arXiv reader

The best constant in the G-N inequality for the mixed local and Nonlocal Laplacian

Hichem Hajaiej and Yu Su

arXiv
2604.05177v1 · PDF · 6 April 2026
Categories
math.AP
Licence
CC BY 4.0 The site shows the text.

Abstract

In this paper, we establish the best constant in the G-N inequality for the mixed local and nonlocal Laplacian. In our problem, classical methods cannot apply directly since regularity results for the operator under study seem to be highly challenging. We build an innovative method that not only enabled us to prove that the optimizer of the best constant is the ground state solution of the equation, but also to establish the best constant.

Items 52

Theorem

Theorem 1.1

5PWU

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Then we have the following results.

  1. (1) For all u\in E, we have

    \begin{aligned} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x \leqslant C_{N,p,s} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{\frac{2N-p(N-2)}{2(1-s)}} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{\frac{p(N-2s)-2N}{2(1-s)}}, \end{aligned}

    where C_{N,p,s}>0 (independent of u) is the best constant.

  2. (2) There exists a radial non-negative optimizer u\in E for inequality (?).

  3. (3) Set Q=\lambda_{1}u(\lambda_{2}x), where

    \begin{aligned} \begin{cases} \lambda_{1}= \left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \right]^{\frac{1}{(1-s)(p-2)}} \left[ \frac{2p(1-s)}{p(N-2s)-2N} \frac{1}{\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}} \right]^{\frac{1}{p-2}},\\ \lambda_{2}= \left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \right]^{\frac{1}{2-2s}}. \end{cases} \end{aligned}

    Then Q is a radial non-negative weak solution of equation (?), and it is also an optimizer for inequality (?). Moreover, we have

    \begin{aligned} \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} =& \frac{p(N-2s)-2N}{2N-p(N-2)} \|Q\|_{D^{s,2}(\mathbb{R}^{N})}^{2}\\ =& \frac{p(N-2s)-2N}{2p(1-s)} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{p}, \end{aligned}
    \begin{aligned} \frac{N-2}{2} \|Q\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|Q\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =\frac{N}{p}\|Q\|_{L^{p}(\mathbb{R}^{N})}^{p}, \end{aligned}

    and

    \begin{aligned} C_{N,p,s}^{-1}= \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{2}}. \end{aligned}
Used by Proof
Theorem

Theorem 1.2

AGQW

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Then equation (?) has a radial ground state solution \Phi\in E, which satisfies

\begin{aligned} I(\Phi)=\inf_{u\in \mathcal{N}}I(u)=c, \end{aligned}

where I is the energy functional of equation (?) defined as follows

\begin{aligned} I(u)= \frac{1}{2} \left( \int_{\mathbb{R}^{N}} |\nabla u|^{2} \mathrm{d}x + \int_{\mathbb{R}^{N}}\int_{\mathbb{R}^{N}}\frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x\mathrm{d}y\right) - \frac{1}{p} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x, \end{aligned}

and \mathcal{N}:=\{u\in E\backslash\{0\}|\langle I'(u),u\rangle=0\} is the Nehari manifold.

Used by Remark 1.1
Proof (Proof of Theorem 1.2) of Theorem 1.2

From Lemma 4.2, we know that there exists a bounded (PS)_{c} sequence \{u_{n}\}\subset \mathcal{N} with c>0. If \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}}|u_{n}|^{p}\mathrm{d}x=0, then

\begin{aligned} c=I(u_{n}) =\frac{1}{2}\|u_{n}\|_{E}^{2}, \end{aligned}

and

\begin{aligned} 0=\langle I'(u_{n}),u_{n}\rangle =\|u_{n}\|_{E}^{2}, \end{aligned}

which gives

\begin{aligned} c=0. \end{aligned}

Therefore, we get \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}}|u_{n}|^{p}\mathrm{d}x>0.

By using Lemma 2.5, there exists y_{n}\subset \mathbb{R}^{N} such that \bar{u}_{n}:=u_{n}(x+y_{n})\rightharpoonup u\not \equiv0 in L^{p}_{loc}(\mathbb{R}^{N}) and

\begin{aligned} c =I(\bar{u}_{n}),~~\mathrm{and}~~0 =\langle I'(\bar{u}_{n}),\varphi\rangle =\langle I'(u),\varphi\rangle. \end{aligned}

Now, by virtue of the Brézis-Lieb Lemma [Brezis-Lieb1983PAMS], one deduces that

\begin{aligned} \bar{c} \leqslant I(u) =& I(u)-\frac{1}{p}\langle I'(u),u\rangle\\ =& \left(\frac{1}{2}-\frac{1}{p}\right) \|u\|_{E}^{2}\\ \leqslant& \lim\limits_{n\to\infty} \left(\frac{1}{2}-\frac{1}{p}\right) \|u_{n}\|_{E}^{2}\\ =& \lim\limits_{n\to\infty} I(u_{n}) -\frac{1}{p} \lim\limits_{n\to\infty} \langle I'(u_{n}),u_{n}\rangle\\ =&c =\bar{c}. \end{aligned}

That is, I(u)=\bar{c}. Therefore, equation (?) has a ground state solution.

Theorem

Theorem 1.3

49JP

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Then equation (?) has a solution \Psi\in E, which satisfies

\begin{aligned} P(\Psi)=\frac{N-2}{2} \|\Psi\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|\Psi\|^{2}_{D^{s,2}(\mathbb{R}^{N})} -\frac{N}{p}\|\Phi\|_{L^{p}(\mathbb{R}^{N})}^{p}=0, \end{aligned}

and

\begin{aligned} I(\Psi)=\inf_{u\in \mathcal{M}}I(u)=c \end{aligned}

where \mathcal{M}:=\{u\in E\backslash\{0\}|P(u)=0\} is the Nehari-Pohožaev manifold.

Used by Remark 1.1
Proof (The proof of Theorem 1.3) of Theorem 1.3

Let \{u_{n}\} be a bounded sequence of c>0 which is in Lemma 5.7. If \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x=0, then

\begin{aligned} c=\frac{1}{2}\lim_{n\rightarrow\infty}\|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{1}{2}\lim_{n\rightarrow\infty}\|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

and

\begin{aligned} 0=\frac{N-2}{2} \lim_{n\rightarrow\infty} \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{N-2s}{2} \lim_{n\rightarrow\infty} \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

which further gives

\begin{aligned} 0=\lim_{n\rightarrow\infty}\|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2}, \end{aligned}
\begin{aligned} 0= \lim_{n\rightarrow\infty} \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

and

\begin{aligned} c=0. \end{aligned}

This is a contradiction. Hence, \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x>0.

By Lemma 2.5 and \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x>0, we know that there exists \{x_{n}\}\subset \mathbb{R}^{N} such that \{\bar{u}_{n}:=u_{n}(x+x_{n})\} convergence strongly and a.e. to \bar{u}\not\equiv0 in L^{p}_{loc}(\mathbb{R}^{N}). Clearly, \{\bar{u}_{n}\} satisfies

\begin{aligned} I(\bar{u}_{n})\to c,~~ P(\bar{u}_{n})\to 0,~~\mathrm{and}~~ \|I'(\bar{u}_{n})\|_{E^{-1}}(1+\|\bar{u}_{n}\|_{E})\to 0,\ \mathrm{as}\ n\rightarrow \infty. \end{aligned}

Up to a subsequence, there exists u such that

\begin{aligned} \bar{u}_{n}\rightharpoonup \bar{u} \; \mathrm{in} ~ E, \;\; \bar{u}_{n}\rightarrow \bar{u} ~ \mathrm{a.e. ~in} ~ \mathbb{R}^{N}. \end{aligned}

Moreover, from Lemma 5.8, we get

\begin{aligned} I(\bar{u})=c,~~P(\bar{u})=0~\mathrm{and}~\|I'(\bar{u})\|_{E^{-1}}=0. \end{aligned}
Remark

Remark 1.1

SH4Z

Due to the appearance of the mixed local and nonlocal operator and the loss of L^{2} term in equation (?), it is hard to get the regularity of weak solutions. Hence, it is not easy to show that all weak solutions of equation (?) satisfy Pohožaev identy. Nevertheless, combining Theorems 1.2 and 1.3, we know that the ground sate solutions of equation (?) satisfy Pohožaev identy.

Theorem

Theorem 1.4

GUEY

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Then we have the following results.

  1. (1) The best constant of (?) is

    \begin{aligned} C_{N,p,s}^{-1} =\left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \left[ \frac{2p}{p-2} c \right]^{\frac{p-2}{2}} \end{aligned}

    where c=\inf\limits_{u\in \mathcal{N}}I(u).

  2. (2) The optimizer Q is not only a weak solution of equation (?), but it is also a ground state solution and satisfies

    \begin{aligned} I(Q)=c~~\mathrm{and}~~P(Q)=0. \end{aligned}
Proof (Proof of Theorem 1.4) of Theorem 1.4

It follows from Theorem 1.1 that

\begin{aligned} I(Q)\geqslant c, ~~\langle I'(Q),Q\rangle=0, ~~\mathrm{and}~~P(Q)=0. \end{aligned}

From Theorem 1.1 and Lemma 6.1 that

\begin{aligned} C_{N,p,s}^{-1} =& \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}\\ \geqslant& \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}. \end{aligned}

On the other hand, by using Lemma 6.2, one has

\begin{aligned} C_{N,p,s}^{-1} \leqslant \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}. \end{aligned}

Hence \|Q\|_{L^{p}(\mathbb{R}^{N})}=\|\phi\|_{L^{p}(\mathbb{R}^{N})}, and

\begin{aligned} C_{N,p,s}^{-1} = \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}. \end{aligned}

Moreover, one deduces that I(Q)=c.

