<ph f="cmr"> </ph><ph f="cmbx">Blowing up Kähler manifolds with constant scalar curvature II</ph>

Claudio Arezzo

Frank Pacard

claudio.arezzo@unipr.it, Dipartimento di Matematica, Universita' di Parma, Via M. D'Azeglio, 85, 43100 Parma, Italy pacard@univ-paris12.fr, University Paris 12 and Institut Universitaire de France, France
1991 Math. Subject Classification: 58E11, 32C17.

1 Introduction

Assume that ( M , ω M )   is a constant scalar curvature compact Kähler manifold. Given m   distinct points p 1 , … , p m ∈ M   , we define Bl p 1 , … , p m M   to be the blow up of M   at the points p 1 , … , p m   . The question we would like to address in the paper is whether Bl p 1 , … , p m M   can be endowed with a constant scalar curvature Kähler form. In [2] , we have already given a positive answer to this question in the case where the manifold has no nontrivial vanishing holomorphic vector field (i.e. has no non zero holomorphic vector field which vanish somewhere). This condition is for example fulfilled when the group of automorphisms of M   is discrete. Under this condition, we have obtained the following :
Theorem 1.1. [2] Assume that ( M , ω M )   is a constant scalar curvature compact Kähler manifold without nontrivial vanishing holomorphic vector field. Then, M ~   , the blow up at finitely many points has a constant scalar curvature Kähler form ω M ~   . In addition, if the scalar curvature of ω M   is not zero then the scalar curvatures of ω M ~   and of ω M   have the same signs.
In the case of 0   -scalar curvature metrics, we also have the :
Theorem 1.2. [2] Assume that ( M , ω M )   is a 0   -scalar curvature compact Kähler manifold without nontrivial vanishing holomorphic vector field. Then, the blow up of M   at finitely many points has a 0   -scalar curvature Kähler form provided the first Chern class of M   is non zero.
This last result compliments, in any dimension, previous constructions which have been obtained in complex dimension n = 2   and for 0   -scalar curvature metrics, by Kim-LeBrunPontecorvo [6] , LeBrun-Singer [13] and Rollin-Singer [15] . Indeed, using twistor theory, Kim-LeBrunPontecorvo and LeBrun-Singer have been able to construct such metrics by desingularizing some quotients of minimal ruled surfaces and, more recently, Rollin-Singer [16] have shown that one can desingularize compact manifolds of zero scalar curvature with cyclic orbifold groups, keeping the scalar curvature zero by solving on the desingularization the hermitian anti-selfdual equation which implies the Kähler scalar flat equation.
Theorem 1.1 and Theorem 1.2 are obtained using a connected sum of the Kähler form ω M   at each p ℓ   with a 0   -scalar curvature Kähler form η   which is defined on N : = Bl 0 C n   , the blow up of C n   at the origin. When n = 2   this model ( N , η )   has been known for long time and it is usually referred to in the literature as Burns metric [9] . In higher dimensions a similar model has been produced by Simanca [17] .
In the present paper, we are interested in the case where M   has nontrivial vanishing holomorphic vector fields X 1 , … , X d   , with d ≥ 1   (this condition in particular implies that M   has a nontrivial continuous automorphisms group) since this is precisely the case which is not covered by our previous result. Our main result states that the blow up of M   at sufficiently many points can be endowed with a constant scalar curvature Kähler metric provided these points are carefully chosen.
Theorem 1.3. Assume that ( M , ω M )   is a constant scalar curvature compact Kähler manifold. Then, there exists m 0 ≥ d + 1   and, for all m ≥ m 0   there exist a nonempty open subset U ⊂ M m   such that, for all ( p 1 , … , p m ) ∈ U   there exist constant scalar curvature Kähler forms on the blow up of Bl p 1 , … , p m M   .
In fact, we can be more precise and show that there exists ω ~ ɛ   , for ɛ ∈ ( 0 , ɛ 0 )   , a one parameter family of Kähler forms on Bl p 1 , … , p m M   which converges in C ∞   norm as ɛ   tends to 0   , to the Kähler form ω M   away from the points p ℓ   . Moreover, the sequence of Kähler forms ɛ − 2 ω ~ ɛ   converges in C ∞   norm to the Kähler form η   on N   the blow up at the origin of C n   . Finally, if ω M   had positive (resp. negative) scalar curvature then the Kähler forms ω ~ ɛ   have positive (resp. negative) scalar curvature. Let us emphasize that, in contrast with the case where the manifold M   does not carry any nontrivial vanishing holomorphic vector field, this time the number and position of the blow up points is not arbitrary.
Remark 1.1. We will see that, in some explicit cases, one can make use of the symmetries of the manifold M   , and construct constant scalar curvature Kähler forms on the blow up of M   at ( p 1 , … , p m )   with m ≤ d   . However this construction will not hold anymore for the choice of the points in some open set of M m   .
Let us mention that, for the 2   -dimensional complex manifolds, Kähler metrics with constant scalar curvature have been obtained by Rollin-Singer in [15] , [16] on blow ups of P 1 × P 1   , P 2   or T 1 × P 1   using a different approach based on both algebraic tools and a connect sum result, while for zero scalar curvature metrics on blow ups of P 1 × Σ g   , g ≥ 2   , the problem was solved by LeBrun-Singer [13] using twistor theory. We will recall these results in § 8   .
In the statement of Theorem 1.3, the value of m 0   and the choice of the blow up points are very delicate issues. To describe this, we still assume that ( M , ω M )   is a constant scalar curvature compact Kähler manifold and we define (see § 3   ) the elliptic fourth order operator L M : = − 1 2 Δ M 2 − Ric M ⋅ ∇ M 2 ,   where all operators are computed with respect to the metric induced by ω M   . We assume that the kernel of L M   is not trivial and we denote by ξ 0 ≡ 1 , ξ 1 , … , ξ d   the linearly independent functions which span this kernel. We can assume, without loss of generality that the functions ξ 1 , … , ξ d   have mean 0   . Observe that d   either denotes the dimension of the space of vanishing holomorphic vector fields or the dimension of the nontrivial kernel of L M   since, by a result of LeBrun-Simanca [12] these two dimensions are the same. This being understood, we define the matrix
M ( p 1 , … , p m ) : = ( ξ 1 ( p 1 ) . . . ξ 1 ( p m ) . . . . . . ξ d ( p 1 ) . . . ξ d ( p m ) ) (1)
as well as the integer valued functions ℭ 1 : = Rank M and ℭ 2 : = dim ( K + m ∩ Ker M )   where K + m   is the cone of vectors with positive entries in R m   .
Our main result is a consequence of the following sequence of results. The first one gives a sufficient condition on the functions ℭ 1   and ℭ 2   for Theorem 1.3 to hold :
Proposition 1.1. Assume that ( M , ω M )   is a constant scalar curvature compact Kähler manifold. Let us assume that m ≥ 1   and p 1 , … , p m   are chosen so that ℭ 1 = d   and that ℭ 2 ≠ 0   . Then there exists ɛ 0 > 0   and for all ɛ ∈ ( 0 , ɛ 0 )   , a constant scalar curvature Kähler form ω ~ ɛ   on Bl p 1 , … , p m M   .
Furthermore, the Kähler form ω ~ ɛ   satisfy the properties described after the statement of Theorem 1.3.
As will be explained in § 6   , the first condition ℭ 1 = d   is easily seen to be generic (and open) in the sense that :
Lemma 1.1. With the above notations, assume that m ≥ d   then the set of the points ( p 1 , … , p m ) ∈ M m   such that ℭ 1 = d   is open and dense in M m   .
A result of LeBrun-Simanca [12] (see also [4] and [13] ) implies that nontrivial vanishing holomorphic vector fields are in one to one correspondence with nonconstant elementsof the kernel of L M   . Furthermore, it is well known that, for a choice of blow up points p 1 , … , p d   in some open and dense subset of M d   , the group of automorphisms of Bl p 1 , … , p d M   is trivial (observe that, by virtue of the above mentioned result of LeBrun and Simanca d   is also equal to the dimension of the identity component of the automorphisms group of M   ). In view of all these results, one is tempted to conjecture that our condition ℭ 1 = d   is equivalent to the fact that the group of automorphisms of Bl p 1 , … , p d ~ M   is trivial.
However, these two conditions turn out to be of a different nature. Indeed, for example when d = 1   the kernel of L M   is spanned by ξ 0 ≡ 1   and ξ 1   , and the set of nontrivial vanishing holomorphic vector fields is spanned by X 1   . Then it is enough to choose p   not in the zero set of ξ 1   to get ℭ 1 = 1   , while the group of automorphisms of Bl p M   is trivial if and only if p   is chosen away from the 0   set of X 1   , but, according to LeBrun and Simanca, the 0   set of X 1   corresponds to the set of critical points of ξ 1   .
The second sufficient condition on the existence of points for which ℭ 2 ≠ 0   is more subtle. We prove, in § 6   , that this condition is always fulfilled for some careful choice of the points, provided their number m   is chosen larger than some value m 0 ≥ d + 1   .
Lemma 1.2. With the above notations, assume that m ≥ d + 1   , then the set of points ( p 1 , … , p m ) ∈ M m   for which ℭ 1 = d   and ℭ 2 ≠ 0   is open. Moreover, there exists m 0 ≥ d + 1   such that, for all m ≥ m 0   the set of points ( p 1 , … , p m ) ∈ M m   for which ℭ 1 = d   and ℭ 2 ≠ 0   is nonempty.
The proof of the Lemma 1.2 is due to E. Sandier, and we are very grateful to him for allowing us to present it here. By opposition to the first condition, it is easy to convince oneself that this condition does not hold for generic choice of the points.
Remark 1.2. We believe that m 0   , the minimal number of points for which the second condition is fulfilled, should be equal to d + 1   . We have only been able to prove this fact for some special manifolds where all elements of the kernel of L M   are explicitly known.
It is interesting to observe that our sufficient conditions depends only on the complex structure of the base manifold and not on riemannian properties of its Kähler metric.
This is shared with all known necessary conditions, from Matsushima-Lichnerovicz's one, Futaki's character and the known stability conditions implied by the existence of canonical metrics thanks to the work of Donaldson, Mabuchi and Paul-Tian. We now give a number of explicit examples for which our result can be applied.
As already mentioned in Remark 1.1, by definition m 0   is larger than the dimension of vanishing holomorphic vector fields on M   and, using some symmetries of the manifolds, one can still construct constant scalar curvature Kähler forms on the blow up of M   at fewer points. To illustrate this fact, we have, when M = P n   , we have :
Lemma 1.3. When M = P n   , then d = n 2 + 2 n   and m 0 ≤ 2 n ( n + 1 )   .
Applying Theorem 1.3, we get the existence of constant scalar curvature Kähler forms on the blow up of P n   at m   points for m ≥ m 0   which belong to some nonempty open set of M m   . One can also work equivariantly using the symmetries of P n   to get the :
Corollary 1.1. For any m ≥ n + 1   there exist points p 1 , . . . , p m   such that the blow up of P n   at p 1 , . . . , p m   carries a constant scalar curvature Kähler metric.
The above corollary is optimal in the number of points because, for m ≤ n   , Bl p 1 , … , p m P n   does not satisfy the Matsushima-Lichnerowicz obstruction. Observe also that Bl p 1 , … , p n + 1 P n   still has vanishing holomorphic vector fields.
On the other hand it is well known that on P n   , n + 2   points forming a projective frame are enough to kill all holomorphic vector fields when blowing up, and indeed wecan prove that this condition also guarantees the existence of Kähler constant scalar curvature metric.
Corollary 1.2. Given p 1 , . . . , p m   points in P n   such that p 1 , . . . , p n + 2   form a projective frame, M ~ = Bl p 1 , … , p m   has a Kähler constant scalar curvature metric and no holomorphic vector fields. Moreover p 1 , . . . , p n + 2   can vary in a dense open subset of ( P n ) n + 2   and p n + 3 , . . . , p m   are arbitrary.
When M = P n × M 0   where M 0   is any manifold with no nontrivial vanishing holomorphic vector field, all the above result carry to M   . In particular :
Corollary 1.3. Assume that ( M 0 , ω 0 )   is a Kähler manifold with constant scalar curvature and without nontrivial vanishing holomorphic vector fields and define M = P n × M 0   , then d = n 2 + 2 n   and m 0 ≤ 2 n ( n + 1 )   . Moreover, the blow up of P n × M 0   at m ≥ n + 1   points has a constant scalar curvature Kähler metric.
As already mentioned, when n = 1   and M 0 = Σ g   is a Riemann surface of genus g   greater than 1   , the existence of zero scalar curvature Kähler metrics on blow ups of P 1 × Σ g   is due to LeBrun and Singer [12] .
Finally, we have also considered the case where M = P n × P m   , for which we have obtained the :
Corollary 1.4. Suppose n ≤ m   . The blow up of P n × P m   at k ≥ m + 2   special points has a constant scalar curvature Kähler metric. Moreover, if the projections of these points to the two factors contain projective frames, these points can move in a dense open subset of ( P n × P m ) m + 2   and the other ones are arbitrary.
The above results for n = 2   translate directly in the following :
Corollary 1.5. For any m ≥ 2   there exist points p 1 , … , p m ∈ P 1 × P 1   such that Bl p 1 , … , p m P 1 × P 1   has a constant scalar curvature Kähler metric.
Rollin-Singer in [16] have already found such constant scalar curvature Kähler metrics on the blow up of P 1 × P 1   at m ≥ 6   points. It is worth mentioning that, for m = 2   , Bl p 1 , p 2 P 1 × P 1   still has nontrivial vanishing holomorphic vector fields. When m ≤ 7   the construction is performed using the symmetries of the manifold, working equivariantly.

