ON ALGEBRAS WITH RETRACTIONS
AND
ON PLANES OVER A DVR
Prosenjit Das
Indian Statistical Institute 2009.
ON ALGEBRAS WITH RETRACTIONS
AND
ON PLANES OVER A DVR
Prosenjit Das
Thesis submitted to the Indian Statistical Institute in partial fulο¬llment of the requirements
for the award of the degree of Doctor of Philosophy
December, 2009
Thesis supervisor: Dr. Amartya Kumar Dutta
Indian Statistical Institute 203, B.T. Road, Kolkata, India.
i
First of all, I would like to express my sincere gratitude to my supervi- sor Dr. Amartya Kumar Dutta. Words are not enough to express what his guidance and constant support have meant to me. He is the person whose stimulating teaching and encouragement motivated me to take up the study of Commutative Algebra. He spent an immense amount of his valuable time in teaching me various aspects of Commutative Algebra, supervising my works and correcting innumerable mistakes. Every aspect of my work has developed under his expert supervision and this thesis would have been impossible but for his guidance. I have learnt a lot from him, not only about mathemat- ics, but about many aspects of everyday life. For his patience, care and help during the last ο¬ve years, I will always be indebted to him.
I would like to thank all my teachers of my school, college and university for introducing me to the world of knowledge. I would especially like to thank Prof. K. C. Chattopadhyay who was ο¬rst to help me in gaining a proper perspective of the world of mathematics and, in the process, opened my eyes to the beauty of the subject. His expert teaching and wonderful personality have deeply inο¬uenced my views of mathematics and of life.
I am also grateful to Prof. Amit Roy for his warm and selο¬ess support, encouragement and advice in many walks of my life. I feel that I am really fortunate to get to know and to be in touch with such a marvellous person and teacher.
My deep respect to Professors S.M. Bhatwadekar, T. Asanuma and N.
Onoda for teaching and discussing many important research topics. Their expert comments, useful suggestions and stimulating discussions enriched my work.
My profound thanks to Miss Neena Gupta for her valuable suggestions, fruitful discussions and constant support. Discussions with her have always been a stimulating experience.
I wish to express my appreciation and gratitude to all the faculty in the Stat-Math Unit, especially Professors Ashok Roy, Somesh Chandra Bagchi, Shashi Mohan Srivastava, Rana Barua, Goutam Mukherjee, Alok Goswami,
iii
ever I needed it.
Heartiest thanks to all of my friends in ISI who have made my life wonderful with such an encouraging and relaxed research environment and also a happy and entertaining hostel life.
I can not simply forget the care and help I have got from Dr. Ashis Mandal during my tough times in my Ph.D. life. His constant encouragement gave me strength during my stay at ISI. My immense thanks to him. I am also thankful to Abhijit Pal for many helpful discussions.
Finally, a very special and heartfelt thanks to my parents, my sisters and my ο¬ancee and her parents for their constant support, love and good wishes which helped me to concentrate on my work throughout my Ph.D. years.
iv
Throughout the thesis all rings will be assumed to be commutative rings with unity. For a commutative ringπ , a prime idealπ ofπ , and anπ -algebra π΄, the following notation will be used:
ππππ(π ) : The set of all prime ideals ofπ .
βπ‘(π) : The height of the idealπ.
ππ‘(π ) : The ο¬eld of fractions of π , when π is an integral domain.
π [π] : Polynomial ring inπvariables overπ .
π β : Group of units of π .
π(π) : Residue ο¬eldπ π/π π π. π΄π : =πβ1π΄where π=π βπ.
ππ¦ππ (π) : Symmetric algebra of anπ -moduleπ overπ .
π΄π’π‘π (π΄) : The group ofπ -algebra automorphisms ofπ΄.
π‘π.ππππ (π΄) : Transcendence degree ofπ΄overπ .
πβ(π ) : Characteristic of π .
Ξ©π (π΄) : The universal module of diο¬erential ofπ΄ overπ .
DVR : Discrete valuation ring.
v
1 Introduction 1
2 Preliminaries 11
3 Codimension-one πΈ1-ο¬bration with retraction 17 3.1 Preview . . . 17 3.2 A version of Russell-Sathaye criterion for an algebra to be a
polynomial algebra . . . 18 3.3 Codimension-oneπΈ1-ο¬bration with retraction . . . 23 4 Factorial πΈ1-form with retraction 33 4.1 Preview . . . 33 4.2 Main Result . . . 35 5 Planes of the form π(π, π)ππβπ(π, π) over a DVR 39 5.1 Preview . . . 39 5.2 Planes of the formπππβπover a ο¬eld . . . 40 5.3 Planes of the formπππβπover a DVR . . . 46 5.4 Planes of the formπππβπover rings containing a ο¬eld . . . . 49
Bibliography 51
vii
Introduction
Aim:
The main aim of this thesis is to study the following problems:
1. For a Noetherian ring π , to ο¬nd a set of minimal suο¬cient ο¬bre condi- tions for an π -algebra with a retraction to π to be an πΈ1-ο¬bration over π .
2. To investigate suο¬cient conditions for a factorialπΈ1-form, with a retrac- tion to the base ring, to be πΈ1.
3. To investigate whether planes of the formπ(π, π)ππβπ(π, π) are co- ordinate planes in the polynomial ring in three variables π, π and π over a discrete valuation ring.
The 1st problem will be discussed in Chapter 3 entitled βCodimension- oneπΈ1-ο¬bration with retractionβ, the 2ndproblem will be studied in Chapter 4 under the heading βπΈ1-form with retractionβ and the 3rd problem will be investigated in Chapter 5 which has the title βPlanes of the formπ(π, π)ππβ π(π, π) over a DVRβ.
