betaFeedback

WorksMathematics

Stochastic differential delay equations with random delay driven by fractional Brownian motion

Abstract

We prove an existence and uniqueness result for stochastic differential delay equations driven by a fractional Brownian motion with Hurst parameter H>12H > \tfrac{1}{2}, where the delay is a random process. Extending the pathwise Riemann–Stieltjes approach of Ferrante and Rovira \cite{FR2010}, we show that if the random delay τ(⋅,ω)τ(\cdot,ω) has almost surely γγ-Hölder continuous trajectories with γδ>(1−H)/Hγδ> (1-H)/H, where δ∈(0,1]δ\in(0,1] is the Hölder exponent of the gradient of the diffusion coefficient, then the equation has a unique solution with paths in C1−α(−r0,T;Rd)⊂W0α,∞(−r0,T;Rd)C^{1-α}(-r_0,T;R^d)\subset W_0^{α,\infty}(-r_0,T;R^d) for appropriate α∈(1−H,12)α\in(1-H,\tfrac12). We also prove that the solution has moments of all orders when γ>1/(2H)γ>1/(2H) and the Hölder constant of the delay is bounded.

Introduction

Stochastic differential delay equations (SDDEs) arise naturally in mathematical finance, population dynamics, control theory, and neuroscience, wherever the present state of a system depends on its past.

In [4], Ferrante and Rovira proved existence, uniqueness and moment estimates for the equation

X(t)=η(0)+∫0tb(s,X(s)) ds+∫0tσ(s,X(s−r)) dWsH,t∈(0,T],(1)X(t) = \eta(0) + \int_0^t b(s, X(s))\,\mathrm{d}s + \int_0^t \sigma(s, X(s-r))\,\mathrm{d}W_s^H,\qquad t \in(0,T], \tag*{(1)}

with a fixed delay r>0r > 0, using the Riemann–Stieltjes integral in the sense of Zähle [20]. In [5] they studied the convergence of solutions of (1) as r→0r \to0.

But it is well-known that sometimes delays are not constants, for instance in neural networks [18], in population biology and epidemiology [1, 13], in mathematical finance [2], or in dynamical systems [6].

Our goal is to extend the results of [5] to the equation

X(t)=η(0)+∫0tb(s,X(s),X(s−τ(s))) ds+∫0tσ(s,X(s),X(s−τ(s))) dWsH,t∈(0,T],X(t)=η(t),t∈[−r0,0].(2)\begin{aligned} X(t) &= \eta(0) + \int_{0}^{t} b(s, X(s), X(s-\tau(s)))\,\mathrm{d}s \\ &\quad+ \int_{0}^{t} \sigma(s, X(s), X(s-\tau(s)))\,\mathrm{d}W^{H}_{s}, \qquad t \in(0,T], \\ X(t) &= \eta(t), \qquad t \in[-r_{0},0]. \tag*{(2)} \end{aligned}

where τ:[0,T]×Ω→[0,r0]\tau: [0,T] \times\Omega\to[0,r_{0}] is a random delay with almost surely Hölder-continuous trajectories, r0>0r_{0} > 0 is the maximum delay, and η:[−r0,0]→Rd\eta: [-r_{0},0] \to\mathbb{R}^{d} is the initial path. Here both coefficients bb and σ\sigma depend pointwise on the current value X(s)X(s) and the delayed value X(s−τ(s))X(s-\tau(s)).

When the driving noise is a standard Brownian motion, the theory on delays, based on Itô calculus, is by now well developed. Stochastic functional differential equations with general path-dependent coefficients are treated in the monographs of Mohammed [14, 15] and Mao [11]. Random or state-dependent delays in the Itô setting have also attracted attention. Random delays appear in Mohammed [14] and in Kolmanovskii and Myshkis [8]. Mao, Shen and Yuan [12] study the almost sure asymptotic stability of neutral stochastic differential delay equations with Markovian switching.

For the fractional Brownian case, the literature on delays is much thinner. Ferrante and Rovira [4, 5] study delay equations by classical methods, using a pathwise Riemann–Stieltjes approach. Neuenkirch, Nourdin and Tindel [16] treat delay equations driven by rough paths, which cover fBm with H>13H > \frac{1}{3}, using the Lyons rough-path theory [10], and León and Tindel [9] study Malliavin calculus for fractional delay equations with fixed delay. As far as the author knows, it is the first study on random delays in the fBm setting.

In this paper, since H>12H > \frac{1}{2}, the integral ∫0tσ dWH\int_{0}^{t}\sigma\,\mathrm{d}W^{H} is defined pathwise as a Riemann–Stieltjes integral; see Young [19], Zähle [20] and Nualart–Răşcanu [17]. For each fixed ω\omega, WH(⋅,ω)W^{H}(\cdot,\omega) is a (H−ε)(H-\varepsilon)-Hölder function, while τ(⋅,ω)\tau(\cdot,\omega) is assumed to be γ\gamma-Hölder with γδ>(1−H)/H\gamma\delta> (1-H)/H. Under these assumptions we work pathwise: for each ω\omega we solve a deterministic integral equation with the specific realizations of (WH(⋅,ω),τ(⋅,ω))(W^{H}(\cdot,\omega),\tau(\cdot,\omega)). As in [17, 5], the stochastic result is then obtained by applying the deterministic one for almost every ω\omega.

The main difference with respect to [17, 5] is that we work with Hölder seminorms, following [7], and we use a contraction argument only on short time intervals. In the above papers the solution is obtained as the fixed point of a contraction on the whole interval [0,T][0,T], with respect to the weighted norms of the spaces W0α,∞W_{0}^{\alpha,\infty}, but when the delay is variable, we do not know how to control the delayed term X(s−τ(s))X(s-\tau(s)) in these norms.

The paper is organized as follows. Section 2 introduces the function spaces, defines the Riemann–Stieltjes integral, and states the hypotheses on the coefficients. Section 3 states the main result, Theorem 3.1. Section 4 contains the key new Lemma 4.1, which controls the regularity of the composition s↦f(s−τ(s))s \mapsto f(s-\tau(s)). Section 5 proves existence and uniqueness for the deterministic version of (2), together with an explicit a priori bound. Section 6 applies these results to the stochastic equation and completes the proof of Theorem 3.1.

Preliminaries

Function spaces

Fix parameters r0>0r_0>0, T>0T>0 and α∈(0,12)\alpha\in(0,\frac{1}{2}). We first recall the Hölder spaces.

Definition 2.1. For λ∈(0,1]\lambda\in(0,1] and an interval I⊂RI\subset\mathbb{R}, the Hölder space Cλ(I;Rd)C^\lambda(I;\mathbb{R}^d) consists of all continuous functions f:I→Rdf:I\to\mathbb{R}^d such that the seminorm

[f]Cλ(I):=sup⁡s,t∈Is≠t∣f(t)−f(s)∣∣t−s∣λ<∞.[f]_{C^\lambda(I)}:=\sup_{\substack{s,t\in I\\s\ne t}}\frac{|f(t)-f(s)|}{|t-s|^\lambda}<\infty.

We write CλC^\lambda for Cλ([−r0,T];Rd)C^\lambda([-r_0,T];\mathbb{R}^d) when the interval and target are clear, and ∥f∥Cλ(I):=∥f∥L∞(I)+[f]Cλ(I)\|f\|_{C^\lambda(I)}:=\|f\|_{L^\infty(I)}+[f]_{C^\lambda(I)}.

We shall also need the spaces W0α,∞W_0^{\alpha,\infty} of [17].

Definition 2.2. For f:[−r0,T]→Rdf:[-r_0,T]\to\mathbb{R}^d measurable, define the norm

∥f∥α,r0:=sup⁡t∈[−r0,T](∣f(t)∣+∫−r0t∣f(t)−f(s)∣(t−s)α+1 ds).(3)\|f\|_{\alpha,r_0}:=\sup_{t\in[-r_0,T]}\left(|f(t)|+\int_{-r_0}^{t}\frac{|f(t)-f(s)|}{(t-s)^{\alpha+1}}\,ds\right). \tag*{(3)}

We denote

W0α,∞(−r0,T;Rd):={f:[−r0,T]→Rd measurable:∥f∥α,r0<∞}.W_0^{\alpha,\infty}(-r_0,T;\mathbb{R}^d):=\{f:[-r_0,T]\to\mathbb{R}^d\ \text{measurable}:\|f\|_{\alpha,r_0}<\infty\}.

For functions defined only on [0,T][0,T], we write Wα,∞(0,T;Rd)W^{\alpha,\infty}(0,T;\mathbb{R}^d) with the same formula but integration from 00, and we denote the corresponding norm by ∥f∥α,0\|f\|_{\alpha,0}.

Remark 2.3. Notice that if ff is β\beta-Hölder continuous with β>α\beta>\alpha, then

∫−r0t∣f(t)−f(s)∣(t−s)α+1 ds≤[f]Cβ∫−r0t(t−s)β−α−1 ds=[f]Cβ(t+r0)β−αβ−α<∞.\int_{-r_0}^{t}\frac{|f(t)-f(s)|}{(t-s)^{\alpha+1}}\,ds\le[f]_{C^\beta}\int_{-r_0}^{t}(t-s)^{\beta-\alpha-1}\,ds=\frac{[f]_{C^\beta}(t+r_0)^{\beta-\alpha}}{\beta-\alpha}<\infty.

In particular, C1−α([−r0,T];Rd)⊂W0α,∞(−r0,T;Rd)C^{1-\alpha}([-r_0,T];\mathbb{R}^d)\subset W_0^{\alpha,\infty}(-r_0,T;\mathbb{R}^d) with

∥f∥α,r0≤Cα,T,r0∥f∥C1−α([−r0,T]).(4)\|f\|_{\alpha,r_0}\le C_{\alpha,T,r_0}\|f\|_{C^{1-\alpha}([-r_0,T])}. \tag*{(4)}

The Riemann–Stieltjes integral

The Riemann–Stieltjes integral ∫0tf(s) dgs\int_0^t f(s)\,dg_s is not well-defined for all pairs (f,g)(f,g) of Hölder functions. The theory of Nualart and Răşcanu [17], following Zähle [20], provides sufficient conditions via fractional calculus.

For the definition of the integral ∫f dg\int f\,dg and of the constant Λα(g)\Lambda_\alpha(g) associated with a function g:[0,T]→Rmg:[0,T]\to\mathbb{R}^m and α∈(0,12)\alpha\in(0,\frac{1}{2}) we refer to [17]; here we only recall the two facts that will be used.

