Abstract
A 2-dimensional stochastic Burgers equation with dissipative term perturbed by Wiener noise is considered. The aim is to prove the well-posedness, existence, and uniqueness of invariant measure as well as strong law of large numbers and convergence to equilibrium.
1. Introduction
The paper is concerned with the 2-dimensional Burgers equation in a bounded domain with Wiener noise as the body forces like this where is the velocity field, is viscid coefficient, denotes the Laplace operator, represents the gradient operator, stands for the -Wiener process, and is a regular bounded open domain of . Burgers equation has received an extensive amount of attention since the studies by Burgers in the 1940s (and it has been considered even earlier by Beteman [1] and Forsyth [2]). But it is well known that the Burgers’ equation is not a good model for turbulence since it does not perform any chaos. Even if a force is added to equation, all solutions will converge to a unique stationary solution as time goes to infinity. However, if the force is a random one, the result is completely different. So, several authors have indeed suggested to use the stochastic Burgers’ equation to model turbulence, see [3–6]. The stochastic equation has also been proposed in [7] to study the dynamics of interfaces.
So far, most of the monographs concerning the equation focus on one-dimensional case, for example, Bertini et al. [8] solved the equation with additive space-time white noise by an adaptation of the Hopf-cole transformation. Da Prato et al. [9] studied the equation via a different approach based on semigroup property for the heat equation on a bounded interval. The more general equation with multiplicative noise was considered by Da Prato and Debussche [10]. With a similar method, Gyöngy and Nualart [11] extended the Burgers equation from bounded interval to real line. A large deviation principle for the solution was obtained by Gourcy [12]. Concerning the ergodicity, an important paper by Weinan et al. [13] proved that there exists a unique stationary distribution for the solutions of the random inviscid Burgers equation, and typical solutions are piecewise smooth with a finite number of jump discontinuities corresponding to shocks. For model with jumps, Dong and Xu [14] proved that the global existence and uniqueness of the strong, weak, and mild solutions for a one-dimensional Burgers equation perturbed by Lévy noise. When the noise is fractal, Wang et al. [15] get the well-posedness.
The main aim in our paper is to study the large time behavior of stochastic system. There are lots of the literature about the topic (see [16–20]).
Burgers system is a well-known model for mechanics problems. But as far as we know, there are no results about the long-term behavior of stochastic Burgers’ system. We think that the difficulty lies in the fact that the dissipative term cannot dominate the nonlinear term . However, in many practical cases, we cannot ignore the energy dissipation and external forces, especially considering the long-term behavior. Therefore, we introduce dissipative term and study the ergodicity of the following equation: where , denote the absolute value or norm for the real number or two-dimensional vector, respectively.
We believe that our work is new and is worth researching. The methods and results in this paper can be applied to stochastic reaction diffusion equations and stochastic real valued Ginzburg Landau equation in high dimensions. But we cannot extend our result to dynamical systems with state-delays. Since in order to show the existence of an invariant measure, we should consider the segments of a solution. In contrast to the scalar solution process, the process of segments is a Markov process. We show that the process of segments is also Feller and that there exists a solution of which the segments are tight. Then, we apply the Krylov-Bogoliubov method. Since the segment process has values in the infinite-dimensional space , boundedness in probability does not automatically imply tightness. For solution processes of infinite-dimensional equations, one often uses compactness of the orbits of the underlying deterministic equation to obtain tightness. For an infinite-dimensional formulation of the functional differential equation, however, such a compactness property does not hold. For ergodicity of stochastic delay equations, we can see [21]. We believe that stochastic Burgers’ system with state-delays is a very interesting problem.
In order to study ergodicity of problem (2), we use a remarkable dissipativity property of the stochastic dynamic to obtain the existence of the invariant measure. For uniqueness, we try to use the method from [22] to prove that the distributions induced by the solution are equivalent. It is well known that the equivalence of the distributions implies uniqueness, a strong law of large numbers, and the convergence to equilibrium.
The remaining of this paper is organized as follows. Some preliminaries are presented in Section 2, the local existence and global existence are presented, respectively, in Sections 3 and 4. In Section 5, we obtain the existence and uniqueness of the invariant measure as well as strong law of large numbers, and convergence to equilibrium. As usual, constants may change from one line to the next; we denote by a constant which depends on some parameter .
