Abstract
Let \({\mathcal {C}}^*_{\epsilon ,\mathbf{S}^2}\) denote the cover time of the two dimensional sphere by a Wiener sausage of radius \(\epsilon \). We prove that
is tight, where \(A_{\mathbf{S}^2}=4\pi \) denotes the Riemannian area of \(\mathbf{S}^2\).
Similar content being viewed by others
Change history
06 March 2020
Lemma 9.1 in the paper is incorrect as stated.
Notes
We expect that the case of the torus \(\mathbf {T}^2\) could be handled by our methods, although we do not address this in the paper.
An earlier version of this article claimed such a result, based on a faulty reduction from general manifolds to planar Brownion motion.
Terms depending on z are absorbed in the factor \(\sqrt{L}\).
The formulation in [25, Theorem 5.2] allows for the independence of \(Z_i\) and \(\bar{Z}_i\) of each other, with the change that the parameter of the Bernoulli \(B_i\) equals the variation distance. By splitting \(B_i\) in that formulation, we arrive at the current statement.
As pointed out by the referee, there is a typo in the latter; in the bottom of page 538 and top of page 539, all sums of the form \(\sum _{i_1,i_2,\ldots : \sum i_j=k}\) are missing the multiplicative factor \(k!/\prod _j i_j!\). With these extra factors, the derivation in [4] gives the result claimed there.
References
Abe, Y.: Second order term of cover time for planar simple random walk. arXiv:1709.08151 (2017)
Arguin, L.-P.: Extrema of log-correlated random variables: principles and examples. In: Contucci, P., Giardina, C. (eds.) Advances in Disordered Systems, Random Processes and Some Applications, pp. 166–204. Cambridge University Press, Cambridge (2016)
Arguin, L.-P., Belius, D., Harper, A.J.: Maxima of a randomized Riemann zeta function, and branching random walks. Ann. Appl. Probab. 27, 178–215 (2017)
Belius, D., Kistler, N.: The subleading order of two dimensional cover times. Probab. Theory Relat. Fields 167, 461–552 (2017)
Belius, D., Rosen, J., Zeitouni, O.: Barrier estimates for a critical Galton–Watson process and the cover time of the binary tree. Annales Inst. H. Poincare 55, 127–154 (2019)
Birkhoff, G.: Tres observaciones sobre el algebra lineal. Univ. Nac. TucumRev Ser. A 5, 147–151 (1946)
Bolthausen, E., Deuschel, J.-D., Giacomin, G.: Entropic repulsion and the maximum of the two-dimensional harmonic crystal. Ann. Probab. 29, 1670–1692 (2001)
Bramson, M.: Minimal displacement of branching random walk. Z. Wahrsch. Verw. Gebiete 45, 89–108 (1978)
Bramson, M., Ding, J., Zeitouni, O.: Convergence in law of the maximum of the two-dimensional discrete Gaussian free field. Commun. Pure Appl. Math. 69, 62–123 (2016)
Biskup, M., Louidor, O.: Extreme local extrema of two-dimensional discrete Gaussian free field. Commun. Math. Phys. 345, 271–304 (2016)
Bramson, M., Zeitouni, O.: Tightness of the recentered maximum of the two-dimensional discrete Gaussian free field. Commun. Pure Appl. Math. 65, 1–20 (2011)
Brillinger, D.: A particle migrating randomly on a sphere. JTP 10, 429–443 (1997)
Chung, K.L.: A Course in Probability Theory, 3rd edn. Academic Press, New York (2001)
Cortines, A., Louidor, O., Saglietti, S.: A Scaling Limit for the Cover Time of the Binary Tree. arXiv:1812.10101 (2018)
Dembo, A., Peres, Y., Rosen, J., Zeitouni, O.: Cover times for Brownian motion and random walks in two dimensions. Ann. Math. 160, 433–467 (2004)
Dembo, A., Peres, Y., Rosen, J., Zeitouni, O.: Thick points for planar Brownian motion and the Erdös–Taylor conjecture for random walk. Acta Math. 186, 239–270 (2001)
Dembo, A., Rosen, J., Zeitouni, O.: Convergence in law of the cover time and cover time excursion counts on the binary tree. arXiv:1906.07276 (2019)
Ding, J.: Asymptotics of cover times via Gaussian free fields: bounded degree graphs and general trees. Ann. Probab. 42, 464–496 (2014)
Ding, J., Lee, J., Peres, Y.: Cover times, blanket times, and majorizing measures. Ann. Math. 175, 1409–1471 (2012)
