betaFeedback

WorksMathematics

Annihilation versus anchoring of a topologically charged pair on a perforated two-dimensional surface

Abstract

Two independent random walks of opposite topological charge move on the discrete torus (Z/nZ)2(\mathbb{Z}/n\mathbb{Z})^2, which contains kk well-separated holes. The walks annihilate when they come within a core distance of each other, and each walk is anchored when it first touches a hole. We prove that the probability of annihilation before anchoring converges to w0/(w0+∑jwj)w_0/(w_0+\sum_j w_j), where the weights are explicit and depend only logarithmically on the sizes of the core and of the holes; for point-like core and holes, the limit is 1/(1+k)1/(1+k). We then prove a fate theorem: the pair either terminates or ends as two opposite charges anchored on two different holes, with probability (k−1)/(k+1)(k-1)/(k+1) in the point-like case, and the hosting pair {j,l}\{j,l\} is chosen with probability proportional to wjwlw_jw_l. The proof combines an abstract competition lemma for rare sets in reversible Markov chains with sharp two-dimensional Green's function estimates.

Model and results

Model

Let Tn=(Z/nZ)2\mathbb{T}_n=(\mathbb{Z}/n\mathbb{Z})^2, N=n2N=n^2, with ∣⋅∣|\cdot| the Euclidean norm of the representative in (−n/2,n/2]2(-n/2,n/2]^2, and π\pi the uniform distribution on Tn\mathbb{T}_n. A walk is a continuous-time simple random walk on Tn\mathbb{T}_n with total jump rate 11; its transition kernel is pu(x,y)=pu(0,y−x)p_u(x,y)=p_u(0,y-x).

Fix k≥1k\ge1 and exponents α0,α1,…,αk∈[0,1)\alpha_0,\alpha_1,\ldots,\alpha_k\in[0,1); put α∗=max⁡iαi\alpha_*=\max_i\alpha_i.

  • Holes. Aj=B(cj,rj)={x:∣x−cj∣≤rj}A_j=B(c_j,r_j)=\{x:|x-c_j|\le r_j\} with rj=⌊nαj⌋r_j=\lfloor n^{\alpha_j}\rfloor if αj>0\alpha_j>0 and rj=0r_j=0 (a single site) if αj=0\alpha_j=0, j=1,…,kj=1,\ldots,k, and dist⁡(Ai,Aj)≥n/log⁡n\operatorname{dist}(A_i,A_j)\ge n/\log n for $i\ne j.

  • Defects. XX (charge +1+1) and YY (charge −1-1) are independent walks started at x0,y0x_0,y_0.

  • Annihilation. With ρ=⌊nα0⌋\rho=\lfloor n^{\alpha_0}\rfloor if α0>0\alpha_0>0 and ρ=0\rho=0 if α0=0\alpha_0=0, let T0=inf⁡{u:∣Xu−Yu∣≤ρ}T_0=\inf\{u:|X_u-Y_u|\le\rho\}.

  • Anchoring. TjX=inf⁡{u:Xu∈Aj}T_j^X=\inf\{u:X_u\in A_j\} and TjY=inf⁡{u:Yu∈Aj}T_j^Y=\inf\{u:Y_u\in A_j\}.

The start (x0,y0)(x_0,y_0) is admissible if ∣x0−y0∣≥ρ+n/log⁡n|x_0-y_0|\ge\rho+n/\log n and dist⁡(x0,Aj)∧dist⁡(y0,Aj)≥n/log⁡n\operatorname{dist}(x_0,A_j)\wedge\operatorname{dist}(y_0,A_j)\ge n/\log n for all jj. Let T=T0∧min⁡j(TjX∧TjY)T=T_0\wedge\min_j(T_j^X\wedge T_j^Y) be the time of the first event. Almost surely exactly one walk jumps at time TT. If the state reached at time TT realises several events simultaneously, we give priority to annihilation, then to XX, then to the smaller index jj. All statements below are unaffected by this convention.

For r≥0r\ge0 write h(r)=1/EπTB(0,r)h(r)=1/\mathbb{E}_{\pi}T_{B(0,r)} for a single walk started from π\pi; by translation invariance EπTB(c,r)\mathbb{E}_{\pi}T_{B(c,r)} does not depend on cc. Put hj=h(rj)h_j=h(r_j) and H=∑j=1khjH=\sum_{j=1}^k h_j.

Results

Theorem 1 (Competition law, rate form). Uniformly over admissible starts and hole positions, as n→∞n \to\infty,

P(annihilation is the first event)=h(ρ)h(ρ)+H+o(1),\mathbb{P}(\mathit{annihilation\ is\ the\ first\ event})=\frac{h(\rho)}{h(\rho)+H}+o(1),
P(the first event is the anchoring of X at Aj)=hj2(h(ρ)+H)+o(1),\mathbb{P}(\mathit{the\ first\ event\ is\ the\ anchoring\ of\ }X\mathit{\ at\ }A_j)=\frac{h_j}{2\left(h(\rho)+H\right)}+o(1),

and the same holds with YY in place of XX.

Theorem 2 (Hitting rates). Uniformly in 0≤r≤nα∗0 \le r \le n^{\alpha_*},

EπTB(0,r)=2πn2log⁡nr∨1(1+o(1)).\mathbb{E}_{\pi}T_{B(0,r)}=\frac{2}{\pi}n^2\log\frac{n}{r\vee1}(1+o(1)).

Set wi=1/(1−αi)w_i=1/(1-\alpha_i), W=∑j=1kwjW=\sum_{j=1}^{k}w_j, Q=∑j=1kwj2Q=\sum_{j=1}^{k}w_j^2. Theorem 2 gives hj/h(ρ)→wj/w0h_j/h(\rho)\to w_j/w_0, hence:

Corollary 1 (Competition law). P(annihilation first)→w0/(w0+W)\mathbb{P}(\mathit{annihilation\ first})\to w_0/(w_0+W). If core and holes are single sites, the limit is 1/(1+k)1/(1+k); in that case h(ρ)=hjh(\rho)=h_j for every nn by symmetry, so Theorem 2 is not needed.

