- Research
- Open access
- Published:
Time-averaging principle for G-SDEs based on Lyapunov condition
Advances in Continuous and Discrete Models volume 2023, Article number: 28 (2023)
Abstract
In this paper, we tame the uncertainty about the volatility in time-averaging principle for stochastic differential equations driven by G-Brownian motion (G-SDEs) based on the Lyapunov condition. That means we treat the time-averaging principle for stochastic differential equations based on the Lyapunov condition in the presence of a family of probability measures, each corresponding to a different scenario for the volatility. The main tool for the mathematical analysis is the G-stochastic calculus, which is introduced in the book by Peng (Nonlinear Expectations and Stochastic Calculus Under Uncertainty. Springer, Berlin, 2019). We show that the solution of a standard equation converges to the solution of the corresponding averaging equation in the sense of sublinear expectation with the help of some properties of G-stochastic calculus. Numerical results obtained using PYTHON illustrate the efficiency of the averaging method.
1 Introduction
In mathematical finance, the traditional way of representing random fluctuations of financial quantities is to use a Brownian motion, which is typically scaled by the volatility constant. However, there is plenty of empirical evidence, derived from market prices, showing that the volatility of financial quantities is not constant and even not deterministic. The presence of volatility uncertainty leads to mathematical difficulties since the family of probability measures representing volatility uncertainty contains mutually singular measures. The canonical process B has a different volatility under each probability measure in a family of probability measures \(\mathcal{P}\). Thus the quadratic variation process \(\langle B \rangle = ( \langle B \rangle _{t} )_{t\geq 0}\) differs among the probability measures in \(\mathcal{P}\). For example, we consider the probability measures \(P^{\bar{\sigma}}\) and \(P^{\underline{\sigma}}\), induced by the constant volatilities σ̄ and \(\underline{\sigma}\), respectively, and we have
Therefore, the set \(\mathcal{P}\) contains mutually singular probability measures, that is, there are probability measures in the set of probability measures that have different null sets. This causes mathematical problems since many results from probability theory and stochastic calculus only hold up to null sets of the underlying measure. Important examples include the time consistent conditional expectations and stochastic integrals.
The averaging principle is an important property in the study of the dynamical behavior for nonlinear dynamical systems. The key technique of the averaging principle is time-scales separation. In particular, the averaging principle provides a powerful tool for simplifying dynamical systems and obtaining approximate solutions to differential equations arising from mechanics, mathematics, physics, control, and other areas. Averaging principles for stochastic systems were proposed by Stratonovich [19, 20] to examine nonlinear oscillation problems in the presence of random noise. Since then there has been a big amount of papers devoted to the study of the averaging principle for stochastic (partial) differential equations, see Khasminskii [12], Freidlin and Wentzell [2], Givon [7], Fu and Liu [3], Xu, Duan and Xu [22], Fu, Wan, Liu [4], Xu, Miao, Liu [21], etc.
From the point of view of fully nonlinear parabolic partial differential equations, in Hu and Wang [10] the authors perfectly established the averaging principle for stochastic differential equations driven by G-Brownian motion, where the condition on the coefficients is the global Lipschitz condition, see assumption (H1) in Hu and Wang [10]. From the same point of view, Hu, Jiang, and Wang [9] extended the one of Hu and Wang [10] to the forward-backward stochastic differential equations driven by G-Brownian motion with global Lipschitz condition. To the authors knowledge, averaging principles for stochastic differential equations with locally Lipschitz coefficients based on a Lyapunov condition under volatility uncertainty, that is, stochastic differential equations driven by G-Brownian motion (G-SDEs) under sublinear expectation, have not been considered. Therefore in this paper we consider these averaging principles in a family of probability measures \(\mathcal{P}\), where the coefficients in the stochastic differential equations have no global Lipschitz assumption. Another important difference from Hu, Jiang, and Wang [9] is that the convergence mode is different. As the reviewer pointed out to the author, the global Lipschitz is not essential, the main difference is that the author obtained a strong convergence result instead of weak convergence compared with Hu et al. [9], but the author needed quite a strong condition (B). Mao, Chen, and You [15] obtained an excellent result about the averaging principle for stochastic differential equations driven by G-Brownian motion with global non-Lipschitz coefficients. Recent important progress in the theory of volatility uncertainty/G-Brownian motion is reviewed by Peng [16] with comments on its explanation, theory, and significance.
In this paper, we study the averaging principle for the following stochastic differential equation with locally Lipschitz coefficients based on a Lyapunov condition under volatility uncertainty:
where the canonical process B is a 1-dimensional G-Brownian motion, which was introduced by Peng [16] and is used to represent the uncertainty about the volatility, and \(\langle B\rangle \). is a quadratic variation process of the G-Brownian motion B. From the construction of G-Brownian motion, Peng proved that the quadratic variation process \((\langle B\rangle _{t})_{t\geq 0}\) is an increasing stochastic process with \((\langle B\rangle _{t})_{0}=0\), and \(\langle B\rangle _{t}\) is not a deterministic process unless \((B_{t})_{t\geq 0}\) is a classical Brownian motion. For more details on G-Brownian motion, we refer to Peng [16].
The main difficulty of this paper is dealing with the local Lipschitz coefficients based on a Lyapunov condition. We shall apply the localization technique to approximate the solution of stochastic differential equation with locally Lipschitz coefficients. Based on the work of Li, Peng [14], Hu, Wang, Zheng [11], the space of suitable integrands is essentially expanded, which requires less regularity and only local integrability. Particularly, we will define the truncated G-SDEs that are uniform Lipschitz and carefully choose the stopping times to construct a consistent localized sequence. However, the Lyapunov-type condition ensures that the G-SDEs with the local Lipschitz coefficients can be approximated pathwisely by the truncated G-SDEs.