Ding, J., Roy, R., Zeitouni, O.: Convergence of the centered maximum of log-correlated Gaussian fields. arXiv:1503.04588 (2015)
Ding, J., Zeitouni, O.: A sharp estimate for cover times on binary trees. Stoch. Process. Appl. 122, 2117–2133 (2012)
Fitzsimmons, P., Pitman, J.: Kac’s moment formula and the Feynman–Kac formula for aditive functionals of a Markov process. Stoch. Proc. Appl. 79, 117–134 (1999)
Fournier, N., Guillin, A.: On the rate of convergence in Wasserstein distance of the empirical measure. Probab. Theory Relat. Fields 162, 707–738 (2015)
Kistler, N.: Derrida’s random energy models. From spin glasses to the extremes of correlated random fields. In: Gayrard, V., Kistler, N. (eds.) Correlated Random Systems: Five Different Methods. Lecture Notes in Mathematics, vol. 2143, pp. 71–120. Springer, Berlin (2015)
Lindvall, T.: Lectures on the Coupling Method. Wiley, New York (1992)
Mirsky, L.: Proofs of two theorems on doubly stochastic matrices. Proc. Am. Math. Soc. 9, 371–374 (1958)
Spivak, M.: A Comprehensive Introduction to Differential Geometry, vol. 2, 2nd edn. Publish or Perish Inc., Berkeley (1975)
Symeonidis, E.: The Poisson integral for a ball in spaces of constant curvature. Comment. Math. Univ. Carol. 44, 437–460 (2003)
Zhai, A.: Exponential concentration of cover times. Electron. J. Probab. (2018). https://doi.org/10.1214/18-EJP149
Zeitouni, O.: Branching random walks and Gaussian fields. In: Probability and Statistical Physics in St. Petersburg, Proceedings of Symposia in Pure Mathematics, vol. 91, pp. 437–471. American Mathematical Society, Providence (2016)
Acknowledgements
We thank two anonymous referees for a detailed reading of the paper. We particularly thank one of the referees for raising doubts concerning our reduction of the general compact manifold case.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
David Belius was supported by SNF grant PP00P2_176918. Part of his work was carried out while he was at the University of Zurich and at the Courant Institute.
Jay Rosen was partially supported by the Simons Foundation. This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No. 692452).
Appendices
Appendix I: Barrier estimates
Recall that \(\alpha (l)=\rho _{L}(L-l) - l^{\gamma }_{L}\), \(\gamma <1/2\), see (1.14) and (1.18), where
and recall that
Recall the notation \(I_{y }=[y,y+1)\). (Note that in [5], \(I_{y }\) is denoted \(H_{y }\).)
Lemma 8.1
For any \(k<L\) and all \( j \ge 0\) and \(0\le z\le 100k\) with \(t_{z}\in \mathbb {N}\),
Proof of Lemma 8.1
Recall that we denote by \(T_{l}, l=0,1,2,\ldots \) the critical Galton–Watson process with geometric distribution, and let \(P^{\mathrm{GW}}_n\) denote its law when \(T_0=n\). With \(v=\rho _{L}(L-k)- k^{ \gamma }_{L}\) we rewrite \(K_{1} \) as
We show below that for any \(0<\gamma <1\), and all \(1\le l\le k-1\)
from which it follows that
Hence
and by (8.2), \(\sqrt{2t_{z}} =\rho _{L}L+ z=\rho _{L}k+v+ z + k^{\gamma }_{L} \).
Thus using [5, Theorem 1.1], with \(a=\rho _{L}k+v, x=\sqrt{2t_{z}}\) and \(b=v, y=v+ j\),
We write out
Using the concavity of the logarithm, we have \(k\log (L)/L \le \log k\), and using \((r-s)^{2}\ge r^{2}/2-s^{2}\) we see that
It follows that
We now prove (8.5). This certainly holds if \(l\le k/2\), since then \(l_{L}^{\gamma }=l_{k}^{\gamma }=l^{\gamma }\). If \(L/2\le l\le k\), then (8.5) says that
which follows from concavity.
Equation (8.5) holds if \(k\le L/2\), since then \(l_{L}^{\gamma }=l^{\gamma }\le k^{\gamma }= k_{L}^{\gamma }\). Finally, if \(k/2\le l\le L/2<k\), then (8.5) says that
Since for \( l\le L/2\) we have \(l^{\gamma }\le (L-l)^{\gamma }\), (8.9) follows from (8.8). \(\square \)
Recall that
see (4.14).
Lemma 8.2
For all L sufficiently large, and all \(0\le z\le 10 L\) with \(t_{z}\in \mathbb {N}\),
Similar estimates hold if we delete the barrier condition on some fixed finite interval.
