November 27, 2006

The author was partially supported by the Swedish Natural Science Research Council .
<ph f="cmbx">The membership problem for polynomial ideals in terms of residue currents</ph>

Mats Andersson

Department of Mathematics, Chalmers University of Technology and the University of Goteborg, S-412 96 GOTEBORG, SWEDEN E-mail address : matsa@math.chalmers.se

1 Introduction

Let F 1 , … , F m   be polynomials in C n   and let Φ   be a polynomial that vanishes on the common zero set of the F j   . By Hilbert's Nullstellensatz, for some power Φ ν   of Φ   , one can find polynomials Q j   such that
∑ j F j Q j = Φ ν . (1.1)
A lot of attention has been paid to find effective versions, i.e., control of ν   and the degrees of Q j   in terms of the degrees of F j   . The breakthrough was in [13] where Brownawell obtained bounds on ν   and deg Q j   not too far from the best possible, using a combination of algebraic and analytic methods, cf., Remark  3 below. Soon after that Kollár [22] obtained by purely algebraic methods the following optimal result.
Theorem (Kollár). Let F 1 , … , F n   and Φ   be polynomials in C n   of degrees d j   , and r   , respectively, and assume that Φ   vanishes on the common zero set of F j   . Then (if d j ≠ 2   ), one can find polynomials Q j   and a natural number s   such that ∑ F j Q j = Φ ν   , and such that ν ≤ N ( d 1 ⋯ d m )   and deg F j Q j ≤ ( 1 + r ) N ( d 1 ⋯ d m )   ; here N ( d 1 ⋯ d m ) = d 1 ⋯ d m   if m ≤ n   ; for the case when m > n   , see [22] .
In particular, if F j   have no common zeros in C n   , then there are polynomials Q j   such that
∑ j F j Q j = 1 , (1.2)
with deg F j Q j ≤ N ( d 1 ⋯ d m ) .   The restriction d j ≠ 2   has recently been removed by Jalonek, [21] , in the case when m = n   .
In [14] Brownawell gave a prime power version of Kollár's theorem which shed more geometric light on these questions, and there is a generalization to smooth algebraic manifolds in [17] .
Kollár's result is optimal as long as one only makes assumptions of the degrees of F j   . However, if one imposes geometric conditions on the zero set one can get sharper results. For instance, assuming that m = n + 1   and F j   have no common zero set even at infinity, then a classical theorem of Macaulay, [23] , states that  1.2 has a solution such that deg F j Q j ≤ ∑ d j − n   .
There is a related result due to Max Nöther, [24] ; see also [19] .
Theorem (Max Nöther, 1873). Assume that the zero set of F 1 , … , F n   is discrete and contained in C n   and that Φ   belongs to the ideal ( F )   .
Then there are polynomials Q j   such that Φ = ∑ 1 n F j Q j   and deg F j Q j ≤ deg Φ .   In this paper we present a more general result about solutions to the equation
Φ = ∑ 1 m F j Q j , (1.3)
where F 1 , … , F m   are given polynomials in C n   , with control of the degrees of F j Q j   . It is formulated in terms of a residue current associated with F j   with support on their common zero set on P n   , and the theorems of Macaulay and Max Nöther are simple consequences. We also provide explicit representation formulas of solutions.
If f j   denote homogenizations of F j   , i.e., f j ( z ) = z 0 d j F j ( z ′ / z 0 )   , where d j ≥ deg F j   , (here z = ( z 0 , z 1 , … , z n )   and z ′ = ( z 1 , … , z n )   ), then each f j   defines a global holomorphic section of the line bundle L d j → P n   , and hence f = ( f 1 , ⋯ , f m )   is a section of the rank m   bundle E * = L d 1 ⊕ ⋯ ⊕ L d m   over P n   (here L s   denotes the line bundle O ( s )   ). If z ∈ C n + 1 \ { 0 }   we let [ z ]   denote the corresponding point in P n   under the natural projection; however, we write f ( z )   rather than f ( [ z ] )   . If E *   is equipped with the natural Hermitian structure, then
∥ f ( z ) ∥ 2 = ∑ 1 m | f j ( z ) | 2 | z | 2 d j . (1.4)
Following [2] we can define the residue current R f   which is an element in ⊕ ℓ D 0 , ℓ ′ ( P n , Λ ℓ E )   and with support on the zero set Z f = { [ z ] ∈ P n ; f ( z ) = 0 } .   If we assume that the polynomials F j   have no common zeros in C n   , then of course Z f   is a subset of the hyperplane at infinity. If c o d i m Z f = m   , i.e., f   is locally a complete intersection, then R f   is a ( 0 , m )   -current with values in det E = L − ∑ d j   ; more precisely it can be identified with the Coleff-Herrera current [ ∂ ¯ 1 f 1 ∧ … ∧ ∂ ¯ 1 f m ] ,   in C n + 1 \ { 0 }   , see Section  2 . We can now formulate our main result in this paper.
Theorem 1.1. Let F 1 , … , F m   be polynomials in C n   , deg F j ≤ d j   , let f = ( f 1 , ⋯ , f m )   be the corresponding section of E * = L d 1 ⊕ ⋯ ⊕ L d m   over P n   , and let R f   be the asssociated residue current. Moreover, assume that
m ≤ n o r r ≥ ∑ j = 1 n + 1 d j − n , (1.5)
where d 1 ≥ d 2 ≥ … ≥ d m   . Let Φ   be a polynomial, deg Φ ≤ r   , and let φ ∈ O ( P n , L r )   denote its r   -homogenization. If
φ R f = 0 , (1.6)
then there are polynomials Q j   such that  1.3 holds and deg F j Q j ≤ r   . If f   is a complete intersection (then the condition  1.5 is fulfilled) and there exist such polynomials Q j   , then  1.6 holds.
