Journal of Algebra 323 (2010) 2209–2225

Contents lists available at ScienceDirect

Journal of Algebra
www.elsevier.com/locate/jalgebra

Reductions and special parts of closures

Neil Epstein
Universität Osnabrück, Institut für Mathematik, 49069 Osnabrück, Germany

a r t i c l e i n f o a b s t r a c t

Article history: We provide an axiomatic framework for working with a wide vari-
Received 28 May 2009 ety of closure operations on ideals and submodules in commutative
Available online 24 February 2010 algebra, including notions of reduction, independence, spread, and
Communicated by Steven Dale Cutkosky
special parts of closures. This framework is applied to tight, Frobe-
Keywords:
nius, and integral closures. Applications are given to evolutions and
Tight closure special Briançon–Skoda theorems.
Integral closure © 2010 Elsevier Inc. All rights reserved.
Evolutions
Briançon–Skoda theorems

1. Introduction

One of the most useful notions in commutative algebra is that of the closure operation on ideals,
or more generally on submodules. Much of the time, authors concentrate on the properties of one
particular closure operation, so the general notion itself is not always given a proper definition.1 The
following encapsulates what most authors mean:

Definition 1.1. Let R be a ring, and let M be a set of R-modules (often either just R, or all (finitely
generated) R-modules). A closure operation c sends any submodule L of a module M ∈ M to another
submodule L cM of M, subject to the following axioms:

(1) For any submodule L of any M ∈ M, L ⊆ L cM = ( L cM )cM .

(2) If K ⊆ L are submodules of some M ∈ M, then K M c
⊆ L cM .

(3) Let g : M → M be a homomorphism of R-modules in M. Then for any submodule L ⊆ M,
g ( L cM ) ⊆ g ( L )cM  .

E-mail address: [email protected].

1
One counterexample is the recent paper [Vas09], which gives structure to sets of closure operations satisfying certain prop-
erties.

0021-8693/$ – see front matter © 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2010.02.015
2210 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

Familiar examples of closure operations on ideals include tight closure, integral closure, and the
radical. The definition above is very broad, so it is useful to identify additional properties that may
not hold for all closure operations. For instance, in [Eps05], we introduced the following notions for
closure operations on ideals, here generalized to the module case:

Definition 1.2. A closure operation c on a class M of R-modules is Nakayama if ( R , m) is local and

for every M ∈ M and submodules K ⊆ L ⊆ M such that K ⊆ L ⊆ ( K + m L )cM , we have K M c
= L cM .
Given M ∈ M and elements z1 , . . . , zt ∈ M, they are c-independent (relative to M) if for all 1 
i  t, we have

 c
zi ∈
/ Rz j .
j =i M

A submodule L ⊆ M is c-independent if it has a c-independent generating set, and it is strongly c-

independent if every minimal generating set for L is c-independent.
Given a submodule L ⊆ M ∈ M, a c-reduction K ⊆ L of L in M is a submodule such that K Mc
= L cM . It
is minimal if there is no proper submodule P  K which is a c-reduction of L. If minimal c-reductions
exist, and if every minimal c-reduction of L in M has the same minimal number of generators, we
call this common number the c-spread of L in M, denoted cM ( L ).

As a closure on ideals in a local ring, it is clear that integral closure is Nakayama, and we showed
in [Eps05] that tight closure in this context is as well. Radical, however, is not Nakayama. Moreover,
we showed that for any Nakayama closure c, minimal c-reductions exist, and they are exactly the
strongly c-independent c-reductions. Although this result was stated for ideals, the exact same proof
shows it to be true in this wider context.
Also in that paper, we used Vraciu’s notion of special tight closure to prove that under mild condi-
tions on the ring, minimal generating sets of minimal ∗-reductions of an ideal all have the same size
generating sets.
Accordingly, in this note we generalize and axiomatize the notion of the special part of a closure, and
we use it to obtain interesting results for Frobenius, integral, and tight closures. Except where otherwise
noted, in this paper R will always denote a Noetherian local ring with maximal ideal m and residue field k.
Most of this work was completed years ago as part of my dissertation. I did not submit the paper
for publication at the time. However, as there are now several papers which use the ideas in the
paper (e.g. [Vra06,EH] and [FV]), I have been convinced to publish it.

2. Axioms for special parts of closures

Definition 2.1. Let c be a closure operation on submodules of an R-module M. Then csp is a special
part of c for M if the following four axioms hold whenever L and N are submodules of M.

csp
(1) L M is a submodule of M.
csp
(2) mL ⊆ L M ⊆ Lc .
csp csp csp
(3) ( L cM )M = L M = ( L M )cM .
csp c
(4) If L ⊆ N ⊆ ( L + N M ) M , then N ⊆ L cM .

If M = R, we say csp is a special part of c for ideals. If the closure operation c is only defined for
ideals, we simply say csp is a special part of c. If c and csp are defined at least on all submodules of
finitely generated R-modules, we say that csp is a special part of c.
If the ambient module is understood, sometimes we write L c in place of L cM . In particular if M = R
(so L is now an ideal), we almost always leave off the ambient module R in the notation.
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2211

Note the following consequences of the definition:

Lemma 2.2. Let c be a closure operation on submodules of M and csp a special part of c. Then

• c is a Nakayama closure.
csp csp
• If L cM ⊆ N cM , then L M ⊆ N M .
csp
• For any c-independent submodule L, m L = L ∩ L M .

