Appendix A
First we consider the case when \(n=1\). We review the definition of CR normal coordinates and recall some of its properties. Give any \(x\in M\), we can find a contact form \({\widehat{\theta }}_x=\varphi _x^{\frac{2}{n}}\theta _0\) conformal to \(\theta _0\), where \({\widehat{\theta }}_x\) is a contact form defined in (z, t) which is the CR normal coordinates centered at x. On the other hand, \({\widehat{\theta }}_x\) satisfies the following properties: (see Theorem 3.7 in P.172 of [15] and Proposition 6.5 in [11])
$$\begin{aligned} {\widehat{\theta }}_x= & {} \Bigg (1+O(\rho _x(y))\Bigg )\theta _{{\mathbb {H}}^n},\nonumber \\ dV_{{\widehat{\theta }}_x}= & {} \Bigg (1+O(\rho _x(y))\Bigg )dV_{\theta _{{\mathbb {H}}^n}},\nonumber \\ W_{k}= & {} \Bigg (1+O(\rho _x(y)^4)\Bigg )Z_k+O(\rho _x(y)^4){\overline{Z}}_k+O(\rho _x(y)^5)\frac{\partial }{\partial t}\quad \text{ for } 1\le k\le n,\nonumber \\ \end{aligned}$$
(A.1)
in \(\{(z,t):(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}\) for some \({\widehat{\rho }}>0\). Here (z, t) represents the point \(y\in M\) in the CR normal coordinates centered at x, \(\rho _x(y)\) is the distance in the CR normal coordinates at x, which implies that
$$\begin{aligned} \rho _{x_k}(y)=(t^2+|z|^4)^{\frac{1}{4}}. \end{aligned}$$
(A.2)
Also, \(\theta _{{\mathbb {H}}^n}=dt + \sqrt{-1}\sum \nolimits _{j=1}^n(z_jd{\overline{z}}_j-{\overline{z}}_jdz_j)\) is the standard contact form of the Heisenberg group \({\mathbb {H}}^n=\{(z,t)=(z_1,\ldots ,z_n,t)\in {\mathbb {C}}^n\times {\mathbb {R}}\}\). Moreover, \(\{W_{k}\}\) is a pseudo-Hermitian frame, i.e. \(\{W_{k}\}\) is a local frame of \({\widehat{\theta }}_x\) satisfying \(-\sqrt{-1}d{\widehat{\theta }}_x(W_{k},{\overline{W}}_{l})=\delta _{kl}\) (see P.165 in [15]), and \(Z_k=\frac{\partial }{\partial z_k}+\sqrt{-1}{\overline{z}}_k\frac{\partial }{\partial t}\). We also have the following expression for the CR conformal sub-Laplacian: (see P.114 in [18])
$$\begin{aligned} -\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_x}+R_{{\widehat{\theta }}_x}=-\left( 2+\frac{2}{n}\right) \Delta _{\theta _{{\mathbb {H}}^n}}+O(\rho _x(y)^2). \end{aligned}$$
(A.3)
On the other hand, the Webster scalar curvature of \({\widehat{\theta }}_x\) satisfies (see P.38 in [11])
$$\begin{aligned} |R_{{\widehat{\theta }}_x}(y)|\le C\rho _x(y)^2, \end{aligned}$$
(A.4)
where \(\rho _x(y)\) is the distance in the CR normal coordinates at x. Let \(G_x\) be the Green’s function with pole at x. Then we have (see (105) in [11])
$$\begin{aligned} -\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_x}G_x(y)+R_{{\widehat{\theta }}_x} G_x(y)=0 \quad \text{ for } y\ne x. \end{aligned}$$
(A.5)
Moreover, the Green’s function satisfies the estimates (see Propositions 5.2 and 5.3 in [11])
$$\begin{aligned} |G_x(y)-\rho _x(y)^{-2n}-A_x|\le C\rho _x(y) \end{aligned}$$
(A.6)
and
$$\begin{aligned} |\nabla _{{\widehat{\theta }}_x}(G_x(y)-\rho _x(y)^{-2n})|_{{\widehat{\theta }}_x}\le C, \end{aligned}$$
(A.7)
where \(A_x\) is the CR mass. Note that \(\rho _x(y)\), the distance in the CR normal coordinates at x, and d(x, y), the Carnot–Carathéodory distance on M with respect to the contact form \(\theta _0\) are equivalent, i.e. there exists a uniform constant \(C_0\) such that
$$\begin{aligned} \frac{1}{C_0}d(x,y)\le \rho _x(y)\le C_0 d(x,y) \quad \text{ for } \text{ all } x, y\in M. \end{aligned}$$
(A.8)
When M is spherical, give any \(x\in M\), we can find a smooth function \(\varphi _x\) such that \(\theta _{{\mathbb {H}}^n}=\varphi _x^{\frac{2}{n}}\theta _0\) in a neighborhood of x. That is,
$$\begin{aligned} {\widehat{\theta }}_x=\theta _{{\mathbb {H}}^n} \end{aligned}$$
(A.9)
in the above notation. Note also that \(R_{{\widehat{\theta }}_x}=R_{\theta _{{\mathbb {H}}^n}}\equiv 0\) in this case. Therefore, (A.1) and (A.4) are still true. On the other hand, the Green’s function \(G_x(y)\) defined as in (A.5) satisfies (A.6) and (A.7) by Lemma 5.1 in [10].
Under the assumptions of Theorem 1.1, the CR mass is positive, i.e. \(A_x>0\) for all \(x\in M\) by the CR positive mass theorem (see Theorem 1.1 in [11] and Corollary C in [10]). Note that the function \(x\mapsto A_x\) is continuous. See Proposition 3.3 in [37] for the proof when M is spherical. See also Remark I.1.2 in [21] and Proposition 3.5 in [20] for the proof of the corresponding statement in the Riemannian case. Hence, we have
$$\begin{aligned} \inf _{x\in M}A_x>0. \end{aligned}$$
Suppose that we are given a set of pairs \((x_k,\varepsilon _k)_{1\le k\le m}\). For every \(1\le k\le m\), we define \({\overline{u}}_{(x_k,\varepsilon _k)}\) by
$$\begin{aligned} {\overline{u}}_{(x_k,\varepsilon _k)}(y)=\varphi _{x_k}(y)\,{\overline{U}}_{(x_k,\varepsilon _k)}(y). \end{aligned}$$
(A.10)
Here
$$\begin{aligned} {\overline{U}}_{(x_k,\varepsilon _k)}(y)= & {} \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n \left[ \frac{\chi _\delta (\rho _{x_k}(y))}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}} \right. \nonumber \\&\left. +\, (1-\chi _\delta (\rho _{x_k}(y))v)G_{x_k}(y)\right] , \end{aligned}$$
(A.11)
where \(\chi _\delta (s)=\chi (\frac{s}{\delta })\) and \(\chi :{\mathbb {R}}\rightarrow [0,1]\) is a cut-off function satisfying \(\chi (s)=1\) for \(s\le 1\) and \(\chi (s)=0\) for \(s\ge 2\). On the other hand, \(\delta \) is a positive real number such that \(\varepsilon _k\ll \delta \) for all \(1\le k\le m\).
Proposition A.1
For \(n=1\), we have
$$\begin{aligned}&\left| \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) -R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) +r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}(y)^{1+\frac{2}{n}}\right. \\&\qquad \left. +\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)) \right| \\&\quad \le C\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n}{2}}\rho _{x_k}(y)^2\,1_{\{\rho _{x_k}(y)\le 2\delta \}} +C\frac{\varepsilon _k^n}{\delta }\,1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}\\&\qquad +\,C\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}}1_{\{\rho _{x_k}(y)\ge \delta \}}. \end{aligned}$$
Proof
By definition of \({\overline{U}}_{(x_k,\varepsilon _k)}\), we have
$$\begin{aligned}&\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) -R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) +r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}(y)^{1+\frac{2}{n}}\\&\qquad +\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\cdot \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)) \\&\quad =\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n\,\Bigg (I_1+I_2+I_3+I_4+I_5+I_6+I_7+I_8\Bigg ). \end{aligned}$$
Here,
$$\begin{aligned} I_1= & {} \chi _\delta (\rho _{x_k}(y))\left[ \phantom {\left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}} \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right. \\&\left. +\,(2n+2)n\varepsilon _k^2 \left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}\right] ,\\ I_2= & {} -\chi _\delta (\rho _{x_k}(y))R_{{\widehat{\theta }}_{x_k}}\frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}},\\ I_3= & {} -\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))\Bigg ( G_{x_k}(y)-\rho _{x_k}(y)^{-2n}-A_{x_k}\Bigg ),\\ I_4= & {} \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))\Bigg ( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\rho _{x_k}(y)^{-2n}\Bigg ),\\ I_5= & {} -2\left( 2+\frac{2}{n}\right) \langle \nabla _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)),\nabla _{{\widehat{\theta }}_{x_k}}(G_x(y)-\rho _x(y)^{-2n})\rangle _{{\widehat{\theta }}_{x_k}},\\ I_6= & {} 2\left( 2+\frac{2}{n}\right) \Bigg \langle \nabla _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)),\nabla _{{\widehat{\theta }}_{x_k}} \Bigg ( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\rho _{x_k}(y)^{-2n}\Bigg )\Big \rangle _{{\widehat{\theta }}_{x_k}},\\ I_7= & {} (2n+2)n\varepsilon _k^2\left[ \left( \frac{\chi _\delta (\rho _{x_k}(y))}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}+ \Bigg (1-\chi _\delta (\rho _{x_k}(y))\Bigg )G_{x_k}(y)\right) ^{1+\frac{2}{n}}\right. \\&\left. -\,\chi _\delta (\rho _{x_k}(y))\left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}\right] ,\\ I_8= & {} 7\Bigg (1-\chi _\delta (\rho _{x_k}(y)\Bigg )\left[ \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}G_{x_k}(y)-R_{{\widehat{\theta }}_{x_k}} G_{x_k}(y)\right] . \end{aligned}$$
Since
$$\begin{aligned} \begin{aligned} \Delta _{\theta _{{\mathbb {H}}^n}}\left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right)&=-\frac{n^2\varepsilon _k^2}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{1+\frac{n}{2}}} \end{aligned} \end{aligned}$$
(A.12)
we have
$$\begin{aligned} |I_1|\le & {} \chi _\delta (\rho _{x_k}(y))\left| \left( 2+\frac{2}{n}\right) (\Delta _{\theta _{{\mathbb {H}}^n}}+O(\rho _{x_k}(y)^2))\left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) \right. \\&\left. +\,(2n+2)n\varepsilon _k^2 \left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}\right| \\\le & {} 1_{\{\rho _{x_k}(y)\le 2\delta \}}\frac{C\rho _{x_k}(y)^2}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}} \end{aligned}$$
by (A.1) and (A.2). By (A.4), we have
$$\begin{aligned} |I_2|\le C\rho _{x_k}(y)^2 \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\,1_{\{\rho _{x_k}(y)\le 2\delta \}}. \end{aligned}$$
By (A.6), we have
$$\begin{aligned} |I_3|\le & {} \left( 2+\frac{2}{n}\right) |\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))|\cdot \Bigg | G_{x_k}(y)-\rho _{x_k}(y)^{-2n}-A_{x_k}\Bigg |\\\le & {} C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}} }{\delta ^2}\rho _{x_k}(y) \le \frac{C}{\delta }1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}. \end{aligned}$$
Note that
$$\begin{aligned} |I_4|= & {} \left( 2+\frac{2}{n}\right) |\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))|\cdot \left| \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\rho _{x_k}(y)^{-2n}\right| \\\le & {} C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}}{\delta ^2} \cdot \left| \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\frac{1}{(t^2+|z|^4)^{\frac{n}{2}}}\right| \\\le & {} C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}}{\delta ^2}\cdot \frac{\varepsilon _k^{2n}}{\delta ^{4n}}\\\le & {} \frac{C}{\delta }1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}} \end{aligned}$$
by (A.2) and the assumption that \(\varepsilon _k\ll \delta \). Note also that
$$\begin{aligned} |I_6|\le & {} C|\nabla _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))|_{{\widehat{\theta }}_{x_k}}\cdot \left| \nabla _{{\widehat{\theta }}_{x_k}}\Bigg ( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\rho _{x_k}(y)^{-2n}\Bigg )\right| _{{\widehat{\theta }}_{x_k}}\\\le & {} C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}}{\delta } \cdot |z|^2 \frac{\varepsilon _k^2}{(t^2+|z|^4)^{\frac{n}{2}+1}}\\\le & {} C\frac{\varepsilon _k^2}{\delta ^{2n+3}}1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}} \le C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}}{\delta } \end{aligned}$$
by (A.1), (A.2), and the assumption that \(\varepsilon _k\ll \delta \). By (A.7), we have
$$\begin{aligned} |I_5|\le C|\nabla _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))|_{{\widehat{\theta }}_{x_k}}\Bigg | \nabla _{{\widehat{\theta }}_{x_k}}( G_{x_k}(y)-\rho _{x_k}(y)^{-2n})\Bigg |_{{\widehat{\theta }}_{x_k}}\le C\frac{1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}}{\delta }. \end{aligned}$$
We also have
$$\begin{aligned} |I_7|= & {} (2n+2)n\varepsilon _k^2\left| \left( \frac{\chi _\delta (\rho _{x_k}(y))}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}+ \Bigg (1-\chi _\delta (\rho _{x_k}(y))\Bigg )G_{x_k}(y)\right) ^{1+\frac{2}{n}}\right. \\&\left. -\,\chi _\delta (\rho _{x_k}(y))\left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}\right| \\\le & {} C\varepsilon _k^2\left( \left( \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}\right) ^{1+\frac{2}{n}}+ G_{x_k}(y)^{1+\frac{2}{n}}\right) 1_{\{\rho _{x_k}(y)\ge \delta \}}\\\le & {} C\frac{\varepsilon _k^2}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{1+\frac{2}{n}}}1_{\{\rho _{x_k}(y)\ge \delta \}} \end{aligned}$$
by (A.6), and
$$\begin{aligned}I_8=0 \end{aligned}$$
by (A.5). From this, the assertion follows. \(\square \)
Proposition A.2
When M is spherical, we have
$$\begin{aligned}&\left| \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) -R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) +r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}(y)^{1+\frac{2}{n}}\right. \\&\qquad \left. +\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)) \right| \\&\quad \le C\frac{\varepsilon _k^n}{\delta }\,1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}} +C\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}}1_{\{\rho _{x_k}(y)\ge \delta \}}. \end{aligned}$$
Proof
The proof is the same as the proof of Proposition A.1, except we need to prove that \(I_1=0\). But this follows from (A.9) and (A.12). \(\square \)
Corollary A.3
We have
$$\begin{aligned}&\left| \left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) -R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}(y) +r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}(y)^{1+\frac{2}{n}}\right| \\&\quad \le C\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n}{2}}\rho _{x_k}(y)^2\,1_{\{\rho _{x_k}(y)\le 2\delta \}} +C\frac{\varepsilon _k^n}{\delta ^2}\,1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}\\&\qquad +\,C\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}}1_{\{\rho _{x_k}(y)\ge \delta \}}. \end{aligned}$$
Proof
Combining Propositions A.1, A.2 and the following estimate:
$$\begin{aligned} \left| \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\left( 2+\frac{2}{n}\right) \Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))\right| \le C\frac{\varepsilon _k^n}{\delta ^2}\,1_{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}, \end{aligned}$$
Corollary A.3 follows. \(\square \)
Proposition A.4
If \(\delta \) is sufficiently small, then
$$\begin{aligned} E({\overline{u}}_{(x_k,\varepsilon _k)})\le Y(S^{2n+1})-cA_{x_k}\varepsilon _k^{2n} +C\delta ^2\varepsilon _k^{2n}+C\delta \varepsilon _k^{2n}+C\delta ^{-2n-2}\varepsilon _k^{2n+2} \end{aligned}$$
for some \(c>0\).