Fate. After the first event the dynamics continue as follows. If the first event is annihilation, the outcome is termination. If a defect is anchored at AiA_i, it stops, and the other defect continues as a walk until it first touches ⋃lAl\bigcup_l A_l, say at AlA_l. If l=il=i the charges cancel and the outcome is termination; if l≠il\ne i the outcome is the sustained pair S{i,l}S_{\{i,l\}}: charges +1+1 and −1-1 anchored on two different holes.

Theorem 3 (Fate of a pair). Uniformly over admissible starts and hole positions,

P(termination)→w0w0+W+QW(w0+W),\mathbb{P}(\mathit{termination})\to\frac{w_0}{w_0+W}+\frac{Q}{W(w_0+W)},
P(S{i,l})→2wiwlW(w0+W)(i<l),\mathbb{P}(S_{\{i,l\}})\to\frac{2w_iw_l}{W(w_0+W)}\qquad(i<l),
P(sustained pair)→W2−QW(w0+W).\mathbb{P}(\mathit{sustained\ pair})\to\frac{W^2-Q}{W(w_0+W)}.

In particular, with a single hole the pair terminates with probability →1\to1, and for kk point-like holes and a point-like core P(sustained pair)→(k−1)/(k+1)\mathbb{P}(\mathit{sustained\ pair})\to(k-1)/(k+1).

Throughout, C,cC,c denote positive constants depending only on kk and α∗\alpha_*, whose value may change from line to line, and nn is assumed large enough. We fix

g=gn=(log⁡n)1/2,t=gn2,s=g1/2n2.g=g_n=(\log n)^{1/2},\qquad t=gn^2,\qquad s=g^{1/2}n^2.

An abstract competition lemma

Let ZZ be an irreducible, reversible, continuous-time Markov chain on a finite set SS with stationary distribution π\pi, relaxation time trelt_{\mathrm{rel}}, and d(u)=max⁡z∥Pz(Zu∈⋅)−π∥TVd(u)=\max_z\lVert\mathbb{P}_z(Z_u\in\cdot)-\pi\rVert_{\mathrm{TV}}, which is non-increasing in uu. Let B0,…,Bm⊂SB_0,\ldots,B_m\subset S be nonempty (not necessarily disjoint), TiT_i the hitting time of BiB_i, T=min⁡iTiT=\min_iT_i, and I=min⁡{i:ZT∈Bi}I=\min\{i:Z_T\in B_i\}. Let fi(z)=Pz(I=i)f_i(z)=\mathbb{P}_z(I=i) and pˉi=Pπ(I=i)\bar p_i=\mathbb{P}_{\pi}(I=i). For fixed 0<s<t0<s<t let

ai=Pπ(Ti≤t),a=∑iai,δ=∑i<lPπ(Ti≤t,Tl≤t).a_i=\mathbb{P}_{\pi}(T_i\le t),\qquad a=\sum_i a_i,\qquad\delta=\sum_{i<l}\mathbb{P}_{\pi}(T_i\le t,T_l\le t).

Lemma 2.1. If δ<a\delta< a, then for every ii

∣pˉi−aia∣≤d(s)+Pπ(T≤s)+δa−δ.\left|\bar{p}_i-\frac{a_i}{a}\right| \le d(s)+\frac{\mathbb{P}_{\pi}(T\le s)+\delta}{a-\delta}.

Proof. On {T>t}\{T>t\} the Markov property gives Pπ(I=i∣Ft)=fi(Zt)\mathbb{P}_{\pi}(I=i\mid\mathcal{F}_t)=f_i(Z_t), and Eπfi(Zt)=pˉi\mathbb{E}_{\pi}f_i(Z_t)=\bar{p}_i by stationarity. Hence

Pπ(T≤t,I=i)=pˉi−Eπ[1{T>t}fi(Zt)]=Eπ[1{T≤t}fi(Zt)].(1)\mathbb{P}_{\pi}(T\le t,I=i)=\bar{p}_i-\mathbb{E}_{\pi}[1_{\{T>t\}}f_i(Z_t)]=\mathbb{E}_{\pi}[1_{\{T\le t\}}f_i(Z_t)]. \tag*{(1)}

On {T≤t−s}\{T\le t-s\} the strong Markov property at TT gives Eπ[fi(Zt)∣FT]=(Pt−Tfi)(ZT)\mathbb{E}_{\pi}[f_i(Z_t)\mid\mathcal{F}_T]=(P_{t-T}f_i)(Z_T), and since 0≤fi≤10\le f_i\le1 and t−T≥st-T\ge s, ∣(Pt−Tfi)(ZT)−pˉi∣≤d(s)\left|(P_{t-T}f_i)(Z_T)-\bar{p}_i\right|\le d(s). On the other hand Pπ(t−s<T≤t)≤Pπ(Z visits Bi in [t−s,t])=Pπ(T≤s)\mathbb{P}_{\pi}(t-s<T\le t)\le\mathbb{P}_{\pi}(Z\text{ visits }B_i\text{ in }[t-s,t])=\mathbb{P}_{\pi}(T\le s) by stationarity. Splitting (1) according to {T≤t−s}\{T\le t-s\} and {t−s<T≤t}\{t-s<T\le t\},

∣Pπ(T≤t,I=i)−pˉiPπ(T≤t)∣≤d(s)Pπ(T≤t)+Pπ(T≤s).(2)\left|\mathbb{P}_{\pi}(T\le t,I=i)-\bar{p}_i\mathbb{P}_{\pi}(T\le t)\right|\le d(s)\mathbb{P}_{\pi}(T\le t)+\mathbb{P}_{\pi}(T\le s). \tag*{(2)}

