Abstract
In this paper we will analyze the blow-up behaviors of solutions to the singular Liouville type equation with exponential Neumann boundary condition. We generalize the Brezis–Merle type concentration-compactness theorem to this Neumann problem. Then along the line of the Li–Shafrir type quantization property we show that the blow-up value \(m(0) \in 2\pi \mathbb N\cup \{ 2\pi (1+\alpha )+2\pi (\mathbb N\cup \{0\})\}\) if the singular point 0 is a blow-up point. In the end, when the boundary value of solutions has an additional condition, we can obtain the precise blow-up value \(m(0)=2\pi (1+\alpha )\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(\Omega \) be a bounded smooth domain in \(\mathbb R^2\). As is well known, topological degree and variational methods can be used to obtain existence results for many Liouville type equations. And this requires the compactness property for the solution set. So it is important to obtain the blow-up analysis for the equations. The asymptotic blow-up analysis for Liouville type equations has already a lot of progresses. In 1991, Brezis and Merle [2] showed a concentration-compactness phenomena of solutions to the following Liouville equation:
And then Li and Shafrir [10] initiated to evaluate the blow-up value at the blow-up point. They showed at the each blow-up point the blow-up value is quantized, i.e., there is no contribution of mass outside the m disjoint balls which contain a contribution of \(8\pi m\) mass for some positive integer m.
In recent years, the Liouville type equation with singular data attracts much attention due to their many applications in Mathematics and Physics, such as cosmic string equation, Chern–Simons and Electroweak self-dual vortices, etc, see [12, 14,15,16,17]. This type equation can be reduced to the following equation:
The Brezis–Merle type concentration-compactness type result has been established in [4] and [3]. Furthermore, Tarantello [13] generalized Li–Shafrir type quantization property to show that the blow-up value \(m(0) \in 8\pi \mathbb N\cup \{ 8\pi (1+\alpha )+8\pi (\mathbb N\cup \{0\})\}\) if the singular point 0 is a blow-up point.
In addition, there have been some progresses in the blow-up analysis of the Liouville type equation under the Neumann boundary condition. Guo and Liu [7] have analyzed the following equation:
Here and in the sequel, \(\nu \) is the out unit normal vector on the boundary. They also obtained Brezis–Merle type concentration-compactness phenomena and Li–Shafrir type quantization property. Later, Bao et al. [5] have studied the following geometric equations on compact Riemnn surface (M, g):
They obtained Brezis–Merle type concentration-compactness phenomena. Recently, Zhang et al. [18] have proved the quantization property of blowing-up solutions for the local equations:
Here L is a proper subset of \(\partial \Omega \), V(x) and h(x) are nonnegative bounded functions.
In this paper we will consider the local singular Liouville type equation with Neumann boundary condition. Without loss of generality, we consider the following boundary value problem in \(B^+_R(0)\):
where \(\alpha \in (-1,+\infty )\) and the coefficient functions \(V_n\) and \(h_n\) satisfy
In the sequel, we always assume that \(V_n(x)\) and \(h_n(x)\) satisfy the above assumptions. We set \( B^+_{R}(x_{0})=\{x=(s,t)\in \mathbb R^2| |x-x_{0}|<R, t>0\}\), \( L_{R}(x_0)=\partial B^+_{R}(x_{0})\cap \partial \mathbb R^2_{+}\) and \( S^+_{R}(x_{0})=\partial B^+_{R}(x_{0})\cap \mathbb R^2_{+}\). We also use the notations \(B^+_{R}\), \(L_{R}\), \(S^+_{R}\) for \(B^+_{R}(0)\), \(L_{R}(0)\), \(S^+_{R}(0)\) respectively.
Our first main result is about “Brezis–Merle type concentration-compactness phenomena Theorem”.
Theorem 1.1
Assume that \(\{u_{n}\}\) is a sequence of solutions of (1) with \(\alpha \in (-1,+\infty )\). If \(\{u_n\}\) satisfies the energy conditions
for the constant C which is independent of n, then there exists a subsequence, denoted still by \(\{u_{n}\}\), satisfying one of the following alternatives:
-
(i)
\( \{u_{n}\} \) is bounded in \( L^{\infty }_{loc} (B^+_{R}\cup L_{R})\),
-
(ii)
\( \{u_{n}\} \rightarrow -\infty \) uniformly on compact subsets of \(B^+_{R}\cup L_{R}\),
-
(iii)
We can define a finite and nonempty blow-up set of \(u_{n}\)
$$\begin{aligned} \quad \quad S=\{ x\in B^+_{R}\cup L_{R}, \text { there is a sequence } y_{n}\rightarrow x \text { such that }u_{n}(y_{n})\rightarrow +\infty \}. \end{aligned}$$such that
$$\begin{aligned} \{u_{n}\} \rightarrow -\infty \text{ uniformly } \text{ on } \text{ compact } \text{ subsets } \text{ of } (B^+_{R}\cup L_{R})\backslash S. \end{aligned}$$
Our second main result is about “Li–Shafrir type quantization property”.
Theorem 1.2
Assume that \(\{u_{n}\}\) is a sequence of solutions of (1) with \(R=1\) and \(\alpha \in (-1,+\infty )\backslash \{2k+1\}, k=0,1,2,\ldots \). If \(\{u_n\}\) satisfies in addition that
i.e. zero is the only blow-up point of \(u_n \) in \(\bar{B}^+_{1}\), then \(m(0)\in 2\pi \mathbb N\cup \{ 2\pi (1+\alpha )+2\pi (\mathbb N\cup \{0\})\}\).
From Theorem 1.2 it is natural to ask what is the precise value of the “mass” m(0). We give an affirmative answer under an extra boundary condition:
with C a suitable positive constant.
Theorem 1.3
Under the assumptions of Theorem 1.2, if we suppose in addition that \(u_{n}\) satisfies (5), then we have \(m(0)=2\pi (1+\alpha )\).
The proof of our main results follow closely the ideas in [4, 10, 13]. Since the problems involve Neumann boundary condition and the singular data, the steps of the blow-up analysis become more delicate. When we prove Theorem 1.1, we need to use the local Green representation formula and the Pohozaev type identity of Neumann problem. For the proof of Theorem 1.2, we will use the approach in [10, 13], which is based on a classification result of bubbling equation
with \(\int _{\mathbb R^2}e^{2u}<\infty \) and a “\(\sup +\inf \)” type inequality
for equation \(-\triangle u=Ve^{2u}\) in \(B_1\). For our problem, we need the corresponding results. On one hand, besides of the above bubbling equation, there exist the other two kinds of bubbling equation, i.e.
with the energy conditions
and
with the energy condition
We will use the classification results shown in [8, 11] to handle our problem. On the other hand, we need to prove a “\(\sup +\inf \)” type inequality for this Neumann problem by using the moving plan method.