It is clear that the conclusion about deg F j Q j   cannot be improved. If Φ = 1   the condition  1.6 means that F j   have no common zeros in C n   and that z 0 r   annihilates the residue R f   at infinity. If Z f   is empty and m = n + 1   (actually any m ≥ n + 1   works) we can choose r = ∑ d j − n   and hence we get a solution to the Bézout equation  1.2 such that deg F j Q j ≤ ∑ d j − n   ; thus we have obtained the theorem of Macaulay mentioned above.
We have the following generalization of Nöther's theorem.
Theorem 1.2. Assume that the projective zero set of F 1 , … , F m   has codimension m   and that there is no irreducible component contained in the hyperplane at infinity. If Φ   belongs to ideal ( F )   , then there are polynomials Q j   such that  1.3 holds and deg F j Q j ≤ deg Φ .  
  • Proof. Since m ≤ n   the condition  1.5 is fulfilled so we can take r = deg Φ   . Since Φ ∈ ( F )   , φ   is in the ideal ( f )   locally in C n   and since f   is a complete intersection, φ R f = 0   in C n   by the duality theorem (see Section  2 ). If m = n   , i.e., as in Nöther's theorem, Z f   is contained in C n ⊂ P n   , so R f   has its support in C n   as well, and hence  1.6 holds in P n   .
    Thus Theorem  1.1 provides the desired solution. In the general case the assumption means that the intersection of Z f   and the hyperplane at infinity has codimension m + 1   , and then Proposition  2.1 in Section  2 implies that φ R f = 0   in P n   . □
Remark 1. Although this theorem is probably known before, we have not found it in the literature. A proof of Nöther's theorem by multivariable residue calculus has previously been obtained by Tsikh, [31] . In [32] is given an argument starting with a representation of Φ   with the Cauchy-Weil formula. Making series expansion of the kernel and using Jacobi formulas (vanishing of certain residues as in [33] ) and the duality theorem, one obtains Nöther's theorem. It is possible that one can prove the general form of Theorem  1.2 in a similar way, following the idea of [4] oto add n − k   linear forms L   such that ( F , L )   has no zeros at infinity, but we have not checked the details.
However some results related to Theorem  1.2 have appeared before.
In [28] , Proposition 2, it is assumed that f j   is a regular sequence in P n   but with no extra condition on the hyperplane at infinity. If Φ   belongs to the ideal ( F )   as above, then there are Q j   solving  1.3 such that deg F j Ψ j ≤ N + deg Φ   , where N = Π 1 m deg F j   . To see this in our setting, recall that (see, e.g., [28] Lemma 2) if F j   is a regular sequence in O x   , then
( ( F ) x ) N ⊂ ( F ) x . (1.7)
Thus z 0 N + deg Φ Φ ( z ′ / z 0 )   annihilates R f   in P n   , and therefore the statement follows from Theorem  1.1 .
Now let F j   be as in Theorem  1.2 and assume that Φ   vanishes on their common zero set in C n   . Then by  1.7 , Φ N   belongs to ( F )   . Therefore we get the following corollary of Theorem  1.2 , which recently appeared in [18] under the slightly stronger assumption that F j   is a strictly regular sequence in C n   .
Corollary 1.3. Assume that the projective zero set of F 1 , … , F m   has codimension m   and that there is no component contained in the hyperplane at infinity. If Φ   vanishes on the zero set of F   in C n   , then there are polynomials Q j   such that deg F j Q j ≤ N deg Φ   and ∑ F j Q j = Φ N   , where N = ( deg F 1 ) ⋯ ( deg F m )   .
If
∥ φ ∥ ≤ C ∥ f ∥ , (1.8)
then, see Section  2 , φ min ( m , n ) R f = 0   , and hence Theorem  1.1 implies
Corollary 1.4. Let F j   and Φ   be as in Theorem  1.1 , r ≥ deg Φ   , and assume that m ≤ n o r r min ( m , n ) ≥ ∑ 1 n + 1 d j − n .   If  1.8 holds, then there are polynomials Q j   such that
∑ F j Q j = Φ min ( m , n ) (1.9)
and deg F j Q j ≤ r min ( m , n )   .
Since there are examples where f   is a complete intersection and the full power min ( m , n )   of φ   is needed to kill R f   , this result is then sharp.
Example 1. Let M   be a given positive integer. Take F j ( z ′ ) = z j M m   in C n   , 1 ≤ j ≤ m ≤ n   , Φ ( z ′ ) = ( z 1 + ⋯ z m ) M m   , and let f j   and φ   be the homogenizations as before ( d j = M m   ). Then  1.8    holds and hence the corollary states that  1.9 has a solution such that deg F j Q j ≤ r min ( m , n ) = M m 2   . This is obvious also by a direct inspection, and one also immediately sees that Φ m − 1   is not in the ideal ( F )   . Thus the corollary is sharp.
It follows that φ m − 1 R f ≠ 0   , and since f   is a complete intersection in P n   , in fact Z f   is the n − m   -plane { [ z ] ∈ P n ; z 1 = ⋯ z m = 0 }   , it also follows that φ m R f = 0   . One can also verify these residue conditions directly. In fact, in the standard affine coordinates z ′   , R f = [ ∂ ¯ 1 z 1 M m ∧ … ∧ ∂ ¯ 1 z m M m ] ∧ ε ,   where ε   is a non-vanishing section of the line bundle det E   , see Section  2 .
Since this residue current is a tensor product of one-variable currents, the residue conditions follow from the one-variable equality z ∂ ¯ [ 1 / z p + 1 ] = ∂ ¯ [ 1 / z p ]   .
