betaFeedback

WorksMathematics

Uniform transport processes and the Brownian sheet:the one-parameter rate of convergence is preserved

Abstract

In~\cite{BFR} we constructed processes WnW_n, built from a family of independent standard Poisson processes, that converge almost surely to the Brownian sheet uniformly on [0,1]2[0,1]^2, and in~\cite{Rov} we proved that P(max⁡[0,1]2∣Wn−W∣>αn−β)=o(n−q)P(\max_{[0,1]^2}|W_n-W|>αn^{-β})=o(n^{-q}) for every β<λ/2β<λ/2, where λ∈(0,15)λ\in(0,\tfrac{1}{5}) is the parameter of the construction; this gives rates n−βn^{-β} with β<110β<\tfrac{1}{10}. In this paper, we show that the same construction, with λ≥1λ\geq 1, satisfies P ⁣(max⁡(s,t)∈[0,1]2∣Wn(s,t)−W(s,t)∣>α n−1/2(log⁡n)3)=o(n−q)for every q>0, P\!\left(\max_{(s,t)\in[0,1]^2} |W_n(s,t)-W(s,t)| > α\, n^{-1/2}(\log n)^3\right) = o(n^{-q}) \quad\text{for every }q>0, which is the one-parameter rate of Gorostiza and Griego~\cite{GG} up to a factor (log⁡n)1/2(\log n)^{1/2}. In particular the restriction λ<15λ<\tfrac{1}{5} is not needed.

Introduction

Let W={W(s,t):(s,t)∈[0,1]2}W=\{W(s,t):(s,t)\in[0,1]^2\} be a Brownian sheet. Griego, Heath and Ruiz-Moncayo [5] constructed realizations of uniform transport processes converging almost surely to Brownian motion, uniformly on [0,1][0,1]; the rate of convergence was obtained by Gorostiza and Griego [4] and by Csörgő and Horváth [3]. In [2] we extended the almost sure convergence to the Brownian sheet: they cut [0,1]2[0,1]^2 into ⌊nλ⌋\lfloor n^\lambda\rfloor vertical strips, observe that the sheet integrated over each strip is, after rescaling, a standard Brownian motion, approximate each of these independent Brownian motions by a uniform transport process as in [5], and sum over the strips. They obtained almost sure uniform convergence for λ∈(0,15)\lambda\in(0,\frac{1}{5}) and left the rate of convergence as an open problem. In [7] we obtained the rate

P(max⁡[0,1]2∣Wn−W∣>αn−β)=o(n−q),β<λ2,λ∈(0,15),P\left(\max_{[0,1]^2}|W_n-W|>\alpha n^{-\beta}\right)=o(n^{-q}),\qquad\beta<\frac{\lambda}{2},\quad\lambda\in(0,\tfrac{1}{5}),

and observed [7 Remark 2.3] that it is worse than the one-parameter rate n−1/2(log⁡n)5/2n^{-1/2}(\log n)^{5/2} of [4], because the latter, valid for each of the ⌊nλ⌋\lfloor n^\lambda\rfloor Brownian motions, could not be kept for their sum.

The purpose of this note is to show that the one-parameter rate is kept for the sum, up to a factor (log⁡n)1/2(\log n)^{1/2}. In [2, 7] the errors of the different strips are added through the triangle inequality, but these errors are independent and centered, so their partial sums behave like a random walk rather than like a sum of absolute values. On the other hand, using the Borell–TIS inequality, it no longer constrains λ\lambda from above, so we can take λ≥1\lambda\ge1.

Section 2 recalls the construction and states precisely, as a hypothesis, the properties of the one-parameter approximation that are used. Section 3 contains the results and Section 4 the proof. The probabilistic tools used in the proof are collected in Section 5.

The construction and the one-parameter input

We follow [2, 7]. Fix λ>0\lambda> 0 and write L=⌊nλ⌋L = \lfloor n^\lambda\rfloor, assuming nn large enough that L≥1L \ge1. It is convenient to assume that WW is the restriction to [0,1]2[0,1]^2 of a Brownian sheet defined on [0,2]×[0,1][0,2] \times[0,1]; this is no loss of generality, since the statements are about the existence of suitable realizations. For k=1,…,Lk = 1, \ldots, L put

Wk(t)=W(knλ,t)−W(k−1nλ,t),W~k(t)=nλ/2Wk(t),t∈[0,1].W_k(t) = W\left(\frac{k}{n^\lambda},t\right) - W\left(\frac{k-1}{n^\lambda},t\right), \qquad\widetilde{W}_k(t) = n^{\lambda/2}W_k(t), \qquad t \in[0,1].

