# Let arXiv:2004.06968v1 [math.PR] 15 Apr 2020 â€؛ pdf â€؛ 2004.06968.pdf AND MARTIN BOUNDARY...

date post

28-Jan-2021Category

## Documents

view

0download

0

Embed Size (px)

### Transcript of Let arXiv:2004.06968v1 [math.PR] 15 Apr 2020 â€؛ pdf â€؛ 2004.06968.pdf AND MARTIN BOUNDARY...

ASYMPTOTIC BEHAVIOR OF THE OCCUPANCY DENSITY FOR

OBLIQUELY REFLECTED BROWNIAN MOTION IN A HALF-PLANE

AND MARTIN BOUNDARY

PHILIP A. ERNST AND SANDRO FRANCESCHI

Abstract. Let π be the occupancy density of an obliquely reflected Brownian motion in the half plane and let (ρ, α) be the polar coordinates of a point in the upper half plane. This work determines the exact asymptotic behavior of π(ρ, α) as ρ→∞ with α ∈ (0, π). We find explicit functions a, b, c such that

π(ρ, α) ∼ ρ→∞

a(α)ρb(α)e−c(α)ρ.

This closes an open problem first stated by Professor J. Michael Harrison in August 2013. We also compute the exact asymptotics for the tail distribution of the boundary occupancy measure and we obtain an explicit integral expression for π. We conclude by finding the Martin boundary of the process and giving all of the corresponding harmonic functions satisfying an oblique Neumann boundary problem.

Contents

1. Introduction 1 2. A kernel functional equation 3 3. Boundary occupancy measure 7 4. Inverse Laplace transform 11 5. Saddle-point method and asymptotics 12 6. Martin boundary 16 Acknowledgments 17 References 18 Appendix A. Generalization of parameters 19 A.1. Generalization to arbitrary covariance matrix 19 A.2. Initial state x 19 A.3. Case µ2 > 0 20 Appendix B. Technical lemmas 20

1. Introduction

In 2013, Professor J. Michael Harrison raised a fundamental question regarding the asymptotic behavior of the occupancy density for reflected Brownian motion (RBM) in the half plane [10]. We shall state Harrison’s problem on the following page after introducing the necessary back- ground for the statement of the problem. The purpose of the present paper is to close this open problem.

Let B(t) + µt be a two-dimensional Brownian motion with identity covariance matrix, drift vector µ = (µ1, µ2), and initial state (0, 0).

1 Let R = (r, 1) be reflection vector and, for all t > 0,

2010 MSC Codes. Primary 60J60, 60K25; Secondary 90B22, 30D05. Key words and phrases. Occupancy density; Green’s function; Obliquely reflected Brownian motion in a half-

plane; Stationary distribution; Exact Asymptotics; Martin boundary; Laplace transform; Saddle-point method. 1Appendix A generalize our results to any covariance matrix and to any starting point.

1

ar X

iv :2

00 4.

06 96

8v 1

[ m

at h.

PR ]

1 5

A pr

2 02

0

2 PHILIP A. ERNST AND SANDRO FRANCESCHI

let

`(t) := − inf 06s6t

(B2(s) + µ2s) and Z(t) := B(t) + µt+R`(t) ∈ R× R+.

It is said that (Z, `) solves the Skorokhod problem for B(t)+µt with respect to upper half-plane and to R. The process Z is a reflected Brownian motion (RBM) in the upper half-plane and ` is the local time of Z on the abscissa. We shall assume throughout that

µ1 + rµ − 2 < 0,

2 (1)

ensuring that Z1(t)→ −∞ as t→∞ (see Appendix B, Lemma 15). Throughout this work, our primary concern shall be the case where

µ2 < 0. 3 (2)

Under (2), `(t) → ∞, µ−2 = −µ2, and (1) is equivalent to rµ2 − µ1 > 0. Figure 1 below gives two examples of parameters satisfying (1) and (2).

Figure 1. Two examples of parameters satisfying the inequality in (1) and (2). µ is the drift and R is the reflection vector

Let pt(z) denote the density function of the random vector Z(t) at the point z in the upper half-plane. For any bounded set A, define

π(z) :=

∫ ∞ 0

pt(z)dt, (3)

and

Π(A) :=

∫ A π(z)dz = E

[∫ ∞ 0

1A(Z(t))dt

] .

We call Π the Green’s measure of the process Z and π the occupancy density (alternatively, the Green’s function) of the process Z. Let (ρ, α) be the polar coordinate representation of a point z in the upper half-plane. The occupancy measure on the boundary (alternatively, the “pushing measure” or the “Green’s measure”) is defined as

ν(A) := E [∫ ∞

0 1A(Z(t))d`(t)

] .

Notice that ` increases only when Z2(t) = 0, which corresponds to the support of ν lying on the abscissa. Indeed, ν is the product measure and has density with respect to Lebesgue measure on the abscissa (see Harrison and Williams [9, §8]). In particular, let ν1 be the density such that ν(dz) = ν1(z1)dz1 × δ0(dz2).

