Abstract
We study robust regularity estimates for local minimizers of nonlocal functionals with non-standard growth of (p, q)-type and for weak solutions to a related class of nonlocal equations. The main results of this paper are local boundedness and Hölder continuity of minimizers and weak solutions. Our approach is based on the study of corresponding De Giorgi classes.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The aim of this paper is to prove regularity properties of local minimizers and weak solutions to a class of nonlocal problems with non-standard growth.
Let \(s \in (0,1)\), \(\Lambda \ge 1\) and \(\Omega \subset {\mathbb {R}}^d\) be an open set. We study energy functionals of the form
where \(f : [0,\infty ) \rightarrow [0,\infty )\) is a convex increasing function and \(k: {\mathbb {R}}^d\times {\mathbb {R}}^d\rightarrow {\mathbb {R}}\) is a measurable function satisfying
When \(f(t) = t^p\) with \(p>1\) and \(\Lambda =1\), the functional (1.1) becomes the standard fractional p-functional whose corresponding operator is the fractional p-Laplacian. The regularity theory for this case is well established, see [17, 19, 20].
Apparently the functional (1.1) is governed by the function f, which controls the growth behavior for large and small values of \(|u(x)-u(y)||x-y|^{-s}\). To establish the regularity theory, we need some growth conditions on f.
Let \(1 \le p \le q\). We say that f satisfies (\(f_{p}^{q}\)) if f is differentiable and satisfies for all \(t\ge 0\)
Condition (\(f_{p}^{q}\)) can be interpreted as a (p, q)-growth condition since it implies that
For a detailed discussion of (\(f_{p}^{q}\)) we refer to Sect. 2.
To study local boundedness of minimizers, we work under the assumption that there exists a constant \(c_0 > 0\) such that for all \(t \ge 0\)
Throughout the paper, we will assume without loss of generality that \(f(0)=0\) and \(f(1)=1\). Note that the assumption \(f(0)=0\) is required in order for \({\mathcal {I}}_f(0)<\infty \), while the second assumption is not restrictive since u minimizes \({\mathcal {I}}_{f}\) if and only if u minimizes \({\mathcal {I}}_{f/f(1)}\).
In the following, we present the first main result of this paper. It is concerned with Hölder estimates and local boundedness for local minimizers of (1.1).
Theorem 1.1
(Local minimizers) Let \(s_0 \in (0,1)\), \(1 < p \le q\), \(\Lambda \ge 1\), \(c_0 > 0\) and assume \(s \in [s_0, 1)\). Let \(f: [0,\infty ) \rightarrow [0,\infty )\) be a convex increasing function satisfying \({(f^{q})}\) and let \(k: {\mathbb {R}}^d\times {\mathbb {R}}^d\rightarrow {\mathbb {R}}\) be a measurable function satisfying (k). Let \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) be a local minimizer of (1.1).
-
(i)
Assume that f satisfies \(({f_{p}})\). Then, there exist \(\alpha \in (0,1)\) and \(C > 0\), depending on d, \(s_0\), p, q and \(\Lambda \), such that for any \(B_{8R}(x_0) \subset \Omega \)
$$\begin{aligned} R^\alpha [u]_{C^{\alpha }(\overline{B_R(x_0)})} \le C \Vert u\Vert _{L^{\infty }(B_{4R}(x_0))} + \textrm{Tail}_{f'}(u; x_0, 4R). \end{aligned}$$(1.3) -
(ii)
Assume that \(sp<d\), \(q < p^{*}:=dp/(d-sp)\) and that f satisfies \(({f > rsim t^p})\). Then, \(u \in L^{\infty }_{\textrm{loc}}(\Omega )\). Moreover, for each \(B_{2R}(x_0) \subset \Omega \) there exists \(C > 0\), depending on d, \(s_0\), p, q, \(p^{*}-q\), \(\Lambda \), \(c_0\) and R, such that for every \(\delta \in (0,1)\)
(1.4)
We refer to Sect. 3 for the definition of the function space \(V^{s, f}(\Omega |{\mathbb {R}}^d)\) and the tail term \(\textrm{Tail}_{f'}\). The proof of Theorem 1.1 and the definition of a local minimizer are given in Sect. 6.
The second main result of this paper is concerned with weak solutions to a related class of nonlocal equations. To motivate our result, we first point out that the Euler–Lagrange equation corresponding to the functional (1.1) is given by
For convex differentiable functions f, it is easy to see that weak solutions to the Euler–Lagrange equation are minimizers of the functional (1.1). In this article, we consider a more general class of equations
with nonlocal operators of the form
where \(h : {\mathbb {R}}^d\times {\mathbb {R}}^d\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) is a measurable function satisfying the structure condition
for a.e. \(x, y \in {\mathbb {R}}^d\) and for all \(t \in {\mathbb {R}}\), and \(f : [0,\infty ) \rightarrow [0,\infty )\) is convex, increasing, differentiable and satisfies \(f(0) = 0\), \(f(1) = 1\).
Note that in the special case \(h(x,y,t) = \text {sign}(t)f'(|t|)k(x,y)\) for some k satisfying (k), the Eqs. (1.5) and (1.6) coincide.
We are ready to state the second main result of this article, which establishes Hölder estimates and local boundedness for weak solution to (1.6).
Theorem 1.2
(Weak solutions) Let \(s_0 \in (0,1)\), \(1 < p \le q\), \(\Lambda \ge 1\), \(c_0 > 0\) and assume \(s \in [s_0, 1)\). Let \(f: [0,\infty ) \rightarrow [0,\infty )\) be a convex increasing function satisfying \({(f^{q})}\) and let \(h: {\mathbb {R}}^d\times {\mathbb {R}}^d\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) be a measurable function satisfying (h). Let \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) be a weak solution to (1.6).
-
(i)
Assume that f satisfies \(({f_{p}})\). Then, there exist \(\alpha \in (0,1)\) and \(C > 0\), depending on d, \(s_0\), p, q and \(\Lambda \), such that for any \(B_{8R}(x_0) \subset \Omega \) the estimate (1.3) holds.
-
(ii)
Assume that \(sp < d\), \(q < p^{*}\) and that f satisfies \(({f > rsim t^p})\). Then, \(u \in L^{\infty }_{\textrm{loc}}(\Omega )\). Moreover, for each \(B_{2R}(x_0) \subset \Omega \) there exists \(C > 0\), depending on d, \(s_0\), p, q, \(p^{*}-q\), \(\Lambda \), \(c_0\) and R, such that for every \(\delta \in (0,1)\) the estimate (1.4) holds.
The proof of Theorem 1.2 and the definition of a weak solution are given in Sect. 7. Note that we can assume without loss of generality that \(f(0) = 0\), \(f(1) = 1\) because a function u solves \({\mathcal {L}}_h u = 0\) if and only if it solves \({\mathcal {L}}_{h/f(1)} u = 0\).
In fact, we prove Hölder estimates and local boundedness for functions in De Giorgi classes (see Sect. 3). The corresponding results, Theorems 4.2 and 5.1, are more general. Theorems 1.1 and 1.2 follow by the observation that minimizers, as well as weak solutions, belong to the corresponding De Giorgi classes, see Sects. 6 and 7.
Note that right-hand sides with suitable growth behaviors can be studied by an adequate extension of De Giorgi classes in analogy to [17].
Remark 1.3
Our results are robust in the sense that the constants C and \(\alpha \) stay uniform as \(s \rightarrow 1^-\), since they depend only on \(s_0\), not on s.
-
(1)
Theorem 1.1 generalizes the results in [47] to nonlocal functionals. We work under the same assumptions on the growth function f as in that article. In this sense, the assumptions on f used in our results are natural.
-
(2)
Theorem 1.2 can be linked to the paper [40]. Some assumptions on the regularity and growth of the function f in [40] are more restrictive than our assumptions, but in return allow Lieberman to prove \(C^{1,\beta }\) regularity of weak solutions to the Euler–Lagrange equations.
Our approach for studying local minimizers and weak solutions is based on so-called De Giorgi classes. We show that minimizers of (1.1) and weak solutions to (1.6) satisfy a suitable improved fractional Caccioppoli inequality, from which the definition of the De Giorgi class emerges. This inequality together with an isoperimetric-type inequality allow us to deduce the Hölder estimates for locally bounded minimizers and weak solutions following the methods from [17, 47]. We emphasize that there is no restriction on the gap between p and q for the Hölder estimates. Furthermore, we derive the local boundedness of functions in De Giorgi classes under the assumption that \(1<p\le q<p^{*}\). We would like to remark that by modification of the proof of Theorem 5.1, it is possible to prove the local boundedness without any restriction on the range of p and q and without the condition \(({f > rsim t^p})\), see [16] and also [5].