The processes W~1,…,W~L\widetilde{W}_1,\ldots,\widetilde{W}_L are independent standard Brownian motions on [0,1][0,1] (they are centered, Gaussian, continuous, and E[W~k(t)W~k′(t′)]=(t∧t′)1{k=k′}\mathbb{E}[\widetilde{W}_k(t)\widetilde{W}_{k'}(t')] = (t\wedge t')\mathbf{1}_{\{k=k'\}}), and W(l/nλ,t)=∑k≤lWk(t)W(l/n^\lambda,t) = \sum_{k\le l}W_k(t).

For each kk, a uniform transport process W~k(n)\widetilde{W}^{(n)}_k is constructed from W~k\widetilde{W}_k by the Skorohod embedding procedure of [5]. Throughout the paper we write

rn=n−1/2(log⁡n)5/2,n≥2,r_n = n^{-1/2}(\log n)^{5/2}, \qquad n \ge2,

for the one-parameter rate of Gorostiza and Griego [4]. In the proofs we will use that

rn≥n−1for all n≥3.(1)r_n \ge n^{-1} \qquad\text{for all } n \ge3. \tag*{(1)}

We use the normalization of [4], in which the transport process moves with speed nn. The transport process of the kk-th strip has the law of

Xn(t)=n(−1)A∫0t(−1)N(n2r) dr=1n(−1)A∫0n2t(−1)N(u) du,t∈[0,1],X_n(t) = n(-1)^A\int_0^t(-1)^{N(n^2r)}\,\mathrm{d}r = \frac{1}{n}(-1)^A\int_0^{n^2t}(-1)^{N(u)}\,\mathrm{d}u, \qquad t \in[0,1],

where NN is a standard Poisson process and AA an independent Bernoulli(12)\left(\frac{1}{2}\right) variable: the process moves at speed nn and reverses its direction at the jump times of a Poisson process of rate n2n^2. It is realized on the same probability space as W~k\widetilde{W}_k as in [5, 2]. This is also the process constructed in the proof of [2 Theorem 2.1]: there the pieces have slope ±n\pm n, durations exponential with parameter 2n22n^2 and i.i.d. symmetric signs, and since each new sign differs from the previous one with probability 12\frac{1}{2}, independently, the direction is reversed at the jump times of a Poisson process of rate n2n^2.

From the construction of the approximations we get the following properties:

(H1) For every nn the pairs (W~k,W~k(n))(\widetilde{W}_k,\widetilde{W}^{(n)}_k), k≤Lk \le L, are independent and identically distributed. In particular the error processes Dk=W~k(n)−W~kD_k = \widetilde{W}^{(n)}_k - \widetilde{W}_k, k≤Lk \le L, are independent and identically distributed.

(H2) For every nn, kk and t∈[0,1]t \in[0,1], EW~k(n)(t)=0\mathbb{E}\widetilde{W}^{(n)}_k(t) = 0; and almost surely, for all t,t′∈[0,1]t,t' \in[0,1],

∣W~k(n)(t)∣≤n,∣W~k(n)(t)−W~k(n)(t′)∣≤n∣t−t′∣.|\widetilde{W}^{(n)}_k(t)| \le n, \qquad|\widetilde{W}^{(n)}_k(t)-\widetilde{W}^{(n)}_k(t')| \le n|t-t'|.