Let F j   be polynomials with no common zeros in C n   . Since the zero set of the section f   (take d j = deg F j   ) is then contained in the hyperplane at infinity it follows from Lojasiewicz' inequality that
∥ z 0 ∥ M ≤ C ∥ f ∥ (1.10)
for some M   , or equivalently,
∑ 1 m | F j ( z ′ ) | 2 ( 1 + | z ′ | 2 ) d j ≥ c 1 ( 1 + | z ′ | 2 ) M . (1.11)
Under this condition z 0 M min ( m , n ) R f = 0   , so we have
Corollary 1.5. Let F 1 , … , F m   be polynomials in C n   of degrees d j   such that  1.11 (or equivalently  1.10 ) holds for some number M   , and assume that m ≤ n o r M min ( m , n ) ≥ ∑ 1 n + 1 d j − n .   Then there is a solution to ∑ F j Q j = 1   with deg F j Q j ≤ min ( m , n ) M   .
Example 2. Also Corollary  1.5 is essentially sharp. Let M   be a given non-negative integer and take F j ( z ′ ) = z j M   , 1 ≤ j < m ≤ n   , and F m ( z ′ ) = ( 1 + z 1 + ⋯ + z m − 1 ) M   . Then f j = z j M   and f m = ( z 0 + ⋯ + z m − 1 ) M   , so  1.11 holds. The corollary thus gives a solution to  1.2 with deg F j Q j ≤ m M   .
Writing 1 = ( 1 + z 1 + ⋯ + z m − 1 ) − z 1 − ⋯ − z m − 1   and taking the power M m − m + 1   we get a solution to  1.2 with deg F j Q j = M m − m + 1   , and it is easily seen to be the best possible, cf., Example  1 . However, for large M   , M m − m + 1   is close to M m   .
Remark 2. Given the estimate  1.11 , one can obtain a solution to the Bézout equation  1.2 by a direct application of Skoda's L 2   -estimate from [29] , as is done in [13] . If we for simplicity assume that all d j = d   , then one gets a solution with deg Q j ≤ min ( m + 1 , n ) M − d   , i.e., deg F j Q j ≤ min ( m + 1 , n ) M   . For m ≤ n   this is the same as in Corollary  1.5 but for m > n   it is strictly weaker. This phenomenon is the same as in the original proof of Briançon-Skoda's theorem, [12] . Under the assumption ∥ φ ∥ ≤ C ∥ f ∥   , the (local) L 2   -estimate immediately implies that φ min ( m , n + 1 )   belongs the the ideal ( f )   locally; to obtain the correct result when m > n   , that the power n   is enough, an additional argument is required. See also Section  2 .
Remark 3. The main step in Brownawell's paper [13] is to obtain good control of the power M   in  1.11 in terms of the degrees of F j   , assuming that they have no common zeros in C n   , and this is done by means of Chow forms, see also [30] .
Kollár's theorem implies that the estimate  1.11 holds with M = N ( d 1 , … , d m )   , see, [22] , and this is in fact best possible. From this estimate one gets, via Corollary  1.5 , a solution to  1.2 with deg F j Q j ≤ min ( m , n ) M   . In view of Kollár's theorem one has then “lost” the factor min ( m , n )   .
Remark 4. Kollár's theorem holds for any field. Berenstein and Yger, [5] , have obtained explicit solutions to the Bézout equation  1.2 in subfields of C   , by means of integral formulas; see also [8] and the more recent survey article [32] for a thorough discussion.
Remark 5. The condition  1.8 means that φ   locally on P n   belongs to the integral closure of the ideal ( f )   . In [20] , Hickel proves that if Φ   is in the integral closure of ( F )   in C n   , then one can solve (assuming m ≤ n   for simplicity) Φ m = ∑ F j Q j   with deg ( F j Q j ) ≤ m deg Φ + m d 1 ⋯ d m   .
This result would follow from Theorem  1.1 if one could prove that the current z 0 m d 1 ⋯ d m φ m R f   vanishes ( φ   is the deg Φ   homogenization of Φ   ).
In C n   it vanishes since | Φ | ≤ C | F |   locally. If the zero set is contained in { z 0 = 0 }   , the current vanishes there by Kollár's theorem. We do not know whether it vanishes in the general case.
Theorem  1.1 is a special case of the following more general result, for which we formulate only the homogeneous version. Let δ f   denote the mapping ℰ ( P n , Λ ν + 1 E ⊗ L r ) → ℰ ( P n , Λ ν E ⊗ L r )   defined as interior multiplication with the section f   of E *   . Thus for instance, if q = ( q 1 , … , q m )   is a section to E ⊗ L r   , then δ f q   is equal to the section ∑ j f j q j   of L r   . Moreover, let ∇ f = δ f − ∂ ¯   .
Theorem 1.6. Let f   be holomorphic section of E * = L d 1 ⊕ ⋯ ⊕ L d m   and assume that ℓ ≥ 0   is given and that m − ℓ ≤ n o r r ≥ ∑ j = 1 n + ℓ + 1 d j − n ,   where d 1 ≥ d 2 ≥ … ≥ d m   . If φ ∈ O ( P n , Λ ℓ E ⊗ L r )   , then φ = δ f ψ   for some ψ ∈ O ( P n , Λ ℓ + 1 E ⊗ L r )   if and only if
∇ f ( w ∧ R f ) = φ ∧ R f (1.12)
for some smooth w   defined in a neighborhood of Z f   .
If ℓ > m − p   ( p = c o d i m Z f   ) then the condition on φ   is void; if ℓ = m − p   , it means that φ ∧ R f = 0   , see the remarks after Theorem  2.3 below. If f   is a complete intersection, then m ≤ n   and therefore we have
Corollary 1.7. Let f   be a holomorphic section of E * = L d 1 ⊕ ⋯ ⊕ L d m   that is a complete intersection, and assume that r ≥ 0   . If φ ∈ O ( P n , L r )   , then φ = f ⋅ q   is solvable with q ∈ O ( P n , E ⊗ L r )   if and only if φ R f = 0   .
  • Proof of Theorem  1.1 . If the hypotheses in Theorem  1.1 are fulfilled, then Theorem  1.6 provides a section q = ( q 1 , … , q m )   of E ⊗ L r   such that ∑ f j q j = δ f q = φ   ; here q j   are sections of L − d j + r   . After dehomogenization this means that Q j   are polynomials such that deg F j Q j ≤ r   . □
In Section  2 we recall the necessary background from [2] about the residue currents, and present a general result about the image of a holomorphic morphism f   . Combined with well-known vanishing results for the line bundles L r → P n   it leads to a proof of Theorem  1.6 .