Applying Lemma 6.1 again, we know

\begin{aligned} c =&I(\phi)\\ =&\frac{p-2}{2p}\int_{\mathbb{R}^{N}}|\phi|^{p}\mathrm{d}x, \end{aligned}

which shows that

\begin{aligned} \|\phi\|_{L^{p}(\mathbb{R}^{N})} =\left[ \frac{2p}{p-2} c \right]^{\frac{1}{p}}. \end{aligned}

Then

\begin{aligned} C_{N,p,s}^{-1} =\left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \left[ \frac{2p}{p-2} c \right]^{\frac{p-2}{2}}. \end{aligned}
Lemma

Lemma 2.1

2B7E

E\hookrightarrow D^{1,2}(\mathbb{R}^{N}) and E\hookrightarrow D^{s,2}(\mathbb{R}^{N}).

Used by Proof, Proof
Proof of Lemma 2.1

It is easy to see that

\begin{aligned} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \leqslant \|u\|_{E}^{2}, \end{aligned}

and

\begin{aligned} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \leqslant \|u\|_{E}^{2}. \end{aligned}

Therefore E\hookrightarrow D^{1,2}(\mathbb{R}^{N}) and E\hookrightarrow D^{s,2}(\mathbb{R}^{N}).

Lemma

Lemma 2.2

H7MC

E\hookrightarrow L^{t}(\mathbb{R}^{N}), t\in [\frac{2N}{N-2s},\frac{2N}{N-2}].

Used by Proof
Proof of Lemma 2.2

Using Hölder’s inequality, we have

\begin{aligned} \int_{\mathbb{R}^{N}} |u|^{t} \mathrm{d}x \leqslant& \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2s}} \mathrm{d}x \right) ^{\frac{(N-2s)[2N-t(N-2)]}{4N(1-s)}} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2}} \mathrm{d}x \right) ^{\frac{(N-2)[t(N-2s)-2N]}{4N(1-s)}}. \end{aligned}

From Lemma 2.1, we know that:

\begin{aligned} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2s}} \mathrm{d}x \right)^{\frac{2}{2_{s}^{*}}} \leqslant \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \leqslant \|u\|_{E}^{2}, \end{aligned}

and

\begin{aligned} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2}} \mathrm{d}x \right)^{\frac{2}{2^{*}}} \leqslant \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \leqslant \|u\|_{E}^{2}. \end{aligned}

Then we get

\begin{aligned} \int_{\mathbb{R}^{N}} |u|^{t} \mathrm{d}x \leqslant& \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2s}} \mathrm{d}x \right) ^{\frac{(tN-2N-2t)(N-2s)}{4N(s-1)}} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2}} \mathrm{d}x \right) ^{\frac{(tN-2N-2ts)(N-2)}{4N(1-s)}}\\ \leqslant&\|u\|_{E}^{t}<\infty. \end{aligned}

The proof is complete.

Lemma

Lemma 2.3

UP28

[Cotsiolis-Tavoularis2004JMAA, Theorem 1.1] Let s\in(0,1] and N>2s. Then there exists a constant S_{s}>0 such that for any u\in D^{s,2}(\mathbb{R}^{N}),

\begin{aligned} \|u\|_{L^{2_{s}^{*}}(\mathbb{R}^{N})}^{2} \leqslant S_{s}^{-1} \|u\|^{2}_{D^{s,2}}(\mathbb{R}^{N}). \end{aligned}
Used by Proof, Proof
Lemma

Lemma 2.4

1CQK

Let N\geqslant 3, s\in(0,1) and q\in(2^{*}_{s},2^{*}). Then the following inequality holds

\begin{aligned} \int_{\mathbb{R}^{N}}|u|^{q}\mathrm{d}x\leqslant2C(N+1)^{2} \left(\sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u|^{q} \mathrm{d}x\right)^{\frac{q-2}{q}} \|u\|_{E}^{2}, \end{aligned}

for all u\in E.

Used by Proof, Proof
Proof of Lemma 2.4

Let u\in E and q\in(2_{s}^{*},2^{*}). From Hölder’s inequality and Lemma 2.3, we have

\begin{aligned} &\int_{B(z,1)}|u|^{q}\mathrm{d}x\\ \leqslant& C \left( \int_{B(z,1)}\int_{B(z,1)}\frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}}\mathrm{d}x\mathrm{d}y + \int_{B(z,1)} |\nabla u|^{2} \mathrm{d}x\right)^{\frac{q}{2}}. \end{aligned}

Applying (Proof), we know that

\begin{aligned} &\int_{B(z,1)}|u|^{q}\mathrm{d}x\\ =& \left(\int_{B(z,1)}|u|^{q}\mathrm{d}x\right)^{\frac{2}{q}} \left(\int_{B(z,1)}|u|^{q}\mathrm{d}x\right) ^{\frac{q-2}{q}}\\ \leqslant& C \left( \int_{B(z,1)}\int_{B(z,1)}\frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}}\mathrm{d}x\mathrm{d}y + \int_{B(z,1)} |\nabla u|^{2} \mathrm{d}x\right) \left(\int_{B(z,1)}|u|^{q}\mathrm{d}x\right) ^{\frac{q-2}{q}}. \end{aligned}

Covering \mathbb{R}^{N} by balls of radius 1, in such a way that each point of \mathbb{R}^{N} is contained in at most N+1 balls. We get

\begin{aligned} &\int_{\mathbb{R}^{N}}|u|^{q}\mathrm{d}x\\ \leqslant& C(N+1) \left(\sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u|^{q} \mathrm{d}x\right)^{\frac{q-2}{q}} \left( \int_{\mathbb{R}^{N}}\int_{\mathbb{R}^{N}}\frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}}\mathrm{d}y\mathrm{d}x + \int_{\mathbb{R}^{N}} |\nabla u|^{2} \mathrm{d}x\right). \end{aligned}
Lemma

Lemma 2.5

IQU9

Let N\geqslant3, 0<s<1, q\in (2^{*}_{s},2^{*}), and \{u_{n}\} be a bounded sequence in E satisfing \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}}|u_{n}|^{q}\mathrm{d}x>0. Then there exists \{z_{n}\}\subset \mathbb{R}^{N} such that \{\bar{u}_{n}:=u_{n}(x+z_{n})\} convergence strongly to \bar{u}\not\equiv0 in L^{q}_{loc}(\mathbb{R}^{N}).

Used by Proof, Proof
Proof of Lemma 2.5

Note that \{u_{n}\} is a bounded sequence in E. Up to a subsequence, we may assume that

\begin{aligned} u_{n}\rightharpoonup u ~ \mathrm{in} ~ E,~ u_{n}\rightarrow u ~ \mathrm{a.e. ~in} ~\mathbb{R}^{N}, u_{n}\rightarrow u ~ \mathrm{in} ~ L^{q}_{loc}(\mathbb{R}^{N}). \end{aligned}

Applying Lemma 2.4 and \lim\limits_{n\to\infty}\int_{\mathbb{R}^{N}}|u_{n}|^{q}\mathrm{d}x>0, there exists C>0 such that

\begin{aligned} \sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u|^{q} \mathrm{d}x\geqslant C>0. \end{aligned}

Note that \{u_{n}\} is bounded in E and E\hookrightarrow L^{q}(\mathbb{R}^{N}), we have

\begin{aligned} \sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u|^{q} \mathrm{d}x\leqslant\int_{\mathbb{R}^{N}} |u_{n}|^{q} \mathrm{d}x \leqslant C. \end{aligned}

Hence, there exists C_{0} such that

\begin{aligned} C_{0} \leqslant \sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u|^{q} \mathrm{d}x\leqslant C_{0}^{-1}. \end{aligned}

From the above inequality, there exists z_{n}\in \mathbb{R}^{N} such that

\begin{aligned} \int_{B(z_{n},1)} |u_{n}|^{q} \mathrm{d}x \geqslant \sup_{z\in \mathbb{R}^{N}} \int_{B(z,1)} |u_{n}|^{q} \mathrm{d}x - \frac{C}{2n} \geqslant C_{1}>0. \end{aligned}

Set \bar{u}_{n}:=u_{n}(x+z_{n}). Then \|\bar{u}_{n}\|_{E}=\|u_{n}\|_{E} and

\begin{aligned} \int_{B(0,1)} |\bar{u}_{n}|^{q} \mathrm{d}x \geqslant C_{1}>0. \end{aligned}

Up to a subsequence, there exists \bar{u} such that

\begin{aligned} \bar{u}_{n}\rightharpoonup \bar{u} \; \mathrm{in} ~ E, \;\; \bar{u}_{n}\rightarrow \bar{u} ~ \mathrm{a.e. ~in} ~ \mathbb{R}^{N}. \end{aligned}

Using the compact embedding E \hookrightarrow L^{q}_{loc}(\mathbb{R}^{N}), we deduce that \bar{u}\not\equiv0.