This paper is organized as follows. Section 2 introduces G-Brownian motion to represent the volatility uncertainty and related stochastic calculus briefly. In Sect. 3, we prove the averaging principle for stochastic differential equations in a family of probability measures. In Sect. 4, we present numerical simulation of stochastic differential equations under volatility uncertainty and give three examples to demonstrate the averaging method using PYTHON.
2 Preliminaries
We introduce in this section some of the basic notions relating to volatility uncertainty and G-Brownian motion and then recall some preliminary results in G-Brownian motion, which are needed in the sequel. More details can be found in Peng [16].
2.1 G-Brownian motion
Definition 2.1
Given a set Ω and a linear space \(\mathcal{H}\) of real-valued functions defined on Ω. Moreover, if \(X_{i}\in \mathcal{H}, i=1,2,\ldots, d\), then \(\varphi (X_{1},\ldots,X_{d}) \in \mathcal{H}\) for all \(\varphi \in C_{b, \mathrm{Lip}}(\mathbb{R}^{d})\), where \(C_{b, \mathrm{Lip}}(\mathbb{R}^{d})\) is the space of all bounded real-valued Lipschitz continuous functions. A sublinear expectation \(\hat{\mathbb{E}}\) on \(\mathcal{H}\) is a functional \(\hat{\mathbb{E}}: \mathcal{H}\to \mathbb{R}\) satisfying the following properties: for all \(X, Y\in \mathcal{H}\),
-
(i)
Monotonicity: If \(X\leq Y\), then \(\hat{\mathbb{E}}[X]\leq \hat{\mathbb{E}}[Y]\);
-
(ii)
Constant preserving: \(\hat{\mathbb{E}}[c]=c\) for any \(c\in \mathbb{R}\);
-
(iii)
Subadditivity: \(\hat{\mathbb{E}}[X+Y]\leq \hat{\mathbb{E}}[X]+\hat{\mathbb{E}}[Y]\);
-
(iv)
Positive homogeneity: \(\hat{\mathbb{E}}[\lambda X]= \lambda \hat{\mathbb{E}}[X] \) for any \(\lambda \geq 0\).
The triple \((\Omega, \mathcal{H}, \hat{\mathbb{E}})\) is called a sublinear expectation space.
Denote by \(\Omega = C_{0}(\mathbb{R}^{+})\) the space of all \(\mathbb{R}\)-valued continuous paths \((\omega _{t})_{t\in \mathbb{R}^{+}}\), with \(\omega _{0}=0\), equipped with the distance
\(\mathcal{B}(\Omega )\) is the Borel σ-algebra of Ω. For each \(t\in [0,\infty )\), we introduce the following spaces:
-
\(\Omega _{t}:= \{\omega (\cdot \wedge t): \omega \in \Omega \}, \mathcal{F}_{t}:= \mathcal{B}(\Omega _{t})\);
-
\(L^{0}(\Omega )\): the space of all \(\mathcal{B}(\Omega )\)-measurable real functions;
-
\(L^{0}(\Omega _{t})\): the space of all \(\mathcal{F}_{t}\)-measurable real functions;
-
\(B_{b}(\Omega )\): all bounded elements in \(L^{0}(\Omega )\), \(B_{b}(\Omega _{t}):= B_{b}(\Omega ) \cap L^{0}(\Omega _{t})\);
-
\(C_{b}(\Omega )\): all continuous elements in \(B_{b}(\Omega )\), \(C_{b}(\Omega _{t}):= C_{b}(\Omega ) \cap L^{0}(\Omega _{t})\).
-
\(C_{b, \mathrm{Lip}}(\mathbb{R}^{n})\): the space of all bounded \(\mathbb{R}\)-valued Lipschitz continuous functions on \(\mathbb{R}^{n}\).
Let \(\Omega = C_{0}\) be the space of all \(\mathbb{R}\)-valued continuous paths \((\omega _{t})_{t\geq 0}\) starting from origin, equipped with local uniformity, and let \(B_{t}(\omega )= \omega _{t}\) be the canonical process. For each \(t\in [0, \infty )\), define \(\Omega _{t}= \{\omega _{\cdot \wedge t}: \omega \in \Omega \}\). Set
and \(\mathrm{Lip}(\Omega )= \bigcup_{t\geq 0} \mathrm{Lip}(\Omega _{t})\). For each \(x\in \mathbb{R}\), we consider any given monotonic and sublinear function
Here, we assume that G is nondegenerate, i.e., \(0< \underline{\sigma}^{2} \leq \bar{\sigma}^{2} <\infty \). In Peng [16], a G-Brownian motion is constructed on a sublinear expectation space \((\Omega, \mathrm{Lip}(\Omega ), \hat{\mathbb{E}}, (\hat{\mathbb{E}}_{t})_{t \geq 0})\), which is called G-expectation space. In this space the corresponding canonical process \(B_{t}(\omega ) = \omega _{t}\) is a G-Brownian motion.
Let \(\mathbb{L}^{p}_{G}(\Omega )\) (respectively \(L_{\ast}^{p}(\Omega )\)) be the completion of \(\mathrm{Lip}(\Omega )\) (respectively \(B_{b}(\Omega )\)) under the natural norm \(\|X\|_{p} = \hat{\mathbb{E}}[|X|^{p}]^{1/p}\). Denis, Hu, and Peng [1] proved that
and there exists a weakly compact family \(\mathcal{P}\) of probability measures defined on \((\Omega, \mathcal{B}(\Omega ))\) such that
Then \(L_{\ast}^{p}(\Omega )\) and \(\mathbb{L}_{G}^{p}(\Omega )\) can be characterized as follows:
and
We will introduce two natural capacities:
and
Definition 2.2
A set \(A\in \mathcal{B}(\Omega )\) is polar if \(V(A) = 0\). The property holds quasi-surely (q.s.) if it holds outside a polar set.