(H3) For every q>0q > 0 there are αq>0\alpha_q > 0 and nq≥3n_q \ge3, depending on qq but not on nn or kk, such that, with rn=n−1/2(log⁡n)5/2r_n = n^{-1/2}(\log n)^{5/2},

P(max⁡0≤t≤1∣W~k(n)(t)−W~k(t)∣>αqrn)≤n−qfor all n≥nq and all k≤L.(2)P\left(\max_{0\le t\le1}|\widetilde{W}^{(n)}_k(t)-\widetilde{W}_k(t)| > \alpha_q r_n\right) \le n^{-q} \qquad\text{for all } n \ge n_q \text{ and all } k \le L. \tag*{(2)}

The approximating process is defined at the strip boundaries by

Wn(lnλ,t)=∑k=1lWk(n)(t),Wk(n)(t)=n−λ/2W~k(n)(t),0≤l≤L,W_n\left(\frac{l}{n^\lambda},t\right)=\sum_{k=1}^{l}W_k^{(n)}(t),\qquad W_k^{(n)}(t)=n^{-\lambda/2}\widetilde{W}_k^{(n)}(t),\qquad0\le l\le L,

by linear interpolation in ss on [0,L/nλ][0,L/n^\lambda], and by Wn(s,t)=Wn(L/nλ,t)W_n(s,t)=W_n(L/n^\lambda,t) for s∈[L/nλ,1]s\in[L/n^\lambda,1].

Main theorem

The new bounds on the errors for these approximations are collected in the following theorem.

Theorem 3.1. Let λ>0\lambda>0. For every q>0q>0 there are constants CC and n0n_0, depending on qq, λ\lambda and on the constants of (H3), such that for all n≥n0n\ge n_0

P(max⁡(s,t)∈[0,1]2∣Wn(s,t)−W(s,t)∣>Clog⁡n((log⁡n)5/2n+n−λ/2))≤n−q.P\left(\max_{(s,t)\in[0,1]^2}|W_n(s,t)-W(s,t)|>C\sqrt{\log n}\left(\frac{(\log n)^{5/2}}{\sqrt{n}}+n^{-\lambda/2}\right)\right)\le n^{-q}.

Finally, the main result of this paper is collected in the following theorem.

Theorem 3.2. For every q>0q>0 there are α>0\alpha>0 and n0n_0 such that for n≥n0n\ge n_0

P(max⁡(s,t)∈[0,1]2∣Wn(s,t)−W(s,t)∣>αn−1/2(log⁡n)3)≤n−q,P\left(\max_{(s,t)\in[0,1]^2}|W_n(s,t)-W(s,t)|>\alpha n^{-1/2}(\log n)^3\right)\le n^{-q},

and max⁡[0,1]2∣Wn−W∣=O(n−1/2(log⁡n)3)\max_{[0,1]^2}|W_n-W|=O(n^{-1/2}(\log n)^3) almost surely. In particular the almost sure convergence of WnW_n to WW holds.

Proof. Recall rn=n−1/2(log⁡n)5/2r_n=n^{-1/2}(\log n)^{5/2} and apply Theorem 3.1. Choose any λ≥1\lambda\ge1. Then n−λ/2≤n−1/2≤rnn^{-\lambda/2}\le n^{-1/2}\le r_n, so the bound is at most 2Crn(log⁡n)1/22Cr_n(\log n)^{1/2}.

Using q=2q=2: with An={max⁡[0,1]2∣Wn−W∣>αn−1/2(log⁡n)3}A_n=\{\max_{[0,1]^2}|W_n-W|>\alpha n^{-1/2}(\log n)^3\} we have

∑n≥1P(An)≤n0+∑n≥n0n−2<∞.\sum_{n\ge1}P(A_n)\le n_0+\sum_{n\ge n_0}n^{-2}<\infty.

By the Borel–Cantelli lemma, almost surely AnA_n occurs for finitely many nn only, i.e. max⁡[0,1]2∣Wn−W∣≤αn−1/2(log⁡n)3\max_{[0,1]^2}|W_n-W|\le\alpha n^{-1/2}(\log n)^3 for all nn large enough. □\square

Remark 3.1. The rates of [7] are n−βn^{-\beta} with β<λ/2<110\beta<\lambda/2<\frac{1}{10}. Theorem 3.2 gives n−1/2(log⁡n)3n^{-1/2}(\log n)^3, the rate n−1/2(log⁡n)5/2n^{-1/2}(\log n)^{5/2} of [4] for one Brownian motion up to a factor (log⁡n)1/2(\log n)^{1/2}. This answers the question raised in [7 Remark 2.3].