If I=iI=i then Ti=TT_i=T; if Ti≤tT_i\le t and I≠iI\ne i then Tl≤Ti≤tT_l\le T_i\le t for some l≠il\ne i. So

ai−δ≤Pπ(T≤t,I=i)≤ai,a−δ≤Pπ(T≤t)≤a,a_i-\delta\le\mathbb{P}_{\pi}(T\le t,I=i)\le a_i,\qquad a-\delta\le\mathbb{P}_{\pi}(T\le t)\le a,

the last by the Bonferroni inequality. Therefore ∣Pπ(T≤t,I=i)/Pπ(T≤t)−ai/a∣≤δ/(a−δ)\left|\mathbb{P}_{\pi}(T\le t,I=i)/\mathbb{P}_{\pi}(T\le t)-a_i/a\right|\le\delta/(a-\delta), and dividing (2) by Pπ(T≤t)≥a−δ\mathbb{P}_{\pi}(T\le t)\ge a-\delta gives the claim.

Lemma 2.2. For every z∈Sz\in S and every ii, ∣fi(z)−pˉi∣≤d(s)+Pz(T≤s)\left|f_i(z)-\bar{p}_i\right|\le d(s)+\mathbb{P}_z(T\le s).

Proof. fi(z)=Pz(T≤s,I=i)+Ez[1{T>s}fi(Zs)]=Ezfi(Zs)+Ez[1{T≤s}(1{I=i}−fi(Zs))]f_i(z)=\mathbb{P}_z(T\le s,I=i)+\mathbb{E}_z[1_{\{T>s\}}f_i(Z_s)]=\mathbb{E}_z f_i(Z_s)+\mathbb{E}_z[1_{\{T\le s\}}(1_{\{I=i\}}-f_i(Z_s))], and ∣Ezfi(Zs)−pˉi∣≤d(s)\left|\mathbb{E}_z f_i(Z_s)-\bar{p}_i\right|\le d(s).

Lemma 2.3 (Aldous–Brown). For nonempty A⊂SA\subset S and all u≥0u\ge0, ∣Pπ(TA>u)−e−u/EπTA∣≤trel/EπTA\left|\mathbb{P}_{\pi}(T_A>u)-e^{-u/\mathbb{E}_{\pi}T_A}\right|\le t_{\mathrm{rel}}/\mathbb{E}_{\pi}T_A. Consequently, with hA=1/EπTAh_A=1/\mathbb{E}_{\pi}T_A,

Pπ(TA≤u)≤(u+trel)hA,∣Pπ(TA≤u)−uhA∣≤trelhA+12(uhA)2.\mathbb{P}_{\pi}(T_A\le u)\le(u+t_{\mathrm{rel}})h_A,\qquad\left|\mathbb{P}_{\pi}(T_A\le u)-uh_A\right|\le t_{\mathrm{rel}}h_A+\frac{1}{2}(uh_A)^2.

Proof. The first statement is [1 Theorem 1]; see also [2 Ch. 3, Sec. 5.4]. The rest follows from

x−12x2≤1−e−x≤x.x-\frac{1}{2}x^2\le1-e^{-x}\le x.

Heat-kernel facts on the torus

Let p~u\widetilde{p}_u be the kernel of the rate-1 continuous-time simple random walk WW on Z2\mathbb{Z}^2 started at 00, so that pu(0,z)=∑m∈Z2p~u(z+nm)p_u(0,z)=\sum_{m\in\mathbb{Z}^2}\widetilde{p}_u(z+nm). We use two standard inputs.

(L1) Local limit theorem [4 Ch. 2]: for u≥1u\ge1 and all x∈Z2x\in\mathbb{Z}^2,

∣p~u(x)−(πu)−1e−∣x∣2/u∣≤Cu−2.\left|\widetilde{p}_u(x)-(\pi u)^{-1}e^{-|x|^2/u}\right|\le Cu^{-2}.

(The covariance of WuW_u is (u/2)I(u/2)I.)

(L2) The coordinates of WW are independent rate-12\frac{1}{2} walks on Z\mathbb{Z} with kernel q~u(k)=e−u/2I∣k∣(u/2)\widetilde{q}_u(k)=e^{-u/2}I_{|k|}(u/2), which is symmetric, non-increasing in ∣k∣|k|, and satisfies q~u(0)≤C(1+u)−1/2\widetilde{q}_u(0)\le C(1+u)^{-1/2}.

Lemma 3.1. There are absolute constants C,cC,c such that for all large nn:

(H0) pu(x,y)≤C/(1+u)p_u(x,y)\le C/(1+u) for 0≤u≤n20\le u\le n^2;

(H1) pu(x,y)≤C/n2p_u(x,y)\le C/n^2 for u≥n2u\ge n^2, and also for all u≤n2u\le n^2 when ∣x−y∣≥n/8|x-y|\ge n/8;

(H2) pn2(x,y)≥c/n2p_{n^2}(x,y) \ge c/n^2 for all x,yx,y;

(H3) G(z):=∫0n2pu(0,z) duG(z) := \int_0^{n^2} p_u(0,z)\,\mathrm{d}u satisfies G(z)=2πlog⁡n∣z∣∨1+O(1)G(z) = \frac{2}{\pi}\log\frac{n}{|z|\vee1} + O(1) for ∣z∣≤n/4|z| \le n/4, and G(z)≤CG(z) \le C for ∣z∣>n/4|z| > n/4.