In what follows, we do not distinguish between two random variables X and Y if \(X = Y\) q.s.
The following inequality is a capacity version of the Markov inequality.
Proposition 2.3
Let \(X\in L^{0}(\Omega )\) and \(\hat{\mathbb{E}}[|X|^{p}]< \infty, p>0\). For any \(x>0\), then
For the proof, see Lemma 6.1.17 in Peng [16].
2.2 G-stochastic calculus
Peng [16] also introduced the related stochastic calculus of Itô type with respect to G-Brownian motion. Now we recall Peng’s G-stochastic calculus from Li and Peng [14] or Chap. 8 in Peng [16], and let \(T>0\) be fixed.
Definition 2.4
Consider the following simple type of processes:
For an element \(\eta \in M_{b,0}(0,T)\) with \(\eta _{t}= \sum_{j=0}^{N-1} \xi _{j}(\omega )I_{[t_{j}, t_{j+1})} (t)\), the related Bochner integral is
Definition 2.5
For each \(p\geq 1\), we denote by \(M_{\ast}^{p} (0,T)\) the completion of \(M_{b,0} (0,T)\) under the norm
Definition 2.6
For each \(\eta \in M_{b,0} (0,T)\) of the form
define
The mapping \(I: M_{b,0}(0,T) \to L_{\ast}^{2}(\Omega _{T})\) can be continuously extended to \(I: M_{\ast}^{2}(0,T) \to L_{\ast}^{2}(\Omega _{T})\). For each \(\eta \in M_{\ast}^{2}(0,T)\), the stochastic integral is defined by
Definition 2.7
For fixed \(p\geq 1\), a stochastic process \((\eta _{t})_{t\geq 0}\) is said to be in \(M_{w}^{p}(0,T)\) if it is associated with a sequence of increasing stopping times \(\{\sigma _{n}\}_{n\in \mathbb{N}}\) such that
and if \(\Omega ^{(n)}:= \{\omega \in \Omega: \sigma _{n}(\omega )\wedge T +T \}\) and \(\hat{\Omega}:= \lim_{n\to \infty} \Omega ^{(n)}\), then \(V(\hat{\Omega}^{c})=0\).
Given \(\eta \in M_{w}^{2}(0,T)\) associated with \(\{\sigma _{n}\}_{n\in \mathbb{N}}\), we note \(\tau _{n}:= \sigma _{n}\wedge T\) and consider the continuous modification of \((\int _{0}^{t} \eta _{s} I_{[0, \tau _{n}]}(s)\,dB_{s})_{0\leq t\leq T}\). For each \(m,n\in \mathbb{N}, n>m\), we can find a polar set \(\hat{A}^{m,n}\) such that for all \(\omega \in (\hat{A}^{m,n})^{c}\) the following equality holds:
for \(0\leq t\leq T\). Define a polar set \(\hat{A}:= \bigcup_{m=1}^{\infty }\bigcup_{n=m+1}^{\infty }\hat{A}^{m,n}\). For each \(n\in \mathbb{N}\) and \((\omega, t) \in \Omega \times [0,T]\), we set
For each \(\omega \in \hat{A}^{c}\) and \(m,n\in \mathbb{N}, n>m, X^{n}(\omega ) \equiv X^{m}(\omega )\) on \([0, \tau _{m}(\omega )]\). Therefore we can define unambiguously a process by stipulating that it is equal to \(X^{m}\) on \([0, \tau _{m}(\omega )]\).
Definition 2.8
Let \(\eta \in M_{w}^{2}(0,T)\) for each \((\omega, t)\in \Omega \times [0,T]\), we define
It is important that the quadratic process of G-Brownian motion B is not always a deterministic process, and it can be formulated by
where \(t_{i}^{N} = (jT)/N\) for each integer \(N\geq 1\).
Definition 2.9
Define a mapping \(Q: M_{b,0}(0,T) \to \mathbb{L}_{\ast}^{1}(\Omega _{T})\):
Then Q can be uniquely extended to \(M_{w}^{1}(0,T)\), we also denote this mapping by
In view of the dual formulation of sublinear expectation as well as the properties of the quadratic variation process \(\langle B\rangle \) in the framework of sublinear expectation, we can generalize the following BDG-type inequalities in Gao [5] to \(\eta \in M_{w}^{p}([0,T])\).
Lemma 2.10
-
(1)
For each \(p\geq 1\) and \(\eta \in M_{w}^{p}(0,T)\),
$$\begin{aligned} \hat{\mathbb{E}}\biggl[\sup_{0\leq t\leq T} \biggl\vert \int _{0}^{t} \eta _{s} \,d \langle B \rangle _{s} \biggr\vert ^{p}\biggr] \leq \bar{ \sigma}^{2p} T^{p-1} \int _{0}^{T} \hat{\mathbb{E}}\bigl[ \vert \eta _{s} \vert ^{p}\bigr] \,ds. \end{aligned}$$ -
(2)
For each \(p\geq 2\) and \(\eta \in M_{w}^{p}(0,T)\), there exists some constant \(C_{p}\) depending only on p and T such that
$$\begin{aligned} \hat{\mathbb{E}}\biggl[\sup_{0\leq t\leq T} \biggl\vert \int _{0}^{t} \eta _{s} \,dB_{s} \biggr\vert ^{p}\biggr] \leq C_{p} \hat{\mathbb{E}} \biggl[ \biggl\vert \int _{0}^{T} \vert \eta _{s} \vert ^{2} \,ds \biggr\vert ^{\frac{p}{2}} \biggr]. \end{aligned}$$
2.3 G-stochastic differential equation
Consider the following SDE driven by a 1-dimensional G-Brownian motion:
where the initial condition \(X_{0}\in \mathbb{R}\) is a given constant.
