, , November 27, 2006

Let

φ, f_{1}, \dots, f_{m}

be holomorphic functions in a neighborhood of the origin in

C^{n}

. The Briançon-Skoda theorem, [8] , states that

φ^{min (n, m)}

belongs to the ideal

(f)

generated by

f_{j}

| φ | \leq C | f |

. This condition is equivalent to that

φ

belongs to the integral closure of the ideal

(f)

The original proof is based on Skoda's

L^{2}

-estimates in [16] , see Remark 1 below, and actually gives the stronger statement that

φ \in (f)

| φ | \leq C | f |^{min (n, m)}

. There are generalizations to more arbitrary rings, see, e.g., [13] .

In general this result cannot be improved but for certain tuples

f_{j}

a much weaker size condition on

φ

is enough to guarantee that

φ

belongs to

(f)

. For instance, the ideal

(f)^{2}

is generated by the

m (m + 1) / 2

functions

g_{j k} = f_{j} f_{k}

, and

| f |^{2} \sim | g |

, so if we apply the previous result we get that

φ \in (f)^{2}

| φ | \leq C | f |^{min (2 n, m (m + 1))}

. However, in this case actually the power

min (n, m) + 1

is enough. In general we have

Theorem 1.1 (Briançon-Skoda). If

f = (f_{1}, \dots, f_{m})

and

φ

are holomorphic at

0

C^{n}

and

| φ | \leq C | f |^{min (m, n) + r - 1}

, then

φ \in (f)^{r}

In [2] we gave a new proof of the case

r = 1

by means of multivariable residue calculus. In this note we extend this method to cover the general case of Theorem 1.1 , and as a by-product we get various related results. In the first one we consider several possibly different tuples.

Theorem 1.2. Let

f_{j}

j = 1, \dots, r

, be

m_{j}

-tuples of holomorphic functions at

0 \in C^{n}

and assume that

| φ | \leq C | f_{1} |^{s_{1}} \dots | f_{r} |^{s_{r}}

for all

s

such that

s_{1} + \dots + s_{r} \leq n + r - 1

and

1 \leq s_{j} \leq m_{j}

Then

φ \in (f_{1}) \dots (f_{r})

Notice that this immediately implies Theorem 1.1 in the case

m \geq n

by just choosing all

f_{j} = f

. In certain cases Theorem 1.2 can be improved, as one can see by taking

f_{j} = f

and

m < n

and compare with Theorem 1.1 . Another case is when all the functions in the various tuples

f_{j}

together form a regular sequence.

Theorem 1.3. Let

f_{j}

j = 1, \dots, m

, be

m_{j}

-tuples of holomorphic functions at

0 \in C^{n}

and assume that the codimension of

{f_{1} = \dots f_{r} = 0}

m_{1} + \dots + m_{r}

. If

| φ | \leq C min (| f_{1} |^{m_{1}}, \dots, | f_{r} |^{m_{r}}),

then

φ \in (f_{1}) \dots (f_{r})

Remark 1. As was mentioned above the Briançon-Skoda theorem follows by direct applications of Skoda's

L^{2}

-estimate if

m \leq n

. In fact, if

ψ

is any plurisubharmonic function, the

L^{2}

-estimate guarantees a holomorphic solution to

f \cdot u = φ

such that

\int_{X \ Z} \frac{| u |^{2}}{| f |^{2 (min (m, n + 1) - 1 + ε)}} e^{- ψ} d V < \infty

provided that

\int_{X \ Z} \frac{| φ |^{2}}{| f |^{2 (min (m, n + 1) + ε)}} e^{- ψ} d V < \infty .

| φ | \leq C | f |^{m}

, the second integral is finite (taking

ψ = 0

) if

ε

is small enough, and thus Skoda's theorem provides the desired solution. The case when

r > 1

is obtained by iteration. If

m > n

a direct use of the

L^{2}

-estimate will not give the desired result. However, see [10] , in this case one can find an

n

-tuple

\tilde{f}

such that

(\tilde{f}) \subset (f)

and

| \tilde{f} | \sim | f |

, and the theorem then follows by applying the

L^{2}

-estimate to the tuple

f^{'}

In the same way, Theorem 1.2 can easily be proved from the

L^{2}

estimate if

m_{1} + \dots + m_{r} \leq n + r - 1

. To see this, assume for simplicity that

r = 2

, and that

| φ | \leq C | f_{1} |^{m_{1}} | f_{2} |^{m_{2}}

. Choosing

ψ = 2 (m_{1} + ε) log | f_{1} |

, Skoda's theorem give a solution to

f_{2} \cdot u = φ

such that

\int_{X \ Z} \frac{| u |^{2}}{| f_{1} |^{2 (m_{1} + ε)}} d V < \infty .

Another application then gives

v_{j}

such that

f_{1} \cdot v_{j} = u_{j}

. This means that

φ

belongs to

(f_{1}) (f_{2})

. However, we do not know whether one can derive Theorem 1.3 from the

L^{2}

-estimate when

m_{1} + \dots + m_{r} > n + r - 1

. □

Now consider a

r \times m

matrix

f_{j}^{k}

of holomorphic functions,

r \leq m

, with rows

f_{1}, \dots, f_{r}

. We let

F

be the

m! / (m - r)! r!

tuple of functions

det (f_{j}^{I_{k}})

for increasing multiindices

I

of length

r

. We will refer to

F

as the determinant of

f

. If

f_{j}

are the rows of the matrix, considered as sections of the trivial bundle

E^{*}

, then

F

is just the section

f_{r} \land \dots f_{1}

of the bundle

Λ^{r} E^{*}

. Our next result is a Briançon-Skoda type result for the tuple

F

. It turns out that it is enough with a much less power than

m! / (m - r)! r!

. Let

Z

be the zero set of

F

and notice that

c o d i m Z \leq m - r + 1

; this is easily seen by Gauss elimination.