2 Weighted spaces

In this section, we describe weighted spaces on the noncompact manifold M   with m   points removed, as well as weighted spaces on the noncompact manifolds N   . We also introduce notations which will be used in the next sections.
For all r > 0   , we agree that
B r : = { z ∈ C n : | z | < r } , B r * : = { z ∈ C n : 0 < | z | ≤ r } , C r : = { z ∈ C n : | z | ≥ r } (2)
Assume that k ∈ N   and α ∈ ( 0 , 1 )   are fixed. Given r ¯ > 0   and a function v ∈ C l o c k , α ( B r ¯ * )   , we define ∥ v ∥ C δ k , α ( B r ¯ * ) : = sup 0 < r ≤ r ¯ r − δ ∥ v ( r ⋅ ) ∥ C k , α ( B ¯ 1 − B 1 / 2 ) .   and, for any function v ∈ C l o c k , α ( C r ¯ )   , we define ∥ v ∥ C δ k , α ( C r ¯ ) : = sup r ≥ r ¯ r − δ ∥ v ( r ⋅ ) ∥ C k , α ( B ¯ 2 − B 1 ) .   Assume that we are given ( M , ω M )   , a Kähler manifold with Kähler form ω M   and that we are also given m   distinct points p 1 , … , p m ∈ M   . By definition, near p ℓ   , the manifold M   is biholomorphic to a neighborhood of 0   in C n   . In particular, we can choose complex coordinates z : = ( z 1 , … , z n )   in a neighborhood of 0   in C n   , to parameterize a neighborhood of p ℓ   in M   . In order to distinguish between the different neighborhoods and coordinate systems, we agree that, for all r   small enough, say r ∈ ( 0 , r 0 )   , B ℓ , r   (resp. B ℓ , r *   ) denotes the ball (resp. the punctured ball) of radius r   in the above defined coordinates z   parameterizing a fixed neighborhood B ℓ , r 0   of p ℓ   . Furthermore, it follows from [5] , that there is no loss of generality in assuming that the Kähler form ω M   can be expended as
ω M : = i ∂ ∂ ¯ ( 1 2 | z | 2 + φ ℓ ) , (3)
where the function φ ℓ ∈ C 4 3 , α ( B ℓ , r 0 * )   . This in particular implies that, in these coordinates, the standard metric on C n   and the metric induced by ω M   agree up to order 2   .
For all r ∈ ( 0 , r 0 )   , we define
M r : = M − ∪ ℓ B ℓ , r . (4)
The weighted space for functions defined on the noncompact manifold
M * : = M − { p ℓ , : ℓ = 1 , … , m } . (5)
is then defined as the set of functions whose decay or blow up near any p ℓ   is controlled by a power of the distance to p ℓ   . More precisely, we have the :
Definition 2.1. Given k ∈ N   , α ∈ ( 0 , 1 )   and δ ∈ R   , we define the weighted space C δ k , α ( M * )   to be the space of functions w ∈ C l o c k , α ( M * )   for which the following norm is finite ∥ w ∥ C δ k , α ( M * ) : = ∥ w ∥ C k , α ( M r 0 / 2 ) + sup ℓ ∥ w | B ℓ , r 0 * ∥ C δ k , α ( B ℓ , r 0 * ) .  
We now turn to the description of weighted space on ( N , η )   , the blow up at the origin of C n   endowed with the Burns-Simanca metric. Away from the exceptional divisor, the Kähler form η   is given by η = i ∂ ∂ ¯ f n ( | v | 2 )   where v = ( v 1 , … , v n )   are complex coordinates in C n − { 0 }   and where the function s ⟶ f n ( s )   is a solution of the ordinary differential equation s 2 ( s ∂ s f n ) n − 1 ∂ s 2 f n + ( n − 1 ) s ∂ s f n − ( n − 2 ) = 0   which satisfies f n ∼ log s   near 0   . It turns out that, when n = 2   , the function P 2   is explicitly given by f 2 ( s ) = log s + λ s + c   where λ > 0   and c ∈ R   , while in dimension n ≥ 3   , even though there is no explicit formula for f n   we have the following :
Lemma 2.1. Assume that n ≥ 3   . Then the function f n   can be expanded as f n ( s ) = λ s + c − λ 2 − n s 2 − n + O ( s 1 − n )   for s > 1   , where λ > 0   and c ∈ R   .
Proof : This result follows from the analysis done in [17] even though this is not explicitly stated. In fact, we define the function ζ   by s ζ : = s ∂ s f n − 1   . Direct computation shows that ζ   solves ( 1 + s ζ ) n − 1 s 2 ∂ s ζ = ( 1 + s ζ ) n − 1 − 1 − ( n − 1 ) s ζ   If in addition we take ζ ( 0 ) = 1   , then ∂ s ζ   remains positive and one can check that ζ   is well defined for all time and converges to some positive constant λ   , as s   tends to ∞   .
This immediately implies that s ∂ s f n = λ s + O ( 1 )   at infinity. The expansion then follows easily. ■   Changing variables u : = 2 λ v   , we see from the previous Lemma that the Kähler form η   can be expanded near ∞   as
η = i ∂ ∂ ¯ ( 1 2 | u | 2 + log | u | 2 + φ ~ ) , (6)
in dimension n = 2   and as
η = i ∂ ∂ ¯ ( 1 2 | u | 2 − 2 n − 2 | u | 4 − 2 n + φ ~ ) , (7)
in dimension n ≥ 3   , where φ ~ ∈ C 3 − 2 n 3 , α ( C 1 )   for all n ≥ 2   . The property which will be crucial for our construction is that, in these two expansions, the coefficient in from of log | u | 2   , in dimension n = 2   and the coefficients in front of | u | 4 − 2 n   are not zero.
For all R > 1   , we define
N R : = N − C R . (8)
We are now in a position to define weighted spaces on the noncompact complete manifold N   . This time, we are interested in functions which decay (or blow up) near infinity at a rate which is controlled by a power of the distance to a fixed point in N   .
More precisely, we have the :
Definition 2.2. Given k ∈ N   , α ∈ ( 0 , 1 )   and δ ∈ R   , we define the weighted space C δ k , α ( N )   to be the space of functions w ∈ C l o c k , α ( N )   for which the following norm is finite ∥ w ∥ C δ k , α ( N ) : = ∥ w ∥ C k , α ( N 2 ) + ∥ w | C 1 ∥ C δ k , α ( C 1 ) .  

3 The geometry of the equation

Recall that ( M , ω M )   is a compact Kähler manifold of complex dimension n   . We will indicate by g M   the Riemannian metric induced by ω M   , Ric M   its Ricci tensor, ρ M   the corresponding Ricci form, and s ( ω M ) = s M   its scalar curvature.
Following LeBrun-Simanca [12] , we want to understand the behavior of the scalar curvature of the metric under deformations of the form ω : = ω M + i ∂ ∂ ¯ φ   where φ   a function defined on M   . The following result is proven, for example in [12]  :
Proposition 3.1. The scalar curvature of ω   can be expanded in terms of φ   as s ( ω ) = s ( ω M ) − 1 2 Δ M 2 φ − Ric M ⋅ ∇ M 2 φ + Q M ( φ ) ,   where Q M   collects all the nonlinear terms and where all operators in the right hand side of this identity are computed with respect to the Kähler form ω M   .
The operator
L M : = − 1 2 Δ M 2 − Ric M ⋅ ∇ M 2 , (9)
which acts on functions is the one which has been defined in the introduction.
For a general Kähler metric it can be very difficult to analyze these operators. Nevertheless geometry comes to the rescue at a constant scalar curvature metric. Indeed, in this case we have
L M = ( ∂ ¯ ∂ # ) ◆ ∂ ¯ ∂ # , (10)
where ∂ # φ   denotes the ( 1 , 0 )   -part of the g M   -gradient of φ   . Using this result, the key observation of LeBrun-Simanca is that to any element φ   of Ker L M   one can associate a holomorphic vector field, namely ∂ # φ   , which must vanish somewhere on M   . This is the crucial relation between constant scalar curvature (or extremal) metrics and the space of holomorphic vector fields H ( M )   (and, in turn, the automorphism group Aut ( M )   ).
We recall the following important result, which has been mentioned in the introduction and which helps to understand the kernel of the operator L M   :
Theorem 3.1. [12] Assume that ( M , ω M )   is a compact Kähler manifold of complex dimension n   , then the space of the elements of the kernel of L M   whose mean over M   is 0   , is in one to one correspondence with the space of holomorphic vector fields which vanish on M   .
The previous considerations extend to ( N , η )   and this implies the following important result of [8] which states that there are no elements in the kernel of the operator L N   which decay at infinity.
Proposition 3.2. [8] There are no nontrivial solution to L N φ = 0   , which belong to C δ 4 , α ( N )   , for some δ < 0   .
Proof : Assume that L N φ = 0   and that φ ∈ C δ 4 , α ( N )   , for some δ < 0   . Then ∂ # φ   is an holomorphic vector field which tends to 0   at infinity. Using Hartogs theorem, the restriction of ∂ # φ   to C 1   can be extended to a holomorphic vector field on C n   . Since this vector field decays at infinity, it has to be identically equal to 0   . This implies that ∂ # φ   is identically equal to 0   on C 1   . However φ   being a real valued function, this implies that ∂ φ = ∂ ¯ φ = 0   in C 1   . Hence the function φ   is constant in C 1   and decays at infinity. This implies that φ   is identically equal to 0   in C 1   and satisfies L N φ = 0   in N   . Now, use the unique continuation theorem to conclude that φ   is identically equal to 0   in N   . ■  

