega(w,z, {\mathbb{D}}^{e}) \), where \({\mathrm{d}}\omega(w,z,{\mathbb{D}}^{e})\) is the harmonic measure at z with respect to the exterior disk \({\mathbb{D}}^{e}:= \{w\in{\mathbb{C}}:\, |w|>1\}\). For boundary points, |z|=1, the Berezin measure \({\mathrm{d}}B^{\langle z\rangle}_{m,n,q}\) converges to the unit point mass at z, as with interior points, but the blow-up to the scale m −1/2 exhibits quite different behavior at boundary points compared with interior points. We also obtain the asymptotic boundary behavior of the 1-point function at the coarser local scale q 1/2 m −1/2.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
1.1 Notation
We will write use standard notation, such as ∂X and int(X) for the boundary and the interior of a subset X of the complex plane \({\mathbb{C}}\). The complex conjugate of a complex number z is usually written as \(\bar{z}\). We write \({\mathbb{R}}\) for the real line, \({\mathbb{D}}:=\{z\in{\mathbb{C}}:|z|<1\}\) for the open unit disk, and \({\mathbb{D}}^{e}:=\{z\in{\mathbb{C}}:|z|>1\}\) for the open exterior (punctured) disk. The characteristic function of a set E is written 1 E . We write
for the normalized area measure in \({\mathbb{C}}\), and use the standard Wirtinger derivatives
We also write Δ for the (quarter) Laplacian
1.2 Determinantal Projection Processes
Given a locally compact topological space X with a positive Radon measure μ, a finite determinantal projection process is a random configuration of n points defined by the following probability measure on Xn:
Here, K n is the integral kernel of a projection operator to an n-dimensional subspace of L 2(X,μ). It is customary to identify all the permutations of the points and think the process as a random measure \(\sum_{j=1}^{n} \delta_{z_{j}}\) on X. It should be mentioned that finite determinantal projection processes presented here are a subclass of more general determinantal point processes which, for instance, can contain infinitely many points. We omit the general definition, as the processes we will consider in this paper are only of the special form (1.1).
Determinantal processes were originally introduced by Macchi [28] to model fermions in quantum mechanics. Indeed, the probability density (1.1) vanishes whenever any two points in the n-tuple (z 1,…,z n ) coincide (fermions are forbidden to be in the same state). We interpret this as saying that the points in the n-tuple repel each other. Point processes of this kind appear in several contexts, e.g., in random matrix theory and combinatorics (for general surveys, see [10, 24]; we should also mention the books [6, 11, 12, 15, 29]).
1.3 Eigenvalues of Random Normal Matrix Ensembles
Our main motivating example comes from the theory of random normal matrices. This topic has in recent years been subject to rather active investigation by physicists as well as by mathematicians. For an introduction, see, e.g., [38]. So, we shall use \(\mathrm{X}={\mathbb{C}}\) and dμ(z)=e−mQ(z)dA(z), for a positive weight function Q satisfying some mild regularity and growth conditions; m is a positive real parameter, and dA(z)=π −1dxdy is the normalized area measure. Let us write \(L^{2}({\mathbb{C}}, \mathrm{e}^{-mQ}):=L^{2}(\mathrm {X},\mu )\) in this situation. The determinantal projection process is associated with an n-dimensional subspace of \(L^{2}({\mathbb{C}}, \mathrm{e}^{-mQ})\), and we will use the space Pol n of all polynomials in z of degree ≤n−1; we write Pol m,n to indicate that we have supplied Pol n with the Hilbert space structure of \(L^{2}({\mathbb{C}}, \mathrm{e}^{-mQ})\). The density of the process is then given by the reproducing kernel K m,n of the space Pol m,n . So, we are talking about the probability measure
In terms of the correlation kernel
which is the integral kernel of an orthogonal projection on \(L^{2}({\mathbb{C}})\), the expression (1.2) simplifies to
The process described by (1.2) and (1.4) represents the eigenvalues of a random normal matrix picked from the distribution
where dvol nm(n)(M) is the natural Riemannian volume form on the n×n normal matrices inherited from the metric of \({\mathbb{C}}^{n^{2}}\) (see [38]); Z m,n is the normalization constant needed to make the total mass equal to 1. We are interested in the limiting behavior of the process as m,n→+∞ while n=mτ+O(1) for some positive real number τ. Without loss of generality, we will consider only τ=1.
1.4 Local Blow-Up Processes
Let \(\mathcal{N}_{+}\) and \(\mathcal{N}_{+,0}\) be the set of points defined by
In the arXiv preprint [20], which appeared later in the expanded form [21], the function \(\widehat{Q}\) was defined as a certain envelope of Q, namely the largest subharmonic function in \({\mathbb{C}}\) which is ≤Q everywhere and has the growth bound
It is known that \(\Delta\widehat{Q}=1_{\mathcal{S}}\Delta Q\) for some compact set \(\mathcal{S}\) (see, e.g., [21]). We assume that \(\mathcal{S}\) is the minimal compact with this property, and call \(\mathcal{S}\) a spectral droplet. We then have \(\mathcal{S}\subset\mathcal{N}_{+,0}\). The point process (1.2) has the following property: as m,n→+∞ while n=m+O(1), the points will tend to accumulate on the set \(\mathcal{S}\) with density ΔQ there. Moreover, the set \(\mathcal{S}\cap\mathcal{N}_{+}\) is rather regular for real-analytic Q, as the Sakai theory applies (cf. [23]; see also the related papers [19] and [22]). Typically we then expect a real-analytic boundary, with the exception of cusps and contact (or kissing) points. Let us refer to the set \(\mathrm{int}(\mathcal{S}\cap\mathcal{N}_{+})\) as the bulk. The results of [3, 4] show that for bulk points z, the local blow-up process at z, with coordinates (ξ 1,…ξ n ),
where (z 1,…,z n ) are from the process (1.2), converges weakly to the translation invariant Ginibre(∞) process, as m,n→+∞ while n=m+o(1). The associated generating kernel is the reproducing kernel \((\xi,\eta)\mapsto\mathrm{e}^{\xi\bar{\eta}}\) of the Bargmann-Fock space. This has the flavor of a universality result. The corresponding statement in the Hermitian random matrix theory is the universality of the sine kernel for bulk (the points z j would in this context be eigenvalues of the random matrix). We observe here that the sine kernel is the reproducing kernel for the Paley-Wiener space (a subspace of \(L^{2}({\mathbb{R}})\) consisting of entire functions). As for the two boundary points in the GUE model, the Tracy-Widom distribution appears, which is generated by the Airy kernel. The Airy kernel is reproducing for another Hilbert space of entire functions (see [27]). This suggests that for real-analytic Q and \(z\in\partial\mathcal {S}\cap \mathcal{N}_{+}\), there should exist a local blow-up
where θ=θ(z) is a suitable positive real, such that as m,n→+∞ while n=m+o(1), the process (ξ 1,…,ξ n ) would converge to a determinantal process whose generating kernel is the reproducing kernel of a Hilbert space of entire functions. We verify this in the context of the Ginibre ensemble (i.e., with Q(z)=|z|2), and identify the associated Hilbert space as the closed subspace of the Bargmann-Fock space characterized by slow growth in a half-plane (see also Sect. 5 where a characterization via the Bargmann transform is presented). In that case we have \(\theta=\frac{1}{2}\) as in the case of interior points. This is in contrast with Hermitian random matrix theory, where a different scaling is required for boundary and for bulk points.