Proof. c is Nakayama because of axioms (4) and (2) of Definition 2.1. For this reason, we call ax-
iom (4) the Nakayama property.
csp csp csp csp
If L cM ⊆ N cM , then by axiom (3) of Definition 2.1, L M = ( L c ) M ⊆ ( N c ) M = N M .
csp
Finally, suppose that L is a c-independent submodule of M, and let z ∈ L ∩ L M . Let z1 , . . . , zn be a
n
c-independent generating set of L. Then there are elements r j ∈ R such that z = j =1 r j z j . If z ∈
/ m L,
then there is some j with r j ∈ / m. Without loss of generality, j = 1, and by dividing by r1 , we may
assume that r1 = 1. That is,


n
csp
z = z1 + r j z j ∈ LM .
j =2

csp
Let N = ( z2 , . . . , zn ). Then by the above equation, we have z1 ∈ N + L M , which implies that L ⊆
csp
N + L M . Then by axiom (4), L ⊆ N cM , so that z1 ∈ N cM . But this contradicts the c-independence of
z1 , . . . , zn .
Thus, z ∈ m L. 2

3. The special part of tight closure

The ideal case of the special part of tight closure was introduced by Vraciu in [Vra02]. Further
work appears in [Eps05] and [Vra06]. Here’s the submodule version:

Definition 3.1. For finitely generated R-modules N ⊆ M, we define the special part of the tight closure
of N in M to be the set

∗sp    [q] ∗

N M = z ∈ M  ∃q such that zq ∈ m N M F e ( M ) .

Most of the proof that ∗sp is a special part of ∗ for ideals is in [Eps05], and the proofs for the
submodule case are identical.
Note also that the special part of tight closure can be computed modulo minimal primes.
In [Vra06], Vraciu introduces the notions of ∗-independence modulo an ideal, ∗-spread modulo an
ideal, and (minimal) ∗-reductions of an ideal modulo another ideal. Note that ∗-independence modulo
J is just ∗-independence in the R-module R / J , (minimal) ∗-reductions of I modulo J correspond
exactly with (minimal) ∗-reductions of I / J in the module R / J , and ∗J ( I ) = ∗R / J ( I / J ) whenever such
a number is defined. She observes that the proof of [Eps05, Theorem 5.1] can be “modified slightly”
to show that ∗J ( I ) exists in her context of a normal local domain. Essentially the same modification
shows that whenever R is excellent and analytically irreducible and L ⊆ M is any inclusion of finitely
generated modules, then ∗M ( L ) exists.

4. Analytic F -independence, and the special part of Frobenius closure

In this section, we assume only that R is a Noetherian local ring of prime characteristic p > 0.
2212 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

Definition 4.1. Let N ⊆ M be finitely generated R-modules. The special part of the Frobenius closure of N
in M is the submodule

  [q]

N M := z ∈ M  ∃q = p e such that zq ∈ m N M .
F sp

[q]
It is equivalent to say that there is some q such that zq ∈ (m N M ) M
F
.

Proposition 4.2. F sp is a special part of the Frobenius closure, in the sense of Definition 2.1, for all finitely
generated R-modules.

[q]
Proof. For property (1) of the definition, if y , z ∈ N M , there is some q with y q ∈ m N M and some q
F sp

 [q ]
with zq ∈ m N M , and without loss of generality q  q . Then


 q/q  [q ] [q/q ]  [q] [q]
zq = zq ∈ m N M F e ( M ) = m[q/q ] N M ⊆ m N M ,

[q] F sp [q]
and thus ( y − z)q = y q − zq ∈ m N M , so y − z ∈ N M . Moreover, for any r ∈ R, (r y )q = r q y q ∈ m N M , so
F sp
that r y ∈ NM .
F sp F sp [q]
For property (2), m N ⊆ N M by taking q = 1 in the definition, and N M ⊆ N M
F
because m N M ⊆
[q]
NM .
F sp
For property (3), suppose L ⊆ N ⊆ M are finitely generated submodules and N ⊆ ( L + N M ) M
F
, then
since N is finitely generated, there is some q such that

[q] [q]  F sp [q]

NM ⊆ LM + NM M
. (1)

F sp [q ] [q ]
Since N M is finitely generated, there is some q such that ( N M ) M ⊆ m N M . Replacing the q in (1)
F sp

by max{q, q }, that containment yields

[q] [q] [q]

N M ⊆ L M + mN M .

[q] [q]
Then by the standard Nakayama lemma, N M ⊆ L M , which proves that N ⊆ L M F
, and hence prop-
erty (3).
F F sp
Finally, for property (4), first suppose z ∈ ( N M )M . Then there is some q such that both zq ∈
F [q] [q] [q] [q] F sp
m( N M ) and ( N M F
)M = N M , which combine to make zq ∈ m N M , and hence that z ∈ N M . Similarly,
F sp F F sp [q] F sp [q] [q]
if z ∈ ( N M ) M , there is some q such that both zq ∈ ( N M ) M and ( N M ) M ⊆ m N M , which combine to
[q] F sp
show that z ∈ m N M , and hence that z ∈ N M . 2
q

The question immediately arises in which situations we have a special Frobenius closure decom-
position:

Proposition 4.3. Let ( R , m) be a Noetherian local ring of characteristic p > 0 with perfect residue field. Then
F sp
for any finitely generated R-module M and any submodule N ⊆ M, N M F
= N + NM .

Proof. The containment ‘⊇’ is obvious. So suppose that z ∈ N M

F
. Then there is some q such that
[q]
zq ∈ N M .
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2213

Let { z1 , . . . , zn } be any generating set of N. Then we have


n
q
zq = ai zi ,
i =1

where ai ∈ R. Let r be the number of ai ’s that are not in m. We can rearrange the zi ’s in such a way
that ai ∈/ m if 1  i  r and ai ∈ m if r < i  n. Since ai ∈
/ m for 1  i  r and R /m is perfect, there
q
exist u i ∈ R \ m and mi ∈ m such that ai = u i + mi whenever i  r. Hence,

 q

r 
r
q

n
q [q]
z− u i zi = mi zi + ai zi ∈ m N M .
i =1 i =1 i =r +1

r F sp F sp
That is, z − i =1 u i z i ∈ N M , so that z ∈ N + N M . 2

By analyzing the proof of [Eps05, Theorem 5.1], if ( R , m) is a local ring, c is any closure operation
on (submodules of) a class of modules M with a special part csp, and if for all R-modules M ∈ M
and submodules L ⊆ M one has L c = L + L csp , then submodules of M ∈ M have spread, in the sense
that every minimal c-reduction has the same minimal number of generators as every other. Hence, if k
is a perfect field, then F -spread is well defined for submodules of any finitely generated R-module.
The assumption on the field can be dropped too.
However, we present below a different way to prove that F -spread is well defined, using notions
analogous to the original definitions of analytic spread and analytic independence from Northcott and
Rees [NR54] for Frobenius closure, inspired also in part by Adela Vraciu’s work on ∗-independence
in [Vra02]:

