Abstract
We study a nonlinear Schrödinger equation with a linear damping, i.e. a zero-order dissipation, and an additive noise. Working in \({\mathbb {R}}^d\) with \(d\le 3\), we prove the uniqueness of the invariant measure when the damping coefficient is sufficiently large.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The nonlinear Schrödinger equation occurs as a basic model in many areas of physics: hydrodynamics, plasma physics, optics, molecular biology, chemical reaction, etc. It describes the propagation of waves in media with both nonlinear and dispersive responses.
In this article, we investigate the long-time behaviour of the following stochastic nonlinear Schrödinger equation
The unknown is \(u:{\mathbb {R}}^d\rightarrow {\mathbb {C}}\). We consider \(\sigma >0\), \(\lambda >0\) and \(\alpha \in \{-1,1\}\); for \(\alpha =1\) this is called the focusing equation and for \(\alpha =-1\) this is the defocusing one. In the r.h.s., there is a stochastic forcing term, which is white in time and coloured in space.
Imposing some condition on the power, i.e. on \(\sigma \), many results are known about existence and uniqueness of solutions, in different spatial domains and with different noises; see [1,2,3, 8, 11, 13, 14]. Basically these results are obtained without damping, i.e. \(\lambda =0\), but can be easily extended to the case with \(\lambda >0\).
When there is no damping and no forcing term (i.e. \(\lambda =0\) and \(\Phi =0\)), the Schrödinger equation is conservative. However, with a noise and a damping term, we expect that the energy injected by the noise is dissipated by the damping term; because of this balance it is meaningful to look for stationary solutions or invariant measures. Ekren et al. [16] and Kim [22] provide the existence of invariant measures of the Eq. (1.1) for any damping coefficient \(\lambda >0\); see also the more general setting of [7] for the two-dimensional case in a different spatial domain and with multiplicative noise and the book [20] for the numerical analysis approach. Notice that the damping \(\lambda u\) is weaker than the dissipation given by a Laplacian \(-\lambda \Delta u\); for this reason we say that \(\lambda u\) is a zero-order dissipation. This implies that the results of existence or uniqueness of invariant measures for the damped Schrödinger equation are less easy than for the stochastic parabolic equations (see, for example, [12]). A similar issue appears in the stochastic damped 2D Euler equations, for which the existence of invariant measures has been recently proven in [5]; there again the difficulty comes from the absence of a strong dissipation, given by the Laplacian in the Navier–Stokes equations.
Let us point out that the existence of invariant measures depends on the damping term as well as on the forcing term. On the other hand, without the damping term it is well known that the stochastic Schrödinger equation has a different long-time behaviour; in [19] it is proved that stochastic solutions may scatter at large time in the subcritical or defocusing case.
The question of the uniqueness of invariant measures is quite challenging for the SPDE’s with a zero-order dissipation. Debussche and Odasso [15] proved the uniqueness of the invariant measure for the cubic focusing Schrödinger equation (1.1), i.e. \(\sigma =\alpha =1\), when the spatial domain is a bounded interval; however, no uniqueness results are known for larger dimension. For the one-dimensional stochastic damped periodic KdV equation, there is a recent result by Glatt-Holtz et al. [17]. However, for nonlinear SPDE’s of parabolic type, i.e. with a stronger dissipation term, the uniqueness issue has been solved in many cases; see, for example, the book [12] by Da Prato and Zabczyk, and the many examples in the paper [18] by Glatt-Holtz, Mattingly and Richards, dealing with the coupling technique. Let us point out that the coupling technique allows for the uniqueness result without restriction on the damping parameter \(\lambda \) but all the examples solved so far are set in a bounded spatial domain and not in \({\mathbb {R}}^d\).
The aim of our paper is to investigate the uniqueness of the invariant measures for Eq. (1.1) in \({\mathbb {R}}^d\) in dimension \(d\le 3\), with some restrictions on the nonlinearity when \(d=3\). However, our technique fails for larger dimension. Notice that also the results for the attractor in the deterministic setting are known for \(d\le 3\) (see [23]). Our main result is Theorem 5.1; it provides a sufficient condition to get the uniqueness of the invariant measure. This condition (5.1) involves \(\lambda \) and the intensity of the noise; to optimize it, in Sects. 2.2 and 3 we perform a detailed analysis on how the solution depends on the damping parameter \(\lambda \).
As far as the contents of this paper are concerned, in Sect. 2 we introduce the mathematical setting and refine known moments estimates on the solution; in Sect. 3 by means of the Strichartz estimates we prove a regularity result on the solutions for \(d=2\) and \(d=3\); this will allow to prove in Sect. 4 that the support of any invariant measure is contained in \(V\cap L^\infty ({\mathbb {R}}^d)\) and some estimates of the moments are given. Finally, Sect. 5 presents the uniqueness result. The four appendices contain auxiliary results.
2 Assumptions and basic results
For \(p\ge 1\), \(L^p({\mathbb {R}}^d)\) is the classical Lebesgue space of complex-valued functions, and the inner product in the real Hilbert space \(L^2({\mathbb {R}}^d)\) is denoted by
We consider the Laplace operator \(\Delta \) as a linear operator in \(L^2({\mathbb {R}}^d)\); so
are nonnegative linear operators and \(\{e^{{i}tA_0 }\}_{t \in {\mathbb {R}}}\) is a unitary group in \(L^2({\mathbb {R}}^d)\). Moreover, for \(s\ge 0\) we consider the power operator \(A_1^{s/2}\) in \(L^2({\mathbb {R}}^d)\) with domain \(H^{s}=\{u \in L^2({\mathbb {R}}^d): \Vert A_1^{s/2}u\Vert _{L^2({\mathbb {R}}^d)}<\infty \}\). Our two main spaces are \(H:=L^2({\mathbb {R}}^d)\) and \(V:=H^1({\mathbb {R}}^d)\). We set \(H^{-s}({\mathbb {R}}^d)\) for the dual space of \(H^s({\mathbb {R}}^d)\) and denote again by \(\langle \cdot ,\cdot \rangle \) the duality bracket.
We define the generalized Sobolev spaces \(H^{s,p}({\mathbb {R}}^d)\) with norm given by \(\Vert u\Vert _{H^{s,p}({\mathbb {R}}^d)}=\Vert A_1^{s/2}u\Vert _{L^p({\mathbb {R}}^d)}\). We recall the Sobolev embedding theorem, see, for example, [4, Theorem 6.5.1]: if \(1<q<p<\infty \) with
then the following inclusion holds
and there exists a constant C such that \(\Vert u\Vert _{H^{s,p}(\mathbb R^d)}\le C \Vert u\Vert _{H^{r,q}({\mathbb {R}}^d)}\) for all \(u \in H^{r,q}({\mathbb {R}}^d)\).