Remark 3.2. (dd-parameter Wiener processes.) The same method applies to the dd-parameter Wiener process, d≥2d\ge2, approximated as in [2 Section 4]: the strips are replaced by the ⌊nλ⌋d−1\lfloor n^\lambda\rfloor^{d-1} boxes ∏i<d((ki−1)/nλ,ki/nλ]×[0,1]\prod_{i<d}((k_i-1)/n^\lambda,k_i/n^\lambda]\times[0,1], each carrying an independent transport process, the normalization n−λ/2n^{-\lambda/2} becomes n−λ(d−1)/2n^{-\lambda(d-1)/2}, and WnW_n is defined by (d−1)(d-1)-linear interpolation.

Proof of Theorem 3.1

Fix q>0q>0. All constants below may depend on qq, λ\lambda and the constants of (H3), but not on nn, and nn is assumed large enough.

Reduction of the problem

For 0≤l≤L0 \le l \le L and t∈[0,1]t \in[0,1] put

Δn(l,t)=Wn(lnλ,t)−W(lnλ,t)=n−λ/2∑k=1lDk(t),Dk=W~k(n)−W~k,\Delta_n(l,t)=W_n\left(\frac{l}{n^\lambda},t\right)-W\left(\frac{l}{n^\lambda},t\right)=n^{-\lambda/2}\sum_{k=1}^{l}D_k(t),\qquad D_k=\widetilde{W}^{(n)}_k-\widetilde{W}_k,

and, for 0≤l≤L0 \le l \le L,

Zl=sup⁡{∣W(s,t)−W(lnλ,t)∣:s∈[lnλ,l+1nλ], t∈[0,1]}.Z_l=\sup\left\{\left|W(s,t)-W\left(\frac{l}{n^\lambda},t\right)\right|:s\in\left[\frac{l}{n^\lambda},\frac{l+1}{n^\lambda}\right],\ t\in[0,1]\right\}.

The following lemma allows us to reduce the study of ∣Wn(s,t)−W(s,t)∣|W_n(s,t)-W(s,t)| to the study of max⁡0≤l≤L∣Δn(l,t)∣\max_{0\le l\le L}|\Delta_n(l,t)| and max⁡0≤l≤LZl\max_{0\le l\le L}Z_l.

Lemma 4.1. For every (s,t)∈[0,1]2(s,t)\in[0,1]^2,

∣Wn(s,t)−W(s,t)∣≤max⁡0≤l≤L∣Δn(l,t)∣+2max⁡0≤l≤LZl.|W_n(s,t)-W(s,t)|\le\max_{0\le l\le L}|\Delta_n(l,t)|+2\max_{0\le l\le L}Z_l.

Proof. If s∈[l/nλ,(l+1)/nλ)s\in[l/n^\lambda,(l+1)/n^\lambda) with l≤L−1l\le L-1, write θ=nλs−l∈[0,1)\theta=n^\lambda s-l\in[0,1). Then

Wn(s,t)−W(s,t)=(1−θ)Δn(l,t)+θΔn(l+1,t)W_n(s,t)-W(s,t)=(1-\theta)\Delta_n(l,t)+\theta\Delta_n(l+1,t)
+(1−θ)[W(lnλ,t)−W(s,t)]+θ[W(l+1nλ,t)−W(s,t)].\qquad+(1-\theta)\left[W\left(\frac{l}{n^\lambda},t\right)-W(s,t)\right]+\theta\left[W\left(\frac{l+1}{n^\lambda},t\right)-W(s,t)\right].

The first bracket is bounded by ZlZ_l and the second, by the triangle inequality through W(l/nλ,t)W(l/n^\lambda,t), by 2Zl2Z_l. If s≥L/nλs\ge L/n^\lambda, then Wn(s,t)−W(s,t)=Δn(L,t)+W(L/nλ,t)−W(s,t)W_n(s,t)-W(s,t)=\Delta_n(L,t)+W(L/n^\lambda,t)-W(s,t), and the last difference is bounded by ZLZ_L.