Proof
When \(n=1\), it follows from Proposition A.1 and integration by parts that
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2-r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}\Bigg )dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad =-\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )\Delta _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)} -R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}+r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}^{1+\frac{2}{n}}\Bigg ) {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\left( 2+\frac{2}{n}\right) \int _M\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)){\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\qquad +\, C\int _{\{\rho _{x_k}(y)\le 2\delta \}}\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n}{2}}\rho _{x_k}(y)^2 {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\qquad +\,C\frac{\varepsilon _k^n}{\delta }\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}{\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\qquad +\,C\int _{\{\rho _{x_k}(y)\ge \delta \}}\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}} {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}. \end{aligned}$$
(A.13)
We are going to estimate each of the terms on the right hand side of (A.13). Since
$$\begin{aligned}&\left| \int _M\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)) \Bigg ({\overline{U}}_{(x_k,\varepsilon _k)}-\rho _{x_k}(y)^{-2n}\Bigg )dV_{{\widehat{\theta }}_{x_k}}\right| \\&\quad \le \frac{C}{\delta ^2}\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}} \left| \frac{1}{(t^2+(\varepsilon _k^2+|z|^2)^2)^{\frac{n}{2}}}-\frac{1}{(t^2+|z|^4)^{\frac{n}{2}}}\right| dV_{{\widehat{\theta }}_{x_k}}\\&\qquad +\, \frac{C}{\delta ^2}\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}|G_{x_k}(y)-\rho _{x_k}(y)^{-2n}|dV_{{\widehat{\theta }}_{x_k}}\\&\quad \le \frac{C}{\delta ^2}\int _\delta ^{2\delta }\frac{\varepsilon _k^{2n}r^{2n-1}dr}{r^{4n}} \le C\delta ^{-2n-2}\varepsilon _k^{2n} \end{aligned}$$
by (A.6) and (A.2), we have
$$\begin{aligned}&\int _M\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)){\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad =-\int _M\Bigg \langle \nabla _{{\widehat{\theta }}_{x_k}}\rho _{x_k}(y)^{-2n}, \nabla _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y))\Bigg \rangle _{{\widehat{\theta }}_{x_k}} dV_{{\widehat{\theta }}_{x_k}} +C\delta ^{-2n-2}\varepsilon _k^{2n}\nonumber \\&\quad =2n\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}\chi '_\delta (\rho _{x_k}(y))\rho _{x_k}(y)^{-2n-1}|\nabla _{{\widehat{\theta }}_{x_k}}\rho _{x_k}(y)|^2_{{\widehat{\theta }}_{x_k}} dV_{{\widehat{\theta }}_{x_k}} +C\delta ^{-2n-2}\varepsilon _k^{2n}\nonumber \\&\quad =-\frac{C}{\delta }\int _\delta ^{2\delta } dr +C\delta ^{-2n-2}\varepsilon _k^{2n}=-C_0 +C\delta ^{-2n-2}\varepsilon _k^{2n} \end{aligned}$$
(A.14)
by integration by parts. Here \(C_0\) is a positive constant depending only on n. On the other hand, it follows form the definition of \({\overline{U}}_{(x_k,\varepsilon _k)}\) that
$$\begin{aligned}&C\int _{\{\rho _{x_k}(y)\le 2\delta \}}\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n}{2}}\rho _{x_k}(y)^2 {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le C\varepsilon _k^{2n}\int _{\{\rho _{x_k}(y)\le 2\delta \}}\frac{\rho _{x_k}(y)^2}{(t^2+|z|^4)^{n}} dV_{{\widehat{\theta }}_{x_k}}=C\delta ^{4-2n}\varepsilon _k^{2n}, \end{aligned}$$
(A.15)
where we have used the assumption that \(n=1\). Note also that
$$\begin{aligned}&C\frac{\varepsilon _k^n}{\delta }\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}{\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le C\frac{\varepsilon _k^{2n}}{\delta } \int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}\Bigg (\rho _{x_k}(y)^{-2n}+A_{x_k}+C\rho _{x_k}(y)\Bigg )dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le C\frac{\varepsilon _k^{2n}}{\delta }\int _\delta ^{2\delta }\frac{r^{2n+1}dr}{r^{2n}}=C\delta \varepsilon _k^{2n}, \end{aligned}$$
(A.16)
and
$$\begin{aligned}&C\int _{\{\rho _{x_k}(y)\ge \delta \}}\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}} {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le C\varepsilon _k^{2n+2}\int _\delta ^\infty \frac{r^{2n+1}dr}{r^{4(\frac{n+2}{2})}\cdot r^{2n}} =C\varepsilon _k^{2n+2}\int _\delta ^\infty \frac{dr}{r^{2n+3}} =C\delta ^{-2n-2}\varepsilon _k^{2n+2},\qquad \qquad \end{aligned}$$
(A.17)
where we have used (A.6) and the definition of \({\overline{U}}_{(x_k,\varepsilon _k)}\). Combining (A.13)–(A.17), we have
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le r_\infty \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}} -\left( \frac{n(2n+2)}{r_\infty }\right) ^{n} \varepsilon _k^{2n} A_{x_k}\left( 4+\frac{4}{n}\right) C_0\nonumber \\&\qquad +\,C\delta ^2\varepsilon _k^{2n}+C\delta \varepsilon _k^{2n}+C\delta ^{-2n-2}\varepsilon _k^{2n+2}\nonumber \\&\quad \le r_\infty \left( \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\right) ^{\frac{n}{n+1}} \left( \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\right) ^{\frac{1}{n+1}} \nonumber \\&\qquad -\left( \frac{n(2n+2)}{r_\infty }\right) ^{n} \varepsilon _k^{2n} A_{x_k}\left( 2+\frac{2}{n}\right) C_0 +C\delta ^2\varepsilon _k^{2n}+C\delta \varepsilon _k^{2n}+C\delta ^{-2n-2}\varepsilon _k^{2n+2}\nonumber \\ \end{aligned}$$
(A.18)
where the last inequality follows from Hölder’s inequality. It follows from the definition of \({\overline{U}}_{(x_k,\varepsilon _k)}\) that
$$\begin{aligned} \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\le & {} \left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1}\left( \int _M\chi _\delta (\rho _{x_k}(y)) \left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{n+1}dV_{{\widehat{\theta }}_{x_k}}\right. \nonumber \\&\left. +\,C \int _M(1-\chi _\delta (\rho _{x_k}(y)))\varepsilon _k^{2n+2}G_{x_k}(y)^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\right) . \end{aligned}$$
(A.19)
Note that
$$\begin{aligned}&\int _M(1-\chi _\delta (\rho _{x_k}(y)))\varepsilon _k^{2n+2}G_{x_k}(y)^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le \varepsilon _k^{2n+2}\int _{\rho _{x_k}(y)\ge \delta }\Bigg (\rho _{x_k}(y)^{-2n}+A_{x_k}+C\rho _{x_k}(y)\Bigg )^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad =C\varepsilon _k^{2n+2}\int _\delta ^\infty \frac{r^{2n+1}dr}{r^{2n\left( 2+\frac{2}{n}\right) }} =C\varepsilon _k^{2n+2}\int _\delta ^\infty \frac{dr}{r^{2n+3}}=C\delta ^{-2n-2}\varepsilon _k^{2n+2}\qquad \qquad \end{aligned}$$
(A.20)
by (A.6). Note also that
$$\begin{aligned}&\int _M\chi _\delta (\rho _{x_k}(y)) \left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{n+1}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad =\int _{{\mathbb {H}}^n}\left( \frac{1}{(t^2+(1+|z|^2)^2)^{\frac{n}{2}}}\right) ^{2+\frac{2}{n}}dV_{\theta _{{\mathbb {H}}^n}} +O(\varepsilon _k^{2n+2})\nonumber \\&\quad =\left( \frac{Y(S^{2n+1})}{n(2n+2)}\right) ^{n+1}+O(\varepsilon _k^{2n+2}), \end{aligned}$$
(A.21)
where we have used the change of variables \((\frac{t}{\varepsilon _k^2},\frac{z}{\varepsilon _k})\mapsto (t,z)\). Combining (A.19)–(A.21), we get
$$\begin{aligned} \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}} \le \left( \frac{Y(S^{2n+1})}{r_\infty }\right) ^{n+1}+C\delta ^{-2n-2}\varepsilon _k^{2n+2}. \end{aligned}$$
Substituting this into (A.18), we obtain
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{{\widehat{\theta }}_{x_k}}\\&\quad \le Y(S^{2n+1}) \left( \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}}\right) ^{\frac{n}{n+1}} -\left( \frac{n(2n+2)}{r_\infty }\right) ^{n} \varepsilon _k^{2n} A_{x_k}\left( 2+\frac{2}{n}\right) C_0 \\&\qquad +\,C\delta ^2\varepsilon _k^{2n}+C\delta \varepsilon _k^{2n}+C\delta ^{-2n-2}\varepsilon _k^{2n+2}. \end{aligned}$$
This proves Proposition A.4 for the case when \(n=1\).
When M is spherical, it follows from Proposition A.2 and integration by parts that
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2-r_\infty {\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}\Bigg )dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\quad \le \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _k^n A_{x_k}\left( 2+\frac{2}{n}\right) \int _M\Delta _{{\widehat{\theta }}_{x_k}}\chi _\delta (\rho _{x_k}(y)){\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}} \nonumber \\&\qquad +\,C\frac{\varepsilon _k^n}{\delta }\int _{\{\delta \le \rho _{x_k}(y)\le 2\delta \}}{\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}\nonumber \\&\qquad +\,C\int _{\{\rho _{x_k}(y)\ge \delta \}}\left( \frac{\varepsilon _k^2}{t^2+(\varepsilon _k^2+|z|^2)^2}\right) ^{\frac{n+2}{2}} {\overline{U}}_{(x_k,\varepsilon _k)}dV_{{\widehat{\theta }}_{x_k}}. \end{aligned}$$
(A.22)
Combining (A.14), (A.16), (A.17), and (A.22), we also obtain (A.18). Now we can follow the same argument as above to finish the proof. \(\square \)
Lemma A.5
We have
$$\begin{aligned} F({\overline{u}}_{(x_{k},\varepsilon _{k})})=r_\infty +o(1). \end{aligned}$$
Proof
It follows from (A.18) that
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{{\widehat{\theta }}_{x_k}}\\&\quad \le r_\infty \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}} -\left( \frac{n(2n+2)}{r_\infty }\right) ^{n} \varepsilon _k^{2n} A_{x_k}\left( 2+\frac{2}{n}\right) C_0\\&\qquad +\,C\delta ^2\varepsilon _k^{2n}+C\delta \varepsilon _k^{2n}+C\delta ^{-2n-2}\varepsilon _k^{2n+2}, \end{aligned}$$
which implies that
$$\begin{aligned} \int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{\theta _0}{\overline{u}}_{(x_k,\varepsilon _k)}|^2_{\theta _0} +R_{\theta _0}{\overline{u}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{\theta _0}\le r_\infty \int _M{\overline{u}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{\theta _0} +o(1). \end{aligned}$$
This implies that \(F({\overline{u}}_{(x_{k},\varepsilon _{k})})\le r_\infty +o(1)\). On the other hand, following the proof of Proposition A.4, one can prove that
$$\begin{aligned}&\int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}|^2_{{\widehat{\theta }}_{x_k}} +R_{{\widehat{\theta }}_{x_k}}{\overline{U}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{{\widehat{\theta }}_{x_k}}\\&\quad \ge r_\infty \int _M{\overline{U}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{{\widehat{\theta }}_{x_k}} -\left( \frac{n(2n+2)}{r_\infty }\right) ^{n} \varepsilon _k^{2n} A_{x_k}\left( 2+\frac{2}{n}\right) C_0\\&\qquad -\,C\delta ^2\varepsilon _k^{2n}-C\delta \varepsilon _k^{2n}-C\delta ^{-2n-2}\varepsilon _k^{2n+2}, \end{aligned}$$
by using the same arguments to obtain (A.18). This implies that
$$\begin{aligned} \int _M\Bigg (\Bigg (2+\frac{2}{n}\Bigg )|\nabla _{\theta _0}{\overline{u}}_{(x_k,\varepsilon _k)}|^2_{\theta _0} +R_{\theta _0}{\overline{u}}_{(x_k,\varepsilon _k)}^2\Bigg )dV_{\theta _0}\ge r_\infty \int _M{\overline{u}}_{(x_k,\varepsilon _k)}^{2+\frac{2}{n}}dV_{\theta _0} +o(1), \end{aligned}$$
which gives \(F({\overline{u}}_{(x_{k},\varepsilon _{k})})\ge r_\infty +o(1)\). This proves the assertion. \(\square \)
Lemma A.6
We have the estimate
$$\begin{aligned} \int _M{\overline{u}}_{(x_i,\varepsilon _i)}{\overline{u}}_{(x_j,\varepsilon _j)}^{1+\frac{2}{n}}dV_{\theta _0} \le C\left( \frac{\varepsilon _i^2\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}}. \end{aligned}$$
Proof
It follows from the definition of \({\overline{u}}_{(x_i,\varepsilon _i)}\) that \({\overline{u}}_{(x_i,\varepsilon _i)}(y)=\varphi _{x_i}(y){\overline{U}}_{(x_i,\varepsilon _i)}(y)\) and
$$\begin{aligned} {\overline{U}}_{(x_i,\varepsilon _i)}(y)\le \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} \varepsilon _i^n \left[ \frac{\chi _\delta (\rho _{x_i}(y))}{(\varepsilon _i^4+d(x_i,y)^4)^{\frac{n}{2}}}+ \Bigg (1-\chi _\delta (\rho _{x_i}(y))\Bigg )G_{x_i}(y)\right] . \end{aligned}$$
On the set \(U=:\{2d(x_i,y)\le \varepsilon _j+d(x_i,x_j)\}\), we have
$$\begin{aligned} \varepsilon _j+d(y,x_j)\ge \varepsilon _j+d(x_i,x_j)-d(x_i,y) \ge \frac{1}{2}(\varepsilon _j+d(x_i,x_j)). \end{aligned}$$
(A.23)
Therefore, we have
$$\begin{aligned}&\int _M\left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{1+\frac{n}{2}}dV_{\theta _0}\\&\quad \le \left( \int _{U}+\int _{M{\setminus } U}\right) \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{1+\frac{n}{2}}dV_{\theta _0}\\&\quad \le C\frac{\varepsilon _i^n\varepsilon _j^n}{(\varepsilon _j^4+d(x_i,x_j)^4)^{\frac{n+1}{2}}} \int _0^{\frac{\varepsilon _j+d(x_i,x_j)}{2}}\frac{r^{2n+1}dr}{(\varepsilon _i^4+r^4)^{\frac{n}{2}}}\\&\qquad +\,C\frac{\varepsilon _i^n\varepsilon _j^{n+2}}{(\varepsilon _j^4+d(x_i,x_j)^4)^{\frac{n}{2}}} \int _0^{\infty }\frac{r^{2n+1}dr}{(\varepsilon _j^4+r^4)^{1+\frac{n}{2}}}\\&\quad \le C\left( \frac{\varepsilon _i^2\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}} \end{aligned}$$
