Schauder-type estimates for higher-order parabolic SPDEs
aa r X i v : . [ m a t h . P R ] M a y Schauder-type estimates for higher-order parabolic SPDEs
Yuxing Wang · Kai DuAbstract
In this paper we consider the Cauchy problem for m -order stochastic partial differential equations ofparabolic type in a class of stochastic Hölder spaces. The Hölder estimates of solutions and their spatial derivativesup to order m are obtained, based on which the existence and uniqueness of solution is proved. An interestingfinding of this paper is that the regularity of solutions relies on a coercivity condition that differs when m is odd oreven: the condition for odd m coincides with the standard parabolicity condition in the literature for higher-orderstochastic partial differential equations, while for even m it depends on the integrability index p . The sharpness ofthe new-found coercivity condition is demonstrated by an example. Keywords higher-order stochastic partial differential equations · coercivity condition · Hölder spaces · Schauderestimates
Mathematics Subject Classification (2010) · · Let ( Ω, F , ( F t ) t ≥ , P ) be a complete filtered probability space and { w k · } a sequence of independent standardWiener processes adapted to the filtration F t . Consider the Cauchy problem for the following m -order stochasticpartial differential equations (SPDEs) of non-divergence form: d u = (cid:20) ( − m +1 X | α | , | β |≤ m A αβ D α + β u + f (cid:21) d t + ∞ X k =1 (cid:20) X | α |≤ m B kα D α u + g k (cid:21) d w kt , (1.1)where the coefficients, the free terms, and the unknown function are all random fields defined on R n × [0 , ∞ ) × Ω and adapted to F t . Typical examples of Equation (1.1) include the Zakai equation (see [38,33] for example),linearised stochastic Cahn–Hilliard equations (see [6,3] for example), and so on. General solvability theory forhigher-order SPDEs of type (1.1) was first investigated in [26] under the framework of Hilbert spaces. This paperconcerns the existence, uniqueness and regularity of solutions of (1.1) in some Hölder-type spaces that will bedefined later. Regularity theory for linear equations often plays an important role in the study of nonlinear stochasticequations, see [36,4,7] and references therein.The weak solution of Equation (1.1), which satisfies the equation in the (analytic) distribution sense, and itsregularity in the framework of Sobolev spaces have been investigated by many researchers. Results for the second-order case (namely m = 1 ) are numerous and fruitful; for instance, a complete L p -theory ( p ≥ of second-order parabolic SPDEs has been developed, see for example [30,25,26,33] for p = 2 and [19,20,21,22,23] for K. Du was partially supported by the National Natural Science Foundation of China (No. 11801084).Yuxing WangSchool of Mathematical Sciences, Fudan University, Shanghai 200433, ChinaE-mail: [email protected] Du (corresponding author)Shanghai Center for Mathematical Sciences, Fudan University, Shanghai 200438, ChinaE-mail: [email protected] Y. Wang, K. Du p ≥ ; degenerate equations were addressed in, for example, [27]; and the Dirichlet problem were also extensivelystudied in many publications such as [24,15,16,5,17,28,14,10]. For higher-order SPDEs, Krylov and Rozovskii[26] applied their abstract result to obtain the existence and uniqueness of solutions in the Sobolev space W m ( R n ) .Recently, van Neeven et al. [35] and Portal and Veraar [32] obtained some maximal L p -regularity results for strongsolutions of abstract stochastic parabolic time-dependent problems, which can also apply to higher-order SPDEswith proper conditions.Another approach to the regularity problem of SPDEs is based on some Hölder spaces, corresponding to thecelebrated Schauder theory for classical elliptic and parabolic PDEs (see [13] and references therein). This paperadopts this strategy to study Equation (1.1), stimulated by recent progress of the related research on second-orderSPDEs. Actually, a C δ -theory for (1.1) with m = 1 was once an open problem proposed by Krylov [20], whichwas partially addressed by Mikulevicius [29], and generally solved by Du and Liu [11] very recently. Introducinga Hölder-type space C δp containing all random fields u satisfying k u k C δp := (cid:20) sup t, x E | u ( x, t ) | p + sup t, x = y E | u ( x, t ) − u ( y, t ) | p | x − y | δp (cid:21) p < ∞ with some constants δ ∈ (0 , and p ∈ [2 , ∞ ) , they proved that, under natural conditions on the coefficients, thesolution u and its derivatives Du and D u belong to C δp , provided that f , g and Dg belong to the same space.In addition, Du and Liu [11] also obtained Hölder continuity in time of D u with time irregular coefficients. Asimilar C δ -theory was also obtained recently for systems of second-order SPDEs in [12].This paper aims to prove a Schauder-type estimate for Equation (1.1) based on the space C δp . To get moreinsight into such a kind of regularity of higher-order equations, let us recall some relevant work on deterministicPDEs. Boccia [1] derived Schauder estimates for solutions of m -order parabolic systems of non-divergence formin the classical C m + δx -space provided that the free term f (there are no terms like g k in deterministic equations)belongs to C δx , and for the divergence form Dong and Zhang [9] obtained C m + δx regularity. Considering the featureof stochastic integral terms in SPDEs, a natural form of Schauder estimates for Equation (1.1) must be like this: the C δp -norms of u and its derivatives up to order m are dominated by the C δp -norms of f and D α g with | α | ≤ m .What surprises us during this work is not the above natural assertion but the structural condition that ensuresthe validity of this assertion. Let us give some explanation. It is well-known that the classical Schauder estimatefor PDEs or PDE systems is based on certain coercivity conditions imposed on the leading coefficients and usu-ally called strong ellipticity or strong parabolicity, and for second-order SPDEs either L p -theory or C δ -theoryrequires a stochastic version of such conditions (see [20,11] for example). The solvability result of higher-orderSPDEs in the space W m ( R n ) obtained [26] relied on the following condition: there is a constant λ > such thatfor all ξ α ∈ R , X | α | = | β | = m A αβ ξ α ξ β − λ X | α | = m | ξ α | ≥ ∞ X k =1 (cid:12)(cid:12)(cid:12)(cid:12) X | α | = m B kα ξ α (cid:12)(cid:12)(cid:12)(cid:12) . (1.2)This is a natural condition as it can reduce to the standard ones for PDEs and for second-order SPDEs. However,things may change when one considers L p -integrability ( p > ) rather than square-integrability; more specifically,the coercivity condition (1.2) being adequate for L -theory seems not to be sufficient for L p -integrability of so-lutions