Abstract
Let \(N\ge 3\), \(R>\rho >0\) and \(A_{\rho }:=\{x\in \mathbb {R}^N;\ \rho<|x|<R\}\). Let \(U^{\pm }_{n,\rho }\), \(n\ge 1\), be a radial solution with n nodal domains of
We show that if \(p=\frac{N+2+2\alpha }{N-2}\), \(\alpha >-2\) and \(N\ge 3\), then \(U^{\pm }_{n,\rho }\) is nondegenerate for small \(\rho >0\) and the Morse index \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) satisfies
where \(\ell =[\frac{\alpha }{2}]+1\). Using Jacobi elliptic functions, we show that if \((p,\alpha )=(3,N-4)\) and \(N\ge 3\), then the Morse index of a positive and negative solutions \({{\textsf{m}}}(U^{\pm }_{1,\rho })\) is completely determined by the ratio \(\rho /R\in (0,1)\). Upper and lower bounds for \({{\textsf{m}}}(U^{\pm }_{n,\rho })\), \(n\ge 1\), are also obtained when \((p,\alpha )=(3,N-4)\) and \(N\ge 3\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and main results
Let \(N\ge 3\), \(R>\rho >0\) and \(A_{\rho }:=\{x\in \mathbb {R}^N;\ \rho<|x|<R\}\). We are concerned with radial solutions of the Dirichlet problem
The equation in (1.1) is called the Hénon equation arising in astrophysics. Hénon [16] studied radial solutions of (1.1) on a ball. We are particularly interested in the case
The exponent \(p_c\) becomes the critical Sobolev exponent provided that \(\alpha =0\).
Let B denote a unit ball in \(\mathbb {R}^N\). First we consider the problem
When \(p\ge p_c\), (1.2) has no nontrivial radial solution due to Pohozaev’s identity. On the other hand, Ni [21] proved that (1.2) has a positive radial solution provided that \(1<p<p_c\). This result was generalized by Bartsch–Willem [7]. They proved that, for each \(n\ge 2\), (1.2) has radial solutions \(U^{\pm }_{n}(r)\) that have n nodal domains, i.e., the set \(\{x\in B;\ U^{\pm }_n(x)\ne 0\}\) has exactly n connected components. The uniqueness of \(U^+_n(r)\) follows from Ni–Nussbaum [22], and \(U^-_n(r)=-U^+_n(r)\). Therefore, \(p_c\) is a threshold between existence and nonexistence of nontrivial radial solutions of (1.2).
The situation, however, is different in annular domains. It is known that for each \(n\ge 1\), (1.1) has exactly two classical radial solutions \(U^{\pm }_{n,\rho }\) with n nodal domains, which satisfies \(U^-_{n,\rho }(r)=-U^+_{n,\rho }(r)\). We let \((U^+_{n,\rho })'(R)<0\). In particular, \(U^+_{1,\rho }\) and \(U^-_{1,\rho }\) are a positive and negative solution, respectively. Existence and uniqueness results hold for arbitrary \(p>1\) and \(\alpha \in \mathbb {R}\). See [22, Theorem 3.8] for existence of radial solutions \(U^{\pm }_{n,\rho }\) and see [22, Theorem 3.1] for the uniqueness of \(U^+_{n,\rho }\). Thus, it is interesting to study what happens in the case \(p=p_c\) as \(\rho \rightarrow 0\). Bandle-Peletier [6] obtained, among other things, an asymptotic expansion of \(\left\| U^{\pm }_{1,\rho }\right\| _{L^{\infty }(A_{\rho })}\) as \(\rho \rightarrow 0\) in the case \(\alpha =0\). Specifically, as \(\rho \rightarrow 0\),
In this paper we study the Morse index of \(U^{\pm }_{n,\rho }\) when \(\rho >0\) is small. We call the Morse index \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) of a solution \(U^{\pm }_{n,\rho }\) of (1.1) the maximal dimension of a subspace \(X\subset H_0^1(A_{\rho })\) where the quadratic form associated to the linearization operator at \(U^{\pm }_{n,\rho }\) is negative definite. In our problem (1.1) \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) is equal to the number of the negative eigenvalues of the problem
counted with their multiplicity, i.e.,
We call that \(U^{\pm }_{n,\rho }\) is nondegenerate if (1.3) does not have a zero eigenvalue.
We prepare some notation to state our main results. Let \([\xi ]\), \(\xi \in \mathbb {R}\), denote the largest integer that does not exceed \(\xi \). Let \(\{\nu _j\}_{j=0}^{\infty }\) denote the set of the eigenvalues of the Laplace-Beltrami operator \(-\Delta _{\mathbb {S}^{N-1}}\) on the sphere \(\mathbb {S}^{N-1}\), i.e.,
The multiplicity of the eigenvalue \(\nu _j\) is given by
The first main result is about the exact Morse index for small \(\rho >0\) and the divergence of the Morse index as \(\rho \rightarrow R\).
Theorem 1.1
Let \(N\ge 3\), \(\alpha >-2\) and \(p=p_c\). Let \(U^{\pm }_{n,\rho }\), \(n\ge 1\), denote radial solutions with n nodal domains of (1.1). Then the following hold:
-
(i)
Let \(\ell :=\left[ \frac{\alpha }{2}\right] +1\). Then,
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho }) {\left\{ \begin{array}{ll} \displaystyle = n\sum _{j=0}^{\ell }M_j(N) =n\frac{(N+2\ell -1)(N+\ell -2)!}{(N-1)!\ell !} &{} \text {for small}\ \rho >0,\\ \displaystyle \ge n\sum _{j=0}^{\ell }M_j(N) =n\frac{(N+2\ell -1)(N+\ell -2)!}{(N-1)!\ell !} &{} \text {for}\ \rho <R, \end{array}\right. } \end{aligned}$$and \(U^{\pm }_{n,\rho }\) is nondegenerate for small \(\rho >0\). In particular, if \(\alpha =0\), then
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho })=n(N+1)\quad \text {for small}\ \ \rho >0. \end{aligned}$$ -
(ii)
As \(\rho \rightarrow R\),
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho })\rightarrow +\infty . \end{aligned}$$ -
(iii)
Let \(R>\rho >0\) be fixed. As \(\alpha \rightarrow \infty \) (and hence \(p=p_c\rightarrow \infty \)),
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho })\rightarrow +\infty . \end{aligned}$$
The Morse index formula given in Theorem 1.1 (i) has the following simpler form
Explicit formulas for \(N=3,4,5\) are given in the following:
Example 1.2
Let \(\alpha >-2\), \(p=p_c\) and \(\ell =\left[ \frac{\alpha }{2}\right] +1\). Then the following hold:
-
(i)
If \(N=3\), then \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\frac{n}{2!}(2\ell +2)(\ell +1)\) for small \(\rho >0\).
-
(ii)
If \(N=4\), then \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\frac{n}{3!}(2\ell +3)(\ell +1)(\ell +2)\) for small \(\rho >0\).
-
(iii)
If \(N=5\), then \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\frac{n}{4!}(2\ell +4)(\ell +1)(\ell +2)(\ell +3)\) for small \(\rho >0\).
Remark 1.3
Let \(\rho =a\), \(R=a+1\), \(\lambda \in (-\infty ,\lambda _1)\) and \(p>1\). Here, \(\lambda _1>0\) denotes the first Dirichlet eigenvalue of \(-\Delta \) on \(A_{\rho }\). It was shown in [14, Proposition 3.6] that the Morse index of a positive radial solution of
diverges as \(a\rightarrow \infty \). The ratio \(\rho /R\) converges to 1 as \(a\rightarrow \infty \). This divergence result of the Morse index corresponds to Theorem 1.1 (ii). Similar phenomena were also observed in previous researches including [9, 17, 24, 25].
