DeepAI
Log In Sign Up

An explicit Euler method for McKean-Vlasov SDEs driven by fractional Brownian motion

09/10/2022
by   Jie He, et al.
0

In this paper, we establish the theory of chaos propagation and propose an Euler-Maruyama scheme for McKean-Vlasov stochastic differential equations driven by fractional Brownian motion with Hurst exponent H ∈ (0,1). Meanwhile, upper bounds for errors in the Euler method is obtained. A numerical example is demonstrated to verify the theoretical results.

READ FULL TEXT VIEW PDF

page 1

page 2

page 3

page 4

05/26/2022

Backward Euler method for stochastic differential equations with non-Lipschitz coefficients

We study the traditional backward Euler method for m-dimensional stochas...
06/28/2020

Symplectic Euler scheme for Hamiltonian stochastic differential equations driven by Levy noise

This paper proposes a general symplectic Euler scheme for a class of Ham...
01/16/2022

Strong convergence rate of the Euler scheme for SDEs driven by additive rough fractional noises

The strong convergence rate of the Euler scheme for SDEs driven by addit...
08/05/2021

Efficient approximation of SDEs driven by countably dimensional Wiener process and Poisson random measure

In this paper we deal with pointwise approximation of solutions of stoch...
04/15/2021

Numerical stability of Grünwald-Letnikov method for time fractional delay differential equations

This paper is concerned with the numerical stability of time fractional ...
04/26/2022

Stopped Brownian-increment tamed Euler method

In this article we propose a new explicit Euler-type approximation metho...

1 Introduction

The pioneering work of McKean–Vlasov stochastic differential equations (SDEs) has been done by McKean in [21, 19, 20] connected with a mathematical foundation of the Boltzmann equation. Due to their widespread applications in many fields, McKean–Vlasov SDEs have been researched by many scholars. In [17], existence and uniqueness are proved for distribution-dependent SDEs with non-degenerate noise under integrability conditions on distribution-dependent coefficients. Some theories about McKean–Vlasov SDEs were investigated, including ergodicity [11], Harnack inequality [27], and the Bismut formula [12, 25]. And the integration by parts formulae on Wiener space for solutions of the SDEs with general McKean–Vlasov interaction was derived in [8]. In [6]

, Buckdahn et al. characterized the function on the coefficients of the stochastic differential equation under appropriate regularity conditions as the unique classical solution of a nonlocal partial differential equation of mean-field type. A complete probabilistic analysis of a large class of stochastic differential games with mean field interactions was provided in

[7].

It is well known that the explicit solutions to McKean–Vlasov SDEs are difficult to be shown. Hence, the numerical methods for McKean–Vlasov SDEs driven by standard Brownian motion are studied by many scholars [1, 2, 4, 5, 9]. Moreover, it should be noted that SDEs driven by fractional Brownian motion (fBm) have wider applications [3, 23, 24]. On the other hand, the numerical methods for SDEs driven by fBm have attracted increasing interest; see [15, 16, 18, 28, 30] for example. Galeati et al. [14] examined the distribution-dependent stochastic differential equations with erratic, potentially distributional drift, driven by an additive fBm of Hurst parameter , and they established strong well-posedness under a variety of assumptions on the drift. To our knowledge, the numerical method for McKean–Vlasov SDEs driven by fBm has not been discussed yet. As we know, propagation of chaos plays a key role to approximate the Mckean-Vlasov SDEs. This paper aims at establishing the theory about propagation of chaos and the strong convergence rate in sense of EM method for McKean–Vlasov SDEs driven by fBm under the globally Lipschitz condition.

In this paper, we consider the following -dimensional McKean–Vlasov SDEs driven by fBm of the form

(1.1)

where the coefficients . Here, the initial value with , and is a -dimensional fBm with Hurst parameter . As we know, the covariance of is

The fBm corresponds to a standard Brownian motion when . Further, the fBm is not a semi-martingale or a Markov process unless . Therefore, when working with the fBm , many of the powerful features are unavailable. We rewrite (1.1) to the system of noninteracting particles

(1.2)

where denotes the law of the process at time . Compared with the standard SDEs, McKean–Vlasov SDEs provide an additional complexity, that is, it is required to approximate the law at each time step. Although there are other technologies, the most common one is the so-called interacting particle system

(1.3)

where the empirical measures is defined by

and denotes the Dirac measure at point .

The structure of this work is as follows. The mathematical preliminaries on the McKean–Vlasov SDEs driven by fBm are presented in Section 2. Section 3 gives the main theorem and its proof. Numerical simulations are provided in Section 4.