Remark 2.4. If g∈Cβ′([0,T];Rm)g\in C^{\beta'}([0,T];\mathbb{R}^m) with β′>1−α\beta'>1-\alpha, then, as shown in [17], Λα(g)<∞\Lambda_\alpha(g)<\infty and, more precisely,

Λα(g)≤Cα,β′,T[g]Cβ′([0,T]).(5)\Lambda_\alpha(g)\le C_{\alpha,\beta',T}[g]_{C^{\beta'}([0,T])}. \tag*{(5)}

In particular, for g=WHg=W^H with H>12H>\frac{1}{2}, choosing α∈(1−H,12)\alpha\in(1-H,\frac{1}{2}) ensures Λα(WH)<∞\Lambda_\alpha(W^H)<\infty a.s.

The estimate that we shall use is the following one, from [17 p. 61]. Let gg satisfy Λα(g)<∞\Lambda_\alpha(g)<\infty, let 0≤u<s≤T0\le u<s\le T and let f:[u,s]→Rd×mf:[u,s]\to\mathbb{R}^{d\times m} be measurable with

∫us(∣f(r)∣(r−u)−α+α∫ur∣f(r)−f(w)∣(r−w)−α−1 dw) dr<∞.\int_u^s \left(|f(r)|(r-u)^{-\alpha}+\alpha\int_u^r|f(r)-f(w)|(r-w)^{-\alpha-1}\,dw\right)\,dr<\infty.

Then the integral ∫usf dg\int_u^s f\,dg is well defined and

∣∫usf(r) dgr∣≤Λα(g)∫us(∣f(r)∣(r−u)α+α∫ur∣f(r)−f(w)∣(r−w)α+1 dw) dr.(6)\left|\int_u^s f(r)\,dg_r\right|\le\Lambda_\alpha(g)\int_u^s\left(\frac{|f(r)|}{(r-u)^\alpha}+\alpha\int_u^r\frac{|f(r)-f(w)|}{(r-w)^{\alpha+1}}\,dw\right)\,dr. \tag*{(6)}

Moreover the integral is additive with respect to the interval of integration: ∫usf dg=∫uvf dg+∫vsf dg\int_u^s f\,dg=\int_u^v f\,dg+\int_v^s f\,dg for u<v<su<v<s. When g=(g1,…,gm)g=(g^1,\ldots,g^m) takes values in Rm\mathbb{R}^m and ff in Rd×m\mathbb{R}^{d\times m}, the integral is understood as ∫f dg=∑j=1m∫f⋅j dgj\int f\,dg=\sum_{j=1}^m\int f^{\cdot j}\,dg^j, where f⋅jf^{\cdot j} is the jj-th column of ff, and we set Λα(g):=∑j=1mΛα(gj)\Lambda_\alpha(g):=\sum_{j=1}^m\Lambda_\alpha(g^j). Since ∣f⋅j∣≤∣f∣|f^{\cdot j}|\le|f|, the estimate (6) then holds as stated, by applying the scalar estimate to each term of the sum.

Hypotheses

For (x,y)∈Rd×Rd(x,y)\in\mathbb{R}^d\times\mathbb{R}^d we use the norm ∣(x,y)∣:=∣x∣+∣y∣|(x,y)|:=|x|+|y|, we write ∇σ:=∇(x,y)σ\nabla\sigma:=\nabla_{(x,y)}\sigma for the gradient of σ\sigma with respect to both spatial arguments, and ∥⋅∥\|\cdot\| for the corresponding operator norm.

(H1) σ:[0,T]×Rd×Rd→Rd×m\sigma:[0,T]\times\mathbb{R}^d\times\mathbb{R}^d\to\mathbb{R}^{d\times m} is measurable, σ(t,⋅,⋅)\sigma(t,\cdot,\cdot) is C1C^1 for each tt, and there exist constants β,δ∈(0,1]\beta,\delta\in(0,1], M0>0M_0>0 and, for each N≥0N\ge0, a constant MN>0M_N>0 such that for all s,t∈[0,T]s,t\in[0,T]:

(i) ∣σ(t,x,y)−σ(t,x′,y′)∣≤M0(∣x−x′∣+∣y−y′∣)|\sigma(t,x,y)-\sigma(t,x',y')|\le M_0(|x-x'|+|y-y'|) for all x,y,x′,y′∈Rdx,y,x',y'\in\mathbb{R}^d;

(ii) ∥∇σ(t,x,y)−∇σ(t,x′,y′)∥≤MN(∣x−x′∣+∣y−y′∣)δ\|\nabla\sigma(t,x,y)-\nabla\sigma(t,x',y')\|\le M_N(|x-x'|+|y-y'|)^\delta for all x,y,x′,y′x,y,x',y' with ∣x∣,∣y∣,∣x′∣,∣y′∣≤N|x|,|y|,|x'|,|y'|\le N;

(iii) ∣σ(t,x,y)−σ(s,x,y)∣+∥∇σ(t,x,y)−∇σ(s,x,y)∥≤M0∣t−s∣β|\sigma(t,x,y)-\sigma(s,x,y)|+\|\nabla\sigma(t,x,y)-\nabla\sigma(s,x,y)\|\le M_0|t-s|^\beta for all x,y∈Rdx,y\in\mathbb{R}^d.

(H2) b:[0,T]×Rd×Rd→Rdb:[0,T]\times\mathbb{R}^d\times\mathbb{R}^d\to\mathbb{R}^d is measurable and there exist a function b0∈L1/α(0,T)b_0\in L^{1/\alpha}(0,T), a constant L0≥0L_0\ge0 and, for each N≥0N\ge0, a constant LN>0L_N>0 such that for all t∈[0,T]t\in[0,T]:

(i) ∣b(t,x,y)−b(t,x′,y′)∣≤LN(∣x−x′∣+∣y−y′∣)|b(t,x,y)-b(t,x',y')|\le L_N(|x-x'|+|y-y'|) for all x,y,x′,y′∈Rdx,y,x',y'\in\mathbb{R}^d such that ∣x∣,∣y∣,∣x′∣,∣y′∣≤N|x|,|y|,|x'|,|y'|\le N;

(ii) ∣b(t,x,y)∣≤L0(∣x∣+∣y∣)+b0(t)|b(t,x,y)|\le L_0(|x|+|y|)+b_0(t) for all x,y∈Rdx,y\in\mathbb{R}^d.

(Hτ_\tau) The random delay τ:[0,T]×Ω→[0,r0]\tau:[0,T]\times\Omega\to[0,r_0] is a measurable stochastic process, and there exist γ∈(0,1]\gamma\in(0,1] with

γδ>1−HH\gamma\delta>\frac{1-H}{H}

and a random variable Cτ≥0C_\tau\ge0, finite almost surely, such that, almost surely,

∣τ(t,ω)−τ(s,ω)∣≤Cτ(ω)∣t−s∣γ,∀s,t∈[0,T].(7)|\tau(t,\omega)-\tau(s,\omega)|\le C_\tau(\omega)|t-s|^\gamma,\qquad\forall s,t\in[0,T]. \tag*{(7)}

Remark 2.5. As in [17, 5], the Lipschitz condition (H1)(i) is global. It implies the linear growth bound

∣σ(t,x,y)∣≤K0+M0(∣x∣+∣y∣),K0:=∣σ(0,0,0)∣+M0Tβ,(8)|\sigma(t,x,y)|\le K_0+M_0(|x|+|y|),\qquad K_0:=|\sigma(0,0,0)|+M_0T^\beta, \tag*{(8)}

so no separate growth hypothesis on σ\sigma is needed.

Remark 2.6. When δ=1\delta= 1, for instance if σ(t,⋅,⋅)\sigma(t,\cdot,\cdot) is C2C^2 with locally bounded second derivatives, the condition in (Hτ)(\mathrm{H}_{\tau}) reads γ>(1−H)/H\gamma> (1-H)/H. It is satisfied, for example, by any Lipschitz delay, for which γ=1\gamma= 1, or by τ(t)=ψ(Yt)\tau(t) = \psi(Y_t) with ψ:R→[0,r0]\psi: \mathbb{R} \to[0,r_0] Lipschitz and YY a process with γ\gamma-Hölder paths.

Main result

Let (Ω,F,P)(\Omega,\mathcal{F},\mathbb{P}) be a complete probability space and WH=(WH,j)j=1mW^H = (W^{H,j})_{j=1}^{m} independent fractional Brownian motions with Hurst parameter H>12H > \frac{1}{2}. Recall that WH,jW^{H,j} is a centered Gaussian process with covariance E[WH,j(t)WH,j(s)]=12(t2H+s2H−∣t−s∣2H)\mathbb{E}[W^{H,j}(t)W^{H,j}(s)] = \frac{1}{2}(t^{2H} + s^{2H} - \lvert t-s\rvert^{2H}). We consider equation (2) under the hypotheses of § 2.3. The main result of the paper is the following.

Theorem 3.1. Let H>12H > \frac{1}{2}. Assume (H1)(\mathrm{H}1) with β>1−H\beta> 1-H, (H2)(\mathrm{H}2) and (Hτ)(\mathrm{H}_{\tau}), and set

α0:=min⁡{12,β,δγ1+δγ}>1−H.\alpha_0 := \min\left\{\frac{1}{2},\beta,\frac{\delta\gamma}{1+\delta\gamma}\right\} > 1-H.

Fix α∈(1−H,α0)\alpha\in(1-H,\alpha_0), assume b0∈L1/α(0,T)b_0 \in L^{1/\alpha}(0,T), and let η\eta be a stochastic process with trajectories in C1−α([−r0,0];Rd)C^{1-\alpha}([-r_0,0];\mathbb{R}^d) almost surely.

(a) There exists a unique stochastic process XX with trajectories in C1−α([−r0,T];Rd)C^{1-\alpha}([-r_0,T];\mathbb{R}^d) almost surely which is a solution of the stochastic equation (2). In particular,

X∈L0(Ω,F,P;W0α,∞(−r0,T;Rd)).X \in L^0(\Omega,\mathcal{F},\mathbb{P};W_0^{\alpha,\infty}(-r_0,T;\mathbb{R}^d)).

(b) Assume in addition that

γ>12H,α<1−12γ,Cτ∈L∞(Ω),E[∥η∥C1−α([−r0,0])p]<∞∀p≥1.(9)\gamma> \frac{1}{2H}, \qquad\alpha< 1-\frac{1}{2\gamma}, \qquad C_{\tau} \in L^\infty(\Omega), \qquad\mathbb{E}\left[\lVert\eta\rVert_{C^{1-\alpha}([-r_0,0])}^{p}\right] < \infty\quad\forall p \ge1. \tag*{(9)}

Then, for all p≥1p \ge1,

E[∥X∥C1−α([−r0,T])p]<∞,and henceE[∥X∥α,r0p]<∞.\mathbb{E}\left[\lVert X\rVert_{C^{1-\alpha}([-r_0,T])}^{p}\right] < \infty, \qquad\text{and hence} \qquad\mathbb{E}\left[\lVert X\rVert_{\alpha,r_0}^{p}\right] < \infty.