We will consider this GSDE, whose coefficients satisfy both a locally Lipschitz condition and a Lyapunov-type condition.
-
(A1)
\(b, h, \sigma: [0,T]\times \mathbb{R}\to \mathbb{R}\) are given deterministic functions satisfying continuous in t and locally Lipschitz in x, i.e., for each \(x, y\in B_{0}(R):= \{a| |a|\leq R\}\), there exists a positive constant \(C_{R}\) that depends only on R such that for each \(t\in [0,T]\),
$$\begin{aligned} \bigl\vert \psi (t,x)-\psi (t,y) \bigr\vert \leq C_{R} \vert x-y \vert \end{aligned}$$and
$$\begin{aligned} \sup_{t\in [0,T]} \bigl\vert \psi (t,0) \bigr\vert \leq L, \end{aligned}$$where \(\psi =b,h,\sigma \), respectively.
-
(A2)
There exists a deterministic nonnegative Lyapunov function \(V\in C^{1,2}([0,T]\times \mathbb{R})\) such that
$$\begin{aligned} \inf_{|x|\geq R} \inf_{t\in [0,T]} V(t,x) \to \infty,\quad \text{as $R\to \infty $}, \end{aligned}$$and for some constant \(C_{L}>0\) and all \((t, x)\in [0,T]\times \mathbb{R}\),
$$\begin{aligned} \mathcal{L}V(t,x) \leq C_{L} V(t,x), \end{aligned}$$where \(\mathcal{L}\) is a differential operator defined by
$$\begin{aligned} \mathcal{L}V = \partial _{t} V + \partial _{x} V b + G \bigl(2 \partial _{x} V h + \partial ^{2}_{x^{2}}V \sigma ^{2} \bigr). \end{aligned}$$Here, \(G(\cdot )\) is a sublinear function defined in (1).
In Li, Lin, and Lin [13], they established the existence and uniqueness of the solution for the above GSDE with locally Lipschitz and Lyapunov conditions through the localization methods. We review their excellent results in what follows.
Theorem 2.11
Under assumptions (A1) and (A2), there exists a unique solution \(X\in M_{w}^{2}(0,T; \mathbb{R})\) to the G-stochastic differential equation (2) and X has t-continuous paths on \([0,T]\).
We notice that the domain of coefficients here is a little larger than the one in Peng [16], where Peng [16] states the following result. We recall the following standard linear growth and Lipschitz assumption:
-
(H1)
There exists some constant L such that
$$\begin{aligned} \bigl\vert \psi (t,x)-\psi (t,y) \bigr\vert \leq L \vert x-y \vert \end{aligned}$$for each \(t\in [0,T], x, y\in \mathbb{R}\), and
$$\begin{aligned} \sup_{t\in [0,T]} \bigl\vert \psi (t, 0) \bigr\vert \leq L, \end{aligned}$$where \(\psi =b,h,\sigma \), respectively.
-
(H2)
\(b, h, \sigma: [0,T]\times \mathbb{R}\to \mathbb{R}\) are given functions satisfying, for each \(x\in \mathbb{R}\), \(b(\cdot, x), h(\cdot, x),\sigma (\cdot, x) \in M_{G}^{2}(0,T)\) and
$$\begin{aligned} \bigl\vert b(\cdot, x) \bigr\vert ^{2} + \bigl\vert h(\cdot, x) \bigr\vert ^{2} + \bigl\vert \sigma (\cdot, x) \bigr\vert ^{2} < C\bigl(1+ \vert x \vert ^{2}\bigr). \end{aligned}$$
We also recall the following excellent results from Peng [16].
Theorem 2.12
Under assumptions (H1) and (H2), there exists a unique solution \(X\in M_{G}^{2}(0,T)\) to the G-stochastic differential equation (2). Denote by \(X_{t}\) the solution starting with \(X_{0}\in \mathbb{R}\), then there exists \(C>0\) that depends on T such that
The following corollary can be deduced from (3), see also Corollary 5.3.2 in Peng [16].
Corollary 2.13
Assume that Lipschitz condition (H1) and linear growth condition (H2) hold, then we have
where the constant C depends only on the Lipschitz constant and the initial value \(X_{0}\).
To apply this theorem, we need to assume that (H2) holds throughout this paper, and C may be a positive constant whose value may change in different occasions.
3 Averaging principle under volatility uncertainty
We now study an averaging principle for a stochastic differential equation driven by a G-Brownian motion in \(\mathbb{R}\):
where \(\epsilon \in (0,1)\) and the initial condition \(X_{0}\in \mathbb{R}\) is a given constant. When the functions \(b, h, \sigma \) satisfy the conditions as in (A1)–(A2), then, by Theorem 2.11, equation (4) has a unique solution \(X^{\epsilon}\in M_{w}^{2}(0,T; \mathbb{R})\).
Our objective is to show that the solution \(X^{\epsilon}\in M_{w}^{2}(0,T; \mathbb{R})\) could be approximated by the solution of some simplified equations. For this, we associate the above stochastic differential equation with the following averaged stochastic differential equation driven by a G-Brownian motion:
Here, the functions \(\bar{b}, \bar{h}, \bar{\sigma}\) are called time averaged functions, and the locally Lipschitz condition and the Lyapunov-type condition (A1) and (A2) are satisfied without time term, where the differential operator \(\mathcal{L}\) is defined by
By Theorem 2.11, the above averaged G-stochastic differential equation has a unique solution \(\bar{X}\in M_{w}^{2}(0,T)\) to (5).