4 Mapping properties

4.1 Analysis of the operator defined on M *  

The results we want to obtain are based on the fact that, near each p ℓ   and in the above defined coordinates, the metric on M   is asymptotic to the Euclidean metric. This implies that, in each B ℓ , r 0   the operator L M   is close to the operator
L 0 : = − 1 2 Δ 0 2 , (11)
where Δ 0   denotes the Laplacian in C n   when endowed with the standard Kähler form.
In particular these two operators have the same indicial roots at p ℓ   . This set of indicial roots is given by Z − { 5 − 2 n , … , − 1 }   when n ≥ 3   and is given by Z   when n = 2   . The mapping properties of L M   when defined between weighed spaces follows from the general theory explained in [14] and this will allow us to construct some right inverse for L M   .
We define in B ℓ , r 0 *   the function G ℓ   by G ℓ ( z ) = − log | z | 2 when n = 2 and G ℓ ( z ) = | z | 4 − 2 n when n ≥ 3 .   Observe that, unless the metric g M   is the Euclidean metric, these functions are not solutions of the homogeneous equation associated to L M   , however they can be perturbed into G ~ ℓ   solutions of the homogeneous problem L M G ~ ℓ = 0   . Indeed, reducing r 0   if this is necessary, we have the following :
Lemma 4.1. There exist functions G ~ ℓ   which are solutions of L M G ~ ℓ = 0   in B ℓ * ( r 0 )   and which are asymptotic to G ℓ   in the sense that G ~ ℓ − G ℓ ∈ C 6 − 2 n 4 , α ( B ℓ , r 0 * )   when n ≥ 4   and G ~ ℓ − G ℓ ∈ C δ 4 , α ( B ℓ , r 0 * )   for any δ < 6 − 2 n   when n = 2 , 3   .
Proof : Observe that L M G ℓ = ( L M − L 0 ) G ℓ   and, thanks to the expansion given in ( 3 ), we conclude that L M G ℓ ∈ C 2 − 2 n 0 , α ( B ℓ , r 0 * )   . The result then follows from the analysis of §6 in [2] . Observe that, as long as 6 − 2 n   is not an indicial root of L M   we conclude that G ~ ℓ − G ℓ ∈ C 6 − 2 n 4 , α ( B ℓ , r 0 * )   while, if 6 − 2 n   is an indicial root of L M   , we only conclude that G ~ ℓ − G ℓ ∈ C δ 4 , α ( B ℓ , r 0 * )   for any δ < 6 − 2 n   . ■   Observe that there is no uniqueness of G ~ ℓ   . With the functions G ~ ℓ   at hand, we define the following deficiency spaces D 0 : = Span { χ 1 , … , χ m } , and D 1 : = Span { χ 1 G ~ 1 , … , χ m G ~ m } ,   where χ ℓ   is a cutoff function which is identically equal to 1   in B ℓ , r 0 / 2   and identically equal to 0   in M − B ℓ , r 0   .
When n ≥ 3   , we fix δ ∈ ( 4 − 2 n , 0 )   and define the operator
L δ : ( C δ 4 , α ( M * ) ⊕ D 1 ) × R ⟶ C δ − 4 0 , α ( M * )
( φ , β ) ↦ L M φ + β ,
Whereas, when n = 2   , we fix δ ∈ ( 0 , 1 )   and define the operator
L δ : ( C δ 4 , α ( M * ) ⊕ D 0 ⊕ D 1 ) × R ⟶ C δ − 4 0 , α ( M * )
( φ , β ) ↦ L M φ + β ,
To keep notations short, it will be convenient to set D : = D 1   when n ≥ 3   and D : = D 0 ⊕ D 1   when n = 2   . The main result of this section reads :
Proposition 4.1. Assume that the points p 1 , … , p m ∈ M   are chosen so that ℭ 1 = d   . Then, the operator L δ   defined above is surjective (and has a ( m + 1 )   -dimensional kernel).
Proof : The proof of this result follows from the general theorem described in [14] , however, we choose here to describe an almost self contained proof. Recall that the kernel of L M   is spanned by the functions ξ 0 ≡ 1 , ξ 1 , … , ξ d   and, by assumption ξ j   , for j = 1 , … , d   have mean 0   . We use the fact that L M   is self adjoint and hence, for ψ ∈ L 1 ( M )   , the problem L M φ = ψ   is solvable if and only if ψ   satisfies ∫ M ψ ξ j = 0   for j = 0 , … , d   .
Observe that C δ − 4 0 , α ( M * ) ⊂ L 1 ( M )   when δ > 4 − 2 n   . Now, given ψ ∈ L 1 ( M )   , we choose a 0 = ∮ M ψ   and, since ℭ 1 = d   , we also choose a 1 , … , a d ∈ R   solution of the system ∀ j = 1 , … , d , ∫ M ψ ξ j = ∑ ℓ = 1 m a ℓ ξ j ( p ℓ )   Then, the problem L M φ = ψ − a 0 − ∑ ℓ = 1 m a ℓ δ p ℓ   is solvable in L 1 ( M )   for all p ∈ [ 1 , n n − 1 )   and uniqueness follows if we impose that φ   is orthogonal to ξ 0 , … , ξ d   . To complete the proof, we invoke regularity theory which implies that φ ∈ C δ 4 , α ( M * ) ⊕ D 1   when n ≥ 3   and C δ 4 , α ( M * ) ⊕ D 0 ⊕ D 1   when n = 2   . The estimate of the dimension of the kernel is left to the reader since it will not be used in the paper. ■   Observe that, when we will solve L φ + β = ψ   , the constant α   is determined by β = ∮ M ψ .  

4.2 Analysis of the operator defined on N  

As above, we use the fact that the metric on N   is asymptotic to Euclidean metrics, as explained in ( 6 ) and ( 7 ). This implies that, in C 1   , the operator L N   is close to the operator L 0   and they have the same set of indicial roots which is given by Z − { 5 − 2 n , … , − 1 }   when n ≥ 3   and is given by Z   when n = 2   . Given δ ∈ R   , we define the operator
L ~ δ : C δ 4 , α ( N ) ⟶ C δ − 4 0 , α ( N )
φ ↦ L N φ
and recall the following result from [2] 
Proposition 4.2. Assume that δ ∈ ( 0 , 1 )   . Then the operator L ~ δ   defined above is surjective and has a one dimensional kernel spanned by the constant function.
Proof : The result of Proposition  3.2 precisely states that the operator L ~ δ ′   is injective when δ ′ < 0   . This implies that the operator L ~ δ   is surjective when δ > 4 − 2 n   . When δ ∈ ( 0 , 1 )   , this also implies that the operator L ~ δ   has a one dimensional kernel, spanned by the constant function. ■  

4.3 Bi-harmonic extensions

We end up this section by the following simple result which follows at once from the application of the maximum principle and whose proof can be found in [2] .
Proposition 4.3. Assume that n ≥ 2   . Given h ∈ C 4 , α ( ∂ B 1 )   , k ∈ C 2 , α ( ∂ B 1 )   there exists a function H h , k i ∈ C 4 , α ( B 1 )   such that L 0 H h , k i = 0 in B 1   with H h , k i = h and Δ 0 H h , k i = k on ∂ B 1 .   Moreover, ∥ H h , k i ∥ C 4 , α ( B 1 ) ≤ c ( ∥ h ∥ C 4 , α ( ∂ B 1 ) + ∥ k ∥ C 2 , α ( ∂ B 1 ) )  
We will also need the following result which differs slightly from what we have already used in [2] . The rational being that there exists a bi-harmonic extension of the data ( h , k )   which is defined on the complement of the unit ball and decays at infinity. Moreover, this function is bounded by a constant times the distance to the origin to the power 4 − 2 n   (when n ≥ 3   ). In the case where the function k   is assumed to have mean 0   , then the rate of decay can be improved and estimated as the distance to the origin to the power 3 − 2 n   :
Proposition 4.4. Assume that n ≥ 2   . Given h ∈ C 4 , α ( ∂ B 1 )   , k ∈ C 2 , α ( ∂ B 1 )   such that ∫ ∂ B 1 k = 0   there exists a function H h , k o ∈ C 3 − 2 n 4 , α ( C 1 )   such that L 0 H h , k o = 0 , in C 1   with H h o = h and Δ 0 H h , k o = k on ∂ B 1 .   Moreover, ∥ H h , k o ∥ C 3 − 2 n 4 , α ( C 1 ) ≤ c ( ∥ h ∥ C 4 , α ( ∂ B 1 ) + ∥ k ∥ C 2 , α ( ∂ B 1 ) )  

