CHENCHEN MOU AND YINGFEI YI

Similar documents
arxiv: v1 [math.ap] 25 Jul 2012

Regularity of Weak Solution to Parabolic Fractional p-laplacian

Some lecture notes for Math 6050E: PDEs, Fall 2016

Laplace s Equation. Chapter Mean Value Formulas

A LOCALIZATION PROPERTY AT THE BOUNDARY FOR MONGE-AMPERE EQUATION

SUBELLIPTIC CORDES ESTIMATES

Integro-differential equations: Regularity theory and Pohozaev identities

arxiv: v1 [math.ca] 15 Dec 2016

MULTIPLE SOLUTIONS FOR CRITICAL ELLIPTIC PROBLEMS WITH FRACTIONAL LAPLACIAN

ON PARABOLIC HARNACK INEQUALITY

Fractional Cahn-Hilliard equation

Richard F. Bass Krzysztof Burdzy University of Washington

FRACTIONAL SOBOLEV SPACES WITH VARIABLE EXPONENTS AND FRACTIONAL P (X)-LAPLACIANS. 1. Introduction

Fractional Sobolev spaces with variable exponents and fractional p(x)-laplacians

1. Introduction Boundary estimates for the second derivatives of the solution to the Dirichlet problem for the Monge-Ampere equation

REGULARITY OF THE MINIMIZER FOR THE D-WAVE GINZBURG-LANDAU ENERGY

ESTIMATES FOR ELLIPTIC HOMOGENIZATION PROBLEMS IN NONSMOOTH DOMAINS. Zhongwei Shen

Brunn-Minkowski inequality for the 1-Riesz capacity and level set convexity for the 1/2-Laplacian

A PROOF OF THE TRACE THEOREM OF SOBOLEV SPACES ON LIPSCHITZ DOMAINS

NONTRIVIAL SOLUTIONS TO INTEGRAL AND DIFFERENTIAL EQUATIONS

Obstacle problems for nonlocal operators

Homogenization of Neuman boundary data with fully nonlinear operator

Regularity for Poisson Equation

ON THE SPECTRUM OF TWO DIFFERENT FRACTIONAL OPERATORS

Math The Laplacian. 1 Green s Identities, Fundamental Solution

THE L 2 -HODGE THEORY AND REPRESENTATION ON R n

hal , version 1-22 Nov 2009

Regularity of the obstacle problem for a fractional power of the laplace operator

Friedrich symmetric systems

Everywhere differentiability of infinity harmonic functions

Metric Spaces and Topology

S chauder Theory. x 2. = log( x 1 + x 2 ) + 1 ( x 1 + x 2 ) 2. ( 5) x 1 + x 2 x 1 + x 2. 2 = 2 x 1. x 1 x 2. 1 x 1.

arxiv:math/ v2 [math.ap] 3 Oct 2006

The continuity method

Fractional order operators on bounded domains

ESTIMATES FOR THE MONGE-AMPERE EQUATION

POINTWISE BOUNDS ON QUASIMODES OF SEMICLASSICAL SCHRÖDINGER OPERATORS IN DIMENSION TWO

SOLUTION OF THE DIRICHLET PROBLEM WITH L p BOUNDARY CONDITION. Dagmar Medková

The Brézis-Nirenberg Result for the Fractional Elliptic Problem with Singular Potential

STABILITY OF THE FELLER PROPERTY FOR NON-LOCAL OPERATORS UNDER BOUNDED PERTURBATIONS

ON THE REGULARITY OF SAMPLE PATHS OF SUB-ELLIPTIC DIFFUSIONS ON MANIFOLDS

Obstacle Problems Involving The Fractional Laplacian

The Fractional Laplacian

SOLUTION OF POISSON S EQUATION. Contents

Some aspects of vanishing properties of solutions to nonlinear elliptic equations

Elementary Theory and Methods for Elliptic Partial Differential Equations. John Villavert

CHAPTER 2. Laplace s equation

On critical systems involving fractional Laplacian

Methods on Nonlinear Elliptic Equations

Liouville Theorems for Integral Systems Related to Fractional Lane-Emden Systems in R N +

A RELATIONSHIP BETWEEN THE DIRICHLET AND REGULARITY PROBLEMS FOR ELLIPTIC EQUATIONS. Zhongwei Shen

Abstract. 1. Introduction

Asymptotic behavior of infinity harmonic functions near an isolated singularity

REGULARITY AND COMPARISON PRINCIPLES FOR p-laplace EQUATIONS WITH VANISHING SOURCE TERM. Contents

Optimal regularity and control of the support for the pullback equation

Sobolev Spaces. Chapter Hölder spaces

A REMARK ON MINIMAL NODAL SOLUTIONS OF AN ELLIPTIC PROBLEM IN A BALL. Olaf Torné. 1. Introduction

Green s Functions and Distributions

Small energy regularity for a fractional Ginzburg-Landau system

Gradient estimates for eigenfunctions on compact Riemannian manifolds with boundary

Traces, extensions and co-normal derivatives for elliptic systems on Lipschitz domains

On polynomially integrable convex bodies

arxiv: v1 [math.ap] 13 Jan 2017

Lecture Notes in Advanced Calculus 1 (80315) Raz Kupferman Institute of Mathematics The Hebrew University

Quantitative Homogenization of Elliptic Operators with Periodic Coefficients

ON THE HÖLDER CONTINUITY OF MATRIX FUNCTIONS FOR NORMAL MATRICES

Brunn-Minkowski inequality for the 1-Riesz capacity and level set convexity for the 1/2-Laplacian

BLOWUP THEORY FOR THE CRITICAL NONLINEAR SCHRÖDINGER EQUATIONS REVISITED

and finally, any second order divergence form elliptic operator

ERRATUM TO AFFINE MANIFOLDS, SYZ GEOMETRY AND THE Y VERTEX

VISCOSITY SOLUTIONS. We follow Han and Lin, Elliptic Partial Differential Equations, 5.