Definition 4.4. Fix a finitely-generated R-module M, as before. Let z1 , . . . , zn ∈ N, where N is a

submodule of M. Then we say that z1 , . . . , zn are analytically F -independent in N [resp. analytically
F -independent] if for any power q of p and any polynomial φ of the form

q q
φ( X 1 , . . . , Xn ) = c 1 X 1 + · · · + cn Xn ,

where q is a power of p, the X i are indeterminates, and the c i are elements of R, such that
[q] [q]
φ(z1 , . . . , zn ) ∈ m N M [resp. such that φ(z1 , . . . , zn ) = 0 ∈ N M ], it follows that the coefficients
c 1 , . . . , cn of φ are all in m.

Lemma 4.5. For elements z1 , . . . , zn ∈ M, with L = ( z1 , . . . , zn ), the following are equivalent:

(1) z1 , . . . , zn are analytically F -independent.

(2) z1 , . . . , zn are analytically F -independent in L.
(3) z1 , . . . , zn are analytically F -independent in any submodule N of M such that L is an F -reduction of N.
(4) z1 , . . . , zn are F -independent.
q q [q]
(5) For any power q of p, z1 , . . . , zn form a minimal set of generators for L M .

q q [q]
Proof. (1) ⇒ (2): Let φ = c 1 X 1 + · · · + cn X n such that φ( z1 , . . . , zn ) ∈ m( z1 , . . . , zn ) M . Then there exist
m1 , . . . , mn ∈ m such that

q q q q
c 1 z1 + · · · + cn zn = m1 z1 + · · · + mn zn .

q q
Let ψ = (c 1 − m1 ) X 1 + · · · + (cn − mn ) X n . Then ψ( z1 , . . . , zn ) = 0, so that by analytic F -independence,
c i − mi ∈ m for all i. Thus, since mi ∈ m for all i, it follows that c i ∈ m for all i.
2214 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

[q ] [q ]
(2) ⇒ (3): Let q be a power of p such that N M = (z1 , . . . , zn )M , and let φ = c 1 X 1 + · · · + cn Xn
q q

[q] [q] [q]

be such that φ( z1 , . . . , zn ) ∈ m N M . If q  q , then N M = ( z1 , . . . , zn ) M , so that by analytic F -
independence in ( z1 , . . . , zn ), all the c i are in m. On the other hand, if q < q , then we have

q /q  [q ]
φ( z1 , . . . , zn ) M ∈ m[q /q] N M
 [q ]
= m[q /q] ( z1 , . . . , zn ) M
[q ]
⊆ m( z1 , . . . , zn ) M .

q /q
Thus by analytic F -independence in ( z1 , . . . , zn ), we have that c i ∈ m for each i. But m is radical,
so c i ∈ m for all i.
[q]
(3) ⇒ (1): Obvious, since 0 ∈ m N M for all N and all q.
(1) ⇒ (4): If z1 , . . . , zn are not F -independent, then without loss of generality z1 ∈ (z2 , . . . , zn )M
F
.
q [q]
Then there is some q with z1 ∈ ( z2 , . . . , zn ) M . Thus, there are choices c 2 , . . . , cn ∈ R such that

q q q
z1 + c 2 z2 + · · · + cn zn .

q q q
If we set φ = X 1 + c 2 X 2 + · · · + cn X n , then φ( z1 , . . . , zn ) = 0 but not all of the coefficients of φ are in m
q
(since the coefficient for X 1 is 1), which shows that z1 , . . . , zn are not analytically F -independent.
(4) ⇒ (1): The proof that (1) implies (4) can pretty much be reversed: If z1 , . . . , zn are not analyti-
q q
cally F -independent, then there is some polynomial φ = c 1 X 1 + · · · + cn X n such that φ( z1 , . . . , zn ) = 0
and at least one of the c i is a unit. We may assume that i = 1. Then

 q
z1 = −c 1−1 c 2 z2 + · · · + cn zn ,
q q

so that z1 ∈ ( z2 , . . . , zn ) M
F
, and z1 , . . . , zn are not F -independent.
(2) ⇔ (5): The elements z1 , . . . , zn are F -independent in L if and only if for any power q of p,
q q q q [q]
whenever c 1 z1 + · · · + cn zn ∈ m( z1 , . . . , zn ) = m L M , it follows that every c i ∈ m. This is in turn equiv-
q q [q]
alent to the statement that for any power q of p, z1 , . . . , zn is a minimal generating set of L M . 2

Lemma 4.6. Let z1 , . . . , zn ∈ M be F -independent. Then the module L that they generate is strongly
F -independent.

 y1 
..
Proof. Let y 1 , . . . , yn be another minimal set of generators. Then the vector . may be obtained
 z1  yn

..
by multiplying the vector . by an invertible n × n matrix of elements of R. Arguing as in Vraciu
zn
[Vra02], we may reduce to the case where y 1 = z1 + dz2 and y i = zi for all i  2. Here d is some
element of R.
Now, it is clear y i ∈ F
/ ( y 1 , . . . , y i −1 , y i +1 , . . . , yn )M as long as i  3, for in those cases y i = zi and
F
the module for which we claim its non-membership is ( z1 , . . . , zi −1 , zi +1 , . . . , zn ) M .
Next, suppose that y 1 ∈ ( y 2 , . . . , yn ) M . Then for some c ∈ R,
F

q  q q [q]
z1 + dq + c z2 = ( z1 + dz2 )q + cz2 ∈ ( z3 , . . . , zn ) M .