4.4 Perturbation of ω M  

We consider the Kähler form
ω = ω M + i ∂ ∂ ¯ φ . (12)
The scalar curvature of ω   is given by s ( ω ) = s ( ω M ) + L M φ + Q M ( φ ) ,   where the operator L M   has been defined in ( 9 ) and where Q M   collects all the nonlinear terms. The structure of Q M   is quite complicated however, it follows from the explicit computation of the Ricci curvature that, near p ℓ   , the nonlinear operator Q M   can be decomposed as
Q M ( φ ) = ∑ j B ℓ , j ( ∇ 4 φ , ∇ 2 φ ) Q ℓ , j ( ∇ 2 φ ) + ∑ j ′ B ~ ℓ , j ′ ( ∇ 3 φ , ∇ 3 φ ) Q ~ ℓ , j ′ ( ∇ 2 φ ) , (13)
where the B ℓ , j   and B ~ ℓ , j ′   are bilinear forms with bounded coefficients and the sums are finite. In addition, identifying the Hessian of φ   with a symmetric matrix, one can check that the nonlinear operators Q ℓ , j   and Q ~ ℓ , j ′   are uniformly bounded and uniformly Lipschitz over the set of symmetric matrices whose entries are bounded by some fixed constant c 0   . More precisely, they satisfy
| Q ℓ , j ( A ) | + | Q ~ ℓ , j ′ ( A ) | ≤ c , (14)
and
| Q ℓ , j ( A ) − Q ℓ , j ( A ′ ) | ≤ c | A − A ′ | , (15)
for all symmetric matrices A   and A ′   satisfying | A | ≤ c 0   and | A ′ | ≤ c 0   .
As usual, we denote by ξ 0 , … , ξ d   , independent functions which span the kernel of L M   . We agree that ξ 0 ≡ 1   is the constant function and that all functions ξ 1 , … , ξ d   are normalized to have mean 0   . Now, assume that we are given a 0 , a 1 , … , a m > 0   such that there exists a solution of L M H a = a 0 − c n ∑ ℓ a ℓ δ p ℓ   where the constant c n   is defined by c n : = 4 ( n − 2 ) ( n − 1 ) | S 2 n − 1 | when n ≥ 3 and c 2 : = 2 | S 3 |   Observe that such a function H a   exists if and only if a 0   is given by a 0 = c n ∑ ℓ = 1 m a ℓ   and the coefficients a 1 , … , a m   are solutions of the system ∑ ℓ = 1 m a ℓ ξ j ( p ℓ ) = 0   for j = 1 , … , d   . This amounts to ask that the vector ( a 1 , … , a m )   has positive entries and is in the kernel of the matrix M ( p 1 , … , p m )   . This is precisely the origin of the condition ℭ 2 ≠ 0   . It is not hard to check that :
Lemma 4.2. Near each p ℓ   , the function H a   satisfies H a + a ℓ G ℓ + b ℓ ∈ C 1 4 , α ( B ℓ , r 0 * )   for some constant b ℓ ∈ R   .
We fix
r ɛ : = ɛ 2 n − 1 2 n + 1 (16)
Assume that we are given h ℓ ∈ C 4 , α ( ∂ B 1 )   and k ℓ ∈ C 2 , α ( ∂ B 1 )   , ℓ = 1 , … , m   , satisfying
∥ h ℓ ∥ C 4 , α ( ∂ B 1 ) + ∥ k ℓ ∥ C 2 , α ( ∂ B 1 ) ≤ κ r ɛ 4 , (17)
where κ > 0   will be fixed later on. Further assume that
∫ ∂ B 1 k ℓ = 0 (18)
We define
H h , k : = ∑ ℓ χ ℓ H h ℓ , k ℓ o ( ⋅ / r ɛ ) , (19)
where h   and k   stand for ( h 1 , … , h m )   and ( k 1 , … , k m )   and where the cutoff functions χ ℓ   are identically equal to 1   in B ℓ , r 0 / 2   and identically equal to 0   in M − B ℓ , r 0   .
We fix δ ∈ ( 4 − 2 n , 5 − 2 n )   . The space C δ 4 , α ( M r ɛ ) ⊕ D   (resp. C δ − 4 0 , α ( M r ɛ )   ) is just defined as the space of restrictions of functions in C δ 4 , α ( M * ) ⊕ D   (resp. C δ − 4 0 , α ( M * )   ) to M r ɛ   , endowed with the induced norm.
We would like to solve the equation s ( ω M + i ∂ ∂ ¯ ( ɛ 2 n − 2 H a + H h , k + φ ) ) = s ( ω M ) − ɛ 2 n − 2 a 0 + β ,   where φ ∈ C δ 4 , α ( M r ɛ ) ⊕ D   and β ∈ R   have to be determined. This amounts to solve the equation
L M ( H h , k + φ ) + Q M ( ɛ 2 n − 2 H a + H h , k + φ ) = β , (20)
in M r ɛ   . We consider an extension (linear) operator ℰ ɛ : C δ − 4 0 , α ( M r ɛ ) ⟶ C δ − 4 0 , α ( M * ) ,   such that ℰ ɛ ψ = ψ   in M r ɛ / 2   , ℰ ɛ ψ   is supported in M r ɛ / 2   and ∥ ℰ ɛ ∥ ≤ 2   .
The equation we would like to solve can be rewritten as
L M φ = − ℰ ɛ ( ( L M H h , k + Q M ( ɛ 2 n − 2 H a + H h , k + φ ) ) | M r ɛ ) + β . (21)
where β   is determined by β = ∮ ℰ ɛ ( ( L M H h , k + Q M ( ɛ 2 n − 2 H a + H h , k + φ ) ) | M r ɛ )   Granted these notations, we prove the :
Lemma 4.3. We fix δ ∈ ( 4 − 2 n , 5 − 2 n )   . There exists c > 0   , c ¯ κ > 0   and there exists ɛ 0 > 0   such that, for all ɛ ∈ ( 0 , ɛ 0 )   ∥ L M H h , k ∥ C δ − 4 0 , α ( M r ɛ ) ≤ c ¯ κ r ɛ 2 n + 1 ,   and ∥ Q M ( ɛ 2 n − 2 H a + H h , k ) ∥ C δ − 4 0 , α ( M r ɛ ) ≤ c ɛ 4 n − 4 r ɛ 6 − 4 n − δ ,   where the norms are the restrictions of the weighted norms in Definition  2.1 to functions defined in M r ɛ   . In addition, we have and ∫ M r ɛ | L M H h , k + Q M ( ɛ 2 n − 2 H a + H h , k ) | ≤ c ¯ κ ( r ɛ 2 n + 1 + ɛ 4 n − 4 r ɛ 2 − 2 n ) .  
Proof : The proof of these estimates can be obtained as in [2] . We briefly recall it here.
First, we use the result of Proposition  4.4 to estimate
∥ H h , k ∥ C 3 − 2 n 4 , α ( M r ɛ ) ≤ c κ r ɛ 2 n + 1 . (22)
Now observe that H h , k = 0   in M r 0   and L 0 H h , k = 0   in each B ℓ , r 0 / 2 − B ℓ , r ɛ   , hence L M H h , k = ( L M − L 0 ) H h , k   in this set. Now, we use the expansion ( 3 ) which reflects the fact that the metric induced by ω M   is in each B ℓ , r 0   asymptotic to the Euclidean metric. This implies that there exists a constant c > 0   such that ∥ ( L M − L 0 ) φ ∥ C δ − 4 0 , α ( B ℓ , r 0 * ) ≤ c ∥ ∇ 2 φ ∥ C δ − 4 2 , α ( B ℓ , r 0 * ) .   Collecting these, and taking into account the effect of the cutoff functions, we conclude that ∥ L M H h , k ∥ C δ − 4 0 , α ( M r ɛ ) ≤ c κ ( r ɛ 2 n + 1 + r ɛ 6 − δ ) .   The first estimate follows at once from the fact that 6 − δ > 2 n + 1   .
The second estimate follows from the structure of the nonlinear operator Q M   described in ( 13 ) together with the estimate ∥ ∇ 2 H a ∥ C 2 − 2 n 2 , α ( M r ɛ ) ≤ c .   and also ( 22 ). Roughly speaking this estimate reflects the fact the most important nonlinear terms in the expression of Q M   are of the form B ℓ , j ( ∇ 4 φ , ∇ 2 φ )   and B ~ ℓ , j ′ ( ∇ 3 φ , ∇ 3 φ )   provided the second derivatives of φ   remain bounded (which is precisely our case). The proof of the second estimate follows easily from these considerations. The third estimate is easy to obtain and left to the reader. ■   We will also need the following result which will be useful for the contraction mapping argument we will apply :
Lemma 4.4. We fix δ ∈ ( 4 − 2 n , 5 − 2 n )   and c ~ κ > 0   . There exists a constant c ^ κ > 0   such that, for all φ , φ ′ ∈ C δ 4 , α ( M r ɛ )   satisfying ∥ φ ∥ C δ 4 , α ( M r ɛ ) ⊕ D + ∥ φ ′ ∥ C δ 4 , α ( M r ɛ ) ⊕ D ≤ c ~ κ ( r ɛ 2 n + 1 + ɛ 4 n − 4 r ɛ 6 − 4 n − δ ) ,   we have
∥ Q M ( ɛ 2 n − 2 H a + H h , k + φ ) − Q M ( ɛ 2 n − 2 H a + H h , k + φ ′ ) ∥ C δ − 4 0 , α ( M r ɛ )
≤ c ^ κ ɛ 2 n − 2 r ɛ 6 − 4 n − δ ∥ φ − φ ′ ∥ C δ 4 , α ( M r ɛ ) ⊕ D ,
and we also have
∫ M ( r ɛ ) | Q M ( ɛ 2 n − 2 H a + H h , k + φ ) − Q M ( ɛ 2 n − 2 H a + H h , k + φ ′ ) |
≤ c ^ κ ɛ 2 n − 2 r ɛ 2 − 2 n ∥ φ − φ ′ ∥ C δ 4 , α ( M r ɛ ) ⊕ D .
The important observation is that lim ɛ → 0 ɛ 2 n − 2 r ɛ 6 − 4 n − δ = lim ɛ → 0 ɛ 2 n − 2 r ɛ 2 − 2 n = 0 .   Using the results of §5.2, we reduce the solvability of the equation ( 21 ) to a fixed point problem in { φ ∈ C δ 4 , α ( M r ɛ ) ⊕ D : ∥ φ ∥ C δ 4 , α ( M r ɛ ) ⊕ D ≤ c κ ( r ɛ 2 n + 1 + ɛ 4 n − 4 r ɛ 6 − 4 n − δ ) } ,   where the constant c κ > 0   is fixed large enough. And in fact, using Lemma  4.3 and Lemma  4.4 it is easy to check that there exists a fixed point solution of ( 21 ) provided ɛ   is chosen small enough, say ɛ ∈ ( 0 , ɛ 0 )   by applying a fixed point theorem for contraction mapping.
Hence we have obtained the :
Proposition 4.5. Assume that n ≥ 2   and choose δ ∈ ( 4 − 2 n , 5 − 2 n )   . Given κ > 0   , there exists c ˘ κ > 0   and ɛ κ > 0   such that, for all ɛ ∈ ( 0 , ɛ κ )   , for all h ℓ ∈ C 4 , α ( ∂ B 1 )   and all k ℓ ∈ C 2 , α ( ∂ B 1 )   satisfying ( 17 ), there exists a Kähler form ω h , k = ω M + i ∂ ∂ ¯ φ h , k ,   defined on M r ɛ   , which has constant scalar curvature equal. Moreover,
∥ φ h , k | B ℓ , 2 r ɛ − B ℓ , r ɛ ( r ɛ ⋅ ) − ɛ 2 n − 2 a ℓ G ℓ ( r ɛ ⋅ ) − ɛ 2 n − 2 b ℓ − H h ℓ , k ℓ o ∥ C 4 , α ( B 2 − B 1 )
≤ c ˘ κ ( r ɛ 5 + ɛ 4 n − 4 r ɛ 10 − 6 n − δ ) ,
and | s ( ω h , k ) − s ( ω M ) + ɛ 2 n − 2 a 0 | ≤ c ˘ κ ( ɛ 2 n − 1 + ɛ 4 n − 4 r ɛ 2 − 2 n ) .  
Recall that φ ℓ   is the function which appears in ( 3 ), the expansion of the Kähler potential ω M   at p ℓ   so that ω h , k = i ∂ ∂ ¯ ( 1 2 | z | 2 + φ ℓ + φ h , k ) ,   in each B ℓ , 2 r ɛ − B ℓ , r ɛ   . We define, ψ ℓ o : = ( φ ℓ + φ h , k ) ( r ɛ ⋅ ) + ɛ 2 n − 2 b ℓ ,   when n ≥ 3   and ψ ℓ o : = ( φ ℓ + φ h , k ) ( r ɛ ⋅ ) + ɛ 2 n − 2 b ℓ − ɛ 2 a ℓ log r ɛ 2 ,   when n = 2   . Observe that, locally, the Kähler potential is not uniquely defined and addition of constant does not modify the corresponding Kähler form.
Thanks to the previous analysis, the following expansions are valid in each B 2 − B 1   .
Lemma 4.5. The following expansions holds
ψ ℓ o ( z ) = − a ℓ ɛ 2 n − 2 r ɛ 4 − 2 n | z | 4 − 2 n + H h ℓ , k ℓ o ( z ) + O ( r ɛ 4 ) + O κ ( r ɛ 5 + ɛ 4 n − 4 r ɛ 10 − 6 n − δ ) , (23)
when n ≥ 3   and
ψ ℓ o ( z ) = a ℓ ɛ 2 log | z | 2 + H h ℓ , k ℓ o ( z ) + O ( r ɛ 4 ) + O κ ( r ɛ 5 + ɛ 4 r ɛ − 2 − δ ) , (24)
when n = 2   .
Similar expansions are valid for the partial derivatives of these functions. The subscript κ   in O κ ( A )   is meant to recall that, as ɛ   tends to 0   , this quantity is bounded by a constant, depending on κ   , times A   , while O ( A )   is bounded by a constant (independent of κ   ) times A   .
The key point is that r ɛ 5 + ɛ 4 n − 4 r ɛ 10 − 6 n − δ ≪ r ɛ 4   as ɛ   tends to 0   (here we have to assume that δ ∈ ( 0 , 2 / 3 )   , when n = 2   ) and this implies that, in the expansion of the boundary data of the potential φ h , k   , the part corresponding to the solution of the nonlinear problem ( 20 ) is much smaller than r ɛ 4   , the characteristic size of the boundary data h   and k   .
It should be clear that the Kähler form provided by Proposition  4.5 depends continuously on the data (such as h   and k   ).