When \((p,\alpha )=(3,N-4)\), the Morse index of positive and negative solutions \(U^{\pm }_{1,\rho }\) can be completely determined by the ratio \(\rho /R\in (0,1)\). In particular, the smallness of \(\rho >0\) is not assumed in the following:
Theorem 1.4
Let \(N\ge 3\) and \((p,\alpha )=(3,N-4)\). Let \(U^+_{1,\rho }\) and \(U^-_{1,\rho }\) be a positive and negative radial solution of (1.1), respectively. Assume that one of the following (a) and (b) holds:
(a) \(\mathcal {R}_{\ell ,1}<\frac{\rho }{R}\le \mathcal {R}_{\ell +1,1}\) for a positive integer \(\ell >\frac{N}{2}-1\),
(b) \(0<\frac{\rho }{R}\le \mathcal {R}_{\ell +1,1}\) for a nonnegative integer \(\ell =\left[ \frac{N}{2}\right] -1\).
Then,
Here,
and K denotes the complete elliptic integral of the first kind whose definition and basic properties are recalled in Sect. 7.1.
Note that if \(\ell >(N-2)/2\), then \(8\nu _{\ell }-3(N-2)^2>0\) and \(0<\sqrt{\frac{4\nu _{\ell }}{8\nu _{\ell }-3(N-2)^2}}<1\). Then \(\mathcal {R}_{\ell ,1}\) and \(\mathcal {R}_{\ell +1,1}\) are well-defined. We can check that \(\mathcal {R}_{[\frac{N}{2}],1}\) is also well-defined. Since
the statements (a) and (b) in Theorem 1.4 cover the whole range (0, 1).
Upper and lower bounds of Morse indices \({{\textsf{m}}}(U^{\pm }_{n\,\rho })\), \(n\ge 1\), can be obtained as follows:
Theorem 1.5
Let \(N\ge 3\) and \((p,\alpha )=(3,N-4)\). Let \(U^{\pm }_{n,\rho }\), \(n\ge 1\), denote radial solutions of (1.1) with n nodal domains. Then, the following holds:
-
(i)
If \(\frac{\rho }{R}>\mathcal {R}_{\ell ,n}\) for a positive integer \(\ell >\frac{N}{2}-1\), then
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho })\ge n\sum _{j=0}^{\ell }M_j(N)=nM_{\ell }(N+1). \end{aligned}$$(1.7)Here, \(\mathcal {R}_{\ell ,n}\) is defined by (1.6).
-
(ii)
If \(\frac{\rho }{R}\le \tilde{\mathcal {R}}_{\ell ,n}\) for a nonnegative integer \(\ell >\frac{N}{2}-2\), then
$$\begin{aligned} {{\textsf{m}}}(U^{\pm }_{n,\rho })\le n\sum _{j=0}^{\ell }M_j(N)=nM_{\ell }(N+1). \end{aligned}$$(1.8)
Here,
Note that if \(\ell >\frac{N}{2}-2\) and \(\ell \ge 0\), then
and hence \(\tilde{\mathcal {R}}_{\ell ,n}\) is well-defined.
When \((p,\alpha )=(3,N-4)\), all the radial solutions of (1.1) can be written explicitly as follows:
Theorem 1.6
Let \(N\ge 3\) and \((p,\alpha )=(3,N-4)\). Let \(U^{\pm }_{n,\rho }\), \(n\ge 1\), denote radial solutions of (1.1) with n nodal domains. Then,
where \(\textrm{sd}(\xi ,k):={\textrm{sn}(\xi ,k)}/{\textrm{dn}(\xi ,k)}\) and \(k\in (\frac{1}{\sqrt{2}},1)\) is the unique solution of
In Theorem 1.6\(\textrm{sn}(\xi ,k)\) and \(\textrm{dn}(\xi ,k)\) denote Jacobi elliptic functions whose definitions and basic properties are summarized in Sect. 7.
Let us compare our theorems with previous results. Morse indices of radial solutions were studied by Amadori–Gladiali [1,2,3,4,5], by De Marchis–Ianni–Pacella [12, 13], by Gladiali–Grossi–Neves [15] and by Moreira dos Santos–Pacella [20]. As mentioned above, (1.2) does not have nontrivial radial solutions if \(p\ge p_c\). Therefore it is appropriate to study differences between (1.1) with \(p=p_c\) and (1.2) with \(p<p_c\). When \(p<p_c\), for each \(n\ge 1\), there exist exactly two radial solutions \(U^{\pm }_n\) of (1.2) with n nodal domains, which satisfy \(U^-_n(r)=-U^+_n(r)\). The following Morse index formula of \(U^{\pm }_n\) was obtained for \(\alpha =0\) in [12] and for \(\alpha >0\) in [2].
Proposition 1.7
Let \(N\ge 3\), \(\alpha \ge 0\) and \(\ell :=\left[ \frac{\alpha }{2}\right] +1\). If \(p(<p_c)\) is close to \(p_c\), then
When \(\alpha \ge 0\) is not an even integer, we see by Theorem 1.1 that \({{\textsf{m}}}(U^{\pm }_{n,\rho })={{\textsf{m}}}(U^{\pm }_{n})\) for small \(\rho >0\). On the other hand, when \(\alpha \ge 0\) is an even integer, we see that \({{\textsf{m}}}(U^{\pm }_{n,\rho })>{{\textsf{m}}}(U^{\pm }_{n})\) for small \(\rho >0\). Hence, we can say that the critical case \(p=p_c\) is more unstable than the subcritical case \(p<p_c\) when \(\alpha \) is an even integer.
In [3, Theorem 1.1] the following lower bounds of Morse indices for (1.1) and (1.2) were obtained: If \(\alpha \ge 0\), then
where \(\ell :=\left[ \frac{\alpha }{2}\right] +1\). A simple proof in the case (1.2) with \(n\ge 2\) can be found in [12]. By Theorem 1.1 we see that \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) does not attain the lower bound (1.11). On the other hand, if \(p(<p_c)\) is close to \(p_c\) and \(\alpha =0\), then we see by Proposition 1.7 that \({{\textsf{m}}}(U^{\pm }_n)\) attains the lower bound (1.11) when \(\ell =1\). We can say that the critical case does not have the most stable solution, while the subcritical case has.
Let us mention technical details. In the critical case by Emden’s transformation we can transform a radial part of (1.1) into the scalar field equation \(u''-u+|u|^{p-1}u=0\). As explained in Sect. 2, the Morse index \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) is equal to the number of the negative eigenvalues of the weighted eigenvalue problem
and all the eigenvalue of (1.12), which is denoted by \(\tilde{\lambda }_{i,j}\), can be written as follows:
Here, \(\tilde{\lambda }_{\textrm{rad},i}\), \(i=1,2,\ldots \), is the i-th radial eigenvalue of (1.12), i.e., the i-th eigenvalue of
Since \(\nu _j\) is explicitly given by (1.5), it becomes important to study all the negative eigenvalues \(\tilde{\lambda }_{\textrm{rad},i}\). In the critical case \(\tilde{\lambda }_{\textrm{rad},i}\), \(i=1,2,\ldots \), are given as the eigenvalues for the linearization of the scalar field equation as explained in Sect. 4. Let \(\ell :=[\frac{\alpha }{2}]+1\). The main part of our analysis is for showing that
and that,
In particular,
The limit (1.15) will be obtained in Remark 5.9. If \(\rho >0\) is small, then all the negative eigenvalues are
which leads to Theorem 1.1. Because of a variational characterization of \(\tilde{\lambda }_{\textrm{rad},n}\), each orthogonal set gives an upper bound of \(\tilde{\lambda }_{\textrm{rad},n}\). We can obtain a sharp upper bound in Remark 5.4. On the other hand, a lower bound of \(\tilde{\lambda }_{\textrm{rad},1}\) is nontrivial. In this paper we use a first Neumann eigenvalue of a linearization problem as a lower bound of \(\tilde{\lambda }_{\textrm{rad},1}\). Then we use a blow-up argument to show that
This limit implies (1.15), because of (1.14).