1.5 The Berezin Measure and the Berezin Density
In [3–5], Ameur, Hedenmalm, and Makarov study the Berezin measure \({\mathrm{d}}B_{m,n}^{\langle z\rangle}\) and Berezin density :
which arise in the study of the Berezin transform. The Berezin measure is a probability measure, which makes it more stable than the reproducing kernel K m,n of Pol m,n itself as we let m,n→+∞. In terms of the point process (1.2), measures the amount of repulsion from z caused by placing one of the points at z. For bulk points z, we have the convergence \({\mathrm{d}}B^{\langle z\rangle}_{m,n}\to{\mathrm{d}}\delta_{z}\) in the weak-star sense of measures, as m,n→+∞ while n=m+o(1). Here, dδ z is the Dirac point mass at z. In fact, there is a better result: the blow-up Berezin density
converges to the standard Gaussian \(\mathrm{e}^{-|\xi|^{2}}\). This corresponds to the convergence of the local blow-up of the point process to the Ginibre(∞) process (cf. [3, 4]). On the other hand, for points z outside the spectral droplet, i.e., for \(z\in{\mathbb{C}}\setminus\mathcal{S}\), the Berezin measure \({\mathrm{d}}B^{\langle z\rangle}_{m,n}\) converges in the weak-star sense of measures to harmonic measure at z with respect to the exterior domain \({\mathbb{C}}\setminus\mathcal{S}\) as m,n→+∞ while n=m+o(1) (see [3] for Q=|z|2, and [4] for the general result).
1.6 The Local Blow-Up of the Point Process and the Berezin Density
It is convenient to think of the point process (1.2) in terms of the k-point intensities
We notice quickly that the intensities are unchanged if the kernel changed to
provided that χ is measurable with |χ(z)|≡1 (we can call this a “gauge transformation”). This can help in the asymptotical analysis of local blow-ups. For k=2, we get the 2-point intensity
where we recognize the Berezin density as a correction to the product of the two 1-point densities; the 1-point intensity is
So, as far as the 2-point intensity goes, we just need the 1-point intensity and the Berezin density. Since the 1-point intensity is just the restriction to z 1=z 2 of the Berezin density, the Berezin density is all we need to describe the 2-point intensity. We will be a little lazy and just work with the Berezin density in the context of local blow-ups, although the asymptotics of the k-point intensity would strictly speaking require a little more work. So, although we state many of our assertions regarding local blow-ups in terms of the Berezin density, we maintain that they generalize to statements about the point processes (cf. [4]).
1.7 The Ginibre Ensemble and Its Polyanalytic Generalization
The case Q(z)=|z|2 of (1.5) (or (1.2)) is known as the Ginibre ensemble. The (probability generating or reproducing) kernel is now particularly simple:
Here, \(\mathcal{S}=\bar{\mathbb{D}}\), the closed unit disk. We will consider a family of generalizations of the Ginibre ensemble, the polyanalytic Ginibre ensembles, which are defined by the reproducing kernels K m,n,q of the subspaces
supplied with the Hilbert space structure of \(L^{2}({\mathbb{C}},\mathrm {e}^{-m|z|^{2}})\). The correlation kernels for the polyanalytic Ginibre ensembles are given by the formula analogous to (1.3):
The parameter value q=1 corresponds to the standard Ginibre process. In general, we now project to the nq-dimensional subspace of the polyanalytic polynomials, where the degree in \(\bar{z}\) is ≤q−1, and the degree in z is ≤n−1. Note that the dimension of the subspace is now nq and not n.
The polyanalytic Ginibre ensembles can be interpreted physically in the following way. In quantum mechanics, the wavefunction of an n-body system consisting of non-interacting fermions with wavefunctions ψ j ,j=1,…,n, is given by the Slater determinant
The important property of this expression is that it satisfies the Pauli exclusion principle, i.e., it is antisymmetric with respect to interchanging two wavefunctions.
In this paper, the wavefunctions of the form \(\psi=u\mathrm{e}^{-\frac{1}{2}m|z|^{2}}\), where u is an eigenfunction of the differential operator (densely defined on \(L^{2}({\mathbb{C}},\mathrm {e}^{-m|z|^{2}})\))
play a special role. This operator represents (after suitable normalizations) a single particle in the complex plane within a uniform magnetic field of which is strength m and perpendicular to the plane. The eigenspaces for this operator are of the form (see, e.g., [30])
where \(A^{2}_{m,r}({\mathbb{C}})\) is the Bargmann-Fock space of polyanalytic functions f satisfying \({\bar{\partial}}^{r} f(z)= 0\) (cf. Sect. 2). These eigenspaces are often referred to as Landau levels (in the literature, the term “Landau level” can also refer to the eigenvalue). The symbol “⊖” means taking the orthogonal difference of the two subspaces of \(A^{2}_{m,r+1}\). We will denote by u r+1,j ,j=1,…,n, some orthonormal basis of the space
and write \(\psi_{r+1,j}:=u_{r+1,j}\mathrm{e}^{-\frac{1}{2}m|z|^{2}}\) for the corresponding wavefunctions. We now form the wavefunction for a system where there are n fermions at each of the q first Landau levels:
Here, we write z=(z 1,1,…,z r,i ,…,z q,n ) for an nq-tuple of complex numbers. The corresponding probability amplitude is \(\frac{1}{Z_{q,n}}|\varPsi_{q,n}|^{2}\), where Z q,n is a normalization constant. By a standard argument in random matrix theory (see [25] p. 47), this expression can be rewritten as (the correlation kernel is given by (1.7))
which is exactly the probability density of the polyanalytic Ginibre ensembles. As a result, the polyanalytic Ginibre ensembles model systems of free fermions where all Landau levels below a given level contain the same number of particles. We will focus on the case where m and n large and kept relatively close to each other: physically, this corresponds to each Landau level being completely filled. We should point out that in our treatment, there is no special significance in the choice of having exactly the same number of particles in each Landau level below the cut-off level—other configurations can be analyzed in a similar manner.
For more background on the free fermion interpretation in the normal matrix (ground state) model, see [37]; more information about the connection between polyanalytic functions and Landau levels is contained in [1, 7, 30, 31, 36].
As for the point process, the points generally repel each other, but for q>1, they also tend to avoid certain geometric configurations, such as circles and lines. We have run a simulation in Fig. 1.
1.8 Results
Macroscopically, we find that the behavior of the polyanalytic Ginibre ensemble is similar to that of the Ginibre ensemble. If we measure this in terms of the Berezin measure, we have that as m,n→+∞ while n=m+O(1),
where ω z is the harmonic measure of z with respect to the exterior disk \({\mathbb{D}}^{e}\). Here, K m,n,q is the reproducing kernel for Pol m,n,q , and we use the notation
for the corresponding Berezin measure and Berezin density. Interestingly, the microscopic behavior of the Berezin measure \({\mathrm{d}}B^{\langle z\rangle}_{m,n,q}\) is quite different for q>1 compared with the Ginibre case q=1. In terms of the blow-up Berezin density (ΔQ(z)=Δ|z|2≡1 here)
we have the following asymptotics as m,n→+∞ while n=m+O(1):
where \(L^{1}_{q-1}\) denotes the generalized Laguerre polynomial of degree q−1 with parameter 1 (see (2.1) for the explicit expression). It is well-known that the Laguerre polynomial \(L^{1}_{q-1}\) has exactly q−1 strictly positive roots, which implies that the Berezin density will exhibit a typical Fresnel-type ring pattern (see Fig. 2). Taking (1.6) into account, we see that the zero rings of the Berezin kernel correspond to circles centered at z where in the limit as m,n→+∞ no repulsion occurs between the center z and points on those circles.
