Abstract
We study the asymptotic behavior of weighted power variations of fractional Brownian motion in Brownian time \(Z_t:= X_{Y_t},t \geqslant 0\), where X is a fractional Brownian motion and Y is an independent Brownian motion.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Our aim in this paper is to study the asymptotic behavior of weighted power variations of the so-called fractional Brownian motion in Brownian time defined as
where X is a two-sided fractional Brownian motion, with Hurst parameter \(H \in (0,1)\), and Y is a standard (one-sided) Brownian motion independent of X. It is a self-similar process (of order H / 2) with stationary increments, which is not Gaussian. When \(H=1/2\), one recovers the celebrated iterated Brownian motion.
In the present paper we follow and we are inspired by the previous papers [2, 4, 5, 9], and our work may be seen as a natural follow-up of [4, 9].
Let \(f:\mathbb {R}\rightarrow \mathbb {R}\) be a function belonging to \(C_b^\infty \), the class of those functions that are \(C^\infty \) and bounded together with their derivatives. Then, for any \(t\geqslant 0\) and any integer \(p\geqslant 1\), the weighted p-variation of Z is defined as
After proper normalization, we may expect the convergence (in some sense) to a non-degenerate limit (to be determined) of
for some \(\kappa \) to be discovered. Due to the fact that one cannot separate X from Y inside Z in the definition of \(S_n^{(p)}\), working directly with (1.1) seems to be a difficult task (see also [3, Problem 5.1]). This is why, following an idea introduced by Khoshnevisan and Lewis [2] in a study of the case \(H=1/2\), we will rather analyze \(S_n^{(p)}\) by means of certain stopping times for Y. The idea is: by stopping Y as it crosses certain levels, and by sampling Z at these times, one can effectively separate X from Y. To be more specific, let us introduce the following collection of stopping times (with respect to the natural filtration of Y), noted
which are in turn expressed in terms of the subsequent hitting times of a dyadic grid cast on the real axis. More precisely, let \(\mathscr {D}_n= \{j2^{-n/2}:\,j\in \mathbb {Z}\}\), \(n\geqslant 0\), be the dyadic partition (of \(\mathbb {R}\)) of order n / 2. For every \(n\geqslant 0\), the stopping times \(T_{k,n}\), appearing in (1.2), are given by the following recursive definition: \(T_{0,n}= 0\), and
Note that the definition of \(T_{k,n}\), and therefore of \(\mathscr {T}_n\), only involves the one-sided Brownian motion Y, and that, for every \(n\geqslant 0\), the discrete stochastic process
defines a simple and symmetric random walk over \(\mathscr {D}_n\). As shown in [2], as n tends to infinity the collection \(\{T_{k,n}:\,1\leqslant k \leqslant 2^nt\}\) approximates the common dyadic partition \(\{k2^{-n}:\,1\leqslant k \leqslant 2^nt\}\) of order n of the time interval [0, t] (see [2, Lemma 2.2] for a precise statement). Based on this fact, one can introduce the counterpart of (1.1) based on \(\mathscr {T}_n\), namely,
for some \(\tilde{\kappa }>0\) to be discovered and with \(\mu _p:= E[N^p]\), where \(N \sim \mathcal {N}(0,1)\). At this stage, it is worthwhile noting that we are dealing with symmetric weighted p-variation of Z, and symmetry will play an important role in our analysis as we will see in Lemma 3.1.
In the particular case where \(H=\frac{1}{2}\), that is when Z is the iterated Brownian motion, the asymptotic behavior of \(\tilde{S}_n^{(p)}(\cdot ) \) has been studied in [4]. In fact, one can deduce the following two finite-dimensional distributions (f.d.d.) convergences in law from [4, Theorem 1.2].
-
1)
For \(f\in C_b^2\) and for any integer \(r \geqslant 1\), we have
where \(L_t^s(Y)\) stands for the local time of Y before time t at level s, W is a two-sided Brownian motion independent of (X, Y) and \(\int _{-\infty }^{+\infty } f(X_s)L_t^s(Y)dW_s\) is the Wiener–Itô integral of \(f(X_{\cdot })L^{\cdot }_t(Y)\) with respect to W.
-
2)
For \(f\in C_b^2\) and for any integer \(r \geqslant 2\), we have
where for all \(t \in \mathbb {R}\), \(\int _0^t f(X_s)d^{\circ }X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in probability of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \), with \(W_{n}^{(1)}(f,t)\) defined in (3.3), W is a two-sided Brownian motion independent of (X, Y) and for \(u \in \mathbb {R}\), \(\int _0^{u} f(X_s)dW_s\) is the Wiener–Itô integral of f(X) with respect to W defined in (5.16).
A natural follow-up of (1.3) and (1.4) is to study the asymptotic behavior of \(\tilde{S}_n^{(p)}(\cdot )\) when \(H \ne \frac{1}{2}\). In fact, the following more general result is our main finding in the present paper.
Theorem 1.1
Let \(f: \mathbb {R}\rightarrow \mathbb {R}\) be a function belonging to \(C_b^{\infty }\) and let W denote a two-sided Brownian motion independent of (X, Y).
-
(1)
For \(H> \frac{1}{6}\), we have
$$\begin{aligned} \sum _{k=0}^{\lfloor 2^n t \rfloor -1} \frac{1}{2}(f(Z_{T_{k,n}})+f(Z_{T_{k+1,n}})) (Z_{T_{k+1,n}}-Z_{T_{k,n}})\underset{n\rightarrow \infty }{\overset{P}{\longrightarrow }} \int _0^{Y_t} f(X_s)d^{\circ }X_s, \nonumber \\ \end{aligned}$$(1.5)where for all \(t \in \mathbb {R}\), \(\int _0^t f(X_s)d^{\circ }X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in probability of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \), with \(W_{n}^{(1)}(f,t)\) defined in (3.3).
For \(H= \frac{1}{6}\), we have
$$\begin{aligned} \sum _{k=0}^{\lfloor 2^n t \rfloor -1} \frac{1}{2}(f(Z_{T_{k,n}})+f(Z_{T_{k+1,n}})) (Z_{T_{k+1,n}}-Z_{T_{k,n}})\underset{n\rightarrow \infty }{\overset{law}{\longrightarrow }} \int _0^{Y_t} f(X_s)d^{*}X_s,\nonumber \\ \end{aligned}$$(1.6)where for all \(t \in \mathbb {R}\), \(\int _0^t f(X_s)d^{*}X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in law of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \).
-
(2)
For \(\frac{1}{6}< H < \frac{1}{2}\) and for any integer \(r\geqslant 2\), we have
where for \(u \in \mathbb {R}\), \(\int _0^{u} f(X_s)dW_s\) is the Wiener–Itô integral of f(X) with respect to W defined in (5.16), \(\beta _{2r-1} = \sqrt{\sum _{l=2}^r\kappa ^2_{r,l}\,\alpha ^2_{2l-1}}\), with \(\alpha _{2l-1}\) defined in (2.18) and \(\kappa _{r,l}\) defined in (3.4).
-
(3)
Fix a time \(t \geqslant 0\), for \(H>\frac{1}{2}\) and for any integer \(r \geqslant 1\), we have
where for all \(t \in \mathbb {R}\), \(\int _0^t f(X_s)d^{\circ }X_s\) is defined as in (1.5).
-
(4)
For \(\frac{1}{4} < H \leqslant \frac{1}{2}\) and for any integer \(r\geqslant 1\), we have
where \(\int _{-\infty }^{+\infty } f(X_s)L_t^s(Y)dW_s\) is the Wiener–Itô integral of \(f(X_{\cdot })L^{\cdot }_t(Y)\) with respect to W, \(\gamma _{2r} := \sqrt{\sum _{a=1}^rb_{2r,a}^2\,\alpha ^2_{2a}}\), with \(\alpha _{2a}\) defined in (2.18) and \(b_{2r,a}\) defined in (7.1).
Theorem 1.1 is also a natural follow-up of [9, Corollary 1.2] where we have studied the asymptotic behavior of the power variations of the fractional Brownian motion in Brownian time. In fact, taking f equal to 1 in (1.8), we deduce the following Corollary.
Corollary 1.2
Assume that \(H>\frac{1}{2}\), for any \(t\geqslant 0\) and any integer \(r \geqslant 1\), we have
thus, we understand the asymptotic behavior of the signed power variations of odd order of the fractional Brownian motion in Brownian time, in the case \(H>\frac{1}{2}\), which was missing in the first point in [9, Corollary 1.2].
Remark 1.3
-
1.
For \(H= \frac{1}{6}\), it has been proved in [8, (3.17)] that
with W a standard two-sided Brownian motion independent of the pair (X, Y) and \(\kappa _3\simeq 2.322\). Thus, (1.7) continues to hold for \(H=\frac{1}{6}\) and \(r=2\).
-
2.
In the particular case where \(H=1/2\) (that is, when Z is the iterated Brownian motion), we emphasize that the fourth point of Theorem 1.1 allows one to recover (1.3). In fact, since \(H=\frac{1}{2}\), then, for any integer \(a\geqslant 1\), by (2.18) and its related explanation, \(\alpha ^2_{2a} = (2a)!\). So, using the decomposition (7.1) and (2.3), the reader can verify that \(\sqrt{\mu _{4r} - \mu _{2r}^2}\) appearing in (1.3) is equal to \(\gamma _{2r}\) appearing in (1.9).
-
3.
The limit process in (1.4) is \(\bigg ( \int _0^{Y_t} f(X_s)(\mu _{2r} d^{\circ }X_s + \sqrt{\mu _{4r-2} - \mu _{2r}^2}\,dW_s \bigg )_{t \geqslant 0} \). Observe that \(\mu _{2r}=E[N^{2r}]= \frac{(2r)!}{r!2^r}\) and since \(H=\frac{1}{2}\), then, for any integer \(l\geqslant 1\), by (2.18) and its related explanation, \(\alpha ^2_{2l-1} = (2l-1)!\). So, using the decomposition (3.4) and (2.3), the reader can verify that \(\sqrt{\mu _{4r-2} - \mu _{2r}^2}\) is equal to \(\beta _{2r-1}\) appearing in (1.7). We deduce that the limit process in (1.4) is \(\bigg ( \frac{(2r)!}{r!2^r}\int _0^{Y_t} f(X_s)d^{\circ }X_s + \beta _{2r-1}\int _0^{Y_t} f(X_s)dW_s \bigg )_{t \geqslant 0} \). Thus, one can say that, for any integer \(r\geqslant 2\), the limit of the weighted \((2r-1)\)-variation of Z for \(H=\frac{1}{2}\) is intermediate between the limit of the weighted \((2r-1)\)-variation of Z for \(H> \frac{1}{2}\) and the limit of the weighted \((2r-1)\)-variation of Z for \(\frac{1}{6}<H<\frac{1}{2}\).
A brief outline of the paper is as follows. In Sect. 2, we give the preliminaries to the proof of Theorem 1.1. In Sect. 3, we start the preparation to our proof. In Sect. 4, we prove (1.5) and (1.6). In Sects. 5, 6 and 7 we prove (1.7), (1.8) and (1.9). Finally, in Sect. 8, we give the proof of a technical lemma.
2 Preliminaries
2.1 Elements of Malliavin Calculus
In this section, we gather some elements of Malliavin calculus we shall need in the sequel. The reader in referred to [6] for details and any unexplained result.
