This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Optimal Regularity for Fully Nonlinear Nonlocal Equations with Unbounded Source Terms

Disson S. dos Prazeres and Makson S. Santos
Abstract.

We prove optimal regularity estimates for viscosity solutions to a class of fully nonlinear nonlocal equations with unbounded source terms. More precisely, depending on the integrability of the source term fLp(B1)f\in L^{p}(B_{1}), we establish that solutions belong to classes ranging from Cσd/pC^{\sigma-d/p} to CσC^{\sigma}, at critical thresholds. We use approximation techniques and Liouville-type arguments. These results represent a novel contribution, providing the first such estimates in the context of not necessarily concave nonlocal equations.

Keywords: Nonlocal operators, Unbounded source, Hölder regularity.

AMS Subject Classifications: 35B65, 35D40, 35R11.

1. Introduction

In this article, we examine the regularity theory of viscosity solutions to a fully nonlinear nonlocal equation of the form

(1.1) σ(u,x)=C(σ)supα𝒜infβdδ(u,x,y)yTAα,β(x)y|y|σ+d+2𝑑y=f(x) in B1,{\mathcal{I}}_{\sigma}(u,x)=C(\sigma)\sup_{\alpha\in{\mathcal{A}}}\inf_{\beta\in{\mathcal{B}}}\int_{\mathbb{R}^{d}}\delta(u,x,y)\dfrac{y^{T}A_{\alpha,\beta}(x)y}{|y|^{\sigma+d+2}}dy=f(x)\;\;\mbox{ in }\;B_{1},

where σ(0,2)\sigma\in(0,2), C(σ)>0C(\sigma)>0 is a normalizing constant, δ(u,x,y):=u(x+y)+u(xy)2u(x)\delta(u,x,y):=u(x+y)+u(x-y)-2u(x), 𝒜{\mathcal{A}} and {\mathcal{B}} are indexes sets, and the source term ff belongs to a suitable Lesbegue space. Moreover, we suppose the following ellipticity-type condition: for each α𝒜\alpha\in{\mathcal{A}} and β\beta\in{\mathcal{B}}, Aα,βA_{\alpha,\beta} is a symmetric d×dd\times d matrix (λ,Λ)(\lambda,\Lambda)-elliptic, i.e.,

(1.2) λIAΛI in B1,\lambda I\leq A\leq\Lambda I\quad\mbox{ in }\,\,B_{1},

for 0<λΛ0<\lambda\leq\Lambda. We establish Hölder regularity for solutions (or their gradient), depending on the range of pp for which fLp(B1)f\in L^{p}(B_{1}).

Regularity results for nonlocal operators have been extensively studied over the years. In [15], the author presents an analytical proof of Hölder continuity and introduces more flexible assumptions on the operator than previous studies. But it is only in [6], where such regularity estimates were given uniformly as the degree of the operator σ2\sigma\to 2, seen as a natural extension of second-order operators. In that paper, the authors also prove properties such as comparison principle, a nonlocal Alexandrov-Bakelman-Pucci estimate (ABP for short), Harnack inequality, and regularity in C1,αC^{1,\alpha} spaces for viscosity solutions to equations of type (1.1).

Numerous results were uncovered based on the findings in [6]. In [8], the authors extended their previous results using perturbative methods, including the C1,αC^{1,\alpha}-regularity for a class of non translation-invariant equations. Around the same time, the authors in [3] also worked with non translation-invariant equations, establishing Hölder regularity of solutions, using different assumptions from [8], allowing first-order terms and some degeneracy in the operators. We also mention the work [12], where the author gives C1,αC^{1,\alpha}-estimates for more general kernels than previous results.

Besides the C1,αC^{1,\alpha} estimates, Evans-Krylov-type results have been explored by the community. In this direction, the authors in [7] proved that viscosity solutions for concave (or convex) equations of the form

(1.3) σ(u,x):=infα𝒜dδ(u,x,y)Kα(y)𝑑y=0 in B1,{\mathcal{I}}_{\sigma}(u,x):=\inf_{\alpha\in{\mathcal{A}}}\int_{\mathbb{R}^{d}}\delta(u,x,y)K_{\alpha}(y)dy=0\;\;\mbox{ in }\;B_{1},

are of class Cσ+βC^{\sigma+\beta}. They assume that, for each α𝒜\alpha\in{\mathcal{A}}, the kernel KαK_{\alpha} belongs to the class 2{\mathcal{L}}_{2}, i.e., they are of class C2C^{2} away from the origin, symmetric, satisfies the ellipticity condition

(1.4) (2σ)λ|y|d+σKα(y)(2σ)Λ|y|d+σ,(2-\sigma)\dfrac{\lambda}{|y|^{d+\sigma}}\leq K_{\alpha}(y)\leq(2-\sigma)\dfrac{\Lambda}{|y|^{d+\sigma}},

and

(1.5) D2Kα(y)C|y|d+2+σ.D^{2}K_{\alpha}(y)\leq\dfrac{C}{|y|^{d+2+\sigma}}.

In [13], the author extended the result above to equations of type (1.3) with rough kernels, i.e., where the kernels KαK_{\alpha} satisfy (1.4) but not necessarily (1.5), for every α𝒜\alpha\in{\mathcal{A}}. Under the additional (and optimal) assumption of CαC^{\alpha} exterior data for the solution, the author provides Cσ+αC^{\sigma+\alpha} a priori estimates. Moreover, results involving non translation-invariant kernels are also given, provided the dependency on the variable xx is of a CαC^{\alpha}-fashion. We also mention the work [18], where the author also provides Cσ+αC^{\sigma+\alpha} for viscosity solutions to nonlocal equations of the type

F(Dσu)=0 in B1F(D^{\sigma}u)=0\;\;\mbox{ in }\;B_{1}

provided the quantities uL(B1)\|u\|_{L^{\infty}(B_{1})} and uLσ1(d)\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})} are sufficiently small.

A natural question is whether one can prove optimal regularity estimates for fully nonlinear nonlocal equations with an unbounded right-hand side. However, such results are relatively scarce in the literature. One of the primary challenges in this context is the lack of compactness properties for viscosity solutions of (1.1) (or even the simpler equation (1.7) below) when fLp(B1)f\in L^{p}(B_{1}). In addition, stability results are also crucial for implementing the now-classic strategy proposed in [8]. In fact, only a few works involving an unbounded norm on the right-hand side are available. Moreover, the existing results consider a class of kernels that differ from the broader 0{\mathcal{L}}_{0} class discussed in [6, 8, 12, 13], which defines the extremal operator

0(u,x)=(2σ)infλa(x,y)Λdδ(u,x,y)a(x,y)|y|d+σ+2.{\mathcal{M}}^{-}_{{\mathcal{L}}_{0}}(u,x)=(2-\sigma)\inf_{\lambda\leq a(x,y)\leq\Lambda}\int_{\mathbb{R}^{d}}\delta(u,x,y)\dfrac{a(x,y)}{|y|^{d+\sigma+2}}.

Instead, they work with kernels of the form

(1.6) Kα(y)=yTAαy|y|d+σ+2dy,K_{\alpha}(y)=\dfrac{y^{T}A_{\alpha}y}{|y|^{d+\sigma+2}}dy,

where AαA_{\alpha} is a symmetric d×dd\times d matrix satisfying

(1.7) λI1d+σ(σAα+Tr(Aα)I)ΛI in B1,\lambda I\leq\dfrac{1}{d+\sigma}\left(\sigma A_{\alpha}+{\rm Tr}(A_{\alpha})I\right)\leq\Lambda I\quad\mbox{ in }\,\,B_{1},

for positive constants 0<λΛ0<\lambda\leq\Lambda. These kernels define a class of extremal operators as

(u,x)=infαdδ(u,x,y)Kα(y)𝑑y.{\mathcal{M}}^{-}(u,x)=\inf_{\alpha}\int_{\mathbb{R}^{d}}\delta(u,x,y)K_{\alpha}(y)dy.

It is important to note that, because they allow some degeneracy (conform (1.7)), although this is a smaller class, it is not necessarily contained within the 0{\mathcal{M}}^{-}_{{\mathcal{L}}_{0}}.

In this direction, the authors in [9] established a quantitative ABP estimate for viscosity supersolutions to

(1.8) {(u,x)f(x) in B1u(x)0 on dB1.\left\{\begin{aligned} {\mathcal{M}}^{-}(u,x)&\leq f(x)\;\;\mbox{ in }\;B_{1}\\ u(x)&\geq 0\;\;\mbox{ on }\;\mathbb{R}^{d}\setminus B_{1}.\end{aligned}\right.

They show that supersolutions to the equation above satisfy

infB1uC(d,λ)(f+L(Ku))(2σ)/2(f+Ld(Ku))σ/2,-\inf_{B_{1}}u\leq C(d,\lambda)(\|f^{+}\|_{L^{\infty}(K_{u})})^{(2-\sigma)/2}(\|f^{+}\|_{L^{d}(K_{u})})^{\sigma/2},

where KuK_{u} is the coincidence set between uu and a type of fractional convex envelope of uu. In [10], the author improved the previous result, by removing the dependency of the LL^{\infty}-norm in the estimate above, in case σ\sigma is sufficiently close to 2. In [11], the author removed the restriction on the degree of the operator present in [10], by proving that supersolutions to (1.8) with fLp(B1)f\in L^{p}(B_{1}), for p(dε0,)p\in(d-\varepsilon_{0},\infty), satisfy

infB1uC(d,σ,λ,Λ)f+Lp(B1).-\inf_{B_{1}}u\leq C(d,\sigma,\lambda,\Lambda)\|f^{+}\|_{L^{p}(B_{1})}.

Moreover, the author also gives Wσ,pW^{\sigma,p}-estimates for viscosity solutions to concave equations of the form

(1.9) σ(u,x)=C(σ)infα𝒜dδ(u,x,y)yTAα(x)y|y|σ+d+2𝑑y=f(x) in B1,{\mathcal{I}}_{\sigma}(u,x)=C(\sigma)\inf_{\alpha\in{\mathcal{A}}}\int_{\mathbb{R}^{d}}\delta(u,x,y)\dfrac{y^{T}A_{\alpha}(x)y}{|y|^{\sigma+d+2}}dy=f(x)\;\;\mbox{ in }\;B_{1},

which, in particular, implies that solutions are of class Cσd/pC^{\sigma-d/p}. To our knowledge, this is the only result in the literature that gives this type of regularity for fully nonlinear nonlocal equations as in (1.9), in the presence of an unbounded right-hand side.

The main purpose of this paper is to establish optimal regularity estimates for viscosity solutions to equation (1.1), where the source term fLp(B1)f\in L^{p}(B_{1}). We emphasize that such estimates have not been previously available for this type of equation. We employ the so-called half-relaxed method, originally introduced in [4], to prove that a sequence (uk)k(u_{k})_{k\in\mathbb{N}} of merely bounded viscosity solutions to (1.1) converges uniformly to a function uu_{\infty} which solves a suitable equation, see Lemma 2.11 below. This method relies on the comparison principle of the operator and has been used previously in the context of Hamilton-Jacobi equations and nonlocal equations with Neumann boundary conditions in a half-space. We refer the reader to [1, 2]. Once compactness and stability are available, standard approximation arguments can be applied to show the regularity properties of the viscosity solutions.

As discussed earlier, the regularity of solutions depends on the range of pp for which fLp(B1)f\in L^{p}(B_{1}), in the spirit of [16]. First, we prove that if p(dε0,dσ1)p\in(d-\varepsilon_{0},\frac{d}{\sigma-1}), then solutions are of class Cσd/pC^{\sigma-d/p}. The constant ε0\varepsilon_{0} is known as the Escauriaza’s exponent. Notice that this type of regularity was previously known from [11] in the context of a concave equation as in (1.9).