Theorem 1.4. Let

F

be the determinant of the holomorphic matrix

f

as above. If

| φ | \leq C | F |^{min (n, m - r + 1)},

then

φ \in (F)

Remark 2. This result is closely related to the following statement which was proved in [3] . Suppose that

φ

is an

r

-tuple of holomorphic functions and let

∥ φ ∥

be the pointwise norm induced by

f

, i.e.,

∥ φ ∥ = det (f f^{*}) 〈 (f f^{*})^{- 1} φ, φ 〉

. If

∥ φ ∥ ≲ | F |^{min (n, m - r + 1)},

then

f ψ = φ

has a local holomorphic solution. □

Remark 3. Another related situation is when

f

is a section of a bundle

E^{*}

φ

takes values in

Λ^{ℓ} E

, and we ask for a holomorphic section

ψ

Λ^{ℓ + 1} E

such that

δ_{f} ψ = φ

, provided that the necessary compatibility condition

δ_{f} φ = 0

is fulfilled. Let

p = c o d i m {f = 0}

. Then a sufficient condition is that

| φ | \leq C | f |^{min (n, m - ℓ)}

ℓ \leq m - p

, whereas there is no condition at all if

ℓ > m - p

, see Theorems 1.2 and 1.4 and Corollary 1.5 in [2] . □

Theorem 1.4 is proved by constructing a certain residue current

R

with support on the analytic set

Z

, such that

R φ = 0

implies that

φ

belongs to the ideal

(F)

locally. The size conditions of

φ

then implies that

R φ = 0

by brutal force, see Theorem 2.3 below. There may be more subtle reasons for annihilation. For instance, in the generic case, i.e., when

c o d i m Z = m - r + 1

, even the converse statement holds; if

φ

is in the ideal

(F)

then actually

R φ = 0

, see Theorem 2.3 (iv). The analogous statement also holds for the equation

f ψ = φ

in Remark 2 , see [3] . These results are therefore extensions of the well-known duality theorem of Dickenstein-Sessa and Passare, [11] and [14] , stating that if

f

is a tuple that defines a complete intersection, i.e.,

c o d i m {f = 0} = m

, then

φ \in (f)

if and only if

φ

annihilates the Coleff-Herrera current defined by

f

. Theorems 1.2 and 1.3 (as well as Theorem 1.1 ) are obtained along the same lines, by an appropriate choice of matrix

f

It has been discussed for several years, see, e.g., [6] and [19] , whether one can prove the Briançon-Skoda theorem with an explicit integral formula. In [2] we discovered such a formula for the case

r = 1

of Theorem 1.1 . In the second part of this paper we construct new completely explicit integral representations of holomorphic functions that provide effective proofs of Theorems 1.1 to 1.4 . In fact, for any holomorphic function

φ

we construct a holomorphic decomposition

\begin{matrix} φ = T φ + S φ, \end{matrix}

(1.1)

such that

T φ

belongs to the ideal in question and

S φ

vanishes as soon as

φ

annihilates the residue current

R

2 The ideal generated by the determinant section

Although we are mainly interested in the local results in this paper it is convenient to adopt an invariant perspective. We therefore assume that we have Hermitian vector bundles

E

and

Q

of ranks

m

and

r \leq m

, respectively, over a complex

n

-dimensional manifold

X

, and a holomorphic morphism

f : E \to Q

. We also assume that

f

is generically surjective, i.e., that the analytic set

Z

where

f

is not surjective has at least codimension

1

. If

ε_{j}

is a local holomorphic frame for

Q

, then

f = f_{1} \otimes ε_{1} + \dots + f_{r} \otimes ε_{r}

, where

f_{j}

are sections of the dual bundle

E^{*}

Moreover,

F = f_{r} \land \dots \land f_{1} \otimes ε_{1} \land \dots \land ε_{r}

is an invariantly defined section of

Λ^{r} E^{*} \otimes det Q^{*}

that we will call the determinant section associated with

f

. Notice that if

e_{j}

is a local frame for

E

with dual frame

e_{j}^{*}

for

E^{*}

, then

f_{j} = \sum_{1}^{m} f_{j}^{k} e_{k}^{*}

, and

F = \sum_{| I | = r}^{'} F_{I} e_{I_{1}}^{*} \land \dots \land e_{I_{r}}^{*},

where the sum runs over increasing multiindices

I

and

F_{I} = det (F_{j}^{I_{k}}) .

Let

S^{ℓ} Q^{*}

be the subbundle of

(Q^{*})^{\otimes ℓ}

consisting of symmetric tensors.

We introduce the complex

(2.1) ⋯ → δ f Λ r + k − 1 E ⊗ S k − 1 Q * ⊗ det Q * → δ f ⋯ → δ f Λ r + 1 E ⊗ Q * ⊗ det Q * → δ f Λ r E ⊗ det Q * → δ F C → 0 ,

where

δ_{f} = \sum_{j} δ_{f_{j}} \otimes δ_{ε_{j}},

δ_{f_{j}}

and

δ_{ε_{j}}

denote interior multiplication on

Λ E

and from the left on

S Q^{*} \otimes det Q^{*}

, respectively, and

δ_{F} = δ_{f}^{r} / r! = δ_{f_{r}} \dots δ_{f_{1}} \otimes δ_{ε_{1}} \dots δ_{ε_{r}} .

It is readily checked that 2.1 actually is a complex. Notice that if

r = 1

, then 2.1 is the usual Koszul complex and therefore exact whenever

f

is pointwise surjective. This is, however, not true when

r > 1

X \ Z

we let

σ_{j}

be the sections of

E

with minimal norms such that

f_{k} σ_{j} = δ_{j k}

. Then

σ = σ_{1} \otimes ε_{1}^{*} + \dots + σ_{r} \otimes ε_{r}^{*}

is the section of

H o m (Q, E)

such that, for each section

φ

Q

v = σ φ

is the solution to

f v = φ

with pointwise minimal norm. We also have the invariantly defined section

σ = σ_{1} \land \dots \land σ_{r} \otimes ε_{r}^{*} \land \dots \land ε_{1}^{*}

Λ E \otimes = det Q^{*}

, and it is in fact the section with minimal norm such that

F σ = 1

, see, e.g., [3] .

Example 1. Assume that

E

and

Q

are trivial and let

ε_{j}

be an ON-frame for

Q

and

e_{j}

an ON-frame for

E

, with dual frame

e_{j}^{*}

. If

F = \sum_{| I | = r}^{'} F_{I} e_{I_{1}}^{*} \land \dots \land e_{I_{r}}^{*}

as above, then

σ = \sum_{| I | = r}^{'} \frac{{\bar{F}}_{I}}{| F |^{2}} e_{I_{1}} \land \dots \land e_{I_{r}} .