2 Mathematical Preliminaries

Throughout the article, we will always work on a finite time interval

and consider an underlying probability space

. More precisely, is the Banach space of continuous functions vanishing at 0 equipped with the supremum norm, is the Borel -algebra and is the unique probability measure on such that the canonical process is a -dimensional fBm with Hurst parameter . For any , let

We use and for the Euclidean norm and inner product, respectively, and let and . The notation for and

means the tensor product of

and . We will denote the set of all probability measures on by

For and any , the -Wasserstein distance is defined by,

where is the set of couplings of and , and is a Polish space under the -Wasserstein metric.

Let with . For and , the left-sided fractional Riemann-Liouville integral of of order on is defined as

and the right-sided fractional Riemann-Liouville integral of of order on is defined as

where and denotes the Gamma function. For further details about fractional integral and derivative, we refer the reader to [3, 26].

Assumption 2.1

There exists a positive constant such that

(2.1)

for all and .

(2.2)

for all . And for initial experience distribution ,

(2.3)

Furthermore, from Assumption 2.1, there exists a positive constant such that

(2.4)

and

(2.5)

3 Main Result

Lemma 3.1

[12] When and hold, the solution of in (1.1) exists and is unique. Moreover, when and is independent of distribution, the solution of in (1.1) exists and is unique.

Lemma 3.2

For two empirical measures and . Then,

This lemma follows from constructing a simple transport plan .

3.1 Case

Theorem 3.3

Let Assumption 2.1 holds and , then

where is a positive constant dependent on .

Proof. From (1.3) and elementary inequality, we get

(3.1)

For the second part at the right side of (3.1), using the Hölder inequality, we obtain

Applying (2.4), we see that

(3.2)

For the last part of the right end of (3.1), apply Theorem 1.1 in [22], we get

(3.3)

Using the Hölder inequality,

By (2.5), we have

(3.4)

Through a similar proof in [10, Lemma 2.3], for one can observe that

(3.5)

Combine (3.2) and (3.4) into (3.1) and (3.5), we get

For the last term, we apply the Minkowski inequality and since all are identically distributed, we have

Thus, we get

Then applying the Grönwall inequality, we obtain

Similarly, we can show

The proof is complete. 

Theorem 3.4 (Propagation of Chaos)

Let Assumption 2.1 be satisfied. If for some , then it holds that

where the constant depends on , , , and but does not depend on N.

Proof. It follows from (1.2) and (1.3) that

Using the elementary inequality, we can show that

(3.6)

For the first part on the right of (3.6), use the Hölder inequality and Assumption 2.1, we get

(3.7)

For the second part on the right of (3.6), through the same technique as (3.3), we get

Apply Assumption 2.1, we obtain

(3.8)

Combining this with (3.7), then

For the part of Wasserstein distance, we note and we obtain

By Lemma 3.2, we see

Thought the fact that for , we have

where we use the Minkowski inequality in the last inequality. Then through simple sorting, we have

What’s particularly interesting is that

is controlled by the Wasserstein distance estimate in

[13, Theorem 1]. Therefore,

where

By Theorem 3.3, we note that . Thus,

Then, applying the Grönwall inequality completes the proof. 

3.1.1 EM Method for Interacting Particle System

Now, we define a uniform mesh with , where for . The numerical solutions are then generated by the EM method

(3.9)

where the empirical measures We show two versions of extension of the numerical solution at the discrete time points to . The first is the piecewise constant extension given by

(3.10)

and the second is the continuous extension of the EM method defined by

(3.11)

Here, From (3.11), for all , we have

(3.12)
Theorem 3.5

Let Assumption 2.1 holds. For some , then

where is a positive constant dependent on but independent of .

Proof. Similar to the proof of Theorem 3.3, we can show

Therefore, for , we have

By the Grönwall inequality, we see

Therefore, the assertion holds. 

Lemma 3.6

Assume (2.4) and (2.5) hold, for a constant , then

(3.13)

where is a positive constant dependent on but independent of .

Proof. From (3.11), we separate the left hand side of (3.13) into two parts

(3.14)

Let us first consider the first part on the right of (3.14), by using the Hölder inequality and (2.4), we obtain

(3.15)

For the second part on the right of (3.14), by choosing , we note the fact that . Then

Thanks to Stochastic Fubini Theorem for the Wiener Integrals with regard to fBm [23, Theorem 1.13.1], we obtain

Therefore, by the Hölder inequality, we get

Applying the Theorem 1.1 in [22], we see

Thus, we obtain

By [12, Lemma 3.2] with , and , we have

where we use the Hölder inequality in the last inequality. Then by (2.5), we know that

(3.16)

Combining (3.15) and (3.16) into (3.14) yields

(3.17)

By Lemma 3.2 and the Minkowski inequality , we note that

(3.18)

Substituting (3.18) into (3.17), we get

Applying the Theorem 3.5 to above inequality gives that

Thus, we obtain the desired result. 

Theorem 3.7

Let Assumption 2.1 holds, for , then