Lemma

Lemma 3.1

G1BF

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). For all u\in E, there exists C_{N,p,s}>0 (independent of u) such that

\begin{aligned} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x \leqslant C_{N,p,s} \left( \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |\nabla u|^{2} \mathrm{d}x \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}. \end{aligned}
Used by Proof, Proof
Proof of Lemma 3.1

From Lemma 2.3, we have

\begin{aligned} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2s}} \mathrm{d}x \right) ^{\frac{N-2s}{N}} \leqslant& S_{s}^{-1} \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y, \end{aligned}

and

\begin{aligned} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2}} \mathrm{d}x \right) ^{\frac{N-2}{N}} \leqslant& S_{1}^{-1} \int_{\mathbb{R}^{N}} |\nabla u|^{2} \mathrm{d}x. \end{aligned}

Plugging (Proof) and (Proof) into (Proof), we deduce that

\begin{aligned} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x \leqslant& \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2s}} \mathrm{d}x \right) ^{\frac{N-2s}{N}\cdot\frac{2N-p(N-2)}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |u|^{\frac{2N}{N-2}} \mathrm{d}x \right) ^{\frac{N-2}{N}\cdot \frac{p(N-2s)-2N}{4(1-s)}}\\ \leqslant& C \left( \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|u(x)-u(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |\nabla u|^{2} \mathrm{d}x \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}. \end{aligned}
Lemma

Lemma 3.2

FUW5

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). There exists a radial minimzing sequence \{u_{n}\}\subset E of C_{N,p,s}^{-1} such that

\begin{aligned} \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} =1. \end{aligned}
Used by Proof
Proof of Lemma 3.2

Step 1. From Lemma 3.1, we know that there exists a minimzing sequence \{w_{n}\}\subset E such that

\begin{aligned} \lim\limits_{n\to\infty} W(w_{n})=C_{N,p,s}^{-1}. \end{aligned}

It is easy to see that \{|w_{n}|\}\subset E and \lim\limits_{n\to\infty} W(|w_{n}|)=C_{N,p,s}^{-1}.

Let |w_{n}|^{*} be the symmetric decreasing rearrangement of |w_{n}|. From [14] and [21], one has

\begin{aligned} \||w_{n}|^{*}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \leqslant \||w_{n}|\|_{D^{1,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

and

\begin{aligned} \||w_{n}|^{*}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \leqslant \||w_{n}|\|_{D^{1,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

and

\begin{aligned} \int_{\mathbb{R}^{N}} ||w_{n}|^{*}|^{p} \mathrm{d}x = \int_{\mathbb{R}^{N}} |w_{n}|^{p} \mathrm{d}x. \end{aligned}

Hence C_{N,p,s}^{-1}\leqslant\lim\limits_{n\to\infty} W(|w_{n}|^{*})\leqslant\lim\limits_{n\to\infty} W(|w_{n}|)=C_{N,p,s}^{-1}. Set

\begin{aligned} v_{n}=|w_{n}|^{*}. \end{aligned}

Then \lim\limits_{n\to\infty} W(v_{n})=C_{N,p,s}^{-1}.

Step 2. We rescale the sequence \{v_{n}\} by

\begin{aligned} u_{n}=\lambda_{1,n}v_{n}(\lambda_{2,n}x), \end{aligned}

where

\begin{aligned} \lambda_{1,n} = \frac {\left[\int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|v_{n}(x)-v_{n}(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y \right]^{\frac{N-2}{1-s}}} {\left[\int_{\mathbb{R}^{N}} |\nabla v_{n}(x)|^{2} \mathrm{d}x \right]^{\frac{N-2s}{1-s}}}, \end{aligned}

and

\begin{aligned} \lambda_{2,n} = \left[ \frac {\int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|v_{n}(x)-v_{n}(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y} {\int_{\mathbb{R}^{N}} |\nabla v_{n}(x)|^{2} \mathrm{d}x} \right]^{\frac{1}{2-2s}}. \end{aligned}

Then

\begin{aligned} 1=& \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|u_{n}(x)-u_{n}(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y\\ =& \lambda_{1,n}^{2} \lambda_{2,n}^{2s-N} \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|v_{n}(x)-v_{n}(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y, \end{aligned}

and

\begin{aligned} 1= \int_{\mathbb{R}^{N}} |\nabla u_{n}|^{2} \mathrm{d}x =& \lambda_{1,n}^{2} \lambda_{2,n}^{2-N} \int_{\mathbb{R}^{N}} |\nabla v_{n}(x)|^{2} \mathrm{d}x, \end{aligned}

and

\begin{aligned} \int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x =&\lambda_{1,n}^{p} \lambda_{2,n}^{-N} \int_{\mathbb{R}^{N}} |v_{n}(x)|^{p} \mathrm{d}x. \end{aligned}

Step 3. The Weinstein functional W(\cdot) is invariant with respect to our scaling:

\begin{aligned} W(u_{n}) =& \frac{\left( \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}} {\int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x}\\ =& \frac{[\lambda_{1,n}^{2} \lambda_{2,n}^{2s-N}]^{\frac{2N-p(N-2)}{4(1-s)}}[\lambda_{1,n}^{2} \lambda_{2,n}^{2-N}]^{\frac{p(N-2s)-2N}{4(1-s)}}}{\lambda_{1,n}^{p} \lambda_{2,n}^{-N}} W(v_{n})\\ =&W(v_{n}). \end{aligned}

This impiles that \{u_{n}\}\subset E is a radial bounded minimizing sequence.

Lemma

Lemma 3.3

778Z

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Let \{u_{n}\}\subset E be a radial minimzing sequence of C_{N,p,s}^{-1} which is in Lemma 3.2. Then u_{n}\to u\not\equiv 0 in E, and

\begin{aligned} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} =1 \end{aligned}

and

\begin{aligned} W(u)=C_{N,p,s}^{-1}. \end{aligned}
Used by Proof
Proof of Lemma 3.3

Step 1. From Lemma 3.2, we have \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} =1 and

\begin{aligned} 0<C_{N,p,s} =&\lim\limits_{n\to\infty} \frac {\int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x} {\left( \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{|u_{n}(x)-u_{n}(y)|^{2}}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |\nabla u_{n}|^{2} \mathrm{d}x \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}}\\ =&\lim\limits_{n\to\infty} \int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x. \end{aligned}

Combining \lim\limits_{n\to\infty} \int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x\geqslant C>0 and Lemma 2.4, one has that there exists a sequence of \{z_{n}\}\subset \mathbb{R}^{N} such that

\begin{aligned} \int_{B(z_{n},1)} |u_{n}|^{p} \mathrm{d}x \geqslant C_{1}>0. \end{aligned}

where C_{1}>0 is a constant independent of u_{n}.

Step 2. We show that \{z_{n}\} is a bounded sequence in \mathbb{R}^{N}. Suppose on the contrary that |z_{n}|\rightarrow\infty as n\rightarrow\infty. We have

\begin{aligned} \sup_{|x|>0} |u_{n}(x)| \leqslant \frac{C}{|x|^{\frac{N-2}{2}}} \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})} \leqslant \frac{C}{|x|^{\frac{N-2}{2}}}. \end{aligned}

For any \left[\frac{C_{1}}{|B(0,1)|}\right]^{\frac{1}{p}}>\varepsilon>0, there exists an M>0 such that for any n>M we can deduce that

\begin{aligned} |u_{n}(x)| \leqslant \frac{C} {||x_{n}|-1|^{\frac{N-2}{2}}} \leqslant \varepsilon, \ ~x\in B^{c}(0,||x_{n}|-1|). \end{aligned}

Note that B(z_{n},1)\subset B^{c}(0,||x_{n}|-1|). Then

\begin{aligned} \int_{B(z_{n},1)} |u_{n}|^{p} \mathrm{d}x \leqslant \varepsilon^{p} \int_{B(z_{n},1)} \mathrm{d}x = \varepsilon^{p} |B(z_{n},1)| = \varepsilon^{p} |B(0,1)| <C_{1}. \end{aligned}

This contradicts (Proof). Then we know that there exists C>0 such that B(z_{n},1)\subset B(0,C) and

\begin{aligned} 0<C_{1}\leqslant\int_{B(z_{n},1)} |u_{n}|^{p} \mathrm{d}x \leqslant \int_{B(0,C)} |u_{n}|^{p} \mathrm{d}x. \end{aligned}

From E\hookrightarrow\hookrightarrow L^{p}(B(0,C)), we know that

\begin{aligned} u_{n}\rightharpoonup u\not\equiv 0. \end{aligned}

Step 3. It follows from Brézis-Lieb Lemma, Lemma 3.1 and Young’s inequality that

\begin{aligned} 0<&C_{N,p,s}\\ =&\lim\limits_{n\to\infty} \int_{\mathbb{R}^{N}} |u_{n}|^{p} \mathrm{d}x\\ =& \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x +\lim\limits_{n\to\infty} \int_{\mathbb{R}^{N}} |u_{n}-u|^{p} \mathrm{d}x\\ =& C_{N,p,s} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{\frac{2N-p(N-2)}{2(1-s)}} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{\frac{p(N-2s)-2N}{2(1-s)}}\\ &+\lim\limits_{n\to\infty} C_{N,p,s} \|u_{n}-u\|_{D^{s,2}(\mathbb{R}^{N})}^{\frac{2N-p(N-2)}{2(1-s)}} \|u_{n}-u\|_{D^{1,2}(\mathbb{R}^{N})}^{\frac{p(N-2s)-2N}{2(1-s)}}\\ \leqslant& C_{N,p,s} \left[ \frac{2N-p(N-2)}{2p(1-s)} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{p} +\frac{p(N-2s)-2N}{2p(1-s)} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{p} \right]\\ &+\lim\limits_{n\to\infty} C_{N,p,s} \left[ \frac{2N-p(N-2)}{2p(1-s)} \|u_{n}-u\|_{D^{s,2}(\mathbb{R}^{N})}^{p} +\frac{p(N-2s)-2N}{2p(1-s)} \|u_{n}-u\|_{D^{1,2}(\mathbb{R}^{N})}^{p} \right]\\ \leqslant& C_{N,p,s} \frac{2N-p(N-2)}{2p(1-s)} \left[ \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} + \|u_{n}-u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \right]^{\frac{p}{2}}\\ &+C_{N,p,s} \frac{p(N-2s)-2N}{2p(1-s)} \lim\limits_{n\to\infty} \left[ \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \|u_{n}-u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \right]^{\frac{p}{2}}\\ =&C_{N,p,s} \frac{2N-p(N-2)}{2p(1-s)} +C_{N,p,s} \frac{p(N-2s)-2N}{2p(1-s)}\\ =&C_{N,p,s}. \end{aligned}

Therefore all the above inequalities must be equalities. Hence,

\begin{aligned} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}=0~ \mathrm{or}~~\|u_{n}-u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}=0. \end{aligned}

Note that u\not\equiv0. We know that \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}\not=0 and \|u_{n}-u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}=0 and

\begin{aligned} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} =1 \end{aligned}

and

\begin{aligned} W(u)=C_{N,p,s}^{-1} \end{aligned}

Thus u is a minimizer of C_{N,p,s}^{-1}.