Remark 2.1
For \(d=1\), the space V is a subset of \(L^\infty ({\mathbb {R}})\) and is a multiplicative algebra. This simplifies the analysis of the Schrödinger equation (1.1). However, for \(d\ge 2\) the analysis is more involved.
We write the nonlinearity as
Lemma C.1 provides a priori estimates on it.
As far as the stochastic term is concerned, we consider a real Hilbert space U with an orthonormal basis \(\{e_j\}_{j\in {\mathbb {N}}}\) and a complete probability space \((\Omega ,{\mathcal {F}},{\mathbb {P}})\). Let W be a U-canonical cylindrical Wiener process adapted to a filtration \({\mathbb {F}}\) satisfying the usual conditions. We can write it as a series
with \(\{W_j\}_j\) a sequence of i.i.d. real Wiener processes (see, for example, [12]). Hence,
for a given linear operator \(\Phi :U\rightarrow V\).
Now, we rewrite the Schrödinger equation (1.1) in the abstract form as
We work under the following assumptions on the noise and the nonlinearity. The initial data \(u_0\) is assumed to be in V.
Assumption 2.2
(on the noise) We assume that \(\Phi :U\rightarrow V\) is a Hilbert–Schmidt operator, i.e.
This means that
and it implies that the series (2.2) converges in V.
In order to compare our setting with the more general one of our previous paper [7] in the two-dimensional setting, we point out that \(\Phi \) is also a Hilbert–Schmidt operator from U to H (and we denote \(\Vert \Phi \Vert _{L_\textrm{HS}(U;H)} := \left( \sum \nolimits _{j \in {\mathbb {N}}}\Vert \Phi e_j\Vert ^2_H\right) ^{1/2}\)) and, for \(d=2\), a \(\gamma \)-radonifying operator from U to \(L^{p}({\mathbb {R}}^2)\) for any finite p.
Assumption 2.3
(on the nonlinearity (2.1))
-
If \(\alpha =1\) (focusing), let \(0\le \sigma <\frac{2}{d}\).
-
If \(\alpha =-1\) (defocusing), let \({\left\{ \begin{array}{ll}0\le \sigma < \frac{2}{d-2},&{} \text { for }d\ge 3\\ \sigma \ge 0, &{} \text { for }d\le 2\end{array}\right. }\)
We recall the continuous embeddings
Hence, for \(\sigma \) chosen as in Assumption 2.3 there is the continuous embedding
Moreover, if \(\sigma d<2(\sigma +1)\), the following Gagliardo–Nirenberg inequality holds
In particular, this holds for the values of \(\sigma \) specified in Assumption 2.3. In the focusing case, thanks to the Young inequality for any \(\epsilon >0\) there exists \(C_\epsilon >0\) such that
Above we denoted by C a generic positive constant which might vary from one line to the other, except G which is the particular constant in the next inequality (2.13) coming from the Gagliardo–Nirenberg inequality. Moreover, we shall use this notation: if \(a,b\ge 0\) satisfy the inequality \(a \le C_A b\) with a constant \(C_A>0\) depending on the expression A, we write \(a \lesssim _A b\); for a generic constant we put no subscript. If we have \(a \lesssim _A b\) and \(b \lesssim _A a\), we write \(a \simeq _A b\).
We recall the classical invariant quantities for the deterministic unforced Schrödinger equation (\(\lambda =0\), \(\Phi =0\)), the mass and the energy (see [9]):
They are both well defined on V, thanks to (2.5).
Remark 2.4
In the defocusing case \(\alpha =-1\), we have
and
In the focusing case \(\alpha =1\), the energy has no positive sign, but we can modify it by adding a term and recover the sign property. We introduce the modified energy
where G is the constant appearing in the following particular form of (2.7)
Even if G depends on \(\sigma \) and d, for short we shall write simply G. Moreover, by Assumption 2.3 we have that \(2-\sigma >0\) in the focusing case.
Therefore,
Moreover, from the definition (2.12) and the continuous embedding (2.5) we get
Next in Sect. 2.1 we recall the known results on solutions and invariant measures; then in Sect. 2.2 we provide the improved estimates for the mass and the energy.
2.1 Basic results
We recall from [14] the basic results on the solutions; for any \(u_0\in V\) there exists a unique global solution \(u=\{u(t;u_0)\}_{t\ge 0}\), which is a continuous V-valued process. Here, uniqueness is meant as pathwise uniqueness. Actually, their result is given without damping but one can easily pass from \(\lambda =0\) to any \(\lambda >0\). Let us state the result from De Bouard and Debussche [14].
Theorem 2.5
Under Assumptions 2.2 and 2.3, for every \(u_0\in V\) there exists a unique V-valued and continuous solution of (2.3). This is a Markov process in V. Moreover, for any finite \(T>0\) and integer \(m\ge 1\) there exist positive constants \(C_1\) and \(C_2\) (depending on T, m and \(\Vert u_0\Vert _V\)) such that
and
We notice that the last estimate can be generalized to consider any power \(m>1\) of the energy in the defocusing case as well as for the modified energy in the focusing case, namely
and
This provides
These estimates are in [16].
As soon as a unique solution in V is defined, we can introduce the Markov semigroup. Let us denote by u(t; x) the solution evaluated at time \(t>0\), with initial value \(x\in V\). We define
for any Borelian and bounded function \(f:V\rightarrow {\mathbb {R}}\).
A probability measure \(\mu \) on the Borelian subsets of V is said to be an invariant measure for (2.3) when
We recall Theorem 3.4 from [16] on existence of invariant measures.
Theorem 2.6
Under Assumptions 2.2 and 2.3, there exists an invariant measure supported in V.
2.2 Mean estimates
In this section, we revise some bounds for \(t\in [0,\infty )\) on the moments of the mass, the energy and the modified energy, in order to see how these quantities depend on the damping coefficient \(\lambda \). This improves the results by [16, Lemma 5.1]. Actually, their Lemma 5.1 has to be modified in the focusing case (see Proposition 2.8).
This is the result for the mass \({\mathcal {M}}(u)=\Vert u\Vert ^2_H\).
Proposition 2.7
Let \(u_0\in V\). Then under Assumptions 2.2 and 2.3, for every \(m\ge 1\) there exists a positive constant C (depending on m) such that
for any \(t\ge 0\).