The borderline case fLd/(σ1)f\in L^{d/(\sigma-1)} is particularly significant. In the local case, i.e., σ=2\sigma=2, this value separates continuity estimates from differentiability properties in the regularity theory. Moreover, this quantity also appears in ABP and Harnack estimates. Meanwhile, in the nonlocal case, this is the first time such a threshold has been explicitly considered, as previous ABP estimates in [9, 10, 11] only consider the LdL^{d}-norm of the right-hand side. We believe this is the correct threshold for future general ABP estimates for the general class 0{\mathcal{L}}_{0}. Nevertheless, in this scenario, we show that viscosity solutions to (1.1) are Log-Lipschitz, which is better than CαC^{\alpha} for every α(0,1)\alpha\in(0,1).

For fLpf\in L^{p} for p(dσ1,+)p\in(\frac{d}{\sigma-1},+\infty), we prove that the solutions belong to the class C1,αC^{1,\alpha}, where α\alpha is defined in (2.2). Finally, for the borderline case where fBMO(B1)f\in\rm{BMO}(B_{1}), we show that viscosity solutions to (1.9) are locally of class CσC^{\sigma}, which is the best regularity we can hope for without assuming further regularity for ff. These results are detailed in Theorems 2.6, 2.7, 2.8 and 2.9 below.

The class of kernels in (1.1) follows the form of (1.6), as described in [9, 10, 11], but with the additional requirement of uniform ellipticity, as in (1.2). In particular, the class of kernels that we deal with is a subset of 0{\mathcal{L}}_{0} and enjoys all its properties. Furthermore, because uniform ellipticity as in (1.2) ensures that the condition (1.7) is satisfied with the same ellipticity constants, our class of kernels is also included in those described in [9, 10, 11]. Developing a similar theory for more general kernels, such as those in 0{\mathcal{L}}_{0}, remains an open challenge, primarily due to the need for an appropriate ABP estimate, as discussed earlier.

The remainder of this paper is structured as follows: In section 2 we gather some auxiliary results and present our main results. The proof of the optimal Hölder regularity is the subject of Section 3. In Section 4 we put forward the Log-Lipschitz regularity of solutions. Section 5 is devoted to the proof of Hölder regularity for the gradient of solutions. Finally, in the last section, we investigate the borderline CσC^{\sigma}-regularity.

2. Preliminaries

2.1. Notations and definitions

This section collects some definitions and notations used throughout the paper. The open ball of radius rr and centered at x0x_{0} in d\mathbb{R}^{d} is denoted by Br(x0)B_{r}(x_{0}). For α(0,1]\alpha\in(0,1], the notation uCα(B1)u\in C^{\alpha^{-}}(B_{1}) means that uCβ(B1)u\in C^{\beta}(B_{1}), for every β<α\beta<\alpha. We proceed by defining the Log-Lipschitz space.

Definition 2.1.

A function uu belongs to CLogLip(Br)C^{\rm Log-Lip}(B_{r}) if there exists a universal constant C>0C>0 such that

supBr/2(x0)|u(x)u(x0)|rlnr1.\sup_{B_{r/2}(x_{0})}|u(x)-u(x_{0})|\leq r\ln{r^{-1}}.
Definition 2.2 (BMO space).

We say that fBMO(B1)f\in{\rm BMO}(B_{1}) if for all Br(x0)B1B_{r}(x_{0})\subset B_{1}, we have

fBMO(B1):=sup0<r1Br(x0)|f(x)fx0,r|𝑑x,\|f\|_{{\rm BMO}(B_{1})}:=\sup_{0<r\leq 1}\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int_{B_{r}(x_{0})}|f(x)-\langle f\rangle_{x_{0},r}|dx,

where fx0,r:=Br(x0)f(x)𝑑x\langle f\rangle_{x_{0},r}:=\displaystyle\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int_{B_{r}(x_{0})}f(x)dx.

Since uu is defined in the whole d\mathbb{R}^{d}, it can behave very widely as |x||x|\to\infty. Hence, we work within a class where we have a certain decay of the solutions as they approach infinity, see also [6].

Definition 2.3 (Growth at infinity).

We say that a function u:du:\mathbb{R}^{d}\rightarrow\mathbb{R} belongs to Lσ1(d)L^{1}_{\sigma}(\mathbb{R}^{d}), if

uLσ1(d):=d|u(x)|11+|x|d+σ𝑑x<+.\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}:=\int_{\mathbb{R}^{d}}|u(x)|\dfrac{1}{1+|x|^{d+\sigma}}dx<+\infty.

In the next, we define viscosity solutions:

Definition 2.4 (Viscosity solution).

We say that an upper (lower) semicontinuous function uLσ1(d)u\in L_{\sigma}^{1}(\mathbb{R}^{d}) is a viscosity subsolution (supersolution) to (1.1), if for any x0B1x_{0}\in B_{1} and φC2(Br(x))\varphi\in C^{2}(B_{r}(x)) such that uφu-\varphi has a local maximum (minimum) at x0x_{0}, then the function

v(x):={φ(x) in Br(x0)u(x) in dBr(x0),v(x):=\left\{\begin{aligned} \varphi(x)&\;\;\mbox{ in }\;B_{r}(x_{0})\\ u(x)&\;\;\mbox{ in }\;\mathbb{R}^{d}\setminus B_{r}(x_{0}),\end{aligned}\right.

satisfies

σ(v,x)()f(x) in B1-{\mathcal{I}}_{\sigma}(v,x)\leq(\geq)f(x)\;\;\mbox{ in }B_{1}

We say a function uC(B¯1)Lσ1(d)u\in C(\overline{B}_{1})\cap L_{\sigma}^{1}(\mathbb{R}^{d}) is a viscosity solution to (1.1) if it is simultaneously a viscosity subsolution and supersolution.

Remark 2.5 (Scaling properties).

Throughout the manuscript, we assume certain smallness conditions on the norms of uu and the source term ff. We want to stress that such conditions are not restrictive. In fact, if uu is a viscosity solution to (1.1), then for ε>0\varepsilon>0 the function

v(x):=u(x)uL(B1)+ε1fLp(B1),v(x):=\dfrac{u(x)}{\|u\|_{L^{\infty}(B_{1})}+\varepsilon^{-1}\|f\|_{L^{p}(B_{1})}},

satisfy vL(B1)1\|v\|_{L^{\infty}(B_{1})}\leq 1 and solves

σ(v,x)=f~(x) in B1,{\mathcal{I}}_{\sigma}(v,x)=\tilde{f}(x)\;\;\mbox{ in }\;B_{1},

where

f~(x):=f(x)uL(B1)+ε1fLp(B1),\tilde{f}(x):=\dfrac{f(x)}{\|u\|_{L^{\infty}(B_{1})}+\varepsilon^{-1}\|f\|_{L^{p}(B_{1})}},

is such that f~L(B1)ε\|\tilde{f}\|_{L^{\infty}(B_{1})}\leq\varepsilon.

2.2. Main results

As mentioned earlier, the regularity of solutions depends on the range of pp under consideration. The critical cases occur when fLdσ1(B1)f\in L^{\frac{d}{\sigma-1}}(B_{1}) for CαC^{\alpha}-regularity and fBMOf\in\mathrm{BMO} for CσC^{\sigma}-regularity. Due to the nonlocal nature of the problem, achieving these critical cases requires ff to possess higher regularity compared to the local case. For instance, in the local case, the first critical threshold is at p=dp=d, leading to ClocLog-LipC^{\text{Log-Lip}}_{\text{loc}}-regularity. Similarly, when fBMOf\in\mathrm{BMO} in the local case, it yields Cloc1,Log-LipC^{1,\text{Log-Lip}}_{\text{loc}}-regularity.

We now present the main results of this article, beginning with a result in Hölder spaces for the case where pp is below dσ1\frac{d}{\sigma-1}.

Theorem 2.6.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a viscosity solution to (1.1), with fLp(B1)f\in L^{p}(B_{1}), p(dε0,dσ1)p\in\left(d-\varepsilon_{0},\frac{d}{\sigma-1}\right). Then, uClocα(B1)u\in C^{\alpha}_{loc}(B_{1}) for any

(2.1) α(0,σpdp].\alpha\in\left(0,\frac{\sigma p-d}{p}\right].

Moreover, there exists a positive constant C=C(p,d,σ,λ,Λ)C=C(p,d,\sigma,\lambda,\Lambda), such that

uCα(B1/2)C(uL(B1)+uLσ1(d)+fLp(B1)).\|u\|_{C^{\alpha}(B_{1/2})}\leq C(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{L^{p}(B_{1})}).

We observe that for σ=2\sigma=2, we have p(dε0,d)p\in(d-\varepsilon_{0},d) and α(0,2pdp)\alpha\in(0,\frac{2p-d}{p}), recovering the regularity result for the local case reported in [16]. Next, we consider the borderline case p=dσ1p=\frac{d}{\sigma-1}. In this scenario, we show that solutions are Log-Lipschitz continuous, achieving the same level of regularity as in [16, Theorem 2], but with the requirement of higher regularity for the source term ff.

Theorem 2.7.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a viscosity solution to (1.1), with fLp(B1)f\in L^{p}(B_{1}), p=dσ1p=\frac{d}{\sigma-1}. Then, uClocLogLip(B1)u\in C^{Log-Lip}_{loc}(B_{1}), and there exists a positive constant C=C(d,σ,λ,Λ)C=C(d,\sigma,\lambda,\Lambda), such that

uCLogLip(B1/2)C(uL(B1)+uLσ1(d)+fLdσ1(B1)).\|u\|_{C^{Log-Lip}(B_{1/2})}\leq C\left(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{L^{\frac{d}{\sigma-1}}(B_{1})}\right).

In what follows, we present our third main theorem. As before, we recover the local regularity in the limit as σ2\sigma\to 2, demonstrated in [16, Theorem 3]. Recall that α0\alpha_{0} comes from the C1,α0C^{1,\alpha_{0}}-regularity of σ{\mathcal{I}}_{\sigma}-harmonic functions.

Theorem 2.8.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a viscosity solution to (1.1), with fLp(B1)f\in L^{p}(B_{1}), p(dσ1,+)p\in\left(\frac{d}{\sigma-1},+\infty\right). Then, uCloc1,α(B1)u\in C^{1,\alpha}_{loc}(B_{1}) for any

(2.2) α(0,min(σ1dp,α0)].\alpha\in\left(0,\min\left(\sigma-1-\frac{d}{p},\alpha_{0}^{-}\right)\right].

Moreover, there exists a positive constant C=C(p,d,σ,λ,Λ)C=C(p,d,\sigma,\lambda,\Lambda), such that

uC1,α(B1/2)C(uL(B1)+uLσ1(d)+fLp(B1)).\|u\|_{C^{1,\alpha}(B_{1/2})}\leq C(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{L^{p}(B_{1})}).

Observe that as pp\to\infty, the corresponding Hölder exponent approaches σ1\sigma-1, indicating that we achieve CσC^{\sigma}-regularity in the case p=p=\infty, which is indeed the case as confirmed in [17]. However, we also show that this result holds under the weaker assumption that fBMO(B1)f\in\mathrm{BMO}(B_{1}), which is a proper subset of L(B1)L^{\infty}(B_{1}). This is the content of our last main result.

Theorem 2.9.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a viscosity solution to (1.9), with fBMO(B1)f\in{\rm BMO}(B_{1}). Then, uClocσ(B1)u\in C^{\sigma}_{loc}(B_{1}) and there exists a positive constant C=C(d,σ,λ,Λ)C=C(d,\sigma,\lambda,\Lambda), such that

uCσ(B1/2)C(uL(B1)+uLσ1(d)+fBMO(B1)).\|u\|_{C^{\sigma}(B_{1/2})}\leq C(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{{\rm BMO}(B_{1})}).