□

We will consider

(0, q)

-forms with values in

Λ^{r + k - 1} E \otimes S^{k - 1} Q^{*} \otimes det Q^{*}

, and it is convenient to consider them as sections of

Λ^{r + k + q - 1} (E \oplus T_{0, 1}^{*} (X)) \otimes S^{k - 1} Q^{*} \otimes det Q^{*}

, so that

δ_{f}

anti-commutes with

\bar{\partial}

, and

δ_{F} \bar{\partial} = (- 1)^{r} δ_{F} \bar{\partial}

. In what follows we let

\otimes

denote usual tensor product all

Q^{*}

-factors, and wedge product of

Λ (E \oplus T_{0, 1}^{*} (X))

-factors. Thus for instance

σ \otimes σ = (\sum_{1}^{r} σ_{j} \otimes ε_{j}^{*}) \otimes (σ_{1} \land \dots \land σ_{r} \otimes ε_{1}^{*} \land \dots \land ε_{r}^{*}) = 0 .

Moreover, for each

k \geq 1

(\bar{\partial} σ)^{\otimes (k - 1)}

is a symmetric tensor; more precisely,

\begin{matrix} (\bar{\partial} σ)^{\otimes (k - 1)} = \sum_{| α | = k - 1} (\bar{\partial} σ_{1})^{α_{1}} \land \dots \land (\bar{\partial} σ_{r})^{α_{r}} \otimes ε_{α}^{*}, \end{matrix}

(2.2)

where

ε_{α}^{*} = \frac{(ε_{1}^{*})^{α_{1}} \dot{\otimes} \dots \dot{\otimes} (ε_{r}^{*})^{α_{r}}}{α_{1}! \dots α_{r}!},

and

\dot{\otimes}

denotes symmetric tensor product. For each

k \geq 1

we define in

X \ Z

the

(0, k - 1)

-forms

\begin{matrix} u_{k} = (\bar{\partial} σ)^{\otimes (k - 1)} \otimes σ = σ_{1} \land \dots \land σ_{r} \land (\bar{\partial} σ)^{\otimes (k - 1)} \otimes ε^{*} \end{matrix}

(2.3)

(where

ε^{*} = ε_{r}^{*} \land \dots \land ε_{1}^{*}

), with values in

Λ^{r + k - 1} E \otimes S^{k - 1} Q^{*} \otimes det Q^{*}

Proposition 2.1. In

X \ Z

we have that

\begin{matrix} δ_{F} u_{1} = 1, δ_{f} u_{k + 1} = \bar{\partial} u_{k}, k \geq 1 . \end{matrix}

(2.4)

Proof. Since $δ_{ε_{j}}$ act from the left, and $δ_{f_{j}} \bar{\partial} σ_{ℓ} = 0$ for all $ℓ$ , we have that
$δ_{f} u_{k + 1} = δ_{f} [σ_{1} \land \dots \land σ_{r} \land (\bar{\partial} σ)^{\otimes k} \otimes ε^{*}] = δ_{f} [σ_{1} \land \dots \land σ_{r} \land \bar{\partial} σ] \otimes (\bar{\partial} σ)^{\otimes (k - 1)} \otimes ε^{*} = \sum_{j = 1}^{r} δ_{f_{j}} (σ_{1} \land \dots \land σ_{r}) \land \bar{\partial} σ_{j} \otimes (\bar{\partial} σ)^{\otimes (k - 1)} \otimes ε^{*} = \bar{\partial} (σ_{1} \land \dots \land σ_{r}) \land (\bar{\partial} σ)^{\otimes (k - 1)} \otimes ε^{*} = \bar{\partial} u_{k} .$
Since $δ_{F} u_{1} = F σ = 1$ , the proposition is proved. □

If we let

u = u_{1} + u_{2} + \dots

, and let

δ

denote either

δ_{f}

δ_{F}

, then 2.4 can be written as

(δ - \bar{\partial}) u = 1

. To analyze the singularities of

u

Z

we will use the following lemma (Lemma 4.1) from [3] .

Lemma 2.2. If

F = F_{0} F^{'}

for some holomorphic function

F_{0}

and non-vanishing holomorphic section

F^{'}

, then

s^{'} = F_{0} σ, S^{'} = F_{0} σ

are smooth across

Z

Notice that

| F |^{2 λ} u

and

\bar{\partial} | F |^{2 λ} \land u

are well-defined forms in

X

for

R e λ > > 0

Theorem 2.3. (i) The forms

| F |^{2 λ} u

and

\bar{\partial} | F |^{2 λ} \land u

have analytic continuations as currents in

X

R e λ > - ε

. If

U = | F |^{2 λ} u |_{λ = 0}

and

R = \bar{\partial} | F |^{2 λ} \land u |_{λ = 0}

, then

(δ - \bar{\partial}) U = 1 - R .

(ii) The current

R

has support on

Z

and

R = R_{p} + \dots + R_{μ}

, where

p = c o d i m Z

and

μ = min (n, m - r + 1)

(iii) If

φ

is a holomorphic function and

R φ = 0

, then locally

F Ψ = φ

has holomorphic solutions.

(iv) If

c o d i m Z = m - r + 1

and

F Ψ = φ

has a holomorphic solution, then

R φ = R_{m - r + 1} φ = 0

(v) If

| φ | \leq C | F |^{μ}

, then

R φ = 0

Here, of course,

R_{k} = \bar{\partial} | F |^{2 λ} \land u_{k} |_{λ = 0}

is the component of

R

which is a

(0, k)

-current with values in

Λ^{r + k - 1} E \otimes S^{k - 1} Q^{*} \otimes det Q^{*}

Proof. In the case

r = 1

, this theorem is contained in Theorems 1.1 to 1.4 in [2] , and most parts of the proof are completely analogous.

Therefore we just point out the necessary modifications. By Hironaka's theorem and a further toric resolution, following the technique developed in [5] and [15] , we may assume that locally

F = F_{0} F^{'}

as in Lemma 2.2 .

Since moreover

σ \otimes σ = 0

, we have then that locally in the resolution

u_{k} = \frac{(\bar{\partial} s^{'})^{\otimes (k - 1)} \otimes S^{'}}{F_{0}^{k}} .

