Abstract
We consider one-dimensional stochastic differential equations (SDEs) with irregular coefficients. The goal of this paper is to estimate the L p(Ω)-difference between two SDEs using a norm associated to the difference of coefficients. In our setting, the (possibly) discontinuous drift coefficient satisfies a one-sided Lipschitz condition and the diffusion coefficient is bounded, uniformly elliptic and Hölder continuous. As an application of this result, we consider the stability problem for this class of SDEs.
2010 Mathematics Subject Classification: 58K25, 41A25, 65C30
Access provided by Autonomous University of Puebla. Download chapter PDF
Similar content being viewed by others
Keywords
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
1 Introduction
Let X = (X t )0 ≤ t ≤ T be a solution of the one-dimensional stochastic differential equation (SDE)
where W: = (W t )0 ≤ t ≤ T is a standard one-dimensional Brownian motion on a probability space \((\varOmega,\mathcal{F}, \mathbb{P})\) with a filtration \((\mathcal{F}_{t})_{0\leq t\leq T}\) satisfying the usual conditions. The drift coefficient b and the diffusion coefficient σ are Borel-measurable functions from \(\mathbb{R}\) into \(\mathbb{R}\). The diffusion process X is used in many fields of application, for example, mathematical finance, optimal control and filtering.
Let X (n) be a solution of the SDE ( 1) with drift coefficient b n and diffusion coefficient σ n . We consider the stability problem for (X, X (n)) when the pair of coefficients (b n , σ n ) converges to (b, σ). Stroock and Varadhan introduced the stability problem in the weak sense in order to consider the martingale problem with continuous and locally bounded coefficients (see Chap. 11 of [17]). In [11], Kawabata and Yamada consider the strong convergence of the stability problem under the condition that the drift coefficients b and b n are Lipschitz continuous functions, the diffusion coefficients σ and σ n are Hölder continuous and (b n , σ n ) locally uniformly converges to (b, σ) (see [11, Example 1]). Kaneko and Nakao [10] prove that if the coefficients b n and σ n are uniformly bounded, σ n is uniformly elliptic and (b n , σ n ) tends to (b, σ) in L 1-sense, then \((X^{(n)})_{n\in \mathbb{N}}\) converges to X in L 2-sense. Moreover they also prove that the solution of the SDE ( 1) can be constructed as the limit of the Euler-Maruyama approximation under the condition that the coefficients b and σ are continuous and of linear growth (see [10, Theorem D]). Recently, under the Nakao-Le Gall condition, Hashimoto and Tsuchiya [8] prove that \((X^{(n)})_{n\in \mathbb{N}}\) converges to X in L p sense for any p ≥ 1 and give the rate of convergence under the condition that b n → b and σ n → σ in L 1 and L 2 sense, respectively. Their proof is based on the Yamada-Watanabe approximation technique which was introduced in [19] and some estimates for the local time.
On a related study, the convergence for the Euler-Maruyama approximation with non-Lipschitz coefficients has been studied recently. Yan [18] has proven that if the sets of discontinuous points of b and σ are countable, then the Euler-Maruyama approximation converges weakly to the unique weak solution of the corresponding SDE. Kohatsu-Higa et al. [12] have studied the weak approximation error for the one-dimensional SDE with the drift 1 (−∞, 0](x) −1 (0, +∞)(x) and constant diffusion. Gyöngy and Rásonyi [7] give the order of the strong rate of convergence for a class of one-dimensional SDEs whose drift is the sum of a Lipschitz continuous function and a monotone decreasing Hölder continuous function and its diffusion coefficient is a Hölder continuous function. The Yamada-Watanabe approximation technique is a key idea to obtain their results. In [15], Ngo and Taguchi extend the results in [7] for SDEs with discontinuous drift. They prove that if the drift coefficient b is bounded and one-sided Lipschitz function, and the diffusion coefficient is bounded, uniformly elliptic and η-Hölder continuous, then there exists a positive constant C such that
where \(\overline{X}^{(n)}\) is the Euler-Maruyama approximation for SDE ( 1). This fact implies that the strong rate of convergence for the stability problem may also depend on the Hölder exponent of the diffusion coefficient.
