Abstract
We prove the existence of infinitely many solutions for a class of elliptic Dirichlet problems with non-symmetric nonlinearities. In particular, in suitable domains of \({\mathbb {R}}^n\) with \(n\ge 3\), this result gives a positive answer to a well known conjecture formulated by A. Bahri and P.L. Lions. The proof is based on a minimization method which does not require the use of techniques of deformation from the symmetry. This method allows us to piece together solutions of Dirichlet problems in suitable subdomains, so we obtain infinitely many nodal solutions with a prescribed nodal structure.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this paper we are concerned with Dirichlet problems of the form
where \(\Omega \) is a bounded domain of \({\mathbb {R}}^n\) with \(n\ge 1\), \(\psi \in L^2(\Omega )\), \(p>1\) and \(p<\frac{n+2}{n-2}\) when \(n\ge 3\).
The solutions of problem (1.1) are the critical points of the energy functional \(E_\psi :H^1_0(\Omega )\rightarrow {\mathbb {R}}\), defined by
where, under our assumptions, the exponent \(p+1\) is less than the critical Sobolev exponent \(2^*=\frac{2n}{n-2}\) for \(n\ge 3\).
If \(\psi \equiv 0\) in \(\Omega \), the functional \(E_\psi \) is even, so the equivariant Lusternik–Schnirelmann theory for \({\mathbb {Z}}_2\)-symmetric sets may be applied and guarantees the existence of infinitely many solutions (see for instance [1, 3, 9, 18,19,20, 29, 30, 37, 39]).
A natural question, which goes back to the beginning of the eighties, is whether the infinite number of solutions still persists for \(\psi \not \equiv 0\).
In particular, this question was raised to the attention by Rabinowitz in his monograph on minmax methods (see [39, Remark 10.58]). In [4] Bahri proved that, if \(n\ge 3\) and \(1<p<\frac{n+2}{n-2}\), then there exists an open dense set of \(\psi \) in \(L^2(\Omega )\) such that problem (1.1) admits infinitely many solutions. In [8] Bahri and Lions proved that, if \(n\ge 3\) and \(1<p<\frac{n}{n-2}\), then problem (1.1) admits infinitely many solutions for every \(\psi \in L^2(\Omega )\). These results suggest the following conjecture, proposed by Bahri and Lions in [8]: the multiplicity result obtained in [8] holds also under the more general assumption \(1<p<\frac{n+2}{n-2}\).
In the present paper we prove that, if the domain \(\Omega \) is a cube of \({\mathbb {R}}^n\), then problem (1.1) has infinitely many solutions for every \(\psi \in L^2(\Omega )\). Thus, for \(n\ge 3\), our result shows that the Bahri–Lions conjecture is true at least when \(\Omega \) is a cube of \({\mathbb {R}}^n\).
In order to show that the infinite number of solutions we have for \(\psi \equiv 0\) persists under perturbations, a detailed analysis was originally carried on in [2, 3, 5,6,7,8, 26, 31, 32, 38, 41, 45] by Ambrosetti, Bahri, Berestycki, Ekeland, Ghoussoub, Krasnoselskyii, Lions, Marino, Prodi, Rabinowitz, Struwe and Tanaka by introducing new perturbation methods.
More recently, a new approach to tackle the break of symmetry in elliptic problems has been developed by Bolle, Chambers, Ghoussoub and Tehrani (see [10, 11, 17]). However, that approach (which works also for more general nonlinear problems) did not allow to solve the Bahri–Lions conjecture.
Related results can be found also in other, more recent, papers (see for example [40] and references therein).
In the present paper we develop a method introduced in [34] in order to construct infinitely many nodal solutions of problem (1.1), having a prescribed nodal structure.
The idea is to piece together the solutions of Dirichlet problems in suitable subdomains of \(\Omega \). A similar idea has been first used by Struwe in earlier papers (see [41,42,43] and references therein). We consider as nodal regions some subdomains of \(\Omega \) that are deformations of cubes by suitable Lipschitz maps (so we obtain nodal solutions having a “check” nodal structure). Notice that Lipschitz conditions combined with the covering of \({\mathbb {R}}^n\) by cubes with vertices in \({\mathbb {Z}}^n\) have been also used in some recent papers by Rabinowitz and Byeon in order to construct solutions with a prescribed pattern for the Allen-Cahn model equation (see [14, 15] and references therein).
The main result of the present paper is stated in Theorem 2.1 (which is a direct consequence of Proposition 2.2) and says that if \(\Omega \) is a cube of \({\mathbb {R}}^n\), \(n\ge 1\), \(p>1\) and \(p<\frac{n+2}{n-2}\) when \(n>2\), then for all \(\psi \in L^2(\Omega )\) there exist infinitely many nodal solutions of problem (1.1), having as nodal structure suitable partitions of \(\Omega \) in subdomains that are Lipschitz deformations of arbitrarily small cubes. More precisely, in Proposition 2.2 we prove that there exists \({\bar{k}}\in {\mathbb {N}}\) such that for all positive integer \(k\ge {\bar{k}}\) there exist at least two solutions \(u_k(x)\) and \(v_k(x)\) of problem (1.1) such that the nodal regions of the functions \(u_k\left( \frac{x}{k}\right) \) and \(v_k\left( \frac{x}{k}\right) \), after translations, tend to the cube \(\Omega \) as \(k\rightarrow \infty \). Moreover, the number of nodal regions of \(u_k\), \(v_k\) and their energy \(E_\psi (u_k)\), \(E_\psi (v_k)\) tend to infinity as \(k\rightarrow \infty \), while the size of the nodal regions tends to zero.
Notice that, in dimension \(n=1\), the existence of infinitely many solutions for all \(\psi \) in \(L^2(\Omega )\) follows from a result obtained by Ehrmann in [25] (see also [24, 28] for related results). However, the method used by Ehrmann relies on a shooting argument, typical of ordinary differential equations, combined with counting the oscillations of the solutions in the interval \(\Omega \). On the contrary, in the present paper we use a method which is more similar to the one introduced by Nehari in [35], that can be in a natural way extended to the case \(n>1\). In fact, Nehari’s method was used by Coffman in [19, 20] and, independently, by Hempel in [29, 30] to study an analogous problem for partial differential equations.
More recently, Nehari’s method has been used also by Conti, Terracini and Verzini to study optimal partition problems, existence of changing sign solutions etc. (see [21,22,23, 46]).
Let us remark that Nehari consider an odd differential operator (so the corresponding energy functional is even) and prove that for every positive integer k there exists a solution having exactly k nodal regions. On the contrary, as Ehrmann in [25], we find solutions with a large number of nodal regions. However, let us point out that our multiplicity result is sharp because, as we proved in [34, Proposition 3.5], \(\psi \) can be chosen in \(L^2(\Omega )\) in such a way that problem (1.1) does not have solutions with a small number of nodal regions: more precisely, for every positive integer h there exists \(\psi _h\) in \(L^2(\Omega )\) such that every solution of problem (1.1) with \(\psi =\psi _h\) has at least h nodal regions.
Now, let us describe the method we use to prove our result. For every cube \(\Omega \) of \({\mathbb {R}}^n\) and every positive integer k, let us consider the \(k^n\) cubic open subdomains \(C_1^k,C_2^k,\ldots ,C_{k^n}^k\), having all the same size, such that \(\overline{\Omega }=\cup _{i=1}^{k^n}{\overline{C}}_i^k\). So, these subdomains are pairwise disjoint and, for all \(i\in \{1,\ldots ,k^n\}\) the cube \(kC_i^k\) is a translation of the cube \(\Omega \).
Moreover, for all \(L\in ]0,1[\), let us consider the set \(D_L\) of all the deformations \(T:\overline{\Omega }\rightarrow \overline{\Omega }\) such that T differs from the identity map in \(\overline{\Omega }\) by a Lipschitz function with Lipschitz constant L, \(T(\overline{\Omega })=\overline{\Omega }\) and \(T(F)=F\) for every face F of the cube \(\overline{\Omega }\). Notice that, since \(L\in ]0,1[\), every deformation \(T\in D_L\) is a bilipschitz map in \(\overline{\Omega }\). Then, for all T in \(D_L\) and k in \({\mathbb {N}}\), by using a Nehari type minmax argument in every subdomain \(T(C_i^k)\) with \(i\in \{1,\ldots ,k^n\}\), we construct two distinct nodal functions \(u^T_k\) and \(v^T_k\) in \(H^1_0(\Omega )\) whose nodal regions are the subdomains \(T(C_i^k)\), for \(i=1,\ldots ,k^n\), and such that, for k large enough, \(u^T_k\) and \(v^T_k\) satisfy equation (1.1) in each nodal region and are solutions of the Dirichlet problem (1.1) in \(\Omega \) when, in addition, they satisfy a suitable stationary property. Moreover, the construction of \(u_k^T\) and \(v_k^T\) shows that \(v^T_k\) behaves as \(-u^T_k\) when \(k\rightarrow \infty \).
Now, for all \(k\in {\mathbb {N}}\), we minimize the energy functional \(E_\psi \) in the set \(\{u^T_k\,\ T\in D_L\}\); moreover, we show that, if the minimum is achieved by a map \(T^L_k\) in \(D_{L_k}\) with \(L_k\in ]0,L[\), then the corresponding function \(u^{T_k^L}_k\) satisfies the stationarity condition which allows us to conclude that it is a solution of problem (1.1) for k large enough.
Indeed, we show that there exists a sequence \((L_k)_k\) of positive numbers such that \(\lim _{k\rightarrow \infty }L_k=0\) and \(T^L_k\in D_{L_k}\) \(\forall k\in {\mathbb {N}}\), so \(L_k\in ]0,L[\) for k large enough and the solution \(u_k=u^{T_k^L}_k\) satisfies all the assertions of Proposition 2.2 (in analogous way one can construct the solutions \(v_k\) that behaves as \(-u_k\) when \(k\rightarrow \infty \)).
In particular, we obtain that \(T^L_k\) tends as \(k\rightarrow \infty \) to the identity map in \(\overline{\Omega }\) and that the rescaled nodal regions \(kT^L_k(C_i^k)\), after translations, tend to the cube \(\Omega \) as \(k\rightarrow \infty \), uniformly with respect to \(i\in \{1,\ldots ,k^n\}\).
The existence of such a sequence \((L_k)_k\), which plays a crucial role in the proof, is strictly related to a minimality property of the cubes in \({\mathbb {R}}^n\). In fact, the functions \(u^T_k\left( \frac{x}{k}\right) \), suitably rescaled, tend as \(k\rightarrow \infty \) to solutions of the equation (1.1) with \(\psi =0\). Therefore, since the effect of the term \(\psi \) tends to vanish as \(k\rightarrow \infty \), the rescaled nodal regions \(kT^L_k(C^k_i)\), suitably translated, tend to polyhedra as \(k\rightarrow \infty \). Among these polyhedra, the cubes of \({\mathbb {R}}^n\) are the unique minimizers of the “shape factor” \({\varphi }(\chi )\) defined by
where \(|\chi |\) is the volume of \(\chi \) and
(notice that \({\varphi }(\chi )\) depends only on the shape of \(\chi \) and not on its size because it is invariant with respect to translations and rescaling of \(\chi \)). Therefore, taking into account the asymptotic behaviour of \(E_\psi (u^{T^L_k}_k)\) as \(k\rightarrow \infty \), the minimality of \(T^L_k\) implies that \({\varphi }(k T^L_k(C^k_i))\rightarrow \overline{{\varphi }}\), where \(\overline{{\varphi }}\) denotes the shape factor of every cube of \({\mathbb {R}}^n\), while the volumes \(|kT^L_k(C^k_i)|\rightarrow 1\) as \(k\rightarrow \infty \), uniformly with respect to \(i\in \{1,\ldots ,k^n\}\).