TRACES FOR FRACTIONAL SOBOLEV SPACES WITH VARIABLE EXPONENTS

Chapter One. The Calderón-Zygmund Theory I: Ellipticity

SOLUTION OF THE DIRICHLET PROBLEM WITH A VARIATIONAL METHOD. 1. Dirichlet integral

SHARP BOUNDARY TRACE INEQUALITIES. 1. Introduction

Robustness for a Liouville type theorem in exterior domains

Heat kernels of some Schrödinger operators

Partial Differential Equations, 2nd Edition, L.C.Evans Chapter 5 Sobolev Spaces

REGULARITY RESULTS FOR THE EQUATION u 11 u 22 = Introduction

Regularity estimates for fully non linear elliptic equations which are asymptotically convex

Regularity of the density for the stochastic heat equation

FOURIER SERIES WITH THE CONTINUOUS PRIMITIVE INTEGRAL

MAXIMAL AVERAGE ALONG VARIABLE LINES. 1. Introduction

The Dirichlet problem for non-divergence parabolic equations with discontinuous in time coefficients in a wedge

Singular Integrals. 1 Calderon-Zygmund decomposition

Boot camp - Problem set

We denote the space of distributions on Ω by D ( Ω) 2.

A Hopf type lemma for fractional equations

WEYL S LEMMA, ONE OF MANY. Daniel W. Stroock

Functional Analysis. Franck Sueur Metric spaces Definitions Completeness Compactness Separability...

Introduction. Christophe Prange. February 9, This set of lectures is motivated by the following kind of phenomena:

A Representation of Excessive Functions as Expected Suprema

SELF-ADJOINTNESS OF SCHRÖDINGER-TYPE OPERATORS WITH SINGULAR POTENTIALS ON MANIFOLDS OF BOUNDED GEOMETRY

Regularity of local minimizers of the interaction energy via obstacle problems

Fact Sheet Functional Analysis

DECAY ESTIMATES FOR THE KLEIN-GORDON EQUATION IN CURVED SPACETIME

Asymptotic Behavior of Infinity Harmonic Functions Near an Isolated Singularity

Large Solutions for Fractional Laplacian Operators

Recent developments in elliptic partial differential equations of Monge Ampère type

Published online: 29 Aug 2007.

Transcription:

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN CHENCHEN MOU AND YINGFEI YI Abstract. In this paper, we study interior regularity properties for the regional fractional Laplacian operator. We obtain the integer order differentiability of the regional fractional Laplacian, which solves a conjecture of Guan and Ma [9]. We further extend the integer order differentiability to the fractional order of the regional fractional Laplacian. Schauder estimates for the regional fractional Laplacian are also provided.. Introduction This paper is devoted to the study of interior regularity for the regional fractional Laplacian operators. Given real numbers 0 < s <, ɛ > 0, and an open set, denote (.) s u(y),ɛ = A(n, s) dy, Bɛ c(x) x y n+s where A(n, s) = s s Γ( n+s ) π n Γ( s ), B ɛ (x) is the open ɛ-ball in centered at x, and u dx (+ x ) n+s ), i.e., L (, is defined as (+ x ) n+s dx <. (.) s = lim ɛ 0 s,ɛ, The regional s-fractional Laplacian s on u L (, dx ), ( + x ) n+s provided that the limit exists. The regional s-fractional Laplacian can be also defined on the closure of by talking in place of in the above. We note that, if x, then s = s. When is a bounded Lipschitz open set, the regional s-fractional Laplacian s is in fact the Feller generator of the so-called reflected symmetric s-stable process (X t ) t 0 on, i.e., a Hunt process associated with the regular Dirichlet form (E, F) on L (, dx): E(u, v) = A(n, s) ( u(y))(v(x) v(y)) x y n+s dxdy, { F = u L ( u(y)) } ( ), x y n+s dxdy <. It is first shown in [] that if 0 < s, then the censored s-stable process in is essentially the reflected s-stable process (X t ) t 0, and if < s <, then the censored s-stable process in is identified as a proper subprocess of (X t ) t 0 killed upon leaving. Later, it is 000 Mathematics Subject Classification. Primary 35J05; Secondary 3B0, 60G5, 60J75. Key words and phrases. Interior regularity, Regional fractional Laplacian, Differentiability, Schauder estimates, Reflected symmetric s-stable process. This research was partially supported by NSF grant DMS090. The second author was also supported by a NSERC discovery grant, a scholarship from Jilin University, and a faculty development fund from the University of Alberta.

C. MOU AND Y. YI shown in [3] that (X t ) t 0 can be refined to be a Feller process starting from each point of which admits a Hölder continuous transition density function. In [9], not only does the Feller generator of (X t ) t 0 on shown to be the regional s-fractional Laplacian, but also a semi-martingale decomposition of (X t ) t 0 is obtained by studying the differentiability of the regional fractional Laplacian and its integration by parts property. For other studies on regional fractional Laplacians, we refer the reader to [7] for a more general integration by parts formula of the regional fractional and fractional-like Laplacian, and to [8] for some boundary Harnack inequalities for the regional fractional Laplacian on C,β (), s < β. If =, the regional fractional Laplacian s becomes the usual fractional Laplacian ( ) s defined via Fourier transform: F(( ) s u)(ξ) = ξ s F(u)(ξ) (see [4]). If we furtherly let s tend to, then the fractional Laplacian ( ) s becomes the classical Laplacian, and it is clear that u C α for some integer α > implies that u C α. In the case that u C α for some α > s with α s not being an integer, one also has ( ) s u C αs ([, Proposition.7]). A natural problem then is whether the regional fractional Laplacian shares similar regularity properties as that of the classical and fractional Laplacian. This problem is first investigated in [9] in which the following results are proved. Theorem. Let be an open set in and u L dx (, ) for some 0 < s <. Then (+ x ) n+s the following holds. a) ([9, Proposition 8.3]) If u C,α () for some α > s when 0 < s < or u C,α () for some α > s when s <, then s u C,0 (). b) ([9, Theorem 8.]) In the case n =, if r is a non-negative integer such that u C r,α () for some some α > s when 0 < s < or u C r+,α () for some α > s when s <, then s u C r,0 (). In the above and also through the rest of the paper, for each non-negative integer r and 0 α, we denote by C r,α () (C r,α ( )) the subspace of C r,0 () (C r,0 ( )) consisting functions whose rth partial derivatives are locally (uniformly) α-hölder continuous in. It is conjectured in [9] that part b) of the above theorem should hold for higher dimensions as well. In this paper, we give an affirmative answer to this conjecture. More precisely, we will prove the following result. Theorem A. Let be an open set and u L (, (+ x ) n+s ) for some 0 < s <. If r is a non-negative integer such that u C r,α () for some α > s or u C r+,α () for some α with + α > s α, then s u C r,0 (). Unlike the fractional Laplacian, the differential operator and the regional fractional Laplacian are not exchangeable in order. To overcome this difficulty in the proof of Theorem A, we derive a class of integral identities (see Lemma.) and use them to conclude that all possible singular terms of D r ( s,ɛ u) as ɛ 0 + are in fact non-singular. Making further estimates, we are able to extend the integer order differentiability result in Theorem A to a fractional order. Throughout the paper, for each j = (j, j j n ) N n, we denote j = j + j + + j n and j u = j u ( x ) j ( x ) j ( x. For any u n) jn Cr,α (), where r

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 3 is a non-negative integer and 0 α, define { supx, j =r j, if α = 0; [u] r,α; = sup j j u(y) x,y,x y, j =r xy, if α > 0. α Then we have the follows result analogous to [, Proposition.7] in the case of regional fractional Laplacian. Theorem B. Let be an open set and u L (, (+ x ) n+s ) for some 0 < s <. Then the following holds. (i) If u C r,α () for some positive number α with s < α and a non-negative integer r, then s C r,αs (), and moreover, [ s u] r,αs; C[u] r,α;. (ii) If u C r,α () for some positive number α with α < s < + α and a positive integer r, then s C r,+αs (), and moreover, [ s u] r,+αs; C[u] r,α;. Schauder estimate is well-known for the classical Laplacian (see [6]) as well as for the fractional Laplacian (see [, 5, 0, ]). We refer the reader to [, 5] for interior and boundary regularity theory for more general fractional operators. Using Schauder estimates for the fractional Laplacian, we are able to show a similar Schauder estimate holds for the regional fractional Laplacian. For each u C r,α ( ), denote u C r,α ( ) = { r j=0 [u] j,0,, if α = 0; u C r,0 ( ) + [u] r,α;, if α > 0. For simplicity, through the rest of the paper, we use the notation C α () (C α ( )), where α > 0, to denote the space C r,α () (C r,α ( )), where r is the largest integer smaller than α and α = α r. We note that if α is an integer r, then C α () = C α, () C α,0 () (C α ( ) = C α, ( ) C α,0 ( )). For a given open set in with, let d x = dist(x, ) and δ = {x ; d x < δ}. We denote Cc () as the space of C functions with compact support in, S as the Schwartz space of rapidly decreasing C function in, and Λ ( ) as the Zygmund space of all bounded functions on such that [u] := u(x + h) + u(x h) Λ ( ) <. h sup x,x+h,xh We equip the space Λ ( ) with the norm u := u Λ ( ) L ( ) +[u] Λ ( ). It is easy to see that the regional fractional Laplacian s is well-defined for functions u C () L (). We can then extend the definition of the regional fractional Laplacian to the space L (, dx For any u L (, ), we define (+ x ) n+s (.3) < s u, ϕ > := u(y) s ϕ(y)dy, for any ϕ Cc (). dx (+ x ) n+s ).

4 C. MOU AND Y. YI In Theorems C and D below, the definition of s is understood in the sense of (.3). Theorem C. Let 0 < α and 0 < s <. If, for some w C α ( ), u L () solves the equation s u = w in, then u C α+s ( δ ) C( u L () + w C α ( ) ), where δ > 0 is sufficiently small and C is a constant depending only on n, s, δ and α. Remark.. The notions of C α+s ( δ ) and C α ( ) we defined earlier have unified different cases for α, α+s being or not being natural numbers in the statement of Theorem C. We note that C ( δ ) = C 0, ( δ ) C,0 ( δ ), C ( ) = C 0, ( ) C,0 ( ), and C ( δ ) = C, ( δ ) C,0 ( δ ). Theorem D. Let 0 < s <. Suppose that, for some w L (), u L () solves the equation s u = w in. Then, for any sufficiently small δ > 0, there exists a constant C > 0 depending only on n, s, and δ such that the following holds: (i) If s, then (ii) If s =, then u C s ( δ ) C( u L () + w L ()). u Λ ( δ ) C( u L () + w L ()). The rest of sections are devoted to the proof of these results. In Section, we will give a class of integral identities and use them to obtain the integer order differentiability of the regional fractional Laplacian s. Both Theorem A and B will be proved in this section. In Section 3, Schauder estimate for the regional fractional Laplacian will be given and Theorem C and D will be proved.. Differentiability In this section, we will study the differentiability of the regional fractional Laplacian in an open set and prove Theorems A, B. The proof will be based on some integral identities in... Integral identities. Given n N, z = (z, z,, z n ) and k = (k, k,, k n ) N n, we denote by z k the monomial n i= zk i i. Also, for each j =,,, n, we let e j to denote the jth standard coordinate vector in. Lemma.. Consider an annulus domain R δɛ (0) := {z : ɛ < z < δ}, where 0 < ɛ < δ. Then for any i, j, m N and k = (k, k,, k n ) N n, we have z k+e i k i + z m dz = z k+e j k j + z m dz. R δɛ (0) R δɛ (0) Proof. The lemma clearly holds if i = j. Let i j, and without loss of generality, we assume that i = n and j = n. Denote z = (z, z,, z n ), k = (k, k,, k n ), and consider polar coordinate (r, θ) in the (z n, z n )-plane. Then under the new coordinate

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 5 R δɛ (0) becomes R δɛ (0) (0, π) = {( z, r) R + : ɛ < z + r < δ } (0, π), and integration by parts yields that z k zn R δɛ (0) z m dz π z k = r k n+k n+3 d zdr cos kn+ θsin R δɛ (0) ( z + r ) m kn θdθ 0 π z k = (k n + ) r k n+k n+3 d zdr cos k R δɛ (0) ( z + r ) m n θsin kn+ θdθ 0 π z k k n r k n+k n+3 d zdr cos kn+ θsin ( z + r ) m kn θdθ R δɛ (0) = (k n + ) R δɛ (0) from which the lemma follows. z k z n z m dz k n R δɛ (0) 0 z k zn z m dz, Lemma.. ([9, Lemma 8.]) Let be an open subset of and u L (). Suppose that u is continuous in an open neighborhood U of x 0 = (x,, x n ) and dist(x 0, U) > ɛ > 0. Then the function f(x) = Bɛ c (x) u(y)dy is differentiable at x 0 and f x i (x 0 ) = B ɛ(x 0 ) u(y) x i y i x y m(dy), where m(dy) is the n dimensional surface Lebesgue measure... Proof of Theorems A. We first prepare some technical lemmas concerning derivatives of s,ɛ. Lemma.3. Let be an open set in and 0 < s <. Suppose that u L (, (+ x ) n+s ) C,0 (). Then for any x, δ < d x = dist(x, ), and 0 < ɛ < δ, we have (.) s A(n, s) x,ɛ i n j= = ( u x j (y) u x j (x))(y j x j )(x i y i ) x y n+s+ dy [u(y) + n j= u x (n ) j (x)(x j y j )](x i y i ) x y n+s+ dy u x i (x) (u(y) )(x i y i ) dy + x y n+s x y n+s+ m(dy) B c δ (x) (n + s) where := B δ (x) B c ɛ(x). B δ (x) B c δ (x) (u(y) )(x i y i ) y x n+s+ dy, Proof. The proof follows from that of [9, Proposition 8.3].

6 C. MOU AND Y. YI Lemma.4. Let be an open set in and 0 < s <. Suppose that u L (, (+ x ) n+s ) C r+,0 () for some positive integer r. For any x, ɛ > 0, δ < d x = dist(x, ), and l = (l, l, l n ), k = (k, k,, k n ) N n, if k i = 0 for some i n, then x i = (n + p m ) l u(y) j =m A j j! (y x)j l+j x y n+s+p (x y) k dy l u(y) j = A j j! (y x)j l+j x y n+s+p+ (x y) k+e i dy [ l+j u(y) j =m A j j! (y x)j l+j +j x y n+s+p+ (y x) j ] (x y) k+e i dy, j! where A j = j!j! j n! for each j = (j, j,, j n ) N and m, p N are such that l + m = r and m + k = p. Proof. Since k i = 0, we have by Lemma. that I : = l u(y) j =m A j j! (y x)j l+j x i x y n+s+p (x y) k dy = I + I + I 3, where I := m! j =m A j l+j+e i (y x) j (x y) k x y n+s+p dy, I := l u(y) j =m A j j! (n + s + p) l+j x y n+s+p+ (x y) k+e i dy, I 3 := B ɛ(x) B δ (x) l u(y) j =m A j j! (y x)j l+j x y n+s+p+ (x y) k+e i m(dy) l u(y) j =m A j j! (y x)j l+j x y n+s+p+ (x y) k+e i m(dy). Using the identity m + k = p, integration by parts yields δ I = (n + s + p) r nsp dr r n+pm (x y) k+e i = ɛ B (x) [ j =m l A j u(x + r(y x)) j! (y x)j l+j r j ] m(dy) δ r n+pm{ ɛ B (x) } (x y) k+e i m(dy) = I 3 (n + p m ) j =m [ l u(x + r(y x)) dr (n+s+p) δ ɛ B (x) r ms (x y) k+e i A j j! (y x)j l+j r j ]

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 7 [ j =m l A j u(x + r(y x)) j! (y x)j l+j r j ] m(dy)dr δ r ms (x y) k+e i ɛ B (x) [ j =m A l+j u(x + r(y x))(y x) j j j! (y +j x)j l+j +j r j ] m(dy)dr l u(y) j =m A j j! = I 3 (n + p m ) (y x)j l+j x y n+s+p+ (x y) k+e i dy [ l+j u(y) j =m A j j! (y x)j l+j +j x y n+s+p+ (y x) j ] (x y) k+e i dy. Therefore, where I () 3 = (n + p m ) I () 3 = n + p I (3) 3 = Observe that I + I () 3 = t N,t m n + p = t N,t m C t m m! j = [ j =m,j i =t t N,t m C t+ I + I 3 = I () 3 + I() 3 + I(3) 3, A j l+j l u(y) j = A j j! (y x)j l+j x y n+s+p+ (x y) k+e i dy, (y x) j (x y) k+e i x y n+s+p+ dy, l+j u(y) j =m A j j! (y x)j l+j +j x y n+s+p+ (y x) j ] (x y) k+e i dy. C t+ j =m,j i =t Rδɛ(x) (y x) j (x y) k x y n+s+p+ n + p t N,t m C t+ l+j+e i j =,j i =t+ Rδɛ(x) (y x) j (x y) k x y n+s+p+ (x y) j dy l+j (y x) j (x y) k+e i x y n+s+p+ dy (t + ) l+j+e i (x y) j dy j =,j i =t+ where j = (j, j,, j n ). An application of Lemma. yields I + I () 3 l+j (y x) j (x y) k+e i x y n+s+p+ dy,

8 C. MOU AND Y. YI = t N,t m C t+ j =m,j i =t (y x) j (x y) k+e i x y n+s+p+ dy = 0. ] [n + p ( k + n + m) l+j+e i Thus, I = I () 3 + I(3) 3 and the lemma is proved. Lemma.5. Let be an open set in and 0 < s <. Suppose that u L (, (+ x ) n+s ) C r+,0 () for some positive integer r. For any x, ɛ > 0, δ < d x = dist(x, ), and l = (l, l, l n ), k = (k, k,, k n ) N n, if k i 0 for some i n, then x i = k i l u(y) j =m A j j! (y x)j l+j x y n+s+p (x y) k dy l u(y) j = A j j! (y x)j l+j x y n+s+p (x y) ke i dy (n + p m ) l u(y) j = A j j! (y x)j l+j x y n+s+p+ (x y) k+e i dy [ l+j u(y) j =m A j j! (y x)j l+j +j x y n+s+p+ (y x) j ] (x y) k+e i dy, where m, p N are such that l + m = r and m + k = p. Proof. Since k i 0, we have by Lemma. that I = I + Ī + I + I 3, where I and I i, i =,, 3, are as in the proof of Lemma.4 and l u(y) j =m A j j! Ī = k (y x)j l+j i x y n+s+p (x y) ke i dy. By the proof of Lemma.4, we have I + I 3 = I () 3 + I() 3 + I(3) 3. where I (i) 3, i =,, 3, are defined in the proof of Lemma.4. Write Ī = Ī () + Ī (), where Note that Ī () = ki Ī () = k i I + Ī () + I () 3 l u(y) j = A j j! (y x)j l+j x y n+s+p (x y) ke i dy, j = A j l+j (y x) j (x y) ke i x y n+s+p dy.

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 9 = m! j =m + k i n + p A j l+j+e i (y x) j (x y) k x y n+s+p+ j = j = A j l+j A j l+j First let k i be an even number. Then I + Ī () + I () 3 = Cm t m! t N,t m + k i n + p = t N,t m j =m,j i =t t N,t m C t+ l+j+e i j =,j i =t+ (y x) j (x y) ke i x y n+s+p+ t N,t m C t+ C t+ j =m,j i =t Rδɛ(x) (y x) j (x y) k x y n+s+p+ + t N,t m C t+ j =,j i =t+ j =m,j i =t It follows from Lemma. that = I + Ī () + I () 3 C t+ t N,t m (x y) j dy (y x) j (x y) ke i x y n+s+p+ (y x) j (x y) k+e i x y n+s+p+ dy. Rδɛ(x) (y x) j (x y) k x y n+s+p+ l+j (x y) j dy (x y) j dy (x y) j dy l+j (y x) j (x y) k+e i x y n+s+p+ dy (t + k i + ) l+j+e i (x y) j dy (n + p) l+j+e i j =m,j i =t (y x) j (x y) k+e i x y n+s+p+ dy = 0. Now let k i be an odd number. Then I + Ī () + I () 3 = t N,t m C t+ m m! j =m,j i =t+ (y x) j (x y) k+e i x y n+s+p+ dy. ] [(n + p) ( k + n + m) l+j+e i l+j+e i (y x) j (x y) k x y n+s+p+ (x y) j dy

0 C. MOU AND Y. YI = + k i n + p k i (n + p) t N,t m + k i n + p t N,t C t j =,j i =t (y x) j (x y) ke i x y n+s+p+ t N,t j =,j i =0 j =,j i =0 C t+ m m! C t j =,j i =t A j l+j A j l+j j =m,j i =t+ (y x) j (x y) k x y n+s+p+ t N,t m C t+ (y x) j (x y) ke i x y n+s+p+ t N,t m C t+ l+j (x y) j dy l+j (y x) j (x y) k+e i x y n+s+p+ dy (y x) j (x y) ke i x y n+s+p+ l+j+e i (x y) j dy j =,j i =t+ It again follows from Lemma. that = I + Ī () + I () 3 t N,t m + t N,t m j =,j i =0 (x y) j dy j =,j i =t+ (y x) j (x y) k+e i x y n+s+p+ dy C t+ (y x) j (x y) k+e i x y n+s+p+ dy l+j (x y) j dy l+j (y x) j (x y) k+e i x y n+s+p+ dy. ] A j [(m + n + k (n + p)) l+j j =m,j i =t+ (y x) j (x y) k x y n+s+p+ C t+ j =m,j i =t+ (y x) j (x y) k+e i x y n+s+p+ dy (t + + k i ) l+j+e i (x y) j dy (n + p) l+j+e i

= INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN t N,t m C t+ j =m,j i =t+ (y x) j (x y) k+e i x y n+s+p+ dy = 0. ] [(n + p) ( k + n + m) l+j+e i Thus, for any k i 0, I = Ī () + I () 3 + I(3) 3 and the lemma is proved. Lemma.6. Let be an open set in and 0 < s <. Suppose that u L (, (+ x ) n+s ) C r,0 () for some positive integer r. Then for any x, ɛ > 0, ˆr = (r, r,, r n ) N n with ˆr = r, δ < d x = dist(x, ), all ɛ-dependent terms of ˆr s,ɛ have the form (.) I ɛ,δ l,k,m,p (x) = l u(y) j =m A j j! (y x)j l+j x y n+s+p (x y) k dy, where l = (l, l, l n ), k = (k, k,, k n ) N n, m, p N are such that l + m = r and m + k = p. Proof. We will prove the lemma by induction. In the case of r =, we observe that the only ɛ-dependent terms on the right hand side of (.) are its first two terms. They clearly have the form (.) with the first term corresponding to l =, m = 0, k = and p =, and the second term corresponding to l = 0, m =, k = and p =. Now suppose that (.) is satisfied when r = q, where q is a fixed positive integer. We want to show that it is also satisfied when r = q +, i.e., for any l, k N with l + m = q and m + k = p, all ɛ-dependent terms of have the form (.3) I = x i l u(y) j =m A j j! (y x)j l+j x y n+s+p (x y) k dy l u(y) j =m A j j! (y x)j l +j (x y) k dy, x y n+s+p where l + m = q + and m + k = p. If k i = 0, then, by Lemma.4, we have I = I () 3 + I(3) 3, where I() 3, I(3) 3 are as in the proof of Lemma.4 which clearly have the form (.3) with l + m = l + m + = q + and m + k = m + k + = (p + ). If k i 0, then, by Lemma.5, we have I = Ī () + I () 3 + I(3) 3, where Ī () is as in the proof of Lemma.5 which is clearly of the form (.3) with l + m = l + m + = q + and m + k = m + k = p. Proof of Theorem A. For any ɛ > 0, x, ˆr = (r, r,, r n ) N n with ˆr = r, and δ < d x = dist(x, ), we have by Lemma.6 that all ɛ-dependent terms of ˆr s,ɛ have the form (.). In the case α > s, we note that any integral of the form (.) is bounded above in absolute value by a constant times δ ɛ ραs dρ which is convergent as ɛ 0. It follows that

C. MOU AND Y. YI ˆr s,ɛ u converges uniformly on any compact subset of as ɛ 0. Thus, ˆr s u C(), i.e., s u C r,0 (). In the case +α > s α, we again consider an integral I ɛ,δ l,k,m,p (x) of the form (.) for some l, k N n, m, p N satisfying l + m = r and m + k = p. Since m + + k = p + is an odd number, we have, for any j N n with j = m +, that (y x) j (x y) k x y n+s+p dy = 0. Hence I ɛ,δ l,k,m,p (x) can be re-written as l u(y) j = A j j! (y x)j l+j x y n+s+p (x y) k dy which is bounded above in absolute value by a constant times δ ɛ ραs dρ that is convergent as ɛ 0. It follows again that ˆr s,ɛ u converges uniformly on any compact subset of as ɛ 0. Thus, ˆr s C(), i.e., s u C r,0 ()..3. Proof of Theorem B. Let u L (, (+ x ) n+s ) C r,α () for some positive integer r and some real numbers 0 < s <, 0 < α. For each ɛ > 0 sufficiently small, x, and any ˆr N n with ˆr r, we write ˆr s,ɛ = I ɛ (x) + I (x), where I ɛ (x) denotes the ɛ-dependent term of ˆr s,ɛ and I (x) denotes the remaining term. Lemma.7. Let u, s, r, α be as in the above. If either α > s or + α > s α, then ˆr s = I 0 (x) + I (x), if ˆr r, where I 0 (x) = lim ɛ 0 I ɛ (x) which consists of terms of the form (.4) I δ l,k,m,p (x) = B δ (x) l u(y) j =m A j j! (y x)j l+j x y n+s+p (x y) k dy, for any δ < d x = dist(x, ) and some l = (l, l, l n ), k = (k, k,, k n ) N n, m, p N with l + m = ˆr and m + k = p. Proof. It follows immediately from Lemma.6 and the proof of Theorem A. Lemma.8. Let u, s, r, α be as in the above. Then the following holds. a) If α > s and ˆr = r, then there exists a constant C > 0 such that I (x) I (y) C[u] r,α. x y αs, x, y, x y. b) If + α > s α and ˆr = r, then there exists a constant C > 0 such that I (x) I (y) C[u] r,α; x y +αs, x, y, x y.

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 3 Proof. The function I can be derived simply by taking higher order derivatives of the right hand side of (.) and identifying all ɛ-independent terms of the derivatives. As these terms involves only regular integrals, the lemma follows from straightforward estimates. Proof of Theorem B. Let x, y and take δ < d x,y = min{d x, d y }. For given l = (l, l, l n ), k = (k, k,, k n ) N n, m, p N, consider where Il,k,m,p δ is as in (.4). It is clear that where J := J := J = I δ l,k,m,p (x) Iδ l,k,m,p (y), B η(0) R δη (0) J = J + J, [ l u(x + z) j =m z n+s+p A j j! zj l+j l u(y + z) j =m A j j! zj l+j u(y) ] z n+s+p z k dz, [ l u(x + z) j =m A j j! zj l+j z n+s+p l u(y + z) j =m A j j! zj l+j u(y) ] z n+s+p z k dz, and η = x y < δ. (i) In this case, we let l +m = r and m+ k = p in J, J. On one hand, since l +m = r and u C r,α (), there exists a constant C > 0 such that j =m l u(x + z) l A j ( u(y + z) l+j l+j u(y) ) z j C [u] r,α; z m+α, z B η (0). j! Using the fact m + k = p, it follows that C [u] r,α; z p+α J dz B η(0) z n+s+p C [u] r,α; η αs = C [u] r,α; x y αs for some constant C > 0. On the other hand, we also have j =m l u(x + z) l A j ( u(y + z) l+j l+j u(y) ) z j j! m C 3 [u] r,α; [ x y m+αi z i + i=0 m z m+αi x y i ], z Bη(0), c where C 3 > 0 is a constant. It follows that C 3 [u] r,α; ( m i=0 J x y m+αi z i + m i= z m+αi x y i ) dz B c η(0) i= z n+s+m

4 C. MOU AND Y. YI m C[u] r,α; ( η ims x y m+αi + i=0 for some constant C 4 > 0. Hence (.5) J J + J (C + C 4 )[u] r,α. x y αs. m η αsi x y i ) C 4 [u] r,α; x y αs Let I 0 be as in Lemma.7. Then Lemma.7 together with (.5) imply that i= I 0 (x) I 0 (y) C 5 [u] r,α. x y αs for some constant C 5 > 0. With this estimate, the proof is now complete by Lemma.7 and Lemma.8 a). (ii) In this case, we let l + m = r and m + k = p in J, J. Since m + + k = p + is an odd number, we have, for any j N n with j = m +, any w, and any ρ < d w that z j+k dy = 0. z n+s+p It follows that J = B η(0) B ρ(w) [ l u(x + z) j = z n+s+p A j j! zj l+j l u(y + z) j = A j j! zj l+j u(y) ] z n+s+p z k dz, J = R δη (0) [ l u(x + z) j = z n+s+p A j j! zj l+j l u(y + z) j = A j j! zj l+j u(y) ] z n+s+p z k dz. The rest of the proof is similar to that of (i). We only note that using facts l + m = r and u C r,α (), the estimate of J follows from the inequality j = l u(x + z) l A j ( u(y + z) l+j l+j u(y) ) z j C 5 [u] r,α; z +α, j! z B η (0), where C 5 > 0 is a constant, while, using facts m + k = p and s < + α, the estimate of J follows from the inequality j = l u(x + z) l A j ( u(y + z) l+j l+j u(y) ) z j j! C 6 [u] r,α; ( x y +αi z i + z +αi x y i ), z Bη(0), c i=0 where C 6 > 0 is a constant. i=

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 5 3. Schauder estimates for the regional fractional Laplacian In this section, we will show the Schauder estimates for the regional fractional Laplacian using those for the fractional Laplacian. Recall that the fractional Laplacian ( ) s is welldefined in S, the Schwartz space of rapidly decreasing C functions in, and we can then extend its definition to the space L ( dx, ) by (+ x ) n+s (3.) < ( ) s u, ϕ >= u(y)( ) s ϕ(y)dy, ϕ S, for any u L (, dx (+ x ) n+s ). 3.. Schauder estimates for the fractional Laplacian. In the following Lemma 3. and 3., the definition of ( ) s is understood in the sense of (3.). We refer the reader to [] for a more general definition of the fractional Laplacian. Lemma 3.. Let 0 < α and 0 < s <. If, for some w C α ( ), u L ( ) solves the equation ( ) s u = w in, then for any δ > 0 sufficiently small there exists a constant C > 0 depending only on n, s, δ and α such that u C α+s ( δ ) C( u L ( ) + w C α ( ) ). Proof. The proof follows from that of [, Proposition.8]. Lemma 3.. Let 0 < s <. Suppose that, for some w L (), u L ( ) solves the equation ( ) s u = w in. Then, for any sufficiently small δ > 0, there exists a constant C > 0 depending only on n, s and δ such that the following holds: (i) If s, then (ii) If s =, then u C s ( δ ) C( u L ( ) + w L ()). u Λ ( δ ) C( u L ( ) + w L ()). Proof. We first use the argument in the proof of [, Proposition.8]. By covering and rescaling arguments, we only need to consider the case δ = B (0) and = B (0). Let (0). η Cc (R) be such that range(η) [0, ], supp(η) B (0), and η(x) = for any x B 3 4 Denote u 0 (x) := A(n, s) η(y)w(y) x y ns dy = ( ) s ηw(x). Then ( ) s u 0 = w = ( ) s u in B 3 (0). It follows that u u 0 C ( B (0)) and 4 u u 0 C ( B (0)) C u u 0 L ( ) C( u L ( ) + w L (B (0))), where C > 0 is a constant depending only on n. We note that C ( B (0)) = C, ( B (0)) C,0 ( B (0)). The lemma now follows from [, Proposition 5.].

6 C. MOU AND Y. YI 3.. Proof of Theorems C and D. Let 0 < s <, w L (), and u L () solves the equation s u = w in. Also let ū L ( ) be such that ū u in and ū 0 outside of. Then for any ϕ S, ( s ),ɛ ū(x) + A(n, s)ū(x) x y n+s dy ϕ(x)dx = A(n, s) ( = A(n, s) ( = A(n, s) \B ɛ(x) \B ɛ(x) \B ɛ(x) = A(n, s) ū(x) c \B ɛ(x) ū(x) ū(y) dyϕ(x)dx x y n+s \B ɛ(x) ū(x)ϕ(x) dydx x y n+s ū(x)ϕ(x) dydx x y n+s In particular, for any ϕ Cc (), we have s,ɛ ϕ(x)dx + A(n, s) = ( s,ɛ ū(x) + A(n, s)ū(x) = A(n, s) ū(x) \B ɛ(x) ϕ(x) ϕ(y) dydx. x y n+s supp(ϕ) c \B ɛ(x) ϕ(x) ϕ(y) dydx. x y n+s \B ɛ(y) \B ɛ(x) ϕ(x) c \B ɛ(x) ) x y n+s dy ϕ(x)dx Letting ɛ 0 in the above, we easily obtain that, for any x, (3.) ( ) s ū(x) = w(x) + A(n, s) dy. c x y n+s ū(y)ϕ(x) ) x y n+s dxdy ū(x)ϕ(y) ) x y n+s dydx dydx x y n+s Proof of Theorems C, D. Let δ > 0 be sufficiently small. We have by Lemma 3. and (3.) that there exists a constant C depending on n, s, and δ such that u C s ( δ ) C( ū L ( ) + w + A(n, s)u c y n+s dy L ( δ )) when s. Similarly, C( u L () + w L ()) u Λ ( δ ) C( u L () + w L ()). when s =. This proves Theorem D. Using bootstrap arguments, Lemma 3. and (3.) for both cases of α, α + s being or not being natural numbers, we have, similarly to the above, that u C α+s ( δ ) C( u L () + w C α ( ) ), where C > 0 is a constant depending on n, s, δ and α. This proves Theorem C. Acknowledgement We would like to thank the referee for pointing out references [, 5] to us and for valuable comments which lead to a significant improvement of the paper.

INTERIOR REGULARITY FOR REGIONAL FRACTIONAL LAPLACIAN 7 References [] K. Bogdan, K. Burdzy and Z. Q. Chen, Censored stable processes, Probab. Theory Related Fields 7 (003), 89 5. [] L. Caffarelli and P. Stinga, Fractional elliptic equations: Caccioppoli estimates and regularity, Ann. Inst. H. Poincaré Anal. Non Linéaire (04), to appear. [3] Z. Q. Chen and T. Kumagai, Heat kernel estimates for stable-like processes on d-sets, Stochastic Process Appl. 08 (003), 7 6. [4] E. Di Nezza, G. Palatucci and E. Valdinoci, Hithchhiker s guide to the fractional Sobolev spaces, Bull. Sci. Math. 36 (0), 5 573. [5] G. Grubb, Regularity of spectral fractional Dirichlet and Neumann problems, arxiv:4.3744 (04). [6] D. Gilbarg and N. S. Trudinger, Elliptic Partial Differential Equations of Second Order, Reprint of the 998 edition, Classics in Mathematics, Springer-Verlag, Berlin, 00. [7] Q. Y. Guan, Integration by parts formula for regional fractional Laplacian, Comm. Math. Phys. 66 (006), 89 39. [8] Q. Y. Guan, Boundary Harnack inequality for regional fractional Laplcian, arxiv:0705.64 (007). [9] Q. Y. Guan and Z. M. Ma, Reflected symmetric α-stable processes and regional fractional Laplacian, Probab. Theory Related Fields 34 (006), 649 694. [0] X. Ros-Oton and J. Serra, The Dirichlet problem for the fractional Laplacian: regularity up to the boundary, J. Math. Pures Appl. 0 (04), 75 30. [] L. Silvestre, Regularity of the obstacle problem for a fractional power of the Laplace operator, Comm. Pure Appl. Math. 60 (007), 67. C. Mou: School of Mathematics, Georgia Institute of Technology, Atlanta, GA 3033, USA E-mail address: cmou3@math.gatech.edu Y. Yi: Department of Mathematical & Statistical Sci, University of Alberta, Edmonton Alberta, Canada T6G G, School of Mathematics, Jilin University, Changchun, 300, PRC, and School of Mathematics, Georgia Institute of Technology, Atlanta, GA 3033, USA E-mail address: yingfei@ualberta.ca,yi@math.gatech.edu