In the last section we construct explicit integral representations of the solutions in Theorem  1.1 . They give essentially the same results except for a small loss of precision. The construction is based on ideas in [1] and [2] .

2 The residue current of a holomorphic section

Let E → X   be a holomorphic Hermitian vector bundle of rank m   over the n   -dimensional complex manifold X   , and let f   be a holomorphic section of the dual bundle E *   , or in other words, a holomorphic morphism f : E → X × C   . Let ℒ r = ⊕ ℓ D 0 , l + r ′ ( X , Λ ℓ E ) ;   we consider ℒ r   as a subbundle to Λ ( T 0 , 1 * ⊕ E )   , so that δ f   (i.e., interior multiplication with f   ) and ∂ ¯   anticommutes. Then ∇ f = δ f − ∂ ¯   induces the complex → ℒ r − 1 → ℒ r →   . It is readily checked that ∇ f   satisfies the Leibniz rule ∇ f ( α ∧ β ) = ∇ f α ∧ β + ( − 1 ) ν α ∧ ∇ f β   , where ν   is the total degree of β   . Let s   be the dual section of E   of f   so that in particular δ f s = ∥ f ∥ 2   . In [2] we defined the current R f = ∂ ¯ ∥ f ∥ 2 λ ∧ s ∇ f s | λ = 0 ;   for large R e λ   the right hand side is integrable and therefore a well defined current, and by a nontrivial argument based on Hironaka's theorem one can make an analytic continuation to λ = 0   . The resulting current is an element in ℒ 0   with support on Z f = { z ; f ( z ) = 0 }   and it satisfies the basic equality
∇ f U f = 1 − R f , (2.1)
where U f ∈ ℒ − 1   is defined as U f = ∥ f ∥ 2 λ s ∇ f s | λ = 0 .   Moreover,
R f = R p , p f + … + R m , m f , (2.2)
where p = c o d i m Z f   ; here lower index ℓ , q   means that the current has bidegree ( 0 , q )   -form and takes values in Λ ℓ E   .
Proposition 2.1. Assume that f   defines a complete intersection and that h   is a holomorphic section of some line bundle such that { h = 0 } ∩ Z f   has codimension m + 1   . If φ   is a holomorphic section such that φ R f = 0   in X \ { h = 0 }   , then φ R f = 0   .
Notice that since f   is a complete intersection, R f = R m f   . The following lemma, which is the core of the proof, states that then R f   is robust in a certain sense.
Lemma 2.2. The current | h | 2 λ R f   has an analytic continuation to R e λ > − ε   and | h | 2 λ R f | λ = 0 = R f .  
  • Proof. Clearly the statement is local. By Hironaka's theorem and a toric resolution we may assume that f = f 0 f ′   , where f 0   is a holomorphic function and f ′   is a non-vanishing section. In this way we can write the action of R f   on a test form ξ   as a finite sum of terms like ∫ ∂ ¯ [ 1 f 0 ℓ ] ∧ α ∧ ξ ~ ρ ,   where [ 1 / f 0 ℓ ]   is the principal value current, α   is a ( 0 , m − 1 )   -form, ξ ~   is the pull-back of ξ   in the given resolution, and ρ   is a cut-off function.
    We may also assume that f 0 = τ k 1 α 1 ⋯ τ k ν α ν ,   in appropriate local coordinates τ j   , and therefore the integral is a sum of terms like
    ∫ [ Π r ≠ j 1 τ k r α r ℓ ∂ ¯ 1 τ k j α j ℓ ] ∧ α ∧ ξ ~ ρ . (2.3)
    We may also assume that h = τ m 1 β 1 ⋯ τ m μ β μ u ,   where u ≠ 0   . Thus | h | 2 λ R . ξ   is a finite sum of terms like
    ∫ | τ m 1 | 2 λ β 1 ⋯ | τ m μ | 2 λ β μ | u | 2 λ [ Π r ≠ j 1 τ k r α r ℓ ∂ ¯ 1 τ k j α j ℓ ] ∧ α ∧ ξ ~ ρ . (2.4)
    If one of the m i   is equal to k j   , then clearly this integral vanishes for R e λ > > 0   , and trivially therefore it has an analytic continuation to λ > − ε   , with the value 0   at λ = 0   . However, since τ k j   is a factor in both h   and f 0   , and c o d i m { h = 0 } ∩ Z = m − 1   , for degree reasons it follows that ξ   vanishes on this set, and therefore, cf. e.g., [7] , [27] or [2] , each term in ξ ~   contains either a factor τ ¯ k j   or d τ ¯ k j   . In any case, this implies that already the integral  2.3 vanishes. On the other hand, if no m i   is equal to τ k j   , it is easy to see that  2.4 has an analytic continuation to R e λ > − ε   and takes the value  2.3 at λ = 0   . In fact, this follows easily since if [ 1 / s ℓ ]   is the usual principal value distribution in C   and v > 0   is smooth and strictly positive, then | s | 2 λ v λ [ 1 / s ℓ ]   has an analytic continuation to R e λ > − ε   and takes the value [ 1 / s ℓ ]   at λ = 0   . Thus the proposition is proved. □
  • Proof of Proposition  2.1 . By assumption φ R f   is a current with support on { h = 0 }   , and hence (locally) | h | 2 λ φ R f = 0   if R e λ > > 0   . From Lemma  2.2 it follows that φ R f = | h | 2 λ φ R f | λ = 0 = 0 .   □
Let L → X   be a holomorphic line bundle and let φ   be a holomorphic section of Λ k E ⊗ L   .
