Introduction and main theorems
Li–Mochizuki [43 ] established a theorem stating that for any r r r -differential q ∈ H 0 ( K X r ) q \in H^0(K_X^r) q ∈ H 0 ( K X r ) with r ≥ 2 r \ge2 r ≥ 2 , which is non-zero unless X X X is hyperbolic, there exists a unique complete harmonic metric with determinant 1 1 1 on the cyclic Higgs bundle associated with q q q . The purpose of this paper is to extend this theorem to complete solutions of a variant of the Hitchin equation associated with more general subharmonic weight functions, and to establish an approximation theorem for them. The objective of this section is to give precise statements of our main theorems. Before presenting the main theorems, we explain the background of cyclic Higgs bundles, harmonic metrics, and their extensions using subharmonic weight functions.
Let X X X be a Riemann surface, K X → X K_X \to X K X → X its canonical bundle, and T X : = K X − 1 → X T_X := K_X^{-1} \to X T X := K X − 1 → X the holomorphic tangent bundle. For each integer r ≥ 2 r \ge2 r ≥ 2 and a choice of a square root K X 1 / 2 K_X^{1/2} K X 1/2 of the canonical bundle, Hitchin [35 , 36 ] introduced a holomorphic vector bundle K r \mathbb{K}_r K r of rank r r r , defined by
K r : = ⨁ j = 1 r K X r − ( 2 j − 1 ) 2 = K X r − 1 2 ⊕ K X r − 3 2 ⊕ ⋯ ⊕ K X − r − 3 2 ⊕ K X − r − 1 2 . \mathbb{K}_r := \bigoplus_{j=1}^{r} K_X^{\frac{r-(2j-1)}{2}} = K_X^{\frac{r-1}{2}} \oplus K_X^{\frac{r-3}{2}} \oplus\cdots\oplus K_X^{-\frac{r-3}{2}} \oplus K_X^{-\frac{r-1}{2}}. K r := j = 1 ⨁ r K X 2 r − ( 2 j − 1 ) = K X 2 r − 1 ⊕ K X 2 r − 3 ⊕ ⋯ ⊕ K X − 2 r − 3 ⊕ K X − 2 r − 1 .
The definition of K r \mathbb{K}_r K r arises from the Weyl vector of type A r − 1 A_{r-1} A r − 1 . For each q ∈ H 0 ( K X r ) q \in H^0(K_X^r) q ∈ H 0 ( K X r ) , one associates a holomorphic section Φ ( q ) ∈ H 0 ( End K r ⊗ K X ) \Phi(q) \in H^0(\operatorname{End}\mathbb{K}_r \otimes K_X) Φ ( q ) ∈ H 0 ( End K r ⊗ K X ) , called a Higgs field, given by
Φ ( q ) : = ( 0 q 1 ⋱ ⋱ ⋱ 1 0 ) . \Phi(q) := \begin{pmatrix}
0 & & & q \\
1 & \ddots& & \\
& \ddots& \ddots& \\
& & 1 & 0
\end{pmatrix}. Φ ( q ) := 0 1 ⋱ ⋱ ⋱ 1 q 0 .
The pair ( K r , Φ ( q ) ) → X (\mathbb{K}_r,\Phi(q)) \to X ( K r , Φ ( q )) → X is called a cyclic Higgs bundle (cf. [4 , 35 , 36 ] ). A diagonal Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on K r \mathbb{K}_r K r induces Hermitian metrics H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 on T X → X T_X \to X T X → X , defined by H j : = h j − 1 ⊗ h j + 1 H_j := h_j^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 for j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 . Moreover, h 1 h_1 h 1 , h r h_r h r , and a holomorphic r r r -differential q q q together determine a degenerate Hermitian metric H r H_r H r on T X → X T_X \to X T X → X , defined by H r : = h r − 1 ⊗ h 1 ⊗ h q H_r := h_r^{-1}\otimes h_1\otimes h_q H r := h r − 1 ⊗ h 1 ⊗ h q , where h q h_q h q is a degenerate Hermitian metric on K X − r K_X^{-r} K X − r given by
h q ( u , v ) : = ⟨ q x , u ⟩ ⟨ q x , v ⟩ ‾ for x ∈ X , u , v ∈ ( K X − r ) x . h_q(u,v) := \langle q_x,u\rangle\overline{\langle q_x,v\rangle}\quad\text{for }x\in X,\ u,v\in(K_X^{-r})_x. h q ( u , v ) := ⟨ q x , u ⟩ ⟨ q x , v ⟩ for x ∈ X , u , v ∈ ( K X − r ) x .
A diagonal Hermitian metric is said to be complete if the Kähler metrics induced by H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 are all complete. A Hermitian metric on ( K r , Φ ( q ) ) → X (\mathbb{K}_r,\Phi(q)) \to X ( K r , Φ ( q )) → X is said to be a harmonic metric if it solves the Hitchin equation. Li–Mochizuki [43 ] proved that for any Riemann surface X X X and any holomorphic section q q q that is non-zero unless X X X is hyperbolic, there exists a unique complete harmonic metric h h h on ( K r , Φ ( q ) ) → X (\mathbb{K}_r,\Phi(q)) \to X ( K r , Φ ( q )) → X with det ( h ) = 1 \det(h)=1 det ( h ) = 1 .
There are two extreme cases in which the Hitchin equation admits explicit solutions: the case where the holomorphic section q q q has no zeros, and the case where q q q is identically zero. If q q q has no zeros, then the diagonal metric h h h constructed from H 1 = ⋯ = H r − 1 = h q 1 / r H_1=\cdots=H_{r-1}=h_q^{1/r} H 1 = ⋯ = H r − 1 = h q 1/ r by the inverse change of variables is a harmonic metric. In particular, if the metric h q 1 / r h_q^{1/r} h q 1/ r is complete, then the harmonic metric h h h is also complete. By contrast, suppose that q q q is identically zero and that there exists a Kähler metric of Gaussian curvature − 2 -2 − 2 on X X X . Let H H H be the Hermitian metric on K X − 1 → X K_X^{-1}\to X K X − 1 → X induced by this Kähler metric. Then the diagonal Hermitian metric h h h on K r \mathbb{K}_r K r constructed from the Hermitian metrics H j : = j ( r − j ) H H_j:=j(r-j)H H j := j ( r − j ) H (j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 ) by the inverse change of variables is a harmonic metric, and it is complete whenever the Kähler metric is complete. From this perspective, complete harmonic metrics on cyclic Higgs bundles may be viewed as interpolating between complete flat Kähler metrics and complete hyperbolic metrics. The zeros of q q q bend a flat metric, while the metric approaches the hyperbolic one in the limiting case q = 0 q=0 q = 0 .
The section q q q induces a semipositive singular Hermitian metric h q − 1 / r = e − 1 r log ∣ q ∣ h ∗ 2 h ∗ h_q^{-1/r}=e^{-\frac{1}{r}\log|q|_{h_*}^2}h_* h q − 1/ r = e − r 1 l o g ∣ q ∣ h ∗ 2 h ∗ on K X K_X K X , and the associated diagonal harmonic metric depends only on this metric. When we consider the harmonic metric as something intermediate between a flat Kähler metric and a Kähler metric with negative constant curvature, rather than as something that produces a harmonic map, and as an object of study in complex analysis, it is therefore natural to extend the theory to a general semipositive singular Hermitian metric e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ , moving beyond the classical setting of holomorphic differentials. Since subharmonic functions can be approximated by logarithms of absolute values of holomorphic functions, this extension may be interpreted as studying Higgs fields with cyclic multivalency and the associated harmonic metrics on ramified coverings, as well as limits of these metrics as the covering degrees tend to infinity.
The main significance of extending the theory of complete solutions to general semipositive singular Hermitian metrics lies in the construction of entropy and free energy , two functions introduced in [50 , 51 ] , and in their application to the approximation theory of equilibrium metrics . The metrics arising in this approximation theory need not be induced by holomorphic differentials, and their weight functions need not be smooth. Studying entropy and free energy in this setting therefore calls for a theory of complete solutions associated with general subharmonic weight functions. Our main theorems provide an analytic foundation for studying approximation of equilibrium metrics through entropy and free energy. We briefly recall the definitions of these functions in § 3.4 ; their applications to approximation of equilibrium metrics will be developed in [52 ] .
In this paper, we study the existence and uniqueness of complete solutions associated with more general subharmonic weight functions. Throughout this paper, when we say that “h h h is a complete solution associated with φ \varphi φ ,” we mean that h h h is a solution to the extended Hitchin equation (5) in § 3.3 associated with φ \varphi φ , and that it is complete in the sense defined in § 3.1 . Our main theorems are as follows:
Theorem 1. Let h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) and h ′ = ( h 1 ′ , … , h r ′ ) h'=(h'_1,\ldots,h'_r) h ′ = ( h 1 ′ , … , h r ′ ) be complete solutions associated with φ \varphi φ satisfying det ( h ) = det ( h ′ ) = 1 \det(h)=\det(h')=1 det ( h ) = det ( h ′ ) = 1 . Then h = h ′ h=h' h = h ′ .
Theorem 2. Let φ \varphi φ be a subharmonic weight function on X X X that is not identically − ∞ -\infty − ∞ unless X X X is hyperbolic. Then there exists a real complete solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) associated with φ \varphi φ . Moreover, if X X X is the complex plane C \mathbb{C} C , h h h satisfies the following comparison property at infinity:
For each R > 0 R>0 R > 0 , let D R \mathbb{D}_R D R be the open ball of radius R R R and h > R h^{>R} h > R the unique complete solution associated with φ ∣ C ∖ D ‾ R \varphi|_{\mathbb{C}\setminus\overline{\mathbb{D}}_R} φ ∣ C ∖ D R satisfying det ( h > R ) = 1 \det(h^{>R})=1 det ( h > R ) = 1 . Then for any R 1 > R 2 > 0 R_1>R_2>0 R 1 > R 2 > 0 , h h h is mutually bounded with h > R 2 h^{>R_2} h > R 2 on C ∖ D R 1 \mathbb{C}\setminus\mathbb{D}_{R_1} C ∖ D R 1 .
Theorem 3. Let X X X be either the complex plane C \mathbb{C} C or the unit disc D \mathbb{D} D , and let φ \varphi φ be a subharmonic weight function on K X K_X K X . Assume that φ ≢ − ∞ \varphi\not\equiv-\infty φ ≡ − ∞ if X = C X=\mathbb{C} X = C . Let ( φ ϵ ) 0 < ϵ < 1 (\varphi_\epsilon)_{0<\epsilon<1} ( φ ϵ ) 0 < ϵ < 1 be a family of smooth subharmonic weight functions, where each φ ϵ \varphi_\epsilon φ ϵ is defined on C \mathbb{C} C in the complex plane case and on the disc D ϵ : = { z ∈ C ∣ ∣ z ∣ < 1 − ϵ } \mathbb{D}_\epsilon:=\{z\in\mathbb{C}\mid|z|<1-\epsilon\} D ϵ := { z ∈ C ∣ ∣ z ∣ < 1 − ϵ } in the unit disc case. Suppose that φ ϵ ′ ≤ φ ϵ \varphi_\epsilon'\leq\varphi_\epsilon φ ϵ ′ ≤ φ ϵ on the domain of φ ϵ \varphi_\epsilon φ ϵ whenever 0 < ϵ ′ < ϵ < 1 0<\epsilon'<\epsilon<1 0 < ϵ ′ < ϵ < 1 , and that φ ϵ \varphi_\epsilon φ ϵ converges pointwise to φ \varphi φ as ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 . Then the corresponding complete solutions h ϵ = ( h ϵ , 1 , … , h ϵ , r ) h_\epsilon=(h_{\epsilon,1},\ldots,h_{\epsilon,r}) h ϵ = ( h ϵ , 1 , … , h ϵ , r ) , normalized by det ( h ϵ ) = 1 \det(h_\epsilon)=1 det ( h ϵ ) = 1 , converge monotonically, as ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 , to the unique complete solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) associated with φ \varphi φ and satisfying det ( h ) = 1 \det(h)=1 det ( h ) = 1 .
The proof of Theorems 1 and 2 follows the general strategy of [43 ] ; however, because issues regarding the regularity of e φ e^\varphi e φ prevent the arguments in [43 ] from being applied directly, some adjustments and modifications are required. In particular, since the Omori–Yau maximum principle and the Cheng–Yau maximum principle, which play a crucial role in [43 ] , cannot be applied in our setting, we provide alternative tools for this purpose in § 4.2 . Also, the existence of a complete solution (Theorem 2 ) is not proven directly for the general case; instead, it is first established for the case where e φ e^{\varphi} e φ possesses good regularity, and then extended to the more general case using Theorem 3 .
Approximating families satisfying the assumptions of Theorem 3 can always be constructed by the standard theory of mollifications (cf. [31 ] ). The proof of Theorem 3 draws on ideas from [43 Proposition 4.6] . In addition to its role in the proof of Theorem 2 , this approximation theorem is used in [50 , 51 ] to establish results on entropy and free energy.
Standard local regularity theory for the Poisson equation (cf. [49 ] ) implies that the solution h h h in Theorem 2 is of class C l o c 1 , α C_{\mathrm{loc}}^{1,\alpha} C loc 1 , α for every 0 < α < 1 0<\alpha<1 0 < α < 1 . In fact, the regularity of h h h can be further improved to C l o c 1 , 1 C_{\mathrm{loc}}^{1,1} C loc 1 , 1 , and it is of class C 2 C^2 C 2 if and only if e φ e^{\varphi} e φ is continuous. These regularity properties are discussed in Appendix § 6 .
This paper is structured as follows: Section 2 provides background. Section 3 is devoted to the proof of the main theorems.
AI Declaration
This paper combines two preprints, arXiv:2508.12848 and arXiv:2610.23441v1, and includes a newly added appendix. In preparing arXiv:2508.12848, I developed all the mathematical arguments myself, without any AI involvement. Conversations with ChatGPT were used solely to improve the English. In preparing arXiv:2610.23441v1, I also discussed mathematical arguments with ChatGPT (OpenAI; GPT-5.6 Sol, GPT-6.1 Sol, and GPT-6 Astra models, accessed through a ChatGPT Plus subscription). The Appendix was prepared with AI assistance, but it does not affect the substance of the main theorems. As can be seen from arXiv:2610.23441v1, every use of the propositions in the Appendix within the main text can be replaced by simpler arguments based on standard results for the Poisson equation. Below, I describe in detail how AI was used in preparing arXiv:2610.23441v1, using the numbering of the present paper. I divide the aspects of the work in which AI assistance was used into the following four categories. I note at the outset that only the contribution described in item 1 is essential to the proofs of the main theorems. In items 2 and 3, AI played an essential role in developing the respective propositions. However, by weakening the statements of these propositions or instead invoking results used in Li–Mochizuki’s paper, one can complete the proofs of the main theorems without the parts of these arguments to which AI contributed. My reason for seeking AI assistance in items 2 and 3 was not that it was indispensable to proving the main theorems, but that I wished to formulate these propositions in a more complete form so that I could cite them in subsequent work. In item 4, I developed the underlying argument myself, and AI identified gaps in it during a subsequent check. I now describe items 1–4 in detail.
I used AI assistance in proving Proposition 24. The specific assistance received and the history leading to the completed proof are as follows. In [43 ] , Li–Mochizuki proved uniqueness by combining the Cheng–Yau and Omori–Yau maximum principles on open Riemann surfaces with a priori estimates derived from Simpson’s inequality. For the equation considered here, however, the versions of these maximum principles used in [43 ] require more regularity than is available for the relevant functions when establishing uniqueness in full generality. I therefore introduced Propositions 22 and 24 as substitutes for the Omori–Yau and Cheng–Yau maximum principles, respectively. In arXiv:2508.12848, I had used extensions of these principles to manifolds with boundary, but the resulting uniqueness theorem required additional assumptions. I proved Proposition 22 without AI assistance, whereas the proof of
Proposition 24 involved AI assistance.
The starting point for Proposition 24 was [11 Theorem 8.20] , and my basic strategy was to extend that result. The initial version of Proposition 24 , which I proved without AI assistance, asserted, roughly speaking, that an upper semicontinuous function u u u is bounded above if there exists a continuous auxiliary function g g g to which u u u is close in a suitable sense.
Consequently, the initial version of Theorem 1 established uniqueness under the assumption that e φ h ∗ − 1 ⊗ H j − 1 e^{\varphi}h_*^{-1} \otimes H_j^{-1} e φ h ∗ − 1 ⊗ H j − 1 is bounded for every j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 . I also proved that this boundedness condition holds whenever e φ h ∗ − 1 e^{\varphi}h_*^{-1} e φ h ∗ − 1 is continuous outside a compact set, so uniqueness follows in that case without any further assumptions.
I had obtained all of these results without AI assistance. At the time, I believed that extending [11 Theorem 8.20] to weak solutions would necessarily require some analogue of continuity to ensure that the relevant strict superlevel set is open at the final step of the contradiction argument. I therefore thought that the result described above might be the limit of what could be achieved by extending that theorem, and I prepared the initial draft with this weaker uniqueness statement. While I was preparing that draft, however, the GPT-6.1 Sol model in ChatGPT suggested the stronger formulation now stated as Proposition 24 . This enabled me to establish Theorem 1 in full generality.
In proving Lemma 45 , which is used in the proof of Proposition 47 , I received repeated assistance from ChatGPT in correcting gaps in my extension of the argument in [20 Proof of Lemma 1.5] . ChatGPT also suggested introducing the truncated function G j # G_j^{\#} G j # in the proof of Proposition 47 . I had not initially recognized that the issue described in Remark 48 could arise. ChatGPT pointed out the resulting gap in my initial draft and proposed this truncation argument to address it. However, this truncation argument is unnecessary if, for example, the supersolutions and subsolutions in Proposition 47 are all continuous, and this continuity assumption suffices for the applications of Proposition 47 in the proofs of the main theorems.
In the proof of Proposition 50 , I initially intended to apply Lemma 49 on Y ˉ i \bar{Y}_i Y ˉ i to show that the solution of the Dirichlet problem lies between the subsolution and the supersolution. ChatGPT pointed out that the lower Ricci curvature bound required by Lemma 49 need not hold even though Y ˉ i \bar{Y}_i Y ˉ i is compact, since the Ricci curvature in the weak sense need not be represented by a continuous function. ChatGPT also explained how the proposition can still be proved by modifying the comparison argument as illustrated in Remark 51. Note, however, that for the applications needed in the proofs of the main theorems, this proposition can be replaced by an application of [43 Proposition 5.2] in the case where e φ e^{\varphi} e φ is smooth.
ChatGPT helped me correct the following minor errors in my initial draft. It brought Ambrose’s theorem, discussed immediately before Proposition 40 , to my attention as a way to fill a gap in the original argument. The constant M φ M_{\varphi} M φ introduced at the beginning of § 4.7 was originally defined as sup X e φ \sup_X e^{\varphi} sup X e φ , but ChatGPT pointed out that this choice causes the argument to fail when φ ≡ − ∞ \varphi\equiv-\infty φ ≡ − ∞ . Finally, the constant c c c in Lemma 66 was originally defined by an equality. ChatGPT pointed out that, for the argument to work, c c c needed to be chosen strictly larger than the value specified in the initial draft.
Acknowledgements. I would like to thank Toshiaki Yachimura for helpful conversations. This work was supported by JSPS KAKENHI Grant Number JP24K16912.
Background
Real, diagonal, and complete Hermitian metrics
Let X X X be a connected, possibly non-compact Riemann surface and K X → X K_X \to X K X → X the canonical bundle. We choose a square root K X 1 / 2 K_X^{1/2} K X 1/2 of the canonical bundle K X → X K_X \to X K X → X . Let K r \mathbb{K}_r K r be a holomorphic vector bundle of rank r r r defined as follows:
K r : = ⨁ j = 1 r K X r − ( 2 j − 1 ) 2 = K X r − 1 2 ⊕ K X r − 3 2 ⊕ ⋯ ⊕ K X − r − 3 2 ⊕ K X − r − 1 2 . \mathbb{K}_r := \bigoplus_{j=1}^{r} K_X^{\frac{r-(2j-1)}{2}} = K_X^{\frac{r-1}{2}} \oplus K_X^{\frac{r-3}{2}} \oplus\cdots\oplus K_X^{-\frac{r-3}{2}} \oplus K_X^{-\frac{r-1}{2}}. K r := j = 1 ⨁ r K X 2 r − ( 2 j − 1 ) = K X 2 r − 1 ⊕ K X 2 r − 3 ⊕ ⋯ ⊕ K X − 2 r − 3 ⊕ K X − 2 r − 1 .
A Hermitian metric h h h on K r → X \mathbb{K}_r \to X K r → X is said to be real (cf. [36 ] ) if the following bundle isomorphism S S S is isometric with respect to h h h :
S : = ( 1 ⋱ 1 ) : K r → K r ∨ , S := \begin{pmatrix}
1 & & \\
& \ddots& \\
& & 1
\end{pmatrix} : \mathbb{K}_r \to\mathbb{K}_r^\vee, S := 1 ⋱ 1 : K r → K r ∨ ,
where K r ∨ \mathbb{K}_r^\vee K r ∨ denotes the dual bundle of K r \mathbb{K}_r K r . Also, h h h is called diagonal (cf. [4 , 17 , 18 ] ) if it splits as h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) . Let T X → X T_X \to X T X → X be the dual bundle of K X → X K_X \to X K X → X . By taking the difference between the adjacent components in a diagonal Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) , we obtain r − 1 r-1 r − 1 Hermitian metrics H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 on T X → X T_X \to X T X → X :
H j : = h j − 1 ⊗ h j + 1 . H_j := h_j^{-1} \otimes h_{j+1}. H j := h j − 1 ⊗ h j + 1 .
A diagonal Hermitian metric h h h is real if and only if h j = h r − j + 1 − 1 h_j=h_{r-j+1}^{-1} h j = h r − j + 1 − 1 for all j = 1 , … , r j=1,\ldots,r j = 1 , … , r . If det ( h ) = 1 \det(h)=1 det ( h ) = 1 , this is also equivalent to the metrics H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 satisfying H j = H r − j H_j=H_{r-j} H j = H r − j for all j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 . We say that a diagonal Hermitian metric h h h is complete (cf. [43 ] ) if the Kähler metrics induced by H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 are all complete.
Remark 4. Note that the notion of completeness introduced above is considerably broader than the original notion in [43 ] . In [43 ] , a reference Kähler metric is fixed, thereby allowing diagonal Hermitian metrics on K r \mathbb{K}_r K r to be identified with r r r -tuples of functions. The term “complete solution” is then used for a diagonal Hermitian metric h h h on K r \mathbb{K}_r K r , viewed as an r r r -tuple of functions, that satisfies det ( h ) = 1 \det(h)=1 det ( h ) = 1 and solves the Hitchin equation for a cyclic Higgs bundle. In contrast, we use the term “complete” for arbitrary diagonal Hermitian metrics, without requiring them to be harmonic or imposing any condition on their determinants, although the main theorem and its proof deal exclusively with metrics satisfying det ( h ) = 1 \det(h)=1 det ( h ) = 1 .
Remark 5. In this paper, we allow Kähler metrics on Riemann surfaces and Hermitian metrics on K r \mathbb{K}_r K r to be continuous but not necessarily smooth. Whenever we say that a (possibly non-smooth) Kähler metric is complete, we mean that the induced distance function defines a complete metric space.
Harmonic metrics on cyclic Higgs bundles
Let q q q be a holomorphic section of K X r → X K_X^r \to X K X r → X . Then we associate a holomorphic section Φ ( q ) ∈ H 0 ( End K X r ⊗ K X ) \Phi(q) \in H^0(\operatorname{End}K_X^r \otimes K_X) Φ ( q ) ∈ H 0 ( End K X r ⊗ K X ) , which is called a Higgs field, as follows:
Φ ( q ) : = ( 0 q 1 ⋱ ⋱ ⋱ 1 0 ) . \Phi(q) := \begin{pmatrix} 0 & & & q \\ 1 & \ddots& & \\ & \ddots& \ddots& \\ & & 1 & 0 \end{pmatrix}. Φ ( q ) := 0 1 ⋱ ⋱ ⋱ 1 q 0 .
The pair ( K X r , Φ ( q ) ) (K_X^r,\Phi(q)) ( K X r , Φ ( q )) is called a cyclic Higgs bundle (cf. [4 , 35 , 36 ] ). The above cyclic Higgs bundle is an example of cyclotomic Higgs bundles introduced in [65 ] . For each Hermitian metric h h h on K X r K_X^r K X r , we associate a connection D q ( h ) D_q(h) D q ( h ) defined as follows:
D q ( h ) : = ∇ h + Φ ( q ) + Φ ( q ) ∗ h . D_q(h) := \nabla^h + \Phi(q) + \Phi(q)^{*h}. D q ( h ) := ∇ h + Φ ( q ) + Φ ( q ) ∗ h .
A Hermitian metric on ( K X r , Φ ( q ) ) → X (K_X^r,\Phi(q)) \to X ( K X r , Φ ( q )) → X is called a harmonic metric if the connection D q ( h ) D_q(h) D q ( h ) is a flat connection. A harmonic metric h h h yields a harmonic map h ^ : X ~ → SL ( r , C ) / SU ( r ) \hat h : \widetilde X \to\operatorname{SL}(r,\mathbb{C})/\operatorname{SU}(r) h ^ : X → SL ( r , C ) / SU ( r ) , where X ~ \widetilde X X is the universal covering space. If h h h is real, then h ^ \hat h h ^ maps into SL ( r , R ) / SO ( r ) \operatorname{SL}(r,\mathbb{R})/\operatorname{SO}(r) SL ( r , R ) / SO ( r ) . Suppose that X X X is a compact Riemann surface of genus at least 2. Then the Higgs bundle ( K X r , Φ ( q ) ) (K_X^r,\Phi(q)) ( K X r , Φ ( q )) is stable for any q ∈ H 0 ( K X r ) q \in H^0(K_X^r) q ∈ H 0 ( K X r ) , and thus there uniquely exists a harmonic metric h h h on ( K X r , Φ ( q ) ) (K_X^r,\Phi(q)) ( K X r , Φ ( q )) such that det ( h ) = 1 \det(h)=1 det ( h ) = 1 . From the uniqueness of the metric and the symmetry of the Higgs field (cf. [4 , 36 ] ), the harmonic metric h h h is diagonal and real. To construct a non-diagonal harmonic metric (resp. a non-real harmonic metric), one may, for example, use a solution to the Dirichlet problem for the Hitchin equations with a non-diagonal (resp. non-real) Hermitian metric prescribed on the boundary (cf. [22 , 44 ] ). Let h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) be a diagonal harmonic metric on a cyclic Higgs bundle ( K X r , Φ ( q ) ) → X (K_X^r,\Phi(q)) \to X ( K X r , Φ ( q )) → X . As mentioned in the previous subsection we obtain r − 1 r-1 r − 1 Hermitian metrics H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 on T X → X TX \to X T X → X defined as
H j : = h j − 1 ⊗ h j + 1 for each j = 1 , … , r − 1. H_j := h_j^{-1} \otimes h_{j+1}\quad\text{for each }j=1,\ldots,r-1. H j := h j − 1 ⊗ h j + 1 for each j = 1 , … , r − 1.
We also obtain a Hermitian metric H r H_r H r on T X → X TX \to X T X → X which is degenerate at the zeros of q q q as follows:
H r : = h r − 1 ⊗ h 1 ⊗ h q , H_r := h_r^{-1} \otimes h_1 \otimes h_q, H r := h r − 1 ⊗ h 1 ⊗ h q ,
where h q h_q h q is a Hermitian metric on K X − r → X K_X^{-r} \to X K X − r → X that is degenerate at the zeros of q q q defined as follows:
( h q ) x ( u , v ) : = ⟨ q x , u ⟩ ⟨ q x , v ⟩ for x ∈ X , u , v ∈ ( K X − r ) x . (h_q)_x(u,v) := \langle q_x,u\rangle\langle q_x,v\rangle\quad\text{for }x\in X,\ u,v\in(K_X^{-r})_x. ( h q ) x ( u , v ) := ⟨ q x , u ⟩ ⟨ q x , v ⟩ for x ∈ X , u , v ∈ ( K X − r ) x .
The connection D q ( h ) D_q(h) D q ( h ) is flat if and only if the Hermitian metric h h h satisfies the following elliptic equation, called the Hitchin equation:
F h + [ Φ ( q ) ∧ Φ ( q ) ∗ h ] = 0. F_h + [\Phi(q)\wedge\Phi(q)^{*h}] = 0. F h + [ Φ ( q ) ∧ Φ ( q ) ∗ h ] = 0.
By using H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r , we can describe the Hitchin equation as follows:
− 1 F H j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , r − 1 , (1) \sqrt{-1}F_{H_j}+\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)=0\ \text{for }j=1,\ldots,r-1, \tag*{(1)} − 1 F H j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , r − 1 , (1)
where for each j = 1 , … , r j=1,\ldots,r j = 1 , … , r , vol ( H j ) \operatorname{vol}(H_j) vol ( H j ) is the volume form of the metric H j H_j H j which is locally described as vol ( H j ) = − 1 H j ( ∂ ∂ z , ∂ ∂ z ˉ ) d z ∧ d z ˉ \operatorname{vol}(H_j)=\sqrt{-1}H_j\left(\frac{\partial}{\partial z},\frac{\partial}{\partial\bar z}\right)dz\wedge d\bar z vol ( H j ) = − 1 H j ( ∂ z ∂ , ∂ z ˉ ∂ ) d z ∧ d z ˉ , and H 0 H_0 H 0 is H r H_r H r . Equation (1) is a kind of Toda equation (cf. [1, 4, 32, 33, 53, 54]). By making a variable change from h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) to H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 , the Hitchin equation becomes the following:
− 1 F H j + 2 vol ( H j ) − vol ( H j − 1 ) − vol ( H j + 1 ) = 0 for j = 1 , … , r − 1. (2) \sqrt{-1}F_{H_j}+2\operatorname{vol}(H_j)-\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_{j+1})=0\ \text{for }j=1,\ldots,r-1. \tag*{(2)} − 1 F H j + 2 vol ( H j ) − vol ( H j − 1 ) − vol ( H j + 1 ) = 0 for j = 1 , … , r − 1. (2)
In the case where r = 2 r=2 r = 2 and q = 0 q=0 q = 0 , the Hitchin equation for the cyclic Higgs bundle is equivalent to the following equation:
− 1 F H + 2 vol ( H ) = 0 , (3) \sqrt{-1}F_H+2\operatorname{vol}(H)=0, \tag*{(3)} − 1 F H + 2 vol ( H ) = 0 , (3)
where the solution H H H to equation (3) is a Hermitian metric H H H on T X → X T_X\to X T X → X . The Kähler metric induced by H H H is a constant negative Gaussian curvature Kähler metric (cf. [35]). We remark on the following two points:
If q q q has no zeros, then H 1 = ⋯ = H r − 1 = h q 1 / r H_1=\cdots=H_{r-1}=h_q^{1/r} H 1 = ⋯ = H r − 1 = h q 1/ r is a solution to equation (2).
Suppose that q = 0 q=0 q = 0 and that there is a solution H H H to equation (3). We define positive constants λ 1 , … , λ r − 1 \lambda_1,\ldots,\lambda_{r-1} λ 1 , … , λ r − 1 as follows:
λ j : = 2 ∑ k = 1 r − 1 ( Λ r − 1 − 1 ) j k = j ( r − j ) , (4) \lambda_j:=2\sum_{k=1}^{r-1}(\Lambda_{r-1}^{-1})_{jk}=j(r-j), \tag*{(4)} λ j := 2 k = 1 ∑ r − 1 ( Λ r − 1 − 1 ) j k = j ( r − j ) , (4)
where Λ r − 1 − 1 \Lambda_{r-1}^{-1} Λ r − 1 − 1 is the inverse matrix of the Cartan matrix of type A r − 1 A_{r-1} A r − 1 . Then H j = λ j H H_j=\lambda_jH H j = λ j H (j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 ) is a solution to equation (2).
Li–Mochizuki [43 ] established the following result:
Theorem 7 ([43 ] ). Suppose that q q q is a non-zero holomorphic section unless X X X is hyperbolic. Then there always exists a unique complete harmonic metric h h h on ( K X r , Φ ( q ) ) → X (K_X^r,\Phi(q))\to X ( K X r , Φ ( q )) → X satisfying det ( h ) = 1 \det(h)=1 det ( h ) = 1 . Moreover, the complete harmonic metric h h h is real.
In addition to [43 ] , we refer the reader to [19 , 23 , 33 , 38 , 39 , 42 , 44 , 53 , 54 , 56 , 66 , 67 ] for the works related to complete harmonic metrics.
More general subharmonic weight functions
For each q ∈ H 0 ( K X r ) q\in H^0(K_X^r) q ∈ H 0 ( K X r ) , the metric h q h_q h q induces a singular metric (cf. [31 ] ) h q − 1 / r h_q^{-1/r} h q − 1/ r that diverges at the zeros of q q q . We can define a curvature for h q − 1 / r h_q^{-1/r} h q − 1/ r in the sense of currents, which is semipositive and has support at the zero points of q q q . We fix a smooth metric h ∗ h_* h ∗ and denote by e − φ q h ∗ e^{-\varphi_q}h_* e − φ q h ∗ the metric h q − 1 / r h_q^{-1/r} h q − 1/ r , where the weight function φ q \varphi_q φ q is defined as φ q : = 1 r log ∣ q ∣ h ∗ 2 \varphi_q:=\frac{1}{r}\log|q|_{h_*}^2 φ q := r 1 log ∣ q ∣ h ∗ 2 . If we choose a local flat reference metric, then the weight function defines a local subharmonic function. Also, although we will not go into details here, there is a way to identify the weight function with a global psh function on the total space of the dual line bundle excluding zero points (see [9 Section 2.1] ). In the sense described above, we call the function φ q \varphi_q φ q a subharmonic weight function , although this is a bit of an abuse of terminology. We consider a more general singular metric e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ with semipositive curvature, where the weight function φ \varphi φ is a function that is locally a sum of a subharmonic function and a smooth function. For each q N ∈ H 0 ( K X N ) q_N \in H^0(K_X^N) q N ∈ H 0 ( K X N ) , let φ q N \varphi_{q_N} φ q N be defined as 1 N log ∣ q N ∣ h ∗ 2 \frac{1}{N}\log|q_N|_{h_*}^2 N 1 log ∣ q N ∣ h ∗ 2 . Then e − φ q N h ∗ e^{-\varphi_{q_N}}h_* e − φ q N h ∗ is a singular metric with semipositive curvature. Moreover, any subharmonic weight function can be approximated, at least in the L l o c 1 L^1_{\mathrm{loc}} L loc 1 sense, by a sequence ( φ q N ) N ∈ N (\varphi_{q_N})_{N\in\mathbb{N}} ( φ q N ) N ∈ N , where q N ∈ H 0 ( K X N ) q_N \in H^0(K_X^N) q N ∈ H 0 ( K X N ) (cf. [31 , 37 ] ). In [48 , 49 ] , for each semipositive singular metric e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ and each r ≥ 2 r \ge2 r ≥ 2 , the following equation for a diagonal metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on K r → X K_r \to X K r → X was introduced:
− 1 F H j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , r − 1 , (5) \sqrt{-1}F_{H_j}+\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)=0 \text{ for } j=1,\ldots,r-1, \tag*{(5)} − 1 F H j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , r − 1 , (5)
where H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 are defined as H j : = h j − 1 ⊗ h j + 1 H_j:=h_j^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , and H r = H 0 H_r=H_0 H r = H 0 is defined as follows:
H r = H 0 : = h r − 1 ⊗ h 1 ⊗ ( e − φ h ∗ ) − r = H 1 − 1 ⊗ ⋯ ⊗ H r − 1 − 1 ⊗ ( e − φ h ∗ ) − r . \begin{aligned}
H_r=H_0&:=h_r^{-1}\otimes h_1\otimes(e^{-\varphi}h_*)^{-r}\\
&=H_1^{-1}\otimes\cdots\otimes H_{r-1}^{-1}\otimes(e^{-\varphi}h_*)^{-r}.
\end{aligned} H r = H 0 := h r − 1 ⊗ h 1 ⊗ ( e − φ h ∗ ) − r = H 1 − 1 ⊗ ⋯ ⊗ H r − 1 − 1 ⊗ ( e − φ h ∗ ) − r .
For φ = φ q N \varphi=\varphi_{q_N} φ = φ q N , a solution of equation (5) pulls back to a harmonic metric on the bundle π ∗ K r \pi^*K_r π ∗ K r over a suitable cyclic branched covering π : Y → X \pi:Y\to X π : Y → X on which q N 1 / N q_N^{1/N} q N 1/ N is single-valued. The corresponding Higgs field is defined using this root and the natural map π ∗ K X → K Y \pi^*K_X\to K_Y π ∗ K X → K Y (see [49 Section 2] ). As mentioned above, more general φ \varphi φ can be approximated by a sequence ( φ q N ) N ∈ N (\varphi_{q_N})_{N\in\mathbb{N}} ( φ q N ) N ∈ N at least in the L l o c 1 L^1_{\mathrm{loc}} L loc 1 -sense (cf. [31 , 37 ] ); therefore, equation (5) can be considered to be an equation obtained as the limit when the covering degree increases infinitely or the number of zeros increases infinitely. By making a variable change from h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) to H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 , the equation becomes the following:
− 1 F H j + 2 vol ( H j ) − vol ( H j − 1 ) − vol ( H j + 1 ) = 0 for j = 1 , … , r − 1. (6) \sqrt{-1}F_{H_j}+2\operatorname{vol}(H_j)-\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_{j+1})=0 \text{ for } j=1,\ldots,r-1. \tag*{(6)} − 1 F H j + 2 vol ( H j ) − vol ( H j − 1 ) − vol ( H j + 1 ) = 0 for j = 1 , … , r − 1. (6)
We remark on the following two points:
Suppose that φ = − ∞ \varphi=-\infty φ = − ∞ . Then equation (6) is the same as equation (2) for the case where q = 0 q=0 q = 0 , and thus for a solution H H H to equation (3) , H j = λ j H H_j=\lambda_jH H j = λ j H (j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 ) is a solution to equation (6) , where λ 1 , … , λ r − 1 \lambda_1,\ldots,\lambda_{r-1} λ 1 , … , λ r − 1 are constants defined in (4) .
Entropy and free energy
This subsection reviews the definitions of the functions called entropy and free energy given in [50 , 51 ] . Let h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) be a solution to equation (5) and let H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r be the Hermitian metrics constructed from h h h and e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ as in Section 2.3. Then we define functions, which we call entropy and free energy, as follows:
Definition 8. For each j = 0 , 1 , … , r − 1 j=0,1,\ldots,r-1 j = 0 , 1 , … , r − 1 and a real number β \beta β , we define a nonnegative function p j ( r , β , φ ) : X → [ 0 , 1 ] p_j(r,\beta,\varphi):X\to[0,1] p j ( r , β , φ ) : X → [ 0 , 1 ] as follows. For each x ∈ X ∖ φ − 1 ( − ∞ ) x\in X\setminus\varphi^{-1}(-\infty) x ∈ X ∖ φ − 1 ( − ∞ ) , we define
p j ( r , β , φ ) ( x ) : = vol ( H j ) β ∑ j = 0 r − 1 vol ( H j ) β ( x ) for each j = 0 , … , r − 1 , p_j(r,\beta,\varphi)(x):=\frac{\operatorname{vol}(H_j)^\beta}{\sum_{j=0}^{r-1}\operatorname{vol}(H_j)^\beta(x)} \text{ for each } j=0,\ldots,r-1, p j ( r , β , φ ) ( x ) := ∑ j = 0 r − 1 vol ( H j ) β ( x ) vol ( H j ) β for each j = 0 , … , r − 1 ,
where vol ( H j ) β / ∑ j = 0 r − 1 vol ( H j ) β \operatorname{vol}(H_j)^\beta/\sum_{j=0}^{r-1}\operatorname{vol}(H_j)^\beta vol ( H j ) β / ∑ j = 0 r − 1 vol ( H j ) β is understood to be ( vol ( H j ) / vol ( H ∗ ) ) β / ∑ j = 0 r − 1 ( vol ( H j ) / vol ( H ∗ ) ) β (\operatorname{vol}(H_j)/\operatorname{vol}(H_*) )^\beta/\sum_{j=0}^{r-1}(\operatorname{vol}(H_j)/\operatorname{vol}(H_*) )^\beta ( vol ( H j ) / vol ( H ∗ ) ) β / ∑ j = 0 r − 1 ( vol ( H j ) / vol ( H ∗ ) ) β for some reference metric H ∗ H_* H ∗ , which does not affect p j ( r , β , φ ) p_j(r,\beta,\varphi) p j ( r , β , φ ) . For each x ∈ φ − 1 ( − ∞ ) x \in\varphi^{-1}(-\infty) x ∈ φ − 1 ( − ∞ ) , we set p 0 ( r , β , φ ) ( x ) = 0 p_0(r,\beta,\varphi)(x)=0 p 0 ( r , β , φ ) ( x ) = 0 and define
p j ( r , β , φ ) ( x ) : = vol ( H j ) β ∑ j = 1 r − 1 vol ( H j ) β ( x ) p_j(r,\beta,\varphi)(x):=\frac{\operatorname{vol}(H_j)^\beta}{\sum_{j=1}^{r-1}\operatorname{vol}(H_j)^\beta}(x) p j ( r , β , φ ) ( x ) := ∑ j = 1 r − 1 vol ( H j ) β vol ( H j ) β ( x )
for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 .
We define the entropy by
S ( r , β , φ ) : = − ∑ j = 0 r − 1 p j ( r , β , φ ) log p j ( r , β , φ ) . S(r,\beta,\varphi):=-\sum_{j=0}^{r-1}p_j(r,\beta,\varphi)\log p_j(r,\beta,\varphi). S ( r , β , φ ) := − j = 0 ∑ r − 1 p j ( r , β , φ ) log p j ( r , β , φ ) .
Definition 9. Let β \beta β be a nonzero real number. For each x ∈ X x \in X x ∈ X , if ( β φ ) ( x ) (\beta\varphi)(x) ( β φ ) ( x ) does not take the value ∞ \infty ∞ , we set
F ( r , β , φ , H ∗ ) ( x ) : = − 1 β log ( ∑ j = 0 r − 1 ( vol ( H j ) vol ( H ∗ ) ) β ) ( x ) , F(r,\beta,\varphi,H_*)(x):=-\frac{1}{\beta}\log\left(\sum_{j=0}^{r-1}\left(\frac{\operatorname{vol}(H_j)}{\operatorname{vol}(H_*)}\right)^\beta\right)(x), F ( r , β , φ , H ∗ ) ( x ) := − β 1 log ( j = 0 ∑ r − 1 ( vol ( H ∗ ) vol ( H j ) ) β ) ( x ) , where H ∗ H_* H ∗ is a fixed reference metric. If ( β φ ) ( x ) (\beta\varphi)(x) ( β φ ) ( x ) takes the value ∞ \infty ∞ , we define
F ( r , β , φ , H ∗ ) ( x ) : = − 1 β log ( ∑ j = 1 r − 1 ( vol ( H j ) vol ( H ∗ ) ) β ) ( x ) . F(r,\beta,\varphi,H_*)(x):=-\frac{1}{\beta}\log\left(\sum_{j=1}^{r-1}\left(\frac{\operatorname{vol}(H_j)}{\operatorname{vol}(H_*)}\right)^\beta\right)(x). F ( r , β , φ , H ∗ ) ( x ) := − β 1 log ( j = 1 ∑ r − 1 ( vol ( H ∗ ) vol ( H j ) ) β ) ( x ) . We call F ( r , β , φ , H ∗ ) F(r,\beta,\varphi,H_*) F ( r , β , φ , H ∗ ) the free energy .
Note that the difference between free energy functions defined using the same reference metric does not depend on the choice of the reference metric. The author expects these functions to provide a new scale for measuring the degree of accuracy of approximations to equilibrium metrics in various situations; see, e.g., [5 , 6 , 7 , 8 , 9 , 10 , 12 , 13 , 14 , 15 , 24 , 26 , 28 , 31 , 34 , 45 , 46 , 47 , 59 , 61 , 62 ] . One of the goals in this direction is to quantitatively establish entropy-increase and free energy-decay principles for processes in which non-equilibrium metrics approximate equilibrium metrics. See [50 , 51 , 52 ] for further details and for the results obtained so far.
Proof
Preliminaries on approximation of non-smooth Kähler metrics
The objective of this subsection is to prove the following proposition:
Proposition 10. Let ( X , g X ) (X,g_X) ( X , g X ) be a Riemann surface equipped with a continuous Kähler metric. Then the following holds:
For each ϵ > 0 \epsilon>0 ϵ > 0 , there exists a smooth Kähler metric g ~ X , ϵ \widetilde{g}_{X,\epsilon} g X , ϵ such that
e − ϵ g X ≤ g ~ X , ϵ ≤ e ϵ g X . e^{-\epsilon}g_X\leq\widetilde{g}_{X,\epsilon}\leq e^\epsilon g_X. e − ϵ g X ≤ g X , ϵ ≤ e ϵ g X .
Suppose moreover that the Ricci curvature of g X g_X g X is bounded below by C C C . Then for each ϵ > 0 \epsilon> 0 ϵ > 0 , there exists a smooth Kähler metric g ~ X , ϵ \widetilde{g}_{X,\epsilon} g X , ϵ such that
e − ϵ g X ≤ g ~ X , ϵ ≤ e ϵ g X , (7) e^{-\epsilon}g_X \leq\widetilde{g}_{X,\epsilon} \leq e^{\epsilon}g_X, \tag*{(7)} e − ϵ g X ≤ g X , ϵ ≤ e ϵ g X , (7)
and g ~ X , ϵ \widetilde{g}_{X,\epsilon} g X , ϵ has Ricci curvature bounded below by a constant C ϵ C_\epsilon C ϵ , where
C ϵ = { e − ϵ C if C > 0 , e ϵ C if C < 0 , − ϵ if C = 0. (8) C_\epsilon=
\begin{cases}
e^{-\epsilon}C & \text{if } C > 0,\\
e^{\epsilon}C & \text{if } C < 0,\\
-\epsilon& \text{if } C = 0.