2. Preliminaries on the Burgers Equation
Let be a row vector valued function on . And it denotes the following: Let be infinitely differentiable 2-dimensional vector field on , and let be infinitely differentiable 2-dimensional vector field with compact support strictly contained in . We denote by the closure of in , whose norms are denoted by , when . Let be the closure of in and whose norms are denoted by and , respectively. Without confusion, set as the inner product in or . For , let be the norm of vector filed in Lebesgue spaces . represents the norm in the usual sobolev spaces for real valued functions on and ; stands for the norm in the usual Lebesgue spaces for real valued functions on . Denote ; then and . Since coincides with , we can endow with the norm . The operator is positive self-adjoint with compact resolvent; we denote by the eigenvalues of , and by the eigenvectors which is a corresponding complete orthonormal system in satisfying for some positive constant C. We remark that . We define the bilinear operator as for all . Then, (2) is equivalent to the following abstract equation: is the Wiener process having the following representative: in which and are a sequence of mutually independent 1-dimensional Brownian motions in a fixed probability space adapted to a filtration .
It can be derived from [23] that the solution to the linear problem corresponding to (2) with the following initial condition: is unique, and when , it has the form of Let then is a solution to (2) if and only if it solves the following evolution equation: So, we see that when is fixed, this equation is in fact a deterministic equation. From now on, we will study the equation of the form (11) to get the existence and uniqueness of the solution a.s. .
3. Local Existence in Time
Definition 1 (see Definition 5.1.1 in [24]). We say a adapted process is a mild solution to (11), if and it satisfies
Lemma 2. For any , if , then has a version which is -Hölder continuous with respect to , with any .
Proof. Let and ; then Then, we have So, by the estimate of and , we arrive at For , , we get Therefore, As is a Gaussian random variable, we obtain for By Kolmogorov’ test theorem, we get the conclusion.
Remark 3. An example of the noise satisfying condition of Lemma 2 is where is a sequence of independent 1-dimensional Brownian motion, and satisfies It is so because the eigenvalues of the operator , in 2-dimensional space, behave like .
Remark 4. Another example of stochastic noise satisfying Lemma 2 is where , is an isomorphism in , and
To prove the local existence of the solution of (1) in sense of Definition 1, we introduce the space defined by where which in fact is a stopping time and , .
Lemma 5. For , and is adapted to , ; then there exists a unique mild solution in sense of Definition 1 to (11) in .
Proof. Choose a in , and set
Then,
For the second term on the right hand side of (25),
In the following, we will estimate , respectively, . Since is contraction on , it is known that
for all , , , , and only depends on , , and . Before calculating each , we outline the Sobolev embedding principle in fractional Sobolev spaces as follows:
when
where is the dimension of the spatial. Let , , , satisfying (29) such that
For , by (27) and Theorem A.8 in [25], we get
where
satisfying
The last inequality follows by (30). For the other term added to , we have
So, by (31)–(34), we have
Similarly, we get for that
For , by Theorem A.8 in [25], we get
where
For , we have
For the first term on the right hand side of (37), by (27), we have
For the second term on the right hand side of (37), by (27), we obtain
From (37) to (41), we get for that
Analogously, for , we get
By (26), (35), (36), (42), and (43), we have
As , by (44), for , we have
Since by Lemma 2,
For the last term on the right hand side of (25), we have
Therefore,
So by (25), (45), and (48), when is small enough,
For each , set , . To simplify the notation in the following calculation, we denote , . Then,
So,
In order to simplify the notation, we set
where
Then, we estimate , , respectively. For , we have
We first consider
For the other term added to ,
By (54)–(56),
Analogously, for ,
For , by (53), we have
For the first term on the right hand side of (59), we have
For ,
For the first term on the right hand side of (60), we arrive at
For the second term on the right hand side of (60), we obtain
By (59)–(63), we get for that
Similarly, we get for that
By (52), (53), (57), (58), (64), and (65), we have
For the second term on the right hand side of (51), we have
where
Then,
Similarly, we can get the same estimate for . So, we have
By (51), (66), and (70), we have
By (49), (71), and fixed point principle, we get the conclusion.
Remark 6. By making some minor modifications in the proof of Lemma 5, we can see that the conclusion in Lemma 5 is also true for (1). Our original aim is to get the global well-posedness of (1), but we find that the dissipative term cannot dominate the nonlinear term . So, we introduce the dissipative term which will also play an important role in obtaining the ergodicity.
4. Global Existence
Theorem 7. With conditions in Lemma 2, for satisfying (12), when , one has Subsequently, one gets the existence of the global solution belonging to .
Proof. Let be a sequence of vectors which satisfies and , , such that in sense of . Let be a sequence of regular process, such that in when or . For , , , where . Then, by (74), we have If satisfies then, is regular, such that Taking inner product with respect to in (78), we have For simplicity, we calculate the third term on the left hand side of (79) first as follows: where . For , we have In the following, we estimate the four terms for , respectively. For the first term, For the second term, by (75), we have similarly, for the third term, For the last term, by (75) and (76), By (81)–(85), it follows that Similarly, For , For the first term on the right hand side of (88), we deduce that where . For the second term on the right hand side of (88), we have Analogously, for the third term on the right hand side of (88), we see that For the last term, by (75) and (76), we have By (88)–(92), we get Analogously, for , it follows that By (80) and the estimates of , and , see (86), (87), (93), and (94), we have For the last term on the left hand side of (79), we have By (79), (95), and (96), we get Rearranging the above inequality, we deduce that Let , and be small enough, such that So, we integrate with respect to on both sides of (98) to obtain where , by Gronwall’s inequality, we arrive at By (100) and (101), we have Multiplying on both sides of (78), and integrating with respect to , we have which is equivalent to We first estimate the second term on the right hand side of (104) as follows: For , we have For , we have By interpolation inequality, there exists some , such that Then, where the last inequality follows from (101). For , we deduce that For , we arrive at For , we obtain By (106) and (109)–(112), Similarly, for , we infer that For , we have By interpolation inequality and (101), we deduce that For , we have Similarly, for , As for , we get By (115)-(119), we arrive at Analogously to , we have By (105) and the estimates of , see (113), (114), (120), and (121), we get that For the first term on the right hand side of (104), we have By (104), (122), and (123), By the Gronwall inequality, we get Let , by Fatou Lemma,
5. Invariant Measures
5.1. Existence
In this section, we will establish the existence of invariant measure for (2). Analogously to [24], we extend the Wiener process to by setting where is another -valued Wiener process satisfying conditions in Lemma 2 and being independent of . For any , we consider the following equation: By Theorem 7, we know that there exists unique solution. In order to obtain the invariant measure, we should show that the family of laws is tight. Since is compact, for any , we only need to show that is bounded in probability in . As we know, is the mild solution of (8) with the following initial condition: Making the classical change of variable , (128) is equivalent to with initial condition In order to get the invariant measure of (131), it is enough to show that is bounded in probability in , for some . That is what we have to do in Theorem 8 below.
Theorem 8. With conditions in Lemma 2, when , there exists an invariant measure for (2).
Proof. Multiplying (131) by and integrating on , we get
For the third term on the left hand side of (133), we deduce that
Substituting (134) into (133), we have
For the third term on the right hand side of (135), we get by the Young inequality that
For the last term on the right hand side of (135),
Since is vector field, we denote it by , where is real valued function, . For , we have
Similarly for ,
Analogously to , we deduce that
For , we have
Since is a Gaussian process, we infer that
Then, with the proof of Lemma 2, we know that is continuous with respect to . By (137)–(141), we have
By (135), (136), and (143), we arrive at
It is equivalent to
Since , let be small enough, such that
Then, the above estimates can be changed into
By the Gronwall inequality, we get
Similarly to the argument of [26], we will prove that has at most polynomial growth, when a.s. So, we conclude that
Multiplying on both sides of (147) and integrating with respect to , we have
As
by (149), we have
By Theorem 7, we know that for problem (131) there exists unique mild solution, which has the following:
Then, for any , where the is the parameter in Lemma 2,
Since
then,
For , we have
In the following, we use Theorem 6.13 in chapter two of [27] to estimate them respectively as follows:
the last inequality follows by Theorem A.8 in [25], where , , and . So, by Hölder inequality and interpolation inequality, we have
For , we have
where
Analogously to estimating , we have
Similarly, we can get the same estimates for and . Therefore,
Analogously to estimating , we can get for , and that
So, by (163)–(164) and (156), we get
For the third term on the right hand side of (154), we obtain
since and are bounded for , the last inequality follows. For the first term on the right hand side of (154), we have
Similar to [26], we can prove that has at most polynomial growth when . For the reader convenience, we sketch a proof. By Lemma 2, we know that is a valued Brownian motion, for . So, by the law of iterated logarithm, we have
Obviously, is a i.i.d sequence. By the law of large numbers, there exists an integer-valued random variable , when , we have
This implies that
for all . In other words,
when . By the law of iterated logarithm, we have
for some positive random variable. By Theorem 5.14 in [23], we know that
So, we have that
since , the fourth inequality follows. By (167) and (174), we know that
If we let , repeating the argument of (174), we can see that also has at most polynomial growth, when a.s., since we have the Sobolev embedding . Consider the second term on the right hand side of (154), by (165),
where the last inequality follows by (152). Analogously, we can prove that
where we used (149) and (152) for the last inequality. By (154) and (175)–(177), we get
for some positive random variable . As is compact, by Prohorov Theorem, we know that the family of laws for taking values in is tight. Since , then so does the law of taking values in the same space. For , set
where . Following the arguments in [24], for all and all , by proving
we can show that is a Markov process. Here, is the -algebra generated by for . So, is the Markov semigroup. Define a dual semigroup in the space of probability measures on as follows:
Let be the law of , which is the solution of (2) with initial condition . Then, we have
where we use the fact that and have the same law, the second equality follows. Therefore,
Since is tight, then by Prokhorov theorem, we know that is relatively compact. We can choose a subsequence of denoted by such that for ,
5.2. Uniqueness
The main result of this part is as follows.
Theorem 9. Assume in Lemma 2 and ; then,(i)the stochastic Burgers equation (2) has a unique invariant measure ;(ii)for all and all Borel measurable functions , , such that , (iii)for every Borel measure on , one has that where stands for the total variation of a measure. In particularly, one has that for every Borel set (the Borel -algebra of ).
In order to prove Theorem 9, we only need Theorem 10 below, see [28, Theorem 4.2.1]. We define , , to be the transition probability measure that is, for .
Theorem 10. Assume that the probability measures , are all equivalent, in the sense that they are mutually absolutely continuous. Then, Theorem 9 holds true.
In the following, we will prove the irreducibility and the strong Feller property in to get the equivalence of the measure . For the two notations, we outline them below. For , let (I) For any , such that for all , for each .(S)For all , every , and all such that in , it holds that
Before checking the condition (I), we need Lemma 11 below. For and , set where is solution of the following equation: for , with initial condition . As it is proved in previously this equation has a unique solution as follows: when and .
Lemma 11. Define ; then, (i)the mapping is continuous, where for Banach space ;(ii)for every and there exists such that .
Proof. (i) is proved by in the Appendix. To prove (ii), let and , define as Obviously, . Define as the solution of the following equation: with initial condition ; then . Set ; then it satisfies all the requirements of the lemma.
Proposition 12. With conditions in Theorem 9, the irreducibility property (I) is satisfied.
Proof. Let and be the same as (ii) in Lemma 11. By the above lemma, we have that for , we can find , such that implies that If in Lemma 2, and denote and the corresponding Ornstein-Uhlenbeck process satisfying conditions in the lemma, then . Choose such that and Then, for , we have that Recall now that the solution of the stochastic Burgers equation is equal to , being the Ornstein-Uhlenbeck process. Then, it remains to show that But this is obviously true. So far, we have proved that for for all , , for all , Next, we will prove for all , the above inequality also holds. Indeed, for , by Chapman-Kolmogorov equation, we have Since , we will extend (204) to the case for all . If this is not true, there exists , , such that Then, we can choose , such that . By (204), we have which is contrary to (205).
In this part, it is time to check the condition (S).
We will first obtain the strong Feller property in for modified Burgers equation (208) below, then let to check the condition (S).
Fix , let satisfy such that , and Consider the following equation:
Proposition 13. There exists a unique mild solution for (208) which is Markov process with the Feller property in , that is for every , , there exists a constant such that holds for all , and all , where is the transition probabilities corresponding to (204).
Proof. The proof of existence and uniqueness is similar to Section 2. Let in , by the Gronwall inequality, we know that is Lipschitz continuous with respect to initial value. Using the method in Proposition 4.3.3 in [24], we can prove that the solution is a Markov process. To prove the Fell property, we first consider the following Galerkin approximations of (208). Let be the orthogonal projection in defined as . Clearly, for every . Consider the equation in as follows:
with initial condition . This is a finite-dimensional equation with globally Lipschitz nonlinear functions, so it has a unique progressively measurable solution with -. trajectory , which is also a Markov process in with associated semigroup defined as
for all and . For every , we can prove that there exists a constant such that
hold for all , and all with . Indeed, the following remarkable formula holds true for the differential in of [29]:
for all , where is a -dimensional standard Wiener process with incremental covariance and is the covariance operator of . Obviously, is nonnegative, adjoint, Hilbert-Schmidt operator with inverse. Since the eigenvalues of the Stokes operator , in -space dimension, behave like , let for some , in Lemma 2, we have , where is the image of . Therefore,
Since for
it follows that
where the last inequality follows by the Estimate 4 of the Appendix (note that is independent of and ). Indeed, is given by , where is the Ornstein-Uhlenbeck process, and is the solution of . Therefore,
In the following step, we will let to get the Fell property for (208). Let and be given. From the Appendix, Remark A.1, we know that converges to strongly in , -a.s.. By the boundedness and continuous of as well as Lebesgue dominated convergence theorem, we have
which implies that for some subsequence ,
for a.e. . Take , by the previous argument, we can find a subsequence such that the previous almost sure convergence in holds true both and .
Thus, from (212), we have
for a.e. . As has continuous trajectories with values in , the above inequality holds for all .
Proposition 14. Under conditions of Theorem 9, (S) holds true.
Proof. Take satisfying in . For every , we have that as by Proposition 13. Then, where the inequality follows by the consistency of and , when , and the limit follows by . Therefore, as .
6. Example
Our theory can be applied to stochastic reaction diffusion equations or stochastic real valued Ginzburg Landau equation in high dimensions as follows: where is the velocity field, denotes the Laplace operator, stands for the -Wiener process, and is a regular bounded open domain of .
Appendix
Fix and let satisfy such that and Consider the following equation: where .
Estimate 1. We have the following estimate in for (A.2): where indicates a constant depending on . Analogously to the derivation of (147), we get Therefore, for all , Then, we get (A.3).
Estimate 2. We obtain the following estimate in for (A.2):
Since we have
the equation is equivalent to
Denote by and ; then
As
so, we have that
For the first term on the right hand side of (A.3), we have
Substitute (A.11) and (A.12) into (A.8), we get
Denote
Then,
where
For , we have
where
So,
Analogously, we can get the same estimate for and . Take advantage of the estimates for , , and , we have
By the Gronwall inequality and (A.3), we get (A.6).
Remark A.1. It is standard to show that, for and , there exists a subsequence which converges to some , strongly in , weekly in , and weak star in . Therefore, we have
Estimate 3. We compare, only in the case . Let be two solutions with the same initial condition but with different functions , , there exists a constant , such that
for every . We have
with initial condition , for . Set . Then,
Take inner product in with respect to , we have
For the third term on the left hand side of (A.23), we have
Similarly, we can get
By (A.23)–(A.27), we have
So, by the Gronwall inequality and (A.6), we get (A.21). By (A.6), we know that converges week star to in , for , we have
Estimate 4. Let us consider only the case , and denote by the solution to (A.2). Let be the differential mapping in the direction at point , defined by, for given as follows:
Set also
so that is also the differential of the mapping in the direction at the point . Thus, satisfies
So,
Therefore,
By the Gronwall inequality and (A.6), we have
And therefore, using again the previous inequality,
Acknowledgments
The authors are thankful to the referee for careful reading and insightful comments which led to many improvements of the earlier version. This work was partially supported by the Fundamental Research Funds for the Central Universities (Grant no. CQDXWL-2013-003).