q [q]
Hence, z1 ∈ ( z2 , . . . , zn ) M , contradicting the fact that the zi are F -independent.
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2215

Finally suppose that y 2 ∈ ( y 1 , y 3 , . . . , yn ) M

F
. Then for some q and some c ∈ R,

q  q [q]
cz1 + 1 + cdq z2 ∈ ( z3 , . . . , zn ) M .

q [q]
If c is a unit, this implies that z1 ∈ ( z2 , . . . , zn ) M , which is a contradiction. If c is not a unit, then
q [q]
(1 + cd ) is a unit, which implies that ∈ (z1 , z3 , . . . , zn )M , also a contradiction.
q
z2
Hence, y 1 , . . . , yn are F -independent elements, as was to be shown. 2

Proposition 4.7. Let N be any submodule of M. Then for any minimal F -reduction L of N, the minimal number
[q]
of generators of L is equal to the eventual minimal number of generators of the modules N M for large enough
choices of the power q of p. Hence, Frobenius closure has spread.

Proof. Let z1 , . . . , zt be a minimal set of generators for L. Since z1 , . . . , zt are F -independent, then
q q [q]
for any power q of p, z1 , . . . , zt form a minimal set of generators for L M . On the other hand, for
[q] [q] [q]
sufficiently large q, L M = N M . Hence the minimal number of generators of such an N M is always
equal to the minimal number of generators of L. 2

5. Special part of integral closure

Note: The paper [EH] generalizes some results of this section (e.g. Proposition 5.3) though the point
of view is very different from the one adopted here, in several respects.
For background on integral closure of ideals, the author recommends the recent book [HS06] of
Huneke and Swanson, and in particular Chapter 10 on Rees valuations.

Definition 5.1. For an ideal I in a Noetherian local ring ( R , m), define the special part of the integral
closure of I to be the set

 

I −sp := x ∈ R  ∃n ∈ N such that xn ∈ m I n .

Proposition 5.2. Let ( R , m) be a Noetherian local ring, and J ⊆ I ideals in R. Then I −sp is an ideal,
J −sp ⊆ I −sp , and if R has prime characteristic p > 0, then I ∗sp ⊆ I −sp .

Proof. For the first statement, let x, y ∈ I −sp and a ∈ R. It is obvious from the definition that ax ∈ I −sp .
So we only need to show that x + y ∈ I −sp .
There exist positive integers r and s such that xr ∈ (m I r )− and y s ∈ (m I s )− . Let n = rs. Then

 s  − s  −  −
xn = xr ∈ mIr ⊆ ms I rs ⊆ m I n ,

and by symmetry we also have yn ∈ (m I n )− . So it suffices to show that if xn , yn ∈ (m I n )− , then

(x + y )n ∈ (m I n )− . Since integral closure may be computed modulo minimal primes, we may assume
from this point on that R is an integral domain.
Now, by one of the equivalent definitions for integral closure in integral domains, there is some
c = 0 such that for all positive integers t,

 t
cxnt , c ynt ∈ m I n . (2)

Note also the general fact that arises from looking at monomials that for any nonnegative integers n
and t:

 t
(x + y )n(t +1) ∈ (x + y )n xn , yn . (3)
2216 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

Let d = c 2 (x + y )n . Clearly d = 0, and we have

 t
d (x + y )n = c 2 (x + y )n(t +1)
 t
∈ c 2 xn , yn by (3)


t
 
= cxnj c yn(t − j )
j =0


t
 j t − j
⊆ mIn mIn by (2)
j =0
 t
= mIn .

−
Hence, (x + y )n ∈ (m I n ) , as was to be shown.
It is obvious that J −sp ⊆ I −sp .
The third statement follows from the fact that

 ∗  ∗
m I [q] ⊆ mIq ⊆ mIq

for all powers q = p e of p. 2

We need the following symbols, following Samuel [HS06]:

• v ( I ) := inf{ v (x) | x ∈ I }, and

• ord I (x) := sup{n ∈ N ∪ ∞ | x ∈ I n }.

First note that for any commutative ring R, any (R0 ∪ ∞)-valued valuation v defined on R, and
any proper ideal J of R, we have v ( J̄ ) = v ( J ).

Proof. Since J ⊆ J̄ , v ( J̄ )  v ( J ). On the other hand, let x ∈ J̄ . Then there is some k such that xn+k ∈
J n ( J , x)k ⊆ J n for all n ∈ N. Hence,

 
(n + k) v (x) = v xn+k  v J n = nv ( J ),

n
so that v (x)  n+ k
· v ( J ) for all n ∈ N. It follows that v (x)  v ( J ), whence v ( J̄ )  v ( J ). 2

Proposition 5.3. Let I be an ideal of R and let v 1 , . . . , v t be the Rees valuations of I , with centers p1 , . . . , pt ,
respectively. Let q = p1 ∩ · · · ∩ pt be their intersection. Then the following are equivalent for any x ∈ R:

(1) There is some n0 ∈ N such that xn ∈ I n+1 for all n  n0 .

(2) There is some n ∈ N such that xn ∈ I n+1 .
−
(3) There is some r ∈ N such that xr ∈ ( I r +1 ) .
(4) There is some n ∈ N such that x ∈ q I .
n n

(5) There is some n ∈ N such that xn ∈ (q I n )− .

(6) v i (x) > v i ( I ) for 1  i  t.
(7) x ∈ ( I R pi )−sp for 1  i  t.

In particular, if I is m-primary, then x ∈ I −sp iff xn ∈ I n+1 for some n iff v (x) > v ( I ) for all Rees valuations v
of I .
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2217

Proof. It is obvious that (1) ⇒ (2), (2) ⇒ (3) (taking n = r) and (4) ⇒ (5). Also, (2) ⇒ (4) because
I ⊆ pi for 1  i  t, which implies that I ⊆ q.
(3) ⇒ (2): There is some integer n0 such that for all positive integers k,

 r n0 +k  k
x ∈ I r +1 .