This resembles what happens in the one-dimensional GUE case, where the zero density points for the Berezin density come from the zeros of the sine kernel. Actually, the analogy is more than a superficial similarity. If we consider rather big values of q, and scale down to local distance (mq)−1/2, with
then by the above we have, for \(z\in{\mathbb{D}}\),
as m,n→+∞ with n=m+O(1). Next, if we let q→+∞, we get that (see [26] 18.11.5)
where J 1 is the standard Bessel function. The identity
shows that we are dealing with a planar analogue of the sine kernel (the sine kernel is the Fourier transform of the characteristic function of the interval [−1,1], the one-dimensional unit ball). We remark here that the asymptotics (1.8) which is obtained here in the model case Q(z)≡|z|2 can be shown to hold at bulk points for rather general weights Q, see [18].
We also investigate the local behaviour of the Berezin transform when |z|=1, i.e., when z is on the boundary of the bulk. Using the same blow-up scale as with an interior point, we show that the blow-up Berezin density tends to a limit which can be expressed in terms of Hermite polynomials (see Theorem 5.10). Here, too, there is a ring-like pattern in the interior direction, but it is not so pronounced as it is for interior points (the bulk). We express the 1-point intensity near a boundary point in terms of a sum of squares of Hermite polynomials. The Wigner semi-circle law then gives the asymptotic behavior of the 1-point function, which tells us the intensity of the process. We find that for big q, but much bigger m,n with m=n+O(1), the 1-point function is nontrivial in the annulus
inside it is essentially constant, and outside it more or less vanishes. Near the outward boundary of the annulus at the scale (mq)−1/2, we expect the statistics of the point process to be related with the well-known Airy point process (in the sense of Tracy and Widom).
We should mention that the model of the present paper has been studied previously in physics literature, see Dunne [13]. However, as far as we know, the exact scaling limits of the blow-up correlation kernels were not obtained earlier.
1.9 Lifting to Two Complex Variables
Analogous results to [3] are obtained on complex manifolds by Berman [8]. We note that the polyanalytic Ginibre processes can also be viewed as processes in \({\mathbb{C}}^{2}\) with the rather singular weight \(\mathrm{e}^{-|z_{1}|^{2}}\delta_{0}(z_{1}-\bar{z}_{2})\), where δ 0 is unit point mass at 0. Berman considers reproducing kernels of polynomial subspaces as the total degree of the polynomials tends to infinity. In contrast, here we discuss the case where one variable has bounded degree and the degree of the other variable goes to infinity.
1.10 The Polyanalytic Ginibre Ensemble and Weighted Interpolation
It is well known in the theory of random matrices that for complex numbers z 1,…,z n , we have
where Δ(z 1,…,z n )=Π i,j:i<j (z j −z i ) is the van der Monde determinant. A point configuration in a compact set which maximizes the van der Monde determinant is known to be a good choice of nodes for Lagrange interpolation [34]. Instead of considering points confined to a compact set, one can add a weight to prevent the points from going off to infinity. This leads to the same expression which arises in the context of random eigenvalues.
We turn to the polyanalytic case. One shows that
where Z m,n,q is a normalization constant
and
So, Δ q (z 1,…,z nq ) is the polyanalytic analogue of the van der Monde determinant. As in the case q=1 (which gives the usual van der Monde determinant), the expression |Δ q (z 1,…,z nq )|2 measures how good the configuration is for Lagrange interpolation by polyanalytic polynomials. So, the polyanalytic Ginibre ensemble is a way to produce random Lagrange interpolation sets, using the Gaussian weight for confinement.
1.11 Further Results and Open Problems
In [33] and [4], the authors showed that the fluctuations of eigenvalues of random normal matrices tend to Gaussian free field. The fluctuations of the polyanalytic Ginibre process will be discussed in a separate paper—the limit is again the Gaussian free field, but the variance depends on the degree of polyanalyticity.
One could naturally address all the questions discussed in this article with a more general weight Q. We conjecture that the spectral droplet will be the same as in the analytic case, i.e. that the points from the process will accumulate on the set \(\mathcal{S}\), with density given by the equilibrium measure \(1_{\mathcal{S}}\Delta Q{\mathrm{d}}A\). It is also likely that the blow-up of the Berezin density at a bulk point z will have a universal limit, which we here computed to be \(q^{-1}L^{\langle1\rangle}_{q-1}(|\xi|^{2})^{2}\mathrm{e}^{-|\xi|^{2}}\).
2 Polyanalytic Bargmann-Fock Spaces
2.1 An Orthogonal Basis
We will consider the Bargmann-Fock space \(A^{2}_{m,q}({\mathbb{C}})\) of poly-analytic functions of degree ≤q−1, i.e., functions of the form
where all the components f r are entire, subject to the norm integrability condition
Here, as always, m>0. We note that \(A^{2}_{1,1}({\mathbb{C}})\) is the standard Bargmann-Fock space. As before, let Pol m,n,q be the closed subspace of \(A^{2}_{m,q}({\mathbb{C}})\), defined by the condition that all the components f r are polynomials of degree less or equal to n−1. Moreover, let K m,q and K m,n,q denote the reproducing kernels for \(A^{2}_{m,q}({\mathbb{C}})\) and Pol m,n,q , respectively. We will be concerned with the asymptotic behaviour of the kernel K m,n,q , as m,n→+∞ while n=m+O(1).
Bargmann-Fock-spaces of polyanalytic functions have been considered in, e.g., [36], where the reproducing kernels and orthonormal bases were identified. We will attempt to supply a self-contained account of these basic matters.
To begin with, we identify an orthonormal basis for the space Pol m,n,q . Here, and later as well, we will need some standard properties of the classical orthogonal polynomials. For details, we refer the reader to [26]. We need the generalized Laguerre polynomials ((x) j :=x(x+1)⋯(x+j−1) is the Pochhammer symbol)
Proposition 2.1
For q≤n, the following functions form an orthonormal basis for Pol m,n,q (i,r,j,k are all integer parameters):
Proof
Clearly, all the above functions belong to the space Pol m,n,q . Also, all the functions \(e^{1}_{i,r}\) are orthogonal to each of the functions \(e^{2}_{j,k}\), for the indicated ranges of the indices, as can be seen by integrating along circles using polar coordinates. Next, we show that the functions \(e^{1}_{i,r}\) form an orthonormal set. Any two functions there having different parameter i are orthogonal, again by integrating along circles using polar coordinates. So, we fix i, and pick two indices r 1,r 2. We compute the inner product of two such functions:
where the delta is in Kronecker’s sense. In a similar fashion, the functions \(e^{2}_{j,k}\) form an orthonormal set. So, the functions \(e^{1}_{i,r},e^{2}_{j,k}\) together form an orthonormal set. Next, the dimension of the span equals the total number of vectors, which we calculate to nq, which equals the known dimension of the space Pol m,n,q . The proof is complete. □
2.2 The Reproducing Kernel
For q≤n, we conclude that the reproducing kernel of Pol m,n,q equals
Remark 2.2
In [2], it is pointed out that Perelomov in his book [32], p. 35, mentions that the formula (2.2)—viewed as the explicit expression for the matrix elements of the displacement operator—had been used by Feynman and Schwinger in a somewhat different form. For this reason, Abreu and Feichtinger [2] suggest naming the orthogonal basis of Proposition 2.1 the Feynman-Schwinger basis for the polyanalytic Bargmann-Fock space.
