Statistics & Probability Letters 64 (2003) 133 – 145 Precise asymptotics in laws of the iterated logarithm for Wiener local time Ji-Wei Wen∗ , Li-Xin Zhang Department of Mathematics, Zhejiang University, Xixi Campus, Hangzhou 310028, China Received May 2002; received in revised form November 2002 Abstract In this paper, we study the asymptotic properties of the upper and lower tail probabilities of the maximum local time L∗ (t) of Wiener process (Brownian motion), and obtain some precise asymptotics in the law of the iterated logarithm and the Chungs-type laws of the iterated logarithm for the supremum of Wiener local time L(x; t); x ∈ R; t ∈ R+ . c 2003 Elsevier B.V. All rights reserved. MSC: 60F17; 60G15 Keywords: The law of the iterated logarithm; Local time; Brownian motion; Precise asymptotics 1. Introduction and main results Let {W (t); t ¿ 0} be a standard one-dimensional Wiener process (Brownian motion) and let L(x; t); x ∈ R; t ∈ R+ be its jointly continuous local time. Put L(0; t)=L(t) and supx∈R L(x; t)=L∗ (t). Kesten (1965) proved the iterated logarithm laws lim sup L∗ (t) =1 (2t log log t)1=2 lim inf L∗ (t)(log log t)1=2 = t 1=2 t →∞ a:s: (1.1) and t →∞ a:s:; Research supported by National Natural Science Foundation of China (No. 10071072). Corresponding author. E-mail addresses: jiweiwen@163.com (J.-W. Wen), lxzhang@mail.hz.zj.cn (L.-X. Zhang). ∗ c 2003 Elsevier B.V. All rights reserved. 0167-7152/03/$ - see front matter doi:10.1016/S0167-7152(03)00142-1 (1.2) 134 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 where the exact value of was not given by Kesten, he only estimated it from both sides. By using a result of Borodin (1982) concerning the Laplace transform of the distribution of L∗ (t), CsEaki and FGoldes (1986) gave the following expression for the exact and asymptotic distributions of L∗ (t). From the scale change of the Wiener process it follows that t −1=2 L∗ (t) has the same distribution as L∗ (1). Theorem A. For any x ¿ 0, P(L∗ (1) ¡ x) = ∞ k=1 ∞ 2j 2 ck 2k 2 2 ; ak exp − 2k + bk + 2 exp − 2 x x x (1.3) k=1 where 0 ¡ j1 ¡ j2 ¡ · · · are the positive zeros of the Bessel function J0 (x) = ∞ (−1)k x 2k (k!)2 2 k=0 and ak = 4 sin2 jk (k = 1; 2; : : :); J1 (k) − bk = 4 −1 + kJ0 (k) ck = 16k J1 (x) = J1 (k) J0 (k) ∞ k=0 Furthermore, J1 (k) J0 (k) (k = 1; 2; : : :); (k = 1; 2; : : :); (−1)k x 2k+1 : k!(k + 1)! 2 2j 2 P(L (1) ¡ x) ∼ a1 exp − 21 x ∗ 2 as x → 0: (1.4) As a consequence due √ √ to CsEaki and FGoldes (1986), the exact value in (1.2) turns out to be equal to 2j1 ≈ 2:4048 2 ≈ 3:4009. CsEaki (1989) studied an integral test for the supremum of Wiener local time, and gave a large deviation estimate for the distribution of L∗ (t). 