because
$$\begin{aligned} \int _0^{\frac{\varepsilon _j+d(x_i,x_j)}{2}}\frac{r^{2n+1}dr}{(\varepsilon _i^4+r^4)^{\frac{n}{2}}}&\le \int _0^{\frac{\varepsilon _j+d(x_i,x_j)}{2}}rdr =\frac{(\varepsilon _j+d(x_i,x_j))^2}{8} \\&\le (\varepsilon _j^4+d(x_i,x_j)^4)^\frac{1}{2} \end{aligned}$$
and
$$\begin{aligned} \begin{aligned} \int _0^{\infty }\frac{r^{2n+1}dr}{(\varepsilon _j^4+r^4)^{1+\frac{n}{2}}}&\le \int _0^{\varepsilon _j}\frac{r^{2n+1}dr}{\varepsilon _j^4r^{4\cdot \frac{n}{2}}} +\int _{\varepsilon _j}^{\infty }\frac{r^{2n+1}dr}{r^{4(1+\frac{n}{2}})}=\frac{C}{\varepsilon _j^2}. \end{aligned} \end{aligned}$$
This proves the assertion. \(\square \)
Lemma A.7
We have the estimate
$$\begin{aligned}&\int _M{\overline{u}}_{(x_i,\varepsilon _i)}\left| \left( 2+\frac{2}{n}\right) \Delta _{\theta _0}{\overline{u}}_{(x_j,\varepsilon _j)} -R_{\theta _0}{\overline{u}}_{(x_j,\varepsilon _j)}+r_\infty {\overline{u}}_{(x_j,\varepsilon _j)}^{1+\frac{2}{n}}\right| dV_{\theta _0}\\&\quad \le C\left( \delta ^4+\delta ^{2n}+\frac{\varepsilon _j^2}{\delta ^2}\right) \left( \frac{\varepsilon _i^2\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}}. \end{aligned}$$
Proof
Recall \({\widehat{\theta }}_x=\varphi _x^{\frac{2}{n}}\theta _0\) for some smooth function \(\varphi _x\) on M. Therefore, we can find a positive constant \(c_0\) such that \(c_0^{-n}\le \varphi _x\le c_0^{n},\) which implies \(\frac{1}{c_0}d(x,y)\le \rho _x(y)\le c_0d(x,y)\) for all \(x, y\in M\). Hence, it follows from Corollary A.3 that
$$\begin{aligned}&\left| \left( 2+\frac{2}{n}\right) \Delta _{\theta _0}{\overline{u}}_{(x_j,\varepsilon _j)} -R_{\theta _0}{\overline{u}}_{(x_j,\varepsilon _j)}+r_\infty {\overline{u}}_{(x_j,\varepsilon _j)}^{1+\frac{2}{n}}\right| \\&\quad \le C(\delta ^2+\delta ^{2n-2})\left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(x_j,y)^4}\right) ^{\frac{n}{2}}1_{\{d(x_j,y)\le 2c_0\delta \}}\\&\qquad +\,C\left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(x_j,y)^4}\right) ^{\frac{n+2}{2}}1_{\left\{ d(x_j,y)\ge \frac{\delta }{c_0}\right\} }. \end{aligned}$$
On the set \(U:=\{2d(x_i,y)\le \varepsilon _j+d(x_i,x_j)\}\cap \{d(y,x_j)\le 2c_0\delta \}\), we have
$$\begin{aligned} d(x_i,y)\le \frac{1}{2}(\varepsilon _j+d(x_i,x_j))\le \varepsilon _j+d(y,x_j)\le 4c_0\delta . \end{aligned}$$
From this, it follows that
$$\begin{aligned}&\int _{\{d(y,x_j)\le 2c_0\delta \}} \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{\frac{n}{2}}dV_{\theta _0}\\&\quad \le \left( \int _{U} + \int _{\{d(y,x_j)\le 2c_0\delta \}{\setminus } U}\right) \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{\frac{n}{2}}dV_{\theta _0}\\&\quad \le C\int _{\{d(x_i,y)\le 4c_0\delta \}} \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}}dV_{\theta _0}\\&\qquad +\,C\int _{\{d(y,x_j)\le 2c_0\delta \}} \left( \frac{\varepsilon _i^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{\frac{n}{2}}dV_{\theta _0}\\&\quad \le C\delta ^2\left( \frac{\varepsilon _i^2\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}}. \end{aligned}$$
Similarly, on the set \(V:=\{2d(x_i,y)\le \varepsilon _j+d(x_i,x_j)\}\cap \{d(y,x_j)\ge \frac{\delta }{c_0}\}\), we have
$$\begin{aligned} \varepsilon _j+d(y,x_j)\ge \varepsilon _j+d(x_i,x_j)-d(x_i,y)\ge \frac{1}{2}(\varepsilon _j+d(x_i,x_j)). \end{aligned}$$
This implies
$$\begin{aligned}&\int _{\left\{ d(y,x_j)\ge \frac{\delta }{c_0}\right\} } \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{\frac{n+2}{2}}dV_{\theta _0}\\&\quad \le \left( \int _{V} + \int _{\left\{ d(y,x_j)\ge \frac{\delta }{c_0}\right\} {\setminus } V}\right) \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \left( \frac{\varepsilon _j^2}{\varepsilon _j^4+d(y,x_j)^4}\right) ^{\frac{n+2}{2}}dV_{\theta _0}\\&\quad \le C\int _{\left\{ 2d(x_i,y)\le \varepsilon _j+d(x_i,x_j)\right\} } \left( \frac{\varepsilon _i^2}{\varepsilon _i^4+d(x_i,y)^4}\right) ^{\frac{n}{2}} \frac{\varepsilon _j^{n+2}}{\delta ^2(\varepsilon _i^4+d(x_i,x_j)^4)^{\frac{n+1}{2}}}dV_{\theta _0}\\&\qquad +\,C\int _{\left\{ d(y,x_j)\ge \frac{\delta }{c_0}\right\} } \left( \frac{\varepsilon _i^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}} \frac{\varepsilon _j^{n+2}}{(\varepsilon _j^4+d(y,x_j)^4)^{1+\frac{n}{2}}}dV_{\theta _0}\\&\quad \le C\frac{\varepsilon _j^2}{\delta ^2}\left( \frac{\varepsilon _i^2\varepsilon _j^2}{\varepsilon _j^4+d(x_i,x_j)^4}\right) ^{\frac{n}{2}} \end{aligned}$$
because
$$\begin{aligned} \begin{aligned} \int _0^{\frac{\varepsilon _j+d(x_i,x_j)}{2}}\frac{r^{2n+1}dr}{(\varepsilon _j^4+r^4)^{\frac{n}{2}}} \le \int _0^{\frac{\varepsilon _j+d(x_i,x_j)}{2}}\frac{r^{2n+1}dr}{r^{4\cdot \frac{n}{2}}} \le (\varepsilon _j^4+d(x_i,x_j)^4)^{\frac{1}{2}} \end{aligned} \end{aligned}$$
and
$$\begin{aligned} \begin{aligned} \int _{\frac{\delta }{c_0}}^{\infty }\frac{r^{2n+1}dr}{(\varepsilon _j^4+r^4)^{1+\frac{n}{2}}} \le \int _{\frac{\delta }{c_0}}^{\infty }\frac{r^{2n+1}dr}{r^{4(1+\frac{n}{2})}}=\frac{C}{\delta ^2}. \end{aligned} \end{aligned}$$
This proves the assertion. \(\square \)
Appendix B
Here we prove Theorem 5.1. We discuss both of the cases when \(n=1\) or M is spherical. For convenience, we denote the CR conformal sub-Laplacian of a contact form \(\theta \) by \(L_\theta \), i.e.
$$\begin{aligned} L_\theta =-\left( 2+\frac{2}{n}\right) \Delta _{\theta }+R_{\theta }. \end{aligned}$$
Let \(\{u_\nu \}\) be a sequence of positive functions satisfying (5.1) and (5.2). Note that \(u_\nu \) is uniformly bounded in \(S_1^2(M)\). To see this, it follows from integration by parts and Hölder’s inequality that
$$\begin{aligned}&\int _M\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}u_\nu |^2_{\theta _0}+R_{\theta _0}u_\nu ^2\right) dV_{\theta _0}\\&\quad \le \left( \int _M|L_{\theta _0}u_\nu -r_\infty u_\nu ^{1+\frac{2}{n}}|^{\frac{2n+2}{n+2}} dV_{\theta _0} \right) ^{\frac{n+2}{2n+2}}\left( \int _Mu_\nu ^{2+\frac{2}{n}}\right) ^{\frac{n}{2n+2}} +r_\infty \int _Mu_\nu ^{2+\frac{2}{n}} dV_{\theta _0}, \end{aligned}$$
which is uniformly bounded by (5.1) and (5.2). Thus, by passing to a subsequence if necessary, we assume that \(u_\nu \) converges to \(u_\infty \) weakly in \(S_1^2(M)\). Since the Folland–Stein embedding \(S_1^2(M)\hookrightarrow L^s(M)\) is compact for \(1<s<2+\displaystyle \frac{2}{n}\) (see Proposition 5.6 in [28]), \(u_\nu \) converges to \(u_\infty \) in \(L^s(M)\) for \(1<s<2+\displaystyle \frac{2}{n}\).
Proposition B.1
The function \(u_\infty \) is a smooth nonnegative function satisfying (5.3).
Proof
Note that for any \(\varphi \in C^\infty (M)\)
$$\begin{aligned} \int _M \varphi u_\nu ^{1+\frac{2}{n}}dV_{\theta _0}\rightarrow \int _M \varphi u_\infty ^{1+\frac{2}{n}}dV_{\theta _0} \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.1)
Indeed, there exists a constant c such that
$$\begin{aligned} \left| u_\nu ^{1+\frac{2}{n}}-u_\infty ^{1+\frac{2}{n}}\right| \le c|u_\nu -u_\infty |\left( |u_\nu |^{\frac{2}{n}}+|u_\infty |^{\frac{2}{n}}\right) \end{aligned}$$
Hence, it follows from Hölder’s inequality that
$$\begin{aligned}&\left| \int _M \varphi u_\nu ^{1+\frac{2}{n}}dV_{\theta _0}-\int _M \varphi u_\infty ^{1+\frac{2}{n}}dV_{\theta _0}\right| \nonumber \\&\quad \le \int _M \left| u_\nu ^{1+\frac{2}{n}}-u_\infty ^{1+\frac{2}{n}}\right| |\varphi |dV_{\theta _0}\nonumber \\&\quad \le c\Vert \varphi \Vert _{L^\infty (M)}\Vert u_\nu -u_\infty \Vert _{L^\beta (M)} \left\| |u_\nu |^{\frac{2}{n}}+|u_\infty |^{\frac{2}{n}}\right\| _{L^{\beta '}(M)} \end{aligned}$$
(B.2)
where \(\beta \) are chosen such that \(1+\displaystyle \frac{2}{n}<\beta <2+\frac{2}{n}\) and \(\displaystyle \frac{1}{\beta }+\frac{1}{\beta '}=1\). Therefore, \(\displaystyle \frac{2}{n}\beta '<2+\frac{2}{n}\). Combining all these, we can conclude that the right hand side of (B.2) tends to 0 as \(\nu \rightarrow \infty \). This proves (B.1). On the other hand, for any \(\varphi \in C^\infty (M)\), we have
$$\begin{aligned} \begin{aligned} \int _M\varphi L_{\theta _0}u_\nu dV_{\theta _0} =\int _M u_\nu L_{\theta _0}\varphi dV_{\theta _0}\rightarrow \int _M u_\infty L_{\theta _0}\varphi dV_{\theta _0}=\int _M\varphi L_{\theta _0}u_\infty dV_{\theta _0} \end{aligned}\nonumber \\ \end{aligned}$$
(B.3)
as \(\nu \rightarrow \infty \). Therefore, it follows from (B.1) and (B.3) that for any \(\varphi \in C^\infty (M)\)
$$\begin{aligned} \int _M\left( L_{\theta _0}u_\nu -r_\infty u_\nu ^{1+\frac{2}{n}}\right) \varphi dV_{\theta _0} \rightarrow \int _M\left( L_{\theta _0}u_\infty -r_\infty u_\infty ^{1+\frac{2}{n}}\right) \varphi dV_{\theta _0} \end{aligned}$$
as \(\nu \rightarrow \infty \). Combining this with (5.2), we can conclude that \(u_\infty \) satisfies (5.3). Since \(u_\nu \) is nonnegative, \(u_\infty \) is also nonnegative. It follows from Theorem 3.22 in [15] that \(u_\infty \) is smooth. This proves the assertion. \(\square \)
Proposition B.2
There holds
$$\begin{aligned} u_\nu ^{1+\frac{2}{n}}-u_\infty ^{1+\frac{2}{n}}-|u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty ) \rightarrow 0 \quad \text{ in } L^{\frac{2n+2}{n+2}}(M) \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.4)
Proof
We use the following inequality:
$$\begin{aligned} \left| |a+b|^{\frac{2}{n}}-|a|^{\frac{2}{n}}\right| \le c\left( |b|^{\frac{2}{n}}+\max \{|a|,|b|\}^{\frac{2}{n}-1}|b|\right) \quad \text{ for } \text{ all } a,b. \end{aligned}$$
(B.5)
Applying (B.5), we get
$$\begin{aligned} u_\nu ^{1+\frac{2}{n}}= & {} u_\nu ^{\frac{2}{n}}u_\infty +(u_\nu -u_\infty +u_\infty )^{\frac{2}{n}}(u_\nu -u_\infty ) \nonumber \\= & {} u_\nu ^{\frac{2}{n}}u_\infty +|u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty ) \nonumber \\&+\,O\left( u_\infty ^{\frac{2}{n}}|u_\nu -u_\infty |+\max \{|u_\nu -u_\infty |,|u_\infty |\}^{\frac{2}{n}-1}|u_\nu -u_\infty |u_\infty \right) \nonumber \\= & {} u_\nu ^{\frac{2}{n}}u_\infty +|u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty ) +O\left( u_\infty ^{\frac{2}{n}}|u_\nu -u_\infty |+|u_\nu -u_\infty |^{\frac{2}{n}}u_\infty \right) .\nonumber \\ \end{aligned}$$
(B.6)
Applying (B.5) again, we obtain
$$\begin{aligned} u_\nu ^{\frac{2}{n}}u_\infty= & {} (u_\nu -u_\infty +u_\infty )^{\frac{2}{n}}u_\infty \nonumber \\= & {} u_\infty ^{1+\frac{2}{n}}+O\left( u_\infty |u_\nu -u_\infty |^{\frac{2}{n}}+\max \{|u_\nu -u_\infty |,|u_\infty |\}^{\frac{2}{n}-1}|u_\nu -u_\infty |u_\infty \right) \nonumber \\= & {} u_\infty ^{1+\frac{2}{n}}+O\left( u_\infty ^{\frac{2}{n}}|u_\nu -u_\infty |+|u_\nu -u_\infty |^{\frac{2}{n}}u_\infty \right) . \end{aligned}$$
(B.7)
Combining (B.6) and (B.7), we get
$$\begin{aligned} u_\nu ^{1+\frac{2}{n}}-u_\infty ^{1+\frac{2}{n}}-|u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty ) =O\left( u_\infty ^{\frac{2}{n}}|u_\nu -u_\infty |+|u_\nu -u_\infty |^{\frac{2}{n}}u_\infty \right) .\nonumber \\ \end{aligned}$$
(B.8)
Since \(u_\infty \) is smooth and \(u_\nu \) converges to \(u_\infty \) in \(L^s(M)\) for \(1<s<2+\displaystyle \frac{2}{n}\), (B.4) follows from (B.8). \(\square \)
If \(u_\nu \) converges to \(u_\infty \) strongly in \(S_1^2(M)\), then Theorem 5.1 follows. Therefore, we assume that \(u_\nu \) converges to \(u_\infty \) weakly in \(S_1^2(M)\), but not strongly in \(S_1^2(M)\). On the other hand, it follows from (5.2), (5.3) and (B.4) that
$$\begin{aligned}&\int _M\Bigg |L_{\theta _0}(u_\nu -u_\infty )-r_\infty |u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty )\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\\&\quad \le C\int _M\Bigg |L_{\theta _0}u_\nu -r_\infty u_\nu ^{1+\frac{2}{n}}\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\\&\qquad +\,C\, r_\infty \int _M\Bigg |u_\nu ^{1+\frac{2}{n}}-u_\infty ^{1+\frac{2}{n}}-|u_\nu -u_\infty |^{\frac{2}{n}}(u_\nu -u_\infty )\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0} \rightarrow 0 \end{aligned}$$
as \(\nu \rightarrow \infty \). That is to say, if we let \(v_\nu :=u_\nu -u_\infty \), then \(\{v_\nu \}\) is a sequence of functions such that \(v_\nu \) converges to 0 weakly in \(S_1^2(M)\), but not strongly in \(S_1^2(M)\), and satisfies
$$\begin{aligned} \int _M\Bigg |L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.9)
We are going to extract bubbles from \(v_\nu \). To do this, we mainly follow the proof of Proposition 8 in [19]. The argument is almost the same except with some small modifications. However, there are some parts in the argument of [19] which are not very precise. So we are going to provide all the details of the proof.