Thus it suffices to prove that for nn large enough,

P(max⁡0≤l≤LZl>Cn−λ/2(log⁡n)1/2)≤n−q.(3)P\left(\max_{0\le l\le L}Z_l>Cn^{-\lambda/2}(\log n)^{1/2}\right)\le n^{-q}. \tag*{(3)}
P(max⁡0≤l≤Lsup⁡t∈[0,1]∣Δn(l,t)∣>Crn(log⁡n)1/2)≤n−q.(4)P\left(\max_{0\le l\le L}\sup_{t\in[0,1]}|\Delta_n(l,t)|>Cr_n(\log n)^{1/2}\right)\le n^{-q}. \tag*{(4)}

Study of (3)

For each ll, the field

Bl(ρ,t)=nλ/2W([lnλ,lnλ+ρnλ]×(0,t]),(ρ,t)∈[0,1]2,B_l(\rho,t)=n^{\lambda/2}W\left(\left[\frac{l}{n^\lambda},\frac{l}{n^\lambda}+\frac{\rho}{n^\lambda}\right]\times(0,t]\right),\qquad(\rho,t)\in[0,1]^2,

is a Brownian sheet, and Zl=n−λ/2sup⁡[0,1]2∣Bl∣Z_l=n^{-\lambda/2}\sup_{[0,1]^2}|B_l|. Let m=Esup⁡[0,1]2∣Bl∣m=\mathbb{E}\sup_{[0,1]^2}|B_l|, which does not depend on ll. By Lemma 5.1 (maximal variance 1) with x=2(q+λ+1)log⁡nx=\sqrt{2(q+\lambda+1)\log n} and a union bound over the L+1≤nλ+1L+1\le n^\lambda+1 strips,

P(max⁡lZl>n−λ/2(m+2(q+λ+1)log⁡n))≤2(nλ+1)n−(q+λ+1)≤4n−q−1.P\left(\max_l Z_l>n^{-\lambda/2}\left(m+\sqrt{2(q+\lambda+1)\log n}\right)\right)\le2(n^\lambda+1)n^{-(q+\lambda+1)}\le4n^{-q-1}.

We get the bound (3).

Study of (4)

We begin studying DkD_k. Let q′=q+λ+7q'=q+\lambda+7, an=αq′rna_n=\alpha_{q'}r_n, with αq′\alpha_{q'} given by (H3), and put

Fk={sup⁡t∈[0,1]∣Dk(t)∣≤an},Gn=⋂k=1LFk.F_k=\left\{\sup_{t\in[0,1]}|D_k(t)|\le a_n\right\},\qquad G_n=\bigcap_{k=1}^{L}F_k.

By (H3) and (H1), πn:=P(Fkc)\pi_n:=P(F_k^c) does not depend on kk and πn≤n−q′\pi_n\le n^{-q'} for n≥nq′n\ge n_{q'}; hence

P(Gnc)≤Lπn≤nλ−q′≤n−q−1.P(G_n^c)\le L\pi_n\le n^{\lambda-q'}\le n^{-q-1}.

We apply Lemma 5.3 to each DkD_k. By (H2), EDk(t)=0\mathbb{E}D_k(t)=0 and sup⁡t∣Dk(t)∣≤Yk:=n+sup⁡t∣W~k(t)∣\sup_t|D_k(t)|\le Y_k:=n+\sup_t|\widetilde{W}_k(t)|, so that EYk2≤2n2+8≤10n2\mathbb{E}Y_k^2\le2n^2+8\le10n^2 by Doob’s inequality. By (H1) the function mn(t)=E[Dk(t)1Fk]m_n(t)=\mathbb{E}[D_k(t)\mathbf{1}_{F_k}] does not depend on kk, and

∣mn(t)∣≤(10n2)1/2πn1/2≤10 n1−q′/2≤10 n−(q+λ+5)/2.(5)|m_n(t)|\le(10n^2)^{1/2}\pi_n^{1/2}\le\sqrt{10}\,n^{1-q'/2}\le\sqrt{10}\,n^{-(q+\lambda+5)/2}. \tag*{(5)}

On GnG_n we have Dk(t)=Dk(t)1Fk=Dk′(t)+mn(t)D_k(t)=D_k(t)\mathbf{1}_{F_k}=D'_k(t)+m_n(t) for every kk and tt, hence