Proof
Let us start by proving the estimate (2.21) for \(m=1\). We apply the Itô formula to \({\mathcal {M}}(u(t))\) (see [16, Theorem 3.2])
Taking the expected value and using the fact that the stochastic integral is a martingale by Theorem 2.5, we obtain, for any \(t \ge 0\),
Solving this ODE, we obtain
which proves (2.21) for \(m=1\).
For \(m\ge 2\), we apply the Itô formula to \({\mathcal {M}}(u(t))^{m}\)
With the Young inequality, we get
for any \(\epsilon >0\). Hence,
By Theorem 2.5, we know that the stochastic integral in (2.23) is a martingale, so taking the expected value on both sides of (2.23) we obtain
Choosing \(\epsilon =1\), by means of Gronwall inequality we get
for any \(t\ge 0\).
If \(1<m<2\), then we use the Hölder inequality and the estimate for \(m=2\):
\(\square \)
Notice that the estimates on the mass do not depend on \(\alpha \), whereas this happens in the next result concerning the energy \({\mathcal {H}}(u)\) given in (2.9) and the modified energy \(\tilde{{\mathcal {H}}}(u)\) given in (2.12). We introduce the functions
and
Both mappings \(\lambda \mapsto \phi _i(\sigma ,\lambda ,\Phi )\) are strictly decreasing. The estimates for the energy in the defocusing case will depend on \(\phi _1\), and the estimates for the modified energy in the focusing case will depend on \(\phi _2\).
This is the result on the power moments of \({\mathcal {H}}(u)\) and \(\tilde{{\mathcal {H}}}(u)\).
Proposition 2.8
Let \(u_0\in V\). Under Assumptions 2.2 and 2.3, we have the following estimates:
- (i):
-
When \(\alpha =-1\), for every \(m\ge 1\) there exists a positive constant \(C=C(d, \sigma ,m)\) such that
$$\begin{aligned} {\mathbb {E}} [{\mathcal {H}}(u(t))^m] \le e^{-\lambda m t}{\mathcal {H}}(u_0)^m+ C \phi _1^{m}\lambda ^{-m} \end{aligned}$$(2.26)for any \(t\ge 0\).
- (ii):
-
When \(\alpha =1\), for every \(m\ge 1\) there exist positive constants \(a=a(d,\sigma )\), \(C_1=C(d, \sigma ,m)\) and \(C_2=C(d, \sigma ,m)\) such that
$$\begin{aligned}{} & {} {\mathbb {E}} [\tilde{{\mathcal {H}}}(u(t))^m] \le e^{-m \frac{2-\sigma }{2+\sigma } \lambda t} \tilde{{\mathcal {H}}}(u_0)^m\nonumber \\{} & {} \quad +C_1 e^{-m a \lambda t} [1+ \mathcal M(u_0)^{m(\frac{1}{2}+\frac{2\sigma }{2-\sigma d})}] \Vert \Phi \Vert ^{m}_{L_\textrm{HS}(U;V)} \lambda ^{-\frac{m}{2}} +C_2 \phi _2^m \lambda ^{-m} \end{aligned}$$(2.27)for any \(t\ge 0\).
Proof
The Itô formula for \({\mathcal {H}}(u(t))\) is (see Theorem 3.2 in [16])
We notice that the stochastic integral is a martingale, because its quadratic variation has finite mean thanks to the moment estimates (2.16)–(2.18). (Computations are similar to those in the next estimate (2.34).)
Below we repeatedly use the Hölder and Young inequalities. In particular,
and
for any positive \(A,B,\lambda , \epsilon \).
\(\bullet \) In the defocusing case \(\alpha =-1\), we neglect the first term in the r.h.s. in (2.28), i.e.
Moreover, thanks to Assumption 2.3 we use Hölder and Young inequalities to get
Now, we insert this estimate in (2.31) and take the mathematical expectation to get rid of the stochastic integral
i.e.
By Gronwall lemma, we get
for any \(t\ge 0\). This proves (2.26) for \(m=1\).
For higher powers \(m \ge 2\), by means of Itô formula we get
We estimate the latter term using the Hölder and the Young inequality:
for any \(\epsilon >0\). Inserting in (2.33) and using the Young inequality (2.30), we get
We estimate \({\mathcal {H}}(u(t))^{m-1} \textrm{d} {\mathcal {H}}(u(t))\) using (2.31), (2.32), and the Young inequality (2.29). Then, we take the mathematical expectation in (2.35) and obtain
By Gronwall lemma, we get (2.26).
For \(1<m<2\), we proceed by means of the Hölder inequality as before, using the estimate for \(m=2\).
\(\bullet \) In the focusing case \(\alpha =1\), we neglect the last two terms in the r.h.s. in (2.28) and get
We write the Itô formula for the modified energy \(\tilde{{\mathcal {H}}}(u)= {{\mathcal {H}}}(u) +G\mathcal M(u)^{1+\frac{2\sigma }{2-\sigma d}}\). Proceeding as in (2.23) for the power \(m=1+\frac{2\sigma }{2-\sigma d}\) of the mass, we have
Since \((1-\frac{2}{2-\sigma d})< 0\) by Assumption 2.3, for \(\epsilon \) small enough we get \(\epsilon (1+\frac{2\sigma }{2-\sigma d})+2\sigma (1-\frac{2}{2-\sigma d})<0 \); hence,
Then,
Notice that the condition \(\sigma <\frac{2}{d}\) implies that \(\sigma <2\). So considering the mathematical expectation, we obtain
By means of the Gronwall lemma, we get
This proves (2.27) for \(m=1\).
For \(m\ge 2\), we have by Itô formula
where we have estimated the quadratic variation of the stochastic integral in (2.38) so to get
Keeping in mind the previous estimate (2.34), we get
Now to estimate the first term in the r.h.s., we use (2.14), i.e. \(\Vert \nabla u(t)\Vert _H^2 \le 4 \tilde{{\mathcal {H}}}(u)\), and for the second term by means of (2.7), we get
for any \(\epsilon >0\). Thus, we estimate the latter term in (2.41) as follows
and by Young inequality
In (2.41), we insert this estimate and the previous estimate (2.39) for \(d \tilde{{\mathcal {H}}}(u(t))\), integrate in time, and take the mathematical expectation to get rid of the stochastic integrals; hence, for \(\epsilon \) small enough we obtain
We use Gronwall lemma and bearing in mind the estimates (2.21) for the mass we get an inequality for \({\mathbb {E}} [\tilde{{\mathcal {H}}}(u(t))^m]\). Computing the time integrals appearing there, with some elementary calculations we get
for any \(t\ge 0\), where
Since \(\frac{2\sigma +1}{\sigma +1}(1+\frac{2\sigma }{2-\sigma d})\frac{1}{2} <\frac{1}{2}+\frac{2\sigma }{2-\sigma d}\), we bound the sum of the two terms with different powers of \({\mathcal {M}}(u_0)\) by putting in evidence only the largest power. Therefore, we obtain (2.27).