In the following, we discuss related literature and describe the novelty of our results.
We first comment on related results for local operators. For this purpose, we consider functionals of the following form
where f is a non-negative function which describes the growth behavior of the functional. If the function f satisfies the so-called p-growth condition, that is
the literature is very rich and many regularity results have been proved. We refer the reader to the classical references [31, 33, 41] and for a more comprehensive treatment to the books [34, 35].
Functionals with non-standard growth of (p, q)-type
where \(1<p<q\), are naturally connected to Orlicz–spaces. The analysis of regularity of minimizers of functionals having non-standard growth of (p, q)-type was initiated by Marcellini’s work [42], where he studies strictly convex \(C^2\)-functions f satisfying (p, q)-growth condition.
To the best of our knowledge, functionals with non-standard growth functions of the type (\(f_{p}^{q}\)) first appeared in the papers [40, 47] in the context of regularity results.
In the paper [40], Lieberman proves several regularity results for bounded weak solutions to a class of elliptic operators in divergence form. Furthermore, he studies quasiminimizers and proves regularity results for functions in corresponding De Giorgi classes.
In [47], Moscariello and Nania prove Hölder continuity of locally bounded minimizers for growth functions satisfying (\(f_{p}^{q}\)) and local boundedness for functions with (p, q)-growth. Their key idea is to introduce an auxiliary function which is comparable to the growth function f and to prove that any function in the De Giorgi class corresponding to the auxiliary function is Hölder continuous. We adapt this idea for the proof of Theorems 1.1 and 1.2 to the nonlocal case, see Sect. 4.
There have been many important contributions to regularity for problems with non-standard growth of (p, q)-type. Papers studying local operators with non-standard growth of (p, q)-type are, among others, [3, 6, 12, 14, 23, 26, 28, 39, 43, 44, 48, 49, 54, 55, 58]. For a more detailed picture on problems with non-standard growth, including double-phase problems, problems with variable exponents, and anisotropic problems, we refer the reader to the surveys [45, 50].
In the case of nonlocal operators, the energy functional is defined by (1.1). Local regularity results for the fractional p-Laplacian, that is \(f(t)=t^p\), were first established in the papers [19, 20] by Di Castro, Kuusi and Palatucci. Another important contribution to regularity is the work [17], where he studies minimizers to nonlocal energy functionals plus a possibly discontinuous potential. The nonlocal energy has p growth for a class of symmetric kernels comparable to \((1-s)|x-y|^{-n-sp}\). Furthermore, he studies weak solutions to the Euler–Lagrange equation. He uses the nonlocality of the functional to prove an improved Caccioppoli inequality with an additional term, which disappears as the fractional order s goes to one. We follow Cozzi’s ideas at several points in the present paper and also make use of some auxiliary results proved in [17] such as an isoperimetric inequality. It is worth emphasizing that the fractional De Giorgi iteration has been first employed in the papers [18, 46]. For further results on the regularity of the fractional p-Laplacian, we refer the reader to [7, 8, 53] and the references therein.
Lately, the interest in the analysis of nonlocal problems with non-standard growth has increased. For instance, regularity results for nonlocal double phase equations and nonlocal equation with variable exponents are proved in [10, 21, 32, 57], respectively [15, 56]. However, we would like to note that both, double phase equations and equations with variable exponents, do not fall into our setup. See also [36,37,38] for further regularity results concerning nonlocal operators with non-standard growth.
As far as we know, first regularity results for fractional order Orlicz–Sobolev spaces have been proved in [13] by Fernández Bonder, Salort and Vivas. The authors establish regularity results for weak solutions to the Dirichlet problem for the fractional g-Laplacian. They prove interior and up to the boundary Hölder regularity to the corresponding Dirichlet problem.
See also [52], where qualitative properties of solutions such as a Liouville type theorem and symmetry results are proved.
The present paper is substantially different from [13]. On the one hand, we do not only study weak solutions but also local minimizers for the functional \({\mathcal {I}}_f\). The present paper also allows for \(p>1\) and is not restricted to the case \(p\ge 2\). Furthermore, we use a completely different approach via De Giorgi classes.
1.1 Notation
We write c and C for strictly positive constants whose exact values are not important and might change from line to line. Furthermore, we use the notations \(c=c(\cdot )\) and \(C=C(\cdot )\) if we want to highlight all quantities the constant depends on.
1.2 Outline
The paper consists of seven sections and is organized as follows. In Sect. 2.1, we introduce an auxiliary growth function and prove several properties for convex functions satisfying (\(f_{p}^{q}\)). Moreover, in Sect. 2.2 we recall some functional inequalities. The fractional De Giorgi classes with general convex functions with non-standard growth are introduced in Sect. 3. Furthermore, we introduce fractional Orlicz–Sobolev spaces. In Sect. 4 we prove Hölder continuity and in Sect. 5 local boundedness for functions in fractional De Giorgi classes. Finally, in Sect. 6 resp. Section 7, we show that minimizers resp. weak solutions belong to the fractional De Giorgi classes and prove Theorems 1.1 and 1.2.
2 Preliminaries
In this section, we study properties of the growth functions f under consideration and collect some functional inequalities.
2.1 Auxiliary results on growth functions
Let us collect several results in order to illustrate the assumption (\(f_{p}^{q}\)). The first two lemmas provide equivalent conditions for the upper and lower bounds in (\(f_{p}^{q}\)), respectively.
Lemma 2.1
Let \(q \ge 1\) and \(f : [0,\infty ) \rightarrow [0,\infty )\) be a differentiable function. Then the following are equivalent:
-
(i)
\({(f^{q})}\),
-
(ii)
\(t \mapsto t^{-q}f(t)\) is decreasing,
-
(iii)
\(f(\lambda t) \le \lambda ^q f(t)\) for all \(\lambda \ge 1\),
-
(iv)
\(\lambda ^q f(t) \le f(\lambda t)\) for all \(\lambda \le 1\).
Proof
(i) \(\Leftrightarrow \) (ii) follows from the observation that \(\frac{\text {d}}{\text {d}t}(t^{-q}f(t)) = t^{-q-1}(t f'(t) - q f(t))\). (i) \(\Leftrightarrow \) (iii) follows from the observation that (iii) can be rewritten as
Since also (iv) can be rewritten as (2.1), the equivalence (iii) \(\Leftrightarrow \) (iv) is trivial. \(\square \)
Lemma 2.2
Let \(p \ge 1\) and \(f : [0,\infty ) \rightarrow [0,\infty )\) be a differentiable function. Then the following are equivalent:
-
(i)
\(({f_{p}})\),
-
(ii)
\(t \mapsto t^{-p}f(t)\) is increasing,
-
(iii)
\(\lambda ^p f(t) \le f(\lambda t)\) for all \(\lambda \ge 1\),
-
(iv)
\(f(\lambda t) \le \lambda ^p f(t)\) for all \(\lambda \le 1\).
Proof
The proof works exactly like the proof of Lemma 2.1. \(\square \)
The following lemma provides a useful property of convex functions.
Lemma 2.3
Let \(f : [0,\infty ) \rightarrow [0,\infty )\) be convex and \(f(0) = 0\). Then, the function \(t \mapsto f(t)/t\) is increasing. If f is differentiable, then f satisfies \(({f_{p}})\) with \(p=1\).
Proof
The assertions follow from \(f(\lambda t) = f(\lambda t + (1-\lambda )0) \le \lambda f(t) + (1-\lambda ) f(0)\) and Lemma 2.2 with \(p=1\). \(\square \)
As a consequence we obtain some doubling-type inequalities for \(f'\). These inequalities play an important role for the tail estimates in the upcoming regularity theory.
Corollary 2.4
Let \(1 \le p \le q\) and let \(f : [0,\infty ) \rightarrow [0,\infty )\) be a function satisfying (\(f_{p}^{q}\)). Then,
and
for all \(t,s \ge 0\).
Proof
For the second inequality in (2.2), we compute using (\(f_{p}^{q}\)) and Lemma 2.1
The first inequality in (2.2) and (2.3) can be proved in the same way. The first estimate in (2.4) is a direct consequence of monotonicity of \(f'\). For the second estimate in (2.4), we may assume that \(t \le s\). Then, we obtain
by using (\(f_{p}^{q}\)), Lemmas 2.1 and 2.2. \(\square \)
Another useful property of convex functions is the following:
Lemma 2.5
Let \(f : [0,\infty ) \rightarrow [0,\infty )\) be convex and \(f(0)=0\). Let \(c > 1\) and assume that for some \(t,s > 0\) it holds that \(f(t) \le c f(s)\). Then \(t \le cs\).