As a consequence, taking also into account the conditions of \(T^L_k\) on \(\partial \Omega \), we infer that, after translations, the rescaled nodal regions \(kT^L_k(C^k_i)\) tend to \(\Omega \) as \(k\rightarrow \infty \) and there exists a sequence \((L_k)_k\) having the desired properties.
It is clear that our method does not require techniques of deformations from the symmetry and may be applied to more general problems. For example, it may be easily adapted to deal with the case where in problem (1.1) the nonlinear term \(|u|^{p-1}u\) is replaced by \(c_+(u^+)^p-c_-(u^-)^p\) with \(c_+\) and \(c_-\) two positive constants. Moreover, this method may be adapted to work even in case of nonlinear elliptic equations involving critical Sobolev exponents. For example, it allows us to obtain in this case a multiplicity result similar to Theorem 2.1, which is announced in Theorem 3.18.
Now, we can compare results and methods of the present paper with those of [34] in order to state clearly a relation between these papers.
In both papers the starting point of our approach is to observe that, for a given \(f\in L^2(\Omega )\), there exists a mountain pass solution in every sufficiently small subdomain of \(\Omega \). Then, we decompose the domain \(\Omega \) into disjoint union of small subdomains which are given by the images of disjoint union of small cubes through a class of admissible Lipschitz maps. Next, on each subdomains we dispose mountain pass solutions, which are perturbations of (positive or negative) mountain pass solutions of problem (1.1) for \(f=0\) (in such a way that the solutions in adjacent subdomains have different signs). Finally, we minimize the energy functional among the class of admissible Lipschitz maps and show that, if a minimizing map is (in a suitable sense) in the interior of the class of the admissible maps, the normal derivatives of the corresponding functions on the interfaces of the subdomains are equal and so we obtain infinitely many solutions (because the cubes may be arbitrarily small).
However, even if the scheme of our approach is similar in both papers, the class of Lipschitz maps considered in [34] is quite different from the class \(D_L\) we use in the present paper. In fact, in [34] we denote by \(P_k\) the union of all the cubes with sides of length \(\frac{1}{k}\) and vertices in \(\frac{1}{k}{\mathbb {Z}}^n\), that are enclosed in \(\Omega \) and, for \(L'>1\), we consider the class \({{\mathcal {C}}}_{L'}(P_k,\overline{\Omega })\) consisting of the bilipschitz maps between \(P_k\) and \(\overline{\Omega }\) with Lipschitz constants in \(\left[ \frac{1}{L'},L'\right] \). Moreover, we show that, if the minimal energy is achieved by a bilipschitz map \(T^{L'}_k\) with Lipschitz constants in \(\left]\frac{1}{L'},L'\right[\), the corresponding function \(u_k^{T_k^{L'}}\) is a solution of problem (1.1).
Thus, in [34] we prove the existence of infinitely many solutions under a geometric condition on the domain \(\Omega \), which guarantees that, for a suitable choice of \(L'>1\), the minimum is achieved by a map \(T^{L'}_k\) which, for k large enough, belongs to \({{\mathcal {C}}}_{L''}(P_k,\overline{\Omega })\) for some \(L''\in \left]\frac{1}{L'},L'\right[\).
This geometric condition may be easily checked in dimension \(n=1\) because, since for all \(L'>1\) the minimality of \(T_k^{L'}\) implies that the subdomains tend to have all the same size as \(k\rightarrow \infty \), for \(n = 1\) we infer that \(T_k^{L'}\in {{\mathcal {C}}}_{L'_k}(P_k,\overline{\Omega })\), where \(L'_k\in [1,L']\) \(\forall k\in {\mathbb {N}}\) and \(L'_k\rightarrow 1\) as \(k\rightarrow \infty \).
On the contrary, for \(n >1\) it is difficult to check this geometric condition because, even if the subdomains tend to have all the same size, for \(n>1\) they might have a shape very different from the cubes, so it is difficult to control the Lipschitz constants of the map \(T^{L'}_k\) as \(k\rightarrow \infty \).
For this reason in the present paper we use the class \(D_L\) of admissible Lipschitz maps.
Moreover, let us point out that the approach used in the present paper and the class \(D_L\) of Lipschitz maps work not only for cubes but also for other domains that can be decomposed into disjoint union of translations of scaling domains (for example, equilateral triangles, triangular prisms, rectangles, rectangular parallelepipeds, etc.).
In fact, in order to apply our method it is sufficient that the domain \(\Omega \) admits tessellations by arbitrarily small congruent subdomains, symmetric by reflection with respect to all the interfaces and colourable with two colors in such a way that adjacent subdomains have different colors, so that we can have a criterion to choose the sign in every subdomain and the functions have the same normal derivatives on their interfaces (for example, tessellations by hexagons or hexagonal prisms do not work in our arguments).
In the present paper we describe our method only for cubic domains, for the sake of simplicity, because for the cubes the same arguments hold in every dimension n (on the contrary, for example, the tetrahedra do not work for \(n=3\) as the equilateral triangles do for \(n = 2\)).
2 Variational framework and statement of the main results
Our aim is to prove the following theorem.
Theorem 2.1
Let \(\Omega \) be a cube of \({\mathbb {R}}^n\) with \(n\ge 1\), let \(p>1\) and \(p<\frac{n+2}{n-2}\) when \(n\ge 3\). Then, for every \(\psi \in L^2(\Omega )\), problem (1.1) admits infinitely many solutions.
Without any loss of generality, we can assume that
For all positive integer k and for all \(z\in {\mathbb {Z}}^n\), let us set
(thus, in particular, we have \(C_0^1=\Omega \)).
Notice that for all \(k\in {\mathbb {N}}\) we have \(C_z^k\subseteq \Omega \) if and only if \(0\le z_i\le k-1\) for \(i=1,\ldots ,n\); moreover, if we set
we have
Then, the following proposition holds (it obviously implies Theorem 2.1).
Proposition 2.2
Under the assumptions of Theorem 2.1, if \(\Omega \) is the cube (2.1), for all \(\psi \in L^2(\Omega )\) there exists \({\bar{k}}\in {\mathbb {N}}\) such that for every \(k\ge {\bar{k}}\) problem (1.1) admits two solutions \(u_k\) and \(v_k\) having the following properties (here we consider \(u_k\) and \(v_k\) extended by the value zero in \({\mathbb {R}}^n\setminus \Omega \)). For all \(k\ge {\bar{k}}\) there exist two bilipschitz maps \(T_{k,u},T_{k,v}:\Omega \rightarrow \Omega \) (with Lipschitz constants independent of k) such that for every choice of \(z^k\) in \(Z_k\) the functions \(U_{z^k}\) and \(V_{z^k}\) defined by
restricted to \(\Omega \), both converge as \(k\rightarrow \infty \) to a positive solution of problem (1.1) with \(\psi \equiv 0\) in \(\Omega \), satisfying
Moreover, the sequences \((T_{k,u})_k\) and \((T_{k,v})_k\) both converge to the identity map uniformly in \(\Omega \), while the domains \(k\left[ T_{k,u}\left( C^k_{z^k}\right) - T_{k,u}\left( \frac{z^k}{k}\right) \right] \) and \(k\left[ T_{k,v}\left( C^k_{z^k}\right) - T_{k,v}\left( \frac{z^k}{k}\right) \right] \) tend to \(\Omega \) as \(k\rightarrow \infty \) for every choice of \(z^k\) in \(Z_k\).
The proof is reported in Sect. 3.
In order to prove Theorem 2.1 and Proposition 2.2, we proceed as follows. For every \(t\in [0,1]\) and \(i\in \{1,\ldots ,n\}\), let us consider the set
(in particular, if \(t=0\) or \(t=1\), \(F_i^t\) is a face of the cube \(\Omega \)).
Now, let us fix \(L\in ]0,1[\) and consider the set \(D_L\) of the admissible deformations of \(\overline{\Omega }\) defined by
Notice that for every deformation \(T\in D_L\) one can write \(T(x)=I(x)+S(x)\) where \(I(x)=x\) \(\forall x\in \overline{\Omega }\) and \(S:\overline{\Omega }\rightarrow {\mathbb {R}}^n\) is a Lipschitz continuous function with Lipschitz constant L. Moreover, we have
where \(1-L>0\) because we assumed \(L\in ]0,1[\). Thus, T is invertible and both T and \(T^{-1}\) are Lipschitz continuous functions in \(\overline{\Omega }\).
Other important consequences of the definition of \(D_L\) are presented in next proposition where we describe some geometrical properties of the deformations \(T(F^t_i)\) of the sets \(F^t_i\) with respect to the straight lines orthogonal to \(F^t_i\) (these properties motivate the introduction of this class of admissible deformations).
Proposition 2.3
Let \(T\in D_L\) and \(L\in ]0,1[\). Then
- (a):
-
for all \(t\in [0,1]\), \(i\in \{1,\ldots ,n\}\) and \(y\in \Omega \) there exists a unique \(x\in F^t_i\) such that \(P_i\circ T(x)=P_i(y)\), where \(P_i\) denotes the orthogonal projection of \({\mathbb {R}}^n\) on the subspace \(\{x=(x_1,\ldots ,x_n)\in {\mathbb {R}}^n\,\ x_i=0\}\) (that is, every straight line orthogonal to \(F^t_i\) meets \(T(F^t_i)\) in a unique point);
- (b):
-
for all \(t'\), \(t''\) in [0, 1] such that \(t'<t''\) and for all \(x'\in F^{t'}_i\) and \(x''\in F^{t''}_i\) such that \(P_i\circ T(x')=P_i\circ T(x'')\), we have \(T_i(x')<T_i(x'')\) (that is, the deformation \(T(F^t_i)\) of the set \(F^t_i\) meets every straight line orthogonal to \(F^t_i\) in a unique point whose \(i^{\textrm{th}}\) coordinate increases as t increases).
Proof
In order to prove (a), first notice that, for all \(t\in [0,1]\), \(i\in \{1,\ldots ,n\}\) and \(y\in \Omega \), there exists \(x\in F^t_i\) such that \(P_i\circ T(x)=P_i(y)\).
In fact, let us consider the function \(P_i\circ T:F^{t}_i\rightarrow F^0_i\), which is a continuous function satisfying
Therefore, since \(P_i(y)\in F^0_i\), there exists \(x\in F^t_i\) such that \(P_i\circ T(x)= P_i(y)\) (as follows from [33]).
Now, let us prove that such a x is unique. Arguing by contradiction, assume that there exists another \({\tilde{x}}\) in \(F^{t}_i\), \({\tilde{x}}\ne x\), such that \(P_i\circ T({\tilde{x}})=P_i(y)\), which implies
Since \(T\in D_L\) with \(L\in ]0,1[\), we infer that
and, as a consequence,
because \(x\ne {\tilde{x}}\). On the other hand, from (2.12) we obtain
in contradiction with (2.14).
Thus, (a) is completely proved.
In order to prove (b), we argue again by contradiction and assume that there exist \(t',t''\) in [0, 1] such that \(t'<t''\) and \(x'\in F^{t'}_i\), \(x''\in F^{t''}_i\) such that
Notice that (2.16) implies
Therefore, we obtain
On the other hand, since \(T\in D_L\) with \(L\in ]0,1[\) and \(x'\ne x''\), we infer that
in contradiction with (2.18).
Thus, we can conclude that, if \(P_i\circ T(x')=P_i\circ T(x'')\) and \(t'<t''\), then \(T_i(x')<T_i(x'')\), so the proof is complete. \(\square \)
Now, we exploit the class of admissible deformations \(D_L\) in order to construct the solutions \(u_k\) and \(v_k\). We first construct the solutions \(u_k\) (then one can proceed in a similar way to construct the solutions \(v_k\)). For all \(k\in {\mathbb {N}}\), \(z\in Z_k\) and \(T\in D_L\) with \(L\in ]0,1[\), let us set
Since \(p<\frac{n+2}{n-2}\) when \(n\ge 3\), one can easily verify that the infimum in (2.20) is achieved.