This paper is organized as follows. In this introduction, we state our main theorems. In Sect. 2, we study the blow-up behaviors for the considered Neumann boundary value problem, and give the proof of corresponding concentration-compactness Theorem 1.1. In Sects. 3 and 4, we give the version of Tarantello’s decomposition Proposition and “\(\sup +\inf \)” type inequality under the Neumann boundary conditions separately. In Sect. 5, we will prove Theorem 1.2. In Sect. 6, we will consider the case \(u_{n}\) satisfy (5) and then we give the the proof of Theorem 1.3.
2 Blow-up analysis
In this section, we will study the blow-up behaviors for the considered Liouville type equation with Neumann boundary value condition and with singular data. We shall analyze the regularity of solutions to (1), (2) and (3). Consequently, we can prove Theorem 1.1. In the sequel, we will handle the problem with \(\alpha \ge 0\) and \(-1<\alpha < 0\) separately.
Proposition 2.1
Let \(\alpha \ge 0\), \(\epsilon _{1}<\frac{\pi }{2}\) and \(\epsilon _{2}<\pi \). Assume that \(\{u_{n}\}\) is a sequence of solutions which satisfies that
and
Then \(u^+_{n}\) is bounded in \(L^{\infty }(\bar{B}^+_{\frac{r}{4}})\).
Proof
Define \(u_{1,n}\), \(u_{2,n}\) by
Extending \(u_{n}\), \(u_{1,n}\) and \(V_n\) evenly we have
Due to \(\epsilon _{1}<\frac{\pi }{2}\) we obtain that
Now by Theorem 1 in [2] we can choose \(\delta _1\) such that
with \(\delta _{1}>0\). Then we have
For \(u_{2,n}\), since \(\epsilon _{2}<\pi \), by Lemma 3.2 in [8] we also can choose \(\delta _{2}>0\), \(\delta _{3}>0\) such that
Let \(u_{3,n}=u_{n}-u_{1,n}-u_{2,n}\). Then we have
Extending \(u_{3,n}\) evenly, \(u_{3,n}\) becomes a harmonic function in \(B_{r}\). Then the mean value theorem for harmonic functions implies that
Notice that
Now we choose \(t>0\) such that \(\int _{B_{r}^+}\frac{1}{|x|^{2\alpha t}}dx\le C\). Set \(s=\frac{t}{t+1}<1\) when \(\alpha >0\) and \(s=1\) when \(\alpha =0\). Then it follows from Holder’s inequality to get
Therefore we have
and consequently we have
Finally, we rewrite the equations as
Since
we know that \( \Vert f_{n}\Vert _{L^{q}(B^+_{\frac{r}{2}}})\le C\) and \( \Vert g_{n}\Vert _{L^{q}(L_{\frac{r}{2}})}\le C\) for some \(q>1\). Then the standard elliptic estimates imply that
\(\square \)
Next we consider the case \(-1<\alpha < 0\). There have subtle differences between the case \(-1<\alpha < 0\) and the case \(\alpha \ge 0\).
Proposition 2.2
Let \(-1<\alpha < 0\), and choose suitable constants \(1<p<\frac{2}{1-\alpha }\), \(p_{1}=\frac{\alpha -1}{2\alpha }\) and \(p_{2}=\frac{1-\alpha }{1+\alpha }\) such that \(\frac{1}{p_{1}}+\frac{1}{p_{2}}=1\). Let \(\epsilon _{1}<\frac{\pi }{2pp_{2}}\), and \(\epsilon _{2}<\frac{\pi }{pp_{2}}\). Assume that \(\{u_{n}\}\) is a sequence of solutions which satisfies that
and
Then \(\Vert u^+_{n}\Vert _{L^{\infty }(\bar{B}^+_{\frac{r}{4}})}\) is bounded.
Proof
Define \(u_{1,n}\), \(u_{2,n}\) by
Extending \(u_{n}\), \(u_{1,n}\) and \(V_n(x)\) evenly we have
As the similar arguments in Proposition 2.1 we can obtain for some \(\delta >0\) that
and
Let \(u_{3,n}=u_{n}-u_{1,n}-u_{2,n}\). Then we have
Extending \(u_{3,n}\) evenly, \(u_{3,n}\) becomes a harmonic function in \(B_{r}\). Then the mean value theorem for harmonic function implies that
Notice that
Since \(\alpha <0\), (7) implies
So we get
And we have
Thus, by Holder’ inequality and \(pp_{1}<-\frac{1}{\alpha }\),
Hence we have \(u_{1,n}\) is uniformly bounded in \(B_r^+\cup L_r\) and consequently
The standard elliptic estimates imply that
\(\square \)
Next we present an inequality which has been established in [7].
Lemma 2.3
[7] Let l be an imbedded \(C^1\) curve in \(\mathbb R^2\). \(f\in L^1(l)\). Set \(\Vert f\Vert _{1}=\int _{l}|f(x)|dx\), and \(\rho =\text {diam}~l\). If we define
then for every \(\delta \in (0,\pi )\) we have
By using Lemma 2.3, we can get the following Lemma.
Lemma 2.4
Set \(f(x)\in L^1(L_r)\). If we define
then for every \(k >0\) we have \(e^{k|\omega |}\in L^{1}(L_r)\) and \(e^{k|\omega |}\in L^{1}(B_{r}^+)\).
Proof
Let \(0<\epsilon <\frac{1}{k}\). Since \(f(x)\in L^{1}(L_{r})\), we can split f(x) as \(f(x)=f_{1}(x)+f_{2}(x)\) with \(\Vert f_{1}\Vert _{1} <\epsilon \) and \(f_{2}\in L^{\infty }(L_r)\). Write \(\omega (x)=\omega _{1}(x)+\omega _{2}(x)\) where
Choosing \(\delta =\pi -1\) in Lemma 2.3 we find \(\int _{L_r}\exp [|\omega _{1}(x)|/\Vert f_{1}\Vert _{1}]dx\le C\). This implies that \(e^{k|\omega _{1}|}\in L^{1}(L_r)\) for every \(k >0\). Thus the conclusion follows the fact \( |\omega |\le |\omega _{1}|+|\omega _{2}|\) and \(\omega _{2}\in L^{\infty }(L_r)\). Using the same method of Lemma 2.3, we can get \(\int _{B_{r}^+}\exp [(2\pi -\delta )|\omega (x)|/\Vert f\Vert _{1}]dx\le C\). Further more we can also obtain \(e^{k|\omega |}\in L^{1}( B_{r}^+)\) for every \(k >0\). \(\square \)
Remark 2.5
If we set \(f(x)\in L^{1}(B_{r}^+)\) and
by using the arguments in Lemma 2.4 again, then we can also obtain \(e^{k|\omega |}\in L^{1}(L_r)\) and \(e^{k|\omega |}\in L^{1}( B_{r}^+)\) for every \(k >0\) .
In addition we need a Harnack inequality for a non-homogenous Neumann-type boundary problem for second-order elliptic equations, which has been established in [9].