We note that by plugging in n=+∞ in Proposition 2.1 we get an orthonormal basis for the space \(A^{2}_{m,q}({\mathbb{C}})\). It follows that the same procedure of plugging in n=+∞ in the above expression for K m,n,q gives us K m,q , the reproducing kernel for \(A^{2}_{m,q}({\mathbb{C}})\). What is probably less obvious is that K m,q may be written in a much simpler form (this representation is, however, known; compare with [7, 31]).
Proposition 2.3
We have that
Proof
It is clear from Proposition 2.1 that the double sum expression equals K m,q (z,w), so the only thing that needs attention is the last equality. We first do the case w=0. Many of the terms in the double sum vanish, so that we are left with
It is well-known that \(L_{r}^{0}(0)=1\) and that \(\sum_{r=0}^{q-1} L^{\alpha}_{r}=L^{\alpha+1}_{q-1}\), so that the above reduces to
We turn to general \(w \in{\mathbb{C}}\). It is well-known that the transformation
acts unitarily on \(A^{2}_{m,q}({\mathbb{C}})\); its adjoint is \({\mathbf{T}} _{w}^{*}={\mathbf{T}}_{-w}\). By the reproducing property of K m,q (z,0), we have, for \(f\in A^{2}_{m,q}({\mathbb{C}})\),
The claim that \(K_{m,q}(z,w)= m\,L^{1}_{q-1}(m|z-w|^{2}) \mathrm{e}^{mz\bar{w}}\) now follows immediately. □
2.3 Szegö Asymptotics
We shall show that the kernel K m,n,q (z,w) is approximated well by K m,q (z,w), provided that \(z,w\in{\mathbb{D}}\) are rather close to one another. It will be instrumental to study the asymptotic behavior of the partial sums
of the Taylor expansion of the exponential function. In [35], Szegö showed that
provided that |ζe1−ζ|<1 and |ζ|<1, whereas
provided that |ζe1−ζ|<1 and |ζ|>1. Here, we have the convergence
uniformly on compact subsets of the respective domains. As for (2.4), the uniform convergence \(\epsilon^{2}_{k}(\zeta)\to0\) as k→+∞ holds also in certain unbounded subdomains; in particular, along the real line, we have uniform convergence on all intervals [a,+∞[ with a>1.
2.4 An Elementary Estimate of Laguerre Polynomials
The following elementary estimate of generalized Laguerre polynomials will prove useful.
Lemma 2.4
Suppose α is a positive real. Then, for k=0,1,2,…, we have that
Moreover, with \(\beta:=\alpha+2k-2+2\sqrt{\frac{1}{4}+(k-1)(k+\alpha-1)}\), we have that
Actually, the related inequality \(|L_{k}^{\alpha}(x)|\le\frac {1}{k!}x^{k}\) holds for all \(x\in[\frac{1}{2}\beta,+\infty\mathclose{[}\).
Proof
We begin with the estimate (0≤i≤k is assumed)
Next, we note that for x≥0 and k=0,1,2,…, we have
and in view of the above estimate,
and, analogously,
By discarding alternatively the even or odd contribution, we arrive at
which is slightly stronger than the first estimate.
It is well-known that \(L_{k}^{\alpha}\) is a polynomial of degree k all of whose zeros are distinct and real, and they all fall in the interval ]0,β[ (see 18.16.13 of [26]). Therefore, we can write
where β 1<⋯<β k are the zeros of \(L^{\alpha}_{k}\). Now, all the zeros satisfy x−β j ≥x−β≥0 for x≥β and |x−β j |≤x for x≥β/2. The remaining estimates of the proposition follow immediately from combining these observations with the representation (2.5). □
2.5 Approximation of the Polynomial Polyanalytic Reproducing Kernel
We estimate the difference K m,n,q −K m,q .
Proposition 2.5
Let \(z,w\in{\mathbb{C}}\) be such that |zw|≤θ 1<1 and |zw|e1−|zw|≤θ 2<1. Let M be a positive real number. Then, as m,n→+∞ while |m−n|≤M,
where the constant C depends on q and M.
Proof
In view of (2.2) and Proposition 2.3, we have that for q≤n,
By Lemma 2.4, we get that
For r confined to 0≤r≤q−1 and for big n, say n≥n 0(q), we have
so that if we use the notation
we obtain from (2.6) that for 0≤r≤q−1 and n≥n 0(q),
Next, we see from Szegö’s asymptotical expansion (2.3) that for large k,l with l=k+O(1), we have
as k,l→+∞ and l=k+O(1). By a careful application of (2.8) to (2.7), and summing over 0≤r≤q−1, the assertion of the proposition follows. □
Corollary 2.6
For \(z,w\in{\mathbb{D}}\), let 1−|zw|≥τ>0. Then, as m,n→+∞, while |m−n|≤M,
where the “O” constant depends on τ, q, and M.
Proof
A Taylor series expansion of the logarithm gives that
and, together with the fact that the exponential function grows faster than any given power, the assertion follows from Proposition 2.5. □
As we shall see, Proposition 2.5 implies that the Berezin density
behaves locally near z like the Berezin density
To make this precise, we recall the definition of the blow-up Berezin density:
Proposition 2.7
Fix \(z\in{\mathbb{D}}\). Then
as m,n→+∞ while n=m+O(1).
Proof
By (2.9), we have
so the comparison is with the blow-up Berezin density for K m,q . We write w:=z+m −1/2 ξ; since \(z\in{\mathbb{D}}\) is fixed, we have 1−|z|2≥τ>0 for some small τ, and we suppose that w is close to z so that 1−|zw|≥τ>0 as well. Then, by Proposition 2.3 and Corollary 2.6,
so that if we expand the square using that
where the “O” depends only on q (this follows from Lemma 2.4), and simplify the expression, we arrive at
which immediate gives that
The corresponding blow-up Berezin density then has
As exponentials grow faster than polynomials, the error terms is negligible for big m. For fixed \(z\in{\mathbb{D}}\), the requirement on ξ so that 1−|zw|≥τ for some fixed τ>0 is fulfilled for big m if, say, |ξ|≤logm is required. So, (2.10) has the immediate consequence that
where more generally \({\mathbb{D}}(z_{0},\rho)\) denotes the open disk of radius ρ about z 0. Since the associated blow-up Berezin measures
are both probability measures, the assertion of the proposition follows from (2.11) once it is noted that the right-hand side of (2.11) tends to 0 as m→+∞. □
We note that the claimed convergence \({\mathrm{d}}B^{\langle z\rangle }_{m,n,q}\to {\mathrm{d}}\delta_{z}\) for \(z\in{\mathbb{D}}\) is an immediate consequence of Proposition 2.7.
3 Berezin Density Asymptotics for an Exterior Point
3.1 Convergence to Harmonic Measure
We show that the Berezin measures have the convergence \({\mathrm{d}}B^{\langle z\rangle}_{m,n,q}\to{\mathrm{d}}\omega_{z}\) for \(z\in {\mathbb{D}}^{e}\) as m,n→+∞ while n=m+O(1). Here, ω z is harmonic measure with respect to the point z and the exterior disk \({\mathbb{D}}^{e}\).
3.2 Concentration of the Berezin Mass
We first study the concentration of the Berezin measure to neighborhoods of the closed unit disk \(\bar{\mathbb{D}}\). We recall the standard notation \({\mathbb{D}} (z_{0},\rho)\) for the open disk of radius ρ centered at z 0.
