Abstract
In a bounded domain, we consider a variable range nonlocal operator, which is maximally isotropic in the sense that its radius of interaction equals the distance to the boundary. We establish \(C^{1,\alpha }\) boundary regularity and existence results for the Dirichlet problem.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
1.1 General setting and the operator
Let \(n\ge 1\) and \(\Omega \subset \mathbb {R}^n\) be a bounded, connected domain of class \(C^{1,1}\). Let \(d_{\Omega }:{\overline{\Omega }}\rightarrow [0,+\infty )\) be the distance to the boundary,
When no confusion arises, we simply write \(d=d_\Omega \). It is convenient to smooth out the distance function by taking \(\delta \in C^{1,1}(\overline{\Omega })\) such that \(\delta =d\) when \(d<d_0\), for a small \(d_0>0\). A similar notation is employed for other domains.
Let \(s\in (0,1)\). Write \(B_r(x)=\left\{ y\in \mathbb {R}^n:|y-x|<r\right\} \) and \(B_r=B_r(0)\). We introduce the operator
where the normalization constant is given by
This is an isotropic regional fractional Laplacian, and the factor \(d(x)^{2s-2}\) is inserted to ensure that \(\mathcal {L}_\Omega u(x)\) converges as \(x\rightarrow \partial \Omega \) to a nontrivial limit, namely \(-\Delta u(x)\). See Lemma A.1.
Probabilistically speaking, the operator \(\mathcal {L}_\Omega \) generates a Lévy type process where a particle at \(x\in \Omega \) jumps randomly and isotropically in the largest possible ball \(B_{d(x)}(x)\) contained inside \(\Omega \).
We list some characteristic properties and consequences.
-
\(\mathcal {L}_\Omega \) enjoys a mid-range maximum principle Proposition 2.2, whose strength lies between the local one for \(-\Delta \) and the global one for \((-\Delta )^s\): no absolute minima exist in the region where \(\mathcal {L}_\Omega u\ge 0\), provided that u is non-negative in the domain of interaction outside of that region. As a result, local barriers suffice to control boundary behaviors.
-
The domain of interaction depends on the point of evaluation. Thus, extra effort is needed in the construction of barriers in Sect. 3, even following the established idea [11].
-
\(\mathcal {L}_{\Omega }\) is of order 2s but scales quadratically and satisfies the classical Hopf lemma, see Lemma A.4 and Lemma 4.6.
-
\(\mathcal {L}_\Omega \) is not variational. To see this, take \(\Omega =(0,1)\). Upon integrating by parts, one immediate sees “hidden boundary terms” at \(x=1/2\). In higher dimensions the “hidden boundary” can be thought of as points having at least two projections to the boundary. Unfortunately, all these points contribute in such a different way that the resulting expression would not be manageable. Consequently, no weak formulations due to integration by parts can be expected. Existence is to be established in the viscosity sense, in Sect. 8.
Generic and universal constants are denoted by C, c. They depend only on n, s and \(\Omega \).
1.2 Main results
Consider the Dirichlet problem
We study the regularity properties of its classical solutions u, meaning that \(u\in C^{2s+}(\Omega )\cap C({\overline{\Omega }})\), where \(C^{2s+}(\Omega )=\bigcup _{\beta >0}C^{2s+\beta }(\Omega )\).
Our main results are the following.
Theorem 1.1
(A priori regularity up to boundary) Suppose \(\Omega \) is a bounded domain of class \(C^{1,1}\) in \(\mathbb {R}^n\) with \(n\ge 1\), \(s\in (0,1)\) and \(u\in C^{2\,s+}(\Omega )\cap C({\overline{\Omega }})\) solves (1.2). Then there exists \(\alpha _0=\alpha _0(n,s)\in (0,1)\) such that for any \(\alpha \in (0,\alpha _0)\), the following holds. If either
-
(1)
\(s\in (\frac{1}{2},1)\), \(f\in L^\infty (\Omega )\); or
-
(2)
\(s\in (0,\frac{1}{2}]\), \(f\in C^{\alpha +1-2s}(\Omega )\),
then \(u\in C^{1,\alpha }(\overline{\Omega })\), with
Here C depends on \(n,s,\Omega \) and the corresponding norm of f.
Higher regularity up to the boundary remains open. There are two difficulties:
-
The condition \(\alpha <1\) is crucially used in the proof of Proposition 6.1, where the quadratic growth (linear in \(x'\) times linear in \(x_n\)) contradicts the control \(|x|^{1+\alpha }\).
-
A complete study of the action of \(\mathcal {L}_{\mathbb {R}^n_+}\) on monomials is missing (Lemma 4.3).
Theorem 1.2
(Existence) Suppose \(\Omega \) is a bounded domain of class \(C^{1,1}\) in \(\mathbb {R}^n\) with \(n\ge 1\), \(s,\beta \in (0,1)\) so that \(\beta +2s\) is not an integer, \(f\in C^\beta (\Omega )\). Then there exists a unique \(u\in C^{2s+\beta }(\Omega )\cap C(\overline{\Omega })\) solving (1.2). Moreover, if \(\beta >(1-2s)_+\), then \(u\in C^{1,\alpha }(\overline{\Omega })\), for some \(\alpha \in (0,2\,s+\beta -1)\) determined by Theorem 1.1.
Remark 1.3
Crucial to construction of the barrier (Proposition 3.1) is the uniform exterior ball condition. This is satisfied by bounded domains of class \(C^{1,1}\), as well as their blow-ups around a boundary point.
1.3 Main ideas
Let us explain the heuristics of the proof. By definition, u is \(C^{1,\alpha }\) at \(0\in \partial \Omega \) if
This is implied (see Lemma C.1) by the expansion
i.e. u/d is \(C^{\alpha }\) up to \(0\in \partial \Omega \). By building suitable barriers in Sect. 3, we will be able to obtain global Hölder regularity (Proposition 5.4). This allows the use of a blow-up argument to reduce the problem to a half plane.
It then suffices to prove a Liouville theorem (Proposition 6.1) for solutions to the homogeneous equation in the half space, namely
under the growth \(u(x)=O(|x|^{1+\alpha })\): the only solution is \(u=cx_n\). Since the homogeneous equation is preserved under scaling and tangential differentiations, solutions can be shown to be one dimensional (1D) (e.g. Lemma 6.4). Hence, we need to show that solutions \(u(x',x_n)\), which is independent of \(x'\), to
which grow no faster than \(x_n^{1+\alpha }\) must be linear (Lemma 6.2). This is in turn implied by the boundary regularity in the half-line (Proposition 4.1), namely
To show this, one simply proves a boundary Harnack inequality (Lemma 4.7) and an improvement of oscillation (Lemma 4.8).
1.4 Related works
The boundary Harnack inequality for a nonlocal elliptic operator in non-divergence form is proved by Ros-Oton–Serra in [12, Theorem 1.2]. A similar type of nonlocal operator with fixed horizon (range of interaction) at every point has been considered by Bellido and Ortega [6].
1.5 Generalizations
We expect that the techniques introduced in this paper should be able to prove similar results in the parabolic setting, and when \(\mathcal {L}_{\Omega }\) is replaced by an analogous integro-differential operator of order 2s with any homogeneous kernel.
2 Preliminary results
Clearly \(\mathcal {L}_\Omega \) satisfies the global maximum principle.