For \(1<m<2\), we proceed as in the previous case. \(\square \)
Merging the results for the mass and the energy, we obtain the result for the V-norm. Indeed, \(\Vert u\Vert _V^2=\Vert \nabla u\Vert _H^2+\Vert u\Vert _H^2\) and
For \(\alpha =-1\), we trivially get
For \(\alpha =1\), we have from (2.14)
Now, we bear in mind the functions \(\phi _1\) and \(\phi _2\) given in (2.24) and (2.25), respectively. This is the result for the moments of the V-norm.
Corollary 2.9
Let \(u_0\in V\). Under Assumptions 2.2 and 2.3, for every \(m\ge 1\) we have the following estimates:
- (i):
-
When \(\alpha =-1\)
$$\begin{aligned} {\mathbb {E}}[ \Vert u(t)\Vert _V^{2m}] \lesssim e^{-m \lambda t}[{{\mathcal {H}}}(u_0)^m +{{\mathcal {M}}}(u_0)^m]+ [\phi _1 + \Vert \Phi \Vert ^{2}_{L_\textrm{HS}(U;H)}]^m \lambda ^{-m}\nonumber \\ \end{aligned}$$(2.42)for any \(t\ge 0\);
- (ii):
-
When \(\alpha =1\), there is a positive constant \(a=a(d,\sigma )\) such that
$$\begin{aligned}{} & {} {\mathbb {E}}[ \Vert u(t)\Vert _V^{2m}] \lesssim e^{-m \frac{2-\sigma }{2+\sigma } \lambda t} \tilde{{\mathcal {H}}}(u_0)^m +e^{-m \lambda t} {{\mathcal {M}}}(u_0)^m\nonumber \\{} & {} \quad +e^{-m a \lambda t} [1+ \mathcal M(u_0)^{m(\frac{1}{2}+\frac{2\sigma }{2-\sigma d})}] \Vert \Phi \Vert ^{m}_{L_\textrm{HS}(U;V)} \lambda ^{-\frac{m}{2}} +[\phi _2 + \Vert \Phi \Vert ^{2}_{L_\textrm{HS}(U;H)}]^m \lambda ^{-m}\nonumber \\ \end{aligned}$$(2.43)for any \(t\ge 0\).
The constants providing the above estimates (\(\lesssim \)) depend on \(m,\sigma \) and d but not on \(\lambda \).
3 Regularity results for the solution
For the solution of Eq. (2.3), we know that \(u\in C([0,+\infty );V)\) a.s. if \(u_0\in V\). Now, we look for the \(L^\infty ({\mathbb {R}}^d)\)-space regularity of the paths. When \(d=1\), this follows directly from the Sobolev embedding \(H^1(\mathbb R)\subset L^\infty ({\mathbb {R}})\). But such an embedding does not hold for \(d>1\). However, for \(d=2\) or \(d=3\) one can obtain the \(L^\infty ({\mathbb {R}}^d)\)-regularity by means of the deterministic and stochastic Strichartz estimates of Appendix A.
Let \(\phi _1\) and \(\phi _2\) be the functions appearing in Proposition 2.8.
Proposition 3.1
Let \(d=2\) or \(d=3\). In addition to Assumptions 2.2 and 2.3, we suppose that \(\sigma <\frac{1+\sqrt{17}}{4}\) when \(d=3\).
Given any finite \(T>0\) and \(u_0\in V\), the solution of Eq. (2.3) is in \( L^{2\sigma }(\Omega ;L^{2\sigma }(0,T;L^\infty ({\mathbb {R}}^d)))\). Moreover, there exists a positive constant \(C=C(\sigma ,d, T)\) such that
where
and
so \(\lambda \mapsto \phi _3(d, \sigma ,\lambda , \Phi )\) is a strictly decreasing function.
Proof
First let us consider \(d=2\). We repeatedly use the embedding \(H^{1,q}({\mathbb {R}}^2)\subset L^\infty ({\mathbb {R}}^2)\) valid for any \(q>2\). So our target is to prove the estimate for the \(L^{2\sigma }(\Omega ;L^{2\sigma }(0,T;H^{1,q}({\mathbb {R}}^2)))\)-norm of u for some \(q>2\).
We introduce the operator \(\Lambda :=-iA_0+\lambda \). It generates the semigroup \(e^{-i\Lambda t}= e^{-\lambda t} e^{iA_0 t}\), \(t\ge 0\).
Let us fix \(T>0\). We write Eq. (2.3) in the mild form (see [14])
and estimate
for some \(q>2\).
For the estimate of \(I_1\), we set
Notice that \(q>2\). Now, before using the homogeneous Strichartz inequality (A.1) we neglect the term \(e^{-\lambda t}\), since \(e^{-\lambda t} \le 1\). First, assuming \(\sigma >1\) we work with the admissible Strichartz pair \((2\sigma ,\frac{2\sigma }{\sigma -1})\) and get
For smaller values, i.e. \(0<\sigma \le 1\), we choose \(\tilde{\sigma }=\frac{3}{\sigma }>2>\sigma \) so \(\frac{2\tilde{\sigma }}{\tilde{\sigma }-1}=\frac{6}{3-\sigma }\) and
by the previous computations.
For the estimate of \( I_2\), we use the Strichartz inequality (A.2) and then the estimate from Lemma C.1 on the nonlinearity. We use the notation \(\gamma ^\prime \) for the conjugate exponent of \(\gamma \in (1,\infty )\), i.e. \(\frac{1}{\gamma }+\frac{1}{\gamma ^\prime }=1\). First, consider \(\sigma >1\); the pair \((2\sigma ,\frac{2\sigma }{\sigma -1})\) is admissible. Then,
For \(0<\sigma \le 1\), we proceed in a similar way; considering the admissible Strichartz pair \((2+\sigma , 2+\frac{4}{\sigma })\), we have
where \((r,\gamma )\) is an admissible Strichartz pair. According to (C.1), we choose
Hence,
In this way by means of the estimate (C.2) of the polynomial nonlinearity \(\Vert F_\alpha (u)\Vert _{H^{1,r^\prime }(\mathbb R^2)}\lesssim \Vert u\Vert ^{1+2\sigma }_V\), we obtain
Summing up, we have shown that for any \(\sigma >0\) there exists \(q>2\) and \(\gamma ^\prime \) such that
Bearing in mind Corollary 2.9, we get the second and third terms in the r.h.s. of (3.1). The details are given in Appendix 5.1.