\end{cases} \tag*{(8)} C ϵ = ⎩ ⎨ ⎧ e − ϵ C e ϵ C − ϵ if C > 0 , if C < 0 , if C = 0. (8)
Note that assertion 1 is an immediate consequence of the well-known Whitney approximation theorem (see [40 Theorem 6.21] ). Therefore, the main part of the proof of Proposition 10 is the proof of assertion 2, which relies on [29 p. 66, Corollary 1] . Before giving the proof, we review this corollary together with [29 p. 66, Theorem 3.1] , from which the corollary is derived. Let ( M , g M ) (M,g_M) ( M , g M ) be a Riemannian manifold of real dimension n n n . We denote by Δ g M = − d ∗ d \Delta_{g_M}=-d^*d Δ g M = − d ∗ d the Laplace–Beltrami operator acting on functions, and by γ p \gamma_p γ p the Riemannian distance function from p ∈ M p \in M p ∈ M . Let O M \mathcal{O}_M O M denote the collection of all open subsets of M M M .
Definition 11 ([29 ] , p. 49 Definition 1.4). Let L = ( L ( U ) ) U ∈ O M \mathcal{L}=(\mathcal{L}(U))_{U\in\mathcal{O}_M} L = ( L ( U ) ) U ∈ O M be a subsheaf of the sheaf of continuous real-valued functions on M M M . We say that L \mathcal{L} L has the semilocal C ∞ C^\infty C ∞ approximation property if the following holds: Let U ∈ O M U\in\mathcal{O}_M U ∈ O M , let K ⊂ U K\subset U K ⊂ U be compact, and let f ∈ L ( U ) f\in\mathcal{L}(U) f ∈ L ( U ) be such that f f f is C ∞ C^\infty C ∞ on a neighborhood of a compact (possibly empty) subset K 1 ⊂ K K_1\subset K K 1 ⊂ K . Then there exists an open subset V ⊂ U V\subset U V ⊂ U containing K K K such that for every ϵ > 0 \epsilon>0 ϵ > 0 there exists a C ∞ C^\infty C ∞ function F ∈ L ( V ) F\in\mathcal{L}(V) F ∈ L ( V ) satisfying
(a) sup p ∈ K ∣ f ( p ) − F ( p ) ∣ < ϵ \sup_{p\in K}|f(p)-F(p)|<\epsilon sup p ∈ K ∣ f ( p ) − F ( p ) ∣ < ϵ
(b) d K 1 ( f , F ) < ϵ d_{K_1}(f,F)<\epsilon d K 1 ( f , F ) < ϵ
where the definition of d K 1 ( ⋅ , ⋅ ) d_{K_1}(\cdot,\cdot) d K 1 ( ⋅ , ⋅ ) can be found in [29 p. 49] .
Definition 12 ([29 ] , p. 49). For each U ∈ O M U\in\mathcal{O}_M U ∈ O M , let C 0 ( U ) C^0(U) C 0 ( U ) be the space of all real-valued continuous functions on U U U . The C 0 C^0 C 0 fine topology on C 0 ( U ) C^0(U) C 0 ( U ) is defined to be the topology generated by the subsets of the form
O f 1 , f 2 : = { f ∈ C 0 ( U ) ∣ ∣ f ( p ) − f 1 ( p ) ∣ < f 2 ( p ) for all p ∈ U } , O_{f_1,f_2}:=\{f\in C^0(U)\mid|f(p)-f_1(p)|<f_2(p)\text{ for all }p\in U\}, O f 1 , f 2 := { f ∈ C 0 ( U ) ∣ ∣ f ( p ) − f 1 ( p ) ∣ < f 2 ( p ) for all p ∈ U } , where f 1 ∈ C 0 ( U ) f_1\in C^0(U) f 1 ∈ C 0 ( U ) , f 2 ∈ C 0 ( U ) f_2\in C^0(U) f 2 ∈ C 0 ( U ) , and f 2 f_2 f 2 is a positive function.
Definition 13 ([29 ] , p. 65). Let η : M → R \eta:M\to\mathbb{R} η : M → R be a continuous function. A continuous function f : M → R f:M\to\mathbb{R} f : M → R is said to be η \eta η -subharmonic if for each point p ∈ M p\in M p ∈ M there is a positive number δ \delta δ such that the following function F : M → R F:M\to\mathbb{R} F : M → R is subharmonic in a neighborhood of p p p :
F ( q ) : = f ( q ) − ( δ + 1 2 n η ( p ) ) γ p ( q ) 2 for q ∈ M . F(q):=f(q)-\left(\delta+\frac{1}{2n}\eta(p)\right)\gamma_p(q)^2\quad\text{for }q\in M. F ( q ) := f ( q ) − ( δ + 2 n 1 η ( p ) ) γ p ( q ) 2 for q ∈ M .
We note that the square γ p 2 \gamma_p^2 γ p 2 of the Riemannian distance function is smooth in a neighborhood of p p p and satisfies Δ g M γ p 2 ∣ p = 2 n \left.\Delta_{g_M}\gamma_p^2\right|_p=2n Δ g M γ p 2 p = 2 n . This can be verified by using normal coordinates at p p p . From this and the definition of η \eta η -subharmonicity, the following characterization immediately follows:
Proposition 14 ([29 ] , p. 66 Lemma 3.2). A continuous function f : M → R f : M \to\mathbb{R} f : M → R is η \eta η -subharmonic if and only if there is a C ∞ C^{\infty} C ∞ function η 1 \eta_{1} η 1 such that Δ g M f ≥ η 1 \Delta_{g_M}f \geq\eta_{1} Δ g M f ≥ η 1 holds in the weak sense and η 1 > η \eta_{1} > \eta η 1 > η on M M M .
We denote by L η s h \mathcal{L}_{\eta\mathrm{sh}} L η sh the sheaf of (continuous) η \eta η -subharmonic functions. The following holds:
Theorem 16 ([29 ] , p. 66 Theorem 3.1). For any continuous η : M → R \eta: M \to\mathbb{R} η : M → R , L η s h \mathcal{L}_{\eta\mathrm{sh}} L η sh has the semilocal C ∞ C^{\infty} C ∞ approximation property.
Since the sheaf L η s h \mathcal{L}_{\eta\mathrm{sh}} L η sh has the C ∞ C^{\infty} C ∞ stability property and the maximum-closure property (cf. [29 Definition 1.1, Definition 1.3, and the arguments immediately below Lemma 3.2] ), it follows from [29 Theorem 1.1] that the following holds:
Corollary 17 ([29 ] , p. 66 Corollary 1). For any open set U ⊂ M U \subset M U ⊂ M , smooth η \eta η -subharmonic functions on U U U are dense in L η s h ( U ) ∗ w i t h r e s p e c t t o t h e ∗ \mathcal{L}_{\eta\mathrm{sh}}(U) *with respect to the* L η sh ( U ) ∗ w i t h r es p ec tt o t h e ∗ C^{0}$ fine topology.
We now prove Proposition 10.
Proof of Proposition 10. We prove assertion 2. It suffices to consider 0 < ϵ ≤ 1 0 < \epsilon\leq1 0 < ϵ ≤ 1 . Fix such an ϵ \epsilon ϵ . Let H H H be the continuous Hermitian metric on K X − 1 → X K^{-1}_{X} \to X K X − 1 → X induced by g X g_X g X . By assumption, there exists a constant C C C such that
− 1 F H ≥ C vol ( H ) . (9) \sqrt{-1}F_H \geq C\operatorname{vol}(H). \tag*{(9)} − 1 F H ≥ C vol ( H ) . (9)
Let H ∗ H_{*} H ∗ be a smooth reference metric on K X − 1 K^{-1}_{X} K X − 1 and write H = e − f H ∗ H=e^{-f}H_{*} H = e − f H ∗ for some continuous function f f f on X X X . Let g ∗ g_{*} g ∗ be the smooth Kähler metric induced by H ∗ H_{*} H ∗ . We apply Corollary 17 on ( X , 2 g ∗ ) (X,2g_{*}) ( X , 2 g ∗ ) , for which Δ 2 g ∗ = − 1 Λ H ∗ ∂ ∂ ˉ \Delta_{2g_{*}}=\sqrt{-1}\Lambda_{H_{*}}\partial\bar{\partial} Δ 2 g ∗ = − 1 Λ H ∗ ∂ ∂ ˉ . It follows from (9) that
− 1 F H ∗ + − 1 Λ H ∗ ∂ ∂ ˉ f ≥ C e − f vol ( H ∗ ) . \sqrt{-1}F_{H_{*}}+\sqrt{-1}\Lambda_{H_{*}}\partial\bar{\partial}f\geq Ce^{-f}\operatorname{vol}(H_{*}). − 1 F H ∗ + − 1 Λ H ∗ ∂ ∂ ˉ f ≥ C e − f vol ( H ∗ ) .
Set η : = C ϵ / 2 e − f − − 1 Λ H ∗ F H ∗ \eta:=C_{\epsilon/2}e^{-f}-\sqrt{-1}\Lambda_{H_{*}}F_{H_{*}} η := C ϵ /2 e − f − − 1 Λ H ∗ F H ∗ , where Λ H ∗ \Lambda_{H_{*}} Λ H ∗ denotes the adjoint of wedging with vol ( H ∗ ) \operatorname{vol}(H_{*}) vol ( H ∗ ) and where C ϵ C_{\epsilon} C ϵ is defined in (8). By applying Corollary 17 with domain X X X , for any positive continuous function f ′ f' f ′ , there exists a smooth η \eta η -subharmonic function u u u such that ∣ f − u ∣ < f ′ |f-u|<f' ∣ f − u ∣ < f ′ on X X X . In particular, there exists a smooth η \eta η -subharmonic function u ϵ u_{\epsilon} u ϵ such that ∣ f − u ϵ ∣ < ϵ / 2 |f-u_{\epsilon}|<\epsilon/2 ∣ f − u ϵ ∣ < ϵ /2 on X X X . Set H ~ ϵ : = e − u ϵ H ∗ \widetilde{H}_{\epsilon}:=e^{-u_{\epsilon}}H_{*} H ϵ := e − u ϵ H ∗ , and let g ~ X , ϵ \widetilde{g}_{X,\epsilon} g X , ϵ be the Kähler metric induced by H ~ ϵ \widetilde{H}_{\epsilon} H ϵ . By construction, g ~ X , ϵ \widetilde{g}_{X,\epsilon} g X , ϵ satisfies (7). Moreover, the curvature satisfies
− 1 F H ~ ϵ = − 1 F H ∗ + − 1 ∂ ∂ ˉ u ϵ ≥ − 1 F H ∗ + η vol ( H ∗ ) = C ϵ / 2 e − f vol ( H ∗ ) ≥ C ϵ e − u ϵ vol ( H ∗ ) = C ϵ vol ( H ~ ϵ ) . \begin{aligned}
\sqrt{-1}F_{\widetilde{H}_{\epsilon}}&=\sqrt{-1}F_{H_{*}}+\sqrt{-1}\partial\bar{\partial}u_{\epsilon}\\
&\geq\sqrt{-1}F_{H_{*}}+\eta\operatorname{vol}(H_{*})\\
&=C_{\epsilon/2}e^{-f}\operatorname{vol}(H_{*})\\
&\geq C_{\epsilon}e^{-u_{\epsilon}}\operatorname{vol}(H_{*})=C_{\epsilon}\operatorname{vol}(\widetilde{H}_{\epsilon}).
\end{aligned} − 1 F H ϵ = − 1 F H ∗ + − 1 ∂ ∂ ˉ u ϵ ≥ − 1 F H ∗ + η vol ( H ∗ ) = C ϵ /2 e − f vol ( H ∗ ) ≥ C ϵ e − u ϵ vol ( H ∗ ) = C ϵ vol ( H ϵ ) .
This completes the proof.
Preliminaries on maximum principles
This subsection provides some preliminaries on maximum principles. Let ( M , g M ) (M,g_M) ( M , g M ) be a connected Riemannian manifold. We denote by Δ g M = − d ∗ d \Delta_{g_M}=-d^*d Δ g M = − d ∗ d the Laplace–Beltrami operator acting on functions. Let p g M ( x , y , t ) p_{g_M}(x,y,t) p g M ( x , y , t ) be the positive minimal heat kernel (cf. [30 , 41 ] ). Throughout this section, we denote by L 1 ( + ∞ ) L^1(+\infty) L 1 ( + ∞ ) the space of functions on R \mathbb{R} R that belong to L 1 ( [ c , ∞ ) ) L^1([c,\infty)) L 1 ([ c , ∞ )) for some c ∈ R c\in\mathbb{R} c ∈ R . We first review the definition of stochastic completeness and several equivalent conditions.
Definition 18 (cf. [2 ] ). We say that ( M , g M ) (M,g_M) ( M , g M ) is stochastically complete if the following holds for some (hence any) ( x , t ) ∈ M × R > 0 (x,t)\in M\times\mathbb{R}_{>0} ( x , t ) ∈ M × R > 0 :
∫ M p g M ( x , y , t ) d μ g M ( y ) = 1 , \int_M p_{g_M}(x,y,t)d\mu_{g_M}(y)=1, ∫ M p g M ( x , y , t ) d μ g M ( y ) = 1 , where we denote by d μ g M d\mu_{g_M} d μ g M the volume measure associated with g M g_M g M .
Proposition 19 (cf. [2 ] ). The following conditions are equivalent:
(i) ( M , g M ) (M,g_M) ( M , g M ) is stochastically complete .
(ii) For every λ > 0 \lambda>0 λ > 0 , the only nonnegative bounded C 2 C^2 C 2 function satisfying Δ g M u ≥ λ u \Delta_{g_M}u\geq\lambda u Δ g M u ≥ λ u on M M M is u ≡ 0 u\equiv0 u ≡ 0 .
(iii) For every λ > 0 \lambda>0 λ > 0 , the only nonnegative bounded C 2 C^2 C 2 solution of Δ g M u = λ u \Delta_{g_M}u=\lambda u Δ g M u = λ u on M M M is u = 0 u=0 u = 0 .
(iv) For every T > 0 T>0 T > 0 , the only bounded solution on M × ( 0 , T ) M\times(0,T) M × ( 0 , T ) of the Cauchy problem
{ d u t d t = 1 2 Δ g M u t lim t → 0 u t = 0 in the L l o c 1 ( M ) sense \begin{cases}
\dfrac{d u_t}{dt}=\dfrac{1}{2}\Delta_{g_M}u_t\\
\lim_{t\to0}u_t=0\ \text{in the }L^1_{\mathrm{loc}}(M)\text{ sense}
\end{cases} ⎩ ⎨ ⎧ d t d u t = 2 1 Δ g M u t lim t → 0 u t = 0 in the L loc 1 ( M ) sense is u ≡ 0 u\equiv0 u ≡ 0 .
The following characterization of stochastic completeness was established by Pigola–Rigoli–Setti [58 ] :
Proposition 20 ([58 ] ). The following are equivalent:
(i) ( M , g M ) (M,g_M) ( M , g M ) is stochastically complete .
(ii) For every C 2 C^2 C 2 function u u u satisfying u ∗ = sup M u < ∞ u^*=\sup_M u<\infty u ∗ = sup M u < ∞ and every ϵ > 0 \epsilon>0 ϵ > 0 , we have
inf Ω ϵ Δ g M u ≤ 0 , \inf_{\Omega_\epsilon}\Delta_{g_M}u\leq0, Ω ϵ inf Δ g M u ≤ 0 , where Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } \Omega_\epsilon=\{x\in M\mid u(x)>u^*-\epsilon\} Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } .
(iii) For every C 2 C^2 C 2 function u u u with u ∗ = sup M u < ∞ u^*=\sup_M u<\infty u ∗ = sup M u < ∞ , there exists a sequence of points ( x k ) k ≥ 1 (x_k)_{k\geq1} ( x k ) k ≥ 1 such that
u ( x k ) > u ∗ − 1 / k , u(x_k)>u^*-1/k, u ( x k ) > u ∗ − 1/ k , ( Δ g M u ) ( x k ) < 1 / k (\Delta_{g_M}u)(x_k)<1/k ( Δ g M u ) ( x k ) < 1/ k for each $k\geq 1.
(iv) For every C 2 C^{2} C 2 function u u u with u ∗ = sup M u < ∞ u^{*}=\sup_{M}u<\infty u ∗ = sup M u < ∞ and every continuous function f f f , if Δ g M u ≥ f ( u ) \Delta_{g_{M}}u\ge f(u) Δ g M u ≥ f ( u ) on the subset Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } \Omega_{\epsilon}=\{x\in M\mid u(x)>u^{*}-\epsilon\} Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } for some ϵ > 0 \epsilon>0 ϵ > 0 , then f ( u ∗ ) ≤ 0 f(u^{*})\le0 f ( u ∗ ) ≤ 0 .
The following is a sufficient condition for a Riemannian manifold ( M , g M ) (M,g_{M}) ( M , g M ) to be stochastically complete:
Proposition 21 (cf. [2 ] ). Suppose that ( M , g M ) (M,g_{M}) ( M , g M ) is a complete Riemannian manifold of dimension m m m . Let p ∈ M p\in M p ∈ M be a fixed origin and denote by γ p ( x ) \gamma_{p}(x) γ p ( x ) the Riemannian distance function from p p p , and by cut ( p ) \operatorname{cut}(p) cut ( p ) its cut locus. Assume that the Ricci curvature satisfies
Ric ( ∇ γ p , ∇ γ p ) ≥ − ( m − 1 ) G 2 ( γ p ) \operatorname{Ric}(\nabla\gamma_{p},\nabla\gamma_{p})\ge-(m-1)G^{2}(\gamma_{p}) Ric ( ∇ γ p , ∇ γ p ) ≥ − ( m − 1 ) G 2 ( γ p ) on M ∖ ( cut ( p ) ∪ { p } ) M\setminus(\operatorname{cut}(p)\cup\{p\}) M ∖ ( cut ( p ) ∪ { p }) for some positive nondecreasing continuous function G G G with
1 G ∉ L 1 ( + ∞ ) . \frac{1}{G}\notin L^{1}(+\infty). G 1 ∈ / L 1 ( + ∞ ) . Then ( M , g M ) (M,g_{M}) ( M , g M ) is stochastically complete.
In particular, it follows from Proposition 21 that any complete Riemannian manifold with Ricci curvature bounded below is stochastically complete. Indeed, if Ric ≥ − ( m − 1 ) K \operatorname{Ric}\ge-(m-1)K Ric ≥ − ( m − 1 ) K for some constant K ≥ 0 K\ge0 K ≥ 0 , then we may take G ≡ K + 1 G\equiv\sqrt{K}+1 G ≡ K + 1 . For later use, we record the following variant of Proposition 20:
Proposition 22. Suppose that ( M , g M ) (M,g_{M}) ( M , g M ) is stochastically complete. Then for every u ∈ C 0 ( M ) ∩ W l o c 1 , 2 ( M ) u\in C^{0}(M)\cap W_{\mathrm{loc}}^{1,2}(M) u ∈ C 0 ( M ) ∩ W loc 1 , 2 ( M ) with u ∗ = sup M u < ∞ u^{*}=\sup_{M}u<\infty u ∗ = sup M u < ∞ , the following cannot occur:
Δ g M u ≥ g in the weak sense on U , (10) \Delta_{g_{M}}u\ge g\text{ in the weak sense on }U, \tag*{(10)} Δ g M u ≥ g in the weak sense on U , (10) g ≥ 2 c on Ω ϵ , (11) g\ge2c\text{ on }\Omega_{\epsilon}, \tag*{(11)} g ≥ 2 c on Ω ϵ , (11) Ω ‾ ϵ ⊆ U , (12) \overline{\Omega}_{\epsilon}\subseteq U, \tag*{(12)} Ω ϵ ⊆ U , (12) where Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } \Omega_{\epsilon}=\{x\in M\mid u(x)>u^{*}-\epsilon\} Ω ϵ = { x ∈ M ∣ u ( x ) > u ∗ − ϵ } .
Proof. The proof is the same as that of the implication (i)⇒ \Rightarrow ⇒ (ii) of Proposition 20. Suppose that (10), (11), and (12) hold for some ϵ > 0 \epsilon>0 ϵ > 0 , c > 0 c>0 c > 0 , and a continuous function g : U → R g:U\to\mathbb{R} g : U → R . We let Ω ∗ = { x ∈ U ∣ g ( x ) > c } \Omega^{*}=\{x\in U\mid g(x)>c\} Ω ∗ = { x ∈ U ∣ g ( x ) > c } , so that Ω ‾ ϵ ⊆ Ω ∗ \overline{\Omega}_{\epsilon}\subseteq\Omega^{*} Ω ϵ ⊆ Ω ∗ . Setting λ = c / ϵ \lambda=c/\epsilon λ = c / ϵ , at each x ∈ Ω ∗ x\in\Omega^{*} x ∈ Ω ∗ we have
g ( x ) > c ≥ c + λ ( u ( x ) − u ∗ ) = λ ( u ( x ) + ϵ − u ∗ ) , g(x)>c\ge c+\lambda(u(x)-u^{*})=\lambda(u(x)+\epsilon-u^{*}), g ( x ) > c ≥ c + λ ( u ( x ) − u ∗ ) = λ ( u ( x ) + ϵ − u ∗ ) ,
so that u + ϵ − u ∗ u+\epsilon-u^{*} u + ϵ − u ∗ is a weak subsolution of
Δ g M u − λ u = 0 (13) \Delta_{g_{M}}u-\lambda u=0 \tag*{(13)} Δ g M u − λ u = 0 (13)
on Ω ∗ \Omega^{*} Ω ∗ . Since the constant function 0 is obviously a subsolution of (13) on M M M , we see that u ϵ : = max { u + ϵ − u ∗ , 0 } u_{\epsilon}:=\max\{u+\epsilon-u^{*},0\} u ϵ := max { u + ϵ − u ∗ , 0 } is also a subsolution on M M M (see, e.g., [20 Lemma 1.5] ). Also, since max : R 2 → R \max:\mathbb{R}^{2}\to\mathbb{R} max : R 2 → R is Lipschitz continuous, hence a W l o c 1 , ∞ W_{\mathrm{loc}}^{1,\infty} W loc 1 , ∞ function (cf. [25 Theorem 4.5] ), u ϵ u_{\epsilon} u ϵ is again a continuous W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 function. Furthermore, u ϵ ≢ 0 u_{\epsilon}\not\equiv0 u ϵ ≡ 0 and 0 ≤ u ϵ ≤ ϵ < ∞ 0\le u_{\epsilon}\le\epsilon<\infty 0 ≤ u ϵ ≤ ϵ < ∞ . Note that any positive constant is a supersolution of (13). Choose a constant u + > ϵ u_{+}>\epsilon u + > ϵ . Applying the monotone iteration scheme (see [60 Proposition 2.4] for the formulation needed here) yields a smooth solution v v v of (13) on M M M such that u ϵ ≤ v ≤ u + u_{\epsilon}\le v\le u_{+} u ϵ ≤ v ≤ u + . Now, since u ϵ u_{\epsilon} u ϵ does not vanish identically, the same holds for v v v , and this contradicts condition (iii) in Proposition 19 and, equivalently, condition (i).
Following the proof of [43 Lemma 3.2] , we extend Proposition 22 to manifolds with smooth compact boundary, under a stronger assumption than stochastic completeness:
Proposition 23. Let ( M , g M ) (M, g_M) ( M , g M ) be a Riemannian manifold with smooth compact boundary ∂ M \partial M ∂ M . Suppose that ( M , g M ) (M, g_M) ( M , g M ) is complete and that the Ricci curvature is bounded below. Then, for every u ∈ C 0 ( M ) ∩ W l o c 1 , 2 ( M ∖ ∂ M ) u \in C^0(M) \cap W_{\mathrm{loc}}^{1,2}(M \setminus\partial M) u ∈ C 0 ( M ) ∩ W loc 1 , 2 ( M ∖ ∂ M ) with u ∗ = sup M u < ∞ u^* = \sup_M u < \infty u ∗ = sup M u < ∞ , if max ∂ M u < u ∗ \max_{\partial M} u < u^* max ∂ M u < u ∗ , the following cannot occur:
Δ g M u ≥ g in the weak sense on U , (14) \Delta_{g_M} u \ge g \quad\text{in the weak sense on } U, \tag*{(14)} Δ g M u ≥ g in the weak sense on U , (14) g ≥ 2 c on Ω ε , (15) g \ge2c \quad\text{on } \Omega_\varepsilon, \tag*{(15)} g ≥ 2 c on Ω ε , (15) Ω ε ‾ ⊆ U , (16) \overline{\Omega_\varepsilon} \subseteq U, \tag*{(16)} Ω ε ⊆ U , (16) where Ω ε = { x ∈ M ∣ u ( x ) > u ∗ − ε } \Omega_\varepsilon= \{x \in M \mid u(x) > u^* - \varepsilon\} Ω ε = { x ∈ M ∣ u ( x ) > u ∗ − ε } .
Proof. Suppose, for contradiction, that there exist ε > 0 \varepsilon> 0 ε > 0 , c > 0 c > 0 c > 0 , and a continuous function g : U → R g : U \to\mathbb{R} g : U → R satisfying (14), (15), and (16). There exists a compact neighborhood K ⊂ M K \subset M K ⊂ M of ∂ M \partial M ∂ M , together with a diffeomorphism Ψ : K → ∂ M × [ 0 , 1 ] \Psi: K \to\partial M \times[0,1] Ψ : K → ∂ M × [ 0 , 1 ] such that Ψ ( ∂ M ) = ∂ M × { 0 } \Psi(\partial M) = \partial M \times\{0\} Ψ ( ∂ M ) = ∂ M × { 0 } . By shrinking K K K if necessary, we may assume that max K u < u ∗ \max_K u < u^* max K u < u ∗ . There exists a complete Riemannian metric g ~ M \widetilde{g}_M g M on M ∖ ∂ M M \setminus\partial M M ∖ ∂ M such that g ~ M = g M \widetilde{g}_M = g_M g M = g M on M ∖ K M \setminus K M ∖ K and g ~ M \widetilde{g}_M g M has Ricci curvature bounded below. After possibly shrinking ε \varepsilon ε and U U U , we may assume that U ⊆ M ∖ K U \subseteq M \setminus K U ⊆ M ∖ K . Then we have
Δ g ~ M u ≥ g in the weak sense on U , \Delta_{\widetilde{g}_M} u \ge g \quad\text{in the weak sense on } U, Δ g M u ≥ g in the weak sense on U ,
g ≥ 2 c on Ω ε , g \ge2c \quad\text{on } \Omega_\varepsilon, g ≥ 2 c on Ω ε ,
Ω ε ‾ ⊆ U . \overline{\Omega_\varepsilon} \subseteq U. Ω ε ⊆ U .
This contradicts Proposition 22, completing the proof.
In order to extend the role of the Cheng–Yau maximum principle in [43 ] to equations with weaker regularity of solutions, following [11 Theorem 8.20] , we prove the following proposition.
Proposition 24. Let ( X , g X ) (X, g_X) ( X , g X ) be a Riemann surface with a complete continuous Kähler metric g X g_X g X whose Ricci curvature is bounded below. Let f : R → R f : \mathbb{R} \to\mathbb{R} f : R → R be continuous and define
F ( t ) : = ∫ 0 t f ( s ) d s . F(t) := \int_0^t f(s)\,\mathrm{d}s. F ( t ) := ∫ 0 t f ( s ) d s . Assume that f ( 0 ) = 0 f(0) = 0 f ( 0 ) = 0 and f ( t ) > 0 f(t) > 0 f ( t ) > 0 for t > 0 t > 0 t > 0 and that there exists a positive constant C C C such that for all t > 0 t > 0 t > 0 the following holds (cf. [11 Definition 1.9] ):
sup s ∈ [ 0 , t ] f ( s ) ≤ C f ( t ) (17) \sup_{s \in[0,t]} f(s) \le C f(t) \tag*{(17)} s ∈ [ 0 , t ] sup f ( s ) ≤ C f ( t ) (17) Suppose moreover that
1 / F ∈ L 1 ( + ∞ ) . (18) 1/\sqrt{F} \in L^1(+\infty). \tag*{(18)} 1/ F ∈ L 1 ( + ∞ ) . (18)
Let u u u be a function belonging to L l o c 1 ( X ) L^1_{\mathrm{loc}}(X) L loc 1 ( X ) such that f ( u ) ∈ L l o c 1 ( X ) f(u) \in L^1_{\mathrm{loc}}(X) f ( u ) ∈ L loc 1 ( X ) and u u u satisfies the following inequality in the sense of distributions:
Δ g X u ≥ f ( u ) . (19) \Delta_{g_X}u \ge f(u). \tag*{(19)} Δ g X u ≥ f ( u ) . (19)
Then ess sup X u < ∞ \operatorname{ess\,sup}_X u < \infty ess sup X u < ∞ . Furthermore, under the stronger assumption that u u u is continuous on X ∖ Y X \setminus Y X ∖ Y for some relatively compact open set Y Y Y with smooth boundary ∂ Y \partial Y ∂ Y , and that u ∈ W l o c 1 , 2 ( X ∖ Y ‾ ) u \in W^{1,2}_{\mathrm{loc}}(X \setminus\overline{Y}) u ∈ W loc 1 , 2 ( X ∖ Y ) , either f ( sup X ∖ Y u ) ≤ 0 f(\sup_{X\setminus Y}u) \le0 f ( sup X ∖ Y u ) ≤ 0 or sup X ∖ Y u = max ∂ Y u \sup_{X\setminus Y}u = \max_{\partial Y}u sup X ∖ Y u = max ∂ Y u holds.
For f ( t ) = e t − 1 f(t)=e^t-1 f ( t ) = e t − 1 , all the above conditions are satisfied. Moreover, f ( u ) = e u − 1 f(u)=e^u-1 f ( u ) = e u − 1 is locally bounded, and hence locally integrable, whenever u u u is upper semicontinuous. For the proof of Proposition 24, we prepare the following lemma, which is a simplified version of [11 Proposition 8.17] :
Lemma 25 (cf. [11 ] ). Let f : R → R f:\mathbb{R}\to\mathbb{R} f : R → R be a continuous function that satisfies the conditions in Proposition 24. Then for each ϵ > 0 \epsilon>0 ϵ > 0 , θ > 0 \theta>0 θ > 0 , 0 < δ < λ 0<\delta<\lambda 0 < δ < λ , and 0 ≤ r 0 < r 1 0\le r_0<r_1 0 ≤ r 0 < r 1 , there exist R 1 > r 1 R_1>r_1 R 1 > r 1 and a C 2 C^2 C 2 function w : [ r 0 , R 1 ) → [ δ , ∞ ) w:[r_0,R_1)\to[\delta,\infty) w : [ r 0 , R 1 ) → [ δ , ∞ ) that solves
{ w ′ ′ ( r ) + θ w ′ ( r ) ≤ ϵ f ( w ( r ) ) on [ r 0 , R 1 ) , w ′ ( r ) > 0 on [ r 0 , R 1 ) , w ( r ) → ∞ as r → R 1 , δ ≤ w ( r ) ≤ λ on [ r 0 , r 1 ] . (20) \begin{cases}
w''(r)+\theta w'(r)\le\epsilon f(w(r)) & \text{on }[r_0,R_1),\\
w'(r)>0 & \text{on }[r_0,R_1),\\
w(r)\to\infty& \text{as }r\to R_1,\\
\delta\le w(r)\le\lambda& \text{on }[r_0,r_1]. \tag*{(20)}
\end{cases} ⎩ ⎨ ⎧ w ′′ ( r ) + θ w ′ ( r ) ≤ ϵ f ( w ( r )) w ′ ( r ) > 0 w ( r ) → ∞ δ ≤ w ( r ) ≤ λ on [ r 0 , R 1 ) , on [ r 0 , R 1 ) , as r → R 1 , on [ r 0 , r 1 ] . (20)
Proof. For a given 0 < σ ≤ 1 0<\sigma\le1 0 < σ ≤ 1 to be specified later, set
C σ : = 1 2 σ ∫ δ ∞ 1 F ( s ) d s , C_\sigma:=\frac{1}{\sqrt{2\sigma}}\int_\delta^\infty\frac{1}{\sqrt{F(s)}}\,ds, C σ := 2 σ 1 ∫ δ ∞ F ( s ) 1 d s ,
which is well defined since we assume condition (18) . Since 1 1 + s ∉ L 1 ( + ∞ ) \frac{1}{1+s}\notin L^1(+\infty) 1 + s 1 ∈ / L 1 ( + ∞ ) , there exists R σ > r 0 R_\sigma>r_0 R σ > r 0 such that
C σ = ∫ r 0 R σ 1 1 + s d s . C_\sigma=\int_{r_0}^{R_\sigma}\frac{1}{1+s}\,ds. C σ = ∫ r 0 R σ 1 + s 1 d s .
Since C σ → ∞ C_\sigma\to\infty C σ → ∞ as σ → 0 \sigma\to0 σ → 0 , we can choose σ \sigma σ small enough that R σ > r 1 R_\sigma>r_1 R σ > r 1 . We let w : [ r 0 , R σ ) → [ δ , ∞ ) w:[r_0,R_\sigma)\to[\delta,\infty) w : [ r 0 , R σ ) → [ δ , ∞ ) be implicitly defined by the equation
∫ r R σ 1 1 + s d s = 1 2 σ ∫ w ( r ) ∞ 1 F ( s ) d s . (21) \int_r^{R_\sigma}\frac{1}{1+s}\,ds=\frac{1}{\sqrt{2\sigma}}\int_{w(r)}^\infty\frac{1}{\sqrt{F(s)}}\,ds. \tag*{(21)} ∫ r R σ 1 + s 1 d s = 2 σ 1 ∫ w ( r ) ∞ F ( s ) 1 d s . (21)
so that, by definition,
w ( r 0 ) = δ , w ( r ) → ∞ as r → R σ . w(r_0)=\delta,\quad w(r)\to\infty\ \text{as }r\to R_\sigma. w ( r 0 ) = δ , w ( r ) → ∞ as r → R σ .
Differentiating (21) yields
w ′ ( r ) = 1 1 + r 2 σ F ( w ( r ) ) . (22) w'(r)=\frac{1}{1+r}\sqrt{2\sigma F(w(r))}. \tag*{(22)} w ′ ( r ) = 1 + r 1 2 σ F ( w ( r )) . (22)
so that w ′ > 0 w' > 0 w ′ > 0 and
2 σ F ( w ( r ) ) = ( ( 1 + r ) w ′ ( r ) ) 2 . 2\sigma F(w(r)) = \left((1+r)w'(r)\right)^2. 2 σ F ( w ( r )) = ( ( 1 + r ) w ′ ( r ) ) 2 .
Differentiating once more, we obtain
σ f ( w ( r ) ) w ′ ( r ) = ( 1 + r ) w ′ ( r ) ( ( 1 + r ) w ′ ( r ) ) ′ , \sigma f(w(r))w'(r) = (1+r)w'(r)\left((1+r)w'(r)\right)', σ f ( w ( r )) w ′ ( r ) = ( 1 + r ) w ′ ( r ) ( ( 1 + r ) w ′ ( r ) ) ′ ,
that is,
σ f ( w ( r ) ) 1 + r = ( ( 1 + r ) w ′ ( r ) ) ′ . (23) \sigma\frac{f(w(r))}{1+r} = \left((1+r)w'(r)\right)'. \tag*{(23)} σ 1 + r f ( w ( r )) = ( ( 1 + r ) w ′ ( r ) ) ′ . (23)
Since f > 0 f > 0 f > 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , we infer that ( 1 + r ) w ′ ( r ) (1+r)w'(r) ( 1 + r ) w ′ ( r ) is increasing. Moreover, we deduce
( w ′ ( r ) ( 1 + r ) ) ′ = w ′ ′ ( r ) ( 1 + r ) + w ′ ( r ) ≥ w ′ ′ ( r ) ( 1 + r ) . \left(w'(r)(1+r)\right)' = w''(r)(1+r) + w'(r)
\ge w''(r)(1+r). ( w ′ ( r ) ( 1 + r ) ) ′ = w ′′ ( r ) ( 1 + r ) + w ′ ( r ) ≥ w ′′ ( r ) ( 1 + r ) .
Inserting this into (23) , we obtain
w ′ ′ ( r ) ≤ σ f ( w ( r ) ) ( 1 + r ) 2 . (24) w''(r) \le\sigma\frac{f(w(r))}{(1+r)^2}. \tag*{(24)} w ′′ ( r ) ≤ σ ( 1 + r ) 2 f ( w ( r )) . (24)
We next integrate (23) on [ r 0 , r ] [r_0,r] [ r 0 , r ] , and we use (17) to deduce
w ′ ( r ) ( 1 + r ) = w ′ ( r 0 ) ( 1 + r 0 ) + σ ∫ r 0 r f ( w ( s ) ) 1 + s d s ≤ w ′ ( r 0 ) ( 1 + r 0 ) + σ C f ( w ( r ) ) ∫ r 0 r 1 1 + s d s , w'(r)(1+r) = w'(r_0)(1+r_0) + \sigma\int_{r_0}^{r} \frac{f(w(s))}{1+s}\,ds
\le w'(r_0)(1+r_0) + \sigma C f(w(r))\int_{r_0}^{r}\frac{1}{1+s}\,ds, w ′ ( r ) ( 1 + r ) = w ′ ( r 0 ) ( 1 + r 0 ) + σ ∫ r 0 r 1 + s f ( w ( s )) d s ≤ w ′ ( r 0 ) ( 1 + r 0 ) + σ C f ( w ( r )) ∫ r 0 r 1 + s 1 d s ,
and thus, using w ( r 0 ) = δ w(r_0)=\delta w ( r 0 ) = δ , we obtain
w ′ ( r ) ≤ w ′ ( r ) ( 1 + r ) ≤ w ′ ( r 0 ) ( 1 + r 0 ) + σ C f ( w ( r ) ) ∫ r 0 r 1 1 + s d s ≤ ( w ′ ( r 0 ) ( 1 + r 0 ) f ( w ( r ) ) + σ C ∫ r 0 r 1 1 + s d s ) f ( w ( r ) ) ≤ ( C w ′ ( r 0 ) ( 1 + r 0 ) f ( δ ) + σ C ∫ r 0 r 1 1 + s d s ) f ( w ( r ) ) = ( C w ′ ( r 0 ) ( 1 + r 0 ) f ( δ ) + σ C 2 σ ∫ δ w ( r ) 1 F ( s ) d s ) f ( w ( r ) ) ≤ ( C w ′ ( r 0 ) ( 1 + r 0 ) f ( δ ) + σ C 2 σ ∫ δ ∞ 1 F ( s ) d s ) f ( w ( r ) ) , (25) \begin{aligned}
w'(r) &\le w'(r)(1+r) \\
&\le w'(r_0)(1+r_0) + \sigma C f(w(r))\int_{r_0}^{r}\frac{1}{1+s}\,ds \\
&\le\left(\frac{w'(r_0)(1+r_0)}{f(w(r))}+\sigma C\int_{r_0}^{r}\frac{1}{1+s}\,ds\right)f(w(r)) \\
&\le\left(\frac{Cw'(r_0)(1+r_0)}{f(\delta)}+\sigma C\int_{r_0}^{r}\frac{1}{1+s}\,ds\right)f(w(r)) \\
&= \left(\frac{Cw'(r_0)(1+r_0)}{f(\delta)}+\frac{\sigma C}{\sqrt{2\sigma}}\int_{\delta}^{w(r)}\frac{1}{\sqrt{F(s)}}\,ds\right)f(w(r)) \\
&\le\left(\frac{Cw'(r_0)(1+r_0)}{f(\delta)}+\frac{\sigma C}{\sqrt{2\sigma}}\int_{\delta}^{\infty}\frac{1}{\sqrt{F(s)}}\,ds\right)f(w(r)), \tag*{(25)}
\end{aligned} w ′ ( r ) ≤ w ′ ( r ) ( 1 + r ) ≤ w ′ ( r 0 ) ( 1 + r 0 ) + σ C f ( w ( r )) ∫ r 0 r 1 + s 1 d s ≤ ( f ( w ( r )) w ′ ( r 0 ) ( 1 + r 0 ) + σ C ∫ r 0 r 1 + s 1 d s ) f ( w ( r )) ≤ ( f ( δ ) C w ′ ( r 0 ) ( 1 + r 0 ) + σ C ∫ r 0 r 1 + s 1 d s ) f ( w ( r )) = ( f ( δ ) C w ′ ( r 0 ) ( 1 + r 0 ) + 2 σ σ C ∫ δ w ( r ) F ( s ) 1 d s ) f ( w ( r )) ≤ ( f ( δ ) C w ′ ( r 0 ) ( 1 + r 0 ) + 2 σ σ C ∫ δ ∞ F ( s ) 1 d s ) f ( w ( r )) , (25)
where we have used (21) in (25) . Then we use (22) with r = r 0 r=r_0 r = r 0 to obtain
w ′ ( r ) ≤ ( C 2 σ F ( w ( r 0 ) ) f ( δ ) + σ C 2 σ ∫ δ ∞ 1 F ( s ) d s ) f ( w ( r ) ) ≤ C ^ σ f ( w ( r ) ) , (26) \begin{aligned}
w'(r) \le\left(\frac{C\sqrt{2\sigma F(w(r_0))}}{f(\delta)}+\frac{\sigma C}{\sqrt{2\sigma}}\int_{\delta}^{\infty}\frac{1}{\sqrt{F(s)}}\,ds\right)f(w(r)) \\
&\le\widehat{C}_{\sigma}f(w(r)), \tag*{(26)}
\end{aligned} w ′ ( r ) ≤ ( f ( δ ) C 2 σ F ( w ( r 0 )) + 2 σ σ C ∫ δ ∞ F ( s ) 1 d s ) f ( w ( r )) ≤ C σ f ( w ( r )) , (26)
where we set
C ^ σ : = C 2 σ F ( w ( r 0 ) ) f ( δ ) + σ C 2 σ ∫ δ ∞ 1 F ( s ) d s . \widehat{C}_{\sigma}:=\frac{C\sqrt{2\sigma F(w(r_0))}}{f(\delta)}+\frac{\sigma C}{\sqrt{2\sigma}}\int_{\delta}^{\infty}\frac{1}{\sqrt{F(s)}}\,ds. C σ := f ( δ ) C 2 σ F ( w ( r 0 )) + 2 σ σ C ∫ δ ∞ F ( s ) 1 d s .
Putting together (24) and (26) , we obtain
w ′ ′ ( r ) + θ w ′ ( r ) ≤ ( σ ( 1 + r 0 ) 2 + θ C ^ σ ) f ( w ( r ) ) . w''(r)+\theta w'(r)\le\left(\frac{\sigma}{(1+r_0)^2}+\theta\widehat{C}_{\sigma}\right)f(w(r)). w ′′ ( r ) + θ w ′ ( r ) ≤ ( ( 1 + r 0 ) 2 σ + θ C σ ) f ( w ( r )) .
Since C ^ σ → 0 \widehat{C}_{\sigma}\to0 C σ → 0 as σ → 0 \sigma\to0 σ → 0 , we can choose σ \sigma σ small enough that the differential inequality in (20) is satisfied. To prove the last condition in (20) , simply observe that by (21)
∫ r 0 r 1 1 1 + s d s = 1 2 σ ∫ δ w ( r 1 ) 1 F ( s ) d s , \int_{r_0}^{r_1}\frac{1}{1+s}\,ds=\frac{1}{\sqrt{2\sigma}}\int_{\delta}^{w(r_1)}\frac{1}{\sqrt{F(s)}}\,ds, ∫ r 0 r 1 1 + s 1 d s = 2 σ 1 ∫ δ w ( r 1 ) F ( s ) 1 d s ,
and it is enough to choose σ \sigma σ in such a way that
∫ r 0 r 1 1 1 + s d s < 1 2 σ ∫ δ λ 1 F ( s ) d s . \int_{r_0}^{r_1}\frac{1}{1+s}\,ds<\frac{1}{\sqrt{2\sigma}}\int_{\delta}^{\lambda}\frac{1}{\sqrt{F(s)}}\,ds. ∫ r 0 r 1 1 + s 1 d s < 2 σ 1 ∫ δ λ F ( s ) 1 d s .
Lemma 26. Under the hypotheses of Proposition 24, set u + : = max { u , 0 } u_+:=\max\{u,0\} u + := max { u , 0 } . Then u + u_+ u + also satisfies the following inequality in the sense of distributions:
Δ g X u + ≥ f ( u + ) . (27) \Delta_{g_X}u_+\ge f(u_+). \tag*{(27)} Δ g X u + ≥ f ( u + ) . (27) Moreover, u + u_+ u + belongs to W l o c 1 , 2 ( X ) ∩ L l o c ∞ ( X ) W_{\mathrm{loc}}^{1,2}(X)\cap L_{\mathrm{loc}}^{\infty}(X) W loc 1 , 2 ( X ) ∩ L loc ∞ ( X ) .