1=2 2 2 ∗ P(L (1) ¿ x) ∼ 4 xe−x =2 as x → ∞: (1.5) Recently, Zhang (2001a, b) studied the precise asymptotics in the lim sup-type and lim inf-type laws of the iterated logarithm for partial sums of i.i.d.r.v’s. In those papers, by using the strong approximation and Feller’s and Einmahl’s truncation method, he showed many sharper results than that of Gut and SpKataru (2000), obtained the best necessary and suLcient conditions. J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 135 Inspired by Gut and SpKataru (2000) and Zhang (2001a, b), we investigate the corresponding asymptotic properties for the local time of Brownian motion, and obtain the following results. Theorem 1.1. Let ¿ − 1 and ¿ − n () log log n → 3 2 and let n () be a function of such that as n → ∞ and √ 1 + ; (1.6) where is a real number. Then lim (2 − − 1)+3=2 √ 1+ =8 ∞ (log n) (log log n) n n=3 P{L∗ (1) ¿ 2 log log n( + n ())} +1 3 1=2 exp{−2(1 + ) } + : 2 (1.7) Here, (·) is the gamma function. Theorem 1.2. For any ¿ − 1, we have lim 2(+1) ∞ (log log n) 0 n=3 n log n P{L∗ (1) ¿ 2 log log n} (1.8) = 2−(+1) ( + 1)−1 E(L∗ (1))2(+1) : Notice that L∗ (1)= 2 log log n and L∗ (n)= 2n log log n have the same distribution, Theorems 1.1 and 1.2 give the asymptotic properties of lim sup-type law of the iterated logarithm of L∗ (·). As corresponding properties of lim inf-type (Chung’s type) law of the iterated logarithm of L∗ (·), we show in Theorems 1.3 and 1.4. Theorem 1.3. Let ¿ − 1 and ¿ − 1 and let n () be a function of such that n () log log n → √ as n → ∞ and 1= 1 + ; (1.9) where is a real number. Then lim √ 1= 1+ = a1 √ 1 − 1+ 1 2(1 + )3=2 +1 ∞ +1 n=3 2 (log n) (log log n) 2j1 P L∗ (1) 6 ( + n ()) n log log n exp {2(1 + )3=2 }( + 1): Here a1 and j1 are de;ned in Theorem A. (1.10) 136 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 Theorem 1.4. For any ¿ − 12 , we have ∞ 2 (log log n) 2j 1 P L∗ (1) 6 = ¡ ∞; lim −2(+1) ∞ n log n log log n (1.11) n=3 where = ∞ k=1 2(+1) 2(+1) j1 j1 ( + 1)ck j1 2(+1) ak ( + 1): + bk + jk k 2k 2 2 k (1.12) Here, jk (k = 1; 2; : : :) are de;ned in Theorem A. 2. Proofs of the theorems In this section, we prove Theorems 1.1–1.4. Proof of Theorem 1.1. First, note that the limit in (1.7) does not depend on any Mnite terms of the inMnite series. Secondly, by (1.5) and condition (1.6) we have P{L∗ (1) ¿ 2 log log n( + n ())} 1=2 2 ∼4 ( + n ()) 2 log log n exp{−( + n ())2 log log n} 8 ∼√ log log n exp{−2 log log n} exp{−2n () log log n} √ √ as n → ∞, uniformly in ∈ ( 1 + ; 1 + + ) √ for some√ ¿ 0. So, for any 0 ¡ ¡ 1, there exist ¿ 0 and n0 such that for all n ¿ n0 and ∈ ( 1 + ; 1 + + ), √ 1 + log log n exp{−2 log log n} exp{−2 1 + − } 8 6 P{L∗ (1) ¿ 2 log log n( + n ())} √ 1 + 68 log log n exp{−2 log log n} exp{−2 1 + + } (2.1) by condition (1.6) again. And lim (2 − − 1)+3=2 √ 1+ ∞ (log n) (log log n)+1=2 n=3 n exp{−2 log log n} J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 2 = lim ( − − 1) √ +3=2 1+ 1+ = ∞ y 0 +1=2 −y e ∞ e (2 − − 1)+3=2 = lim √ ∞ 0 137 (log x) (log log x)+1=2 exp{−2 log log x} d x x y+1=2 exp{−y(2 − − 1)} dy 3 dy = + 2 : (2.2) So, combining (2.1) and (2.2), we have 8 √ 1+ 3 exp{−2 1 + − } + 2 6 lim (2 − − 1)+3=2 √ 1+ 68 ∞ (log n) (log log n) n=3 n P{L∗ (1) ¿ 2 log log n( + n ())} √ 1+ 3 : exp{−2 1 + + } + 2 Now, let → 0, (1.7) is proved. Proof of Theorem 1.2. We have lim 2(+1) 0 ∞ (log log n) n log n n=3 = lim 0 = 2− 2(+1) 0 e ∞ ∞ P{L∗ (1) ¿ 2 log log n} (log log x) P{L∗ (1) ¿ 2 log log x} d x x log x y2+1 P{L∗ (1) ¿ y} dy = 2−(+1) ( + 1)−1 E(L∗ (1))2(+1) : The theorem is now proved. Now, we turn to prove Theorem 1.3. Proof of Theorem 1.3. As in the proof of Theorem 1.1, we note that the limit in (1.10) does not depend on any Mnite terms of the inMnite series. By (1.4) and condition (1.9) we have 2 log log n 2j1 P L∗ (1) 6 ( + n ()) ∼ a1 exp − log log n ( + n ())2 138 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 log log n 2 + 2n () + n2 () 2 log log n exp 3 n () log log n ∼ a1 exp − 2 = a1 exp − √ √ for any as n → ∞, uniformly in ∈ (1= 1 + − ; 1= 1 + ) for some ¿ √0. So, by (1.9), √ 0 ¡ ¡ 1, there exist ¿ 0 and n0 such that for all n ¿ n0 and ∈ (1= 1 + − ; 1= 1 + ), log log n exp{2(1 + )3=2 − } a1 exp − 2 2 2j1 ( + n ()) 6 P L∗ (1) 6 log log n log log n 6 a1 exp − exp{2(1 + )3=2 + } 2 (2.3) by condition (1.9) again. And lim √ 1= 1+ 1 √ − 1+ +1 ∞ log log n (log n) (log log n) exp − n 2 n=3 log log x (log x) (log log x) exp − = lim dx √ x 2 1= 1+ e +1 ∞ 1 1 √ − y exp −y 2 − 1 − dy = lim √ 1+ 1= 1+ 0 +1 −−1 ∞ 1 1 √ − −1− y e−y dy = lim √ 2 1+ 1= 1+ 0 +1 ∞ +1 1 1 −y y e dy = ( + 1): = 2(1 + )3=2 2(1 + )3=2 0 1 √ − 1+ +1 ∞ (2.4) Hence, combining (2.3) and (2.4), we have a1 1 2(1 + )3=2 +1 6 lim √ 1= 1+ √ exp{2(1 + )3=2 − }( + 1) 1 − 1+ +1 ∞ n=3 2 (log n) (log log n) 2j1 P L∗ (1) 6 ( + n ()) n log log n J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 6 a1 1 2(1 + )3=2 +1 139 exp{2(1 + )3=2 + }( + 1): Let → 0, (1.10) is now proved. To prove Theorem 1.4, we need some lemmas. The Mrst two of them can be found in Zwillinger (1996, Section 6.18). Lemma 2.1. Let 0 ¡ j1 ¡ j2 ¡ · · · be the positive zeros of the Bessel function J0 (x)= (k!)2 )(x=2)2k . Then 1 ; n → ∞; j n = !n + O n ∞ k=0 ((−1) k = (2.5) where !n = (n − 14 ). Lemma 2.2. Let Jv (x) (v = 0; 1) be de;ned as in Theorem A. Then 1 2 1 1 cos x − v − + O ; x → ∞: Jv (x) = x 2 4 x Lemma 2.3. We have sin jn = (−1) n+1 (2.6) √ 2 1 ; +O 2 n n → ∞: (2.7) Proof. By Lemma 2.1, write jn = !n + n . Then sin jn = sin !n cos n + cos !n sin n √ 1 n+1 2 = (−1) cos n + O 2 n √ √ 1 n+1 2 n+1 2 = (−1) + (−1) (cos n − 1) + O 2 2 n √ √ 1 2 n = (−1)n+1 + (−1)n 2 sin2 +O 2 2 n √ 1 n+1 2 = (−1) +O ; n → ∞: 2 n Lemma 2.4. Let ak ; bk ; ck (k = 1; 2; : : :) be de;ned as in Theorem A. Then ck = −16: lim ak = 8; lim bk = −8; lim k →∞ k →∞ k →∞ k (2.8) 140 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 Proof. By Lemma 2.3, √ sin jk = (−1) k+1 we have 4 ak = = sin2 jk 1 2 1 2 +O 2 k (k → ∞); 4 1 ; =8+O k + O(1=k) k→∞ and by Lemma 2.2 √ J1 (k) cos (k − 34 ) + O(1=k) (−1)k+1 ( 2=2) + O(1=k) 1 √ = ; = −1 + O = 1 J0 (k) cos (k − 4 ) + O(1=k) k (−1)k ( 2=2) + O(1=k) k → ∞: So, ck = 16k J1 (k)=J0 (k) ∼ −16k; k → ∞; J1 (k) 2 1 J1 (k) − ; = −8 + O bk = 4 −1 + kJ0 (k) J0 (k) k k → ∞: Lemma 2.5. For any x ¿ 0; ¿ − 12 , the following series are all absolutely convergent: ∞ ∞ ∞ 2jk2 2k 2 2 2k 2 2 ; ; ak exp − 2 ; ak exp − 2 bk exp − 2 x x x k=1 ∞ k=1 2k 2 2 ; ck exp − 2 x k=1 k=1 ∞ −2(+1) a k jk ; k=1 ∞ ak k −2(+1) ; k=1 ∞ ck −2(+1) k : k 2 2 k=1 Proof. By Lemmas 2.1 and 2.4, it is obvious. Lemma 2.6. For 0 ¡ x 6 x0 , we have ∞ ∞ 2 2 2 2j c 2k k ak exp − 2k + bk + 2 exp − 2 x x x k=k0 +1 k=k0 +1 1 22 2j12 #(k0 ; x0 ); 6 exp − 2 + 1 + 2 exp − 2 x x x where #(k0 ; x0 ) = exp 2j12 x02 + exp ∞ 2j 2 |ak | exp − 2k x0 k=k +1 0 ∞ 22 x02 (2.9) 2k 2 2 (|bk | + |ck |) exp − 2 x0 k=k +1 0 (2.10) J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 141 and #(k0 ; x0 ) → 0 as k0 → ∞ for fixed x0 ¿ 0: Proof. Note that, for 0 ¡ x 6 x0 , 2jk2 2j12 2(jk2 − j12 ) exp − 2 = exp − 2 exp − x x x2 2j12 2(jk2 − j12 ) 6 exp − 2 exp − x x02 2 2j1 2jk2 2j12 exp − 2 ; = exp − 2 exp x x02 x0 22 2(k 2 − 1)2 2k 2 2 = exp − 2 exp − exp − 2 x x x2 22 2(k 2 − 1)2 6 exp − 2 exp − x x02 2 2 2k 2 2 22 exp − 2 : 6 exp − 2 exp x x02 x0 The proof is completed by Lemma 2.5. Proof of Theorem 1.4. We recall which is deMned in (1.12). By Lemma 2.5, it is obvious that ¡ ∞. Now, let M be a positive real number. For any ¿ 0, put B() = exp(exp(2 =M )); k0 ¿ 1, then ∞ 2 −2(+1) (log log n) 2j1 ∗ − (2.11) I () =: P L (1) 6 n log n log log n n=3 6 I1 (; M ) + I2 (; M; k0 ) + I3 (M; k0 ) + I4 (; M; k0 ) + I5 (k0 ); where I1 (; M ) = −2(+1) 2 (log log n) 2j1 P L∗ (1) 6 ; n log n log log n n¡B() 2 k0 (log log n) j log log n I2 (; M; k0 ) = −2(+1) ak exp − k 2 2 n log n j1 n¿B() k=1 + ck log log n bk + 22 j12 k 2 2 exp − 2 2 log log n j1 (2.12) 142 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 k0 2(+1) 2(+1) ∞ j1 j1 y e−y dy + bk y e−y dy 2 2 2 j k 2 2 k jk =j1 M k =j1 M k=1 2(+1) ∞ j1 ck +1 −y y e dy + ; 2 2 2 2k k 2 2 k =j1 M − ∞ ak k 2(+1) ∞ 0 j1 2(+1) ∞ j1 −y ak y e dy + bk y e−y dy I3 (M; k0 ) = 2 2 2 jk k 2 2 jk =j1 M k =j1 M k=1 ck + 2 2 2k − k0 j1 k 2(+1) ∞ k 2 2 =j12 M y+1 e−y dy 2(+1) 2(+1) j1 j1 ( + 1)ck j1 2(+1) ak ( + 1) + bk + ; jk k 2k 2 2 k k=1 I4 (; M; k0 ) = −2(+1) (log log n) n log n n¿B() 2 ∞ jk log log n × ak exp − 2 j12 k=k0 +1 + I5 (k0 ) = ck log log n bk + 22 j12 ; k 2 2 exp − 2 2 log log n j1 ∞ k=k0 +1 2(+1) 2(+1) j1 j1 ( + 1)|ck | j1 2(+1) |ak | ( + 1): + |bk | + jk k 2k 2 2 k Notice that for each k ¿ 1, lim −2(+1) ∞ 2 (log log n) j log log n ak exp − k 2 2 n log n j1 n¿B() 2 2 k ck log log n exp − 2 2 log log n + bk + 22 j12 j1 ∞ (log log x) jk2 −2(+1) ak exp − 2 2 log log x = lim ∞ x log x j1 B() 2 2 k ck log log x exp − 2 2 log log x dx + bk + 22 j12 j1 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 143 2(+1) 