A tamed-adaptive Milstein scheme for stochastic differential equations with low regularity coefficients
Thi-Huong Vu, Hoang-Long Ngo, Duc-Trong Luong, Ngoc Khue Tran · 2024
Read A tamed-adaptive Milstein scheme for stochastic differential equations with low regularity coefficients by Thi-Huong Vu, Hoang-Long Ngo, Duc-Trong Luong, Ngoc Khue Tran (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2411-01849.txt.
Description
A 2024 arXiv research paper by Thi-Huong Vu, Hoang-Long Ngo, Duc-Trong Luong, Ngoc Khue Tran. We propose a tamed-adaptive Milstein scheme for stochastic differential equations in which the first-order derivatives of the coefficients are locally Hölder continuous of order α \alpha .
Catalog record
| CREATOR | Thi-Huong Vu, Hoang-Long Ngo, Duc-Trong Luong, Ngoc Khue Tran |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 103 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2411-01849
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A tamed-adaptive Milstein scheme for stochastic differential equations with low regularity coefficients Thi-Huong Vu, Hoang-Long Ngo, Duc-Trong Luong, Ngoc Khue Tran Abstract We propose a tamed-adaptive Milstein scheme for stochastic differential equations in which the first-order derivatives of the coefficients are locally Hölder continuous of order α \alpha . We show that the scheme converges in the L 2 L_{2} -norm with a rate of ( 1 + α ) / 2 (1+\alpha)/2 over both finite intervals [ 0 , T ] [0,T] and the infinite interval ( 0 , + ∞ ) (0,+\infty) , under certain growth conditions on the coefficients. Keywords: Milstein approximation; Tamed Milstein; Adaptive Milstein; Stochastic differential equations; Locally Lipschitz continuous; Locally Hölder continuous; Polynomial growth coefficient. 1 Introduction This paper considers the numerical approximation for a one-dimensional process X=(X_{t})_{t\geq 0} which is a solution to the following stochastic differential equation (SDE) X_{t}=x_{0}+\int_{0}^{t}\mu(X_{s})ds+\int_{0}^{t}\sigma(X_{s})dW_{s},\quad x_{0}\in\mathbb{R},\quad t\geq 0, (1) where W=(W_{t})_{t\geq 0} is a one-dimensional standard Brownian motion defined on a complete filtered probability space (\Omega,\mathcal{F},(\mathcal{F}_{t}),\mathbb{P}) . Stochastic differential equations are widely employed in modeling systems across various fields, including biology, physics, and economics [ 1 , 21 , 25 ] . Due to the limited availability of explicit solutions, numerical approximation methods have become indispensable for many applications. The convergence rates of these numerical schemes typically depend on the growth conditions and smoothness of the coefficients. It is well-known that for SDEs whose coefficients \mu and \sigma satisfy global Lipschitz continuity and linear growth conditions, the Euler–Maruyama method attains a strong convergence rate of order 1/2 . Furthermore, if \mu is once differentiable and \sigma is twice differentiable, then the Milstein scheme attains a higher strong convergence rate of order 1 (see [ 7 , 15 , 23 ] ). However, recent studies [ 8 ] revealed that for SDEs with superlinearly growing coefficients, classical explicit schemes such as the Euler-Maruyama and Milstein schemes may fail to converge in the L_{p} -norm. This limitation has motivated the development of various alternative numerical methods tailored for SDEs with non-globally Lipschitz coefficients. The tamed Euler-Maruyama scheme was introduced in [ 9 ] and further developed in [ 29 , 30 , 24 ] . Another notable approach is the truncated Euler scheme introduced in [ 22 ] . These modified schemes are designed to handle superlinear growth and have been shown to achieve a strong convergence rate of order 1/2 for certain classes of SDEs, under appropriate one-sided Lipschitz and monotonicity conditions. Several implicit Milstein-type schemes have been developed for nonlinear SDEs, which achieve strong convergence of order 1 ; see, for example, [ 6 , 26 , 33 ] . However, it is important to emphasize that implicit methods generally entail significant computational cost, as they require solving nonlinear algebraic equations at each time step. To overcome this limitation, explicit variants have been proposed. Notably, the tamed Milstein scheme introduced in [ 31 ] attains strong convergence of order 1 for a class of SDEs with superlinearly growing coefficients. Besides, the truncated Milstein scheme, initially