Moreover, for all \(k\in {\mathbb {N}}\) and \(L\in ]0,1[\), also the infimum
is achieved, as one can prove by standard arguments using Ascoli-Arzelà Theorem.
For the construction of the functions \(u_k\) we need the following Lemmas.
Lemma 2.4
For all \(L\in ]0,1[\) we have
and there exists \(k(L)\in {\mathbb {N}}\) such that, for all \(k\ge k(L)\), \(z\in Z_k\) and \(T\in D_L\), the infimum
is achieved by a unique minimizing function \({\tilde{u}}^T_{k,z}\). Moreover, we have
Proof
For all \(k\in {\mathbb {N}}\), let us consider \(z^k\in Z_k\) and \(T_k\in D_L\) realizing the minimum (2.22) and \({\bar{u}}_k\in H^1_0(T(C^k_{z^k}))\) realizing the minimum \(E_\psi (k,z^k,T_k)\).
Let us extend the function \({\bar{u}}_k\) in all of \(\Omega \) by the value zero in \(\Omega \setminus C^k_{z^k}\). Since \(T_k\in D_L\) \(\forall k\in {\mathbb {N}}\), taking into account the second inequality in (2.10), we obtain
so (up to a subsequence) \({\bar{u}}_k\rightarrow 0\) almost everywhere in \(\Omega \). It follows that
otherwise, since \(p<\frac{n+2}{n-2}\) for \(n\ge 3\), \({\bar{u}}_k\rightarrow 0\) also in \(L^{p+1}(\Omega )\), which is impossible because \(\int _{\Omega }|{\bar{u}}_k|^{p+1}dx=1\) \(\forall k\in {\mathbb {N}}\). As a consequence, since
we obtain (2.22).
Notice that (2.22) implies that for all \(L\in ]0,1[\) there exists \(k(L)\in {\mathbb {N}}\) satisfying
Since \(E_\psi (0)=0\), it follows by standard arguments that for all \(k\ge k(L)\), \(z\in Z_k\) and \(T\in D_L\) there exists \({\tilde{u}}^T_{k,z}\in H^1_0(T(C^k_z))\) such that
Taking into account that
it follows that
In order to prove (2.24), we argue by contradiction and assume that for all \(k\ge k(L)\) there exist \(z^k\in Z_k\) and \(T_k\in D_L\) such that
From (2.31) we infer that the sequence \(({\tilde{u}}^{T_k}_{k,z^k})_k\) (with \({\tilde{u}}^{T_k}_{k,z^k}\) extended by the value zero outside \(T_k(C^k_{z^k})\)) is bounded in \(H^1_0(\Omega )\). Moreover, up to a subsequence, \({\tilde{u}}^{T_k}_{k,z^k}\rightarrow 0\) as \(k\rightarrow \infty \) almost everywhere in \(\Omega \) because \(T_k\in D_L\), so \(\textrm{meas}(T_k(C^k_{z^k}))\rightarrow 0\) as \(k\rightarrow \infty \). Therefore, \({\tilde{u}}^{T_k}_{k,z^k}\rightarrow 0\) as \(k\rightarrow \infty \) also in \(L^{p+1}(\Omega )\). Then, from \(E_\psi ({\tilde{u}}^{T_k}_{k,z^k})\le 0\) \(\forall k\in {\mathbb {N}}\) it follows easily that
in contradiction with (2.32). Thus, we can conclude that (2.24) holds.
Finally, notice that the functional \(E_\psi \) is strictly convex in a suitable neighborhood of zero. Therefore, for k large enough, \({\tilde{u}}^{T_k}_{k,z^k}\) is the unique minimizing function for (2.23) for all \(z\in Z_k\) and \(T\in D_L\). So the proof is complete. \(\square \)
Lemma 2.5
For all \(k\ge k(L)\), \(z\in Z_k\) and \(T\in D_L\), there exists a function \( u^{T}_{k,z}\) in \(H^1_0(T(C^k_z))\) such that \(u^{T}_{k,z}\not \equiv {\tilde{u}}^{T}_{k,z}\), \(\sigma (z)[u^{T}_{k,z}-{\tilde{u}}^{T}_{k,z}]\ge 0\) in \(T(C^k_z)\) and
where, for all \(u\in H^1_0(T(C^k_z))\), \(M_\psi (u)\) is defined by
Proof
First notice that the maximum in (2.35) is achieved for all \(u\in H^1_0(T(C^k_z))\) because \(p>1\). Now, let us consider a sequence \((u_i)_i\) in \(H^1_0(T(C^k_z))\) such that \(u_i\not \equiv {\tilde{u}}^{T}_{k,z}\), \(\sigma (z)[ u_i-{\tilde{u}}^{T}_{k,z} ]\ge 0\) in \(T(C^k_z)\) \(\forall i\in {\mathbb {N}}\) and
Then, let us set \(w_i=\Vert u_i-{\tilde{u}}^{T}_{k,z}\Vert ^{-1}_{L^{p+1}}( u_i-{\tilde{u}}^{T}_{k,z})\) and notice that, obviously, \(M_\psi (u_i)=M_\psi ({\tilde{u}}^{T}_{k,z}+w_i)\). Moreover notice that, since the sequence \((w_i)_i\) is bounded in \(L^{p+1}\), (2.36) implies that it is bounded also in \(H^1_0\). Since \(p<\frac{n+2}{n-2}\) when \(n\ge 3\), it follows that (up to a subsequence) \((w_i)_i\) converges weakly in \(H^1_0\), in \(L^{p+1}\) and almost everywhere to a function \({\hat{w}}\in H^1_0(T(C^k_z))\). As a consequence, \(\Vert {\hat{w}}\Vert _{L^{p+1}}=1\) and \(\sigma (z){\hat{w}}\ge 0\) in \(T(C^k_z)\). Indeed, \(w_i\rightarrow {\hat{w}}\) as \(i\rightarrow \infty \) strongly in \(H^1_0(T(C^k_z))\). In fact, since we have the weak convergence, arguing by contradiction assume that \(\Vert w_i\Vert ^2_{H^1_0}\) does not converge to \(\Vert {\hat{w}}\Vert ^2_{H^1_0}\) as \(i\rightarrow \infty \), that is
which, combined with the weak convergence, implies \(M_\psi ({\tilde{u}}+{\hat{w}})<\lim _{i\rightarrow \infty }M_\psi ({\tilde{u}}+w_i)\). Therefore, we obtain a contradiction because \({\hat{w}}\not \equiv 0\) and, as a consequence, \( \lim _{i\rightarrow \infty } M_\psi ({\tilde{u}}+w_i)\le M_\psi ({\tilde{u}}+{\hat{w}}) \) because of (2.36). Thus, we can conclude that \(w_i\rightarrow {\hat{w}}\) in \(H^1_0( T(C^k_z))\) as \(i\rightarrow \infty \), which imples \(\lim _{i\rightarrow \infty }M_\psi ({\tilde{u}}+w_i)=M_\psi ({\tilde{u}}+{\hat{w}})\).
Moreover, since \(p>1\), there exists \({\hat{t}}>0\) such that \(E_\psi ({\tilde{u}}^T_{k,z}+{\hat{t}}{\hat{w}})=M_\psi ({\tilde{u}}^T_{k,z}+{\hat{t}}{\hat{w}})\), so all the assertions in Lemma 2.5 hold with \(u^T_{k,z}={\tilde{u}}^T_{k,z}+{\hat{t}}{\hat{w}}\). \(\square \)
Remark 2.6
Notice that the function \(u^T_{k,z}\) given by Lemma 2.5, for k large enough, satisfies \(E_\psi (u^T_{k,z})\ge E_\psi (k,z,T)\) because \(u^T_{k,z}\not \equiv {\tilde{u}}^T_{k,z}\) in \(T(C^k_z)\).
Thus, by (2.22) we get
Now, we extend every function \(u^T_{k,z}\) in all of \(\Omega \) by the value zero outside \( T(C^k_z)\) and we consider the function \(u^T_k\in H^1_0(\Omega )\) defined by \(u^T_k=\sum _{z\in Z_k}u^T_{k,z}\). Using Ascoli–Arzelà Theorem, one can verify that for all \(k\ge k(L)\) there exists an admissible deformation \(T^L_k\in D_L\) such that
In next section we show that \(u^{T^L_k}_k\) is a solution of problem (1.1) for k large enough and that Proposition 2.2 holds with \(u_k=u^{T^L_k}_k\) and \(T_{k,u}=T^L_k\). In order to construct the solutions \(v_k\), we proceed in analogous way. In fact, as in Lemma 2.5, for all \(k\ge k(L)\), \(z\in Z_k\) and \(T\in D_L\), there exists also a function \(v^T_{k,z}\) in \(H^1_0(T(C^k_z))\) such that \(v^T_{k,z}\ne {\tilde{u}}^T_{k,z}\), \(\sigma (z)[v^T_{k,z}-{\tilde{u}}^T_{k,z}]\le 0\) in \(T(C^k_z)\) and
Then, we set \(v^T_k=\sum _{z\in Z_k}v^T_{k,z}\) (where \(v^T_{k,z}\) is extended in \(\Omega \) by the value zero outside \(T(C^k_z)\)) and, using Ascoli–Arzelà Theorem, we minimize \(E_\psi (v^T_k)\) with respect to T in \(D_L\). If \(T_{k,v}\in D_L\) is a minimizing admissible deformation, the function \(v_k^{T_{k,v}}\) is a solution of problem (1.1) for k large enough and Proposition 2.2 holds with \(v_k=v_k^{T_{k,v}}\), as we show in next section.
3 Asymptotic estimates and proof of the main results
In this section we describe the asymptotic behaviour as \(k\rightarrow \infty \) of the functions \(u_k\) and \(v_k\), arising in Proposition 2.2, we constructed in Sect. 2. Then, we show that these functions are solutions of problem (1.1) for k large enough and satisfy all the assertions of Proposition 2.2.
As follows from Proposition 2.3, for all \(T\in D_L\), \(i\in \{1,\ldots ,n\}\) and \(t\in [0,1]\), the set \(T(F^t_i)\) is the graph of a function \(f^{t,T}_i:F^0_i\rightarrow {\mathbb {R}}\) and
In next lemma we prove that \(f^{t,T}_i\) is a Lipschitz continuous function.
Lemma 3.1
If \(T\in D_L\) with \(L\in ]0,1[\), then for all \(i\in \{1,\ldots ,n\}\) and \(t\in [0,1]\) we have
Proof
For all x, y in \(F^0_i\), there exist \(x^t,y^t\) in \(F^t_i\) such that \(P_i\circ T(x^t)=x\), \(P_i\circ T(y^t)=y\) and, as a consequence, \(f^{t,T}_i(x)=T_i(x^t)\), \(f^{t,T}_i(y)=T_i(y^t)\).
Thus, since \(x^t_i=y^t_i=t\) and \(T\in D_L\), we obtain
Moreover, since \(L\in ]0,1[\), we obtain
which, combined with (3.3), implies (3.2). \(\square \)
Let us denote by \(\textrm{Lip}(f^{t,T}_i)\) the best Lipschitz constant of the function \(f^{t,T}_i\), that is
Then, from (3.2) it follows that \(\textrm{Lip}(f^{t,T}_i)\rightarrow 0\) as \(L\rightarrow 0\).