Lemma 3.1
Suppose \(z\in{\mathbb{D}}^{e}\) and that ρ>1. Then
as m,n→+∞ while n≤m+O(1).
Proof
Let \(w\in{\mathbb{C}}\setminus{\mathbb{D}}(0,\rho)\). We note that
for sufficiently big m, and so, by Lemma 2.4,
If we replace (n,r) by (q,j) in (3.3), we obtain
We may restrict to q≤n; after, we are considering the limit as n→+∞. As E q−1≤E n−1 on the positive half-axis for q≤n, an application of (3.3) and (3.4) to (2.2) gives
Next, we observe that Pol m,n,1⊂Pol m,n,q (since q≥1), which implies that
We conclude that the Berezin density may be estimated as follows:
Finally, we see from Szegö’s asymptotical expansion (2.4) that
Here, k≤l+O(1), where the convergence \(\epsilon^{2}_{k}(l\zeta/k)\to0\) is uniform if ζ is real with ζ≥a for some fixed a>1. This leads to
where the “o” term is uniform in the convergence. As we implement this estimate in (3.6), and integrate over \({\mathbb{C}} \setminus {\mathbb{D}}(0,\rho)\), the claim follows. □
3.3 A Principal Value Calculation
We follow the approach of [3], and calculate a certain principal value integral. As on the real line, principal value integrals in the complex plane are defined by first removing a disk of radius ϵ around the singularity and then letting ϵ→0.
Lemma 3.2
Fix \(z\in{\mathbb{D}}^{e}\). For any l=0,1,2,…, we have
as m,n→+∞ with n=m+O(1).
Proof
The case q=1 was treated in [3], so we may from now on assume that q≥2. We write
where
and
It follows that the expression |K m,n,q |2 decomposes accordingly:
We first consider the following integral involving \(|K_{m,n,q}^{I}|^{2}\):
where the delta is understood in Kronecker’s sense. The identity (for p=1,2,3,…)
plus the standard orthogonality properties of the Laguerre polynomials gives that
where the right-hand side should be interpreted as 0 for r 2<r 1. By polar coordinates, then, we have
and (3.9) simplifies to
provided n is so big that q+l≤n. Next, we apply Lemma 2.4 and (3.2) (using that \(z\in {\mathbb{D}}^{e}\)) to arrive at
On the other hand, by the estimate from below in Lemma 2.4,
provided that m|z|2≥β(n), where
As we assume \(z\in{\mathbb{D}}^{e}\) and n=m+O(1), we must have
for big enough m,n, and so, by (3.13),
Here, we used that the E n−1(m|z|2) is much bigger than E q−2(m|z|2) as m,n both grow. Let us look at the contribution from 0≤r 1≤q−2 in the right-hand side of (3.11) using the estimate (3.12):
for an appropriate positive constant C 1(q,l). As we combine this estimate with (3.14), we obtain
as m,n→+∞ in the prescribed fashion. So the contribution to (3.11) which comes from 0≤r 1≤q−2 is negligible from the point of view of the Berezin density. It remains to consider the contribution from r 1=q−1. The corresponding part of the sum in the right-hand side of (3.11) equals
and we now claim that
as m,n→+∞ in the given fashion. We use the recurrence relation (3.10) to write
A straightforward argument allows us to show that as insert this into (3.15), the sum that is subtracted on the right-hand side of (3.16) makes asymptotically no contribution to the sum in (3.15). Consequently, (3.15) is equivalent to having
as m,n→+∞ in the given fashion. If we insert the expression (2.2) defining K m,n,q (z,z), we see that (there is some cancellation of terms)
By careful application of Lemma 2.4 to all the Laguerre polynomials in this expression, while inserting (3.13) to control the denominator, we indeed get (3.17). So, after a lot of effort, we have obtained that
as m,n→+∞ with n=m+O(1). Analogous but slightly easier arguments (left to the interested reader) show that
and
again as m,n→+∞ with n=m+O(1). Finally, we put everything together based on the decomposition (3.8):
as m,n→+∞ with n=m+O(1). This ends the proof. □
3.4 Convergence to Harmonic Measure
We now show that even in the principal value sense, the Berezin density tends to avoid the interior of the unit disk.
Lemma 3.3
Fix a real parameter ρ with 0<ρ<1 and a point \(z\in {\mathbb{D}}^{e}\). Then, for l=0,1,2,…, we have the convergence
as m,n→+∞ with n=m+O(1).
Proof
In terms of the decomposition (3.8), we will focus on the term \(|K^{I}_{m,n,q}|^{2}\) and leave the other two to the reader (the necessary arguments are similar but slightly easier). The analogue of (3.9) reads
By Lemma 2.4, we have
In terms of the function
which takes values in the interval [0,1[, the estimate becomes
and if we use that a↦χ(a,b) is decreasing for fixed b (a direct calculation involving derivatives suffices to verify this), we get
Next, since \(z\in{\mathbb{D}}^{e}\), n=m+O(1), and i 1+l≤n−1, we may use another aspect of Lemma 2.4 to see that
provided m,n are big enough. As we combine the equality (3.20) with the estimates (3.21) and (3.22), and use some well-understood comparisons of factorials and powers, we arrive at
for some appropriate positive constant C 3(q,l). The function χ(i 1,mρ 2) drops off exponentially quickly to 0 as i 1 exceeds mρ 2 by a margin greater than O(m 1/2), as can be seen, e.g., by an application of the Central Limit Theorem (compare with the next section). This means that effectively we are summing up to m|ρ|2+O(m 1/2) in the right-hand side expression of (3.23), which does not permit the sum to compare with the size of K m,n,q (z,z); cf. (3.14). This results in the convergence
as m,n→+∞ with n=m+O(1). Together with the estimates which were left as an exercise to the reader, we get
as m,n→+∞ with n=m+O(1), which amounts to the assertion of the lemma. □
As in the proof of the Theorem 2.10 in [3], we may now conclude the following.
Theorem 3.4
Fix \(z\in{\mathbb{D}}^{e}\) and a bounded continuous function g on \({\mathbb{C}}\). Then
as m,n→+∞ with n=m+O(1). Here, \({\mathrm{d}}\omega(w,z,{\mathbb{D}} ^{e})\) is harmonic measure with respect to the point z and the domain \({\mathbb{D}}^{e}\).
Naturally, the harmonic measure in question can be written explicitly with the Poisson kernel:
Here, ds(w) denotes the arc-length measure on the unit circle.
4 Poly-Bargmann Transforms
4.1 Purpose of the Section
In this section, we discuss the poly-Bargmann transforms, a generalization of the classical Bargmann transform, which are needed later when we analyze the Berezin density at a boundary point. The poly-Bargmann transforms appeared in Vasilevski’s paper [36], where the basic properties were presented.
4.2 The Hermite Polynomials and the Bargmann Transform
We denote by H j the j-th Hermite polynomial with respect to the Gaussian weight \(\mathrm{e}^{-\frac{1}{2}t^{2}}\) (“probabilistic (monic) Hermite polynomials”). The generating function identity
allows us to write
We recall the standard definition of the Bargmann transform:
As the function systems
form orthonormal bases for \(L^{2}({\mathbb{R}})\) and the Bargmann-Fock space \(A^{2}_{1,1}({\mathbb{C}})\) (this is \(A^{2}_{m,q}({\mathbb{C}})\) with m=q=1), respectively, we obtain the following well-known fact.