Remark 3.2. The inequality α0>1−H\alpha_0 > 1-H holds because β>1−H\beta> 1-H and because δγ/(1+δγ)>1−H\delta\gamma/(1+\delta\gamma) > 1-H is equivalent to δγ>(1−H)/H\delta\gamma> (1-H)/H, which is assumed in (Hτ)(\mathrm{H}_{\tau}). For a Lipschitz delay, that is, for γ=1\gamma= 1, the first two conditions in (9) hold automatically.

The proof of Theorem 3.1 is given in § 6. It relies on the composition lemma of § 4 and on the deterministic results of § 5.

The composition lemma

The following lemma is the key new ingredient of this paper.

Lemma 4.1. Let α∈(0,12)\alpha\in(0,\frac{1}{2}), γ>α/(1−α)\gamma> \alpha/(1-\alpha), r0,T>0r_0,T > 0. Suppose f∈C1−α(−r0,T;Rd)f \in C^{1-\alpha}(-r_0,T;\mathbb{R}^d) and τ:[0,T]→[0,r0]\tau: [0,T] \to[0,r_0] is γ\gamma-Hölder with constant [τ]γ:=[τ]Cγ([0,T])[\tau]_{\gamma} := [\tau]_{C^\gamma([0,T])}. Define fτ:[0,T]→Rdf^\tau: [0,T] \to\mathbb{R}^d by fτ(s):=f(s−τ(s))f^\tau(s) := f(s-\tau(s)). Then fτ∈W0α,∞(0,T;Rd)f^\tau\in W_0^{\alpha,\infty}(0,T;\mathbb{R}^d); more precisely:

(a) For all s,t∈[0,T]s,t\in[0,T],

∣fτ(t)−fτ(s)∣≤[f]C1−α(T1−γ+[τ]γ)1−α∣t−s∣γ(1−α).(10)|f^\tau(t)-f^\tau(s)|\le[f]_{C^{1-\alpha}}(T^{1-\gamma}+[\tau]_\gamma)^{1-\alpha}|t-s|^{\gamma(1-\alpha)}. \tag*{(10)}

(b) For all t∈[0,T]t\in[0,T],

∫0t∣fτ(t)−fτ(s)∣(t−s)α+1 ds≤(T1−γ+[τ]γ)1−α[f]C1−αTγ(1−α)−αγ(1−α)−α.(11)\int_0^t\frac{|f^\tau(t)-f^\tau(s)|}{(t-s)^{\alpha+1}}\,ds\le\frac{(T^{1-\gamma}+[\tau]_\gamma)^{1-\alpha}[f]_{C^{1-\alpha}}T^{\gamma(1-\alpha)-\alpha}}{\gamma(1-\alpha)-\alpha}. \tag*{(11)}

(c) ∥fτ∥α,0≤∥f∥α,r0+(T1−γ+[τ]γ)1−α[f]C1−αTγ(1−α)−αγ(1−α)−α\|f^\tau\|_{\alpha,0}\le\|f\|_{\alpha,r_0}+\dfrac{(T^{1-\gamma}+[\tau]_\gamma)^{1-\alpha}[f]_{C^{1-\alpha}}T^{\gamma(1-\alpha)-\alpha}}{\gamma(1-\alpha)-\alpha}.

Proof. We first obtain the uniform bound. Since t−τ(t)∈[−r0,T]t-\tau(t)\in[-r_0,T], we have ∣fτ(t)∣≤∥f∥L∞(−r0,T)≤∥f∥α,r0|f^\tau(t)|\le\|f\|_{L^\infty(-r_0,T)}\le\|f\|_{\alpha,r_0}.

We now prove the Hölder estimate (a). Since f∈C1−α(−r0,T)f\in C^{1-\alpha}(-r_0,T),

∣fτ(t)−fτ(s)∣=∣f(t−τ(t))−f(s−τ(s))∣≤[f]C1−α∣(t−τ(t))−(s−τ(s))∣1−α.(12)|f^\tau(t)-f^\tau(s)|=|f(t-\tau(t))-f(s-\tau(s))|\le[f]_{C^{1-\alpha}}|(t-\tau(t))-(s-\tau(s))|^{1-\alpha}. \tag*{(12)}

Since γ≤1\gamma\le1 and ∣t−s∣≤T|t-s|\le T, we have ∣t−s∣=∣t−s∣1−γ⋅∣t−s∣γ≤T1−γ∣t−s∣γ|t-s|=|t-s|^{1-\gamma}\cdot|t-s|^\gamma\le T^{1-\gamma}|t-s|^\gamma, so

∣(t−τ(t))−(s−τ(s))∣≤∣t−s∣+∣τ(t)−τ(s)∣≤T1−γ∣t−s∣γ+[τ]γ∣t−s∣γ=(T1−γ+[τ]γ)∣t−s∣γ.(13)\begin{aligned} |(t-\tau(t))-(s-\tau(s))|&\le|t-s|+|\tau(t)-\tau(s)|\\ &\le T^{1-\gamma}|t-s|^\gamma+[\tau]_\gamma|t-s|^\gamma=(T^{1-\gamma}+[\tau]_\gamma)|t-s|^\gamma. \tag*{(13)} \end{aligned}

Inserting into (12) gives (a).

To obtain (b), we use part (a):

∫0t∣fτ(t)−fτ(s)∣(t−s)α+1 ds≤[f]C1−α(T1−γ+[τ]γ)1−α∫0t(t−s)γ(1−α)−α−1 ds.\int_0^t\frac{|f^\tau(t)-f^\tau(s)|}{(t-s)^{\alpha+1}}\,ds\le[f]_{C^{1-\alpha}}(T^{1-\gamma}+[\tau]_\gamma)^{1-\alpha}\int_0^t(t-s)^{\gamma(1-\alpha)-\alpha-1}\,ds.

The exponent satisfies γ(1−α)−α−1>−1\gamma(1-\alpha)-\alpha-1>-1 if and only if γ(1−α)>α\gamma(1-\alpha)>\alpha, i.e. γ>α/(1−α)\gamma>\alpha/(1-\alpha), which holds by assumption. The integral equals tγ(1−α)−α/(γ(1−α)−α)≤Tγ(1−α)−α/(γ(1−α)−α)t^{\gamma(1-\alpha)-\alpha}/(\gamma(1-\alpha)-\alpha)\le T^{\gamma(1-\alpha)-\alpha}/(\gamma(1-\alpha)-\alpha), giving (b).

Finally, combining the uniform bound with (b) we obtain (c). □

The deterministic equation

In this section we fix ω∈Ω\omega\in\Omega and work deterministically with the realizations g:=WH(⋅,ω)g:=W^H(\cdot,\omega) and τ:=τ(⋅,ω)\tau:=\tau(\cdot,\omega). Thus g:[0,T]→Rmg:[0,T]\to\mathbb{R}^m is a fixed function with Λα(g)<∞\Lambda_\alpha(g)<\infty and τ:[0,T]→[0,r0]\tau:[0,T]\to[0,r_0] is a fixed γ\gamma-Hölder function.

Notation

Throughout this section (H1) and (H2) hold, and

α∈(0,α0),α0:=min⁡{12,β,δγ1+δγ}.(14)\alpha\in(0,\alpha_0),\qquad\alpha_0:=\min\left\{\frac{1}{2},\beta,\frac{\delta\gamma}{1+\delta\gamma}\right\}. \tag*{(14)}

We write

μ:=1−α,Λ:=Λα(g),T:=T1−γ+[τ]γ,\mu:=1-\alpha,\qquad\Lambda:=\Lambda_\alpha(g),\qquad\mathcal{T}:=T^{1-\gamma}+[\tau]_\gamma,
κ:=min⁡{β,δγμ},c0:=1κ−α,C1:=∥b0∥L1/α(0,T)+K0+M016M0,\kappa:=\min\{\beta,\delta\gamma\mu\},\qquad c_0:=\frac{1}{\kappa-\alpha},\qquad C_1:=\|b_0\|_{L^{1/\alpha}(0,T)}+\frac{K_0+M_0}{16M_0},

where K0K_0 is the constant in (8). By (14) we have β>α\beta> \alpha and δγμ>α\delta\gamma\mu> \alpha, so κ>α\kappa> \alpha and c0<∞c_0 < \infty; note also that every exponent θ∈{1,β,μ,γμ,δγμ}\theta\in\{1,\beta,\mu,\gamma\mu,\delta\gamma\mu\} satisfies 1/(θ−α)≤c01/(\theta-\alpha) \le c_0.

For an interval II we write ∥f∥∞,I:=sup⁡I∣f∣\|f\|_{\infty,I} := \sup_I |f| and [f]μ,I:=[f]Cμ(I)[f]_{\mu,I} := [f]_{C^\mu(I)}. For 0<b≤T0 < b \le T and f:[−r0,b]→Rdf : [-r_0,b] \to\mathbb{R}^d we set fτ(s):=f(s−τ(s))f^\tau(s) := f(s-\tau(s)), Φf(s):=(f(s),fτ(s))\Phi_f(s) := (f(s),f^\tau(s)), s∈[0,b]s \in[0,b].

By Lemma 4.1(a) and ∣s−u∣μ≤T(1−γ)μ∣s−u∣γμ≤Tμ∣s−u∣γμ|s-u|^\mu\le T^{(1-\gamma)\mu}|s-u|^{\gamma\mu} \le T^\mu|s-u|^{\gamma\mu}, for all u,s∈[0,b]u,s \in[0,b]:

∣fτ(s)−fτ(u)∣≤Tμ[f]μ,[−r0,b]∣s−u∣γμ,∣Φf(s)−Φf(u)∣≤2Tμ[f]μ,[−r0,b]∣s−u∣γμ.(15)|f^\tau(s)-f^\tau(u)| \le T^\mu[f]_{\mu,[-r_0,b]}|s-u|^{\gamma\mu}, \qquad|\Phi_f(s)-\Phi_f(u)| \le2T^\mu[f]_{\mu,[-r_0,b]}|s-u|^{\gamma\mu}. \tag*{(15)}

Given $0 \le a < b \le T$ and f∈Cμ([−r0,b];Rd)f \in C^\mu([-r_0,b];\mathbb{R}^d) we define, for t∈[a,b]t \in[a,b],

Fa(f)(t):=∫atb(r,f(r),fτ(r)) dr.(16)F_a(f)(t) := \int_a^t b(r,f(r),f^\tau(r))\,dr. \tag*{(16)}
Ga(f)(t):=∫atσ(r,f(r),fτ(r)) dgr.(17)G_a(f)(t) := \int_a^t \sigma(r,f(r),f^\tau(r))\,d g_r. \tag*{(17)}

Estimates on the intervals

We start with two basic lemmas. The first lemma is the form of the estimate (6) that we shall use, for Hölder continuous integrands.