It is then easy to see that the proposed analytic extensions exist and we have that

\begin{matrix} U_{k} = [\frac{1}{F_{0}^{k}}] (\bar{\partial} s^{'})^{\otimes (k - 1)} \otimes S^{'}, \end{matrix}

(2.5)

and

\begin{matrix} R_{k} = \bar{\partial} [\frac{1}{F_{0}^{k}}] \land (\bar{\partial} s^{'})^{\otimes (k - 1)} \otimes S^{'}, \end{matrix}

(2.6)

where

[1 / F_{0}^{k}]

is the usual principal value current. If

R φ = 0

, then

(δ - \bar{\partial}) U φ = φ

, and hence by successively solving the

\bar{\partial}

-equations

\bar{\partial} w_{k} = U_{k} φ + δ w_{k + 1},

we finally get the holomorphic solution

Ψ = U_{1} φ + δ w_{2}

. All parts but

(i v)

now follow in a similar way as in [2] . Notice in particular, that

k \leq min (n, m - r + 1)

in 2.6 for degree reasons, so that

R φ = 0

if the hypothesis in (v) is satisfied. As for

(i v)

, let us assume that we have a holomorphic section

Ψ

Λ^{r} E \otimes det Q^{*}

such that

F Ψ = φ

. If

Ψ = ψ \otimes ε_{*}

, then

F Ψ = δ_{f_{r}} \dots δ_{f_{1}} ψ

. Since

u_{m - r + 1}

has full degree in

e_{j}

we have that

u_{m - r + 1} φ = φ σ_{1} \land \dots \land σ_{r} \land (\bar{\partial} σ)^{\otimes (m - r)} \otimes ε^{*} = (δ_{f_{r}} \dots δ_{f_{1}} ψ) σ_{1} \land \dots \land σ_{r} \land (\bar{\partial} σ)^{\otimes (m - r)} \otimes ε^{*} = ψ \land (\bar{\partial} σ)^{\otimes (m - r)} \otimes ε^{*} |_{λ = 0} = (\bar{\partial} σ)^{\otimes (m - r)} \otimes Ψ = \bar{\partial} (σ \otimes (\bar{\partial} σ)^{\otimes (m - r + 1)}) \otimes Ψ = \bar{\partial} u_{m - r}^{'} \otimes Ψ .

Since

c o d i m Z = m - r + 1

we have that

R = R_{m - r + 1}

according to part (ii), so

R φ = R_{m - r + 1} φ = \bar{\partial} | F |^{2 λ} \land u_{m - r + 1} φ |_{λ = 0} = - \bar{\partial} (\bar{\partial} | F |^{2 λ} \land u_{m - r}^{'} \otimes Ψ |_{λ = 0}) .

However,

\bar{\partial} | F |^{2 λ} \land u_{m - r}^{'} \otimes Ψ |_{λ = 0}

vanishes for degree reasons, precisely in the same way as

R_{k}

vanishes for

k \leq m - r

. □

Proof of Theorem 1.4 . If we consider the matrix $f$ as a morphism $E : Q$ , for trivial bundles $E$ and $Q$ , the theorem immediately follows from parts (v) and (iii) of Theorem 2.3 . □

Remark 4. As we have seen, the reason for the power

m - r + 1

in Theorem 1.4 (and in part (v) of Theorem 2.3 ) when

n

is large, is that the complex 2.1 terminates at

k = m - r + 1

. If one tries to analyse the section

F

by means of the usual Koszul complex with respect to the basis

(e_{I})_{| I | = r}^{'}

, then one could hope that for some miraculous reason the corresponding forms

u_{k}

would vanish when

k > m - r + 1

, although one has

m! / (m - r)! r!

dimensions (basis elements). However, this is not the case in general. Take for instance the simplest non-trivial case,

m = 3

and

r = 2

, and choose

f_{1} = (1, 0, ξ_{1})

f_{2} = (0, 1, ξ_{2})

and choose the trivial metric. Then

F_{12} = 1, F_{13} = ξ_{2}, F_{23} = ξ_{1},

and

σ = \bar{F} / | F |^{2}

, so that

σ_{12} = \frac{1}{1 + | ξ_{1} |^{2} + | ξ_{2} |^{2}}, σ_{13} = \frac{{\bar{ξ}}_{2}}{1 + | ξ_{1} |^{2} + | ξ_{2} |^{2}}, σ_{23} = \frac{{\bar{ξ}}_{1}}{1 + | ξ_{1} |^{2} + | ξ_{2} |^{2}} .

Now

m - r + 1 = 2

, but if we form the usual Koszul complex, with say that basis

ε_{1}, ε_{2}, ε_{3}

, so that

σ = σ_{12} ε_{1} + σ_{13} ε_{2} + σ_{23} ε_{3} = \frac{1}{| F |^{2}} (ε_{1} + {\bar{ξ}}_{2} ε_{2} + {\bar{ξ}}_{1} ε_{3}),

we have

σ \land (\bar{\partial} σ)^{2} = \frac{2}{| F |^{6}} d {\bar{ξ}}_{1} \land d {\bar{ξ}}_{2} \land ε_{1} \land ε_{2} \land ε_{3},

and this form is not zero. To get an example where

Z

is non-empty, one can multiply

f

with a function

f_{0}

. □

3 Products of ideals

For

j = 1, . . ., r

, let

E_{j} \to X

be a Hermitian vector bundle of rank

m_{j}

and let

f_{j}

be a section of

E_{j}^{*}

. Moreover, let

E = \oplus_{1}^{r} E_{j}

and let

Q ≃ C^{r}

with

O N

-basis

ε_{1}, \dots, ε_{r}

. If we consider

f_{j}

as sections of

E

, then

f = \sum_{1}^{r} f_{j} \otimes ε_{j}

is a morphism

E \to Q

. Moreover,

F Ψ = φ

with

Ψ = ψ \otimes ε^{*}

as before, means that

δ_{f_{r}} \dots δ_{f_{1}} ψ = φ

, and hence that

φ

belongs to the product ideal

(f_{1}) \dots (f_{r})

. To obtain such a solution

Ψ

we proceed as in the previous section. Notice that now

σ_{j}

can be identified with the section of

E_{j}

with minimal norm such that

f_{j} σ_{j} = 1

Moreover,

| F | = | f_{1} | \dots | f_{r} |

. In this case we therefore have

R_{k} = \bar{\partial} | F |^{2 λ} \land u_{k} = \bar{\partial} (| f_{1} |^{2 λ} \dots | f_{r} |^{2 λ}) \land σ_{1} \land \dots \land σ_{r} \land \sum_{| α | = k - 1} (\bar{\partial} σ_{1})^{α_{1}} \land \dots \land (\bar{\partial} σ_{r})^{α_{r}} \otimes ε_{α}^{*} \otimes ε^{*} |_{λ = 0} .

