Small time Chung-type LIL for Lévy processes (Q1940751): Difference between revisions

General Lévy random processes are considered and Chung-type laws of the iterated logarithm (LIL) at zero are investigated. Given the Lévy process \(X = (X_t)\), \(t \geq 0\), it is characterized a norming function \({b}\), satisfying \[ \lim \inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{b(t)} = 1, \quad\text{where } \parallel X\parallel_t := \sup_{0\leq s\leq t} |X_s|. \] Via the Borel-Cantelli lemma, Chung-type LIL for a general random process is connected to the so-called small deviation rate of the process \[ -\log \operatorname{P}(\parallel X\parallel_t \leq \varepsilon), \quad\text{as } \varepsilon \rightarrow 0 \text{ and } t \rightarrow 0. \] In this paper, it is show how the asymptotics of (2) imply explicit LIL. It is not sufficient to have estimates for (2) for fixed \(t\) only. The paper is organized as follows. The introduction presents the problem statements and an overview of earlier results on small deviations of Lévy processes, the LIL for special Lévy processes that have already appeared in the literature. In Section 2, the main results which manage the transfer between small deviations and LIL are stated. Several examples of LIL for concrete Lévy processes are considered in Section 3. The proofs are given in Section 4. Let \({X}\) be a Lévy process with characteristic triplet (\(\gamma, \sigma^2, \Pi\)), where \(\gamma \in \mathbb{R}\), \(\sigma^2 \geq 0\) and the Lévy measure \(\Pi\) has no atom at zero and satisfies \[ \int (1 \wedge x^2) \Pi(dx) < \infty. \] As only the behavior for small times is of interest, it is assumed that \(\operatorname{E} e^{izX_i} =: e^{t\psi(z)}\), has the form \[ \psi(z) = i\gamma z - \frac{\sigma^2 z^2}{2} + \int^1_{-1} (e^{izx} - 1 - izx) \Pi(dx), \quad z \in \mathbb{R}. \] Denote by \(\Phi\) the Laplace exponent of a subordinator \({A}\), \(\operatorname{E} e^{-uA_1} = e^{-\Phi(u)}\), \[ \Phi(u) = u\gamma_A + \int^\infty_0 (1 - e^{-ux}) \Pi_A(dx). \] For the two-sided tail of the Lévy measure \(\Pi\), set \(\bar{\Pi}(\varepsilon) := \Pi([-\varepsilon, \varepsilon]^c)\). In the following, \(f \sim g\) denotes the strong asymptotic equivalence, that is, \(\lim f/g = 1\), and \(f \approx g\) the weak asymptotic equivalence, \(0 < \lim \inf f/g \leq \lim \sup f/g < \infty\). The main result is Theorem 2.1. Let \({X}\) be a Lévy process that has jumps of absolute value smaller than one. Let \({F}\) be a function increasing to infinity at zero, such that, for some \(0 < \lambda_1 \leq \lambda_2 < \infty\), \[ \lambda_1 F(\varepsilon)t \leq -\log \operatorname{P}(\parallel X\parallel_t < \varepsilon) \leq \lambda_2 F(\varepsilon)t \quad\text{for all } \varepsilon < \varepsilon_0\text{ and } t < t_0. \] Let \[ b_\lambda(t) := F^{-1} \left(\frac{\log |\log t|}{\lambda t}\right) \] for \(\lambda>0\). Assume that, as \(n \rightarrow \infty\), \[ (n+1)^{-(n+1)^\beta} \left|\int_{|x|>b_{\lambda^\prime_2} (n^{-n^\beta})} x\Pi(dx) - \gamma \right| = o(b_{\lambda^\prime_2}(n^{-n^\beta}))\tag{3} \] for all \(\beta>1\) and \(\lambda^\prime_2>\lambda_2\). Then, almost surely, for any \(\lambda^\prime_1>\lambda_1\) and \(\lambda^\prime_2>\lambda_2\), \[ 1 \leq \lim \inf_{t \rightarrow 0} \frac{\parallel X \parallel_t}{b_{\lambda^\prime_1}(t)} \quad\text{and}\quad \lim \inf_{t \rightarrow 0} \frac{\parallel X \parallel_t}{b_{\lambda^\prime_2}(t)} \leq 1. \] As corollaries of Theorem 2.1, the optimal limiting constants are obtained in the case when the function \({F}\) is regularly varying at zero with non-positive exponent and the process \({X}\) is a symmetric Lévy process with \(F(\varepsilon) = \varepsilon^{-2} U(\varepsilon)\), where \(U(\varepsilon)\) is the variance of \({X}\) with jumps larger than \(\varepsilon\) replaced by jumps of size \(\varepsilon\). Theorem 2.2 instead of (3) demands that there is a constant \(C>0\) such that \[ C b_\lambda(t) \leq b_\lambda \left(\frac{t}{2}\right), \quad 0<t\leq t_0,\;\lambda \in \left(\frac{\lambda_1}{2}, 2\lambda_2\right) \] in order for the same statement as in Theorem 2.1 to hold. As a corollary of Theorem 2.2, it follows that, almost surely, \[ \lim\inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{b_\lambda (t)} \in (0, \infty) \] if \({F}\) is regularly varying at zero with negative exponent. In Corollary 2.4 and Proposition 2.1, the conditions on the function \({F}\) are expressed in terms of (\(\gamma, \sigma^2, \Pi\)). In Chapter 3, explicit LIL are obtained for certain Lévy processes. It is proved that, for a Lévy process \({X}\) with \(\sigma \neq 0\), \[ \lim\inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{\sqrt{t/\log |\log t|}} = \frac{\pi \sigma}{\sqrt{8}} \quad\text{a.s.} \] Similarly to Lévy processes with non-zero Brownian component, symmetric processes of smaller deviation order are dominated by stable Lévy processes. When \(\Pi\) behaves as a regularly varying function at zero and is symmetric and \(\bar{\Pi}(\varepsilon) \approx \varepsilon^{-\alpha} |\log \varepsilon|^{-\gamma}\), as \(\varepsilon \rightarrow 0\) with \(0<\alpha<2\), or \(\alpha=2\), \(\gamma>1\), then (4) holds for certain \(b(t)\). For a Lévy process with triplet (\(\gamma, 0, \Pi\)), where \[ \frac{\Pi(dx)}{dx} = \frac{C_1 \mathbf{1}_{(0,1]}(x)}{x^{1+\alpha_1}} + \frac{C_2 \mathbf{1}_{[-1,0)}(x)}{(-x)^{1+\alpha_2}} \] with \(2>\alpha_1\geq \alpha_2\) and \(C_1, C_2\geq 0, C_1+C_2 \neq 0\), the path-wise behavior at zero is analyzed in the cases \(\alpha_1>1\), \(\alpha_1=1\), \(0<\alpha_1<1\), respectively, \(\alpha_2\) can be even negative. The path-wise behavior is considered for Lévy processes obtained from the Brownian motion \({B}\) by subordination, that is, \(X_t = \sigma B_{A_t}\). As a specific example, let \({A}\) be a gamma process. The so-called variance-gamma process is \[ X_t = \sigma B_{A_t} + \mu A_t \] for some constants \(\sigma \neq 0\), \(\mu \in \mathbb{R}\). Then, for \(\mu = 0\), there are constants \(0<\lambda_1 \leq \lambda_2<\infty\) such that almost surely \[ 1 \leq \lim \inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{e^{-\lambda_1 \log |\log t|/t}} \quad\text{and}\quad \lim \inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{e^{-\lambda_2 \log |\log t|/t}} \leq 1, \] and, for \(\mu \neq 0\), \[ \lim\inf_{t \rightarrow 0} \frac{\parallel X\parallel_t}{t} = |\mu| \operatorname{E}(A_1) \quad\text{a.s.} \]