Lemma 1.4
(Global maximum principle) Suppose \(u\in C^{2\,s+}(\Omega )\cap C({\overline{\Omega }})\) solves
Then either \(u\equiv 0\) in \({\overline{\Omega }}\) or \(u>0\) in \(\Omega \).
Proof
At any interior minimum \(x_0\), \(\mathcal {L}_\Omega u(x_0)\le 0\), with strict inequality unless \(u\equiv u(x_0)\) in \(B_{d(x_0)}(x_0)\). But the latter ball contains a sequence converging to \(\partial \Omega \) where \(u\ge 0\). \(\square \)
A more careful examination yields the following version of maximum principle. It is especially useful to study the blow-up equation when the domain becomes unbounded.
Proposition 1.5
(Mid-range strong maximum principle) Let \({U}\subset \mathbb {R}^n\) be a domain that is not necessarily bounded. Suppose G is non-empty, bounded, open in U. The domain interacting with G is
Suppose \(u\in C^{2s+}(G)\cap C(\overline{G_*})\), is a solution to
Then \(u\ge 0\) in G.
Proof
If \(\min _{\overline{G}}u\) is attained on \(\partial G \subset \overline{G_*}{\setminus } G\) and not in G, then \(u\ge 0\) in G by the Dirichlet condition. If \(u(x_0)=\min _{G}u\le 0\), then
Thus equality holds and \(u\equiv u(x_0)=0\) in G. \(\square \)
The interior Harnack inequality is known to DiCastro–Kuusi–Palatucci [2].
Lemma 1.6
(Interior Harnack inequality) Suppose \(B_r(z)\subset \Omega \subset \mathbb {R}^n\), \(n\ge 1\). If \(u\ge 0\) in \(B_r(z)_*\) (as defined in (2.2)) and
then for any \(\eta >0\), there exists \(C_0(n,s,\Omega ,\eta )>0\) such that
Now we state the interior estimates which follows from the corresponding result for the restricted fractional Laplacian [9, 10]. We denote
Lemma 1.7
(Interior estimates) Suppose \(U\subset \mathbb {R}^n\) is not necessarily bounded, and \(B_1\subset B_4\subset U\). Suppose \(u\in C^{2\,s+}(\overline{B_1})\cap C({\overline{U}})\) is a solution to
for \(f\in L^\infty (B_1)\). Then there exists a constant \(C=C(n,s,\epsilon )>0\) such that
Moreover, if \(f\in C^\beta (\overline{B_1})\) for \(\beta \in (0,1)\) and \(\beta +2s\) is not an integer, then there exists \(C=C(n,s,\beta )>0\) such that
where \(d_1=\left\| {d_U}\right\| _{L^\infty (B_1)}+2\).
Remark 2.5
We notice that, because the operator \(\mathcal {L}_U\) degenerates away from the boundary, so does the estimate (through the term \(\left\| {d_U}\right\| _{L^\infty (B_1)}^{2-2s}\)).
Proof
Let us extend u by zero outside U. In this proof we write \(d=d_U\). At each interior point \(x\in U\), one can rewrite the equation in terms of the restricted fractional Laplacian, namely
where
with \(c_{n,s}=2^{2s}\pi ^{-\frac{n}{2}}\Gamma (\frac{n+2s}{2})/|\Gamma (-s)|\) being the normalization constant for \((-\Delta )^s\). We first prove the \(C^{2s}\) (or \(C^{1-\epsilon }\)) regularity. Note that for \(x\in B_1\), we have \(d(x)\ge 3\) and \(B_{d(x)}(x)^c\subset B_1^c\), so
Suppose first \(s\ne \frac{1}{2}\). By [10, Theorem 1.1(a)],
When \(s=\frac{1}{2}\), using again [10, Theorem 1.1(a)], we replace accordingly the \(C^{2s}\) norm by \(C^{1-\epsilon }\) norm for any \(\epsilon \in (0,1)\), with the constant depending also on \(\epsilon \).
Now we prove the higher regularity. Up to multiplicative constants, we decompose \(g=g_1+g_2+g_3\) where
Since \(\left[ {\varphi ^p}\right] _{C^\beta }\le p \left\| {\varphi ^{p-1}}\right\| _{L^\infty }\left[ {\varphi }\right] _{C^\beta }\) for \(\beta \in (0,1)\) and \(p\in \mathbb {R}\), we use the bounds \(3\le d(x)\le {{\,\textrm{inrad}\,}}(U)\) and \(\left\| {d}\right\| _{C^{0,1}({\overline{U}})}\le 1\) to control
For \(x,y\in \overline{B_1}\), we express
For the integral in the first line, note that \(3\le d(x)\le |z|\). By mean value theorem, there exists \(x_*\in B_1\) such that
For the second line we note that there is a nontrivial contribution only in the symmetric difference \(B_{d(x)}(x)^c \triangle B_{d(y)}(y)^c\), which lies in an annulus of width at most of order \(|x-y|\). More precisely, we have
Therefore, using \(|z|\ge d(x) \ge 3 \ge 3|y|\) and \({{\,\textrm{supp}\,}}u\subset {\overline{\Omega }}\),
where \(d_1:=\left\| {d}\right\| _{L^\infty (B_1)}+2\). In summary,
Now, by [9, Corollary 2.4],
where we have absorbed \(\left\| {u}\right\| _{L^1_{2s+1}(U)}\) by \(\left\| {u}\right\| _{L^1_{2s}(U)}\). \(\square \)
3 The barriers
3.1 Super-solution near the boundary
We construct a barrier in the spirit of [10]. Since the domain of interaction varies from point to point, we must compute at all points.
The idea is to consider powers of the distance function to a ball which touches the domain from the outside, since we want the super-solution to be strictly positive except at the contact point, however near the boundary.
Proposition 1.9
(Super-solution) Let \({U}\subset \mathbb {R}^n\) be a possibly unbounded domain of class \(C^{1,1}\). Suppose \(x_0\in \partial {U}\) can be touched by an exterior ball of radius \(b>0\). Then, there exists a constant \(r_0>0\) and a function \(\varphi ^{(x_0)}\in C^2 \bigl ({U\cap B_{2r_0}(x_0)}\bigr ) \cap C^{0,1}\bigl (\overline{U\cap B_{2r_0}(x_0)}\bigr )\) satisfying
Here \(\nu (x_0)\) is the (outer) normal at \(x_0\in \partial \Omega \), and \(r_0\) and C depend only on n, s and b.
Upon a translation and a rotation, the exterior ball condition states that \(B_b(-be_n)\) touches \(\partial {U}\) at \(x_0=0\in \partial {U}\) from the outside.
The super-solution will be built from
for \(p=1\) and \(p\sim 2^-\). As in [10], at each point \(x\in {U}\) we compare \(d_{B_b(-be_n)}(y)^p\) with the one-dimensional function
where
are the hyperplanes which are orthogonal to
and contain respectively \(-be_n+b{\varvec{v}}\) and \(-be_n\).
First we consider the planar barrier. This uses the fact that \(-\mathcal {L}_{U}\) behaves like the Laplacian near \(\partial {U}\). (Notice that this computation is valid for \(d_{\mathcal {T}_x}^p\) only at the point x, although it is all we will need.)
Lemma 1.10
For \(p>0\) and \(x\in U\), we have
Here the constant in \(O(|p-2|)\) depends only on n and s.