In particular, letting k = n0 r + 1 and n = n0 r 2 + n0 r + r, we have

 n0 +k  r +1 k
xn = xn0 r +n0 r +r = xr = I n0 r +n0 r +r +1 = I n+1 .
2 2
∈ I

(5) ⇒ (6): Suppose xn ∈ q I n , and let v = v i for some 1  i  t. Then we have:

   
nv (x) = v xn  v q I n = v q I n  v pi I n = v (pi ) + nv ( I ) > nv ( I ).

Hence, v i (x) > v i ( I ) for 1  i  t.2

(6) ⇒ (1): By the Rees valuation theorem [Ree88, Theorem 4.16],

ord I (xn ) v i (x)

lim = min > 1.
n→∞ n 1i t v i (I )

ord (x )
So there is some n0 ∈ N such that I n
n
> 1 for all n  n0 . Hence for such n, ord I (xn ) > n, whence
since ord I is integer-valued, ord I (xn )  n + 1, which means that xn ∈ I n+1 .
(7) ⇒ (6) is clear from the definitions. (5) ⇒ (7) is because integral closure is persistent and q ⊆ pi
for 1  i  t.
The last statement follows from the fact that if I is m-primary then each of its Rees valuations has
center m. 2

At this point, the reader may protest that we haven’t yet shown that −sp is a special part of
the integral closure operation. That situation will soon be remedied, but first we note the following
important lemma of Lipman’s from Huneke’s paper:

Lemma 5.4. (See [Hun86, Lemma 3.4].) Let R be a Noetherian local integral domain, let I be an ideal of R, let
K be the quotient field of R, and let x ∈ R. Then if x is in I V for each discrete valuation ring V between R and
K whose center on R is m, then x ∈ I − .

Next, note the following ‘asymptotic’ property associated with the definition of −sp.

Lemma 5.5. Let ( R , m) be a Noetherian local ring, I ⊆ R a proper ideal, x ∈ R, and n0 ∈ N+ . If xn0 ∈ m I n0 ,
then xn ∈ m I n for all n  n0 .

Proof. First assume that R is an integral domain. Let V be any discrete valuation ring between R and
the quotient field K of R whose center on R is m, and let v be its associated discrete valuation on K .
Then we have

2
We have proved a bit more here, actually. In particular,

v i (pi )
v i ( x) − v i ( I )  .
n
2218 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

 
n0 v (x) = v xn0  v m I n0 = v (m) + n0 v ( I )

so that v (x)  v (m)

n
+ v ( I ). Then for any n  n0 , we have v (x)  v (m)
n
+ v ( I ), so that
0

 
v xn = nv (x)  v (m) + nv ( I ) = v m I n .

That is, xn ∈ m I n V for all such V . Hence, by Lemma 5.4, xn ∈ m I n .

If we drop the assumption that R is a domain, the result follows immediately from the fact that
integral closure in R can be computed modulo the minimal primes of R. 2

Lemma 5.6. x ∈ I −sp if and only if this is true modulo all minimal primes of R.

Proof. This follows immediately from the above lemma and the fact that the corresponding statement
is true for integral closure. 2

Proposition 5.7. −sp is a special part of integral closure, in the sense of Definition 2.1.

Proof. We already showed that I −sp is an ideal (i.e. axiom (1) of the definition), and it is clear from
the definitions that m I ⊆ I −sp ⊆ I − (axiom (2)).
Now for (3), suppose that J ⊆ I ⊆ ( J + I −sp )− . First, we may assume without loss of generality that
I is integrally closed. Next, recall that when we were proving that I −sp is an ideal, we showed that
for any n and any x, y ∈ R, if xn , yn ∈ (m I n )− then (x + y )n ∈ (m I n )− . It follows easily from this fact
along with Lemma 5.5 that there is some n such that for any x ∈ I −sp , xn ∈ (m I n )− . Thus, if μ = μ( I ),
then

 μn  n  n(μ−1)  n − n(μ−1)  μn −
I −sp = I −sp I −sp ⊆ mI I ⊆ mI .

There is some r with I r +1 ⊆ ( J + I −sp ) I r . Then letting μ = μ( I ) and m = μn,

 2r +2  2m  m  −sp m  m
Im = I r +1 ⊆ J + I J + I −sp I 2rm
  −  m 2r +1
⊆ J m + mIm I .

Now, after modding out by a minimal prime, we may assume that R is a domain. Let v be any
m-centered valuation. Then

  2r +2   − 
(2r + 2) v I m = v I m  v J m + mIm + (2r + 1) v I m ,

so that

    −
  

v I m  min v J m , v m I m  min v J m , 1 + v I m .

Thus, mv ( I ) = v ( I m )  v ( J m ) = mv ( J ), which means that v ( I )  v ( J ). Since this holds for all m-

centered valuations v, it follows from Lemma 5.4 that I ⊆ J − .
Finally we prove (axiom (4)) for integral closure. Note first that for any minimal prime p,

 −sp
I −sp + p I +p
=
p p
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2219

and

 −
I− + p I +p
= .
p p

Hence, if it holds for integral domains, then we have for any minimal prime p of R:

−  −sp −  −sp −sp

( I −sp ) + p I +p I +p (I −) +p
= = = .
p p p p

Thus (axiom (4)) for integral closure holds in R. So we may assume from now on that R is an integral
domain.
−sp −
Suppose x ∈ ( I − ) . Then for some positive integer n, we have xn ∈ (m( Ī )n ) . Hence, by
Lemma 5.4, for any valuation v on K centered on m in R, where K is the fraction field of R, we
have:

  
v xn  v m( Ī )n = v (m) + nv ( Ī ) = v (m) + nv ( I ) = v m I n .

Hence, by Lemma 5.4 again, xn ∈ (m I n ), so that x ∈ I −sp .

−
Now let x ∈ ( I −sp ) . Then there is some integer r and some elements ai ∈ ( I −sp )i for 1  i  r such
that


r
xr = ai xr −i .
i =1

Take any valuation v of K centered on m in R. Then

  

r v (x) = v xr  min v (ai ) + (r − i ) v (x)  1  i  r .