It remains to estimate the term \( I_3\). We choose q as in (3.5). Using the stochastic Strichartz estimate (A.3), we get for \(\sigma >1\)
For smaller values of \(\sigma \), we proceed as before for \(I_1\).
Now, consider \(d=3\). The additional assumption on \(\sigma \) appears because of the stronger conditions on the parameters given later on.
For \(q\ge 1\), we have \(H^{\theta ,q}({\mathbb {R}}^3)\subset L^\infty ({\mathbb {R}}^3)\) when \(\theta q>3\). So for each \(I_i\) in (3.4) we look for an estimate in the norm \(L^{2\sigma }(0,T; H^{\theta ,q}({\mathbb {R}}^3))\) for some parameters with \(\theta q>3\).
We estimate \( I_1\) for any \(0<\sigma <2\). When \(0<\sigma \le 1\), we consider the admissible Strichartz pair (2, 6). By means of the homogeneous Strichartz estimate (A.1), we proceed as before
When \(\sigma >1\), we work with the admissible Strichartz pair \((2\sigma , \frac{6\sigma }{3\sigma -2})\) and get
since \(\frac{6\sigma }{3\sigma -2}>3\) for \(1<\sigma <2\), we obtain the \(L^\infty ({\mathbb {R}}^3)\)-norm estimate.
The estimate for \(I_2\) is more involved, and we postpone it to Appendix 5.2, where condition (D.1) leads to the upper bound \(\sigma <\frac{1+\sqrt{17}}{4}\).
It remains to estimate the term \( I_3\). For any \(\sigma >0\), we use the Hölder inequality and the stochastic Strichartz estimate (A.3) for the admissible pair \((2+\frac{\sigma ^2}{2},6\frac{4+\sigma ^2}{4+3\sigma ^2})\); therefore,
\(\square \)
Notice that the restriction \(\sigma <\frac{1+\sqrt{17}}{4}\) on the power of the nonlinearity affects only the defocusing case, since by Assumption 2.3 in the focusing case we already require the stronger bound \(\sigma <\frac{2}{3}\) when \(d=3\).
We conclude this section by remarking that there is no similar result for \(d\ge 4\).
Remark 3.2
For larger dimension, there is no result similar to those in this section. Indeed, if one looks for \(u\in L^{2\sigma }(0,T; H^{1,q}({\mathbb {R}}^d))\subset L^{2\sigma }(0,T; L^\infty (\mathbb R^d))\), it is necessary that
in order to have \( H^{1,q}({\mathbb {R}}^d)\subset L^\infty (\mathbb R^d)\). Already the estimate for \(I_1\) does not hold under this assumption. Indeed, the homogeneous Strichartz estimate (A.1) provides
if
Since \(\frac{2d}{d-2}\le 4\) for \(d\ge 4\), the latter condition \(q\le \frac{2d}{d-2}\) and the condition \(q>d\) are incompatible for \(d\ge 4\).
Let us notice that also in the deterministic setting the results on the attractors are known for \(d\le 3\), see [23].
4 The support of the invariant measures
From Theorem 2.6 we know that there exist invariant measures supported on V. Now, we show some more properties on these invariant measures. In dimension \(d=2\) and \(d=3\), thanks to the regularity results of Sect. 3 we provide an estimate for the moments in the V and \(L^\infty ({\mathbb {R}}^d)\)-norm.
Set
The function \(\phi _4=\phi _4(d, \sigma ,\lambda ,\Phi )\) is strictly decreasing w.r.t. \(\lambda \).
Proposition 4.1
Let \(d\le 3\) and Assumptions 2.2 and 2.3 hold.
Let \(\mu \) be an invariant measure for Eq. (2.3), given by Theorem 2.6. Then, for any finite \(m\ge 1\) we have
Moreover, supposing in addition that \(\sigma <\frac{1+\sqrt{17}}{4}\) when \(d=3\), we have
where \(\lambda \mapsto \phi _5(d, \sigma ,\lambda ,\Phi )\) is a smooth decreasing function.
Proof
As far as (4.1) is concerned, we define the bounded mapping \(\Psi _k\) on V as
for \(k \in {\mathbb {N}}\).
By the invariance of \(\mu \) and the boundedness of \(\Psi _k\), we have
So
Moreover, from Corollary 2.9 we get an estimate for \({\mathbb {E}} \Vert u(s;x)\Vert _V^{2m}\), and letting \(s \rightarrow +\infty \) the exponential terms in the r.h.s. of (2.42) and (2.43) vanish so we get
By Fatou lemma, we have that the same holds for the integral, that is
From (4.3), we get
as well. Since \(\Psi _k\) converges pointwise and monotonically from below to \(\Vert \cdot \Vert _V^{2m}\), the monotone convergence theorem yields (4.1).
As far as (4.2) is concerned, for \(d=1\) this is a consequence of estimate (4.1), because of the Sobolev embedding \(H^1({\mathbb {R}})\subset L^\infty ({\mathbb {R}})\). However, for \(d>1\) we consider the estimate (3.1) for \(T=1\) and set \(\tilde{\Psi }( u)= \Vert u\Vert _{L^\infty ({\mathbb {R}}^d)}^{2\sigma }\); this defines a mapping \(\tilde{\Psi }:V \rightarrow {\mathbb {R}}_+ \cup \{+\infty \}\). Its approximation \(\tilde{\Psi }_k:V\rightarrow {\mathbb {R}}_+ \), given by
defines a bounded mapping \(\tilde{\Psi }_k:V\rightarrow {\mathbb {R}}_+ \) for any \(k \in {\mathbb {N}}\).
It obviously holds
By the invariance of \(\mu \) and the boundedness of \(\tilde{\Psi }_k\), it also holds
Thus, by Fubini–Tonelli theorem, since \(\tilde{\Psi }_k(u)=\Vert u\Vert ^{2\sigma }_{L^\infty ({\mathbb {R}}^d)} \wedge k^{2\sigma } \le \Vert u\Vert ^{2\sigma }_{L^\infty ({\mathbb {R}}^d)}\), we get
where we used (3.1) from Proposition 3.1 for \(T=1\).