Lemma

Lemma 3.4

JL2A

Let N\geqslant3, s\in(0,1) and p\in (2^{*}_{s},2^{*}). Let u be a minimizer of C_{N,p,s}^{-1}. Then Q=\lambda_{1}u(\lambda_{2}x) is a minimizer of C_{N,p,s}^{-1}, and also is a weak solution of

\begin{aligned} -\Delta Q +(-\Delta)^{s}Q = |Q|^{p-2}Q, \ \ x\in\mathbb{R}^{N}, \end{aligned}

where

\begin{aligned} \begin{cases} \lambda_{1}= \left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \right]^{\frac{1}{(1-s)(p-2)}} \left[ \frac{2p(1-s)}{p(N-2s)-2N} \frac{1}{\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}} \right]^{\frac{1}{p-2}},\\ \lambda_{2}= \left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \right]^{\frac{1}{2-2s}}. \end{cases} \end{aligned}
Used by Proof
Proof of Lemma 3.4

From Lemma 3.3, one has that u is minimizer of W(u)=C_{N,p,s}^{-1}. Then

\begin{aligned} \left.\frac{\mathrm{d}} {\mathrm{d}\varepsilon}\right|_{\varepsilon=0}W(u+\varepsilon \varphi)=0, \end{aligned}

for all \varphi\in E. Thus, u is a weak solution of the Euler-Lagrange equation

\begin{aligned} &-\Delta u +\frac{2N-p(N-2)}{p(N-2s)-2N} \frac{ \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} } {\|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}} (-\Delta)^{s}u\\ =& \frac{2p(1-s)}{p(N-2s)-2N} \frac{\|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}} {\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}} |u|^{p-2}u, \ \ x\in\mathbb{R}^{N}, \end{aligned}

Set Q=\lambda_{1}u(\lambda_{2}x), where

\begin{aligned} \begin{cases} \lambda_{2}= \left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \frac{ \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} } {\|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}}\right]^{\frac{1}{2-2s}}\\ \lambda_{1}=\left[ \frac{p(N-2s)-2N}{2N-p(N-2)} \frac{ \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} } {\|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}}\right]^{\frac{1}{(1-s)(p-2)}} \left[ \frac{2p(1-s)}{p(N-2s)-2N} \frac{\|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}} {\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}} \right]^{\frac{1}{p-2}} \end{cases} \end{aligned}

Then Q is a weak solution of

\begin{aligned} -\Delta Q +(-\Delta)^{s}Q = |Q|^{p-2}Q, \ \ x\in\mathbb{R}^{N}, \end{aligned}

and

\begin{aligned} W(Q)=W(u). \end{aligned}
Lemma

Lemma 3.5

NGPH

We have

\begin{aligned} \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2N-p(N-2)} \|Q\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2p(1-s)} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{p}, \end{aligned}
\begin{aligned} \frac{N-2}{2} \|Q\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|Q\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =\frac{N}{p}\int_{\mathbb{R}^{N}}|Q|^{p}\mathrm{d}x, \end{aligned}

and

\begin{aligned} C_{N,p,s}^{-1}= \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |Q|^{p} \mathrm{d}x \right)^{\frac{p-2}{2}}. \end{aligned}
Proof of Lemma 3.5

From Lemma 3.4 and \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2}=\|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}=1, we compute

\begin{aligned} \frac{ \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} } {\|Q\|_{D^{s,2}(\mathbb{R}^{N})}^{2}} = \frac{ \lambda_{1}^{2}\lambda_{2}^{2-N} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} } { \lambda_{1}^{2}\lambda_{2}^{2s-N} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}} = \lambda_{2}^{2-2s} = \frac{p(N-2s)-2N}{2N-p(N-2)}, \end{aligned}

and

\begin{aligned} \frac{ \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} } {\|Q\|_{L^{p}(\mathbb{R}^{N})}^{p}} =& \frac{ \lambda_{1}^{2}\lambda_{2}^{2-N} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} } { \lambda_{1}^{p}\lambda_{2}^{-N} \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}}\\ =& \frac{ \lambda_{1}^{2-p}\lambda_{2}^{2} } {\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}}\\ =& \frac{p(N-2s)-2N}{2p(1-s)}. \end{aligned}

Hence

\begin{aligned} \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2N-p(N-2)} \|Q\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2p(1-s)} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{p}. \end{aligned}

Using the fact that (Proof), we have

\begin{aligned} \frac{N-2}{2} \|Q\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|Q\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =\frac{N}{p}\int_{\mathbb{R}^{N}}|Q|^{p}\mathrm{d}x. \end{aligned}

Since Q is a weak solution, we have

\begin{aligned} \|Q\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \|Q\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =\int_{\mathbb{R}^{N}}|Q|^{p}\mathrm{d}x. \end{aligned}

Then

\begin{aligned} C_{N,p,s}^{-1} =& \frac {\left( \|Q\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \|Q\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}} {\int_{\mathbb{R}^{N}} |Q|^{p} \mathrm{d}x}\\ =& \frac {\left( \frac{2N-p(N-2)}{2p(1-s)} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{p} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \|Q\|_{L^{p}(\mathbb{R}^{N})}^{p} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}} {\int_{\mathbb{R}^{N}} |Q|^{p} \mathrm{d}x}\\ =& \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \left( \int_{\mathbb{R}^{N}} |Q|^{p} \mathrm{d}x \right)^{\frac{p-2}{2}}. \end{aligned}
Lemma

Lemma 4.1

89FF

Let N\geqslant3 and 0<s<1. Then we have

  1. (1) the functional I has mountain pass geometric structure;

  2. (2) for any u\in E\setminus\{0\}, there exists a unique t_{u}>0 such that t_{u}u\in \mathcal{N} and I(t_{u}u)=\max\limits_{t>0}I(tu);

  3. (3) \bar{c}=\inf\limits_{u\in\mathcal{N}}I(u)>0;

  4. (4) c=\bar{c}=\bar{\bar{c}}, where

    \begin{aligned} \bar{\bar{c}}:=\inf\limits_{u\in E\setminus\{0\}}\sup\limits_{t\geqslant 0}I(tu). \end{aligned}
  5. (5) For u\in \mathcal{N}, we have \Phi'(u)\not=0, where

    \begin{aligned} \Phi(u)=\langle I'(u),u\rangle=\|u\|_{E}^{2}-\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x, \end{aligned}

    and

    \begin{aligned} \langle\Phi'(u),u\rangle =&2\|u\|_{E}^{2} -p\beta\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

    Moreover, if \bar{u}\in \mathcal{N} and I(\bar{u})=c, then u is a ground state solution for equation (?).

Used by Proof
Proof (Proof of Lemma 4.1) of Lemma 4.1

(1) Using Lemma 2.2, one has

\begin{aligned} I(u) \geqslant \|u\|_{E}^{2} -C\|u\|_{E}^{p}. \end{aligned}

We keep in mind that 2<p. Then there exists a sufficiently small positive number \rho such that

\begin{aligned} \varsigma := \inf_{\|u\|_{E}=\rho} I(u)>0=I(0). \end{aligned}

For u\in E\setminus\{0\}, we have

\begin{aligned} I(tu) = \frac{t^{2}}{2} \|u\|_{E}^{2} -\frac{t^{p}}{p} \int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

From 2<p, it follows that I(tu)<0 for t large enough.

From above, we can choose t_{1}>0 corresponding to u such that I(t_{1}u)<0 for t>t_{1} and \|t_{1}u\|_{E}>\rho.

(2) For any u\in E\setminus\{0\} and t\in (0,\infty), we define

\begin{aligned} f_{1}(t)=I(tu)=&\frac{t^{2}}{2}\|u\|_{E}^{2}-\frac{t^{p}}{p}\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

We compute

\begin{aligned} f_{1}'(t) =&t\|u\|_{E}^{2} -t^{p-1} \int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

We know that f_{1}'(\cdot)=0 iff

\begin{aligned} \|u\|_{E}^{2} =t^{p-2} \int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

Let

\begin{aligned} f_{2}(t) =t^{p-2} \int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

Clearly, \lim\limits_{t\rightarrow0}f_{2}(t)\rightarrow0, \lim\limits_{t\rightarrow+\infty}f_{2}(t)\rightarrow+\infty . By using the intermediate value theorem, we know that there exists a 0<t_{u}<\infty such that

\begin{aligned} f_{2}(t_{u})=\|u\|_{E}^{2}. \end{aligned}

Furthermore, it is easy to see that f_{2}(\cdot) is strictly increasing on (0,\infty). Then we get the uniqueness of t_{u}. And then,

\begin{aligned} \|u\|_{E}^{2} =t_{u}^{p-2} \int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x, \end{aligned}

which gives

\begin{aligned} \|t_{u}u\|_{E}^{2} =\beta\int_{\mathbb{R}^{N}}|t_{u}u|^{p}\mathrm{d}x +\int_{\mathbb{R}^{N}}|t_{u}u|^{2^{*}}\mathrm{d}x. \end{aligned}

This implies that t_{u}u\in \mathcal{N}.