Brief introductions to the topics of the problems and precise statements of the main results obtained are given below:
β Codimension-one πΈ
1-ο¬bration with retraction
Let π be a ring. A ο¬nitely generated ο¬at π -algebra π΄ is said to be an πΈ1- ο¬bration over π ifπ΄βπ π(π) = π(π)[1] for all prime ideals π of π . A very
1
interesting and important phenomenon is that the generic and codimension- one ο¬bres determine an πΈ1-ο¬bration. To get a feel for this striking feature of πΈ1-ο¬bration, here is a nice result by Bhatwadekar-Dutta ( [BD95]):
Theorem 1.0.1. Let π be a Noetherian domain with ο¬eld of fractionsπΎ and π΄ an π -subalgebra of π [π1, π2,β β β , ππ]such that π΄ is ο¬at over π , π΄βπ πΎ = πΎ[1] and π΄βπ π(π) is an integral domain for every prime ideal π in π of height one. Then
(i) If π is normal, then π΄βΌ=ππ¦ππ (πΌ) for an invertible ideal πΌ of π .
(ii) If π containsβ, then π΄ is anπΈ1-ο¬bration over π .
(iii) Ifπ is seminormal and containsβ, thenπ΄βΌ=ππ¦ππ (πΌ) for an invertible idealπΌ of π .
An analogous result has also been obtained by Dutta ( [Dut95]) for ο¬nitely generated faithfully ο¬atπ -subalgebras:
Theorem 1.0.2. Let π be a Noetherian domain with ο¬eld of fractionsπΎ and π΄ a faithfully ο¬at ο¬nitely generated π -algebra such that π΄βπ πΎ =πΎ[1] and π΄βπ π(π) is geometrically integral for every prime idealπ in R of height one.
Then
(i) If π is normal, then π΄βΌ=ππ¦ππ (πΌ) for an invertible ideal πΌ of π .
(ii) If π containsβ, then π΄ is anπΈ1-ο¬bration over π .
(iii) Ifπ is seminormal and containsβ, thenπ΄βΌ=ππ¦ππ (πΌ) for an invertible idealπΌ of π .
We will call anπ -algebraπ΄aCodimension-oneπΈ1-ο¬bration ifπ΄βπ π(π) = π(π)[1] for each prime ideal π of π with βπ‘(π) β€ 1. In view of the above theorems it is easy to see that
1. For a Noetherian normal domainπ or a Noetherian domainπ containing β, a ο¬atπ -subalgebraπ΄of a polynomial algebra overπ is anπΈ1-ο¬bration overπ if and only if π΄ is a codimension-oneπΈ1-ο¬bration over π .
2. For a Noetherian normal domainπ or a Noetherian domainπ containing β, a faithfully ο¬at ο¬nitely generatedπ -algebraπ΄is anπΈ1-ο¬bration over π if and only if π΄is a codimension-one πΈ1-ο¬bration over π .
In ( [Asa87], Theorem 3.4), Asanuma has given a structure theorem for πΈπ-ο¬brations over a Noetherian ring. The statement of Asanumaβs theorem shows that
A necessary condition for an algebra π΄ over a Noetherian ring π to be πΈπ-ο¬bration is that π΄ is isomorphic, as an π -algebra, to an π -subalgebra of some polynomial ring overπ .
As a consequence of this result we get that anyπΈπ-ο¬bration over a Noethe- rian ring has a retraction toπ . Therefore, whenπ is Noetherian, it is natural to ask for minimal suο¬cient ο¬bre conditions which ensure that an π -algebra with a retraction to π will be a codimension-oneπΈ1-ο¬bration overπ .
Recently, in [BDO], Bhatwadekar-Dutta-Onoda have shown, as a conse- quence of a general structure theorem for any faithfully ο¬at π -algebra over a Noetherian normal domain which is locallyπΈ1 in codimension-one, that for a Noetherian normal domainπ , a ο¬at π -algebra π΄ with a retraction toπ is an πΈ1-ο¬bration overπ (in fact,ππππ(π΄) is an algebraic line bundle overππππ(π )) ifπ΄ is locallyπΈ1 in codimension-one; more precisely,
Theorem 1.0.3. Letπ be a Noetherian normal domain with ο¬eld of fractions πΎ and π΄ a Noetherian ο¬at π -algebra such that π΄π = π π[1] for each prime idealπ ofπ of height one. Suppose that there exists a retractionΞ¦ :π΄βββ π .
Then π΄βΌ=ππ¦ππ (πΌ) for an invertible ideal πΌ in π .
In view of the above results, naturally one asks the following questions:
(1) Is Theorem 1.0.1 true when the condition βπ΄ is an π -subalgebra of π [π1, π2,β β β , ππ]β is replaced by the condition βπ΄ has a retraction to π β?
(2) Is Theorem 1.0.2 true when the condition βπ΄ is a faithfully ο¬at ο¬nitely generatedπ -algebraβ is replaced by the condition βπ΄is a ο¬atπ -algebra with a retraction toπ β?
(3) How far can the hypothesis βπ is normalβ in Theorem 1.0.3 be relaxed?
In Chapter 3 of the thesis, we investigate the above questions. We will show that questions (1) and (2) have answers in the aο¬rmative when π΄ is Noetherian; the results also show that Theorem 1.0.3 holds in more generality.
The main results of this study are listed below (Proposition 3.3.4, Theorem 3.3.5, Theorem 3.3.7, Theorem 3.3.9):
Proposition A. Let π be either a Noetherian domain or a Krull domain with ο¬eld of fractionsπΎ and π΄ a ο¬at π -algebra with a retraction Ξ¦ :π΄βββ π such that
(1) πΎππ Ξ¦ is ο¬nitely generated.
(2) π΄π =π π[1] for every prime ideal π of π satisfying ππππ‘β(π π) = 1.
Then there exists an invertible ideal πΌ of π such that π΄βΌ=ππ¦ππ (πΌ).
Theorem A. Let π be a Krull domain with ο¬eld of fractions πΎ and π΄ a ο¬at π -algebra with a retraction Ξ¦ :π΄βββ π such that
(1) Ker Ξ¦ is ο¬nitely generated.
(2) π΄βπ πΎ=πΎ[1].
(3) π΄βπ π(π)is an integral domain for each height one prime ideal π ofπ .