For degree reasons

R_{k}

will vanish unless

\begin{matrix} 0 \leq α_{j} \leq m_{j} - 1 and α_{1} + \dots + α_{r} \leq n - 1 . \end{matrix}

(3.1)

Proof of Theorem 1.2 . Consider the tuples $f_{j}$ as sections of $E_{j}$ . For each $j$ , let $e_{j i}$ , $i = 1, \dots, m_{j}$ , be a local frame for $E_{j}$ so that $f_{j} = \sum_{i = 1}^{m_{j}} f_{j}^{i} e_{j i}^{*}$ . After a suitable resolution we may assume that for each $j$ , $f_{j} = f_{j}^{0} f_{j}^{'}$ , where $f_{j}^{0}$ is holomorphic, and $f_{j}^{'}$ is a non-vanishing section of $E_{j}^{*}$ . Therefore, $R_{k}$ is a sum of terms like $\bar{\partial} (| f_{1}^{0} |^{2 λ} \dots | f_{r}^{0} |^{2 λ} v^{λ}) \land \frac{β}{(f_{1}^{0})^{α_{1} + 1} \dots (f_{r}^{0})^{α_{r} + 1}} |_{λ = 0},$ where $v$ is smooth and non-vanishing. By the same argument as before this current is annihilated by $φ$ if $| φ | \leq C | f_{1} |^{α_{1} + 1} \dots | f_{r} |^{α_{r} + 1}$ , and in view of 3.1 and the hypothesis in the theorem, taking $s_{j} = α_{j} + 1$ , therefore $φ$ annihilates $R$ . It now follows from Theorem 2.3 (iii) that $F Ψ = φ$ has a holomorphic solution, and thus $ψ \in (f_{1}) \dots (f_{r})$ . □

We can also easily obtain the Briançon-Skoda theorem.

Proof of Theorem 1.1 . Assume that the tuple $f = (f^{1}, \dots, f^{m})$ is given.
Choose disjoint isomorphic bundles $E_{j} ≃ C^{m}$ with isomorphic bases $e_{j i}$ , and let $f_{j} = \sum_{i = 1}^{m} f^{i} e_{j i}^{*}$ . Outside $Z = {f = 0}$ we have $σ_{j} = \sum_{1}^{m} σ^{i} e_{j i}$ .
Now $\bar{\partial} σ^{i}$ are linearly dependent, since $\sum_{1}^{m} f^{i} \bar{\partial} σ^{i} = \bar{\partial} \sum_{1}^{m} f^{i} σ^{i} = \bar{\partial} 1 = 0$ . Thus the form $u_{k}$ must vanish if $k - 1 > m - 1$ , and therefore $R_{k}$ vanishes unless $k \leq min (n, m)$ . Since $| f_{j} | = | f |$ , locally in the resolution, we have $R_{k} = \bar{\partial} | f |^{2 r λ} \land \frac{β}{(f^{0})^{k + r - 1}} |_{λ = 0},$ and hence it is annihilated by $φ$ if $| φ | \leq C | f |^{min (m, n) + r - 1}$ . □

It remains to consider the case when the

f_{j}

together define a complete intersection. The proof is very much inspired by similar proofs in [20] .

Proof of Theorem 1.4 . We now assume that $c o d i m {f_{1} = \dots = f_{r} = 0} = m_{1} + \dots + m_{r}$ . In particular, $m_{1} + \dots + m_{r} \leq n$ . Let $ξ$ be a test form times $φ$ . If the support is small enough, after a resolution of singularities and further localization, $R . ξ$ becomes a sum of terms, the worst of which are like $\int \bar{\partial} (| f_{1}^{0} |^{2 λ} \dots | f_{r}^{0} |^{2 λ}) \land \frac{s_{1}^{'} \land \dots \land s_{r}^{'} \land (\bar{\partial} s_{1}^{'})^{m_{1} - 1} \land \dots \land (\bar{\partial} s_{r}^{'})^{m_{r} - 1} \land \tilde{ξ} ρ}{(f_{1}^{0})^{m_{1}} \dots (f_{r}^{0})^{m_{r}}} |_{λ = 0},$ where $\tilde{ξ}$ is the pull-back of $ξ$ and $ρ$ is a cut-off function in the resolution.
We may assume that each $f_{j}^{0}$ is a monomial times a non-vanishing factor in a local coordinate system $τ_{k}$ . Let $τ$ be one of the coordinate factors in, say, $f_{1}$ (with order $ℓ$ ), and consider the integral that appears when $\bar{\partial}$ falls on $| τ^{ℓ} |^{2 λ}$ . If $τ$ does not occur in any other $f_{j}^{0}$ , then the assumption $| φ | \leq C | f_{1} |^{m_{1}}$ implies that $\tilde{φ}$ is divisible by $τ^{ℓ m_{1}}$ . Hence $\tilde{φ}$ and therefore also $\tilde{ξ}$ annihilates the singularity as before, so that the integral vanishes. We now claim that if, on the other hand, $τ$ occurs in some of the other factors, then the integral vanishes because of the complete intersection assumption. Thus let us assume that $τ$ occurs in $f_{2}^{0}, \dots, f_{k}^{0}$ but not in $f_{k + 1}^{0}, \dots, f_{r}^{0}$ . The forms $s_{j} = | f_{j} |^{2} σ_{j}$ are smooth and, moreover, $\tilde{γ} = \frac{s_{k + 1}^{'} \land \dots \land s_{r}^{'} \land (\bar{\partial} s_{k + 1}^{'})^{m_{k + 1} - 1} \land \dots \land (\bar{\partial} s_{r}^{'})^{m_{r} - 1} \land \tilde{ξ}}{(f_{k + 1}^{0})^{m_{k + 1}} \dots (f_{r}^{0})^{m_{r}}}$ is the pull-back of $γ = \frac{s_{k + 1} \land \dots \land s_{r} \land (\bar{\partial} s_{k + 1})^{m_{k + 1} - 1} \land \dots \land (\bar{\partial} s_{r})^{m_{r} - 1} \land ξ}{| f_{k + 1} |^{2 m_{k + 1}} \dots | f_{r}^{0} |^{2 m_{r}}} .$ Since the form $γ$ has codegree $1 + (m_{1} - 1) + \dots + (m_{k} - 1)$ in $d \bar{z}$ , which is strictly less than $m_{1} + \dots + m_{k} = c o d i m {f_{1} = \dots = f_{k} = 0}$ , the anti-holomorphic factor of the denominator vanishes on ${f_{1} = \dots = f_{k} = 0}$ . Therefore, each term of its pull-back vanishes where $τ = 0$ , so it must contain either a factor $\bar{τ}$ or $d \bar{τ}$ . However, because of the assumption, the (pull-back) of the denominator contains no factor $\bar{τ}$ , so each term of $\tilde{γ}$ will contain $\bar{τ}$ or $d \bar{τ}$ . Therefore, the integral that appears when $\bar{\partial}$ falls on $| τ |^{2 λ ℓ}$ will vanish when $λ = 0$ . □