To get the averaging principle, we need the following time averaging conditions for functions \(b, h, \sigma \) and \(\bar{b}, \bar{h}, \bar{\sigma}\):
- (B):
-
For any \(T_{1}\in [0,T]\) and all x, there exists a function φ such that
$$\begin{aligned} &\sup_{t\geq 0} \biggl\vert \frac{1}{T_{1}} \int _{t}^{t+T_{1}} \bigl(b(s,x)- \bar{b}(x) \bigr) \,ds \biggr\vert ^{2} \leq \varphi (T_{1}) \bigl(1+ \vert x \vert ^{2}\bigr), \\ &\sup_{t\geq 0} \frac{1}{T_{1}} \int _{t}^{t+T_{1}} \bigl\vert h(s,x)-\bar{h}(x) \bigr\vert ^{2} \,ds \leq \varphi (T_{1}) \bigl(1+ \vert x \vert ^{2}\bigr), \end{aligned}$$(6)and
$$\begin{aligned} \sup_{t\geq 0} \frac{1}{T_{1}} \int _{t}^{t+T_{1}} \bigl\vert \sigma (s,x)- \bar{ \sigma}(x) \bigr\vert ^{2} \,ds \leq \varphi (T_{1}) \bigl(1+ \vert x \vert ^{2}\bigr). \end{aligned}$$
Here, \(\varphi (T_{1})\) is a positive bounded function with \(\lim_{T_{1}\to \infty} \varphi (T_{1}) =0\). The averaged functions f̄ (where \(\bar{f} = \bar{b}, \bar{h}, \bar{\sigma}\)) have been given many different definitions in the literature, for instance, we can choose \(\bar{f}(x) = \frac{1}{T} \int _{0}^{T} f(s,x) \,ds\).
In fact, it follows from assumption (B) that the averaged functions f̄ (where \(\bar{f} = \bar{b}, \bar{h}, \bar{\sigma}\)) satisfy (locally) Lipschitz conditions with the same Lipschitz constant as f, provided that f satisfy (locally) Lipschitz conditions. We only state that the function b̄ satisfies the locally Lipschitz condition as b. In fact, for every \(x, y\in B_{0}(R)\) and every \(T>0\), we have
Then, taking T tending to infinity in the inequality, we get the function b̄ is locally Lipschitz. Similar discussion and the following remark can be found in Gao [6], Guo, Lv, Wei [8], Shen, Song, Wu [17], and Shen, Xiang, Wu [18].
Remark 3.1
Due to
we can claim that (6) in assumption (B) is weaker than the following traditional averaging condition:
With several preliminary assumptions at our hands, we are in a position to present our main results. We first introduce a lemma which is important for our averaging principle, and then we consider the averaging principle of the GSDE with standard linear growth and Lipschitz assumption. Finally, we extend those assumptions to a locally Lipschitz condition and a Lyapunov-type condition.
Lemma 3.2
Suppose that assumptions (H1), (H2), and (B) are satisfied. Then
Proof
Let \(\{t_{1}, t_{2}, \ldots, t_{N}\}\) be a partition of \([0,T]\):
and
Then it is easy to obtain that \(T\leq N\sqrt{\epsilon} < T+\sqrt{\epsilon}\). Let
then we have
Using Hölder’s inequality and linear growth assumption on b and b̄, we get
By Theorem 2.12, (8), and (9), we have
Here, in the last inequality, we have used the subadditivity of sublinear expectation. By assumption (B), the Lipschitz conditions of b and b̄, we obtain
Hence, by Corollary 2.13, we have
Then, substituting (12) in (10), we finally obtain the estimate
Finally, the required inequality follows by letting ϵ tend to zero in the inequality. □
Lemma 3.3
Suppose that assumptions (H1), (H2), and (B) are satisfied. Then
Proof
Let \(\{t_{1}, t_{2}, \ldots, t_{N}\}\) be a partition of \([0,T]\), \(t_{i} = i\sqrt{\epsilon}, i=0,1,2,\ldots, N-1; t_{N} =T, 0< T-t_{N-1} \leq \sqrt{\epsilon}\). Hence \(T\leq N\sqrt{\epsilon} < T+\sqrt{\epsilon}\). Let
then, by subadditivity of the sublinear expectation, we have
By assumptions (H1), (H2), and (B), we get
Hence, using Corollary 2.13 again, we have
Then, substituting (16) in (14), we obtain the following estimate:
Finally, the required inequality follows by letting ϵ tend to zero in the inequality. □
Lemma 3.4
Suppose that assumptions (H1), (H2), and (B) are satisfied. Then
The proof is the same as Lemma 3.3, here we omit the proof.
Now, we present our first main result, the averaging principle of the GSDE with standard linear growth and Lipschitz assumption.
Theorem 3.5
Assume that assumptions (H1), (H2), and (B) are satisfied. Then
Proof
Starting with
and employing the simple arithmetic inequality
we arrive at
where \(u\in [0,T]\), and \(I_{i}, i=1,2,3\), denote the above terms respectively. Now we present some useful estimates for \(I_{i}, i=1,2,3\).
Firstly, we apply the arithmetic inequality and Hölder’s inequality to obtain
Here, in the last inequality, we have used the Lipschitz condition of b̄.
Second, for \(I_{2}\), we take the expectation on \(I_{2}\):
Here, we have used the BDG-type inequality (see Lemma 2.10).
Finally, we take expectation on \(I_{3}\) using the BDG-type inequality in Lemma 2.10(2):
Here, in the last inequality, we have used the Lipschitz condition of σ̄.