Δn(l,t)=n−λ/2∑k=1lDk′(t)+n−λ/2l mn(t),∣n−λ/2l mn(t)∣≤nλ/2∣mn(t)∣≤n−3/2,(6)\Delta_n(l,t)=n^{-\lambda/2}\sum_{k=1}^{l}D'_k(t)+n^{-\lambda/2}l\,m_n(t),\qquad|n^{-\lambda/2}l\,m_n(t)|\le n^{\lambda/2}|m_n(t)|\le n^{-3/2}, \tag*{(6)}

using l≤nλl\le n^\lambda and (5). Moreover, for fixed tt, the variables D1′(t),…,DL′(t)D'_1(t),\ldots,D'_L(t) are independent (by (H1)) and centered, and by Lemma 5.3

∣Dk′(t)∣≤an+∣mn(t)∣≤2ansurely, for every k≤L and t∈[0,1],(7)|D'_k(t)|\le a_n+|m_n(t)|\le2a_n\quad\text{surely, for every }k\le L\text{ and }t\in[0,1], \tag*{(7)}

because ∣mn(t)∣≤10 n−5/2≤αq′n−1≤an|m_n(t)|\le\sqrt{10}\,n^{-5/2}\le\alpha_{q'}n^{-1}\le a_n for nn large, by (5) and rn≥n−1r_n\ge n^{-1}.

Partial sums on a grid

Let M=⌈nλ+3⌉M=\lceil n^{\lambda+3}\rceil, ti=i/Mt_i=i/M, i=0,…,Mi=0,\ldots,M. For fixed l≤Ll\le L and i≤Mi\le M, the variables Xk=Dk′(ti)X_k=D'_k(t_i), k≤lk\le l, are independent, centered and, by (7), bounded by the deterministic constant c=2anc=2a_n, uniformly in kk, ll and ii. Lemma 5.2 therefore applies with c=2anc=2a_n and gives, since l≤nλl\le n^\lambda,

P(n−λ/2∣∑k=1lDk′(ti)∣>y)≤2exp⁡(−y2nλ2l(2an)2)≤2exp⁡(−y28an2).P\left(n^{-\lambda/2}\left|\sum_{k=1}^{l}D'_k(t_i)\right|>y\right)\le2\exp\left(-\frac{y^2n^\lambda}{2l(2a_n)^2}\right)\le2\exp\left(-\frac{y^2}{8a_n^2}\right).

Choose yn=an8(q+2λ+4)log⁡ny_n=a_n\sqrt{8(q+2\lambda+4)\log n}. The number of pairs (l,i)(l,i) is at most (nλ+1)(M+1)≤4n2λ+3(n^\lambda+1)(M+1)\le4n^{2\lambda+3}, so a union bound yields

P(max⁡l≤Lmax⁡i≤Mn−λ/2∣∑k=1lDk′(ti)∣>yn)≤8n2λ+3n−(q+2λ+4)=8n−q−1.P\left(\max_{l\le L}\max_{i\le M}n^{-\lambda/2}\left|\sum_{k=1}^{l}D'_k(t_i)\right|>y_n\right)\le8n^{2\lambda+3}n^{-(q+2\lambda+4)}=8n^{-q-1}.

Together with (6) and the bound on P(Gnc)P(G_n^c), outside an event of probability at most 9n−q−19n^{-q-1},

max⁡l≤Lmax⁡i≤M∣Δn(l,ti)∣≤yn+n−3/2≤Crn(log⁡n)1/2,(8)\max_{l\le L}\max_{i\le M}|\Delta_n(l,t_i)|\le y_n+n^{-3/2}\le Cr_n(\log n)^{1/2}, \tag*{(8)}

where we used n−3/2≤rnn^{-3/2}\le r_n.

Between grid points

Let t∈[ti,ti+1]t \in[t_i,t_{i+1}]. Then

∣Δn(l,t)−Δn(l,ti)∣≤∣Wn(lnλ,t)−Wn(lnλ,ti)∣+∣W(lnλ,t)−W(lnλ,ti)∣.|\Delta_n(l,t)-\Delta_n(l,t_i)| \leq|W_n(\frac{l}{n^\lambda},t)-W_n(\frac{l}{n^\lambda},t_i)|+|W(\frac{l}{n^\lambda},t)-W(\frac{l}{n^\lambda},t_i)|.

