A Bando-Mabuchi Uniqueness Theorem Li YI Institut Elie Cartan, Nancy 3 Abstract. In this paper we prove a uniqueness theorem on generalized 1 0 K¨ahler-Einstein metrics on Fano manifolds. Our result generalize the one 2 shown by Berndtsson using the convexity properties of Bergman kernels. The n same technics as well as that of the regularization of closed positive currents a J and an estimate due to Demailly-Koll´ar will be used in our proof. 5 1 §0. Introduction. ] V C Let X be an n-dimensional projective manifold, here n 3; we denote by ≥ . c (X) its first Chern class, i.e. the Chern class corresponding to the anti h 1 t canonical bundle K . Let α be a (1,1) class on X; here we denote by a − X { } m α a non-singular representative. We assume that the following requirements [ are satisfied. 2 (i) The class c (X) α is Ka¨hler, i.e. it contains a Ka¨hler metric which v 1 −{ } 7 we denote by ω; 4 8 2 (ii) The class α is pseudo-effective; let Θ α be a closed positive { } ∈ { } . current; 1 0 3 (iii) If we denote Ric(ω) the Ricci curvature associated to the metric ω, 1 : then we have v Ric(ω) = ω +Θ √ 1∂∂f i Θ X − − for some function f which is unique up to an additive constant. We r Θ a assume that there exists a real number ε > 0 such that 0 e−2(1+ε0)fΘdV < . Z ∞ X In other words, the singularities of the current Θ are not too wide. Following Bedford and Taylor (c.f. [5]), if φ is a psh bounded function defined on an open subset Ω Cn, then the quantity ⊂ √ 1∂∂φ n − (cid:0) (cid:1) 1 is a positive measure of locally finite mass on Ω. Hence it makes sense to consider the equation (⋆) (ω +√ 1∂∂ϕ)n = e−ϕ−fΘωn − where the solution ϕ is assumed to be bounded, and such that ω := ω +√ 1∂∂ϕ ϕ − is a Ka¨hler current (i.e. it is greater than a Ka¨hler metric). We remark that if Θ is non-singular, then the equation (⋆) is equivalent with the following identity Ric(ω ) = ω +Θ. ϕ ϕ The uniqueness of the solution ϕ up to a biholomorphism of X was estab- lished by Bando-Mabuchi in [4]. If Θ is the current of integration on a divisor D, such that the pair (X,D) is klt, then the question was settled by Berndts- son in [2], who found a proof within the framework of the psh variation of Bergman kernels. In this article we show that the following generalization of the result in [2] holds. Theorem 1. Let ϕ and ϕ be two bounded solutions of the equation (⋆). 0 1 Then there exists a biholomorphism F : X X such that → F (ω ) = ω and F (Θ) = Θ. ∗ ϕ1 ϕ0 ∗ If we assume that the anti-canonical class has further positivity proper- ties, for example, K 0, then one proves in [2] that the proceeding result X − ≥ holds under the assumption e−fΘ ∈ L1loc(X). However, we remark that in order to have this result, we have to assume two strong positivity properties: c (X) α is a Ka¨hler class, and c (X) is a semi-positive class. 1 1 −{ } The arguments we will invoke in order to prove this result generalize the approach by Berndtsson in [2]. It is based on the convexity properties of Bergman kernel, which we recall next. 2 Let L X be a line bundle, which is endowed with a family of non- → singular metrics h = e ϕt, t D, the unit disk. We assume that the function t − ∈ ϕ(t,x) = ϕ (x) t is non-singular and psh. We can see L as a bundle over the family X D; × the bundle structure is independent of t D, but the hermitian structure ∈ varies with respect to t. Let E := p (K +L) X D/D ∗ × be the direct image of the relative adjoint bundle of L, and let (ut)t D be a ∈ holomorphic section of E. Of course, in general (u ) does not correspond to t a holomorphic n-form on X D : we