Let us again compare our problem with a problem on a ball. It was shown in [3, Proposition 3.3] that, for a certain class of nonlinear terms including \(|x|^{\alpha }|U|^{p-1}U\), radial eigenvalues of a wighted eigenvalue problem on a ball satisfy
It follows from (1.14) and (1.16) that the n-th radial eigenvalue is larger than that of our problem. This causes a difference of lower bounds of Morse indices. A lower bound obtained in [3, Theorem 1.1] is smaller than that in Theorem 1.1 (i).
When \((p,\alpha )=(3,N-4)\), \(\tilde{\lambda }_{\textrm{rad},n}\) can be explicitly written as
where k is the solution of (1.10). This explicit eigenvalue relates \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) to the ratio \(\rho /R\), and hence it plays a crucial role in the proof of Theorems 1.4 and 1.5 (i). There are various results about lower bounds of the Morse index, while few results are known for upper bounds. A rather explicit upper bound of the Morse index is obtained in Theorem 1.5 (ii). An upper bound of the Morse index is in general not easy to obtain, because a lower bound of \(\tilde{\lambda }_{\textrm{rad},1}\) is needed. In this paper we use the following explicit lower bound of \(\tilde{\lambda }_{\textrm{rad},1}\):
Recently, an exact expression of all the eigenvalues for the linearization of a Neumann problem \(u''-u+u^3=0\) is obtained in [19]. The same method is applicable to the Dirichlet problem. However, we do not use those exact expression in this paper.
In summary, thanks to the critical exponent \(p_c\), we can perform these detailed analysis of eigenvalues \(\tilde{\lambda }_{\textrm{rad},1},\dots ,\tilde{\lambda }_{\textrm{rad},n+1}\).
The paper consists of seven sections. In Sect. 2 we recall fundamental results about eigenvalues of (1.3) and (1.12). In Sect. 3 we use Emden’s transformation and transform (1.1) into the scalar filed equation on an interval. Then, we prove Theorem 1.6. In Sect. 4 we compare the weighted eigenvalue problem (1.12) with (4.1) which is an eigenvalue problem associated with the scalar field equation. In Sect. 5 we compute the Morse index of \(U^{\pm }_{n,\rho }\) and prove Theorem 1.1. In Sect. 6 we consider the case \((p,\alpha )=(3,N-4)\) and prove Theorems 1.4 and 1.5. Section 7 is an appendix. We recall the definition and basic properties of the complete elliptic integral K(k) and Jacobi elliptic functions \(\textrm{sn}(\xi ,k)\), \(\textrm{cn}(\xi ,k)\), \(\textrm{dn}(\xi ,k)\) and \(\textrm{sd}(\xi ,k)\).
2 Preliminaries
Let \(U^{\pm }_{n,\rho }\) be solutions of (1.1) with n nodal domains. In this paper we mainly study eigenvalues of the wighted eigenvalue problem (1.12). We define the number of the negative eigenvalues of (1.12) counted with their multiplicity by
The following proposition plays a crucial role in the study of the Morse index for radial solutions. It was extensively used in previous researches including [1,2,3,4,5, 12, 13, 15].
Proposition 2.1
Let \({{\textsf{m}}}(U^{\pm }_{n,\rho })\) and \(\widetilde{{\textsf{m}}}(U^{\pm }_{n,\rho })\) defined by (1.4) and (2.1), respectively. Then the following holds:
The proof of Proposition 2.1 is the same as [13, Lemma 4.2 (a)], which proves the case \(\alpha =0\). See also [4, Proposition 1.1]. We omit the proof.
Let \(\tilde{L}_n^{\pm }:=|x|^2\left( \Delta +p|x|^{\alpha }|U^{\pm }_{n,\rho }|^{p-1}\right) \) defined on \(H_0^1(A_{\rho })\) and \(\tilde{L}_{\textrm{rad},n}^{\pm }:=|x|^2\big (\Delta +p|x|^{\alpha }|U^{\pm }_{n,\rho }|^{p-1}\big )\) defined on \(H_{0,\textrm{rad}}^1(A_{\rho })\). The eigenvalue problem \(\tilde{L}^{\pm }_{\textrm{rad},n}\tilde{\Phi }_{\textrm{rad}}=-{\tilde{\lambda }_{\textrm{rad}}}\tilde{\Phi }_{\textrm{rad}}\) can be also written as (1.13). From now on, \(\sigma (\tilde{L}^{\pm }_n)\) and \(\sigma (\tilde{L}^{\pm }_{\textrm{rad},n})\) denote the set of the eigenvalues of (1.12) and (1.13), respectively. Let \(\sigma (-\Delta _{\mathbb {S}^{N-1}})\) denote the set of eigenvalues of \(-\Delta _{\mathbb {S}^{N-1}}\), i.e., \(\sigma (-\Delta _{\mathbb {S}^{N-1}})=\{\nu _j\}_{j=0}^{\infty }\), where \(\nu _j\) is defined by (1.5). Let \(\tilde{\lambda }_{\textrm{rad},i}\), \(i=1,2,\ldots \), denote the i-th eigenvalue of (1.13).
Because of Proposition 2.1, we count the number of the negative eigenvalues of (1.12) instead of (1.3). The eigenvalue problem (1.12) is easier to study, since all the eigenvalues of (1.12) can be decomposed into a radial and spherical parts.
Proposition 2.2
The eigenvalues of (1.12) satisfy the following:
Specifically, each eigenvalue of (1.12), which is denoted by \(\tilde{\lambda }_{i,j}\), can be written as
where \(\tilde{\lambda }_{\textrm{rad},i}\in \sigma (\tilde{L}^{\pm }_{\textrm{rad}})\) and \(\nu _j\in \sigma (-\Delta _{\mathbb {S}^{N-1}})\).
The proof of Proposition 2.2 is the same as [8, Lemma 3.1], which studies (1.2). The relation (2.2) was also extensively used in [1,2,3,4,5, 8, 12,13,14,15, 17, 18].
The multiplicity of an eigenvalue of (1.12) can be calculated by the following proposition:
Proposition 2.3
Let \(\tilde{\lambda }\in \sigma (\tilde{L}^{\pm }_n)\) be fixed. Let \(m(\tilde{\lambda })\) denote the multiplicity of \(\tilde{\lambda }\). Then,
where the summation takes all pairs (i, j) satisfying
Moreover, the eigenspace of (1.12) associated to \(\tilde{\lambda }\) is spanned by
where \(\tilde{\Phi }_{\textrm{rad},i}\) denotes the i-th eigenfunction of (1.13), \(\omega _j(\theta )\) denotes an eigenfunction of \(-\Delta _{\mathbb {S}^{N-1}}\) associated to \(\nu _j\) and the pair (i, j) satisfies (2.4).
We call the eigenvalue \(\tilde{\lambda }_{\textrm{rad},i}+\nu _0\) of (1.12) a radial eigenvalue and \(\tilde{\lambda }_{\textrm{rad},i}+\nu _j\), \(j\ge 1\), a nonradial eigenvalue.
Since \(\nu _j\) is explicitly given by (1.5), it is important to study the negative eigenvalues of (1.13). As mentioned in Sect. 1, we show that
3 Exact solutions
Let \(N\ge 3\), \(\alpha >-2\) and \(p=p_c\). The problem
has an exact positive radial singular solution
where
We use the so-called Emden transformation
where
Then, we see in the following lemma that u(t) is a solution of the problem
where
Lemma 3.1
Let \(N\ge 3\), \(\alpha >-2\) and \(p=p_c\). The radial function \(U(r)\in C^2((\rho ,R))\cap C([\rho ,R])\) is a solution of (1.1) if and only if u(t) is a solution of (3.6).