We continue to denote by \(X = (X_{t})_{t \in \mathbb {R}}\) a two-sided fractional Brownian motion with Hurst parameter \( H \in (0,1).\) That is, X is a zero mean Gaussian process, defined on a complete probability space \((\Omega , \mathscr {A}, P)\), with covariance function,
We suppose that \(\mathscr {A}\) is the \(\sigma \)-field generated by X. For all \(n \in \mathbb {N}^*\), we let \(\mathscr {E}_n\) be the set of step functions on \([-n,n]\), and \(\displaystyle {\mathscr {E}:= \cup _n \mathscr {E}_n}\). Set \(\varepsilon _t = \mathbf 1 _{[0,t]}\) (resp. \(\mathbf 1 _{[t,0]}\)) if \(t \geqslant 0\) (resp. \(t < 0\)). Let \(\mathscr {H}\) be the Hilbert space defined as the closure of \(\mathscr {E}\) with respect to the inner product
The mapping \(\varepsilon _t \mapsto X_{t}\) can be extended to an isometry between \(\mathscr {H}\) and the Gaussian space \(\mathbb {H}_{1}\) associated with X. We will denote this isometry by \(\varphi \mapsto X(\varphi ).\)
Let \(\mathscr {F}\) be the set of all smooth cylindrical random variables, i.e., of the form
where \(l \in \mathbb {N}^*\), \(\phi : \mathbb {R}^{l}\rightarrow \mathbb {R}\) is a \(C^{\infty }\)-function such that f and its partial derivatives have at most polynomial growth, and \( t_{1}< . . . <t_{l}\) are some real numbers. The derivative of F with respect to X is the element of \(L^{2}(\Omega , \mathscr {H})\) defined by
In particular \(D_{s}X_{t} = \varepsilon _t(s)\). For any integer \(k \geqslant 1\), we denote by \(\mathbb {D}^{k,2}\) the closure of \(\mathscr {F}\) with respect to the norm
The Malliavin derivative D satisfies the chain rule. If \(\varphi : \mathbb {R}^{n} \rightarrow \mathbb {R}\) is \(C_{b}^{1}\) and if \(F_1,\ldots ,F_n\) are in \(\mathbb {D}^{1,2}\), then \(\varphi (F_{1}, \ldots ,F_{n}) \in \mathbb {D}^{1,2}\) and we have
We have the following Leibniz formula, whose proof is straightforward by induction on q. Let \(\varphi , \psi \in C_{b}^{q}\) \((q\geqslant 1)\), and fix \(0 \leqslant u<v \) and \(0\leqslant s<t .\) Then \(( \varphi (X_{s})+ \varphi (X_{t}) ) \big (\psi (X_{u})+ \psi (X_{v}) ) \in \mathbb {D}^{q,2}\) and
where \(\tilde{\otimes }\) stands for the symmetric tensor product and \( \varphi ^{(l)}\) (resp. \( \psi ^{(q-l)}\)) means that \(\varphi \) is differentiated l times (resp. \(\psi \) is differentiated \(q-l\) times). A similar statement holds fo \( u<v \leqslant 0 \) and \( s<t \leqslant 0\).
If a random element \(u \in L^{2}(\Omega , \mathscr {H})\) belongs to the domain of the divergence operator, that is, if it satisfies
then I(u) is defined by the duality relationship
for every \(F \in \mathbb {D}^{1,2}.\)
For every \(n\geqslant 1\), let \(\mathbb {H}_{n}\) be the nth Wiener chaos of X, that is, the closed linear subspace of \( L^{2}(\Omega , \mathscr {A},P)\) generated by the random variables \(\lbrace H_{n}(X(h)), h \in \mathscr {H}, \Vert h\Vert _{\mathscr {H}}=1 \rbrace ,\) where \(H_{n}\) is the nth Hermite polynomial. Recall that \(H_0=0\), \(H_p(x)= (-1)^p \exp (\frac{x^2}{2})\frac{d^p}{dx^p}\exp (-\frac{x^2}{2})\) for \(p\geqslant 1\), and that
for jointly Gaussian M, N and integers \(p,q \geqslant 1\). The mapping
provides a linear isometry between the symmetric tensor product \(\mathscr {H}^{\odot n}\) and \(\mathbb {H}_{n}\). For \(H =\frac{1}{2}\), \(I_{n}\) coincides with the multiple Wiener–Itô integral of order n. The following duality formula holds
for any element \( h\in \mathscr {H}^{\odot n}\) and any random variable \(F \in \mathbb {D}^{n,2}.\)
Let \(\lbrace e_{k}, k \geqslant 1\rbrace \) be a complete orthonormal system in \(\mathscr {H}.\) Given \(f \in \mathscr {H}^{\odot n}\) and \(g \in \mathscr {H}^{\odot m},\) for every \(r= 0, \ldots ,n\wedge m,\) the contraction of f and g of order r is the element of \( \mathscr {H}^{\otimes (n+m-2r)}\) defined by
Finally, we recall the following product formula: If \(f\in \mathscr {H}^{\odot n}\) and \(g\in \mathscr {H}^{\odot m}\), then
2.2 Some Technical Results
For all \(k\in \mathbb {Z}\) and \(n \in \mathbb {N}\), we write
The following lemma will play a pivotal role in the proof of Theorem 1.1. The reader can find an original version of this lemma in [5, Lemma 5, Lemma 6].
Lemma 2.1
-
1.
If \(H\leqslant \frac{1}{2}\), for all integer \(q\geqslant 1\), for all \(j \in \mathbb {N}\) and \(u\in \mathbb {R}\),
$$\begin{aligned} \left| \left\langle \varepsilon _u^{\otimes q}, \delta _{(j+1)2^{-n/2}}^{\otimes q} \right\rangle _{\mathscr {H}^{\otimes q}} \right|\leqslant & {} 2^{-nqH}. \end{aligned}$$(2.7) -
2.
If \(H > \frac{1}{2}\), for all integer \(q\geqslant 1\), for all \(t \in \mathbb {R}_+\) and \(j, j' \in \{0, \ldots , \lfloor 2^{n/2}t\rfloor -1\}\),
$$\begin{aligned} \left| \left\langle \varepsilon _{j2^{-n/2}}^{\otimes q}, \delta _{(j'+1)2^{-n/2}}^{\otimes q}\right\rangle _{\mathscr {H}^{\otimes q}} \right|\leqslant & {} 2^q 2^{-\frac{nq}{2}}t^{(2H-1)q}, \end{aligned}$$(2.8)$$\begin{aligned} \left| \left\langle \varepsilon _{(j+1)2^{-n/2}}^{\otimes q}, \delta _{(j'+1)2^{-n/2}}^{\otimes q} \right\rangle _{\mathscr {H}^{\otimes q}} \right|\leqslant & {} 2^q 2^{-\frac{nq}{2}}t^{(2H-1)q}. \end{aligned}$$(2.9) -
3.
For all integers \(r,n \geqslant 1\) and \(t\in \mathbb {R}_+\), and with \(C_{H,r}\) a constant depending only on H and r (but independent of t and n),
-
(a)
if \(H< 1-\frac{1}{2r}\),
$$\begin{aligned} \sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\left| \left\langle \delta _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\right\rangle _\mathscr {H}\right| ^r\leqslant & {} C_{H,r} \, t\, 2^{n\left( \frac{1}{2}-rH\right) } \end{aligned}$$(2.10) -
(b)
if \(H=1 - \frac{1}{2r}\),
$$\begin{aligned} \sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\left| \left\langle \delta _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\right\rangle _\mathscr {H}\right| ^r\leqslant & {} C_{H,r}\,2^{n\left( \frac{1}{2}-rH\right) }(t(1+n) +t^2)\nonumber \\ \end{aligned}$$(2.11) -
(c)
if \(H> 1 - \frac{1}{2r}\),
$$\begin{aligned}&\sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\left| \left\langle \delta _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\right\rangle _\mathscr {H}\right| ^r \leqslant C_{H,r}\big (t\, 2^{n\left( \frac{1}{2} -rH\right) }\nonumber \\&\quad + \,t^{2-(2-2H)r}\,2^{n(1-r)}\big ). \end{aligned}$$(2.12)
-
(a)
-
4.
For \(H\in (0,1)\). For all integer \(n \geqslant 1\) and \(t\in \mathbb {R}_+\),
$$\begin{aligned} \sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\left| \left\langle \varepsilon _{k2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\right\rangle _\mathscr {H}\right|\leqslant & {} 2^{\frac{n}{2} +1} t^{2H + 1}, \end{aligned}$$(2.13)$$\begin{aligned} \sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\left| \left\langle \varepsilon _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\right\rangle _\mathscr {H}\right|\leqslant & {} 2^{\frac{n}{2}+1} t^{2H + 1}. \end{aligned}$$(2.14)
Proof
The proof, which is quite long and technical, is postponed in Sect. 8. \(\square \)
It has been mentioned in [2] that \(\{\Vert Y_{T_{\lfloor 2^n t\rfloor ,n}}\Vert _{4} : n \geqslant 0\}\) is a bounded sequence. More generally, we have the following result.
Lemma 2.2
For any integer \(k \geqslant 1\), \(\{\Vert Y_{T_{\lfloor 2^n t\rfloor ,n}}\Vert _{2k} : n \geqslant 0\}\) is a bounded sequence.
Proof
Recall from the introduction that \(\{Y_{T_{k,n}} : k \geqslant 0\}\) is a simple and symmetric random walk on \(\mathscr {D}_n\), and observe that \(Y_{T_{\lfloor 2^n t\rfloor ,n}} = \sum _{l=0}^{\lfloor 2^n t\rfloor -1} (Y_{T_{l+1,n}} - Y_{T_{l,n}})\). So, we have
where \(\forall i\in \{1, \ldots ,m\}\) \(a_i\) is an even integer, \( \forall m \in \{1, \ldots , k\}\) \(C_{a_1,\ldots ,a_m} \geqslant 0\), is some combinatorial constant whose explicit value is immaterial here. Now observe that the quantity in (2.15) is equal to
so, since \(1\leqslant m \leqslant k\), we deduce that \(\big \{ E\big [\big (Y_{T_{\lfloor 2^n t\rfloor ,n}}\big )^{2k}\big ] : n\geqslant 0 \big \}\) is a bounded sequence, which proves the lemma. \(\square \)
Also, in order to prove the fourth point of Theorem 1.1, we will need estimates on the local time of Y taken from [2], that we collect in the following statement.
Proposition 2.3
-
1.
For every \(x\in \mathbb {R}\), \(p \in \mathbb {N}^{*}\) and \(t > 0\), we have
$$\begin{aligned} E[(L_t^{x}(Y))^p] \leqslant 2\,E[(L_1^0(Y))^p]\,t^{p/2}\, {\mathrm{exp}}\left( -\frac{x^2}{2t}\right) . \end{aligned}$$ -
2.
There exists a positive constant \(\mu \) such that, for every \(a,b\in \mathbb {R}\) with \(ab\geqslant 0\) and \(t > 0\),
$$\begin{aligned} E[|L_t^b(Y)-L_t^a(Y)|^2]^{1/2} \leqslant \mu \,\sqrt{|b-a|}\,t^{1/4}\, {\mathrm{exp}}\left( -\frac{a^2}{4t}\right) . \end{aligned}$$ -
3.
There exists a positive random variable \(K\in L^8\) such that, for every \(j\in \mathbb {Z}\), every \(n\geqslant 0\) and every \(t > 0\), one has that
$$\begin{aligned} \left| {\mathcal {L}}_{j,n}(t)-L_t^{j2^{-n/2}}(Y)\right| \leqslant 2Kn2^{-n/4}\sqrt{L_t^{j2^{-n/2}}(Y)}, \end{aligned}$$where \({\mathcal {L}}_{j,n}(t)=2^{-n/2}(U_{j,n}(t)+D_{j,n}(t))\).
2.3 Notation
Throughout all the forthcoming proofs, we shall use the following notation. For all \(t\in \mathbb {R}\) and \(n \in \mathbb {N}\), we define \(X^{(n)}_t := 2^{\frac{nH}{2}}X_{t2^{-\frac{n}{2}}}\). For all \(k \in \mathbb {Z}\) and \(H \in (0,1)\), we write
it is clear that \(\rho (-k)=\rho (k)\). Observe that, by (2.1), we have
If \(H\leqslant \frac{1}{2}\), for all \(r \in \mathbb {N}^*\), we define
Note that \(\sum _{a\in \mathbb {Z}}|\rho (a)|^{r} < \infty \) if and only if \(H < 1- 1/(2r)\), which is satisfied for all \(r \geqslant 1 \) if we suppose that \(H \leqslant 1/2\) (in the case \(H=1/2\), we have \(\rho (0)=1\) and \(\rho (a) =0\) for all \(a \ne 0\). So, for any \(r \in \mathbb {N}^*\), we have \(\sum _{a\in \mathbb {Z}}|\rho (a)|^r=1\)).
For simplicity, throughout the paper we remove the subscript \(\mathscr {H}\) in the inner product defined in (2.1), that is, we write \(\langle \,\, ;\,\, \rangle \) instead of \(\langle \,\, ;\,\, \rangle _\mathscr {H}\).
For any sufficiently smooth function \(f : \mathbb {R}\rightarrow \mathbb {R}\), the notation \(\partial ^{l}f\) means that f is differentiated l times. We denote for any \(j \in \mathbb {Z}\) , \(\Delta _{j,n} f(X):= \frac{1}{2}(f(X_{j2^{-n/2}})+ f(X_{(j+1)2^{-n/2}})).\)
In the proofs contained in this paper, C shall denote a positive, finite constant that may change value from line to line.
3 Preparation to the proof of Theorem 1.1
3.1 A Key Algebraic Lemma
For each integer \(n\geqslant 1\), \(k\in \mathbb {Z}\) and real number \(t\geqslant 0\), let \(U_{j,n}(t)\) (resp. \(D_{j,n}(t)\)) denote the number of upcrossings (resp. downcrossings) of the interval \([j2^{-n/2},(j+1)2^{-n/2}]\) within the first \(\lfloor 2^n t\rfloor \) steps of the random walk \(\{Y_{T_{k,n}}\}_{k\geqslant 0}\), that is,
The following lemma taken from [2, Lemma 2.4] is going to be the key when studying the asymptotic behavior of the weighted power variation \(V_n^{(r)}(f,t)\) of order \(r\geqslant 1\), defined as:
where \(\mu _r:= E[N^r]\), with \(N \sim \mathcal {N}(0,1)\). Its main feature is to separate X from Y, thus providing a representation of \(V_n^{(r)}(f,t)\) which is amenable to analysis.