The proof of Theorem 2.9 differs from the strategies used for the previous theorems. This is primarily because scaling of the form xρσu(ρx)x\mapsto\rho^{-\sigma}u(\rho x), for ρ1\rho\ll 1, leads to a growth rate of |x|σ|x|^{-\sigma} at infinity, which increases too rapidly to be integrable with respect to the tails of our kernel. Consequently, we employ techniques similar to those in [13, 14], where Liouville-type results are used to establish interior regularity of solutions through blow-up arguments.

2.3. Auxiliary results

In this subsection, we prove some results used throughout the paper. Since we did not find any references stating exactly what we needed, we start with a comparison principle for viscosity solutions of (1.1) (and also (1.9)), in the case where f0f\equiv 0. See also [6, Theorem 5.2].

Proposition 2.10 (Comparison principle).

Let u,vLσ1(d)u,v\in L^{1}_{\sigma}(\mathbb{R}^{d}), uu upper semicontinuous and vv lower semicontinuous, be respectively a viscosity subsolution and supersolution to the equation

(2.3) σ(w,x)=0 in B1,{\mathcal{I}}_{\sigma}(w,x)=0\;\;\mbox{ in }\;B_{1},

such that uvu\leq v in NB1\mathbb{R}^{N}\setminus B_{1}. Then, uvu\leq v in B1B_{1}.

Proof.

Suppose by contradiction that

θ=supB1(uv)>0.\theta=\sup_{B_{1}}(u-v)>0.

For ε>0\varepsilon>0 we define the auxiliary function

Φ(x,y)=u(x)v(y)|xy|22ε2,\Phi(x,y)=u(x)-v(y)-\frac{|x-y|^{2}}{2\varepsilon^{2}},

and consider (xε,yε)B¯1×B¯1(x_{\varepsilon},y_{\varepsilon})\in\overline{B}_{1}\times\overline{B}_{1} such that

Φ(xε,yε)=supx,yB1Φ(x,y).\Phi(x_{\varepsilon},y_{\varepsilon})=\sup_{x,y\in B_{1}}\Phi(x,y).

Notice that Φ(xε,yε)supxB1Φ(x,x)=θ\Phi(x_{\varepsilon},y_{\varepsilon})\geq\sup_{x\in B_{1}}\Phi(x,x)=\theta, which yields to

(2.4) |xεyε|2ε2u(xε)v(yε)θ.\frac{|x_{\varepsilon}-y_{\varepsilon}|}{2\varepsilon^{2}}\leq u(x_{\varepsilon})-v(y_{\varepsilon})-\theta.

By compactness, we have that (xε,yε)(x¯,y¯)B¯1×B¯1(x_{\varepsilon},y_{\varepsilon})\rightarrow(\bar{x},\bar{y})\in\overline{B}_{1}\times\overline{B}_{1} as ε0\varepsilon\to 0, and by using (2.4) we obtain x¯=y¯\bar{x}=\bar{y}. Therefore

0limε0|xεyε|22ε2=u(x¯)v(x¯)θ0,0\leq\lim_{\varepsilon\rightarrow 0}\frac{|x_{\varepsilon}-y_{\varepsilon}|^{2}}{2\varepsilon^{2}}=u(\bar{x})-v(\bar{x})-\theta\leq 0,

which implies

u(x¯)v(x¯)=θ>0.u(\bar{x})-v(\bar{x})=\theta>0.

Moreover, since uvu\leq v on dB1\mathbb{R}^{d}\setminus B_{1}, we have x¯B1\bar{x}\in B_{1}. We set φ1(x):=v(yε)+|xyε|/2ε2\varphi_{1}(x):=v(y_{\varepsilon})+|x-y_{\varepsilon}|/2\varepsilon^{2} and φ2(y):=u(xε)+|yxε|/2ε2\varphi_{2}(y):=u(x_{\varepsilon})+|y-x_{\varepsilon}|/2\varepsilon^{2}, and observe that uφ1u-\varphi_{1} has a local maximum at xεx_{\varepsilon}, while vφ2v-\varphi_{2} has a local minimum at yεy_{\varepsilon}. Hence by using that uu is a subsolution and vv is a supersolution of (2.3), we have the viscosity inequalities

supα𝒜infβ(Brδ(φ1,xε,y)yTAα,β(xε)y|y|σ+d+2𝑑y+dBrδ(u,xε,y)yTAα,β(xε)y|y|σ+d+2𝑑y)0,\sup_{\alpha\in{\mathcal{A}}}\inf_{\beta\in{\mathcal{B}}}\left(\int_{B_{r}}\delta(\varphi_{1},x_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(x_{\varepsilon})y}{|y|^{\sigma+d+2}}dy+\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(u,x_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(x_{\varepsilon})y}{|y|^{\sigma+d+2}}dy\right)\geq 0,

and

supα𝒜infβ(Brδ(φ2,yε,y)yTAα,β(yε)y|y|σ+d+2𝑑y+dBrδ(v,yε,y)yTAα,β(yε)y|y|σ+d+2𝑑y)0,\sup_{\alpha\in{\mathcal{A}}}\inf_{\beta\in{\mathcal{B}}}\left(\int_{B_{r}}\delta(\varphi_{2},y_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(y_{\varepsilon})y}{|y|^{\sigma+d+2}}dy+\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(v,y_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(y_{\varepsilon})y}{|y|^{\sigma+d+2}}dy\right)\leq 0,

for rr small enough. Hence, from the definition of sup\sup and inf\inf, there exist α𝒜\alpha\in{\mathcal{A}} and β\beta\in{\mathcal{B}} such that

ε2or(1)+dBrδ(u,xε,y)yTAα,β(xε)y|y|σ+d+2𝑑yγ/2,\varepsilon^{-2}o_{r}(1)+\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(u,x_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(x_{\varepsilon})y}{|y|^{\sigma+d+2}}dy\geq-\gamma/2,

and

ε2or(1)+dBrδ(v,yε,y)yTAα,β(yε)y|y|σ+d+2𝑑yγ/2,\varepsilon^{-2}o_{r}(1)+\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(v,y_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(y_{\varepsilon})y}{|y|^{\sigma+d+2}}dy\leq\gamma/2,

for γ>0\gamma>0 sufficiently small. Subtracting the inequalities above yields to

ε2or(1)+dBrδ(u,xε,y)yTAα,β(xε)y|y|σ+d+2𝑑ydBrδ(v,yε,y)yTAα,β(yε)y|y|σ+d+2𝑑yγ.\varepsilon^{-2}o_{r}(1)+\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(u,x_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(x_{\varepsilon})y}{|y|^{\sigma+d+2}}dy-\int_{\mathbb{R}^{d}\setminus B_{r}}\delta(v,y_{\varepsilon},y)\dfrac{y^{T}A_{\alpha,\beta}(y_{\varepsilon})y}{|y|^{\sigma+d+2}}dy\geq-\gamma.

Notice that, When ε0\varepsilon\rightarrow 0, by the contradiction hypotheses we have

B1Brδ(u,x¯,y)yTAα,β(x¯)y|y|σ+d+2𝑑yB1Brδ(v,x¯,y)yTAα,β(x¯)y|y|σ+d+2𝑑y0.\int_{B_{1}\setminus B_{r}}\delta(u,\bar{x},y)\dfrac{y^{T}A_{\alpha,\beta}(\bar{x})y}{|y|^{\sigma+d+2}}dy-\int_{B_{1}\setminus B_{r}}\delta(v,\bar{x},y)\dfrac{y^{T}A_{\alpha,\beta}(\bar{x})y}{|y|^{\sigma+d+2}}dy\leq 0.

Moreover,

dB1δ(u,x¯,y)yTAα,β(x¯)y|y|σ+d+2𝑑ydB1δ(v,x¯,y)yTAα,β(x¯)y|y|σ+d+2𝑑yλθdB1|z|(d+σ)𝑑z.\int_{\mathbb{R}^{d}\setminus B_{1}}\delta(u,\bar{x},y)\dfrac{y^{T}A_{\alpha,\beta}(\bar{x})y}{|y|^{\sigma+d+2}}dy-\int_{\mathbb{R}^{d}\setminus B_{1}}\delta(v,\bar{x},y)\dfrac{y^{T}A_{\alpha,\beta}(\bar{x})y}{|y|^{\sigma+d+2}}dy\leq-\lambda\theta\int_{\mathbb{R}^{d}\setminus B_{1}}|z|^{-(d+\sigma)}dz.

Therefore

λθdB1|z|(d+σ)𝑑zγ,-\lambda\theta\int_{\mathbb{R}^{d}\setminus B_{1}}|z|^{-(d+\sigma)}dz\geq-\gamma,

which is a contradiction for γ\gamma small enough. This finishes the proof. ∎

We now focus on one of the main contributions of this paper: the stability of solutions to (1.1) when fLp(B1)f\in L^{p}(B_{1}). To the best of our knowledge, this is the first time such a result has been established in the fully nonlinear nonlocal context. For comparison, see [6, Corollary 4.7].

Proposition 2.11 (Compactness and stability).

Let uku_{k} be a normalized viscosity solution to

σ(uk,x)=fk in B1.{\mathcal{I}}_{\sigma}(u_{k},x)=f_{k}\;\;\mbox{ in }\;B_{1}.

Suppose that there exists a positive constant MM such that

(2.5) |uk(x)|M(1+|x|)1+α for all xd.|u_{k}(x)|\leq M(1+|x|)^{1+\alpha}\;\;\mbox{ for all }\;x\in\mathbb{R}^{d}.

Suppose further that

fkLp(B1)0,\|f_{k}\|_{L^{p}(B_{1})}\to 0,

for p(dε0,+)p\in(d-\varepsilon_{0},+\infty). Then there exists uC(B1)Lσ1(d)u_{\infty}\in C(B_{1})\cap L_{\sigma}^{1}(\mathbb{R}^{d}) such that

ukuL(B4/5)0.\|u_{k}-u_{\infty}\|_{L^{\infty}(B_{4/5})}\to 0.

Moreover, uu_{\infty} solves

(2.6) σ(u,x)=0 in B1.{\mathcal{I}}_{\sigma}(u_{\infty},x)=0\;\;\mbox{ in }\;B_{1}.
Proof.

First, observe that given R>0R>0, we obtain from (2.5)

|uk(x)|C(R) for any xBR.|u_{k}(x)|\leq C(R)\;\;\mbox{ for any }\;x\in B_{R}.

Hence, Given any compact set Ωd\Omega\subset\mathbb{R}^{d}, we have that the a.e. limits

u¯(x):=lim supk,ykxuk(yk),x in Ω,\bar{u}(x):=\limsup_{k\to\infty,y_{k}\to x}u_{k}(y_{k}),\;\;x\mbox{ in }\;\Omega,

and

u¯(x):=lim infk,ykxuk(yk),x in Ω,\underline{u}(x):=\liminf_{k\to\infty,y_{k}\to x}u_{k}(y_{k}),\;\;x\mbox{ in }\;\Omega,

are well-defined. Since the a.e. convergence holds for every compact set of d\mathbb{R}^{d}, we also have the a.e. convergence in the whole d\mathbb{R}^{d}. Using this fact and once again (2.5), the Dominated Convergence Theorem ensures that

(2.7) u¯ukLσ1(d)0,\|\bar{u}-u_{k}\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}\to 0,

and

(2.8) u¯ukLσ1(d)0,\|\underline{u}-u_{k}\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}\to 0,

through the respective subsequences. We are going to show that u¯\bar{u} is a viscosity subsolution to (2.6), and u¯\underline{u} is a viscosity supersolution to (2.6). We will prove the subsolution case since the supersolution case is analogous. Let x0B1x_{0}\in B_{1} and φC2(Br(x0))\varphi\in C^{2}(B_{r}(x_{0})) be such that u¯φ\bar{u}-\varphi has a maximum at x0x_{0}. Without loss of generality, we can assume that φ\varphi is defined by

φ={P,inBr(x0),u¯,onBr(x0)c,\varphi=\left\{\begin{aligned} P,&\quad\mbox{in}\quad B_{r}(x_{0}),\\ \bar{u},&\quad\mbox{on}\quad B_{r}(x_{0})^{c},\end{aligned}\right.

for some paraboloid PP. We need to show that

(2.9) σ(φ,x)0 in B1.{\mathcal{I}}_{\sigma}(\varphi,x)\geq 0\;\;\mbox{ in }\;B_{1}.