Proof. Let U ⊆ X U\subseteq X U ⊆ X be a coordinate neighborhood identified with the unit disc and let g E g_E g E be the Euclidean metric on U U U . Writing g X = e F U g E g_X=e^{F_U}g_E g X = e F U g E on U U U , where F U ∈ C 0 ( U ) F_U\in C^0(U) F U ∈ C 0 ( U ) , inequality (19) reduces to the following on U U U :
Δ g E u ≥ e F U f ( u ) . (28) \Delta_{g_E}u\ge e^{F_U}f(u). \tag*{(28)} Δ g E u ≥ e F U f ( u ) . (28)
Since f ( 0 ) = 0 f(0)=0 f ( 0 ) = 0 , the constant function 0 0 0 also satisfies inequality (28) , and hence the claim in the first part of the lemma follows by applying Lemma 45 (ii), which we will prove later, to u u u and the constant function 0 0 0 , with G ( x , t ) = e F U ( x ) f ( t ) G(x,t)=e^{F_U(x)}f(t) G ( x , t ) = e F U ( x ) f ( t ) . Since f ( t ) ≥ 0 f(t)\ge0 f ( t ) ≥ 0 for t ≥ 0 t\ge0 t ≥ 0 , we observe that u + u_+ u + agrees almost everywhere with a subharmonic function. Since u + ≥ 0 u_+\ge0 u + ≥ 0 and subharmonic functions are locally bounded above, u + u_+ u + belongs to L l o c ∞ ( X ) L_{\mathrm{loc}}^{\infty}(X) L loc ∞ ( X ) . Moreover, it is well-known that every locally bounded subharmonic function belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 (cf. [21 Section 1.1.1] ), and therefore u + ∈ W l o c 1 , 2 ( X ) u_+\in W_{\mathrm{loc}}^{1,2}(X) u + ∈ W loc 1 , 2 ( X ) . This completes the proof.
Proof of Proposition 24. We suppose that the conclusion does not hold, i.e., for each positive real number λ \lambda λ , the set
E λ : = { x ∈ X ∣ u + ( x ) > λ } E_\lambda:= \{x \in X \mid u_+(x) > \lambda\} E λ := { x ∈ X ∣ u + ( x ) > λ }
has positive measure with respect to every smooth positive measure. Fix ϵ 0 > 0 \epsilon_0 > 0 ϵ 0 > 0 . By Proposition 10, there exists a smooth Kähler metric g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 such that
e − ϵ 0 g X ≤ g ~ X , ϵ 0 ≤ e ϵ 0 g X , (29) e^{-\epsilon_0}g_X \le\widetilde{g}_{X,\epsilon_0} \le e^{\epsilon_0}g_X, \tag*{(29)} e − ϵ 0 g X ≤ g X , ϵ 0 ≤ e ϵ 0 g X , (29)
and the Ricci curvature of g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 is bounded below. Since g X g_X g X is complete, g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 is also complete. From here on, geodesic balls and the Riemannian distance function are taken with respect to this smooth Kähler metric. Let B r 0 B_{r_0} B r 0 be an open geodesic ball centered at o o o with radius r 0 > 0 r_0 > 0 r 0 > 0 . Set
M 0 : = e s s s u p B 2 r 0 u + . M_0 := \mathop{\mathrm{ess\,sup}}_{B_{2r_0}} u_+. M 0 := ess sup B 2 r 0 u + .
Since u + u_+ u + belongs to L l o c ∞ ( X ) L^\infty_{\mathrm{loc}}(X) L loc ∞ ( X ) and B 2 r 0 B_{2r_0} B 2 r 0 is relatively compact, M 0 M_0 M 0 is finite. Fix positive real numbers δ \delta δ and λ \lambda λ so that M 0 < δ < λ M_0 < \delta< \lambda M 0 < δ < λ . Also, choose r 1 > 2 r 0 r_1 > 2r_0 r 1 > 2 r 0 large enough so that
E λ ∩ ( B r 1 ∖ B r 0 ) E_\lambda\cap(B_{r_1} \setminus B_{r_0}) E λ ∩ ( B r 1 ∖ B r 0 )
has positive measure. Since the Ricci curvature of g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 is bounded below, the Laplacian comparison theorem (see, e.g., [2 Theorem 1.2] ) yields
Δ g ~ X , ϵ 0 r ≤ A weakly on X ∖ B r 0 / 2 , (30) \Delta_{\widetilde{g}_{X,\epsilon_0}} r \le A \quad\text{weakly on } X \setminus B_{r_0/2}, \tag*{(30)} Δ g X , ϵ 0 r ≤ A weakly on X ∖ B r 0 /2 , (30)
for some constant A > 0 A > 0 A > 0 , where Δ g ~ X , ϵ 0 = − d ∗ d \Delta_{\widetilde{g}_{X,\epsilon_0}} = -d^*d Δ g X , ϵ 0 = − d ∗ d denotes the Laplace–Beltrami operator, and r r r is the Riemannian distance function from o o o . Applying Lemma 25 with
ϵ = e − ϵ 0 C , \epsilon= \frac{e^{-\epsilon_0}}{C}, ϵ = C e − ϵ 0 ,
where C C C is the constant in condition (17), we deduce the existence of w w w satisfying (20). Setting w ‾ ( x ) = w ( r ( x ) ) \overline{w}(x) = w(r(x)) w ( x ) = w ( r ( x )) and taking into account w ′ > 0 w' > 0 w ′ > 0 , (30), and the well-known formula ∣ ∇ g ~ X , ϵ 0 r ∣ g ~ X , ϵ 0 = 1 \lvert\nabla_{\widetilde{g}_{X,\epsilon_0}}r\rvert_{\widetilde{g}_{X,\epsilon_0}} = 1 ∣ ∇ g X , ϵ 0 r ∣ g X , ϵ 0 = 1 almost everywhere, w ‾ \overline{w} w solves
Δ g ~ X , ϵ 0 w ‾ ≤ e − ϵ 0 C f ( w ‾ ) on B R 1 ∖ B r 0 in the weak sense , w ′ ( r ) > 0 on B R 1 ∖ B r 0 , w ‾ → ∞ as x → ∂ B R 1 , δ ≤ w ‾ ≤ λ on B r 1 ∖ B r 0 . (31) \begin{aligned}
\Delta_{\widetilde{g}_{X,\epsilon_0}}\overline{w} &\le\frac{e^{-\epsilon_0}}{C}f(\overline{w}) && \text{on } B_{R_1} \setminus B_{r_0} \text{ in the weak sense},\\
w'(r) &> 0 && \text{on } B_{R_1} \setminus B_{r_0},\quad\overline{w}\to\infty\text{ as } x\to\partial B_{R_1},\\
\delta&\le\overline{w}\le\lambda&& \text{on } B_{r_1}\setminus B_{r_0}. \tag*{(31)}
\end{aligned} Δ g X , ϵ 0 w w ′ ( r ) δ ≤ C e − ϵ 0 f ( w ) > 0 ≤ w ≤ λ on B R 1 ∖ B r 0 in the weak sense , on B R 1 ∖ B r 0 , w → ∞ as x → ∂ B R 1 , on B r 1 ∖ B r 0 . (31)
We compare u + u_+ u + and w ‾ \overline{w} w on U : = B R 1 ∖ B ‾ r 0 U := B_{R_1} \setminus\overline{B}_{r_0} U := B R 1 ∖ B r 0 . We set
v : = max { u + − w ‾ , 0 } . v := \max\{u_+ - \overline{w}, 0\}. v := max { u + − w , 0 } .
Observe that v v v vanishes almost everywhere outside a compact subset of U U U . Indeed, on B 2 r 0 ∖ B ‾ r 0 B_{2r_0}\setminus\overline{B}_{r_0} B 2 r 0 ∖ B r 0 , by the assumption M 0 < δ M_0 < \delta M 0 < δ , v v v is almost everywhere 0, and near the boundary of B R 1 B_{R_1} B R 1 , v v v is also almost everywhere 0 since u + ∈ L l o c ∞ ( X ) u_+ \in L^\infty_{\mathrm{loc}}(X) u + ∈ L loc ∞ ( X ) and w ‾ \overline{w} w diverges at the boundary of B R 1 B_{R_1} B R 1 . Since X X X is a Riemann surface, the metrics g X g_X g X and g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 are conformally equivalent, and we may write g ~ X , ϵ 0 = ρ g X \widetilde{g}_{X,\epsilon_0}=\rho g_X g X , ϵ 0 = ρ g X with e − ϵ 0 ≤ ρ ≤ e ϵ 0 e^{-\epsilon_0}\leq\rho\leq e^{\epsilon_0} e − ϵ 0 ≤ ρ ≤ e ϵ 0 . By (27) , we obtain
Δ g ~ X , ϵ 0 u + ≥ ρ − 1 f ( u + ) ≥ e − ϵ 0 f ( u + ) . (32) \Delta_{\widetilde{g}_{X,\epsilon_0}}u_+\geq\rho^{-1}f(u_+)\geq e^{-\epsilon_0}f(u_+). \tag*{(32)} Δ g X , ϵ 0 u + ≥ ρ − 1 f ( u + ) ≥ e − ϵ 0 f ( u + ) . (32)
Subtracting the first inequality in (31) from (32) , we obtain the following:
Δ g ~ X , ϵ 0 ( u + − w ‾ ) ≥ e − ϵ 0 ( f ( u + ) − 1 C f ( w ‾ ) ) . \Delta_{\widetilde{g}_{X,\epsilon_0}}(u_+-\overline{w})\geq e^{-\epsilon_0}\left(f(u_+)-\frac{1}{C}f(\overline{w})\right). Δ g X , ϵ 0 ( u + − w ) ≥ e − ϵ 0 ( f ( u + ) − C 1 f ( w ) ) .
Note that the function v v v belongs to W 0 1 , 2 ( U ) W^{1,2}_0(U) W 0 1 , 2 ( U ) since u + ∈ W l o c 1 , 2 u_+\in W^{1,2}_{\mathrm{loc}} u + ∈ W loc 1 , 2 and w ‾ ∈ W l o c 1 , ∞ \overline{w}\in W^{1,\infty}_{\mathrm{loc}} w ∈ W loc 1 , ∞ and since v v v vanishes almost everywhere outside a compact subset of U U U . Therefore, v v v can be used as a test function, and multiplying both sides by v v v and integrating with respect to the measure induced by g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 , we have
− ∫ U ( d ( u + − w ‾ ) , d v ) g ~ X , ϵ 0 d μ g ~ X , ϵ 0 ≥ ∫ U e − ϵ 0 ( f ( u + ) − 1 C f ( w ‾ ) ) v d μ g ~ X , ϵ 0 , (33) -\int_U (d(u_+-\overline{w}),dv)_{\widetilde{g}_{X,\epsilon_0}}\,d\mu_{\widetilde{g}_{X,\epsilon_0}}\geq\int_U e^{-\epsilon_0}\left(f(u_+)-\frac{1}{C}f(\overline{w})\right)v\,d\mu_{\widetilde{g}_{X,\epsilon_0}}, \tag*{(33)} − ∫ U ( d ( u + − w ) , d v ) g X , ϵ 0 d μ g X , ϵ 0 ≥ ∫ U e − ϵ 0 ( f ( u + ) − C 1 f ( w ) ) v d μ g X , ϵ 0 , (33)
where we denote by ( ⋅ , ⋅ ) g ~ X , ϵ 0 (\cdot,\cdot)_{\widetilde{g}_{X,\epsilon_0}} ( ⋅ , ⋅ ) g X , ϵ 0 the inner product induced by g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 , and by d μ g ~ X , ϵ 0 d\mu_{\widetilde{g}_{X,\epsilon_0}} d μ g X , ϵ 0 the volume measure of g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 . From the weak differential formula for the function max { ⋅ , 0 } \max\{\cdot,0\} max { ⋅ , 0 } and the chain rule for Sobolev functions, we obtain d v = 1 { u + > w ‾ } d ( u + − w ‾ ) dv=1_{\{u_+>\overline{w}\}}d(u_+-\overline{w}) d v = 1 { u + > w } d ( u + − w ) almost everywhere on U U U , where we denote by 1 { u + > w ‾ } 1_{\{u_+>\overline{w}\}} 1 { u + > w } the indicator function of the set { u + > w ‾ } \{u_+>\overline{w}\} { u + > w } . Since 1 { u + > w ‾ } 2 = 1 { u + > w ‾ } 1^2_{\{u_+>\overline{w}\}}=1_{\{u_+>\overline{w}\}} 1 { u + > w } 2 = 1 { u + > w } , from (33) , we have
− ∫ U ( d v , d v ) g ~ X , ϵ 0 d μ g ~ X , ϵ 0 ≥ ∫ U e − ϵ 0 ( f ( u + ) − 1 C f ( w ‾ ) ) v d μ g ~ X , ϵ 0 . (34) -\int_U (dv,dv)_{\widetilde{g}_{X,\epsilon_0}}\,d\mu_{\widetilde{g}_{X,\epsilon_0}}\geq\int_U e^{-\epsilon_0}\left(f(u_+)-\frac{1}{C}f(\overline{w})\right)v\,d\mu_{\widetilde{g}_{X,\epsilon_0}}. \tag*{(34)} − ∫ U ( d v , d v ) g X , ϵ 0 d μ g X , ϵ 0 ≥ ∫ U e − ϵ 0 ( f ( u + ) − C 1 f ( w ) ) v d μ g X , ϵ 0 . (34)
The right-hand side of inequality (34) is nonnegative. Indeed, for almost every point x ∈ U x\in U x ∈ U such that v ( x ) > 0 v(x)>0 v ( x ) > 0 , since u + ( x ) > w ‾ ( x ) u_+(x)>\overline{w}(x) u + ( x ) > w ( x ) , by (17) , we obtain f ( u + ( x ) ) ≥ 1 C f ( w ‾ ( x ) ) f(u_+(x))\geq\frac{1}{C}f(\overline{w}(x)) f ( u + ( x )) ≥ C 1 f ( w ( x )) . This implies that
∫ U ( d v , d v ) g ~ X , ϵ 0 d μ g ~ X , ϵ 0 = 0. \int_U(dv,dv)_{\widetilde{g}_{X,\epsilon_0}}\,d\mu_{\widetilde{g}_{X,\epsilon_0}}=0. ∫ U ( d v , d v ) g X , ϵ 0 d μ g X , ϵ 0 = 0.
Since v v v vanishes almost everywhere outside a compact subset of U U U , there exists a function v ~ \widetilde{v} v on X X X that belongs to W l o c 1 , 2 ( X ) W^{1,2}_{\mathrm{loc}}(X) W loc 1 , 2 ( X ) and such that
v ~ = { v on U 0 on X ∖ U . \widetilde{v}=
\begin{cases}
v & \text{on }U\\
0 & \text{on }X\setminus U.
\end{cases} v = { v 0 on U on X ∖ U .
Since d v = 0 dv=0 d v = 0 almost everywhere on U U U , we have d v ~ = 0 d\widetilde{v}=0 d v = 0 almost everywhere on X X X . Hence v ~ \widetilde{v} v is almost everywhere constant on the connected surface X X X . Since v ~ = 0 \widetilde{v}=0 v = 0 on B r 0 B_{r_0} B r 0 , this constant is 0. Therefore, v = 0 v=0 v = 0 almost everywhere on U U U . Since w ‾ \overline{w} w satisfies w ‾ ≤ λ \overline{w}\leq\lambda w ≤ λ on B r 1 ∖ B r 0 B_{r_1}\setminus B_{r_0} B r 1 ∖ B r 0 , this in particular implies that E λ ∩ ( B r 1 ∖ B r 0 ) E_\lambda\cap(B_{r_1}\setminus B_{r_0}) E λ ∩ ( B r 1 ∖ B r 0 ) has measure 0, which contradicts the assumption.
Finally, we prove the last assertion of the proposition. Suppose that u u u is continuous on X ∖ Y X\setminus Y X ∖ Y for some relatively compact open set Y Y Y with smooth boundary ∂ Y \partial Y ∂ Y , and that u ∈ W l o c 1 , 2 ( X ∖ Y ) u\in W^{1,2}_{\mathrm{loc}}(X\setminus Y) u ∈ W loc 1 , 2 ( X ∖ Y ) . Assume further that sup X ∖ Y u > max ∂ Y u \sup_{X\setminus Y}u>\max_{\partial Y}u sup X ∖ Y u > max ∂ Y u . Fix ϵ 0 > 0 \epsilon_0>0 ϵ 0 > 0 . By Proposition 10 , there exists a complete Kähler metric g ~ X , ϵ 0 \widetilde{g}_{X,\epsilon_0} g X , ϵ 0 satisfying (29) and whose Ricci curvature is bounded below. By (19) , we obtain
Δ g ~ X , ϵ 0 u ≥ ρ − 1 f ( u ) , \Delta_{\widetilde{g}_{X,\epsilon_0}}u\geq\rho^{-1}f(u), Δ g X , ϵ 0 u ≥ ρ − 1 f ( u ) ,
where ρ \rho ρ is defined by g ~ X , ϵ 0 = ρ g X \widetilde{g}_{X,\epsilon_0}=\rho g_X g X , ϵ 0 = ρ g X . By Proposition 23 , for each integer k ≥ 1 k\geq1 k ≥ 1 , there exists a point
x k ∈ Ω 1 / k = { x ∈ X ∖ Y ∣ u ( x ) > sup X ∖ Y u − 1 k } x_k\in\Omega_{1/k}=\left\{x\in X\setminus Y\mid u(x)>\sup_{X\setminus Y}u-\frac{1}{k}\right\} x k ∈ Ω 1/ k = { x ∈ X ∖ Y ∣ u ( x ) > X ∖ Y sup u − k 1 }
such that ρ ( x k ) − 1 f ( u ( x k ) ) < 1 / k \rho(x_k)^{-1}f(u(x_k))<1/k ρ ( x k ) − 1 f ( u ( x k )) < 1/ k . Since ρ \rho ρ is bounded, letting k → ∞ k\to\infty k → ∞ , we obtain f ( sup X ∖ Y u ) ≤ 0 f(\sup_{X\setminus Y}u)\leq0 f ( sup X ∖ Y u ) ≤ 0 .
Preliminaries on linear algebra
The purpose of this subsection is to reformulate the linear-algebraic lemmas established in [43 ] (see also [64 ] ). All the proofs can be found in [43 ] , except for Lemma 30. Let h 1 , … , h r h_1,\ldots,h_r h 1 , … , h r be positive real numbers satisfying ∏ i = 1 r h i = h 1 ⋅ h 2 ⋯ h r − 1 ⋅ h r = 1 \prod_{i=1}^{r}h_i=h_1\cdot h_2\cdots h_{r-1}\cdot h_r=1 ∏ i = 1 r h i = h 1 ⋅ h 2 ⋯ h r − 1 ⋅ h r = 1 . We set H j : = h j − 1 ⋅ h j + 1 H_j:=h_j^{-1}\cdot h_{j+1} H j := h j − 1 ⋅ h j + 1 for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 . Let ϕ \phi ϕ be a real number or − ∞ -\infty − ∞ . We set a non-negative real number H r H_r H r as H r : = H 1 − 1 ⋅ H 2 − 1 ⋯ H r − 1 − 1 ⋅ e r ϕ = h r − 1 ⋅ h 1 ⋅ e r ϕ H_r:=H_1^{-1}\cdot H_2^{-1}\cdots H_{r-1}^{-1}\cdot e^{r\phi}=h_r^{-1}\cdot h_1\cdot e^{r\phi} H r := H 1 − 1 ⋅ H 2 − 1 ⋯ H r − 1 − 1 ⋅ e r ϕ = h r − 1 ⋅ h 1 ⋅ e r ϕ . We also set H 0 : = H r H_0:=H_r H 0 := H r . The following Lemma 27 and Lemma 28 are special cases of well-known inequalities which are formulated in more general settings (cf. [43 , 64 ] ).
Lemma 27 (cf. [43 , 64 ] ). There exists a positive constant C C C which depends only on r r r such that the following holds for any h 1 , … , h r h_1,\ldots,h_r h 1 , … , h r satisfying ∏ i = 1 r h i = h 1 ⋅ h 2 ⋯ h r − 1 ⋅ h r = 1 \prod_{i=1}^{r}h_i=h_1\cdot h_2\cdots h_{r-1}\cdot h_r=1 ∏ i = 1 r h i = h 1 ⋅ h 2 ⋯ h r − 1 ⋅ h r = 1 and ϕ \phi ϕ :
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C ( ∑ j = 1 r H j − r e ϕ ) 2 . (35) \sum_{j=1}^{r}(H_{j-1}-H_j)^2\ge C\left(\sum_{j=1}^{r}H_j-r e^\phi\right)^2. \tag*{(35)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ C ( j = 1 ∑ r H j − r e ϕ ) 2 . (35)
Proof. We give a proof directly without using the general theory. Since the variable transformation from h 1 , … , h r − 1 h_1,\ldots,h_{r-1} h 1 , … , h r − 1 to H j = h j − 1 ⋅ h j + 1 H_j=h_j^{-1}\cdot h_{j+1} H j = h j − 1 ⋅ h j + 1 is invertible, it is sufficient to state that inequality (35) holds for any positive real numbers H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 . We first suppose that ϕ \phi ϕ is not − ∞ -\infty − ∞ . Set H ~ j : = e − ϕ H j \widetilde{H}_j:=e^{-\phi}H_j H j := e − ϕ H j (j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 ) and H ~ r : = H ~ 1 − 1 ⋯ H ~ r − 1 − 1 = e − ϕ H r \widetilde{H}_r:=\widetilde{H}_1^{-1}\cdots\widetilde{H}_{r-1}^{-1}=e^{-\phi}H_r H r := H 1 − 1 ⋯ H r − 1 − 1 = e − ϕ H r . Then, from inequality (35) for H ~ 1 , … , H ~ r \widetilde{H}_1,\ldots,\widetilde{H}_r H 1 , … , H r in the case where ϕ = 0 \phi=0 ϕ = 0 , we obtain inequality (35) for H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r with respect to ϕ \phi ϕ . Therefore, from the beginning, we may assume that ϕ = 0 \phi=0 ϕ = 0 . We arrange the method given in [43 Proof of Proposition 3.8] for our proof. We set c max : = max { H j ∣ 1 ≤ j ≤ r } c_{\max}:=\max\{H_j\mid1\leq j\leq r\} c m a x := max { H j ∣ 1 ≤ j ≤ r } and c min : = min { H j ∣ 1 ≤ j ≤ r } c_{\min}:=\min\{H_j\mid1\leq j\leq r\} c m i n := min { H j ∣ 1 ≤ j ≤ r } . We take j max j_{\max} j m a x and j min j_{\min} j m i n so that c max = H j max c_{\max}=H_{j_{\max}} c m a x = H j m a x and c min = H j min c_{\min}=H_{j_{\min}} c m i n = H j m i n . In what follows, for simplicity we will only consider the case where they are 1 1 1 and r r r , respectively. Even if this were not the case, a simple consideration shows that the following estimates would hold just as well, with a slight modification of the argument. The following holds:
c max − c min = ∑ j = 2 r ( H j − 1 − H j ) ≤ ( ∑ j = 2 r ( H j − 1 − H j ) 2 ) 1 / 2 ( ∑ j = 2 r 1 2 ) 1 / 2 ≤ r − 1 ( ∑ j = 1 r ( H j − 1 − H j ) 2 ) 1 / 2 , \begin{aligned}
c_{\max}-c_{\min}
&=\sum_{j=2}^{r}(H_{j-1}-H_j)\\
&\leq\left(\sum_{j=2}^{r}(H_{j-1}-H_j)^2\right)^{1/2}\left(\sum_{j=2}^{r}1^2\right)^{1/2}\\
&\leq\sqrt{r-1}\left(\sum_{j=1}^{r}(H_{j-1}-H_j)^2\right)^{1/2},
\end{aligned} c m a x − c m i n = j = 2 ∑ r ( H j − 1 − H j ) ≤ ( j = 2 ∑ r ( H j − 1 − H j ) 2 ) 1/2 ( j = 2 ∑ r 1 2 ) 1/2 ≤ r − 1 ( j = 1 ∑ r ( H j − 1 − H j ) 2 ) 1/2 ,
where we have used the Cauchy–Schwarz inequality. Therefore we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ 1 r − 1 ( c max − c min ) 2 = 1 r − 1 c max 2 ( 1 − c min / c max ) 2 . (36) \begin{aligned}
\sum_{j=1}^{r}(H_{j-1}-H_j)^2
&\geq\frac{1}{r-1}(c_{\max}-c_{\min})^2\\
&=\frac{1}{r-1}c_{\max}^2(1-c_{\min}/c_{\max})^2. \tag*{(36)}
\end{aligned} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ r − 1 1 ( c m a x − c m i n ) 2 = r − 1 1 c m a x 2 ( 1 − c m i n / c m a x ) 2 . (36)
On the other hand, we have
∑ j = 1 r H j − r ≤ r c max − r . \sum_{j=1}^{r}H_j-r\leq rc_{\max}-r. j = 1 ∑ r H j − r ≤ r c m a x − r .
Since ∏ i = 1 r H i = H 1 ⋅ H 2 ⋯ H r = 1 \prod_{i=1}^{r} H_i = H_1 \cdot H_2 \cdots H_r = 1 ∏ i = 1 r H i = H 1 ⋅ H 2 ⋯ H r = 1 , it holds that ∑ j = 1 r H j − r ≥ 0 \sum_{j=1}^{r} H_j-r \ge0 ∑ j = 1 r H j − r ≥ 0 . Hence we have
( ∑ j = 1 r H j − r ) 2 ≤ r 2 ( c max − 1 ) 2 = r 2 c max 2 ( 1 − 1 / c max ) 2 . (37) \left(\sum_{j=1}^{r} H_j-r\right)^2 \le r^2(c_{\max}-1)^2
= r^2c_{\max}^2(1-1/c_{\max})^2. \tag*{(37)} ( j = 1 ∑ r H j − r ) 2 ≤ r 2 ( c m a x − 1 ) 2 = r 2 c m a x 2 ( 1 − 1/ c m a x ) 2 . (37)
We compare (36) and (37) . Since c min ≤ 1 c_{\min}\le1 c m i n ≤ 1 , we have
1 − c min / c max ≥ 1 − 1 / c max . 1-c_{\min}/c_{\max}\ge1-1/c_{\max}. 1 − c m i n / c m a x ≥ 1 − 1/ c m a x .
Furthermore, since 1 − 1 / c max ≥ 0 1-1/c_{\max}\ge0 1 − 1/ c m a x ≥ 0 , we have
( 1 − c min / c max ) 2 ≥ ( 1 − 1 / c max ) 2 . (38) (1-c_{\min}/c_{\max})^2\ge(1-1/c_{\max})^2. \tag*{(38)} ( 1 − c m i n / c m a x ) 2 ≥ ( 1 − 1/ c m a x ) 2 . (38)
Then from (36) , (37) , and (38) , for the case where ϕ ≠ − ∞ \phi\ne-\infty ϕ = − ∞ , we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ 1 r 2 ( r − 1 ) ( ∑ j = 1 r H j − r e ϕ ) 2 . (39) \sum_{j=1}^{r}(H_{j-1}-H_j)^2\ge\frac{1}{r^2(r-1)}\left(\sum_{j=1}^{r}H_j-re^\phi\right)^2. \tag*{(39)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ r 2 ( r − 1 ) 1 ( j = 1 ∑ r H j − r e ϕ ) 2 . (39)
We next consider the case where ϕ = − ∞ \phi=-\infty ϕ = − ∞ , i.e., e ϕ = 0 e^\phi=0 e ϕ = 0 . Let u 1 , … , u r − 1 u_1,\ldots,u_{r-1} u 1 , … , u r − 1 be the canonical basis of R r − 1 \mathbb{R}^{r-1} R r − 1 . We define an invertible ( r − 1 ) × ( r − 1 ) (r-1)\times(r-1) ( r − 1 ) × ( r − 1 ) -matrix A A A as follows:
A u j : = u j − u j + 1 ( j = 1 , … , r − 2 ) , Au_j:=u_j-u_{j+1}\quad(j=1,\ldots,r-2), A u j := u j − u j + 1 ( j = 1 , … , r − 2 ) ,
A u r − 1 : = u r − 1 , Au_{r-1}:=u_{r-1}, A u r − 1 := u r − 1 ,
where u 1 , … , u r − 1 u_1,\ldots,u_{r-1} u 1 , … , u r − 1 are considered to be column vectors. We set H ^ : = ( H 1 , … , H r − 1 ) ∈ R r − 1 \widehat{H}:=(H_1,\ldots,H_{r-1})\in\mathbb{R}^{r-1} H := ( H 1 , … , H r − 1 ) ∈ R r − 1 . Then we have
∑ j = 1 r ( H j − 1 − H j ) 2 = H 1 2 + ∑ j = 2 r − 1 ( H j − 1 − H j ) 2 + H r − 1 2 ≥ H 1 2 + ∑ j = 2 r − 1 ( H j − 1 − H j ) 2 = ∣ A H ^ ∣ 2 , \begin{aligned}
\sum_{j=1}^{r}(H_{j-1}-H_j)^2
&=H_1^2+\sum_{j=2}^{r-1}(H_{j-1}-H_j)^2+H_{r-1}^2\\
&\ge H_1^2+\sum_{j=2}^{r-1}(H_{j-1}-H_j)^2\\
&=|A\widehat{H}|^2,
\end{aligned} j = 1 ∑ r ( H j − 1 − H j ) 2 = H 1 2 + j = 2 ∑ r − 1 ( H j − 1 − H j ) 2 + H r − 1 2 ≥ H 1 2 + j = 2 ∑ r − 1 ( H j − 1 − H j ) 2 = ∣ A H ∣ 2 ,
where the norm ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ denotes the Hilbert–Schmidt norm. On the other hand, by the Cauchy–Schwarz inequality, we have
( ∑ j = 1 r − 1 H j ) 2 ≤ ( r − 1 ) ∣ H ^ ∣ 2 ≤ ( r − 1 ) ∣ A − 1 ⋅ A H ^ ∣ 2 ≤ ( r − 1 ) ∣ A − 1 ∣ 2 ∣ A H ^ ∣ 2 = ( r − 1 ) ⋅ 1 2 r ( r − 1 ) ∣ A H ^ ∣ 2 = 1 2 r ( r − 1 ) 2 ∣ A H ^ ∣ 2 . \begin{aligned}
\left(\sum_{j=1}^{r-1}H_j\right)^2
&\le(r-1)|\widehat{H}|^2\\
&\le(r-1)|A^{-1}\cdot A\widehat{H}|^2\\
&\le(r-1)|A^{-1}|^2|A\widehat{H}|^2\\
&=(r-1)\cdot\frac{1}{2}r(r-1)|A\widehat{H}|^2\\
&=\frac{1}{2}r(r-1)^2|A\widehat{H}|^2.
\end{aligned} ( j = 1 ∑ r − 1 H j ) 2 ≤ ( r − 1 ) ∣ H ∣ 2 ≤ ( r − 1 ) ∣ A − 1 ⋅ A H ∣ 2 ≤ ( r − 1 ) ∣ A − 1 ∣ 2 ∣ A H ∣ 2 = ( r − 1 ) ⋅ 2 1 r ( r − 1 ) ∣ A H ∣ 2 = 2 1 r ( r − 1 ) 2 ∣ A H ∣ 2 .
Therefore we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ 2 r 2 ( r − 1 ) ( ∑ j = 1 r − 1 H j ) 2 (40) \sum_{j=1}^{r}(H_{j-1}-H_j)^2 \ge\frac{2}{r^2(r-1)}\left(\sum_{j=1}^{r-1}H_j\right)^2 \tag*{(40)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ r 2 ( r − 1 ) 2 ( j = 1 ∑ r − 1 H j ) 2 (40)
for the case where ϕ = − ∞ \phi=-\infty ϕ = − ∞ . From inequalities (39) and (40), we have inequality (35) with C = 1 r 2 ( r − 1 ) C=\frac{1}{r^2(r-1)} C = r 2 ( r − 1 ) 1 .
Lemma 28 (cf. [43 , 64 ] ). Let C C C be a positive constant as in Lemma 27. Suppose that there exists a positive real number H > 0 H>0 H > 0 such that e ϕ ≤ H e^\phi\le H e ϕ ≤ H and that ∑ j = 1 r H j ≥ 2 r H \sum_{j=1}^{r}H_j\ge2rH ∑ j = 1 r H j ≥ 2 r H . Then the following holds:
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C 4 ( ∑ j = 1 r H j ) 2 . \sum_{j=1}^{r}(H_{j-1}-H_j)^2\ge\frac{C}{4}\left(\sum_{j=1}^{r}H_j\right)^2. j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ 4 C ( j = 1 ∑ r H j ) 2 .
Proof. The proof is the same as in the proof of [43 Lemma 3.7] . From the assumption, we have
∑ j = 1 r H j − r e ϕ ≥ r H ≥ r e ϕ . \sum_{j=1}^{r}H_j-re^\phi\ge rH\ge re^\phi. j = 1 ∑ r H j − r e ϕ ≥ r H ≥ r e ϕ .
This implies
1 2 ∑ j = 1 r H j ≥ r e ϕ , \frac{1}{2}\sum_{j=1}^{r}H_j\ge re^\phi, 2 1 j = 1 ∑ r H j ≥ r e ϕ ,
and thus
∑ j = 1 r H j − r e ϕ ≥ 1 2 ∑ j = 1 r H j . \sum_{j=1}^{r}H_j-re^\phi\ge\frac{1}{2}\sum_{j=1}^{r}H_j. j = 1 ∑ r H j − r e ϕ ≥ 2 1 j = 1 ∑ r H j .
From this we have the desired inequality.
For the proof of the following Lemma 29, we refer the reader to [43 Proposition 3.8] .
Lemma 29 ([43 ] ). For any 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , there exists δ > 0 \delta>0 δ > 0 that depends only on ϵ \epsilon ϵ and r r r , such that the following holds:
Let h 1 ′ , … , h r ′ h'_1,\ldots,h'_r h 1 ′ , … , h r ′ be positive real numbers satisfying ∏ i = 1 r h i ′ = h 1 ′ ⋅ h 2 ′ ⋯ h r − 1 ′ ⋅ h r ′ = 1 \prod_{i=1}^{r}h'_i=h'_1\cdot h'_2\cdots h'_{r-1}\cdot h'_r=1 ∏ i = 1 r h i ′ = h 1 ′ ⋅ h 2 ′ ⋯ h r − 1 ′ ⋅ h r ′ = 1 . We set s j : = h j − 1 h j ′ s_j:=h_j^{-1}h'_j s j := h j − 1 h j ′ for each j = 1 , … , r j=1,\ldots,r j = 1 , … , r . We also set s : = ( s 1 , … , s r ) ∈ ( R > 0 ) r s:=(s_1,\ldots,s_r)\in(\mathbb{R}_{>0})^r s := ( s 1 , … , s r ) ∈ ( R > 0 ) r . We assume that there exist positive constants B B B and C C C such that the following conditions hold:
H j − 1 ≤ B H_j^{-1}\le B H j − 1 ≤ B for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 .
$\sum_{j=1}^{r}s_j\le C.```
The following lemma is a variant of [43 Lemma 3.10] :
Lemma 30. For each ϵ > 0 \epsilon> 0 ϵ > 0 , if ∑ j = 1 r − 1 s j − 1 ( s j + 1 − s j ) 2 H j ≤ ϵ 2 \sum_{j=1}^{r-1}s_j^{-1}(s_{j+1}-s_j)^2H_j \le\epsilon^2 ∑ j = 1 r − 1 s j − 1 ( s j + 1 − s j ) 2 H j ≤ ϵ 2 , then
∑ j = 1 r ∣ s j − 1 ∣ ≤ r ( r − 1 ) B C ⋅ ϵ . \sum_{j=1}^{r}|s_j-1| \le r\sqrt{(r-1)BC}\cdot\epsilon. j = 1 ∑ r ∣ s j − 1∣ ≤ r ( r − 1 ) B C ⋅ ϵ .
Proof. By assumption, we have
1 B C ∑ j = 1 r − 1 ( s j + 1 − s j ) 2 ≤ ∑ j = 1 r − 1 s j − 1 ( s j + 1 − s j ) 2 H j ≤ ϵ 2 . \frac{1}{BC}\sum_{j=1}^{r-1}(s_{j+1}-s_j)^2 \le\sum_{j=1}^{r-1}s_j^{-1}(s_{j+1}-s_j)^2H_j \le\epsilon^2. B C 1 j = 1 ∑ r − 1 ( s j + 1 − s j ) 2 ≤ j = 1 ∑ r − 1 s j − 1 ( s j + 1 − s j ) 2 H j ≤ ϵ 2 .
For each i < k i<k i < k , we have
s k − s i = ∑ j = i k − 1 ( s j + 1 − s j ) , s_k-s_i=\sum_{j=i}^{k-1}(s_{j+1}-s_j), s k − s i = j = i ∑ k − 1 ( s j + 1 − s j ) ,
and thus, by the Cauchy-Schwarz inequality, we obtain
∣ s k − s i ∣ ≤ k − i ( ∑ j = i k − 1 ( s j + 1 − s j ) 2 ) 1 / 2 ≤ r − 1 B C ϵ . (41) |s_k-s_i|\le\sqrt{k-i}\left(\sum_{j=i}^{k-1}(s_{j+1}-s_j)^2\right)^{1/2}\le\sqrt{r-1}\sqrt{BC}\epsilon. \tag*{(41)} ∣ s k − s i ∣ ≤ k − i ( j = i ∑ k − 1 ( s j + 1 − s j ) 2 ) 1/2 ≤ r − 1 B C ϵ . (41)
Since ∏ j = 1 r s j = 1 \prod_{j=1}^{r}s_j=1 ∏ j = 1 r s j = 1 , min 1 ≤ j ≤ r s j ≤ 1 ≤ max 1 ≤ j ≤ r s j \min_{1\le j\le r}s_j\le1\le\max_{1\le j\le r}s_j min 1 ≤ j ≤ r s j ≤ 1 ≤ max 1 ≤ j ≤ r s j . Therefore, we obtain
∣ s j − 1 ∣ ≤ max 1 ≤ j ≤ r s j − min 1 ≤ j ≤ r s j . (42) |s_j-1|\le\max_{1\le j\le r}s_j-\min_{1\le j\le r}s_j. \tag*{(42)} ∣ s j − 1∣ ≤ 1 ≤ j ≤ r max s j − 1 ≤ j ≤ r min s j . (42)
From (41) and (42), we have
∑ j = 1 r ∣ s j − 1 ∣ ≤ r ( r − 1 ) B C ϵ . \sum_{j=1}^{r}|s_j-1|\le r\sqrt{(r-1)BC}\epsilon. j = 1 ∑ r ∣ s j − 1∣ ≤ r ( r − 1 ) B C ϵ .
This implies the claim. □
Throughout the remainder of the paper, we write n : = ⌊ r / 2 ⌋ n := \lfloor r/2\rfloor n := ⌊ r /2 ⌋ , where ⌊ x ⌋ \lfloor x\rfloor ⌊ x ⌋ denotes the greatest integer not exceeding x x x .
Lemma 32 ([43 ] , part of Theorem 4.4). Suppose that H j = H r − j H_j = H_{r-j} H j = H r − j for all j = 1 , … , r j = 1,\ldots,r j = 1 , … , r and H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r satisfy
H j − 1 ⋅ H j − 1 < 1 for all 1 ≤ j ≤ n . (43) H_{j-1}\cdot H_j^{-1} < 1 \text{ for all } 1 \leq j \leq n. \tag*{(43)} H j − 1 ⋅ H j − 1 < 1 for all 1 ≤ j ≤ n . (43) Then the following holds:
h j < e − 1 2 ( r − ( 2 j − 1 ) ) ϕ for all j = 1 , … , n . h_j < e^{-\frac{1}{2}(r-(2j-1))\phi} \text{ for all } j = 1,\ldots,n. h j < e − 2 1 ( r − ( 2 j − 1 )) ϕ for all j = 1 , … , n .
Proof. We give an alternative proof based on a slightly different perspective from that in [43 p. 27] . It is enough to consider the case where ϕ ∈ R \phi\in\mathbb{R} ϕ ∈ R . Note that from the assumption, we have h j = h r − j + 1 − 1 h_j = h_{r-j+1}^{-1} h j = h r − j + 1 − 1 for all j = 1 , … , r j = 1,\ldots,r j = 1 , … , r (see § 3.1 ). We set
v j : = log ( h j ) + 1 2 ( r − ( 2 j − 1 ) ) ϕ for j = 1 , … , r , v_j := \log(h_j) + \frac{1}{2}(r-(2j-1))\phi\text{ for } j = 1,\ldots,r, v j := log ( h j ) + 2 1 ( r − ( 2 j − 1 )) ϕ for j = 1 , … , r ,
A j : = log ( H j ) − ϕ for j = 0 , 1 , … , r − 1. A_j := \log(H_j) - \phi\text{ for } j = 0,1,\ldots,r-1. A j := log ( H j ) − ϕ for j = 0 , 1 , … , r − 1.
Then for each j = 1 , … , n j = 1,\ldots,n j = 1 , … , n , by a direct calculation, we have
2 v j = A 0 + 2 ∑ k = 1 j − 1 A k . 2v_j = A_0 + 2\sum_{k=1}^{j-1} A_k. 2 v j = A 0 + 2 k = 1 ∑ j − 1 A k .
It is enough to show that v j < 0 v_j < 0 v j < 0 for all j = 1 , … , n j = 1,\ldots,n j = 1 , … , n . Suppose, for contradiction, that v j ≥ 0 v_j \geq0 v j ≥ 0 for some 1 ≤ j ≤ n 1 \leq j \leq n 1 ≤ j ≤ n . By (43), we have A 0 < A 1 < ⋯ < A n A_0 < A_1 < \cdots< A_n A 0 < A 1 < ⋯ < A n . Hence A j − 1 ≥ 0 A_{j-1} \geq0 A j − 1 ≥ 0 ; otherwise, A k < 0 A_k < 0 A k < 0 for all 0 ≤ k ≤ j − 1 0 \leq k \leq j-1 0 ≤ k ≤ j − 1 , and the above identity would imply v j < 0 v_j < 0 v j < 0 , a contradiction. From (43), A k > 0 A_k > 0 A k > 0 for all j ≤ k ≤ n j \leq k \leq n j ≤ k ≤ n . This implies
∑ k = 0 r − 1 A k = A 0 + ∑ k = 1 j − 1 A k + ∑ k = j n A k + ∑ k = n + 1 r − 1 A k = A 0 + ∑ k = 1 j − 1 A k + ∑ k = j n A k + ∑ k = 1 r − ( n + 1 ) A k = 2 v j + 2 ∑ k = j n − 1 A k + ( r + 1 − 2 n ) A n > 0 , \begin{aligned}
\sum_{k=0}^{r-1} A_k = A_0 + \sum_{k=1}^{j-1} A_k + \sum_{k=j}^{n} A_k + \sum_{k=n+1}^{r-1} A_k \\
&= A_0 + \sum_{k=1}^{j-1} A_k + \sum_{k=j}^{n} A_k + \sum_{k=1}^{r-(n+1)} A_k \\
&= 2v_j + 2\sum_{k=j}^{n-1} A_k + (r+1-2n)A_n > 0,
\end{aligned} k = 0 ∑ r − 1 A k = A 0 + k = 1 ∑ j − 1 A k + k = j ∑ n A k + k = n + 1 ∑ r − 1 A k = A 0 + k = 1 ∑ j − 1 A k + k = j ∑ n A k + k = 1 ∑ r − ( n + 1 ) A k = 2 v j + 2 k = j ∑ n − 1 A k + ( r + 1 − 2 n ) A n > 0 ,
which contradicts the fact ∑ j = 0 r − 1 A j = 0 \sum_{j=0}^{r-1} A_j = 0 ∑ j = 0 r − 1 A j = 0 .
Proof of Theorem 1
Let h = ( h 1 , … , h r ) h = (h_1,\ldots,h_r) h = ( h 1 , … , h r ) be a weak solution to equation (5) satisfying det ( h ) = 1 \det(h) = 1 det ( h ) = 1 . By Proposition 69, h h h is of class C l o c 1 , 1 C^{1,1}_{\mathrm{loc}} C loc 1 , 1 . We denote by H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r the Hermitian metrics on K X − 1 → X K_X^{-1} \to X K X − 1 → X induced by h h h as in § 3.3 . The following is an extension of the well-known inequality for Higgs bundles established by Simpson [63 Lemma 10.1] (for the proof, see [49 Proposition 11] and [51 Proposition 23] ):
Lemma 33. The following inequality holds in the sense of distributions:
− 1 ∂ ∂ ˉ log ( ∑ j = 1 r vol ( H j ) / vol ( H ∗ ) ) ≥ ∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 ∑ j = 1 r vol ( H j ) + − 1 F H ∗ . (44) \sqrt{-1}\partial\bar\partial\log\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)/\operatorname{vol}(H_*)\right) \ge\frac{\sum_{j=1}^{r}\left(\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)\right)^2}{\sum_{j=1}^{r}\operatorname{vol}(H_j)}+\sqrt{-1}F_{H_*}. \tag*{(44)} − 1 ∂ ∂ ˉ log ( j = 1 ∑ r vol ( H j ) / vol ( H ∗ ) ) ≥ ∑ j = 1 r vol ( H j ) ∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 + − 1 F H ∗ . (44)
By using the above Lemma 33, we can show the following as in [43 Lemma 3.13] :
Lemma 34. Suppose that there exists a Hermitian metric H H H on K X − 1 → X K_X^{-1}\to X K X − 1 → X such that
e φ h ∗ − 1 ≤ H . e^{\varphi}h_*^{-1}\le H. e φ h ∗ − 1 ≤ H . Then there exist positive constants C 1 C_1 C 1 and C 2 C_2 C 2 independent of H H H such that
− 1 Λ H ∂ ∂ ˉ log ( ∑ j = 1 r vol ( H j ) / vol ( H ) ) ≥ C 1 ( ∑ j = 1 r vol ( H j ) / vol ( H ) ) − C 2 + − 1 Λ H F H , (45) \sqrt{-1}\Lambda_H\partial\bar\partial\log\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)/\operatorname{vol}(H)\right)\ge C_1\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)/\operatorname{vol}(H)\right)-C_2+\sqrt{-1}\Lambda_HF_H, \tag*{(45)} − 1 Λ H ∂ ∂ ˉ log ( j = 1 ∑ r vol ( H j ) / vol ( H ) ) ≥ C 1 ( j = 1 ∑ r vol ( H j ) / vol ( H ) ) − C 2 + − 1 Λ H F H , (45) where we denote by Λ H = ( vol ( H ) ∧ ) ∗ \Lambda_H=(\operatorname{vol}(H)\wedge)^* Λ H = ( vol ( H ) ∧ ) ∗ the contraction operator induced by H H H .