4 Explicit integral representation

We are now going to supply explicit proofs of Theorems 1.1 to 1.4 .

Since all of them are local, we assume that the functions

f

and

φ

are defined in a convex neighborhood

X

of the closure of the unit ball

B

C^{n}

. We first recall the construction of weighted representation formulas for holomorphic functions from [1] . For fixed

z \in X

, let

δ_{η}

denote interior multiplication with the vector field

2 π i \sum (ζ_{j} - z_{j}) \frac{\partial}{\partial ζ_{j}},

and let

\nabla_{η} = δ_{η} - \bar{\partial}

. Then we have (lower indices denote bidegree), see [1] ,

Proposition 4.1. Assume that

z

is a fixed point in

X

and

g = g_{0, 0} + \dots + g_{n, n}

is a smooth form in

X

with compact support such that

\nabla_{η} g = 0

and

g_{0, 0} (z) = 1

. Then

\begin{matrix} φ (z) = \int g φ = \int g_{n, n} φ \end{matrix}

(4.1)

for each holomorphic function

φ

X

For further reference we also notice, see [4] , that:

(i) if

g^{1}

and

g^{2}

satisfy the assumptions in the proposition (it is enough that one of them has compact support), then also

g = g^{1} \land g^{2}

does.

(ii) it is enough that

g

us smooth in a neighborhood of the point

z

Example 2. Let

χ

be a cutoff function in

X

that is identically

1

in a neighborhood of

\bar{B}

. Moreover, let

s (ζ, z) = \frac{1}{2 π i} \frac{\partial | ζ |^{2}}{| ζ |^{2} - \bar{ζ} \cdot z} .

Then for each

z \in B

g = χ - \bar{\partial} χ \land \frac{s}{\nabla_{η} s} = χ - \bar{\partial} χ \land [s + s \land \bar{\partial} s + s \land (\bar{\partial} s)^{2} + \dots + s \land (\bar{\partial} s)^{n - 1}] = χ - \bar{\partial} χ \land \sum_{k = 1}^{n} \frac{1}{(2 π i)^{k}} \frac{\partial | ζ |^{2} \land (\bar{\partial} \partial | ζ |^{2})^{k - 1}}{(| ζ |^{2} - \bar{ζ} \cdot z)^{k}}

a compactly supported form such that

\nabla_{η}

-closed and

g_{0, 0} (z) = 1

Moreover,

g

depends holomorphically on

z

. □

Example 3. Another possible choice is

\begin{matrix} g = {(1 + \nabla_{ζ - z} \frac{\bar{ζ} \cdot d ζ}{2 π i (1 - | ζ |^{2})})}^{- ν} = {(\frac{1 - \bar{ζ} \cdot z}{1 - | ζ |^{2}} - ω)}^{- ν} \end{matrix}

(4.2)

for positive

ν

, where

ω = (i / 2 π) \partial \bar{\partial} log (1 / (1 - | ζ |^{2}))

. It is

O ((1 - | ζ |^{2})^{ν})

near the boundary and therefore at least of class

C^{ν - 1}

. □

Now assume that

f

is a holomorphic

r \times m

-matrix, that we consider as a holomorphic morphism

f : E \to Q

with respect to some fixed ON-bases for the trivial bundles

E ≃ C^{m}

and

Q ≃ C^{r}

. We will construct the decomposition 1.1 from the currents

U

and

R

in Section 2 , following an idea in [4] . First we choose holomorphic

(1, 0)

-forms

h_{j}

X

, Hefer forms, such that

δ_{η} h_{j} = f_{j} (ζ) - f_{j} (z),

and let

h = \sum_{1}^{m} h_{j} \otimes ε_{j}^{*}

. We may also assume that

h_{j}

, and hence

h

, depend holomorphically on the parameter

z

. Now

δ_{h} : E_{k + 1} \to E_{k}

, for

k \geq 1

, and hence

(δ_{h})_{k} : E_{k + 1} \to E_{1}, k \geq 0,

(δ_{h})_{ℓ} = δ_{h}^{ℓ} / ℓ!

. It is easily seen that

\begin{matrix} δ_{η} (δ_{h})_{k} = (δ_{h})_{k - 1} δ_{f} - δ_{f (z)} (δ_{h})_{k - 1} . \end{matrix}

(4.3)

So far

δ_{F}

has only acted on

(0, 0)

-forms with values in

Λ^{r} E

. We now extend it to general

(p, q)

-forms, with the convention that one insert a minus sign when

p + q

is odd. Thus we let

δ_{F} α = (- 1)^{(r + 1) (deg α + 1)} δ_{f_{r}} \dots δ_{f_{1}} \otimes δ_{ε_{1}} \dots δ_{ε_{r}},

where

deg α

is the degree of

α

Λ (E \otimes T^{*} (X))

. With this convention

δ_{F}

, as well as

δ_{f}

, will anti-commute with

\bar{\partial}

and

δ_{η}

It is possible to find

(1, 0)

-form-valued mappings

H_{k}^{0} : E_{k} \to C

, such that

\begin{matrix} δ_{η} H_{1}^{0} = δ_{F} (ζ) - δ_{F (z)}, δ_{η} H_{k}^{0} = H_{k - 1}^{0} δ_{f (ζ)} - δ_{F (z)} (δ_{h})_{k - 1}, k \geq 2 . \end{matrix}

(4.4)

The form

H_{1}^{0}

is a usual Hefer form. The right hand side of the second equation for

k = 2

is now holomorphic and

δ_{η}

-closed, and it is well-known then that there exists a holomorphic solution

H_{2}^{0}

. We may as well assume that it depends holomorphically on the parameter

z

X

. The existence of

H_{k}^{0}

in general follows by induction. For explicit choices of solutions in

X

, see [4] . We now define

H^{1} U = \sum_{k = 1}^{min (n + 1, m - r + 1)} (δ_{h})_{k - 1} U_{k},

and

H^{0} R = \sum_{k = 1}^{min (n, m - r + 1)} H_{k}^{0} R_{k} .