The goal of this paper is to estimate the difference between two SDEs using the norm of the difference of coefficients. More precisely, let us consider another SDE given by
We will prove the following inequality:
where η is the Hölder exponent of the diffusion coefficients, C is a positive constant and | | ⋅ | | p is a L p-norm which will be defined by ( 4). We will also estimate \(\mathbb{E}[\sup _{0\leq t\leq T}\vert X_{t} -\hat{ X}_{t}\vert ^{p}]\) for any p ≥ 1. It is worth noting that in the papers [10] and [11], the authors only prove the strong convergence for the stability problem. On the other hand, applying our main results, we are able to establish the strong rate of convergence for the stability problem (see Sect. 4). In order to obtain ( 3), we use the Yamada-Watanabe approximation technique and a Gaussian upper bound for the density of SDE ( 2) (see [2, 16] and [14]).
Finally, we note that SDEs with discontinuous drift coefficient have many applications in mathematical finance [1] and [9], optimal control problems [4] and other domains (see also [5] and [13]).
This paper is organized as follows: Sect. 2 introduces our framework and main results. All the proofs are shown in Sect. 3. In Sect. 4, we apply the main results to the stability problem.
2 Main Results
2.1 Notations and Assumptions
We will assume that the drift coefficient b belongs to the class of one-sided Lipschitz functions which is defined as follows.
Definition 1
A function \(f: \mathbb{R} \rightarrow \mathbb{R}\) is called a one-sided Lipschitz function if there exists a positive constant L such that for any \(x,y \in \mathbb{R}\),
Let \(\mathcal{L}\) be the class of all one-sided Lipschitz functions.
Remark 1
By the definition of the class \(\mathcal{L}\), if \(f,g \in \mathcal{L}\) and α ≥ 0, then f + g, \(\alpha f \in \mathcal{L}\). The one-sided Lipschitz property is closely related to the monotonicity condition. Actually, any monotone decreasing function is one-sided Lipschitz. Moreover, any Lipschitz continuous function is also a one-sided Lipschitz.
Now we give assumptions for the coefficients \(b,\hat{b},\sigma\) and \(\hat{\sigma }\).
Assumption 1
We assume that the coefficients \(b,\hat{b},\sigma\) and \(\hat{\sigma }\) satisfy the following conditions:
-
A-(i): \(b \in \mathcal{L}\).
-
A-(ii): b and \(\hat{b}\) are measurable and there exists K > 0 such that
$$\displaystyle{ \sup _{x\in \mathbb{R}}\left (\vert b(x)\vert \vee \vert \hat{b}(x)\vert \right ) \leq K. }$$ -
A-(iii): σ and \(\hat{\sigma }\) are η: = (1∕2 +α)-Hölder continuous with some α ∈ [0, 1∕2], i.e., there exists K > 0 such that
$$\displaystyle{ \sup _{x,y\in \mathbb{R},x\neq y}\left (\frac{\vert \sigma (x) -\sigma (y)\vert } {\vert x - y\vert ^{\eta }} \vee \frac{\vert \hat{\sigma }(x) -\hat{\sigma } (y)\vert } {\vert x - y\vert ^{\eta }} \right ) \leq K. }$$ -
A-(iv): a = σ 2 and \(\hat{a} =\hat{\sigma } ^{2}\) are bounded and uniformly elliptic, i.e., there exists λ ≥ 1 such that for any \(x \in \mathbb{R}\),
$$\displaystyle{ \lambda ^{-1} \leq a(x) \leq \lambda \text{ and }\lambda ^{-1} \leq \hat{ a}(x) \leq \lambda. }$$
Remark 2
Assume that A-(ii), A-(iii) and A-(iv) hold. Then the SDE ( 1) and the SDE ( 2) have unique strong solution (see [20]). Note that the one-sided Lipschitz property is used only in ( 11) for b, so we don’t need to assume \(\hat{b} \in \mathcal{L}\).