As before, we denote
$$\begin{aligned} B_r(x)=\{y\in M:d(x,y)<r\}, \end{aligned}$$
where d is the Carnot–Carathéodory distance on M with respect to the contact form \(\theta _0\).
Lemma B.3
There exists \(x\in M\) such that for every \(\rho >0\), there exists \(\delta (\rho )>0\) such that
$$\begin{aligned} \liminf _{\nu \rightarrow \infty }\int _{B_\rho (x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} \ge \delta (\rho ). \end{aligned}$$
(B.10)
Proof
Suppose that for all \(x\in M\), there is \(\rho (x)>0\) such that
$$\begin{aligned} \liminf _{\nu \rightarrow \infty }\int _{B_{\rho (x)}(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} =0. \end{aligned}$$
Since M is compact, we can cover M with finite number of \(B_{\rho (x_i)}(x_i)\) with \(i=1, \ldots , L\). As a result, there exists a subsequence of \(\{v_\nu \}\), which is still denoted by \(\{v_\nu \}\), such that
$$\begin{aligned} \int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} \rightarrow 0 \end{aligned}$$
as \(\nu \rightarrow \infty \). This contradicts to the assumption that \(v_\nu \) does not converge to 0 strongly in \(S_1^2(M)\). This proves the assertion. \(\square \)
Lemma B.4
The constant \(\delta (\rho )\) in Lemma B.3 can be taken to be \(a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1}\), where \(a_0\) is any positive real number strictly less than 1.
Before we give the proof of Lemma B.4, we remark that our constant in Lemma B.4 is different from that of [19]. But one can see from the following arguments that a uniform constant will be sufficient for our purpose.
Proof of Lemma B.4
The proof is similar to the proof of Lemma 10 in [19]. Let \(\psi _\nu :{\mathbb {R}}\rightarrow [0,1]\) be a \(C^1\) cut-off function such that
$$\begin{aligned} \psi _\nu (s)=\left\{ \begin{array}{ll} 1, &{} \hbox { for }s\le \rho _\nu ; \\ 0, &{} \hbox { for }s\ge \rho _\nu +\delta _\nu , \end{array} \right. \end{aligned}$$
where \(\rho<\rho _\nu <2\rho \) and \(0<\delta _\nu <\rho \). Fix a point \(x\in M\) where (B.10) holds for x. Define the function \(\phi _\nu \) by \(\phi _\nu (y)=\psi _\nu (d(x,y))\) where d is the Carnot–Carathéodory distance on M with respect to the contact form \(\theta _0\). Then we have
$$\begin{aligned} |\nabla _{\theta _0}\phi _\nu |_{\theta _0}\le \frac{C}{\delta _\nu }. \end{aligned}$$
(B.11)
By Hölder’s inequality, we have
$$\begin{aligned} \int _M\Bigg (L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \Bigg )v_\nu \phi _\nu dV_{\theta _0} \le \Vert \alpha _\nu \Vert _{L^{\frac{2n+2}{n+2}}(M)}\left( \int _M |v_\nu \phi _\nu |^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n}{2n+2}}\nonumber \\ \end{aligned}$$
(B.12)
where \(\alpha _\nu =L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu .\) By integration by parts, the left hand side of (B.12) can be written as
$$\begin{aligned}&\int _M\Bigg (L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \Bigg )v_\nu \phi _\nu dV_{\theta _0}\nonumber \\&\quad =\left( 2+\frac{2}{n}\right) \int _M|\nabla _{\theta _0} v_\nu |_{\theta _0}^2\phi _\nu dV_{\theta _0} +\left( 2+\frac{2}{n}\right) \int _Mv_\nu \langle \nabla _{\theta _0} v_\nu ,\nabla _{\theta _0} \phi _\nu \rangle _{\theta _0} dV_{\theta _0}\nonumber \\&\qquad +\int _MR_{\theta _0}v_\nu ^2\phi _\nu dV_{\theta _0}-r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0}. \end{aligned}$$
(B.13)
It follows from (B.12) and (B.13) that
$$\begin{aligned}&\left( 2+\frac{2}{n}\right) \int _M|\nabla _{\theta _0} v_\nu |_{\theta _0}^2\phi _\nu dV_{\theta _0} +\left( 2+\frac{2}{n}\right) \int _Mv_\nu \langle \nabla _{\theta _0} v_\nu ,\nabla _{\theta _0} \phi _\nu \rangle _{\theta _0} dV_{\theta _0}\nonumber \\&\qquad +\int _MR_{\theta _0}v_\nu ^2\phi _\nu dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0} +\Vert \alpha _\nu \Vert _{L^{\frac{2n+2}{n+2}}(M)}\left( \int _M |v_\nu \phi _\nu |^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n}{2n+2}}.\qquad \qquad \end{aligned}$$
(B.14)
Also, by Folland–Stein embedding theorem in Proposition 2.1, we have
$$\begin{aligned} \left( \int _M |v_\nu \phi _\nu |^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n}{2n+2}} \le C\left( \int _M\left( |\nabla _{\theta _0}(v_\nu \phi _\nu )|^2_{\theta _0}+R_{\theta _0}(v_\nu \phi _\nu )^2\right) dV_{\theta _0} \right) ^{\frac{1}{2}}.\nonumber \\ \end{aligned}$$
(B.15)
By Cauchy–Schwarz inequality, we have
$$\begin{aligned} \left( \int _M|\nabla _{\theta _0}(v_\nu \phi _\nu )|^2_{\theta _0}dV_{\theta _0}\right) ^{\frac{1}{2}}\le & {} \left( \int _M|\nabla _{\theta _0}v_\nu |^2_{\theta _0}\phi _\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}}\nonumber \\&+\,\left( \int _M|\nabla _{\theta _0}\phi _\nu |^2_{\theta _0}v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}}. \end{aligned}$$
(B.16)
Substituting (B.15) and (B.16) into (B.14), we get
$$\begin{aligned}&\left( 2+\frac{2}{n}\right) \int _M|\nabla _{\theta _0} v_\nu |_{\theta _0}^2\phi _\nu dV_{\theta _0} +\left( 2+\frac{2}{n}\right) \int _Mv_\nu \langle \nabla _{\theta _0} v_\nu ,\nabla _{\theta _0} \phi _\nu \rangle _{\theta _0} dV_{\theta _0}\nonumber \\&\qquad +\int _MR_{\theta _0}v_\nu ^2\phi _\nu dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0} +C\,\Vert \alpha _\nu \Vert _{L^{\frac{2n+2}{n+2}}(M)}\left[ \left( \int _M|\nabla _{\theta _0}v_\nu |^2_{\theta _0}\phi _\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}}\right. \nonumber \\&\qquad \left. +\left( \int _M|\nabla _{\theta _0}\phi _\nu |^2_{\theta _0}v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}} +\left( \int _MR_{\theta _0}(v_\nu \phi _\nu )^2dV_{\theta _0}\right) ^{\frac{1}{2}}\right] . \end{aligned}$$
(B.17)
By passing to a subsequence, we obtain from (B.10) that
$$\begin{aligned} \int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0} \ge \frac{\delta (\rho )}{2}. \end{aligned}$$
(B.18)
Since \(\{v_\nu \}\) is uniformly bounded in \(S_1^2(M)\), it follows from (B.9) that
$$\begin{aligned} \Vert \alpha _\nu \Vert _{L^{\frac{2n+2}{n+2}}(M)}\left[ \left( \int _M|\nabla _{\theta _0}v_\nu |^2_{\theta _0}\phi _\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}} +\left( \int _MR_{\theta _0}(v_\nu \phi _\nu )^2dV_{\theta _0}\right) ^{\frac{1}{2}}\right] \rightarrow 0\nonumber \\ \end{aligned}$$
(B.19)
as \(\nu \rightarrow \infty \). By (B.18) and (B.19), we can rewrite (B.17) as follows:
$$\begin{aligned}&(1+o(1))\int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0}\nonumber \\&\qquad +\left( 2+\frac{2}{n}\right) \int _Mv_\nu \langle \nabla _{\theta _0} v_\nu ,\nabla _{\theta _0} \phi _\nu \rangle _{\theta _0} dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0} +C\,\Vert \alpha _\nu \Vert _{L^{\frac{2n+2}{n+2}}(M)}\left( \int _M|\nabla _{\theta _0}\phi _\nu |^2_{\theta _0}v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}}.\qquad \qquad \end{aligned}$$
(B.20)
Note that
$$\begin{aligned} \int _M|\nabla _{\theta _0}\phi _\nu |^2_{\theta _0}v_\nu ^2dV_{\theta _0} \le \frac{C}{\delta _\nu ^2}\int _Mv_\nu ^2dV_{\theta _0} \end{aligned}$$
(B.21)
by (B.11). Hence,
$$\begin{aligned}&\left| \int _Mv_\nu \langle \nabla _{\theta _0} v_\nu ,\nabla _{\theta _0} \phi _\nu \rangle _{\theta _0} dV_{\theta _0}\right| \nonumber \\&\quad \le \left( \int _M|\nabla _{\theta _0}v_\nu |^2_{\theta _0}dV_{\theta _0}\right) ^{\frac{1}{2}} \left( \int _M|\nabla _{\theta _0}\phi _\nu |^2_{\theta _0}v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}} \le \frac{C}{\delta _\nu }\left( \int _Mv_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}}\qquad \qquad \end{aligned}$$
(B.22)
by Hölder’s inequality and the fact that \(\{v_\nu \}\) is uniformly bounded in \(S_1^2(M)\).
If we choose \(\delta _\nu =\displaystyle \left( \int _M v_\nu ^2 dV_{\theta _0}\right) ^{\frac{1}{4}}\), then
$$\begin{aligned} \frac{C}{\delta _\nu }\left( \int _Mv_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty , \end{aligned}$$
(B.23)
since \(v_\nu \rightarrow 0\) in \(L^2(M)\) as \(\nu \rightarrow \infty \). Using (B.21)–(B.23), we can rewrite (B.20) as
$$\begin{aligned}&(1+o(1))\int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0}+o(1). \end{aligned}$$
(B.24)
In view of (B.18), we deduce from (B.24) that
$$\begin{aligned} (1+o(1))\int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0} \le r_\infty \int _M|v_\nu |^{2+\frac{2}{n}}\phi _\nu dV_{\theta _0}.\nonumber \\ \end{aligned}$$
(B.25)
We introduce \(\gamma _\nu >0\) which will be chosen latter. If \(\gamma _\nu \) is sufficiently small, it follows from (B.25) that
$$\begin{aligned}&(1+o(1))\int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _{M}|v_\nu |^{2+\frac{2}{n}}(\phi _\nu +\gamma _\nu ) dV_{\theta _0}. \end{aligned}$$
(B.26)
By the definition of the CR Yamabe constant in (3.1), we have
$$\begin{aligned}\int _{M} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}w|_{\theta _0}^2 +R_{\theta _0}w^2\right) dV_{\theta _0}\ge Y(M,\theta _0)\Vert w\Vert _{L^{2+\frac{2}{n}}(M)}^2. \end{aligned}$$
Applying this to the function \(w=|v_\nu |(\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\), we obtain
$$\begin{aligned}&Y(M,\theta _0) \left( \int _{M}|v_\nu |^{2+\frac{2}{n}}(\phi _\nu +\gamma _\nu ) dV_{\theta _0}\right) ^{\frac{n}{n+1}}\nonumber \\&\quad \le \left( 2+\frac{2}{n}\right) \int _{M} \Bigg (|\nabla _{\theta _0}v_\nu |_{\theta _0}^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}} +v_\nu ^2\Bigg |\nabla _{\theta _0}\Bigg ((\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\Bigg )\Bigg |_{\theta _0}^2\nonumber \\&\qquad +\,2|v_\nu |(\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\langle \nabla _{\theta _0}v_\nu ,\nabla _{\theta _0}\Bigg ((\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\Bigg )\rangle _{\theta _0} \Bigg )dV_{\theta _0}\nonumber \\&\qquad +\int _{M}R_{\theta _0}v_\nu ^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}}dV_{\theta _0}. \end{aligned}$$
(B.27)
By (B.21), (B.23) and the fact that \(\displaystyle \frac{n}{n+1}-2<0\), we can choose \(\gamma _k\) sufficiently small such that
$$\begin{aligned} \int _{M}v_\nu ^2\gamma _\nu ^{\frac{n}{n+1}-2}|\nabla _{\theta _0}\phi _\nu |_{\theta _0}^2=o(1), \end{aligned}$$
which implies that
$$\begin{aligned} \int _{M} v_\nu ^2\Bigg |\nabla _{\theta _0}\Bigg ((\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\Bigg )\big |_{\theta _0}^2dV_{\theta _0}= & {} \int _{M}v_\nu ^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}-2}|\nabla _{\theta _0}\phi _\nu |_{\theta _0}^2dV_{\theta _0} \nonumber \\\le & {} \int _{M}v_\nu ^2\gamma _\nu ^{\frac{n}{n+1}-2}|\nabla _{\theta _0}\phi _\nu |_{\theta _0}^2dV_{\theta _0}=o(1).\nonumber \\ \end{aligned}$$
(B.28)
It follows from (B.28), Hölder’s inequality and the fact that \(\{v_\nu \}\) is uniformly bounded in \(S_1^2(M)\) that
$$\begin{aligned}&\left| \int _{M} |v_\nu |(\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\langle \nabla _{\theta _0}v_\nu ,\nabla _{\theta _0}\Bigg ((\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\Bigg )\rangle _{\theta _0}dV_{\theta _0}\right| \nonumber \\&\quad \le \left( \int _{M} |\nabla _{\theta _0}v_\nu |_{\theta _0}^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}}dV_{\theta _0}\right) ^{\frac{1}{2}}\left( \int _{M} v_\nu ^2\Bigg |\nabla _{\theta _0}\Bigg ((\phi _\nu +\gamma _\nu )^{\frac{n}{2n+2}}\Bigg )\big |_{\theta _0}^2dV_{\theta _0}\right) ^{\frac{1}{2}}\nonumber \\&\quad =o(1). \end{aligned}$$
(B.29)
Putting (B.28) and (B.29) into (B.27), we get
$$\begin{aligned}&Y(M,\theta _0) \left( \int _{M}|v_\nu |^{2+\frac{2}{n}}(\phi _\nu +\gamma _\nu ) dV_{\theta _0}\right) ^{\frac{n}{n+1}}\\&\quad \le \int _{M} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}}+ R_{\theta _0}v_\nu ^2(\phi _\nu +\gamma _\nu )^{\frac{n}{n+1}}\right) dV_{\theta _0}+o(1). \end{aligned}$$
Combining this with (B.26) and using (B.18), we obtain
$$\begin{aligned}&(1+o(1))\int _{M}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu +R_{\theta _0}v_\nu ^2\phi _\nu \right) dV_{\theta _0}\nonumber \\&\quad \le r_\infty Y(M,\theta _0)^{-\frac{n+1}{n}} \left( \int _{M} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2\phi _\nu ^{\frac{n}{n+1}}+ R_{\theta _0}v_\nu ^2\phi _\nu ^{\frac{n}{n+1}}\right) dV_{\theta _0}\right) ^{\frac{n+1}{n}}.\nonumber \\ \end{aligned}$$
(B.30)
It follows from (B.30) and the definition of \(\phi _\nu \) that
$$\begin{aligned}&(1+o(1))\int _{B_{\rho _\nu }(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\nonumber \\&\quad \le r_\infty Y(M,\theta _0)^{-\frac{n+1}{n}} \left( \int _{B_{\rho _\nu }(x)} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+ R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\right) ^{\frac{n+1}{n}}\nonumber \\&\qquad +\,r_\infty Y(M,\theta _0)^{-\frac{n+1}{n}}\left( \int _{B_{\rho _\nu +\delta _\nu }(x)-B_{\rho _\nu }(x)} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+ R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\right) ^{\frac{n+1}{n}}\nonumber \\&\quad :=r_\infty Y(M,\theta _0)^{-\frac{n+1}{n}}(I+II). \end{aligned}$$
(B.31)
It follows from (B.18) that \(I\ge C\delta (\rho )^{\frac{n+1}{n}}\). We have the following two cases:
Case (i). If \(II=o(I)\), then it follows from (B.31) that
$$\begin{aligned}&(1+o(1))\int _{B_{\rho _\nu }(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\\&\quad \le r_\infty Y(M,\theta _0)^{-\frac{n+1}{n}} \left( \int _{B_{\rho _\nu }(x)} \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+ R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\right) ^{\frac{n+1}{n}}, \end{aligned}$$
which implies that
$$\begin{aligned} \int _{B_{\rho _\nu }(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\\ \ge a_0r_\infty ^{-n} Y(M,\theta _0)^{n+1} \end{aligned}$$
where \(a_0\) is any positive real number strictly less than 1.