Theorem 4.2. If

φ

is holomorphic in

X

and

g

is the smooth form in Example 2 , then we have the holomorphic decomposition

\begin{matrix} φ (z) = δ_{F (z)} \int H^{1} U \land g φ + \int H^{0} R \land g φ, z \in B . \end{matrix}

(4.5)

Proof. First we assume that $z \in B \ Z$ , and consider the form $g^{'} = δ_{F (z)} H^{1} U + H^{0} R .$ We have that $g_{0, 0} = δ_{F} U_{1}$ , and hence $g_{0, 0} (z) = 1$ . Moreover, using 4.3 and 4.4 it is readily verified that $\nabla_{η} g^{'} = 0$ . Since $g^{'} \land g$ is smooth in a neighborhood of $z$ it follows form Proposition 4.1 and the subsequent remarks that 4.5 holds for this $z$ . However, since both sides of 4.5 are holomorphic in $B$ the theorem is proved. □

In particular,

Ψ (z) = \int H^{1} U \land g φ

is an explicit solution to

δ_{F (z)} Ψ = φ

R φ = 0

. We now consider this solution in more detail. In view of 2.2 and 2.3 we have, outside

Z

, that

(δ_{h})_{k - 1} u_{k} = \sum_{| α | = k - 1} (δ_{h_{1}})_{α_{1}} \dots (δ_{h_{r}})_{α_{r}} [σ_{1} \land \dots \land σ_{r} \land (\bar{\partial} σ_{1})^{α_{1}} \land \dots \land (\bar{\partial} σ_{r})^{α_{r}}] \otimes ε^{*} .

Moreover, since we have the trivial metric,

σ_{j} = \sum_{i = 1}^{m} σ_{i j} e_{j}, j = 1, \dots, r,

are just the columns in the matrix

f^{*} (f f^{*})^{- 1}

. Suppressing the non-vanishing section

ε^{*}

, we have

Corollary 4.3. Let

f

be a generically surjective holomorphic

r \times m

-matrix in

X

with rows

f_{j}

, considered as sections of the trivial bundle

E^{*}

, and assume that the hypothesis of Theorem 1.4 is fulfilled. Then

ψ (z) = \int H^{1} U \land g φ

is an explicit solution to

δ_{F (z)} ψ (z) = δ_{f_{1} (z)} \dots δ_{f_{r} (z)} ψ (z) = φ (z)

B

, where

H^{1} U φ

is the value at

λ = 0

of (the analytic continuation of )

(4.6) | f | 2 λ ∑ k = 1 min ( n + 1 , m − r + 1 ) ∑ | α | = k − 1 ( δ h 1 ) α 1 ⋯ ( δ h r ) α r [ σ 1 ∧ … ∧ σ r ∧ ( ∂ ¯ σ 1 ) α 1 ∧ … ∧ ( ∂ ¯ σ r ) α r ] φ

and

g

is the form in Example 2 . If

m - r + 1 \leq n

, then

H^{1} U φ

is locally integrable, and the value at

λ = 0

exists in the ordinary sense.

Proof. It remains to verify the claim about the local integrability. In fact, after a resolution of singularities, cf., 2.5 , it follows that $U_{k} φ$ is locally integrable if $| φ | ≲ | F |^{k}$ . If $m - r + 1 \leq n$ , then the sum terminates at $k = m - r + 1$ , and therefore the current is locally integrable; otherwise the worst term is like $U_{n + 1} φ$ , and it will not we locally integrable in general. □

If all the

f_{j}

takes values in different bundles

E_{j}^{*}

and

E = \oplus E_{j}

, then we can simplify the expression for

H^{1} U

further. In this case, cf., Section 3 ,

σ_{j} = \sum_{i = 1}^{m_{j}} \frac{{\bar{f}}_{j}^{i}}{| f_{j} |^{2}} e_{i j}, j = 1, \dots, r .

Moreover, with natural choices of Hefer forms

h_{j}

δ_{h_{j}}

will vanish on forms with values in

E_{k}

for

k \neq j

, and hence we get

Corollary 4.4. Let

f_{j}

m_{j}

-tuples of functions, considered as sections of the trivial bundles

E_{j}^{*}

over

X

. If the conditions of Theorem 1.2 or 1.3 are fulfilled, or if all

f_{j}

are equal to some fixed

m

-tuple

f

, and the condition in Theorem 1.1 is fulfilled, then

ψ (z) = \int H^{1} U φ \land g

is an explicit solution to

δ_{f_{1} (z)} \dots δ_{f_{r} (z)} ψ (z) = φ (z)

B

, where

H^{1} U φ

is the value at

λ = 0

of (the analytic continuation of )

(4.7) | f | 2 λ ∑ k = 1 n + 1 ∑ | α | = k − 1 ( δ h 1 ) α 1 [ σ 1 ∧ ( ∂ ¯ σ 1 ) α 1 ] ∧ … ∧ ( δ h r ) α r [ σ r ∧ ( ∂ ¯ σ r ) α r ] φ .

N = min (n + 1, m - r + 1)

, and

g

is the form in Example 2 .

In the case of Theorems 1.2 and 1.3 , only terms such that

α_{j} \leq m_{j}

actually occur. In the case of Theorem 1.1 we have only terms such that

k \leq m

We conclude this paper with some brief comments on Berndtsson's classical division formula from [7] . As mentioned in the introduction, the first known explicit formula for the Briançon-Skoda theorem (

r = 1

) was in Theorem 9.5 in [4] ; in fact, it is identical to the formula above in the case

r = 1

, and it is different from Berndtsson's formula.