In particular, for each v there exists some i between 1 and r (dependent on v) such that r v (x) 
v (ai ) + (r − i ) v (x). Hence,

v (ai )
v (x)  .
i

−
But there exists some t such that for all j, atj ∈ (m I jt ) , so that in particular,

v (m)
v (ai )  + i v ( I ).
t

Combining the latest two displayed equations, we have

 r v (m) 
v xrt = rt v (x)  + rt v ( I )  v (m) + rt v ( I ) = v m I rt ,
i

since r  i. Noting that r and t are independent of the choice of v, Lemma 5.4 then implies that
xrt ∈ (m I rt )− , so that x ∈ I −sp . 2
2220 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

6. Evolutions and Fermat’s last theorem

After Eisenbud and Mazur [EM97] connected “evolutionary stability” and the Wiles–Taylor proof of
Fermat’s last theorem with symbolic squares, Hübl [Hüb99] related their methods to certain questions
about integral closure of ideals, as well as the fiber cone of and associated graded ring to an ideal.
In particular, he showed that if k is a field of characteristic 0, and S is a reduced local algebra
essentially of finite type over k, then S /k is evolutionarily stable if and only if it has a presentation
S = R / I , R /k smooth, such that ( R , I ) satisfies the following condition “(NN)”:

 

I ∩ f ∈ R  ∃n such that f n ∈ I n+1 = m I .

In Section 3 of his paper, Hübl considers the following conditions on a ring and an ideal ( R , I ).
(MR) says that if f ∈ I \ m I , then f is contained in some minimal reduction of I . Condition (AR) says
that

 

I ∩ f ∈ R  ∃n such that xn ∈ m I n = m I .

Call the following condition (SP):

I ∩ I −sp = m I .

Clearly (MR) ⇒ (SP) ⇒ (AR) ⇒ (NN), with none of the arrows reversible. Moreover, since −sp is in
fact a special part of integral closure (Proposition 5.7), if follows from Lemma 2.2 that whenever I is
bar-independent, it satisfies (SP), hence also (NN).
Thus, if R is a regular local ring essentially of finite type over a field k of characteristic 0, and I
is a radical bar-independent ideal (e.g. I may be a radical ideal with no proper reductions), R / I is an
evolutionarily stable algebra over k.

7. Special tight closure Briançon–Skoda theorems

The history of “Briançon–Skoda theorems” goes back more than 35 years and could itself be the
subject of a short essay. The original theorem, proved in 1974 by Briançon and Skoda is as follows
(with notation slightly altered):

Theorem 7.1. (See [BS74, Théorème 3].) Let I be an n-generated ideal in the convergent power series ring
R = C{ z1 , . . . , zd }. Then if t = inf{n, d}, ( I − )t ⊆ I .

An algebraic proof, which generalized the theorem to all regular local rings, was given in 1981 by
Lipman and Sathaye:

Theorem 7.2. (See [LS81, special case of Theorem 1].) Let I be an ideal in a regular local ring R, let  be the
−
analytic spread of I , and let w  0 be an integer. Then ( I + w ) ⊆ I w +1 .

This is a generalization because the analytic spread of an ideal is bounded above by both the
number of generators of the ideal and the dimension of the ring.
In 1990, Hochster and Huneke gave a tight closure proof, generalizing the Briançon–Skoda theorem
to all rings of characteristic p (and later, for rings of equal characteristic zero, after tight closure was
well defined for such rings), but not including the mixed characteristic case:

Theorem 7.3. (See [HH90, Theorem 5.6] and [HH99, Theorem 4.1.5].) Let I be an ideal in a Noetherian local ring
R of equal characteristic, let I be an ideal of analytic spread , and let w  0 be an integer. Then ( I + w )− ⊆
( I w +1 )∗ .
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2221

The reason why this generalizes Theorem 7.1 (and Theorem 7.2 when R contains a field) is that
( I w +1 )∗ = I w +1 if R is regular. Theorem 7.3 is really a theorem about the tight closure of an ideal
capturing the integral closure of a not-much-higher power of that ideal. It is noteworthy that although
Theorem 7.2 has a very difficult proof, Theorem 7.3 (at least in characteristic p) is extremely easy once
the foundations of tight closure theory are laid down.
Similarly, we can prove “special” versions, as follows:

Proposition 7.4 (Special tight closure Briançon–Skoda theorem). Let ( R , m) be a Noetherian local ring of char-
acteristic p > 0, and I a proper ideal of R. If n = μ( I ) and w is any nonnegative integer, then

 −sp  ∗sp
I n+ w ⊆ I w +1 .

Proof. Without loss of generality R is an integral domain, since the special parts of both the integral
and tight closures can be computed modulo minimal primes.
−sp
Suppose 0 = x ∈ ( I n+ w ) , where n = μ( I ). Then by Lemma 5.5, there is some power q1 of p with
q1  μ( I ) such that xq1 ∈ m I q1 (n+ w ) . Let q0 be a power of p such that q0  μ(m). Then there exists
some integer k such that for all powers q of p,

 q  qq
xq1 k xq1 q0 = xq1 (k+qq0 ) ∈ m I (n+ w )q1 0
 qq q  q −μ(m)+1  [qq q ] w +1
⊆ mqq0 I n+ w 1 0 ⊆ m[q] 0 I 1 0
  [q q ] [q]
⊆ m I w +1 1 0 .

∗sp
Hence, xq1 q0 ∈ (m( I w +1 )[q1 q0 ] )∗ , which shows that x ∈ ( I w +1 ) . 2

Corollary 7.5 (Special Briançon–Skoda theorem in characteristic p). Let ( R , m) be a Noetherian regular (or
weakly F -regular) local ring of characteristic p > 0, n = μ( I ), and w any nonnegative integer. Then

 −sp
I n+ w ⊆ m I w +1 .

Proof. This follows from Proposition 7.4 and the fact that in a weakly F -regular local ring ( R , m),
J ∗sp = m J for any proper ideal J of R. 2

It would be interesting to prove the above corollary in the equicharacteristic zero case as well,
by reduction to characteristic p, or even in mixed characteristic (perhaps using methods of Lipman,
Sathaye, Teissier, et al.).

8. The special part of the integral closure of monomial ideals

For a standard graded ring S over a field k, there is a unique homogeneous maximal ideal m,
and we may define the special part of the integral closure of a homogeneous ideal J in analogous
fashion, namely let J −sp be the ideal generated by all homogeneous elements x of S such that for
−
some integer t, xt ∈ (m J t ) . Then one can show (routinely) that all homogeneous elements of J −sp .
Moreover:

Lemma 8.1. Let S be a standard N-graded Noetherian domain over a field k, with irrelevant maximal ideal m.
Let J be a homogeneous ideal of S, and let n be the lowest degree among degrees of elements generating J .
Then J −sp contains no homogeneous elements of degree less than or equal to n.
2222 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

Proof. Let x be a homogeneous element of J −sp , and let d be its degree. Then there is some integer
t > 0 with xt ∈ (m J t )− . Hence there is some positive integer k such that

 k−1
xtk ∈ m I t m J t , xt .

The expression on the left-hand side has degree dtk. On the other hand, any element of the expression
on the right-hand side has degree greater than or equal to

1 + nt + (k − 1) min{1 + nt , dt }.

So if d  n, then

dtk  1 + nt + (k − 1)(dt )  1 + kdt ,

which is a contradiction. 2

Convention: For the rest of this section, we will fix a polynomial ring R = k[x1 , . . . , xn ] in n
variables, using the standard N-grading, with k a field. Let m be the homogeneous maximal ideal
(x1 , . . . , xn ).
Note that if f ∈ I , where I is a monomial ideal, then if we express f = c 1 m1 + · · · + cr mr where
the mi are monomials in the variables x j and the c i ∈ k (any such f has a unique such expression, of
course), then mi ∈ I for all i. This is due to the Nn -(multi)graded nature of the polynomial ring R. We
will use this fact repeatedly, sometimes without mentioning it.
It is folk knowledge (see, e.g. [Eis95, Exercises 4.22–4.23]) that if we let Γ ( I ) denote the set of
exponents (as elements of Nn ) of the monomials in a monomial ideal I , then

 
Γ I − = conv Γ ( I ) ∩ Nn ,

where “conv” denotes the convex hull of a subset of Rn .

Another way of expressing this set is as follows. Let {xβ1 , . . . , xβr } be a minimal set of generators
of I . (Here we use double subscripting, so that βi , j is the exponent of x j in the monomial mi .) Then
for a monomial xα , α ∈ Γ ( I − ) if and only if there exist nonnegative rational numbers c 1 , . . . , cr such
that


r 
r
ci = 1 and α c i βi . (4)
i =1 i =1

The partial ordering on Rn we use is the standard one, where γ  δ if γi  δi for all i, and γ > δ
means both that γ  δ and that γ = δ .
With this latter characterization of integral closure of a monomial ideal, we are ready to describe
the special part of the integral closure in similar terms.

Proposition 8.2. Let R = k[x1 , . . . , xn ], where k is a field, and m = (x1 , . . . , xn ). Let I be a monomial ideal of R
contained in m, minimally generated by monomials {xβ1 , . . . , xβr }. Then I −sp is also a monomial ideal, and for
a monomial xα , α ∈ Γ ( I −sp ) if and only if there exist nonnegative rational numbers c 1 , . . . , cr such that


r 
r
c i = 1 and α> c i βi . (5)
i =1 i =1
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2223

Proof. First we show that I −sp is a monomial ideal. Let f = b1 m1 + · · · + b u mu ∈ I −sp , where the
mi are distinct monomials and 0 = b i ∈ k for all i. Then there is some positive integer t such that
f t ∈ m I t . In particular, since the latter is a monomial ideal, mti ∈ m I t for i = 1, . . . , r, which means
that mi ∈ I −sp for each i.
Now, let α ∈ Nn , and suppose that xα ∈ I −sp . Then there is some t with xt α ∈ m I t . Then by Vitulli
[Vit03, Corollary 3.2], there is some s with xst α ∈ ms I st . In particular, there exists a positive integer q
with xqα ∈ m I q . This means that xqα is a multiple of one of the generating monomials of m I q . In
particular, there exist nonnegative integers p 1 , . . . , p r such that


r
pi = q and qα > p 1 β1 + · · · + p r βr . (6)
i =1

(The “>” is because xqα ∈ m I q , and not merely in I q .) Then dividing through by q and letting c i = p i /q,
we get (5).
Conversely, suppose that α and c 1 , . . . , cr satisfy (5). Since the c i are rational, they have a common
denominator, say q, so that there are nonnegative integers p 1 , . . . , p r such that c i = p i /q for each i,
satisfying (6). Hence xqα ∈ m I q , which means that xα ∈ I −sp . 2

We use this to tell us exactly when special decomposition of integral closure fails for monomial
ideals. First, for any subset C of Rn , let

low(C ) := { P ∈ C | Q ≮ P for all Q ∈ C },

the “lowest points” of C .

Corollary 8.3. Let R = k[x1 , . . . , xn ], where k is a field, and m = (x1 , . . . , xn ). Let I be an ideal of R minimally
generated by distinct nontrivial monomials {xβ1 , . . . , xβr }. Let S = {β1 , . . . , βr }. Then Γ ( I − ) is the disjoint
union of Γ ( I −sp ) with low(conv( S )) ∩ Nn . Hence, I − = I + I −sp if and only if S = low(conv( S )) ∩ Nn .

For example, if I = (xt , yt ), we have I −sp = m I , but I − = (x, y )t , so the decomposition fails if t > 1.
p p
In general, if I = (x1 1 , . . . , xn n ) for integers p i , then the decomposition holds if and only if whenever
1  i < j  n, gcd( p i , p j ) = 1.