Proof
Let \(t,s > 0\) be such that \(f(t) \le c f(s)\). We assume that \(t > cs\). Then by Lemmas 2.3 and 2.2 with \(p=1\), we have
This is a contradiction, so it must hold that \(t \le cs\), as desired. \(\square \)
One of the key ideas of proving Hölder regularity in [47] is to construct F, which is a convex increasing function satisfying some growth conditions and the comparability of \(g(t) := F(t^p)\) and f(t). These properties are important in our framework as well but we also need the comparability of the derivatives of these functions for the regularity estimates.
Proposition 2.6
(c.f. [30]) Let \(1 \le p \le q\) and \(f : [0,\infty ) \rightarrow [0,\infty )\) be a convex, increasing, and differentiable function satisfying \(f(0)=0\). Define \(F, g : [0,\infty ) \rightarrow [0,\infty )\) by
If f satisfies (\(f_{p}^{q}\)), then F is a convex increasing function satisfying
and g is an increasing function satisfying
Proof
First of all, the function F is well-defined by Lemma 2.3. The functions F and g are increasing by definition. Moreover, F is convex since
by \(({f_{p}})\). Thus, the first inequality in (2.6) follows from Lemma 2.3. The second inequality follows from
for \(\lambda \ge 1\), where we used Lemma 2.1. By (\(f_{p}^{q}\)) we have
and after integrating from 0 to t and using that \(f(0) = 0\), we deduce (2.7). Finally, we compute
using (\(f_{p}^{q}\)) from where (2.8) follows. \(\square \)
We close this subsection with two estimates for convex functions. Lemmas 2.7 and 2.8 are generalizations of [17, Lemma 4.1 and 4.2].
Lemma 2.7
Let \(f : [0,\infty ) \rightarrow [0,\infty )\) be convex, differentiable and \(f(0)=0\). Then, for any \(\theta \in [0,1]\) and \(a, b \ge 0\):
Proof
The result is clear for \(\theta = 0\) by the superadditivity of convex functions with \(f(0)=0\). For \(\theta = 1\), we compute
where we used the fact that \(t \mapsto f'(t)\) is increasing since f is convex. The result for \(\theta \in (0,1)\) follows by interpolation. \(\square \)
Lemma 2.8
Let \(f : [0,\infty ) \rightarrow [0,\infty )\) be convex, increasing and differentiable. Then, for every \(\mu \in [0,1]\) and \(a, b \ge 0\):
Proof
Let us first assume that \(b \ge a\). Then
If on the other hand \(\mu a \le b < a\), then
It remains to consider the case \(b < \mu a\), but then \(f(\vert \mu a-b \vert ) - f(\vert a-b \vert ) \le 0\), so there is nothing to prove. \(\square \)
2.2 Functional inequalities
In this section, we collect some well-known functional inequalities which are useful for the application of De Giorgi’s methods for nonlocal operators. While the first three results are embeddings for fractional Sobolev spaces, the last proposition is a fractional isoperimetric inequality.
Lemma 2.9
[17, Lemma 4.6] Let \(0< {\tilde{\sigma }}< \sigma < 1\) and \(1 \le {\tilde{p}} < p\). Let \(\Omega ' \subset \Omega \subset {\mathbb {R}}^d\) be two bounded measurable sets, then for any \(u \in W^{\sigma , p}(\Omega )\)
where
Theorem 2.10
[17, Corollary 4.9] Let \(0< s_0 \le s < 1\) and \(p \ge 1\) be such that \(sp < d\). Let \(u \in W^{s, p}_{0}(B_R)\) and assume \(u=0\) on a set \(\Omega \subset B_R\) with \(|\Omega | \ge \gamma |B_R|\) for some \(\gamma \in (0,1]\). Then, there exists a constant \(C > 0\), depending on d, \(s_0\), p and \(\gamma \), such that
Theorem 2.11
[4, 25, 51] Let \(0< s_0 \le s < 1\) and \(p \ge 1\) be such that \(sp < d\). Let \(u \in W^{s, p}(B_R)\). Then, there exists a constant \(C > 0\), depending on d, \(s_0\) and p, such that
Proposition 2.12
[17, Proposition 5.1] Let \(p > 1\), \(C_0 > 0\) and \(\gamma , \gamma _0 \in (0,1)\). Then, there exist constants \({\bar{s}} \in (0,1)\) and \(C > 0\), depending on d, p, \(\gamma \), \(\gamma _0\) and \(C_0\), such that if \(s \in [{\bar{s}},1)\) and if \(u \in W^{s, p}(B_R)\) satisfies
then
3 De Giorgi classes
In this section, we introduce fractional order Orlicz–Sobolev spaces and define fractional De Giorgi classes governed by convex functions having non-standard growth.
Fractional order Orlicz–Sobolev spaces have been introduced in [29] by Fernández Bonder and Salort. The authors prove several properties of the spaces including that the fractional order Orlicz–Sobolev space approximates some Orlicz–Sobolev space as the fractional parameter goes to 1. For further results concerning fractional Orlicz–Sobolev spaces, we refer the reader to [1, 2, 9, 11, 24] and the references therein.
Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying \({(f^{q})}\) and \(f(0)=0\). Let \(s \in (0, 1)\) and \(\Omega \subset {\mathbb {R}}^d\) be open. We define the Orlicz and Orlicz–Sobolev spaces by
where \(\Phi _{L^f(\Omega )}\), \(\Phi _{W^{s, f}(\Omega )}\) and \(\Phi _{V^{s,f}(\Omega |{\mathbb {R}}^d)}\) are modulars defined by
Under more restrictive assumptions on f (for instance f being an Orlicz function), \(L^f(\Omega )\), \(W^{s, f}(\Omega )\), and \(V^{s, f}(\Omega |{\mathbb {R}}^d)\) are Banach spaces endowed with the norms
Let us next define nonlocal tails, which capture the behavior of functions \(u\in V^{s, f}(\Omega |{\mathbb {R}}^d)\) at large scales. For this purpose, we consider a convex and differentiable function f and a generalized inverse function of its derivative \(f'\). There are several definitions of generalized inverse functions in the literature, see [22, 27], but we use the following definition for a generalized inverse of \(f'\):
The advantage of this definition is that (3.1) enjoys the following properties, which play an important role in the proof of regularity estimates.
Proposition 3.1
Let \(f : [0, \infty ) \rightarrow [0, \infty )\) be a convex and differentiable function. Then
Proof
To prove (3.2), let \(y \ge 0\) and \(t = (f')^{-1}(y)\). Then, there exists a sequence \((t_n)_{n\ge 1}\) such that \(t_n \ge t\), \(t_n \rightarrow t\), and \(f'(t_n) \ge y\). Since \(f'\) is continuous by Darboux’s theorem, we obtain \(f'(t) \ge y\) by taking the limit \(n \rightarrow \infty \). Assertion (3.3) is obvious by definition of \((f')^{-1}\). \(\square \)
We define the nonlocal \(f'\)-Tail by
Note that \(\textrm{Tail}_{f'}\) coincides with the standard tail considered in [17, 19, 20] in the special case \(f(t) \asymp t^p\). Moreover, it is natural in the sense that the following scaling property is satisfied:
We claim that the nonlocal \(f'\)-Tail is well-defined for functions in the fractional Orlicz–Sobolev space \(V^{s, f}(\Omega |{\mathbb {R}}^d)\). To this end, we define the Legendre transform \(f^{*}: [0, \infty ) \rightarrow [0, \infty )\) by \(f^{*}(s) = \sup _{t \in [0, \infty )} (st-f(t))\). It is well known that
and that
Inequality (3.6) is called Fenchel’s inequality.
Proposition 3.2
Let \(q >1\), \(s \in (0,1)\) and \(\Omega \subset {\mathbb {R}}^d\) be open. Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying \({(f^{q})}\) and \(f(0) = 0\). If \(u \in V^{s, f}(\Omega |{\mathbb {R}}^d)\) and \(B_R(x_0) \subset \Omega \), then \(\textrm{Tail}_{f'}(u; x_0, R) < \infty \).