Case (ii). Suppose that there exists a fixed constant C such that \(II\ge C\delta (\rho )^{\frac{n+1}{n}}\) for all choices of \(\rho _\nu \in [\rho , 2\rho ]\) with \(\delta _\nu =\displaystyle \left( \int _M v_\nu ^2 dV_{\theta _0}\right) ^{\frac{1}{4}}\). In fact, it can not occur since \(\delta _\nu \rightarrow 0\) as \(\nu \rightarrow 0\), and \(\int _M\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0}\) is uniformly bounded. This proves Lemma B.4. \(\square \)
It follows from Lemma B.4 that for any \(x\in M\) satisfying (B.10) and for any given \(a_0<1\), and \(\nu \) sufficiently large, there exists a \(\rho _\nu (x)\) such that
$$\begin{aligned} \int _{B_{\rho _\nu (x)}(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} = a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1}. \end{aligned}$$
Which means for and \(\rho >\rho _\nu (x)\), we have
$$\begin{aligned} \int _{B_{\rho _\nu (x)}(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} >a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1}. \end{aligned}$$
Then for every \(\nu \) sufficiently large, we define:
$$\begin{aligned} \rho _{1,\nu }=\inf \rho _\nu (x), \end{aligned}$$
(B.32)
where the infimum is taken among all \(x\in M\) satisfying (B.10), which is a closed set. Thus the infimum is attained. That is, there exists \(x^*_{1,\nu }\in M\) such that
$$\begin{aligned} \rho _{1,\nu }=\rho _\nu (x^*_{1,\nu }). \end{aligned}$$
(B.33)
For any \(\rho _0>0\), by the proof of Lemma B.4, we have
$$\begin{aligned} \int _{B_{{\widetilde{\rho }}}(x)}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} \end{aligned}$$
for all \(\nu \) sufficiently large. Since \(\rho _0\) is arbitrary, and \(\rho _\nu (x)\le \rho _0\), then we have the following lemma:
Lemma B.5
The sequence \(\{\rho _{1,\nu }\}\) converges to zero as \(\nu \rightarrow \infty \).
As explained in Appendix A, we can find \(\rho >0\) which is independent of x (since M is compact) with \({\widehat{\theta }}_x=\varphi _x^{\frac{2}{n}}\theta _0\) in a neighborhood \(B_{3\delta }(x)\) of x such that (A.1) and (A.4) are satisfied when \(n=1\) and (A.9) is satisfied when M is spherical. We define a sequence of functions \(\{ {\widetilde{v}}_\nu \}\) in \(B_{2\delta }(x)\) as follows:
$$\begin{aligned} {\widetilde{v}}_\nu =\varphi _x^{-1}v_\nu . \end{aligned}$$
Since \({\widehat{\theta }}_x=\varphi _x^{\frac{2}{n}}\theta _0\), by the CR transformation law, we have
$$\begin{aligned} L_{\theta _0}v_\nu =\varphi _x^{1+\frac{2}{n}}L_{{\widehat{\theta }}_x}{\widetilde{v}}_\nu , \end{aligned}$$
(B.34)
which holds in \(B_{2\delta }(x)\subset M\). It follows from (A.8) that (B.34) holds in the CR normal coordinates \(\{(z,t):(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}\subset {\mathbb {H}}^n\), where \({\widehat{\rho }}>0\) is independent of x. By (B.9) and (B.34), we have
$$\begin{aligned} \int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\Bigg |L_{{\widehat{\theta }}_x}{\widetilde{v}}_\nu -r_\infty |{\widetilde{v}}_\nu |^{\frac{2}{n}}{\widetilde{v}}_\nu \Bigg |^{\frac{2n+2}{n+2}}dV_{{\widehat{\theta }}_x} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.35)
By the properties of \(v_\nu \), we know that \({\widetilde{v}}_\nu \) is bounded in \(L^{2+\frac{2}{n}}\) and \({\widetilde{v}}_\nu \rightarrow 0\) in \(L^s\) for all \(s<2+\displaystyle \frac{2}{n}\) as \(\nu \rightarrow \infty \).
When \(n=1\), it follows from (A.1), (A.3), (A.4) and (B.35) that
$$\begin{aligned}&\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left| L_{\theta _{{\mathbb {H}}^n}}{\widetilde{v}}_\nu -r_\infty |{\widetilde{v}}_\nu |^{\frac{2}{n}}{\widetilde{v}}_\nu \right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad \le C\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left| L_{\theta _{{\mathbb {H}}^n}} {\widetilde{v}}_\nu -L_{{\widehat{\theta }}_x}{\widetilde{v}}_\nu \right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\qquad +\,C\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left| L_{{\widehat{\theta }}_x}{\widetilde{v}}_\nu -r_\infty {\widetilde{v}}_\nu ^{1+\frac{2}{n}}\right| ^{\frac{2n+2}{n+2}}\Bigg |dV_{\theta _{{\mathbb {H}}^n}}-dV_{{\widehat{\theta }}_x}\Bigg |\nonumber \\&\qquad +\,C\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left| L_{{\widehat{\theta }}_x}{\widetilde{v}}_\nu -r_\infty |{\widetilde{v}}_\nu |^{\frac{2}{n}}{\widetilde{v}}_\nu \right| ^{\frac{2n+2}{n+2}}dV_{{\widehat{\theta }}_x} \rightarrow 0 \end{aligned}$$
(B.36)
as \(\nu \rightarrow \infty \). When M is spherical, then it follows from (A.9) and (B.35) that
$$\begin{aligned} \int _{\{(t^2+|z|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\Bigg |L_{\theta _{{\mathbb {H}}^n}}{\widetilde{v}}_\nu -r_\infty |{\widetilde{v}}_\nu |^{\frac{2}{n}}{\widetilde{v}}_\nu \Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty , \end{aligned}$$
(B.37)
since \(R_{\theta _{{\mathbb {H}}^n}}=0\).
Let \({\widehat{\chi }}\) be a cut off function such that \({\widehat{\chi }}(s)=1\) if \(0\le s\le \displaystyle \frac{{\widehat{\rho }}}{2}\) and 0 if \(s\ge {\widehat{\rho }}\). Let \(\{{\widetilde{V}}_\nu \}\) be a sequence of functions in \({\mathbb {H}}^n\) defined by
$$\begin{aligned} {\widetilde{V}}_\nu (z,t)= & {} \left\{ \begin{array}{ll} (\rho _{1,\nu })^n\,{\widehat{\chi }}\Bigg (\rho _{1,\nu }(t^2+|z|^4)^{\frac{1}{4}}\Bigg )\,{\widetilde{v}}_\nu \Bigg (\rho _{1,\nu }z,(\rho _{1,\nu })^2t\Bigg ), &{} \hbox { for } (t^2+|z|^4)^{\frac{1}{4}}<\displaystyle \frac{{\widehat{\rho }}}{\rho _{1,\nu }}; \\ 0, &{} \hbox { otherwise,} \end{array} \right. \nonumber \\ \end{aligned}$$
(B.38)
where \(\rho _{1,\nu }\) is defined as in (B.32). Then we have the following:
Proposition B.6
-
(i)
For any fixed ball B of \({\mathbb {H}}^n\), we have
$$\begin{aligned} \int _{B}\left| L_{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu -r_\infty |{\widetilde{V}}_\nu |^{\frac{2}{n}}{\widetilde{V}}_\nu \right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}} \rightarrow 0\quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
-
(ii)
There exists a constant C such that
$$\begin{aligned} \int _{{\mathbb {H}}^n}\left( |\nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu |^2_{\theta _{{\mathbb {H}}^n}}+ |{\widetilde{V}}_\nu |^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\le C \quad \text{ for } \text{ all } \nu . \end{aligned}$$
Proof
To prove (i), we fix \(\rho >0\). Since \(\rho _{1,\nu }\rightarrow 0\) as \(\nu \rightarrow \infty \) by Lemma B.5, there exists N such that \(\displaystyle \rho \le \frac{{\widehat{\rho }}}{2\rho _{1,\nu }}\) when \(\nu \ge N\). Hence, if \(\nu \ge N\), we have
$$\begin{aligned}&\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<\rho \}}\left| L_{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu (z,t)-r_\infty |{\widetilde{V}}_\nu (z,t)|^{\frac{2}{n}}{\widetilde{V}}_\nu (z,t)\right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\\&\quad =\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<\rho \}}\left| L_{\theta _{{\mathbb {H}}^n}} \Bigg ((\rho _{1,\nu })^n\,{\widetilde{v}}_\nu \Bigg (\rho _{1,\nu }z,(\rho _{1,\nu })^2t\Bigg )\Bigg )\right. \\&\qquad \left. -\,r_\infty \Bigg |(\rho _{1,\nu })^n\,{\widetilde{v}}_\nu \Bigg (\rho _{1,\nu }z,(\rho _{1,\nu })^2t\Bigg )\Bigg |^{\frac{2}{n}} (\rho _{1,\nu })^n\,{\widetilde{v}}_\nu \Bigg (\rho _{1,\nu }z,(\rho _{1,\nu })^2t\Bigg )\right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\\&\quad =\int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<\rho \rho _{1,\nu }\}} \left| L_{\theta _{{\mathbb {H}}^n}} {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}}) -r_\infty |{\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})|^{\frac{2}{n}}{\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})\right| ^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}} =o(1), \end{aligned}$$
where the first equality follows from (B.38) and the fact that \(\displaystyle \rho \le \frac{{\widehat{\rho }}}{2\rho _{1,\nu }}\), the second equality follows from the change of variables \(({\tilde{z}},{\tilde{t}})=(\rho _{1,\nu }z,(\rho _{1,\nu })^2t)\), and the last equality follows from (B.36), (B.37) and the fact that \(\displaystyle \rho \,\rho _{1,\nu }\le \frac{{\widehat{\rho }}}{2}\). This proves (i). For (ii), we have
$$\begin{aligned}&\int _{{\mathbb {H}}^n}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu |^2_{\theta _{{\mathbb {H}}^n}}+ r_\infty |{\widetilde{V}}_\nu |^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\\&\quad =\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<\frac{{\widehat{\rho }}}{\rho _{1,\nu }}\}}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu |^2_{\theta _{{\mathbb {H}}^n}}+ r_\infty |{\widetilde{V}}_\nu |^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\\&\quad =\int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left( \left( 2+\frac{2}{n}\right) \left| \nabla _{\theta _{{\mathbb {H}}^n}}\Bigg ({\widehat{\chi }}\Bigg (({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}\Bigg ) {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})\Bigg )\right| ^2_{\theta _{{\mathbb {H}}^n}}\right. \\&\qquad \left. +\, r_\infty \Bigg |{\widehat{\chi }}\Bigg (({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}\Bigg ) {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})\Bigg |^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\\&\quad \le \int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}} \left( 2+\frac{2}{n}\right) {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})^2 \left| \nabla _{\theta _{{\mathbb {H}}^n}}{\widehat{\chi }}\Bigg (({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}\Bigg ) \right| ^2_{\theta _{{\mathbb {H}}^n}}dV_{\theta _{{\mathbb {H}}^n}}\\&\qquad + \int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left( \left( 2+\frac{2}{n}\right) \Bigg ({\widehat{\chi }}\Bigg (({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}\Bigg )\Bigg )^2 |\nabla _{\theta _{{\mathbb {H}}^n}} {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})|^2_{\theta _{{\mathbb {H}}^n}}\right. \\&\qquad \left. +\, r_\infty \Bigg |{\widehat{\chi }}\Bigg (({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}\Bigg ) {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})\Bigg |^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\\&\quad \le \int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}} \left( 2+\frac{2}{n}\right) {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})^2dV_{\theta _{{\mathbb {H}}^n}}\\&\qquad +\, \int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<{\widehat{\rho }}\}}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _{{\mathbb {H}}^n}} {\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})|^2_{\theta _{{\mathbb {H}}^n}}+ r_\infty |{\widetilde{v}}_\nu ({\tilde{z}},{\tilde{t}})|^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\\&\quad \le C, \end{aligned}$$
where the first equality follows from (B.38), the second equality follows from the change of variables \(({\tilde{z}},{\tilde{t}})=(\rho _{1,\nu }z,(\rho _{1,\nu })^2t)\), the third inequality follows from the property of the cut-off function \({\widehat{\chi }}\), and the last inequality follows from the fact that \({\widetilde{v}}_\nu \) is uniformly bounded in \(S_1^2({\mathbb {H}}^n)\). This proves (ii). \(\square \)
It follows from Proposition B.6(ii) that, by passing to subsequence if necessary, \({\widetilde{V}}_\nu \) converges to \({\widetilde{V}}\) weakly in \(S_1^2(B)\) as \(\nu \rightarrow \infty \) on each ball B of \({\mathbb {H}}^n\). Since the Folland–Stein embedding \(S_1^2(B)\hookrightarrow L^s(B)\) is compact for \(1<s<2+\displaystyle \frac{2}{n}\) on each ball B of \({\mathbb {H}}^n\), \({\widetilde{V}}_\nu \) converges to \({\widetilde{V}}\) in \(L^s(B)\) for \(1<s<2+\displaystyle \frac{2}{n}\). On the other hand, it follows from Proposition B.6(i) that \({\widetilde{V}}\) satisfies
$$\begin{aligned} \left( 2+\frac{2}{n}\right) \Delta _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}=r_\infty |{\widetilde{V}}|^{\frac{2}{n}}{\widetilde{V}}\quad \text{ in } {\mathbb {H}}^n. \end{aligned}$$
(B.39)
Lemma B.7
-
(i)
For every ball B in \({\mathbb {H}}^n\), there holds
$$\begin{aligned} \int _B\Bigg |L_{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})-r_\infty |{\widetilde{V}}_\nu -{\widetilde{V}}|^{\frac{2}{n}}({\widetilde{V}}_\nu -{\widetilde{V}})\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}} \rightarrow 0 \text{ as } \nu \rightarrow \infty . \end{aligned}$$
-
(ii)
There exists a constant C such that
$$\begin{aligned} \int _{{\mathbb {H}}^n}\left( |\nabla _{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})|^2_{\theta _{{\mathbb {H}}^n}}+ |{\widetilde{V}}_\nu -{\widetilde{V}}|^{2+\frac{2}{n}}\right) dV_{\theta _{{\mathbb {H}}^n}}\le C \quad \text{ for } \text{ all } \nu . \end{aligned}$$
Proof
By the same proof of Proposition B.2, we have
$$\begin{aligned} {\widetilde{V}}_\nu ^{1+\frac{2}{n}}-{\widetilde{V}}^{1+\frac{2}{n}}-|{\widetilde{V}}_\nu -{\widetilde{V}}|^{\frac{2}{n}}({\widetilde{V}}_\nu -{\widetilde{V}}) \rightarrow 0 \quad \text{ in } L^{\frac{2n+2}{n+2}}(B) \text{ as } \nu \rightarrow \infty . \end{aligned}$$
This together with (B.39) and Proposition B.6(i) implies that
$$\begin{aligned}&\int _B\Bigg |L_{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})-r_\infty |{\widetilde{V}}_\nu -{\widetilde{V}}|^{\frac{2}{n}}({\widetilde{V}}_\nu -{\widetilde{V}})\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\\&\quad \le C\int _B\Bigg |L_{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu -r_\infty |{\widetilde{V}}_\nu |^{\frac{2}{n}}{\widetilde{V}}_\nu \Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _{{\mathbb {H}}^n}}\\&\qquad +\,C\, r_\infty \int _B\Bigg | {\widetilde{V}}_\nu ^{1+\frac{2}{n}}-{\widetilde{V}}^{1+\frac{2}{n}}-|{\widetilde{V}}_\nu -{\widetilde{V}}|^{\frac{2}{n}}({\widetilde{V}}_\nu -{\widetilde{V}})\Bigg |^{\frac{2n+2}{n+2}} dV_{\theta _{{\mathbb {H}}^n}} \rightarrow 0 \end{aligned}$$
as \(\nu \rightarrow \infty \). This proves (i). On the other hand, (ii) follows from (B.39) and Proposition B.6(ii). This proves the assertion. \(\square \)
Lemma B.8
For every ball B in \({\mathbb {H}}^n\), \({\widetilde{V}}_\nu \) converges to \({\widetilde{V}}\) strongly in \(S_1^2(B)\) as \(\nu \rightarrow \infty \).