2.2 Gaussian Upper Bound for the Density of SDE
A Gaussian upper bounded for the density of X t is well-known under suitable conditions for the coefficients. If coefficients b and σ are Hölder continuous and σ is bounded and uniformly elliptic, then a Gaussian type estimate holds for the fundamental solution of parabolic type partial differential equations (see [6, Theorem 11, Chap. 1]). Under A-(ii), (iii) and (iv), the density function p t (x 0, ⋅ ) of X t exists for any t ∈ (0, T] and there exist positive constants \(\overline{C}\) and c ∗ such that for any \(y \in \mathbb{R}\) and t ∈ (0, T],
where \(p_{c}(t,x,y):= \frac{e^{-\frac{(y-x)^{2}} {2ct} }} {\sqrt{2\pi ct}}\) (see [14, Remark 4.1]).
Using a Gaussian upper bound for the density of X t , we can prove the following estimate.
Lemma 1
Let p ≥ 1. Assume that A-(ii), A-(iii) and A-(iv) hold. Then we have
and
where \(C_{T}:= \overline{C}\sqrt{\frac{2T} {\pi c_{{\ast}}}}\) and for any bounded measurable function f, ||⋅|| p is defined by
Proof
We only prove the first estimate. The second one can be obtained by using a similar argument. From a Gaussian upper bound for the density of \(\hat{X}_{t}\), for any \(x \in \mathbb{R}\) and s ∈ (0, T], we have
where \(\hat{p}_{s}(x_{0},\cdot )\) is a density function of \(\hat{X}_{s}\). Hence we obtain
This concludes the proof.
Remark 3
Our proof of Lemma 1 is based on the fact that we are in the one-dimensional setting. In multi-dimensional case, the integrand of ( 5) is not integrable with respect to s in general. This is the main reason for restricting our discussion to the one-dimensional SDE case.
2.3 Rate of Convergence
For any p ≥ 1, we define
Then we have the following estimate for the difference between two SDEs.
Theorem 1
Suppose that Assumption 1 holds. We assume that ɛ 1 < 1 if α ∈ (0,1∕2] and 1∕log (1∕ɛ 1 ) < 1 if α = 0. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,\alpha,\lambda\) and x 0 such that
where \(\mathcal{T}\) is the set of all stopping times τ ≤ T.
Theorem 2
Suppose that Assumption 1 holds. We assume that ɛ 1 < 1 if α ∈ (0,1∕2] and 1∕log (1∕ɛ 1 ) < 1 if α = 0. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,\alpha,\lambda\) and x 0 such that
Theorem 3
Suppose that Assumption 1 holds and p ≥ 2. We assume that ɛ p < 1 if α ∈ (0,1∕2] and 1∕log (1∕ɛ p ) < 1 if α = 0. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,p,\alpha,\lambda\) and x 0 such that
Using Jensen’s inequality, we can extend Theorem 3 as follows.
Corollary 1
Suppose that Assumption 1 holds and p ∈ (1,2). We assume that ɛ 2p < 1 if α ∈ (0,1∕2] and 1∕log (1∕ɛ 2p ) < 1 if α = 0. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,p,\alpha,\lambda\) and x 0 such that
Next, we will find a bound for \(\mathbb{E}[\vert g(X_{T}) - g(\hat{X}_{T})\vert ^{r}]\) where g is a function of bounded variation and r ≥ 1.
Definition 2
For a function \(f: \mathbb{R} \rightarrow \mathbb{R}\), we define
Here the supremum is taken over all positive integers N and all partitions −∞ < x 0 < x 1 < ⋯ < x N = x < ∞. We call f a function of bounded variation, if
Denote by BV the class of all functions of bounded variation.
Corollary 2
Suppose that Assumption 1 holds. Furthermore assume that ɛ 1 < 1 if α ∈ (0,1∕2] and 1∕log (1∕ɛ 1 ) < 1 if α = 0. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,\alpha,\lambda\) and x 0 such that for any g ∈ BV and r ≥ 1,
Remark 4
In the proof of all results, we calculate the constant C explicitly. In Theorems 1– 3 and Corollary 1, the constant C does not blow up when T → 0. On the other hand, in Corollary 2, the constant C may tend to infinity as T → 0 because we use a Gaussian upper bound for the density of X T in ( 17).
