Abstract
We investigate the convergence, in the small mass limit, of the stationary solutions of a class of stochastic damped wave equations, where the friction coefficient depends on the state and the noisy perturbation is of multiplicative type. We show that the Smoluchowski–Kramers approximation that has been previously shown to be true in any fixed time interval, is still valid in the long time regime. Namely, we prove that the first marginals of any sequence of stationary solutions for the damped wave equation converge to the unique invariant measure of the limiting stochastic quasilinear parabolic equation. The convergence is proved with respect to the Wasserstein distance associated with the \(H^{-1}\) norm.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this article we deal with the following stochastic wave equation with state-dependent damping, on a bounded smooth domain \({\mathcal {O}}\subset {\mathbb {R}}^d\), with \(d\ge 1\),
depending on a parameter \(0<\mu<<1\). The friction coefficient \(\gamma \) is a strictly positive, bounded and continuously differentiable function. The diffusion coefficient \(\sigma \) is bounded and Lipschitz-continuous and the noise \(w^Q(t)\) is a cylindrical Q-Wiener process, white in time and colored in space. The nonlinearity \(f:{\mathcal {O}}\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) is Lipschtz-continuous with respect to the second variable and the identically zero function is globally asymptotically stable in the absence of the stochastic perturbation. Here and in what follows, we denote \(H:=L^2({\mathcal {O}})\), \(H^{-1}:=H^{-1}({\mathcal {O}})\), and \(H^1:=H^1_0({\mathcal {O}})\) (the set of functions in the Sobolev space \(H^1({\mathcal {O}})\) with zero-trace).
The solution \(u_\mu (t,x)\) of Eq. (1.1) can be interpreted as the displacement of the particles of a material continuum in a domain \({\mathcal {O}}\), subject to a random external force field \(\partial _t w^Q(t,x)\) and a damping force which is proportional to the velocity field and depends on the state \(u_\mu \). The second order differential operator takes into account of the interaction forces between neighboring particles, in the presence of a non-linear reaction given by f. Here \(\mu \) represents the constant density of the particles and we are interested in the regime when \(\mu \rightarrow 0\), known as the Smoluchowski–Kramers approximation limit (Refs. [25, 31]).
In [3, 4] it has been proven that, when \(\gamma \) is constant, for every \(T>0\) and \(\eta >0\)
where \(u \in \,L^2(\Omega ; C([0,T];H)\cap L^2(0,T;H^1))\) is the solution of the parabolic problem
When the friction coefficient \(\gamma \) is state-dependent, the situation is more complicated and, because of the interplay between the noise and the non-constant friction, in the small-mass limit an extra drift term is created. In this regard, in [10] it has been proven that for every \({\mathfrak {u}}_0 \in \,H^1\), \(T>0\) and \(p<\infty \), and for every \(\eta >0\)
where u is the unique solution of the stochastic quasi-linear equation
Notice that the case of a non-constant damping coefficient is not the sole instance in which, within the context of a small mass limit, an additional drift term appears. For example, in the case of a damped stochastic wave equation, constrained to live on the unitary sphere of H, in the limit the Smoluchowski–Kramers approximation yields a stochastic parabolic problem also constrained to live on the unitary sphere of H, where an extra-drift term emerges, and that drift does not encompass the Itô-to-Stratonovich correction (see [2]). For a partial list of references where this type of limit has been addressed in a variety of different contexts, see [1, 5, 16, 17, 22,23,24, 26, 32], in finite dimension, and [2,3,4, 9, 10, 27,28,29], in infinite dimension.
After establishing the validity of the small mass limits within a fixed time interval [0, T], the next step of interest is to compare the long-term dynamics of the second-order system with that of the first-order system (to this purpose, see e.g. [6,7,8, 11, 12, 24]).
In [3], a comparative analysis of the long-term behavior of equations (1.1) (with a constant \(\gamma \)) and (1.3) was conducted, assuming both systems to be of gradient type. Notably, in the case where the noise is white in both space and time (\(Q=I\)) and the dimension is \(d=1\), an explicit expression for the Boltzmann distribution of the process \(z_\mu (t):=(u_{\mu }(t), \partial u_{\mu }/\partial t(t))\) in the phase space \({{\mathcal {H}}}:=L^2(0,L)\times H^{-1}(0,L)\) was derived. Since there is no equivalent of the Lebesgue measure in the functional space \({{\mathcal {H}}}\), an auxiliary Gaussian measure was introduced, and the density of the Boltzmann distribution was then expressed with respect to such auxiliary Gaussian measure, which itself corresponds to the stationary measure of the linear wave equation associated with problem (1.1). In particular, it was shown that the first marginal of the invariant measure linked to the process \(z_{\mu }(t)\) remains independent of \(\mu >0\) and coincides with the invariant measure for the heat equation (1.3).
In the case of non-gradient systems, that is when the noise is colored in space and/or of multiplicative type, there is no explicit expression for the invariant measure \(\nu _\mu \) associated with system (1.1) and there is no reason to expect that the first marginal of \(\nu _\mu \) does not depend on \(\mu \) or coincides with the invariant measure \(\nu \) of system (1.3). Nonetheless, in [6] it was proved that, as the mass parameter \(\mu \) tends to zero, the first marginal of any invariant measure \(\nu _\mu \) associated with the second-order system (1.1) converges in a suitable manner to the invariant measure \(\nu \) of the first-order system (1.3). Specifically, the following convergence was established
where \((\Pi _1\nu _\mu )^\prime \) denotes the extension of the first marginal of the invariant measure \(\nu _\mu \) to H, and the metric \({\mathcal {W}}_\alpha \) corresponds to the Wasserstein metric on \({\mathcal {P}}(H)\) associated with a distance metric \(\alpha \) in H, which was determined based on the characteristics of the non-linearity function f under consideration.
In the present paper, we want to see if any of the results proved in [6] in the case of a constant friction \(\gamma \), can be proven for a state-dependent \(\gamma \), where the Smoluchowski–Kramers approximation gives the stochastic quasi-linear parabolic problem (1.5), instead of the simpler parabolic semi-linear problem (1.3).
One of the key ingredients used in [6] for the proof of (1.6) is the fact that the transition semigroup \(P^H_t\) associated with equation (1.3) admits a unique invariant measure \(\nu \in \,{\mathcal {P}}(H)\) and the following contraction property holds
for some \(c, \delta >0\). In the case of Eq. (1.3), these types of problems have been studied extensively and a wide variety of results is available. However, in the case of the quasi-linear problem (1.5) the situation is considerably more delicate and several fundamental facts are not known, as for one whether the semigroup associated is Feller in H or not. In particular, even the use of the Krylov–Bogoliubov theorem for the proof of the existence of an invariant measure in H is not possible. Thus, in the present paper we have to follow a different path, that in particular brings us to study Eqs. (1.1) and (1.5) in spaces of lower regularity than \(H^1\times H\) and H, respectively.
In [10], it has been proved that Eq. (1.1) is well-posed in \({\mathcal {H}}_1:=H^1\times H\), for every \(\mu >0\), so that the associated Markov transition semigroup \(P^{\mu , {\mathcal {H}}_1}_t\) can be introduced. Our first step is showing that in fact (1.1) is well-posed also in \({\mathcal {H}}:=H\times H^{-1}\), for every \(\mu >0\), and there exists an invariant measure \(\nu ^{\mu , {\mathcal {H}}}\) for the corresponding transition semigroup \(P^{\mu , {\mathcal {H}}}_t\). We show that such invariant measure is supported in \({\mathcal {H}}_1\) and its restriction to \({\mathcal {H}}_1\) is invariant for \(P^{\mu , {\mathcal {H}}_1}_t\). Moreover, we prove suitable uniform bounds for the moments of \(\nu ^{\mu , {\mathcal {H}}}\) and \(\nu ^{\mu , {\mathcal {H}}_1}\), which are fundamental for the proof of the limit.
Next, we move our analysis to the limiting equation (1.5). To this purpose, we do not work directly with (1.5), but rather with its equivalent formulation
where g is the antiderivative of \(\gamma \) vanishing at zero. Since we are assuming that \(\gamma \) is strictly positive, bounded and continuously differentiable, the mappings
are both homeomorphisms and the coefficients in (1.8) are all well defined and regular. Moreover, as shown in [10], by using a generalized Itô’s formula, for every \({\mathfrak {r}}_0=g({\mathfrak {u}}_0) \in \,H^1\) and \(t\ge 0\) we have that
In particular, Eq. (1.8) is well posed in \(C([0,T];H)\cap L^2(0,T;H^1)\) if and only if Eq. (1.5) is.
As a consequence of limit (1.4), we have that for every initial condition \({\mathfrak {r}}_0 \in \,H^1\) Eq. (1.8) has a unique solution \(\rho ^{{\mathfrak {r}}_0} \in \,L^2(\Omega ;L^p(0,T;H^1))\), with \(p<\infty \). However, since the long time behavior of (1.8) in \(H^1\) and H is not well understood, we need to study its well-posedness in H and \(H^{-1}\), so that we can introduce the corresponding transition semigroups \(R^{H}_t\) and \(R^{H^{-1}}_t\). Due to the equivalence of problems (1.5) and (1.8) in H this allows us to introduce the transition semigroup \(P^H_t\) associated with Eq. (1.5).
Next, we prove that there exists some constant \(\lambda >0\) such that for every \({\mathfrak {r}}_1, {\mathfrak {r}}_2 \in \,H^{-1}\) and \(t\ge 0\)
To this purpose, we would like to mention that in [19, 20], it was proved that under suitable conditions on the initial conditions, the following property holds
Such bound gives in particular the Feller property in \(L^1({\mathcal {O}})\) but, unfortunately, this is not useful to our analysis, as it is not clear how to handle the proof of our limiting problem in a \(L^1({\mathcal {O}})\) setting. As far as we know, it is not clear if such a bound like (1.11) is satisfied in H. As we already mentioned above, this is why it becomes important to work in \(H^{-1}\), where we have the validity of even stronger condition (1.10).