Proof
By direct calculation we have
Then,
We see by (3.3) that \(2\beta -N+2=0\). By (3.8) we obtain
Since \(\rho<r<R\), we have that \(0<t<t_{\rho }\) and that u(t) satisfies the Dirichlet boundary condition. Then, u(t) satisfies (3.6). It is clear that the converse is true. The proof is complete. \(\square \)
It is well known that a solution of (3.6) corresponds to an orbit of the system
Since a Dirichlet boundary condition is imposed in (3.6), a corresponding orbit (u(t), v(t)) starts from a point on the v-axis and arrives a point on the v-axis. The system (3.9) has three equilibria \((-1,0)\), (0, 0) and (1, 0). Then, \((\pm 1,0)\) are centers and (0, 0) is a saddle. Multiplying the equation in (3.6) by \(u'\) and integrating it over [0, x], we see that each solution (u(t), v(t)) is on a level set
We see that (3.9) has two homoclinic loops connecting (0, 0) to itself which are on \(v^2-u^2+2|u|^{p+1}/(p+1)=0\). One loop surrounds (1, 0) and the other loop surround \((-1,0)\). Hence two loops consist of a figure eight. Let
Two loops satisfy \(v^2=u^2-2|u|^{p+1}/(p+1)\), i.e., the boundary of \(\Omega \). It is obvious that there is no orbit in \(\Omega \) satisfying the boundary condition of (3.6). Therefore, a solution orbit of (3.6) is in \(\mathbb {R}^2\setminus \Omega \), and they are periodic orbits. If (u(t), v(t)) satisfies (3.9), then \((-u(t),-v(t))\) also satisfies (3.9). This indicates that all times from a point on the v-axis to the next point of the v-axis are equal, and hence the length of each nodal domain of u(t) is equal to each other. Hence, if u has n nodal domains, then the length of each nodal domain is \(t_{\rho }/n\).
When \((p,\alpha )=(3,N-4)\), the radial solutions of (1.1) can be written explicitly in terms of Jacobi elliptic functions
Proof of Theorem 1.6
Because of Lemma 3.1, it is enough to obtain an exact solution of (3.6) with n nodal domains. Since u satisfies \(u''-u+u^3=0\) and u satisfies the Dirichlet boundary condition, a general solution u(t) can be written in terms of elliptic functions as follows:
See [10, Chapter 7, Section 10] for details of this formula and see Sect. 7 for the definition of \(\textrm{cn}(\xi ,k)\). Since \(u(0)=u(t_{\rho })=0\) and u has n nodal domains, we see
where K denotes the complete elliptic integral of the first kind whose definition and basic properties are recalled in Sect. 7.1. Then,
By the addition formula
we have that \(\textrm{cn}(x\mp K(k),k)=\pm \sqrt{1-k^2}\textrm{sd}(x,k)\). We obtain
We return to the original variables. Then we obtain (1.9). By the first equality in (3.11) we obtain (1.10). The function K(k) is increasing in \(k\in (0,1)\). See Sect. 7. It is clear that \(\sqrt{2k^2-1}K(k)\) is strictly increasing in \(k\in (\frac{1}{\sqrt{2}},1)\),
Thus, (1.10) has a unique solution \(k\in \left( \frac{1}{\sqrt{2}},1\right) \). \(\square \)
4 Eigenvalue problem
Let \(n>1\) and let u(t) be a solution of (3.6) with n nodal domains. We consider the linearized eigenvalue problem
Here \(f'(u)=-1+p|u|^{p-1}\). Let \(\mu _i\), \(i\ge 1\), denote the i-th eigenvalue and let \(\phi _i\) denote an eigenfunction associated with \(\mu _i\).
We use the same change of variables as (3.4), i.e., let \(t:=-\frac{1}{m}\log \frac{r}{R}\) and we define
where \(U^*(r)\) is the singular solution of (3.1) given by (3.2) and \(\phi \) is an eigenfunction of (4.1). Then U(r) is a solution of (1.1) and \(\tilde{\Phi }(r)\) satisfies
For \(i\ge 1\), let
Then, the pair \((\tilde{\lambda }_{\textrm{rad},i},\tilde{\Phi })\) satisfies (1.12), and hence the radial part of the eigenvalue problem (1.12), which is (1.13), is equivalent to (4.1). All the eigenvalues of (1.12) can be obtained by (2.3). Since all the eigenvalues of \(-\Delta _{\mathbb {S}^{N-1}}\) are explicitly given by (1.5), it is crucial to study eigenvalues of (4.1).
5 Morse index
5.1 Fundamental results for the scalar field equation
Hereafter, we use a time map. A reader can consult [23, Chapters 1 and 2] for details about relations of a time map and a solution structure of two point boundary value problems.
Let
First, we consider a positive solution of
We use the change of variables
where m and \(U^*(r)\) are defined in (3.5) and (3.2), respectively. Then v(t) satisfies
where \(T_0:=\frac{1}{m}\log \sqrt{\frac{R}{\rho }}\). Note that \(T_0\rightarrow 0\) as \(\rho \rightarrow R\) and \(T_0\rightarrow \infty \) as \(\rho \rightarrow 0\). A solution of (5.1) corresponds to a solution (5.2). A solution of (5.2) corresponds to an orbit of (3.9) in the right half-plane that starts from a point \((0,v'(-T_0))\) and arrives \((0,v'(T_0))\). This orbit goes across the horizontal axis at (a, 0), and a is the maximum value of v(t) for \(-T_0<t<T_0\). The orbit is in \(\mathbb {R}^2\setminus \Omega \), where \(\Omega \) is defined by (3.10). Hence \(a>a_0\). We study the time of this orbit. Multiplying the equation in (5.2) with \(v'\) and integrating it over \([T_0,t]\), we have
Integrating \(1=v'(t)/\sqrt{2(F(a)-F(v(t))}\) over \([0,T_0]\), we have
Hence, \(T_0\) can be related to a which is \(\max _{-T_0<t<T_0}v(t)\). It was shown in [22] that, for each pair \((\rho ,R)\), \(0<\rho <R\), (5.1) has a unique solution. Therefore, (5.2) also has a unique solution for each \(T_0>0\). Since \(T_0(a)\) corresponds to a solution of (5.2) with \(T_0=T_0(a)\), the uniqueness of a solution of (5.2) indicates that \(T_0(a)\) is monotone. Since \(a\rightarrow a_0\), the corresponding orbit converges to a solution corresponding to a homoclinic loop in \(C_{loc}(\mathbb {R})\), and hence \(T(a)\rightarrow \infty \) as \(a\rightarrow a_0\). This limit, together with the existence of a solution of (5.2) for all \(T_0>0\), indicates that \(T_0(a)\) is decreasing, and \(T_0(a)\rightarrow 0\) as \(a\rightarrow \infty \). In summary,
Hence there exists an inverse function \(a=a(T_0)\) for \(0<T_0<\infty \).
We consider the following limit problem of (5.2):
Then w can be explicitly written as
In particular, \(w(0)=a_0\). We easily see that \(\lim _{a\rightarrow a_0}T_0=\infty \) and
Moreover,
Lemma 5.1
Let v be a solution of (5.2). Then there exists \(C>0\) such that, for \(a\in (a_0,2a_0)\),
Proof
Since v(t) is even, it is enough to prove (5.5) for \(0\le t\le T_0\). We define
Then \(\tilde{v}(t)\rightarrow w(t)\) in \(C_{loc}([0,\infty ))\) as \(a\rightarrow a_0\), \(\tilde{v}(t)\le a\) for \(t\ge 0\) and \(\tilde{v}(t)\) is nonincreasing in t. There exists \(t_0>0\) independent of \(a\in (a_0,2a_0)\) such that \(0\le \tilde{v}(t)\le 2^{-1/(p-1)}\) for \(t\ge t_0\). Let \(\bar{v}(t):=2a_0\exp (-(t-t_0)/\sqrt{2})\). Then,
Note that \(a_0=\tilde{v}(t_0)\le \bar{v}(t_0)=2a_0\) and \(0=\tilde{v}(T_0)\le \bar{v}(T_0)\). We see that \(\bar{v}(t)-\tilde{v}(t)\) does not have a negative minimum in \(t_0<t<T_0\). Thus,
It is clear that
By (5.7) and (5.6) we see that \(0\le \tilde{v}(t)\le \bar{v}(t)\) for \(t\ge 0\) and that \(\bar{v}(t)\) is independent of \(a\in (a_0,2a_0)\). The proof is complete. \(\square \)
A linearization problem of (5.3) in \(L^2(\mathbb {R})\) becomes
The spectra of (5.8) is known as follows:
Proposition 5.2
Let \(L:=\frac{d^2}{dt^2}+f'(w)\). The problem (5.8) has a continuous spectrum \([1,\infty )\). Moreover, the first eigenvalue of (5.8) is \(-(p-1)(p+3)/4\), the second eigenvalue is 0 and the third eigenvalue is \((p-1)(5-p)/4\) if \(1<p<3\). Specifically, the following hold:
See e.g. [11, p.9] for Proposition 5.2.