(3) By applying \langle I'(u),u\rangle=0, we know

\begin{aligned} 0 =\langle I'(u),u\rangle\geqslant \|u\|_{E}^{2} -C\|u\|_{E}^{p}, \end{aligned}

which implies

\begin{aligned} C\|u\|_{E}^{p-2} \geqslant 1, \end{aligned}

and

\begin{aligned} \|u\|_{E}^{2} \geqslant C. \end{aligned}

Then, for u\in \mathcal{N}, we get

\begin{aligned} I(u) =& I(u)-\frac{1}{p}\langle I'(u),u\rangle\\ =&\frac{1}{2}\|u\|_{E}^{2} -\beta\frac{1}{p}\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x -\frac{1}{p} \left( \|u\|_{E}^{2} -\beta\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x \right)\\ =& \left(\frac{1}{2}-\frac{1}{ p}\right) \|u\|_{E}^{2} \geqslant C. \end{aligned}

Hence, I is bounded from below on \mathcal{N}. And then \bar{c}>0.

(4) By virtue of Lemma 4.1 (2), we have the following result directly

\begin{aligned} \bar{c}=\bar{\bar{c}}. \end{aligned}

For any u\in E\setminus\{0\}, there exists some \tilde{t}>0 large, such that I(\tilde{t}u)<0. Define a path \gamma:[0, 1]\to E by \gamma(t) = t\tilde{t}u. Clearly, \gamma\in\Gamma and

\begin{aligned} c\leqslant \bar{\bar{c}}. \end{aligned}

On the other hand, for each path \gamma\in\Gamma, let g(t):=\langle I'(\gamma(t)),\gamma(t)\rangle. Then, g(0)=0 and g(t)>0 for t>0 small. A direct calculation gives

\begin{aligned} I(\gamma(1)) -\frac{1}{p}\langle I'(\gamma(1)),\gamma(1)\rangle \geqslant\left(\frac{1}{2}-\frac{1}{p}\right)\|\gamma(1)\|_{E}^{2} \geqslant 0, \end{aligned}

which implies

\begin{aligned} \langle I'(\gamma(1)),\gamma(1)\rangle \leqslant p\cdot I(\gamma(1)) =p\cdot I(\tilde{t}u)<0. \end{aligned}

Thus, there exists \tilde{\tilde{t}} \in(0,1) such that g(\tilde{\tilde{t}}) = 0, i.e. \gamma(\tilde{\tilde{t}})\in \mathcal{N} and c\geqslant \bar{c}. This deduces c=\bar{c}=\bar{\bar{c}}.

(5) For u\in \mathcal{N}, it follows from (?) and (?) that

\begin{aligned} \langle\Phi'(u),u\rangle =&\langle\Phi'(u),u\rangle-p\Phi(u)\\ =&\left(2\|u\|_{E}^{2}-p\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x\right) -p\left(\|u\|_{E}^{2} -\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x\right)\\ =&(2-p)\|u\|_{E}^{2} <0. \end{aligned}

Thus, \Phi'(u)\not=0 for u\in \mathcal{N}.

If u\in \mathcal{N} and I(u)=\bar{c}, from \bar{c} is the minimum of I on \mathcal{N}, and then using the Lagrange multiplier theorem, we know that there exists \lambda\in \mathbb{R} such that I'(u)=\lambda \Phi'(u). So

\begin{aligned} \langle \lambda \Phi'(u),u\rangle =\langle I'(u),u\rangle=\Phi(u)=0. \end{aligned}

This shows \lambda=0 and I'(u)=0. Thus, u is a ground state solution for equation (?).

Definition

Definition 4.1

WB6K

Let the sequence \{u_{n}\}\subset E satisfy the condition

\begin{aligned} I(u_{n})\rightarrow c ~\mathrm{and}~I'(u_{n})\rightarrow 0 ~\mathrm{in}~E^{-1}, ~{\rm as}~n\to\infty. \end{aligned}

Then \{u_{n}\} is called the Palais-Smale sequence of I at level c, short for (PS)_{c} sequence, where E^{-1} is the dual space of E.

Lemma

Lemma 4.2

5YSL

Let N\geqslant3 and 0<s<1. Then there exists a bounded (PS)_{c} sequence \{u_{n}\}\subset \mathcal{N} such that

\begin{aligned} I(u_{n})\rightarrow c \ \,\mathrm{and}\ \, \|I'(u_n)\|_{E^{-1}}\rightarrow 0,\ \mathrm{as}\ n\rightarrow \infty. \end{aligned}
Used by Proof
Proof of Lemma 4.2

From Lemmas 4.1 (2) and (4), we know that \mathcal{N}\not=\emptyset and \inf\limits_{u\in \mathcal{N}}I(u)=\bar{c}=c. By the Ekeland’s variational principle, there exist \{u_{n}\}\subset \mathcal{N} and \lambda_{n}\in \mathbb{R} such that

\begin{aligned} I(u_{n})\rightarrow \bar{c} ~\mathrm{and}~I'(u_{n})-\lambda_{n}\Phi'(u_{n})\rightarrow 0 ~\mathrm{in}~E^{-1}, ~{\rm as}~n\to\infty. \end{aligned}

So

\begin{aligned} \bar{c}=I(u_{n}) =I(u_{n}) -\frac{1}{p}\langle I'(u_{n}),u_{n}\rangle \geqslant\left(\frac{1}{2}-\frac{1}{p}\right) \|u_{n}\|_{E}^{2}, \end{aligned}

which implies that \{u_{n}\} is bounded in E.

Taking n\to\infty, we have

\begin{aligned} |\langle I'(u_{n}),u_{n}\rangle-\langle\lambda_{n}\Phi'(u_{n}),u_{n}\rangle| \leqslant \|I'(u_{n}) -\lambda_{n}\Phi'(u_{n})\|_{E^{-1}}\|u_{n}\|_{E} \rightarrow 0, \end{aligned}

we have

\begin{aligned} \langle I'(u_{n}),u_{n}\rangle -\lambda_{n}\langle \Phi'(u_{n}),u_{n}\rangle\rightarrow 0, ~{\rm as}~n\to\infty. \end{aligned}

Note that \{u_{n}\}\subset \mathcal{N}. From Lemma 4.1 (5), we have

\begin{aligned} \langle I'(u_{n}),u_{n}\rangle=0, \end{aligned}

and

\begin{aligned} \langle \Phi'(u_{n}),u_{n}\rangle\not=0. \end{aligned}

Combining (?)-(?), we conclude that \lambda_{n}\rightarrow0.

It follows from Hölder’s and Sobolev’s inequalities that

\begin{aligned} &\|I'(u_{n})\|_{E^{-1}}\\ =&\sup_{\varphi\in E,\|\varphi\|_{E}=1} |\langle \Phi'(u_{n}),\varphi\rangle|\\ =&\sup_{\varphi\in E,\|\varphi\|_{E}=1} \left| 2\int_{\mathbb{R}^{N}} \nabla u \nabla \varphi \mathrm{d}x -p \int_{\mathbb{R}^{N}} |u_{n}|^{p-2}u_{n}\varphi \mathrm{d}x \right.\\ &\left.+2 \int_{\mathbb{R}^{N}} \int_{\mathbb{R}^{N}} \frac{(u_{n}(x)-u_{n}(y))(\varphi(x)-\varphi(y))}{|x-y|^{N+2s}} \mathrm{d}x \mathrm{d}y \right|\\ \leqslant&C. \end{aligned}

Then we obtain

\begin{aligned} \|I'(u_{n})\|_{E^{-1}} \leqslant\|I'(u_{n})-\lambda_{n}\Phi'(u_{n})\|_{E^{-1}} +|\lambda_{n}|\|\Phi'(u_{n})\|_{E^{-1}} =o_{n}(1). \end{aligned}

That is, I'(u_{n})\rightarrow0 in E^{-1}. Hence, \{u_{n}\} is a (PS)_{{c}} sequence of I.

Lemma

Lemma 5.1

MUKN

Let C_{1},C_{2},C_{3}>0 and g_{1}:[0,\infty)\to \mathbb{R} as

\begin{aligned} g_{1}(t) = C_{1}t^{N-2} + C_{2}t^{N-2s} - C_{3}t^{N}. \end{aligned}

Then g_{1} has a unique critical point which corresponds to its maximum.

Used by Proof, Proof
Proof of Lemma 5.1

We compute

\begin{aligned} g_{1}'(t) = C_{1}(N-2)t^{N-3} + C_{2}(N-2s)t^{N-2s-1} - C_{3}Nt^{N-1}. \end{aligned}

CLearly, g_{1}'(t)>0 for t>0 small, g_{1}'(t)<0 for t>0 large. Then there exists \bar{t}>0 such that g_{1}'(\bar{t})=0.

To prove the uniqueness of \bar{t}, we suppose that 0<\bar{t}<\bar{\bar{t}} satisfy g_{1}'(\bar{t})=g_{1}'(\bar{\bar{t}})=0. Then

\begin{aligned} \frac{g_{1}'(\bar{t})}{\bar{t}^{N-2s-1}} - \frac{g_{1}'(\bar{\bar{t}})}{\bar{\bar{t}}^{N-2s-1}}=0, \end{aligned}

which shows

\begin{aligned} C_{1}(N-2) (\bar{t}^{2s-2}-\bar{\bar{t}}^{2s-2}) = C_{3}N (\bar{t}^{2s}-\bar{\bar{t}}^{2s}). \end{aligned}

This yields a contradiction.

Lemma

Lemma 5.2

GA9A

For all u\in E\backslash\{0\}, there exists a unique t_{u}> 0 such that u(\frac{x}{t_{u}})\in \mathcal{M}.