Theorem 2.3. Let ℓ ≥ 0   and suppose that H 0 , s ( X , Λ s + ℓ + 1 E ⊗ L ) = 0   for all 1 ≤ s ≤ m − ℓ − 1   . Moreover, let φ ∈ O ( X , Λ ℓ E ⊗ L )   .
Then δ f ψ = φ   has a solution ψ ∈ O ( ( X , Λ ℓ + 1 E ⊗ L ) )   if and only if there is a smooth solution w   , defined in a neighborhood of Z f   , to
∇ f ( w ∧ R f ) = φ ∧ R f . (2.5)
In view of  2.2 , the condition on φ   is void if ℓ > m − p   . Moreover, since w = w ℓ + 1 , 0 + w ℓ + 2 , 1 + ⋯   the condition means precisely that φ ∧ R f = 0   if ℓ = m − p   . In the case ℓ = 0   and p = m   , i.e., f   defines a complete intersection, we get back the well-known duality theorem, first proved in [16] and [25] .
It was also proved in [2] that h min ( m , n ) R f = 0   if h   is holomorphic and ∥ h ∥ ≤ C ∥ f ∥   . The local version of Theorem  2.3 therefore immediately implies the Briançon-Skoda theorem, [12] ,: If ∥ φ ∥ ≤ C ∥ f ∥   , then locally φ min ( m , n )   belongs to the ideal ( f )   . There is also an explicit representation formula in [2] .
  • Proof of Theorem  2.3 . First suppose that the holomorphic solution ψ   exists. Then ∇ f ψ = φ   and hence ∇ f ( ψ ∧ R f ) = φ ∧ R f   since ∇ f R f = 0   .
    Conversely, if  2.5 holds for some smooth w   , we claim that ∇ f v = φ   , if v = ( − 1 ) ℓ φ ∧ U f + w ∧ R f .   In fact, since ∇ f φ = 0   , ∇ v = φ ∧ ∇ f U f + ∇ f ( w ∧ R f ) = φ ∧ ( 1 − R f ) + φ ∧ R f = φ .   This means that ∂ ¯ v m , m − ℓ − 1 = 0 a n d δ f v k + 1 , k − ℓ = ∂ ¯ v k , k − ℓ − 1 .   By the assumption on the Dolbeault cohomology, we can successively solve the equations ∂ ¯ η m , m − ℓ − 2 = v m , m − ℓ − 1 , ∂ ¯ η k , k − ℓ − 2 = v k , k − ℓ − 1 + δ f η k + 1 , k − ℓ − 1 , k ≥ ℓ ,   and then finally ψ = v ℓ , 0 + δ f η ℓ + 1 , 0   is the desired holomorphic solution.
    □
Example 3. Suppose that X   is a compact and L   is a strictly positive line bundle. Then there is an r 0 >   such that H 0 , k ( X , Λ E ∙ ⊗ L r ) = 0   for all k ≥ 1   if r ≥ r 0   . If f   is a holomorphic section of E *   , then a holomorphic section φ ∈ O ( Λ ℓ E ⊗ L r )   , r ≥ r 0   , is in the image of the morphism
O ( X , Λ ℓ + 1 E ⊗ L r ) → O ( X , Λ ℓ E ⊗ L r ) (2.6)
if φ ∧ R f = 0   . If ℓ = m − p   the condition is necessary.
We shall now focus on the case where X = P n   and E   is the Hermitian vector bundle from Section 1. Let E 1 , … , E m   be trivial line bundles over P n   with basis elements ε 1 , … , ε m   , and let E j *   be the dual bundles, with bases ε j *   . Then we have that E * = ( L d 1 ⊗ E 1 * ) ⊕ ⋯ ⊕ ( L d m ⊗ E m * ) ,   E = ( L − d 1 ⊗ E 1 ) ⊕ ⋯ ⊕ ( L − d m ⊗ E m ) ,   and for instance our section f   can be written f = ∑ 1 m f j ε j * .   Its dual section s   is then, cf.,  1.4 , s = ∑ j f j ( z ) ¯ | z | 2 d j ε j ,   so R f = ∂ ¯ ∥ f ∥ 2 λ ∧ ∑ ℓ + 1 m s ∧ ( ∂ ¯ s ) ℓ − 1 ∥ f ∥ 2 ℓ | λ = 0 .   In C n = { z 0 ≠ 0 } ⊂ P n   we have the coordinates z ′   and the natural holomorphic frame e j = z 0 − d j ε j   and its dual e j * = z 0 d j ε j *   . If f j ′ ( z ′ ) = f j ( 1 , z ′ )   then f = ∑ 1 m f j ′ e j *   and s = ∑ 1 m f j ′ ( z ′ ) ¯ ( 1 + | z ′ | ) d j e j .   When c o d i m Z f = m   , the residue current R f   is independent of the metric, it just contains the top degree term R m , m f   , and in fact, see [2] , R f = [ ∂ ¯ 1 f m ′ ∧ … ∧ ∂ ¯ 1 f 1 ′ ] ∧ e 1 ∧ … ∧ e m ,   where the expression in brackets is a Coleff-Herrera residue current.
Choosing the local coordinates z 0 , ζ 1 , … , ζ n   in C n + 1 \ { 0 }   , where ζ j = z j / z 0   , it is easy to see that π * [ ∂ ¯ 1 f m ′ ∧ … ∧ ∂ ¯ 1 f 1 ′ ] = z 0 ∑ d j [ ∂ ¯ 1 f m ∧ … ∧ ∂ ¯ 1 f 1 ] ,   and hence we can identify R f   with the Coleff-Herrera current [ ∂ ¯ 1 f m ∧ … ∧ ∂ ¯ 1 f 1 ] ∧ ε 1 ∧ … ∧ ε m   in C n + 1 \ { 0 }   .