Surprisingly enough it was recently discovered, [12] , that the general case of the Briançon-Skoda theorem, i.e., Theorem 1.1 , actually can be obtained from Berndtsson's classical formula, and we will sketch the proof below; for more details, see [12] . However, we see no way of proving any of the variations discussed in this paper by Berndtsson type formulas.

As before we consider the given

m

-tuple

f

as a section of the trivial bundle

E^{*}

over

X

. Let

s

be the section of

E

with minimal norm such that

f s = | f |

. Then

s = | f |^{2} σ

in the previous notation. If the metric is trivial and

f = \sum f_{j} e_{j}^{*}

, then

s = \sum {\bar{f}}_{j} e_{j}

. For

ε > 0

, let

σ^{ε} = \frac{s}{| f |^{2} + ε},

let

h = h_{j} e_{j}^{*}

be a Hefer form as before, and let

g^{'} = 1 - \nabla_{η} h \cdot σ^{ε} = \frac{ε}{| f |^{2} + ε} + f (z) \cdot σ^{ε} + h \cdot \bar{\partial} σ^{ε} .

By Proposition 4.1 we have the representation formula

\begin{matrix} φ (z) = \int {(\frac{ε}{| f |^{2} + ε} + f (z) \cdot σ^{ε} + h \cdot \bar{\partial} σ^{ε})}^{min (n + 1, m) + r - 1} φ \land g . \end{matrix}

(4.8)

At least if the form

g

from Example 3 is used, the resulting formula is precisely of the type in [7] though derived in a somewhat different way.

Proposition 4.5. Assume that

f, φ

are holomorphic in

X

and that

\begin{matrix} | φ | ≲ | f |^{min (n, m) + r - 1}, \end{matrix}

(4.9)

holds. When

ε \to 0

the formula 4.8 converges to an explicit representation of

φ

B

as an element in the ideal

(f)^{r}

Sketch of proof. To begin with we assume that

m > n

. Then the power of

g^{'}

n + r

, and expanding we get that

(g^{'})^{n + r} φ = \sum_{ℓ = 1}^{r} c_{ℓ} {(\frac{ε}{| f |^{2} + ε} + h \cdot \bar{\partial} σ^{ε})}^{n + ℓ} (f (z) \cdot σ^{ε})^{r - ℓ} φ + \dots,

where

\dots

denote terms in

(f)^{r}

. Taking for granted that these latter terms actually converge to currents with values in

(f)^{r}

when

ε \to 0

, we have to prove that the first terms tend to zero. When expanding further, for degree reasons, the worst term that appears is

\frac{ε}{| f |^{2} + ε} (h \cdot \bar{\partial} σ^{ε})^{n} (f (z) \cdot σ^{ε})^{r - 1} φ .

Using the technique in Section 2 we may assume that

f = f_{0} f^{'}

where

f^{'} \neq 0

, and then this term is dominated by

\begin{matrix} ε \frac{| d f_{0} | | f_{0} |}{(| f_{0} |^{2} | f^{'} |^{2} + ε)^{2}} . \end{matrix}

(4.10)

Assuming furthermore, as we may, that

f_{0}

is a monomial, it is readily checked that the expression 4.10 tends to

0

L_{loc}^{1}

when

ε \to 0

. We then consider the case when

m \leq n

. When expanding

(g^{'})^{m + r - 1} φ

, besides terms in

(f)^{r}

, the worst term that appears is

\begin{matrix} (h \cdot \bar{\partial} σ^{ε})^{m} (f (z) \cdot σ^{ε})^{r - 1} φ . \end{matrix}

(4.11)

Using that

{(h \cdot \bar{\partial} \frac{s}{| f |^{2} + ε})}^{m} = h_{1} \land \dots \land h_{m} \land ε \frac{\bar{\partial} s_{m} \land \dots \land \bar{\partial} s_{1}}{(| f |^{2} + ε)^{m + 1}},

it follows that also 4.11 is dominated by 4.10 , and so the proposition is proved. □

References

M. Andersson: Integral representation with weights I, Math. Ann. 326 (2003), 1–18.
M. Andersson: Residue currents and ideals of holomorphic functions, Bull. Sci. Math., 128, (2004), 481–512.
M. Andersson: Residue currents of holomorphic morphisms, Preprint Gothenburg (2004).
M. Andersson: Integral representation with weights II, In preparation.
C. Berenstein & A. Yger: Green currents and analytic continuation, J. Anal. Math. 75 (1998), 1–50.
C. Berenstein & R. Gay & A. Vidras & A. Yger: Residue Currents and Bézout Identities, Birkhäuser (1993).
B. Berndtsson: A formula for division and interpolation, Math. Ann. 263 (1983), 113-160.
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.
J-P Demailly: Estimations $L^{2}$ pour l'opérateur $\bar{\partial}$ d'un fibré vectoriel holomorphe semi-positif au-dessus d'une variété kählérienne complète, Ann. Sci. École Norm. Sup. (4) 15 (1982), 457–511.
J-P Demailly: Complex Analytic and Differential Geometry, Monograph Grenoble (1997).
A. Dickenstein & C. Sessa: Canonical representatives in moderate cohomology, Invent. Math. 80 (1985), 417–434..
E. Gotmark: , Licentiate thesis (2005), in preparation.
J. Lipman & B. Teissier: Pseudorational local rings and a theorem of Briançon-Skoda about integral closures of ideals, Michigan Math. J. 28 (1981), 97–116.
M. Passare: Residues, currents, and their relation to ideals of holomorphic functions, Math. Scand. 62 (1988), 75–152.
M. Passare & A. Tsikh & A. Yger: Residue currents of the Bochner-Martinelli type, Publ. Mat. 44 (2000), 85-117.
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.
H. Skoda: Morphismes surjectifs de fibrés vectoriels semi-positifs, Ann. Sci. École Norm. Sup. 11 (1978), 577–611.
B. Teissier: Résultats récents d'algèbre commutative effective, Séminaire Bourbaki, Vol. 1989/90. Astérisque No. 189-190 (1990).
A. Tsikh & A. Yger: Residue currents. Complex analysis, J. Math. Sci. (N. Y.) 120 (2004), 1916–1971.
E. Wulcan: Products of residue currents of Cauchy-Fantappiè-Leray type, Preprint Gothenburg (2004).

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

November 27, 2006

Mats Andersson