3 Proofs
3.1 Yamada-Watanabe Approximation Technique
In this section, we introduce the approximation method of Yamada and Watanabe (see [19] and [7]) which is the key technique for our proof. We define an approximation for the function ϕ(x) = | x | . For each δ ∈ (1, ∞) and κ ∈ (0, 1), there exists a continuous function \(\psi _{\delta,\kappa }: \mathbb{R} \rightarrow \mathbb{R}^{+}\) with supp ψ δ, κ ⊂ [κ∕δ, κ] such that
For example, we can take
where \(\mu _{\delta,\kappa }^{-1}:=\int _{ \kappa /\delta }^{\kappa }\exp (- \frac{1} {(\kappa -z)(z-\kappa /\delta )})dz\). We define a function \(\phi _{\delta,\kappa } \in C^{2}(\mathbb{R}; \mathbb{R})\) by
It is easy to verify that ϕ δ, κ has the following useful properties:
The property ( 8) implies that the function ϕ δ, κ approximates ϕ.
3.2 Proof of Theorem 1
To simplify the discussion, we set
Proof (Proof of Theorem 1)
Let δ ∈ (1, ∞) and κ ∈ (0, 1). From Itô’s formula, ( 7) and ( 8), we have
where
Note that since σ, \(\hat{\sigma }\) and ϕ′ δ, κ are bounded, (M t δ, κ)0 ≤ t ≤ T is a martingale so \(\mathbb{E}[M_{t}^{\delta,\kappa }] = 0\). Since \(b \in \mathcal{L}\), for any \(x,y \in \mathbb{R}\) with x ≠ y, we have, from ( 6) and ( 7),
Therefore we get
Using Lemma 1 with p = 1, we have
From ( 9) and (x + y)2 ≤ 2x 2 + 2y 2 for any x, y ≥ 0, we have
Again using Lemma 1 with p = 1, we have
Since σ is (1∕2 +α)-Hölder continuous, we have
Let τ be a stopping time with τ ≤ T and Z t : = | Y t∧τ | . From ( 10), ( 12), ( 13), ( 15) and ( 16), we obtain
If α ∈ (0, 1∕2], then since ɛ 1 < 1, by choosing δ = 2 and κ = ɛ 1 1∕(2α+1), we have
where
By Gronwall’s inequality, we get
Therefore by the dominated convergence theorem, we conclude the statement by taking t → T.
If α = 0, then since 1∕log(1∕ɛ 1) < 1, by choosing δ = ɛ 1 −1∕2 and κ = 1∕log(1∕ɛ 1), we have
where
By Gronwall’s inequality, we obtain
Therefore by the dominated convergence theorem, we conclude the statement by taking t → T.
3.3 Proof of Corollary 2
To prove Corollary 2, we recall the upper bound for \(\mathbb{E}[\vert g(X) - g(\hat{X})\vert ^{r}]\) where g is a function of bounded variation, r ≥ 1, X and \(\hat{X}\) are random variables.
Lemma 2 ([3], Theorem 4.3)
Let X and \(\hat{X}\) be random variables. Assume that X has a bounded density p X . If g ∈ BV and r ≥ 1, then for every q ≥ 1, we have
Using the above Lemma, we can prove Corollary 2.
Proof (Proof of Corollary 2)
From the Gaussian upper bound for the density p T (x 0, ⋅ ) of X T , we have for any \(y \in \mathbb{R}\),
This means that the density p T (x 0, ⋅ ) of X T is bounded. Hence from Lemma 2 with q = 1 and Theorem 1 with τ = T, for any g ∈ BV and r ≥ 1, we have
where
This concludes the proof of statement.