Proposition 2.6
Let \(f\in L^p(B_{r}^{+})\) for some \(1<p\le +\infty \), \(g\in L^q(B_{r}^{+}\cap \partial \mathbb R^2_{+})\) for some \(1<q\le +\infty \), and u satisfy
Then for any \(0<\theta <1\), there exist a constant \(\beta \in (0,1)\) depending on r, \(\theta \) only, and a constant \(\gamma >0\) depending on r, p, q only such that
When the energy \(\int _{B_{R}^{+}}V_{n}|x|^{2\alpha }e^{2u_{n}}\) and \(\int _{L_{R}}h_{n}|x|^{\alpha }e^{u_{n}} \) are large, the blow-up phenomenon may occur, which is declared in Theorem 1.1. Next we give the proof of Theorem 1.1.
Proof of Theorem 1.1
Firstly we treat the case \(\alpha \ge 0\). Since \(V_{n}|x|^{2\alpha }e^{2u_{n}}\) is bounded in \(L^{1}(B_{R}^{+})\) and \(h_{n}|x|^{\alpha }e^{u_{n}}\) is bounded in \(L^{1}(L_{R})\), along a subsequence (still denoted by \(u_{n}\)), such that
for every \(\varphi \in C_c(B_{R}^{+}\cup L_R)\) and \(\phi \in C_c(L_{R})\). Here \(\mu \) and \(\vartheta \) are two nonnegative bounded measures. A point \(x\in B_{R}^{+}\cup L_{R}\) is called an \(\epsilon -\) regular point with respect to \(\mu \) and \(\vartheta \) if there is a function \(\varphi \in C(B_{R}^{+}\cup L_{R})\), \(supp \varphi \in B_{r}(x)\cap (B_{R}^{+}\cup L_{R})\) with \(0\le \varphi \le 1\) and \(\varphi =1\) in a neighborhood of x such that
\(\square \)
We define the
By \(\int _{B_{R}^{+}}V_{n}|x|^{2\alpha }e^{2u_{n}}\le C\) and \(\int _{L_{R}}h_{n}|x|^{\alpha }e^{u_{n}}\le C\), we have \(\Sigma (\epsilon )\) is finite. Furthermore we have \(S=\Sigma (\epsilon _{0})\) by using the similar arguments in [2, 5], where \(\epsilon _{0}=\min \{\epsilon _{1},\epsilon _{2}\}\) as in Proposition 2.1.
When \(S=\emptyset \), it follows that (i) or (ii) holds. \(S=\emptyset \) means that \(u_{n}^+\) is uniformly bounded in \(L^\infty (B_{R}^{+}\cup L_{R})\). Thus \(f_{n}=V_{n}|x|^{2\alpha }e^{2u_{n}}\) is bounded in \(L^p(B_{R}^{+})\) for any \(p>1\), and \(g_{n}=h_{n}|x|^{\alpha }e^{u_{n}}\) is bounded in \(L^p(L_{R})\) for any \(p>1\). Apply Harnack inequality in Proposition 2.6, we know that (i) or (ii) holds.
For the case \(-1<\alpha <0\), we will use Proposition 2.2 instead of Proposition 2.1. Then similar with the case \(\alpha \ge 0\), we can show (i) or (ii) holds when \(S=\emptyset \).
When \(S\ne \emptyset \), we can show that (iii) holds. Actually in this case, we know that \(u_{n}^+\) is uniformly bounded in \(L^\infty _{loc}(B_{R}^{+}\cup L_{R}\backslash S)\) and therefore \(f_{n}\) is bounded in \(L^p_{loc}(B_{R}^{+}\backslash S)\) for some \(p>1\) and \(g_{n}\) is bounded in \(L^p_{loc}(L_{R}\backslash S)\) for some \(p>1\). Then we have that either
or
We should show that (9) does not happen when \(S\ne \emptyset \). To this purpose, we can take a point \(p\in S\) and choose a small \(r_0>0\) such that p is the only blow-up point in \( \bar{B}_{r_0}^{+}\). Then it is suffice to prove that
If \(p\ne 0\), this is a smooth case and (11) has been shown in [5]. So next we suppose \(p=0\). Since \(u_{n}\) is uniformly bounded in \(L^{\infty }_{loc}( \bar{B}_{r_0}^+\backslash \{0\})\), then we use elliptic estimates, and along a subsequence, we may assume that
Noticing that, by Fatou’s lemma, \(V(x)|x|^{2\alpha }e^{2\xi }\in L^{1}( B_{r_0}^+)\) and \(h(x)|x|^{\alpha }e^{\xi }\in L^{1}(L_{r_0})\), we have for any \(0<r\le r_0\)
here \(\beta \) is the blow-up value for the blow-up point \(p=0\), which is defined by
Set
By Green’s representation formula for \(u_{n}\) in \(\bar{B}_{r_{0}}^+\) and (13) we derive that
with
and
Clearly,
For \(\phi (x)\), we want to estimate the decay of \(\phi \) near the zero. we observe first that \(\phi (x)\) is bounded from below on \(\bar{B}^+_{r_{0}}\), as we have,
By (2) we find
and
Thus by the integrability of \(\varphi _{1}\) and \(\varphi _{2}\), we see that necessarily
On the other hand, let us set \(s=\frac{\beta }{\pi }-\alpha \) and split \(\phi =\phi _1+\phi _2\), where
Noticing that, in view of (15), \(s<1\), it follows that
and
By Lemma 2.4 and Remark 2.5, for every \(k >0\) we have \(e^{k|\phi _{1}|}\in L^{1}(L_{r_{0}})\), \(e^{k|\phi _{2}|}\in L^{1}(L_{r_{0}})\), \(e^{k|\phi _{1}|}\in L^{1}(B^+_{r_{0}})\) and \(e^{k|\phi _{2}|}\in L^{1}(B^+_{r_{0}})\). By Holder’s inequality it follows that \(\varphi _1(x)\in L ^{t}(B^+_{r_{0}})\) for any \(t \in (1, \frac{1}{s})\) if \(s>0\), and \(V(x)|x|^{2\alpha }e^{2\xi }\in L ^{t}(B^+_{r_{0}})\) for any \(t >1\) if \(s\le 0\). We also have \(\varphi _2(x) \in L ^{t}(L_{r_{0}})\) for any \(t \in (1, \frac{1}{s})\) if \(s>0\), and \(\varphi _2(x) \in L ^{t}(L_{r_{0}})\) for any \(t >1\) if \(s\le 0\). But if \(-1<\alpha <0\), we have \(0<s<1\). Since \(\phi \) satisfies that
we get that \(\phi \) is in \(L^\infty (B^+_{r_0}\cap L_{r_0})\). Furthermore, if \(s\le 0\), then \(\phi \) is in \(C^1(B^+_{r_0}\cap L_{r_0})\). If \(s>0\), \(\nabla \phi (x)\) will have a decay when \(x\rightarrow 0\). Without loss of generality, we assume that \(0<s<1\) in the sequel. We estimate \(\nabla \phi (x)\) for \(x\in B_{r_0}^+(0)\).