Used by Proof, Proof
Proof of Lemma 5.2

Let u=E\setminus\{0\} and t>0. Set

\begin{aligned} g_{2}(t) =& I\left(u\left(\frac{x}{t}\right)\right)\\ =& \frac{t^{N-2}}{2} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{t^{N-2s}}{2} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} - \frac{t^{N}}{p} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x. \end{aligned}

Clearly,

\begin{aligned} g_{2}'(t) =& \frac{N-2}{2}t^{N-3} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{N-2s}{2} t^{N-2s-1} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} - \frac{N}{p}t^{N-1} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x \end{aligned}

and

\begin{aligned} tg_{2}'(t)=P\left(u\left(\frac{x}{t}\right)\right). \end{aligned}

Combining (Proof) and Lemma 5.1, the result is valid.

Lemma

Lemma 5.3

SBGZ
  1. (i) \mathcal{M}\not=\emptyset and for all u\in \mathcal{M}, I(u)=\max\limits_{t>0}I\left(u\left(\frac{x}{t}\right)\right);

  2. (ii) \inf\limits_{u\in \mathcal{M}}I(u)= \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right)=c>0, where

    \begin{aligned} c=\inf\limits_{\gamma\in \Gamma}\sup\limits_{t\in[0, 1]}I(\gamma (t)), \end{aligned}

    and

    \begin{aligned} \Gamma=\{\gamma\in C\left([0, 1], E\right)| \gamma(0)=0, I(\gamma (1))<0\}. \end{aligned}
Proof of Lemma 5.3

(i) From Lemmas 5.1 and 5.2, we can deduce that.

(ii) First, we show \inf\limits_{u\in \mathcal{M}}I(u)>0. For u\in \mathcal{M}, by Lemma 2.1, we have

\begin{aligned} 0 =& P(u) = \frac{N-2}{2} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{N-2s}{2} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} - \frac{N}{p} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x\\ \geqslant& \frac{N-2}{2} \|u\|_{E}^{2} - \frac{N}{p} C_{N,p,s} \|u\|_{E}^{p}, \end{aligned}

which gives

\begin{aligned} \|u\|_{E} >C>0. \end{aligned}

Then

\begin{aligned} I(u) =&I(u)-\frac{1}{N}P(u)\\ =& \frac{1}{2} \left(1-\frac{N-2}{N}\right) \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{1}{2} \left(1-\frac{N-2s}{N}\right) \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}\\ \geqslant& \frac{s}{N} \|u\|_{E}^{2} \geqslant C>0. \end{aligned}

This gives \inf\limits_{u\in \mathcal{M}}I(u)>0.

Second, we prove that \inf\limits_{u\in \mathcal{M}}I(u)= \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right). Clearly,

\begin{aligned} \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right) \leqslant \inf\limits_{u\in \mathcal{M}}\max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right) = \inf\limits_{u\in \mathcal{M}}I(u). \end{aligned}

On the other hand, for any u\in E\setminus\{0\}, by Lemma 5.2, we have

\begin{aligned} \max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right) \geqslant I\left(u\left(\frac{x}{t_{u}}\right)\right) \geqslant \inf\limits_{u\in \mathcal{M}}I(u). \end{aligned}

This implies

\begin{aligned} \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} J\left(u\left(\frac{x}{t}\right)\right) \geqslant \inf\limits_{u\in \mathcal{M}}J(u). \end{aligned}

Thus, the conclusion holds.

Third, we prove that \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} I\left(u\left(\frac{x}{t}\right)\right)=c: For any u\in E\setminus\{0\}, there exists some \tilde{t}>0 large, such that I(\tilde{t}u)<0. Define a path \gamma:[0, 1]\to E by \gamma(t) = t\tilde{t}u. Clearly, \gamma\in\Gamma and

\begin{aligned} c= \inf\limits_{\gamma\in \Gamma}\sup\limits_{t\in[0, 1]}I(\gamma (t)) \leqslant \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} J\left(u\left(\frac{x}{t}\right)\right). \end{aligned}

On the other hand, for each path \gamma\in\Gamma, a direct calculation gives

\begin{aligned} I(\gamma(1)) -\frac{1}{N} P(\gamma(1)) \geqslant \left(\frac{1}{2}-\frac{N-2}{2N}\right)\|\gamma(1)\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \left(\frac{1}{2}-\frac{N-2s}{2N}\right)\|\gamma(1)\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \geqslant 0, \end{aligned}

which implies

\begin{aligned} P(\gamma(1)) \leqslant N\cdot I(\gamma(1)) =N\cdot I(\tilde{t}u)<0. \end{aligned}

Note that P(\gamma(t))>0 for t small enough. Then there exists \tilde{\tilde{t}} \in(0,1) such that P(\gamma(\tilde{\tilde{t}})) = 0, i.e. \gamma(\tilde{\tilde{t}})\in \mathcal{M} and

\begin{aligned} c= \inf\limits_{\gamma\in \Gamma}\sup\limits_{t\in[0, 1]}I(\gamma (t)) \geqslant \inf\limits_{u\in E\setminus\{0\}}\max \limits_{t\geqslant0} J\left(u\left(\frac{x}{t}\right)\right). \end{aligned}
Lemma

Lemma 5.4

0FHU

For every u\in E, t>0,

\begin{aligned} J(u) - \frac{1}{N}P(u) - J\left(u\left(\frac{x}{t}\right)\right) + \frac{t^{N}}{N}P(u) = g_{3}(t) \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + g_{4}(t) \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

where

\begin{aligned} g_{3}(t) =& \frac{1}{N} + \frac{N-2}{2N}t^{N} - \frac{1}{2} t^{N-2} > g_{3}(1)=0, \ \, t\in(0,1)\cup (1,\infty), \\ g_{4}(t) =& \frac{s}{N} + + \frac{N-2s}{2N}t^{N} - \frac{1}{2} t^{N-2s} > g_{4}(1)=0, \ \, t\in(0,1)\cup(1,\infty). \end{aligned}
Proof of Lemma 5.4

Clearly,

\begin{aligned} J(u)-\frac{1}{N}P(u) = \frac{1}{N} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{s}{N} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}, \end{aligned}

and

\begin{aligned} &\frac{t^{N}}{N} P(u) - J\left(u\left(\frac{x}{t}\right)\right)\\ =& \frac{1}{2} \left( \frac{N-2}{N}t^{N} - t^{N-2} \right) \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{1}{2} \left( \frac{N-2s}{N}t^{N} - t^{N-2s}\right) \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}. \end{aligned}

Combining the above two expressions, we have the desired results.

Lemma

Lemma 5.5

TVHU

For all u\in \mathcal{M}, we have P'(u)\not=0.

Proof of Lemma 5.5

For all u\in \mathcal{M}, we recall some useful identities, k\geqslant c,

\begin{aligned} k= \frac{1}{2} \|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{1}{2} \|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})} -\frac{1}{p}\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x, \end{aligned}

and

\begin{aligned} 0=P(u)= \frac{N-2}{2} \|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})} -\frac{N}{p}\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

We prove that P'(u)\not=0. Assume on the contrary that

\begin{aligned} 0=\langle P'(u),u\rangle= (N-2) \|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + (N-2s) \|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})} -N\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

Set

\begin{aligned} x_{1}=\|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})}, ~~x_{2}=\|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})}, x_{3}=\int_{\mathbb{R}^{N}}|u|^{p}\mathrm{d}x. \end{aligned}

Then we have

\begin{aligned} \begin{cases} \frac{1}{2}x_{1}+\frac{1}{2}x_{2}-\frac{1}{p}x_{3}=k\\ \frac{N-2}{2}x_{1}+\frac{N-2s}{2}x_{2}-\frac{N}{p}x_{3}=0\\ (N-2)x_{1}+(N-2s)x_{2}-Nx_{3}=0\\ \end{cases} \Leftrightarrow D \begin{bmatrix} x_{1}\\ x_{2}\\ x_{3}\\ \end{bmatrix} = \begin{bmatrix} k\\ 0\\ 0\\ \end{bmatrix}, \end{aligned}

where D is the coefficient matrix of the system. In the system, the first equation comes from (Proof), the second one holds since (Proof), and the third equation is (Proof).

It is easy to see that

\begin{aligned} \mathrm{det} D = \left| \begin{matrix} \frac{1}{2}&\frac{1}{2}&-\frac{1}{p}\\ \frac{N-2}{2}&\frac{N-2s}{2}&-\frac{N}{p}\\ N-2&N-2s&-N\\ \end{matrix} \right| =&-\frac{N(p-2)(1-s)}{2p} \end{aligned}

and

\begin{aligned} \mathrm{det} D_{1} = \left| \begin{matrix} k&\frac{1}{2}&-\frac{1}{p}\\ 0&\frac{N-2s}{2}&-\frac{N}{p}\\ 0&N-2s&-N\\ \end{matrix} \right| =&-k\frac{N(p-2)(N-2s)}{2p} \end{aligned}

and

\begin{aligned} \mathrm{det} D_{2} = \left| \begin{matrix} \frac{1}{2}&k&-\frac{1}{p}\\ \frac{N-2}{2}&0&-\frac{N}{p}\\ N-2&0&-N\\ \end{matrix} \right| =&k\frac{N(p-2)(N-2)}{2p} \end{aligned}

and

\begin{aligned} \mathrm{det} D_{3} = \left| \begin{matrix} \frac{1}{2}&\frac{1}{2}&k\\ \frac{N-2}{2}&\frac{N-2s}{2}&0\\ N-2&N-2s&0\\ \end{matrix} \right| = 0. \end{aligned}

Then

\begin{aligned} x_{1}=\frac{\mathrm{det} D_{1}}{\mathrm{det} D} = \frac{2p}{N(p-2)(1-s)}\cdot k\cdot \frac{N(p-2)(N-2s)}{2p} = k\frac{N-2s}{1-s}>0 \end{aligned}

and

\begin{aligned} x_{2}=\frac{\mathrm{det} D_{2}}{\mathrm{det} D} = - \frac{2p}{N(p-2)(1-s)}\cdot k\cdot \frac{N(p-2)(N-2)}{2p} = -k\frac{N-2}{1-s}<0 \end{aligned}

and

\begin{aligned} x_{3}=\frac{\mathrm{det} D_{1}}{\mathrm{det} D} = 0 \end{aligned}

Note that N>2s, 1>s and p>2. Then x_{2}<0 and x_{3}=0, what are not possible. Hence, one has P'(u)\not=0 for all u\in \mathcal{M}.