3.4 Proof of Theorem 2
Let V t : = sup0 ≤ s ≤ t | Y s | . Recall that for each δ ∈ (1, ∞) and κ ∈ (0, 1),
Hence the quadratic variation of M t δ, κ is given by
Before proving Theorem 2, we estimate the expectation of sup0 ≤ s ≤ t | M s δ, κ | for any t ∈ [0, T], δ ∈ (1, ∞) and κ ∈ (0, 1).
Lemma 3
Suppose that the assumption of Theorem 2 hold. Then for any t ∈ [0,T], δ ∈ (1,∞) and κ ∈ (0,1), we have
where
and \(\hat{C}_{p}\) is the constant of Burkholder-Davis-Gundy’s inequality with p > 0.
Proof
From Burkholder-Davis-Gundy’s inequality, we have
From Jensen’s inequality and Lemma 1, we have
Since σ is (1∕2 +α)-Hölder continuous, we obtain
If α ∈ (0, 1∕2], then we get
Using Young’s inequality \(xy \leq \frac{x^{2}} {2\sqrt{2}\hat{C}_{1}K} + \frac{\sqrt{2}\hat{C}_{1}Ky^{2}} {2}\) for any x, y ≥ 0 and Jensen’s inequality, we obtain
From Theorem 1 with τ = s, we have
Since 4α 2∕(2α + 1) ≤ α ≤ 1∕2, from ( 18) and ( 19), we get
which concludes the statement for α ∈ (0, 1∕2].
If α = 0, then from Jensen’s inequality and Theorem 1 with τ = s, we get
Therefore we have
This concludes the statement for α = 0.
Using the above estimate, we can prove Theorem 2.
Proof (Proof of Theorem 2)
From ( 10), ( 12), ( 14) and ( 16), we have
If α ∈ (0, 1∕2], then from ( 20), Lemmas 1 and 3, we have
Hence we get
Note that 0 < 4α 2∕(2α + 1) ≤ α ≤ 1∕2. Taking δ = 2 and κ = ɛ 1 1∕2, we have
where
By Gronwall’s inequality, we obtain
If α = 0, then from ( 20), Lemmas 1 and 3, we have
Taking δ = ɛ 1 −1∕2 and κ = 1∕log(1∕ɛ 1), we get
where
By Gronwall’s inequality, we obtain
Hence we conclude the proof of Theorem 2.
3.5 Proof of Theorem 3
In this section, we also estimate the expectation of sup0 ≤ s ≤ t | M s δ, κ | p for any p ≥ 2, t ∈ [0, T], δ ∈ (1, ∞) and κ ∈ (0, 1).
Lemma 4
Let p ≥ 2. Assume that A-(ii), A-(iii) and A-(iv) hold. Then for any t ∈ [0,T], δ ∈ (1,∞) and κ ∈ (0,1), we have
where C 5 (p,T):= 2 p∕2 C p K p and \(C_{6}(p,T):= 2^{p/2}T^{\frac{p-1} {2} }C_{p}C_{T}^{1/2}\) . In particular, if α = 1∕2, we have
Proof (Proof of Lemma 4)
From Burkholder-Davis-Gundy’s inequality, we have
From Jensen’s inequality and Lemma 1, we have
Since σ is (1∕2 +α)-Hölder continuous, we get
This concludes the first statement.
In particular, if α = 1∕2, then we get from definition of V t ,
Using Young’s inequality \(xy \leq \frac{x^{2}} {2\cdot 5^{p-1}C_{5}(p,T)} + \frac{5^{p-1}C_{ 5}(p,T)y^{2}} {2}\) for any x, y ≥ 0 and Jensen’s inequality, we obtain
which concludes the second statement.
To prove Theorem 3, we recall the following Gronwall type inequality.
Lemma 5 ([7] Lemma 3.2.-(ii))
Let (A t ) 0≤t≤T be a nonnegative continuous stochastic process and set B t := sup0≤s≤t A s . Assume that for some r > 0, q ≥ 1, ρ ∈ [1,q] and C 1 ,ξ ≥ 0,
for all t ∈ [0,T]. If r ≥ q or q + 1 −ρ < r < q hold, then there exists constant C 2 depending on r,q,ρ,T and C 1 such that
Now using Lemmas 4 and 5, we can prove Theorem 3.