2(+1) ∞ j1 j1 −y = ak y e dy + bk y e−y dy jk k jk2 =j12 M k 2 2 =j12 M 2(+2) ∞ j1 ck + y+1 e−y dy: 2 2 k k 2 2 =j12 M 2k ∞ It follows that for any M ¿ 1; k0 ¿ 1: lim I2 (; M; k0 ) = 0: (2.13) ∞ For I1 (; M ), we have (log log n) n log n I1 (; M ) 6 −2(+1) n6B() 6 +1 = −2(+1) +1 +1 B() (log log x) +1 d x 6 −2(+1) x log x +1 e 1 M +1 2 M +1 ; (2.14) where +1 is an absolute constant. For I4 (; M; k0 ), since x= 2j12 =log log n 6 x0 = : 2j12 =log log B()= 2Mj12 , whenever n ¿ B(). So by Lemma 2.6, I4 (; M; k0 ) 6 −2(+1) (log log n) 2j12 1 22 exp − 2 + 1 + 2 exp − 2 #(k0 ; 2Mj12 ) n log n x x x n¿B() 6 −2(+1) (log log n) n log n n¿B() log log n × exp − 2 log log n + 1+ 22 j12 2 log log n exp − 2 j12 #(k0 ; It follows that lim sup I4 (; M; k0 ) ∞ −2(+1) 2 6 #(k0 ; 2Mj1 ) lim ∞ (log log x) ∞ x log x e 2 log log x log log x log log x × exp − + 1+ exp − dx 2 22 j12 2 j12 2Mj12 ): 144 J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 = #(k0 ; 2Mj12 ) ∞ 0 6 +2 #(k0 ; 2Mj12 ); y 2 2 y e−y + 1 + 2 e−( =j1 )y dy 2j1 (2.15) where +2 = +2 () = (1 + (j1 =)2(+1) + [( + 1)=22 ](j1 =)2(+1) )( + 1) ¡ ∞. So, combining (2.11) – (2.15), we obtain +1 lim sup I () 6 +1 ∞ 1 M +1 + +2 #(k0 ; 2Mj12 ) + I3 (M; k0 ) + I5 (k0 ): Notice that by Lemmas 2.5 and 2.6, limk0 →∞ #(k0 ; show that I3 (M; k0 ) → 0 It is obvious that sup I3 (M; k0 ) 6 k0 2Mj12 ) = 0; limk0 →∞ I5 (k0 ) = 0. It remains to as M → ∞ uniformly in k0 : ∞ 0 (2.16) y e−y dy = ( + 1); ( + 2) = ( + 1)( + 1). Hence, we have ∞ k=1 2(+1) j2 =j2 M k 1 j1 |ak | y e−y dy jk 0 2 2 2 j1 2(+1) (k )=( j1 M ) −y y e dy + |bk | k 0 2(+1) (k 2 2 )=( j2 M ) 1 |ck | j1 +1 −y + 2 2 y e dy : 2k k 0 By Lemma 2.5, it follows that lim sup I3 (M; k0 ) 6 M →∞ k0 ∞ k=1 + 2(+1) jk2 =j12 M j1 |ak | lim y e−y dy M →∞ 0 jk ∞ k=1 |bk | j1 k 2(+1) lim M →∞ 0 (k 2 2 )=( j12 M ) y e−y dy 2(+1) (k 2 2 )=( j12 M ) ∞ j1 |ck | + lim y+1 e−y dy M →∞ 0 2k 2 2 k k=1 = 0: The proof is now completed. (2.17) J.-W. Wen, L.-X. Zhang / Statistics & Probability Letters 64 (2003) 133 – 145 145 References Borodin, A.N., 1982. Distribution of integral functionals of Brownian motion. Zap. NauScn. Sem. Leningrad. Otdel. Mat. Inst. Steklova 119, 19–38. CsEaki, E., 1989. An integral test for the supremum of Wiener local time. Probab. Theory Related Fields 83, 207–217. CsEaki, E., FGoldes, A., 1986. How small are the increments of the local time of a Wiener process? Ann. Probab. 14, 533–546. Gut, A., SpKataru, A., 2000. Precise asymptotics in the law of the iterated logarithm. Ann. Probab. 28, 1870–1883. Kesten, H., 1965. An iterated logarithm law for local time. Duke Math. J. 32, 447–456. Zhang, L.X., 2001a. Precise rates in the law of the iterated logarithm. Preprint. Zhang, L.X., 2001b. Precise asymptotics in Chung’s law of the iterated logarithm. Preprint. Zwillinger, D., 1996. CRC Standard Mathematical Tables and Formulae, 30th Edition. CRC Press, Boca Raton, FL.