Proof
By contradiction, we assume that \({\widetilde{V}}_\nu \) does not converge to \({\widetilde{V}}\) strongly in \(S_1^2(B)\) as \(\nu \rightarrow \infty \) for some B in \({\mathbb {H}}^n\). Therefore, it follows from Lemma B.7 that the sequence \(\{{\widetilde{V}}_\nu -{\widetilde{V}}\}\) satisfied the same properties of the sequence \(\{v_\nu \}\). In particular, it follows from Lemma B.4 that there exists a sequence \(({\widetilde{x}}_\nu ,{\widetilde{\rho }}_\nu )\) with \({\widetilde{x}}_\nu \in B_{\rho _{1,\nu }}(x^*_{1,\nu })\) and \({\widetilde{\rho }}_\nu \rightarrow 0\) as \(\nu \rightarrow \infty \) such that
$$\begin{aligned} \left( 2+\frac{2}{n}\right) \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )}|\nabla _{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})|^2_{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}}\ge \frac{1+a_0}{2} r_\infty ^{-n} Y(B,\theta _{{\mathbb {H}}^n})^{n+1}\nonumber \\ \end{aligned}$$
(B.40)
for \(\nu \) sufficiently large. It follows from Lemma B.7 that
$$\begin{aligned} \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )} \langle \nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}},\nabla _{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})\rangle _{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}}\rightarrow 0 \end{aligned}$$
(B.41)
as \(\nu \rightarrow \infty \). Combining (B.40) and (B.41), we get
$$\begin{aligned}&\left( 2+\frac{2}{n}\right) \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )}|\nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}_\nu |^2_{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad =\left( 2+\frac{2}{n}\right) \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )}|\nabla _{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})|^2_{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\qquad +\left( 2+\frac{2}{n}\right) \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )}|\nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}}|^2_{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}} \nonumber \\&\qquad -\,2\left( 2+\frac{2}{n}\right) \int _{\exp _{x^*_{1,\nu }}^{-1}\Bigg (B_{{\widetilde{\rho }}_\nu }({\widetilde{x}}_\nu )\Bigg )} \langle \nabla _{\theta _{{\mathbb {H}}^n}}{\widetilde{V}},\nabla _{\theta _{{\mathbb {H}}^n}}({\widetilde{V}}_\nu -{\widetilde{V}})\rangle _{\theta _{{\mathbb {H}}^n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad \ge \frac{1+a_0}{2} r_\infty ^{-n} Y(B,\theta _{{\mathbb {H}}^n})^{n+1}+o(1) \nonumber \\&\quad \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} \end{aligned}$$
(B.42)
for \(\nu \) sufficiently large, where the last inequality follows from
$$\begin{aligned} Y(B,\theta _{{\mathbb {H}}^n})=Y({\mathbb {H}}^n,\theta _{{\mathbb {H}}^n})=Y(S^{2n+1},\theta _{S^{2n+1}})\ge Y(M,\theta _0). \end{aligned}$$
Since \({\widetilde{V}}_\nu \) has support in \(\{(t^2+|z|^4)^{\frac{1}{4}}<\frac{{\widehat{\rho }}}{\rho _{1,\nu }}\}\), it follows from (B.42) that there exists \(\widetilde{{\widetilde{x}}}_\nu \in B_{\rho _{1,\nu }}(x^*_{1,\nu })\) such that
$$\begin{aligned} \int _{B_{{\widetilde{\rho }}_\nu \rho _{1,\nu }}(\widetilde{{\widetilde{x}}}_\nu )}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} \end{aligned}$$
for \(\nu \) sufficiently large. This implies that
$$\begin{aligned} \rho _{\nu }(\widetilde{{\widetilde{x}}}_\nu )\le {\widetilde{\rho }}_\nu \rho _{1,\nu } <\rho _{1,\nu }, \end{aligned}$$
where we have used the fact that \({\widetilde{\rho }}_\nu \rightarrow 0\) as \(\nu \rightarrow \infty \). But this contradicts to (B.32). This proves Lemma B.8. \(\square \)
Since \({\widetilde{V}}\) satisfies (B.39), it follows from the result of Jerison and Lee in [26] that there exists \((z_0,t_0)\in {\mathbb {H}}^n\) and \(\gamma _1>0\) such that
$$\begin{aligned} {\widetilde{V}}(z,t)=W\circ T_{(z_0,t_0)}(z,t), \end{aligned}$$
where
$$\begin{aligned} W(z,t)=\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}}\left( \frac{\gamma _1^2}{\gamma _1^4t^2+(\gamma _1^2|z|^2+1)^2}\right) ^{\frac{n}{2}} \end{aligned}$$
and
$$\begin{aligned} T_{(z_0,t_0)}(z,t)=(z+z_0,t+t_0+2Im(z\cdot z_0)) \text{ for } (z,t)\in {\mathbb {H}}^n \end{aligned}$$
is the translation in \({\mathbb {H}}^n\). By the optimality of \((x^*_{1,\nu }, \rho _{1,\nu })\), we can conclude that \((z_0,t_0)=(0,0)\); for if \((z_0,t_0)\ne (0,0)\), we can find \(({\widetilde{x}}^*_{1,\nu }, {\widetilde{\rho }}_{1,\nu })\) with \({\widetilde{\rho }}_{1,\nu }<\rho _{1,\nu }\) such that
$$\begin{aligned} \int _{B_{{\widetilde{\rho }}_{1,\nu }}({\widetilde{x}}^*_{1,\nu })}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+R_{\theta _0}v_\nu ^2\right) dV_{\theta _0} \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1}. \end{aligned}$$
Therefore, we have
$$\begin{aligned} {\widetilde{V}}(z,t)=\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}}\left( \frac{\gamma _1^2}{\gamma _1^4t^2+(\gamma _1^2|z|^2+1)^2}\right) ^{\frac{n}{2}}. \end{aligned}$$
(B.43)
We remark that it was claimed in [19] that \(\gamma _1=1\) (see the last line in P.146 in [19]), which does not seem to be true. In fact, we will show that \(\gamma _1\ge C_1\). Here \(C_1\) is a positive constant depending only on \(a_0\), \(r_{\infty }\), and M. We need the following:
Proposition B.9
For any \(x\in M\) and for any \(r>0\), there holds
$$\begin{aligned} \int _{B_r(x)}v_\nu \Delta _{\theta _0}v_\nu dV_{\theta _0} =\int _{B_r(x)}|\nabla _{\theta _0}v_\nu |^2_{\theta _0}dV_{\theta _0}+o(1). \end{aligned}$$
Proof
We consider the following sequence of cut-off functions:
$$\begin{aligned} \chi _\nu (y)=\left\{ \begin{array}{ll} 1, &{} \hbox { if }d(x,y)\le r; \\ 0, &{} \hbox { if }d(x,y)\ge r+r_\nu , \end{array} \right. \end{aligned}$$
(B.44)
such that
$$\begin{aligned} 0\le \chi _\nu \le 1,\quad |\nabla _{\theta _0}\chi _\nu |_{\theta _0}\le \frac{C}{r_\nu } \quad \text{ and } \quad |\Delta _{\theta _0}\chi _\nu |\le \frac{C}{r_\nu ^2}, \end{aligned}$$
(B.45)
where \(r_\nu \) will be chosen later. Since the function \(\chi _\nu v_\nu \) has support in \(B_{r+2r_\nu }(x)\), it follows from integration by parts that
$$\begin{aligned} \begin{aligned} \int _{B_{r+2r_\nu }(x)}\chi _\nu v_\nu \Delta _{\theta _0}(\chi _\nu v_\nu ) dV_{\theta _0} =\int _{B_{r+2r_\nu }(x)}|\nabla _{\theta _0}(\chi _\nu v_\nu )|^2_{\theta _0} dV_{\theta _0}. \end{aligned} \end{aligned}$$
(B.46)
We are going to expand the left and right hand side of (B.46). By (B.44), the left hand side of (B.46) can be written as
$$\begin{aligned}&\int _{B_{r+2r_\nu }(x)}\chi _\nu v_\nu \Delta _{\theta _0}(\chi _\nu v_\nu ) dV_{\theta _0}\\&\quad =\int _{B_{r}(x)}v_\nu \Delta _{\theta _0}v_\nu dV_{\theta _0} +\int _{B_{r+2r_\nu }(x)-B_{r}(x)}\chi _\nu ^2 v_\nu \Delta _{\theta _0} v_\nu dV_{\theta _0}\\&\qquad +\int _{B_{r+2r_\nu }(x)-B_{r}(x)}\chi _\nu v_\nu ^2 \Delta _{\theta _0} \chi _\nu dV_{\theta _0}\\&\qquad +\,2\int _{B_{r+2r_\nu }(x)-B_{r}(x)}\chi _\nu v_\nu \langle \nabla _{\theta _0} \chi _\nu , \nabla _{\theta _0} v_\nu \rangle _{\theta _0} dV_{\theta _0}\\&\quad :=\int _{B_{r}(x)}v_\nu \Delta _{\theta _0}v_\nu dV_{\theta _0} +I+II+III. \end{aligned}$$
By (B.44) and (B.45), we have
$$\begin{aligned} |II|\le \int _{B_{r+2r_\nu }(x)-B_{r}(x)}\left| \chi _\nu v_\nu ^2 \Delta _{\theta _0} \chi _\nu \right| dV_{\theta _0}\le \frac{C}{r_\nu ^2}\int _M v_\nu ^2dV_{\theta _0}. \end{aligned}$$
Since \(v_\nu \) converges to 0 in \(L^2(M)\) as \(\nu \rightarrow \infty \), if we choose
$$\begin{aligned} r_\nu =\left( \int _M v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{4}}\rightarrow 0\quad \text{ as } \nu \rightarrow \infty , \end{aligned}$$
(B.47)
then \(|II|\rightarrow 0\) as \(\nu \rightarrow \infty \). Since \(v_\nu \) converges to 0 in \(L^2(M)\) as \(\nu \rightarrow \infty \), we have
$$\begin{aligned} \int _M \left| R_{\theta _0}v_\nu ^2 \right| dV_{\theta _0}\rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.48)
Since \(r_\nu \rightarrow 0\) as \(\nu \rightarrow \infty \) by (B.47) and \(v_\nu \) is uniformly bounded in \(S_1^2(M)\hookrightarrow L^{2+\frac{2}{n}}(M)\), we have
$$\begin{aligned} \int _{B_{r+2r_\nu }(x)-B_{r}(x)} |v_\nu |^{2+\frac{2}{n}} dV_{\theta _0}\rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
(B.49)
Therefore, we have
$$\begin{aligned}&\left| \left( 2+\frac{2}{n}\right) I\right| \\&\quad \le \left( \int _{B_{r+2r_\nu }(x)-B_{r}(x)} v_\nu ^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n}{2n+2}} \left( \int _M\left| L_{\theta _0} v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \right| ^{\frac{2n+2}{n+2}}dV_{\theta _0}\right) ^{\frac{n+2}{2n+2}}\\&\qquad +\int _M \left| R_{\theta _0}v_\nu ^2 \right| dV_{\theta _0}+r_\infty \int _{B_{r+2r_\nu }(x)-B_{r}(x)} |v_\nu |^{2+\frac{2}{n}} dV_{\theta _0} \\&\quad =o(1), \end{aligned}$$
where we have used (B.9), (B.48) and (B.49) in the last equality. By (B.47), Cauchy–Schwarz inequality and the fact that \(v_\nu \) is bounded in \(S_1^2(M)\), we have
$$\begin{aligned} |III|\le & {} 2\int _{B_{r+2r_\nu }(x)-B_{r}(x)}\left| \chi _\nu v_\nu \langle \nabla _{\theta _0} \chi _\nu , \nabla _{\theta _0} v_\nu \rangle _{\theta _0}\right| dV_{\theta _0}\nonumber \\\le & {} \frac{C}{r_\nu }\left( \int _M v_\nu ^2dV_{\theta _0}\right) ^{\frac{1}{2}} \left( \int _M |\nabla _{\theta _0} v_\nu |^2_{\theta _0}dV_{\theta _0}\right) ^{\frac{1}{2}} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty .\qquad \end{aligned}$$
(B.50)
Combining all these, we can see that the left hand side of (B.46) is equal to
$$\begin{aligned} \int _{B_{r+2r_\nu }(x)}\chi _\nu v_\nu \Delta _{\theta _0}(\chi _\nu v_\nu ) dV_{\theta _0}=\int _{B_{r}(x)}v_\nu \Delta _{\theta _0}v_\nu dV_{\theta _0}+o(1). \end{aligned}$$
(B.51)
On the other hand, the right hand side of (B.46) can be written as
$$\begin{aligned}&\int _{B_{r+2r_\nu }(x)}|\nabla _{\theta _0}(\chi _\nu v_\nu )|^2_{\theta _0} dV_{\theta _0}\\&\quad =\int _{B_{r}(x)}|\nabla _{\theta _0} v_\nu |^2_{\theta _0} dV_{\theta _0} +\int _{B_{r+2r_\nu }(x)-B_r(x)}\chi _\nu ^2|\nabla _{\theta _0}v_\nu |^2_{\theta _0} dV_{\theta _0}\\&\qquad +\int _{B_{r+2r_\nu }(x)-B_r(x)}v_\nu ^2|\nabla _{\theta _0}\chi _\nu |^2_{\theta _0} dV_{\theta _0} +\int _{B_{r+2r_\nu }(x)-B_r(x)}\chi _\nu v_\nu \langle \nabla _{\theta _0}\chi _\nu ,\nabla _{\theta _0}v_\nu \rangle _{\theta _0} dV_{\theta _0}\\&\quad :=\int _{B_{r}(x)}|\nabla _{\theta _0} v_\nu |^2_{\theta _0} dV_{\theta _0} +I'+II'+III'. \end{aligned}$$
Since \(r_\nu \rightarrow 0\) as \(\nu \rightarrow \infty \) by (B.47) and \(v_\nu \) is uniformly bounded in \(S_1^2(M)\), we have
$$\begin{aligned} |I'|\le \int _{B_{r+2r_\nu }(x)-B_r(x)}|\nabla _{\theta _0}v_\nu |^2_{\theta _0} dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
By (B.45) and (B.47), we have
$$\begin{aligned} |II'|=\int _{B_{r+2r_\nu }(x)-B_r(x)}\chi _\nu ^2|\nabla _{\theta _0}v_\nu |^2_{\theta _0} dV_{\theta _0} \le \frac{C}{r_\nu ^2}\int _M v_\nu ^2dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
Since \(III=III'\), it follows from (B.50) that \(|III'|\rightarrow 0\) as \(\nu \rightarrow \infty \). Combining all these, we can see that the right hand side of (B.46) is equal to
$$\begin{aligned} \int _{B_{r+2r_\nu }(x)}|\nabla _{\theta _0}(\chi _\nu v_\nu )|^2_{\theta _0} dV_{\theta _0} =\int _{B_{r}(x)}|\nabla _{\theta _0} v_\nu |^2_{\theta _0} dV_{\theta _0}+o(1). \end{aligned}$$
(B.52)
The assertion follows from combining (B.46), (B.51) and (B.52). \(\square \)
Proposition B.10
There exists a positive constant \(C_1\) depending only on \(a_0\), \(r_\infty \) and M such that
$$\begin{aligned} \gamma _1\ge C_1. \end{aligned}$$
Proof
Since \(\rho _{1,\nu }\rightarrow 0\) as \(\nu \rightarrow \infty \) by Lemma B.5, we can find N such that
$$\begin{aligned} C_0\le \frac{{\widehat{\rho }}}{2\rho _{1,\nu }} \quad \text{ whenever } \nu \ge N, \end{aligned}$$
(B.53)
where \(C_0\) is the uniform constant given in (A.8). Note that
$$\begin{aligned}&\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}}|{\widetilde{V}}_\nu (z,t)|^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad =\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}} \Bigg |(\rho _{1,\nu })^n\,{\widetilde{v}}_\nu \Bigg (\rho _{1,\nu }z,(\rho _{1,\nu })^2t\Bigg )\Bigg |^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad =\int _{\{\rho _{x^*_{1,\nu }}(y)<C_0\rho _{1,\nu }\}}|{\widetilde{v}}_\nu (y)|^{2+\frac{2}{n}} dV_{{\widehat{\theta }}_{x^*_{1,\nu }}}+o(1)\nonumber \\&\quad \ge \int _{\{d(x^*_{1,\nu },y)<\rho _{1,\nu }\}}|v_\nu (y)|^{2+\frac{2}{n}} dV_{\theta _0}+o(1), \end{aligned}$$
(B.54)
where the first equality follows from (B.53) and the definition of \({\widetilde{V}}_\nu \) in (B.38), and the second equality follows from (A.1) and the change of variables \(({\tilde{z}},{\tilde{t}})=(\rho _{1,\nu }z,(\rho _{1,\nu })^2t)\), and the last inequality follows from (A.8) and the definition of \({\widetilde{v}}_\nu \). For sufficiently large \(\nu \), we have