4.5 Perturbation of η  

Given a ~ > 0   , we now consider on N   , the Kähler form η a ~ : = ( 2 2 − n a ~ ) 1 n − 1 η   which, after a change of variable, can be expanded in C 1   as
η a ~ = i ∂ ∂ ¯ ( 1 2 | u | 2 − a ~ | u | 4 − 2 n + φ ~ a ~ ) , (25)
when n ≥ 3   and
η a ~ = i ∂ ∂ ¯ ( 1 2 | u | 2 + a ~ log | u | 2 + φ ~ a ~ ) , (26)
when n = 2   , where the function φ a ~ ∈ C 3 − 2 n 3 , α ( C 1 )   enjoys properties similar to the one enjoyed by φ ~   . Everything being uniform in a ~   as long as this parameter remains both bounded from above and bounded away from 0   . In fact we will need to choose a ~   so that 1 2 inf ℓ a ℓ < a ~ < 2 sup ℓ a ℓ ,   where the coefficients a ℓ   are the one which have been defined at the beginning of the previous section.
As in the previous section, we consider the perturbed Kähler form η ~ = η a ~ + i ∂ ∂ ¯ φ ,   The scalar curvature of η ~   is given by s ( η ~ ) = L N φ + Q N ( φ ) ,   since the scalar curvature of η   is identically equal to 0   . Even though this is not explicit in the notations, the operators L N   and Q N   both depend on t   . Again, the structure of the nonlinear operator Q N   is quite complicated but, in C 1   , it enjoys a decomposition similar to the ones described in ( 13 ) and the estimates ( 14 ) and ( 15 ) hold uniformly in each C 1   and in a ~   , as long as a ~   remains bounded from above and bounded away from 0   .
We define
R ɛ : = r ɛ ɛ (27)
Given h ~ ∈ C 4 , α ( ∂ B 1 )   and k ~ ∈ C 2 , α ( ∂ B 1 )   satisfying
∥ h ~ ∥ C 4 , α ( ∂ B 1 ) + ∥ k ~ ∥ C 2 , α ( ∂ B 1 ) ≤ κ R ɛ 3 − 2 n , (28)
where κ > 0   will be fixed later on, we define
H ~ h ~ , k ~ : = χ ~ ( H h ~ , k ~ i ( ⋅ / R ɛ ) − H h ~ , k ~ i ( 0 ) ) + H h ~ , k ~ i ( 0 ) , (29)
where χ ~   are cutoff functions which are identically equal to 1   in C 2   and identically equal to 0   in N 1   .
We fix δ ∈ ( 0 , 1 )   . The space C δ 4 , α ( N R ɛ )   (resp. C δ − 4 0 , α ( N R ɛ )   ) is defined as the space of restrictions of functions in C δ 4 , α ( N )   (resp. C δ − 4 0 , α ( N )   ) to N R ɛ   , endowed with the induced norm.
We would like to solve the equation L N ( H ~ h ~ ℓ , k ~ ℓ + φ ) + Q N ( H ~ h ~ ℓ , k ~ ℓ + φ ) = ɛ 2 s   where s ∈ R   satisfies | s | ≤ 1 + | s ( ω M ) |   .
We consider an extension (linear) operator ℰ ~ ɛ : C δ − 4 0 , α ( N R ɛ ) ⟶ C δ − 4 0 , α ( N ) ,   such that ℰ ~ ɛ ψ = ψ   in N R ɛ   , ℰ ~ ɛ ψ   is supported in N 2 R ɛ   and ∥ ℰ ~ ɛ ∥ ≤ 2   . The equation we would like to solve can be rewritten as
L N φ = ℰ ~ ɛ ( ( ɛ 2 s − L N H ~ h ~ , k ~ − Q N ( H ~ h ~ , k ~ + φ ) ) | N R ɛ ) . (30)
We prove the :
Lemma 4.6. Assume that δ ∈ ( 0 , 1 )   . There exists c , c ¯ κ > 0   such that, for all ɛ ∈ ( 0 , ɛ 0 )   ∥ L N H ~ h ~ , k ~ ∥ C δ − 4 0 , α ( N R ɛ ) ≤ c ¯ κ R ɛ 2 − 2 n , ∥ ɛ 2 s ∥ C δ − 4 0 , α ( N R ɛ ) ≤ c R ɛ 3 − 2 n − δ ,   and ∥ Q N ( H ~ h ~ , k ~ ) ∥ C δ − 4 0 , α ( N R ɛ ) ≤ c ¯ κ R ɛ 4 − 4 n .  
Proof : Again, the proof of these estimates can be obtained as in [2] . First, we use the result of Proposition  4.3 to estimate
∥ ∇ 2 H ~ h ~ , k ~ ∥ C 0 2 , α ( N R ɛ − N 2 ) ≤ c κ R ɛ 1 − 2 n and ∥ ∇ 2 H ~ h ~ , k ~ ∥ C 0 2 , α ( N 2 − N 1 ) ≤ c κ R ɛ 2 − 2 n . (31)
Now observe that L 0 H ~ h ~ , k ~ = 0   in each C 1   , hence L N H ~ h ~ , k ~ = ( L N − L 0 ) H ~ h ~ , k ~   in this set.
Now, we use the expansion ( 6 ) or the expansion ( 7 ) which reflect the fact that the metric induced by ω N   is, in each C 1   asymptotic to the Euclidean metric. This implies that there exists a constant c > 0   such that ∥ ( L N − L 0 ) φ ∥ C δ − 4 0 , α ( N R ɛ − N 2 ) ≤ c ∥ ∇ 2 φ ∥ C δ + 2 n − 4 2 , α ( N R ɛ − N 2 ) ≤ c ∥ ∇ 2 φ ∥ C 0 2 , α ( N R ɛ − N 2 ) .   Collecting these, and taking into account the effect of the cutoff functions, we conclude that ∥ L M H h , k ∥ C δ − 4 0 , α ( N R ɛ ) ≤ c κ R ɛ 2 − 2 n .   Observe that the main contribution comes from the effect of the cutoff function χ ~   .
The second estimate is easy to derive. The third estimate follows from the structure of the nonlinear operator Q N   , which is similar to the one described in ( 13 ), together with ( 31 ). Again, the most important nonlinear terms in the expression of Q N   are of the form B j ( ∇ 4 φ , ∇ 2 φ )   and B ~ j ′ ( ∇ 3 φ , ∇ 3 φ )   provided the second derivatives of φ   remain bounded (which is precisely our case). The proof of the second estimate follows easily from these considerations and, again, the main contribution comes from the effect of the cutoff function χ ~   . ■   We will also need the :
Lemma 4.7. We fix δ ∈ ( 0 , 1 )   and c ~ > 0   . There exists c ^ > 0   such that, for all φ , φ ′ ∈ C δ 4 , α ( N R ɛ )   , satisfying ∥ φ ∥ C δ 4 , α ( N R ɛ ) + ∥ φ ′ ∥ C δ 4 , α ( N R ɛ ) ≤ c ~ R ɛ 3 − 2 n − δ ,   we have ∥ Q N ( H ~ h ~ , k ~ + φ ) − Q N ( H ~ h ~ , k ~ + φ ′ ) ∥ C δ − 4 0 , α ( N R ɛ ) ≤ c ^ R ɛ 3 − 2 n − δ ∥ φ − φ ′ ∥ C δ 4 , α ( N R ɛ ) .  
Using the result of §5.3, we reduce the solvability of the equation ( 30 ) to a fixed point problem in { φ ∈ C δ 4 , α ( N R ɛ ) : ∥ φ ∥ C δ 4 , α ( N R ɛ ) ≤ c ~ R ɛ 3 − 2 n − δ } ,   where the constant c ~ > 0   is fixed large enough. In fact, using Lemma  4.6 and Lemma  4.7 it is easy to check that there exists a fixed point solution of ( 30 ) provided ɛ   is chosensmall enough, say ɛ ∈ ( 0 , ɛ κ )   , which is obtained by applying a fixed point theorem for contraction mapping.
We have obtained the :
Proposition 4.6. Given κ > 0   , a ~ > 0   and s ∈ R   , there exists ɛ κ > 0   such that, for all ɛ ∈ ( 0 , ɛ κ )   , for all h ~ ∈ C 4 , α ( ∂ B 1 )   and k ~ ∈ C 2 , α ( ∂ B 1 )   , there exists a Kähler form η h ~ , k ~ , s , a ~ = η a ~ + i ∂ ∂ ¯ φ ~ h ~ , k ~ , s , a ~ ,   defined on N ( R ɛ )   , which has constant scalar curvature. Moreover ∥ φ ~ h ~ , k ~ , s , a ~ | C R ɛ − C R ɛ / 2 ( R ɛ ⋅ ) − H h ~ , k ~ i ∥ C 4 , α ( B 1 − B 1 / 2 ) ≤ c R ɛ 3 − 2 n ,   for some constant c > 0   independent of κ   . In addition, s ( ω ~ h ~ , k ~ , s ) = ɛ 2 s   .
We now apply this result when s = s ( ω h , k ) ,   is the scalar curvature of the metric defined in the previous section. The corresponding Kähler form will be denoted by η h ~ , k ~ , a ~   even though it also depends on h   and k   . Recall that φ ~ a ~   is the function which appears in ( 25 ) and ( 26 ), the expansion of the Kähler potential η a ~   at ∞   . Hence, the Kähler form η h ~ , k ~ , a ~   has constant scalar curvature equal to ɛ 2 s ( ω h , k )   and, near ∞   , it can be expanded as η h ~ , k ~ , a ~ = i ∂ ∂ ¯ ( 1 2 | u | 2 − a ~ | u | 4 − 2 n + φ t + φ ~ h ~ , k ~ , s ) ,   when n ≥ 3   and η h ~ , k ~ , a ~ = i ∂ ∂ ¯ ( 1 2 | u | 2 + a ~ log | u | 2 + φ t + φ ~ h ~ , k ~ , s ) ,   when n = 2   .
As in the previous section, we define, ψ i : = − a ~ R ɛ 4 − 2 n | z | 4 − 2 n + ( φ t + φ ~ h ~ , k ~ , s ) ( R ɛ ⋅ ) ,   when n ≥ 3   and ψ o : = a ~ log | z | 2 + ( φ t + φ ~ h ~ , k ~ , s ) ( R ɛ ⋅ ) ,   when n = 2   . Observe that, locally, the Kähler potential is not uniquely defined and addition of constant does not modify the corresponding Kähler form.
Thanks to the previous analysis, the following expansions are valid in each B 1 − B 1 / 2   .
Lemma 4.8. The following expansion holds
ψ i ( z ) : = − a ~ R ɛ 4 − 2 n | z | 4 − 2 n + H h ~ , k ~ i ( z ) + O ( R ɛ 3 − 2 n ) , (32)
when n ≥ 3   and
ψ i ( z ) = a ~ log | z | 2 + H h ~ , k ~ i ( z ) + O ( R ɛ − 1 ) , (33)
when n = 2   .
Similar expansions are valid for the partial derivatives of these functions.
The key point is that the quantity O ( R ɛ 3 − 2 n )   is bounded by a constant, independent of κ   and of a ~   , times R ɛ 3 − 2 n   , as ɛ   tends to 0   . Again, it should be clear that the Kähler form provided by Proposition  4.6 depends continuously on the data (such as h , h ~ , k   and k ~   ).