Corollary 3.3 shows, in some sense, that also the converse in true. Notice that, if we set \(S_T(x)=T(x)-x\) \(\forall x\in \Omega \), then \(T\in D_L\) if and only if
Moreover, it is obvious that the set \(D_L\) may be also written as
Lemma 3.2
Let \(T\in D_L\) with \(L\in ]0,1[\) and assume that there exists \(\Lambda \in \left]0,\frac{1}{n}\right[\) such that
and
Then,
that is \(\textrm{Lip}(S_T)\le \frac{(n+1)\sqrt{n} \,\Lambda }{1-n\Lambda }\), so \(T\in D_{L(\Lambda )}\) with \(L(\Lambda )=\frac{(n+1)\sqrt{n} \, \Lambda }{1-n\Lambda }\).
Proof
Notice that \(T_i(x)=f_i^{x_i,T}(P_i\circ T(x))\) for all \(x\in \overline{\Omega }\) and \(i\in \{1,\ldots ,n\}\).
Thus, for \(x,y\in \Omega \) and \(h=y-x\), we obtain
where, for all i and j in \(\{1,\ldots ,n\}\), \(\mu _i\) and \(\nu _i^j\) are suitable numbers in \([-\Lambda ,\Lambda ]\) because of our assumptions on the functions \(f^{t,T}_i\).
It follows that
and, summing up,
Since \(\Lambda <\frac{1}{n}\), we obtain
which implies
So the proof is complete. \(\square \)
The following corollary is a direct consequence of Lemma 3.2.
Corollary 3.3
Let \((T_k)_k\) be a sequence in \(D_L\) with \(L\in ]0,1[\) and assume that, for a suitable sequence \((\Lambda _k)_k\) in \(\left]0,\frac{1}{n} \right[\), the same conditions as in Lemma 3.2 are satisfied with T replaced by \(T_k\) and \(\Lambda \) by \(\Lambda _k\) for all \(k\in {\mathbb {N}}\).
Then, \(\lim _{k\rightarrow \infty }\Lambda _k=0\) implies \(\lim _{k\rightarrow \infty }\textrm{Lip}(S_{T_k})=0\).
Remark 3.4
Notice that, if \(\textrm{Lip}(S_{T_k})\mathop {\longrightarrow }0\) as \(k\rightarrow \infty \), then \(S_{T_k}\) converges to a constant function \(S_\infty \) uniformly in \(\Omega \). Moreover, taking into account that \(T_k\in D_L\) \(\forall k\in {\mathbb {N}}\) so \(T_k\) must satisfy suitable conditions on \(\partial \Omega \), we can say that \(S_\infty \equiv 0\), that is \(T_k\) converges to the identity function in \(\Omega \).
Now, let us prove the assertions of Proposition 2.2 for the function \(u_k=u^{T^L_k}_k\) (in a similar way one can proceed for the function \(v_k=v^{T_k,v}_k\)). First, we prove the following proposition (here we use the notation introduced in Lemmas 2.4 and 2.5).
Proposition 3.5
For all \(k\ge k(L)\) the function \(u_k=u^{T^L_k}_k\) (extended to \({\mathbb {R}}^n\) by the value zero in \({\mathbb {R}}^n\setminus \Omega \)) has the following asymptotic behaviour.
For every choice of \(z^k\) in \(Z_k\), there exists a function \({\hat{T}}:\overline{\Omega }\rightarrow {\mathbb {R}}^n\) such that
and, if we set \(\chi :={\hat{T}}(\Omega )\), the function \(U_{z^k}\) defined by
restricted to \(\chi \), as \(k\rightarrow \infty \) converges in \(H^1(\chi )\) to a positive solution \(U_\chi \) of the Dirichlet problem
satisfying
where
Moreover, we have
Proof
For all \(k\in {\mathbb {N}}\), let us rescale problem (1.1) by replacing every function \(u\in H^1_0(\Omega )\) by the function \(R^ku\in H^1_0(k\Omega )\) defined by
(here \(R^ku\) is extended by the value zero outside \(k\Omega \)).
Then, our problem becomes
where \(\psi _k\in L^2(k\Omega )\) is defined by
Moreover, the corresponding functional becomes
defined for all \(u\in H^1_0(k\Omega )\).
Since \(T^L_k\in D_L\) \(\forall k\in {\mathbb {N}}\), so in particular it satisfies (2.10), also the function \(kT^L_k\left( \frac{x+z^k}{k}\right) \), defined for all \(x\in \overline{\Omega }\), satisfies (2.10) and, as a consequence,
Therefore, using Ascoli–Arzelà Theorem, we infer that (up to a subsequence) the function \(kT^L_k\left( \frac{\cdot \, +z^k}{k}\right) -kT^L_k\left( \frac{z^k}{k}\right) \) converges as \(k\rightarrow \infty \) to a function \({\hat{T}}:\overline{\Omega }\mathop {\longrightarrow }B(0,(1+L)\sqrt{n})\) uniformly in \(\overline{\Omega }\). As a consequence, \({\hat{T}}\) satisfies (3.16) in \(\overline{\Omega }\).
From Lemmas 2.4 and 2.5 we infer that \(R^k{\tilde{u}}^{T^L_k}_{k,z^k}\) and \(R^k u^{T^L_k}_{k,z^k}\) belong to \(H^1_0(kC^k_{z^k})\) and that
Moreover,
where \(M^k(U)\) is defined by
Notice that
where \(n<\frac{4p}{p-1}\) under our assumptions on p. In fact, for \(n\le 4\) it is obviously true because \(p>1\) while for \(n>4\) it is true because \(1<p<\frac{n+2}{n-2}\), as one can easily verify by direct computation (taking into account that \(\frac{n+2}{n-2}<\frac{n}{n-4}\)). As a consequence, we obtain in particular
Therefore, we infer that the function \(U_{z^k}\) satisfies all the assertions in Proposition 3.5, that is its restriction to \(\chi \) converges to a positive solution \(U_\chi \) of the asymptotic problem (3.18), satisfying the minimality condition (3.19).
In fact, (3.28) and (3.32) imply
and
It follows that
so, up to a subsequence, \(\left( \int _\chi |U_{z^k}|^{p+1}dx\right) ^{-\frac{1}{p+1}}U_{z^k}\) converges as \(k\rightarrow \infty \) to a positive function \({\overline{U}}_\chi \in H^1_0(\chi )\) almost everywhere in \(\chi \), strongly in \(L^{p+1}(\chi )\) and weakly in \(H^1(\chi )\).
Moreover, the minimality property of \(u^{T^L_k}_{k,z^k}\) implies, by standard arguments, that
and that, as \(k\rightarrow \infty \), \(U_{z^k}\) converges strongly in \(H^1(\chi )\) to the function \(U_\chi =m(\chi )^{\frac{1}{p-1}}{\overline{U}}_\chi \), which is a positive solution of problem (3.18).
Therefore, taking also into account (3.24), we obtain
So the proof is complete. \(\square \)
In next lemma we describe other properties of the function \({\hat{T}}\) and of the domain \({\hat{T}}(\Omega )\) arising in Proposition 3.5.
Lemma 3.6
Let \((z^k)_k\), \({\hat{T}}\) and \(\chi \) be as in Proposition 3.5. Then, the function \(S_{{\hat{T}}}:\overline{\Omega }\rightarrow {\mathbb {R}}^n\) defined by \(S_{{\hat{T}}}(x)={\hat{T}}(x)-x\) \(\forall x\in \overline{\Omega }\), satisfies the Lipschitz condition
Moreover, for every \(i\in \{1,\ldots ,n\}\) there exist two functions \(f^0_i,f^1_i:P_i(\chi )\rightarrow {\mathbb {R}}\), Lipschitz continuous with Lipschitz constant \(\frac{L}{1-L}\), such that \(f^0_i\circ P_i(0)=0\), \(f^0_i\circ P_i(x)<f^1_i\circ P_i(x)\) \(\forall x\in \chi \) and
Proof
Notice that, as the functions \(S_{T^L_k}:\overline{\Omega }\rightarrow {\mathbb {R}}^n\) defined by \(S_{T^L_k}(x)=T^L_k(x)-x\) \(\forall x\in \Omega \), also the functions \(k\left[ T^L_k\left( \frac{x+z^k}{k}\right) -T^L_k\left( \frac{z^k}{k}\right) -\frac{x}{k}\right] \) are Lipschitz continuous with Lipschitz constant L for all \(k\in {\mathbb {N}}\). Therefore, as \(k\rightarrow \infty \), we infer that the function \(S_{{\hat{T}}}\) satisfies (3.38). In order to obtain the functions \(f^0_i\) and \(f^1_i\), we use Lemma 3.1. From (3.1) it follows that
Now, notice that, as the functions \(f_i^{\frac{z^k_i}{k},T^L_k}\) and \(f_i^{\frac{z^k_i+1}{k},T^L_k}\), also the functions \(f^0_{i,k}\) and \(f^1_{i,k}\) defined by
and
are both Lipschitz continuous with Lipschitz constant \(\frac{L}{1-L}\). Moreover, for all \(k\in {\mathbb {N}}\) we have
and
where \(x^1_{i,k}\) and \(x^0_{i,k}\) are the points in \(\overline{\Omega }\) such that
and
which implies \(|x^1_{i,k}-x^0_{i,k}|\ge 1\).
Therefore, by Ascoli–Arzelà Theorem we can say that, up to a subsequence, the functions \(f^1_{i,k}\) and \(f^0_{i,k}\) converge as \(k\rightarrow \infty \) uniformly in \(P_i(\overline{\chi })\) respectively to functions \(f^1_i\) and \(f^0_i\) satisfying all the assertions in Lemma 3.6. \(\square \)
Lemma 3.7
Let \((z^k)_k\) and \(\chi \) be as in Proposition 3.5. Then, for every choice of \(z^k\) in \(Z_k\), the domain \(\chi \) is a cube of \({\mathbb {R}}^n\) having a vertex in the origin and the sides of lenght 1. Moreover, we have
(where \(|T^L_k(C^k_z)|\) denotes the volume of \(T^L_k(C^k_z)\)) and
where \({\bar{m}}=m(\Omega )\).
Proof
Notice that, as we pointed out in the proof of Proposition 3.5, the effect of the term \(\psi \) in problem (1.1) tends to vanish as \(k\rightarrow \infty \) because in the rescaled problem (3.23) \(\psi \) is replaced by the function \(\psi _k\) defined by (3.31) and, since \(n<\frac{4p}{p-1}\),
As a consequence, taking into account the minimality of \(T^L_k\), the interfaces between the domains \(kT^L_k(C^k_z)\), with \(z\in Z_k\), tend to be flat, so these domains tend as \(k\rightarrow \infty \) to polyhedra with 2n faces, having minimality properties inherited by the analogous properties of the domains \(kT^L_k(C^k_z)\), related to the minimality of \(T^L_k\).
In particular, arguing as in the proof of Proposition 3.5, one can show in addition that, for every \(\rho >0\), the function \(k\left[ T^L_k\left( \frac{x+z^k}{k}\right) -T^L_k\left( \frac{z^k}{k}\right) \right] \) (up to a subsequence) converges as \(k\rightarrow \infty \) to a function \({\hat{T}}\) uniformly in the domain \(\Omega _\rho =\cup _{z\in {\mathbb {Z}}_\rho }{\overline{C}}^1_z\), where
(not only in \(\overline{\Omega }\), which is strictly enclosed in \(\Omega _\rho \) for \(\rho >\sqrt{n}\)).
Moreover, if we set \(\chi _z:={\hat{T}}(C^1_z)\) \(\forall z\in {\mathbb {Z}}_\rho \), the number \(\sum _{z\in {\mathbb {Z}}_\rho }[m(\chi _z)]^{\frac{p+1}{p-1}}\) has to be as small as possible for all \(\rho >0\). By symmetry reasons, among this polyhedra \(\chi \), the cubes of \({\mathbb {R}}^n\) are the unique minimizers of the value \({\varphi }(\chi ):=m(\chi )|\chi |^{2\left( \frac{1}{p+1}-\frac{1}{2^*}\right) }\) (where \(|\chi |\) is the volume of \(\chi \) and \(m(\chi )\) is defined in (3.20)) that is, if we set \({\bar{{\varphi }}}={\varphi }(\Omega )\), \({\varphi }(\chi )={\bar{{\varphi }}}\) if \(\chi \) is a cube and \({\varphi }(\chi )>{\bar{{\varphi }}}\) otherwise (notice that \({\varphi }(\chi )\) depends only on the shape of \(\chi \) and not on its size because it is invariant with respect to translations and rescaling of \(\chi \)).