Proof
Let \(u \in V^{s, f}(\Omega |{\mathbb {R}}^d)\) and \(B_R(x_0) \subset \Omega \). We may assume that \(x_0 = 0\). In order to prove finiteness of the tail, it is sufficient to show that
Since \(B_R \subset \Omega \) and \(|x-y|\le 2|y|\) for \(x\in B_{R}\), \(y\in B_{R}^c\), we have
by using \({(f^{q})}\) and Lemma 2.1. Our aim is to estimate the right-hand side of (3.8) from below by the expression in (3.7). Note that by Lemmas 2.3, 2.2 and \({(f^{q})}\): \( f(|u(x)|)(|y|^{-s} \vee |y|^{-sq} ) \ge f(|y|^{-s}|u(x)|)\). Therefore,
where we also used \(f(t + s) \le 2^q(f(t) + f(s))\). This is a direct consequence of monotonicity and \({(f^{q})}\). Since \(\int _{B_R^c} |y|^{-d}(|y|^{-s} \vee |y|^{-sq}) \,\textrm{d}y< \infty \), we estimate the right-hand side of (3.8) by
Altogether,
Note that we have \((f^{*}(f'))(t) \le (q-1) f(t)\) by (3.5) and \({(f^{q})}\). Therefore, we obtain
where we used Fenchel’s inequality (3.6). Combining the previous estimate with (3.9) finishes the proof. \(\square \)
Having defined the fractional order Orlicz–Sobolev spaces, we are ready to introduce De Giorgi classes that are suitable to our setting.
Definition 3.3
Let \(q > 1\), \(c > 0\), \(s \in (0,1)\), and let \(\Omega \) be open. Let \(f : [0,\infty ) \mapsto [0,\infty )\) be convex and differentiable with \(f(0)=0\), \(f(1) = 1\). We say that \(u \in G_+(\Omega ; q, c, s, f)\) if \(u \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) and if for every \(x_0 \in \Omega \), \(0< r < R \le d(x_0, \partial \Omega )\), \(k \in {\mathbb {R}}\), it holds
where \(w_{\pm }(x) = (u(x)- k)_{\pm }\) and \(A_{k}^{-} = \lbrace y \in {\mathbb {R}}^d: u(y) < k \rbrace \). We say that \(u \in G_-(\Omega ; q, c, s, f)\) if (3.10) holds true with \(w_+\), \(w_-\) and \(A_{k}^{-}\) replaced by \(w_-\), \(w_+\) and \(A_{k}^{+} = \lbrace y \in {\mathbb {R}}^d: u(y) > k \rbrace \), respectively. Moreover, we denote by \(G(\Omega ; q, c, s, f) = G_+(\Omega ; q, c, s, f) \cap G_-(\Omega ; q, c, s, f)\).
The De Giorgi classes under consideration will contain minimizers of (1.1) and weak solutions for (1.7) under suitable additional assumptions on f, see Theorems 6.2 and 7.3.
The following proposition allows us to infer Hölder regularity of minimizers of \({\mathcal {I}}_f\) from regularity of functions in \(G(\Omega ; q, c, s, F(\cdot ^p))\), where F is as in Proposition 2.6.
Proposition 3.4
Let \(1 < p \le q\), \(s \in (0,1)\) and \(\Omega \subset {\mathbb {R}}^d\) be open. Let \(f : [0,\infty ) \mapsto [0,\infty )\) be convex and increasing with (\(f_{p}^{q}\)). Then for every \(c_1 > 0\), there exists \(c_2 = c_2(c_1, p, q) > 0\) such that \(G_{\pm }(\Omega ; q, c_1, s, f) \subset G_{\pm }(\Omega ; q, c_2, s, F(\cdot ^p))\).
Proof
The proof follows directly from (2.7) and (2.8). \(\square \)
4 Hölder estimate
In this section, we prove Hölder estimates for functions in the De Giorgi class \(G(\Omega ; q, c, s, g)\), where g is the function given by (2.5). Let us first prove a growth lemma for functions in \(G_-(\Omega ; q, c, s, g)\).
Theorem 4.1
Let \(1<p\le q\), \(c, H > 0\), \(R>0\), \(s_0\in (0,1)\) and assume \(s \in [s_0,1)\). Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying (\(f_{p}^{q}\)). Suppose that \(B_{4R} = B_{4R}(x_0) \subset \Omega \). Let \(u \in G_-(\Omega ; q, c, s, g)\) satisfy \(u \ge 0\) in \(B_{4R}\) and
for some \(\gamma \in (0,1)\). There exists \(\delta \in (0,1)\) such that if
then
The constant \(\delta \) depends only on d, \(s_0\), p, q, c and \(\gamma \).
Proof
Within the proof we use \(C>0\) to denote a constant depending on d, \(s_0\), p, q, c and \(\gamma \) and whose value might change from line to line. We may assume that \(x_0 = 0\).
Let us assume
for some \(\gamma _0 \in (0, 2^{-d-1}]\). We first prove the assertion of the lemma under the assumption (4.4) and then verify (4.4) using (4.1).
Let \(0< {\tilde{\sigma }}:=\max \lbrace s_0/4, 2s-1 \rbrace< \sigma := \max \lbrace s_0/2, (3s-1)/2 \rbrace < s\). Then, we have
for some \(C = C(s_0) > 0\). Indeed, for the last inequality in (4.5), we observe that
Let \(\delta \in (0, 1/8)\) to be determined later. Let \(\delta H \le h < k \le 2\delta H\), \(R \le \rho < \tau \le 2R\) and define \(w_\pm = (u-k)_\pm \), \(A^-_{k,R} = \{ x \in B_R : u(x) < k \} = {{\,\textrm{supp}\,}}(w_-) \cap B_R\). By (4.4) we have
Thus, we apply Theorem 2.10 to w to obtain
Moreover, by (4.5) and Lemma 2.9 we have
or equivalently,
where \(\mu (\text {d}X) = (1-s) |x-y|^{-d+(s-\sigma )p} \,\textrm{d}y\,\textrm{d}x\). Since F is increasing and convex, Jensen’s inequality yields
By definition of \(G_-(\Omega ; q, c, s, g)\), we have
where \(A_{k}^{+} = \lbrace y \in {\mathbb {R}}^d: u(y) < k \rbrace \). Let us estimate the right-hand side of this inequality. Using the assumptions that \(u \ge 0\) in \(B_{4R}\) and the fact that F is increasing, we estimate \(\Vert w_-\Vert _{L^1(B_\tau )} \le C |A_{k,\tau }^-| k\) and
Moreover, using (2.8) and (2.4), we obtain
The first term is controlled by \(CR^{-s} f'(k/R^s)\). For the second term, we use (4.2), (3.2) and \(\delta H < k\) to obtain
Note that the integral over \(B_{4R}\setminus B_{\rho }\) vanishes since \(u_-=0\) in \(B_{4R}\) and \(f'(0)=0\) by \(({f_{p}})\). Furthermore, we have
by \({(f^{q})}\) and (2.7). Therefore, we have estimated
Combining (4.6) and (4.8), we can find a constant \(C>0\), depending on d, \(s_0\), p, q, c and \(\gamma \), such that
Using Lemma 2.5, we deduce that
Note that Lemma 2.5 is applicable because \(\mu (A_{k, \rho }^- \times B_\rho ) \le C |A_{k, \rho }^-|\rho ^{(s-\sigma )p}\) and therefore
We iterate inequality (4.9) with \(k = k_j\), \(h = k_{j+1}\), \(\rho = R_{j+1}\), and \(\tau = R_j\), where
for \(j \in {\mathbb {N}} \cup \lbrace 0 \rbrace \). Let \(y_j = |A_{k_{j}, R_{j}}^{-}| / |B_{R_j}|\), then
In other words, we have \(y_{j+1} \le C b^j y_j^{d/(d-{\tilde{\sigma }})} \le C b^j \max \lbrace y_j^{1+\beta _1}, y_j^{1+\beta _2} \rbrace \), where
Thus, \(y_j \rightarrow 0\) as \(j \rightarrow \infty \), provided that \(y_0 \le \min \lbrace C^{-1/\beta _2} b^{-1/\beta _2^2}, C^{-1/\beta _1} b^{-1/\beta _1^2} \rbrace \). See [15, Lemma 4.4]. By taking \(\gamma _0 \le \min \lbrace C^{-1/\beta _2} b^{-1/\beta _2^2}, C^{-1/\beta _1} b^{-1/\beta _1^2}\rbrace \), we conclude (4.3) from (4.4).