Proof
Using \(d_{\mathcal {T}_x}(x+ty)=d_{\mathcal {T}_x}(x)+ty\cdot {\varvec{v}}\) for \(t\in [-1,1]\), we have
Changing variable to \(y=d_{\mathcal {T}_x}(x)z\) (recall that \(d_{\mathcal {T}_x}(x)>0\) for \(x\ne 0\in \partial U\)) and choosing another coordinate system for z such that \({\varvec{v}}\) is the direction of the last coordinate axis, we have (here \(\psi \) is defined in (B.1))
By Lemma B.1,
as desired. \(\square \)
Next we compare \(d_{B_b(-be_n)}^p\) and \(d_{\mathcal {T}_x}^p\) pointwise. For each fixed \(x\in U\), hence \({\varvec{v}}\in \mathbb {S}^{n-1}\), we denote the projection of a vector \(y\in \mathbb {R}^n\) onto \(\mathcal {P}_x\) by
Lemma 1.11
For \(x\in U\) and \(z\in B_{d_U(x)}(x)\),
Proof
For \(z\in U\), we express
to see it is non-negative. We make the following observations:
-
Since \((x+be_n)'=0\), \((z+be_n)'=(z-x)'\).
-
The sum of radii of the interior disjoint balls \(B_{d_U(x)}(x)\) and \(B_b(-be_n)\) is at most the distance between the centers, giving \(d_U(x)+b \le |x+be_n|\). This implies
$$\begin{aligned} (z+be_n)\cdot {\varvec{v}} =(z-x)\cdot {\varvec{v}}+|x+be_n| \ge |x+be_n| - |z-x| \ge |x+be_n| - d_U(x) \ge b. \end{aligned}$$
Thus
Now we can compute \(\mathcal {L}_{U}(d_{-be_n}^p)\) locally near the boundary. \(\square \)
Lemma 1.12
Let \(p\in [1,2]\) and \(x\in U\). Then we have
and
Here C depends only on n and s.
Proof
We split
Since \(\mathcal {T}_x\) is chosen such that \(d_{B_b(-be_n)}(x)=d_{\mathcal {T}_x}(x)\) and \(d_{B_b(-be_n)}\ge d_{\mathcal {T}_x}\) on \(B_{d_U(x)}(x)\),
Then (3.2) follows from Lemma 3.2.
For (3.3), since \(d_{\mathcal {T}_x}\) is linear hence \(\mathcal {L}_U\)-harmonic, by Lemma 3.3 we have
We are ready to prove Proposition 3.1. \(\square \)
Proof of Proposition 3.1
Let \({\widetilde{\varphi }}(x)=2d_{B_b(-be_n)}(x)-d_{B_b(-be_n)}^{p}(x)\), where \(p<2\) is chosen (using Lemma 3.4) such that
Then
whenever x is close enough to \(B_b(-be_n)\), say \(d_{B_b(-be_n)}(x)\le 2r_0<1\). On the other hand, we verify that
provided that \(d_{B_b(-be_n)}(x)\le 1\). Moreover, on \(\left\{ r_0 \le d_{B_b(-be_n)} \le 2r_0\right\} \cap U\) where \(2r_0<1\),
Therefore, \(\varphi ^{(0)}=r_0^{-1}{\widetilde{\varphi }}\) is the desired super-solution. \(\square \)
3.2 Super-solution for a bounded domain
A concave paraboloid serves as a simple global super-solution. Choose a coordinate system such that \(0\in \Omega \). Let \(M={{\,\textrm{diam}\,}}\Omega \) so that \(\Omega \subset B_M\). Consider the positive, strictly concave function
When \(\Omega =B_M\), this is known as the torsion function.
Lemma 1.13
There holds
Proof
For any \(x\in \Omega \) and \(y\in B_{d(x)}\), the parallelogram law implies
Thus
\(\square \)
Remark 3.6
By requiring that \(\Omega \) be compactly contained in \(B_M\), one obtains a strict super-solution. However, we will not need this.
4 Boundary Harnack inequality in 1D
We are interested in one-dimensional (1D) Dirichlet problems on the half space \(\mathbb {R}^n_+\). Note that for \(x=(x',x_n)\in \mathbb {R}^n_+\),
Throughout this section we assume that u is 1D (i.e. u depends only on \(x_n\)). Then
It is convenient to write \(x\in \mathbb {R}_+\) in place of \(x_n\). In this section we will prove the following
Proposition 1.15
(Boundary regularity in 1D) Suppose \(u\in C^{2s+}((0,1))\cap C([0,2))\) is a 1D solution to
There exists \(\alpha _*\in (0,2\,s \wedge 1)\) and \(C>0\) such that
Here the C and \(\alpha _*\) depend only on n and s.
Remark 4.2
Note that the function x is a model solution to (4.1), and in fact the unique solution on \(\mathbb {R}_+\) up to a constant multiple, as we will show in Lemma 6.2. In other words, any two solutions are comparable up to the boundary in a Hölder continuous way.
4.1 Preliminaries
The scaling property Lemma A.4 allows us to compute the action of \(\mathcal {L}_{\mathbb {R}_+}\) on monomials.
Lemma 1.17
(Monomials on the half line) For any \(p\ge 0\),
where \(a(p)=-\psi (p,1)\), as defined in (B.1). In particular, \(a(0)=a(1)=0\) and \(a(2)=-2\).
Remark 4.4
When \(n=1\), by a series expansion,
However, it is not clear from this expression if a(p) is monotone or signed for large p.
Proof
Let \(r>0\). Applying Lemma A.4 to \(\Omega =r^{-1}\Omega =\mathbb {R}^n_+\) and \(u(x)=x^p\), we see that
By linearity,
Thus \(a(p)=\mathcal {L}_{\mathbb {R}^n_+}x^p\big |_{x=1}\). \(\square \)
We will use the following version of strong maximum principle for functions with non-negative data in the adjacent interval of the same length.
Lemma 1.19
(Strong maximum principle) Suppose \(u\in C^{2s+}((0,1))\cap C([0,2))\) solves
Then either \(u\equiv 0\) on (0, 1), or
Proof
This is simply Proposition 2.2 with \(G=\mathbb {R}^{n-1}\times (0,1)\) and \(G_*=\mathbb {R}^{n-1}\times (0,2)\). \(\square \)
4.2 Boundary Harnack inequality
First of all we show Proposition 4.1 for \(\alpha =0\), using interior Harnack inequality and comparison arguments.
Lemma 1.20
(Two-sided estimate) Suppose \(u\in C^{2s+}((0,1))\cap C([0,2))\) solves
then there exists \(C>0\) universal such that
Proof
By replacing u by u/u(1) if necessary, we may assume that \(u(1)=1\). By Lemma 2.3, there exists \(C>0\) universal such that
Applying Lemma 4.5 to \(u-C^{-1}x\) and \(2Cx-u\) on \((0,\tfrac{1}{2})\) yields the result. \(\square \)
Corollary 1.21
(Boundary Harnack inequality) Let \(u\in C^{2s+}((0,1))\cap C([0,2))\) be a solution to
Then there exists \(C>0\) universal such that
4.3 Boundary Hölder regularity
Lemma 1.22
(Improvement of oscillation) Suppose \(u\in C^{2s+}((0,1))\cap C([0,2))\) solves
For \(k=1,2,\dots \), denote
Then there exists a universal constant \(c\in (0,1)\) such that for any \(k\ge 1\),
Proof
By replacing u by u/u(1) if necessary, we may assume that \(u(1)=1\). By Lemma 4.7, we can take \(m_1=C_3^{-1}\) and \(M_1=C_3\). We assume in the following that u(x)/x is not a constant; otherwise we can trivially take \(M_k=m_k\) for all \(k\ge 2\).