The integral \( \int _V \Vert x\Vert _V^{2\sigma } {\textrm{d}}\mu (x)\) can be estimated by means of (4.1). The same holds for the integral of the second term, by bearing in mind the expression (3.2) of \(\psi \) and the bounds (2.11), (2.15); let us denote by \(\phi _\psi =\phi _\psi (d, \sigma , \lambda ,\Phi )\) the new function estimating \(\int _V \psi (x)^{\sigma (2\sigma +1)} {\textrm{d}}\mu (x)\).
Therefore, we have proved that
where \(\phi _5\) is proportional to
This holds for any k. Since \(\tilde{\Psi }_k\) converges pointwise and monotonically from below to \(\tilde{\Psi }\), the monotone convergence theorem yields the same bound for \( \int _{V} \Vert x\Vert ^{2\sigma }_{L^\infty ({\mathbb {R}}^d)}\, {\textrm{d}}\mu (x)\). This proves (4.2). \(\square \)
5 Uniqueness of the invariant measure for sufficiently large damping
We will prove that if the damping coefficient \(\lambda \) is sufficiently large, then the invariant measure is unique.
Theorem 5.1
Let \(d\le 3\). In addition to Assumptions 2.2 and 2.3, we suppose that \(\sigma <\frac{1+\sqrt{17}}{4}\) when \(d=3\).
If
where \(\phi _5\) is the function appearing in Proposition 4.1, then for Eq. (2.3) there exists a unique invariant measure supported in V.
Proof
The existence of an invariant measure comes from Theorem 2.6. Now, we prove the uniqueness by means of a reductio ad absurdum. Let us suppose that there exists more than one invariant measure. In particular, there exist two different ergodic invariant measures \(\mu _1\) and \(\mu _2\). For both of them, Proposition 4.1 holds. Fix either \(i=1\) or \(i=2\) and consider any \(f\in L^1(\mu _i)\). Then, by the Birkhoff ergodic theorem (see, for example, [10]) for \(\mu _i\)-a.e. \(x_i\in V\) we have
Here, u(t; x) is the solution at time t, with initial value \(u(0)=x\in V\).
Now, fix two initial data \(x_1\) and \(x_2\) belonging, respectively, to the support of the measure \(\mu _1\) and \(\mu _2\). We have
\({\mathbb {P}}\)-a.s.. Taking any arbitrary f in the set \( \mathcal G_0\) defined in (B.2), we get
If we prove that
then we conclude that
so \(\mu _1=\mu _2\) thanks to Lemma B.1. So let us focus on the limit (5.3).
With a short notation, we write \(u_i(t)=u(t;x_i)\). Then, consider the difference \(w=u_1-u_2\) fulfilling
so
Using the elementary estimate
we bound the nonlinear term in the r.h.s. as
Therefore,
Gronwall inequality gives
that is
This is a pathwise estimate.
We know from Proposition 4.1 that \(f(x)=\Vert x\Vert _{L^\infty ({\mathbb {R}}^d)}^{2\sigma } \in L^1(\mu _i)\); therefore, (5.2) becomes
\({\mathbb {P}}\)-a.s., for either \(i=1\) or \(i=2\). Therefore, if
the exponential term in the r.h.s. of (5.4) vanishes as \(t\rightarrow +\infty \). This proves (5.3) and concludes the proof. \(\square \)
Data availability
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
References
V. Barbu, M. Röckner and D. Zhang. Stochastic nonlinear Schrödinger equations with linear multiplicative noise: rescaling approach. J. Nonlinear Sci., 24 (3):383–409, 2014.
V. Barbu, M. Röckner and D. Zhang. Stochastic nonlinear Schrödinger equations. Nonlinear Analysis: Theory, Methods & Applications, 136:168–194, 2016.
V. Barbu, M. Röckner and D. Zhang. Stochastic nonlinear Schrödinger equations: no blow-up in the non-conservative case. J. Differential Equations, 263 (11):7919–7940, 2017.
J. Bergh and J. Löfström. Interpolation Spaces. An Introduction. Grundlehren der Mathematischen Wissenschaften, vol. 223. Springer-Verlag, Berlin-New York, 1976.
H. Bessaih and B. Ferrario. Invariant measures for stochastic damped 2D Euler equations. Commun. Math. Phys., 377:531–549, 2020.
P. Billingsley, Convergence of probability measures, John Wiley & Sons, 2013.
Z. Brzeźniak, B. Ferrario and M. Zanella. Invariant measures for a stochastic nonlinear and damped 2D Schrödinger equation. arXiv:2106.07043
Z. Brzeźniak, F. Hornung and L. Weis. Martingale solutions for the stochastic nonlinear Schrödinger equation in the energy space. Probab. Theory Related Fields, 174(3-4):1273–1338, 2019.
T. Cazenave. Semilinear Schrödinger Equations, volume 10. American Mathematical Soc., 2003.
I.P. Cornfeld, S.V. Fomin and Ya.G. Sinai. Ergodic Theory, Grundlehren der mathematischen Wissenschaften 245. Springer-Verlag, 1982.
J. Cui, J. Hong and L. Sun. On global existence and blow-up for damped stochastic nonlinear Schrödinger equation. Discrete Contin. Dyn. Syst., Ser. B 24(12): 6837–6854, 2019.
G. Da Prato and J. Zabczyk. Ergodicity for infinite-dimensional systems. London Mathematical Society Lecture Note Series, 229. Cambridge University Press, Cambridge, 1996.
A. De Bouard and A. Debussche. A stochastic nonlinear Schrödinger equation with multiplicative noise. Commun. Math. Phys., 205(1):161–181, 1999.
A. De Bouard and A. Debussche, The Stochastic Nonlinear Schrödinger Equation in \(H^1\). Stoch. Anal. Appl. 21(1): 97-126, 2003.
A. Debussche and C. Odasso. Ergodicity for a weakly damped stochastic non-linear Schrödinger equation. J. Evol. Equ. 5(3):317–356, 2005.
I. Ekren, I. Kukavica and M. Ziane. Existence of invariant measures for the stochastic damped Schrödinger equation. Stoch. Partial Differ. Equ. Anal. Comput. 5(3): 343–367, 2017.
N.E. Glatt-Holtz, V.R. Martinez and G.H. Richards. On the long-time statistical behavior of smooth solutions of the weakly damped, stochastically-driven KdV equation. arXiv:2103.12942v1
N.E. Glatt-Holtz, J. Mattingly and G.H. Richards. On Unique Ergodicity in Nonlinear Stochastic Partial Differential Equations. J. Stat. Phys. 166: 618–649, 2017.