As a consequence of (1.10), we have that \(R^{H^{-1}}_t\) is Feller. This, together with suitable uniform bounds in \(H^1\), allows to conclude that \(R^{H^{-1}}_t\) has an invariant measure \(\nu ^{H^{-1}}\), supported in \(H^1\). Moreover (1.10) implies that for every \(\varphi \in \text {Lip}_{b}(H^{-1})\) and \({\mathfrak {r}}_1, {\mathfrak {r}}_2\in H^{-1}\)
so that \(\nu ^{H^{-1}}\) is the unique invariant measure of \(R^{H^{-1}}_t\), and \(\nu ^H\), its restriction to H, turns out to be the unique invariant measure of \(R^H_t\). Finally, due to the equivalence between Eqs. (1.5) and (1.8), we show that this implies that \(P^{H}_t\) has a unique invariant measure \(\nu \).
By using a general argument developed in [15], and already used in [6] in a similar context, all this allows to obtain our main result. The idea introduced in [15] is quite simple and general. If \(\{\nu _n\}_{n \in \,{\mathbb {N}}}\) is a sequence of invariant measures for a sequence of Markov processes \(\{X_n(t)\}_{n \in \,{\mathbb {N}}}\) on some Banach space E, with transition semigroups \(\{P^n_t\}_{n \in \,{\mathbb {N}}}\), and \(\nu \) is the invariant measure for a Markov process X(t) on E, with transition semigroup \(P_t\), in order to study the convergence of \(\nu _n\) to \(\nu \) with respect to some Wasserstein distance \(\rho _\alpha \), associated to some distance \(\alpha \) on E, we first notice that, due to the invariance of \(\nu _n\) and \(\nu \),
Thus, if there exists some \(\delta >0\) such that for every probability measures \(\nu ^1\) and \(\nu ^2\) on E
from (1.13) we get
This implies that, if we pick \(t_\star >0\) such that \(c\,e^{-\delta t_\star }<1/2\), we have
Now, thanks to the Kantorovich–Rubinstein duality, we have
where \(\gamma _n\) is a E-valued random variable, distributed as \(\nu _n\), and \(X_n^{\gamma _n}(t)\) and \(X^{\gamma _n}(t)\) are the processes \(X_n(t)\) and X(t) with initial condition \(\gamma _n\). In particular, this implies that the proof of the converge of \(\nu _n\) to \(\nu \) with respect to the Wasserstein distance \(\rho _\alpha \) reduces to the proof of the following limit
for a fixed time t sufficiently large.
In the present paper, our goal is showing that if we define
then it holds
Due to (1.12) and the invariance of \(\nu _\mu ^{{\mathcal {H}}}\) and \(\nu ^{H^{-1}}\) we have
and then, if we pick \({\bar{t}}>0\) such that \(c e^{-\lambda {\bar{t}}}\le 1/2\), we obtain
(here we are using the notation \([\cdot ]^\prime \) to denote the extension to \(H^{-1}\) of an arbitrary probability measure defined in H). As we have seen above, the Kantorovich–Rubinstein duality gives
for every \({\mathcal {F}}_{0}\)-measurable \({\mathcal {H}}_{1}\)-valued random variable \(\zeta _\mu :=(\xi _\mu ,\eta _\mu )\), distributed as the invariant measure \(\nu _{\mu }^{ {\mathcal {H}}}\). Hence, once we prove that for every \(t\ge 0\) large enough
we obtain that
Our last steps consists in showing that this implies (), which also implies that \(\Pi _1 \nu _\mu ^{{\mathcal {H}}}\) converges to \(\nu \), weakly in H, as \(\mu \downarrow 0\).
2 Notations and Assumptions
Throughout the present paper \({\mathcal {O}}\) is a bounded domain in \({\mathbb {R}}^d\), with \(d\ge 1\), having a smooth boundary. We denote by H the Hilbert space \(L^2({\mathcal {O}})\) and by \(\Vert \cdot \Vert _H\) and \(\langle \cdot , \cdot \rangle _H\) the corresponding norm and inner product.
Given the domain \({\mathcal {O}}\), we denote by A the realization of the Laplace operator \(\Delta \), endowed with Dirichlet boundary conditions. As known there exists a complete orthonormal basis \(\{e_i\}_{i\in {\mathbb {N}}}\) of H which diagonalizes A. In what follows, we denote by \(\{-\alpha _i\}_{i\in {\mathbb {N}}}\) the corresponding sequence of eigenvalues, and for every \(\delta \in \,{\mathbb {R}}\), we define \(H^\delta \) as the completion of \(C^{\infty }_0({\mathcal {O}})\) with respect to the norm
Notice that with this definition \(H^0=H\) and, if \(\delta _1<\delta _2\), then \(H^{\delta _2}\hookrightarrow H^{\delta _1}\) with compact embedding. We also define
Next, for every two separable Hilbert spaces E and F, we denote by \({\mathcal {L}}(E,F)\) the space of bounded linear operators from E into F and by \({\mathcal {L}}_2(E,F)\) the subspace of Hilbert–Schmidt operators. \({\mathcal {L}}_2(E,F)\) is a Hilbert space, endowed with the inner product
and, as well known, \({\mathcal {L}}_2(E,F) \subset {\mathcal {L}}(E,F)\), with
Finally, if X is any Polish space, we denote by \(B_b(X)\) the space of bounded Borel measurable functions \(\varphi :X\rightarrow {\mathbb {R}}\), endowed with the sup-norm
Moreover, we denote by \(C_b(X)\) the subspace of uniformly continuous and bounded functions.
2.1 Assumptions
We assume that \(w^Q(t)\) is a cylindrical Q-Wiener process, for some \(Q \in \,{\mathcal {L}}(H)\), defined on a complete stochastic basis \((\Omega ,{\mathcal {F}},({\mathcal {F}}_t)_{t\ge 0},{\mathbb {P}})\). This means that \(w^Q(t)\) can be formally written as
where \(\{\beta _i\}_{i\in {\mathbb {N}}}\) is a sequence of independent standard Brownian motions on \((\Omega ,{\mathcal {F}},({\mathcal {F}}_t)_{t\ge 0},{\mathbb {P}})\), and \(\{e_i\}_{i\in \,{\mathbb {N}}}\) is the complete orthonormal system introduced above that diagonalizes the Laplace operator, endowed with Dirichlet boundary conditions. When \(Q=I\), the process \(w^I(t)\) will be denoted by w(t). In particular, we have \(w^Q(t)=Qw(t)\).
In what follows we shall denote by \(H_Q\) the set Q(H). \(H_Q\) is the reproducing kernel of the noise \(w^Q\) and is a Hilbert space, endowed with the inner product
Notice that the sequence \(\{Q e_i\}_{i \in \,{\mathbb {N}}}\) is a complete orthonormal system in \(H_Q\). Moreover, if U is any Hilbert space containing \(H_Q\) such that the embedding of \(H_Q\) into U is Hilbert–Schmidt, we have that
Hypothesis 1
The mapping \(\sigma :H\rightarrow {\mathcal {L}}_2(H_Q,H)\) is defined by
for some measurable mappings \(\sigma _i:{\mathcal {O}}\times {\mathbb {R}}\rightarrow {\mathbb {R}}\). We assume that there exists \(L_\sigma >0\) such that
Moreover, we assume \(\sigma \) is bounded, that is,
Remark 2.1
-
1.
Condition (2.1) implies that \(\sigma :H\rightarrow {\mathcal {L}}_2(H_Q,H)\) is Lipschitz continuous. Namely, for any \(h_1,h_2\in H\)
$$\begin{aligned} \Vert \sigma (h_1)-\sigma (h_2)\Vert _{{\mathcal {L}}_2(H_Q,H)} \le \sqrt{L_\sigma }\, \Vert h_1 -h_2\Vert _H. \end{aligned}$$(2.3) -
2.
If the noise is additive, then Hypothesis 1 is satisfied when \(\text {Tr}\,Q^{2}<+\infty \).
Hypothesis 2
The mapping \(\gamma \) belongs to \(C^1_b({\mathbb {R}})\) and there exist \(\gamma _0\) and \(\gamma _1\) such that
If we define
the function \(g:{\mathbb {R}}\rightarrow {\mathbb {R}}\) is differentiable, strictly increasing and invertible so that its inverse \(g^{-1}:{\mathbb {R}}\rightarrow {\mathbb {R}}\) is differentiable, with
Hypothesis 3
The mapping \(f:{\mathcal {O}}\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) is measurable and there exists a positive constant \(L_f\) such that
where \(\alpha _1\) is the smallest eigenvalue of \(-A\), and
Moreover,
In what follows, for every \(x\in {\mathcal {O}}\) and \(r\in {\mathbb {R}}\) we denote
and for every function \(h:{\mathcal {O}}\rightarrow {\mathbb {R}}\), we denote
Remark 2.2
-
1.
Condition (2.8) implies that \(F:H\rightarrow H\) is Lipschitz continuous. Namely for any \(h_{1},h_{2}\in H\)
$$\begin{aligned} \left\| F(h_{1})-F(h_{2})\right\| _{H}\le L_f\left\| h_{1}-h_{2}\right\| _{H}. \end{aligned}$$Moreover, there exists \(c>0\) such that
$$\begin{aligned} \Vert F(h)\Vert _{H}\le L_f\,\Vert h\Vert _H+c. \end{aligned}$$(2.9) -
2.
If the friction coefficient \(\gamma \) is constant, then \(\gamma _{0}=\gamma _{1}\), and condition (2.7) becomes
$$\begin{aligned} L_f<\alpha _{1}. \end{aligned}$$ -
3.
It is immediate to check that if for every \(h \in \,H\) we define
$$\begin{aligned} \Lambda (h):=\int \limits _{{\mathcal {O}}}f(x,h(x))\,dx, \end{aligned}$$then \(\Lambda :H\rightarrow {\mathbb {R}}\) is differentiable and its differential is given by
$$\begin{aligned} {[}D\Lambda (h)](x)=f(x,h(x)),\ \ \ \ \ x \in \,{\mathcal {O}}. \end{aligned}$$(2.10)
Hypothesis 4
We assume
Remark 2.3
Condition (2.11) is assumed in order to have the well-posedness of Eq. (1.8) in \(H^{-1}\) and to prove limit (1.15). If the diffusion coefficient \(\sigma \) is constant, then \(L_\sigma =0\) and Hypothesis 4 reduces to condition (2.7) in Hypothesis 3. However, in the case \(\sigma \) is non constant, condition (2.7) alone is not enough, as also the Lipschitz constant of g has to be small, compared to the eigenvalue \(\alpha _1\) and the constants \(\gamma _0\) and \(\gamma _1\).
3 The Main Result
For every \(\mu >0\), we denote \(v_{\mu }:=\partial _{t}u_{\mu }\), and rewrite Eq. (1.1) as the following system
where A is the realization in H of the Laplacian \(\Delta \), endowed with Dirichlet boundary conditions. In [10] it has been proven that, under Hypotheses 1, 2 and 3 (without condition (2.7)), for every \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in L^{2}(\Omega ;{{\mathcal {H}}}_{1})\), and for every \(\mu , T>0\), there exists a unique process \(z_\mu \in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) which solves system (3.1), in the sense that
In particular, we can introduce the transition semigroup \(P_{t}^{\mu , {\mathcal {H}}_1}\) associated with Eq. (3.1) in \({\mathcal {H}}_1\), by setting
for every \(\varphi \in B_{b}({\mathcal {H}}_1)\) and \({\mathfrak {z}}\in {\mathcal {H}}_1\).
In what follows, we will need to study system (3.1) also in the space of lower regularity \({\mathcal {H}}\), and for this reason we introduce the following notion of generalized solution.
Definition 3.1
For every \(\mu ,T>0\) and every \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in {\mathcal {H}}\), we say that the process \(z_{\mu }\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}))\) is a generalized solution of system (3.1) if for every sequence \(\{{\mathfrak {u}}_{0,n},{\mathfrak {v}}_{0,n}\}_{n\in {\mathbb {N}}}\subset {{\mathcal {H}}}_{1}\) converging to \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\) in \({{\mathcal {H}}}\), as \(n\rightarrow +\infty \), it holds
where \(z_{\mu , n}\in L^{2}(\Omega ;C([0,T];{{\mathcal {H}}}_{1}))\) is the unique solution of Eq. (3.1) with initial conditions \(({\mathfrak {u}}_{0,n},{\mathfrak {v}}_{0,n})\).
Notice that if \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in {\mathcal {H}}_1\), then the weak solution coincides with the solution defined above in \({\mathcal {H}}_1\) in the sense of (3.2).
In Sect. 5 we will prove that, under Hypotheses 1, 2 and 3, for every \(\mu , T>0\) there exists a unique generalized solution \(z_\mu \in L^{2}(\Omega ;C([0,T];{\mathcal {H}}))\) for system (3.1). This will allow us to introduce the transition semigroup associated with (3.1) in \({\mathcal {H}}\), which will be denoted by \({P}_{t}^{\mu , {\mathcal {H}}}\). Clearly, if \(\varphi \in \,B_b({\mathcal {H}})\), for every \(\mu >0\) and \({\mathfrak {z}}\in {{\mathcal {H}}}_{1}\) we have
In Sect. 5 we will also show that under the same Hypotheses, for every \(\mu >0\), the semigroup \( P_t^{\mu , {\mathcal {H}}}\) admits an invariant measure \(\nu ^{{\mathcal {H}}}_{\mu }\) in \({\mathcal {H}}\), with \(\text {supp}\,(\nu _{\mu }^{{\mathcal {H}}})\subset {\mathcal {H}}_{1}\). In particular, since \(\text {supp}\,(\nu _{\mu }^{{\mathcal {H}}})\subset {{\mathcal {H}}}_{1}\) and \(\mathcal{B}({{\mathcal {H}}}_{1})\subset \mathcal{B}({{\mathcal {H}}})\), we will have that \(\nu ^{{\mathcal {H}}}_\mu \) is also a probability measure on \({{\mathcal {H}}}_{1}\). In what follows, it will be convenient to denote the restriction of \(\nu ^{{\mathcal {H}}}_\mu \) to \({\mathcal {H}}_1\) by \(\nu ^{{\mathcal {H}}_1}_{\mu }\).
Now, we recall that, given a lower semicontinuous metric \(\alpha \) on \(H^{-1}\), it is possible to introduce the distance \({\mathcal {W}}_{\alpha }:{\mathcal {P}}(H^{-1})\times {\mathcal {P}}(H^{-1})\rightarrow [0,+\infty ]\) defined by
where
Notice that, if \({\mathcal {C}}(\nu _1, \nu _2)\) is the set of all couplings of \((\nu _{1},\nu _{2})\), the following Kantorovich–Rubinstein identity holds
and in fact it is possible to prove that the infimum above is attained at some \({\bar{\lambda }}\).
At this point, we are ready to state the main result of this paper.
Theorem 3.2
Assume Hypotheses 1 to 4, and define
Then we have
where \(\nu \) is the unique invariant measure for \(P^H_t\), the transition semigroup associated to the limiting equation (1.5). Moreover,
4 Plan of the Paper and List of Symbols
In Sect. 5, we will study the well-posedness of system (3.1) in \({\mathcal {H}}\). Namely, we will prove that for every initial condition \(({\mathfrak {u}}_0,{\mathfrak {v}}_0) \in \,{\mathcal {H}}\) system (3.1) admits a unique generalized solution in \({\mathcal {H}}\). This will allow us to introduce the transition semigroup \(P^{\mu \,{\mathcal {H}}}_t\), \(t\ge 0\), for every \(\mu >0\). Moreover, we will prove that \(P^{\mu \,{\mathcal {H}}}_t\) admits an invariant measure \(\nu _\mu ^{{\mathcal {H}}}\) supported in \({\mathcal {H}}_1\).
In Sect. 6, we will prove suitable a-priori bounds for the solutions of system (3.1). In particular, we will prove some uniform bounds for the momenta of the invariant measures \(\nu _\mu ^{{\mathcal {H}}}\).
In Sect. 7, we will consider the limiting problem (1.5) in the space H. To this purpose, we will introduce the auxiliary problem (1.8) and we will first study its well-posedness in H. Due to (1.9), we will get the analogous results for problem (1.5). Moreover, we will study the well-posedness of (5.1) in \(H^{-1}\).
In Sect. 8, we will investigate the ergodic behavior of the limiting Eq. (1.5) in \(H^{-1}\). We will prove that the corresponding semigroup has a contractive property in \(H^{-1}\). In particular, we it admits a unique invariant measure in \(H^{-1}\). Moreover, we will show that such invariant measure is supported in H and its restriction to H is the unique invariant measure for the semigroup associated with Eq. (1.5) in H. Finally, we will show that this implies that the semigroup \(P^H_t\) admits a unique invariant measure.
In Sect. 9, we will finally prove the main result of this paper, Theorem 3.2.
4.1 Symbols Used Throughout the Paper
Almost all the symbols listed below are introduced for the first time in the Introduction. However, they are introduced again with all the needed details throughout the paper. In what follows, we describe what is their meaning and where their definition is given.
-
\(H=L^2({\mathcal {O}})\), \(H^1=H^1_0({\mathcal {O}})\), and \(H^{-1}=H^{-1}({\mathcal {O}})\), Sect. 2.
-
\({\mathcal {H}}= H\times H^{-1},\) and \({\mathcal {H}}_1=H^1\times H\), Sect. 2.
-
For every \(\mu >0\), \(z_\mu =(u_\mu ,v_\mu )\) denotes the solution of system (3.1), both in \({\mathcal {H}}\) and \({\mathcal {H}}_1\), Sect. 3.
-
For every \(\mu >0\), \(P^{\mu ,{\mathcal {H}}_1}_t\) is the semigroup associated with system (3.1) in \({\mathcal {H}}_1\), and \(P^{\mu ,{\mathcal {H}}}_t\) is the semigroup associated with system (3.1) in \({\mathcal {H}}\), Sect. 3.
-
For every \(\mu >0\), \(\zeta _\mu =(u_\mu ,\eta _\mu )\) denotes the solution of system (1.8), both in \({\mathcal {H}}\) and \({\mathcal {H}}_1\), Sect. 5.
-
For every \(\mu >0\), \(\nu ^{{\mathcal {H}}}_\mu \) is an invariant measure for the semigroup \(P^{\mu ,{\mathcal {H}}}_t\), and \(\nu ^{{\mathcal {H}}_1}_\mu \) is its restriction to \({\mathcal {H}}_1\), which is invariant for \(P^{\mu ,{\mathcal {H}}_1}_t\), Sect. 5.3.
-
For every \(\mu >0\), \({\mathcal {N}}_\mu \) is the Kolmogorov operator associated with system (3.1) in \({\mathcal {H}}_1\) and its associated semigroup \(P^{\mu ,{\mathcal {H}}_1}_t\), Sect. 6.
-
u denotes the solution of the limiting problem (1.5), both in H and \(H^{-1}\), Sect. 7.
-
\(P^H_t\) is the transition semigroup associated with the limiting problem (1.5) in H, Sect. 7.2.
-
\(\rho \) denotes the solution of the limiting problem (1.8), both in H and \(H^{-1}\), Sect. 7.
-
\(R^{H}_t\) is the transition semigroup associated with the auxiliary problem (1.8) in H and \(R^{H^{-1}}_t\) is the transition semigroup associated with (1.8) in \(H^{-1}\), Sect. 8.
-
\(\nu ^H\) is the unique invariant measure of the semigroup \(R^H_t\) and \(\nu ^{H^{-1}}\) is the unique invariant measure of the semigroup \(R^{H^{-1}}_t\), Sect. 8.
-
\(\nu ^H\circ g\) is the unique invariant measure of the semigroup \(P^H_t\), Sect. 9.
5 Generalized Solutions for System (3.1) and Invariant Measures
We have seen in Sect. 3 that Eq. (1.1) is equivalent to system (3.1). In fact, we can give another equivalent formulation for system (3.1). Actually, if g is the function introduced in (2.5) and we define
then system (3.1) can be rewritten as
where we have denoted
and
This means that, for every \(\mu >0\) and every \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in {{\mathcal {H}}}_{1}\), the adapted \({{\mathcal {H}}}_{1}\)-valued process \(\zeta _{\mu }=(u_\mu , \eta _\mu )\) is the unique solution of Eq. (5.1), with \(\zeta _{\mu }(0)=\big ({\mathfrak {u}}_0,\sqrt{\mu }{\mathfrak {v}}_0+g({\mathfrak {u}}_0)/\sqrt{\mu }\big )\), if and only if the adapted \({{\mathcal {H}}}_{1}\)-valued process
is the unique solution of system (3.1), with \(z_\mu (0)={\mathfrak {z}}_0:=({\mathfrak {u}}_0,{\mathfrak {v}}_0)\). The reason why we have introduced the equivalent problem (5.1) is that, in the presence of a non-constant friction \(\gamma \), while it is not clear how to study the well-posedness of system (3.1), the analogous problem for (5.1) can be handled in a more direct way, both in \({\mathcal {H}}_1\) and in \({\mathcal {H}}\), thanks to the theory of non-linear quasi-dissipative operators.
As a matter of fact, in [10] it has been proven that, under Hypotheses 1, 2 and 3 (without condition (2.7)), for every \(\zeta ^0\in L^{2}(\Omega ;{\mathcal {H}}_{1})\) and for every \(\mu , T>0\), there exists a unique solution \(\zeta _{\mu }\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) for Eq. (5.1), with \(\zeta _{\mu }(0)=\zeta ^0\), and this has allowed to conclude that for every \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in L^{2}(\Omega ;{{\mathcal {H}}}_{1})\), and for every \(\mu , T>0\), there exists a unique solution \(z_\mu \in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) to Eq. (3.1), with \(z_\mu (0)=({\mathfrak {u}}_0,{\mathfrak {v}}_0)\).
5.1 Generalized Solutions for System (5.1)
In order to study the existence and uniqueness of generalized solutions for system (3.1), we study the analogous problem for system (5.1). As for (3.1), we have the following definition of generalized solution for system (5.1).
Definition 5.1
For every \(\mu ,T>0\) and every \(\zeta ^0\in {\mathcal {H}}\), we say that \(\zeta _{\mu }\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}))\) is a generalized solution of problem (5.1), with initial condition \(\zeta ^0\), if for every sequence \(\{\zeta ^0_n\}_{n\in {\mathbb {N}}}\subset {{\mathcal {H}}}_{1}\) converging to \(\zeta ^0\) in \({{\mathcal {H}}}\), as \(n\rightarrow +\infty \), it holds
where \(\zeta _{\mu ,n}\in L^{2}(\Omega ;C([0,T];{{\mathcal {H}}}_{1}))\) is the unique solution of Eq. (5.1) with initial condition \(\zeta ^0_n\).
The following result holds.
Lemma 5.2
Under Hypotheses 1, 2 and 3, for every \(\mu ,T>0\) and every \(\zeta ^0\in {\mathcal {H}}\), there exists a unique generalized solution \(\zeta _{\mu }\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}))\) for Eq. (5.1). Moreover, if \(\zeta _{\mu }^{1}, \zeta _{\mu }^{2}\) are two generalized solutions of (5.1), with initial conditions \(\zeta ^1, \zeta ^2\in {{\mathcal {H}}}\), respectively, then
for some constant \(c_{\mu }\).
Proof
Without any loss of generality, we assume \(\mu =1\), and for simplicity of notation, we denote \({\mathcal {A}}_{1}\) and \(\Sigma _1\) by \({\mathcal {A}}\) and \(\Sigma \), respectively. In [10], it is proved that the operator \({{\mathcal {A}}}\) is quasi-m-dissipative in \({\mathcal {H}}\). Namely, there exists \(\eta \ge 0\) such that for every \(\zeta ,\theta \in D({\mathcal {A}})\)
and there exists \(\lambda _{0}>0\) such that
Now, let \(\zeta ^0\in {{\mathcal {H}}}\) and let \(\{\zeta ^0_n\}_{n\in {\mathbb {N}}}\subset {{\mathcal {H}}}_{1}\) be any sequence converging to \(\zeta ^0\) in \({{\mathcal {H}}}\). For every \(n\in {\mathbb {N}}\), we denote by \(\zeta _{n}\) the unique solution of Eq. (5.1) with initial condition \(\zeta _{n}(0)=\zeta ^0_n\). By applying Itô’s formula, thanks to (2.1) and (5.3), we get
Due to (2.3) we have
Thus, if we first integrate both sides in (5.4) with respect to time, and then take the supremum and the expectation, we get
and the Gronwall’s inequality gives
for some constant c. In particular, this implies that the sequence \(\{\zeta _{n}\}_{n\in {\mathbb {N}}}\) is Cauchy in the space \(L^{2}(\Omega ;C([0,T];{{\mathcal {H}}}))\), so there exists a limit \(\zeta \in L^{2}(\Omega ;C([0,T];{{\mathcal {H}}}))\). It is easy to see that the limit \(\zeta \) does not depend on the choice of the sequence \(\{\zeta ^0_n\}\subset {{\mathcal {H}}}_{1}\), which implies the uniqueness of generalized solutions. Finally, by using a similar argument as above, we obtain (5.2). \(\square \)
Remark 5.3
When \(\zeta ^0\in {{\mathcal {H}}}_{1}\), the unique generalized solution \(\zeta _{\mu }\) of Eq. (5.1) coincides with its unique classical solution.
5.2 Generalized Solutions for System (3.1)
Due to Hypothesis 2, it is immediate to check that \(z_\mu =(u_{\mu },v_{\mu })\) is a generalized solution to (3.1), with initial condition \({\mathfrak {z}}_0=({\mathfrak {u}}_0,{\mathfrak {v}}_0)\), if and only if \(\zeta _{\mu }=\big (u_{\mu },\sqrt{\mu }v_{\mu }+g(u_{\mu })/\sqrt{\mu }\big )\) is a generalized solution for system (5.1), with initial condition \(\zeta _{\mu }(0)=({\mathfrak {u}}_0,\sqrt{\mu }{\mathfrak {v}}_0+g({\mathfrak {u}}_0)/\sqrt{\mu })\). In this case, we have
Thus, as a consequence of Lemma 5.2 and Remark 5.3, we have the following result.
Proposition 5.4
Fix \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in {\mathcal {H}}\) and assume Hypotheses 1, 2 and 3. Then, for every \(\mu , T>0\) there exists a unique generalized solution \(z_\mu \in L^{2}(\Omega ;C([0,T];{\mathcal {H}}))\) for system (3.1).
5.3 Existence of Invariant Measures for System (3.1)
We are proving now that, for every fixed \(\mu >0\), system (3.1) admits an invariant measure \(\nu _\mu ^{{\mathcal {H}}}\) in \({\mathcal {H}}\), which is supported on \({\mathcal {H}}_1\).
Proposition 5.5
Assume Hypotheses 1, 2 and 3. Then, for every \(\mu >0\), the semigroup \(P_t^{\mu , {\mathcal {H}}}\) admits an invariant measure \(\nu ^{{\mathcal {H}}}_{\mu }\) in \({\mathcal {H}}\), with \(\text {supp}\,(\nu _{\mu }^{{\mathcal {H}}})\subset {\mathcal {H}}_{1}\).
Proof
First, if \(z_\mu ^{{\mathfrak {z}}_1}\) and \(z_\mu ^{{\mathfrak {z}}_2}\) are generalized solutions to system (3.1), with initial conditions \({\mathfrak {z}}_1, {\mathfrak {z}}_2\in {\mathcal {H}}\), respectively, then due to (5.2) and (5.5), it is easy to see that for every \(t\ge 0\)
for some \(c_\mu (t)>0\). This means that the transition semigroup \(P^{\mu , {\mathcal {H}}}_{t}\) is Feller on \({\mathcal {H}}\).
Now, for every \({\mathfrak {z}} \in \,{\mathcal {H}}\) we introduce the following family of measures on \({\mathcal {H}}\)
and for every \(R>0\) we define the set
Then, from (6.1) and (6.2) with \(\mu =1\), we have
and, due to the compactness of the embedding of \({\mathcal {H}}_1\) into \({\mathcal {H}}\), this implies that the family of measures \(\{\Gamma ^\mu _{t}(0,\cdot )\}_{t\ge 0}\) is tight in \({\mathcal {H}}\). By the Prokhorov theorem, there exists some sequence \(t_n\uparrow \infty \) such that \(\Gamma ^\mu _{t_n}(0,\cdot )\) converges weakly, as \(n\rightarrow +\infty \), to a probability measure \(\nu _\mu ^{{\mathcal {H}}}\) that is invariant for \(P^{\mu , {\mathcal {H}}}_{t}\). Moreover, since
it follows that \(\text {supp}\,(\nu _\mu ^{ {\mathcal {H}}})\subset {\mathcal {H}}_{1}\). \(\square \)
Remark 5.6
Since \(\text {supp}\,(\nu _{\mu }^{{\mathcal {H}}})\subset {{\mathcal {H}}}_{1}\) and \(\mathcal{B}({{\mathcal {H}}}_{1})\subset \mathcal{B}({{\mathcal {H}}})\), we have that \(\nu ^{{\mathcal {H}}}_\mu \) is also a probability measure on \({{\mathcal {H}}}_{1}\). In what follows, it will be convenient to denote the restriction of \(\nu ^{{\mathcal {H}}}_\mu \) to \({\mathcal {H}}_1\) by \(\nu ^{{\mathcal {H}}_1}_{\mu }\).
6 Some Uniform Bounds for System (3.1)
We have seen that for every \(\mu > 0\) Eq. (3.1) has an invariant measure. In this section we will prove some uniform bounds for the moments of such family of invariant measures. To this purpose, we need to start with suitable uniform bounds for the solution \((u_{\mu },v_{\mu })\) of system (3.1). Some of them have been already proved in [10, Proposition 4.2, Remark 4.3]. In what follows, we show how those bounds depend on time and on random initial conditions in \(L^2(\Omega ;{\mathcal {H}}_1)\).
Lemma 6.1
Assume Hypotheses 1, 2 and 3, and fix \((\xi ,\eta )\in L^{2}(\Omega ;{\mathcal {H}}_{1})\). For every \(\mu , T>0\), let \((u_{\mu },v_{\mu })\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) be the unique solution to system (3.1) with initial conditions \((\xi , \eta )\). Then there exist two constants \(\mu _{0}\in (0,1)\) and \(c>0\), independent of \(T>0\), such that for every \(\mu \in (0,\mu _{0})\) and \(t \in \,[0,T]\)
and
Proof
Let \(({\mathfrak {u}}_0,{\mathfrak {v}}_0)\in L^{2}(\Omega ;{\mathcal {H}}_{1})\) and let \((u_{\mu },v_{\mu })\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) be the unique solution to system (3.1). By proceeding as in the proof of [10, Lemma 4.2], we have for every \(\mu \in (0,1)\)
Moreover, by proceeding as in the proof of [10, Lemma 4.1], we have \({{\mathbb {P}}}\)-a.s.
Due to (2.9), for every \(\delta >0\) we have
Moreover, due to (2.2) we have
According to (2.7), we can fix \(\delta >0\) such that
and this yields
Thus (6.2) holds by combining (6.4) with (6.3). Finally, by combining (6.2) with (6.3), we complete the proof of (6.1). \(\square \)
Lemma 6.2
Let \(\{(\xi _{\mu },\eta _{\mu })\}_{\mu \in \,(0,1)}\subset L^{2}(\Omega ;{{\mathcal {H}}}_{1})\) be a family of random variables such that
If \((u_{\mu },v_{\mu })\in L^{2}(\Omega ;C([0,T];{\mathcal {H}}_{1}))\) is the solution to system (3.1) with initial condition \((\xi _{\mu },\eta _{\mu })\), then there exist \(\mu _{T}\in (0,\mu _{0})\) and \(c_{T}>0\) such that for every \(\mu \in (0,\mu _{T})\)
Proof
If for every \(\mu \in (0,\mu _{0})\) and \(t\in [0,T]\), we define
then (6.6) is equivalent to
for some constants \(\mu _{T}\in (0,1)\) and \(c_{T}>0\).
Now, if we assume (6.7) is not true, there exists a sequence \((\mu _{k})_{k\in {\mathbb {N}}}\subset (0,\mu _{0})\) converging to 0, as \(k\rightarrow \infty \), such that
For every \(k\in {\mathbb {N}}\), the mapping \(t\mapsto L_{\mu _k}(t)\) is continuous \({\mathbb {P}}\)-a.s., so that there exists a random time \(t_{k}\in [0,T]\) such that
As a consequence of Itô’s formula, we have
Hence, if s is any random time such that \({\mathbb {P}}(s\le t_{k})=1\), we have
where
If we define
this implies that there exists some constant \(\lambda >0\), independent of k, such that
and since \(L_{\mu _{k}}(0)=0\), if we take \(s=0\) we get
Now, on the set \(E_{k}:=\big \{\theta _{k}>0\big \}\), we fix an arbitrary \(s\in \big [t_{k}-\mu _{k}\theta _{k}/(2\lambda ),t_{k}\big ]\). Since \(t_k-s\le \mu _k\theta _k/(2\lambda )\), by using again (6.9) and recalling the definition of \(\theta _k\), we have
Hence, if we define
due to (6.10) we have
so that
Now, according to (6.1), (6.2) and (6.5)
and
so that
Thanks to (6.8) this gives
and hence
Now, according to (6.11), we have
and due to (6.12), this implies
However, as a consequence of (6.1), (6.2) and (6.5), we have
and this gives a contradiction, since \({{\mathbb {E}}}(I_{k};E_{k})\le {{\mathbb {E}}}(I_{k})\) for every \(k\in {\mathbb {N}}\). In particular, this means that claim (6.7) is true, and (6.6) holds. \(\square \)
Lemma 6.3
For every \(\mu >0\), if \(\nu _{\mu }^{{\mathcal {H}}}\in {\mathcal {P}}({{\mathcal {H}}})\) is any invariant measure for \( P_t^{\mu , {\mathcal {H}}}\) supported in \({{\mathcal {H}}}_{1}\), then \({\nu }_{\mu }^{{\mathcal {H}}_1}\in {\mathcal {P}}({{\mathcal {H}}}_{1})\) is invariant for \({P}_{t}^{\mu , {\mathcal {H}}_1}\). Moreover,
Proof
First, we show the invariance of \(\nu _{\mu }^{{\mathcal {H}}_1}\) for \({P}_{t}^{\mu , {\mathcal {H}}_1}\). Due to the invariance of \(\nu _{\mu }^{ {\mathcal {H}}}\) in \({\mathcal {H}}\), for every \(\varphi \in C_{b}({{\mathcal {H}}})\) we have
Thus, since \(\text {supp}\,(\nu _{\mu }^{ {\mathcal {H}}})\subset {{\mathcal {H}}}_{1}\) and \(\mathcal{B}({{\mathcal {H}}}_{1})\subset \mathcal{B}({{\mathcal {H}}})\), for every \(\varphi \in C_{b}({{\mathcal {H}}})\) we get
If \((\hat{e}_i)_{i \in \,{\mathbb {N}}}\subset {\mathcal {H}}_1\) is an orthonormal basis of \({\mathcal {H}}\), for every \(n\in {\mathbb {N}}\) we denote by \(\Pi _n\) the projection of \({{\mathcal {H}}}\) onto \({\mathcal {H}}(n):=\text {span}(\hat{e}_1,\ldots ,\hat{e}_n)\). We have that \(\Pi _n:{{\mathcal {H}}}\rightarrow {{\mathcal {H}}}_{1}\) is continuous and
Hence, if for any \(\varphi \in C_{b}({{\mathcal {H}}}_{1})\) and \(n\in {\mathbb {N}}\), we define \(\varphi _{n}:=\varphi \circ \Pi _n\), we have \(\varphi _{n}\in C_{b}({{\mathcal {H}}})\) and
For every \(n\in {\mathbb {N}}\), we have \(\sup _{n\in {\mathbb {N}}}\left\| \varphi _{n}\right\| _\infty \le \left\| \varphi \right\| _\infty \), and the dominated convergence theorem implies that for any given \(\mu >0\) and \(\varphi \in C_{b}({{\mathcal {H}}}_{1})\)
In particular, by taking the limit as n goes to infinity in both sides of
we conclude that
and this implies the invariance of \(\nu _{\mu }^{{\mathcal {H}}_1}\).
Next, in order to prove (6.13), we consider the Kolmogorov operator associated to \({P}_{t}^{\mu , {\mathcal {H}}_1}\) in \({{\mathcal {H}}}_{1}\)
If, with the notations of Sect. 2, we define
due to (2.10) and to the fact that \(\Vert {\mathfrak {u}}\Vert _{H^1}^2=\langle (-A) {\mathfrak {u}},{\mathfrak {u}}\rangle _H\), we have
Then, we have
By the invariance of \(\nu _{\mu }^{{\mathcal {H}}_1}\) in \({\mathcal {H}}_{1}\), we have
and thus, due to (6.14) and (2.2),
Next, we consider the function
We have
so that
Note that for every \(\delta >0\) and \(\mu \in \,(0,1)\)
and
so that, due to the invariance of \(\nu _\mu ^{{\mathcal {H}}}\), we have
Thanks to (2.7), this implies that we can take \(\delta >0\) sufficiently small so that
and then
By combining this with (6.15), we complete the proof of (6.13). \(\square \)
Remark 6.4
-
1.
In Proposition 5.5, we have seen that for every \(\mu >0\) the semigroup \( P_t^{\mu , {\mathcal {H}}}\) admits an invariant measure. Thanks to Lemma 6.3, this implies that for every \(\mu >0\) the transition semigroup \({P}_{t}^{\mu , {\mathcal {H}}_1}\) admits an invariant measure in \({{\mathcal {H}}}_{1}\).
-
2.
As a consequence of (6.13), we have
$$\begin{aligned} \sup _{\mu \in \,(0,1)} \int \limits _{{\mathcal {H}}}\left( \left\| {\mathfrak {u}}\right\| _{H^{1}}^{2}+\mu \left\| {\mathfrak {v}}\right\| _{H}^{2}\right) \nu _{\mu }^{ {\mathcal {H}}}(d{\mathfrak {u}},d{\mathfrak {v}})<\infty . \end{aligned}$$(6.16)
7 The Limiting Equation
As we have mentioned in the Introduction, in order to study the limiting problem
we consider first the following quasilinear stochastic parabolic equation
where for every \(r\in {\mathbb {R}}\) and \(x\in {\mathcal {O}}\)
and for every \(h\in H\)
The rationale behind this approach stems from the inherent advantage of initially establishing the small-mass limit of \(g(u_\mu )\) to \(\rho \), alongside their stationary counterparts, before moving back to the original problem involving \(u_\mu \) and u. As explained in [10], due to a generalized Itô’s formula, the solutions u and \(\rho \) of Eqs. (7.1) and (7.2), respectively, are related by
From Hypothesis 2, we know
Moreover, if we define
due to Hypotheses 1, 2 and 3, and due to (6.14), for every \(h_{1},h_{2}\in H\) we have
and
Moreover, for every \(\delta >0\)
and
Finally, thanks to (2.2) we have
7.1 Well-Posedness of Eq. (7.2) in H
Throughout this subsection we will not need to assume condition (2.7) in Hypothesis 3. Namely, we will just assume that the mapping \(f:{\mathcal {O}}\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) is measurable, with
As a consequence of the limiting result proved in [10], the well-posedness of Eq. (7.2) has been established when the initial condition \({\mathfrak {r}}_0\in H^{1}\). Here we want to prove the existence and uniqueness of the solution of (7.2) when \({\mathfrak {r}}_0\in L^{2}(\Omega ;H)\).
Definition 7.1
Let \({\mathfrak {r}}_0\in L^{2}(\Omega ;H)\). An adapted process \(\rho \in L^{2}(\Omega ;C([0,T];H)\cap L^{2}(0,T;H^{1}))\) is a solution of Eq. (7.2) if for every \(\varphi \in C^{\infty }_{0}({\mathcal {O}})\)
In order to study Eq. (7.2), we first consider the following approximating problem
with \(0<\epsilon<<1\) (for a similar approach see e.g. [14]).
Lemma 7.2
Assume Hypotheses 1 and 2 and condition (7.7). Then, for every \(\epsilon , T>0\) and every \(\rho _{0}\in L^{2}(\Omega ;H)\), Eq. (7.9) admits a unique solution
Moreover, there exists some \(c_T>0\) such that for every \(\epsilon >0\)
Proof
The uniqueness and the existence of solutions for Eq. (7.9) can be proven by proceeding as in the proof of [18, Theorem 5.1].
In order to prove the energy estimate (7.10), we apply Itô’s formula and we get
Note that
and hence
Therefore, the Gronwall lemma gives (7.10). \(\square \)
Proposition 7.3
Assume Hypotheses 1, 2 and condition (7.7), and fix \({\mathfrak {r}}_0\in L^{2}(\Omega ;H)\). Then, for every \(T>0\), there exists a unique solution
of Eq. (7.2). Moreover, there exists some constant \(c_T>0\) such that
Proof
By proceeding as in [11, Theorem 6.2], we can show that equation (7.2) admits at most one solution in \(L^{2}(\Omega ;C([0,T];H))\cap L^{2}(0,T;H^{1}))\). Hence, if we show that there exists a probabilistically weak solution
such that \({\hat{\rho }} \in \,L^{2}({\hat{\Omega }};C([0,T];H))\cap L^{2}(0,T;H^{1}))\), the existence and uniqueness of a probabilistically strong solution for Eq. (7.2) follows.
Step 1 There exists a filtered probability space \(({\hat{\Omega }}, \hat{{\mathcal {F}}}, \{\hat{{\mathcal {F}}}\}_t, \hat{{\mathbb {P}}})\), a cylindrical Wiener process \(\hat{w}^Q\) associated with \(\{\hat{{\mathcal {F}}}\}_t\) and a process \({\hat{\rho }}\in L^{2}(\Omega ;L^{\infty }(0,T;H))\cap L^{2}(0,T;H^{1}))\) such that
for every \(\varphi \in \,C^\infty _0({\mathcal {O}})\).
Proof of Step 1. According to Proposition 7.2, we know that for every \(\epsilon >0\) there exists a unique solution \(\rho _{\epsilon }\) to Eq. (7.9), and
For every \( h \in \,(0,T)\) and \(t \in \,[0,T-h]\) we have
We have
For \(I^\epsilon _{2}(t,h)\), if \(\epsilon \in \,(0,1)\) we have
and for \(I^\epsilon _3(t,h)\) we have
Finally, for \(I^\epsilon _4(t,h)\), by using a factorization argument as in [13, Theorems 5.11, 5.15], due to the boundedness of \(\sigma _g\) in \({\mathcal {L}}_2(H_Q,H)\) we obtain that for some \(\theta \in \,(0,1)\)
Therefore, by putting together (7.13), (7.14), (7.15) and (7.16), thanks to (7.12) we conclude
and together with the bound
due to [30, Theorem 7] this implies that \(\{\rho ^\epsilon \}_{\epsilon \in \,(0,1)}\) is tight in \(L^\infty (0,T;H^{-\alpha })\), for every \(\alpha >0\). Moreover, since for every \(\beta \in \,(-3,1)\) we have
and the bound
holds, thanks again to [30, Theorem 7], we have that the family \(\{\rho ^\epsilon \}_{\epsilon \in \,(0,1)}\) is tight also in the space \(L^{8/(3+\beta )}(0,T;H^\beta )\), for every \(\beta \in \,(-3,1)\).
In what follows, for every \(\alpha >0\) and \(\beta \in \,(-3,1)\) we denote
where U is any Hilbert space such that the embedding \(H_Q\hookrightarrow U\) is Hilbert–Schmidt. Due to the tightness of \(\{\rho ^\epsilon , w^Q\}_{\epsilon \in \,(0,1)}\) in \({\mathcal {X}}_{\alpha , \beta }(T)\), there exists a sequence \(\epsilon _n\downarrow 0\) such that \({\mathcal {L}}(\rho ^{\epsilon _n}, w^Q)\) is weakly convergent in \({\mathcal {X}}_{\alpha , \beta }(T)\). Due to Skorohod’s Theorem this implies that there exists a probability space \(({\hat{\Omega }}, \hat{{\mathcal {F}}}, \hat{P})\), a sequence of \({\mathcal {X}}_{\alpha , \beta }(T)\)-valued random variables \({\mathcal {Y}}_n=({\hat{\rho }}_n, \hat{w}^Q_n)\) and a \({\mathcal {X}}_{\alpha , \beta }(T)\)-valued random variable \({\mathcal {Y}}=({\hat{\rho }}, \hat{w}^Q)\), all defined on the probability space \(({\hat{\Omega }}, \hat{{\mathcal {F}}}, \hat{P})\), such that
and
Now, we have
and thanks to (7.17) and (7.18), this gives for every \(\varphi \in C^{\infty }_{0}({\mathcal {O}})\),
Thus, by using the general argument introduced in [14, proof of Theorem 4.1], thanks to (7.18) we can take the limit as \(n\rightarrow \infty \) of both sides in the equality above, and we obtain that \({\hat{\rho }}\) satisfies (7.8), with \(w^Q\) replaced by \(\hat{w}^Q\). Moreover, \({\hat{\rho }}\in L^{2}({\hat{\Omega }};L^{\infty }(0,T;H))\cap L^{2}(0,T;H^{1}))\) and satisfies (7.11), with \({\mathbb {E}}\) replaced by \(\hat{{\mathbb {E}}}\).
Step 2 We have that there exists a unique solution \(\rho \in \,L^2(\Omega ;C([0,T];H)\cap L^2(0,T;H^1))\) that satisfies (7.11).
Proof of Step 2 Due to what we have seen above, there exists a unique solution
that satisfies (7.11). It only remains to prove that \(\rho \in \,C([0,T];H)\), \({\mathbb {P}}\)-a.s. By proceeding as in [14, Sect. 4.3] we consider the problem
whose unique solution \(\xi \) belongs to \(L^2(\Omega ;C([0,T];H)\cap L^2(0,T;H^1))\). Then, if we denote \(\eta (t):=\rho (t)-\xi (t)\), we have that \(\eta \in L^{\infty }(0,T;H)\cap L^{2}(0,T;H^{1})\), \({\mathbb {P}}\)-a.s., and solves
Now, if we denote by U(t, s) the evolution family associated with the time-dependent differential operator
we have that
and since \(\xi \in \,L^2(0,T;H^1)\) and \(\rho \in \,L^\infty (0,T;H)\), \({\mathbb {P}}\)-a.s., we get that \(\eta \in \,C([0,T];H)\), \({\mathbb {P}}\)-a.s. In particular, \(\rho =\eta +\xi \) belongs to C([0, T]; H), \({\mathbb {P}}\)-a.s. \(\square \)
7.2 Well-Posedness of Eq. (7.1) in H
From the well-posedness of the quasilinear stochastic parabolic Eq. (7.2), we get the well-poseness of Eq. (7.1) in H. By proceeding as in the proof of [10, Theorem 7.1], we can show that \(u\in L^{2}(\Omega ;C([0,T];H)\cap L^{2}([0,T];H^{1}))\) is a solution to Eq. (7.1) with initial condition \({\mathfrak {u}}_0\in L^{2}(\Omega ;H)\) if and only if \(\rho :=g(u)\) is a weak solution to equation (7.2) with initial value \({\mathfrak {r}}_0=g({\mathfrak {u}}_0)\in L^{2}(\Omega ;H)\). Moreover, as a consequence of the Lipschitz continuity of g and \(g^{-1}\) on \({\mathbb {R}}\), we have the following result.
Proposition 7.4
Assume Hypotheses 1, 2 and condition (7.7). For every \(T>0\) and every \({\mathfrak {u}}_0\in L^{2}(\Omega ;H)\), there exists a unique weak solution \(u\in L^{2}(\Omega ;C([0,T];H)\cap L^{2}(0,T;H^{1}))\), to Eq. (7.1) such that
In what follows, we shall denote
for every \(\varphi \in B_{b}(H)\).
7.3 Some Bounds for \(\rho \) and u
Once established the existence of a unique weak solution in \(L^{2}(\Omega ;C([0,T];H)\cap L^{2}([0,T];H^{1}))\), both for (7.1) and (7.2), we prove some bounds for their solutions h and \(\rho \).
Lemma 7.5
Under Hypotheses 1, 2 and 3, there exist some \(\lambda >0\) and \(c>0\), such that for every \(t\ge 0\)
Proof
We apply Itô’s formula to the process \(\rho (t)\) and the function \(K({\mathfrak {r}})=\Vert {\mathfrak {r}}\Vert _H^2\) and we get
Then thanks to (7.4) and (2.7), together with 7.6, we can find some constant \(\lambda >0\) such that
and this allows we complete the proof. \(\square \)
Due to estimates (7.20) and the Lipschitz continuity of g and \(g^{-1}\) on \({\mathbb {R}}\), estimates analogous to (7.20) holds for the solution u.
Proposition 7.6
Assume Hypotheses 1, 2 and 3. For every \(T>0\) and every \({\mathfrak {u}}_0\in L^{2}(\Omega ;H)\), there exists a unique \(u\in L^{2}(\Omega ;C([0,T];H)\cap L^{2}(0,T;H^{1}))\) which solves Eq. (7.1) in the following sense
for any \(\varphi \in C^\infty _0({\mathcal {O}})\). Moreover, for every \(t\ge 0\)
7.4 Well-Posedness of Eq. (7.2) in \(H^{-1}\)
Here, we will use the results we have just mentioned about the well-posedness of Eq. (7.2) in H, to study its well-posedness in \(H^{-1}\).
Definition 7.7
For every fixed \({\mathfrak {r}}_0\in H^{-1}\) and \(T>0\), an adapted process \(\rho \in L^{2}(\Omega ;L^{2}(0,T;H))\) is a solution of Eq. (7.2) with initial condition \(\rho _0\) if for every \(\varphi \in C^{\infty }_{0}({\mathcal {O}})\)
\({\mathbb {P}}\)-a.s., where
Proposition 7.8
Assume Hypotheses 1, 2, 3 and 4. Then, for every \({\mathfrak {r}}_{0}\in H^{-1}\) and every \(T>0\), there exists a unique solution
to Eq. (7.2). Moreover, there exist \(c, \lambda >0\) independent of \(T>0\) such that for every \(t \in \,[0,T]\)
Proof
We fix an arbitrary sequence \(\{{\mathfrak {r}}_{\epsilon }\}_{\epsilon >0}\subset H\) converging to \({\mathfrak {r}}_0\) strongly in \(H^{-1}\), as \(\epsilon \rightarrow 0\). Thanks to Proposition 7.3, for each \(\epsilon >0\) there exists a solution \(\rho _{\epsilon } \in \,L^{2}(\Omega ;C([0,T];H)\cap L^{2}([0,T];H^{1}))\) for problem (7.2) with initial condition \({\mathfrak {r}}_{\epsilon }\). If for every \(\epsilon ,\delta >0\) we define
we have
Since
we have
so that
where
last inequality following from (2.11).
Hence, if we first integrate both sides in (7.22) with respect to time and then take the expectation, we get
and this implies that the sequence \((\rho _{\epsilon })\) is Cauchy in \(C([0,T];L^2(\Omega ;H^{-1}))\cap L^2(\Omega ;L^{2}(0,T;H)))\). In particular, it converges to some \(\rho \) in \(C([0,T];L^2(\Omega ;H^{-1}))\cap L^2(\Omega ;L^{2}(0,T;H)))\), as \(\epsilon \rightarrow 0\). For every \(\epsilon >0\) and \(\varphi \in \,C^\infty _0({\mathcal {O}})\) we have
Then, due to the Lipschitz continuity of B, \(F_{g}\) and \(\sigma _{g}\), we can take the limit in both sides of the identity above, as \(\epsilon \rightarrow 0\), and we get that \(\rho \) is a solution for (7.2).
To prove the uniqueness, assume that \(\rho _{1},\rho _{2}\) are two solutions to (7.2). By proceeding as above, we have
which gives \(\rho _{1}=\rho _{2}\).
Next, we prove that \(\rho \in \,L^2(\Omega ;C([0,T];H^{-1})\cap L^{2}(0,T;H)))\). We apply Itô’s formula to \(\left\| \rho \right\| _{H^{-1}}^{2}\) and we get
Due to (2.7) there exists \({\bar{\delta }}>0\) such that
so that, thanks to (7.5) we have
Now, since we have
if we integrate both sides in (7.23) and then take the supremum with respect to time and the expectation, we get
which, in particular implies that \(\rho \in \,L^2(\Omega ;L^\infty (0,T;H^{-1})\cap L^{2}(0,T;H)))\). Moreover, since \(\rho \) solves equation (7.2), it belongs to \(C([0,T];H^{-1})\), \({\mathbb {P}}\)-a.s.
Finally, in order to prove (7.21), we take the expectation of both sides of (7.23) and we get
and this implies that there exist some \(c, \lambda >0\) such that (7.21) holds. \(\square \)
8 Ergodic Behavior of the Limiting Equation
We first study the existence of a unique invariant measure for \(R^H_t\) and \(R^{H^{-1}}_t\), and then we show how this implies the existence of a unique invariant measure for \(P^H_t\).
8.1 Ergodicity of Eq. (7.2)
In what follows, we denote by \(R^{H^{-1}}_{t}\) the transition semigroup associated to Eq. (7.2) on \(H^{-1}\)
for every \(\varphi \in B_{b}(H^{-1})\). Similarly, we denote by \(R^{H}_{t}\) the transition semigroup associated to equation (7.2) on H,
for every \(\varphi \in B_{b}(H)\). Clearly, if \({\mathfrak {r}}\in H\) and \(\varphi \in B_{b}(H^{-1})\), then
For every \(A \in \,{\mathcal {B}}(H^{-1})\) we have that \(A\cap H \in \,{\mathcal {B}}(H)\). Thus, if \(\nu \in {\mathcal {P}}(H)\), we can define its extension \(\nu '\in {\mathcal {P}}(H^{-1})\) by setting
With this definition, \(\text {supp}\,(\nu ')\subset H\). Indeed, if we denote by \(B_{H}({\mathfrak {r}},R)\) the closed ball in H centered at \({\mathfrak {r}}\in H\) with radius \(R>0\), then \(B_{H}^{c}({\mathfrak {r}},R)\in {\mathcal {B}}(H)\), so that
which implies that \(\text {supp}\,(\nu ')\subset H\).
Proposition 8.1
Assume Hypotheses 1, 2, 3 and 4, and define \(\alpha ({\mathfrak {r}},{\mathfrak {s}}):=|{\mathfrak {r}}-{\mathfrak {s}}|_{H^{-1}}\). Then, there exist some positive constant \(\lambda _{0},t_{0}\) and c such that
Moreover, \(R^{H^{-1}}_{t}\) has a unique invariant measure \(\nu ^{H^{-1}}\) such that \(\text {{supp}}\,(\nu ^{H^{-1}})\subset H^{1}\) and
Proof
Let \(\rho ^{{\mathfrak {r}}_1},\rho ^{{\mathfrak {r}}_2}\) be two solutions of (7.2), with initial conditions \({\mathfrak {r}}_1, {\mathfrak {r}}_2 \in \,H^{-1}\), respectively. By proceeding as in the proof of Proposition 7.8, we have
for some constant \(\lambda >0\). In particular, the semigroup \(R_t^{H^{-1}}\) is Feller in \(H^{-1}\) and for every \(\varphi \in \text {Lip}_{b}(H^{-1})\) and \({\mathfrak {r}}_1, {\mathfrak {r}}_2\in H^{-1}\)
As shown e.g. in [21, Theorem 2.5], (8.3) implies (8.1). Moreover, it implies that \(R_t^{H^{-1}}\) has at most one invariant measure.
If for every \(R>0\) and \(t>0\) we denote
Then, thanks to (7.20), for every \(R>0\) and \(t>0\) we have
Since \(B_R\) is compactly embedded in \(H^{-1}\), this implies that the family of measures \(\{\Gamma _t\}_{t>0}\), is tight in \(H^{-1}\). Then, by Prokhorov’s Theorem, there exists \(t_n\uparrow \infty \) such that \(\Gamma _{t_n}\) converges weakly to some probability measure in \({\mathcal {P}}(H^{-1})\) which is invariant for \(R^{H^{-1}}_t\) and, due to what we have seen above, such measure is the unique invariant measure \(\nu ^{H^{-1}}\) of \(R^{H^{-1}}_t\). Moreover, (8.4) gives
so that \(\text {supp}\,(\nu ^{H^{-1}})\subset H^{1}\).
Finally, in order to prove (8.2), we first notice that due to the invariance of \(\nu ^{H^{-1}}\) and (7.21)
Thus, if we take \({\bar{t}}>0\) such that \(c e^{-\lambda {\bar{t}}}=1/2\), we get
Then, in view of (8.3), for every \(\varphi \in \,\text {Lip}_b(H^{-1})\) we have
and (8.5) allows to obtain (8.2), with \(\lambda _0=\lambda /2\). \(\square \)
Remark 8.2
Based on the fact that \({\mathcal {B}}(H)\subset {\mathcal {B}}(H^{-1})\) and the fact that \(\text {supp}\,(\nu ^{H^{-1}})\subset H^{1}\), we have that \(\nu ^{H^{-1}}\in {\mathcal {P}}(H^{-1})\) is also a probability measure on H. In what follows, it will be convenient to distinguish the restriction of \(\nu ^{H^{-1}}\) to H from \(\nu ^{H^{-1}}\) itself and for this reason we will denote it by \(\nu ^{H}\).
Proposition 8.3
The probability measure \(\nu ^{H}\) is the unique invariant measure for the transition semigroup \(R_t^{H}\). Moreover, \(\text {{supp}}\,(\nu ^{H})\subset H^{1}\) and
Proof
By proceeding as in the proof of Lemma 6.3, it is possible to show that \(\nu ^{H}\) is invariant for \(R_{t}^{H}\), and from Proposition 8.1 we get that \(\text {supp}(\nu ^{H})\subset H^{1}\).
To prove its uniqueness, we notice that if \(\nu \in {\mathcal {P}}(H)\) is any invariant measure for \(R_{t}^{H}\), then its extension \(\nu '\in {\mathcal {P}}(H^{-1})\), with the support in H, is invariant for \(R_t^{H^{-1}}\). From Proposition 8.1, we have \(\nu '=\nu ^{H^{-1}}\), and hence
which implies that \(\nu =\nu ^{H}\).
Finally, in order to prove (8.6), we consider the Komolgov operator associated to \(R_{t}^{H}\)
We consider the function \(\varphi ({\mathfrak {r}}):=\left\| {\mathfrak {r}}\right\| _{H}^{2}/2\), then
so thanks to (7.4) and (2.7), by the invariance of \(\nu ^{H}\) on H we have
\(\square \)
Remark 8.4
As a direct consequence of (8.6), we have
8.2 Ergodicity for Eq. (7.1)
Now, we recall that we denoted by \({P}^{H}_{t}\) the transition semigroup associated to the limiting problem (7.1)
for every \(\varphi \in B_{b}(H)\). For every \({\mathfrak {r}}, {\mathfrak {u}} \in \,H\) and \(t\ge 0\) we have
Hence, if we define the operator \(T_{g}:C_{b}(H)\rightarrow C_{b}(H)\) by
we have \(T_{g}^{-1}:=T_{g^{-1}}\),
and for every \(\varphi \in \,C_b(H)\)
Lemma 8.5
\(\nu \in \,{\mathcal {P}}(H)\) is invariant for \(P^H_t\) if and only \(\nu \circ g^{-1}\) is invariant for \(R^H_t\). In particular, \(\nu ^H\circ g\) is the unique invariant measure for \(P^H_t\).
Proof
Assume \(\nu \in \,{\mathcal {P}}(H)\) is invariant for \(P^H_t\). Then, thanks to (8.9) and (8.8), for every \(\varphi \in \,C_b(H)\) and \(t\ge 0\) we have
This implies that \(\nu \circ g^{-1}\) is invariant for \(R^H_t\). In the same way, if \(\lambda \in \,{\mathcal {P}}(H)\) is invariant for \(R^H_t\), then \(\lambda \circ g\) is invariant for \(P^H_t\). Hence, we can conclude due to (9.1).
Our statement can be rephrased by saying that there exists a unique invariant measure for \(R^H_t\) if and only is there exists a unique invariant measure for \(P^H_t\). Therefore, since we have shown in Corollary 8.3 that \(\nu ^H\) is the unique invariant measure for \(R^H_t\), we obtain that \(\nu ^H\circ g\) is the unique invariant measure for \(P^H_t\). \(\square \)
9 Proof of Theorem 3.2
Due to Hypothesis 2, with an abuse of notation in this section we will look at g and \(g^{-1}\) as mappings on H
For every probability measure \(\nu \in {\mathcal {P}}(H)\), we define probability measures \(\nu \circ g\) and \(\nu \circ g^{-1}\) \(\in {\mathcal {P}}(H)\) by
Clearly, we have
We notice that Theorem 3.2 is proved once we can show that if \((\nu _{\mu }^{ {\mathcal {H}}})_{\mu >0}\subset {\mathcal {P}}({{\mathcal {H}}})\) is a family of invariant measures for the transition semigroups \( P_t^{\mu , {\mathcal {H}}}\), such that \(\text {supp}(\nu _{\mu }^{ {\mathcal {H}}})\subset {{\mathcal {H}}}_{1}\), then
where \(\nu ^{H^{-1}}\) is the unique invariant measure for \(R_t^{H^{-1}}\) in \(H^{-1}\).
Actually, in view of (6.16), the family of probability measures \((\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}})_{\mu \in (0,1)}\) is tight in \(H^{\delta }\), for every \(\delta <1\). If \(\nu \) is any weak limit of \(\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}}\) in H, as \(\mu \rightarrow 0\), we have \((\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}})\circ g^{-1}\) converges weakly to \(\nu \circ g^{-1}\) on H. Due to the continuity of the embedding of \(H^{-1}\) into H,
as measures on \(H^{-1}\). On the other hand, according to (9.2) we have that \(\left[ (\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}})\circ g^{-1}\right] '\) converges weakly to \(\nu ^{H^{-1}}\) in \(H^{-1}\), so that \((\nu \circ g^{-1})'=\nu ^{H^{-1}}\) in \(H^{-1}\). This implies that \(\nu \circ g^{-1}=\nu ^{H}\in {\mathcal {P}}(H)\), and thus \(\nu =\nu ^{H}\circ g\in {\mathcal {P}}(H)\). Since this holds for every weak limit \(\nu \) of \(\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}}\), we conclude that \(\Pi _{1}\nu _{\mu }^{ {\mathcal {H}}}\) converges weakly to \(\nu ^{H}\circ g\) in H, as \(\mu \rightarrow 0\), and, due to Lemma 8.5, \(\nu ^{H}\circ g\) is the unique invariant measure for \(P_{t}^{H}\).
9.1 Proof of (9.2)
Due to the invariance of \(\nu _\mu ^{{\mathcal {H}}}\) and \(\nu ^{H^{-1}}\), we have
According to (8.1), we have
and then, if we pick \({\bar{t}}>0\) such that \(c e^{-\lambda _0 {\bar{t}}}\le 1/2\), we obtain
Now, if we fix a \({\mathcal {F}}_{0}\)-measurable \({\mathcal {H}}_{1}\)-valued random variable \(\vartheta _\mu :=(\xi _\mu ,\eta _\mu )\), distributed as the invariant measure \(\nu _{\mu }^{ {\mathcal {H}}}\), the Kantorovich–Rubinstein identity (3.3) gives for every \(t\ge 0\)
Thus, (9.2) follows once we prove that for every \(t\ge 0\) large enough
According to (6.16) we have that \(\vartheta _\mu \in L^{2}(\Omega ;{{\mathcal {H}}}_{1})\), for every \(\mu \in (0,1)\). Hence, if we denote \(\rho _\mu (t):=g(u_\mu ^{\vartheta _\mu }(t))\), by proceeding as in [10, Sect. 5], we can rewrite equation (3.1) in the following way
where the identity holds in \(H^{-1}\) sense. Since \(\rho ^{\,g(\xi _\mu )}\) solves Eq. (7.2) with initial condition \(g(\xi _\mu )\in L^{2}(\Omega ;H)\) in \(H^{-1}\) sense, we have
If we define \(\vartheta _\mu (t):=\rho _\mu (t)-\rho ^{\,g(\xi _{\mu })}(t)\), as a consequence of Itô’s formula, we have
\(\vartheta _\mu (t):=\rho _\mu (t)-\rho ^{\,g(\xi _{\mu })}(t)\), as a consequence of Itô’s formula, we have
so that
where
Since B has linear growth, thanks to (6.1) and (6.2) for every \(\mu \in (0,\mu _{0})\) we have
Similarly, thanks to the linear growth of \(F_{g}\), we have for every \(\mu \in (0,\mu _{0})\)
Moreover, due to (6.16) we know the family of random variable \(\vartheta _\mu \) satisfies (6.5). Then, from (6.6) we obtain that for every \(\mu \in (0,\mu _{t})\)
Therefore, from (2.11) and once again (6.16), we conclude that for every \(\mu \in \,(0,\mu _t)\)
and (9.3) follows.
References
Birrell, J., Hottovy, S., Volpe, G., Wehr, J.: Small mass limit of a Langevin equation on a manifold. Ann. Henri Poincaré. Theor. Math. Phys. 18, 707–755 (2017)
Brzezniak, Z., Cerrai, S.: Stochastic wave equations with constraints: well-posedness and Smoluchowski–Kramers diffusion approximation. http://arxiv.org/abs/2303.09717
Cerrai, S., Freidlin, M.: On the Smoluchowski–Kramers approximation for a system with an infinite number of degrees of freedom. Probab. Theory Relat. Fields 135, 363–394 (2006)
Cerrai, S., Freidlin, M.: Smoluchowski–Kramers approximation for a general class of SPDE’s. J. Evol. Equ. 6, 657–689 (2006)
Cerrai, S., Freidlin, M.: Small mass asymptotics for a charged particle in a magnetic field and longtime influence of small perturbations. J. Stat. Phys. 144, 101–123 (2011)
Cerrai, S., Glatt-Holtz, N.: On the convergence of stationary solutions in the Smoluchowski–Kramers approximation of infinite dimensional systems. J. Funct. Anal. 278, 1–38 (2020)
Cerrai, S., Salins, M.: Smoluchowski–Kramers approximation and large deviations for infinite dimensional gradient systems. Asympt. Anal. 88, 201–215 (2013)
Cerrai, S., Salins, M.: Smoluchowski–Kramers approximation and large deviations for infinite dimensional non-gradient systems with applications to the exit problem. Ann. Probab. 44, 2591–2642 (2016)
Cerrai, S., Salins, M.: On the Smoluchowski–Kramers approximation for a system with an infinite number of degrees of freedom subject to a magnetic field. Stoch. Process. Appl. 127, 273–303 (2017)
Cerrai, S., Xi, G.: A Smoluchowski–Kramers approximation for an infinite dimensional system with state-dependent damping. Ann. Probab. 50, 874–904 (2022)
Cerrai, S., Xie, M.: On the small noise limit in the Smoluchowski–Kramers approximation of nonlinear wave equations with variable friction. Preprint at http://arxiv.org/abs/2203.05923
Cerrai, S., Wehr, J., Zhu, Y.: An averaging approach to the Smoluchowski–Kramers approximation in the presence of a varying magnetic field. J. Stat. Phys. 181, 132–148 (2020)
Da Prato, G., Zabczyk, J.: Stochastic Equations in Infinite Dimensions, 2nd edn. Cambridge University Press, Cambridge (2014)
Debussche, A., Hofmanovà, M., Vovelle, J.: Degenerate parabolic stochastic partial differential equations: quasilinear case. Ann. Probab. 44, 1916–1955 (2016)
Földes, J., Glatt-Holtz, N., Richards, G., Thomann, E.: Ergodic and mixing properties of the Boussinesq equations with a degenerate random forcing. J. Funct. Anal. 269, 2427–2504 (2015)
Freidlin, M.: Some remarks on the Smoluchowski–Kramers approximation. J. Stat. Phys. 117, 617–634 (2004)
Freidlin, M., Hu, W.: Smoluchowski–Kramers approximation in the case of variable friction. J. Math. Sci. 179, 184–207 (2011)
Frid, H., Li, Y., Marroquin, D., Nariyoshi, J.F., Zeng, Z.: The strong trace property and the Neumann problem for stochastic conservation laws. Stoch. Part. Differ. Equ. Anal. Comput. 10, 1–59 (2022)
Frid, H., Li, Y., Marroquin, D., Nariyoshi, J.F., Zeng, Z.: The Dirichlet problem for stochastic degenerate parabolic-hyperbolic equations, communications in mathematical. Anal. Appl. 1, 1–71 (2022)
Gess, B., Hofmanovà, M.: Well-posedness and regularity for quasilinear degenerate parabolic-hyperbolic SPDE. Ann. Probab. 46, 2495–2544 (2018)
Hairer, M., Mattingly, J.: Spectral gaps in Wasserstein distances and the 2D stochastic Navier–Stokes equations. Ann. Probab. 36, 2050–2091 (2008)
Herzog, D., Hottovy, S., Volpe, G.: The small-mass limit for Langevin dynamics with unbounded coefficients and positive friction. J. Stat. Phys. 163, 659–673 (2016)
Hottovy, S., McDaniel, A., Volpe, G., Wehr, J.: The Smoluchowski–Kramers limit of stochastic differential equations with arbitrary state-dependent friction. Commun. Math. Phys. 336, 1259–1283 (2015)
Hu, W., Spiliopoulos, K.: Hypoelliptic multiscale Langevin diffusions: large deviations, invariant measures and small mass asymptotics. Electron. J. Probab. 22, 1 (2017)
Kramers, H.: Brownian motion in a field of force and the diffusion model of chemical reactions. Physica 7, 284–304 (1940)
Lee, J.J.: Small mass asymptotics of a charged particle in a variable magnetic field. Asympt. Anal. 86, 99–121 (2014)
Lv, Y., Roberts, A.: Averaging approximation to singularly perturbed nonlinear stochastic wave equations. J. Math. Phys. 53, 1–11 (2012)
Nguyen, H.: The small-mass limit and white-noise limit of an infinite dimensional generalized Langevin equation. J. Stat. Phys. 173, 411–437 (2018)
Salins, M.: Smoluchowski–Kramers approximation for the damped stochastic wave equation with multiplicative noise in any spatial dimension. Stoch. Part. Differ. Equ. Anal. Comput. 7, 86–122 (2019)
Simon, J.: Compact sets in the space \(L^p(0, T;B)\). Ann. Matematica Pura Appl. 146, 65–96 (1986)
Smoluchowski, M.: Drei Vortage über Diffusion Brownsche Bewegung und Koagulation von Kolloidteilchen. Phys. Zeit. 17, 557–585 (1916)
Spiliopoulos, K.: A note on the Smoluchowski–Kramers approximation for the Langevin equation with reflection. Stoch. Dyn. 7, 141–152 (2007)
Funding
Sandra Cerrai—Partially supported by the NSF Grant DMS-1954299—Multiscale analysis of infinite-dimensional stochastic systems.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
There are no conflict of interest, nor conflict of interest related to this manuscript.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Cerrai, S., Xie, M. On the Small-Mass Limit for Stationary Solutions of Stochastic Wave Equations with State Dependent Friction. Appl Math Optim 90, 7 (2024). https://doi.org/10.1007/s00245-024-10153-2
Accepted:
Published:
DOI: https://doi.org/10.1007/s00245-024-10153-2