5 Gluing the pieces together

We are now in a position to describe the connected sum construction. For all ɛ > 0   small enough, we define a complex manifold M ɛ   be removing small balls centered at the points p ℓ   , ℓ = 1 , … , m   and replacing them by properly rescaled versions of the N   . We define
r ɛ = ɛ 2 n − 1 2 n + 1 , and R ɛ : = r ɛ ɛ . (34)
By construction M ɛ : = M ⊔ p 1 , ɛ N ⊔ p 2 , ɛ ⋅ ⋅ ⋅ ⊔ p m , ɛ N ,   is obtained by performing a connected sum of M r ɛ   (which has m   boundaries) with the truncated ALE spaces N R ɛ , … , N R ɛ   . This connected sum is obtained by identifying ∂ B ℓ r ɛ   with ∂ C R ɛ   using ( z 1 , … , z n ) = ɛ ( u 1 , … , u n ) ,   where ( z 1 , … , z n )   are coordinates in B ℓ ( r 0 )   and ( u 1 , … , u n )   are coordinates in C ℓ ( R 0 )   .
Next, a Kähler form is defined on M ɛ   by gluing together the Kähler form w h , k   define in §4.4 and the rescalled Kähler forms ɛ 2 η h ~ 1 , k ~ 1 , a ~ 1 , … , ɛ 2 η h ~ m , k ~ m , a ~ m   defined in §4.5 on each copy of N R ɛ   .
It remains to explain to choose the data h : = ( h 1 , … , h m )   , k : = ( k 0 , … , k m )   satisfying ( 17 ) and ( 18 ), h ~ : = ( h ~ 1 , … , h ~ m )   , k ~ : = ( k ~ 1 , … , k ~ m )   satisfying ( 28 ), and a ~ : = ( a ~ 1 , … , a ~ m )   in such a way that, for each ℓ = 1 , … , m   , the function ψ ℓ o   defined in B 2 − B 1   as in §4.4 on the one hand and the function ɛ 2 ψ ℓ i   defined in B 1 − B 1 / 2   as in §4.5 with data h ~ ℓ , k ~ ℓ   and a ~ ℓ   on the other hand, have their derivatives up to order 4   which coincide on ∂ B 1   .
Now, assume that the functions ψ ℓ o   and ɛ 2 ψ ℓ i   are C 4 , α   functions which satisfy
ψ ℓ o = ɛ 2 ψ ℓ i , ∂ r ψ ℓ o = ɛ 2 ∂ r ψ ℓ i , Δ 0 ψ ℓ o = ɛ 2 Δ 0 ψ ℓ i , ∂ r Δ 0 ψ ℓ o = ɛ 2 ∂ r Δ 0 ψ ℓ i , (35)
on ∂ B 1   where r = | u |   . Consider polar coordinates and decompose Δ 0 = ∂ r 2 + 1 r 2 Δ S 2 n − 1 + 2 n − 1 r ∂ r .   Using the decomposition of Δ 0   in polar coordinates, it is clear that these identities guaranty that ψ ℓ o   and ɛ 2 ψ ℓ i   have their partial derivatives up to order 3   which coincide on ∂ B 1   .
Assuming for the moment that we have already found the construction data in such a way that ( 35 ) are satisfied, then the Kähler form i ∂ ∂ ¯ ( 1 2 r ɛ 2 | u | 2 + ψ ℓ o ) ,   defined in B 2 − B 1   and the Kähler form i ɛ 2 ∂ ∂ ¯ ( 1 2 R ɛ 2 | u | 2 + ψ ℓ i ) ,   defined in B 1 − B 1 / 2   have constant scalar curvature which are equal. This implies that the potential function defined by ψ : = ψ ℓ o   in B 2 − B 1   and ψ : = ɛ 2 ψ ℓ i   in B 1 − B 1 / 2   is a weak solution of the nonlinear elliptic partial differential equation s ( i ∂ ∂ ¯ ( 1 2 r ɛ 2 | u | 2 + ψ ) ) = c t e .   in B 2 − B 1 / 2   . In addition ψ   is C 3 , α   and smooth away from ∂ B 1   . It then follows from elliptic regularity theory together with a bootstrap argument that this function is in fact smooth. Hence, by gluing the Kähler metrics ω h , k   and ɛ 2 η h ~ ℓ , k ~ ℓ , a ~ ℓ   on the different pieces constituting M ɛ   , we have produced a Kähler metric on M ɛ   which has constant scalar curvature. This will end the proof of the main Theorem. It remains to explain how to find the functions h = ( h 1 , … , h m )   , k = ( k 1 , … , k m )   , h ~ = ( h ~ 1 , … , h ~ m )   , k ~ = ( k ~ 1 , … , k ~ m )   and the parameters a ~ = ( a ~ 1 , … , a ~ m )   . To this aim, let us assume that n ≥ 3   since only notational changes are needed to handle the general case n = 2   . It follows from the result of Lemma  4.5 and Lemma  4.8 that the following expansions hold
ψ ℓ o ( z ) = − a ℓ ɛ 2 n − 2 r ɛ 4 − 2 n | z | 4 − 2 n + H h ℓ , k ℓ o ( z ) + O ( r ɛ 4 )
ɛ 2 ψ ℓ i ( z ) = − a ~ ℓ ɛ 2 n − 2 r ɛ 4 − 2 n | z | 4 − 2 n + ɛ 2 H h ~ , k ~ i ( z ) + O ( r ɛ 4 ) ,
We change parameters and define the functions h ℓ ′ , k ℓ ′ , h ~ ℓ ′   and k ~ ℓ ′   by
h ℓ ′ : = ( a ~ ℓ − a ℓ ) r ɛ 4 − 2 n ɛ 2 n − 2 + h ℓ
k ℓ ′ : = 2 ( 4 − 2 n ) ( a ~ ℓ − a ℓ ) ɛ 2 n − 2 r ɛ 4 − 2 n + k ℓ
and also h ~ ℓ ′ : = ɛ 2 h ~ ℓ k ~ ℓ ′ : = ɛ 2 k ~ ℓ   Recall that the functions k ℓ   are assumed to satisfy ( 18 ) while the function k ℓ ′   is not assumed to satisfy such a constraint. The role of the scalar a ~ ℓ − a ℓ   is precisely to recover this lost degree of freedom in the assignment of the boundary data. With these new variables, the expansions for both ψ ℓ o   and ɛ 2 ψ ℓ i   can now be written as
ψ ℓ o ( z ) = − a ~ ℓ r ɛ 4 − 2 n | z | 4 − 2 n + H h ℓ ′ , k ℓ ′ o ( z ) + O ( r ɛ 4 )
ɛ 2 ψ ℓ i ( z ) = − a ~ ℓ r ɛ 4 − 2 n | z | 4 − 2 n + H h ~ ′ , k ~ ′ i ( z ) + O ( r ɛ 4 ) .
The data functions h ′ : = ( h 1 ′ , … , h m ′ )   , h ~ 1 ′ , … , h ~ m ′ )   are assumed to be bounded by a constant κ   times r ɛ 4   in C 4 , α ( ∂ B 1 )   and the data functions k ′ : = ( k 1 ′ , … , k m ′ )   and k ~ ′ : = ( k ~ 1 ′ , … , k ~ m ′ )   are assumed to be bounded by a constant κ   times r ɛ 4   in C 2 , α ( ∂ B 1 )   . The terms O ( r ɛ 4 )   are nonlinear terms of h ′ , h ~ ′ , k ′ , k ~ ′   which are bounded by a constant (independent of κ   times r ɛ 4   provided ɛ   is chosen small enough.
We recall the following result :
Lemma 5.1. [2] The mapping
P : C 4 , α ( ∂ B 1 ) × C 2 , α ( ∂ B 1 ) ⟶ C 3 , α ( ∂ B 1 ) × C 1 , α ( ∂ B 1 )
( h , k ) ↦ ( ∂ r ( H h , k o − H h , k i ) , ∂ r Δ 0 ( H h , k o − H h , k i ) ) ,
is an isomorphism.
Using this Lemma, it is straightforward to check that the solvability of ( 35 ) reduces to a fixed point problem which can be written as ( h ′ , h ~ ′ , k ′ , k ~ ′ ) = S ɛ ( h ′ , h ~ ′ , k ′ , k ~ ) ,   where the nonlinear operator S ɛ   satisfies ∥ S ɛ ( h ′ , h ~ ′ , k ′ , k ~ ′ ) ∥ ( C 4 , α ) 2 m × ( C 2 , α ) 2 m ≤ c 0 r ɛ 4 ,   for some constant c 0 > 0   which does not depend on κ   , provided ɛ   is small enough. We finally choose κ = 2 c 0 ,   and ɛ ∈ ( 0 , ɛ κ )   , where ɛ κ   is fixed small enough. We have therefore proved that S ɛ   is a map from A ɛ : = { ( h ′ , h ~ ′ , k ′ , k ~ ′ ) ∈ ( C 4 , α ) 2 m × ( C 2 , α ) 2 m : ∥ ( h ′ , h ~ ′ , k ′ , k ~ ′ ) ∥ ( C 4 , α ) 2 m × ( C 2 , α ) 2 m ≤ κ r ɛ 4 } ,   into itself. This mapping is clearly continuous and if in addition it were compact, the application of Leray-Schauder's fixed point theorem would directly guaranty the existence of a fixed point. However, the nonlinear equations we have solved being fully nonlinear, there is no gain of regularity and hence To overcome this last difficulty, we define a family of smoothing mappings D q   [1] such that
| | D q u | | C r , α ≤ c | | u | | C r ′ , α ′ for r + α ≤ r ′ + α ′ | | D q u | | C r , α ≤ c q r ′ + α ′ − r − α | | u | | C r ′ , α ′ for r + α ≥ r ′ + α ′ | | u − D q u | | C r , α ≤ c q r ′ + α ′ − r − α | | u | | C r ′ , α ′ for r + α ≤ r ′ + α ′ . (36)
and replace the nonlinearities S ɛ   by S ɛ , q : = D q ∘ S ɛ   . This time the corresponding operator S ɛ , q   is compact and for ɛ   sufficiently small, maps the ball of radius r ɛ 4   to itself.
Hence it has a fixed point ( h q ′ , h ~ q ′ , k q ′ , k ~ q ′ )   when ɛ   is small enough. Finally, the fixed points are bounded uniformly in q ∈ ( 0 , 1 )   , so for any fixed α ′ < α   we may extract a sequence q j → 0   such that ( h q j ′ , h ~ q j ′ , k q j ′ , k ~ q j ′ )   converges in ( C 4 , α ′ ) 2 m × ( C 2 , α ′ ) 2 m   to a fixed point of S ɛ   . This completes our proof of the existence the Kähler metric on M ɛ   which has constant scalar curvature.
Observe that the use of smoothing operator is not strictly necessary and the fixed point for S ɛ   could have been obtained through the application of a fixed point theorem for contraction mappings, but this would have required more work in §4.4 and §4.5 to prove that S ɛ   is a contraction for ɛ   small enough.

6 Understanding the constraints

In this section, we give the proof of Lemma 1.1 and Lemma 1.2.
As usual, let us denote by ξ 0 ≡ 1 , ξ 1 , … , ξ d   the set of independent functions which span the kernel of L M   . We assume that, for j = 1 , … , d   , the functions ξ j   are normalized to have mean 0   and to be mutually L 2   -orthogonal. We keep the notations of the introduction and define the matrix M   as in ( 1 ) and the integer values functions ℭ 1   and ℭ 2   as in ( 1 ).
Lemma 6.1. Assume that m ≥ d   . Then the set of points ( p 1 , … , p m ) ∈ M m   such that ℭ 1 = d   is open and dense in M m   .
Proof : Observe that it is sufficient to consider the case where m = d   , since increasing the number of columns of the matrix can only increase its rank !
The proof is by induction on d   . When d = 0   there is nothing to prove. Now, assume that the result if true for d − 1   functions. We write the function ( p 1 , … , p d ) ⟶ det M ( p 1 , … , p d )   as a linear combination of functions depending on p d   , the coefficients of which are functions which depend on p 1 , … , p d − 1   . In other words, we expand the determinant det M   with respect to the last row. We obtain det M ( p 1 , … , p d ) = M 1 ( p 1 , … , p d − 1 ) ξ 1 ( p d ) + … + M d ( p 1 , … , p d 1 ) ξ d ( p d )   where M i   is (up to factor ± 1   ) equal to the determinant of the matrix ( ξ a ( p b ) ) a , b   where a = 1 , … , i ˇ , … , d   and b = 1 , … , d − 1   . By assumption, the set of ( p 1 , … , p d − 1 ) ∈ M d − 1   for which ( M 1 , … , M d ) ≠ 0   is open and dense in M d   . We fix ( p 1 , … , p d − 1 ) ∈ M d   such that ( M 1 , … M d ) ≠ 0   . The functions ξ 1 , … , ξ d   being independent, the set of p d ∈ M   for which det M ≠ 0   is open and non empty. Furthermore, unique continuation theorem for the solutions of L M ξ = 0   implies that the the set of p d ∈ M   for which det M ≠ 0   is dense. This completes the proof of the result. ■   The second condition for our construction to work asks for the existence p 1 , … , p m ∈ M   , with m ≥ d + 1   , such that there exists a solution a 1 , … , a m > 0   to the system ∀ i = 1 , … , d , ∑ j = 1 m a j ξ i ( p j ) = 0 ,   This amounts to ask for the existence of points p 1 , … , p m ∈ M   such that the kernel of M : = M ( p 1 , … , p m )   contains a vector whose entries are all positive, i.e. an element of K + m   , the positive cone in R n   . Equivalently, we have to prove that it is possible to find points p 1 , … , p m ∈ M   such that the image of t M   is included in a hyperplane of R m   whose normal belongs to K + m   .
Observe that, according to Lemma  6.1 , for all m ≥ 1   , there exists an open and dense set U m ⊂ M m   such that for all ( p 1 , … , p m ) ∈ U m   , the kernel of M   is ( m − d )   -dimensional and in fact varies continuously as the points change in U m   . Equivalently, for any choice of the points in U m   , the image of t M   is a d   -dimensional subspace of R m   which also varies continuously as the points change in U m   . This clearly shows that the set of points ( p 1 , … , p m ) ∈ M m   such that ℭ 1 = d   and ℭ 2 ≠ 0   is an open (probably empty !) set. Also, it should now be clear that once we have found p 1 , … , p m   satisfying both conditions, then the conditions remain fulfilled after any adjunction of points to this list.
Now, the condition that Im ( t M )   is included in a hyperplane of R m   whose normal belongs to K + m   is equivalent to the requirement that Im ( t M )   does not contain any vector of K + m   , i.e. that all nonzero elements of Im ( t M )   have entries which change sign.
This being understood, for all Λ : = ( λ 1 , … , λ d ) ∈ S d − 1   , we define on M   the function f Λ : = ∑ j = 1 d λ j ξ j .   The previous discussion can be summarized as follows : We have to prove that, it is possible to find ( p 1 , … , p m ) ∈ U m   such that, for all Λ ∈ S d − 1   , the vector ( f Λ ( p 1 ) , … , f Λ ( p m ) )   has entries which change sign.
As stated in the introduction, except in special cases, we have not been able to prove that the minimal number of points for which the above condition is satisfied is d + 1   even though we suspect that this is the case. Nevertheless, we have the general result :
Lemma 6.2. There exists m 0 ≥ d + 1   and, for all m ≥ m 0   , there exists an nonempty open set V m ⊂ M m   such that ℭ 2 ≠ 0   , for all ( p 1 , … , p m ) ∈ V m   .
Proof : We keep the above notations. Pick Λ ∈ S d − 1   . Then the function p ∈ M ⟶ f Λ ( p )   has mean 0   (since it is a linear combination of the functions ξ 1 , … , ξ d   which are are assumed to have mean 0   ). Therefore it is possible to find p Λ , p ~ Λ ∈ M   such that f Λ ( p Λ ) < 0 < f Λ ( p ~ Λ )   Now, by continuity, we also have f Λ ′ ( p ) < 0 < f Λ ′ ( p ~ )   for all Λ ′   in some open neighborhood O Λ   of Λ   in S d − 1   , all p   in some open neighborhood o Λ   of p Λ   in M   and all p ~   in some open neighborhood o ~ Λ   of p ~ Λ   in M   . The sets O Λ   constitute an open cover of S d − 1   , and by compactness one can find a finite sub-cover S d − 1 = ∪ j = 1 J O Λ j   Given any ( p 1 , … , p J , p ~ 1 , … , p ~ J ) ∈ Π j = 1 J o Λ j × Π j = 1 J o ~ Λ j   and given any Λ ∈ S d − 1   , it belongs to some O Λ j   and hence the j   -th and the ( J + j )   -th entries of the vector ( f ( p 1 ) , … , f ( p J ) , f ( p ~ 1 ) , … , f ( p ~ J ) )   do not have the same sign. Therefore, we have found m = 2 J   points satisfying the required conditions. ■   One can then define m 0 ≥ d + 1   to be the least number of points for which ℭ 1 = d   and ℭ 2 ≠ 0   .