5.2 Proof of Theorem 1.1
Let \(T_0>0\) and \(u_1(t)\) be a positive solution of
Let \(\mu ^D\) denote the first eigenvalue of the Dirichlet problem
Lemma 5.3
Let \(\mu ^D\) be the first Dirichlet eigenvalue of (5.10). Then the following hold:
(i) For \(T_0>0\),
(ii) As \(T_0\rightarrow 0\),
Proof
(i) Let \(u_1\) be a positive solution of (5.9). Let
Multiplying the equation of (5.9) by \(u'\) and integrating it over \([T_0,x]\), we have
Since \(b=\sqrt{2F(a)}\), we have
Let \(I_1:=(0,2T_0)\) and
By the variational characterization we see that
We take a test function \(\psi :=u_1^{\frac{p+1}{2}}\). Using (5.13) and \(u_1''=u_1-u_1^p\), we have
By (5.13) we have
Thus, by (5.15) and (5.14) we have
Note that the largeness of \(T_0>0\) is not necessary in (5.16). We have shown that (5.11) holds.
(ii) In this case we take a test function \(u_1\). We show that
If (5.17) holds, then by (5.14) we have
and hence (5.12) holds.
Hereafter, we prove (5.17). We need an apriori estimate to prove (5.17). Using Hölder’s inequality, we have
Using this inequality, we have
Multiplying the equation in (5.9) with \(u_1\) and integrating it by parts, we have
Then, by (5.20) we have
Therefore,
By Hölder’s inequality we have
and hence
By (5.23), (5.22) and (5.19) we have
We have shown that (5.17) holds. The proof of (ii) is complete. \(\square \)
Remark 5.4
Let v(t) be a solution of (5.2) and let \(a:=v(0)\). We can obtain a sharp upper bound of \(\mu ^D\) in the following way: We can prove that
using the dominated convergence theorem with Lemma 5.1. Here,
\(\Gamma \) denotes the Gamma function defined by \(\Gamma (z):=\int _0^{\infty }t^{z-1}e^{-z}dt\) and B denotes the Beta function, which satisfies \(B(\xi ,\eta ):={\Gamma (\xi )\Gamma (\eta )}/{\Gamma (\xi +\eta )}\). Using the same calculation as in the proof of [6, Lemma 3.2], we can obtain
By (5.25), (5.24) and (5.15) we have
However, we do not use (5.26) in this paper.
Let \(u_1\) be a positive solution of (5.9). Let \(\mu ^N\) denote the first eigenvalue of the Neumann problem
Note that \(u_1(0)=u_1(2T_0)=0\), while \(\phi '(0)=\phi '(2T_0)=0\).
Lemma 5.5
Let \(\mu ^N\) be the first Neumann eigenvalue of (5.27). Then,
Proof
Let v(t) be a solution of (5.2). Then the first eigenvalue of (5.27) is equal to that of
Hereafter, we consider (5.29).
Let \(I_{T_0}:=(-T_0,T_0)\). Let \(\mu ^N\) be the first eigenvalue of (5.29) and let \(\phi \) be a first eigenfunction of (5.29). We may assume that \(\phi >0\) in \(\mathbb {R}\) and \(\left\| \phi \right\| _{L^2(I_{T_0})}=1\).
We see that
and that \(\phi \) attains the infimum of (5.30). Hereafter, we define \(\phi =0\) on \(\mathbb {R}{\setminus } I_{T_0}\) and \(\chi _{I_{T_0}}\) denotes the indicator function of \(I_{T_0}\). We also define \(v(t)=0\) on \(\mathbb {R}\setminus I_{T_0}\) to extend the domain of v(t). We see that \(1-p\left\| v\right\| _{\infty }^{p-1}\le 1-p|v|^{p-1}\) on \(I_{T_0}\). By Lemma 5.3 (i) we have
where we used
Since \(\left\| v\right\| _{\infty }\) is bounded for large \(T_0>0\), by (5.31) we see that \(\mu ^N\) is bounded for large \(T_0>0\). Hence, \(1+\mu ^N+p\left\| v\right\| _{\infty }^{p-1}\) is also bounded for large \(T_0>0\). Since
uniformly for large \(T_0>0\), we see that, for each compact set K, there are \(C_{K_1}>0\) and a compact set \(K_1\) such that K is in the interior set of \(K_1\) and that \(\left\| \phi \right\| _{H^1(K_1)}<C_{K_1}\) for large \(T_0>0\). Since \(H^1(K_1)\hookrightarrow C^{\gamma }(K_1)\), \(0<\gamma <1/2\), is continuous, \(\{\phi \}\) is bounded in \(C^{\gamma }(K_1)\). Note that \(\left\| \phi \right\| _{L^{\infty }(\mathbb {R})}\) is bounded uniformly for large \(T_0>0\) because \(\left\| \phi \right\| _{H^1(\mathbb {R})}\) is bounded uniformly for large \(T_0>0\). Since \(\phi \) satisfies the equation in (5.29), by Schauder estimates we see that \(\{\phi \}\) is bounded in \(C^{2,\gamma }(K)\). It follows from Ascoli-Arzelá theorem with a diagonal argument that there exists \(\phi _*\in C^2(\mathbb {R})\) such that \(\phi \rightarrow \phi _*\) in \(C^2_{loc}(\mathbb {R})\) as \(T_0\rightarrow \infty \). Moreover, \(|\phi _*(t)|\le C_2\) for \(t\in \mathbb {R}\), because \(H^1(K_1)\hookrightarrow L^{\infty }(K_1)\) is a continuous inclusion and \(\left\| \phi \right\| _{H^1(I_{T_0})}\) is bounded uniformly for large \(T_0>0\). We show that
where w is a unique solution of (5.3). Since \(\left\| \phi \right\| _{\infty }\) is bounded uniformly for large \(T_0>0\), by Lemma 5.1 we see that \(|v|^{p-1}\phi ^2\) is dominated by an \(L^1(\mathbb {R})\)-function which is independent of \(T_0>0\) large. Note that \(T_0\rightarrow \infty \) if and only if \(a\rightarrow a_0\). The function \(|v|^{p-1}\phi ^2\) converges pointwise to \(w^{p-1}\phi _*^2\) in \(\mathbb {R}\) as \(T_0\rightarrow \infty \). By the dominated convergence theorem we obtain (5.34). Using (5.34), Fatou’s lemma and Lemma 5.3 (i), we have
where w is a solution of (5.3) and we used (5.32). Because of (5.35), \(\phi _*\not \equiv 0\) in \(\mathbb {R}\). Since \(\{\mu ^N\}\) is bounded, there exists \(\mu ^N_*\) and a subsequence of \(\{\mu ^N\}\), which is still denoted by \(\{\mu ^N\}\), such that \(\mu ^N\rightarrow \mu ^N_*\) as \(T_0\rightarrow \infty \). Applying Fatou’s lemma to the LHS of (5.33), we see that \(\phi _*'\in L^2(\mathbb {R})\). Since \(\left\| \phi \right\| _2=1\), again by Fatou’s lemma we see that \(\phi _*\in L^2(\mathbb {R})\), and hence \(\phi _*\in H^1(\mathbb {R})\). Since \(\phi \) satisfies (5.29), \(\phi _*\) satisfies the problem
Since \(\phi _*\not \equiv 0\) in \(\mathbb {R}\), by the strong maximum principle \(\phi _*>0\) in \(\mathbb {R}\). Thus, \(\phi _*\) is a first eigenfunction. By Proposition 5.2 we see that \(\mu _*^N=-(p-1)(p+3)/4\). This indicates that (5.28) holds. \(\square \)
The following elementary inequality will be used later.