Proof. (H0). pu(0,z)=∏i=1,2∑m∈Zq~u(zi+nm)p_u(0,z) = \prod_{i=1,2}\sum_{m\in\mathbb{Z}}\widetilde q_u(z_i+nm). Since q~u\widetilde q_u is non-increasing in ∣k∣|k|, for ∣zi∣≤n/2|z_i| \le n/2 and m≠0m \ne0 the term q~u(zi+nm)\widetilde q_u(z_i+nm) is at most (2/n)(2/n) times the sum of q~u\widetilde q_u over a block of ⌈n/2⌉\lceil n/2\rceil consecutive sites, and these blocks are disjoint for distinct mm of the same sign. Hence ∑mq~u(zi+nm)≤q~u(0)+4/n≤C(1+u)−1/2\sum_m\widetilde q_u(z_i+nm) \le\widetilde q_u(0)+4/n \le C(1+u)^{-1/2} for u≤n2u \le n^2.

(H1), off-diagonal part. The moment generating function of a coordinate is EeθWv(1)=exp⁡(v2(cosh⁡θ−1))≤exp⁡(v2θ2)\mathbb{E}e^{\theta W_v^{(1)}} = \exp\left(\frac{v}{2}(\cosh\theta-1)\right) \le\exp\left(\frac{v}{2}\theta^2\right) for ∣θ∣≤1|\theta| \le1, so Chernoff’s bound with θ=min⁡(1,b/v)\theta= \min(1,b/v) gives P(Wv(1)≥b)≤exp⁡(−min⁡(b2/(2v),b/2))\mathbb{P}(W_v^{(1)} \ge b) \le\exp(-\min(b^2/(2v),b/2)). The torus walk is the projection of WW, so Px(∣Xv−x∣≥n/16)≤4exp⁡(−min⁡(b2/(2v),b/2))\mathbb{P}_x(|X_v-x| \ge n/16) \le4\exp(-\min(b^2/(2v),b/2)) with b=n/32b=n/32. Let ∣x−z∣≥n/8|x-z| \ge n/8 and u≤n2u \le n^2. By Chapman–Kolmogorov at time u/2u/2, and since every yy has ∣x−y∣≥n/16|x-y| \ge n/16 or ∣y−z∣≥n/16|y-z| \ge n/16, (H0) and symmetry give

pu(x,z)≤2sup⁡y,y′pu/2(y,y′)Px(∣Xu/2−x∣≥n/16)≤C1+uexp⁡(−min⁡(b2u,b2))≤Cn2.p_u(x,z) \le2\sup_{y,y'}p_{u/2}(y,y')\mathbb{P}_x(|X_{u/2}-x| \ge n/16) \le\frac{C}{1+u}\exp\left(-\min\left(\frac{b^2}{u},\frac{b}{2}\right)\right) \le\frac{C}{n^2}.

(H1), u≥n2u \ge n^2, and (H2). Let z~\widetilde z be the representative of zz with ∣z~∣≤n/2|\widetilde z| \le n/\sqrt{2}. By (L1), p~n2(z~)≥(πn2)−1e−1/2−Cn−4≥c/n2\widetilde p_{n^2}(\widetilde z) \ge(\pi n^2)^{-1}e^{-1/2}-Cn^{-4} \ge c/n^2, which gives (H2) because pn2(0,z)≥p~n2(z~)p_{n^2}(0,z) \ge\widetilde p_{n^2}(\widetilde z). By (H0), sup⁡pn2≤C/n2\sup p_{n^2} \le C/n^2, and for u>n2u > n^2 Chapman–Kolmogorov gives pu(x,y)≤sup⁡wpn2(w,y)≤C/n2p_u(x,y) \le\sup_w p_{n^2}(w,y) \le C/n^2.

(H3). Let ∣z∣≤n/4|z| \le n/4 and eu(z)=pu(0,z)−p~u(z)≥0e_u(z) = p_u(0,z)-\widetilde p_u(z) \ge0. On the event that the torus walk is at zz while Wu≠zW_u \ne z, the time σ=inf⁡{v:∣Wv∣∞≥n/2}\sigma= \inf\{v:|W_v|_\infty\ge n/2\} is at most uu and ∣Xσ−z∣≥n/4−1≥n/8|X_\sigma-z| \ge n/4-1 \ge n/8. By the strong Markov property and (H1), eu(z)≤sup⁡v≤n2, ∣x−z∣≥n/8pv(x,z)≤C/n2e_u(z) \le\sup_{v\le n^2,\ |x-z|\ge n/8}p_v(x,z) \le C/n^2 for u≤n2u \le n^2, so 0≤∫0n2eu(z) du≤C0 \le\int_0^{n^2}e_u(z)\,\mathrm{d}u \le C. By (L1),

∫0n2p~u(z) du=∫1n2e−∣z∣2/uπu du+O(1),\int_0^{n^2}\widetilde p_u(z)\,\mathrm{d}u = \int_1^{n^2}\frac{e^{-|z|^2/u}}{\pi u}\,\mathrm{d}u + O(1),

the error being at most 1+∫1∞Cu−2 du1+\int_1^\infty Cu^{-2}\,\mathrm{d}u. For z=0z=0 the integral is 2πlog⁡n\frac{2}{\pi}\log n. For ∣z∣≥1|z|\ge1 substitute v=∣z∣2/uv=|z|^2/u:

∫1n2e−∣z∣2/uπu du=1π∫∣z∣2/n2∣z∣2e−vv dv=1πlog⁡n2∣z∣2+O(1),\int_1^{n^2}\frac{e^{-|z|^2/u}}{\pi u}\,\mathrm{d}u = \frac{1}{\pi}\int_{|z|^2/n^2}^{|z|^2}\frac{e^{-v}}{v}\,\mathrm{d}v = \frac{1}{\pi}\log\frac{n^2}{|z|^2}+O(1),

because ∫01(1−e−v)v−1 dv≤1\int_0^1(1-e^{-v})v^{-1}\,\mathrm{d}v \le1 and ∫1∞e−vv−1 dv≤1\int_1^\infty e^{-v}v^{-1}\,\mathrm{d}v \le1. This proves the first part. If ∣z∣>n/4|z|>n/4, (H1) gives G(z)≤n2⋅C/n2G(z)\le n^2\cdot C/n^2. ∎

Lemma 3.2 (Mixing). For the single walk, d(mn2)≤(1−c)md(mn^2) \le(1-c)^m and trel≤n2/4t_{\mathrm{rel}} \le n^2/4. For the pair (X,Y)(X,Y) on Tn×Tn\mathbb{T}_n\times\mathbb{T}_n, d(mn2)≤(1−c2)md(mn^2) \le(1-c^2)^m and trel≤n2/4t_{\mathrm{rel}} \le n^2/4.