$$\begin{aligned}&r_\infty \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}|v_\nu |^{2+\frac{2}{n}} dV_{\theta _0}\nonumber \\&\quad \ge \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}v_\nu L_{\theta _0}v_\nu dV_{\theta _0}\nonumber \\&\qquad - \left( \int _{M}\left| L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \right| ^{\frac{2n+2}{n+2}}dV_{\theta _0}\right) ^{\frac{n+2}{2n+2}} \left( \int _M|v_\nu |^{2+\frac{2}{n}} dV_{\theta _0}\right) ^{\frac{n}{2n+2}}\nonumber \\&\quad =\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu |^2_{\theta _0}+R_{\theta _0}v_\nu ^{2+\frac{2}{n}} \right) dV_{\theta _0}+o(1)\nonumber \\&\quad \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} +o(1), \end{aligned}$$
(B.55)
where the first inequality follows from Hölder’s inequality, the last equality follows from Proposition B.9, (B.9) and the fact that \(v_\nu \) is bounded in \(L^{2+\frac{2}{n}}(M)\), and the last inequality follows from the definition of \((x^*_{1,\nu },\rho _{1,\nu })\) in (B.33).
Combining (B.54) and (B.55), we obtain
$$\begin{aligned} \int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}}|{\widetilde{V}}_\nu (z,t)|^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}} \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} +o(1). \end{aligned}$$
Combining this with Lemma B.8, we get
$$\begin{aligned} \int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}}{\widetilde{V}}(z,t)^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}} \ge a_0 r_\infty ^{-n} Y(M,\theta _0)^{n+1} +o(1). \end{aligned}$$
(B.56)
We compute
$$\begin{aligned}&\int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}}{\widetilde{V}}(z,t)^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad = \left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\{(t^2+|z|^4)^{\frac{1}{4}}<C_0\}}\left( \frac{\gamma _1^2}{\gamma _1^4t^2+(\gamma _1^2|z|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad = \left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\{({\tilde{t}}^2+|{\tilde{z}}|^4)^{\frac{1}{4}}<\gamma _1 C_0\}}\left( \frac{1}{{\tilde{t}}^2+(|{\tilde{z}}|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}} \end{aligned}$$
(B.57)
where the first equality follows from (B.43) and the second equality follows from change of variables \(({\tilde{z}},{\tilde{t}})=(\gamma _1z,\gamma _1^2t)\). Note that the last term in (B.57) tends to zero as \(\gamma _1\rightarrow 0^+\). Hence, combining (B.56) and (B.57), we can conclude that \(\gamma _1\) is bounded below by a positive constant \(C_1\) depending only on \(a_0\), \(r_\infty \), and M. This proves the assertion. \(\square \)
For any \((x,\lambda )\in M\times (0,\infty )\), we can find a unique solution \({\widehat{\omega }}(x,\lambda )\) of the following equation:
$$\begin{aligned} L_{\theta _0}{\widehat{\omega }}(x,\lambda )=r_\infty \omega '(x,\lambda )^{1+\frac{2}{n}} \quad \text{ in } M, \end{aligned}$$
(B.58)
where \(\omega '(x,\lambda )\) is defined as
$$\begin{aligned} \omega '(x,\lambda )(y)= \left\{ \begin{array}{ll} \chi _\delta (\rho _x(y))\varphi _x(y)\omega (x,\lambda )(y), &{} \hbox { for } y\in B_{2\delta }(x); \\ 0, &{} \hbox { otherwise.} \end{array} \right. \end{aligned}$$
Here \(\chi _\delta \) is the cut-off function defined in (A.11), \(\varphi _x\) is the conformal factor such that \({\widehat{\theta }}_x=\varphi _x^{\frac{2}{n}}\theta _0\) in a neighborhood \(B_{3\delta }(x)\) of x. Moreover, \(\omega (x,\lambda )(y)\) is given by
$$\begin{aligned} \omega (x,\lambda )(y)=\left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}}\left( \frac{\lambda ^2}{\lambda ^4t^2+(\lambda ^2|z|^2+1)^2}\right) ^{\frac{n}{2}}, \end{aligned}$$
(B.59)
where (z, t) is CR normal coordinates of y centered at x. It follows from the definition of \({\widehat{\omega }}(x,\lambda )\) that
$$\begin{aligned} {\widehat{\omega }}(x,\lambda )(y)=\omega '(x,\lambda )(y)=0 \quad \hbox { for }y\in M-B_{2\delta }(x). \end{aligned}$$
(B.60)
When \(n=1\), there holds (see Proposition 1 in [18])
$$\begin{aligned} |{\widehat{\omega }}(x,\lambda )(y)-\omega '(x,\lambda )(y)|\le C\lambda ^{-1}(1+|\log (\lambda ^{-2}+\rho _x(y)^2)|) \quad \text{ for } y\in B_{2\delta }(x)\nonumber \\ \end{aligned}$$
(B.61)
and (see (3.6) in [18])
$$\begin{aligned} \Bigg |L_{\theta _0}\Bigg ({\widehat{\omega }}(x,\lambda )(y)-\omega '(x,\lambda )(y)\Bigg )\Bigg |\le \inf \Bigg \{1,\frac{C}{\rho _x(y)^2+\lambda ^{-2}}\Bigg \} \quad \text{ for } y\in B_{2\delta }(x).\nonumber \\ \end{aligned}$$
(B.62)
It follows from (B.60)–(B.62) that
$$\begin{aligned} \Vert {\widehat{\omega }}(x,\lambda )-\omega '(x,\lambda )\Vert _{S_1^2(M)}\rightarrow 0 \quad \text{ as } \lambda \rightarrow \infty . \end{aligned}$$
(B.63)
Similarly, when M is spherical, it follows from Lemmas 3 and 4 in [19] that
$$\begin{aligned} |{\widehat{\omega }}(x,\lambda )(y)-\omega '(x,\lambda )(y)|\le \frac{C}{\lambda ^n} \quad \text{ for } y\in B_{2\delta }(x) \end{aligned}$$
(B.64)
and
$$\begin{aligned} \Vert {\widehat{\omega }}(x,\lambda )-\omega '(x,\lambda )\Vert _{S_1^2(M)}=O\left( \frac{1}{\lambda }\right) . \end{aligned}$$
(B.65)
We have the following:
Proposition B.11
There holds
$$\begin{aligned} \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })} \Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{2+\frac{2}{n}}dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
Proof
By Lemma B.5, we can find N such that
$$\begin{aligned} C_0 \rho _{1,\nu }\le \delta \quad \text{ for } \nu \ge N, \end{aligned}$$
where \(C_0\) is the constant in (A.8). Therefore, for \(\nu \ge N\), we have
$$\begin{aligned}&\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })} \Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{2+\frac{2}{n}}dV_{\theta _0}\nonumber \\&\quad =\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })} \Bigg |{\widetilde{v}}_\nu (x)- \chi _\delta (\rho _{x^*_{1,\nu }}(x))\omega \Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )(x) \Bigg |^{2+\frac{2}{n}} dV_{{\widehat{\theta }}_{x^*_{1,\nu }}}\nonumber \\&\quad \le \int _{\{(|z|^4+t^2)^{\frac{1}{4}}\le C_0\rho _{1,\nu }\}} \Bigg |{\widetilde{v}}_\nu (z,t)- \omega \Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )(z,t)\Bigg |^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}}\nonumber \\&\quad =\int _{\{(|z|^4+t^2)^{\frac{1}{4}}\le C_0\}} \Bigg |{\widetilde{V}}_\nu ({\tilde{z}},{\tilde{t}})-{\widetilde{V}}_\nu ({\tilde{z}},{\tilde{t}})\Bigg |^{2+\frac{2}{n}} dV_{\theta _{{\mathbb {H}}^n}}, \end{aligned}$$
(B.66)
where the first inequality follows from (A.8), the first equality from the definition of \(\omega '(x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }})\) in (B.58), and the last equality follows from (B.43), (B.59) and the change of variables \(({\tilde{z}},{\tilde{t}})=(\frac{z}{\rho _{1,\nu }},\frac{t}{(\rho _{1,\nu })^2})\). Thanks to Lemma B.8, the last expression in (B.66) tends to zero as \(\nu \rightarrow \infty \). This proves the assertion. \(\square \)
We can now extract from \(\{v_\nu \}\) the first bubble and consider the following new sequence of functions:
$$\begin{aligned} v_\nu ^1(x)=v_\nu (x)-{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) (x). \end{aligned}$$
Lemma B.12
-
(i)
The sequence \(\{v_\nu ^1\}\) satisfies
$$\begin{aligned} \int _M\Bigg |L_{\theta _0}v_\nu ^1-r_\infty |v_\nu ^1|^{\frac{2}{n}}v_\nu ^1\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty \end{aligned}$$
-
(ii)
There exists a constant C such that
$$\begin{aligned} \int _{M}\left( |\nabla _{\theta _0}v^1_\nu |_{\theta _0}^2+(v^1_\nu )^2\right) dV_{\theta _0}\le C \quad \text{ for } \text{ all } \nu . \end{aligned}$$
Proof
We follow the proof of Lemma 14 in [19]. By Lemma B.5, we can choose N such that \(\rho _{1,\nu }\le 2\delta \) for \(\nu \ge N.\) For \(\nu \ge N\), it follows from (B.60) that
$$\begin{aligned}&\int _M\Bigg |L_{\theta _0}v_\nu ^1-r_\infty |v_\nu ^1|^{\frac{2}{n}}v_\nu ^1\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\nonumber \\&\quad \le C\int _{M}\Bigg |L_{\theta _0}v_\nu -r_\infty |v_\nu |^{\frac{2}{n}}v_\nu \Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\nonumber \\&\qquad +\,C\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\Bigg |L_{\theta _0}{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )-r_\infty {\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )^{1+\frac{2}{n}}\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\nonumber \\&\qquad +\,C\,r_\infty \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\left| |v_\nu |^{\frac{2}{n}}v_\nu -{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}\right. \nonumber \\&\qquad \left. -\Bigg |v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| ^{\frac{2n+2}{n+2}}dV_{\theta _0}. \end{aligned}$$
(B.67)
It follows from (B.60), (B.61), (B.64), Lemma B.5 and the definition of \({\widehat{\omega }}(x,\lambda )\) that
$$\begin{aligned}&\int _M\Bigg |L_{\theta _0}{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )-r_\infty {\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )^{1+\frac{2}{n}}\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0}\nonumber \\&\quad =r_\infty \int _M\Bigg |\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )^{1+\frac{2}{n}}- {\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )^{1+\frac{2}{n}}\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0} =O(\rho _{1,\nu })=o(1).\nonumber \\ \end{aligned}$$
(B.68)
On the other hand,
$$\begin{aligned}&\left| |v_\nu |^{\frac{2}{n}}v_\nu -{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}-\Bigg |v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| \\&\quad \le \left| |v_\nu |^{\frac{2}{n}}v_\nu -\omega '\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}\right. \\&\qquad \left. -\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| \\&\qquad +\left| \omega '\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}} -{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}\right| \\&\qquad +\left| \omega '\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) -{\widetilde{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) \right| ^{1+\frac{2}{n}}\\&\qquad +\left| \Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg ({\widetilde{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )-\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| \\&\qquad +\left| \Bigg |{\widetilde{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )-\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| \\&\quad :=I+II+III+IV+V. \end{aligned}$$
It follows from the proof of (B.8) that
$$\begin{aligned} I=O\left( |v_\nu |^{\frac{2}{n}} \Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg | +\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} |v_\nu |\right) . \end{aligned}$$
This together with Hölder’s inequality implies that
$$\begin{aligned}&\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })} \left| |v_\nu |^{\frac{2}{n}}v_\nu -\omega '\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}\right. \\&\qquad \left. -\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| ^{\frac{2n+2}{n+2}}dV_{\theta _0}\\&\quad \le C\left( \int _M|v_\nu |^{2+\frac{2}{n}}\right) ^{\frac{2}{n+2}}\left( \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n}{n+2}}\\&\qquad +\,C\left( \int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{2}{n+2}} \left( \int _M|v_\nu |^{2+\frac{2}{n}}\right) ^{\frac{n}{n+2}}\\&\quad =o(1), \end{aligned}$$
where the last equality follows from Proposition B.11 and the fact that \(v_\nu \) is bounded in \(L^{2+\frac{2}{n}}(M)\). Also, it follows from (B.60), (B.61) and (B.64) that
$$\begin{aligned} II\le C\frac{\rho _{1,\nu }}{\gamma _1} \quad \text{ and } \quad III\le C\frac{\rho _{1,\nu }}{\gamma _1}. \end{aligned}$$
It follows from (B.60), (B.61), (B.64) and Hölder’s inequality that
$$\begin{aligned} IV\le C\frac{\rho _{1,\nu }}{\gamma _1}\Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\varepsilon ^*_{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \quad \text{ and } \quad V\le C\Bigg (\frac{\varepsilon ^*_{1,\nu }}{\gamma _1}\Bigg )^{\frac{2}{n}} \Bigg |v_\nu -\omega '\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |. \end{aligned}$$
Combining all these with Lemma B.5, Propositions B.10 and B.11, we can conclude that
$$\begin{aligned}&\int _{B_{\rho _{1,\nu }}(x^*_{1,\nu })}\left| |v_\nu |^{\frac{2}{n}}v_\nu -{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^{1+\frac{2}{n}}\right. \nonumber \\&\quad \left. -\Bigg |v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg |^{\frac{2}{n}} \Bigg (v_\nu -{\widehat{\omega }}\Bigg (x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\Bigg )\Bigg )\right| ^{\frac{2n+2}{n+2}}dV_{\theta _0} =o(1).\qquad \qquad \end{aligned}$$
(B.69)
Now (i) follows from (B.9), (B.67), (B.68) and (B.69).