Therefore, taking sublinear expectation on (19), substituting (20)–(22) in (19), we get
An application of the Gronwall inequality in (23) implies that
Consequently, the required result follows by applying Lemmas 3.2–3.4. The proof is complete. □
Next we turn our attention to the time-averaging principle of the GSDE with a locally Lipschitz condition and a Lyapunov-type condition. Our method of argument of the time-averaging principle is based on the localization method.
Theorem 3.6
Assume that assumptions (A1, A2) and (B) are satisfied. Then
Proof
We first consider the following truncated GSDE of (4) and (5), respectively, with coefficients that satisfy assumptions (A1, A2) and (B) for each \(N\in \mathbb{N}\), \(0\leq t\leq T\):
and
and then we consider the following truncated GSDEs of (18):
where \(b^{N}, h^{N}, \sigma ^{N}, \bar{b}^{N}, \bar{h}^{N}, \bar{\sigma}^{N}\) are defined in the following way:
It is easy to verify that \(b^{N}, h^{N}, \sigma ^{N}, \bar{b}^{N}, \bar{h}^{N}, \bar{\sigma}^{N}\) are all bounded functions and uniformly Lipschitz in x. Then, by the result of Lipschitz GSDE with coefficients in \(M_{G}^{p}(0,T; \mathbb{R})\) in Theorem 3.5, for the truncated GSDEs (25) and (26), we obtain the following result:
Define two sequences of stopping times by
and
which satisfy \(\{\tau _{N}\leq t\}\cup \{\bar{\tau}_{N}\leq t\} \in \mathcal{F}_{t}\). Based on stopping times, we can deduce from (25) and (26) that
and
On the other hand, by the representations of \(X^{N+1}, \bar{X}^{N+1}\) and the continuity of process \(X^{N+1}, \bar{X}^{N+1}\), we have
and
By the uniqueness of the solution to the truncated GSDEs (25) and (26), for each \(N\in \mathbb{N}\), \(X^{N} \) and \(X^{N+1} \) are distinguishable on \([0,\tau _{n}]\); at the same time, \(\bar{X}^{N}\) and \(\bar{X}^{N+1}\) are distinguishable on \([0,\bar{\tau}_{n}]\). This also implies that the sequences \(\{\tau _{N}\}_{N\in \mathbb{N}}\) and \(\{\bar{\tau}_{N}\}_{N\in \mathbb{N}}\) are q.s. increasing. Now we claim that
and
By the definition of \(\tau _{N}\) we know that \(|X^{\epsilon, N+1}_{\cdot}| \leq N\) on \([0, \tau _{n}]\), hence we have for \(0\leq y\leq T\)
and
where \(f= b, h,\sigma \) and \(\bar{f}= \bar{b}, \bar{h}, \bar{\sigma}\). It follows from Lemma 4.2 in Li and Peng [14] that both \(f^{N}(\frac{t}{\epsilon}, X^{\epsilon, N}_{t}) I_{[0, \tau _{n}]}(t)\) and \(\bar{f}^{N}( \bar{X}^{N}_{t} ) I_{[0, \bar{\tau}_{n}]}(t)\) are \(M_{G}^{p} (0,T; \mathbb{R})\)-processes for any \(p\geq 2\). Therefore, for any \(t\in [0, T]\), we have
and, similarly, we have
Applying G-Itô’s formula to
and
respectively, where we can take \(V(t, x) =1+ |x|^{2}\), we obtain
and
Letting
we have \(\eta _{s}(\Phi, X^{\epsilon, N}), \bar{\eta}_{s}(\bar{\Phi}, \bar{X}^{N}) \in M_{w}^{2}(0,T; \mathbb{R})\). Hence, substituting (33) in (31), we arrive at
By assumption (A2), \(\mathcal{L} V\leq C_{L} V\) implies \(\mathcal{L} \Phi \leq 0\). It follows from Proposition 1.4 in IV-1 of Peng [16] that, for \(\eta \in M_{w}^{1}(0,T; \mathbb{R})\),
Taking expectation on (35), we obtain
In particular, we have
Since \(\tau _{N}< T\) implies \(|X^{\epsilon, N}_{T\wedge \tau _{N}}| = N \), q.s., from which we deduce
Letting \(N\to \infty \), by (A2), we obtain
Since \(\{\omega: \tau _{N}(\omega ) =T\}\) is increasing, the upwards convergence theorem yields (29). Similarly, substituting (34) in (32), after same discussion as above, we get that (30) holds.
Therefore, there exists a polar set A such that for all \(\omega \in A^{c}\) the following assertion holds: one can find an \(N_{0}(\omega )\) that depends on ω such that for all \(N\geq N_{0}(\omega ), N\in \mathbb{N}, \tau _{N}(\omega ) =T\). Then we define, for \(t\in [0, T]\),
Similarly, one can find an \(N_{1}(\omega )\) that depends on ω such that for all \(N\geq N_{1}(\omega ), N\in \mathbb{N}, \bar{\tau}_{N}(\omega ) =T\). Then we define, for \(t\in [0, T]\),
From the argument above, we have \(X^{\epsilon }I_{[0, \tau _{N} ]} = X^{\epsilon, N} I_{[0, \tau _{N}]} \in M_{G}^{2}(0, T; \mathbb{R})\), and thus \(X^{\epsilon}\in M_{w}^{2}(0,T; \mathbb{R})\). Also, we have \(\bar{X} \in M_{w}^{2}(0,T; \mathbb{R})\). Moreover,
and
Hence, the two equations imply that \(X^{\epsilon}, \bar{X}\) satisfy (4), (5), respectively. Furthermore, we obtain that \(X^{\epsilon } - \bar{X} \) satisfies
Therefore, letting \(N\to \infty \) in (28), we can deduce
which completes the proof of this theorem. □
As a consequence of Theorem 3.6 and Markov-type inequality of capacity, the convergence also holds in the sense of capacity.