Proof. In what follows, we will identify the Hermitian metrics on K X − 1 → X K_X^{-1}\to X K X − 1 → X with positive real numbers at each point by fixing one trivialization for each point. Note, however, that all the following discussions are independent of the choice of trivialization. From Lemma 27, there exists a positive constant C C C depending only on r r r such that
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C ( ∑ j = 1 r H j − r e φ h ∗ − 1 ) . \sqrt{\sum_{j=1}^{r}(H_{j-1}-H_j)^2}\ge\sqrt{C}\left(\sum_{j=1}^{r}H_j-re^{\varphi}h_*^{-1}\right). j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ C ( j = 1 ∑ r H j − r e φ h ∗ − 1 ) .
From the assumption e φ h ∗ − 1 ≤ H e^{\varphi}h_*^{-1}\le H e φ h ∗ − 1 ≤ H , we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C ( ∑ j = 1 r H j − r e φ h ∗ − 1 ) ≥ C ( ∑ j = 1 r H j − 2 r H ) . (46) \sqrt{\sum_{j=1}^{r}(H_{j-1}-H_j)^2}\ge\sqrt{C}\left(\sum_{j=1}^{r}H_j-re^{\varphi}h_*^{-1}\right)\ge\sqrt{C}\left(\sum_{j=1}^{r}H_j-2rH\right). \tag*{(46)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ C ( j = 1 ∑ r H j − r e φ h ∗ − 1 ) ≥ C ( j = 1 ∑ r H j − 2 r H ) . (46)
On the other hand, from Lemma 28, under the assumption of ∑ j = 1 r H j ≥ 2 r H \sum_{j=1}^{r}H_j\ge2rH ∑ j = 1 r H j ≥ 2 r H , we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C 2 ∑ j = 1 r H j . (47) \sqrt{\sum_{j=1}^{r}(H_{j-1}-H_j)^2}\ge\frac{\sqrt{C}}{2}\sum_{j=1}^{r}H_j. \tag*{(47)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ 2 C j = 1 ∑ r H j . (47)
From inequalities (46) and (47), we have
∑ j = 1 r ( H j − 1 − H j ) 2 ≥ C 2 ( ∑ j = 1 r H j ) ( ∑ j = 1 r H j − 2 r H ) . (48) \sum_{j=1}^{r}(H_{j-1}-H_j)^2\ge\frac{C}{2}\left(\sum_{j=1}^{r}H_j\right)\left(\sum_{j=1}^{r}H_j-2rH\right). \tag*{(48)} j = 1 ∑ r ( H j − 1 − H j ) 2 ≥ 2 C ( j = 1 ∑ r H j ) ( j = 1 ∑ r H j − 2 r H ) . (48)
Therefore, the following holds globally over X X X :
∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 ∑ j = 1 r vol ( H j ) ≥ C 2 ( ∑ j = 1 r vol ( H j ) − 2 r vol ( H ) ) . (49) \frac{\sum_{j=1}^{r}\left(\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)\right)^2}{\sum_{j=1}^{r}\operatorname{vol}(H_j)}
\ge\frac{C}{2}\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)-2r\operatorname{vol}(H)\right). \tag*{(49)} ∑ j = 1 r vol ( H j ) ∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 ≥ 2 C ( j = 1 ∑ r vol ( H j ) − 2 r vol ( H ) ) . (49)
From the above inequality (49) and inequality (44), we have the desired conclusion.
Throughout this section, we follow the notation in Lemma 34 and denote by Λ H \Lambda_H Λ H the contraction operator induced by a Hermitian metric H H H on K X − 1 → X K_X^{-1} \to X K X − 1 → X . For the Kähler metric g H g_H g H induced by H H H , we have Δ 2 g H = − 1 Λ H ∂ ∂ ˉ \Delta_{2g_H}=\sqrt{-1}\Lambda_H\partial\bar{\partial} Δ 2 g H = − 1 Λ H ∂ ∂ ˉ . In this subsection, we apply Propositions 22 and 24 to inequalities involving this operator with respect to 2 g H 2g_H 2 g H . The following Corollary 35 is an extension of the crucial insight discovered by Li–Mochizuki [43 Section 3.3] to more general subharmonic weight functions:
Corollary 35. Let g X g_X g X be a complete continuous Kähler metric on X X X whose Ricci curvature is bounded below and H H H the Hermitian metric on K X − 1 → X K_X^{-1} \to X K X − 1 → X induced by g X g_X g X . Then for the Kähler metric g X g_X g X , the following (i) and (ii) are equivalent:
(i) The function e φ h ∗ − 1 ⊗ H − 1 e^\varphi h_*^{-1}\otimes H^{-1} e φ h ∗ − 1 ⊗ H − 1 is a bounded function.
(ii) For each j = 1 , … , r j=1,\ldots,r j = 1 , … , r , H j ⊗ H − 1 H_j\otimes H^{-1} H j ⊗ H − 1 is a bounded function.
Proof. It is clear that (ii) implies (i). We prove the converse. We suppose that (i) holds. For simplicity, we assume that we can take 1 as an upper bound on the function e φ h ∗ − 1 ⊗ H − 1 e^\varphi h_*^{-1}\otimes H^{-1} e φ h ∗ − 1 ⊗ H − 1 . Since the Ricci curvature of g X g_X g X is bounded below, there exists a positive constant C 3 C_3 C 3 such that − 1 Λ H F H ≥ − C 3 \sqrt{-1}\Lambda_HF_H\ge-C_3 − 1 Λ H F H ≥ − C 3 . We set
C 4 : = C 1 C 2 + C 3 , C_4:=\frac{C_1}{C_2+C_3}, C 4 := C 2 + C 3 C 1 ,
u : = log ( ∑ j = 1 r vol ( H j ) / vol ( H ) ) + log C 4 . u:=\log\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)/\operatorname{vol}(H)\right)+\log C_4. u := log ( j = 1 ∑ r vol ( H j ) / vol ( H ) ) + log C 4 .
From inequality (45), we obtain
− 1 Λ H ∂ ∂ ˉ u ≥ ( C 2 + C 3 ) ( e u − 1 ) . (50) \sqrt{-1}\Lambda_H\partial\bar{\partial}u\ge(C_2+C_3)(e^u-1). \tag*{(50)} − 1 Λ H ∂ ∂ ˉ u ≥ ( C 2 + C 3 ) ( e u − 1 ) . (50)
By applying Proposition 24 to inequality (50) , we obtain the claim. □ \square □
Following [43 Lemma 3.22] , we can show the following:
Lemma 36. There exist positive constants C 1 C_1 C 1 and C 2 C_2 C 2 depending only on r r r such that the following holds for each i = 1 , … , r − 1 i=1,\ldots,r-1 i = 1 , … , r − 1 :
− 1 Λ H i ∂ ∂ ˉ log ( ∑ j = 1 r ( vol ( H j ) / vol ( H i ) ) ) ≥ C 1 ( ∑ j = 1 r vol ( H j ) / vol ( H i ) ) − C 2 + − 1 Λ H i F H i . \sqrt{-1}\Lambda_{H_i}\partial\bar{\partial}\log\left(\sum_{j=1}^{r}\left(\operatorname{vol}(H_j)/\operatorname{vol}(H_i)\right)\right)\ge C_1\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)/\operatorname{vol}(H_i)\right)-C_2+\sqrt{-1}\Lambda_{H_i}F_{H_i}. − 1 Λ H i ∂ ∂ ˉ log ( j = 1 ∑ r ( vol ( H j ) / vol ( H i ) ) ) ≥ C 1 ( j = 1 ∑ r vol ( H j ) / vol ( H i ) ) − C 2 + − 1 Λ H i F H i .
Proof. As in the proof of Lemma 34 , we identify the Hermitian metrics on K X − 1 → X K_X^{-1}\to X K X − 1 → X with positive real numbers at each point by fixing one trivialization for each point. For each i = 1 , … , r − 1 i=1,\ldots,r-1 i = 1 , … , r − 1 , if 2 H i ≥ e φ h ∗ − 1 2H_i\ge e^\varphi h_*^{-1} 2 H i ≥ e φ h ∗ − 1 , then by the same argument as in the proof of Lemma 34 , we have
∑ j = 1 r ( H j − 1 − H j ) 2 ∑ j = 1 r H j ≥ C 2 ( ∑ j = 1 r H j − 4 r H i ) . \frac{\sum_{j=1}^{r}(H_{j-1}-H_j)^2}{\sum_{j=1}^{r}H_j}\ge\frac{C}{2}\left(\sum_{j=1}^{r}H_j-4rH_i\right). ∑ j = 1 r H j ∑ j = 1 r ( H j − 1 − H j ) 2 ≥ 2 C ( j = 1 ∑ r H j − 4 r H i ) .
where C C C is a positive constant as in Lemma 27 and Lemma 28 . On the other hand, if 2 H i < ϵ φ h ∗ − 1 2H_i<\epsilon^{\varphi}h_*^{-1} 2 H i < ϵ φ h ∗ − 1 , then from Lemma 29 , there exists a positive constant δ \delta δ depending only on r r r (and the choice of ϵ = 1 2 \epsilon=\frac{1}{2} ϵ = 2 1 ) such that
∑ j = 1 r ( H j − 1 − H j ) 2 ∑ j = 1 r H j ≥ δ ( ∑ j = 1 r H j ) . \frac{\sum_{j=1}^{r}(H_{j-1}-H_j)^2}{\sum_{j=1}^{r}H_j}\geq\delta\left(\sum_{j=1}^{r}H_j\right). ∑ j = 1 r H j ∑ j = 1 r ( H j − 1 − H j ) 2 ≥ δ ( j = 1 ∑ r H j ) .
From the above argument, the following holds globally on X X X :
∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 ∑ j = 1 r vol ( H j ) ≥ min { C / 2 , δ } ( ∑ j = 1 r vol ( H j ) ) − 2 r C vol ( H i ) . (51) \frac{\sum_{j=1}^{r}(\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j))^2}{\sum_{j=1}^{r}\operatorname{vol}(H_j)}\geq\min\{C/2,\delta\}\left(\sum_{j=1}^{r}\operatorname{vol}(H_j)\right)-2rC\operatorname{vol}(H_i). \tag*{(51)} ∑ j = 1 r vol ( H j ) ∑ j = 1 r ( vol ( H j − 1 ) − vol ( H j ) ) 2 ≥ min { C /2 , δ } ( j = 1 ∑ r vol ( H j ) ) − 2 r C vol ( H i ) . (51)
From inequality (51) and inequality (44), we have the desired conclusion. □ \square □
From Lemma 36, we have the following:
Corollary 37. Suppose that an H i H_i H i among H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 is complete. Then ∑ j = 1 r H j ⊗ H i − 1 \sum_{j=1}^{r}H_j\otimes H_i^{-1} ∑ j = 1 r H j ⊗ H i − 1 is a bounded function. If moreover e φ e^\varphi e φ is continuous, then ∑ j = 1 r H j ⊗ H i − 1 ≤ C 2 + 2 C 1 2 \sum_{j=1}^{r}H_j\otimes H_i^{-1}\leq\frac{C_2+2}{C_1^2} ∑ j = 1 r H j ⊗ H i − 1 ≤ C 1 2 C 2 + 2 with constants C 1 C_1 C 1 and C 2 C_2 C 2 in Lemma 36.
Proof. From equation (6), we have − 1 Λ H i F H i ≥ − 2 \sqrt{-1}\Lambda_{H_i}F_{H_i}\geq-2 − 1 Λ H i F H i ≥ − 2 (cf. [43 Lemma 2.4] ). We set
C : = C 2 + 2 , C:=C_2+2, C := C 2 + 2 ,
u : = log ( ∑ j = 1 r ( vol ( H j ) / vol ( H i ) ) ) + log ( C 1 / ( C 2 + 2 ) ) . u:=\log\left(\sum_{j=1}^{r}(\operatorname{vol}(H_j)/\operatorname{vol}(H_i))\right)+\log(C_1/(C_2+2)). u := log ( j = 1 ∑ r ( vol ( H j ) / vol ( H i )) ) + log ( C 1 / ( C 2 + 2 )) .
Then we have
− 1 Λ H i ∂ ‾ ∂ u ≥ C ( e u − 1 ) . (52) \sqrt{-1}\Lambda_{H_i}\overline{\partial}\partial u\geq C(e^u-1). \tag*{(52)} − 1 Λ H i ∂ ∂ u ≥ C ( e u − 1 ) . (52)
By applying Proposition 24 to inequality (52), it follows that u u u is essentially bounded above and hence so is ∑ j = 1 r H j ⊗ H i − 1 \sum_{j=1}^{r}H_j\otimes H_i^{-1} ∑ j = 1 r H j ⊗ H i − 1 . Since ∑ j = 1 r H j ⊗ H i − 1 \sum_{j=1}^{r}H_j\otimes H_i^{-1} ∑ j = 1 r H j ⊗ H i − 1 is locally a sum of a smooth function and a subharmonic function, a local mean value inequality implies that it is indeed a bounded function. Suppose moreover that e φ e^\varphi e φ is continuous. Since a locally bounded subharmonic function belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 (cf. [21 Section 1.1.1] ), e φ e^\varphi e φ belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . Therefore u u u is a continuous function belonging to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 , and thus by applying Proposition 24 to inequality (52), we have u ≤ 0 u\leq0 u ≤ 0 . This implies that ∑ j = 1 r H j ⊗ H i − 1 ≤ C 2 + 2 C 1 2 \sum_{j=1}^{r}H_j\otimes H_i^{-1}\leq\frac{C_2+2}{C_1^2} ∑ j = 1 r H j ⊗ H i − 1 ≤ C 1 2 C 2 + 2 . □ \square □
Let h ′ = ( h 1 ′ , … , h r ′ ) h'=(h'_1,\ldots,h'_r) h ′ = ( h 1 ′ , … , h r ′ ) be another solution to equation (5) satisfying det ( h ′ ) = 1 \det(h')=1 det ( h ′ ) = 1 . We denote by s = ( s 1 , … , s r ) s=(s_1,\ldots,s_r) s = ( s 1 , … , s r ) the endomorphism determined by h ( s ⋅ , ⋅ ) = h ′ ( ⋅ , ⋅ ) h(s\cdot,\cdot)=h'(\cdot,\cdot) h ( s ⋅ , ⋅ ) = h ′ ( ⋅ , ⋅ ) . We set s 0 : = s r s_0:=s_r s 0 := s r .
Lemma 38. The endomorphism s s s satisfies the following inequality in the sense of distributions:
− 1 ∂ ∂ ‾ ∑ j = 1 r s j ≥ ∑ j = 1 r s j − 1 − 1 ( s j − 1 − s j ) 2 vol ( H j − 1 ) . (53) \sqrt{-1}\partial\overline{\partial}\sum_{j=1}^{r}s_j\geq\sum_{j=1}^{r}s_{j-1}^{-1}(s_{j-1}-s_j)^2\operatorname{vol}(H_{j-1}). \tag*{(53)} − 1 ∂ ∂ j = 1 ∑ r s j ≥ j = 1 ∑ r s j − 1 − 1 ( s j − 1 − s j ) 2 vol ( H j − 1 ) . (53)
Proof. By a direct computation, we have
− 1 ∂ ∂ ˉ ∑ j = 1 r s j = − 1 ∂ ∂ ˉ ∑ j = 1 r e log s j = − 1 ∂ ∑ j = 1 r ( ∂ ˉ log s j ) e log s j = − 1 ∑ j = 1 r ∂ ( log s j ) ∧ ∂ ˉ ( log s j ) e log s j + − 1 ∑ j = 1 r ( ∂ ∂ ˉ log s j ) e log s j ≥ − 1 ∑ j = 1 r s j ∂ ∂ ˉ ( log s j ) = ∑ j = 1 r s j ( − 1 F h j − − 1 F h j ′ ) = ∑ j = 1 r s j ( − vol ( H j − 1 ) + vol ( H j ) + vol ( H j − 1 ′ ) − vol ( H j ′ ) ) . \begin{aligned}
\sqrt{-1}\partial\bar\partial\sum_{j=1}^{r}s_j
&=\sqrt{-1}\partial\bar\partial\sum_{j=1}^{r}e^{\log s_j}\\
&=\sqrt{-1}\partial\sum_{j=1}^{r}(\bar\partial\log s_j)e^{\log s_j}\\
&=\sqrt{-1}\sum_{j=1}^{r}\partial(\log s_j)\wedge\bar\partial(\log s_j)e^{\log s_j}
+\sqrt{-1}\sum_{j=1}^{r}(\partial\bar\partial\log s_j)e^{\log s_j}\\
&\geq\sqrt{-1}\sum_{j=1}^{r}s_j\partial\bar\partial(\log s_j)\\
&=\sum_{j=1}^{r}s_j(\sqrt{-1}F_{h_j}-\sqrt{-1}F_{h'_j})\\
&=\sum_{j=1}^{r}s_j(-\operatorname{vol}(H_{j-1})+\operatorname{vol}(H_j)+\operatorname{vol}(H'_{j-1})-\operatorname{vol}(H'_j)).
\end{aligned} − 1 ∂ ∂ ˉ j = 1 ∑ r s j = − 1 ∂ ∂ ˉ j = 1 ∑ r e l o g s j = − 1 ∂ j = 1 ∑ r ( ∂ ˉ log s j ) e l o g s j = − 1 j = 1 ∑ r ∂ ( log s j ) ∧ ∂ ˉ ( log s j ) e l o g s j + − 1 j = 1 ∑ r ( ∂ ∂ ˉ log s j ) e l o g s j ≥ − 1 j = 1 ∑ r s j ∂ ∂ ˉ ( log s j ) = j = 1 ∑ r s j ( − 1 F h j − − 1 F h j ′ ) = j = 1 ∑ r s j ( − vol ( H j − 1 ) + vol ( H j ) + vol ( H j − 1 ′ ) − vol ( H j ′ )) .
On the other hand, since we have s j ⋅ s j − 1 − 1 = H j − 1 − 1 ⊗ H j − 1 ′ s_j\cdot s_{j-1}^{-1}=H_{j-1}^{-1}\otimes H'_{j-1} s j ⋅ s j − 1 − 1 = H j − 1 − 1 ⊗ H j − 1 ′ for each j = 1 , … , r j=1,\ldots,r j = 1 , … , r , it holds that
s j − 1 − 1 ( s j − s j − 1 ) 2 vol ( H j − 1 ) = ( s j − s j − 1 ) ( s j s j − 1 − 1 − 1 ) vol ( H j − 1 ) = ( s j − s j − 1 ) ( vol ( H j − 1 ′ ) − vol ( H j − 1 ) ) . \begin{aligned}
s_{j-1}^{-1}(s_j-s_{j-1})^2\operatorname{vol}(H_{j-1})
&=(s_j-s_{j-1})(s_js_{j-1}^{-1}-1)\operatorname{vol}(H_{j-1})\\
&=(s_j-s_{j-1})(\operatorname{vol}(H'_{j-1})-\operatorname{vol}(H_{j-1})).
\end{aligned} s j − 1 − 1 ( s j − s j − 1 ) 2 vol ( H j − 1 ) = ( s j − s j − 1 ) ( s j s j − 1 − 1 − 1 ) vol ( H j − 1 ) = ( s j − s j − 1 ) ( vol ( H j − 1 ′ ) − vol ( H j − 1 )) .
Therefore, we have
∑ j = 1 r s j − 1 − 1 ( s j − s j − 1 ) 2 vol ( H j − 1 ) = ∑ j = 1 r ( s j − s j − 1 ) ( vol ( H j − 1 ′ ) − vol ( H j − 1 ) ) = ∑ j = 1 r ( s j vol ( H j − 1 ′ ) + s j − 1 vol ( H j − 1 ) − s j − 1 vol ( H j − 1 ′ ) − s j vol ( H j − 1 ) ) = ∑ j = 1 r s j ( vol ( H j − 1 ′ ) + vol ( H j ) − vol ( H j ′ ) − vol ( H j − 1 ) ) = ∑ j = 1 r s j ( − vol ( H j − 1 ) + vol ( H j ) + vol ( H j − 1 ′ ) − vol ( H j ′ ) ) . \begin{aligned}
\sum_{j=1}^{r}s_{j-1}^{-1}(s_j-s_{j-1})^2\operatorname{vol}(H_{j-1})
&=\sum_{j=1}^{r}(s_j-s_{j-1})(\operatorname{vol}(H'_{j-1})-\operatorname{vol}(H_{j-1}))\\
&=\sum_{j=1}^{r}(s_j\operatorname{vol}(H'_{j-1})+s_{j-1}\operatorname{vol}(H_{j-1})-s_{j-1}\operatorname{vol}(H'_{j-1})-s_j\operatorname{vol}(H_{j-1}))\\
&=\sum_{j=1}^{r}s_j(\operatorname{vol}(H'_{j-1})+\operatorname{vol}(H_j)-\operatorname{vol}(H'_j)-\operatorname{vol}(H_{j-1}))\\
&=\sum_{j=1}^{r}s_j(-\operatorname{vol}(H_{j-1})+\operatorname{vol}(H_j)+\operatorname{vol}(H'_{j-1})-\operatorname{vol}(H'_j)).
\end{aligned} j = 1 ∑ r s j − 1 − 1 ( s j − s j − 1 ) 2 vol ( H j − 1 ) = j = 1 ∑ r ( s j − s j − 1 ) ( vol ( H j − 1 ′ ) − vol ( H j − 1 )) = j = 1 ∑ r ( s j vol ( H j − 1 ′ ) + s j − 1 vol ( H j − 1 ) − s j − 1 vol ( H j − 1 ′ ) − s j vol ( H j − 1 )) = j = 1 ∑ r s j ( vol ( H j − 1 ′ ) + vol ( H j ) − vol ( H j ′ ) − vol ( H j − 1 )) = j = 1 ∑ r s j ( − vol ( H j − 1 ) + vol ( H j ) + vol ( H j − 1 ′ ) − vol ( H j ′ )) .
This implies the claim.
Then we prove Theorem 1.
Proof of Theorem 1. Let H 1 ′ , … , H r ′ H'_1,\ldots,H'_r H 1 ′ , … , H r ′ be the Hermitian metrics constructed from h ′ h' h ′ as in Section 2.3. From Corollaries 35 and 37, it follows that H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 and H 1 ′ , … , H r − 1 ′ H'_1,\ldots,H'_{r-1} H 1 ′ , … , H r − 1 ′ are mutually bounded. In particular, s 1 , … , s r s_1,\ldots,s_r s 1 , … , s r are bounded functions on X X X . We choose one of H 1 , … , H r − 1 , H 1 ′ , … , H r − 1 ′ H_1,\ldots,H_{r-1},H'_1,\ldots,H'_{r-1} H 1 , … , H r − 1 , H 1 ′ , … , H r − 1 ′ and denote it by H H H . Since H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 and H 1 ′ , … , H r − 1 ′ H'_1,\ldots,H'_{r-1} H 1 ′ , … , H r − 1 ′ are solutions to equation (6), the Gaussian curvature associated with H H H is bounded below (cf. [43 Lemma 2.4] ). Fix ϵ > 0 \epsilon>0 ϵ > 0 . By Proposition 10 , there exists a smooth Hermitian metric H ~ \widetilde{H} H on K X − 1 → X K_X^{-1} \to X K X − 1 → X such that e − ϵ H ≤ H ~ ≤ e ϵ H e^{-\epsilon H} \leq\widetilde{H} \leq e^{\epsilon H} e − ϵH ≤ H ≤ e ϵH and the Ricci curvature of H ~ \widetilde{H} H is bounded below. From inequality (53) , we obtain
− 1 Λ H ~ ∂ ∂ ‾ ∑ j = 1 r s j ≥ ∑ j = 1 r s j − 1 ( s j − 1 − s j ) 2 H j − 1 ⊗ H ~ − 1 ≥ ∑ j = 1 r − 1 s j − 1 ( s j − s j + 1 ) 2 H j ⊗ H ~ − 1 . (54) \sqrt{-1}\Lambda_{\widetilde{H}}\partial\overline{\partial}\sum_{j=1}^{r}s_j \geq\sum_{j=1}^{r}s_j^{-1}(s_{j-1}-s_j)^2H_{j-1}\otimes\widetilde{H}^{-1} \geq\sum_{j=1}^{r-1}s_j^{-1}(s_j-s_{j+1})^2H_j\otimes\widetilde{H}^{-1}. \tag*{(54)} − 1 Λ H ∂ ∂ j = 1 ∑ r s j ≥ j = 1 ∑ r s j − 1 ( s j − 1 − s j ) 2 H j − 1 ⊗ H − 1 ≥ j = 1 ∑ r − 1 s j − 1 ( s j − s j + 1 ) 2 H j ⊗ H − 1 . (54)
Applying Proposition 22 to inequality (54) , we obtain, for each m ∈ Z > 0 m \in\mathbb{Z}_{>0} m ∈ Z > 0 , a point x m ∈ X x_m \in X x m ∈ X such that
( ∑ j = 1 r s j ) ( x m ) ≥ sup X ( ∑ j = 1 r s j ) − 1 / m , \left(\sum_{j=1}^{r}s_j\right)(x_m) \geq\sup_X\left(\sum_{j=1}^{r}s_j\right)-1/m, ( j = 1 ∑ r s j ) ( x m ) ≥ X sup ( j = 1 ∑ r s j ) − 1/ m ,
∑ j = 1 r − 1 ( s j − 1 ( s j − s j + 1 ) 2 ) ( x m ) ( H j ⊗ H ~ − 1 ) ( x m ) ≤ 1 / m . \sum_{j=1}^{r-1}\left(s_j^{-1}(s_j-s_{j+1})^2\right)(x_m)(H_j\otimes\widetilde{H}^{-1})(x_m) \leq1/m. j = 1 ∑ r − 1 ( s j − 1 ( s j − s j + 1 ) 2 ) ( x m ) ( H j ⊗ H − 1 ) ( x m ) ≤ 1/ m .
Around each x m x_m x m , we choose a local coordinate z z z such that H ~ ( ∂ ∂ z , ∂ ∂ z ) ( x m ) = 1 \widetilde{H}\left(\frac{\partial}{\partial z},\frac{\partial}{\partial z}\right)(x_m)=1 H ( ∂ z ∂ , ∂ z ∂ ) ( x m ) = 1 . Since H 1 , … , H r − 1 , H ~ H_1,\ldots,H_{r-1},\widetilde{H} H 1 , … , H r − 1 , H are mutually bounded, there exists a constant B > 0 B>0 B > 0 such that H j ( ∂ ∂ z , ∂ ∂ z ) ( x m ) − 1 ≤ B H_j\left(\frac{\partial}{\partial z},\frac{\partial}{\partial z}\right)(x_m)^{-1}\leq B H j ( ∂ z ∂ , ∂ z ∂ ) ( x m ) − 1 ≤ B for all j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 and all m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 . By Lemma 30 , there exists a positive constant C 1 > 0 C_1>0 C 1 > 0 such that
∑ j = 1 r ∣ s j ( x m ) − 1 ∣ ≤ C 1 / m 1 / 2 \sum_{j=1}^{r}|s_j(x_m)-1|\leq C_1/m^{1/2} j = 1 ∑ r ∣ s j ( x m ) − 1∣ ≤ C 1 / m 1/2
for all m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 . This implies that sup X ∑ j = 1 r s j ≤ r \sup_X\sum_{j=1}^{r}s_j\leq r sup X ∑ j = 1 r s j ≤ r . On the other hand, since s 1 ⋯ s r = 1 s_1\cdots s_r=1 s 1 ⋯ s r = 1 , we have ∑ j = 1 r s j ≥ r \sum_{j=1}^{r}s_j\geq r ∑ j = 1 r s j ≥ r . Therefore, ∑ j = 1 r s j = r \sum_{j=1}^{r}s_j=r ∑ j = 1 r s j = r on X X X . By the arithmetic mean–geometric mean inequality, we conclude that s 1 = ⋯ = s r = 1 s_1=\cdots=s_r=1 s 1 = ⋯ = s r = 1 . Hence h = h ′ h=h' h = h ′ . □
Pullback
Extending [43 Section 3.4.7] , this section establishes some properties of pullbacks of solutions to equation (5) by holomorphic maps. Let π : Y → X \pi:Y\to X π : Y → X be a holomorphic map whose derivative is nowhere vanishing on Y Y Y . Since the derivative is nowhere vanishing on Y Y Y , we have a natural isomorphism π ∗ K X ≃ K Y \pi^*K_X\simeq K_Y π ∗ K X ≃ K Y . Choose a square root K X 1 / 2 K_X^{1/2} K X 1/2 of the holomorphic vector bundle K X → X K_X\to X K X → X . Since we have π ∗ K X 1 / 2 ⊗ π ∗ K X 1 / 2 ≃ π ∗ K X ≃ K Y \pi^*K_X^{1/2}\otimes\pi^*K_X^{1/2}\simeq\pi^*K_X\simeq K_Y π ∗ K X 1/2 ⊗ π ∗ K X 1/2 ≃ π ∗ K X ≃ K Y , by setting K Y 1 / 2 : = π ∗ K X 1 / 2 K_Y^{1/2}:=\pi^*K_X^{1/2} K Y 1/2 := π ∗ K X 1/2 , there is a natural isomorphism
π ∗ ( K X , r ) ≃ K Y , r , \pi^*(\mathbb{K}_{X,r})\simeq\mathbb{K}_{Y,r}, π ∗ ( K X , r ) ≃ K Y , r ,
where we have used the notation K X , r : = ⨁ j = 1 r K X r − ( 2 j − 1 ) 2 \mathbb{K}_{X,r}:=\bigoplus_{j=1}^{r}K_X^{\frac{r-(2j-1)}{2}} K X , r := ⨁ j = 1 r K X 2 r − ( 2 j − 1 ) and K Y , r : = ⨁ j = 1 r K Y r − ( 2 j − 1 ) 2 \mathbb{K}_{Y,r}:=\bigoplus_{j=1}^{r}K_Y^{\frac{r-(2j-1)}{2}} K Y , r := ⨁ j = 1 r K Y 2 r − ( 2 j − 1 ) to explicitly denote the underlying Riemann surfaces. This notation is adopted solely for this subsection and Section 3.9. For each diagonal Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on K X , r \mathbb{K}_{X,r} K X , r , we denote by π ∗ h = ( π ∗ h 1 , … , π ∗ h r ) \pi^*h=(\pi^*h_1,\ldots,\pi^*h_r) π ∗ h = ( π ∗ h 1 , … , π ∗ h r ) the Hermitian metric on K Y , r \mathbb{K}_{Y,r} K Y , r induced by the above isomorphism. Let φ \varphi φ be a subharmonic weight function on K X K_X K X . We denote by π ∗ φ \pi^*\varphi π ∗ φ the induced weight function on K Y K_Y K Y . Let H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r be the Hermitian metrics constructed from a diagonal Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) and the semipositive metric e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ as in Section 2.3. Pulling back equation (5) for h h h by π \pi π , we obtain
− 1 F π ∗ h j + vol ( π ∗ H j − 1 ) − vol ( π ∗ H j ) = 0 for j = 1 , … , r − 1. \sqrt{-1}F_{\pi^*h_j}+\operatorname{vol}(\pi^*H_{j-1})-\operatorname{vol}(\pi^*H_j)=0 \quad\text{for }j=1,\ldots,r-1. − 1 F π ∗ h j + vol ( π ∗ H j − 1 ) − vol ( π ∗ H j ) = 0 for j = 1 , … , r − 1.
This implies
Proposition 39. If h h h solves equation (5) associated to φ \varphi φ , then π ∗ h \pi^*h π ∗ h solves equation (5) associated to π ∗ φ \pi^*\varphi π ∗ φ . If π \pi π is surjective, the converse also holds.
We next discuss completeness. By Ambrose’s theorem [3 ] (see also [57 ] ), a local isometry from a complete Riemannian manifold to a connected Riemannian manifold is a surjective covering map, and the target manifold is complete. Conversely, completeness is preserved under Riemannian coverings. These facts also apply to pullbacks of continuous Hermitian metrics on Riemann surfaces: for each continuous metric g g g on X X X , choose a smooth metric g ~ \widetilde{g} g satisfying 1 2 g ≤ g ~ ≤ 2 g \frac{1}{2}g \leq\widetilde{g} \leq2g 2 1 g ≤ g ≤ 2 g . These comparisons preserve completeness and remain valid after pullback. Applying this criterion to the metrics induced by H 1 , … , H r − 1 H_1,\ldots,H_{r-1} H 1 , … , H r − 1 yields the following:
Proposition 40. For each diagonal Hermitian metric h h h on K X , r K_{X,r} K X , r , the pullback π ∗ h \pi^*h π ∗ h is complete if and only if π \pi π is a surjective covering map and h h h is complete. In particular, if π \pi π is a surjective covering map and h h h is a solution to equation (5) associated to φ \varphi φ , then h h h is a complete solution if and only if π ∗ h \pi^*h π ∗ h is a complete solution associated to π ∗ φ \pi^*\varphi π ∗ φ .
Supersolution and subsolution to equation (5)
In this subsection, we establish some properties of subsolutions and supersolutions to equation (5) . Let e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ be a semipositive singular Hermitian metric on K X → X K_X \to X K X → X with a smooth reference metric h ∗ h_* h ∗ . For each j = 1 , … , r j=1,\ldots,r j = 1 , … , r , we denote by h ∗ , j : = h ∗ r − ( 2 j − 1 ) 2 h_{*,j}:=h_*^{\frac{r-(2j-1)}{2}} h ∗ , j := h ∗ 2 r − ( 2 j − 1 ) the real diagonal Hermitian metric on K X r − ( 2 j − 1 ) 2 K_X^{\frac{r-(2j-1)}{2}} K X 2 r − ( 2 j − 1 ) induced by h ∗ h_* h ∗ . Let h = ( h 1 , … , h r ) = ( e − φ 1 h ∗ , 1 , e − φ 2 h ∗ , 2 , … , e − φ r h ∗ , r ) h=(h_1,\ldots,h_r)=(e^{-\varphi_1}h_{*,1},e^{-\varphi_2}h_{*,2},\ldots,e^{-\varphi_r}h_{*,r}) h = ( h 1 , … , h r ) = ( e − φ 1 h ∗ , 1 , e − φ 2 h ∗ , 2 , … , e − φ r h ∗ , r ) be a real diagonal Hermitian metric on K r → X K_r\to X K r → X . We set H j : = h j − 1 ⊗ h j + 1 H_j:=h_j^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 and H 0 = H r : = e r φ h ∗ − r ⊗ h r − 1 ⊗ h 1 H_0=H_r:=e^{r\varphi}h_*^{-r}\otimes h_r^{-1}\otimes h_1 H 0 = H r := e r φ h ∗ − r ⊗ h r − 1 ⊗ h 1 . We suppose that φ 1 , … , φ n , H 0 , … , H n \varphi_1,\ldots,\varphi_n,H_0,\ldots,H_n φ 1 , … , φ n , H 0 , … , H n belong to L l o c 1 L^1_{\mathrm{loc}} L loc 1 . The notions of supersolutions and subsolutions to equation (5) are then defined as follows (see also [43 Definition 5.1] ):
Definition 41. We say that h h h is a supersolution (resp. subsolution ) to equation (5) if it satisfies the following differential inequalities in the weak sense:
− 1 F h j + vol ( H j − 1 ) − vol ( H j ) ≥ 0 ( resp. ≤ 0 ) for all j = 1 , … , n . \sqrt{-1}F_{h_j}+\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)\geq0 \quad(\text{resp. }\leq0)\quad\text{for all }j=1,\ldots,n. − 1 F h j + vol ( H j − 1 ) − vol ( H j ) ≥ 0 ( resp. ≤ 0 ) for all j = 1 , … , n .
Example 42. Define a diagonal Hermitian metric h ( φ ) : = ( h 1 ( φ ) , … , h r ( φ ) ) h^{(\varphi)}:=(h_1^{(\varphi)},\ldots,h_r^{(\varphi)}) h ( φ ) := ( h 1 ( φ ) , … , h r ( φ ) ) by h j ( φ ) : = ( e − φ h ∗ ) r − ( 2 j − 1 ) 2 h_j^{(\varphi)}:=(e^{-\varphi}h_*)^{\frac{r-(2j-1)}{2}} h j ( φ ) := ( e − φ h ∗ ) 2 r − ( 2 j − 1 ) for each j = 1 , … , r j=1,\ldots,r j = 1 , … , r . For each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , define H j ( φ ) : = h j ( φ ) − 1 ⊗ h j + 1 ( φ ) H_j^{(\varphi)}:=h_j^{(\varphi)-1}\otimes h_{j+1}^{(\varphi)} H j ( φ ) := h j ( φ ) − 1 ⊗ h j + 1 ( φ ) and define H r ( φ ) : = ( e − φ h ∗ ) − r ⊗ H 1 ( φ ) − 1 ⊗ ⋯ ⊗ H r − 1 ( φ ) − 1 H_r^{(\varphi)}:=(e^{-\varphi}h_*)^{-r}\otimes H_1^{(\varphi)-1}\otimes\cdots\otimes H_{r-1}^{(\varphi)-1} H r ( φ ) := ( e − φ h ∗ ) − r ⊗ H 1 ( φ ) − 1 ⊗ ⋯ ⊗ H r − 1 ( φ ) − 1 . Then H 1 ( φ ) = ⋯ = H r ( φ ) = e φ h ∗ − 1 H_1^{(\varphi)}=\cdots=H_r^{(\varphi)}=e^{\varphi}h_*^{-1} H 1 ( φ ) = ⋯ = H r ( φ ) = e φ h ∗ − 1 . Since each h j ( φ ) h_j^{(\varphi)} h j ( φ ) , j = 1 , … , n j=1,\ldots,n j = 1 , … , n , has semipositive curvature, h ( φ ) h^{(\varphi)} h ( φ ) is a supersolution to equation (5) .
Example 43. Let h + h_+ h + and h − h_- h − be real diagonal Hermitian metrics such that h + h_+ h + is a supersolution and h − h_- h − is a subsolution to equation (5) associated with φ \varphi φ . Suppose that φ ≤ φ ′ \varphi\leq\varphi' φ ≤ φ ′ . Then h + h_+ h + is also a supersolution to equation (5) associated with φ ′ \varphi' φ ′ . Similarly, if φ ≥ φ ′ \varphi\geq\varphi' φ ≥ φ ′ , then h − h_- h − is also a subsolution to equation (5) associated with φ ′ \varphi' φ ′ .
Example 44. Let h + = ( h + , 1 , … , h + , r ) h_+=(h_{+,1},\ldots,h_{+,r}) h + = ( h + , 1 , … , h + , r ) be a real diagonal supersolution to equation (5) , and let h − = ( h − , 1 , … , h − , r ) h_-=(h_{-,1},\ldots,h_{-,r}) h − = ( h − , 1 , … , h − , r ) be a real diagonal subsolution to equation (5) . Let φ + \varphi_+ φ + be a nonpositive
bounded subharmonic function, and let ϕ − \phi_{-} ϕ − be a nonnegative bounded superharmonic function. For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , define
h ~ + , j : = e − ϕ + h + , j , \widetilde{h}_{+,j}:=e^{-\phi_{+}}h_{+,j}, h + , j := e − ϕ + h + , j ,
h ~ − , j : = e − ϕ − h − , j . \widetilde{h}_{-,j}:=e^{-\phi_{-}}h_{-,j}. h − , j := e − ϕ − h − , j .
For the remaining indices j = n + 1 , … , r j=n+1,\ldots,r j = n + 1 , … , r , we define h ~ + , j \widetilde{h}_{+,j} h + , j and h ~ − , j \widetilde{h}_{-,j} h − , j so that h ~ + : = ( h ~ + , 1 , … , h ~ + , r ) \widetilde{h}_{+}:=(\widetilde{h}_{+,1},\ldots,\widetilde{h}_{+,r}) h + := ( h + , 1 , … , h + , r ) and h ~ − : = ( h ~ − , 1 , … , h ~ − , r ) \widetilde{h}_{-}:=(\widetilde{h}_{-,1},\ldots,\widetilde{h}_{-,r}) h − := ( h − , 1 , … , h − , r ) are real diagonal Hermitian metrics. Then h ~ + \widetilde{h}_{+} h + and h ~ − \widetilde{h}_{-} h − are a supersolution and a subsolution to equation (5) , respectively (cf. [43 Proof of Proposition 5.9] ).
For the proof of Proposition 47, following the argument in [20 Proof of Lemma 1.5] , we first establish the following lemma.
Lemma 45. Let U ⊆ R d U \subseteq\mathbb{R}^{d} U ⊆ R d be an open subset, and let F , G : U × R → R F,G:U\times\mathbb{R}\to\mathbb{R} F , G : U × R → R be Borel measurable functions. We denote by g E g_{\mathrm{E}} g E the Euclidean metric and by Δ g E : = − d ∗ d \Delta_{g_{\mathrm{E}}}:=-d^{*}d Δ g E := − d ∗ d the Laplacian. Then the following holds:
(i) Let u 0 , u 1 : U → R u_{0},u_{1}:U\to\mathbb{R} u 0 , u 1 : U → R be locally integrable functions. Suppose that F ( ⋅ , u i ( ⋅ ) ) F(\cdot,u_{i}(\cdot)) F ( ⋅ , u i ( ⋅ )) is locally integrable for i = 0 , 1 i=0,1 i = 0 , 1 , and u 0 u_{0} u 0 and u 1 u_{1} u 1 satisfy the following in the sense of distributions for i = 0 , 1 i=0,1 i = 0 , 1 :
Δ g E u i ≤ F ( ⋅ , u i ( ⋅ ) ) \Delta_{g_{\mathrm{E}}}u_{i}\leq F(\cdot,u_{i}(\cdot)) Δ g E u i ≤ F ( ⋅ , u i ( ⋅ )) Then u min : = min { u 0 , u 1 } u_{\min}:=\min\{u_{0},u_{1}\} u m i n := min { u 0 , u 1 } and F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) are also locally integrable functions and satisfy
Δ g E u min ≤ F ( ⋅ , u min ( ⋅ ) ) (55) \Delta_{g_{\mathrm{E}}}u_{\min}\leq F(\cdot,u_{\min}(\cdot)) \tag*{(55)} Δ g E u m i n ≤ F ( ⋅ , u m i n ( ⋅ )) (55) (ii) Let v 0 , v 1 : U → R v_{0},v_{1}:U\to\mathbb{R} v 0 , v 1 : U → R be locally integrable functions. Suppose that G ( ⋅ , v i ( ⋅ ) ) G(\cdot,v_{i}(\cdot)) G ( ⋅ , v i ( ⋅ )) is locally integrable for i = 0 , 1 i=0,1 i = 0 , 1 , and v 0 v_{0} v 0 and v 1 v_{1} v 1 satisfy the following in the sense of distributions for i = 0 , 1 i=0,1 i = 0 , 1 :
Δ g E v i ≥ G ( ⋅ , v i ( ⋅ ) ) \Delta_{g_{\mathrm{E}}}v_{i}\geq G(\cdot,v_{i}(\cdot)) Δ g E v i ≥ G ( ⋅ , v i ( ⋅ )) Then v max : = max { v 0 , v 1 } v_{\max}:=\max\{v_{0},v_{1}\} v m a x := max { v 0 , v 1 } and G ( ⋅ , v max ( ⋅ ) ) G(\cdot,v_{\max}(\cdot)) G ( ⋅ , v m a x ( ⋅ )) are also locally integrable functions and satisfy
Δ g E v max ≥ G ( ⋅ , v max ( ⋅ ) ) . \Delta_{g_{\mathrm{E}}}v_{\max}\geq G(\cdot,v_{\max}(\cdot)). Δ g E v m a x ≥ G ( ⋅ , v m a x ( ⋅ )) .
Proof. We first note that assertion (ii) follows immediately from assertion (i). Indeed, assuming assertion (i), set
u 0 : = − v 0 , u 1 : = − v 1 , F ( x , t ) : = − G ( x , − t ) . u_{0}:=-v_{0},\quad u_{1}:=-v_{1},\quad F(x,t):=-G(x,-t). u 0 := − v 0 , u 1 := − v 1 , F ( x , t ) := − G ( x , − t ) .