For \(I_1\), we fix \(t\in (1,\frac{1}{s})\) and choose \(\tau _{1}>0\) such that \(\frac{\tau _{1} t}{t-1}<2\), and hence we have \(0<\tau _{1} <2-2s\). By Holder’s inequality we obtain,
For \(I_{2}\), since \(|x-y|\le \frac{|x|}{2}\) implies that \(|y|\ge \frac{|x|}{2}\), we have
Similarly, for \(I_3\), we fix \(t\in (1,\frac{1}{s})\) and choose \(\tau _{2}>0\) such that \(\frac{\tau _{2} t}{t-1}<1\). and hence we have \(0<\tau _{2} <1-s\). By Holder’s inequality we obtain,
For \(I_{4}\) we have
for some \(\tau _{3}\) with \(0<\tau _{3}<1\).
In conclusion, for all \(x\in B_{r_0}^+(0)\) we have
for suitable constants \(0<\tau _{1} <2-s\), \(0<\tau _{2} <1-s\) and \(0<\tau _{3}<1\).
At this point we are ready to derive our contradiction by means of a Pohozaev type identity. We multiply all terms in (1) by \(x\cdot \nabla u_n\) and integrate over \(B_{r}^+(0)\) for any \(r\in (0,r_0)\) to get
Passing to the limit in (17) to derive the following identity
Set \(\eta = \phi +\gamma \). Since \( \nabla \xi (x)=-\frac{\beta }{\pi }\frac{x}{|x|^2}+\nabla \eta (x)\), we have
Since \(\gamma \in C^{1}(B_{r}^+)\), by (16) we have
with \(0<\tau _{1} <2-s\), \(0<\tau _{2} <1-s\) and \(0<\tau _{3}<1\). So,
Similarly, letting \(r\rightarrow 0\) on the right side of (18) we also can obtain that
Comparing (19) and (20), we see that necessarily \(\beta =2\pi (1+\alpha )\), in contradiction with (15). Therefore, the proof of Theorem 1.1 is finished.
3 A version of Tarantello’s decomposition proposition
In this section, we would like to show the new version of Tarantello’ decomposition Proposition for Liouville equation under the Neumann boundary condition. Firstly we give the “Minimal-Mass Lemma”, which is frequently used in the following Proposition.
Lemma 3.1
Assume that \(\{u_{n}\}\) is a sequence of solutions to (1) for \(R=1\) and \(u_n \) satisfies (2) and (4). If there exists a sequence \(\{x_{n}\}\subset \bar{B}^+_{1}\backslash \{0\}\) such that
Then we must have \( x_{0}=0 \) and
for every small \( \delta >0\).
Proof
Noticing that 0 is the only blow-up point for \(u_{n}\) in \(\bar{B}^+_{1}\) and \(u_{n}(x_{n})\rightarrow +\infty ,\) we have \(x_{0}=0\). Next we consider the new function
Then \(v_n(x)\) satisfies
with the energy conditions
Suppose that along a subsequence \(\frac{x_{n}}{|x_{n}|}\rightarrow x_0\in \bar{\mathbb R}^2_+\) with \(|x_{0}|=1\). Hence \(x_{0}\) define a blow-up point for \(v_{n}\) as we have
Moreover functions \(V_{n}(|x_{n}|x)|x|^{2\alpha }\) and \(h_{n}(|x_{n}|x)|x|^{\alpha }\) are uniformly bounded from above and below near \(x_{0}\).
Consequently, if \(x_0\in \mathbb R^2_+\), by [2] we have for sufficiently small \(\delta >0\),
and if \(x_0\in \partial \mathbb R^2_+\), by [5] we have for sufficiently small \(\delta >0\),
A simple change of variables leads to the conclusion. \(\square \)
On the other hand, if (21) fails to hold, i.e. \(\sup \limits _{\bar{B}^+_{R}} \{ u_{n}(x)+(\alpha +1)\log |x|\}\le C\), we will treat this situation in the following Lemma.
Lemma 3.2
Assume that \(\{u_{n}\}\) is a sequence of solutions to (1) for \(R>0\) and \(u_n \) satisfies (2) and (4). If
then we have
Proof
Let \(u_{n}(x_{n})=\max \limits _{\bar{B}^+_{R}} u_{n}\rightarrow +\infty \) and \(\varepsilon _{n}=e^{-\frac{u_{n}(x_{n})}{\alpha +1}}\rightarrow 0\). By (24) we get
In \(\bar{B}^+_{\frac{R}{\varepsilon _{n}}}\) we define
Then \(\xi _{n}\) satisfies
with the energy conditions
Then necessarily alternative (i) in Theorem 1.1 must hold, in other words, \(\xi _{n}\) is uniformly bounded in \(L^{\infty }_{loc}(\bar{\mathbb R}^{2}_{+})\). In particular,
\(\square \)
Remark 3.3
In fact in Lemma 3.2, we have additionally that along a subsequence \(\xi _{n}\rightarrow \xi \) uniformly in \(C^2_{loc}({\mathbb R}^2_{+})\cap C^1_{loc}(\bar{\mathbb R}^2_{+}\backslash \{0\})\cap C^0_{loc}(\bar{\mathbb R}^2_{+})\). Without loss of generality we always assume that
in the sequel. Then by the classification results in [8] we know that \(\xi \) takes the form
In addition, \(\int _{\mathbb R^2_{+}}|x|^{2\alpha }e^{2\xi }+\int _{\partial \mathbb R^2_{+}}|x|^{\alpha }e^{\xi }=2\pi (1+\alpha )\). In the forth section, we can further obtain by assistant with the Harnack inequality
provided that the assumptions of Lemma hold.
In general, the assumption “\(\sup \limits _{\bar{B}^+_{R}} \{ u_{n}(x)+(\alpha +1)\log |x|\}\le C\)” does not always hold. So we must distinguish between the situation whether (24) holds or not. In particular we have the following Tarantello’s decomposition Proposition.
Proposition 3.4
Assume that \(\{u_{n}\}\) is a sequence of solutions to (1) for \(R=1\) and \(u_n \) satisfies (2) and (4). Then there exists \(\varepsilon _{0}\in \left( 0,\frac{1}{2}\right) \) such that the following alternatives hold:
Proof
If (27) fails to hold for every \(\varepsilon _{0}\in (0,\frac{1}{2})\), then we find a sequence \(x_{n}\subset \bar{B}^+_{1}\) and
Then by (22), we have
\(\forall \delta >0.\) Setting
Next we consider the new sequences \(v_{n}\) in \(\bar{B}^+_{2\varepsilon _{0}}\). We repeat the alternative above for the sequence \(v_{n}\) in \(\bar{B}^+_{2\varepsilon _{0}}\). If \(\sup \limits _{\bar{B}^+_{2\varepsilon _{0}}} \{ v_{n}(x)+(\alpha +1)\log |x|\}\le C \) holds with a suitable \(\varepsilon _{0}\in (0,\frac{1}{2})\), then in this case we can set \(x_{1,n}=x_{n}\). If there exists a sequence \(x'_{n}\), \(v_{n}(x'_{n})+(\alpha +1)\log |x'_{n}|\rightarrow +\infty \), then in this case there exists a second sequence \(\tilde{x}_{n}=|x_n|x'_n\subset \bar{B}^+_{1}\), such that
Consequently by (22),
In addition, notice that \(\bar{B}^+_{\delta |\tilde{x}_{n}|}(\tilde{x}_{n})\) and \(\bar{B}^+_{\delta |x_{n}|}(x_{n})\) do not intersect for \(\delta \in (0,1)\) and n large.