investigated in [ 22 ] for SDEs with commutative noise, was shown in [ 4 ] to achieve the same convergence rate under the assumption that both the drift and diffusion coefficients possess continuous second-order derivatives and are polynomially bounded. Further refinements and generalizations of the tamed and truncated Milstein schemes can be found in [ 2 , 4 , 16 , 17 , 18 , 19 , 20 , 28 , 10 ] , and references therein. An alternative to fixed-step numerical methods is the class of adaptive schemes, in which each time step is dynamically determined based on the current state of the approximated process. Fang and Giles [ 3 ] proposed an adaptive Euler–Maruyama scheme tailored for SDEs with polynomially growing drift, satisfying a one-sided Lipschitz condition, and with bounded, globally Lipschitz continuous diffusion coefficients. Building on this framework, a tamed-adaptive Euler-Maruyama scheme was developed in [ 12 , 13 , 14 ] by integrating the taming technique into the adaptive time-stepping strategy, thereby extending the applicability of explicit methods to a broader class of SDEs with superlinear growth. The tamed-adaptive Euler-Maruyama scheme is applicable to SDEs whose drift coefficients are both locally Lipschitz continuous and one-sided Lipschitz continuous, and whose diffusion coefficients are locally (\alpha+1/2) -Hölder continuous. In [ 11 ] , the authors introduced a class of path-dependent time-stepping strategies, wherein the step size is adaptively reduced when the numerical solution approaches the boundary of a prescribed sphere. This approach was employed to construct an explicit adaptive Milstein scheme applicable to SDEs with non-commutative noise, i.e., where the commutativity condition is not satisfied. It was demonstrated that this scheme achieves strong convergence of order 1 under the assumption that the drift coefficient satisfies a one-sided Lipschitz condition. In this paper, we propose a tamed-adaptive Milstein scheme for equation ( 1 ), which synthesizes key elements from both the tamed Milstein method and adaptive time-stepping strategies. Notably, due to the adaptive control of the step size, it is not necessary to tame the drift coefficient b or the diffusion coefficient \sigma . Instead, only the composite term \sigma\sigma^{\prime} requires taming. This distinction sets our approach apart from the methods introduced in [ 12 , 13 , 14 ] , highlighting a fundamental difference in the treatment of nonlinearities. The novel contribution of this work lies in the significant relaxation of the regularity assumptions imposed on the coefficients of the SDE. In contrast to prior studies that typically require global Lipschitz continuity, we assume only local \alpha -Hölder continuity of the first derivatives of the drift and diffusion coefficients, \mu^{\prime} and \sigma^{\prime} . Under these relaxed conditions, we prove that the proposed tamed-adaptive Milstein scheme strongly converges in the L_{2} -norm with a rate of (1+\alpha)/2 , thereby quantifying the influence of the regularity of \mu^{\prime} and \sigma^{\prime} on the convergence behavior of the Milstein method. Notably, when \alpha=1 , the strong convergence rate reaches the optimal order of 1 . Moreover, we conduct a detailed analysis of the strong convergence rate on both finite time intervals [0,T] , for fixed T>0 , and the infinite time horizon [0,\infty) . The study of numerical schemes over infinite time intervals has only recently attracted attention in the literature (see [ 3 , 12 , 13 , 14 ] ). In this context, the integrability properties of both the exact and approximate solutions play a critical role in determining the attainable convergence rate. The remainder of this paper is structured as follows. In Section 2 , we outline the assumptions and main results of the model. Specifically, we define the tamed-adaptive Milstein scheme and present key theorems regarding moment estimates for the approximate process and the strong convergence of the scheme. All proofs are deferred to Section 3 . In Section 4 , we provide a numerical analysis of the tamed-adaptive Milstein scheme to illustrate the theoretical findings. We also compare the performance of the tamed-adaptive Milstein scheme with that of the tamed Milstein scheme proposed in [ 17 ] . Technical estimates for the increments of both exact and approximate solutions are provided in Section 5 . 2 Model assumptions and