S. Herr, M. Röckner and D. Zhang. Scattering for stochastic nonlinear Schrödinger equations. Comm. Math. Phys. 368(2): 843–884, 2019.
J. Hong and X. Wang. Invariant measures for stochastic nonlinear Schrödinger equations. Numerical Approximations and Symplectic Structures, Springer, 2019.
F. Hornung. The nonlinear stochastic Schrödinger equation via stochastic Strichartz estimates. J. Evol. Equ. 18: 1085–1114, 2018.
J.U. Kim. Invariant Measures for a Stochastic Nonlinear Schrödinger Equation. Indiana Univ. Math. J., 55(2): 687–717, 2006.
P. Laurençot. Long-time behaviour for weakly damped driven nonlinear Schrödinger equations in \({\mathbb{R}}^{N}, N\le 3\).NoDEA 2: 357–369, 1995.
Acknowledgements
We thank an anonymous referee for her/his remarks, which helped improve the presentation of our results. Moreover, B. Ferrario and M. Zanella gratefully acknowledge financial support from GNAMPA-INdAM.
Funding
Open access funding provided by Universitá degli Studi di Pavia within the CRUI-CARE Agreement. Funding was provided by Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
A Strichartz estimates
In this section, we recall the deterministic and stochastic Strichartz estimates on \({\mathbb {R}}^d\).
Definition A.1
We say that a pair (p, r) is admissible if
and
If (p, r) is an admissible pair, then \(2\le p\le \infty \).
Given \(1\le \gamma \le \infty \), we denote by \(\gamma ^\prime \) its conjugate exponent, i.e. \(\frac{1}{\gamma }+\frac{1}{\gamma ^\prime }=1\).
Lemma A.1
Let (p, r) be an admissible pair of exponents. Then, the following properties hold
- (i):
-
For every \(\varphi \in L^2({\mathbb {R}}^d)\), the function \(t \mapsto e^{itA_0}\varphi \) belongs to \(L^p({\mathbb {R}};L^r({\mathbb {R}}^d)) \cap C({\mathbb {R}} ;L^2({\mathbb {R}}^d))\). Furthermore, there exists a constant C such that
$$\begin{aligned} \Vert e^{i \cdot A_0}\varphi \Vert _{L^p({\mathbb {R}};L^r({\mathbb {R}}^d))} \le C\Vert \varphi \Vert _{L^2({\mathbb {R}}^d)}, \qquad \forall \varphi \in L^2({\mathbb {R}}^d). \end{aligned}$$(A.1) - (ii):
-
Let I be an interval of \({\mathbb {R}}\) and \(0\in J ={\overline{I}}\). If \((\gamma ,\rho )\) is an admissible pair and \(f\in L^{\gamma ^\prime }(I;L^{\rho ^\prime }({\mathbb {R}} ^d))\), then the function \(t \mapsto G_f(t)=\int _0^t e^{i(t-s)A_0} f(s) \textrm{d}s\) belongs to \(L^q(I;L^r({\mathbb {R}}^d)) \cap C(J ;L^2({\mathbb {R}}^d))\). Furthermore, there exists a constant C, independent of I, such that
$$\begin{aligned} \Vert G_f\Vert _{L^p(I;L^r({\mathbb {R}}^d)) }\le C \Vert f\Vert _{L^{\gamma ^\prime }(I;L^{\rho ^\prime }({\mathbb {R}} ^d))} , \qquad \forall f \in L^{\gamma ^\prime }(I;L^{\rho ^\prime }({\mathbb {R}} ^d)). \end{aligned}$$(A.2)
Proof
See [9, Proposition 2.3.3]. \(\square \)
Lemma A.3
(stochastic Strichartz estimate) Let (p, r) be an admissible pair. Then, for any \(a\in (1,\infty )\) and \(T<\infty \) there exists a constant C such that
for any \(\Psi \in L^2(0,T;L_\textrm{HS}(U,L^2({\mathbb {R}}^d)))\).
Proof
See [21, Proposition 2]. \(\square \)
B Determining sets
The set
is a determining set for measures on V (see, for example, [6, Theorem 1.2]). This means that given two probability measures \(\mu _1\) and \(\mu _2\) on V we have
Following Remark 2.2 in [18], we can consider as a determining set for measures on V the set
involving the weaker H-norm instead of the V-norm. So \(\mathcal G_0\subset {\mathcal {G}}_1\). Let us show that \({\mathcal {G}}_0\) is a determining set for measures on V as well.
Lemma B.1
Let \(\mu _1\) and \(\mu _2\) be two invariant measures. If
then \(\mu _1=\mu _2\).
Proof
We show the proof since we work in \({\mathbb {R}}^d\), whereas [18] deals with a bounded domain.
Set \(P_Nx \) to be the element whose Fourier transform is \(1_{|\xi |\le N}{\mathcal {F}}(x)\); hence, \(\Vert P_N x\Vert _V\le \sqrt{1+N^2}\Vert x\Vert _H\). Now, we show that any function \(f\in \mathcal G_1\) can be approximated by a function \(f_N\in {\mathcal {G}}_0\) by setting \(f_N(x)=f(P_N x)\). Indeed,
By assumption, we know that
Taking the limit as \(N\rightarrow +\infty \), by the bounded convergence theorem we get the same identity for \(f \in {\mathcal {G}}_1\). Hence, \(\mu _1=\mu _2\). \(\square \)
C Estimate of the nonlinearity
We consider \(F(u)=|u|^{2\sigma }u\).
Lemma C.1
Let \(d=2\). For any \(\sigma >0\), if \(p\in (1,2)\) is defined as
then
Let \(d=3\). For any \(\sigma \in (0,\frac{3}{2}]\), we have
and for any \(\sigma \in [1,\frac{3}{2}]\) we have
Proof
We start with the case \(d=2\). To estimate the \(H^{1,p}\)-norm of F, it is enough to deal with \(\Vert F\Vert _{L^p({\mathbb {R}}^d)}\) and \(\Vert \partial F\Vert _{L^p({\mathbb {R}}^d)}\). We compute
and thus, \(|\partial F(u)| \lesssim _{\sigma }|u|^{2\sigma } |\partial u|\).
We have
and the Hölder inequality, for \(1\le p<2\), gives
We recall the Sobolev embedding
Therefore, if
then both the r.h.s. of (C.6) and (C.7) can be estimated by a quantity involving the \(H^1({\mathbb {R}}^2)\)-norm. The two latter inequalities are the same as
so one easily sees that the choice (C.1) allows to fulfil the two required estimates, i.e. \(\Vert F(u)\Vert _{L^p(\mathbb R^d)}\lesssim \Vert u\Vert _V^{2\sigma +1}\) and \(\Vert \partial F(u)\Vert _{L^p({\mathbb {R}}^d)}\lesssim \Vert u\Vert _V^{2\sigma +1}\). This proves (C.2).