Then there exists an invertible ideal πΌ of π such that π΄βΌ=ππ¦ππ (πΌ).
Theorem B. Let π be a Noetherian domain with ο¬eld of fractionsπΎ and π΄ a ο¬at π -algebra with a retraction Ξ¦ :π΄βββ π such that
(1) Ker Ξ¦ is ο¬nitely generated.
(2) π΄βπ πΎ=πΎ[1].
(3) π΄βπ π(π) is geometrically integral overπ(π) for each height one prime idealπ of π .
Thenπ΄is ο¬nitely generated overπ and there exists a ο¬nite birational extension π β² of π and an invertible ideal πΌ of π β² such that π΄βπ π β²βΌ=ππ¦ππ β²(πΌ).
Theorem C. Let π be a Noetherian domain containing βwith ο¬eld of frac- tions πΎ andπ΄ a ο¬at π -algebra with a retractionΞ¦ :π΄βββ π such that
(1) Ker Ξ¦ is ο¬nitely generated.
(2) π΄βπ πΎ=πΎ[1].
(3) π΄βπ π(π)is an integral domain for each height one prime ideal π ofπ .
Then π΄ is an πΈ1-ο¬bration over π . Thus, if π is seminormal, then π΄ βΌ= ππ¦ππ (πΌ) for some invertible ideal πΌ of π .
As a consequence of Theorem A, we get the following L¨uroth-type result (see Corollary 3.3.6):
Corollary A. Letπ be a UFD with ο¬eld of fractionsπΎ andπ΄a ο¬atπ -algebra with a retraction Ξ¦ :π΄βββ π such that
(1) πΎππ Ξ¦ is ο¬nitely generated.
(2) π΄βπ πΎ=πΎ[1].
(3) π΄βπ π(π)is an integral domain for each height one prime ideal π ofπ .
Then there exists π₯βπΎππ Ξ¦ such thatπ΄=π [π₯] =π [1].
β Factorial πΈ
1-form with retraction
Let π be a ο¬eld with algebraic closure Β―π and let π ,β π΄ be π-algebras. We shall call π΄ an πΈ1-form overπ ifπ΄βπΒ―π= (π βπΒ―π)[1]. It is well known that any separable πΈ1-form over any ο¬eld is trivial. More generally, the following result ( [Dut00], Theorem 7) shows that a separableπΈ1-form over any arbitrary commutative algebra is trivial.
Theorem 1.0.4. Let π be a ο¬eld, πΏ a separable ο¬eld extension of π, π a π- algebra and π΄ an π -algebra such that π΄βππΏβΌ=ππ¦π(π βππΏ)(πβ²) for a ο¬nitely generated rank one projective module πβ² over π βππΏ. Then π΄ βΌ= ππ¦ππ (π) for a ο¬nitely generated rank one projective moduleπ over π .
If π is not perfect, there exist non-trivial purely inseparable πΈ1-forms.
Asanuma gave a complete structure theorem for purely inseparable πΈ1-forms over a ο¬eld π of characteristic π >2 ( [Asa05], Theorem 8.1). However, from Asanumaβs results, it can be deduced that any factorialπΈ1-form over a ο¬eldπ with a π-rational point is trivial, i.e, we have the following result:
Theorem 1.0.5. Let k be a ο¬eld and π΄ an πΈ1-form overπ such that (1) π΄ is a UFD.
(2) π΄ has a π-rational point.
Then π΄=π[1].
In Chapter 4 of this thesis we prove the following generalization (see Theo- rem 4.2.2) of the above result. Our result also gives a simple proof of Theorem 1.0.5 without using Asanumaβs intricate structure theorem.
Theorem D.Let π be a ο¬eld and let π ,βπ΄ be π-algebras such that (1) π΄ is a UFD.
(2) There is a retraction Ξ¦ :π΄ββπ .
(3) π΄ is an πΈ1-form over π .
Then π΄=π [1].
β Planes of the form π(π, π )π
πβ π(π, π ) over a DVR
Let π be a ο¬eld and π β π[π1, π2, . . . , ππ](= π[π]). We say π is a vari- able in π[π1, π2, . . . , ππ] if there exist elements π1, π2, . . . , ππβ1 such that π[π1, π2, . . . , ππ] = π[π][π1, π2, . . . , ππβ1] = π[π][πβ1]. It is obvious that if π β π[π1, π2, . . . , ππ](= π[π]) is a variable, then π[π1, π2, . . . , ππ]/(π) = π[πβ1]. Naturally one asks whether the converse holds:
Problem 1. Let π be a ο¬eld,π β₯2 an integer and πβπ[π1, π2, . . . , ππ](=
π[π])be such thatπ[π1, π2, . . . , ππ]/(π) =π[πβ1]. Is thenπ[π1, π2, . . . , ππ] = π[π][πβ1]?
In aο¬ne algebraic geometry, this problem is generally known as the Epi- morphism problem. While the problem is open in general, a few special cases have been investigated by some mathematicians. For such cases, one also considers the corresponding generalized epimorphism problem.
Problem 1β². Let π be an integral domain, π β₯ 2 an integer and π β π [π1, π2, . . . , ππ](=π [π]) be an element such thatπ [π1, π2, . . . , ππ]/(π) = π [πβ1]. Is then π [π1, π2, . . . , ππ] =π [π][πβ1]?
The ο¬rst major breakthrough in this area was got, independently, by Abhyankar-Moh ( [AM75]) and Suzuki ( [Suz74]). They showed that Problem 1 has an aο¬rmative answer for the case π= 2 when the characteristic of the ο¬eld πis 0:
Theorem 1.0.6. Let π be a ο¬eld of characteristic 0. Suppose that π β π[π, π](=π[2]) is such that π[π, π]/(π) =π[1]. Then π[π, π] =π[π][1].
This theorem is known as the famous Abhyankar-Moh and Suzuki Epi- morphism Theorem. The following well known counter example shows that Theorem 1.0.6 does not hold over ο¬elds of positive characteristic.