can only represent (ut)t D by an n-form U on X D such that ∂U×is equal to some multiple of dt∈, and such that × U = u . In order to evaluate the curvature Θ of E in the direction of X t t E | ×{} the section (u ), we introduce a family (v ) of (n 1)-forms with values in L t t − such that ∂v ω = 0, t ∗ ∧ ∂ϕtv = P (ϕ˙ u ) t t t ∗ ⊥ where P is the projection on the orthogonal complement of the holomorphic n-forms on each fiber, then we have the following formula Θ u ,u = p (c √ 1∂∂ϕ u ue ϕt)+ ∂v 2e ϕt√ 1dt dt E t t n t − t − h i ∗ − ∧ ∧ Z k k − ∧ X b b where we denote by u = (u dt v ) , another representation of our initial t t t − ∧ section (u ). t b We assume now that for some geometric reasons, the curvature of E vanishes. As a consequence, v will be holomorphic, as well as the first term t on the right hand side of the formula above. But then we can define a vector field by the formula v = V u ; t t t − ⌋ it will be holomorphic outside u = 0. Actually, under the condition Θ = 0 t E one can see that V is the complex gradient of √ 1∂ϕ˙ ; finally, a simple com- t t − putation shows that the flow associated to V , F, is the holomorphic map we 1 3 seek. While trying to implement this scheme in our setting, we have to face two important difficulties. In our case, the bundle L will be just K , endowed X − with the following family of metrics. We connect ϕ and ϕ with a geodesic, 0 1 (this is known to exist), and this will give us a metric ϕ whose curvature t ω = ω +√ 1∂∂ ϕ t X t − belongs to the class c (X) α . Then the current 1 −{ } ω +Θ t with Θ α is closed, positive, and belongs to c (X) for each t. Our main 1 ∈ { } concerns are: Θ is not necessarily smooth; ∗ the current ω = ω+√ 1∂∂ ϕ is not necessarily Ka¨hler, it is only known t X t ∗ − to be positive. The plan of our article will be as follows: we approximate ω and Θ with t non-singular objects (section 1), and then we construct the analogue vν of t the (n 1)-form v mentioned above. Thanks to the fact that ϕ and ϕ t 0 1 − are solutions of (⋆), we show that the curvature of E, endowed with the regularized metric, converges to zero. An uniformity argument (which is quite subtle, given that the Lelong set of Θ could be quite complicated) shows that vν converges to a holomorphic (n 1)-form. Then the vector field t − will be defined as above. §1. Regularization process. Fix a non-singular real (1,1)-form ω in the Ka¨hler class c (X) α , a 1 − { } quasi-psh function ϕ is said to be ω-psh if ω + √ 1∂∂ϕ 0. We denote − ≥ PSH(X,ω) L the convex space of all ω-psh functions which are bounded ∞ ∩ on X. Let ϕ and ϕ PSH(X,ω) L , we call the path of semi-positive 0 1 ∞ ∈ ∩ forms ω = ω +√ 1∂∂ϕ witht [0,1] t t − ∈ a (generalized) geodesic in PSH(X,ω). The following theorem implies the existence of a continuous geodesic which connects ω and ω (cf. [3]): 0 1 4 Theorem 2. Assume that the semi-positive closed (1,1)-forms ω and ω 0 1 belong to the same K¨ahler class c (X) α and have bounded coefficients. 1 −{ } Then there exists a geodesic ω connecting ω and ω with the properties t 0 1 that it is continuous on [0,1] X and that there is a constant C such that × ω Cω, i.e. ω has uniformly bounded coefficients. t t ≤ Remark 3. The geodesic ω can be constructed in such a way that it is t Lipschitz in t. Indeed, like shown in [2], define first ( ) ϕ = sup κ t t ∗ { } where the supremum is taken over all psh κ with t lim κ ϕ ; t 0,1 t 0,1 ≤ → then construct a barrier χ = max(ϕ ARet,ϕ +A(Ret 1)) t 0 1 − − with A sufficiently large. This χ is psh and we have t lim χ = ϕ . t 0,1 t 0,1 → Therefore the