Let us next prove (4.4) by contradiction using the assumption (4.1). Suppose that (4.4) does not hold, i.e.,
Let \({\bar{s}}\) be the constant given in Proposition 2.12. We distinguish two cases \(s \in [{\bar{s}}, 1)\) and \(s \in (0, {\bar{s}})\). For the first case, we let l be the unique integer such that \(2^{-l-1} \le \delta < 2^{-l}\) and set \(k_i = 2^{-i}H\) for \(i=0, 1, \dots , l-1\). To apply Proposition 2.12 to \((u-k_{i-1})_-\) with \(h = k_{i-1} - k_i\) and \(k = k_{i-1} - k_{i+1}\), we check the following conditions: By (4.1) and (4.10)
and
for \(i=1, \dots , l-2\). Moreover, we prove that there is a constant \(C>0\) such that
Indeed, it follows from \(u \ge 0\) in \(B_{4R}\) that
The estimate
follows from (4.5) and the computation
which can be obtained along the lines of the first part of this proof. Estimating \(\vert B_{2R} \cap \{ u \le k_{i-1} \}\vert \le CR^d\) and applying Lemma 2.5, we obtain (4.12), as desired.
Therefore, by applying Proposition 2.12 and using (4.11) we have
where \(D_i = B_{2R} \cap \lbrace h \le (u-k_{i-1})_-< k \rbrace = B_{2R} \cap \lbrace k_{i+1} < u \le k_i \rbrace \). Using (4.12) and \(k_{i+1} \ge 2\delta H\), we obtain
We sum up this inequality over \(i=1, \dots , l-2\) to derive
from which we conclude by definition of l
Therefore, we arrive at a contradiction by using (4.10) and taking \(\delta \) sufficiently small. The proof for the case \(s \in [{\bar{s}}, 1)\) is finished.
For the case \(s \in (0, {\bar{s}})\), we use the estimate (4.8) with \(k=4\delta H\) to obtain
By (4.1) and (4.10), we obtain
where we also used (2.8), \(({f_{p}})\) and (2.7). Therefore, by Lemma 2.5 we obtain \(\delta \le C \delta ^p\). Since \(p > 1\), we arrive at a contradiction by taking \(\delta \) sufficiently small. \(\square \)
Using Theorem 4.1, we prove Hölder estimates for functions in \(G(\Omega ; q, c, s, g)\).
Theorem 4.2
Let \(1<p\le q\), \(c > 0\), \(s_0 \in (0,1)\), and assume \(s \in [s_0,1)\). Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying (\(f_{p}^{q}\)). Then, there exist \(\alpha \in (0,1)\) and \(C > 0\), depending on d, \(s_0\), p, q and c, such that for every \(u \in G(\Omega ; q, c, s, g)\) and any \(B_{8R}(x_0) \subset \Omega \),
Proof
Let \(B_{8R}(x_0) \subset \Omega \). We may assume that \(x_0 = 0\) and that \(\Vert u\Vert _{L^{\infty }(B_{4R})} < \infty \). The idea of the proof is to find a small constant \(\alpha \in (0,1)\) and to construct a non-increasing sequence \((M_j)\) and a non-decreasing sequence \((m_j)\) satisfying
for all \(j \ge 0\), where \(R_j = 4^{-j}R\) and
for some \(C_0 > 0\). Once we construct such sequences, the desired result follows by a standard argument.
We set \(M_j = 4^{-\alpha j} L/2\) and \(m_j = -4^{-\alpha j}L/2\) for \(j=0, 1, \dots , j_0\) for some \(j_0 \in {\mathbb {N}}\) to be determined later. Moreover, we take \(C_0\) sufficiently large so that \(C_0 \ge 2 \cdot 4^{\alpha j_0}\). This ensures that \(M_j\) and \(m_j\) satisfy (4.14) up to \(j_0\). Let us now fix \(j \ge j_0\) and suppose that the sequences \((M_j)\) and \((m_j)\) have been constructed up to j. It is enough to construct \(M_{j+1}\) and \(m_{j+1}\) satisfying (4.14). We first assume
In this case, we define \(v=u-m_j\) and set \(H=(M_j-m_j)/2 = 4^{-\alpha j}L/2\). Then, \(0 \le v \le 2H\) in \(B_{4R_j}\) and
To apply Theorem 4.1 to v, we let \(\delta \) be the constant in Theorem 4.1 and verify (4.2). Indeed, it is easy to see that
for \(y \in B_{4R} \setminus B_{4R_j}\) and \(v(y) \ge -|u(y)| - L/2\) for \(y \in {\mathbb {R}}^d\setminus B_{4R}\). Thus, using (2.4)
Using the change of variables, we obtain
and hence
Taking \(\alpha = \alpha (d, s_0, p, q) \in (0,1)\) sufficiently small so that
we obtain
For \(J_2\), we use (2.4) to deduce
Since \(L \ge \textrm{Tail}_{f'}(u; 0, 4R)\), it follows from (3.2) and the definition of H
Choosing \(j_0\) sufficiently large so that \(8\cdot 4^{(\alpha -s_0)j_0} \le {\widetilde{\delta }}\), for some \({\widetilde{\delta }} < \delta \) to be determined later, we have
Similarly, by (2.3)
for some \(C_1,C_2 \ge 1\) depending on d, \(s_0\) and p. We now choose \({\widetilde{\delta }} = (2^{q+1} C_2)^{-\frac{1}{p-1}}\delta > 0\), and obtain:
This inequality together with (3.3) verify (4.2) and allow us to apply Theorem 4.1 to v. Therefore, we obtain \(v \ge \delta H\) in \(B_{R_j}\), which implies
upon choosing \(\alpha \in (0,1)\) so small that it satisfies (4.16) and \(\alpha < \log _4\left( \frac{2}{2-\delta } \right) \).
Therefore, we define \(M_{j+1} = M_j\) and \(m_{j+1} = m_j + 4^{-\alpha j} (1-4^{-\alpha })L\) in the case (4.15). The other case can be proved in a similar way. \(\square \)
5 Local boundedness
The goal of this section is to prove local boundedness of functions \(u \in G(\Omega ; q, c, s, f)\). More precisely, we prove that a function \(u \in G_+(\Omega ; q, c, s, f)\) is locally bounded from above. Similarly, one can prove that functions \(u \in G_-(\Omega ; q, c, s, f)\) are locally bounded from below by considering \(-u\).
Theorem 5.1
Let \(1 \le p \le q < p^{*}\), \(s_0 \in (0,1)\), \(c_0, c_1 > 0\) and assume \(s \in [s_0, 1)\). Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying \({(f^{q})}\) and \(({f > rsim t^p})\). If \(u \in G_+(\Omega ; q, c_1, s, f)\), then u is locally bounded from above. Moreover, for each \(B_{2R}(x_0) \subset \Omega \), there exists \(C > 0\) such that for every \(\delta \in (0,1)\)
The constant C depends on d, \(s_0\), p, q, \(p^{*}-q\), \(c_0\), \(c_1\) and R.
Note that \(p=1\) is allowed in the proof of local boundedness.
Proof
Let \(x_0 \in \Omega \) and \(R > 0\) be such that \(B_{2R}(x_0) \subset \Omega \). We assume without loss of generality that \(x_0 = 0\). For \(j=0,1,\dots \), let
where k is an arbitrary positive number that will be determined later. We define \(w_j = (u-k_j)_+\), \({\tilde{w}}_j = (u-{\tilde{k}}_j)_+\), \(A_{k,R}^+ = \{ x \in B_R : u(x) > k \}\), and
Since \(u \in G_+(\Omega ;q,c_1,s,f)\), using the assumptions \({(f^{q})}\) and \(({f > rsim t^p})\) we have
for some \(C = C(d, q, c_0, c_1) > 0\). Since
and \({\tilde{w}}_j \le w_j\), by assuming \(k \ge \delta ^{\frac{q-1}{q}}\) we have
for some \(C = C(d, q, c_0, c_1, R) > 0\). For \(J_2\), we observe
and \(\textrm{Tail}_{f'}({\tilde{w}}_j; 0, R_{j+1}) \le \textrm{Tail}_{f'}(u_+; 0, R)\). Fix \(\delta \in (0,1)\) and assume
Then, using (5.1), (3.2), (1.2) and \(f(1) = 1\), we deduce
for some constant \(C = C(d, q, c_0, c_1,s_0) > 0\). Combining the estimates of \(J_1\) and \(J_2\), we arrive at
Using Theorem 2.11 and the inequality \({\tilde{w}}_j^{p^{*}} \ge (k_{j+1}-{\tilde{k}}_j)^{p^{*}-q} w_{j+1}^{q}\), we deduce
or
where \(b = 2^{p^{*}-q+(d+2q)p^{*}/p}\) and \(\beta = \frac{p^{*}}{p}-1\). If \(Y_0 \le (C k^{q-p^{*}} \delta ^{-(q-1)p^{*}/p})^{-1/\beta } b^{-1/\beta ^2}\), then \(Y_j \rightarrow 0\) as \(j \rightarrow \infty \). Thus, if we assume
then
We now take
with \(C_0 = (Cb^{1/\beta })^{1/(p^{*}-q)}\), which is in accordance with (5.1) and (5.2). \(\square \)
6 Application to minimizers
In this section, we prove Theorem 1.1 by showing that local minimizers of (1.1) belong to the De Giorgi class \(G(\Omega ; q, c, s, f)\) under some assumptions on f and using several results from previous sections.