  • Proof of Theorem  1.6 . It is well-known, see, e.g., [15] , that H 0 , k ( P n , L ν ) = 0   for all ν   if 1 ≤ k ≤ n − 1   and that H 0 , n ( P n , L ν ) = 0   if (and only if ) ν ≥ − n   . Since E = L − d 1 ⊕ ⋯ ⊕ L − d m   we have that Λ ν E ⊗ L r = ′ ⊕ | J | = ν L − d J 1 ⊗ ⋯ ⊗ L − d J ν ⊗ L r = ′ ⊕ | J | = ν L r − d J 1 ⋯ − d J ν .   Thus H 0 , s ( P n , Λ s + ℓ + 1 E ⊗ L r ) = 0   for 1 ≤ s ≤ m − ℓ − 1   if either m − ℓ − 1 ≤ n − 1   or r − ∑ 1 n + ℓ + 1 d j ≥ − n .   Now Theorem  1.6 follows from Theorem  2.3 . □

3 Integral representation

The aim of this section is to present an explicit integral representation of the solution Q j   to the division problem in Theorem  1.1 . We have
Theorem 3.1. Let F 1 , … , F m , Φ   be polynomials in C n   , let f   and R f   be as before, and let φ   be the r   -homogenization of Φ   ( deg Φ ≤ r   ). Then there is an explicit decomposition
Φ ( z ′ ) = ∑ 1 m F j ( z ′ ) ∫ P n T j ( ζ , z ′ ) φ ( ζ ) + ∫ P n S ( ζ , z ′ ) ∧ R f ( ζ ) φ ( ζ ) , (3.1)
where T j ( ζ , z ′ ) , S ( ζ , z ′ )   are smooth forms (in [ ζ ]   ) on P n   and holomorphic polynomials in z ′   , such that deg z ′ ( F j ( z ′ ) T j ( ζ , z ′ ) ) ≤ d 1 + d 2 + ⋯ + d μ + 1 + r ,   if μ = min ( n , m − 1 )   and d 1 ≥ d 2 ≥ ⋯ ≥ d m   .
Thus, if φ R f = 0   we get back the conclusion of Theorem  1.1 but with the extra term d 1 + ⋯ + d μ + 1   in the estimate of the degree.
For fixed z ∈ C n   , η = 2 π i ∑ 0 n z j ∂ ∂ ζ j   is an L z ⊗ L ζ − 1   -valued ( 1 , 0 )   -form on P n   , and if δ η   denotes interior multiplication with η   , then δ η : D ℓ + 1 , 0 ′ ( P n , L r + 1 ) → D ℓ , 0 ′ ( P n , L r ) .  
Remark 6. When we say that η   is a section of L z ⊗ L ζ − 1   rather than L − 1 = L ζ − 1   , we just indicate that it is 1   -homogeneous in z   ; it would be more correct, but less convenient, to consider η   as a section of the bundle L z ⊗ L ζ − 1 ⊗ ( T ζ * ) 0 , 1   over P z n × P ζ n   .
Let ∇ η = δ η − ∂ ¯   . Notice that if α = α 0 + α 1 = z ⋅ ζ ¯ | ζ | 2 − ∂ ¯ ζ ¯ ⋅ d ζ 2 π i | ζ | 2 ,   then the first term, α 0   , is a section of L z ⊗ L ζ − 1   and the second term, α 1   , is a projective form (since δ ζ α 1 = 0   ); moreover
∇ η α = 0 . (3.2)
We have the following basic integral representation of global holomorphic sections of L r   .
Proposition 3.2. Assume that r ≥ 0   and that φ ∈ O ( P n , L r )   .
Then φ ( z ) = ∫ P n α n + r φ .  
For degree reasons, actually φ ( z ) = ( n + r ) ! n ! r ! ∫ P n α 0 r ∧ α 1 n φ ;   this formula appeared already in [11] ; expressed in affine coordinates it is the well-known weighted Bergman representation formula for polynomials in C n   . However, we prefer to supply a direct proof on P n   , following the ideas in [1] .
  • Proof. Let σ   be the L z − 1 ⊗ L ζ ⊗ T 1 , 0 * ( P ζ n )   valued ( 1 , 0 )   -form on P n   that is dual, with respect to the natural metric, to η   . Then, since η   has a first order zero at [ z ]   (and no others), it follows (see [1] ) that ∇ η σ ∇ η σ = 1 − [ [ z ] ] .   The rightmost term is the L z − n ⊗ L ζ n   -valued ( n , n )   -current point evaluation at [ z ]   for sections of L − n   . If φ   is a global holomorphic section of L r   it follows by  3.2 that ∇ η ( σ ∇ η σ ∧ α n + r φ ) = φ α n + r − φ [ [ z ] ] ,   where this time the last term is φ   times the L z r ⊗ L ζ − r   -valued current point evaluation at [ z ]   . If we integrate this equality over P n   we get the desired representation formula. □
Let E 1 , … , E m   be the trivial line bundles over P n   with basis elements ε 1 , … , ε m   , so that E = ( L − d 1 ⊗ E 1 ) ⊕ ⋯ ⊕ ( L − d m ⊗ E m )   as in Section  2 .
We also introduce disjoint copies E ~ j   of E j   with bases ε ~ j   and the bundle E ~ = ( L − d 1 ⊗ E ~ 1 ) ⊕ ⋯ ⊕ ( L − d m ⊗ E ~ m ) .   Let Λ   be the exterior algebra bundle over the direct sum of all the bundles E   , E ~   , E *   , and T * ( P n )   . Any form γ   with values in Λ   can be written uniquely as γ = γ ′ ∧ ( ∑ ε j * ∧ ε j ) m / m ! + γ ′ ′   where γ ′ ′   denotes terms that do not contain a factor ( ∑ ε j * ∧ ε j ) m / m !   , and we define ∫ ε γ = γ ′ .   We have a globally defined form τ = ∑ 1 m ε j * ∧ ( ε j − ε ~ j ) .   From now on we consider [ z ]   as a fixed arbitrary point in C n ⊂ P n   , and let z = ( 1 , z ′ )   . We also introduce the section f z = ∑ j ζ 0 d j f j ( 1 , z ) ε j * = ∑ ζ 0 d j F j ( z ′ ) ε j *   of E *   and let f ~ z   be the corresponding section of E ~ *   .
