![MathJax Logo](/templates/jsp/_style2/_tandf/pb2/images/math-jax.gif)
ABSTRACT
In this article, we obtain a central limit theorem and prove a moderate deviation principle for stochastic reaction-diffusion systems with multiplicative noise and non-Lipschitz reaction term.
1. Introduction
Consider the following stochastic reaction-diffusion system:
(1)
(1)
with
, where
is a bounded open set of
, with
, having a
boundary.
For each ,
where the coefficients
are taken in
, the matrices
are non-negative and symmetric for each
and fulfil a uniform ellipticity condition, i.e.
for some positive constants
, and coefficients
are continuous. The operators
act on the boundary of
and are assumed to be either of Dirichlet or of co-normal type. The linear operators
are bounded on
and may be equal to the identity operator if d = 1. Let
be a probability space with an increasing family
of the sub-σ-fields of
satisfying the usual conditions. The noises
are independent cylindrical Wiener processes on
.
Moderate deviations trace back to Cramér (Citation1938) established expansions of tail probabilities about sums of independent random variables defying the normal distribution. Comparing to large deviation, which offers a accurate estimate associated with the law of large number, moderate deviation is commonly applied to provide further estimates concerning the central limit theorem and the law of iterated logarithm.
Let be the solution of a small perturbation of the equation of (Equation1
(1)
(1) ), that is
(2)
(2)
The solution of Equation (Equation2
(2)
(2) ) will tend to the solution, denoted by
, of the following deterministic equation, as
,
(3)
(3)
Assume that the mapping
is locally Lipschitz-continuous and has polynomial growth. The mapping
is globally Lipschitz, without any assumption of boundedness and non-degeneracy. The existence and uniqueness of the solution for Equation (Equation1
(1)
(1) ) is proved in Cerrai (Citation2003). The existence and uniqueness of the solutions for Equation (Equation2
(2)
(2) ) and Equation (Equation3
(3)
(3) ) can be established as that of Equation (Equation1
(1)
(1) ). In Cerrai and Röckner (Citation2004), Cerrai and Röckner had shown that the process
obeys a large deviation principle (abbreviated LDP) in
, for any T>0. Also, they established the large deviations for invariant measures for
in Cerrai and Röckner (Citation2005). Based on these works, in this paper, we then consider the precise deviations of
from
as
in the following way: the asymptotic behaviour of the trajectory
We emphasize that
it is related to central limit theorem (abbreviated CLT) when
, see Theorem 3.4 below;
it provides moderate deviation principle (MDP for short, cf. Dembo & Zeitouni, Citation2000) when
is the deviation scale verifying
(4)
(4) see Theorem 4.3 below. In the sequel, we assume that (Equation4
(4)
(4) ) is in place.
Moderate deviation is an intermediate estimation between the large deviation with deviation scale and the CLT with
. The MDP provides rates of convergence and a way to construct asymptotic confidence interval, see Gao and Zhao (Citation2011) and the references therein. In recent years, there is an increasing interest on the study of MDP. For independent and identically distributed random sequences, Chen (Citation1990, Citation1991, Citation1993) and Ledoux (Citation1992) found the necessary and sufficient conditions for MDP. Moderate deviation was discussed by Djellout and Guillin (Citation2001), Gao (Citation1996, Citation2003), and Wu (Citation1995, Citation1999) for Markov processes as well as Guillin and Liptser (Citation2006) for diffusion processes. For mean field interacting particle models, MDP was worked by Douc et al. (Citation2001). Wang and Zhang (Citation2015) and Wang et al. (Citation2015) discussed moderate deviations for stochastic reaction-diffusion equation and 2D stochastic Navier-Stokes equations, respectively. Recently, there are several works on moderate deviations for SPDEs with jump, see Budhiraja et al. (Citation2016) and Dong et al. (Citation2017).
In this paper, we prove a CLT and establish a MDP for the class of reaction-diffusion systems of the form (Equation1(1)
(1) ). Since we consider the solutions in spaces of continuous functions and there is no Itô formula, we will work directly on heat kernels. Also, because of the weak assumptions on drift and diffusion coefficients, we have to deal with the estimates on the solutions delicately.