Suppose by contradiction that

(2.10) σ(φ,x)<η,{\mathcal{I}}_{\sigma}(\varphi,x)<-\eta,

for some η>0\eta>0. Now, let ψk\psi_{k} be a viscosity solution to

(2.11) {λ,1+(ψk,x)=|fk(x)| in Br(x0)ψk=0 on Br(x0),\left\{\begin{aligned} {\mathcal{M}}^{+}_{\lambda^{*},1}(\psi_{k},x)&=&-|f_{k}(x)|&\;\;\mbox{ in }\;B_{r}(x_{0})\\ \psi_{k}&=&0&\;\;\mbox{ on }\;\partial B_{r}(x_{0}),\end{aligned}\right.

for some λ<1\lambda^{*}<1 to be chosen later. Here, the maximal operator +{\mathcal{M}}^{+} is defined with respect to the class 0{\mathcal{L}}_{0}, as in [6] (and defined below). We have

σ(φ+ψk,x)σ(φ,x)\displaystyle{\mathcal{I}}_{\sigma}(\varphi+\psi_{k},x)-{\mathcal{I}}_{\sigma}(\varphi,x) λ,Λ+(ψk,x)\displaystyle\leq{\mathcal{M}}^{+}_{\lambda,\Lambda}(\psi_{k},x)
=Λnδ+(ψk,x,y)|y|n+σ𝑑yλnδ(ψk,x,y)|y|n+σ𝑑y\displaystyle=\Lambda\int_{\mathbb{R}^{n}}\dfrac{\delta^{+}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy-\lambda\int_{\mathbb{R}^{n}}\dfrac{\delta^{-}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy
=Λnδ+(ψk,x,y)|y|n+σ𝑑yλλ(nδ+(ψk,x,y)|y|n+σ𝑑y+|fk(x)|)\displaystyle=\Lambda\int_{\mathbb{R}^{n}}\dfrac{\delta^{+}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy-\dfrac{\lambda}{\lambda^{*}}\left(\int_{\mathbb{R}^{n}}\dfrac{\delta^{+}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy+|f_{k}(x)|\right)
=(Λλλ)nδ+(ψk,x,y)|y|n+σ𝑑yλλ|fk(x)|,\displaystyle=\left(\Lambda-\dfrac{\lambda}{\lambda^{*}}\right)\int_{\mathbb{R}^{n}}\dfrac{\delta^{+}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy-\dfrac{\lambda}{\lambda^{*}}|f_{k}(x)|,

where we have used (2.11) to conclude

nδ+(ψk,x,y)|y|n+σ𝑑yλnδ(ψk,x,y)|y|n+σ𝑑y=|fk(x)|.\int_{\mathbb{R}^{n}}\dfrac{\delta^{+}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy-\lambda^{*}\int_{\mathbb{R}^{n}}\dfrac{\delta^{-}(\psi_{k},x,y)}{|y|^{n+\sigma}}dy=-|f_{k}(x)|.

Now, by choosing λ\lambda^{*} such that Λλλ0\Lambda-\dfrac{\lambda}{\lambda^{*}}\leq 0 and λλ1-\dfrac{\lambda}{\lambda^{*}}\leq-1, we obtain

(2.12) σ(φ+ψk,x)σ(φ,x)+fk(x).{\mathcal{I}}_{\sigma}(\varphi+\psi_{k},x)\leq{\mathcal{I}}_{\sigma}(\varphi,x)+f_{k}(x).

Let PkP_{k} be defined by

(2.13) φk={P,inBr(x0),uk,onBr(x0)c,\varphi_{k}=\left\{\begin{aligned} P,&\quad\mbox{in}\quad B_{r}(x_{0}),\\ u_{k},&\quad\mbox{on}\quad B_{r}(x_{0})^{c},\end{aligned}\right.

and

(2.14) P~={c|xx0|2,inBr(x0),0,onBr(x0)c.\tilde{P}=\left\{\begin{aligned} c|x-x_{0}|^{2},&\quad\mbox{in}\quad B_{r}(x_{0}),\\ 0,&\quad\mbox{on}\quad B_{r}(x_{0})^{c}.\end{aligned}\right.

By using the ABP estimates in [11, Theorem 3.1] we have ψk0\|\psi_{k}\|_{\infty}\rightarrow 0, as kk\rightarrow\infty. Then, there exists a xkBr(x0)x_{k}\in B_{r}(x_{0}) such that φk+ψk+P~\varphi_{k}+\psi_{k}+\tilde{P} touch uku_{k} by above in Br(x0)B_{r}(x_{0}). Therefore, we have the viscosity inequality

(2.15) σ(φk+ψk+P~)f(xk).{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k}+\tilde{P})\geq f(x_{k}).

By ellipticity and using (2.10) and (2.12) we obtain that

(2.16) σ(φk+ψk+P~)σ(φk+ψk)+σ+(P~)σ(φk+ψk)σ(φ+ψk)η+f(xk)+σ+(P~).{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k}+\tilde{P})\leq{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k})+{\mathcal{M}}^{+}_{\sigma}(\tilde{P})\leq{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k})-{\mathcal{I}}_{\sigma}(\varphi+\psi_{k})-\eta+f(x_{k})+{\mathcal{M}}^{+}_{\sigma}(\tilde{P}).

We observe that

|Nδ(φk+ψk,x,y)𝑑yNδ(φ+ψk,x,y)𝑑y|\displaystyle\Big{|}\int_{\mathbb{R}^{N}}\delta(\varphi_{k}+\psi_{k},x,y)dy-\int_{\mathbb{R}^{N}}\delta(\varphi+\psi_{k},x,y)dy\Big{|} NBr(x0)|δ(uk,x,y)δ(u¯,x,y)|𝑑y\displaystyle\leq\int_{\mathbb{R}^{N}\setminus B_{r}(x_{0})}|\delta(u_{k},x,y)-\delta(\bar{u},x,y)|dy
C(r)uku¯Lσ1(d),\displaystyle\leq C(r)\|u_{k}-\bar{u}\|_{L^{1}_{\sigma}(\mathbb{R}^{d})},

and by using (2.7), we can conclude that for kk sufficiently large

(2.17) |σ(φk+ψk)σ(φ+ψk)|η4.|{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k})-{\mathcal{I}}_{\sigma}(\varphi+\psi_{k})|\leq\dfrac{\eta}{4}.

Hence, from (2.16) and (2.17), we obtain

σ(φk+ψk+P~)η4η+f(xk)+σ+(P~).{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k}+\tilde{P})\leq\dfrac{\eta}{4}-\eta+f(x_{k})+{\mathcal{M}}^{+}_{\sigma}(\tilde{P}).

Now, choose c(Λ,N,σ)c(\Lambda,N,\sigma) sufficiently small so that

σ+(P~)η/4.{\mathcal{M}}^{+}_{\sigma}(\tilde{P})\leq\eta/4.

Finally, for kk sufficiently large we get

σ(φk+ψk+P~,xk)η/2+f(xk),{\mathcal{I}}_{\sigma}(\varphi_{k}+\psi_{k}+\tilde{P},x_{k})\leq-\eta/2+f(x_{k}),

which is a contradiction with (2.15). This finishes the proof of (2.9), i.e., u¯\bar{u} solves in the viscosity sense

σ(u¯,x)0 in B1,{\mathcal{I}}_{\sigma}(\bar{u},x)\geq 0\;\;\mbox{ in }\;B_{1},

We similarly show that

σ(u¯,x)0 in B1.{\mathcal{I}}_{\sigma}(\underline{u},x)\leq 0\;\;\mbox{ in }\;B_{1}.

Now, from the definition of u¯\bar{u}, u¯\underline{u} and the viscosity inequalities above we can infer from Proposition 2.10, that in fact

u¯=u¯=u,\bar{u}=\underline{u}=u_{\infty},

and hence up to a subsequence, ukuu_{k}\to u_{\infty} locally uniformly in B1B_{1} (see for instance [2, Lemma 6.2]). Moreover, from the viscosity inequalities satisfied by u¯\bar{u} and u¯\underline{u}, we conclude that uu_{\infty} solves

σ(u,x)=0 in B1,{\mathcal{I}}_{\sigma}(u_{\infty},x)=0\;\;\mbox{ in }\;B_{1},

in the viscosity sense. ∎

Using the stability result above, we can prove the following Approximation Lemma, which relates the solutions to our problem with σ{\mathcal{I}}_{\sigma}-harmonic functions.

Lemma 2.12 (Approximation Lemma).

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized viscosity solution to (1.1), with p(dε0,+)p\in(d-\varepsilon_{0},+\infty). Suppose that

|u(x)|M(1+|x|)1+α for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

Given δ>0\delta>0 there exist ε>0\varepsilon>0, such that if

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

we can find a function hC1,α0(B4/5)h\in C^{1,\alpha_{0}}(B_{4/5}) satisfying

supB3/4|uh|δ.\sup_{B_{3/4}}|u-h|\leq\delta.
Proof.

Suppose not, then there exist δ0>0\delta_{0}>0 and sequences (uk)k(u_{k})_{k\in\mathbb{N}}, (fk)k(f_{k})_{k\in\mathbb{N}} such that

(2.18) σ(uk,x)=fk in B1,{\mathcal{I}}_{\sigma}(u_{k},x)=f_{k}\;\;\mbox{ in }\;B_{1},\\
(2.19) fkLp(B1)1k,\|f_{k}\|_{L^{p}(B_{1})}\leq\dfrac{1}{k},

and

(2.20) |uk(x)|M(1+|x|)1+α for all xd,|u_{k}(x)|\leq M(1+|x|)^{1+\alpha}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d},

but,

(2.21) |ukh|>δ0,|u_{k}-h|>\delta_{0},

for all hC1,α0(B4/5)h\in C^{1,\alpha_{0}}(B_{4/5}). From the contradiction hypotheses (2.18), (2.19), (2.20) and Proposition 2.11, we can guarantee the existence of a function uC(B1)Lσ1(d)u_{\infty}\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) such that ukuu_{k}\to u_{\infty} locally uniformly in B1B_{1} satisfying

(2.22) σ(u,x)=0 in B1.{\mathcal{I}}_{\sigma}(u_{\infty},x)=0\;\;\mbox{ in }\;B_{1}.\\

Now, the regularity available for (2.22), see [6], implies that uC1,α0(B4/5)u_{\infty}\in C^{1,\alpha_{0}}(B_{4/5}). By taking huh\equiv u_{\infty}, we reach a contradiction with (2.21) for kk sufficiently large. ∎

3. Hölder regularity

In this section, we detail the proof of Theorem 2.6, namely, the optimal ClocαC^{\alpha}_{loc}-regularity, for

α(0,σpdp],\alpha\in\left(0,\frac{\sigma p-d}{p}\right],

where p(dε0,dσ1)p\in\left(d-\varepsilon_{0},\frac{d}{\sigma-1}\right). We start by applying Lemma 2.12 and showing the existence of a constant close to uu in sufficiently small balls.