7 Geometric interpretations of the constraints

Let us go back to a well known interpretation of the kernel of the operator L M   in terms of holomorphic vector fields. We recall some important results (which go back to Matsushima and Lichnerowicz) and emphasize only the part relevant to the geometric interpretation of the analysis we have carried out in the previous sections. We refer the readers to the papers [13] and [12] and the book of Futaki [4] for a complete introduction and other applications. Given a real valued function ξ   satisfying L M ξ = 0   and having called X ξ = ∂ # ξ   the ( 1 , 0 )   part of the gradient of ξ   , one can use ( 10 ) to conclude that ∂ ¯ X ξ = 0   , i.e. X ξ   is a holomorphic vector field which vanishes somewhere on the manifold (since ξ   certainly has critical points !). Conversely, every Killing vector field which vanishes somewhere is the imaginary part of a holomorphic vector field of the form X ξ   for some ξ ∈ Ker L M   (see for example [4] , [13] ,[12] ). If in addition ( M , ω M )   has constant scalar curvature, then, modulo the space of parallel holomorphic vector fields, every holomorphic vector field vanishing somewhere on M   arises as X ξ   for some ξ ∈ Ker L M   [4] , [13] , [12] . Of course the parallel part of the Lie algebra of holomorphic vector fields does not come in since the requirement of vanishing somewhere makes it trivial.
We will therefore disregard (and in fact we have already done it in the statements of our results) these vector fields.
The correspondence just described should not shadow some important differences in dealing with holomorphic vector fields and with function annihilating the operator L M   . This appears clearly when comparing our sufficient conditions with the existence of holomorphic vector fields on the blow up manifold. The following is standard, but we include it for reader's convenience.
Proposition 7.1. Let X   be a holomorphic vector field on M   and p   any point in M   . Then X   lifts to a holomorphic vector field X ~   on Bl p M   if and only if X ( p ) = 0   . Moreover any holomorphic vector field on Bl p M   projects to a holomorphic vector field on M   vanishing at p   .
Proof : If X   lifts to X ~   on Bl p − M   , then the (holomorphic) flow associated to X ~   preserves Bl p M \ E   , where E   is the exceptional divisor, and E   is fixed. Therefore X ( p ) = 0   . Moreover being the blow up map an isomorphism away from the exceptional divisor, the same observation proves the last part of the proposition, since extending the projected vector field to zero at p   gives a continuous vector field and hence a holomorphic one.
Conversely, we can look directly on a local chart in C n   , with p   identified with the origin, since away from the origin the problem is trivial. Then Bl 0 C n = { ( ( z 1 , . . . , z n ) , [ l 1 , . . . , l n ] ) ∈ C n × P n − 1 : z i l j = z j l i , i , j = 1 , . . . , n } .   We consider the chart given by u i   for which the defining equations become z 1 = u 1 u n , . . . , z n − 1 = u n − 1 u n , z n = u n   . In this chart, ∂ z i = 1 u n ∂ u 1 i = 1 , . . . , n − 1   ∂ z n = − u 1 u n ∂ u 1 − ⋅ ⋅ ⋅ − u n − 1 u n ∂ u n − 1 + ∂ u n .   Hence a holomorphic vector field on C n   , which can be written as X = X α ∂ z α   lifts to Bl 0 − C n − { Exceptional divisor }   as the vector field Y = X 1 u n ∂ u 1 + ⋅ ⋅ ⋅ + X n − 1 u n ∂ u n − 1 + X n ( − u 1 u n ∂ u 1 − ⋅ ⋅ ⋅ − u n − 1 u n ∂ u n − 1 + ∂ u n ) .   In order to extend Y   to the exceptional divisor we have to verify that the functions X α − u α X n u n   are bounded for any α   near u α = 0   . This clearly forces X α ( 0 ) = 0   for any α   , and therefore the vanishing of the vector field X   at the blow up point. Once this is verified the extension is continuous and hence holomorphic. ■   Whether a holomorphic vector field (vanishing somewhere) lifts to a blow up depends only on its value at the point one is blowing up. Therefore, if the space of holomorphic vector fields with zeros has complex dimension d   , blowing up d   points in generic position (generic in the sense described by the above proposition) one gets a manifold without such vector fields. In terms of the potentials used throughout this paper this translates in looking for m   points such that Rank ( ∂ # ( ξ i ) ( p j ) ) = d .   Our first condition ( ℭ 1 = d   ) is sensitive to the zeros of the bounded functions in Ker L M   with mean zero instead of their critical values.
The second condition is certainly encoding a more subtle phenomenon. In fact, this second condition should be related to some suitable stability property of the blown up manifold. The fact that some positivity condition must hold is present in all known examples in different veins, and has been deeply investigated in the case of complex surfaces with zero scalar curvature by LeBrun-Singer [13] , Rollin-Singer [15] , and for Del Pezzo surfaces by Rollin-Singer [16] . In our construction it appears naturally from the analytical approach we developed, but a geometric interpretation is certainly worth seeking.