Definition

Definition 5.1

IYXW

If sequence \{u_{n}\}\subset E satisfies the condition

\begin{aligned} I(u_{n})\rightarrow c ~\mathrm{and}~\|I'(u_{n})\|_{E^{-1}}(1+\|u_{n}\|_{E})\rightarrow 0, ~{\rm as}~n\to\infty. \end{aligned}

Then \{u_{n}\} is called the Cerami sequence of I with respect to c, short for (C)_{c} sequence.

Lemma

Lemma 5.6

V9DF

Let X be a Banach space and M a metric space. Let M_{0} be a closed subspace of M and \Gamma_{0}\subset C(M_{0},X). Define

\begin{aligned} \Gamma:=\{\gamma\in C(M,X):\gamma|_{M_{0}}\in\Gamma_{0}\}. \end{aligned}

If \varphi\in C^{1}(X,\mathbb{R}) satisfies

\begin{aligned} a:=\sup\limits_{\gamma_{0}\in \Gamma_{0}}\sup\limits_{u\in M_{0}}I(\gamma (t)) < c:=\inf\limits_{\gamma\in \Gamma}\sup\limits_{u\in M}I(\gamma (t))<\infty, \end{aligned}

then, for every \varepsilon\in(0,\frac{c-a}{2}), \delta>0 and \gamma\in\Gamma such that

\begin{aligned} \sup\limits_{M}\varphi\circ\gamma\leqslant c+\varepsilon, \end{aligned}

there exists u\in X such that

  1. (i) c-2\varepsilon \leqslant \varphi(u)\leqslant c+2\varepsilon;

  2. (ii) \mathrm{dist}(u,\gamma(M))\leqslant 2\delta;

  3. (iii) (1+\|u_{n}\|_{X})\|\varphi'(u)\|_{X^{-1}}<\frac{8\varepsilon}{\delta}, where X^{-1} is the dual space of X.

Used by Proof
Lemma

Lemma 5.7

O8Q8

Let N\geqslant3 and 0<s<1. Then there exists a bounded sequence \{u_{n}\}\subset E such that

\begin{aligned} I(u_{n})\rightarrow c,\ \ P(u_{n})\rightarrow 0, \ \,\mathrm{and}\ \, \|I'(u_{n})\|_{E^{-1}}(1+\|u_{n}\|_{E})\rightarrow 0,\ \mathrm{as}\ n\rightarrow \infty. \end{aligned}
Used by Proof
Proof of Lemma 5.7

Let us define the continuous map h:\mathbb{R}\times E\to E as

\begin{aligned} h(z,v)=v(e^{-z}x),~~z\in\mathbb{R},~~v\in E,~~x\in\mathbb{R}^{N}, \end{aligned}

where \mathbb{R}\times E is a Banach space, with the product norm \|(z,v)\|_{\mathbb{R}\times E}^{2}=|z|^{2}+\|v\|_{E}^{2}. We consider the auxiliary functional

\begin{aligned} J(z,v)=I(h(z,v))= \frac{e^{(N-2)z}}{2} \|v\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{e^{(N-2s)z}}{2} \|v\|^{2}_{D^{s,2}(\mathbb{R}^{N})} -\frac{e^{Nz}}{p}\int_{\mathbb{R}^{N}}|v|^{p}\mathrm{d}x. \end{aligned}

Clearly,

\begin{aligned} \frac{d}{\mathrm{d}z}J(z,v)=P(h(z,v)) \end{aligned}

and

\begin{aligned} \langle\frac{d}{\mathrm{d}v}J(z_{1},v),(z_{2},v)\rangle = \langle\frac{d}{\mathrm{d}v}I(h(z_{1},v)),h(z_{2},w)\rangle = \langle I'(h(z_{1},v)),h(z_{2},w)\rangle \end{aligned}

and

\begin{aligned} \langle J'(z_{1},v),(z_{2},w)\rangle =& \left.\frac{\mathrm{d}}{\mathrm{d}t}\right|_{t=0}J((z_{1}+tz_{2},v+tw))\\ =& \left.\frac{\mathrm{d}}{\mathrm{d}t}\right|_{t=0}I(h(z_{1}+tz_{2},v+tw))\\ =& \langle\frac{d}{\mathrm{d}v}I(h(z_{1},v)),h(z_{2},w)\rangle + P(h(z_{1},v))z_{2}\\ =& \langle I'(h(z_{1},v)),h(z_{2},w)\rangle + P(h(z_{1},v))z_{2}. \end{aligned}

for all z,z_{1},z_{2}\in \mathbb{R} and v,w\in E.

Set the minimax value

\begin{aligned} \tilde{c}:=\inf\limits_{\tilde{\gamma}\in \tilde{\Gamma}}\sup\limits_{t\in [0,1]}J(\tilde{\gamma}(t)), \end{aligned}

where

\begin{aligned} \tilde{\Gamma}=\{\tilde{\gamma}\in C\left([0, 1], \mathbb{R}\times E\right)| \tilde{\gamma}(0)=(0,0), J(\tilde{\gamma}(1))<0\}. \end{aligned}

Note that \Gamma:=\{h\circ \tilde{\gamma}:\tilde{\gamma}\in \tilde{\Gamma}\}. Then

\begin{aligned} c=\tilde{c}. \end{aligned}

By the definition of c, for every n\in \mathbb{N}, there exists \gamma_{n}\in \Gamma such that

\begin{aligned} \sup\limits_{t\in [0,1]}J(0,\gamma(t)) = \sup\limits_{t\in [0,1]}I(\gamma(t)) \leqslant c+\frac{1}{n^{2}}. \end{aligned}

We apply Lemma 5.6 with X=\mathbb{R}\times E, M=[0,1], M_{0}=\{0,1\} and \varphi=J. Let \varepsilon_{n}=\frac{1}{n^{2}}, \delta_{n}=\frac{1}{n} and \tilde{\gamma}(t)=(0,\gamma(t)). Clearly, \varepsilon_{n}=\frac{1}{n^{2}}\in(0,\frac{c}{2}) for n large. Then Lemma 5.6 yields the existence of (z_{n},v_{n})\in \mathbb{R}\times E such that

\begin{aligned} (i)&J(z_{n},v_{n})\to c,\\ (ii)&\|J'(z_{n},v_{n})\|_{(\mathbb{R}\times E)^{-1}}(1+\|(z_{n},v_{n})\|_{\mathbb{R}\times E})< \frac{8}{n},\\ (iii)&\mathrm{dist}((z_{n},v_{n}),\{0\}\times \gamma_{n}([0,1]))\leqslant \frac{2}{n}. \end{aligned}

By using (iii), one has z_{n}\to0 as

\begin{aligned} |z_{n}|=|z_{n}-0|\leqslant \mathrm{dist}((z_{n},v_{n}),\{0\}\times \gamma_{n}([0,1]))\leqslant \frac{2}{n}. \end{aligned}

From (ii) we know

\begin{aligned} \|J'(z_{n},v_{n})\|_{(\mathbb{R}\times E)^{-1}}< \frac{8}{n} \end{aligned}

Furthermore, taking z_{2}=1 and w=0 in (Proof), we have

\begin{aligned} P(h(z_{n},v_{n}))\to 0, ~~\mathrm{as}~~n\to\infty. \end{aligned}

Let u_{n}:=h(z_{n},v_{n}). We have I(u_{n})\to c and P(u_{n})\to 0 as n\to \infty. We now claim that

\begin{aligned} \|I'(u_{n})\|_{E^{-1}}(1+\|u_{n}\|_{E})\to 0,~~\mathrm{as}~~n\to\infty. \end{aligned}

For any \phi\in E, we set \tilde{\phi}=h(-z_{n},\phi). Then

\begin{aligned} \|(0,\tilde{\phi})\|_{\mathbb{R}\times E}^{2} =\|\tilde{\phi}\|_{E}^{2} =e^{-(N-2)z_{n}}\|\phi\|^{2}_{D^{1,2}(\mathbb{R}^{N})} +e^{-(N-2s)z_{n}}\|\phi\|^{2}_{D^{s,2}(\mathbb{R}^{N})} \leqslant C\|\phi\|_{E}^{2}. \end{aligned}

Then

\begin{aligned} | \langle I'(u_{n}),\phi \rangle | (1+\|u_{n}\|_{E}) =& | \langle J'(z_{n},v_{n}),(0,\tilde{\phi}) \rangle | (1+\|(z_{n},v_{n})\|_{\mathbb{R}\times E})\\ \leqslant& \|J'(z_{n},v_{n})\|_{(\mathbb{R}\times E)^{-1}} \|(0,\tilde{\phi})\|_{\mathbb{R}\times E} (1+\|(z_{n},v_{n})\|_{\mathbb{R}\times E})\\ <&\frac{8}{n}C\|\phi\|_{E}^{2}. \end{aligned}
Lemma

Lemma 5.8

K3VC

Let \{u_{n}\} be a sequence of c>0 as follows:

\begin{aligned} I(u_{n})\to c,~~ P(u_{n})\to 0,~~\mathrm{and}~~ \|I'(u_{n})\|_{E^{-1}}(1+\|u_{n}\|_{E})\to 0,\ \mathrm{as}\ n\rightarrow \infty. \end{aligned}

Then it is a bounded sequence in E. If u_{n}\rightharpoonup u in E, then \|I'(u)\|_{E^{-1}}=0 and P(u)=0. Moreover, if u\not\equiv0, then I(u)=c.