Lemma 3.1
Fix \(f\in C^\infty _b\), \(t\geqslant 0\) and \(r\in \mathbb {N}^{*}\). Then
3.2 Transforming the Weighted Power Variations of Odd Order
By [2, Lemma 2.5], one has
where \(j^*(n,t)=2^{n/2}Y_{T_{\lfloor 2^n t\rfloor ,n}}\). As a consequence, \(V_n^{(2r-1)}(f,t)\) is equal to
where \(X^+_t := X_t\) for \(t\geqslant 0\), \(X^-_{-t} :=X_t\) for \(t<0\), \(X^{n,+}_{t} := 2^{\frac{nH}{2}}X^+_{2^{-\frac{n}{2}}t}\) for \( t \geqslant 0\) and \(X^{n,-}_{-t} := 2^{\frac{nH}{2}}X^-_{2^{-\frac{n}{2}}(-t)}\) for \(t < 0\).
Let us now introduce the following sequence of processes \(W_{\pm ,n}^{(2r-1)}\), in which \(H_p\) stands for the pth Hermite polynomial (\(H_1(x) =x\), \(H_2(x) =x^2 -1\), etc.):
We then have, using the decomposition
(with \(\kappa _{r,r}=1\), and \(\kappa _{r,1}=\frac{(2r)!}{r!2^r}=E[N^{2r}]\), with \(N\sim \mathcal {N}(0,1)\). If interested, the reader can find the explicit value of \(\kappa _{r,i}\), for \(1<i<r\), e.g., in [9, Corollary 1.2]),
4 Proofs of (1.5) and (1.6)
4.1 Proof of (1.5)
In [8, Theorem 2.1], we have proved that for \(H > \frac{1}{6}\) and \(f \in C_b^{\infty }\), the following change-of-variable formula holds true
where F is a primitive of f and \(\int _0^t f(Z_s)d^\circ Z_s\) is the limit in probability of \(2^{-\frac{nH}{2}}V_n^{(1)}(f,t)\) as \(n \rightarrow \infty \), with \(V_n^{(1)}(f,t)\) defined in (3.1). On the other hand, it has been proved in [5, Theorem 4] (see also [10, Theorem 1.3] for an extension of this formula to the bi-dimensional case) that for all \(t\in \mathbb {R}\), the following change-of-variable formula holds true for \(H > \frac{1}{6}\)
where \(\int _0^t f(X_s)d^{\circ }X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in probability of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \), with \(W_{n}^{(1)}(f,t)\) defined in (3.3). Thanks to (4.2), we deduce that
by combining this last equality with (4.1), we get \(\int _0^t f(Z_s)d^\circ Z_s = \int _0^{Y_t} f(X_s)d^{\circ }X_s\). So, we deduce that, for \(H>\frac{1}{6}\),
thus (1.5) holds true.
4.2 Proof of (1.6)
In [8, Theorem 2.1], we have proved that for \(H = \frac{1}{6}\) and \(f \in C_b^{\infty }\), the following change-of-variable formula holds true
where F is a primitive of f, W is a standard two-sided Brownian motion independent of the pair (X, Y), \(\kappa _3\simeq 2.322\) and \(\int _0^{t}f(Z_s)d^{\circ }Z_s\) is the limit in law of \(2^{-\frac{nH}{2}}V_n^{(1)}(f,t)\) as \(n\rightarrow \infty \), with \(V_n^{(1)}(f,t)\) defined in (3.1). On the other hand, it has been proved in (2.19) in [7] that for all \(t\in \mathbb {R}\), the following change-of-variable formula holds true for \(H = \frac{1}{6}\)
where \(\kappa _3\) and W are the same as in (4.3), \(\int _0^t f(X_s)d^{*}X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in law of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \), with \(W_{n}^{(1)}(f,t)\) defined in (3.3). Thanks to (4.4), we deduce that
By combining this last equality with (4.3), we get \(\int _0^t f(Z_s)d^\circ Z_s \overset{law}{=} \int _0^{Y_t} f(X_s)d^{*}X_s\). So, we deduce that, for \(H=\frac{1}{6}\),
thus (1.6) holds true.
5 Proof of (1.7)
Thanks to (3.1) and (3.5), for any integer \(r\geqslant 2\), we have
The proof of (1.7) will be done in several steps.
5.1 Step 1: Limit of \(2^{-n/4}\sum _{l=2}^{r}\kappa _{r,l}W_{n}^{(2l-1)}(f,t)\)
Observe that, by (3.4), we have
We have the following proposition:
Proposition 5.1
If \(H \in ( \frac{1}{6}, \frac{1}{2})\), if \(r\geqslant 2\) then, for any \(f \in C_b^{\infty }\),
where \(\beta _{2r-1} = \sqrt{\sum _{l=2}^r\kappa ^2_{r,l}\,\alpha ^2_{2l-1}}\), \(\alpha _{2l-1}\) is given by (2.18), \(W^{+}_t=W_t\) if \(t>0\) and \(W^{-}_t=W_{-t}\) if \(t<0\), with W a two-sided Brownian motion independent of (X, Y), and where \(\int _0^t f(X^{\pm }_s)dW^{\pm }_s\) must be understood in the Wiener–Itô sense.
Proof
For all \(t\geqslant 0\), we define \(F_{\pm ,n}^{(2r-1)}(f,t):= 2^{-\frac{n}{4}}\sum _{l=2}^{r} \kappa _{r,l} W_{\pm ,n}^{(2l-1)}(f,t)\). In what follows we may study separately the finite-dimensional distributions convergence in law of \(\big (X, F_{+,n}^{(2r-1)}(f, \cdot ),\) \( F_{-,n}^{(2r-1)}(f,\cdot )\big )\) when n is even and when n is odd. For the sake of simplicity, we will only consider the even case, the analysis when n is odd being mutatis mutandis the same. So, assume that n is even and let m be another even integer such that \(n \geqslant m \geqslant 0\). We shall apply a coarse gaining argument. We have
Observe that \(2^{\frac{n-m}{2}}\) is an integer precisely because we have assumed that n and m are even numbers. We have
where
Here is a sketch of what remains to be done in order to complete the proof of (5.2). Firstly, we will prove (a) the f.d.d. convergence in law of \((X, A^+_{n,m}, A^-_{n,m})\) to \((X, \beta _{2r-1}\int _0^{\cdot }f(X^+_s)dW^+_s,\) \( \beta _{2r-1}\int _0^{\cdot }f(X^-_s)dW^-_s)\) as \(n{\rightarrow }\infty \) and then \(m\rightarrow \infty \). Secondly, we will show that (b) \(B^{\pm }_{n,m}(t)\) converges to 0 in \(L^2(\Omega )\) as \(n{\rightarrow }\infty \) and then \(m{\rightarrow }\infty \). By applying the same techniques, we would also obtain that the same holds with \(C^{\pm }_{n,m}(t)\). Thirdly, we will prove that (c) (5.4) converges to 0 in \(L^2(\Omega )\) as \(n{\rightarrow }\infty \) and then \(m{\rightarrow }\infty \). Once this has been done, one can easily deduce the f.d.d. convergence in law of \((X, F_{+,n}^{(2r-1)}(f, \cdot ), F_{-,n}^{(2r-1)}(f,\cdot ))\) to \((X, \beta _{2r-1}\int _0^{\cdot }f(X^+_s)dW^+_s, \beta _{2r-1}\int _0^{\cdot }f(X^-_s)dW^-_s)\) as \(n{\rightarrow }\infty \), which is equivalent to (5.2).
(a) Finite-dimensional distributions convergence in law of \((X, A^+_{n,m}, A^-_{n,m})\)
Fix m. Showing the f.d.d. convergence in law of \((X, A^+_{n,m}, A^-_{n,m})\) as \(n\rightarrow \infty \) can be easily reduced to checking the f.d.d. convergence in law of the following random-vector valued process:
Thanks to (3.27) in [9] (see also (3.4) in [9] and page 1073 in [5]), we have
where \((B^{(2)}, \ldots , B^{(r)})\) is a \((r-1 )\)-dimensional two-sided Brownian motion and \(\alpha _{2l-1}\) is defined in (2.18), for all \(t \geqslant 0\), \(B_t^{r,+} := B^{(r)}_t\), \(B_t^{r,-}:= B^{(r)}_{-t}\).
Since \(E[ X_x H_{2r-1}(X_{j+1}^{n,\pm } - X_j^{n,\pm })] = 0\) when \(r \geqslant 2\) (Hermite polynomials of different orders are orthogonal), Peccati–Tudor Theorem (see, e.g., [6, Theorem 6.2.3]) applies and yields
with \((B^{(2)}, \ldots , B^{(r-1)})\) is independent of X (and independent of Y as well). We then have, as \(n\rightarrow \infty \) and m is fixed,
with \(\beta _{2r-1} := \sqrt{\sum _{l=2}^r\kappa ^2_{r,l}\,\alpha ^2_{2l-1}}\) and W is a two-sided Brownian motion independent of X (and independent of Y as well). One can write
with
It is clear that \(\displaystyle {K^{\pm }_m(t) \underset{m\rightarrow \infty }{\overset{L^2}{\longrightarrow }} \int _0^t f(X^{\pm }_s)dW^{\pm }_s}\). On the other hand, \(L^{\pm }_m(t)\) converges to 0 in \(L^2\) as \(m\rightarrow \infty \). Indeed, by independence,
where \(\theta _i\) denotes a random real number satisfying \(i2^{-m/2}< \theta _i < (i+1)2^{-m/2}\). Since \(f \in C_b^{\infty }\) and by Cauchy–Schwarz inequality, we deduce that
from which the claim follows. Summarizing, we just showed that
as \(n\rightarrow \infty \) then \(m\rightarrow \infty \).
(b) \(B^{\pm }_{n,m}(t)\) converges to 0 in \(L^2(\Omega )\) as \(n\rightarrow \infty \) and then \(m\rightarrow \infty \).
It suffices to prove that for all \(k \in \{ 2, \ldots ,r\}\),
as \(n\rightarrow \infty \) and then \(m\rightarrow \infty \), where \(B^{\pm ,k}_{n,m}(t)\) is defined as follows
With obvious notation, we have that
It suffices to prove the convergence to 0 of \(B^{+,k}_{n,m}(t)\), the proof for \(B^{-,k}_{n,m}(t)\) being exactly the same. In fact, the reader can find this proof in the proof of [5, Theorem 1, (1.15)] at page 1073.
(c) (5.4) converges to 0 in \(L^2(\Omega )\) as \(n\rightarrow \infty \) and then \(m\rightarrow \infty \).
It suffices to prove that for all \(k \in \{ 2, \ldots ,r\}\), \(\displaystyle { J^{\pm ,k}_{n,m}(t)\overset{L^2}{\longrightarrow } 0 }\) as \(n\rightarrow \infty \) and then \(m\rightarrow \infty \), where \(J^{\pm ,k}_{n,m}(t)\) is defined as follows,
with obvious notation. We will only prove the convergence to 0 of \(J_{n,m}^{+,k}(t)\), the proof for \(J_{n,m}^{-,k}(t)\) being exactly the same. Using the relationship between Hermite polynomials and multiple stochastic integrals, namely \(H_r(2^{nH/2}(X^{+}_{(j+1)2^{-n/2}} - X^{+}_{j2^{-n/2}}) ) = 2^{nrH/2}I_r(\delta _{(j+1)2^{-n/2}}^{\otimes r})\), we obtain, using (2.6) as well,
with obvious notation. Thanks both to the duality formula (2.5) and to (2.2), we have
At this stage, the proof of the claim \(\mathbf{(c)}\) is going to be different according to the value of l:
-
If \(l= 2k-1\) in (5.7) then
$$\begin{aligned}&Q_{n,m}^{+,2k-1}(t) = 2^{-n/2} 2^{nH(2k-1)}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \bigg |E\big [ \Theta _j^n f(X^+)\Theta _{j'}^n f(X^+)\big ]\bigg | \nonumber \\&\quad \times \big |\langle \delta _{(j+1)2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\rangle \big |^{2k-1} \nonumber \\&\quad \leqslant C_f 2^{-n/2} 2^{nH(2k-1)}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \big |\big \langle \delta _{(j+1)2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\big \rangle \big |^{2k-1} \nonumber \\&\quad = C_f 2^{-n/2}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \big |\frac{1}{2}\big (\big |j-j'+1\big |^{2H} + \big |j-j'-1\big |^{2H} - 2 \big |j-j'\big |^{2H}\big ) \big |^{2k-1}\nonumber \\&\quad = C_f 2^{-n/2}\sum _{j = \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1}\sum _{p =j - \lfloor 2^{n/2}t \rfloor + 1}^{j - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}} \big |\frac{1}{2}\big (\big |p+1\big |^{2H} + \big |p-1\big |^{2H} - 2 \big |p\big |^{2H}\big ) \big |^{2k-1}\nonumber \\ \end{aligned}$$(5.8)where we have the first inequality since f belongs to \(C_b^{\infty }\) and the last one follows by the change of variable \(p=j-j'\). Using the notation (2.16), and by a Fubini argument, we get that the quantity given in (5.8) is equal to
$$\begin{aligned}&C_f 2^{-n/2} \sum _{p = \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}} - \lfloor 2^{n/2}t \rfloor +1}^{\lfloor 2^{n/2}t \rfloor - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}} -1} \big |\rho (p)\big |^{2k-1} \big (\big (p+ \lfloor 2^{n/2}t \rfloor \big )\wedge \lfloor 2^{n/2}t \rfloor \nonumber \\&\quad - \big ( p+ \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big )\vee \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}} \big ). \end{aligned}$$(5.9)By separating the cases when \( 0 \leqslant p \leqslant \lfloor 2^{n/2}t \rfloor - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}} -1\) or when \(\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}} - \lfloor 2^{n/2}t \rfloor +1 \leqslant p < 0\) we deduce that
$$\begin{aligned} 0\leqslant & {} \bigg (\frac{(p+ \lfloor 2^{n/2}t \rfloor )}{2^{n/2}}\wedge \frac{(\lfloor 2^{n/2}t \rfloor )}{2^{n/2}} - \frac{( p+ \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}})}{2^{n/2}}\vee \lfloor 2^{m/2}t\rfloor 2^{-m/2}\bigg ) \\\leqslant & {} \lfloor 2^{n/2}t \rfloor 2^{-n/2} - \lfloor 2^{m/2}t \rfloor 2^{-m/2} = \big |\lfloor 2^{n/2}t \rfloor 2^{-n/2} - \lfloor 2^{m/2}t \rfloor 2^{-m/2} \big |\\\leqslant & {} \big |\lfloor 2^{n/2}t \rfloor 2^{-n/2} -t \big |+ \big |t- \lfloor 2^{m/2}t \rfloor 2^{-m/2} \big | \leqslant 2^{-n/2} + 2^{-m/2}. \end{aligned}$$As a result, the quantity given in (5.9) is bounded by
$$\begin{aligned} C_f \sum _{p \in \mathbb {Z}} \big |\rho (p)\big |^{2k-1} \big ( 2^{-n/2} + 2^{-m/2}\big ), \end{aligned}$$with \(\sum _{p \in \mathbb {Z}} |\rho (p)|^{2k-1} < \infty \) (because \(H< 1/2 \leqslant 1 - \frac{1}{4k-2}\)). Finally, we have
$$\begin{aligned} Q_{n,m}^{+,2k-1}(t) \leqslant C\big (2^{-n/2} + 2^{-m/2}\big ). \end{aligned}$$(5.10) -
Preparation to the cases \(0\leqslant l \leqslant 2k-2\): In order to handle the terms \(Q_{n,m}^{+,l}(t)\) whenever \( 0\leqslant l \leqslant 2k-2\), we will make use of the following decomposition:
$$\begin{aligned} \big |d_n^{(+ ,l)}(j,j')\big | \leqslant \frac{1}{4} \big (\Omega ^{(1,l)}_n(j,j') + \Omega ^{(2,l)}_n(j,j') + \Omega ^{(3,l)}_n(j,j') + \Omega ^{(4,l)}_n(j,j')\big ), \end{aligned}$$(5.11)where
$$\begin{aligned} \Omega ^{(1,l)}_n(j,j')= & {} \sum _{a=0}^{2(2k-1-l)}\left( {\begin{array}{c}2(2k-1-l)\\ a\end{array}}\right) \bigg | E\bigg [f^{(a)}\big (X^+_{j2^{-n/2}}\big )f^{(2(2k-1-l)-a)}\big (X^+_{j'2^{-n/2}}\big )\bigg ]\bigg |\\&\times \bigg | \bigg \langle \varepsilon _{j2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{j'2^{-n/2}}^{\otimes (2(2k-1-l)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1-l)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1-l)}\bigg \rangle \bigg |\\ \Omega ^{(2,l)}_n(j,j')= & {} \sum _{a=0}^{2(2k-1-l)}\left( {\begin{array}{c}2(2k-1-l)\\ a\end{array}}\right) \bigg | E\bigg [f^{(a)}\big (X^+_{j2^{-n/2}}\big )f^{\big (2(2k-1-l)-a\big )} \big (X^+_{(j'+1)2^{-n/2}}\big )\bigg ]\bigg |\\&\times \bigg | \bigg \langle \varepsilon _{j2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2}}^{\otimes (2(2k-1-l)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1-l)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1-l)}\bigg \rangle \bigg |\\ \Omega ^{(3,l)}_n(j,j')= & {} \sum _{a=0}^{2(2k-1-l)}\left( {\begin{array}{c}2(2k-1-l)\\ a\end{array}}\right) \bigg | E\bigg [f^{(a)}\big (X^+_{(j+1)2^{-n/2}}\big ) f^{\big (2(2k-1-l)-a\big )}\big (X^+_{j'2^{-n/2}}\big )\bigg ]\bigg |\\&\times \bigg | \bigg \langle \varepsilon _{(j+1)2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{j'2^{-n/2}}^{\otimes (2(2k-1-l)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1-l)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1-l)}\bigg \rangle \bigg |\\ \Omega ^{(4,l)}_n(j,j')= & {} \sum _{a=0}^{2(2k-1-l)}\left( {\begin{array}{c}2(2r-1-l)\\ a\end{array}}\right) \bigg | E\bigg [f^{(a)}\big (X^+_{(j+1)2^{-n/2}}\big )f^{\big (2(2k-1-l)-a\big )} \big (X^+_{(j'+1)2^{-n/2}}\big )\bigg ]\bigg |\\&\times \bigg | \bigg \langle \varepsilon _{(j+1)2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2}}^{\otimes (2(2k-1-l)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1-l)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1-l)}\bigg \rangle \bigg |. \end{aligned}$$ -
For \(1\leqslant l \leqslant 2k-2:\) Since f belongs to \(C_b^{\infty }\) and thanks to (2.7), we deduce that
$$\begin{aligned} d_n^{(+,l)}(j,j') \leqslant C \big (2^{-nH}\big )^{2(2k-1-l)}. \end{aligned}$$As a consequence of this previous inequality, we have
$$\begin{aligned}&Q_{n,m}^{+,l}(t) \nonumber \\&\quad \leqslant C \big (2^{-nH}\big )^{2(2k-2)}\,2^{-n/2}\, 2^{nH(2k-1)}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \big |\langle \delta _{(j+1)2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\rangle \big |^l \nonumber \\&\quad \leqslant C \big (2^{-nH}\big )^{2(2k-2)}\, 2^{nH(2k-1)} 2^{-nHl} \bigg (\sum _{p \in \mathbb {Z}}|\rho (p)|^l\bigg ) \big ( 2^{-n/2} + 2^{-m/2}\big )\nonumber \\&\quad \leqslant C\, 2^{-nH(2k-2)}\big (2^{-n/2} + 2^{-m/2}\big ), \end{aligned}$$(5.12)where we have the second inequality by the same arguments that have been used previously in the case \(l =2k-1\).
-
For \(l=0:\) Thanks to the decomposition (5.11) we get
$$\begin{aligned} Q_{n,m}^{+,0}(t) \leqslant \frac{1}{4} 2^{-n/2}2^{nH(2k-1)}\sum _{k'=1}^4 \sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1}\Omega ^{(k',0)}_n(j,j') \end{aligned}$$(5.13)We will study only the term corresponding to \(\Omega ^{(2,0)}_n(j,j')\) in (5.13), which is representative to the difficulty. It is given by
$$\begin{aligned}&\frac{1}{4} 2^{-n/2}2^{nH(2k-1)}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \sum _{a=0}^{2(2k-1)}\left( {\begin{array}{c}2(2k-1)\\ a\end{array}}\right) \bigg | E\big [f^{(a)}\big (X^+_{j2^{-n/2}}\big )\\&\quad \times f^{(2(2k-1)-a)}\big (X^+_{(j'+1)2^{-n/2}}\big )\bigg ]\bigg | \bigg | \bigg \langle \varepsilon _{j2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2}}^{\otimes (2(2k-1)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1)}\bigg \rangle \bigg |\\&\quad \leqslant C 2^{-n/2}2^{nH(2k-1)}\sum _{j,j'= \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t \rfloor -1} \sum _{a=0}^{2(2k-1)}\bigg | \bigg \langle \varepsilon _{j2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2}}^{\otimes (2(2k-1)-a)}; \\&\qquad \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1)}\bigg \rangle \bigg |. \end{aligned}$$We define \(E_n^{(a,k)}(j,j'):= \big | \big \langle \varepsilon _{j2^{-n/2}}^{\otimes a}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2}}^{\otimes (2(2k-1)-a)}; \delta _{(j+1)2^{-n/2}}^{\otimes (2k-1)}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes (2k-1)}\big \rangle \big |.\) By (2.7), we thus get, with \(\tilde{c}_a\) some combinatorial constants,
$$\begin{aligned} E_n^{(a,k)}(j,j')\leqslant & {} \tilde{c}_a\,2^{-nH(4k-3)} \big ( \big |\big \langle \varepsilon _{j2^{-n/2}}; \delta _{(j+1)2^{-n/2}}\big \rangle \big | + | \langle \varepsilon _{j2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\rangle |\\&+ \big |\big \langle \varepsilon _{(j'+1)2^{-n/2}}; \delta _{(j+1)2^{-n/2}}\big \rangle \big | + \big |\big \langle \varepsilon _{(j'+1)2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\big \rangle \big |\big ). \end{aligned}$$For instance, we can write
$$\begin{aligned}&\sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big |\big \langle \varepsilon _{(j'+1)2^{-n/2}}; \delta _{(j+1)2^{-n/2}}\big \rangle \big |\\&\quad =2^{-nH-1} \sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big | (j+1)^{2H}-j^{2H}+\big |j'-j+1\big |^{2H}-\big |j'-j\big |^{2H} \big |\\&\quad \leqslant 2^{-nH-1} \sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big ((j+1)^{2H}-j^{2H}\big )\\&\qquad + \,2^{-nH-1} \sum _{\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\leqslant j\leqslant j'\leqslant \lfloor 2^{n/2}t\rfloor -1} \big ((j'-j+1)^{2H}-(j'-j)^{2H}\big )\\&\qquad +\, 2^{-nH-1} \sum _{\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\leqslant j'<j\leqslant \lfloor 2^{n/2}t\rfloor -1} \big ((j-j')^{2H}-(j-j'-1)^{2H}\big )\\&\quad \leqslant \frac{3}{2}\,2^{-nH}\big (\lfloor 2^{n/2}t\rfloor - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big ) \lfloor 2^{n/2}t\rfloor ^{2H} \leqslant \frac{3t^{2H}}{2}\big ( 2^{n/2}t - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big ). \end{aligned}$$Similarly,
$$\begin{aligned} \sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big |\big \langle \varepsilon _{j2^{-n/2}}; \delta _{(j+1)2^{-n/2}}\big \rangle \big |\leqslant & {} \frac{3t^{2H}}{2}\big ( 2^{n/2}t -\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big );\\ \sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big |\big \langle \varepsilon _{j2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\big \rangle \big |\leqslant & {} \frac{3t^{2H}}{2}\big ( 2^{n/2}t -\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big );\\ \sum _{j,j'=\lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}}^{\lfloor 2^{n/2}t\rfloor -1} \big |\big \langle \varepsilon _{(j'+1)2^{-n/2}}; \delta _{(j'+1)2^{-n/2}}\big \rangle \big |\leqslant & {} \frac{3t^{2H}}{2}\big ( 2^{n/2}t - \lfloor 2^{m/2}t\rfloor 2^{\frac{n-m}{2}}\big ). \end{aligned}$$As a consequence, we deduce
$$\begin{aligned} Q_{n,m}^{(+,0)}(t)\leqslant C\,2^{-nH(2k-2)} \left( t - \lfloor 2^{m/2}t\rfloor 2^{\frac{-m}{2}}\right) \leqslant C\,2^{-nH(2k-2)}2^{-m/2}. \end{aligned}$$(5.14)
Combining (5.10), (5.12) and (5.14) finally shows
So, we deduce that \(J^{+,k}_{n,m}(t)\) converges to 0 in \(L^2(\Omega )\) as \(n\rightarrow \infty \) and then \(m\rightarrow \infty \). Finally, thanks to (a), (b) and (c), (5.2) holds true. \(\square \)
5.2 Step 2: Limit of \(2^{-n/4}W_{n}^{(1)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}})\)
Thanks to (1.5), for \(H> \frac{1}{6}\), \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}}) \underset{n\rightarrow \infty }{\overset{P}{\longrightarrow }}\int _0^{Y_t} f(X_s)d^{\circ }X_s\). Thus, since \(H< \frac{1}{2}\), we deduce that
5.3 Step 3: Moment bounds for \(W_n^{(2r-1)}(f,\cdot )\)
We recall the following result from [8]. Fix an integer \(r\geqslant 1\) as well as a function \(f\in C^\infty _b\). There exists a constant \(c >0\) such that, for all real numbers \(s<t\) and all \(n\in \mathbb {N}\),
5.4 Step 4: Last step in the proof of (1.7)
Following [2], we introduce the following natural definition for two-sided stochastic integrals: for \(u\in \mathbb {R}\), let
where \(W^+\) and \(W^-\) are defined in Proposition 5.1, \(X^+\) and \(X^-\) are defined in Sect. 4, and \(\int _0^\cdot f(X^{\pm }_s)dW^{\pm }_s\) must be understood in the Wiener–Itô sense.
Using (3.5), (5.15), the conclusion of Step 3 (to pass from \(Y_{T_{\lfloor 2^n t\rfloor ,n}}\) to \(Y_t\)) and since by [2, Lemma 2.3], we have \(Y_{T_{\lfloor 2^n t\rfloor ,n}} \overset{L^2}{\longrightarrow } Y_t\) as \(n \rightarrow \infty \), we deduce that the limit of \(2^{-n/4}V_n^{(2r-1)}(f,t) \) is the same as that of
Thus, the proof of (1.7) follows directly from (5.2), the definition of the integral in (5.16), as well as the fact that X, W and Y are independent.
6 Proof of (1.8)
We suppose that \(H > \frac{1}{2}\). The proof of (1.8) will be done in several steps:
6.1 Step 1: Limits and moment bounds for \(W_n^{(2i-1)}(f,\cdot )\)
We recall the following Itô-type formula from [5, Theorem 4] (see also [10, Theorem 1.3] for an extension of this formula to the bi-dimensional case). For all \(t\in \mathbb {R}\), the following change-of-variable formula holds true for \(H > \frac{1}{2}\)
where F is a primitive of f and \(\int _0^t f(X_s)d^{\circ }X_s\) is the Stratonovich integral of f(X) with respect to X defined as the limit in probability of \(2^{-\frac{nH}{2}}W_{n}^{(1)}(f,t)\) as \(n \rightarrow \infty \).
For the rest of the proof, we suppose that \(f\in C_b^\infty \). The following proposition will play a pivotal role in the proof of (1.8).
Proposition 6.1
There exists a positive constant C, independent of n and t, such that for all \(i\geqslant 1\) and \(t\in \mathbb {R}\), we have
where, we have
Proof
Set \( \phi _n(j,j'):= \Delta _{j,n} f(X)\Delta _{j',n} f(X),\) where we recall that \(\Delta _{j,n} f(X):= \frac{1}{2}(f(X_{j2^{-n/2}})+ f(X_{(j+1)2^{-n/2}})\). Fix \(t \geqslant 0\) (the proof in the case \(t < 0\) is similar), for all \(i\geqslant 1\), we have
with obvious notation at the last equality and with the third equality following from (2.4), the fourth one from (2.6) and the fifth one from (2.5). We have the following estimates.
-
Case \(a= 2i-1\)
$$\begin{aligned} \big |Q_n^{(i,2i-1)}(t)\big |\leqslant & {} 2^{-nH(2 - 2i)}\sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1}E\big ( \big |\phi _n(j,j')\big | \big )\\&\times \big |\langle \delta _{(j+1)2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \rangle \big |^{2i-1}\\\leqslant & {} C 2^{-nH(2-2i)}\sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1}\big |\langle \delta _{(j+1)2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \rangle \big |^{2i-1}. \end{aligned}$$Now, we distinguish three cases:
-
(a)
If \(H < 1 - \frac{1}{(4i-2)}:\) by (2.10) we have
$$\begin{aligned} \left| Q_n^{(i,2i-1)}(t)\right|\leqslant & {} C\, t\, 2^{-nH(2-2i)}\, 2^{n\left( \frac{1}{2} -(2i-1)H\right) }= C\, t\, 2^{-n\left( H- \frac{1}{2}\right) }. \end{aligned}$$ -
(b)
If \(H = 1 - \frac{1}{(4i-2)}:\) by (2.11) we have
$$\begin{aligned} \left| Q_n^{(i,2i-1)}(t)\right|\leqslant & {} C \left[ t(1+n) +t^2\right] 2^{-nH(2-2i)}\, 2^{n\left( \frac{1}{2} -(2i-1)H\right) }\\= & {} C \left[ t(1+n) +t^2\right] 2^{-n\left( H- \frac{1}{2}\right) }. \end{aligned}$$ -
(c)
If \(H > 1 - \frac{1}{(4i-2)}:\) by (2.12) we have
$$\begin{aligned} \left| Q_n^{(i,2i-1)}(t)\right|\leqslant & {} C \,t \,2^{-nH(2-2i)}\, 2^{n\left( \frac{1}{2} -(2i-1)H\right) }\\&+\, C\,t^{2-(2-2H)(2i-1)}\,2^{-nH(2-2i)}\,2^{n(1-(2i-1))}\\= & {} C\,t\, 2^{-n(H- \frac{1}{2})} + C\,t^{2(1-(1-H)(2i-1))}\, 2^{-n(1-H)(2i-2)}. \end{aligned}$$So, we deduce that
$$\begin{aligned} \big |Q_n^{(i,2i-1)}(t)\big |\leqslant & {} C \big [ \big |t\big |(1+n) +t^2\big ] 2^{-n\big (H- \frac{1}{2}\big )} \nonumber \\&+\, C\,\big |t\big |^{2\big (1-(1-H)(2i-1)\big )}\, 2^{-n(1-H)(2i-2)}{} \mathbf{1}_{\big \{H > 1 - \frac{1}{(4i-2)}\big \}}\nonumber \\ \end{aligned}$$(6.4)
-
(a)
-
Preparation to the cases where \(0\leqslant a \leqslant 2i-2\)
Thanks to (2.2) we have
$$\begin{aligned}&D^{4i-2-2a}(\phi _n(j,j'))= D^{4i-2-2a}( \Delta _{j,n} f(X)\Delta _{j',n} f(X))\leqslant C\sum _{l=0}^{4i-2-2a}\nonumber \\&\quad \bigg (f^{(l)}(X_{j2^{-n/2}})\varepsilon _{j2^{-n/2}}^{\otimes l} + f^{(l)}(X_{(j+1)2^{-n/2}})\varepsilon _{(j+1)2^{-n/2} }^{\otimes l}) \tilde{\otimes }\nonumber \\&\quad (f^{(4i-2-2a-l)}(X_{j'2^{-n/2}})\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2-2a -l} + f^{(4i-2-2a-l)}(X_{(j'+1)2^{-n/2}}) \varepsilon _{(j'+1)2^{-n/2} }^{\otimes 4i-2-2a -l})\nonumber \\&\qquad = C \sum _{l=0}^{4i-2-2a}(f^{(l)}(X_{j2^{-n/2}}) f^{(4i-2-2a-l)}(X_{j'2^{-n/2}}) \varepsilon _{j2^{-n/2}}^{\otimes l}\tilde{\otimes }\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2-2a -l}+ f^{(l)}(X_{j2^{-n/2}})\nonumber \\&\qquad \times f^{(4i-2-2a-l)}(X_{(j'+1)2^{-n/2}}) \varepsilon _{j2^{-n/2}}^{\otimes l}\tilde{\otimes }\varepsilon _{(j'+1)2^{-n/2} }^{\otimes 4i-2-2a -l}\nonumber \\&\qquad \quad +f^{(l)}\big (X_{(j+1)2^{-n/2}}\big )f^{(4i-2-2a-l)}(X_{j'2^{-n/2}})\nonumber \\&\qquad \times \varepsilon _{(j+1)2^{-n/2} }^{\otimes l}\tilde{\otimes }\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2-2a -l} + f^{(l)}\big (X_{(j+1)2^{-n/2}}\big )f^{(4i-2-2a-l)} (X_{(j'+1)2^{-n/2}})\varepsilon _{(j+1)2^{-n/2} }^{\otimes l}\tilde{\otimes }\nonumber \\&\quad \varepsilon _{(j'+1)2^{-n/2} }^{\otimes 4i-2-2a -l}\bigg ) \end{aligned}$$(6.5)So, we have
-
Case \(1\leqslant a \leqslant 2i-2\)
$$\begin{aligned}&\left| Q_n^{(i,a)}(t)\right| \\&\quad \leqslant C 2^{-nH(2-2i)}\sum _{l=0}^{4i-2-2a}\sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1} \\&\qquad \bigg | \bigg \langle \bigg (\varepsilon _{j2^{-n/2}}^{\otimes l} + \varepsilon _{(j+1)2^{-n/2} }^{\otimes l}\bigg ) \tilde{\otimes }\bigg (\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2-2a -l} + \varepsilon _{(j'+1)2^{-n/2} }^{\otimes 4i-2-2a -l}\bigg ),\\&\qquad \delta _{(j+1)2^{-n/2}}^{\otimes 2i-1-a}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes 2i-1-a} \bigg \rangle \bigg |\bigg |\bigg \langle \delta _{(j+1)2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \bigg \rangle \bigg |^a\\&\quad \leqslant C \,t^{(2H-1)(4i-2-2a)}\,2^{-nH(2-2i)} \big (2^{-\frac{n}{2}}\big )^{4i-2-2a}\\&\qquad \times \sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1}\bigg |\bigg \langle \delta _{(j+1)2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \bigg \rangle \bigg |^a, \end{aligned}$$where we have the first inequality because \(f\in C^\infty _b\) and thanks to (6.5), and the second one thanks to (2.8) and (2.9). Now, we distinguish three cases:
-
(a)
If \(H < 1 - \frac{1}{2a}:\) by (2.10) we have
$$\begin{aligned} \left| Q_n^{(i,a)}(t)\right|\leqslant & {} C\, t\,t^{(2H-1)(4i-2-2a)}\,2^{-nH(2-2i)}2^{-n(2i-1-a)}\, 2^{n\left( \frac{1}{2} -aH\right) }\\= & {} C\,t^{2(2H-1)(2i-1-a)+1}\,2^{-\frac{n}{2}(2H-1)}\,2^{-n(1-H)\left[ 2i-1-a\right] }. \end{aligned}$$ -
(b)
If \(H = 1 - \frac{1}{2a}:\) by (2.11) we have
$$\begin{aligned}&\left| Q_n^{(i,a)}(t)\right| \\&\quad \leqslant C \left[ t(1+n) +t^2\right] \,t^{(2H-1)(4i-2-2a)}\,2^{-nH(2-2i)}2^{-n(2i-1-a)}\, 2^{n\left( \frac{1}{2} -aH\right) }\\&\quad = C\left[ t(1+n) +t^2\right] \,t^{2(2H-1)(2i-1-a)}\, 2^{-\frac{n}{2}(2H-1)}\,2^{-n(1-H)\left[ 2i-1-a\right] }. \end{aligned}$$ -
(c)
If \(H > 1 - \frac{1}{2a}:\) by (2.12) we have
$$\begin{aligned} \left| Q_n^{(i,a)}(t)\right|\leqslant & {} C\, t\,t^{(2H-1)(4i-2-2a)}\,2^{-nH(2-2i)}2^{-n(2i-1-a)}\, 2^{n\left( \frac{1}{2} -aH\right) }\\&+\, C\, t^{2-(2-2H)a}\,t^{(2H-1)(4i-2-2a)}\,2^{-nH(2-2i)}2^{-n(2i-1-a)}\, 2^{n(1-a)}\\= & {} C\,t^{2(2H-1)(2i-1-a)+1}\,2^{-\frac{n}{2}(2H-1)}\,2^{-n(1-H)\left[ 2i-1-a\right] }\\&+\, C\, t^{2(1-(1-H)a)}\,t^{2(2H-1)(2i-1-a)}\,2^{-n(1-H)\left[ 2i-2\right] }. \end{aligned}$$So, we deduce that
$$\begin{aligned} \big |Q_n^{(i,a)}(t)\big |\leqslant & {} C\big [ \big |t\big |(1+n) +t^2\big ] \,\big |t\big |^{2(2H-1)(2i-1-a)}\,2^{-\frac{n}{2}(2H-1)}\,2^{-n(1-H)\big [2i-1-a\big ]}\nonumber \\&+\, C\, \big |t\big |^{2(1-(1-H)a)}\,\big |t\big |^{2(2H-1)(2i-1-a)}\, 2^{-n(1-H)\big [2i-2\big ]}\mathbf{1}_{\big \{H > 1 - \frac{1}{2a}\big \}}\nonumber \\ \end{aligned}$$(6.6)
-
(a)
-
Case \(a=0\)
$$\begin{aligned} Q_n^{(i,0)}(t)= 2^{-nH(2 - 2i)}\sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1}E\bigg (\bigg \langle D^{4i-2}\big (\phi _n(j,j')\big ) ,\delta _{(j+1)2^{-n/2}}^{\otimes 2i-1}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes 2i-1}\bigg \rangle \bigg ). \end{aligned}$$By (6.5) we deduce that
$$\begin{aligned}&\big |Q_n^{(i,0)}(t)\big |\leqslant C 2^{-nH(2-2i)}\sum _{l=0}^{4i-2}\sum _{j,j'=0}^{\lfloor 2^{\frac{n}{2}} t \rfloor -1} \bigg | \bigg \langle \bigg (\varepsilon _{j2^{-n/2}}^{\otimes l} + \varepsilon _{(j+1)2^{-n/2} }^{\otimes l}\bigg ) \nonumber \\&\quad \tilde{\otimes }\bigg (\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2 -l} + \varepsilon _{(j'+1)2^{-n/2} }^{\otimes 4i-2 -l}\bigg ), \delta _{(j+1)2^{-n/2}}^{\otimes 2i-1}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes 2i-1} \bigg \rangle \bigg |. \end{aligned}$$(6.7)We define
$$\begin{aligned} E_n^{(i,l)}(j,j'):= & {} \bigg | \bigg \langle \bigg (\varepsilon _{j2^{-n/2}}^{\otimes l} + \varepsilon _{(j+1)2^{-n/2}}^{\otimes l}\bigg ) \tilde{\otimes }\bigg (\varepsilon _{j'2^{-n/2}}^{\otimes 4i-2 -l} + \varepsilon _{(j'+1)2^{-n/2}}^{\otimes 4i-2 -l}\bigg ),\\&\quad \delta _{(j+1)2^{-n/2}}^{\otimes 2i-1}\otimes \delta _{(j'+1)2^{-n/2}}^{\otimes 2i-1} \bigg \rangle \bigg |. \end{aligned}$$Observe that by (2.8) and (2.9), we have
$$\begin{aligned}&E_n^{(i,l)}(j,j') \leqslant C t^{(2H-1)(4i-3)}\big (2^{-\frac{n}{2}}\big )^{4i-3}\, \big (\big |\big \langle \big (\varepsilon _{j2^{-n/2}} + \varepsilon _{(j+1)2^{-n/2} }\big ),\delta _{(j'+1)2^{-n/2}} \big \rangle \big |\\&\quad + \big |\big \langle \big (\varepsilon _{j'2^{-n/2}} + \varepsilon _{(j'+1)2^{-n/2} }\big ),\delta _{(j+1)2^{-n/2}} \big \rangle \big | + \big |\big \langle \big (\varepsilon _{j2^{-n/2}}\\&\qquad +\, \varepsilon _{(j+1)2^{-n/2} }\big ),\delta _{(j+1)2^{-n/2}} \big \rangle \big |\\&\quad + \big |\big \langle \big (\varepsilon _{j'2^{-n/2}} + \varepsilon _{(j'+1)2^{-n/2} }\big ),\delta _{(j'+1)2^{-n/2}} \big \rangle \big | \big ). \end{aligned}$$By combining these previous estimates with (6.7), (2.13) and (2.14), we deduce that
$$\begin{aligned} \left| Q_n^{(i,0)}(t)\right|\leqslant & {} C\,\left| t\right| ^{(2H-1)(4i-3)}\,\left| t\right| ^{2H + 1}\,2^{-nH(2-2i)}\,\left( 2^{-\frac{n}{2}}\right) ^{4i-3}\,2^{\frac{n}{2} }\nonumber \\= & {} C\,\left| t\right| ^{(2H-1)(4i-3)}\,\left| t\right| ^{2H + 1}\,2^{-n(2i-2)(1-H)}. \end{aligned}$$(6.8)By combining (6.3) with (6.4), (6.6) and (6.8), we deduce that (6.2) holds true.
\(\square \)
6.2 Step 2: Limit of \(2^{-\frac{nH}{2}}W_{n}^{(2i-1)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}})\)
Let us prove that for \(i\geqslant 2\),
Due to the independence between X and Y and thanks to (6.2), we have
It suffices to prove that
For simplicity, we write \(Y_n(t)\) instead of \(Y_{T_{\lfloor 2^n t\rfloor ,n}}\). We have
Let us prove that, for all \(1 \leqslant a \leqslant 2i-2\)
(the proof of the convergence to 0 of the other terms in (6.11) is similar). In fact, by Hölder inequality, we have
Observe that for \( H > 1 - \frac{1}{2a}\) we have \(2< 4(1-(1-H)a)< 4\). So, by Hölder inequality, we deduce that \(E[|Y_n(t)|^{4(1-(1-H)a)}]^\frac{1}{2} \leqslant E[(Y_n(t))^4]^{\frac{1}{2}(1-(1-H)a)} \leqslant C\) for all \(n \in \mathbb {N}\), where we have the last inequality by Lemma 2.2. On the other hand since \( H> \frac{1}{2}\) we have \(4(2H-1)(2i-1-a) >0\), and it is clear that there exists an integer \(k_0> 1\) such that \(\frac{2k_0}{4(2H-1)(2i-1-a)} > 1\). Thus, by Hölder inequality, we have \(E[|Y_n(t)|^{4(2H-1)(2i-1-a)}]^\frac{1}{2} \leqslant E[ (Y_n(t))^{2k_0}]^{\frac{(2H-1)(2i-1-a)}{k_0}}\leqslant C\) for all \(n \in \mathbb {N}\), where we have the last inequality by Lemma 2.2. Finally, we deduce that
Thus, (6.10) holds true.
6.3 Step 3: Limit of \(V_n^{(1)}(f,\cdot )\)
Recall that for all \(t \geqslant 0\) and \(r\geqslant 1\),
We claim that
We will make use of the following Taylor’s type formula (if interested the reader can find a proof of this formula, e.g., in [1] page 1788). Fix \(f \in C_b^{\infty }\), let F be a primitive of f. For any \(a,\, b \in \mathbb {R}\),
where \(|O ( |b-a|^{5})| \leqslant C_{F}|b-a|^{5}\), \(C_F\) being a constant depending only on F. One can thus write
Thanks to the Minkowski inequality, we have
Due to the independence between X and Y, the self-similarity and the stationarity of increments of X, we have
Finally, thanks to the previous calculation and since \(H> \frac{1}{2}\), we deduce that
By (3.5), we have \(2^{-\frac{3nH}{2}}V_n^{(3)}(f,t) = 2^{-\frac{3nH}{2}}W_{n}^{(3)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}}) + 32^{-\frac{3nH}{2}}W_{n}^{(1)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}}).\) By (6.9), we have that \(2^{-\frac{3nH}{2}}W_{n}^{(3)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}})\) converges to 0 in \(L^2\) as \(n \rightarrow \infty \). By (6.2) and thanks to the independence of X and Y, we deduce that
by Hölder inequality and thanks to Lemma 2.2, we can prove easily that the last quantity converges to 0 as \(n \rightarrow \infty \). Finally, we get
Now, let us prove that
In fact, as it has been mentioned in the introduction, \(T_{\lfloor 2^n t\rfloor ,n} \overset{a.s.}{\longrightarrow } t\) as \(n\rightarrow \infty \) (see [2, Lemma 2.2] for a precise statement), and thanks to the continuity of F as well as the continuity of the paths of Z, we have
In addition, by the mean value theorem, and since f is bounded, we have that \( \big | F(Z_{T_{\lfloor 2^n t\rfloor ,n}}) - F(0) \big | \leqslant \sup _{x\in \mathbb {R}}|f(x)| |Z_{T_{\lfloor 2^n t\rfloor ,n}}|, \) so, we deduce that
Due to independence between X and Y, and to the self-similarity of X, we have \(\Vert Z_{T_{\lfloor 2^n t\rfloor ,n}}\Vert _4 = \Vert X_{Y_{T_{\lfloor 2^n t\rfloor ,n}}}\Vert _4 = \Vert |Y_{T_{\lfloor 2^n t\rfloor ,n}}|^H X_1\Vert _4 = \Vert |Y_{T_{\lfloor 2^n t\rfloor ,n}}|^H\Vert _4 \Vert X_1\Vert _4\). By Hölder inequality, we have \(\Vert |Y_{T_{\lfloor 2^n t\rfloor ,n}}|^H\Vert _4 \leqslant (\Vert Y_{T_{\lfloor 2^n t\rfloor ,n}}\Vert _4)^H.\) Finally, we have
Thanks to Lemma 2.2 and to the previous inequality, we deduce that the sequence \(\big (F(Z_{T_{\lfloor 2^n t\rfloor ,n}}) - F(0)\big )_{n \in \mathbb {N}}\) is bounded in \(L^4\). Combining this fact with (6.17) we deduce that (6.16) holds true.
Finally, combining (6.13) with (6.14), (6.15) and (6.16), we deduce that
By (6.1), we have \(F(X_t)-F(0) = \int _0^t f(X_s)d^{\circ }X_s\) which implies that \(F(Z_t) -F(0)= \int _0^{Y_t} f(X_s)d^{\circ }X_s\). So, we deduce finally that (6.12) holds true.
6.4 Step 4: Last step in the proof of (1.8)
Thanks to (3.5), we have
For \(r=1\), (1.8) holds true by (6.12). For \(r\geqslant 2\), we have \(2^{-\frac{nH}{2}} V_n^{(2r-1)}(f,t) = \kappa _{r,1} 2^{-\frac{nH}{2}}V_n^{(1)}(f,t) + \sum _{i=2}^{r}\kappa _{r,i}2^{-\frac{nH}{2}}W_{n}^{(2i-1)}(f,Y_{T_{\lfloor 2^n t\rfloor ,n}})\). Combining this equality with (6.9) and (6.12), we deduce that (1.8) holds true.
7 Proof of (1.9)
Recall that for all \(t \geqslant 0\) and \(r\geqslant 1\),
and for all \(i \in \mathbb {Z}\), \(\Delta _{i,n} f(X):= \frac{1}{2}(f(X_{i2^{-n/2}})+ f(X_{(i+1)2^{-n/2}})\). Thanks to Lemma 3.1, we have
with obvious notation at the last line. Fix \(t\geqslant 0\). In order to study the asymptotic behavior of \(2^{-\frac{n}{2}}V_n^{(2r)}(t)\) as n tends to infinity (after using the adequate normalization according to the value of the Hurst parameter H) , we shall consider (separately) the cases when n is even and when n is odd.
When n is even, for any even integers \(n\geqslant m\geqslant 0\) and any integer \(p\geqslant 0\), one can decompose \(2^{-\frac{n}{2}}V_n^{(2r)}(t)\) as
where
We can see that since we have taken even integers \(n\geqslant m\geqslant 0\) then \(2^{m/2}\), \(2^{\frac{n-m}{2}}\) and \(2^{n/2}\) are integers as well. This justifies the validity of the previous decomposition.
When n is odd, for any odd integers \(n \geqslant m \geqslant 0\) we can work with the same decomposition for \(V_n^{(2r)}(t)\). The only difference is that we have to replace the sum \(\sum _{-p 2^{m/2} +1\leqslant j\leqslant p 2^{m/2}}\) in \(A_{m,n,p}^{(2r)}(t)\), \(B_{m,n,p}^{(2r)}(t)\), \(C_{m,n,p}^{(2r)}(t)\) and \(D_{m,n,p}^{(2r)}(t)\) by \(\sum _{-p 2^{\frac{m+1}{2}} +1\leqslant j\leqslant p 2^{\frac{m+1}{2}}}\). And instead of \(\sum _{i\geqslant p 2^{n/2}}\) and \(\sum _{i < -p2^{n/2}}\) in \(E_{n,p}^{(2r)}(t)\), we must consider \(\sum _{i\geqslant p 2^{\frac{n+1}{2}}}\) and \(\sum _{i < -p2^{\frac{n+1}{2}}}\) respectively. The analysis can then be done mutatis mutandis.
Suppose that \(\frac{1}{4}<H\leqslant \frac{1}{2}\). Firstly, we will prove that \(2^{-\frac{n}{4}}A_{m,n,p}^{(2r)}(t)\), \(2^{-\frac{n}{4}}B_{m,n,p}^{(2r)}(t)\), \(2^{-\frac{n}{4}}C_{m,n,p}^{(2r)}(t)\) and \(2^{-\frac{n}{4}}E_{n,p}^{(2r)}(t)\) converge to 0 in \(L^2\) by letting n, then m, then p tends to infinity. Secondly, we will study the f.d.d. convergence in law of \((2^{-\frac{n}{4}}D_{m,n,p}^{(2r)}(t) )_{t\geqslant 0}\), which will then be equivalent to the f.d.d. convergence in law of \(( 2^{-\frac{3n}{4}}V_n^{(2r)}(t))_{t\geqslant 0}\).
-
(1)
\(2^{-\frac{n}{4}}A_{m,n,p}^{(2r)}(t)\underset{n\rightarrow \infty }{\overset{L^2}{\longrightarrow }} 0\,:\)
We have, for all \(r \in \mathbb {N}^*\),
where \(H_n\) is the nth Hermite polynomial, \(\mu _{2r}=E[N^{2r}]\) with \(N \sim \mathcal {N}(0,1)\), and \(b_{2r,a}\) are some explicit constants (if interested, the reader can find these explicit constants, e.g., in [9, Corollary 1.2]). We deduce that
with obvious notation at the last line. It suffices to prove that for any fixed m and p and for all \(a\in \{1, \ldots ,r\}\)
Set \( \phi _n(i,i'):= \Delta _{i,n} f(X)\Delta _{i',n} f(X).\) Thanks to (2.4), (2.5), (2.6) and to the independence of X and Y, we have
by obvious notation at the last line. By the points 2 and 3 of Proposition 2.3, see also (3.14) in [9] for the detailed proof, we have
Since \( -p 2^{m/2}+1\leqslant j \leqslant p 2^{m/2}\) and \((j-1)2^{\frac{n-m}{2}}\leqslant i\leqslant j2^{\frac{n-m}{2}}-1\), we deduce that \(-p 2^{n/2} \leqslant i\leqslant p 2^{n/2} -1\). So, \(|i| \leqslant p2^{n/2}\). Consequently we have that \(|i|^{1/4} \leqslant p^{1/4}2^{n/8}\), which shows that \( \Vert \mathcal {L}_{i,n}(t) - L_t^{i2^{-\frac{n}{2}}}(Y)\Vert _2 \leqslant C(p^{1/4} +1)n2^{-\frac{n}{4}}\). Finally, we deduce that
Now, observe that, by the same arguments that has been used to show (6.5) and since \(f\in C_b^\infty \), we have
Since \(H\leqslant \frac{1}{2}\), thanks to (2.7), we have \(\Theta _{i,i',n}^{(a,l)}\leqslant C 2^{-nH(4a-2l)}\). So, by combining (7.4) with (7.5), for \(l=0\), we have
for \(l \ne 0\), we have
By the same arguments that has been used in the proof of (2.10), one can prove that for \(H< 1- \frac{1}{2l}\), we have
For \(H=\frac{1}{2}\), thanks to (2.17) and to the discussion of the case \(H= \frac{1}{2}\) after (2.18), we have
thus, (7.7) holds true for \(l=1\) and \(H= \frac{1}{2}\). So, since \(H\leqslant \frac{1}{2}\), we deduce that
By combining (7.4) with (7.6) and (7.8), we deduce that (7.3) holds true for \(H\leqslant \frac{1}{2}\).
-
(2)
\(2^{-\frac{n}{4}}B_{m,n,p}^{(2r)}(t)\overset{L^2}{\longrightarrow } 0\) as \(m\rightarrow \infty \), uniformly on \(n\,:\)
Using (7.1), we get
with obvious notation at the last line. It suffices to prove that for any fixed p and for all \(a\in \{1, \ldots ,r\}\)
uniformly on n. By the same arguments that has been used to prove (7.4), we get
by Proposition 2.3 (point 2) and Cauchy–Schwarz, we have
So, we deduce that
by obvious notation at the last line.
By the same arguments that has been used in the proof of (7.3), we have, for \(\frac{1}{4}<H\leqslant \frac{1}{2}\), and \(l=0\)
for \(l \ne 0\), we have
So, thanks to (7.7), we deduce that
By combining (7.11) with (7.12) and (7.13), we deduce that (7.10) holds true for \(\frac{1}{4}<H\leqslant \frac{1}{2}\).
-
(3)
\(2^{-\frac{n}{4}}C_{m,n,p}^{(2r)}(t)\overset{L^2}{\longrightarrow } 0\) as \(n\rightarrow \infty \), then \(m\rightarrow \infty \,:\)
Using (7.1), we get
with obvious notation. It suffices to prove that for any fixed p and for all \(a\in \{1, \ldots ,r\}\)
as \(n \rightarrow \infty \), then \(m \rightarrow \infty \). By obvious notation, we have
Thanks to the independence of X and Y, and to the first point of Proposition 2.3, we have
by the same arguments that has been used previously for several times, we deduce that
with obvious notation. Following the proof of (5.6), we get that
-
If \(l=2a\) then the term \(O_{n,m}^{2a}(t)\) in (7.15) can be bounded by
$$\begin{aligned}&\frac{1}{4} \sup _{\left| x-y\right| \leqslant 2^{-m/2}} E\left( \left| f(X_x) - f(X_y)\right| ^2\right) \sum _{i,i'= -p2^{\frac{n}{2}}}^{p2^{\frac{n}{2}} -1}\left| \left\langle \delta _{(i+1)2^{-n/2}} ; \delta _{(i'+1)2^{-n/2}} \right\rangle \right| ^{2a}. \end{aligned}$$Since \(H\leqslant \frac{1}{2}\) and thanks to (7.7), observe that
$$\begin{aligned} O_{n,m}^{2a}(t) \leqslant C\, p\, 2^{n\left( \frac{1}{2} - 2Ha\right) } \sup _{\left| x-y\right| \leqslant 2^{-m/2}} E\left( \left| f(X_x) - f(X_y)\right| ^2\right) . \end{aligned}$$(7.16) -
If \(1 \leqslant l \leqslant 2a-1\) then, by (7.7) among other things used in the proof of (5.6), we have
$$\begin{aligned} O_{n,m}^{l}(t)\leqslant & {} C \left( 2^{-nH}\right) ^{(4a-2l)}\sum _{i,i'= -p2^{\frac{n}{2}}}^{p2^{\frac{n}{2}} -1}|\langle \delta _{(i+1)2^{-n/2}}; \delta _{(i'+1)2^{-n/2}}\rangle |^l \nonumber \\\leqslant & {} C\,p\, 2^{-nH(4a-2l)}\, 2^{n\left( \frac{1}{2} - lH\right) }. \end{aligned}$$(7.17) -
If \(l=0\) then
$$\begin{aligned} O_{n,m}^{0}(t) \leqslant C \left( 2^{-nH}\right) ^{4a}\left( 2p2^{\frac{n}{2}}\right) ^2 \leqslant C\,p^2\, 2^{-4nHa} 2^n. \end{aligned}$$(7.18)
By combining (7.15) with (7.16), (7.17) and (7.18), we get
it is then clear that, since \(\frac{1}{4} < H \leqslant \frac{1}{2}\), the last quantity converges to 0 as \(n \rightarrow \infty \) and then \(m\rightarrow \infty \). Finally, we have proved that (7.14) holds true.
-
(4)
\(2^{-\frac{n}{4}}E_{n,p}^{(2r)}(t)\overset{L^2}{\longrightarrow } 0\) as \(p\rightarrow \infty \), uniformly on \(n\,:\)
Using (7.1), we get
with obvious notation at the last line. It suffices to prove that for all \(a\in \{1, \ldots ,r\}\)
uniformly on n. By the same arguments that has been used previously, we have
It suffices to prove the convergence to 0 of the quantity given in (7.21). We have,
with obvious notation at the last line. It is enough to prove that, for all \(l \in \{0, \ldots ,2a\}\):
uniformly on n. By the same arguments that has been used in the proof of (7.3), for \(\frac{1}{4}<H\leqslant \frac{1}{2}\), we have
For \(l=0:\)
By the third point of Proposition 2.3, we have
so that
which implies
On the other hand, thanks to the point 1 of Proposition 2.3, we have
Consequently, we get
By combining (7.23) with (7.24) and (7.25), we deduce that
Observe that, by Cauchy–Schwarz inequality, we have
Thanks to (7.26), we get
But, for \(k \in \{4,8\},\)
On the other hand, since \(H>\frac{1}{4}\), we have
Finally, we deduce that
uniformly on n.
For \(l\ne 0:\) By the same arguments that has been used in the proof of (7.3) and thanks to (2.17), the Cauchy–Schwarz inequality and (7.26), we have
uniformly on n, and we have the fourth inequality because , since \(H\leqslant \frac{1}{2}\leqslant 1 -\frac{1}{2l}\), \(\sum _{a\in \mathbb {Z}} |\rho (a)|^l< \infty .\) By combining (7.27) and (7.28), we deduce that (7.22) holds true for \(\frac{1}{4} < H\leqslant \frac{1}{2}\).
-
(5)
The convergence in law of \(D_{m,n,p}^{(2r)}(t)\) as \(n\rightarrow \infty \), then \(m \rightarrow \infty \), then \(p \rightarrow \infty \,:\)
Let us prove that
as \(n\rightarrow \infty \), then \(m\rightarrow \infty \), then \(p\rightarrow \infty \), where \(\gamma _{2r}\) and \(\int _{-\infty }^{+\infty } f(X_s)L_t^s(Y)dW_s\) are defined in the point (3) of Theorem 1.1. In fact, using the decomposition (7.1), we have
It was been proved in (3.27) in [9] that
where \((B^{(1)}, \ldots , B^{(r)})\) is a r-dimensional two-sided Brownian motion and \(\alpha _{2a}\) is defined in (2.18). Since for any \(x\in \mathbb {R}\), \(E[ X_x H_{2a}(X_{j+1}^{n,\pm } - X_j^{n,\pm })] = 0\) (Hermite polynomials of different orders are orthogonal), and thanks to the independence between X and Y, Peccati–Tudor Theorem (see, e.g., [6, Theorem 6.2.3]) applies and yields
where \((B^{(1)}, \ldots , B^{(r)})\) is a r-dimensional two-sided Brownian motion independent of X and Y. Hence, for any fixed m and p, we have
where \(\gamma _{2r}:= \sqrt{\sum _{a=1}^rb_{2r,a}^2 \alpha _{2a}^2}\) and W is a two-sided Brownian motion. Fix \(t\geqslant 0\), observe that
with obvious notation at the last line. Since \(E[\int _{-\infty }^{+\infty }\big (f(X_s)L^{s}_{t}(Y)\big )^2ds]\leqslant C\int _{-\infty }^{+\infty }E[(L^{s}_{t}(Y))^2]ds\) \(\leqslant C\int _{-\infty }^{+\infty }\exp {(\frac{-s^2}{2t})}ds < \infty \), where we have the second inequality by the point 1 of Proposition 2.3, and thanks to the independence between (X, Y) and W and the a.s. continuity of \(s\rightarrow f(X_s)\) and \(s\rightarrow L_t^{s}(Y)\), we deduce that
Now, let us prove that, for any fixed p,
In fact, since \(f(X_{(j+1)2^{-\frac{m}{2}}})-f(X_{j2^{-\frac{m}{2}}}) = f'(X_{\theta _j})(X_{(j+1)2^{-\frac{m}{2}}}-X_{j2^{-\frac{m}{2}}})\) where \(\theta _j\) is a random real number satisfying \(j2^{-\frac{m}{2}}< \theta _j <(j+1)2^{-\frac{m}{2}}\), and thanks to the independence of X , Y and W, the independence of the increments of W, and the point 1 of Proposition 2.3, we have
Thus (7.33) holds true. Thanks to (7.30), (7.31), (7.32) and (7.33), we deduce that (7.29) holds true.
Finally, by combining (7.3) with (7.10), (7.14), (7.20) and (7.29), we deduce that (1.9) holds true.
8 Proof of Lemma 2.1
-
1.
We have, \(\langle \varepsilon _u^{\otimes q}, \delta _{(j+1)2^{-n/2}}^{\otimes q} \rangle _{\mathscr {H}^{\otimes q}}= \langle \varepsilon _u, \delta _{(j+1)2^{-n/2}} \rangle _\mathscr {H}^q. \) Thanks to (2.1), we have
$$\begin{aligned} \langle \varepsilon _u, \delta _{(j+1)2^{-n/2}} \rangle _\mathscr {H} = E\big ( X_{u}\big (X_{(j+1)2^{-n/2}}-X_{j2^{-n/2}}\big )\big ). \end{aligned}$$Observe that, for all \(0 \leqslant s \leqslant t\) and \(u\in \mathbb {R}\),
$$\begin{aligned} E\big ( X_{u}(X_{t}-X_{s})\big ) = \frac{1}{2}\big ( t^{2H} -s^{2H}\big ) + \frac{1}{2}\big (\big |s-u\big |^{2H} -\big |t-u\big |^{2H}\big ). \end{aligned}$$Since for \(H \leqslant 1/2\) one has \(|b^{2H} - a^{2H}|\leqslant |b-a|^{2H}\) for any \(a, b \in \mathbb {R}_{+}\), we immediately deduce (2.7).
-
2.
By (2.1), for all \(j, j'\in \{0, \ldots , \lfloor 2^{n/2}t\rfloor -1\}\),
$$\begin{aligned}&\big |\big \langle \varepsilon _{j2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \big \rangle _\mathscr {H} \big |=\big |E\big [X_{j2^{-n/2}}\big (X_{(j'+1)2^{-n/2}} -X_{j'2^{-n/2}}\big )\big ]\big |\nonumber \\&\quad = \big |2^{-nH -1}\big (\big |j'+1\big |^{2H} - \big |j'\big |^{2H}\big ) + 2^{-nH-1}\big ( \big |j-j'\big |^{2H} - \big |j-j'-1\big |^{2H}\big ) \big |\nonumber \\&\quad \leqslant 2^{-nH-1}\big | \big |j'+1\big |^{2H} - \big |j'\big |^{2H}\big | + 2^{-nH-1}\big | \big |j-j'\big |^{2H} - \big |j-j'-1\big |^{2H}\big | .\nonumber \\ \end{aligned}$$(8.1)We consider the function \(f: [a,b] \rightarrow \mathbb {R}\) defined by
$$\begin{aligned} f(x)= \big |x\big |^{2H}. \end{aligned}$$Applying the mean value theorem to f, we have that
$$\begin{aligned} \big |\big |b\big |^{2H} - \big |a\big |^{2H}\big | \leqslant 2H \big (\big |a\big |\vee \big |b\big |\big )^{2H-1}\big |b-a\big | \leqslant 2 \big (\big |a\big |\vee \big |b\big |\big )^{2H-1}\big |b-a\big |. \nonumber \\ \end{aligned}$$(8.2)We deduce from (8.2) that
$$\begin{aligned}&2^{-nH-1}\big |\big |j'+1\big |^{2H} - \big |j'\big |^{2H}\big | \leqslant 2^{-nH}\big |j'+1\big |^{2H-1}\\&\quad \leqslant 2^{-nH}\big | \lfloor 2^{n/2}t\rfloor \big |^{2H-1}\leqslant 2^{-n/2}t^{2H-1}, \end{aligned}$$similarly we have,
$$\begin{aligned} 2^{-nH-1}\big |\big |j-j'\big |^{2H} - \big |j-j'-1\big |^{2H}\big | \leqslant 2^{-nH}\big |\lfloor 2^{n/2}t\rfloor \big |^{2H-1}\leqslant 2^{-n/2}t^{2H-1}. \end{aligned}$$Combining the last two inequalities with (8.1), and since \(\langle \varepsilon _{j2^{-n/2}}^{\otimes q}, \delta _{(j'+1)2^{-n/2}}^{\otimes q} \rangle _{\mathscr {H}^{\otimes q}} = \langle \varepsilon _{j2^{-n/2}}, \delta _{(j'+1)2^{-n/2}} \rangle _\mathscr {H}^q\), we deduce that (2.8) holds true. The proof of (2.9) may be done similarly.
-
3.
By (2.1) we have
$$\begin{aligned}&\big |\langle \delta _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\rangle _\mathscr {H}\big |^r = \big |E\big [\big (X_{(k+1)2^{-n/2}}-X_{k2^{-n/2}}\big ) \big (X_{(l+1)2^{-n/2}}-X_{l2^{-n/2}}\big )\big ] \big |^r\\&\quad = \big |2^{-nH-1}\big (\big |k-l+1\big |^{2H} + \big |k-l-1\big |^{2H} -2\big |k-l\big |^{2H}\big )\big |^r = 2^{-nrH}\big |\rho (k-l)\big |^r, \end{aligned}$$where we have the last equality by the notation (2.16). So, we deduce that
$$\begin{aligned}&\sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\big |\big \langle \delta _{(k+1)2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\big \rangle _\mathscr {H}\big |^r = 2^{-nrH} \sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1} \big |\rho (k-l)\big |^r \nonumber \\&\quad = 2^{-nrH} \sum _{k=0}^{\lfloor 2^{n/2} t \rfloor - 1}\sum _{p=k -\lfloor 2^{n/2} t \rfloor +1}^{k}\big |\rho (p)\big |^r \nonumber \\&\quad = 2^{-nrH} \sum _{p=1 -\lfloor 2^{n/2} t \rfloor }^{\lfloor 2^{n/2} t \rfloor -1}\big |\rho (p)\big |^r \big (\big (p+ \lfloor 2^{n/2} t \rfloor \big )\wedge \lfloor 2^{n/2} t \rfloor - p \vee 0\big )\nonumber \\&\quad \leqslant 2^{-nrH}\lfloor 2^{n/2} t \rfloor \sum _{p=1 -\lfloor 2^{n/2} t \rfloor }^{\lfloor 2^{n/2} t \rfloor -1}\big |\rho (p)\big |^r \leqslant 2^{n\big (\frac{1}{2} -rH\big )}t\sum _{p=1 -\lfloor 2^{n/2} t \rfloor }^{\lfloor 2^{n/2} t \rfloor -1}|\rho (p)|^r, \nonumber \\ \end{aligned}$$(8.3)where we have the second equality by the change of variable \(p=k-l\) and the third equality by a Fubini argument. Observe that \(|\rho (p)|^r \sim (H(2H-1))^r p^{(2H-2)r}\) as \(p \rightarrow +\infty \). So, we deduce that
-
(a)
if \(H< 1 - \frac{1}{2r}:\) \(\sum _{p \in \mathbb {Z}}|\rho (p)|^r < \infty \), by combining this fact with (8.3) we deduce that (2.10) holds true.
-
(b)
If \(H = 1- \frac{1}{2r}:\) \(|\rho (p)|^r \sim \frac{(H(2H-1))^r}{p}\) as \(p \rightarrow +\infty \). So, we deduce that there exists a constant \(C_{H,r} >0\) independent of n and t such that for all integer \(n \geqslant 1\) and all \(t \in \mathbb {R}_+\)
$$\begin{aligned}&\sum _{p=1 -\lfloor 2^{n/2} t \rfloor }^{\lfloor 2^{n/2} t \rfloor -1}\big |\rho (p)\big |^r \leqslant C_{H,r} \left( 1+ \sum _{p=2}^{\lfloor 2^{n/2} t \rfloor } \frac{1}{p}\right) \leqslant C_{H,r} \left( 1+ \int _1^{2^{n/2}t} \frac{1}{x}dx\right) \\&\quad = C_{H,r} \left( 1+ \frac{n\log (2)}{2} + \log (t)\right) \leqslant C_{H,r} \big (1+ n + t\big ). \end{aligned}$$By combining this last inequality with (8.3) we deduce that (2.11) holds true.
-
(c)
If \(H> 1- \frac{1}{2r}:\) \(|\rho (p)|^r \sim \frac{(H(2H-1))^r}{p^{(2-2H)r}}\) as \(p \rightarrow +\infty \) where \(0<(2-2H)r<1\). So, we deduce that there exists a constant \(K_{H,r} >0\) independent of n and t such that for all integer \(n \geqslant 1\) and all \(t \in \mathbb {R}_+\)
$$\begin{aligned}&\sum _{p=1 -\lfloor 2^{n/2} t \rfloor }^{\lfloor 2^{n/2} t \rfloor -1}\big |\rho (p)\big |^r \leqslant K_{H,r} \left( 1+ \sum _{p=1}^{\lfloor 2^{n/2} t \rfloor } \frac{1}{p^{(2-2H)r}}\right) \\&\qquad \leqslant K_{H,r} \left( 1+ \int _0^{2^{n/2}t} \frac{1}{x^{(2-2H)r}}dx\right) \\&\quad =K_{H,r}\left( 1+ \frac{2^{\frac{n}{2}\big (1-(2-2H)r\big )} t^{1-(2-2H)r}}{1-(2-2H)r}\right) \\&\qquad \leqslant C_{H,r}\big (1+2^{\frac{n}{2}\big (1-(2-2H)r\big )} t^{1-(2-2H)r}\big ), \end{aligned}$$where \(C_{H,r}= K_{H,r} \vee \frac{K_{H,r}}{1-(2-2H)r}\). By combining the last inequality with (8.3) we deduce that (2.12) holds true.
-
(a)
-
4.
As it has been proved in (8.1), we have
$$\begin{aligned}&\big |\big \langle \varepsilon _{k2^{-n/2}}, \delta _{(l+1)2^{-n/2}} \big \rangle _\mathscr {H} \big |=\big |E\big [X_{k2^{-n/2}} \big (X_{(l+1)2^{-n/2}}-X_{l2^{-n/2}}\big )\big ]\big |\\&\quad \leqslant 2^{-nH-1}\big | \big |l+1\big |^{2H} - \big |l\big |^{2H}\big | + 2^{-nH-1}\big | \big |k-l\big |^{2H} - \big |k-l-1\big |^{2H}\big |, \end{aligned}$$so, by a telescoping argument we get
$$\begin{aligned}&\sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\big |\langle \varepsilon _{k2^{-n/2}} ; \delta _{(l+1)2^{-n/2}}\big \rangle _\mathscr {H}|\nonumber \\&\quad \leqslant 2^{\frac{n}{2}-1} t^{2H+1} + 2^{-nH -1}\sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\big |\big |k-l\big |^{2H} - \big |k-l-1\big |^{2H}\big |, \end{aligned}$$(8.4)by using the change of variable \(p=k-l\) and a Fubini argument, among other things that has been used in the previous proof, we deduce that
$$\begin{aligned} 2^{-nH -1}\sum _{k,l=0}^{\lfloor 2^{n/2} t \rfloor - 1}\big |\big |k-l\big |^{2H} - \big |k-l-1\big |^{2H}\big | \leqslant 2^{\frac{n}{2}} t^{2H+1}. \end{aligned}$$By combining this last inequality with (8.4) we deduce that (2.13) holds true. The proof of (2.14) may be done similarly.
References
Gradinaru, M., Russo, F., Vallois, P.: Generalized covariations, local time and Stratonovich Itô’s formula for fractional Brownian motion with Hurst index \(H \ge 1/4\). Ann. Probab. 31(4), 1772–1820 (2003)
Khoshnevisan, D., Lewis, T.M.: Stochastic calculus for Brownian motion on a Brownian fracture. Ann. Appl. Probab. 9(3), 629–667 (1999)
Khoshnevisan, D., Lewis, T.M.: Iterated Brownian Motion and its Intrinsic skeletal structure. In: Dalang R.C., Dozzi M., Russo F. (eds) Seminar on Stochastic Analysis, Random Fields and Applications. Progress in Probability, vol 45. Birkhäuser, Basel (1999)
Nourdin, I., Peccati, G.: Weighted power variations of iterated Brownian motion. Electron. J. Probab. 13(43), 1229–1256 (2008)
Nourdin, I., Nualart, D., Tudor, C.: Central and non-central limit theorems for weighted power variations of the fractional Brownian motion. Ann. Inst. Henri Poincaré Probab. Stat. 46(4), 1055–1079 (2009)
Nourdin, I., Peccati, G.: Normal Approximations Using Malliavin Calculus: From Stein’s Method to the Universality. Cambridge University Press, Cambridge (2012)
Nourdin, I., Réveillac, A., Swanson, J.: The weak Stratonovich integral with respect to fractional Brownian motion with Hurst parameter 1/6. Electron. J. Probab. 15(70), 2117–2162 (2010)
Nourdin, I., Zeineddine, R.: An Itô-type formula for the fractional Brownian motion in Brownian time. Electron. J. Probab. 19(99), 1–15 (2014)
Zeineddine, R.: Fluctuations of the power variation of fractional Brownian motion in Brownian time. Bernoulli 21(2), 760–780 (2015)
Zeineddine, R.: Change-of-variable formula for the bi-dimensional fractional Brownian motion in Brownian time. ALEA Lat. Am. J. Probab. Math. Stat. 12(2), 597–683 (2015)
Acknowledgements
We are thankful to the referees for their careful reading of the original manuscript and for a number of suggestions. The financial support of the DFG (German Science Foundations) Research Training Group 2131 is gratefully acknowledged.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Zeineddine, R. Asymptotic Behavior of Weighted Power Variations of Fractional Brownian Motion in Brownian Time. J Theor Probab 31, 1539–1589 (2018). https://doi.org/10.1007/s10959-017-0749-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10959-017-0749-1
Keywords
- Weighted power variations
- Limit theorem
- Malliavin calculus
- Fractional Brownian motion
- Fractional Brownian motion in Brownian time