Proof. By (H2), pn2(x,⋅)≥cπ(⋅)p_{n^2}(x,\cdot) \ge c\pi(\cdot); Doeblin’s argument gives the bound on dd, and the pair satisfies the same minorisation with c2c^2. The generator of the walk has eigenvalues 1−12(cos⁡2πan+cos⁡2πbn)1-\frac{1}{2}\left(\cos\frac{2\pi a}{n}+\cos\frac{2\pi b}{n}\right); the smallest non-zero one is 12(1−cos⁡2πn)≥4/n2\frac{1}{2}\left(1-\cos\frac{2\pi}{n}\right) \ge4/n^2 because 1−cos⁡x≥2x2/π21-\cos x \ge2x^2/\pi^2 on [−π,π][-\pi,\pi]. The spectral gap of a product of independent chains is the minimum of the gaps. ∎

Hitting estimates for a single walk

Let A=B(a0,r)A=B(a_0,r) with 0≤r≤nα∗0\le r\le n^{\alpha_*}, and L=log⁡nr∨1≥(1−α∗)log⁡nL=\log\frac{n}{r\vee1}\ge(1-\alpha_*)\log n. For v≥0v\ge0 and x∈Tnx\in\mathbb{T}_n let Γv(x)=Ex∫0v1A(Xu) du=∑y∈A∫0vpu(x,y) du\Gamma_v(x)=\mathbb{E}_x\int_0^v\mathbf{1}_A(X_u)\,\mathrm{d}u=\sum_{y\in A}\int_0^v p_u(x,y)\,\mathrm{d}u.

Lemma 4.1 (Occupation). For a∈Aa \in A and n2≤v≤2tn^2 \le v \le2t, Γv(a)=∣A∣[2πL+O(1+g)]\Gamma_v(a)=|A|[\frac{2}{\pi}L+O(1+g)], and for all v≤2tv \le2t, Γv(a)≤∣A∣[2πL+C(1+g)]\Gamma_v(a)\le|A|[\frac{2}{\pi}L+C(1+g)]. If dist⁡(x,A)≥n/log⁡n\operatorname{dist}(x,A)\ge n/\log n and v≤4tv\le4t, then Γv(x)≤∣A∣C(log⁡log⁡n+g)\Gamma_v(x)\le|A|C(\log\log n+g).

Proof. For a,y∈Aa,y\in A we have ∣a−y∣≤2r≤n/4|a-y|\le2r\le n/4, so by (H3)

∑y∈AG(y−a)=∣A∣2πL+2π∑y∈Alog⁡r∨1∣a−y∣∨1+O(∣A∣).\sum_{y\in A}G(y-a)=|A|\frac{2}{\pi}L+\frac{2}{\pi}\sum_{y\in A}\log\frac{r\vee1}{|a-y|\vee1}+O(|A|).

Each term of the last sum is at least −log⁡2-\log2. Grouping yy by ∣a−y∣∈(r2−j−1,r2−j]|a-y|\in(r2^{-j-1},r2^{-j}], j≥0j\ge0, the sum is at most Cr2∑j(j+1)4−j≤C∣A∣Cr^2\sum_j(j+1)4^{-j}\le C|A| when r≥1r\ge1 (and it vanishes when r=0r=0). By (H1),

0≤∑y∈A∫n2vpu(a,y) du≤C∣A∣(v−n2)/n2≤C∣A∣g.0\le\sum_{y\in A}\int_{n^2}^{v}p_u(a,y)\,du\le C|A|(v-n^2)/n^2\le C|A|g.

This proves the first two claims. If dist⁡(x,A)≥n/log⁡n\operatorname{dist}(x,A)\ge n/\log n, then by (H3) G(y−x)≤2πlog⁡log⁡n+CG(y-x)\le\frac{2}{\pi}\log\log n+C for every y∈Ay\in A, and the contribution of u∈[n2,v]u\in[n^2,v] is at most C∣A∣⋅4gC|A|\cdot4g.

Lemma 4.2 (Hitting bounds). (a) For every xx and v>0v>0, Px(TA≤v)≤Γ2v(x)/inf⁡a∈AΓv(a)\mathbb{P}_x(T_A\le v)\le\Gamma_{2v}(x)/\inf_{a\in A}\Gamma_v(a).

(b) Pπ(TA≤n2)≤C/L\mathbb{P}_\pi(T_A\le n^2)\le C/L and EπTA≥cn2L\mathbb{E}_\pi T_A\ge cn^2L.

(c) If dist⁡(x,A)≥n/log⁡n\operatorname{dist}(x,A)\ge n/\log n and v≤2tv\le2t, then Px(TA≤v)≤κn:=C(log⁡log⁡n+g)/log⁡n→0\mathbb{P}_x(T_A\le v)\le\kappa_n:=C(\log\log n+g)/\log n\to0.