Next we consider the new scaling sequences
We make the same alternative above for the new sequence \(v'_{n}\). We see that each time the new iterated sequence \(v'_{n}\) fails to verify (27), we contribute with at least an account of \(2\pi \) to the blow-up value m(0). So necessarily after a number of steps we find \(\varepsilon _{0}\in \left( 0,\frac{1}{2}\right) \) and a sequence \(\{ x_{1,n} \}\subset \bar{B}^+_{1} \):
Now, for \(\varepsilon _{0}\in \left( 0,\frac{1}{2}\right) \), we repeat an analogous alternative for \(u_{n}\) on the set \(\bar{B}^+_{1}\backslash \bar{B}^+_{\frac{1}{2\varepsilon _{0}} |x_{1,n}|}\). If
in this case we then obtain the first sequences \(x_{1,n}\). If there exists a sequence \(\{ y_{n} \}\subset \bar{B}^+_{1}\backslash \{0\} \) such that
By (22) we have
for \(\forall \delta >0\). Our next task is to obtain the second sequence \(x_{2,n}\) for \(\varepsilon _{0}\in (0,\frac{1}{2})\). In this direction, we consider
If (34) is uniformly bounded for any \(\varepsilon _{0}\in (0,\frac{1}{2})\) then we would let \(x_{2,n}=y_{n}\) and adjust according \(\varepsilon _{0}\) in order to ensure (30) with \(j=1\). Otherwise we would replace \(y_{n}\) with a new sequence \(y'_{n}\) which have the same properties (32) and (33), but \(\frac{|y'_{n}|}{|y_{n}|}\rightarrow 0, \text { as } n \rightarrow \infty \). Moreover each time when such a new sequences exist, we at least contribute an amount of \(2\pi \) to the blow-up value m(0). So making the same alternative for such new sequence, this procedure must stop a number of steps. And we arrive to one for which (34) is uniformly bounded for every \(\varepsilon _{0}\in (0,\frac{1}{2})\). Such sequence will define \(x_{2,n}\). So we obtain the desired proproties (29) (30) for \(j=1\) by adjust \(\varepsilon _{0}\in (0,\frac{1}{2})\). At this point we iterate the argument above by replacing \(x_{1,n}\) with the new sequence \(x_{2,n}\).
Either (28) (29) (30) hold for \(m=2\), or we obtain a third sequence for which we can verify (29) (30) for \(j=1,2\). Since the blow-up value m(0) is finite, so only a finite number of sequence \(x_{j,n}\) satisfying (29) (30) are allowed. Then after a finite number of steps we arrive to the desired conclusion. \(\square \)
4 A version of “\( \sup +\inf \)” type inequality
In this section we will show a version of “\( \sup +\inf \)” type inequality for Liouville equation under the Neumann boundary condition. This inequality concerns the case where the sequence \(u_n\) is subject to alternative (ii) of Proposition 3.4. It is the key part for the proof of Theorem 1.2.
Proposition 4.1
Assume that \(\{u_{n}\}\) is a sequence of solutions to (1) with \(R=1\) which satisfying (2) and (3). Suppose that there exists \(\varepsilon _{0}>0\) and a sequence \(\{ x_{n} \}\subset \bar{B}^+_{1} \) such that
-
(i)
\(x_{n}\rightarrow 0, u_{n}(x_{n})+(\alpha +1)\log |x_{n}|\rightarrow +\infty \);
-
(ii)
\(\sup \limits _{\bar{B}^+_{2\varepsilon _{0}|x_{n}|}} \{ u_{n}(x)+(\alpha +1)\log |x|\}\le C\);
Set \(v_{n}(x)=u_{n}(|x_{n}|x)+(\alpha +1)\log |x_{n}|\). Then passing to a subsequence, we have
for suitable constant \(r_{0}>0\).
Proof
We use a moving plane technique to obtain our conclusion. Similar arguments also be used in [1, 13]. As usual we identify \(x=(x_{1},x_{2})\in \mathbb R^2\) and \(x_{1}+ix_{2}\in \mathbb C\), where \(\mathbb C\) is complex plane. Recalling (2), without loss of generality, suppose \(A\ge B\) and \(a\le b\).
In polar coordinates, define
for \((t,\theta )\in Q=(-\infty ,0]\times [0,\pi ]\). A simple calculation shows that
where \(\tilde{V}_{n}(t,\theta )=V_{n}(e^{t+i\theta })e^{\frac{2A}{a}e^{t}}\) and \(\tilde{h}_{n}(t,\theta )=h_{n}(e^{t+i\theta })e^{\frac{A}{a}e^{t}}\).
Since for fixed n
we have
Furthermore,
for suitable \(C(n)>0\) depending on n. Thus we can choose \(\lambda \) sufficiently negative (depending on n) such that \(\forall \mu \le \lambda \):
Therefore we get, for fixed n, there exists \(\lambda <0\)(depending on n) such that
Consequently we can define
We claim that
To prove the claim, we let \(\psi _{n}(t,\theta )=\omega _{n}(2\lambda _{n}-t,\theta )-\omega _{n}(t,\theta )\). Hence by (37) we get \(\psi _{n}\le 0\). By using assumption (2) we obtain
for \(\forall \xi \in \mathbb R\). By virtue of (39) we have
for \((t,\theta )\in [\lambda _{n},0]\times [0,\pi ]\). Suppose by the contradiction that
By the strong maximum principle, Hopf Lemma and a result in Appendix we have \(\psi _{n}(t,\theta )<0\) in \((\lambda _{n},0)\times [0,\pi ]\) and \(\frac{\partial \psi _{n}(t,\theta )}{\partial t}|_{t=\lambda _n}<0\) for \(\theta \in [0,\pi ]\). On the other hand, from the definition of \(\lambda _{n}\), there exists a sequence \(\lambda _{n,k}\rightarrow \lambda _{n}\), as \(k\rightarrow +\infty ,\) such that
Set \(x_{k}\) is the maximum point of \(\omega _{n}(2\lambda _{n,k}-t,\theta )-\omega _{n}(t,\theta )\) in \([\lambda _{n,k},0]\times [0,\pi ]\). From continuity, we have \(x_{k}\rightarrow x_{0}\) and \(x_{0}\) lies on \(\{\lambda _{n}\}\times [0,\pi ]\). In addition, we have \(\frac{\partial \psi _{n}(t,\theta )}{\partial t}|_{x_{0}} = 0\). Thus we get a contradiction. So we arrive to the conclusion (38).
Next we want to estimate \(\lambda _{n}\). To this purpose, let us note that \(v_{n}\) satisfies
and
and
Thus in view of (42) and Lemma 3.2 we have
In order to proceed further, we distinguish two cases.