By (H2), the first term is at most n−λ/2⋅L⋅n⋅M−1≤nλ/2+1n−(λ+3)≤n−2n^{-\lambda/2}\cdot L\cdot n\cdot M^{-1}\leq n^{\lambda/2+1}n^{-(\lambda+3)}\leq n^{-2}, deterministically. For the second, for each i<Mi<M the field (s,ρ)↦M1/2W((0,s]×(ti,ti+ρ/M])(s,\rho)\mapsto M^{1/2}W((0,s]\times(t_i,t_i+\rho/M]) on [0,1]2[0,1]^2 is a Brownian sheet, and the second term is bounded by M−1/2M^{-1/2} times its supremum. By Lemma 5.1 with x=2(q+λ+4)log⁡nx=\sqrt{2(q+\lambda+4)\log n} and a union bound over the M≤2nλ+3M\leq2n^{\lambda+3} intervals,

P(max⁡i<Msup⁡s∈[0,1],t∈[ti,ti+1]∣W(s,t)−W(s,ti)∣>M−1/2(m+2(q+λ+4)log⁡n))≤4n−q−1.P\left(\max_{i<M}\sup_{\substack{s\in[0,1],\,t\in[t_i,t_{i+1}]}}|W(s,t)-W(s,t_i)|>M^{-1/2}\left(m+\sqrt{2(q+\lambda+4)\log n}\right)\right)\leq4n^{-q-1}.

Since M−1/2≤n−1≤rnM^{-1/2}\leq n^{-1}\leq r_n, combining with (8) we obtain the bound (4).

Remark 4.1. The construction of Section 2 makes sense for every λ>0\lambda>0; nothing in the definition of the strips, of the strip Brownian motions W~k\widetilde{W}_k, of the transport processes W~k(n)\widetilde{W}^{(n)}_k or of the interpolated process WnW_n requires λ<15\lambda<\frac{1}{5}. The restriction λ∈(0,15)\lambda\in(0,\frac{1}{5}) appears only in the proof of almost sure convergence in [2 pp. 324–325], where Markov’s inequality of order MM leads to an exponent that can be made smaller than −1-1 only if λ<15\lambda<\frac{1}{5}. The present paper uses from [2] only the construction (hypotheses (H1)–(H2)) and from [4] only the one-parameter rate of each strip (hypothesis (H3)); none of these depends on λ\lambda. Therefore Theorem 3.1 and Theorem 3.2 hold for every λ>0\lambda>0, and in particular for λ≥1\lambda\geq1, which is the range giving the rate n−1/2(log⁡n)3n^{-1/2}(\log n)^3.

Probabilistic tools

We collect here the probabilistic inequalities used in § 4. We begin with a Borell–TIS inequality.

Lemma 5.1. Let Γ\Gamma be a centered Gaussian field with continuous paths on a compact set TT, let σ2=sup⁡x∈TE Γ(x)2\sigma^2=\sup_{x\in T}\mathbb{E}\,\Gamma(x)^2 and m=Esup⁡T∣Γ∣m=\mathbb{E}\sup_T|\Gamma|. Then m<∞m<\infty and, for every x>0x>0,

P(sup⁡T∣Γ∣>m+x)≤2e−x2/(2σ2).P\left(\sup_T|\Gamma|>m+x\right)\leq2e^{-x^2/(2\sigma^2)}.

Proof. Since Γ\Gamma has continuous paths on the compact set TT, it is a.s. bounded: for almost every ω\omega the map t↦Γ(t,ω)t\mapsto\Gamma(t,\omega) is continuous on a compact set, hence bounded. The Borell–TIS inequality [1 Theorem 2.1.1], applied to Γ\Gamma and to −Γ-\Gamma, then gives

P(sup⁡TΓ>Esup⁡TΓ+x)≤e−x2/(2σ2),P(inf⁡TΓ<Einf⁡TΓ−x)≤e−x2/(2σ2).P\left(\sup_T\Gamma>\mathbb{E}\sup_T\Gamma+x\right)\leq e^{-x^2/(2\sigma^2)},\qquad P\left(\inf_T\Gamma<\mathbb{E}\inf_T\Gamma-x\right)\leq e^{-x^2/(2\sigma^2)}.