Lemma 5.6
Let \(\xi _j\ge 0\), \(j=0,1,\ldots ,n\), and \(\eta _j>0\), \(j=0,1,\ldots ,n\). If
then
Proof
Without loss of generality we assume that
Since \(\frac{\xi _j}{\eta _j}\ge \frac{\xi _1}{\eta _1}\) for \(j=1,2,\cdots ,n\), we see that \(\xi _j\eta _1\ge \xi _1\eta _j\). Then
and hence
\(\square \)
Let \(n\ge 1\). We define
Then (3.6) can be written as follows:
Hereafter, \(u^{\pm }_n\) denotes a solution of (5.37) with n nodal domains such that \((u^+_n)'(0)>0\) and \((u^-_n)'(0)<0\). It is easy to see that \(u^-_n(t)=-u^+_n(t)\). We do not distinguish \(u^+_n\) and \(u^-_n\). Note that \(u^{\pm }_n(t)\) can be extended to a \(2T_0\)-antiperiodic and \(4T_0\)-periodic function and that all the nodal domains of \(u^{\pm }_n(t)\) are
The linearization problem of (5.37) becomes
Let \(\{\mu _i\}_{i=1}^{\infty }\) denote the set of the eigenvalues of (5.38) associated with \(u^{\pm }_n\). Since \(2nT_0=\frac{1}{m}\log \frac{R}{\rho }\), we see that
Lemma 5.7
Let \(n\ge 1\). Then the following hold:
(i) For \(\rho <R\),
(ii) For each small \(\varepsilon >0\), there is \(\rho _{\varepsilon }>0\) such that, for \(0<\rho <\rho _{\varepsilon }\),
Proof
(i) First we show that \(\mu _n<-(p-1)(p+3)/4\). Recall that \(T=2nT_0\). Let \(I:=(0,T)\) and let \(I_j:=(2(j-1)T_0,2jT_0)\), \(j=1,2,\ldots ,n\). We use the variational characterization of \(\mu _n\)
where
Let \(\phi ^D(t)\) be the first eigenfunction of (5.10) defined on \(I_1\). For \(j=1,2,\ldots ,n\), let
Then, \(\textrm{supp}\phi _j=\bar{I_j}\). Since \(\phi _j(t)\) is continuous on I, we can check that the weak derivative of \(\phi _j(t)\) is \(\phi '_j(t)\) for almost everywhere in I. The function \(\phi '_j(t)\) is bounded on \(I_j\), and it is identically equal to 0 on \(I\setminus \bar{I_j}\). Then, the weak derivative of \(\phi _j(t)\) is in \(L^2(I)\), and hence \(\phi _j\in H^1(I)\). Since \(\phi _j(0)=\phi _j(2nT_0)=0\), we see that \(\phi _j(t)\in H^1_0(I)\). Moreover, \(\phi _j\) satisfies
We take a set of functions \(\phi _1,\ldots ,\phi _n\) which are orthogonal in \(L^2(I)\), and define
where \((c_1,\ldots ,c_n)\ne (0,\ldots ,0)\). Then, \(\psi \in \textrm{span}(\phi _1,\ldots ,\phi _n)\). We see that \(\psi \in H_0^1(I)\), since \(\phi _j\), \(1\le j\le n\), is in \(H_0^1(I)\). Since \(\textrm{supp}\phi _j\cap \textrm{supp}\phi _k\), \(j\ne k\), has zero measure, by (5.42) we have
By (5.44) and Lemma 5.3 (i) we have
The proof of (i) is complete. Note that the smallness \(\rho >0\) is not used.
(ii) Let \(\phi ^N(t)\) be the first eigenfunction of (5.27) defined on \(I_1\) and \(\tilde{\phi }^N(t)\) be the first eigenfunction of (5.38) defined on I. Let \(\xi _0:=H_1(\phi ^N)\), \(\eta _0:=\Vert \phi ^N\Vert _{L^2(I_1)}^2\),
where, for \(j=1,2,\ldots ,n\),
We restrict a domain of the function \(\tilde{\phi }^N\) to \(I_j\) for \(j=1,2,\ldots \). Then, \(\left. \tilde{\phi }^N\right| _{I_j}\) does not necessarily satisfy a Dirichlet boundary condition on \(\partial I_j\), though \(\tilde{\phi }^N\) satisfies a Dirichlet boundary condition on \(\partial I.\) However, as seen below, a space of test functions is \(H^1(I_j)\). Hence, we can use \(\left. \tilde{\phi }^N\right| _{I_j}\) as a test function. By the unique continuation theorem for linear elliptic PDEs that neither \(\phi ^N\) nor \(\tilde{\phi }^N\) vanishes on an open set, and hence \(\eta _j>0\) for \(j=0,1,2,\ldots ,n\). Since \(\phi ^N\) is the first eigenfunction of (5.27), we see that
By Lemma 5.6 we see that
Since \(\tilde{\phi }^N\) is a first eigenfunction of (5.38), by (5.45) we see that
By Lemma 5.5 we see that for each \(\varepsilon >0\), there exists \(\rho _\varepsilon >0\) such that
for \(0<\rho <\rho _\varepsilon \). The proof is complete. \(\square \)
Let \(\{\tilde{\lambda }_{\textrm{rad},i}\}_{i=1}^{\infty }\) denote the set of the eigenvalues of (1.13) associated with the solution \(u^{\pm }_n\).
Corollary 5.8
Let \(\ell :=\left[ \frac{\alpha }{2}\right] +1\). Then the following holds:
(i) For \(\rho <R\),
(ii) There exists \(\rho _\varepsilon >0\) such that, for \(0<\rho <\rho _\varepsilon \),
Proof
(i) We use (4.2), i.e., \(\tilde{\lambda }_{\textrm{rad},i}=\left( \frac{N-2}{2}\right) ^2\mu _i\). Multiplying (5.39) by \(\left( \frac{N-2}{2}\right) ^2\), we have
Since \(2\ell -2\le \alpha <2\ell \), by elementary calculation we have
By (5.47) and (5.46) we see that (i) holds. Note that the smallness of \(\rho >0\) is not used.