Proposition 4.1
The Bargmann transform \({\mathbf{B}}:L^{2}({\mathbb{R}})\to A_{1,1}^{2}({\mathbb{C}})\) acts isometrically and bijectively, and for each j=0,1,2,…, the basis function \((2\pi)^{-1/4}(j!)^{-1/2}H_{j}(t)\mathrm{e}^{-\frac{1}{4}t^{2}}\) is mapped to the basis function (j!)−1/2 z j.
4.3 A Class of Auxiliary Operators
Let \(\partial_{z},\bar{\partial}_{z}\) denote the standard Wirtinger differential operators
For r=0,1,2,…, we introduce the operator
with the semi-group property T 1∘T r−1=r 1/2 T r . We also consider the dilated variant
where \(f_{m^{-1/2}}(z)=f(m^{-1/2}z)\). It has the semi-group property T m,1∘T m,r−1=r 1/2 T m,r , and may be expressed in more concrete terms:
We now study the effect of T r on the basis elements (j!)−1/2 z j.
Proposition 4.2
For j≥r, we have
while for j≤r,
Proof
The proof is based on an induction argument. The statement is obviously true for r=0 and all j. So, by induction, we assume that the statement holds for some r−1≥0 and all j. In case j≥r, we then have
if we use the standard identity \(rL^{\alpha}_{r}(x) = (\alpha+ 1 -x) L^{\alpha+1}_{r-1}(x) - xL^{\alpha+2}_{r-2}(x)\). Next, in case j≤r−1, we have instead (see [17], 8.971(5))
which completes the proof. □
4.4 Pure Poly-analytic Fock Spaces
Write e j (z):=(j!)−1/2 z j for j=0,1,2,…, which functions form the standard orthonormal basis for the space \(A^{2}_{1,1}({\mathbb{C}})\). The function \({\mathbf{T}}_{r}[e_{j}]\in A^{2}_{1,r+1}({\mathbb{C}})\) is then a polynomial in \(z,\bar{z}\), where the degree in z remains equal to j, and the degree in \(\bar{z}\) equals r. For general positive m, the functions \(e_{j,m}(z):= (j!)^{-1/2}m^{\frac{1}{2}(j+1)}z^{j}\), with j=0,1,2,…, form an orthonormal basis for the space \(A^{2}_{m,1}({\mathbb{C}})\). The functions T m,r [e j,m ] are computed using Proposition 4.2 above, and we then recognize that we can identify them with the basis elements which appear in Proposition 2.1. We clearly have that
and in view of the orthogonality properties in Proposition 2.1, we also must have
where the “⊖” is with respect to the inner product in \(L^{2}({\mathbb{C}},\mathrm{e}^{-m|z|^{2}})\). Following Vasilevski [36], then, we define the pure (true) polyanalytic Fock space of level r+1:
with the understanding that Pol m,n,0:={0} and \(A^{2}_{m,0}({\mathbb{C}}):=\{0\}\). The operator T m,r now becomes an isometric isomorphism
We obtain the orthogonal decompositions
As a consequence, if K δ:m,n,r+1 and K δ:m,r+1 denote the reproducing kernels for the spaces δPol m,n,r+1 and \(\delta A^{2}_{m,r+1}({\mathbb{C}})\), respectively, we must have that
4.5 The Poly-Bargmann Transforms
For m=1, the poly-Bargmann transform of level r is defined by
Proposition 4.3
We have
Proof
We proceed by induction. For r=0, the formula reduces to the usual Bargmann transform (if we recall that H 0=1). By the induction hypothesis, we suppose therefore that the formula is valid for all integers up to r−1. From the semi-group property T 1∘T r−1=r 1/2 T r , we see that
Since the formula holds for r−1, the following calculation shows that it holds for r as well:
here, we used the standard identity \(H_{r}(x)=xH_{r-1}(x)-H'_{r-1}(x)\). The proof is complete. □
5 Reproducing Kernel and Berezin Density Asymptotics for a Boundary Point
5.1 Purpose of the Section
In this section, we will calculate the limit of the blow-up Berezin transform at a boundary point z, that is, |z|=1. There is no loss of generality to take z=1. Our strategy is to investigate the blow-up of the reproducing kernel of the space of analytic polynomials Pol m,n,1 (with q=1) first, and then use this information together with poly-Bargmann transform to lift the asymptotics to the context of the general polyanalytic spaces Pol m,n,q .
5.2 The Central Limit Theorem Revisited
The following improvement of the central limit theorem will be needed. Let cdf X denote the cumulative distribution function of a real-valued random variable X and let
be the error function. Here, for complex z, we may think of the contour of integration as following the real line from −∞ to 0, and then from going from 0 to z along any smooth curve.
We shall write i.i.d. as shorthand for independent identically distributed in the context of random variables. The following result is from [9, 14].
Theorem 5.1
(Berry-Esséen)
Let X 1,X 2,… be i.i.d. real-valued random variables with E(X j )=0, \(E(X_{j}^{2})=1\) and E(|X j |3)=ρ<+∞ for all j. Also, let \(Y_{n} = n^{-1/2}\sum_{j=1}^{n} X_{j}\). Then there exists an absolute constant C such that
The Berry-Esséen theorem gives the following asymptotics for the partial Taylor sums of the exponential function.
Lemma 5.2
We have
as n→+∞, uniformly in x∈[0,+∞[.
Proof
Let X 1,…,X n be independent exponentially distributed random variables on [0,+∞[ with density e−x. It is well known that the sum \(\sum_{j=1}^{n}X_{j}\) obeys a gamma distribution with the cumulative distribution function 1−e−x E n−1(x). The random variables X 1−1,…,X n −1 all have zero mean and variance 1, and the third moment is finite, so by the Berry-Esséen theorem, we have
where the “O” term is uniform in x as n→+∞. □
Results of this nature are well known. A reader who prefers a more direct approach might use Taylor’s formula with remainder term to obtain
and then analyze the integral by Laplace’s method.
This allows us to blow up the function \(E_{n-1}(mz\bar{w})\) when z and w are close to the point 1 and m,n→+∞ with n=m+O(1).
Lemma 5.3
Fix a positive real ε. For complex \(\xi,\eta\in{\mathbb{C}}\), we then have
as m,n→+∞ while n=m+O(1). Here, the “O” expression on the right-hand side is uniform on compact subsets of \({\mathbb{C}}\).
Proof
If we put
then
where the “O” term is uniform on compact subsets. If we use that E n−1 has only nonnegative Taylor coefficients, we get from Lemma 5.2 that
with a uniform “O” term. So, for real ξ,η, we may deduce from (5.2) the assertion of the lemma with ε=0. For general complex ξ,η, we note that
uniformly in the domain where max{|ξ|,|η|}≤m 1/8. As the right-hand side of (5.2) is \(\le\frac{3}{2}\) for big n, we see that
holds in the domain where max{|ξ|,|η|}≤m 1/8, provided m is big enough. We need to show that the difference
is of order O(m −1/2+ε) uniformly as ξ,η remain confined to some compact subset of \({\mathbb{C}}\). We know that F m,n (ξ,η)=O(m −1/2) uniformly when \(\xi,\eta \in{\mathbb{R}} \) with confined to max{|ξ|,|η|}≤m 1/8. In view of the calculation we just made, we also have a good uniform estimate of F m,n (ξ,η) when \(\xi,\eta\in{\mathbb{C}}\) with max{|ξ|,|η|}≤m 1/8. With an elementary estimate of harmonic measure (see exercise II.3a in [16]), we can show that F m,n (ξ,η)=O(m ϵ−1/2) holds uniformly when \(\xi,\eta\in{\mathbb{C}}\) belong to a compact subset of \({\mathbb{C}}\), and in addition, \(\eta\in{\mathbb{R}}\). Here, ϵ is a positive number which we can get as small as we like. A similar argument with η in place ξ worsens the control to F m,n (ξ,η)=O(m 2ϵ−1/2), but now the control is uniform when both ξ,η are both complex and confined to some compact subset. The proof is complete with ε=2ϵ. □
5.3 The Reproducing Kernel for a Subspace of the Fock Space
We shall identify both the right-hand and the left-hand side expressions appearing in Lemma 5.3 with reproducing kernels of certain Hilbert spaces of entire functions. This will be the case r=0 of the proposition below.