Case 1. (43) holds, and necessarily \(u_{n}(0)\rightarrow +\infty \).
In this case, we set
We also set \(\xi _{n}(x)=u_{n}(\varepsilon _{n}x)+(\alpha +1)\log \varepsilon _{n}\). Then in \(\bar{B}^+_{2\varepsilon _{0}\frac{|x_{n}|}{\varepsilon _{n}}}\), \(\xi _{n}\) satisfies
and \(\xi _{n}(0)=0\). In addition, in view of (43), we have
Therefore we argue as in Lemma 3.2 and Remark 3.3 to conclude that
uniformly in \( C^2_{loc}({\mathbb R}^2_{+})\cap C^1_{loc}(\bar{\mathbb R}^2_{+}\backslash \{0\})\cap C^0_{loc}(\bar{\mathbb R}^2_{+})\), where \(\xi \) takes the form (26) and satisfies \(\xi (0)=0\).
Claim.
with \(\varepsilon _{n}=e^{-\frac{u_{n}(0)}{\alpha +1}}\) and
as \(n\rightarrow +\infty \), and then we obtain part (b) of our Proposition.
To establish this Claim, recalling (45), we may use Case (1) to obtain
uniformly in \( C^2_{loc}({\mathbb R}^2_{+})\cap C^1_{loc}(\bar{\mathbb R}^2_{+}\backslash \{0\})\cap C^0_{loc}(\bar{\mathbb R}^2_{+})\), with \(\lambda >0\), \(y_{0}\in C\) such that \(2(\alpha +1)\lambda ^{\alpha +1}=|y_{0}|^2+\lambda ^{2(\alpha +1)}\).
For \((t,\theta )\in Q\), let
Set \(y_{0}=(|y_{0}|\cos \theta _{0}, |y_{0}|\sin \theta _{0})\) and \(\tau =(\frac{1}{2(\alpha +1)\lambda ^{\alpha +1}})^{\frac{1}{\alpha +1}}\). Since
we have
and further have
Then by a direct computation we can obtain
This implies \(\omega (\log \frac{1}{\sqrt{\tau }}-t,\theta )=\omega (\log \frac{1}{\sqrt{\tau }}+t,\theta )\) for \((t,\theta )\in Q\) and \(\omega (t,\theta )\) is symmetric with respect to \( t=\log \frac{1}{\sqrt{\tau }}\), \(\tau =(\frac{1}{2(\alpha +1)\lambda ^{\alpha +1}})^{\frac{1}{\alpha +1}}\). On the other hand, if we let \(t_{1}<t_{2}< \log \frac{1}{\sqrt{\tau }}\), then we have
Furthermore we have
and
Then by a direct calculation we can get
This implies that \(\omega (t,\theta )\) is increasing for \( t<\log \frac{1}{\sqrt{\tau }}\) and then attain its maximum at \( t=\log \frac{1}{\sqrt{\tau }}\).
By the definition of \(\omega (t,\theta )\), we have
In addition, by (35),
Then in view of (48), (49), for every fixed \(s\in \mathbb R\) we have
From (51) and \(\omega (t,\theta )\) attain its maximum at \( t=\log \frac{1}{\sqrt{\tau }}\), for large n, we have
and
By (53), we see that for large n, if we set \(\lambda =\log \varepsilon _{n}+\log \frac{1}{\sqrt{\tau }}+2\) and \(t=\log \varepsilon _{n}+\log \frac{1}{\sqrt{\tau }}+4\), (37) fails to hold. As a consequence, (46) follows. Hence using (46), (50), (52) for large n, we can estimate
Then in view of (35), (38), we have
Case 2. (44) holds.
In this case, duo to the assumption (i) we have firstly
Suppose that along a subsequence,
Therefore \(v_{n}\) admits a blow-up point \(x_{0}\). Then we apply Theorem 1.1 to \(v_{n}\) to get that \(v_n\) must verify alternative (iii) in Theorem 1.1. Moreover by (44), \(0 \notin S\). Consequently,
We choose \(s_{0}\) small enough such that \(x_{0}\) is an only blow-up point for \(v_{n}\) in \({B}_{s_{0}}(x_{0})\cap \bar{\mathbb R}^2_+\).
If \(x_0\in \mathbb R^2_+\), we can choose \(s_0\) small enough such that \({B}_{s_{0}}(x_{0})\subset \mathbb R^2_+\). Let \(y_{n}\in \bar{B}_{s_{0}}(x_{0})\), and \( v_{n}(y_{n})=\max \limits _{\bar{B}_{s_{0}}(x_{0})}v_{n}. \) Then \(y_{n}\rightarrow x_{0}\) and \(v_{n}(y_{n})\rightarrow +\infty \). Set
Then we have
with the energy condition
where \(U_{n}(x)=|y_{n}+\delta _{n}x|^{2\alpha }V_{n}(|x_{n}|y_{n}+|x_{n}|\delta _{n}x)\rightarrow 1\) in \(B_L(0)\) for all \(L>0\). Then along a subsequence, by the classification results in [6] we have
Now we need consider the following two situations.
If \(x_0\in \partial \mathbb R^2_+\), then \({B}_{s_{0}}(x_{0})\cap \mathbb R^2_+=B^+_{s_{0}}(x_{0})\). Let \(y_{n}\in \bar{B}^+_{s_{0}}(x_{0})\), and \( v_{n}(y_{n})=\max \limits _{\bar{B}^+_{s_{0}}(x_{0})}v_{n}. \) Denote \(y_{n}=(y_{n,1},y_{n,2})\). Then \(y_{n}\rightarrow x_{0}\) and \(v_{n}(y_{n})\rightarrow +\infty \). Set
Then we have
with the energy coniditions
where \(U_{n}(x)=|y_{n}+\delta _{n}x|^{2\alpha }V_{n}(|x_{n}|y_{n}+|x_{n}|\delta _{n}x)\rightarrow 1\) in \(B^+_S\) and \(H_{n}(x)=|y_{n}+\delta _{n}x|^{\alpha }h_{n}(|x_{n}|y_{n}+|x_{n}|\delta _{n}x)\rightarrow 1\) on \(L_S \) for all \(S>0\). Now we need consider the following two situations.
(1): \(\frac{|y_{n,2}|}{\delta _{n}}\rightarrow +\infty \). Then along a subsequence, by the classification results in [6] we have
with \(\xi (0)=\max \limits _{\mathbb R^2} \xi =0\).
(2): \(\frac{|y_{n,2}|}{\delta _{n}}\rightarrow \Lambda < +\infty \). Also by the classification results in [11] we have
with \(\xi (0)=\max \limits _{\bar{\mathbb R}^2_{-\Lambda }} \xi =0\).
Claim.
To establish this claim, we first notice that (54). By using the convergence properties (56), (58) and (59), we have for suitable small \(\sigma >0\) and n large,
Let \(\rho _{n}\in (0,+\infty )\) and \(\theta _{n}\in [0,2\pi ]\) be the polar coordinate for \(y_{n}\), i.e.