Proof. (a) XTA∈AX_{T_A}\in A, so by the strong Markov property Γ2v(x)≥Ex[1{TA≤v}Γv(XTA)]\Gamma_{2v}(x)\ge\mathbb{E}_x[1_{\{T_A\le v\}}\Gamma_v(X_{T_A})]. (b) Averaging (a) over x∼πx\sim\pi with v=n2v=n^2 and using EπΓ2n2=2∣A∣\mathbb{E}_\pi\Gamma_{2n^2}=2|A| and Lemma 4.1, Pπ(TA≤n2)≤2/(2πL−C)≤C1/L\mathbb{P}_\pi(T_A\le n^2)\le2/(\frac{2}{\pi}L-C)\le C_1/L. By stationarity and the union bound, Pπ(TA≤mn2)≤mC1/L\mathbb{P}_\pi(T_A\le mn^2)\le mC_1/L; with m=⌊L/(2C1)⌋m=\lfloor L/(2C_1)\rfloor this is at most 1/21/2, so EπTA≥12mn2\mathbb{E}_\pi T_A\ge\frac12mn^2. (c) Since Px(TA≤v)\mathbb{P}_x(T_A\le v) is non-decreasing in vv we may take n2≤v≤2tn^2\le v\le2t, and (a) with Lemma 4.1 gives the bound C(log⁡log⁡n+g)/(2πL−C(1+g))C(\log\log n+g)/(\frac{2}{\pi}L-C(1+g)).

Proof of Theorem 2. Write E=EπTAE=\mathbb{E}_\pi T_A. By stationarity and the strong Markov property,

t∣A∣n2=Eπ∫0t1A(Xu) du=Eπ[1{TA≤t}Γt−TA(XTA)].\frac{t|A|}{n^2}=\mathbb{E}_\pi\int_0^t1_A(X_u)\,du=\mathbb{E}_\pi[1_{\{T_A\le t\}}\Gamma_{t-T_A}(X_{T_A})].

By Lemma 4.1 the right-hand side is at most Pπ(TA≤t)∣A∣[2πL+C(1+g)]\mathbb{P}_\pi(T_A\le t)|A|[\frac{2}{\pi}L+C(1+g)] and at least Pπ(TA≤t−n2)∣A∣[2πL−C(1+g)]\mathbb{P}_\pi(T_A\le t-n^2)|A|[\frac{2}{\pi}L-C(1+g)]. Since Pπ(t−n2<TA≤t)≤Pπ(TA≤n2)≤C/L\mathbb{P}_\pi(t-n^2<T_A\le t)\le\mathbb{P}_\pi(T_A\le n^2)\le C/L by stationarity and Lemma 4.2(b),

g2πL+C(1+g)≤Pπ(TA≤t)≤g2πL−C(1+g)+CL.\frac{g}{\frac{2}{\pi}L+C(1+g)}\le\mathbb{P}_\pi(T_A\le t)\le\frac{g}{\frac{2}{\pi}L-C(1+g)}+\frac{C}{L}.

As g→∞g\to\infty and g/L→0g/L\to0, this gives Pπ(TA≤t)=πg2L(1+o(1))\mathbb{P}_\pi(T_A\le t)=\frac{\pi g}{2L}(1+o(1)), uniformly in rr. By Lemma 4.2(b), t/E≤g/(cL)→0t/E\le g/(cL)\to0, and by Lemmas 2.3 and 3.2, ∣Pπ(TA≤t)−t/E∣≤tE(14g+t2E)|\mathbb{P}_\pi(T_A\le t)-t/E|\le\frac{t}{E}(\frac{1}{4g}+\frac{t}{2E}). Hence

tE=πg2L(1+o(1)),\frac{t}{E}=\frac{\pi g}{2L}(1+o(1)),

i.e. E=2πn2L(1+o(1))E=\frac{2}{\pi}n^2L(1+o(1)).

Proof of Theorem 1

Let Z=(X,Y)Z=(X,Y) on S=Tn×TnS=\mathbb{T}_n\times\mathbb{T}_n, with stationary law π⊗π\pi\otimes\pi (again written π\pi), and

B0=D0={(x,y):∣x−y∣≤ρ},DjX=Aj×Tn,DjY=Tn×Aj,B_0=D_0=\{(x,y):|x-y|\le\rho\},\qquad D_j^X=A_j\times\mathbb{T}_n,\qquad D_j^Y=\mathbb{T}_n\times A_j,

ordered as D0,D1X,…,DkX,D1Y,…,DkYD_0,D_1^X,\ldots,D_k^X,D_1^Y,\ldots,D_k^Y to match the priority convention.

Rates. X−YX-Y is a rate-2 simple random walk on Tn\mathbb{T}_n, started uniformly under π\pi, so EπTD0=12EπTB(0,ρ)\mathbb{E}_{\pi}T_{D_0}=\frac{1}{2}\mathbb{E}_{\pi}T_{B(0,\rho)}, i.e. the rate of D0D_0 is 2h(ρ)2h(\rho). The rates of DjXD_j^X and DjYD_j^Y are hjh_j. By Lemma 4.2(b) every rate h′h' in this list satisfies th′≤Cg/log⁡n→0th'\le Cg/\log n\to0, and trel/t≤1/(4g)t_{\mathrm{rel}}/t\le1/(4g). Hence, by Lemma 2.3, each ai=Pπ(Ti≤t)a_i=\mathbb{P}_{\pi}(T_i\le t) equals tt times its rate times 1+ηi1+\eta_i, where η:=max⁡i∣ηi∣≤14g+Cg/log⁡n→0\eta:=\max_i|\eta_i|\le\frac{1}{4g}+Cg/\log n\to0. In particular a=2t(h(ρ)+H)(1+O(η))a=2t(h(\rho)+H)(1+O(\eta)).

Pairs. We show δ=o(a)\delta=o(a). There are finitely many pairs, of three types.

(i) DjXD_j^X and DlYD_l^Y. XX and YY are independent under π\pi, so the probability that both are hit by time tt is ajXalY≤Cthj⋅aa_j^X a_l^Y\le Cth_j\cdot a.