Example 1.0.7. Let π be a ο¬eld of characteristicπ > 0 and π =πππ βπβ ππ π β π[π, π](= π[2]) where π β€ π and π β₯ 2. Then π[π, π]/(π) = π[1] but π[π, π]β=π[π][1] (see [Abh77], Example 9.12, pg. 72).
In ( [RS79], Theorem 2.6.2), Russell-Sathaye showed that Theorem 1.0.6 holds over locally factorial Krull domains of characteristic 0. The most gen- eralized version of Theorem 1.0.6 has been obtained by Bhatwadekar. He has shown that the theorem can be extended to any seminormal domain of char- acteristic 0 and to any integral domain containing a ο¬eld of characteristic 0 ( [Bha88], Theorem 3.7 and Theorem 3.9):
Theorem 1.0.8. Let π be a seminormal domain of characteristic 0 or an integral domain containing β. Let π β π [π, π](= π [2]) be such that π [π, π]/(π) =π [1]. Then π [π, π] =π [π][1].
The case π = 3 of Problem 1 is still open in general. Among the partial results in this direction, the following theorem of Kaliman ( [Kal02]) deserves a special mention.
Theorem 1.0.9. Let π β β[π, π, π] be such that β[π, π, π]/(πβπ) for all but ο¬nitely many πββ. Then β[π, π, π] =β[π][2].
For certain speciο¬c forms of π, aο¬rmative answers (to the caseπ = 3 of Problem 1) had been obtained by Sathaye, Russell and Wright. In particular, whenπis of the formπ(π, π)ππβπ(π, π), aο¬rmative answers were obtained in the following cases:
(1) π= 1, πa ο¬eld of characteristic 0 (A. Sathaye, [Sat76]).
(2) π= 1, πa ο¬eld of any characteristic (P. Russell, [Rus76]).
(3) π β₯ 2 and π an algebraically closed ο¬eld of characteristic π β₯ 0 with πβ€π(D. Wright, [Wri78]).
In Chapter 5 of the thesis, we ο¬rst show that the result (3) of D. Wright can be generalized to any ο¬eld, not necessarily algebraically closed, in the following form (see Theorem 5.2.5):
Theorem E. Let π be a ο¬eld of characteristic π β₯ 0 and let π βπ[π, π, π]
be of the form πππβπ where π, π β π[π, π] with π β= 0 and π is an integer
β₯ 2 not divisible by π. Suppose that π΅ := π[π, π, π]/(π) = π[2] and identify π[π, π] with its image in π΅. Then there exist variables π, π in π΅ such that π is the image of π in π΅, π β π[π, π], π β π[π], π[π, π] = π[π, π] and π[π, π, π] =π[π, π, π].
We will then discuss how far the result of David Wright can be generalized to the case of DVR and more general rings so that we can get some answers to Problem 1β² forπ= 3 whenπ=π(π, π)ππβπ(π, π),πβ₯2.
The study of Epimorphism problem (Problem 1β²) for π = 3 over a DVR containing βhas an additional importance in that it is closely related to the study ofπΈ2-ο¬bration over a regular local ring of dimension 2. We recall below the connection.
Let π be a ring and π΄ an π -algebra. If π΄ = π [2], it is obvious that π΄ is an πΈ2-ο¬bration over π . Now, what about the converse? If A is an πΈ2- ο¬bration over π , is thenπ΄=π [2]? Till now this is an open problem whenπ is a regular local ring containing β. However, some partial results have been obtained in this direction. In ( [Sat83]), Sathaye showed that an πΈ2-ο¬bration over a DVR containing β is πΈ2. It can be seen by a result of Bass-Connell- Wright ( [BCW77]) that over a PID containing β, an πΈ2-ο¬bration is πΈ2. An immediate question occurring after this result is the following:
Problem 2. Let π be a regular local ring of dimension two containing β.
Suppose π΄ is an πΈ2-ο¬bration over π . Is then π΄=π [2]?
Though Problem 2 is open till now, Bhatwadekar-Dutta showed in ( [BD94b], section 4) that this problem is closely related to the following Epi- morphism problem (a special case of Problem 1β²) in the sense that a counter example to this Epimorphism problem (Problem 3) will give rise to a counter example to Problem 2:
Problem 3. Let (π , π‘) be a DVR containingβand let πβπ [π, π, π](=π [3]) be such that π [π, π, π]/(π) =π [2]. Is thenπ [π, π, π] =π [π][2]?
Hence, to explore Problem 2, it is relevant to explore Problem 3 at least for polynomials like π = π(π, π)ππβπ(π, π) for which the corresponding Problem 1 (withπ= 3) has already been settled.
The ο¬rst investigation in this direction was made by Bhatwadekar-Dutta in [BD94a]. They showed ( [BD94a], Theorem 3.5) that Problem 3 has an aο¬rmative answer (in any characteristic) whenπ=π(π, π)πβπ(π, π) withπ‘β€ π(π, π), thereby partially generalizing A. Sathayeβs theorem on linear planes over a ο¬eld ( [Sat76]).
In Chapter 5 we will show that Problem 3 has an aο¬rmative answer for polynomials of the form π=π(π, π)ππβπ(π, π), whereπβ₯2 is an integer not divisible by the characteristic ofπ /π‘π , thereby obtaining a generalization of D. Wrightβs theorem ( [Wri78], Theorem). More precisely, we will prove the following (see Theorem 5.3.3 ):
Theorem F. Let (π , π‘) be a DVR with residue ο¬eldπ. Letπβπ [π, π, π](=
π [3]) be of the formπ=πππβπwhere π, πβπ [π, π] with πβ= 0 andπis an integerβ₯2 such thatπis not divisible by the characteristic ofπ /π‘π . Suppose that π [π, π, π]/(π) = π [2]. Then π [π, π, π] = π [π, π][1], π [π, π] = π [π][1]
and πβπ [π0] whereπΎ[π, π] =πΎ[π0, π].