Since \(y_{n}\rightarrow x_{0}\) and \(|x_{0}|=1\), we have \(\rho _{n}\rightarrow 1\) as \(n \rightarrow +\infty \). Recalling the definition of \(\omega _{n}\), we have for all \(s>0\)
Thus we obtain
and
Since \(\delta _{n}\rightarrow 0\), then for n large, we can use (61) to obtain
Consequently, for \(\theta =\theta _{n}\), when \(\lambda = \log |x_{n}|+\log \rho _{n}+\log (1+\delta _{n})\) and \(t=\log |x_{n}|+\log \rho _{n}+2\log (1+\delta _{n})\), (37) fails to hold. And (60) is established.
for suitable constant \(r_{0}>0\). The Proposition is completely established. \(\square \)
We shall need the following version of Proposition 4.1.
Corollary 4.2
Under the assumptions of Proposition 4.1, for every \(r\in (0,1]\), we have
for suitable \(r_{0}>0\) and C.
Proof
For \(r\in (0,1)\), in \( \bar{B}^+_{\frac{1}{r}}\) we define
Then \(u_{n,r}\) satisfies
where \(V_{n,r}(x)=V_{n}(rx)\) and \(h_{n,r}(x)=h_{n}(rx)\). Notice that
and \(V_{n,r}(x)\) and \(h_{n,r}(x)\) still satisfy (2) in \(\bar{B}^+_{1}\). If we set \(x_{n,r}=\frac{x_{n}}{r}\) and
then by applying Proposition 4.1 to \(u_{n,r}(x)\) and \(v_{n,r}(x)\) we conclude that
So when case (a) holds, we have
When case (b) holds, then
\(\square \)
As already mentioned, Proposition 4.1 play a crucial role in proving Theorem 1.2 as it also implies the following result.
Corollary 4.3
In addition to the assumptions of Proposition 4.1, we suppose further that
with
for \(\gamma >0\) suitable constant. Then along a subsequence,
Proof
For given \(r\in (\delta _{n} , r_{n})\), define \(u_{n,r}\) as in (63). So
And by (64) we have
Thus we can use Harnack inequality to conclude that there exists a constant \(\beta \in (0,1)\) such that
According to Corollary 4.2, we must treat two situations.
For case (54), i.e.
and Corollary 4.2 implies that
Hence if we insert (68) in (66) we obtain
Similarly,
In case (43), we have
and Corollary 4.2 implies that
Inserting (70) in (66), we obtain
Similarly,
Thus we achieve the proof of this Corollary. \(\square \)
5 Proof of Theorem 1.2
Now we come to prove Theorem 1.2.
Proof of theorem 1.2
We will consider separately alternatives (i) and (ii) for \(u_{n}\) in Proposition 3.4. Firstly we consider the case where alternative (i) holds in Proposition 3.4. We have the following Claim.
Claim 1
If (27) holds, then \(m(0)= 2\pi (1+\alpha )\).
Note that by Lemma 3.2, the validity of (27) implies
So we set
and as in Lemma 3.2 and Remark 3.3, along a subsequence,
here \(\xi \) described in (26).
Since
we can find \(R_{n}\rightarrow +\infty \) such that, along a subsequence,
For every \(r\in (R_{n}\varepsilon _{n}, \varepsilon _{0})\), we can also apply Harnack inequality as in the proof of Corollary 4.3 to derive
with \(\beta \in (0,1)\).
Moreover by the alternative (b) in Proposition 4.1 and Corollary 4.2 we derive that
Combine (72) with (71), we get the estimate
on \(S_r^+\). Consequently,
So we have
Hence by (4), letting \(n\rightarrow +\infty \), the desired conclusion follows.
We are left to treat the case where alternative (ii) holds in Proposition 3.4. In this case, we can apply Corollary 4.3 and derive
And similarly for \(m\ge 2\),
as \( n\rightarrow +\infty \) for \( j=1,\ldots ,m-1\). Consequently,
as \(n \rightarrow +\infty \). Set
And set
Then we see that
with the energy conditions
where \(V_{j,n}(x)=|x|^{2\alpha }V_{n}(|x_{j,n}|x)\) and \(h_{j,n}(x)=|x|^{\alpha }h_{n}(|x_{j,n}|x)\) satisfy
Now we set
So that by (74), we have
Claim 2
Either \( \beta _{0}=0\) or \(\beta _{0}=2\pi (1+\alpha )\).
In fact, by Proposition 4.1, we see that either \(\max \limits _{\bar{B}^+_{2\varepsilon _{0}}} v_{1,n}\rightarrow -\infty \) or \(v_{1,n}(0)\rightarrow +\infty \). If \(\max \limits _{\bar{B}^+_{2\varepsilon _{0}}} v_{1,n}\rightarrow -\infty \), then \(\beta _{0}=0\) in this case. If \(v_{1,n}(0)\rightarrow +\infty \), we see that 0 is the only blow-up point of \(v_{1,n}\) in \( \bar{B}^+_{2\varepsilon _{0}}\) since \(\sup \limits _{\bar{B}^+_{2\varepsilon _{0}}} \{ v_{1,n}+(\alpha +1)\log |x| \}\le C\). We can apply Theorem 1.1 and conclude that \( \lim \limits _{n\rightarrow +\infty } (\int _{B^+_{2\varepsilon _{0}}}V_{1,n}e^{2v_{1,n}}+\int _{L_{2\varepsilon _{0}}}h_{1,n}e^{v_{1,n}})=\beta _{0}\). Furthermore, since \(\sup \limits _{\bar{B}^+_{2\varepsilon _{0}}} \{ v_{1,n}+(\alpha +1)\log |x| \}\le C\) we can use Claim 1 above for \(v_{1,n}\) and obtain \(\beta _{0}=2\pi (1+\alpha )\) in this case.
Claim 3
\(\beta _{j}\in 2\pi \mathbb N, \quad j=1,2,\ldots ,m\).
In fact, (28) implies
Therefore, the blow-up set \(S_{j}\) of \(v_{j,n}\) is nonempty and satisfy:
At this point, we are in position to apply Li–Shafrir and Zhang–Zhou–Zhou (see [10, 18]) results around each point \(S_{j}\) and derive \(\beta _{j}\in 2\pi \mathbb N\), \(\forall j=1,\ldots ,m\).
Thus by the Claims above, Theorem 1.2 is completely established. \(\square \)
6 Proof of Theorem 1.3
In this section, we will obtain the precise blow-up value at the singular blow-up point when \(u_{n}\) have the following mild boundary condition:
As we shall see, the behavior of \(u_{n}\) around the blow-up point 0 is very seriously affected by the validity of (76). Now we give the proof of Theorem 1.3.