By Proposition 3.5, for every choice of \(z^k\) in \(Z_k\), the corresponding limit domain \(\chi \) satisfies
and
which implies
Therefore, taking into account the minimality of \(T^L_k\), it follows that, for every choice of \(z^k\) in \(Z_k\), the limit domain \(\chi \) is a cube, that is \({\varphi }(\chi )={\bar{{\varphi }}}\). As a consequence, since it is true for every choice of \(z^k\) in \(Z_k\), we can say that
Now, let us set
Then, we have
Taking into account the minimality of \(T^L_k\), since \(\sum _{z\in Z_k}|T^L_k(C^k_z)|=1\), we obtain
where \(\mu _k>0\) is a suitable Lagrange multiplier. It follows that
which, summing up, yields
Since
(because of (3.54)), we obtain
which implies
From (3.57) we get
Thus, we can easily obtain (3.47) from (3.60) and (3.62) and then (3.48) from (3.47) and (3.54). Finally, we can say that, for every choice of \(z^k\) in \({\mathbb {Z}}_k\), the corresponding limit domain \(\chi \) is a cube of \({\mathbb {R}}^n\) with sides of lenght 1. Moreover, the construction of \(\chi \) shows that this cube has a vertex in the origin, so the proof is complete. \(\square \)
Lemma 3.8
For all \(k\in {\mathbb {N}}\) and \(L\in ]0,1[\), let \(T^L_k \in D_L\) be a minimizing deformation as in Sect. 2. Then, for all \(i\in \{1,\ldots ,n\}\), we have
Proof
Arguing by contradiction, assume that for some \(i\in \{1,\ldots ,n\}\) there exist sequences \((h_k)_k\), \((z^k)_k\), \((\zeta ^k)_k\) such that \(h_k\in \{0,1,\ldots ,k\}\), \(z^k\in Z_k\), \(\zeta ^k\in Z_k\), \(z^k\ne \zeta ^k\), \(z_i^k=\zeta ^k_i=h_k\) \(\forall k\in {\mathbb {N}}\) and (up to a subsequence)
We say that there exist two sequences \(({\hat{z}}^k)_k\) and \(({\hat{\zeta ^k}})_k\) in \(Z_k\) such that \({\hat{z}}^k_i={\hat{\zeta }}^k_i=h_k\), \(|{\hat{z}}^k-{\hat{\zeta ^k}}|=1\) \(\forall k\in {\mathbb {N}}\) and
In fact, if \(\limsup _{k\rightarrow \infty }|z^k-\zeta ^k|<2\), it is obvious because in this case \(|z^k-\zeta ^k|=1\) for k large enough (so we can set \({\hat{z}}^k=z^k\) and \({\hat{\zeta }}^k=\zeta ^k\)). On the contrary, if \(\limsup _{k\rightarrow \infty }|z^k-\zeta ^k|\ge 2\), let us set \(\nu ^k=\sum _{j=1}^n|z^k_i-\zeta _i^k|\). Then, one can choose \(\nu _k+1\) points \(\pi _0,\pi _1,\ldots ,\pi _{{\nu _{k}}}\) in \(Z_k\) such that
(notice that all the points \(\frac{\pi _0}{k}\), \(\frac{\pi _1}{k},\ldots ,\frac{\pi _{\nu _k}}{k}\) must belong to \(F_i^{\frac{h_k}{k}}\) because of the choice of \(\nu _k\)).
Therefore, we obtain
where
which implies
where
![](http://media.springernature.com/lw529/springer-static/image/art%3A10.1007%2Fs00526-023-02507-5/MediaObjects/526_2023_2507_Equ115_HTML.png)
and, by Lemma 3.1
Therefore, (3.65) implies
So, if the maximum in (3.73) is achieved for \(j= j_k\), our assertion (3.66) holds for \({\hat{z}}^k=\pi _{j_k}\) and \({\hat{\zeta }}^k=\pi _{j_k-1}\).
Now, for all \(i\in \{1,\ldots ,n\}\) let us consider the vector \(e^i=(e^i_1,\ldots ,e^i_n)\in {\mathbb {R}}^n\) such that \(e^i_i=1\), \(e^i_j=0\) for \(j\ne i\), \(i,j\in \{1,\ldots ,n\}\) and the function \(\delta _i^k:\Omega \rightarrow {\mathbb {R}}^n\) defined by
Notice that the set \(Z_\Omega =\cup _{k\in {\mathbb {N}}}\frac{1}{k} Z_k\) is a subset of \(\overline{\Omega }\), \({\overline{Z}}_\Omega =\overline{\Omega }\) and, for all \(i\in \{1,\ldots ,n\}\), the sequence of functions \({\delta _i^k}_{|_{Z_\Omega }}\), up to a subsequence, converges as \(k\rightarrow \infty \) to a function \(\delta _i:Z_\Omega \rightarrow {\mathbb {R}}^n\).
Taking into account Lemma 3.7, for all \(x\in Z_\Omega \) we have
Moreover, we infer that
because of (3.66), while
because of the conditions satisfied by \(T^L_k\) on the boundary of \(\Omega \).
Therefore, since
and \(\delta _i\left[ P_i\left( \frac{{\hat{z}}^k}{k}\right) \right] =e^i\) \(\forall k\in {\mathbb {N}}\), it follows that
On the other hand, since \({\hat{z}}_i^k\le k\), we obtain
We say that the last term tends to zero as \(k\rightarrow \infty \) that is, if the maximum in (3.80) is achieved for \(j=j_k\), we have
In fact, assume that (up to a subsequence) the sequence \(k\left[ \delta _i\left( \frac{{\hat{z}}^k-(j_k-1)e^i}{k}\right) -\delta _i\left( \frac{{\hat{z}}^k-j_ke^i}{k}\right) \right] \) converges as \(k\rightarrow \infty \) to a vector in \({\mathbb {R}}^n\), we denote by \(D_i{\hat{\delta _i}}\), and the sequences \(\delta _i\left( \frac{{\hat{z}}^k-(j_k-1)e^i}{k}\right) \) converge to some vectors \({\hat{\delta _i}}\) in \({\mathbb {R}}^n\). Then, we have that \({\hat{\delta }_i\cdot {\hat{\delta }_j}}=0\) for \(i\ne j\) and \({\hat{\delta }_i\cdot {\hat{\delta }_i}}=1\) for \(i,j\in \{1,\ldots ,n\}\) (because of Lemma 3.7). Therefore, in order to prove that \(D_i{\hat{\delta }_i}=0\), it suffices to prove that \(D_i{\hat{\delta }_i\cdot {\hat{\delta }_j}}=0\) \(\forall j\in \{1,\ldots ,n\}\).
First, notice that \(D_i{\hat{\delta }_i\cdot {\hat{\delta }_i}}=0\). In fact, we have
where the limit is equal to zero because
In order to prove that \(D_i{{\hat{\delta }_i}\cdot {\hat{\delta }_j}}=0\) for \(j\ne i\), notice that, since \(\delta _i\cdot \delta _j\equiv 0\) in \(Z_\Omega \), we have
Moreover, we have
and, as a consequence,
because \(\delta _i\cdot \delta _i\equiv 1\) in \(Z_\Omega \). Therefore, from (3.84) we obtain \(D_i{\hat{\delta }_i\cdot \hat{\delta }_j}=0\) also for \(i\ne j\), so \(D_i{\hat{\delta }_i}=0\).
Then, from (3.80) we infer that
in contradiction with (3.79).
Thus, we can conclude that (3.65) cannot hold and (3.79) is true. So the proof is complete. \(\square \)
Indeed, the minimality of \(T^L_k\) allows us to prove a stronger result, stated in the following corollary.
Corollary 3.9
Under the same assumptions of Lemma 3.8, for all \(i\in \{1,\ldots ,n\}\) we have
Proof
Since, under our assumptions on the values of \(T^L_k\) on \(\partial \Omega \), \(P_i\circ T^L_k\) is a one-to-one map between \(F^{\frac{h}{k}}_0\) and \(F^0_i\), (3.88) is equivalent to
Arguing by contradiction, assume that there exist sequences \((h_k)_k\), \((x^k)_k\), \((y^k)_k\) such that \(h_k\in \{0,1,\ldots ,k\}\), \(x^k\) and \(y^k\) belong to \(F_i^{\frac{h}{k}}\), \(x^k\ne y^k\) \(\forall k\in {\mathbb {N}}\) and (up to a subsequence)
Notice that the interfaces between the domains \(kT^L_k(C^k_z)\), \(z\in Z_k\), tend to be flat because of the minimality of the admissible deformation \(T^L_k\) and, as follows from Lemmas 3.7 and 3.8, up to translations these domains tend to \(\Omega \) that is, for every choice of \(z^k\) in \(Z_k\), the domain \(k\left[ T^L_k(C^k_z)-T^L_k\left( \frac{z^k}{k}\right) \right] \) tends to \(C^1_0=\Omega \).
Therefore, (3.90) is possible only if \(\lim _{k\rightarrow \infty }k|x^k-y^k|=\infty \) (otherwise, up to a subsequence, the segment \(\{x^k+t(y^k-x^k)\,\ t\in [0,1]\}\) meets only a finite number of subdomains \({\overline{C}}^k_z\) with \(z\in Z_k\)). In this case, if \(x^k\in C^k_{z^k}\) and \(y^k\in C^k_{\zeta ^k}\) for suitable \(z^k\) and \(\zeta ^k\) in \(Z_k\), we have
and
where
(as follows from Lemma 3.8) and
because the segments \(\left\{ x^k+t\left( \frac{z^k}{k}-x^k\right) \,\ t\in [0,1]\right\} \) and \(\left\{ y^k+t\left( \frac{\zeta ^k}{k}-y^k\right) \,\ t\in [0,1]\right\} \) are respectively enclosed in the subdomains \({\overline{C}}^k_{z^k}\) and \({\overline{C}}^k_{\zeta ^k}\).
Moreover, for k large enough, we have
and, analogously,
with
where
because of Lemma 3.1.
In a similar way we obtain
Therefore, it follows easily that (3.90) cannot be true.
Thus, we have a contradiction, so (3.88) holds and the proof is complete. \(\square \)
Lemma 3.10
Under the same assumptions as in Lemma 3.8, for all \(i\in \{1,\ldots ,n\}\) we have also
Proof
Arguing by contradiction, assume that for all \(k\in {\mathbb {N}}\) there exist \(h_k\in \{1,\ldots ,k\}\) and \(x^k\in F^0_i\) such that
for some \(i\in \{1,\ldots ,n\}\).
For all \(k\in {\mathbb {N}}\), let us choose \(y^k\in \Omega \) and \(z^k\in Z_k\) such that
Therefore, we have
Taking into account Lemmas 3.7, 3.8 and Corollary 3.9, the domain \(k\left[ T^k_L({\overline{C}}^k_{z^k})-T^L_k\left( \frac{z^k}{k}\right) \right] \) tends to \({\overline{C}}^1_0=\overline{\Omega }\) as \(k\rightarrow \infty \).
On the other hand, this convergence is not possible if (3.102) holds for some \(i\in \{1,\ldots ,n\}\).