The proof of Bhatwadekar-Duttaβs theorem on linear planes over a DVR is highly technical. However, in the case of planes of the form πππβπ with π β₯ 2, the proof turns out to be much simpler due to the fact that π is a variablealong with π.
Using theorems on residual variables of Bhatwadekar-Dutta ( [BD93]), we shall show that Theorem F can be further generalized over (i) any integral domain containingβ and (ii) any Noetherian UFD containing a ο¬eld of char- acteristicπβ₯0 whereπβ€π. We shall prove (see Theorem 5.4.1 and Theprem 5.4.2):
Theorem G. Letπ be an integral domain containingβ. Letπβπ [π, π, π](=
π [3]) be of the form π = πππβπ where π, π β π [π, π] and π is an integer
β₯ 2. Suppose that π [π, π, π]/(π) = π [2]. Then π [π, π, π] = π [π, π][1] and π [π, π] =π [π][1].
Theorem H. Let π be a Noetherian UFD containing a ο¬eld of characteristic πβ₯0 andπβπ [π, π, π](=π [3]) be of the formπππβπwhere π, πβπ [π, π], πβ= 0 and π is an integer β₯2 such that πβ€π. Suppose that π [π, π, π]/(π) = π [2]. Then π [π, π, π] =π [π, π][1] and π [π, π] =π [π][1].
The results obtained in Chapter 3 and Chapter 5 were obtained in two
joint works with my supervisor Dr. Amartya K. Dutta ( [DDa], [DDb]); and the results of Chapter 4 was obtained in my independent work [Das].
Preliminaries
Throughout the thesis π will denote a commutative ring with unity. The notation π΄ = π [π] will mean that π΄ is isomorphic, as an π -algebra, to a polynomial ring inπ variables over π .
Deο¬nitions
1. An π -algebra π΄ is said to be anπ΄π-ο¬bration over π if (i) π΄ is ο¬nitely generated overπ .
(ii) π΄ is ο¬at overπ .
(iii) π΄βπ π(π) =π(π)[π] for all prime ideals π ofπ .
2. Letπbe a ο¬eld, Β―πdenote the algebraic closure ofπandπ be aπ-algebra.
An π -algebra π΄ is said to be an πΈπ-form over π (with respect to π) if π΄βπΒ―π= (π βππ)Β― [π].
3. Letπbe a ο¬eld and Β―πdenote the algebraic closure ofπ. Aπ-algebraπ is said to begeometrically integral over πifπ βππΒ― is an integral domain.
4. Let π be a ο¬eld. A π-algebra π΄ is said to be geometrically normal if π΄βπΒ―πis a normal domain.
5. A reduced ring π is said to beseminormal if it satisο¬es the condition : forπ, πβπ withπ2 =π3, there existsπ‘βπ such thatπ‘3 =πandπ‘2 =π.
6. Letπ΄be a ring and π be a subring ofπ΄. An π -algebra homomorphism πΌ:π΄βββ π is called a retraction from π΄ to π and π is called aretract of π΄.
11
7. Letπbe a ο¬eld. Aπ-algebra π΄is said to have aπ-rational point if there is a retraction fromπ΄ toπ.
Results
We state some results which have been used subsequently. The ο¬rst result occurs in ( [BD95], Lemma 3.4).
Lemma 2.0.10. Letπ be a Noetherian ring andπ 1 a ring containingπ which is ο¬nitely generated as anπ -module. Ifπ΄is a ο¬atπ -algebra such thatπ΄βπ π 1 is a ο¬nitely generated π 1-algebra, then π΄ is a ο¬nitely generated π -algebra.
The following result follows from ( [BD95], Lemma 3.3 and Corollary 3.5).
Lemma 2.0.11. Letπ be a Noetherian ring andπ΄a ο¬atπ -algebra such that, for every minimal prime idealπ of π ,π π΄ is a prime ideal ofπ΄,π π΄β©π =π and π΄/π π΄ is ο¬nitely generated over π /π. Then π΄ is ο¬nitely generated over π .
We now quote a theorem on ο¬nite generation due to N. Onoda ( [Ono84], Theorem 2.20).
Theorem 2.0.12. Let π be a Noetherian domain and let π΄ be an integral domain containing π such that
(1) There exists a non zero element π‘ β π΄ for which π΄[1/π‘] is a ο¬nitely generated π -algebra.
(2) π΄πͺ is a ο¬nitely generated π πͺ-algebra for each maximal ideal πͺof π .
Then π΄ is a ο¬nitely generated π -algebra.
The results onπΈ1-ο¬brations in ( [BD95], [Dut95], [DO07]) crucially involve certain patching techniques. We state below one such βpatching lemmaβ ( [DO07], Corollary 3.2).
Lemma 2.0.13. Let π β π΄ be integral domains with π΄ being faithfully ο¬at over π . Suppose that there exists a non-zero element π‘βπ such that
(1) π΄[1/π‘] =π [1/π‘][1].
(2) πβ1π΄= (πβ1π )[1], where π ={π βπ β£π is not a zero-divisor in π /π‘π }.
Then there exists an invertible ideal πΌ in π such thatπ΄βΌ=ππ¦ππ (πΌ).
Now, we state the result of D. Wright ( [Wri78], Pg. 95) which we will generalize in Chapter 5.
Theorem 2.0.14. Letπbe an algebraically closed ο¬eld of characteristicπβ₯0.
Letπβπ[π, π, π](=π[3])be of the formπππβπwhereπ, πβπ[π, π]withπβ= 0 and π is an integer β₯2 not divisible by π. Suppose thatπ[π, π, π]/(π) =π[2]. Then there exist variables π,Λ πΛ in π[π, π]such that π=πΛ and πβπ[π]Λ and π[π, π, π] =π[π, π, π].Λ
We also mention some relevant result on π΄π’π‘π(π[2]) over a ο¬eld π (see [Wri78], Appendix, Theorems 2 and 3).
Theorem 2.0.15. Let π be a ο¬eld and π΄ = π[π, π](= π[2]). Let πΊπ΄2(π) denote the group of π-automorphisms of π΄, π΄π2(π) the subgroup of πΊπ΄2(π) deο¬ned by π΄π2(π) ={(π, π)7β(πΌ1π+π½1π +πΎ1, πΌ2π+π½2π +πΎ2)β£πΌπ, π½π, πΎπ β π and πΌ1π½2 βπΌ2π½1 β= 0}, β°2(π) the subgroup of πΊπ΄2(π) deο¬ned by β°2(π) = {(π, π) 7β (πΌπ +β(π), π½π +πΎ)β£ πΌ, π½ β πβ, πΎ β π and β(π) β π[π]} and π΅π2(π) =π΄π2(π)β© β°2(π). Then πΊπ΄2(π) =π΄π2(π)βπ΅π2(π)β°2(π). Moreover, if π β πΊπ΄2(π) is of ο¬nite order, then there exists π βπΊπ΄2(π) such that either π ππβ1 βπ΄π2(π) or π ππβ1 β β°2(π).
The next result is due to A. Sathaye ( [Sat76], Corollary 1). We will use it to prove Lemma 5.2.2.
Theorem 2.0.16. Let πΏβ£π be a separable ο¬eld extension. Assume that there exist ββπ[π, π]and elements π’π βπΏ[π, π]for 1β€πβ€π such that
1 πΏ[π, π]/(π’π) =πΏ[1] for each π.
2 (π’π, π’π)πΏ[π, π] =πΏ[π, π]for πβ=π.
3 β= βπ
π=1
π’πππ, ππ>0.
Then there exist π’βπ[π, π], ππ βπΏβ and ππ βπΏ such that π’π =πππ’+ππ for 1β€πβ€π .
We will also use the following special case of the result ( [Dut00], Theorem 7).
Theorem 2.0.17. Let π be a ο¬eld, πΏ a separable ο¬eld extension of π, π΄ a factorial π-domain andπ΅ anπ΄-algebra such that π΅βππΏ= (π΄βππΏ)[1]. Then π΅ =π΄[1].
The following version of Abhyankar-Eakin-Heinzerβs cancellation theorem ( [AEH72], Theorem 3.3) will be used in the proofs.
Theorem 2.0.18. Let π΄ be an aο¬ne domain over a ο¬eld π such that π is algebraically closed in π΄ and π‘π.ππππ(π΄) = 1. Suppose that π΅ is a π-algebra such that π΄[π]=π΅[π] for some πβ₯1. Then either π΅=π΄ or π΅ βΌ=π΄=π[1].
We now state a version of the Russell-Sathaye criterion ( [RS79], Theorem 2.3.1) for a ring to be a polynomial algebra over a subring (see [BD94a], Theorem 2.6).
Theorem 2.0.19. Let π βπ΄be integral domains withπ΄being ο¬nitely gener- ated over π . Suppose that there exist primes π1, π2, . . . , ππ in π such that for each π,1β€πβ€π,
(1) ππ remains prime in π΄, (2) πππ΄β©π =πππ ,
(3) π΄[π 1
1π2...ππ] =π [π 1
1π2...ππ][1] and
(4) π /πππ is algebraically closed in π΄/πππ΄.
Then π΄=π [1].
The following result from ( [BD94a], Lemma 2.5) will enable us to apply Theorem 2.0.19.
Lemma 2.0.20. Let π be an integral domain and πΉ β π [π, π](= π [2]) be such that π [π, π]/(πΉ) =π [1]. Then π [πΉ]is algebraically closed in π [π, π].
We now quote a result of E. Hamann ( [Ham75], Theorem 2.6).
Theorem 2.0.21. Let π be a Noetherian ring such that π red is seminormal.
Then π [1] is R-invariant, i.e., if π΄ is an π -algebra such that π΄[π] = π [π+1]
as π -algebras, thenπ΄=π [1].
Finally, we state a result on residual variables which will be our main tool to prove Theorem G and Theorem H. It comes as a direct consequence of Theorem 3.1, Theorem 3.2 and Remark 3.4 in [BD93].
Theorem 2.0.22. Letπ be a Noetherian domain such that either π contains β or π is seminormal, π΄ be a polynomial algebra in π variables over π and π1, π2, . . . , ππβ1 βπ΄. Then the following are equivalent:
1. π΄=π [π1, π2, . . . , ππβ1][1].
2. π΄βπ π(π) = (π [π1, π2, . . . , ππβ1]βπ π(π))[1] for every prime ideal π of π .
Codimension-one πΈ 1 -ο¬bration with retraction
3.1 Preview
The following result on πΈ1-ο¬brations was proved in ( [Dut95], Theorem 3.4, Theorem 3.5):
Theorem 3.1.1. Let π be a Noetherian domain with ο¬eld of fractionsπΎ and π΄ a faithfully ο¬at ο¬nitely generated π -algebra such that π΄βπ πΎ =πΎ[1] and π΄βπ π(π) is geometrically integral overπ(π) for each height one prime ideal π of π . Under these hypotheses, we have the following results:
(i) If π is normal, then π΄βΌ=ππ¦ππ (πΌ) for an invertible ideal πΌ of π .
(ii) If π containsβ, then π΄ is anπΈ1-ο¬bration over π .
A striking feature of this result is that conditions on merely the generic and codimension-one ο¬bres imply that all ο¬bres are πΈ1. Analogous results were proved for subalgebras of polynomial algebras ( [BD95], 3.10, 3.12) without the hypothesis βπ΄is ο¬nitely generated overπ β. In this chapter we investigate whether the condition βπ΄ is ο¬nitely generatedβ in Theorem 3.1.1 can be re- placed by a weaker hypothesis like βπ΄ is Noetherianβ when the π algebra π΄ is known to have a retraction to π . Recently, in [BDO], Bhatwadekar-Dutta- Onoda have shown the following:
Theorem 3.1.2. Letπ be a Noetherian normal domain with ο¬eld of fractions πΎ and π΄ a Noetherian ο¬at π -algebra such that π΄π = π π[1] for each prime
17
idealπ ofπ of height one. Suppose that there exists a retractionΞ¦ :π΄βββ π .
Then π΄βΌ=ππ¦ππ (πΌ) for an invertible ideal πΌ in π .
The above theorem occurs in [BDO] as a consequence of a general structure theorem for any faithfully ο¬at algebra over a Noetherian normal domain π which is locally πΈ1 in codimension-one. The statements and proofs in [BDO]
are quite technical. In this chapter, we will ο¬rst prove (see Theorem 3.3.5) an analogue of Theorem 3.1.1 (i). Our approach, which is more in the spirit of the proof in ( [Dut95], 3.4), will provide a short and direct proof of Theorem 3.1.2.
Next we will prove an analogous version of Theorem 3.1.1 (ii) (see Theorem 3.3.9).
3.2 A version of Russell-Sathaye criterion for an al- gebra to be a polynomial algebra
In this section we present a version of Russell-Sathaye criterion ( [RS79], The- orem 2.3.1) for an algebra to be a polynomial algebra. Our version is an extension of the version given by Dutta-Onoda ( [DO07], Theorem 2.4) and suitable for algebras which are known to have retractions to the base ring. For convenience, we ο¬rst record a few preliminary results. The ο¬rst result is easy.
Lemma 3.2.1. Let π΅ β π΄ be integral domains. Suppose that there exists a non-zero element π in π΅ such that π΅[1/π] =π΄[1/π] and ππ΄β©π΅ =ππ΅. Then π΅ =π΄.
Lemma 3.2.2. Let πΆ be aπ·-algebra such that π· is a retract of πΆ. Then the following hold:
(I) ππΆβ©π·=ππ· for all πβπ·.
(II) If π·βπΆ are domains, thenπ· is algebraically closed in πΆ.
Proof. Proof of (I): Let π βπ·. Note that ππΆβ©π·=ππ· is equivalent to say that the map π·/ππ· ββ πΆ/ππΆ is injective. Now since π· is a retract of πΆ, the composite mapπ·/ππ·ββπΆ/ππΆβββ π·/ππ· is identity and hence the map π·/ππ·ββπΆ/ππΆ is injective.
Proof of (II): Let π : πΆβββ π· be the retraction and let π‘ β πΆβ{0} be algebraic over π·. Then there exits a polynomial π(π) βπ·[π] (unique upto
a constant multiple) of least degree such that π(π‘) = 0. Note that π(π‘) β= 0.
Since π(π(π‘)) =π(π(π‘)) = 0 and since π(π‘) βπ·, we must have π(π) = (πβ π(π‘))π(π) whereπππ(π(π))< πππ(π(π)). Now, since π(π) is a polynomial of least degree such that π(π‘) = 0, we get π(π‘) β= 0 and hence from the relation π(π‘) = (π‘βπ(π‘))π(π‘) = 0 we have π‘=π(π‘), i.e.,π‘βπ·. Thusπ·is algebraically closed inπΆ.
Lemma 3.2.3. Let π be a ring and π΄ be anπ -algebra with a generating set π = {π₯π : π β Ξ} where Ξ is some indexing set. Suppose that there is a retraction Ξ¦ :π΄βββ π . ThenπΎππ Ξ¦ = ({π₯πβππ :πβΞ})π΄ where ππ = Ξ¦(π₯π) for each πβΞ.
Proof. LetπΛ={π₯πβππ:πβΞ} andπΌ be the ideal ofπ΄ generated byπ. NoteΛ that π [π] = π [π]. It is easy to see thatΛ π΄ = π βπΎππ Ξ¦ = π βπΌ. Since πΌ βπΎππ Ξ¦, it follows that πΎππ Ξ¦ =πΌ.
Lemma 3.2.4. Let π βπ΄be integral domains andΞ¦ :π΄βββ π be a retraction with ο¬nitely generated kernel. Suppose that there exists an element π which is a non-zero non-unit in π such that π΄[1/π] = π [1/π][1]. Then there exists π₯βπΎππ Ξ¦such that π₯ /βππ΄and π΄[1/π] =π [1/π][π₯].
Proof. Suppose, if possible, that π₯ β ππ΄ for every π₯ β πΎππ Ξ¦ for which π΄[1/π] =π [1/π][π₯].
Let πΎππ Ξ¦ = (π1, π2, . . . , ππ)π΄. Choose π₯0 β πΎππ Ξ¦ such that π΄[1/π] = π [1/π][π₯0]. Note that Ξ¦ extends to a retraction Ξ¦π : π΄[1/π]βββ π [1/π] with kernelπ₯0(π΄[1/π]). By our assumption,π₯0 =ππ₯1 for someπ₯1 βπ΄. Obviously, π₯1 β πΎππ Ξ¦ and π΄[1/π] = π [1/π][π₯1] and hence π₯1 β ππ΄. Arguing in a similar manner, we get π₯2 βπΎππ Ξ¦ such that π₯1 =ππ₯2, π΄[1/π] =π [1/π][π₯2] and π₯2 β ππ΄. Continuing this process we get a sequence {π₯π}πβ₯0 such that π₯π β πΎππ Ξ¦, π΄[1/π] = π [1/π][π₯π] and π₯π = ππ₯π+1. Thus π₯0 = πππ₯π for all πβ₯1.
Note that (π₯0, π₯1, . . . , π₯π, . . .)π΄ β(π1, π2, . . . , ππ)π΄. But since ππ β π΄ β π΄[1/π] = π [1/π][π₯0], there exist ππ β β and πΌππ β π [1/π] such that ππ =
ππ
β
π=0
πΌπππ₯0π. Choose π β β such that πΌπππππ β π for all π, π and set πππ :=
πΌπππππ. Now since π₯0, ππ β πΎππ Ξ¦π, we have πΌπ0 = 0 for all π and hence ππ = βππ
π=1
πΌπππ₯0π. Thus ππ = βππ
π=1
ππππ₯ππ βπ₯ππ [π₯π]βπ₯ππ΄ for allπ, 1β€πβ€π.
So, we have πΎππ Ξ¦ = (π1, π2, . . . , ππ)π΄=π₯ππ΄. Nowπ₯π+1 βπΎππ Ξ¦ = π₯ππ΄,
which implies thatπ₯π+1=πΌπ₯π for someπΌβπ΄. Sinceπ₯π =ππ₯π+1, it follows that πΌπ = 1, which is a contradiction to the fact that π is not a unit in π΄.
Thus there exists π₯βπΎππ Ξ¦ such that π₯ /βππ΄and π΄[1/π] =π [1/π][π₯].
Now we present a version of Russell-Sathaye criterion when there exists a retraction.
Proposition 3.2.5. Let π β π΄ be integral domains such that there exists a retraction Ξ¦ :π΄βββ π . Suppose that there exists a prime π in π such that
(1) π is a prime in π΄.
(2) π΄[1/π] =π [1/π][1].
Then ππ΄β©π =ππ ,π /ππ is algebraically closed in π΄/ππ΄ and there exists an increasing chain π΄0 β π΄1 β π΄2 ... β π΄π β ... of subrings of A and a sequence of elements {π₯π}πβ₯0 in πΎππ Ξ¦ with π₯0π΄ βπ₯1π΄ β β β β β π₯ππ΄ β. . . such that
(a) π΄π=π [π₯π] =π [1] for all πβ₯0.
(b) π΄[1/π] =π΄π[1/π]for all πβ₯0.
(c) ππ΄β©π΄πβππ΄π+1 for all πβ₯0.
(d) π΄= βͺ
πβ₯0π΄π=π [π₯1, π₯2, . . . , π₯π, . . .].
(e) πΎππ Ξ¦ = (π₯0, π₯1, π₯2, . . . , π₯π, . . .)π΄.
Moreover the following are equivalent:
(i) πΎππ Ξ¦ is ο¬nitely generated.
(ii) πΎππ Ξ¦ =π₯ππ΄ for some π β₯0.
(iii) π΄ is ο¬nitely generated over π .
(iv) π΄=π [π₯π]for some π β₯0.
(v) There exists π₯βπΎππ Ξ¦βππ΄such that π΄=π [π₯] =π [1]. The conditions (i)β(v) will be satisο¬ed if β©
πβ₯0πππ΄= (0).
Proof. ππ΄β©π = ππ by Lemma 3.2.2. Since Ξ¦ induces a retraction Ξ¦π : π΄/ππ΄βββ π /ππ ,π /ππ is algebraically closed in π΄/ππ΄by Lemma 3.2.2.
By condition (2), there exists π₯β²0 βπ΄ such that π΄[1/π] =π [1/π][π₯β²0]. Let π₯0 =π₯β²0βΞ¦(π₯β²0). Thenπ₯0βπΎππΞ¦ and π΄[1/π] =π [1/π][π₯0] =π [1/π][1]. Set π΄0 :=π [π₯0](=π [1]). Then π΄0 βπ΄ and π΄[1/π] =π΄0[1/π] =π [1/π][π₯0].
Now suppose that we have obtained elementsπ₯0, π₯1, . . . , π₯πβπΎππΞ¦ such that setting π΄π := π [π₯π](=π [1]) for all π, 0β€πβ€π, we have π΄0 βπ΄1 β π΄2 β β β βπ΄πβπ΄ and π΄π[1/π] =π΄[1/π]; 0β€πβ€π.
We now describe our choice ofπ₯π+1:
Let π₯π denote the image of π₯π inπ΄/ππ΄. We consider separately the two possibilities:
(I) π₯π is transcendental overπ /ππ .
(II) π₯π is algebraic over π /ππ .
Case I : π₯π is transcendental over π /ππ . In this case the map π΄π/ππ΄π(=
π [π₯π]/ππ [π₯π]) ββ π΄/ππ΄ is injective, i.e., ππ΄π = ππ΄β©π΄π. Since π΄π[1/π] = π΄[1/π], we get π΄π=π΄ by Lemma 3.2.1. Now we setπ₯π+1 :=π₯πand π΄π+1 :=
π [π₯π+1](=π΄π=π΄).
Case II: π₯π is algebraic over π /ππ . Since π /ππ is algebraically closed in π΄/ππ΄, we see thatπ₯πβπ /ππ . Thusπ₯π=ππ’π+ππ for someπ’πβπ΄ andππβ π . Applying Ξ¦, we get 0 = Ξ¦(π₯π) =πΞ¦(π’π) +ππ showing that ππ β ππ and hence π₯π βππ΄. Set π₯π+1 :=π₯π/π(βπ΄). Clearly π₯π+1 βπΎππ Ξ¦. Now setting π΄π+1 :=π [π₯π+1](=π [1]), we see that π΄0 β π΄1 βπ΄2 β β β β π΄π β π΄π+1 β π΄ and π΄π+1[1/π] =π΄π[1/π] =π΄[1/π].
Thus we set π₯π+1 :=π₯π orπ₯π+1:=π₯π/πdepending on whether the image of π₯π in π΄/ππ΄ is transcendental or algebraic over π /ππ . By construction, conditions (a) and (b) hold. We now verify (c).
If π₯π = π₯π+1, i.e., π΄π+1 = π΄π = π΄, then ππ΄β© π΄π = ππ΄ = ππ΄π+1. Now consider the case π₯π = ππ₯π+1 β ππ΄π+1. Let π β ππ΄β©π΄π. Then π = π0+π1(ππ₯π+1) +β β β +ππ(ππ₯π+1)π for some π β₯0 and π0, π1, . . . , ππ βπ . Then π0 βππ΄β©π =ππ βππ΄π+1. Therefore, πβππ΄π+1. Thusππ΄β©π΄πβπ΄π+1.
We now prove (d). Let π΅ = βͺ
πβ₯0π΄π. Obviously, π΅ β π΄ and π΅[1/π] = π΄[1/π]. Hence, by Lemma 3.2.1, it is enough to show that ππ΄β©π΅=ππ΅.