With the above preparations now in hand, we now state Harrison’s open problem.

Harrison’s Problem [10]: Determine the exact asymptotic behavior of π(ρ, α) with ρ → ∞ and α fixed.

Theorem 6 of this paper closes this problem. In the process of finding the exact asymptotic

2The symmetrical case µ1 + rµ − 2 > 0 ensures that Z1(t)→∞. It can be treated in the same way.

3See Appendix A.3 for the case µ2 > 0.

ASYMPTOTIC BEHAVIOR OF THE OCCUPANCY DENSITY FOR RBM IN A HALF-PLANE 3

behavior of π(ρ, α) with ρ→∞ and α fixed, we also determine the exact tail asymptotic behav- ior of the boundary occupancy measure ν (Proposition 4) and an explicit integral expression for the occupancy density π (Proposition 5). These asymptotics lead us to explicitly determine all harmonic functions of the Martin compactification and to obtain the Martin boundary of the process (Proposition 13).

The significance of Harrison’s problem is directly related to the task of finding the exact asymptotic behavior of the stationary density of RBM in a quadrant. Referring to this task, Harrison remarks that “given the ‘cones of boundary influence’ discovered by Avram et al. [1], one may plausibly hope to crack the problem by piecing together the asymptotic analyses of occuupancy densities for three much simpler processes: a RBM in the upper half-plane that is obtained by removing the left-hand boundary of the quadrant; a RBM in the right half-plane that is obtained by removing the lower boundary of the quadrant; and the unrestricted Brow- nian motion that is obtained by removing both of the quadrant’s boundaries.” ([10]). Harrison further emphasizes the importance of the problem at hand by writing that “at the very least, the solution of the problem posed above may provide a deeper understanding or alternative interpretation of recent results on the asymptotic behavior of various quantities associated with the stationary distribution of RBM in a quadrant,” as in Dai and Miyazawa [4, 5], Franceschi and Kourkova [8].

The tools in this paper are, in part, inspired by methods introduced by the seminal work of Malyshev [19], which studies the asymptotic behavior of the stationary distribution for random walks in the quadrant. Subsequent works studying asymptotics in the spirit of Malshev’s ap- proach include Kourkova and Malyshev [13], which studies the Martin boundary of random walks in the quadrant and in the half-plane; Kourkova and Suhov [14], which extends the methods of Malyshev to the join-the-shorter-queue paradigm; Kourkova and Raschel [12], which studies the asymptotics of the Green’s functions of random walks in the quadrant with non-zero drift absorbed at the axes, and Franceschi and Kourkova [8], which extends Malyshev’s method to computing asymptotics in the continuous case.

A second group of literature closely relating to the present paper is that which concerns the asymptotics of the stationary distribution of semi-martingale reflecting Brownian motion (SRBM) in the quadrant (Dai and Miyazawa [4, 5]) or in the orthant (Miyazawa and Kobayashi [21]). These three papers develop a similar analytic method and contain similar asymptotic results to those for SRBM arising from a tandem queue (Lieshout and Mandjes [17, 18], Miyazawa and Rolski [22]).

The remainder of the paper is organized as follows. Proposition 2 of Section 2 establishes a kernel functional equation linking the moment generating functions of the measures π and ν. Section 3 is concerned with the boundary occupancy measure. An explicit expression for its moment generating function is established in Lemma 3 and its singularities are studied. The exact tail asymptotics of ν are subsequently given in Proposition 4. Proposition 5 of Section 4 expresses the occupancy density π as a simple integral via Laplace transform inversion. Theorem 6 in Section 5 provides the paper’s key result on the exact asymptotic behavior of π(ρ, α) as ρ → ∞ with α ∈ (0, π). Section 6 is devoted to the study of the Martin boundary and to the corresponding harmonic functions.

2. A kernel functional equation

We begin by defining the moment generating function (MGF) (alternatively, bilateral Laplace transform) of the measures π and ν. For θ = (θ1, θ2) ∈ C2, let

f(θ) := π̂(θ) =

∫ R×R+

eθ·zπ(z)dz = E [∫ ∞

0 eθ·Z(s)ds

] ,

and

g(θ1) := ν̂(θ) = ν̂1(θ1) =

∫ R eθ1·z1ν1(z1)dz1 = E

[∫ ∞ 0

eθ·Z(s)d`(s)

] .

4 PHILIP A. ERNST AND SANDRO FRANCESCHI

We note that g depends only on θ1; it does not depend on θ2 since the support of ν lies on the abscissa. Further, f is a two-dimensional Laplace transform which is bilateral for one dimension. We wish to establish a kernel functional equation linking the moment generating functions f and g (Proposition 2).

Consider the kernel

Q(θ) := 1

2 (θ21 + θ

2 2) + µ1θ1 + µ2θ2 =

1

2 (|θ + µ|2 − (µ21 + µ22)). (4)

Note that Q(θ)t = logE[eθ·Xt ] is the cumulant-generating function of Xt. The kernel Q is also called the “char

*View more*