Then assertion (ii) follows by applying assertion (i) to u 0 u_{0} u 0 and u 1 u_{1} u 1 . Therefore, it suffices to prove assertion (i). We begin by proving u min u_{\min} u m i n and F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) are both locally integrable. Indeed, since
∣ u min ∣ ≤ ∣ u 0 ∣ + ∣ u 1 ∣ , |u_{\min}|\leq|u_{0}|+|u_{1}|, ∣ u m i n ∣ ≤ ∣ u 0 ∣ + ∣ u 1 ∣ ,
we have u min ∈ L l o c 1 ( U ) u_{\min}\in L^{1}_{\mathrm{loc}}(U) u m i n ∈ L loc 1 ( U ) . Also, since u min ( x ) u_{\min}(x) u m i n ( x ) coincides with either u 0 ( x ) u_{0}(x) u 0 ( x ) or u 1 ( x ) u_{1}(x) u 1 ( x ) for each x ∈ U x\in U x ∈ U , we have
∣ F ( x , u min ( x ) ) ∣ ≤ ∣ F ( x , u 0 ( x ) ) ∣ + ∣ F ( x , u 1 ( x ) ) ∣ , |F(x,u_{\min}(x))|\leq|F(x,u_{0}(x))|+|F(x,u_{1}(x))|, ∣ F ( x , u m i n ( x )) ∣ ≤ ∣ F ( x , u 0 ( x )) ∣ + ∣ F ( x , u 1 ( x )) ∣ ,
which proves the local integrability of F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) . We next prove (55) . Let V ⊆ U V \subseteq U V ⊆ U be an arbitrary relatively compact open subset. Let ( ρ ε ) ε > 0 (\rho_\varepsilon)_{\varepsilon>0} ( ρ ε ) ε > 0 be a standard mollifier. For each ε > 0 \varepsilon>0 ε > 0 and each i = 0 , 1 i=0,1 i = 0 , 1 , we set
u i , ε : = u i ∗ ρ ε , u_{i,\varepsilon}:=u_i*\rho_\varepsilon, u i , ε := u i ∗ ρ ε ,
f i , ε : = F ( ⋅ , u i ( ⋅ ) ) ∗ ρ ε , f_{i,\varepsilon}:=F(\cdot,u_i(\cdot))*\rho_\varepsilon, f i , ε := F ( ⋅ , u i ( ⋅ )) ∗ ρ ε ,
where ∗ * ∗ denotes the convolution. We take ε \varepsilon ε small enough so that u i , ε u_{i,\varepsilon} u i , ε and f i , ε f_{i,\varepsilon} f i , ε are defined on V V V . Note that, by the standard properties of mollification, for each i = 0 , 1 i=0,1 i = 0 , 1 , u i , ε → u i u_{i,\varepsilon}\to u_i u i , ε → u i and f i , ε → f i f_{i,\varepsilon}\to f_i f i , ε → f i in L 1 ( V ) L^1(V) L 1 ( V ) as ε ↘ 0 \varepsilon\searrow0 ε ↘ 0 , where f i : = F ( ⋅ , u i ( ⋅ ) ) f_i:=F(\cdot,u_i(\cdot)) f i := F ( ⋅ , u i ( ⋅ )) . Moreover, u i , ε ( x ) → u i ( x ) u_{i,\varepsilon}(x)\to u_i(x) u i , ε ( x ) → u i ( x ) , i = 0 , 1 i=0,1 i = 0 , 1 , at every common Lebesgue point x ∈ V x\in V x ∈ V of u 0 u_0 u 0 and u 1 u_1 u 1 , and hence for almost every x ∈ V x\in V x ∈ V . Also, since Δ g E u i ≤ f i \Delta_{g_E}u_i\leq f_i Δ g E u i ≤ f i in the sense of distributions and ρ ε \rho_\varepsilon ρ ε is nonnegative, convolution with ρ ε \rho_\varepsilon ρ ε yields Δ g E u i , ε ≤ f i , ε \Delta_{g_E}u_{i,\varepsilon}\leq f_{i,\varepsilon} Δ g E u i , ε ≤ f i , ε pointwise on V V V . We define
u ε : = min { u 0 , ε , u 1 , ε } = 1 2 ( u 0 , ε + u 1 , ε − ∣ u 0 , ε − u 1 , ε ∣ ) . u_\varepsilon:=\min\{u_{0,\varepsilon},u_{1,\varepsilon}\}=\frac{1}{2}\left(u_{0,\varepsilon}+u_{1,\varepsilon}-\lvert u_{0,\varepsilon}-u_{1,\varepsilon}\rvert\right). u ε := min { u 0 , ε , u 1 , ε } = 2 1 ( u 0 , ε + u 1 , ε − ∣ u 0 , ε − u 1 , ε ∣ ) .
Applying the inequality ∣ min { a , b } − min { c , d } ∣ ≤ ∣ a − c ∣ + ∣ b − d ∣ \lvert\min\{a,b\}-\min\{c,d\}\rvert\leq\lvert a-c\rvert+\lvert b-d\rvert ∣ min { a , b } − min { c , d }∣ ≤ ∣ a − c ∣ + ∣ b − d ∣ for arbitrary real numbers a , b , c , d a,b,c,d a , b , c , d , we obtain
∣ u ε − u min ∣ ≤ ∣ u 0 , ε − u 0 ∣ + ∣ u 1 , ε − u 1 ∣ . \lvert u_\varepsilon-u_{\min}\rvert\leq\lvert u_{0,\varepsilon}-u_0\rvert+\lvert u_{1,\varepsilon}-u_1\rvert. ∣ u ε − u m i n ∣ ≤ ∣ u 0 , ε − u 0 ∣ + ∣ u 1 , ε − u 1 ∣ .
and hence, u ε u_\varepsilon u ε converges to u min u_{\min} u m i n in L 1 ( V ) L^1(V) L 1 ( V ) . Set
w ε : = u 0 , ε − u 1 , ε . w_\varepsilon:=u_{0,\varepsilon}-u_{1,\varepsilon}. w ε := u 0 , ε − u 1 , ε .
By Kato’s inequality,
Δ g E ∣ w ε ∣ ≥ sgn ( w ε ) Δ g E w ε \Delta_{g_E}\lvert w_\varepsilon\rvert\geq\operatorname{sgn}(w_\varepsilon)\Delta_{g_E}w_\varepsilon Δ g E ∣ w ε ∣ ≥ sgn ( w ε ) Δ g E w ε
in the sense of distributions on V V V , where sgn ( t ) \operatorname{sgn}(t) sgn ( t ) (t ∈ R t\in\mathbb{R} t ∈ R ) is defined by
sgn ( t ) : = { t ∣ t ∣ if t ≠ 0 , 0 if t = 0. \operatorname{sgn}(t):=
\begin{cases}
\frac{t}{\lvert t\rvert} & \text{if } t\ne0,\\
0 & \text{if } t=0.
\end{cases} sgn ( t ) := { ∣ t ∣ t 0 if t = 0 , if t = 0.
Hence,
Δ g E u ε ≤ α ε Δ g E u 0 , ε + ( 1 − α ε ) Δ g E u 1 , ε \Delta_{g_E}u_\varepsilon\leq\alpha_\varepsilon\Delta_{g_E}u_{0,\varepsilon}+(1-\alpha_\varepsilon)\Delta_{g_E}u_{1,\varepsilon} Δ g E u ε ≤ α ε Δ g E u 0 , ε + ( 1 − α ε ) Δ g E u 1 , ε
in the sense of distributions on V V V , where α ε : = 1 2 ( 1 − sgn ( w ε ) ) \alpha_\varepsilon:=\frac{1}{2}(1-\operatorname{sgn}(w_\varepsilon)) α ε := 2 1 ( 1 − sgn ( w ε )) . On the other hand, since
Δ g E u i , ε ≤ f i , ε \Delta_{g_E}u_{i,\varepsilon}\leq f_{i,\varepsilon} Δ g E u i , ε ≤ f i , ε
pointwise on V V V and 0 ≤ α ε ≤ 1 0\leq\alpha_\varepsilon\leq1 0 ≤ α ε ≤ 1 , we have
α ε Δ g E u 0 , ε + ( 1 − α ε ) Δ g E u 1 , ε ≤ α ε f 0 , ε + ( 1 − α ε ) f 1 , ε \alpha_\varepsilon\Delta_{g_E}u_{0,\varepsilon}+(1-\alpha_\varepsilon)\Delta_{g_E}u_{1,\varepsilon}\leq\alpha_\varepsilon f_{0,\varepsilon}+(1-\alpha_\varepsilon)f_{1,\varepsilon} α ε Δ g E u 0 , ε + ( 1 − α ε ) Δ g E u 1 , ε ≤ α ε f 0 , ε + ( 1 − α ε ) f 1 , ε
everywhere on V V V . Therefore,
Δ g E u ε ≤ α ε f 0 , ε + ( 1 − α ε ) f 1 , ε (56) \Delta_{g_E}u_\varepsilon\leq\alpha_\varepsilon f_{0,\varepsilon}+(1-\alpha_\varepsilon)f_{1,\varepsilon} \tag*{(56)} Δ g E u ε ≤ α ε f 0 , ε + ( 1 − α ε ) f 1 , ε (56)
in the sense of distributions on V V V . We prove that α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ \alpha_\epsilon f_{0,\epsilon}+(1-\alpha_\epsilon)f_{1,\epsilon} α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ converges to F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) in L 1 ( V ) L^1(V) L 1 ( V ) . A simple calculation yields
α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ = α ϵ ( f 0 , ϵ − f 0 ) + ( 1 − α ϵ ) ( f 1 , ϵ − f 1 ) + α ϵ f 0 + ( 1 − α ϵ ) f 1 . \alpha_\epsilon f_{0,\epsilon}+(1-\alpha_\epsilon)f_{1,\epsilon}=\alpha_\epsilon(f_{0,\epsilon}-f_0)+(1-\alpha_\epsilon)(f_{1,\epsilon}-f_1)+\alpha_\epsilon f_0+(1-\alpha_\epsilon)f_1. α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ = α ϵ ( f 0 , ϵ − f 0 ) + ( 1 − α ϵ ) ( f 1 , ϵ − f 1 ) + α ϵ f 0 + ( 1 − α ϵ ) f 1 .
Since 0 ≤ α ϵ ≤ 1 0\leq\alpha_\epsilon\leq1 0 ≤ α ϵ ≤ 1 and f i , ϵ → f i f_{i,\epsilon}\to f_i f i , ϵ → f i in L 1 ( V ) L^1(V) L 1 ( V ) for i = 0 , 1 i=0,1 i = 0 , 1 , the first two terms converge to 0 0 0 in L 1 ( V ) L^1(V) L 1 ( V ) . Therefore, it is enough to prove that g ϵ : = α ϵ f 0 + ( 1 − α ϵ ) f 1 g_\epsilon:=\alpha_\epsilon f_0+(1-\alpha_\epsilon)f_1 g ϵ := α ϵ f 0 + ( 1 − α ϵ ) f 1 converges to F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) in L 1 ( V ) L^1(V) L 1 ( V ) . Let x ∈ V x\in V x ∈ V be a common Lebesgue point of u 0 u_0 u 0 and u 1 u_1 u 1 . If u 0 ( x ) > u 1 ( x ) u_0(x)>u_1(x) u 0 ( x ) > u 1 ( x ) , then we obtain α ϵ ( x ) = 0 \alpha_\epsilon(x)=0 α ϵ ( x ) = 0 for sufficiently small ϵ \epsilon ϵ and thus g ϵ ( x ) g_\epsilon(x) g ϵ ( x ) converges to f 1 ( x ) = F ( x , u min ( x ) ) f_1(x)=F(x,u_{\min}(x)) f 1 ( x ) = F ( x , u m i n ( x )) . Similarly, if u 0 ( x ) < u 1 ( x ) u_0(x)<u_1(x) u 0 ( x ) < u 1 ( x ) , then g ϵ ( x ) g_\epsilon(x) g ϵ ( x ) converges to f 0 ( x ) = F ( x , u min ( x ) ) f_0(x)=F(x,u_{\min}(x)) f 0 ( x ) = F ( x , u m i n ( x )) . Finally, if u 0 ( x ) = u 1 ( x ) u_0(x)=u_1(x) u 0 ( x ) = u 1 ( x ) , then
g ϵ ( x ) = α ϵ ( x ) f 0 ( x ) + ( 1 − α ϵ ( x ) ) f 1 ( x ) = α ϵ ( x ) f 0 ( x ) + ( 1 − α ϵ ( x ) ) f 0 ( x ) = f 0 ( x ) = F ( x , u min ( x ) ) . \begin{aligned}
g_\epsilon(x)&=\alpha_\epsilon(x)f_0(x)+(1-\alpha_\epsilon(x))f_1(x)\\
&=\alpha_\epsilon(x)f_0(x)+(1-\alpha_\epsilon(x))f_0(x)\\
&=f_0(x)\\
&=F(x,u_{\min}(x)).
\end{aligned} g ϵ ( x ) = α ϵ ( x ) f 0 ( x ) + ( 1 − α ϵ ( x )) f 1 ( x ) = α ϵ ( x ) f 0 ( x ) + ( 1 − α ϵ ( x )) f 0 ( x ) = f 0 ( x ) = F ( x , u m i n ( x )) .
Therefore, we conclude that g ϵ ( x ) g_\epsilon(x) g ϵ ( x ) converges to F ( x , u min ( x ) ) F(x,u_{\min}(x)) F ( x , u m i n ( x )) in every case, and hence, g ϵ g_\epsilon g ϵ converges to F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) almost everywhere. Moreover, we have
∣ g ϵ ∣ ≤ ∣ f 0 ∣ + ∣ f 1 ∣ . |g_\epsilon|\leq|f_0|+|f_1|. ∣ g ϵ ∣ ≤ ∣ f 0 ∣ + ∣ f 1 ∣.
Therefore, by the dominated convergence theorem, g ϵ g_\epsilon g ϵ converges to F ( ⋅ , u min ( ⋅ ) ) F(\cdot,u_{\min}(\cdot)) F ( ⋅ , u m i n ( ⋅ )) in L 1 ( V ) L^1(V) L 1 ( V ) . Let ϕ : U → R ≥ 0 \phi:U\to\mathbb{R}_{\geq0} ϕ : U → R ≥ 0 be a nonnegative smooth function supported in V V V . From (56) , we obtain
∫ V u ϵ Δ g E ϕ ≤ ∫ V ( α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ ) ϕ . \int_V u_\epsilon\Delta_{g_E}\phi\leq\int_V\left(\alpha_\epsilon f_{0,\epsilon}+(1-\alpha_\epsilon)f_{1,\epsilon}\right)\phi. ∫ V u ϵ Δ g E ϕ ≤ ∫ V ( α ϵ f 0 , ϵ + ( 1 − α ϵ ) f 1 , ϵ ) ϕ .
By taking the limit ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 , we have
∫ V u min Δ g E ϕ ≤ ∫ V F ( ⋅ , u min ( ⋅ ) ) ϕ . \int_V u_{\min}\Delta_{g_E}\phi\leq\int_V F(\cdot,u_{\min}(\cdot))\phi. ∫ V u m i n Δ g E ϕ ≤ ∫ V F ( ⋅ , u m i n ( ⋅ )) ϕ .
Since V V V is an arbitrary relatively compact open subset, this proves the assertion.
Remark 46. In Lemma 45, the inequalities are understood in the sense of distributions. In particular, we do not assume that the functions under consideration belong to W l o c 1 , 2 ( U ) W_{\mathrm{loc}}^{1,2}(U) W loc 1 , 2 ( U ) . We emphasize this point because, in some literature on Poisson equations, weak solutions are defined from the outset as functions belonging to W l o c 1 , 2 ( U ) W_{\mathrm{loc}}^{1,2}(U) W loc 1 , 2 ( U ) .
The following proposition is an extension of [43 Lemma 5.4] :
Proposition 47. Let h + = ( h + , 1 , … , h + , r ) h_+=(h_{+,1},\ldots,h_{+,r}) h + = ( h + , 1 , … , h + , r ) and h + ′ = ( h + , 1 ′ , … , h + , r ′ ) h'_+=(h'_{+,1},\ldots,h'_{+,r}) h + ′ = ( h + , 1 ′ , … , h + , r ′ ) be two supersolutions to equation (5), and let h − = ( h − , 1 , … , h − , r ) h_-=(h_{-,1},\ldots,h_{-,r}) h − = ( h − , 1 , … , h − , r ) and h − ′ = ( h − , 1 ′ , … , h − , r ′ ) h'_-=(h'_{-,1},\ldots,h'_{-,r}) h − ′ = ( h − , 1 ′ , … , h − , r ′ ) be two subsolutions to equation (5). For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , define
h ~ + , j : = min { h + , j , h + , j ′ } , \widetilde h_{+,j}:=\min\{h_{+,j},h'_{+,j}\}, h + , j := min { h + , j , h + , j ′ } , h ~ − , j : = max { h − , j , h − , j ′ } . \widetilde h_{-,j}:=\max\{h_{-,j},h'_{-,j}\}. h − , j := max { h − , j , h − , j ′ } . For the remaining indices j = n + 1 , … , r j=n+1,\ldots,r j = n + 1 , … , r , we define h ~ + , j \widetilde h_{+,j} h + , j and h ~ − , j \widetilde h_{-,j} h − , j so that h ~ + : = ( h ~ + , 1 , … , h ~ + , r ) \widetilde h_+:=(\widetilde h_{+,1},\ldots,\widetilde h_{+,r}) h + := ( h + , 1 , … , h + , r ) and h ~ − : = ( h ~ − , 1 , … , h ~ − , r ) \widetilde h_-:=(\widetilde h_{-,1},\ldots,\widetilde h_{-,r}) h − := ( h − , 1 , … , h − , r ) are real diagonal Hermitian metrics. Suppose that for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , h ~ + , j ⊗ h ~ + , j + 1 \widetilde h_{+,j}\otimes\widetilde h_{+,j+1} h + , j ⊗ h + , j + 1 and h ~ − , j ⊗ h ~ − , j + 1 \widetilde h_{-,j}\otimes\widetilde h_{-,j+1} h − , j ⊗ h − , j + 1 belong to L l o c 1 L^1_{\mathrm{loc}} L loc 1 . Then h ~ + \widetilde h_+ h + and h ~ − \widetilde h_- h − are a supersolution and a subsolution to equation (5), respectively.
Proof. The proof is based on the argument in [43 Proof of Lemma 5.4] . For simplicity, we only prove that h ~ + \tilde{h}_{+} h ~ + is a supersolution since the proof for showing that h ~ − \tilde{h}_{-} h ~ − is a subsolution is similar. Since the assertion is local, we can assume from the outset that X X X is the unit disc. Let g E g_{E} g E be the Euclidean metric of the unit disc. Let h ∗ h_{*} h ∗ be the reference metric on K X K_{X} K X induced by the Euclidean metric and denote by h ∗ , j : = h ∗ r − ( 2 j − 1 ) 2 h_{*,j}:=h_{*}^{\frac{r-(2j-1)}{2}} h ∗ , j := h ∗ 2 r − ( 2 j − 1 ) the Hermitian metric on K X r − ( 2 j − 1 ) 2 K_{X}^{\frac{r-(2j-1)}{2}} K X 2 r − ( 2 j − 1 ) induced by h ∗ h_{*} h ∗ . Using h ∗ , j h_{*,j} h ∗ , j , we denote the j j j -th components of h + h_{+} h + and h + ′ h'_{+} h + ′ as follows:
h + , j = e − ϕ + , j h ∗ , j , h_{+,j}=e^{-\phi_{+,j}}h_{*,j}, h + , j = e − ϕ + , j h ∗ , j ,
h + , j ′ = e − ϕ + , j ′ h ∗ , j . h'_{+,j}=e^{-\phi'_{+,j}}h_{*,j}. h + , j ′ = e − ϕ + , j ′ h ∗ , j .
We also set ϕ ^ + , j : = max { ϕ + , j , ϕ + , j ′ } \hat{\phi}_{+,j}:=\max\{\phi_{+,j},\phi'_{+,j}\} ϕ ^ + , j := max { ϕ + , j , ϕ + , j ′ } for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Then by the definition of supersolutions, we obtain
{ 1 2 Δ g E ϕ + , j + e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ≥ 0 for j = 2 , … , n − 1 , 1 2 Δ g E ϕ + , n + e ϕ + , n − 1 − ϕ + , n − e ( 2 n + 2 − r ) ϕ + , n ≥ 0 1 2 Δ g E ϕ + , 1 + e r ϕ − 2 ϕ + , 1 − e ϕ + , 1 − ϕ + , 2 ≥ 0 , (57) \begin{cases}
\frac{1}{2}\Delta_{g_{E}}\phi_{+,j}+e^{\phi_{+,j-1}-\phi_{+,j}}-e^{\phi_{+,j}-\phi_{+,j+1}}\geq0 & \text{for }j=2,\ldots,n-1,\\
\frac{1}{2}\Delta_{g_{E}}\phi_{+,n}+e^{\phi_{+,n-1}-\phi_{+,n}}-e^{(2n+2-r)\phi_{+,n}}\geq0\\
\frac{1}{2}\Delta_{g_{E}}\phi_{+,1}+e^{r\phi-2\phi_{+,1}}-e^{\phi_{+,1}-\phi_{+,2}}\geq0, \tag*{(57)}
\end{cases} ⎩ ⎨ ⎧ 2 1 Δ g E ϕ + , j + e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ≥ 0 2 1 Δ g E ϕ + , n + e ϕ + , n − 1 − ϕ + , n − e ( 2 n + 2 − r ) ϕ + , n ≥ 0 2 1 Δ g E ϕ + , 1 + e r ϕ − 2 ϕ + , 1 − e ϕ + , 1 − ϕ + , 2 ≥ 0 , for j = 2 , … , n − 1 , (57)
{ 1 2 Δ g E ϕ + , j ′ + e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ≥ 0 for j = 2 , … , n − 1 , 1 2 Δ g E ϕ + , n ′ + e ϕ + , n − 1 ′ − ϕ + , n ′ − e ( 2 n + 2 − r ) ϕ + , n ′ ≥ 0 1 2 Δ g E ϕ + , 1 ′ + e r ϕ − 2 ϕ + , 1 ′ − e ϕ + , 1 ′ − ϕ + , 2 ′ ≥ 0. (58) \begin{cases}
\frac{1}{2}\Delta_{g_{E}}\phi'_{+,j}+e^{\phi'_{+,j-1}-\phi'_{+,j}}-e^{\phi'_{+,j}-\phi'_{+,j+1}}\geq0 & \text{for }j=2,\ldots,n-1,\\
\frac{1}{2}\Delta_{g_{E}}\phi'_{+,n}+e^{\phi'_{+,n-1}-\phi'_{+,n}}-e^{(2n+2-r)\phi'_{+,n}}\geq0\\
\frac{1}{2}\Delta_{g_{E}}\phi'_{+,1}+e^{r\phi-2\phi'_{+,1}}-e^{\phi'_{+,1}-\phi'_{+,2}}\geq0. \tag*{(58)}
\end{cases} ⎩ ⎨ ⎧ 2 1 Δ g E ϕ + , j ′ + e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ≥ 0 2 1 Δ g E ϕ + , n ′ + e ϕ + , n − 1 ′ − ϕ + , n ′ − e ( 2 n + 2 − r ) ϕ + , n ′ ≥ 0 2 1 Δ g E ϕ + , 1 ′ + e r ϕ − 2 ϕ + , 1 ′ − e ϕ + , 1 ′ − ϕ + , 2 ′ ≥ 0. for j = 2 , … , n − 1 , (58)
Strictly speaking, the case n = 1 n=1 n = 1 requires separate consideration; however, since the following discussion can easily be modified or extended to cover that case, we assume n ≥ 2 n\geq2 n ≥ 2 for simplicity. For each j = 2 , … , n − 1 j=2,\ldots,n-1 j = 2 , … , n − 1 , we set
G j ( x , t ) : = − 2 ( e ϕ ^ + , j − 1 − t − e t − ϕ ^ + , j + 1 ) , G_{j}(x,t):=-2\left(e^{\hat{\phi}_{+,j-1}-t}-e^{t-\hat{\phi}_{+,j+1}}\right), G j ( x , t ) := − 2 ( e ϕ ^ + , j − 1 − t − e t − ϕ ^ + , j + 1 ) ,
G j # ( x , t ) : = max { G j ( x , t ) , B j ( x ) } , G_{j}^{\#}(x,t):=\max\{G_{j}(x,t),B_{j}(x)\}, G j # ( x , t ) := max { G j ( x , t ) , B j ( x )} ,
where B j B_{j} B j is defined by
B j : = − 2 ∣ e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ∣ − 2 ∣ e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ∣ . (59) B_{j}:=-2\left|e^{\phi_{+,j-1}-\phi_{+,j}}-e^{\phi_{+,j}-\phi_{+,j+1}}\right|-2\left|e^{\phi'_{+,j-1}-\phi'_{+,j}}-e^{\phi'_{+,j}-\phi'_{+,j+1}}\right|. \tag*{(59)} B j := − 2 e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 − 2 e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ . (59)
Clearly,
G j ( x , t ) ≤ G j # ( x , t ) , G_{j}(x,t)\leq G_{j}^{\#}(x,t), G j ( x , t ) ≤ G j # ( x , t ) ,
G j ( x , ϕ + , j ) ≤ − 2 ( e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ) , (60) G_{j}(x,\phi_{+,j})\leq-2\left(e^{\phi_{+,j-1}-\phi_{+,j}}-e^{\phi_{+,j}-\phi_{+,j+1}}\right), \tag*{(60)} G j ( x , ϕ + , j ) ≤ − 2 ( e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ) , (60)
G j ( x , ϕ + , j ′ ) ≤ − 2 ( e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ) . (61) G_{j}(x,\phi'_{+,j})\leq-2\left(e^{\phi'_{+,j-1}-\phi'_{+,j}}-e^{\phi'_{+,j}-\phi'_{+,j+1}}\right). \tag*{(61)} G j ( x , ϕ + , j ′ ) ≤ − 2 ( e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ) . (61)
Moreover, from (59), (60), and (61), we deduce
G j # ( x , ϕ + , j ) ≤ − 2 ( e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ) , (62) G_{j}^{\#}(x,\phi_{+,j})\leq-2\left(e^{\phi_{+,j-1}-\phi_{+,j}}-e^{\phi_{+,j}-\phi_{+,j+1}}\right), \tag*{(62)} G j # ( x , ϕ + , j ) ≤ − 2 ( e ϕ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ + , j + 1 ) , (62)
G j # ( x , ϕ + , j ′ ) ≤ − 2 ( e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ) . (63) G_{j}^{\#}(x,\phi'_{+,j})\leq-2\left(e^{\phi'_{+,j-1}-\phi'_{+,j}}-e^{\phi'_{+,j}-\phi'_{+,j+1}}\right). \tag*{(63)} G j # ( x , ϕ + , j ′ ) ≤ − 2 ( e ϕ + , j − 1 ′ − ϕ + , j ′ − e ϕ + , j ′ − ϕ + , j + 1 ′ ) . (63)
This in particular implies that G j # ( x , ϕ + , j ) G_{j}^{\#}(x,\phi_{+,j}) G j # ( x , ϕ + , j ) and G j # ( x , ϕ + , j ′ ) G_{j}^{\#}(x,\phi'_{+,j}) G j # ( x , ϕ + , j ′ ) are locally integrable, since they are bounded below by a locally integrable function B j B_{j} B j . From (57), (58), (62), and (63), we obtain
Δ g E ϕ + , j ≥ G j # ( x , ϕ + , j ) , \Delta_{g_{E}}\phi_{+,j}\geq G_{j}^{\#}(x,\phi_{+,j}), Δ g E ϕ + , j ≥ G j # ( x , ϕ + , j ) ,
Δ g E ϕ + , j ′ ≥ G j # ( x , ϕ + , j ′ ) . \Delta_{g_{E}}\phi'_{+,j}\geq G_{j}^{\#}(x,\phi'_{+,j}). Δ g E ϕ + , j ′ ≥ G j # ( x , ϕ + , j ′ ) .
By Lemma 45 (ii), we have
Δ g E ϕ ^ + , j ≥ G j # ( x , ϕ ^ + , j ) ≥ G j ( x , ϕ ^ + , j ) . \Delta_{g_E}\hat{\phi}_{+,j} \ge G_j^{\#}(x,\hat{\phi}_{+,j}) \ge G_j(x,\hat{\phi}_{+,j}). Δ g E ϕ ^ + , j ≥ G j # ( x , ϕ ^ + , j ) ≥ G j ( x , ϕ ^ + , j ) .
The above discussion can also be easily extended to the cases j = 1 , n j=1,n j = 1 , n . Since h ~ + , j = min { h + , j , h + , j ′ } = e − ϕ ^ + , j h ∗ , j \widetilde{h}_{+,j}=\min\{h_{+,j},h'_{+,j}\}=e^{-\hat{\phi}_{+,j}}h_{*,j} h + , j = min { h + , j , h + , j ′ } = e − ϕ ^ + , j h ∗ , j for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , this implies the claim.
Remark 48. Following the comparison argument in [43 Proof of Lemma 5.4] , inequalities (57) and (58) also yield the following (64) and (65) :
1 2 Δ g E ϕ + , j + e ϕ ^ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ ^ + , j + 1 ≥ 0 , (64) \frac{1}{2}\Delta_{g_E}\phi_{+,j}+e^{\hat{\phi}_{+,j-1}-\phi_{+,j}}-e^{\phi_{+,j}-\hat{\phi}_{+,j+1}}\ge0, \tag*{(64)} 2 1 Δ g E ϕ + , j + e ϕ ^ + , j − 1 − ϕ + , j − e ϕ + , j − ϕ ^ + , j + 1 ≥ 0 , (64)
1 2 Δ g E ϕ + , j ′ + e ϕ ^ + , j − 1 − ϕ + , j ′ − e ϕ + , j ′ − ϕ ^ + , j + 1 ≥ 0 , (65) \frac{1}{2}\Delta_{g_E}\phi'_{+,j}+e^{\hat{\phi}_{+,j-1}-\phi'_{+,j}}-e^{\phi'_{+,j}-\hat{\phi}_{+,j+1}}\ge0, \tag*{(65)} 2 1 Δ g E ϕ + , j ′ + e ϕ ^ + , j − 1 − ϕ + , j ′ − e ϕ + , j ′ − ϕ ^ + , j + 1 ≥ 0 , (65)
for each j = 2 , … , n − 1 j=2,\ldots,n-1 j = 2 , … , n − 1 . In the present setting, these inequalities are interpreted after testing against nonnegative smooth compactly supported functions, allowing the left-hand sides to take the value + ∞ +\infty + ∞ . This interpretation is well defined because the exponential terms with a minus sign are locally integrable, whereas those with a plus sign need not be. However, the local integrability required by Lemma 45 is not guaranteed, so the lemma cannot be applied directly. The truncation argument above, based on (57) and (58) , provides inequalities with locally integrable right-hand sides to which Lemma 45 applies.
The following lemma is an extension of [43 Proposition 4.6] :
Lemma 49. Suppose that e φ e^{\varphi} e φ is continuous. Let X X X be a Riemann surface with a smooth compact boundary ∂ X \partial X ∂ X , possibly empty. Let h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) be a continuous real diagonal complete subsolution to equation (5), and let h ′ = ( h 1 ′ , … , h r ′ ) h'=(h'_1,\ldots,h'_r) h ′ = ( h 1 ′ , … , h r ′ ) be a continuous real diagonal supersolution to equation (5). For each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , set H j : = h j − 1 ⊗ h j + 1 H_j:=h_j^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 , H j ′ : = h j ′ − 1 ⊗ h j + 1 ′ H'_j:=h_j'^{-1}\otimes h'_{j+1} H j ′ := h j ′ − 1 ⊗ h j + 1 ′ and H 0 = H r : = h r − 1 ⊗ h 1 ⊗ ( e − φ h ∗ ) − r H_0=H_r:=h_r^{-1}\otimes h_1\otimes(e^{-\varphi}h_*)^{-r} H 0 = H r := h r − 1 ⊗ h 1 ⊗ ( e − φ h ∗ ) − r , H 0 ′ = H r ′ : = h r ′ − 1 ⊗ h 1 ′ ⊗ ( e − φ h ∗ ) − r H'_0=H'_r:=h_r'^{-1}\otimes h'_1\otimes(e^{-\varphi}h_*)^{-r} H 0 ′ = H r ′ := h r ′ − 1 ⊗ h 1 ′ ⊗ ( e − φ h ∗ ) − r . Suppose that the curvatures of H 1 , … , H n H_1,\ldots,H_n H 1 , … , H n are bounded below and that H j − 1 ⊗ H j − 1 H_{j-1}\otimes H_j^{-1} H j − 1 ⊗ H j − 1 is bounded for every j = 2 , … , n j=2,\ldots,n j = 2 , … , n . Suppose moreover that log ( h j ⊗ h j ′ − 1 ) \log(h_j\otimes h_j'^{-1}) log ( h j ⊗ h j ′ − 1 ) is bounded above and belongs to W l o c 2 ( X ∖ ∂ X ) W_{\mathrm{loc}}^2(X\setminus\partial X) W loc 2 ( X ∖ ∂ X ) for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Then one of the following holds:
(i)
max 1 ≤ j ≤ n sup X ( h j ⊗ h j ′ − 1 ) ≤ 1 ; \max_{1\le j\le n}\sup_X(h_j\otimes h_j'^{-1})\le1; 1 ≤ j ≤ n max X sup ( h j ⊗ h j ′ − 1 ) ≤ 1 ; (ii) The boundary ∂ X \partial X ∂ X is nonempty and
max 1 ≤ j ≤ n sup X ( h j ⊗ h j ′ − 1 ) ≤ max 1 ≤ j ≤ n max ∂ X ( h j ⊗ h j ′ − 1 ) . \max_{1\le j\le n}\sup_X(h_j\otimes h_j'^{-1})\le\max_{1\le j\le n}\max_{\partial X}(h_j\otimes h_j'^{-1}). 1 ≤ j ≤ n max X sup ( h j ⊗ h j ′ − 1 ) ≤ 1 ≤ j ≤ n max ∂ X max ( h j ⊗ h j ′ − 1 ) . Moreover, if log ( h j ⊗ h j ′ − 1 ) \log(h_j\otimes h_j'^{-1}) log ( h j ⊗ h j ′ − 1 ) is of class C 2 C^2 C 2 for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n , and if (ii) does not occur, then either h j ⊗ h j ′ − 1 < 1 h_j\otimes h_j'^{-1}<1 h j ⊗ h j ′ − 1 < 1 for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n or h j = h j ′ h_j=h_j' h j = h j ′ for all $j=1,\ldots,n.
Proof. For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , we define ξ j : = log ( h j ⊗ h j − 1 ) \xi_j:=\log(h_j\otimes h_j^{-1}) ξ j := log ( h j ⊗ h j − 1 ) , M j : = sup X e ξ j M_j:=\sup_X e^{\xi_j} M j := sup X e ξ j , M : = max 1 ≤ j ≤ n M j M:=\max_{1\leq j\leq n}M_j M := max 1 ≤ j ≤ n M j . Assume that (ii) does not occur. A direct calculation yields
− 1 ∂ ∂ ˉ ξ j ≥ vol ( H j − 1 ) − vol ( H j ) − vol ( H j − 1 ′ ) + vol ( H j ′ ) for j = 1 , … , n . (66) \sqrt{-1}\partial\bar\partial\xi_j\geq\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)-\operatorname{vol}(H'_{j-1})+\operatorname{vol}(H'_j)\quad\text{for }j=1,\ldots,n. \tag*{(66)} − 1 ∂ ∂ ˉ ξ j ≥ vol ( H j − 1 ) − vol ( H j ) − vol ( H j − 1 ′ ) + vol ( H j ′ ) for j = 1 , … , n . (66)
Suppose that n = 1 n=1 n = 1 . From (66) , we have
− 1 Λ H 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ( 4 − r ) ξ 1 ) . (67) \sqrt{-1}\Lambda_{H_1}\partial\bar\partial\xi_1\geq(1-e^{-2\xi_1})H_0\otimes H_1^{-1}-(1-e^{(4-r)\xi_1}). \tag*{(67)} − 1 Λ H 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ( 4 − r ) ξ 1 ) . (67)
By assumption, the curvature of H 1 H_1 H 1 is bounded below. Fix ϵ > 0 \epsilon>0 ϵ > 0 . From Proposition 10, there exists a smooth Hermitian metric H ~ 1 \widetilde{H}_1 H 1 on K X − 1 → X ∖ ∂ X K_X^{-1}\to X\setminus\partial X K X − 1 → X ∖ ∂ X such that
e − ϵ H 1 ≤ H ~ 1 ≤ e ϵ H 1 (68) e^{-\epsilon}H_1\leq\widetilde{H}_1\leq e^\epsilon H_1 \tag*{(68)} e − ϵ H 1 ≤ H 1 ≤ e ϵ H 1 (68)
and the curvature of H ~ 1 \widetilde{H}_1 H 1 is bounded below. If ∂ X ≠ ∅ \partial X\neq\varnothing ∂ X = ∅ , we choose a compact neighbourhood K K K of ∂ X \partial X ∂ X satisfying max K ξ 1 < sup X ξ 1 \max_K\xi_1<\sup_X\xi_1 max K ξ 1 < sup X ξ 1 , together with a diffeomorphism Ψ : K → ∂ X × [ 0 , 1 ] \Psi:K\to\partial X\times[0,1] Ψ : K → ∂ X × [ 0 , 1 ] such that Ψ ( ∂ X ) = ∂ X × { 0 } \Psi(\partial X)=\partial X\times\{0\} Ψ ( ∂ X ) = ∂ X × { 0 } . Set Y : = X ∖ Ψ − 1 ( ∂ X × [ 0 , 1 ) ) Y:=X\setminus\Psi^{-1}(\partial X\times[0,1)) Y := X ∖ Ψ − 1 ( ∂ X × [ 0 , 1 )) . Then ∂ Y = Ψ − 1 ( ∂ X × { 1 } ) ⊆ X ∖ ∂ X \partial Y=\Psi^{-1}(\partial X\times\{1\})\subseteq X\setminus\partial X ∂ Y = Ψ − 1 ( ∂ X × { 1 }) ⊆ X ∖ ∂ X , and sup Y ξ 1 = sup X ξ 1 \sup_Y\xi_1=\sup_X\xi_1 sup Y ξ 1 = sup X ξ 1 . We apply Proposition 23 on Y Y Y with the metric 2 g H ~ 1 2g_{\widetilde{H}_1} 2 g H 1 , where g H ~ 1 g_{\widetilde{H}_1} g H 1 denotes the Kähler metric induced by H ~ 1 \widetilde{H}_1 H 1 . This metric is smooth up to ∂ Y \partial Y ∂ Y and complete by (68), with Ricci curvature bounded below. From (67), we have
− 1 Λ H ~ 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H ~ 1 − 1 − ( 1 − e ( 4 − r ) ξ 1 ) H 1 ⊗ H ~ 1 − 1 . (69) \sqrt{-1}\Lambda_{\widetilde{H}_1}\partial\bar\partial\xi_1\geq(1-e^{-2\xi_1})H_0\otimes\widetilde{H}_1^{-1}-(1-e^{(4-r)\xi_1})H_1\otimes\widetilde{H}_1^{-1}. \tag*{(69)} − 1 Λ H 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ( 4 − r ) ξ 1 ) H 1 ⊗ H 1 − 1 . (69)
By assumption, ξ 1 \xi_1 ξ 1 is bounded above, continuous, and belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . Moreover, the right-hand side of inequality (69) is continuous, and, by assumption, sup X ξ 1 > max ∂ X ξ 1 \sup_X\xi_1>\max_{\partial X}\xi_1 sup X ξ 1 > max ∂ X ξ 1 if ∂ X \partial X ∂ X is nonempty. Therefore, by applying Propositions 22 and 23 to (69), we obtain, for each m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 , a point x m ∈ X x_m\in X x m ∈ X such that
ξ 1 ( x m ) ≥ sup X ξ 1 − 1 / m , (70) \xi_1(x_m)\geq\sup_X\xi_1-1/m, \tag*{(70)} ξ 1 ( x m ) ≥ X sup ξ 1 − 1/ m , (70)
1 / m ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H ~ 1 − 1 ) ( x m ) − ( 1 − e ( 4 − r ) ξ 1 ( x m ) ) ( H 1 ⊗ H ~ 1 − 1 ) ( x m ) . (71) 1/m\geq(1-e^{-2\xi_1(x_m)})(H_0\otimes\widetilde{H}_1^{-1})(x_m)-(1-e^{(4-r)\xi_1(x_m)})(H_1\otimes\widetilde{H}_1^{-1})(x_m). \tag*{(71)} 1/ m ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H 1 − 1 ) ( x m ) − ( 1 − e ( 4 − r ) ξ 1 ( x m ) ) ( H 1 ⊗ H 1 − 1 ) ( x m ) . (71)
Suppose that sup X ξ 1 > 0 \sup_X\xi_1>0 sup X ξ 1 > 0 . From (70), for all sufficiently large m m m , we have ξ 1 ( x m ) ≥ sup X ξ 1 − 1 / m > 0 \xi_1(x_m)\geq\sup_X\xi_1-1/m>0 ξ 1 ( x m ) ≥ sup X ξ 1 − 1/ m > 0 . Therefore, for all sufficiently large m m m , inequality (71) yields
1 / m ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H ~ 1 − 1 ) ( x m ) − ( 1 − e ( 4 − r ) ξ 1 ( x m ) ) ( H 1 ⊗ H ~ 1 − 1 ) ( x m ) ≥ − ( 1 − e ( 4 − r ) ( sup X ξ 1 − 1 / m ) ) e − ϵ . \begin{aligned}
1/m&\geq(1-e^{-2\xi_1(x_m)})(H_0\otimes\widetilde{H}_1^{-1})(x_m)-(1-e^{(4-r)\xi_1(x_m)})(H_1\otimes\widetilde{H}_1^{-1})(x_m)\\
&\geq-(1-e^{(4-r)(\sup_X\xi_1-1/m)})e^{-\epsilon}.
\end{aligned} 1/ m ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H 1 − 1 ) ( x m ) − ( 1 − e ( 4 − r ) ξ 1 ( x m ) ) ( H 1 ⊗ H 1 − 1 ) ( x m ) ≥ − ( 1 − e ( 4 − r ) ( s u p X ξ 1 − 1/ m ) ) e − ϵ .