For \(d=3\), first we show that for any \(\sigma \in (0,\frac{3}{2}]\)
with \(p=\frac{6}{2 \sigma +3}\in [1,2)\).
To this end, we notice that the r.h.s. of (C.6) and (C.7) is estimated by a quantity involving the \(H^1({\mathbb {R}}^3)\)-norm if \( H^1({\mathbb {R}}^3)\subset L^{(2\sigma +1)p}({\mathbb {R}}^3)\) and \(H^1({\mathbb {R}}^3)\subset L^{\frac{4 \sigma p}{2-p}}({\mathbb {R}}^3)\). Recalling the Sobolev embedding
we get the conditions
Notice that (C.10) is stronger than (C.9); moreover, (C.10) has a solution \(p \in [1,2)\) only if \(\sigma \in (0,\frac{3}{2}]\). Choosing
we fulfil all the requirements and so we have proved (C.3).
Now for \(1\le \sigma \le 2\) there is the continuous embedding \(H^{1,\frac{6}{2 \sigma +3}}({\mathbb {R}}^3)\subseteq H^{2-\sigma ,\frac{6}{5}}({\mathbb {R}}^3)\). Hence, from (C.3) we get (C.4). \(\square \)
D Computations in the proof of Proposition 3.1
1.1 From (3.8) to (3.1)
From (3.8), we proceed as follows. We distinguish different values of the parameter \(\sigma \).
\(\bullet \sigma \in (0,\frac{1}{4})\): we have \(\gamma ^\prime =\frac{1}{1-\sigma }\), so \(\frac{2\sigma }{\gamma ^\prime }=2\sigma (1-\sigma )<\frac{3}{8}\) and \(\gamma ^\prime (2\sigma +1)=\frac{2\sigma +1}{1-\sigma }<2\). With the Hölder inequality twice
We conclude by means of the estimate of Corollary 2.9 for \(m=1\); for instance, in case (i)
where we used \(\frac{1-e^{-\lambda T}}{\lambda }\le T\).
\(\bullet \sigma \in [\frac{1}{4}, \frac{1}{2})\): we have \(\gamma ^\prime =\frac{1}{1-\sigma }\), so \(\frac{2\sigma }{\gamma ^\prime }=2\sigma (1-\sigma )\le \frac{1}{2}\) and \(\gamma ^\prime (2\sigma +1)=\frac{2\sigma +1}{1-\sigma }\ge 2\). With the Hölder inequality
and then we conclude by means of the estimate of Corollary 2.9 for \(2m=\gamma ^\prime (2\sigma +1)\); for instance, in case (ii)
where we used \(\int _0^T e^{-bt}\textrm{d}t=\frac{1-e^{-bT}}{b}\le T\) for any \(b>0\).
\(\bullet \sigma \in [\frac{1}{2},\frac{2}{3})\): we have \(\gamma ^\prime =\frac{4}{3}\), so \(\frac{2\sigma }{\gamma ^\prime }=\frac{3}{2} \sigma <1\) and \(\gamma ^\prime (2\sigma +1)=\frac{4}{3}(2\sigma +1) \ge \frac{8}{3}\). So we proceed as in the previous case.
\(\bullet \sigma \ge \frac{2}{3}\): we have \(\gamma ^\prime =\frac{4}{3}\), so \(\frac{2\sigma }{\gamma ^\prime }\ge 1\) and \(\gamma ^\prime (2\sigma +1)\ge \frac{28}{9}\). With the Hölder inequality
and then we conclude by means of the estimate of Corollary 2.9 for \(2m=2\sigma (2\sigma +1)\).
1.2 Estimate of \(I_2\) when \(d=3\)
We distinguish two ranges of values for \(\sigma \).
\(\bullet \) For \(0<\sigma \le 1\) we have \(L^2(0,T;L^6(\mathbb R^3))\subseteq L^{2\sigma }(0,T;L^6({\mathbb {R}}^3))\). So we consider the admissible Strichartz pair (2, 6) and get for any admissible Strichartz pair \((\gamma ,r)\)
The parameters are such that \(\gamma ^\prime =\frac{4r^\prime }{7r^\prime -6}\). From Definition A.1, we have the condition \(2\le r\le 6\), equivalent to \(\frac{6}{5}\le r^\prime \le 2\). Choosing
we have \( r^\prime \in [\tfrac{6}{5},2)\) when \(0<\sigma \le 1\) and \(\gamma ^\prime =\frac{2}{2-\sigma }\in (1,2]\); thus, we can use (C.3) to estimate the nonlinearity \(F_\alpha (u)\). Summing up, we have
Hence,
by Hölder inequality since \(\sigma (2-\sigma )\le 1\).
From here, bearing in mind Corollary 2.9 we conclude as in the previous subsection and we obtain the second and third terms in the r.h.s. of (3.1).
\(\bullet \) For \( \sigma >1\), we use the admissible Strichartz pair \((2\sigma ,\frac{6\sigma }{3\sigma -2})\) so
where we used (A.2) with \(\gamma ^\prime =2\) and \(\rho ^\prime =\frac{6}{5}\), corresponding to the admissible Strichartz pair \((\gamma ,\rho )\) with \(\gamma =2\) and \(\rho =6\). Notice that \(\frac{6}{5}\) is the minimal allowed value for \(\rho ^\prime \) when \(d=3\).
Now, assuming \(1<\sigma \le \frac{3}{2}\) we use the estimate (C.4) with \(\theta =2-\sigma \). Summing up, we obtain
When
we have \(H^{2-\sigma ,\frac{6\sigma }{3\sigma -2}}({\mathbb {R}}^3)\subset L^{\infty }({\mathbb {R}}^3)\). This gives the condition \( \sigma <\frac{1+\sqrt{17}}{4}\). Hence,
Since \(\sigma >1\), we conclude with the Hölder inequality that
Finally, we obtain the second and third term of (3.1) by means of Corollary 2.9 as before.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Brzeźniak, Z., Ferrario, B. & Zanella, M. Ergodic results for the stochastic nonlinear Schrödinger equation with large damping. J. Evol. Equ. 23, 19 (2023). https://doi.org/10.1007/s00028-023-00870-6
Accepted:
Published:
DOI: https://doi.org/10.1007/s00028-023-00870-6