Suppose \(M_k>m_k>0\) for \(k\ge 1\) is known, such that
By Lemma A.4, both the functions \((u-m_k)(2^{-1}4^{-k}x)\) and \((M_k-u)(2^{-1}4^{-k}x)\) are \(\mathcal {L}_{\mathbb {R}_+}\)-harmonic and non-negative on (0, 2). As they solve (4.2), the strong maximum principle Lemma 4.5, they are strictly positive on (0, 1). This means that
Similarly, the functions \((u-m_k)(4^{-k-1}x)\) and \((M_k-u)(4^{-k-1}x)\) solve (4.3), so that Lemma 4.7 implies that
Rescaling and multiplying throughout by the normalizing factor \(4^{k+1}\), we have
This means that
Adding up these two inequalities,
Thus
Now a standard iteration yields the Hölder continuity of the quotient. \(\square \)
Proof of Proposition 4.1
By replacing u by u/u(1) if necessary, we assume that \(u(1)=1\). By Lemma 2.4, we know that \(u\in C^{\beta }((0,\tfrac{3}{4}))\) for some \(\beta >0\) and \(\left\| {u}\right\| _{C^{\beta }(B_{d/2}(d))} \le Cd^{-\beta }\) for \(d>0\). Fix \(\theta>(1+\beta )/\beta >1\). Let \(x,y\in [0,1/4)\). Write \(r=|x-y|\), \(d=x\wedge y\). If \(r\le d^{\theta }/2\), then by Lemma 2.4,
If \(r\ge \frac{d^\theta }{2}\), then \(x,y\in (0,d+r)\) and by iterating Lemma 4.8, we have
Hence,
as desired. \(\square \)
5 Hölder regularity up to boundary
5.1 Pointwise boundary Harnack inequality
Using the global maximum principle, we obtain a direct pointwise bound which is good for controlling the interior behavior.
Lemma 1.23
(Interior control) Let \(u\in C^{2\,s+}(\Omega )\cap C({\overline{\Omega }})\) be a solution to
Then
Proof
Use Lemma 2.1 on \( \left\| {f}\right\| _{L^\infty (\Omega )}\varphi ^{(1)} +\left\| {g}\right\| _{L^\infty (\partial \Omega )} \pm u \), with \(\varphi ^{(1)}\) given in Lemma 3.5. \(\square \)
We can now control a solution by the distance function.
Lemma 1.24
(Global boundary Harnack principle) Suppose \(u\in C^{2\,s+}(\Omega )\cap C({\overline{\Omega }})\) solves
Then there exists a universal constant C such that
Proof
Since \(\Omega \) is a bounded domain of class \(C^{1,1}\), there exists \(b>0\) such that an exterior tangent ball of radius b exists at each point \(x_0\in \partial \Omega \). By Proposition 3.1, in a suitable coordinate system there exists \(\varphi ^{(x_0)}\) such that
Since
Proposition 2.2 applies, we have
Since \(\varphi ^{(x_0)}\) grows linearly away from the boundary, we have
The interior estimate simply follows from Lemma 5.1. \(\square \)
We present a local analogue in a half ball \(B_r^+=B_r \cap \left\{ x_n>0\right\} \), where \(r>0\).
Lemma 1.25
(Local boundary Harnack principle) Suppose \(u\in C^{2\,s+}(B_1^+) \cap C(\overline{B_2^+})\) solves
Then
Here C depends only on n and s.
Proof
Let \(x_0\in \partial \mathbb {R}^n_+\cap \partial B_1^+\). By Proposition 3.1 with \(b=1\), there is a universal \(r_0\in (0,1/2)\) such that
By Proposition 2.2,
Now for each \(x\in B_1^+ \cap \left\{ 0<x_n<r_0\right\} \) we choose \(x_0=(x',0)\) to obtain
for C universal. The result follows by combining it with the trivial estimate
5.2 Hölder regularity up to boundary
As above, we give a global and a local result. While practically having an order of 2s in the interior, the operator satisfies the classical Hopf boundary lemma. Thus the minimum of the two yields the combined regularity. This effect is analogously seen with the spectral fractional Laplacian.
Proposition 1.26
(Global boundary regularity) Suppose \(u\in C^{2\,s+}(\Omega )\cap C({\overline{\Omega }})\) solves
Then for any \(\epsilon \in (0,1)\), there exists a constant \(C=C(n,s,\Omega ,\epsilon )>0\) such that
Proof
By dividing by \(\left\| {f}\right\| _{L^\infty (\Omega )}\) if necessary, we can assume \( \left\| {f}\right\| _{L^\infty (\Omega )}\le 1. \) By Lemma 5.2,
Let
We need to show that
Write \(\rho =\min \left\{ d(x),d(y)\right\} =d(x)\), by interchanging x and y if necessary.
- Case 1:
-
\(4|x-y|<\rho \). Then \(y\in B_{\rho /4}(x)\subset B_\rho (x)\subset \Omega \). By Lemma A.3 and Lemma A.4, the rescaled function \(u_\rho (z)=u(x+\rho z)\) satisfies
$$\begin{aligned} \mathcal {L}_{\rho ^{-1}(\Omega -x)} u_\rho (z) =f_\rho (z):=\rho ^2 f(x+\rho z) \quad \text { in }B_{1/4}\subset B_1\subset \Omega . \end{aligned}$$(5.2)Using the interior estimates Lemma 2.4, we have
$$\begin{aligned}{} & {} \left\| {u_\rho }\right\| _{C^{\beta }(B_{1/4})} \le C\left( \left\| {u_\rho }\right\| _{L^\infty (B_1)} +\left\| {u_\rho }\right\| _{L^1_{2s}(\Omega )}\right. \\{} & {} \left. +\left\| { d_{ \rho ^{-1}(\Omega -x_0) } }\right\| _{L^\infty (B_1)}^{2-2s} \left\| {f_\rho }\right\| _{L^\infty (B_1)} \right) \end{aligned}$$In view of (5.1) we observe that
$$\begin{aligned} \begin{aligned} \left[ {u_\rho }\right] _{C^{\beta }(B_{1/4})}&= \rho ^\beta \left[ {u}\right] _{C^{\beta }(B_{\rho /4}(x))}\\ \left\| {u_\rho }\right\| _{L^\infty (B_1)}&\le C \left\| {d}\right\| _{L^\infty (B_\rho (x))} \le C \rho \le C\rho ^\beta \\ \end{aligned}\\ \begin{aligned}&\left\| {u_\rho }\right\| _{L^1_{2s}(\Omega )} = \left\| {u(x+\rho \cdot )}\right\| _{L^1_{2s}(\Omega )}\\&\le C\left\| {d(x+\rho \cdot )}\right\| _{L^1_{2s}(\Omega )}\\&\le Cd(x)\left\| {1}\right\| _{L^1_{2s}(\Omega )} +C\rho \left\| {1}\right\| _{L^1_{2s-1}(\Omega )}\\&\le C\rho \left( 1+\int _{ 1\le |z| \le \rho ^{-1} {{\,\textrm{diam}\,}}\Omega } \dfrac{1}{|z|^{n+2s-1}} \,dz \right) \\&\le {\left\{ \begin{array}{ll} C\rho &{} \text { for }s\in (\frac{1}{2},1),\\ C\rho (1+\log \frac{1}{\rho }) &{} \text { for }s=\frac{1}{2},\\ C\rho (1+\rho ^{2s-1}) &{} \text { for }s\in (0,\frac{1}{2}),\\ \end{array}\right. }\\&\le C\rho ^\beta . \end{aligned}\\ \begin{aligned}&\left\| { d_{ \rho ^{-1}(\Omega -x_0) } }\right\| _{L^\infty (B_1)}^{2-2s} \left\| {f_\rho }\right\| _{L^\infty (B_1)}\\&\le \left\| {d}\right\| _{L^\infty (\Omega )}^{2-2s}\rho ^{2s-2} \cdot \rho ^2 \le C \rho ^{2s} \le C \rho ^\beta . \end{aligned}\end{aligned}$$We conclude that