Taking the limit inferior as m → ∞ m\to\infty m → ∞ , we obtain sup X ξ 1 ≤ 0 \sup_X\xi_1\leq0 sup X ξ 1 ≤ 0 , contradicting the assumption sup X ξ 1 > 0 \sup_X\xi_1>0 sup X ξ 1 > 0 . Therefore we obtain sup X ξ 1 ≤ 0 \sup_X\xi_1\leq0 sup X ξ 1 ≤ 0 , and thus we have assertion (i) for the case where n = 1 n=1 n = 1 . Suppose that ξ 1 \xi_1 ξ 1 is of class C 2 C^2 C 2 and (ii) does not occur. Then inequality (67), the strong maximum principle, and the assumption that (ii) does not occur immediately yield ξ 1 < 0 \xi_1<0 ξ 1 < 0 or ξ 1 = 0 \xi_1=0 ξ 1 = 0 . We next consider the case where n ≥ 2 n\geq2 n ≥ 2 . From (66), we have
− 1 Λ H 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ξ 1 − ξ 2 ) , (72) \sqrt{-1}\Lambda_{H_1}\partial\bar\partial\xi_1\geq(1-e^{-2\xi_1})H_0\otimes H_1^{-1}-(1-e^{\xi_1-\xi_2}), \tag*{(72)} − 1 Λ H 1 ∂ ∂ ˉ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ξ 1 − ξ 2 ) , (72)
− 1 Λ H j ∂ ∂ ˉ ξ j ≥ ( 1 − e ξ j − 1 − ξ j ) H j − 1 ⊗ H j − 1 − ( 1 − e ξ j − ξ j + 1 ) for j = 2 , … , n − 1 , (73) \sqrt{-1}\Lambda_{H_j}\partial\bar\partial\xi_j\geq(1-e^{\xi_{j-1}-\xi_j})H_{j-1}\otimes H_j^{-1}-(1-e^{\xi_j-\xi_{j+1}})\quad\text{for }j=2,\ldots,n-1, \tag*{(73)} − 1 Λ H j ∂ ∂ ˉ ξ j ≥ ( 1 − e ξ j − 1 − ξ j ) H j − 1 ⊗ H j − 1 − ( 1 − e ξ j − ξ j + 1 ) for j = 2 , … , n − 1 , (73)
− 1 Λ H n ∂ ∂ ˉ ξ n ≥ ( 1 − e ξ n − 1 − ξ n ) H n − 1 ⊗ H n − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) . (74) \sqrt{-1}\Lambda_{H_n}\partial\bar\partial\xi_n\geq(1-e^{\xi_{n-1}-\xi_n})H_{n-1}\otimes H_n^{-1}-(1-e^{(2n+2-r)\xi_n}). \tag*{(74)} − 1 Λ H n ∂ ∂ ˉ ξ n ≥ ( 1 − e ξ n − 1 − ξ n ) H n − 1 ⊗ H n − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) . (74)
By assumption, the curvatures of H 1 , … , H n H_1,\ldots,H_n H 1 , … , H n are bounded below. Fix ϵ > 0 \epsilon>0 ϵ > 0 . From Proposition 10, there exist smooth Hermitian metrics H ~ 1 , … , H ~ n \widetilde{H}_1,\ldots,\widetilde{H}_n H 1 , … , H n on K X − 1 → X ∖ ∂ X K_X^{-1}\to X\setminus\partial X K X − 1 → X ∖ ∂ X such that
e − ϵ H j ≤ H ~ j ≤ e ϵ H j for j = 1 , … , n (75) e^{-\epsilon}H_j\leq\widetilde{H}_j\leq e^\epsilon H_j\quad\text{for }j=1,\ldots,n \tag*{(75)} e − ϵ H j ≤ H j ≤ e ϵ H j for j = 1 , … , n (75)
and the curvatures of H ~ 1 , … , H ~ n \widetilde{H}_1,\ldots,\widetilde{H}_n H 1 , … , H n are bounded below. If ∂ X ≠ ∅ \partial X\neq\emptyset ∂ X = ∅ , then M > max 1 ≤ j ≤ n max ∂ X e ξ j M>\max_{1\leq j\leq n}\max_{\partial X}e^{\xi_j} M > max 1 ≤ j ≤ n max ∂ X e ξ j . For each component with M j = M M_j=M M j = M , we therefore use the same construction as above, replacing ξ 1 \xi_1 ξ 1 and H ~ 1 \widetilde{H}_1 H 1 by ξ j \xi_j ξ j and H ~ j \widetilde{H}_j H j , respectively, and using (75) in place of (68). From inequalities (72), (73), and (74), we have
− 1 Λ H ~ 1 ∂ ˉ ∂ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H ~ 1 − 1 − ( 1 − e ξ 1 − ξ 2 ) H 1 ⊗ H ~ 1 − 1 , (76) \sqrt{-1}\Lambda_{\widetilde{H}_1}\bar{\partial}\partial\xi_1\geq(1-e^{-2\xi_1})H_0\otimes\widetilde{H}_1^{-1}-(1-e^{\xi_1-\xi_2})H_1\otimes\widetilde{H}_1^{-1}, \tag*{(76)} − 1 Λ H 1 ∂ ˉ ∂ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ξ 1 − ξ 2 ) H 1 ⊗ H 1 − 1 , (76)
− 1 Λ H ~ j ∂ ˉ ∂ ξ j ≥ ( 1 − e ξ j − 1 − ξ j ) H j − 1 ⊗ H ~ j − 1 − ( 1 − e ξ j − ξ j + 1 ) H j ⊗ H ~ j − 1 for j = 2 , … , n − 1 , (77) \sqrt{-1}\Lambda_{\widetilde{H}_j}\bar{\partial}\partial\xi_j\geq(1-e^{\xi_{j-1}-\xi_j})H_{j-1}\otimes\widetilde{H}_j^{-1}-(1-e^{\xi_j-\xi_{j+1}})H_j\otimes\widetilde{H}_j^{-1}\quad\text{for }j=2,\ldots,n-1, \tag*{(77)} − 1 Λ H j ∂ ˉ ∂ ξ j ≥ ( 1 − e ξ j − 1 − ξ j ) H j − 1 ⊗ H j − 1 − ( 1 − e ξ j − ξ j + 1 ) H j ⊗ H j − 1 for j = 2 , … , n − 1 , (77)
− 1 Λ H ~ n ∂ ˉ ∂ ξ n ≥ ( 1 − e ξ n − 1 − ξ n ) H n − 1 ⊗ H ~ n − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) H n ⊗ H ~ n − 1 . (78) \sqrt{-1}\Lambda_{\widetilde{H}_n}\bar{\partial}\partial\xi_n\geq(1-e^{\xi_{n-1}-\xi_n})H_{n-1}\otimes\widetilde{H}_n^{-1}-(1-e^{(2n+2-r)\xi_n})H_n\otimes\widetilde{H}_n^{-1}. \tag*{(78)} − 1 Λ H n ∂ ˉ ∂ ξ n ≥ ( 1 − e ξ n − 1 − ξ n ) H n − 1 ⊗ H n − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) H n ⊗ H n − 1 . (78)
By assumption, ξ 1 , … , ξ n \xi_1,\ldots,\xi_n ξ 1 , … , ξ n are bounded above, continuous, and belong to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . Moreover, the right-hand sides of inequalities (76), (77), and (78) are all continuous. We set k : = max { j ∣ M j = M } k:=\max\{j\mid M_j=M\} k := max { j ∣ M j = M } . Assume M = M k > 1 M=M_k>1 M = M k > 1 and derive a contradiction. Suppose that k = 1 k=1 k = 1 . By applying Proposition 22 when the boundary is empty, and Proposition 23 together with the assumption that (ii) does not occur when the boundary is nonempty, to (76), we obtain, for each m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 , a point x m ∈ X x_m\in X x m ∈ X such that
ξ 1 ( x m ) ≥ sup X ξ 1 − 1 m , (79) \xi_1(x_m)\geq\sup_X\xi_1-\frac{1}{m}, \tag*{(79)} ξ 1 ( x m ) ≥ X sup ξ 1 − m 1 , (79)
1 m ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H ~ 1 − 1 ) ( x m ) − ( 1 − e ξ 1 ( x m ) − ξ 2 ( x m ) ) ( H 1 ⊗ H ~ 1 − 1 ) ( x m ) ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H ~ 1 − 1 ) ( x m ) − ( 1 − M 2 − 1 e ξ 1 ( x m ) ) ( H 1 ⊗ H ~ 1 − 1 ) ( x m ) . (80) \begin{aligned}
\frac{1}{m}\geq(1-e^{-2\xi_1(x_m)})(H_0\otimes\widetilde{H}_1^{-1})(x_m)-(1-e^{\xi_1(x_m)-\xi_2(x_m)})(H_1\otimes\widetilde{H}_1^{-1})(x_m) \\
&\geq(1-e^{-2\xi_1(x_m)})(H_0\otimes\widetilde{H}_1^{-1})(x_m)-(1-M_2^{-1}e^{\xi_1(x_m)})(H_1\otimes\widetilde{H}_1^{-1})(x_m). \tag*{(80)}
\end{aligned} m 1 ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H 1 − 1 ) ( x m ) − ( 1 − e ξ 1 ( x m ) − ξ 2 ( x m ) ) ( H 1 ⊗ H 1 − 1 ) ( x m ) ≥ ( 1 − e − 2 ξ 1 ( x m ) ) ( H 0 ⊗ H 1 − 1 ) ( x m ) − ( 1 − M 2 − 1 e ξ 1 ( x m ) ) ( H 1 ⊗ H 1 − 1 ) ( x m ) . (80)
By the assumptions k = 1 k=1 k = 1 and M k = M > 1 M_k=M>1 M k = M > 1 , the first term is nonnegative for all sufficiently large m m m . Hence, from (75), (79), and (80), taking the limit inferior as m → ∞ m\to\infty m → ∞ , we obtain M 1 ≤ M 2 M_1\leq M_2 M 1 ≤ M 2 , which contradicts the assumption k = 1 k=1 k = 1 . Suppose that 2 ≤ k ≤ n − 1 2\leq k\leq n-1 2 ≤ k ≤ n − 1 . By applying Proposition 22 when the boundary is empty, and Proposition 23 together with the assumption that (ii) does not occur when the boundary is nonempty, to (77), we obtain, for each m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 , a point x m ∈ X x_m\in X x m ∈ X such that
ξ k ( x m ) ≥ sup X ξ k − 1 m , \xi_k(x_m)\geq\sup_X\xi_k-\frac{1}{m}, ξ k ( x m ) ≥ X sup ξ k − m 1 ,
1 m ≥ ( 1 − e ξ k − 1 ( x m ) − ξ k ( x m ) ) ( H k − 1 ⊗ H ~ k − 1 ) ( x m ) − ( 1 − e ξ k ( x m ) − ξ k + 1 ( x m ) ) ( H k ⊗ H ~ k − 1 ) ( x m ) . \frac{1}{m}\geq(1-e^{\xi_{k-1}(x_m)-\xi_k(x_m)})(H_{k-1}\otimes\widetilde{H}_k^{-1})(x_m)-(1-e^{\xi_k(x_m)-\xi_{k+1}(x_m)})(H_k\otimes\widetilde{H}_k^{-1})(x_m). m 1 ≥ ( 1 − e ξ k − 1 ( x m ) − ξ k ( x m ) ) ( H k − 1 ⊗ H k − 1 ) ( x m ) − ( 1 − e ξ k ( x m ) − ξ k + 1 ( x m ) ) ( H k ⊗ H k − 1 ) ( x m ) .
By taking the limit inferior as m → ∞ m\to\infty m → ∞ , together with the assumption that H k − 1 ⊗ H k − 1 H_{k-1}\otimes H_k^{-1} H k − 1 ⊗ H k − 1 is bounded and with (75), we obtain M k ≤ M k + 1 M_k\leq M_{k+1} M k ≤ M k + 1 , which contradicts the definition of k k k . Therefore, k = n k=n k = n . By applying Proposition 22 when the boundary is empty, and Proposition 23 together with the assumption that (ii) does not occur when the boundary is nonempty, to (78), we obtain, for each m ∈ Z > 0 m\in\mathbb{Z}_{>0} m ∈ Z > 0 , a point x m ∈ X x_m\in X x m ∈ X such that
ξ n ( x m ) ≥ sup X ξ n − 1 m \xi_n(x_m)\geq\sup_X\xi_n-\frac{1}{m} ξ n ( x m ) ≥ X sup ξ n − m 1
1 m ≥ ( 1 − e ξ n − 1 ( x m ) − ξ n ( x m ) ) ( H n − 1 ⊗ H ~ n − 1 ) ( x m ) − ( 1 − e ( 2 n + 2 − r ) ξ n ( x m ) ) ( H n ⊗ H ~ n − 1 ) ( x m ) . \frac{1}{m}\geq(1-e^{\xi_{n-1}(x_m)-\xi_n(x_m)})(H_{n-1}\otimes\widetilde{H}_n^{-1})(x_m)-(1-e^{(2n+2-r)\xi_n(x_m)})(H_n\otimes\widetilde{H}_n^{-1})(x_m). m 1 ≥ ( 1 − e ξ n − 1 ( x m ) − ξ n ( x m ) ) ( H n − 1 ⊗ H n − 1 ) ( x m ) − ( 1 − e ( 2 n + 2 − r ) ξ n ( x m ) ) ( H n ⊗ H n − 1 ) ( x m ) .
By taking the limit inferior as m → ∞ m \to\infty m → ∞ , together with the assumption H k − 1 ⊗ H ~ k − 1 H_{k-1} \otimes\widetilde{H}_{k}^{-1} H k − 1 ⊗ H k − 1 is bounded and with (75), we obtain M = M n ≤ 1 M = M_n \le1 M = M n ≤ 1 , which contradicts the assumption M > 1 M > 1 M > 1 . Therefore, we obtain M ≤ 1 M \le1 M ≤ 1 , and thus assertion (i) follows. We prove the latter half of the statement. From (72), (73), and (74), we have
− 1 Λ H ~ 1 ∂ ˉ ∂ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H ~ 1 − 1 − ( 1 − e ξ 1 ) H 1 ⊗ H ~ 1 − 1 . (81) \sqrt{-1}\Lambda_{\widetilde{H}_1}\bar{\partial}\partial\xi_1 \ge(1-e^{-2\xi_1})H_0\otimes\widetilde{H}_1^{-1}-(1-e^{\xi_1})H_1\otimes\widetilde{H}_1^{-1}. \tag*{(81)} − 1 Λ H 1 ∂ ˉ ∂ ξ 1 ≥ ( 1 − e − 2 ξ 1 ) H 0 ⊗ H 1 − 1 − ( 1 − e ξ 1 ) H 1 ⊗ H 1 − 1 . (81)
− 1 Λ H ~ j ∂ ˉ ∂ ξ j ≥ ( 1 − e − ξ j ) H j − 1 ⊗ H ~ j − 1 − 1 − ( 1 − e ξ j ) H j ⊗ H ~ j − 1 for j = 2 , … , n − 1. (82) \sqrt{-1}\Lambda_{\widetilde{H}_j}\bar{\partial}\partial\xi_j \ge(1-e^{-\xi_j})H_{j-1}\otimes\widetilde{H}_{j-1}^{-1}-(1-e^{\xi_j})H_j\otimes\widetilde{H}_j^{-1}\quad\text{for }j=2,\ldots,n-1. \tag*{(82)} − 1 Λ H j ∂ ˉ ∂ ξ j ≥ ( 1 − e − ξ j ) H j − 1 ⊗ H j − 1 − 1 − ( 1 − e ξ j ) H j ⊗ H j − 1 for j = 2 , … , n − 1. (82)
− 1 Λ H ~ n ∂ ˉ ∂ ξ n ≥ ( 1 − e − ξ n ) H n − 1 ⊗ H ~ n − 1 − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) H n ⊗ H ~ n − 1 . (83) \sqrt{-1}\Lambda_{\widetilde{H}_n}\bar{\partial}\partial\xi_n \ge(1-e^{-\xi_n})H_{n-1}\otimes\widetilde{H}_{n-1}^{-1}-(1-e^{(2n+2-r)\xi_n})H_n\otimes\widetilde{H}_n^{-1}. \tag*{(83)} − 1 Λ H n ∂ ˉ ∂ ξ n ≥ ( 1 − e − ξ n ) H n − 1 ⊗ H n − 1 − 1 − ( 1 − e ( 2 n + 2 − r ) ξ n ) H n ⊗ H n − 1 . (83)
Suppose that ξ 1 , … , ξ n \xi_1,\ldots,\xi_n ξ 1 , … , ξ n are all of class C 2 C^2 C 2 . Assume that (ii) does not occur. Then by this assumption and by the strong maximum principle applied to (81), (82), and (83), we obtain either ξ j < 0 \xi_j<0 ξ j < 0 or ξ j = 0 \xi_j=0 ξ j = 0 for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . If ξ 1 = 0 \xi_1=0 ξ 1 = 0 , then from (72) and M 2 ≤ 1 M_2\le1 M 2 ≤ 1 , we have ξ 2 = 0 \xi_2=0 ξ 2 = 0 . If ξ j = 0 \xi_j=0 ξ j = 0 for some j = 2 , … , n − 1 j=2,\ldots,n-1 j = 2 , … , n − 1 , then from (73), M j − 1 ≤ 1 M_{j-1}\le1 M j − 1 ≤ 1 , and M j + 1 ≤ 1 M_{j+1}\le1 M j + 1 ≤ 1 , we have ξ j − 1 = ξ j + 1 = 0 \xi_{j-1}=\xi_{j+1}=0 ξ j − 1 = ξ j + 1 = 0 . If ξ n = 0 \xi_n=0 ξ n = 0 , then from (74) and M n − 1 ≤ 1 M_{n-1}\le1 M n − 1 ≤ 1 , we have ξ n − 1 = 0 \xi_{n-1}=0 ξ n − 1 = 0 . We conclude that either e ξ j < 1 e^{\xi_j}<1 e ξ j < 1 for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n , or ξ j = 0 \xi_j=0 ξ j = 0 for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n . □
At the end of this subsection, we prove the following proposition:
Proposition 50. Suppose that e φ e^{\varphi} e φ is continuous. Let h + = ( h + , 1 , … , h + , r ) h_+=(h_{+,1},\ldots,h_{+,r}) h + = ( h + , 1 , … , h + , r ) be a continuous supersolution to equation (5), and let h − = ( h − , 1 , … , h − , r ) h_-=(h_{-,1},\ldots,h_{-,r}) h − = ( h − , 1 , … , h − , r ) be a continuous subsolution to equation (5) such that h − , j ≤ h + , j h_{-,j}\le h_{+,j} h − , j ≤ h + , j for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Then there exists a real solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) to equation (5) such that h − , j ≤ h j ≤ h + , j h_{-,j}\le h_j\le h_{+,j} h − , j ≤ h j ≤ h + , j for all $j=1,\ldots,n.
Proof. We first suppose that X X X is noncompact. Let Y 1 ⋐ Y 2 ⋐ ⋯ ⋐ Y i ⋐ Y i + 1 ⋐ ⋯ ⋐ X Y_1\Subset Y_2\Subset\cdots\Subset Y_i\Subset Y_{i+1}\Subset\cdots\Subset X Y 1 ⋐ Y 2 ⋐ ⋯ ⋐ Y i ⋐ Y i + 1 ⋐ ⋯ ⋐ X be a family of relatively compact open subsets with smooth boundaries ∂ Y i \partial Y_i ∂ Y i such that ⋃ i = 1 ∞ Y i = X \bigcup_{i=1}^{\infty}Y_i=X ⋃ i = 1 ∞ Y i = X . For each i i i , let ∂ h ( i ) = ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) \partial h^{(i)}=(\partial h_1^{(i)},\ldots,\partial h_r^{(i)}) ∂ h ( i ) = ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) be a real diagonal Hermitian metric on the boundary ∂ Y i \partial Y_i ∂ Y i of Y i Y_i Y i such that for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , h − , j ∣ ∂ Y i ≤ ∂ h j ( i ) ≤ h + , j ∣ ∂ Y i h_{-,j}|_{\partial Y_i}\le\partial h_j^{(i)}\le h_{+,j}|_{\partial Y_i} h − , j ∣ ∂ Y i ≤ ∂ h j ( i ) ≤ h + , j ∣ ∂ Y i on ∂ Y i \partial Y_i ∂ Y i . Let h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) h^{(i)}=(h_1^{(i)},\ldots,h_r^{(i)}) h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) be the solution to the Dirichlet problem (cf. [49 ] ) for equation (5) with boundary condition h ( i ) ∣ ∂ Y i = ∂ h ( i ) h^{(i)}|_{\partial Y_i}=\partial h^{(i)} h ( i ) ∣ ∂ Y i = ∂ h ( i ) satisfying det ( h ( i ) ) = 1 \det(h^{(i)})=1 det ( h ( i ) ) = 1 . Although the curvature lower bounds assumed in Lemma 49 need not hold here, the compactness of Y i ‾ \overline{Y_i} Y i allows us to apply the same comparison argument at an attained maximum, using the local maximum principle for subharmonic functions (see Remark 51 for more details). Hence h − , j ≤ h j ( i ) ≤ h + , j h_{-,j}\le h_j^{(i)}\le h_{+,j} h − , j ≤ h j ( i ) ≤ h + , j for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n . We obtain h − , j ≤ h j ( i ) ≤ h + , j h_{-,j}\le h_j^{(i)}\le h_{+,j} h − , j ≤ h j ( i ) ≤ h + , j for each i ≥ 1 i\ge1 i ≥ 1 and each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . By the interior L p L^p L p -estimate (cf. [27 Theorem 9.11] ), for each i ≥ 1 i\ge1 i ≥ 1 , 1 < p < ∞ 1<p<\infty 1 < p < ∞ , and l ≥ 1 l\ge1 l ≥ 1 , there exist positive constants C 1 C_1 C 1 and C 2 C_2 C 2 independent of l l l such that
∣ log h ( i + l ) ∣ L 2 p ( Y i ) ≤ C 1 ( ∣ log h ( i + l ) ∣ L p ( Y i + 1 ) + ∣ F h ( i + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 . \left|\log h^{(i+l)}\right|_{L^p_2(Y_i)}\le C_1\left(\left|\log h^{(i+l)}\right|_{L^p(Y_{i+1})}+\left|F_{h^{(i+l)}}\right|_{L^p(Y_{i+1})}\right)\le C_2. log h ( i + l ) L 2 p ( Y i ) ≤ C 1 ( log h ( i + l ) L p ( Y i + 1 ) + ∣ F h ( i + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 .
Since bounded subsets of reflexive Banach spaces are weakly relatively compact, for each i ≥ 2 i\ge2 i ≥ 2 , we can find a subsequence of ( h ( i + l ) ) l ≥ 1 (h^{(i+l)})_{l\ge1} ( h ( i + l ) ) l ≥ 1 that converges weakly in L 2 p ( Y i ) L^p_2(Y_i) L 2 p ( Y i ) to a Hermitian metric h h h on Y i Y_i Y i . Using the diagonal trick, we can find a subsequence of ( h ( k ) ) k ≥ 1 (h^{(k)})_{k\ge1} ( h ( k ) ) k ≥ 1 that converges weakly in L 2 , l o c p L^p_{2,\mathrm{loc}} L 2 , loc p to a Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on X X X belonging to L 2 , l o c p L^p_{2,\mathrm{loc}} L 2 , loc p . After passing to a further subsequence if necessary, we may assume that h j ( k ) h_j^{(k)} h j ( k ) converges to h j h_j h j almost everywhere for every j = 1 , … , r j=1,\ldots,r j = 1 , … , r . Since the h j ( k ) h_j^{(k)} h j ( k ) are uniformly bounded above and away from zero on compact subsets, Lebesgue’s dominated convergence theorem allows us to pass to the limit in the nonlinear terms of equation (5). Therefore, the Hermitian metric h h h solves equation (5) in the weak sense. Suppose now that X X X is compact. Fix p ∈ X p \in X p ∈ X and apply the noncompact case to X ∖ { p } X \setminus\{p\} X ∖ { p } to obtain a real solution h h h satisfying the required bounds. These bounds and the continuity of e φ e^{\varphi} e φ imply that the local logarithmic weights of h h h and the right-hand sides of their Poisson equations are bounded near p p p . Subtracting local Newtonian potentials of these right-hand sides yields bounded harmonic functions on a punctured coordinate disc. Their singularities are removable, so h h h extends to a real weak solution on X X X . The required bounds hold at p p p by continuity.
Remark 51. We illustrate the comparison used in Proposition 50 when n = 1 n=1 n = 1 . Let h h h be a continuous subsolution and h ′ h' h ′ a continuous supersolution on Y i ‾ \overline{Y_i} Y i , satisfying h 1 ≤ h ′ h_1 \le h' h 1 ≤ h ′ on ∂ Y i \partial Y_i ∂ Y i . Using the notation of Lemma 49 , set ξ : = log ( h 1 ⊗ h ′ − 1 ) \xi:= \log(h_1 \otimes h'^{-1}) ξ := log ( h 1 ⊗ h ′ − 1 ) . Then, in the sense of distributions,
− 1 ∂ ∂ ‾ ξ ≥ ( 1 − e − 2 ξ ) vol ( H 0 ) + ( e ( 4 − r ) ξ − 1 ) vol ( H 1 ) . \sqrt{-1}\partial\overline{\partial}\xi\ge(1-e^{-2\xi})\operatorname{vol}(H_0)+(e^{(4-r)\xi}-1)\operatorname{vol}(H_1). − 1 ∂ ∂ ξ ≥ ( 1 − e − 2 ξ ) vol ( H 0 ) + ( e ( 4 − r ) ξ − 1 ) vol ( H 1 ) .
If max Y i ‾ ξ > 0 \max_{\overline{Y_i}}\xi>0 max Y i ξ > 0 , this maximum is attained at some p ∈ Y i p \in Y_i p ∈ Y i . The right-hand side is strictly positive near p p p , so ξ \xi ξ is subharmonic there. The maximum principle for subharmonic functions forces ξ \xi ξ to be constant near p p p , contradicting the displayed inequality. Thus ξ ≤ 0 \xi\le0 ξ ≤ 0 on Y i ‾ \overline{Y_i} Y i . Applying this comparison to ( h − , h ( i ) ) (h_-,h^{(i)}) ( h − , h ( i ) ) and ( h ( i ) , h + ) (h^{(i)},h_+) ( h ( i ) , h + ) yields the inequalities used in Proposition 50 .
Proof of Theorems 2 and 3 for the unit disc case
Suppose that X X X is the unit disc D : = { z ∈ C ∣ ∣ z ∣ < 1 } \mathbb{D}:=\{z\in\mathbb{C}\mid|z|<1\} D := { z ∈ C ∣ ∣ z ∣ < 1 } . Let g X g_X g X be the Poincaré metric, i.e., the complete Kähler metric whose Gaussian curvature equals − 1 -1 − 1 . We denote by h X h_X h X the corresponding Hermitian metric on K X → X K_X\to X K X → X and by ω X \omega_X ω X the Kähler form. Note that h X h_X h X and ω X \omega_X ω X satisfy the following:
− 1 F h X = ω X . \sqrt{-1}F_{h_X}=\omega_X. − 1 F h X = ω X .
We choose the reference metric h ∗ h_* h ∗ as h X h_X h X . First, following [43 Section 5] , we prove the existence of a complete solution to equation (5) under the assumption that e φ e^{\varphi} e φ is continuous. We set a constant M φ M_\varphi M φ , which may be infinite, as follows:
M φ : = max { sup X e φ , 1 } . M_\varphi:=\max\{\sup_X e^\varphi,1\}. M φ := max { X sup e φ , 1 } .
For each 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , let D ϵ : = { z ∈ C ∣ ∣ z ∣ < 1 − ϵ } \mathbb{D}_\epsilon:=\{z\in\mathbb{C}\mid|z|<1-\epsilon\} D ϵ := { z ∈ C ∣ ∣ z ∣ < 1 − ϵ } be the open disc of radius 1 − ϵ 1-\epsilon 1 − ϵ centered at 0 0 0 . Then we define a family of open subsets ( Y i ) i ≥ 2 (Y_i)_{i\ge2} ( Y i ) i ≥ 2 of X = D X=\mathbb{D} X = D by Y i : = D 1 / i Y_i:=\mathbb{D}_{1/i} Y i := D 1/ i for i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … . For each i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … , we denote by ∂ Y i \partial Y_i ∂ Y i the boundary of Y i Y_i Y i . Given each i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … and each diagonal continuous Hermitian metric ∂ h ( i ) = ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) \partial h^{(i)}=(\partial h^{(i)}_1,\ldots,\partial h^{(i)}_r) ∂ h ( i ) = ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) on ∂ Y i \partial Y_i ∂ Y i with det ( ∂ h ( i ) ) = 1 \det(\partial h^{(i)})=1 det ( ∂ h ( i ) ) = 1 , there exists a solution h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) h^{(i)}=(h^{(i)}_1,\ldots,h^{(i)}_r) h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) to equation (5) such that lim z → ζ h ( i ) ( z ) = ∂ h ( i ) ( ζ ) \lim_{z\to\zeta}h^{(i)}(z)=\partial h^{(i)}(\zeta) lim z → ζ h ( i ) ( z ) = ∂ h ( i ) ( ζ ) for any ζ ∈ ∂ Y i \zeta\in\partial Y_i ζ ∈ ∂ Y i and that det ( h ( i ) ) = 1 \det(h^{(i)})=1 det ( h ( i ) ) = 1 (cf. [49 ] ). Moreover, if ∂ h ( i ) \partial h^{(i)} ∂ h ( i ) is real, then h ( i ) h^{(i)} h ( i ) is also real (see [49 Remark 5] ). From h ( i ) h^{(i)} h ( i ) , we define Hermitian metrics H j ( i ) H_j^{(i)} H j ( i ) on K Y i − 1 → Y i K_{Y_i}^{-1}\to Y_i K Y i − 1 → Y i for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 by
H j ( i ) : = h j ( i ) − 1 ⊗ h j + 1 ( i ) , H_j^{(i)}:=h_j^{(i)-1}\otimes h_{j+1}^{(i)}, H j ( i ) := h j ( i ) − 1 ⊗ h j + 1 ( i ) ,
and define H 0 ( i ) = H r ( i ) H_0^{(i)}=H_r^{(i)} H 0 ( i ) = H r ( i ) by
H 0 ( i ) = H r ( i ) : = e r φ h ∗ − r ∣ Y i ⊗ h r ( i ) − 1 ⊗ h 1 ( i ) . H_0^{(i)}=H_r^{(i)}:=e^{r\varphi}h_*^{-r}|_{Y_i}\otimes h_r^{(i)-1}\otimes h_1^{(i)}. H 0 ( i ) = H r ( i ) := e r φ h ∗ − r ∣ Y i ⊗ h r ( i ) − 1 ⊗ h 1 ( i ) .
Lemma 52. Suppose that e φ e^{\varphi} e φ is continuous and a bounded function. Then there exists a positive constant C M φ C_{M_{\varphi}} C M φ depending only on M φ M_{\varphi} M φ and r r r , such that for each i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … , by choosing the boundary condition appropriately, a real solution h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) h^{(i)}=(h_1^{(i)},\ldots,h_r^{(i)}) h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) to the Dirichlet problem for equation (5) on Y i Y_i Y i satisfies the following estimate:
1 2 ω X ≤ vol ( H j ( i ) ) ≤ C M φ ω X for each j = 1 , … , n . (84) \frac{1}{2}\omega_X\leq\operatorname{vol}(H_j^{(i)})\leq C_{M_{\varphi}}\omega_X\quad\text{for each }j=1,\ldots,n. \tag*{(84)} 2 1 ω X ≤ vol ( H j ( i ) ) ≤ C M φ ω X for each j = 1 , … , n . (84)
Proof. Let C 1 C_1 C 1 and C 2 C_2 C 2 be constants as in inequality (45). We set C M φ C_{M_{\varphi}} C M φ as follows:
C M φ : = max { C 2 M φ + 1 C 1 , r − 1 2 + 2 r − 1 M φ r } . C_{M_{\varphi}}:=\max\left\{\frac{C_2M_{\varphi}+1}{C_1},\frac{r-1}{2}+2^{r-1}M_{\varphi}^{r}\right\}. C M φ := max { C 1 C 2 M φ + 1 , 2 r − 1 + 2 r − 1 M φ r } .
For each i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … , we define a Hermitian metric ∂ h ( i ) : = ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) \partial h^{(i)}:=(\partial h_1^{(i)},\ldots,\partial h_r^{(i)}) ∂ h ( i ) := ( ∂ h 1 ( i ) , … , ∂ h r ( i ) ) on the boundary ∂ Y i \partial Y_i ∂ Y i of Y i Y_i Y i as the unique one such that
( ∂ h j ( i ) ) − 1 ⊗ ∂ h j + 1 ( i ) = 1 2 h X − 1 ∣ ∂ Y i (\partial h_j^{(i)})^{-1}\otimes\partial h_{j+1}^{(i)}=\frac{1}{2}h_X^{-1}\big|_{\partial Y_i} ( ∂ h j ( i ) ) − 1 ⊗ ∂ h j + 1 ( i ) = 2 1 h X − 1 ∂ Y i
holds for all j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 and that det ( ∂ h ( i ) ) = 1 \det(\partial h^{(i)})=1 det ( ∂ h ( i ) ) = 1 . Since ∂ h ( i ) \partial h^{(i)} ∂ h ( i ) is real, h ( i ) h^{(i)} h ( i ) is also real. We show that the Hermitian metrics H 1 ( i ) , … , H r − 1 ( i ) H_1^{(i)},\ldots,H_{r-1}^{(i)} H 1 ( i ) , … , H r − 1 ( i ) , constructed from the solution h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) h^{(i)}=(h_1^{(i)},\ldots,h_r^{(i)}) h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) to the Dirichlet problem for equation (5) with boundary condition lim z → ζ h ( i ) ( z ) = ∂ h ( i ) ( ζ ) \lim_{z\to\zeta}h^{(i)}(z)=\partial h^{(i)}(\zeta) lim z → ζ h ( i ) ( z ) = ∂ h ( i ) ( ζ ) , satisfy the estimate (84). We first prove the estimate 1 2 ω X ≤ vol ( H j ( i ) ) \frac{1}{2}\omega_X\leq\operatorname{vol}(H_j^{(i)}) 2 1 ω X ≤ vol ( H j ( i ) ) for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , H j ( i ) H_j^{(i)} H j ( i ) satisfies the following:
− 1 ∂ ∂ ˉ log ( H j ( i ) ⊗ h X ) = − − 1 F H j ( i ) − − 1 F h X = 2 vol ( H j ( i ) ) − vol ( H j − 1 ( i ) ) − vol ( H j + 1 ( i ) ) − ω X ≤ 2 vol ( H j ( i ) ) − ω X . \begin{aligned}
\sqrt{-1}\partial\bar{\partial}\log(H_j^{(i)}\otimes h_X)
&=-\sqrt{-1}F_{H_j^{(i)}}-\sqrt{-1}F_{h_X}\\
&=2\operatorname{vol}(H_j^{(i)})-\operatorname{vol}(H_{j-1}^{(i)})-\operatorname{vol}(H_{j+1}^{(i)})-\omega_X\\
&\leq2\operatorname{vol}(H_j^{(i)})-\omega_X.
\end{aligned} − 1 ∂ ∂ ˉ log ( H j ( i ) ⊗ h X ) = − − 1 F H j ( i ) − − 1 F h X = 2 vol ( H j ( i ) ) − vol ( H j − 1 ( i ) ) − vol ( H j + 1 ( i ) ) − ω X ≤ 2 vol ( H j ( i ) ) − ω X .
Therefore we obtain the following:
− 1 Λ h X − 1 ∂ ∂ ˉ log ( H j ( i ) ⊗ h X ) ≤ 2 H j ( i ) ⊗ h X − 1. (85) \sqrt{-1}\Lambda_{h_X^{-1}}\partial\bar{\partial}\log(H_j^{(i)}\otimes h_X)\leq2H_j^{(i)}\otimes h_X-1. \tag*{(85)} − 1 Λ h X − 1 ∂ ∂ ˉ log ( H j ( i ) ⊗ h X ) ≤ 2 H j ( i ) ⊗ h X − 1. (85)
Note that by Proposition 69 and by the assumption that e φ e^{\varphi} e φ is continuous, H j ( i ) H_j^{(i)} H j ( i ) (j = 1 , … , n j=1,\ldots,n j = 1 , … , n ) are of class C 2 C^2 C 2 on Y i Y_i Y i , and thus we can apply the strong maximum principle to inequality (85). By the boundary condition and the strong maximum principle, we have 1 2 ω X ≤ vol ( H j ( i ) ) \frac{1}{2}\omega_X\leq\operatorname{vol}(H_j^{(i)}) 2 1 ω X ≤ vol ( H j ( i ) ) on Y i Y_i Y i . We next prove the estimate vol ( H j ( i ) ) ≤ C M φ ω X \operatorname{vol}(H_j^{(i)})\leq C_{M_{\varphi}}\omega_X vol ( H j ( i ) ) ≤ C M φ ω X . Since a locally bounded subharmonic function belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 (cf. [21 Section 1.1.1] ), for each smooth Hermitian metric H H H on K Y i − 1 K_{Y_i}^{-1} K Y i − 1 , log ( ∑ j = 1 r H j ( i ) ⊗ H − 1 ) \log(\sum_{j=1}^{r}H_j^{(i)}\otimes H^{-1}) log ( ∑ j = 1 r H j ( i ) ⊗ H − 1 ) is continuous and belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . Then by applying Proposition 23 to inequality (45) for H j = H j ( i ) H_j=H_j^{(i)} H j = H j ( i ) (j = 1 , … , r j=1,\ldots,r j = 1 , … , r ) and H = M φ h X − 1 H=M_{\varphi}h_X^{-1} H = M φ h X − 1 , one of the following holds:
0 ≥ C 1 ∑ j = 1 r H j ( i ) ⊗ H − 1 − C 2 − M φ − 1 , 0\geq C_1\sum_{j=1}^{r}H_j^{(i)}\otimes H^{-1}-C_2-M_{\varphi}^{-1}, 0 ≥ C 1 j = 1 ∑ r H j ( i ) ⊗ H − 1 − C 2 − M φ − 1 ,
r − 1 2 M φ − 1 + 2 r − 1 ( max ∂ Y i e r φ ) M φ − 1 ≥ ∑ j = 1 r H j ( i ) ⊗ H − 1 . \frac{r-1}{2}M_{\varphi}^{-1}+2^{r-1}(\max_{\partial Y_i}e^{r\varphi})M_{\varphi}^{-1}\geq\sum_{j=1}^{r}H_j^{(i)}\otimes H^{-1}. 2 r − 1 M φ − 1 + 2 r − 1 ( ∂ Y i max e r φ ) M φ − 1 ≥ j = 1 ∑ r H j ( i ) ⊗ H − 1 .
In either case, we have vol ( H j ( i ) ) ≤ ∑ j = 1 r vol ( H j ( i ) ) ≤ C M φ ω X \operatorname{vol}(H_j^{(i)})\leq\sum_{j=1}^{r}\operatorname{vol}(H_j^{(i)})\leq C_{M_{\varphi}}\omega_X vol ( H j ( i ) ) ≤ ∑ j = 1 r vol ( H j ( i ) ) ≤ C M φ ω X for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . This implies the claim. ∎
Then following [43 Proposition 5.6] , we show the following:
Lemma 53. Suppose that e φ e^{\varphi} e φ is continuous and bounded. Then there exists a real C 2 C^{2} C 2 complete solution to equation (5).
Proof. By Lemma 52 and the interior L p L^{p} L p -estimate (cf. [27 Theorem 9.11] ), for each i ≥ 2 i \ge2 i ≥ 2 , 1 < p < ∞ 1 < p < \infty 1 < p < ∞ , and l ≥ 0 l \ge0 l ≥ 0 , there exist positive constants C 1 C_{1} C 1 and C 2 C_{2} C 2 that are independent of l l l such that
∣ log h ( i + 1 + l ) ∣ L 2 p ( Y i ) ≤ C 1 ( ∣ log h ( i + 1 + l ) ∣ L p ( Y i + 1 ) + ∣ F h ( i + 1 + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 , \left|\log h^{(i+1+l)}\right|_{L_{2}^{p}(Y_{i})} \le C_{1}\left(\left|\log h^{(i+1+l)}\right|_{L^{p}(Y_{i+1})}+\left|F_{h^{(i+1+l)}}\right|_{L^{p}(Y_{i+1})}\right) \le C_{2}, log h ( i + 1 + l ) L 2 p ( Y i ) ≤ C 1 ( log h ( i + 1 + l ) L p ( Y i + 1 ) + ∣ F h ( i + 1 + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 , where the L 2 p L_{2}^{p} L 2 p -norm and the L p L^{p} L p -norm are measured by the Poincaré metric. Since L 2 p ( Y i ) L_{2}^{p}(Y_{i}) L 2 p ( Y i ) is reflexive, every bounded sequence in L 2 p ( Y i ) L_{2}^{p}(Y_{i}) L 2 p ( Y i ) admits a weakly convergent subsequence. Hence, for each i ≥ 2 i \ge2 i ≥ 2 , we can extract a subsequence of ( h ( i + 1 + l ) ) l ≥ 0 \left(h^{(i+1+l)}\right)_{l\ge0} ( h ( i + 1 + l ) ) l ≥ 0 that converges weakly in L 2 p ( Y i ) L_{2}^{p}(Y_{i}) L 2 p ( Y i ) to a Hermitian metric on Y i Y_{i} Y i . By a diagonal argument, we can then extract a subsequence of ( h ( k ) ) k ≥ 2 \left(h^{(k)}\right)_{k\ge2} ( h ( k ) ) k ≥ 2 that converges weakly in L 2 , l o c p L_{2,\mathrm{loc}}^{p} L 2 , loc p on X X X to an L 2 , l o c p L_{2,\mathrm{loc}}^{p} L 2 , loc p -Hermitian metric h = ( h 1 , … , h r ) h=(h_{1},\ldots,h_{r}) h = ( h 1 , … , h r ) . The Hermitian metric h h h solves equation (5) in the weak sense. By Proposition 69, h h h has a C 2 C^{2} C 2 representative and solves equation (5) in the strong sense. Since h ( i ) h^{(i)} h ( i ) (i = 2 , 3 , … , i=2,3,\ldots, i = 2 , 3 , … , ) are real, h h h is also real. By Lemma 52, h h h satisfies the following estimate:
1 2 ω X ≤ vol ( H j ) ≤ C M φ ω X for each j = 1 , … , r − 1 , (86) \frac{1}{2}\omega_{X}\le\operatorname{vol}(H_{j})\le C_{M_{\varphi}}\omega_{X}\quad\text{for each }j=1,\ldots,r-1, \tag*{(86)} 2 1 ω X ≤ vol ( H j ) ≤ C M φ ω X for each j = 1 , … , r − 1 , (86) where for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , we have defined H j H_{j} H j as H j : = h j − 1 ⊗ h j + 1 H_{j}:=h_{j}^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 . In particular, h h h is complete.
Following [43 Proposition 5.6] , we drop the assumption that e φ e^{\varphi} e φ is a bounded function.
Lemma 54. Suppose that e φ e^{\varphi} e φ is continuous. Then there exists a real C 2 C^{2} C 2 complete solution h h h to equation (5).
Proof. For each i ≥ 2 i \ge2 i ≥ 2 , let h c ( i ) = ( h c , 1 ( i ) , … , h c , r ( i ) ) h_{c}^{(i)}=(h_{c,1}^{(i)},\ldots,h_{c,r}^{(i)}) h c ( i ) = ( h c , 1 ( i ) , … , h c , r ( i ) ) be the unique complete solution to equation (5) on Y i Y_{i} Y i associated with e − φ h ∗ ∣ Y i e^{-\varphi}h_{\ast}|_{Y_{i}} e − φ h ∗ ∣ Y i . By (86), h c ( i ) h_{c}^{(i)} h c ( i ) is dominated by the complete hyperbolic metric as follows:
h c , k ( i ) ≤ ( 2 h Y i ) r + 1 − 2 k 2 for each k = 1 , … , n , (87) h_{c,k}^{(i)}\le(2h_{Y_{i}})^{\frac{r+1-2k}{2}}\quad\text{for each }k=1,\ldots,n, \tag*{(87)} h c , k ( i ) ≤ ( 2 h Y i ) 2 r + 1 − 2 k for each k = 1 , … , n , (87) where we denote by h Y i h_{Y_{i}} h Y i the Hermitian metric on K Y i → Y i K_{Y_{i}}\to Y_{i} K Y i → Y i induced by the complete hyperbolic metric on Y i Y_{i} Y i . Since h Y i h_{Y_{i}} h Y i tends to zero as z z z approaches ∂ Y i \partial Y_{i} ∂ Y i , it follows that h c , j ( k ) − 1 ⊗ h c , j ( i ) ( z ) h_{c,j}^{(k)^{-1}}\otimes h_{c,j}^{(i)}(z) h c , j ( k ) − 1 ⊗ h c , j ( i ) ( z ) goes to 0 as z z z approaches ∂ Y i \partial Y_{i} ∂ Y i , for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . In particular, h c , j ( k ) − 1 ⊗ h c , j ( i ) h_{c,j}^{(k)^{-1}}\otimes h_{c,j}^{(i)} h c , j ( k ) − 1 ⊗ h c , j ( i ) is bounded above for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Then by Lemma 49, we have h c , j ( i ) < h c , j ( k ) h_{c,j}^{(i)}<h_{c,j}^{(k)} h c , j ( i ) < h c , j ( k ) for each i < k i<k i < k and each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . By (87), we obtain the following uniform estimate:
h c , k ( i ) ≤ ( 2 h X ) r + 1 − 2 k 2 for each k = 1 , … , n . h_{c,k}^{(i)}\le(2h_{X})^{\frac{r+1-2k}{2}}\quad\text{for each }k=1,\ldots,n. h c , k ( i ) ≤ ( 2 h X ) 2 r + 1 − 2 k for each k = 1 , … , n . Therefore in particular we see that ( h c ( k ) ) k ≥ i + 1 \left(h_{c}^{(k)}\right)_{k\ge i+1} ( h c ( k ) ) k ≥ i + 1 is uniformly bounded above and below on Y i Y_{i} Y i . Then by the interior L p L^{p} L p -estimate ([27 Theorem 9.11] ), for each i ≥ 2 i\ge2 i ≥ 2 , 1 < p < ∞ 1<p<\infty 1 < p < ∞ , and l ≥ 0 l\ge0 l ≥ 0 , there exist positive constants C 1 C_{1} C 1 and C 2 C_{2} C 2 that are independent of l l l such that
∣ log h c ( i + 2 + l ) ∣ L 2 p ( Y i ) ≤ C 1 ( ∣ log h c ( i + 2 + l ) ∣ L p ( Y i + 1 ) + ∣ F h c ( i + 2 + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 , \left|\log h_{c}^{(i+2+l)}\right|_{L_{2}^{p}(Y_{i})}\le C_{1}\left(\left|\log h_{c}^{(i+2+l)}\right|_{L^{p}(Y_{i+1})}+\left|F_{h_{c}^{(i+2+l)}}\right|_{L^{p}(Y_{i+1})}\right)\le C_{2}, log h c ( i + 2 + l ) L 2 p ( Y i ) ≤ C 1 ( log h c ( i + 2 + l ) L p ( Y i + 1 ) + F h c ( i + 2 + l ) L p ( Y i + 1 ) ) ≤ C 2 ,
where the L p L^{p} L p -norm and the L 2 p L^{p}_{2} L 2 p -norms are measured by the Poincaré metric on X X X . Since every bounded sequence in a reflexive Banach space has a weakly convergent subsequence, for each i ≥ 2 i \ge2 i ≥ 2 , we can find a subsequence of ( h c ( i + 2 + l ) ) l ≥ 0 (h_{c}^{(i+2+l)})_{l\ge0} ( h c ( i + 2 + l ) ) l ≥ 0 that converges to a Hermitian metric h h h on Y i Y_i Y i in the L 2 p ( Y i ) L^{p}_{2}(Y_i) L 2 p ( Y i ) -weak sense. By using the diagonal trick, we can find a subsequence of ( h c ( k ) ) k ≥ 2 (h_{c}^{(k)})_{k\ge2} ( h c ( k ) ) k ≥ 2 that converges to an L 2 , l o c p L^{p}_{2,\mathrm{loc}} L 2 , loc p -Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on X X X in the L 2 , l o c p L^{p}_{2,\mathrm{loc}} L 2 , loc p -weak sense. The Hermitian metric h h h solves equation (5) in the weak sense. By Proposition 69, h h h is C 2 C^2 C 2 and solves equation (5) in the strong sense. Since h c ( i ) h_c^{(i)} h c ( i ) (i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … ) are real, h h h is also real. By (86), for each i i i , the complete solution h c ( i ) h_c^{(i)} h c ( i ) satisfies the following estimate:
( h c , j ( i ) ) − 1 ⊗ h c , j + 1 ( i ) ≥ 1 2 h Y i − 1 for each j = 1 , … , n . (h_{c,j}^{(i)})^{-1}\otimes h_{c,j+1}^{(i)} \ge\frac{1}{2}h_{Y_i}^{-1}\ \text{for each }j=1,\ldots,n. ( h c , j ( i ) ) − 1 ⊗ h c , j + 1 ( i ) ≥ 2 1 h Y i − 1 for each j = 1 , … , n .
Therefore, the solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) satisfies the following estimate:
h j − 1 ⊗ h j + 1 ≥ 1 2 h X − 1 for each j = 1 , … , n . (88) h_j^{-1}\otimes h_{j+1}\ge\frac{1}{2}h_X^{-1}\ \text{for each }j=1,\ldots,n. \tag*{(88)} h j − 1 ⊗ h j + 1 ≥ 2 1 h X − 1 for each j = 1 , … , n . (88)
In particular, h h h is complete.
By the proof of Lemma 54, we directly obtain the following lemma:
Lemma 55. Suppose that e φ e^\varphi e φ is continuous. Then the unique complete solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) to equation (5) associated with φ \varphi φ satisfying det ( h ) = 1 \det(h)=1 det ( h ) = 1 satisfies the following estimate:
h j − 1 ⊗ h j + 1 ≥ 1 2 h X − 1 for each j = 1 , … , n . h_j^{-1}\otimes h_{j+1}\ge\frac{1}{2}h_X^{-1}\ \text{for each }j=1,\ldots,n. h j − 1 ⊗ h j + 1 ≥ 2 1 h X − 1 for each j = 1 , … , n . In particular, the following holds:
h j ≤ ( 2 h X ) r + 1 − 2 j 2 for each j = 1 , … , n . h_j\le(2h_X)^{\frac{r+1-2j}{2}}\ \text{for each }j=1,\ldots,n. h j ≤ ( 2 h X ) 2 r + 1 − 2 j for each j = 1 , … , n .
Let ( φ ϵ ) 0 < ϵ < 1 (\varphi_\epsilon)_{0<\epsilon<1} ( φ ϵ ) 0 < ϵ < 1 be a family of smooth subharmonic weight functions satisfying the condition in Theorem 3, where each φ ϵ \varphi_\epsilon φ ϵ is defined on D ϵ \mathbb{D}_\epsilon D ϵ . By Theorem 1 and Lemma 54, for each 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , there exists a unique real complete solution h ϵ h_\epsilon h ϵ associated with φ ϵ \varphi_\epsilon φ ϵ . At the end of this subsection, we prove
Lemma 56. As ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 , h ϵ h_\epsilon h ϵ converges monotonically to a real complete solution h h h associated with φ \varphi φ .
Proof. As we remarked in Example 43, for each ϵ ′ < ϵ \epsilon'<\epsilon ϵ ′ < ϵ , h ϵ ′ h_{\epsilon'} h ϵ ′ is a supersolution to equation (5) associated with φ ϵ \varphi_\epsilon φ ϵ . Since D ϵ ⊆ D ϵ ′ \mathbb{D}_\epsilon\subseteq\mathbb{D}_{\epsilon'} D ϵ ⊆ D ϵ ′ , the metric h ϵ ′ h_{\epsilon'} h ϵ ′ is smooth and positive on a neighborhood of D ‾ ϵ \overline{\mathbb{D}}_\epsilon D ϵ . On the other hand, by Lemma 55, h ϵ h_\epsilon h ϵ is dominated by the complete hyperbolic metric on D ϵ \mathbb{D}_\epsilon D ϵ , and hence h ϵ , j ⊗ h ϵ , j − 1 ( z ) h_{\epsilon,j}\otimes h_{\epsilon,j}^{-1}(z) h ϵ , j ⊗ h ϵ , j − 1 ( z ) goes to 0 as z z z approaches ∂ D ϵ \partial\mathbb{D}_\epsilon ∂ D ϵ , for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . In particular, h ϵ , j ⊗ h ϵ , j − 1 h_{\epsilon,j}\otimes h_{\epsilon,j}^{-1} h ϵ , j ⊗ h ϵ , j − 1 is bounded above on D ϵ \mathbb{D}_\epsilon D ϵ . Therefore by Lemma 49, we obtain h ϵ , j ⊗ h ϵ ′ , j − 1 < 1 h_{\epsilon,j}\otimes h_{\epsilon',j}^{-1}<1 h ϵ , j ⊗ h ϵ ′ , j − 1 < 1 for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . For each i = 2 , 3 , … i=2,3,\ldots i = 2 , 3 , … , we denote by h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) h^{(i)}=(h_1^{(i)},\ldots,h_r^{(i)}) h ( i ) = ( h 1 ( i ) , … , h r ( i ) ) the complete solution h 1 / i h_{1/i} h 1/ i defined on Y i = D 1 / i Y_i=\mathbb{D}_{1/i} Y i = D 1/ i . Then by monotonicity, Lemma 55, and the interior L p L^p L p -estimate ([27, Theorem 9.11), for each i ≥ 2 i\ge2 i ≥ 2 , 1 < p < ∞ 1<p<\infty 1 < p < ∞ , and l ≥ 0 l\ge0 l ≥ 0 , there exist positive constants C 1 C_1 C 1 and C 2 C_2 C 2 that are independent of l l l such that
∣ log h ( i + 2 + l ) ∣ L 2 p ( Y i ) ≤ C 1 ( ∣ log h ( i + 2 + l ) ∣ L p ( Y i + 1 ) + ∣ F h ( i + 2 + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 , \left|\log h^{(i+2+l)}\right|_{L^p_2(Y_i)}
\le C_1\left(\left|\log h^{(i+2+l)}\right|_{L^p(Y_{i+1})}
+\left|F_{h^{(i+2+l)}}\right|_{L^p(Y_{i+1})}\right)\le C_2, log h ( i + 2 + l ) L 2 p ( Y i ) ≤ C 1 ( log h ( i + 2 + l ) L p ( Y i + 1 ) + ∣ F h ( i + 2 + l ) ∣ L p ( Y i + 1 ) ) ≤ C 2 ,
where the L p L^p L p -norm and the L 2 p L^p_2 L 2 p -norms are measured by the Poincaré metric on X X X . Then by compact Sobolev embedding, for each i ≥ 2 i \ge2 i ≥ 2 , we can find a subsequence of ( h ( i + 2 + l ) ) l ≥ 0 (h^{(i+2+l)})_{l\ge0} ( h ( i + 2 + l ) ) l ≥ 0 that converges to a Hermitian metric h h h on Y i ‾ \overline{Y_i} Y i in C 0 ( Y i ‾ ) C^0(\overline{Y_i}) C 0 ( Y i ) . By using the diagonal trick, we can find a subsequence of ( h ( k ) ) k ≥ 2 (h^{(k)})_{k\ge2} ( h ( k ) ) k ≥ 2 that locally uniformly converges to a continuous Hermitian metric h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) on X X X . Monotonicity then implies that the whole family ( h ϵ ) 0 < ϵ < 1 (h_\epsilon)_{0<\epsilon<1} ( h ϵ ) 0 < ϵ < 1 converges pointwise to h h h as ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 . By dominated convergence, h h h solves equation (5) associated with φ \varphi φ in the weak sense. Since h ( i ) h^{(i)} h ( i ) is real for each i i i , h h h is also real. By (88) in the proof of Lemma 54, for each 0 < ϵ < 1 0<\epsilon<1 0 < ϵ < 1 , we have the following estimate:
h ϵ , j − 1 ⊗ h ϵ , j + 1 ≥ 1 2 h D ϵ − 1 for each j = 1 , … , n , h_{\epsilon,j}^{-1}\otimes h_{\epsilon,j+1}\ge\frac{1}{2}h_{\mathbb{D}_\epsilon}^{-1}\quad\text{for each }j=1,\ldots,n, h ϵ , j − 1 ⊗ h ϵ , j + 1 ≥ 2 1 h D ϵ − 1 for each j = 1 , … , n ,
where h D ϵ h_{\mathbb{D}_\epsilon} h D ϵ is the Hermitian metric induced by the hyperbolic metric. Therefore in the limit ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 , we obtain
h j − 1 ⊗ h j + 1 ≥ 1 2 h X − 1 for each j = 1 , … , n . h_j^{-1}\otimes h_{j+1}\ge\frac{1}{2}h_X^{-1}\quad\text{for each }j=1,\ldots,n. h j − 1 ⊗ h j + 1 ≥ 2 1 h X − 1 for each j = 1 , … , n .
In particular, the metric h h h is complete.
As a corollary, we obtain
Corollary 57. On the unit disc, for any subharmonic weight function φ \varphi φ , there exists a unique real complete solution h h h associated with φ \varphi φ . Moreover, for each smooth approximation ( φ ϵ ) 0 < ϵ < 1 (\varphi_\epsilon)_{0<\epsilon<1} ( φ ϵ ) 0 < ϵ < 1 of φ \varphi φ satisfying the condition in Theorem 3, where each φ ϵ \varphi_\epsilon φ ϵ is defined on the disc D ϵ \mathbb{D}_\epsilon D ϵ , the corresponding family ( h ϵ ) 0 < ϵ < 1 (h_\epsilon)_{0<\epsilon<1} ( h ϵ ) 0 < ϵ < 1 of real complete solutions monotonically converges to the complete solution h h h .
Proof of Theorems 2 and 3 for the complex plane case
Throughout this subsection, we assume that the base Riemann surface is the complex plane C \mathbb{C} C . We also assume that the weight function φ \varphi φ is not identically − ∞ -\infty − ∞ . By Lemma 56, on the unit disc, there exists a complete solution to equation (5) . Therefore, by combining Theorem 1 with the standard argument in [43 Proposition 5.7] , on every hyperbolic Riemann surface, there exists a unique real complete solution to equation (5) ; a detailed proof will be given in Section 3.9. The estimate in Lemma 55 also holds on these surfaces for arbitrary subharmonic weight functions, by Lemma 56 and the same covering argument. We denote by D R = { z ∈ C ∣ ∣ z ∣ < R } \mathbb{D}_R=\{z\in\mathbb{C}\mid|z|<R\} D R = { z ∈ C ∣ ∣ z ∣ < R } the open ball of radius R R R centered at the origin. We take positive real numbers 0 < R 2 < R 1 0<R_2<R_1 0 < R 2 < R 1 . Let h < R 1 = ( h 1 < R 1 , … , h r < R 1 ) h^{<R_1}=(h_1^{<R_1},\ldots,h_r^{<R_1}) h < R 1 = ( h 1 < R 1 , … , h r < R 1 ) be the real complete solution on D R 1 \mathbb{D}_{R_1} D R 1 associated with φ ∣ D R 1 \varphi|_{\mathbb{D}_{R_1}} φ ∣ D R 1 , and let h > R 2 = ( h 1 > R 2 , … , h r > R 2 ) h^{>R_2}=(h_1^{>R_2},\ldots,h_r^{>R_2}) h > R 2 = ( h 1 > R 2 , … , h r > R 2 ) be the real complete solution on C ∖ D ‾ R 2 \mathbb{C}\setminus\overline{\mathbb{D}}_{R_2} C ∖ D R 2 associated with φ ∣ C ∖ D ‾ R 2 \varphi|_{\mathbb{C}\setminus\overline{\mathbb{D}}_{R_2}} φ ∣ C ∖ D R 2 . Following the argument in [43 Proposition 5.9] , we first establish the following lemma:
Lemma 58. For each 0 < R 2 < R 1 0<R_2<R_1 0 < R 2 < R 1 , define a diagonal Hermitian metric h − ( R 1 , R 2 ) = ( h − 1 , … , h − r ) h_-^{(R_1,R_2)}=(h_{-1},\ldots,h_{-r}) h − ( R 1 , R 2 ) = ( h − 1 , … , h − r ) as follows. For j = 1 , … , n j=1,\ldots,n j = 1 , … , n , set
h − , j ( R 1 , R 2 ) : = { h j < R 1 on D ‾ R 2 , max { h j > R 2 , h j < R 1 } on { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } , h j > R 2 on C ∖ D R 1 . h_{-,j}^{(R_1,R_2)}:=
\begin{cases}
h_j^{<R_1} & \text{on }\overline{\mathbb{D}}_{R_2},\\
\max\{h_j^{>R_2},h_j^{<R_1}\} & \text{on }\{z\in\mathbb{C}\mid R_2<|z|<R_1\},\\
h_j^{>R_2} & \text{on }\mathbb{C}\setminus\mathbb{D}_{R_1}.
\end{cases} h − , j ( R 1 , R 2 ) := ⎩ ⎨ ⎧ h j < R 1 max { h j > R 2 , h j < R 1 } h j > R 2 on D R 2 , on { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } , on C ∖ D R 1 . The remaining components h − , j ( R 1 , R 2 ) h_{-,j}^{(R_1,R_2)} h − , j ( R 1 , R 2 ) , j = n + 1 , … , r j=n+1,\ldots,r j = n + 1 , … , r , are defined so that h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) is real. Then h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) is a continuous subsolution to equation (5) and belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 .
Proof. By Proposition 47, h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) is a subsolution to equation (5) on { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } \{z \in\mathbb{C} \mid R_2 < |z| < R_1\} { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } . Since max : R 2 → R \max: \mathbb{R}^2 \to\mathbb{R} max : R 2 → R is Lipschitz continuous, it belongs to W l o c 1 , ∞ W_{\mathrm{loc}}^{1,\infty} W loc 1 , ∞ (cf. [25 Theorem 4.5] ). Therefore, since h > R 2 h_{>R_2} h > R 2 and h < R 1 h_{<R_1} h < R 1 belong to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 , by the chain rule for Sobolev functions, h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 on { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } \{z \in\mathbb{C} \mid R_2 < |z| < R_1\} { z ∈ C ∣ R 2 < ∣ z ∣ < R 1 } . Moreover, h j < R 1 h_j^{<R_1} h j < R 1 and h j > R 2 h_j^{>R_2} h j > R 2 , j = 1 , … , n j = 1,\ldots,n j = 1 , … , n , satisfy the estimates in terms of the hyperbolic metrics given in Lemma 55 as noted at the beginning of this subsection. Although that lemma assumes that e φ e^\varphi e φ is continuous, we obtain the same estimates for the corresponding limiting complete solutions by passing to the limit along the smooth approximations in Lemma 56. Therefore, h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) coincides with h < R 1 h^{<R_1} h < R 1 near the boundary of D R 2 \mathbb{D}_{R_2} D R 2 and with h > R 2 h^{>R_2} h > R 2 near the boundary of D R 1 \mathbb{D}_{R_1} D R 1 . Hence, h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) is continuous and belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . This completes the proof.
In addition to R 1 R_1 R 1 and R 2 R_2 R 2 , we take positive real numbers R 1 ′ R'_1 R 1 ′ , R 2 ′ R'_2 R 2 ′ satisfying 0 < R 2 < R 2 ′ < R 1 ′ < R 1 0 < R_2 < R'_2 < R'_1 < R_1 0 < R 2 < R 2 ′ < R 1 ′ < R 1 . Let h ( φ ) = ( h 1 ( φ ) , … , h r ( φ ) ) h^{(\varphi)} = (h_1^{(\varphi)}, \ldots, h_r^{(\varphi)}) h ( φ ) = ( h 1 ( φ ) , … , h r ( φ ) ) be the supersolution constructed from e − φ h ∗ e^{-\varphi}h_* e − φ h ∗ as in Example 42. For each positive constant c c c , define a diagonal Hermitian metric h + ( R 2 ′ , R 2 ) = ( h + , 1 ( R 2 ′ , R 2 ) , … , h + , r ( R 2 ′ , R 2 ) ) h_+^{(R'_2,R_2)} = (h_{+,1}^{(R'_2,R_2)}, \ldots, h_{+,r}^{(R'_2,R_2)}) h + ( R 2 ′ , R 2 ) = ( h + , 1 ( R 2 ′ , R 2 ) , … , h + , r ( R 2 ′ , R 2 ) ) as follows. For j = 1 , … , n j = 1,\ldots,n j = 1 , … , n , set
h + , j ( R 2 ′ , R 2 ) : = { h j ( φ ) on D R 1 ′ , min { h j ( φ ) , e c h j > R 2 } on C ∖ D R 2 ′ . h_{+,j}^{(R'_2,R_2)} :=
\begin{cases}
h_j^{(\varphi)} & \text{on }\mathbb{D}_{R'_1},\\
\min\{h_j^{(\varphi)},e^c h_j^{>R_2}\} & \text{on }\mathbb{C}\setminus\mathbb{D}_{R'_2}.
\end{cases} h + , j ( R 2 ′ , R 2 ) := { h j ( φ ) min { h j ( φ ) , e c h j > R 2 } on D R 1 ′ , on C ∖ D R 2 ′ .
The remaining components h + , j ( R 2 ′ , R 2 ) h_{+,j}^{(R'_2,R_2)} h + , j ( R 2 ′ , R 2 ) , j = n + 1 , … , r j = n+1,\ldots,r j = n + 1 , … , r , are defined so that h + ( R 2 ′ , R 2 ) h_+^{(R'_2,R_2)} h + ( R 2 ′ , R 2 ) is real. By Examples 42 and 44 and Proposition 47, outside the boundary of D R 2 ′ \mathbb{D}_{R'_2} D R 2 ′ , h + ( R 2 ′ , R 2 ) h_+^{(R'_2,R_2)} h + ( R 2 ′ , R 2 ) is a supersolution to equation (5). Following the argument in [43 Proposition 5.9] , we prove the following:
Lemma 59. The following holds:
(i) Suppose that the following conditions hold:
On D ‾ R 1 ′ \overline{\mathbb{D}}_{R'_1} D R 1 ′ , e − φ e^{-\varphi} e − φ is finite and continuous.
On C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) , φ \varphi φ is continuous and belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 .
The positive constant c c c is large enough so that the following holds on { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ ≤ R 1 ′ } \{z \in\mathbb{C} \mid R'_2 \le|z| \le R'_1\} { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ ≤ R 1 ′ } :
h j ( φ ) < e c h j > R 2 for j = 1 , … , n . (89) h_j^{(\varphi)} < e^c h_j^{>R_2} \quad\text{for } j = 1,\ldots,n. \tag*{(89)} h j ( φ ) < e c h j > R 2 for j = 1 , … , n . (89) Then h + ( R 2 ′ , R 2 ) h_+^{(R'_2,R_2)} h + ( R 2 ′ , R 2 ) is continuous everywhere and belongs to W l o c 1 , 2 ( C ) W_{\mathrm{loc}}^{1,2}(\mathbb{C}) W loc 1 , 2 ( C ) . In particular, h + ( R 2 ′ , R 2 ) h_+^{(R'_2,R_2)} h + ( R 2 ′ , R 2 ) is a supersolution on the entire complex plane, including the boundary of D R 2 ′ \mathbb{D}_{R'_2} D R 2 ′ .
(ii) Suppose that c c c is large enough so that the following holds on { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ < R 1 } \{z \in\mathbb{C} \mid R'_2 \le|z| < R_1\} { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ < R 1 } :
h j < R 1 < e c h j > R 2 for j = 1 , … , n . (90) h_j^{<R_1} < e^c h_j^{>R_2} \quad\text{for } j = 1,\ldots,n. \tag*{(90)} h j < R 1 < e c h j > R 2 for j = 1 , … , n . (90) Then we obtain
h − , j ( R 1 , R 2 ) < h + , j ( R 2 ′ , R 2 ) for j = 1 , … , n . (91) h_{-,j}^{(R_1,R_2)} < h_{+,j}^{(R'_2,R_2)} \quad\text{for } j = 1,\ldots,n. \tag*{(91)} h − , j ( R 1 , R 2 ) < h + , j ( R 2 ′ , R 2 ) for j = 1 , … , n . (91)
Proof. We first prove assertion (i). We begin by noting the following two points:
By the subharmonicity of φ \varphi φ and its continuity outside the polar set φ − 1 ( − ∞ ) \varphi^{-1}(-\infty) φ − 1 ( − ∞ ) , the polar set is a closed set.
By the upper semicontinuity of φ \varphi φ and continuity of h j > R 2 h^{>R_2}_j h j > R 2 , for each x ∈ φ − 1 ( − ∞ ) x \in\varphi^{-1}(-\infty) x ∈ φ − 1 ( − ∞ ) , there exists an open neighborhood U U U of x x x such that min { h j ( φ ) , e c h j > R 2 } = e c h j > R 2 \min\{h^{(\varphi)}_j,e^c h^{>R_2}_j\}=e^c h^{>R_2}_j min { h j ( φ ) , e c h j > R 2 } = e c h j > R 2 on U U U .
Therefore, since e c h > R 2 e^c h^{>R_2} e c h > R 2 is continuous and belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 , it is enough to prove, on the open set C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) , h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) is continuous and belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 . We then prove the global continuity of h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) . It remains only to check continuity across the boundary of D R 2 ′ \mathbb{D}_{R'_2} D R 2 ′ . Since e − φ e^{-\varphi} e − φ is continuous on D ‾ R 1 ′ \overline{\mathbb{D}}_{R'_1} D R 1 ′ , so is h ( φ ) h^{(\varphi)} h ( φ ) . Moreover, by (89), h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) coincides with h ( φ ) h^{(\varphi)} h ( φ ) in a neighborhood of ∂ D R 2 ′ : = { z ∈ C ∣ ∣ z ∣ = R 2 ′ } \partial\mathbb{D}_{R'_2}:=\{z\in\mathbb{C}\mid|z|=R'_2\} ∂ D R 2 ′ := { z ∈ C ∣ ∣ z ∣ = R 2 ′ } . This implies the continuity of h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) . We then prove that h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 . Since φ \varphi φ is continuous on C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) , it is locally bounded on C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) . Therefore, the assumption that φ \varphi φ belongs to W l o c 1 , 2 ( C ∖ φ − 1 ( − ∞ ) ) W^{1,2}_{\mathrm{loc}}(\mathbb{C}\setminus\varphi^{-1}(-\infty)) W loc 1 , 2 ( C ∖ φ − 1 ( − ∞ )) implies that h ( φ ) h^{(\varphi)} h ( φ ) also belongs to W l o c 1 , 2 ( C ∖ φ − 1 ( − ∞ ) ) . S i n c e W^{1,2}_{\mathrm{loc}}(\mathbb{C}\setminus\varphi^{-1}(-\infty)). Since W loc 1 , 2 ( C ∖ φ − 1 ( − ∞ )) . S in ce h^{(R'_2,R_2)}_+$ coincides with h ( φ ) h^{(\varphi)} h ( φ ) on a neighbourhood of D ‾ R 2 ′ \overline{\mathbb{D}}_{R'_2} D R 2 ′ , it is enough to prove the assertion on C ∖ φ − 1 ( − ∞ ) ∩ C ∖ D ‾ R 2 ′ \mathbb{C}\setminus\varphi^{-1}(-\infty)\cap\mathbb{C}\setminus\overline{\mathbb{D}}_{R'_2} C ∖ φ − 1 ( − ∞ ) ∩ C ∖ D R 2 ′ . Since h > R 2 h^{>R_2} h > R 2 belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 and min : R 2 → R \min:\mathbb{R}^2\to\mathbb{R} min : R 2 → R is Lipschitz continuous, the chain rule for Sobolev functions implies that min { h j ( φ ) , e c h j > R 2 } \min\{h^{(\varphi)}_j,e^c h^{>R_2}_j\} min { h j ( φ ) , e c h j > R 2 } belongs to W l o c 1 , 2 W^{1,2}_{\mathrm{loc}} W loc 1 , 2 for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Therefore, it follows that h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) belongs to W l o c 1 , 2 ( C ) W^{1,2}_{\mathrm{loc}}(\mathbb{C}) W loc 1 , 2 ( C ) . The supersolution property on all of C \mathbb{C} C follows immediately, given that h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) coincides with h ( φ ) h^{(\varphi)} h ( φ ) in the neighbourhood of the boundary of D R 2 ′ \mathbb{D}_{R'_2} D R 2 ′ . This proves assertion (i). We next prove assertion (ii). From Lemma 32 and [50 Proposition 27] , we have
h j < R 1 < h j ( φ ) and h j > R 2 < h j ( φ ) (92) h_j^{<R_1}<h_j^{(\varphi)} \quad\text{and}\quad h_j^{>R_2}<h_j^{(\varphi)} \tag*{(92)} h j < R 1 < h j ( φ ) and h j > R 2 < h j ( φ ) (92)
for all j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Therefore, on D R 2 ′ \mathbb{D}_{R'_2} D R 2 ′ , we obtain (91) from (92). On D R 1 ∖ D R 2 ′ \mathbb{D}_{R_1}\setminus\mathbb{D}_{R'_2} D R 1 ∖ D R 2 ′ , by (90), (92), and c > 0 c>0 c > 0 , we have h j < R 1 < min { h j ( φ ) , e c h j > R 2 } h_j^{<R_1}<\min\{h_j^{(\varphi)},e^c h_j^{>R_2}\} h j < R 1 < min { h j ( φ ) , e c h j > R 2 } and h j > R 2 < min { h j ( φ ) , e c h j > R 2 } h_j^{>R_2}<\min\{h_j^{(\varphi)},e^c h_j^{>R_2}\} h j > R 2 < min { h j ( φ ) , e c h j > R 2 } . Hence,
h − , j ( R 1 , R 2 ) = max { h j < R 1 , h j > R 2 } < min { h j ( φ ) , e c h j > R 2 } = h + , j ( R 2 ′ , R 2 ) . h^{(R_1,R_2)}_{-,j}=\max\{h_j^{<R_1},h_j^{>R_2}\}<\min\{h_j^{(\varphi)},e^c h_j^{>R_2}\}=h^{(R'_2,R_2)}_{+,j}. h − , j ( R 1 , R 2 ) = max { h j < R 1 , h j > R 2 } < min { h j ( φ ) , e c h j > R 2 } = h + , j ( R 2 ′ , R 2 ) .
Finally, on C ∖ D R 1 \mathbb{C}\setminus\mathbb{D}_{R_1} C ∖ D R 1 , we have h − , j ( R 1 , R 2 ) = h j > R 2 h^{(R_1,R_2)}_{-,j}=h_j^{>R_2} h − , j ( R 1 , R 2 ) = h j > R 2 . Since c > 0 c>0 c > 0 , it follows from (92) that
h j > R 2 < min { h j ( φ ) , e c h j > R 2 } = h + , j ( R 2 ′ , R 2 ) . h_j^{>R_2}<\min\{h_j^{(\varphi)},e^c h_j^{>R_2}\}=h^{(R'_2,R_2)}_{+,j}. h j > R 2 < min { h j ( φ ) , e c h j > R 2 } = h + , j ( R 2 ′ , R 2 ) .
Thus (91) holds on C \mathbb{C} C , which proves assertion (ii).
Lemma 60. The subsolution h − ( R 1 , R 2 ) h^{(R_1,R_2)}_- h − ( R 1 , R 2 ) is complete. For each positive constant c c c , h − ( R 1 , R 2 ) h^{(R_1,R_2)}_- h − ( R 1 , R 2 ) and h + ( R 2 ′ , R 2 ) h^{(R'_2,R_2)}_+ h + ( R 2 ′ , R 2 ) are mutually bounded on C ∖ D R 1 \mathbb{C}\setminus\mathbb{D}_{R_1} C ∖ D R 1 .
Proof. Since h > R 2 h^{>R_2} h > R 2 is complete, h − ( R 1 , R 2 ) h^{(R_1,R_2)}_- h − ( R 1 , R 2 ) is also complete. As in the proof of Lemma 59, for each c > 0 c>0 c > 0 , we have the following estimate on C ∖ D R 1 \mathbb{C}\setminus\mathbb{D}_{R_1} C ∖ D R 1 :
h j > R 2 = h − , j ( R 1 , R 2 ) < h + , j ( R 2 ′ , R 2 ) ≤ e c h j > R 2 for j = 1 , … , n . h_j^{>R_2}=h^{(R_1,R_2)}_{-,j}<h^{(R'_2,R_2)}_{+,j}\le e^c h_j^{>R_2}\quad\text{for }j=1,\ldots,n. h j > R 2 = h − , j ( R 1 , R 2 ) < h + , j ( R 2 ′ , R 2 ) ≤ e c h j > R 2 for j = 1 , … , n .
This proves the latter half of the statement.
Corollary 61. Suppose that
On C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) , φ \varphi φ is continuous and belongs to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 .
Then there exists a real complete solution h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) to equation (5) . Furthermore, the complete solution h h h has the following property concerning its size and rate of divergence: After a translation, we may assume that the closed ball is centered at the origin. Let R 2 , R 2 ′ , R 1 ′ , R 1 , c R_2,R'_2,R'_1,R_1,c R 2 , R 2 ′ , R 1 ′ , R 1 , c be positive real numbers satisfying the following properties:
0 < R 2 < R 2 ′ < R 1 ′ < R 1 0<R_2<R'_2<R'_1<R_1 0 < R 2 < R 2 ′ < R 1 ′ < R 1 .
The following holds on { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ ≤ R 1 ′ } \{z\in\mathbb{C}\mid R'_2\leq|z|\leq R'_1\} { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ ≤ R 1 ′ } :
h j ( φ ) < e c h j > R 2 for j = 1 , … , n . h_j^{(\varphi)}<e^{c}h_j^{>R_2}\quad\text{for }j=1,\ldots,n. h j ( φ ) < e c h j > R 2 for j = 1 , … , n .
The following holds on { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ < R 1 } \{z\in\mathbb{C}\mid R'_2\leq|z|<R_1\} { z ∈ C ∣ R 2 ′ ≤ ∣ z ∣ < R 1 } :
h j < R 1 < e − c h j > R 2 for j = 1 , … , n . h_j^{<R_1}<e^{-c}h_j^{>R_2}\quad\text{for }j=1,\ldots,n. h j < R 1 < e − c h j > R 2 for j = 1 , … , n .
Then we have the following estimate on C ∖ D R 1 \mathbb{C}\setminus D_{R_1} C ∖ D R 1 :
h j > R 2 ≤ h j ≤ e c h j > R 2 for j = 1 , … , n . (93) h_j^{>R_2}\leq h_j\leq e^{c}h_j^{>R_2}\quad\text{for }j=1,\ldots,n. \tag*{(93)} h j > R 2 ≤ h j ≤ e c h j > R 2 for j = 1 , … , n . (93)
Proof. We may assume from the beginning that the center of the closed ball of radius R 1 ′ R'_1 R 1 ′ is the origin. Take positive real numbers R 2 , R 2 ′ , R 1 ′ R_2,R'_2,R'_1 R 2 , R 2 ′ , R 1 ′ so that R 2 < R 2 ′ < R 1 ′ < R 1 R_2<R'_2<R'_1<R_1 R 2 < R 2 ′ < R 1 ′ < R 1 . Take a positive constant c c c large enough so that (89) and (90) hold. Then the supersolution h + ( R 2 ′ , R 2 ) h_+^{(R'_2,R_2)} h + ( R 2 ′ , R 2 ) and a subsolution h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) associated with these real numbers constructed as in Lemmas 58 and 59 satisfy (91), are globally continuous, and belong to W l o c 1 , 2 W_{\mathrm{loc}}^{1,2} W loc 1 , 2 . By the assumption that φ \varphi φ is continuous on C ∖ φ − 1 ( − ∞ ) \mathbb{C}\setminus\varphi^{-1}(-\infty) C ∖ φ − 1 ( − ∞ ) , e φ e^\varphi e φ is continuous. Therefore by Proposition 50 , there exists a real solution h h h to equation (5) satisfying
h − , j ( R 1 , R 2 ) ≤ h j ≤ h + , j ( R 2 ′ , R 2 ) . h_{-,j}^{(R_1,R_2)}\leq h_j\leq h_{+,j}^{(R'_2,R_2)}. h − , j ( R 1 , R 2 ) ≤ h j ≤ h + , j ( R 2 ′ , R 2 ) .
By the construction, h h h is real, and in particular det ( h ) = 1 \det(h)=1 det ( h ) = 1 . By Lemma 60 and the above inequalities, h h h is mutually bounded with the complete metric h − ( R 1 , R 2 ) h_-^{(R_1,R_2)} h − ( R 1 , R 2 ) outside a compact set. Thus h h h is complete. Since h h h is real, det ( h ) = 1 \det(h)=1 det ( h ) = 1 , and Theorem 1 shows that h h h is independent of the choices of R 2 , R 2 ′ , R 1 ′ , c R_2,R'_2,R'_1,c R 2 , R 2 ′ , R 1 ′ , c . Estimate (93) follows from the definitions. □ \square □
Let ( φ ϵ ) ϵ > 0 (\varphi_\epsilon)_{\epsilon>0} ( φ ϵ ) ϵ > 0 be a smooth approximation of the weight function φ \varphi φ satisfying the condition in Theorem 3 . From Corollary 61 and Theorem 1 , for each ϵ > 0 \epsilon>0 ϵ > 0 , there exists a unique real smooth complete solution h ϵ = ( h ϵ , 1 , … , h ϵ , r ) h_\epsilon=(h_{\epsilon,1},\ldots,h_{\epsilon,r}) h ϵ = ( h ϵ , 1 , … , h ϵ , r ) associated with φ ϵ \varphi_\epsilon φ ϵ . We prove the corresponding family of real complete solutions ( h ϵ ) ϵ > 0 (h_\epsilon)_{\epsilon>0} ( h ϵ ) ϵ > 0 converges to a real complete solution h h h associated with φ \varphi φ . For each ϵ > 0 \epsilon>0 ϵ > 0 and each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , we set H ϵ , j : = h ϵ , j − 1 ⊗ h ϵ , j + 1 H_{\epsilon,j}:=h_{\epsilon,j}^{-1}\otimes h_{\epsilon,j+1} H ϵ , j := h ϵ , j − 1 ⊗ h ϵ , j + 1 , and we also define H ϵ , r : = ( e − φ ϵ h ∗ ) − r ⊗ H ϵ , 1 − 1 ⊗ ⋯ ⊗ H ϵ , r − 1 − 1 H_{\epsilon,r}:=(e^{-\varphi_\epsilon}h_*)^{-r}\otimes H_{\epsilon,1}^{-1}\otimes\cdots\otimes H_{\epsilon,r-1}^{-1} H ϵ , r := ( e − φ ϵ h ∗ ) − r ⊗ H ϵ , 1 − 1 ⊗ ⋯ ⊗ H ϵ , r − 1 − 1 . We also take a positive real number R 2 R_2 R 2 . We denote by h ϵ > R 2 = ( h ϵ , 1 > R 2 , … , h ϵ , r > R 2 ) h_\epsilon^{>R_2}=(h_{\epsilon,1}^{>R_2},\ldots,h_{\epsilon,r}^{>R_2}) h ϵ > R 2 = ( h ϵ , 1 > R 2 , … , h ϵ , r > R 2 ) the unique real complete solution associated with φ ϵ ∣ C ∖ D R 2 \varphi_\epsilon|_{\mathbb{C}\setminus D_{R_2}} φ ϵ ∣ C ∖ D R 2 . We also set H ϵ , j > R 2 : = ( h ϵ , j > R 2 ) − 1 ⊗ h ϵ , j + 1 > R 2 H_{\epsilon,j}^{>R_2}:=(h_{\epsilon,j}^{>R_2})^{-1}\otimes h_{\epsilon,j+1}^{>R_2} H ϵ , j > R 2 := ( h ϵ , j > R 2 ) − 1 ⊗ h ϵ , j + 1 > R 2 for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , and set H ϵ , r > R 2 : = ( e − φ ϵ h ∗ ∣ C ∖ D R 2 ) − r ⊗ H ϵ , 1 > R 2 , − 1 ⊗ ⋯ ⊗ H ϵ , r − 1 > R 2 , − 1 H_{\epsilon,r}^{>R_2}:=(e^{-\varphi_\epsilon}h_*|_{\mathbb{C}\setminus D_{R_2}})^{-r}\otimes H_{\epsilon,1}^{>R_2,-1}\otimes\cdots\otimes H_{\epsilon,r-1}^{>R_2,-1} H ϵ , r > R 2 := ( e − φ ϵ h ∗ ∣ C ∖ D R 2 ) − r ⊗ H ϵ , 1 > R 2 , − 1 ⊗ ⋯ ⊗ H ϵ , r − 1 > R 2 , − 1 . We divide the proof of Theorems 2 and 3 for the complex plane case into the following five lemmas.
Lemma 62. For each 0 < ϵ < ϵ ′ 0 < \epsilon< \epsilon^{\prime} 0 < ϵ < ϵ ′ and j = 1 , … , n j = 1,\ldots,n j = 1 , … , n , the following holds:
H ϵ , j ⊗ H ϵ ′ , j − 1 H_{\epsilon,j} \otimes H_{\epsilon^{\prime},j}^{-1} H ϵ , j ⊗ H ϵ ′ , j − 1 and H ϵ , j > R 2 ⊗ H ϵ ′ , j > R 2 − 1 H_{\epsilon,j}^{>R_2} \otimes H_{\epsilon^{\prime},j}^{>R_2-1} H ϵ , j > R 2 ⊗ H ϵ ′ , j > R 2 − 1 are bounded above.
It holds that
h ϵ ′ , j ≤ h ϵ , j and h ϵ ′ , j > R 2 ≤ h ϵ , j > R 2 . h_{\epsilon^{\prime},j} \le h_{\epsilon,j} \quad\text{and}\quad h_{\epsilon^{\prime},j}^{>R_2} \le h_{\epsilon,j}^{>R_2}. h ϵ ′ , j ≤ h ϵ , j and h ϵ ′ , j > R 2 ≤ h ϵ , j > R 2 . Proof. Fix j = 1 , … , n j = 1,\ldots,n j = 1 , … , n . We first prove assertion 1. Since the proofs are the same, we only prove the assertion for H ϵ , j H_{\epsilon,j} H ϵ , j and H ϵ ′ , j H_{\epsilon^{\prime},j} H ϵ ′ , j for simplicity. Because e φ ϵ ′ h ∗ − 1 ⊗ H ϵ ′ , j − 1 e^{\varphi_{\epsilon^{\prime}}}h_{*}^{-1} \otimes H_{\epsilon^{\prime},j}^{-1} e φ ϵ ′ h ∗ − 1 ⊗ H ϵ ′ , j − 1 is bounded, e φ ϵ h ∗ − 1 ⊗ H ϵ ′ , j − 1 e^{\varphi_{\epsilon}}h_{*}^{-1} \otimes H_{\epsilon^{\prime},j}^{-1} e φ ϵ h ∗ − 1 ⊗ H ϵ ′ , j − 1 is also bounded. Then by Corollary 35, we observe that H ϵ , j ⊗ H ϵ ′ , j − 1 H_{\epsilon,j} \otimes H_{\epsilon^{\prime},j}^{-1} H ϵ , j ⊗ H ϵ ′ , j − 1 is bounded. We next prove assertion 2. Since the proofs are the same, we only prove the assertion for h ϵ , j h_{\epsilon,j} h ϵ , j and h ϵ ′ , j h_{\epsilon^{\prime},j} h ϵ ′ , j for simplicity. The following holds:
( h ϵ ′ , j ⊗ h ϵ , j − 1 ) 2 = ∏ k = j r − j ( H ϵ ′ , k − 1 ⊗ H ϵ , k ) for each j = 1 , … , n . \left(h_{\epsilon^{\prime},j} \otimes h_{\epsilon,j}^{-1}\right)^2 = \prod_{k=j}^{r-j}\left(H_{\epsilon^{\prime},k}^{-1} \otimes H_{\epsilon,k}\right) \quad\text{for each }j = 1,\ldots,n. ( h ϵ ′ , j ⊗ h ϵ , j − 1 ) 2 = k = j ∏ r − j ( H ϵ ′ , k − 1 ⊗ H ϵ , k ) for each j = 1 , … , n . Therefore, by assertion 1, h ϵ ′ , j ⊗ h ϵ , j − 1 h_{\epsilon^{\prime},j} \otimes h_{\epsilon,j}^{-1} h ϵ ′ , j ⊗ h ϵ , j − 1 is bounded, and thus by Lemma 49, we have the claim.
Lemma 63. The following holds for all ϵ > 0 \epsilon> 0 ϵ > 0 and j = 1 , … , n j = 1,\ldots,n j = 1 , … , n :
h ϵ , j ≤ h j ( φ ) = ( e − φ h ∗ ) r − ( 2 j − 1 ) 2 . h_{\epsilon,j} \le h_j^{(\varphi)} = \left(e^{-\varphi}h_*\right)^{\frac{r-(2j-1)}{2}}. h ϵ , j ≤ h j ( φ ) = ( e − φ h ∗ ) 2 r − ( 2 j − 1 ) . Proof. From Lemma 32, [50 Proposition 27] , and the inequality φ ≤ φ ϵ \varphi\le\varphi_{\epsilon} φ ≤ φ ϵ , we obtain
h ϵ , j ≤ h j ( φ ϵ ) ≤ h j ( φ ) , h_{\epsilon,j} \le h_j^{(\varphi_{\epsilon})} \le h_j^{(\varphi)}, h ϵ , j ≤ h j ( φ ϵ ) ≤ h j ( φ ) , and thus we have the claim.
Lemma 64. The following holds:
There exists a positive constant C ( R 2 ) C(R_2) C ( R 2 ) such that on C ∖ D ‾ R 2 \mathbb{C} \setminus\overline{D}_{R_2} C ∖ D R 2 , the following holds for all ϵ > 0 \epsilon> 0 ϵ > 0 and j = 1 , … , n j = 1,\ldots,n j = 1 , … , n :
h ϵ , j − 1 ⊗ h ϵ , j > R 2 ≤ C ( R 2 ) . (94) h_{\epsilon,j}^{-1} \otimes h_{\epsilon,j}^{>R_2} \le C(R_2). \tag*{(94)} h ϵ , j − 1 ⊗ h ϵ , j > R 2 ≤ C ( R 2 ) . (94) For each real number R 1 R_1 R 1 which is greater than R 2 R_2 R 2 and each ϵ > 0 \epsilon> 0 ϵ > 0 , there exists a positive constant C ( ϵ , R 1 , R 2 ) C(\epsilon,R_1,R_2) C ( ϵ , R 1 , R 2 ) such that on C ∖ D R 1 \mathbb{C} \setminus D_{R_1} C ∖ D R 1 , the following holds for all j = 1 , … , n j = 1,\ldots,n j = 1 , … , n :
h ϵ , j ⊗ h ϵ , j > R 2 − 1 ≤ C ( ϵ , R 1 , R 2 ) . (95) h_{\epsilon,j} \otimes h_{\epsilon,j}^{>R_2-1} \le C(\epsilon,R_1,R_2). \tag*{(95)} h ϵ , j ⊗ h ϵ , j > R 2 − 1 ≤ C ( ϵ , R 1 , R 2 ) . (95) Proof. We first prove assertion 1. By Corollary 37, there exists a constant C C C independent of ϵ \epsilon ϵ such that
e φ ϵ h ∗ − 1 ⊗ H ϵ , j > R 2 − 1 ≤ C for each j = 1 , … , n . e^{\varphi_{\epsilon}}h_*^{-1} \otimes H_{\epsilon,j}^{>R_2-1} \le C \quad\text{for each }j = 1,\ldots,n. e φ ϵ h ∗ − 1 ⊗ H ϵ , j > R 2 − 1 ≤ C for each j = 1 , … , n . Applying Lemma 34 to H = C H ϵ , j > R 2 H = C H_{\epsilon,j}^{>R_2} H = C H ϵ , j > R 2 , we have
− 1 Λ H ϵ , j > R 2 ∂ ˉ ∂ log ( ∑ k = 1 r vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) ≥ C 1 ( ∑ k = 1 r vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) − C C 2 − 2. (96) \sqrt{-1}\Lambda_{H_{\epsilon,j}^{>R_2}}\bar{\partial}\partial\log\left(\sum_{k=1}^{r}\operatorname{vol}(H_{\epsilon,k})/\operatorname{vol}(H_{\epsilon,j}^{>R_2})\right) \ge C_1\left(\sum_{k=1}^{r}\operatorname{vol}(H_{\epsilon,k})/\operatorname{vol}(H_{\epsilon,j}^{>R_2})\right) - CC_2 - 2. \tag*{(96)} − 1 Λ H ϵ , j > R 2 ∂ ˉ ∂ log ( k = 1 ∑ r vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) ≥ C 1 ( k = 1 ∑ r vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) − C C 2 − 2. (96)
By applying Proposition 24 to inequality (96) , we obtain
∑ k = 1 r ( vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) ≤ C C 2 + 2 C 1 . \sum_{k=1}^{r}\left(\operatorname{vol}(H_{\epsilon,k})/\operatorname{vol}(H_{\epsilon,j}^{>R_2})\right)\leq\frac{CC_2+2}{C_1}. k = 1 ∑ r ( vol ( H ϵ , k ) / vol ( H ϵ , j > R 2 ) ) ≤ C 1 C C 2 + 2 .
In particular, it follows that H ϵ , k ⊗ H ϵ , j > R 2 − 1 H_{\epsilon,k}\otimes H_{\epsilon,j}^{>R_2-1} H ϵ , k ⊗ H ϵ , j > R 2 − 1 is bounded above by a constant independent of ϵ \epsilon ϵ , j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , and k = 1 , … , r k=1,\ldots,r k = 1 , … , r . As a consequence, it follows that
( h ϵ , j > R 2 ⊗ h ϵ , j − 1 ) 2 = ∏ k = j r − j ( H ϵ , k > R 2 − 1 ⊗ H ϵ , k ) \left(h_{\epsilon,j}^{>R_2}\otimes h_{\epsilon,j}^{-1}\right)^2=\prod_{k=j}^{r-j}\left(H_{\epsilon,k}^{>R_2-1}\otimes H_{\epsilon,k}\right) ( h ϵ , j > R 2 ⊗ h ϵ , j − 1 ) 2 = k = j ∏ r − j ( H ϵ , k > R 2 − 1 ⊗ H ϵ , k )
is bounded above by a constant independent of ϵ \epsilon ϵ and j j j . This establishes inequality (94) . We next provide a proof of assertion 2, which is similar to the proof of assertion 1. For each ϵ > 0 \epsilon>0 ϵ > 0 and j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 , let H ϵ , j ( R 1 , R 2 ) H_{\epsilon,j}^{(R_1,R_2)} H ϵ , j ( R 1 , R 2 ) be a smooth complete Hermitian metric on the anticanonical bundle of C ∖ D ‾ R 2 \mathbb{C}\setminus\overline{D}_{R_2} C ∖ D R 2 satisfying the following properties:
On C ∖ D ‾ R 1 \mathbb{C}\setminus\overline{D}_{R_1} C ∖ D R 1 , H ϵ , j ( R 1 , R 2 ) H_{\epsilon,j}^{(R_1,R_2)} H ϵ , j ( R 1 , R 2 ) coincides with H ϵ , j H_{\epsilon,j} H ϵ , j .
On D ‾ ( R 1 + R 2 ) / 2 ∖ D ‾ R 2 \overline{D}_{(R_1+R_2)/2}\setminus\overline{D}_{R_2} D ( R 1 + R 2 ) /2 ∖ D R 2 , H ϵ , j ( R 1 , R 2 ) H_{\epsilon,j}^{(R_1,R_2)} H ϵ , j ( R 1 , R 2 ) coincides with the hyperbolic metric on C ∖ D ‾ R 2 \mathbb{C}\setminus\overline{D}_{R_2} C ∖ D R 2 .