Proof of Theorem 1.3
Let \(p_{n}\) satisfy
From (76), the maximum principle and Hopf Lemma we have
Set \(w_{n}=u_{n}-\min \limits _{S^+_{1}} u_{n}-p_{n}\), then \(w_{n}\) satisfies
where \(W_{n}(x)=e^{\min \limits _{S^+_{1}} 2u_{n}+2p_{n}}\) and \(G_{n}(x)=e^{\min \limits _{S^+_{1}} u_{n}+p_{n}}\). In addition, by (4) we have
By Green’s representation formula,
where R(x, y) is the regular part of the Green function. Passing to the limit in (80), we have
Set \(g(x)=\beta R(x,0)\in C^{1}(\bar{B}^+_{1})\) and
By the Pohozaev identity for \(w_{n}\) in \(\bar{B}^+_{r}\), we have
Let \(n\rightarrow \infty \) in (83), and using (79) and (81) we find the identity:
Inserting (82) into (84), we obtain
Letting \(r\rightarrow 0\), then we have \(m(0)=2\pi (1+\alpha )\). \(\square \)
References
Brezis, H., Li, Y.Y., Shafrir, I.: A sup+inf inequality for some nonlinear elliptic equations involving exponential nonlinearities. J. Funct. Anal. 115, 344–358 (1993)
Brezis, H., Merle, F.: Uniform estimates and blow-up behavior for solutions of \(-\Delta u=V(x)e^{u }\) in two dimensions. Comm. Partial Differ. Equ. 16, 1223–1253 (1991)
Bartolucci, D., Montefusco, E.: Blow-up analysis, existence and qualitative properties of solutions for the two-dimensional Emden–Fowler equation with sigular potential. Math. Methods Appl. Sci. 30, 2309–2327 (2007)
Bartolucci, D., Tarantello, G.: Liouville type equations with singular data and their applications to periodic multivortices for the electroweak theory. Commun. Math. Phys. 229, 3–47 (2002)
Bao, J., Wang, L.H., Zhou, C.Q.: Blow up analysis for solutions to Neumann boundary value problem. J. Math. Anal. Appl. 418, 142–162 (2014)
Chen, W., Li, C.: Classification of solutions of some nonlinear elliptic equations. Duke Math. J. 63, 615–623 (1991)
Guo, Y.X., Liu, J.Q.: Blow-up analysis for solutions of the laplacian equation with exponential Neumann boundary condition in dimension two. Commun. Contemp. Math. 8, 737–761 (2006)
Jost, J., Wang, G., Zhou, C.O.: Metrics of constant curvature on a Riemann surface with two corners on the boundary. Ann. I.H. Poincare-AN 26, 437–456 (2009)
Jost, J., Wang, G., Zhou, C.Q., Zhu, M.M.: The boundary value problem for the super-Liouville equation. Ann. I.H. Poincare-AN 31, 685–706 (2014)
Li, Y.Y., Shafrir, I.: Blow-up analysis for solutions of \(-\Delta u=Ve^u\) in dimension two. Indiana Univ. Math. J. 43, 1255–1270 (1994)
Li, Y.Y., Zhu, M.J.: Uniqueness theorems through the method of moving spheres. Duke Math. J. 80, 383–417 (1995)
Poliakovsky, A., Tarantello, G.: On a plannar Liouville-type problem in the study of selfgravitating strings. J. Differential Equ. 252, 3668–3693 (2012)
Tarantello, G.: A quantization property for blow up solutions of singular Liouville-type equations. J. Funct. Anal. 219, 368–399 (2005)
Tarantello, G.: Multiple condensate solutions for the Chern–Simons–Higgs theory. J. Math. Phys. 37, 3769–3796 (1996)
Tarantello, G.: Self-Dual Gauge Field Vortices, an Analytical Approach, PNLDE, vol. 72. Birkhauser Boston Inc, Boston, MA (2007)
Tarantello, G.: Blow-up analysis for a cosmic string equation. J. Funct. Anal. 272(1), 255–338 (2017)
Yang, Y.: Solitons in Field Theory and Nonlinear Analysis. Springer Monographs in Mathematics, Springer, New York (2001)
Zhang, T., Zhou, C.L., Zhou, C.Q.: Quantization of the blow-up value for the Liouville equation with exponential Neumann boundary condition. Commun. Math. Stat. 6, 29–48 (2018)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by J. Jost.
The authors are supported partially by NSFC of China (No. 11771285).
Appendix
Appendix
Identifying \(x=(x_{1},x_{2})\in \mathbb R^2\). Suppose \(\Omega =(a,0)\times (0,b)\) with \(a<0\), \(b>0\) is an open rectangle in \(\mathbb R^2\). Let \(x_{0}=(a,b)\). Suppose u satisfy
Then we have \(\frac{\partial u}{\partial x_{1}}|_{x_{0}}<0\).
Proof
Firstly we choose a point \(y\in (a,0) \times \{ b \}\) which is the center of the circle whose radius is \(|x_{0}-y|=R\). And let \(R<\min \{b,\frac{|a|}{2}\}\). Set \(B_{R}(y)\cap \Omega =B^-_{R}(y)\). For \(0<\rho <R\), we introduce an auxiliary function v by defining
where \(r=|x-y|>\rho \) and \(\gamma \) is a positive constant yet to be determined. Direct calculation gives
Hence \(\gamma \) may be chosen large enough so that \(\Delta v\ge 0\) throughout the annular region \(A=B^-_{R}(y)-B^-_{\rho }(y)\). By the strong maximum principle and Hopf Lemma we have \(u(x)<0\) in \(\Omega \) and \(u(x)<0\) in \((a,0)\times \{b\}\). Since \(u-u(x_{0})<0\) on \(\partial B^-_{\rho }(y)\cap \{ 0<x_{2}<b \}\), there is a constant \(\varepsilon >0\) small enough for which \(u-u(x_{0})+\varepsilon v\le 0\) on \(\partial B^-_{\rho }(y)\cap \{ 0<x_{2}<b \}\). This inequality is also satisfied on \(\partial B^-_{R}(y)\cap \{ 0<x_{2}<b \}\). Suppose there exists a point \(y_{1}\) on \(\partial B^-_{R}(y)\backslash \partial B^-_{\rho }(y) \cap \{ x_{2}=b \}\) satisfies \(u(y_{1})-u(x_{0})+\varepsilon v(y_{1})= \max \limits _{\partial B^-_{R}(y)\backslash \partial B^-_{\rho }(y) \cap \{ x_{2}=b \}} (u-u(x_{0})+\varepsilon v)\). Since \(\Delta (u-u(x_{0})+\varepsilon v)\ge 0\) in A, so we have \(\frac{\partial (u-u(x_{0})+\varepsilon v)}{\partial x_{2}}|_{y_{1}}>0\) by Hopf Lemma. But since \(\frac{\partial v }{\partial x_{2}}|_{x_{2}=b}= 0\) then we have \(\frac{\partial (u-u(x_{0})+\varepsilon v) }{\partial x_{2}}|_{x_{2}=b}\le 0\), which is the desired contradiction. The weak maximum principle implies that \(u-u(x_{0})+\varepsilon v\le 0\) throughout A. Then we have
\(\square \)
Rights and permissions
About this article
Cite this article
Zhang, T., Zhou, C. Liouville type equation with exponential Neumann boundary condition and with singular data. Calc. Var. 57, 163 (2018). https://doi.org/10.1007/s00526-018-1442-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00526-018-1442-7