or their derivatives when m ≥ . An indirect evidence is that, when the abstract maximal L p -regularityresults obtained in [35,32] applied to higher-order SPDEs of type (1.1) the coefficients B α with | α | = m wererequired to either be sufficiently small or have some additional analytic properties (see [32] for details). Similarphenomena have been found also in complex valued SPDEs (see [2]) and systems of second-order SPDEs (see [18,12]). This seems to be a unique feature of stochastic equations in contrast to deterministic PDEs.A major contribution of this paper is the finding of a p -dependent coercivity condition that is just a smallmodification of (1.2) but perfectly works for the Schauder theory for Equation (1.1) based on C δp . Let us state thiscondition as below: with some constants λ > and p ≥ it holds that X | α | = | β | = m A αβ ξ α ξ β − λ X | α | = m | ξ α | ≥ p + ( − m ( p − ∞ X k =1 (cid:12)(cid:12)(cid:12)(cid:12) X | α | = m B kα ξ α (cid:12)(cid:12)(cid:12)(cid:12) (1.3) = ∞ X k =1 (cid:12)(cid:12)(cid:12)(cid:12) X | α | = m B kα ξ α (cid:12)(cid:12)(cid:12)(cid:12) when m is odd , ( p − ∞ X k =1 (cid:12)(cid:12)(cid:12)(cid:12) X | α | = m B kα ξ α (cid:12)(cid:12)(cid:12)(cid:12) when m is even . igher-order SPDEs 3 Obviously, this condition is really p -dependent only when m is even, and for odd m it turns to be the same with(1.2). Though it might look strange at first glance, the following example demonstrates its sharpness to some extent. Example 1.1
Given µ ∈ R , we consider the following equation on the torus T := R / (2 π Z ) :d u = ( − m +1 D m u d t + µD m u d w t (1.4)with the initial condition u ( x,
0) = X n ∈ Z e − n m · e √− nx , x ∈ T . If µ < , from Theorem 3.2.1 in [26] this equation admits a unique solution u in L ( Ω ; C ([0 , T ]; H l ( T ))) forany integer l . However, we have the following lemma. Lemma 1.2
Let m be even and µ < . If p > /µ , then E k u ( · , t ) k pL ( T ) = + ∞ for any t > /ε with ε = ( p − µ − . Consequently, sup x ∈ T E | u ( x, t ) | p = + ∞ for any t > /ε . The proof of Lemma 1.2 is presented in Section 6. This result indicates that the coefficient p − in the evencase of the condition (1.3) couldn’t get any smaller if one wants to always ensure the finiteness of sup x E | u ( x, t ) | p ,and this, of course, is a basic requirement in our theory.Although our main result, Theorem 2.3 below, is stated (and also proved) only for linear equations of form(1.1), we point out that it is not difficult to extend it to the semilinear case where f and g depend on the unknown u and are Lipschitz continuous with respect to all D α u with | α | < m and to all D β u with | β | < m , respectively.Besides, it is also interesting to ask if the coercivity condition (1.3) is sufficient or not to construct an L p -theoryfor Equation (1.1).Our approach to Schauder estimates, following the strategy used in [11,12], combines a perturbation schemeof Wang [37] with some integral-type estimates that were also used in [34]. The effect of the p -dependent condition(1.3) can be seen in the proof of the mixed norm estimates (Lemma 3.1); the latter leads to a local boundedness es-timate that plays a key role in proving the fundamental interior Schauder estimate for the model equation (see (3.1)below).This paper is organised as follows. In the next section we state our main theorem after introducing some nota-tion and assumptions. Sections 3 and 4 are both devoted to the estimates for the model equation whose coefficientsdepend on t and ω but not on x ; we prove some auxiliary estimates in Section 3, and establish the interior Hölderestimate in Section 4. The proof of the main theorem is completed in Section 5. In the final section we proveLemma 1.2. Before stating the main results, we introduce some notation and the working spaces. For a function f of x =( x , . . . , x n ) ∈ R n and a multi-index β = ( β , . . . , β n ) ∈ N n , we define D β f = ∂ β · · · ∂ β n ∂x β · · · ∂x β n n f, | β | = β + · · · + β n . For k ∈ N = { , , , . . . } , D k f is regarded as the set of all k -order derivatives of f and k D k f k E = P | α | = k k D α f k E where k · k E is the norm of a normed space E . All the derivatives of E -valued functions are defined with respectto the spatial variables in the strong sense as in [31].A Banach space-valued Hölder continuous function is a natural extension of the classical Hölder continuousfunction. Let E be a Banach space, O be a domain in R n , I ⊂ R be an interval, and Q := O × I . For a function h : O → E , we define | h | Ek ; O := max | β |≤ k sup x ∈O (cid:13)(cid:13)(cid:13) D β h ( x ) (cid:13)(cid:13)(cid:13) E , [ h ] Ek + δ ; O := max | β | = k sup x,y ∈O ,x = y (cid:13)(cid:13) D β h ( x ) − D β h ( y ) (cid:13)(cid:13) E | x − y | δ , | h | Ek + δ ; O := | h | Ek ; O + [ h ] Ek + δ ; O . Y. Wang, K. Du with k ∈ N and α ∈ (0 , . For a function u : Q → E , we define [ u ] Ek + δ ; Q := sup t ∈ I [ u ( · , t )] Ek + δ ; O , | u | Ek + δ ; Q := sup t ∈ I | u ( · , t ) | Ek + δ ; O . Moreover, we define the parabolic modulus | X | p = | ( x, t ) | p := | x | + | t | m and [ u ] E ( k + δ,δ/ m ); Q := max | β | = k sup X,Y ∈ Q,X = Y (cid:13)(cid:13) D β u ( X ) − D β u ( Y ) (cid:13)(cid:13) E | X − Y | δ p , | u | E ( k + δ,δ/ m ); Q := | u | Ek ; Q + [ u ] E ( k + δ,δ/ m ); Q . In this paper, E is either i) R , ii) l or iii) L pω := L p ( Ω ) . We omit the superscript in cases i) and ii), and in case iii)we denote · ... := |· | L pω ... , J · K ... := [ · ] L pω ... for simplicity. Definition 2.1
The Hölder-type spaces C k + δx ( Q ; L pω ) and C k + δ,δ/ mx,t ( Q ; L pω ) are defined as all predictable ran-dom fields u defined on Q × Ω and taking values in an Euclidean space or l such that u ( · , t ) is an L pω -valuedstrongly continuous function for each t , and u k + δ ; Q and u ( k + δ,δ/ m ); Q are finite respectively.Obviously, a function u in C k + δx ( Q ; L pω ) means that itself and its spatial derivatives up to order k lie in thespace C δp defined in the previous section.In this paper we adopt a concept of quasi-classical solutions introduced in [11]. Definition 2.2
A predictable random field u is called a quasi-classical solution of (1.1) if(i) for each t ∈ (0 , ∞ ) , u ( · , t ) is an m times strongly differentiable function from R n to L pω for some p ≥ ;and (ii) for each x ∈ R n , the process u ( x, · ) is stochastically continuous and satisfies the integral equation u ( x, T ) − u ( x, T ) = ˆ T T (cid:20) − ( − m X | α | , | β |≤ m A αβ D α + β u ( x, t ) + f ( x, t ) (cid:21) d t (2.1) + ˆ T T ∞ X k =1 (cid:20) X | α |≤ m B kα D α u ( x, t ) + g k ( x, t ) (cid:21) d w kt almost surely (a.s.) for all ≤ T < T < ∞ .In particular, if u ( · , t, ω ) ∈ C m ( R n ) for each ( t, ω ) ∈ [0 , ∞ ) × Ω , then u is a classical solution of (1.1).Next we will introduce some notations for the domains: B r ( x ) := { y ∈ R n : | y − x | < r } , Q r ( x, t ) := B r ( x ) × ( t − r m , t ] and simply write B r := B r (0) , Q r := Q r (0 , . Also we denote Q r,T ( x ) := B r ( x ) × (0 , T ] , Q r,T := Q r,T (0) , Q T := R n × (0 , T ] . Assumption 1
The following conditions hold throughout the paper unless otherwise stated:1) The coercivity condition (1.3) is satisfied with some λ > and p ≥ .2) The random fields A αβ and f are real-valued, and B α and g are l -valued; all of them are predictable. Theclassical C δx -norms of A αβ ( · , t, ω ) and C m + δx -norms of B α ( · , t, ω ) are all dominated by a constant K > uniformly in ( t, ω ) .3) The free terms f ∈ C δx ( Q T ; L pω ) and g ∈ C m + δx ( Q T ; L pω ) .Now we are ready to state the main result in this paper which consists of the global Hölde estimate and thesolvability. Theorem 2.3
Under Assumptions 1, there exists a unique quasi-classical solution u ∈ C m + δ,δ/ mx,t ( Q T ; L pω ) toEquation (1.1) with the initial condition u ( · ,
0) = 0 . Moreover, there is a constant
C > depending only on n , m, λ , p, δ and K such that u (2 m + δ,δ/ m ); Q T ≤ Ce CT ( f δ ; Q T + g m + δ ; Q T ) . (2.2) igher-order SPDEs 5 In the proof of Theorem 2.3 the global Hölde estimate (2.2) is derived first, and then the existence and unique-ness of solutions of Equation (1.1) is obtained by the standard method of continuity.We remark that the Cauchy problem with nonzero initial condition can be reduced into the case of zero initialcondition by some simple calculation. Also, if p is large enough one can obtain a modification of the solution that isHölder continuous jointly in space and time by means of the Kolmogorov continuity theorem (see [8] for example). In Sections 3 and 4 we always assume that the coefficients A αβ and B α with | α | = | β | = m are all bounded pre-dictable processes (dominated by the constant K ), independent of the spatial variable x , and satisfy the coercivitycondition (1.3). Consider the following model equation d u ( x, t ) = (cid:20) − ( − m X | α | = | β | = m A αβ ( t ) D α + β u ( x, t ) + f ( x, t ) (cid:21) d t + ∞ X k =1 (cid:20) X | α | = m B kα ( t ) D α u ( x, t ) + g k ( x, t ) (cid:21) d w kt (3.1)with ( x, t ) ∈ R n × [ − , + ∞ ) . Let
O ⊂ R n and H k ( O ) = W k, ( O ) be the usual Sobolev spaces. Let I ⊂ R and Q = O × I . For p, q ∈ [1 , ∞ ] , define L pω L qt H mx ( Q ) := L p ( Ω ; L q ( I ; H m ( O ))) , and the domain Q in the notation will be often omitted if there is no confusion. Lemma 3.1
Let Q T = R n × [0 , T ] , p ≥ and the integer l ≥ m . Suppose f ∈ L pω L t H l − mx ( Q T ) and g ∈ L pω L t H lx ( Q T ) . Then Equation (3.1) with zero initial value admits a unique weak solution u ∈ L pω L ∞ t H lx ( Q T ) ∩ L pω L t H l + mx ( Q T ) . Moreover, for any multi-index β such that | β | ≤ l , (cid:13)(cid:13)(cid:13) D β u (cid:13)(cid:13)(cid:13) L pω L ∞ t L x + (cid:13)(cid:13)(cid:13) D β D m u (cid:13)(cid:13)(cid:13) L pω L t L x ≤ C ( (cid:13)(cid:13)(cid:13) D β f (cid:13)(cid:13)(cid:13) L pω L t H − mx + (cid:13)(cid:13)(cid:13) D β g (cid:13)(cid:13)(cid:13) L pω L t L x ) (3.2) where the constant C depends only on n , p , m , T , λ , and K .Proof If p = 2 , the existence and uniqueness of the weak solution has already been obtained in [26] and [33]. So itremains to prove estimate (3.2) for general p ≥ . Since we can differentiate (3.1) with order β , it suffices to provethe estimate in the case | β | = 0 .By an Itô formula from [26, Theorem 1.3.1], one can derive d k u ( · , t ) k L x (3.3) = ˆ R n [ − X | α | = | β | = m A αβ D α uD β u + 2 uf + ∞ X k =1 | X | α | = m B kα D α u + g k | ]d x d t + ∞ X k =1 ˆ R n u ( X | α | = m B kα D α u + g k )d x d w kt . Note that in the last term one has ˆ R n u (cid:16) X | α | = m B kα D α u (cid:17) d x = 0 when m is odd, (3.4)but it is not true for even m .Take a stopping time τ such that E "(cid:18) sup t ∈ [0 ,τ ] k u ( t ) k L x + ˆ τ k D m u ( t ) k L x d t (cid:19) p < + ∞ . Let us consider two cases:
Y. Wang, K. Du
Case 1. m is odd . Using the fact (3.4), and by Condition (1.3), the Sobolev–Gagliargo–Nirenberg inequalityand Young’s inequality, we have sup t ∈ [0 ,τ ] k u ( t ) k L x + ˆ τ k D m u ( t ) k L x d t ≤ C ˆ τ (cid:2) k f ( t ) k H − mx + k g ( t ) k L x (cid:3) d t + C (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) sup t ∈ [0 ,τ ] X k ˆ t ˆ R n ug k d x d w ks (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . Then computing E [ · ] p/ on both sides of the above inequality, and using the Burkholder–Davis–Gundy (BDG)inequality and Young’s inequality, one can obtain that E sup t ∈ [0 ,τ ] k u ( t ) k pL x + E (cid:18) ˆ τ k D m u k L x d t (cid:19) p ≤ C E (cid:12)(cid:12)(cid:12)(cid:12) ˆ τ ( k f k H − mx + k g k L x )d t (cid:12)(cid:12)(cid:12)(cid:12) p , (3.5)where the constant C depends only on n , p , m , T , and λ . Case 2. m is even . Applying Itô’s formula to k u ( · , t ) k pL x , one can derive d k u ( · , t ) k pL x (3.6) = p k u k p − L x ˆ R n (cid:20)(cid:16) − X | α | = | β | = m A αβ D α uD β u + + ∞ X k =1 | X | α | = m B kα D α u | (cid:17) +2 uf + 2 X k X | α | = m ( B kα D α u ) g k + | g | (cid:21) d x d t + p k u k p − L x X k ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x d w kt + p ( p − {k u k L x =0 } k u k p − L x X k (cid:20) ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x (cid:21) d t. With the help of Hölder inequality, one can obtain ( p − X k (cid:20) ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x (cid:21) (3.7) ≤ ( p − ǫ ) X k (cid:18) ˆ R n X | α | = m uB kα D α u d x (cid:19) + C ( ǫ, p, m ) X k (cid:18) ˆ R n ug k d x (cid:19) ≤ (1 + ǫ )( p − k u k L x · X k ˆ R n (cid:12)(cid:12)(cid:12)(cid:12) X | α | = m B kα D α u (cid:12)(cid:12)(cid:12)(cid:12) d x + C X k (cid:18) ˆ R n ug k d x (cid:19) We choose ǫ > so small that ( p − Kǫ ≤ λ/ . Then combining with (1.3), (3.6) (3.7) and Sobolev-Gagliargo-Nirenberg inequality, we have d k u ( · , t ) k pL x (3.8) ≤ p k u k p − L x (cid:20) − λ X | η | = m k D η u k L x + k u k H mx k f k H − mx + k g k L x + k g k L x X | η | = m k D η u k L x (cid:21) d t + p k u k p − L x X k ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x d w kt ≤ − ε k u k p − L x k D m u k L x + C k u k p − L x ( k u k L x + k f k H − mx + k g k L x )+ p k u k p − L x X k ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x d w kt . igher-order SPDEs 7 For simplicity, we denote k D m u k L x = P | α | = m k D α u k L x . Integrating with respect to t on interval [0 , s ] for any s ∈ [0 , T ] , we can obtain that k u ( s ) k pL x + ε ˆ s k u k p − L x k D m u k L x d t (3.9) ≤ C ˆ s k u k p − L x ( k u k L x + k f k H − mx + k g k L x )d t + p X k ˆ s k u k p − L x ˆ R n u ( X | α | = m B kα D α u + g k )d x d w kt , a.s. where ε = ε ( m, λ ) > . Choosing the stopping time τ as before and taking expectation on both sides of (3.9) andby Gronwall’s inequality, one can derive sup t ∈ [0 ,T ] E k u ( t ∧ τ ) k pL x + E ˆ τ k u k p − L x k D m u k L x d t (3.10) ≤ C E ˆ τ k u k p − L x ( k f k H − mx + k g k L x )d t Then we can estimate E sup t ∈ [0 ,τ ] k u ( t ) k pL x from (3.9) by the BDG inequality E sup t ∈ [0 ,τ ] k u ( t ) k pL x + E ˆ τ k u k p − L x k D m u k L x d t (3.11) ≤ C E ˆ τ k u k p − L x ( k u k L x + k f k H − mx + k g k L x )d t + C E (cid:26) ˆ τ k u k p − L x X k (cid:20) ˆ R n u ( X | α | = m B kα D α u + g k )d x (cid:21) d t (cid:27) . The last term of the above inequality is dominated by C E (cid:20) ˆ τ k u k p − L x ( k u k L x k D m u k L x + k u k L x k g k L x )d t (cid:21) ≤ C E (cid:26) sup t ∈ [0 ,τ ] k u k p L x (cid:20) ˆ τ ( k u k p − L x k D m u k L x + k u k p − L x k g k L x )d t (cid:21) (cid:27) ≤ E sup t ∈ [0 ,τ ] k u ( t ) k pL x + C E ˆ τ ( k u k p − L x k D m u k L x + k u k p − L x k g k L x )d t, which along with (3.10) and (3.11) yields that E sup t ∈ [0 ,τ ] k u ( t ) k pL x ≤ C E ˆ τ k u k p − L x ( k f k H − mx + k g k L x )d t ≤ E sup t ∈ [0 ,τ ] k u ( t ) k pL x + C E (cid:20) ˆ τ ( k f k H − mx + k g k L x )d t (cid:21) p . Thus we obtain the estimate E sup t ∈ [0 ,τ ] k u ( t ) k pL x ≤ C E (cid:20) ˆ τ ( k f k H − mx + k g k L x )d t (cid:21) p . (3.12)Next we need to estimate E ( ´ τ k D m u k L x d t ) p/ . Back to (3.3) and integrating with respect to time, one can easilyget that k u ( τ ) k L x + λ ˆ τ k D m u k L x d t ≤ ˆ τ ˆ R n (cid:16) uf + X k X | α | = m B kα D α ug k + | g | (cid:17) d x d t + X k ˆ τ ˆ R n u (cid:16) X | α | = m B kα D α u + g k (cid:17) d x d w kt . Y. Wang, K. Du
Computing E [ · ] p/ on both sides of the above inequality and by the Hölder’s inequality and BDG inequality, wederive that E (cid:18) ˆ τ k D m u k L x d t (cid:19) p ≤ ǫ E (cid:18) ˆ τ k u k H mx d t (cid:19) p + C E (cid:20) ˆ τ ( k f k H − mx + k g k L x )d t (cid:21) p + C E (cid:26) X k ˆ τ (cid:20) ˆ R n u ( X | α | = m B kα D α u + g k )d x (cid:21) d t (cid:27) p ≤ ǫ E (cid:18) ˆ τ k u k H mx d t (cid:19) p + C E (cid:20) ˆ τ ( k f k H − mx + k g k L x )d t (cid:21) p + C E (cid:20) ˆ τ k u k L x ( k D m u k L x + k g k L x )d t (cid:21) p ≤ ǫ E (cid:18) ˆ τ k D m u k L x d t (cid:19) p + C E (cid:20) ˆ τ ( k f k H − mx + k g k L x )d t (cid:21) p + C E sup t ∈ [0 ,τ ] k u ( t ) k pL x which along with (3.12) implies the estimate (3.5) in this case. Here the constant C further depends on K .Finally, we replace τ in (3.5) by the following sequence of stopping times τ k := inf (cid:26) s ≥ t ∈ [0 ,s ] k u ( s ) k L x + ˆ s k D m u ( s ) k L x d s > k (cid:27) ∧ T, and send k to infinity. Then (3.5) yields the desired estimate for l = 0 and the lemma is proved. ⊓⊔ Proposition 3.2
Let l be a positive integer, l ≥ m , p ≥ , r ∈ (0 , and θ ∈ (0 , . Let u ∈ L pω L t H l + mx ( Q r ) solve (3.1) in Q r with free terms f ∈ L pω L t H l − mx ( Q r ) and g ∈ L pω L t H lx ( Q r ) . Then there exists a constant C depending only on n , p , l , m , λ , K , and θ such that k D l u k L pω L ∞ t L x ( Q θr ) + k D l + m u k L pω L t L x ( Q θr ) (3.13) ≤ C m − X k =0 r − m − l + k k D k u k L pω L t L x ( Q r ) + C l − m X k =0 r m + k − l k D k f k L pω L t L x ( Q r ) + C l X k =0 r k − l k D k g k L pω L t L x ( Q r ) . Consequently, for l − | β | ) > n , r n + | β | k sup Q θr | D β u |k L pω ≤ C m − X k =0 r − m + k k D k u k L pω L t L x ( Q r ) (3.14) + C l − m X k =0 r m + k k D k f k L pω L t L x ( Q r ) + C l X k =0 r k k D k g k L pω L t L x ( Q r ) where the constant C further depends on | β | .Proof By Sobolev’s embedding theorem, (3.14) can be derived directly from (3.13). Also we can reduce the prob-lem for general r > to the case r = 1 by rescaling. Indeed, for general r > , we can apply the obtainedestimates for r = 1 to the rescaled function v ( x, t ) := u ( rx, r m t ) , ∀ ( x, t ) ∈ R n × [ − , + ∞ ) igher-order SPDEs 9 which solves the equation d v ( x, t ) = h − ( − m X | α | = | β | = m A αβ ( r m t ) D α + β v ( x, t ) + F ( x, t ) i d t + ∞ X k =1 h X | α | = m B kα ( r m t ) D α v ( x, t ) + G k ( x, t ) i d W kt with free terms F ( x, t ) = r m f ( rx, r m t ) , G ( x, t ) = r m g ( rx, r m t ) , W kt = r − m w kr m t and obviously, W k are mutually independent Wiener processes. So it suffices to prove (3.13) for r = 1 . Byinduction, we shall only consider the case l = m .For any θ ∈ (0 , , choose m + 1 cut-off functions ξ i ∈ C ∞ ( R n +1 ) with i = 1 , , · · · , m + 1 , satisfying i) ≤ ξ i ≤ , ii) ξ i = 1 in Q θ i and ξ i = 0 outside Q θ i +1 , where θ i = i √ θ . Let v i = ξ i u which satisfy d v i ( x, t ) = h − ( − m X | α | = | β | = m A αβ D α + β v i + f i ( x, t ) i d t (3.15) + ∞ X k =1 (cid:16) X | η | = m B kη D η v i + g ki (cid:17) d w kt , i = 1 , , . . . , m + 1 where f i = ξ i f + u∂ t ξ i − X | α | = | β | = m X γ + η = α + β, | η | > C γη A αβ D γ uD η ξ i g ki = ξ i g k − X | α | = m X γ + η = α, | η | > C γη B kα D γ uD η ξ i where C γη are the constants that can be derived from the Leibniz formula, depending only on γ , η , and m .Applying Lemma 3.1 to (3.15) for | β | = m + 1 − i with i = 1 , , · · · , m + 1 , we have k D m +1 − i u k L pω L ∞ t L x ( Q θi ) + k D m +1 − i u k L pω L t L x ( Q θi ) ≤ C (cid:20) m +1 − i X k =0 k D k f k L pω L t H − mx ( Q θi +1 ) + m +1 − i X k =0 k D k g k L pω L t L x ( Q θi +1 ) + k u k L pω L t H m − ix ( Q θi +1 ) (cid:21) ≤ C (cid:20) k f k L pω L t L x ( Q θi +1 ) + m +1 − i X k =0 k D k g k L pω L t L x ( Q θi +1 ) + k u k L pω L t H m − ix ( Q θi +1 ) (cid:21) . From the above inequalities, one can prove (3.13) for l = m . Higher-order estimates follow from induction. Theproof is complete. ⊓⊔ Next we shall give an estimate for equation (3.1) with the Dirichlet boundary conditions ( u ( x,
0) = 0 , ∀ x ∈ R n D α u ( x, t ) | ∂B r = 0 , | α | ≤ m − . (3.16) Proposition 3.3
Let f and g be in L pω L t H lx ( B r × (0 , r m )) for all l ∈ N . Then the Dirichlet problem (3.1)with (3.16) admits a unique weak solution u ∈ L ω L t H mx ( B r × (0 , r m )) . For each t ∈ (0 , r m ) , u ( · , t ) ∈ L p ( Ω ; C l ( B ε )) for all l ≥ and ε ∈ (0 , r ) . Moreover, there is a constant C = C ( n, p, λ, m, K ) such that m − X i =0 r i k D i u k L pω L t L x ( B r × (0 ,r m )) (3.17) ≤ C " r m k f k L pω L t L x ( B r × (0 ,r m )) + m − X i =0 r m + i k D i g k L pω L t L x ( B r × (0 ,r m )) Proof
The existence and uniqueness of the weak solution of the Dirichlet problem (3.1) and (3.16) follow from[26, Section 3.2]. Then we choose a cut-off function ϕ ∈ C ∞ ( R n ) such that ϕ ( x ) = 1 if x ∈ B ε and ϕ ( x ) = 0 if | x | > ( r + ε ) / where ε ∈ (0 , r ) . Applying lemma 3.1 to v := ϕD α u with Sobolev’s embedding theorem, theinterior regularity can be obtained. We omit the proof of the estimate (3.17) because it’s analogous to the proof of(3.2) with the help of rescaling and Sobolev-Gagliargo-Nirenberg inequality. ⊓⊔ In this section we assume that f ∈ C x ( R n × [ − , ∞ ); L pω ) and g ∈ C mx ( R n × [ − , ∞ ); L pω ) , and f ( x, t ) and D m g ( x, t ) are Dini continuous with respect to x uniformly in t , namely, the modulus of continuity defined by ̟ ( r ) = sup t ≥− , | x − y |≤ r ( k f ( x, t ) − f ( y, t ) k L pω + k D m g ( x, t ) − D m g ( y, t ) k L pω ) satisfies that ˆ ̟ ( r ) r d r < ∞ Theorem 4.1
Let u be a quasi-classical solution to (3.1) in Q . Under the above settings, there is a constant Cdepending only on n, λ , p, m and K, such that for any X, Y ∈ Q / , k D m u ( X ) − D m u ( Y ) k L pω ≤ C (cid:20) ̺M + ˆ ̺ ̟ ( r ) r d r + ̺ ˆ ̺ ̟ ( r ) r d r (cid:21) (4.1) where ̺ = | X − Y | p and M = u m − Q + f Q + g m ; Q .Proof Firstly we mollify the functions u , f and g in the spatial variables. We choose a nonnegative and symmetricmollifier ϕ : R n → R and define ϕ ε ( x ) = ε n ϕ ( x/ε ) , u ε = u ∗ ϕ ε , f ε = f ∗ ϕ ε and g ε = g ∗ ϕ ε . It is easyto check that f ε and D m g ε are Dini continuous and have the same modulus of continuity ̟ with f and D m g andsatisfy f ε − f R n + g ε − g m ; R n → (cid:13)(cid:13)(cid:13) D m u ε ( X ) − D m u ( X ) (cid:13)(cid:13)(cid:13) L pω → ∀ X ∈ R n × R , as ε → . On the other hand, from Fubini’s theorem one can check that u ε satisfies the model equation (3.1) in theclassical sense with free terms f ε and g ε . Therefore it suffices to prove the theorem for the mollified functions, andthe general case is straightforward by passing the limits. The readers are referred to the appendix of [11] for moredetails. Then based on the smoothness of mollified functions, we can assume that f and g satisfy the followingadditional condition: ( A ) f, g ∈ L pω L t H kx ( Q R ) ∩ C kx ( Q R ; L pω ) for all k ∈ N and R > . From the definition of ̟ , one can see that for any x, y ∈ R n and t ∈ R , k f ( x, t ) − f ( y, t ) k L pω + k D m g ( x, t ) − D m g ( y, t ) k L pω ≤ ̟ ( | x − y | ) (cid:13)(cid:13)(cid:13)(cid:13) D β g ( y, t ) − X | α |≤ m −| β | D α + β g ( x, t ) α ! ( y − x ) α (cid:13)(cid:13)(cid:13)(cid:13) L pω ≤ C ( m ) | x − y | m −| β | ̟ ( | x − y | ) , | β | ≤ m (4.2)By translation we may suppose that X = (0 , and prove the theorem for any Y ∈ Q / . Given Y = ( y, s ) ∈ Q / , and ˜ κ ∈ N such that ̺ := | Y | p ∈ [ ρ ˜ κ +2 , ρ ˜ κ +1 ) . With ρ = 1 / , we denote Q κ = Q ρ κ (0 , , κ = 0 , , , · · · . Let us introduce the following Dirichlet problems: d u κ = h − ( − m X | γ | =2 m A γ D γ u κ + f (0 , t ) i d t + ∞ X k =1 (cid:20) X | η | = m B kη D η u κ + X | α |≤ m D α g k (0 , t ) α ! x α (cid:21) d w kt in Q κ D α u κ = D α u on ∂ p Q κ , | α | ≤ m − where ∂ p Q κ denotes the parabolic boundary of the cylinder Q κ for κ = 0 , , , . . . . Then the solvability andinterior regularity of each u κ can be obtained by applying Proposition 3.3 to u κ − u . igher-order SPDEs 11 We have the following decomposition (cid:13)(cid:13)(cid:13) D m u ( Y ) − D m u (0) (cid:13)(cid:13)(cid:13) L pω ≤ (cid:13)(cid:13)(cid:13) D m u ˜ κ (0) − D m u (0) (cid:13)(cid:13)(cid:13) L pω + (cid:13)(cid:13)(cid:13) D m u ˜ κ ( Y ) − D m u ˜ κ (0) (cid:13)(cid:13)(cid:13) L pω + (cid:13)(cid:13)(cid:13) D m u ˜ κ ( Y ) − D m u ( Y ) (cid:13)(cid:13)(cid:13) L pω =: K + K + K . (4.3)The next step is to estimate the three terms respectively. We split it into three lemmas. Lemma 4.2 K ≤ C ˆ ρ ˜ κ − ̟ ( r ) r d r. Proof
Apply (3.14) to u κ − u κ +1 with | β | = l, r = ρ κ +1 , θ = to get I κ,l := D l ( u κ − u κ +1 ) Q κ +2 ≤ C m − X i =0 ρ ( i − l )( κ +1) (cid:13)(cid:13)(cid:13)(cid:13) Q κ +1 | D i ( u κ − u κ +1 ) | d X (cid:13)(cid:13)(cid:13)(cid:13) / L p/ ω . In what follows, we define ffl Q = | Q | ´ Q , where | Q | is the Lebesgue measure of the set Q ⊂ R n +1 .On the other hand, from (3.17) one can obtain J κ := m − X i =0 ρ iκ (cid:13)(cid:13)(cid:13)(cid:13) Q κ | D i ( u κ − u ) | d X (cid:13)(cid:13)(cid:13)(cid:13) / L p/ ω ≤ Cρ mκ ̟ ( ρ κ ) . Combining the above we derive I κ,l ≤ Cρ − l ( κ +1) ( J κ + J κ +1 ) ≤ Cρ (2 m − l ) κ − l ̟ ( ρ κ ) (4.4)where C is independent of κ . Choose l = 2 m , then X κ ≥ D m ( u κ − u κ +1 ) Q κ +2 ≤ Cρ − m X κ ≥ ̟ ( ρ κ ) ≤ C ˆ ̟ ( r ) r d r < ∞ , which implies that D m u κ (0) converges in L pω as κ → ∞ . Here is the zero vector in R n +1 . Next we shall provethat the limit is D m u (0) . It suffices to prove lim κ →∞ (cid:13)(cid:13)(cid:13) D m u κ (0) − D m u (0) (cid:13)(cid:13)(cid:13) L ω = 0 (4.5)as p ≥ . Applying (3.14) to u κ − u with | β | = 2 m , l = n + 2 m , r = ρ κ , θ = 1 / and p = 2 , we have sup Q κ +1 (cid:13)(cid:13)(cid:13) D m ( u κ − u ) (cid:13)(cid:13)(cid:13) L ω ≤ C m − X i =0 ρ − mκ +2 iκ E Q κ | D i ( u κ − u ) | d X + C E Q κ | f ( x, t ) − f (0 , t ) | d X + C X i ≤ m ρ (2 i − m ) κ E Q κ (cid:13)(cid:13)(cid:13)(cid:13) D i g ( x, t ) − X | α |≤ m − i D α D i g (0 , t ) α ! x α (cid:13)(cid:13)(cid:13)(cid:13) d X + C n + m X i =1 ρ iκ E Q κ (cid:18)(cid:12)(cid:12)(cid:12) D i f (cid:12)(cid:12)(cid:12) + (cid:13)(cid:13)(cid:13) D i + m g (cid:13)(cid:13)(cid:13) (cid:19) d X ≤ C m − X i =0 ρ − mκ +2 iκ E Q κ | D i ( u κ − u ) | d X + C̟ ( ρ κ ) + C n + m X i =1 ρ iκ ( J f K i ; Q κ + J g K i + m ; Q κ ) where the last two terms tend to 0 as κ → ∞ . From (3.17) and (4.2) we have m − X i =0 ρ − mκ +2 iκ E Q κ | D i ( u κ − u ) | d X ≤ C E Q κ (cid:18) | f ( x, t ) − f (0 , t ) | + m − X i =0 ρ (2 i − m ) κ (cid:13)(cid:13)(cid:13)(cid:13) D i g ( x, t ) − X | α |≤ m − i D α D i g (0 , t ) α ! x α (cid:13)(cid:13)(cid:13)(cid:13) (cid:19) d X ≤ C̟ ( ρ κ ) → , as κ → ∞ . Therefore D m u κ (0) converges strongly to D m u (0) in L pω . Moreover, we have K = (cid:13)(cid:13)(cid:13) D m u ˜ κ (0) − D m u (0) (cid:13)(cid:13)(cid:13) L pω ≤ X j ≥ ˜ κ D m ( u j − u j +1 ) Q κ +2 (4.6) ≤ C ˆ ρ ˜ κ − ̟ ( r ) r d r where C = C ( n, m, λ, p, K ) . ⊓⊔ Lemma 4.3 K ≤ C̺M + C̺ ˆ ̺ ̟ ( r ) r d r. Proof
Define h ι := u ι − u ι − , for ι = 1 , , . . . , ˜ κ. Then we decompose K by K = (cid:13)(cid:13)(cid:13) D m u ˜ κ ( Y ) − D m u ˜ κ (0) (cid:13)(cid:13)(cid:13) L pω ≤ (cid:13)(cid:13)(cid:13) D m u ( Y ) − D m u (0) (cid:13)(cid:13)(cid:13) L pω + ˜ κ X ι =1 (cid:13)(cid:13)(cid:13) D m h ι ( Y ) − D m h ι (0) (cid:13)(cid:13)(cid:13) L pω . As D m +1 u satisfies the following homogeneous equation: d( D m +1 u ) = − ( − m X | γ | =2 m A γ D γ ( D m +1 u ) d t + ∞ X k =1 X | η | = m B kη D η ( D m +1 u ) d w kt (4.7)in Q / . Using (3.14) to D m +1 u , one has D m u Q / + D m u Q / ≤ C m − X i =0 − i (cid:13)(cid:13)(cid:13) D i + m +1 u (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) ≤ C m − X i =0 − i (cid:20)(cid:13)(cid:13)(cid:13) D i + m +1 ( u − u ) (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) + (cid:13)(cid:13)(cid:13) D i + m +1 u (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) (cid:21) Applying (3.13) to u , one can get m − X i =0 (cid:13)(cid:13)(cid:13) D i + m +1 u (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) ≤ C m − X i =0 (cid:13)(cid:13)(cid:13) D i u (cid:13)(cid:13)(cid:13) L pω L t L x ( Q ) + k f k L pω L t L x ( Q ) + m X i =0 (cid:13)(cid:13)(cid:13) D i g (cid:13)(cid:13)(cid:13) L pω L t L x ( Q ) ! . igher-order SPDEs 13 Applying (3.13) and (3.17) to u − u one can obtain m − X i =0 (cid:13)(cid:13)(cid:13) D i + m +1 ( u − u ) (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) ≤ C m − X i =0 (cid:13)(cid:13)(cid:13) D i (cid:16) u − u (cid:17)(cid:13)(cid:13)(cid:13) L pω L t L x ( Q ) + C k f ( x, t ) − f (0 , t ) k L pω L t L x ( Q ) + C m X i =0 (cid:13)(cid:13)(cid:13)(cid:13) D i (cid:18) g ( x, t ) − X | α |≤ m − i D α g (0 , t ) α ! x α (cid:19)(cid:13)(cid:13)(cid:13)(cid:13) L pω L t L x ( Q ) ≤ C k f ( x, t ) − f (0 , t ) k L pω L t L x ( Q ) + C m X i =0 (cid:13)(cid:13)(cid:13)(cid:13) D i (cid:18) g ( x, t ) − X | α |≤ m D α g (0 , t ) α ! x α (cid:19)(cid:13)(cid:13)(cid:13)(cid:13) L pω L t L x ( Q ) . Therefore, D m u Q / + D m u Q / ≤ CM . Hence, for − − m < s ≤ t ≤ and x ∈ B / , (cid:13)(cid:13)(cid:13) D m u ( x, t ) − D m u ( x, s ) (cid:13)(cid:13)(cid:13) L pω = (cid:13)(cid:13)(cid:13)(cid:13) ˆ ts − ( − m X | γ | =2 m A γ D γ ( D m u )d t + ˆ ts X | η | = m B kη D η ( D m u )d w kt (cid:13)(cid:13)(cid:13)(cid:13) L pω ≤ C √ t − s (cid:16) D m u Q / + D m u Q / (cid:17) ≤ C √ t − sM ≤ C ( t − s ) m M Analogous to the above steps we can get (cid:13)(cid:13)(cid:13) D m +1 u (cid:13)(cid:13)(cid:13) L pω L t L x ( Q / ) ≤ CM . Thus we get (cid:13)(cid:13)(cid:13) D m u ( X ) − D m u ( Y ) (cid:13)(cid:13)(cid:13) L pω ≤ CM | X − Y | p , ∀ X, Y ∈ Q / . (4.8)Note that h ι satisfies d h ι = − ( − m X | γ | =2 m A γ D γ h ι d t + ∞ X k =1 X | η | = m B kη D η h ι d w kt . (4.9)in Q ι . By (4.4) we have ρ − mι D m h ι Q ι +1 + D m h ι Q ι +1 ≤ Cρ − mι ̟ ( ρ ι − ) , D m +1 h ι Q ι +1 ≤ Cρ − ι ̟ ( ρ ι − ) . Hence for − ρ m (˜ κ +1) ≤ t ≤ and | x | ≤ ρ ˜ κ +1 , (cid:13)(cid:13)(cid:13) D m h ι ( x, t ) − D m h ι ( x, (cid:13)(cid:13)(cid:13) L pω ≤ C (cid:16) ρ m ˜ κ D m h ι Q ι +1 + ρ m ˜ κ D m h ι Q ι +1 (cid:17) ≤ Cρ m (˜ κ − ι ) ̟ ( ρ ι − ) , and (cid:13)(cid:13)(cid:13) D m h ι ( x, − D m h ι (0 , (cid:13)(cid:13)(cid:13) L pω ≤ Cρ ˜ κ − ι ̟ ( ρ ι − ) . Combining the last two estimates and (4.8), we can obtain K = (cid:13)(cid:13)(cid:13) D m u ˜ κ ( Y ) − D m u ˜ κ (0) (cid:13)(cid:13)(cid:13) L pω ≤ (cid:13)(cid:13)(cid:13) D m u ( Y ) − D m u (0) (cid:13)(cid:13)(cid:13) L pω + ˜ κ X ι =1 (cid:13)(cid:13)(cid:13) D m h ι ( Y ) − D m h ι (0) (cid:13)(cid:13)(cid:13) L pω ≤ CM ρ ˜ κ +1 + C ˜ κ X ι =1 ρ ˜ κ − ι ̟ ( ρ ι − ) ≤ C̺M + C̺ ˆ ̺ ̟ ( r ) r d r. The lemma is proved. ⊓⊔ Lemma 4.4 K ≤ C̟ ( ̺ ) + C ˆ ̺ ̟ ( r ) r d r. Proof
We consider the following sequence of equations d u Y,κ = (cid:20) − ( − m X | γ | =2 m A γ D γ u Y,κ + f ( y, t ) (cid:21) d t + ∞ X k =1 (cid:20) X | η | = m B kη D η u Y,κ + X | α |≤ m D α g k ( y, t ) α ! ( x − y ) α (cid:21) d w kt in Q κ ( Y ) D α u Y,κ = D α u on ∂ p Q κ ( Y ) , | α | ≤ m − , with κ = 0 , , . . . , ˜ κ − , ˜ κ + 2 , . . . ; the equations associated with ˜ κ and ˜ κ + 1 are replaced by the following single equation d u Y, ˜ κ = (cid:20) − ( − m X | γ | =2 m A γ D γ u Y, ˜ κ + f ( y, t ) (cid:21) d t + ∞ X k =1 (cid:20) X | η | = m B kη D η u Y, ˜ κ + X | α |≤ m D α g k ( y, t ) α ! ( x − y ) α (cid:21) d w kt in Q κ D α u Y, ˜ κ = D α u on ∂ p Q κ (0) , | α | ≤ m − . As | Y | p ∈ [ ρ ˜ κ +2 , ρ ˜ κ +1 ) , it is easily seen that Q ˜ κ +2 ( Y ) ⊂ Q ˜ κ (0) . So analogous to the proof of (4.6) we have (cid:13)(cid:13)(cid:13) D m u Y, ˜ κ ( Y ) − D m u ( Y ) (cid:13)(cid:13)(cid:13) L pω ≤ C ˆ ρ ˜ κ − ̟ ( r ) r d r, where C = C ( n, m, λ, K, p ) . On the other hand, combining (3.14), (3.17) and (4.2), one can derive (cid:13)(cid:13)(cid:13) D m u Y, ˜ κ ( Y ) − D m u ˜ κ ( Y ) (cid:13)(cid:13)(cid:13) L pω ≤ C̟ ( ̺ ) . Then we have (cid:13)(cid:13)(cid:13) D m u ˜ κ ( Y ) − D m u ( Y ) (cid:13)(cid:13)(cid:13) L pω ≤ C̟ ( ̺ ) + C ˆ ̺ ̟ ( r ) r d r. The lemma is proved. ⊓⊔ Now recalling (4.3) and combining Lemmas 4.2, 4.3 and 4.4, one has that k D m u ( Y ) − D m u (0) k L pω ≤ C (cid:20) ̺M + ˆ ̺ ̟ ( r ) r d r + ̺ ˆ ̺ ̟ ( r ) r d r (cid:21) . The proof of Theorem 4.1 is complete. ⊓⊔ igher-order SPDEs 15 From the above theorem, one can easily derive the following interior Hölder estimate for (3.1), where we denote Q r,T = B r × [0 , T ] for r , T > . Corollary 4.5
If u is a quasi-classical solution of (3.1) in R n × [0 , ∞ ) with zero initial condition and δ ∈ (0 , .Then there is a positive constant C depending only on n, m, p, K, λ and δ , such that r D m u z ( δ,δ/ m ); Q / ,T ≤ C (cid:20) u m − Q ,T + f δ ; Q ,T + g m + δ ; Q ,T δ (1 − δ ) (cid:21) (4.10) for any T > , provided the right-hand side is finite.Proof Because of the zero initial condition, define ˜ u ( x, t ) , ˜ f ( x, t ) and ˜ g ( x, t ) to be zero whenever t ∈ [ − , ,and be equal to u ( x, t ) , f ( x, t ) and g ( x, t ) , respectively, whenever t ≥ . Obviously, ˜ u is a quasi-classical solutionto (3.1) in R n × [ − , ∞ ) . From (4.1) we have r D m ˜ u z ( δ,δ/ m ); Q / ( X ) ≤ C " ˜ u m − Q ( X ) + ˜ f δ ; Q ( X ) + ˜ g m + δ ; Q ( X ) δ (1 − δ ) for any X = ( x, t ) ∈ R n × [0 , ∞ ) . Using the localization property of Hölder norms (see Lemma 4.1.1 in [19]),we obtain r D m u z ( δ,δ/ m ); Q / ,T ≤ r D m ˜ u z ( δ,δ/ m ); Q / ,T ≤ C sup t ∈ [0 ,T ] (cid:18) r D m ˜ u z ( δ,δ/ m ); Q / (0 ,t ) + ˜ u Q / (0 ,t ) (cid:19) ≤ C " u m − Q ,T + ˜ f δ ; Q ( X ) + ˜ g m + δ ; Q ( X ) δ (1 − δ ) . The proof is complete. ⊓⊔ This section is devoted to the proof of Theorem 2.3. We need two technical lemmas; readers are referred to [11]for their proofs.
Lemma 5.1
Let ϕ be a bounded nonnegative function from [0 , T ] to [0 , ∞ ) satisfying ϕ ( t ) ≤ θϕ ( s ) + k X i =1 C i ( s − t ) − θ i , ∀ ≤ t < s ≤ T, for some nonnegative constants θ, θ i and C i (i=1, . . . ,k), where θ < . Then ϕ (0) ≤ C k X i =1 C i T − θ i , where C depends only on θ , . . . , θ k and θ . Lemma 5.2
Let B R = { x ∈ R n : | x | < R } with R > , p ≥ , and ≤ s < r . There exists a positive constant C , depending only on n and p , such that J u K s ; B R ≤ Cε r − s J u K r ; B R + Cε − s − n/p k u k L p ( B R ; L pω ) for any u ∈ C r ( B R ; L pω ) and ε ∈ (0 , R ) . Now we are in a position to complete the proof of Theorem 2.3.
Proof (Proof of Theorem 2.3)
The proof is divided into two steps.
Step 1 . Global Hölder estimate (2.2).Suppose u is the quasi-classical solution to (1.1) with zero initial condition. Let ρ/ ≤ r < R ≤ ρ with ρ ∈ (0 , / to be determined. Choose a nonnegative function ζ ∈ C ∞ ( R n ) such that ζ ( x ) = 1 on B r , ζ ( x ) = 0 outside B R , and for δ > , [ ζ ] δ ; R n ≤ C ( R − r ) − δ . Set v = ζu , and A αβ, ( t ) = A αβ (0 , t ) , B kα, ( t ) = B kα (0 , t ) . Then v satisfies d v = (cid:18) − ( − m X | α | = | β | = m A αβ, D α + β v + ˜ f (cid:19) d t + ∞ X k =1 (cid:18) X | α | = m B kα, D α u + ˜ g k (cid:19) d w kt where ˜ f = ( − m X | α | = | β | = m A αβ, D α + β ( ζu ) − ( − m X | α | , | β |≤ m ζA αβ D α + β u + ζf, ˜ g k = − X | α | = m B kα, D α ( ζu ) + X | α |≤ m ζB kα D α u + ζg k . We denote Q R,τ = B R × (0 , τ ) for τ > and define M τx,r ( u ) = sup t ∈ [0 ,τ ] (cid:18) B r ( x ) E | u ( y, t ) | p d y (cid:19) /p , M τr ( u ) = sup x ∈ R n M τx,r ( u ) . Then following from Lemma 5.2, we directly derive ˜ f δ ; Q R,τ ≤ (cid:16) ε + KR δ (cid:17) J u K m + δ ; Q R,τ + C ( R − r ) − m − δ − n/p M τ ,R ( u )+ J f K δ ; Q R,τ + C ( R − r ) − δ f Q R,τ , ˜ g m + δ ; Q R,τ ≤ (cid:16) ε + KR δ (cid:17) J u K m + δ ; Q R,τ + C ( R − r ) − m − δ − n/p M τ ,R ( u )+ J g K m + δ ; Q R,τ + C ( R − r ) − m − δ g Q R,τ . In the above two inequalities, C = C ( n, p, K, ε, ρ ) . Applying Corollary 4.5 and taking positive ρ , ε so smallthat ε + KR δ ≤ δ (1 − δ )4 C where C is the constant in the corollary, then we get that J u K (2 m + δ,δ/ m ); Q r,τ ≤ J u K m + δ ; Q R,τ + C ( R − r ) − m − δ − n/p M τ ,R ( u )+ C ( R − r ) − δ f δ ; Q R,τ + C ( R − r ) − m − δ g m + δ ; Q R,τ . Then by Lemma 5.1, we obtain J u K (2 m + δ,δ/ m ); Q ρ/ ,τ ≤ C (cid:0) M τ ,ρ ( u ) + f δ ; Q ρ,τ + g m + δ ; Q ρ,τ (cid:1) . Note that the above inequality is true for any point x ∈ R n instead of . Therefore, applying Lemma 5.2, we have sup x ∈ R n u (2 m + δ,δ/ m ); Q ρ/ ,τ ( x ) ≤ C (cid:0) M τρ ( u ) + f δ ; Q ρ,τ + g m + δ ; Q ρ,τ (cid:1) ≤ C (cid:0) M τρ/ ( u ) + f δ ; Q τ + g m + δ ; Q τ (cid:1) . The next step is to estimate M τρ/ ( u ) . Applying Itô’s formula to | u | p and integrating in Q ρ/ ,τ ( x ) × Ω with theuse of Sobolev-Gagliargo-Nirenberg inequality, we get M τρ/ ( u ) ≤ C τ (cid:18) sup x ∈ R n u m ; Q ρ/ ,τ ( x ) + f Q τ + g Q τ (cid:19) . igher-order SPDEs 17 Taking τ = 2 ( CC ) − , the above two inequalities yield sup x ∈ R n u (2 m + δ,δ/ m ); Q ρ/ ,τ ( x ) ≤ C ( f δ ; Q τ + g m + δ ; Q τ ) . Following from the localization property of Hölder norms, we get u (2 m + δ,δ/ m ); Q τ ≤ C τ ( f δ ; Q τ + g m + δ ; Q τ ) (5.1)with C τ = C τ ( n, m, δ, λ, K, p ) ≥ . Finally, we conclude the proof by induction. Assume that there is a constant C S ≥ for some S > such that u (2 m + δ,δ/ m ); Q S ≤ C S ( f δ ; Q S + g m + δ ; Q S ) . Then applying (5.1) to v ( x, t ) := u ( x, t + S ) − u ( x, S ) for t ≥ , we can derive that v (2 m + δ,δ/ m ); Q τ ≤ C τ (cid:16) f δ ; Q S + τ + g m + δ ; Q S + τ + ˜ C u (2 m + δ,δ/ m ); Q S (cid:17) ≤ C τ (1 + ˜ CC S ) (cid:0) f δ ; Q S + τ + g m + δ ; Q S + τ (cid:1) where ˜ C = ˜ C ( m, K ) ≥ . Thus we get u (2 m + δ,δ/ m ); Q S + τ ≤ v (2 m + δ,δ/ m ); Q τ + u (2 m + δ,δ/ m ); Q S ≤ CC τ C S (cid:0) f δ ; Q S + τ + g m + δ ; Q S + τ (cid:1) which means C S + τ ≤ CC τ C S . As τ is fixed, by iteration we have C S ≤ C e CS where C = C ( n, m, δ, λ, p, K ) .This completes the proof of (2.2). Step 2 . The solvability.For simplicity, we denote L = − ( − m X | α | , | β |≤ m A αβ D α + β , Λ k = X | α |≤ m B kα D α . Define L s = sL + (1 − s ) ∆ m , Λ ks = sΛ k where s ∈ [0 , and ∆ m := P | γ | =2 m δ γ D γ where δ γ := ( , γ i = 2 m for some ≤ i ≤ n , other . Then consider the equation d u = ( L s u + f )d t + ∞ X k =1 ( Λ ks u + g k )d w kt in Q u ( · ,
0) = 0 in R n (5.2)where Q := R n × [0 , ∞ ) . Evidently, the solutions of the above equations enjoy the estimate (2.2) with the samedominating constant C (independent of s ). So by the standard method of continuity (see [13, Theorem 5.2]), itsuffices to show the solvability of the following equation (the case s = 0 ): d u = ( ∆ m u + f )d t + ∞ X k =1 g k d w kt , u ( · ,
0) = 0 . (5.3)Letting ϕ : R n → R be a nonnegative and symmetric mollifier (see Appendix in [11]) and ϕ ε ( x ) = ε n ϕ ( x/ε ) ,we define f ε = ϕ ε ∗ f and g ε = ϕ ε ∗ g . From the results of Appendix in [11], we obtain that f ε ∈ C δx ( Q ; L pω ) and g ε ∈ C m + δx ( Q ; L pω ) satisfying f ε − f δ/ Q T + g ε − g m + δ/ Q T → as ε → . (5.4) Moreover, f ε ( x, t, ω ) and g ε ( x, t, ω ) are smooth in x for any ( t, ω ) , and f ε , g ε ∈ C k ( Q T ; L pω ) for all k ∈ N .For any k ∈ N and r > n , we have E (cid:12)(cid:12)(cid:12)(cid:12) ˆ Q T (1 + | x | ) − r ( | D k f ε ( x, t ) | + | D k g ε ( x, t ) | )d x d t (cid:12)(cid:12)(cid:12)(cid:12) p ≤ C (cid:12)(cid:12)(cid:12)(cid:12) ˆ Q T (cid:16) | x | (cid:17) − r (cid:16) E | D k f ε | p + E | D k g ε | p (cid:17) p d x d t (cid:12)(cid:12)(cid:12)(cid:12) p ≤ CT p (cid:16) f ε pk ; Q T + g ε pk ; Q T (cid:17) < ∞ . Considering the weighted Sobolev spaces and analogously to proving Lemma 3.1 but only with minor changes, wederive that (5.3) with free terms f ε and g ε admits a unique weak solution u ε satisfying E sup t ∈ [0 ,T ] (cid:12)(cid:12)(cid:12)(cid:12) ˆ R n (1 + | x | ) − r (cid:12)(cid:12)(cid:12) D k u ε ( x, t ) (cid:12)(cid:12)(cid:12) d x (cid:12)(cid:12)(cid:12)(cid:12) p < ∞ ∀ k ∈ N for any large r . Following from Sobolev’s embedding theorem, u ε is smooth in x and moreover, E sup ( x,t ) ∈Q T (1 + | x | ) − rp | D k u ε ( x, t ) | p ≤ C E sup t ∈ [0 ,T ] (cid:13)(cid:13)(cid:13) (1 + | x | ) − r D k u ε (cid:13)(cid:13)(cid:13) pH nx ≤ C E sup t ∈ [0 ,T ] n X i =0 (cid:12)(cid:12)(cid:12)(cid:12) ˆ R n (1 + | x | ) − r | D k + i u ε ( x, t ) | d x (cid:12)(cid:12)(cid:12)(cid:12) p < ∞ . Then we have E (cid:12)(cid:12) D k u ε ( x, t ) (cid:12)(cid:12) p < ∞ for each ( x, t ) ∈ Q T and k ∈ N . From global estimate (2.2) with δ/ insteadof δ and (5.4), we obtain u ε − u ε ′ m ; Q T ≤ C (cid:16) f ε − f ε ′ δ/ Q T + g ε − g ε ′ m + δ/ Q T (cid:17) → as ε, ε ′ → . Hence, u ε converges to a function u ∈ C m, x,t ( Q T ; L pω ) which is apparently a quasi-classical solutionto (5.3). Then we can derive the uniqueness and regularity from the estimate (2.2). The solvability is proved.To sum up, the proof of Theorem 2.3 is complete. ⊓⊔ Recall Equation (1.4): d u = ( − m +1 D m u d t + µD m u d w t with the initial condition with the initial condition u ( x,
0) = X n ∈ Z e − n m · e √− nx , x ∈ T = R / π Z . Since µ < , this equation admits a unique solution u ∈ L ( Ω ; C ([0 , T ]; H l ( T ))) for any integer l (cf. [26]).We shall prove that if p > /µ , then E k u ( · , t ) k pL ( T ) = + ∞ for any t > /ε , where ε = ( p − µ − > .Since u ∈ L ( Ω ; C ([0 , T ]; H l ( T ))) for any integer l , one can express u in the Fourier series u ( x, t ) = X n ∈ Z u n ( t )e √− nx where u n ( t ) satisfies d u n = ( − m +1 ( √− n ) m u n d t + µ ( √− n ) m u n d w t = u n ( − n m d t + µ ( √− m n m d w t ) ,u n (0) = e − n m . igher-order SPDEs 19 Then we obtain that u n ( t ) = exp (cid:26) − n m (1 + t + ( − m µ t ) + µ ( √− m n m w t (cid:27) . (6.1)Set f ( t ) := 2 + 2 t + ( − m µ t , then | u n ( t ) | = (cid:12)(cid:12)(cid:12) exp n − n m f ( t ) + 2 µ ( √− m n m w t o(cid:12)(cid:12)(cid:12) = (cid:12)(cid:12)(cid:12)(cid:12) exp (cid:26) − f ( t ) (cid:16) n m − µ ( √− m w t f ( t ) (cid:17) + µ ( − m | w t | f ( t ) (cid:27)(cid:12)(cid:12)(cid:12)(cid:12) . Using the condition that m is even , one can obtain | u n ( t ) | = exp ( − f ( t ) n m − µ ( − m/ w t f ( t ) ! + µ | w t | f ( t ) ) . By Parseval’s identity, k u ( · , t ) k L ( T ) = 2 π X n ∈ Z | u n ( t ) | = 2 π X n ∈ Z exp ( − f ( t ) n m − µ ( − m/ w t f ( t ) ! + µ | w t | f ( t ) ) . Therefore, we have E k u ( · , t ) k pL ( T ) = (2 π ) p E X n ∈ Z exp ( − f ( t ) n m − µ ( − m/ w t f ( t ) ! + µ | w t | f ( t ) )! p = (2 π ) p − ˆ + ∞−∞ exp (cid:26) − y (cid:27) X n ∈ Z exp ( − f ( t ) n m − µ ( − m/ yf ( t ) / √ t ! + µ y f ( t ) /t )! p d y = (2 π ) p − ˆ + ∞−∞ exp (cid:26) − y (cid:18) − pµ f ( t ) /t (cid:19)(cid:27) X n ∈ Z exp ( − f ( t ) n m − µ ( − m/ yf ( t ) / √ t ! )! p d y. Noticing the fact that µ ( − m/ y is positive on one of the intervals ( −∞ , and (0 , + ∞ ) , one has E k u ( · , t ) k pL ( T ) ≥ (2 π ) p − ˆ + ∞ exp (cid:26) − y (cid:18) − pµ f ( t ) /t (cid:19)(cid:27) X n ∈ Z exp ( − f ( t ) (cid:18) n m − | µ | yf ( t ) / √ t (cid:19) )! p d y ≥ (2 π ) p − X n ∈ Z ˆ + ∞ exp (cid:26) − y (cid:18) − pµ f ( t ) /t (cid:19)(cid:27) exp ( − p f ( t ) (cid:18) n m − | µ | yf ( t ) / √ t (cid:19) ) d y. Since (cid:12)(cid:12)(cid:12) n m − | µ | yf ( t ) / √ t (cid:12)(cid:12)(cid:12) ≤ when y ∈ h n m f ( t ) | µ |√ t , ( n m + 1) f ( t ) | µ |√ t i , one can further derive that E k u ( · , t ) k pL ( T ) ≥ (2 π ) p − X n ∈ Z ˆ ( n m +1) f ( t ) / ( | µ |√ t ) n m f ( t ) / ( | µ |√ t ) exp (cid:26) − y (cid:18) − pµ f ( t ) /t (cid:19)(cid:27) exp n − p f ( t ) o d y ≥ (2 π ) p − e − p f ( t ) X n ∈ Z ˆ f ( t ) | µ |√ t exp (cid:26) − ( y + n m ) (cid:18) − pµ f ( t ) /t (cid:19)(cid:27) d y. Obviously, the last term is infinite if − pµ f ( t ) /t < , (6.2)which is satisfied when t > /ε , where ε = ( p − µ − > . The lemma is proved. Remark 6.1
When m is odd, it follows from (6.1) that | u n ( t ) | = exp n − n m f ( t ) o where f ( t ) = 2 + (2 − µ ) t . Furthermore, one can obtain E k u ( · , t ) k pL ( T ) = (2 π ) p X n ∈ Z exp n − n m f ( t ) o! p , which means that the condition µ < is sufficient to ensure E k u ( · , t ) k pL ( T ) < + ∞ for any p ≥ . References
1. Boccia, S.: Schauder estimates for solutions of higher-order parabolic systems. Methods Appl. Anal. (1), 47–68 (2013)2. Brze´zniak, Z., Veraar, M.: Is the stochastic parabolicity condition dependent on p and q ? Electron. J. Probab (56), 1–24 (2012)3. Cardon-Weber, C.: Cahn-Hilliard stochastic equation: existence of the solution and of its density. Bernoulli (5), 777–816 (2001)4. Chow, P.L.: Stochastic Partial Differential Equations. CRC Press (2014)5. Cioica, P.A., Kim, K.H., Lee, K., Lindner, F.: On the L q ( L p ) -regularity and Besov smoothness of stochastic parabolic equationson bounded Lipschitz domains. Electron. J. Probab (82), 1–41 (2013)6. Da Prato, G., Debussche, A.: Stochastic Cahn-Hilliard equation. Nonlinear Anal.-Theory Methods Appl. (2), 241–263 (1996)7. Da Prato, G., Zabczyk, J.: Stochastic equations in infinite dimensions. Cambridge university press (2014)8. Dalang, R.C., Khoshnevisan, D., Nualart, E.: Hitting probabilities for systems of non-linear stochastic heat equations with additivenoise. ALEA Latin Am. J. Probab. Math. Stat. , 231–271 (2007)9. Dong, H., Zhang, H.: Schauder estimates for higher-order parabolic systems with time irregular coefficients. Calc. Var. PartialDiffer. Equ. (1), 47–74 (2015)10. Du, K.: W ,p -solutions of parabolic SPDEs in general domains. Stoch. Process. Their Appl. (2018)11. Du, K., Liu, J.: On the Cauchy problem for stochastic parabolic equations in Hölder spaces. Trans. Am. Math. Soc (4), 2643–2664 (2019)12. Du, K., Liu, J., Zhang, F.: Stochastic continuity of random fields governed by a system of stochastic PDEs. to appear in Annales del’Institut Henri Poincaré, Probabilités et Statistiques (arXiv:1706.01588)13. Gilbarg, D., Trudinger, N.S.: Elliptic partial differential equations of second order. springer (2015)14. Kim, I.: A BMO estimate for stochastic singular integral operators and its application to SPDEs. J. Funct. Anal. (5), 1289–1309(2015)15. Kim, K.H.: On L p -theory of stochastic partial differential equations of divergence form in C domains. Probab. Theory Relat.Fields (4), 473–492 (2004)16. Kim, K.H.: On stochastic partial differential equations with variable coefficients in C domains. Stoch. Process. Their Appl. (2),261–283 (2004)17. Kim, K.H.: A weighted Sobolev space theory of parabolic stochastic PDEs on non-smooth domains. J. Theor. Probab. (1),107–136 (2014)18. Kim, K.H., Lee, K.: A note on W γp -theory of linear stochastic parabolic partial differential systems. Stoch. Process. Their Appl. (1), 76–90 (2013)19. Krylov, N.V.: On L p -theory of stochastic partial differential equations in the whole space. SIAM J. Math. Anal. (2), 313–340(1996)20. Krylov, N.V.: An analytic approach to SPDEs. Stochastic partial differential equations: six perspectives , 185–242 (1999)21. Krylov, N.V.: SPDEs in L q ((0 , τ ] , L p ) Spaces. Electron. J. Probab. (2000)22. Krylov, N.V.: On the foundation of the L p -theory of stochastic partial differential equations. Stochastic Partial Differential Equa-tions and Applications-VII pp. 179–191 (2005)23. Krylov, N.V.: A brief overview of the L p -theory of SPDEs. Theory Stoch. Process. (2), 71–78 (2008)24. Krylov, N.V., Lototsky, S.V.: A Sobolev space theory of SPDEs with constant coefficients in a half space. SIAM J. Math. Anal. (1), 19–33 (1999)25. Krylov, N.V., Rozovskii, B.L.: On the Cauchy problem for linear stochastic partial differential equations. Math. USSR Izv. (6),1267–1284 (1977)26. Krylov, N.V., Rozovskii, B.L.: Stochastic evolution equations. Itogi Nauki i Tekhniki. Seriya" Sovremennye Problemy Matematiki.Noveishie Dostizheniya" , 71–146 (1979)27. Krylov, N.V., Rozovskii, B.L.: Characteristics of degenerating second-order parabolic Itô equations. J. Soviet Math. (4), 336–348(1986)28. Lindner, F.: Singular behavior of the solution to the stochastic heat equation on a polygonal domain. Stoch. Partial Differ. Equ.,Anal. Computat. (2), 146–195 (2014)29. Mikulevicius, R.: On the Cauchy problem for parabolic SPDEs in Hölder classes. Ann. Probab. pp. 74–103 (2000)igher-order SPDEs 2130. Pardoux, E.: Equations aux derives partielles stochastiques non lineaires monotones. Etude de solutions fortes de type Ito,(1975).Ph.D. thesis, Thése Doct. Sci. Math. Univ. Paris Sud31. Phillips, R.S., Hille, E.: Functional analysis and semi-groups. RI (1957)32. Portal, P., Veraar, M.: Stochastic maximal regularity for rough time-dependent problems. Stoch. Partial Differ. Equ., Anal. Compu-tat. pp. 1–57 (2019)33. Rozovskii, B.L.: Stochastic evolution systems: linear theory and applications to non-linear filtering. Springer (1990)34. Trudinger, N.S.: A new approach to the Schauder estimates for linear elliptic equations. In: Miniconference on Operator Theoryand Partial Differential Equations, pp. 52–59. Centre for Mathematics and its Applications, Mathematical Sciences Institute, TheAustralian National University (1986)35. Van Neerven, J., Veraar, M., Weis, L.: Maximal L p -regularity for stochastic evolution equations. SIAM J. Math. Anal. (3),1372–1414 (2012)36. Walsh, J.B.: An introduction to stochastic partial differential equations. In: École d’Été de Probabilités de Saint Flour XIV-1984,pp. 265–439. Springer (1986)37. Wang, X.J.: Schauder estimates for elliptic and parabolic equations. Chin. Ann. Math. Ser. B (6), 637–642 (2006)38. Zakai, M.: On the optimal filtering of diffusion processes. Probab. Theory Relat. Fields11