Thus, we have a contradiction, (3.102) cannot hold for any \(i\in \{1,\ldots ,n\}\) and (3.101) is true. So the proof is complete. \(\square \)
Now, for all \(i\in \{1,\ldots ,n\}\), \(t\in [0,1]\) and \(k\in {\mathbb {N}}\), let us consider the function \({\tilde{f}}_i^{t,k,L}:F^0_i\rightarrow [0,1]\) defined by
Proposition 3.11
For all \(i\in \{1,\ldots ,n\}\), \(t\in [0,1]\) and \(k\in {\mathbb {N}}\), the functions \({\tilde{f}}_i^{t,k,L}\) defined by (3.105) have the following properties:
Proof
Taking into account the definition of \({\tilde{f}}_i^{t,k,L}\), properties (3.106) and (3.107) follow by direct computation respectively from Corollary 3.9 and Lemma 3.10. \(\square \)
Lemma 3.12
Let \({\tilde{f}}_i^{t,k,L}\) be the functions defined in (3.105). Then, for all \(x\in \overline{\Omega }\) there exists \(y\in \overline{\Omega }\) such that \({\tilde{f}}_i^{x_i,k,L}\circ P_i(y)=y_i\) \(\forall i\in \{1,\ldots ,n\}\) (that is, y belongs to the graph of \({\tilde{f}}_i^{x_i,k,L}\) for \(i=1,\ldots ,n\)).
Proof
From Proposition 2.3 and (3.105) we infer that \({\tilde{f}}_i^{t,k,L}\circ P_i(y)\) is strictly increasing with respect to t in the interval [0, 1] for all \(i\in \{1,\ldots ,n\}\), \(k\in {\mathbb {N}}\), \(y\in \overline{\Omega }\). Moreover, we have
so for all \(y\in \overline{\Omega }\) there exists a unique \(t_i(y)\in [0,1]\) such that \({\tilde{f}}_i^{t_i(y),k,L}\circ P_i(y)=y_i\). Let us set \(t(y)=(t_1(y),\ldots ,t_n(y))\). Then, the function t(y), defined for all \(y\in \overline{\Omega }\), is continuous in \(\overline{\Omega }\) and satisfies
Therefore, from [33] we infer that for all \(x\in \overline{\Omega }\) there exists at least one \(y\in \overline{\Omega }\) such that \(t(y)=x\), that is \(t_i(y)=x_i\) for all \(i\in \{1,\ldots ,n\}\).
Thus, since \(t_i(y)=x_i\) is equivalent to \({\tilde{f}}_i^{x_i,k,L}\circ P_i(y)=y_i\), the proof is complete. \(\square \)
Lemma 3.13
Let \({\tilde{f}}_i^{t,k,L}\) be the functions defined in (3.105). Then, there exists \({\tilde{k}}_1\in {\mathbb {N}}\) such that for all \(k\ge {\tilde{k}}_1\) the following property holds: for all \(x\in \overline{\Omega }\) there exists a unique \(y\in \overline{\Omega }\) such that
Proof
In Lemma 3.12 we proved that for all \(k\in {\mathbb {N}}\) and for all \(x\in \overline{\Omega }\) there exists at least one \(y\in \overline{\Omega }\) satisfying (3.110). Now, we have to prove that for k large enough such a y is unique.
For all \({{\mathcal {L}}}\ge 0\) let us set \(C_i({{\mathcal {L}}})=\{x\in {\mathbb {R}}^n\,\ |x_i|\le {{\mathcal {L}}}|P_i(x)|\}\) and notice that the graph of \({\tilde{f}}_i^{x_i,k,L}\) is enclosed in \(y+C_i(\textrm{Lip}({\tilde{f}}_i^{x_i,k,L}))\).
One can verify by direct computation that \(\cap _{1\le i\le n}C_i({{\mathcal {L}}}_i)=\{0\}\) when \({{\mathcal {L}}}_i<(n-1)^{-\frac{1}{2}}\) \(\forall i\in \{1,\ldots ,n\}\).
Therefore, if \(\textrm{Lip}({\tilde{f}}_i^{x_i,k,L})<(n-1)^{-\frac{1}{2}}\) \(\forall i\in \{1,\ldots ,n\}\), y is the unique point in \({\mathbb {R}}^n\) satisfying (3.110). On the other hand, taking into account (3.106) of Proposition 3.11, we infer that there exists \({\tilde{k}}_1\in {\mathbb {N}}\) such that
Thus, the assertion of Lemma 3.13 holds for such a \({\tilde{k}}_1\), so the proof is complete. \(\square \)
Definition 3.14
Taking into account Lemma 3.13, for all \(k\ge {\tilde{k}}_1\) we can define a function \({\widetilde{T}}_k^L:\overline{\Omega }\rightarrow \overline{\Omega }\) in the following way. For all \(x\in \overline{\Omega }\) we set \({\widetilde{T}}^L_k(x)=y\) where y is the unique point in \(\overline{\Omega }\) satisfying (3.110), given by Lemma 3.13
Taking into account the properties of the function \({\tilde{f}}_i^{t,k,L}\) defined by (3.105) one can verify by standard arguments that \({\widetilde{T}}_k^L\) is a one-to-one continuous function and that \(({\widetilde{T}}_k^L)^{-1}\) is continuous too. Moreover, for all \(i\in \{1,\ldots ,n\}\) and \(t\in [0,1]\), \({\widetilde{T}}^L_k(F^t_i)\) is the graph of the function \({\tilde{f}}_i^{t,k,L}\) (that is \(f_i^{t,{\widetilde{T}}^L_k}={\tilde{f}}_i^{t,k,L}\)), \({\widetilde{T}}^L_k(F^t_i)=F^t_i\) for \(t\in \{0,1\}\) and
Proposition 3.15
For all \(k\ge {\tilde{k}}_1\) and \(L\in ]0,1[\), let \({\widetilde{T}}_k^L\) be the function introduced in Definition 3.14. Then, there exists \({\tilde{k}}_2\in {\mathbb {N}}\) such that \({\widetilde{T}}^L_k\in D_L\) \(\forall k\ge {\tilde{k}}_2\). Moreover, there exists a sequence of positive numbers \((L_k)_k\) such that \(\lim _{k\rightarrow \infty } L_k=0\) and \({{\widetilde{T}}}_k^L\subseteq D_{L_k}\) \(\forall k\ge {\tilde{k}}_2\).
Proof
From Proposition 3.11 we infer that there exists a sequence of positive numbers \((\Lambda _k)_k\) such that \(\lim _{k\rightarrow \infty }\Lambda _k=0\) and
Since \(\Lambda _k\rightarrow 0\), we can choose \(k_n\in {\mathbb {N}}\) such that \(\Lambda _k<\frac{1}{n}\) \(\forall k\ge k_n\). Hence, taking into account that \({\tilde{f}}_i^{t,k,L}=f_i^{t,{{\widetilde{T}}}_k^L}\), from Lemma 3.2 we infer that, for all \(k\ge k_n\), \({{\widetilde{T}}}_k^L\in D_{L_k}\) where \(L_k=\frac{(n+1)\sqrt{n}\, \Lambda _k}{1-n\Lambda _k}\). Since \(L_k\rightarrow 0\) as \(k\rightarrow \infty \), we can choose \({\tilde{k}}_2\) such that \(L_k\le L\) \(\forall k\ge {\tilde{k}}_2\). So the proof is complete. \(\square \)
Now, we can show that the function \(u_k=u_k^{T^L_k}\) is a solution of problem (1.1) for k large enough and satisfies all the assertions of Proposition 2.2 (in a similar way one can argue for the function \(v_k\)).
Notice that \(u_k^{T_k^L}=u_k^{{{\widetilde{T}}}_k^L}\), where \({{\widetilde{T}}}_k^L\in D_L\) is the function introduced in Definition 3.14, because \(T^L_k(C_z^k)={{\widetilde{T}}}_k^L(C_z^L)\) \(\forall z\in Z_k\).
First, we prove that \(u_k^{{{\widetilde{T}}}_k^L}\) is a solution of the Dirichlet problem in every subdomain \({{\widetilde{T}}}_k^L(C_z^k)\) for all \(z\in Z_k\) and then we show that it satisfies a suitable stationarity condition which allows us to prove that, indeed, it is a solution of the Dirichlet problem (1.1) in the domain \(\Omega \).
Lemma 3.16
There exists \(k_1(L)\in {\mathbb {N}}\) such that, for all \(k\ge k_1(L)\) and \(z\in Z_k\), the function \(u_{k,z}^{{{\widetilde{T}}}_k^L}\) is a solution of the Dirichlet problem
Proof
For all \(k\in {\mathbb {N}}\) and \(z\in Z_k\), let us consider the function \(G_{k,z}^{{{\widetilde{T}}}_k^L}:{{\widetilde{T}}}_k^L(C_z^k)\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) such that \(G_{k,z}^{{{\widetilde{T}}}_k^L}(x,\cdot )\in {{\mathcal {C}}}^2({\mathbb {R}})\) \(\forall x\in {{\widetilde{T}}}_k^L(C_z^k)\) and
where \( {\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\) is the function given by Lemma 2.4 which, for k large enough, is a solution of problem (3.115) because it is a local minimum of the functional \(E_\psi \) in \(H^1_0({{\widetilde{T}}}_k^L(C_z^k))\).
Moreover, let us set \(g_{k,z}^{{{\widetilde{T}}}_k^L}(x,t)=\frac{\partial G_{k,z}^{{{\widetilde{T}}}_k^L}}{\partial t}(x,t).\) Then, let us consider the functional \(E_{k,z,{{\widetilde{T}}}_k^L}:H^1_0({{\widetilde{T}}}_k^L(C_z^k))\rightarrow {\mathbb {R}}\) defined by
Since \(p>1\), for k large enough one can verify that for all \(u\not \equiv {\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\) there exists \(t_u>0\) such that
if and only if \(\sigma (z)[u-{\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}(x)]\vee 0\not \equiv 0\); in this case such a \(t_u\) is unique; in the other case we have
and
When \(\sigma (z)[u-{\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}(x)]\vee 0\not \equiv 0\), one can verify by direct computation that \(E'_{k,z,{{\widetilde{T}}}_k^L}\big ({\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}+t(u-{\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}) \big )[u-{\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}]\) is positive for \(t\in ]0,t_u[\) and negative for \(t>t_u\), so
Moreover, one can verify that
Assume, for example, that \(\sigma (z)=1\) (in a similar way one can argue when \(\sigma (z)=-1\)). In this case, we have
where
because \({\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\) is a solution of problem (3.115).
Notice that
where, as follows from Lemma 2.4,
Moreover, we have
where, for all \(k\in {\mathbb {N}}\),
We say that \(\lim _{k\rightarrow \infty }\lambda _k=\infty \). In fact, otherwise, there exist suitable sequences \((z^k)_k\) in \({\mathbb {R}}^n\), and \((v_k)_k\) in \(H^1_0(\Omega )\) such that \(z^k\in Z_k\), \(v_k\equiv 0\) in \(\Omega {\setminus } C_{z^k}^k\), \(\int _\Omega |v_k|^{p+1}dx=1\) \(\forall k\in {\mathbb {N}}\) and (up to a subsequence) \(\lim _{k\rightarrow \infty }\int _\Omega |{\nabla }v_k|^2dx<\infty \).
As a consequence, since \(p<\frac{n+2}{n-2}\) when \(n\ge 3\), there exists \({\bar{v}}\in H^1_0(\Omega )\) such that (up to a subsequence) \(v_k\rightarrow {\bar{v}}\) as \(k\rightarrow \infty \) weakly in \(H^1_0(\Omega )\), in \(L^{p+1}(\Omega )\) and almost everywhere in \(\Omega \).