Let us first define local minimizers of (1.1). We assume that f is a convex increasing function and k satisfies (k), i.e.,
Definition 6.1
[minimizer] We say that \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) is a local subminimizer (superminimizer) of (1.1) if for every measurable function \(v : {\mathbb {R}}^d\rightarrow {\mathbb {R}}\) with \(v = u\) a.e. in \({\mathbb {R}}^d\setminus \Omega \) and \(v \le u\) (\(v \ge u\)) a.e. in \(\Omega \), it holds that \({\mathcal {I}}_f(u) \le {\mathcal {I}}_f(v)\). We call \(u \in V^{s, f}(\Omega |{\mathbb {R}}^d)\) a local minimizer of (1.1) if it is a subminimizer and a superminimizer.
Recall that we always assume \(f(0) = 0\) and \(f(1)=1\).
Theorem 6.2
Let \(s \in (0,1)\), \(q \ge 1\) and \(\Lambda \ge 1\). Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying \({(f^{q})}\) and let \(k: {\mathbb {R}}^d\times {\mathbb {R}}^d\rightarrow {\mathbb {R}}\) be a measurable function satisfying (K). Let \(u \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) be a local subminimizer of (1.1). Then, \(u \in G_+(\Omega ;q,c,s,f)\) for some \(c = c(d, q, \Lambda )> 0\).
It is worth emphasizing that the result of the theorem is true in the case \(q=1\).
Proof
We follow the strategy carried out in [17, Proposition 7.5]. Let \(x_0 \in \Omega \), \(0< r < R \le d(x_0, \partial \Omega )\) and \(k \in {\mathbb {R}}\). Without loss of generality, we take \(x_0 = 0\). Let \(r \le \rho < \tau \le R\) and let \(\eta \in C_c^{\infty }({\mathbb {R}}^d)\) be a cutoff function with \(0 \le \eta \le 1\), \({{\,\textrm{supp}\,}}(\eta ) = B_{\frac{\tau +\rho }{2}}\), \(\eta \equiv 1\) in \(B_{\rho }\), and \(\Vert \nabla \eta \Vert _{\infty } \le \frac{4}{\tau -\rho }\). Let \(w_\pm (x)\) and \(A_k^\pm \) be as in Definition 3.3. We define \(A_{k,R}^+ = \{ x \in B_R : u(x) > k \}\) and \(A_{k,R}^- = \{ x \in B_R : u(x) < k \}\).
We set \(v := u - \eta ^qw_+\), then \(u \equiv v\) in \({\mathbb {R}}^d\setminus B_{\tau }\) and \(u \ge v\) a.e. in \({\mathbb {R}}^d\). Since u is a local subminimizer of (1.1), it holds that \({\mathcal {I}}_f(u) \le {\mathcal {I}}_f(v)\), i.e.,
where
To estimate A(x, y), we distinguish four different cases and prove the following.
If \(x \in A_k^-\) or \(y \in A_k^-\), then
Furthermore, if \(x,y\in A_{k,\rho }^+\), then
If \(x \in A_{k,\rho }^+\) and \(y \in A_{k}^-\), then
Finally, if \(x,y \in A_k^+\), then we have
for some \(c = c(q)>0\), where \(a\vee b := \max \{a,b\}\). In the following, we prove (6.2)–(6.5). The proof of (6.2) is a direct consequence of monotonicity of f. Namely, if \(x \not \in A_k^+\) and \(y\in A_k^+\), then
By symmetry, one can treat the case \(x \in A_k^+\) and \(y\not \in A_k^+\). The remaining case is trivial.
To see (6.3), observe that for \(x,y \in A_{k,\rho }^+\) it holds that \(\eta (x) = \eta (y) = 1\), and therefore
Let us prove (6.4). For \(x \in A_{k,\rho }^+\) and \(y \not \in A_k^+\) it holds that
By application of Lemma 2.7 with \(\theta = \frac{1}{2}\), \(a = \frac{w_-(y)}{\vert x-y \vert ^s}\) and \(b = \frac{w_+(x)}{\vert x-y \vert ^s}\), we obtain
which implies (6.4) since \(w_+(y) = 0\).
To prove (6.5) let us take \(x,y \in A_{k}^+\). We compute
Let us assume without loss of generality that \(\eta (x) \ge \eta (y)\). Then, we have \(\vert \eta ^q(y)-\eta ^q(x) \vert \le q \eta ^{q-1}(x) \vert \eta (y)-\eta (x) \vert \). We estimate, using monotonicity and convexity of f, as well as \({(f^{q})}\)
which implies (6.5).
By putting together the information from (6.2)–(6.4) and using assumptions on k, we deduce
Moreover, from (6.2), (6.4) and (6.5) we obtain
Note that by monotonicity of f and Lemma 2.1, we have
for some \(c = c(d, q) > 0\). We use Lemmas 2.3 and 2.2 to obtain
Since the map \(s \mapsto (1-s)/(1-2^{-(1-s)})\) is bounded on (0, 1) from above, we have
for some \(c = c(d, q) > 0\).
By combination of the estimates (6.6), (6.7) and (6.8), we get:
Let us deduce two more estimates for A(x, y):
The proof of (6.10) is trivial since \({{\,\textrm{supp}\,}}(\eta ) = B_{\frac{\tau +\rho }{2}}\) and \(w_+=0\) on \(A_k^-\). To see (6.11), we compute
and apply Lemma 2.8 with \(\mu = 1-\eta ^q(x)\), \(a = \frac{w_+(x)}{\vert x-y \vert ^s}\) and \(b=\frac{w_+(y)}{\vert x-y \vert ^s}\). Consequently, by (6.2), (6.4), (6.10) and (6.11), it holds that
Moreover, using Lemma 2.3 and (2.2) with \(p=1\) we observe that
By combining estimates (6.1), (6.9), (6.12), and (6.13), we derive
for some \(c = c(d, q, \Lambda ) > 0\). By setting
we can deduce from the above line that
We “fill the hole" by adding \(c\phi (\rho )\) to both sides. After dividing by \(1+c\), we get that
where \(\gamma \in (0,1)\) and \(c=c(d, q, \Lambda ) > 0\). The desired result follows now from a standard iteration argument, see Lemma 4.11 in [17]. \(\square \)
Remark 6.3
Similar to the proof of Theorem 6.2, it is possible to show that local superminimizer (minimizers, respectively) \(u \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) satisfies \(u \in G_-(\Omega ;q,c,s,f)\) (\(u \in G(\Omega ;q,c,s,f)\), respectively) for some \(c = c(d, q, \Lambda ) > 0\).
Proof (Proof of Theorem 1.1)
[Proof of Theorem 1.1] By Theorem 6.2, it follows that \(u \in G(\Omega ;q,c_1,s,f)\) for some \(c_1 = c_1(d, q, \Lambda )> 0\). According to Proposition 3.4, it holds that \(u \in G(\Omega ;q,c_2,s,g)\) for some \(c_2 = c_2(d, p, q, \Lambda ) > 0\). From Theorems 4.2 and 5.1 we deduce the desired result. \(\square \)
7 Application to weak solutions
In this section we aim to study weak solutions to nonlocal equations (1.6) and prove Theorem 1.2. Throughout this section we assume that f is a convex increasing function satisfying \({(f^{q})}\) and h is a measurable function satisfying the structure condition (h), i.e.,
for a.e. \(x, y \in {\mathbb {R}}^d\) and for all \(t \in {\mathbb {R}}\). We define weak solutions to (1.6) as follows:
Definition 7.1
(Weak solution) We say that \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) is a weak subsolution to (1.6) if for every \(\phi \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) with \(\phi = 0\) a.e. in \({\mathbb {R}}^d\setminus \Omega \) and \(\phi \ge 0\) a.e. in \(\Omega \), it holds that
We say that \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) is a weak supersolution if \(-u\) is a weak subsolution. A function \(u \in V^{s,f}(\Omega | {\mathbb {R}}^d)\) is called a weak solution if it is a weak subsolution and a weak supersolution.