Proposition 3.1.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized a viscosity solution to (1.1), with p(dε0,dσ1)p\in\left(d-\varepsilon_{0},\frac{d}{\sigma-1}\right). Assume that

|u(x)|M(1+|x|)1+α for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then there exist constants 0<ρ1/20<\rho\ll 1/2 and AA satisfying, |A|C|A|\leq C and

supBρ|uA|ρα,\sup_{B_{\rho}}|u-A|\leq\rho^{\alpha},

where C>0C>0 is a universal constant.

Proof.

Fix δ>0\delta>0 (to be chosen later) and let hh be the function from Lemma 2.12. Since hC1,α0(B4/5)h\in C^{1,\alpha_{0}}(B_{4/5}), for ρ\rho sufficiently small, we have

|hh(0)|Cρ.|h-h(0)|\leq C\rho.

Now, from Lemma 2.12 and the Triangular inequality we obtain

supBρ|uh(0)|\displaystyle\sup_{B_{\rho}}|u-h(0)| supBρ|uh|+supBρ|hh(0)|\displaystyle\leq\sup_{B_{\rho}}|u-h|+\sup_{B_{\rho}}|h-h(0)|
δ+Cρ.\displaystyle\leq\delta+C\rho.

Now, we make the universal choices

(3.1) ρ=min[(12C)11α,(1(1+C)100)1α0α] and δ=ρα2,\rho=\min\left[\left(\frac{1}{2C}\right)^{\frac{1}{1-\alpha}},\left(\frac{1}{(1+C)100}\right)^{\frac{1}{\alpha_{0}-\alpha}}\right]\;\;\mbox{ and }\;\;\delta=\frac{\rho^{\alpha}}{2},

and by setting A=h(0)A=h(0), we conclude that

supBρ|uA|ρα.\sup_{B_{\rho}}|u-A|\leq\rho^{\alpha}.

Notice that the choice of δ\delta determines the value of ε\varepsilon via Lemma 2.12. ∎

In what follows, we iterate the previous proposition to find a sequence of constants that approaches uu at the origin.

Proposition 3.2.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized a viscosity solution to (1.1), with p(dε0,dσ1)p\in\left(d-\varepsilon_{0},\frac{d}{\sigma-1}\right). Assume that

|u(x)|M(1+|x|)1+α for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then we can find a sequence (Ak)k(A_{k})_{k\in\mathbb{N}} satisfying

(3.2) supBρk|u(x)Ak|ρkα,\sup_{B_{\rho^{k}}}|u(x)-A_{k}|\leq\rho^{k\alpha},

with

(3.3) |Ak+1Ak|Cρkα.|A_{k+1}-A_{k}|\leq C\rho^{k\alpha}.
Proof.

We argue by an induction argument. By setting A0=A1=0A_{0}=A_{1}=0, the case k=0k=0 follows immediately. Suppose we have verified the statement for k=1,,nk=1,\ldots,n, and let us prove the case k=n+1k=n+1. We introduce the auxiliary function vk:dv_{k}:\mathbb{R}^{d}\longrightarrow\mathbb{R}

vk(x):=u(ρkx)Akρkα.v_{k}(x):=\frac{u(\rho^{k}x)-A_{k}}{\rho^{k\alpha}}.

Notice that by (3.2) we have |vk(x)|1|v_{k}(x)|\leq 1 in B1B_{1}. In addition, vkv_{k} solves

σ(v,x)=f~(x) in B1,\mathcal{I}_{\sigma}(v,x)=\tilde{f}(x)\;\;\mbox{ in }\;B_{1},

where f~(x)=ρk(σα)f(ρkx)\tilde{f}(x)=\rho^{k(\sigma-\alpha)}f(\rho^{k}x). Moreover, our choice of α\alpha in (2.1) assures f~Lp(B1)ε\|\tilde{f}\|_{L^{p}(B_{1})}\leq\varepsilon. Next, we are going to show that vkv_{k} satisfies

(3.4) |v(x)|M(1+|x|1+α0) for all xd,|v(x)|\leq M(1+|x|^{1+\alpha_{0}})\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d},

for some universal constant MM. In fact, we resort again to an induction argument. For k=0k=0, we have v0=uv_{0}=u and (3.4) is verified. Now, assume that the case k=1,,nk=1,\ldots,n is already verified. We shall prove the case k=n+1k=n+1. Observe that

vn+1=vn(ρx)A~nρα,v_{n+1}=\dfrac{v_{n}(\rho x)-\tilde{A}_{n}}{\rho^{\alpha}},

where A~n\tilde{A}_{n} comes from Lemma (2.12) applied to vnv_{n}. Now, for 2|x|ρ>12|x|\rho>1 we estimate

|vn+1(x)|\displaystyle|v_{n+1}(x)| ρα(|vn(ρx)|+|A~n|)\displaystyle\leq\rho^{-\alpha}\left(|v_{n}(\rho x)|+|\tilde{A}_{n}|\right)
ρ(1+α)[(1+ρ1+α0|x|1+α0)+C(1+ρ|x|)]\displaystyle\leq\rho^{-(1+\alpha)}\left[(1+\rho^{1+\alpha_{0}}|x|^{1+\alpha_{0}})+C(1+\rho|x|)\right]
ρ(α0α)(5+9C)|x|1+α0\displaystyle\leq\rho^{(\alpha_{0}-\alpha)}(5+9C)|x|^{1+\alpha_{0}}
|x|1+α0,\displaystyle\leq|x|^{1+\alpha_{0}},

where in the last inequality we used (3.1). On the other hand, if 2|x|ρ12|x|\rho\leq 1, we obtain

|vn+1(x)|\displaystyle|v_{n+1}(x)| ρα(|vn(ρx)h~(ρx)|+|h~(ρx)A~n|\displaystyle\leq\rho^{-\alpha}(|v_{n}(\rho x)-\tilde{h}(\rho x)|+|\tilde{h}(\rho x)-\tilde{A}_{n}|
ρα(ρα2+Cρ|x|)\displaystyle\leq\rho^{-\alpha}\left(\dfrac{\rho^{\alpha}}{2}+C\rho|x|\right)
12+C2ρα\displaystyle\leq\frac{1}{2}+\frac{C}{2\rho^{\alpha}}
M(1+|x|1+α0),\displaystyle\leq M(1+|x|^{1+\alpha_{0}}),

where M:=1/2+C/(2ρα)M:=1/2+C/(2\rho^{\alpha}), and hence (3.4) is proved. Finally, we now can apply Proposition 3.1 to vkv_{k} and we obtain

supBρ|vkA~k|ρα,\sup_{B_{\rho}}|v_{k}-\tilde{A}_{k}|\leq\rho^{\alpha},

and rescaling back to uu we conclude

supBρk+1|uAk+1|ρ(k+1)α,\sup_{B_{\rho^{k+1}}}|u-A_{k+1}|\leq\rho^{(k+1)\alpha},

where Ak+1=Ak+ρkαA~kA_{k+1}=A_{k}+\rho^{k\alpha}\tilde{A}_{k}, which satisfies (3.3). This finishes the proof. ∎

We are now ready to prove Theorem 2.6.

Proof of Theorem 2.6.

Notice that from (3.3) we have that (Ak)k(A_{k})_{k\in\mathbb{N}} is a Cauchy sequence, and hence there exists AA_{\infty} such that AkAA_{k}\to A_{\infty}, as kk\to\infty. Moreover, we also have from (3.3)

|AkA|Cρkα.|A_{k}-A_{\infty}|\leq C\rho^{k\alpha}.

Now, fix 0<r10<r\ll 1 and let kk\in\mathbb{N} be such that ρk+1rρk\rho^{k+1}\leq r\leq\rho^{k}. We estimate,

supBr|u(x)A|\displaystyle\sup_{B_{r}}|u(x)-A_{\infty}| supBρk|u(x)Ak|+supBρk|AkA|\displaystyle\leq\sup_{B_{\rho^{k}}}|u(x)-A_{k}|+\sup_{B_{\rho^{k}}}|A_{k}-A_{\infty}|
ρkα+Cρkα\displaystyle\leq\rho^{k\alpha}+C\rho^{k\alpha}
(C+1)ραρ(k+1)α\displaystyle\leq\frac{(C+1)}{\rho^{\alpha}}\rho^{(k+1)\alpha}
Crα.\displaystyle\leq Cr^{\alpha}.

By taking the limit as kk\to\infty in (3.2), we obtain A=u(0)A_{\infty}=u(0). This finishes the proof. ∎

4. Log-Lipschitz continuity

This section addresses the first critical case p=dσ1p=\frac{d}{\sigma-1}, which yields the desired Log-Lipschitz regularity. In particular, solutions are of class ClocαC_{loc}^{\alpha} for every α(0,1)\alpha\in(0,1). As before, we begin by demonstrating the existence of a linear approximation of uu within sufficiently small balls.

Proposition 4.1.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized viscosity solution to (1.1), with p=dσ1p=\frac{d}{\sigma-1}. Suppose further that

|u(x)|M(1+|x|)1+α0 for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha_{0}}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then, there exist a constant 0<ρ1/20<\rho\ll 1/2 and an affine function \ell of the form

(x)=A+Bx,\ell(x)=A+B\cdot x,

satisfying |A|,|B|C|A|,|B|\leq C and

supBρ|u(x)(x)|ρ.\sup_{B_{\rho}}|u(x)-\ell(x)|\leq\rho.
Proof.

The proof is similar to Proposition 3.1. We fix δ>0\delta>0 to be determined later. For ρ1/2\rho\ll 1/2, we have that

|h(x)h(0)Dh(0)x|Cρ1+α0,|h(x)-h(0)-Dh(0)\cdot x|\leq C\rho^{1+\alpha_{0}},

where hC1,α0(B3/4)h\in C^{1,\alpha_{0}}(B_{3/4}) comes from Lemma 2.12. By setting (x)=h(0)+Dh(0)x\ell(x)=h(0)+Dh(0)\cdot x, we obtain from the Triangular inequality that

supBρ|u(x)(x)|\displaystyle\sup_{B_{\rho}}|u(x)-\ell(x)| supBρ|u(x)h(x)|+supBρ|h(x)(x)|\displaystyle\leq\sup_{B_{\rho}}|u(x)-h(x)|+\sup_{B_{\rho}}|h(x)-\ell(x)|
δ+Cρ1+α0.\displaystyle\leq\delta+C\rho^{1+\alpha_{0}}.

As before, we make universal choices

(4.1) ρ=(12C)1α0 and δ=ρ2,\rho=\left(\frac{1}{2C}\right)^{\frac{1}{\alpha_{0}}}\;\;\mbox{ and }\;\;\delta=\frac{\rho}{2},

which determines the value of ε\varepsilon through Lemma 2.12. Therefore

supBρ|u(x)(x)|ρ.\sup_{B_{\rho}}|u(x)-\ell(x)|\leq\rho.

Proposition 4.2.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized viscosity solution to (1.1), with p=dσ1p=\frac{d}{\sigma-1}. Suppose further that

|u(x)|M(1+|x|)1+α0 for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha_{0}}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then, there exists a sequence of affine function (k)k(\ell_{k})_{k\in\mathbb{N}} of the form

k(x)=Ak+Bkx,\ell_{k}(x)=A_{k}+B_{k}\cdot x,

satisfying

(4.2) |Ak+1Ak|ρk+|Bk+1Bk|C,\frac{|A_{k+1}-A_{k}|}{\rho^{k}}+|B_{k+1}-B_{k}|\leq C,

and

supBρ|u(x)(x)|ρk.\sup_{B_{\rho}}|u(x)-\ell(x)|\leq\rho^{k}.
Proof.