For (ii), note that
$$\begin{aligned}&\int _M\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}{\widehat{\omega }}(x,\lambda )|_{\theta _0}^2+R_{\theta _0}{\widehat{\omega }}(x,\lambda )^2\right) dV_{\theta _0}\nonumber \\&\quad =r_\infty \int _M\omega '(x,\lambda )^{1+\frac{2}{n}}{\widehat{\omega }}(x,\lambda )dV_{\theta _0}\nonumber \\&\quad =r_\infty \int _M\omega '(x,\lambda )^{2+\frac{2}{n}}dV_{\theta _0}+O\left( \frac{1}{\lambda }\right) \int _M\omega '(x,\lambda )^{1+\frac{2}{n}}dV_{\theta _0}\nonumber \\&\quad \le r_\infty \int _M\omega '(x,\lambda )^{2+\frac{2}{n}}dV_{\theta _0}+O\left( \frac{1}{\lambda }\right) \left( \int _M\omega '(x,\lambda )^{2+\frac{2}{n}}dV_{\theta _0}\right) ^{\frac{n+2}{2n+2}}\qquad \qquad \end{aligned}$$
(B.70)
where the first equality follows from the definition of \({\widehat{\omega }}(x,\lambda )\), the second equality follows from (B.60), (B.61) and (B.64), the last inequality follows from the Hölder’s inequality. Note also that it follows from the definition of \(\omega '(x,\lambda )\) that
$$\begin{aligned} \begin{aligned} \int _M\omega '(x,\lambda )^{2+\frac{2}{n}}dV_{\theta _0} \le \int _{B_{2\delta }(x)}\omega (x,\lambda )^{2+\frac{2}{n}}dV_{\theta _0} \le \int _M\omega (x,1)^{2+\frac{2}{n}}dV_{\theta _0}, \end{aligned}\qquad \qquad \end{aligned}$$
(B.71)
where the last inequality follows from the change of variables \(({\tilde{z}},{\tilde{t}})=(\lambda z,\lambda ^2t)\) in the CR normal coordinates. We derive from (B.70) and (B.71) that there exists a uniform constant C such that
$$\begin{aligned} \int _M\left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}{\widehat{\omega }}(x,\lambda )|_{\theta _0}^2+R_{\theta _0}{\widehat{\omega }}(x,\lambda )^2\right) dV_{\theta _0}\le C \end{aligned}$$
(B.72)
when \(\lambda \) is sufficiently large. By Cauchy–Schwarz inequality, we have
$$\begin{aligned}&\int _{M}\left( |\nabla _{\theta _0}v^1_\nu |_{\theta _0}^2+(v^1_\nu )^2\right) dV_{\theta _0}\nonumber \\&\quad \le 2\int _{M}\left( |\nabla _{\theta _0}v_\nu |_{\theta _0}^2+v_\nu ^2\right) dV_{\theta _0}\nonumber \\&\qquad +\,2\int _{M}\left( \left| \nabla _{\theta _0}{\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) \right| _{\theta _0}^2+ {\widehat{\omega }}\left( x^*_{1,\nu },\frac{\gamma _1}{\rho _{1,\nu }}\right) ^2\right) dV_{\theta _0}. \end{aligned}$$
(B.73)
Now (ii) follows from combining (B.72), (B.73) and the assumption that \(\{v_\nu \}\) is uniformly bounded in \(S_1^2(M)\). This proves Lemma B.12. \(\square \)
Iterating the above procedure, either \(v_\nu ^1\) converges strongly to 0 in \(S_1^2(M)\) as \(\nu \rightarrow \infty \), or we can find a new sequence \((x^*_{2,\nu },\rho _{2,\nu })\) and extract another bubble by defining
$$\begin{aligned} v_\nu ^2(x)=v_\nu ^1(x)-{\widehat{\omega }}\left( x^*_{2,\nu },\frac{\gamma _2}{\rho _{2,\nu }}\right) (x) \end{aligned}$$
and show that
$$\begin{aligned} \int _M\Bigg |L_{\theta _0}v_\nu ^2-r_\infty |v_\nu ^2|^{\frac{2}{n}}v_\nu ^2\Bigg |^{\frac{2n+2}{n+2}}dV_{\theta _0} \rightarrow 0 \quad \text{ as } \nu \rightarrow \infty . \end{aligned}$$
On the other hand, it can be shown that (see Lemmas 15 and 16 in [19])
$$\begin{aligned} \rho _{2,\nu }\ge \frac{1}{2}\rho _{1,\nu } \quad \text{ and } \quad \frac{\rho _{2,\nu }}{\rho _{1,\nu }}+\frac{d(x^*_{1,\nu },x^*_{2,\nu })^2}{\rho _{1,\nu }\,\rho _{2,\nu }} \rightarrow \infty \end{aligned}$$
(B.74)
as \(\nu \rightarrow \infty \). Here d is Carnot–Carathéodory distance on M with respect to the contact form \(\theta _0\). This argument can be iterated as long as the new sequence \(\{v_\nu ^l\}\) does not coverage strongly to 0 in \(S_1^2(M)\). And we claim that the iteration must terminate in finite steps. To see this, note that
$$\begin{aligned}&\int _M \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu ^{l}|^2_{\theta _0} +R_{\theta _0}(v_\nu ^{l})^2\right) dV_{\theta _0}\\&\quad =\int _M \left( \left( 2+\frac{2}{n}\right) \Bigg |\nabla _{\theta _0}\Bigg (v_\nu ^{l-1}-{\widehat{\omega }}\Bigg (x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\Bigg )\Bigg )\Bigg |^2_{\theta _0}\right. \\&\qquad \left. +\,R_{\theta _0}\Bigg (v_\nu ^{l-1}-{\widehat{\omega }}\Bigg (x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\Bigg )\Bigg )^2\right) dV_{\theta _0}\\&\quad =\int _M \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu ^{l-1}|^2_{\theta _0} +R_{\theta _0}(v_\nu ^{l-1})^2\right) dV_{\theta _0}\\&\qquad +\,r_\infty \int _M{\widehat{\omega }}\left( x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\right) \omega '\left( x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\right) ^{1+\frac{2}{n}}dV_{\theta _0} +o(1), \end{aligned}$$
where the first equality follows from the fact that \(v_\nu ^{l-1}\) converges to 0 weakly in \(S_1^2(M)\) as \(\nu \rightarrow \infty \), and the last equality follows from (B.58). We compute
$$\begin{aligned}&\int _M{\widehat{\omega }}\left( x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\right) \omega '\left( x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\right) ^{1+\frac{2}{n}}dV_{\theta _0}\\&\quad \ge \int _{B_{\delta }(x^*_{l,\nu })}\Bigg (\varphi _{x^*_{l,\nu }}(y)\omega \Bigg (x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\Bigg )(y)\Bigg )^{2+\frac{2}{n}}dV_{\theta _0}+o(1)\\&\quad \ge \int _{\left\{ (|z|^4+t^2)^{\frac{1}{4}}\le \frac{\delta }{C_0}\right\} }\omega \left( x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }}\right) (z,t)^{2+\frac{2}{n}}dV_{\theta _{{\mathbb {H}}^n}}+o(1)\\&\quad =\left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\left\{ (|{\tilde{z}}|^4+{\tilde{t}}^2)^{\frac{1}{4}}\le \frac{\delta }{C_0\rho _{l,\nu }}\right\} } \left( \frac{\gamma _{l}^2}{\gamma _{l}^4{\tilde{t}}^2+(\gamma _{l}^2|{\tilde{z}}|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}}+o(1)\\&\quad =\left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\left\{ (|{\widehat{z}}|^4+{\widehat{t}}^2)^{\frac{1}{4}}\le \frac{\gamma _{l}\delta }{C_0\rho _{l,\nu }}\right\} } \left( \frac{1}{{\widehat{t}}^2+(|{\widehat{z}}|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}}+o(1)\\&\quad \ge \left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\left\{ (|{\widehat{z}}|^4+{\widehat{t}}^2)^{\frac{1}{4}}\le \frac{C_1\delta }{C_0}\right\} } \left( \frac{1}{{\widehat{t}}^2+(|{\widehat{z}}|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}}+o(1), \end{aligned}$$
where the first inequality follows from (B.60), (B.61), (B.64) and the definition of \(\omega '(x^*_{l,\nu },\frac{\gamma _{l}}{\rho _{l,\nu }})\), the second inequality follows from (A.1) and (A.8), the first equality follows from the change of variables \(({\tilde{z}},{\tilde{t}})=(\frac{z}{\rho _{l,\nu }},\frac{t}{(\rho _{l,\nu })^2})\), the second equality follows from the change of variables \(({\widehat{z}},{\widehat{t}})=(\gamma _l{\tilde{z}},\gamma _{l}^2{\tilde{t}})\), and the last inequality follows from Proposition B.10 and Lemma B.5. Hence, if we let
$$\begin{aligned} C_2=r_\infty \left( \frac{n(2n+2)}{r_\infty }\right) ^{n+1} \int _{\{(|{\widehat{z}}|^4+{\widehat{t}}^2)^{\frac{1}{4}}\le \frac{C_1\delta }{C_0}\}} \left( \frac{1}{{\widehat{t}}^2+(|{\widehat{z}}|^2+1)^2}\right) ^{n+1}dV_{\theta _{{\mathbb {H}}^n}}, \end{aligned}$$
then it follows from the above computation that
$$\begin{aligned}&\int _M \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu ^{l}|^2_{\theta _0} +R_{\theta _0}(v_\nu ^{l})^2\right) dV_{\theta _0}\\&\quad \ge \int _M \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu ^{l-1}|^2_{\theta _0} +R_{\theta _0}(v_\nu ^{l-1})^2\right) dV_{\theta _0}+C_2+o(1). \end{aligned}$$
That is to say, the quantity \(\int _M \left( \left( 2+\frac{2}{n}\right) |\nabla _{\theta _0}v_\nu ^{l-1}|^2_{\theta _0} +R_{\theta _0}(v_\nu ^{l-1})^2\right) dV_{\theta _0}\) at the l-th step decreases by at least \(C_2\) after extraction of a bubble. Therefore, the iteration must stop after finite steps.
Therefore, there exists an integer m and a sequence of m-tuples \((x^*_{k,\nu },\varepsilon ^*_{k,\nu })_{1\le k\le m}\) where \(\varepsilon ^*_{k,\nu }=\displaystyle \frac{\rho _{k,\nu }}{\gamma _k}\) such that
$$\begin{aligned} \varepsilon ^*_{k,\nu }\rightarrow 0 \quad \text{ as } \nu \rightarrow \infty \text{ for } \text{ all } 1\le k\le m, \end{aligned}$$
by Lemma B.5 and Proposition B.10. Also, we have
$$\begin{aligned} \Bigg \Vert v_\nu -\sum _{k=1}^m{\widehat{\omega }}\left( x^*_{k,\nu },\frac{1}{\varepsilon ^*_{k,\nu }}\right) \Bigg \Vert _{S^2_1(M)}\rightarrow 0 \quad \text{ as } \quad \nu \rightarrow \infty . \end{aligned}$$
(B.75)
Now (5.4) follows from (B.74) and Proposition B.10. On the other hand, we have
$$\begin{aligned}&\Bigg \Vert u_\nu -u_\infty -\sum _{k=1}^m{\overline{u}}_{\Bigg (x^*_{k,\nu },\varepsilon ^*_{k,\nu }\Bigg )}\Bigg \Vert _{S^2_1(M)}\\&\quad =\left\| v_\nu -\sum _{k=1}^m{\widehat{\omega }}\left( x^*_{k,\nu },\frac{1}{\varepsilon ^*_{k,\nu }}\right) -\sum _{k=1}^m\left( \omega '\left( x^*_{k,\nu },\frac{1}{\varepsilon ^*_{k,\nu }}\right) -{\widehat{\omega }}\left( x^*_{k,\nu },\frac{1}{\varepsilon ^*_{k,\nu }}\right) \right) \right. \\&\qquad \left. -\sum _{k=1}^m \varphi _{x^*_{k,\nu }}(y) \left( \frac{n(2n+2)}{r_\infty }\right) ^{\frac{n}{2}} (\varepsilon ^*_{k,\nu })^n \Bigg (1-\chi _\delta (\rho _{x^*_{k,\nu }}(y))\Bigg )G_{x^*_{k,\nu }}(y)\right\| _{S^2_1(M)}=o(1) \end{aligned}$$
where the first equality follows from (A.10) and (A.11), and the last equality follows from (B.59), (B.63), (B.65), (B.75), Lemma B.5 and the fact that the Green’s function \(G_{x^*_{k,\nu }}(y)\) is bounded in \(S_1^2(K)\) for any compact set \(K\subset M-\{x^*_{k,\nu }\}\) (see (A.6) and (A.7)). This proves (5.5) and this completes the proof of Theorem 5.1.