Lemma 3.3. There is a holomorphic section H = ∑ H j ∧ ε j   of E * ⊗ L ⊗ T 1 , 0 *   , thus H j   are sections of L d j ⊗ L ⊗ T 1 , 0 *   , such that δ η H = f − f z ,   and such that the coefficients in H j   are polynomials in z ′ / z 0   of degrees (at most) d j − 1   .
  • Proof. For each F j ( z ′ )   we can find Hefer functions h j k ( ζ ′ , z ′ )   , polynomials of degree d j − 1   in ( ζ ′ , z ′ )   , such that ∑ k = 1 n h j k ( ζ ′ , z ′ ) ( ζ k − z k ) = F j ( ζ ′ ) − F j ( z ′ ) .   If we then take H j = ζ 0 d j + 1 2 π i ∑ 1 n h j k ( ζ ′ / ζ 0 , z ′ ) d ( ζ k / ζ 0 ) ,   then clearly H j   is a projective ( 1 , 0 )   -form, and moreover, δ η H j = f j ( ζ ) − ζ 0 d j F j ( z ′ )   as wanted. □
Let δ F   denote interior multiplication with the section F = f + f ~ z   of E * ⊕ E ~ *   . Then δ F τ = f − f z = − δ η H   . If ∇ = δ F + δ η − ∂ ¯ ,   thus
∇ ( τ + H ) = 0 . (3.3)
We are now ready to define the explicit division formula.
  • Proof of Theorem  3.1 . From  3.3 it follows that
    ( ∇ η + δ F ) ( e τ + H ∧ U f ) = e τ + H ∧ ( 1 − R f ) . (3.4)
    We can rewrite this as
    δ F ( e τ + H ∧ U f ) + e τ + H ∧ R f = e τ + H − ∇ η ( e τ + H ∧ U f ) . (3.5)
    We claim that the component of full bidegree ( n , n )   of
    ∫ ε [ e τ + H − ∇ η ( e τ + H ∧ U f ) ] ∧ α n + r φ (3.6)
    is equal to ( n + r ) ! n ! r ! α 1 n α 0 r φ + ∂ ¯ ( ⋯ )   where ( ⋯ )   is a scalar-valued ( n , n − 1 )   -form. In fact, since α n + r   has bidegree ( * , * )   the factor U ℓ , ℓ − 1   must be combined with H ℓ   , and then it follows that τ   can be replaced by ω = ∑ j ε j * ∧ ε j   . Observe that the component of U ℓ , ℓ − 1   with basis element ε J 1 ∧ … ∧ ε J ℓ   takes values in L − ( d J 1 + ⋯ + d J ℓ )   , whereas the component of H ℓ   with basis element ε J 1 * ∧ … ∧ ε J ℓ *   takes values in L d J 1 + ⋯ + d J ℓ ⊗ L ℓ   . The product of these two factors must be combined with α 1 n − ℓ α 0 ℓ + r φ   which gives a scalar-valued ( n , n )   -form as claimed. Thus we can integrate  3.6 over P n   , and by Proposition  3.2 and Stokes' theorem it is equal to φ ( z )   .
    We now consider the left hand side of  3.5 multiplied with α n + r φ   . To begin with, ∫ P n ∫ ε e τ + H ∧ R f ∧ α n + r φ   is well defined with the same argument as above, and again one can replace τ   by ω   . Moreover, since α n + r φ   contains no ε j   , ∫ ε δ f ( e τ + H ∧ U f ) ∧ α n + r φ = ∫ ε δ f ( e τ + H ∧ U f ∧ α n + r φ ) = 0 .   Since δ f ~ z ∑ j ε ~ j ∧ ε j * = ∑ j F ( z ′ ) ζ 0 d j ε j * = f z ,   another computation shows that the component of bidegree ( n , n )   of ∫ ε δ f ~ z ( e τ + H ∧ U f ) ∧ α n + r φ   is equal to ∫ ε f z ∧ ∑ k = 0 m − 1 ω m − k − 1 ∧ H k ∧ U k + 1 , k ∧ α 1 n − k α 0 k + r φ .   Again one can check that this form is scalar valued. Summing up we have the desired decomposition  3.1 with
    S ( ζ , z ′ ) ∧ R f ( ζ ) = ∫ ε e ω + H ∧ R f ∧ α n + r = ∑ k = c o d i m Z f m ∫ ε ( n + r ) ! ( n − k ) ! ( k + r ) ! ω m − k ∧ H k ∧ R k , k f α 1 n − k α 0 k + r ,  
    and
    T j ( ζ , z ′ ) = ∫ ε ε j * ζ 0 d j ∧ ∑ k = 1 m − 1 ( n + r ) ! ( n − k ) ! ( k + r ) ! I ~ n − k − 1 ∧ H k ∧ U k + 1 , k ∧ α 1 n − k α 0 k + r φ ,  
    Both α   and H   are polynomials in z ′   so it just remains to check the degrees of T j   . The worst case occur when k   is as large as possible which is k = μ = min ( m − 1 , n )   . Then the factor α 0 k + r   has degree k + r   . Recall that H = ∑ H ℓ ∧ ε ℓ *   and that deg H ℓ = d ℓ − 1   . The term H j   cannot occur, because of the presence of ε j *   , and thus we get that d j + deg Q j   is at most d 1 − 1 + d 2 − 1 + ⋯ d μ + 1 − 1 + 1 + μ + r = d 1 + ⋯ + d μ + 1 + r   . □
The division formula constructed here, Theorem  3.1 , is a generalization to P n   of the formula in [2] , which was used to give an explicit representation of the solutions in the local version of Theorem  2.3 ; in particular it provided the first known explicit proof of the Briançon-Skoda theorem.