Taking into account that \(\lim _{k\rightarrow \infty }\textrm{meas}{{\widetilde{T}}}_k^L(C_z^k)=0\), the almost everywhere convegence implies that \({\bar{v}}\equiv 0\) in \(\Omega \), in contradiction with the convergence in \(L^{p+1}(\Omega )\) because \(\int _{{{\widetilde{T}}}_k^L(C_z^k) }|v_k|^{p+1}\) \(dx=1\) \(\forall k\in {\mathbb {N}}\). Thus, we can conclude that \(\lim _{k\rightarrow \infty }\lambda _k=\infty \).
It follows that, for k large enough,
As a consequence, if we set
we have \( u_{k,z}^{{{\widetilde{T}}}_k^L}\in \Gamma \) and
Therefore, there exists a Lagrange multiplier \(\mu \in {\mathbb {R}}\) such that
In particular, if we choose \(v= u_{k,z}^{{{\widetilde{T}}}_k^L}-{\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\), we obtain \(\mu =0\) because
while
Thus, \(u_{k,z}^{{{\widetilde{T}}}_k^L}\) is a weak solution of the Dirichlet problem
On the other hand, since \(u_{k,z}^{{{\widetilde{T}}}_k^L}\ge {\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\) in \({{\widetilde{T}}}_k^L(C^k_z)\), we have
So \( u_{k,z}^{{{\widetilde{T}}}_k^L}\) is a solution of problem (3.115) and the proof is complete. \(\square \)
Proposition 3.17
Under the assumptions of Proposition 2.2, there exists \({\bar{k}}\in {\mathbb {N}}\) such that the function \(u^{{{\widetilde{T}}}_k^L}_k=\sum _{z\in Z_k}u_{k,z}^{{{\widetilde{T}}}_k^L}\) is a solution of problem (1.1) for all \(k\ge {\bar{k}}\).
Proof
From Lemma 3.16 we infer that, for a suitable \(k_1(L)\in {\mathbb {N}}\), \(E_\psi '\left( u_{k}^{{{\widetilde{T}}}_k^L}\right) [v]=0\) for all \(v\in H^1_0({{\widetilde{T}}}_k^L(C_z^k))\), \(z\in Z_k\), \(k\ge k_1(L)\).
Now, we have to prove that \(E'_\psi (u_{k}^{{{\widetilde{T}}}_k^L})[v]=0\) \(\forall v\in H^1_0( \Omega )\). Taking into account Lemma 3.16, we obtain
where \(\nu _{k,z}\) denotes the outward normal on \(\partial {{\widetilde{T}}}_k^L(C_z^k)\). Thus, in order to obtain \(E'_\psi (u_{k}^{{{\widetilde{T}}}_k^L})[v]=0\), we prove that
for all \(z_1,z_2\in Z_k\) such that \(|z_1-z_2|=1\) (that is when \({{\widetilde{T}}}_k^L(C_{z_1}^k)\) and \({{\widetilde{T}}}_k^L(C_{z_2}^k)\) are adjacent subdomains of \(\Omega \)).
Notice that, since for all \(k\ge k_1(L)\) and \(z\in Z_k\) the function \(u_{k,z}^{{{\widetilde{T}}}_k^L}\) is a solution of problem (3.115), for all vector field \(\Phi \in {{\mathcal {C}}}^1_0(\overline{\Omega },{\mathbb {R}}^n)\) we obtain
Thus, it is easy to verify that in order to prove (3.139) it suffices to show that there exists \({\bar{k}}\) such that
From Proposition 3.15 we infer that there exists \({\bar{k}}\ge k_1(L)\) such that \({{\widetilde{T}}}_k^L\in D_{L/2}\) \(\forall k\ge {\bar{k}}\). Now, for all \(\tau \in {\mathbb {R}}\) and \(\Phi \in {{\mathcal {C}}}^1_0(\overline{\Omega },{\mathbb {R}}^n)\), let us consider the function \(T_{\tau ,\Phi }:\overline{\Omega }\rightarrow \overline{\Omega }\) defined by the Cauchy problem
One can verify by standard arguments that for all \(\Phi \in {{\mathcal {C}}}^1_0(\overline{\Omega },{\mathbb {R}}^n)\) there exists \({\bar{\tau }}_\Phi >0\) such that \(T_{\tau ,\Phi }\circ {{\widetilde{T}}}_k^L\in D_L\) \(\forall \tau \in [-{\bar{\tau }}_\Phi ,{\bar{\tau }}_\Phi ]\). It follows that
because of the minimality of \({{\widetilde{T}}}_k^L\). Moreover, notice that
so we have to prove that
Arguing by contradiction, assume that (3.145) does not hold. We can assume, for example, that
(otherwise we replace \(\Phi \) by \(-\Phi \)). Therefore, there exists a sequence of positive numbers \((\tau _i)_i\) such that \(\lim _{i\rightarrow \infty }\tau _i=0\) and
We say that, as a consequence of (3.146), for i large enough we have
In fact, arguing by contradiction, assume that (up to a subsequence still denoted by \((\tau _i)_i\)) the inequality (3.148) does not hold.
Then, for all \(i\in {\mathbb {N}}\) and \(z\in Z_k\), there exists \(t_{z,i}\ge 0\) such that
Since \(p>1\), the sequence \((t_{z,i})_i\) is bounded \(\forall z\in Z_k\). Moreover, taking into account that
we infer that \(\lim _{i\rightarrow \infty }t_{z,i}=1\) \(\forall z\in Z_k\) and
As a consequence, for all \(i\in {\mathbb {N}}\) there exists \(\tau '_i\in ]0,\tau _i[\) such that
which, as \(i\rightarrow \infty \), implies
in contradiction with (3.146). Thus, (3.148) holds. From Lemma 2.4 we infer that, if we choose \({\bar{k}}\) large enough, for all \(k\ge {\bar{k}}\), \(z\in Z_k\) and \(i\in {\mathbb {N}}\) there exists a unique minimizing function \({\tilde{u}}_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\). Moreover, \({\tilde{u}}_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\rightarrow {\tilde{u}}_{k,z}^{{{\widetilde{T}}}_k^L}\) in \(H^1_0(\Omega )\), as \(i\rightarrow \infty \), \(\forall k\ge {\bar{k}}\), \(\forall z\in Z_k\).
Then, using the functions \({\tilde{u}}_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\) and arguing as in the proof of Lemma 3.16, for i large enough we obtain the functions \( u_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\).
The construction of the functions \({\tilde{u}}_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\) and \( u_{k,z}^{T_{\tau _i,\Phi }\circ {{\widetilde{T}}}_k^L}\) shows also that
and
Therefore, from (3.148) and (3.155) we obtain
for i large enough, in contradiction with (3.143).
So we can conclude that \(\frac{d\ }{d\tau }E_\psi \left( u_{k}^{{{\widetilde{T}}}_k^L}\circ T^{-1}_{\tau _i,\Phi }\right) =0\), that is \(E'_\psi (u_{k}^{{{\widetilde{T}}}_k^L})[\Phi \cdot {\nabla }u_{k}^{{{\widetilde{T}}}_k^L}]=0\) for all vector field \(\Phi \in {{\mathcal {C}}}^1_0(\overline{\Omega },{\mathbb {R}}^n)\).
Thus, \(u_{k}^{{{\widetilde{T}}}_k^L}\) is a solution of problem (1.1) for all \(k\ge {\bar{k}}\). \(\square \)
Proof of Proposition 2.2 (conclusion)
If \(\Omega \) is the cube (2.1), all the assertions of Proposition 2.2 hold for k large enough if we set \(u_k=u_{k}^{{{\widetilde{T}}}_k^L}\) and \(T_{k,u}=T^L_k\) (or \(T_{k,u}={{\widetilde{T}}}_k^L\)) where the function \(u_k^{T_k^L}\) and the admissible deformation \(T^L_k\) are obtained by the minimizing method described in Sect. 2 (the functions \(v_k=v_k^{T_{k,v}}\) are obtained in a similar way: it suffices to replace \(\sigma (z)\) by \(\sigma (z)+1\)). Notice that we have \(u_k^{T^L_k}=u_k^{{{\widetilde{T}}}_k^L}\) (where \({{\widetilde{T}}}_k^L:\overline{\Omega }\rightarrow \overline{\Omega }\) is the function introduced in Definition 3.14) because \(T^L_k(C^k_z)={{\widetilde{T}}}^L_k(C^k_z)\) \(\forall z\in Z_k\).
In fact, Proposition 3.17 guarantees that there exists \({\bar{k}}\in {\mathbb {N}}\) such that \(u_k^{{{\widetilde{T}}}_k^L}\) is a solution of problem (1.1) for all \(k\ge {\bar{k}}\).
The asymptotic behaviour of \(u_k\) as \(k\rightarrow \infty \) is described by Proposition 3.5, Lemma 3.7 and Proposition 3.15. In fact, Proposition 3.5 shows that for every choice of \(z^k\) in \(Z_k\), up to a subsequence, the function \(U_{z^k}\), as \(k\rightarrow \infty \) converges in \(H^1(\chi )\) to a positive solution \(U_\chi \) of the Dirichlet problem \(-\Delta U=|U|^{p-1}U\) in \(\chi \), \(U=0\) on \(\partial \chi \), satisfying
where \(\chi \) is a bounded domain of \({\mathbb {R}}^n\). Lemma 3.7 says that the minimality of the admissible deformation \(T^L_k\) implies that \(\chi \) must be a cube of \({\mathbb {R}}^n\) having a vertex in the origin and the sides of length 1 and finally Proposition 3.15 guarantees that \({{\widetilde{T}}}_k^L\in D_{L_k}\) for a suitable sequence \((L_k)_k\) in ]0, 1[ such that \(\lim _{k\rightarrow \infty }L_k=0\) so, as a consequence, \(\chi =C_0^1=\Omega \) and \({{\widetilde{T}}}_k^L\) converges as \(k\rightarrow \infty \) to the identity function uniformly in \(\Omega \) (as pointed out in Remark 3.4).
Notice that \(T^L_k\in D_L\) \(\forall k\in {\mathbb {N}}\) but, unlike \({{\widetilde{T}}}_k^L\), we cannot say that \(T^L_k\in D_{L_k}\) \(\forall k\in {\mathbb {N}}\). However, we can say that also \(T^L_k\) converges to the identity function uniformly in \(\Omega \) because, taking into account the definition of \({{\widetilde{T}}}_k^L\), we have
so, as a consequence,
Therefore, all the assertions in Proposition 2.2 hold for \(T_{k,u}={{\widetilde{T}}}_k^L\) and also for \(T_{k,u}=T^L_k\), so the proof is complete. \(\square \)
Theorem 2.1 is a direct consequence of Proposition 2.2.
Notice that our method to construct solutions having this checked nodal structure does not require any technique of deformation from the symmetry and it works in case of more general nonlinearities, even when they are not perturbations of symmetric nonlinearities by lower order terms.
For example, it works when in problem (1.1) the term \(|u|^{p-1}u+\psi \) is replaced by \(c_+(u^+)^p-c_-(u^-)^p+\psi \) with \(c_+>0\), \(c_->0\) and \(c_+\ne c_-\).
Moreover, notice that our method works also when the nonlinear term has critical growth. For example, for \(n>2\) and \(\lambda \in {\mathbb {R}}\) let us consider the Dirichlet problem
whose solutions are critical points of the energy functional \({{\mathcal {F}}}:H^1_0(\Omega )\rightarrow {\mathbb {R}}\) defined by
It is well known that, if \(\lambda =0\) and \(\psi \equiv 0\) in \(\Omega \), problem (3.160) has only the trivial solution \(u\equiv 0\) for every bounded starshaped domain \(\Omega \), as a consequence of the Pohozaev identity (see [36]).
When \(n\ge 4\), \(\psi \equiv 0\) in \(\Omega \) and \(\lambda \) is positive and strictly less than the first eigenvalue of the Laplace operator \(-\Delta \) in \(H^1_0(\Omega )\), there exists a positive solution that concentrates as a Dirach mass as \(\lambda \rightarrow 0\) (see [12, 13] etc.); the existence of nodal solutions is studied for example in [16] etc..