Let us agree to identify
In the following proposition and later, we encounter integrals ranging from −∞ to some complex number w. We can use any contour that first goes from −∞ to 0 along the real axis and then connects 0 to w. Our integrands are usually holomorphic, so we are allowed us to perturb the contour of integration when needed.
Proposition 5.4
For r=0,1,2,…, the function
is the reproducing kernel for the Hilbert space \({\mathbf{B}}_{r}[L^{2}({\mathbb{R}} _{-})]\subset A^{2}_{1,r+1}({\mathbb{C}})\).
Proof
Let M(ξ,η)=M η (ξ) be the reproducing kernel for \({\mathbf{B}}_{r}[ L^{2}({\mathbb{R}}_{-})]\). This kernel has \(M_{\eta}\in {\mathbf{B}}_{r}[L^{2}({\mathbb{R}} _{-})]\) and
for all \(\eta\in{\mathbb{C}}\) and all \(f\in L^{2}({\mathbb{R}}_{-})\), which allows us to conclude that
After applying the operator B r to both sides, we see that (cf. Proposition 4.3)
□
Remark 5.5
The special case r=0 of the kernel in Proposition 5.4 is \(\mathrm{e}^{\xi\bar{\eta}}\mathrm{erf}(-\xi-\bar{\eta})\), which is what we have on the right-hand side in Lemma 5.3.
5.4 The Blow-Up of the Polynomial Space at the Boundary Point
We turn to the polyanalytic analogue of the left-hand side of Lemma 5.3.
Definition 5.6
We introduce the blow-up space at 1,
which we equip with the norm
For 0≤r≤q−1, we denote by \({\delta\mathrm{Pol}}^{\langle 1\rangle }_{m,n,r+1}\) the subspace
equipped with the same norm.
An elementary change of variables argument allows us to identify the norm on \(\mathrm{Pol}^{\langle1\rangle}_{m,n,q}\) with that of \(A^{2}_{1,q}({\mathbb{C}})\):
As a consequence, we may regard \(\mathrm{Pol}^{\langle1\rangle}_{m,n,q}\) and \({{\delta\mathrm{Pol}}}^{\langle1\rangle}_{m,n,r+1}\) as norm closed subspaces of \(A^{2}_{1,q}({\mathbb{C}})\). We may read off from the definition of the norm in \({\mathrm{Pol}}^{\langle1\rangle}_{m,n,q}\) that the kernel on the left-hand side in Lemma 5.3 is the reproducing kernel for the space \(\mathrm{Pol}^{\langle1\rangle}_{m,n,1}\). So, Lemma 5.3 can be understood as saying that
where \(K_{\mathrm{Pol}_{m,n,1}}\) and \(K_{{\mathbf{B}}_{0}[L^{2}({\mathbb{R}}_{-})]}\) denote the reproducing kernels of the spaces in the subscripts, and the bound is locally uniform on compact subsets. We want to generalize (5.4) beyond q=1. To this end, we make use of the operators T r .
Proposition 5.7
For r=1,2,3,…, we have that
is an isometric isomorphism.
Proof
From the isometry properties of T r−1 and T r together with the semi-group property r −1/2 T 1∘T r−1=T r , we get that
is an isometric isomorphism. In view of (5.3), the isometry part of the assertion follows. It remains to show that the operator is onto. This is an algebraic exercise which we leave to the reader. □
By iterating Proposition 5.7, we obtain the following.
Corollary 5.8
For r=1,2,3…, we have that
is an isometric isomorphism.
5.5 The Blow-Up of the Polynomial Reproducing Kernel at a Boundary Point
From Corollary 5.8 above, we get that
where the subscripts z and w are used to indicate that the operator is acting with respect to that variable, and the bar means complex conjugation of the operator. In the above formula, T r is as before, and \(\bar{\mathbf{T}}_{e}\) is given by
We would like to plug in the approximation (5.4) into (5.5). We recall that the operator T r is given by the analogous formula, which expands as a sum of certain polynomials in \(\bar{z}\) of degree ≤r times powers of the differential operator ∂ z of degree ≤r (a similar assertion can be made about \(\bar{\mathbf{T}}_{r}\) as well). The Cauchy integral formula allows us to control the size of the derivatives on a compact subset in terms of the size of the functions on a slightly bigger compact subset. This means that the approximation (5.4) carries over, and we find that
with uniform control on compact subsets. Next, as
we get that
again with uniform control on compact subsets. Next, it should be rather clear that
is the reproducing kernel for the space \({\mathbf{T}}_{r}{\mathbf{B}}_{0}[L^{2}({\mathbb{R}} _{-})]={\mathbf{B}}_{r} [L^{2}({\mathbb{R}}_{-})]\), which was identified in terms of Hermite polynomials back in Proposition 5.4. We write this down as a proposition.
Proposition 5.9
Fix a positive real number ε. Then the reproducing kernel for \(\mathrm{Pol}^{\langle1\rangle}_{m,n,q}\) has the following form:
as m,n→+∞ while n=m+O(1), where the control is uniform on compact subsets.
If we like, we may use the classical Christoffel-Darboux identity
to rewrite the above sum. Also, we should note that reproducing kernel for the blow-up space \(\mathrm{Pol}^{\langle1\rangle}_{m,n,q}\) is connected with the reproducing kernel K m,n,q for Pol m,n,q via the identity
5.6 The Blow-Up of the 1-Point Intensity Near a Boundary Point
The 1-point intensity function is
and the localized version with z=1+m −1/2 ξ is
where we throw in a factor of m −1 to compensate for the expected number of Jacobian. In view of (5.9) together with Proposition 5.9, we obtain
So, essentially, the 1-point intensity function is determined by the density
which corresponds to filling the lowest energy eigenstates of the harmonic oscillator. By the Wigner semi-circle law, then, we get the approximation
valid for big m,n with n=m+O(1), and big q (but much smaller than m,n). So, if we rescale to characteristic distance q 1/2 m −1/2 we find an interesting law in the limit. To be more precise, we find that
for ξ′ confined to a compact set with \(-1\le\operatorname{Re}\xi '\le1\). Here, it is assumed that the limit m,n→+∞ while m=n+O(1) is taken first, and the limit q→+∞ is taken afterwards. Asymptotically, then, (5.10) asserts that the one-point function is the definite integral of the semi-circle function.
5.7 The Blow-Up Berezin Density at a Boundary Point
The blow-up Berezin density at 1 is given by
From (5.9), we have that
while Proposition 5.9 gives
as m,n→+∞ with n=m+O(1). Now, as each Hermite polynomial H r is either even or odd,
which leads to
and
A similar calculation gives that
Putting things together, we obtain the following asymptotics for the blow-up Berezin density.