supremum in ( ) has the same property if we restrict to those ∗ κ which are larger that χ. For such κ the onesided derivative at 0 is larger than A and the onesided derivative at 1 is smaller than A. We assume from − now on that κ is independent of the imaginary part of t, κ is convex in t so the derivative with respect to t increases, and must lie between A and A. − Therefore ϕ satisfies t ϕ ARet ϕ ϕ +ARet 0 t 0 − ≤ ≤ at 0 and a similar estimate at 1. Thus lim ϕ = ϕ t 0,1 t 0,1 → uniformly on X. In addition, the upper semicontinuous regularization ϕ of ∗t ϕ must satisfy the same estimate. Since ϕ is psh it belongs to the class of t ∗t competitors for ϕ and must coincide with ϕ , so ϕ is psh. t t t 5 Our proof begins with the regularization for ϕ , the ampleness of c (X) α t 1 −{ } allows us to write ϕ as a uniform limit of a decreasing sequence of non- t singular functions ϕν with the property that t ων := ω +√ 1∂∂ϕν 0. t − t ≥ In order to regularize the current Θ = α+√ 1∂∂θ, − we use the following result: Theorem 4. ([7]) Let T be a closed almost positive (1,1)-current and let α be a non-singular real (1,1)-form in the same ∂∂-cohomology class as Θ, i.e. Θ = α+√ 1∂∂θ where θ is an almost psh function. Let γ be a continuous − real (1,1)-form such that Θ γ. Suppose that (1) is equipped with a ≥ OTX non-singular hermitian metric such that the Chern curvature form satisfies c( (1))+π u 0 OTX X∗ ≥ with π : P(T X) X and with some nonnegative non-singular (1,1)-form X ∗ → u on X. Fix a hermitian metric ω0 on X. Then for every c > 0, there is a sequence of closed almost positive (1,1)-current Θ = α+√ 1∂∂θ such c,k c,k − that θ is non-singular on X E (Θ) and decreases to θ as k tends to + c,k c \ ∞ (in particular, the current Θ is non-singular on X E (Θ) and converges c,k c \ weakly to Θ on X) and such that (1) Θ γ min λ ,c u ε ω0 where: c,k k k ≥ − { } − (2) λ (x) is a decreasing sequence of continuous functions on X such that k lim λ (x) = ν(Θ,X) at every point. k k + → ∞ (3) ε is positive decreasing and lim ε = 0. k k k + → ∞ (4) ν(Θ ,x) = (ν(Θ,x) c) at every point x X. c,k + − ∈ Recall that E (Θ) = x X;ν(Θ,x) c c { ∈ ≥ } 6 is the c-upperlevel set of Lelong numbers of Θ. A well-known theorem of Siu [15] asserts that E (Θ) is an analytic subset of X for any c > 0. c Choose u = Aω0 where A is a positive constant large enough and choose c = 1 with ν N a decreasing sequence, according to the above theorem, we ν ∈ get a sequence of functions θν such that they are non-singular on X E (Θ) 1 \ ν and 1 Θν := α+√ 1∂∂θν Amin λ , ω0 ε ω0 ν ν − ≥ − { ν} − holds on X for every ν, where λ (x) is a decreasing sequence of continuous ν functions on X such that lim λ (x) = ν(Θ,x) at every point and ε are ν ν ν + positive decreasing such tha→t ∞lim ε = 0. ν ν + → ∞ During the proof we need to use the ”metric” form ”e ” concerning the −· currents ω and Θ. So we revise next the notations. Denote φ = φ0+ϕ with t t t √ 1∂∂φ0 = ω. So ω can be written as t − ω = √ 1∂∂φ . t t − Similarly, denote ψ = ψ0 +θ with √ 1∂∂ψ0 = α. Then Θ can be presented − as Θ = √ 1∂∂ψ. − Hence if we denote τ = φ + ψ and respectively τν = φν + ψν, the above t t t t discussions imply that the sequence of currents ων +Θν = √ 1∂∂τν t − t decreases to ω +Θ = √ 1∂∂τ t t − as ν tends to + and is non-singular on X E (Θ). We evaluate next the 1 lower bounds of∞ων + Θν. To this end, deno\teνfirst a neighborhood set of t E (Θ) by 1 ν U = x X : eψν < δ δν { ∈ ν} (the choice of δ to beprecise ina moment). Remark that according to Theo- ν rem 2, ω has uniformly bounded coefficients, since ων converge