Then by definition, we can easily confirm that e φ ϵ h ∗ − 1 ∣ C ∖ D ‾ R 2 ⊗ H ϵ , j ( R 1 , R 2 ) − 1 e^{\varphi_\epsilon}h_*^{-1}|_{\mathbb{C}\setminus\overline{D}_{R_2}}\otimes H_{\epsilon,j}^{(R_1,R_2)-1} e φ ϵ h ∗ − 1 ∣ C ∖ D R 2 ⊗ H ϵ , j ( R 1 , R 2 ) − 1 is bounded. Since the Gaussian curvature of H ϵ , j ( R 1 , R 2 ) H_{\epsilon,j}^{(R_1,R_2)} H ϵ , j ( R 1 , R 2 ) is bounded below, by Lemma 34 and Corollary 35, H ϵ , j > R 2 ⊗ H ϵ , j ( R 1 , R 2 ) − 1 H_{\epsilon,j}^{>R_2}\otimes H_{\epsilon,j}^{(R_1,R_2)-1} H ϵ , j > R 2 ⊗ H ϵ , j ( R 1 , R 2 ) − 1 is bounded above by a constant that may depend on ϵ \epsilon ϵ . As a consequence, since H ϵ , j ( R 1 , R 2 ) H_{\epsilon,j}^{(R_1,R_2)} H ϵ , j ( R 1 , R 2 ) coincides with H ϵ , j H_{\epsilon,j} H ϵ , j on C ∖ D ‾ R 1 \mathbb{C}\setminus\overline{D}_{R_1} C ∖ D R 1 , for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , it follows that
( h ϵ , j > R 2 − 1 ⊗ h ϵ , j ) 2 = ∏ k = j r − j ( H ϵ , k > R 2 ⊗ H ϵ , k − 1 ) \left(h_{\epsilon,j}^{>R_2-1}\otimes h_{\epsilon,j}\right)^2=\prod_{k=j}^{r-j}\left(H_{\epsilon,k}^{>R_2}\otimes H_{\epsilon,k}^{-1}\right) ( h ϵ , j > R 2 − 1 ⊗ h ϵ , j ) 2 = k = j ∏ r − j ( H ϵ , k > R 2 ⊗ H ϵ , k − 1 )
is bounded above on C ∖ D ‾ R 1 \mathbb{C}\setminus\overline{D}_{R_1} C ∖ D R 1 . This completes the proof. □
Lemma 65. For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , define h j h_j h j and h ~ j > R 2 \widetilde{h}_j^{>R_2} h j > R 2 pointwise by
h j ( x ) : = sup ϵ > 0 h ϵ , j ( x ) for x ∈ C , h_j(x):=\sup_{\epsilon>0}h_{\epsilon,j}(x)\quad\text{for }x\in\mathbb{C}, h j ( x ) := ϵ > 0 sup h ϵ , j ( x ) for x ∈ C , h ~ j > R 2 ( x ) : = sup ϵ > 0 h ϵ , j > R 2 ( x ) for x ∈ C ∖ D ‾ R 2 . \widetilde{h}_j^{>R_2}(x):=\sup_{\epsilon>0}h_{\epsilon,j}^{>R_2}(x)\quad\text{for }x\in\mathbb{C}\setminus\overline{D}_{R_2}. h j > R 2 ( x ) := ϵ > 0 sup h ϵ , j > R 2 ( x ) for x ∈ C ∖ D R 2 . For the remaining indices j = n + 1 , … , r j=n+1,\ldots,r j = n + 1 , … , r , we define h j h_j h j and h ~ j > R 2 \widetilde{h}_j^{>R_2} h j > R 2 so that h = ( h 1 , … , h r ) h=(h_1,\ldots,h_r) h = ( h 1 , … , h r ) and h ~ > R 2 = ( h ~ 1 > R 2 , … , h ~ r > R 2 ) \widetilde{h}^{>R_2}=(\widetilde{h}_1^{>R_2},\ldots,\widetilde{h}_r^{>R_2}) h > R 2 = ( h 1 > R 2 , … , h r > R 2 ) are real. Then the following holds:
For each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , h j h_j h j is of class C l o c 2 j − 1 , 1 C_{\mathrm{loc}}^{2j-1,1} C loc 2 j − 1 , 1 . h h h solves equation (5) in the sense of distributions.
h ~ > R 2 = h > R 2 \widetilde{h}^{>R_2}=h^{>R_2} h > R 2 = h > R 2 .
Proof. We first prove assertion 1. We set H j : = h j − 1 ⊗ h j + 1 H_j:=h_j^{-1}\otimes h_{j+1} H j := h j − 1 ⊗ h j + 1 for each j = 1 , … , r − 1 j=1,\ldots,r-1 j = 1 , … , r − 1 and H r : = ( e − φ h ∗ ) − r ⊗ H 1 − 1 ⊗ ⋯ ⊗ H r − 1 − 1 H_r:=(e^{-\varphi}h_*)^{-r}\otimes H_1^{-1}\otimes\cdots\otimes H_{r-1}^{-1} H r := ( e − φ h ∗ ) − r ⊗ H 1 − 1 ⊗ ⋯ ⊗ H r − 1 − 1 . Then by Lemma 63 , h h h is finite outside the polar set of φ \varphi φ . Take a smooth reference metric h ∗ h_* h ∗ on K C K_{\mathbb{C}} K C and define weight functions φ ϵ , j \varphi_{\epsilon,j} φ ϵ , j (ϵ > 0 \epsilon>0 ϵ > 0 ) and φ j \varphi_j φ j by
h j = e − φ j h ∗ r − ( 2 j − 1 ) 2 and h ϵ , j = e − φ ϵ , j h ∗ r − ( 2 j − 1 ) 2 . (97) h_j=e^{-\varphi_j}h_*^{\frac{r-(2j-1)}{2}}\quad\text{and}\quad h_{\epsilon,j}=e^{-\varphi_{\epsilon,j}}h_*^{\frac{r-(2j-1)}{2}}. \tag*{(97)} h j = e − φ j h ∗ 2 r − ( 2 j − 1 ) and h ϵ , j = e − φ ϵ , j h ∗ 2 r − ( 2 j − 1 ) . (97)
for j = 1 , … , n j=1,\ldots,n j = 1 , … , n . By Lemma 63 and the monotonicity of h ϵ , j h_{\epsilon,j} h ϵ , j , for every fixed ϵ 0 > 0 \epsilon_0>0 ϵ 0 > 0 and 0 < ϵ < ϵ 0 0<\epsilon<\epsilon_0 0 < ϵ < ϵ 0 , we have
r − ( 2 j − 1 ) 2 φ ≤ ϕ ϵ , j ≤ ϕ ϵ 0 , j . (98) \frac{r-(2j-1)}{2}\varphi\leq\phi_{\epsilon,j}\leq\phi_{\epsilon_0,j}. \tag*{(98)} 2 r − ( 2 j − 1 ) φ ≤ ϕ ϵ , j ≤ ϕ ϵ 0 , j . (98)
Since both r − ( 2 j − 1 ) 2 φ \frac{r-(2j-1)}{2}\varphi 2 r − ( 2 j − 1 ) φ and ϕ ϵ 0 , j \phi_{\epsilon_0,j} ϕ ϵ 0 , j belong to L l o c 1 L^1_{\mathrm{loc}} L loc 1 , the dominated convergence theorem implies that ϕ ϵ , j \phi_{\epsilon,j} ϕ ϵ , j converges to ϕ j \phi_j ϕ j in L l o c 1 L^1_{\mathrm{loc}} L loc 1 for j = 1 , … , n j=1,\ldots,n j = 1 , … , n . Therefore, the curvature F h ϵ , j F_{h_{\epsilon,j}} F h ϵ , j converges to F h j F_{h_j} F h j in the sense of distributions. On the other hand, for each fixed ϵ 0 > 0 \epsilon_0>0 ϵ 0 > 0 and each 0 < ϵ < ϵ 0 0<\epsilon<\epsilon_0 0 < ϵ < ϵ 0 , it holds that
0 < H ϵ , n = h ϵ , n − ( 2 n + 2 − r ) ≤ h ϵ 0 , n − ( 2 n + 2 − r ) . 0<H_{\epsilon,n}=h_{\epsilon,n}^{-(2n+2-r)}\leq h_{\epsilon_0,n}^{-(2n+2-r)}. 0 < H ϵ , n = h ϵ , n − ( 2 n + 2 − r ) ≤ h ϵ 0 , n − ( 2 n + 2 − r ) .
Also, by [50 Proposition 27] , we obtain
0 ≤ H ϵ , j ≤ H ϵ , n ≤ h ϵ 0 , n − ( 2 n + 2 − r ) 0\leq H_{\epsilon,j}\leq H_{\epsilon,n}\leq h_{\epsilon_0,n}^{-(2n+2-r)} 0 ≤ H ϵ , j ≤ H ϵ , n ≤ h ϵ 0 , n − ( 2 n + 2 − r )
for each j = 0 , 1 , … , n − 1 j=0,1,\ldots,n-1 j = 0 , 1 , … , n − 1 and 0 < ϵ < ϵ 0 0<\epsilon<\epsilon_0 0 < ϵ < ϵ 0 . Therefore, H 1 , … , H r H_1,\ldots,H_r H 1 , … , H r belong to L l o c ∞ ( C ) L^\infty_{\mathrm{loc}}(\mathbb{C}) L loc ∞ ( C ) and by the Lebesgue dominated convergence theorem, for each j = 1 , … , r j=1,\ldots,r j = 1 , … , r , as ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 , H ϵ , j H_{\epsilon,j} H ϵ , j converges to H j H_j H j in the L l o c 1 L^1_{\mathrm{loc}} L loc 1 sense. Therefore, by taking the limit ϵ ↘ 0 \epsilon\searrow0 ϵ ↘ 0 in the equation
− 1 F h ϵ , j + vol ( H ϵ , j − 1 ) − vol ( H ϵ , j ) = 0 for j = 1 , … , n , \sqrt{-1}F_{h_{\epsilon,j}}+\operatorname{vol}(H_{\epsilon,j-1})-\operatorname{vol}(H_{\epsilon,j})=0\quad\text{for }j=1,\ldots,n, − 1 F h ϵ , j + vol ( H ϵ , j − 1 ) − vol ( H ϵ , j ) = 0 for j = 1 , … , n ,
we conclude that
− 1 F h j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , n , \sqrt{-1}F_{h_j}+\operatorname{vol}(H_{j-1})-\operatorname{vol}(H_j)=0\quad\text{for }j=1,\ldots,n, − 1 F h j + vol ( H j − 1 ) − vol ( H j ) = 0 for j = 1 , … , n ,
in the sense of distributions. By reality, the remaining equations follow as well. Therefore, h h h solves equation (5) in the sense of distributions. By Proposition 69, for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , h j h_j h j agrees almost everywhere with a C l o c 2 j − 1 , 1 C^{2j-1,1}_{\mathrm{loc}} C loc 2 j − 1 , 1 Hermitian metric. We claim that the equality holds everywhere. Indeed, since every subharmonic function that is not identically − ∞ -\infty − ∞ belongs to L l o c p L^p_{\mathrm{loc}} L loc p for every 1 ≤ p < ∞ 1\leq p<\infty 1 ≤ p < ∞ , by (97), for a fixed ϵ 0 > 0 \epsilon_0>0 ϵ 0 > 0 , the L l o c p L^p_{\mathrm{loc}} L loc p -norm of ( ϕ ϵ , j ) 0 < ϵ < ϵ 0 (\phi_{\epsilon,j})_{0<\epsilon<\epsilon_0} ( ϕ ϵ , j ) 0 < ϵ < ϵ 0 is uniformly bounded, and thus, the standard interior L p L^p L p -estimate together with the uniform L p L^p L p -bounds of the curvatures − 1 F h ϵ , j \sqrt{-1}F_{h_{\epsilon,j}} − 1 F h ϵ , j gives a uniform bound on the W l o c 2 , p W^{2,p}_{\mathrm{loc}} W loc 2 , p -norms of ( ϕ ϵ , j ) 0 < ϵ < ϵ 0 (\phi_{\epsilon,j})_{0<\epsilon<\epsilon_0} ( ϕ ϵ , j ) 0 < ϵ < ϵ 0 . Taking p = 2 p=2 p = 2 , compact Sobolev embedding and a diagonal argument show that every sequence ϵ i → 0 \epsilon_i\to0 ϵ i → 0 admits a subsequence along which the local weights ϕ ϵ i , j \phi_{\epsilon_i,j} ϕ ϵ i , j , and hence the metrics h ϵ i , j h_{\epsilon_i,j} h ϵ i , j , converge locally uniformly. By uniqueness of the L l o c 1 L^1_{\mathrm{loc}} L loc 1 -limit of the weight function, the limiting metric is the C l o c 1 , 1 C^{1,1}_{\mathrm{loc}} C loc 1 , 1 -representative of h j h_j h j . Therefore h ϵ , j h_{\epsilon,j} h ϵ , j converges to this representative in C l o c 0 C^0_{\mathrm{loc}} C loc 0 , and the pointwise limit h j h_j h j agrees with it everywhere. This completes the proof of assertion 1.
We next prove assertion 2. The proof of assertion 2 is essentially the same as that of assertion 1, so we only indicate the necessary modifications. By the same arguments as in Lemmas 62 and 63, together with [50 Proposition 27] , we can verify that h ~ > R 2 \widetilde{h}^{>R_2} h > R 2 solves equation (5) in the sense of distributions. By Proposition 69, for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n , h ~ j > R 2 \widetilde{h}^{>R_2}_j h j > R 2 agrees almost everywhere with a C l o c 2 j − 1 , 1 C^{2j-1,1}_{\mathrm{loc}} C loc 2 j − 1 , 1 Hermitian metric. By repeating the argument above, we can verify that h ϵ > R 2 h^{>R_2}_{\epsilon} h ϵ > R 2 converges to h ~ > R 2 \widetilde{h}^{>R_2} h > R 2 in C l o c 0 C^0_{\mathrm{loc}} C loc 0 . Hence, the pointwise limit h ~ j > R 2 \widetilde{h}^{>R_2}_j h j > R 2 agrees everywhere with this representative. We set
H ~ j > R 2 : = h ~ j − 1 > R 2 ⊗ h ~ j > R 2 for j = 1 , … , r − 1. \widetilde{H}^{>R_2}_j:=\widetilde{h}^{>R_2}_{j-1}\otimes\widetilde{h}^{>R_2}_j\quad\text{for }j=1,\ldots,r-1. H j > R 2 := h j − 1 > R 2 ⊗ h j > R 2 for j = 1 , … , r − 1.
By Lemma 55, H ϵ , j > R 2 H^{>R_2}_{\epsilon,j} H ϵ , j > R 2 is bounded below by the hyperbolic metric with a factor of 1 / 2 1/2 1/2 . Therefore, H ~ j > R 2 \widetilde{H}^{>R_2}_j H j > R 2 is bounded below by the hyperbolic metric with a factor of 1 / 2 1/2 1/2 and is, in particular, complete. Finally, by reality, det ( h ~ > R 2 ) = 1 \det(\widetilde{h}^{>R_2})=1 det ( h > R 2 ) = 1 . Therefore, by the uniqueness of the complete solution, we obtain h ~ > R 2 = h > R 2 \widetilde{h}^{>R_2}=h^{>R_2} h > R 2 = h > R 2 .
Lemma 66. The Hermitian metric h h h in Lemma 65 is mutually bounded with h > R 2 h^{>R_2} h > R 2 on C ∖ D R 1 \mathbb{C} \setminus\mathbb{D}_{R_1} C ∖ D R 1 , and in particular, it is complete.
Proof. By Lemma 65 , as ε ↘ 0 \varepsilon\searrow0 ε ↘ 0 , h ε , j h_{\varepsilon,j} h ε , j and h ε , j > R 2 h_{\varepsilon,j}^{>R_2} h ε , j > R 2 converge to h j h_j h j and h j > R 2 h_j^{>R_2} h j > R 2 , respectively. Therefore, by taking the limit ε ↘ 0 \varepsilon\searrow0 ε ↘ 0 in inequality (94) in Lemma 64 , we obtain
h j − 1 ⊗ h j > R 2 ≤ C ( R 2 ) for j = 1 , … , n . h_j^{-1} \otimes h_j^{>R_2} \leq C(R_2) \text{ for } j=1,\ldots,n. h j − 1 ⊗ h j > R 2 ≤ C ( R 2 ) for j = 1 , … , n .
We prove the converse inequality. Fix ε 0 > 0 \varepsilon_0>0 ε 0 > 0 . Take a constant c c c so that
c > max { max 1 ≤ j ≤ n sup z ∈ ∂ D R 1 log ( h j ( z ) ⊗ h ε 0 , j > R 2 − 1 ( z ) ) , 0 } . c > \max\left\{\max_{1\leq j\leq n}\sup_{z\in\partial\mathbb{D}_{R_1}}\log\left(h_j(z)\otimes h_{\varepsilon_0,j}^{>R_2-1}(z)\right),0\right\}. c > max { 1 ≤ j ≤ n max z ∈ ∂ D R 1 sup log ( h j ( z ) ⊗ h ε 0 , j > R 2 − 1 ( z ) ) , 0 } .
where ∂ D R 1 \partial\mathbb{D}_{R_1} ∂ D R 1 denotes the boundary { z ∈ C ∣ ∣ z ∣ = R 1 } \{z\in\mathbb{C}\mid|z|=R_1\} { z ∈ C ∣ ∣ z ∣ = R 1 } of D R 1 \mathbb{D}_{R_1} D R 1 . By inequality (95) in Lemma 64 , h ε , j ⊗ h ε , j > R 2 − 1 h_{\varepsilon,j}\otimes h_{\varepsilon,j}^{>R_2-1} h ε , j ⊗ h ε , j > R 2 − 1 is bounded above. Then by Lemma 65 , we have the following for each j = 1 , … , n j=1,\ldots,n j = 1 , … , n and 0 < ε < ε 0 0<\varepsilon<\varepsilon_0 0 < ε < ε 0 :
( h ε , j ⊗ h ε , j > R 2 − 1 ) ( z ) ≤ e c for all z ∈ ∂ D R 1 . \left(h_{\varepsilon,j}\otimes h_{\varepsilon,j}^{>R_2-1}\right)(z)\leq e^c \text{ for all } z\in\partial\mathbb{D}_{R_1}. ( h ε , j ⊗ h ε , j > R 2 − 1 ) ( z ) ≤ e c for all z ∈ ∂ D R 1 .
Then by Example 44 and Lemma 49 , we obtain
e − c h ε , j ⊗ ( h ε , j > R 2 − 1 ) ≤ 1. e^{-c}h_{\varepsilon,j}\otimes\left(h_{\varepsilon,j}^{>R_2-1}\right)\leq1. e − c h ε , j ⊗ ( h ε , j > R 2 − 1 ) ≤ 1.
Since the constant c c c does not depend on 0 < ε < ε 0 0<\varepsilon<\varepsilon_0 0 < ε < ε 0 , taking the limit ε ↘ 0 \varepsilon\searrow0 ε ↘ 0 , we obtain
h j ⊗ h j > R 2 − 1 ≤ e c . h_j\otimes h_j^{>R_2-1}\leq e^c. h j ⊗ h j > R 2 − 1 ≤ e c .
This implies the claim. □
As a corollary, we obtain
Corollary 67. On the complex plane, for any subharmonic weight function φ \varphi φ that is not identically − ∞ -\infty − ∞ , there exists a unique real complete solution h h h associated with φ \varphi φ . Moreover, for each smooth approximation ( φ ε ) 0 < ε < 1 (\varphi_\varepsilon)_{0<\varepsilon<1} ( φ ε ) 0 < ε < 1 of φ \varphi φ satisfying the condition in Theorem 3 , the corresponding family ( h ε ) 0 < ε < 1 (h_\varepsilon)_{0<\varepsilon<1} ( h ε ) 0 < ε < 1 of real complete solutions monotonically converges to the complete solution h h h . Furthermore, the complete solution h h h has the property in Theorem 2 .
Proof of Theorem 2 for the general case
Theorem 68. For any Riemann surface X X X and any subharmonic weight function φ \varphi φ on K X K_X K X which is not identically − ∞ -\infty − ∞ unless X X X is hyperbolic, there exists a real complete solution associated with φ \varphi φ .
Proof. The argument follows [43 Proof of Propositions 5.7 and 5.10] . Let π : X ~ → X \pi:\widetilde{X}\to X π : X → X be the universal covering map. Let φ \varphi φ be a subharmonic weight function which is not identically − ∞ -\infty − ∞ unless X X X is hyperbolic. The case X = C P 1 X=\mathbb{CP}^1 X = CP 1 is excluded by the assumption on φ \varphi φ . Thus, by uniformization, X ~ \widetilde{X} X is biholomorphic to either D \mathbb{D} D or C \mathbb{C} C . We denote by π ∗ φ \pi^*\varphi π ∗ φ the pullback of this weight function by π \pi π . As in § 4.5 , by setting K X ~ 1 / 2 : = π ∗ K X 1 / 2 K_{\widetilde{X}}^{1/2}:=\pi^*K_X^{1/2} K X 1/2 := π ∗ K X 1/2 , there is a natural isomorphism
π ∗ ( K X , r ) ≃ K X ~ , r . \pi^*(K_{X,r})\simeq K_{\widetilde{X},r}. π ∗ ( K X , r ) ≃ K X , r .
where we have used the notation K X , r : = ⨁ j = 1 r K X r − ( 2 j − 1 ) 2 \mathcal{K}_{X,r}:=\bigoplus_{j=1}^{r}K_{X}^{\frac{r-(2j-1)}{2}} K X , r := ⨁ j = 1 r K X 2 r − ( 2 j − 1 ) and K ~ X , r : = ⨁ j = 1 r K X ~ r − ( 2 j − 1 ) 2 \widetilde{\mathcal{K}}_{X,r}:=\bigoplus_{j=1}^{r}K_{\widetilde{X}}^{\frac{r-(2j-1)}{2}} K X , r := ⨁ j = 1 r K X 2 r − ( 2 j − 1 ) to explicitly denote the underlying Riemann surfaces. Then by Theorem 1 and Corollaries 57 and 67, there exists a unique real complete solution h ~ \widetilde{h} h on K ~ X , r \widetilde{\mathcal{K}}_{X,r} K X , r associated with π ∗ φ \pi^{*}\varphi π ∗ φ . For each γ ∈ π 1 ( X ) \gamma\in\pi_{1}(X) γ ∈ π 1 ( X ) , we contend that h ~ \widetilde{h} h is γ \gamma γ -invariant. Indeed, since π ∗ φ \pi^{*}\varphi π ∗ φ is γ \gamma γ -invariant, by Proposition 39, γ ∗ h ~ \gamma^{*}\widetilde{h} γ ∗ h also solves equation (5) associated to π ∗ φ \pi^{*}\varphi π ∗ φ , where γ ∗ K ~ X , r \gamma^{*}\widetilde{\mathcal{K}}_{X,r} γ ∗ K X , r and K ~ X , r \widetilde{\mathcal{K}}_{X,r} K X , r are naturally identified. By Theorem 1, we have h ~ = γ ∗ h ~ \widetilde{h}=\gamma^{*}\widetilde{h} h = γ ∗ h . Therefore, h ~ \widetilde{h} h is π 1 ( X ) \pi_{1}(X) π 1 ( X ) -invariant, and thus, there exists a unique Hermitian metric h h h on K X , r \mathcal{K}_{X,r} K X , r such that h ~ = π ∗ h \widetilde{h}=\pi^{*}h h = π ∗ h . By Proposition 39, h h h is a solution to equation (5), and moreover, by Proposition 40, h h h is a complete solution. Since h ~ \widetilde{h} h is real, h h h is also real. This implies the claim. □
From Corollaries 57 and 67 and Theorem 68, we obtain Theorems 2 and 3.
Regularity of solutions to equation (5)
The purpose of this section is to prove the following proposition:
Proposition 69. Let h = ( h 1 , … , h r ) h=(h_{1},\ldots,h_{r}) h = ( h 1 , … , h r ) be a solution to equation (5) with det ( h ) = 1 \det(h)=1 det ( h ) = 1 , possibly incomplete. Then h h h has the following regularity properties:
h h h is of class C l o c 1 , 1 C_{\mathrm{loc}}^{1,1} C loc 1 , 1 .
h h h is of class C 2 C^{2} C 2 if and only if e φ e^{\varphi} e φ is continuous.
By invoking well-established general theory for the Poisson equation, it is immediately apparent that the regularity of the solution extends to C l o c 1 , α C_{\mathrm{loc}}^{1,\alpha} C loc 1 , α for any 0 < α < 1 0<\alpha<1 0 < α < 1 (cf. [49 ] ). In fact, Proposition 69 asserts that the regularity can be further improved to C l o c 1 , 1 C_{\mathrm{loc}}^{1,1} C loc 1 , 1 while, in general, it cannot reach C 2 C^{2} C 2 . The proof of Proposition 69 is based on the following lemma:
Lemma 70. Let U ⊆ C U\subseteq\mathbb{C} U ⊆ C be a connected open subset and Δ : = ∂ 2 ∂ x 2 + ∂ 2 ∂ y 2 = 4 ∂ 2 ∂ z ∂ z ‾ \Delta:=\frac{\partial^{2}}{\partial x^{2}}+\frac{\partial^{2}}{\partial y^{2}}=4\frac{\partial^{2}}{\partial z\partial\overline{z}} Δ := ∂ x 2 ∂ 2 + ∂ y 2 ∂ 2 = 4 ∂ z ∂ z ∂ 2 the standard Laplacian. Let u , f ∈ L l o c 1 ( U ) u,f\in L_{\mathrm{loc}}^{1}(U) u , f ∈ L loc 1 ( U ) satisfy the following in the sense of distributions:
Suppose that Δ f \Delta f Δ f is a locally finite signed measure. Then the following holds:
If f ∈ L l o c ∞ ( U ) f\in L_{\mathrm{loc}}^{\infty}(U) f ∈ L loc ∞ ( U ) , then u u u has a C l o c 1 , 1 C_{\mathrm{loc}}^{1,1} C loc 1 , 1 representative.
If f ∈ C 0 ( U ) f\in C^{0}(U) f ∈ C 0 ( U ) , then u u u has a C 2 C^{2} C 2 representative.
Proof. We begin by recalling the fundamental solution of the bi-Laplacian Δ 2 \Delta^{2} Δ 2 (cf. [16 , 55 ] ). Define F : C → R F:\mathbb{C}\to\mathbb{R} F : C → R by
F ( z ) : = { 1 8 π ∣ z ∣ 2 log ∣ z ∣ if z ≠ 0 , 0 if z = 0. F(z):=
\begin{cases}
\frac{1}{8\pi}|z|^{2}\log|z| & \text{if } z\ne0,\\
0 & \text{if } z=0.
\end{cases} F ( z ) := { 8 π 1 ∣ z ∣ 2 log ∣ z ∣ 0 if z = 0 , if z = 0.
Then one can easily verify that
Δ 2 F = δ 0 \Delta^{2}F=\delta_{0} Δ 2 F = δ 0
in the sense of distributions, where δ 0 \delta_0 δ 0 denotes the Dirac measure (cf. [16 ] and [55 Lemma 3.1] ). Furthermore, outside the origin, for the Hessian of F F F , we obtain
D 2 F : = ( ∂ 2 F ∂ x 2 ∂ 2 F ∂ x ∂ y ∂ 2 F ∂ y ∂ x ∂ 2 F ∂ y 2 ) = 1 8 π r 2 ( ( 2 log r + 1 ) r 2 + 2 x 2 2 x y 2 x y ( 2 log r + 1 ) r 2 + 2 y 2 ) . (99) D^2F:=\begin{pmatrix}
\frac{\partial^2F}{\partial x^2} & \frac{\partial^2F}{\partial x\partial y}\\
\frac{\partial^2F}{\partial y\partial x} & \frac{\partial^2F}{\partial y^2}
\end{pmatrix}
=\frac{1}{8\pi r^2}
\begin{pmatrix}
(2\log r+1)r^2+2x^2 & 2xy\\
2xy & (2\log r+1)r^2+2y^2
\end{pmatrix}. \tag*{(99)} D 2 F := ( ∂ x 2 ∂ 2 F ∂ y ∂ x ∂ 2 F ∂ x ∂ y ∂ 2 F ∂ y 2 ∂ 2 F ) = 8 π r 2 1 ( ( 2 log r + 1 ) r 2 + 2 x 2 2 x y 2 x y ( 2 log r + 1 ) r 2 + 2 y 2 ) . (99)
where r r r is defined by r : = x 2 + y 2 r:=\sqrt{x^2+y^2} r := x 2 + y 2 . Since F ∈ W l o c 2 , 1 ( C ) F\in W_{\mathrm{loc}}^{2,1}(\mathbb{C}) F ∈ W loc 2 , 1 ( C ) by [55 Lemma 3.1] , (98) also holds in the sense of distributions on C \mathbb{C} C . Let v = ( x y ) T v=(x\ y)^{\mathrm{T}} v = ( x y ) T be a 2 × 1 2\times1 2 × 1 column vector with components x x x and y y y and I 2 I_2 I 2 the 2 × 2 2\times2 2 × 2 unit matrix. Then (98) can be rewritten as follows:
D 2 F = ( 1 2 Δ F − 1 8 π ) I 2 + 2 8 π r 2 v v T (100) D^2F=\left(\frac{1}{2}\Delta F-\frac{1}{8\pi}\right)I_2+\frac{2}{8\pi r^2}vv^{\mathrm{T}} \tag*{(100)} D 2 F = ( 2 1 Δ F − 8 π 1 ) I 2 + 8 π r 2 2 v v T (100)
= 1 2 ( Δ F ) I 2 + 1 8 π ( 2 r 2 v v T − I 2 ) . (101) =\frac{1}{2}(\Delta F)I_2+\frac{1}{8\pi}\left(\frac{2}{r^2}vv^{\mathrm{T}}-I_2\right). \tag*{(101)} = 2 1 ( Δ F ) I 2 + 8 π 1 ( r 2 2 v v T − I 2 ) . (101)
Let U ′ ⊆ U U'\subseteq U U ′ ⊆ U be an arbitrary relatively compact open subset. Let χ : U → R \chi:U\to\mathbb{R} χ : U → R be a smooth function with compact support such that χ ∣ U ′ ≡ 1 \chi|_{U'}\equiv1 χ ∣ U ′ ≡ 1 . Let ν \nu ν be the finite signed measure on R 2 \mathbb{R}^2 R 2 extending χ Δ f \chi\Delta f χ Δ f to the outside of U U U by setting it to 0 0 0 . We set h : = u − F ∗ ν h:=u-F*\nu h := u − F ∗ ν , where ∗ * ∗ denotes the convolution. Then h h h satisfies Δ 2 h = 0 \Delta^2h=0 Δ 2 h = 0 on U ′ U' U ′ , and thus in particular it is smooth on U ′ U' U ′ . By (100), we have the following in the sense of distributions:
D 2 u = D 2 h + D 2 ( F ∗ ν ) = D 2 h + ( D 2 F ) ∗ ν = D 2 h + { 1 2 ( Δ F ) I 2 + 1 8 π ( 2 r 2 v v T − I 2 ) } ∗ ν = D 2 h + 1 2 { ( Δ u ) − ( Δ h ) } I 2 + 1 8 π ( 2 r 2 v v T − I 2 ) ∗ ν = 1 2 f I 2 + ( D 2 h − 1 2 ( Δ h ) I 2 ) + 1 8 π ( 2 r 2 v v T − I 2 ) ∗ ν . (102) \begin{aligned}
D^2u&=D^2h+D^2(F*\nu)\\
&=D^2h+(D^2F)*\nu\\
&=D^2h+\left\{\frac{1}{2}(\Delta F)I_2+\frac{1}{8\pi}\left(\frac{2}{r^2}vv^{\mathrm{T}}-I_2\right)\right\}*\nu\\
&=D^2h+\frac{1}{2}\{(\Delta u)-(\Delta h)\}I_2+\frac{1}{8\pi}\left(\frac{2}{r^2}vv^{\mathrm{T}}-I_2\right)*\nu\\
&=\frac{1}{2}fI_2+\left(D^2h-\frac{1}{2}(\Delta h)I_2\right)+\frac{1}{8\pi}\left(\frac{2}{r^2}vv^{\mathrm{T}}-I_2\right)*\nu. \tag*{(102)}
\end{aligned} D 2 u = D 2 h + D 2 ( F ∗ ν ) = D 2 h + ( D 2 F ) ∗ ν = D 2 h + { 2 1 ( Δ F ) I 2 + 8 π 1 ( r 2 2 v v T − I 2 ) } ∗ ν = D 2 h + 2 1 {( Δ u ) − ( Δ h )} I 2 + 8 π 1 ( r 2 2 v v T − I 2 ) ∗ ν = 2 1 f I 2 + ( D 2 h − 2 1 ( Δ h ) I 2 ) + 8 π 1 ( r 2 2 v v T − I 2 ) ∗ ν . (102)
The matrix-valued function 2 r 2 v v T − I 2 \frac{2}{r^2}vv^{\mathrm{T}}-I_2 r 2 2 v v T − I 2 is defined everywhere except at the origin; we extend it to the origin by setting its value there to 0 0 0 , and denote this extended function by K K K . Since ν \nu ν is a finite signed measure, its total variation ∣ ν ∣ |\nu| ∣ ν ∣ is finite and satisfies ∣ ∫ C b d ν ∣ ≤ ∫ C ∣ b ∣ d ∣ ν ∣ \left|\int_{\mathbb{C}}b\,d\nu\right|\leq\int_{\mathbb{C}}|b|\,d|\nu| ∫ C b d ν ≤ ∫ C ∣ b ∣ d ∣ ν ∣ for every b ∈ L 1 ( ∣ ν ∣ ) b\in L^1(|\nu|) b ∈ L 1 ( ∣ ν ∣ ) . This implies the following estimate for each x ∈ C x\in\mathbb{C} x ∈ C :
∣ K ∗ ν ∣ ( x ) = ∣ ∫ C K ( x − y ) d ν ( y ) ∣ ≤ ∫ C ∣ K ( x − y ) ∣ d ∣ ν ∣ ( y ) ≤ ∣ K ∣ L ∞ ∫ C d ∣ ν ∣ . \begin{aligned}
|K*\nu|(x)&=\left|\int_{\mathbb{C}}K(x-y)d\nu(y)\right|\\
&\leq\int_{\mathbb{C}}|K(x-y)|d|\nu|(y)\\
&\leq|K|_{L^\infty}\int_{\mathbb{C}}d|\nu|.
\end{aligned} ∣ K ∗ ν ∣ ( x ) = ∫ C K ( x − y ) d ν ( y ) ≤ ∫ C ∣ K ( x − y ) ∣ d ∣ ν ∣ ( y ) ≤ ∣ K ∣ L ∞ ∫ C d ∣ ν ∣.
Suppose that f ∈ L l o c ∞ f\in L^\infty_{\mathrm{loc}} f ∈ L loc ∞ . Then from the argument above, the right-hand side of (101) belongs to L l o c ∞ ( U ′ ) L^\infty_{\mathrm{loc}}(U') L loc ∞ ( U ′ ) . Since U ′ U' U ′ is an arbitrary relatively compact open subset, u u u belongs to W l o c 2 , ∞ ( U ) W_{\mathrm{loc}}^{2,\infty}(U) W loc 2 , ∞ ( U ) and therefore it has a C l o c 1 , 1 C_{\mathrm{loc}}^{1,1} C loc 1 , 1 representative. Suppose that f f f is continuous. We first show that μ : = Δ f \mu:=\Delta f μ := Δ f has no atoms. Fix a ∈ U a \in U a ∈ U . We decompose μ \mu μ as μ = μ + − μ − \mu= \mu^{+} - \mu^{-} μ = μ + − μ − as a difference of two locally finite nonnegative measures μ + \mu^{+} μ + and μ − \mu^{-} μ − (for instance, one may take the Jordan decomposition). Choose local subharmonic potentials p ± p_{\pm} p ± satisfying Δ p ± = μ ± \Delta p_{\pm} = \mu^{\pm} Δ p ± = μ ± . By Weyl’s lemma, p + − p − − f p_{+} - p_{-} - f p + − p − − f has a harmonic representative. Since f f f is continuous, the Lelong numbers of p + p_{+} p + and p − p_{-} p − coincide. The standard characterization of Lelong numbers therefore gives μ + ( { a } ) = μ − ( { a } ) \mu^{+}(\{a\}) = \mu^{-}(\{a\}) μ + ({ a }) = μ − ({ a }) . Consequently, μ ( { a } ) = 0 \mu(\{a\}) = 0 μ ({ a }) = 0 . As above, we take an arbitrary relatively compact open subset U ′ ⊆ U U^{\prime} \subseteq U U ′ ⊆ U and let χ : U → R \chi: U \to\mathbb{R} χ : U → R be a smooth function with compact support such that χ ∣ U ′ ≡ 1 \chi|_{U^{\prime}} \equiv1 χ ∣ U ′ ≡ 1 . We define ν \nu ν to be the finite signed measure on R 2 \mathbb{R}^{2} R 2 extending χ Δ f \chi\Delta f χ Δ f to the outside of U U U by setting it to 0 0 0 . Since μ \mu μ has no atoms, neither do ν \nu ν and ∣ ν ∣ |\nu| ∣ ν ∣ . This implies that K ∗ ν K * \nu K ∗ ν is continuous. Indeed, let x ∈ C x \in\mathbb{C} x ∈ C be an arbitrary point and ( x n ) n ≥ 1 (x_n)_{n\geq1} ( x n ) n ≥ 1 an arbitrary sequence that converges to x x x . Since ∣ ν ∣ |\nu| ∣ ν ∣ has no atoms and K K K is continuous except for the origin, K ( x n − y ) K(x_n-y) K ( x n − y ) converges to K ( x − y ) K(x-y) K ( x − y ) for ∣ ν ∣ |\nu| ∣ ν ∣ -almost every y y y . Also, we have a uniform estimate ∣ K ( x n − y ) − K ( x − y ) ∣ ≤ 2 sup C ∣ K ∣ |K(x_n-y)-K(x-y)| \leq2\sup_{\mathbb{C}}|K| ∣ K ( x n − y ) − K ( x − y ) ∣ ≤ 2 sup C ∣ K ∣ for all n ≥ 1 n \geq1 n ≥ 1 and y ∈ C y \in\mathbb{C} y ∈ C . This estimate and Lebesgue’s dominated convergence theorem imply that
∣ ( K ∗ ν ) ( x n ) − ( K ∗ ν ) ( x ) ∣ ≤ ∫ C ∣ K ( x n − y ) − K ( x − y ) ∣ d ∣ ν ∣ ( y ) \left|(K * \nu)(x_n) - (K * \nu)(x)\right| \leq\int_{\mathbb{C}} |K(x_n-y)-K(x-y)|\,d|\nu|(y) ∣ ( K ∗ ν ) ( x n ) − ( K ∗ ν ) ( x ) ∣ ≤ ∫ C ∣ K ( x n − y ) − K ( x − y ) ∣ d ∣ ν ∣ ( y )
→ 0 as n → ∞ . \to0 \text{ as } n \to\infty. → 0 as n → ∞.
Therefore, by (102) , u u u has a C 2 C^{2} C 2 representative on U ′ U^{\prime} U ′ , and since U ′ U^{\prime} U ′ is an arbitrary relatively compact open subset, u u u has a C 2 C^{2} C 2 representative on U U U . □
We establish a sufficient condition on f f f for the requirement in Lemma 70 to hold:
Lemma 71. Let U ⊆ C U \subseteq\mathbb{C} U ⊆ C be open, and let f ∈ L l o c ∞ ( U ) f \in L^{\infty}_{\mathrm{loc}}(U) f ∈ L loc ∞ ( U ) be positive almost everywhere with log f ∈ L l o c 1 ( U ) \log f \in L^{1}_{\mathrm{loc}}(U) log f ∈ L loc 1 ( U ) . If Δ log f \Delta\log f Δ log f is a locally finite signed measure, then Δ f \Delta f Δ f is also a locally finite signed measure.
Proof. Let V ⋐ U V \Subset U V ⋐ U be an arbitrary relatively compact open subset and ( ρ ε ) ε > 0 (\rho_{\varepsilon})_{\varepsilon>0} ( ρ ε ) ε > 0 a family of standard mollifiers for mollification on V V V . We set f ~ : = log f \widetilde{f} := \log f f := log f and for each ε > 0 \varepsilon> 0 ε > 0 , we denote by f ~ ε : = f ~ ∗ ρ ε \widetilde{f}_{\varepsilon} := \widetilde{f} * \rho_{\varepsilon} f ε := f ∗ ρ ε the mollified function defined on an open set smaller than V V V . We take an arbitrary relatively compact open subset V ′ ⋐ V V^{\prime} \Subset V V ′ ⋐ V and ε 0 > 0 \varepsilon_{0} > 0 ε 0 > 0 small enough so that f ~ ε 0 \widetilde{f}_{\varepsilon_{0}} f ε 0 is defined on V ′ V^{\prime} V ′ . Then by the assumption that log f \log f log f is locally essentially bounded above, there is a constant M M M such that f ~ ε ≤ M \widetilde{f}_{\varepsilon} \leq M f ε ≤ M on V ′ V^{\prime} V ′ for all 0 < ε < ε 0 0 < \varepsilon< \varepsilon_{0} 0 < ε < ε 0 . By a direct calculation, we have
Δ ( e f ~ ε ) = e f ~ ε ( ( Δ f ~ ) ∗ ρ ε ) + e f ~ ε ∣ d f ~ ε ∣ 2 ≥ − e M ( ∣ Δ f ~ ∣ ∗ ρ ε ) , (103) \Delta\left(e^{\widetilde{f}_{\varepsilon}}\right)
= e^{\widetilde{f}_{\varepsilon}}\left((\Delta\widetilde{f}) * \rho_{\varepsilon}\right)
+ e^{\widetilde{f}_{\varepsilon}}|d\widetilde{f}_{\varepsilon}|^{2}
\geq-e^{M}\left(|\Delta\widetilde{f}| * \rho_{\varepsilon}\right), \tag*{(103)} Δ ( e f ε ) = e f ε ( ( Δ f ) ∗ ρ ε ) + e f ε ∣ d f ε ∣ 2 ≥ − e M ( ∣Δ f ∣ ∗ ρ ε ) , (103)
where ∣ Δ f ~ ∣ |\Delta\widetilde{f}| ∣Δ f ∣ denotes the total variation of Δ f ~ \Delta\widetilde{f} Δ f . Since e f ~ ε e^{\widetilde{f}_{\varepsilon}} e f ε converges to f = e f ~ f = e^{\widetilde{f}} f = e f in L l o c 1 L^{1}_{\mathrm{loc}} L loc 1 , taking the limit as ε ↘ 0 \varepsilon\searrow0 ε ↘ 0 in (103) , we obtain
Δ f + e M ∣ Δ f ~ ∣ ≥ 0 (104) \Delta f + e^{M}|\Delta\widetilde{f}| \geq0 \tag*{(104)} Δ f + e M ∣Δ f ∣ ≥ 0 (104)
on V ′ V^{\prime} V ′ . Since positive distributions are locally finite positive measures, (104) shows that Δ f + e M ∣ Δ f ~ ∣ \Delta f + e^{M}|\Delta\widetilde{f}| Δ f + e M ∣Δ f ∣ is a locally finite positive measure on V ′ V^{\prime} V ′ . Hence Δ f \Delta f Δ f is a locally finite signed measure on V ′ V^{\prime} V ′ . Since V ⋐ U V \Subset U V ⋐ U and V ′ ⋐ V V^{\prime} \Subset V V ′ ⋐ V were arbitrary, the conclusion follows. □
Proof of Proposition 69. If φ ≡ − ∞ \varphi\equiv-\infty φ ≡ − ∞ , then H r ≡ 0 H_r \equiv0 H r ≡ 0 , and standard elliptic bootstrapping shows that h h h is smooth. We may therefore assume that φ ≢ − ∞ \varphi\not\equiv-\infty φ ≡ − ∞ . It is enough to prove the assertions only for h 1 h_1 h 1 and h r h_r h r since by the general theory of the Poisson equation (cf. [49 ] ) and the standard elliptic regularity, h j h_j h j (2 ≤ j ≤ r − 1 2 \leq j \leq r - 1 2 ≤ j ≤ r − 1 ) are at least in the class C l o c 0 , α C^{0,\alpha}_{\mathrm{loc}} C loc 0 , α for any 0 < α < 1 0 < \alpha< 1 0 < α < 1 . Since the proofs are the same, we prove that h 1 h_1 h 1 belongs to C l o c 1 , 1 C^{1,1}_{\mathrm{loc}} C loc 1 , 1 for simplicity. We work in a local holomorphic coordinate and identify the metrics with their coefficient functions; Lemma 70 is applied to log h 1 \log h_1 log h 1 . Note that by Lemma 71 , the Laplacians of H 1 H_1 H 1 and H r H_r H r are locally finite because the curvatures of H 1 H_1 H 1 and H r H_r H r are locally finite. Therefore we can apply Lemma 70 to the equation
− 1 F h 1 + vol ( H r ) − vol ( H 1 ) = 0 (105) \sqrt{-1}F_{h_1}+\operatorname{vol}(H_r)-\operatorname{vol}(H_1)=0 \tag*{(105)} − 1 F h 1 + vol ( H r ) − vol ( H 1 ) = 0 (105)
and it follows that h 1 h_1 h 1 belongs to C l o c 1 , 1 C^{1,1}_{\mathrm{loc}} C loc 1 , 1 . Suppose that e φ e^\varphi e φ is continuous. Then H 1 H_1 H 1 and H r H_r H r are continuous, and thus by applying Lemma 70 to (104), h 1 h_1 h 1 is of class C 2 C^2 C 2 . Conversely, suppose that h 1 h_1 h 1 is of class C 2 C^2 C 2 . By this assumption, the curvature term in (104) is continuous. Therefore equality (104) implies that e φ e^\varphi e φ is continuous, which concludes the proof.
E-mail address 1: natsuo.miyatake.e8@tohoku.ac.jp
E-mail address 2: natsuo.m.math@gmail.com
Mathematical Science Center for Co-creative Society, Tohoku University, 468-1 Aramaki Azaaoba, Aoba-ku, Sendai 980-0845, Japan.