(ii) Since \(2\ell -2\le \alpha <2\ell \), by elementary calculation we have
By (5.48) we see that if \(\varepsilon >0\) is small, then
Multiplying (5.40) by \(\left( \frac{N-2}{2}\right) ^2\), we have
when \(\rho >0\) is small. Here, \(\varepsilon >0\) in (5.50) is the same value as \(\varepsilon >0\) in (5.49), since by Lemma 5.7 (ii) we can take an arbitrarily small \(\varepsilon >0\) in (5.50). Thus, by (5.49) and (5.50) we see that the conclusion of (ii) holds. \(\square \)
Remark 5.9
By (5.50) and (5.46) we see that, for each \(1\le i\le n\),
Lemma 5.10
For \(\rho <R\), \(\mu _{n+1}>0\) and hence
Proof
Since \(\tilde{\lambda }_{\textrm{rad},n+1}=\left( \frac{N-2}{2}\right) ^2\mu _{n+1}\), it is enough to show that \(\mu _{n+1}>0\). We prove the lemma by contradiction. Suppose that \(\mu _{n+1}\le 0\). Then, the associated eigenfunction \(\phi _{n+1}\) has \(n+2\) zeros on [0, T]. Let \(\phi \) be a solution of the initial value problem
By Sturm’s comparison principle \(\phi \) oscillates more rapidly than \(\phi _{n+1}\) or \(\phi =c\phi _{n+1}\) for some \(c\ne 0\). Hence, \(\phi \) has at least \(n+2\) zeros on [0, T]. Let \(0=z_0<z_1<z_2<\cdots <z_{n+1}\) denote the first \(n+2\) zeros of \(\phi \). Since \(\psi :=(u^{\pm }_n)'(t)\) satisfies the same equation \(\psi ''+(-1+p|u^{\pm }_n|^{p-1})\psi =0\), by Sturm separation theorem we see that \(\psi (t)\) has one zero in \((z_i,z_{i+1})\) for \(i=0,1,2,\ldots ,n\). The function \(\psi (t)\) has exactly n zeros in (0, T), while there are at least \(n+1\) intervals \(\{(z_i,z_{i+1})\}_{i=0}^n\). We obtain a contradiction. Thus, \(\mu _{n+1}\le 0\) does not occur, and hence \(\mu _{n+1}>0\). \(\square \)
Proof of Theorem 1.1
(i) First, we consider the case where \(\rho >0\) is small. When \(\rho >0\) is small, by Corollary 5.8 and Lemma 5.10 we have
Therefore, if \(\rho >0\) is small, then (1.12) has no zero eigenvalue, and hence (1.3) has also no zero eigenvalue. Thus, \(U^{\pm }_{n,\rho }\) is nondegenerate for small \(\rho >0\).
By (5.51) we see that all negative eigenvalues are
Moreover, it follows from Proposition 2.3 that a multiplicity of each eigenvalue is \(M_j(N)\). Thus,
By Proposition 2.1 we see that \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\widetilde{{{\textsf{m}}}}(U^{\pm }_{n,\rho })=n\sum _{j=0}^{\ell }M_j(N)\) for small \(\rho >0\).
We calculate \(\sum _{j=0}^{\ell }M_j(N)\). We use the following form of the multiplicity formula of \(\nu _j\):
where
We assume that \(\ell \) is even. Then
We obtain the same formula in the odd case.
Next, we consider the case where \(\rho >0\) is not necessarily small. Even in this case, by Corollary 5.8 (i) we have
Thus, by Proposition 2.1 we have
The proof of (i) is complete.
(ii) Since \(2nT_0=\frac{1}{m}\log \frac{R}{\rho }\), it follows from Lemma 5.3 (ii) that \(\mu ^D\rightarrow -\infty \) as \(\rho \rightarrow R\). By (5.44) and (4.2) we see that
Therefore, it is clear that, for each large integer \(\ell >0\), there exists \(\rho _{\ell }<R\) such that if \(\rho _{\ell }<\rho <R\), then \(\tilde{\lambda }_{\textrm{rad},n}+\nu _j<0\) for \(0\le j\le \ell \). The integer \(\ell \) can be arbitrary large if \(\rho \) is close to R. This indicates that \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\widetilde{{{\textsf{m}}}}(U^{\pm }_{n,\rho })\rightarrow \infty \) as \(\rho \rightarrow R\).
(iii) It follows from (5.46) that
Note that the smallness of \(\rho >0\) is not used in (5.46). By the same argument as in (ii) we see that \({{\textsf{m}}}(U^{\pm }_{n,\rho })=\widetilde{{{\textsf{m}}}}(U^{\pm }_{n,\rho })\rightarrow \infty \) as \(\alpha \rightarrow \infty \). \(\square \)
6 The case \((p,\alpha )=(3,N-4)\)
6.1 Proof of Theorem 1.4
We consider the case \((p,\alpha )=(3,N-4)\). In this section u denotes a solution of (3.6) with n nodal domains for simplicity. Let \(a:=\max _{0\le t\le t_{\rho }}|u(t)|\). Then, \(a>a_0=\sqrt{2}\) and
for \(0\le t\le t_{\rho }\).
The following lemma says that the n-th eigenvalue of (4.1) with respect to a solution u with n nodal domains can be written explicitly.
Lemma 6.1
Let \(p=3\) and let \(a>\sqrt{2}\). Then,
are the n-th eigenvalue of (4.1) and an associated eigenfunction, respectively. In particular,
where \(t_{\rho }\) is defined by (3.7).
Proof
Substituting \(\mu _n\) and \(\phi _n\) into \(\phi ''+(-1+3u^2+\mu )\phi \), we have
Using \(u''=u-u^3\) and (6.1), we can check that the RHS of (6.2) is equal to 0. Since \(u(t)^2-2+a^2>0\) for \(0\le t\le t_{\rho }\) and u(t) has n nodal domains, we see that \(\phi _n(t)\) has \(n-1\) zeros on \((0,t_{\rho })\). It follows from Sturm-Liouville theory that \(\phi _n\) is an n-th eigenfunction, and hence \(\mu _n\) is the n-th eigenvalue. \(\square \)
Proof of Theorem 1.4
We consider the case \(n=1\). Using Lemma 6.1 with \(n=1\) and Lemma 5.10 with \(n=1\), we see that
By (4.2) we see that \(\tilde{\lambda }_{\textrm{rad},i}=\left( \frac{N-2}{2}\right) ^2\mu _i\). Therefore, \(\tilde{\lambda }_{\textrm{rad},1}\) is the only negative radial eigenvalue of (1.12). It follows from Propositions 2.1–2.3 that
First we consider the case (a). Let \(\ell \) be given in (a). Since u(t) can be written explicitly as (3.12), we see that \(a=\sqrt{2k^2/(2k^2-1)}\), and hence
Since \(\ell >\frac{N}{2}-1\), we see that \(8\nu _{\ell }-3(N-2)^2>0\) and \(8\nu _{\ell +1}-3(N-2)^2>0\), and hence \(\mathcal {R}_{\ell ,1}\) and \(\mathcal {R}_{\ell +1,1}\) are well-defined. Since \(\mathcal {R}_{\ell ,1}<\frac{\rho }{R}\le \mathcal {R}_{\ell +1,1}\) for a positive integer \(\ell >\frac{N}{2}-1\), we have
By direct calculation we can check that (6.4) is equivalent to
Thus, by (6.3) we see that the conclusion holds.
Second we consider the case (b). Let \(\ell \) be given in (b). Since \(\ell =\left[ \frac{N}{2}\right] -1\), we see that \(\ell +1>(N-2)/2\) and \(\ell \le (N-2)/2\), and hence \(8\nu _{\ell +1}-3(N-2)^2>0\). Then, \(0=\mathcal {R}_{\ell ,1}<\mathcal {R}_{\ell +1,1}\). Since
by the same argument as in (a) we can check that (6.6) is equivalent to (6.5). Thus, by (6.3) we see that the conclusion holds. \(\square \)
6.2 Upper and lower bounds of the Morse index
Proof of Theorem 1.5
(i) It follows from Proposition 2.2 that if \(\tilde{\lambda }_{\textrm{rad},n}+\nu _{\ell }<0\), then the following are negative eigenvalues of (1.12):
By Propositions 2.1–2.3 we see that
Let \(\ell >0\) be given in Theorem 1.5 (i). In a similar way to the proof of Theorem 1.4 we see that \(\tilde{\lambda }_{\textrm{rad},n}+\nu _{\ell }<0\) is equivalent to \(\frac{\rho }{R}>\mathcal {R}_{\ell ,n}\). Thus, we have shown that (1.7) holds. \(\square \)
Proof of Theorem 1.5 (ii)
Let u(t) be a solution of (3.6) with n nodal domains. Let \(a:=\Vert u\Vert _{\infty }\). Then u(t) can be written explicitly as (3.12) and
We define
By direct calculation we see
Let \(\mu _1\) be the first eigenvalue of (4.1) and let \(\phi _1\) be a positive eigenfunction associated to \(\mu _1\). We define \(\varphi :=\psi -c\phi _1\). Here \(c>0\) can be taken such that the following holds: There exists \(t_0\in (0,t_{\rho })\) such that \(\varphi (t_0)=0\) and \(\varphi (t)\ge 0\) for \(0\le t\le t_{\rho }\), since \(\varphi (0)=\varphi (t_{\rho })>0\).