The organization of this paper is as follows. The assumptions will be given in Section 2. Section 3 is devoted to prove the CLT. In Section 4, we establish the MDP.
Throughout this paper, c and are positive constants independent of ε and those values may be different from line to line.
2. Preliminaries
In this section, we formulate the equation and state the precise conditions on the coefficients.
Let H be the separable Hilbert space endowed with the scalar product
. We shall denote by A the realization in H of the differential operator
, endowed with the boundary conditions
, i.e.
,
. Set
and
coincides with
or
.
We set and define
. Denoted by C the realization in H of the second-order elliptic operator
, endowed with the boundary conditions
. In Davies (Citation1989), it is proved that C is a non-positive and self-adjoint operator which generates an analytic semigroup
with dense domain given by
, where
is the realization in
of
with the boundary condition
. Moreover, denoted by
the realization of the operator
in
for any
. We see that
and denote all of them by L if there is no confusion.
The covariance operator Q of the Wiener process is a positive symmetric, trace class operator on H. Let
. Then
is a Hilbert space with the inner product
(5)
(5)
Let
denote the norm in
. Clearly, the embedding of
in H is Hilbert-Schmidt, since Q is a trace class operator. Let
denote the space of linear operators S such that
is a Hilbert-Schmidt operator from H to H. Define the norm on the space
by
. Set
the space of all bounded linear operators from
into H and denote by
its norm.
The hypothesis concerns the eigenvalues of A:
Hypothesis 2.1
The complete orthonormal system of H which diagonalizes A is equi-bounded in the sup-norm.
Assume that is a bounded linear operator which satisfies:
Hypothesis 2.2
Q is non-negative and diagonal with respect to the complete orthonormal basis which diagonalizes A, with eigenvalues . Moreover, if
, there exists ϱ (
if d = 2 and
if d>2) such that
.
The assumptions on the coefficients f and g are as follows.
Hypothesis 2.3
The mapping is continuous and
is Lipschitz-continuous, uniformly with respect to
and t in bounded sets of
, i.e., for some
.
We set for any and
,
and for
, define for any
the composition operator
by setting, for any
,
Hypothesis 2.4
The mapping
is locally Lipschitz-continuous, locally uniformly for
, and there exists
and
such that
There exists
such that, for each
, and for some
One of the following two conditions holds:
, where
is as in Hypothesis 2.4 (1) and
;
there exists some a>0,
and
such that, for each
and for some
,
Then the system (Equation1(1)
(1) ) can be rewritten in the following type:
(6)
(6)
where the operator A can be written as A = C + L.
3. Central limit theorem
In this section, our task is to establish the CLT. We will need some results in Cerrai (Citation2003), and we list them here for the convenience of readers.
Proposition 3.1
Under Hypotheses 2.1–2.3 on the coefficients, define the following mappings (7)
(7)
(8)
(8)
Then, (1) (the inequality (3.7) in Cerrai, Citation2003) for any
and
(9)
(9)
where
is any constant;
(2) (the inequality (4.6) in Cerrai, Citation2003) there exists such that γ maps
into itself for any
and for any
(10)
(10)
for some continuous increasing function
such that
.
Proposition 3.2
Theorem 5.3 in Cerrai, Citation2003
Under Hypotheses 2.1–2.4 on the coefficients, for any and for any
and T>0 such a problem admits a unique mild solution in
which is the Banach space of all adapted processes u in
, such that
, where
. Moreover,
(11)
(11)
Similarly as the proof of (Equation11(11)
(11) ), under Hypotheses 2.1–2.4, we have
. The following result is concerned with the convergence of
as
.
Proposition 3.3
Under Hypotheses 2.1–2.4, there exists a constant such that, for all
,
(12)
(12)
Proof.
Define
(13)
(13)
It is immediate to check that
is Lipschitz-continuous in E. Then if we consider
(14)
(14)
and
(15)
(15)
there exist unique mild solutions
and
in
for Equations (Equation14
(14)
(14) ) and (Equation15
(15)
(15) ), respectively. Define
So,
From (Equation9
(9)
(9) ),
(16)
(16)
and one obtains
in view of the Lipschitz continuity of
.
If we take such that
, this yields
Gronwall's inequality implies
Therefore, due to (Equation10
(10)
(10) ),
Next we can repeat the same arguments in the intervals
,
and so on and for any T>s,
(17)
(17)
with the help of (Equation11
(11)
(11) ) for
and
.
Define and
. We can see that
is non-decreasing and denote
. So (Equation17
(17)
(17) ) with p = 1 yields
i.e.
. Thus, Fatou's lemma implies
(18)
(18)
which completes the proof.
Let be the solution of the following SPDE:
with initial value
. Let
be the solution of the following SPDE:
where we need an additional assumption about
.
Hypothesis 3.1
is Fréchet derivative of F w.r.t. the second variable satisfying
for some positive
,
and there exists
such that, for each
and
.
We now come to the main result of this section.
Theorem 3.4
Central Limit Theorem
Under Hypotheses 2.1–3.1, converges to
in the space
in probability, that is, there exists a constant
such that, for all
,
(19)
(19)
Proof.
Define
and we see that
is Lipschitz-continuous in E. Then if we consider
and
There exists unique mild solutions
and
in
, respectively. Define the following mappings
Then
Due to inequality (Equation16
(16)
(16) ), one obtains
If we take
such that
, this yields
(20)
(20)
Consider
There exists a random field
taking values in
such that
Then,
In view of last inequality, the Lipschitz continuity of
and Hypothesis 3.1, it yields
(21)
(21)
Inequalities (Equation20
(20)
(20) ), (Equation21
(21)
(21) ) and Gronwall's inequality implies
Furthermore,
according to inequality (Equation17
(17)
(17) ). Also,
is finitely controlled by some positive constant M from
.
Then, we take such that
,
(22)
(22)
Then we can repeat the same arguments in the intervals
,
and so on and for any T>s,
(23)
(23)
with the help of (Equation11
(11)
(11) ).
Similarly as inequality (Equation18(18)
(18) ) replacing
with
, the proof is completed.
4. Moderate deviations
Let . Then
satisfies the following SPDE:
(24)
(24)
with initial value
. This equation admits a unique solution
, where
stands for the solution functional from
into
.
In this part, we will prove that satisfies an LDP on
with
satisfying (Equation4
(4)
(4) ). This special type of LDP is usually called the MDP of
(cf. Dembo & Zeitouni, Citation2000).
Firstly we recall the general criteria for an LDP given in Budhiraja and Dupuis (Citation2000). Let be a Polish space with the Borel σ-field
.
Definition 4.1
Rate function
A function is called a rate function on
, if for each
, the level set
is a compact subset of
.
Definition 4.2
LDP
Let I be a rate function on . A family
of
-valued random elements is said to satisfy the LDP on
with rate function I, if the following two conditions hold.
(Large deviation upper bound) For each closed subset F of
,
(Large deviation lower bound) For each open subset G of
,
Next, we introduce the Skeleton Equations. The Cameron-Martin space associated with the Wiener process is given by
The space
is a Hilbert space with inner product
Let
denote the class of
-valued
-predictable processes ϕ belonging to
a.s.. Let
. The set
endowed with the weak topology is a Polish space. Define
.
For any , consider the deterministic equation
(25)
(25)
with initial value
and for any
, consider
(26)
(26)
Now we are ready to state the second main result.
Theorem 4.3
Moderate Deviation Principle
Under Hypotheses 2.1–3.1, obeys an LDP on
with speed
and with rate function I given by
with the convention
.
We will adopt the following weak convergence method to prove the MDP.
Theorem 4.4
Budhiraja & Dupuis, Citation2000
For , let
be a measurable mapping from
into
. Let
. Suppose that
satisfies the following assumptions: there exists a measurable map
such that
for every
and any family
satisfying that
converge in distribution as
-valued random elements to h as
,
converges in distribution to
as
;
for every
, the set
is a compact subset of
.
Then the family satisfies an LDP in
with the rate function I given by
with the convention
.
For , set
and define
satisfying
To prove Theorem 4.3, we only need to verify the following two propositions according to Theorem 4.4.
Proposition 4.5
Under the same conditions as Theorem 4.3, for every fixed , let
be such that
converges in distribution to ϕ as
. Then
in
as
.
Proposition 4.6
Under the same conditions as Theorem 4.3, for every positive number , the family
is compact in
.
We start to prove Proposition 4.5 and we need the following lemma and inequalities.
Lemma 4.1
Under the same conditions as Theorem 4.3, for any and
, (Equation25
(25)
(25) ) and (Equation26
(26)
(26) ) replacing F and
with
and
respectively, admit the unique solutions, respectively,
,
in
. Moreover, for any N>0 and
, there exist constants
and
such that for any
,
,
(27)
(27)
and
(28)
(28)
Proof.
Similarly as Theorem 2.4 in Chueshov and Millet (Citation2010), the existence and uniqueness of the solution can be proved. Here, we will prove inequalities (Equation27(27)
(27) ) and (Equation28
(28)
(28) ).
Define the following mappings
Then
In view of the Lipschitz continuity of
and inequality (Equation16
(16)
(16) ), one obtains
If we take
such that
, this yields
Gronwall's inequality implies
Then,
Furthermore,
as similar proof as Theorem 4.2 in Cerrai (Citation2003); due to (4.6) and the proof in Theorem 5.3 in Cerrai (Citation2003), it yields
(29)
(29)
Therefore, for some constant
,
Next we can repeat the same arguments in the intervals
,
and so on and for any T>s, there exists some constant
,
(30)
(30)
Here choosing ε small enough, we get (Equation27
(27)
(27) ).
The proof of (Equation28(28)
(28) ) is very similar to that of (Equation27
(27)
(27) ), we omit it here. The proof of this lemma is complete.
Under the conditions of Lemma 4.1, we have the following three inequalities. Let . Then, from inequality (Equation29
(29)
(29) ),
(31)
(31)
(32)
(32)
and
(33)
(33)
Proof of Proposition 4.5.
Let , where
is the solution of (Equation25
(25)
(25) ) replaced h by ϕ. Then
and
, where
,
satisfy, respectively,
with initial value
and
with initial value
. Define the following mappings
Then
Thus,
There exists a random field
taking values in
such that
which yields
We take
such that
. One obtains
Gronwall's inequality implies
Thus,
With the help of inequalities (Equation30
(30)
(30) )–(Equation33
(33)
(33) ), we get
(34)
(34)
Then we can repeat the same arguments in the intervals
,
and so on and for any T>s,
(35)
(35)
Furthermore, by Chebysev inequality and (Equation30
(30)
(30) ),
(36)
(36)
Finally,
Taking
and for
,
. Then,
(37)
(37)
Similarly as inequality (Equation18
(18)
(18) ), replacing
with
, the proof of the proposition is completed.
Proof of Proposition 4.6
The proof is similar as that of Proposition 4.5 and easier. The proof will be omitted.
Disclosure statement
No potential conflict of interest was reported by the authors.
Additional information
Funding
References
- Budhiraja, A., & Dupuis, P. (2000). A variational representation for positive functionals of infinite dimensional Brownian motion. Probability and Mathematical Statistics-Wroclaw University, 20(1), 39–61. https://doi.org/10.1080/13550280152537111
- Budhiraja, A., Dupuis, P., & Ganguly, A. (2016). Moderate deviation principle for stochastic differential equations with jump. Annals of Probability, 44(3), 1–40. https://doi.org/10.1214/15-AOP1007
- Cerrai, S. (2003). Stochastic reaction-diffusion systems with multiplicative noise and non-Lipschitz reaction term. Probability Theory and Related Fields, 125(2), 271–304. https://doi.org/10.1007/s00440-002-0230-6
- Cerrai, S., & Röckner, M. (2004). Large deviations for stochastic reaction-diffusion systems with multiplicative noise and non-Lipschitz reaction term. Annals of Probability, 32(1), 1–40. https://doi.org/10.1214/aop/1079021473
- Cerrai, S., & Röckner, M. (2005). Large deviations for invariant measures of stochastic reaction-diffusion systems with multiplicative noise and non-Lipschitz reaction term. Annales de l'Institut Henri Poincare (B) Probability and Statistics, 41(1), 69–105. https://doi.org/10.1016/j.anihpb.2004.03.001
- Chen, X. (1990). Probabilities of moderate deviations for B-valued independent random vectors. Chinese Journal of Contemporary Mathematics, 11, 381–393.
- Chen, X. (1991). Probabilities of moderate deviations for independent random vectors in a Banach space. Chinese Journal of Applied Probability and Statistics, 7, 24–33.
- Chen, X. (1993). On lower bound of the moderate deviations of i.i.d. random variables in Banach space. Chinese Journal of Applied Probability and Statistics, 9, 386–393.
- Chueshov, I., & Millet, A. (2010). Stochastic 2D hydrodynamics type systems: Well posedness and large deviations. Applied Mathematics and Optimization, 61(3), 379–420. https://doi.org/10.1007/s00245-009-9091-z
- Cramér, H. (1938). Sur un nouveau théorème limite de la théorie des probabilités. Actualités Scientifiques et Industrielles, 736, 5–23. https://doi.org/10.1007/978-3-642-40607-2_8
- Davies, E. B. (1989). Heat kernels and spectral theory. Cambridge University Press.
- Dembo, A., & Zeitouni, O. (2000). Large deviations techniques and applications. Springer-Verlag.
- Djellout, H., & Guillin, A. (2001). Moderate deviations for Markov chains with atom stochastic processes and their applications. Stochastic Processes and Their Applications, 95(2), 203–221. https://doi.org/10.1016/S0304-4149(01)00100-4
- Dong, Z., Xiong, J., Zhai, J. L., & Zhang, T. S. (2017). A moderate deviation principle for 2-D stochastic Navier-Stokes equations driven by multiplicative Lévy noises. Journal of Functional Analysis, 272(1), 227–254. https://doi.org/10.1016/j.jfa.2016.10.012
- Douc, R., Guillin, A., & Najim, J. (2001). Moderate deviations for particle filtering. The Annals of Applied Probability, 15(1B), 587–614. https://doi.org/10.1214/105051604000000657
- Gao, F. Q. (1996). Moderate deviations for martingales and mixing random processes. Stochastic Processes and Their Applications, 61(2), 263–275. https://doi.org/10.1016/0304-4149(95)00078-X
- Gao, F. Q. (2003). Moderate deviations and large deviations for kernel density estimators. Journal of Theoretical Probability, 16(2), 401–418. https://doi.org/10.1023/A:1023574711733
- Gao, F. Q., & Zhao, X. (2011). Delta method in large deviations and moderate deviations for estimators. The Annals of Statistics, 39(2), 1211–1240. https://doi.org/10.1214/10-AOS865
- Guillin, A., & Liptser, R. (2006). Examples of moderate deviation principle for diffusion processes. Discrete and Continuous Dynamical Systems -- Series B, 6(4), 803–828. https://doi.org/10.3934/dcdsb.2006.6.803
- Ledoux, M. (1992). Sur les déviations modérées des sommes de variables aléatoires vectorielles indépendants de même loi. Annales de l'IHP Probabilités et statistiques, 28, 267–280.
- Wang, R., & Zhang, T. S. (2015). Moderated deviations for stochastic reaction-diffusion equations with multiplicative noise. Potential Analysis, 42(1), 99–113. https://doi.org/10.1007/s11118-014-9425-6
- Wang, R., Zhai, J. L., & Zhang, T. S. (2015). A moderate deviation principle for 2-D stochastic Navier-Stokes equations. Journal of Differential Equations, 258(10), 3363–3390. https://doi.org/10.1016/j.jde.2015.01.008
- Wu, L. M. (1995). Moderate deviations of dependent random variables related to CLT. Annals of Probability, 23(1), 420–445. https://doi.org/10.1214/aop/1176988393
- Wu, L. M. (1999). Forward-Backward martingale decomposition and compactness results for additive functionals of stationary ergodic Markov processes. Annales De L'Institut Henri Poincare (B) Probability and Statistics, 35(2), 121–141. https://doi.org/10.1016/S0246-0203(99)80008-9