$$\begin{aligned} \left[ {u}\right] _{C^{\beta }(B_{\rho /4}(x))} \le C \quad \textit{ i.e. } \quad |u(x)-u(y)|\le C |x-y| \quad \text { for }|x-y|<\frac{\rho }{4}. \end{aligned}$$ - Case 2:
-
\(|x-y|\ge \frac{\rho }{4}\). Then
$$\begin{aligned}\begin{aligned} |u(x)-u(y)| \le |u(x)|+|u(y)|&\le C \left( d(x)+d(y)\right) \\&\le C (2d(x)+|x-y|)\\&\le C |x-y| \le C|x-y|^\beta . \end{aligned}\end{aligned}$$
\(\square \)
Proposition 1.27
(Local boundary regularity) Suppose \(u\in C^{2\,s+\varepsilon }_{\text {loc}}(B_1^+)\cap C(\overline{B_2^+})\) solves
Then for any \(\epsilon \in (0,1)\), there exists a constant \(C=C(n,s,\epsilon )>0\) such that
Proof
By normalizing if necessary, we assume \(\left\| {u}\right\| _{L^\infty (B_2^+)}\le 1\). Let
We need to show that
Without loss of generality let \(\rho =x_n\le y_n\). By Lemma 5.3 we have
Case 1: \(4|x-y|<\rho \). Then \(y\in B_{\rho /4}(x)\subset B_\rho (x)\subset B_1^+\). As in the proof of Proposition 5.4, the rescaled function \(u_\rho (z)=u(x+\rho z)\) they satisfy the equation (note \(\rho ^{-1}\ge 16\))
By (5.3),
From Lemma 2.4 we have the estimate
Therefore, by (5.3),
Case 2: \(4|x-y|\ge \rho \). Then by (5.3),
\(\square \)
6 Liouville-type results
In this section we classify solutions to homoegenous Dirichlet problems in a half space with controlled growth. We write \(\mathbb {R}^n_+=\left\{ x=(x',x_n)\in \mathbb {R}^{n-1}\times \mathbb {R}_+\right\} \).
Proposition 1.28
(Liouville-type result) Let v be a solution to
which satisfies the growth condition
for some \(\alpha \in (0,\alpha _*)\) with \(\alpha _*\in (0,2s\wedge 1)\) given in Proposition 4.1. Then v is a 1D and linear, i.e.
for some constant \(b_0\in \mathbb {R}\).
Lemma 1.29
(Liouville in 1D) If \({\bar{v}}\) solves
and satisfies the growth condition
where \(\alpha \in (0,\alpha _*)\) with \(\alpha _*\in (0,2s\wedge 1)\) given in Proposition 4.1. Then
for some \(c_0\in \mathbb {R}\).
Proof
Let
which satisfies the growth condition
In particular,
Applying Proposition 4.1 to \({\bar{v}}_R\), we see that
as \(R\rightarrow +\infty \). Hence \({\bar{v}}/x\) is a constant \(c_0\in \mathbb {R}\). \(\square \)
Lemma 1.30
(Solutions with slow growth vanish) Suppose v solves (6.1) with
for \(\beta \in [0,\beta _0)\) where
Then \(v\equiv 0\).
Proof
The rescaled function \(v_R(x)=R^{-\beta _0}v(Rx)\) satisfies (6.1) and the growth condition
By Proposition 5.5,
as \(R\rightarrow \infty \). Hence, \(v\equiv v(0)=0\). \(\square \)
Lemma 1.31
(Solutions with mild growth are 1D) Suppose v satisfies (6.1) and the growth condition
for \(\beta \in [\beta _0,2\beta _0)\cap (0,1+\alpha _*)\) where \(\beta _0\) is as in (6.3) and \(\alpha _* \in (0,2s\wedge 1)\) is given in Proposition 4.1. Then v is a 1D, i.e.
for some \(b_0\in \mathbb {R}\).
Proof
Let \(h\in (0,1]\) and \(\omega \in \mathbb {S}^{n-1}\cap \left\{ x_n=0\right\} \). Write
which satisfies (6.1) and the growth condition (via the rescaling as in Lemma 6.3)
Since \(\beta -\beta _0\in [0,\beta _0)\), Lemma 6.3 implies \(w\equiv 0\). Then \(v(x+h\omega )=v(x)\) for any \(x\in \mathbb {R}^n_+\), \(h\in (0,1]\), \(\omega \in \mathbb {S}^{n-1}\cap \left\{ x_n=0\right\} \). Since \((h\omega )\mathbb {Z}\) is arbitrary on \(\left\{ x_n=0\right\} \), v depends only on \(x_n\). By Lemma 6.2, \(v(x)=b_0x_n\) for some \(b_0\in \mathbb {R}\). \(\square \)
Proof of Proposition 6.1
We will prove by induction in k the following claim: if v satisfies (6.1) and the growth condition
and \(k\beta _0<1+\alpha _*\), then v is 1D and linear.
By Lemma 6.4, this is true for \(k=1\). Suppose the claim is true for k and v is a solution to (6.1) satisfying
By the rescaling argument and boundary regularity (e.g. in Lemma 6.4), the Hölder difference quotient \(\frac{v(x+h\omega )-v(x)}{h^{\beta _0}}\) satisfies (6.1) and (6.4). Hence, there exists \(b_0(h,\omega )\) such that
By iterating (6.5) for \(h=1\), we have
for any \(R\in {\mathbb {N}}\). But then for \(x=(0,R)\), by (6.4) (recall that \((k+1)\beta _0<1+\alpha _*<2\)) we have
contradicting (6.4) unless \(b_0(1,\omega )\equiv 0\) for all \(\omega \in \mathbb {S}^{n-1}\cap \left\{ x_n=0\right\} \). In view of (6.5), v depends only on \(x_n\) and the result follows from Lemma 6.2. \(\square \)
7 Proof of the higher regularity
Proof of Theorem 1.1
In view of the interior estimates in Lemma 2.4, we just need to prove the following expansion: for some \(\alpha =\alpha (n,s)\in (0,1)\) so small that Proposition 6.1 holds, for any \(z\in \partial \Omega \), there exists \(Q_z\in \mathbb {R}\), \(r>0\) such that for any \(x\in \Omega \cap B_r(z)\),
Indeed, as in the proof of Proposition 4.1, one can interpolate the (degenerate) interior estimate with (7.1) to obtain the full \(C^{1,\alpha '}(\overline{\Omega })\) regularity (for some \(\alpha '\in (0,\alpha )\)).