Since Esup⁡T(−Γ)=−Einf⁡TΓ≤Esup⁡T∣Γ∣=m\mathbb{E}\sup_T(-\Gamma)=-\mathbb{E}\inf_T\Gamma\leq\mathbb{E}\sup_T|\Gamma|=m, combining the two inequalities yields the stated bound.

We will also use the following Hoeffding inequality (see [6]).

Lemma 5.2. If X1,…,XlX_1,\ldots,X_l are independent, centered and ∣Xk∣≤c|X_k|\leq c almost surely for every kk, then for every x>0x>0

P(∣∑k=1lXk∣>x)≤2exp⁡(−x22lc2).P\left(\left|\sum_{k=1}^{l}X_k\right|>x\right)\leq2\exp\left(-\frac{x^2}{2lc^2}\right).

The last technical result is a truncation lemma.

Lemma 5.3. Let D=(D(t))t∈[0,1]D=(D(t))_{t\in[0,1]} be a process with ED(t)=0\mathbb{E}D(t)=0 for every tt and sup⁡t∣D(t)∣≤Y\sup_t\lvert D(t)\rvert\le Y with EY2<∞\mathbb{E}Y^2<\infty. For a>0a>0 let F={sup⁡t∣D(t)∣≤a}F=\{\sup_t\lvert D(t)\rvert\le a\} and

m(t)=E[D(t)1F],D′(t)=D(t)1F−m(t).m(t)=\mathbb{E}[D(t)\mathbf{1}_F],\qquad D'(t)=D(t)\mathbf{1}_F-m(t).

Then ED′(t)=0\mathbb{E}D'(t)=0, ∣m(t)∣≤(EY2)1/2P(Fc)1/2\lvert m(t)\rvert\le(\mathbb{E}Y^2)^{1/2}P(F^c)^{1/2} and, surely, ∣D′(t)∣≤a+∣m(t)∣\lvert D'(t)\rvert\le a+\lvert m(t)\rvert for every tt.

Proof. Since ED(t)=0\mathbb{E}D(t)=0, we have m(t)=−E[D(t)1Fc]m(t)=-\mathbb{E}[D(t)\mathbf{1}_{F^c}], and Cauchy–Schwarz gives the bound on m(t)m(t). The bound on D′(t)D'(t) holds on every sample point, because ∣D(t)1F∣≤a\lvert D(t)\mathbf{1}_F\rvert\le a everywhere.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process.

During the preparation of this work the author used Claude (Anthropic) to assist with LaTeX typesetting of formulas, translation and language editing, and the verification of proofs and calculations. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the published article.

References

  1. [1]R. J. Adler and J. E. Taylor. Random Fields and Geometry. Springer Monographs in Mathematics. Springer, New York, 2007.DOI
  2. [2]X. Bardina, M. Ferrante and C. Rovira. Strong approximations of Brownian sheet by uniform transport processes. Collect. Math. 71 (2020), 319–329.
  3. [3]M. Csörgő and L. Horváth. Rate of convergence of transport processes with an application to stochastic differential equations. Probab. Theory Related Fields 78 (1988), no. 3, 379–387.DOI
  4. [4]L. G. Gorostiza and R. J. Griego. Rate of convergence of uniform transport processes to Brownian motion and application to stochastic integrals. Stochastics 3 (1980), 291–303.DOI
  5. [5]R. J. Griego, D. Heath and A. Ruiz-Moncayo. Almost sure convergence of uniform transport processes to Brownian motion. Ann. Math. Statist. 42 (1971), no. 3, 1129–1131.DOI
  6. [6]W. Hoeffding. Probability inequalities for sums of bounded random variables. J. Amer. Statist. Assoc. 58 (1963), 13–30.
  7. [7]C. Rovira. Rate of convergence of uniform transport processes to a Brownian sheet. Open Math. 18 (2020), 731–737.arxiv.org/abs/1909.01981

Paper details

vlnc:W_5tfa8xr3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Claude (Anthropic)

View statement
Looking for related papers…
Contents