Proof (Proof of Theorem 3)
From ( 20) and the inequality \(\left (\sum _{i=1}^{m}a_{i}\right )^{p} \leq m^{p-1}\sum _{i=1}^{m}a_{i}^{p}\) for any p ≥ 2 a i > 0 and \(m \in \mathbb{N}\), and Jensen’s inequality, we have
From Lemma 1 with p ≥ 2, we have
If α = 1∕2, using Lemma 4, we have
Hence we get
Taking δ = 2 and κ = ɛ p 1∕(2p), we have
where
By Gronwall’s inequality, we obtain
If α ∈ [0, 1∕2), using Lemma 4, we have
Now we apply Theorem 1 with τ = s and Lemma 5 with r = p, q = 2, ρ = 1 + 2α and
Then there exists C 7(α, p, T) which depends on p, α, T, L and C 5(p, T) such that
Taking δ = 2 and κ = ɛ p 1∕(2p) if α ∈ (0, 1∕2) and δ = ɛ p −1∕(2p) and κ = 1∕log(1∕ɛ p ) if α = 0, we get
where
Hence we conclude the proof of Theorem 3.
4 Application to the Stability Problem
In this section, we apply our main results to the stability problem. For any \(n \in \mathbb{N}\), we consider the one-dimensional stochastic differential equation
Assumption 2
We assume that the coefficients b, σ and the sequence of coefficients \((b_{n})_{n\in \mathbb{N}}\) and \((\sigma _{n})_{n\in \mathbb{N}}\) satisfy the following conditions:
-
A′-(i): \(b \in \mathcal{L}\).
-
A′-(ii): b and b n are bounded measurable i.e., there exists K > 0 such that
$$\displaystyle{ \sup _{n\in \mathbb{N},x\in \mathbb{R}}\left (\vert b_{n}(x)\vert \vee \vert b(x)\vert \right ) \leq K. }$$ -
A′-(iii): σ and σ n are η = 1∕2 +α-Hölder continuous with α ∈ [0, 1∕2], i.e., there exists K > 0 such that
$$\displaystyle{ \sup _{n\in \mathbb{N},x,y\in \mathbb{R},x\neq y}\left (\frac{\vert \sigma (x) -\sigma (y)\vert } {\vert x - y\vert ^{\eta }} \vee \frac{\vert \sigma _{n}(x) -\sigma _{n}(y)\vert } {\vert x - y\vert ^{\eta }} \right ) \leq K. }$$ -
A′-(iv): a = σ and a n : = σ n 2 are bounded and uniformly elliptic, i.e., there exists λ ≥ 1 such that for any \(x \in \mathbb{R}\) and \(n \in \mathbb{N}\),
$$\displaystyle\begin{array}{rcl} \lambda ^{-1} \leq a(x) \leq \lambda \mathit{and}\lambda ^{-1} \leq a_{ n}(x) \leq \lambda.& & {}\\ \end{array}$$ -
A′-(p): For given p > 0,
$$\displaystyle{ \varepsilon _{p,n}:= \vert \vert b - b_{n}\vert \vert _{p}^{p} \vee \vert \vert \sigma -\sigma _{ n}\vert \vert _{2p}^{2p} \rightarrow 0 }$$as n → ∞.
For p ≥ 1 and α ∈ [0, 1∕2], we define N α, p by
Then using Theorem 1– 3 and Corollary 1, 2, we have the following corollaries.
Corollary 3
Suppose that Assumption 2 holds with p = 1. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,\alpha,\lambda\) and x 0 such that for any n ≥ N α,1 ,
and
and for any g ∈ BV and r ≥ 1, we have
Corollary 4
Suppose that Assumption 2 holds with p ≥ 2. Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,p,\alpha,\lambda\) and x 0 such that for any n ≥ N α,p ,
Corollary 5
Suppose that Assumption 2 holds with 2p for p ∈ (1,2). Then there exists a positive constant C which depends on \(\overline{C},c_{{\ast}},K,L,T,p,\alpha,\lambda\) and x 0 such that for any n ≥ N α,2p ,
The next proposition shows that there exist the sequences \((b_{n})_{n\in \mathbb{N}}\) and \((\sigma _{n})_{n\in \mathbb{N}}\) satisfying Assumption 2.