(ii) DjXD_j^X and DlXD_l^X, j≠lj\ne l (and likewise for YY). By the strong Markov property at TjXT_j^X and Lemma 4.2(c), as dist⁡(Aj,Al)≥n/log⁡n\operatorname{dist}(A_j,A_l)\ge n/\log n,

Pπ(TjX≤TlX≤t)≤ajXsup⁡x∈AjPx(TAl≤t)≤ajXκn,\mathbb{P}_{\pi}(T_j^X\le T_l^X\le t)\le a_j^X\sup_{x\in A_j}\mathbb{P}_x(T_{A_l}\le t)\le a_j^X\kappa_n,

so the pair contributes at most 2κna2\kappa_n a.

(iii) D0D_0 and DjXD_j^X (and likewise for YY). Write Du=Xu−YuD_u=X_u-Y_u. Realise the stationary pair as follows: X0X_0 and D0D_0 are independent and uniform; jumps occur at the points of a rate-2 Poisson process; at the ii-th jump an independent uniform direction eie_i and an independent fair sign σi\sigma_i are drawn, and XX moves by eie_i if σi=+1\sigma_i=+1 while YY moves by eie_i if σi=−1\sigma_i=-1. Then DD jumps by fi=σieif_i=\sigma_i e_i. The directions (fi)(f_i) are i.i.d. uniform and independent of (σi)(\sigma_i), so the path of DD is a function of (D0,jump times,(fi))(D_0,\text{jump times},(f_i)), which is independent of (X0,(σi))(X_0,(\sigma_i)). Since Xu=X0+∑i≤Nufi1{σi=+1}X_u=X_0+\sum_{i\le N_u}f_i\mathbf{1}_{\{\sigma_i=+1\}} and X0X_0 is uniform and independent of everything else, XT0X_{T_0} is uniform and independent of the path of DD. By the strong Markov property of the pair at T0T_0,

Pπ(T0≤2t, X visits Aj during [T0,T0+2t])=Pπ(T0≤2t)Pπ(TAj≤2t).(3)\mathbb{P}_{\pi}(T_0\le2t,\ X\text{ visits }A_j\text{ during }[T_0,T_0+2t])=\mathbb{P}_{\pi}(T_0\le2t)\mathbb{P}_{\pi}(T_{A_j}\le2t). \tag*{(3)}

Now let F={∃u≤v≤2t: Xu∈Aj, Dv∈B(0,ρ)}F=\{\exists u\le v\le2t:\ X_u\in A_j,\ D_v\in B(0,\rho)\} and F~={∃v≤u≤2t: Dv∈B(0,ρ), Xu∈Aj}\widetilde F=\{\exists v\le u\le2t:\ D_v\in B(0,\rho),\ X_u\in A_j\}. The stationary pair is reversible, so its time reversal on [0,2t][0,2t] has the same law and Pπ(F)=Pπ(F~)\mathbb{P}_{\pi}(F)=\mathbb{P}_{\pi}(\widetilde F). The event F~\widetilde F is contained in the event in (3), and {TjX≤t,T0≤t}⊂F∪F~\{T_j^X\le t,T_0\le t\}\subset F\cup\widetilde F. With Lemma 2.3,

Pπ(TjX≤t, T0≤t)≤2(2t+trel)22h(ρ)hj≤Cth(ρ)⋅a.\mathbb{P}_{\pi}(T_j^X\le t,\ T_0\le t)\le2(2t+t_{\mathrm{rel}})^2 2h(\rho)h_j\le Cth(\rho)\cdot a.

Altogether δ≤C(κn+g/log⁡n)a=o(a)\delta\le C(\kappa_n+g/\log n)a=o(a).

Short times. By Lemma 2.3, Pπ(T≤s)≤(s+trel)⋅2(h(ρ)+H)≤C(s/t)a=Cg−1/2a\mathbb{P}_{\pi}(T\le s)\le(s+t_{\mathrm{rel}})\cdot2(h(\rho)+H)\le C(s/t)a=Cg^{-1/2}a. By Lemma 3.2, d(s)≤(1−c1/2)⌊g1/2⌋→0d(s)\le(1-c^{1/2})^{\lfloor g^{1/2}\rfloor}\to0.

Conclusion from stationarity. Lemma 2.1 now gives pˉi=ai/a+o(1)\bar p_i=a_i/a+o(1), and ai/aa_i/a equals the ratio of the corresponding rates up to a factor 1+O(η)1+O(\eta). Thus pˉ0=h(ρ)/(h(ρ)+H)+o(1)\bar p_0=h(\rho)/(h(\rho)+H)+o(1) and pˉjX=pˉjY=hj/(2(h(ρ)+H))+o(1)\bar p_j^X=\bar p_j^Y=h_j/(2(h(\rho)+H))+o(1).

Admissible starts. Let z0=(x0,y0)z_0=(x_0,y_0) be admissible. By Lemma 4.2(c), Px0(TAj≤s)≤κn\mathbb{P}_{x_0}(T_{A_j}\le s)\le\kappa_n, and the same holds for y0y_0. The difference X−YX-Y is a rate-2 walk started at distance at least n/log⁡nn/\log n from B(0,ρ)B(0,\rho), so Pz0(T0≤s)=Px0−y0(TB(0,ρ)≤2s)≤κn\mathbb{P}_{z_0}(T_0\le s)=\mathbb{P}_{x_0-y_0}(T_{B(0,\rho)}\le2s)\le\kappa_n for a rate-1 walk, since 2s≤2t2s\le2t. Hence Pz0(T≤s)≤(2k+1)κn\mathbb{P}_{z_0}(T\le s)\le(2k+1)\kappa_n, and Lemma 2.2 gives ∣fi(z0)−pˉi∣≤d(s)+(2k+1)κn→0|f_i(z_0)-\bar p_i|\le d(s)+(2k+1)\kappa_n\to0. All bounds depend on the configuration only through kk and α∗\alpha_*.

Remark 5.1. In the point-like case, EπT{a}\mathbb{E}_{\pi}T_{\{a\}} is the same for every site aa, so the three rates are 2h(0),h(0),h(0)2h(0),h(0),h(0) exactly and the limit 1/(1+k)1/(1+k) uses only Lemma 4.2(b),(c).