Proof of Lemma 8.2
The left hand side of (8.11) can be written in terms of the critical Galton–Watson process \(T_{l}, l=0,1,2,\ldots \) as
Equation (8.11) follows immediately from [5, Theorem 1.1], with \(a=\rho _{L}L,\)\(x=\rho _{L}L+z\), \( b=y=0\), since \(\{ T_{L} =0\}=\{ T_{L} \in [0,1] \}\).
For the last statement, we simply note that following the proof of [5, Lemma 2.3] we can show that the analogue of [5, Theorem 1.1] holds where we skip some fixed finite interval. \(\square \)
Lemma 8.3
If \((L-k) \log L/L=o_{L}(1)\) then for all L sufficiently large, and all \(0\le z \le \log L\) with \(t_{z}\in \mathbb {N}\),
Proof of Lemma 8.3
We rewrite this in terms of the critical Galton–Watson process with geometric offspring distribution, \(T_{l}, l=0,1,2,\ldots \), as
We condition on \(T_{k}\) as follows: let \(v=\rho _{L}\left( L-k\right) \). Then
Since, by (3.21),
we obtain
By [5, Theorem 1.1], with \(a=\rho _{L}L\), \(x=\rho _{L}L+z\) recall (8.2), and \(b=v, y=v+j\),
But
which follows by considering separately \(j\ge 4(L-k)z/k\) and \(j< 4(L-k)z/k\), since in this case \(j<z/4\) because our condition on k implies that \((L-k)/k \) is tiny. It then follows that
\(\square \)
Lemma 8.4
If \(k\log L/L=o_{L}(1)\) and \(m=\rho _{L}(L-k)+j\), then for all L sufficiently large, with \(m^{ 2}/2\in \mathbb {N}\),
and if \(j\le \eta L\) and we skip the barrier from \(k+1\) to \(k+s\), with \(s\le \eta ' \log L\), for some \(\eta , \eta '<\infty \), then
Proof
We first turn to the probability in (8.15). By the Markov property, the probability in question can be rewritten in terms of the critical Galton–Watson process \(T_{l}, l=0,1,2,\ldots \) as
By [5, Theorem 1.1], with \(a=\rho _{L}\left( L-k\right) , x=m\) and \(b=y=0\),
We have
Using the concavity of the logarithm, we get that
which gives (8.15).
For (8.16) with \(k'=k+ s\), \(v= \rho _{L}(L-k')\) we bound
where
By [5, Proposition 1.4, Remark 2.2]
and as in the first part of this proof
Thus
Considering separately the cases where \(j'\le 1.1 j\) and \(j'> 1.1 j\) we see that for L sufficiently large,
and (8.16) follows. \(\square \)
Lemma 8.5
If \(v=\rho _{L}(L-k)+u\) then for L large with \(v^{2}/2\in \mathbb {N}\), and \(1\le k,{\widetilde{k}}, u,j\le L^{3/4}\),
In addition, if we skip the barrier from \(k+1\) to \(k+3\),
Proof
By the Markov property, the probability in question can be rewritten in terms of the critical Galton–Watson process \(T_{l}, l=0,1,2,\ldots \) as
This is a linear barrier of length \({\widetilde{k}}\). At the start of the barrier it is at distance u from the starting point, and at the end it is at distance j from the end point. Therefore by [5, Theorem 1.1] we have that the probability is at most
where we have bounded the square root using the fact that \(k,{\widetilde{k}}, u,j<L^{3/4}\) so that the ratio inside the square root is \(\asymp 2L/2L=1\). Using \(k,{\widetilde{k}}, u,j<L^{3/4}\) again we see that
This gives (8.23).
Equation (8.24) follows as in the proof of the previous Lemma. \(\square \)
Appendix II: Conditional excursion probabilities
The following is a modification of [16] adapted to our situation. Fix \(k'>k>2\) and let \({{\mathcal {G}}}_k^y\) to be the \(\sigma \)-algebra generated by the excursions from \(\partial B_{d}(y,h_{k-1})\) to \(\partial B_{d}(y,h_{k})\) (and if we start outside \(\partial B_{d}(y,h_{k-1})\) we include the initial excursion to \(\partial B_{d}(y,h_{k-1})\)). Note that \({{\mathcal {G}}}_k^y\) includes the information on the end points of the excursions from \(\partial B_{d}(y,h_{k})\) to \(\partial B_{d}(y,h_{k-1})\).
Recall that \(T_{y,l-1\rightarrow l}^{y,k\overset{n}{\rightarrow }k-1} \) is the number of excursions from \(\partial B_{d}(y,h_{l-1})\rightarrow \partial B_{d}(y,h_{l})\) during the first n excursions from \(\partial B_{d}(y,h_{k })\rightarrow \partial B_{d}(y,h_{k-1})\), see (2.11).