Lemma 5.1. Let $0 \le u < s \le T$ and let φ:[u,s]→Rd×m\varphi: [u,s] \to\mathbb{R}^{d\times m} be bounded and such that

∣φ(r)−φ(w)∣≤∑i=1kai∣r−w∣θi,r,w∈[u,s],|\varphi(r)-\varphi(w)| \le\sum_{i=1}^k a_i|r-w|^{\theta_i}, \qquad r,w \in[u,s],

for some ai≥0a_i \ge0 and θi∈(α,1]\theta_i \in(\alpha,1]. Then ∫usφ dg\int_u^s \varphi\,dg is well defined and

∣∫usφ(r) dgr∣≤Λ(s−u)1−α(∥φ∥∞,[u,s]1−α+∑i=1kai(s−u)θiθi−α).(18)\left|\int_u^s \varphi(r)\,d g_r\right| \le\Lambda(s-u)^{1-\alpha}\left(\frac{\|\varphi\|_{\infty,[u,s]}}{1-\alpha}+\sum_{i=1}^k\frac{a_i(s-u)^{\theta_i}}{\theta_i-\alpha}\right). \tag*{(18)}

Proof. We insert the assumptions into (6). First,

∫us∣φ(r)∣(r−u)−α dr≤∥φ∥∞,[u,s](s−u)1−α/(1−α).\int_u^s |\varphi(r)|(r-u)^{-\alpha}\,dr \le\|\varphi\|_{\infty,[u,s]}(s-u)^{1-\alpha}/(1-\alpha).

Second, for each ii,

α∫us∫ur(r−w)θi−α−1 dw dr=α(s−u)1+θi−α(θi−α)(1+θi−α)≤(s−u)1−α(s−u)θiθi−α,\alpha\int_u^s\int_u^r(r-w)^{\theta_i-\alpha-1}\,dw\,dr = \frac{\alpha(s-u)^{1+\theta_i-\alpha}}{(\theta_i-\alpha)(1+\theta_i-\alpha)} \le\frac{(s-u)^{1-\alpha}(s-u)^{\theta_i}}{\theta_i-\alpha},

because α<1≤1+θi−α\alpha< 1 \le1+\theta_i-\alpha. □\square

The second lemma collects the estimates for the increments of σ\sigma.

Lemma 5.2. Assume (H1)(H1) and write z=(x,y)z=(x,y), ∣z∣=∣x∣+∣y∣|z|=|x|+|y|. For all s,t∈[0,T]s,t \in[0,T]:

(a) ∣σ(t,z)−σ(s,z′)∣≤M0∣t−s∣β+M0∣z−z′∣|\sigma(t,z)-\sigma(s,z')| \le M_0|t-s|^\beta+M_0|z-z'| for all z,z′z,z';

(b) ∣[σ(t,z1)−σ(s,z1)]−[σ(t,z3)−σ(s,z3)]∣≤M0∣t−s∣β∣z1−z3∣|[\sigma(t,z_1)-\sigma(s,z_1)]-[\sigma(t,z_3)-\sigma(s,z_3)]| \le M_0|t-s|^\beta|z_1-z_3| for all z1,z3z_1,z_3;

(c) if zi=(xi,yi)z_i=(x_i,y_i) with ∣xi∣,∣yi∣≤N|x_i|,|y_i|\le N for i=1,…,4i=1,\ldots,4, then

∣σ(t,z1)−σ(t,z2)−σ(t,z3)+σ(t,z4)∣≤M0∣z1−z2−z3+z4∣+MN∣z1−z3∣(∣z1−z2∣δ+∣z3−z4∣δ).(19)\left|\sigma(t,z_1)-\sigma(t,z_2)-\sigma(t,z_3)+\sigma(t,z_4)\right| \le M_0|z_1-z_2-z_3+z_4|+M_N|z_1-z_3|\left(|z_1-z_2|^\delta+|z_3-z_4|^\delta\right). \tag*{(19)}

Proof. (a) follows from (H1)(i) and (H1)(iii). Note that (H1)(i) gives ∥∇σ(t,z)∥≤M0\|\nabla\sigma(t,z)\|\le M_0 for all zz.

(b) Apply the mean value theorem in integral form to the C1C^1 map ψ(z):=σ(t,z)−σ(s,z)\psi(z):=\sigma(t,z)-\sigma(s,z) on the segment [z3,z1][z_3,z_1]:

ψ(z1)−ψ(z3)=∫01∇ψ(z3+θ(z1−z3)) dθ (z1−z3),\psi(z_1)-\psi(z_3)=\int_0^1\nabla\psi\left(z_3+\theta(z_1-z_3)\right)\,\mathrm{d}\theta\,(z_1-z_3),

and by (H1)(iii) the gradient of ψ\psi is bounded in norm by M0∣t−s∣βM_0|t-s|^\beta. Here the gradient is taken with respect to both arguments xx and yy, which is why (H1)(iii) is stated for the full gradient.

(c) By the mean value theorem in integral form, applied in the variable z=(x,y)z=(x,y),

σ(t,z1)−σ(t,z3)=∫01∇σ(t,z3+θ(z1−z3)) dθ (z1−z3),\sigma(t,z_1)-\sigma(t,z_3)=\int_0^1\nabla\sigma\left(t,z_3+\theta(z_1-z_3)\right)\,\mathrm{d}\theta\,(z_1-z_3),
σ(t,z2)−σ(t,z4)=∫01∇σ(t,z4+θ(z2−z4)) dθ (z2−z4).\sigma(t,z_2)-\sigma(t,z_4)=\int_0^1\nabla\sigma\left(t,z_4+\theta(z_2-z_4)\right)\,\mathrm{d}\theta\,(z_2-z_4).

Subtracting, and adding and subtracting ∫01∇σ(t,z4+θ(z2−z4)) dθ (z1−z3)\int_0^1\nabla\sigma\left(t,z_4+\theta(z_2-z_4)\right)\,\mathrm{d}\theta\,(z_1-z_3), we get

σ(t,z1)−σ(t,z2)−σ(t,z3)+σ(t,z4)=∫01∇σ(t,z4+θ(z2−z4)) dθ[(z1−z3)−(z2−z4)]+∫01[∇σ(t,z3+θ(z1−z3))−∇σ(t,z4+θ(z2−z4))] dθ (z1−z3).\begin{aligned} &\sigma(t,z_1)-\sigma(t,z_2)-\sigma(t,z_3)+\sigma(t,z_4)\\ &=\int_0^1\nabla\sigma\left(t,z_4+\theta(z_2-z_4)\right)\,\mathrm{d}\theta\left[(z_1-z_3)-(z_2-z_4)\right]\\ &\quad+\int_0^1\left[\nabla\sigma\left(t,z_3+\theta(z_1-z_3)\right)-\nabla\sigma\left(t,z_4+\theta(z_2-z_4)\right)\right]\,\mathrm{d}\theta\,(z_1-z_3). \end{aligned}

The first term is bounded by M0∣z1−z2−z3+z4∣M_0|z_1-z_2-z_3+z_4|. In the second term all the points involved lie in the convex set {∣x∣≤N, ∣y∣≤N}\{|x|\le N,\ |y|\le N\} and

∣z3+θ(z1−z3)−z4−θ(z2−z4)∣=∣(1−θ)(z3−z4)+θ(z1−z2)∣≤∣z3−z4∣+∣z1−z2∣,\left|z_3+\theta(z_1-z_3)-z_4-\theta(z_2-z_4)\right| =\left|(1-\theta)(z_3-z_4)+\theta(z_1-z_2)\right| \le|z_3-z_4|+|z_1-z_2|,

so by (H1)(ii) and (p+q)δ≤pδ+qδ(p+q)^\delta\le p^\delta+q^\delta the second term is bounded by

MN∣z1−z3∣(∣z1−z2∣δ+∣z3−z4∣δ).M_N|z_1-z_3|\left(|z_1-z_2|^\delta+|z_3-z_4|^\delta\right).

We can now estimate the Hölder seminorms of Fa(f)F_a(f) and Ga(f)G_a(f) on a short interval.

Proposition 5.3. Let 0≤a<b≤T0\le a<b\le T with ℓ:=b−a≤1\ell:=b-a\le1 and f∈Cμ([−r0,b];Rd)f\in C^\mu([-r_0,b];\mathbb{R}^d). Then Fa(f),Ga(f)∈Cμ([a,b];Rd)F_a(f),G_a(f)\in C^\mu([a,b];\mathbb{R}^d) and

[Fa(f)]μ,[a,b]≤2L0ℓα∥f∥∞,[−r0,b]+∥b0∥L1/α(0,T),(20)[F_a(f)]_{\mu,[a,b]}\le2L_0\ell^\alpha\|f\|_{\infty,[-r_0,b]}+\|b_0\|_{L^{1/\alpha}(0,T)}, \tag*{(20)}
[Ga(f)]μ,[a,b]≤c0Λ(K0+M0+2M0∥f∥∞,[−r0,b]+M0(1+Tμ)ℓγμ[f]μ,[−r0,b]).(21)[G_a(f)]_{\mu,[a,b]}\le c_0\Lambda\left(K_0+M_0+2M_0\|f\|_{\infty,[-r_0,b]}+M_0(1+T^\mu)\ell^{\gamma\mu}[f]_{\mu,[-r_0,b]}\right). \tag*{(21)}

Proof. Let a≤u<s≤ba \le u < s \le b. By (H2)(ii) and Hölder’s inequality,

∣Fa(f)(s)−Fa(f)(u)∣≤2L0∥f∥∞,[−r0,b](s−u)+∥b0∥L1/α(s−u)1−α,|F_a(f)(s) - F_a(f)(u)| \le2L_0\|f\|_{\infty,[-r_0,b]}(s-u) + \|b_0\|_{L^{1/\alpha}}(s-u)^{1-\alpha},

and (s−u)≤ℓα(s−u)μ(s-u) \le\ell^\alpha(s-u)^\mu; this gives (20).