Notice that also when \(\psi \equiv 0\) in \(\Omega \), so that the functional \({{\mathcal {F}}}\) is even, the problem of finding infinitely many solutions is difficult because the well known Palais-Smale compactness condition is not satisfied, as a consequence of the presence of the critical Sobolev exponent (see [12, 13, 44] etc.).
When \(\Omega \) is a cube of \({\mathbb {R}}^n\), our method, combined with some estimates as in [13], allows us to construct infinitely many solutions with many nodal regions and arbitrarily large energy level for all \(\lambda >0\) and \(\psi \in L^2(\Omega )\).
In fact, as we prove in a paper in preparation, the following theorem holds (see also [27] for the particular case where \(\Omega \) is a cube and \(\psi \equiv 0\) in \(\Omega \)).
Theorem 3.18
Assume that \(\Omega \) is a cube of \({\mathbb {R}}^n\) with \(n\ge 4\) and \(\lambda >0\). Then, for every \(\psi \in L^2(\Omega )\), problem (3.160) admits infinitely many solutions.
More precisely, if \(\Omega \) is for example the cube (2.1), for all \(\psi \in L^2(\Omega )\) there exists \({\bar{k}}\in {\mathbb {N}}\) such that, for every \(k\ge {\bar{k}}\), problem (3.160) admits a solution \(u_k\) having the following properties.
For all \(k\ge {\bar{k}}\) there exists \(T_k\in D_L\) such that, for every choice of \(z^k\) in \(Z_k\), the function \(u_{k,z^k}:={u_k}_{|_{T_k(C^k_{z^k})}}\) belongs to \(H^1_0\big (T_k(C^k_{z^k})\big )\) (here we consider \(u_{k,z^k}\) extended by the value zero in \({\mathbb {R}}^n\setminus \Omega \)).
Moreover, there exist \({\varepsilon }_k>0\) and \(m_k\in C^1_0=\Omega \) such that \({\varepsilon }_k\rightarrow 0\) as \(k\rightarrow \infty \) and the function \(U_k\) defined by
converges as \(k\rightarrow \infty \) to a function \({\overline{U}}\in {{\mathcal {D}}}({\mathbb {R}}^n)\) such that
The sequence \((T_k)_k\) converges to the identity map uniformly in \(\Omega \) while the domains \(k\Big [T_k(C_{z^k}^k)\) \(-T_k\left( \frac{z^k}{k}\right) \Big ]\) tend to the cube \(\Omega \) as \(k\rightarrow \infty \) for every choice of \(z^k\) in \(Z_k\).
Furthermore, for all \(k\ge {\bar{k}}\) there exists also another solution \(v_k\) of problem (3.160) such that the function \(-v_k\) presents an asymptotic behaviour as \(u_k\) when \(k\rightarrow \infty \).
Data availibility
Data sharing not applicable to this article as no datasets were generated or analysed during the current study.
References
Ambrosetti, A.: On the existence of multiple solutions for a class of nonlinear boundary value problems. Rend. Sem. Mat. Univ. Padova 49, 195–204 (1973)
Ambrosetti, A.: A perturbation theorem for superlinear boundary value problems. Tech. Sum. Report 1446, Wisconsin Univ Madison Mathematics Research Center (1974)
Ambrosetti, A., Rabinowitz, P.H.: Dual variational methods in critical point theory and applications. J. Funct. Anal. 14, 349–381 (1973)
Bahri, A.: Topological results on a certain class of functionals and application. J. Funct. Anal. 41(3), 397–427 (1981)
Bahri, A., Berestycki, H.: Points critiques de perturbations de fonctionnelles paires et applications. C. R. Acad. Sci. Paris Sér A–B 291(3), A189–A192 (1980)
Bahri, A., Berestycki, H.: A perturbation method in critical point theory and applications. Trans. Amer. Math. Soc. 267(1), 1–32 (1981)
Bahri, A., Berestycki, H.: Forced vibrations of superquadratic Hamiltonian systems. Acta Math. 152(3–4), 143–197 (1984)
Bahri, A., Lions, P.-L.: Morse index of some min-max critical points. I. Application to multiplicity results. Comm. Pure Appl. Math. 41(8), 1027–1037 (1988)
Bartsch, T., Clapp, M.: Critical point theory for indefinite functionals with symmetries. J. Funct. Anal. 138(1), 107–136 (1996)
Bolle, P.: On the Bolza problem. J. Diff. Equs. 152(2), 274–288 (1999)
Bolle, P., Ghoussoub, N., Tehrani, H.: The multiplicity of solutions in non-homogeneous boundary value problems. Manuscripta Math. 101(3), 325–350 (2000)
Brezis, H. (1985) Elliptic equations with limiting Sobolev exponents—the impact of topology. vol. 39, pp. S17–S39. 1986. Frontiers of the mathematical sciences, New York (1985)
Brézis, H., Nirenberg, L.: Positive solutions of nonlinear elliptic equations involving critical Sobolev exponents. Comm. Pure Appl. Math. 36(4), 437–477 (1983)
Byeon, J., Rabinowitz, P.H.: On a phase transition model. Calc. Var. Part. Differ. Equs. 47(1–2), 1–23 (2013)
Byeon, J., Rabinowitz, P.H.: Unbounded solutions for a periodic phase transition model. J. Differ. Equs. 260(2), 1126–1153 (2016)
Cerami, G., Solimini, S., Struwe, M.: Some existence results for superlinear elliptic boundary value problems involving critical exponents. J. Funct. Anal. 69(3), 289–306 (1986)
Chambers, C., Ghoussoub, N.: Deformation from symmetry and multiplicity of solutions in non-homogeneous problems. Discrete Contin. Dyn. Syst. 8(1), 267–281 (2002)
Clapp, M., Puppe, D.: Critical point theory with symmetries. J. Reine Angew. Math. 418, 1–29 (1991)
Coffman, C.V.: A minimum-maximum principle for a class of non-linear integral equations. J. Analyse Math. 22, 391–419 (1969)
Coffman, C.V.: On a class of non-linear elliptic boundary value problems. J. Math. Mech. 19, 351–356 (1969/1970)
Conti, M., Terracini, S., Verzini, G.: Nodal solutions to a class of nonstandard superlinear equations on \( {\mathbb{R} }^N\). Adv. Differ. Equs. 7(3), 297–318 (2002)
Conti, M., Terracini, S., Verzini, G.: An optimal partition problem related to nonlinear eigenvalues. J. Funct. Anal. 198(1), 160–196 (2003)
Conti, M., Terracini, S., Verzini, G.: On a class of optimal partition problems related to the Fučík spectrum and to the monotonicity formulae. Calc. Var. Part. Differ. Equs. 22(1), 45–72 (2005)
Ehrmann, H.: Nachweis periodischer Lösungen bei gewissen nichtlinearen Schwingungsdifferentialgleichungen. Arch. Ratl. Mech. Anal. 1, 124–138 (1957)
Ehrmann, H.: Über die Existenz der Lösungen von Randwertaufgaben bei gewöhnlichen nichtlinearen Differentialgleichungen zweiter Ordnung. Math. Ann. 134, 167–194 (1957)
Ekeland, I., Ghoussoub, N.: \(Z_2\)-equivariant Ljusternik–Schnirelman theory for non-even functionals. Ann. Inst. H. Poincaré C Anal. Non Linéaire 15(3), 341–370 (1998)
Fortunato, D., Jannelli, E.: Infinitely many solutions for some nonlinear elliptic problems in symmetrical domains. Proc. Roy. Soc. Edinburgh Sect. A 105, 205–213 (1987)
Fučík, S., Lovicar, V.: Periodic solutions of the equation \(x^{^{\prime \prime }}(t)+g(x(t))=p(t)\). Časopis Pěst. Mat. 100(2), 160–175 (1975)
Hempel, J.A.: Superlinear variational boundary value problems and nonuniqueness. Ph.d. thesis, University of New England, (1970)
Hempel, J.A.: Multiple solutions for a class of nonlinear boundary value problems. Indiana Univ. Math. J. 20, 983–996 (1970/71)
Krasnosel’skii, M.A., Topological methods in the theory of nonlinear integral equations. A Pergamon Press Book. The Macmillan Company, New York,: Translated by A. H. Armstrong; translation edited by J, Burlak (1964)
Marino, A., Prodi, G.: Metodi perturbativi nella teoria di Morse. Boll. Un. Mat. Ital. (4), 11(3, suppl.):1–32, (1975). Collection of articles dedicated to Giovanni Sansone on the occasion of his eighty-fifth birthday
Miranda, C.: Un’osservazione su un teorema di Brouwer. Boll. Un. Mat. Ital. 2(3), 5–7 (1940)
Molle, R., Passaseo, D.: Infinitely many solutions for elliptic equations with non-symmetric nonlinearities. Calc. Var. Part. Differ. Equs. 61(3), 115 (2022)
Nehari, Z.: Characteristic values associated with a class of non-linear second-order differential equations. Acta Math. 105, 141–175 (1961)
Pohožaev, S.I.: On the eigenfunctions of the equation \(\Delta u+\lambda f(u)=0\). Soviet Math. Dokl. 6, 1408–1411 (1965)
Rabinowitz, P.H.: Variational methods for nonlinear eigenvalue problems. In Eigenvalues of non-linear problems (Centro Internaz. Mat. Estivo (C.I.M.E.), III Ciclo, Varenna, 1974), pp. 139–195. (1974)
Rabinowitz, P.H.: Multiple critical points of perturbed symmetric functionals. Trans. Amer. Math. Soc. 272(2), 753–769 (1982)
Rabinowitz, P.H.: Minimax methods in critical point theory with applications to differential equations, volume 65 of CBMS Regional Conference Series in Mathematics. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, (1986)
Ramos, M., Tavares, H., Zou, W.: A Bahri–Lions theorem revisited. Adv. Math. 222(6), 2173–2195 (2009)
Struwe, M.: Infinitely many critical points for functionals which are not even and applications to superlinear boundary value problems. Manuscripta Math. 32(3–4), 335–364 (1980)
Struwe, M.: Infinitely many solutions of superlinear boundary value problems with rotational symmetry. Arch. Math. (Basel) 36(4), 360–369 (1981)
Struwe, M.: Superlinear elliptic boundary value problems with rotational symmetry. Arch. Math. (Basel) 39(3), 233–240 (1982)
Struwe, M.: A global compactness result for elliptic boundary value problems involving limiting nonlinearities. Math. Z. 187(4), 511–517 (1984)
Tanaka, K.: Morse indices at critical points related to the symmetric mountain pass theorem and applications. Comm. Part. Differ. Equs. 14(1), 99–128 (1989)
Terracini, S., Verzini, G.: Oscillating solutions to second-order ODEs with indefinite superlinear nonlinearities. Nonlinearity 13(5), 1501–1514 (2000)
Acknowledgements
The authors are very grateful to the reviewer of this paper for the deep comments and suggestions contained in his/her report.
Funding
Open access funding provided by Universitá degli Studi di Roma Tor Vergata within the CRUI-CARE Agreement. The authors have been supported by the “Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA)” of the Istituto Nazionale di Alta Matematica (INdAM). R.M. acknowledges also the MIUR Excellence Department Project awarded to the Department of Mathematics, University of Rome Tor Vergata, CUP E83C18000100006.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
We wish to confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome.
Additional information
Communicated by P. H. Rabinowitz.
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
Molle, R., Passaseo, D. On the Bahri–Lions conjecture for elliptic equations with non-symmetric nonlinearities. Calc. Var. 62, 177 (2023). https://doi.org/10.1007/s00526-023-02507-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00526-023-02507-5