8 Examples

8.1 The case of P n  

A convenient way to study our problems on the blow up at points of the projective spaces is to look at P n   as the quotient of the unit sphere in C n + 1   with complex coordinates ( z 1 , . . . , z n + 1 )   via the standard S 1   -action given by the restriction of complex scalar multiplication.
It is well known that the automorphism group of P n   is given by the projectivization of GL ( n + 1 , C )   , whose complex dimension is d = ( n + 1 ) 2 − 1   . We therefore seek for d   real functions whose ( 1 , 0 )   -part of the gradient generate the Lie algebra of the automorphism group as explained in the previous subsection. This can be done in two equivalent ways: either by explicit computation on the automorphism group, or by relying on the equivalence described in the previous section between this and the study of the kernel of the operator L M = − 1 2 Δ M 2 − Ric M ⋅ ∇ M 2   which for P n   with its Fubini-Study metric induced by the Hopf fibration becomes L M = − 1 2 Δ P n 2 − 2 ( n + 1 ) Δ P n = − 1 2 Δ P n ( Δ P n + 4 ( n + 1 ) ) .   Our problem reduces to seeking a basis of functions with mean zero of the eigenspace of the Laplacian Δ P n   associated to the eigenvalue 4 ( n + 1 )   (i.e. the eigenspace of the Laplacian Δ S 2 n + 1   which are associated to the eigenvalue 4 ( n + 1 )   and are invariant under the S 1   action), and this is clearly given by the n 2 + 2 n   functions ξ a b = z a z ¯ b + z b z ¯ a , ξ ^ a b = i ( z a z ¯ b − z b z ¯ a )   for 1 ≤ a < b ≤ n + 1   and ξ ~ a = | z a | 2 − | z a + 1 | 2 ,   for a = 1 , … , n   . Recall that we should add ξ 0 ≡ 1   to this list.
Obviously any explicit calculation will be rather troublesome. It is hence very convenient (and giving best results) to introduce symmetries acting on the projective space in order to reduce as much as possible the elements of the kernel of L P n   which are invariant under these symmetries.
Example 1: Let us consider the group G   acting on P n   generated by the transformations ( z 1 , … , z n + 1 ) ⟶ ( ± z 1 , … , ± z n + 1 ) .   Of course, the action of any element of the group on C n + 1   maps the unit sphere into itself. The space of elements of the kernel of L P n   which are invariant under the action of the elements of G   is generated by { 1 , ξ ~ 1 , … , ξ ~ n }   .
Let us fix the following set of blow up points p 1 = ( 1 , 0 , … , 0 ) , p 2 = ( 0 , 1 , … , 0 ) , … , p n + 1 = ( 0 , … , 0 , 1 ) .   We find the matrix M = ( 1 − 1 0 . . . 0 0 1 − 1 . . . . . . . . . ⋱ . . . ⋱ 0 0 . . . 0 1 − 1 )   whose rank is clearly equal to n   and has a one dimensional kernel spanned by the vector ( 1 , … , 1 )   which has positive entries ! This proves that, working equivariantly with respect to the action of the group G   , the blow up of P n   at the above n + 1   points carries a constant scalar curvature Kähler form.
We have then proved
Corollary 8.1. The blow up of P n   at p 1 = ( 1 , 0 , … , 0 ) , … , p n + 1 = ( 0 , … , 0 , 1 )   has a constant scalar curvature Kähler metric.
It is worth remarking that this result is optimal in the number of points to be blown up, since for fewer points the manifold would have nonreductive automorphisms group, and hence no Kähler metrics of constant scalar curvature by the Mathushima-Lichnerovicz obstruction. Another interesting aspect of this example is that the manifold obtained still has non trivial (in fact n   dimensional) automorphism group. The point is that the surviving automorphisms are precisely those which are not G   -invariant.
Exemple 2 : Now, we still work equivariantly with respect to the action of the group G   defined above but we fix the following set of blow up points p 1 = ( 1 , 0 , … , 0 ) , p 2 = ( 0 , 1 , 0 , … , 0 ) , … , p n = ( 0 , 0 , … , 1 , 0 )   and p n + 1 = ( 0 , … , − α , β ) p n + 2 = ( 0 , 0 , … , α , β ) .   where α 2 + β 2 = 1   . This time, we find the matrix M = ( 1 − 1 0 . . . 0 0 0 0 1 − 1 . . . 0 0 0 . . . . . . . . . . . . 0 . . . 1 − 1 − α 2 − α 2 0 . . . 0 1 α 2 − β 2 α 2 − β 2 )   whose rank is again equal to n   and has a two dimensional kernel containing the vector ( 1 , … , 1 , 1 4 β 2 , 1 − 1 2 β 2 , 1 − 1 2 β 2 )   which has positive entries, provided 0 < β < 1 2   ! This proves that, working equivariantly with respect to the action of the group G   , the blow up of P n   at the above n + 2   points carries a constant scalar curvature Kähler form.
It is an easy general observation that the addition of points to a list of points satisfying our sufficient conditions preserves these conditions satisfied. In particular we can add one point to the above lists, so, for example, the G   -orbit of any point p   (which does not initially belong to the list) and keep the two conditions fulfilled. Observe that, for generic choice of the point p   the G   orbit of p   has 2 n   points, so this substantially increases the number of points one can blow.
However, one can add to one of the above lists k   points of the form ( 0 , … , 0 , α , 0 , … , ± β , 0 , … , 0 )   where α 2 + β 2 = 1   , α ≠ 0   β ≠ 0   , so that the list of points remains invariant under the action of G   . This clearly increases the number of blow up points by 2 k   . Using this idea and starting from the list of points given in example 1, one shows that the blow up of P n   at n + 1 + 2 k   points carries a constant scalar curvature Kähler form and starting from the second list of points, one shows that the blow up of P n   at n + 2 + 2 k   points carries a constant scalar curvature Kähler form. Therefore, we have obtained :
Corollary 8.2. The blow up of P n   at m ≥ n + 1   (special) points has a constant scalar curvature Kähler metric.
Example 3 : The important observation is that, so far we have worked equivariantly and the m   blow up points cannot be chosen into some open set of M m   . We now give a upper bound for the number m 0   which corresponds to the least number of points (largerthan d   ) for which the conditions ℭ 1 = d   and ℭ 2 ≠ 0   are fulfilled. Given α , β ∈ R − { 0 }   satisfying α 2 + β 2 = 1   and α 2 − β 2 ≠ 0   , we consider the following set of points :
p i j = ( 0 , … , 0 , α , 0 , … , 0 , β , 0 … , 0 )
p ~ i j = ( 0 , … , 0 , α , 0 , … , 0 , − β , 0 … , 0 )
p ^ i j = ( 0 , … , 0 , β , 0 , … , 0 , i α , 0 … , 0 )
p ˘ i j = ( 0 , … , 0 , β , 0 , … , 0 , − i α , 0 … , 0 )
where 1 ≤ i < j ≤ n + 1   correspond to the indices of the nonzero entries. There are exactly 2 n ( n + 1 )   such points and it is easy to check that both conditions are fulfilled.
Indeed, the first condition ℭ 1 = d   is easy to check and left to the reader. Concerning the second condition, we show that ( 1 , … , 1 )   is in the kernel of M   . Observe that this condition can be translated into the fact that for each ξ = ξ a b , ξ ^ a b   or ξ ~ a   ∑ p ξ ( p ) = 0   where summation over p ∈ { p i j , p ~ i j , p ^ i j , p ˘ i j : i , j }   is understood. Observe that it is enough to check that the formula holds for the functions z a z ¯ b   and the functions | z a + 1 | 2 − | z a | 2   . It is now easy too check that ℭ 2 ≠ 0   .
Therefore, we have obtained the :
Corollary 8.3. When M = P n   , we have the estimate m 0 ≤ 2 n ( n + 1 )   .
Example 4 : For general Kähler manifolds with space of holomorphic vector fields of dimension d   , one needs to blow up d   points in general position to have a manifold without holomorphic vector fields. In this respect P n   is very special since it is easy to observe that n + 2   points suffice provided they form a so called projective frame, namely any choice of n + 1   of them are linearly independent in C n + 1   (such sets of points are also often called in generic position with respect to hyperplanes). The freedom of choices of projective frames ranges clearly in an open and dense subset of ( P n ) n + 2   .
Suppose that we find a projective frame p 1 , . . . , p n + 2   for which we can prove that B l p 1 , . . . p n + 2 P n   has a Kähler constant scalar curvature metric necessarely, being n + 2 < d   , using some equivariant construction. The first simple observation is that B l p 1 , . . . , p n + 2 P n   is biholomorphic to B l q 1 , . . . , q n + 2 P n   for any other projective frame q 1 , . . . , q n + 2   . The manifold obtained is also with at most discrete automorphisms and we can then apply to it the results of [2] to blow up any other set of points and still get Kähler constant scalar curvature metrics.
For all these reasons we now seek for a projective frame for which some equivariant construction works. To this aim, consider the group G   of permutations on the n + 1   affine coordinates and the points p 1 = 1 n + 1 ( 1 , . . . , 1 ) , p 2 = 1 n + α 2 ( 1 , , . . . , 1 , − α )   with α > 0   . The equivariant kernel is spanned by { 1 , ∑ a < b ξ a b }   and therefore the matrix we obtain is M = ( n n ( n − 1 − 2 α ) n + α 2 )   Our sufficient conditions are then fulfilled for α > n − 1 2   . It is easy to see that the points given by the orbits of p 1   (which is fixed), and p 2   (which consists of n + 1   elements) form a projective frame and hence the resulting manifold is without holomorphic vector fields.
Corollary 8.4. Given m ≥ n + 2   points in P n   which contain a projective frame, then the blow up of P n   at these m   points has a constant scalar curvature Kähler metric. Moreover the n + 2   of these points which form a projective frame can move in an open set of M n + 2   and the remaining m − n − 2   are arbitrary.
This example naturally rises an important comment. For 3 ≤ m ≤ 8   Tian's solution of the Calabi Conjecture tells us that Bl p 1 , … , p m P 2   admits Kähler-Einstein metrics (of course in the classes [ ω F S ] − [ E 1 ] − ⋯ − [ E m ]   ) as long as no three collinear points are blown up, no 5   of them lie on a quadric and 8   on a cubic. These conditions of course trivially prevent the blown up manifold to have a Kähler-Einstein metric. Moreover it is well known that, under these assumptions, for 4 ≤ m ≤ 8   , Bl p 1 , … , p m P 2   has only discrete automorphisms.
Therefore, as noted in [2] , we can use these metrics for successive blow ups and apply the main result in [2] to represent the classes [ ω F S ] − [ E 1 ] − ⋯ − [ E m ] − ε [ E m + 1 ]   with metrics of constant scalar curvature. What we have gained with the present construction is firstly to represent also the classes [ ω F S ] − ε [ ( E 1 + ⋯ + E m ) ]   with canonical positively curved metrics, therefore getting another open set (by perturbation arguments) in the Kähler cone. Applying their technique on parabolically stable bundles over riemann surfaces Rollin-Singer proved the same result as ours for m ≥ 7   .
Moreover for m > 4   the open set of points for which our construction works is larger than the above mentioned conditions for the existence of a Kähler -Einstein metric. For example once four points forming a projective frame are blown up, we can add a fifth aligned to two of the previous ones still getting a canonical metric.
Example 5 : The case of M = P 1 × M 0   . The type of manifolds we now treat are of particular interest since they have been a central object of study in the papers of LeBrun-Singer, Kim-LeBrun-Pontecorvo up to the recent works of Rollin-Singer, when M 0   is taken to be a Riemann surface. The point is that a complete understanding of these examples leads via algebraic geometric techniques to the relation with stability of rank two vector bundles over Riemann surfaces. By understanding via a different approach these models in our more general setting, we hope to give a tool to the study of similar approach to higher rank vector bundles over any Kähler constant scalar curvature manifold.
We assume throughout this example that M 0   is a Kähler manifold of any dimension with no nonvanishing holomorphic vector fields. Moreover on P 1 × M 0   we consider the product metric, where P 1   is endowed with the Fubini-Study metric normalized as in the previous subsection though for any positive value of the scalar curvature the results we obtain would hold verbatim.
With these conventions the bounded kernel of the operator L M = − 1 2 Δ M 2 − Ric M ⋅ ∇ M 2   is naturally identified with ξ 12   , ξ ^ 12   and ξ ~ a   the functions on P 1   which have been described above.
We then look at the group generated by ( z 1 , z 2 ) → ( z 2 , − z 1 ) ,   which reduces the invariant kernel to be generated by { 1 , ξ ^ 12 }   . Let us choose the points p 1 = 1 2 ( i , 1 )   and p 1 = 1 2 ( 1 , i )   . The first condition then requires the matrix M = ( − 1 1 )   Obviously the two conditions are fulfilled. Since the G   -orbits of the points chosen are projectively the points themselves, we have proved :
Corollary 8.5. Given any two points q 1   and q 2   in M 0   (possibly coinciding), the blow up of P 1 × M 0   at ( p 1 , q 1 )   , ( p 2 , q 2 )   has a constant scalar curvature Kähler metric.
A stronger version of the above Corollary was proved by LeBrun-Singer when M 0   is a Riemann surface of genus at least 1   in [13] (in the case M 0   is a torus one must recall that all holomorphic vector fields are parallel so they do not interfere in the analysis of the operator L M   ). In fact they proved that the metric on the blow up can be chosen to have zero scalar curvature. For M 0 = Σ g   we have gained the freedom of assigning any sign to the constant scalar curvature abtained.
Example 6 : The case of P n × P m   . The results for this case easily follow from the corresponding results for the previous examples.
The case of P 1 × P 1   falls directly in the previous discussion since the blow up of this manifold at m ≥ 1   points is biholomorphic to the blow up at m + 1   suitably choosen points of P 2   . By the previous calculation we already know that for m = 2   our problem has a positive solution, and that this is the least number of points for which this can happen. Moreover we have already seen that for 3   points onwards (suitably chosen in an open set) the manifold has only discrete automorphisms, therefore iteration can start.
Concerning the higher dimensional examples, for sake of simplicity we illustrate the case of P 1 × P 2   , but everything can be easily extended to any of these products. Let us consider affine coordinates ( z 1 , z 2 )   on P 1   and ( w 1 , w 2 , w 3 )   on P 2   and let us denote by ξ a b , ξ ^ a b , ξ ~ a   the functions elements of the kernel on P 1   as described above and by χ a b , χ ^ a b , χ ~ a   the functions elements of the kernel on P 2   . To get the optimal result we have to work equivariantly and we consider the diagonal action used in the first example. The equivariant bounded kernel is then spanned by the union of the generators of each single factor, hence by ξ ~ 1   , χ ~ 1   and χ ~ 2   .
Let us choose the points p 1 = ( ( 1 , 0 ) , ( 0 , 1 , 0 ) ) p 2 = ( ( 0 , 1 ) , ( 1 , 0 , 0 ) )   p 3 = ( ( 1 , 0 ) , ( 0 , 0 , 1 ) ) p 4 = ( 1 5 ( 1 , 2 ) , ( 1 3 ( 1 , 1 , 1 ) ) .   The matrix is then M = ( − 1 1 1 − 3 5 − 1 1 0 0 1 0 − 1 0 )   which is clearly of rank 3   and whose kernel contains elements with positive entries.
Therefore the blow up has a Kähler constant scalar curvature metric. It is moreover not difficult to see that the blow up of P 1 × P 2   has no holomorphic vector fields and therefore can be used both for moving these points in some open set and for starting iterations of further blow ups. Note that this examples generalizes immediately to any dimension in the following
Corollary 8.6. Suppose n ≤ m   . The blow up of P n × P m   at k ≥ m + 2   special points has a constant scalar curvature Kähler metric. Moreover, if the projections of these points to the two factors contain projective frames, these points can move in an open set of ( P n × P m ) m + 2   and the other ones are arbitrary.
References

  1. S. Alinhac and P. Gérard, Opérateurs pseudo-différentiels et théorème de Nash-Moser, InterEditions/Editions du CNRS (1991).
  2. C. Arezzo and F. Pacard F., Blowing up and desingularizing Kähler orbifolds with constant scalar curvature, math.DG/041240522.
  3. E. Calabi, Métriques kählériennes et fibrés holomorphes, Ann. Sci. Icole Norm. Sup. 4 12 (1979), no. 2, 269-294.
  4. A. Futaki, Kähler-Einstein metrics and Integral Invariants, Lecture Notes in Math. 1314 (1988), Springer-Verlag.
  5. P. Griffiths and J. Harris, Principles of algebraic geometry, Wiley-Interscience Publ., 1978.
  6. J. Kim, C. LeBrun and M. Pontecorvo, Scalar-flat Kähler surfaces of all genera, J. Reine Angew. Math. 486 (1997), 69-95.
  7. J. Kim and M. Pontecorvo, A new method of constructing scalar-flat Kähler surfaces, J. Differential Geom. 41 (1995), no. 2, 449-477.
  8. A. Kovalev and M. Singer, Gluing theorems for anti-self-dual spaces, GAFA 11 (2001), 1229-1281.
  9. C. LeBrun, Counter-examples to the generalized positive action conjecture Comm. Math. Phys. 118 (1988), no. 4, 591-596.
  10. C. LeBrun, Scalar-flat Kähler metrics on blown-up ruled surfaces J. Reine Angew. Math. 420 (1991), 161-177.
  11. C. LeBrun, On the scalar curvature of complex surfaces, Geom. Funct. Anal. 5 (1995), 619-628.
  12. C. LeBrun and S. Simanca, Extremal Kähler metrics and complex deformation theory, Geom. Funct. Anal. 4 (1994), 298-336.
  13. C. LeBrun and M. Singer, Existence and deformation theory for scalar flat Kähler metrics on compact complex surfaces Invent. Math. 112 (1993), 273-313.
  14. R. Melrose, The Atiyah-Patodi-singer index theorem, Research notes in Math 4 (1993).
  15. Y. Rollin and M. Singer, Non-minimal scalar-flat Kaehler surfaces and parabolic stability, math.DG/0404423, to appear in Invent. Math.
  16. Y. Rollin and M. Singer, Construction of Kaehler surfaces with constant scalar curvature, math.DG/0412405.
  17. S. Simanca, Kähler metrics of constant scalar curvature on bundles over C P n − 1   , Math. Ann. 291 (1991), no. 2, 239-246.

claudio.arezzo@unipr.it, Dipartimento di Matematica, Universita' di Parma, Via M. D'Azeglio, 85, 43100 Parma, Italy pacard@univ-paris12.fr, University Paris 12 and Institut Universitaire de France, France