As before, we resort to an induction argument. By considering 0=1=0\ell_{0}=\ell_{1}=0, the case k=0k=0 follows trivially. Now, suppose that the cases k=1,,nk=1,\ldots,n have been verified, and let us prove the case k=n+1k=n+1. We define the auxiliary function vk:dv_{k}:\mathbb{R}^{d}\longrightarrow\mathbb{R} by

vk(x):=u(ρkx)(ρkx)ρk.v_{k}(x):=\frac{u(\rho^{k}x)-\ell(\rho^{k}x)}{\rho^{k}}.

We have that vk(x)1v_{k}(x)\leq 1 in B1B_{1}, and solves

σ(x,v)=f~(x) in B1,{\mathcal{I}_{\sigma}}(x,v)=\tilde{f}(x)\;\;\mbox{ in }\;\;B_{1},

where f~(x)=ρk(σ1)f(ρkx)\tilde{f}(x)=\rho^{k(\sigma-1)}f(\rho^{k}x). Notice that as σ1>0\sigma-1>0, we have f~Lp(B1)ε\|\tilde{f}\|_{L^{p}(B_{1})}\leq\varepsilon. Arguing similarly as in Proposition 3.2, we can also show that

|vk(x)|1+|x|1+α0.|v_{k}(x)|\leq 1+|x|^{1+\alpha_{0}}.

Hence, we can apply Proposition 5.1 to vkv_{k} to conclude that there exists ~k=A~+B~x\tilde{\ell}_{k}=\tilde{A}+\tilde{B}x such that

supBρ|vk(x)~k(x)|ρ.\sup_{B_{\rho}}|v_{k}(x)-\tilde{\ell}_{k}(x)|\leq\rho.

Rescaling back to uu, we obtain

supBρ|u(x)k+1(x)|ρk,\sup_{B_{\rho}}|u(x)-\ell_{k+1}(x)|\leq\rho^{k},

where k+1=k(x)+ρk~k(ρ1x)\ell_{k+1}=\ell_{k}(x)+\rho^{k}\tilde{\ell}_{k}(\rho^{-1}x). Observe that

|Ak+1Ak|=|A~ρk|Cρk,|A_{k+1}-A_{k}|=|\tilde{A}\rho^{k}|\leq C\rho^{k},

and

|Bk+1Bk|=|B~|C|B_{k+1}-B_{k}|=|\tilde{B}|\leq C

which prove the condition (4.2). ∎

We now present the proof of Theorem 2.7

Proof of Theorem 2.7.

Notice that, by (4.2), (Ak)k(A_{k})_{k\in\mathbb{N}} is a Cauchy sequence, and hence, there exists AA_{\infty} such that AkAA_{k}\to A_{\infty}, as kk\to\infty. Moreover, we have

|AkA|Cρkα.|A_{k}-A_{\infty}|\leq C\rho^{k\alpha}.

Now, fix 0<r10<r\ll 1 and let kk\in\mathbb{N} be such that ρk+1rρk\rho^{k+1}\leq r\leq\rho^{k}. We have,

supBr|u(x)A|\displaystyle\sup_{B_{r}}|u(x)-A_{\infty}| supBρk|u(x)AkBkx|+supBρk|Bkx|\displaystyle\leq\sup_{B_{\rho^{k}}}|u(x)-A_{k}-B_{k}x|+\sup_{B_{\rho^{k}}}|B_{k}x|
ρk+Ckρk\displaystyle\leq\rho^{k}+Ck\rho^{k}
1ρ(ρk+1+Ckρk+1)\displaystyle\leq\frac{1}{\rho}\left(\rho^{k+1}+Ck\rho^{k+1}\right)
Cr+lnrlnρCr\displaystyle\leq Cr+\dfrac{\ln{r}}{\ln{\rho}}Cr
Crlnr.\displaystyle\leq-Cr\ln{r}.

Finally, by taking the limit as kk\to\infty in (3.2), we obtain A=u(0)A_{\infty}=u(0). This finishes the proof. ∎

5. Höder continuity of the gradient

In this section, we give the proof of Theorem 2.8, in which we prove Cloc1,αC_{loc}^{1,\alpha}-regularity for

α(0,min(σ1dp,α0)].\alpha\in\left(0,\min\left(\sigma-1-\frac{d}{p}\,,\alpha_{0}^{-}\right)\right].

The proof follows the general lines of the proof of Theorem 2.6, but now at the gradient level.

Proposition 5.1.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized viscosity solution to (1.1), with p(dσ1,+)p\in\left(\frac{d}{\sigma-1},+\infty\right). Suppose further that

|u(x)|M(1+|x|)1+α0 for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha_{0}}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then, there exist a constant 0<ρ1/20<\rho\ll 1/2 and an affine function \ell of the form

(x)=A+Bx,\ell(x)=A+B\cdot x,

satisfying |A|,|B|C|A|,|B|\leq C and

supBρ|u(x)(x)|ρ1+α.\sup_{B_{\rho}}|u(x)-\ell(x)|\leq\rho^{1+\alpha}.
Proof.

We fix δ>0\delta>0 to be determined later. For ρ1/2\rho\ll 1/2, we have that

|h(x)h(0)Dh(0)x|Cρ1+α0,|h(x)-h(0)-Dh(0)\cdot x|\leq C\rho^{1+\alpha_{0}},

where hC1,α0(B3/4)h\in C^{1,\alpha_{0}}(B_{3/4}) comes from Lemma 2.12. By setting (x)=h(0)+Dh(0)x\ell(x)=h(0)+Dh(0)\cdot x, we obtain from the Triangular inequality that

supBρ|u(x)(x)|\displaystyle\sup_{B_{\rho}}|u(x)-\ell(x)| supBρ|u(x)h(x)|+supBρ|h(x)(x)|\displaystyle\leq\sup_{B_{\rho}}|u(x)-h(x)|+\sup_{B_{\rho}}|h(x)-\ell(x)|
δ+Cρ1+α0.\displaystyle\leq\delta+C\rho^{1+\alpha_{0}}.

As before, we make universal choices

(5.1) ρ=min[(12C)1α0α,(1(1+C)100)1α0α] and δ=ρ1+α2,\rho=\min\left[\left(\frac{1}{2C}\right)^{\frac{1}{\alpha_{0}-\alpha}},\left(\frac{1}{(1+C)100}\right)^{\frac{1}{\alpha_{0}-\alpha}}\right]\;\;\mbox{ and }\;\;\delta=\frac{\rho^{1+\alpha}}{2},

which determines the value of ε\varepsilon through Lemma 2.12. Therefore

supBρ|u(x)(x)|ρ1+α.\sup_{B_{\rho}}|u(x)-\ell(x)|\leq\rho^{1+\alpha}.

Proposition 5.2.

Let uC(B1)Lσ1(d)u\in C(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}) be a normalized viscosity solution to (1.1), with p(dσ1,+)p\in\left(\frac{d}{\sigma-1},+\infty\right). Suppose further that

|u(x)|M(1+|x|)1+α0 for all xd.|u(x)|\leq M(1+|x|)^{1+\alpha_{0}}\;\;\mbox{ for all }\;\;x\in\mathbb{R}^{d}.

If

fLp(B1)ε,\|f\|_{L^{p}(B_{1})}\leq\varepsilon,

then, there exists a sequence of affine function (k)k(\ell_{k})_{k\in\mathbb{N}} of the form

k(x)=Ak+Bkx,\ell_{k}(x)=A_{k}+B_{k}\cdot x,

satisfying

(5.2) |Ak+1Ak|+ρk|Bk+1Bk|Cρk(1+α),|A_{k+1}-A_{k}|+\rho^{k}|B_{k+1}-B_{k}|\leq C\rho^{k(1+\alpha)},

and

supBρk|u(x)k(x)|ρk(1+α).\sup_{B_{\rho^{k}}}|u(x)-\ell_{k}(x)|\leq\rho^{k(1+\alpha)}.
Proof.

As before, we resort to an induction argument. By considering 0=1=0\ell_{0}=\ell_{1}=0, the case k=0k=0 follows trivially. Now, suppose that the cases k=1,,nk=1,\ldots,n have been verified, and let us prove the case k=n+1k=n+1. We define the auxiliary function vk:dv_{k}:\mathbb{R}^{d}\longrightarrow\mathbb{R} by

vk(x):=u(ρkx)(ρkx)ρk(1+α).v_{k}(x):=\frac{u(\rho^{k}x)-\ell(\rho^{k}x)}{\rho^{k(1+\alpha)}}.

We have that vk(x)1v_{k}(x)\leq 1 in B1B_{1}, and solves

σ(x,v)=f~(x) in B1,{\mathcal{I}_{\sigma}}(x,v)=\tilde{f}(x)\;\;\mbox{ in }\;\;B_{1},

where f~(x)=ρk(σα1)f(ρkx)\tilde{f}(x)=\rho^{k(\sigma-\alpha-1)}f(\rho^{k}x). Notice that for our choice of α\alpha in (2.2), we have f~Lp(B1)ε\|\tilde{f}\|_{L^{p}(B_{1})}\leq\varepsilon. Arguing similarly as in Proposition 3.2, we can also show that

|vk(x)|1+|x|1+α0.|v_{k}(x)|\leq 1+|x|^{1+\alpha_{0}}.

Hence, we can apply Proposition 5.1 to vkv_{k} to conclude that there exists ~k\tilde{\ell}_{k} such that

supBρ|vk(x)~k(x)|ρ1+α.\sup_{B_{\rho}}|v_{k}(x)-\tilde{\ell}_{k}(x)|\leq\rho^{1+\alpha}.

Rescaling back to uu, we obtain

supBρ|u(x)k+1(x)|ρ1+α,\sup_{B_{\rho}}|u(x)-\ell_{k+1}(x)|\leq\rho^{1+\alpha},

where k+1=k(x)+ρk(1+α)~k(ρ1x)\ell_{k+1}=\ell_{k}(x)+\rho^{k(1+\alpha)}\tilde{\ell}_{k}(\rho^{-1}x). From the definition of k+1\ell_{k+1}, it is immediate that condition (5.2) is also satisfied. This finishes the proof. ∎

Proof of Theorem 2.8.

The proof follows the same lines as in Theorem 1.1 and Theorem 1.2. Notice that from (5.2) we have that (Ak)k(A_{k})_{k\in\mathbb{N}} and (Bk)k(B_{k})_{k\in\mathbb{N}} are a Cauchy sequence, and hence, we can find AA_{\infty} and BB_{\infty} satisfying AkAA_{k}\to A_{\infty} and BkBB_{k}\to B_{\infty}, as kk\to\infty. Moreover, we have

|AkA|Cρk(1+α) and |BkB|Cρkα.|A_{k}-A_{\infty}|\leq C\rho^{k(1+\alpha)}\;\;\;\mbox{ and }\;\;\;|B_{k}-B_{\infty}|\leq C\rho^{k\alpha}.

Now, fix 0<r10<r\ll 1 and let kk\in\mathbb{N} be such that ρk+1rρk\rho^{k+1}\leq r\leq\rho^{k}. We have,

supBr|u(x)ABx|\displaystyle\sup_{B_{r}}|u(x)-A_{\infty}-B_{\infty}\cdot x| supBρk|u(x)AkBkx|+supBρk|AkA|+ρksupBρk|BkB|\displaystyle\leq\sup_{B_{\rho^{k}}}|u(x)-A_{k}-B_{k}\cdot x|+\sup_{B_{\rho^{k}}}|A_{k}-A_{\infty}|+\rho^{k}\sup_{B_{\rho^{k}}}|B_{k}-B_{\infty}|
ρk(1+α)+Cρk(1+α)+Cρk(1+α)\displaystyle\leq\rho^{k(1+\alpha)}+C\rho^{k(1+\alpha)}+C\rho^{k(1+\alpha)}
Cρ1+αρ(k+1)(1+α)\displaystyle\leq\frac{C}{\rho^{1+\alpha}}\rho^{(k+1)(1+\alpha)}
Cr1+α.\displaystyle\leq Cr^{1+\alpha}.