Put φ(r):=σ(r,Φf(r))\varphi(r) := \sigma(r,\Phi_f(r)). By (8), ∥φ∥∞,[a,b]≤K0+2M0∥f∥∞,[−r0,b]\|\varphi\|_{\infty,[a,b]} \le K_0 + 2M_0\|f\|_{\infty,[-r_0,b]}. By Lemma 5.2(a) and (15), for r,w∈[a,b]r,w \in[a,b],

∣φ(r)−φ(w)∣≤M0∣r−w∣β+M0[f]μ,[−r0,b]∣r−w∣μ+M0Tμ[f]μ,[−r0,b]∣r−w∣γμ.|\varphi(r)-\varphi(w)| \le M_0|r-w|^\beta+ M_0[f]_{\mu,[-r_0,b]}|r-w|^\mu+ M_0T^\mu[f]_{\mu,[-r_0,b]}|r-w|^{\gamma\mu}.

Lemma 5.1 with exponents β,μ,γμ\beta,\mu,\gamma\mu, all larger than α\alpha, together with s−u≤ℓ≤1s-u \le\ell\le1 and the definition of c0c_0, yields

∣∫usφ dg∣≤c0Λ(s−u)μ(K0+2M0∥f∥∞,[−r0,b]+M0+M0(1+Tμ)ℓγμ[f]μ,[−r0,b]),\left|\int_u^s \varphi\,dg\right| \le c_0\Lambda(s-u)^\mu\left(K_0+2M_0\|f\|_{\infty,[-r_0,b]}+M_0+M_0(1+T^\mu)\ell^{\gamma\mu}[f]_{\mu,[-r_0,b]}\right),

which is (21). □

Next we prove the corresponding Lipschitz estimate, for two functions that coincide up to time aa.

Proposition 5.4. Let $0 \le a < b \le T$ with ℓ:=b−a≤1\ell:= b-a \le1, and let f,h∈C1([−r0,b];Rd)f,h \in C^1([-r_0,b];\mathbb{R}^d) be such that

f=hf=h on [−r0,a][-r_0,a], ∥f∥∞,[−r0,b],∥h∥∞,[−r0,b]≤N\|f\|_{\infty,[-r_0,b]},\|h\|_{\infty,[-r_0,b]} \le N, [f]μ,[−r0,b],[h]μ,[−r0,b]≤R[f]_{\mu,[-r_0,b]},[h]_{\mu,[-r_0,b]} \le R.

Then

[Fa(f)−Fa(h)]μ,[a,b]≤2LNℓ[f−h]μ,[a,b],(22)[F_a(f)-F_a(h)]_{\mu,[a,b]} \le2LN\ell[f-h]_{\mu,[a,b]}, \tag*{(22)}
[Ga(f)−Ga(h)]μ,[a,b]≤c0Λℓγμ(5M0+M0Tμ+8MNTδμRδ)[f−h]μ,[a,b].(23)[G_a(f)-G_a(h)]_{\mu,[a,b]} \le c_0\Lambda\ell^{\gamma\mu}\left(5M_0+M_0T^\mu+8MN T^{\delta\mu}R^\delta\right)[f-h]_{\mu,[a,b]}. \tag*{(23)}

Proof. Write Δ:=f−h\Delta:= f-h and d:=[Δ]μ,[a,b]d := [\Delta]_{\mu,[a,b]}. Since Δ=0\Delta=0 on [−r0,a][-r_0,a],

∥Δ∥∞,[−r0,b]≤ℓμd,[Δ]μ,[−r0,b]=d.(24)\|\Delta\|_{\infty,[-r_0,b]} \le\ell^\mu d,\qquad[\Delta]_{\mu,[-r_0,b]}=d. \tag*{(24)}

For the second identity, note that if u≤a<su \le a < s then

∣Δ(s)−Δ(u)∣=∣Δ(s)−Δ(a)∣≤d(s−a)μ≤d(s−u)μ.|\Delta(s)-\Delta(u)|=|\Delta(s)-\Delta(a)|\le d(s-a)^\mu\le d(s-u)^\mu.

Consequently, by Lemma 4.1(a) applied to Δ\Delta, for all u,s∈[a,b]u,s \in[a,b],

∣Δτ(s)∣≤ℓμd,∣Δτ(s)−Δτ(u)∣≤Tμd∣s−u∣γμ.(25)|\Delta^\tau(s)|\le\ell^\mu d,\qquad|\Delta^\tau(s)-\Delta^\tau(u)|\le T^\mu d|s-u|^{\gamma\mu}. \tag*{(25)}

By (H2)(i), for a≤u<s≤ba \le u < s \le b,

∣(Fa(f)−Fa(h))(s)−(Fa(f)−Fa(h))(u)∣≤LN∫us(∣Δ(r)∣+∣Δτ(r)∣) dr≤2LNℓμd(s−u),|(F_a(f)-F_a(h))(s)-(F_a(f)-F_a(h))(u)|\le L_N\int_u^s(|\Delta(r)|+|\Delta^\tau(r)|)\,dr\le2LN\ell^\mu d(s-u),

and (s−u)≤ℓα(s−u)μ(s-u)\le\ell^\alpha(s-u)^\mu; this gives (22).

Put v(s):=σ(s,Φf(s))−σ(s,Φh(s))v(s):=\sigma(s,\Phi_f(s))-\sigma(s,\Phi_h(s)), so that Ga(f)−Ga(h)=∫av dgG_a(f)-G_a(h)=\int_a^v\,dg. By (H1)(i), (24) and (25),

∣v(s)∣≤M0(∣Δ(s)∣+∣Δτ(s)∣)≤2M0ℓμd.(26)|v(s)|\le M_0(|\Delta(s)|+|\Delta^\tau(s)|)\le2M_0\ell^\mu d. \tag*{(26)}

For u,s∈[a,b]u,s \in[a,b] we add and subtract σ(u,Φf(s))−σ(u,Φh(s))\sigma(u,\Phi_f(s))-\sigma(u,\Phi_h(s)) and obtain the exact identity

v(s)−v(u)=J1(s,u)+J2(s,u),(27)v(s)-v(u)=J_1(s,u)+J_2(s,u), \tag*{(27)}

where

J1(s,u):=[σ(s,Φf(s))−σ(u,Φf(s))]−[σ(s,Φh(s))−σ(u,Φh(s))],(28)J_1(s,u):=\left[\sigma(s,\Phi_f(s))-\sigma(u,\Phi_f(s))\right]-\left[\sigma(s,\Phi_h(s))-\sigma(u,\Phi_h(s))\right], \tag*{(28)}
J2(s,u):=σ(u,Φf(s))−σ(u,Φf(u))−σ(u,Φh(s))+σ(u,Φh(u)).(29)J_2(s,u):=\sigma(u,\Phi_f(s))-\sigma(u,\Phi_f(u))-\sigma(u,\Phi_h(s))+\sigma(u,\Phi_h(u)). \tag*{(29)}

We can get a bound on J1J_1. By Lemma 5.2(b) with z1=Φf(s)z_1=\Phi_f(s), z3=Φh(s)z_3=\Phi_h(s),

∣J1(s,u)∣≤M0∣s−u∣β(∣Δ(s)∣+∣Δτ(s)∣)≤2M0ℓμd∣s−u∣β.(30)\lvert J_1(s,u)\rvert\leq M_0\lvert s-u\rvert^\beta\left(\lvert\Delta(s)\rvert+\lvert\Delta^\tau(s)\rvert\right)\leq2M_0\ell^\mu d\lvert s-u\rvert^\beta. \tag*{(30)}

On the other hand, to bound J2J_2, we apply Lemma 5.2(c) at time uu with

z1=Φf(s),z2=Φf(u),z3=Φh(s),z4=Φh(u),z_1=\Phi_f(s),\qquad z_2=\Phi_f(u),\qquad z_3=\Phi_h(s),\qquad z_4=\Phi_h(u),

all of whose components are bounded by NN. Here

∣z1−z2−z3+z4∣=∣Δ(s)−Δ(u)∣+∣Δτ(s)−Δτ(u)∣≤d∣s−u∣μ+Tμd∣s−u∣γμ,\lvert z_1-z_2-z_3+z_4\rvert=\lvert\Delta(s)-\Delta(u)\rvert+\lvert\Delta^\tau(s)-\Delta^\tau(u)\rvert\leq d\lvert s-u\rvert^\mu+T^\mu d\lvert s-u\rvert^{\gamma\mu},
∣z1−z3∣=∣Δ(s)∣+∣Δτ(s)∣≤2ℓμd,\lvert z_1-z_3\rvert=\lvert\Delta(s)\rvert+\lvert\Delta^\tau(s)\rvert\leq2\ell^\mu d,
∣z1−z2∣δ+∣z3−z4∣δ≤2(2TμR)δ∣s−u∣δγμ,\lvert z_1-z_2\rvert^\delta+\lvert z_3-z_4\rvert^\delta\leq2(2T^\mu R)^\delta\lvert s-u\rvert^{\delta\gamma\mu},

where we used (24), (25) and (15). Hence, since 22+δ≤82^{2+\delta}\leq8,

∣J2(s,u)∣≤M0d∣s−u∣μ+M0Tμd∣s−u∣γμ+8MNTδμRδℓμd∣s−u∣δγμ.(31)\lvert J_2(s,u)\rvert\leq M_0d\lvert s-u\rvert^\mu+M_0T^\mu d\lvert s-u\rvert^{\gamma\mu}+8M_NT^{\delta\mu}R^\delta\ell^\mu d\lvert s-u\rvert^{\delta\gamma\mu}. \tag*{(31)}

Then, by (27), (30) and (31), vv satisfies the assumption of Lemma 5.1 on every [u,s]⊂[a,b][u,s]\subset[a,b] with the exponents β\beta, μ\mu, γμ\gamma\mu, δγμ\delta\gamma\mu, all larger than α\alpha by (14). Using (18), (26), the definition of c0c_0 and s−u≤ℓ≤1s-u\leq\ell\leq1,

∣∫usv dg∣≤c0Λ(s−u)μd(2M0ℓμ+2M0ℓμ+β+M0ℓμ+M0Tμℓγμ+8MNTδμRδℓμ+δγμ)≤c0Λ(s−u)μdℓγμ(5M0+M0Tμ+8MNTδμRδ),\begin{aligned} \left\lvert\int_u^s v\,dg\right\rvert\leq c_0\Lambda(s-u)^\mu d\left(2M_0\ell^\mu+2M_0\ell^{\mu+\beta}+M_0\ell^\mu+M_0T^\mu\ell^{\gamma\mu}+8M_NT^{\delta\mu}R^\delta\ell^{\mu+\delta\gamma\mu}\right) \\ &\leq c_0\Lambda(s-u)^\mu d\ell^{\gamma\mu}\left(5M_0+M_0T^\mu+8M_NT^{\delta\mu}R^\delta\right), \end{aligned}

which is (23).