Recall that we always assume \(f(0) = 0\) and \(f(1)=1\). This assumption can be made without loss of generality since u solves \({\mathcal {L}}_h u = 0\) if and only if u solves \({\mathcal {L}}_{h/f(1)} u = 0\) and one can always choose \(f(0) = 0\).
Remark 7.2
Let us prove that the weak formulation (7.1) is well-defined. Let \(u,\phi \in V^{s,f}(\Omega | {\mathbb {R}}^d)\). Then, by (h) and Fenchel’s inequality (3.6), we have
where we used that by (3.5) and \({(f^{q})}\): \(f^{*}(f'(t)) \le (q-1) f(t)\).
The following theorem yields that weak solutions to (1.6) belong to the De Giorgi classes introduced in Sect. 3.
Theorem 7.3
Let \(s \in (0,1)\), \(q \ge 1\) and \(\Lambda \ge 1\). Let \(f: [0, \infty ) \rightarrow [0, \infty )\) be a convex increasing function satisfying \({(f^{q})}\) and let \(h: {\mathbb {R}}^d\times {\mathbb {R}}^d\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) be a measurable function satisfying (h). Let \(u \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) be a weak subsolution to (1.6). Then, \(u \in G_+(\Omega ;q,c,s,f)\) for some \(c > 0\) depending on d, q and \(\Lambda \).
Note that the result remains true for \(q=1\).
Proof
The desired result follows from a similar argument as in the proof of Theorem 6.2. Let \(x_0 \in \Omega \), \(0< r < R \le d(x_0, \partial \Omega )\), and \(k \in {\mathbb {R}}\). We may assume without loss of generality that \(x_0 = 0\). Let \(r \le \rho < \tau \le R\) and let \(\eta \in C_c^{\infty }({\mathbb {R}}^d)\) be a cutoff function with \(0 \le \eta \le 1\), \({{\,\textrm{supp}\,}}(\eta ) = B_{\frac{\tau +\rho }{2}}\), \(\eta \equiv 1\) in \(B_{\rho }\), and \(\Vert \nabla \eta \Vert _{\infty } \le \frac{4}{\tau -\rho }\). We define \(w_\pm (x)\), \(A_k^{\pm }\) and \(A_{k, R}^\pm \) as in Theorem 6.2. We set \(v = \eta ^q w_+\). Since u is a weak subsolution to (1.6), we have
where
Let us estimate B(x, y). If \(x, y \in A_k^-\), then
If \(x \in A_{k}^+\) and \(y \in A_{k}^-\), then by (h)
Thus, we obtain
where we used (2.4) and Lemma 2.3.
If \(x, y \in A_{k}^+\), we prove
for any \(\varepsilon \in (0,1)\), where \(c = c(q, \varepsilon ) > 0\). It is enough to prove (7.5) for the case \(w_+(x) \ge w_+(y)\). If \(\eta (x) \ge \eta (y)\), then (7.5) follows from
where we used (h) and Lemma 2.3. When \(\eta (y) \ge \eta (x)\), we observe that
where we used Lemma 2.3 again. Note that
Thus, for \(\varepsilon \in (0,1)\) we use Lemma 2.1 and the Fenchel’s inequality (3.6) to obtain
By (3.5), \({(f^{q})}\) and Lemma 2.2 (iv) with \(p=1\), we deduce that
for some \(c=c(q, \varepsilon ) > 0\). Therefore, (7.5) follows from (7.6) and (7.7).
Combining (7.3), (7.4) and (7.5), we have
Let us now take into account the pairs \((x,y) \in ({\mathbb {R}}^d\times {\mathbb {R}}^d) \setminus (B_{\tau } \times B_{\tau })\). Using (h) we compute
By monotonicity of \(f'\), we have that
Moreover, by (6.13) we obtain
Therefore, it follows from (7.2), (7.8)–(7.11) and (6.8) that
where \(c = c(d, q,\Lambda , \varepsilon ) > 0\). We take \(\varepsilon = (4\Lambda ^2(q-1))^{-1} \wedge 2^{-1} \in (0,1)\) so that (7.12) boils down to (6.14) with \(\gamma =1/2\) and \(c = c(d,q,\Lambda ) > 0\). This finishes the proof. \(\square \)
Remark 7.4
Similar to the proof of Theorem 7.3, it is possible to show that weak solutions \(u \in V^{s,f}(\Omega |{\mathbb {R}}^d)\) to \({\mathcal {L}}_h u = 0\) in \(\Omega \) satisfy \(u \in G(\Omega ;q,c,s,f)\) for some \(c > 0\) depending on d, q and \(\Lambda \).
Proof (Proof of Theorem 1.2)
[Proof of Theorem 1.2] By Theorem 7.3, it follows that \(u \in G(\Omega ;q,c_1,s,f)\) for some \(c_1 > 0\) depending on d, q and \(\Lambda \). According to Proposition 3.4, it holds that \(u \in G(\Omega ;q,c_2,s,g)\) for some \(c_2 = c_2(d, p, q, \Lambda ) > 0\). From Theorems 4.2 and 5.1 we deduce the desired result. \(\square \)
Data Availibility Statement
Data sharing is not applicable to this article as no data sets were generated or analysed.
Change history
21 January 2023
Missing Open Access funding information has been added in the Funding Note.
References
Alberico, A., Cianchi, A., Pick, L., Slavíková, L.: Fractional Orlicz-Sobolev embeddings. J. Math. Pures Appl. 9(149), 216–253 (2021)
Alberico, A., Cianchi, A., Pick, L., Slavíková, L.: On fractional Orlicz–Sobolev spaces. Anal. Math. Phys. 11(2):Paper No. 84, 21 (2021)
Acerbi, E., Mingione, G.: Regularity results for a class of functionals with non-standard growth. Arch. Ration. Mech. Anal. 156(2), 121–140 (2001)
Bourgain, J., Brezis, H., Mironescu, P.: Limiting Embedding Theorems for \(W^{s,p}\) When \(s\uparrow 1\) and Applications, vol. 87, pp 77–101 (2002). Dedicated to the memory of Thomas H. Wolff
Byun, S.-S., Kim, H., Ok, J.: Local Hölder continuity for fractional nonlocal equations with general growth. arXiv preprint arXiv:2112.13958 (2021)
Bhattacharya, T., Leonetti, F.: A new Poincaré inequality and its application to the regularity of minimizers of integral functionals with nonstandard growth. Nonlinear Anal. 17(9), 833–839 (1991)
Brasco, L., Lindgren, E.: Higher Sobolev regularity for the fractional \(p\)-Laplace equation in the superquadratic case. Adv. Math. 304, 300–354 (2017)
Brasco, L., Lindgren, E., Schikorra, A.: Higher Hölder regularity for the fractional \(p\)-Laplacian in the superquadratic case. Adv. Math. 338, 782–846 (2018)
Bahrouni, S., Ounaies, H.: Embedding theorems in the fractional Orlicz-Sobolev space and applications to non-local problems. Discrete Contin. Dyn. Syst. 40(5), 2917–2944 (2020)
Byun, S.-S., Ok, J., Song, K.: Hölder regularity for weak solutions to nonlocal double phase problems. arXiv preprint arXiv:2108.09623 (2021)
Bahrouni, S., Ounaies, H., Tavares, L.S.: Basic results of fractional Orlicz-Sobolev space and applications to non-local problems. Topol. Methods Nonlinear Anal. 55(2), 681–695 (2020)
Breit, D., Stroffolini, B., Verde, A.: A general regularity theorem for functionals with \(\phi \)-growth. J. Math. Anal. Appl. 383(1), 226–233 (2011)
Bonder, J.F., Salort, A., Vivas, H.: Interior and up to the boundary regularity for the fractional \(g\)-Laplacian: the convex case. arXiv preprint arXiv:2008.05543 (2020)
Cianchi, A.: Boundedness of solutions to variational problems under general growth conditions. Commun. Partial Differ. Equ. 22(9–10), 1629–1646 (1997)
Chaker, J., Kim, M.: Local regularity for nonlocal equations with variable exponents. arXiv preprint arXiv:2107.06043 (2021)
Chaker, J., Kim, M., Weidner, M.: Harnack inequality for nonlocal problems with non-standard growth. Math. Ann. (2022). https://doi.org/10.1007/s00208-022-02405-9
Cozzi, M.: Regularity results and Harnack inequalities for minimizers and solutions of nonlocal problems: a unified approach via fractional De Giorgi classes. J. Funct. Anal. 272(11), 4762–4837 (2017)
Caffarelli, L.A., Vasseur, A.: Drift diffusion equations with fractional diffusion and the quasi-geostrophic equation. Ann. Math. (2) 171(3), 1903–1930 (2010)