Corollary 3.7
Under the same assumptions as in Theorem 3.6, for all \(\delta >0\), we have
Proof
By Proposition 2.3 and Theorem 3.6, for any number \(\delta >0\), one can find
Noting that \(v(A)=1- \mathbb{V}(A^{c})\), we have
Let \(\epsilon \to 0\) and the required result follows from Theorem 3.6. □
4 Numerical simulation
In this section, we present numerical simulation of GSDE obtained using PYTHON and give three examples to demonstrate the averaging principle for the GSDE driven by G-Brownian motion.
Example 1
(Without quadratic variation term)
Consider the following standard GSDE:
and the averaged GSDE
with the same initial condition \(X^{\epsilon}_{0} = Z_{0}= X_{0}\), where \(B_{t}\) is a G-Brownian motion and satisfies
Obviously, \(\frac{1}{\pi} \int _{0}^{\pi }\lambda X^{\epsilon }\sin ^{2}(t) \,dt = \frac{1}{2} \lambda X^{\epsilon }\), all coefficients of the standard GSDE and averaged GSDE satisfy conditions (A1)–(A2) and (B1)–(B3) for the functions \(b, h, \sigma, \bar{b}, \bar{h}, \bar{\sigma}\). Thus Theorem 3.5 and Theorem 3.6 hold, that is,
and
Now we carry out the numerical simulation to get the solutions of GSDE (36) and averaged GSDE (37) under conditions \(X_{0} = 1, \lambda =1.0, \varepsilon =0.01\), (a) \(\underline{\sigma} = 0.1, \bar{\sigma}=0.5\); (b) \(\underline{\sigma} = 1, \bar{\sigma}=5\); (c) \(\underline{\sigma} = 0.5, \bar{\sigma}=2\), and (d) Brownian motion \(\sigma =1\), respectively. Figure 1 depicts a sample average of 5000 trajectories of the SDE \(X^{\epsilon}\), a sample average of 5000 trajectories of the averaged SDE Z, and a sample average of 5000 trajectories of the error \(X^{\epsilon}-Z\). Not only do we see a good agreement between solutions of the equation and the averaged equation, but we are also aware of the fact: Larger volatility can cause greater fluctuation!
Example 2
(With quadratic variation term)
Consider the following standard GSDE:
and the averaged GSDE
with the same initial condition \(X^{\epsilon}_{0} = Z_{0}= X_{0}\), where \(B_{t}\) is a G-Brownian motion and satisfies
We can easily calculate the following claim for any \(T_{1}\in [0,T]\):
where φ is a positive bounded function with \(\lim_{T_{1} \to \infty} \varphi (T_{1})=0\). In fact, due to \(|\sin (x)|\leq 1\), for any \(T_{1}\in [0,T], \epsilon \in (0,1)\), we have
Hence all the coefficients of GSDE and averaged GSDE satisfy conditions (A1)–(A2) and (B) for the functions \(b, h, \sigma, \bar{b}, \bar{h}, \bar{\sigma}\). Thus we can use the solution Z of GSDE (39) to approximate the original solution \(X^{\epsilon}\) of GSDE (38), and the convergence will be assured.
Now we carry out the numerical simulation to get the solutions of GSDE (38) and averaged GSDE (39) under conditions \(X_{0} = 1, \lambda =1.0, \varepsilon =0.01\), (a) \(\underline{\sigma} = 0.1, \bar{\sigma}=0.5\); (b) \(\underline{\sigma} = 1, \bar{\sigma}=5\); (c) \(\underline{\sigma} = 0.5, \bar{\sigma}=2\), and (d) \(\underline{\sigma} = \bar{\sigma}=1\), respectively. Figure 2 depicts a sample average of 5000 trajectories of the SDE \(X^{\epsilon}\), a sample average of 5000 trajectories of the averaged SDE Z, and a sample average of 5000 trajectories of the error \(X^{\epsilon}-Z\). We can see a good agreement between solutions of the equation and the averaged equation. Comparing (b)(c) with (d), we can observe the following fact: because GSDE contains quadratic variation term, the solution of the equation can better reflect the change of trend: The solution of this equation decays faster as the uncertainty of the volatility increases.
Example 3
(With diffusion coefficient term)
Consider the following standard GSDE, where the diffusion coefficient is not a constant:
and the averaged GSDE
with the same initial condition \(X^{\epsilon}_{0} = Z_{0}= X_{0}\), where \(B_{t}\) is a G-Brownian motion and satisfies
Here,
Due to \(|\sin (x)|\leq 1\), for any \(\epsilon \in (0,1)\), we have
Hence, all the coefficients of GSDE and averaged GSDE satisfy conditions (A1)–(A2) and (B) for the functions \(b, h, \sigma, \bar{b}, \bar{h}, \bar{\sigma}\). Thus we can use the solution Z of GSDE (41) to approximate the original solution \(X^{\epsilon}\) of GSDE (40), and the convergence will be assured.
Now we carry out the numerical simulation to get the solutions of GSDE (40) and averaged GSDE (41) under conditions \(X_{0} = 1, \lambda =1.0, \varepsilon =0.01\), (a) \(\underline{\sigma} = 0.1, \bar{\sigma}=0.5\); (b) \(\underline{\sigma} = 1, \bar{\sigma}=5\); (c) \(\underline{\sigma} = 0.5, \bar{\sigma}=2\), and (d) \(\underline{\sigma} = \bar{\sigma}=1\), respectively. Figure 3 depicts a sample average of 5000 trajectories of the SDE \(X^{\epsilon}\), a sample average of 5000 trajectories of the averaged SDE Z, and a sample average of 5000 trajectories of the error \(X^{\epsilon}-Z\). We can see a good agreement between solutions of the equation and the averaged equation, and the error is approximately zero. The numerical verification is consistent with the theoretical results.