Suppose on the contrary that there exists \(z\in \partial \Omega \) such that (7.1) does not hold for any \(Q\in \mathbb {R}\), i.e.
We split the proof by contradiction into a number of steps.
Step 1: Choosing one Q for each r.
For each \(r>0\) small, we choose a Q(r) that minimizers \(\left\| {u-Qd}\right\| _{L^2(B_r(z))}\), i.e.
We claim that
Suppose on the contrary that (7.2) does not hold, i.e there exists a (large) \({\bar{C}}>0\) such that
Then, for any \(x\in B_r(z)\),
Since \(\sup _{B_r(z)}d=r\),
Since for any \(j\ge i \ge 0\),
the limit \(Q_0:=\lim _{r\searrow 0}Q(r)\) exists, and by fixing \(i=0\) and letting \(j\rightarrow \infty \),
In particular, putting \(r=1\) implies \(|Q_0|\le C({\bar{C}}+1)\), since \(|Q(1)|\le C\). Hence, for all \(r\in (0,1]\),
a contradiction. This proves (7.2).
Step 2: The blow-up sequence and growth bound.
Now we define the monotone quantity
From \(\lim _{r\searrow 0}\theta (r)=\infty \), there is a sequence \(r_m\rightarrow 0\) such that
Define the blow-up sequence \(v_m:(r_m)^{-1}(\Omega -z)\rightarrow \mathbb {R}\),
which satisfies
and, from the choice of \(Q(r_m)\),
We claim the following growth control
Indeed, the arguments in Step 1 (replacing \(\theta \) by \(\theta (r)\), and the interval (0, 1] by [r, 1]) shows that
Also since \(\theta \) is non-increasing,
Then (here we implicitly extend suitable functions by 0 outside \(\Omega \))
This proves (7.5).
Step 3: Equation for the blow-up sequence.
Let \(\Omega _m=(r_m)^{-1}(\Omega -z)\), which converges to a halfspace \(\left\{ x\cdot e>0\right\} \) as \(m\rightarrow +\infty \), for \(e=-\nu (z)\), the inward normal at \(z\in \partial \Omega \). By the properties in Lemma A.3 and Lemma A.4, the functions \(v_m\) satisfy
since \(\mathcal {L}_\Omega u\) and \(\mathcal {L}_\Omega d=\mathcal {L}_{\Omega }\delta \) are bounded in view of Lemma A.2. Now, by Proposition 5.4, \(\left\| {v_m}\right\| _{C^{2\beta }(\Omega _m)}\le C\) for some \(\beta >0\). So Arzelà–Ascoli Theorem asserts a subsequence of \(v_m\) uniformly converging on compact sets in \(\left\{ x\cdot e>0\right\} \) to some function \(v\in C^\beta (\left\{ x\cdot e>0\right\} )\), \(\beta \in (0,1)\). Passing to the limit in (7.5) and (7.6) yields
and
Step 4: Classification of the limit, and the contradiction.
By Proposition 6.1, \(v(x)=c_0(x\cdot e)\) for some constant \(c_0\in \mathbb {R}\). Using the fact that
we pass to the limit in (7.4) (upon dividing by \(r_m\)) to see that
But this implies \(c_0=0\) and hence \(v=0\), contradicting (7.3) in the limit \(m\rightarrow +\infty \). Therefore (7.1) holds and the proof is complete. \(\square \)
8 Existence of viscosity solution
Consider the Dirichlet problem
We will establish the existence of a continuous viscosity solution using Perron’s method, carefully exploiting the mid-range maximum principle that \(\mathcal {L}_{\Omega }\) satisfies. Throughout the section we assume \(f\in C^{\alpha }(\Omega )\) and \(g\in C(\partial \Omega )\), for some \(\alpha >0\).
Our goal is to prove the following.
Proposition 1.32
Let \(\Omega \subset \mathbb {R}^n\) be a bounded domain of class \(C^{1,1}\). Let \(s\in (0,1)\) and \(f\in C^\alpha \) for some \(\alpha >0\). For any \(f\in C^\alpha (\Omega )\) and \(g\in C(\partial \Omega )\), There exists a unique \(u\in C(\overline{\Omega })\) satisfying (8.1) in the viscosity sense. Moreover, \(u\in C^{2\,s+\alpha }(\Omega )\cap C(\overline{\Omega })\) is a classical solution to (8.1).
The notion of viscosity solutions has been successfully used in nonlocal equations, see for example [1, 7, 11, 13]. For the proof, we extend the clean arguments described in [4] using the barrier constructed in Sect. 3.
Definition 1.33
(Semi-continuous functions) We denote by
the space of upper (resp. lower) semi-continuous functions in \(\overline{\Omega }\). For \(u\in L^\infty (\overline{\Omega })\) we define the USC (resp. LSC) envelope as
We also need a localized definition based on Proposition 2.2.
Definition 1.34
(Viscosity solutions) Let G be an non-empty, open set in \(\Omega \). The domain of interaction of G is
(Thus, \(\mathcal {L}_{\Omega }=\mathcal {L}_{G_*}\) in G.) Let \(f\in C(\Omega )\). We say that \(u\in USC(\overline{G_*})\) (resp. \(u\in LSC(\overline{G_*})\)) is a (mid-range) viscosity sub-solution of
if for any \(x\in \Omega \), any neighborhood \(N_x\) of x in \(\Omega \) and any \(\varphi \in C^2(N_x)\cap L^1(\overline{G_*})\) with
we have
In particular, when \(G=\Omega \), \(\overline{G_*}=\overline{\Omega }\). We say that \(u\in C(\overline{\Omega })\) is a (global) viscosity solution in \(\Omega \) if it is both a sub-solution and a super-solution in \(\Omega \).
Remark 8.4
Global sub-(resp. super-) solutions are necessarily mid-range sub-(resp. super-) solutions (but not vice versa). This is because the test function \(\varphi \in L^1(\overline{\Omega })\) can be extended to keep the sign of \(u-\varphi \) without affecting the computation of \(\mathcal {L}_{\Omega }\varphi \) at the contact point.
8.1 Comparison principle for viscosity solutions
We generalize Proposition 2.2 to viscosity solutions.
Lemma 1.36
(Mid-range maximum principle) Let \(G\subseteq \Omega \) be open. Suppose \(u\in LSC(\overline{G_*})\) solves, in the viscosity sense,
Then \(u\ge 0\) in G.
Proof
If not, \(\min _{\overline{\Omega }}u=-\delta \) for some \(\delta >0\). Using a translated coordinate system if necessary, we assume that \(0\in G\). The convex paraboloid
takes values in \([-\delta /2,-\delta /4]\) and so (by moving down then up) there exists \(c>0\) such that
touches u from below at some \(x_0\in G\), i.e. \(u(x_0)=\varphi (x_0)\) and \(u\ge \varphi \) in G. By construction \(u\ge 0\ge {{\tilde{\varphi }}}\ge \varphi \) in \(\overline{G_*}{\setminus } G\). On the one hand, by Definition 8.3,
On the other hand,
a contradiction. \(\square \)
Corollary 1.37
(Mid-range comparison principle) Let \(G\subseteq \Omega \) be open. Suppose \(u\in USC(\overline{G_*})\), \(v\in LSC(\overline{G_*})\) are respectively super- and sub-solutions to (8.1), i.e.
in the viscosity sense, then
8.2 Supremum of sub-solutions
Define the family of admissible sub-solutions as
The pointwise supremum of all sub-solutions in \(\mathcal {A}\) is defined as
We will prove that u is a viscosity solution by showing that \(u^*=g\) on \(\partial \Omega \) so that \(u^*=u\), and then verify that \(u_*\) is a super-solution so that, by comparison, \(u_*=u\).