Theorem 5.10
Fix a positive real number ε. Then the blow-up Berezin density at 1 has the following form:
as m,n→+∞ while n=m+O(1), where the constant of the error term is uniform on compact subsets.
Remark 5.11
When we make some explicit calculations based on Theorem 5.10, we see that the Fresnel zone pattern is less pronounced for a boundary point z.
References
Abreu, L.D.: Sampling and interpolation in Bargmann-Fock spaces of polyanalytic functions. Appl. Comput. Harmon. Anal. 29, 287–302 (2010)
Abreu, L.D., Feichtinger, H.G.: Function Spaces of Polyanalytic Functions. Harmonic and Complex Analysis and Its Applications. Springer, Berlin (2013). 32 pp., in press
Ameur, Y., Hedenmalm, H., Makarov, N.: Berezin transform in polynomial Bergman spaces. Commun. Pure Appl. Math. 63(12), 1533–1584 (2010)
Ameur, Y., Hedenmalm, H., Makarov, N.: Fluctuations of eigenvalues of random normal matrices. Duke Math. J. 159, 31–81 (2011)
Ameur, Y., Hedenmalm, H., Makarov, N.: Random normal matrices and Ward identities. Submitted
Anderson, G.W., Guionnet, A., Zeitouni, O.: An Introduction to Random Matrices. Cambridge Studies in Advanced Mathematics, vol. 118. Cambridge University Press, Cambridge (2010)
Askour, N., Intissar, A., Mouayn, Z.: Espaces de Bargmann généralisés et formules explicites pour leurs noyaux reproduisants. C. R. Math. Acad. Sci. Sér. I Math. 325(7), 707–712 (1997)
Berman, R.: Determinantal point processes and fermions on complex manifolds: bulk universality. arXiv:0811.3341v1
Berry, A.C.: The accuracy of the Gaussian approximation to the sum of independent variates. Trans. Am. Math. Soc. 49(1), 122–136 (1941)
Borodin, A.: Determinantal point processes. arXiv:0911.1153v1
Deift, P.A.: Orthogonal Polynomials and Random Matrices: A Riemann-Hilbert Approach. Courant Lecture Notes in Mathematics, vol. 3. New York University, Courant Institute of Mathematical Sciences/American Mathematical Society, New York/Providence (1999)
Deift, P., Gioev, D.: Random Matrix Theory: Invariant Ensembles and Universality. Courant Lecture Notes in Mathematics, vol. 18. Courant Institute of Mathematical Sciences/American Mathematical Society, New York/Providence (2009)
Dunne, G.V.: Edge asymptotics of planar electron densities. Int. J. Mod. Phys. B 8, 1625–1638 (1994)
Esséen, C.-G.: On the Liapunoff limit of error in the theory of probability. Ark. Mat. Astron. Fys. A 28(9), 1–19 (1942)
Forrester, P.J.: Log-Gases and Random Matrices. London Mathematical Society Monograph Series, vol. 34. Princeton University Press, Princeton (2010)
Garnett, J., Marshall, D.: Harmonic Measure. Cambridge University Press, Cambridge (2005)
Gradshteyn, I.S., Ryzhik, I.M.: Table of Integrals, Series, and Products. Academic Press, New York (1980)
Haimi, A., Hedenmalm, H.: Asymptotic expansion of polyanalytic Bergman kernels. Submitted
Hedenmalm, H., Olofsson, A.: Hele-Shaw flow on weakly hyperbolic surfaces. Indiana Univ. Math. J. 54(4), 1161–1180 (2005)
Hedenmalm, H., Makarov, N.: Quantum Hele-Shaw flow. arXiv:math/0411437
Hedenmalm, H., Makarov, N.: Coulomb gas ensembles and Laplacian growth. Proc. Lond. Math. Soc. (3) 106, 859–907 (2013)
Hedenmalm, H., Perdomo-G., Y.: Mean value surfaces with prescribed curvature form. J. Math. Pures Appl. 83(9), 1075–1107 (2004)
Hedenmalm, H., Shimorin, S.: Hele-Shaw flow on hyperbolic surfaces. J. Math. Pures Appl. (9) 81(3), 187–222 (2002)
Hough, J.B., Krishnapur, M., Peres, Y., Virág, B.: Determinantal processes and independence. Probab. Surv. 3, 206–229 (2006)
Hough, J.B., Krishnapur, M., Peres, Y., Virág, B.: Zeros of Gaussian Analytic Functions and Determinantal Point Processes. University Lecture Series, vol. 51. American Mathematical Society, Providence (2009)
Koornwinder, T.H., Wong, R.S., Koekoek, R., Swarttouw, R.F.: Orthogonal Polynomials. NIST Handbook of Mathematical Functions, pp. 435–484. U.S. Dept. of Commerce, Washington (2010)
Levin, E., Lubinsky, D.: On the airy reproducing kernel, and associated sampling series and quadrature formula. Integral Equ. Oper. Theory 63, 427–438 (2009)
Macchi, O.: The coincidence approach to stochastic point processes. Adv. Appl. Probab. 7, 83–122 (1975)
Mehta, M.L.: Random Matrices, 3rd edn. Pure and Applied Mathematics (Amsterdam), vol. 142. Elsevier/Academic Press, Amsterdam (2004)
Mouayn, Z.: Coherent state transforms attached to generalized Bargmann spaces on the complex plane. Math. Nachr. 284(14–15), 1948–1954 (2011)
Mouayn, Z.: New formulae representing magnetic Berezin transforms as functions of the Laplacian on \({\mathbb{C}}^{N}\). arXiv:1101.0379
Perelomov, A.: Generalized Coherent States and Their Applications. Texts and Monographs in Physics. Springer, Berlin (1986)
Rider, B., Virág, B.: The noise in the circular law and the Gaussian free field. Int. Math. Res. Not. 2, mm006 (2007). 33 pp.
Saff, E., Totik, V.: Logarithmic Potentials with External Fields. Grundlehren der mathematischen Wissenschaften, vol. 316. Springer, Berlin (1997)
Szegö, G.: Über eine Eigenschaft der Exponentialreihe. Sitzungber. Berlin Math. Gesellschaftwiss. 23, 50–64 (1923)
Vasilevski, N.L.: Poly-Fock spaces. In: Differential Operators and Related Topics, Vol. I, Odessa 1997. Oper. Theory Adv. Appl., vol. 117, pp. 371–386. Birkhäuser, Basel (2000)
Zabrodin, A.: Matrix models and growth processes: from viscous flows to the quantum Hall effect. In: Applications of Random Matrices in Physics. NATO Sci. Ser. II Math. Phys. Chem., vol. 221, pp. 261–318. Springer, Dordrecht (2006). arXiv:hep-th/0412219v1
Zabrodin, A.: Random matrices and Laplacian growth. arXiv:0907.4929v1 [math-ph]
Acknowledgements
We would like to thank Philip Kennerberg for sharing his MATLAB-code and discussing the practicalities in simulating the process. We also thank the referees for helpful suggestions.
Author information
Authors and Affiliations
Corresponding author
Additional information
The second author is supported by the Göran Gustafsson Foundation (KVA) and Vetenskapsrådet (VR).
Rights and permissions
About this article
Cite this article
Haimi, A., Hedenmalm, H. The Polyanalytic Ginibre Ensembles. J Stat Phys 153, 10–47 (2013). https://doi.org/10.1007/s10955-013-0813-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10955-013-0813-x