Proof of Theorem 3

Single-walk competition. Apply § 2 to one walk with the sets A1,…,AkA_1,\ldots,A_k. The pair bound (ii), the short-time bound and the mixing bound hold verbatim, so with φl(y)=Py(the first hole touched is Al)\varphi_l(y)=\mathbb{P}_y(\text{the first hole touched is }A_l) and φˉl=Eπφl\bar{\varphi}_l=\mathbb{E}_{\pi}\varphi_l,

φˉl=hlH+o(1),∣φl(y)−φˉl∣≤d(s)+kκnif dist⁡(y,Aj)≥nlog⁡n∀j.(4)\bar{\varphi}_l=\frac{h_l}{H}+o(1),\qquad\left|\varphi_l(y)-\bar{\varphi}_l\right|\le d(s)+k\kappa_n\quad\text{if }\operatorname{dist}(y,A_j)\ge\frac{n}{\log n}\quad\forall j. \tag*{(4)}

Position of the free defect. Let EiXE_i^X be the event that the first event is the anchoring of XX at AiA_i. Let τX=min⁡jTjX\tau^X=\min_j T_j^X, which is a function of the path of XX only. On EiXE_i^X we have T=τXT=\tau^X, and YTY_T lies in no hole. Let N={y:dist⁡(y,Aj)<n/log⁡n for some j}\mathcal{N}=\{y:\operatorname{dist}(y,A_j)<n/\log n\text{ for some }j\}, so that π(N)≤Ck(n/log⁡n+nα∗)2/n2→0\pi(\mathcal{N})\le Ck(n/\log n+n^{\alpha_*})^2/n^2\to0. Since τX\tau^X is independent of YY,

P(EiX,YT∈N)≤P(YτX∈N)≤Px0(τX≤s)+sup⁡u≥sPy0(Yu∈N)≤kκn+π(N)+d(s)→0.\mathbb{P}(E_i^X,Y_T\in\mathcal{N})\le\mathbb{P}(Y_{\tau^X}\in\mathcal{N})\le\mathbb{P}_{x_0}(\tau^X\le s)+\sup_{u\ge s}\mathbb{P}_{y_0}(Y_u\in\mathcal{N})\le k\kappa_n+\pi(\mathcal{N})+d(s)\to0.

Second phase. By the strong Markov property at TT, on EiXE_i^X the free defect YY continues as a walk from YTY_T, so P(EiX,Y next touches Al)=E[1EiXφl(YT)]\mathbb{P}(E_i^X,Y\text{ next touches }A_l)=\mathbb{E}[1_{E_i^X}\varphi_l(Y_T)]. By (4) and the previous display,

P(EiX,Y next touches Al)=P(EiX)hlH+o(1)=hihl2H(h(ρ)+H)+o(1),\mathbb{P}(E_i^X,Y\text{ next touches }A_l)=\mathbb{P}(E_i^X)\frac{h_l}{H}+o(1)=\frac{h_i h_l}{2H\bigl(h(\rho)+H\bigr)}+o(1),

using Theorem 1. The same holds with the roles of XX and YY exchanged.

Summation. S{i,l}S_{\{i,l\}} occurs in four disjoint ways (XX or YY anchored first, at AiA_i or at AlA_l), so P(S{i,l})=2hihlH(h(ρ)+H)+o(1)\mathbb{P}(S_{\{i,l\}})=\frac{2h_i h_l}{H(h(\rho)+H)}+o(1). Termination occurs by annihilation or by both defects reaching the same hole: P(termination)=h(ρ)h(ρ)+H+∑ihi2H(h(ρ)+H)+o(1)\mathbb{P}(\text{termination})=\frac{h(\rho)}{h(\rho)+H}+\frac{\sum_i h_i^2}{H(h(\rho)+H)}+o(1). By Theorem 2, hj/h(ρ)→wj/w0h_j/h(\rho)\to w_j/w_0, which gives the stated limits; they sum to 1. For k=1k=1, W2=QW^2=Q. For point-like core and holes, w0=wj=1w_0=w_j=1, W=Q=kW=Q=k, and (W2−Q)/(W(1+W))=(k−1)/(k+1)(W^2-Q)/(W(1+W))=(k-1)/(k+1).

Remarks

Remark 7.1 (Numerical check). Monte Carlo simulation on tori with n=32,64,128n=32,64,128 and k=1,2,4k=1,2,4 point holes (3,000–6,000 runs each, stationary starts) gives P(annihilation first)\mathbb{P}(\text{annihilation first}) and P(sustained pair)\mathbb{P}(\text{sustained pair}) within 0.03 of 1/(1+k)1/(1+k) and (k−1)/(k+1)(k-1)/(k+1); for k=1k=1 no sustained pair occurs.

Remark 7.2 (Scope). (i) The error terms are O((log⁡n)−1/4)O((\log n)^{-1/4}) from the choice of gg and are not optimised. (ii) The proof uses the torus only through Lemmas 3.1 and 3.2 and through translation invariance (for the identity of rates and for step (iii), which uses the group structure). (iii) Partial anchoring (a hole anchors with probability q<1q<1 per visit), creation of new pairs, detachment, interaction between defects and anisotropic conduction are not covered by these theorems.

References

  1. [1]D. J. Aldous and M. Brown, Inequalities for rare events in time-reversible Markov chains I, in Stochastic Inequalities, IMS Lecture Notes 22 (1992), 1–16.DOI
  2. [2]D. J. Aldous and J. A. Fill, Reversible Markov Chains and Random Walks on Graphs, monograph, Chapter 3.
  3. [4]G. F. Lawler and V. Limic, Random Walk: A Modern Introduction, Cambridge University Press, 2010.DOI

Paper details

vlnc:W_7s3a7keg
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

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

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents