PROCEEDINGS OF THE

AMERICAN MATHEMATICAL SOCIETY

Volume 68, Number 1, January 1978

ON THE GEOMETRICINTERPRETATIONOF
SEMINORMALHTY
EDWARDD. DAVIS

ABsTRAmC. We give a new characterization of the propertyof seminormality

for 1-dimensionallocal rings; and with the help of this resultwe develop
extensionsof the theoremsof Salmon and Bombieriwhich interpretthis
property in terms of singularities of algebraic curves.

1. Introduction.Let R be a reduced, 1-dimensionalnoetherianring with

normalization(in its total quotientring) a finite R-module.It is known that
Pic(R) -- Pic(R[X]) is an isomorphismif, and only if, the conductorof the
normalization,viewed as an ideal of the largerring, is its own radical[B-M,
7.12], [E, 4.3]. These conditions are equivalent to what is now called
seminormality[T, 3.6]. If R is the local ring of a point of a scheme, then the
point is called seminormalprovidedthat R is seminormal.Assume thatp is a
singularpoint of an algebraiccurveover an algebraicallyclosed field. Salmon
[S] provesthat in the plane curvecase,p is seminormalif, and only if, it is an
ordinarydouble point. Bombieri[B] generalizesthis result by droppingthe
planarrestrictionand provingthatp is seminormalif, and only if, analytically
the curve locally at p consists of smooth brancheswith linearlyindependent
tangents,i.e., the curve has "normalcrossings"at p. Now in [S] it is claimed
that the theoremis true over arbitrarygroundfield providedthat "ordinary
double point" is interpretedto mean that the tangent cone splits over the
algebraicclosure into two distinct lines. This is not quite true. In [B] it is
stated,but not reallyproved,that the seminormalityof p is equivalentto "p is
an ordinaryn-fold point, where n is the dimension of the Zariski tangent
space at p". We shall give a new characterizationof seminormalityof a
1-dimensionallocal ring which shows that this propertymay be described
purely algebraically in terms of the associated graded ring, or purely
geometricallyin terms of the projectivizedtangent cone. This result has as
corollary the theorem stated in [B] and the correct version of the theorem
claimed in [S]. We conclude with a brief exposition of how the "normal
crossings"resultof [B] extendsto the abstractcase.
Standingnotation.R is a reduced,noetherian,1-dimensionallocal ring,with
normalizationD a finite R-module.M, k, e(R) and emdim(R) denote the

Receivedby the editorsJanuary23, 1976and, in revisedform,July28, 1976and May 28, 1977.

AMS (MOS) subjectclassifications(1970).Primary13H15, 14H20.
Keywordsandphrases.1-dimensionallocal ring, seminormality,algebraiccurve,tangentcone,
ordinarypoint, branches,normalcrossings.
t) American
Mathematical
Society1978

This content downloaded from 128.235.251.160 on Thu, 8 Jan 2015 07:40:12 AM

All use subject to JSTOR Terms and Conditions
2 E. D. DAVIS

maximalideal, residuefield, multiplicityand embeddingdimensionof R. Let

J denote the Jacobsonradicalof D and K = D/J. For any semilocalringA,
G(A) denotes the associated graded ring (with respect to its Jacobson
radical).
Observethat R is seminormalif, and only if, M = J.

viewpoint.The first theoremis the resultreferred

2. The algebraic-geometric
to in ?1.
1. Thefollowingstatementsare equivalent.
THEOREM
(1) R is seminormal.
(2) G(R) is reducedand e(R) = emdim(R).
(3) Proj(G(R)) is reducedande(R) = emdim(R).
Before proceeding with the proof we state and discuss the following
geometriccorollary.
COROLLARY1. Letp be a closedpoint of an algebraic(or algebroid)curveat
whichthe Zariskitangentspacehas dimensionn. Then:
(1) (cf. [S])p is seminormalif, and only if, it is an n-foldpoint at whichthe
projectivizedtangentconeis reduced.
(2) (cf. [B]) For algebraicallyclosedgroundfield,p is seminormalif, and only
if, it is an ordinaryn-foldpoint (i.e., a point of multiplicityn with n distinct
tangents).
(1) is clearly a special case of (1) *-* (3) of the theorem, and (2) is valid
because over algebraicallyclosed groundfield "ordinarypoint"and "projec-
tivized tangent cone reduced" are equivalent. Something weaker than
algebraicclosuresuffices: it is enough that all points on the normalizationof
the curvelying overp be rationalover the residuefield atp. Henceforthif our
local ring R has this property,we say that R has k-rationalnormalization.In
connectionwith (1) and [S] we consideran example.Supposethat char(k) =
2 and that b2 = a E k, with b E k - k. We considerthe plane curve over k
defined by the equation y2 = aX2 + X3. This curve is absolutely reduced
and irreducible(and thereforea variety in anyone's sense of the term) since
over k the equationbecomes (Y + bX)2 = X3. Let R be the local ring of the
origin on this curve. Clearly e(R) = emdim(R) = 2, and G(R) =
k[X, y]/ y2 + aX2 is reduced;hence R is seminormal.But y2 + aX2 =
(Y + bX)2; i.e., the affine tangentcone does not split over k, as claimed in
[S], into two distinctlines. But this is in some sense the uniquecounterexam-
ple. Observethat, in general,G(R) will remainreducedover k if, and only if,
K is a separablek-algebra;and this is the case whenevere(R) < char(k).
Now for the proof of Theorem 1 we require the following technical
characterizationof seminormality.
LEMMA.R is seminormalif, and only if, the followingconditionshold. (i)
e(R) = emdim(R);(ii) MJ n M = M2.

This content downloaded from 128.235.251.160 on Thu, 8 Jan 2015 07:40:12 AM

All use subject to JSTOR Terms and Conditions
 GEOMETRICINTERPRETATIONOF SEMINORMALITY 3

PROOF. Since dim(R) = 1, e(R) = k-dim(D/MD). But J and D are

isomorphicR-modulessince J = xD for some regularelementx of J. Hence
e(R) = k-dim(J/MJ). Suppose that R is seminormal;i.e., M = J. Then
e(R) = k-dim(J/MJ) = k-dim(M/M2) emdim(R), and MJ n M = M2.
To prove the converse observe that the inclusion M -- J induces a k-linear
map M/M2 -- J/MJ with kernelMJ n M/M2. By (ii) this map is injective;
and by (i) k-dim(M/M2) = k-dim(J/MJ). So (i) and (ii) imply that the map
is surective. ThusJ = M + MJ; whenceM = J (Nakayama'slemma).
PROOFOF THEOREM1. (1) -- (2). Observethat the k-linearmaps Mn/Mn+ '
Jnl/Jn+l induce a graded k-algebra homomorphism G(R) -- G (D). In
any event G(D) is reduced: G(D)_ K[T], T an indeterminate.If R is
seminormal,then e(R) = emdim(R) (Lemma),and G(R) is reducedbecause
the map G (R) --G (D) is injective when M = J. (2) -- (1). Suppose that R is
not seminormal.Then by the Lemma,there is x E MJ n M, x M M2. Now
xn E MnJn (n > 0), and for sufficientlylarge n, Jn c M because J is the
radical of the conductor. For such n, x n E Mn + 1; i.e., the leading form of x
in G (R) is nilpotent. (2) -- (3) is clear. (3) -- (2). Since the points of
Proj(G(R)) correspondto the minimal prime ideals of G(R), the fact that
Proj(G(R)) is reducedimplies that the minimalprimarycomponentsof the
0-ideal in G(R) are prime. Hence in orderto prove that G(R) is reduced,it
suffices to show that the irrelevantmaximalideal of G(R) is not an associ-
ated primeof the 0-ideal. One can prove this directly;but insteadwe quote a
more generalresultof J. Sally [Sy] valid for Cohen-Macaulaylocal ringsR of
arbitrarydimension: G(R) is Cohen-Macaulayprovided that emdim(R)=
e(R) + dim(R) - 1.
REMARK. One can prove (3) -* (1) by using the "M = J" criterion and the
methodsof ?1 of Lipman'spaper [L].
We conclude this section by giving a more explicit descriptionof G(R).
For T an indeterminatelet G = {f E K[TIIf(0) E k}.

COROLLARY 2. R is seminormalif, and only if, G(R) = G (isomorphismof

gradedk-algebras).
PROOF. Let t E G(D) be the leading form of x, where J = xD. Then
t " T induces a graded K-algebra isomorphismof G(D) with K[T]. If
M = J, then this isomorphismidentifies G(R) with G. Note that if G(R) _
G, then k-dim(Mn/Mn+l) = k-dim(K) (n > 0). Since in any event we have
e(R) = k-dim(Mn/MMn+)for all large n, it follows that e(R) = emdim(R) if
G(R) G. And G is clearlyreduced.
REMARK.(Thanks to the referee for bringingto light this supplementto
Corollary2.) Observefirst that G-say by the conductorcriterion-isseminor-
mal with normalizationK[ T]. It follows that G is the seminormalizationof
each of its graded k-subalgebrashaving normalizationK[T]. (See [T] for
"seminormalization.")Next note that G(R) is reduced if, and only if, the
homomorphismG(R) -* G(D) is injective.(Proof: essentiallyidentical with

This content downloaded from 128.235.251.160 on Thu, 8 Jan 2015 07:40:12 AM

All use subject to JSTOR Terms and Conditions
4 E. D. DAVIS

the "reduced"part of the proof of (1) *-*(2) of Theorem 1.) It follows that
when G(R) is reduced,the identificationof G(D) with K[ T] identifiesG(R)
with a k-subalgebraof G having normalizationK[ T]. These comments,
together with Corollary2, prove: R is seminormalif, and only if, G(R) is
reducedand seminormal.Consequently:D is obtainedfrom R by glueingover
M if, and only if, G(D) is obtainedfrom G(R) by glueingover the homoge-
neousmaximalideal. (See [TI for "glueing.")
Let {MjII < i < n) be the set of maximalideals of D, and K, = D/M,.
For each i let k-dim(K,)= d(i) and let {a,u11< j < d(i)) be a k-basis of K,.
Since K is the k-algebra direct sum of (K,), the set {a.) is canonically
identifiedwith a subsetof K. Let X = {X,jl < i < n, 1 < j < d (i)) be a set
of indeterminates,and define a k-algebrahomomorphismk[X] -* K[ T] by
X, - a. T. One easily verifies that the image is G and the kernel is of the
form Sk[X], where S = SI u ... u Sn u {X#Xuvii# u) and S, is a set of
homogeneous generators of the kernel of the k-algebra homomorphism
k[Xl, ** *,X,d(,l]-* K[ T]. Hence:

COROLLARY 3. R is seminormalif, and only if, G(R). k[X]/Sk[X]

(graded k-algebra isomorphism). And, in particular,when R has k-rational
normalization,
R is seminormalif, andonlyif,
G(R) k[X, Xn]/ (X,Xjli7I}k[Xi,...
... , *X]* ,

3. The algebroid-geometricviewpoint.We retain in force the notation

introduced at the end of ?2. Let G = {f E KffT]] f (0) E k). One easily
verifies: G is local with normalizationK[[T]]; G is seminormal;Xy - aa,T
defines a continuous k-algebra homomorphismwith image G and kernel
Sk[[X]]. Assume for the moment that R is complete and equicharacteristic.
Then: k is contained in R; K is contained in D; and we have a continuous
K-algebraisomorphismD = K[[ T]], whichidentifiesR with G if, and only if,
R is seminormal.Consequently,for complete equicharacteristicR, we have
that R is seminormalif, and only if, R k[[X]]/Sk[[X]] (continuous k-
algebraisomorphism).All this is noted in [B], albeit for the case in which R
has k-rationalnormalization.
Now let "^"denote completion. Let P, denote the minimal prime ideal of R
correspondingto M,; i.e., Pi = Q, n R, whereQ,is the uniqueminimalprime
ideal of D contained in M.. Recall that the set of (algebroid,or analytic)
branches of R is {R/P,}. (Properly speaking we should say that
{Spec(R/P,)) is the set of branchesof Spec(R).) Since the Zariski tangent
space Z(R) of R is the k-linear dual of M/M2 = M/M2, we see that
Z(R) = Z(R). Furthermore,since Z(R/P) is the dual of M/P + M2 = the
annihilatorin Z (R) of the subspaceP + M2/M2, the Zariskitangentspaces
of the branchesare canonicallyembedded in Z(R), and we may therefore
meaningfullyspeakof theirlineardependenceor independence.
Observethat by the "M = J" criterion,R is seminormalif, and only if R,
is seminormal.With this fact and the above explicit descriptionof complete

This content downloaded from 128.235.251.160 on Thu, 8 Jan 2015 07:40:12 AM

All use subject to JSTOR Terms and Conditions
 GEOMETRICINTERPRETATIONOF SEMINORMALITY 5

equicharacteristicR, one easily shows that the following theoremis valid in

case.
the equicharacteristic
THEOREM 2. Thefollowingare equivalent.
(1) R is seminormal.
(2) The branchesof R are seminormaland the Zariskitangentspace of R is
the directsum of the Zariskitangentspacesof the branches.
(3) The branchesof R are seminormaland M has a minimalbasis {x. I1 < i
< n, 1 < j < d (i)), with {xujIu# i) a basisfor P, (for each i).
Claim.Theorem2 is valid without the equicharacteristic assumption.(3)
(1) is valid since from (3) one can deduce the explicit descriptionof G(R)
given in Corollary 3. (1) -- (2) --* (3) can be proved in the spirit of the
equicharacteristiccase, but the technical details must necessarily be
somewhatdifferentowing to the absence of a coefficientfield. We omit these
details which are tedious, but quite straightforward,manipulationsof stan-
dardtechniquesof elementarylinearand local algebra.
Finally we single out the special case wherein R has k-rational norma-
lization.As in the classicalcase, "normalcrossings"means "smoothbranches
with linearlyindependenttangents".
COROLLARY 4. Suppose that R has k-rational normalization.Then R is
seminormalif, and onlyif, R has normalcrossings.
Recent resultsof F. Orecchia[0, 1.1, 2.3, 2.9] imply the validity
REMARKs.
of (1) +->(2) of Theorem 2. Theorem 1 remains valid if the standing hypothesis
"reducedwith finite normalization"is weakenedto "Cohen-Macaulay".But
this is not a true generalization:under the weakerhypothesis(1), (2) and (3)
each imply the strongerhypothesis.Theorem2 does not remainvalid under
the weakerhypothesis;e.g., R = k[[X, Y]]/XY2k[[X, Y]] satisfies(2) and (3)
but not (1). If "branches"means {R/P,}, {P,) the set of primarycomponents
of 0, then Theorem2 is valid under the weakerhypothesis.But the criticism
of the above generalizationof Theorem 1 also applieshere.
REFERENCES
[B-.M]H. Bassand M. P. Murthy,Grothendieck groupsandPicardgroupsof abeliangrouprings,
Ann. of Math.(2) 86 (1967),16-73.
[B] E. Bombieri,Seminormalita e singolaritacordinarie,SymposiaMathematicaXI, Academic
Press,New York, 1972,pp. 205-210.
[E]S. Endo,Projectivemodulesoverpolynomialrings,J. Math.Soc. Japan15 (1963),339-352.
[L]J. Lipman,StableidealsandArf rings,Amer.J. Math.93 (1971),649-685.
[Sy] J. Sally, On the associatedgradedringof a local Cohen-Macaulay ring,Kyoto J. Math. 17
(1977), 19-21.
[S] P. Salmon,Singolaritae gruppodi Picard,SymposiaMathematicaII, AcademicPress,New
York, 1969,pp. 341-345.
[T] C. Traverso,Seminormality and the Picardgroup,Ann. ScuolaNorm. Sup. Pisa 24 (1970),
585-595.
[0] F. Orecchia,Sullaseminormalita di certevarietaaffiniriducibili,Boll. Un. Mat. Ital. (5) 13B
(1976),588-600.
DEPARTMENTOF MATHEMATICS,STATEUNIVERSITYOF NEW YoRK, ALBANY, NEW YORK 12222

This content downloaded from 128.235.251.160 on Thu, 8 Jan 2015 07:40:12 AM

All use subject to JSTOR Terms and Conditions