9. Intersections and compatibility

Lemma 9.1. Let I be a proper ideal in the local ring ( R , m) of characteristic p > 0. Then I −sp ∩ I ∗ = I ∗sp .

Proof. Let f ∈ I −sp ∩ I ∗ . Then there is some q1 with f q1 ∈ m I q1 , and some q0 and some c ∈ R o with
c f q ∈ I [q] for all q  q0 . Thus, there is some d ∈ R o such that for all powers q, q2 of p, df qq1 q2 ∈
mqq2 I qq1 q2 . Then we have:

 q
cd f q1 q2 = cdf qq1 q2 ∈ mqq2 I qq1 q2 ∩ I [qq1 q2 ]
 [qq ]
⊆ mqq2 I q2 −μ( I )+1 1 ∩ I [qq1 q2 ]
⊆ mqq2 I [qq1 ] ∩ I [qq1 q2 ]
⊆ mqq2 −r I [qq1 q2 ]
 [q]
⊆ m[q] I [qq1 q2 ] = m I [q1 q2 ] .

Thus, f q1 q2 ∈ (m I [q1 q2 ] )∗ , so f ∈ I ∗sp . 2

2224 N. Epstein / Journal of Algebra 323 (2010) 2209–2225

Lemma 9.2. If ( R , m) is an excellent analytically irreducible local ring and I is a ∗-independent ideal in R,
then I ∗sp ∩ I F = I F sp .

Proof. Let f ∈ I ∗sp ∩ I F , and let f 1 , . . . , f d be a ∗-independent generating set of I . Then there is some
power q1 of p such that f q1 ∈ I [q1 ] . Hence


d
q
f q1 = ai f i 1 .
i =1

Also, there is some q0 such that c f q ∈ m[q/q0 ] I [q] for all q  0. That is,


d
q
cf q = miq f i ,
i =1

where miq ∈ m[q/q0 ] for all such q. On the other hand, from the first displayed equation we also have


d
q/q1 q
cf q = cai fi .
i =1

Combining the previous two displayed equations, we have


d
 q/q1 q
cai − miq f i = 0.
i =1

Since f 1 , . . . , f d are ∗-independent and the colon criterion [Abe01, Proposition 2.4] holds in R, there
is some power q2  max{q0 , q1 } of p such that cai 1 − miq ∈ m[q/q2 ] , so that cai 1 ∈ m[q/q2 ] for all
q/q q/q

∈ m∗ = m, which implies that ai ∈ m. So we have f q1 ∈ m I [q1 ] ,

q /q1
q  0 and all 1  i  d. Hence ai 2
whence f ∈ I F sp . 2

Acknowledgment

The author wishes to thank Professor Craig Huneke for constant support and advice during the
development of these results.

References

[Abe01] Ian M. Aberbach, Extension of weakly and strongly F -regular rings by flat maps, J. Algebra 241 (2001) 799–807.
[BS74] Joël Briançon, Henri Skoda, Sur la clôture intégrale d’un idéal de germes de fonctions holomorphes en un point de Cn ,
C. R. Acad. Sci. Paris Sér. A 278 (1974) 949–951.
[EH] Neil Epstein, Mel Hochster, Special integral closure, interior integral closure, and continuous closure, 2009, in prepara-
tion.
[Eis95] David Eisenbud, Commutative Algebra, Grad. Texts in Math., vol. 150, Springer-Verlag, New York, 1995.
[EM97] David Eisenbud, Barry Mazur, Evolutions, symbolic squares, and Fitting ideals, J. Reine Angew. Math. 488 (1997) 189–
201.
[Eps05] Neil Epstein, A tight closure analogue of analytic spread, Math. Proc. Cambridge Philos. Soc. 139 (2) (2005) 371–383.
[FV] Louiza Fouli, Janet Vassilev, The cl-core of an ideal, Math. Proc. Cambridge Philos. Soc., 2008, in press.
[HH90] Melvin Hochster, Craig Huneke, Tight closure, invariant theory, and the Briançon–Skoda theorem, J. Amer. Math.
Soc. 3 (1) (1990) 31–116.
[HH99] Melvin Hochster, Craig Huneke, Tight closure in equal characteristic zero, 1999, book-length preprint.
[HS06] Craig Huneke, Irena Swanson, Integral Closure of Ideals, Rings, and Modules, London Math. Soc. Lecture Note Ser.,
vol. 336, Cambridge Univ. Press, Cambridge, 2006.
 N. Epstein / Journal of Algebra 323 (2010) 2209–2225 2225

[Hüb99] Reinhold Hübl, Evolutions and valuations associated to an ideal, J. Reine Angew. Math. 517 (1999) 81–100.
[Hun86] Craig Huneke, The primary components of and integral closures of ideals in 3-dimensional regular local rings, Math.
Ann. 275 (1986) 617–635.
[LS81] Joseph Lipman, Avinash Sathaye, Jacobian ideals and a theorem of Briançon–Skoda, Michigan Math. J. 28 (2) (1981)
199–222.
[NR54] D.G. Northcott, D. Rees, Reductions of ideals in local rings, Proc. Cambridge Philos. Soc. 50 (2) (1954) 145–158.
[Ree88] D. Rees, Lectures on the Asymptotic Theory of Ideals, London Math. Soc. Lecture Note Ser., vol. 113, Cambridge Univ.
Press, Cambridge, 1988.
[Vas09] Janet C. Vassilev, Structure on the set of closure operations of a commutative ring, J. Algebra 321 (10) (2009) 2737–
2753.
[Vit03] Marie A. Vitulli, Some normal monomial ideals, in: Topics in Algebraic and Noncommutative Geometry, Luminy/
Annapolis, MD, 2001, in: Contemp. Math., vol. 324, Amer. Math. Soc., Providence, RI, 2003, pp. 205–217.
[Vra02] Adela Vraciu, ∗-Independence and special tight closure, J. Algebra 249 (2002) 544–565.
[Vra06] Adela Vraciu, Chains and families of tightly closed ideals, Bull. Lond. Math. Soc. 38 (2006) 201–208.