5 Conclusion
In this paper we have studied the time-averaging principle for stochastic differential equations based on the Lyapunov condition in the presence of a family of probability measures, each corresponding to a different scenario for the volatility. To overcome the difficulty from the locally Lipschitz coefficients of G-SDEs, the G-stochastic calculus and the localization technique have been used. We show that the solution of a standard equation converges to the solution of the corresponding averaging equation in the sense of sublinear expectation with the help of some properties of G-stochastic calculus. We present numerical simulations of G-SDEs and give three examples to demonstrate the averaging method. The numerical results exhibit that there is a good agreement between solutions of the equation and the averaged equation, and the error is approximately zero. The numerical verification is consistent with the theoretical results.
Availability of data and materials
Not applicable.
References
Denis, L., Hu, M., Peng, S.: Function spaces and capacity related to a sublinear expectation: application to G-Brownian motion paths. Potential Anal. 34, 139–161 (2011)
Freidlin, M.I., Wentzell, A.D.: Random perturbations of dynamical systems. In: Grundlehren der Mathematischen Wissenschaften, vol. 260. Springer, Berlin (1998)
Fu, H., Liu, J.: Strong convergence in stochastic averaging principle for two time-scales stochastic partial differential equations. J. Math. Anal. Appl. 384(1), 70–86 (2011)
Fu, H., Wan, L., Liu, J.: Strong convergence in averaging principle for stochastic hyperbolic-parabolic equations with two time-scales. Stoch. Process. Appl. 125(8), 3255–3279 (2015)
Gao, F.: Pathwise properties and homeomorphic flows for stochastic differential equations driven by G-Brownian motion. Stoch. Process. Appl. 119(10), 3356–3382 (2009)
Gao, P.: Averaging principle for stochastic Korteweg-de Vries equation. J. Differ. Equ. 267(12), 6872–6909 (2019)
Givon, D.: Strong convergence rate for two-time-scale jump-diffusion stochastic differential systems. Multiscale Model. Simul. 6(2), 577–594 (2007)
Guo, Z., Lv, G., Wei, J.: Averaging principle for stochastic differential equations under a weak condition. Chaos, Interdiscip. J. Nonlinear Sci. 30(12), 123139 (2020)
Hu, M., Jiang, L., Wang, F.: An averaging principle for nonlinear parabolic PDEs via FBSDEs driven by G-Brownian motion. J. Math. Anal. Appl. 508, Article IDÂ 125893 (2022)
Hu, M., Wang, F.: Probabilistic approach to singular perturbations of viscosity solutions to nonlinear parabolic PDEs. Stoch. Process. Appl. 141, 139–171 (2021)
Hu, M., Wang, F., Zheng, G.: Quasi-continuous random variables and processes under the G-expectation framework. Stoch. Process. Appl. 126(8), 2367–2387 (2016)
Khasminskii, R.Z.: On the principle of averaging the Itô’s stochastic differential equations. Kybernetika 4, 260–279 (1968)
Li, X., Lin, X., Lin, Y.: Lyapunov-type conditions and stochastic differential equations driven by G-Brownian motion. J. Math. Anal. Appl. 439(1), 235–255 (2016)
Li, X., Peng, S.: Stopping times and related Itôs calculus with G-Brownian motion. Stoch. Process. Appl. 121(7), 1492–1508 (2011)
Mao, W., Chen, B., You, S.: On the averaging principle for SDEs driven by G-Brownian motion with non-Lipschitz coefficients. Adv. Differ. Equ. 2021(1), 1 (2021)
Peng, S.: Nonlinear Expectations and Stochastic Calculus Under Uncertainty. Probability Theory and Stochastic Modelling, vol. 65. Springer, Berlin (2019)
Shen, G., Song, J., Wu, J.L.: Stochastic averaging principle for distribution dependent stochastic differential equations. Appl. Math. Lett. 125, 107761 (2022)
Shen, G., Xiang, J., Wu, J.L.: Averaging principle for distribution dependent stochastic differential equations driven by fractional Brownian motion and standard Brownian motion. J. Differ. Equ. 321, 381–414 (2022)
Stratonovich, R.L.: Topics in the Theory of Random Noise (Vol. 2). CRC Press, Boca Raton (1967)
Stratonovich, R.L.: Conditional Markov Processes and Their Application to the Theory of Optimal Control (1968)
Xu, J., Miao, Y., Liu, J.: A note on strong convergence rate in averaging principle for stochastic FitzHugh–Nagumo system with two time-scales. Stoch. Anal. Appl. 34(1), 178–181 (2016)
Xu, Y., Duan, J., Xu, W.: An averaging principle for stochastic dynamical systems with Lévy noise. Phys. D, Nonlinear Phenom. 240(17), 1395–1401 (2011)
Acknowledgements
I am grateful to the anonymous reviewer for carefully reading the manuscript and giving many valuable suggestions. IÂ also wish to thank Prof. Wang, Falei for his helpful discussions and suggestions.
Funding
This work is supported by the National Natural Science Foundation of China (Grant No. 11501325, 71871129), the China Postdoctoral Science Foundation (Grant No. 2018T110706, 2018M642641) and Shandong Province Key R&D Plan (Major Science and Technology Innovation) Project (Grant No. 2021CXGC010108).
Author information
Authors and Affiliations
Contributions
GZ is the unique author of the manuscript. GZ prepared the manuscript initially and performed all the steps of the proofs in this research. GZ read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The author declares that he has no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zong, G. Time-averaging principle for G-SDEs based on Lyapunov condition. Adv Cont Discr Mod 2023, 28 (2023). https://doi.org/10.1186/s13662-023-03772-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13662-023-03772-6