Proposition 1
-
(i)
Assume \(\sup _{x\in \mathbb{R}}\vert b(x)\vert \leq K\) . If the set of discontinuity points of b is a null set with respect to the Lebesgue measure, then there exists a differentiable and bounded sequence \((b_{n})_{n\in \mathbb{N}}\) such that for any p ≥ 1,
$$\displaystyle{ \int _{\mathbb{R}}\vert b(x) - b_{n}(x)\vert ^{p}e^{-\frac{\vert x-x_{0}\vert ^{2}} {2c_{{\ast}}T} }dx \rightarrow 0 }$$(21)as n →∞. Moreover, if b is a one-sided Lipschitz function, we can construct an explicit sequence \((b_{n})_{n\in \mathbb{N}}\) which satisfies a one-sided Lipschitz condition.
-
(ii)
If the diffusion coefficient σ satisfies A′-(ii) and A′-(iii), then there exists a differentiable sequence \((\sigma _{n})_{n\in \mathbb{N}}\) such that for any \(n \in \mathbb{N}\) , σ n satisfies A′-(iii), A′-(iv) and for any p ≥ 1,
$$\displaystyle{ \int _{\mathbb{R}}\vert \sigma (x) -\sigma _{n}(x)\vert ^{2p}e^{-\frac{\vert x-x_{0}\vert ^{2}} {2c_{{\ast}}T} }dx \leq \frac{K^{2p}\sqrt{2\pi c_{ {\ast}}T}} {n^{2p\eta }}. }$$
Proof
Let \(\rho (x):=\mu e^{-1/(1-\vert x\vert ^{2}) }\mathbf{1}(\vert x\vert <1)\) with \(\mu ^{-1} =\int _{\vert x\vert <1}e^{-1/(1-\vert x\vert ^{2}) }dx\) and a sequence \((\rho _{n})_{n\in \mathbb{N}}\) be defined by ρ n (x): = n ρ(nx). We set \(b_{n}(x):=\int _{\mathbb{R}}b(y)\rho _{n}(x - y)dy\) and \(\sigma _{n}(x):=\int _{\mathbb{R}}\sigma (y)\rho _{n}(x - y)dy\). Then for any \(n \in \mathbb{N}\) and \(x \in \mathbb{R}\), we have | b n (x) | ≤ K and λ −1 ≤ a n (x): = σ n 2(x) ≤ λ, b n and σ n are differentiable.
Proof of (i). From Jensen’s inequality, we have
Since b is bounded, we have
On the other hand, since the set of discontinuity points of b is a null set with respect to the Lebesgue measure, b is continuous almost everywhere. From ( 22), using the dominated convergence theorem, we have
as n → ∞. From this fact and the dominated convergence theorem, \((b_{n})_{n\in \mathbb{N}}\) satisfies ( 21).
Let b be a one-sided Lipschitz function. Then, we have
which implies that \((b_{n})_{n\in \mathbb{N}}\) satisfies the one-sided Lipschitz condition.
Proof of (ii). In the same way as in the proof of (i), we have from Hölder continuity of σ
Finally, we show that σ n is η-Hölder continuous. For any \(x,y \in \mathbb{R}\),
which implies that σ n is η-Hölder continuous. This concludes that \((\sigma _{n})_{n\in \mathbb{N}}\) satisfies (ii).