We prove by contradiction that
Suppose the contrary, i.e.,
Since \(\phi _1(t)>0\) for \(0<t<t_{\rho }\), we see that
for \(0<t<t_{\rho }\). Since \(t_0\) is a minimum point of \(\varphi \), we have \(\varphi ''(t_0)\ge 0\), and hence
We obtain a contradiction. Thus, (6.8) does not occur, and \(\underline{\mu }\le \mu _1\).
We define
Multiplying (6.7) by \(\left( \frac{N-2}{2}\right) ^2\), by (4.2) we have that \(\underline{\tilde{\lambda }}_{\textrm{rad},1}\le \tilde{\lambda }_{\textrm{rad},1}\). Hereafter, let \(\ell \ge 1\). If \(\tilde{\lambda }_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\), then the following are nonnegative eigenvalues of (1.12):
Hence the following eigenvalues can be negative:
Therefore, by Propositions 2.1 and 2.3 we see that
provided that \(\tilde{\lambda }_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\). If \(\underline{\tilde{\lambda }}_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\), then \(\tilde{\lambda }_{\textrm{rad},1}+\nu _{\ell +1}\ge \underline{\tilde{\lambda }}_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\), and hence (6.9) holds. Hence \(\underline{\tilde{\lambda }}_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\) is a sufficient condition for (6.9). Let \(\ell \) be given in Theorem 1.5 (ii). By direct calculation we see that \(\underline{\tilde{\lambda }}_{\textrm{rad},1}+\nu _{\ell +1}\ge 0\) is equivalent to \(\frac{\rho }{R}\le \tilde{\mathcal {R}}_{\ell ,n}\). Thus, (1.8) holds. \(\square \)
Data Availability
The author confirms that the data supporting the findings of this study are available within the article.
References
Amadori, A., Gladiali, F.: Bifurcation and symmetry breaking for the Henon equation. Adv. Differ. Equ. 19, 755–782 (2014)
Amadori, A., Gladiali, F.: Asymptotic profile and Morse index of nodal radial solutions to the Hénon problem. Calc. Var. Partial Differ. Equ. 58, 168, 47 (2019)
Amadori, A., Gladiali, F.: On a singular eigenvalue problem and its applications in computing the Morse index of solutions to semilinear PDE’s: II. Nonlinearity 33, 2541–2561 (2020)
Amadori, A., Gladiali, F.: On a singular eigenvalue problem and its applications in computing the Morse index of solutions to semilinear PDE’s. Nonlinear Anal. Real World Appl. 55, 103133, 13 (2020)
Amadori, A., Gladiali, F.: The Hénon problem with large exponent in the disc. J. Differ. Equ. 268, 5892–5944 (2020)
Bandle, C., Peletier, L.: Nonlinear elliptic problems with critical exponent in shrinking annuli. Math. Ann. 280, 1–19 (1988)
Bartsch, T., Willem, M.: Infinitely many radial solutions of a semilinear elliptic problem on \(\mathbb{R} ^N\). Arch. Ration. Mech. Anal. 124, 261–276 (1993)
Bartsch, T., Clapp, M., Grossi, M., Pacella, F.: Asymptotically radial solutions in expanding annular domains. Math. Ann. 352, 485–515 (2012)
Ben Ayed, M., Mehdi, K.. El., Ould Ahmedou, M., Pacella, F.: Energy and Morse index of solutions of Yamabe type problems on thin annuli. J. Eur. Math. Soc. 7, 283–304 (2005)
Davis, H.: Introduction to Nonlinear Differential and Integral Equations. Dover, New York (1962)
de Groen, P., Karadzhov, G.: Metastability in the shadow system for Gierer–Meinhardt’s equations. Electron. J. Differ. Equ. 2002(50), 1–22 (2002)
De Marchis, F., Ianni, I., Pacella, F.: A Morse index formula for radial solutions of Lane–Emden problems. Adv. Math. 322, 682–737 (2017)
De Marchis, F., Ianni, I., Pacella, F.: Exact Morse index computation for nodal radial solutions of Lane–Emden problems. Math. Ann. 367, 185–227 (2017)
Gladiali, F., Grossi, M., Pacella, F., Srikanth, P.: Bifurcation and symmetry breaking for a class of semilinear elliptic equations in an annulus. Calc. Var. Partial Differ. Equ. 40, 295–317 (2011)
Gladiali, F., Grossi, M., Neves, S.: Symmetry breaking and Morse index of solutions of nonlinear elliptic problems in the plane. Commun. Contemp. Math. 18, 1550087, 31 (2016)
Hénon, M.: Numerical experiments on the stability oh spherical stellar systems. Astron. Astrophys. 24, 229–238 (1973)
Lin, S.: Existence of positive nonradial solutions for nonlinear elliptic equations in annular domains. Trans. Am. Math. Soc. 332, 775–791 (1992)
Lin, S.: Asymptotic behavior of positive solutions to semilinear elliptic equations on expanding annuli. J. Differ. Equ. 120, 255–288 (1995)
Miyamoto, Y., Takemura, H., Wakasa, T.: Asymptotic formulas of the eigenvalues for the linearization of the scalar field equation (submitted)
Moreira dos Santos, E., Pacella, F.: Morse index of radial nodal solutions of Hénon type equations in dimension two. Commun. Contemp. Math. 19, 1650042, 16 (2017)
Ni, W.: A nonlinear Dirichlet problem on the unit ball and its applications. Indiana Univ. Math. J. 31, 801–807 (1982)
Ni, W., Nussbaum, R.: Uniqueness and nonuniqueness for positive radial solutions of \(\Delta u+f(u, r)=0\). Commun. Pure Appl. Math. 38, 67–108 (1985)
Schaaf, R.: Global Solution Branches of Two-Point Boundary Value Problems. Lecture Notes in Mathematics, 1458, p. XX + 141. Springer, Berlin (1990)
Shioji, N., Watanabe, K.: Uniqueness of positive radial solutions of the Brezis–Nirenberg problem on thin annular domains on \(\mathbb{S} ^n\) and symmetry breaking bifurcations. Commun. Pure Appl. Anal. 19, 4727–4770 (2020)
Srikanth, P.: Symmetry breaking for a class of semilinear elliptic problems. Ann. Inst. H. Poincaré Anal. Non Linéaire 7, 107–112 (1990)
Acknowledgements
The author would like to thank the referee for helpful comments and for bringing [3, Proposition 3.3] to author’s attention.
Funding
Open access funding provided by The University of Tokyo.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by JSPS KAKENHI Grant numbers 19H01797, 19H05599.
Appendix: Complete elliptic integral and Jacobi elliptic functions
Appendix: Complete elliptic integral and Jacobi elliptic functions
1.1 Complete elliptic integral
Let \(0\le k<1\). The complete elliptic integral of the first kind is denoted by
One can easily see that K(k) is monotonically increasing in \(k\in [0,1)\),
1.2 Jacobi elliptic functions
Let \(0<k<1\). The Jacobi elliptic function \(\textrm{sn}(\xi ,k)\) is an odd, periodic and analytic function with period 4K(k) as a function for the real domain, and is defined locally by
for \(\xi \in [0,K(k)]\). The function \(\textrm{cn}(\xi ,k)\) is an even and 4K(k)-periodic function defined locally by
for \(\xi \in [0,K(k)]\) and \(\textrm{dn}(\xi ,k)\) is an even and 2K(k)-periodic function defined by
In particular,
for \(\xi \in \mathbb {R}\) and \(k\in (0,1)\). The function \(\textrm{sd}(\xi ,k)\) is defined by
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
Miyamoto, Y. Exact Morse index of radial solutions for semilinear elliptic equations with critical exponent on annuli. Math. Z. 304, 65 (2023). https://doi.org/10.1007/s00209-023-03328-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00209-023-03328-0