Finally, by taking the limit as kk\to\infty in (3.2), we obtain A=u(0)A_{\infty}=u(0). We can also show that B=Du(0)B_{\infty}=Du(0), see for instance [5]. This finishes the proof. ∎

6. The borderline case

This section deals with the CσC^{\sigma}-regularity for viscosity solutions of (1.9). As previously discussed, since 1+α=σ1+\alpha=\sigma, we can no longer follow the strategy employed above. In this case, we follow the ideas put forward in [13] (see also[18]). We begin with a technical lemma, that can be found in [13, Claim 3.2].

Lemma 6.1.

Let 0<α¯<α<10<\overline{\alpha}<\alpha<1 and uC1,α¯(B1)u\in C^{1,\overline{\alpha}}(B_{1}). If u(0)=|Du(0)|=0u(0)=|Du(0)|=0 and

sup0<r<1/2rα¯α[u]C1,α¯(Br)A,\sup_{0<r<1/2}r^{\overline{\alpha}-\alpha}[u]_{C^{1,\overline{\alpha}}(B_{r})}\leq A,

for a constant A>0A>0, then

[u]C1+α(B1/2)2A.[u]_{C^{1+\alpha}(B_{1/2})}\leq 2A.
Proposition 6.2.

Let uC1+α(B1)Lσ1(d)u\in C^{1+\alpha}(B_{1})\cap L^{1}_{\sigma}(\mathbb{R}^{d}), with α<σ1\alpha<\sigma-1, be a normalized viscosity solution to (1.9) with fBMO(B1)f\in{\rm BMO}(B_{1}). Suppose further that

uC1+α(B1)M and fBMO(B1)ε,\|u\|_{C^{1+\alpha}(B_{1})}\leq M\quad\mbox{ and }\quad\|f\|_{{\rm BMO}(B_{1})}\leq\varepsilon,

then uCσ(B1/2)u\in C^{\sigma}(B_{1/2}) and

uCσ(B1/2)C.\|u\|_{C^{\sigma}(B_{1/2})}\leq C.
Proof.

We argue by contradiction. Suppose that the result is false, then we can find sequences (uk)k,(fk)k(u_{k})_{k\in\mathbb{N}},(f_{k})_{k\in\mathbb{N}}, such that for every kk\in\mathbb{N},

(6.1) [uk]C1+α(B1)MfkBMO(B1)<εk,\displaystyle\begin{split}[u_{k}]_{C^{1+\alpha}(B_{1})}\leq M\\ \|f_{k}\|_{{\rm BMO}(B_{1})}<\varepsilon_{k},\end{split}

but

ukCσ(B1/2)>k,\|u_{k}\|_{C^{\sigma}(B_{1/2})}>k,

where εk0\varepsilon_{k}\to 0, as kk\to\infty. For kk\in\mathbb{N}, we define the quantity

θk(r)=supr<r<1/2supzB1/2r1+ασ[u]C1+α(Br(z))\theta_{k}(r^{\prime})=\sup_{r^{\prime}<r<1/2}\sup_{z\in B_{1/2}}r^{1+\alpha-\sigma}[u]_{C^{1+\alpha}(B_{r}(z))}

and observe that

limr0θk(r)=supr>0θk(r).\lim_{r^{\prime}\rightarrow 0}\theta_{k}(r^{\prime})=\sup_{r^{\prime}>0}\theta_{k}(r^{\prime}).

Moreover, if r1r2r_{1}\leq r_{2}, then θk(r2)θk(r1)\theta_{k}(r_{2})\leq\theta_{k}(r_{1}). By the Lemma 6.1 we have that

supr>0θk(r)k/2,\sup_{r^{\prime}>0}\theta_{k}(r^{\prime})\geq k/2,

therefore there exists a rk>1/kr_{k}>1/k and zkB1/2z_{k}\in B_{1/2} such that

(6.2) rk1+ασ[uk]C1+α(Brk(zk))>θk(1/k)>θk(rk).r_{k}^{1+\alpha-\sigma}[u_{k}]_{C^{1+\alpha}(B_{r_{k}}(z_{k}))}>\theta_{k}(1/k)>\theta_{k}(r_{k})\rightarrow\infty.

Since from (6.1) [uk]C1+α(B1)<M[u_{k}]_{C^{1+\alpha}(B_{1})}<M, we have rk0r_{k}\rightarrow 0 as kk\to\infty. Now, for R[1,12rk]R\in\left[1,\frac{1}{2r_{k}}\right], we define the blow-up vk:BRv_{k}:B_{R}\rightarrow\mathbb{R}

vk(x)=1θk(rk)1rkσuk(rkx+zk).v_{k}(x)=\frac{1}{\theta_{k}(r_{k})}\frac{1}{r_{k}^{\sigma}}u_{k}(r_{k}x+z_{k}).

Notice that

(6.3) [vk]C1+α(BR)=rk1+ασθk(rk)[uk]C1+α(BrkR(zk))=(Rrk)1+ασ[uk]BrkR(zk)θk(rk)Rσ1αRσ1α.\displaystyle\begin{split}[v_{k}]_{C^{1+\alpha}(B_{R})}&=\frac{r_{k}^{1+\alpha-\sigma}}{\theta_{k}(r_{k})}[u_{k}]_{C^{1+\alpha}(B_{r_{k}R}(z_{k}))}\\ &=\frac{(Rr_{k})^{1+\alpha-\sigma}[u_{k}]_{B_{r_{k}R}(z_{k})}}{\theta_{k}(r_{k})}R^{\sigma-1-\alpha}\\ &\leq R^{\sigma-1-\alpha}.\end{split}

Consider the auxiliary function wk:dw_{k}:\mathbb{R}^{d}\to\mathbb{R} defined as

wk(x):=(vklk)(x),w_{k}(x):=(v_{k}-l_{k})(x),

where

lk=vk(0)+Dvk(0)x.l_{k}=v_{k}(0)+Dv_{k}(0)x.

Since that [lk]C1+α(BR)=0[l_{k}]_{C^{1+\alpha}(B_{R})}=0, it follows from (6.3) that

(6.4) [wk]C1+α(BR)Rσ1α.[w_{k}]_{C^{1+\alpha}(B_{R})}\leq R^{\sigma-1-\alpha}.

In particular, for R=1R=1 we have

[Dwk]Cα(B1)1,[Dw_{k}]_{C^{\alpha}(B_{1})}\leq 1,

therefore

|Dwk(x)|=|Dwk(x)Dwk(0)||x|α, for xB1,|Dw_{k}(x)|=|Dw_{k}(x)-Dw_{k}(0)|\leq|x|^{\alpha},\quad\mbox{ for }\quad x\in B_{1},

which implies

(6.5) |wk(x)|=|wk(x)wk(0)|DwkL(B|x|)|x||x|1+α,|w_{k}(x)|=|w_{k}(x)-w_{k}(0)|\leq\|Dw_{k}\|_{L^{\infty}(B_{|x|})}|x|\leq|x|^{1+\alpha},

for all xB1x\in B_{1}. Let η\eta be a smooth function such that η=1\eta=1 in B1/2B_{1/2} and η=0\eta=0 outside B1B_{1}. For e𝕊ne\in\mathbb{S}^{n} we have

B1ηDewk𝑑x=B1Deηwk𝑑xC(n).\int_{B_{1}}\eta\cdot D_{e}w_{k}dx=\int_{B_{1}}D_{e}\eta\cdot w_{k}dx\leq C(n).

Hence, there exists zB1z\in B_{1} such that |Dwk(z)|C(n)|Dw_{k}(z)|\leq C(n) and by (6.4) we have

|Dewk(x)Dewk(z)|Rσ1α|xz|α.|D_{e}w_{k}(x)-D_{e}w_{k}(z)|\leq R^{\sigma-1-\alpha}|x-z|^{\alpha}.

Therefore, for xBRx\in B_{R} and 1R12rk1\leq R\leq\frac{1}{2r_{k}},

(6.6) |Dewk(x)|C(n)+Rσ1α|xz|αCRσ1.|D_{e}w_{k}(x)|\leq C(n)+R^{\sigma-1-\alpha}|x-z|^{\alpha}\leq CR^{\sigma-1}.

Our goal now is to show that

(6.7) [wk]Cβ(BR)Rσβ,[w_{k}]_{C^{\beta}(B_{R})}\leq R^{\sigma-\beta},

for all β[0,1+α]\beta\in[0,1+\alpha] and R[1,12rk]R\in\left[1,\frac{1}{2r_{k}}\right].

Case β=0\beta=0: Notice that for 1|x|R1\leq|x|\leq R, we have from 6.6

(6.8) |wk(x)|=|wk(x)wk(0)|Dewk(x)L(B|x|)|x|CRσ.\displaystyle\begin{split}|w_{k}(x)|&=|w_{k}(x)-w_{k}(0)|\\ &\leq\|D_{e}w_{k}(x)\|_{L^{\infty}(B_{|x|})}|x|\\ &\leq CR^{\sigma}.\end{split}

The estimate for xB1x\in B_{1} follows from (6.5).

Case β(0,1)\beta\in(0,1): In this case, we estimate for x,x¯BRx,\bar{x}\in B_{R}

|wk(x)wk(x¯)|\displaystyle|w_{k}(x)-w_{k}(\bar{x})| DwkL(BR)|xx¯|\displaystyle\leq\|Dw_{k}\|_{L^{\infty}(B_{R})}|x-\bar{x}|
Rσ1|xx¯|1β|xx¯|β\displaystyle\leq R^{\sigma-1}|x-\bar{x}|^{1-\beta}|x-\bar{x}|^{\beta}
CRσβ|xx¯|β,\displaystyle\leq CR^{\sigma-\beta}|x-\bar{x}|^{\beta},

which gives

(6.9) [wk]Cβ(BR)CRσβ.[w_{k}]_{C^{\beta}(B_{R})}\leq CR^{\sigma-\beta}.

Case β[1,1+α]\beta\in[1,1+\alpha]: Finally, we have

|Dwk(x)Dwk(x¯)|\displaystyle|Dw_{k}(x)-Dw_{k}(\bar{x})| [Dwk]Cα(R)|xx¯|α\displaystyle\leq[Dw_{k}]_{C^{\alpha}(R)}|x-\bar{x}|^{\alpha}
CRσ1αRαβ+1|xx¯|β1,\displaystyle\leq CR^{\sigma-1-\alpha}R^{\alpha-\beta+1}|x-\bar{x}|^{\beta-1},

which implies

[Dwk]Cβ1(BR)CRσβ,[Dw_{k}]_{C^{\beta-1}(B_{R})}\leq CR^{\sigma-\beta},

or equivalently

(6.10) [wk]Cβ(BR)CRσβ.[w_{k}]_{C^{\beta}(B_{R})}\leq CR^{\sigma-\beta}.

Therefore, (6.7) follows from (6.8), (6.9) and (6.10). Thus, there exists a wC1+α(d)w\in C^{1+\alpha}(\mathbb{R}^{d}) such that wkww_{k}\rightarrow w locally uniformly in the C1+αC^{1+\alpha}-norm. Now, from (6.2), we have that

[wk]C1+α(B1)12[w_{k}]_{C^{1+\alpha}(B_{1})}\geq\frac{1}{2}

and therefore

(6.11) [w]C1+α(B1)12.[w]_{C^{1+\alpha}(B_{1})}\geq\frac{1}{2}.