This division formula is based on the ideas in [1] and it differs from Berndtsson's classical formula, [10] , in some respects. To begin with our formula works also for sections with values in Λ ℓ E   , although in this paper we have only generalized the scalar-valued part to P n   . The more interesting novelty with regard to this paper, is that the residue term contains precisely the factor φ R f   , so that our formula provides a solution of the division problem as soon as φ R f = 0   (or φ R f = ∇ f ( w ∧ R f )   for some smooth w   ). One can obtain a similar formula involving residues (but not precisely R f   except for the complete intersection case) from Berndtsson's formula; this was first done by Passare in [25] , and various variants have been used by several authors since then, see [8] and the references given there. These formulas all go back to the construction of weighted integral formulas in [9] . However, the division formula in [2] , even in the simplest case, when f   is nonvanishing, could not have been obtained from [9] , because the required choice of weight, see formula (2.12) in Remark 3 in [3] , is not encompassed by the method in [9] , but the more general construction in [1] is needed.
Ackowledgement I am indebted to the referee for his careful reading and for his many important remarks and constructive suggestions that have helped to clarify and improve the final version of this paper.
References

  1. M. Andersson: Integral representation with weights I, Math. Ann. 326 (2003), 1–18.
  2. M. Andersson: Residue currents and ideals of holomorphic functions, Bull. Sci. Math. 128, (2004), 481–512.
  3. M. Andersson: Ideals of smooth functions and residue currents, J. Funtional Anal. 212 (2004), 76–88.
  4. C. Berenstein & B. A. Taylor: Interpolation problems in C n   with applications to harmonic analysis, J. Analyse Math. 38 (1980), 188–254.
  5. C. Berenstein & A. Yger: Effective Bézout Identities in Q [ z 1 , … , z n ]   , Acta Mathematica, 166 (1991), 69–120.
  6. C. Berenstein & A. Yger: Green currents and analytic continuation, J. Anal. Math. 75 (1998), 1–50.
  7. C. Berenstein & A. Yger: Analytic residue theory in the non-complete intersection case, J. Reine Angew. Math. 527 (2000), 203–235.
  8. C. Berenstein & R. Gay & A. Vidras & A. Yger: Residue Currents and Bézout Identities, Birkhäuser (1993).
  9. B. Berndtsson & M. Andersson: Henkin-Ramirez formulas with weights, Ann. Inst. Fourier 32 (1982), 91–110.
  10. B. Berndtsson: A formula for division and interpolation, Math. Ann. 263 (1983), 113-160.
  11. B. Berndtsson: Integral formulas on projective space and the Radon transform of Gindikin-Henkin-Polyakov, Publ. Mat. 32 (1988), 7–41.
  12. J. Brianc on & H. Skoda: Sur la clôture intégrale d'un idéal de germes de fonctions holomorphes en un point de C n   ., C. R. Acad. Sci. Paris Sér. A 278 (1974), 949–951.
  13. W. D. Brownawell: Bounds for the degrees in Nullstellensatz, Ann. of Math. 126 (1987), 577–592.
  14. W. D. Brownawell: A pure prime product version of the Hilbert Nullstellensatz, Mich. Math, J. 45 (1998), 581–597.
  15. J-P Demailly: Complex Analytic and Differential Geometry, Monograph Grenoble (1997).
  16. A. Dickenstein & C. Sessa: Canonical representatives in moderate cohomology, Invent. Math. 80 (1985), 417–434..
  17. L. Ein & R. Lazarsfeld: A geometric effective Nullstellensatz, Invent. math. 135 (1999), 427–448.
  18. A. Fabiano & A. Ploski & P. Tworzewski: Effective Nullstellensatz for strictly regular sequences, Univ. Iagel. Acta Math. 38 (2000), 163–167.
  19. P. Griffiths & J. Harris: Principles of algebraic geometry, Pure and Applied Mathematics. Wiley-Interscience [John Wiley & Sons], New York, (1978).
  20. M. Hickel: Solution d'une conjecture de C. Berenstein-A. Yger et invariants de contact à l'infini, Ann. Inst. Fourier 51 (2001), 707–744.
  21. Z. Jelonek: On the effective Nullstellensatz, Preprint (2004).
  22. J. Kollar: Sharp effective Nullstellensatz, J. American Math. Soc. 1 (1988), 963–975.
  23. F.S. Macaulay: The algebraic theory of modular systems, Cambridge Univ. Press, Cambridge 1916.
  24. M. Nother: Über einen Satz aus der Theorie der algebraischen Functionen, Math. Ann. (1873), 351–359.
  25. M. Passare: Residues, currents, and their relation to ideals of holomorphic functions, Math. Scand. 62 (1988), 75–152.
  26. M. Passare: A calculus for meromorphic currents, J. Reine Angew. Math. 392 (1988), 37-56.
  27. M. Passare & A. Tsikh & A. Yger: Residue currents of the Bochner-Martinelli type, Publ. Mat. 44 (2000), 85-117.
  28. B. Shiffman: Degree bounds for the division problem in polynomial ideals, Michigan Math. J. 36 (1989), 163–171.
  29. H. Skoda: Application des techniques L 2   à la théorie des idéaux d'une algèbre de fonctions holomorphes avec poids, Ann. Sci. École Norm. Sup. 5 (1972), 545–579.
  30. B. Teissier: Résultats récents d'algèbre commutative effective, Séminaire Bourbaki, Vol. 1989/90. Astérisque No. 189-190 (1990).
  31. A. Tsikh: Multidimensional residues and their applications, Transl. Amer. Math. Soc. 103 (1992).
  32. A. Tsikh & A. Yger: Residue currents. Complex analysis, J. Math. Sci. (N. Y.) 120 (2004), 1916–1971.
  33. A. Vidras & A. Yger: On some generalizations of Jacobi's residue formula, Ann. Sci. Ecole Norm. Sup. 34 (2001), 131–157.

Department of Mathematics, Chalmers University of Technology and the University of Goteborg, S-412 96 GOTEBORG, SWEDEN E-mail address : matsa@math.chalmers.se