A priori bound

Consider the deterministic integral equation

x(t)=η(0)+∫0tb(r,x(r),xτ(r)) dr+∫0tσ(r,x(r),xτ(r)) dgr,t∈(0,T],x(t)=\eta(0)+\int_0^t b(r,x(r),x^\tau(r))\,dr+\int_0^t \sigma(r,x(r),x^\tau(r))\,dg_r,\qquad t\in(0,T],
x(t)=η(t),t∈[−r0,0],(32)x(t)=\eta(t),\qquad t\in[-r_0,0], \tag*{(32)}

where η∈Cμ([−r0,0];Rd)\eta\in C^\mu([-r_0,0];\mathbb{R}^d) is the initial path. For T′∈(0,T]T'\in(0,T], a solution on [0,T′][0,T'] is a function x∈Cμ([−r0,T′];Rd)x\in C^\mu([-r_0,T'];\mathbb{R}^d) satisfying (32) for t∈(0,T′]t\in(0,T']. Clearly, a solution on [0,T′][0,T'] satisfies, for every 0≤a<b≤T′0\leq a<b\leq T',

x(t)=x(a)+Fa(x)(t)+Ga(x)(t),t∈[a,b].(33)x(t)=x(a)+F_a(x)(t)+G_a(x)(t),\qquad t\in[a,b]. \tag*{(33)}

The following proposition gives an a priori bound for the solutions of (32).

Proposition 5.5. Define

ℓ:=min⁡{1,116L0,(16c0M0(1+Tμ)Λ)−1/(γμ)},n:=⌈T/ℓ⌉.(34)\ell:= \min\left\{1,\frac{1}{16L_0},\left(16c_0M_0(1+T^\mu)\Lambda\right)^{-1/(\gamma\mu)}\right\},\qquad n:=\lceil T/\ell\rceil. \tag*{(34)}

Here we use the convention 1/0=+∞1/0=+\infty. Let T′∈(0,T]T'\in(0,T] and let xx be a solution of (32) on [0,T′][0,T']. Then

∥x∥∞,[−r0,T′]+ℓμ[x]μ,[−r0,T′]≤2n(∥η∥Cμ([−r0,0])+4C1).(35)\|x\|_{\infty,[-r_0,T']}+\ell^\mu[x]_{\mu,[-r_0,T']}\le2^n\left(\|\eta\|_{C^\mu([-r_0,0])}+4C_1\right). \tag*{(35)}

Moreover,

∥x∥C1−α([−r0,T′])≤(1+ℓ−μ)1+T/ℓ(∥η∥C1−α([−r0,0])+4C1).(36)\|x\|_{C^{1-\alpha}([-r_0,T'])}\le(1+\ell^{-\mu})^{1+T/\ell}\left(\|\eta\|_{C^{1-\alpha}([-r_0,0])}+4C_1\right). \tag*{(36)}

Proof. For t∈[0,T′]t\in[0,T'] set

At:=∥x∥∞,[−r0,t],Ht:=[x]μ,[−r0,t],Zt:=At+ℓμHt.A_t:=\|x\|_{\infty,[-r_0,t]},\qquad H_t:=[x]_{\mu,[-r_0,t]},\qquad Z_t:=A_t+\ell^\mu H_t.

which are finite because x∈Cμ([−r0,T′])x\in C^\mu([-r_0,T']). Let 0≤a<b≤T′0\le a<b\le T' with ℓ′:=b−a≤ℓ\ell':=b-a\le\ell and put h:=[x]μ,[a,b]h:=[x]_{\mu,[a,b]}. Since x(t)−x(a)x(t)-x(a) is controlled by hh on [a,b][a,b], and the Hölder seminorm on the union of two adjacent intervals is at most the sum of the two seminorms,

Ab≤Aa+ℓμh,Hb≤Ha+h,soZb≤Za+2ℓμh.(37)A_b\le A_a+\ell^\mu h,\qquad H_b\le H_a+h,\qquad\text{so}\qquad Z_b\le Z_a+2\ell^\mu h. \tag*{(37)}

By (33), h≤[Fa(x)]μ,[a,b]+[Ga(x)]μ,[a,b]h\le[F_a(x)]_{\mu,[a,b]}+[G_a(x)]_{\mu,[a,b]}. We apply Proposition 5.3 on [a,b][a,b], whose length is ℓ′≤ℓ≤1\ell'\le\ell\le1, with f=xf=x, so that ∥f∥∞,[−r0,b]=Ab\|f\|_{\infty,[-r_0,b]}=A_b and [f]μ,[−r0,b]=Hb[f]_{\mu,[-r_0,b]}=H_b, and we bound ℓ′\ell' by ℓ\ell:

h≤2L0ℓαAb+∥b0∥L1/α+c0Λ(K0+M0)+2c0ΛM0Ab+c0ΛM0(1+Tμ)ℓγμHb.(38)h\le2L_0\ell^\alpha A_b+\|b_0\|_{L^{1/\alpha}}+c_0\Lambda(K_0+M_0)+2c_0\Lambda M_0A_b+c_0\Lambda M_0(1+T^\mu)\ell^{\gamma\mu}H_b. \tag*{(38)}

The choice (34) of ℓ\ell gives the two conditions

2L0ℓ≤18,c0ΛM0(1+Tμ)ℓγμ≤116.(39)2L_0\ell\le\frac18,\qquad c_0\Lambda M_0(1+T^\mu)\ell^{\gamma\mu}\le\frac1{16}. \tag*{(39)}

The first one follows from ℓ≤1/(16L0)\ell\le1/(16L_0) and the second one from the third entry of the minimum. Moreover, since 0<ℓ≤10<\ell\le1 and γ≤1\gamma\le1 we have γμ≤μ\gamma\mu\le\mu and hence

ℓμ≤ℓγμ.\ell^\mu\le\ell^{\gamma\mu}.

Together with 1+Tμ≥11+T^\mu\ge1, the second condition in (39) therefore implies

c0ΛM0ℓμ≤c0ΛM0(1+Tμ)ℓγμ≤116.(40)c_0\Lambda M_0\ell^\mu\le c_0\Lambda M_0(1+T^\mu)\ell^{\gamma\mu}\le\frac1{16}. \tag*{(40)}

Notice that the terms (III) and (IV) carry no power of ℓ\ell in (38). Multiplying (38) by ℓμ\ell^\mu and treating each term separately,

ℓμ⋅(I)=2L0ℓμ+αAb=2L0ℓAb≤18Abby μ+α=1 and (39),\ell^\mu\cdot\mathrm{(I)}=2L_0\ell^{\mu+\alpha}A_b=2L_0\ell A_b\le\frac18A_b\qquad\text{by }\mu+\alpha=1\text{ and (39),}
ℓμ⋅(II)≤∥b0∥L1/αby ℓ≤1,\ell^\mu\cdot\mathrm{(II)}\le\|b_0\|_{L^{1/\alpha}}\qquad\text{by }\ell\le1,
ℓμ⋅(III)=c0Λℓμ(K0+M0)≤K0+M016M0by (40),\ell^\mu\cdot\mathrm{(III)}=c_0\Lambda\ell^\mu(K_0+M_0)\le\frac{K_0+M_0}{16M_0}\qquad\text{by (40),}
ℓμ⋅(IV)=2c0ΛM0ℓμAb≤18Abby (40),\ell^\mu\cdot\mathrm{(IV)}=2c_0\Lambda M_0\ell^\mu A_b\le\frac18A_b\qquad\text{by (40),}
ℓμ⋅(V)≤116ℓμHbby (39).\ell^\mu\cdot\mathrm{(V)}\le\frac1{16}\ell^\mu H_b\qquad\text{by (39).}

Adding up, and recalling the definitions of C1C_1 and Zb=Ab+ℓμHbZ_b=A_b+\ell^\mu H_b,

ℓμh≤18Ab+18Ab+116ℓμHb+C1≤14Zb+C1.\ell^\mu h \le\frac{1}{8}A_b+\frac{1}{8}A_b+\frac{1}{16}\ell^\mu H_b+C_1 \le\frac{1}{4}Z_b+C_1.

Inserting this into (37) gives Zb≤Za+12Zb+2C1Z_b\le Z_a+\frac{1}{2}Z_b+2C_1, that is

Zb+4C1≤2(Za+4C1).(41)Z_b+4C_1\le2(Z_a+4C_1). \tag*{(41)}

Now split [0,T′][0,T'] into at most nn intervals of length at most ℓ\ell and iterate (41): ZT′+4C1≤2n(Z0+4C1)Z_{T'}+4C_1\le2^n(Z_0+4C_1). Since Z0=∥η∥∞+ℓμ[η]μ≤∥η∥Cμ([−r0,0])Z_0=\|\eta\|_\infty+\ell^\mu[\eta]_\mu\le\|\eta\|_{C^\mu([-r_0,0])}, this is (35).

Finally, from (34),

1ℓ≤1+16L0+(16c0M0(1+Tμ)Λ)1/(γμ),n≤1+Tℓ.(42)\frac{1}{\ell}\le1+16L_0+\left(16c_0M_0(1+T^\mu)\Lambda\right)^{1/(\gamma\mu)},\qquad n\le1+\frac{T}{\ell}. \tag*{(42)}

Since μ=1−α\mu=1-\alpha and ∥x∥∞,[−r0,T′]+[x]μ,[−r0,T′]≤(1+ℓ−μ)(∥x∥∞,[−r0,T′]+ℓμ[x]μ,[−r0,T′])\|x\|_{\infty,[-r_0,T']}+[x]_{\mu,[-r_0,T']}\le(1+\ell^{-\mu})(\|x\|_{\infty,[-r_0,T']}+\ell^\mu[x]_{\mu,[-r_0,T']}), the bound (35) and n≤1+T/ℓn\le1+T/\ell give (36). □\square

Existence and uniqueness: deterministic case

We can now state and prove the existence and uniqueness result for the deterministic equation (32).

Theorem 5.6. Assume (H1)(H1) and (H2)(H2), let α∈(0,α0)\alpha\in(0,\alpha_0) with α0\alpha_0 as in (14), let τ:[0,T]→[0,r0]\tau:[0,T]\to[0,r_0] be γ\gamma-Hölder, let g:[0,T]→Rmg:[0,T]\to\mathbb{R}^m satisfy Λα(g)<∞\Lambda_\alpha(g)<\infty and let η∈C1−α([−r0,0];Rd)\eta\in C^{1-\alpha}([-r_0,0];\mathbb{R}^d). Then equation (32) has a unique solution

x∈C1−α([−r0,T];Rd)⊂W0α,∞(−r0,T;Rd),x\in C^{1-\alpha}([-r_0,T];\mathbb{R}^d)\subset W_0^{\alpha,\infty}(-r_0,T;\mathbb{R}^d),

and xx satisfies the bound (36) with T′=TT'=T.