Moreover, from (6.7),

(6.12) [w]Cβ(d)Rσβ,[w]_{C^{\beta}(\mathbb{R}^{d})}\leq R^{\sigma-\beta},

for all β[0,1+α]\beta\in[0,1+\alpha] and R>1R>1. Notice that wkw_{k} solves

σ(wk,x)=f~k(x),{\mathcal{I}}_{\sigma}(w_{k},x)=\tilde{f}_{k}(x),

where f~k(x)=1θk(rk)fk(rkx+zk)\tilde{f}_{k}(x)=\dfrac{1}{\theta_{k}(r_{k})}f_{k}(r_{k}x+z_{k}), for pdp\geq d, we have

f~k(x)Lp(B1\displaystyle\|\tilde{f}_{k}(x)\|_{L^{p}(B_{1}} =\displaystyle= (B1|f~k(x)|p𝑑x)1/p\displaystyle\left(\int_{B_{1}}|\tilde{f}_{k}(x)|^{p}dx\right)^{1/p}
=\displaystyle= |B1|1/p( Brk(zk)|fk(x)|pdx)1/p\displaystyle|B_{1}|^{1/p}\left(\mathchoice{\mathop{\vrule width=6.0pt,height=3.0pt,depth=-2.5pt\kern-8.0pt\intop}\nolimits_{\kern-6.0ptB_{r_{k}}(z_{k})}}{\mathop{\vrule width=5.0pt,height=3.0pt,depth=-2.6pt\kern-6.0pt\intop}\nolimits_{B_{r_{k}}(z_{k})}}{\mathop{\vrule width=5.0pt,height=3.0pt,depth=-2.6pt\kern-6.0pt\intop}\nolimits_{B_{r_{k}}(z_{k})}}{\mathop{\vrule width=5.0pt,height=3.0pt,depth=-2.6pt\kern-6.0pt\intop}\nolimits_{B_{r_{k}}(z_{k})}}|f_{k}(x)|^{p}dx\right)^{1/p}
\displaystyle\leq |B1|1/pfkBMO(B1)|B1|1/pεk.\displaystyle|B_{1}|^{1/p}\|f_{k}\|_{\rm BMO(B_{1})}\leq|B_{1}|^{1/p}\varepsilon_{k}.

Hence,

0+(wk(+h)wk)\displaystyle{\mathcal{M}}^{+}_{{\mathcal{L}}_{0}}(w_{k}(\cdot+h)-w_{k}) σ(wk,x+h)σ(wk,x)\displaystyle\geq{\mathcal{I}}_{\sigma}(w_{k},x+h)-{\mathcal{I}}_{\sigma}(w_{k},x)
=1θk(rk)[fk(rk(x+h)+zk)fk(rkx+zk)].\displaystyle=\dfrac{1}{\theta_{k}(r_{k})}\left[f_{k}(r_{k}(x+h)+z_{k})-f_{k}(r_{k}x+z_{k})\right].

Moreover,

|wk(x+h)wk|[wk]Cα(BR)|h|αC|x|σα|w_{k}(x+h)-w_{k}|\leq[w_{k}]_{C^{\alpha}(B_{R})}|h|^{\alpha}\leq C|x|^{\sigma-\alpha}

and as [fk]BMO(B1)0[f_{k}]_{\rm BMO(B_{1})}\rightarrow 0 (and therefore fkLp(B1)0\|f_{k}\|_{L^{p}(B_{1})}\to 0), we have from Proposition 2.11 that

0+(w(+h)w)0 in d.{\mathcal{M}}^{+}_{{\mathcal{L}}_{0}}(w(\cdot+h)-w)\geq 0\;\;\ \mbox{ in }\;\;\mathbb{R}^{d}.

Similarly, we can prove that

0(w(+h)w)0 in d.{\mathcal{M}}^{-}_{{\mathcal{L}}_{0}}(w(\cdot+h)-w)\leq 0\;\;\ \mbox{ in }\;\;\mathbb{R}^{d}.

Therefore,

(6.13) 0(w(+h)w)00+(w(+h)w) in d.{\mathcal{M}}^{-}_{{\mathcal{L}}_{0}}(w(\cdot+h)-w)\leq 0\leq{\mathcal{M}}^{+}_{{\mathcal{L}}_{0}}(w(\cdot+h)-w)\;\;\ \mbox{ in }\;\;\mathbb{R}^{d}.

Finally, for

w~k=wk(+h)dμ(h)wk,\tilde{w}_{k}=\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int w_{k}(\cdot+h)d\mu(h)-w_{k},

we have that

|w~k||wk(+h)wk|dμ(h)C|x|σ1α.|\tilde{w}_{k}|\leq\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int|w_{k}(\cdot+h)-w_{k}|d\mu(h)\leq C|x|^{\sigma-1-\alpha}.

The concavity of σ{\mathcal{I}}_{\sigma} yields

0+(w~k(+h)dμ(h)wk,x)\displaystyle{\mathcal{M}}^{+}_{{\mathcal{L}}_{0}}\left(\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int\tilde{w}_{k}(\cdot+h)d\mu(h)-w_{k},x\right) σ(w~k(+h)dμ(h),x)σ(wk,x)\displaystyle\geq{\mathcal{I}}_{\sigma}\left(\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int\tilde{w}_{k}(\cdot+h)d\mu(h),x\right)-{\mathcal{I}}_{\sigma}(w_{k},x)
σ(wk,x+h)𝑑μ(h)f~k(x)\displaystyle\geq\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int{\mathcal{I}}_{\sigma}(w_{k},x+h)d\mu(h)-\tilde{f}_{k}(x)
=f~k(x+h)f~k(x)dμ(h).\displaystyle=\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int\tilde{f}_{k}(x+h)-\tilde{f}_{k}(x)d\mu(h).

Hence, by passing the limit as kk\to\infty we get

(6.14) 0+(w(+h)dμ(h)w,x)0.{\mathcal{M}}^{+}_{{\mathcal{L}}_{0}}\left(\mathchoice{{\vbox{\hbox{$\textstyle-$ }}\kern-7.83337pt}}{{\vbox{\hbox{$\scriptstyle-$ }}\kern-5.90005pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.75003pt}}{{\vbox{\hbox{$\scriptscriptstyle-$ }}\kern-4.25003pt}}\!\int w(\cdot+h)d\mu(h)-w,x\right)\geq 0.

Therefore, from (6.12), (6.13) and (6.14), we can apply Theorem [13, Theorem 2.1], to conclude that ww is a polynomial of degree 1, which is a contradiction with (6.11).

Proof of Theorem 2.9.

Recall that by Remark 2.5 we can assume fBMO(B1)ε\|f\|_{BMO(B_{1})}\leq\varepsilon, where ε>0\varepsilon>0 is the constant from the previous lemma. Since fLp(B1)CfBMO(B1)\|f\|_{L^{p}(B_{1})}\leq C\|f\|_{\rm BMO(B_{1})}, we can use Theorem 2.8 to conclude

uC1,α(B1/2)C(uL(B1)+uLσ1(d)+fBMO(B1)).\|u\|_{C^{1,\alpha}(B_{1/2})}\leq C(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{{\rm BMO}(B_{1})}).

Therefore from Proposition 6.2 we have

uCσ(B1/2)C(uL(B1)+uLσ1(d)+fBMO(B1)),\|u\|_{C^{\sigma}(B_{1/2})}\leq C(\|u\|_{L^{\infty}(B_{1})}+\|u\|_{L^{1}_{\sigma}(\mathbb{R}^{d})}+\|f\|_{BMO(B_{1})}),

proving the result. ∎


Acknowledgement: D. dos Prazeres was partially supported by CNPq and CAPES/Fapitec. M. Santos was partially supported by the Portuguese government through FCT-Fundação para a Ciência e a Tecnologia, I.P., under the projects UID/MAT/04459/2020, and PTDC/MAT-PUR/1788/2020. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brazil (CAPES) - Finance Code 001.

References

  • [1] G. Barles, E. Chasseigne, C. Georgelin, and E. Jakobsen. On neumann type problems for nonlocal equations set in a half space. arXiv preprint arXiv:1112.0476, 2011.
  • [2] Guy Barles. An Introduction to the Theory of Viscosity Solutions for First-Order Hamilton–Jacobi Equations and Applications, pages 49–109. Springer Berlin Heidelberg, Berlin, Heidelberg, 2013.
  • [3] Guy Barles, Emmanuel Chasseigne, and Cyril Imbert. Hölder continuity of solutions of second-order non-linear elliptic integro-differential equations. J. Eur. Math. Soc. (JEMS), 13(1):1–26, 2011.
  • [4] Guy. Barles and Benoît Perthame. Discontinuous solutions of deterministic optimal stopping time problems. RAIRO Modél. Math. Anal. Numér., 21(4):557–579, 1987.
  • [5] Anne Bronzi, Edgard Pimentel, Giane Rampasso, and Eduardo Teixeira. Regularity of solutions to a class of variable-exponent fully nonlinear elliptic equations. J. Funct. Anal., 279(12):108781, 31, 2020.
  • [6] Luis Caffarelli and Luis Silvestre. Regularity theory for fully nonlinear integro-differential equations. Comm. Pure Appl. Math., 62(5):597–638, 2009.
  • [7] Luis Caffarelli and Luis Silvestre. The Evans-Krylov theorem for nonlocal fully nonlinear equations. Ann. of Math. (2), 174(2):1163–1187, 2011.
  • [8] Luis Caffarelli and Luis Silvestre. Regularity results for nonlocal equations by approximation. Arch. Ration. Mech. Anal., 200(1):59–88, 2011.
  • [9] Nestor Guillen and Russell W. Schwab. Aleksandrov-Bakelman-Pucci type estimates for integro-differential equations. Arch. Ration. Mech. Anal., 206(1):111–157, 2012.
  • [10] Shuhei Kitano. ABP maximum principles for fully nonlinear integro-differential equations with unbounded inhomogeneous terms. Partial Differ. Equ. Appl., 1(4):Paper No. 16, 11, 2020.
  • [11] Shuhei Kitano. Wσ,pW^{\sigma,p} a priori estimates for fully nonlinear integro-differential equations. Calc. Var. Partial Differential Equations, 61(4):Paper No. 153, 19, 2022.
  • [12] Dennis Kriventsov. C1,αC^{1,\alpha} interior regularity for nonlinear nonlocal elliptic equations with rough kernels. Comm. Partial Differential Equations, 38(12):2081–2106, 2013.
  • [13] Joaquim Serra. cσ+αc^{\sigma+\alpha} regularity for concave nonlocal fully nonlinear elliptic equations with rough kernels. Calculus of Variations and Partial Differential Equations, 54:3571–3601, 2014.
  • [14] Joaquim Serra. Regularity for fully nonlinear nonlocal parabolic equations with rough kernels. Calc. Var. Partial Differential Equations, 54(1):615–629, 2015.
  • [15] Luis Silvestre. Hölder estimates for solutions of integro-differential equations like the fractional Laplace. Indiana Univ. Math. J., 55(3):1155–1174, 2006.
  • [16] E. Teixeira. Universal moduli of continuity for solutions to fully nonlinear elliptic equations. Arch. Ration. Mech. Anal., 211(3):911–927, 2014.
  • [17] H. Vivas. c2sc^{2s} regularity for fully nonlinear nonlocal equations with bounded right hand side. ArXiv preprint arXiv:1907.02455, 2019.
  • [18] Hui Yu. Small perturbation solutions for nonlocal elliptic equations. Communications in Partial Differential Equations, 42(1):142–158, 2017.

Makson S. Santos
Departamento de Matemática do Instituto Superior Técnico
Universidade de Lisboa
1049-001 Lisboa, Portugal
[email protected]

Disson dos Prazeres
Department of Mathematics
Universidade Federal de Sergipe - UFS,
49100-000, Jardim Rosa Elze, São Cristóvão - SE, Brazil
[email protected]