References
J. Akahori, Y. Imamura, On a symmetrization of diffusion processes. Quant. Finan. 14 (7), 1211–1216 (2014). doi:10.1080/14697688.2013.825923
D.G. Aronson, Bounds for the fundamental solution of a parabolic equation. Bull. Am. Math. Soc. 73, 890–896 (1967)
R. Avikainen, On irregular functionals of SDEs and the Euler scheme. Finance Stochast. 13, 381–401 (2009)
V.E. Beneš, L.A. Shepp, H.S. Witsenhausen, Some Solvable stochastic control problems. Stochastics 4, 39–83 (1980)
K.S. Chan, O. Stramer, Weak consistency of the Euler method for numerically solving stochastic differential equations with discontinuous coefficient. Stochast. Process. Appl. 76, 33–44 (1998)
A. Friedman, Partial Differential Equations of Parabolic Type (Dover, New York, 1964)
I. Gyöngy, M. Rásonyi, A note on Euler approximations for SDEs with Hölder continuous diffusion coefficients. Stochast. Process. Appl. 121, 2189–2200 (2011)
H. Hashimoto, T. Tsuchiya, Convergence rate of stability problems of SDEs with (Dis-) continuous coefficients. Preprint arXiv:1401.4542v1 (2014)
Y. Imamura, Y. Ishigaki, T. Kawagoe, T. Okumura, A numerical scheme based on semi-static hedging strategy. Monte Carlo Methods Appl. 20 (4), 223–235 (2014) doi:10.1515/mcma-2014-0002
H. Kaneko, S. Nakao, A note on approximation for stochastic differential equations. Séminaire de Probabiliteés de Strasbourg 22, 155–162 (1988)
S. Kawabata, T. Yamada, On some limit theorems for solutions of stochastic differential equations, in Seminaire de Probabilities XVI, University of Strasbourg 1980/81. Lecture Notes in Mathematics, vol. 920 (Springer, New York, 1982), pp. 412–441
A. Kohatsu-Higa, A. Lejay, K. Yasuda, Weak approximation errors for stochastic differential equations with non-regular drift. Preprint (2013)
N.V. Krylov, M. Röckner, Strong solutions of stochastic equations with singular time dependent drift. Probab. Theory Relat. Fields 131, 154–196 (2005)
V. Lemaire, S. Menozzi, On some non asymptotic bounds for the Euler scheme. Electron. J. Probab. 15, 1645–1681 (2010)
H.-L. Ngo, D. Taguchi, Strong rate of convergence for the Euler-Maruyama approximation of stochastic differential equations with irregular coefficients. Math. Comput. 85 (300), 1793–1819 (2016)
D.W. Stroock, Diffusion semigroups corresponding to uniformly elliptic divergence form operators, in Séminaire de Probabilités, XXII (Springer, Berlin, 1988), pp. 316–347
D.W. Stroock, R.S. Varadhan, in Multidimensional Diffusion Processes. Die Grundlehren der Mathematischen Wissenschaften (Springer, Berlin/Heidelberg/New York, 1979)
B.L. Yan, The Euler scheme with irregular coefficients. Ann. Probab. 30 (3), 1172–1194 (2002)
T. Yamada, S. Watanabe, On the uniqueness of solutions of stochastic differential equations. J. Math. Kyoto Univ. 11, 155–167 (1971)
A.K. Zvonkin, A transformation of the phase space of a diffusion process that removes the drift. Math. USSR Sb. 22, 129–148 (1974)
Acknowledgements
The author is very grateful to Professor Arturo Kohatsu-Higa for his supports and fruitful discussions. The author would also like to thank Hideyuki Tanaka and Takahiro Tsuchiya for their useful advices. The author would like to express my thanks to Professor Toshio Yamada for his encouragement and comments. The author also thanks the referees for their comments which helped to improve the paper.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2016 Springer International Publishing Switzerland
About this chapter
Cite this chapter
Taguchi, D. (2016). Stability Problem for One-Dimensional Stochastic Differential Equations with Discontinuous Drift. In: Donati-Martin, C., Lejay, A., Rouault, A. (eds) Séminaire de Probabilités XLVIII. Lecture Notes in Mathematics(), vol 2168. Springer, Cham. https://doi.org/10.1007/978-3-319-44465-9_4
Download citation
DOI: https://doi.org/10.1007/978-3-319-44465-9_4
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-44464-2
Online ISBN: 978-3-319-44465-9
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)