Di Castro, A., Kuusi, T., Palatucci, G.: Nonlocal Harnack inequalities. J. Funct. Anal. 267(6), 1807–1836 (2014)
Di Castro, A., Kuusi, T., Palatucci, G.: Local behavior of fractional \(p\)-minimizers. Ann. Inst. H. Poincaré Anal. Non Linéaire 33(5), 1279–1299 (2016)
De Filippis, C., Palatucci, G.: Hölder regularity for nonlocal double phase equations. J. Differ. Equ. 267(1), 547–586 (2019)
de La Fortelle, A.: Generalized inverses of increasing functions and Lebesgue decomposition. Markov Process. Relat. Fields 26(4), 637–658 (2020)
Dall’Aglio, A., Mascolo, E., Papi, G.: Local boundedness for minima of functionals with nonstandard growth conditions. Rend. Mat. Appl. (7) 18(2), 305–326 (1998)
De Nápoli, P., Fernández Bonder, J., Salort, A.: A Pólya-Szegö principle for general fractional Orlicz-Sobolev spaces. Complex Var. Elliptic Equ. 66(4), 546–568 (2021)
Di Nezza, E., Palatucci, G., Valdinoci, E.: Hitchhiker’s guide to the fractional Sobolev spaces. Bull. Sci. Math. 136(5), 521–573 (2012)
Diening, L., Stroffolini, B., Verde, A.: Everywhere regularity of functionals with \(\phi \)-growth. Manuscripta Math. 129(4), 449–481 (2009)
Embrechts, P., Hofert, M.: A note on generalized inverses. Math. Methods Oper. Res. 77(3), 423–432 (2013)
Esposito, L., Leonetti, F., Mingione, G.: Sharp regularity for functionals with \((p, q)\) growth. J. Differ. Equ. 204(1), 5–55 (2004)
Fernández Bonder, J., Salort, A.M.: Fractional order Orlicz-Sobolev spaces. J. Funct. Anal. 277(2), 333–367 (2019)
Fiorenza, A.: An inequality for Jensen means. Nonlinear Anal. 16(2), 191–198 (1991)
Frehse, J.: A note on the Hölder continuity of solutions of variational problems. Abh. Math. Sem. Univ. Hamburg 43, 59–63 (1975)
Fang, Y., Zhang, C.: On weak and viscosity solutions of nonlocal double phase equations. arXiv preprint arXiv:2106.04412 (2021)
Giaquinta, M., Giusti, E.: On the regularity of the minima of variational integrals. Acta Math. 148, 31–46 (1982)
Giaquinta, M.: Multiple integrals in the calculus of variations and nonlinear elliptic systems. Ann. Math. Stud. 105. Princeton University Press, Princeton, NJ (1983)
Giusti, E.: Direct methods in the calculus of variations. World Scientific Publishing Co., Inc, River Edge (2003)
Goel, D., Kumar, D., Sreenadh, K.: Regularity and multiplicity results for fractional \((p,q)\)-Laplacian equations. Commun. Contemp. Math., 22(8):1950065, 37, 2020
Giacomoni, J., Kumar, D., Sreenadh, K.: Global regularity results for non-homogeneous growth fractional problems. arXiv preprint arXiv:2106.02501 (2021)
Giacomoni, J., Kumar, D., Sreenadh, K.: Interior and boundary regularity results for strongly nonhomogeneous \( p, q \)-fractional problems. arXiv preprint arXiv:2102.06080 (2021)
Giannetti, F., Passarelli di Napoli, A.: Regularity results for a new class of functionals with non-standard growth conditions. J. Differ. Equ. 254(3), 1280–1305 (2013)
Lieberman, G.M.: The natural generalization of the natural conditions of Ladyzhenskaya and Ural’tseva for elliptic equations. Commun. Partial Differ. Equ. 16(2–3), 311–361 (1991)
Ladyzhenskaya, O.A., Ural’tseva, Ni.N.: Linear and Quasilinear Elliptic Equations. Academic Press, New York-London (1968). Translated from the Russian by Scripta Technica, Inc, Translation editor: Leon Ehrenpreis
Marcellini, P.: Regularity of minimizers of integrals of the calculus of variations with nonstandard growth conditions. Arch. Rational Mech. Anal. 105(3), 267–284 (1989)
Marcellini, P.: Regularity and existence of solutions of elliptic equations with \(p, q\)-growth conditions. J. Differ. Equ. 90(1), 1–30 (1991)
Marcellini, P.: Regularity for elliptic equations with general growth conditions. J. Differ. Equ. 105(2), 296–333 (1993)
Mingione, G.: Regularity of minima: an invitation to the dark side of the calculus of variations. Appl. Math. 51(4), 355–426 (2006)
Mingione, G.: Gradient potential estimates. J. Eur. Math. Soc. (JEMS) 13(2), 459–486 (2011)
Moscariello, G., Nania, L.: Hölder continuity of minimizers of functionals with nonstandard growth conditions. Ricerche Mat. 40(2):259–273 (1992), 1991
Mascolo, E., Papi, G.: Local boundedness of minimizers of integrals of the calculus of variations. Ann. Mat. Pura Appl. 4(167), 323–339 (1994)
Marcellini, P., Papi, G.: Nonlinear elliptic systems with general growth. J. Differ. Equ. 221(2), 412–443 (2006)
Mingione, G., Rǎdulescu, V.: Recent developments in problems with nonstandard growth and nonuniform ellipticity. J. Math. Anal. Appl. 501(1):Paper No. 125197, 41 (2021)
Maz’ya, V.G., Shaposhnikova, T.O.: On the Bourgain, Brezis, and Mironescu theorem concerning limiting embeddings of fractional Sobolev spaces. J. Funct. Anal. 195(2), 230–238 (2002)
Molina, S., Salort, A., Vivas, H.: Maximum principles, Liouville theorem and symmetry results for the fractional \(g\)-Laplacian. Nonlinear Anal. 212, Paper No. 112465, 24 (2021)
Nowak, S.: \(H^{s,p}\) regularity theory for a class of nonlocal elliptic equations. Nonlinear Anal. 195, 111730, 28 (2020)
Ok, J.: Regularity of \(\omega \)-minimizers for a class of functionals with non-standard growth. Calc. Var. Partial Differ. Equ. 56(2):Paper No. 48, 31 (2017)
Ok, J.: Regularity for double phase problems under additional integrability assumptions. Nonlinear Anal., 194:111408, 13 (2020)
Ok, J.: Local Hölder regularity for nonlocal equations with variable powers. arXiv preprint arXiv:2107.06611 (2021)
Scott, J.M., Mengesha, T.: Self-improving inequalities for bounded weak solutions to nonlocal double phase equations. Commun. Pure Appl. Anal. 21(1), 183–212 (2022)
Zhou, J., Tan, Z.: Regularity of weak solutions to a class of nonlinear problem with non-standard growth conditions. J. Math. Phys. 61(9):091509, 23 (2020)
Funding
Open Access funding enabled and organized by Projekt DEAL. Jamil Chaker gratefully acknowledges funding by the Deutsche Forschungsgemeinschaft (SFB 1283/2 2021 - 317210226). Minhyun Kim and Marvin Weidner gratefully acknowledge funding by the Deutsche Forschungsgemeinschaft (GRK 2235/2 2021 - 282638148). Marvin Weidner has received funding from the European Research Council (ERC) under the Grant Agreement No 801867.
Author information
Authors and Affiliations
Contributions
All authors developed and discussed the results and contributed to the final manuscript.
Corresponding author
Ethics declarations
Conflict of interest
All authors state no conflict of interest.
Informed consent
Informed consent has been obtained from all individuals included in this research work.
Additional information
Communicated by A. Mondino.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
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/.
About this article
Cite this article
Chaker, J., Kim, M. & Weidner, M. Regularity for nonlocal problems with non-standard growth. Calc. Var. 61, 227 (2022). https://doi.org/10.1007/s00526-022-02364-8
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00526-022-02364-8