Proof. By Proposition 5.5, every solution on any [0,T′][0,T'], T′≤TT'\le T, satisfies

∥x∥∞,[−r0,T′]≤N∗,[x]μ,[−r0,T′]≤R∗,\|x\|_{\infty,[-r_0,T']}\le N_*,\qquad[x]_{\mu,[-r_0,T']}\le R_*,

with N∗:=2n(∥η∥Cμ+4C1)N_*:=2^n(\|\eta\|_{C^\mu}+4C_1) and R∗:=ℓ−μN∗R_*:=\ell^{-\mu}N_*, which do not depend on T′T' or on the solution. Define

R:=2[(2L0+2c0ΛM0)N∗+∥b0∥L1/α+c0Λ(K0+M0)+116R∗],R:=2\left[(2L_0+2c_0\Lambda M_0)N_*+\|b_0\|_{L^{1/\alpha}}+c_0\Lambda(K_0+M_0)+\frac{1}{16}R_*\right],
N:=N∗+R,R‾:=R∗+R,N:=N_*+R,\qquad\overline{R}:=R_*+R,

and

ℓ∗:=min⁡{ℓ,18LN,(4c0Λ(5M0+M0Tμ+8MNTδμR‾δ))−1/(γμ)},\ell_*:=\min\left\{\ell,\frac{1}{8L_N},\left(4c_0\Lambda(5M_0+M_0T^\mu+8M_NT^{\delta\mu}\overline{R}^{\delta})\right)^{-1/(\gamma\mu)}\right\},

so that

Θ:=2LNℓ∗+c0Λℓ∗γμ(5M0+M0Tμ+8MNTδμR‾δ)≤12.(43)\Theta:=2L_N\ell_*+c_0\Lambda\ell_*^{\gamma\mu}(5M_0+M_0T^\mu+8M_NT^{\delta\mu}\overline{R}^{\delta})\le\frac{1}{2}. \tag*{(43)}

We first prove local existence. Let 0≤a<b≤T0\le a<b\le T with b−a≤ℓ∗b-a\le\ell_* and suppose that xx is a solution on [0,a][0,a]; if a=0a=0, we take x=ηx=\eta. Let

X:={y∈Cμ([−r0,b];Rd):y=x on [−r0,a], [y]μ,[a,b]≤R},d(y,y′)=[y−y′]μ,[a,b].\mathcal{X}:=\{y\in C^\mu([-r_0,b];\mathbb{R}^d):y=x\text{ on }[-r_0,a],\ [y]_{\mu,[a,b]}\le R\},\qquad d(y,y')=[y-y']_{\mu,[a,b]}.

The set X\mathcal{X} is non-empty, since it contains the extension of xx by the constant x(a)x(a) on [a,b][a,b]. Then it is easy to check that (X,d)(\mathcal{X},d) is a complete metric space, with underlying Banach space Cμ([−r0,b];Rd)C^{\mu}([-r_{0},b];\mathbb{R}^{d}) with the norm ∥y∥Cμ([−r0,b])=∥y∥∞,[−r0,b]+[y]μ,[−r0,b]\|y\|_{C^{\mu}([-r_{0},b])}=\|y\|_{\infty,[-r_{0},b]}+[y]_{\mu,[-r_{0},b]}. Notice also that every y∈Xy\in\mathcal{X} satisfies, by the a priori bounds above,

∥y∥∞,[−r0,b]≤N∗+R=N,[y]μ,[−r0,b]≤R∗+R=R‾.(44)\|y\|_{\infty,[-r_{0},b]}\le N_{*}+R=N,\qquad[y]_{\mu,[-r_{0},b]}\le R_{*}+R=\overline{R}. \tag*{(44)}

Define L(y):=x\mathcal{L}(y):=x on [−r0,a][-r_{0},a] and

L(y)(t):=x(a)+Fa(y)(t)+Ga(y)(t)\mathcal{L}(y)(t):=x(a)+F_{a}(y)(t)+G_{a}(y)(t)

for t∈[a,b]t\in[a,b].

Let y∈Xy\in\mathcal{X} and ℓ′:=b−a≤ℓ∗≤ℓ≤1\ell^{\prime}:=b-a\le\ell_{*}\le\ell\le1. We have the following bounds:

  • since y=xy=x on [−r0,a][-r_{0},a], ∣x∣≤N∗|x|\le N_{*} there, and ∣y(t)−y(a)∣≤R(t−a)μ|y(t)-y(a)|\le R(t-a)^{\mu} on [a,b][a,b],

∥y∥∞,[−r0,b]≤N∗+ℓ′μR;\|y\|_{\infty,[-r_{0},b]}\le N_{*}+\ell^{\prime\mu}R;
  • [y]μ,[−r0,b]≤R∗+R[y]_{\mu,[-r_{0},b]}\le R_{*}+R, as in (44);

  • the values of the delayed term yτy^{\tau} on [a,b][a,b] are values of yy on [−r0,b][-r_{0},b], so ∥yτ∥∞,[a,b]≤N∗+ℓ′μR\|y^{\tau}\|_{\infty,[a,b]}\le N_{*}+\ell^{\prime\mu}R, and by (15) ∣yτ(s)−yτ(u)∣≤Tμ(R∗+R)∣s−u∣γμ|y^{\tau}(s)-y^{\tau}(u)|\le T^{\mu}(R_{*}+R)|s-u|^{\gamma\mu} for u,s∈[a,b]u,s\in[a,b].

Inserting the three bounds above into (20) and (21) with f=yf=y gives

[L(y)]μ,[a,b]≤2L0ℓ′α(N∗+ℓ′μR)+∥b0∥L1/α+c0Λ(K0+M0)+2c0ΛM0(N∗+ℓ′μR)+c0ΛM0(1+Tμ)ℓ′γμ(R∗+R)=E0+ER,[\mathcal{L}(y)]_{\mu,[a,b]}\le2L_{0}\ell^{\prime\alpha}(N_{*}+\ell^{\prime\mu}R)+\|b_{0}\|_{L^{1/\alpha}}+c_{0}\Lambda(K_{0}+M_{0}) +2c_{0}\Lambda M_{0}(N_{*}+\ell^{\prime\mu}R)+c_{0}\Lambda M_{0}(1+T^{\mu})\ell^{\prime\gamma\mu}(R_{*}+R)=E_{0}+E_{R},

where

E0:=2L0ℓ′αN∗+∥b0∥L1/α+c0Λ(K0+M0)+2c0ΛM0N∗+c0ΛM0(1+Tμ)ℓ′γμR∗,E_{0}:=2L_{0}\ell^{\prime\alpha}N_{*}+\|b_{0}\|_{L^{1/\alpha}}+c_{0}\Lambda(K_{0}+M_{0})+2c_{0}\Lambda M_{0}N_{*}+c_{0}\Lambda M_{0}(1+T^{\mu})\ell^{\prime\gamma\mu}R_{*},
ER:=2L0ℓ′α+μR+2c0ΛM0ℓ′μR+c0ΛM0(1+Tμ)ℓ′γμR.E_{R}:=2L_{0}\ell^{\prime\alpha+\mu}R+2c_{0}\Lambda M_{0}\ell^{\prime\mu}R+c_{0}\Lambda M_{0}(1+T^{\mu})\ell^{\prime\gamma\mu}R.

For E0E_{0} we use ℓ′≤1\ell^{\prime}\le1 and the second condition in (39), which remains valid with ℓ′≤ℓ\ell^{\prime}\le\ell in place of ℓ\ell:

E0≤(2L0+2c0ΛM0)N∗+∥b0∥L1/α+c0Λ(K0+M0)+116R∗=12R.E_{0}\le(2L_{0}+2c_{0}\Lambda M_{0})N_{*}+\|b_{0}\|_{L^{1/\alpha}}+c_{0}\Lambda(K_{0}+M_{0})+\frac{1}{16}R_{*}=\frac{1}{2}R.

On the other hand, for ERE_{R} we use (39) and (40), again with ℓ′≤ℓ\ell^{\prime}\le\ell in place of ℓ\ell, term by term:

2L0ℓ′α+μR=2L0ℓ′R≤18R,2L_{0}\ell^{\prime\alpha+\mu}R=2L_{0}\ell^{\prime}R\le\frac{1}{8}R,
2c0ΛM0ℓ′μR≤18R,2c_{0}\Lambda M_{0}\ell^{\prime\mu}R\le\frac{1}{8}R,
c0ΛM0(1+Tμ)ℓ′γμR≤116R,c_{0}\Lambda M_{0}(1+T^{\mu})\ell^{\prime\gamma\mu}R\le\frac{1}{16}R,

so that ER≤(18+18+116)R=516RE_{R}\le\left(\frac{1}{8}+\frac{1}{8}+\frac{1}{16}\right)R=\frac{5}{16}R. Therefore

[L(y)]μ,[a,b]≤E0+ER≤12R+516R=1316R≤R.[\mathcal{L}(y)]_{\mu,[a,b]}\le E_{0}+E_{R}\le\frac{1}{2}R+\frac{5}{16}R=\frac{13}{16}R\le R.

Hence L(X)⊂X\mathcal{L}(\mathcal{X})\subset\mathcal{X}. By Proposition 5.4, which is applicable with the constants NN and R‾\overline{R} thanks to (44), and by (43),

d(L(y),L(y′))≤Θd(y,y′)≤12d(y,y′),y,y′∈X.d(\mathcal{L}(y),\mathcal{L}(y^{\prime}))\le\Theta d(y,y^{\prime})\le\frac{1}{2}d(y,y^{\prime}),\qquad y,y^{\prime}\in\mathcal{X}.

By the Banach fixed point theorem L\mathcal{L} has a fixed point y∈Xy \in\mathcal{X}. By (33) for xx at time aa and the additivity of the integrals, yy is a solution on [0,b][0,b].

Once local existence is proved, global existence follows by an iterative procedure. Starting from a=0a=0 and applying the local existence result successively on the intervals [kℓ∗,(k+1)ℓ∗∧T][k\ell_\ast,(k+1)\ell_\ast\wedge T] we obtain, after finitely many steps, a solution on [0,T][0,T]. Note that ℓ∗\ell_\ast is the same at every step, because the constants are uniform.