uniformly to t t 7 ωt, we have actually eτtν < Cδν for x ∈ Uδν with C > 0 a numerical constant. Then by Theorem 4, we have 1 √ 1∂∂τν Amin λ , +ε ω0 (1) − t ≥ −(cid:18) { ν ν} ν(cid:19) 1 on X U . As ν + , the quantity Amin λ , +ε ω0 tends to \ δν → ∞ −(cid:18) { ν ν} ν(cid:19) zero; Globally on X, we have 1 √ 1∂∂τν 2A ω0 3Aω0. (2) − t ≥ −(cid:18) ν(cid:19) ≥ − In the following proof we need to solve some ∂τtν-equations, to insure that the solutions of such equations we will deal with satisfy certain L2-properties, we need to modify further our approximations in order that the new ”metrics” are non-singular on X. To this end, define 1 τν = log eτtν + . t (cid:18) ν(cid:19) e 1 Since eτtν + > 0, functions τν constructed in this way are therefore non- ν t singular on X. We analyze next their Hessian properties. e 1 √ 1∂∂log eτtν + − (cid:18) ν(cid:19) eτtν∂τν =√ 1∂ t − eτtν + 1 ν (eτtν∂∂τν +eτtν∂τν ∂τν)(eτtν + 1) e2τtν∂τν ∂τν =√ 1 t t ∧ t ν − t ∧ t − (eτtν + 1)2 ν (eτtν + 1)eτtν∂∂τν + 1eτtν∂τν ∂τν =√ 1 ν t ν t ∧ t − (eτtν + 1)2 ν eτtν 1eτtν =√ 1 ∂∂τν +√ 1 ν ∂τν ∂τν. − eτtν + 1 · t − (eτtν + 1)2 · t ∧ t ν ν We hope that Hessians of these new functions τν have similar properties as t τν, that is, t √ 1∂∂τν e − t e 8 would be uniformly bounded from below on X and almost positive (in other words, thelower boundstendtozeroasν tendstoinfinity) onX U .Indeed, \ δν this will be true if we choose properly the value of δ . Observe first that we ν have eτtν 0 < 1, eτtν + 1 ≤ ν so the first term in the last equality satisfies our requirements. We now analyze the second term. Notice that we have √ 1∂τν ∂τν = ∂τν 2 0 − t ∧ t | t | ≥ and 1eτtν ν 0 (3) (eτtν + 1)2 ≥ ν hold on X. So it is sufficient to check that the function in (3) goes to zero on X U . Indeed, for x X U , we have \ δν ∈ \ δν 1eτtν 1 eτtν 1 1 C ν = . (eτtν + ν1)2 ≤ ν e2τtν ν eτtν ≤ νδν 1 Hence if we choose for example δ = , we obtain that ν √ν C C = νδ √ν ν tends to zero as ν tends to infinity. Therefore the second term fulfills also our requirements. We can conclude from (1), (2) and the above discussions that √ 1∂∂τν A ω0 (4) − t ≥ − 1 on X with A > 0 a numerical constant as well as 1 e √ 1∂∂τν A ω0 (5) − t ≥ − δν,ν on X U with the property that eA tends to zero as ν + . For the \ δν δν,ν → ∞ sake of simplifying the notations, we denote from now on τν still by τν with t t the understanding that it is now a sequence of non-singular functions on X e satisfying (4) and (5) on the relevant sets. 9 §2. Constructions of vν. t Like explained in the introduction, finding solutions of the ∂τtν-equations ∂τtνvν = Pν,t(τ˙νu ) (6) t t t ⊥ is important in the search of holomorphic vector fields. In this expression, u t is the section of E described in the introduction, ∂τν τ˙ν = t t ∂t and Pν,t is the projection on the orthogonal complement of the holomorphic n-form⊥s on each fiber, and finally the operator ∂τtν is defined by ∂τtν := eτtν∂e−τtν = ∂ −∂τtν ∧. We will explain why equation (6) can always be solved for τν fixed in a t moment. In order to analyze the uniformity properties of these solutions, the following lemma will be important (it is a slight variation of Lemma 6.2 in [2]). Lemma 5. Let L be a holomorphic line bundle over X with a metric ξ satisfying √ 1∂∂ξ c ω with c > 0 a numerical constant. Let ξ be a 0 0 0 − ≥ − smooth metric on L with ξ ξ and assume 0 ≤ I := eξ0 ξ < . − Z ∞ X Then there is a constant B, only depending on I and ξ such that if f is a 0 ∂-exact L valued (n,1)-form with f 2e ξ 1, (7) − Z | | ≤ X there is a solution u to ∂u = f with u 2e ξ B. − Z | | ≤ X 10