Used by Proof
Proof of Lemma 5.8

From \|I'(u_{n})\|_{E^{-1}}(1+\|u_{n}\|_{E})\to 0, we have \|I'(u_{n})\|_{E^{-1}}\to 0 as n\to\infty. Then

\begin{aligned} c+o_{n}(1)=& \frac{1}{2}\|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} +\frac{1}{2}\|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} -\frac{1}{p}\|u_{n}\|_{L^{p}(\mathbb{R}^{N})}^{p}\\ &-\frac{1}{p} \left(\|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} +\|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2} -\|u_{n}\|_{L^{p}(\mathbb{R}^{N})}^{p}\right)\\ =& \frac{p-2}{2p}\|u_{n}\|_{E}^{2}. \end{aligned}

This shows that \{u_{n}\} is bounded in E. Up to a subsequence, there exists u such that

\begin{aligned} u_{n}\rightharpoonup u \; \mathrm{in} ~ E, \;\; u_{n}\rightarrow u ~ \mathrm{a.e. ~in} ~ \mathbb{R}^{N}. \end{aligned}

From \|I'(u_{n})\|_{E^{-1}}\to 0, we have \|I'(u)\|_{E^{-1}}=0.

For u\equiv0, we have P(u)=0.

From now on, we suppose u\not\equiv0. Firstly, we show P(u)\geqslant0. By the way of contradiction, we suppose that P(u)<0. In terms of Lemma 5.2, there exists a unique \bar{t}>0 such that

\begin{aligned} u\left(\frac{x}{\bar{t}}\right)\in\mathcal{M} ~~\mathrm{and}~~ P\left(u\left(\frac{x}{\bar{t}}\right)\right)=0. \end{aligned}

In view of (Proof) and P(u)<0, we obtain

\begin{aligned} g_{2}'(1)=P(u)<0. \end{aligned}

By the expression of g_{2}'(\cdot), one can choose t_{0}>0 small such that

g_{2}'(t_{0})>0.

Using the intermediate value theorem, we know there exists \bar{\bar{t}}\in(t_{0},1) such that

g_{2}'(\bar{\bar{t}})=0.

By the uniqueness of \bar{t}, one has

\begin{aligned} \bar{t} = \bar{\bar{t}} \in(t_{0},1). \end{aligned}

Making use of (?), there holds

\begin{aligned} c \leqslant& I\left(u\left(\frac{x}{\bar{t}}\right)\right) = I\left(u\left(\frac{x}{\bar{t}}\right)\right) - \frac{1}{N} P\left(u\left(\frac{x}{\bar{t}}\right)\right)\\ =& \frac{1}{N} \bar{t}^{N-2} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{s}{N} \bar{t}^{N-2s} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \\ <& \frac{1}{N} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{s}{N} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2}\\ \leqslant& \frac{1}{N} \lim_{n\rightarrow\infty} \|u_{n}\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{s}{N} \lim_{n\rightarrow\infty} \|u_{n}\|_{D^{s,2}(\mathbb{R}^{N})}^{2}\\ =& \lim_{n\rightarrow\infty} \bigg[ I(u_{n}) - \frac{1}{N} P(u_{n}) \bigg]\\ =& \lim_{n\rightarrow\infty} I(u_{n}) = c. \end{aligned}

This is a contradiction. Hence, P(u)\geqslant0.

We now prove that P(u)=0. Suppose that P(u)>0. From u\not\equiv0, we know

\begin{aligned} I(u) - \frac{1}{N} P(u)>0. \end{aligned}

Set v_{n}=u_{n}-u. According to Brézis-Lieb lemma [Brezis-Lieb1983PAMS], we know

c = \lim_{n\rightarrow\infty} I(u_{n})=\lim_{n\rightarrow\infty} I(v_{n})+I(u)

and

0=\lim_{n\rightarrow\infty}P(u_{n})=\lim_{n\rightarrow\infty}P(v_{n})+P(u),

which further implies that:

\begin{aligned} c=& \lim_{n\rightarrow\infty} I(u_{n})- \frac{1}{N} \lim_{n\rightarrow\infty}P(u_{n})\\ =& \lim_{n\rightarrow\infty} \left[ I(v_{n}) - \frac{1}{N} P(v_{n}) \right] + \left[ I(u)-\frac{1}{N}P(u) \right] \end{aligned}

and

\lim_{n\rightarrow\infty} P(v_{n})=-P(u)<0.

Similar to (?), there exists a unique sequence \bar{t}_{n} such that

v_{n}\left(\frac{x}{\bar{t}_{n}}\right)\in \mathcal{M} ~~\mathrm{and}~~ P\left(v_{n}\left(\frac{x}{\bar{t}_{n}}\right)\right)=0 \ \, \mathrm{and} \ \, \bar{t}_{n}\in(0,1].

From (?)-(?) and \bar{t}_{n}\in(0,1], we obtain

\begin{aligned} c \leqslant& \lim_{n\rightarrow\infty} I\left(v_{n}\left(\frac{x}{\bar{t}_{n}}\right)\right)\\ =& \lim_{n\rightarrow\infty} \left[ I\left(v_{n}\left(\frac{x}{\bar{t}_{n}}\right)\right) - \frac{1}{N} P\left(v_{n}\left(\frac{x}{\bar{t}_{n}}\right)\right) \right]\\ <& \lim_{n\rightarrow\infty} \left[ I(v_{n}) - \frac{1}{N} P(v_{n}) \right]\\ <& \lim_{n\rightarrow\infty} \left[ I(v_{n}) - \frac{1}{N} P(v_{n}) \right] + \left[ I(u)-\frac{1}{N}P(u) \right]\\ =& \lim_{n\rightarrow\infty} I(u_{n}) - \frac{1}{N} \lim_{n\rightarrow\infty} P(u_{n}) = c. \end{aligned}

This is a contradiction. Hence, P(u)=0.

Moreover, if u\not\equiv0, then

\begin{aligned} c \leqslant I(u) =I(u) -\frac{1}{N}P(u) \leqslant \lim_{n\rightarrow\infty}I(u_{n}) -\frac{1}{N} \lim_{n\rightarrow\infty} P(u_{n}) =c, \end{aligned}

which implies I(u)=c.

Lemma

Lemma 6.1

0BSC

Let u be a weak solution of equation (?) with P(u)=0 and I(u)\geqslant c. Then

\begin{aligned} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2N-p(N-2)} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} = \frac{p(N-2s)-2N}{2p(1-s)} \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}, \end{aligned}

and

\begin{aligned} I(u)=\frac{p-2}{2p} \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}. \end{aligned}

Moreover, let \phi be a weak solution of equation (?) with P(\phi)=0 and I(\phi)=c. Then

\begin{aligned} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{p} \leqslant \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}. \end{aligned}
Used by Proof, Proof
Proof of Lemma 6.1

It follows from \langle I'(u),u\rangle=0 and P(u)=0 that

\begin{aligned} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} - \|u\|_{L^{p}(\mathbb{R}^{N})}^{p} =0 \end{aligned}

and

\begin{aligned} \frac{N-2}{2} \|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \frac{N-2s}{2} \|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =\frac{N}{p}\|u\|_{L^{p}(\mathbb{R}^{N})}^{p}. \end{aligned}

Then

\begin{aligned} \left( \frac{N-2}{2} -\frac{N}{p} \right) \|u\|^{2}_{D^{1,2}(\mathbb{R}^{N})} + \left( \frac{N-2s}{2} -\frac{N}{p} \right) \|u\|^{2}_{D^{s,2}(\mathbb{R}^{N})} =0. \end{aligned}

This is the first equation of (6.1). Putting this result back, we get the second equation of (6.1). Furthermore, one deduces

\begin{aligned} I(u) =& \frac{1}{2} \|u\|_{D^{1,2}(\mathbb{R}^{N})}^{2} + \frac{1}{2} \|u\|_{D^{s,2}(\mathbb{R}^{N})}^{2} - \frac{1}{p} \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}\\ =& \frac{p-2}{2p} \int_{\mathbb{R}^{N}} |u|^{p} \mathrm{d}x. \end{aligned}

Then

\begin{aligned} \frac{p-2}{2p} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{p} = I(\phi)=c \leqslant I(u) = \frac{p-2}{2p} \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}, \end{aligned}

which gives

\begin{aligned} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{p} \leqslant \|u\|_{L^{p}(\mathbb{R}^{N})}^{p}. \end{aligned}
Lemma

Lemma 6.2

QUFQ

Let \phi be a weak solution of equation (?) with P(\phi)=0 and I(\phi)=c. Then

\begin{aligned} C_{N,p,s}^{-1}\leqslant \left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}. \end{aligned}
Used by Proof
Proof of Lemma 6.2

It follows from (6.1) that

\begin{aligned} C_{N,p,s}^{-1}\leqslant& W(\phi)\\ =& \frac {\left( \|\phi\|_{D^{s,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \|\phi\|_{D^{1,2}(\mathbb{R}^{N})}^{2} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}}} {\int_{\mathbb{R}^{N}} |\phi|^{p} \mathrm{d}x}\\ =&\left( \frac{2N-p(N-2)}{2p(1-s)} \right) ^{\frac{2N-p(N-2)}{4(1-s)}} \left( \frac{p(N-2s)-2N}{2p(1-s)} \right) ^{\frac{p(N-2s)-2N}{4(1-s)}} \|\phi\|_{L^{p}(\mathbb{R}^{N})}^{\frac{p(p-2)}{p}}. \end{aligned}

Source and licence

The text on this page is from arXiv:2604.05177v1 by Hichem Hajaiej and Yu Su, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2604/2604.05177.jsonl.