Proposition 1.38
(Perron’s method) The function u defined in (8.2) lies in \(C(\overline{\Omega })\) and is a viscosity solution to (8.1).
Lemma 1.39
The USC envelope of u defined by (8.2) is a sub-solution in the interior, i.e.
As a result, \(\sup _{\overline{\Omega }}u^*\le C\), for \(C>0\) depending only on n, s, \(\left\| {f}\right\| _{L^\infty (\Omega )}\) and \(\Omega \).
Proof
The proof is the same as [4, Lemma 4.15], except that the test function \(\phi \in C^2\) is chosen such that \(u-\phi \) attains its global maximum in \(\overline{\Omega }\). \(\square \)
Lemma 1.40
The USC envelope of u defined by (8.2) satisfies the boundary condition, i.e.
Proof
The proof is similar to [4, Proof of Theorem 4.17, Step 1], but a mid-range comparison is to be employed. Indeed, for each \(x_0\in \partial \Omega \), let \(r_0\) be as in Proposition 3.1 and define the barrier
where \(k_\varepsilon \), depending not only on \(\varepsilon \) but also on g, \(\Omega \) and \(\sup _{\overline{\Omega }}{u^*}\), is chosen such that
By Proposition 3.1,
By Corollary 8.6,
In particular, since \(\varphi ^{(x_0)} \in C(\overline{\Omega })\), there exists \(\delta (\varepsilon )>0\) such that
This implies \(\lim _{x_k\rightarrow x_0} u^*(x_k)=g(x_0)\) and hence \(u^*|_{\partial \Omega }=g \in C(\partial \Omega )\). \(\square \)
Lemma 1.41
It holds that \(u=u^* \in USC(\overline{\Omega })\).
Proof
By definition, \(u\le u^*\). By Lemmas 8.8 and 8.9, \(u^*\in \mathcal {A}\), so \(u^* \le u\). \(\square \)
Lemma 1.42
Let G be an open subset of \(\Omega \). Suppose \(u\in USC(\overline{\Omega })\) satisfies
in the viscosity sense, and \(v\in C^2(G) \cap L^\infty (\overline{\Omega })\) satisfies pointwise
Then, the maximum \(w=u\vee v\) is also a sub-solution in \(\Omega \).
Proof
Suppose \(x\in N_x\subset \Omega \) and \(\phi \in C^2(N_x)\cap L^1(\Omega )\) is such that \(w(x)=\phi (x)\) and \(w\le \phi \) in \(\overline{\Omega }\). We want to show that \(\mathcal {L}_{\Omega }\phi (x) \le f(x)\). If \(w(x)=u(x)\), then since \(u\le w\le \phi \), the result follows from the fact that u is a viscosity sub-solution. If \(w(x)=v(x) \ne u(x)\), then \(x\in G\) and (using \(v\le w\le \phi \)) the pointwise computation also gives \(\mathcal {L}_{\Omega }\phi (x) \le \mathcal {L}_{\Omega }v \le f\), as desired. \(\square \)
Lemma 1.43
The LSC envelope of u defined by (8.2) is a super-solution in the interior, i.e.
Proof
If \(u_*\) is not a super-solution in \(\Omega \), then there exists \(x\in N_x\subset \Omega \), \(\varphi \in C^2(N_x)\cap L^1(\overline{\Omega })\), such that \(u(x)=\varphi (x)\), \(u_*\ge \varphi \) in \(\overline{\Omega }\), while \(\mathcal {L}_{\Omega }\varphi (x)<f(x)\). By replacing \(\varphi \) by \({{\tilde{\varphi }}}=\varphi -\varepsilon |\cdot -x|^2\) if necessary (where \(\varepsilon \) depends on \(\varphi \) and f), we can assume that \(u_*>\varphi \) in \(\overline{\Omega }\setminus \left\{ x\right\} \). By the continuity of \(\mathcal {L}_{\Omega }\varphi \) and f at x, there exist \(\delta ,\rho >0\) such that
Now, define \(u_\delta =u\vee (\varphi +\delta )\), which is a sub-solution in \(\Omega \) due to Lemma 8.11. Now \(u_\delta \in \mathcal {A}\) and so \(u_\delta \le u\). But this means that \(\varphi +\delta \le u\) in all of \(\overline{\Omega }\) including \(x_0\), a contradiction.
Suppose, on the contrary, that \(u_*\) is not a super-solution in \(\Omega \). Then there exists \(x_0\in \Omega \) and \(\varphi \in C(\overline{\Omega }) \cap C^2(B_{d_{\Omega }(x_0)}(x_0))\) such that \(u_*(x_0)=\varphi (x_0)\), \(u_* \ge \varphi \) in \(\overline{\Omega }\), while \(\mathcal {L}_{\Omega }\varphi (x_0)<f(x_0)\). By replacing \(\varphi \) by \({\tilde{\varphi }}=\varphi -\varepsilon |x-x_0|^2\) if necessary, where \(\varepsilon \) depends on \(\varphi \) and f, we can assume that \(u_*>\varphi \) in \(\overline{\Omega } \setminus \left\{ x_0\right\} \). By continuity of \(\varphi \), f and \(\mathcal {L}_{\Omega }\varphi \) at \(x_0\), there exists \(\delta ,\rho >0\) such that
Now, define \(u_\delta =u \vee (\varphi +\delta )\), which is a sub-solution in \(\Omega \) due to Lemma 8.11. Now \(u_\delta \in \mathcal {A}\) and so \(u_\delta \le u\). But this means that \(\varphi +\delta \le u\) in all of \(\overline{\Omega }\) including \(x_0\), a contradiction. \(\square \)
Lemma 1.44
It holds that \(u=u_*\in LSC(\overline{\Omega })\).
Proof
By definition \(u_*\le u\). In view of Lemmas 8.9 and 8.12, comparing \(u_*\) to u via Corollary 8.6 gives \(u\le u_*\). \(\square \)
Proof of Proposition 8.7
By Lemmas 8.9, 8.10 and 8.13, \(u\in C(\overline{\Omega })\) is both a sub- and super-solution, and \(u=g\) on \(\partial \Omega \). \(\square \)
8.3 Regularity
Since it suffices to obtain qualitative interior regularity, we compare to the (restricted) fractional Laplacian in \(\mathbb {R}^n\) as in Lemma 2.4, and invoke the ccorresponding regualrity results in [3, Chapter 3].
Lemma 1.45
Let \(u \in C(\overline{\Omega })\) be as in Proposition 8.7, with \(f\in C^\alpha (\Omega )\). Then \(u\in C^{2s+\alpha }(\Omega )\).