We finish by proving the uniqueness of the solution. Let x,x′x,x' be two solutions on [0,T][0,T] and let t0:=sup⁡{t∈[0,T]:x=x′ on [−r0,t]}t_0:=\sup\{t\in[0,T]:x=x' \text{ on }[-r_0,t]\}. By continuity, x=x′x=x' on [−r0,t0][-r_0,t_0]. Suppose t0<Tt_0<T. Both solutions satisfy the a priori bounds above, so by (33) and Proposition 5.4 with a=t0a=t_0 and b=(t0+ℓ∗)∧Tb=(t_0+\ell_\ast)\wedge T,

[x−x′]μ,[a,b]=[Fa(x)−Fa(x′)+Ga(x)−Ga(x′)]μ,[a,b]≤12[x−x′]μ,[a,b].[x-x']_{\mu,[a,b]}=[F_a(x)-F_a(x')+G_a(x)-G_a(x')]_{\mu,[a,b]}\leq\frac{1}{2}[x-x']_{\mu,[a,b]}.

Hence x−x′x-x' is constant on [a,b][a,b], and it vanishes at aa; so x=x′x=x' on [−r0,b][-r_0,b] with b>t0b>t_0, a contradiction. Therefore x=x′x=x' on [−r0,T][-r_0,T].

Proof of the main theorem

Let WHW^H be the fractional Brownian motion of § 3. We know that Λα(WH)<∞\Lambda_\alpha(W^H)<\infty a.s. for all α∈(1−H,12)\alpha\in(1-H,\frac{1}{2}), by Remark 2.4. For α∈(1−H,12)\alpha\in(1-H,\frac{1}{2}) we write Λα:=Λα(WH)\Lambda_\alpha:=\Lambda_\alpha(W^H).

The following integrability result is a consequence of Fernique’s theorem [3].

Theorem 6.1. For every α∈(1−H,12)\alpha\in(1-H,\frac{1}{2}), every θ∈(0,2)\theta\in(0,2) and every c>0c>0,

E[exp⁡(cΛαθ)]<∞,(45)\mathbb{E}\left[\exp\left(c\Lambda_\alpha^\theta\right)\right]<\infty, \tag*{(45)}

and in particular Λα∈⋂p≥1Lp(Ω)\Lambda_\alpha\in\bigcap_{p\geq1}L^p(\Omega).

Proof of Theorem 3.1. (a) Let Ω∗\Omega^* be the set of full probability on which (7) holds with Cτ(ω)<∞C_\tau(\omega)<\infty, Λα(WH(⋅,ω))<∞\Lambda_\alpha(W^H(\cdot,\omega))<\infty and η(⋅,ω)∈C1−α([−r0,0];Rd)\eta(\cdot,\omega)\in C^{1-\alpha}([-r_0,0];\mathbb{R}^d). For every ω∈Ω∗\omega\in\Omega^* the assumptions of Theorem 5.6 are satisfied with g=WH(⋅,ω)g=W^H(\cdot,\omega), τ=τ(⋅,ω)\tau=\tau(\cdot,\omega) and initial path η(⋅,ω)\eta(\cdot,\omega). Hence, for every ω∈Ω∗\omega\in\Omega^*, the deterministic equation (32) has a unique solution

X(⋅,ω)∈C1−α([−r0,T];Rd)⊂W0α,∞(−r0,T;Rd),X(\cdot,\omega)\in C^{1-\alpha}([-r_0,T];\mathbb{R}^d)\subset W_0^{\alpha,\infty}(-r_0,T;\mathbb{R}^d),

and this is precisely equation (2) for that ω\omega. This gives the existence of the solution and its uniqueness among the processes with trajectories in C1−α([−r0,T];Rd)C^{1-\alpha}([-r_0,T];\mathbb{R}^d).

(b) Let cˉ:=∥Cτ∥L∞(Ω)\bar{c}:=\|C_\tau\|_{L^\infty(\Omega)}, so that a.s. T(ω)=T1−γ+[τ(⋅,ω)]γ≤T1−γ+cˉ=:Tˉ\mathcal{T}(\omega)=T^{1-\gamma}+[\tau(\cdot,\omega)]_\gamma\leq T^{1-\gamma}+\bar{c}=:\bar{T}, a deterministic constant. Write μ=1−α\mu=1-\alpha. By (36), and using ℓ≤1\ell\leq1, 1+ℓ−μ≤1+ℓ−1≤e1/ℓ1+\ell^{-\mu}\leq1+\ell^{-1}\leq e^{1/\ell} and 2T/ℓ≤eT/ℓ2^{T/\ell}\leq e^{T/\ell},

∥X∥C1−α([−r0,T])≤2e(1+T)/ℓ(∥η∥C1−α([−r0,0])+4C1).\|X\|_{C^{1-\alpha}([-r_0,T])}\leq2e^{(1+T)/\ell}\left(\|\eta\|_{C^{1-\alpha}([-r_0,0])}+4C_1\right).

By (42) and T≤Tˉ\mathcal{T}\leq\bar{T},

1ℓ≤1+16L0+C2Λα1/(γμ),C2:=(16c0M0(1+Tˉμ))1/(γμ),\frac{1}{\ell}\leq1+16L_0+C_2\Lambda_\alpha^{1/(\gamma\mu)},\qquad C_2:=\left(16c_0M_0(1+\bar{T}^{\mu})\right)^{1/(\gamma\mu)},

where C2C_2 is deterministic. Therefore, almost surely,

∥X∥C1−α([−r0,T])≤C3exp⁡(C4Λα1/(γ(1−α)))(∥η∥C1−α([−r0,0])+4C1),(46)\|X\|_{C^{1-\alpha}([-r_0,T])}\leq C_3\exp\left(C_4\Lambda_\alpha^{1/(\gamma(1-\alpha))}\right)\left(\|\eta\|_{C^{1-\alpha}([-r_0,0])}+4C_1\right), \tag*{(46)}

with deterministic constants C3:=2e(1+T)(1+16L0)C_3 := 2e^{(1+T)(1+16L_0)} and C4:=(1+T)C2C_4 := (1 + T)C_2. The second condition in (9) says that

θ:=1γ(1−α)<2.\theta:= \frac{1}{\gamma(1-\alpha)} < 2.

Hence, by (45) and using the Cauchy–Schwarz inequality and the moment assumption on η\eta,

E[∥X∥C1−αp]≤C3p(E[e2pC4Λθ])1/2(E[(∥η∥C1−α+4C1)2p])1/2<∞.\mathbb{E}\left[\|X\|_{C^{1-\alpha}}^p\right] \le C_3^p\left(\mathbb{E}\left[e^{2pC_4\Lambda\theta}\right]\right)^{1/2}\left(\mathbb{E}\left[\left(\|\eta\|_{C^{1-\alpha}}+4C_1\right)^{2p}\right]\right)^{1/2}<\infty.

The bound for ∥X∥α,r0\|X\|_{\alpha,r_0} follows from (4).

References

  1. [1]B. Boukanjime, T. Caraballo, M. El Fatini and M. El Khalifi, Dynamics of a stochastic coronavirus (COVID-19) epidemic model with Markovian switching, Chaos Solitons Fractals 141 (2020), 110361.DOI
  2. [2]Á. Cartea and L. Sánchez-Betancourt, Optimal execution with stochastic delay, Finance Stoch. 27 (2023), 1–47.DOI
  3. [3]X. Fernique, Regularité des trajectoires des fonctions aléatoires gaussiennes, in École d’Été de Probabilités de Saint-Flour, IV-1974, Lecture Notes in Math. 480 (1975), 1–96.DOI
  4. [4]M. Ferrante and C. Rovira, Stochastic delay differential equations driven by fractional Brownian motion with Hurst parameter H > 1/2, Bernoulli 12 (2006), 85–100.
  5. [5]M. Ferrante and C. Rovira, Convergence of delay differential equations driven by fractional Brownian motion, J. Evol. Equ. 10 (2010), 761–783.arxiv.org/abs/0903.5498
  6. [6]M.J. Garrido-Atienza, A. Ogrowsky and B. Schmalfuß, Random differential equations with random delays, Stoch. Dyn. 11 (2011), 675–699.
  7. [7]Y. Hu and D. Nualart, Differential equations driven by Hölder continuous functions of order greater than 1/2, in Stochastic Analysis and Applications (Abel Symposium 2005), Springer, 2007, 399–413.DOI
  8. [8]V. Kolmanovskii and A. Myshkis, Introduction to the Theory and Applications of Functional Differential Equations, Springer, 1999.DOI
  9. [9]J. León and S. Tindel, Malliavin calculus for fractional delay equations, J. Theoret. Probab. 25 (2012), 854–889.arxiv.org/abs/0912.2180
  10. [10]T. Lyons, Differential equations driven by rough signals (I): An extension of an inequality of L.C. Young, Math. Res. Lett. 1 (1994), 451–464.DOI
  11. [11]X. Mao, Stochastic Differential Equations and Applications, Horwood, 1997.
  12. [12]X. Mao, Y. Shen and C. Yuan, Almost surely asymptotic stability of neutral stochastic differential delay equations with Markovian switching, Stochastic Process. Appl. 118 (2008), 1385–1406.DOI
  13. [13]M. Martcheva, N. Tuncer and C. N. Ngonghala, Mathematical models of epidemics with infection time, AppliedMath 5 (2025), no. 2, 47.
  14. [14]S.-E.A. Mohammed, Stochastic Functional Differential Equations, Research Notes in Mathematics 99, Pitman, 1984.
  15. [15]S.-E.A. Mohammed, Stochastic differential systems with memory: theory, examples and applications, in Stochastic Analysis and Related Topics VI, Progr. Probab. 42, Birkhäuser, 1998, 1–77.DOI
  16. [16]A. Neuenkirch, I. Nourdin and S. Tindel, Delay equations driven by rough paths, Electron. J. Probab. 13 (2008), 2031–2068.arxiv.org/abs/0711.2633
  17. [17]D. Nualart and A. Răşcanu, Differential equations driven by fractional Brownian motion, Collect. Math. 53 (2002), 55–81.
  18. [18]J. Touboul, Limits and dynamics of stochastic neuronal networks with random heterogeneous delays, J. Stat. Phys. 149 (2012), 569–597.DOI
  19. [19]L.C. Young, An inequality of the Hölder type connected with Stieltjes integration, Acta Math. 67 (1936), 251–282.DOI
  20. [20]M. Zähle, Integration with respect to fractal functions and stochastic calculus I, Prob. Theory Relat. Fields 111 (1998), 333–374.DOI

Paper details

vlnc:W_y3u93enq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents