Abstract
We consider the Navier–Stokes equation on the 2D torus, with a stochastic forcing term which is a cylindrical fractional Wiener noise of Hurst parameter H. Following Albeverio and Ferrario (Ann Probab 32(2):1632–1649, 2004) and Da Prato and Debussche (J Funct Anal 196(1):180–210, 2002) which dealt with the case \(H=\frac{1}{2}\), we prove a local existence and uniqueness result when \(\frac{7}{16}< H<\frac{1}{2}\) and a global existence and uniqueness result when \(\frac{1}{2}<H<1\).
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
Avoid common mistakes on your manuscript.
1 Introduction
An incompressible fluid flow dynamics is described by the so-called incompressible Navier–Stokes equations. In this paper, we consider the Navier–Stokes equations on the torus, i.e., we work on the square \({\mathbb {T}}=[0,2\pi ]^2\) with periodic boundary conditions; we add a stochastic forcing term. These are the equations
where for \(t\in [0,T]\) and \(\xi \in {\mathbb {T}}\), \(v=v(t,\xi )\) is the vector velocity, \(p=p(t,\xi )\) the scalar pressure, \(\nu > 0\) the viscosity coefficient, and \(W^H=W^H(t,\xi )\) a cylindrical fractional Brownian process.
Stochastic perturbations in the equations of motions are commonly used to model small perturbations (numerical, empirical, and physical uncertainties) or thermodynamic fluctuations present in fluid flows. We refer to the lecture notes by Flandoli [14], the monograph of Kuksin and Shirikyan [17] as well as the references cited therein for a recent overview.
Different noise terms have been considered so far. The contribution of this paper is to study Eq. (3) with a cylindrical fraction Brownian motion \(W^H\) for \(H \ne \frac{1}{2}\). Indeed the case \(H=\frac{1}{2}\) has been studied in [1, 3, 8, 9]. Let us point out that with a colored (not cylindrical) noise, the analysis of Eq. (1) is easier; some results on a more general bidimensional domain can be found in [12]. Moreover, the bigger is H the more regular is the fractional Brownian motion. Hence it is worth to ask whether the cylindrical fractional Brownian motion with \(H<\frac{1}{2}\) can be considered; in addition also the analysis for \(H>\frac{1}{2}\) is interesting in order to compare the results for different values of the Hurst parameter. In this paper we shall prove local existence and uniqueness of solutions for \(\frac{7}{16}<H<\frac{1}{2}\) and global existence and uniqueness for \(\frac{1}{2}<H<1\).
As far as the contents of the paper are concerned, in Sect. 2 we introduce the mathematical setting, in Sect. 3 we analyze the linear Stokes problem, whereas Sect. 4 and Sect. 5 analyze the bilinear term and the Navier–Stokes problem, respectively. In Appendices, we present some proofs.
2 Mathematical setting
In this section we introduce the basic tools.
2.1 The spaces
For a complex number \(b=\mathfrak {R}b+i\mathfrak {I}b\), we denote by \({\overline{b}}\) the complex conjugate (\({\overline{b}}=\mathfrak {R}B-i \mathfrak {I}b\)) and by |b| the absolute value (\(|b|=\sqrt{(\mathfrak {R}b)^2+(\mathfrak {I}b)^2}\)).
We consider subspaces of \({\mathbb {Z}}^2\):
and
When \(k=(k^{(1)},k^{(2)})\in {\mathbb {Z}}^2\), we denote by |k| the absolute value (\(|k|=\sqrt{(k^{(1)})^2+(k^{(2)})^2}\)).
We consider the separable Hilbert space \({\mathcal {H}}^0\) which is the \(L^2\)-closure of the space of smooth vectors which are periodic, zero mean value, and divergence free. Let \(\{h_k\}_{k}\) be the basis for \({\mathcal {H}}^0\), given by \(h_k(\xi )=\frac{1}{2\pi } \frac{k^\perp }{|k|}e^{i k \cdot \xi }\) for \(k \in {\mathbb {Z}}^2_0\) and \(\xi \in {\mathbb {T}}\). Notice that, for any \(k \in {\mathbb {Z}}^2_0\), \(h_{-k}(\xi )=-\overline{h}_k(\xi )\) and \(\varDelta h_k=-|k|^2 h_k\). Therefore,
Notice that the complex coefficients \(v_k\) must satisfy \(v_{-k}=-{\overline{v}}_k\) in order to get a real vector v.
More generally, for \(r\in {\mathbb {R}}\) we define
This is a Hilbert space with scalar product
Following [4], we define the periodic divergence-free vector Sobolev spaces (\(r\in {\mathbb {R}}, 1\le p\le \infty \))
and the periodic divergence-free vector Besov spaces as real interpolation spaces
In particular, \({\mathcal {B}}^r_{2\,2}={\mathcal {H}}^r_2 ={\mathcal {H}}^r\). Moreover, (see [4])
Here C is a generic constant. We make the convention to denote different constants by the same symbol C, unless we want to mark them for further reference.
One interesting result in Besov spaces is given by the following estimate of Chemin (see Corollary 1.3.1 in [7]):
if
and \(p,q \in [1,\infty ]\).
2.2 The abstract equation
Let us consider a unitary viscosity \(\nu =1\) in system (1). Then we write the evolution in abstract form as
with the operators formally defined as \(A=\varDelta \) and \(B(u,v)=P[(u\cdot \nabla )v]\), where P is the projector operator onto the space of divergence-free vector fields. We can represent the stochastic forcing term as
where \(\{b^H_k\}_{k \in {\mathbb {Z}}^2_+}\) are a sequence of i.i.d. complex fractional Brownian processes defined on a complete probability space \((\varOmega ,{\mathbb {F}}, {\mathbb {P}})\) with filtration \(\{{\mathbb {F}}_t\}_{\{t \in {\mathbb {R}}\}}\) and \(b^H_{-k}=\overline{b^H_k}\) for all \(k \in {{\mathbb {Z}} ^2_+}\). We denote by \({\mathbb {E}}\) the mathematical expectation with respect to \({\mathbb {P}}\). This means that \(b^H_k(t)=\mathfrak {R}b_k^H(t)+i\mathfrak {I}b_k^H(t)\) and \(\{\mathfrak {R}b^H_k, \mathfrak {I}b_k^H(t)\}_{k \in {\mathbb {Z}}^2_+}\) is a sequence of i.i.d. standard real fractional Brownian processes (fBm) with Hurst parameter H. Each element of the sequence is a centered Gaussian process whose covariance is
\(w^H\) is called an \({\mathcal {H}}^0\)-cylindrical fractional Brownian motion, and one can prove that the series (4) converges in any space U with continuous embedding \({\mathcal {H}}^0\subset U\) of Hilbert–Schmidt type.
Now let us define rigorously the operators A and B.
The Stokes operator A, as a linear operator in \({\mathcal {B}}^{s}_{pq}\) with domain \({\mathcal {B}}^{s+2}_{pq}\), generates an analytic semigroup \(\{e^{tA}\}_{t\ge 0}\) in \({\mathcal {B}}^{s}_{pq}\) and
for any \(t\ge 0\), \(s_1>s_2\).
As far as the bilinear term \(B(u,v)=P[(u\cdot \nabla )v]\) is considered, we recall some basic properties (See [20]). Let \(\langle \cdot ,\cdot \rangle \) denote the \({\mathcal {H}}^{-r}-{\mathcal {H}}^r\) duality bracket. One checks by integration by parts that
and taking \(u_2=u_3\)
These relationships are true with regular entries and then are extended to more general vectors by density.
A basic estimate is (see [15, Lemma 2.2])
when
Other estimates have been given before in (2); indeed, by the divergence-free condition, we have \(B(u,v)=P[\text {div}\ (u \otimes v)]\); therefore, \(\Vert B(u,v)\Vert _{{\mathcal {B}}^s_{pq}}\le \Vert u \otimes v\Vert _{{\mathcal {B}}^{s+1}_{pq}}\).
Moreover, as done in [3], we can develop the bilinear term in Fourier series. Given \(v=\sum _{l\in {{\mathbb {Z}} ^2_0}} v_l h_l\) and \(u=\sum _{h \in {{\mathbb {Z}} ^2_0}} u_h h_h\), we have formally
Using that the projector P acts on the kth component as \(P_k a=\frac{a\cdot k^\perp }{|k|^2}k^\perp \), we get
Summing up, the bilinear term can be written in Fourier series as
with
Notice that \({\overline{B}}_k=-B_{-k}\). The convergence of the series (9) will be analyzed in the next section.
Our aim is to study Eq. (3) for \(H \ne \frac{1}{2}\). Indeed, the case \(H=\frac{1}{2}\) has been studied in [1, 3, 8, 9]: Da Prato and Debussche proved the existence of a strong mild solution for \(\mu \)-a.e. initial condition (where \(\mu \) is the Gibbs measure of the enstrophy, introduced in [2] which is an invariant measure for Eq. (3)), whereas Albeverio and Ferrario proved pathwise uniqueness of these solutions.
We shall prove a local existence and uniqueness result for \(\frac{7}{16}<H<\frac{1}{2}\) and a global existence and uniqueness result for \(H>\frac{1}{2}\). This latter result improves that of [12]; indeed, the case of cylindrical fBm is included in [12] but only for \(H>\frac{3}{4}\). (See Theorem 5.1 and Corollary 4.3 there.) By the way, there are other differences with respect to [12]: in [12] the spatial domain is not the torus but a generic smooth bounded subset D of \({\mathbb {R}}^2\) (and the Dirichlet boundary condition is assumed), and the solution is a process with values in \(L^4\) in time and space, whereas our solution is more regular since a.a. paths are at least in \(C([0,T];{\mathcal {H}}^{\frac{1}{2}})\) (see next Theorem 2 with \(H>\frac{3}{4}\)) and one knows that \({\mathcal {H}}^{\frac{1}{2}}\subset L^4(D)\).
In order to analyze Eq. (3), we introduce as in [8] two subproblems: the linear Stokes equation
and the equation for \(u=v-z\)
which is a Navier–Stokes-type equation with random coefficients.
First we deal with the linear problem for z, then we define the bilinear term B(z, z) a.s. as in [3, 8] and finally face the nonlinear equation for u. At the end, we recover the existence result for v from the representation \(v=z+u\).
3 The Stokes equation
If we neglect the bilinear term in (3), we obtain the linear Stokes equation
We consider its stationary mild solution; this is the process
We can write
First, we provide a result for each stochastic convolution integral appearing in the Fourier series representation.
Lemma 1
Let \(\lambda >0\) and \(b^H\) be a real fBm of Hurst parameter \(H\in (0,1)\). Then
is a stationary centered Gaussian process whose variance is
where \(C_H\) is the positive constant given in (14).
Proof
Following the proof of Lemma 4.1 in [12], we have that the random variables
and
have the same law. Moreover, by self-similarity of the fBm, the latter random variable has the same law as
Therefore,
We estimate \({\mathbb {E}}\left( \int _0^{+\infty } e^{-r}\mathrm{d}b^H(r)\right) ^2\) using the representation
This comes from the formula on a finite time interval
and the fact that by the law of iterated logarithm (see [5]), we get
Hence
By elementary calculations, one shows that the latter integral is finite. We set
\(\square \)
Now we come back to the stationary process z given in (12). We have the following result
Proposition 1
For any \(r<2\left( H-\frac{1}{2}\right) \), we have
Proof
First we show that for any fixed time, the random variable \(z(t)\in {\mathcal {H}}^r\), \({\mathbb {P}}\)-a.s. Indeed, using (13) and the previous lemma, we have
The latter series is convergent for \(4H-2r>2\), i.e., \(r<2\left( H-\frac{1}{2}\right) \).
It follows that for any finite \(m\ge 1\), we have \(z\in L^m_{\text {loc}}({\mathbb {R}};{\mathcal {H}}^r)\), \({\mathbb {P}}\)-a.s. Indeed, z(t) is a Gaussian random variable; so all the moments are finite, i.e., for any \(m\ge 2\), there exists a finite constant \(e_m\) such that
for any t. Moreover, the process z is a stationary process, and by interchanging the integrals, for any \(T_1<T_2\), we get
Since the expectation is finite, then \(\int _{T_0}^{T_1} \Vert z(t)\Vert _{{\mathcal {H}}^r}^m \mathrm{d}t<\infty \), \({\mathbb {P}}\)-a.s.
The continuity in time of the trajectories has been proved in [11] when \(H>\frac{1}{2}\) and in [19] when \(H<\frac{1}{2}\). \(\square \)
Remark 1
We see that when \(H\le \frac{1}{2}\), the process z at any fixed time takes values in a distributional space. This is the source of the difficulty in our problem.
Remark 2
From the proof of Proposition 1, we obtain that the process z is a stationary process, and for any time t, the law of z(t) is the centered Gaussian measure \(\mu ^H\sim {\mathcal {N}}(0,C_H(-A)^{-2H})\). More precisely, we assign the measure \(\mu ^H\) on the sequences \(\{(\mathfrak {R}v_k,\mathfrak {I}v_k)\}_{k \in {{\mathbb {Z}} ^2_+}}\) as
with
When we identify the space \({\mathcal {H}}^r\) with that of the sequences \(\{(\mathfrak {R}v_k,\mathfrak {I}v_k)\}_{k \in {{\mathbb {Z}} ^2_+}}\) such that \(\sum _{k\in {{\mathbb {Z}} ^2_+}} |k|^{2r}[(\mathfrak {R}v_k)^2+(\mathfrak {I}v_k)^2]<\infty \), we get \(\mu ^H({\mathcal {H}}^r)=1\) for any \(r<2\left( H-\frac{1}{2}\right) \) and \(\mu ^H({\mathcal {H}}^r)=0\) for any \(r\ge 2\left( H-\frac{1}{2}\right) \) (See [16]). Similarly, \(\mu ^H({\mathcal {B}}^r_{pq})=1\) for any \(r<2\left( H-\frac{1}{2}\right) \) and \(\mu ^H({\mathcal {B}}^r_{pq})=0\) for any \(r\ge 2\left( H-\frac{1}{2}\right) \).
We finish this section with a result on the deterministic Stokes equation, that will be used in the sequel. Given the deterministic linear problem
we represent its mild solution as
and we have (see [6, Proposition 4.1], based on [10])
Proposition 2
Let \(1<p,q,r<\infty \) and \(s \in {\mathbb {R}}\).
For any \(f \in L^r(0,T;{\mathcal {B}}^{s}_{p\,q})\) and \(x_0 \in {\mathcal {B}}^{s+2-\frac{2}{r}}_{p\,r}\), there exists a unique solution \(x \in W^{1,r}(0,T)\equiv \{ x \in L^r(0,T;{\mathcal {B}}^{s+2}_{p\,q}): \frac{\mathrm{d}x}{\mathrm{d}t} \in L^r(0,T;{\mathcal {B}}^{s}_{p\,q})\}\).
Moreover, the functions \(x,\frac{\mathrm{d}x}{\mathrm{d}t}\) depend continuously on the data f and \(x_0\), that is there exists a positive constant C such that
Finally, the space \(W^{1,r}(0,T)\) is continuously embedded into the space \(C([0,T];{\mathcal {B}}^{s+2-\frac{2}{r}}_{p\,r})\), that is there exists a positive constant C such that
and therefore the initial condition makes sense.
All the constants depend only on p, q, r, s.
4 The bilinear term
When we study equation for the auxiliary process \(u=v-z\), there appears B(z, z). We analyze the space regularity of this term. Following [1,2,3, 8], we estimate it with respect to the Gaussian measure \(\mu ^H\).
Proposition 3
Let \(\frac{1}{4}<H<1\) and
Then, for any \(m \in {\mathbb {N}}\)
Proof
Let us begin to perform computations for \(m=1\).
First, we explain why we need the lower bound \(H>\frac{1}{4}\). By (10) we have
From (10) we have that \(\gamma _{h,k}=\gamma _{k-h,k}\) and \(\gamma ^2_{h,k}\le |k|^2\); then we can bound \(\int \Vert B(z,z)\Vert _{{\mathcal {H}}^\rho }^2\ \mathrm{d}\mu ^H(z)\) by
For any fixed k, the latter series (over h) is convergent if and only if \(8H>2\). Therefore we require
The inner series depends on k as proved in Lemma 5 in Appendix A. Therefore the double series (21) is estimated by
The first series converges when \(\rho <4H-3\), the second one when \(\rho <-1\) and the third one when \(\rho <2H-2\). This provides the summability (19) under conditions (17), (18).
Now, let us consider higher powers \(m>1\). We have that (19) holds also for the other powers, since \(\mu ^H\) is Gaussian and therefore the higher moments are expressed by means of the second moments. For completeness we provide computations for \(m=2\) in Appendix B. \(\square \)
Using the stationarity we can write (19) also as
for an \(t\in {\mathbb {R}}\). As an easy consequence, we obtain
for any \(\infty<t_0<t_1<\infty \). Hence
Corollary 1
Let \(m\ge 1\) and \(T>0\). Choosing \(\rho \) as in (17), (18) we get
\({\mathbb {P}}\)-a.s.
Remark 3
Notice that for \(\frac{1}{2}<H<1\) the quadratic term B(z, z) is in \(L^2(0,T;{\mathcal {H}}^{-1})\), \({\mathbb {P}}\)-a.s.
5 The nonlinear auxiliary equation
Let v be the unknown for our equation (3) and let z be the stationary Stokes process given by (12). The process \(u=v-z\) solves the equation
For \(r<2\left( H-\frac{1}{2}\right) \) we have \(z(0)\in {\mathcal {B}}^r_{pq}\), \({\mathbb {P}}\)-a.s. and we take \(u(0)=v(0)-z(0)\).
We shall prove that Eq. (22) has a local solution when \(\frac{7}{16}<H<\frac{1}{2}\) whereas we have a global result when \(\frac{1}{2}<H<1\). This implies results for the unknown \(v=z+u\).
5.1 \(\frac{1}{4}<H<\frac{1}{2}\)
We consider a mild solution u to Eq. (22). We want to show local existence (and uniqueness) by means of a fixed point argument. Thus we define the mapping \({\mathcal {I}}\)
A fixed point of \({\mathcal {I}}\) is a mild solution of Eq. (22).
Given \(T>0\), let
First, we want to show that \({\mathcal {I}}: {\mathcal {E}}_T\rightarrow {\mathcal {E}}_T\) for suitable values of the parameters \(\alpha , \beta ,\sigma ,p,q,H\).
Define
Given \(u_0\in {\mathcal {B}}^\sigma _{pq}\), it is an easy result that \(I_0 \in {\mathcal {E}}_T\) when
Indeed, \(\Vert e^{tA}u_0\Vert _{{\mathcal {B}}^\sigma _{pq}}\le \Vert u_0\Vert _{{\mathcal {B}}^\sigma _{pq}}\); by (5) we have
and the latter integral is finite when \(\alpha <\sigma +\frac{2}{\beta }\).
To study the integrals involving B(u, u), B(z, u) and B(u, z) we define
Lemma 2
Let \(\alpha ,\sigma \in {\mathbb {R}}\) and \(\beta ,p,q\ge 1\) be such that
If \(u \in L^\beta (0,T;{\mathcal {B}}^\alpha _{pq})\) and \({\tilde{u}} \in C([0,T];{\mathcal {B}}^\sigma _{pq})\), then
and
where the constant C is independent of the time T.
Proof
We consider the first estimate, since the second one is obtained in the same way interchanging u and \({\tilde{u}}\).
We use (2) with \(s_1=\alpha \) and \( s_2=\sigma \):
where \(\alpha <\frac{2}{p}\), \(\sigma <\frac{2}{p}\) and \(\alpha +\sigma >0\). Then we get that \(B(u,{\tilde{u}})\in L^\beta (0,T;{\mathcal {B}}^{\alpha +\sigma -\frac{2}{p}-1}_{pq})\).
Moreover
Now, we perform estimates using (5) and the Hölder inequality:
Integrating in time over the interval [0, T], we conclude the proof. \(\square \)
Now we consider the other norm for \(I_1\).
Lemma 3
Let \(\alpha ,\sigma \in {\mathbb {R}}\) and \(\beta ,p,q\ge 1\) be such that
If \(u \in L^\beta (0,T;{\mathcal {B}}^\alpha _{pq})\) and \({\tilde{u}} \in C([0,T];{\mathcal {B}}^\sigma _{pq})\), then
and
where the constant C is independent of the time T.
Proof
First, from the previous proof we know that \(B(u,{\tilde{u}})\in L^\beta (0,T; {\mathcal {B}}^{\alpha +\sigma -\frac{2}{p}-1}_{pq})\); when \(\beta \ge q\) we also have \(B(u,{\tilde{u}})\in L^q(0,T; {\mathcal {B}}^{\alpha +\sigma -\frac{2}{p}-1}_{pq})\) and Proposition 2 provides \(I_1(u,{\tilde{u}})\in C([0,T];{\mathcal {B}}^{\alpha +\sigma -\frac{2}{p}+1-\frac{2}{q}}_{pq}) \) and finally we use that \({\mathcal {B}}^{\alpha +\sigma -\frac{2}{p}+1-\frac{2}{q}}_{pq}\subseteq {\mathcal {B}}^{\sigma }_{pq}\) when \(\frac{2}{p}+\frac{2}{q}\le \alpha +1 \).
Now, we perform estimates using (5) and the Hölder inequality.
The latter internal is finite when \((\frac{\beta }{\beta -1})(\frac{1}{2}+\frac{1}{p}-\frac{\alpha }{2})<1\), i.e.,
This inequality is true when \(\beta \ge q \) and \(\frac{2}{q}+\frac{2}{p}< \alpha +1\), that is our assumptions imply (24).
Computing the time integral and taking the supremum over \(t\in [0,T]\) we get the required estimate. \(\square \)
For the integral involving B(z, z) we define the process
where z is the Stokes process given in (12).
Lemma 4
Let \(\frac{1}{4}<H<\frac{1}{2}\), \(\beta , p\ge 1\), \(q\ge 2\) and \(\alpha , \sigma \in {\mathbb {R}}\) be such that
Then \(I_2\in {\mathcal {E}}_T\), \({\mathbb {P}}\)-a.s.
Proof
We proceed pathwise. First we show that \(I_2\in L^\beta (0,T;{\mathcal {B}}^\alpha _{pq})\). From Corollary 1 we know that the paths of B(z, z) are in \(L^\beta (0,T;{\mathcal {B}}^{\sigma -1}_{pq})\) for any \(\beta \ge 1\) and for \(\sigma <4H-2\). Therefore, according to Proposition 2 the paths of \(I_2\) are in \(L^\beta (0,T;{\mathcal {B}}^{\sigma +1}_{pq})\). When \(\alpha \le \sigma +1 \), the embedding theorem gives \(I_2 \in L^\beta (0,T;{\mathcal {B}}^{\alpha }_{pq})\).
Now, we show that \(I_2 \in C([0,T];{\mathcal {B}}^\sigma _{pq})\). Again by Corollary 1, for any \(q\ge 1\) and \(\sigma <4H-2\) the paths of B(z, z) are in \(L^q(0,T;{\mathcal {B}}^{\sigma -1}_{pq})\). We bear in mind Proposition 2 and we get that \(I_2\in C([0,T];{\mathcal {B}}^{\sigma +1-\frac{2}{q}}_{pq})\). When \(q\ge 2\) this finishes the proof. \(\square \)
Summing up, we have proved estimates for all the terms in the r.h.s. of (23). Let us point out that merging these results we have to satisfy two conditions:
which comes from Proposition 1 and provides \(z \in C([0,T];{\mathcal {H}}^\sigma )\), \({\mathbb {P}}\)-a.s., so to apply Lemmas 2 and 3 for the integrals involving B(z, u) and B(u, z), and
which comes from Lemma 4 to estimate the integral involving B(z, z).
When \(H<\frac{1}{2}\), the latter condition is stronger and we will write only this one in the following.
Proposition 4
Let \(\frac{1}{4}<H<\frac{1}{2}\), \(\alpha ,\sigma \in {\mathbb {R}}\), \(\beta ,p\ge 1\) and \(q\ge 2\) be such that
Then, for any finite T we have that \({\mathcal {I}}: {\mathcal {E}}_T\rightarrow {\mathcal {E}}_T\), \({\mathbb {P}}\)-a.s.
Remark 4
(How to fulfill conditions) Notice that by condition (33) we have \(\sigma <0\) when \(H<\frac{1}{2}\). Therefore, condition (30) requires \(\alpha >0\).
Moreover, conditions (30) and (32) provide
thus it is necessary that \(\sigma >-\frac{1}{2}\), i.e., \(H>\frac{3}{8}\).
Actually we are going to show that H must be bigger than \(\frac{3}{8}\) in order to satisfy all the conditions (25)–(33). Indeed, we write a system equivalent to the previous one. Condition (29) is trivially satisfied when \(\sigma <0\) and can be neglected. Taking \(\beta =q\), condition (26) is weaker than condition (25), that is (25) implies (26). In addition, since condition (25) requires \(\frac{2}{\beta }<1\), we have that condition (32) is weaker than condition (31), that is (31) implies (32). Therefore, in the case \(\beta =q\) the previous system of conditions is equivalent to
which is simpler to analyze. Let us notice that
This sequence of inequalities is meaningful only when \(-3\sigma < \sigma +1\), i.e., \(\sigma >-\frac{1}{4}\). Taking into account the last condition (33), we see that in order to fulfill all the above conditions it is necessary that \(H>\frac{7}{16}\).
Therefore it is possible to fulfill all the conditions when \(\frac{7}{16}<H<\frac{1}{2}\). Setting \(H=\frac{7}{16}+c\) with \(0<c<\frac{1}{16}\), we can choose for instance
in order to satisfy (25)–(33).
Now we can prove the local existence result for u, proving that \({\mathcal {I}}\) is a contraction for T small enough.
Proposition 5
Let \(\frac{7}{16}< H<\frac{1}{2}\) and the parameters fulfill the conditions (25)–(33). Then, given \(u_0 \in {\mathcal {B}}^\sigma _{pq}\) there exist a stopping time \(\tau \in ]0,T]\) and for \({\mathbb {P}}\)-a.e. \(\omega \) a unique mild solution \(u(\omega ,\cdot )\) of Eq. (22) with values in \(C([0,\tau (\omega )];{\mathcal {B}}^\sigma _{pq})\cap L^\beta (0,\tau (\omega );{\mathcal {B}}^\alpha _{pq})\).
Proof
Using the bilinearity of the operator B, we get
Let us work in the subspace of \({\mathcal {E}}_T\) with \(\Vert u\Vert _{{\mathcal {E}}_T}\le M\). The initial data \(u(0) \in {\mathcal {B}}^\sigma _{pq}\) is fixed. Therefore, according to Lemmas 2 and 3, we have
for a suitable constant \({\overline{C}}\) independent of T.
When T is such that
the mapping \({\mathcal {I}}\) is a contraction and hence has a unique fixed point, which is the unique solution of Eq. (22).
Notice that T is a random time, since inequality (35) involves the random process z. It can be chosen to be a stopping time. \(\square \)
Since \(v=z+u\), we also get existence of a local mild solution v to Eq. (3) where the bilinear term B(v, v) has to be understood as the sum of four terms, that is
Theorem 1
Let \(\frac{7}{16}< H<\frac{1}{2}\) and the parameters fulfill the conditions (25)–(33). Then, given \(v_0 \in {\mathcal {B}}^\sigma _{pq}\) there exist a stopping time \(\tau \in ]0,T]\) and for \({\mathbb {P}}\)-a.e. \(\omega \) a unique mild solution \(v(\omega ,\cdot )\) of Eq. (36) with values in \(C([0,\tau (\omega )];{\mathcal {B}}^\sigma _{pq})\).
Remark 5
We cannot get a global existence result for v as in [1, 8]; indeed, when \(H=\frac{1}{2}\) the Gaussian measure \(\mu ^H\) defined by (15) is invariant for the Navier–Stokes equation (3). This allows to define B(v, v) and to get global existence. However, when \(H\ne \frac{1}{2}\) the measure \(\mu ^H\) is invariant for the Stokes equation (11) but not for the Navier–Stokes equation (3); this depends eventually on the fact that for \(0<H<1\) the Gaussian measure \(\mu ^H\) is (formally) invariant for the deterministic Euler dynamics
only when \(H=\frac{1}{2}\) (and in this case \(\mu ^{\frac{1}{2}}\) is called the enstrophy measure, see [2]).
5.2 \(\frac{1}{2}<H<1\)
When \(\frac{1}{2}<H<1\) the fBm \(w^H\) and the Stokes process z are more regular and we expect more regularity of the processes u and v too. Actually we can obtain an a priori energy estimate; this will lead to global existence. Let us notice that now we deal with solutions which are weak in the sense of PDE’s; for instance the solution u of Eq. (22) has paths at least in \(L^\infty (0,T;{\mathcal {H}}^0)\cap L^2(0,T;{\mathcal {H}}^{1})\) and fulfills for any \(t>0\) and any \(\varphi \in {\mathcal {H}}^1\)
\({\mathbb {P}}\)-a.s. This is obtained from (22) by using (6).
Since the paths of the process u are in \(L^\infty (0,T;{\mathcal {H}}^0)\cap L^2(0,T;{\mathcal {H}}^{1})\) and those of z are in \(C([0,T];{\mathcal {H}}^\sigma )\) for some \(\sigma >0\), then all the terms in the latter relationship are well defined. Let us check the trilinear terms, by using Hölder inequality, interpolation inequality and Sobolev embeddings:
The third trilinear term can be dealt with as with the second term. And finally the latter term is well defined as soon as \(B(z(s),z(s))\in L^1(0,T;{\mathcal {H}}^{-1})\). (See Remark 3.)
Now, let \(0<\sigma <2\left( H-\frac{1}{2}\right) \) for \(\frac{1}{2}<H<1\). Taking the \({\mathcal {H}}^0\)-scalar product of Eq. (22) with u, we get the usual energy estimate (See [20]). We make use of (7) and (8):
Moreover, by interpolation and Young inequality
Since \(\Vert u\Vert ^2_{{\mathcal {H}}^1}=\Vert u\Vert ^2_{{\mathcal {H}}^0}+\Vert \nabla u\Vert ^2_{L^2}\), collecting all the estimates we have found
According to Remark 3, \(B(z,z)\in L^2(0,T;{\mathcal {H}}^{-1})\). Moreover \(z \in C([0,T];{\mathcal {H}}^\sigma )\) by Proposition 1. This provides as usual by means of Gronwall Lemma that \(u \in L^\infty (0,T;{\mathcal {H}}^0)\cap L^2(0,T;{\mathcal {H}}^1)\), \({\mathbb {P}}\)-a.s. The reader can see all the details of this standard procedure in [20]. First one has to work on the finite dimensional approximation and then pass to the limit. By interpolation
Hence, \(v=z+u \in L^\infty (0,T;{\mathcal {H}}^0)\cap L^{\frac{2}{\sigma }}(0,T;{\mathcal {H}}^\sigma )\).
We can improve the estimates, now getting \(u \in C([0,T];{\mathcal {H}}^\sigma )\cap L^2(0,T;{\mathcal {H}}^{1+\sigma })\). This gives global existence for the process \(v=z+u\) in the space \(C([0,T];{\mathcal {H}}^\sigma )\) for \(0<\sigma <2\left( H-\frac{1}{2}\right) \). Therefore the term B(v, v) is well defined. Actually the process v is a weak solution (in the sense of PDE’s) to Eq. (3), that is it solves for any \(t>0\) and any \(\varphi \in {\mathcal {H}}^{2-\sigma }\)
\({\mathbb {P}}\)-a.s. We leave to the reader to check that all term are well defined (use that \(2-\sigma >1\)). Similarly, the process z can be considered as a weak solution of the stochastic Stokes equation (11).
Theorem 2
(Global existence) Let \(\frac{1}{2}<H<1\) and
Given \(v_0\in {\mathcal {H}}^\sigma \) there exists a \(C([0,T];{\mathcal {H}}^\sigma )\cap L^2(0,T;{\mathcal {H}}^{1+\sigma })\)-valued process u solving Eq. (22) with \(u(0)=v_0-z(0)\). Therefore there exists a \(C([0,T];{\mathcal {H}}^\sigma )\)-valued process v solving Eq. (3) with \(v(0)=v_0\).
Proof
We have to work on
We performed the estimates on the terms in the r.h.s. Notice that \(0<\sigma <1\).
From (8), the interpolation inequality \(\Vert u\Vert _{\mathcal {H}^{1}}\le C\Vert u\Vert ^{\sigma }_{\mathcal {H}^{\sigma }}\Vert u\Vert ^{1-\sigma }_{\mathcal {H}^{1+\sigma }}\) and Young inequality we get
where \(0<\epsilon \ll 1\), and
Now, \( \Vert u\Vert ^2_{{\mathcal {H}}^{1+\sigma }}=\Vert u\Vert ^2_{{\mathcal {H}}^\sigma }+\Vert \nabla u\Vert ^2_{H^\sigma }\). Summing up
Now we bear in mind (37), Proposition 1 with \(\epsilon \ll 1\) and Proposition 3 to deal with the sum in the r.h.s. So by means of Gronwall lemma, we conclude that \(u \in L^\infty (0,T;{\mathcal {H}}^{\sigma })\cap L^2(0,T;{\mathcal {H}}^{\sigma +1})\). In addition, by means of the previous estimates we get
Since \(u\in L^2(0,T;{\mathcal {H}}^{\sigma +1})\), one gets that \(u \in C([0,T];{\mathcal {H}}^{\sigma })\). (See [20].) Hence \(v=u+z\in C([0,T];{\mathcal {H}}^{\sigma })\). \(\square \)
The solution obtained is also unique. We have a pathwise uniqueness result.
Theorem 3
(Uniqueness) Let \(\frac{1}{2}<H<1\) and \(0<\sigma <2\left( H-\frac{1}{2}\right) \). Given \(v_0\in {\mathcal {H}}^\sigma \) there exists a unique \(C([0,T];{\mathcal {H}}^\sigma )\)-valued process solving (3).
Proof
Let \(v_1,v_2 \in C([0,T];{\mathcal {H}}^\sigma )\) be solutions of (3). Then the difference \(V=v_1-v_2\) fulfills
with \(V(0)=0\). We are going to prove that \(V(t)=0\) for all \(t\ge 0\) and this is obtained by means of the a priori estimate of the energy. Actually the paths of V are more regular than those of \(v_1\) and \(v_2\), since the noise term has disappeared in (39); this was remarked already in [13]. More precisely, we state that any solution V of (39) with \(V(0)=0\) is such that \( V\in C([0,T];\mathcal {H}^0)\cap L^2(0,T; \mathcal {H}^1)\) and \(\frac{\mathrm{d}V}{\mathrm{d}t}\in L^2(0,T;\mathcal {H}^{-1})\); therefore the equality \(\frac{\mathrm{d}}{\mathrm{d}t}\Vert V(t)\Vert ^2_{\mathcal {H}^0}=2\langle \frac{\mathrm{d}V}{\mathrm{d}t}(t),V(t)\rangle \) holds and the energy estimates (coming later) are justified.
Indeed we are given \(v_1,v_2 \in C([0,T];\mathcal {H}^\sigma )\) with \(\sigma \in (0,1)\). The r.h.s. of (39) belongs to \(L^\infty (0,T;\mathcal {H}^{2\sigma -2})\), since
by (8). According to Proposition 2 (used with vanishing initial data and \(r=2\)) we get that any solution V will be in \(L^2(0,T;\mathcal {H}^{2\sigma })\).
If \(2\sigma \ge 1\) (i.e., when \(\frac{1}{2}\le \sigma <1\)) we have obtained that \(V \in L^2(0,T; \mathcal {H}^1)\); moreover, \(\frac{\mathrm{d}V}{\mathrm{d}t}=-AV+B(v_1,v_1)-B(v_2,v_2)\) and therefore \(\frac{\mathrm{d}V}{\mathrm{d}t}\in L^2(0,T;\mathcal {H}^{-1})\).
Otherwise, when \(0<\sigma <\frac{1}{2}\) we proceed as follows; by the bilinearity of B we get that (39) can be written as
Let us look at the regularity of the r.h.s., knowing that \(v_1,v_2 \in C([0,T];\mathcal {H}^\sigma )\) and \(V\in C([0,T];\mathcal {H}^\sigma )\cap L^2(0,T;\mathcal {H}^{2\sigma })\).
Thanks to (8) we get, for \(0<\sigma <\frac{1}{2}\)
Hence the r.h.s. of (40) belongs to \(L^2(0,T;\mathcal {H}^{3\sigma -2})\) and therefore thanks to Proposition 2 any solution V belongs to \(L^2(0,T;\mathcal {H}^{3\sigma })\).
If \(3\sigma \ge 1\) (i.e., when \(\frac{1}{3}\le \sigma <\frac{1}{2}\)) we have obtained that \(V \in L^2(0,T; \mathcal {H}^1)\) and moreover the r.h.s. of (40) belongs to \(L^2(0,T; \mathcal {H}^{3\sigma -2})\subseteq L^2(0,T; \mathcal {H}^{-1})\). Hence we conclude as in the previous case about \(\frac{\mathrm{d}V}{\mathrm{d}t}\).
Otherwise, for smaller values of \(\sigma \) we proceed again with the bootstrap argument. We conclude that, given \(\sigma \in (0,1)\) and \(v_i \in C([0,T]; \mathcal {H}^\sigma )\), any solution V to (40) is in \( L^2(0,T; \mathcal {H}^1)\) and \(\frac{\mathrm{d}V}{\mathrm{d}t}\in L^2(0,T;\mathcal {H}^{-1})\).
Now, we look for the a priori energy estimate. Keeping in mind (7), (8) and the interpolation inequality \(\Vert V\Vert _{\mathcal {H}^{1-\sigma }}\le C \Vert V(t)\Vert ^\sigma _{\mathcal {H}^{0}} \Vert V(t)\Vert ^{1-\sigma }_{\mathcal {H}^1}\), we get
From
we conclude by Gronwall Lemma that \({\sup _{0\le t\le T}} \Vert V(t)\Vert _{\mathcal {H}^0}=0\). This proves pathwise uniqueness. \(\square \)
References
Albeverio, S., Cruzeiro, A.B.: Global flows with invariant (Gibbs) measures for Euler and Navier–Stokes two-dimensional fluids. Commun. Math. Phys. 129(3), 431–444 (1990)
Albeverio, S., Ribeiro de Faria, M., Høegh-Krohn, R.: Stationary measures for the periodic Euler flow in two dimensions. J. Stat. Phys. 20(6), 585–595 (1979)
Albeverio, S., Ferrario, B.: Uniqueness of solutions of the stochastic Navier–Stokes equation with invariant measure given by the enstrophy. Ann. Probab. 32(2), 1632–1649 (2004)
Bergh, L., Löfström, J.: Interpolation Spaces. An introduction. Springer, Berlin (1976)
Biagini, F., Hu, Y., Øksendal, B., Zhang, T.: Stochastic Calculus for Fractional Brownian Motion and Applications, Probability and Its Applications (New York). Springer, London (2008)
Brzeźniak, Z.: Stochastic partial differential equations in M-type 2 Banach spaces. Potential Anal. 4(1), 1–45 (1995)
Chemin J.-Y.: About Navier–Stokes system. Publication du Laboratoire d’Analyse Numérique R 96023 (1996)
Da Prato, G., Debussche, A.: Two-dimensional Navier–Stokes equations driven by a space–time white noise. J. Funct. Anal. 196(1), 180–210 (2002)
Debussche, A.: The 2D-Navier–Stokes Equations Perturbed by a Delta Correlated Noise, Probabilistic Methods in Fluids, 115129. World Scientific, River Edge (2003)
Dore, G., Venni, A.: On the closedness of the sum of two closed operators. Math. Z. 196(2), 189–201 (1987)
Duncan, T.E., Pasik-Duncan, B., Maslowski, B.: Fractional Brownian motion and stochastic equations in Hilbert spaces. Stoch. Dyn. 2(2), 225–250 (2002)
Fang, L., Sundar, P., Viens, F.: Two-dimensional stochastic Navier–Stokes equations with fractional Brownian noise. Random Oper. Stoch. Equ. 21(2), 135–158 (2013)
Ferrario, B.: Uniqueness result for the 2D Navier–Stokes equation with additive noise. Stoch. Stoch. Rep. 75(6), 435–442 (2003)
Flandoli, F.: An introduction to 3D stochastic fluid dynamics. In: SPDE in Hydrodynamic: Recent Progress and Prospects. Lecture Notes in Mathematics, vol. 1942, pp. 51–150. Springer, Berlin (2008)
Giga, Y., Miyakawa, T.: Solutions in \(L_r\) of the Navier–Stokes initial value problem. Arch. Ration. Mech. Anal. 89(3), 267–281 (1985)
Kuo, H.H.: Gaussian measures in Banach spaces. In: Lecture Notes in Mathematics, vol. 463. Springer, Berlin (1975)
Kuksin, S., Shirikyan, A.: Mathematics of Two-Dimensional Turbulence, vol. 194 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge (2012)
Maslowski, B., van Neerven, J.: Equivalence of laws and null controllability for SPDEs driven by a fractional Brownian motion. Nonlinear Differ. Equ. Appl. 20, 1473–1498 (2013)
Pasik-Duncan, B., Duncan, T.E., Maslowski, B.: Linear stochastic equations in a Hilbert space with a fractional Brownian motion. In: Yan, H., Yin, G., Zhang, Q. (eds.) Stochastic Processes, Optimization, and Control Theory: Applications in Financial Engineering, Queueing Networks, and Manufacturing Systems, International Series in Operations Research & Management Science, vol. 94, pp. 201–221. Springer, New York (2006)
Temam, R.: Navier-Stokes Equations., Theory and Numerical Analysis, Studies in Mathematics and Its Applications, vol. 2. North-Holland, Amsterdam (1977)
Author information
Authors and Affiliations
Corresponding author
Additional information
Benedetta Ferrario is partially supported by INdAM-GNAMPA, by MIUR-Dipartimenti di Eccellenza Program (2018–2022) and by PRIN 2015 “Deterministic and stochastic evolution equations”. Christian Olivera is partially supported by FAPESP—Grants 2017/17670-0 and 2015/07278-0.
Appendices
A Auxiliary results
We prove some results about convergence of series.
Lemma 5
For any \(k \in {{\mathbb {Z}} ^2_0}\), the series
converges if \(H>\frac{1}{4} \) and its sum \(S_1(k)\) depends on k as follows
for some positive constants M and \(M_H\) independent of k.
Proof
The series can be estimated by the following integral
over the region \(D_k\) which is the plane without two unitary balls around the points k and 0. Therefore it is enough to evaluate
for \(a\ge 2\), where \({\tilde{D}}_a={\mathbb {R}}^2\setminus (B_1((0,0))\cup B_1((a,0)))\) and \(B_r((c,d))\) denotes the ball of radius r with center (c, d).
Now we make a change of variables: \(u=\frac{x}{a}\) and \(v=\frac{y}{a}\). The domain \(D_a\) becomes the domain \(R_a={\mathbb {R}}^2\setminus (B_{\frac{1}{a}}((0,0))\cup B_{\frac{1}{a}}((1,0)))\). Hence
We split the integral region into three disjoint regions: \(R^0_a=B_{\frac{1}{2}}((0,0))\setminus B_{\frac{1}{a}}((0,0))\), \(R^1_a=B_{\frac{1}{2}}((1,0))\setminus B_{\frac{1}{a}}((1,0))\) and \(R^\infty =R_a\setminus (R^0_a\cup R^1_a)\).
By symmetry the integral over \(R^0_a\) is the same as that over \(R^1_a\) and we have
The integral over \(R^\infty \) is a constant \({\tilde{C}}_H\) independent of a, when \(H>\frac{1}{4}\). Summing up, we get that there exist positive constants M and \(M_H\) such that
Returning to the notation with k we get our result. \(\square \)
In the next lemma, we consider a restricted range for H; the assumption on \(\rho \) is a restriction of that in (18).
Lemma 6
We are given \(\frac{1}{2}<H<1\). Let us assume \(-1<\rho <2(H-1)\).
Then, for any \(k \in {{\mathbb {Z}} ^2_0}\) the series
converges and its sum \(S_2(k)\) is bounded by
for a suitable constant \(C_H\).
Proof
By assumption we have \(4H+4H-2\rho -2>2\) and therefore the series is convergent. To estimate its sum we proceed as in the proof of the previous lemma. First the series can be estimated by an integral and it is enough to evaluate
for \(a\ge 2\). By the change of variables \(u=\frac{x}{a}\) and \(v=\frac{y}{a}\) we obtain
We split the integral over \(R_a\) into three parts by setting \(R_a=R^0_a\cup R^1_a\cup R^\infty \) with disjoint unions:
-
over \(R^0_a=B_{\frac{1}{2}}((0,0))\setminus B_{\frac{1}{a}}((0,0))\)
$$\begin{aligned} \begin{aligned} \iint _{R^0_a} \frac{\mathrm{d}u \ \mathrm{d}v }{(u^{2}+ v^{2})^{2H-\rho -1} \ ((u-1)^{2}+ v^{2})^{2H}}&\le 2^{4H} \iint _{R^0_a} \frac{\mathrm{d}u \ \mathrm{d}v }{(u^{2}+ v^{2})^{2H-\rho -1} } \\&= 2^{4H} 2\pi \int _{1/a}^{1/2}\frac{r\ \mathrm{d}r}{r^{2(2H-\rho -1)}} \end{aligned} \end{aligned}$$By assumption we get that \(2(2H-\rho -1)-1>1\); therefore the latter integral is bounded by
$$\begin{aligned} C_H a^{4H-2\rho -4} \end{aligned}$$ -
over \(R^1_a=B_{\frac{1}{2}}((1,0))\setminus B_{\frac{1}{a}}((1,0))\) we proceed as in the proof of the previous lemma and get that the integral over \(R^1_a\) is bounded by
$$\begin{aligned} C_H a^{4H-2} \end{aligned}$$ -
over \(R^\infty =R_a\setminus (R^0_a\cup R^1_a)\): this integral is bounded, uniformly in a.
Now we compare the exponents of a; since \(\rho >-1\), we have \(4H-2\rho -4 <4H-2\). Summing the three contributions and noticing that \(4H-2>0\) we conclude that the integral over the region \(R_a\) is bounded by \(C_H a^{4H-2}\) for all \(a\ge 2\). Thus
for a suitable constant \(C_H\). \(\square \)
Lemma 7
Let \(\frac{1}{4}<H<1\) and
Then
Proof
First let us prove it when \(2\rho +2>0\); this is possible only when \(\frac{1}{2}<H<1\). In this case we have
which holds for any \(h,l \in {{\mathbb {Z}} ^2_0}\). It comes from
this is equivalent to
Thus, by triangle inequality we obtain (41).
Hence with a positive power we get
This implies that we study the series
The inner series is estimated by Lemma 6; thus
The assumption (18) implies that \(2(4H-3\rho -3)>2\) and therefore this latter series is convergent.
Now let us consider the case \(2\rho +2\le 0\). We have \(|h-l|^{(2\rho +2)}\le 1\). Therefore we are left with
We handle this contribution according to Lemma 5. Indeed it is bounded by
Thus there is convergence if
These conditions are fulfilled under assumptions (17) and (18). \(\square \)
B The case \(m=2\)
We present the proof of Proposition 3 for the case \(m=2\).
First we shall use many times that \(\gamma _{h,k}=\gamma _{k-h,k}\) and \(\gamma ^2_{h,k}\le |k|^2\) for any h. We have
Now we consider the series. There are indeed 6 sums but we have to consider only the non vanishing integrals: \(\mu ^H\) is the product of centered Gaussian measures and all odd powers give zero contribution in the integral with respect to the measure \(\mu ^H\). In particular we analyze
We get different contributions according to the choice of equal indices. Let us list all these possible contributions; we choose h equal to some subsequent index but the case leading to \(z_h=\overline{z_{k-h}}\) is not possible since \(k \ne 0\). Therefore there are 6 possible cases. In the following we do not specify that the sum involves indices belonging to \({{\mathbb {Z}} ^2_0}\) in order to shorten the notation.
-
1.
When \(h=h^\prime \):
-
for \(l=l^\prime \) we get
$$\begin{aligned}\begin{aligned}&\sum _{k}\sum _{j} |k|^{2\rho }|j|^{2\rho } \sum _{h \ne k} \sum _{l\ne j} \int \gamma ^2_{h,k} |z_h|^2 |z_{k-h}|^2 \gamma ^2_{l,j} |z_l|^2 |z_{j-l}|^2 \ \mu ^H(\mathrm{d}z) \\&\quad =\left( \sum _{k} |k|^{2\rho } \sum _{h \ne k} \int \gamma ^2_{h,k} |z_h|^2 |z_{k-h}|^2 \ \mu ^H(\mathrm{d}z)\right) ^2 \end{aligned} \end{aligned}$$We know from (20) that the quantity in the parenthesis is finite under assumptions (17) and (18).
-
for \(l=j-l^\prime \) we get as before.
-
-
2.
When \(h=k-h^\prime \):
-
for \(l=l^\prime \) we get as before;
-
for \(l=j-l^\prime \) we get as before.
-
-
3.
When \(h=-l\) we get
$$\begin{aligned}&\sum _{k} \sum _{j} |k|^{2\rho }|j|^{2\rho } \sum _{h \ne k}\sum _{h^\prime \ne k} \sum _{l^\prime \ne j} \gamma _{h,k} \gamma _{h^\prime ,k} \gamma _{-h,j}\gamma _{l^\prime ,j} \int |z_h|^2 z_{k-h} \overline{z_{h^\prime }}\ \overline{z_{k-h^\prime }} z_j\\&\quad +\,h \overline{ z_{l^\prime }}\ \overline{ z_{j-l^\prime }} \ \mu ^H(\mathrm{d}z) \end{aligned}$$This gives
-
for \(k-h=h^\prime \) or \(h=h^\prime \)
$$\begin{aligned}\begin{aligned}&\sum _{k} \sum _{j} |k|^{2\rho }|j|^{2\rho } \sum _{h \ne k,-j} \sum _{l^\prime \ne j} \int \gamma _{h,k}^2 |z_h|^2 |z_{k-h}|^2 \overline{z_{h}} \gamma _{-h,j} z_{j+h} \gamma _{l^\prime ,j} \overline{ z_{l^\prime }}\ \overline{ z_{j-l^\prime }} \ \mu ^H(\mathrm{d}z) \\&\quad =\sum _{k}\sum _{j} |k|^{2\rho }|j|^{2\rho } \sum _{h \ne k,-j} \gamma _{h,k}^2 \left( \gamma _{-h,j}^2+\gamma _{-h,j} \gamma _{h+j,j}\right) \\&\quad \quad \int |z_h|^4 |z_{k-h}|^2 |z_{j+h}|^2 \ \mu ^H(\mathrm{d}z) \\&\quad \le C \sum _{k}\sum _{j} |k|^{2\rho +2}|j|^{2\rho +2} \sum _{h \ne k,-j} \frac{1}{|h|^{8H} |k-h|^{4H} |j+h|^{4H} } \\&\quad = C \sum _{h} \frac{1}{|h|^{8H} } \left( \sum _{k \ne h}\frac{|k|^{2\rho +2}}{ |k-h|^{4H} }\right) ^2 \end{aligned} \end{aligned}$$The inner series converges for \(\rho <2(H-1)\) (i.e., \(2\rho +2-4H<-2\)); from (17) and (18) we know that this condition is fulfilled. Moreover its sum depends on h in such a way that it vanishes when \(|h| \rightarrow \infty \); hence it is bounded. Thus we are left with the convergence of the first series \(\sum _{h} |h|^{-8H} \), which holds for any \(H>\frac{1}{4}\). So this contribution is finite.
-
for \(k=-j\)
$$\begin{aligned} \begin{aligned} \sum _{k}&|k|^{4\rho } \sum _{h \ne k} \sum _{h^\prime \ne k} \sum _{l^\prime \ne -k} \int \gamma _{h,k} |z_h|^2 |z_{k-h}|^2 \gamma _{h^\prime ,k} \overline{z_{h^\prime }}\ \overline{z_{k-h^\prime }} \gamma _{-h,-k} \gamma _{l^\prime ,-k} \overline{ z_{l^\prime }}\ \overline{ z_{-k-l^\prime }} \ \mu ^H(\mathrm{d}z) \\&= \sum _{k} |k|^{4\rho } \sum _{h \ne k} \sum _{h^\prime \ne k} \gamma _{h,k} \gamma _{h^\prime ,k} \gamma _{-h,-k} (\gamma _{-h^\prime ,-k} +\gamma _{h^\prime -k,-k}) \\&\quad \quad \int |z_h|^2 |z_{k-h}|^2 |z_{h^\prime }|^2 |z_{k-h^\prime }|^2 \ \mu ^H(\mathrm{d}z) \\&\le C \sum _{k} |k|^{4\rho +4}\sum _{h \ne k} \sum _{h^\prime \ne k} \frac{1}{|h|^{4H} |k-h|^{4H} |h^\prime |^{4H} |k-h^\prime |^{4H}} \\&=C \sum _{k} |k|^{4\rho +4}\left( \sum _{h \ne k} \frac{1}{|h|^{4H} |k-h|^{4H}}\right) ^2 \end{aligned} \end{aligned}$$We handle this contribution according to Lemma 5. Indeed it is bounded by the series
$$\begin{aligned} \sum _{k} |k|^{4\rho +4} \times {\left\{ \begin{array}{ll} M_H\frac{1}{ |k|^{8H-2}} &{} \text { if } \frac{1}{4}< H<\frac{1}{2}\\ M\frac{\ln |k|}{ |k|^{2}} &{} \text { if } H=\frac{1}{2}\\ M_H\frac{1}{ |k|^{4H}} &{} \text { if } H> \frac{1}{2} \end{array}\right. } \end{aligned}$$Thus there is convergence if
$$\begin{aligned} {\left\{ \begin{array}{ll} 8H-2-4\rho -4>2&{} \text { if } \frac{1}{4}< H<\frac{1}{2}\\ 2-4\rho -4>2 &{} \text { if } H=\frac{1}{2}\\ 4H-4\rho -4>2 &{} \text { if } H> \frac{1}{2} \end{array}\right. } \end{aligned}$$These conditions are fulfilled under assumptions (17) and (18).
-
for \(k-h=l^\prime \)
$$\begin{aligned} \begin{aligned}&\sum _{j} |j|^{2\rho } \sum _{h \ne -j} \sum _{l^\prime \ne -h,j} \sum _{h^\prime \ne h+l^\prime } |h+l^\prime |^{2\rho } \gamma _{h,h+l^\prime } \gamma _{h^\prime ,h+l^\prime } \gamma _{-h,j} \gamma _{l^\prime ,j} \\&\quad \int |z_h|^2 |z_{l^\prime }|^2 \overline{z_{h^\prime }}\ \overline{z_{h+l^\prime -h^\prime }} z_{j+h} \ \overline{ z_{j-l^\prime }} \ \mu ^H(\mathrm{d}z) \\&\quad = \sum _{j} |j|^{2\rho } \sum _{h \ne -j} \sum _{l^\prime \ne -h,j} |h+l^\prime |^{2\rho } \gamma _{h,h+l^\prime }( \gamma _{h^\prime ,h+l^\prime }+\gamma _{l^\prime -j,h+l^\prime }) \gamma _{-h,j} \gamma _{l^\prime ,j} \\&\qquad \int |z_h|^2 |z_{l^\prime }|^2 |z_{j+h}|^2 |z_{l^\prime -j}|^2 \ \mu ^H(\mathrm{d}z) \\&\quad \le C \sum _{j} |j|^{2\rho +2} \sum _{h \ne -j} \sum _{l^\prime \ne -h,j} \frac{|h+l^\prime |^{2\rho +2}}{|h|^{4H} |l^\prime |^{4H} |j+h|^{4H} |l^\prime -j|^{4H}}. \end{aligned} \end{aligned}$$The convergence of this sums is done in Lemma 7 in Appendix.
-
For \(k-h=j-l^\prime \), we proceed as in the previous case.
-
-
4.
When \(h=l-j\) we get
$$\begin{aligned}&\sum _{k}\sum _{j} |k|^{2\rho }|j|^{2\rho } \sum _{h\ne k,-j}\sum _{h^\prime \ne k} \sum _{l^\prime \ne j} \gamma _{h,k} \gamma _{h^\prime ,k} \gamma _{h+j,j} \gamma _{l^\prime ,j} \\&\quad \int |z_h|^2 z_{k-h} \overline{z_{h^\prime }}\ \overline{z_{k-h^\prime }} z_{h+j} \overline{ z_{l^\prime }}\ \overline{z_{j-l^\prime }} \ \mu ^H(\mathrm{d}z) \end{aligned}$$Estimating the \(\gamma \)’s (that is \(|\gamma _{\cdot ,k}|\le C|k|\)), we get the same computations as in step 3.
-
5.
When \(h=l^\prime \), we get the same computations as in step 3.
-
6.
When \(h=j-l^\prime \), we get the same computations as in step 3.
This concludes the proof of (19) for \(m=2\).
Rights and permissions
About this article
Cite this article
Ferrario, B., Olivera, C. 2D Navier–Stokes equation with cylindrical fractional Brownian noise. Annali di Matematica 198, 1041–1067 (2019). https://doi.org/10.1007/s10231-018-0809-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10231-018-0809-x
Keywords
- Stochastic partial differential equation
- Navier–Stokes equations
- Cylindrical fractional Brownian motion