Proof
We verify that u, when extended continuously to a bounded function with compact support outside \(\Omega \), is a viscosity solution to
where u is and
Recalling the definition of viscosity solution in [3, Chapter 3], suppose \(x\in N_x\subset \Omega \) and \(\phi \in L^1_{2\,s}(\mathbb {R}^n) \cap C^2(N_x)\) is such that
In particular, \(\phi \in L^1(\Omega )\) and \(u\le \phi \) in \(\overline{\Omega }\). By Definition 8.3, \(\mathcal {L}_{\Omega }\phi (x)\le f(x)\). This pointwise inequality rearranges to \((-\Delta )^s\phi \le F[\phi ](x)\). Hence, u is a viscosity sub-solution to (8.3)–(8.4). Similarly, u is also a viscosity super-solution. By bootstrapping the regularity result in [3, Chapter 3] (recall that F[u] is as regular as u, as in the proof of Lemma 2.4), \(u\in C^{2s+\alpha }(\Omega )\). \(\square \)
Proof of Proposition 8.1
By Proposition 8.7, there exists a viscosity solution \(u\in C(\overline{\Omega })\). By Lemma 8.14, \(u\in C^{2\,s+\alpha }(\Omega )\), so it is also a classical solution. \(\square \)
Proof of Theorem 1.2
It follows immediately from Proposition 8.1 and Theorem 1.1. \(\square \)
Data availability statement
Data sharing not applicable to this article as no datasets were generated or analysed during the current study.
References
Caffarelli, L., Silvestre, L.: Regularity theory for fully nonlinear integro-differential equations. Commun. Pure Appl. Math. 62(5), 597–638 (2009)
Di Castro, A., Kuusi, T., Palatucci, G.: Nonlocal Harnack inequalities. J. Funct. Anal. 267(6), 1807–1836 (2014)
Fernández-Real, X., Ros-Oton, X.: Integro-differential elliptic equations (To appear)
Fernández-Real, X., Ros-Oton, X.: Regularity Theory for Elliptic PDE. Zurich Lectures in Advanced Mathematics. EMS books (2022)
Gilbarg, D., Trudinger N.S.: Elliptic partial differential equations of second order. Reprint of the 1998 edition. Classics in Mathematics. Springer, Berlin. xiv+517 pp. ISBN: 3-540-41160-7 35-02 (2001)
José, C., Bellido, A., Ortega.: A restricted nonlocal operator bridging together the Laplacian and the Fractional Laplacian. arXiv:2004.12160 (2020)
Korvenpää, J., Kuusi, T., Palatucci, G.: Fractional superharmonic functions and the Perron method for nonlinear integro-differential equations. Math. Ann. 369(3–4), 1443–1489 (2017)
Ros-Oton, X., Serra, J.: The Pohozaev identity for the fractional Laplacian. Arch. Rat. Mech. Anal. 213, 587–628 (2014)
Ros-Oton, X., Serra, J.: The Dirichlet problem for the fractional Laplacian: regularity up to the boundary. J. Math. Pures Appl. (9) 101(3), 275–302 (2014)
Ros-Oton, X., Serra, J.: Regularity theory for general stable operators. J. Differ. Equ. 260(12), 8675–8715 (2016)
Ros-Oton, X., Serra, J.: Boundary regularity for fully nonlinear integro-differential equations. Duke Math. J. 165(11), 2079–2154 (2016)
Ros-Oton, X., Serra, J.: The boundary Harnack principle for nonlocal elliptic operators in non-divergence form. Potential Anal. 51(3), 315–331 (2019)
Servadei, R., Valdinoci, E.: Weak and viscosity solutions to the fractional Laplace equation. Publ. Mat. 58(1), 133–154 (2014)
Acknowledgements
The author has received funding from the Swiss National Science Foundation under the Grant PZ00P2_202012/1. He thanks Xavier Fernández-Real and Xavier Ros-Oton for simulating discussions. as well as the anonymous referee for useful comments.
Funding
Open access funding provided by University of Basel
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by X. Ros-Oton.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Basic properties of \(\mathcal {L}_\Omega \)
We show that \(\mathcal {L}_\Omega \) reduces to the classical Laplace operator at the boundary, with the choice of the normalization \(C_{n,s}d(x)^{2-2s}\).
Lemma A.1
(Limit operator) If \(u\in C^{2,\beta }(\Omega )\) for some \(\beta >0\), then
Proof
We compute
as \(d(x)\rightarrow 0\). \(\square \)
A nice bound is available for \(\mathcal {L}_\Omega \) on \(C^{1,1}\) functions. We will need it only for \(\delta \), a smooth function that agrees with \(d_\Omega \) near the boundary.
Lemma A.2
We have
In particular, \(\mathcal {L}_\Omega \delta \) is universally bounded.
Proof
Since \(\delta \in C^{1,1}(\overline{\Omega })\), by a Taylor expansion with quadratic error and (1.1),
\(\square \)
We collect the effect of translation and scaling on \(L_\Omega \), since the operator depends heavily on the domain. When various domains are in consideration, we put the domain as a subscript.
Let \(z\in \mathbb {R}^n\). For \(u:\Omega \rightarrow \mathbb {R}\), define \(u(\cdot ;z):\Omega -z\rightarrow \mathbb {R}\) by
Lemma A.3
(Translation) Let \(u\in C^{2s+}(\Omega )\). For any \(z\in \mathbb {R}^n\),
Proof
Since
we have
\(\square \)
Let \(r>0\). For \(u:\Omega \rightarrow \mathbb {R}\), consider the rescaling \(u_r:r^{-1}\Omega \rightarrow \mathbb {R}\) given by
Lemma A.4
(Scaling) Let \(u\in C^{2s+}(\Omega )\). For any \(r>0\),
Proof
Note that
Therefore
\(\square \)
Appendix B: An auxiliary function
Let
Note that, by (1.1), \(\psi (2,t)=2\) for all \(t>0\).
Lemma B.1
For \(p>0\) and \(t\in [0,1]\),
Consequently,
Here the constant C depends only on n, s and p and it remains bounded as \(p\rightarrow 2\).
Proof
It suffices to bound
When \(t\in (0,1/2)\) or \(|y_n|<1/2\), we have \(|ty_n|<1/2\) and so Taylor expansion gives
When \(t\in [1/2,1]\) and \(|y_n|\ge 1/2\), we have also \(|y|\ge 1/2\) so the integrand is bounded (by the boundedness of the function \(x\mapsto x^p\log x\) on [0, 2]). so
\(\square \)
Lemma B.2
There exists \(c>0\) depending only on n and s such that as \(q\rightarrow 0^+\),
uniformly in \(t\in [0,1]\).
Proof
Using the Taylor expansion
we have
where the error \(O(q^2)\) is bounded independently of \(t\in [0,1]\). To see that the coefficient of q remains strictly positive as \(t\rightarrow 0\), we observe that for \(|y|<1/2\), \(\log (1-y^2) \le Cy^2\) and the homogeneity similar to (1.1). \(\square \)
Appendix C: Boundary expansions
Lemma C.1
Suppose \(\Omega \) is \(C^{1,1}\), \(0\in \partial \Omega \) and \(u\in C^{1,\gamma }(\Omega )\). Let \(\nu (x)\) denotes the inward normal of the parallel surface containing x, and assume \(\nu (0)=e_n\). Suppose
Then
Proof
Represent \(\partial \Omega \) by a graph \(x_n=\Phi (x')\) near \(x=(x',x_n)=0\), then \(\Phi (0)=|\nabla \Phi (0)|=0\) and
Suppose (C.1) holds. Then
Clearly \(d(x)=O(|x|)\) as \(x\rightarrow 0\). Thus (C.2) holds. \(\square \)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.