Lemma 9.1
For any \(L-2k\ge k '>k\ge 2\) and \(n>1\), let \(\mathcal {A}_{k'}\) denote an event, measurable on the excursions of the Brownian motion from \(\partial B_{d}(y,h_{k'})\) to \(\partial B_{d}(y,h_{k'-1})\) and on the collection \(\{T_{y,l-1\rightarrow l}^{y,k\overset{n}{\rightarrow }k-1} \}_{l=k',\ldots ,L}\) . Then,
In particular, for all \(m_l ;l=k',\ldots L\), and all \(y \in \mathbf{S}^2\),
The key to the proof of Lemma 9.1 is the following Lemma.
Lemma 9.2
For \(k'>k \ge 2\) and a Brownian path \(X_\cdot \) starting at \(z\in \partial B_{d}(y,h_{k'-1})\), let \(Z_l=T_{y,l-1\rightarrow l}^{y,k'-1\overset{1}{\rightarrow }k} \), \(l=k',\ldots L\), denote the number of excursions of the path from \(\partial B_{d}(y,h_{l-1})\rightarrow \partial B_{d}(y,h_{l})\), prior to \(\bar{\tau }=\inf \{t>0\, :\,X_t\in \partial B_{d}(y,h_{k})\}\), and let F denote an event measurable with respect to the path of the Brownian motion inside \(B_d(y,h_{k'})\), prior to \(\bar{\tau }\). Then, for some \(c<\infty \) and all \(\{ m_l : l=k',\ldots L \}\), uniformly in \(v\in \partial B_{d}(y,h_{k})\) and y,
In words, conditioning by the endpoint of the excursion at level k has only minor influence on the probability of events involving pieces of excursions at levels larger than \(k'\).
Proof of Lemma 9.2
Without loss of generality we can take \(y=0\). Fixing \(k \ge 2\) and \(z \in \partial B_{d}(0, h_{k'-1})\) it suffices to consider \(\{m_l ,\, l=k',\ldots L \}\) for which \(\mathbb {P}^z\left( Z_{l} =m_l,\, l=k',\ldots L\right) >0\). Fix such \(\{m_l ,\, l=k',\ldots L \}\) and a positive continuous function g on \(\partial B_{d}(0,h_{k})\). Let \(\bar{\tau } = \inf \{ t:\, w_t \in \partial B_{d}(0,h_{k})\}\), \(\tau _0=0\) and for \(i=0,1,\ldots \) define
Set \(j=m_{k'}\) and let \(Z^j_l\), \(l = k',\ldots L\) be the corresponding number of excursions by the Brownian path prior to time \(\tau _{2j}\). Then, by the strong Markov property at \(\tau _{2j}\),
and, substituting \(g=1\),
Consequently,
and, using again the strong Markov property at time \(\tau _{2}\),
with the reversed inequality if the \(\inf \) is replaced by \(\sup \). Writing \(p=\mathbb {P}^{x} (Z_{k'} \ge 1) =1-1/(k'-k)\) whenever \(x\in \partial B_{d}(0,h_{k'-1})\), c.f. (2.7), it thus follows that
with the reversed inequality if the \(\inf \) and \(\sup \) are interchanged.
As in (2.8), let \(p_{B_{d}( 0, h_{k})}( z,x)\) denote the Poisson kernel for \(B_{d}( 0, h_{k})\subseteq \mathbf{S}^{ 2}\), see (2.8).
Recall that
Therefore, we get the Harnack inequality
Writing \(\alpha =(h_k-h_{k'-1})/2\), \(\beta =h_{k'-1}\) we have
Using the bounds, valid for \(|\alpha |,|\beta |<0.1\), \(|\sin (\alpha )|\ge |2\alpha /3|\), \(|\cos \beta -1|<|\beta |/2\), we obtain that the right side of (9.6) is bounded by \(1+4h_{k'-1}/h_k\), and therefore the right side of (9.5) is bounded by \(1+10h_{k'-1}/h_k\). Substituting this bound into (9.4) and using the value of p yields the claim. \(\square \)
Proof of Lemma 9.1
This follows from Lemma 9.2 in the same manner as [16, Lemma 7.3] was derived from [16, Lemma 7.4]: using the strong Markov property, each excursion from \(\partial B_{d}(y,h_{k})\) to \(\partial B_{d}(y,h_{k-1})\) will contribute a multiplicative factor \((1+10 (k'-k)h_{k'-1}/{h_{k}})\) to the probability. We omit further details. \(\square \)
About this article
Cite this article
Belius, D., Rosen, J. & Zeitouni, O. Tightness for the cover time of the two dimensional sphere. Probab. Theory Relat. Fields 176, 1357–1437 (2020). https://doi.org/10.1007/s00440-019-00940-2
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-019-00940-2